Code reorganization: move files into their places.

This is in a separate commit to ensure renames are properly preserved.

342 files changed, 90658 insertions, 0 deletions

diff --git a/src/gallium/Makefile b/src/gallium/Makefile
new file mode 100644
index 0000000000..d880d090c1
--- /dev/null
+++ b/src/gallium/Makefile
@@ -0,0 +1,28 @@
+TOP = ../../..
+include $(TOP)/configs/current
+
+
+ifeq ($(CONFIG_NAME), linux-cell)
+CELL_DIR = cell
+endif
+
+ifeq ($(CONFIG_NAME), linux-llvm)
+LLVM_DIR = llvm
+endif
+
+SUBDIRS = softpipe i915simple i965simple failover pipebuffer $(CELL_DIR) $(LLVM_DIR)
+
+
+default: subdirs
+
+
+subdirs:
+	@for dir in $(SUBDIRS) ; do \
+		if [ -d $$dir ] ; then \
+			(cd $$dir && $(MAKE)) || exit 1 ; \
+		fi \
+	done
+
+
+clean:
+	rm -f `find . -name \*.[oa]`
diff --git a/src/gallium/Makefile.template b/src/gallium/Makefile.template
new file mode 100644
index 0000000000..8e84f8eb2d
--- /dev/null
+++ b/src/gallium/Makefile.template
@@ -0,0 +1,63 @@
+# -*-makefile-*-
+
+
+# We still have a dependency on the "dri" buffer manager.  Most likely
+# the interface can be reused in non-dri environments, and also as a
+# frontend to simpler memory managers.
+#
+COMMON_SOURCES = 
+
+OBJECTS = $(C_SOURCES:.c=.o) \
+          $(CPP_SOURCES:.cpp=.o) \
+	  $(ASM_SOURCES:.S=.o) 
+
+
+### Include directories
+INCLUDES = \
+	-I. \
+	-I$(TOP)/src/mesa/pipe \
+	-I$(TOP)/src/mesa \
+	-I$(TOP)/include \
+        $(DRIVER_INCLUDES)
+
+
+##### RULES #####
+
+.c.o:
+	$(CC) -c $(INCLUDES) $(CFLAGS) $(DRIVER_DEFINES) $< -o $@
+
+.cpp.o:
+	$(CXX) -c $(INCLUDES) $(CXXFLAGS) $(DRIVER_DEFINES) $< -o $@
+
+.S.o:
+	$(CC) -c $(INCLUDES) $(CFLAGS) $(DRIVER_DEFINES)  $< -o $@
+
+
+##### TARGETS #####
+
+default: depend symlinks $(LIBNAME)
+
+
+$(LIBNAME): $(OBJECTS) Makefile $(TOP)/src/mesa/pipe/Makefile.template
+	$(TOP)/bin/mklib -o $@ -static $(OBJECTS) $(DRIVER_LIBS)
+
+
+depend: $(C_SOURCES) $(CPP_SOURCES) $(ASM_SOURCES) $(SYMLINKS)
+	rm -f depend
+	touch depend
+	$(MKDEP) $(MKDEP_OPTIONS) $(DRIVER_DEFINES) $(INCLUDES) $(C_SOURCES) $(CPP_SOURCES) \
+		$(ASM_SOURCES) 2> /dev/null
+
+
+# Emacs tags
+tags:
+	etags `find . -name \*.[ch]` `find ../include`
+
+
+# Remove .o and backup files
+clean::
+	-rm -f *.o */*.o *~ *.so *~ server/*.o $(SYMLINKS)
+	-rm -f depend depend.bak
+
+
+include depend
diff --git a/src/gallium/README.portability b/src/gallium/README.portability
new file mode 100644
index 0000000000..c70ca774da
--- /dev/null
+++ b/src/gallium/README.portability
@@ -0,0 +1,43 @@
+	      CROSS-PLATFORM PORTABILITY GUIDELINES FOR GALLIUM3D 
+
+
+= General Considerations =
+
+The state tracker and winsys driver support a rather limited number of
+platforms. However, the pipe drivers are meant to run in a wide number of
+platforms. Hence the pipe drivers, the auxiliary modules, and all public
+headers in general, should stricly follow these guidelines to ensure
+
+
+= Compiler Support =
+
+* Include the p_compiler.h.
+
+* Don't use the 'inline' keyword, use the INLINE macro in p_compiler.h instead.
+
+* Cast explicitly when converting to integer types of smaller sizes.
+
+* Cast explicitly when converting between float, double and integral types.
+
+* Don't use named struct initializers.
+
+* Don't use variable number of macro arguments. Use static inline functions
+instead.
+
+
+= Standard Library =
+
+* Avoid including standard library headers. Most standard library functions are
+not available in Windows Kernel Mode. Use the appropriate p_*.h include.
+
+== Memory Allocation ==
+
+* Use MALLOC, CALLOC, FREE instead of the malloc, calloc, free functions.
+
+* Use align_pointer() function defined in p_util.h for aligning pointers in a
+portable way.
+
+== Debugging ==
+
+TODO
+
diff --git a/src/gallium/SConscript b/src/gallium/SConscript
new file mode 100644
index 0000000000..d9c20e0100
--- /dev/null
+++ b/src/gallium/SConscript
@@ -0,0 +1,9 @@
+Import('*')
+
+#env = env.Clone()
+
+SConscript([
+	'softpipe/SConscript',
+	'i915simple/SConscript',
+	'i965simple/SConscript',
+])
diff --git a/src/gallium/aux/cso_cache/cso_cache.c b/src/gallium/aux/cso_cache/cso_cache.c
new file mode 100644
index 0000000000..9e77e0774d
--- /dev/null
+++ b/src/gallium/aux/cso_cache/cso_cache.c
@@ -0,0 +1,181 @@
+/**************************************************************************
+ *
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ **************************************************************************/
+
+/* Authors:  Zack Rusin <zack@tungstengraphics.com>
+ */
+
+#include "cso_cache.h"
+#include "cso_hash.h"
+
+#if 1
+static unsigned hash_key(const void *key, unsigned key_size)
+{
+   unsigned *ikey = (unsigned *)key;
+   unsigned hash = 0, i;
+
+   assert(key_size % 4 == 0);
+
+   /* I'm sure this can be improved on:
+    */
+   for (i = 0; i < key_size/4; i++)
+      hash ^= ikey[i];
+
+   return hash;
+}
+#else
+static unsigned hash_key(const unsigned char *p, int n)
+{
+   unsigned h = 0;
+   unsigned g;
+
+   while (n--) {
+      h = (h << 4) + *p++;
+      if ((g = (h & 0xf0000000)) != 0)
+         h ^= g >> 23;
+      h &= ~g;
+   }
+   return h;
+}
+#endif
+
+unsigned cso_construct_key(void *item, int item_size)
+{
+   return hash_key((item), item_size);
+}
+
+static struct cso_hash *_cso_hash_for_type(struct cso_cache *sc, enum cso_cache_type type)
+{
+   struct cso_hash *hash = 0;
+
+   switch(type) {
+   case CSO_BLEND:
+      hash = sc->blend_hash;
+      break;
+   case CSO_SAMPLER:
+      hash = sc->sampler_hash;
+      break;
+   case CSO_DEPTH_STENCIL_ALPHA:
+      hash = sc->depth_stencil_hash;
+      break;
+   case CSO_RASTERIZER:
+      hash = sc->rasterizer_hash;
+      break;
+   case CSO_FRAGMENT_SHADER:
+      hash = sc->fs_hash;
+      break;
+   case CSO_VERTEX_SHADER:
+      hash = sc->vs_hash;
+      break;
+   }
+
+   return hash;
+}
+
+static int _cso_size_for_type(enum cso_cache_type type)
+{
+   switch(type) {
+   case CSO_BLEND:
+      return sizeof(struct pipe_blend_state);
+   case CSO_SAMPLER:
+      return sizeof(struct pipe_sampler_state);
+   case CSO_DEPTH_STENCIL_ALPHA:
+      return sizeof(struct pipe_depth_stencil_alpha_state);
+   case CSO_RASTERIZER:
+      return sizeof(struct pipe_rasterizer_state);
+   case CSO_FRAGMENT_SHADER:
+      return sizeof(struct pipe_shader_state);
+   case CSO_VERTEX_SHADER:
+      return sizeof(struct pipe_shader_state);
+   }
+   return 0;
+}
+
+struct cso_hash_iter
+cso_insert_state(struct cso_cache *sc,
+                 unsigned hash_key, enum cso_cache_type type,
+                 void *state)
+{
+   struct cso_hash *hash = _cso_hash_for_type(sc, type);
+   return cso_hash_insert(hash, hash_key, state);
+}
+
+struct cso_hash_iter
+cso_find_state(struct cso_cache *sc,
+               unsigned hash_key, enum cso_cache_type type)
+{
+   struct cso_hash *hash = _cso_hash_for_type(sc, type);
+
+   return cso_hash_find(hash, hash_key);
+}
+
+struct cso_hash_iter cso_find_state_template(struct cso_cache *sc,
+                                             unsigned hash_key, enum cso_cache_type type,
+                                             void *templ)
+{
+   struct cso_hash_iter iter = cso_find_state(sc, hash_key, type);
+   int size = _cso_size_for_type(type);
+   while (!cso_hash_iter_is_null(iter)) {
+      void *iter_data = cso_hash_iter_data(iter);
+      if (!memcmp(iter_data, templ, size))
+         return iter;
+      iter = cso_hash_iter_next(iter);
+   }
+   return iter;
+}
+
+void * cso_take_state(struct cso_cache *sc,
+                      unsigned hash_key, enum cso_cache_type type)
+{
+   struct cso_hash *hash = _cso_hash_for_type(sc, type);
+   return cso_hash_take(hash, hash_key);
+}
+
+struct cso_cache *cso_cache_create(void)
+{
+   struct cso_cache *sc = malloc(sizeof(struct cso_cache));
+
+   sc->blend_hash         = cso_hash_create();
+   sc->sampler_hash       = cso_hash_create();
+   sc->depth_stencil_hash = cso_hash_create();
+   sc->rasterizer_hash    = cso_hash_create();
+   sc->fs_hash            = cso_hash_create();
+   sc->vs_hash            = cso_hash_create();
+
+   return sc;
+}
+
+void cso_cache_delete(struct cso_cache *sc)
+{
+   assert(sc);
+   cso_hash_delete(sc->blend_hash);
+   cso_hash_delete(sc->sampler_hash);
+   cso_hash_delete(sc->depth_stencil_hash);
+   cso_hash_delete(sc->rasterizer_hash);
+   cso_hash_delete(sc->fs_hash);
+   cso_hash_delete(sc->vs_hash);
+   free(sc);
+}
diff --git a/src/gallium/aux/cso_cache/cso_cache.h b/src/gallium/aux/cso_cache/cso_cache.h
new file mode 100644
index 0000000000..116e2eaa2c
--- /dev/null
+++ b/src/gallium/aux/cso_cache/cso_cache.h
@@ -0,0 +1,107 @@
+/**************************************************************************
+ *
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ **************************************************************************/
+
+ /*
+  * Authors:
+  *   Zack Rusin <zack@tungstengraphics.com>
+  */
+
+#ifndef CSO_CACHE_H
+#define CSO_CACHE_H
+
+#include "pipe/p_context.h"
+#include "pipe/p_state.h"
+
+
+struct cso_hash;
+
+struct cso_cache {
+   struct cso_hash *blend_hash;
+   struct cso_hash *depth_stencil_hash;
+   struct cso_hash *fs_hash;
+   struct cso_hash *vs_hash;
+   struct cso_hash *rasterizer_hash;
+   struct cso_hash *sampler_hash;
+};
+
+struct cso_blend {
+   struct pipe_blend_state state;
+   void   *data;
+};
+
+struct cso_depth_stencil_alpha {
+   struct pipe_depth_stencil_alpha_state state;
+   void *data;
+};
+
+struct cso_rasterizer {
+   struct pipe_rasterizer_state state;
+   void *data;
+};
+
+struct cso_fragment_shader {
+   struct pipe_shader_state state;
+   void *data;
+};
+
+struct cso_vertex_shader {
+   struct pipe_shader_state state;
+   void *data;
+};
+
+struct cso_sampler {
+   struct pipe_sampler_state state;
+   void *data;
+};
+
+
+enum cso_cache_type {
+   CSO_BLEND,
+   CSO_SAMPLER,
+   CSO_DEPTH_STENCIL_ALPHA,
+   CSO_RASTERIZER,
+   CSO_FRAGMENT_SHADER,
+   CSO_VERTEX_SHADER
+};
+
+unsigned cso_construct_key(void *item, int item_size);
+
+struct cso_cache *cso_cache_create(void);
+void cso_cache_delete(struct cso_cache *sc);
+
+struct cso_hash_iter cso_insert_state(struct cso_cache *sc,
+                                      unsigned hash_key, enum cso_cache_type type,
+                                      void *state);
+struct cso_hash_iter cso_find_state(struct cso_cache *sc,
+                                    unsigned hash_key, enum cso_cache_type type);
+struct cso_hash_iter cso_find_state_template(struct cso_cache *sc,
+                                             unsigned hash_key, enum cso_cache_type type,
+                                             void *templ);
+void * cso_take_state(struct cso_cache *sc, unsigned hash_key,
+                      enum cso_cache_type type);
+
+#endif
diff --git a/src/gallium/aux/cso_cache/cso_hash.c b/src/gallium/aux/cso_cache/cso_hash.c
new file mode 100644
index 0000000000..0338cb3b47
--- /dev/null
+++ b/src/gallium/aux/cso_cache/cso_hash.c
@@ -0,0 +1,388 @@
+/**************************************************************************
+ *
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ **************************************************************************/
+
+ /*
+  * Authors:
+  *   Zack Rusin <zack@tungstengraphics.com>
+  */
+
+#include "cso_hash.h"
+
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include <assert.h>
+
+#define MAX(a, b) ((a > b) ? (a) : (b))
+
+static const int MinNumBits = 4;
+
+static const unsigned char prime_deltas[] = {
+   0,  0,  1,  3,  1,  5,  3,  3,  1,  9,  7,  5,  3,  9, 25,  3,
+   1, 21,  3, 21,  7, 15,  9,  5,  3, 29, 15,  0,  0,  0,  0,  0
+};
+
+static int primeForNumBits(int numBits)
+{
+   return (1 << numBits) + prime_deltas[numBits];
+}
+
+/*
+    Returns the smallest integer n such that
+    primeForNumBits(n) >= hint.
+*/
+static int countBits(int hint)
+{
+   int numBits = 0;
+   int bits = hint;
+
+   while (bits > 1) {
+      bits >>= 1;
+      numBits++;
+   }
+
+   if (numBits >= (int)sizeof(prime_deltas)) {
+      numBits = sizeof(prime_deltas) - 1;
+   } else if (primeForNumBits(numBits) < hint) {
+      ++numBits;
+   }
+   return numBits;
+}
+
+struct cso_node {
+   struct cso_node *next;
+   unsigned key;
+   void *value;
+};
+
+struct cso_hash_data {
+   struct cso_node *fakeNext;
+   struct cso_node **buckets;
+   int size;
+   int nodeSize;
+   short userNumBits;
+   short numBits;
+   int numBuckets;
+};
+
+struct cso_hash {
+   union {
+      struct cso_hash_data *d;
+      struct cso_node      *e;
+   } data;
+};
+
+static void *cso_data_allocate_node(struct cso_hash_data *hash)
+{
+   return malloc(hash->nodeSize);
+}
+
+static void cso_data_free_node(struct cso_node *node)
+{
+   /* XXX still a leak here.
+    * Need to cast value ptr to original cso type, then free the
+    * driver-specific data hanging off of it.  For example:
+   struct cso_sampler *csamp = (struct cso_sampler *) node->value;
+   free(csamp->data);
+   */
+   free(node->value);
+   free(node);
+}
+
+static struct cso_node *
+cso_hash_create_node(struct cso_hash *hash,
+                      unsigned akey, void *avalue,
+                      struct cso_node **anextNode)
+{
+   struct cso_node *node = cso_data_allocate_node(hash->data.d);
+   node->key = akey;
+   node->value = avalue;
+
+   node->next = (struct cso_node*)(*anextNode);
+   *anextNode = node;
+   ++hash->data.d->size;
+   return node;
+}
+
+static void cso_data_rehash(struct cso_hash_data *hash, int hint)
+{
+   if (hint < 0) {
+      hint = countBits(-hint);
+      if (hint < MinNumBits)
+         hint = MinNumBits;
+      hash->userNumBits = hint;
+      while (primeForNumBits(hint) < (hash->size >> 1))
+         ++hint;
+   } else if (hint < MinNumBits) {
+      hint = MinNumBits;
+   }
+
+   if (hash->numBits != hint) {
+      struct cso_node *e = (struct cso_node *)(hash);
+      struct cso_node **oldBuckets = hash->buckets;
+      int oldNumBuckets = hash->numBuckets;
+      int  i = 0;
+
+      hash->numBits = hint;
+      hash->numBuckets = primeForNumBits(hint);
+      hash->buckets = malloc(sizeof(struct cso_node*) * hash->numBuckets);
+      for (i = 0; i < hash->numBuckets; ++i)
+         hash->buckets[i] = e;
+
+      for (i = 0; i < oldNumBuckets; ++i) {
+         struct cso_node *firstNode = oldBuckets[i];
+         while (firstNode != e) {
+            unsigned h = firstNode->key;
+            struct cso_node *lastNode = firstNode;
+            while (lastNode->next != e && lastNode->next->key == h)
+               lastNode = lastNode->next;
+
+            struct cso_node *afterLastNode = lastNode->next;
+            struct cso_node **beforeFirstNode = &hash->buckets[h % hash->numBuckets];
+            while (*beforeFirstNode != e)
+               beforeFirstNode = &(*beforeFirstNode)->next;
+            lastNode->next = *beforeFirstNode;
+            *beforeFirstNode = firstNode;
+            firstNode = afterLastNode;
+         }
+      }
+      free(oldBuckets);
+   }
+}
+
+static void cso_data_might_grow(struct cso_hash_data *hash)
+{
+   if (hash->size >= hash->numBuckets)
+      cso_data_rehash(hash, hash->numBits + 1);
+}
+
+static void cso_data_has_shrunk(struct cso_hash_data *hash)
+{
+   if (hash->size <= (hash->numBuckets >> 3) &&
+       hash->numBits > hash->userNumBits) {
+      int max = MAX(hash->numBits-2, hash->userNumBits);
+      cso_data_rehash(hash,  max);
+   }
+}
+
+static struct cso_node *cso_data_first_node(struct cso_hash_data *hash)
+{
+   struct cso_node *e = (struct cso_node *)(hash);
+   struct cso_node **bucket = hash->buckets;
+   int n = hash->numBuckets;
+   while (n--) {
+      if (*bucket != e)
+         return *bucket;
+      ++bucket;
+   }
+   return e;
+}
+
+static struct cso_node **cso_hash_find_node(struct cso_hash *hash, unsigned akey)
+{
+   struct cso_node **node;
+
+   if (hash->data.d->numBuckets) {
+      node = (struct cso_node **)(&hash->data.d->buckets[akey % hash->data.d->numBuckets]);
+      assert(*node == hash->data.e || (*node)->next);
+      while (*node != hash->data.e && (*node)->key != akey)
+         node = &(*node)->next;
+   } else {
+      node = (struct cso_node **)((const struct cso_node * const *)(&hash->data.e));
+   }
+   return node;
+}
+
+struct cso_hash_iter cso_hash_insert(struct cso_hash *hash,
+                                       unsigned key, void *data)
+{
+   cso_data_might_grow(hash->data.d);
+
+   struct cso_node **nextNode = cso_hash_find_node(hash, key);
+   struct cso_node *node = cso_hash_create_node(hash, key, data, nextNode);
+   struct cso_hash_iter iter = {hash, node};
+   return iter;
+}
+
+struct cso_hash * cso_hash_create(void)
+{
+   struct cso_hash *hash = malloc(sizeof(struct cso_hash));
+   hash->data.d = malloc(sizeof(struct cso_hash_data));
+   hash->data.d->fakeNext = 0;
+   hash->data.d->buckets = 0;
+   hash->data.d->size = 0;
+   hash->data.d->nodeSize = sizeof(struct cso_node);
+   hash->data.d->userNumBits = MinNumBits;
+   hash->data.d->numBits = 0;
+   hash->data.d->numBuckets = 0;
+
+   return hash;
+}
+
+void cso_hash_delete(struct cso_hash *hash)
+{
+   struct cso_node *e_for_x = (struct cso_node *)(hash->data.d);
+   struct cso_node **bucket = (struct cso_node **)(hash->data.d->buckets);
+   int n = hash->data.d->numBuckets;
+   while (n--) {
+      struct cso_node *cur = *bucket++;
+      while (cur != e_for_x) {
+         struct cso_node *next = cur->next;
+         cso_data_free_node(cur);
+         cur = next;
+      }
+   }
+   free(hash->data.d->buckets);
+   free(hash->data.d);
+   free(hash);
+}
+
+struct cso_hash_iter cso_hash_find(struct cso_hash *hash,
+                                     unsigned key)
+{
+   struct cso_node **nextNode = cso_hash_find_node(hash, key);
+   struct cso_hash_iter iter = {hash, *nextNode};
+   return iter;
+}
+
+unsigned cso_hash_iter_key(struct cso_hash_iter iter)
+{
+   if (!iter.node || iter.hash->data.e == iter.node)
+      return 0;
+   return iter.node->key;
+}
+
+void * cso_hash_iter_data(struct cso_hash_iter iter)
+{
+   if (!iter.node || iter.hash->data.e == iter.node)
+      return 0;
+   return iter.node->value;
+}
+
+static struct cso_node *cso_hash_data_next(struct cso_node *node)
+{
+   union {
+      struct cso_node *next;
+      struct cso_node *e;
+      struct cso_hash_data *d;
+   } a;
+   a.next = node->next;
+   if (!a.next) {
+      fprintf(stderr, "iterating beyond the last element\n");
+      return 0;
+   }
+   if (a.next->next)
+      return a.next;
+
+   int start = (node->key % a.d->numBuckets) + 1;
+   struct cso_node **bucket = a.d->buckets + start;
+   int n = a.d->numBuckets - start;
+   while (n--) {
+      if (*bucket != a.e)
+         return *bucket;
+      ++bucket;
+   }
+   return a.e;
+}
+
+
+static struct cso_node *cso_hash_data_prev(struct cso_node *node)
+{
+   union {
+      struct cso_node *e;
+      struct cso_hash_data *d;
+   } a;
+
+   a.e = node;
+   while (a.e->next)
+      a.e = a.e->next;
+
+   int start;
+   if (node == a.e)
+      start = a.d->numBuckets - 1;
+   else
+      start = node->key % a.d->numBuckets;
+
+   struct cso_node *sentinel = node;
+   struct cso_node **bucket = a.d->buckets + start;
+   while (start >= 0) {
+      if (*bucket != sentinel) {
+         struct cso_node *prev = *bucket;
+         while (prev->next != sentinel)
+            prev = prev->next;
+         return prev;
+      }
+
+      sentinel = a.e;
+      --bucket;
+      --start;
+   }
+   fprintf(stderr, "iterating backward beyond first element\n");
+   return a.e;
+}
+
+struct cso_hash_iter cso_hash_iter_next(struct cso_hash_iter iter)
+{
+   struct cso_hash_iter next = {iter.hash, cso_hash_data_next(iter.node)};
+   return next;
+}
+
+int cso_hash_iter_is_null(struct cso_hash_iter iter)
+{
+   if (!iter.node || iter.node == iter.hash->data.e)
+      return 1;
+   return 0;
+}
+
+void * cso_hash_take(struct cso_hash *hash,
+                      unsigned akey)
+{
+   struct cso_node **node = cso_hash_find_node(hash, akey);
+   if (*node != hash->data.e) {
+      void *t = (*node)->value;
+      struct cso_node *next = (*node)->next;
+      cso_data_free_node(*node);
+      *node = next;
+      --hash->data.d->size;
+      cso_data_has_shrunk(hash->data.d);
+      return t;
+   }
+   return 0;
+}
+
+struct cso_hash_iter cso_hash_iter_prev(struct cso_hash_iter iter)
+{
+   struct cso_hash_iter prev = {iter.hash,
+                                 cso_hash_data_prev(iter.node)};
+   return prev;
+}
+
+struct cso_hash_iter cso_hash_first_node(struct cso_hash *hash)
+{
+   struct cso_hash_iter iter = {hash, cso_data_first_node(hash->data.d)};
+   return iter;
+}
diff --git a/src/gallium/aux/cso_cache/cso_hash.h b/src/gallium/aux/cso_cache/cso_hash.h
new file mode 100644
index 0000000000..b4aa111860
--- /dev/null
+++ b/src/gallium/aux/cso_cache/cso_hash.h
@@ -0,0 +1,62 @@
+/**************************************************************************
+ *
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ **************************************************************************/
+
+ /*
+  * Authors:
+  *   Zack Rusin <zack@tungstengraphics.com>
+  */
+
+#ifndef CSO_HASH_H
+#define CSO_HASH_H
+
+struct cso_hash;
+struct cso_node;
+
+struct cso_hash_iter {
+   struct cso_hash *hash;
+   struct cso_node  *node;
+};
+
+struct cso_hash *cso_hash_create(void);
+void              cso_hash_delete(struct cso_hash *hash);
+
+struct cso_hash_iter cso_hash_insert(struct cso_hash *hash, unsigned key,
+                                     void *data);
+void  *cso_hash_take(struct cso_hash *hash, unsigned key);
+
+struct cso_hash_iter cso_hash_first_node(struct cso_hash *hash);
+struct cso_hash_iter cso_hash_find(struct cso_hash *hash, unsigned key);
+
+
+int       cso_hash_iter_is_null(struct cso_hash_iter iter);
+unsigned  cso_hash_iter_key(struct cso_hash_iter iter);
+void     *cso_hash_iter_data(struct cso_hash_iter iter);
+
+struct cso_hash_iter cso_hash_iter_next(struct cso_hash_iter iter);
+struct cso_hash_iter cso_hash_iter_prev(struct cso_hash_iter iter);
+
+#endif
diff --git a/src/gallium/aux/draw/Makefile b/src/gallium/aux/draw/Makefile
new file mode 100644
index 0000000000..451911a354
--- /dev/null
+++ b/src/gallium/aux/draw/Makefile
@@ -0,0 +1,2 @@
+default:
+	cd .. ; make
diff --git a/src/gallium/aux/draw/draw_clip.c b/src/gallium/aux/draw/draw_clip.c
new file mode 100644
index 0000000000..e3051507ea
--- /dev/null
+++ b/src/gallium/aux/draw/draw_clip.c
@@ -0,0 +1,488 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+/**
+ * \brief  Clipping stage
+ *
+ * \author  Keith Whitwell <keith@tungstengraphics.com>
+ */
+
+
+#include "pipe/p_util.h"
+#include "pipe/p_shader_tokens.h"
+
+#include "draw_context.h"
+#include "draw_private.h"
+
+
+#ifndef IS_NEGATIVE
+#define IS_NEGATIVE(X) ((X) < 0.0)
+#endif
+
+#ifndef DIFFERENT_SIGNS
+#define DIFFERENT_SIGNS(x, y) ((x) * (y) <= 0.0F && (x) - (y) != 0.0F)
+#endif
+
+#ifndef MAX_CLIPPED_VERTICES
+#define MAX_CLIPPED_VERTICES ((2 * (6 + PIPE_MAX_CLIP_PLANES))+1)
+#endif
+
+
+
+struct clipper {
+   struct draw_stage stage;      /**< base class */
+
+   /* Basically duplicate some of the flatshading logic here:
+    */
+   boolean flat;
+   uint num_color_attribs;
+   uint color_attribs[4];  /* front/back primary/secondary colors */
+
+   float (*plane)[4];
+};
+
+
+/* This is a bit confusing:
+ */
+static INLINE struct clipper *clipper_stage( struct draw_stage *stage )
+{
+   return (struct clipper *)stage;
+}
+
+
+#define LINTERP(T, OUT, IN) ((OUT) + (T) * ((IN) - (OUT)))
+
+
+/* All attributes are float[4], so this is easy:
+ */
+static void interp_attr( float *fdst,
+			 float t,
+			 const float *fin,
+			 const float *fout )
+{  
+   fdst[0] = LINTERP( t, fout[0], fin[0] );
+   fdst[1] = LINTERP( t, fout[1], fin[1] );
+   fdst[2] = LINTERP( t, fout[2], fin[2] );
+   fdst[3] = LINTERP( t, fout[3], fin[3] );
+}
+
+static void copy_colors( struct draw_stage *stage,
+			 struct vertex_header *dst,
+			 const struct vertex_header *src )
+{
+   const struct clipper *clipper = clipper_stage(stage);
+   uint i;
+   for (i = 0; i < clipper->num_color_attribs; i++) {
+      const uint attr = clipper->color_attribs[i];
+      COPY_4FV(dst->data[attr], src->data[attr]);
+   }
+}
+
+
+
+/* Interpolate between two vertices to produce a third.  
+ */
+static void interp( const struct clipper *clip,
+		    struct vertex_header *dst,
+		    float t,
+		    const struct vertex_header *out, 
+		    const struct vertex_header *in )
+{
+   const unsigned nr_attrs = clip->stage.draw->num_vs_outputs;
+   unsigned j;
+
+   /* Vertex header.
+    */
+   {
+      dst->clipmask = 0;
+      dst->edgeflag = 0;
+      dst->pad = 0;
+      dst->vertex_id = UNDEFINED_VERTEX_ID;
+   }
+
+   /* Clip coordinates:  interpolate normally
+    */
+   {
+      interp_attr(dst->clip, t, in->clip, out->clip);
+   }
+
+   /* Do the projective divide and insert window coordinates:
+    */
+   {
+      const float *pos = dst->clip;
+      const float *scale = clip->stage.draw->viewport.scale;
+      const float *trans = clip->stage.draw->viewport.translate;
+      const float oow = 1.0f / pos[3];
+
+      dst->data[0][0] = pos[0] * oow * scale[0] + trans[0];
+      dst->data[0][1] = pos[1] * oow * scale[1] + trans[1];
+      dst->data[0][2] = pos[2] * oow * scale[2] + trans[2];
+      dst->data[0][3] = oow;
+   }
+
+   /* Other attributes
+    * Note: start at 1 to skip winpos (data[0]) since we just computed
+    * it above.
+    */
+   for (j = 1; j < nr_attrs; j++) {
+      interp_attr(dst->data[j], t, in->data[j], out->data[j]);
+   }
+}
+
+
+static void emit_poly( struct draw_stage *stage,
+		       struct vertex_header **inlist,
+		       unsigned n,
+		       const struct prim_header *origPrim)
+{
+   struct prim_header header;
+   unsigned i;
+
+   /* later stages may need the determinant, but only the sign matters */
+   header.det = origPrim->det;
+
+   for (i = 2; i < n; i++) {
+      header.v[0] = inlist[i-1];
+      header.v[1] = inlist[i];
+      header.v[2] = inlist[0];	/* keep in v[2] for flatshading */
+	
+      {
+	 unsigned tmp1 = header.v[1]->edgeflag;
+	 unsigned tmp2 = header.v[2]->edgeflag;
+
+	 if (i != n-1) header.v[1]->edgeflag = 0;
+	 if (i != 2)   header.v[2]->edgeflag = 0;
+
+         header.edgeflags = ((header.v[0]->edgeflag << 0) | 
+                             (header.v[1]->edgeflag << 1) | 
+                             (header.v[2]->edgeflag << 2));
+
+	 stage->next->tri( stage->next, &header );
+
+	 header.v[1]->edgeflag = tmp1;
+	 header.v[2]->edgeflag = tmp2;
+      }
+   }
+}
+
+
+
+
+/* Clip a triangle against the viewport and user clip planes.
+ */
+static void
+do_clip_tri( struct draw_stage *stage, 
+	     struct prim_header *header,
+	     unsigned clipmask )
+{
+   struct clipper *clipper = clipper_stage( stage );
+   struct vertex_header *a[MAX_CLIPPED_VERTICES];
+   struct vertex_header *b[MAX_CLIPPED_VERTICES];
+   struct vertex_header **inlist = a;
+   struct vertex_header **outlist = b;
+   unsigned tmpnr = 0;
+   unsigned n = 3;
+   unsigned i;
+
+   inlist[0] = header->v[0];
+   inlist[1] = header->v[1];
+   inlist[2] = header->v[2];
+
+   while (clipmask && n >= 3) {
+      const unsigned plane_idx = ffs(clipmask)-1;
+      const float *plane = clipper->plane[plane_idx];
+      struct vertex_header *vert_prev = inlist[0];
+      float dp_prev = dot4( vert_prev->clip, plane );
+      unsigned outcount = 0;
+
+      clipmask &= ~(1<<plane_idx);
+
+      inlist[n] = inlist[0]; /* prevent rotation of vertices */
+
+      for (i = 1; i <= n; i++) {
+	 struct vertex_header *vert = inlist[i];
+
+	 float dp = dot4( vert->clip, plane );
+
+	 if (!IS_NEGATIVE(dp_prev)) {
+	    outlist[outcount++] = vert_prev;
+	 }
+
+	 if (DIFFERENT_SIGNS(dp, dp_prev)) {
+	    struct vertex_header *new_vert = clipper->stage.tmp[tmpnr++];
+	    outlist[outcount++] = new_vert;
+
+	    if (IS_NEGATIVE(dp)) {
+	       /* Going out of bounds.  Avoid division by zero as we
+		* know dp != dp_prev from DIFFERENT_SIGNS, above.
+		*/
+	       float t = dp / (dp - dp_prev);
+	       interp( clipper, new_vert, t, vert, vert_prev );
+	       
+	       /* Force edgeflag true in this case:
+		*/
+	       new_vert->edgeflag = 1;
+	    } else {
+	       /* Coming back in.
+		*/
+	       float t = dp_prev / (dp_prev - dp);
+	       interp( clipper, new_vert, t, vert_prev, vert );
+
+	       /* Copy starting vert's edgeflag:
+		*/
+	       new_vert->edgeflag = vert_prev->edgeflag;
+	    }
+	 }
+
+	 vert_prev = vert;
+	 dp_prev = dp;
+      }
+
+      {
+	 struct vertex_header **tmp = inlist;
+	 inlist = outlist;
+	 outlist = tmp;
+	 n = outcount;
+      }
+   }
+
+   /* If flat-shading, copy color to new provoking vertex.
+    */
+   if (clipper->flat && inlist[0] != header->v[2]) {
+      if (1) {
+	 inlist[0] = dup_vert(stage, inlist[0], tmpnr++);
+      }
+
+      copy_colors(stage, inlist[0], header->v[2]);
+   }
+
+
+
+   /* Emit the polygon as triangles to the setup stage:
+    */
+   if (n >= 3)
+      emit_poly( stage, inlist, n, header );
+}
+
+
+/* Clip a line against the viewport and user clip planes.
+ */
+static void
+do_clip_line( struct draw_stage *stage,
+	      struct prim_header *header,
+	      unsigned clipmask )
+{
+   const struct clipper *clipper = clipper_stage( stage );
+   struct vertex_header *v0 = header->v[0];
+   struct vertex_header *v1 = header->v[1];
+   const float *pos0 = v0->clip;
+   const float *pos1 = v1->clip;
+   float t0 = 0.0F;
+   float t1 = 0.0F;
+   struct prim_header newprim;
+
+   while (clipmask) {
+      const unsigned plane_idx = ffs(clipmask)-1;
+      const float *plane = clipper->plane[plane_idx];
+      const float dp0 = dot4( pos0, plane );
+      const float dp1 = dot4( pos1, plane );
+
+      if (dp1 < 0.0F) {
+	 float t = dp1 / (dp1 - dp0);
+         t1 = MAX2(t1, t);
+      } 
+
+      if (dp0 < 0.0F) {
+	 float t = dp0 / (dp0 - dp1);
+         t0 = MAX2(t0, t);
+      }
+
+      if (t0 + t1 >= 1.0F)
+	 return; /* discard */
+
+      clipmask &= ~(1 << plane_idx);  /* turn off this plane's bit */
+   }
+
+   if (v0->clipmask) {
+      interp( clipper, stage->tmp[0], t0, v0, v1 );
+
+      if (clipper->flat)
+	 copy_colors(stage, stage->tmp[0], v0);
+
+      newprim.v[0] = stage->tmp[0];
+   }
+   else {
+      newprim.v[0] = v0;
+   }
+
+   if (v1->clipmask) {
+      interp( clipper, stage->tmp[1], t1, v1, v0 );
+      newprim.v[1] = stage->tmp[1];
+   }
+   else {
+      newprim.v[1] = v1;
+   }
+
+   stage->next->line( stage->next, &newprim );
+}
+
+
+static void
+clip_point( struct draw_stage *stage, 
+	    struct prim_header *header )
+{
+   if (header->v[0]->clipmask == 0) 
+      stage->next->point( stage->next, header );
+}
+
+
+static void
+clip_line( struct draw_stage *stage,
+	   struct prim_header *header )
+{
+   unsigned clipmask = (header->v[0]->clipmask | 
+                        header->v[1]->clipmask);
+   
+   if (clipmask == 0) {
+      /* no clipping needed */
+      stage->next->line( stage->next, header );
+   }
+   else if ((header->v[0]->clipmask &
+             header->v[1]->clipmask) == 0) {
+      do_clip_line(stage, header, clipmask);
+   }
+   /* else, totally clipped */
+}
+
+
+static void
+clip_tri( struct draw_stage *stage,
+	  struct prim_header *header )
+{
+   unsigned clipmask = (header->v[0]->clipmask | 
+                        header->v[1]->clipmask | 
+                        header->v[2]->clipmask);
+   
+   if (clipmask == 0) {
+      /* no clipping needed */
+      stage->next->tri( stage->next, header );
+   }
+   else if ((header->v[0]->clipmask & 
+             header->v[1]->clipmask & 
+             header->v[2]->clipmask) == 0) {
+      do_clip_tri(stage, header, clipmask);
+   }
+}
+
+/* Update state.  Could further delay this until we hit the first
+ * primitive that really requires clipping.
+ */
+static void 
+clip_init_state( struct draw_stage *stage )
+{
+   struct clipper *clipper = clipper_stage( stage );
+
+   clipper->flat = stage->draw->rasterizer->flatshade ? TRUE : FALSE;
+
+   if (clipper->flat) {
+      const struct pipe_shader_state *vs = stage->draw->vertex_shader->state;
+      uint i;
+
+      clipper->num_color_attribs = 0;
+      for (i = 0; i < vs->num_outputs; i++) {
+	 if (vs->output_semantic_name[i] == TGSI_SEMANTIC_COLOR ||
+	     vs->output_semantic_name[i] == TGSI_SEMANTIC_BCOLOR) {
+	    clipper->color_attribs[clipper->num_color_attribs++] = i;
+	 }
+      }
+   }
+   
+   stage->tri = clip_tri;
+   stage->line = clip_line;
+}
+
+
+
+static void clip_first_tri( struct draw_stage *stage,
+			    struct prim_header *header )
+{
+   clip_init_state( stage );
+   stage->tri( stage, header );
+}
+
+static void clip_first_line( struct draw_stage *stage,
+			     struct prim_header *header )
+{
+   clip_init_state( stage );
+   stage->line( stage, header );
+}
+
+
+static void clip_flush( struct draw_stage *stage, 
+			     unsigned flags )
+{
+   stage->tri = clip_first_tri;
+   stage->line = clip_first_line;
+   stage->next->flush( stage->next, flags );
+}
+
+
+static void clip_reset_stipple_counter( struct draw_stage *stage )
+{
+   stage->next->reset_stipple_counter( stage->next );
+}
+
+
+static void clip_destroy( struct draw_stage *stage )
+{
+   draw_free_temp_verts( stage );
+   FREE( stage );
+}
+
+
+/**
+ * Allocate a new clipper stage.
+ * \return pointer to new stage object
+ */
+struct draw_stage *draw_clip_stage( struct draw_context *draw )
+{
+   struct clipper *clipper = CALLOC_STRUCT(clipper);
+
+   draw_alloc_temp_verts( &clipper->stage, MAX_CLIPPED_VERTICES+1 );
+
+   clipper->stage.draw = draw;
+   clipper->stage.point = clip_point;
+   clipper->stage.line = clip_first_line;
+   clipper->stage.tri = clip_first_tri;
+   clipper->stage.flush = clip_flush;
+   clipper->stage.reset_stipple_counter = clip_reset_stipple_counter;
+   clipper->stage.destroy = clip_destroy;
+
+   clipper->plane = draw->plane;
+
+   return &clipper->stage;
+}
diff --git a/src/gallium/aux/draw/draw_context.c b/src/gallium/aux/draw/draw_context.c
new file mode 100644
index 0000000000..4be3830316
--- /dev/null
+++ b/src/gallium/aux/draw/draw_context.c
@@ -0,0 +1,293 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+ /*
+  * Authors:
+  *   Keith Whitwell <keith@tungstengraphics.com>
+  */
+
+
+#include "pipe/p_util.h"
+#include "draw_context.h"
+#include "draw_private.h"
+
+
+
+struct draw_context *draw_create( void )
+{
+   struct draw_context *draw = CALLOC_STRUCT( draw_context );
+
+#if defined(__i386__) || defined(__386__)
+   draw->use_sse = GETENV( "GALLIUM_NOSSE" ) == NULL;
+#else
+   draw->use_sse = FALSE;
+#endif
+
+   /* create pipeline stages */
+   draw->pipeline.wide      = draw_wide_stage( draw );
+   draw->pipeline.stipple   = draw_stipple_stage( draw );
+   draw->pipeline.unfilled  = draw_unfilled_stage( draw );
+   draw->pipeline.twoside   = draw_twoside_stage( draw );
+   draw->pipeline.offset    = draw_offset_stage( draw );
+   draw->pipeline.clip      = draw_clip_stage( draw );
+   draw->pipeline.flatshade = draw_flatshade_stage( draw );
+   draw->pipeline.cull      = draw_cull_stage( draw );
+   draw->pipeline.validate  = draw_validate_stage( draw );
+   draw->pipeline.first     = draw->pipeline.validate;
+
+   ASSIGN_4V( draw->plane[0], -1,  0,  0, 1 );
+   ASSIGN_4V( draw->plane[1],  1,  0,  0, 1 );
+   ASSIGN_4V( draw->plane[2],  0, -1,  0, 1 );
+   ASSIGN_4V( draw->plane[3],  0,  1,  0, 1 );
+   ASSIGN_4V( draw->plane[4],  0,  0,  1, 1 ); /* yes these are correct */
+   ASSIGN_4V( draw->plane[5],  0,  0, -1, 1 ); /* mesa's a bit wonky */
+   draw->nr_planes = 6;
+
+   /* Statically allocate maximum sized vertices for the cache - could be cleverer...
+    */
+   {
+      uint i;
+      const unsigned size = (MAX_VERTEX_SIZE + 0x0f) & ~0x0f;
+      char *tmp = align_malloc(Elements(draw->vcache.vertex) * size, 16);
+
+      for (i = 0; i < Elements(draw->vcache.vertex); i++)
+	 draw->vcache.vertex[i] = (struct vertex_header *)(tmp + i * size);
+   }
+
+   draw->shader_queue_flush = draw_vertex_shader_queue_flush;
+
+   draw->convert_wide_points = TRUE;
+   draw->convert_wide_lines = TRUE;
+
+   draw->reduced_prim = ~0; /* != any of PIPE_PRIM_x */
+
+   draw_vertex_cache_invalidate( draw );
+   draw_set_mapped_element_buffer( draw, 0, NULL );
+
+   return draw;
+}
+
+
+void draw_destroy( struct draw_context *draw )
+{
+   draw->pipeline.wide->destroy( draw->pipeline.wide );
+   draw->pipeline.stipple->destroy( draw->pipeline.stipple );
+   draw->pipeline.unfilled->destroy( draw->pipeline.unfilled );
+   draw->pipeline.twoside->destroy( draw->pipeline.twoside );
+   draw->pipeline.offset->destroy( draw->pipeline.offset );
+   draw->pipeline.clip->destroy( draw->pipeline.clip );
+   draw->pipeline.flatshade->destroy( draw->pipeline.flatshade );
+   draw->pipeline.cull->destroy( draw->pipeline.cull );
+   draw->pipeline.validate->destroy( draw->pipeline.validate );
+   if (draw->pipeline.rasterize)
+      draw->pipeline.rasterize->destroy( draw->pipeline.rasterize );
+   tgsi_exec_machine_free_data(&draw->machine);
+   align_free( draw->vcache.vertex[0] ); /* Frees all the vertices. */
+   FREE( draw );
+}
+
+
+
+void draw_flush( struct draw_context *draw )
+{
+   draw_do_flush( draw, DRAW_FLUSH_BACKEND );
+}
+
+
+
+/**
+ * Register new primitive rasterization/rendering state.
+ * This causes the drawing pipeline to be rebuilt.
+ */
+void draw_set_rasterizer_state( struct draw_context *draw,
+                                const struct pipe_rasterizer_state *raster )
+{
+   draw_do_flush( draw, DRAW_FLUSH_STATE_CHANGE );
+
+   draw->rasterizer = raster;
+}
+
+
+/** 
+ * Plug in the primitive rendering/rasterization stage (which is the last
+ * stage in the drawing pipeline).
+ * This is provided by the device driver.
+ */
+void draw_set_rasterize_stage( struct draw_context *draw,
+                               struct draw_stage *stage )
+{
+   draw_do_flush( draw, DRAW_FLUSH_STATE_CHANGE );
+
+   draw->pipeline.rasterize = stage;
+}
+
+
+/**
+ * Set the draw module's clipping state.
+ */
+void draw_set_clip_state( struct draw_context *draw,
+                          const struct pipe_clip_state *clip )
+{
+   draw_do_flush( draw, DRAW_FLUSH_STATE_CHANGE );
+
+   assert(clip->nr <= PIPE_MAX_CLIP_PLANES);
+   memcpy(&draw->plane[6], clip->ucp, clip->nr * sizeof(clip->ucp[0]));
+   draw->nr_planes = 6 + clip->nr;
+}
+
+
+/**
+ * Set the draw module's viewport state.
+ */
+void draw_set_viewport_state( struct draw_context *draw,
+                              const struct pipe_viewport_state *viewport )
+{
+   draw_do_flush( draw, DRAW_FLUSH_STATE_CHANGE );
+   draw->viewport = *viewport; /* struct copy */
+}
+
+
+
+void
+draw_set_vertex_buffer(struct draw_context *draw,
+                       unsigned attr,
+                       const struct pipe_vertex_buffer *buffer)
+{
+   draw_do_flush( draw, DRAW_FLUSH_VERTEX_CACHE/*STATE_CHANGE*/ );
+   assert(attr < PIPE_ATTRIB_MAX);
+   draw->vertex_buffer[attr] = *buffer;
+}
+
+
+void
+draw_set_vertex_element(struct draw_context *draw,
+                        unsigned attr,
+                        const struct pipe_vertex_element *element)
+{
+   draw_do_flush( draw, DRAW_FLUSH_VERTEX_CACHE/*STATE_CHANGE*/ );
+   assert(attr < PIPE_ATTRIB_MAX);
+   draw->vertex_element[attr] = *element;
+}
+
+
+/**
+ * Tell drawing context where to find mapped vertex buffers.
+ */
+void
+draw_set_mapped_vertex_buffer(struct draw_context *draw,
+                              unsigned attr, const void *buffer)
+{
+   draw_do_flush( draw, DRAW_FLUSH_VERTEX_CACHE/*STATE_CHANGE*/ );
+   draw->user.vbuffer[attr] = buffer;
+}
+
+
+void
+draw_set_mapped_constant_buffer(struct draw_context *draw,
+                                const void *buffer)
+{
+   draw_do_flush( draw, DRAW_FLUSH_VERTEX_CACHE/*STATE_CHANGE*/ );
+   draw->user.constants = buffer;
+}
+
+
+/**
+ * Tells the draw module whether to convert wide points (size != 1)
+ * into triangles.
+ */
+void
+draw_convert_wide_points(struct draw_context *draw, boolean enable)
+{
+   draw_do_flush( draw, DRAW_FLUSH_STATE_CHANGE );
+   draw->convert_wide_points = enable;
+}
+
+
+/**
+ * Tells the draw module whether to convert wide lines (width != 1)
+ * into triangles.
+ */
+void
+draw_convert_wide_lines(struct draw_context *draw, boolean enable)
+{
+   draw_do_flush( draw, DRAW_FLUSH_STATE_CHANGE );
+   draw->convert_wide_lines = enable;
+}
+
+
+/**
+ * Allocate space for temporary post-transform vertices, such as for clipping.
+ */
+void draw_alloc_temp_verts( struct draw_stage *stage, unsigned nr )
+{
+   assert(!stage->tmp);
+
+   stage->nr_tmps = nr;
+
+   if (nr) {
+      ubyte *store = (ubyte *) MALLOC( MAX_VERTEX_SIZE * nr );
+      unsigned i;
+
+      stage->tmp = (struct vertex_header **) MALLOC( sizeof(struct vertex_header *) * nr );
+      
+      for (i = 0; i < nr; i++)
+	 stage->tmp[i] = (struct vertex_header *)(store + i * MAX_VERTEX_SIZE);
+   }
+}
+
+
+void draw_free_temp_verts( struct draw_stage *stage )
+{
+   if (stage->tmp) {
+      FREE( stage->tmp[0] );
+      FREE( stage->tmp );
+      stage->tmp = NULL;
+   }
+}
+
+
+boolean draw_use_sse(struct draw_context *draw)
+{
+   return (boolean) draw->use_sse;
+}
+
+
+void draw_reset_vertex_ids(struct draw_context *draw)
+{
+   struct draw_stage *stage = draw->pipeline.first;
+   
+   while (stage) {
+      unsigned i;
+
+      for (i = 0; i < stage->nr_tmps; i++)
+	 stage->tmp[i]->vertex_id = UNDEFINED_VERTEX_ID;
+
+      stage = stage->next;
+   }
+
+   draw_vertex_cache_reset_vertex_ids(draw);
+}
diff --git a/src/gallium/aux/draw/draw_context.h b/src/gallium/aux/draw/draw_context.h
new file mode 100644
index 0000000000..ddeb184497
--- /dev/null
+++ b/src/gallium/aux/draw/draw_context.h
@@ -0,0 +1,142 @@
+
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+/**
+ * \brief  Public interface into the drawing module.
+ */
+
+/* Authors:  Keith Whitwell <keith@tungstengraphics.com>
+ */
+
+
+#ifndef DRAW_CONTEXT_H
+#define DRAW_CONTEXT_H
+
+
+#include "pipe/p_state.h"
+
+
+struct vertex_buffer;
+struct vertex_info;
+struct draw_context;
+struct draw_stage;
+struct draw_vertex_shader;
+
+
+/**
+ * Clipmask flags
+ */
+/*@{*/
+#define CLIP_RIGHT_BIT   0x01
+#define CLIP_LEFT_BIT    0x02
+#define CLIP_TOP_BIT     0x04
+#define CLIP_BOTTOM_BIT  0x08
+#define CLIP_NEAR_BIT    0x10
+#define CLIP_FAR_BIT     0x20
+/*@}*/
+
+/**
+ * Bitshift for each clip flag
+ */
+/*@{*/
+#define CLIP_RIGHT_SHIFT 	0
+#define CLIP_LEFT_SHIFT 	1
+#define CLIP_TOP_SHIFT  	2
+#define CLIP_BOTTOM_SHIFT       3
+#define CLIP_NEAR_SHIFT  	4
+#define CLIP_FAR_SHIFT  	5
+/*@}*/
+
+
+struct draw_context *draw_create( void );
+
+void draw_destroy( struct draw_context *draw );
+
+void draw_set_viewport_state( struct draw_context *draw,
+                              const struct pipe_viewport_state *viewport );
+
+void draw_set_clip_state( struct draw_context *pipe,
+                          const struct pipe_clip_state *clip );
+
+void draw_set_rasterizer_state( struct draw_context *draw,
+                                const struct pipe_rasterizer_state *raster );
+
+void draw_set_rasterize_stage( struct draw_context *draw,
+                               struct draw_stage *stage );
+
+void draw_convert_wide_points(struct draw_context *draw, boolean enable);
+
+void draw_convert_wide_lines(struct draw_context *draw, boolean enable);
+
+
+struct draw_vertex_shader *
+draw_create_vertex_shader(struct draw_context *draw,
+                          const struct pipe_shader_state *shader);
+void draw_bind_vertex_shader(struct draw_context *draw,
+                             struct draw_vertex_shader *dvs);
+void draw_delete_vertex_shader(struct draw_context *draw,
+                               struct draw_vertex_shader *dvs);
+
+boolean draw_use_sse(struct draw_context *draw);
+
+void draw_set_vertex_buffer(struct draw_context *draw,
+			    unsigned attr,
+			    const struct pipe_vertex_buffer *buffer);
+
+void draw_set_vertex_element(struct draw_context *draw,
+			     unsigned attr,
+			     const struct pipe_vertex_element *element);
+
+void draw_set_mapped_element_buffer( struct draw_context *draw,
+                                     unsigned eltSize, void *elements );
+
+void draw_set_mapped_vertex_buffer(struct draw_context *draw,
+                                   unsigned attr, const void *buffer);
+
+void draw_set_mapped_constant_buffer(struct draw_context *draw,
+                                     const void *buffer);
+
+
+/***********************************************************************
+ * draw_prim.c 
+ */
+
+void draw_arrays(struct draw_context *draw, unsigned prim,
+		 unsigned start, unsigned count);
+
+void draw_flush(struct draw_context *draw);
+
+/***********************************************************************
+ * draw_debug.c 
+ */
+boolean draw_validate_prim( unsigned prim, unsigned length );
+unsigned draw_trim_prim( unsigned mode, unsigned count );
+
+
+
+#endif /* DRAW_CONTEXT_H */
diff --git a/src/gallium/aux/draw/draw_cull.c b/src/gallium/aux/draw/draw_cull.c
new file mode 100644
index 0000000000..8177b0ac86
--- /dev/null
+++ b/src/gallium/aux/draw/draw_cull.c
@@ -0,0 +1,150 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+/**
+ * \brief  Drawing stage for polygon culling
+ */
+
+/* Authors:  Keith Whitwell <keith@tungstengraphics.com>
+ */
+
+
+#include "pipe/p_util.h"
+#include "pipe/p_defines.h"
+#include "draw_private.h"
+
+
+struct cull_stage {
+   struct draw_stage stage;
+   unsigned winding;  /**< which winding(s) to cull (one of PIPE_WINDING_x) */
+};
+
+
+static INLINE struct cull_stage *cull_stage( struct draw_stage *stage )
+{
+   return (struct cull_stage *)stage;
+}
+
+
+
+
+static void cull_tri( struct draw_stage *stage,
+		      struct prim_header *header )
+{
+   /* Window coords: */
+   const float *v0 = header->v[0]->data[0];
+   const float *v1 = header->v[1]->data[0];
+   const float *v2 = header->v[2]->data[0];
+
+   /* edge vectors e = v0 - v2, f = v1 - v2 */
+   const float ex = v0[0] - v2[0];
+   const float ey = v0[1] - v2[1];
+   const float fx = v1[0] - v2[0];
+   const float fy = v1[1] - v2[1];
+   
+   /* det = cross(e,f).z */
+   header->det = ex * fy - ey * fx;
+
+   if (header->det != 0) {
+      /* if (det < 0 then Z points toward camera and triangle is 
+       * counter-clockwise winding.
+       */
+      unsigned winding = (header->det < 0) ? PIPE_WINDING_CCW : PIPE_WINDING_CW;
+
+      if ((winding & cull_stage(stage)->winding) == 0) {
+         /* triangle is not culled, pass to next stage */
+	 stage->next->tri( stage->next, header );
+      }
+   }
+}
+
+static void cull_first_tri( struct draw_stage *stage, 
+			    struct prim_header *header )
+{
+   struct cull_stage *cull = cull_stage(stage);
+
+   cull->winding = stage->draw->rasterizer->cull_mode;
+
+   stage->tri = cull_tri;
+   stage->tri( stage, header );
+}
+
+
+
+static void cull_line( struct draw_stage *stage,
+		       struct prim_header *header )
+{
+   stage->next->line( stage->next, header );
+}
+
+
+static void cull_point( struct draw_stage *stage,
+			struct prim_header *header )
+{
+   stage->next->point( stage->next, header );
+}
+
+
+static void cull_flush( struct draw_stage *stage, unsigned flags )
+{
+   stage->tri = cull_first_tri;
+   stage->next->flush( stage->next, flags );
+}
+
+static void cull_reset_stipple_counter( struct draw_stage *stage )
+{
+   stage->next->reset_stipple_counter( stage->next );
+}
+
+
+static void cull_destroy( struct draw_stage *stage )
+{
+   draw_free_temp_verts( stage );
+   FREE( stage );
+}
+
+
+/**
+ * Create a new polygon culling stage.
+ */
+struct draw_stage *draw_cull_stage( struct draw_context *draw )
+{
+   struct cull_stage *cull = CALLOC_STRUCT(cull_stage);
+
+   draw_alloc_temp_verts( &cull->stage, 0 );
+
+   cull->stage.draw = draw;
+   cull->stage.next = NULL;
+   cull->stage.point = cull_point;
+   cull->stage.line = cull_line;
+   cull->stage.tri = cull_first_tri;
+   cull->stage.flush = cull_flush;
+   cull->stage.reset_stipple_counter = cull_reset_stipple_counter;
+   cull->stage.destroy = cull_destroy;
+
+   return &cull->stage;
+}
diff --git a/src/gallium/aux/draw/draw_debug.c b/src/gallium/aux/draw/draw_debug.c
new file mode 100644
index 0000000000..d6220b5f62
--- /dev/null
+++ b/src/gallium/aux/draw/draw_debug.c
@@ -0,0 +1,113 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+ /*
+  * Authors:
+  *   Keith Whitwell <keith@tungstengraphics.com>
+  */
+
+#include "draw_private.h"
+#include "draw_context.h"
+
+
+
+static void
+draw_prim_info(unsigned prim, unsigned *first, unsigned *incr)
+{
+   assert(prim >= PIPE_PRIM_POINTS);
+   assert(prim <= PIPE_PRIM_POLYGON);
+
+   switch (prim) {
+   case PIPE_PRIM_POINTS:
+      *first = 1;
+      *incr = 1;
+      break;
+   case PIPE_PRIM_LINES:
+      *first = 2;
+      *incr = 2;
+      break;
+   case PIPE_PRIM_LINE_STRIP:
+      *first = 2;
+      *incr = 1;
+      break;
+   case PIPE_PRIM_LINE_LOOP:
+      *first = 2;
+      *incr = 1;
+      break;
+   case PIPE_PRIM_TRIANGLES:
+      *first = 3;
+      *incr = 3;
+      break;
+   case PIPE_PRIM_TRIANGLE_STRIP:
+      *first = 3;
+      *incr = 1;
+      break;
+   case PIPE_PRIM_TRIANGLE_FAN:
+   case PIPE_PRIM_POLYGON:
+      *first = 3;
+      *incr = 1;
+      break;
+   case PIPE_PRIM_QUADS:
+      *first = 4;
+      *incr = 4;
+      break;
+   case PIPE_PRIM_QUAD_STRIP:
+      *first = 4;
+      *incr = 2;
+      break;
+   default:
+      assert(0);
+      *first = 1;
+      *incr = 1;
+      break;
+   }
+}
+
+
+unsigned 
+draw_trim_prim( unsigned mode, unsigned count )
+{
+   unsigned length, first, incr;
+
+   draw_prim_info( mode, &first, &incr );
+
+   if (count < first)
+      length = 0;
+   else
+      length = count - (count - first) % incr; 
+
+   return length;
+}
+
+
+boolean
+draw_validate_prim( unsigned mode, unsigned count )
+{
+   return (count > 0 &&
+           count == draw_trim_prim( mode, count ));
+}
+
diff --git a/src/gallium/aux/draw/draw_flatshade.c b/src/gallium/aux/draw/draw_flatshade.c
new file mode 100644
index 0000000000..4398abbc60
--- /dev/null
+++ b/src/gallium/aux/draw/draw_flatshade.c
@@ -0,0 +1,205 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+/* Authors:  Keith Whitwell <keith@tungstengraphics.com>
+ */
+
+#include "pipe/p_util.h"
+#include "pipe/p_shader_tokens.h"
+#include "draw_private.h"
+
+
+/** subclass of draw_stage */
+struct flat_stage
+{
+   struct draw_stage stage;
+
+   uint num_color_attribs;
+   uint color_attribs[4];  /* front/back primary/secondary colors */
+};
+
+
+static INLINE struct flat_stage *
+flat_stage(struct draw_stage *stage)
+{
+   return (struct flat_stage *) stage;
+}
+
+
+/** Copy all the color attributes from 'src' vertex to 'dst' vertex */
+static INLINE void copy_colors( struct draw_stage *stage,
+                                struct vertex_header *dst,
+                                const struct vertex_header *src )
+{
+   const struct flat_stage *flat = flat_stage(stage);
+   uint i;
+   for (i = 0; i < flat->num_color_attribs; i++) {
+      const uint attr = flat->color_attribs[i];
+      COPY_4FV(dst->data[attr], src->data[attr]);
+   }
+}
+
+
+/** Copy all the color attributes from src vertex to dst0 & dst1 vertices */
+static INLINE void copy_colors2( struct draw_stage *stage,
+                                 struct vertex_header *dst0,
+                                 struct vertex_header *dst1,
+                                 const struct vertex_header *src )
+{
+   const struct flat_stage *flat = flat_stage(stage);
+   uint i;
+   for (i = 0; i < flat->num_color_attribs; i++) {
+      const uint attr = flat->color_attribs[i];
+      COPY_4FV(dst0->data[attr], src->data[attr]);
+      COPY_4FV(dst1->data[attr], src->data[attr]);
+   }
+}
+
+
+/**
+ * Flatshade tri.  Required for clipping and when unfilled tris are
+ * active, otherwise handled by hardware.
+ */
+static void flatshade_tri( struct draw_stage *stage,
+			   struct prim_header *header )
+{
+   struct prim_header tmp;
+
+   tmp.det = header->det;
+   tmp.edgeflags = header->edgeflags;
+   tmp.v[0] = dup_vert(stage, header->v[0], 0);
+   tmp.v[1] = dup_vert(stage, header->v[1], 1);
+   tmp.v[2] = header->v[2];
+
+   copy_colors2(stage, tmp.v[0], tmp.v[1], tmp.v[2]);
+   
+   stage->next->tri( stage->next, &tmp );
+}
+
+
+/**
+ * Flatshade line.  Required for clipping.
+ */
+static void flatshade_line( struct draw_stage *stage,
+			    struct prim_header *header )
+{
+   struct prim_header tmp;
+
+   tmp.v[0] = dup_vert(stage, header->v[0], 0);
+   tmp.v[1] = header->v[1];
+
+   copy_colors(stage, tmp.v[0], tmp.v[1]);
+   
+   stage->next->line( stage->next, &tmp );
+}
+
+
+static void flatshade_point( struct draw_stage *stage,
+                             struct prim_header *header )
+{
+   stage->next->point( stage->next, header );
+}
+
+
+static void flatshade_init_state( struct draw_stage *stage )
+{
+   struct flat_stage *flat = flat_stage(stage);
+   const struct pipe_shader_state *vs = stage->draw->vertex_shader->state;
+   uint i;
+
+   /* Find which vertex shader outputs are colors, make a list */
+   flat->num_color_attribs = 0;
+   for (i = 0; i < vs->num_outputs; i++) {
+      if (vs->output_semantic_name[i] == TGSI_SEMANTIC_COLOR ||
+          vs->output_semantic_name[i] == TGSI_SEMANTIC_BCOLOR) {
+         flat->color_attribs[flat->num_color_attribs++] = i;
+      }
+   }
+
+   stage->line = flatshade_line;
+   stage->tri = flatshade_tri;
+}
+
+static void flatshade_first_tri( struct draw_stage *stage,
+				 struct prim_header *header )
+{
+   flatshade_init_state( stage );
+   stage->tri( stage, header );
+}
+
+static void flatshade_first_line( struct draw_stage *stage,
+				  struct prim_header *header )
+{
+   flatshade_init_state( stage );
+   stage->line( stage, header );
+}
+
+
+static void flatshade_flush( struct draw_stage *stage, 
+			     unsigned flags )
+{
+   stage->tri = flatshade_first_tri;
+   stage->line = flatshade_first_line;
+   stage->next->flush( stage->next, flags );
+}
+
+
+static void flatshade_reset_stipple_counter( struct draw_stage *stage )
+{
+   stage->next->reset_stipple_counter( stage->next );
+}
+
+
+static void flatshade_destroy( struct draw_stage *stage )
+{
+   draw_free_temp_verts( stage );
+   FREE( stage );
+}
+
+
+/**
+ * Create flatshading drawing stage.
+ */
+struct draw_stage *draw_flatshade_stage( struct draw_context *draw )
+{
+   struct flat_stage *flatshade = CALLOC_STRUCT(flat_stage);
+
+   draw_alloc_temp_verts( &flatshade->stage, 2 );
+
+   flatshade->stage.draw = draw;
+   flatshade->stage.next = NULL;
+   flatshade->stage.point = flatshade_point;
+   flatshade->stage.line = flatshade_first_line;
+   flatshade->stage.tri = flatshade_first_tri;
+   flatshade->stage.flush = flatshade_flush;
+   flatshade->stage.reset_stipple_counter = flatshade_reset_stipple_counter;
+   flatshade->stage.destroy = flatshade_destroy;
+
+   return &flatshade->stage;
+}
+
+
diff --git a/src/gallium/aux/draw/draw_offset.c b/src/gallium/aux/draw/draw_offset.c
new file mode 100644
index 0000000000..dbc676deae
--- /dev/null
+++ b/src/gallium/aux/draw/draw_offset.c
@@ -0,0 +1,186 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+/**
+ * \brief  polygon offset state
+ *
+ * \author  Keith Whitwell <keith@tungstengraphics.com>
+ * \author  Brian Paul
+ */
+
+#include "pipe/p_util.h"
+#include "draw_private.h"
+
+
+
+struct offset_stage {
+   struct draw_stage stage;
+
+   float scale;
+   float units;
+};
+
+
+
+static INLINE struct offset_stage *offset_stage( struct draw_stage *stage )
+{
+   return (struct offset_stage *) stage;
+}
+
+
+
+
+
+/**
+ * Offset tri Z.  Some hardware can handle this, but not usually when
+ * doing unfilled rendering.
+ */
+static void do_offset_tri( struct draw_stage *stage,
+			   struct prim_header *header )
+{
+   struct offset_stage *offset = offset_stage(stage);   
+   float inv_det = 1.0f / header->det;
+
+   /* Window coords:
+    */
+   float *v0 = header->v[0]->data[0];
+   float *v1 = header->v[1]->data[0];
+   float *v2 = header->v[2]->data[0];
+
+   /* edge vectors e = v0 - v2, f = v1 - v2 */
+   float ex = v0[0] - v2[0];
+   float ey = v0[1] - v2[1];
+   float ez = v0[2] - v2[2];
+   float fx = v1[0] - v2[0];
+   float fy = v1[1] - v2[1];
+   float fz = v1[2] - v2[2];
+
+   /* (a,b) = cross(e,f).xy */
+   float a = ey*fz - ez*fy;
+   float b = ez*fx - ex*fz;
+
+   float dzdx = FABSF(a * inv_det);
+   float dzdy = FABSF(b * inv_det);
+
+   float zoffset = offset->units + MAX2(dzdx, dzdy) * offset->scale;
+
+   /*
+    * Note: we're applying the offset and clamping per-vertex.
+    * Ideally, the offset is applied per-fragment prior to fragment shading.
+    */
+   v0[2] = CLAMP(v0[2] + zoffset, 0.0f, 1.0f);
+   v1[2] = CLAMP(v1[2] + zoffset, 0.0f, 1.0f);
+   v2[2] = CLAMP(v2[2] + zoffset, 0.0f, 1.0f);
+
+   stage->next->tri( stage->next, header );
+}
+
+
+static void offset_tri( struct draw_stage *stage,
+			struct prim_header *header )
+{
+   struct prim_header tmp;
+
+   tmp.det = header->det;
+   tmp.edgeflags = header->edgeflags;
+   tmp.v[0] = dup_vert(stage, header->v[0], 0);
+   tmp.v[1] = dup_vert(stage, header->v[1], 1);
+   tmp.v[2] = dup_vert(stage, header->v[2], 2);
+
+   do_offset_tri( stage, &tmp );
+}
+
+
+static void offset_first_tri( struct draw_stage *stage, 
+			      struct prim_header *header )
+{
+   struct offset_stage *offset = offset_stage(stage);
+   float mrd = 1.0f / 65535.0f; /* XXX this depends on depthbuffer bits! */
+
+   offset->units = stage->draw->rasterizer->offset_units * mrd;
+   offset->scale = stage->draw->rasterizer->offset_scale;
+
+   stage->tri = offset_tri;
+   stage->tri( stage, header );
+}
+
+
+static void offset_line( struct draw_stage *stage,
+			 struct prim_header *header )
+{
+   stage->next->line( stage->next, header );
+}
+
+
+static void offset_point( struct draw_stage *stage,
+			  struct prim_header *header )
+{
+   stage->next->point( stage->next, header );
+}
+
+
+static void offset_flush( struct draw_stage *stage,
+			  unsigned flags )
+{
+   stage->tri = offset_first_tri;
+   stage->next->flush( stage->next, flags );
+}
+
+
+static void offset_reset_stipple_counter( struct draw_stage *stage )
+{
+   stage->next->reset_stipple_counter( stage->next );
+}
+
+
+static void offset_destroy( struct draw_stage *stage )
+{
+   draw_free_temp_verts( stage );
+   FREE( stage );
+}
+
+
+/**
+ * Create polygon offset drawing stage.
+ */
+struct draw_stage *draw_offset_stage( struct draw_context *draw )
+{
+   struct offset_stage *offset = CALLOC_STRUCT(offset_stage);
+
+   draw_alloc_temp_verts( &offset->stage, 3 );
+
+   offset->stage.draw = draw;
+   offset->stage.next = NULL;
+   offset->stage.point = offset_point;
+   offset->stage.line = offset_line;
+   offset->stage.tri = offset_first_tri;
+   offset->stage.flush = offset_flush;
+   offset->stage.reset_stipple_counter = offset_reset_stipple_counter;
+   offset->stage.destroy = offset_destroy;
+
+   return &offset->stage;
+}
diff --git a/src/gallium/aux/draw/draw_prim.c b/src/gallium/aux/draw/draw_prim.c
new file mode 100644
index 0000000000..51e2242719
--- /dev/null
+++ b/src/gallium/aux/draw/draw_prim.c
@@ -0,0 +1,482 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+ /*
+  * Authors:
+  *   Keith Whitwell <keith@tungstengraphics.com>
+  */
+
+#include "pipe/p_debug.h"
+
+#include "draw_private.h"
+#include "draw_context.h"
+
+
+
+#define RP_NONE  0
+#define RP_POINT 1
+#define RP_LINE  2
+#define RP_TRI   3
+
+
+static unsigned reduced_prim[PIPE_PRIM_POLYGON + 1] = {
+   RP_POINT,
+   RP_LINE,
+   RP_LINE,
+   RP_LINE,
+   RP_TRI,
+   RP_TRI,
+   RP_TRI,
+   RP_TRI,
+   RP_TRI,
+   RP_TRI
+};
+
+
+static void draw_prim_queue_flush( struct draw_context *draw )
+{
+   unsigned i;
+
+   if (0)
+      debug_printf("Flushing with %d prims, %d verts\n",
+                   draw->pq.queue_nr, draw->vs.queue_nr);
+
+   assert (draw->pq.queue_nr != 0);
+
+   /* NOTE: we cannot save draw->pipeline->first in a local var because
+    * draw->pipeline->first is often changed by the first call to tri(),
+    * line(), etc.
+    */
+   if (draw->rasterizer->line_stipple_enable) {
+      switch (draw->reduced_prim) {
+      case RP_TRI:
+	 for (i = 0; i < draw->pq.queue_nr; i++) {
+	    if (draw->pq.queue[i].reset_line_stipple)
+	       draw->pipeline.first->reset_stipple_counter( draw->pipeline.first );
+	    
+	    draw->pipeline.first->tri( draw->pipeline.first, &draw->pq.queue[i] );
+	 }
+	 break;
+      case RP_LINE:
+	 for (i = 0; i < draw->pq.queue_nr; i++) {
+	    if (draw->pq.queue[i].reset_line_stipple)
+	       draw->pipeline.first->reset_stipple_counter( draw->pipeline.first );
+	    
+	    draw->pipeline.first->line( draw->pipeline.first, &draw->pq.queue[i] );
+	 }
+	 break;
+      case RP_POINT:
+	 draw->pipeline.first->reset_stipple_counter( draw->pipeline.first );
+	 for (i = 0; i < draw->pq.queue_nr; i++)
+	    draw->pipeline.first->point( draw->pipeline.first, &draw->pq.queue[i] );
+	 break;
+      }
+   }
+   else {
+      switch (draw->reduced_prim) {
+      case RP_TRI:
+	 for (i = 0; i < draw->pq.queue_nr; i++) 
+	    draw->pipeline.first->tri( draw->pipeline.first, &draw->pq.queue[i] );
+	 break;
+      case RP_LINE:
+	 for (i = 0; i < draw->pq.queue_nr; i++) 
+	    draw->pipeline.first->line( draw->pipeline.first, &draw->pq.queue[i] );
+	 break;
+      case RP_POINT:
+	 for (i = 0; i < draw->pq.queue_nr; i++)
+	    draw->pipeline.first->point( draw->pipeline.first, &draw->pq.queue[i] );
+	 break;
+      }
+   }
+
+   draw->pq.queue_nr = 0;   
+   draw_vertex_cache_unreference( draw );
+}
+
+
+
+void draw_do_flush( struct draw_context *draw, unsigned flags )
+{
+   if (0)
+      debug_printf("Flushing with %d verts, %d prims\n",
+                   draw->vs.queue_nr,
+                   draw->pq.queue_nr );
+
+
+   if (flags >= DRAW_FLUSH_SHADER_QUEUE) {
+      if (draw->vs.queue_nr)
+         (*draw->shader_queue_flush)(draw);
+
+      if (flags >= DRAW_FLUSH_PRIM_QUEUE) {
+	 if (draw->pq.queue_nr)
+	    draw_prim_queue_flush(draw);
+
+	 if (flags >= DRAW_FLUSH_VERTEX_CACHE) {
+            draw_vertex_cache_invalidate(draw);
+
+	    if (flags >= DRAW_FLUSH_STATE_CHANGE) {
+               draw->pipeline.first->flush( draw->pipeline.first, flags );
+               draw->pipeline.first = draw->pipeline.validate;
+               draw->reduced_prim = ~0;
+	    }
+	 }
+      }    
+   }
+}
+
+
+
+/* Return a pointer to a freshly queued primitive header.  Ensure that
+ * there is room in the vertex cache for a maximum of "nr_verts" new
+ * vertices.  Flush primitive and/or vertex queues if necessary to
+ * make space.
+ */
+static struct prim_header *get_queued_prim( struct draw_context *draw,
+					    unsigned nr_verts )
+{
+   if (!draw_vertex_cache_check_space( draw, nr_verts )) {
+//      debug_printf("v");
+      draw_do_flush( draw, DRAW_FLUSH_VERTEX_CACHE );
+   }
+   else if (draw->pq.queue_nr == PRIM_QUEUE_LENGTH) {
+//      debug_printf("p");
+      draw_do_flush( draw, DRAW_FLUSH_PRIM_QUEUE );
+   }
+
+   assert(draw->pq.queue_nr < PRIM_QUEUE_LENGTH);
+
+   return &draw->pq.queue[draw->pq.queue_nr++];
+}
+
+
+
+/**
+ * Add a point to the primitive queue.
+ * \param i0  index into user's vertex arrays
+ */
+static void do_point( struct draw_context *draw,
+		      unsigned i0 )
+{
+   struct prim_header *prim = get_queued_prim( draw, 1 );
+   
+   prim->reset_line_stipple = 0;
+   prim->edgeflags = 1;
+   prim->pad = 0;
+   prim->v[0] = draw->vcache.get_vertex( draw, i0 );
+}
+
+
+/**
+ * Add a line to the primitive queue.
+ * \param i0  index into user's vertex arrays
+ * \param i1  index into user's vertex arrays
+ */
+static void do_line( struct draw_context *draw,
+		     boolean reset_stipple,
+		     unsigned i0,
+		     unsigned i1 )
+{
+   struct prim_header *prim = get_queued_prim( draw, 2 );
+   
+   prim->reset_line_stipple = reset_stipple;
+   prim->edgeflags = 1;
+   prim->pad = 0;
+   prim->v[0] = draw->vcache.get_vertex( draw, i0 );
+   prim->v[1] = draw->vcache.get_vertex( draw, i1 );
+}
+
+/**
+ * Add a triangle to the primitive queue.
+ */
+static void do_triangle( struct draw_context *draw,
+			 unsigned i0,
+			 unsigned i1,
+			 unsigned i2 )
+{
+   struct prim_header *prim = get_queued_prim( draw, 3 );
+   
+   prim->reset_line_stipple = 1;
+   prim->edgeflags = ~0;
+   prim->pad = 0;
+   prim->v[0] = draw->vcache.get_vertex( draw, i0 );
+   prim->v[1] = draw->vcache.get_vertex( draw, i1 );
+   prim->v[2] = draw->vcache.get_vertex( draw, i2 );
+}
+			  
+static void do_ef_triangle( struct draw_context *draw,
+			    boolean reset_stipple,
+			    unsigned ef_mask,
+			    unsigned i0,
+			    unsigned i1,
+			    unsigned i2 )
+{
+   struct prim_header *prim = get_queued_prim( draw, 3 );
+   struct vertex_header *v0 = draw->vcache.get_vertex( draw, i0 );
+   struct vertex_header *v1 = draw->vcache.get_vertex( draw, i1 );
+   struct vertex_header *v2 = draw->vcache.get_vertex( draw, i2 );
+
+   prim->reset_line_stipple = reset_stipple;
+
+   prim->edgeflags = ef_mask & ((v0->edgeflag << 0) | 
+				(v1->edgeflag << 1) | 
+				(v2->edgeflag << 2));
+   prim->pad = 0;
+   prim->v[0] = v0;
+   prim->v[1] = v1;
+   prim->v[2] = v2;
+}
+
+
+static void do_ef_quad( struct draw_context *draw,
+		     unsigned v0,
+		     unsigned v1,
+		     unsigned v2,
+		     unsigned v3 )
+{
+   const unsigned omitEdge2 = ~(1 << 1);
+   const unsigned omitEdge3 = ~(1 << 2);
+   do_ef_triangle( draw, 1, omitEdge2, v0, v1, v3 );
+   do_ef_triangle( draw, 0, omitEdge3, v1, v2, v3 );
+}
+
+static void do_quad( struct draw_context *draw,
+		     unsigned v0,
+		     unsigned v1,
+		     unsigned v2,
+		     unsigned v3 )
+{
+   do_triangle( draw, v0, v1, v3 );
+   do_triangle( draw, v1, v2, v3 );
+}
+
+
+/**
+ * Main entrypoint to draw some number of points/lines/triangles
+ */
+static void
+draw_prim( struct draw_context *draw, 
+	   unsigned prim, unsigned start, unsigned count )
+{
+   unsigned i;
+   boolean unfilled = (draw->rasterizer->fill_cw != PIPE_POLYGON_MODE_FILL ||
+		       draw->rasterizer->fill_ccw != PIPE_POLYGON_MODE_FILL);
+
+//   debug_printf("%s (%d) %d/%d\n", __FUNCTION__, draw->prim, start, count );
+
+   switch (prim) {
+   case PIPE_PRIM_POINTS:
+      for (i = 0; i < count; i ++) {
+	 do_point( draw,
+		   start + i );
+      }
+      break;
+
+   case PIPE_PRIM_LINES:
+      for (i = 0; i+1 < count; i += 2) {
+	 do_line( draw, 
+		  TRUE,
+		  start + i + 0,
+		  start + i + 1);
+      }
+      break;
+
+   case PIPE_PRIM_LINE_LOOP:  
+      if (count >= 2) {
+	 for (i = 1; i < count; i++) {
+	    do_line( draw, 
+		     i == 1, 	/* XXX: only if vb not split */
+		     start + i - 1,
+		     start + i );
+	 }
+
+	 do_line( draw, 
+		  0,
+		  start + count - 1,
+		  start + 0 );
+      }
+      break;
+
+   case PIPE_PRIM_LINE_STRIP:
+      for (i = 1; i < count; i++) {
+	 do_line( draw,
+		  i == 1,
+		  start + i - 1,
+		  start + i );
+      }
+      break;
+
+   case PIPE_PRIM_TRIANGLES:
+      if (unfilled) {
+	 for (i = 0; i+2 < count; i += 3) {
+	    do_ef_triangle( draw,
+			    1, 
+			    ~0,
+			    start + i + 0,
+			    start + i + 1,
+			    start + i + 2 );
+	 }
+      } 
+      else {
+	 for (i = 0; i+2 < count; i += 3) {
+	    do_triangle( draw,
+			 start + i + 0,
+			 start + i + 1,
+			 start + i + 2 );
+	 }
+      }
+      break;
+
+   case PIPE_PRIM_TRIANGLE_STRIP:
+      for (i = 0; i+2 < count; i++) {
+	 if (i & 1) {
+	    do_triangle( draw,
+			 start + i + 1,
+			 start + i + 0,
+			 start + i + 2 );
+	 }
+	 else {
+	    do_triangle( draw,
+			 start + i + 0,
+			 start + i + 1,
+			 start + i + 2 );
+	 }
+      }
+      break;
+
+   case PIPE_PRIM_TRIANGLE_FAN:
+      if (count >= 3) {
+	 for (i = 0; i+2 < count; i++) {
+	    do_triangle( draw,
+			 start + 0,
+			 start + i + 1,
+			 start + i + 2 );
+	 }
+      }
+      break;
+
+
+   case PIPE_PRIM_QUADS:
+      if (unfilled) {
+	 for (i = 0; i+3 < count; i += 4) {
+	    do_ef_quad( draw,
+			start + i + 0,
+			start + i + 1,
+			start + i + 2,
+			start + i + 3);
+	 }
+      }
+      else {
+	 for (i = 0; i+3 < count; i += 4) {
+	    do_quad( draw,
+		     start + i + 0,
+		     start + i + 1,
+		     start + i + 2,
+		     start + i + 3);
+	 }
+      }
+      break;
+
+   case PIPE_PRIM_QUAD_STRIP:
+      if (unfilled) {
+	 for (i = 0; i+3 < count; i += 2) {
+	    do_ef_quad( draw,
+			start + i + 2,
+			start + i + 0,
+			start + i + 1,
+			start + i + 3);
+	 }
+      }
+      else {
+	 for (i = 0; i+3 < count; i += 2) {
+	    do_quad( draw,
+		     start + i + 2,
+		     start + i + 0,
+		     start + i + 1,
+		     start + i + 3);
+	 }
+      }
+      break;
+
+   case PIPE_PRIM_POLYGON:
+      if (unfilled) {
+	 unsigned ef_mask = (1<<2) | (1<<0);
+
+	 for (i = 0; i+2 < count; i++) {
+
+            if (i + 3 >= count)
+	       ef_mask |= (1<<1);
+
+	    do_ef_triangle( draw,
+			    i == 0,
+			    ef_mask,
+			    start + i + 1,
+			    start + i + 2,
+			    start + 0);
+
+	    ef_mask &= ~(1<<2);
+	 }
+      }
+      else {
+	 for (i = 0; i+2 < count; i++) {
+	    do_triangle( draw,
+			 start + i + 1,
+			 start + i + 2,
+			 start + 0);
+	 }
+      }
+      break;
+
+   default:
+      assert(0);
+      break;
+   }
+}
+
+
+
+
+/**
+ * Draw vertex arrays
+ * This is the main entrypoint into the drawing module.
+ * \param prim  one of PIPE_PRIM_x
+ * \param start  index of first vertex to draw
+ * \param count  number of vertices to draw
+ */
+void
+draw_arrays(struct draw_context *draw, unsigned prim,
+            unsigned start, unsigned count)
+{
+   if (reduced_prim[prim] != draw->reduced_prim) {
+      draw_do_flush( draw, DRAW_FLUSH_STATE_CHANGE );
+      draw->reduced_prim = reduced_prim[prim];
+   }
+
+   /* drawing done here: */
+   draw_prim(draw, prim, start, count);
+}
+
+
diff --git a/src/gallium/aux/draw/draw_private.h b/src/gallium/aux/draw/draw_private.h
new file mode 100644
index 0000000000..b17eaaed65
--- /dev/null
+++ b/src/gallium/aux/draw/draw_private.h
@@ -0,0 +1,346 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+/**
+ * Private data structures, etc for the draw module.
+ */
+
+
+/**
+ * Authors:
+ * Keith Whitwell <keith@tungstengraphics.com>
+ * Brian Paul
+ */
+
+
+#ifndef DRAW_PRIVATE_H
+#define DRAW_PRIVATE_H
+
+
+#include "pipe/p_state.h"
+#include "pipe/p_defines.h"
+
+#include "x86/rtasm/x86sse.h"
+#include "pipe/tgsi/exec/tgsi_exec.h"
+
+
+struct gallivm_prog;
+struct gallivm_cpu_engine;
+
+/**
+ * Basic vertex info.
+ * Carry some useful information around with the vertices in the prim pipe.  
+ */
+struct vertex_header {
+   unsigned clipmask:12;
+   unsigned edgeflag:1;
+   unsigned pad:3;
+   unsigned vertex_id:16;
+
+   float clip[4];
+
+   float data[][4];		/* Note variable size */
+};
+
+/* NOTE: It should match vertex_id size above */
+#define UNDEFINED_VERTEX_ID 0xffff
+
+/* XXX This is too large */
+#define MAX_VERTEX_SIZE ((2 + PIPE_MAX_SHADER_OUTPUTS) * 4 * sizeof(float))
+
+
+
+/**
+ * Basic info for a point/line/triangle primitive.
+ */
+struct prim_header {
+   float det;                 /**< front/back face determinant */
+   unsigned reset_line_stipple:1;
+   unsigned edgeflags:3;
+   unsigned pad:28;
+   struct vertex_header *v[3];  /**< 1 to 3 vertex pointers */
+};
+
+
+
+struct draw_context;
+
+/**
+ * Base class for all primitive drawing stages.
+ */
+struct draw_stage
+{
+   struct draw_context *draw;   /**< parent context */
+
+   struct draw_stage *next;     /**< next stage in pipeline */
+
+   struct vertex_header **tmp;  /**< temp vert storage, such as for clipping */
+   unsigned nr_tmps;
+
+   void (*point)( struct draw_stage *,
+		  struct prim_header * );
+
+   void (*line)( struct draw_stage *,
+		 struct prim_header * );
+
+   void (*tri)( struct draw_stage *,
+		struct prim_header * );
+
+   void (*flush)( struct draw_stage *,
+		  unsigned flags );
+
+   void (*reset_stipple_counter)( struct draw_stage * );
+
+   void (*destroy)( struct draw_stage * );
+};
+
+
+#define PRIM_QUEUE_LENGTH      16
+#define VCACHE_SIZE            32
+#define VCACHE_OVERFLOW        4
+#define VS_QUEUE_LENGTH        (VCACHE_SIZE + VCACHE_OVERFLOW + 1)	/* can never fill up */
+
+/**
+ * Private version of the compiled vertex_shader
+ */
+struct draw_vertex_shader {
+   const struct pipe_shader_state   *state;
+#if defined(__i386__) || defined(__386__)
+   struct x86_function              sse2_program;
+#endif
+#ifdef MESA_LLVM
+   struct gallivm_prog *llvm_prog;
+#endif
+};
+
+
+/* Internal function for vertex fetch.
+ */
+typedef void (*fetch_func)(const void *ptr, float *attrib);
+typedef void (*full_fetch_func)( struct draw_context *draw,
+				 struct tgsi_exec_machine *machine,
+				 const unsigned *elts,
+				 unsigned count );
+
+
+
+/**
+ * Private context for the drawing module.
+ */
+struct draw_context
+{
+   /** Drawing/primitive pipeline stages */
+   struct {
+      struct draw_stage *first;  /**< one of the following */
+
+      struct draw_stage *validate; 
+
+      /* stages (in logical order) */
+      struct draw_stage *flatshade;
+      struct draw_stage *clip;
+      struct draw_stage *cull;
+      struct draw_stage *twoside;
+      struct draw_stage *offset;
+      struct draw_stage *unfilled;
+      struct draw_stage *stipple;
+      struct draw_stage *wide;
+      struct draw_stage *rasterize;
+   } pipeline;
+
+   /* pipe state that we need: */
+   const struct pipe_rasterizer_state *rasterizer;
+   struct pipe_viewport_state viewport;
+   struct pipe_vertex_buffer vertex_buffer[PIPE_ATTRIB_MAX];
+   struct pipe_vertex_element vertex_element[PIPE_ATTRIB_MAX];
+   const struct draw_vertex_shader *vertex_shader;
+
+   uint num_vs_outputs;  /**< convenience, from vertex_shader */
+
+   /* user-space vertex data, buffers */
+   struct {
+      /** vertex element/index buffer (ex: glDrawElements) */
+      const void *elts;
+      /** bytes per index (0, 1, 2 or 4) */
+      unsigned eltSize;
+
+      /** vertex arrays */
+      const void *vbuffer[PIPE_ATTRIB_MAX];
+
+      /** constant buffer (for vertex shader) */
+      const void *constants;
+   } user;
+
+   /* Clip derived state:
+    */
+   float plane[12][4];
+   unsigned nr_planes;
+
+   boolean convert_wide_points; /**< convert wide points to tris? */
+   boolean convert_wide_lines;  /**< convert side lines to tris? */
+
+   unsigned reduced_prim;
+
+   /** TGSI program interpreter runtime state */
+   struct tgsi_exec_machine machine;
+
+   /* Vertex fetch internal state
+    */
+   struct {
+      const ubyte *src_ptr[PIPE_ATTRIB_MAX];
+      unsigned pitch[PIPE_ATTRIB_MAX];
+      fetch_func fetch[PIPE_ATTRIB_MAX];
+      unsigned nr_attrs;
+      full_fetch_func fetch_func;
+   } vertex_fetch;
+
+   /* Post-tnl vertex cache:
+    */
+   struct {
+      unsigned referenced;  /**< bitfield */
+      unsigned idx[VCACHE_SIZE + VCACHE_OVERFLOW];
+      struct vertex_header *vertex[VCACHE_SIZE + VCACHE_OVERFLOW];
+      unsigned overflow;
+
+      /** To find space in the vertex cache: */
+      struct vertex_header *(*get_vertex)( struct draw_context *draw,
+                                           unsigned i );
+   } vcache;
+
+   /* Vertex shader queue:
+    */
+   struct {
+      struct {
+	 unsigned elt;   /**< index into the user's vertex arrays */
+	 struct vertex_header *dest; /**< points into vcache.vertex[] array */
+      } queue[VS_QUEUE_LENGTH];
+      unsigned queue_nr;
+   } vs;
+
+   /**
+    * Run the vertex shader on all vertices in the vertex queue.
+    */
+   void (*shader_queue_flush)(struct draw_context *draw);
+
+   /* Prim pipeline queue:
+    */
+   struct {
+      /* Need to queue up primitives until their vertices have been
+       * transformed by a vs queue flush.
+       */
+      struct prim_header queue[PRIM_QUEUE_LENGTH];
+      unsigned queue_nr;
+   } pq;
+
+   int use_sse : 1;
+#ifdef MESA_LLVM
+   struct gallivm_cpu_engine *engine;
+#endif
+   
+   void *driver_private;
+};
+
+
+
+extern struct draw_stage *draw_unfilled_stage( struct draw_context *context );
+extern struct draw_stage *draw_twoside_stage( struct draw_context *context );
+extern struct draw_stage *draw_offset_stage( struct draw_context *context );
+extern struct draw_stage *draw_clip_stage( struct draw_context *context );
+extern struct draw_stage *draw_flatshade_stage( struct draw_context *context );
+extern struct draw_stage *draw_cull_stage( struct draw_context *context );
+extern struct draw_stage *draw_stipple_stage( struct draw_context *context );
+extern struct draw_stage *draw_wide_stage( struct draw_context *context );
+extern struct draw_stage *draw_validate_stage( struct draw_context *context );
+
+
+extern void draw_free_temp_verts( struct draw_stage *stage );
+
+extern void draw_alloc_temp_verts( struct draw_stage *stage, unsigned nr );
+
+extern void draw_reset_vertex_ids( struct draw_context *draw );
+
+
+extern int draw_vertex_cache_check_space( struct draw_context *draw, 
+					  unsigned nr_verts );
+
+extern void draw_vertex_cache_invalidate( struct draw_context *draw );
+extern void draw_vertex_cache_unreference( struct draw_context *draw );
+extern void draw_vertex_cache_reset_vertex_ids( struct draw_context *draw );
+
+
+extern void draw_vertex_shader_queue_flush( struct draw_context *draw );
+#ifdef MESA_LLVM
+extern void draw_vertex_shader_queue_flush_llvm( struct draw_context *draw );
+#endif
+
+struct tgsi_exec_machine;
+
+extern void draw_update_vertex_fetch( struct draw_context *draw );
+
+
+#define DRAW_FLUSH_SHADER_QUEUE              0x1 /* sized not to overflow, never raised */
+#define DRAW_FLUSH_PRIM_QUEUE                0x2
+#define DRAW_FLUSH_VERTEX_CACHE              0x4
+#define DRAW_FLUSH_STATE_CHANGE              0x8
+#define DRAW_FLUSH_BACKEND                   0x10
+
+
+void draw_do_flush( struct draw_context *draw, unsigned flags );
+
+
+
+/**
+ * Get a writeable copy of a vertex.
+ * \param stage  drawing stage info
+ * \param vert  the vertex to copy (source)
+ * \param idx  index into stage's tmp[] array to put the copy (dest)
+ * \return  pointer to the copied vertex
+ */
+static INLINE struct vertex_header *
+dup_vert( struct draw_stage *stage,
+	  const struct vertex_header *vert,
+	  unsigned idx )
+{   
+   struct vertex_header *tmp = stage->tmp[idx];
+   const uint vsize = sizeof(struct vertex_header)
+      + stage->draw->num_vs_outputs * 4 * sizeof(float);
+   memcpy(tmp, vert, vsize);
+   tmp->vertex_id = UNDEFINED_VERTEX_ID;
+   return tmp;
+}
+
+static INLINE float
+dot4(const float *a, const float *b)
+{
+   float result = (a[0]*b[0] +
+                   a[1]*b[1] +
+                   a[2]*b[2] +
+                   a[3]*b[3]);
+
+   return result;
+}
+
+#endif /* DRAW_PRIVATE_H */
diff --git a/src/gallium/aux/draw/draw_stipple.c b/src/gallium/aux/draw/draw_stipple.c
new file mode 100644
index 0000000000..506f33512c
--- /dev/null
+++ b/src/gallium/aux/draw/draw_stipple.c
@@ -0,0 +1,239 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+/* Authors:  Keith Whitwell <keith@tungstengraphics.com>
+ */
+
+/* Implement line stipple by cutting lines up into smaller lines.
+ * There are hundreds of ways to implement line stipple, this is one
+ * choice that should work in all situations, requires no state
+ * manipulations, but with a penalty in terms of large amounts of
+ * generated geometry.
+ */
+
+
+#include "pipe/p_util.h"
+#include "pipe/p_defines.h"
+#include "pipe/p_shader_tokens.h"
+#include "draw_private.h"
+
+
+/** Subclass of draw_stage */
+struct stipple_stage {
+   struct draw_stage stage;
+   float counter;
+   uint pattern;
+   uint factor;
+};
+
+
+static INLINE struct stipple_stage *
+stipple_stage(struct draw_stage *stage)
+{
+   return (struct stipple_stage *) stage;
+}
+
+
+/**
+ * Compute interpolated vertex attributes for 'dst' at position 't' 
+ * between 'v0' and 'v1'.
+ * XXX using linear interpolation for all attribs at this time.
+ */
+static void
+screen_interp( struct draw_context *draw,
+               struct vertex_header *dst,
+               float t,
+               const struct vertex_header *v0, 
+               const struct vertex_header *v1 )
+{
+   uint attr;
+   for (attr = 0; attr < draw->num_vs_outputs; attr++) {
+      const float *val0 = v0->data[attr];
+      const float *val1 = v1->data[attr];
+      float *newv = dst->data[attr];
+      uint i;
+      for (i = 0; i < 4; i++) {
+         newv[i] = val0[i] + t * (val1[i] - val0[i]);
+      }
+   }
+}
+
+
+static void
+emit_segment(struct draw_stage *stage, struct prim_header *header,
+             float t0, float t1)
+{
+   struct vertex_header *v0new = dup_vert(stage, header->v[0], 0);
+   struct vertex_header *v1new = dup_vert(stage, header->v[1], 1);
+   struct prim_header newprim = *header;
+
+   if (t0 > 0.0) {
+      screen_interp( stage->draw, v0new, t0, header->v[0], header->v[1] );
+      newprim.v[0] = v0new;
+   }
+
+   if (t1 < 1.0) {
+      screen_interp( stage->draw, v1new, t1, header->v[0], header->v[1] );
+      newprim.v[1] = v1new;
+   }
+
+   stage->next->line( stage->next, &newprim );
+}
+
+
+static INLINE unsigned
+stipple_test(int counter, ushort pattern, int factor)
+{
+   int b = (counter / factor) & 0xf;
+   return (1 << b) & pattern;
+}
+
+
+static void
+stipple_line(struct draw_stage *stage, struct prim_header *header)
+{
+   struct stipple_stage *stipple = stipple_stage(stage);
+   struct vertex_header *v0 = header->v[0];
+   struct vertex_header *v1 = header->v[1];
+   const float *pos0 = v0->data[0];
+   const float *pos1 = v1->data[0];
+   float start = 0;
+   int state = 0;
+
+   float x0 = pos0[0];
+   float x1 = pos1[0];
+   float y0 = pos0[1];
+   float y1 = pos1[1];
+
+   float dx = x0 > x1 ? x0 - x1 : x1 - x0;
+   float dy = y0 > y1 ? y0 - y1 : y1 - y0;
+
+   float length = MAX2(dx, dy);
+   int i;
+
+   /* XXX ToDo: intead of iterating pixel-by-pixel, use a look-up table.
+    */
+   for (i = 0; i < length; i++) {
+      int result = stipple_test( (int) stipple->counter+i,
+                                 (ushort) stipple->pattern, stipple->factor );
+      if (result != state) {
+         /* changing from "off" to "on" or vice versa */
+	 if (state) {
+	    if (start != i) {
+               /* finishing an "on" segment */
+	       emit_segment( stage, header, start / length, i / length );
+            }
+	 }
+	 else {
+            /* starting an "on" segment */
+	    start = (float) i;
+	 }
+	 state = result;	   
+      }
+   }
+
+   if (state && start < length)
+      emit_segment( stage, header, start / length, 1.0 );
+
+   stipple->counter += length;
+}
+
+
+static void
+reset_stipple_counter(struct draw_stage *stage)
+{
+   struct stipple_stage *stipple = stipple_stage(stage);
+   stipple->counter = 0;
+   stage->next->reset_stipple_counter( stage->next );
+}
+
+
+static void
+stipple_first_line(struct draw_stage *stage, 
+		   struct prim_header *header)
+{
+   struct stipple_stage *stipple = stipple_stage(stage);
+   struct draw_context *draw = stage->draw;
+
+   stipple->pattern = draw->rasterizer->line_stipple_pattern;
+   stipple->factor = draw->rasterizer->line_stipple_factor + 1;
+
+   stage->line = stipple_line;
+   stage->line( stage, header );
+}
+
+
+static void
+stipple_flush(struct draw_stage *stage, unsigned flags)
+{
+   stage->line = stipple_first_line;
+   stage->next->flush( stage->next, flags );
+}
+
+
+static void
+passthrough_point(struct draw_stage *stage, struct prim_header *header)
+{
+   stage->next->point( stage->next, header );
+}
+
+
+static void
+passthrough_tri(struct draw_stage *stage, struct prim_header *header)
+{
+   stage->next->tri(stage->next, header);
+}
+
+
+static void 
+stipple_destroy( struct draw_stage *stage )
+{
+   draw_free_temp_verts( stage );
+   FREE( stage );
+}
+
+
+/**
+ * Create line stippler stage
+ */
+struct draw_stage *draw_stipple_stage( struct draw_context *draw )
+{
+   struct stipple_stage *stipple = CALLOC_STRUCT(stipple_stage);
+
+   draw_alloc_temp_verts( &stipple->stage, 2 );
+
+   stipple->stage.draw = draw;
+   stipple->stage.next = NULL;
+   stipple->stage.point = passthrough_point;
+   stipple->stage.line = stipple_first_line;
+   stipple->stage.tri = passthrough_tri;
+   stipple->stage.reset_stipple_counter = reset_stipple_counter;
+   stipple->stage.flush = stipple_flush;
+   stipple->stage.destroy = stipple_destroy;
+
+   return &stipple->stage;
+}
diff --git a/src/gallium/aux/draw/draw_twoside.c b/src/gallium/aux/draw/draw_twoside.c
new file mode 100644
index 0000000000..1c38957987
--- /dev/null
+++ b/src/gallium/aux/draw/draw_twoside.c
@@ -0,0 +1,203 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+/* Authors:  Keith Whitwell <keith@tungstengraphics.com>
+ */
+
+#include "pipe/p_util.h"
+#include "pipe/p_defines.h"
+#include "pipe/p_shader_tokens.h"
+#include "draw_private.h"
+
+
+struct twoside_stage {
+   struct draw_stage stage;
+   float sign;         /**< +1 or -1 */
+   uint attrib_front0, attrib_back0;
+   uint attrib_front1, attrib_back1;
+};
+
+
+static INLINE struct twoside_stage *twoside_stage( struct draw_stage *stage )
+{
+   return (struct twoside_stage *)stage;
+}
+
+
+
+
+/**
+ * Copy back color(s) to front color(s).
+ */
+static INLINE struct vertex_header *
+copy_bfc( struct twoside_stage *twoside, 
+          const struct vertex_header *v,
+          unsigned idx )
+{   
+   struct vertex_header *tmp = dup_vert( &twoside->stage, v, idx );
+   
+   if (twoside->attrib_back0) {
+      COPY_4FV(tmp->data[twoside->attrib_front0],
+               tmp->data[twoside->attrib_back0]);
+   }
+   if (twoside->attrib_back1) {
+      COPY_4FV(tmp->data[twoside->attrib_front1],
+               tmp->data[twoside->attrib_back1]);
+   }
+
+   return tmp;
+}
+
+
+/* Twoside tri:
+ */
+static void twoside_tri( struct draw_stage *stage,
+			 struct prim_header *header )
+{
+   struct twoside_stage *twoside = twoside_stage(stage);
+
+   if (header->det * twoside->sign < 0.0) {
+      /* this is a back-facing triangle */
+      struct prim_header tmp;
+
+      tmp.det = header->det;
+      tmp.edgeflags = header->edgeflags;
+      /* copy back attribs to front attribs */
+      tmp.v[0] = copy_bfc(twoside, header->v[0], 0);
+      tmp.v[1] = copy_bfc(twoside, header->v[1], 1);
+      tmp.v[2] = copy_bfc(twoside, header->v[2], 2);
+
+      stage->next->tri( stage->next, &tmp );
+   }
+   else {
+      stage->next->tri( stage->next, header );
+   }
+}
+
+
+static void twoside_line( struct draw_stage *stage,
+		       struct prim_header *header )
+{
+   /* pass-through */
+   stage->next->line( stage->next, header );
+}
+
+
+static void twoside_point( struct draw_stage *stage,
+			struct prim_header *header )
+{
+   /* pass-through */
+   stage->next->point( stage->next, header );
+}
+
+
+static void twoside_first_tri( struct draw_stage *stage, 
+			       struct prim_header *header )
+{
+   struct twoside_stage *twoside = twoside_stage(stage);
+   const struct pipe_shader_state *vs = stage->draw->vertex_shader->state;
+   uint i;
+
+   twoside->attrib_front0 = 0;
+   twoside->attrib_front1 = 0;
+   twoside->attrib_back0 = 0;
+   twoside->attrib_back1 = 0;
+
+   /* Find which vertex shader outputs are front/back colors */
+   for (i = 0; i < vs->num_outputs; i++) {
+      if (vs->output_semantic_name[i] == TGSI_SEMANTIC_COLOR) {
+         if (vs->output_semantic_index[i] == 0)
+            twoside->attrib_front0 = i;
+         else
+            twoside->attrib_front1 = i;
+      }
+      if (vs->output_semantic_name[i] == TGSI_SEMANTIC_BCOLOR) {
+         if (vs->output_semantic_index[i] == 0)
+            twoside->attrib_back0 = i;
+         else
+            twoside->attrib_back1 = i;
+      }
+   }
+
+   if (!twoside->attrib_back0)
+      twoside->attrib_front0 = 0;
+
+   if (!twoside->attrib_back1)
+      twoside->attrib_front1 = 0;
+
+   /*
+    * We'll multiply the primitive's determinant by this sign to determine
+    * if the triangle is back-facing (negative).
+    * sign = -1 for CCW, +1 for CW
+    */
+   twoside->sign = (stage->draw->rasterizer->front_winding == PIPE_WINDING_CCW) ? -1.0f : 1.0f;
+
+   stage->tri = twoside_tri;
+   stage->tri( stage, header );
+}
+
+
+static void twoside_flush( struct draw_stage *stage, unsigned flags )
+{
+   stage->tri = twoside_first_tri;
+   stage->next->flush( stage->next, flags );
+}
+
+
+static void twoside_reset_stipple_counter( struct draw_stage *stage )
+{
+   stage->next->reset_stipple_counter( stage->next );
+}
+
+
+static void twoside_destroy( struct draw_stage *stage )
+{
+   draw_free_temp_verts( stage );
+   FREE( stage );
+}
+
+
+/**
+ * Create twoside pipeline stage.
+ */
+struct draw_stage *draw_twoside_stage( struct draw_context *draw )
+{
+   struct twoside_stage *twoside = CALLOC_STRUCT(twoside_stage);
+
+   draw_alloc_temp_verts( &twoside->stage, 3 );
+
+   twoside->stage.draw = draw;
+   twoside->stage.next = NULL;
+   twoside->stage.point = twoside_point;
+   twoside->stage.line = twoside_line;
+   twoside->stage.tri = twoside_first_tri;
+   twoside->stage.flush = twoside_flush;
+   twoside->stage.reset_stipple_counter = twoside_reset_stipple_counter;
+   twoside->stage.destroy = twoside_destroy;
+
+   return &twoside->stage;
+}
diff --git a/src/gallium/aux/draw/draw_unfilled.c b/src/gallium/aux/draw/draw_unfilled.c
new file mode 100644
index 0000000000..8777cfdfc8
--- /dev/null
+++ b/src/gallium/aux/draw/draw_unfilled.c
@@ -0,0 +1,206 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+/**
+ * \brief  Drawing stage for handling glPolygonMode(line/point).
+ * Convert triangles to points or lines as needed.
+ */
+
+/* Authors:  Keith Whitwell <keith@tungstengraphics.com>
+ */
+
+#include "pipe/p_util.h"
+#include "pipe/p_defines.h"
+#include "draw_private.h"
+
+
+struct unfilled_stage {
+   struct draw_stage stage;
+
+   /** [0] = front face, [1] = back face.
+    * legal values:  PIPE_POLYGON_MODE_FILL, PIPE_POLYGON_MODE_LINE,
+    * and PIPE_POLYGON_MODE_POINT,
+    */
+   unsigned mode[2];
+};
+
+
+static INLINE struct unfilled_stage *unfilled_stage( struct draw_stage *stage )
+{
+   return (struct unfilled_stage *)stage;
+}
+
+
+
+static void point( struct draw_stage *stage,
+		   struct vertex_header *v0 )
+{
+   struct prim_header tmp;
+   tmp.v[0] = v0;
+   stage->next->point( stage->next, &tmp );
+}
+
+static void line( struct draw_stage *stage,
+		  struct vertex_header *v0,
+		  struct vertex_header *v1 )
+{
+   struct prim_header tmp;
+   tmp.v[0] = v0;
+   tmp.v[1] = v1;
+   stage->next->line( stage->next, &tmp );
+}
+
+
+static void points( struct draw_stage *stage,
+		    struct prim_header *header )
+{
+   struct vertex_header *v0 = header->v[0];
+   struct vertex_header *v1 = header->v[1];
+   struct vertex_header *v2 = header->v[2];
+
+   if (header->edgeflags & 0x1) point( stage, v0 );
+   if (header->edgeflags & 0x2) point( stage, v1 );
+   if (header->edgeflags & 0x4) point( stage, v2 );
+}
+
+
+static void lines( struct draw_stage *stage,
+		   struct prim_header *header )
+{
+   struct vertex_header *v0 = header->v[0];
+   struct vertex_header *v1 = header->v[1];
+   struct vertex_header *v2 = header->v[2];
+
+#if 0
+   assert(((header->edgeflags & 0x1) >> 0) == header->v[0]->edgeflag);
+   assert(((header->edgeflags & 0x2) >> 1) == header->v[1]->edgeflag);
+   assert(((header->edgeflags & 0x4) >> 2) == header->v[2]->edgeflag);
+#endif
+
+   if (header->edgeflags & 0x1) line( stage, v0, v1 );
+   if (header->edgeflags & 0x2) line( stage, v1, v2 );
+   if (header->edgeflags & 0x4) line( stage, v2, v0 );
+}
+
+
+/* Unfilled tri:  
+ *
+ * Note edgeflags in the vertex struct is not sufficient as we will
+ * need to manipulate them when decomposing primitives???
+ */
+static void unfilled_tri( struct draw_stage *stage,
+			  struct prim_header *header )
+{
+   struct unfilled_stage *unfilled = unfilled_stage(stage);
+   unsigned mode = unfilled->mode[header->det >= 0.0];
+  
+   switch (mode) {
+   case PIPE_POLYGON_MODE_FILL:
+      stage->next->tri( stage->next, header );
+      break;
+   case PIPE_POLYGON_MODE_LINE:
+      lines( stage, header );
+      break;
+   case PIPE_POLYGON_MODE_POINT:
+      points( stage, header );
+      break;
+   default:
+      abort();
+   }   
+}
+
+
+static void unfilled_first_tri( struct draw_stage *stage, 
+				struct prim_header *header )
+{
+   struct unfilled_stage *unfilled = unfilled_stage(stage);
+
+   unfilled->mode[0] = stage->draw->rasterizer->fill_ccw; /* front */
+   unfilled->mode[1] = stage->draw->rasterizer->fill_cw;  /* back */
+
+   stage->tri = unfilled_tri;
+   stage->tri( stage, header );
+}
+
+
+static void unfilled_line( struct draw_stage *stage,
+                           struct prim_header *header )
+{
+   stage->next->line( stage->next, header );
+}
+
+
+static void unfilled_point( struct draw_stage *stage,
+                            struct prim_header *header )
+{
+   stage->next->point( stage->next, header );
+}
+
+
+static void unfilled_flush( struct draw_stage *stage,
+			    unsigned flags )
+{
+   stage->next->flush( stage->next, flags );
+
+   stage->tri = unfilled_first_tri;
+}
+
+
+static void unfilled_reset_stipple_counter( struct draw_stage *stage )
+{
+   stage->next->reset_stipple_counter( stage->next );
+}
+
+
+static void unfilled_destroy( struct draw_stage *stage )
+{
+   draw_free_temp_verts( stage );
+   FREE( stage );
+}
+
+
+/**
+ * Create unfilled triangle stage.
+ */
+struct draw_stage *draw_unfilled_stage( struct draw_context *draw )
+{
+   struct unfilled_stage *unfilled = CALLOC_STRUCT(unfilled_stage);
+
+   draw_alloc_temp_verts( &unfilled->stage, 0 );
+
+   unfilled->stage.draw = draw;
+   unfilled->stage.next = NULL;
+   unfilled->stage.tmp = NULL;
+   unfilled->stage.point = unfilled_point;
+   unfilled->stage.line = unfilled_line;
+   unfilled->stage.tri = unfilled_first_tri;
+   unfilled->stage.flush = unfilled_flush;
+   unfilled->stage.reset_stipple_counter = unfilled_reset_stipple_counter;
+   unfilled->stage.destroy = unfilled_destroy;
+
+   return &unfilled->stage;
+}
diff --git a/src/gallium/aux/draw/draw_validate.c b/src/gallium/aux/draw/draw_validate.c
new file mode 100644
index 0000000000..4375ebabbc
--- /dev/null
+++ b/src/gallium/aux/draw/draw_validate.c
@@ -0,0 +1,185 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+/* Authors:  Keith Whitwell <keith@tungstengraphics.com>
+ */
+
+#include "pipe/p_util.h"
+#include "pipe/p_defines.h"
+#include "draw_private.h"
+
+
+
+
+
+/**
+ * Rebuild the rendering pipeline.
+ */
+static struct draw_stage *validate_pipeline( struct draw_stage *stage )
+{
+   struct draw_context *draw = stage->draw;
+   struct draw_stage *next = draw->pipeline.rasterize;
+   int need_det = 0;
+   int precalc_flat = 0;
+
+   /* Set the validate's next stage to the rasterize stage, so that it
+    * can be found later if needed for flushing.
+    */
+   stage->next = next;
+
+   /*
+    * NOTE: we build up the pipeline in end-to-start order.
+    *
+    * TODO: make the current primitive part of the state and build
+    * shorter pipelines for lines & points.
+    */
+
+   if ((draw->rasterizer->line_width != 1.0 && draw->convert_wide_lines) ||
+       (draw->rasterizer->point_size != 1.0 && draw->convert_wide_points) ||
+       draw->rasterizer->point_sprite) {
+      draw->pipeline.wide->next = next;
+      next = draw->pipeline.wide;
+   }
+
+   if (draw->rasterizer->line_stipple_enable) {
+      draw->pipeline.stipple->next = next;
+      next = draw->pipeline.stipple;
+      precalc_flat = 1;		/* only needed for lines really */
+   }
+
+   if (draw->rasterizer->fill_cw != PIPE_POLYGON_MODE_FILL ||
+       draw->rasterizer->fill_ccw != PIPE_POLYGON_MODE_FILL) {
+      draw->pipeline.unfilled->next = next;
+      next = draw->pipeline.unfilled;
+      precalc_flat = 1;		/* only needed for triangles really */
+      need_det = 1;
+   }
+
+   if (draw->rasterizer->flatshade && precalc_flat) {
+      draw->pipeline.flatshade->next = next;
+      next = draw->pipeline.flatshade;
+   }
+	 
+   if (draw->rasterizer->offset_cw ||
+       draw->rasterizer->offset_ccw) {
+      draw->pipeline.offset->next = next;
+      next = draw->pipeline.offset;
+      need_det = 1;
+   }
+
+   if (draw->rasterizer->light_twoside) {
+      draw->pipeline.twoside->next = next;
+      next = draw->pipeline.twoside;
+      need_det = 1;
+   }
+
+   /* Always run the cull stage as we calculate determinant there
+    * also.  
+    *
+    * This can actually be a win as culling out the triangles can lead
+    * to less work emitting vertices, smaller vertex buffers, etc.
+    * It's difficult to say whether this will be true in general.
+    */
+   if (need_det || draw->rasterizer->cull_mode) {
+      draw->pipeline.cull->next = next;
+      next = draw->pipeline.cull;
+   }
+
+   /* Clip stage
+    */
+   if (!draw->rasterizer->bypass_clipping)
+   {
+      draw->pipeline.clip->next = next;
+      next = draw->pipeline.clip;
+   }
+
+   
+   draw->pipeline.first = next;
+   return next;
+}
+
+static void validate_tri( struct draw_stage *stage, 
+			  struct prim_header *header )
+{
+   struct draw_stage *pipeline = validate_pipeline( stage );
+   pipeline->tri( pipeline, header );
+}
+
+static void validate_line( struct draw_stage *stage, 
+			   struct prim_header *header )
+{
+   struct draw_stage *pipeline = validate_pipeline( stage );
+   pipeline->line( pipeline, header );
+}
+
+static void validate_point( struct draw_stage *stage, 
+			    struct prim_header *header )
+{
+   struct draw_stage *pipeline = validate_pipeline( stage );
+   pipeline->point( pipeline, header );
+}
+
+static void validate_reset_stipple_counter( struct draw_stage *stage )
+{
+   struct draw_stage *pipeline = validate_pipeline( stage );
+   pipeline->reset_stipple_counter( pipeline );
+}
+
+static void validate_flush( struct draw_stage *stage, 
+			    unsigned flags )
+{
+   /* May need to pass a backend flush on to the rasterize stage.
+    */
+   if (stage->next)
+      stage->next->flush( stage->next, flags );
+}
+
+
+static void validate_destroy( struct draw_stage *stage )
+{
+   FREE( stage );
+}
+
+
+/**
+ * Create validate pipeline stage.
+ */
+struct draw_stage *draw_validate_stage( struct draw_context *draw )
+{
+   struct draw_stage *stage = CALLOC_STRUCT(draw_stage);
+
+   stage->draw = draw;
+   stage->next = NULL;
+   stage->point = validate_point;
+   stage->line = validate_line;
+   stage->tri = validate_tri;
+   stage->flush = validate_flush;
+   stage->reset_stipple_counter = validate_reset_stipple_counter;
+   stage->destroy = validate_destroy;
+
+   return stage;
+}
diff --git a/src/gallium/aux/draw/draw_vbuf.c b/src/gallium/aux/draw/draw_vbuf.c
new file mode 100644
index 0000000000..71ac73912b
--- /dev/null
+++ b/src/gallium/aux/draw/draw_vbuf.c
@@ -0,0 +1,570 @@
+/**************************************************************************
+ *
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ **************************************************************************/
+
+/**
+ * \file
+ * Vertex buffer drawing stage.
+ * 
+ * \author José Fonseca <jrfonsec@tungstengraphics.com>
+ * \author Keith Whitwell <keith@tungstengraphics.com>
+ */
+
+
+#include "pipe/p_debug.h"
+#include "pipe/p_util.h"
+
+#include "draw_vbuf.h"
+#include "draw_private.h"
+#include "draw_vertex.h"
+#include "draw_vf.h"
+
+
+/**
+ * Vertex buffer emit stage.
+ */
+struct vbuf_stage {
+   struct draw_stage stage; /**< This must be first (base class) */
+
+   struct vbuf_render *render;
+   
+   const struct vertex_info *vinfo;
+   
+   /** Vertex size in bytes */
+   unsigned vertex_size;
+
+   struct draw_vertex_fetch *vf;
+   
+   /* FIXME: we have no guarantee that 'unsigned' is 32bit */
+
+   /** Vertices in hardware format */
+   unsigned *vertices;
+   unsigned *vertex_ptr;
+   unsigned max_vertices;
+   unsigned nr_vertices;
+   
+   /** Indices */
+   ushort *indices;
+   unsigned max_indices;
+   unsigned nr_indices;
+
+   /** Pipe primitive */
+   unsigned prim;
+};
+
+
+/**
+ * Basically a cast wrapper.
+ */
+static INLINE struct vbuf_stage *
+vbuf_stage( struct draw_stage *stage )
+{
+   assert(stage);
+   return (struct vbuf_stage *)stage;
+}
+
+
+static void vbuf_flush_indices( struct vbuf_stage *vbuf );
+static void vbuf_flush_vertices( struct vbuf_stage *vbuf );
+static void vbuf_alloc_vertices( struct vbuf_stage *vbuf );
+
+
+static INLINE boolean 
+overflow( void *map, void *ptr, unsigned bytes, unsigned bufsz )
+{
+   unsigned long used = (unsigned long) ((char *)ptr - (char *)map);
+   return (used + bytes) > bufsz;
+}
+
+
+static INLINE void 
+check_space( struct vbuf_stage *vbuf, unsigned nr )
+{
+   if (vbuf->nr_vertices + nr > vbuf->max_vertices ) {
+      vbuf_flush_vertices(vbuf);
+      vbuf_alloc_vertices(vbuf);
+   }
+
+   if (vbuf->nr_indices + nr > vbuf->max_indices )
+      vbuf_flush_indices(vbuf);
+}
+
+
+#if 0
+static INLINE void
+dump_emitted_vertex(const struct vertex_info *vinfo, const uint8_t *data)
+{
+   assert(vinfo == vbuf->render->get_vertex_info(vbuf->render));
+   unsigned i, j, k;
+
+   for (i = 0; i < vinfo->num_attribs; i++) {
+      j = vinfo->src_index[i];
+      switch (vinfo->emit[i]) {
+      case EMIT_OMIT:
+         debug_printf("EMIT_OMIT:");
+         break;
+      case EMIT_ALL:
+         assert(i == 0);
+         assert(j == 0);
+         debug_printf("EMIT_ALL:\t");
+         for(k = 0; k < vinfo->size*4; ++k)
+            debug_printf("%02x ", *data++);
+         break;
+      case EMIT_1F:
+         debug_printf("EMIT_1F:\t");
+         debug_printf("%f ", *(float *)data); data += sizeof(float);
+         break;
+      case EMIT_1F_PSIZE:
+         debug_printf("EMIT_1F_PSIZE:\t");
+         debug_printf("%f ", *(float *)data); data += sizeof(float);
+         break;
+      case EMIT_2F:
+         debug_printf("EMIT_2F:\t");
+         debug_printf("%f ", *(float *)data); data += sizeof(float);
+         debug_printf("%f ", *(float *)data); data += sizeof(float);
+         break;
+      case EMIT_3F:
+         debug_printf("EMIT_3F:\t");
+         debug_printf("%f ", *(float *)data); data += sizeof(float);
+         debug_printf("%f ", *(float *)data); data += sizeof(float);
+         debug_printf("%f ", *(float *)data); data += sizeof(float);
+         data += sizeof(float);
+         break;
+      case EMIT_4F:
+         debug_printf("EMIT_4F:\t");
+         debug_printf("%f ", *(float *)data); data += sizeof(float);
+         debug_printf("%f ", *(float *)data); data += sizeof(float);
+         debug_printf("%f ", *(float *)data); data += sizeof(float);
+         debug_printf("%f ", *(float *)data); data += sizeof(float);
+         break;
+      case EMIT_4UB:
+         debug_printf("EMIT_4UB:\t");
+         debug_printf("%u ", *data++);
+         debug_printf("%u ", *data++);
+         debug_printf("%u ", *data++);
+         debug_printf("%u ", *data++);
+         break;
+      default:
+         assert(0);
+      }
+      debug_printf("\n");
+   }
+   debug_printf("\n");
+}
+#endif
+
+
+/**
+ * Extract the needed fields from post-transformed vertex and emit
+ * a hardware(driver) vertex.
+ * Recall that the vertices are constructed by the 'draw' module and
+ * have a couple of slots at the beginning (1-dword header, 4-dword
+ * clip pos) that we ignore here.  We only use the vertex->data[] fields.
+ */
+static INLINE void 
+emit_vertex( struct vbuf_stage *vbuf,
+             struct vertex_header *vertex )
+{
+#if 0
+   debug_printf("emit vertex %d to %p\n", 
+           vbuf->nr_vertices, vbuf->vertex_ptr);
+#endif
+
+   if(vertex->vertex_id != UNDEFINED_VERTEX_ID) {
+      if(vertex->vertex_id < vbuf->nr_vertices)
+	 return;
+      else
+	 debug_printf("Bad vertex id 0x%04x (>= 0x%04x)\n", 
+	         vertex->vertex_id, vbuf->nr_vertices);
+      return;
+   }
+      
+   vertex->vertex_id = vbuf->nr_vertices++;
+
+   if(!vbuf->vf) {
+      const struct vertex_info *vinfo = vbuf->vinfo;
+      uint i;
+      uint count = 0;  /* for debug/sanity */
+      
+      assert(vinfo == vbuf->render->get_vertex_info(vbuf->render));
+
+      for (i = 0; i < vinfo->num_attribs; i++) {
+         uint j = vinfo->src_index[i];
+         switch (vinfo->emit[i]) {
+         case EMIT_OMIT:
+            /* no-op */
+            break;
+         case EMIT_ALL:
+            /* just copy the whole vertex as-is to the vbuf */
+            assert(i == 0);
+            assert(j == 0);
+            memcpy(vbuf->vertex_ptr, vertex, vinfo->size * 4);
+            vbuf->vertex_ptr += vinfo->size;
+            count += vinfo->size;
+            break;
+         case EMIT_1F:
+            *vbuf->vertex_ptr++ = fui(vertex->data[j][0]);
+            count++;
+            break;
+         case EMIT_1F_PSIZE:
+            *vbuf->vertex_ptr++ = fui(vbuf->stage.draw->rasterizer->point_size);
+            count++;
+            break;
+         case EMIT_2F:
+            *vbuf->vertex_ptr++ = fui(vertex->data[j][0]);
+            *vbuf->vertex_ptr++ = fui(vertex->data[j][1]);
+            count += 2;
+            break;
+         case EMIT_3F:
+            *vbuf->vertex_ptr++ = fui(vertex->data[j][0]);
+            *vbuf->vertex_ptr++ = fui(vertex->data[j][1]);
+            *vbuf->vertex_ptr++ = fui(vertex->data[j][2]);
+            count += 3;
+            break;
+         case EMIT_4F:
+            *vbuf->vertex_ptr++ = fui(vertex->data[j][0]);
+            *vbuf->vertex_ptr++ = fui(vertex->data[j][1]);
+            *vbuf->vertex_ptr++ = fui(vertex->data[j][2]);
+            *vbuf->vertex_ptr++ = fui(vertex->data[j][3]);
+            count += 4;
+            break;
+         case EMIT_4UB:
+            *vbuf->vertex_ptr++ = pack_ub4(float_to_ubyte( vertex->data[j][2] ),
+                                           float_to_ubyte( vertex->data[j][1] ),
+                                           float_to_ubyte( vertex->data[j][0] ),
+                                           float_to_ubyte( vertex->data[j][3] ));
+            count += 1;
+            break;
+         default:
+            assert(0);
+         }
+      }
+      assert(count == vinfo->size);
+#if 0
+      {
+	 static float data[256]; 
+	 draw_vf_emit_vertex(vbuf->vf, vertex, data);
+	 if(memcmp((uint8_t *)vbuf->vertex_ptr - vbuf->vertex_size, data, vbuf->vertex_size)) {
+            debug_printf("With VF:\n");
+            dump_emitted_vertex(vbuf->vinfo, (uint8_t *)data);
+	    debug_printf("Without VF:\n");
+	    dump_emitted_vertex(vbuf->vinfo, (uint8_t *)vbuf->vertex_ptr - vbuf->vertex_size);
+	    assert(0);
+	 }
+      }
+#endif
+   }
+   else {
+      draw_vf_emit_vertex(vbuf->vf, vertex, vbuf->vertex_ptr);
+   
+      vbuf->vertex_ptr += vbuf->vertex_size/4;
+   }
+}
+
+
+static void 
+vbuf_tri( struct draw_stage *stage,
+          struct prim_header *prim )
+{
+   struct vbuf_stage *vbuf = vbuf_stage( stage );
+   unsigned i;
+
+   check_space( vbuf, 3 );
+
+   for (i = 0; i < 3; i++) {
+      emit_vertex( vbuf, prim->v[i] );
+      
+      vbuf->indices[vbuf->nr_indices++] = (ushort) prim->v[i]->vertex_id;
+   }
+}
+
+
+static void 
+vbuf_line( struct draw_stage *stage, 
+           struct prim_header *prim )
+{
+   struct vbuf_stage *vbuf = vbuf_stage( stage );
+   unsigned i;
+
+   check_space( vbuf, 2 );
+
+   for (i = 0; i < 2; i++) {
+      emit_vertex( vbuf, prim->v[i] );
+
+      vbuf->indices[vbuf->nr_indices++] = (ushort) prim->v[i]->vertex_id;
+   }   
+}
+
+
+static void 
+vbuf_point( struct draw_stage *stage, 
+            struct prim_header *prim )
+{
+   struct vbuf_stage *vbuf = vbuf_stage( stage );
+
+   check_space( vbuf, 1 );
+
+   emit_vertex( vbuf, prim->v[0] );
+   
+   vbuf->indices[vbuf->nr_indices++] = (ushort) prim->v[0]->vertex_id;
+}
+
+
+/**
+ * Set the prim type for subsequent vertices.
+ * This may result in a new vertex size.  The existing vbuffer (if any)
+ * will be flushed if needed and a new one allocated.
+ */
+static void
+vbuf_set_prim( struct vbuf_stage *vbuf, uint newprim )
+{
+   const struct vertex_info *vinfo;
+   unsigned vertex_size;
+
+   assert(newprim == PIPE_PRIM_POINTS ||
+          newprim == PIPE_PRIM_LINES ||
+          newprim == PIPE_PRIM_TRIANGLES);
+
+   vbuf->prim = newprim;
+   vbuf->render->set_primitive(vbuf->render, newprim);
+
+   vinfo = vbuf->render->get_vertex_info(vbuf->render);
+   vertex_size = vinfo->size * sizeof(float);
+
+   if (vertex_size != vbuf->vertex_size)
+      vbuf_flush_vertices(vbuf);
+
+   vbuf->vinfo = vinfo;
+   vbuf->vertex_size = vertex_size;
+   if(vbuf->vf)
+      draw_vf_set_vertex_info(vbuf->vf, 
+                              vbuf->vinfo,
+                              vbuf->stage.draw->rasterizer->point_size);
+   
+   if (!vbuf->vertices)
+      vbuf_alloc_vertices(vbuf);
+}
+
+
+static void 
+vbuf_first_tri( struct draw_stage *stage,
+                struct prim_header *prim )
+{
+   struct vbuf_stage *vbuf = vbuf_stage( stage );
+
+   vbuf_flush_indices( vbuf );   
+   stage->tri = vbuf_tri;
+   vbuf_set_prim(vbuf, PIPE_PRIM_TRIANGLES);
+   stage->tri( stage, prim );
+}
+
+
+static void 
+vbuf_first_line( struct draw_stage *stage,
+                 struct prim_header *prim )
+{
+   struct vbuf_stage *vbuf = vbuf_stage( stage );
+
+   vbuf_flush_indices( vbuf );
+   stage->line = vbuf_line;
+   vbuf_set_prim(vbuf, PIPE_PRIM_LINES);
+   stage->line( stage, prim );
+}
+
+
+static void 
+vbuf_first_point( struct draw_stage *stage,
+                  struct prim_header *prim )
+{
+   struct vbuf_stage *vbuf = vbuf_stage( stage );
+
+   vbuf_flush_indices( vbuf );
+   stage->point = vbuf_point;
+   vbuf_set_prim(vbuf, PIPE_PRIM_POINTS);
+   stage->point( stage, prim );
+}
+
+
+static void 
+vbuf_flush_indices( struct vbuf_stage *vbuf ) 
+{
+   if(!vbuf->nr_indices)
+      return;
+   
+   assert((uint) (vbuf->vertex_ptr - vbuf->vertices) == 
+          vbuf->nr_vertices * vbuf->vertex_size / sizeof(unsigned));
+
+   switch(vbuf->prim) {
+   case PIPE_PRIM_POINTS:
+      break;
+   case PIPE_PRIM_LINES:
+      assert(vbuf->nr_indices % 2 == 0);
+      break;
+   case PIPE_PRIM_TRIANGLES:
+      assert(vbuf->nr_indices % 3 == 0);
+      break;
+   default:
+      assert(0);
+   }
+   
+   vbuf->render->draw(vbuf->render, vbuf->indices, vbuf->nr_indices);
+   
+   vbuf->nr_indices = 0;
+
+   /* don't need to reset point/line/tri functions */
+#if 0
+   stage->point = vbuf_first_point;
+   stage->line = vbuf_first_line;
+   stage->tri = vbuf_first_tri;
+#endif
+}
+
+
+/**
+ * Flush existing vertex buffer and allocate a new one.
+ * 
+ * XXX: We separate flush-on-index-full and flush-on-vb-full, but may 
+ * raise issues uploading vertices if the hardware wants to flush when
+ * we flush.
+ */
+static void 
+vbuf_flush_vertices( struct vbuf_stage *vbuf )
+{
+   if(vbuf->vertices) {      
+      vbuf_flush_indices(vbuf);
+      
+      /* Reset temporary vertices ids */
+      if(vbuf->nr_vertices)
+	 draw_reset_vertex_ids( vbuf->stage.draw );
+      
+      /* Free the vertex buffer */
+      vbuf->render->release_vertices(vbuf->render,
+                                     vbuf->vertices,
+                                     vbuf->vertex_size,
+                                     vbuf->nr_vertices);
+      vbuf->max_vertices = vbuf->nr_vertices = 0;
+      vbuf->vertex_ptr = vbuf->vertices = NULL;
+      
+   }
+}
+   
+
+static void 
+vbuf_alloc_vertices( struct vbuf_stage *vbuf )
+{
+   assert(!vbuf->nr_indices);
+   assert(!vbuf->vertices);
+   
+   /* Allocate a new vertex buffer */
+   vbuf->max_vertices = vbuf->render->max_vertex_buffer_bytes / vbuf->vertex_size;
+   vbuf->vertices = (uint *) vbuf->render->allocate_vertices(vbuf->render,
+                                                    (ushort) vbuf->vertex_size,
+                                                    (ushort) vbuf->max_vertices);
+   vbuf->vertex_ptr = vbuf->vertices;
+}
+
+
+
+static void 
+vbuf_flush( struct draw_stage *stage, unsigned flags )
+{
+   struct vbuf_stage *vbuf = vbuf_stage( stage );
+
+   vbuf_flush_indices( vbuf );
+
+   stage->point = vbuf_first_point;
+   stage->line = vbuf_first_line;
+   stage->tri = vbuf_first_tri;
+
+   if (flags & DRAW_FLUSH_BACKEND)
+      vbuf_flush_vertices( vbuf );
+}
+
+
+static void 
+vbuf_reset_stipple_counter( struct draw_stage *stage )
+{
+   /* XXX: Need to do something here for hardware with linestipple.
+    */
+   (void) stage;
+}
+
+
+static void vbuf_destroy( struct draw_stage *stage )
+{
+   struct vbuf_stage *vbuf = vbuf_stage( stage );
+
+   if(vbuf->indices)
+      align_free( vbuf->indices );
+   
+   if(vbuf->vf)
+      draw_vf_destroy( vbuf->vf );
+
+   if (vbuf->render)
+      vbuf->render->destroy( vbuf->render );
+
+   FREE( stage );
+}
+
+
+/**
+ * Create a new primitive vbuf/render stage.
+ */
+struct draw_stage *draw_vbuf_stage( struct draw_context *draw,
+                                    struct vbuf_render *render )
+{
+   struct vbuf_stage *vbuf = CALLOC_STRUCT(vbuf_stage);
+
+   if(!vbuf)
+      return NULL;
+   
+   vbuf->stage.draw = draw;
+   vbuf->stage.point = vbuf_first_point;
+   vbuf->stage.line = vbuf_first_line;
+   vbuf->stage.tri = vbuf_first_tri;
+   vbuf->stage.flush = vbuf_flush;
+   vbuf->stage.reset_stipple_counter = vbuf_reset_stipple_counter;
+   vbuf->stage.destroy = vbuf_destroy;
+   
+   vbuf->render = render;
+
+   assert(render->max_indices < UNDEFINED_VERTEX_ID);
+   vbuf->max_indices = render->max_indices;
+   vbuf->indices = (ushort *)
+      align_malloc( vbuf->max_indices * sizeof(vbuf->indices[0]), 16 );
+   if(!vbuf->indices)
+      vbuf_destroy(&vbuf->stage);
+   
+   vbuf->vertices = NULL;
+   vbuf->vertex_ptr = vbuf->vertices;
+
+   vbuf->prim = ~0;
+   
+   if(!GETENV("GALLIUM_NOVF"))
+      vbuf->vf = draw_vf_create();
+   
+   return &vbuf->stage;
+}
diff --git a/src/gallium/aux/draw/draw_vbuf.h b/src/gallium/aux/draw/draw_vbuf.h
new file mode 100644
index 0000000000..cfd2b9820c
--- /dev/null
+++ b/src/gallium/aux/draw/draw_vbuf.h
@@ -0,0 +1,106 @@
+/**************************************************************************
+ *
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ **************************************************************************/
+
+/**
+ * \file
+ * Vertex buffer drawing stage.
+ * 
+ * \author Keith Whitwell <keith@tungstengraphics.com>
+ * \author José Fonseca <jrfonsec@tungstengraphics.com>
+ */
+
+#ifndef DRAW_VBUF_H_
+#define DRAW_VBUF_H_
+
+
+#include "pipe/p_util.h"
+
+
+struct draw_context;
+struct vertex_info;
+
+
+/**
+ * Interface for hardware vertex buffer rendering.
+ */
+struct vbuf_render {
+
+   /**
+    * Driver limits.  May be tuned lower to improve cache hits on
+    * index list.
+    */
+   unsigned max_indices;
+   unsigned max_vertex_buffer_bytes;
+
+   /**
+    * Get the hardware vertex format.
+    * 
+    * XXX: have this in draw_context instead?
+    */ 
+   const struct vertex_info *(*get_vertex_info)( struct vbuf_render * );
+	 
+   /**
+    * Request a destination for vertices.
+    * Hardware renderers will use ttm memory, others will just malloc
+    * something.
+    */
+   void *(*allocate_vertices)( struct vbuf_render *,
+			       ushort vertex_size,
+			       ushort nr_vertices );
+
+   /**
+    * Notify the renderer of the current primitive when it changes.
+    * Prim is restricted to TRIANGLES, LINES and POINTS.
+    */
+   void (*set_primitive)( struct vbuf_render *, unsigned prim );
+
+   /**
+    * DrawElements, note indices are ushort:
+    */
+   void (*draw)( struct vbuf_render *,
+		 const ushort *indices,
+		 uint nr_indices );
+
+   /**
+    * Called when vbuf is done with this set of vertices:
+    */
+   void (*release_vertices)( struct vbuf_render *,
+			     void *vertices, 
+			     unsigned vertex_size,
+			     unsigned vertices_used );
+
+   void (*destroy)( struct vbuf_render * );
+};
+
+
+
+struct draw_stage *
+draw_vbuf_stage( struct draw_context *draw,
+                 struct vbuf_render *render );
+
+
+#endif /*DRAW_VBUF_H_*/
diff --git a/src/gallium/aux/draw/draw_vertex.c b/src/gallium/aux/draw/draw_vertex.c
new file mode 100644
index 0000000000..2d6592150f
--- /dev/null
+++ b/src/gallium/aux/draw/draw_vertex.c
@@ -0,0 +1,79 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+/*
+ * Functions for specifying the post-transformation vertex layout.
+ *
+ * Author:
+ *    Brian Paul
+ *    Keith Whitwell
+ */
+
+
+#include "pipe/draw/draw_private.h"
+#include "pipe/draw/draw_vertex.h"
+
+
+/**
+ * Compute the size of a vertex, in dwords/floats, to update the
+ * vinfo->size field.
+ */
+void
+draw_compute_vertex_size(struct vertex_info *vinfo)
+{
+   uint i;
+
+   vinfo->size = 0;
+   for (i = 0; i < vinfo->num_attribs; i++) {
+      switch (vinfo->emit[i]) {
+      case EMIT_OMIT:
+         break;
+      case EMIT_4UB:
+         /* fall-through */
+      case EMIT_1F_PSIZE:
+         /* fall-through */
+      case EMIT_1F:
+         vinfo->size += 1;
+         break;
+      case EMIT_2F:
+         vinfo->size += 2;
+         break;
+      case EMIT_3F:
+         vinfo->size += 3;
+         break;
+      case EMIT_4F:
+         vinfo->size += 4;
+         break;
+      case EMIT_ALL:
+         /* fall-through */
+      default:
+         assert(0);
+      }
+   }
+
+   assert(vinfo->size * 4 <= MAX_VERTEX_SIZE);
+}
diff --git a/src/gallium/aux/draw/draw_vertex.h b/src/gallium/aux/draw/draw_vertex.h
new file mode 100644
index 0000000000..267c74203b
--- /dev/null
+++ b/src/gallium/aux/draw/draw_vertex.h
@@ -0,0 +1,111 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+/**
+ * Post-transform vertex format info.  The vertex_info struct is used by
+ * the draw_vbuf code to emit hardware-specific vertex layouts into hw
+ * vertex buffers.
+ *
+ * Author:
+ *    Brian Paul
+ */
+
+
+#ifndef DRAW_VERTEX_H
+#define DRAW_VERTEX_H
+
+
+#include "pipe/p_state.h"
+
+
+/**
+ * Vertex attribute emit modes
+ */
+enum attrib_emit {
+   EMIT_OMIT,      /**< don't emit the attribute */
+   EMIT_ALL,       /**< emit whole post-xform vertex, w/ header */
+   EMIT_1F,
+   EMIT_1F_PSIZE,  /**< insert constant point size */
+   EMIT_2F,
+   EMIT_3F,
+   EMIT_4F,
+   EMIT_4UB  /**< XXX may need variations for RGBA vs BGRA, etc */
+};
+
+
+/**
+ * Attribute interpolation mode
+ */
+enum interp_mode {
+   INTERP_NONE,      /**< never interpolate vertex header info */
+   INTERP_POS,       /**< special case for frag position */
+   INTERP_CONSTANT,
+   INTERP_LINEAR,
+   INTERP_PERSPECTIVE
+};
+
+
+/**
+ * Information about hardware/rasterization vertex layout.
+ */
+struct vertex_info
+{
+   uint num_attribs;
+   uint hwfmt[4];      /**< hardware format info for this format */
+   enum interp_mode interp_mode[PIPE_MAX_SHADER_INPUTS];
+   enum attrib_emit emit[PIPE_MAX_SHADER_INPUTS];   /**< EMIT_x */
+   uint src_index[PIPE_MAX_SHADER_INPUTS]; /**< map to post-xform attribs */
+   uint size;          /**< total vertex size in dwords */
+};
+
+
+
+/**
+ * Add another attribute to the given vertex_info object.
+ * \param src_index  indicates which post-transformed vertex attrib slot
+ *                   corresponds to this attribute.
+ * \return slot in which the attribute was added
+ */
+static INLINE uint
+draw_emit_vertex_attr(struct vertex_info *vinfo,
+                      enum attrib_emit emit, enum interp_mode interp,
+                      uint src_index)
+{
+   const uint n = vinfo->num_attribs;
+   assert(n < PIPE_MAX_SHADER_INPUTS);
+   vinfo->emit[n] = emit;
+   vinfo->interp_mode[n] = interp;
+   vinfo->src_index[n] = src_index;
+   vinfo->num_attribs++;
+   return n;
+}
+
+
+extern void draw_compute_vertex_size(struct vertex_info *vinfo);
+
+
+#endif /* DRAW_VERTEX_H */
diff --git a/src/gallium/aux/draw/draw_vertex_cache.c b/src/gallium/aux/draw/draw_vertex_cache.c
new file mode 100644
index 0000000000..44427999cc
--- /dev/null
+++ b/src/gallium/aux/draw/draw_vertex_cache.c
@@ -0,0 +1,196 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+ /*
+  * Authors:
+  *   Keith Whitwell <keith@tungstengraphics.com>
+  */
+
+#include "pipe/p_util.h"
+#include "draw_private.h"
+#include "draw_context.h"
+
+
+void draw_vertex_cache_invalidate( struct draw_context *draw )
+{
+   assert(draw->pq.queue_nr == 0);
+   assert(draw->vs.queue_nr == 0);
+   assert(draw->vcache.referenced == 0);
+
+   memset(draw->vcache.idx, ~0, sizeof(draw->vcache.idx));
+}
+
+
+/**
+ * Check if vertex is in cache, otherwise add it.  It won't go through
+ * VS yet, not until there is a flush operation or the VS queue fills up.  
+ *
+ * Note that cache entries are basically just two pointers: the first
+ * an index into the user's vertex arrays, the second a location in
+ * the vertex shader cache for the post-transformed vertex.
+ *
+ * \return pointer to location of (post-transformed) vertex header in the cache
+ */
+static struct vertex_header *get_vertex( struct draw_context *draw,
+					 unsigned i )
+{
+   unsigned slot = (i + (i>>5)) % VCACHE_SIZE;
+   
+   assert(slot < 32); /* so we don't exceed the bitfield size below */
+
+   /* Cache miss?
+    */
+   if (draw->vcache.idx[slot] != i) {
+
+      /* If slot is in use, use the overflow area:
+       */
+      if (draw->vcache.referenced & (1 << slot)) {
+	 slot = VCACHE_SIZE + draw->vcache.overflow++;
+      }
+
+      assert(slot < Elements(draw->vcache.idx));
+
+      draw->vcache.idx[slot] = i;
+
+      /* Add to vertex shader queue:
+       */
+      assert(draw->vs.queue_nr < VS_QUEUE_LENGTH);
+      draw->vs.queue[draw->vs.queue_nr].dest = draw->vcache.vertex[slot];
+      draw->vs.queue[draw->vs.queue_nr].elt = i;
+      draw->vs.queue_nr++;
+
+      /* Need to set the vertex's edge flag here.  If we're being called
+       * by do_ef_triangle(), that function needs edge flag info!
+       */
+      draw->vcache.vertex[slot]->clipmask = 0;
+      draw->vcache.vertex[slot]->edgeflag = 1; /*XXX use user's edge flag! */
+      draw->vcache.vertex[slot]->pad = 0;
+      draw->vcache.vertex[slot]->vertex_id = UNDEFINED_VERTEX_ID;
+   }
+
+
+   /* primitive flushing may have cleared the bitfield but did not
+    * clear the idx[] array values.  Set the bit now.  This fixes a
+    * bug found when drawing long triangle fans.
+    */
+   draw->vcache.referenced |= (1 << slot);
+   return draw->vcache.vertex[slot];
+}
+
+
+static struct vertex_header *get_uint_elt_vertex( struct draw_context *draw,
+                                                  unsigned i )
+{
+   const unsigned *elts = (const unsigned *) draw->user.elts;
+   return get_vertex( draw, elts[i] );
+}
+
+
+static struct vertex_header *get_ushort_elt_vertex( struct draw_context *draw,
+						    unsigned i )
+{
+   const ushort *elts = (const ushort *) draw->user.elts;
+   return get_vertex( draw, elts[i] );
+}
+
+
+static struct vertex_header *get_ubyte_elt_vertex( struct draw_context *draw,
+                                                   unsigned i )
+{
+   const ubyte *elts = (const ubyte *) draw->user.elts;
+   return get_vertex( draw, elts[i] );
+}
+
+
+void draw_vertex_cache_reset_vertex_ids( struct draw_context *draw )
+{
+   unsigned i;
+
+   for (i = 0; i < Elements(draw->vcache.vertex); i++)
+      draw->vcache.vertex[i]->vertex_id = UNDEFINED_VERTEX_ID;
+}
+
+
+void draw_vertex_cache_unreference( struct draw_context *draw )
+{
+   draw->vcache.referenced = 0;
+   draw->vcache.overflow = 0;
+}
+
+
+int draw_vertex_cache_check_space( struct draw_context *draw,
+				   unsigned nr_verts )
+{
+   if (draw->vcache.overflow + nr_verts < VCACHE_OVERFLOW) {
+      /* The vs queue is sized so that this can never happen:
+       */
+      assert(draw->vs.queue_nr + nr_verts < VS_QUEUE_LENGTH);
+      return TRUE;
+   }
+   else
+      return FALSE;
+}
+
+
+
+/**
+ * Tell the drawing context about the index/element buffer to use
+ * (ala glDrawElements)
+ * If no element buffer is to be used (i.e. glDrawArrays) then this
+ * should be called with eltSize=0 and elements=NULL.
+ *
+ * \param draw  the drawing context
+ * \param eltSize  size of each element (1, 2 or 4 bytes)
+ * \param elements  the element buffer ptr
+ */
+void
+draw_set_mapped_element_buffer( struct draw_context *draw,
+                                unsigned eltSize, void *elements )
+{
+//   draw_statechange( draw );
+
+   /* choose the get_vertex() function to use */
+   switch (eltSize) {
+   case 0:
+      draw->vcache.get_vertex = get_vertex;
+      break;
+   case 1:
+      draw->vcache.get_vertex = get_ubyte_elt_vertex;
+      break;
+   case 2:
+      draw->vcache.get_vertex = get_ushort_elt_vertex;
+      break;
+   case 4:
+      draw->vcache.get_vertex = get_uint_elt_vertex;
+      break;
+   default:
+      assert(0);
+   }
+   draw->user.elts = elements;
+   draw->user.eltSize = eltSize;
+}
+
diff --git a/src/gallium/aux/draw/draw_vertex_fetch.c b/src/gallium/aux/draw/draw_vertex_fetch.c
new file mode 100644
index 0000000000..e13df04605
--- /dev/null
+++ b/src/gallium/aux/draw/draw_vertex_fetch.c
@@ -0,0 +1,510 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+ /*
+  * Authors:
+  *   Keith Whitwell <keith@tungstengraphics.com>
+  */
+
+#include "pipe/p_util.h"
+#include "pipe/p_shader_tokens.h"
+#include "draw_private.h"
+#include "draw_context.h"
+
+
+#define DRAW_DBG 0
+
+
+/**
+ * Fetch a float[4] vertex attribute from memory, doing format/type
+ * conversion as needed.
+ *
+ * This is probably needed/dupliocated elsewhere, eg format
+ * conversion, texture sampling etc.
+ */
+#define FETCH_ATTRIB( NAME, SZ, CVT )			\
+static void						\
+fetch_##NAME(const void *ptr, float *attrib)		\
+{							\
+   static const float defaults[4] = { 0,0,0,1 };	\
+   int i;						\
+							\
+   for (i = 0; i < SZ; i++) {				\
+      attrib[i] = CVT;					\
+   }							\
+							\
+   for (; i < 4; i++) {					\
+      attrib[i] = defaults[i];				\
+   }							\
+}
+
+#define CVT_64_FLOAT   (float) ((double *) ptr)[i]
+#define CVT_32_FLOAT   ((float *) ptr)[i]
+
+#define CVT_8_USCALED  (float) ((unsigned char *) ptr)[i]
+#define CVT_16_USCALED (float) ((unsigned short *) ptr)[i]
+#define CVT_32_USCALED (float) ((unsigned int *) ptr)[i]
+
+#define CVT_8_SSCALED  (float) ((char *) ptr)[i]
+#define CVT_16_SSCALED (float) ((short *) ptr)[i]
+#define CVT_32_SSCALED (float) ((int *) ptr)[i]
+
+#define CVT_8_UNORM    (float) ((unsigned char *) ptr)[i] / 255.0f
+#define CVT_16_UNORM   (float) ((unsigned short *) ptr)[i] / 65535.0f
+#define CVT_32_UNORM   (float) ((unsigned int *) ptr)[i] / 4294967295.0f
+
+#define CVT_8_SNORM    (float) ((char *) ptr)[i] / 127.0f
+#define CVT_16_SNORM   (float) ((short *) ptr)[i] / 32767.0f
+#define CVT_32_SNORM   (float) ((int *) ptr)[i] / 2147483647.0f
+
+FETCH_ATTRIB( R64G64B64A64_FLOAT,   4, CVT_64_FLOAT )
+FETCH_ATTRIB( R64G64B64_FLOAT,      3, CVT_64_FLOAT )
+FETCH_ATTRIB( R64G64_FLOAT,         2, CVT_64_FLOAT )
+FETCH_ATTRIB( R64_FLOAT,            1, CVT_64_FLOAT )
+
+FETCH_ATTRIB( R32G32B32A32_FLOAT,   4, CVT_32_FLOAT )
+FETCH_ATTRIB( R32G32B32_FLOAT,      3, CVT_32_FLOAT )
+FETCH_ATTRIB( R32G32_FLOAT,         2, CVT_32_FLOAT )
+FETCH_ATTRIB( R32_FLOAT,            1, CVT_32_FLOAT )
+
+FETCH_ATTRIB( R32G32B32A32_USCALED, 4, CVT_32_USCALED )
+FETCH_ATTRIB( R32G32B32_USCALED,    3, CVT_32_USCALED )
+FETCH_ATTRIB( R32G32_USCALED,       2, CVT_32_USCALED )
+FETCH_ATTRIB( R32_USCALED,          1, CVT_32_USCALED )
+
+FETCH_ATTRIB( R32G32B32A32_SSCALED, 4, CVT_32_SSCALED )
+FETCH_ATTRIB( R32G32B32_SSCALED,    3, CVT_32_SSCALED )
+FETCH_ATTRIB( R32G32_SSCALED,       2, CVT_32_SSCALED )
+FETCH_ATTRIB( R32_SSCALED,          1, CVT_32_SSCALED )
+
+FETCH_ATTRIB( R32G32B32A32_UNORM, 4, CVT_32_UNORM )
+FETCH_ATTRIB( R32G32B32_UNORM,    3, CVT_32_UNORM )
+FETCH_ATTRIB( R32G32_UNORM,       2, CVT_32_UNORM )
+FETCH_ATTRIB( R32_UNORM,          1, CVT_32_UNORM )
+
+FETCH_ATTRIB( R32G32B32A32_SNORM, 4, CVT_32_SNORM )
+FETCH_ATTRIB( R32G32B32_SNORM,    3, CVT_32_SNORM )
+FETCH_ATTRIB( R32G32_SNORM,       2, CVT_32_SNORM )
+FETCH_ATTRIB( R32_SNORM,          1, CVT_32_SNORM )
+
+FETCH_ATTRIB( R16G16B16A16_USCALED, 4, CVT_16_USCALED )
+FETCH_ATTRIB( R16G16B16_USCALED,    3, CVT_16_USCALED )
+FETCH_ATTRIB( R16G16_USCALED,       2, CVT_16_USCALED )
+FETCH_ATTRIB( R16_USCALED,          1, CVT_16_USCALED )
+
+FETCH_ATTRIB( R16G16B16A16_SSCALED, 4, CVT_16_SSCALED )
+FETCH_ATTRIB( R16G16B16_SSCALED,    3, CVT_16_SSCALED )
+FETCH_ATTRIB( R16G16_SSCALED,       2, CVT_16_SSCALED )
+FETCH_ATTRIB( R16_SSCALED,          1, CVT_16_SSCALED )
+
+FETCH_ATTRIB( R16G16B16A16_UNORM, 4, CVT_16_UNORM )
+FETCH_ATTRIB( R16G16B16_UNORM,    3, CVT_16_UNORM )
+FETCH_ATTRIB( R16G16_UNORM,       2, CVT_16_UNORM )
+FETCH_ATTRIB( R16_UNORM,          1, CVT_16_UNORM )
+
+FETCH_ATTRIB( R16G16B16A16_SNORM, 4, CVT_16_SNORM )
+FETCH_ATTRIB( R16G16B16_SNORM,    3, CVT_16_SNORM )
+FETCH_ATTRIB( R16G16_SNORM,       2, CVT_16_SNORM )
+FETCH_ATTRIB( R16_SNORM,          1, CVT_16_SNORM )
+
+FETCH_ATTRIB( R8G8B8A8_USCALED,   4, CVT_8_USCALED )
+FETCH_ATTRIB( R8G8B8_USCALED,     3, CVT_8_USCALED )
+FETCH_ATTRIB( R8G8_USCALED,       2, CVT_8_USCALED )
+FETCH_ATTRIB( R8_USCALED,         1, CVT_8_USCALED )
+
+FETCH_ATTRIB( R8G8B8A8_SSCALED,  4, CVT_8_SSCALED )
+FETCH_ATTRIB( R8G8B8_SSCALED,    3, CVT_8_SSCALED )
+FETCH_ATTRIB( R8G8_SSCALED,      2, CVT_8_SSCALED )
+FETCH_ATTRIB( R8_SSCALED,        1, CVT_8_SSCALED )
+
+FETCH_ATTRIB( R8G8B8A8_UNORM,  4, CVT_8_UNORM )
+FETCH_ATTRIB( R8G8B8_UNORM,    3, CVT_8_UNORM )
+FETCH_ATTRIB( R8G8_UNORM,      2, CVT_8_UNORM )
+FETCH_ATTRIB( R8_UNORM,        1, CVT_8_UNORM )
+
+FETCH_ATTRIB( R8G8B8A8_SNORM,  4, CVT_8_SNORM )
+FETCH_ATTRIB( R8G8B8_SNORM,    3, CVT_8_SNORM )
+FETCH_ATTRIB( R8G8_SNORM,      2, CVT_8_SNORM )
+FETCH_ATTRIB( R8_SNORM,        1, CVT_8_SNORM )
+
+FETCH_ATTRIB( A8R8G8B8_UNORM,       4, CVT_8_UNORM )
+//FETCH_ATTRIB( R8G8B8A8_UNORM,       4, CVT_8_UNORM )
+
+
+
+static fetch_func get_fetch_func( enum pipe_format format )
+{
+#if 0
+   {
+      char tmp[80];
+      pf_sprint_name(tmp, format);
+      debug_printf("%s: %s\n", __FUNCTION__, tmp);
+   }
+#endif
+
+   switch (format) {
+   case PIPE_FORMAT_R64_FLOAT:
+      return fetch_R64_FLOAT;
+   case PIPE_FORMAT_R64G64_FLOAT:
+      return fetch_R64G64_FLOAT;
+   case PIPE_FORMAT_R64G64B64_FLOAT:
+      return fetch_R64G64B64_FLOAT;
+   case PIPE_FORMAT_R64G64B64A64_FLOAT:
+      return fetch_R64G64B64A64_FLOAT;
+
+   case PIPE_FORMAT_R32_FLOAT:
+      return fetch_R32_FLOAT;
+   case PIPE_FORMAT_R32G32_FLOAT:
+      return fetch_R32G32_FLOAT;
+   case PIPE_FORMAT_R32G32B32_FLOAT:
+      return fetch_R32G32B32_FLOAT;
+   case PIPE_FORMAT_R32G32B32A32_FLOAT:
+      return fetch_R32G32B32A32_FLOAT;
+
+   case PIPE_FORMAT_R32_UNORM:
+      return fetch_R32_UNORM;
+   case PIPE_FORMAT_R32G32_UNORM:
+      return fetch_R32G32_UNORM;
+   case PIPE_FORMAT_R32G32B32_UNORM:
+      return fetch_R32G32B32_UNORM;
+   case PIPE_FORMAT_R32G32B32A32_UNORM:
+      return fetch_R32G32B32A32_UNORM;
+
+   case PIPE_FORMAT_R32_USCALED:
+      return fetch_R32_USCALED;
+   case PIPE_FORMAT_R32G32_USCALED:
+      return fetch_R32G32_USCALED;
+   case PIPE_FORMAT_R32G32B32_USCALED:
+      return fetch_R32G32B32_USCALED;
+   case PIPE_FORMAT_R32G32B32A32_USCALED:
+      return fetch_R32G32B32A32_USCALED;
+
+   case PIPE_FORMAT_R32_SNORM:
+      return fetch_R32_SNORM;
+   case PIPE_FORMAT_R32G32_SNORM:
+      return fetch_R32G32_SNORM;
+   case PIPE_FORMAT_R32G32B32_SNORM:
+      return fetch_R32G32B32_SNORM;
+   case PIPE_FORMAT_R32G32B32A32_SNORM:
+      return fetch_R32G32B32A32_SNORM;
+
+   case PIPE_FORMAT_R32_SSCALED:
+      return fetch_R32_SSCALED;
+   case PIPE_FORMAT_R32G32_SSCALED:
+      return fetch_R32G32_SSCALED;
+   case PIPE_FORMAT_R32G32B32_SSCALED:
+      return fetch_R32G32B32_SSCALED;
+   case PIPE_FORMAT_R32G32B32A32_SSCALED:
+      return fetch_R32G32B32A32_SSCALED;
+
+   case PIPE_FORMAT_R16_UNORM:
+      return fetch_R16_UNORM;
+   case PIPE_FORMAT_R16G16_UNORM:
+      return fetch_R16G16_UNORM;
+   case PIPE_FORMAT_R16G16B16_UNORM:
+      return fetch_R16G16B16_UNORM;
+   case PIPE_FORMAT_R16G16B16A16_UNORM:
+      return fetch_R16G16B16A16_UNORM;
+
+   case PIPE_FORMAT_R16_USCALED:
+      return fetch_R16_USCALED;
+   case PIPE_FORMAT_R16G16_USCALED:
+      return fetch_R16G16_USCALED;
+   case PIPE_FORMAT_R16G16B16_USCALED:
+      return fetch_R16G16B16_USCALED;
+   case PIPE_FORMAT_R16G16B16A16_USCALED:
+      return fetch_R16G16B16A16_USCALED;
+
+   case PIPE_FORMAT_R16_SNORM:
+      return fetch_R16_SNORM;
+   case PIPE_FORMAT_R16G16_SNORM:
+      return fetch_R16G16_SNORM;
+   case PIPE_FORMAT_R16G16B16_SNORM:
+      return fetch_R16G16B16_SNORM;
+   case PIPE_FORMAT_R16G16B16A16_SNORM:
+      return fetch_R16G16B16A16_SNORM;
+
+   case PIPE_FORMAT_R16_SSCALED:
+      return fetch_R16_SSCALED;
+   case PIPE_FORMAT_R16G16_SSCALED:
+      return fetch_R16G16_SSCALED;
+   case PIPE_FORMAT_R16G16B16_SSCALED:
+      return fetch_R16G16B16_SSCALED;
+   case PIPE_FORMAT_R16G16B16A16_SSCALED:
+      return fetch_R16G16B16A16_SSCALED;
+
+   case PIPE_FORMAT_R8_UNORM:
+      return fetch_R8_UNORM;
+   case PIPE_FORMAT_R8G8_UNORM:
+      return fetch_R8G8_UNORM;
+   case PIPE_FORMAT_R8G8B8_UNORM:
+      return fetch_R8G8B8_UNORM;
+   case PIPE_FORMAT_R8G8B8A8_UNORM:
+      return fetch_R8G8B8A8_UNORM;
+
+   case PIPE_FORMAT_R8_USCALED:
+      return fetch_R8_USCALED;
+   case PIPE_FORMAT_R8G8_USCALED:
+      return fetch_R8G8_USCALED;
+   case PIPE_FORMAT_R8G8B8_USCALED:
+      return fetch_R8G8B8_USCALED;
+   case PIPE_FORMAT_R8G8B8A8_USCALED:
+      return fetch_R8G8B8A8_USCALED;
+
+   case PIPE_FORMAT_R8_SNORM:
+      return fetch_R8_SNORM;
+   case PIPE_FORMAT_R8G8_SNORM:
+      return fetch_R8G8_SNORM;
+   case PIPE_FORMAT_R8G8B8_SNORM:
+      return fetch_R8G8B8_SNORM;
+   case PIPE_FORMAT_R8G8B8A8_SNORM:
+      return fetch_R8G8B8A8_SNORM;
+
+   case PIPE_FORMAT_R8_SSCALED:
+      return fetch_R8_SSCALED;
+   case PIPE_FORMAT_R8G8_SSCALED:
+      return fetch_R8G8_SSCALED;
+   case PIPE_FORMAT_R8G8B8_SSCALED:
+      return fetch_R8G8B8_SSCALED;
+   case PIPE_FORMAT_R8G8B8A8_SSCALED:
+      return fetch_R8G8B8A8_SSCALED;
+
+   case PIPE_FORMAT_A8R8G8B8_UNORM:
+      return fetch_A8R8G8B8_UNORM;
+
+   case 0:
+      return NULL;		/* not sure why this is needed */
+
+   default:
+      assert(0);
+      return NULL;
+   }
+}
+
+
+static void 
+transpose_4x4( float *out, const float *in )
+{
+   /* This can be achieved in 12 sse instructions, plus the final
+    * stores I guess.  This is probably a bit more than that - maybe
+    * 32 or so?
+    */
+   out[0] = in[0];  out[1] = in[4];  out[2] = in[8];   out[3] = in[12];
+   out[4] = in[1];  out[5] = in[5];  out[6] = in[9];   out[7] = in[13];
+   out[8] = in[2];  out[9] = in[6];  out[10] = in[10]; out[11] = in[14];
+   out[12] = in[3]; out[13] = in[7]; out[14] = in[11]; out[15] = in[15];
+}
+
+
+
+static void fetch_xyz_rgb( struct draw_context *draw,
+			   struct tgsi_exec_machine *machine,
+			   const unsigned *elts,
+			   unsigned count )
+{
+   const unsigned *pitch   = draw->vertex_fetch.pitch;
+   const ubyte **src       = draw->vertex_fetch.src_ptr;
+   int i;
+
+   assert(count <= 4);
+
+//   debug_printf("%s\n", __FUNCTION__);
+
+   /* loop over vertex attributes (vertex shader inputs)
+    */
+
+   for (i = 0; i < 4; i++) {
+      {
+	 const float *in = (const float *)(src[0] + elts[i] * pitch[0]);
+	 float *out = &machine->Inputs[0].xyzw[0].f[i];
+	 out[0] = in[0];
+	 out[4] = in[1];
+	 out[8] = in[2];
+ 	 out[12] = 1.0f;
+      }
+
+      {
+	 const float *in = (const float *)(src[1] + elts[i] * pitch[1]);
+	 float *out = &machine->Inputs[1].xyzw[0].f[i];
+	 out[0] = in[0];
+	 out[4] = in[1];
+	 out[8] = in[2];
+ 	 out[12] = 1.0f;
+      }
+   }
+}
+
+
+
+
+static void fetch_xyz_rgb_st( struct draw_context *draw,
+			      struct tgsi_exec_machine *machine,
+			      const unsigned *elts,
+			      unsigned count )
+{
+   const unsigned *pitch   = draw->vertex_fetch.pitch;
+   const ubyte **src       = draw->vertex_fetch.src_ptr;
+   int i;
+
+   assert(count <= 4);
+
+   /* loop over vertex attributes (vertex shader inputs)
+    */
+
+   for (i = 0; i < 4; i++) {
+      {
+	 const float *in = (const float *)(src[0] + elts[i] * pitch[0]);
+	 float *out = &machine->Inputs[0].xyzw[0].f[i];
+	 out[0] = in[0];
+	 out[4] = in[1];
+	 out[8] = in[2];
+ 	 out[12] = 1.0f;
+      }
+
+      {
+	 const float *in = (const float *)(src[1] + elts[i] * pitch[1]);
+	 float *out = &machine->Inputs[1].xyzw[0].f[i];
+	 out[0] = in[0];
+	 out[4] = in[1];
+	 out[8] = in[2];
+ 	 out[12] = 1.0f;
+      }
+
+      {
+	 const float *in = (const float *)(src[2] + elts[i] * pitch[2]);
+	 float *out = &machine->Inputs[2].xyzw[0].f[i];
+	 out[0] = in[0];
+	 out[4] = in[1];
+	 out[8] = 0.0f;
+ 	 out[12] = 1.0f;
+      }
+   }
+}
+
+
+
+
+/**
+ * Fetch vertex attributes for 'count' vertices.
+ */
+static void generic_vertex_fetch( struct draw_context *draw,
+				  struct tgsi_exec_machine *machine,
+				  const unsigned *elts,
+				  unsigned count )
+{
+   unsigned nr_attrs = draw->vertex_fetch.nr_attrs;
+   unsigned attr;
+
+   assert(count <= 4);
+
+//   debug_printf("%s %d\n", __FUNCTION__, count);
+
+   /* loop over vertex attributes (vertex shader inputs)
+    */
+   for (attr = 0; attr < nr_attrs; attr++) {
+
+      const unsigned pitch   = draw->vertex_fetch.pitch[attr];
+      const ubyte *src = draw->vertex_fetch.src_ptr[attr];
+      const fetch_func fetch = draw->vertex_fetch.fetch[attr];
+      unsigned i;
+      float p[4][4];
+
+
+      /* Fetch four attributes for four vertices.  
+       * 
+       * Could fetch directly into AOS format, but this is meant to be
+       * a prototype for an sse implementation, which would have
+       * difficulties doing that.
+       */
+      for (i = 0; i < count; i++) 
+	 fetch( src + elts[i] * pitch, p[i] );
+
+      /* Be nice and zero out any missing vertices: 
+       */
+      for ( ; i < 4; i++) 
+	 p[i][0] = p[i][1] = p[i][2] = p[i][3] = 0;
+      
+      /* Transpose/swizzle into sse-friendly format.  Currently
+       * assuming that all vertex shader inputs are float[4], but this
+       * isn't true -- if the vertex shader only wants tex0.xy, we
+       * could optimize for that.
+       *
+       * To do so fully without codegen would probably require an
+       * excessive number of fetch functions, but we could at least
+       * minimize the transpose step:
+       */
+      transpose_4x4( (float *)&machine->Inputs[attr].xyzw[0].f[0], (float *)p );
+   }
+}
+
+
+			       
+void draw_update_vertex_fetch( struct draw_context *draw )
+{
+   unsigned nr_attrs, i;
+
+//   debug_printf("%s\n", __FUNCTION__);
+   
+   /* this may happend during context init */
+   if (!draw->vertex_shader)
+      return;
+
+   nr_attrs = draw->vertex_shader->state->num_inputs;
+
+   for (i = 0; i < nr_attrs; i++) {
+      unsigned buf = draw->vertex_element[i].vertex_buffer_index;
+      enum pipe_format format  = draw->vertex_element[i].src_format;
+
+      draw->vertex_fetch.src_ptr[i] = (const ubyte *) draw->user.vbuffer[buf] + 
+						       draw->vertex_buffer[buf].buffer_offset + 
+						       draw->vertex_element[i].src_offset;
+
+      draw->vertex_fetch.pitch[i] = draw->vertex_buffer[buf].pitch;
+      draw->vertex_fetch.fetch[i] = get_fetch_func( format );
+   }
+
+   draw->vertex_fetch.nr_attrs = nr_attrs;
+
+   draw->vertex_fetch.fetch_func = generic_vertex_fetch;
+
+   switch (nr_attrs) {
+   case 2:
+      if (draw->vertex_element[0].src_format == PIPE_FORMAT_R32G32B32_FLOAT &&
+	  draw->vertex_element[1].src_format == PIPE_FORMAT_R32G32B32_FLOAT)
+	 draw->vertex_fetch.fetch_func = fetch_xyz_rgb;
+      break;
+   case 3:
+      if (draw->vertex_element[0].src_format == PIPE_FORMAT_R32G32B32_FLOAT &&
+	  draw->vertex_element[1].src_format == PIPE_FORMAT_R32G32B32_FLOAT &&
+	  draw->vertex_element[2].src_format == PIPE_FORMAT_R32G32_FLOAT)
+	 draw->vertex_fetch.fetch_func = fetch_xyz_rgb_st;
+      break;
+   default:
+      break;
+   }
+
+}
diff --git a/src/gallium/aux/draw/draw_vertex_shader.c b/src/gallium/aux/draw/draw_vertex_shader.c
new file mode 100644
index 0000000000..c824c1407e
--- /dev/null
+++ b/src/gallium/aux/draw/draw_vertex_shader.c
@@ -0,0 +1,325 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+ /*
+  * Authors:
+  *   Keith Whitwell <keith@tungstengraphics.com>
+  *   Brian Paul
+  */
+
+#include "pipe/p_util.h"
+#include "pipe/p_shader_tokens.h"
+#if defined(__i386__) || defined(__386__)
+#include "pipe/tgsi/exec/tgsi_sse2.h"
+#endif
+#include "draw_private.h"
+#include "draw_context.h"
+
+#include "x86/rtasm/x86sse.h"
+#include "pipe/llvm/gallivm.h"
+
+
+#define DBG_VS 0
+
+
+static INLINE unsigned
+compute_clipmask(const float *clip, /*const*/ float plane[][4], unsigned nr)
+{
+   unsigned mask = 0;
+   unsigned i;
+
+   /* Do the hardwired planes first:
+    */
+   if (-clip[0] + clip[3] < 0) mask |= CLIP_RIGHT_BIT;
+   if ( clip[0] + clip[3] < 0) mask |= CLIP_LEFT_BIT;
+   if (-clip[1] + clip[3] < 0) mask |= CLIP_TOP_BIT;
+   if ( clip[1] + clip[3] < 0) mask |= CLIP_BOTTOM_BIT;
+   if (-clip[2] + clip[3] < 0) mask |= CLIP_FAR_BIT;
+   if ( clip[2] + clip[3] < 0) mask |= CLIP_NEAR_BIT;
+
+   /* Followed by any remaining ones:
+    */
+   for (i = 6; i < nr; i++) {
+      if (dot4(clip, plane[i]) < 0) 
+         mask |= (1<<i);
+   }
+
+   return mask;
+}
+
+
+typedef void (XSTDCALL *codegen_function) (
+   const struct tgsi_exec_vector *input,
+   struct tgsi_exec_vector *output,
+   float (*constant)[4],
+   struct tgsi_exec_vector *temporary );
+
+
+/**
+ * Transform vertices with the current vertex program/shader
+ * Up to four vertices can be shaded at a time.
+ * \param vbuffer  the input vertex data
+ * \param elts  indexes of four input vertices
+ * \param count  number of vertices to shade [1..4]
+ * \param vOut  array of pointers to four output vertices
+ */
+static void
+run_vertex_program(struct draw_context *draw,
+                   unsigned elts[4], unsigned count,
+                   struct vertex_header *vOut[])
+{
+   struct tgsi_exec_machine *machine = &draw->machine;
+   unsigned int j;
+
+   ALIGN16_DECL(struct tgsi_exec_vector, inputs, PIPE_ATTRIB_MAX);
+   ALIGN16_DECL(struct tgsi_exec_vector, outputs, PIPE_ATTRIB_MAX);
+   const float *scale = draw->viewport.scale;
+   const float *trans = draw->viewport.translate;
+
+   assert(count <= 4);
+   assert(draw->vertex_shader->state->output_semantic_name[0]
+          == TGSI_SEMANTIC_POSITION);
+
+   /* Consts does not require 16 byte alignment. */
+   machine->Consts = (float (*)[4]) draw->user.constants;
+
+   machine->Inputs = ALIGN16_ASSIGN(inputs);
+   machine->Outputs = ALIGN16_ASSIGN(outputs);
+
+   draw->vertex_fetch.fetch_func( draw, machine, elts, count );
+
+   /* run shader */
+#ifdef MESA_LLVM
+   if (1) {
+   struct gallivm_prog  *prog  = draw->vertex_shader->llvm_prog;
+   gallivm_cpu_vs_exec(prog,
+                       machine->Inputs,
+                       machine->Outputs,
+                       machine->Consts,
+                       machine->Temps);
+   } else
+#elif defined(__i386__) || defined(__386__)
+   if (draw->use_sse) {
+      /* SSE */
+      /* cast away const */
+      struct draw_vertex_shader *shader
+         = (struct draw_vertex_shader *)draw->vertex_shader;
+      codegen_function func
+         = (codegen_function) x86_get_func( &shader->sse2_program );
+
+      if (func)
+         func(
+            machine->Inputs,
+            machine->Outputs,
+            machine->Consts,
+            machine->Temps );
+      else
+         /* interpreter */
+         tgsi_exec_machine_run( machine );
+   }
+   else
+#endif
+   {
+      /* interpreter */
+      tgsi_exec_machine_run( machine );
+   }
+
+   /* store machine results */
+   for (j = 0; j < count; j++) {
+      unsigned slot;
+      float x, y, z, w;
+
+      /* Handle attr[0] (position) specially:
+       *
+       * XXX: Computing the clipmask should be done in the vertex
+       * program as a set of DP4 instructions appended to the
+       * user-provided code.
+       */
+      x = vOut[j]->clip[0] = machine->Outputs[0].xyzw[0].f[j];
+      y = vOut[j]->clip[1] = machine->Outputs[0].xyzw[1].f[j];
+      z = vOut[j]->clip[2] = machine->Outputs[0].xyzw[2].f[j];
+      w = vOut[j]->clip[3] = machine->Outputs[0].xyzw[3].f[j];
+
+      vOut[j]->clipmask = compute_clipmask(vOut[j]->clip, draw->plane, draw->nr_planes);
+      vOut[j]->edgeflag = 1;
+
+      /* divide by w */
+      w = 1.0f / w;
+      x *= w;
+      y *= w;
+      z *= w;
+
+      /* Viewport mapping */
+      vOut[j]->data[0][0] = x * scale[0] + trans[0];
+      vOut[j]->data[0][1] = y * scale[1] + trans[1];
+      vOut[j]->data[0][2] = z * scale[2] + trans[2];
+      vOut[j]->data[0][3] = w;
+
+#if DBG_VS
+      debug_printf("output[%d]win: %f %f %f %f\n", j,
+             vOut[j]->data[0][0],
+             vOut[j]->data[0][1],
+             vOut[j]->data[0][2],
+             vOut[j]->data[0][3]);
+#endif
+      /* Remaining attributes are packed into sequential post-transform
+       * vertex attrib slots.
+       */
+      for (slot = 1; slot < draw->num_vs_outputs; slot++) {
+         vOut[j]->data[slot][0] = machine->Outputs[slot].xyzw[0].f[j];
+         vOut[j]->data[slot][1] = machine->Outputs[slot].xyzw[1].f[j];
+         vOut[j]->data[slot][2] = machine->Outputs[slot].xyzw[2].f[j];
+         vOut[j]->data[slot][3] = machine->Outputs[slot].xyzw[3].f[j];
+#if DBG_VS
+         debug_printf("output[%d][%d]: %f %f %f %f\n", j, slot,
+                vOut[j]->data[slot][0],
+                vOut[j]->data[slot][1],
+                vOut[j]->data[slot][2],
+                vOut[j]->data[slot][3]);
+#endif
+      }
+   } /* loop over vertices */
+}
+
+
+/**
+ * Run the vertex shader on all vertices in the vertex queue.
+ * Called by the draw module when the vertx cache needs to be flushed.
+ */
+void
+draw_vertex_shader_queue_flush(struct draw_context *draw)
+{
+   unsigned i;
+
+   assert(draw->vs.queue_nr != 0);
+
+   /* XXX: do this on statechange: 
+    */
+   draw_update_vertex_fetch( draw );
+
+//   fprintf(stderr, " q(%d) ", draw->vs.queue_nr );
+
+   /* run vertex shader on vertex cache entries, four per invokation */
+   for (i = 0; i < draw->vs.queue_nr; i += 4) {
+      struct vertex_header *dests[4];
+      unsigned elts[4];
+      int j, n = MIN2(4, draw->vs.queue_nr - i);
+
+      for (j = 0; j < n; j++) {
+         elts[j] = draw->vs.queue[i + j].elt;
+         dests[j] = draw->vs.queue[i + j].dest;
+      }
+
+      for ( ; j < 4; j++) {
+	 elts[j] = elts[0];
+	 dests[j] = dests[0];
+      }
+
+      assert(n > 0);
+      assert(n <= 4);
+
+      run_vertex_program(draw, elts, n, dests);
+   }
+
+   draw->vs.queue_nr = 0;
+}
+
+
+struct draw_vertex_shader *
+draw_create_vertex_shader(struct draw_context *draw,
+                          const struct pipe_shader_state *shader)
+{
+   struct draw_vertex_shader *vs;
+
+   vs = CALLOC_STRUCT( draw_vertex_shader );
+   if (vs == NULL) {
+      return NULL;
+   }
+
+   vs->state = shader;
+
+#ifdef MESA_LLVM
+   struct gallivm_ir *ir = gallivm_ir_new(GALLIVM_VS);
+   gallivm_ir_set_layout(ir, GALLIVM_SOA);
+   gallivm_ir_set_components(ir, 4);
+   gallivm_ir_fill_from_tgsi(ir, shader->tokens);
+   vs->llvm_prog = gallivm_ir_compile(ir);
+   gallivm_ir_delete(ir);
+
+   draw->engine = gallivm_global_cpu_engine();
+   if (!draw->engine) {
+      draw->engine = gallivm_cpu_engine_create(vs->llvm_prog);
+   }
+   else {
+      gallivm_cpu_jit_compile(draw->engine, vs->llvm_prog);
+   }
+#elif defined(__i386__) || defined(__386__)
+   if (draw->use_sse) {
+      /* cast-away const */
+      struct pipe_shader_state *sh = (struct pipe_shader_state *) shader;
+
+      x86_init_func( &vs->sse2_program );
+      if (!tgsi_emit_sse2( (struct tgsi_token *) sh->tokens,
+                           &vs->sse2_program )) {
+         x86_release_func( (struct x86_function *) &vs->sse2_program );
+	 fprintf(stdout /*err*/,
+		 "tgsi_emit_sse2() failed, falling back to interpreter\n");
+      }
+   }
+#endif
+
+   return vs;
+}
+
+
+void
+draw_bind_vertex_shader(struct draw_context *draw,
+                        struct draw_vertex_shader *dvs)
+{
+   draw_do_flush( draw, DRAW_FLUSH_STATE_CHANGE );
+
+   draw->vertex_shader = dvs;
+   draw->num_vs_outputs = dvs->state->num_outputs;
+
+   /* specify the vertex program to interpret/execute */
+   tgsi_exec_machine_init(&draw->machine,
+                          draw->vertex_shader->state->tokens,
+                          PIPE_MAX_SAMPLERS,
+                          NULL /*samplers*/ );
+}
+
+
+void
+draw_delete_vertex_shader(struct draw_context *draw,
+                          struct draw_vertex_shader *dvs)
+{
+#if defined(__i386__) || defined(__386__)
+   x86_release_func( (struct x86_function *) &dvs->sse2_program );
+#endif
+
+   FREE( dvs );
+}
diff --git a/src/gallium/aux/draw/draw_vf.c b/src/gallium/aux/draw/draw_vf.c
new file mode 100644
index 0000000000..dc3a5ecd21
--- /dev/null
+++ b/src/gallium/aux/draw/draw_vf.c
@@ -0,0 +1,428 @@
+/*
+ * Copyright 2003 Tungsten Graphics, inc.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * on the rights to use, copy, modify, merge, publish, distribute, sub
+ * license, and/or sell copies of the Software, and to permit persons to whom
+ * the Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the next
+ * paragraph) shall be included in all copies or substantial portions of the
+ * Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.  IN NO EVENT SHALL
+ * TUNGSTEN GRAPHICS AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
+ * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
+ * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
+ * USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors:
+ *    Keith Whitwell <keithw@tungstengraphics.com>
+ */
+
+
+#include <stddef.h>
+
+#include "pipe/p_compiler.h"
+#include "pipe/p_util.h"
+
+#include "draw_vf.h"
+
+
+#define DRAW_VF_DBG 0
+
+
+/* TODO: remove this */
+extern void 
+_mesa_exec_free( void *addr );
+
+
+static boolean match_fastpath( struct draw_vertex_fetch *vf,
+				 const struct draw_vf_fastpath *fp)
+{
+   unsigned j;
+
+   if (vf->attr_count != fp->attr_count) 
+      return FALSE;
+
+   for (j = 0; j < vf->attr_count; j++) 
+      if (vf->attr[j].format != fp->attr[j].format ||
+	  vf->attr[j].inputsize != fp->attr[j].size ||
+	  vf->attr[j].vertoffset != fp->attr[j].offset) 
+	 return FALSE;
+      
+   if (fp->match_strides) {
+      if (vf->vertex_stride != fp->vertex_stride)
+	 return FALSE;
+
+      for (j = 0; j < vf->attr_count; j++) 
+	 if (vf->attr[j].inputstride != fp->attr[j].stride) 
+	    return FALSE;
+   }
+   
+   return TRUE;
+}
+
+static boolean search_fastpath_emit( struct draw_vertex_fetch *vf )
+{
+   struct draw_vf_fastpath *fp = vf->fastpath;
+
+   for ( ; fp ; fp = fp->next) {
+      if (match_fastpath(vf, fp)) {
+         vf->emit = fp->func;
+	 return TRUE;
+      }
+   }
+
+   return FALSE;
+}
+
+void draw_vf_register_fastpath( struct draw_vertex_fetch *vf,
+			     boolean match_strides )
+{
+   struct draw_vf_fastpath *fastpath = CALLOC_STRUCT(draw_vf_fastpath);
+   unsigned i;
+
+   fastpath->vertex_stride = vf->vertex_stride;
+   fastpath->attr_count = vf->attr_count;
+   fastpath->match_strides = match_strides;
+   fastpath->func = vf->emit;
+   fastpath->attr = (struct draw_vf_attr_type *)
+      MALLOC(vf->attr_count * sizeof(fastpath->attr[0]));
+
+   for (i = 0; i < vf->attr_count; i++) {
+      fastpath->attr[i].format = vf->attr[i].format;
+      fastpath->attr[i].stride = vf->attr[i].inputstride;
+      fastpath->attr[i].size = vf->attr[i].inputsize;
+      fastpath->attr[i].offset = vf->attr[i].vertoffset;
+   }
+
+   fastpath->next = vf->fastpath;
+   vf->fastpath = fastpath;
+}
+
+
+
+
+/***********************************************************************
+ * Build codegen functions or return generic ones:
+ */
+static void choose_emit_func( struct draw_vertex_fetch *vf, 
+			      unsigned count, 
+			      uint8_t *dest)
+{
+   vf->emit = NULL;
+   
+   /* Does this match an existing (hardwired, codegen or known-bad)
+    * fastpath?
+    */
+   if (search_fastpath_emit(vf)) {
+      /* Use this result.  If it is null, then it is already known
+       * that the current state will fail for codegen and there is no
+       * point trying again.
+       */
+   }
+   else if (vf->codegen_emit) {
+      vf->codegen_emit( vf );
+   }
+
+   if (!vf->emit) {
+      draw_vf_generate_hardwired_emit(vf);
+   }
+
+   /* Otherwise use the generic version:
+    */
+   if (!vf->emit)
+      vf->emit = draw_vf_generic_emit;
+
+   vf->emit( vf, count, dest );
+}
+
+
+
+
+
+/***********************************************************************
+ * Public entrypoints, mostly dispatch to the above:
+ */
+
+
+
+static unsigned 
+draw_vf_set_vertex_attributes( struct draw_vertex_fetch *vf, 
+                               const struct draw_vf_attr_map *map,
+                               unsigned nr, 
+                               unsigned vertex_stride )
+{
+   unsigned offset = 0;
+   unsigned i, j;
+
+   assert(nr < PIPE_ATTRIB_MAX);
+
+   for (j = 0, i = 0; i < nr; i++) {
+      const unsigned format = map[i].format;
+      if (format == DRAW_EMIT_PAD) {
+#if (DRAW_VF_DBG)
+	    debug_printf("%d: pad %d, offset %d\n", i,  
+			 map[i].offset, offset);  
+#endif
+
+	 offset += map[i].offset;
+
+      }
+      else {
+	 vf->attr[j].attrib = map[i].attrib;
+	 vf->attr[j].format = format;
+	 vf->attr[j].insert = draw_vf_format_info[format].insert;
+	 vf->attr[j].vertattrsize = draw_vf_format_info[format].attrsize;
+	 vf->attr[j].vertoffset = offset;
+	 vf->attr[j].isconst = draw_vf_format_info[format].isconst;
+	 if(vf->attr[j].isconst)
+	    memcpy(vf->attr[j].data, &map[i].data, vf->attr[j].vertattrsize);
+	 
+#if (DRAW_VF_DBG)
+	    debug_printf("%d: %s, offset %d\n", i,  
+			 draw_vf_format_info[format].name,
+			 vf->attr[j].vertoffset);   
+#endif
+
+	 offset += draw_vf_format_info[format].attrsize;
+	 j++;
+      }
+   }
+
+   vf->attr_count = j;
+   vf->vertex_stride = vertex_stride ? vertex_stride : offset;
+   vf->emit = choose_emit_func;
+
+   assert(vf->vertex_stride >= offset);
+   return vf->vertex_stride;
+}
+
+
+void draw_vf_set_vertex_info( struct draw_vertex_fetch *vf, 
+                              const struct vertex_info *vinfo,
+                              float point_size )
+{
+   unsigned i, j, k;
+   struct draw_vf_attr *a = vf->attr;
+   struct draw_vf_attr_map attrs[PIPE_MAX_SHADER_INPUTS];
+   unsigned count = 0;  /* for debug/sanity */
+   unsigned nr_attrs = 0;
+   
+   for (i = 0; i < vinfo->num_attribs; i++) {
+      j = vinfo->src_index[i];
+      switch (vinfo->emit[i]) {
+      case EMIT_OMIT:
+         /* no-op */
+         break;
+      case EMIT_ALL: {
+         /* just copy the whole vertex as-is to the vbuf */
+	 unsigned s = vinfo->size;
+         assert(i == 0);
+         assert(j == 0);
+         /* copy the vertex header */
+         /* XXX: we actually don't copy the header, just pad it */
+	 attrs[nr_attrs].attrib = 0;
+	 attrs[nr_attrs].format = DRAW_EMIT_PAD;
+	 attrs[nr_attrs].offset = offsetof(struct vertex_header, data);
+	 s -= offsetof(struct vertex_header, data)/4;
+         count += offsetof(struct vertex_header, data)/4;
+	 nr_attrs++;
+	 /* copy the vertex data */
+         for(k = 0; k < (s & ~0x3); k += 4) {
+      	    attrs[nr_attrs].attrib = k/4;
+      	    attrs[nr_attrs].format = DRAW_EMIT_4F;
+      	    attrs[nr_attrs].offset = 0;
+      	    nr_attrs++;
+            count += 4;
+         }
+         /* tail */
+         /* XXX: actually, this shouldn't be needed */
+ 	 attrs[nr_attrs].attrib = k/4;
+  	 attrs[nr_attrs].offset = 0;
+         switch(s & 0x3) {
+         case 0:
+            break;
+         case 1:
+      	    attrs[nr_attrs].format = DRAW_EMIT_1F;
+      	    nr_attrs++;
+            count += 1;
+            break;
+         case 2:
+      	    attrs[nr_attrs].format = DRAW_EMIT_2F;
+      	    nr_attrs++;
+            count += 2;
+            break;
+         case 3:
+      	    attrs[nr_attrs].format = DRAW_EMIT_3F;
+      	    nr_attrs++;
+            count += 3;
+            break;
+         }
+         break;
+      }
+      case EMIT_1F:
+	 attrs[nr_attrs].attrib = j;
+	 attrs[nr_attrs].format = DRAW_EMIT_1F;
+	 attrs[nr_attrs].offset = 0;
+	 nr_attrs++;
+         count++;
+         break;
+      case EMIT_1F_PSIZE:
+	 attrs[nr_attrs].attrib = j;
+	 attrs[nr_attrs].format = DRAW_EMIT_1F_CONST;
+	 attrs[nr_attrs].offset = 0;
+	 attrs[nr_attrs].data.f[0] = point_size;
+	 nr_attrs++;
+         count++;
+         break;
+      case EMIT_2F:
+	 attrs[nr_attrs].attrib = j;
+	 attrs[nr_attrs].format = DRAW_EMIT_2F;
+	 attrs[nr_attrs].offset = 0;
+	 nr_attrs++;
+         count += 2;
+         break;
+      case EMIT_3F:
+	 attrs[nr_attrs].attrib = j;
+	 attrs[nr_attrs].format = DRAW_EMIT_3F;
+	 attrs[nr_attrs].offset = 0;
+	 nr_attrs++;
+         count += 3;
+         break;
+      case EMIT_4F:
+	 attrs[nr_attrs].attrib = j;
+	 attrs[nr_attrs].format = DRAW_EMIT_4F;
+	 attrs[nr_attrs].offset = 0;
+	 nr_attrs++;
+         count += 4;
+         break;
+      case EMIT_4UB:
+	 attrs[nr_attrs].attrib = j;
+	 attrs[nr_attrs].format = DRAW_EMIT_4UB_4F_BGRA;
+	 attrs[nr_attrs].offset = 0;
+	 nr_attrs++;
+         count += 1;
+         break;
+      default:
+         assert(0);
+      }
+   }
+   
+   assert(count == vinfo->size);  
+   
+   draw_vf_set_vertex_attributes(vf, 
+                                 attrs, 
+                                 nr_attrs, 
+                                 vinfo->size * sizeof(float) );
+
+   for (j = 0; j < vf->attr_count; j++) {
+      a[j].inputsize = 4;
+      a[j].do_insert = a[j].insert[4 - 1];
+      if(a[j].isconst) {
+	 a[j].inputptr = a[j].data;
+	 a[j].inputstride = 0;
+      }
+   }
+}
+
+
+#if 0
+/* Set attribute pointers, adjusted for start position:
+ */
+void draw_vf_set_sources( struct draw_vertex_fetch *vf,
+		     GLvector4f * const sources[],
+		     unsigned start )
+{
+   struct draw_vf_attr *a = vf->attr;
+   unsigned j;
+   
+   for (j = 0; j < vf->attr_count; j++) {
+      const GLvector4f *vptr = sources[a[j].attrib];
+      
+      if ((a[j].inputstride != vptr->stride) ||
+	  (a[j].inputsize != vptr->size))
+	 vf->emit = choose_emit_func;
+      
+      a[j].inputstride = vptr->stride;
+      a[j].inputsize = vptr->size;
+      a[j].do_insert = a[j].insert[vptr->size - 1]; 
+      a[j].inputptr = ((uint8_t *)vptr->data) + start * vptr->stride;
+   }
+}
+#endif
+
+
+/**
+ * Emit a vertex to dest.  
+ */
+void draw_vf_emit_vertex( struct draw_vertex_fetch *vf,
+                          struct vertex_header *vertex,
+                          void *dest )
+{
+   struct draw_vf_attr *a = vf->attr;
+   unsigned j;
+   
+   for (j = 0; j < vf->attr_count; j++) {
+      if (!a[j].isconst) {
+	 a[j].inputptr = (uint8_t *)&vertex->data[a[j].attrib][0];
+	 a[j].inputstride = 0; /* XXX: one-vertex-max ATM */
+      }
+   }
+   
+   vf->emit( vf, 1, (uint8_t*) dest );
+}
+
+
+
+struct draw_vertex_fetch *draw_vf_create( void )
+{
+   struct draw_vertex_fetch *vf = CALLOC_STRUCT(draw_vertex_fetch);
+   unsigned i;
+
+   for (i = 0; i < PIPE_ATTRIB_MAX; i++)
+      vf->attr[i].vf = vf;
+
+   vf->identity[0] = 0.0;
+   vf->identity[1] = 0.0;
+   vf->identity[2] = 0.0;
+   vf->identity[3] = 1.0;
+
+   vf->codegen_emit = NULL;
+
+#ifdef USE_SSE_ASM
+   if (!GETENV("GALLIUM_NO_CODEGEN"))
+      vf->codegen_emit = draw_vf_generate_sse_emit;
+#endif
+
+   return vf;
+}
+
+
+void draw_vf_destroy( struct draw_vertex_fetch *vf )
+{
+   struct draw_vf_fastpath *fp, *tmp;
+
+   for (fp = vf->fastpath ; fp ; fp = tmp) {
+      tmp = fp->next;
+      FREE(fp->attr);
+
+      /* KW: At the moment, fp->func is constrained to be allocated by
+       * _mesa_exec_alloc(), as the hardwired fastpaths in
+       * t_vertex_generic.c are handled specially.  It would be nice
+       * to unify them, but this probably won't change until this
+       * module gets another overhaul.
+       */
+      //_mesa_exec_free((void *) fp->func);
+      FREE(fp);
+   }
+   
+   vf->fastpath = NULL;
+   FREE(vf);
+}
diff --git a/src/gallium/aux/draw/draw_vf.h b/src/gallium/aux/draw/draw_vf.h
new file mode 100644
index 0000000000..011c8f0ff1
--- /dev/null
+++ b/src/gallium/aux/draw/draw_vf.h
@@ -0,0 +1,236 @@
+/*
+ * Copyright 2008 Tungsten Graphics, inc.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * on the rights to use, copy, modify, merge, publish, distribute, sub
+ * license, and/or sell copies of the Software, and to permit persons to whom
+ * the Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the next
+ * paragraph) shall be included in all copies or substantial portions of the
+ * Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.  IN NO EVENT SHALL
+ * TUNGSTEN GRAPHICS AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
+ * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
+ * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
+ * USE OR OTHER DEALINGS IN THE SOFTWARE.
+ */
+
+
+/**
+ * Vertex fetch/store/convert code.  This functionality is used in two places:
+ * 1. Vertex fetch/convert - to grab vertex data from incoming vertex
+ *    arrays and convert to format needed by vertex shaders.
+ * 2. Vertex store/emit - to convert simple float[][4] vertex attributes
+ *    (which is the organization used throughout the draw/prim pipeline) to
+ *    hardware-specific formats and emit into hardware vertex buffers.
+ *
+ *
+ * Authors:
+ *    Keith Whitwell <keithw@tungstengraphics.com>
+ */
+
+#ifndef DRAW_VF_H
+#define DRAW_VF_H
+
+
+#include "pipe/p_compiler.h"
+#include "pipe/p_state.h"
+
+#include "draw_vertex.h"
+#include "draw_private.h" /* for vertex_header */
+
+
+enum draw_vf_attr_format {
+   DRAW_EMIT_1F,
+   DRAW_EMIT_2F,
+   DRAW_EMIT_3F,
+   DRAW_EMIT_4F,
+   DRAW_EMIT_3F_XYW,			/**< for projective texture */
+   DRAW_EMIT_1UB_1F,			/**< for fog coordinate */
+   DRAW_EMIT_3UB_3F_RGB,		/**< for specular color */
+   DRAW_EMIT_3UB_3F_BGR,		/**< for specular color */
+   DRAW_EMIT_4UB_4F_RGBA,		/**< for color */
+   DRAW_EMIT_4UB_4F_BGRA,		/**< for color */
+   DRAW_EMIT_4UB_4F_ARGB,		/**< for color */
+   DRAW_EMIT_4UB_4F_ABGR,		/**< for color */
+   DRAW_EMIT_1F_CONST,
+   DRAW_EMIT_2F_CONST,
+   DRAW_EMIT_3F_CONST,
+   DRAW_EMIT_4F_CONST,
+   DRAW_EMIT_PAD,			/**< leave a hole of 'offset' bytes */
+   DRAW_EMIT_MAX
+};
+
+struct draw_vf_attr_map 
+{
+   /** Input attribute number */
+   unsigned attrib;
+   
+   enum draw_vf_attr_format format;
+   
+   unsigned offset;
+   
+   /** 
+    * Constant data for DRAW_EMIT_*_CONST 
+    */
+   union {
+      uint8_t ub[4];
+      float f[4];
+   } data;
+};
+
+struct draw_vertex_fetch;
+
+
+
+#if 0
+unsigned 
+draw_vf_set_vertex_attributes( struct draw_vertex_fetch *vf,
+                               const struct draw_vf_attr_map *map,
+                               unsigned nr, 
+                               unsigned vertex_stride );
+#endif
+
+void draw_vf_set_vertex_info( struct draw_vertex_fetch *vf, 
+                              const struct vertex_info *vinfo,
+                              float point_size );
+
+#if 0
+void 
+draw_vf_set_sources( struct draw_vertex_fetch *vf,
+		     GLvector4f * const attrib[],
+		     unsigned start );
+#endif
+
+void 
+draw_vf_emit_vertex( struct draw_vertex_fetch *vf,
+                     struct vertex_header *vertex,
+                     void *dest );
+
+struct draw_vertex_fetch *
+draw_vf_create( void );
+
+void 
+draw_vf_destroy( struct draw_vertex_fetch *vf );
+
+
+
+/***********************************************************************
+ * Internal functions and structs:
+ */
+
+struct draw_vf_attr;
+
+typedef void (*draw_vf_extract_func)( const struct draw_vf_attr *a, 
+				      float *out, 
+				      const uint8_t *v );
+
+typedef void (*draw_vf_insert_func)( const struct draw_vf_attr *a, 
+				     uint8_t *v, 
+				     const float *in );
+
+typedef void (*draw_vf_emit_func)( struct draw_vertex_fetch *vf,
+      				   unsigned count, 
+      				   uint8_t *dest );
+
+
+
+/**
+ * Describes how to convert/move a vertex attribute from a vertex
+ * array to a vertex structure.
+ */
+struct draw_vf_attr
+{
+   struct draw_vertex_fetch *vf;
+
+   unsigned format;
+   unsigned inputsize;
+   unsigned inputstride;
+   unsigned vertoffset;      /**< position of the attrib in the vertex struct */
+   
+   boolean isconst;              /**< read from const data below */
+   uint8_t data[16];
+
+   unsigned attrib;          /**< which vertex attrib (0=position, etc) */
+   unsigned vertattrsize;    /**< size of the attribute in bytes */
+
+   uint8_t *inputptr;
+   const draw_vf_insert_func *insert;
+   draw_vf_insert_func do_insert;
+   draw_vf_extract_func extract;
+};
+
+struct draw_vertex_fetch
+{
+   struct draw_vf_attr attr[PIPE_ATTRIB_MAX];
+   unsigned attr_count;
+   unsigned vertex_stride;
+
+   draw_vf_emit_func emit;
+
+   /* Parameters and constants for codegen:
+    */
+   float identity[4];
+
+   struct draw_vf_fastpath *fastpath;
+   
+   void (*codegen_emit)( struct draw_vertex_fetch *vf );
+};
+
+
+struct draw_vf_attr_type {
+   unsigned format;
+   unsigned size;
+   unsigned stride;
+   unsigned offset;
+};
+
+/** XXX this could be moved into draw_vf.c */
+struct draw_vf_fastpath {
+   unsigned vertex_stride;
+   unsigned attr_count;
+   boolean match_strides;
+
+   struct draw_vf_attr_type *attr;
+
+   draw_vf_emit_func func;
+   struct draw_vf_fastpath *next;
+};
+
+
+void 
+draw_vf_register_fastpath( struct draw_vertex_fetch *vtx,
+                           boolean match_strides );
+
+void 
+draw_vf_generic_emit( struct draw_vertex_fetch *vf,
+                      unsigned count,
+                      uint8_t *v );
+
+void 
+draw_vf_generate_hardwired_emit( struct draw_vertex_fetch *vf );
+
+void 
+draw_vf_generate_sse_emit( struct draw_vertex_fetch *vf );
+
+
+/** XXX this type and function could probably be moved into draw_vf.c */
+struct draw_vf_format_info {
+   const char *name;
+   draw_vf_insert_func insert[4];
+   const unsigned attrsize;
+   const boolean isconst;
+};
+
+extern const struct draw_vf_format_info 
+draw_vf_format_info[DRAW_EMIT_MAX];
+
+
+#endif
diff --git a/src/gallium/aux/draw/draw_vf_generic.c b/src/gallium/aux/draw/draw_vf_generic.c
new file mode 100644
index 0000000000..7a60a9db9c
--- /dev/null
+++ b/src/gallium/aux/draw/draw_vf_generic.c
@@ -0,0 +1,585 @@
+
+/*
+ * Copyright 2003 Tungsten Graphics, inc.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * on the rights to use, copy, modify, merge, publish, distribute, sub
+ * license, and/or sell copies of the Software, and to permit persons to whom
+ * the Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the next
+ * paragraph) shall be included in all copies or substantial portions of the
+ * Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.  IN NO EVENT SHALL
+ * TUNGSTEN GRAPHICS AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
+ * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
+ * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
+ * USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors:
+ *    Keith Whitwell <keithw@tungstengraphics.com>
+ */
+
+
+#include "pipe/p_compiler.h"
+#include "pipe/p_debug.h"
+#include "pipe/p_util.h"
+
+#include "draw_vf.h"
+
+
+
+static INLINE void insert_4f_4( const struct draw_vf_attr *a, uint8_t *v, const float *in )
+{
+   float *out = (float *)(v);
+   (void) a;
+   
+   out[0] = in[0];
+   out[1] = in[1];
+   out[2] = in[2];
+   out[3] = in[3];
+}
+
+static INLINE void insert_4f_3( const struct draw_vf_attr *a, uint8_t *v, const float *in )
+{
+   float *out = (float *)(v);
+   (void) a;
+   
+   out[0] = in[0];
+   out[1] = in[1];
+   out[2] = in[2];
+   out[3] = 1;
+}
+
+static INLINE void insert_4f_2( const struct draw_vf_attr *a, uint8_t *v, const float *in )
+{
+   float *out = (float *)(v);
+   (void) a;
+   
+   out[0] = in[0];
+   out[1] = in[1];
+   out[2] = 0;
+   out[3] = 1;
+}
+
+static INLINE void insert_4f_1( const struct draw_vf_attr *a, uint8_t *v, const float *in )
+{
+   float *out = (float *)(v);
+   (void) a;
+   
+   out[0] = in[0];
+   out[1] = 0;
+   out[2] = 0;
+   out[3] = 1;
+}
+
+static INLINE void insert_3f_xyw_4( const struct draw_vf_attr *a, uint8_t *v, const float *in )
+{
+   float *out = (float *)(v);
+   (void) a;
+   
+   out[0] = in[0];
+   out[1] = in[1];
+   out[2] = in[3];
+}
+
+static INLINE void insert_3f_xyw_err( const struct draw_vf_attr *a, uint8_t *v, const float *in )
+{
+   (void) a; (void) v; (void) in;
+   assert(0);
+}
+
+static INLINE void insert_3f_3( const struct draw_vf_attr *a, uint8_t *v, const float *in )
+{
+   float *out = (float *)(v);
+   (void) a;
+   
+   out[0] = in[0];
+   out[1] = in[1];
+   out[2] = in[2];
+}
+
+static INLINE void insert_3f_2( const struct draw_vf_attr *a, uint8_t *v, const float *in )
+{
+   float *out = (float *)(v);
+   (void) a;
+   
+   out[0] = in[0];
+   out[1] = in[1];
+   out[2] = 0;
+}
+
+static INLINE void insert_3f_1( const struct draw_vf_attr *a, uint8_t *v, const float *in )
+{
+   float *out = (float *)(v);
+   (void) a;
+   
+   out[0] = in[0];
+   out[1] = 0;
+   out[2] = 0;
+}
+
+
+static INLINE void insert_2f_2( const struct draw_vf_attr *a, uint8_t *v, const float *in )
+{
+   float *out = (float *)(v);
+   (void) a;
+   
+   out[0] = in[0];
+   out[1] = in[1];
+}
+
+static INLINE void insert_2f_1( const struct draw_vf_attr *a, uint8_t *v, const float *in )
+{
+   float *out = (float *)(v);
+   (void) a;
+   
+   out[0] = in[0];
+   out[1] = 0;
+}
+
+static INLINE void insert_1f_1( const struct draw_vf_attr *a, uint8_t *v, const float *in )
+{
+   float *out = (float *)(v);
+   (void) a;
+
+   out[0] = in[0];
+}
+
+static INLINE void insert_null( const struct draw_vf_attr *a, uint8_t *v, const float *in )
+{
+   (void) a; (void) v; (void) in;
+}
+
+static INLINE void insert_4ub_4f_rgba_4( const struct draw_vf_attr *a, uint8_t *v, 
+					 const float *in )
+{
+   (void) a;
+   UNCLAMPED_FLOAT_TO_UBYTE(v[0], in[0]);
+   UNCLAMPED_FLOAT_TO_UBYTE(v[1], in[1]);
+   UNCLAMPED_FLOAT_TO_UBYTE(v[2], in[2]);
+   UNCLAMPED_FLOAT_TO_UBYTE(v[3], in[3]);
+}
+
+static INLINE void insert_4ub_4f_rgba_3( const struct draw_vf_attr *a, uint8_t *v, 
+					 const float *in )
+{
+   (void) a;
+   UNCLAMPED_FLOAT_TO_UBYTE(v[0], in[0]);
+   UNCLAMPED_FLOAT_TO_UBYTE(v[1], in[1]);
+   UNCLAMPED_FLOAT_TO_UBYTE(v[2], in[2]);
+   v[3] = 0xff;
+}
+
+static INLINE void insert_4ub_4f_rgba_2( const struct draw_vf_attr *a, uint8_t *v, 
+					 const float *in )
+{
+   (void) a;
+   UNCLAMPED_FLOAT_TO_UBYTE(v[0], in[0]);
+   UNCLAMPED_FLOAT_TO_UBYTE(v[1], in[1]);
+   v[2] = 0;
+   v[3] = 0xff;
+}
+
+static INLINE void insert_4ub_4f_rgba_1( const struct draw_vf_attr *a, uint8_t *v, 
+					 const float *in )
+{
+   (void) a;
+   UNCLAMPED_FLOAT_TO_UBYTE(v[0], in[0]);
+   v[1] = 0;
+   v[2] = 0;
+   v[3] = 0xff;
+}
+
+static INLINE void insert_4ub_4f_bgra_4( const struct draw_vf_attr *a, uint8_t *v, 
+					 const float *in )
+{
+   (void) a;
+   UNCLAMPED_FLOAT_TO_UBYTE(v[2], in[0]);
+   UNCLAMPED_FLOAT_TO_UBYTE(v[1], in[1]);
+   UNCLAMPED_FLOAT_TO_UBYTE(v[0], in[2]);
+   UNCLAMPED_FLOAT_TO_UBYTE(v[3], in[3]);
+}
+
+static INLINE void insert_4ub_4f_bgra_3( const struct draw_vf_attr *a, uint8_t *v, 
+					 const float *in )
+{
+   (void) a;
+   UNCLAMPED_FLOAT_TO_UBYTE(v[2], in[0]);
+   UNCLAMPED_FLOAT_TO_UBYTE(v[1], in[1]);
+   UNCLAMPED_FLOAT_TO_UBYTE(v[0], in[2]);
+   v[3] = 0xff;
+}
+
+static INLINE void insert_4ub_4f_bgra_2( const struct draw_vf_attr *a, uint8_t *v, 
+					 const float *in )
+{
+   (void) a;
+   UNCLAMPED_FLOAT_TO_UBYTE(v[2], in[0]);
+   UNCLAMPED_FLOAT_TO_UBYTE(v[1], in[1]);
+   v[0] = 0;
+   v[3] = 0xff;
+}
+
+static INLINE void insert_4ub_4f_bgra_1( const struct draw_vf_attr *a, uint8_t *v, 
+					 const float *in )
+{
+   (void) a;
+   UNCLAMPED_FLOAT_TO_UBYTE(v[2], in[0]);
+   v[1] = 0;
+   v[0] = 0;
+   v[3] = 0xff;
+}
+
+static INLINE void insert_4ub_4f_argb_4( const struct draw_vf_attr *a, uint8_t *v, 
+					 const float *in )
+{
+   (void) a;
+   UNCLAMPED_FLOAT_TO_UBYTE(v[1], in[0]);
+   UNCLAMPED_FLOAT_TO_UBYTE(v[2], in[1]);
+   UNCLAMPED_FLOAT_TO_UBYTE(v[3], in[2]);
+   UNCLAMPED_FLOAT_TO_UBYTE(v[0], in[3]);
+}
+
+static INLINE void insert_4ub_4f_argb_3( const struct draw_vf_attr *a, uint8_t *v, 
+					 const float *in )
+{
+   (void) a;
+   UNCLAMPED_FLOAT_TO_UBYTE(v[1], in[0]);
+   UNCLAMPED_FLOAT_TO_UBYTE(v[2], in[1]);
+   UNCLAMPED_FLOAT_TO_UBYTE(v[3], in[2]);
+   v[0] = 0xff;
+}
+
+static INLINE void insert_4ub_4f_argb_2( const struct draw_vf_attr *a, uint8_t *v, 
+					 const float *in )
+{
+   (void) a;
+   UNCLAMPED_FLOAT_TO_UBYTE(v[1], in[0]);
+   UNCLAMPED_FLOAT_TO_UBYTE(v[2], in[1]);
+   v[3] = 0x00;
+   v[0] = 0xff;
+}
+
+static INLINE void insert_4ub_4f_argb_1( const struct draw_vf_attr *a, uint8_t *v, 
+					 const float *in )
+{
+   (void) a;
+   UNCLAMPED_FLOAT_TO_UBYTE(v[1], in[0]);
+   v[2] = 0x00;
+   v[3] = 0x00;
+   v[0] = 0xff;
+}
+
+static INLINE void insert_4ub_4f_abgr_4( const struct draw_vf_attr *a, uint8_t *v, 
+					 const float *in )
+{
+   (void) a;
+   UNCLAMPED_FLOAT_TO_UBYTE(v[3], in[0]);
+   UNCLAMPED_FLOAT_TO_UBYTE(v[2], in[1]);
+   UNCLAMPED_FLOAT_TO_UBYTE(v[1], in[2]);
+   UNCLAMPED_FLOAT_TO_UBYTE(v[0], in[3]);
+}
+
+static INLINE void insert_4ub_4f_abgr_3( const struct draw_vf_attr *a, uint8_t *v, 
+					 const float *in )
+{
+   (void) a;
+   UNCLAMPED_FLOAT_TO_UBYTE(v[3], in[0]);
+   UNCLAMPED_FLOAT_TO_UBYTE(v[2], in[1]);
+   UNCLAMPED_FLOAT_TO_UBYTE(v[1], in[2]);
+   v[0] = 0xff;
+}
+
+static INLINE void insert_4ub_4f_abgr_2( const struct draw_vf_attr *a, uint8_t *v, 
+					 const float *in )
+{
+   (void) a;
+   UNCLAMPED_FLOAT_TO_UBYTE(v[3], in[0]);
+   UNCLAMPED_FLOAT_TO_UBYTE(v[2], in[1]);
+   v[1] = 0x00;
+   v[0] = 0xff;
+}
+
+static INLINE void insert_4ub_4f_abgr_1( const struct draw_vf_attr *a, uint8_t *v, 
+					 const float *in )
+{
+   (void) a;
+   UNCLAMPED_FLOAT_TO_UBYTE(v[3], in[0]);
+   v[2] = 0x00;
+   v[1] = 0x00;
+   v[0] = 0xff;
+}
+
+static INLINE void insert_3ub_3f_rgb_3( const struct draw_vf_attr *a, uint8_t *v, 
+					const float *in )
+{
+   (void) a;
+   UNCLAMPED_FLOAT_TO_UBYTE(v[0], in[0]);
+   UNCLAMPED_FLOAT_TO_UBYTE(v[1], in[1]);
+   UNCLAMPED_FLOAT_TO_UBYTE(v[2], in[2]);
+}
+
+static INLINE void insert_3ub_3f_rgb_2( const struct draw_vf_attr *a, uint8_t *v, 
+					const float *in )
+{
+   (void) a;
+   UNCLAMPED_FLOAT_TO_UBYTE(v[0], in[0]);
+   UNCLAMPED_FLOAT_TO_UBYTE(v[1], in[1]);
+   v[2] = 0;
+}
+
+static INLINE void insert_3ub_3f_rgb_1( const struct draw_vf_attr *a, uint8_t *v, 
+					const float *in )
+{
+   (void) a;
+   UNCLAMPED_FLOAT_TO_UBYTE(v[0], in[0]);
+   v[1] = 0;
+   v[2] = 0;
+}
+
+static INLINE void insert_3ub_3f_bgr_3( const struct draw_vf_attr *a, uint8_t *v, 
+					const float *in )
+{
+   (void) a;
+   UNCLAMPED_FLOAT_TO_UBYTE(v[2], in[0]);
+   UNCLAMPED_FLOAT_TO_UBYTE(v[1], in[1]);
+   UNCLAMPED_FLOAT_TO_UBYTE(v[0], in[2]);
+}
+
+static INLINE void insert_3ub_3f_bgr_2( const struct draw_vf_attr *a, uint8_t *v, 
+					const float *in )
+{
+   (void) a;
+   UNCLAMPED_FLOAT_TO_UBYTE(v[2], in[0]);
+   UNCLAMPED_FLOAT_TO_UBYTE(v[1], in[1]);
+   v[0] = 0;
+}
+
+static INLINE void insert_3ub_3f_bgr_1( const struct draw_vf_attr *a, uint8_t *v, 
+					const float *in )
+{
+   (void) a;
+   UNCLAMPED_FLOAT_TO_UBYTE(v[2], in[0]);
+   v[1] = 0;
+   v[0] = 0;
+}
+
+
+static INLINE void insert_1ub_1f_1( const struct draw_vf_attr *a, uint8_t *v, 
+				    const float *in )
+{
+   (void) a;
+   UNCLAMPED_FLOAT_TO_UBYTE(v[0], in[0]);
+}
+
+
+const struct draw_vf_format_info draw_vf_format_info[DRAW_EMIT_MAX] = 
+{
+   { "1f",
+     { insert_1f_1, insert_1f_1, insert_1f_1, insert_1f_1 },
+     sizeof(float), FALSE },
+
+   { "2f",
+     { insert_2f_1, insert_2f_2, insert_2f_2, insert_2f_2 },
+     2 * sizeof(float), FALSE },
+
+   { "3f",
+     { insert_3f_1, insert_3f_2, insert_3f_3, insert_3f_3 },
+     3 * sizeof(float), FALSE },
+
+   { "4f",
+     { insert_4f_1, insert_4f_2, insert_4f_3, insert_4f_4 },
+     4 * sizeof(float), FALSE },
+
+   { "3f_xyw",
+     { insert_3f_xyw_err, insert_3f_xyw_err, insert_3f_xyw_err, 
+       insert_3f_xyw_4 },
+     3 * sizeof(float), FALSE },
+
+   { "1ub_1f",
+     { insert_1ub_1f_1, insert_1ub_1f_1, insert_1ub_1f_1, insert_1ub_1f_1 },
+     sizeof(uint8_t), FALSE },
+
+   { "3ub_3f_rgb",
+     { insert_3ub_3f_rgb_1, insert_3ub_3f_rgb_2, insert_3ub_3f_rgb_3,
+       insert_3ub_3f_rgb_3 },
+     3 * sizeof(uint8_t), FALSE },
+
+   { "3ub_3f_bgr",
+     { insert_3ub_3f_bgr_1, insert_3ub_3f_bgr_2, insert_3ub_3f_bgr_3,
+       insert_3ub_3f_bgr_3 },
+     3 * sizeof(uint8_t), FALSE },
+
+   { "4ub_4f_rgba",
+     { insert_4ub_4f_rgba_1, insert_4ub_4f_rgba_2, insert_4ub_4f_rgba_3, 
+       insert_4ub_4f_rgba_4 },
+     4 * sizeof(uint8_t), FALSE },
+
+   { "4ub_4f_bgra",
+     { insert_4ub_4f_bgra_1, insert_4ub_4f_bgra_2, insert_4ub_4f_bgra_3,
+       insert_4ub_4f_bgra_4 },
+     4 * sizeof(uint8_t), FALSE },
+
+   { "4ub_4f_argb",
+     { insert_4ub_4f_argb_1, insert_4ub_4f_argb_2, insert_4ub_4f_argb_3,
+       insert_4ub_4f_argb_4 },
+     4 * sizeof(uint8_t), FALSE },
+
+   { "4ub_4f_abgr",
+     { insert_4ub_4f_abgr_1, insert_4ub_4f_abgr_2, insert_4ub_4f_abgr_3,
+       insert_4ub_4f_abgr_4 },
+     4 * sizeof(uint8_t), FALSE },
+
+   { "1f_const",
+     { insert_1f_1, insert_1f_1, insert_1f_1, insert_1f_1 },
+     sizeof(float), TRUE },
+   
+   { "2f_const",
+     { insert_2f_1, insert_2f_2, insert_2f_2, insert_2f_2 },
+     2 * sizeof(float), TRUE },
+   
+   { "3f_const",
+     { insert_3f_1, insert_3f_2, insert_3f_3, insert_3f_3 },
+     3 * sizeof(float), TRUE },
+   
+   { "4f_const",
+     { insert_4f_1, insert_4f_2, insert_4f_3, insert_4f_4 },
+     4 * sizeof(float), TRUE },
+
+   { "pad",
+     { NULL, NULL, NULL, NULL },
+     0, FALSE },
+
+};
+
+
+
+    
+/***********************************************************************
+ * Hardwired fastpaths for emitting whole vertices or groups of
+ * vertices
+ */
+#define EMIT5(NR, F0, F1, F2, F3, F4, NAME)				\
+static void NAME( struct draw_vertex_fetch *vf,				\
+		  unsigned count,						\
+		  uint8_t *v )						\
+{									\
+   struct draw_vf_attr *a = vf->attr;				\
+   unsigned i;								\
+									\
+   for (i = 0 ; i < count ; i++, v += vf->vertex_stride) {		\
+      if (NR > 0) {							\
+	 F0( &a[0], v + a[0].vertoffset, (float *)a[0].inputptr );	\
+	 a[0].inputptr += a[0].inputstride;				\
+      }									\
+      									\
+      if (NR > 1) {							\
+	 F1( &a[1], v + a[1].vertoffset, (float *)a[1].inputptr );	\
+	 a[1].inputptr += a[1].inputstride;				\
+      }									\
+      									\
+      if (NR > 2) {							\
+	 F2( &a[2], v + a[2].vertoffset, (float *)a[2].inputptr );	\
+	 a[2].inputptr += a[2].inputstride;				\
+      }									\
+      									\
+      if (NR > 3) {							\
+	 F3( &a[3], v + a[3].vertoffset, (float *)a[3].inputptr );	\
+	 a[3].inputptr += a[3].inputstride;				\
+      }									\
+									\
+      if (NR > 4) {							\
+	 F4( &a[4], v + a[4].vertoffset, (float *)a[4].inputptr );	\
+	 a[4].inputptr += a[4].inputstride;				\
+      }									\
+   }									\
+}
+
+   
+#define EMIT2(F0, F1, NAME) EMIT5(2, F0, F1, insert_null, \
+				  insert_null, insert_null, NAME)
+
+#define EMIT3(F0, F1, F2, NAME) EMIT5(3, F0, F1, F2, insert_null, \
+				      insert_null, NAME)
+   
+#define EMIT4(F0, F1, F2, F3, NAME) EMIT5(4, F0, F1, F2, F3, \
+				          insert_null, NAME)
+   
+
+EMIT2(insert_3f_3, insert_4ub_4f_rgba_4, emit_xyz3_rgba4)
+
+EMIT3(insert_4f_4, insert_4ub_4f_rgba_4, insert_2f_2, emit_xyzw4_rgba4_st2)
+
+EMIT4(insert_4f_4, insert_4ub_4f_rgba_4, insert_2f_2, insert_2f_2, emit_xyzw4_rgba4_st2_st2)
+
+
+/* Use the codegen paths to select one of a number of hardwired
+ * fastpaths.
+ */
+void draw_vf_generate_hardwired_emit( struct draw_vertex_fetch *vf )
+{
+   draw_vf_emit_func func = NULL;
+
+   /* Does it fit a hardwired fastpath?  Help! this is growing out of
+    * control!
+    */
+   switch (vf->attr_count) {
+   case 2:
+      if (vf->attr[0].do_insert == insert_3f_3 &&
+	  vf->attr[1].do_insert == insert_4ub_4f_rgba_4) {
+ 	 func = emit_xyz3_rgba4; 
+      }
+      break;
+   case 3:
+      if (vf->attr[2].do_insert == insert_2f_2) {
+	 if (vf->attr[1].do_insert == insert_4ub_4f_rgba_4) {
+	    if (vf->attr[0].do_insert == insert_4f_4) 
+	       func = emit_xyzw4_rgba4_st2;
+	 }
+      }
+      break;
+   case 4:
+      if (vf->attr[2].do_insert == insert_2f_2 &&
+	  vf->attr[3].do_insert == insert_2f_2) {
+	 if (vf->attr[1].do_insert == insert_4ub_4f_rgba_4) {
+	    if (vf->attr[0].do_insert == insert_4f_4) 
+	       func = emit_xyzw4_rgba4_st2_st2;
+	 }
+      }
+      break;
+   }
+
+   vf->emit = func;
+}
+
+/***********************************************************************
+ * Generic (non-codegen) functions for whole vertices or groups of
+ * vertices
+ */
+
+void draw_vf_generic_emit( struct draw_vertex_fetch *vf,
+		      unsigned count,
+		      uint8_t *v )
+{
+   struct draw_vf_attr *a = vf->attr;
+   const unsigned attr_count = vf->attr_count;
+   const unsigned stride = vf->vertex_stride;
+   unsigned i, j;
+
+   for (i = 0 ; i < count ; i++, v += stride) {
+      for (j = 0; j < attr_count; j++) {
+	 float *in = (float *)a[j].inputptr;
+	 a[j].inputptr += a[j].inputstride;
+	 a[j].do_insert( &a[j], v + a[j].vertoffset, in );
+      }
+   }
+}
+
+
diff --git a/src/gallium/aux/draw/draw_vf_sse.c b/src/gallium/aux/draw/draw_vf_sse.c
new file mode 100644
index 0000000000..1ad2ae756d
--- /dev/null
+++ b/src/gallium/aux/draw/draw_vf_sse.c
@@ -0,0 +1,614 @@
+/*
+ * Copyright 2003 Tungsten Graphics, inc.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * on the rights to use, copy, modify, merge, publish, distribute, sub
+ * license, and/or sell copies of the Software, and to permit persons to whom
+ * the Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the next
+ * paragraph) shall be included in all copies or substantial portions of the
+ * Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.  IN NO EVENT SHALL
+ * TUNGSTEN GRAPHICS AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
+ * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
+ * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
+ * USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors:
+ *    Keith Whitwell <keithw@tungstengraphics.com>
+ */
+
+
+#include "simple_list.h"
+
+#include "pipe/p_compiler.h"
+
+#include "draw_vf.h"
+
+
+#if defined(USE_SSE_ASM)
+
+#include "x86/rtasm/x86sse.h"
+#include "x86/common_x86_asm.h"
+
+
+#define X    0
+#define Y    1
+#define Z    2
+#define W    3
+
+
+struct x86_program {
+   struct x86_function func;
+
+   struct draw_vertex_fetch *vf;
+   boolean inputs_safe;
+   boolean outputs_safe;
+   boolean have_sse2;
+   
+   struct x86_reg identity;
+   struct x86_reg chan0;
+};
+
+
+static struct x86_reg get_identity( struct x86_program *p )
+{
+   return p->identity;
+}
+
+static void emit_load4f_4( struct x86_program *p, 			   
+			   struct x86_reg dest,
+			   struct x86_reg arg0 )
+{
+   sse_movups(&p->func, dest, arg0);
+}
+
+static void emit_load4f_3( struct x86_program *p, 
+			   struct x86_reg dest,
+			   struct x86_reg arg0 )
+{
+   /* Have to jump through some hoops:
+    *
+    * c 0 0 0
+    * c 0 0 1
+    * 0 0 c 1
+    * a b c 1
+    */
+   sse_movss(&p->func, dest, x86_make_disp(arg0, 8));
+   sse_shufps(&p->func, dest, get_identity(p), SHUF(X,Y,Z,W) );
+   sse_shufps(&p->func, dest, dest, SHUF(Y,Z,X,W) );
+   sse_movlps(&p->func, dest, arg0);
+}
+
+static void emit_load4f_2( struct x86_program *p, 
+			   struct x86_reg dest,
+			   struct x86_reg arg0 )
+{
+   /* Initialize from identity, then pull in low two words:
+    */
+   sse_movups(&p->func, dest, get_identity(p));
+   sse_movlps(&p->func, dest, arg0);
+}
+
+static void emit_load4f_1( struct x86_program *p, 
+			   struct x86_reg dest,
+			   struct x86_reg arg0 )
+{
+   /* Pull in low word, then swizzle in identity */
+   sse_movss(&p->func, dest, arg0);
+   sse_shufps(&p->func, dest, get_identity(p), SHUF(X,Y,Z,W) );
+}
+
+
+
+static void emit_load3f_3( struct x86_program *p, 			   
+			   struct x86_reg dest,
+			   struct x86_reg arg0 )
+{
+   /* Over-reads by 1 dword - potential SEGV if input is a vertex
+    * array.
+    */
+   if (p->inputs_safe) {
+      sse_movups(&p->func, dest, arg0);
+   } 
+   else {
+      /* c 0 0 0
+       * c c c c
+       * a b c c 
+       */
+      sse_movss(&p->func, dest, x86_make_disp(arg0, 8));
+      sse_shufps(&p->func, dest, dest, SHUF(X,X,X,X));
+      sse_movlps(&p->func, dest, arg0);
+   }
+}
+
+static void emit_load3f_2( struct x86_program *p, 
+			   struct x86_reg dest,
+			   struct x86_reg arg0 )
+{
+   emit_load4f_2(p, dest, arg0);
+}
+
+static void emit_load3f_1( struct x86_program *p, 
+			   struct x86_reg dest,
+			   struct x86_reg arg0 )
+{
+   emit_load4f_1(p, dest, arg0);
+}
+
+static void emit_load2f_2( struct x86_program *p, 
+			   struct x86_reg dest,
+			   struct x86_reg arg0 )
+{
+   sse_movlps(&p->func, dest, arg0);
+}
+
+static void emit_load2f_1( struct x86_program *p, 
+			   struct x86_reg dest,
+			   struct x86_reg arg0 )
+{
+   emit_load4f_1(p, dest, arg0);
+}
+
+static void emit_load1f_1( struct x86_program *p, 
+			   struct x86_reg dest,
+			   struct x86_reg arg0 )
+{
+   sse_movss(&p->func, dest, arg0);
+}
+
+static void (*load[4][4])( struct x86_program *p, 
+			   struct x86_reg dest,
+			   struct x86_reg arg0 ) = {
+   { emit_load1f_1, 
+     emit_load1f_1, 
+     emit_load1f_1, 
+     emit_load1f_1 },
+
+   { emit_load2f_1, 
+     emit_load2f_2, 
+     emit_load2f_2, 
+     emit_load2f_2 },
+
+   { emit_load3f_1, 
+     emit_load3f_2, 
+     emit_load3f_3, 
+     emit_load3f_3 },
+
+   { emit_load4f_1, 
+     emit_load4f_2, 
+     emit_load4f_3, 
+     emit_load4f_4 } 
+};
+
+static void emit_load( struct x86_program *p,
+		       struct x86_reg dest,
+		       unsigned sz,
+		       struct x86_reg src,
+		       unsigned src_sz)
+{
+   load[sz-1][src_sz-1](p, dest, src);
+}
+
+static void emit_store4f( struct x86_program *p, 			   
+			  struct x86_reg dest,
+			  struct x86_reg arg0 )
+{
+   sse_movups(&p->func, dest, arg0);
+}
+
+static void emit_store3f( struct x86_program *p, 
+			  struct x86_reg dest,
+			  struct x86_reg arg0 )
+{
+   if (p->outputs_safe) {
+      /* Emit the extra dword anyway.  This may hurt writecombining,
+       * may cause other problems.
+       */
+      sse_movups(&p->func, dest, arg0);
+   }
+   else {
+      /* Alternate strategy - emit two, shuffle, emit one.
+       */
+      sse_movlps(&p->func, dest, arg0);
+      sse_shufps(&p->func, arg0, arg0, SHUF(Z,Z,Z,Z) ); /* NOTE! destructive */
+      sse_movss(&p->func, x86_make_disp(dest,8), arg0);
+   }
+}
+
+static void emit_store2f( struct x86_program *p, 
+			   struct x86_reg dest,
+			   struct x86_reg arg0 )
+{
+   sse_movlps(&p->func, dest, arg0);
+}
+
+static void emit_store1f( struct x86_program *p, 
+			  struct x86_reg dest,
+			  struct x86_reg arg0 )
+{
+   sse_movss(&p->func, dest, arg0);
+}
+
+
+static void (*store[4])( struct x86_program *p, 
+			 struct x86_reg dest,
+			 struct x86_reg arg0 ) = 
+{
+   emit_store1f, 
+   emit_store2f, 
+   emit_store3f, 
+   emit_store4f 
+};
+
+static void emit_store( struct x86_program *p,
+			struct x86_reg dest,
+			unsigned sz,
+			struct x86_reg temp )
+
+{
+   store[sz-1](p, dest, temp);
+}
+
+static void emit_pack_store_4ub( struct x86_program *p,
+				 struct x86_reg dest,
+				 struct x86_reg temp )
+{
+   /* Scale by 255.0
+    */
+   sse_mulps(&p->func, temp, p->chan0);
+
+   if (p->have_sse2) {
+      sse2_cvtps2dq(&p->func, temp, temp);
+      sse2_packssdw(&p->func, temp, temp);
+      sse2_packuswb(&p->func, temp, temp);
+      sse_movss(&p->func, dest, temp);
+   }
+   else {
+      struct x86_reg mmx0 = x86_make_reg(file_MMX, 0);
+      struct x86_reg mmx1 = x86_make_reg(file_MMX, 1);
+      sse_cvtps2pi(&p->func, mmx0, temp);
+      sse_movhlps(&p->func, temp, temp);
+      sse_cvtps2pi(&p->func, mmx1, temp);
+      mmx_packssdw(&p->func, mmx0, mmx1);
+      mmx_packuswb(&p->func, mmx0, mmx0);
+      mmx_movd(&p->func, dest, mmx0);
+   }
+}
+
+static int get_offset( const void *a, const void *b )
+{
+   return (const char *)b - (const char *)a;
+}
+
+/* Not much happens here.  Eventually use this function to try and
+ * avoid saving/reloading the source pointers each vertex (if some of
+ * them can fit in registers).
+ */
+static void get_src_ptr( struct x86_program *p,
+			 struct x86_reg srcREG,
+			 struct x86_reg vfREG,
+			 struct draw_vf_attr *a )
+{
+   struct draw_vertex_fetch *vf = p->vf;
+   struct x86_reg ptr_to_src = x86_make_disp(vfREG, get_offset(vf, &a->inputptr));
+
+   /* Load current a[j].inputptr
+    */
+   x86_mov(&p->func, srcREG, ptr_to_src);
+}
+
+static void update_src_ptr( struct x86_program *p,
+			 struct x86_reg srcREG,
+			 struct x86_reg vfREG,
+			 struct draw_vf_attr *a )
+{
+   if (a->inputstride) {
+      struct draw_vertex_fetch *vf = p->vf;
+      struct x86_reg ptr_to_src = x86_make_disp(vfREG, get_offset(vf, &a->inputptr));
+
+      /* add a[j].inputstride (hardcoded value - could just as easily
+       * pull the stride value from memory each time).
+       */
+      x86_lea(&p->func, srcREG, x86_make_disp(srcREG, a->inputstride));
+      
+      /* save new value of a[j].inputptr 
+       */
+      x86_mov(&p->func, ptr_to_src, srcREG);
+   }
+}
+
+
+/* Lots of hardcoding
+ *
+ * EAX -- pointer to current output vertex
+ * ECX -- pointer to current attribute 
+ * 
+ */
+static boolean build_vertex_emit( struct x86_program *p )
+{
+   struct draw_vertex_fetch *vf = p->vf;
+   unsigned j = 0;
+
+   struct x86_reg vertexEAX = x86_make_reg(file_REG32, reg_AX);
+   struct x86_reg srcECX = x86_make_reg(file_REG32, reg_CX);
+   struct x86_reg countEBP = x86_make_reg(file_REG32, reg_BP);
+   struct x86_reg vfESI = x86_make_reg(file_REG32, reg_SI);
+   struct x86_reg temp = x86_make_reg(file_XMM, 0);
+   uint8_t *fixup, *label;
+
+   /* Push a few regs?
+    */
+   x86_push(&p->func, countEBP);
+   x86_push(&p->func, vfESI);
+
+
+   /* Get vertex count, compare to zero
+    */
+   x86_xor(&p->func, srcECX, srcECX);
+   x86_mov(&p->func, countEBP, x86_fn_arg(&p->func, 2));
+   x86_cmp(&p->func, countEBP, srcECX);
+   fixup = x86_jcc_forward(&p->func, cc_E);
+
+   /* Initialize destination register. 
+    */
+   x86_mov(&p->func, vertexEAX, x86_fn_arg(&p->func, 3));
+
+   /* Move argument 1 (vf) into a reg:
+    */
+   x86_mov(&p->func, vfESI, x86_fn_arg(&p->func, 1));
+
+   
+   /* always load, needed or not:
+    */
+   sse_movups(&p->func, p->identity, x86_make_disp(vfESI, get_offset(vf, &vf->identity[0])));
+
+   /* Note address for loop jump */
+   label = x86_get_label(&p->func);
+
+   /* Emit code for each of the attributes.  Currently routes
+    * everything through SSE registers, even when it might be more
+    * efficient to stick with regular old x86.  No optimization or
+    * other tricks - enough new ground to cover here just getting
+    * things working.
+    */
+   while (j < vf->attr_count) {
+      struct draw_vf_attr *a = &vf->attr[j];
+      struct x86_reg dest = x86_make_disp(vertexEAX, a->vertoffset);
+
+      /* Now, load an XMM reg from src, perhaps transform, then save.
+       * Could be shortcircuited in specific cases:
+       */
+      switch (a->format) {
+      case DRAW_EMIT_1F:
+      case DRAW_EMIT_1F_CONST:
+	 get_src_ptr(p, srcECX, vfESI, a);
+	 emit_load(p, temp, 1, x86_deref(srcECX), a->inputsize);
+	 emit_store(p, dest, 1, temp);
+	 update_src_ptr(p, srcECX, vfESI, a);
+	 break;
+      case DRAW_EMIT_2F:
+      case DRAW_EMIT_2F_CONST:
+	 get_src_ptr(p, srcECX, vfESI, a);
+	 emit_load(p, temp, 2, x86_deref(srcECX), a->inputsize);
+	 emit_store(p, dest, 2, temp);
+	 update_src_ptr(p, srcECX, vfESI, a);
+	 break;
+      case DRAW_EMIT_3F:
+      case DRAW_EMIT_3F_CONST:
+	 /* Potentially the worst case - hardcode 2+1 copying:
+	  */
+	 if (0) {
+	    get_src_ptr(p, srcECX, vfESI, a);
+	    emit_load(p, temp, 3, x86_deref(srcECX), a->inputsize);
+	    emit_store(p, dest, 3, temp);
+	    update_src_ptr(p, srcECX, vfESI, a);
+	 }
+	 else {
+	    get_src_ptr(p, srcECX, vfESI, a);
+	    emit_load(p, temp, 2, x86_deref(srcECX), a->inputsize);
+	    emit_store(p, dest, 2, temp);
+	    if (a->inputsize > 2) {
+	       emit_load(p, temp, 1, x86_make_disp(srcECX, 8), 1);
+	       emit_store(p, x86_make_disp(dest,8), 1, temp);
+	    }
+	    else {
+	       sse_movss(&p->func, x86_make_disp(dest,8), get_identity(p));
+	    }
+	    update_src_ptr(p, srcECX, vfESI, a);
+	 }
+	 break;
+      case DRAW_EMIT_4F:
+      case DRAW_EMIT_4F_CONST:
+	 get_src_ptr(p, srcECX, vfESI, a);
+	 emit_load(p, temp, 4, x86_deref(srcECX), a->inputsize);
+	 emit_store(p, dest, 4, temp);
+	 update_src_ptr(p, srcECX, vfESI, a);
+	 break;
+      case DRAW_EMIT_3F_XYW:
+	 get_src_ptr(p, srcECX, vfESI, a);
+	 emit_load(p, temp, 4, x86_deref(srcECX), a->inputsize);
+	 sse_shufps(&p->func, temp, temp, SHUF(X,Y,W,Z));
+	 emit_store(p, dest, 3, temp);
+	 update_src_ptr(p, srcECX, vfESI, a);
+	 break;
+
+      case DRAW_EMIT_1UB_1F:	 
+	 /* Test for PAD3 + 1UB:
+	  */
+	 if (j > 0 &&
+	     a[-1].vertoffset + a[-1].vertattrsize <= a->vertoffset - 3)
+	 {
+	    get_src_ptr(p, srcECX, vfESI, a);
+	    emit_load(p, temp, 1, x86_deref(srcECX), a->inputsize);
+	    sse_shufps(&p->func, temp, temp, SHUF(X,X,X,X));
+	    emit_pack_store_4ub(p, x86_make_disp(dest, -3), temp); /* overkill! */
+	    update_src_ptr(p, srcECX, vfESI, a);
+	 }
+	 else {
+	    debug_printf("Can't emit 1ub %x %x %d\n", 
+	            a->vertoffset, a[-1].vertoffset, a[-1].vertattrsize );
+	    return FALSE;
+	 }
+	 break;
+      case DRAW_EMIT_3UB_3F_RGB:
+      case DRAW_EMIT_3UB_3F_BGR:
+	 /* Test for 3UB + PAD1:
+	  */
+	 if (j == vf->attr_count - 1 ||
+	     a[1].vertoffset >= a->vertoffset + 4) {
+	    get_src_ptr(p, srcECX, vfESI, a);
+	    emit_load(p, temp, 3, x86_deref(srcECX), a->inputsize);
+	    if (a->format == DRAW_EMIT_3UB_3F_BGR)
+	       sse_shufps(&p->func, temp, temp, SHUF(Z,Y,X,W));
+	    emit_pack_store_4ub(p, dest, temp);
+	    update_src_ptr(p, srcECX, vfESI, a);
+	 }
+	 /* Test for 3UB + 1UB:
+	  */
+	 else if (j < vf->attr_count - 1 &&
+		  a[1].format == DRAW_EMIT_1UB_1F &&
+		  a[1].vertoffset == a->vertoffset + 3) {
+	    get_src_ptr(p, srcECX, vfESI, a);
+	    emit_load(p, temp, 3, x86_deref(srcECX), a->inputsize);
+	    update_src_ptr(p, srcECX, vfESI, a);
+
+	    /* Make room for incoming value:
+	     */
+	    sse_shufps(&p->func, temp, temp, SHUF(W,X,Y,Z));
+
+	    get_src_ptr(p, srcECX, vfESI, &a[1]);
+	    emit_load(p, temp, 1, x86_deref(srcECX), a[1].inputsize);
+	    update_src_ptr(p, srcECX, vfESI, &a[1]);
+
+	    /* Rearrange and possibly do BGR conversion:
+	     */
+	    if (a->format == DRAW_EMIT_3UB_3F_BGR)
+	       sse_shufps(&p->func, temp, temp, SHUF(W,Z,Y,X));
+	    else
+	       sse_shufps(&p->func, temp, temp, SHUF(Y,Z,W,X));
+
+	    emit_pack_store_4ub(p, dest, temp);
+	    j++;		/* NOTE: two attrs consumed */
+	 }
+	 else {
+	    debug_printf("Can't emit 3ub\n");
+	 }
+	 return FALSE;	/* add this later */
+	 break;
+
+      case DRAW_EMIT_4UB_4F_RGBA:
+	 get_src_ptr(p, srcECX, vfESI, a);
+	 emit_load(p, temp, 4, x86_deref(srcECX), a->inputsize);
+	 emit_pack_store_4ub(p, dest, temp);
+	 update_src_ptr(p, srcECX, vfESI, a);
+	 break;
+      case DRAW_EMIT_4UB_4F_BGRA:
+	 get_src_ptr(p, srcECX, vfESI, a);
+	 emit_load(p, temp, 4, x86_deref(srcECX), a->inputsize);
+	 sse_shufps(&p->func, temp, temp, SHUF(Z,Y,X,W));
+	 emit_pack_store_4ub(p, dest, temp);
+	 update_src_ptr(p, srcECX, vfESI, a);
+	 break;
+      case DRAW_EMIT_4UB_4F_ARGB:
+	 get_src_ptr(p, srcECX, vfESI, a);
+	 emit_load(p, temp, 4, x86_deref(srcECX), a->inputsize);
+	 sse_shufps(&p->func, temp, temp, SHUF(W,X,Y,Z));
+	 emit_pack_store_4ub(p, dest, temp);
+	 update_src_ptr(p, srcECX, vfESI, a);
+	 break;
+      case DRAW_EMIT_4UB_4F_ABGR:
+	 get_src_ptr(p, srcECX, vfESI, a);
+	 emit_load(p, temp, 4, x86_deref(srcECX), a->inputsize);
+	 sse_shufps(&p->func, temp, temp, SHUF(W,Z,Y,X));
+	 emit_pack_store_4ub(p, dest, temp);
+	 update_src_ptr(p, srcECX, vfESI, a);
+	 break;
+      default:
+	 debug_printf("unknown a[%d].format %d\n", j, a->format);
+	 return FALSE;	/* catch any new opcodes */
+      }
+      
+      /* Increment j by at least 1 - may have been incremented above also:
+       */
+      j++;
+   }
+
+   /* Next vertex:
+    */
+   x86_lea(&p->func, vertexEAX, x86_make_disp(vertexEAX, vf->vertex_stride));
+
+   /* decr count, loop if not zero
+    */
+   x86_dec(&p->func, countEBP);
+   x86_test(&p->func, countEBP, countEBP); 
+   x86_jcc(&p->func, cc_NZ, label);
+
+   /* Exit mmx state?
+    */
+   if (p->func.need_emms)
+      mmx_emms(&p->func);
+
+   /* Land forward jump here:
+    */
+   x86_fixup_fwd_jump(&p->func, fixup);
+
+   /* Pop regs and return
+    */
+   x86_pop(&p->func, x86_get_base_reg(vfESI));
+   x86_pop(&p->func, countEBP);
+   x86_ret(&p->func);
+
+   vf->emit = (draw_vf_emit_func)x86_get_func(&p->func);
+   return TRUE;
+}
+
+
+
+void draw_vf_generate_sse_emit( struct draw_vertex_fetch *vf )
+{
+   struct x86_program p;   
+
+   if (!cpu_has_xmm) {
+      vf->codegen_emit = NULL;
+      return;
+   }
+
+   memset(&p, 0, sizeof(p));
+
+   p.vf = vf;
+   p.inputs_safe = 0;		/* for now */
+   p.outputs_safe = 1;		/* for now */
+   p.have_sse2 = cpu_has_xmm2;
+   p.identity = x86_make_reg(file_XMM, 6);
+   p.chan0 = x86_make_reg(file_XMM, 7);
+
+   x86_init_func(&p.func);
+
+   if (build_vertex_emit(&p)) {
+      draw_vf_register_fastpath( vf, TRUE );
+   }
+   else {
+      /* Note the failure so that we don't keep trying to codegen an
+       * impossible state:
+       */
+      draw_vf_register_fastpath( vf, FALSE );
+      x86_release_func(&p.func);
+   }
+}
+
+#else
+
+void draw_vf_generate_sse_emit( struct draw_vertex_fetch *vf )
+{
+   /* Dummy version for when USE_SSE_ASM not defined */
+}
+
+#endif
diff --git a/src/gallium/aux/draw/draw_wide_prims.c b/src/gallium/aux/draw/draw_wide_prims.c
new file mode 100644
index 0000000000..655774b155
--- /dev/null
+++ b/src/gallium/aux/draw/draw_wide_prims.c
@@ -0,0 +1,432 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+/* Authors:  Keith Whitwell <keith@tungstengraphics.com>
+ */
+
+#include "pipe/p_util.h"
+#include "pipe/p_defines.h"
+#include "pipe/p_shader_tokens.h"
+#include "draw_private.h"
+
+
+struct wide_stage {
+   struct draw_stage stage;
+
+   float half_line_width;
+   float half_point_size;
+
+   uint texcoord_slot[PIPE_MAX_SHADER_OUTPUTS];
+   uint texcoord_mode[PIPE_MAX_SHADER_OUTPUTS];
+   uint num_texcoords;
+
+   int psize_slot;
+};
+
+
+
+static INLINE struct wide_stage *wide_stage( struct draw_stage *stage )
+{
+   return (struct wide_stage *)stage;
+}
+
+
+static void passthrough_point( struct draw_stage *stage,
+                               struct prim_header *header )
+{
+   stage->next->point( stage->next, header );
+}
+
+static void passthrough_line( struct draw_stage *stage,
+                              struct prim_header *header )
+{
+   stage->next->line(stage->next, header);
+}
+
+static void passthrough_tri( struct draw_stage *stage,
+                             struct prim_header *header )
+{
+   stage->next->tri(stage->next, header);
+}
+
+
+/**
+ * Draw a wide line by drawing a quad (two triangles).
+ * XXX need to disable polygon stipple.
+ */
+static void wide_line( struct draw_stage *stage,
+		       struct prim_header *header )
+{
+   const struct wide_stage *wide = wide_stage(stage);
+   const float half_width = wide->half_line_width;
+
+   struct prim_header tri;
+
+   struct vertex_header *v0 = dup_vert(stage, header->v[0], 0);
+   struct vertex_header *v1 = dup_vert(stage, header->v[0], 1);
+   struct vertex_header *v2 = dup_vert(stage, header->v[1], 2);
+   struct vertex_header *v3 = dup_vert(stage, header->v[1], 3);
+
+   float *pos0 = v0->data[0];
+   float *pos1 = v1->data[0];
+   float *pos2 = v2->data[0];
+   float *pos3 = v3->data[0];
+
+   const float dx = FABSF(pos0[0] - pos2[0]);
+   const float dy = FABSF(pos0[1] - pos2[1]);
+   
+   /*
+    * Draw wide line as a quad (two tris) by "stretching" the line along
+    * X or Y.
+    * We need to tweak coords in several ways to be conformant here.
+    */
+
+   if (dx > dy) {
+      /* x-major line */
+      pos0[1] = pos0[1] - half_width - 0.25f;
+      pos1[1] = pos1[1] + half_width - 0.25f;
+      pos2[1] = pos2[1] - half_width - 0.25f;
+      pos3[1] = pos3[1] + half_width - 0.25f;
+      if (pos0[0] < pos2[0]) {
+         /* left to right line */
+         pos0[0] -= 0.5f;
+         pos1[0] -= 0.5f;
+         pos2[0] -= 0.5f;
+         pos3[0] -= 0.5f;
+      }
+      else {
+         /* right to left line */
+         pos0[0] += 0.5f;
+         pos1[0] += 0.5f;
+         pos2[0] += 0.5f;
+         pos3[0] += 0.5f;
+      }
+   }
+   else {
+      /* y-major line */
+      pos0[0] = pos0[0] - half_width + 0.25f;
+      pos1[0] = pos1[0] + half_width + 0.25f;
+      pos2[0] = pos2[0] - half_width + 0.25f;
+      pos3[0] = pos3[0] + half_width + 0.25f;
+      if (pos0[1] < pos2[1]) {
+         /* top to bottom line */
+         pos0[1] -= 0.5f;
+         pos1[1] -= 0.5f;
+         pos2[1] -= 0.5f;
+         pos3[1] -= 0.5f;
+      }
+      else {
+         /* bottom to top line */
+         pos0[1] += 0.5f;
+         pos1[1] += 0.5f;
+         pos2[1] += 0.5f;
+         pos3[1] += 0.5f;
+      }
+   }
+
+   tri.det = header->det;  /* only the sign matters */
+   tri.v[0] = v0;
+   tri.v[1] = v2;
+   tri.v[2] = v3;
+   stage->next->tri( stage->next, &tri );
+
+   tri.v[0] = v0;
+   tri.v[1] = v3;
+   tri.v[2] = v1;
+   stage->next->tri( stage->next, &tri );
+}
+
+
+/**
+ * Draw a wide line by drawing a quad, using geometry which will
+ * fullfill GL's antialiased line requirements.
+ */
+static void wide_line_aa(struct draw_stage *stage,
+                         struct prim_header *header)
+{
+   const struct wide_stage *wide = wide_stage(stage);
+   const float half_width = wide->half_line_width;
+   struct prim_header tri;
+   struct vertex_header *v[4];
+   float *pos;
+   float dx = header->v[1]->data[0][0] - header->v[0]->data[0][0];
+   float dy = header->v[1]->data[0][1] - header->v[0]->data[0][1];
+   const float len = (float) sqrt(dx * dx + dy * dy);
+   uint i;
+
+   dx = dx * half_width / len;
+   dy = dy * half_width / len;
+
+   /* allocate/dup new verts */
+   for (i = 0; i < 4; i++) {
+      v[i] = dup_vert(stage, header->v[i/2], i);
+   }
+
+   /*
+    * Quad for line from v0 to v1:
+    *
+    *  1                         3
+    *  +-------------------------+
+    *  |                         |
+    *  *v0                     v1*
+    *  |                         |
+    *  +-------------------------+
+    *  0                         2
+    */
+
+   pos = v[0]->data[0];
+   pos[0] += dy;
+   pos[1] -= dx;
+
+   pos = v[1]->data[0];
+   pos[0] -= dy;
+   pos[1] += dx;
+
+   pos = v[2]->data[0];
+   pos[0] += dy;
+   pos[1] -= dx;
+
+   pos = v[3]->data[0];
+   pos[0] -= dy;
+   pos[1] += dx;
+
+   tri.det = header->det;  /* only the sign matters */
+
+   tri.v[0] = v[2];  tri.v[1] = v[1];  tri.v[2] = v[0];
+   stage->next->tri( stage->next, &tri );
+
+   tri.v[0] = v[3];  tri.v[1] = v[1];  tri.v[2] = v[2];
+   stage->next->tri( stage->next, &tri );
+
+}
+
+
+/**
+ * Set the vertex texcoords for sprite mode.
+ * Coords may be left untouched or set to a right-side-up or upside-down
+ * orientation.
+ */
+static void set_texcoords(const struct wide_stage *wide,
+                          struct vertex_header *v, const float tc[4])
+{
+   uint i;
+   for (i = 0; i < wide->num_texcoords; i++) {
+      if (wide->texcoord_mode[i] != PIPE_SPRITE_COORD_NONE) {
+         uint j = wide->texcoord_slot[i];
+         v->data[j][0] = tc[0];
+         if (wide->texcoord_mode[i] == PIPE_SPRITE_COORD_LOWER_LEFT)
+            v->data[j][1] = 1.0f - tc[1];
+         else
+            v->data[j][1] = tc[1];
+         v->data[j][2] = tc[2];
+         v->data[j][3] = tc[3];
+      }
+   }
+}
+
+
+/* If there are lots of sprite points (and why wouldn't there be?) it
+ * would probably be more sensible to change hardware setup to
+ * optimize this rather than doing the whole thing in software like
+ * this.
+ */
+static void wide_point( struct draw_stage *stage,
+			struct prim_header *header )
+{
+   const struct wide_stage *wide = wide_stage(stage);
+   const boolean sprite = (boolean) stage->draw->rasterizer->point_sprite;
+   float half_size;
+   float left_adj, right_adj;
+
+   struct prim_header tri;
+
+   /* four dups of original vertex */
+   struct vertex_header *v0 = dup_vert(stage, header->v[0], 0);
+   struct vertex_header *v1 = dup_vert(stage, header->v[0], 1);
+   struct vertex_header *v2 = dup_vert(stage, header->v[0], 2);
+   struct vertex_header *v3 = dup_vert(stage, header->v[0], 3);
+
+   float *pos0 = v0->data[0];
+   float *pos1 = v1->data[0];
+   float *pos2 = v2->data[0];
+   float *pos3 = v3->data[0];
+
+   /* point size is either per-vertex or fixed size */
+   if (wide->psize_slot >= 0) {
+      half_size = 0.5f * header->v[0]->data[wide->psize_slot][0];
+   }
+   else {
+      half_size = wide->half_point_size;
+   }
+
+   left_adj = -half_size + 0.25f;
+   right_adj = half_size + 0.25f;
+
+   pos0[0] += left_adj;
+   pos0[1] -= half_size;
+
+   pos1[0] += left_adj;
+   pos1[1] += half_size;
+
+   pos2[0] += right_adj;
+   pos2[1] -= half_size;
+
+   pos3[0] += right_adj;
+   pos3[1] += half_size;
+
+   if (sprite) {
+      static const float tex00[4] = { 0, 0, 0, 1 };
+      static const float tex01[4] = { 0, 1, 0, 1 };
+      static const float tex11[4] = { 1, 1, 0, 1 };
+      static const float tex10[4] = { 1, 0, 0, 1 };
+      set_texcoords( wide, v0, tex00 );
+      set_texcoords( wide, v1, tex01 );
+      set_texcoords( wide, v2, tex10 );
+      set_texcoords( wide, v3, tex11 );
+   }
+
+   tri.det = header->det;  /* only the sign matters */
+   tri.v[0] = v0;
+   tri.v[1] = v2;
+   tri.v[2] = v3;
+   stage->next->tri( stage->next, &tri );
+
+   tri.v[0] = v0;
+   tri.v[1] = v3;
+   tri.v[2] = v1;
+   stage->next->tri( stage->next, &tri );
+}
+
+
+static void wide_first_point( struct draw_stage *stage, 
+			      struct prim_header *header )
+{
+   struct wide_stage *wide = wide_stage(stage);
+   struct draw_context *draw = stage->draw;
+
+   wide->half_point_size = 0.5f * draw->rasterizer->point_size;
+
+   if (draw->rasterizer->point_size != 1.0) {
+      stage->point = wide_point;
+   }
+   else {
+      stage->point = passthrough_point;
+   }
+
+   if (draw->rasterizer->point_sprite) {
+      /* find vertex shader texcoord outputs */
+      const struct draw_vertex_shader *vs = draw->vertex_shader;
+      uint i, j = 0;
+      for (i = 0; i < vs->state->num_outputs; i++) {
+         if (vs->state->output_semantic_name[i] == TGSI_SEMANTIC_GENERIC) {
+            wide->texcoord_slot[j] = i;
+            wide->texcoord_mode[j] = draw->rasterizer->sprite_coord_mode[j];
+            j++;
+         }
+      }
+      wide->num_texcoords = j;
+   }
+
+   wide->psize_slot = -1;
+
+   if (draw->rasterizer->point_size_per_vertex) {
+      /* find PSIZ vertex output */
+      const struct draw_vertex_shader *vs = draw->vertex_shader;
+      uint i;
+      for (i = 0; i < vs->state->num_outputs; i++) {
+         if (vs->state->output_semantic_name[i] == TGSI_SEMANTIC_PSIZE) {
+            wide->psize_slot = i;
+            break;
+         }
+      }
+   }
+   
+   stage->point( stage, header );
+}
+
+
+
+static void wide_first_line( struct draw_stage *stage,
+			     struct prim_header *header )
+{
+   struct wide_stage *wide = wide_stage(stage);
+   struct draw_context *draw = stage->draw;
+
+   wide->half_line_width = 0.5f * draw->rasterizer->line_width;
+
+   if (draw->rasterizer->line_width != 1.0) {
+      if (draw->rasterizer->line_smooth)
+         wide->stage.line = wide_line_aa;
+      else
+         wide->stage.line = wide_line;
+   }
+   else {
+      wide->stage.line = passthrough_line;
+   }
+   
+   stage->line( stage, header );
+}
+
+
+static void wide_flush( struct draw_stage *stage, unsigned flags )
+{
+   stage->line = wide_first_line;
+   stage->point = wide_first_point;
+   stage->next->flush( stage->next, flags );
+}
+
+
+static void wide_reset_stipple_counter( struct draw_stage *stage )
+{
+   stage->next->reset_stipple_counter( stage->next );
+}
+
+
+static void wide_destroy( struct draw_stage *stage )
+{
+   draw_free_temp_verts( stage );
+   FREE( stage );
+}
+
+
+struct draw_stage *draw_wide_stage( struct draw_context *draw )
+{
+   struct wide_stage *wide = CALLOC_STRUCT(wide_stage);
+
+   draw_alloc_temp_verts( &wide->stage, 4 );
+
+   wide->stage.draw = draw;
+   wide->stage.next = NULL;
+   wide->stage.point = wide_first_point;
+   wide->stage.line = wide_first_line;
+   wide->stage.tri = passthrough_tri;
+   wide->stage.flush = wide_flush;
+   wide->stage.reset_stipple_counter = wide_reset_stipple_counter;
+   wide->stage.destroy = wide_destroy;
+
+   return &wide->stage;
+}
diff --git a/src/gallium/aux/llvm/Makefile b/src/gallium/aux/llvm/Makefile
new file mode 100644
index 0000000000..9c6e16d86b
--- /dev/null
+++ b/src/gallium/aux/llvm/Makefile
@@ -0,0 +1,83 @@
+# -*-makefile-*-
+TOP = ../../../..
+include $(TOP)/configs/current
+
+LIBNAME = gallivm
+
+
+GALLIVM_SOURCES = \
+        gallivm.cpp  \
+        gallivm_cpu.cpp \
+        instructions.cpp  \
+        loweringpass.cpp \
+        tgsitollvm.cpp \
+        storage.cpp \
+        storagesoa.cpp \
+        instructionssoa.cpp
+
+INC_SOURCES = gallivm_builtins.cpp
+
+CPP_SOURCES = \
+	$(GALLIVM_SOURCES)
+
+C_SOURCES =
+ASM_SOURCES =
+
+OBJECTS = $(C_SOURCES:.c=.o) \
+          $(CPP_SOURCES:.cpp=.o) \
+	  $(ASM_SOURCES:.S=.o)
+
+### Include directories
+INCLUDES = \
+	-I. \
+	-I$(TOP)/src/mesa/pipe \
+	-I$(TOP)/src/mesa \
+	-I$(TOP)/include
+
+
+##### RULES #####
+
+.c.o:
+	$(CC) -c $(INCLUDES) $(LLVM_CFLAGS) $(CFLAGS) $(DRIVER_DEFINES) $< -o $@
+
+.cpp.o:
+	$(CXX) -c $(INCLUDES) $(LLVM_CXXFLAGS) $(CXXFLAGS) $(DRIVER_DEFINES) $< -o $@
+
+.S.o:
+	$(CC) -c $(INCLUDES) $(CFLAGS) $(DRIVER_DEFINES)  $< -o $@
+
+##### TARGETS #####
+
+default:: depend symlinks $(LIBNAME)
+
+
+$(LIBNAME): $(OBJECTS) Makefile
+	$(TOP)/bin/mklib -o $@ -static $(OBJECTS)
+
+
+depend: $(C_SOURCES) $(CPP_SOURCES) $(ASM_SOURCES) $(INC_SOURCES)
+	rm -f depend
+	touch depend
+	$(MKDEP) $(MKDEP_OPTIONS) $(DRIVER_DEFINES) $(INCLUDES) $(C_SOURCES) $(CPP_SOURCES) \
+		$(ASM_SOURCES) $(INC_SOURCES) 2> /dev/null
+
+
+gallivm_builtins.cpp: llvm_builtins.c
+	clang --emit-llvm $< |llvm-as|opt -std-compile-opts|llvm2cpp -gen-contents -o=$@ -f -for=shader -funcname=createGallivmBuiltins
+
+
+# Emacs tags
+tags:
+	etags `find . -name \*.[ch]` `find ../include`
+
+
+# Remove .o and backup files
+clean:
+	-rm -f *.o */*.o *~ *.so *~ server/*.o
+	-rm -f depend depend.bak
+	-rm -f gallivm_builtins.cpp
+
+symlinks:
+
+
+include depend
diff --git a/src/gallium/aux/llvm/gallivm.cpp b/src/gallium/aux/llvm/gallivm.cpp
new file mode 100644
index 0000000000..da0105c2c9
--- /dev/null
+++ b/src/gallium/aux/llvm/gallivm.cpp
@@ -0,0 +1,327 @@
+/**************************************************************************
+ *
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ **************************************************************************/
+
+ /*
+  * Authors:
+  *   Zack Rusin zack@tungstengraphics.com
+  */
+#ifdef MESA_LLVM
+
+#include "gallivm.h"
+#include "gallivm_p.h"
+
+#include "instructions.h"
+#include "loweringpass.h"
+#include "storage.h"
+#include "tgsitollvm.h"
+
+#include "pipe/p_context.h"
+#include "pipe/p_shader_tokens.h"
+
+#include "pipe/tgsi/exec/tgsi_exec.h"
+#include "pipe/tgsi/util/tgsi_dump.h"
+
+#include <llvm/Module.h>
+#include <llvm/CallingConv.h>
+#include <llvm/Constants.h>
+#include <llvm/DerivedTypes.h>
+#include <llvm/Instructions.h>
+#include <llvm/ModuleProvider.h>
+#include <llvm/Pass.h>
+#include <llvm/PassManager.h>
+#include <llvm/ParameterAttributes.h>
+#include <llvm/Support/PatternMatch.h>
+#include <llvm/ExecutionEngine/JIT.h>
+#include <llvm/ExecutionEngine/Interpreter.h>
+#include <llvm/ExecutionEngine/GenericValue.h>
+#include <llvm/Support/MemoryBuffer.h>
+#include <llvm/LinkAllPasses.h>
+#include <llvm/Analysis/Verifier.h>
+#include <llvm/Analysis/LoopPass.h>
+#include <llvm/Target/TargetData.h>
+#include <llvm/Bitcode/ReaderWriter.h>
+#include <llvm/Transforms/Utils/Cloning.h>
+
+#include <sstream>
+#include <fstream>
+#include <iostream>
+
+static int GLOBAL_ID = 0;
+
+using namespace llvm;
+
+static inline
+void AddStandardCompilePasses(PassManager &PM)
+{
+   PM.add(new LoweringPass());
+   PM.add(createVerifierPass());                  // Verify that input is correct
+
+   PM.add(createLowerSetJmpPass());          // Lower llvm.setjmp/.longjmp
+
+   //PM.add(createStripSymbolsPass(true));
+
+   PM.add(createRaiseAllocationsPass());     // call %malloc -> malloc inst
+   PM.add(createCFGSimplificationPass());    // Clean up disgusting code
+   PM.add(createPromoteMemoryToRegisterPass());// Kill useless allocas
+   PM.add(createGlobalOptimizerPass());      // Optimize out global vars
+   PM.add(createGlobalDCEPass());            // Remove unused fns and globs
+   PM.add(createIPConstantPropagationPass());// IP Constant Propagation
+   PM.add(createDeadArgEliminationPass());   // Dead argument elimination
+   PM.add(createInstructionCombiningPass()); // Clean up after IPCP & DAE
+   PM.add(createCFGSimplificationPass());    // Clean up after IPCP & DAE
+
+   PM.add(createPruneEHPass());              // Remove dead EH info
+
+   PM.add(createFunctionInliningPass());   // Inline small functions
+   PM.add(createArgumentPromotionPass());    // Scalarize uninlined fn args
+
+   PM.add(createTailDuplicationPass());      // Simplify cfg by copying code
+   PM.add(createInstructionCombiningPass()); // Cleanup for scalarrepl.
+   PM.add(createCFGSimplificationPass());    // Merge & remove BBs
+   PM.add(createScalarReplAggregatesPass()); // Break up aggregate allocas
+   PM.add(createInstructionCombiningPass()); // Combine silly seq's
+   PM.add(createCondPropagationPass());      // Propagate conditionals
+
+   PM.add(createTailCallEliminationPass());  // Eliminate tail calls
+   PM.add(createCFGSimplificationPass());    // Merge & remove BBs
+   PM.add(createReassociatePass());          // Reassociate expressions
+   PM.add(createLoopRotatePass());
+   PM.add(createLICMPass());                 // Hoist loop invariants
+   PM.add(createLoopUnswitchPass());         // Unswitch loops.
+   PM.add(createLoopIndexSplitPass());       // Index split loops.
+   PM.add(createInstructionCombiningPass()); // Clean up after LICM/reassoc
+   PM.add(createIndVarSimplifyPass());       // Canonicalize indvars
+   PM.add(createLoopUnrollPass());           // Unroll small loops
+   PM.add(createInstructionCombiningPass()); // Clean up after the unroller
+   PM.add(createGVNPass());                  // Remove redundancies
+   PM.add(createSCCPPass());                 // Constant prop with SCCP
+
+   // Run instcombine after redundancy elimination to exploit opportunities
+   // opened up by them.
+   PM.add(createInstructionCombiningPass());
+   PM.add(createCondPropagationPass());      // Propagate conditionals
+
+   PM.add(createDeadStoreEliminationPass()); // Delete dead stores
+   PM.add(createAggressiveDCEPass());        // SSA based 'Aggressive DCE'
+   PM.add(createCFGSimplificationPass());    // Merge & remove BBs
+   PM.add(createSimplifyLibCallsPass());     // Library Call Optimizations
+   PM.add(createDeadTypeEliminationPass());  // Eliminate dead types
+   PM.add(createConstantMergePass());        // Merge dup global constants
+}
+
+void gallivm_prog_delete(struct gallivm_prog *prog)
+{
+   delete prog->module;
+   prog->module = 0;
+   prog->function = 0;
+   free(prog);
+}
+
+static inline void
+constant_interpolation(float (*inputs)[16][4],
+                       const struct tgsi_interp_coef *coefs,
+                       unsigned attrib,
+                       unsigned chan)
+{
+   unsigned i;
+
+   for (i = 0; i < QUAD_SIZE; ++i) {
+      inputs[i][attrib][chan] = coefs[attrib].a0[chan];
+   }
+}
+
+static inline void
+linear_interpolation(float (*inputs)[16][4],
+                     const struct tgsi_interp_coef *coefs,
+                     unsigned attrib,
+                     unsigned chan)
+{
+   unsigned i;
+
+   for( i = 0; i < QUAD_SIZE; i++ ) {
+      const float x = inputs[i][0][0];
+      const float y = inputs[i][0][1];
+
+      inputs[i][attrib][chan] =
+         coefs[attrib].a0[chan] +
+         coefs[attrib].dadx[chan] * x +
+         coefs[attrib].dady[chan] * y;
+   }
+}
+
+static inline void
+perspective_interpolation(float (*inputs)[16][4],
+                          const struct tgsi_interp_coef *coefs,
+                          unsigned attrib,
+                          unsigned chan )
+{
+   unsigned i;
+
+   for( i = 0; i < QUAD_SIZE; i++ ) {
+      const float x = inputs[i][0][0];
+      const float y = inputs[i][0][1];
+      /* WPOS.w here is really 1/w */
+      const float w = 1.0f / inputs[i][0][3];
+      assert(inputs[i][0][3] != 0.0);
+
+      inputs[i][attrib][chan] =
+         (coefs[attrib].a0[chan] +
+          coefs[attrib].dadx[chan] * x +
+          coefs[attrib].dady[chan] * y) * w;
+   }
+}
+
+void gallivm_ir_dump(struct gallivm_ir *ir, const char *file_prefix)
+{
+   if (!ir || !ir->module)
+      return;
+
+   if (file_prefix) {
+      std::ostringstream stream;
+      stream << file_prefix;
+      stream << ir->id;
+      stream << ".ll";
+      std::string name = stream.str();
+      std::ofstream out(name.c_str());
+      if (!out) {
+         std::cerr<<"Can't open file : "<<stream.str()<<std::endl;;
+         return;
+      }
+      out << (*ir->module);
+      out.close();
+   } else {
+      const llvm::Module::FunctionListType &funcs = ir->module->getFunctionList();
+      llvm::Module::FunctionListType::const_iterator itr;
+      std::cout<<"; ---------- Start shader "<<ir->id<<std::endl;
+      for (itr = funcs.begin(); itr != funcs.end(); ++itr) {
+         const llvm::Function &func = (*itr);
+         std::string name = func.getName();
+         const llvm::Function *found = 0;
+         if (name.find("vs_shader") != std::string::npos ||
+             name.find("fs_shader") != std::string::npos ||
+             name.find("function") != std::string::npos)
+            found = &func;
+         if (found) {
+            std::cout<<*found<<std::endl;
+         }
+      }
+      std::cout<<"; ---------- End shader "<<ir->id<<std::endl;
+   }
+}
+
+
+void gallivm_prog_inputs_interpolate(struct gallivm_prog *prog,
+                                     float (*inputs)[16][4],
+                                     const struct tgsi_interp_coef *coef)
+{
+   for (int i = 0; i < prog->num_interp; ++i) {
+      const gallivm_interpolate &interp = prog->interpolators[i];
+      switch (interp.type) {
+      case TGSI_INTERPOLATE_CONSTANT:
+         constant_interpolation(inputs, coef, interp.attrib, interp.chan);
+         break;
+
+      case TGSI_INTERPOLATE_LINEAR:
+         linear_interpolation(inputs, coef, interp.attrib, interp.chan);
+         break;
+
+      case TGSI_INTERPOLATE_PERSPECTIVE:
+         perspective_interpolation(inputs, coef, interp.attrib, interp.chan);
+         break;
+
+      default:
+         assert( 0 );
+      }
+   }
+}
+
+
+struct gallivm_ir * gallivm_ir_new(enum gallivm_shader_type type)
+{
+   struct gallivm_ir *ir =
+      (struct gallivm_ir *)calloc(1, sizeof(struct gallivm_ir));
+   ++GLOBAL_ID;
+   ir->id   = GLOBAL_ID;
+   ir->type = type;
+
+   return ir;
+}
+
+void gallivm_ir_set_layout(struct gallivm_ir *ir,
+                           enum gallivm_vector_layout layout)
+{
+   ir->layout = layout;
+}
+
+void gallivm_ir_set_components(struct gallivm_ir *ir, int num)
+{
+   ir->num_components = num;
+}
+
+void gallivm_ir_fill_from_tgsi(struct gallivm_ir *ir,
+                               const struct tgsi_token *tokens)
+{
+   std::cout << "Creating llvm from: " <<std::endl;
+   tgsi_dump(tokens, 0);
+
+
+   llvm::Module *mod = tgsi_to_llvmir(ir, tokens);
+
+   //llvm::Module *mod = tgsi_to_llvm(ir, tokens);
+   ir->module = mod;
+   gallivm_ir_dump(ir, 0);
+}
+
+void gallivm_ir_delete(struct gallivm_ir *ir)
+{
+   delete ir->module;
+   free(ir);
+}
+
+struct gallivm_prog * gallivm_ir_compile(struct gallivm_ir *ir)
+{
+   struct gallivm_prog *prog =
+      (struct gallivm_prog *)calloc(1, sizeof(struct gallivm_prog));
+   llvm::Module *mod = llvm::CloneModule(ir->module);
+   prog->num_consts = ir->num_consts;
+   memcpy(prog->interpolators, ir->interpolators, sizeof(prog->interpolators));
+   prog->num_interp = ir->num_interp;
+
+   /* Run optimization passes over it */
+   PassManager passes;
+   passes.add(new TargetData(mod));
+   AddStandardCompilePasses(passes);
+   passes.run(*mod);
+   prog->module = mod;
+
+   std::cout << "After optimizations:"<<std::endl;
+   mod->dump();
+
+   return prog;
+}
+
+#endif /* MESA_LLVM */
diff --git a/src/gallium/aux/llvm/gallivm.h b/src/gallium/aux/llvm/gallivm.h
new file mode 100644
index 0000000000..92da4bca7f
--- /dev/null
+++ b/src/gallium/aux/llvm/gallivm.h
@@ -0,0 +1,103 @@
+/**************************************************************************
+ *
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ **************************************************************************/
+
+ /*
+  * Authors:
+  *   Zack Rusin zack@tungstengraphics.com
+  */
+
+#ifndef GALLIVM_H
+#define GALLIVM_H
+
+#if defined __cplusplus
+extern "C" {
+#endif
+
+#include "pipe/p_state.h"
+
+#ifdef MESA_LLVM
+
+struct tgsi_token;
+
+struct gallivm_ir;
+struct gallivm_prog;
+struct gallivm_cpu_engine;
+struct tgsi_interp_coef;
+struct tgsi_sampler;
+struct tgsi_exec_vector;
+
+enum gallivm_shader_type {
+   GALLIVM_VS,
+   GALLIVM_FS
+};
+
+enum gallivm_vector_layout {
+   GALLIVM_AOS,
+   GALLIVM_SOA
+};
+
+struct gallivm_ir *gallivm_ir_new(enum gallivm_shader_type type);
+void               gallivm_ir_set_layout(struct gallivm_ir *ir,
+                                         enum gallivm_vector_layout layout);
+void               gallivm_ir_set_components(struct gallivm_ir *ir, int num);
+void               gallivm_ir_fill_from_tgsi(struct gallivm_ir *ir,
+                                             const struct tgsi_token *tokens);
+void               gallivm_ir_delete(struct gallivm_ir *ir);
+
+
+struct gallivm_prog *gallivm_ir_compile(struct gallivm_ir *ir);
+
+void gallivm_prog_inputs_interpolate(struct gallivm_prog *prog,
+                                     float (*inputs)[PIPE_MAX_SHADER_INPUTS][4],
+                                     const struct tgsi_interp_coef *coefs);
+void gallivm_prog_dump(struct gallivm_prog *prog, const char *file_prefix);
+
+
+struct gallivm_cpu_engine *gallivm_cpu_engine_create(struct gallivm_prog *prog);
+struct gallivm_cpu_engine *gallivm_global_cpu_engine();
+int gallivm_cpu_vs_exec(struct gallivm_prog *prog,
+                        struct tgsi_exec_vector       *inputs,
+                        struct tgsi_exec_vector       *dests,
+                        float (*consts)[4],
+                        struct tgsi_exec_vector       *temps);
+int gallivm_cpu_fs_exec(struct gallivm_prog *prog,
+                        float x, float y,
+                        float (*dests)[PIPE_MAX_SHADER_INPUTS][4],
+                        float (*inputs)[PIPE_MAX_SHADER_INPUTS][4],
+                        float (*consts)[4],
+                        struct tgsi_sampler *samplers);
+void gallivm_cpu_jit_compile(struct gallivm_cpu_engine *ee, struct gallivm_prog *prog);
+void gallivm_cpu_engine_delete(struct gallivm_cpu_engine *ee);
+
+
+#endif /* MESA_LLVM */
+
+#if defined __cplusplus
+} // extern "C"
+#endif
+
+#endif
diff --git a/src/gallium/aux/llvm/gallivm_builtins.cpp b/src/gallium/aux/llvm/gallivm_builtins.cpp
new file mode 100644
index 0000000000..1796f0a177
--- /dev/null
+++ b/src/gallium/aux/llvm/gallivm_builtins.cpp
@@ -0,0 +1,567 @@
+// Generated by llvm2cpp - DO NOT MODIFY!
+
+
+Module* createGallivmBuiltins(Module *mod) {
+
+mod->setModuleIdentifier("shader");
+
+// Type Definitions
+ArrayType* ArrayTy_0 = ArrayType::get(IntegerType::get(8), 25);
+
+PointerType* PointerTy_1 = PointerType::get(ArrayTy_0, 0);
+
+std::vector<const Type*>FuncTy_2_args;
+FuncTy_2_args.push_back(Type::FloatTy);
+FuncTy_2_args.push_back(Type::FloatTy);
+FunctionType* FuncTy_2 = FunctionType::get(
+  /*Result=*/Type::FloatTy,
+  /*Params=*/FuncTy_2_args,
+  /*isVarArg=*/false);
+
+PointerType* PointerTy_3 = PointerType::get(FuncTy_2, 0);
+
+VectorType* VectorTy_4 = VectorType::get(Type::FloatTy, 4);
+
+std::vector<const Type*>FuncTy_5_args;
+FuncTy_5_args.push_back(VectorTy_4);
+FunctionType* FuncTy_5 = FunctionType::get(
+  /*Result=*/VectorTy_4,
+  /*Params=*/FuncTy_5_args,
+  /*isVarArg=*/false);
+
+std::vector<const Type*>FuncTy_6_args;
+FuncTy_6_args.push_back(VectorTy_4);
+FuncTy_6_args.push_back(VectorTy_4);
+FuncTy_6_args.push_back(VectorTy_4);
+FunctionType* FuncTy_6 = FunctionType::get(
+  /*Result=*/VectorTy_4,
+  /*Params=*/FuncTy_6_args,
+  /*isVarArg=*/false);
+
+VectorType* VectorTy_7 = VectorType::get(IntegerType::get(32), 4);
+
+std::vector<const Type*>FuncTy_9_args;
+FunctionType* FuncTy_9 = FunctionType::get(
+  /*Result=*/IntegerType::get(32),
+  /*Params=*/FuncTy_9_args,
+  /*isVarArg=*/true);
+
+PointerType* PointerTy_8 = PointerType::get(FuncTy_9, 0);
+
+PointerType* PointerTy_10 = PointerType::get(IntegerType::get(8), 0);
+
+std::vector<const Type*>FuncTy_12_args;
+FuncTy_12_args.push_back(Type::FloatTy);
+FunctionType* FuncTy_12 = FunctionType::get(
+  /*Result=*/Type::FloatTy,
+  /*Params=*/FuncTy_12_args,
+  /*isVarArg=*/false);
+
+PointerType* PointerTy_11 = PointerType::get(FuncTy_12, 0);
+
+std::vector<const Type*>FuncTy_13_args;
+FuncTy_13_args.push_back(VectorTy_4);
+FunctionType* FuncTy_13 = FunctionType::get(
+  /*Result=*/IntegerType::get(32),
+  /*Params=*/FuncTy_13_args,
+  /*isVarArg=*/false);
+
+
+// Function Declarations
+
+Function* func_approx = new Function(
+  /*Type=*/FuncTy_2,
+  /*Linkage=*/GlobalValue::WeakLinkage,
+  /*Name=*/"approx", mod); 
+func_approx->setCallingConv(CallingConv::C);
+const ParamAttrsList *func_approx_PAL = 0;
+func_approx->setParamAttrs(func_approx_PAL);
+
+Function* func_powf = new Function(
+  /*Type=*/FuncTy_2,
+  /*Linkage=*/GlobalValue::ExternalLinkage,
+  /*Name=*/"powf", mod); // (external, no body)
+func_powf->setCallingConv(CallingConv::C);
+const ParamAttrsList *func_powf_PAL = 0;
+func_powf->setParamAttrs(func_powf_PAL);
+
+Function* func_lit = new Function(
+  /*Type=*/FuncTy_5,
+  /*Linkage=*/GlobalValue::WeakLinkage,
+  /*Name=*/"lit", mod); 
+func_lit->setCallingConv(CallingConv::C);
+const ParamAttrsList *func_lit_PAL = 0;
+func_lit->setParamAttrs(func_lit_PAL);
+
+Function* func_cmp = new Function(
+  /*Type=*/FuncTy_6,
+  /*Linkage=*/GlobalValue::WeakLinkage,
+  /*Name=*/"cmp", mod); 
+func_cmp->setCallingConv(CallingConv::C);
+const ParamAttrsList *func_cmp_PAL = 0;
+{
+  ParamAttrsVector Attrs;
+  ParamAttrsWithIndex PAWI;
+  PAWI.index = 0; PAWI.attrs = 0  | ParamAttr::NoUnwind;
+  Attrs.push_back(PAWI);
+  func_cmp_PAL = ParamAttrsList::get(Attrs);
+  
+}
+func_cmp->setParamAttrs(func_cmp_PAL);
+
+Function* func_vcos = new Function(
+  /*Type=*/FuncTy_5,
+  /*Linkage=*/GlobalValue::WeakLinkage,
+  /*Name=*/"vcos", mod); 
+func_vcos->setCallingConv(CallingConv::C);
+const ParamAttrsList *func_vcos_PAL = 0;
+func_vcos->setParamAttrs(func_vcos_PAL);
+
+Function* func_printf = new Function(
+  /*Type=*/FuncTy_9,
+  /*Linkage=*/GlobalValue::ExternalLinkage,
+  /*Name=*/"printf", mod); // (external, no body)
+func_printf->setCallingConv(CallingConv::C);
+const ParamAttrsList *func_printf_PAL = 0;
+func_printf->setParamAttrs(func_printf_PAL);
+
+Function* func_cosf = new Function(
+  /*Type=*/FuncTy_12,
+  /*Linkage=*/GlobalValue::ExternalLinkage,
+  /*Name=*/"cosf", mod); // (external, no body)
+func_cosf->setCallingConv(CallingConv::C);
+const ParamAttrsList *func_cosf_PAL = 0;
+func_cosf->setParamAttrs(func_cosf_PAL);
+
+Function* func_scs = new Function(
+  /*Type=*/FuncTy_5,
+  /*Linkage=*/GlobalValue::WeakLinkage,
+  /*Name=*/"scs", mod); 
+func_scs->setCallingConv(CallingConv::C);
+const ParamAttrsList *func_scs_PAL = 0;
+func_scs->setParamAttrs(func_scs_PAL);
+
+Function* func_sinf = new Function(
+  /*Type=*/FuncTy_12,
+  /*Linkage=*/GlobalValue::ExternalLinkage,
+  /*Name=*/"sinf", mod); // (external, no body)
+func_sinf->setCallingConv(CallingConv::C);
+const ParamAttrsList *func_sinf_PAL = 0;
+func_sinf->setParamAttrs(func_sinf_PAL);
+
+Function* func_vsin = new Function(
+  /*Type=*/FuncTy_5,
+  /*Linkage=*/GlobalValue::WeakLinkage,
+  /*Name=*/"vsin", mod); 
+func_vsin->setCallingConv(CallingConv::C);
+const ParamAttrsList *func_vsin_PAL = 0;
+func_vsin->setParamAttrs(func_vsin_PAL);
+
+Function* func_kilp = new Function(
+  /*Type=*/FuncTy_13,
+  /*Linkage=*/GlobalValue::WeakLinkage,
+  /*Name=*/"kilp", mod); 
+func_kilp->setCallingConv(CallingConv::C);
+const ParamAttrsList *func_kilp_PAL = 0;
+{
+  ParamAttrsVector Attrs;
+  ParamAttrsWithIndex PAWI;
+  PAWI.index = 0; PAWI.attrs = 0  | ParamAttr::NoUnwind;
+  Attrs.push_back(PAWI);
+  func_kilp_PAL = ParamAttrsList::get(Attrs);
+  
+}
+func_kilp->setParamAttrs(func_kilp_PAL);
+
+// Global Variable Declarations
+
+
+GlobalVariable* gvar_array__str = new GlobalVariable(
+/*Type=*/ArrayTy_0,
+/*isConstant=*/true,
+/*Linkage=*/GlobalValue::InternalLinkage,
+/*Initializer=*/0, // has initializer, specified below
+/*Name=*/".str",
+mod);
+
+GlobalVariable* gvar_array__str1 = new GlobalVariable(
+/*Type=*/ArrayTy_0,
+/*isConstant=*/true,
+/*Linkage=*/GlobalValue::InternalLinkage,
+/*Initializer=*/0, // has initializer, specified below
+/*Name=*/".str1",
+mod);
+
+// Constant Definitions
+Constant* const_array_14 = ConstantArray::get("VEC IN   is %f %f %f %f\x0A", true);
+Constant* const_array_15 = ConstantArray::get("VEC OUT  is %f %f %f %f\x0A", true);
+ConstantFP* const_float_16 = ConstantFP::get(Type::FloatTy, APFloat(-1.280000e+02f));
+ConstantFP* const_float_17 = ConstantFP::get(Type::FloatTy, APFloat(1.280000e+02f));
+Constant* const_float_18 = Constant::getNullValue(Type::FloatTy);
+Constant* const_int32_19 = Constant::getNullValue(IntegerType::get(32));
+std::vector<Constant*> const_packed_20_elems;
+ConstantFP* const_float_21 = ConstantFP::get(Type::FloatTy, APFloat(1.000000e+00f));
+const_packed_20_elems.push_back(const_float_21);
+UndefValue* const_float_22 = UndefValue::get(Type::FloatTy);
+const_packed_20_elems.push_back(const_float_22);
+const_packed_20_elems.push_back(const_float_22);
+const_packed_20_elems.push_back(const_float_21);
+Constant* const_packed_20 = ConstantVector::get(VectorTy_4, const_packed_20_elems);
+ConstantInt* const_int32_23 = ConstantInt::get(APInt(32,  "1", 10));
+ConstantInt* const_int32_24 = ConstantInt::get(APInt(32,  "3", 10));
+ConstantInt* const_int32_25 = ConstantInt::get(APInt(32,  "2", 10));
+std::vector<Constant*> const_packed_26_elems;
+const_packed_26_elems.push_back(const_float_21);
+const_packed_26_elems.push_back(const_float_18);
+const_packed_26_elems.push_back(const_float_18);
+const_packed_26_elems.push_back(const_float_21);
+Constant* const_packed_26 = ConstantVector::get(VectorTy_4, const_packed_26_elems);
+Constant* const_double_27 = Constant::getNullValue(Type::DoubleTy);
+std::vector<Constant*> const_packed_28_elems;
+const_packed_28_elems.push_back(const_int32_19);
+ConstantInt* const_int32_29 = ConstantInt::get(APInt(32,  "5", 10));
+const_packed_28_elems.push_back(const_int32_29);
+const_packed_28_elems.push_back(const_int32_25);
+const_packed_28_elems.push_back(const_int32_24);
+Constant* const_packed_28 = ConstantVector::get(VectorTy_7, const_packed_28_elems);
+std::vector<Constant*> const_packed_30_elems;
+const_packed_30_elems.push_back(const_int32_19);
+const_packed_30_elems.push_back(const_int32_23);
+ConstantInt* const_int32_31 = ConstantInt::get(APInt(32,  "6", 10));
+const_packed_30_elems.push_back(const_int32_31);
+const_packed_30_elems.push_back(const_int32_24);
+Constant* const_packed_30 = ConstantVector::get(VectorTy_7, const_packed_30_elems);
+std::vector<Constant*> const_packed_32_elems;
+const_packed_32_elems.push_back(const_int32_19);
+const_packed_32_elems.push_back(const_int32_23);
+const_packed_32_elems.push_back(const_int32_25);
+ConstantInt* const_int32_33 = ConstantInt::get(APInt(32,  "7", 10));
+const_packed_32_elems.push_back(const_int32_33);
+Constant* const_packed_32 = ConstantVector::get(VectorTy_7, const_packed_32_elems);
+std::vector<Constant*> const_ptr_34_indices;
+const_ptr_34_indices.push_back(const_int32_19);
+const_ptr_34_indices.push_back(const_int32_19);
+Constant* const_ptr_34 = ConstantExpr::getGetElementPtr(gvar_array__str, &const_ptr_34_indices[0], const_ptr_34_indices.size() );
+UndefValue* const_packed_35 = UndefValue::get(VectorTy_4);
+std::vector<Constant*> const_ptr_36_indices;
+const_ptr_36_indices.push_back(const_int32_19);
+const_ptr_36_indices.push_back(const_int32_19);
+Constant* const_ptr_36 = ConstantExpr::getGetElementPtr(gvar_array__str1, &const_ptr_36_indices[0], const_ptr_36_indices.size() );
+
+// Global Variable Definitions
+gvar_array__str->setInitializer(const_array_14);
+gvar_array__str1->setInitializer(const_array_15);
+
+// Function Definitions
+
+// Function: approx (func_approx)
+{
+  Function::arg_iterator args = func_approx->arg_begin();
+  Value* float_a = args++;
+  float_a->setName("a");
+  Value* float_b = args++;
+  float_b->setName("b");
+  
+  BasicBlock* label_entry = new BasicBlock("entry",func_approx,0);
+  
+  // Block entry (label_entry)
+  FCmpInst* int1_cmp = new FCmpInst(FCmpInst::FCMP_OLT, float_b, const_float_16, "cmp", label_entry);
+  SelectInst* float_b_addr_0 = new SelectInst(int1_cmp, const_float_16, float_b, "b.addr.0", label_entry);
+  FCmpInst* int1_cmp3 = new FCmpInst(FCmpInst::FCMP_OGT, float_b_addr_0, const_float_17, "cmp3", label_entry);
+  SelectInst* float_b_addr_1 = new SelectInst(int1_cmp3, const_float_17, float_b_addr_0, "b.addr.1", label_entry);
+  FCmpInst* int1_cmp7 = new FCmpInst(FCmpInst::FCMP_OLT, float_a, const_float_18, "cmp7", label_entry);
+  SelectInst* float_a_addr_0 = new SelectInst(int1_cmp7, const_float_18, float_a, "a.addr.0", label_entry);
+  std::vector<Value*> float_call_params;
+  float_call_params.push_back(float_a_addr_0);
+  float_call_params.push_back(float_b_addr_1);
+  CallInst* float_call = new CallInst(func_powf, float_call_params.begin(), float_call_params.end(), "call", label_entry);
+  float_call->setCallingConv(CallingConv::C);
+  float_call->setTailCall(true);const ParamAttrsList *float_call_PAL = 0;
+  float_call->setParamAttrs(float_call_PAL);
+  
+  new ReturnInst(float_call, label_entry);
+  
+}
+
+// Function: lit (func_lit)
+{
+  Function::arg_iterator args = func_lit->arg_begin();
+  Value* packed_tmp = args++;
+  packed_tmp->setName("tmp");
+  
+  BasicBlock* label_entry_38 = new BasicBlock("entry",func_lit,0);
+  BasicBlock* label_ifthen = new BasicBlock("ifthen",func_lit,0);
+  BasicBlock* label_UnifiedReturnBlock = new BasicBlock("UnifiedReturnBlock",func_lit,0);
+  
+  // Block entry (label_entry_38)
+  ExtractElementInst* float_tmp6 = new ExtractElementInst(packed_tmp, const_int32_19, "tmp6", label_entry_38);
+  FCmpInst* int1_cmp_39 = new FCmpInst(FCmpInst::FCMP_OGT, float_tmp6, const_float_18, "cmp", label_entry_38);
+  new BranchInst(label_ifthen, label_UnifiedReturnBlock, int1_cmp_39, label_entry_38);
+  
+  // Block ifthen (label_ifthen)
+  InsertElementInst* packed_tmp10 = new InsertElementInst(const_packed_20, float_tmp6, const_int32_23, "tmp10", label_ifthen);
+  ExtractElementInst* float_tmp12 = new ExtractElementInst(packed_tmp, const_int32_23, "tmp12", label_ifthen);
+  ExtractElementInst* float_tmp14 = new ExtractElementInst(packed_tmp, const_int32_24, "tmp14", label_ifthen);
+  std::vector<Value*> float_call_41_params;
+  float_call_41_params.push_back(float_tmp12);
+  float_call_41_params.push_back(float_tmp14);
+  CallInst* float_call_41 = new CallInst(func_approx, float_call_41_params.begin(), float_call_41_params.end(), "call", label_ifthen);
+  float_call_41->setCallingConv(CallingConv::C);
+  float_call_41->setTailCall(true);const ParamAttrsList *float_call_41_PAL = 0;
+  float_call_41->setParamAttrs(float_call_41_PAL);
+  
+  InsertElementInst* packed_tmp16 = new InsertElementInst(packed_tmp10, float_call_41, const_int32_25, "tmp16", label_ifthen);
+  new ReturnInst(packed_tmp16, label_ifthen);
+  
+  // Block UnifiedReturnBlock (label_UnifiedReturnBlock)
+  new ReturnInst(const_packed_26, label_UnifiedReturnBlock);
+  
+}
+
+// Function: cmp (func_cmp)
+{
+  Function::arg_iterator args = func_cmp->arg_begin();
+  Value* packed_tmp0 = args++;
+  packed_tmp0->setName("tmp0");
+  Value* packed_tmp1 = args++;
+  packed_tmp1->setName("tmp1");
+  Value* packed_tmp2 = args++;
+  packed_tmp2->setName("tmp2");
+  
+  BasicBlock* label_entry_44 = new BasicBlock("entry",func_cmp,0);
+  BasicBlock* label_cond__14 = new BasicBlock("cond.?14",func_cmp,0);
+  BasicBlock* label_cond_cont20 = new BasicBlock("cond.cont20",func_cmp,0);
+  BasicBlock* label_cond__28 = new BasicBlock("cond.?28",func_cmp,0);
+  BasicBlock* label_cond_cont34 = new BasicBlock("cond.cont34",func_cmp,0);
+  BasicBlock* label_cond__42 = new BasicBlock("cond.?42",func_cmp,0);
+  BasicBlock* label_cond_cont48 = new BasicBlock("cond.cont48",func_cmp,0);
+  
+  // Block entry (label_entry_44)
+  ExtractElementInst* float_tmp3 = new ExtractElementInst(packed_tmp0, const_int32_19, "tmp3", label_entry_44);
+  CastInst* double_conv = new FPExtInst(float_tmp3, Type::DoubleTy, "conv", label_entry_44);
+  FCmpInst* int1_cmp_45 = new FCmpInst(FCmpInst::FCMP_OLT, double_conv, const_double_27, "cmp", label_entry_44);
+  ExtractElementInst* float_tmp11 = new ExtractElementInst(packed_tmp0, const_int32_23, "tmp11", label_entry_44);
+  CastInst* double_conv12 = new FPExtInst(float_tmp11, Type::DoubleTy, "conv12", label_entry_44);
+  FCmpInst* int1_cmp13 = new FCmpInst(FCmpInst::FCMP_OLT, double_conv12, const_double_27, "cmp13", label_entry_44);
+  SelectInst* packed_tmp1_tmp2 = new SelectInst(int1_cmp_45, packed_tmp1, packed_tmp2, "tmp1.tmp2", label_entry_44);
+  new BranchInst(label_cond__14, label_cond_cont20, int1_cmp13, label_entry_44);
+  
+  // Block cond.?14 (label_cond__14)
+  ShuffleVectorInst* packed_tmp233 = new ShuffleVectorInst(packed_tmp1_tmp2, packed_tmp1, const_packed_28, "tmp233", label_cond__14);
+  ExtractElementInst* float_tmp254 = new ExtractElementInst(packed_tmp0, const_int32_25, "tmp254", label_cond__14);
+  CastInst* double_conv265 = new FPExtInst(float_tmp254, Type::DoubleTy, "conv265", label_cond__14);
+  FCmpInst* int1_cmp276 = new FCmpInst(FCmpInst::FCMP_OLT, double_conv265, const_double_27, "cmp276", label_cond__14);
+  new BranchInst(label_cond__28, label_cond_cont34, int1_cmp276, label_cond__14);
+  
+  // Block cond.cont20 (label_cond_cont20)
+  ShuffleVectorInst* packed_tmp23 = new ShuffleVectorInst(packed_tmp1_tmp2, packed_tmp2, const_packed_28, "tmp23", label_cond_cont20);
+  ExtractElementInst* float_tmp25 = new ExtractElementInst(packed_tmp0, const_int32_25, "tmp25", label_cond_cont20);
+  CastInst* double_conv26 = new FPExtInst(float_tmp25, Type::DoubleTy, "conv26", label_cond_cont20);
+  FCmpInst* int1_cmp27 = new FCmpInst(FCmpInst::FCMP_OLT, double_conv26, const_double_27, "cmp27", label_cond_cont20);
+  new BranchInst(label_cond__28, label_cond_cont34, int1_cmp27, label_cond_cont20);
+  
+  // Block cond.?28 (label_cond__28)
+  PHINode* packed_tmp23_reg2mem_0 = new PHINode(VectorTy_4, "tmp23.reg2mem.0", label_cond__28);
+  packed_tmp23_reg2mem_0->reserveOperandSpace(2);
+  packed_tmp23_reg2mem_0->addIncoming(packed_tmp233, label_cond__14);
+  packed_tmp23_reg2mem_0->addIncoming(packed_tmp23, label_cond_cont20);
+  
+  ShuffleVectorInst* packed_tmp378 = new ShuffleVectorInst(packed_tmp23_reg2mem_0, packed_tmp1, const_packed_30, "tmp378", label_cond__28);
+  ExtractElementInst* float_tmp399 = new ExtractElementInst(packed_tmp0, const_int32_24, "tmp399", label_cond__28);
+  CastInst* double_conv4010 = new FPExtInst(float_tmp399, Type::DoubleTy, "conv4010", label_cond__28);
+  FCmpInst* int1_cmp4111 = new FCmpInst(FCmpInst::FCMP_OLT, double_conv4010, const_double_27, "cmp4111", label_cond__28);
+  new BranchInst(label_cond__42, label_cond_cont48, int1_cmp4111, label_cond__28);
+  
+  // Block cond.cont34 (label_cond_cont34)
+  PHINode* packed_tmp23_reg2mem_1 = new PHINode(VectorTy_4, "tmp23.reg2mem.1", label_cond_cont34);
+  packed_tmp23_reg2mem_1->reserveOperandSpace(2);
+  packed_tmp23_reg2mem_1->addIncoming(packed_tmp233, label_cond__14);
+  packed_tmp23_reg2mem_1->addIncoming(packed_tmp23, label_cond_cont20);
+  
+  ShuffleVectorInst* packed_tmp37 = new ShuffleVectorInst(packed_tmp23_reg2mem_1, packed_tmp2, const_packed_30, "tmp37", label_cond_cont34);
+  ExtractElementInst* float_tmp39 = new ExtractElementInst(packed_tmp0, const_int32_24, "tmp39", label_cond_cont34);
+  CastInst* double_conv40 = new FPExtInst(float_tmp39, Type::DoubleTy, "conv40", label_cond_cont34);
+  FCmpInst* int1_cmp41 = new FCmpInst(FCmpInst::FCMP_OLT, double_conv40, const_double_27, "cmp41", label_cond_cont34);
+  new BranchInst(label_cond__42, label_cond_cont48, int1_cmp41, label_cond_cont34);
+  
+  // Block cond.?42 (label_cond__42)
+  PHINode* packed_tmp37_reg2mem_0 = new PHINode(VectorTy_4, "tmp37.reg2mem.0", label_cond__42);
+  packed_tmp37_reg2mem_0->reserveOperandSpace(2);
+  packed_tmp37_reg2mem_0->addIncoming(packed_tmp378, label_cond__28);
+  packed_tmp37_reg2mem_0->addIncoming(packed_tmp37, label_cond_cont34);
+  
+  ShuffleVectorInst* packed_tmp5113 = new ShuffleVectorInst(packed_tmp37_reg2mem_0, packed_tmp1, const_packed_32, "tmp5113", label_cond__42);
+  new ReturnInst(packed_tmp5113, label_cond__42);
+  
+  // Block cond.cont48 (label_cond_cont48)
+  PHINode* packed_tmp37_reg2mem_1 = new PHINode(VectorTy_4, "tmp37.reg2mem.1", label_cond_cont48);
+  packed_tmp37_reg2mem_1->reserveOperandSpace(2);
+  packed_tmp37_reg2mem_1->addIncoming(packed_tmp378, label_cond__28);
+  packed_tmp37_reg2mem_1->addIncoming(packed_tmp37, label_cond_cont34);
+  
+  ShuffleVectorInst* packed_tmp51 = new ShuffleVectorInst(packed_tmp37_reg2mem_1, packed_tmp2, const_packed_32, "tmp51", label_cond_cont48);
+  new ReturnInst(packed_tmp51, label_cond_cont48);
+  
+}
+
+// Function: vcos (func_vcos)
+{
+  Function::arg_iterator args = func_vcos->arg_begin();
+  Value* packed_val = args++;
+  packed_val->setName("val");
+  
+  BasicBlock* label_entry_53 = new BasicBlock("entry",func_vcos,0);
+  
+  // Block entry (label_entry_53)
+  ExtractElementInst* float_tmp1 = new ExtractElementInst(packed_val, const_int32_19, "tmp1", label_entry_53);
+  CastInst* double_conv_54 = new FPExtInst(float_tmp1, Type::DoubleTy, "conv", label_entry_53);
+  ExtractElementInst* float_tmp3_55 = new ExtractElementInst(packed_val, const_int32_23, "tmp3", label_entry_53);
+  CastInst* double_conv4 = new FPExtInst(float_tmp3_55, Type::DoubleTy, "conv4", label_entry_53);
+  ExtractElementInst* float_tmp6_56 = new ExtractElementInst(packed_val, const_int32_25, "tmp6", label_entry_53);
+  CastInst* double_conv7 = new FPExtInst(float_tmp6_56, Type::DoubleTy, "conv7", label_entry_53);
+  ExtractElementInst* float_tmp9 = new ExtractElementInst(packed_val, const_int32_24, "tmp9", label_entry_53);
+  CastInst* double_conv10 = new FPExtInst(float_tmp9, Type::DoubleTy, "conv10", label_entry_53);
+  std::vector<Value*> int32_call_params;
+  int32_call_params.push_back(const_ptr_34);
+  int32_call_params.push_back(double_conv_54);
+  int32_call_params.push_back(double_conv4);
+  int32_call_params.push_back(double_conv7);
+  int32_call_params.push_back(double_conv10);
+  CallInst* int32_call = new CallInst(func_printf, int32_call_params.begin(), int32_call_params.end(), "call", label_entry_53);
+  int32_call->setCallingConv(CallingConv::C);
+  int32_call->setTailCall(true);const ParamAttrsList *int32_call_PAL = 0;
+  int32_call->setParamAttrs(int32_call_PAL);
+  
+  CallInst* float_call13 = new CallInst(func_cosf, float_tmp1, "call13", label_entry_53);
+  float_call13->setCallingConv(CallingConv::C);
+  float_call13->setTailCall(true);const ParamAttrsList *float_call13_PAL = 0;
+  float_call13->setParamAttrs(float_call13_PAL);
+  
+  InsertElementInst* packed_tmp15 = new InsertElementInst(const_packed_35, float_call13, const_int32_19, "tmp15", label_entry_53);
+  CallInst* float_call18 = new CallInst(func_cosf, float_tmp1, "call18", label_entry_53);
+  float_call18->setCallingConv(CallingConv::C);
+  float_call18->setTailCall(true);const ParamAttrsList *float_call18_PAL = 0;
+  float_call18->setParamAttrs(float_call18_PAL);
+  
+  InsertElementInst* packed_tmp20 = new InsertElementInst(packed_tmp15, float_call18, const_int32_23, "tmp20", label_entry_53);
+  CallInst* float_call23 = new CallInst(func_cosf, float_tmp1, "call23", label_entry_53);
+  float_call23->setCallingConv(CallingConv::C);
+  float_call23->setTailCall(true);const ParamAttrsList *float_call23_PAL = 0;
+  float_call23->setParamAttrs(float_call23_PAL);
+  
+  InsertElementInst* packed_tmp25 = new InsertElementInst(packed_tmp20, float_call23, const_int32_25, "tmp25", label_entry_53);
+  CallInst* float_call28 = new CallInst(func_cosf, float_tmp1, "call28", label_entry_53);
+  float_call28->setCallingConv(CallingConv::C);
+  float_call28->setTailCall(true);const ParamAttrsList *float_call28_PAL = 0;
+  float_call28->setParamAttrs(float_call28_PAL);
+  
+  InsertElementInst* packed_tmp30 = new InsertElementInst(packed_tmp25, float_call28, const_int32_24, "tmp30", label_entry_53);
+  CastInst* double_conv33 = new FPExtInst(float_call13, Type::DoubleTy, "conv33", label_entry_53);
+  CastInst* double_conv36 = new FPExtInst(float_call18, Type::DoubleTy, "conv36", label_entry_53);
+  CastInst* double_conv39 = new FPExtInst(float_call23, Type::DoubleTy, "conv39", label_entry_53);
+  CastInst* double_conv42 = new FPExtInst(float_call28, Type::DoubleTy, "conv42", label_entry_53);
+  std::vector<Value*> int32_call43_params;
+  int32_call43_params.push_back(const_ptr_36);
+  int32_call43_params.push_back(double_conv33);
+  int32_call43_params.push_back(double_conv36);
+  int32_call43_params.push_back(double_conv39);
+  int32_call43_params.push_back(double_conv42);
+  CallInst* int32_call43 = new CallInst(func_printf, int32_call43_params.begin(), int32_call43_params.end(), "call43", label_entry_53);
+  int32_call43->setCallingConv(CallingConv::C);
+  int32_call43->setTailCall(true);const ParamAttrsList *int32_call43_PAL = 0;
+  int32_call43->setParamAttrs(int32_call43_PAL);
+  
+  new ReturnInst(packed_tmp30, label_entry_53);
+  
+}
+
+// Function: scs (func_scs)
+{
+  Function::arg_iterator args = func_scs->arg_begin();
+  Value* packed_val_58 = args++;
+  packed_val_58->setName("val");
+  
+  BasicBlock* label_entry_59 = new BasicBlock("entry",func_scs,0);
+  
+  // Block entry (label_entry_59)
+  ExtractElementInst* float_tmp2 = new ExtractElementInst(packed_val_58, const_int32_19, "tmp2", label_entry_59);
+  CallInst* float_call_60 = new CallInst(func_cosf, float_tmp2, "call", label_entry_59);
+  float_call_60->setCallingConv(CallingConv::C);
+  float_call_60->setTailCall(true);const ParamAttrsList *float_call_60_PAL = 0;
+  float_call_60->setParamAttrs(float_call_60_PAL);
+  
+  InsertElementInst* packed_tmp5 = new InsertElementInst(const_packed_35, float_call_60, const_int32_19, "tmp5", label_entry_59);
+  CallInst* float_call7 = new CallInst(func_sinf, float_tmp2, "call7", label_entry_59);
+  float_call7->setCallingConv(CallingConv::C);
+  float_call7->setTailCall(true);const ParamAttrsList *float_call7_PAL = 0;
+  float_call7->setParamAttrs(float_call7_PAL);
+  
+  InsertElementInst* packed_tmp9 = new InsertElementInst(packed_tmp5, float_call7, const_int32_23, "tmp9", label_entry_59);
+  new ReturnInst(packed_tmp9, label_entry_59);
+  
+}
+
+// Function: vsin (func_vsin)
+{
+  Function::arg_iterator args = func_vsin->arg_begin();
+  Value* packed_val_62 = args++;
+  packed_val_62->setName("val");
+  
+  BasicBlock* label_entry_63 = new BasicBlock("entry",func_vsin,0);
+  
+  // Block entry (label_entry_63)
+  ExtractElementInst* float_tmp2_64 = new ExtractElementInst(packed_val_62, const_int32_19, "tmp2", label_entry_63);
+  CallInst* float_call_65 = new CallInst(func_sinf, float_tmp2_64, "call", label_entry_63);
+  float_call_65->setCallingConv(CallingConv::C);
+  float_call_65->setTailCall(true);const ParamAttrsList *float_call_65_PAL = 0;
+  float_call_65->setParamAttrs(float_call_65_PAL);
+  
+  InsertElementInst* packed_tmp6 = new InsertElementInst(const_packed_35, float_call_65, const_int32_19, "tmp6", label_entry_63);
+  InsertElementInst* packed_tmp9_66 = new InsertElementInst(packed_tmp6, float_call_65, const_int32_23, "tmp9", label_entry_63);
+  InsertElementInst* packed_tmp12 = new InsertElementInst(packed_tmp9_66, float_call_65, const_int32_25, "tmp12", label_entry_63);
+  InsertElementInst* packed_tmp15_67 = new InsertElementInst(packed_tmp12, float_call_65, const_int32_24, "tmp15", label_entry_63);
+  new ReturnInst(packed_tmp15_67, label_entry_63);
+  
+}
+
+// Function: kilp (func_kilp)
+{
+  Function::arg_iterator args = func_kilp->arg_begin();
+  Value* packed_val_69 = args++;
+  packed_val_69->setName("val");
+  
+  BasicBlock* label_entry_70 = new BasicBlock("entry",func_kilp,0);
+  BasicBlock* label_lor_rhs = new BasicBlock("lor_rhs",func_kilp,0);
+  BasicBlock* label_lor_rhs5 = new BasicBlock("lor_rhs5",func_kilp,0);
+  BasicBlock* label_lor_rhs11 = new BasicBlock("lor_rhs11",func_kilp,0);
+  BasicBlock* label_UnifiedReturnBlock_71 = new BasicBlock("UnifiedReturnBlock",func_kilp,0);
+  
+  // Block entry (label_entry_70)
+  ExtractElementInst* float_tmp1_72 = new ExtractElementInst(packed_val_69, const_int32_19, "tmp1", label_entry_70);
+  FCmpInst* int1_cmp_73 = new FCmpInst(FCmpInst::FCMP_OLT, float_tmp1_72, const_float_18, "cmp", label_entry_70);
+  new BranchInst(label_UnifiedReturnBlock_71, label_lor_rhs, int1_cmp_73, label_entry_70);
+  
+  // Block lor_rhs (label_lor_rhs)
+  ExtractElementInst* float_tmp3_75 = new ExtractElementInst(packed_val_69, const_int32_23, "tmp3", label_lor_rhs);
+  FCmpInst* int1_cmp4 = new FCmpInst(FCmpInst::FCMP_OLT, float_tmp3_75, const_float_18, "cmp4", label_lor_rhs);
+  new BranchInst(label_UnifiedReturnBlock_71, label_lor_rhs5, int1_cmp4, label_lor_rhs);
+  
+  // Block lor_rhs5 (label_lor_rhs5)
+  ExtractElementInst* float_tmp7 = new ExtractElementInst(packed_val_69, const_int32_25, "tmp7", label_lor_rhs5);
+  FCmpInst* int1_cmp8 = new FCmpInst(FCmpInst::FCMP_OLT, float_tmp7, const_float_18, "cmp8", label_lor_rhs5);
+  new BranchInst(label_UnifiedReturnBlock_71, label_lor_rhs11, int1_cmp8, label_lor_rhs5);
+  
+  // Block lor_rhs11 (label_lor_rhs11)
+  ExtractElementInst* float_tmp13 = new ExtractElementInst(packed_val_69, const_int32_24, "tmp13", label_lor_rhs11);
+  FCmpInst* int1_cmp14 = new FCmpInst(FCmpInst::FCMP_OLT, float_tmp13, const_float_18, "cmp14", label_lor_rhs11);
+  CastInst* int32_retval = new ZExtInst(int1_cmp14, IntegerType::get(32), "retval", label_lor_rhs11);
+  new ReturnInst(int32_retval, label_lor_rhs11);
+  
+  // Block UnifiedReturnBlock (label_UnifiedReturnBlock_71)
+  new ReturnInst(const_int32_23, label_UnifiedReturnBlock_71);
+  
+}
+
+return mod;
+
+}
diff --git a/src/gallium/aux/llvm/gallivm_cpu.cpp b/src/gallium/aux/llvm/gallivm_cpu.cpp
new file mode 100644
index 0000000000..dc4d92a72a
--- /dev/null
+++ b/src/gallium/aux/llvm/gallivm_cpu.cpp
@@ -0,0 +1,202 @@
+/**************************************************************************
+ *
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ **************************************************************************/
+
+ /*
+  * Authors:
+  *   Zack Rusin zack@tungstengraphics.com
+  */
+#ifdef MESA_LLVM
+
+#include "gallivm.h"
+#include "gallivm_p.h"
+
+#include "instructions.h"
+#include "loweringpass.h"
+#include "storage.h"
+#include "tgsitollvm.h"
+
+#include "pipe/p_context.h"
+#include "pipe/p_shader_tokens.h"
+
+#include "pipe/tgsi/exec/tgsi_exec.h"
+#include "pipe/tgsi/util/tgsi_dump.h"
+
+#include <llvm/Module.h>
+#include <llvm/CallingConv.h>
+#include <llvm/Constants.h>
+#include <llvm/DerivedTypes.h>
+#include <llvm/Instructions.h>
+#include <llvm/ModuleProvider.h>
+#include <llvm/Pass.h>
+#include <llvm/PassManager.h>
+#include <llvm/ParameterAttributes.h>
+#include <llvm/Support/PatternMatch.h>
+#include <llvm/ExecutionEngine/JIT.h>
+#include <llvm/ExecutionEngine/Interpreter.h>
+#include <llvm/ExecutionEngine/GenericValue.h>
+#include <llvm/Support/MemoryBuffer.h>
+#include <llvm/LinkAllPasses.h>
+#include <llvm/Analysis/Verifier.h>
+#include <llvm/Analysis/LoopPass.h>
+#include <llvm/Target/TargetData.h>
+#include <llvm/Bitcode/ReaderWriter.h>
+#include <llvm/Transforms/Utils/Cloning.h>
+
+#include <sstream>
+#include <fstream>
+#include <iostream>
+
+struct gallivm_cpu_engine {
+   llvm::ExecutionEngine *engine;
+};
+
+static struct gallivm_cpu_engine *CPU = 0;
+
+typedef int (*fragment_shader_runner)(float x, float y,
+                                      float (*dests)[16][4],
+                                      float (*inputs)[16][4],
+                                      int num_attribs,
+                                      float (*consts)[4], int num_consts,
+                                      struct tgsi_sampler *samplers);
+
+int gallivm_cpu_fs_exec(struct gallivm_prog *prog,
+                        float fx, float fy,
+                        float (*dests)[16][4],
+                        float (*inputs)[16][4],
+                        float (*consts)[4],
+                        struct tgsi_sampler *samplers)
+{
+   fragment_shader_runner runner = reinterpret_cast<fragment_shader_runner>(prog->function);
+   assert(runner);
+
+   return runner(fx, fy, dests, inputs, prog->num_interp,
+                 consts, prog->num_consts,
+                 samplers);
+}
+
+static inline llvm::Function *func_for_shader(struct gallivm_prog *prog)
+{
+   llvm::Module *mod = prog->module;
+   llvm::Function *func = 0;
+
+   switch (prog->type) {
+   case GALLIVM_VS:
+      func = mod->getFunction("vs_shader");
+      break;
+   case GALLIVM_FS:
+      func = mod->getFunction("fs_shader");
+      break;
+   default:
+      assert(!"Unknown shader type!");
+      break;
+   }
+   return func;
+}
+
+/*!
+  This function creates a CPU based execution engine for the given gallivm_prog.
+  gallivm_cpu_engine should be used as a singleton throughout the library. Before
+  executing gallivm_prog_exec one needs to call gallivm_cpu_jit_compile.
+  The gallivm_prog instance which is being passed to the constructor is being
+  automatically JIT compiled so one shouldn't call gallivm_cpu_jit_compile
+  with it again.
+ */
+struct gallivm_cpu_engine * gallivm_cpu_engine_create(struct gallivm_prog *prog)
+{
+   struct gallivm_cpu_engine *cpu = (struct gallivm_cpu_engine *)
+                                    calloc(1, sizeof(struct gallivm_cpu_engine));
+   llvm::Module *mod = static_cast<llvm::Module*>(prog->module);
+   llvm::ExistingModuleProvider *mp = new llvm::ExistingModuleProvider(mod);
+   llvm::ExecutionEngine *ee = llvm::ExecutionEngine::create(mp, false);
+   ee->DisableLazyCompilation();
+   cpu->engine = ee;
+
+   llvm::Function *func = func_for_shader(prog);
+
+   prog->function = ee->getPointerToFunction(func);
+   CPU = cpu;
+   return cpu;
+}
+
+
+/*!
+  This function JIT compiles the given gallivm_prog with the given cpu based execution engine.
+  The reference to the generated machine code entry point will be stored
+  in the gallivm_prog program. After executing this function one can call gallivm_prog_exec
+  in order to execute the gallivm_prog on the CPU.
+ */
+void gallivm_cpu_jit_compile(struct gallivm_cpu_engine *cpu, struct gallivm_prog *prog)
+{
+   llvm::Module *mod = static_cast<llvm::Module*>(prog->module);
+   llvm::ExistingModuleProvider *mp = new llvm::ExistingModuleProvider(mod);
+   llvm::ExecutionEngine *ee = cpu->engine;
+   assert(ee);
+   /*FIXME : remove */
+   ee->DisableLazyCompilation();
+   ee->addModuleProvider(mp);
+
+   llvm::Function *func = func_for_shader(prog);
+   prog->function = ee->getPointerToFunction(func);
+}
+
+void gallivm_cpu_engine_delete(struct gallivm_cpu_engine *cpu)
+{
+   free(cpu);
+}
+
+struct gallivm_cpu_engine * gallivm_global_cpu_engine()
+{
+   return CPU;
+}
+
+
+typedef void (*vertex_shader_runner)(void *ainputs,
+                                     void *dests,
+                                     float (*aconsts)[4],
+                                     void *temps);
+
+
+/*!
+  This function is used to execute the gallivm_prog in software. Before calling
+  this function the gallivm_prog has to be JIT compiled with the gallivm_cpu_jit_compile
+  function.
+ */
+int gallivm_cpu_vs_exec(struct gallivm_prog *prog,
+                        struct tgsi_exec_vector       *inputs,
+                        struct tgsi_exec_vector       *dests,
+                        float (*consts)[4],
+                        struct tgsi_exec_vector       *temps)
+{
+   vertex_shader_runner runner = reinterpret_cast<vertex_shader_runner>(prog->function);
+   assert(runner);
+   /*FIXME*/
+   runner(inputs, dests, consts, temps);
+
+   return 0;
+}
+
+#endif
diff --git a/src/gallium/aux/llvm/gallivm_p.h b/src/gallium/aux/llvm/gallivm_p.h
new file mode 100644
index 0000000000..cfe7b1901b
--- /dev/null
+++ b/src/gallium/aux/llvm/gallivm_p.h
@@ -0,0 +1,110 @@
+#ifndef GALLIVM_P_H
+#define GALLIVM_P_H
+
+#ifdef MESA_LLVM
+
+#include "gallivm.h"
+#include "pipe/p_shader_tokens.h"
+#include "pipe/p_compiler.h"
+
+namespace llvm {
+   class Module;
+}
+
+#if defined __cplusplus
+extern "C" {
+#endif
+
+enum gallivm_shader_type;
+enum gallivm_vector_layout;
+
+struct gallivm_interpolate {
+   int attrib;
+   int chan;
+   int type;
+};
+
+struct gallivm_ir {
+   llvm::Module *module;
+   int id;
+   enum gallivm_shader_type type;
+   enum gallivm_vector_layout layout;
+   int num_components;
+   int   num_consts;
+
+   //FIXME: this might not be enough for some shaders
+   struct gallivm_interpolate interpolators[32*4];
+   int   num_interp;
+};
+
+struct gallivm_prog {
+   llvm::Module *module;
+   void *function;
+
+   int   id;
+   enum gallivm_shader_type type;
+
+   int   num_consts;
+
+   //FIXME: this might not be enough for some shaders
+   struct gallivm_interpolate interpolators[32*4];
+   int   num_interp;
+};
+
+static INLINE void gallivm_swizzle_components(int swizzle,
+                                              int *xc, int *yc,
+                                              int *zc, int *wc)
+{
+   int x = swizzle / 1000; swizzle -= x * 1000;
+   int y = swizzle / 100;  swizzle -= y * 100;
+   int z = swizzle / 10;   swizzle -= z * 10;
+   int w = swizzle;
+
+   if (xc) *xc = x;
+   if (yc) *yc = y;
+   if (zc) *zc = z;
+   if (wc) *wc = w;
+}
+
+static INLINE boolean gallivm_is_swizzle(int swizzle)
+{
+   const int NO_SWIZZLE = TGSI_SWIZZLE_X * 1000 + TGSI_SWIZZLE_Y * 100 +
+                          TGSI_SWIZZLE_Z * 10 + TGSI_SWIZZLE_W;
+   return swizzle != NO_SWIZZLE;
+}
+
+static INLINE int gallivm_x_swizzle(int swizzle)
+{
+   int x;
+   gallivm_swizzle_components(swizzle, &x, 0, 0, 0);
+   return x;
+}
+
+static INLINE int gallivm_y_swizzle(int swizzle)
+{
+   int y;
+   gallivm_swizzle_components(swizzle, 0, &y, 0, 0);
+   return y;
+}
+
+static INLINE int gallivm_z_swizzle(int swizzle)
+{
+   int z;
+   gallivm_swizzle_components(swizzle, 0, 0, &z, 0);
+   return z;
+}
+
+static INLINE int gallivm_w_swizzle(int swizzle)
+{
+   int w;
+   gallivm_swizzle_components(swizzle, 0, 0, 0, &w);
+   return w;
+}
+
+#endif /* MESA_LLVM */
+
+#if defined __cplusplus
+} // extern "C"
+#endif
+
+#endif
diff --git a/src/gallium/aux/llvm/instructions.cpp b/src/gallium/aux/llvm/instructions.cpp
new file mode 100644
index 0000000000..55d39fa5f1
--- /dev/null
+++ b/src/gallium/aux/llvm/instructions.cpp
@@ -0,0 +1,889 @@
+/**************************************************************************
+ *
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ **************************************************************************/
+
+ /*
+  * Authors:
+  *   Zack Rusin zack@tungstengraphics.com
+  */
+#ifdef MESA_LLVM
+
+#include "instructions.h"
+
+#include "storage.h"
+
+#include <llvm/CallingConv.h>
+#include <llvm/Constants.h>
+#include <llvm/DerivedTypes.h>
+#include <llvm/Function.h>
+#include <llvm/InstrTypes.h>
+#include <llvm/Instructions.h>
+#include <llvm/ParameterAttributes.h>
+
+#include <sstream>
+#include <fstream>
+#include <iostream>
+
+using namespace llvm;
+
+#include "gallivm_builtins.cpp"
+
+static inline std::string createFuncName(int label)
+{
+   std::ostringstream stream;
+   stream << "function";
+   stream << label;
+   return stream.str();
+}
+
+Instructions::Instructions(llvm::Module *mod, llvm::Function *func, llvm::BasicBlock *block,
+                           Storage *storage)
+   :  m_mod(mod), m_func(func), m_builder(block), m_idx(0),
+      m_storage(storage)
+{
+   m_floatVecType = VectorType::get(Type::FloatTy, 4);
+
+   m_llvmFSqrt = 0;
+   m_llvmFAbs  = 0;
+   m_llvmPow   = 0;
+   m_llvmFloor = 0;
+   m_llvmFlog  = 0;
+   m_llvmLit  = 0;
+   m_fmtPtr = 0;
+
+   createGallivmBuiltins(m_mod);
+}
+
+llvm::Value * Instructions::add(llvm::Value *in1, llvm::Value *in2)
+{
+   return m_builder.CreateAdd(in1, in2, name("add"));
+}
+
+llvm::Value * Instructions::madd(llvm::Value *in1, llvm::Value *in2,
+                                 llvm::Value *in3)
+{
+   Value *mulRes = mul(in1, in2);
+   return add(mulRes, in3);
+}
+ 
+llvm::Value * Instructions::mul(llvm::Value *in1, llvm::Value *in2)
+{
+   return m_builder.CreateMul(in1, in2, name("mul"));
+}
+
+const char * Instructions::name(const char *prefix)
+{
+   ++m_idx;
+   snprintf(m_name, 32, "%s%d", prefix, m_idx);
+   return m_name;
+}
+
+llvm::Value * Instructions::dp3(llvm::Value *in1, llvm::Value *in2)
+{
+   Value *mulRes = mul(in1, in2);
+   Value *x = m_builder.CreateExtractElement(mulRes,
+                                                          m_storage->constantInt(0),
+                                                          name("extractx"));
+   Value *y = m_builder.CreateExtractElement(mulRes,
+                                                          m_storage->constantInt(1),
+                                                          name("extracty"));
+   Value *z = m_builder.CreateExtractElement(mulRes,
+                                                          m_storage->constantInt(2),
+                                                          name("extractz"));
+   Value *xy = m_builder.CreateAdd(x, y,name("xy"));
+   Value *dot3 = m_builder.CreateAdd(xy, z, name("dot3"));
+   return vectorFromVals(dot3, dot3, dot3, dot3);
+}
+
+llvm::Value *Instructions::callFSqrt(llvm::Value *val)
+{
+   if (!m_llvmFSqrt) {
+      // predeclare the intrinsic
+      std::vector<const Type*> fsqrtArgs;
+      fsqrtArgs.push_back(Type::FloatTy);
+      ParamAttrsList *fsqrtPal = 0;
+      FunctionType* fsqrtType = FunctionType::get(
+         /*Result=*/Type::FloatTy,
+         /*Params=*/fsqrtArgs,
+         /*isVarArg=*/false);
+      m_llvmFSqrt = new Function(
+         /*Type=*/fsqrtType,
+         /*Linkage=*/GlobalValue::ExternalLinkage,
+         /*Name=*/"llvm.sqrt.f32", m_mod);
+      m_llvmFSqrt->setCallingConv(CallingConv::C);
+      m_llvmFSqrt->setParamAttrs(fsqrtPal);
+   }
+   CallInst *call = m_builder.CreateCall(m_llvmFSqrt, val,
+                                         name("sqrt"));
+   call->setCallingConv(CallingConv::C);
+   call->setTailCall(false);
+   return call;
+}
+
+llvm::Value * Instructions::rsq(llvm::Value *in1)
+{
+   Value *x = m_builder.CreateExtractElement(in1,
+                                             m_storage->constantInt(0),
+                                             name("extractx"));
+   Value *abs  = callFAbs(x);
+   Value *sqrt = callFSqrt(abs);
+
+   Value *rsqrt = m_builder.CreateFDiv(ConstantFP::get(Type::FloatTy,
+                                                                APFloat(1.f)),
+                                                sqrt,
+                                                name("rsqrt"));
+   return vectorFromVals(rsqrt, rsqrt, rsqrt, rsqrt);
+}
+
+llvm::Value * Instructions::vectorFromVals(llvm::Value *x, llvm::Value *y,
+                                           llvm::Value *z, llvm::Value *w)
+{
+   Constant *const_vec = Constant::getNullValue(m_floatVecType);
+   Value *res = m_builder.CreateInsertElement(const_vec, x,
+                                              m_storage->constantInt(0),
+                                              name("vecx"));
+   res = m_builder.CreateInsertElement(res, y, m_storage->constantInt(1),
+                               name("vecxy"));
+   res = m_builder.CreateInsertElement(res, z, m_storage->constantInt(2),
+                               name("vecxyz"));
+   if (w)
+      res = m_builder.CreateInsertElement(res, w, m_storage->constantInt(3),
+                                          name("vecxyzw"));
+   return res;
+}
+
+llvm::Value *Instructions::callFAbs(llvm::Value *val)
+{
+   if (!m_llvmFAbs) {
+      // predeclare the intrinsic
+      std::vector<const Type*> fabsArgs;
+      fabsArgs.push_back(Type::FloatTy);
+      ParamAttrsList *fabsPal = 0;
+      FunctionType* fabsType = FunctionType::get(
+         /*Result=*/Type::FloatTy,
+         /*Params=*/fabsArgs,
+         /*isVarArg=*/false);
+      m_llvmFAbs = new Function(
+         /*Type=*/fabsType,
+         /*Linkage=*/GlobalValue::ExternalLinkage,
+         /*Name=*/"fabs", m_mod);
+      m_llvmFAbs->setCallingConv(CallingConv::C);
+      m_llvmFAbs->setParamAttrs(fabsPal);
+   }
+   CallInst *call = m_builder.CreateCall(m_llvmFAbs, val,
+                                         name("fabs"));
+   call->setCallingConv(CallingConv::C);
+   call->setTailCall(false);
+   return call;
+}
+
+llvm::Value * Instructions::lit(llvm::Value *in)
+{
+   if (!m_llvmLit) {
+      m_llvmLit = m_mod->getFunction("lit");
+   }
+   CallInst *call = m_builder.CreateCall(m_llvmLit, in, name("litres"));
+   call->setCallingConv(CallingConv::C);
+   call->setTailCall(false);
+   return call;
+}
+
+llvm::Value * Instructions::sub(llvm::Value *in1, llvm::Value *in2)
+{
+   Value *res = m_builder.CreateSub(in1, in2, name("sub"));
+   return res;
+}
+
+llvm::Value * Instructions::callPow(llvm::Value *val1, llvm::Value *val2)
+{
+   if (!m_llvmPow) {
+      // predeclare the intrinsic
+      std::vector<const Type*> powArgs;
+      powArgs.push_back(Type::FloatTy);
+      powArgs.push_back(Type::FloatTy);
+      ParamAttrsList *powPal = 0;
+      FunctionType* powType = FunctionType::get(
+         /*Result=*/Type::FloatTy,
+         /*Params=*/powArgs,
+         /*isVarArg=*/false);
+      m_llvmPow = new Function(
+         /*Type=*/powType,
+         /*Linkage=*/GlobalValue::ExternalLinkage,
+         /*Name=*/"llvm.pow.f32", m_mod);
+      m_llvmPow->setCallingConv(CallingConv::C);
+      m_llvmPow->setParamAttrs(powPal);
+   }
+   std::vector<Value*> params;
+   params.push_back(val1);
+   params.push_back(val2);
+   CallInst *call = m_builder.CreateCall(m_llvmPow, params.begin(), params.end(),
+                                         name("pow"));
+   call->setCallingConv(CallingConv::C);
+   call->setTailCall(false);
+   return call;
+}
+
+llvm::Value * Instructions::pow(llvm::Value *in1, llvm::Value *in2)
+{
+   Value *x1 = m_builder.CreateExtractElement(in1,
+                                              m_storage->constantInt(0),
+                                              name("x1"));
+   Value *x2 = m_builder.CreateExtractElement(in2,
+                                              m_storage->constantInt(0),
+                                              name("x2"));
+   llvm::Value *val = callPow(x1, x2);
+   return vectorFromVals(val, val, val, val);
+}
+
+llvm::Value * Instructions::rcp(llvm::Value *in1)
+{
+   Value *x1 = m_builder.CreateExtractElement(in1,
+                                              m_storage->constantInt(0),
+                                              name("x1"));
+   Value *res = m_builder.CreateFDiv(ConstantFP::get(Type::FloatTy,
+                                                              APFloat(1.f)),
+                                              x1, name("rcp"));
+   return vectorFromVals(res, res, res, res);
+}
+
+llvm::Value * Instructions::dp4(llvm::Value *in1, llvm::Value *in2)
+{
+   Value *mulRes = mul(in1, in2);
+   std::vector<llvm::Value*> vec = extractVector(mulRes);
+   Value *xy = m_builder.CreateAdd(vec[0], vec[1], name("xy"));
+   Value *xyz = m_builder.CreateAdd(xy, vec[2], name("xyz"));
+   Value *dot4 = m_builder.CreateAdd(xyz, vec[3], name("dot4"));
+   return vectorFromVals(dot4, dot4, dot4, dot4);
+}
+
+llvm::Value * Instructions::dph(llvm::Value *in1, llvm::Value *in2)
+{
+   Value *mulRes = mul(in1, in2);
+   std::vector<llvm::Value*> vec1 = extractVector(mulRes);
+   Value *xy = m_builder.CreateAdd(vec1[0], vec1[1], name("xy"));
+   Value *xyz = m_builder.CreateAdd(xy, vec1[2], name("xyz"));
+   Value *dph = m_builder.CreateAdd(xyz, vec1[3], name("dph"));
+   return vectorFromVals(dph, dph, dph, dph);
+}
+
+llvm::Value * Instructions::dst(llvm::Value *in1, llvm::Value *in2)
+{
+   Value *y1 = m_builder.CreateExtractElement(in1,
+                                              m_storage->constantInt(1),
+                                              name("y1"));
+   Value *z = m_builder.CreateExtractElement(in1,
+                                             m_storage->constantInt(2),
+                                             name("z"));
+   Value *y2 = m_builder.CreateExtractElement(in2,
+                                              m_storage->constantInt(1),
+                                              name("y2"));
+   Value *w = m_builder.CreateExtractElement(in2,
+                                             m_storage->constantInt(3),
+                                             name("w"));
+   Value *ry = m_builder.CreateMul(y1, y2, name("tyuy"));
+   return vectorFromVals(ConstantFP::get(Type::FloatTy, APFloat(1.f)),
+                         ry, z, w);
+}
+
+llvm::Value * Instructions::ex2(llvm::Value *in)
+{
+   llvm::Value *val = callPow(ConstantFP::get(Type::FloatTy, APFloat(2.f)),
+                              m_builder.CreateExtractElement(
+                                 in, m_storage->constantInt(0),
+                                 name("x1")));
+   return vectorFromVals(val, val, val, val);
+}
+
+llvm::Value * Instructions::callFloor(llvm::Value *val)
+{
+   if (!m_llvmFloor) {
+      // predeclare the intrinsic
+      std::vector<const Type*> floorArgs;
+      floorArgs.push_back(Type::FloatTy);
+      ParamAttrsList *floorPal = 0;
+      FunctionType* floorType = FunctionType::get(
+         /*Result=*/Type::FloatTy,
+         /*Params=*/floorArgs,
+         /*isVarArg=*/false);
+      m_llvmFloor = new Function(
+         /*Type=*/floorType,
+         /*Linkage=*/GlobalValue::ExternalLinkage,
+         /*Name=*/"floorf", m_mod);
+      m_llvmFloor->setCallingConv(CallingConv::C);
+      m_llvmFloor->setParamAttrs(floorPal);
+   }
+   CallInst *call =  m_builder.CreateCall(m_llvmFloor, val,
+                                          name("floorf"));
+   call->setCallingConv(CallingConv::C);
+   call->setTailCall(false);
+   return call;
+}
+
+llvm::Value * Instructions::floor(llvm::Value *in)
+{
+   std::vector<llvm::Value*> vec = extractVector(in);
+   return vectorFromVals(callFloor(vec[0]), callFloor(vec[1]),
+                         callFloor(vec[2]), callFloor(vec[3]));
+}
+
+llvm::Value * Instructions::arl(llvm::Value *in)
+{
+   return floor(in);
+}
+
+llvm::Value * Instructions::frc(llvm::Value *in)
+{
+   llvm::Value *flr = floor(in);
+   return sub(in, flr);
+}
+
+llvm::Value * Instructions::callFLog(llvm::Value *val)
+{
+   if (!m_llvmFlog) {
+      // predeclare the intrinsic
+      std::vector<const Type*> flogArgs;
+      flogArgs.push_back(Type::FloatTy);
+      ParamAttrsList *flogPal = 0;
+      FunctionType* flogType = FunctionType::get(
+         /*Result=*/Type::FloatTy,
+         /*Params=*/flogArgs,
+         /*isVarArg=*/false);
+      m_llvmFlog = new Function(
+         /*Type=*/flogType,
+         /*Linkage=*/GlobalValue::ExternalLinkage,
+         /*Name=*/"logf", m_mod);
+      m_llvmFlog->setCallingConv(CallingConv::C);
+      m_llvmFlog->setParamAttrs(flogPal);
+   }
+   CallInst *call = m_builder.CreateCall(m_llvmFlog, val,
+                                         name("logf"));
+   call->setCallingConv(CallingConv::C);
+   call->setTailCall(false);
+   return call;
+}
+
+llvm::Value * Instructions::lg2(llvm::Value *in)
+{
+   std::vector<llvm::Value*> vec = extractVector(in);
+   llvm::Value *const_vec = constVector(1.442695f, 1.442695f,
+                                        1.442695f, 1.442695f);
+   return mul(vectorFromVals(callFLog(vec[0]), callFLog(vec[1]),
+                             callFLog(vec[2]), callFLog(vec[3])), const_vec);
+}
+
+llvm::Value * Instructions::min(llvm::Value *in1, llvm::Value *in2)
+{
+   std::vector<llvm::Value*> vec1 = extractVector(in1);
+   std::vector<llvm::Value*> vec2 = extractVector(in2);
+
+   Value *xcmp  = m_builder.CreateFCmpOLT(vec1[0], vec2[0], name("xcmp"));
+   Value *selx = m_builder.CreateSelect(xcmp, vec1[0], vec2[0],
+                                        name("selx"));
+
+   Value *ycmp  = m_builder.CreateFCmpOLT(vec1[1], vec2[1], name("ycmp"));
+   Value *sely = m_builder.CreateSelect(ycmp, vec1[1], vec2[1],
+                                        name("sely"));
+
+   Value *zcmp  = m_builder.CreateFCmpOLT(vec1[2], vec2[2], name("zcmp"));
+   Value *selz = m_builder.CreateSelect(zcmp, vec1[2], vec2[2],
+                                        name("selz"));
+
+   Value *wcmp  = m_builder.CreateFCmpOLT(vec1[3], vec2[3], name("wcmp"));
+   Value *selw = m_builder.CreateSelect(wcmp, vec1[3], vec2[3],
+                                        name("selw"));
+
+   return vectorFromVals(selx, sely, selz, selw);
+}
+
+llvm::Value * Instructions::max(llvm::Value *in1, llvm::Value *in2)
+{
+   std::vector<llvm::Value*> vec1 = extractVector(in1);
+   std::vector<llvm::Value*> vec2 = extractVector(in2);
+
+   Value *xcmp  = m_builder.CreateFCmpOGT(vec1[0], vec2[0],
+                                          name("xcmp"));
+   Value *selx = m_builder.CreateSelect(xcmp, vec1[0], vec2[0],
+                                        name("selx"));
+
+   Value *ycmp  = m_builder.CreateFCmpOGT(vec1[1], vec2[1],
+                                          name("ycmp"));
+   Value *sely = m_builder.CreateSelect(ycmp, vec1[1], vec2[1],
+                                        name("sely"));
+
+   Value *zcmp  = m_builder.CreateFCmpOGT(vec1[2], vec2[2],
+                                          name("zcmp"));
+   Value *selz = m_builder.CreateSelect(zcmp, vec1[2], vec2[2],
+                                        name("selz"));
+
+   Value *wcmp  = m_builder.CreateFCmpOGT(vec1[3], vec2[3],
+                                          name("wcmp"));
+   Value *selw = m_builder.CreateSelect(wcmp, vec1[3], vec2[3],
+                                        name("selw"));
+
+   return vectorFromVals(selx, sely, selz, selw);
+}
+
+void Instructions::printVector(llvm::Value *val)
+{
+   static const char *frmt = "Vector is [%f, %f, %f, %f]\x0A";
+
+   if (!m_fmtPtr) {
+      Constant *format = ConstantArray::get(frmt, true);
+      ArrayType *arrayTy = ArrayType::get(IntegerType::get(8), strlen(frmt) + 1);
+      GlobalVariable* globalFormat = new GlobalVariable(
+         /*Type=*/arrayTy,
+         /*isConstant=*/true,
+         /*Linkage=*/GlobalValue::InternalLinkage,
+         /*Initializer=*/0, // has initializer, specified below
+         /*Name=*/name(".str"),
+         m_mod);
+      globalFormat->setInitializer(format);
+
+      Constant* const_int0 = Constant::getNullValue(IntegerType::get(32));
+      std::vector<Constant*> const_ptr_21_indices;
+      const_ptr_21_indices.push_back(const_int0);
+      const_ptr_21_indices.push_back(const_int0);
+      m_fmtPtr = ConstantExpr::getGetElementPtr(globalFormat,
+                                                &const_ptr_21_indices[0], const_ptr_21_indices.size());
+   }
+
+   Function *func_printf = m_mod->getFunction("printf");
+   if (!func_printf)
+      func_printf = declarePrintf();
+   assert(func_printf);
+   std::vector<llvm::Value*> vec = extractVector(val);
+   Value *dx = m_builder.CreateFPExt(vec[0], Type::DoubleTy, name("dx"));
+   Value *dy = m_builder.CreateFPExt(vec[1], Type::DoubleTy, name("dy"));
+   Value *dz = m_builder.CreateFPExt(vec[2], Type::DoubleTy, name("dz"));
+   Value *dw = m_builder.CreateFPExt(vec[3], Type::DoubleTy, name("dw"));
+   std::vector<Value*> params;
+   params.push_back(m_fmtPtr);
+   params.push_back(dx);
+   params.push_back(dy);
+   params.push_back(dz);
+   params.push_back(dw);
+   CallInst *call = m_builder.CreateCall(func_printf, params.begin(), params.end(),
+                                         name("printf"));
+   call->setCallingConv(CallingConv::C);
+   call->setTailCall(true);
+}
+
+llvm::Function * Instructions::declarePrintf()
+{
+   std::vector<const Type*> args;
+   ParamAttrsList *params = 0;
+   FunctionType* funcTy = FunctionType::get(
+      /*Result=*/IntegerType::get(32),
+      /*Params=*/args,
+      /*isVarArg=*/true);
+   Function* func_printf = new Function(
+      /*Type=*/funcTy,
+      /*Linkage=*/GlobalValue::ExternalLinkage,
+      /*Name=*/"printf", m_mod);
+   func_printf->setCallingConv(CallingConv::C);
+   func_printf->setParamAttrs(params);
+   return func_printf;
+}
+
+
+llvm::Value * Instructions::sgt(llvm::Value *in1, llvm::Value *in2)
+{
+   Constant *const1f = ConstantFP::get(Type::FloatTy, APFloat(1.000000e+00f));
+   Constant *const0f = Constant::getNullValue(Type::FloatTy);
+
+   std::vector<llvm::Value*> vec1 = extractVector(in1);
+   std::vector<llvm::Value*> vec2 = extractVector(in2);
+   Value *xcmp = m_builder.CreateFCmpOGT(vec1[0], vec2[0], name("xcmp"));
+   Value *x = m_builder.CreateSelect(xcmp, const1f, const0f, name("xsel"));
+
+   Value *ycmp = m_builder.CreateFCmpOGT(vec1[1], vec2[1], name("ycmp"));
+   Value *y = m_builder.CreateSelect(ycmp, const1f, const0f, name("ysel"));
+
+   Value *zcmp = m_builder.CreateFCmpOGT(vec1[2], vec2[2], name("zcmp"));
+   Value *z = m_builder.CreateSelect(zcmp, const1f, const0f, name("zsel"));
+
+   Value *wcmp = m_builder.CreateFCmpOGT(vec1[3], vec2[3], name("wcmp"));
+   Value *w = m_builder.CreateSelect(wcmp, const1f, const0f, name("wsel"));
+
+   return vectorFromVals(x, y, z, w);
+}
+llvm::Value * Instructions::sge(llvm::Value *in1, llvm::Value *in2)
+{
+   Constant *const1f = ConstantFP::get(Type::FloatTy, APFloat(1.000000e+00f));
+   Constant *const0f = Constant::getNullValue(Type::FloatTy);
+
+   std::vector<llvm::Value*> vec1 = extractVector(in1);
+   std::vector<llvm::Value*> vec2 = extractVector(in2);
+
+   Value *xcmp = m_builder.CreateFCmpOGE(vec1[0], vec2[0], name("xcmp"));
+   Value *x = m_builder.CreateSelect(xcmp, const1f, const0f, name("xsel"));
+
+   Value *ycmp = m_builder.CreateFCmpOGE(vec1[1], vec2[1], name("ycmp"));
+   Value *y = m_builder.CreateSelect(ycmp, const1f, const0f, name("ysel"));
+
+   Value *zcmp = m_builder.CreateFCmpOGE(vec1[2], vec2[2], name("zcmp"));
+   Value *z = m_builder.CreateSelect(zcmp, const1f, const0f, name("zsel"));
+
+   Value *wcmp = m_builder.CreateFCmpOGE(vec1[3], vec2[3], name("wcmp"));
+   Value *w = m_builder.CreateSelect(wcmp, const1f, const0f, name("wsel"));
+
+   return vectorFromVals(x, y, z, w);
+}
+
+
+llvm::Value * Instructions::slt(llvm::Value *in1, llvm::Value *in2)
+{
+   Constant *const1f = ConstantFP::get(Type::FloatTy, APFloat(1.000000e+00f));
+   Constant *const0f = Constant::getNullValue(Type::FloatTy);
+
+   std::vector<llvm::Value*> vec1 = extractVector(in1);
+   std::vector<llvm::Value*> vec2 = extractVector(in2);
+
+   Value *xcmp = m_builder.CreateFCmpOLT(vec1[0], vec2[0], name("xcmp"));
+   Value *x = m_builder.CreateSelect(xcmp, const1f, const0f, name("xsel"));
+
+   Value *ycmp = m_builder.CreateFCmpOLT(vec1[1], vec2[1], name("ycmp"));
+   Value *y = m_builder.CreateSelect(ycmp, const1f, const0f, name("ysel"));
+
+   Value *zcmp = m_builder.CreateFCmpOLT(vec1[2], vec2[2], name("zcmp"));
+   Value *z = m_builder.CreateSelect(zcmp, const1f, const0f, name("zsel"));
+
+   Value *wcmp = m_builder.CreateFCmpOLT(vec1[3], vec2[3], name("wcmp"));
+   Value *w = m_builder.CreateSelect(wcmp, const1f, const0f, name("wsel"));
+
+   return vectorFromVals(x, y, z, w);
+}
+
+llvm::Value * Instructions::cross(llvm::Value *in1, llvm::Value *in2)
+{
+   Value *x1 = m_builder.CreateExtractElement(in1,
+                                              m_storage->constantInt(0),
+                                              name("x1"));
+   Value *y1 = m_builder.CreateExtractElement(in1,
+                                              m_storage->constantInt(1),
+                                              name("y1"));
+   Value *z1 = m_builder.CreateExtractElement(in1,
+                                              m_storage->constantInt(2),
+                                              name("z1"));
+
+   Value *x2 = m_builder.CreateExtractElement(in2,
+                                              m_storage->constantInt(0),
+                                              name("x2"));
+   Value *y2 = m_builder.CreateExtractElement(in2,
+                                              m_storage->constantInt(1),
+                                              name("y2"));
+   Value *z2 = m_builder.CreateExtractElement(in2,
+                                              m_storage->constantInt(2),
+                                              name("z2"));
+   Value *y1z2 = mul(y1, z2);
+   Value *z1y2 = mul(z1, y2);
+
+   Value *z1x2 = mul(z1, x2);
+   Value *x1z2 = mul(x1, z2);
+
+   Value *x1y2 = mul(x1, y2);
+   Value *y1x2 = mul(y1, x2);
+
+   return vectorFromVals(sub(y1z2, z1y2), sub(z1x2, x1z2), sub(x1y2, y1x2));
+}
+
+
+llvm::Value * Instructions::abs(llvm::Value *in)
+{
+   std::vector<llvm::Value*> vec = extractVector(in);
+   Value *xabs  = callFAbs(vec[0]);
+   Value *yabs  = callFAbs(vec[1]);
+   Value *zabs  = callFAbs(vec[2]);
+   Value *wabs  = callFAbs(vec[3]);
+   return vectorFromVals(xabs, yabs, zabs, wabs);
+}
+
+void Instructions::ifop(llvm::Value *in)
+{
+   BasicBlock *ifthen = new BasicBlock(name("ifthen"), m_func,0);
+   BasicBlock *ifend = new BasicBlock(name("ifthenend"), m_func,0);
+
+   //BasicBlock *yblock = new BasicBlock(name("yblock"), m_func,0);
+   //BasicBlock *zblock = new BasicBlock(name("zblock"), m_func,0);
+   //BasicBlock *wblock = new BasicBlock(name("wblock"), m_func,0);
+
+   Constant *float0 = Constant::getNullValue(Type::FloatTy);
+
+   Value *x = m_builder.CreateExtractElement(in, m_storage->constantInt(0),
+                                             name("extractx"));
+   Value *xcmp = m_builder.CreateFCmpUNE(x, float0, name("xcmp"));
+   m_builder.CreateCondBr(xcmp, ifthen, ifend);
+   //m_builder.SetInsertPoint(yblock);
+
+   m_builder.SetInsertPoint(ifthen);
+   m_ifStack.push(ifend);
+}
+
+llvm::BasicBlock * Instructions::currentBlock() const
+{
+   return m_builder.GetInsertBlock();
+}
+
+void Instructions::elseop()
+{
+   assert(!m_ifStack.empty());
+   BasicBlock *ifend = new BasicBlock(name("ifend"), m_func,0);
+   m_builder.CreateBr(ifend);
+   m_builder.SetInsertPoint(m_ifStack.top());
+   currentBlock()->setName(name("ifelse"));
+   m_ifStack.pop();
+   m_ifStack.push(ifend);
+}
+
+void Instructions::endif()
+{
+   assert(!m_ifStack.empty());
+   m_builder.CreateBr(m_ifStack.top());
+   m_builder.SetInsertPoint(m_ifStack.top());
+   m_ifStack.pop();
+}
+
+llvm::Value * Instructions::lerp(llvm::Value *in1, llvm::Value *in2,
+                                 llvm::Value *in3)
+{
+   llvm::Value *m = mul(in1, in2);
+   llvm::Value *vec1 = constVector(1.f, 1.f, 1.f, 1.f);
+   llvm::Value *s = sub(vec1, in1);
+   return add(m, mul(s, in3));
+}
+
+void Instructions::beginLoop()
+{
+   BasicBlock *begin = new BasicBlock(name("loop"), m_func,0);
+   BasicBlock *end = new BasicBlock(name("endloop"), m_func,0);
+
+   m_builder.CreateBr(begin);
+   Loop loop;
+   loop.begin = begin;
+   loop.end   = end;
+   m_builder.SetInsertPoint(begin);
+   m_loopStack.push(loop);
+}
+
+void Instructions::endLoop()
+{
+   assert(!m_loopStack.empty());
+   Loop loop = m_loopStack.top();
+   m_builder.CreateBr(loop.begin);
+   loop.end->moveAfter(currentBlock());
+   m_builder.SetInsertPoint(loop.end);
+   m_loopStack.pop();
+}
+
+void Instructions::brk()
+{
+   assert(!m_loopStack.empty());
+   BasicBlock *unr = new BasicBlock(name("unreachable"), m_func,0);
+   m_builder.CreateBr(m_loopStack.top().end);
+   m_builder.SetInsertPoint(unr);
+}
+
+llvm::Value * Instructions::trunc(llvm::Value *in)
+{
+   std::vector<llvm::Value*> vec = extractVector(in);
+   Value *icastx = m_builder.CreateFPToSI(vec[0], IntegerType::get(32),
+                                          name("ftoix"));
+   Value *icasty = m_builder.CreateFPToSI(vec[1], IntegerType::get(32),
+                                          name("ftoiy"));
+   Value *icastz = m_builder.CreateFPToSI(vec[2], IntegerType::get(32),
+                                          name("ftoiz"));
+   Value *icastw = m_builder.CreateFPToSI(vec[3], IntegerType::get(32),
+                                          name("ftoiw"));
+   Value *fx = m_builder.CreateSIToFP(icastx, Type::FloatTy,
+                                      name("fx"));
+   Value *fy = m_builder.CreateSIToFP(icasty, Type::FloatTy,
+                                      name("fy"));
+   Value *fz = m_builder.CreateSIToFP(icastz, Type::FloatTy,
+                                      name("fz"));
+   Value *fw = m_builder.CreateSIToFP(icastw, Type::FloatTy,
+                                      name("fw"));
+   return vectorFromVals(fx, fy, fz, fw);
+}
+
+void Instructions::end()
+{
+   m_builder.CreateRetVoid();
+}
+
+void Instructions::cal(int label, llvm::Value *input)
+{
+   std::vector<Value*> params;
+   params.push_back(input);
+   llvm::Function *func = findFunction(label);
+
+   m_builder.CreateCall(func, params.begin(), params.end());
+}
+
+llvm::Function * Instructions::declareFunc(int label)
+{
+   PointerType *vecPtr = PointerType::getUnqual(m_floatVecType);
+   std::vector<const Type*> args;
+   args.push_back(vecPtr);
+   args.push_back(vecPtr);
+   args.push_back(vecPtr);
+   args.push_back(vecPtr);
+   ParamAttrsList *params = 0;
+   FunctionType *funcType = FunctionType::get(
+      /*Result=*/Type::VoidTy,
+      /*Params=*/args,
+      /*isVarArg=*/false);
+   std::string name = createFuncName(label);
+   Function *func = new Function(
+      /*Type=*/funcType,
+      /*Linkage=*/GlobalValue::ExternalLinkage,
+      /*Name=*/name.c_str(), m_mod);
+   func->setCallingConv(CallingConv::C);
+   func->setParamAttrs(params);
+   return func;
+}
+
+void Instructions::bgnSub(unsigned label)
+{
+   llvm::Function *func = findFunction(label);
+
+   Function::arg_iterator args = func->arg_begin();
+   Value *ptr_INPUT = args++;
+   ptr_INPUT->setName("INPUT");
+   m_storage->pushArguments(ptr_INPUT);
+
+   llvm::BasicBlock *entry = new BasicBlock("entry", func, 0);
+
+   m_func = func;
+   m_builder.SetInsertPoint(entry);
+}
+
+void Instructions::endSub()
+{
+   m_func = 0;
+   m_builder.SetInsertPoint(0);
+}
+
+llvm::Function * Instructions::findFunction(int label)
+{
+   llvm::Function *func = m_functions[label];
+   if (!func) {
+      func = declareFunc(label);
+      m_functions[label] = func;
+   }
+   return func;
+}
+
+llvm::Value * Instructions::constVector(float x, float y, float z, float w)
+{
+   std::vector<Constant*> vec(4);
+   vec[0] = ConstantFP::get(Type::FloatTy, APFloat(x));
+   vec[1] = ConstantFP::get(Type::FloatTy, APFloat(y));
+   vec[2] = ConstantFP::get(Type::FloatTy, APFloat(z));
+   vec[3] = ConstantFP::get(Type::FloatTy, APFloat(w));
+   return ConstantVector::get(m_floatVecType, vec);
+}
+
+
+std::vector<llvm::Value*> Instructions::extractVector(llvm::Value *vec)
+{
+   std::vector<llvm::Value*> elems(4);
+   elems[0] = m_builder.CreateExtractElement(vec, m_storage->constantInt(0),
+                                             name("x"));
+   elems[1] = m_builder.CreateExtractElement(vec, m_storage->constantInt(1),
+                                             name("y"));
+   elems[2] = m_builder.CreateExtractElement(vec, m_storage->constantInt(2),
+                                             name("z"));
+   elems[3] = m_builder.CreateExtractElement(vec, m_storage->constantInt(3),
+                                             name("w"));
+   return elems;
+}
+
+llvm::Value * Instructions::cmp(llvm::Value *in1, llvm::Value *in2, llvm::Value *in3)
+{
+   llvm::Function *func = m_mod->getFunction("cmp");
+   assert(func);
+
+   std::vector<Value*> params;
+   params.push_back(in1);
+   params.push_back(in2);
+   params.push_back(in3);
+   CallInst *call = m_builder.CreateCall(func, params.begin(), params.end(), name("cmpres"));
+   call->setTailCall(false);
+   return call;
+}
+
+llvm::Value * Instructions::cos(llvm::Value *in)
+{
+#if 0
+   llvm::Function *func = m_mod->getFunction("vcos");
+   assert(func);
+
+   CallInst *call = m_builder.CreateCall(func, in, name("cosres"));
+   call->setTailCall(false);
+   return call;
+#else
+   std::vector<llvm::Value*> elems = extractVector(in);
+   Function *func = m_mod->getFunction("cosf");
+   assert(func);
+   CallInst *cos = m_builder.CreateCall(func, elems[0], name("cosres"));
+   cos->setCallingConv(CallingConv::C);
+   cos->setTailCall(true);
+   return vectorFromVals(cos, cos, cos, cos);
+#endif
+}
+
+llvm::Value * Instructions::scs(llvm::Value *in)
+{
+   llvm::Function *func = m_mod->getFunction("scs");
+   assert(func);
+
+   CallInst *call = m_builder.CreateCall(func, in, name("scsres"));
+   call->setTailCall(false);
+   return call;
+}
+
+llvm::Value * Instructions::kilp(llvm::Value *in)
+{
+   llvm::Function *func = m_mod->getFunction("kilp");
+   assert(func);
+
+   CallInst *call = m_builder.CreateCall(func, in, name("kilpres"));
+   call->setTailCall(false);
+   return call;
+}
+
+llvm::Value * Instructions::sin(llvm::Value *in)
+{
+   llvm::Function *func = m_mod->getFunction("vsin");
+   assert(func);
+
+   CallInst *call = m_builder.CreateCall(func, in, name("sinres"));
+   call->setTailCall(false);
+   return call;
+}
+#endif //MESA_LLVM
+
+
diff --git a/src/gallium/aux/llvm/instructions.h b/src/gallium/aux/llvm/instructions.h
new file mode 100644
index 0000000000..9ebc17dd8e
--- /dev/null
+++ b/src/gallium/aux/llvm/instructions.h
@@ -0,0 +1,152 @@
+/**************************************************************************
+ *
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ **************************************************************************/
+
+ /*
+  * Authors:
+  *   Zack Rusin zack@tungstengraphics.com
+  */
+
+#ifndef INSTRUCTIONS_H
+#define INSTRUCTIONS_H
+
+#include <llvm/BasicBlock.h>
+#include <llvm/Module.h>
+#include <llvm/Value.h>
+#include <llvm/Support/LLVMBuilder.h>
+
+#include <map>
+#include <stack>
+
+namespace llvm {
+   class VectorType;
+   class Function;
+}
+
+class Storage;
+
+class Instructions
+{
+public:
+   Instructions(llvm::Module *mod, llvm::Function *func, llvm::BasicBlock *block,
+                Storage *storage);
+
+   llvm::BasicBlock *currentBlock() const;
+
+   llvm::Value *abs(llvm::Value *in1);
+   llvm::Value *arl(llvm::Value *in1);
+   llvm::Value *add(llvm::Value *in1, llvm::Value *in2);
+   void         beginLoop();
+   void         bgnSub(unsigned);
+   void         brk();
+   void         cal(int label, llvm::Value *input);
+   llvm::Value *cmp(llvm::Value *in1, llvm::Value *in2, llvm::Value *in3);
+   llvm::Value *cos(llvm::Value *in);
+   llvm::Value *cross(llvm::Value *in1, llvm::Value *in2);
+   llvm::Value *dp3(llvm::Value *in1, llvm::Value *in2);
+   llvm::Value *dp4(llvm::Value *in1, llvm::Value *in2);
+   llvm::Value *dph(llvm::Value *in1, llvm::Value *in2);
+   llvm::Value *dst(llvm::Value *in1, llvm::Value *in2);
+   void         elseop();
+   void         endif();
+   void         endLoop();
+   void         end();
+   void         endSub();
+   llvm::Value *ex2(llvm::Value *in);
+   llvm::Value *floor(llvm::Value *in);
+   llvm::Value *frc(llvm::Value *in);
+   void         ifop(llvm::Value *in);
+   llvm::Value *kilp(llvm::Value *in);
+   llvm::Value *lerp(llvm::Value *in1, llvm::Value *in2,
+                     llvm::Value *in3);
+   llvm::Value *lit(llvm::Value *in);
+   llvm::Value *lg2(llvm::Value *in);
+   llvm::Value *madd(llvm::Value *in1, llvm::Value *in2,
+                     llvm::Value *in2);
+   llvm::Value *min(llvm::Value *in1, llvm::Value *in2);
+   llvm::Value *max(llvm::Value *in1, llvm::Value *in2);
+   llvm::Value *mul(llvm::Value *in1, llvm::Value *in2);
+   llvm::Value *pow(llvm::Value *in1, llvm::Value *in2);
+   llvm::Value *rcp(llvm::Value *in);
+   llvm::Value *rsq(llvm::Value *in);
+   llvm::Value *scs(llvm::Value *in);
+   llvm::Value *sge(llvm::Value *in1, llvm::Value *in2);
+   llvm::Value *sgt(llvm::Value *in1, llvm::Value *in2);
+   llvm::Value *sin(llvm::Value *in);
+   llvm::Value *slt(llvm::Value *in1, llvm::Value *in2);
+   llvm::Value *sub(llvm::Value *in1, llvm::Value *in2);
+   llvm::Value *trunc(llvm::Value *in);
+
+   void printVector(llvm::Value *val);
+private:
+   const char *name(const char *prefix);
+
+   llvm::Value *callFAbs(llvm::Value *val);
+   llvm::Value *callFloor(llvm::Value *val);
+   llvm::Value *callFSqrt(llvm::Value *val);
+   llvm::Value *callFLog(llvm::Value *val);
+   llvm::Value *callPow(llvm::Value *val1, llvm::Value *val2);
+
+   llvm::Value *vectorFromVals(llvm::Value *x, llvm::Value *y,
+                               llvm::Value *z, llvm::Value *w=0);
+
+   llvm::Value *constVector(float x, float y, float z, float w);
+
+   llvm::Function *declarePrintf();
+   llvm::Function *declareFunc(int label);
+
+   llvm::Function *findFunction(int label);
+
+   std::vector<llvm::Value*> extractVector(llvm::Value *vec);
+private:
+   llvm::Module             *m_mod;
+   llvm::Function           *m_func;
+   char                      m_name[32];
+   llvm::LLVMFoldingBuilder  m_builder;
+   int                       m_idx;
+
+   llvm::VectorType *m_floatVecType;
+
+   llvm::Function   *m_llvmFSqrt;
+   llvm::Function   *m_llvmFAbs;
+   llvm::Function   *m_llvmPow;
+   llvm::Function   *m_llvmFloor;
+   llvm::Function   *m_llvmFlog;
+   llvm::Function   *m_llvmLit;
+
+   llvm::Constant   *m_fmtPtr;
+
+   std::stack<llvm::BasicBlock*> m_ifStack;
+   struct Loop {
+      llvm::BasicBlock *begin;
+      llvm::BasicBlock *end;
+   };
+   std::stack<Loop> m_loopStack;
+   std::map<int, llvm::Function*> m_functions;
+   Storage *m_storage;
+};
+
+#endif
diff --git a/src/gallium/aux/llvm/instructionssoa.cpp b/src/gallium/aux/llvm/instructionssoa.cpp
new file mode 100644
index 0000000000..a4d5046637
--- /dev/null
+++ b/src/gallium/aux/llvm/instructionssoa.cpp
@@ -0,0 +1,121 @@
+#include "instructionssoa.h"
+
+#include "storagesoa.h"
+
+#include <llvm/Constants.h>
+
+using namespace llvm;
+
+InstructionsSoa::InstructionsSoa(llvm::Module *mod, llvm::Function *func,
+                                 llvm::BasicBlock *block, StorageSoa *storage)
+   : m_builder(block),
+     m_storage(storage),
+     m_idx(0)
+{
+}
+
+const char * InstructionsSoa::name(const char *prefix) const
+{
+   ++m_idx;
+   snprintf(m_name, 32, "%s%d", prefix, m_idx);
+   return m_name;
+}
+
+llvm::Value * InstructionsSoa::vectorFromVals(llvm::Value *x, llvm::Value *y,
+                                              llvm::Value *z, llvm::Value *w)
+{
+   VectorType  *vectorType = VectorType::get(Type::FloatTy, 4);
+   Constant *constVector = Constant::getNullValue(vectorType);
+   Value *res = m_builder.CreateInsertElement(constVector, x,
+                                              m_storage->constantInt(0),
+                                              name("vecx"));
+   res = m_builder.CreateInsertElement(res, y, m_storage->constantInt(1),
+                               name("vecxy"));
+   res = m_builder.CreateInsertElement(res, z, m_storage->constantInt(2),
+                               name("vecxyz"));
+   if (w)
+      res = m_builder.CreateInsertElement(res, w, m_storage->constantInt(3),
+                                          name("vecxyzw"));
+   return res;
+}
+
+std::vector<llvm::Value*> InstructionsSoa::arl(const std::vector<llvm::Value*> in)
+{
+   std::vector<llvm::Value*> res(4);
+
+   //Extract x's
+   llvm::Value *x1 = m_builder.CreateExtractElement(in[0],
+                                                    m_storage->constantInt(0),
+                                                    name("extractX"));
+   //cast it to an unsigned int
+   x1 = m_builder.CreateFPToUI(x1, IntegerType::get(32), name("x1IntCast"));
+
+   res[0] = x1;//vectorFromVals(x1, x2, x3, x4);
+   //only x is valid. the others shouldn't be necessary
+   /*
+   res[1] = Constant::getNullValue(m_floatVecType);
+   res[2] = Constant::getNullValue(m_floatVecType);
+   res[3] = Constant::getNullValue(m_floatVecType);
+   */
+
+   return res;
+}
+
+
+std::vector<llvm::Value*> InstructionsSoa::add(const std::vector<llvm::Value*> in1,
+                                               const std::vector<llvm::Value*> in2)
+{
+   std::vector<llvm::Value*> res(4);
+
+   res[0] = m_builder.CreateAdd(in1[0], in2[0], name("addx"));
+   res[1] = m_builder.CreateAdd(in1[1], in2[1], name("addy"));
+   res[2] = m_builder.CreateAdd(in1[2], in2[2], name("addz"));
+   res[3] = m_builder.CreateAdd(in1[3], in2[3], name("addw"));
+
+   return res;
+}
+
+std::vector<llvm::Value*> InstructionsSoa::mul(const std::vector<llvm::Value*> in1,
+                                               const std::vector<llvm::Value*> in2)
+{
+   std::vector<llvm::Value*> res(4);
+
+   res[0] = m_builder.CreateMul(in1[0], in2[0], name("mulx"));
+   res[1] = m_builder.CreateMul(in1[1], in2[1], name("muly"));
+   res[2] = m_builder.CreateMul(in1[2], in2[2], name("mulz"));
+   res[3] = m_builder.CreateMul(in1[3], in2[3], name("mulw"));
+
+   return res;
+}
+
+void InstructionsSoa::end()
+{
+   m_builder.CreateRetVoid();
+}
+
+std::vector<llvm::Value*> InstructionsSoa::madd(const std::vector<llvm::Value*> in1,
+                                                const std::vector<llvm::Value*> in2,
+                                                const std::vector<llvm::Value*> in3)
+{
+   std::vector<llvm::Value*> res = mul(in1, in2);
+   return add(res, in3);
+}
+
+std::vector<llvm::Value*> InstructionsSoa::extractVector(llvm::Value *vector)
+{
+   std::vector<llvm::Value*> res(4);
+   res[0] = m_builder.CreateExtractElement(vector,
+                                           m_storage->constantInt(0),
+                                           name("extract1X"));
+   res[1] = m_builder.CreateExtractElement(vector,
+                                           m_storage->constantInt(1),
+                                           name("extract2X"));
+   res[2] = m_builder.CreateExtractElement(vector,
+                                           m_storage->constantInt(2),
+                                           name("extract3X"));
+   res[3] = m_builder.CreateExtractElement(vector,
+                                           m_storage->constantInt(3),
+                                           name("extract4X"));
+
+   return res;
+}
diff --git a/src/gallium/aux/llvm/instructionssoa.h b/src/gallium/aux/llvm/instructionssoa.h
new file mode 100644
index 0000000000..4169dcbb2e
--- /dev/null
+++ b/src/gallium/aux/llvm/instructionssoa.h
@@ -0,0 +1,74 @@
+/**************************************************************************
+ *
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ **************************************************************************/
+
+#ifndef INSTRUCTIONSSOA_H
+#define INSTRUCTIONSSOA_H
+
+#include <llvm/Support/LLVMBuilder.h>
+
+#include <vector>
+
+namespace llvm {
+   class Module;
+   class Function;
+   class BasicBlock;
+   class Value;
+}
+class StorageSoa;
+
+class InstructionsSoa
+{
+public:
+   InstructionsSoa(llvm::Module *mod, llvm::Function *func,
+                   llvm::BasicBlock *block, StorageSoa *storage);
+
+   std::vector<llvm::Value*> arl(const std::vector<llvm::Value*> in);
+
+   std::vector<llvm::Value*> add(const std::vector<llvm::Value*> in1,
+                                 const std::vector<llvm::Value*> in2);
+   std::vector<llvm::Value*> madd(const std::vector<llvm::Value*> in1,
+                                  const std::vector<llvm::Value*> in2,
+                                  const std::vector<llvm::Value*> in3);
+   std::vector<llvm::Value*> mul(const std::vector<llvm::Value*> in1,
+                                 const std::vector<llvm::Value*> in2);
+   void         end();
+
+   std::vector<llvm::Value*> extractVector(llvm::Value *vector);
+private:
+   const char * name(const char *prefix) const;
+   llvm::Value *vectorFromVals(llvm::Value *x, llvm::Value *y,
+                               llvm::Value *z, llvm::Value *w);
+private:
+   llvm::LLVMFoldingBuilder  m_builder;
+   StorageSoa *m_storage;
+private:
+   mutable int  m_idx;
+   mutable char m_name[32];
+};
+
+
+#endif
diff --git a/src/gallium/aux/llvm/llvm_builtins.c b/src/gallium/aux/llvm/llvm_builtins.c
new file mode 100644
index 0000000000..4f98d754ba
--- /dev/null
+++ b/src/gallium/aux/llvm/llvm_builtins.c
@@ -0,0 +1,115 @@
+/*clang --emit-llvm llvm_builtins.c |llvm-as|opt -std-compile-opts|llvm2cpp -gen-contents -o=gallivm_builtins.cpp -f -for=shader -funcname=createGallivmBuiltins*/
+/**************************************************************************
+ *
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ **************************************************************************/
+
+ /*
+  * Authors:
+  *   Zack Rusin zack@tungstengraphics.com
+  */
+typedef __attribute__(( ocu_vector_type(4) )) float float4;
+
+extern float powf(float a, float b);
+
+inline float approx(float a, float b)
+{
+    if (b < -128.0f) b = -128.0f;
+    if (b > 128.0f)   b = 128.0f;
+    if (a < 0) a = 0;
+    return powf(a, b);
+}
+
+inline float4 lit(float4 tmp)
+{
+    float4 result;
+    result.x = 1.0;
+    result.w = 1.0;
+    if (tmp.x > 0) {
+        result.y = tmp.x;
+        result.z = approx(tmp.y, tmp.w);
+    } else {
+        result.y = 0;
+        result.z = 0;
+    }
+    return result;
+}
+
+inline float4 cmp(float4 tmp0, float4 tmp1, float4 tmp2)
+{
+   float4 result;
+
+   result.x = (tmp0.x < 0.0) ? tmp1.x : tmp2.x;
+   result.y = (tmp0.y < 0.0) ? tmp1.y : tmp2.y;
+   result.z = (tmp0.z < 0.0) ? tmp1.z : tmp2.z;
+   result.w = (tmp0.w < 0.0) ? tmp1.w : tmp2.w;
+
+   return result;
+}
+
+extern float cosf(float  val);
+extern float sinf(float  val);
+
+inline float4 vcos(float4 val)
+{
+   float4 result;
+   printf("VEC IN   is %f %f %f %f\n", val.x, val.y, val.z, val.w);
+   result.x = cosf(val.x);
+   result.y = cosf(val.x);
+   result.z = cosf(val.x);
+   result.w = cosf(val.x);
+   printf("VEC OUT  is %f %f %f %f\n", result.x, result.y, result.z, result.w);
+   return result;
+}
+
+inline float4 scs(float4 val)
+{
+   float4 result;
+   float tmp = val.x;
+   result.x = cosf(tmp);
+   result.y = sinf(tmp);
+   return result;
+}
+
+
+inline float4 vsin(float4 val)
+{
+   float4 result;
+   float tmp = val.x;
+   float res = sinf(tmp);
+   result.x = res;
+   result.y = res;
+   result.z = res;
+   result.w = res;
+   return result;
+}
+
+inline int kilp(float4 val)
+{
+   if (val.x < 0 || val.y < 0 || val.z < 0 || val.w < 0)
+      return 1;
+   else
+      return 0;
+}
diff --git a/src/gallium/aux/llvm/loweringpass.cpp b/src/gallium/aux/llvm/loweringpass.cpp
new file mode 100644
index 0000000000..556dbec366
--- /dev/null
+++ b/src/gallium/aux/llvm/loweringpass.cpp
@@ -0,0 +1,17 @@
+#include "loweringpass.h"
+
+using namespace llvm;
+
+char LoweringPass::ID = 0;
+RegisterPass<LoweringPass> X("lowering", "Lowering Pass");
+
+LoweringPass::LoweringPass()
+   :  ModulePass((intptr_t)&ID)
+{
+}
+
+bool LoweringPass::runOnModule(Module &m)
+{
+   llvm::cerr << "Hello: " << m.getModuleIdentifier() << "\n";
+   return false;
+}
diff --git a/src/gallium/aux/llvm/loweringpass.h b/src/gallium/aux/llvm/loweringpass.h
new file mode 100644
index 0000000000..f62dcf6ba7
--- /dev/null
+++ b/src/gallium/aux/llvm/loweringpass.h
@@ -0,0 +1,15 @@
+#ifndef LOWERINGPASS_H
+#define LOWERINGPASS_H
+
+#include "llvm/Pass.h"
+#include "llvm/Module.h"
+
+struct LoweringPass : public llvm::ModulePass
+{
+   static char ID;
+   LoweringPass();
+
+   virtual bool runOnModule(llvm::Module &m);
+};
+
+#endif
diff --git a/src/gallium/aux/llvm/storage.cpp b/src/gallium/aux/llvm/storage.cpp
new file mode 100644
index 0000000000..c4326de8c5
--- /dev/null
+++ b/src/gallium/aux/llvm/storage.cpp
@@ -0,0 +1,364 @@
+/**************************************************************************
+ *
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ **************************************************************************/
+
+ /*
+  * Authors:
+  *   Zack Rusin zack@tungstengraphics.com
+  */
+#ifdef MESA_LLVM
+
+#include "storage.h"
+
+#include "gallivm_p.h"
+
+#include "pipe/p_shader_tokens.h"
+#include <llvm/BasicBlock.h>
+#include <llvm/Module.h>
+#include <llvm/Value.h>
+
+#include <llvm/CallingConv.h>
+#include <llvm/Constants.h>
+#include <llvm/DerivedTypes.h>
+#include <llvm/InstrTypes.h>
+#include <llvm/Instructions.h>
+
+using namespace llvm;
+
+Storage::Storage(llvm::BasicBlock *block, llvm::Value *input)
+   : m_block(block),
+     m_INPUT(input),
+     m_addrs(32),
+     m_idx(0)
+{
+   m_floatVecType = VectorType::get(Type::FloatTy, 4);
+   m_intVecType   = VectorType::get(IntegerType::get(32), 4);
+
+   m_undefFloatVec = UndefValue::get(m_floatVecType);
+   m_undefIntVec   = UndefValue::get(m_intVecType);
+   m_extSwizzleVec = 0;
+
+   m_numConsts = 0;
+}
+
+//can only build vectors with all members in the [0, 9] range
+llvm::Constant *Storage::shuffleMask(int vec)
+{
+   if (!m_extSwizzleVec) {
+      std::vector<Constant*> elems;
+      elems.push_back(ConstantFP::get(Type::FloatTy, APFloat(0.f)));
+      elems.push_back(ConstantFP::get(Type::FloatTy, APFloat(1.f)));
+      elems.push_back(ConstantFP::get(Type::FloatTy, APFloat(0.f)));
+      elems.push_back(ConstantFP::get(Type::FloatTy, APFloat(1.f)));
+      m_extSwizzleVec = ConstantVector::get(m_floatVecType, elems);
+   }
+
+   if (m_intVecs.find(vec) != m_intVecs.end()) {
+      return m_intVecs[vec];
+   }
+   int origVec = vec;
+   Constant* const_vec = 0;
+   if (origVec == 0) {
+      const_vec = Constant::getNullValue(m_intVecType);
+   } else {
+      int x = gallivm_x_swizzle(vec);
+      int y = gallivm_y_swizzle(vec);
+      int z = gallivm_z_swizzle(vec);
+      int w = gallivm_w_swizzle(vec);
+      std::vector<Constant*> elems;
+      elems.push_back(constantInt(x));
+      elems.push_back(constantInt(y));
+      elems.push_back(constantInt(z));
+      elems.push_back(constantInt(w));
+      const_vec = ConstantVector::get(m_intVecType, elems);
+   }
+
+   m_intVecs[origVec] = const_vec;
+   return const_vec;
+}
+
+llvm::ConstantInt *Storage::constantInt(int idx)
+{
+   if (m_constInts.find(idx) != m_constInts.end()) {
+      return m_constInts[idx];
+   }
+   ConstantInt *const_int = ConstantInt::get(APInt(32,  idx));
+   m_constInts[idx] = const_int;
+   return const_int;
+}
+
+llvm::Value *Storage::inputElement(int idx, llvm::Value *indIdx)
+{
+   Value *val = element(InputsArg, idx, indIdx);
+   LoadInst *load = new LoadInst(val, name("input"), false, m_block);
+   load->setAlignment(8);
+
+   return load;
+}
+
+llvm::Value *Storage::constElement(int idx, llvm::Value *indIdx)
+{
+   m_numConsts = ((idx + 1) > m_numConsts) ? (idx + 1) : m_numConsts;
+
+   Value *elem = element(ConstsArg, idx, indIdx);
+   LoadInst *load = new LoadInst(elem, name("const"), false, m_block);
+   load->setAlignment(8);
+   return load;
+}
+
+llvm::Value *Storage::shuffleVector(llvm::Value *vec, int shuffle)
+{
+   Constant *mask = shuffleMask(shuffle);
+   ShuffleVectorInst *res =
+      new ShuffleVectorInst(vec, m_extSwizzleVec, mask,
+                            name("shuffle"), m_block);
+   return res;
+}
+
+
+llvm::Value *Storage::tempElement(int idx, llvm::Value *indIdx)
+{
+   Value *elem = element(TempsArg, idx, indIdx);
+
+   LoadInst *load = new LoadInst(elem, name("temp"), false, m_block);
+   load->setAlignment(8);
+
+   return load;
+}
+
+void Storage::setTempElement(int idx, llvm::Value *val, int mask)
+{
+   if (mask != TGSI_WRITEMASK_XYZW) {
+      llvm::Value *templ = 0;
+      if (m_tempWriteMap[idx])
+         templ = tempElement(idx);
+      val = maskWrite(val, mask, templ);
+   }
+   Value *elem = element(TempsArg, idx);
+   StoreInst *st = new StoreInst(val, elem, false, m_block);
+   st->setAlignment(8);
+   m_tempWriteMap[idx] = true;
+}
+
+void Storage::setOutputElement(int dstIdx, llvm::Value *val, int mask)
+{
+   if (mask != TGSI_WRITEMASK_XYZW) {
+      llvm::Value *templ = 0;
+      if (m_destWriteMap[dstIdx])
+         templ = outputElement(dstIdx);
+      val = maskWrite(val, mask, templ);
+   }
+
+   Value *elem = element(DestsArg, dstIdx);
+   StoreInst *st = new StoreInst(val, elem, false, m_block);
+   st->setAlignment(8);
+   m_destWriteMap[dstIdx] = true;
+}
+
+llvm::Value *Storage::maskWrite(llvm::Value *src, int mask, llvm::Value *templ)
+{
+   llvm::Value *dst = templ;
+   if (!dst)
+      dst = Constant::getNullValue(m_floatVecType);
+   if ((mask & TGSI_WRITEMASK_X)) {
+      llvm::Value *x = new ExtractElementInst(src, unsigned(0),
+                                              name("x"), m_block);
+      dst = new InsertElementInst(dst, x, unsigned(0),
+                                  name("dstx"), m_block);
+   }
+   if ((mask & TGSI_WRITEMASK_Y)) {
+      llvm::Value *y = new ExtractElementInst(src, unsigned(1),
+                                              name("y"), m_block);
+      dst = new InsertElementInst(dst, y, unsigned(1),
+                                  name("dsty"), m_block);
+   }
+   if ((mask & TGSI_WRITEMASK_Z)) {
+      llvm::Value *z = new ExtractElementInst(src, unsigned(2),
+                                              name("z"), m_block);
+      dst = new InsertElementInst(dst, z, unsigned(2),
+                                  name("dstz"), m_block);
+   }
+   if ((mask & TGSI_WRITEMASK_W)) {
+      llvm::Value *w = new ExtractElementInst(src, unsigned(3),
+                                              name("w"), m_block);
+      dst = new InsertElementInst(dst, w, unsigned(3),
+                                  name("dstw"), m_block);
+   }
+   return dst;
+}
+
+const char * Storage::name(const char *prefix)
+{
+   ++m_idx;
+   snprintf(m_name, 32, "%s%d", prefix, m_idx);
+   return m_name;
+}
+
+int Storage::numConsts() const
+{
+   return m_numConsts;
+}
+
+llvm::Value * Storage::addrElement(int idx) const
+{
+   Value *ret = m_addrs[idx];
+   if (!ret)
+      return m_undefFloatVec;
+   return ret;
+}
+
+void Storage::setAddrElement(int idx, llvm::Value *val, int mask)
+{
+   if (mask != TGSI_WRITEMASK_XYZW) {
+      llvm::Value *templ = m_addrs[idx];
+      val = maskWrite(val, mask, templ);
+   }
+   m_addrs[idx] = val;
+}
+
+llvm::Value * Storage::extractIndex(llvm::Value *vec)
+{
+   llvm::Value *x = new ExtractElementInst(vec, unsigned(0),
+                                           name("x"), m_block);
+   return new FPToSIInst(x, IntegerType::get(32), name("intidx"), m_block);
+}
+
+void Storage::setCurrentBlock(llvm::BasicBlock *block)
+{
+   m_block = block;
+}
+
+llvm::Value * Storage::outputElement(int idx, llvm::Value *indIdx)
+{
+   Value *elem = element(DestsArg, idx, indIdx);
+   LoadInst *load = new LoadInst(elem, name("output"), false, m_block);
+   load->setAlignment(8);
+
+   return load;
+}
+
+llvm::Value * Storage::inputPtr() const
+{
+   return m_INPUT;
+}
+
+void Storage::pushArguments(llvm::Value *input)
+{
+   m_argStack.push(m_INPUT);
+
+   m_INPUT = input;
+}
+
+void Storage::popArguments()
+{
+   m_INPUT = m_argStack.top();
+   m_argStack.pop();
+}
+
+void Storage::pushTemps()
+{
+   m_extSwizzleVec = 0;
+}
+
+void Storage::popTemps()
+{
+}
+
+llvm::Value * Storage::immediateElement(int idx)
+{
+   return m_immediates[idx];
+}
+
+void Storage::addImmediate(float *val)
+{
+   std::vector<Constant*> vec(4);
+   vec[0] = ConstantFP::get(Type::FloatTy, APFloat(val[0]));
+   vec[1] = ConstantFP::get(Type::FloatTy, APFloat(val[1]));
+   vec[2] = ConstantFP::get(Type::FloatTy, APFloat(val[2]));
+   vec[3] = ConstantFP::get(Type::FloatTy, APFloat(val[3]));
+   m_immediates.push_back(ConstantVector::get(m_floatVecType, vec));
+}
+
+
+llvm::Value * Storage::elemPtr(Args arg)
+{
+   std::vector<Value*> indices;
+   indices.push_back(constantInt(0));
+   indices.push_back(constantInt(static_cast<int>(arg)));
+   GetElementPtrInst *getElem = new GetElementPtrInst(m_INPUT,
+                                                      indices.begin(),
+                                                      indices.end(),
+                                                      name("input_ptr"),
+                                                      m_block);
+   return new LoadInst(getElem, name("input_field"), false, m_block);
+}
+
+llvm::Value * Storage::elemIdx(llvm::Value *ptr, int idx,
+                               llvm::Value *indIdx )
+{
+   GetElementPtrInst *getElem = 0;
+
+   if (indIdx) {
+      getElem = new GetElementPtrInst(ptr,
+                                      BinaryOperator::create(Instruction::Add,
+                                                             indIdx,
+                                                             constantInt(idx),
+                                                             name("add"),
+                                                             m_block),
+                                      name("field"),
+                                      m_block);
+   } else {
+      getElem = new GetElementPtrInst(ptr,
+                                      constantInt(idx),
+                                      name("field"),
+                                      m_block);
+   }
+   return getElem;
+}
+
+llvm::Value * Storage::element(Args arg, int idx, llvm::Value *indIdx )
+{
+   Value *val = elemPtr(arg);
+   return elemIdx(val, idx, indIdx);
+}
+
+void Storage::setKilElement(llvm::Value *val)
+{
+   std::vector<Value*> indices;
+   indices.push_back(constantInt(0));
+   indices.push_back(constantInt(static_cast<int>(KilArg)));
+   GetElementPtrInst *elem = new GetElementPtrInst(m_INPUT,
+                                                   indices.begin(),
+                                                   indices.end(),
+                                                   name("kil_ptr"),
+                                                   m_block);
+   StoreInst *st = new StoreInst(val, elem, false, m_block);
+   st->setAlignment(8);
+}
+
+#endif //MESA_LLVM
+
+
diff --git a/src/gallium/aux/llvm/storage.h b/src/gallium/aux/llvm/storage.h
new file mode 100644
index 0000000000..8574f7554e
--- /dev/null
+++ b/src/gallium/aux/llvm/storage.h
@@ -0,0 +1,133 @@
+/**************************************************************************
+ *
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ **************************************************************************/
+
+ /*
+  * Authors:
+  *   Zack Rusin zack@tungstengraphics.com
+  */
+
+#ifndef STORAGE_H
+#define STORAGE_H
+
+#include <map>
+#include <set>
+#include <stack>
+#include <vector>
+
+namespace llvm {
+   class BasicBlock;
+   class Constant;
+   class ConstantInt;
+   class LoadInst;
+   class Value;
+   class VectorType;
+}
+
+class Storage
+{
+public:
+   Storage(llvm::BasicBlock *block,
+           llvm::Value *input);
+
+   llvm::Value *inputPtr() const;
+
+   void setCurrentBlock(llvm::BasicBlock *block);
+
+   llvm::ConstantInt *constantInt(int);
+   llvm::Constant *shuffleMask(int vec);
+   llvm::Value *inputElement(int idx, llvm::Value *indIdx =0);
+   llvm::Value *constElement(int idx, llvm::Value *indIdx =0);
+   llvm::Value *outputElement(int idx, llvm::Value *indIdx =0);
+   llvm::Value *tempElement(int idx, llvm::Value *indIdx =0);
+   llvm::Value *immediateElement(int idx);
+
+   void setOutputElement(int dstIdx, llvm::Value *val, int mask);
+   void setTempElement(int idx, llvm::Value *val, int mask);
+
+   llvm::Value *addrElement(int idx) const;
+   void setAddrElement(int idx, llvm::Value *val, int mask);
+
+   void setKilElement(llvm::Value *val);
+
+   llvm::Value *shuffleVector(llvm::Value *vec, int shuffle);
+
+   llvm::Value *extractIndex(llvm::Value *vec);
+
+   int numConsts() const;
+
+   void pushArguments(llvm::Value *input);
+   void popArguments();
+   void pushTemps();
+   void popTemps();
+
+   void addImmediate(float *val);
+
+private:
+   llvm::Value *maskWrite(llvm::Value *src, int mask, llvm::Value *templ);
+   const char *name(const char *prefix);
+
+   enum Args {
+      DestsArg   = 0,
+      InputsArg  = 1,
+      TempsArg   = 2,
+      ConstsArg  = 3,
+      KilArg     = 4
+   };
+   llvm::Value *elemPtr(Args arg);
+   llvm::Value *elemIdx(llvm::Value *ptr, int idx,
+                        llvm::Value *indIdx = 0);
+   llvm::Value *element(Args arg, int idx, llvm::Value *indIdx = 0);
+
+private:
+   llvm::BasicBlock *m_block;
+   llvm::Value *m_INPUT;
+
+   std::map<int, llvm::ConstantInt*> m_constInts;
+   std::map<int, llvm::Constant*>    m_intVecs;
+   std::vector<llvm::Value*>         m_addrs;
+   std::vector<llvm::Constant*>      m_immediates;
+
+   llvm::VectorType *m_floatVecType;
+   llvm::VectorType *m_intVecType;
+
+   char        m_name[32];
+   int         m_idx;
+
+   int         m_numConsts;
+
+   std::map<int, bool > m_destWriteMap;
+   std::map<int, bool > m_tempWriteMap;
+
+   llvm::Value      *m_undefFloatVec;
+   llvm::Value      *m_undefIntVec;
+   llvm::Value      *m_extSwizzleVec;
+
+   std::stack<llvm::Value*> m_argStack;
+   std::stack<std::vector<llvm::Value*> > m_tempStack;
+};
+
+#endif
diff --git a/src/gallium/aux/llvm/storagesoa.cpp b/src/gallium/aux/llvm/storagesoa.cpp
new file mode 100644
index 0000000000..ed0674a96f
--- /dev/null
+++ b/src/gallium/aux/llvm/storagesoa.cpp
@@ -0,0 +1,389 @@
+/**************************************************************************
+ *
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ **************************************************************************/
+
+#include "storagesoa.h"
+
+#include "gallivm_p.h"
+
+#include "pipe/p_shader_tokens.h"
+#include "pipe/p_debug.h"
+
+#include <llvm/BasicBlock.h>
+#include <llvm/Module.h>
+#include <llvm/Value.h>
+
+#include <llvm/CallingConv.h>
+#include <llvm/Constants.h>
+#include <llvm/DerivedTypes.h>
+#include <llvm/InstrTypes.h>
+#include <llvm/Instructions.h>
+
+using namespace llvm;
+
+
+StorageSoa::StorageSoa(llvm::BasicBlock *block,
+                       llvm::Value *input,
+                       llvm::Value *output,
+                       llvm::Value *consts,
+                       llvm::Value *temps)
+   : m_block(block),
+     m_input(input),
+     m_output(output),
+     m_consts(consts),
+     m_temps(temps),
+     m_immediates(0),
+     m_idx(0)
+{
+}
+
+void StorageSoa::addImmediate(float *vec)
+{
+   std::vector<float> vals(4);
+   vals[0] = vec[0];
+   vals[1] = vec[1];
+   vals[2] = vec[2];
+   vals[3] = vec[3];
+   m_immediatesToFlush.push_back(vals);
+}
+
+void StorageSoa::declareImmediates()
+{
+   if (m_immediatesToFlush.empty())
+      return;
+
+   VectorType *vectorType = VectorType::get(Type::FloatTy, 4);
+   ArrayType  *vectorChannels = ArrayType::get(vectorType, 4);
+   ArrayType  *arrayType = ArrayType::get(vectorChannels, m_immediatesToFlush.size());
+
+   m_immediates = new GlobalVariable(
+      /*Type=*/arrayType,
+      /*isConstant=*/false,
+      /*Linkage=*/GlobalValue::ExternalLinkage,
+      /*Initializer=*/0, // has initializer, specified below
+      /*Name=*/name("immediates"),
+      currentModule());
+
+   std::vector<Constant*> arrayVals;
+   for (unsigned int i = 0; i < m_immediatesToFlush.size(); ++i) {
+      std::vector<float> vec = m_immediatesToFlush[i];
+      std::vector<float> vals(4);
+      std::vector<Constant*> channelArray;
+
+      vals[0] = vec[0]; vals[1] = vec[0]; vals[2] = vec[0]; vals[3] = vec[0];
+      llvm::Constant *xChannel = createConstGlobalVector(vals);
+
+      vals[0] = vec[1]; vals[1] = vec[1]; vals[2] = vec[1]; vals[3] = vec[1];
+      llvm::Constant *yChannel = createConstGlobalVector(vals);
+
+      vals[0] = vec[2]; vals[1] = vec[2]; vals[2] = vec[2]; vals[3] = vec[2];
+      llvm::Constant *zChannel = createConstGlobalVector(vals);
+
+      vals[0] = vec[3]; vals[1] = vec[3]; vals[2] = vec[3]; vals[3] = vec[3];
+      llvm::Constant *wChannel = createConstGlobalVector(vals);
+      channelArray.push_back(xChannel);
+      channelArray.push_back(yChannel);
+      channelArray.push_back(zChannel);
+      channelArray.push_back(wChannel);
+      Constant *constChannels = ConstantArray::get(vectorChannels,
+                                                   channelArray);
+      arrayVals.push_back(constChannels);
+   }
+   Constant *constArray = ConstantArray::get(arrayType, arrayVals);
+   m_immediates->setInitializer(constArray);
+
+   m_immediatesToFlush.clear();
+}
+
+llvm::Value *StorageSoa::addrElement(int idx) const
+{
+   std::map<int, llvm::Value*>::const_iterator itr = m_addresses.find(idx);
+   if (itr == m_addresses.end()) {
+      debug_printf("Trying to access invalid shader 'address'\n");
+      return 0;
+   }
+   llvm::Value * res = (*itr).second;
+
+   res = new LoadInst(res, name("addr"), false, m_block);
+
+   return res;
+}
+
+std::vector<llvm::Value*> StorageSoa::inputElement(llvm::Value *idx)
+{
+   std::vector<llvm::Value*> res(4);
+
+   res[0] = element(m_input, idx, 0);
+   res[1] = element(m_input, idx, 1);
+   res[2] = element(m_input, idx, 2);
+   res[3] = element(m_input, idx, 3);
+
+   return res;
+}
+
+std::vector<llvm::Value*> StorageSoa::constElement(llvm::Value *idx)
+{
+   std::vector<llvm::Value*> res(4);
+   llvm::Value *xChannel, *yChannel, *zChannel, *wChannel;
+
+   xChannel = elementPointer(m_consts, idx, 0);
+   yChannel = elementPointer(m_consts, idx, 1);
+   zChannel = elementPointer(m_consts, idx, 2);
+   wChannel = elementPointer(m_consts, idx, 3);
+
+   res[0] = alignedArrayLoad(xChannel);
+   res[1] = alignedArrayLoad(yChannel);
+   res[2] = alignedArrayLoad(zChannel);
+   res[3] = alignedArrayLoad(wChannel);
+
+   return res;
+}
+
+std::vector<llvm::Value*> StorageSoa::outputElement(llvm::Value *idx)
+{
+   std::vector<llvm::Value*> res(4);
+
+   res[0] = element(m_output, idx, 0);
+   res[1] = element(m_output, idx, 1);
+   res[2] = element(m_output, idx, 2);
+   res[3] = element(m_output, idx, 3);
+
+   return res;
+}
+
+std::vector<llvm::Value*> StorageSoa::tempElement(llvm::Value *idx)
+{
+   std::vector<llvm::Value*> res(4);
+
+   res[0] = element(m_temps, idx, 0);
+   res[1] = element(m_temps, idx, 1);
+   res[2] = element(m_temps, idx, 2);
+   res[3] = element(m_temps, idx, 3);
+
+   return res;
+}
+
+std::vector<llvm::Value*> StorageSoa::immediateElement(llvm::Value *idx)
+{
+   std::vector<llvm::Value*> res(4);
+
+   res[0] = element(m_immediates, idx, 0);
+   res[1] = element(m_immediates, idx, 1);
+   res[2] = element(m_immediates, idx, 2);
+   res[3] = element(m_immediates, idx, 3);
+
+   return res;
+}
+
+llvm::Value * StorageSoa::elementPointer(llvm::Value *ptr, llvm::Value *index,
+                                         int channel) const
+{
+   std::vector<Value*> indices;
+   if (m_immediates == ptr)
+      indices.push_back(constantInt(0));
+   indices.push_back(index);
+   indices.push_back(constantInt(channel));
+
+   GetElementPtrInst *getElem = new GetElementPtrInst(ptr,
+                                                      indices.begin(),
+                                                      indices.end(),
+                                                      name("ptr"),
+                                                      m_block);
+   return getElem;
+}
+
+llvm::Value * StorageSoa::element(llvm::Value *ptr, llvm::Value *index,
+                                  int channel) const
+{
+   llvm::Value *res = elementPointer(ptr, index, channel);
+   LoadInst *load = new LoadInst(res, name("element"), false, m_block);
+   //load->setAlignment(8);
+   return load;
+}
+
+const char * StorageSoa::name(const char *prefix) const
+{
+   ++m_idx;
+   snprintf(m_name, 32, "%s%d", prefix, m_idx);
+   return m_name;
+}
+
+llvm::ConstantInt * StorageSoa::constantInt(int idx) const
+{
+   if (m_constInts.find(idx) != m_constInts.end()) {
+      return m_constInts[idx];
+   }
+   ConstantInt *constInt = ConstantInt::get(APInt(32,  idx));
+   m_constInts[idx] = constInt;
+   return constInt;
+}
+
+llvm::Value *StorageSoa::alignedArrayLoad(llvm::Value *val)
+{
+   VectorType  *vectorType = VectorType::get(Type::FloatTy, 4);
+   PointerType *vectorPtr  = PointerType::get(vectorType, 0);
+
+   CastInst *cast = new BitCastInst(val, vectorPtr, name("toVector"), m_block);
+   LoadInst *load = new LoadInst(cast, name("alignLoad"), false, m_block);
+   load->setAlignment(8);
+   return load;
+}
+
+llvm::Module * StorageSoa::currentModule() const
+{
+    if (!m_block || !m_block->getParent())
+       return 0;
+
+    return m_block->getParent()->getParent();
+}
+
+llvm::Constant * StorageSoa::createConstGlobalVector(const std::vector<float> &vec)
+{
+   VectorType *vectorType = VectorType::get(Type::FloatTy, 4);
+   std::vector<Constant*> immValues;
+   ConstantFP *constx = ConstantFP::get(Type::FloatTy, APFloat(vec[0]));
+   ConstantFP *consty = ConstantFP::get(Type::FloatTy, APFloat(vec[1]));
+   ConstantFP *constz = ConstantFP::get(Type::FloatTy, APFloat(vec[2]));
+   ConstantFP *constw = ConstantFP::get(Type::FloatTy, APFloat(vec[3]));
+   immValues.push_back(constx);
+   immValues.push_back(consty);
+   immValues.push_back(constz);
+   immValues.push_back(constw);
+   Constant  *constVector = ConstantVector::get(vectorType, immValues);
+
+   return constVector;
+}
+
+std::vector<llvm::Value*> StorageSoa::load(Argument type, int idx, int swizzle,
+                                           llvm::Value *indIdx)
+{
+   std::vector<llvm::Value*> val(4);
+
+   //if we have an indirect index, always use that
+   //   if not use the integer offset to create one
+   llvm::Value *realIndex = 0;
+   if (indIdx)
+      realIndex = indIdx;
+   else
+      realIndex = constantInt(idx);
+   debug_printf("XXXXXXXXX realIdx = %p, indIdx = %p\n", realIndex, indIdx);
+
+   switch(type) {
+   case Input:
+      val = inputElement(realIndex);
+      break;
+   case Output:
+      val = outputElement(realIndex);
+      break;
+   case Temp:
+      val = tempElement(realIndex);
+      break;
+   case Const:
+      val = constElement(realIndex);
+      break;
+   case Immediate:
+      val = immediateElement(realIndex);
+      break;
+   case Address:
+      debug_printf("Address not handled in the load phase!\n");
+      assert(0);
+      break;
+   }
+   if (!gallivm_is_swizzle(swizzle))
+      return val;
+
+   std::vector<llvm::Value*> res(4);
+
+   res[0] = val[gallivm_x_swizzle(swizzle)];
+   res[1] = val[gallivm_y_swizzle(swizzle)];
+   res[2] = val[gallivm_z_swizzle(swizzle)];
+   res[3] = val[gallivm_w_swizzle(swizzle)];
+   return res;
+}
+
+void StorageSoa::store(Argument type, int idx, const std::vector<llvm::Value*> &val,
+                       int mask)
+{
+   llvm::Value *out = 0;
+   switch(type) {
+   case Output:
+      out = m_output;
+      break;
+   case Temp:
+      out = m_temps;
+      break;
+   case Input:
+      out = m_input;
+      break;
+   case Address: {
+      llvm::Value *addr = m_addresses[idx];
+      if (!addr) {
+         addAddress(idx);
+         addr = m_addresses[idx];
+         assert(addr);
+      }
+      new StoreInst(val[0], addr, false, m_block);
+      return;
+      break;
+   }
+   default:
+      debug_printf("Can't save output of this type: %d !\n", type);
+      assert(0);
+      break;
+   }
+   llvm::Value *realIndex = constantInt(idx);
+   if ((mask & TGSI_WRITEMASK_X)) {
+      llvm::Value *xChannel = elementPointer(out, realIndex, 0);
+      new StoreInst(val[0], xChannel, false, m_block);
+   }
+   if ((mask & TGSI_WRITEMASK_Y)) {
+      llvm::Value *yChannel = elementPointer(out, realIndex, 1);
+      new StoreInst(val[1], yChannel, false, m_block);
+   }
+   if ((mask & TGSI_WRITEMASK_Z)) {
+      llvm::Value *zChannel = elementPointer(out, realIndex, 2);
+      new StoreInst(val[2], zChannel, false, m_block);
+   }
+   if ((mask & TGSI_WRITEMASK_W)) {
+      llvm::Value *wChannel = elementPointer(out, realIndex, 3);
+      new StoreInst(val[3], wChannel, false, m_block);
+   }
+}
+
+void StorageSoa::addAddress(int idx)
+{
+   GlobalVariable *val = new GlobalVariable(
+      /*Type=*/IntegerType::get(32),
+      /*isConstant=*/false,
+      /*Linkage=*/GlobalValue::ExternalLinkage,
+      /*Initializer=*/0, // has initializer, specified below
+      /*Name=*/name("address"),
+      currentModule());
+   val->setInitializer(Constant::getNullValue(IntegerType::get(32)));
+
+   debug_printf("adding to %d\n", idx);
+   m_addresses[idx] = val;
+}
diff --git a/src/gallium/aux/llvm/storagesoa.h b/src/gallium/aux/llvm/storagesoa.h
new file mode 100644
index 0000000000..6443351f27
--- /dev/null
+++ b/src/gallium/aux/llvm/storagesoa.h
@@ -0,0 +1,111 @@
+/**************************************************************************
+ *
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ **************************************************************************/
+
+#ifndef STORAGESOA_H
+#define STORAGESOA_H
+
+#include <vector>
+#include <list>
+#include <map>
+
+namespace llvm {
+   class BasicBlock;
+   class Constant;
+   class ConstantInt;
+   class GlobalVariable;
+   class LoadInst;
+   class Value;
+   class VectorType;
+   class Module;
+}
+
+class StorageSoa
+{
+public:
+   enum Argument {
+      Input,
+      Output,
+      Temp,
+      Const,
+      Immediate,
+      Address
+   };
+public:
+   StorageSoa(llvm::BasicBlock *block,
+              llvm::Value *input,
+              llvm::Value *output,
+              llvm::Value *consts,
+              llvm::Value *temps);
+
+
+   std::vector<llvm::Value*> load(Argument type, int idx, int swizzle, 
+                                  llvm::Value *indIdx =0);
+   void store(Argument type, int idx, const std::vector<llvm::Value*> &val,
+              int mask);
+
+   void addImmediate(float *vec);
+   void declareImmediates();
+
+   void addAddress(int idx);
+
+   llvm::Value  * addrElement(int idx) const;
+
+   llvm::ConstantInt *constantInt(int) const;
+private:
+   llvm::Value *elementPointer(llvm::Value *ptr, llvm::Value *indIdx,
+                               int channel) const;
+   llvm::Value *element(llvm::Value *ptr, llvm::Value *idx,
+                        int channel) const;
+   const char *name(const char *prefix) const;
+   llvm::Value  *alignedArrayLoad(llvm::Value *val);
+   llvm::Module *currentModule() const;
+   llvm::Constant  *createConstGlobalVector(const std::vector<float> &vec);
+
+   std::vector<llvm::Value*> inputElement(llvm::Value *indIdx);
+   std::vector<llvm::Value*> constElement(llvm::Value *indIdx);
+   std::vector<llvm::Value*> outputElement(llvm::Value *indIdx);
+   std::vector<llvm::Value*> tempElement(llvm::Value *indIdx);
+   std::vector<llvm::Value*> immediateElement(llvm::Value *indIdx);
+private:
+   llvm::BasicBlock *m_block;
+
+   llvm::Value *m_input;
+   llvm::Value *m_output;
+   llvm::Value *m_consts;
+   llvm::Value *m_temps;
+   llvm::GlobalVariable *m_immediates;
+
+   std::map<int, llvm::Value*> m_addresses;
+
+   std::vector<std::vector<float> > m_immediatesToFlush;
+
+   mutable std::map<int, llvm::ConstantInt*> m_constInts;
+   mutable char        m_name[32];
+   mutable int         m_idx;
+};
+
+#endif
diff --git a/src/gallium/aux/llvm/tgsitollvm.cpp b/src/gallium/aux/llvm/tgsitollvm.cpp
new file mode 100644
index 0000000000..0de595e678
--- /dev/null
+++ b/src/gallium/aux/llvm/tgsitollvm.cpp
@@ -0,0 +1,1221 @@
+#include "tgsitollvm.h"
+
+#include "gallivm.h"
+#include "gallivm_p.h"
+
+#include "storage.h"
+#include "instructions.h"
+#include "storagesoa.h"
+#include "instructionssoa.h"
+
+#include "pipe/p_shader_tokens.h"
+
+#include "pipe/tgsi/util/tgsi_parse.h"
+#include "pipe/tgsi/exec/tgsi_exec.h"
+#include "pipe/tgsi/util/tgsi_util.h"
+#include "pipe/tgsi/util/tgsi_build.h"
+#include "pipe/tgsi/util/tgsi_dump.h"
+
+
+#include <llvm/Module.h>
+#include <llvm/CallingConv.h>
+#include <llvm/Constants.h>
+#include <llvm/DerivedTypes.h>
+#include <llvm/Instructions.h>
+#include <llvm/ModuleProvider.h>
+#include <llvm/Pass.h>
+#include <llvm/PassManager.h>
+#include <llvm/ParameterAttributes.h>
+#include <llvm/Support/PatternMatch.h>
+#include <llvm/ExecutionEngine/JIT.h>
+#include <llvm/ExecutionEngine/Interpreter.h>
+#include <llvm/ExecutionEngine/GenericValue.h>
+#include <llvm/Support/MemoryBuffer.h>
+#include <llvm/LinkAllPasses.h>
+#include <llvm/Analysis/Verifier.h>
+#include <llvm/Analysis/LoopPass.h>
+#include <llvm/Target/TargetData.h>
+#include <llvm/Bitcode/ReaderWriter.h>
+#include <llvm/Transforms/Utils/Cloning.h>
+
+
+#include <sstream>
+#include <fstream>
+#include <iostream>
+
+using namespace llvm;
+
+static inline FunctionType *vertexShaderFunctionType()
+{
+   //Function takes three arguments,
+   // the calling code has to make sure the types it will
+   // pass are castable to the following:
+   // [4 x <4 x float>] inputs,
+   // [4 x <4 x float>] output,
+   // [4 x [4 x float]] consts,
+   // [4 x <4 x float>] temps
+
+   std::vector<const Type*> funcArgs;
+   VectorType *vectorType = VectorType::get(Type::FloatTy, 4);
+   ArrayType *vectorArray = ArrayType::get(vectorType, 4);
+   PointerType *vectorArrayPtr = PointerType::get(vectorArray, 0);
+
+   ArrayType   *floatArray     = ArrayType::get(Type::FloatTy, 4);
+   ArrayType   *constsArray    = ArrayType::get(floatArray, 4);
+   PointerType *constsArrayPtr = PointerType::get(constsArray, 0);
+
+   funcArgs.push_back(vectorArrayPtr);//inputs
+   funcArgs.push_back(vectorArrayPtr);//output
+   funcArgs.push_back(constsArrayPtr);//consts
+   funcArgs.push_back(vectorArrayPtr);//temps
+
+   FunctionType *functionType = FunctionType::get(
+      /*Result=*/Type::VoidTy,
+      /*Params=*/funcArgs,
+      /*isVarArg=*/false);
+
+   return functionType;
+}
+
+static inline void
+add_interpolator(struct gallivm_ir *ir,
+                 struct gallivm_interpolate *interp)
+{
+   ir->interpolators[ir->num_interp] = *interp;
+   ++ir->num_interp;
+}
+
+static void
+translate_declaration(struct gallivm_ir *prog,
+                      llvm::Module *module,
+                      Storage *storage,
+                      struct tgsi_full_declaration *decl,
+                      struct tgsi_full_declaration *fd)
+{
+   if (decl->Declaration.File == TGSI_FILE_INPUT) {
+      unsigned first, last, mask;
+      uint interp_method;
+
+      assert(decl->Declaration.Declare == TGSI_DECLARE_RANGE);
+
+      first = decl->u.DeclarationRange.First;
+      last = decl->u.DeclarationRange.Last;
+      mask = decl->Declaration.UsageMask;
+
+      /* Do not touch WPOS.xy */
+      if (first == 0) {
+         mask &= ~TGSI_WRITEMASK_XY;
+         if (mask == TGSI_WRITEMASK_NONE) {
+            first++;
+            if (first > last) {
+               return;
+            }
+         }
+      }
+
+      interp_method = decl->Interpolation.Interpolate;
+
+      if (mask == TGSI_WRITEMASK_XYZW) {
+         unsigned i, j;
+
+         for (i = first; i <= last; i++) {
+            for (j = 0; j < NUM_CHANNELS; j++) {
+               //interp( mach, i, j );
+               struct gallivm_interpolate interp;
+               interp.type = interp_method;
+               interp.attrib = i;
+               interp.chan = j;
+               add_interpolator(prog, &interp);
+            }
+         }
+      } else {
+         unsigned i, j;
+         for( j = 0; j < NUM_CHANNELS; j++ ) {
+            if( mask & (1 << j) ) {
+               for( i = first; i <= last; i++ ) {
+                  struct gallivm_interpolate interp;
+                  interp.type = interp_method;
+                  interp.attrib = i;
+                  interp.chan = j;
+                  add_interpolator(prog, &interp);
+               }
+            }
+         }
+      }
+   }
+}
+
+static void
+translate_declarationir(struct gallivm_ir *,
+                      llvm::Module *,
+                      StorageSoa *storage,
+                      struct tgsi_full_declaration *decl,
+                      struct tgsi_full_declaration *)
+{
+   if (decl->Declaration.File == TGSI_FILE_ADDRESS) {
+      int idx = decl->u.DeclarationRange.First;
+      storage->addAddress(idx);
+   }
+}
+
+static void
+translate_immediate(Storage *storage,
+                    struct tgsi_full_immediate *imm)
+{
+   float vec[4];
+   int i;
+   for (i = 0; i < imm->Immediate.Size - 1; ++i) {
+      switch (imm->Immediate.DataType) {
+      case TGSI_IMM_FLOAT32:
+         vec[i] = imm->u.ImmediateFloat32[i].Float;
+         break;
+      default:
+         assert(0);
+      }
+   }
+   storage->addImmediate(vec);
+}
+
+
+static void
+translate_immediateir(StorageSoa *storage,
+                      struct tgsi_full_immediate *imm)
+{
+   float vec[4];
+   int i;
+   for (i = 0; i < imm->Immediate.Size - 1; ++i) {
+      switch (imm->Immediate.DataType) {
+      case TGSI_IMM_FLOAT32:
+         vec[i] = imm->u.ImmediateFloat32[i].Float;
+         break;
+      default:
+         assert(0);
+      }
+   }
+   storage->addImmediate(vec);
+}
+
+static inline int
+swizzleInt(struct tgsi_full_src_register *src)
+{
+   int swizzle = 0;
+   int start = 1000;
+
+   for (int k = 0; k < 4; ++k) {
+      swizzle += tgsi_util_get_full_src_register_extswizzle(src, k) * start;
+      start /= 10;
+   }
+   return swizzle;
+}
+
+static inline llvm::Value *
+swizzleVector(llvm::Value *val, struct tgsi_full_src_register *src,
+              Storage *storage)
+{
+   int swizzle = swizzleInt(src);
+
+   if (gallivm_is_swizzle(swizzle)) {
+      /*fprintf(stderr, "XXXXXXXX swizzle = %d\n", swizzle);*/
+      val = storage->shuffleVector(val, swizzle);
+   }
+   return val;
+}
+
+static void
+translate_instruction(llvm::Module *module,
+                      Storage *storage,
+                      Instructions *instr,
+                      struct tgsi_full_instruction *inst,
+                      struct tgsi_full_instruction *fi,
+                      unsigned instno)
+{
+   llvm::Value *inputs[4];
+   inputs[0] = 0;
+   inputs[1] = 0;
+   inputs[2] = 0;
+   inputs[3] = 0;
+
+   for (int i = 0; i < inst->Instruction.NumSrcRegs; ++i) {
+      struct tgsi_full_src_register *src = &inst->FullSrcRegisters[i];
+      llvm::Value *val = 0;
+      llvm::Value *indIdx = 0;
+
+      if (src->SrcRegister.Indirect) {
+         indIdx = storage->addrElement(src->SrcRegisterInd.Index);
+         indIdx = storage->extractIndex(indIdx);
+      }
+      if (src->SrcRegister.File == TGSI_FILE_CONSTANT) {
+         val = storage->constElement(src->SrcRegister.Index, indIdx);
+      } else if (src->SrcRegister.File == TGSI_FILE_INPUT) {
+         val = storage->inputElement(src->SrcRegister.Index, indIdx);
+      } else if (src->SrcRegister.File == TGSI_FILE_TEMPORARY) {
+         val = storage->tempElement(src->SrcRegister.Index);
+      } else if (src->SrcRegister.File == TGSI_FILE_OUTPUT) {
+         val = storage->outputElement(src->SrcRegister.Index, indIdx);
+      } else if (src->SrcRegister.File == TGSI_FILE_IMMEDIATE) {
+         val = storage->immediateElement(src->SrcRegister.Index);
+      } else {
+         fprintf(stderr, "ERROR: not supported llvm source %d\n", src->SrcRegister.File);
+         return;
+      }
+
+      inputs[i] = swizzleVector(val, src, storage);
+   }
+
+   /*if (inputs[0])
+     instr->printVector(inputs[0]);
+     if (inputs[1])
+     instr->printVector(inputs[1]);*/
+   llvm::Value *out = 0;
+   switch (inst->Instruction.Opcode) {
+   case TGSI_OPCODE_ARL: {
+      out = instr->arl(inputs[0]);
+   }
+      break;
+   case TGSI_OPCODE_MOV: {
+      out = inputs[0];
+   }
+      break;
+   case TGSI_OPCODE_LIT: {
+      out = instr->lit(inputs[0]);
+   }
+      break;
+   case TGSI_OPCODE_RCP: {
+      out = instr->rcp(inputs[0]);
+   }
+      break;
+   case TGSI_OPCODE_RSQ: {
+      out = instr->rsq(inputs[0]);
+   }
+      break;
+   case TGSI_OPCODE_EXP:
+      break;
+   case TGSI_OPCODE_LOG:
+      break;
+   case TGSI_OPCODE_MUL: {
+      out = instr->mul(inputs[0], inputs[1]);
+   }
+      break;
+   case TGSI_OPCODE_ADD: {
+      out = instr->add(inputs[0], inputs[1]);
+   }
+      break;
+   case TGSI_OPCODE_DP3: {
+      out = instr->dp3(inputs[0], inputs[1]);
+   }
+      break;
+   case TGSI_OPCODE_DP4: {
+      out = instr->dp4(inputs[0], inputs[1]);
+   }
+      break;
+   case TGSI_OPCODE_DST: {
+      out = instr->dst(inputs[0], inputs[1]);
+   }
+      break;
+   case TGSI_OPCODE_MIN: {
+      out = instr->min(inputs[0], inputs[1]);
+   }
+      break;
+   case TGSI_OPCODE_MAX: {
+      out = instr->max(inputs[0], inputs[1]);
+   }
+      break;
+   case TGSI_OPCODE_SLT: {
+      out = instr->slt(inputs[0], inputs[1]);
+   }
+      break;
+   case TGSI_OPCODE_SGE: {
+      out = instr->sge(inputs[0], inputs[1]);
+   }
+      break;
+   case TGSI_OPCODE_MAD: {
+      out = instr->madd(inputs[0], inputs[1], inputs[2]);
+   }
+      break;
+   case TGSI_OPCODE_SUB: {
+      out = instr->sub(inputs[0], inputs[1]);
+   }
+      break;
+   case TGSI_OPCODE_LERP: {
+      out = instr->lerp(inputs[0], inputs[1], inputs[2]);
+   }
+      break;
+   case TGSI_OPCODE_CND:
+      break;
+   case TGSI_OPCODE_CND0:
+      break;
+   case TGSI_OPCODE_DOT2ADD:
+      break;
+   case TGSI_OPCODE_INDEX:
+      break;
+   case TGSI_OPCODE_NEGATE:
+      break;
+   case TGSI_OPCODE_FRAC: {
+      out = instr->frc(inputs[0]);
+   }
+      break;
+   case TGSI_OPCODE_CLAMP:
+      break;
+   case TGSI_OPCODE_FLOOR: {
+      out = instr->floor(inputs[0]);
+   }
+      break;
+   case TGSI_OPCODE_ROUND:
+      break;
+   case TGSI_OPCODE_EXPBASE2: {
+      out = instr->ex2(inputs[0]);
+   }
+      break;
+   case TGSI_OPCODE_LOGBASE2: {
+      out = instr->lg2(inputs[0]);
+   }
+      break;
+   case TGSI_OPCODE_POWER: {
+      out = instr->pow(inputs[0], inputs[1]);
+   }
+      break;
+   case TGSI_OPCODE_CROSSPRODUCT: {
+      out = instr->cross(inputs[0], inputs[1]);
+   }
+      break;
+   case TGSI_OPCODE_MULTIPLYMATRIX:
+      break;
+   case TGSI_OPCODE_ABS: {
+      out = instr->abs(inputs[0]);
+   }
+      break;
+   case TGSI_OPCODE_RCC:
+      break;
+   case TGSI_OPCODE_DPH: {
+      out = instr->dph(inputs[0], inputs[1]);
+   }
+      break;
+   case TGSI_OPCODE_COS: {
+      out = instr->cos(inputs[0]);
+   }
+      break;
+   case TGSI_OPCODE_DDX:
+      break;
+   case TGSI_OPCODE_DDY:
+      break;
+   case TGSI_OPCODE_KILP: {
+      out = instr->kilp(inputs[0]);
+      storage->setKilElement(out);
+      return;
+   }
+      break;
+   case TGSI_OPCODE_PK2H:
+      break;
+   case TGSI_OPCODE_PK2US:
+      break;
+   case TGSI_OPCODE_PK4B:
+      break;
+   case TGSI_OPCODE_PK4UB:
+      break;
+   case TGSI_OPCODE_RFL:
+      break;
+   case TGSI_OPCODE_SEQ:
+      break;
+   case TGSI_OPCODE_SFL:
+      break;
+   case TGSI_OPCODE_SGT: {
+      out = instr->sgt(inputs[0], inputs[1]);
+   }
+      break;
+   case TGSI_OPCODE_SIN: {
+      out = instr->sin(inputs[0]);
+   }
+      break;
+   case TGSI_OPCODE_SLE:
+      break;
+   case TGSI_OPCODE_SNE:
+      break;
+   case TGSI_OPCODE_STR:
+      break;
+   case TGSI_OPCODE_TEX:
+      break;
+   case TGSI_OPCODE_TXD:
+      break;
+   case TGSI_OPCODE_UP2H:
+      break;
+   case TGSI_OPCODE_UP2US:
+      break;
+   case TGSI_OPCODE_UP4B:
+      break;
+   case TGSI_OPCODE_UP4UB:
+      break;
+   case TGSI_OPCODE_X2D:
+      break;
+   case TGSI_OPCODE_ARA:
+      break;
+   case TGSI_OPCODE_ARR:
+      break;
+   case TGSI_OPCODE_BRA:
+      break;
+   case TGSI_OPCODE_CAL: {
+      instr->cal(inst->InstructionExtLabel.Label, storage->inputPtr());
+      return;
+   }
+      break;
+   case TGSI_OPCODE_RET: {
+      instr->end();
+      return;
+   }
+      break;
+   case TGSI_OPCODE_SSG:
+      break;
+   case TGSI_OPCODE_CMP: {
+      out = instr->cmp(inputs[0], inputs[1], inputs[2]);
+   }
+      break;
+   case TGSI_OPCODE_SCS: {
+      out = instr->scs(inputs[0]);
+   }
+      break;
+   case TGSI_OPCODE_TXB:
+      break;
+   case TGSI_OPCODE_NRM:
+      break;
+   case TGSI_OPCODE_DIV:
+      break;
+   case TGSI_OPCODE_DP2:
+      break;
+   case TGSI_OPCODE_TXL:
+      break;
+   case TGSI_OPCODE_BRK: {
+      instr->brk();
+      return;
+   }
+      break;
+   case TGSI_OPCODE_IF: {
+      instr->ifop(inputs[0]);
+      storage->setCurrentBlock(instr->currentBlock());
+      return;  //just update the state
+   }
+      break;
+   case TGSI_OPCODE_LOOP:
+      break;
+   case TGSI_OPCODE_REP:
+      break;
+   case TGSI_OPCODE_ELSE: {
+      instr->elseop();
+      storage->setCurrentBlock(instr->currentBlock());
+      return; //only state update
+   }
+      break;
+   case TGSI_OPCODE_ENDIF: {
+      instr->endif();
+      storage->setCurrentBlock(instr->currentBlock());
+      return; //just update the state
+   }
+      break;
+   case TGSI_OPCODE_ENDLOOP:
+      break;
+   case TGSI_OPCODE_ENDREP:
+      break;
+   case TGSI_OPCODE_PUSHA:
+      break;
+   case TGSI_OPCODE_POPA:
+      break;
+   case TGSI_OPCODE_CEIL:
+      break;
+   case TGSI_OPCODE_I2F:
+      break;
+   case TGSI_OPCODE_NOT:
+      break;
+   case TGSI_OPCODE_TRUNC: {
+      out = instr->trunc(inputs[0]);
+   }
+      break;
+   case TGSI_OPCODE_SHL:
+      break;
+   case TGSI_OPCODE_SHR:
+      break;
+   case TGSI_OPCODE_AND:
+      break;
+   case TGSI_OPCODE_OR:
+      break;
+   case TGSI_OPCODE_MOD:
+      break;
+   case TGSI_OPCODE_XOR:
+      break;
+   case TGSI_OPCODE_SAD:
+      break;
+   case TGSI_OPCODE_TXF:
+      break;
+   case TGSI_OPCODE_TXQ:
+      break;
+   case TGSI_OPCODE_CONT:
+      break;
+   case TGSI_OPCODE_EMIT:
+      break;
+   case TGSI_OPCODE_ENDPRIM:
+      break;
+   case TGSI_OPCODE_BGNLOOP2: {
+      instr->beginLoop();
+      storage->setCurrentBlock(instr->currentBlock());
+      return;
+   }
+      break;
+   case TGSI_OPCODE_BGNSUB: {
+      instr->bgnSub(instno);
+      storage->setCurrentBlock(instr->currentBlock());
+      storage->pushTemps();
+      return;
+   }
+      break;
+   case TGSI_OPCODE_ENDLOOP2: {
+      instr->endLoop();
+      storage->setCurrentBlock(instr->currentBlock());
+      return;
+   }
+      break;
+   case TGSI_OPCODE_ENDSUB: {
+      instr->endSub();
+      storage->setCurrentBlock(instr->currentBlock());
+      storage->popArguments();
+      storage->popTemps();
+      return;
+   }
+      break;
+   case TGSI_OPCODE_NOISE1:
+      break;
+   case TGSI_OPCODE_NOISE2:
+      break;
+   case TGSI_OPCODE_NOISE3:
+      break;
+   case TGSI_OPCODE_NOISE4:
+      break;
+   case TGSI_OPCODE_NOP:
+      break;
+   case TGSI_OPCODE_TEXBEM:
+      break;
+   case TGSI_OPCODE_TEXBEML:
+      break;
+   case TGSI_OPCODE_TEXREG2AR:
+      break;
+   case TGSI_OPCODE_TEXM3X2PAD:
+      break;
+   case TGSI_OPCODE_TEXM3X2TEX:
+      break;
+   case TGSI_OPCODE_TEXM3X3PAD:
+      break;
+   case TGSI_OPCODE_TEXM3X3TEX:
+      break;
+   case TGSI_OPCODE_TEXM3X3SPEC:
+      break;
+   case TGSI_OPCODE_TEXM3X3VSPEC:
+      break;
+   case TGSI_OPCODE_TEXREG2GB:
+      break;
+   case TGSI_OPCODE_TEXREG2RGB:
+      break;
+   case TGSI_OPCODE_TEXDP3TEX:
+      break;
+   case TGSI_OPCODE_TEXDP3:
+      break;
+   case TGSI_OPCODE_TEXM3X3:
+      break;
+   case TGSI_OPCODE_TEXM3X2DEPTH:
+      break;
+   case TGSI_OPCODE_TEXDEPTH:
+      break;
+   case TGSI_OPCODE_BEM:
+      break;
+   case TGSI_OPCODE_M4X3:
+      break;
+   case TGSI_OPCODE_M3X4:
+      break;
+   case TGSI_OPCODE_M3X3:
+      break;
+   case TGSI_OPCODE_M3X2:
+      break;
+   case TGSI_OPCODE_NRM4:
+      break;
+   case TGSI_OPCODE_CALLNZ:
+      break;
+   case TGSI_OPCODE_IFC:
+      break;
+   case TGSI_OPCODE_BREAKC:
+      break;
+   case TGSI_OPCODE_KIL:
+      break;
+   case TGSI_OPCODE_END:
+      instr->end();
+      return;
+      break;
+   default:
+      fprintf(stderr, "ERROR: Unknown opcode %d\n",
+              inst->Instruction.Opcode);
+      assert(0);
+      break;
+   }
+
+   if (!out) {
+      fprintf(stderr, "ERROR: unsupported opcode %d\n",
+              inst->Instruction.Opcode);
+      assert(!"Unsupported opcode");
+   }
+
+   /* # not sure if we need this */
+   switch( inst->Instruction.Saturate ) {
+   case TGSI_SAT_NONE:
+      break;
+   case TGSI_SAT_ZERO_ONE:
+      /*TXT( "_SAT" );*/
+      break;
+   case TGSI_SAT_MINUS_PLUS_ONE:
+      /*TXT( "_SAT[-1,1]" );*/
+      break;
+   default:
+      assert( 0 );
+   }
+
+   /* store results  */
+   for (int i = 0; i < inst->Instruction.NumDstRegs; ++i) {
+      struct tgsi_full_dst_register *dst = &inst->FullDstRegisters[i];
+
+      if (dst->DstRegister.File == TGSI_FILE_OUTPUT) {
+         storage->setOutputElement(dst->DstRegister.Index, out, dst->DstRegister.WriteMask);
+      } else if (dst->DstRegister.File == TGSI_FILE_TEMPORARY) {
+         storage->setTempElement(dst->DstRegister.Index, out, dst->DstRegister.WriteMask);
+      } else if (dst->DstRegister.File == TGSI_FILE_ADDRESS) {
+         storage->setAddrElement(dst->DstRegister.Index, out, dst->DstRegister.WriteMask);
+      } else {
+         fprintf(stderr, "ERROR: unsupported LLVM destination!");
+         assert(!"wrong destination");
+      }
+   }
+}
+
+
+static void
+translate_instructionir(llvm::Module *module,
+                        StorageSoa *storage,
+                        InstructionsSoa *instr,
+                        struct tgsi_full_instruction *inst,
+                        struct tgsi_full_instruction *fi,
+                        unsigned instno)
+{
+   std::vector< std::vector<llvm::Value*> > inputs(inst->Instruction.NumSrcRegs);
+
+   for (int i = 0; i < inst->Instruction.NumSrcRegs; ++i) {
+      struct tgsi_full_src_register *src = &inst->FullSrcRegisters[i];
+      std::vector<llvm::Value*> val;
+      llvm::Value *indIdx = 0;
+      int swizzle = swizzleInt(src);
+
+      if (src->SrcRegister.Indirect) {
+         indIdx = storage->addrElement(src->SrcRegisterInd.Index);
+      }
+      if (src->SrcRegister.File == TGSI_FILE_CONSTANT) {
+         val = storage->load(StorageSoa::Const,
+                             src->SrcRegister.Index, swizzle, indIdx);
+      } else if (src->SrcRegister.File == TGSI_FILE_INPUT) {
+         val = storage->load(StorageSoa::Input,
+                             src->SrcRegister.Index, swizzle, indIdx);
+      } else if (src->SrcRegister.File == TGSI_FILE_TEMPORARY) {
+         val = storage->load(StorageSoa::Temp,
+                             src->SrcRegister.Index, swizzle, indIdx);
+      } else if (src->SrcRegister.File == TGSI_FILE_OUTPUT) {
+         val = storage->load(StorageSoa::Output,
+                             src->SrcRegister.Index, swizzle, indIdx);
+      } else if (src->SrcRegister.File == TGSI_FILE_IMMEDIATE) {
+         val = storage->load(StorageSoa::Immediate,
+                             src->SrcRegister.Index, swizzle, indIdx);
+      } else {
+         fprintf(stderr, "ERROR: not supported llvm source %d\n", src->SrcRegister.File);
+         return;
+      }
+
+      inputs[i] = val;
+   }
+
+   std::vector<llvm::Value*> out(4);
+   switch (inst->Instruction.Opcode) {
+   case TGSI_OPCODE_ARL: {
+      out = instr->arl(inputs[0]);
+   }
+      break;
+   case TGSI_OPCODE_MOV: {
+      out = inputs[0];
+   }
+      break;
+   case TGSI_OPCODE_LIT: {
+   }
+      break;
+   case TGSI_OPCODE_RCP: {
+   }
+      break;
+   case TGSI_OPCODE_RSQ: {
+   }
+      break;
+   case TGSI_OPCODE_EXP:
+      break;
+   case TGSI_OPCODE_LOG:
+      break;
+   case TGSI_OPCODE_MUL: {
+      out = instr->mul(inputs[0], inputs[1]);
+   }
+      break;
+   case TGSI_OPCODE_ADD: {
+      out = instr->add(inputs[0], inputs[1]);
+   }
+      break;
+   case TGSI_OPCODE_DP3: {
+   }
+      break;
+   case TGSI_OPCODE_DP4: {
+   }
+      break;
+   case TGSI_OPCODE_DST: {
+   }
+      break;
+   case TGSI_OPCODE_MIN: {
+   }
+      break;
+   case TGSI_OPCODE_MAX: {
+   }
+      break;
+   case TGSI_OPCODE_SLT: {
+   }
+      break;
+   case TGSI_OPCODE_SGE: {
+   }
+      break;
+   case TGSI_OPCODE_MAD: {
+      out = instr->madd(inputs[0], inputs[1], inputs[2]);
+   }
+      break;
+   case TGSI_OPCODE_SUB: {
+   }
+      break;
+   case TGSI_OPCODE_LERP: {
+   }
+      break;
+   case TGSI_OPCODE_CND:
+      break;
+   case TGSI_OPCODE_CND0:
+      break;
+   case TGSI_OPCODE_DOT2ADD:
+      break;
+   case TGSI_OPCODE_INDEX:
+      break;
+   case TGSI_OPCODE_NEGATE:
+      break;
+   case TGSI_OPCODE_FRAC: {
+   }
+      break;
+   case TGSI_OPCODE_CLAMP:
+      break;
+   case TGSI_OPCODE_FLOOR: {
+   }
+      break;
+   case TGSI_OPCODE_ROUND:
+      break;
+   case TGSI_OPCODE_EXPBASE2: {
+   }
+      break;
+   case TGSI_OPCODE_LOGBASE2: {
+   }
+      break;
+   case TGSI_OPCODE_POWER: {
+   }
+      break;
+   case TGSI_OPCODE_CROSSPRODUCT: {
+   }
+      break;
+   case TGSI_OPCODE_MULTIPLYMATRIX:
+      break;
+   case TGSI_OPCODE_ABS: {
+   }
+      break;
+   case TGSI_OPCODE_RCC:
+      break;
+   case TGSI_OPCODE_DPH: {
+   }
+      break;
+   case TGSI_OPCODE_COS: {
+   }
+      break;
+   case TGSI_OPCODE_DDX:
+      break;
+   case TGSI_OPCODE_DDY:
+      break;
+   case TGSI_OPCODE_KILP: {
+   }
+      break;
+   case TGSI_OPCODE_PK2H:
+      break;
+   case TGSI_OPCODE_PK2US:
+      break;
+   case TGSI_OPCODE_PK4B:
+      break;
+   case TGSI_OPCODE_PK4UB:
+      break;
+   case TGSI_OPCODE_RFL:
+      break;
+   case TGSI_OPCODE_SEQ:
+      break;
+   case TGSI_OPCODE_SFL:
+      break;
+   case TGSI_OPCODE_SGT: {
+   }
+      break;
+   case TGSI_OPCODE_SIN: {
+   }
+      break;
+   case TGSI_OPCODE_SLE:
+      break;
+   case TGSI_OPCODE_SNE:
+      break;
+   case TGSI_OPCODE_STR:
+      break;
+   case TGSI_OPCODE_TEX:
+      break;
+   case TGSI_OPCODE_TXD:
+      break;
+   case TGSI_OPCODE_UP2H:
+      break;
+   case TGSI_OPCODE_UP2US:
+      break;
+   case TGSI_OPCODE_UP4B:
+      break;
+   case TGSI_OPCODE_UP4UB:
+      break;
+   case TGSI_OPCODE_X2D:
+      break;
+   case TGSI_OPCODE_ARA:
+      break;
+   case TGSI_OPCODE_ARR:
+      break;
+   case TGSI_OPCODE_BRA:
+      break;
+   case TGSI_OPCODE_CAL: {
+   }
+      break;
+   case TGSI_OPCODE_RET: {
+   }
+      break;
+   case TGSI_OPCODE_SSG:
+      break;
+   case TGSI_OPCODE_CMP: {
+   }
+      break;
+   case TGSI_OPCODE_SCS: {
+   }
+      break;
+   case TGSI_OPCODE_TXB:
+      break;
+   case TGSI_OPCODE_NRM:
+      break;
+   case TGSI_OPCODE_DIV:
+      break;
+   case TGSI_OPCODE_DP2:
+      break;
+   case TGSI_OPCODE_TXL:
+      break;
+   case TGSI_OPCODE_BRK: {
+   }
+      break;
+   case TGSI_OPCODE_IF: {
+   }
+      break;
+   case TGSI_OPCODE_LOOP:
+      break;
+   case TGSI_OPCODE_REP:
+      break;
+   case TGSI_OPCODE_ELSE: {
+   }
+      break;
+   case TGSI_OPCODE_ENDIF: {
+   }
+      break;
+   case TGSI_OPCODE_ENDLOOP:
+      break;
+   case TGSI_OPCODE_ENDREP:
+      break;
+   case TGSI_OPCODE_PUSHA:
+      break;
+   case TGSI_OPCODE_POPA:
+      break;
+   case TGSI_OPCODE_CEIL:
+      break;
+   case TGSI_OPCODE_I2F:
+      break;
+   case TGSI_OPCODE_NOT:
+      break;
+   case TGSI_OPCODE_TRUNC: {
+   }
+      break;
+   case TGSI_OPCODE_SHL:
+      break;
+   case TGSI_OPCODE_SHR:
+      break;
+   case TGSI_OPCODE_AND:
+      break;
+   case TGSI_OPCODE_OR:
+      break;
+   case TGSI_OPCODE_MOD:
+      break;
+   case TGSI_OPCODE_XOR:
+      break;
+   case TGSI_OPCODE_SAD:
+      break;
+   case TGSI_OPCODE_TXF:
+      break;
+   case TGSI_OPCODE_TXQ:
+      break;
+   case TGSI_OPCODE_CONT:
+      break;
+   case TGSI_OPCODE_EMIT:
+      break;
+   case TGSI_OPCODE_ENDPRIM:
+      break;
+   case TGSI_OPCODE_BGNLOOP2: {
+   }
+      break;
+   case TGSI_OPCODE_BGNSUB: {
+   }
+      break;
+   case TGSI_OPCODE_ENDLOOP2: {
+   }
+      break;
+   case TGSI_OPCODE_ENDSUB: {
+   }
+      break;
+   case TGSI_OPCODE_NOISE1:
+      break;
+   case TGSI_OPCODE_NOISE2:
+      break;
+   case TGSI_OPCODE_NOISE3:
+      break;
+   case TGSI_OPCODE_NOISE4:
+      break;
+   case TGSI_OPCODE_NOP:
+      break;
+   case TGSI_OPCODE_TEXBEM:
+      break;
+   case TGSI_OPCODE_TEXBEML:
+      break;
+   case TGSI_OPCODE_TEXREG2AR:
+      break;
+   case TGSI_OPCODE_TEXM3X2PAD:
+      break;
+   case TGSI_OPCODE_TEXM3X2TEX:
+      break;
+   case TGSI_OPCODE_TEXM3X3PAD:
+      break;
+   case TGSI_OPCODE_TEXM3X3TEX:
+      break;
+   case TGSI_OPCODE_TEXM3X3SPEC:
+      break;
+   case TGSI_OPCODE_TEXM3X3VSPEC:
+      break;
+   case TGSI_OPCODE_TEXREG2GB:
+      break;
+   case TGSI_OPCODE_TEXREG2RGB:
+      break;
+   case TGSI_OPCODE_TEXDP3TEX:
+      break;
+   case TGSI_OPCODE_TEXDP3:
+      break;
+   case TGSI_OPCODE_TEXM3X3:
+      break;
+   case TGSI_OPCODE_TEXM3X2DEPTH:
+      break;
+   case TGSI_OPCODE_TEXDEPTH:
+      break;
+   case TGSI_OPCODE_BEM:
+      break;
+   case TGSI_OPCODE_M4X3:
+      break;
+   case TGSI_OPCODE_M3X4:
+      break;
+   case TGSI_OPCODE_M3X3:
+      break;
+   case TGSI_OPCODE_M3X2:
+      break;
+   case TGSI_OPCODE_NRM4:
+      break;
+   case TGSI_OPCODE_CALLNZ:
+      break;
+   case TGSI_OPCODE_IFC:
+      break;
+   case TGSI_OPCODE_BREAKC:
+      break;
+   case TGSI_OPCODE_KIL:
+      break;
+   case TGSI_OPCODE_END:
+      instr->end();
+      return;
+      break;
+   default:
+      fprintf(stderr, "ERROR: Unknown opcode %d\n",
+              inst->Instruction.Opcode);
+      assert(0);
+      break;
+   }
+
+   if (!out[0]) {
+      fprintf(stderr, "ERROR: unsupported opcode %d\n",
+              inst->Instruction.Opcode);
+      assert(!"Unsupported opcode");
+   }
+
+   /* store results  */
+   for (int i = 0; i < inst->Instruction.NumDstRegs; ++i) {
+      struct tgsi_full_dst_register *dst = &inst->FullDstRegisters[i];
+
+      if (dst->DstRegister.File == TGSI_FILE_OUTPUT) {
+         storage->store(StorageSoa::Output,
+                        dst->DstRegister.Index, out, dst->DstRegister.WriteMask);
+      } else if (dst->DstRegister.File == TGSI_FILE_TEMPORARY) {
+         storage->store(StorageSoa::Temp,
+                        dst->DstRegister.Index, out, dst->DstRegister.WriteMask);
+      } else if (dst->DstRegister.File == TGSI_FILE_ADDRESS) {
+         storage->store(StorageSoa::Address,
+                        dst->DstRegister.Index, out, dst->DstRegister.WriteMask);
+      } else {
+         fprintf(stderr, "ERROR: unsupported LLVM destination!");
+         assert(!"wrong destination");
+      }
+   }
+}
+
+llvm::Module *
+tgsi_to_llvm(struct gallivm_ir *ir, const struct tgsi_token *tokens)
+{
+   llvm::Module *mod = new Module("shader");
+   struct tgsi_parse_context parse;
+   struct tgsi_full_instruction fi;
+   struct tgsi_full_declaration fd;
+   unsigned instno = 0;
+   Function* shader = mod->getFunction("execute_shader");
+   std::ostringstream stream;
+   if (ir->type == GALLIVM_VS) {
+      stream << "vs_shader";
+   } else {
+      stream << "fs_shader";
+   }
+   stream << ir->id;
+   std::string func_name = stream.str();
+   shader->setName(func_name.c_str());
+
+   Function::arg_iterator args = shader->arg_begin();
+   Value *ptr_INPUT = args++;
+   ptr_INPUT->setName("input");
+
+   BasicBlock *label_entry = new BasicBlock("entry", shader, 0);
+
+   tgsi_parse_init(&parse, tokens);
+
+   fi = tgsi_default_full_instruction();
+   fd = tgsi_default_full_declaration();
+   Storage storage(label_entry, ptr_INPUT);
+   Instructions instr(mod, shader, label_entry, &storage);
+   while(!tgsi_parse_end_of_tokens(&parse)) {
+      tgsi_parse_token(&parse);
+
+      switch (parse.FullToken.Token.Type) {
+      case TGSI_TOKEN_TYPE_DECLARATION:
+         translate_declaration(ir, mod, &storage,
+                               &parse.FullToken.FullDeclaration,
+                               &fd);
+         break;
+
+      case TGSI_TOKEN_TYPE_IMMEDIATE:
+         translate_immediate(&storage,
+                             &parse.FullToken.FullImmediate);
+         break;
+
+      case TGSI_TOKEN_TYPE_INSTRUCTION:
+         translate_instruction(mod, &storage, &instr,
+                               &parse.FullToken.FullInstruction,
+                               &fi, instno);
+         ++instno;
+         break;
+
+      default:
+         assert(0);
+      }
+   }
+
+   tgsi_parse_free(&parse);
+
+   ir->num_consts = storage.numConsts();
+   return mod;
+}
+
+llvm::Module * tgsi_to_llvmir(struct gallivm_ir *ir,
+                              const struct tgsi_token *tokens)
+{
+   llvm::Module *mod = new Module("shader");
+   struct tgsi_parse_context parse;
+   struct tgsi_full_instruction fi;
+   struct tgsi_full_declaration fd;
+   unsigned instno = 0;
+   std::ostringstream stream;
+   if (ir->type == GALLIVM_VS) {
+      stream << "vs_shader";
+   } else {
+      stream << "fs_shader";
+   }
+   //stream << ir->id;
+   std::string func_name = stream.str();
+   Function *shader = llvm::cast<Function>(mod->getOrInsertFunction(
+                                              func_name.c_str(),
+                                              vertexShaderFunctionType()));
+
+   Function::arg_iterator args = shader->arg_begin();
+   Value *input = args++;
+   input->setName("inputs");
+   Value *output = args++;
+   output->setName("outputs");
+   Value *consts = args++;
+   consts->setName("consts");
+   Value *temps = args++;
+   temps->setName("temps");
+
+   BasicBlock *label_entry = new BasicBlock("entry", shader, 0);
+
+   tgsi_parse_init(&parse, tokens);
+
+   fi = tgsi_default_full_instruction();
+   fd = tgsi_default_full_declaration();
+
+   StorageSoa storage(label_entry, input, output, consts, temps);
+   InstructionsSoa instr(mod, shader, label_entry, &storage);
+
+   while(!tgsi_parse_end_of_tokens(&parse)) {
+      tgsi_parse_token(&parse);
+
+      switch (parse.FullToken.Token.Type) {
+      case TGSI_TOKEN_TYPE_DECLARATION:
+         translate_declarationir(ir, mod, &storage,
+                                 &parse.FullToken.FullDeclaration,
+                                 &fd);
+         break;
+
+      case TGSI_TOKEN_TYPE_IMMEDIATE:
+         translate_immediateir(&storage,
+                             &parse.FullToken.FullImmediate);
+         break;
+
+      case TGSI_TOKEN_TYPE_INSTRUCTION:
+         storage.declareImmediates();
+         translate_instructionir(mod, &storage, &instr,
+                                 &parse.FullToken.FullInstruction,
+                                 &fi, instno);
+         ++instno;
+         break;
+
+      default:
+         assert(0);
+      }
+   }
+
+   tgsi_parse_free(&parse);
+
+   return mod;
+}
diff --git a/src/gallium/aux/llvm/tgsitollvm.h b/src/gallium/aux/llvm/tgsitollvm.h
new file mode 100644
index 0000000000..7ada04d629
--- /dev/null
+++ b/src/gallium/aux/llvm/tgsitollvm.h
@@ -0,0 +1,20 @@
+#ifndef TGSITOLLVM_H
+#define TGSITOLLVM_H
+
+
+namespace llvm {
+   class Module;
+}
+
+struct gallivm_ir;
+struct tgsi_token;
+
+
+llvm::Module * tgsi_to_llvm(struct gallivm_ir *ir,
+                            const struct tgsi_token *tokens);
+
+
+llvm::Module * tgsi_to_llvmir(struct gallivm_ir *ir,
+                              const struct tgsi_token *tokens);
+
+#endif
diff --git a/src/gallium/aux/pipebuffer/Makefile b/src/gallium/aux/pipebuffer/Makefile
new file mode 100644
index 0000000000..75764a9a18
--- /dev/null
+++ b/src/gallium/aux/pipebuffer/Makefile
@@ -0,0 +1,23 @@
+
+TOP = ../../../..
+include $(TOP)/configs/current
+
+LIBNAME = pipebuffer
+
+DRIVER_SOURCES = \
+	pb_buffer_fenced.c \
+	pb_buffer_malloc.c \
+	pb_bufmgr_fenced.c \
+	pb_bufmgr_mm.c \
+	pb_bufmgr_pool.c \
+	pb_winsys.c
+
+C_SOURCES = \
+	$(DRIVER_SOURCES)
+
+ASM_SOURCES = 
+
+include ../Makefile.template
+
+symlinks:
+
diff --git a/src/gallium/aux/pipebuffer/linked_list.h b/src/gallium/aux/pipebuffer/linked_list.h
new file mode 100644
index 0000000000..e99817fb13
--- /dev/null
+++ b/src/gallium/aux/pipebuffer/linked_list.h
@@ -0,0 +1,91 @@
+/**************************************************************************
+ * 
+ * Copyright 2006 Tungsten Graphics, Inc., Bismarck, ND. USA.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDERS, AUTHORS AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, 
+ * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR 
+ * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE 
+ * USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ **************************************************************************/
+
+/**
+ * \file
+ * List macros heavily inspired by the Linux kernel
+ * list handling. No list looping yet.
+ * 
+ * Is not threadsafe, so common operations need to
+ * be protected using an external mutex.
+ */
+
+#ifndef LINKED_LIST_H_
+#define LINKED_LIST_H_
+
+
+#include <stddef.h>
+
+
+struct list_head
+{
+    struct list_head *prev;
+    struct list_head *next;
+};
+
+
+#define LIST_INITHEAD(__item)			\
+  do {						\
+    (__item)->prev = (__item);			\
+    (__item)->next = (__item);			\
+  } while (0)
+
+#define LIST_ADD(__item, __list)		\
+  do {						\
+    (__item)->prev = (__list);			\
+    (__item)->next = (__list)->next;		\
+    (__list)->next->prev = (__item);		\
+    (__list)->next = (__item);			\
+  } while (0)
+
+#define LIST_ADDTAIL(__item, __list)		\
+  do {						\
+    (__item)->next = (__list);			\
+    (__item)->prev = (__list)->prev;		\
+    (__list)->prev->next = (__item);		\
+    (__list)->prev = (__item);			\
+  } while(0)
+
+#define LIST_DEL(__item)			\
+  do {						\
+    (__item)->prev->next = (__item)->next;	\
+    (__item)->next->prev = (__item)->prev;	\
+  } while(0)
+
+#define LIST_DELINIT(__item)			\
+  do {						\
+    (__item)->prev->next = (__item)->next;	\
+    (__item)->next->prev = (__item)->prev;	\
+    (__item)->next = (__item);			\
+    (__item)->prev = (__item);			\
+  } while(0)
+
+#define LIST_ENTRY(__type, __item, __field)   \
+    ((__type *)(((char *)(__item)) - offsetof(__type, __field)))
+
+
+#endif /*LINKED_LIST_H_*/
diff --git a/src/gallium/aux/pipebuffer/pb_buffer.h b/src/gallium/aux/pipebuffer/pb_buffer.h
new file mode 100644
index 0000000000..97beb5f72a
--- /dev/null
+++ b/src/gallium/aux/pipebuffer/pb_buffer.h
@@ -0,0 +1,202 @@
+/**************************************************************************
+ *
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ **************************************************************************/
+
+/**
+ * \file
+ * Generic code for buffers.
+ * 
+ * Behind a pipe buffle handle there can be DMA buffers, client (or user) 
+ * buffers, regular malloced buffers, etc. This file provides an abstract base 
+ * buffer handle that allows the driver to cope with all those kinds of buffers 
+ * in a more flexible way.
+ * 
+ * There is no obligation of a winsys driver to use this library. And a pipe
+ * driver should be completly agnostic about it.
+ * 
+ * \author José Fonseca <jrfonseca@tungstengraphics.com>
+ */
+
+#ifndef PB_BUFFER_H_
+#define PB_BUFFER_H_
+
+
+#include "pipe/p_compiler.h"
+#include "pipe/p_debug.h"
+#include "pipe/p_state.h"
+#include "pipe/p_inlines.h"
+
+
+struct pb_vtbl;
+
+/**
+ * Buffer description.
+ * 
+ * Used when allocating the buffer.
+ */
+struct pb_desc
+{
+   unsigned alignment;
+   unsigned usage;
+};
+
+
+/**
+ * Base class for all pb_* buffers.
+ */
+struct pb_buffer 
+{
+   struct pipe_buffer base;
+
+   /**
+    * Pointer to the virtual function table.
+    *
+    * Avoid accessing this table directly. Use the inline functions below 
+    * instead to avoid mistakes. 
+    */
+   const struct pb_vtbl *vtbl;
+};
+
+
+/**
+ * Virtual function table for the buffer storage operations.
+ * 
+ * Note that creation is not done through this table.
+ */
+struct pb_vtbl
+{
+   void (*destroy)( struct pb_buffer *buf );
+
+   /** 
+    * Map the entire data store of a buffer object into the client's address.
+    * flags is bitmask of PIPE_BUFFER_FLAG_READ/WRITE. 
+    */
+   void *(*map)( struct pb_buffer *buf, 
+                 unsigned flags );
+   
+   void (*unmap)( struct pb_buffer *buf );
+
+   /**
+    * Get the base buffer and the offset.
+    * 
+    * A buffer can be subdivided in smaller buffers. This method should return
+    * the underlaying buffer, and the relative offset.
+    * 
+    * Buffers without an underlaying base buffer should return themselves, with 
+    * a zero offset.
+    * 
+    * Note that this will increase the reference count of the base buffer.
+    */
+   void (*get_base_buffer)( struct pb_buffer *buf,
+                            struct pb_buffer **base_buf,
+                            unsigned *offset );
+};
+
+
+static INLINE struct pipe_buffer *
+pb_pipe_buffer( struct pb_buffer *pbuf )
+{
+   assert(pbuf);
+   return &pbuf->base;
+}
+
+
+static INLINE struct pb_buffer *
+pb_buffer( struct pipe_buffer *buf )
+{
+   assert(buf);
+   /* Could add a magic cookie check on debug builds.
+    */
+   return (struct pb_buffer *)buf;
+}
+
+
+/* Accessor functions for pb->vtbl:
+ */
+static INLINE void *
+pb_map(struct pb_buffer *buf, 
+       unsigned flags)
+{
+   assert(buf);
+   return buf->vtbl->map(buf, flags);
+}
+
+
+static INLINE void 
+pb_unmap(struct pb_buffer *buf)
+{
+   assert(buf);
+   buf->vtbl->unmap(buf);
+}
+
+
+static INLINE void
+pb_get_base_buffer( struct pb_buffer *buf,
+		    struct pb_buffer **base_buf,
+		    unsigned *offset )
+{
+   buf->vtbl->get_base_buffer(buf, base_buf, offset);
+}
+
+
+static INLINE void 
+pb_destroy(struct pb_buffer *buf)
+{
+   assert(buf);
+   buf->vtbl->destroy(buf);
+}
+
+
+/* XXX: thread safety issues!
+ */
+static INLINE void
+pb_reference(struct pb_buffer **dst,
+             struct pb_buffer *src)
+{
+   if (src) 
+      src->base.refcount++;
+
+   if (*dst && --(*dst)->base.refcount == 0)
+      pb_destroy( *dst );
+
+   *dst = src;
+}
+
+
+/**
+ * Malloc-based buffer to store data that can't be used by the graphics 
+ * hardware.
+ */
+struct pb_buffer *
+pb_malloc_buffer_create(size_t size, 
+                        const struct pb_desc *desc);
+
+
+void 
+pb_init_winsys(struct pipe_winsys *winsys);
+
+
+#endif /*PB_BUFFER_H_*/
diff --git a/src/gallium/aux/pipebuffer/pb_buffer_fenced.c b/src/gallium/aux/pipebuffer/pb_buffer_fenced.c
new file mode 100644
index 0000000000..f4fc3f6d71
--- /dev/null
+++ b/src/gallium/aux/pipebuffer/pb_buffer_fenced.c
@@ -0,0 +1,299 @@
+/**************************************************************************
+ *
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ **************************************************************************/
+
+/**
+ * \file
+ * Implementation of fenced buffers.
+ * 
+ * \author José Fonseca <jrfonseca-at-tungstengraphics-dot-com>
+ * \author Thomas Hellström <thomas-at-tungstengraphics-dot-com>
+ */
+
+
+#include "linked_list.h"
+
+#include "p_compiler.h"
+#include "p_debug.h"
+#include "p_winsys.h"
+#include "p_thread.h"
+#include "p_util.h"
+
+#include "pb_buffer.h"
+#include "pb_buffer_fenced.h"
+
+#ifndef __MSC__
+#include <unistd.h>
+#endif
+
+
+/**
+ * Convenience macro (type safe).
+ */
+#define SUPER(__derived) (&(__derived)->base)
+
+
+struct fenced_buffer_list
+{
+   _glthread_Mutex mutex;
+   
+   struct pipe_winsys *winsys;
+   
+   size_t numDelayed;
+   size_t checkDelayed;
+   
+   struct list_head delayed;
+};
+
+
+/**
+ * Wrapper around a pipe buffer which adds fencing and reference counting.
+ */
+struct fenced_buffer
+{
+   struct pb_buffer base;
+   
+   struct pb_buffer *buffer;
+
+   struct pipe_fence_handle *fence;
+
+   struct list_head head;
+   struct fenced_buffer_list *list;
+};
+
+
+static INLINE struct fenced_buffer *
+fenced_buffer(struct pb_buffer *buf)
+{
+   assert(buf);
+   assert(buf->vtbl == &fenced_buffer_vtbl);
+   return (struct fenced_buffer *)buf;
+}
+
+
+
+
+static void
+_fenced_buffer_list_check_free(struct fenced_buffer_list *fenced_list, 
+                               int wait)
+{
+   struct pipe_winsys *winsys = fenced_list->winsys;
+   struct fenced_buffer *fenced_buf;   
+   struct list_head *list, *prev;
+   int signaled = -1;
+
+   list = fenced_list->delayed.next;
+
+   if (fenced_list->numDelayed > 3) {
+      unsigned i;
+
+      for (i = 0; i < fenced_list->numDelayed; i += 3) {
+         list = list->next;
+      }
+   }
+
+   prev = list->prev;
+   for (; list != &fenced_list->delayed; list = prev, prev = list->prev) {
+
+      fenced_buf = LIST_ENTRY(struct fenced_buffer, list, head);
+
+      if (signaled != 0) {
+         if (wait) {
+            signaled = winsys->fence_finish(winsys, fenced_buf->fence, 0);
+         }
+         else {
+            signaled = winsys->fence_signalled(winsys, fenced_buf->fence, 0);
+         }
+      }
+
+      if (signaled != 0)
+	 /* XXX: we are assuming that buffers are freed in the same order they 
+	  * are fenced which may not always be true... 
+	  */
+         break;
+
+      winsys->fence_reference(winsys, &fenced_buf->fence, NULL);
+      
+      LIST_DEL(list);
+      fenced_list->numDelayed--;
+
+      /* Do the delayed destroy:
+       */
+      pb_reference(&fenced_buf->buffer, NULL);
+      FREE(fenced_buf);
+   }
+}
+
+
+static void
+fenced_buffer_destroy(struct pb_buffer *buf)
+{
+   struct fenced_buffer *fenced_buf = fenced_buffer(buf);   
+   struct fenced_buffer_list *fenced_list = fenced_buf->list;
+
+   if (fenced_buf->fence) {
+      LIST_ADDTAIL(&fenced_buf->head, &fenced_list->delayed);
+      fenced_list->numDelayed++;
+   }
+   else {
+      pb_reference(&fenced_buf->buffer, NULL);
+      FREE(fenced_buf);
+   }
+   
+   if ((fenced_list->numDelayed % fenced_list->checkDelayed) == 0)
+      _fenced_buffer_list_check_free(fenced_list, 0);
+}
+
+
+static void *
+fenced_buffer_map(struct pb_buffer *buf, 
+                  unsigned flags)
+{
+   struct fenced_buffer *fenced_buf = fenced_buffer(buf);   
+   return pb_map(fenced_buf->buffer, flags);
+}
+
+
+static void
+fenced_buffer_unmap(struct pb_buffer *buf)
+{
+   struct fenced_buffer *fenced_buf = fenced_buffer(buf);   
+   pb_unmap(fenced_buf->buffer);
+}
+
+
+static void
+fenced_buffer_get_base_buffer(struct pb_buffer *buf,
+                              struct pb_buffer **base_buf,
+                              unsigned *offset)
+{
+   struct fenced_buffer *fenced_buf = fenced_buffer(buf);
+   pb_get_base_buffer(fenced_buf->buffer, base_buf, offset);
+}
+
+
+const struct pb_vtbl 
+fenced_buffer_vtbl = {
+      fenced_buffer_destroy,
+      fenced_buffer_map,
+      fenced_buffer_unmap,
+      fenced_buffer_get_base_buffer
+};
+
+
+struct pb_buffer *
+fenced_buffer_create(struct fenced_buffer_list *fenced_list, 
+                     struct pb_buffer *buffer)
+{
+   struct fenced_buffer *buf;
+   
+   if(!buffer)
+      return NULL;
+   
+   buf = CALLOC_STRUCT(fenced_buffer);
+   if(!buf)
+      return NULL;
+   
+   buf->base.base.refcount = 1;
+   buf->base.base.alignment = buffer->base.alignment;
+   buf->base.base.usage = buffer->base.usage;
+   buf->base.base.size = buffer->base.size;
+   
+   buf->base.vtbl = &fenced_buffer_vtbl;
+   buf->buffer = buffer;
+   buf->list = fenced_list;
+   
+   return &buf->base;
+}
+
+
+void
+buffer_fence(struct pb_buffer *buf,
+             struct pipe_fence_handle *fence)
+{
+   struct fenced_buffer *fenced_buf = fenced_buffer(buf);
+   struct fenced_buffer_list *fenced_list = fenced_buf->list;
+   struct pipe_winsys *winsys = fenced_list->winsys;
+   
+   _glthread_LOCK_MUTEX(fenced_list->mutex);
+   winsys->fence_reference(winsys, &fenced_buf->fence, fence);
+   _glthread_UNLOCK_MUTEX(fenced_list->mutex);
+}
+
+
+struct fenced_buffer_list *
+fenced_buffer_list_create(struct pipe_winsys *winsys) 
+{
+   struct fenced_buffer_list *fenced_list;
+
+   fenced_list = (struct fenced_buffer_list *)CALLOC(1, sizeof(*fenced_list));
+   if (!fenced_list)
+      return NULL;
+
+   fenced_list->winsys = winsys;
+
+   LIST_INITHEAD(&fenced_list->delayed);
+
+   fenced_list->numDelayed = 0;
+   
+   /* TODO: don't hard code this */ 
+   fenced_list->checkDelayed = 5;
+
+   _glthread_INIT_MUTEX(fenced_list->mutex);
+
+   return fenced_list;
+}
+
+
+void
+fenced_buffer_list_check_free(struct fenced_buffer_list *fenced_list, 
+                              int wait)
+{
+   _glthread_LOCK_MUTEX(fenced_list->mutex);
+   _fenced_buffer_list_check_free(fenced_list, wait);
+   _glthread_UNLOCK_MUTEX(fenced_list->mutex);
+}
+
+
+void
+fenced_buffer_list_destroy(struct fenced_buffer_list *fenced_list)
+{
+   _glthread_LOCK_MUTEX(fenced_list->mutex);
+
+   /* Wait on outstanding fences */
+   while (fenced_list->numDelayed) {
+      _glthread_UNLOCK_MUTEX(fenced_list->mutex);
+      sched_yield();
+      _fenced_buffer_list_check_free(fenced_list, 1);
+      _glthread_LOCK_MUTEX(fenced_list->mutex);
+   }
+
+   _glthread_UNLOCK_MUTEX(fenced_list->mutex);
+   
+   FREE(fenced_list);
+}
+
+
diff --git a/src/gallium/aux/pipebuffer/pb_buffer_fenced.h b/src/gallium/aux/pipebuffer/pb_buffer_fenced.h
new file mode 100644
index 0000000000..c40b9c75e1
--- /dev/null
+++ b/src/gallium/aux/pipebuffer/pb_buffer_fenced.h
@@ -0,0 +1,117 @@
+/**************************************************************************
+ *
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ **************************************************************************/
+
+/**
+ * \file
+ * Buffer fencing.
+ * 
+ * "Fenced buffers" is actually a misnomer. They should be referred as 
+ * "fenceable buffers", i.e, buffers that can be fenced, but I couldn't find
+ * the word "fenceable" in the dictionary.
+ * 
+ * A "fenced buffer" is a decorator around a normal buffer, which adds two 
+ * special properties:
+ * - the ability for the destruction to be delayed by a fence;
+ * - reference counting.
+ * 
+ * Usually DMA buffers have a life-time that will extend the life-time of its 
+ * handle. The end-of-life is dictated by the fence signalling. 
+ * 
+ * Between the handle's destruction, and the fence signalling, the buffer is 
+ * stored in a fenced buffer list.
+ * 
+ * \author José Fonseca <jrfonseca@tungstengraphics.com>
+ */
+
+#ifndef PB_BUFFER_FENCED_H_
+#define PB_BUFFER_FENCED_H_
+
+
+#include "pipe/p_debug.h"
+
+
+struct pipe_winsys;
+struct pipe_buffer;
+struct pipe_fence_handle;
+
+
+/**
+ * List of buffers which are awaiting fence signalling.
+ */
+struct fenced_buffer_list;
+
+
+/**
+ * The fenced buffer's virtual function table.
+ * 
+ * NOTE: Made public for debugging purposes.
+ */
+extern const struct pb_vtbl fenced_buffer_vtbl;
+
+
+/**
+ * Create a fenced buffer list.
+ * 
+ * See also fenced_bufmgr_create for a more convenient way to use this.
+ */
+struct fenced_buffer_list *
+fenced_buffer_list_create(struct pipe_winsys *winsys);
+
+
+/**
+ * Walk the fenced buffer list to check and free signalled buffers.
+ */ 
+void
+fenced_buffer_list_check_free(struct fenced_buffer_list *fenced_list, 
+                              int wait);
+
+void
+fenced_buffer_list_destroy(struct fenced_buffer_list *fenced_list);
+
+
+/**
+ * Wrap a buffer in a fenced buffer.
+ * 
+ * NOTE: this will not increase the buffer reference count.
+ */
+struct pb_buffer *
+fenced_buffer_create(struct fenced_buffer_list *fenced, 
+                     struct pb_buffer *buffer);
+
+
+/**
+ * Set a buffer's fence.
+ * 
+ * NOTE: Although it takes a generic pb_buffer argument, it will fail
+ * on everything but buffers returned by fenced_buffer_create.
+ */
+void
+buffer_fence(struct pb_buffer *buf,
+             struct pipe_fence_handle *fence);
+
+
+#endif /*PB_BUFFER_FENCED_H_*/
diff --git a/src/gallium/aux/pipebuffer/pb_buffer_malloc.c b/src/gallium/aux/pipebuffer/pb_buffer_malloc.c
new file mode 100644
index 0000000000..9e8244f909
--- /dev/null
+++ b/src/gallium/aux/pipebuffer/pb_buffer_malloc.c
@@ -0,0 +1,127 @@
+/**************************************************************************
+ *
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ **************************************************************************/
+
+/**
+ * \file
+ * Implementation of malloc-based buffers to store data that can't be processed
+ * by the hardware. 
+ * 
+ * \author José Fonseca <jrfonseca@tungstengraphics.com>
+ */
+
+
+#include "pipe/p_debug.h"
+#include "pipe/p_util.h"
+#include "pb_buffer.h"
+
+
+struct malloc_buffer 
+{
+   struct pb_buffer base;
+   void *data;
+};
+
+
+extern const struct pb_vtbl malloc_buffer_vtbl;
+
+static INLINE struct malloc_buffer *
+malloc_buffer(struct pb_buffer *buf)
+{
+   assert(buf);
+   assert(buf->vtbl == &malloc_buffer_vtbl);
+   return (struct malloc_buffer *)buf;
+}
+
+
+static void
+malloc_buffer_destroy(struct pb_buffer *buf)
+{
+   align_free(malloc_buffer(buf)->data);
+   FREE(buf);
+}
+
+
+static void *
+malloc_buffer_map(struct pb_buffer *buf, 
+                  unsigned flags)
+{
+   return malloc_buffer(buf)->data;
+}
+
+
+static void
+malloc_buffer_unmap(struct pb_buffer *buf)
+{
+   /* No-op */
+}
+
+
+static void
+malloc_buffer_get_base_buffer(struct pb_buffer *buf,
+                              struct pb_buffer **base_buf,
+                              unsigned *offset)
+{
+   *base_buf = buf;
+   *offset = 0;
+}
+
+
+const struct pb_vtbl 
+malloc_buffer_vtbl = {
+      malloc_buffer_destroy,
+      malloc_buffer_map,
+      malloc_buffer_unmap,
+      malloc_buffer_get_base_buffer
+};
+
+
+struct pb_buffer *
+pb_malloc_buffer_create(size_t size,
+                   	const struct pb_desc *desc) 
+{
+   struct malloc_buffer *buf;
+   
+   /* TODO: do a single allocation */
+   
+   buf = CALLOC_STRUCT(malloc_buffer);
+   if(!buf)
+      return NULL;
+
+   buf->base.base.refcount = 1;
+   buf->base.base.alignment = desc->alignment;
+   buf->base.base.usage = desc->usage;
+   buf->base.base.size = size;
+   buf->base.vtbl = &malloc_buffer_vtbl;
+
+   buf->data = align_malloc(size, desc->alignment < sizeof(void*) ? sizeof(void*) : desc->alignment);
+   if(!buf->data) {
+      align_free(buf);
+      return NULL;
+   }
+
+   return &buf->base;
+}
diff --git a/src/gallium/aux/pipebuffer/pb_bufmgr.h b/src/gallium/aux/pipebuffer/pb_bufmgr.h
new file mode 100644
index 0000000000..1ddf784c97
--- /dev/null
+++ b/src/gallium/aux/pipebuffer/pb_bufmgr.h
@@ -0,0 +1,126 @@
+/**************************************************************************
+ *
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ **************************************************************************/
+
+/**
+ * \file
+ * Buffer management.
+ * 
+ * A buffer manager does only one basic thing: it creates buffers. Actually,
+ * "buffer factory" would probably a more accurate description.
+ * 
+ * You can chain buffer managers so that you can have a finer grained memory
+ * management and pooling.
+ * 
+ * For example, for a simple batch buffer manager you would chain:
+ * - the native buffer manager, which provides DMA memory from the graphics
+ * memory space;
+ * - the pool buffer manager, which keep around a pool of equally sized buffers
+ * to avoid latency associated with the native buffer manager; 
+ * - the fenced buffer manager, which will delay buffer destruction until the 
+ * the moment the card finishing processing it. 
+ * 
+ * \author José Fonseca <jrfonseca@tungstengraphics.com>
+ */
+
+#ifndef PB_BUFMGR_H_
+#define PB_BUFMGR_H_
+
+
+#include <stddef.h>
+
+
+struct pb_desc;
+struct pipe_buffer;
+struct pipe_winsys;
+
+
+/** 
+ * Abstract base class for all buffer managers.
+ */
+struct pb_manager
+{
+   /* XXX: we will likely need more allocation flags */
+   struct pb_buffer *
+   (*create_buffer)( struct pb_manager *mgr, 
+	             size_t size,
+	             const struct pb_desc *desc);
+
+   void
+   (*destroy)( struct pb_manager *mgr );
+};
+
+
+/** 
+ * Static buffer pool manager.
+ * 
+ * Manages the allocation of equally sized buffers. It does so by allocating
+ * a single big buffer and divide it equally sized buffers. 
+ * 
+ * It is meant to manage the allocation of batch buffer pools.
+ */
+struct pb_manager *
+pool_bufmgr_create(struct pb_manager *provider, 
+                   size_t n, size_t size,
+                   const struct pb_desc *desc);
+
+
+/** 
+ * Wraper around the old memory manager.
+ * 
+ * It managers buffers of different sizes. It does so by allocating a buffer
+ * with the size of the heap, and then using the old mm memory manager to manage
+ * that heap. 
+ */
+struct pb_manager *
+mm_bufmgr_create(struct pb_manager *provider, 
+                 size_t size, size_t align2);
+
+/**
+ * Same as mm_bufmgr_create.
+ * 
+ * Buffer will be release when the manager is destroyed.
+ */
+struct pb_manager *
+mm_bufmgr_create_from_buffer(struct pb_buffer *buffer, 
+                             size_t size, size_t align2);
+
+
+/** 
+ * Fenced buffer manager.
+ *
+ * This manager is just meant for convenience. It wraps the buffers returned
+ * by another manager in fenced buffers, so that  
+ * 
+ * NOTE: the buffer manager that provides the buffers will be destroyed
+ * at the same time.
+ */
+struct pb_manager *
+fenced_bufmgr_create(struct pb_manager *provider,
+                     struct pipe_winsys *winsys);
+
+
+#endif /*PB_BUFMGR_H_*/
diff --git a/src/gallium/aux/pipebuffer/pb_bufmgr_fenced.c b/src/gallium/aux/pipebuffer/pb_bufmgr_fenced.c
new file mode 100644
index 0000000000..c535d3276c
--- /dev/null
+++ b/src/gallium/aux/pipebuffer/pb_bufmgr_fenced.c
@@ -0,0 +1,131 @@
+/**************************************************************************
+ * 
+ * Copyright 2006 Tungsten Graphics, Inc., Bismarck, ND., USA
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDERS, AUTHORS AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM,
+ * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR 
+ * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE 
+ * USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * 
+ **************************************************************************/
+
+/**
+ * \file
+ * A buffer manager that wraps buffers in fenced buffers.
+ * 
+ * \author José Fonseca <jrfonseca@tungstengraphics.dot.com>
+ */
+
+
+#include "p_debug.h"
+#include "p_util.h"
+
+#include "pb_buffer.h"
+#include "pb_buffer_fenced.h"
+#include "pb_bufmgr.h"
+
+
+struct fenced_pb_manager
+{
+   struct pb_manager base;
+
+   struct pb_manager *provider;
+   
+   struct fenced_buffer_list *fenced_list;
+};
+
+
+static INLINE struct fenced_pb_manager *
+fenced_pb_manager(struct pb_manager *mgr)
+{
+   assert(mgr);
+   return (struct fenced_pb_manager *)mgr;
+}
+
+
+static struct pb_buffer *
+fenced_bufmgr_create_buffer(struct pb_manager *mgr, 
+                            size_t size,
+                            const struct pb_desc *desc)
+{
+   struct fenced_pb_manager *fenced_mgr = fenced_pb_manager(mgr);
+   struct pb_buffer *buf;
+   struct pb_buffer *fenced_buf;
+
+   /* check for free buffers before allocating new ones */
+   fenced_buffer_list_check_free(fenced_mgr->fenced_list, 0);
+   
+   buf = fenced_mgr->provider->create_buffer(fenced_mgr->provider, size, desc);
+   if(!buf) {
+      /* try harder to get a buffer */
+      fenced_buffer_list_check_free(fenced_mgr->fenced_list, 1);
+      
+      buf = fenced_mgr->provider->create_buffer(fenced_mgr->provider, size, desc);
+      if(!buf) {
+         /* give up */
+         return NULL;
+      }
+   }
+   
+   fenced_buf = fenced_buffer_create(fenced_mgr->fenced_list, buf);
+   if(!fenced_buf) {
+      assert(buf->base.refcount == 1);
+      pb_destroy(buf);
+   }
+   
+   return fenced_buf;
+}
+
+
+static void
+fenced_bufmgr_destroy(struct pb_manager *mgr)
+{
+   struct fenced_pb_manager *fenced_mgr = fenced_pb_manager(mgr);
+
+   fenced_buffer_list_destroy(fenced_mgr->fenced_list);
+
+   fenced_mgr->provider->destroy(fenced_mgr->provider);
+   
+   FREE(fenced_mgr);
+}
+
+
+struct pb_manager *
+fenced_bufmgr_create(struct pb_manager *provider, 
+                     struct pipe_winsys *winsys) 
+{
+   struct fenced_pb_manager *fenced_mgr;
+
+   fenced_mgr = (struct fenced_pb_manager *)CALLOC(1, sizeof(*fenced_mgr));
+   if (!fenced_mgr)
+      return NULL;
+
+   fenced_mgr->base.destroy = fenced_bufmgr_destroy;
+   fenced_mgr->base.create_buffer = fenced_bufmgr_create_buffer;
+
+   fenced_mgr->provider = provider;
+   fenced_mgr->fenced_list = fenced_buffer_list_create(winsys);
+   if(!fenced_mgr->fenced_list) {
+      FREE(fenced_mgr);
+      return NULL;
+   }
+      
+   return &fenced_mgr->base;
+}
diff --git a/src/gallium/aux/pipebuffer/pb_bufmgr_mm.c b/src/gallium/aux/pipebuffer/pb_bufmgr_mm.c
new file mode 100644
index 0000000000..8b1b51c0e2
--- /dev/null
+++ b/src/gallium/aux/pipebuffer/pb_bufmgr_mm.c
@@ -0,0 +1,593 @@
+/**************************************************************************
+ *
+ * Copyright 2006 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * Copyright 1999 Wittawat Yamwong
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ **************************************************************************/
+
+/**
+ * \file
+ * Buffer manager using the old texture memory manager.
+ * 
+ * \author José Fonseca <jrfonseca@tungstengraphics.com>
+ */
+
+
+#include "linked_list.h"
+
+#include "p_defines.h"
+#include "p_debug.h"
+#include "p_thread.h"
+#include "p_util.h"
+#include "pb_buffer.h"
+#include "pb_bufmgr.h"
+
+
+/**
+ * Convenience macro (type safe).
+ */
+#define SUPER(__derived) (&(__derived)->base)
+
+
+struct mem_block 
+{
+   struct mem_block *next, *prev;
+   struct mem_block *next_free, *prev_free;
+   struct mem_block *heap;
+   int ofs, size;
+   unsigned int free:1;
+   unsigned int reserved:1;
+};
+
+
+#ifdef DEBUG
+/**
+ * For debugging purposes.
+ */
+static void
+mmDumpMemInfo(const struct mem_block *heap)
+{
+   debug_printf("Memory heap %p:\n", (void *)heap);
+   if (heap == 0) {
+      debug_printf("  heap == 0\n");
+   } else {
+      const struct mem_block *p;
+
+      for(p = heap->next; p != heap; p = p->next) {
+	 debug_printf("  Offset:%08x, Size:%08x, %c%c\n",p->ofs,p->size,
+		 p->free ? 'F':'.',
+		 p->reserved ? 'R':'.');
+      }
+
+      debug_printf("\nFree list:\n");
+
+      for(p = heap->next_free; p != heap; p = p->next_free) {
+	 debug_printf(" FREE Offset:%08x, Size:%08x, %c%c\n",p->ofs,p->size,
+		 p->free ? 'F':'.',
+		 p->reserved ? 'R':'.');
+      }
+
+   }
+   debug_printf("End of memory blocks\n");
+}
+#endif
+
+
+/** 
+ * input: total size in bytes
+ * return: a heap pointer if OK, NULL if error
+ */
+static struct mem_block *
+mmInit(int ofs, int size)
+{
+   struct mem_block *heap, *block;
+  
+   if (size <= 0) 
+      return NULL;
+
+   heap = CALLOC_STRUCT(mem_block);
+   if (!heap) 
+      return NULL;
+   
+   block = CALLOC_STRUCT(mem_block);
+   if (!block) {
+      FREE(heap);
+      return NULL;
+   }
+
+   heap->next = block;
+   heap->prev = block;
+   heap->next_free = block;
+   heap->prev_free = block;
+
+   block->heap = heap;
+   block->next = heap;
+   block->prev = heap;
+   block->next_free = heap;
+   block->prev_free = heap;
+
+   block->ofs = ofs;
+   block->size = size;
+   block->free = 1;
+
+   return heap;
+}
+
+
+static struct mem_block *
+SliceBlock(struct mem_block *p, 
+           int startofs, int size, 
+           int reserved, int alignment)
+{
+   struct mem_block *newblock;
+
+   /* break left  [p, newblock, p->next], then p = newblock */
+   if (startofs > p->ofs) {
+      newblock = CALLOC_STRUCT(mem_block);
+      if (!newblock)
+	 return NULL;
+      newblock->ofs = startofs;
+      newblock->size = p->size - (startofs - p->ofs);
+      newblock->free = 1;
+      newblock->heap = p->heap;
+
+      newblock->next = p->next;
+      newblock->prev = p;
+      p->next->prev = newblock;
+      p->next = newblock;
+
+      newblock->next_free = p->next_free;
+      newblock->prev_free = p;
+      p->next_free->prev_free = newblock;
+      p->next_free = newblock;
+
+      p->size -= newblock->size;
+      p = newblock;
+   }
+
+   /* break right, also [p, newblock, p->next] */
+   if (size < p->size) {
+      newblock = CALLOC_STRUCT(mem_block);
+      if (!newblock)
+	 return NULL;
+      newblock->ofs = startofs + size;
+      newblock->size = p->size - size;
+      newblock->free = 1;
+      newblock->heap = p->heap;
+
+      newblock->next = p->next;
+      newblock->prev = p;
+      p->next->prev = newblock;
+      p->next = newblock;
+
+      newblock->next_free = p->next_free;
+      newblock->prev_free = p;
+      p->next_free->prev_free = newblock;
+      p->next_free = newblock;
+	 
+      p->size = size;
+   }
+
+   /* p = middle block */
+   p->free = 0;
+
+   /* Remove p from the free list: 
+    */
+   p->next_free->prev_free = p->prev_free;
+   p->prev_free->next_free = p->next_free;
+
+   p->next_free = 0;
+   p->prev_free = 0;
+
+   p->reserved = reserved;
+   return p;
+}
+
+
+/**
+ * Allocate 'size' bytes with 2^align2 bytes alignment,
+ * restrict the search to free memory after 'startSearch'
+ * depth and back buffers should be in different 4mb banks
+ * to get better page hits if possible
+ * input:	size = size of block
+ *       	align2 = 2^align2 bytes alignment
+ *		startSearch = linear offset from start of heap to begin search
+ * return: pointer to the allocated block, 0 if error
+ */
+static struct mem_block *
+mmAllocMem(struct mem_block *heap, int size, int align2, int startSearch)
+{
+   struct mem_block *p;
+   const int mask = (1 << align2)-1;
+   int startofs = 0;
+   int endofs;
+
+   if (!heap || align2 < 0 || size <= 0)
+      return NULL;
+
+   for (p = heap->next_free; p != heap; p = p->next_free) {
+      assert(p->free);
+
+      startofs = (p->ofs + mask) & ~mask;
+      if ( startofs < startSearch ) {
+	 startofs = startSearch;
+      }
+      endofs = startofs+size;
+      if (endofs <= (p->ofs+p->size))
+	 break;
+   }
+
+   if (p == heap) 
+      return NULL;
+
+   assert(p->free);
+   p = SliceBlock(p,startofs,size,0,mask+1);
+
+   return p;
+}
+
+
+#if 0
+/**
+ * Free block starts at offset
+ * input: pointer to a heap, start offset
+ * return: pointer to a block
+ */
+static struct mem_block *
+mmFindBlock(struct mem_block *heap, int start)
+{
+   struct mem_block *p;
+
+   for (p = heap->next; p != heap; p = p->next) {
+      if (p->ofs == start) 
+	 return p;
+   }
+
+   return NULL;
+}
+#endif
+
+
+static INLINE int
+Join2Blocks(struct mem_block *p)
+{
+   /* XXX there should be some assertions here */
+
+   /* NOTE: heap->free == 0 */
+
+   if (p->free && p->next->free) {
+      struct mem_block *q = p->next;
+
+      assert(p->ofs + p->size == q->ofs);
+      p->size += q->size;
+
+      p->next = q->next;
+      q->next->prev = p;
+
+      q->next_free->prev_free = q->prev_free; 
+      q->prev_free->next_free = q->next_free;
+     
+      FREE(q);
+      return 1;
+   }
+   return 0;
+}
+
+
+/**
+ * Free block starts at offset
+ * input: pointer to a block
+ * return: 0 if OK, -1 if error
+ */
+static int
+mmFreeMem(struct mem_block *b)
+{
+   if (!b)
+      return 0;
+
+   if (b->free) {
+      debug_printf("block already free\n");
+      return -1;
+   }
+   if (b->reserved) {
+      debug_printf("block is reserved\n");
+      return -1;
+   }
+
+   b->free = 1;
+   b->next_free = b->heap->next_free;
+   b->prev_free = b->heap;
+   b->next_free->prev_free = b;
+   b->prev_free->next_free = b;
+
+   Join2Blocks(b);
+   if (b->prev != b->heap)
+      Join2Blocks(b->prev);
+
+   return 0;
+}
+
+
+/**
+ * destroy MM
+ */
+static void
+mmDestroy(struct mem_block *heap)
+{
+   struct mem_block *p;
+
+   if (!heap)
+      return;
+
+   for (p = heap->next; p != heap; ) {
+      struct mem_block *next = p->next;
+      FREE(p);
+      p = next;
+   }
+
+   FREE(heap);
+}
+
+
+struct mm_pb_manager
+{
+   struct pb_manager base;
+   
+   _glthread_Mutex mutex;
+   
+   size_t size;
+   struct mem_block *heap;
+   
+   size_t align2;
+   
+   struct pb_buffer *buffer;
+   void *map;
+};
+
+
+static INLINE struct mm_pb_manager *
+mm_pb_manager(struct pb_manager *mgr)
+{
+   assert(mgr);
+   return (struct mm_pb_manager *)mgr;
+}
+
+
+struct mm_buffer
+{
+   struct pb_buffer base;
+   
+   struct mm_pb_manager *mgr;
+   
+   struct mem_block *block;
+};
+
+
+static INLINE struct mm_buffer *
+mm_buffer(struct pb_buffer *buf)
+{
+   assert(buf);
+   return (struct mm_buffer *)buf;
+}
+
+
+static void
+mm_buffer_destroy(struct pb_buffer *buf)
+{
+   struct mm_buffer *mm_buf = mm_buffer(buf);
+   struct mm_pb_manager *mm = mm_buf->mgr;
+   
+   assert(buf->base.refcount == 0);
+   
+   _glthread_LOCK_MUTEX(mm->mutex);
+   mmFreeMem(mm_buf->block);
+   FREE(buf);
+   _glthread_UNLOCK_MUTEX(mm->mutex);
+}
+
+
+static void *
+mm_buffer_map(struct pb_buffer *buf,
+              unsigned flags)
+{
+   struct mm_buffer *mm_buf = mm_buffer(buf);
+   struct mm_pb_manager *mm = mm_buf->mgr;
+
+   return (unsigned char *) mm->map + mm_buf->block->ofs;
+}
+
+
+static void
+mm_buffer_unmap(struct pb_buffer *buf)
+{
+   /* No-op */
+}
+
+
+static void
+mm_buffer_get_base_buffer(struct pb_buffer *buf,
+                          struct pb_buffer **base_buf,
+                          unsigned *offset)
+{
+   struct mm_buffer *mm_buf = mm_buffer(buf);
+   struct mm_pb_manager *mm = mm_buf->mgr;
+   pb_get_base_buffer(mm->buffer, base_buf, offset);
+   *offset += mm_buf->block->ofs;
+}
+
+
+static const struct pb_vtbl 
+mm_buffer_vtbl = {
+      mm_buffer_destroy,
+      mm_buffer_map,
+      mm_buffer_unmap,
+      mm_buffer_get_base_buffer
+};
+
+
+static struct pb_buffer *
+mm_bufmgr_create_buffer(struct pb_manager *mgr, 
+                        size_t size,
+                        const struct pb_desc *desc)
+{
+   struct mm_pb_manager *mm = mm_pb_manager(mgr);
+   struct mm_buffer *mm_buf;
+
+   /* We don't handle alignments larger then the one initially setup */
+   assert(desc->alignment % (1 << mm->align2) == 0);
+   if(desc->alignment % (1 << mm->align2))
+      return NULL;
+   
+   _glthread_LOCK_MUTEX(mm->mutex);
+
+   mm_buf = CALLOC_STRUCT(mm_buffer);
+   if (!mm_buf) {
+      _glthread_UNLOCK_MUTEX(mm->mutex);
+      return NULL;
+   }
+
+   mm_buf->base.base.refcount = 1;
+   mm_buf->base.base.alignment = desc->alignment;
+   mm_buf->base.base.usage = desc->usage;
+   mm_buf->base.base.size = size;
+   
+   mm_buf->base.vtbl = &mm_buffer_vtbl;
+   
+   mm_buf->mgr = mm;
+   
+   mm_buf->block = mmAllocMem(mm->heap, size, mm->align2, 0);
+   if(!mm_buf->block) {
+      debug_printf("warning: heap full\n");
+#if 0
+      mmDumpMemInfo(mm->heap);
+#endif
+      
+      mm_buf->block = mmAllocMem(mm->heap, size, mm->align2, 0);
+      if(!mm_buf->block) {
+	 assert(0);
+         FREE(mm_buf);
+         _glthread_UNLOCK_MUTEX(mm->mutex);
+         return NULL;
+      }
+   }
+   
+   /* Some sanity checks */
+   assert(0 <= mm_buf->block->ofs && mm_buf->block->ofs < mm->size);
+   assert(size <= mm_buf->block->size && mm_buf->block->ofs + mm_buf->block->size <= mm->size);
+   
+   _glthread_UNLOCK_MUTEX(mm->mutex);
+   return SUPER(mm_buf);
+}
+
+
+static void
+mm_bufmgr_destroy(struct pb_manager *mgr)
+{
+   struct mm_pb_manager *mm = mm_pb_manager(mgr);
+   
+   _glthread_LOCK_MUTEX(mm->mutex);
+
+   mmDestroy(mm->heap);
+   
+   pb_unmap(mm->buffer);
+   pb_reference(&mm->buffer, NULL);
+   
+   _glthread_UNLOCK_MUTEX(mm->mutex);
+   
+   FREE(mgr);
+}
+
+
+struct pb_manager *
+mm_bufmgr_create_from_buffer(struct pb_buffer *buffer, 
+                             size_t size, size_t align2) 
+{
+   struct mm_pb_manager *mm;
+
+   if(!buffer)
+      return NULL;
+   
+   mm = CALLOC_STRUCT(mm_pb_manager);
+   if (!mm)
+      return NULL;
+
+   mm->base.create_buffer = mm_bufmgr_create_buffer;
+   mm->base.destroy = mm_bufmgr_destroy;
+
+   mm->size = size;
+   mm->align2 = align2; /* 64-byte alignment */
+
+   _glthread_INIT_MUTEX(mm->mutex);
+
+   mm->buffer = buffer; 
+
+   mm->map = pb_map(mm->buffer, 
+		    PIPE_BUFFER_USAGE_CPU_READ |
+		    PIPE_BUFFER_USAGE_CPU_WRITE);
+   if(!mm->map)
+      goto failure;
+
+   mm->heap = mmInit(0, size); 
+   if (!mm->heap)
+      goto failure;
+
+   return SUPER(mm);
+   
+failure:
+if(mm->heap)
+   mmDestroy(mm->heap);
+   if(mm->map)
+      pb_unmap(mm->buffer);
+   if(mm)
+      FREE(mm);
+   return NULL;
+}
+
+
+struct pb_manager *
+mm_bufmgr_create(struct pb_manager *provider, 
+                 size_t size, size_t align2) 
+{
+   struct pb_buffer *buffer;
+   struct pb_manager *mgr;
+   struct pb_desc desc;
+
+   assert(provider);
+   assert(provider->create_buffer);
+   
+   memset(&desc, 0, sizeof(desc));
+   desc.alignment = 1 << align2;
+   
+   buffer = provider->create_buffer(provider, size, &desc); 
+   if (!buffer)
+      return NULL;
+   
+   mgr = mm_bufmgr_create_from_buffer(buffer, size, align2);
+   if (!mgr) {
+      pb_reference(&buffer, NULL);
+      return NULL;
+   }
+
+  return mgr;
+}
diff --git a/src/gallium/aux/pipebuffer/pb_bufmgr_pool.c b/src/gallium/aux/pipebuffer/pb_bufmgr_pool.c
new file mode 100644
index 0000000000..04477a865a
--- /dev/null
+++ b/src/gallium/aux/pipebuffer/pb_bufmgr_pool.c
@@ -0,0 +1,288 @@
+/**************************************************************************
+ * 
+ * Copyright 2006 Tungsten Graphics, Inc., Bismarck, ND., USA
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDERS, AUTHORS AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM,
+ * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR 
+ * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE 
+ * USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * 
+ **************************************************************************/
+
+/**
+ * \file
+ * Batch buffer pool management.
+ * 
+ * \author José Fonseca <jrfonseca-at-tungstengraphics-dot-com>
+ * \author Thomas Hellström <thomas-at-tungstengraphics-dot-com>
+ */
+
+
+#include "linked_list.h"
+
+#include "p_compiler.h"
+#include "p_debug.h"
+#include "p_thread.h"
+#include "p_defines.h"
+#include "p_util.h"
+
+#include "pb_buffer.h"
+#include "pb_bufmgr.h"
+
+
+/**
+ * Convenience macro (type safe).
+ */
+#define SUPER(__derived) (&(__derived)->base)
+
+
+struct pool_pb_manager
+{
+   struct pb_manager base;
+   
+   _glthread_Mutex mutex;
+   
+   size_t bufSize;
+   size_t bufAlign;
+   
+   size_t numFree;
+   size_t numTot;
+   
+   struct list_head free;
+   
+   struct pb_buffer *buffer;
+   void *map;
+   
+   struct pool_buffer *bufs;
+};
+
+
+static INLINE struct pool_pb_manager *
+pool_pb_manager(struct pb_manager *mgr)
+{
+   assert(mgr);
+   return (struct pool_pb_manager *)mgr;
+}
+
+
+struct pool_buffer
+{
+   struct pb_buffer base;
+   
+   struct pool_pb_manager *mgr;
+   
+   struct list_head head;
+   
+   size_t start;
+};
+
+
+static INLINE struct pool_buffer *
+pool_buffer(struct pb_buffer *buf)
+{
+   assert(buf);
+   return (struct pool_buffer *)buf;
+}
+
+
+
+static void
+pool_buffer_destroy(struct pb_buffer *buf)
+{
+   struct pool_buffer *pool_buf = pool_buffer(buf);
+   struct pool_pb_manager *pool = pool_buf->mgr;
+   
+   assert(pool_buf->base.base.refcount == 0);
+
+   _glthread_LOCK_MUTEX(pool->mutex);
+   LIST_ADD(&pool_buf->head, &pool->free);
+   pool->numFree++;
+   _glthread_UNLOCK_MUTEX(pool->mutex);
+}
+
+
+static void *
+pool_buffer_map(struct pb_buffer *buf, unsigned flags)
+{
+   struct pool_buffer *pool_buf = pool_buffer(buf);
+   struct pool_pb_manager *pool = pool_buf->mgr;
+   void *map;
+
+   _glthread_LOCK_MUTEX(pool->mutex);
+   map = (unsigned char *) pool->map + pool_buf->start;
+   _glthread_UNLOCK_MUTEX(pool->mutex);
+   return map;
+}
+
+
+static void
+pool_buffer_unmap(struct pb_buffer *buf)
+{
+   /* No-op */
+}
+
+
+static void
+pool_buffer_get_base_buffer(struct pb_buffer *buf,
+                            struct pb_buffer **base_buf,
+                            unsigned *offset)
+{
+   struct pool_buffer *pool_buf = pool_buffer(buf);
+   struct pool_pb_manager *pool = pool_buf->mgr;
+   pb_get_base_buffer(pool->buffer, base_buf, offset);
+   *offset += pool_buf->start;
+}
+
+
+static const struct pb_vtbl 
+pool_buffer_vtbl = {
+      pool_buffer_destroy,
+      pool_buffer_map,
+      pool_buffer_unmap,
+      pool_buffer_get_base_buffer
+};
+
+
+static struct pb_buffer *
+pool_bufmgr_create_buffer(struct pb_manager *mgr,
+                          size_t size,
+                          const struct pb_desc *desc)
+{
+   struct pool_pb_manager *pool = pool_pb_manager(mgr);
+   struct pool_buffer *pool_buf;
+   struct list_head *item;
+
+   assert(size == pool->bufSize);
+   assert(pool->bufAlign % desc->alignment == 0);
+   
+   _glthread_LOCK_MUTEX(pool->mutex);
+
+   if (pool->numFree == 0) {
+      _glthread_UNLOCK_MUTEX(pool->mutex);
+      debug_printf("warning: out of fixed size buffer objects\n");
+      return NULL;
+   }
+
+   item = pool->free.next;
+
+   if (item == &pool->free) {
+      _glthread_UNLOCK_MUTEX(pool->mutex);
+      debug_printf("error: fixed size buffer pool corruption\n");
+      return NULL;
+   }
+
+   LIST_DEL(item);
+   --pool->numFree;
+
+   _glthread_UNLOCK_MUTEX(pool->mutex);
+   
+   pool_buf = LIST_ENTRY(struct pool_buffer, item, head);
+   assert(pool_buf->base.base.refcount == 0);
+   pool_buf->base.base.refcount = 1;
+   pool_buf->base.base.alignment = desc->alignment;
+   pool_buf->base.base.usage = desc->usage;
+   
+   return SUPER(pool_buf);
+}
+
+
+static void
+pool_bufmgr_destroy(struct pb_manager *mgr)
+{
+   struct pool_pb_manager *pool = pool_pb_manager(mgr);
+   _glthread_LOCK_MUTEX(pool->mutex);
+
+   FREE(pool->bufs);
+   
+   pb_unmap(pool->buffer);
+   pb_reference(&pool->buffer, NULL);
+   
+   _glthread_UNLOCK_MUTEX(pool->mutex);
+   
+   FREE(mgr);
+}
+
+
+struct pb_manager *
+pool_bufmgr_create(struct pb_manager *provider, 
+                   size_t numBufs, 
+                   size_t bufSize,
+                   const struct pb_desc *desc) 
+{
+   struct pool_pb_manager *pool;
+   struct pool_buffer *pool_buf;
+   size_t i;
+
+   pool = (struct pool_pb_manager *)CALLOC(1, sizeof(*pool));
+   if (!pool)
+      return NULL;
+
+   pool->base.destroy = pool_bufmgr_destroy;
+   pool->base.create_buffer = pool_bufmgr_create_buffer;
+
+   LIST_INITHEAD(&pool->free);
+
+   pool->numTot = numBufs;
+   pool->numFree = numBufs;
+   pool->bufSize = bufSize;
+   pool->bufAlign = desc->alignment; 
+   
+   _glthread_INIT_MUTEX(pool->mutex);
+
+   pool->buffer = provider->create_buffer(provider, numBufs*bufSize, desc); 
+   if (!pool->buffer)
+      goto failure;
+
+   pool->map = pb_map(pool->buffer,
+                          PIPE_BUFFER_USAGE_CPU_READ |
+                          PIPE_BUFFER_USAGE_CPU_WRITE);
+   if(!pool->map)
+      goto failure;
+
+   pool->bufs = (struct pool_buffer *)CALLOC(numBufs, sizeof(*pool->bufs));
+   if (!pool->bufs)
+      goto failure;
+
+   pool_buf = pool->bufs;
+   for (i = 0; i < numBufs; ++i) {
+      pool_buf->base.base.refcount = 0;
+      pool_buf->base.base.alignment = 0;
+      pool_buf->base.base.usage = 0;
+      pool_buf->base.base.size = bufSize;
+      pool_buf->base.vtbl = &pool_buffer_vtbl;
+      pool_buf->mgr = pool;
+      pool_buf->start = i * bufSize;
+      LIST_ADDTAIL(&pool_buf->head, &pool->free);
+      pool_buf++;
+   }
+
+   return SUPER(pool);
+   
+failure:
+   if(pool->bufs)
+      FREE(pool->bufs);
+   if(pool->map)
+      pb_unmap(pool->buffer);
+   if(pool->buffer)
+      pb_reference(&pool->buffer, NULL);
+   if(pool)
+      FREE(pool);
+   return NULL;
+}
diff --git a/src/gallium/aux/pipebuffer/pb_winsys.c b/src/gallium/aux/pipebuffer/pb_winsys.c
new file mode 100644
index 0000000000..978944091f
--- /dev/null
+++ b/src/gallium/aux/pipebuffer/pb_winsys.c
@@ -0,0 +1,170 @@
+/**************************************************************************
+ *
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ **************************************************************************/
+
+/**
+ * \file
+ * Implementation of client buffer (also designated as "user buffers"), which
+ * are just state-tracker owned data masqueraded as buffers.
+ * 
+ * \author José Fonseca <jrfonseca@tungstengraphics.com>
+ */
+
+
+#include "pipe/p_winsys.h"
+#include "pipe/p_util.h"
+
+#include "pb_buffer.h"
+
+
+/**
+ * User buffers are special buffers that initially reference memory
+ * held by the user but which may if necessary copy that memory into
+ * device memory behind the scenes, for submission to hardware.
+ *
+ * These are particularly useful when the referenced data is never
+ * submitted to hardware at all, in the particular case of software
+ * vertex processing.
+ */
+struct pb_user_buffer 
+{
+   struct pb_buffer base;
+   void *data;
+};
+
+
+extern const struct pb_vtbl pb_user_buffer_vtbl;
+
+
+static INLINE struct pb_user_buffer *
+pb_user_buffer(struct pb_buffer *buf)
+{
+   assert(buf);
+   assert(buf->vtbl == &pb_user_buffer_vtbl);
+   return (struct pb_user_buffer *)buf;
+}
+
+
+static void
+pb_user_buffer_destroy(struct pb_buffer *buf)
+{
+   assert(buf);
+   FREE(buf);
+}
+
+
+static void *
+pb_user_buffer_map(struct pb_buffer *buf, 
+                   unsigned flags)
+{
+   return pb_user_buffer(buf)->data;
+}
+
+
+static void
+pb_user_buffer_unmap(struct pb_buffer *buf)
+{
+   /* No-op */
+}
+
+
+static void
+pb_user_buffer_get_base_buffer(struct pb_buffer *buf,
+                               struct pb_buffer **base_buf,
+                               unsigned *offset)
+{
+   *base_buf = buf;
+   *offset = 0;
+}
+
+
+const struct pb_vtbl 
+pb_user_buffer_vtbl = {
+      pb_user_buffer_destroy,
+      pb_user_buffer_map,
+      pb_user_buffer_unmap,
+      pb_user_buffer_get_base_buffer
+};
+
+
+static struct pipe_buffer *
+pb_winsys_user_buffer_create(struct pipe_winsys *winsys,
+                             void *data, 
+                             unsigned bytes) 
+{
+   struct pb_user_buffer *buf = CALLOC_STRUCT(pb_user_buffer);
+
+   if(!buf)
+      return NULL;
+   
+   buf->base.base.refcount = 1;
+   buf->base.base.size = bytes;
+   buf->base.base.alignment = 0;
+   buf->base.base.usage = 0;
+
+   buf->base.vtbl = &pb_user_buffer_vtbl;   
+   buf->data = data;
+   
+   return &buf->base.base;
+}
+
+
+static void *
+pb_winsys_buffer_map(struct pipe_winsys *winsys,
+                     struct pipe_buffer *buf,
+                     unsigned flags)
+{
+   (void)winsys;
+   return pb_map(pb_buffer(buf), flags);
+}
+
+
+static void
+pb_winsys_buffer_unmap(struct pipe_winsys *winsys,
+                       struct pipe_buffer *buf)
+{
+   (void)winsys;
+   pb_unmap(pb_buffer(buf));
+}
+
+
+static void
+pb_winsys_buffer_destroy(struct pipe_winsys *winsys,
+                         struct pipe_buffer *buf)
+{
+   (void)winsys;
+   pb_destroy(pb_buffer(buf));
+}
+
+
+void 
+pb_init_winsys(struct pipe_winsys *winsys)
+{
+   winsys->user_buffer_create = pb_winsys_user_buffer_create;
+   winsys->buffer_map = pb_winsys_buffer_map;
+   winsys->buffer_unmap = pb_winsys_buffer_unmap;
+   winsys->buffer_destroy = pb_winsys_buffer_destroy;
+}
diff --git a/src/gallium/aux/tgsi/Makefile b/src/gallium/aux/tgsi/Makefile
new file mode 100644
index 0000000000..12a8bd0409
--- /dev/null
+++ b/src/gallium/aux/tgsi/Makefile
@@ -0,0 +1,3 @@
+default:
+	cd ../.. ; make
+
diff --git a/src/gallium/aux/tgsi/exec/Makefile b/src/gallium/aux/tgsi/exec/Makefile
new file mode 100644
index 0000000000..eb8b14e0e8
--- /dev/null
+++ b/src/gallium/aux/tgsi/exec/Makefile
@@ -0,0 +1,3 @@
+default:
+	cd ../../.. ; make
+
diff --git a/src/gallium/aux/tgsi/exec/tgsi_exec.c b/src/gallium/aux/tgsi/exec/tgsi_exec.c
new file mode 100644
index 0000000000..37e6007068
--- /dev/null
+++ b/src/gallium/aux/tgsi/exec/tgsi_exec.c
@@ -0,0 +1,2485 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+/**
+ * TGSI interpretor/executor.
+ *
+ * Flow control information:
+ *
+ * Since we operate on 'quads' (4 pixels or 4 vertices in parallel)
+ * flow control statements (IF/ELSE/ENDIF, LOOP/ENDLOOP) require special
+ * care since a condition may be true for some quad components but false
+ * for other components.
+ *
+ * We basically execute all statements (even if they're in the part of
+ * an IF/ELSE clause that's "not taken") and use a special mask to
+ * control writing to destination registers.  This is the ExecMask.
+ * See store_dest().
+ *
+ * The ExecMask is computed from three other masks (CondMask, LoopMask and
+ * ContMask) which are controlled by the flow control instructions (namely:
+ * (IF/ELSE/ENDIF, LOOP/ENDLOOP and CONT).
+ *
+ *
+ * Authors:
+ *   Michal Krol
+ *   Brian Paul
+ */
+
+#include "pipe/p_compiler.h"
+#include "pipe/p_state.h"
+#include "pipe/p_util.h"
+#include "pipe/p_shader_tokens.h"
+#include "pipe/tgsi/util/tgsi_parse.h"
+#include "pipe/tgsi/util/tgsi_util.h"
+#include "tgsi_exec.h"
+
+#define TILE_TOP_LEFT     0
+#define TILE_TOP_RIGHT    1
+#define TILE_BOTTOM_LEFT  2
+#define TILE_BOTTOM_RIGHT 3
+
+/*
+ * Shorthand locations of various utility registers (_I = Index, _C = Channel)
+ */
+#define TEMP_0_I           TGSI_EXEC_TEMP_00000000_I
+#define TEMP_0_C           TGSI_EXEC_TEMP_00000000_C
+#define TEMP_7F_I          TGSI_EXEC_TEMP_7FFFFFFF_I
+#define TEMP_7F_C          TGSI_EXEC_TEMP_7FFFFFFF_C
+#define TEMP_80_I          TGSI_EXEC_TEMP_80000000_I
+#define TEMP_80_C          TGSI_EXEC_TEMP_80000000_C
+#define TEMP_FF_I          TGSI_EXEC_TEMP_FFFFFFFF_I
+#define TEMP_FF_C          TGSI_EXEC_TEMP_FFFFFFFF_C
+#define TEMP_1_I           TGSI_EXEC_TEMP_ONE_I
+#define TEMP_1_C           TGSI_EXEC_TEMP_ONE_C
+#define TEMP_2_I           TGSI_EXEC_TEMP_TWO_I
+#define TEMP_2_C           TGSI_EXEC_TEMP_TWO_C
+#define TEMP_128_I         TGSI_EXEC_TEMP_128_I
+#define TEMP_128_C         TGSI_EXEC_TEMP_128_C
+#define TEMP_M128_I        TGSI_EXEC_TEMP_MINUS_128_I
+#define TEMP_M128_C        TGSI_EXEC_TEMP_MINUS_128_C
+#define TEMP_KILMASK_I     TGSI_EXEC_TEMP_KILMASK_I
+#define TEMP_KILMASK_C     TGSI_EXEC_TEMP_KILMASK_C
+#define TEMP_OUTPUT_I      TGSI_EXEC_TEMP_OUTPUT_I
+#define TEMP_OUTPUT_C      TGSI_EXEC_TEMP_OUTPUT_C
+#define TEMP_PRIMITIVE_I   TGSI_EXEC_TEMP_PRIMITIVE_I
+#define TEMP_PRIMITIVE_C   TGSI_EXEC_TEMP_PRIMITIVE_C
+#define TEMP_R0            TGSI_EXEC_TEMP_R0
+
+#define FOR_EACH_CHANNEL(CHAN)\
+   for (CHAN = 0; CHAN < 4; CHAN++)
+
+#define IS_CHANNEL_ENABLED(INST, CHAN)\
+   ((INST).FullDstRegisters[0].DstRegister.WriteMask & (1 << (CHAN)))
+
+#define IS_CHANNEL_ENABLED2(INST, CHAN)\
+   ((INST).FullDstRegisters[1].DstRegister.WriteMask & (1 << (CHAN)))
+
+#define FOR_EACH_ENABLED_CHANNEL(INST, CHAN)\
+   FOR_EACH_CHANNEL( CHAN )\
+      if (IS_CHANNEL_ENABLED( INST, CHAN ))
+
+#define FOR_EACH_ENABLED_CHANNEL2(INST, CHAN)\
+   FOR_EACH_CHANNEL( CHAN )\
+      if (IS_CHANNEL_ENABLED2( INST, CHAN ))
+
+
+/** The execution mask depends on the conditional mask and the loop mask */
+#define UPDATE_EXEC_MASK(MACH) \
+      MACH->ExecMask = MACH->CondMask & MACH->LoopMask & MACH->ContMask & MACH->FuncMask
+
+
+#define CHAN_X  0
+#define CHAN_Y  1
+#define CHAN_Z  2
+#define CHAN_W  3
+
+
+
+static void
+tgsi_exec_prepare( struct tgsi_exec_machine *mach )
+{
+   struct tgsi_exec_labels *labels = &mach->Labels;
+   struct tgsi_parse_context parse;
+   struct tgsi_full_instruction *instructions;
+   struct tgsi_full_declaration *declarations;
+   uint maxInstructions = 10, numInstructions = 0;
+   uint maxDeclarations = 10, numDeclarations = 0;
+   uint k;
+   uint instno = 0;
+
+   mach->ImmLimit = 0;
+   labels->count = 0;
+
+   declarations = (struct tgsi_full_declaration *)
+      MALLOC( maxDeclarations * sizeof(struct tgsi_full_declaration) );
+
+   instructions = (struct tgsi_full_instruction *)
+      MALLOC( maxInstructions * sizeof(struct tgsi_full_instruction) );
+
+   k = tgsi_parse_init( &parse, mach->Tokens );
+   if (k != TGSI_PARSE_OK) {
+      debug_printf("Problem parsing!\n");
+      return;
+   }
+
+   while( !tgsi_parse_end_of_tokens( &parse ) ) {
+      uint pointer = parse.Position;
+      uint i;
+
+      tgsi_parse_token( &parse );
+      switch( parse.FullToken.Token.Type ) {
+      case TGSI_TOKEN_TYPE_DECLARATION:
+         /* save expanded declaration */
+         if (numDeclarations == maxDeclarations) {
+            declarations = REALLOC(declarations,
+                                   maxDeclarations
+                                   * sizeof(struct tgsi_full_declaration),
+                                   (maxDeclarations + 10)
+                                   * sizeof(struct tgsi_full_declaration));
+            maxDeclarations += 10;
+         }
+         memcpy(declarations + numDeclarations,
+                &parse.FullToken.FullDeclaration,
+                sizeof(declarations[0]));
+         numDeclarations++;
+         break;
+
+      case TGSI_TOKEN_TYPE_IMMEDIATE:
+         {
+            uint size = parse.FullToken.FullImmediate.Immediate.Size - 1;
+            assert( size % 4 == 0 );
+            assert( mach->ImmLimit + size / 4 <= TGSI_EXEC_NUM_IMMEDIATES );
+
+            for( i = 0; i < size; i++ ) {
+               mach->Imms[mach->ImmLimit + i / 4][i % 4] = parse.FullToken.FullImmediate.u.ImmediateFloat32[i].Float;
+            }
+            mach->ImmLimit += size / 4;
+         }
+         break;
+
+      case TGSI_TOKEN_TYPE_INSTRUCTION:
+         assert( labels->count < 128 );
+
+         labels->labels[labels->count][0] = instno;
+         labels->labels[labels->count][1] = pointer;
+         labels->count++;
+
+         /* save expanded instruction */
+         if (numInstructions == maxInstructions) {
+            instructions = REALLOC(instructions,
+                                   maxInstructions
+                                   * sizeof(struct tgsi_full_instruction),
+                                   (maxInstructions + 10)
+                                   * sizeof(struct tgsi_full_instruction));
+            maxInstructions += 10;
+         }
+         memcpy(instructions + numInstructions,
+                &parse.FullToken.FullInstruction,
+                sizeof(instructions[0]));
+         numInstructions++;
+         break;
+
+      default:
+         assert( 0 );
+      }
+   }
+   tgsi_parse_free (&parse);
+
+   if (mach->Declarations) {
+      FREE( mach->Declarations );
+   }
+   mach->Declarations = declarations;
+   mach->NumDeclarations = numDeclarations;
+
+   if (mach->Instructions) {
+      FREE( mach->Instructions );
+   }
+   mach->Instructions = instructions;
+   mach->NumInstructions = numInstructions;
+}
+
+
+/**
+ * Initialize machine state by expanding tokens to full instructions,
+ * allocating temporary storage, setting up constants, etc.
+ * After this, we can call tgsi_exec_machine_run() many times.
+ */
+void
+tgsi_exec_machine_init(
+   struct tgsi_exec_machine *mach,
+   const struct tgsi_token *tokens,
+   uint numSamplers,
+   struct tgsi_sampler *samplers)
+{
+   uint i, k;
+   struct tgsi_parse_context parse;
+
+#if 0
+   tgsi_dump(tokens, 0);
+#endif
+
+   mach->Tokens = tokens;
+
+   mach->Samplers = samplers;
+
+   k = tgsi_parse_init (&parse, mach->Tokens);
+   if (k != TGSI_PARSE_OK) {
+      debug_printf( "Problem parsing!\n" );
+      return;
+   }
+
+   mach->Processor = parse.FullHeader.Processor.Processor;
+   tgsi_parse_free (&parse);
+
+   mach->Temps = (struct tgsi_exec_vector *) tgsi_align_128bit( mach->_Temps);
+   mach->Addrs = &mach->Temps[TGSI_EXEC_NUM_TEMPS];
+
+   /* Setup constants. */
+   for( i = 0; i < 4; i++ ) {
+      mach->Temps[TEMP_0_I].xyzw[TEMP_0_C].u[i] = 0x00000000;
+      mach->Temps[TEMP_7F_I].xyzw[TEMP_7F_C].u[i] = 0x7FFFFFFF;
+      mach->Temps[TEMP_80_I].xyzw[TEMP_80_C].u[i] = 0x80000000;
+      mach->Temps[TEMP_FF_I].xyzw[TEMP_FF_C].u[i] = 0xFFFFFFFF;
+      mach->Temps[TEMP_1_I].xyzw[TEMP_1_C].f[i] = 1.0f;
+      mach->Temps[TEMP_2_I].xyzw[TEMP_2_C].f[i] = 2.0f;
+      mach->Temps[TEMP_128_I].xyzw[TEMP_128_C].f[i] = 128.0f;
+      mach->Temps[TEMP_M128_I].xyzw[TEMP_M128_C].f[i] = -128.0f;
+   }
+
+   tgsi_exec_prepare( mach );
+}
+
+
+void
+tgsi_exec_machine_free_data(struct tgsi_exec_machine *mach)
+{
+   if (mach->Instructions) {
+      FREE(mach->Instructions);
+      mach->Instructions = NULL;
+      mach->NumInstructions = 0;
+   }
+   if (mach->Declarations) {
+      FREE(mach->Declarations);
+      mach->Declarations = NULL;
+      mach->NumDeclarations = 0;
+   }
+}
+
+
+static void
+micro_abs(
+   union tgsi_exec_channel *dst,
+   const union tgsi_exec_channel *src )
+{
+   dst->f[0] = (float) fabs( (double) src->f[0] );
+   dst->f[1] = (float) fabs( (double) src->f[1] );
+   dst->f[2] = (float) fabs( (double) src->f[2] );
+   dst->f[3] = (float) fabs( (double) src->f[3] );
+}
+
+static void
+micro_add(
+   union tgsi_exec_channel *dst,
+   const union tgsi_exec_channel *src0,
+   const union tgsi_exec_channel *src1 )
+{
+   dst->f[0] = src0->f[0] + src1->f[0];
+   dst->f[1] = src0->f[1] + src1->f[1];
+   dst->f[2] = src0->f[2] + src1->f[2];
+   dst->f[3] = src0->f[3] + src1->f[3];
+}
+
+static void
+micro_iadd(
+   union tgsi_exec_channel *dst,
+   const union tgsi_exec_channel *src0,
+   const union tgsi_exec_channel *src1 )
+{
+   dst->i[0] = src0->i[0] + src1->i[0];
+   dst->i[1] = src0->i[1] + src1->i[1];
+   dst->i[2] = src0->i[2] + src1->i[2];
+   dst->i[3] = src0->i[3] + src1->i[3];
+}
+
+static void
+micro_and(
+   union tgsi_exec_channel *dst,
+   const union tgsi_exec_channel *src0,
+   const union tgsi_exec_channel *src1 )
+{
+   dst->u[0] = src0->u[0] & src1->u[0];
+   dst->u[1] = src0->u[1] & src1->u[1];
+   dst->u[2] = src0->u[2] & src1->u[2];
+   dst->u[3] = src0->u[3] & src1->u[3];
+}
+
+static void
+micro_ceil(
+   union tgsi_exec_channel *dst,
+   const union tgsi_exec_channel *src )
+{
+   dst->f[0] = (float) ceil( (double) src->f[0] );
+   dst->f[1] = (float) ceil( (double) src->f[1] );
+   dst->f[2] = (float) ceil( (double) src->f[2] );
+   dst->f[3] = (float) ceil( (double) src->f[3] );
+}
+
+static void
+micro_cos(
+   union tgsi_exec_channel *dst,
+   const union tgsi_exec_channel *src )
+{
+   dst->f[0] = (float) cos( (double) src->f[0] );
+   dst->f[1] = (float) cos( (double) src->f[1] );
+   dst->f[2] = (float) cos( (double) src->f[2] );
+   dst->f[3] = (float) cos( (double) src->f[3] );
+}
+
+static void
+micro_ddx(
+   union tgsi_exec_channel *dst,
+   const union tgsi_exec_channel *src )
+{
+   dst->f[0] =
+   dst->f[1] =
+   dst->f[2] =
+   dst->f[3] = src->f[TILE_BOTTOM_RIGHT] - src->f[TILE_BOTTOM_LEFT];
+}
+
+static void
+micro_ddy(
+   union tgsi_exec_channel *dst,
+   const union tgsi_exec_channel *src )
+{
+   dst->f[0] =
+   dst->f[1] =
+   dst->f[2] =
+   dst->f[3] = src->f[TILE_TOP_LEFT] - src->f[TILE_BOTTOM_LEFT];
+}
+
+static void
+micro_div(
+   union tgsi_exec_channel *dst,
+   const union tgsi_exec_channel *src0,
+   const union tgsi_exec_channel *src1 )
+{
+   dst->f[0] = src0->f[0] / src1->f[0];
+   dst->f[1] = src0->f[1] / src1->f[1];
+   dst->f[2] = src0->f[2] / src1->f[2];
+   dst->f[3] = src0->f[3] / src1->f[3];
+}
+
+static void
+micro_udiv(
+   union tgsi_exec_channel *dst,
+   const union tgsi_exec_channel *src0,
+   const union tgsi_exec_channel *src1 )
+{
+   dst->u[0] = src0->u[0] / src1->u[0];
+   dst->u[1] = src0->u[1] / src1->u[1];
+   dst->u[2] = src0->u[2] / src1->u[2];
+   dst->u[3] = src0->u[3] / src1->u[3];
+}
+
+static void
+micro_eq(
+   union tgsi_exec_channel *dst,
+   const union tgsi_exec_channel *src0,
+   const union tgsi_exec_channel *src1,
+   const union tgsi_exec_channel *src2,
+   const union tgsi_exec_channel *src3 )
+{
+   dst->f[0] = src0->f[0] == src1->f[0] ? src2->f[0] : src3->f[0];
+   dst->f[1] = src0->f[1] == src1->f[1] ? src2->f[1] : src3->f[1];
+   dst->f[2] = src0->f[2] == src1->f[2] ? src2->f[2] : src3->f[2];
+   dst->f[3] = src0->f[3] == src1->f[3] ? src2->f[3] : src3->f[3];
+}
+
+static void
+micro_ieq(
+   union tgsi_exec_channel *dst,
+   const union tgsi_exec_channel *src0,
+   const union tgsi_exec_channel *src1,
+   const union tgsi_exec_channel *src2,
+   const union tgsi_exec_channel *src3 )
+{
+   dst->i[0] = src0->i[0] == src1->i[0] ? src2->i[0] : src3->i[0];
+   dst->i[1] = src0->i[1] == src1->i[1] ? src2->i[1] : src3->i[1];
+   dst->i[2] = src0->i[2] == src1->i[2] ? src2->i[2] : src3->i[2];
+   dst->i[3] = src0->i[3] == src1->i[3] ? src2->i[3] : src3->i[3];
+}
+
+static void
+micro_exp2(
+   union tgsi_exec_channel *dst,
+   const union tgsi_exec_channel *src)
+{
+   dst->f[0] = (float) pow( 2.0, (double) src->f[0] );
+   dst->f[1] = (float) pow( 2.0, (double) src->f[1] );
+   dst->f[2] = (float) pow( 2.0, (double) src->f[2] );
+   dst->f[3] = (float) pow( 2.0, (double) src->f[3] );
+}
+
+static void
+micro_f2it(
+   union tgsi_exec_channel *dst,
+   const union tgsi_exec_channel *src )
+{
+   dst->i[0] = (int) src->f[0];
+   dst->i[1] = (int) src->f[1];
+   dst->i[2] = (int) src->f[2];
+   dst->i[3] = (int) src->f[3];
+}
+
+static void
+micro_f2ut(
+   union tgsi_exec_channel *dst,
+   const union tgsi_exec_channel *src )
+{
+   dst->u[0] = (uint) src->f[0];
+   dst->u[1] = (uint) src->f[1];
+   dst->u[2] = (uint) src->f[2];
+   dst->u[3] = (uint) src->f[3];
+}
+
+static void
+micro_flr(
+   union tgsi_exec_channel *dst,
+   const union tgsi_exec_channel *src )
+{
+   dst->f[0] = (float) floor( (double) src->f[0] );
+   dst->f[1] = (float) floor( (double) src->f[1] );
+   dst->f[2] = (float) floor( (double) src->f[2] );
+   dst->f[3] = (float) floor( (double) src->f[3] );
+}
+
+static void
+micro_frc(
+   union tgsi_exec_channel *dst,
+   const union tgsi_exec_channel *src )
+{
+   dst->f[0] = src->f[0] - (float) floor( (double) src->f[0] );
+   dst->f[1] = src->f[1] - (float) floor( (double) src->f[1] );
+   dst->f[2] = src->f[2] - (float) floor( (double) src->f[2] );
+   dst->f[3] = src->f[3] - (float) floor( (double) src->f[3] );
+}
+
+static void
+micro_ge(
+   union tgsi_exec_channel *dst,
+   const union tgsi_exec_channel *src0,
+   const union tgsi_exec_channel *src1,
+   const union tgsi_exec_channel *src2,
+   const union tgsi_exec_channel *src3 )
+{
+   dst->f[0] = src0->f[0] >= src1->f[0] ? src2->f[0] : src3->f[0];
+   dst->f[1] = src0->f[1] >= src1->f[1] ? src2->f[1] : src3->f[1];
+   dst->f[2] = src0->f[2] >= src1->f[2] ? src2->f[2] : src3->f[2];
+   dst->f[3] = src0->f[3] >= src1->f[3] ? src2->f[3] : src3->f[3];
+}
+
+static void
+micro_i2f(
+   union tgsi_exec_channel *dst,
+   const union tgsi_exec_channel *src )
+{
+   dst->f[0] = (float) src->i[0];
+   dst->f[1] = (float) src->i[1];
+   dst->f[2] = (float) src->i[2];
+   dst->f[3] = (float) src->i[3];
+}
+
+static void
+micro_lg2(
+   union tgsi_exec_channel *dst,
+   const union tgsi_exec_channel *src )
+{
+   dst->f[0] = (float) log( (double) src->f[0] ) * 1.442695f;
+   dst->f[1] = (float) log( (double) src->f[1] ) * 1.442695f;
+   dst->f[2] = (float) log( (double) src->f[2] ) * 1.442695f;
+   dst->f[3] = (float) log( (double) src->f[3] ) * 1.442695f;
+}
+
+static void
+micro_lt(
+   union tgsi_exec_channel *dst,
+   const union tgsi_exec_channel *src0,
+   const union tgsi_exec_channel *src1,
+   const union tgsi_exec_channel *src2,
+   const union tgsi_exec_channel *src3 )
+{
+   dst->f[0] = src0->f[0] < src1->f[0] ? src2->f[0] : src3->f[0];
+   dst->f[1] = src0->f[1] < src1->f[1] ? src2->f[1] : src3->f[1];
+   dst->f[2] = src0->f[2] < src1->f[2] ? src2->f[2] : src3->f[2];
+   dst->f[3] = src0->f[3] < src1->f[3] ? src2->f[3] : src3->f[3];
+}
+
+static void
+micro_ilt(
+   union tgsi_exec_channel *dst,
+   const union tgsi_exec_channel *src0,
+   const union tgsi_exec_channel *src1,
+   const union tgsi_exec_channel *src2,
+   const union tgsi_exec_channel *src3 )
+{
+   dst->i[0] = src0->i[0] < src1->i[0] ? src2->i[0] : src3->i[0];
+   dst->i[1] = src0->i[1] < src1->i[1] ? src2->i[1] : src3->i[1];
+   dst->i[2] = src0->i[2] < src1->i[2] ? src2->i[2] : src3->i[2];
+   dst->i[3] = src0->i[3] < src1->i[3] ? src2->i[3] : src3->i[3];
+}
+
+static void
+micro_ult(
+   union tgsi_exec_channel *dst,
+   const union tgsi_exec_channel *src0,
+   const union tgsi_exec_channel *src1,
+   const union tgsi_exec_channel *src2,
+   const union tgsi_exec_channel *src3 )
+{
+   dst->u[0] = src0->u[0] < src1->u[0] ? src2->u[0] : src3->u[0];
+   dst->u[1] = src0->u[1] < src1->u[1] ? src2->u[1] : src3->u[1];
+   dst->u[2] = src0->u[2] < src1->u[2] ? src2->u[2] : src3->u[2];
+   dst->u[3] = src0->u[3] < src1->u[3] ? src2->u[3] : src3->u[3];
+}
+
+static void
+micro_max(
+   union tgsi_exec_channel *dst,
+   const union tgsi_exec_channel *src0,
+   const union tgsi_exec_channel *src1 )
+{
+   dst->f[0] = src0->f[0] > src1->f[0] ? src0->f[0] : src1->f[0];
+   dst->f[1] = src0->f[1] > src1->f[1] ? src0->f[1] : src1->f[1];
+   dst->f[2] = src0->f[2] > src1->f[2] ? src0->f[2] : src1->f[2];
+   dst->f[3] = src0->f[3] > src1->f[3] ? src0->f[3] : src1->f[3];
+}
+
+static void
+micro_imax(
+   union tgsi_exec_channel *dst,
+   const union tgsi_exec_channel *src0,
+   const union tgsi_exec_channel *src1 )
+{
+   dst->i[0] = src0->i[0] > src1->i[0] ? src0->i[0] : src1->i[0];
+   dst->i[1] = src0->i[1] > src1->i[1] ? src0->i[1] : src1->i[1];
+   dst->i[2] = src0->i[2] > src1->i[2] ? src0->i[2] : src1->i[2];
+   dst->i[3] = src0->i[3] > src1->i[3] ? src0->i[3] : src1->i[3];
+}
+
+static void
+micro_umax(
+   union tgsi_exec_channel *dst,
+   const union tgsi_exec_channel *src0,
+   const union tgsi_exec_channel *src1 )
+{
+   dst->u[0] = src0->u[0] > src1->u[0] ? src0->u[0] : src1->u[0];
+   dst->u[1] = src0->u[1] > src1->u[1] ? src0->u[1] : src1->u[1];
+   dst->u[2] = src0->u[2] > src1->u[2] ? src0->u[2] : src1->u[2];
+   dst->u[3] = src0->u[3] > src1->u[3] ? src0->u[3] : src1->u[3];
+}
+
+static void
+micro_min(
+   union tgsi_exec_channel *dst,
+   const union tgsi_exec_channel *src0,
+   const union tgsi_exec_channel *src1 )
+{
+   dst->f[0] = src0->f[0] < src1->f[0] ? src0->f[0] : src1->f[0];
+   dst->f[1] = src0->f[1] < src1->f[1] ? src0->f[1] : src1->f[1];
+   dst->f[2] = src0->f[2] < src1->f[2] ? src0->f[2] : src1->f[2];
+   dst->f[3] = src0->f[3] < src1->f[3] ? src0->f[3] : src1->f[3];
+}
+
+static void
+micro_imin(
+   union tgsi_exec_channel *dst,
+   const union tgsi_exec_channel *src0,
+   const union tgsi_exec_channel *src1 )
+{
+   dst->i[0] = src0->i[0] < src1->i[0] ? src0->i[0] : src1->i[0];
+   dst->i[1] = src0->i[1] < src1->i[1] ? src0->i[1] : src1->i[1];
+   dst->i[2] = src0->i[2] < src1->i[2] ? src0->i[2] : src1->i[2];
+   dst->i[3] = src0->i[3] < src1->i[3] ? src0->i[3] : src1->i[3];
+}
+
+static void
+micro_umin(
+   union tgsi_exec_channel *dst,
+   const union tgsi_exec_channel *src0,
+   const union tgsi_exec_channel *src1 )
+{
+   dst->u[0] = src0->u[0] < src1->u[0] ? src0->u[0] : src1->u[0];
+   dst->u[1] = src0->u[1] < src1->u[1] ? src0->u[1] : src1->u[1];
+   dst->u[2] = src0->u[2] < src1->u[2] ? src0->u[2] : src1->u[2];
+   dst->u[3] = src0->u[3] < src1->u[3] ? src0->u[3] : src1->u[3];
+}
+
+static void
+micro_umod(
+   union tgsi_exec_channel *dst,
+   const union tgsi_exec_channel *src0,
+   const union tgsi_exec_channel *src1 )
+{
+   dst->u[0] = src0->u[0] % src1->u[0];
+   dst->u[1] = src0->u[1] % src1->u[1];
+   dst->u[2] = src0->u[2] % src1->u[2];
+   dst->u[3] = src0->u[3] % src1->u[3];
+}
+
+static void
+micro_mul(
+   union tgsi_exec_channel *dst,
+   const union tgsi_exec_channel *src0,
+   const union tgsi_exec_channel *src1 )
+{
+   dst->f[0] = src0->f[0] * src1->f[0];
+   dst->f[1] = src0->f[1] * src1->f[1];
+   dst->f[2] = src0->f[2] * src1->f[2];
+   dst->f[3] = src0->f[3] * src1->f[3];
+}
+
+static void
+micro_imul(
+   union tgsi_exec_channel *dst,
+   const union tgsi_exec_channel *src0,
+   const union tgsi_exec_channel *src1 )
+{
+   dst->i[0] = src0->i[0] * src1->i[0];
+   dst->i[1] = src0->i[1] * src1->i[1];
+   dst->i[2] = src0->i[2] * src1->i[2];
+   dst->i[3] = src0->i[3] * src1->i[3];
+}
+
+static void
+micro_imul64(
+   union tgsi_exec_channel *dst0,
+   union tgsi_exec_channel *dst1,
+   const union tgsi_exec_channel *src0,
+   const union tgsi_exec_channel *src1 )
+{
+   dst1->i[0] = src0->i[0] * src1->i[0];
+   dst1->i[1] = src0->i[1] * src1->i[1];
+   dst1->i[2] = src0->i[2] * src1->i[2];
+   dst1->i[3] = src0->i[3] * src1->i[3];
+   dst0->i[0] = 0;
+   dst0->i[1] = 0;
+   dst0->i[2] = 0;
+   dst0->i[3] = 0;
+}
+
+static void
+micro_umul64(
+   union tgsi_exec_channel *dst0,
+   union tgsi_exec_channel *dst1,
+   const union tgsi_exec_channel *src0,
+   const union tgsi_exec_channel *src1 )
+{
+   dst1->u[0] = src0->u[0] * src1->u[0];
+   dst1->u[1] = src0->u[1] * src1->u[1];
+   dst1->u[2] = src0->u[2] * src1->u[2];
+   dst1->u[3] = src0->u[3] * src1->u[3];
+   dst0->u[0] = 0;
+   dst0->u[1] = 0;
+   dst0->u[2] = 0;
+   dst0->u[3] = 0;
+}
+
+static void
+micro_movc(
+   union tgsi_exec_channel *dst,
+   const union tgsi_exec_channel *src0,
+   const union tgsi_exec_channel *src1,
+   const union tgsi_exec_channel *src2 )
+{
+   dst->u[0] = src0->u[0] ? src1->u[0] : src2->u[0];
+   dst->u[1] = src0->u[1] ? src1->u[1] : src2->u[1];
+   dst->u[2] = src0->u[2] ? src1->u[2] : src2->u[2];
+   dst->u[3] = src0->u[3] ? src1->u[3] : src2->u[3];
+}
+
+static void
+micro_neg(
+   union tgsi_exec_channel *dst,
+   const union tgsi_exec_channel *src )
+{
+   dst->f[0] = -src->f[0];
+   dst->f[1] = -src->f[1];
+   dst->f[2] = -src->f[2];
+   dst->f[3] = -src->f[3];
+}
+
+static void
+micro_ineg(
+   union tgsi_exec_channel *dst,
+   const union tgsi_exec_channel *src )
+{
+   dst->i[0] = -src->i[0];
+   dst->i[1] = -src->i[1];
+   dst->i[2] = -src->i[2];
+   dst->i[3] = -src->i[3];
+}
+
+static void
+micro_not(
+   union tgsi_exec_channel *dst,
+   const union tgsi_exec_channel *src )
+{
+   dst->u[0] = ~src->u[0];
+   dst->u[1] = ~src->u[1];
+   dst->u[2] = ~src->u[2];
+   dst->u[3] = ~src->u[3];
+}
+
+static void
+micro_or(
+   union tgsi_exec_channel *dst,
+   const union tgsi_exec_channel *src0,
+   const union tgsi_exec_channel *src1 )
+{
+   dst->u[0] = src0->u[0] | src1->u[0];
+   dst->u[1] = src0->u[1] | src1->u[1];
+   dst->u[2] = src0->u[2] | src1->u[2];
+   dst->u[3] = src0->u[3] | src1->u[3];
+}
+
+static void
+micro_pow(
+   union tgsi_exec_channel *dst,
+   const union tgsi_exec_channel *src0,
+   const union tgsi_exec_channel *src1 )
+{
+   dst->f[0] = (float) pow( (double) src0->f[0], (double) src1->f[0] );
+   dst->f[1] = (float) pow( (double) src0->f[1], (double) src1->f[1] );
+   dst->f[2] = (float) pow( (double) src0->f[2], (double) src1->f[2] );
+   dst->f[3] = (float) pow( (double) src0->f[3], (double) src1->f[3] );
+}
+
+static void
+micro_rnd(
+   union tgsi_exec_channel *dst,
+   const union tgsi_exec_channel *src )
+{
+   dst->f[0] = (float) floor( (double) (src->f[0] + 0.5f) );
+   dst->f[1] = (float) floor( (double) (src->f[1] + 0.5f) );
+   dst->f[2] = (float) floor( (double) (src->f[2] + 0.5f) );
+   dst->f[3] = (float) floor( (double) (src->f[3] + 0.5f) );
+}
+
+static void
+micro_shl(
+   union tgsi_exec_channel *dst,
+   const union tgsi_exec_channel *src0,
+   const union tgsi_exec_channel *src1 )
+{
+   dst->i[0] = src0->i[0] << src1->i[0];
+   dst->i[1] = src0->i[1] << src1->i[1];
+   dst->i[2] = src0->i[2] << src1->i[2];
+   dst->i[3] = src0->i[3] << src1->i[3];
+}
+
+static void
+micro_ishr(
+   union tgsi_exec_channel *dst,
+   const union tgsi_exec_channel *src0,
+   const union tgsi_exec_channel *src1 )
+{
+   dst->i[0] = src0->i[0] >> src1->i[0];
+   dst->i[1] = src0->i[1] >> src1->i[1];
+   dst->i[2] = src0->i[2] >> src1->i[2];
+   dst->i[3] = src0->i[3] >> src1->i[3];
+}
+
+static void
+micro_trunc(
+   union tgsi_exec_channel *dst,
+   const union tgsi_exec_channel *src0 )
+{
+   dst->f[0] = (float) (int) src0->f[0];
+   dst->f[1] = (float) (int) src0->f[1];
+   dst->f[2] = (float) (int) src0->f[2];
+   dst->f[3] = (float) (int) src0->f[3];
+}
+
+static void
+micro_ushr(
+   union tgsi_exec_channel *dst,
+   const union tgsi_exec_channel *src0,
+   const union tgsi_exec_channel *src1 )
+{
+   dst->u[0] = src0->u[0] >> src1->u[0];
+   dst->u[1] = src0->u[1] >> src1->u[1];
+   dst->u[2] = src0->u[2] >> src1->u[2];
+   dst->u[3] = src0->u[3] >> src1->u[3];
+}
+
+static void
+micro_sin(
+   union tgsi_exec_channel *dst,
+   const union tgsi_exec_channel *src )
+{
+   dst->f[0] = (float) sin( (double) src->f[0] );
+   dst->f[1] = (float) sin( (double) src->f[1] );
+   dst->f[2] = (float) sin( (double) src->f[2] );
+   dst->f[3] = (float) sin( (double) src->f[3] );
+}
+
+static void
+micro_sqrt( union tgsi_exec_channel *dst,
+            const union tgsi_exec_channel *src )
+{
+   dst->f[0] = (float) sqrt( (double) src->f[0] );
+   dst->f[1] = (float) sqrt( (double) src->f[1] );
+   dst->f[2] = (float) sqrt( (double) src->f[2] );
+   dst->f[3] = (float) sqrt( (double) src->f[3] );
+}
+
+static void
+micro_sub(
+   union tgsi_exec_channel *dst,
+   const union tgsi_exec_channel *src0,
+   const union tgsi_exec_channel *src1 )
+{
+   dst->f[0] = src0->f[0] - src1->f[0];
+   dst->f[1] = src0->f[1] - src1->f[1];
+   dst->f[2] = src0->f[2] - src1->f[2];
+   dst->f[3] = src0->f[3] - src1->f[3];
+}
+
+static void
+micro_u2f(
+   union tgsi_exec_channel *dst,
+   const union tgsi_exec_channel *src )
+{
+   dst->f[0] = (float) src->u[0];
+   dst->f[1] = (float) src->u[1];
+   dst->f[2] = (float) src->u[2];
+   dst->f[3] = (float) src->u[3];
+}
+
+static void
+micro_xor(
+   union tgsi_exec_channel *dst,
+   const union tgsi_exec_channel *src0,
+   const union tgsi_exec_channel *src1 )
+{
+   dst->u[0] = src0->u[0] ^ src1->u[0];
+   dst->u[1] = src0->u[1] ^ src1->u[1];
+   dst->u[2] = src0->u[2] ^ src1->u[2];
+   dst->u[3] = src0->u[3] ^ src1->u[3];
+}
+
+static void
+fetch_src_file_channel(
+   const struct tgsi_exec_machine *mach,
+   const uint file,
+   const uint swizzle,
+   const union tgsi_exec_channel *index,
+   union tgsi_exec_channel *chan )
+{
+   switch( swizzle ) {
+   case TGSI_EXTSWIZZLE_X:
+   case TGSI_EXTSWIZZLE_Y:
+   case TGSI_EXTSWIZZLE_Z:
+   case TGSI_EXTSWIZZLE_W:
+      switch( file ) {
+      case TGSI_FILE_CONSTANT:
+         chan->f[0] = mach->Consts[index->i[0]][swizzle];
+         chan->f[1] = mach->Consts[index->i[1]][swizzle];
+         chan->f[2] = mach->Consts[index->i[2]][swizzle];
+         chan->f[3] = mach->Consts[index->i[3]][swizzle];
+         break;
+
+      case TGSI_FILE_INPUT:
+         chan->u[0] = mach->Inputs[index->i[0]].xyzw[swizzle].u[0];
+         chan->u[1] = mach->Inputs[index->i[1]].xyzw[swizzle].u[1];
+         chan->u[2] = mach->Inputs[index->i[2]].xyzw[swizzle].u[2];
+         chan->u[3] = mach->Inputs[index->i[3]].xyzw[swizzle].u[3];
+         break;
+
+      case TGSI_FILE_TEMPORARY:
+         chan->u[0] = mach->Temps[index->i[0]].xyzw[swizzle].u[0];
+         chan->u[1] = mach->Temps[index->i[1]].xyzw[swizzle].u[1];
+         chan->u[2] = mach->Temps[index->i[2]].xyzw[swizzle].u[2];
+         chan->u[3] = mach->Temps[index->i[3]].xyzw[swizzle].u[3];
+         break;
+
+      case TGSI_FILE_IMMEDIATE:
+         assert( index->i[0] < (int) mach->ImmLimit );
+         chan->f[0] = mach->Imms[index->i[0]][swizzle];
+         assert( index->i[1] < (int) mach->ImmLimit );
+         chan->f[1] = mach->Imms[index->i[1]][swizzle];
+         assert( index->i[2] < (int) mach->ImmLimit );
+         chan->f[2] = mach->Imms[index->i[2]][swizzle];
+         assert( index->i[3] < (int) mach->ImmLimit );
+         chan->f[3] = mach->Imms[index->i[3]][swizzle];
+         break;
+
+      case TGSI_FILE_ADDRESS:
+         chan->u[0] = mach->Addrs[index->i[0]].xyzw[swizzle].u[0];
+         chan->u[1] = mach->Addrs[index->i[1]].xyzw[swizzle].u[1];
+         chan->u[2] = mach->Addrs[index->i[2]].xyzw[swizzle].u[2];
+         chan->u[3] = mach->Addrs[index->i[3]].xyzw[swizzle].u[3];
+         break;
+
+      case TGSI_FILE_OUTPUT:
+         /* vertex/fragment output vars can be read too */
+         chan->u[0] = mach->Outputs[index->i[0]].xyzw[swizzle].u[0];
+         chan->u[1] = mach->Outputs[index->i[1]].xyzw[swizzle].u[1];
+         chan->u[2] = mach->Outputs[index->i[2]].xyzw[swizzle].u[2];
+         chan->u[3] = mach->Outputs[index->i[3]].xyzw[swizzle].u[3];
+         break;
+
+      default:
+         assert( 0 );
+      }
+      break;
+
+   case TGSI_EXTSWIZZLE_ZERO:
+      *chan = mach->Temps[TEMP_0_I].xyzw[TEMP_0_C];
+      break;
+
+   case TGSI_EXTSWIZZLE_ONE:
+      *chan = mach->Temps[TEMP_1_I].xyzw[TEMP_1_C];
+      break;
+
+   default:
+      assert( 0 );
+   }
+}
+
+static void
+fetch_source(
+   const struct tgsi_exec_machine *mach,
+   union tgsi_exec_channel *chan,
+   const struct tgsi_full_src_register *reg,
+   const uint chan_index )
+{
+   union tgsi_exec_channel index;
+   uint swizzle;
+
+   index.i[0] =
+   index.i[1] =
+   index.i[2] =
+   index.i[3] = reg->SrcRegister.Index;
+
+   if (reg->SrcRegister.Indirect) {
+      union tgsi_exec_channel index2;
+      union tgsi_exec_channel indir_index;
+
+      index2.i[0] =
+      index2.i[1] =
+      index2.i[2] =
+      index2.i[3] = reg->SrcRegisterInd.Index;
+
+      swizzle = tgsi_util_get_src_register_swizzle( &reg->SrcRegisterInd, CHAN_X );
+      fetch_src_file_channel(
+         mach,
+         reg->SrcRegisterInd.File,
+         swizzle,
+         &index2,
+         &indir_index );
+
+      index.i[0] += indir_index.i[0];
+      index.i[1] += indir_index.i[1];
+      index.i[2] += indir_index.i[2];
+      index.i[3] += indir_index.i[3];
+   }
+
+   if( reg->SrcRegister.Dimension ) {
+      switch( reg->SrcRegister.File ) {
+      case TGSI_FILE_INPUT:
+         index.i[0] *= 17;
+         index.i[1] *= 17;
+         index.i[2] *= 17;
+         index.i[3] *= 17;
+         break;
+      case TGSI_FILE_CONSTANT:
+         index.i[0] *= 4096;
+         index.i[1] *= 4096;
+         index.i[2] *= 4096;
+         index.i[3] *= 4096;
+         break;
+      default:
+         assert( 0 );
+      }
+
+      index.i[0] += reg->SrcRegisterDim.Index;
+      index.i[1] += reg->SrcRegisterDim.Index;
+      index.i[2] += reg->SrcRegisterDim.Index;
+      index.i[3] += reg->SrcRegisterDim.Index;
+
+      if (reg->SrcRegisterDim.Indirect) {
+         union tgsi_exec_channel index2;
+         union tgsi_exec_channel indir_index;
+
+         index2.i[0] =
+         index2.i[1] =
+         index2.i[2] =
+         index2.i[3] = reg->SrcRegisterDimInd.Index;
+
+         swizzle = tgsi_util_get_src_register_swizzle( &reg->SrcRegisterDimInd, CHAN_X );
+         fetch_src_file_channel(
+            mach,
+            reg->SrcRegisterDimInd.File,
+            swizzle,
+            &index2,
+            &indir_index );
+
+         index.i[0] += indir_index.i[0];
+         index.i[1] += indir_index.i[1];
+         index.i[2] += indir_index.i[2];
+         index.i[3] += indir_index.i[3];
+      }
+   }
+
+   swizzle = tgsi_util_get_full_src_register_extswizzle( reg, chan_index );
+   fetch_src_file_channel(
+      mach,
+      reg->SrcRegister.File,
+      swizzle,
+      &index,
+      chan );
+
+   switch (tgsi_util_get_full_src_register_sign_mode( reg, chan_index )) {
+   case TGSI_UTIL_SIGN_CLEAR:
+      micro_abs( chan, chan );
+      break;
+
+   case TGSI_UTIL_SIGN_SET:
+      micro_abs( chan, chan );
+      micro_neg( chan, chan );
+      break;
+
+   case TGSI_UTIL_SIGN_TOGGLE:
+      micro_neg( chan, chan );
+      break;
+
+   case TGSI_UTIL_SIGN_KEEP:
+      break;
+   }
+
+   if (reg->SrcRegisterExtMod.Complement) {
+      micro_sub( chan, &mach->Temps[TEMP_1_I].xyzw[TEMP_1_C], chan );
+   }
+}
+
+static void
+store_dest(
+   struct tgsi_exec_machine *mach,
+   const union tgsi_exec_channel *chan,
+   const struct tgsi_full_dst_register *reg,
+   const struct tgsi_full_instruction *inst,
+   uint chan_index )
+{
+   union tgsi_exec_channel *dst;
+
+   switch( reg->DstRegister.File ) {
+   case TGSI_FILE_NULL:
+      return;
+
+   case TGSI_FILE_OUTPUT:
+      dst = &mach->Outputs[mach->Temps[TEMP_OUTPUT_I].xyzw[TEMP_OUTPUT_C].u[0]
+                           + reg->DstRegister.Index].xyzw[chan_index];
+      break;
+
+   case TGSI_FILE_TEMPORARY:
+      dst = &mach->Temps[reg->DstRegister.Index].xyzw[chan_index];
+      break;
+
+   case TGSI_FILE_ADDRESS:
+      dst = &mach->Addrs[reg->DstRegister.Index].xyzw[chan_index];
+      break;
+
+   default:
+      assert( 0 );
+      return;
+   }
+
+   switch (inst->Instruction.Saturate)
+   {
+   case TGSI_SAT_NONE:
+      if (mach->ExecMask & 0x1)
+         dst->i[0] = chan->i[0];
+      if (mach->ExecMask & 0x2)
+         dst->i[1] = chan->i[1];
+      if (mach->ExecMask & 0x4)
+         dst->i[2] = chan->i[2];
+      if (mach->ExecMask & 0x8)
+         dst->i[3] = chan->i[3];
+      break;
+
+   case TGSI_SAT_ZERO_ONE:
+      /* XXX need to obey ExecMask here */
+      micro_max(dst, chan, &mach->Temps[TEMP_0_I].xyzw[TEMP_0_C]);
+      micro_min(dst, dst, &mach->Temps[TEMP_1_I].xyzw[TEMP_1_C]);
+      break;
+
+   case TGSI_SAT_MINUS_PLUS_ONE:
+      assert( 0 );
+      break;
+
+   default:
+      assert( 0 );
+   }
+}
+
+#define FETCH(VAL,INDEX,CHAN)\
+    fetch_source (mach, VAL, &inst->FullSrcRegisters[INDEX], CHAN)
+
+#define STORE(VAL,INDEX,CHAN)\
+    store_dest (mach, VAL, &inst->FullDstRegisters[INDEX], inst, CHAN )
+
+
+/**
+ * Execute ARB-style KIL which is predicated by a src register.
+ * Kill fragment if any of the four values is less than zero.
+ */
+static void
+exec_kilp(struct tgsi_exec_machine *mach,
+          const struct tgsi_full_instruction *inst)
+{
+   uint uniquemask;
+   uint chan_index;
+   uint kilmask = 0; /* bit 0 = pixel 0, bit 1 = pixel 1, etc */
+   union tgsi_exec_channel r[1];
+
+   /* This mask stores component bits that were already tested. Note that
+    * we test if the value is less than zero, so 1.0 and 0.0 need not to be
+    * tested. */
+   uniquemask = (1 << TGSI_EXTSWIZZLE_ZERO) | (1 << TGSI_EXTSWIZZLE_ONE);
+
+   for (chan_index = 0; chan_index < 4; chan_index++)
+   {
+      uint swizzle;
+      uint i;
+
+      /* unswizzle channel */
+      swizzle = tgsi_util_get_full_src_register_extswizzle (
+                        &inst->FullSrcRegisters[0],
+                        chan_index);
+
+      /* check if the component has not been already tested */
+      if (uniquemask & (1 << swizzle))
+         continue;
+      uniquemask |= 1 << swizzle;
+
+      FETCH(&r[0], 0, chan_index);
+      for (i = 0; i < 4; i++)
+         if (r[0].f[i] < 0.0f)
+            kilmask |= 1 << i;
+   }
+
+   mach->Temps[TEMP_KILMASK_I].xyzw[TEMP_KILMASK_C].u[0] |= kilmask;
+}
+
+
+/*
+ * Fetch a texel using STR texture coordinates.
+ */
+static void
+fetch_texel( struct tgsi_sampler *sampler,
+             const union tgsi_exec_channel *s,
+             const union tgsi_exec_channel *t,
+             const union tgsi_exec_channel *p,
+             float lodbias,  /* XXX should be float[4] */
+             union tgsi_exec_channel *r,
+             union tgsi_exec_channel *g,
+             union tgsi_exec_channel *b,
+             union tgsi_exec_channel *a )
+{
+   uint j;
+   float rgba[NUM_CHANNELS][QUAD_SIZE];
+
+   sampler->get_samples(sampler, s->f, t->f, p->f, lodbias, rgba);
+
+   for (j = 0; j < 4; j++) {
+      r->f[j] = rgba[0][j];
+      g->f[j] = rgba[1][j];
+      b->f[j] = rgba[2][j];
+      a->f[j] = rgba[3][j];
+   }
+}
+
+
+static void
+exec_tex(struct tgsi_exec_machine *mach,
+         const struct tgsi_full_instruction *inst,
+         boolean biasLod)
+{
+   const uint unit = inst->FullSrcRegisters[1].SrcRegister.Index;
+   union tgsi_exec_channel r[8];
+   uint chan_index;
+   float lodBias;
+
+   /*   debug_printf("Sampler %u unit %u\n", sampler, unit); */
+
+   switch (inst->InstructionExtTexture.Texture) {
+   case TGSI_TEXTURE_1D:
+
+      FETCH(&r[0], 0, CHAN_X);
+
+      switch (inst->FullSrcRegisters[0].SrcRegisterExtSwz.ExtDivide) {
+      case TGSI_EXTSWIZZLE_W:
+         FETCH(&r[1], 0, CHAN_W);
+         micro_div( &r[0], &r[0], &r[1] );
+         break;
+
+      case TGSI_EXTSWIZZLE_ONE:
+         break;
+
+      default:
+         assert (0);
+      }
+
+      if (biasLod) {
+         FETCH(&r[1], 0, CHAN_W);
+         lodBias = r[2].f[0];
+      }
+      else
+         lodBias = 0.0;
+
+      fetch_texel(&mach->Samplers[unit],
+                  &r[0], NULL, NULL, lodBias,  /* S, T, P, BIAS */
+                  &r[0], &r[1], &r[2], &r[3]); /* R, G, B, A */
+      break;
+
+   case TGSI_TEXTURE_2D:
+   case TGSI_TEXTURE_RECT:
+
+      FETCH(&r[0], 0, CHAN_X);
+      FETCH(&r[1], 0, CHAN_Y);
+      FETCH(&r[2], 0, CHAN_Z);
+
+      switch (inst->FullSrcRegisters[0].SrcRegisterExtSwz.ExtDivide) {
+      case TGSI_EXTSWIZZLE_W:
+         FETCH(&r[3], 0, CHAN_W);
+         micro_div( &r[0], &r[0], &r[3] );
+         micro_div( &r[1], &r[1], &r[3] );
+         micro_div( &r[2], &r[2], &r[3] );
+         break;
+
+      case TGSI_EXTSWIZZLE_ONE:
+         break;
+
+      default:
+         assert (0);
+      }
+
+      if (biasLod) {
+         FETCH(&r[3], 0, CHAN_W);
+         lodBias = r[3].f[0];
+      }
+      else
+         lodBias = 0.0;
+
+      fetch_texel(&mach->Samplers[unit],
+                  &r[0], &r[1], &r[2], lodBias,  /* inputs */
+                  &r[0], &r[1], &r[2], &r[3]);  /* outputs */
+      break;
+
+   case TGSI_TEXTURE_3D:
+   case TGSI_TEXTURE_CUBE:
+
+      FETCH(&r[0], 0, CHAN_X);
+      FETCH(&r[1], 0, CHAN_Y);
+      FETCH(&r[2], 0, CHAN_Z);
+
+      switch (inst->FullSrcRegisters[0].SrcRegisterExtSwz.ExtDivide) {
+      case TGSI_EXTSWIZZLE_W:
+         FETCH(&r[3], 0, CHAN_W);
+         micro_div( &r[0], &r[0], &r[3] );
+         micro_div( &r[1], &r[1], &r[3] );
+         micro_div( &r[2], &r[2], &r[3] );
+         break;
+
+      case TGSI_EXTSWIZZLE_ONE:
+         break;
+
+      default:
+         assert (0);
+      }
+
+      if (biasLod) {
+         FETCH(&r[3], 0, CHAN_W);
+         lodBias = r[3].f[0];
+      }
+      else
+         lodBias = 0.0;
+
+      fetch_texel(&mach->Samplers[unit],
+                  &r[0], &r[1], &r[2], lodBias,
+                  &r[0], &r[1], &r[2], &r[3]);
+      break;
+
+   default:
+      assert (0);
+   }
+
+   FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
+      STORE( &r[chan_index], 0, chan_index );
+   }
+}
+
+
+/**
+ * Evaluate a constant-valued coefficient at the position of the
+ * current quad.
+ */
+static void
+eval_constant_coef(
+   struct tgsi_exec_machine *mach,
+   unsigned attrib,
+   unsigned chan )
+{
+   unsigned i;
+
+   for( i = 0; i < QUAD_SIZE; i++ ) {
+      mach->Inputs[attrib].xyzw[chan].f[i] = mach->InterpCoefs[attrib].a0[chan];
+   }
+}
+
+/**
+ * Evaluate a linear-valued coefficient at the position of the
+ * current quad.
+ */
+static void
+eval_linear_coef(
+   struct tgsi_exec_machine *mach,
+   unsigned attrib,
+   unsigned chan )
+{
+   const float x = mach->QuadPos.xyzw[0].f[0];
+   const float y = mach->QuadPos.xyzw[1].f[0];
+   const float dadx = mach->InterpCoefs[attrib].dadx[chan];
+   const float dady = mach->InterpCoefs[attrib].dady[chan];
+   const float a0 = mach->InterpCoefs[attrib].a0[chan] + dadx * x + dady * y;
+   mach->Inputs[attrib].xyzw[chan].f[0] = a0;
+   mach->Inputs[attrib].xyzw[chan].f[1] = a0 + dadx;
+   mach->Inputs[attrib].xyzw[chan].f[2] = a0 + dady;
+   mach->Inputs[attrib].xyzw[chan].f[3] = a0 + dadx + dady;
+}
+
+/**
+ * Evaluate a perspective-valued coefficient at the position of the
+ * current quad.
+ */
+static void
+eval_perspective_coef(
+   struct tgsi_exec_machine *mach,
+   unsigned attrib,
+   unsigned chan )
+{
+   const float x = mach->QuadPos.xyzw[0].f[0];
+   const float y = mach->QuadPos.xyzw[1].f[0];
+   const float dadx = mach->InterpCoefs[attrib].dadx[chan];
+   const float dady = mach->InterpCoefs[attrib].dady[chan];
+   const float a0 = mach->InterpCoefs[attrib].a0[chan] + dadx * x + dady * y;
+   const float *w = mach->QuadPos.xyzw[3].f;
+   /* divide by W here */
+   mach->Inputs[attrib].xyzw[chan].f[0] = a0 / w[0];
+   mach->Inputs[attrib].xyzw[chan].f[1] = (a0 + dadx) / w[1];
+   mach->Inputs[attrib].xyzw[chan].f[2] = (a0 + dady) / w[2];
+   mach->Inputs[attrib].xyzw[chan].f[3] = (a0 + dadx + dady) / w[3];
+}
+
+
+typedef void (* eval_coef_func)(
+   struct tgsi_exec_machine *mach,
+   unsigned attrib,
+   unsigned chan );
+
+static void
+exec_declaration(
+   struct tgsi_exec_machine *mach,
+   const struct tgsi_full_declaration *decl )
+{
+   if( mach->Processor == TGSI_PROCESSOR_FRAGMENT ) {
+      if( decl->Declaration.File == TGSI_FILE_INPUT ) {
+         unsigned first, last, mask;
+         eval_coef_func eval;
+
+         assert( decl->Declaration.Declare == TGSI_DECLARE_RANGE );
+
+         first = decl->u.DeclarationRange.First;
+         last = decl->u.DeclarationRange.Last;
+         mask = decl->Declaration.UsageMask;
+
+         switch( decl->Interpolation.Interpolate ) {
+         case TGSI_INTERPOLATE_CONSTANT:
+            eval = eval_constant_coef;
+            break;
+
+         case TGSI_INTERPOLATE_LINEAR:
+            eval = eval_linear_coef;
+            break;
+
+         case TGSI_INTERPOLATE_PERSPECTIVE:
+            eval = eval_perspective_coef;
+            break;
+
+         default:
+            assert( 0 );
+         }
+
+         if( mask == TGSI_WRITEMASK_XYZW ) {
+            unsigned i, j;
+
+            for( i = first; i <= last; i++ ) {
+               for( j = 0; j < NUM_CHANNELS; j++ ) {
+                  eval( mach, i, j );
+               }
+            }
+         }
+         else {
+            unsigned i, j;
+
+            for( j = 0; j < NUM_CHANNELS; j++ ) {
+               if( mask & (1 << j) ) {
+                  for( i = first; i <= last; i++ ) {
+                     eval( mach, i, j );
+                  }
+               }
+            }
+         }
+      }
+   }
+}
+
+static void
+exec_instruction(
+   struct tgsi_exec_machine *mach,
+   const struct tgsi_full_instruction *inst,
+   int *pc )
+{
+   uint chan_index;
+   union tgsi_exec_channel r[8];
+
+   (*pc)++;
+
+   switch (inst->Instruction.Opcode) {
+   case TGSI_OPCODE_ARL:
+      FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
+	 FETCH( &r[0], 0, chan_index );
+	 micro_f2it( &r[0], &r[0] );
+	 STORE( &r[0], 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_MOV:
+   /* TGSI_OPCODE_SWZ */
+      FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
+         FETCH( &r[0], 0, chan_index );
+         STORE( &r[0], 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_LIT:
+      if (IS_CHANNEL_ENABLED( *inst, CHAN_X )) {
+	 STORE( &mach->Temps[TEMP_1_I].xyzw[TEMP_1_C], 0, CHAN_X );
+      }
+
+      if (IS_CHANNEL_ENABLED( *inst, CHAN_Y ) || IS_CHANNEL_ENABLED( *inst, CHAN_Z )) {
+	 FETCH( &r[0], 0, CHAN_X );
+	 if (IS_CHANNEL_ENABLED( *inst, CHAN_Y )) {
+	    micro_max( &r[0], &r[0], &mach->Temps[TEMP_0_I].xyzw[TEMP_0_C] );
+	    STORE( &r[0], 0, CHAN_Y );
+	 }
+
+	 if (IS_CHANNEL_ENABLED( *inst, CHAN_Z )) {
+	    FETCH( &r[1], 0, CHAN_Y );
+	    micro_max( &r[1], &r[1], &mach->Temps[TEMP_0_I].xyzw[TEMP_0_C] );
+
+	    FETCH( &r[2], 0, CHAN_W );
+	    micro_min( &r[2], &r[2], &mach->Temps[TEMP_128_I].xyzw[TEMP_128_C] );
+	    micro_max( &r[2], &r[2], &mach->Temps[TEMP_M128_I].xyzw[TEMP_M128_C] );
+	    micro_pow( &r[1], &r[1], &r[2] );
+	    micro_lt( &r[0], &mach->Temps[TEMP_0_I].xyzw[TEMP_0_C], &r[0], &r[1], &mach->Temps[TEMP_0_I].xyzw[TEMP_0_C] );
+	    STORE( &r[0], 0, CHAN_Z );
+	 }
+      }
+
+      if (IS_CHANNEL_ENABLED( *inst, CHAN_W )) {
+	 STORE( &mach->Temps[TEMP_1_I].xyzw[TEMP_1_C], 0, CHAN_W );
+      }
+      break;
+
+   case TGSI_OPCODE_RCP:
+   /* TGSI_OPCODE_RECIP */
+      FETCH( &r[0], 0, CHAN_X );
+      micro_div( &r[0], &mach->Temps[TEMP_1_I].xyzw[TEMP_1_C], &r[0] );
+      FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
+	 STORE( &r[0], 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_RSQ:
+   /* TGSI_OPCODE_RECIPSQRT */
+      FETCH( &r[0], 0, CHAN_X );
+      micro_sqrt( &r[0], &r[0] );
+      micro_div( &r[0], &mach->Temps[TEMP_1_I].xyzw[TEMP_1_C], &r[0] );
+      FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
+	 STORE( &r[0], 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_EXP:
+      assert (0);
+      break;
+
+   case TGSI_OPCODE_LOG:
+      assert (0);
+      break;
+
+   case TGSI_OPCODE_MUL:
+      FOR_EACH_ENABLED_CHANNEL( *inst, chan_index )
+      {
+         FETCH(&r[0], 0, chan_index);
+         FETCH(&r[1], 1, chan_index);
+
+         micro_mul( &r[0], &r[0], &r[1] );
+
+         STORE(&r[0], 0, chan_index);
+      }
+      break;
+
+   case TGSI_OPCODE_ADD:
+      FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
+         FETCH( &r[0], 0, chan_index );
+         FETCH( &r[1], 1, chan_index );
+         micro_add( &r[0], &r[0], &r[1] );
+         STORE( &r[0], 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_DP3:
+   /* TGSI_OPCODE_DOT3 */
+      FETCH( &r[0], 0, CHAN_X );
+      FETCH( &r[1], 1, CHAN_X );
+      micro_mul( &r[0], &r[0], &r[1] );
+
+      FETCH( &r[1], 0, CHAN_Y );
+      FETCH( &r[2], 1, CHAN_Y );
+      micro_mul( &r[1], &r[1], &r[2] );
+      micro_add( &r[0], &r[0], &r[1] );
+
+      FETCH( &r[1], 0, CHAN_Z );
+      FETCH( &r[2], 1, CHAN_Z );
+      micro_mul( &r[1], &r[1], &r[2] );
+      micro_add( &r[0], &r[0], &r[1] );
+
+      FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
+         STORE( &r[0], 0, chan_index );
+      }
+      break;
+
+    case TGSI_OPCODE_DP4:
+    /* TGSI_OPCODE_DOT4 */
+       FETCH(&r[0], 0, CHAN_X);
+       FETCH(&r[1], 1, CHAN_X);
+
+       micro_mul( &r[0], &r[0], &r[1] );
+
+       FETCH(&r[1], 0, CHAN_Y);
+       FETCH(&r[2], 1, CHAN_Y);
+
+       micro_mul( &r[1], &r[1], &r[2] );
+       micro_add( &r[0], &r[0], &r[1] );
+
+       FETCH(&r[1], 0, CHAN_Z);
+       FETCH(&r[2], 1, CHAN_Z);
+
+       micro_mul( &r[1], &r[1], &r[2] );
+       micro_add( &r[0], &r[0], &r[1] );
+
+       FETCH(&r[1], 0, CHAN_W);
+       FETCH(&r[2], 1, CHAN_W);
+
+       micro_mul( &r[1], &r[1], &r[2] );
+       micro_add( &r[0], &r[0], &r[1] );
+
+      FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
+	 STORE( &r[0], 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_DST:
+      if (IS_CHANNEL_ENABLED( *inst, CHAN_X )) {
+	 STORE( &mach->Temps[TEMP_1_I].xyzw[TEMP_1_C], 0, CHAN_X );
+      }
+
+      if (IS_CHANNEL_ENABLED( *inst, CHAN_Y )) {
+	 FETCH( &r[0], 0, CHAN_Y );
+	 FETCH( &r[1], 1, CHAN_Y);
+	 micro_mul( &r[0], &r[0], &r[1] );
+	 STORE( &r[0], 0, CHAN_Y );
+      }
+
+      if (IS_CHANNEL_ENABLED( *inst, CHAN_Z )) {
+	 FETCH( &r[0], 0, CHAN_Z );
+	 STORE( &r[0], 0, CHAN_Z );
+      }
+
+      if (IS_CHANNEL_ENABLED( *inst, CHAN_W )) {
+	 FETCH( &r[0], 1, CHAN_W );
+	 STORE( &r[0], 0, CHAN_W );
+      }
+      break;
+
+   case TGSI_OPCODE_MIN:
+      FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
+         FETCH(&r[0], 0, chan_index);
+         FETCH(&r[1], 1, chan_index);
+
+         /* XXX use micro_min()?? */
+         micro_lt( &r[0], &r[0], &r[1], &r[0], &r[1] );
+
+         STORE(&r[0], 0, chan_index);
+      }
+      break;
+
+   case TGSI_OPCODE_MAX:
+      FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
+         FETCH(&r[0], 0, chan_index);
+         FETCH(&r[1], 1, chan_index);
+
+         /* XXX use micro_max()?? */
+         micro_lt( &r[0], &r[0], &r[1], &r[1], &r[0] );
+
+         STORE(&r[0], 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_SLT:
+   /* TGSI_OPCODE_SETLT */
+      FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
+         FETCH( &r[0], 0, chan_index );
+         FETCH( &r[1], 1, chan_index );
+         micro_lt( &r[0], &r[0], &r[1], &mach->Temps[TEMP_1_I].xyzw[TEMP_1_C], &mach->Temps[TEMP_0_I].xyzw[TEMP_0_C] );
+         STORE( &r[0], 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_SGE:
+   /* TGSI_OPCODE_SETGE */
+      FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
+         FETCH( &r[0], 0, chan_index );
+         FETCH( &r[1], 1, chan_index );
+         micro_ge( &r[0], &r[0], &r[1], &mach->Temps[TEMP_1_I].xyzw[TEMP_1_C], &mach->Temps[TEMP_0_I].xyzw[TEMP_0_C] );
+         STORE( &r[0], 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_MAD:
+   /* TGSI_OPCODE_MADD */
+      FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
+         FETCH( &r[0], 0, chan_index );
+         FETCH( &r[1], 1, chan_index );
+         micro_mul( &r[0], &r[0], &r[1] );
+         FETCH( &r[1], 2, chan_index );
+         micro_add( &r[0], &r[0], &r[1] );
+         STORE( &r[0], 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_SUB:
+      FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
+         FETCH(&r[0], 0, chan_index);
+         FETCH(&r[1], 1, chan_index);
+
+         micro_sub( &r[0], &r[0], &r[1] );
+
+         STORE(&r[0], 0, chan_index);
+      }
+      break;
+
+   case TGSI_OPCODE_LERP:
+   /* TGSI_OPCODE_LRP */
+      FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
+         FETCH(&r[0], 0, chan_index);
+         FETCH(&r[1], 1, chan_index);
+         FETCH(&r[2], 2, chan_index);
+
+         micro_sub( &r[1], &r[1], &r[2] );
+         micro_mul( &r[0], &r[0], &r[1] );
+         micro_add( &r[0], &r[0], &r[2] );
+
+         STORE(&r[0], 0, chan_index);
+      }
+      break;
+
+   case TGSI_OPCODE_CND:
+      assert (0);
+      break;
+
+   case TGSI_OPCODE_CND0:
+      assert (0);
+      break;
+
+   case TGSI_OPCODE_DOT2ADD:
+      /* TGSI_OPCODE_DP2A */
+      assert (0);
+      break;
+
+   case TGSI_OPCODE_INDEX:
+      assert (0);
+      break;
+
+   case TGSI_OPCODE_NEGATE:
+      assert (0);
+      break;
+
+   case TGSI_OPCODE_FRAC:
+   /* TGSI_OPCODE_FRC */
+      FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
+         FETCH( &r[0], 0, chan_index );
+         micro_frc( &r[0], &r[0] );
+         STORE( &r[0], 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_CLAMP:
+      assert (0);
+      break;
+
+   case TGSI_OPCODE_FLOOR:
+   /* TGSI_OPCODE_FLR */
+      FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
+         FETCH( &r[0], 0, chan_index );
+         micro_flr( &r[0], &r[0] );
+         STORE( &r[0], 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_ROUND:
+      FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
+         FETCH( &r[0], 0, chan_index );
+         micro_rnd( &r[0], &r[0] );
+         STORE( &r[0], 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_EXPBASE2:
+    /* TGSI_OPCODE_EX2 */
+      FETCH(&r[0], 0, CHAN_X);
+
+      micro_pow( &r[0], &mach->Temps[TEMP_2_I].xyzw[TEMP_2_C], &r[0] );
+
+      FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
+	 STORE( &r[0], 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_LOGBASE2:
+   /* TGSI_OPCODE_LG2 */
+      FETCH( &r[0], 0, CHAN_X );
+      micro_lg2( &r[0], &r[0] );
+      FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
+         STORE( &r[0], 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_POWER:
+      /* TGSI_OPCODE_POW */
+      FETCH(&r[0], 0, CHAN_X);
+      FETCH(&r[1], 1, CHAN_X);
+
+      micro_pow( &r[0], &r[0], &r[1] );
+
+      FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
+	 STORE( &r[0], 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_CROSSPRODUCT:
+      /* TGSI_OPCODE_XPD */
+      FETCH(&r[0], 0, CHAN_Y);
+      FETCH(&r[1], 1, CHAN_Z);
+
+      micro_mul( &r[2], &r[0], &r[1] );
+
+      FETCH(&r[3], 0, CHAN_Z);
+      FETCH(&r[4], 1, CHAN_Y);
+
+      micro_mul( &r[5], &r[3], &r[4] );
+      micro_sub( &r[2], &r[2], &r[5] );
+
+      if (IS_CHANNEL_ENABLED( *inst, CHAN_X )) {
+         STORE( &r[2], 0, CHAN_X );
+      }
+
+      FETCH(&r[2], 1, CHAN_X);
+
+      micro_mul( &r[3], &r[3], &r[2] );
+
+      FETCH(&r[5], 0, CHAN_X);
+
+      micro_mul( &r[1], &r[1], &r[5] );
+      micro_sub( &r[3], &r[3], &r[1] );
+
+      if (IS_CHANNEL_ENABLED( *inst, CHAN_Y )) {
+         STORE( &r[3], 0, CHAN_Y );
+      }
+
+      micro_mul( &r[5], &r[5], &r[4] );
+      micro_mul( &r[0], &r[0], &r[2] );
+      micro_sub( &r[5], &r[5], &r[0] );
+
+      if (IS_CHANNEL_ENABLED( *inst, CHAN_Z )) {
+         STORE( &r[5], 0, CHAN_Z );
+      }
+
+      if (IS_CHANNEL_ENABLED( *inst, CHAN_W )) {
+         STORE( &mach->Temps[TEMP_1_I].xyzw[TEMP_1_C], 0, CHAN_W );
+      }
+      break;
+
+    case TGSI_OPCODE_MULTIPLYMATRIX:
+       assert (0);
+       break;
+
+    case TGSI_OPCODE_ABS:
+       FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
+          FETCH(&r[0], 0, chan_index);
+
+          micro_abs( &r[0], &r[0] );
+
+          STORE(&r[0], 0, chan_index);
+       }
+       break;
+
+   case TGSI_OPCODE_RCC:
+      assert (0);
+      break;
+
+   case TGSI_OPCODE_DPH:
+      FETCH(&r[0], 0, CHAN_X);
+      FETCH(&r[1], 1, CHAN_X);
+
+      micro_mul( &r[0], &r[0], &r[1] );
+
+      FETCH(&r[1], 0, CHAN_Y);
+      FETCH(&r[2], 1, CHAN_Y);
+
+      micro_mul( &r[1], &r[1], &r[2] );
+      micro_add( &r[0], &r[0], &r[1] );
+
+      FETCH(&r[1], 0, CHAN_Z);
+      FETCH(&r[2], 1, CHAN_Z);
+
+      micro_mul( &r[1], &r[1], &r[2] );
+      micro_add( &r[0], &r[0], &r[1] );
+
+      FETCH(&r[1], 1, CHAN_W);
+
+      micro_add( &r[0], &r[0], &r[1] );
+
+      FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
+	 STORE( &r[0], 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_COS:
+      FETCH(&r[0], 0, CHAN_X);
+
+      micro_cos( &r[0], &r[0] );
+
+      FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
+	 STORE( &r[0], 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_DDX:
+      FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
+         FETCH( &r[0], 0, chan_index );
+         micro_ddx( &r[0], &r[0] );
+         STORE( &r[0], 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_DDY:
+      FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
+         FETCH( &r[0], 0, chan_index );
+         micro_ddy( &r[0], &r[0] );
+         STORE( &r[0], 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_KILP:
+      exec_kilp (mach, inst);
+      break;
+
+   case TGSI_OPCODE_KIL:
+      /* for enabled ExecMask bits, set the killed bit */
+      mach->Temps[TEMP_KILMASK_I].xyzw[TEMP_KILMASK_C].u[0] |= mach->ExecMask;
+      break;
+
+   case TGSI_OPCODE_PK2H:
+      assert (0);
+      break;
+
+   case TGSI_OPCODE_PK2US:
+      assert (0);
+      break;
+
+   case TGSI_OPCODE_PK4B:
+      assert (0);
+      break;
+
+   case TGSI_OPCODE_PK4UB:
+      assert (0);
+      break;
+
+   case TGSI_OPCODE_RFL:
+      assert (0);
+      break;
+
+   case TGSI_OPCODE_SEQ:
+      FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
+         FETCH( &r[0], 0, chan_index );
+         FETCH( &r[1], 1, chan_index );
+         micro_eq( &r[0], &r[0], &r[1],
+                   &mach->Temps[TEMP_1_I].xyzw[TEMP_1_C],
+                   &mach->Temps[TEMP_0_I].xyzw[TEMP_0_C] );
+         STORE( &r[0], 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_SFL:
+      assert (0);
+      break;
+
+   case TGSI_OPCODE_SGT:
+      FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
+         FETCH( &r[0], 0, chan_index );
+         FETCH( &r[1], 1, chan_index );
+         micro_lt( &r[0], &r[0], &r[1], &mach->Temps[TEMP_0_I].xyzw[TEMP_0_C], &mach->Temps[TEMP_1_I].xyzw[TEMP_1_C] );
+         STORE( &r[0], 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_SIN:
+      FETCH( &r[0], 0, CHAN_X );
+      micro_sin( &r[0], &r[0] );
+      FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
+         STORE( &r[0], 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_SLE:
+      FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
+         FETCH( &r[0], 0, chan_index );
+         FETCH( &r[1], 1, chan_index );
+         micro_ge( &r[0], &r[0], &r[1], &mach->Temps[TEMP_0_I].xyzw[TEMP_0_C], &mach->Temps[TEMP_1_I].xyzw[TEMP_1_C] );
+         STORE( &r[0], 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_SNE:
+      FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
+         FETCH( &r[0], 0, chan_index );
+         FETCH( &r[1], 1, chan_index );
+         micro_eq( &r[0], &r[0], &r[1], &mach->Temps[TEMP_0_I].xyzw[TEMP_0_C], &mach->Temps[TEMP_1_I].xyzw[TEMP_1_C] );
+         STORE( &r[0], 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_STR:
+      assert (0);
+      break;
+
+   case TGSI_OPCODE_TEX:
+      /* simple texture lookup */
+      /* src[0] = texcoord */
+      /* src[1] = sampler unit */
+      exec_tex(mach, inst, FALSE);
+      break;
+
+   case TGSI_OPCODE_TXB:
+      /* Texture lookup with lod bias */
+      /* src[0] = texcoord (src[0].w = LOD bias) */
+      /* src[1] = sampler unit */
+      exec_tex(mach, inst, TRUE);
+      break;
+
+   case TGSI_OPCODE_TXD:
+      /* Texture lookup with explict partial derivatives */
+      /* src[0] = texcoord */
+      /* src[1] = d[strq]/dx */
+      /* src[2] = d[strq]/dy */
+      /* src[3] = sampler unit */
+      assert (0);
+      break;
+
+   case TGSI_OPCODE_TXL:
+      /* Texture lookup with explit LOD */
+      /* src[0] = texcoord (src[0].w = LOD) */
+      /* src[1] = sampler unit */
+      exec_tex(mach, inst, TRUE);
+      break;
+
+   case TGSI_OPCODE_UP2H:
+      assert (0);
+      break;
+
+   case TGSI_OPCODE_UP2US:
+      assert (0);
+      break;
+
+   case TGSI_OPCODE_UP4B:
+      assert (0);
+      break;
+
+   case TGSI_OPCODE_UP4UB:
+      assert (0);
+      break;
+
+   case TGSI_OPCODE_X2D:
+      assert (0);
+      break;
+
+   case TGSI_OPCODE_ARA:
+      assert (0);
+      break;
+
+   case TGSI_OPCODE_ARR:
+      assert (0);
+      break;
+
+   case TGSI_OPCODE_BRA:
+      assert (0);
+      break;
+
+   case TGSI_OPCODE_CAL:
+      /* skip the call if no execution channels are enabled */
+      if (mach->ExecMask) {
+         /* do the call */
+
+         /* push the Cond, Loop, Cont stacks */
+         assert(mach->CondStackTop < TGSI_EXEC_MAX_COND_NESTING);
+         mach->CondStack[mach->CondStackTop++] = mach->CondMask;
+         assert(mach->LoopStackTop < TGSI_EXEC_MAX_LOOP_NESTING);
+         mach->LoopStack[mach->LoopStackTop++] = mach->LoopMask;
+         assert(mach->ContStackTop < TGSI_EXEC_MAX_LOOP_NESTING);
+         mach->ContStack[mach->ContStackTop++] = mach->ContMask;
+
+         assert(mach->FuncStackTop < TGSI_EXEC_MAX_CALL_NESTING);
+         mach->FuncStack[mach->FuncStackTop++] = mach->FuncMask;
+
+         /* note that PC was already incremented above */
+         mach->CallStack[mach->CallStackTop++] = *pc;
+         *pc = inst->InstructionExtLabel.Label;
+      }
+      break;
+
+   case TGSI_OPCODE_RET:
+      mach->FuncMask &= ~mach->ExecMask;
+      UPDATE_EXEC_MASK(mach);
+
+      if (mach->ExecMask == 0x0) {
+         /* really return now (otherwise, keep executing */
+
+         if (mach->CallStackTop == 0) {
+            /* returning from main() */
+            *pc = -1;
+            return;
+         }
+         *pc = mach->CallStack[--mach->CallStackTop];
+
+         /* pop the Cond, Loop, Cont stacks */
+         assert(mach->CondStackTop > 0);
+         mach->CondMask = mach->CondStack[--mach->CondStackTop];
+         assert(mach->LoopStackTop > 0);
+         mach->LoopMask = mach->LoopStack[--mach->LoopStackTop];
+         assert(mach->ContStackTop > 0);
+         mach->ContMask = mach->ContStack[--mach->ContStackTop];
+         assert(mach->FuncStackTop > 0);
+         mach->FuncMask = mach->FuncStack[--mach->FuncStackTop];
+
+         UPDATE_EXEC_MASK(mach);
+      }
+      break;
+
+   case TGSI_OPCODE_SSG:
+      assert (0);
+      break;
+
+   case TGSI_OPCODE_CMP:
+      FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
+         FETCH(&r[0], 0, chan_index);
+         FETCH(&r[1], 1, chan_index);
+         FETCH(&r[2], 2, chan_index);
+
+         micro_lt( &r[0], &r[0], &mach->Temps[TEMP_0_I].xyzw[TEMP_0_C], &r[1], &r[2] );
+
+         STORE(&r[0], 0, chan_index);
+      }
+      break;
+
+   case TGSI_OPCODE_SCS:
+      if( IS_CHANNEL_ENABLED( *inst, CHAN_X ) || IS_CHANNEL_ENABLED( *inst, CHAN_Y ) ) {
+         FETCH( &r[0], 0, CHAN_X );
+      }
+      if( IS_CHANNEL_ENABLED( *inst, CHAN_X ) ) {
+         micro_cos( &r[1], &r[0] );
+         STORE( &r[1], 0, CHAN_X );
+      }
+      if( IS_CHANNEL_ENABLED( *inst, CHAN_Y ) ) {
+         micro_sin( &r[1], &r[0] );
+         STORE( &r[1], 0, CHAN_Y );
+      }
+      if( IS_CHANNEL_ENABLED( *inst, CHAN_Z ) ) {
+         STORE( &mach->Temps[TEMP_0_I].xyzw[TEMP_0_C], 0, CHAN_Z );
+      }
+      if( IS_CHANNEL_ENABLED( *inst, CHAN_W ) ) {
+         STORE( &mach->Temps[TEMP_1_I].xyzw[TEMP_1_C], 0, CHAN_W );
+      }
+      break;
+
+   case TGSI_OPCODE_NRM:
+      assert (0);
+      break;
+
+   case TGSI_OPCODE_DIV:
+      assert( 0 );
+      break;
+
+   case TGSI_OPCODE_DP2:
+      FETCH( &r[0], 0, CHAN_X );
+      FETCH( &r[1], 1, CHAN_X );
+      micro_mul( &r[0], &r[0], &r[1] );
+
+      FETCH( &r[1], 0, CHAN_Y );
+      FETCH( &r[2], 1, CHAN_Y );
+      micro_mul( &r[1], &r[1], &r[2] );
+      micro_add( &r[0], &r[0], &r[1] );
+
+      FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
+         STORE( &r[0], 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_IF:
+      /* push CondMask */
+      assert(mach->CondStackTop < TGSI_EXEC_MAX_COND_NESTING);
+      mach->CondStack[mach->CondStackTop++] = mach->CondMask;
+      FETCH( &r[0], 0, CHAN_X );
+      /* update CondMask */
+      if( ! r[0].u[0] ) {
+         mach->CondMask &= ~0x1;
+      }
+      if( ! r[0].u[1] ) {
+         mach->CondMask &= ~0x2;
+      }
+      if( ! r[0].u[2] ) {
+         mach->CondMask &= ~0x4;
+      }
+      if( ! r[0].u[3] ) {
+         mach->CondMask &= ~0x8;
+      }
+      UPDATE_EXEC_MASK(mach);
+      /* Todo: If CondMask==0, jump to ELSE */
+      break;
+
+   case TGSI_OPCODE_ELSE:
+      /* invert CondMask wrt previous mask */
+      {
+         uint prevMask;
+         assert(mach->CondStackTop > 0);
+         prevMask = mach->CondStack[mach->CondStackTop - 1];
+         mach->CondMask = ~mach->CondMask & prevMask;
+         UPDATE_EXEC_MASK(mach);
+         /* Todo: If CondMask==0, jump to ENDIF */
+      }
+      break;
+
+   case TGSI_OPCODE_ENDIF:
+      /* pop CondMask */
+      assert(mach->CondStackTop > 0);
+      mach->CondMask = mach->CondStack[--mach->CondStackTop];
+      UPDATE_EXEC_MASK(mach);
+      break;
+
+   case TGSI_OPCODE_END:
+      /* halt execution */
+      *pc = -1;
+      break;
+
+   case TGSI_OPCODE_REP:
+      assert (0);
+      break;
+
+   case TGSI_OPCODE_ENDREP:
+       assert (0);
+       break;
+
+   case TGSI_OPCODE_PUSHA:
+      assert (0);
+      break;
+
+   case TGSI_OPCODE_POPA:
+      assert (0);
+      break;
+
+   case TGSI_OPCODE_CEIL:
+      FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
+         FETCH( &r[0], 0, chan_index );
+         micro_ceil( &r[0], &r[0] );
+         STORE( &r[0], 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_I2F:
+      FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
+         FETCH( &r[0], 0, chan_index );
+         micro_i2f( &r[0], &r[0] );
+         STORE( &r[0], 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_NOT:
+      FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
+         FETCH( &r[0], 0, chan_index );
+         micro_not( &r[0], &r[0] );
+         STORE( &r[0], 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_TRUNC:
+      FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
+         FETCH( &r[0], 0, chan_index );
+         micro_trunc( &r[0], &r[0] );
+         STORE( &r[0], 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_SHL:
+      FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
+         FETCH( &r[0], 0, chan_index );
+         FETCH( &r[1], 1, chan_index );
+         micro_shl( &r[0], &r[0], &r[1] );
+         STORE( &r[0], 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_SHR:
+      FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
+         FETCH( &r[0], 0, chan_index );
+         FETCH( &r[1], 1, chan_index );
+         micro_ishr( &r[0], &r[0], &r[1] );
+         STORE( &r[0], 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_AND:
+      FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
+         FETCH( &r[0], 0, chan_index );
+         FETCH( &r[1], 1, chan_index );
+         micro_and( &r[0], &r[0], &r[1] );
+         STORE( &r[0], 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_OR:
+      FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
+         FETCH( &r[0], 0, chan_index );
+         FETCH( &r[1], 1, chan_index );
+         micro_or( &r[0], &r[0], &r[1] );
+         STORE( &r[0], 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_MOD:
+      assert (0);
+      break;
+
+   case TGSI_OPCODE_XOR:
+      FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
+         FETCH( &r[0], 0, chan_index );
+         FETCH( &r[1], 1, chan_index );
+         micro_xor( &r[0], &r[0], &r[1] );
+         STORE( &r[0], 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_SAD:
+      assert (0);
+      break;
+
+   case TGSI_OPCODE_TXF:
+      assert (0);
+      break;
+
+   case TGSI_OPCODE_TXQ:
+      assert (0);
+      break;
+
+   case TGSI_OPCODE_EMIT:
+      mach->Temps[TEMP_OUTPUT_I].xyzw[TEMP_OUTPUT_C].u[0] += 16;
+      mach->Primitives[mach->Temps[TEMP_PRIMITIVE_I].xyzw[TEMP_PRIMITIVE_C].u[0]]++;
+      break;
+
+   case TGSI_OPCODE_ENDPRIM:
+      mach->Temps[TEMP_PRIMITIVE_I].xyzw[TEMP_PRIMITIVE_C].u[0]++;
+      mach->Primitives[mach->Temps[TEMP_PRIMITIVE_I].xyzw[TEMP_PRIMITIVE_C].u[0]] = 0;
+      break;
+
+   case TGSI_OPCODE_LOOP:
+      /* fall-through (for now) */
+   case TGSI_OPCODE_BGNLOOP2:
+      /* push LoopMask and ContMasks */
+      assert(mach->LoopStackTop < TGSI_EXEC_MAX_LOOP_NESTING);
+      mach->LoopStack[mach->LoopStackTop++] = mach->LoopMask;
+      assert(mach->ContStackTop < TGSI_EXEC_MAX_LOOP_NESTING);
+      mach->ContStack[mach->ContStackTop++] = mach->ContMask;
+      break;
+
+   case TGSI_OPCODE_ENDLOOP:
+      /* fall-through (for now at least) */
+   case TGSI_OPCODE_ENDLOOP2:
+      /* Restore ContMask, but don't pop */
+      assert(mach->ContStackTop > 0);
+      mach->ContMask = mach->ContStack[mach->ContStackTop - 1];
+      if (mach->LoopMask) {
+         /* repeat loop: jump to instruction just past BGNLOOP */
+         *pc = inst->InstructionExtLabel.Label + 1;
+      }
+      else {
+         /* exit loop: pop LoopMask */
+         assert(mach->LoopStackTop > 0);
+         mach->LoopMask = mach->LoopStack[--mach->LoopStackTop];
+         /* pop ContMask */
+         assert(mach->ContStackTop > 0);
+         mach->ContMask = mach->ContStack[--mach->ContStackTop];
+      }
+      UPDATE_EXEC_MASK(mach);
+      break;
+
+   case TGSI_OPCODE_BRK:
+      /* turn off loop channels for each enabled exec channel */
+      mach->LoopMask &= ~mach->ExecMask;
+      /* Todo: if mach->LoopMask == 0, jump to end of loop */
+      UPDATE_EXEC_MASK(mach);
+      break;
+
+   case TGSI_OPCODE_CONT:
+      /* turn off cont channels for each enabled exec channel */
+      mach->ContMask &= ~mach->ExecMask;
+      /* Todo: if mach->LoopMask == 0, jump to end of loop */
+      UPDATE_EXEC_MASK(mach);
+      break;
+
+   case TGSI_OPCODE_BGNSUB:
+      /* no-op */
+      break;
+
+   case TGSI_OPCODE_ENDSUB:
+      /* no-op */
+      break;
+
+   case TGSI_OPCODE_NOISE1:
+      assert( 0 );
+      break;
+
+   case TGSI_OPCODE_NOISE2:
+      assert( 0 );
+      break;
+
+   case TGSI_OPCODE_NOISE3:
+      assert( 0 );
+      break;
+
+   case TGSI_OPCODE_NOISE4:
+      assert( 0 );
+      break;
+
+   case TGSI_OPCODE_NOP:
+      break;
+
+   default:
+      assert( 0 );
+   }
+}
+
+
+/**
+ * Run TGSI interpreter.
+ * \return bitmask of "alive" quad components
+ */
+uint
+tgsi_exec_machine_run( struct tgsi_exec_machine *mach )
+{
+   uint i;
+   int pc = 0;
+
+   mach->CondMask = 0xf;
+   mach->LoopMask = 0xf;
+   mach->ContMask = 0xf;
+   mach->FuncMask = 0xf;
+   mach->ExecMask = 0xf;
+
+   mach->CondStackTop = 0; /* temporarily subvert this assertion */
+   assert(mach->CondStackTop == 0);
+   assert(mach->LoopStackTop == 0);
+   assert(mach->ContStackTop == 0);
+   assert(mach->CallStackTop == 0);
+
+   mach->Temps[TEMP_KILMASK_I].xyzw[TEMP_KILMASK_C].u[0] = 0;
+   mach->Temps[TEMP_OUTPUT_I].xyzw[TEMP_OUTPUT_C].u[0] = 0;
+
+   if( mach->Processor == TGSI_PROCESSOR_GEOMETRY ) {
+      mach->Temps[TEMP_PRIMITIVE_I].xyzw[TEMP_PRIMITIVE_C].u[0] = 0;
+      mach->Primitives[0] = 0;
+   }
+
+
+   /* execute declarations (interpolants) */
+   for (i = 0; i < mach->NumDeclarations; i++) {
+      exec_declaration( mach, mach->Declarations+i );
+   }
+
+   /* execute instructions, until pc is set to -1 */
+   while (pc != -1) {
+      assert(pc < mach->NumInstructions);
+      exec_instruction( mach, mach->Instructions + pc, &pc );
+   }
+
+#if 0
+   /* we scale from floats in [0,1] to Zbuffer ints in sp_quad_depth_test.c */
+   if (mach->Processor == TGSI_PROCESSOR_FRAGMENT) {
+      /*
+       * Scale back depth component.
+       */
+      for (i = 0; i < 4; i++)
+         mach->Outputs[0].xyzw[2].f[i] *= ctx->DrawBuffer->_DepthMaxF;
+   }
+#endif
+
+   return ~mach->Temps[TEMP_KILMASK_I].xyzw[TEMP_KILMASK_C].u[0];
+}
+
+
diff --git a/src/gallium/aux/tgsi/exec/tgsi_exec.h b/src/gallium/aux/tgsi/exec/tgsi_exec.h
new file mode 100644
index 0000000000..1fb66ee960
--- /dev/null
+++ b/src/gallium/aux/tgsi/exec/tgsi_exec.h
@@ -0,0 +1,239 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+#if !defined TGSI_EXEC_H
+#define TGSI_EXEC_H
+
+#include "pipe/p_compiler.h"
+
+#if defined __cplusplus
+extern "C" {
+#endif
+
+#define NUM_CHANNELS 4  /* R,G,B,A */
+#define QUAD_SIZE    4  /* 4 pixel/quad */
+
+/**
+  * Registers may be treated as float, signed int or unsigned int.
+  */
+union tgsi_exec_channel
+{
+   float    f[QUAD_SIZE];
+   int      i[QUAD_SIZE];
+   unsigned u[QUAD_SIZE];
+};
+
+/**
+  * A vector[RGBA] of channels[4 pixels]
+  */
+struct tgsi_exec_vector
+{
+   union tgsi_exec_channel xyzw[NUM_CHANNELS];
+};
+
+/**
+ * For fragment programs, information for computing fragment input
+ * values from plane equation of the triangle/line.
+ */
+struct tgsi_interp_coef
+{
+   float a0[NUM_CHANNELS];	/* in an xyzw layout */
+   float dadx[NUM_CHANNELS];
+   float dady[NUM_CHANNELS];
+};
+
+
+struct softpipe_tile_cache;  /**< Opaque to TGSI */
+
+/**
+ * Information for sampling textures, which must be implemented
+ * by code outside the TGSI executor.
+ */
+struct tgsi_sampler
+{
+   const struct pipe_sampler_state *state;
+   struct pipe_texture *texture;
+   /** Get samples for four fragments in a quad */
+   void (*get_samples)(struct tgsi_sampler *sampler,
+                       const float s[QUAD_SIZE],
+                       const float t[QUAD_SIZE],
+                       const float p[QUAD_SIZE],
+                       float lodbias,
+                       float rgba[NUM_CHANNELS][QUAD_SIZE]);
+   void *pipe; /*XXX temporary*/
+   struct softpipe_tile_cache *cache;
+};
+
+/**
+ * For branching/calling subroutines.
+ */
+struct tgsi_exec_labels
+{
+   unsigned labels[128][2];
+   unsigned count;
+};
+
+/*
+ * Locations of various utility registers (_I = Index, _C = Channel)
+ */
+#define TGSI_EXEC_TEMP_00000000_I   32
+#define TGSI_EXEC_TEMP_00000000_C   0
+
+#define TGSI_EXEC_TEMP_7FFFFFFF_I   32
+#define TGSI_EXEC_TEMP_7FFFFFFF_C   1
+
+#define TGSI_EXEC_TEMP_80000000_I   32
+#define TGSI_EXEC_TEMP_80000000_C   2
+
+#define TGSI_EXEC_TEMP_FFFFFFFF_I   32
+#define TGSI_EXEC_TEMP_FFFFFFFF_C   3
+
+#define TGSI_EXEC_TEMP_ONE_I        33
+#define TGSI_EXEC_TEMP_ONE_C        0
+
+#define TGSI_EXEC_TEMP_TWO_I        33
+#define TGSI_EXEC_TEMP_TWO_C        1
+
+#define TGSI_EXEC_TEMP_128_I        33
+#define TGSI_EXEC_TEMP_128_C        2
+
+#define TGSI_EXEC_TEMP_MINUS_128_I  33
+#define TGSI_EXEC_TEMP_MINUS_128_C  3
+
+#define TGSI_EXEC_TEMP_KILMASK_I    34
+#define TGSI_EXEC_TEMP_KILMASK_C    0
+
+#define TGSI_EXEC_TEMP_OUTPUT_I     34
+#define TGSI_EXEC_TEMP_OUTPUT_C     1
+
+#define TGSI_EXEC_TEMP_PRIMITIVE_I  34
+#define TGSI_EXEC_TEMP_PRIMITIVE_C  2
+
+#define TGSI_EXEC_TEMP_R0           35
+
+#define TGSI_EXEC_NUM_TEMPS   (32 + 4)
+#define TGSI_EXEC_NUM_ADDRS   1
+#define TGSI_EXEC_NUM_IMMEDIATES  256
+
+#define TGSI_EXEC_MAX_COND_NESTING  10
+#define TGSI_EXEC_MAX_LOOP_NESTING  10
+#define TGSI_EXEC_MAX_CALL_NESTING  10
+
+/**
+ * Run-time virtual machine state for executing TGSI shader.
+ */
+struct tgsi_exec_machine
+{
+   /*
+    * 32 program temporaries
+    * 4  internal temporaries
+    * 1  address
+    * 1  temporary of padding to align to 16 bytes
+    */
+   struct tgsi_exec_vector       _Temps[TGSI_EXEC_NUM_TEMPS + TGSI_EXEC_NUM_ADDRS + 1];
+
+   /*
+    * This will point to _Temps after aligning to 16B boundary.
+    */
+   struct tgsi_exec_vector       *Temps;
+   struct tgsi_exec_vector       *Addrs;
+
+   struct tgsi_sampler           *Samplers;
+
+   float                         Imms[TGSI_EXEC_NUM_IMMEDIATES][4];
+   unsigned                      ImmLimit;
+   float                         (*Consts)[4];
+   struct tgsi_exec_vector       *Inputs;
+   struct tgsi_exec_vector       *Outputs;
+   const struct tgsi_token       *Tokens;
+   unsigned                      Processor;
+
+   /* GEOMETRY processor only. */
+   unsigned                      *Primitives;
+
+   /* FRAGMENT processor only. */
+   const struct tgsi_interp_coef *InterpCoefs;
+   struct tgsi_exec_vector       QuadPos;
+
+   /* Conditional execution masks */
+   uint CondMask;  /**< For IF/ELSE/ENDIF */
+   uint LoopMask;  /**< For BGNLOOP/ENDLOOP */
+   uint ContMask;  /**< For loop CONT statements */
+   uint FuncMask;  /**< For function calls */
+   uint ExecMask;  /**< = CondMask & LoopMask */
+
+   /** Condition mask stack (for nested conditionals) */
+   uint CondStack[TGSI_EXEC_MAX_COND_NESTING];
+   int CondStackTop;
+
+   /** Loop mask stack (for nested loops) */
+   uint LoopStack[TGSI_EXEC_MAX_LOOP_NESTING];
+   int LoopStackTop;
+
+   /** Loop continue mask stack (see comments in tgsi_exec.c) */
+   uint ContStack[TGSI_EXEC_MAX_LOOP_NESTING];
+   int ContStackTop;
+
+   /** Function execution mask stack (for executing subroutine code) */
+   uint FuncStack[TGSI_EXEC_MAX_CALL_NESTING];
+   int FuncStackTop;
+
+   /** Function call stack for saving/restoring the program counter */
+   uint CallStack[TGSI_EXEC_MAX_CALL_NESTING];
+   int CallStackTop;
+
+   struct tgsi_full_instruction *Instructions;
+   uint NumInstructions;
+
+   struct tgsi_full_declaration *Declarations;
+   uint NumDeclarations;
+
+   struct tgsi_exec_labels Labels;
+};
+
+
+void
+tgsi_exec_machine_init(
+   struct tgsi_exec_machine *mach,
+   const struct tgsi_token *tokens,
+   unsigned numSamplers,
+   struct tgsi_sampler *samplers);
+
+uint
+tgsi_exec_machine_run(
+   struct tgsi_exec_machine *mach );
+
+
+void
+tgsi_exec_machine_free_data(struct tgsi_exec_machine *mach);
+
+
+#if defined __cplusplus
+} /* extern "C" */
+#endif
+
+#endif /* TGSI_EXEC_H */
diff --git a/src/gallium/aux/tgsi/exec/tgsi_sse2.c b/src/gallium/aux/tgsi/exec/tgsi_sse2.c
new file mode 100755
index 0000000000..1e56e4afb6
--- /dev/null
+++ b/src/gallium/aux/tgsi/exec/tgsi_sse2.c
@@ -0,0 +1,2378 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+#include "pipe/p_util.h"
+#include "pipe/p_shader_tokens.h"
+#include "pipe/tgsi/util/tgsi_parse.h"
+#include "pipe/tgsi/util/tgsi_util.h"
+#include "tgsi_exec.h"
+#include "tgsi_sse2.h"
+
+#include "x86/rtasm/x86sse.h"
+
+#if defined(__i386__) || defined(__386__)
+
+#define DUMP_SSE  0
+
+#if DUMP_SSE
+
+static void
+_print_reg(
+   struct x86_reg reg )
+{
+   if (reg.mod != mod_REG) 
+      debug_printf( "[" );
+      
+   switch( reg.file ) {
+   case file_REG32:
+      switch( reg.idx ) {
+      case reg_AX:
+         debug_printf( "EAX" );
+         break;
+      case reg_CX:
+         debug_printf( "ECX" );
+         break;
+      case reg_DX:
+         debug_printf( "EDX" );
+         break;
+      case reg_BX:
+         debug_printf( "EBX" );
+         break;
+      case reg_SP:
+         debug_printf( "ESP" );
+         break;
+      case reg_BP:
+         debug_printf( "EBP" );
+         break;
+      case reg_SI:
+         debug_printf( "ESI" );
+         break;
+      case reg_DI:
+         debug_printf( "EDI" );
+         break;
+      }
+      break;
+   case file_MMX:
+      assert( 0 );
+      break;
+   case file_XMM:
+      debug_printf( "XMM%u", reg.idx );
+      break;
+   case file_x87:
+      assert( 0 );
+      break;
+   }
+
+   if (reg.mod == mod_DISP8 ||
+       reg.mod == mod_DISP32)
+      debug_printf("+%d", reg.disp);
+
+   if (reg.mod != mod_REG) 
+      debug_printf( "]" );
+}
+
+static void
+_fill(
+   const char  *op )
+{
+   unsigned count = 10 - strlen( op );
+
+   while( count-- ) {
+      debug_printf( " " );
+   }
+}
+
+#define DUMP_START() debug_printf( "\nsse-dump start ----------------" )
+#define DUMP_END() debug_printf( "\nsse-dump end ----------------\n" )
+#define DUMP( OP ) debug_printf( "\n%s", OP )
+#define DUMP_I( OP, I ) do {\
+   debug_printf( "\n%s", OP );\
+   _fill( OP );\
+   debug_printf( "%u", I ); } while( 0 )
+#define DUMP_R( OP, R0 ) do {\
+   debug_printf( "\n%s", OP );\
+   _fill( OP );\
+   _print_reg( R0 ); } while( 0 )
+#define DUMP_RR( OP, R0, R1 ) do {\
+   debug_printf( "\n%s", OP );\
+   _fill( OP );\
+   _print_reg( R0 );\
+   debug_printf( ", " );\
+   _print_reg( R1 ); } while( 0 )
+#define DUMP_RRI( OP, R0, R1, I ) do {\
+   debug_printf( "\n%s", OP );\
+   _fill( OP );\
+   _print_reg( R0 );\
+   debug_printf( ", " );\
+   _print_reg( R1 );\
+   debug_printf( ", " );\
+   debug_printf( "%u", I ); } while( 0 )
+
+#else
+
+#define DUMP_START()
+#define DUMP_END()
+#define DUMP( OP )
+#define DUMP_I( OP, I )
+#define DUMP_R( OP, R0 )
+#define DUMP_RR( OP, R0, R1 )
+#define DUMP_RRI( OP, R0, R1, I )
+
+#endif
+
+#define FOR_EACH_CHANNEL( CHAN )\
+   for( CHAN = 0; CHAN < 4; CHAN++ )
+
+#define IS_DST0_CHANNEL_ENABLED( INST, CHAN )\
+   ((INST).FullDstRegisters[0].DstRegister.WriteMask & (1 << (CHAN)))
+
+#define IF_IS_DST0_CHANNEL_ENABLED( INST, CHAN )\
+   if( IS_DST0_CHANNEL_ENABLED( INST, CHAN ))
+
+#define FOR_EACH_DST0_ENABLED_CHANNEL( INST, CHAN )\
+   FOR_EACH_CHANNEL( CHAN )\
+      IF_IS_DST0_CHANNEL_ENABLED( INST, CHAN )
+
+#define CHAN_X 0
+#define CHAN_Y 1
+#define CHAN_Z 2
+#define CHAN_W 3
+
+#define TEMP_R0   TGSI_EXEC_TEMP_R0
+
+/**
+ * X86 utility functions.
+ */
+
+static struct x86_reg
+make_xmm(
+   unsigned xmm )
+{
+   return x86_make_reg(
+      file_XMM,
+      (enum x86_reg_name) xmm );
+}
+
+/**
+ * X86 register mapping helpers.
+ */
+
+static struct x86_reg
+get_const_base( void )
+{
+   return x86_make_reg(
+      file_REG32,
+      reg_CX );
+}
+
+static struct x86_reg
+get_input_base( void )
+{
+   return x86_make_reg(
+      file_REG32,
+      reg_AX );
+}
+
+static struct x86_reg
+get_output_base( void )
+{
+   return x86_make_reg(
+      file_REG32,
+      reg_DX );
+}
+
+static struct x86_reg
+get_temp_base( void )
+{
+#ifdef WIN32
+   return x86_make_reg(
+      file_REG32,
+      reg_BX );
+#else
+   return x86_make_reg(
+      file_REG32,
+      reg_SI );
+#endif
+}
+
+static struct x86_reg
+get_coef_base( void )
+{
+   return get_output_base();
+}
+
+/**
+ * Data access helpers.
+ */
+
+static struct x86_reg
+get_argument(
+   unsigned index )
+{
+   return x86_make_disp(
+      x86_make_reg( file_REG32, reg_SP ),
+      (index + 1) * 4 );
+}
+
+static struct x86_reg
+get_const(
+   unsigned vec,
+   unsigned chan )
+{
+   return x86_make_disp(
+      get_const_base(),
+      (vec * 4 + chan) * 4 );
+}
+
+static struct x86_reg
+get_input(
+   unsigned vec,
+   unsigned chan )
+{
+   return x86_make_disp(
+      get_input_base(),
+      (vec * 4 + chan) * 16 );
+}
+
+static struct x86_reg
+get_output(
+   unsigned vec,
+   unsigned chan )
+{
+   return x86_make_disp(
+      get_output_base(),
+      (vec * 4 + chan) * 16 );
+}
+
+static struct x86_reg
+get_temp(
+   unsigned vec,
+   unsigned chan )
+{
+   return x86_make_disp(
+      get_temp_base(),
+      (vec * 4 + chan) * 16 );
+}
+
+static struct x86_reg
+get_coef(
+   unsigned vec,
+   unsigned chan,
+   unsigned member )
+{
+   return x86_make_disp(
+      get_coef_base(),
+      ((vec * 3 + member) * 4 + chan) * 4 );
+}
+
+/**
+ * X86 rtasm wrappers.
+ */
+
+static void
+emit_addps(
+   struct x86_function  *func,
+   struct x86_reg       dst,
+   struct x86_reg       src )
+{
+   DUMP_RR( "ADDPS", dst, src );
+   sse_addps( func, dst, src );
+}
+
+static void
+emit_andnps(
+   struct x86_function  *func,
+   struct x86_reg       dst,
+   struct x86_reg       src )
+{
+   DUMP_RR( "ANDNPS", dst, src );
+   sse_andnps( func, dst, src );
+}
+
+static void
+emit_andps(
+   struct x86_function  *func,
+   struct x86_reg       dst,
+   struct x86_reg       src )
+{
+   DUMP_RR( "ANDPS", dst, src );
+   sse_andps( func, dst, src );
+}
+
+static void
+emit_call(
+   struct x86_function  *func,
+   void                 (* addr)() )
+{
+   struct x86_reg ecx = x86_make_reg( file_REG32, reg_CX );
+
+   DUMP_I( "CALL", addr );
+   x86_mov_reg_imm( func, ecx, (unsigned long) addr );
+   x86_call( func, ecx );
+}
+
+static void
+emit_cmpps(
+   struct x86_function  *func,
+   struct x86_reg       dst,
+   struct x86_reg       src,
+   enum sse_cc          cc )
+{
+   DUMP_RRI( "CMPPS", dst, src, cc );
+   sse_cmpps( func, dst, src, cc );
+}
+
+static void
+emit_cvttps2dq(
+   struct x86_function  *func,
+   struct x86_reg       dst,
+   struct x86_reg       src )
+{
+   DUMP_RR( "CVTTPS2DQ", dst, src );
+   sse2_cvttps2dq( func, dst, src );
+}
+
+static void
+emit_maxps(
+   struct x86_function  *func,
+   struct x86_reg       dst,
+   struct x86_reg       src )
+{
+   DUMP_RR( "MAXPS", dst, src );
+   sse_maxps( func, dst, src );
+}
+
+static void
+emit_minps(
+   struct x86_function  *func,
+   struct x86_reg       dst,
+   struct x86_reg       src )
+{
+   DUMP_RR( "MINPS", dst, src );
+   sse_minps( func, dst, src );
+}
+
+static void
+emit_mov(
+   struct x86_function  *func,
+   struct x86_reg       dst,
+   struct x86_reg       src )
+{
+   DUMP_RR( "MOV", dst, src );
+   x86_mov( func, dst, src );
+}
+
+static void
+emit_movaps(
+   struct x86_function  *func,
+   struct x86_reg       dst,
+   struct x86_reg       src )
+{
+   DUMP_RR( "MOVAPS", dst, src );
+   sse_movaps( func, dst, src );
+}
+
+static void
+emit_movss(
+   struct x86_function  *func,
+   struct x86_reg       dst,
+   struct x86_reg       src )
+{
+   DUMP_RR( "MOVSS", dst, src );
+   sse_movss( func, dst, src );
+}
+
+static void
+emit_movups(
+   struct x86_function  *func,
+   struct x86_reg       dst,
+   struct x86_reg       src )
+{
+   DUMP_RR( "MOVUPS", dst, src );
+   sse_movups( func, dst, src );
+}
+
+static void
+emit_mulps(
+   struct x86_function  *func,
+   struct x86_reg       dst,
+   struct x86_reg       src )
+{
+   DUMP_RR( "MULPS", dst, src );
+   sse_mulps( func, dst, src );
+}
+
+static void
+emit_or(
+   struct x86_function  *func,
+   struct x86_reg       dst,
+   struct x86_reg       src )
+{
+   DUMP_RR( "OR", dst, src );
+   x86_or( func, dst, src );
+}
+
+static void
+emit_orps(
+   struct x86_function  *func,
+   struct x86_reg       dst,
+   struct x86_reg       src )
+{
+   DUMP_RR( "ORPS", dst, src );
+   sse_orps( func, dst, src );
+}
+
+static void
+emit_pmovmskb(
+   struct x86_function  *func,
+   struct x86_reg       dst,
+   struct x86_reg       src )
+{
+   DUMP_RR( "PMOVMSKB", dst, src );
+   sse_pmovmskb( func, dst, src );
+}
+
+static void
+emit_pop(
+   struct x86_function  *func,
+   struct x86_reg       dst )
+{
+   DUMP_R( "POP", dst );
+   x86_pop( func, dst );
+}
+
+static void
+emit_push(
+   struct x86_function  *func,
+   struct x86_reg       dst )
+{
+   DUMP_R( "PUSH", dst );
+   x86_push( func, dst );
+}
+
+static void
+emit_rcpps(
+   struct x86_function  *func,
+   struct x86_reg       dst,
+   struct x86_reg       src )
+{
+   DUMP_RR( "RCPPS", dst, src );
+   sse2_rcpps( func, dst, src );
+}
+
+#ifdef WIN32
+static void
+emit_retw(
+   struct x86_function  *func,
+   unsigned             size )
+{
+   DUMP_I( "RET", size );
+   x86_retw( func, size );
+}
+#else
+static void
+emit_ret(
+   struct x86_function  *func )
+{
+   DUMP( "RET" );
+   x86_ret( func );
+}
+#endif
+
+static void
+emit_rsqrtps(
+   struct x86_function  *func,
+   struct x86_reg       dst,
+   struct x86_reg       src )
+{
+   DUMP_RR( "RSQRTPS", dst, src );
+   sse_rsqrtps( func, dst, src );
+}
+
+static void
+emit_shufps(
+   struct x86_function  *func,
+   struct x86_reg       dst,
+   struct x86_reg       src,
+   unsigned char        shuf )
+{
+   DUMP_RRI( "SHUFPS", dst, src, shuf );
+   sse_shufps( func, dst, src, shuf );
+}
+
+static void
+emit_subps(
+   struct x86_function  *func,
+   struct x86_reg       dst,
+   struct x86_reg       src )
+{
+   DUMP_RR( "SUBPS", dst, src );
+   sse_subps( func, dst, src );
+}
+
+static void
+emit_xorps(
+   struct x86_function  *func,
+   struct x86_reg       dst,
+   struct x86_reg       src )
+{
+   DUMP_RR( "XORPS", dst, src );
+   sse_xorps( func, dst, src );
+}
+
+/**
+ * Data fetch helpers.
+ */
+
+static void
+emit_const(
+   struct x86_function *func,
+   unsigned xmm,
+   unsigned vec,
+   unsigned chan )
+{
+   emit_movss(
+      func,
+      make_xmm( xmm ),
+      get_const( vec, chan ) );
+   emit_shufps(
+      func,
+      make_xmm( xmm ),
+      make_xmm( xmm ),
+      SHUF( 0, 0, 0, 0 ) );
+}
+
+static void
+emit_inputf(
+   struct x86_function *func,
+   unsigned xmm,
+   unsigned vec,
+   unsigned chan )
+{
+   emit_movups(
+      func,
+      make_xmm( xmm ),
+      get_input( vec, chan ) );
+}
+
+static void
+emit_output(
+   struct x86_function *func,
+   unsigned xmm,
+   unsigned vec,
+   unsigned chan )
+{
+   emit_movups(
+      func,
+      get_output( vec, chan ),
+      make_xmm( xmm ) );
+}
+
+static void
+emit_tempf(
+   struct x86_function *func,
+   unsigned xmm,
+   unsigned vec,
+   unsigned chan )
+{
+   emit_movaps(
+      func,
+      make_xmm( xmm ),
+      get_temp( vec, chan ) );
+}
+
+static void
+emit_coef(
+   struct x86_function *func,
+   unsigned xmm,
+   unsigned vec,
+   unsigned chan,
+   unsigned member )
+{
+   emit_movss(
+      func,
+      make_xmm( xmm ),
+      get_coef( vec, chan, member ) );
+   emit_shufps(
+      func,
+      make_xmm( xmm ),
+      make_xmm( xmm ),
+      SHUF( 0, 0, 0, 0 ) );
+}
+
+/**
+ * Data store helpers.
+ */
+
+static void
+emit_inputs(
+   struct x86_function *func,
+   unsigned xmm,
+   unsigned vec,
+   unsigned chan )
+{
+   emit_movups(
+      func,
+      get_input( vec, chan ),
+      make_xmm( xmm ) );
+}
+
+static void
+emit_temps(
+   struct x86_function *func,
+   unsigned xmm,
+   unsigned vec,
+   unsigned chan )
+{
+   emit_movaps(
+      func,
+      get_temp( vec, chan ),
+      make_xmm( xmm ) );
+}
+
+static void
+emit_addrs(
+   struct x86_function *func,
+   unsigned xmm,
+   unsigned vec,
+   unsigned chan )
+{
+   emit_temps(
+      func,
+      xmm,
+      vec + TGSI_EXEC_NUM_TEMPS,
+      chan );
+}
+
+/**
+ * Coefficent fetch helpers.
+ */
+
+static void
+emit_coef_a0(
+   struct x86_function *func,
+   unsigned xmm,
+   unsigned vec,
+   unsigned chan )
+{
+   emit_coef(
+      func,
+      xmm,
+      vec,
+      chan,
+      0 );
+}
+
+static void
+emit_coef_dadx(
+   struct x86_function *func,
+   unsigned xmm,
+   unsigned vec,
+   unsigned chan )
+{
+   emit_coef(
+      func,
+      xmm,
+      vec,
+      chan,
+      1 );
+}
+
+static void
+emit_coef_dady(
+   struct x86_function *func,
+   unsigned xmm,
+   unsigned vec,
+   unsigned chan )
+{
+   emit_coef(
+      func,
+      xmm,
+      vec,
+      chan,
+      2 );
+}
+
+/**
+ * Function call helpers.
+ */
+
+static void
+emit_push_gp(
+   struct x86_function *func )
+{
+   emit_push(
+      func,
+      get_const_base() );
+   emit_push(
+      func,
+      get_input_base() );
+   emit_push(
+      func,
+      get_output_base() );
+
+   /* It is important on non-win32 platforms that temp base is pushed last.
+    */
+   emit_push(
+      func,
+      get_temp_base() );
+}
+
+static void
+emit_pop_gp(
+   struct x86_function *func )
+{
+   /* Restore GP registers in a reverse order.
+    */
+   emit_pop(
+      func,
+      get_temp_base() );
+   emit_pop(
+      func,
+      get_output_base() );
+   emit_pop(
+      func,
+      get_input_base() );
+   emit_pop(
+      func,
+      get_const_base() );
+}
+
+static void
+emit_func_call_dst(
+   struct x86_function *func,
+   unsigned xmm_dst,
+   void (*code)() )
+{
+   emit_movaps(
+      func,
+      get_temp( TEMP_R0, 0 ),
+      make_xmm( xmm_dst ) );
+
+   emit_push_gp(
+      func );
+
+#ifdef WIN32
+   emit_push(
+      func,
+      get_temp( TEMP_R0, 0 ) );
+#endif
+
+   emit_call(
+      func,
+      code );
+
+   emit_pop_gp(
+      func );
+
+   emit_movaps(
+      func,
+      make_xmm( xmm_dst ),
+      get_temp( TEMP_R0, 0 ) );
+}
+
+static void
+emit_func_call_dst_src(
+   struct x86_function *func,
+   unsigned xmm_dst,
+   unsigned xmm_src,
+   void (*code)() )
+{
+   emit_movaps(
+      func,
+      get_temp( TEMP_R0, 1 ),
+      make_xmm( xmm_src ) );
+
+   emit_func_call_dst(
+      func,
+      xmm_dst,
+      code );
+}
+
+/**
+ * Low-level instruction translators.
+ */
+
+static void
+emit_abs(
+   struct x86_function *func,
+   unsigned xmm )
+{
+   emit_andps(
+      func,
+      make_xmm( xmm ),
+      get_temp(
+         TGSI_EXEC_TEMP_7FFFFFFF_I,
+         TGSI_EXEC_TEMP_7FFFFFFF_C ) );
+}
+
+static void
+emit_add(
+   struct x86_function *func,
+   unsigned xmm_dst,
+   unsigned xmm_src )
+{
+   emit_addps(
+      func,
+      make_xmm( xmm_dst ),
+      make_xmm( xmm_src ) );
+}
+
+static void XSTDCALL
+cos4f(
+   float *store )
+{
+#ifdef WIN32
+   store[0] = (float) cos( (double) store[0] );
+   store[1] = (float) cos( (double) store[1] );
+   store[2] = (float) cos( (double) store[2] );
+   store[3] = (float) cos( (double) store[3] );
+#else
+   const unsigned X = TEMP_R0 * 16;
+   store[X + 0] = cosf( store[X + 0] );
+   store[X + 1] = cosf( store[X + 1] );
+   store[X + 2] = cosf( store[X + 2] );
+   store[X + 3] = cosf( store[X + 3] );
+#endif
+}
+
+static void
+emit_cos(
+   struct x86_function *func,
+   unsigned xmm_dst )
+{
+   emit_func_call_dst(
+      func,
+      xmm_dst,
+      cos4f );
+}
+
+static void XSTDCALL
+ex24f(
+   float *store )
+{
+#ifdef WIN32
+   store[0] = (float) pow( 2.0, (double) store[0] );
+   store[1] = (float) pow( 2.0, (double) store[1] );
+   store[2] = (float) pow( 2.0, (double) store[2] );
+   store[3] = (float) pow( 2.0, (double) store[3] );
+#else
+   const unsigned X = TEMP_R0 * 16;
+   store[X + 0] = powf( 2.0f, store[X + 0] );
+   store[X + 1] = powf( 2.0f, store[X + 1] );
+   store[X + 2] = powf( 2.0f, store[X + 2] );
+   store[X + 3] = powf( 2.0f, store[X + 3] );
+#endif
+}
+
+static void
+emit_ex2(
+   struct x86_function *func,
+   unsigned xmm_dst )
+{
+   emit_func_call_dst(
+      func,
+      xmm_dst,
+      ex24f );
+}
+
+static void
+emit_f2it(
+   struct x86_function *func,
+   unsigned xmm )
+{
+   emit_cvttps2dq(
+      func,
+      make_xmm( xmm ),
+      make_xmm( xmm ) );
+}
+
+static void XSTDCALL
+flr4f(
+   float *store )
+{
+#ifdef WIN32
+   const unsigned X = 0;
+#else
+   const unsigned X = TEMP_R0 * 16;
+#endif
+   store[X + 0] = (float) floor( (double) store[X + 0] );
+   store[X + 1] = (float) floor( (double) store[X + 1] );
+   store[X + 2] = (float) floor( (double) store[X + 2] );
+   store[X + 3] = (float) floor( (double) store[X + 3] );
+}
+
+static void
+emit_flr(
+   struct x86_function *func,
+   unsigned xmm_dst )
+{
+   emit_func_call_dst(
+      func,
+      xmm_dst,
+      flr4f );
+}
+
+static void XSTDCALL
+frc4f(
+   float *store )
+{
+#ifdef WIN32
+   const unsigned X = 0;
+#else
+   const unsigned X = TEMP_R0 * 16;
+#endif
+   store[X + 0] -= (float) floor( (double) store[X + 0] );
+   store[X + 1] -= (float) floor( (double) store[X + 1] );
+   store[X + 2] -= (float) floor( (double) store[X + 2] );
+   store[X + 3] -= (float) floor( (double) store[X + 3] );
+}
+
+static void
+emit_frc(
+   struct x86_function *func,
+   unsigned xmm_dst )
+{
+   emit_func_call_dst(
+      func,
+      xmm_dst,
+      frc4f );
+}
+
+static void XSTDCALL
+lg24f(
+   float *store )
+{
+#ifdef WIN32
+   const unsigned X = 0;
+#else
+   const unsigned X = TEMP_R0 * 16;
+#endif
+   store[X + 0] = LOG2( store[X + 0] );
+   store[X + 1] = LOG2( store[X + 1] );
+   store[X + 2] = LOG2( store[X + 2] );
+   store[X + 3] = LOG2( store[X + 3] );
+}
+
+static void
+emit_lg2(
+   struct x86_function *func,
+   unsigned xmm_dst )
+{
+   emit_func_call_dst(
+      func,
+      xmm_dst,
+      lg24f );
+}
+
+static void
+emit_MOV(
+   struct x86_function *func,
+   unsigned xmm_dst,
+   unsigned xmm_src )
+{
+   emit_movups(
+      func,
+      make_xmm( xmm_dst ),
+      make_xmm( xmm_src ) );
+}
+
+static void
+emit_mul (struct x86_function *func,
+          unsigned xmm_dst,
+          unsigned xmm_src)
+{
+   emit_mulps(
+      func,
+      make_xmm( xmm_dst ),
+      make_xmm( xmm_src ) );
+}
+
+static void
+emit_neg(
+   struct x86_function *func,
+   unsigned xmm )
+{
+   emit_xorps(
+      func,
+      make_xmm( xmm ),
+      get_temp(
+         TGSI_EXEC_TEMP_80000000_I,
+         TGSI_EXEC_TEMP_80000000_C ) );
+}
+
+static void XSTDCALL
+pow4f(
+   float *store )
+{
+#ifdef WIN32
+   store[0] = (float) pow( (double) store[0], (double) store[4] );
+   store[1] = (float) pow( (double) store[1], (double) store[5] );
+   store[2] = (float) pow( (double) store[2], (double) store[6] );
+   store[3] = (float) pow( (double) store[3], (double) store[7] );
+#else
+   const unsigned X = TEMP_R0 * 16;
+   store[X + 0] = powf( store[X + 0], store[X + 4] );
+   store[X + 1] = powf( store[X + 1], store[X + 5] );
+   store[X + 2] = powf( store[X + 2], store[X + 6] );
+   store[X + 3] = powf( store[X + 3], store[X + 7] );
+#endif
+}
+
+static void
+emit_pow(
+   struct x86_function *func,
+   unsigned xmm_dst,
+   unsigned xmm_src )
+{
+   emit_func_call_dst_src(
+      func,
+      xmm_dst,
+      xmm_src,
+      pow4f );
+}
+
+static void
+emit_rcp (
+   struct x86_function *func,
+   unsigned xmm_dst,
+   unsigned xmm_src )
+{
+   emit_rcpps(
+      func,
+      make_xmm( xmm_dst ),
+      make_xmm( xmm_src ) );
+}
+
+static void
+emit_rsqrt(
+   struct x86_function *func,
+   unsigned xmm_dst,
+   unsigned xmm_src )
+{
+   emit_rsqrtps(
+      func,
+      make_xmm( xmm_dst ),
+      make_xmm( xmm_src ) );
+}
+
+static void
+emit_setsign(
+   struct x86_function *func,
+   unsigned xmm )
+{
+   emit_orps(
+      func,
+      make_xmm( xmm ),
+      get_temp(
+         TGSI_EXEC_TEMP_80000000_I,
+         TGSI_EXEC_TEMP_80000000_C ) );
+}
+
+static void XSTDCALL
+sin4f(
+   float *store )
+{
+#ifdef WIN32
+   store[0] = (float) sin( (double) store[0] );
+   store[1] = (float) sin( (double) store[1] );
+   store[2] = (float) sin( (double) store[2] );
+   store[3] = (float) sin( (double) store[3] );
+#else
+   const unsigned X = TEMP_R0 * 16;
+   store[X + 0] = sinf( store[X + 0] );
+   store[X + 1] = sinf( store[X + 1] );
+   store[X + 2] = sinf( store[X + 2] );
+   store[X + 3] = sinf( store[X + 3] );
+#endif
+}
+
+static void
+emit_sin (struct x86_function *func,
+          unsigned xmm_dst)
+{
+   emit_func_call_dst(
+      func,
+      xmm_dst,
+      sin4f );
+}
+
+static void
+emit_sub(
+   struct x86_function *func,
+   unsigned xmm_dst,
+   unsigned xmm_src )
+{
+   emit_subps(
+      func,
+      make_xmm( xmm_dst ),
+      make_xmm( xmm_src ) );
+}
+
+/**
+ * Register fetch.
+ */
+
+static void
+emit_fetch(
+   struct x86_function *func,
+   unsigned xmm,
+   const struct tgsi_full_src_register *reg,
+   const unsigned chan_index )
+{
+   unsigned swizzle = tgsi_util_get_full_src_register_extswizzle( reg, chan_index );
+
+   switch( swizzle ) {
+   case TGSI_EXTSWIZZLE_X:
+   case TGSI_EXTSWIZZLE_Y:
+   case TGSI_EXTSWIZZLE_Z:
+   case TGSI_EXTSWIZZLE_W:
+      switch( reg->SrcRegister.File ) {
+      case TGSI_FILE_CONSTANT:
+         emit_const(
+            func,
+            xmm,
+            reg->SrcRegister.Index,
+            swizzle );
+         break;
+
+      case TGSI_FILE_INPUT:
+         emit_inputf(
+            func,
+            xmm,
+            reg->SrcRegister.Index,
+            swizzle );
+         break;
+
+      case TGSI_FILE_TEMPORARY:
+         emit_tempf(
+            func,
+            xmm,
+            reg->SrcRegister.Index,
+            swizzle );
+         break;
+
+      default:
+         assert( 0 );
+      }
+      break;
+
+   case TGSI_EXTSWIZZLE_ZERO:
+      emit_tempf(
+         func,
+         xmm,
+         TGSI_EXEC_TEMP_00000000_I,
+         TGSI_EXEC_TEMP_00000000_C );
+      break;
+
+   case TGSI_EXTSWIZZLE_ONE:
+      emit_tempf(
+         func,
+         xmm,
+         TGSI_EXEC_TEMP_ONE_I,
+         TGSI_EXEC_TEMP_ONE_C );
+      break;
+
+   default:
+      assert( 0 );
+   }
+
+   switch( tgsi_util_get_full_src_register_sign_mode( reg, chan_index ) ) {
+   case TGSI_UTIL_SIGN_CLEAR:
+      emit_abs( func, xmm );
+      break;
+
+   case TGSI_UTIL_SIGN_SET:
+      emit_setsign( func, xmm );
+      break;
+
+   case TGSI_UTIL_SIGN_TOGGLE:
+      emit_neg( func, xmm );
+      break;
+
+   case TGSI_UTIL_SIGN_KEEP:
+      break;
+   }
+}
+
+#define FETCH( FUNC, INST, XMM, INDEX, CHAN )\
+   emit_fetch( FUNC, XMM, &(INST).FullSrcRegisters[INDEX], CHAN )
+
+/**
+ * Register store.
+ */
+
+static void
+emit_store(
+   struct x86_function *func,
+   unsigned xmm,
+   const struct tgsi_full_dst_register *reg,
+   const struct tgsi_full_instruction *inst,
+   unsigned chan_index )
+{
+   switch( reg->DstRegister.File ) {
+   case TGSI_FILE_OUTPUT:
+      emit_output(
+         func,
+         xmm,
+         reg->DstRegister.Index,
+         chan_index );
+      break;
+
+   case TGSI_FILE_TEMPORARY:
+      emit_temps(
+         func,
+         xmm,
+         reg->DstRegister.Index,
+         chan_index );
+      break;
+
+   case TGSI_FILE_ADDRESS:
+      emit_addrs(
+         func,
+         xmm,
+         reg->DstRegister.Index,
+         chan_index );
+      break;
+
+   default:
+      assert( 0 );
+   }
+
+   switch( inst->Instruction.Saturate ) {
+   case TGSI_SAT_NONE:
+      break;
+
+   case TGSI_SAT_ZERO_ONE:
+//      assert( 0 );
+      break;
+
+   case TGSI_SAT_MINUS_PLUS_ONE:
+      assert( 0 );
+      break;
+   }
+}
+
+#define STORE( FUNC, INST, XMM, INDEX, CHAN )\
+   emit_store( FUNC, XMM, &(INST).FullDstRegisters[INDEX], &(INST), CHAN )
+
+/**
+ * High-level instruction translators.
+ */
+
+static void
+emit_kil(
+   struct x86_function *func,
+   const struct tgsi_full_src_register *reg )
+{
+   unsigned uniquemask;
+   unsigned registers[4];
+   unsigned nextregister = 0;
+   unsigned firstchan = ~0;
+   unsigned chan_index;
+
+   /* This mask stores component bits that were already tested. Note that
+    * we test if the value is less than zero, so 1.0 and 0.0 need not to be
+    * tested. */
+   uniquemask = (1 << TGSI_EXTSWIZZLE_ZERO) | (1 << TGSI_EXTSWIZZLE_ONE);
+
+   FOR_EACH_CHANNEL( chan_index ) {
+      unsigned swizzle;
+
+      /* unswizzle channel */
+      swizzle = tgsi_util_get_full_src_register_extswizzle(
+         reg,
+         chan_index );
+
+      /* check if the component has not been already tested */
+      if( !(uniquemask & (1 << swizzle)) ) {
+         uniquemask |= 1 << swizzle;
+
+         /* allocate register */
+         registers[chan_index] = nextregister;
+         emit_fetch(
+            func,
+            nextregister,
+            reg,
+            chan_index );
+         nextregister++;
+
+         /* mark the first channel used */
+         if( firstchan == ~0 ) {
+            firstchan = chan_index;
+         }
+      }
+   }
+
+   emit_push(
+      func,
+      x86_make_reg( file_REG32, reg_AX ) );
+   emit_push(
+      func,
+      x86_make_reg( file_REG32, reg_DX ) );
+
+   FOR_EACH_CHANNEL( chan_index ) {
+      if( uniquemask & (1 << chan_index) ) {
+         emit_cmpps(
+            func,
+            make_xmm( registers[chan_index] ),
+            get_temp(
+               TGSI_EXEC_TEMP_00000000_I,
+               TGSI_EXEC_TEMP_00000000_C ),
+            cc_LessThan );
+
+         if( chan_index == firstchan ) {
+            emit_pmovmskb(
+               func,
+               x86_make_reg( file_REG32, reg_AX ),
+               make_xmm( registers[chan_index] ) );
+         }
+         else {
+            emit_pmovmskb(
+               func,
+               x86_make_reg( file_REG32, reg_DX ),
+               make_xmm( registers[chan_index] ) );
+            emit_or(
+               func,
+               x86_make_reg( file_REG32, reg_AX ),
+               x86_make_reg( file_REG32, reg_DX ) );
+         }
+      }
+   }
+
+   emit_or(
+      func,
+      get_temp(
+         TGSI_EXEC_TEMP_KILMASK_I,
+         TGSI_EXEC_TEMP_KILMASK_C ),
+      x86_make_reg( file_REG32, reg_AX ) );
+
+   emit_pop(
+      func,
+      x86_make_reg( file_REG32, reg_DX ) );
+   emit_pop(
+      func,
+      x86_make_reg( file_REG32, reg_AX ) );
+}
+
+static void
+emit_setcc(
+   struct x86_function *func,
+   struct tgsi_full_instruction *inst,
+   enum sse_cc cc )
+{
+   unsigned chan_index;
+
+   FOR_EACH_DST0_ENABLED_CHANNEL( *inst, chan_index ) {
+      FETCH( func, *inst, 0, 0, chan_index );
+      FETCH( func, *inst, 1, 1, chan_index );
+      emit_cmpps(
+         func,
+         make_xmm( 0 ),
+         make_xmm( 1 ),
+         cc );
+      emit_andps(
+         func,
+         make_xmm( 0 ),
+         get_temp(
+            TGSI_EXEC_TEMP_ONE_I,
+            TGSI_EXEC_TEMP_ONE_C ) );
+      STORE( func, *inst, 0, 0, chan_index );
+   }
+}
+
+static void
+emit_cmp(
+   struct x86_function *func,
+   struct tgsi_full_instruction *inst )
+{
+   unsigned chan_index;
+
+   FOR_EACH_DST0_ENABLED_CHANNEL( *inst, chan_index ) {
+      FETCH( func, *inst, 0, 0, chan_index );
+      FETCH( func, *inst, 1, 1, chan_index );
+      FETCH( func, *inst, 2, 2, chan_index );
+      emit_cmpps(
+         func,
+         make_xmm( 0 ),
+         get_temp(
+            TGSI_EXEC_TEMP_00000000_I,
+            TGSI_EXEC_TEMP_00000000_C ),
+         cc_LessThan );
+      emit_andps(
+         func,
+         make_xmm( 1 ),
+         make_xmm( 0 ) );
+      emit_andnps(
+         func,
+         make_xmm( 0 ),
+         make_xmm( 2 ) );
+      emit_orps(
+         func,
+         make_xmm( 0 ),
+         make_xmm( 1 ) );
+      STORE( func, *inst, 0, 0, chan_index );
+   }
+}
+
+static int
+emit_instruction(
+   struct x86_function *func,
+   struct tgsi_full_instruction *inst )
+{
+   unsigned chan_index;
+
+   switch( inst->Instruction.Opcode ) {
+   case TGSI_OPCODE_ARL:
+      FOR_EACH_DST0_ENABLED_CHANNEL( *inst, chan_index ) {
+         FETCH( func, *inst, 0, 0, chan_index );
+         emit_f2it( func, 0 );
+         STORE( func, *inst, 0, 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_MOV:
+   /* TGSI_OPCODE_SWZ */
+      FOR_EACH_DST0_ENABLED_CHANNEL( *inst, chan_index ) {
+         FETCH( func, *inst, 0, 0, chan_index );
+         STORE( func, *inst, 0, 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_LIT:
+      if( IS_DST0_CHANNEL_ENABLED( *inst, CHAN_X ) ||
+          IS_DST0_CHANNEL_ENABLED( *inst, CHAN_W ) ) {
+         emit_tempf(
+            func,
+            0,
+            TGSI_EXEC_TEMP_ONE_I,
+            TGSI_EXEC_TEMP_ONE_C);
+         if( IS_DST0_CHANNEL_ENABLED( *inst, CHAN_X ) ) {
+            STORE( func, *inst, 0, 0, CHAN_X );
+         }
+         if( IS_DST0_CHANNEL_ENABLED( *inst, CHAN_W ) ) {
+            STORE( func, *inst, 0, 0, CHAN_W );
+         }
+      }
+      if( IS_DST0_CHANNEL_ENABLED( *inst, CHAN_Y ) ||
+          IS_DST0_CHANNEL_ENABLED( *inst, CHAN_Z ) ) {
+         if( IS_DST0_CHANNEL_ENABLED( *inst, CHAN_Y ) ) {
+            FETCH( func, *inst, 0, 0, CHAN_X );
+            emit_maxps(
+               func,
+               make_xmm( 0 ),
+               get_temp(
+                  TGSI_EXEC_TEMP_00000000_I,
+                  TGSI_EXEC_TEMP_00000000_C ) );
+            STORE( func, *inst, 0, 0, CHAN_Y );
+         }
+         if( IS_DST0_CHANNEL_ENABLED( *inst, CHAN_Z ) ) {
+            FETCH( func, *inst, 1, 0, CHAN_Y );
+            emit_maxps(
+               func,
+               make_xmm( 1 ),
+               get_temp(
+                  TGSI_EXEC_TEMP_00000000_I,
+                  TGSI_EXEC_TEMP_00000000_C ) );
+            FETCH( func, *inst, 2, 0, CHAN_W );
+            emit_minps(
+               func,
+               make_xmm( 2 ),
+               get_temp(
+                  TGSI_EXEC_TEMP_128_I,
+                  TGSI_EXEC_TEMP_128_C ) );
+            emit_maxps(
+               func,
+               make_xmm( 2 ),
+               get_temp(
+                  TGSI_EXEC_TEMP_MINUS_128_I,
+                  TGSI_EXEC_TEMP_MINUS_128_C ) );
+            emit_pow( func, 1, 2 );
+            FETCH( func, *inst, 0, 0, CHAN_X );
+            emit_xorps(
+               func,
+               make_xmm( 2 ),
+               make_xmm( 2 ) );
+            emit_cmpps(
+               func,
+               make_xmm( 2 ),
+               make_xmm( 0 ),
+               cc_LessThanEqual );
+            emit_andps(
+               func,
+               make_xmm( 2 ),
+               make_xmm( 1 ) );
+            STORE( func, *inst, 2, 0, CHAN_Z );
+         }
+      }
+      break;
+
+   case TGSI_OPCODE_RCP:
+   /* TGSI_OPCODE_RECIP */
+      FETCH( func, *inst, 0, 0, CHAN_X );
+      emit_rcp( func, 0, 0 );
+      FOR_EACH_DST0_ENABLED_CHANNEL( *inst, chan_index ) {
+         STORE( func, *inst, 0, 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_RSQ:
+   /* TGSI_OPCODE_RECIPSQRT */
+      FETCH( func, *inst, 0, 0, CHAN_X );
+      emit_rsqrt( func, 0, 0 );
+      FOR_EACH_DST0_ENABLED_CHANNEL( *inst, chan_index ) {
+         STORE( func, *inst, 0, 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_EXP:
+      return 0;
+      break;
+
+   case TGSI_OPCODE_LOG:
+      return 0;
+      break;
+
+   case TGSI_OPCODE_MUL:
+      FOR_EACH_DST0_ENABLED_CHANNEL( *inst, chan_index ) {
+         FETCH( func, *inst, 0, 0, chan_index );
+         FETCH( func, *inst, 1, 1, chan_index );
+         emit_mul( func, 0, 1 );
+         STORE( func, *inst, 0, 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_ADD:
+      FOR_EACH_DST0_ENABLED_CHANNEL( *inst, chan_index ) {
+         FETCH( func, *inst, 0, 0, chan_index );
+         FETCH( func, *inst, 1, 1, chan_index );
+         emit_add( func, 0, 1 );
+         STORE( func, *inst, 0, 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_DP3:
+   /* TGSI_OPCODE_DOT3 */
+      FETCH( func, *inst, 0, 0, CHAN_X );
+      FETCH( func, *inst, 1, 1, CHAN_X );
+      emit_mul( func, 0, 1 );
+      FETCH( func, *inst, 1, 0, CHAN_Y );
+      FETCH( func, *inst, 2, 1, CHAN_Y );
+      emit_mul( func, 1, 2 );
+      emit_add( func, 0, 1 );
+      FETCH( func, *inst, 1, 0, CHAN_Z );
+      FETCH( func, *inst, 2, 1, CHAN_Z );
+      emit_mul( func, 1, 2 );
+      emit_add( func, 0, 1 );
+      FOR_EACH_DST0_ENABLED_CHANNEL( *inst, chan_index ) {
+         STORE( func, *inst, 0, 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_DP4:
+   /* TGSI_OPCODE_DOT4 */
+      FETCH( func, *inst, 0, 0, CHAN_X );
+      FETCH( func, *inst, 1, 1, CHAN_X );
+      emit_mul( func, 0, 1 );
+      FETCH( func, *inst, 1, 0, CHAN_Y );
+      FETCH( func, *inst, 2, 1, CHAN_Y );
+      emit_mul( func, 1, 2 );
+      emit_add( func, 0, 1 );
+      FETCH( func, *inst, 1, 0, CHAN_Z );
+      FETCH( func, *inst, 2, 1, CHAN_Z );
+      emit_mul(func, 1, 2 );
+      emit_add(func, 0, 1 );
+      FETCH( func, *inst, 1, 0, CHAN_W );
+      FETCH( func, *inst, 2, 1, CHAN_W );
+      emit_mul( func, 1, 2 );
+      emit_add( func, 0, 1 );
+      FOR_EACH_DST0_ENABLED_CHANNEL( *inst, chan_index ) {
+         STORE( func, *inst, 0, 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_DST:
+      IF_IS_DST0_CHANNEL_ENABLED( *inst, CHAN_X ) {
+         emit_tempf(
+            func,
+            0,
+            TGSI_EXEC_TEMP_ONE_I,
+            TGSI_EXEC_TEMP_ONE_C );
+         STORE( func, *inst, 0, 0, CHAN_X );
+      }
+      IF_IS_DST0_CHANNEL_ENABLED( *inst, CHAN_Y ) {
+         FETCH( func, *inst, 0, 0, CHAN_Y );
+         FETCH( func, *inst, 1, 1, CHAN_Y );
+         emit_mul( func, 0, 1 );
+         STORE( func, *inst, 0, 0, CHAN_Y );
+      }
+      IF_IS_DST0_CHANNEL_ENABLED( *inst, CHAN_Z ) {
+         FETCH( func, *inst, 0, 0, CHAN_Z );
+         STORE( func, *inst, 0, 0, CHAN_Z );
+      }
+      IF_IS_DST0_CHANNEL_ENABLED( *inst, CHAN_W ) {
+         FETCH( func, *inst, 0, 1, CHAN_W );
+         STORE( func, *inst, 0, 0, CHAN_W );
+      }
+      break;
+
+   case TGSI_OPCODE_MIN:
+      FOR_EACH_DST0_ENABLED_CHANNEL( *inst, chan_index ) {
+         FETCH( func, *inst, 0, 0, chan_index );
+         FETCH( func, *inst, 1, 1, chan_index );
+         emit_minps(
+            func,
+            make_xmm( 0 ),
+            make_xmm( 1 ) );
+         STORE( func, *inst, 0, 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_MAX:
+      FOR_EACH_DST0_ENABLED_CHANNEL( *inst, chan_index ) {
+         FETCH( func, *inst, 0, 0, chan_index );
+         FETCH( func, *inst, 1, 1, chan_index );
+         emit_maxps(
+            func,
+            make_xmm( 0 ),
+            make_xmm( 1 ) );
+         STORE( func, *inst, 0, 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_SLT:
+   /* TGSI_OPCODE_SETLT */
+      emit_setcc( func, inst, cc_LessThan );
+      break;
+
+   case TGSI_OPCODE_SGE:
+   /* TGSI_OPCODE_SETGE */
+      emit_setcc( func, inst, cc_NotLessThan );
+      break;
+
+   case TGSI_OPCODE_MAD:
+   /* TGSI_OPCODE_MADD */
+      FOR_EACH_DST0_ENABLED_CHANNEL( *inst, chan_index ) {
+         FETCH( func, *inst, 0, 0, chan_index );
+         FETCH( func, *inst, 1, 1, chan_index );
+         FETCH( func, *inst, 2, 2, chan_index );
+         emit_mul( func, 0, 1 );
+         emit_add( func, 0, 2 );
+         STORE( func, *inst, 0, 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_SUB:
+      FOR_EACH_DST0_ENABLED_CHANNEL( *inst, chan_index ) {
+         FETCH( func, *inst, 0, 0, chan_index );
+         FETCH( func, *inst, 1, 1, chan_index );
+         emit_sub( func, 0, 1 );
+         STORE( func, *inst, 0, 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_LERP:
+   /* TGSI_OPCODE_LRP */
+      FOR_EACH_DST0_ENABLED_CHANNEL( *inst, chan_index ) {
+         FETCH( func, *inst, 0, 0, chan_index );
+         FETCH( func, *inst, 1, 1, chan_index );
+         FETCH( func, *inst, 2, 2, chan_index );
+         emit_sub( func, 1, 2 );
+         emit_mul( func, 0, 1 );
+         emit_add( func, 0, 2 );
+         STORE( func, *inst, 0, 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_CND:
+      return 0;
+      break;
+
+   case TGSI_OPCODE_CND0:
+      return 0;
+      break;
+
+   case TGSI_OPCODE_DOT2ADD:
+   /* TGSI_OPCODE_DP2A */
+      return 0;
+      break;
+
+   case TGSI_OPCODE_INDEX:
+      return 0;
+      break;
+
+   case TGSI_OPCODE_NEGATE:
+      return 0;
+      break;
+
+   case TGSI_OPCODE_FRAC:
+   /* TGSI_OPCODE_FRC */
+      FOR_EACH_DST0_ENABLED_CHANNEL( *inst, chan_index ) {
+         FETCH( func, *inst, 0, 0, chan_index );
+         emit_frc( func, 0 );
+         STORE( func, *inst, 0, 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_CLAMP:
+      return 0;
+      break;
+
+   case TGSI_OPCODE_FLOOR:
+   /* TGSI_OPCODE_FLR */
+      FOR_EACH_DST0_ENABLED_CHANNEL( *inst, chan_index ) {
+         FETCH( func, *inst, 0, 0, chan_index );
+         emit_flr( func, 0 );
+         STORE( func, *inst, 0, 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_ROUND:
+      return 0;
+      break;
+
+   case TGSI_OPCODE_EXPBASE2:
+   /* TGSI_OPCODE_EX2 */
+      FETCH( func, *inst, 0, 0, CHAN_X );
+      emit_ex2( func, 0 );
+      FOR_EACH_DST0_ENABLED_CHANNEL( *inst, chan_index ) {
+         STORE( func, *inst, 0, 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_LOGBASE2:
+   /* TGSI_OPCODE_LG2 */
+      FETCH( func, *inst, 0, 0, CHAN_X );
+      emit_lg2( func, 0 );
+      FOR_EACH_DST0_ENABLED_CHANNEL( *inst, chan_index ) {
+         STORE( func, *inst, 0, 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_POWER:
+   /* TGSI_OPCODE_POW */
+      FETCH( func, *inst, 0, 0, CHAN_X );
+      FETCH( func, *inst, 1, 1, CHAN_X );
+      emit_pow( func, 0, 1 );
+      FOR_EACH_DST0_ENABLED_CHANNEL( *inst, chan_index ) {
+         STORE( func, *inst, 0, 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_CROSSPRODUCT:
+   /* TGSI_OPCODE_XPD */
+      if( IS_DST0_CHANNEL_ENABLED( *inst, CHAN_X ) ||
+          IS_DST0_CHANNEL_ENABLED( *inst, CHAN_Y ) ) {
+         FETCH( func, *inst, 1, 1, CHAN_Z );
+         FETCH( func, *inst, 3, 0, CHAN_Z );
+      }
+      if( IS_DST0_CHANNEL_ENABLED( *inst, CHAN_X ) ||
+          IS_DST0_CHANNEL_ENABLED( *inst, CHAN_Z ) ) {
+         FETCH( func, *inst, 0, 0, CHAN_Y );
+         FETCH( func, *inst, 4, 1, CHAN_Y );
+      }
+      IF_IS_DST0_CHANNEL_ENABLED( *inst, CHAN_X ) {
+         emit_MOV( func, 2, 0 );
+         emit_mul( func, 2, 1 );
+         emit_MOV( func, 5, 3 );
+         emit_mul( func, 5, 4 );
+         emit_sub( func, 2, 5 );
+         STORE( func, *inst, 2, 0, CHAN_X );
+      }
+      if( IS_DST0_CHANNEL_ENABLED( *inst, CHAN_Y ) ||
+          IS_DST0_CHANNEL_ENABLED( *inst, CHAN_Z ) ) {
+         FETCH( func, *inst, 2, 1, CHAN_X );
+         FETCH( func, *inst, 5, 0, CHAN_X );
+      }
+      IF_IS_DST0_CHANNEL_ENABLED( *inst, CHAN_Y ) {
+         emit_mul( func, 3, 2 );
+         emit_mul( func, 1, 5 );
+         emit_sub( func, 3, 1 );
+         STORE( func, *inst, 3, 0, CHAN_Y );
+      }
+      IF_IS_DST0_CHANNEL_ENABLED( *inst, CHAN_Z ) {
+         emit_mul( func, 5, 4 );
+         emit_mul( func, 0, 2 );
+         emit_sub( func, 5, 0 );
+         STORE( func, *inst, 5, 0, CHAN_Z );
+      }
+      IF_IS_DST0_CHANNEL_ENABLED( *inst, CHAN_W ) {
+         FETCH( func, *inst, 0, TGSI_EXEC_TEMP_ONE_I, TGSI_EXEC_TEMP_ONE_C );
+         STORE( func, *inst, 0, 0, CHAN_W );
+      }
+      break;
+
+   case TGSI_OPCODE_MULTIPLYMATRIX:
+      return 0;
+      break;
+
+   case TGSI_OPCODE_ABS:
+      FOR_EACH_DST0_ENABLED_CHANNEL( *inst, chan_index ) {
+         FETCH( func, *inst, 0, 0, chan_index );
+         emit_abs( func, 0) ;
+
+         STORE( func, *inst, 0, 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_RCC:
+      return 0;
+      break;
+
+   case TGSI_OPCODE_DPH:
+      FETCH( func, *inst, 0, 0, CHAN_X );
+      FETCH( func, *inst, 1, 1, CHAN_X );
+      emit_mul( func, 0, 1 );
+      FETCH( func, *inst, 1, 0, CHAN_Y );
+      FETCH( func, *inst, 2, 1, CHAN_Y );
+      emit_mul( func, 1, 2 );
+      emit_add( func, 0, 1 );
+      FETCH( func, *inst, 1, 0, CHAN_Z );
+      FETCH( func, *inst, 2, 1, CHAN_Z );
+      emit_mul( func, 1, 2 );
+      emit_add( func, 0, 1 );
+      FETCH( func, *inst, 1, 1, CHAN_W );
+      emit_add( func, 0, 1 );
+      FOR_EACH_DST0_ENABLED_CHANNEL( *inst, chan_index ) {
+         STORE( func, *inst, 0, 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_COS:
+      FETCH( func, *inst, 0, 0, CHAN_X );
+      emit_cos( func, 0 );
+      FOR_EACH_DST0_ENABLED_CHANNEL( *inst, chan_index ) {
+         STORE( func, *inst, 0, 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_DDX:
+      return 0;
+      break;
+
+   case TGSI_OPCODE_DDY:
+      return 0;
+      break;
+
+   case TGSI_OPCODE_KIL:
+      emit_kil( func, &inst->FullSrcRegisters[0] );
+      break;
+
+   case TGSI_OPCODE_PK2H:
+      return 0;
+      break;
+
+   case TGSI_OPCODE_PK2US:
+      return 0;
+      break;
+
+   case TGSI_OPCODE_PK4B:
+      return 0;
+      break;
+
+   case TGSI_OPCODE_PK4UB:
+      return 0;
+      break;
+
+   case TGSI_OPCODE_RFL:
+      return 0;
+      break;
+
+   case TGSI_OPCODE_SEQ:
+      return 0;
+      break;
+
+   case TGSI_OPCODE_SFL:
+      return 0;
+      break;
+
+   case TGSI_OPCODE_SGT:
+      return 0;
+      break;
+
+   case TGSI_OPCODE_SIN:
+      FETCH( func, *inst, 0, 0, CHAN_X );
+      emit_sin( func, 0 );
+      FOR_EACH_DST0_ENABLED_CHANNEL( *inst, chan_index ) {
+         STORE( func, *inst, 0, 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_SLE:
+      return 0;
+      break;
+
+   case TGSI_OPCODE_SNE:
+      return 0;
+      break;
+
+   case TGSI_OPCODE_STR:
+      return 0;
+      break;
+
+   case TGSI_OPCODE_TEX:
+      emit_tempf(
+         func,
+         0,
+         TGSI_EXEC_TEMP_ONE_I,
+         TGSI_EXEC_TEMP_ONE_C );
+      FOR_EACH_DST0_ENABLED_CHANNEL( *inst, chan_index ) {
+         STORE( func, *inst, 0, 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_TXD:
+      return 0;
+      break;
+
+   case TGSI_OPCODE_UP2H:
+      return 0;
+      break;
+
+   case TGSI_OPCODE_UP2US:
+      return 0;
+      break;
+
+   case TGSI_OPCODE_UP4B:
+      return 0;
+      break;
+
+   case TGSI_OPCODE_UP4UB:
+      return 0;
+      break;
+
+   case TGSI_OPCODE_X2D:
+      return 0;
+      break;
+
+   case TGSI_OPCODE_ARA:
+      return 0;
+      break;
+
+   case TGSI_OPCODE_ARR:
+      return 0;
+      break;
+
+   case TGSI_OPCODE_BRA:
+      return 0;
+      break;
+
+   case TGSI_OPCODE_CAL:
+      return 0;
+      break;
+
+   case TGSI_OPCODE_RET:
+   case TGSI_OPCODE_END:
+#ifdef WIN32
+      emit_retw( func, 16 );
+#else
+      emit_ret( func );
+#endif
+      break;
+
+   case TGSI_OPCODE_SSG:
+      return 0;
+      break;
+
+   case TGSI_OPCODE_CMP:
+      emit_cmp (func, inst);
+      break;
+
+   case TGSI_OPCODE_SCS:
+      IF_IS_DST0_CHANNEL_ENABLED( *inst, CHAN_X ) {
+         FETCH( func, *inst, 0, 0, CHAN_X );
+         emit_cos( func, 0 );
+         STORE( func, *inst, 0, 0, CHAN_X );
+      }
+      IF_IS_DST0_CHANNEL_ENABLED( *inst, CHAN_Y ) {
+         FETCH( func, *inst, 0, 0, CHAN_Y );
+         emit_sin( func, 0 );
+         STORE( func, *inst, 0, 0, CHAN_Y );
+      }
+      IF_IS_DST0_CHANNEL_ENABLED( *inst, CHAN_Z ) {
+         FETCH( func, *inst, 0, TGSI_EXEC_TEMP_00000000_I, TGSI_EXEC_TEMP_00000000_C );
+         STORE( func, *inst, 0, 0, CHAN_Z );
+      }
+      IF_IS_DST0_CHANNEL_ENABLED( *inst, CHAN_W ) {
+         FETCH( func, *inst, 0, TGSI_EXEC_TEMP_ONE_I, TGSI_EXEC_TEMP_ONE_C );
+         STORE( func, *inst, 0, 0, CHAN_W );
+      }
+      break;
+
+   case TGSI_OPCODE_TXB:
+      return 0;
+      break;
+
+   case TGSI_OPCODE_NRM:
+      return 0;
+      break;
+
+   case TGSI_OPCODE_DIV:
+      return 0;
+      break;
+
+   case TGSI_OPCODE_DP2:
+      return 0;
+      break;
+
+   case TGSI_OPCODE_TXL:
+      return 0;
+      break;
+
+   case TGSI_OPCODE_BRK:
+      return 0;
+      break;
+
+   case TGSI_OPCODE_IF:
+      return 0;
+      break;
+
+   case TGSI_OPCODE_LOOP:
+      return 0;
+      break;
+
+   case TGSI_OPCODE_REP:
+      return 0;
+      break;
+
+   case TGSI_OPCODE_ELSE:
+      return 0;
+      break;
+
+   case TGSI_OPCODE_ENDIF:
+      return 0;
+      break;
+
+   case TGSI_OPCODE_ENDLOOP:
+      return 0;
+      break;
+
+   case TGSI_OPCODE_ENDREP:
+      return 0;
+      break;
+
+   case TGSI_OPCODE_PUSHA:
+      return 0;
+      break;
+
+   case TGSI_OPCODE_POPA:
+      return 0;
+      break;
+
+   case TGSI_OPCODE_CEIL:
+      return 0;
+      break;
+
+   case TGSI_OPCODE_I2F:
+      return 0;
+      break;
+
+   case TGSI_OPCODE_NOT:
+      return 0;
+      break;
+
+   case TGSI_OPCODE_TRUNC:
+      return 0;
+      break;
+
+   case TGSI_OPCODE_SHL:
+      return 0;
+      break;
+
+   case TGSI_OPCODE_SHR:
+      return 0;
+      break;
+
+   case TGSI_OPCODE_AND:
+      return 0;
+      break;
+
+   case TGSI_OPCODE_OR:
+      return 0;
+      break;
+
+   case TGSI_OPCODE_MOD:
+      return 0;
+      break;
+
+   case TGSI_OPCODE_XOR:
+      return 0;
+      break;
+
+   case TGSI_OPCODE_SAD:
+      return 0;
+      break;
+
+   case TGSI_OPCODE_TXF:
+      return 0;
+      break;
+
+   case TGSI_OPCODE_TXQ:
+      return 0;
+      break;
+
+   case TGSI_OPCODE_CONT:
+      return 0;
+      break;
+
+   case TGSI_OPCODE_EMIT:
+      return 0;
+      break;
+
+   case TGSI_OPCODE_ENDPRIM:
+      return 0;
+      break;
+
+   default:
+      return 0;
+   }
+   
+   return 1;
+}
+
+static void
+emit_declaration(
+   struct x86_function *func,
+   struct tgsi_full_declaration *decl )
+{
+   if( decl->Declaration.File == TGSI_FILE_INPUT ) {
+      unsigned first, last, mask;
+      unsigned i, j;
+
+      assert( decl->Declaration.Declare == TGSI_DECLARE_RANGE );
+
+      first = decl->u.DeclarationRange.First;
+      last = decl->u.DeclarationRange.Last;
+      mask = decl->Declaration.UsageMask;
+
+      /* Do not touch WPOS.xy */
+      if( first == 0 ) {
+         mask &= ~TGSI_WRITEMASK_XY;
+         if( mask == TGSI_WRITEMASK_NONE ) {
+            first++;
+         }
+      }
+
+      for( i = first; i <= last; i++ ) {
+         for( j = 0; j < NUM_CHANNELS; j++ ) {
+            if( mask & (1 << j) ) {
+               switch( decl->Interpolation.Interpolate ) {
+               case TGSI_INTERPOLATE_CONSTANT:
+                  emit_coef_a0( func, 0, i, j );
+                  emit_inputs( func, 0, i, j );
+                  break;
+
+               case TGSI_INTERPOLATE_LINEAR:
+                  emit_inputf( func, 0, 0, TGSI_SWIZZLE_X );
+                  emit_coef_dadx( func, 1, i, j );
+                  emit_inputf( func, 2, 0, TGSI_SWIZZLE_Y );
+                  emit_coef_dady( func, 3, i, j );
+                  emit_mul( func, 0, 1 );    /* x * dadx */
+                  emit_coef_a0( func, 4, i, j );
+                  emit_mul( func, 2, 3 );    /* y * dady */
+                  emit_add( func, 0, 4 );    /* x * dadx + a0 */
+                  emit_add( func, 0, 2 );    /* x * dadx + y * dady + a0 */
+                  emit_inputs( func, 0, i, j );
+                  break;
+
+               case TGSI_INTERPOLATE_PERSPECTIVE:
+                  emit_inputf( func, 0, 0, TGSI_SWIZZLE_X );
+                  emit_coef_dadx( func, 1, i, j );
+                  emit_inputf( func, 2, 0, TGSI_SWIZZLE_Y );
+                  emit_coef_dady( func, 3, i, j );
+                  emit_mul( func, 0, 1 );    /* x * dadx */
+                  emit_inputf( func, 4, 0, TGSI_SWIZZLE_W );
+                  emit_coef_a0( func, 5, i, j );
+                  emit_rcp( func, 4, 4 );    /* 1.0 / w */
+                  emit_mul( func, 2, 3 );    /* y * dady */
+                  emit_add( func, 0, 5 );    /* x * dadx + a0 */
+                  emit_add( func, 0, 2 );    /* x * dadx + y * dady + a0 */
+                  emit_mul( func, 0, 4 );    /* (x * dadx + y * dady + a0) / w */
+                  emit_inputs( func, 0, i, j );
+                  break;
+
+               default:
+                  assert( 0 );
+		  break;
+               }
+            }
+         }
+      }
+   }
+}
+
+unsigned
+tgsi_emit_sse2(
+   struct tgsi_token *tokens,
+   struct x86_function *func )
+{
+   struct tgsi_parse_context parse;
+   unsigned ok = 1;
+
+   DUMP_START();
+
+   func->csr = func->store;
+
+   emit_mov(
+      func,
+      get_input_base(),
+      get_argument( 0 ) );
+   emit_mov(
+      func,
+      get_output_base(),
+      get_argument( 1 ) );
+   emit_mov(
+      func,
+      get_const_base(),
+      get_argument( 2 ) );
+   emit_mov(
+      func,
+      get_temp_base(),
+      get_argument( 3 ) );
+
+   tgsi_parse_init( &parse, tokens );
+
+   while( !tgsi_parse_end_of_tokens( &parse ) && ok ) {
+      tgsi_parse_token( &parse );
+
+      switch( parse.FullToken.Token.Type ) {
+      case TGSI_TOKEN_TYPE_DECLARATION:
+         break;
+
+      case TGSI_TOKEN_TYPE_INSTRUCTION:
+         ok = emit_instruction(
+	    func,
+	    &parse.FullToken.FullInstruction );
+
+	 if (!ok) {
+	    debug_printf("failed to translate tgsi opcode %d\n", 
+			 parse.FullToken.FullInstruction.Instruction.Opcode );
+	 }
+         break;
+
+      case TGSI_TOKEN_TYPE_IMMEDIATE:
+         /* XXX implement this */
+	 ok = 0;
+	 debug_printf("failed to emit immediate value\n");
+	 break;
+
+      default:
+         assert( 0 );
+	 ok = 0;
+	 break;
+      }
+   }
+
+   tgsi_parse_free( &parse );
+
+   DUMP_END();
+
+   return ok;
+}
+
+/**
+ * Fragment shaders are responsible for interpolating shader inputs. Because on
+ * x86 we have only 4 GP registers, and here we have 5 shader arguments (input,
+ * output, const, temp and coef), the code is split into two phases --
+ * DECLARATION and INSTRUCTION phase.
+ * GP register holding the output argument is aliased with the coeff argument,
+ * as outputs are not needed in the DECLARATION phase.
+ */
+unsigned
+tgsi_emit_sse2_fs(
+   struct tgsi_token *tokens,
+   struct x86_function *func )
+{
+   struct tgsi_parse_context parse;
+   boolean instruction_phase = FALSE;
+
+   DUMP_START();
+
+   func->csr = func->store;
+
+   /* DECLARATION phase, do not load output argument. */
+   emit_mov(
+      func,
+      get_input_base(),
+      get_argument( 0 ) );
+   emit_mov(
+      func,
+      get_const_base(),
+      get_argument( 2 ) );
+   emit_mov(
+      func,
+      get_temp_base(),
+      get_argument( 3 ) );
+   emit_mov(
+      func,
+      get_coef_base(),
+      get_argument( 4 ) );
+
+   tgsi_parse_init( &parse, tokens );
+
+   while( !tgsi_parse_end_of_tokens( &parse ) ) {
+      tgsi_parse_token( &parse );
+
+      switch( parse.FullToken.Token.Type ) {
+      case TGSI_TOKEN_TYPE_DECLARATION:
+         emit_declaration(
+            func,
+            &parse.FullToken.FullDeclaration );
+         break;
+
+      case TGSI_TOKEN_TYPE_INSTRUCTION:
+         if( !instruction_phase ) {
+            /* INSTRUCTION phase, overwrite coeff with output. */
+            instruction_phase = TRUE;
+            emit_mov(
+               func,
+               get_output_base(),
+               get_argument( 1 ) );
+         }
+         emit_instruction(
+            func,
+            &parse.FullToken.FullInstruction );
+         break;
+
+      case TGSI_TOKEN_TYPE_IMMEDIATE:
+         /* XXX implement this */
+         assert(0);
+         break;
+
+      default:
+         assert( 0 );
+      }
+   }
+
+   tgsi_parse_free( &parse );
+
+   DUMP_END();
+
+   return 1;
+}
+
+#endif /* i386 */
diff --git a/src/gallium/aux/tgsi/exec/tgsi_sse2.h b/src/gallium/aux/tgsi/exec/tgsi_sse2.h
new file mode 100755
index 0000000000..9bee371766
--- /dev/null
+++ b/src/gallium/aux/tgsi/exec/tgsi_sse2.h
@@ -0,0 +1,26 @@
+#if !defined TGSI_SSE2_H
+#define TGSI_SSE2_H
+
+#if defined __cplusplus
+extern "C" {
+#endif // defined __cplusplus
+
+struct tgsi_token;
+struct x86_function;
+
+unsigned
+tgsi_emit_sse2(
+   struct tgsi_token *tokens,
+   struct x86_function *function );
+
+unsigned
+tgsi_emit_sse2_fs(
+   struct tgsi_token *tokens,
+   struct x86_function *function );
+
+#if defined __cplusplus
+} // extern "C"
+#endif // defined __cplusplus
+
+#endif // !defined TGSI_SSE2_H
+
diff --git a/src/gallium/aux/tgsi/util/tgsi_build.c b/src/gallium/aux/tgsi/util/tgsi_build.c
new file mode 100644
index 0000000000..a00ff1c2a5
--- /dev/null
+++ b/src/gallium/aux/tgsi/util/tgsi_build.c
@@ -0,0 +1,1371 @@
+#include "pipe/p_debug.h"
+#include "pipe/p_util.h"
+#include "pipe/p_shader_tokens.h"
+#include "tgsi_build.h"
+#include "tgsi_parse.h"
+
+/*
+ * version
+ */
+
+struct tgsi_version
+tgsi_build_version( void )
+{
+   struct tgsi_version  version;
+
+   version.MajorVersion = 1;
+   version.MinorVersion = 1;
+   version.Padding = 0;
+
+   return version;
+}
+
+/*
+ * header
+ */
+
+struct tgsi_header
+tgsi_build_header( void )
+{
+   struct tgsi_header header;
+
+   header.HeaderSize = 1;
+   header.BodySize = 0;
+
+   return header;
+}
+
+static void
+header_headersize_grow( struct tgsi_header *header )
+{
+   assert( header->HeaderSize < 0xFF );
+   assert( header->BodySize == 0 );
+
+   header->HeaderSize++;
+}
+
+static void
+header_bodysize_grow( struct tgsi_header *header )
+{
+   assert( header->BodySize < 0xFFFFFF );
+
+   header->BodySize++;
+}
+
+struct tgsi_processor
+tgsi_default_processor( void )
+{
+   struct tgsi_processor processor;
+
+   processor.Processor = TGSI_PROCESSOR_FRAGMENT;
+   processor.Padding = 0;
+
+   return processor;
+}
+
+struct tgsi_processor
+tgsi_build_processor(
+   unsigned type,
+   struct tgsi_header *header )
+{
+   struct tgsi_processor processor;
+
+   processor = tgsi_default_processor();
+   processor.Processor = type;
+
+   header_headersize_grow( header );
+
+   return processor;
+}
+
+/*
+ * declaration
+ */
+
+struct tgsi_declaration
+tgsi_default_declaration( void )
+{
+   struct tgsi_declaration declaration;
+
+   declaration.Type = TGSI_TOKEN_TYPE_DECLARATION;
+   declaration.Size = 1;
+   declaration.File = TGSI_FILE_NULL;
+   declaration.Declare = TGSI_DECLARE_RANGE;
+   declaration.UsageMask = TGSI_WRITEMASK_XYZW;
+   declaration.Interpolate = 0;
+   declaration.Semantic = 0;
+   declaration.Padding = 0;
+   declaration.Extended = 0;
+
+   return declaration;
+}
+
+struct tgsi_declaration
+tgsi_build_declaration(
+   unsigned file,
+   unsigned declare,
+   unsigned usage_mask,
+   unsigned interpolate,
+   unsigned semantic,
+   struct tgsi_header *header )
+{
+   struct tgsi_declaration declaration;
+
+   assert( file <= TGSI_FILE_IMMEDIATE );
+   assert( declare <= TGSI_DECLARE_MASK );
+
+   declaration = tgsi_default_declaration();
+   declaration.File = file;
+   declaration.Declare = declare;
+   declaration.UsageMask = usage_mask;
+   declaration.Interpolate = interpolate;
+   declaration.Semantic = semantic;
+
+   header_bodysize_grow( header );
+
+   return declaration;
+}
+
+static void
+declaration_grow(
+   struct tgsi_declaration *declaration,
+   struct tgsi_header *header )
+{
+   assert( declaration->Size < 0xFF );
+
+   declaration->Size++;
+
+   header_bodysize_grow( header );
+}
+
+struct tgsi_full_declaration
+tgsi_default_full_declaration( void )
+{
+   struct tgsi_full_declaration  full_declaration;
+
+   full_declaration.Declaration  = tgsi_default_declaration();
+   full_declaration.Interpolation = tgsi_default_declaration_interpolation();
+   full_declaration.Semantic = tgsi_default_declaration_semantic();
+
+   return full_declaration;
+}
+
+unsigned
+tgsi_build_full_declaration(
+   const struct tgsi_full_declaration *full_decl,
+   struct tgsi_token *tokens,
+   struct tgsi_header *header,
+   unsigned maxsize )
+{
+   unsigned size = 0;
+   struct tgsi_declaration *declaration;
+
+   if( maxsize <= size )
+     return 0;
+   declaration = (struct tgsi_declaration *) &tokens[size];
+   size++;
+
+   *declaration = tgsi_build_declaration(
+      full_decl->Declaration.File,
+      full_decl->Declaration.Declare,
+      full_decl->Declaration.UsageMask,
+      full_decl->Declaration.Interpolate,
+      full_decl->Declaration.Semantic,
+      header );
+
+   switch( full_decl->Declaration.Declare )  {
+   case TGSI_DECLARE_RANGE:
+   {
+      struct tgsi_declaration_range *dr;
+
+      if( maxsize <= size )
+         return 0;
+      dr = (struct tgsi_declaration_range *) &tokens[size];
+      size++;
+
+      *dr = tgsi_build_declaration_range(
+         full_decl->u.DeclarationRange.First,
+         full_decl->u.DeclarationRange.Last,
+         declaration,
+         header );
+      break;
+    }
+
+   case TGSI_DECLARE_MASK:
+   {
+      struct tgsi_declaration_mask *dm;
+
+      if( maxsize <= size )
+         return 0;
+      dm = (struct tgsi_declaration_mask  *) &tokens[size];
+      size++;
+
+      *dm = tgsi_build_declaration_mask(
+         full_decl->u.DeclarationMask.Mask,
+         declaration,
+         header );
+      break;
+   }
+
+   default:
+      assert( 0 );
+   }
+
+   if( full_decl->Declaration.Interpolate ) {
+      struct tgsi_declaration_interpolation *di;
+
+      if( maxsize <= size )
+         return  0;
+      di = (struct tgsi_declaration_interpolation *) &tokens[size];
+      size++;
+
+      *di = tgsi_build_declaration_interpolation(
+         full_decl->Interpolation.Interpolate,
+         declaration,
+         header );
+   }
+
+   if( full_decl->Declaration.Semantic ) {
+      struct tgsi_declaration_semantic *ds;
+
+      if( maxsize <= size )
+         return  0;
+      ds = (struct tgsi_declaration_semantic *) &tokens[size];
+      size++;
+
+      *ds = tgsi_build_declaration_semantic(
+         full_decl->Semantic.SemanticName,
+         full_decl->Semantic.SemanticIndex,
+         declaration,
+         header );
+   }
+
+   return size;
+}
+
+struct tgsi_declaration_range
+tgsi_build_declaration_range(
+   unsigned first,
+   unsigned last,
+   struct tgsi_declaration *declaration,
+   struct tgsi_header *header )
+{
+   struct tgsi_declaration_range declaration_range;
+
+   assert( last >= first );
+   assert( last <= 0xFFFF );
+
+   declaration_range.First = first;
+   declaration_range.Last = last;
+
+   declaration_grow( declaration, header );
+
+   return declaration_range;
+}
+
+struct tgsi_declaration_mask
+tgsi_build_declaration_mask(
+   unsigned mask,
+   struct tgsi_declaration *declaration,
+   struct tgsi_header *header )
+{
+   struct tgsi_declaration_mask  declaration_mask;
+
+   declaration_mask.Mask = mask;
+
+   declaration_grow( declaration, header );
+
+   return declaration_mask;
+}
+
+struct tgsi_declaration_interpolation
+tgsi_default_declaration_interpolation( void )
+{
+   struct tgsi_declaration_interpolation di;
+
+   di.Interpolate = TGSI_INTERPOLATE_CONSTANT;
+   di.Padding = 0;
+
+   return di;
+}
+
+struct tgsi_declaration_interpolation
+tgsi_build_declaration_interpolation(
+   unsigned interpolate,
+   struct tgsi_declaration *declaration,
+   struct tgsi_header *header )
+{
+   struct tgsi_declaration_interpolation di;
+
+   assert( interpolate <= TGSI_INTERPOLATE_PERSPECTIVE );
+
+   di = tgsi_default_declaration_interpolation();
+   di.Interpolate = interpolate;
+
+   declaration_grow( declaration, header );
+
+   return di;
+}
+
+struct tgsi_declaration_semantic
+tgsi_default_declaration_semantic( void )
+{
+   struct tgsi_declaration_semantic ds;
+
+   ds.SemanticName = TGSI_SEMANTIC_POSITION;
+   ds.SemanticIndex = 0;
+   ds.Padding = 0;
+
+   return ds;
+}
+
+struct tgsi_declaration_semantic
+tgsi_build_declaration_semantic(
+   unsigned semantic_name,
+   unsigned semantic_index,
+   struct tgsi_declaration *declaration,
+   struct tgsi_header *header )
+{
+   struct tgsi_declaration_semantic ds;
+
+   assert( semantic_name <= TGSI_SEMANTIC_COUNT );
+   assert( semantic_index <= 0xFFFF );
+
+   ds = tgsi_default_declaration_semantic();
+   ds.SemanticName = semantic_name;
+   ds.SemanticIndex = semantic_index;
+
+   declaration_grow( declaration, header );
+
+   return ds;
+}
+
+/*
+ * immediate
+ */
+
+struct tgsi_immediate
+tgsi_default_immediate( void )
+{
+   struct tgsi_immediate immediate;
+
+   immediate.Type = TGSI_TOKEN_TYPE_IMMEDIATE;
+   immediate.Size = 1;
+   immediate.DataType = TGSI_IMM_FLOAT32;
+   immediate.Padding = 0;
+   immediate.Extended = 0;
+
+   return immediate;
+}
+
+struct tgsi_immediate
+tgsi_build_immediate(
+   struct tgsi_header *header )
+{
+   struct tgsi_immediate immediate;
+
+   immediate = tgsi_default_immediate();
+
+   header_bodysize_grow( header );
+
+   return immediate;
+}
+
+struct tgsi_full_immediate
+tgsi_default_full_immediate( void )
+{
+   struct tgsi_full_immediate fullimm;
+
+   fullimm.Immediate = tgsi_default_immediate();
+   fullimm.u.Pointer = (void *) 0;
+
+   return fullimm;
+}
+
+static void
+immediate_grow(
+   struct tgsi_immediate *immediate,
+   struct tgsi_header *header )
+{
+   assert( immediate->Size < 0xFF );
+
+   immediate->Size++;
+
+   header_bodysize_grow( header );
+}
+
+struct tgsi_immediate_float32
+tgsi_build_immediate_float32(
+   float value,
+   struct tgsi_immediate *immediate,
+   struct tgsi_header *header )
+{
+   struct tgsi_immediate_float32 immediate_float32;
+
+   immediate_float32.Float = value;
+
+   immediate_grow( immediate, header );
+
+   return immediate_float32;
+}
+
+unsigned
+tgsi_build_full_immediate(
+   const struct tgsi_full_immediate *full_imm,
+   struct tgsi_token *tokens,
+   struct tgsi_header *header,
+   unsigned maxsize )
+{
+   unsigned size = 0, i;
+   struct tgsi_immediate *immediate;
+
+   if( maxsize <= size )
+      return 0;
+   immediate = (struct tgsi_immediate *) &tokens[size];
+   size++;
+
+   *immediate = tgsi_build_immediate( header );
+
+   for( i = 0; i < full_imm->Immediate.Size - 1; i++ ) {
+      struct tgsi_immediate_float32 *if32;
+
+      if( maxsize <= size )
+         return  0;
+      if32 = (struct tgsi_immediate_float32 *) &tokens[size];
+      size++;
+
+      *if32 = tgsi_build_immediate_float32(
+         full_imm->u.ImmediateFloat32[i].Float,
+         immediate,
+         header );
+   }
+
+   return size;
+}
+
+/*
+ * instruction
+ */
+
+struct tgsi_instruction
+tgsi_default_instruction( void )
+{
+   struct tgsi_instruction instruction;
+
+   instruction.Type = TGSI_TOKEN_TYPE_INSTRUCTION;
+   instruction.Size = 1;
+   instruction.Opcode = TGSI_OPCODE_MOV;
+   instruction.Saturate = TGSI_SAT_NONE;
+   instruction.NumDstRegs = 1;
+   instruction.NumSrcRegs = 1;
+   instruction.Padding  = 0;
+   instruction.Extended = 0;
+
+   return instruction;
+}
+
+struct tgsi_instruction
+tgsi_build_instruction(
+   unsigned opcode,
+   unsigned saturate,
+   unsigned num_dst_regs,
+   unsigned num_src_regs,
+   struct tgsi_header *header )
+{
+   struct tgsi_instruction instruction;
+
+   assert (opcode <= TGSI_OPCODE_LAST);
+   assert (saturate <= TGSI_SAT_MINUS_PLUS_ONE);
+   assert (num_dst_regs <= 3);
+   assert (num_src_regs <= 15);
+
+   instruction = tgsi_default_instruction();
+   instruction.Opcode = opcode;
+   instruction.Saturate = saturate;
+   instruction.NumDstRegs = num_dst_regs;
+   instruction.NumSrcRegs = num_src_regs;
+
+   header_bodysize_grow( header );
+
+   return instruction;
+}
+
+static void
+instruction_grow(
+   struct tgsi_instruction *instruction,
+   struct tgsi_header *header )
+{
+   assert (instruction->Size <   0xFF);
+
+   instruction->Size++;
+
+   header_bodysize_grow( header );
+}
+
+struct tgsi_full_instruction
+tgsi_default_full_instruction( void )
+{
+   struct tgsi_full_instruction full_instruction;
+   unsigned i;
+
+   full_instruction.Instruction = tgsi_default_instruction();
+   full_instruction.InstructionExtNv = tgsi_default_instruction_ext_nv();
+   full_instruction.InstructionExtLabel = tgsi_default_instruction_ext_label();
+   full_instruction.InstructionExtTexture = tgsi_default_instruction_ext_texture();
+   for( i = 0;  i < TGSI_FULL_MAX_DST_REGISTERS; i++ ) {
+      full_instruction.FullDstRegisters[i] = tgsi_default_full_dst_register();
+   }
+   for( i = 0;  i < TGSI_FULL_MAX_SRC_REGISTERS; i++ ) {
+      full_instruction.FullSrcRegisters[i] = tgsi_default_full_src_register();
+   }
+
+   return full_instruction;
+}
+
+unsigned
+tgsi_build_full_instruction(
+   const struct tgsi_full_instruction *full_inst,
+   struct  tgsi_token *tokens,
+   struct  tgsi_header *header,
+   unsigned  maxsize )
+{
+   unsigned size = 0;
+   unsigned i;
+   struct tgsi_instruction *instruction;
+   struct tgsi_token *prev_token;
+
+   if( maxsize <= size )
+      return 0;
+   instruction = (struct tgsi_instruction *) &tokens[size];
+   size++;
+
+   *instruction = tgsi_build_instruction(
+      full_inst->Instruction.Opcode,
+      full_inst->Instruction.Saturate,
+      full_inst->Instruction.NumDstRegs,
+      full_inst->Instruction.NumSrcRegs,
+      header );
+   prev_token = (struct tgsi_token  *) instruction;
+
+   if( tgsi_compare_instruction_ext_nv(
+         full_inst->InstructionExtNv,
+         tgsi_default_instruction_ext_nv() ) ) {
+      struct tgsi_instruction_ext_nv *instruction_ext_nv;
+
+      if( maxsize <= size )
+         return 0;
+      instruction_ext_nv =
+         (struct  tgsi_instruction_ext_nv *) &tokens[size];
+      size++;
+
+      *instruction_ext_nv  = tgsi_build_instruction_ext_nv(
+         full_inst->InstructionExtNv.Precision,
+         full_inst->InstructionExtNv.CondDstIndex,
+         full_inst->InstructionExtNv.CondFlowIndex,
+         full_inst->InstructionExtNv.CondMask,
+         full_inst->InstructionExtNv.CondSwizzleX,
+         full_inst->InstructionExtNv.CondSwizzleY,
+         full_inst->InstructionExtNv.CondSwizzleZ,
+         full_inst->InstructionExtNv.CondSwizzleW,
+         full_inst->InstructionExtNv.CondDstUpdate,
+         full_inst->InstructionExtNv.CondFlowEnable,
+         prev_token,
+         instruction,
+         header );
+      prev_token = (struct tgsi_token  *) instruction_ext_nv;
+   }
+
+   if( tgsi_compare_instruction_ext_label(
+         full_inst->InstructionExtLabel,
+         tgsi_default_instruction_ext_label() ) ) {
+      struct tgsi_instruction_ext_label *instruction_ext_label;
+
+      if( maxsize <= size )
+         return 0;
+      instruction_ext_label =
+         (struct  tgsi_instruction_ext_label *) &tokens[size];
+      size++;
+
+      *instruction_ext_label = tgsi_build_instruction_ext_label(
+         full_inst->InstructionExtLabel.Label,
+         prev_token,
+         instruction,
+         header );
+      prev_token = (struct tgsi_token  *) instruction_ext_label;
+   }
+
+   if( tgsi_compare_instruction_ext_texture(
+         full_inst->InstructionExtTexture,
+         tgsi_default_instruction_ext_texture() ) ) {
+      struct tgsi_instruction_ext_texture *instruction_ext_texture;
+
+      if( maxsize <= size )
+         return 0;
+      instruction_ext_texture =
+         (struct  tgsi_instruction_ext_texture *) &tokens[size];
+      size++;
+
+      *instruction_ext_texture = tgsi_build_instruction_ext_texture(
+         full_inst->InstructionExtTexture.Texture,
+         prev_token,
+         instruction,
+         header   );
+      prev_token = (struct tgsi_token  *) instruction_ext_texture;
+   }
+
+   for( i = 0;  i <   full_inst->Instruction.NumDstRegs; i++ ) {
+      const struct tgsi_full_dst_register *reg = &full_inst->FullDstRegisters[i];
+      struct tgsi_dst_register *dst_register;
+      struct tgsi_token *prev_token;
+
+      if( maxsize <= size )
+         return 0;
+      dst_register = (struct tgsi_dst_register *) &tokens[size];
+      size++;
+
+      *dst_register = tgsi_build_dst_register(
+         reg->DstRegister.File,
+         reg->DstRegister.WriteMask,
+         reg->DstRegister.Index,
+         instruction,
+         header );
+      prev_token = (struct tgsi_token  *) dst_register;
+
+      if( tgsi_compare_dst_register_ext_concode(
+            reg->DstRegisterExtConcode,
+            tgsi_default_dst_register_ext_concode() ) ) {
+         struct tgsi_dst_register_ext_concode *dst_register_ext_concode;
+
+         if( maxsize <= size )
+            return 0;
+         dst_register_ext_concode =
+            (struct  tgsi_dst_register_ext_concode *) &tokens[size];
+         size++;
+
+         *dst_register_ext_concode =   tgsi_build_dst_register_ext_concode(
+            reg->DstRegisterExtConcode.CondMask,
+            reg->DstRegisterExtConcode.CondSwizzleX,
+            reg->DstRegisterExtConcode.CondSwizzleY,
+            reg->DstRegisterExtConcode.CondSwizzleZ,
+            reg->DstRegisterExtConcode.CondSwizzleW,
+            reg->DstRegisterExtConcode.CondSrcIndex,
+            prev_token,
+            instruction,
+            header );
+         prev_token = (struct tgsi_token  *) dst_register_ext_concode;
+      }
+
+      if( tgsi_compare_dst_register_ext_modulate(
+            reg->DstRegisterExtModulate,
+            tgsi_default_dst_register_ext_modulate() ) ) {
+         struct tgsi_dst_register_ext_modulate *dst_register_ext_modulate;
+
+         if( maxsize <= size )
+            return 0;
+         dst_register_ext_modulate =
+            (struct  tgsi_dst_register_ext_modulate *) &tokens[size];
+         size++;
+
+         *dst_register_ext_modulate = tgsi_build_dst_register_ext_modulate(
+            reg->DstRegisterExtModulate.Modulate,
+            prev_token,
+            instruction,
+            header );
+         prev_token = (struct tgsi_token  *) dst_register_ext_modulate;
+      }
+   }
+
+   for( i = 0;  i < full_inst->Instruction.NumSrcRegs; i++ ) {
+      const struct tgsi_full_src_register *reg = &full_inst->FullSrcRegisters[i];
+      struct tgsi_src_register *src_register;
+      struct tgsi_token *prev_token;
+
+      if( maxsize <= size )
+         return 0;
+      src_register = (struct tgsi_src_register *)  &tokens[size];
+      size++;
+
+      *src_register = tgsi_build_src_register(
+         reg->SrcRegister.File,
+         reg->SrcRegister.SwizzleX,
+         reg->SrcRegister.SwizzleY,
+         reg->SrcRegister.SwizzleZ,
+         reg->SrcRegister.SwizzleW,
+         reg->SrcRegister.Negate,
+         reg->SrcRegister.Indirect,
+         reg->SrcRegister.Dimension,
+         reg->SrcRegister.Index,
+         instruction,
+         header );
+      prev_token = (struct tgsi_token  *) src_register;
+
+      if( tgsi_compare_src_register_ext_swz(
+            reg->SrcRegisterExtSwz,
+            tgsi_default_src_register_ext_swz() ) ) {
+         struct tgsi_src_register_ext_swz *src_register_ext_swz;
+
+         if( maxsize <= size )
+            return 0;
+         src_register_ext_swz =
+            (struct  tgsi_src_register_ext_swz *) &tokens[size];
+         size++;
+
+         *src_register_ext_swz = tgsi_build_src_register_ext_swz(
+            reg->SrcRegisterExtSwz.ExtSwizzleX,
+            reg->SrcRegisterExtSwz.ExtSwizzleY,
+            reg->SrcRegisterExtSwz.ExtSwizzleZ,
+            reg->SrcRegisterExtSwz.ExtSwizzleW,
+            reg->SrcRegisterExtSwz.NegateX,
+            reg->SrcRegisterExtSwz.NegateY,
+            reg->SrcRegisterExtSwz.NegateZ,
+            reg->SrcRegisterExtSwz.NegateW,
+            reg->SrcRegisterExtSwz.ExtDivide,
+            prev_token,
+            instruction,
+            header );
+         prev_token = (struct tgsi_token  *) src_register_ext_swz;
+      }
+
+      if( tgsi_compare_src_register_ext_mod(
+            reg->SrcRegisterExtMod,
+            tgsi_default_src_register_ext_mod() ) ) {
+         struct tgsi_src_register_ext_mod *src_register_ext_mod;
+
+         if( maxsize <= size )
+            return 0;
+         src_register_ext_mod =
+            (struct  tgsi_src_register_ext_mod *) &tokens[size];
+         size++;
+
+         *src_register_ext_mod = tgsi_build_src_register_ext_mod(
+            reg->SrcRegisterExtMod.Complement,
+            reg->SrcRegisterExtMod.Bias,
+            reg->SrcRegisterExtMod.Scale2X,
+            reg->SrcRegisterExtMod.Absolute,
+            reg->SrcRegisterExtMod.Negate,
+            prev_token,
+            instruction,
+            header );
+         prev_token = (struct tgsi_token  *) src_register_ext_mod;
+      }
+
+      if( reg->SrcRegister.Indirect ) {
+         struct  tgsi_src_register *ind;
+
+         if( maxsize <= size )
+            return 0;
+         ind = (struct tgsi_src_register *) &tokens[size];
+         size++;
+
+         *ind = tgsi_build_src_register(
+            reg->SrcRegisterInd.File,
+            reg->SrcRegisterInd.SwizzleX,
+            reg->SrcRegisterInd.SwizzleY,
+            reg->SrcRegisterInd.SwizzleZ,
+            reg->SrcRegisterInd.SwizzleW,
+            reg->SrcRegisterInd.Negate,
+            reg->SrcRegisterInd.Indirect,
+            reg->SrcRegisterInd.Dimension,
+            reg->SrcRegisterInd.Index,
+            instruction,
+            header );
+      }
+
+      if( reg->SrcRegister.Dimension ) {
+         struct  tgsi_dimension *dim;
+
+         assert( !reg->SrcRegisterDim.Dimension );
+
+         if( maxsize <= size )
+            return 0;
+         dim = (struct tgsi_dimension *) &tokens[size];
+         size++;
+
+         *dim = tgsi_build_dimension(
+            reg->SrcRegisterDim.Indirect,
+            reg->SrcRegisterDim.Index,
+            instruction,
+            header );
+
+         if( reg->SrcRegisterDim.Indirect ) {
+            struct tgsi_src_register *ind;
+
+            if( maxsize <= size )
+               return 0;
+            ind = (struct tgsi_src_register *) &tokens[size];
+            size++;
+
+            *ind = tgsi_build_src_register(
+               reg->SrcRegisterDimInd.File,
+               reg->SrcRegisterDimInd.SwizzleX,
+               reg->SrcRegisterDimInd.SwizzleY,
+               reg->SrcRegisterDimInd.SwizzleZ,
+               reg->SrcRegisterDimInd.SwizzleW,
+               reg->SrcRegisterDimInd.Negate,
+               reg->SrcRegisterDimInd.Indirect,
+               reg->SrcRegisterDimInd.Dimension,
+               reg->SrcRegisterDimInd.Index,
+               instruction,
+               header );
+         }
+      }
+   }
+
+   return size;
+}
+
+struct tgsi_instruction_ext_nv
+tgsi_default_instruction_ext_nv( void )
+{
+   struct tgsi_instruction_ext_nv instruction_ext_nv;
+
+   instruction_ext_nv.Type = TGSI_INSTRUCTION_EXT_TYPE_NV;
+   instruction_ext_nv.Precision = TGSI_PRECISION_DEFAULT;
+   instruction_ext_nv.CondDstIndex = 0;
+   instruction_ext_nv.CondFlowIndex = 0;
+   instruction_ext_nv.CondMask = TGSI_CC_TR;
+   instruction_ext_nv.CondSwizzleX = TGSI_SWIZZLE_X;
+   instruction_ext_nv.CondSwizzleY = TGSI_SWIZZLE_Y;
+   instruction_ext_nv.CondSwizzleZ = TGSI_SWIZZLE_Z;
+   instruction_ext_nv.CondSwizzleW = TGSI_SWIZZLE_W;
+   instruction_ext_nv.CondDstUpdate = 0;
+   instruction_ext_nv.CondFlowEnable = 0;
+   instruction_ext_nv.Padding = 0;
+   instruction_ext_nv.Extended = 0;
+
+   return instruction_ext_nv;
+}
+
+union token_u32
+{
+   unsigned u32;
+};
+
+unsigned
+tgsi_compare_instruction_ext_nv(
+   struct tgsi_instruction_ext_nv a,
+   struct tgsi_instruction_ext_nv b )
+{
+   a.Padding = b.Padding = 0;
+   a.Extended = b.Extended = 0;
+   return ((union token_u32 *) &a)->u32 != ((union token_u32 *) &b)->u32;
+}
+
+struct tgsi_instruction_ext_nv
+tgsi_build_instruction_ext_nv(
+   unsigned precision,
+   unsigned cond_dst_index,
+   unsigned cond_flow_index,
+   unsigned cond_mask,
+   unsigned cond_swizzle_x,
+   unsigned cond_swizzle_y,
+   unsigned cond_swizzle_z,
+   unsigned cond_swizzle_w,
+   unsigned cond_dst_update,
+   unsigned cond_flow_update,
+   struct tgsi_token *prev_token,
+   struct tgsi_instruction *instruction,
+   struct tgsi_header *header )
+{
+   struct tgsi_instruction_ext_nv instruction_ext_nv;
+
+   instruction_ext_nv = tgsi_default_instruction_ext_nv();
+   instruction_ext_nv.Precision = precision;
+   instruction_ext_nv.CondDstIndex = cond_dst_index;
+   instruction_ext_nv.CondFlowIndex = cond_flow_index;
+   instruction_ext_nv.CondMask = cond_mask;
+   instruction_ext_nv.CondSwizzleX = cond_swizzle_x;
+   instruction_ext_nv.CondSwizzleY = cond_swizzle_y;
+   instruction_ext_nv.CondSwizzleZ = cond_swizzle_z;
+   instruction_ext_nv.CondSwizzleW = cond_swizzle_w;
+   instruction_ext_nv.CondDstUpdate = cond_dst_update;
+   instruction_ext_nv.CondFlowEnable = cond_flow_update;
+
+   prev_token->Extended = 1;
+   instruction_grow( instruction, header );
+
+   return instruction_ext_nv;
+}
+
+struct tgsi_instruction_ext_label
+tgsi_default_instruction_ext_label( void )
+{
+   struct tgsi_instruction_ext_label instruction_ext_label;
+
+   instruction_ext_label.Type = TGSI_INSTRUCTION_EXT_TYPE_LABEL;
+   instruction_ext_label.Label = 0;
+   instruction_ext_label.Padding = 0;
+   instruction_ext_label.Extended = 0;
+
+   return instruction_ext_label;
+}
+
+unsigned
+tgsi_compare_instruction_ext_label(
+   struct tgsi_instruction_ext_label a,
+   struct tgsi_instruction_ext_label b )
+{
+   a.Padding = b.Padding = 0;
+   a.Extended = b.Extended = 0;
+   return ((union token_u32 *) &a)->u32 != ((union token_u32 *) &b)->u32;
+}
+
+struct tgsi_instruction_ext_label
+tgsi_build_instruction_ext_label(
+   unsigned label,
+   struct tgsi_token  *prev_token,
+   struct tgsi_instruction *instruction,
+   struct tgsi_header *header )
+{
+   struct tgsi_instruction_ext_label instruction_ext_label;
+
+   instruction_ext_label = tgsi_default_instruction_ext_label();
+   instruction_ext_label.Label = label;
+
+   prev_token->Extended = 1;
+   instruction_grow( instruction, header );
+
+   return instruction_ext_label;
+}
+
+struct tgsi_instruction_ext_texture
+tgsi_default_instruction_ext_texture( void )
+{
+   struct tgsi_instruction_ext_texture instruction_ext_texture;
+
+   instruction_ext_texture.Type = TGSI_INSTRUCTION_EXT_TYPE_TEXTURE;
+   instruction_ext_texture.Texture = TGSI_TEXTURE_UNKNOWN;
+   instruction_ext_texture.Padding = 0;
+   instruction_ext_texture.Extended = 0;
+
+   return instruction_ext_texture;
+}
+
+unsigned
+tgsi_compare_instruction_ext_texture(
+   struct tgsi_instruction_ext_texture a,
+   struct tgsi_instruction_ext_texture b )
+{
+   a.Padding = b.Padding = 0;
+   a.Extended = b.Extended = 0;
+   return ((union token_u32 *) &a)->u32 != ((union token_u32 *) &b)->u32;
+}
+
+struct tgsi_instruction_ext_texture
+tgsi_build_instruction_ext_texture(
+   unsigned texture,
+   struct tgsi_token *prev_token,
+   struct tgsi_instruction *instruction,
+   struct tgsi_header *header )
+{
+   struct tgsi_instruction_ext_texture instruction_ext_texture;
+
+   instruction_ext_texture = tgsi_default_instruction_ext_texture();
+   instruction_ext_texture.Texture = texture;
+
+   prev_token->Extended = 1;
+   instruction_grow( instruction, header );
+
+   return instruction_ext_texture;
+}
+
+struct tgsi_src_register
+tgsi_default_src_register( void )
+{
+   struct tgsi_src_register src_register;
+
+   src_register.File = TGSI_FILE_NULL;
+   src_register.SwizzleX = TGSI_SWIZZLE_X;
+   src_register.SwizzleY = TGSI_SWIZZLE_Y;
+   src_register.SwizzleZ = TGSI_SWIZZLE_Z;
+   src_register.SwizzleW = TGSI_SWIZZLE_W;
+   src_register.Negate = 0;
+   src_register.Indirect = 0;
+   src_register.Dimension = 0;
+   src_register.Index = 0;
+   src_register.Extended = 0;
+
+   return src_register;
+}
+
+struct tgsi_src_register
+tgsi_build_src_register(
+   unsigned file,
+   unsigned swizzle_x,
+   unsigned swizzle_y,
+   unsigned swizzle_z,
+   unsigned swizzle_w,
+   unsigned negate,
+   unsigned indirect,
+   unsigned dimension,
+   int index,
+   struct tgsi_instruction *instruction,
+   struct tgsi_header *header )
+{
+   struct tgsi_src_register   src_register;
+
+   assert( file <= TGSI_FILE_IMMEDIATE );
+   assert( swizzle_x <= TGSI_SWIZZLE_W );
+   assert( swizzle_y <= TGSI_SWIZZLE_W );
+   assert( swizzle_z <= TGSI_SWIZZLE_W );
+   assert( swizzle_w <= TGSI_SWIZZLE_W );
+   assert( negate <= 1 );
+   assert( index >= -0x8000 && index <= 0x7FFF );
+
+   src_register = tgsi_default_src_register();
+   src_register.File = file;
+   src_register.SwizzleX = swizzle_x;
+   src_register.SwizzleY = swizzle_y;
+   src_register.SwizzleZ = swizzle_z;
+   src_register.SwizzleW = swizzle_w;
+   src_register.Negate = negate;
+   src_register.Indirect = indirect;
+   src_register.Dimension = dimension;
+   src_register.Index = index;
+
+   instruction_grow( instruction, header );
+
+   return src_register;
+}
+
+struct tgsi_full_src_register
+tgsi_default_full_src_register( void )
+{
+   struct tgsi_full_src_register full_src_register;
+
+   full_src_register.SrcRegister = tgsi_default_src_register();
+   full_src_register.SrcRegisterExtSwz = tgsi_default_src_register_ext_swz();
+   full_src_register.SrcRegisterExtMod = tgsi_default_src_register_ext_mod();
+   full_src_register.SrcRegisterInd = tgsi_default_src_register();
+   full_src_register.SrcRegisterDim = tgsi_default_dimension();
+   full_src_register.SrcRegisterDimInd = tgsi_default_src_register();
+
+   return full_src_register;
+}
+
+struct tgsi_src_register_ext_swz
+tgsi_default_src_register_ext_swz( void )
+{
+   struct tgsi_src_register_ext_swz src_register_ext_swz;
+
+   src_register_ext_swz.Type = TGSI_SRC_REGISTER_EXT_TYPE_SWZ;
+   src_register_ext_swz.ExtSwizzleX = TGSI_EXTSWIZZLE_X;
+   src_register_ext_swz.ExtSwizzleY = TGSI_EXTSWIZZLE_Y;
+   src_register_ext_swz.ExtSwizzleZ = TGSI_EXTSWIZZLE_Z;
+   src_register_ext_swz.ExtSwizzleW = TGSI_EXTSWIZZLE_W;
+   src_register_ext_swz.NegateX = 0;
+   src_register_ext_swz.NegateY = 0;
+   src_register_ext_swz.NegateZ = 0;
+   src_register_ext_swz.NegateW = 0;
+   src_register_ext_swz.ExtDivide = TGSI_EXTSWIZZLE_ONE;
+   src_register_ext_swz.Padding = 0;
+   src_register_ext_swz.Extended = 0;
+
+   return src_register_ext_swz;
+}
+
+unsigned
+tgsi_compare_src_register_ext_swz(
+   struct tgsi_src_register_ext_swz a,
+   struct tgsi_src_register_ext_swz b )
+{
+   a.Padding = b.Padding = 0;
+   a.Extended = b.Extended = 0;
+   return ((union token_u32 *) &a)->u32 != ((union token_u32 *) &b)->u32;
+}
+
+struct tgsi_src_register_ext_swz
+tgsi_build_src_register_ext_swz(
+   unsigned ext_swizzle_x,
+   unsigned ext_swizzle_y,
+   unsigned ext_swizzle_z,
+   unsigned ext_swizzle_w,
+   unsigned negate_x,
+   unsigned negate_y,
+   unsigned negate_z,
+   unsigned negate_w,
+   unsigned ext_divide,
+   struct tgsi_token *prev_token,
+   struct tgsi_instruction *instruction,
+   struct tgsi_header *header )
+{
+   struct tgsi_src_register_ext_swz src_register_ext_swz;
+
+   assert( ext_swizzle_x <= TGSI_EXTSWIZZLE_ONE );
+   assert( ext_swizzle_y <= TGSI_EXTSWIZZLE_ONE );
+   assert( ext_swizzle_z <= TGSI_EXTSWIZZLE_ONE );
+   assert( ext_swizzle_w <= TGSI_EXTSWIZZLE_ONE );
+   assert( negate_x <= 1 );
+   assert( negate_y <= 1 );
+   assert( negate_z <= 1 );
+   assert( negate_w <= 1 );
+   assert( ext_divide <= TGSI_EXTSWIZZLE_ONE );
+
+   src_register_ext_swz = tgsi_default_src_register_ext_swz();
+   src_register_ext_swz.ExtSwizzleX = ext_swizzle_x;
+   src_register_ext_swz.ExtSwizzleY = ext_swizzle_y;
+   src_register_ext_swz.ExtSwizzleZ = ext_swizzle_z;
+   src_register_ext_swz.ExtSwizzleW = ext_swizzle_w;
+   src_register_ext_swz.NegateX = negate_x;
+   src_register_ext_swz.NegateY = negate_y;
+   src_register_ext_swz.NegateZ = negate_z;
+   src_register_ext_swz.NegateW = negate_w;
+   src_register_ext_swz.ExtDivide = ext_divide;
+
+   prev_token->Extended = 1;
+   instruction_grow( instruction, header );
+
+   return src_register_ext_swz;
+}
+
+struct tgsi_src_register_ext_mod
+tgsi_default_src_register_ext_mod( void )
+{
+   struct tgsi_src_register_ext_mod src_register_ext_mod;
+
+   src_register_ext_mod.Type = TGSI_SRC_REGISTER_EXT_TYPE_MOD;
+   src_register_ext_mod.Complement = 0;
+   src_register_ext_mod.Bias = 0;
+   src_register_ext_mod.Scale2X = 0;
+   src_register_ext_mod.Absolute = 0;
+   src_register_ext_mod.Negate = 0;
+   src_register_ext_mod.Padding = 0;
+   src_register_ext_mod.Extended = 0;
+
+   return src_register_ext_mod;
+}
+
+unsigned
+tgsi_compare_src_register_ext_mod(
+   struct tgsi_src_register_ext_mod a,
+   struct tgsi_src_register_ext_mod b )
+{
+   a.Padding = b.Padding = 0;
+   a.Extended = b.Extended = 0;
+   return ((union token_u32 *) &a)->u32 != ((union token_u32 *) &b)->u32;
+}
+
+struct tgsi_src_register_ext_mod
+tgsi_build_src_register_ext_mod(
+   unsigned complement,
+   unsigned bias,
+   unsigned scale_2x,
+   unsigned absolute,
+   unsigned negate,
+   struct tgsi_token *prev_token,
+   struct tgsi_instruction *instruction,
+   struct tgsi_header *header )
+{
+   struct tgsi_src_register_ext_mod src_register_ext_mod;
+
+   assert( complement <= 1 );
+   assert( bias <= 1 );
+   assert( scale_2x <= 1 );
+   assert( absolute <= 1 );
+   assert( negate <= 1 );
+
+   src_register_ext_mod = tgsi_default_src_register_ext_mod();
+   src_register_ext_mod.Complement = complement;
+   src_register_ext_mod.Bias = bias;
+   src_register_ext_mod.Scale2X = scale_2x;
+   src_register_ext_mod.Absolute = absolute;
+   src_register_ext_mod.Negate = negate;
+
+   prev_token->Extended = 1;
+   instruction_grow( instruction, header );
+
+   return src_register_ext_mod;
+}
+
+struct tgsi_dimension
+tgsi_default_dimension( void )
+{
+   struct tgsi_dimension dimension;
+
+   dimension.Indirect = 0;
+   dimension.Dimension = 0;
+   dimension.Padding = 0;
+   dimension.Index = 0;
+   dimension.Extended = 0;
+
+   return dimension;
+}
+
+struct tgsi_dimension
+tgsi_build_dimension(
+   unsigned indirect,
+   unsigned index,
+   struct tgsi_instruction *instruction,
+   struct tgsi_header *header )
+{
+   struct tgsi_dimension dimension;
+
+   dimension = tgsi_default_dimension();
+   dimension.Indirect = indirect;
+   dimension.Index = index;
+
+   instruction_grow( instruction, header );
+
+   return dimension;
+}
+
+struct tgsi_dst_register
+tgsi_default_dst_register( void )
+{
+   struct tgsi_dst_register dst_register;
+
+   dst_register.File = TGSI_FILE_NULL;
+   dst_register.WriteMask = TGSI_WRITEMASK_XYZW;
+   dst_register.Indirect = 0;
+   dst_register.Dimension = 0;
+   dst_register.Index = 0;
+   dst_register.Padding = 0;
+   dst_register.Extended = 0;
+
+   return dst_register;
+}
+
+struct tgsi_dst_register
+tgsi_build_dst_register(
+   unsigned file,
+   unsigned mask,
+   int index,
+   struct tgsi_instruction *instruction,
+   struct tgsi_header *header )
+{
+   struct tgsi_dst_register dst_register;
+
+   assert( file <= TGSI_FILE_IMMEDIATE );
+   assert( mask <= TGSI_WRITEMASK_XYZW );
+   assert( index >= -32768 && index <= 32767 );
+
+   dst_register = tgsi_default_dst_register();
+   dst_register.File = file;
+   dst_register.WriteMask = mask;
+   dst_register.Index = index;
+
+   instruction_grow( instruction, header );
+
+   return dst_register;
+}
+
+struct tgsi_full_dst_register
+tgsi_default_full_dst_register( void )
+{
+   struct tgsi_full_dst_register full_dst_register;
+
+   full_dst_register.DstRegister = tgsi_default_dst_register();
+   full_dst_register.DstRegisterExtConcode =
+      tgsi_default_dst_register_ext_concode();
+   full_dst_register.DstRegisterExtModulate =
+      tgsi_default_dst_register_ext_modulate();
+
+   return full_dst_register;
+}
+
+struct tgsi_dst_register_ext_concode
+tgsi_default_dst_register_ext_concode( void )
+{
+   struct tgsi_dst_register_ext_concode dst_register_ext_concode;
+
+   dst_register_ext_concode.Type = TGSI_DST_REGISTER_EXT_TYPE_CONDCODE;
+   dst_register_ext_concode.CondMask = TGSI_CC_TR;
+   dst_register_ext_concode.CondSwizzleX = TGSI_SWIZZLE_X;
+   dst_register_ext_concode.CondSwizzleY = TGSI_SWIZZLE_Y;
+   dst_register_ext_concode.CondSwizzleZ = TGSI_SWIZZLE_Z;
+   dst_register_ext_concode.CondSwizzleW = TGSI_SWIZZLE_W;
+   dst_register_ext_concode.CondSrcIndex = 0;
+   dst_register_ext_concode.Padding = 0;
+   dst_register_ext_concode.Extended = 0;
+
+   return dst_register_ext_concode;
+}
+
+unsigned
+tgsi_compare_dst_register_ext_concode(
+   struct tgsi_dst_register_ext_concode a,
+   struct tgsi_dst_register_ext_concode b )
+{
+   a.Padding = b.Padding = 0;
+   a.Extended = b.Extended = 0;
+   return ((union token_u32 *) &a)->u32 != ((union token_u32 *) &b)->u32;
+}
+
+struct tgsi_dst_register_ext_concode
+tgsi_build_dst_register_ext_concode(
+   unsigned cc,
+   unsigned swizzle_x,
+   unsigned swizzle_y,
+   unsigned swizzle_z,
+   unsigned swizzle_w,
+   int index,
+   struct tgsi_token *prev_token,
+   struct tgsi_instruction *instruction,
+   struct tgsi_header *header )
+{
+   struct tgsi_dst_register_ext_concode dst_register_ext_concode;
+
+   assert( cc <= TGSI_CC_FL );
+   assert( swizzle_x <= TGSI_SWIZZLE_W );
+   assert( swizzle_y <= TGSI_SWIZZLE_W );
+   assert( swizzle_z <= TGSI_SWIZZLE_W );
+   assert( swizzle_w <= TGSI_SWIZZLE_W );
+   assert( index >= -32768 && index <= 32767 );
+
+   dst_register_ext_concode = tgsi_default_dst_register_ext_concode();
+   dst_register_ext_concode.CondMask = cc;
+   dst_register_ext_concode.CondSwizzleX = swizzle_x;
+   dst_register_ext_concode.CondSwizzleY = swizzle_y;
+   dst_register_ext_concode.CondSwizzleZ = swizzle_z;
+   dst_register_ext_concode.CondSwizzleW = swizzle_w;
+   dst_register_ext_concode.CondSrcIndex = index;
+
+   prev_token->Extended = 1;
+   instruction_grow( instruction, header );
+
+   return dst_register_ext_concode;
+}
+
+struct tgsi_dst_register_ext_modulate
+tgsi_default_dst_register_ext_modulate( void )
+{
+   struct tgsi_dst_register_ext_modulate dst_register_ext_modulate;
+
+   dst_register_ext_modulate.Type = TGSI_DST_REGISTER_EXT_TYPE_MODULATE;
+   dst_register_ext_modulate.Modulate = TGSI_MODULATE_1X;
+   dst_register_ext_modulate.Padding = 0;
+   dst_register_ext_modulate.Extended = 0;
+
+   return dst_register_ext_modulate;
+}
+
+unsigned
+tgsi_compare_dst_register_ext_modulate(
+   struct tgsi_dst_register_ext_modulate a,
+   struct tgsi_dst_register_ext_modulate b )
+{
+   a.Padding = b.Padding = 0;
+   a.Extended = b.Extended = 0;
+   return ((union token_u32 *) &a)->u32 != ((union token_u32 *) &b)->u32;
+}
+
+struct tgsi_dst_register_ext_modulate
+tgsi_build_dst_register_ext_modulate(
+   unsigned modulate,
+   struct tgsi_token *prev_token,
+   struct tgsi_instruction *instruction,
+   struct tgsi_header *header )
+{
+   struct tgsi_dst_register_ext_modulate dst_register_ext_modulate;
+
+   assert( modulate <= TGSI_MODULATE_EIGHTH );
+
+   dst_register_ext_modulate = tgsi_default_dst_register_ext_modulate();
+   dst_register_ext_modulate.Modulate = modulate;
+
+   prev_token->Extended = 1;
+   instruction_grow( instruction, header );
+
+   return dst_register_ext_modulate;
+}
+
diff --git a/src/gallium/aux/tgsi/util/tgsi_build.h b/src/gallium/aux/tgsi/util/tgsi_build.h
new file mode 100644
index 0000000000..116c78abf3
--- /dev/null
+++ b/src/gallium/aux/tgsi/util/tgsi_build.h
@@ -0,0 +1,320 @@
+#if !defined TGSI_BUILD_H
+#define TGSI_BUILD_H
+
+#if defined __cplusplus
+extern "C" {
+#endif // defined __cplusplus
+
+/*
+ * version
+ */
+
+struct tgsi_version
+tgsi_build_version( void );
+
+/*
+ * header
+ */
+
+struct tgsi_header
+tgsi_build_header( void );
+
+struct tgsi_processor
+tgsi_default_processor( void );
+
+struct tgsi_processor
+tgsi_build_processor(
+   unsigned processor,
+   struct tgsi_header *header );
+
+/*
+ * declaration
+ */
+
+struct tgsi_declaration
+tgsi_default_declaration( void );
+
+struct tgsi_declaration
+tgsi_build_declaration(
+   unsigned file,
+   unsigned declare,
+   unsigned usage_mask,
+   unsigned interpolate,
+   unsigned semantic,
+   struct tgsi_header *header );
+
+struct tgsi_full_declaration
+tgsi_default_full_declaration( void );
+
+unsigned
+tgsi_build_full_declaration(
+   const struct tgsi_full_declaration *full_decl,
+   struct tgsi_token *tokens,
+   struct tgsi_header *header,
+   unsigned maxsize );
+
+struct tgsi_declaration_range
+tgsi_build_declaration_range(
+   unsigned first,
+   unsigned last,
+   struct tgsi_declaration *declaration,
+   struct tgsi_header *header );
+
+struct tgsi_declaration_mask
+tgsi_build_declaration_mask(
+   unsigned mask,
+   struct tgsi_declaration *declaration,
+   struct tgsi_header *header );
+
+struct tgsi_declaration_interpolation
+tgsi_default_declaration_interpolation( void );
+
+struct tgsi_declaration_interpolation
+tgsi_build_declaration_interpolation(
+   unsigned interpolate,
+   struct tgsi_declaration *declaration,
+   struct tgsi_header *header );
+
+struct tgsi_declaration_semantic
+tgsi_default_declaration_semantic( void );
+
+struct tgsi_declaration_semantic
+tgsi_build_declaration_semantic(
+   unsigned semantic_name,
+   unsigned semantic_index,
+   struct tgsi_declaration *declaration,
+   struct tgsi_header *header );
+
+/*
+ * immediate
+ */
+
+struct tgsi_immediate
+tgsi_default_immediate( void );
+
+struct tgsi_immediate
+tgsi_build_immediate(
+   struct tgsi_header *header );
+
+struct tgsi_full_immediate
+tgsi_default_full_immediate( void );
+
+struct tgsi_immediate_float32
+tgsi_build_immediate_float32(
+   float value,
+   struct tgsi_immediate *immediate,
+   struct tgsi_header *header );
+
+unsigned
+tgsi_build_full_immediate(
+   const struct tgsi_full_immediate *full_imm,
+   struct tgsi_token *tokens,
+   struct tgsi_header *header,
+   unsigned maxsize );
+
+/*
+ * instruction
+ */
+
+struct tgsi_instruction
+tgsi_default_instruction( void );
+
+struct tgsi_instruction
+tgsi_build_instruction(
+   unsigned opcode,
+   unsigned saturate,
+   unsigned num_dst_regs,
+   unsigned num_src_regs,
+   struct tgsi_header *header );
+
+struct tgsi_full_instruction
+tgsi_default_full_instruction( void );
+
+unsigned
+tgsi_build_full_instruction(
+   const struct tgsi_full_instruction *full_inst,
+   struct tgsi_token *tokens,
+   struct tgsi_header *header,
+   unsigned maxsize );
+
+struct tgsi_instruction_ext_nv
+tgsi_default_instruction_ext_nv( void );
+
+unsigned
+tgsi_compare_instruction_ext_nv(
+   struct tgsi_instruction_ext_nv a,
+   struct tgsi_instruction_ext_nv b );
+
+struct tgsi_instruction_ext_nv
+tgsi_build_instruction_ext_nv(
+   unsigned precision,
+   unsigned cond_dst_index,
+   unsigned cond_flow_index,
+   unsigned cond_mask,
+   unsigned cond_swizzle_x,
+   unsigned cond_swizzle_y,
+   unsigned cond_swizzle_z,
+   unsigned cond_swizzle_w,
+   unsigned cond_dst_update,
+   unsigned cond_flow_update,
+   struct tgsi_token *prev_token,
+   struct tgsi_instruction *instruction,
+   struct tgsi_header *header );
+
+struct tgsi_instruction_ext_label
+tgsi_default_instruction_ext_label( void );
+
+unsigned
+tgsi_compare_instruction_ext_label(
+   struct tgsi_instruction_ext_label a,
+   struct tgsi_instruction_ext_label b );
+
+struct tgsi_instruction_ext_label
+tgsi_build_instruction_ext_label(
+   unsigned label,
+   struct tgsi_token *prev_token,
+   struct tgsi_instruction *instruction,
+   struct tgsi_header *header );
+
+struct tgsi_instruction_ext_texture
+tgsi_default_instruction_ext_texture( void );
+
+unsigned
+tgsi_compare_instruction_ext_texture(
+   struct tgsi_instruction_ext_texture a,
+   struct tgsi_instruction_ext_texture b );
+
+struct tgsi_instruction_ext_texture
+tgsi_build_instruction_ext_texture(
+   unsigned texture,
+   struct tgsi_token *prev_token,
+   struct tgsi_instruction *instruction,
+   struct tgsi_header *header );
+
+struct tgsi_src_register
+tgsi_default_src_register( void );
+
+struct tgsi_src_register
+tgsi_build_src_register(
+   unsigned file,
+   unsigned swizzle_x,
+   unsigned swizzle_y,
+   unsigned swizzle_z,
+   unsigned swizzle_w,
+   unsigned negate,
+   unsigned indirect,
+   unsigned dimension,
+   int index,
+   struct tgsi_instruction *instruction,
+   struct tgsi_header *header );
+
+struct tgsi_full_src_register
+tgsi_default_full_src_register( void );
+
+struct tgsi_src_register_ext_swz
+tgsi_default_src_register_ext_swz( void );
+
+unsigned
+tgsi_compare_src_register_ext_swz(
+   struct tgsi_src_register_ext_swz a,
+   struct tgsi_src_register_ext_swz b );
+
+struct tgsi_src_register_ext_swz
+tgsi_build_src_register_ext_swz(
+   unsigned ext_swizzle_x,
+   unsigned ext_swizzle_y,
+   unsigned ext_swizzle_z,
+   unsigned ext_swizzle_w,
+   unsigned negate_x,
+   unsigned negate_y,
+   unsigned negate_z,
+   unsigned negate_w,
+   unsigned ext_divide,
+   struct tgsi_token *prev_token,
+   struct tgsi_instruction *instruction,
+   struct tgsi_header *header );
+
+struct tgsi_src_register_ext_mod
+tgsi_default_src_register_ext_mod( void );
+
+unsigned
+tgsi_compare_src_register_ext_mod(
+   struct tgsi_src_register_ext_mod a,
+   struct tgsi_src_register_ext_mod b );
+
+struct tgsi_src_register_ext_mod
+tgsi_build_src_register_ext_mod(
+   unsigned complement,
+   unsigned bias,
+   unsigned scale_2x,
+   unsigned absolute,
+   unsigned negate,
+   struct tgsi_token *prev_token,
+   struct tgsi_instruction *instruction,
+   struct tgsi_header *header );
+
+struct tgsi_dimension
+tgsi_default_dimension( void );
+
+struct tgsi_dimension
+tgsi_build_dimension(
+   unsigned indirect,
+   unsigned index,
+   struct tgsi_instruction *instruction,
+   struct tgsi_header *header );
+
+struct tgsi_dst_register
+tgsi_default_dst_register( void );
+
+struct tgsi_dst_register
+tgsi_build_dst_register(
+   unsigned file,
+   unsigned mask,
+   int index,
+   struct tgsi_instruction *instruction,
+   struct tgsi_header *header );
+
+struct tgsi_full_dst_register
+tgsi_default_full_dst_register( void );
+
+struct tgsi_dst_register_ext_concode
+tgsi_default_dst_register_ext_concode( void );
+
+unsigned
+tgsi_compare_dst_register_ext_concode(
+   struct tgsi_dst_register_ext_concode a,
+   struct tgsi_dst_register_ext_concode b );
+
+struct tgsi_dst_register_ext_concode
+tgsi_build_dst_register_ext_concode(
+   unsigned cc,
+   unsigned swizzle_x,
+   unsigned swizzle_y,
+   unsigned swizzle_z,
+   unsigned swizzle_w,
+   int index,
+   struct tgsi_token *prev_token,
+   struct tgsi_instruction *instruction,
+   struct tgsi_header *header );
+
+struct tgsi_dst_register_ext_modulate
+tgsi_default_dst_register_ext_modulate( void );
+
+unsigned
+tgsi_compare_dst_register_ext_modulate(
+   struct tgsi_dst_register_ext_modulate a,
+   struct tgsi_dst_register_ext_modulate b );
+
+struct tgsi_dst_register_ext_modulate
+tgsi_build_dst_register_ext_modulate(
+   unsigned modulate,
+   struct tgsi_token *prev_token,
+   struct tgsi_instruction *instruction,
+   struct tgsi_header *header );
+
+#if defined __cplusplus
+} // extern "C"
+#endif // defined __cplusplus
+
+#endif // !defined TGSI_BUILD_H
+
diff --git a/src/gallium/aux/tgsi/util/tgsi_dump.c b/src/gallium/aux/tgsi/util/tgsi_dump.c
new file mode 100644
index 0000000000..b5c54847e0
--- /dev/null
+++ b/src/gallium/aux/tgsi/util/tgsi_dump.c
@@ -0,0 +1,1581 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+#include <stdio.h> 
+
+#include "pipe/p_debug.h"
+#include "pipe/p_util.h"
+#include "pipe/p_shader_tokens.h"
+#include "tgsi_dump.h"
+#include "tgsi_parse.h"
+#include "tgsi_build.h"
+
+struct gen_dump
+{
+   unsigned tabs;
+   void  (* write)(
+               struct gen_dump   *dump,
+               const void        *data,
+               unsigned          size );
+};
+
+struct text_dump
+{
+   struct gen_dump   base;
+   char              *text;
+   unsigned          length;
+   unsigned          capacity;
+};
+
+static void
+_text_dump_write(
+   struct gen_dump   *dump,
+   const void        *data,
+   unsigned          size )
+{
+   struct text_dump  *td = (struct text_dump *) dump;
+   unsigned          new_length = td->length + size;
+
+   if( new_length >= td->capacity ) {
+      unsigned new_capacity = td->capacity;
+
+      do {
+         if( new_capacity == 0 ) {
+            new_capacity = 256;
+         }
+         else {
+            new_capacity *= 2;
+         }
+      } while( new_length >= new_capacity );
+      td->text = (char *) REALLOC(
+         td->text,
+         td->capacity,
+         new_capacity );
+      td->capacity = new_capacity;
+   }
+   memcpy(
+      &td->text[td->length],
+      data,
+      size );
+   td->length = new_length;
+   td->text[td->length] = '\0';
+}
+
+struct file_dump
+{
+   struct gen_dump   base;
+   FILE              *file;
+};
+
+static void
+_file_dump_write(
+   struct gen_dump   *dump,
+   const void        *data,
+   unsigned          size )
+{
+   struct file_dump  *fd = (struct file_dump *) dump;
+
+#if 0
+   fwrite( data, 1, size, fd->file );
+#else
+   {
+      unsigned i;
+
+      for (i = 0; i < size; i++ ) {
+         fprintf( fd->file, "%c", ((const char *) data)[i] );
+      }
+   }
+#endif
+}
+
+static void
+gen_dump_str(
+   struct gen_dump   *dump,
+   const char        *str )
+{
+   unsigned i;
+   size_t   len = strlen( str );
+
+   for (i = 0; i < len; i++) {
+      dump->write( dump, &str[i], 1 );
+      if (str[i] == '\n') {
+         unsigned i;
+
+         for (i = 0; i < dump->tabs; i++) {
+            dump->write( dump, "    ", 4 );
+         }
+      }
+   }
+}
+
+static void
+gen_dump_chr(
+   struct gen_dump   *dump,
+   const char        chr )
+{
+   dump->write( dump, &chr, 1 );
+}
+
+static void
+gen_dump_uix(
+   struct gen_dump   *dump,
+   const unsigned    ui )
+{
+   char  str[36];
+
+   sprintf( str, "0x%x", ui );
+   gen_dump_str( dump, str );
+}
+
+static void
+gen_dump_uid(
+   struct gen_dump   *dump,
+   const unsigned    ui )
+{
+   char  str[16];
+
+   sprintf( str, "%u", ui );
+   gen_dump_str( dump, str );
+}
+
+static void
+gen_dump_sid(
+   struct gen_dump   *dump,
+   const int         si )
+{
+   char  str[16];
+
+   sprintf( str, "%d", si );
+   gen_dump_str( dump, str );
+}
+
+static void
+gen_dump_flt(
+   struct gen_dump   *dump,
+   const float       flt )
+{
+   char  str[48];
+
+   sprintf( str, "%10.4f", flt );
+   gen_dump_str( dump, str );
+}
+
+static void
+gen_dump_enum(
+   struct gen_dump   *dump,
+   const unsigned    e,
+   const char        **enums,
+   const unsigned    enums_count )
+{
+   if (e >= enums_count) {
+      gen_dump_uid( dump, e );
+   }
+   else {
+      gen_dump_str( dump, enums[e] );
+   }
+}
+
+static void
+gen_dump_tab(
+   struct gen_dump   *dump )
+{
+   ++dump->tabs;
+}
+
+static void
+gen_dump_untab(
+   struct gen_dump   *dump )
+{
+   assert( dump->tabs > 0 );
+
+   --dump->tabs;
+}
+
+#define TXT(S)          gen_dump_str( dump, S )
+#define CHR(C)          gen_dump_chr( dump, C )
+#define UIX(I)          gen_dump_uix( dump, I )
+#define UID(I)          gen_dump_uid( dump, I )
+#define SID(I)          gen_dump_sid( dump, I )
+#define FLT(F)          gen_dump_flt( dump, F )
+#define TAB()           gen_dump_tab( dump )
+#define UNT()           gen_dump_untab( dump )
+#define ENM(E,ENUMS)    gen_dump_enum( dump, E, ENUMS, sizeof( ENUMS ) / sizeof( *ENUMS ) )
+
+static const char *TGSI_PROCESSOR_TYPES[] =
+{
+   "PROCESSOR_FRAGMENT",
+   "PROCESSOR_VERTEX",
+   "PROCESSOR_GEOMETRY"
+};
+
+static const char *TGSI_PROCESSOR_TYPES_SHORT[] =
+{
+   "FRAG",
+   "VERT",
+   "GEOM"
+};
+
+static const char *TGSI_TOKEN_TYPES[] =
+{
+   "TOKEN_TYPE_DECLARATION",
+   "TOKEN_TYPE_IMMEDIATE",
+   "TOKEN_TYPE_INSTRUCTION"
+};
+
+static const char *TGSI_FILES[] =
+{
+   "FILE_NULL",
+   "FILE_CONSTANT",
+   "FILE_INPUT",
+   "FILE_OUTPUT",
+   "FILE_TEMPORARY",
+   "FILE_SAMPLER",
+   "FILE_ADDRESS",
+   "FILE_IMMEDIATE"
+};
+
+static const char *TGSI_FILES_SHORT[] =
+{
+   "NULL",
+   "CONST",
+   "IN",
+   "OUT",
+   "TEMP",
+   "SAMP",
+   "ADDR",
+   "IMM"
+};
+
+static const char *TGSI_DECLARES[] =
+{
+   "DECLARE_RANGE",
+   "DECLARE_MASK"
+};
+
+static const char *TGSI_INTERPOLATES[] =
+{
+   "INTERPOLATE_CONSTANT",
+   "INTERPOLATE_LINEAR",
+   "INTERPOLATE_PERSPECTIVE",
+   "INTERPOLATE_ATTRIB"
+};
+
+static const char *TGSI_INTERPOLATES_SHORT[] =
+{
+   "CONSTANT",
+   "LINEAR",
+   "PERSPECTIVE",
+   "ATTRIB"
+};
+
+static const char *TGSI_SEMANTICS[] =
+{
+   "SEMANTIC_POSITION",
+   "SEMANTIC_COLOR",
+   "SEMANTIC_BCOLOR",
+   "SEMANTIC_FOG",
+   "SEMANTIC_PSIZE",
+   "SEMANTIC_GENERIC,"
+};
+
+static const char *TGSI_SEMANTICS_SHORT[] =
+{
+   "POSITION",
+   "COLOR",
+   "BCOLOR",
+   "FOG",
+   "PSIZE",
+   "GENERIC",
+};
+
+static const char *TGSI_IMMS[] =
+{
+   "IMM_FLOAT32"
+};
+
+static const char *TGSI_IMMS_SHORT[] =
+{
+   "FLT32"
+};
+
+static const char *TGSI_OPCODES[] =
+{
+   "OPCODE_ARL",
+   "OPCODE_MOV",
+   "OPCODE_LIT",
+   "OPCODE_RCP",
+   "OPCODE_RSQ",
+   "OPCODE_EXP",
+   "OPCODE_LOG",
+   "OPCODE_MUL",
+   "OPCODE_ADD",
+   "OPCODE_DP3",
+   "OPCODE_DP4",
+   "OPCODE_DST",
+   "OPCODE_MIN",
+   "OPCODE_MAX",
+   "OPCODE_SLT",
+   "OPCODE_SGE",
+   "OPCODE_MAD",
+   "OPCODE_SUB",
+   "OPCODE_LERP",
+   "OPCODE_CND",
+   "OPCODE_CND0",
+   "OPCODE_DOT2ADD",
+   "OPCODE_INDEX",
+   "OPCODE_NEGATE",
+   "OPCODE_FRAC",
+   "OPCODE_CLAMP",
+   "OPCODE_FLOOR",
+   "OPCODE_ROUND",
+   "OPCODE_EXPBASE2",
+   "OPCODE_LOGBASE2",
+   "OPCODE_POWER",
+   "OPCODE_CROSSPRODUCT",
+   "OPCODE_MULTIPLYMATRIX",
+   "OPCODE_ABS",
+   "OPCODE_RCC",
+   "OPCODE_DPH",
+   "OPCODE_COS",
+   "OPCODE_DDX",
+   "OPCODE_DDY",
+   "OPCODE_KILP",
+   "OPCODE_PK2H",
+   "OPCODE_PK2US",
+   "OPCODE_PK4B",
+   "OPCODE_PK4UB",
+   "OPCODE_RFL",
+   "OPCODE_SEQ",
+   "OPCODE_SFL",
+   "OPCODE_SGT",
+   "OPCODE_SIN",
+   "OPCODE_SLE",
+   "OPCODE_SNE",
+   "OPCODE_STR",
+   "OPCODE_TEX",
+   "OPCODE_TXD",
+   "OPCODE_UP2H",
+   "OPCODE_UP2US",
+   "OPCODE_UP4B",
+   "OPCODE_UP4UB",
+   "OPCODE_X2D",
+   "OPCODE_ARA",
+   "OPCODE_ARR",
+   "OPCODE_BRA",
+   "OPCODE_CAL",
+   "OPCODE_RET",
+   "OPCODE_SSG",
+   "OPCODE_CMP",
+   "OPCODE_SCS",
+   "OPCODE_TXB",
+   "OPCODE_NRM",
+   "OPCODE_DIV",
+   "OPCODE_DP2",
+   "OPCODE_TXL",
+   "OPCODE_BRK",
+   "OPCODE_IF",
+   "OPCODE_LOOP",
+   "OPCODE_REP",
+   "OPCODE_ELSE",
+   "OPCODE_ENDIF",
+   "OPCODE_ENDLOOP",
+   "OPCODE_ENDREP",
+   "OPCODE_PUSHA",
+   "OPCODE_POPA",
+   "OPCODE_CEIL",
+   "OPCODE_I2F",
+   "OPCODE_NOT",
+   "OPCODE_TRUNC",
+   "OPCODE_SHL",
+   "OPCODE_SHR",
+   "OPCODE_AND",
+   "OPCODE_OR",
+   "OPCODE_MOD",
+   "OPCODE_XOR",
+   "OPCODE_SAD",
+   "OPCODE_TXF",
+   "OPCODE_TXQ",
+   "OPCODE_CONT",
+   "OPCODE_EMIT",
+   "OPCODE_ENDPRIM",
+   "OPCODE_BGNLOOP2",
+   "OPCODE_BGNSUB",
+   "OPCODE_ENDLOOP2",
+   "OPCODE_ENDSUB",
+   "OPCODE_NOISE1",
+   "OPCODE_NOISE2",
+   "OPCODE_NOISE3",
+   "OPCODE_NOISE4",
+   "OPCODE_NOP",
+   "OPCODE_TEXBEM",
+   "OPCODE_TEXBEML",
+   "OPCODE_TEXREG2AR",
+   "OPCODE_TEXM3X2PAD",
+   "OPCODE_TEXM3X2TEX",
+   "OPCODE_TEXM3X3PAD",
+   "OPCODE_TEXM3X3TEX",
+   "OPCODE_TEXM3X3SPEC",
+   "OPCODE_TEXM3X3VSPEC",
+   "OPCODE_TEXREG2GB",
+   "OPCODE_TEXREG2RGB",
+   "OPCODE_TEXDP3TEX",
+   "OPCODE_TEXDP3",
+   "OPCODE_TEXM3X3",
+   "OPCODE_TEXM3X2DEPTH",
+   "OPCODE_TEXDEPTH",
+   "OPCODE_BEM",
+   "OPCODE_M4X3",
+   "OPCODE_M3X4",
+   "OPCODE_M3X3",
+   "OPCODE_M3X2",
+   "OPCODE_NRM4",
+   "OPCODE_CALLNZ",
+   "OPCODE_IFC",
+   "OPCODE_BREAKC",
+   "OPCODE_KIL",
+   "OPCODE_END"
+};
+
+static const char *TGSI_OPCODES_SHORT[] =
+{
+   "ARL",
+   "MOV",
+   "LIT",
+   "RCP",
+   "RSQ",
+   "EXP",
+   "LOG",
+   "MUL",
+   "ADD",
+   "DP3",
+   "DP4",
+   "DST",
+   "MIN",
+   "MAX",
+   "SLT",
+   "SGE",
+   "MAD",
+   "SUB",
+   "LERP",
+   "CND",
+   "CND0",
+   "DOT2ADD",
+   "INDEX",
+   "NEGATE",
+   "FRAC",
+   "CLAMP",
+   "FLOOR",
+   "ROUND",
+   "EXPBASE2",
+   "LOGBASE2",
+   "POWER",
+   "CROSSPRODUCT",
+   "MULTIPLYMATRIX",
+   "ABS",
+   "RCC",
+   "DPH",
+   "COS",
+   "DDX",
+   "DDY",
+   "KILP",
+   "PK2H",
+   "PK2US",
+   "PK4B",
+   "PK4UB",
+   "RFL",
+   "SEQ",
+   "SFL",
+   "SGT",
+   "SIN",
+   "SLE",
+   "SNE",
+   "STR",
+   "TEX",
+   "TXD",
+   "UP2H",
+   "UP2US",
+   "UP4B",
+   "UP4UB",
+   "X2D",
+   "ARA",
+   "ARR",
+   "BRA",
+   "CAL",
+   "RET",
+   "SSG",
+   "CMP",
+   "SCS",
+   "TXB",
+   "NRM",
+   "DIV",
+   "DP2",
+   "TXL",
+   "BRK",
+   "IF",
+   "LOOP",
+   "REP",
+   "ELSE",
+   "ENDIF",
+   "ENDLOOP",
+   "ENDREP",
+   "PUSHA",
+   "POPA",
+   "CEIL",
+   "I2F",
+   "NOT",
+   "TRUNC",
+   "SHL",
+   "SHR",
+   "AND",
+   "OR",
+   "MOD",
+   "XOR",
+   "SAD",
+   "TXF",
+   "TXQ",
+   "CONT",
+   "EMIT",
+   "ENDPRIM",
+   "BGNLOOP2",
+   "BGNSUB",
+   "ENDLOOP2",
+   "ENDSUB",
+   "NOISE1",
+   "NOISE2",
+   "NOISE3",
+   "NOISE4",
+   "NOP",
+   "TEXBEM",
+   "TEXBEML",
+   "TEXREG2AR",
+   "TEXM3X2PAD",
+   "TEXM3X2TEX",
+   "TEXM3X3PAD",
+   "TEXM3X3TEX",
+   "TEXM3X3SPEC",
+   "TEXM3X3VSPEC",
+   "TEXREG2GB",
+   "TEXREG2RGB",
+   "TEXDP3TEX",
+   "TEXDP3",
+   "TEXM3X3",
+   "TEXM3X2DEPTH",
+   "TEXDEPTH",
+   "BEM",
+   "M4X3",
+   "M3X4",
+   "M3X3",
+   "M3X2",
+   "NRM4",
+   "CALLNZ",
+   "IFC",
+   "BREAKC",
+   "KIL",
+   "END"
+};
+
+static const char *TGSI_SATS[] =
+{
+   "SAT_NONE",
+   "SAT_ZERO_ONE",
+   "SAT_MINUS_PLUS_ONE"
+};
+
+static const char *TGSI_INSTRUCTION_EXTS[] =
+{
+   "INSTRUCTION_EXT_TYPE_NV",
+   "INSTRUCTION_EXT_TYPE_LABEL",
+   "INSTRUCTION_EXT_TYPE_TEXTURE"
+};
+
+static const char *TGSI_PRECISIONS[] =
+{
+   "PRECISION_DEFAULT",
+   "TGSI_PRECISION_FLOAT32",
+   "TGSI_PRECISION_FLOAT16",
+   "TGSI_PRECISION_FIXED12"
+};
+
+static const char *TGSI_CCS[] =
+{
+   "CC_GT",
+   "CC_EQ",
+   "CC_LT",
+   "CC_UN",
+   "CC_GE",
+   "CC_LE",
+   "CC_NE",
+   "CC_TR",
+   "CC_FL"
+};
+
+static const char *TGSI_SWIZZLES[] =
+{
+   "SWIZZLE_X",
+   "SWIZZLE_Y",
+   "SWIZZLE_Z",
+   "SWIZZLE_W"
+};
+
+static const char *TGSI_SWIZZLES_SHORT[] =
+{
+   "x",
+   "y",
+   "z",
+   "w"
+};
+
+static const char *TGSI_TEXTURES[] =
+{
+   "TEXTURE_UNKNOWN",
+   "TEXTURE_1D",
+   "TEXTURE_2D",
+   "TEXTURE_3D",
+   "TEXTURE_CUBE",
+   "TEXTURE_RECT",
+   "TEXTURE_SHADOW1D",
+   "TEXTURE_SHADOW2D",
+   "TEXTURE_SHADOWRECT"
+};
+
+static const char *TGSI_SRC_REGISTER_EXTS[] =
+{
+   "SRC_REGISTER_EXT_TYPE_SWZ",
+   "SRC_REGISTER_EXT_TYPE_MOD"
+};
+
+static const char *TGSI_EXTSWIZZLES[] =
+{
+   "EXTSWIZZLE_X",
+   "EXTSWIZZLE_Y",
+   "EXTSWIZZLE_Z",
+   "EXTSWIZZLE_W",
+   "EXTSWIZZLE_ZERO",
+   "EXTSWIZZLE_ONE"
+};
+
+static const char *TGSI_EXTSWIZZLES_SHORT[] =
+{
+   "x",
+   "y",
+   "z",
+   "w",
+   "0",
+   "1"
+};
+
+static const char *TGSI_WRITEMASKS[] =
+{
+   "0",
+   "WRITEMASK_X",
+   "WRITEMASK_Y",
+   "WRITEMASK_XY",
+   "WRITEMASK_Z",
+   "WRITEMASK_XZ",
+   "WRITEMASK_YZ",
+   "WRITEMASK_XYZ",
+   "WRITEMASK_W",
+   "WRITEMASK_XW",
+   "WRITEMASK_YW",
+   "WRITEMASK_XYW",
+   "WRITEMASK_ZW",
+   "WRITEMASK_XZW",
+   "WRITEMASK_YZW",
+   "WRITEMASK_XYZW"
+};
+
+static const char *TGSI_DST_REGISTER_EXTS[] =
+{
+   "DST_REGISTER_EXT_TYPE_CONDCODE",
+   "DST_REGISTER_EXT_TYPE_MODULATE"
+};
+
+static const char *TGSI_MODULATES[] =
+{
+   "MODULATE_1X",
+   "MODULATE_2X",
+   "MODULATE_4X",
+   "MODULATE_8X",
+   "MODULATE_HALF",
+   "MODULATE_QUARTER",
+   "MODULATE_EIGHTH"
+};
+
+static void
+dump_declaration_short(
+   struct gen_dump               *dump,
+   struct tgsi_full_declaration  *decl )
+{
+   TXT( "\nDCL " );
+   ENM( decl->Declaration.File, TGSI_FILES_SHORT );
+
+   switch( decl->Declaration.Declare ) {
+   case TGSI_DECLARE_RANGE:
+      CHR( '[' );
+      UID( decl->u.DeclarationRange.First );
+      if( decl->u.DeclarationRange.First != decl->u.DeclarationRange.Last ) {
+         TXT( ".." );
+         UID( decl->u.DeclarationRange.Last );
+      }
+      CHR( ']' );
+      break;
+   default:
+      assert( 0 );
+   }
+
+   if( decl->Declaration.UsageMask != TGSI_WRITEMASK_XYZW ) {
+      CHR( '.' );
+      if( decl->Declaration.UsageMask & TGSI_WRITEMASK_X ) {
+         CHR( 'x' );
+      }
+      if( decl->Declaration.UsageMask & TGSI_WRITEMASK_Y ) {
+         CHR( 'y' );
+      }
+      if( decl->Declaration.UsageMask & TGSI_WRITEMASK_Z ) {
+         CHR( 'z' );
+      }
+      if( decl->Declaration.UsageMask & TGSI_WRITEMASK_W ) {
+         CHR( 'w' );
+      }
+   }
+
+   if( decl->Declaration.Interpolate ) {
+      TXT( ", " );
+      ENM( decl->Interpolation.Interpolate, TGSI_INTERPOLATES_SHORT );
+   }
+
+   if( decl->Declaration.Semantic ) {
+      TXT( ", " );
+      ENM( decl->Semantic.SemanticName, TGSI_SEMANTICS_SHORT );
+      CHR( '[' );
+      UID( decl->Semantic.SemanticIndex );
+      CHR( ']' );
+   }
+}
+
+static void
+dump_declaration_verbose(
+   struct gen_dump               *dump,
+   struct tgsi_full_declaration  *decl,
+   unsigned                      ignored,
+   unsigned                      deflt,
+   struct tgsi_full_declaration  *fd )
+{
+   TXT( "\nFile       : " );
+   ENM( decl->Declaration.File, TGSI_FILES );
+   TXT( "\nDeclare    : " );
+   ENM( decl->Declaration.Declare, TGSI_DECLARES );
+   if( deflt || fd->Declaration.UsageMask != decl->Declaration.UsageMask ) {
+      TXT( "\nUsageMask  : " );
+      if( decl->Declaration.UsageMask & TGSI_WRITEMASK_X ) {
+         CHR( 'X' );
+      }
+      if( decl->Declaration.UsageMask & TGSI_WRITEMASK_Y ) {
+         CHR( 'Y' );
+      }
+      if( decl->Declaration.UsageMask & TGSI_WRITEMASK_Z ) {
+         CHR( 'Z' );
+      }
+      if( decl->Declaration.UsageMask & TGSI_WRITEMASK_W ) {
+         CHR( 'W' );
+      }
+   }
+   if( deflt || fd->Declaration.Interpolate != decl->Declaration.Interpolate ) {
+      TXT( "\nInterpolate: " );
+      UID( decl->Declaration.Interpolate );
+   }
+   if( deflt || fd->Declaration.Semantic != decl->Declaration.Semantic ) {
+      TXT( "\nSemantic   : " );
+      UID( decl->Declaration.Semantic );
+   }
+   if( ignored ) {
+      TXT( "\nPadding    : " );
+      UIX( decl->Declaration.Padding );
+   }
+
+   CHR( '\n' );
+   switch( decl->Declaration.Declare ) {
+   case TGSI_DECLARE_RANGE:
+      TXT( "\nFirst: " );
+      UID( decl->u.DeclarationRange.First );
+      TXT( "\nLast : " );
+      UID( decl->u.DeclarationRange.Last );
+      break;
+
+   case TGSI_DECLARE_MASK:
+      TXT( "\nMask: " );
+      UIX( decl->u.DeclarationMask.Mask );
+      break;
+
+   default:
+      assert( 0 );
+   }
+
+   if( decl->Declaration.Interpolate ) {
+      CHR( '\n' );
+      TXT( "\nInterpolate: " );
+      ENM( decl->Interpolation.Interpolate, TGSI_INTERPOLATES );
+      if( ignored ) {
+         TXT( "\nPadding    : " );
+         UIX( decl->Interpolation.Padding );
+      }
+   }
+
+   if( decl->Declaration.Semantic ) {
+      CHR( '\n' );
+      TXT( "\nSemanticName : " );
+      ENM( decl->Semantic.SemanticName, TGSI_SEMANTICS );
+      TXT( "\nSemanticIndex: " );
+      UID( decl->Semantic.SemanticIndex );
+      if( ignored ) {
+         TXT( "\nPadding      : " );
+         UIX( decl->Semantic.Padding );
+      }
+   }
+}
+
+static void
+dump_immediate_short(
+   struct gen_dump            *dump,
+   struct tgsi_full_immediate *imm )
+{
+   unsigned i;
+
+   TXT( "\nIMM " );
+   ENM( imm->Immediate.DataType, TGSI_IMMS_SHORT );
+
+   TXT( " { " );
+   for( i = 0; i < imm->Immediate.Size - 1; i++ ) {
+      switch( imm->Immediate.DataType ) {
+      case TGSI_IMM_FLOAT32:
+         FLT( imm->u.ImmediateFloat32[i].Float );
+         break;
+
+      default:
+         assert( 0 );
+      }
+
+      if( i < imm->Immediate.Size - 2 ) {
+         TXT( ", " );
+      }
+   }
+   TXT( " }" );
+}
+
+static void
+dump_immediate_verbose(
+   struct gen_dump            *dump,
+   struct tgsi_full_immediate *imm,
+   unsigned                   ignored )
+{
+   unsigned i;
+
+   TXT( "\nDataType   : " );
+   ENM( imm->Immediate.DataType, TGSI_IMMS );
+   if( ignored ) {
+      TXT( "\nPadding    : " );
+      UIX( imm->Immediate.Padding );
+   }
+
+   for( i = 0; i < imm->Immediate.Size - 1; i++ ) {
+      CHR( '\n' );
+      switch( imm->Immediate.DataType ) {
+      case TGSI_IMM_FLOAT32:
+         TXT( "\nFloat: " );
+         FLT( imm->u.ImmediateFloat32[i].Float );
+         break;
+
+      default:
+         assert( 0 );
+      }
+   }
+}
+
+static void
+dump_instruction_short(
+   struct gen_dump               *dump,
+   struct tgsi_full_instruction  *inst,
+   unsigned                      instno )
+{
+   unsigned i;
+   boolean  first_reg = TRUE;
+
+   CHR( '\n' );
+   UID( instno );
+   CHR( ':' );
+   ENM( inst->Instruction.Opcode, TGSI_OPCODES_SHORT );
+
+   switch( inst->Instruction.Saturate ) {
+   case TGSI_SAT_NONE:
+      break;
+   case TGSI_SAT_ZERO_ONE:
+      TXT( "_SAT" );
+      break;
+   case TGSI_SAT_MINUS_PLUS_ONE:
+      TXT( "_SAT[-1,1]" );
+      break;
+   default:
+      assert( 0 );
+   }
+
+   for( i = 0; i < inst->Instruction.NumDstRegs; i++ ) {
+      struct tgsi_full_dst_register *dst = &inst->FullDstRegisters[i];
+
+      if( !first_reg ) {
+         CHR( ',' );
+      }
+      CHR( ' ' );
+
+      ENM( dst->DstRegister.File, TGSI_FILES_SHORT );
+
+      CHR( '[' );
+      SID( dst->DstRegister.Index );
+      CHR( ']' );
+
+      if( dst->DstRegister.WriteMask != TGSI_WRITEMASK_XYZW ) {
+         CHR( '.' );
+         if( dst->DstRegister.WriteMask & TGSI_WRITEMASK_X ) {
+            CHR( 'x' );
+         }
+         if( dst->DstRegister.WriteMask & TGSI_WRITEMASK_Y ) {
+            CHR( 'y' );
+         }
+         if( dst->DstRegister.WriteMask & TGSI_WRITEMASK_Z ) {
+            CHR( 'z' );
+         }
+         if( dst->DstRegister.WriteMask & TGSI_WRITEMASK_W ) {
+            CHR( 'w' );
+         }
+      }
+
+      first_reg = FALSE;
+   }
+
+   for( i = 0; i < inst->Instruction.NumSrcRegs; i++ ) {
+      struct tgsi_full_src_register *src = &inst->FullSrcRegisters[i];
+
+      if( !first_reg ) {
+         CHR( ',' );
+      }
+      CHR( ' ' );
+
+      if( src->SrcRegisterExtMod.Complement ) {
+         TXT( "(1 - " );
+      }
+      if( src->SrcRegisterExtMod.Negate  ) {
+         CHR( '-' );
+      }
+      if( src->SrcRegisterExtMod.Absolute ) {
+         CHR( '|' );
+      }
+      if( src->SrcRegister.Negate ) {
+         CHR( '-' );
+      }
+
+      ENM( src->SrcRegister.File, TGSI_FILES_SHORT );
+
+      CHR( '[' );
+      SID( src->SrcRegister.Index );
+      CHR( ']' );
+
+      if (src->SrcRegister.Extended) {
+         if (src->SrcRegisterExtSwz.ExtSwizzleX != TGSI_EXTSWIZZLE_X ||
+             src->SrcRegisterExtSwz.ExtSwizzleY != TGSI_EXTSWIZZLE_Y ||
+             src->SrcRegisterExtSwz.ExtSwizzleZ != TGSI_EXTSWIZZLE_Z ||
+             src->SrcRegisterExtSwz.ExtSwizzleW != TGSI_EXTSWIZZLE_W) {
+            CHR( '.' );
+            ENM( src->SrcRegisterExtSwz.ExtSwizzleX, TGSI_EXTSWIZZLES_SHORT );
+            ENM( src->SrcRegisterExtSwz.ExtSwizzleY, TGSI_EXTSWIZZLES_SHORT );
+            ENM( src->SrcRegisterExtSwz.ExtSwizzleZ, TGSI_EXTSWIZZLES_SHORT );
+            ENM( src->SrcRegisterExtSwz.ExtSwizzleW, TGSI_EXTSWIZZLES_SHORT );
+         }
+      }
+      else if( src->SrcRegister.SwizzleX != TGSI_SWIZZLE_X ||
+               src->SrcRegister.SwizzleY != TGSI_SWIZZLE_Y ||
+               src->SrcRegister.SwizzleZ != TGSI_SWIZZLE_Z ||
+               src->SrcRegister.SwizzleW != TGSI_SWIZZLE_W ) {
+         CHR( '.' );
+         ENM( src->SrcRegister.SwizzleX, TGSI_SWIZZLES_SHORT );
+         ENM( src->SrcRegister.SwizzleY, TGSI_SWIZZLES_SHORT );
+         ENM( src->SrcRegister.SwizzleZ, TGSI_SWIZZLES_SHORT );
+         ENM( src->SrcRegister.SwizzleW, TGSI_SWIZZLES_SHORT );
+      }
+
+      if( src->SrcRegisterExtMod.Absolute ) {
+         CHR( '|' );
+      }
+      if( src->SrcRegisterExtMod.Complement ) {
+         CHR( ')' );
+      }
+
+      first_reg = FALSE;
+   }
+
+   switch( inst->Instruction.Opcode ) {
+   case TGSI_OPCODE_IF:
+   case TGSI_OPCODE_ELSE:
+   case TGSI_OPCODE_BGNLOOP2:
+   case TGSI_OPCODE_ENDLOOP2:
+   case TGSI_OPCODE_CAL:
+      TXT( " :" );
+      UID( inst->InstructionExtLabel.Label );
+      break;
+   }
+}
+
+static void
+dump_instruction_verbose(
+   struct gen_dump               *dump,
+   struct tgsi_full_instruction  *inst,
+   unsigned                      ignored,
+   unsigned                      deflt,
+   struct tgsi_full_instruction  *fi )
+{
+   unsigned i;
+
+   TXT( "\nOpcode     : " );
+   ENM( inst->Instruction.Opcode, TGSI_OPCODES );
+   if( deflt || fi->Instruction.Saturate != inst->Instruction.Saturate ) {
+      TXT( "\nSaturate   : " );
+      ENM( inst->Instruction.Saturate, TGSI_SATS );
+   }
+   if( deflt || fi->Instruction.NumDstRegs != inst->Instruction.NumDstRegs ) {
+      TXT( "\nNumDstRegs : " );
+      UID( inst->Instruction.NumDstRegs );
+   }
+   if( deflt || fi->Instruction.NumSrcRegs != inst->Instruction.NumSrcRegs ) {
+      TXT( "\nNumSrcRegs : " );
+      UID( inst->Instruction.NumSrcRegs );
+   }
+   if( ignored ) {
+      TXT( "\nPadding    : " );
+      UIX( inst->Instruction.Padding );
+   }
+
+   if( deflt || tgsi_compare_instruction_ext_nv( inst->InstructionExtNv, fi->InstructionExtNv ) ) {
+      CHR( '\n' );
+      TXT( "\nType          : " );
+      ENM( inst->InstructionExtNv.Type, TGSI_INSTRUCTION_EXTS );
+      if( deflt || fi->InstructionExtNv.Precision != inst->InstructionExtNv.Precision ) {
+         TXT( "\nPrecision     : " );
+         ENM( inst->InstructionExtNv.Precision, TGSI_PRECISIONS );
+      }
+      if( deflt || fi->InstructionExtNv.CondDstIndex != inst->InstructionExtNv.CondDstIndex ) {
+         TXT( "\nCondDstIndex  : " );
+         UID( inst->InstructionExtNv.CondDstIndex );
+      }
+      if( deflt || fi->InstructionExtNv.CondFlowIndex != inst->InstructionExtNv.CondFlowIndex ) {
+         TXT( "\nCondFlowIndex : " );
+         UID( inst->InstructionExtNv.CondFlowIndex );
+      }
+      if( deflt || fi->InstructionExtNv.CondMask != inst->InstructionExtNv.CondMask ) {
+         TXT( "\nCondMask      : " );
+         ENM( inst->InstructionExtNv.CondMask, TGSI_CCS );
+      }
+      if( deflt || fi->InstructionExtNv.CondSwizzleX != inst->InstructionExtNv.CondSwizzleX ) {
+         TXT( "\nCondSwizzleX  : " );
+         ENM( inst->InstructionExtNv.CondSwizzleX, TGSI_SWIZZLES );
+      }
+      if( deflt || fi->InstructionExtNv.CondSwizzleY != inst->InstructionExtNv.CondSwizzleY ) {
+         TXT( "\nCondSwizzleY  : " );
+         ENM( inst->InstructionExtNv.CondSwizzleY, TGSI_SWIZZLES );
+      }
+      if( deflt || fi->InstructionExtNv.CondSwizzleZ != inst->InstructionExtNv.CondSwizzleZ ) {
+         TXT( "\nCondSwizzleZ  : " );
+         ENM( inst->InstructionExtNv.CondSwizzleZ, TGSI_SWIZZLES );
+      }
+      if( deflt || fi->InstructionExtNv.CondSwizzleW != inst->InstructionExtNv.CondSwizzleW ) {
+         TXT( "\nCondSwizzleW  : " );
+         ENM( inst->InstructionExtNv.CondSwizzleW, TGSI_SWIZZLES );
+      }
+      if( deflt || fi->InstructionExtNv.CondDstUpdate != inst->InstructionExtNv.CondDstUpdate ) {
+         TXT( "\nCondDstUpdate : " );
+         UID( inst->InstructionExtNv.CondDstUpdate );
+      }
+      if( deflt || fi->InstructionExtNv.CondFlowEnable != inst->InstructionExtNv.CondFlowEnable ) {
+         TXT( "\nCondFlowEnable: " );
+         UID( inst->InstructionExtNv.CondFlowEnable );
+      }
+      if( ignored ) {
+         TXT( "\nPadding       : " );
+         UIX( inst->InstructionExtNv.Padding );
+         if( deflt || fi->InstructionExtNv.Extended != inst->InstructionExtNv.Extended ) {
+            TXT( "\nExtended      : " );
+            UID( inst->InstructionExtNv.Extended );
+         }
+      }
+   }
+
+   if( deflt || tgsi_compare_instruction_ext_label( inst->InstructionExtLabel, fi->InstructionExtLabel ) ) {
+      CHR( '\n' );
+      TXT( "\nType    : " );
+      ENM( inst->InstructionExtLabel.Type, TGSI_INSTRUCTION_EXTS );
+      if( deflt || fi->InstructionExtLabel.Label != inst->InstructionExtLabel.Label ) {
+         TXT( "\nLabel   : " );
+         UID( inst->InstructionExtLabel.Label );
+      }
+      if( ignored ) {
+         TXT( "\nPadding : " );
+         UIX( inst->InstructionExtLabel.Padding );
+         if( deflt || fi->InstructionExtLabel.Extended != inst->InstructionExtLabel.Extended ) {
+            TXT( "\nExtended: " );
+            UID( inst->InstructionExtLabel.Extended );
+         }
+      }
+   }
+
+   if( deflt || tgsi_compare_instruction_ext_texture( inst->InstructionExtTexture, fi->InstructionExtTexture ) ) {
+      CHR( '\n' );
+      TXT( "\nType    : " );
+      ENM( inst->InstructionExtTexture.Type, TGSI_INSTRUCTION_EXTS );
+      if( deflt || fi->InstructionExtTexture.Texture != inst->InstructionExtTexture.Texture ) {
+         TXT( "\nTexture : " );
+         ENM( inst->InstructionExtTexture.Texture, TGSI_TEXTURES );
+      }
+      if( ignored ) {
+         TXT( "\nPadding : " );
+         UIX( inst->InstructionExtTexture.Padding );
+         if( deflt || fi->InstructionExtTexture.Extended != inst->InstructionExtTexture.Extended ) {
+            TXT( "\nExtended: " );
+            UID( inst->InstructionExtTexture.Extended );
+         }
+      }
+   }
+
+   for( i = 0; i < inst->Instruction.NumDstRegs; i++ ) {
+      struct tgsi_full_dst_register *dst = &inst->FullDstRegisters[i];
+      struct tgsi_full_dst_register *fd = &fi->FullDstRegisters[i];
+
+      CHR( '\n' );
+      TXT( "\nFile     : " );
+      ENM( dst->DstRegister.File, TGSI_FILES );
+      if( deflt || fd->DstRegister.WriteMask != dst->DstRegister.WriteMask ) {
+         TXT( "\nWriteMask: " );
+         ENM( dst->DstRegister.WriteMask, TGSI_WRITEMASKS );
+      }
+      if( ignored ) {
+         if( deflt || fd->DstRegister.Indirect != dst->DstRegister.Indirect ) {
+            TXT( "\nIndirect : " );
+            UID( dst->DstRegister.Indirect );
+         }
+         if( deflt || fd->DstRegister.Dimension != dst->DstRegister.Dimension ) {
+            TXT( "\nDimension: " );
+            UID( dst->DstRegister.Dimension );
+         }
+      }
+      if( deflt || fd->DstRegister.Index != dst->DstRegister.Index ) {
+         TXT( "\nIndex    : " );
+         SID( dst->DstRegister.Index );
+      }
+      if( ignored ) {
+         TXT( "\nPadding  : " );
+         UIX( dst->DstRegister.Padding );
+         if( deflt || fd->DstRegister.Extended != dst->DstRegister.Extended ) {
+            TXT( "\nExtended : " );
+            UID( dst->DstRegister.Extended );
+         }
+      }
+
+      if( deflt || tgsi_compare_dst_register_ext_concode( dst->DstRegisterExtConcode, fd->DstRegisterExtConcode ) ) {
+         CHR( '\n' );
+         TXT( "\nType        : " );
+         ENM( dst->DstRegisterExtConcode.Type, TGSI_DST_REGISTER_EXTS );
+         if( deflt || fd->DstRegisterExtConcode.CondMask != dst->DstRegisterExtConcode.CondMask ) {
+            TXT( "\nCondMask    : " );
+            ENM( dst->DstRegisterExtConcode.CondMask, TGSI_CCS );
+         }
+         if( deflt || fd->DstRegisterExtConcode.CondSwizzleX != dst->DstRegisterExtConcode.CondSwizzleX ) {
+            TXT( "\nCondSwizzleX: " );
+            ENM( dst->DstRegisterExtConcode.CondSwizzleX, TGSI_SWIZZLES );
+         }
+         if( deflt || fd->DstRegisterExtConcode.CondSwizzleY != dst->DstRegisterExtConcode.CondSwizzleY ) {
+            TXT( "\nCondSwizzleY: " );
+            ENM( dst->DstRegisterExtConcode.CondSwizzleY, TGSI_SWIZZLES );
+         }
+         if( deflt || fd->DstRegisterExtConcode.CondSwizzleZ != dst->DstRegisterExtConcode.CondSwizzleZ ) {
+            TXT( "\nCondSwizzleZ: " );
+            ENM( dst->DstRegisterExtConcode.CondSwizzleZ, TGSI_SWIZZLES );
+         }
+         if( deflt || fd->DstRegisterExtConcode.CondSwizzleW != dst->DstRegisterExtConcode.CondSwizzleW ) {
+            TXT( "\nCondSwizzleW: " );
+            ENM( dst->DstRegisterExtConcode.CondSwizzleW, TGSI_SWIZZLES );
+         }
+         if( deflt || fd->DstRegisterExtConcode.CondSrcIndex != dst->DstRegisterExtConcode.CondSrcIndex ) {
+            TXT( "\nCondSrcIndex: " );
+            UID( dst->DstRegisterExtConcode.CondSrcIndex );
+         }
+         if( ignored ) {
+            TXT( "\nPadding     : " );
+            UIX( dst->DstRegisterExtConcode.Padding );
+            if( deflt || fd->DstRegisterExtConcode.Extended != dst->DstRegisterExtConcode.Extended ) {
+               TXT( "\nExtended    : " );
+               UID( dst->DstRegisterExtConcode.Extended );
+            }
+         }
+      }
+
+      if( deflt || tgsi_compare_dst_register_ext_modulate( dst->DstRegisterExtModulate, fd->DstRegisterExtModulate ) ) {
+         CHR( '\n' );
+         TXT( "\nType    : " );
+         ENM( dst->DstRegisterExtModulate.Type, TGSI_DST_REGISTER_EXTS );
+         if( deflt || fd->DstRegisterExtModulate.Modulate != dst->DstRegisterExtModulate.Modulate ) {
+            TXT( "\nModulate: " );
+            ENM( dst->DstRegisterExtModulate.Modulate, TGSI_MODULATES );
+         }
+         if( ignored ) {
+            TXT( "\nPadding : " );
+            UIX( dst->DstRegisterExtModulate.Padding );
+            if( deflt || fd->DstRegisterExtModulate.Extended != dst->DstRegisterExtModulate.Extended ) {
+               TXT( "\nExtended: " );
+               UID( dst->DstRegisterExtModulate.Extended );
+            }
+         }
+      }
+   }
+
+   for( i = 0; i < inst->Instruction.NumSrcRegs; i++ ) {
+      struct tgsi_full_src_register *src = &inst->FullSrcRegisters[i];
+      struct tgsi_full_src_register *fs = &fi->FullSrcRegisters[i];
+
+      CHR( '\n' );
+      TXT( "\nFile     : ");
+      ENM( src->SrcRegister.File, TGSI_FILES );
+      if( deflt || fs->SrcRegister.SwizzleX != src->SrcRegister.SwizzleX ) {
+         TXT( "\nSwizzleX : " );
+         ENM( src->SrcRegister.SwizzleX, TGSI_SWIZZLES );
+      }
+      if( deflt || fs->SrcRegister.SwizzleY != src->SrcRegister.SwizzleY ) {
+         TXT( "\nSwizzleY : " );
+         ENM( src->SrcRegister.SwizzleY, TGSI_SWIZZLES );
+      }
+      if( deflt || fs->SrcRegister.SwizzleZ != src->SrcRegister.SwizzleZ ) {
+         TXT( "\nSwizzleZ : " );
+         ENM( src->SrcRegister.SwizzleZ, TGSI_SWIZZLES );
+      }
+      if( deflt || fs->SrcRegister.SwizzleW != src->SrcRegister.SwizzleW ) {
+         TXT( "\nSwizzleW : " );
+         ENM( src->SrcRegister.SwizzleW, TGSI_SWIZZLES );
+      }
+      if( deflt || fs->SrcRegister.Negate != src->SrcRegister.Negate ) {
+         TXT( "\nNegate   : " );
+         UID( src->SrcRegister.Negate );
+      }
+      if( ignored ) {
+         if( deflt || fs->SrcRegister.Indirect != src->SrcRegister.Indirect ) {
+            TXT( "\nIndirect : " );
+            UID( src->SrcRegister.Indirect );
+         }
+         if( deflt || fs->SrcRegister.Dimension != src->SrcRegister.Dimension ) {
+            TXT( "\nDimension: " );
+            UID( src->SrcRegister.Dimension );
+         }
+      }
+      if( deflt || fs->SrcRegister.Index != src->SrcRegister.Index ) {
+         TXT( "\nIndex    : " );
+         SID( src->SrcRegister.Index );
+      }
+      if( ignored ) {
+         if( deflt || fs->SrcRegister.Extended != src->SrcRegister.Extended ) {
+            TXT( "\nExtended : " );
+            UID( src->SrcRegister.Extended );
+         }
+      }
+
+      if( deflt || tgsi_compare_src_register_ext_swz( src->SrcRegisterExtSwz, fs->SrcRegisterExtSwz ) ) {
+         CHR( '\n' );
+         TXT( "\nType       : " );
+         ENM( src->SrcRegisterExtSwz.Type, TGSI_SRC_REGISTER_EXTS );
+         if( deflt || fs->SrcRegisterExtSwz.ExtSwizzleX != src->SrcRegisterExtSwz.ExtSwizzleX ) {
+            TXT( "\nExtSwizzleX: " );
+            ENM( src->SrcRegisterExtSwz.ExtSwizzleX, TGSI_EXTSWIZZLES );
+         }
+         if( deflt || fs->SrcRegisterExtSwz.ExtSwizzleY != src->SrcRegisterExtSwz.ExtSwizzleY ) {
+            TXT( "\nExtSwizzleY: " );
+            ENM( src->SrcRegisterExtSwz.ExtSwizzleY, TGSI_EXTSWIZZLES );
+         }
+         if( deflt || fs->SrcRegisterExtSwz.ExtSwizzleZ != src->SrcRegisterExtSwz.ExtSwizzleZ ) {
+            TXT( "\nExtSwizzleZ: " );
+            ENM( src->SrcRegisterExtSwz.ExtSwizzleZ, TGSI_EXTSWIZZLES );
+         }
+         if( deflt || fs->SrcRegisterExtSwz.ExtSwizzleW != src->SrcRegisterExtSwz.ExtSwizzleW ) {
+            TXT( "\nExtSwizzleW: " );
+            ENM( src->SrcRegisterExtSwz.ExtSwizzleW, TGSI_EXTSWIZZLES );
+         }
+         if( deflt || fs->SrcRegisterExtSwz.NegateX != src->SrcRegisterExtSwz.NegateX ) {
+            TXT( "\nNegateX   : " );
+            UID( src->SrcRegisterExtSwz.NegateX );
+         }
+         if( deflt || fs->SrcRegisterExtSwz.NegateY != src->SrcRegisterExtSwz.NegateY ) {
+            TXT( "\nNegateY   : " );
+            UID( src->SrcRegisterExtSwz.NegateY );
+         }
+         if( deflt || fs->SrcRegisterExtSwz.NegateZ != src->SrcRegisterExtSwz.NegateZ ) {
+            TXT( "\nNegateZ   : " );
+            UID( src->SrcRegisterExtSwz.NegateZ );
+         }
+         if( deflt || fs->SrcRegisterExtSwz.NegateW != src->SrcRegisterExtSwz.NegateW ) {
+            TXT( "\nNegateW   : " );
+            UID( src->SrcRegisterExtSwz.NegateW );
+         }
+         if( deflt || fs->SrcRegisterExtSwz.ExtDivide != src->SrcRegisterExtSwz.ExtDivide ) {
+            TXT( "\nExtDivide  : " );
+            ENM( src->SrcRegisterExtSwz.ExtDivide, TGSI_EXTSWIZZLES );
+         }
+         if( ignored ) {
+            TXT( "\nPadding   : " );
+            UIX( src->SrcRegisterExtSwz.Padding );
+            if( deflt || fs->SrcRegisterExtSwz.Extended != src->SrcRegisterExtSwz.Extended ) {
+               TXT( "\nExtended   : " );
+               UID( src->SrcRegisterExtSwz.Extended );
+            }
+         }
+      }
+
+      if( deflt || tgsi_compare_src_register_ext_mod( src->SrcRegisterExtMod, fs->SrcRegisterExtMod ) ) {
+         CHR( '\n' );
+         TXT( "\nType     : " );
+         ENM( src->SrcRegisterExtMod.Type, TGSI_SRC_REGISTER_EXTS );
+         if( deflt || fs->SrcRegisterExtMod.Complement != src->SrcRegisterExtMod.Complement ) {
+            TXT( "\nComplement: " );
+            UID( src->SrcRegisterExtMod.Complement );
+         }
+         if( deflt || fs->SrcRegisterExtMod.Bias != src->SrcRegisterExtMod.Bias ) {
+            TXT( "\nBias     : " );
+            UID( src->SrcRegisterExtMod.Bias );
+         }
+         if( deflt || fs->SrcRegisterExtMod.Scale2X != src->SrcRegisterExtMod.Scale2X ) {
+            TXT( "\nScale2X   : " );
+            UID( src->SrcRegisterExtMod.Scale2X );
+         }
+         if( deflt || fs->SrcRegisterExtMod.Absolute != src->SrcRegisterExtMod.Absolute ) {
+            TXT( "\nAbsolute  : " );
+            UID( src->SrcRegisterExtMod.Absolute );
+         }
+         if( deflt || fs->SrcRegisterExtMod.Negate != src->SrcRegisterExtMod.Negate ) {
+            TXT( "\nNegate   : " );
+            UID( src->SrcRegisterExtMod.Negate );
+         }
+         if( ignored ) {
+            TXT( "\nPadding   : " );
+            UIX( src->SrcRegisterExtMod.Padding );
+            if( deflt || fs->SrcRegisterExtMod.Extended != src->SrcRegisterExtMod.Extended ) {
+               TXT( "\nExtended  : " );
+               UID( src->SrcRegisterExtMod.Extended );
+            }
+         }
+      }
+   }
+}
+
+static void
+dump_gen(
+   struct gen_dump         *dump,
+   const struct tgsi_token *tokens,
+   unsigned                flags )
+{
+   struct tgsi_parse_context parse;
+   struct tgsi_full_instruction fi;
+   struct tgsi_full_declaration fd;
+   unsigned verbose = flags & TGSI_DUMP_VERBOSE;
+   unsigned ignored = !(flags & TGSI_DUMP_NO_IGNORED);
+   unsigned deflt = !(flags & TGSI_DUMP_NO_DEFAULT);
+   unsigned instno = 0;
+
+   dump->tabs = 0;
+
+   /* sanity check */
+   assert(strcmp(TGSI_OPCODES[TGSI_OPCODE_CONT], "OPCODE_CONT") == 0);
+
+   tgsi_parse_init( &parse, tokens );
+
+   TXT( "tgsi-dump begin -----------------" );
+
+   CHR( '\n' );
+   ENM( parse.FullHeader.Processor.Processor, TGSI_PROCESSOR_TYPES_SHORT );
+   CHR( ' ' );
+   UID( parse.FullVersion.Version.MajorVersion );
+   CHR( '.' );
+   UID( parse.FullVersion.Version.MinorVersion );
+
+   if( verbose ) {
+      TXT( "\nMajorVersion: " );
+      UID( parse.FullVersion.Version.MajorVersion );
+      TXT( "\nMinorVersion: " );
+      UID( parse.FullVersion.Version.MinorVersion );
+      CHR( '\n' );
+
+      TXT( "\nHeaderSize: " );
+      UID( parse.FullHeader.Header.HeaderSize );
+      TXT( "\nBodySize  : " );
+      UID( parse.FullHeader.Header.BodySize );
+      TXT( "\nProcessor : " );
+      ENM( parse.FullHeader.Processor.Processor, TGSI_PROCESSOR_TYPES );
+      CHR( '\n' );
+   }
+
+   fi = tgsi_default_full_instruction();
+   fd = tgsi_default_full_declaration();
+
+   while( !tgsi_parse_end_of_tokens( &parse ) ) {
+      tgsi_parse_token( &parse );
+
+      switch( parse.FullToken.Token.Type ) {
+      case TGSI_TOKEN_TYPE_DECLARATION:
+         dump_declaration_short(
+            dump,
+            &parse.FullToken.FullDeclaration );
+         break;
+
+      case TGSI_TOKEN_TYPE_IMMEDIATE:
+         dump_immediate_short(
+            dump,
+            &parse.FullToken.FullImmediate );
+         break;
+
+      case TGSI_TOKEN_TYPE_INSTRUCTION:
+         dump_instruction_short(
+            dump,
+            &parse.FullToken.FullInstruction,
+            instno );
+         instno++;
+         break;
+
+      default:
+         assert( 0 );
+      }
+
+      if( verbose ) {
+         TXT( "\nType       : " );
+         ENM( parse.FullToken.Token.Type, TGSI_TOKEN_TYPES );
+         if( ignored ) {
+            TXT( "\nSize       : " );
+            UID( parse.FullToken.Token.Size );
+            if( deflt || parse.FullToken.Token.Extended ) {
+               TXT( "\nExtended   : " );
+               UID( parse.FullToken.Token.Extended );
+            }
+         }
+
+         switch( parse.FullToken.Token.Type ) {
+         case TGSI_TOKEN_TYPE_DECLARATION:
+            dump_declaration_verbose(
+               dump,
+               &parse.FullToken.FullDeclaration,
+               ignored,
+               deflt,
+               &fd );
+            break;
+
+         case TGSI_TOKEN_TYPE_IMMEDIATE:
+            dump_immediate_verbose(
+               dump,
+               &parse.FullToken.FullImmediate,
+               ignored );
+            break;
+
+         case TGSI_TOKEN_TYPE_INSTRUCTION:
+            dump_instruction_verbose(
+               dump,
+               &parse.FullToken.FullInstruction,
+               ignored,
+               deflt,
+               &fi );
+            break;
+
+         default:
+            assert( 0 );
+         }
+
+         CHR( '\n' );
+      }
+   }
+
+   TXT( "\ntgsi-dump end -------------------\n" );
+
+   tgsi_parse_free( &parse );
+}
+
+
+static void
+sanity_checks(void)
+{
+   assert(strcmp(TGSI_OPCODES[TGSI_OPCODE_END], "OPCODE_END") == 0);
+   assert(strcmp(TGSI_OPCODES_SHORT[TGSI_OPCODE_END], "END") == 0);
+}
+
+
+void
+tgsi_dump(
+   const struct tgsi_token *tokens,
+   unsigned                flags )
+{
+   struct file_dump  dump;
+
+   sanity_checks();
+
+   dump.base.write = _file_dump_write;
+#if 0
+   {
+      static unsigned   counter = 0;
+      char              buffer[64];
+      sprintf( buffer, "tgsi-dump-%.4u.txt", counter++ );
+      dump.file = fopen( buffer, "wt" );
+   }
+#else
+   dump.file = stderr;
+#endif
+
+   dump_gen(
+      &dump.base,
+      tokens,
+      flags );
+
+#if 0
+   fclose( dump.file );
+#endif
+}
+
+void
+tgsi_dump_str(
+   char                    **str,
+   const struct tgsi_token *tokens,
+   unsigned                flags )
+{
+   struct text_dump  dump;
+
+   dump.base.write = _text_dump_write;
+   dump.text = NULL;
+   dump.length = 0;
+   dump.capacity = 0;
+
+   dump_gen(
+      &dump.base,
+      tokens,
+      flags );
+
+   *str = dump.text;
+}
diff --git a/src/gallium/aux/tgsi/util/tgsi_dump.h b/src/gallium/aux/tgsi/util/tgsi_dump.h
new file mode 100644
index 0000000000..1adc9db251
--- /dev/null
+++ b/src/gallium/aux/tgsi/util/tgsi_dump.h
@@ -0,0 +1,28 @@
+#if !defined TGSI_DUMP_H
+#define TGSI_DUMP_H
+
+#if defined __cplusplus
+extern "C" {
+#endif // defined __cplusplus
+
+#define TGSI_DUMP_VERBOSE       1
+#define TGSI_DUMP_NO_IGNORED    2
+#define TGSI_DUMP_NO_DEFAULT    4
+
+void
+tgsi_dump(
+   const struct tgsi_token *tokens,
+   unsigned                flags );
+
+void
+tgsi_dump_str(
+   char                    **str,
+   const struct tgsi_token *tokens,
+   unsigned                flags );
+
+#if defined __cplusplus
+} // extern "C"
+#endif // defined __cplusplus
+
+#endif // !defined TGSI_DUMP_H
+
diff --git a/src/gallium/aux/tgsi/util/tgsi_parse.c b/src/gallium/aux/tgsi/util/tgsi_parse.c
new file mode 100644
index 0000000000..bf6b89ce56
--- /dev/null
+++ b/src/gallium/aux/tgsi/util/tgsi_parse.c
@@ -0,0 +1,319 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+#include "pipe/p_debug.h"
+#include "pipe/p_util.h"
+#include "pipe/p_shader_tokens.h"
+#include "tgsi_parse.h"
+#include "tgsi_build.h"
+
+void
+tgsi_full_token_init(
+   union tgsi_full_token *full_token )
+{
+   full_token->Token.Type = TGSI_TOKEN_TYPE_DECLARATION;
+}
+
+void
+tgsi_full_token_free(
+   union tgsi_full_token *full_token )
+{
+   if( full_token->Token.Type == TGSI_TOKEN_TYPE_IMMEDIATE ) {
+      FREE( full_token->FullImmediate.u.Pointer );
+   }
+}
+
+unsigned
+tgsi_parse_init(
+   struct tgsi_parse_context *ctx,
+   const struct tgsi_token *tokens )
+{
+   ctx->FullVersion.Version = *(struct tgsi_version *) &tokens[0];
+   if( ctx->FullVersion.Version.MajorVersion > 1 ) {
+      return TGSI_PARSE_ERROR;
+   }
+
+   ctx->FullHeader.Header = *(struct tgsi_header *) &tokens[1];
+   if( ctx->FullHeader.Header.HeaderSize >= 2 ) {
+      ctx->FullHeader.Processor = *(struct tgsi_processor *) &tokens[2];
+   }
+   else {
+      ctx->FullHeader.Processor = tgsi_default_processor();
+   }
+
+   ctx->Tokens = tokens;
+   ctx->Position = 1 + ctx->FullHeader.Header.HeaderSize;
+
+   tgsi_full_token_init( &ctx->FullToken );
+
+   return TGSI_PARSE_OK;
+}
+
+void
+tgsi_parse_free(
+   struct tgsi_parse_context *ctx )
+{
+   tgsi_full_token_free( &ctx->FullToken );
+}
+
+boolean
+tgsi_parse_end_of_tokens(
+   struct tgsi_parse_context *ctx )
+{
+   return ctx->Position >=
+      1 + ctx->FullHeader.Header.HeaderSize + ctx->FullHeader.Header.BodySize;
+}
+
+static void
+next_token(
+   struct tgsi_parse_context *ctx,
+   void *token )
+{
+   assert( !tgsi_parse_end_of_tokens( ctx ) );
+
+   *(struct tgsi_token *) token = ctx->Tokens[ctx->Position++];
+}
+
+void
+tgsi_parse_token(
+   struct tgsi_parse_context *ctx )
+{
+   struct tgsi_token token;
+   unsigned i;
+
+   tgsi_full_token_free( &ctx->FullToken );
+   tgsi_full_token_init( &ctx->FullToken );
+
+   next_token( ctx, &token );
+
+   switch( token.Type ) {
+   case TGSI_TOKEN_TYPE_DECLARATION:
+   {
+      struct tgsi_full_declaration *decl = &ctx->FullToken.FullDeclaration;
+
+      *decl = tgsi_default_full_declaration();
+      decl->Declaration = *(struct tgsi_declaration *) &token;
+
+      switch( decl->Declaration.Type ) {
+      case TGSI_DECLARE_RANGE:
+         next_token( ctx, &decl->u.DeclarationRange );
+         break;
+
+      case TGSI_DECLARE_MASK:
+         next_token( ctx, &decl->u.DeclarationMask );
+         break;
+
+      default:
+         assert (0);
+      }
+
+      if( decl->Declaration.Interpolate ) {
+         next_token( ctx, &decl->Interpolation );
+      }
+
+      if( decl->Declaration.Semantic ) {
+         next_token( ctx, &decl->Semantic );
+      }
+
+      break;
+   }
+
+   case TGSI_TOKEN_TYPE_IMMEDIATE:
+   {
+      struct tgsi_full_immediate *imm = &ctx->FullToken.FullImmediate;
+
+      *imm = tgsi_default_full_immediate();
+      imm->Immediate = *(struct tgsi_immediate *) &token;
+
+      assert( !imm->Immediate.Extended );
+
+      switch (imm->Immediate.DataType) {
+      case TGSI_IMM_FLOAT32:
+         imm->u.Pointer = MALLOC(
+            sizeof( struct tgsi_immediate_float32 ) * (imm->Immediate.Size - 1) );
+         for( i = 0; i < imm->Immediate.Size - 1; i++ ) {
+            next_token( ctx, &imm->u.ImmediateFloat32[i] );
+         }
+         break;
+
+      default:
+         assert( 0 );
+      }
+
+      break;
+   }
+
+   case TGSI_TOKEN_TYPE_INSTRUCTION:
+   {
+      struct tgsi_full_instruction *inst = &ctx->FullToken.FullInstruction;
+      unsigned extended;
+
+      *inst = tgsi_default_full_instruction();
+      inst->Instruction = *(struct tgsi_instruction *) &token;
+
+      extended = inst->Instruction.Extended;
+
+      while( extended ) {
+         struct tgsi_src_register_ext token;
+
+         next_token( ctx, &token );
+
+         switch( token.Type ) {
+         case TGSI_INSTRUCTION_EXT_TYPE_NV:
+            inst->InstructionExtNv =
+               *(struct tgsi_instruction_ext_nv *) &token;
+            break;
+
+         case TGSI_INSTRUCTION_EXT_TYPE_LABEL:
+            inst->InstructionExtLabel =
+               *(struct tgsi_instruction_ext_label *) &token;
+            break;
+
+         case TGSI_INSTRUCTION_EXT_TYPE_TEXTURE:
+            inst->InstructionExtTexture =
+               *(struct tgsi_instruction_ext_texture *) &token;
+            break;
+
+         default:
+            assert( 0 );
+         }
+
+         extended = token.Extended;
+      }
+
+      assert( inst->Instruction.NumDstRegs <= TGSI_FULL_MAX_DST_REGISTERS );
+
+      for(  i = 0; i < inst->Instruction.NumDstRegs; i++ ) {
+         unsigned extended;
+
+         next_token( ctx, &inst->FullDstRegisters[i].DstRegister );
+
+         /*
+          * No support for indirect or multi-dimensional addressing.
+          */
+         assert( !inst->FullDstRegisters[i].DstRegister.Indirect );
+         assert( !inst->FullDstRegisters[i].DstRegister.Dimension );
+
+         extended = inst->FullDstRegisters[i].DstRegister.Extended;
+
+         while( extended ) {
+            struct tgsi_src_register_ext token;
+
+            next_token( ctx, &token );
+
+            switch( token.Type ) {
+            case TGSI_DST_REGISTER_EXT_TYPE_CONDCODE:
+               inst->FullDstRegisters[i].DstRegisterExtConcode =
+                  *(struct tgsi_dst_register_ext_concode *) &token;
+               break;
+
+            case TGSI_DST_REGISTER_EXT_TYPE_MODULATE:
+               inst->FullDstRegisters[i].DstRegisterExtModulate =
+                  *(struct tgsi_dst_register_ext_modulate *) &token;
+               break;
+
+            default:
+               assert( 0 );
+            }
+
+            extended = token.Extended;
+         }
+      }
+
+      assert( inst->Instruction.NumSrcRegs <= TGSI_FULL_MAX_SRC_REGISTERS );
+
+      for( i = 0; i < inst->Instruction.NumSrcRegs; i++ ) {
+         unsigned extended;
+
+         next_token( ctx, &inst->FullSrcRegisters[i].SrcRegister );
+
+         extended = inst->FullSrcRegisters[i].SrcRegister.Extended;
+
+         while( extended ) {
+            struct tgsi_src_register_ext token;
+
+            next_token( ctx, &token );
+
+            switch( token.Type ) {
+            case TGSI_SRC_REGISTER_EXT_TYPE_SWZ:
+               inst->FullSrcRegisters[i].SrcRegisterExtSwz =
+                  *(struct tgsi_src_register_ext_swz *) &token;
+               break;
+
+            case TGSI_SRC_REGISTER_EXT_TYPE_MOD:
+               inst->FullSrcRegisters[i].SrcRegisterExtMod =
+                  *(struct tgsi_src_register_ext_mod *) &token;
+               break;
+
+            default:
+               assert( 0 );
+            }
+
+            extended = token.Extended;
+         }
+
+         if( inst->FullSrcRegisters[i].SrcRegister.Indirect ) {
+            next_token( ctx, &inst->FullSrcRegisters[i].SrcRegisterInd );
+
+            /*
+             * No support for indirect or multi-dimensional addressing.
+             */
+            assert( !inst->FullSrcRegisters[i].SrcRegisterInd.Indirect );
+            assert( !inst->FullSrcRegisters[i].SrcRegisterInd.Dimension );
+            assert( !inst->FullSrcRegisters[i].SrcRegisterInd.Extended );
+         }
+
+         if( inst->FullSrcRegisters[i].SrcRegister.Dimension ) {
+            next_token( ctx, &inst->FullSrcRegisters[i].SrcRegisterDim );
+
+            /*
+             * No support for multi-dimensional addressing.
+             */
+            assert( !inst->FullSrcRegisters[i].SrcRegisterDim.Dimension );
+            assert( !inst->FullSrcRegisters[i].SrcRegisterDim.Extended );
+
+            if( inst->FullSrcRegisters[i].SrcRegisterDim.Indirect ) {
+               next_token( ctx, &inst->FullSrcRegisters[i].SrcRegisterDimInd );
+
+               /*
+               * No support for indirect or multi-dimensional addressing.
+               */
+               assert( !inst->FullSrcRegisters[i].SrcRegisterInd.Indirect );
+               assert( !inst->FullSrcRegisters[i].SrcRegisterInd.Dimension );
+               assert( !inst->FullSrcRegisters[i].SrcRegisterInd.Extended );
+            }
+         }
+      }
+
+      break;
+   }
+
+   default:
+      assert( 0 );
+   }
+}
+
diff --git a/src/gallium/aux/tgsi/util/tgsi_parse.h b/src/gallium/aux/tgsi/util/tgsi_parse.h
new file mode 100644
index 0000000000..9372da8d5d
--- /dev/null
+++ b/src/gallium/aux/tgsi/util/tgsi_parse.h
@@ -0,0 +1,121 @@
+#if !defined TGSI_PARSE_H
+#define TGSI_PARSE_H
+
+#if defined __cplusplus
+extern "C" {
+#endif // defined __cplusplus
+
+struct tgsi_full_version
+{
+   struct tgsi_version  Version;
+};
+
+struct tgsi_full_header
+{
+   struct tgsi_header      Header;
+   struct tgsi_processor   Processor;
+};
+
+struct tgsi_full_dst_register
+{
+   struct tgsi_dst_register               DstRegister;
+   struct tgsi_dst_register_ext_concode   DstRegisterExtConcode;
+   struct tgsi_dst_register_ext_modulate  DstRegisterExtModulate;
+};
+
+struct tgsi_full_src_register
+{
+   struct tgsi_src_register         SrcRegister;
+   struct tgsi_src_register_ext_swz SrcRegisterExtSwz;
+   struct tgsi_src_register_ext_mod SrcRegisterExtMod;
+   struct tgsi_src_register         SrcRegisterInd;
+   struct tgsi_dimension            SrcRegisterDim;
+   struct tgsi_src_register         SrcRegisterDimInd;
+};
+
+struct tgsi_full_declaration
+{
+   struct tgsi_declaration Declaration;
+   union
+   {
+      struct tgsi_declaration_range DeclarationRange;
+      struct tgsi_declaration_mask  DeclarationMask;
+   } u;
+   struct tgsi_declaration_interpolation  Interpolation;
+   struct tgsi_declaration_semantic       Semantic;
+};
+
+struct tgsi_full_immediate
+{
+   struct tgsi_immediate   Immediate;
+   union
+   {
+      void                          *Pointer;
+      struct tgsi_immediate_float32 *ImmediateFloat32;
+   } u;
+};
+
+#define TGSI_FULL_MAX_DST_REGISTERS 2
+#define TGSI_FULL_MAX_SRC_REGISTERS 3
+
+struct tgsi_full_instruction
+{
+   struct tgsi_instruction             Instruction;
+   struct tgsi_instruction_ext_nv      InstructionExtNv;
+   struct tgsi_instruction_ext_label   InstructionExtLabel;
+   struct tgsi_instruction_ext_texture InstructionExtTexture;
+   struct tgsi_full_dst_register       FullDstRegisters[TGSI_FULL_MAX_DST_REGISTERS];
+   struct tgsi_full_src_register       FullSrcRegisters[TGSI_FULL_MAX_SRC_REGISTERS];
+};
+
+union tgsi_full_token
+{
+   struct tgsi_token             Token;
+   struct tgsi_full_declaration  FullDeclaration;
+   struct tgsi_full_immediate    FullImmediate;
+   struct tgsi_full_instruction  FullInstruction;
+};
+
+void
+tgsi_full_token_init(
+   union tgsi_full_token *full_token );
+
+void
+tgsi_full_token_free(
+   union tgsi_full_token *full_token );
+
+struct tgsi_parse_context
+{
+   const struct tgsi_token    *Tokens;
+   unsigned                   Position;
+   struct tgsi_full_version   FullVersion;
+   struct tgsi_full_header    FullHeader;
+   union tgsi_full_token      FullToken;
+};
+
+#define TGSI_PARSE_OK      0
+#define TGSI_PARSE_ERROR   1
+
+unsigned
+tgsi_parse_init(
+   struct tgsi_parse_context *ctx,
+   const struct tgsi_token *tokens );
+
+void
+tgsi_parse_free(
+   struct tgsi_parse_context *ctx );
+
+boolean
+tgsi_parse_end_of_tokens(
+   struct tgsi_parse_context *ctx );
+
+void
+tgsi_parse_token(
+   struct tgsi_parse_context *ctx );
+
+#if defined __cplusplus
+} // extern "C"
+#endif // defined __cplusplus
+
+#endif // !defined TGSI_PARSE_H
+
diff --git a/src/gallium/aux/tgsi/util/tgsi_transform.c b/src/gallium/aux/tgsi/util/tgsi_transform.c
new file mode 100644
index 0000000000..357f77b05a
--- /dev/null
+++ b/src/gallium/aux/tgsi/util/tgsi_transform.c
@@ -0,0 +1,199 @@
+/**************************************************************************
+ * 
+ * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+/**
+ * TGSI program transformation utility.
+ *
+ * Authors:  Brian Paul
+ */
+
+
+#include "tgsi_transform.h"
+
+
+
+static void
+emit_instruction(struct tgsi_transform_context *ctx,
+                 const struct tgsi_full_instruction *inst)
+{
+   uint ti = ctx->ti;
+
+   ti += tgsi_build_full_instruction(inst,
+                                     ctx->tokens_out + ti,
+                                     ctx->header,
+                                     ctx->max_tokens_out - ti);
+   ctx->ti = ti;
+}
+
+
+static void
+emit_declaration(struct tgsi_transform_context *ctx,
+                 const struct tgsi_full_declaration *decl)
+{
+   uint ti = ctx->ti;
+
+   ti += tgsi_build_full_declaration(decl,
+                                     ctx->tokens_out + ti,
+                                     ctx->header,
+                                     ctx->max_tokens_out - ti);
+   ctx->ti = ti;
+}
+
+
+static void
+emit_immediate(struct tgsi_transform_context *ctx,
+               const struct tgsi_full_immediate *imm)
+{
+   uint ti = ctx->ti;
+
+   ti += tgsi_build_full_immediate(imm,
+                                   ctx->tokens_out + ti,
+                                   ctx->header,
+                                   ctx->max_tokens_out - ti);
+   ctx->ti = ti;
+}
+
+
+
+/**
+ * Apply user-defined transformations to the input shader to produce
+ * the output shader.
+ * For example, a register search-and-replace operation could be applied
+ * by defining a transform_instruction() callback that examined and changed
+ * the instruction src/dest regs.
+ *
+ * \return number of tokens emitted
+ */
+int
+tgsi_transform_shader(const struct tgsi_token *tokens_in,
+                      struct tgsi_token *tokens_out,
+                      uint max_tokens_out,
+                      struct tgsi_transform_context *ctx)
+{
+   uint procType;
+
+   /* input shader */
+   struct tgsi_parse_context parse;
+
+   /* output shader */
+   struct tgsi_processor *processor;
+
+
+   /**
+    ** callback context init
+    **/
+   ctx->emit_instruction = emit_instruction;
+   ctx->emit_declaration = emit_declaration;
+   ctx->emit_immediate = emit_immediate;
+   ctx->tokens_out = tokens_out;
+   ctx->max_tokens_out = max_tokens_out;
+
+
+   /**
+    ** Setup to begin parsing input shader
+    **/
+   if (tgsi_parse_init( &parse, tokens_in ) != TGSI_PARSE_OK) {
+      debug_printf("tgsi_parse_init() failed in tgsi_transform_shader()!\n");
+      return -1;
+   }
+   procType = parse.FullHeader.Processor.Processor;
+   assert(procType == TGSI_PROCESSOR_FRAGMENT ||
+          procType == TGSI_PROCESSOR_VERTEX ||
+          procType == TGSI_PROCESSOR_GEOMETRY);
+
+
+   /**
+    **  Setup output shader
+    **/
+   *(struct tgsi_version *) &tokens_out[0] = tgsi_build_version();
+
+   ctx->header = (struct tgsi_header *) (tokens_out + 1);
+   *ctx->header = tgsi_build_header();
+
+   processor = (struct tgsi_processor *) (tokens_out + 2);
+   *processor = tgsi_build_processor( procType, ctx->header );
+
+   ctx->ti = 3;
+
+
+   /**
+    ** Loop over incoming program tokens/instructions
+    */
+   while( !tgsi_parse_end_of_tokens( &parse ) ) {
+
+      tgsi_parse_token( &parse );
+
+      switch( parse.FullToken.Token.Type ) {
+      case TGSI_TOKEN_TYPE_INSTRUCTION:
+         {
+            struct tgsi_full_instruction *fullinst
+               = &parse.FullToken.FullInstruction;
+
+            if (ctx->transform_instruction)
+               ctx->transform_instruction(ctx, fullinst);
+            else
+               ctx->emit_instruction(ctx, fullinst);
+         }
+         break;
+
+      case TGSI_TOKEN_TYPE_DECLARATION:
+         {
+            struct tgsi_full_declaration *fulldecl
+               = &parse.FullToken.FullDeclaration;
+
+            if (ctx->transform_declaration)
+               ctx->transform_declaration(ctx, fulldecl);
+            else
+               ctx->emit_declaration(ctx, fulldecl);
+         }
+         break;
+
+      case TGSI_TOKEN_TYPE_IMMEDIATE:
+         {
+            struct tgsi_full_immediate *fullimm
+               = &parse.FullToken.FullImmediate;
+
+            if (ctx->transform_immediate)
+               ctx->transform_immediate(ctx, fullimm);
+            else
+               ctx->emit_immediate(ctx, fullimm);
+         }
+         break;
+
+      default:
+         assert( 0 );
+      }
+   }
+
+   if (ctx->epilog) {
+      ctx->epilog(ctx);
+   }
+
+   tgsi_parse_free (&parse);
+
+   return ctx->ti;
+}
diff --git a/src/gallium/aux/tgsi/util/tgsi_transform.h b/src/gallium/aux/tgsi/util/tgsi_transform.h
new file mode 100644
index 0000000000..365d8c298c
--- /dev/null
+++ b/src/gallium/aux/tgsi/util/tgsi_transform.h
@@ -0,0 +1,93 @@
+/**************************************************************************
+ * 
+ * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+#ifndef TGSI_TRANSFORM_H
+#define TGSI_TRANSFORM_H
+
+
+#include "pipe/p_util.h"
+#include "pipe/p_shader_tokens.h"
+#include "pipe/tgsi/util/tgsi_parse.h"
+#include "pipe/tgsi/util/tgsi_build.h"
+
+
+
+/**
+ * Subclass this to add caller-specific data
+ */
+struct tgsi_transform_context
+{
+/**** PUBLIC ***/
+
+   /**
+    * User-defined callbacks invoked per instruction.
+    */
+   void (*transform_instruction)(struct tgsi_transform_context *ctx,
+                                 struct tgsi_full_instruction *inst);
+
+   void (*transform_declaration)(struct tgsi_transform_context *ctx,
+                                 struct tgsi_full_declaration *decl);
+
+   void (*transform_immediate)(struct tgsi_transform_context *ctx,
+                               struct tgsi_full_immediate *imm);
+
+   /**
+    * Called at end of input program to allow caller to append extra
+    * instructions.  Return number of tokens emitted.
+    */
+   void (*epilog)(struct tgsi_transform_context *ctx);
+
+
+/*** PRIVATE ***/
+
+   /**
+    * These are setup by tgsi_transform_shader() and cannot be overridden.
+    * Meant to be called from in the above user callback functions.
+    */
+   void (*emit_instruction)(struct tgsi_transform_context *ctx,
+                            const struct tgsi_full_instruction *inst);
+   void (*emit_declaration)(struct tgsi_transform_context *ctx,
+                            const struct tgsi_full_declaration *decl);
+   void (*emit_immediate)(struct tgsi_transform_context *ctx,
+                          const struct tgsi_full_immediate *imm);
+
+   struct tgsi_header *header;
+   uint max_tokens_out;
+   struct tgsi_token *tokens_out;
+   uint ti;
+};
+
+
+
+extern int
+tgsi_transform_shader(const struct tgsi_token *tokens_in,
+                      struct tgsi_token *tokens_out,
+                      uint max_tokens_out,
+                      struct tgsi_transform_context *ctx);
+
+
+#endif /* TGSI_TRANSFORM_H */
diff --git a/src/gallium/aux/tgsi/util/tgsi_util.c b/src/gallium/aux/tgsi/util/tgsi_util.c
new file mode 100644
index 0000000000..4cdd89182a
--- /dev/null
+++ b/src/gallium/aux/tgsi/util/tgsi_util.c
@@ -0,0 +1,274 @@
+#include "pipe/p_debug.h"
+#include "pipe/p_util.h"
+#include "pipe/p_shader_tokens.h"
+#include "tgsi_parse.h"
+#include "tgsi_build.h"
+#include "tgsi_util.h"
+
+union pointer_hack
+{
+   void *pointer;
+   unsigned long long uint64;
+};
+
+void *
+tgsi_align_128bit(
+   void *unaligned )
+{
+   union pointer_hack ph;
+
+   ph.uint64 = 0;
+   ph.pointer = unaligned;
+   ph.uint64 = (ph.uint64 + 15) & ~15;
+   return ph.pointer;
+}
+
+unsigned
+tgsi_util_get_src_register_swizzle(
+   const struct tgsi_src_register *reg,
+   unsigned component )
+{
+   switch( component ) {
+   case 0:
+      return reg->SwizzleX;
+   case 1:
+      return reg->SwizzleY;
+   case 2:
+      return reg->SwizzleZ;
+   case 3:
+      return reg->SwizzleW;
+   default:
+      assert( 0 );
+   }
+   return 0;
+}
+
+unsigned
+tgsi_util_get_src_register_extswizzle(
+   const struct tgsi_src_register_ext_swz *reg,
+   unsigned component )
+{
+   switch( component ) {
+   case 0:
+      return reg->ExtSwizzleX;
+   case 1:
+      return reg->ExtSwizzleY;
+   case 2:
+      return reg->ExtSwizzleZ;
+   case 3:
+      return reg->ExtSwizzleW;
+   default:
+      assert( 0 );
+   }
+   return 0;
+}
+
+unsigned
+tgsi_util_get_full_src_register_extswizzle(
+   const struct tgsi_full_src_register  *reg,
+   unsigned component )
+{
+   unsigned swizzle;
+
+   /*
+    * First, calculate  the   extended swizzle for a given channel. This will give
+    * us either a channel index into the simple swizzle or  a constant 1 or   0.
+    */
+   swizzle = tgsi_util_get_src_register_extswizzle(
+      &reg->SrcRegisterExtSwz,
+      component );
+
+   assert (TGSI_SWIZZLE_X == TGSI_EXTSWIZZLE_X);
+   assert (TGSI_SWIZZLE_Y == TGSI_EXTSWIZZLE_Y);
+   assert (TGSI_SWIZZLE_Z == TGSI_EXTSWIZZLE_Z);
+   assert (TGSI_SWIZZLE_W == TGSI_EXTSWIZZLE_W);
+   assert (TGSI_EXTSWIZZLE_ZERO > TGSI_SWIZZLE_W);
+   assert (TGSI_EXTSWIZZLE_ONE > TGSI_SWIZZLE_W);
+
+   /*
+    * Second, calculate the simple  swizzle  for   the   unswizzled channel index.
+    * Leave the constants intact, they are   not   affected by the   simple swizzle.
+    */
+   if( swizzle <= TGSI_SWIZZLE_W ) {
+      swizzle = tgsi_util_get_src_register_swizzle(
+         &reg->SrcRegister,
+         component );
+   }
+
+   return swizzle;
+}
+
+void
+tgsi_util_set_src_register_swizzle(
+   struct tgsi_src_register *reg,
+   unsigned swizzle,
+   unsigned component )
+{
+   switch( component ) {
+   case 0:
+      reg->SwizzleX = swizzle;
+      break;
+   case 1:
+      reg->SwizzleY = swizzle;
+      break;
+   case 2:
+      reg->SwizzleZ = swizzle;
+      break;
+   case 3:
+      reg->SwizzleW = swizzle;
+      break;
+   default:
+      assert( 0 );
+   }
+}
+
+void
+tgsi_util_set_src_register_extswizzle(
+   struct tgsi_src_register_ext_swz *reg,
+   unsigned swizzle,
+   unsigned component )
+{
+   switch( component ) {
+   case 0:
+      reg->ExtSwizzleX = swizzle;
+      break;
+   case 1:
+      reg->ExtSwizzleY = swizzle;
+      break;
+   case 2:
+      reg->ExtSwizzleZ = swizzle;
+      break;
+   case 3:
+      reg->ExtSwizzleW = swizzle;
+      break;
+   default:
+      assert( 0 );
+   }
+}
+
+unsigned
+tgsi_util_get_src_register_extnegate(
+   const  struct tgsi_src_register_ext_swz *reg,
+   unsigned component )
+{
+   switch( component ) {
+   case 0:
+      return reg->NegateX;
+   case 1:
+      return reg->NegateY;
+   case 2:
+      return reg->NegateZ;
+   case 3:
+      return reg->NegateW;
+   default:
+      assert( 0 );
+   }
+   return 0;
+}
+
+void
+tgsi_util_set_src_register_extnegate(
+   struct tgsi_src_register_ext_swz *reg,
+   unsigned negate,
+   unsigned component )
+{
+   switch( component ) {
+   case 0:
+      reg->NegateX = negate;
+      break;
+   case 1:
+      reg->NegateY = negate;
+      break;
+   case 2:
+      reg->NegateZ = negate;
+      break;
+   case 3:
+      reg->NegateW = negate;
+      break;
+   default:
+      assert( 0 );
+   }
+}
+
+unsigned
+tgsi_util_get_full_src_register_sign_mode(
+   const struct  tgsi_full_src_register *reg,
+   unsigned component )
+{
+   unsigned sign_mode;
+
+   if( reg->SrcRegisterExtMod.Absolute ) {
+      /* Consider only the post-abs negation. */
+
+      if( reg->SrcRegisterExtMod.Negate ) {
+         sign_mode = TGSI_UTIL_SIGN_SET;
+      }
+      else {
+         sign_mode = TGSI_UTIL_SIGN_CLEAR;
+      }
+   }
+   else {
+      /* Accumulate the three negations. */
+
+      unsigned negate;
+
+      negate = reg->SrcRegister.Negate;
+      if( tgsi_util_get_src_register_extnegate( &reg->SrcRegisterExtSwz, component ) ) {
+         negate = !negate;
+      }
+      if( reg->SrcRegisterExtMod.Negate ) {
+         negate = !negate;
+      }
+
+      if( negate ) {
+         sign_mode = TGSI_UTIL_SIGN_TOGGLE;
+      }
+      else {
+         sign_mode = TGSI_UTIL_SIGN_KEEP;
+      }
+   }
+
+   return sign_mode;
+}
+
+void
+tgsi_util_set_full_src_register_sign_mode(
+   struct tgsi_full_src_register *reg,
+   unsigned sign_mode )
+{
+   reg->SrcRegisterExtSwz.NegateX = 0;
+   reg->SrcRegisterExtSwz.NegateY = 0;
+   reg->SrcRegisterExtSwz.NegateZ = 0;
+   reg->SrcRegisterExtSwz.NegateW = 0;
+
+   switch (sign_mode)
+   {
+   case TGSI_UTIL_SIGN_CLEAR:
+      reg->SrcRegister.Negate = 0;
+      reg->SrcRegisterExtMod.Absolute = 1;
+      reg->SrcRegisterExtMod.Negate = 0;
+      break;
+
+   case TGSI_UTIL_SIGN_SET:
+      reg->SrcRegister.Negate = 0;
+      reg->SrcRegisterExtMod.Absolute = 1;
+      reg->SrcRegisterExtMod.Negate = 1;
+      break;
+
+   case TGSI_UTIL_SIGN_TOGGLE:
+      reg->SrcRegister.Negate = 1;
+      reg->SrcRegisterExtMod.Absolute = 0;
+      reg->SrcRegisterExtMod.Negate = 0;
+      break;
+
+   case TGSI_UTIL_SIGN_KEEP:
+      reg->SrcRegister.Negate = 0;
+      reg->SrcRegisterExtMod.Absolute = 0;
+      reg->SrcRegisterExtMod.Negate = 0;
+      break;
+
+   default:
+      assert( 0 );
+   }
+}
+
diff --git a/src/gallium/aux/tgsi/util/tgsi_util.h b/src/gallium/aux/tgsi/util/tgsi_util.h
new file mode 100644
index 0000000000..ef14446f0e
--- /dev/null
+++ b/src/gallium/aux/tgsi/util/tgsi_util.h
@@ -0,0 +1,70 @@
+#if !defined TGSI_UTIL_H
+#define TGSI_UTIL_H
+
+#if defined __cplusplus
+extern "C" {
+#endif // defined __cplusplus
+
+void *
+tgsi_align_128bit(
+   void *unaligned );
+
+unsigned
+tgsi_util_get_src_register_swizzle(
+   const struct tgsi_src_register *reg,
+   unsigned component );
+
+unsigned
+tgsi_util_get_src_register_extswizzle(
+   const struct tgsi_src_register_ext_swz *reg,
+   unsigned component);
+
+unsigned
+tgsi_util_get_full_src_register_extswizzle(
+   const struct tgsi_full_src_register *reg,
+   unsigned component );
+
+void
+tgsi_util_set_src_register_swizzle(
+   struct tgsi_src_register *reg,
+   unsigned swizzle,
+   unsigned component );
+
+void
+tgsi_util_set_src_register_extswizzle(
+   struct tgsi_src_register_ext_swz *reg,
+   unsigned swizzle,
+   unsigned component );
+
+unsigned
+tgsi_util_get_src_register_extnegate(
+   const struct tgsi_src_register_ext_swz *reg,
+   unsigned component );
+
+void
+tgsi_util_set_src_register_extnegate(
+   struct tgsi_src_register_ext_swz *reg,
+   unsigned negate,
+   unsigned component );
+
+#define TGSI_UTIL_SIGN_CLEAR    0   /* Force positive */
+#define TGSI_UTIL_SIGN_SET      1   /* Force negative */
+#define TGSI_UTIL_SIGN_TOGGLE   2   /* Negate */
+#define TGSI_UTIL_SIGN_KEEP     3   /* No change */
+
+unsigned
+tgsi_util_get_full_src_register_sign_mode(
+   const struct tgsi_full_src_register *reg,
+   unsigned component );
+
+void
+tgsi_util_set_full_src_register_sign_mode(
+   struct tgsi_full_src_register *reg,
+   unsigned sign_mode );
+
+#if defined __cplusplus
+} // extern "C"
+#endif // defined __cplusplus
+
+#endif // !defined TGSI_UTIL_H
+
diff --git a/src/gallium/aux/util/p_debug.c b/src/gallium/aux/util/p_debug.c
new file mode 100644
index 0000000000..b9607a6ba7
--- /dev/null
+++ b/src/gallium/aux/util/p_debug.c
@@ -0,0 +1,76 @@
+/**************************************************************************
+ * 
+ * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+
+#include <stdarg.h>
+
+#ifdef WIN32
+#include <windows.h>
+#include <winddi.h>
+#else
+#include <stdio.h>
+#include <stdlib.h>
+#endif
+
+#include "pipe/p_debug.h" 
+#include "pipe/p_compiler.h" 
+
+
+void debug_vprintf(const char *format, va_list ap)
+{
+#ifdef WIN32
+   EngDebugPrint("Gallium3D: ", (PCHAR)format, ap);
+#else
+   vfprintf(stderr, format, ap);
+#endif
+}
+
+
+void debug_printf(const char *format, ...)
+{
+   va_list ap;
+   va_start(ap, format);
+   debug_vprintf(format, ap);
+   va_end(ap);
+}
+
+
+static INLINE void debug_abort(void) 
+{
+#ifdef WIN32
+   EngDebugBreak();
+#else
+   abort();
+#endif
+}
+
+
+void debug_assert_fail(const char *expr, const char *file, unsigned line) 
+{
+   debug_printf("%s:%i: Assertion `%s' failed.\n", file, line, expr);
+   debug_abort();
+}
diff --git a/src/gallium/aux/util/p_tile.c b/src/gallium/aux/util/p_tile.c
new file mode 100644
index 0000000000..3f795a3898
--- /dev/null
+++ b/src/gallium/aux/util/p_tile.c
@@ -0,0 +1,699 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+/**
+ * RGBA/float tile get/put functions.
+ * Usable both by drivers and state trackers.
+ * Surfaces should already be in a mapped state.
+ */
+
+
+#include "pipe/p_defines.h"
+#include "pipe/p_util.h"
+#include "pipe/p_inlines.h"
+
+#include "p_tile.h"
+
+
+
+/**
+ * Move raw block of pixels from surface to user memory.
+ * This should be usable by any hw driver that has mappable surfaces.
+ */
+void
+pipe_get_tile_raw(struct pipe_context *pipe,
+                  struct pipe_surface *ps,
+                  uint x, uint y, uint w, uint h,
+                  void *p, int dst_stride)
+{
+   const uint cpp = ps->cpp;
+   const ubyte *pSrc;
+   const uint src_stride = ps->pitch * cpp;
+   ubyte *pDest;
+   uint i;
+
+   if (dst_stride == 0) {
+      dst_stride = w * cpp;
+   }
+
+   if (pipe_clip_tile(x, y, &w, &h, ps))
+      return;
+
+   pSrc = (const ubyte *) pipe_surface_map(ps) + (y * ps->pitch + x) * cpp;
+   pDest = (ubyte *) p;
+
+   for (i = 0; i < h; i++) {
+      memcpy(pDest, pSrc, w * cpp);
+      pDest += dst_stride;
+      pSrc += src_stride;
+   }
+
+   pipe_surface_unmap(ps);
+}
+
+
+/**
+ * Move raw block of pixels from user memory to surface.
+ * This should be usable by any hw driver that has mappable surfaces.
+ */
+void
+pipe_put_tile_raw(struct pipe_context *pipe,
+                  struct pipe_surface *ps,
+                  uint x, uint y, uint w, uint h,
+                  const void *p, int src_stride)
+{
+   const uint cpp = ps->cpp;
+   const ubyte *pSrc;
+   const uint dst_stride = ps->pitch * cpp;
+   ubyte *pDest;
+   uint i;
+
+   if (src_stride == 0) {
+      src_stride = w * cpp;
+   }
+
+   if (pipe_clip_tile(x, y, &w, &h, ps))
+      return;
+
+   pSrc = (const ubyte *) p;
+   pDest = (ubyte *) pipe_surface_map(ps) + (y * ps->pitch + x) * cpp;
+
+   for (i = 0; i < h; i++) {
+      memcpy(pDest, pSrc, w * cpp);
+      pDest += dst_stride;
+      pSrc += src_stride;
+   }
+
+   pipe_surface_unmap(ps);
+}
+
+
+
+
+/** Convert short in [-32768,32767] to GLfloat in [-1.0,1.0] */
+#define SHORT_TO_FLOAT(S)   ((2.0F * (S) + 1.0F) * (1.0F/65535.0F))
+
+#define UNCLAMPED_FLOAT_TO_SHORT(us, f)  \
+   us = ( (short) ( CLAMP((f), -1.0, 1.0) * 32767.0F) )
+
+
+
+/*** PIPE_FORMAT_A8R8G8B8_UNORM ***/
+
+static void
+a8r8g8b8_get_tile_rgba(unsigned *src,
+                       unsigned w, unsigned h,
+                       float *p,
+                       unsigned dst_stride)
+{
+   unsigned i, j;
+
+   for (i = 0; i < h; i++) {
+      float *pRow = p;
+      for (j = 0; j < w; j++, pRow += 4) {
+         const unsigned pixel = *src++;
+         pRow[0] = UBYTE_TO_FLOAT((pixel >> 16) & 0xff);
+         pRow[1] = UBYTE_TO_FLOAT((pixel >>  8) & 0xff);
+         pRow[2] = UBYTE_TO_FLOAT((pixel >>  0) & 0xff);
+         pRow[3] = UBYTE_TO_FLOAT((pixel >> 24) & 0xff);
+      }
+      p += dst_stride;
+   }
+}
+
+
+static void
+a8r8g8b8_put_tile_rgba(unsigned *dst,
+                       unsigned w, unsigned h,
+                       const float *p,
+                       unsigned src_stride)
+{
+   unsigned i, j;
+
+   for (i = 0; i < h; i++) {
+      const float *pRow = p;
+      for (j = 0; j < w; j++, pRow += 4) {
+         unsigned r, g, b, a;
+         UNCLAMPED_FLOAT_TO_UBYTE(r, pRow[0]);
+         UNCLAMPED_FLOAT_TO_UBYTE(g, pRow[1]);
+         UNCLAMPED_FLOAT_TO_UBYTE(b, pRow[2]);
+         UNCLAMPED_FLOAT_TO_UBYTE(a, pRow[3]);
+         *dst++ = (a << 24) | (r << 16) | (g << 8) | b;
+      }
+      p += src_stride;
+   }
+}
+
+
+/*** PIPE_FORMAT_B8G8R8A8_UNORM ***/
+
+static void
+b8g8r8a8_get_tile_rgba(unsigned *src,
+                       unsigned w, unsigned h,
+                       float *p,
+                       unsigned dst_stride)
+{
+   unsigned i, j;
+
+   for (i = 0; i < h; i++) {
+      float *pRow = p;
+      for (j = 0; j < w; j++, pRow += 4) {
+         const unsigned pixel = *src++;
+         pRow[0] = UBYTE_TO_FLOAT((pixel >>  8) & 0xff);
+         pRow[1] = UBYTE_TO_FLOAT((pixel >> 16) & 0xff);
+         pRow[2] = UBYTE_TO_FLOAT((pixel >> 24) & 0xff);
+         pRow[3] = UBYTE_TO_FLOAT((pixel >>  0) & 0xff);
+      }
+      p += dst_stride;
+   }
+}
+
+
+static void
+b8g8r8a8_put_tile_rgba(unsigned *dst,
+                       unsigned w, unsigned h,
+                       const float *p,
+                       unsigned src_stride)
+{
+   unsigned i, j;
+
+   for (i = 0; i < h; i++) {
+      const float *pRow = p;
+      for (j = 0; j < w; j++, pRow += 4) {
+         unsigned r, g, b, a;
+         UNCLAMPED_FLOAT_TO_UBYTE(r, pRow[0]);
+         UNCLAMPED_FLOAT_TO_UBYTE(g, pRow[1]);
+         UNCLAMPED_FLOAT_TO_UBYTE(b, pRow[2]);
+         UNCLAMPED_FLOAT_TO_UBYTE(a, pRow[3]);
+         *dst++ = (b << 24) | (g << 16) | (r << 8) | a;
+      }
+      p += src_stride;
+   }
+}
+
+
+/*** PIPE_FORMAT_A1R5G5B5_UNORM ***/
+
+static void
+a1r5g5b5_get_tile_rgba(ushort *src,
+                       unsigned w, unsigned h,
+                       float *p,
+                       unsigned dst_stride)
+{
+   unsigned i, j;
+
+   for (i = 0; i < h; i++) {
+      float *pRow = p;
+      for (j = 0; j < w; j++, pRow += 4) {
+         const ushort pixel = *src++;
+         pRow[0] = ((pixel >> 10) & 0x1f) * (1.0f / 31.0f);
+         pRow[1] = ((pixel >>  5) & 0x1f) * (1.0f / 31.0f);
+         pRow[2] = ((pixel      ) & 0x1f) * (1.0f / 31.0f);
+         pRow[3] = ((pixel >> 15)       ) * 1.0f;
+      }
+      p += dst_stride;
+   }
+}
+
+
+/*** PIPE_FORMAT_A4R4G4B4_UNORM ***/
+
+static void
+a4r4g4b4_get_tile_rgba(ushort *src,
+                       unsigned w, unsigned h,
+                       float *p,
+                       unsigned dst_stride)
+{
+   unsigned i, j;
+
+   for (i = 0; i < h; i++) {
+      float *pRow = p;
+      for (j = 0; j < w; j++, pRow += 4) {
+         const ushort pixel = *src++;
+         pRow[0] = ((pixel >>  8) & 0xf) * (1.0f / 15.0f);
+         pRow[1] = ((pixel >>  4) & 0xf) * (1.0f / 15.0f);
+         pRow[2] = ((pixel      ) & 0xf) * (1.0f / 15.0f);
+         pRow[3] = ((pixel >> 12)      ) * (1.0f / 15.0f);
+      }
+      p += dst_stride;
+   }
+}
+
+
+/*** PIPE_FORMAT_R5G6B5_UNORM ***/
+
+static void
+r5g6b5_get_tile_rgba(ushort *src,
+                     unsigned w, unsigned h,
+                     float *p,
+                     unsigned dst_stride)
+{
+   unsigned i, j;
+
+   for (i = 0; i < h; i++) {
+      float *pRow = p;
+      for (j = 0; j < w; j++, pRow += 4) {
+         const ushort pixel = *src++;
+         pRow[0] = ((pixel >> 11) & 0x1f) * (1.0f / 31.0f);
+         pRow[1] = ((pixel >>  5) & 0x3f) * (1.0f / 63.0f);
+         pRow[2] = ((pixel      ) & 0x1f) * (1.0f / 31.0f);
+         pRow[3] = 1.0f;
+      }
+      p += dst_stride;
+   }
+}
+
+
+static void
+r5g5b5_put_tile_rgba(ushort *dst,
+                     unsigned w, unsigned h,
+                     const float *p,
+                     unsigned src_stride)
+{
+   unsigned i, j;
+
+   for (i = 0; i < h; i++) {
+      const float *pRow = p;
+      for (j = 0; j < w; j++, pRow += 4) {
+         uint r = (uint) (CLAMP(pRow[0], 0.0, 1.0) * 31.0);
+         uint g = (uint) (CLAMP(pRow[1], 0.0, 1.0) * 63.0);
+         uint b = (uint) (CLAMP(pRow[2], 0.0, 1.0) * 31.0);
+         *dst++ = (r << 11) | (g << 5) | (b);
+      }
+      p += src_stride;
+   }
+}
+
+
+
+/*** PIPE_FORMAT_Z16_UNORM ***/
+
+/**
+ * Return each Z value as four floats in [0,1].
+ */
+static void
+z16_get_tile_rgba(ushort *src,
+                  unsigned w, unsigned h,
+                  float *p,
+                  unsigned dst_stride)
+{
+   const float scale = 1.0f / 65535.0f;
+   unsigned i, j;
+
+   for (i = 0; i < h; i++) {
+      float *pRow = p;
+      for (j = 0; j < w; j++, pRow += 4) {
+         pRow[0] =
+         pRow[1] =
+         pRow[2] =
+         pRow[3] = *src++ * scale;
+      }
+      p += dst_stride;
+   }
+}
+
+
+
+
+/*** PIPE_FORMAT_U_L8 ***/
+
+static void
+l8_get_tile_rgba(ubyte *src,
+                 unsigned w, unsigned h,
+                 float *p,
+                 unsigned dst_stride)
+{
+   unsigned i, j;
+
+   for (i = 0; i < h; i++) {
+      float *pRow = p;
+      for (j = 0; j < w; j++, src++, pRow += 4) {
+         pRow[0] =
+         pRow[1] =
+         pRow[2] = UBYTE_TO_FLOAT(*src);
+         pRow[3] = 1.0;
+      }
+      p += dst_stride;
+   }
+}
+
+
+/*** PIPE_FORMAT_U_A8 ***/
+
+static void
+a8_get_tile_rgba(ubyte *src,
+                 unsigned w, unsigned h,
+                 float *p,
+                 unsigned dst_stride)
+{
+   unsigned i, j;
+
+   for (i = 0; i < h; i++) {
+      float *pRow = p;
+      for (j = 0; j < w; j++, src++, pRow += 4) {
+         pRow[0] =
+         pRow[1] =
+         pRow[2] = 0.0;
+         pRow[3] = UBYTE_TO_FLOAT(*src);
+      }
+      p += dst_stride;
+   }
+}
+
+
+/*** PIPE_FORMAT_R16G16B16A16_SNORM ***/
+
+static void
+r16g16b16a16_get_tile_rgba(short *src,
+                           unsigned w, unsigned h,
+                           float *p,
+                           unsigned dst_stride)
+{
+   unsigned i, j;
+
+   for (i = 0; i < h; i++) {
+      float *pRow = p;
+      for (j = 0; j < w; j++, src += 4, pRow += 4) {
+         pRow[0] = SHORT_TO_FLOAT(src[0]);
+         pRow[1] = SHORT_TO_FLOAT(src[1]);
+         pRow[2] = SHORT_TO_FLOAT(src[2]);
+         pRow[3] = SHORT_TO_FLOAT(src[3]);
+      }
+      p += dst_stride;
+   }
+}
+
+
+static void
+r16g16b16a16_put_tile_rgba(short *dst,
+                           unsigned w, unsigned h,
+                           const float *p,
+                           unsigned src_stride)
+{
+   unsigned i, j;
+
+   for (i = 0; i < h; i++) {
+      const float *pRow = p;
+      for (j = 0; j < w; j++, dst += 4, pRow += 4) {
+         UNCLAMPED_FLOAT_TO_SHORT(dst[0], pRow[0]);
+         UNCLAMPED_FLOAT_TO_SHORT(dst[1], pRow[1]);
+         UNCLAMPED_FLOAT_TO_SHORT(dst[2], pRow[2]);
+         UNCLAMPED_FLOAT_TO_SHORT(dst[3], pRow[3]);
+      }
+      p += src_stride;
+   }
+}
+
+
+
+/*** PIPE_FORMAT_U_I8 ***/
+
+static void
+i8_get_tile_rgba(ubyte *src,
+                 unsigned w, unsigned h,
+                 float *p,
+                 unsigned dst_stride)
+{
+   unsigned i, j;
+
+   for (i = 0; i < h; i++) {
+      float *pRow = p;
+      for (j = 0; j < w; j++, src++, pRow += 4) {
+         pRow[0] =
+         pRow[1] =
+         pRow[2] =
+         pRow[3] = UBYTE_TO_FLOAT(*src);
+      }
+      p += dst_stride;
+   }
+}
+
+
+/*** PIPE_FORMAT_U_A8_L8 ***/
+
+static void
+a8_l8_get_tile_rgba(ushort *src,
+                    unsigned w, unsigned h,
+                    float *p,
+                    unsigned dst_stride)
+{
+   unsigned i, j;
+
+   for (i = 0; i < h; i++) {
+      float *pRow = p;
+      for (j = 0; j < w; j++, pRow += 4) {
+         ushort p = *src++;
+         pRow[0] =
+         pRow[1] =
+         pRow[2] = UBYTE_TO_FLOAT(p & 0xff);
+         pRow[3] = UBYTE_TO_FLOAT(p >> 8);
+      }
+      p += dst_stride;
+   }
+}
+
+
+
+
+/*** PIPE_FORMAT_Z32_UNORM ***/
+
+/**
+ * Return each Z value as four floats in [0,1].
+ */
+static void
+z32_get_tile_rgba(unsigned *src,
+                  unsigned w, unsigned h,
+                  float *p,
+                  unsigned dst_stride)
+{
+   const double scale = 1.0 / (double) 0xffffffff;
+   unsigned i, j;
+
+   for (i = 0; i < h; i++) {
+      float *pRow = p;
+      for (j = 0; j < w; j++, pRow += 4) {
+         pRow[0] =
+         pRow[1] =
+         pRow[2] =
+         pRow[3] = (float) (*src++ * scale);
+      }
+      p += dst_stride;
+   }
+}
+
+
+/*** PIPE_FORMAT_S8Z24_UNORM ***/
+
+/**
+ * Return Z component as four float in [0,1].  Stencil part ignored.
+ */
+static void
+s8z24_get_tile_rgba(unsigned *src,
+                    unsigned w, unsigned h,
+                    float *p,
+                    unsigned dst_stride)
+{
+   const double scale = 1.0 / ((1 << 24) - 1);
+   unsigned i, j;
+
+   for (i = 0; i < h; i++) {
+      float *pRow = p;
+      for (j = 0; j < w; j++, pRow += 4) {
+         pRow[0] =
+         pRow[1] =
+         pRow[2] =
+         pRow[3] = (float) (scale * (*src++ & 0xffffff));
+      }
+      p += dst_stride;
+   }
+}
+
+
+/*** PIPE_FORMAT_Z24S8_UNORM ***/
+
+/**
+ * Return Z component as four float in [0,1].  Stencil part ignored.
+ */
+static void
+z24s8_get_tile_rgba(unsigned *src,
+                    unsigned w, unsigned h,
+                    float *p,
+                    unsigned dst_stride)
+{
+   const double scale = 1.0 / ((1 << 24) - 1);
+   unsigned i, j;
+
+   for (i = 0; i < h; i++) {
+      float *pRow = p;
+      for (j = 0; j < w; j++, pRow += 4) {
+         pRow[0] =
+         pRow[1] =
+         pRow[2] =
+         pRow[3] = (float) (scale * (*src++ >> 8));
+      }
+      p += dst_stride;
+   }
+}
+
+
+void
+pipe_get_tile_rgba(struct pipe_context *pipe,
+                   struct pipe_surface *ps,
+                   uint x, uint y, uint w, uint h,
+                   float *p)
+{
+   unsigned dst_stride = w * 4;
+   void *packed;
+
+   if (pipe_clip_tile(x, y, &w, &h, ps))
+      return;
+
+   packed = MALLOC(h * w * ps->cpp);
+
+   if (!packed)
+      return;
+
+   pipe_get_tile_raw(pipe, ps, x, y, w, h, packed, w * ps->cpp);
+
+   switch (ps->format) {
+   case PIPE_FORMAT_A8R8G8B8_UNORM:
+      a8r8g8b8_get_tile_rgba((unsigned *) packed, w, h, p, dst_stride);
+      break;
+   case PIPE_FORMAT_B8G8R8A8_UNORM:
+      b8g8r8a8_get_tile_rgba((unsigned *) packed, w, h, p, dst_stride);
+      break;
+   case PIPE_FORMAT_A1R5G5B5_UNORM:
+      a1r5g5b5_get_tile_rgba((ushort *) packed, w, h, p, dst_stride);
+      break;
+   case PIPE_FORMAT_A4R4G4B4_UNORM:
+      a4r4g4b4_get_tile_rgba((ushort *) packed, w, h, p, dst_stride);
+      break;
+   case PIPE_FORMAT_R5G6B5_UNORM:
+      r5g6b5_get_tile_rgba((ushort *) packed, w, h, p, dst_stride);
+      break;
+   case PIPE_FORMAT_U_L8:
+      l8_get_tile_rgba((ubyte *) packed, w, h, p, dst_stride);
+      break;
+   case PIPE_FORMAT_U_A8:
+      a8_get_tile_rgba((ubyte *) packed, w, h, p, dst_stride);
+      break;
+   case PIPE_FORMAT_U_I8:
+      i8_get_tile_rgba((ubyte *) packed, w, h, p, dst_stride);
+      break;
+   case PIPE_FORMAT_U_A8_L8:
+      a8_l8_get_tile_rgba((ushort *) packed, w, h, p, dst_stride);
+      break;
+   case PIPE_FORMAT_R16G16B16A16_SNORM:
+      r16g16b16a16_get_tile_rgba((short *) packed, w, h, p, dst_stride);
+      break;
+   case PIPE_FORMAT_Z16_UNORM:
+      z16_get_tile_rgba((ushort *) packed, w, h, p, dst_stride);
+      break;
+   case PIPE_FORMAT_Z32_UNORM:
+      z32_get_tile_rgba((unsigned *) packed, w, h, p, dst_stride);
+      break;
+   case PIPE_FORMAT_S8Z24_UNORM:
+      s8z24_get_tile_rgba((unsigned *) packed, w, h, p, dst_stride);
+      break;
+   case PIPE_FORMAT_Z24S8_UNORM:
+      z24s8_get_tile_rgba((unsigned *) packed, w, h, p, dst_stride);
+      break;
+   default:
+      assert(0);
+   }
+
+   FREE(packed);
+}
+
+
+void
+pipe_put_tile_rgba(struct pipe_context *pipe,
+                   struct pipe_surface *ps,
+                   uint x, uint y, uint w, uint h,
+                   const float *p)
+{
+   unsigned src_stride = w * 4;
+   void *packed;
+
+   if (pipe_clip_tile(x, y, &w, &h, ps))
+      return;
+
+   packed = MALLOC(h * w * ps->cpp);
+
+   if (!packed)
+      return;
+
+   switch (ps->format) {
+   case PIPE_FORMAT_A8R8G8B8_UNORM:
+      a8r8g8b8_put_tile_rgba((unsigned *) packed, w, h, p, src_stride);
+      break;
+   case PIPE_FORMAT_B8G8R8A8_UNORM:
+      b8g8r8a8_put_tile_rgba((unsigned *) packed, w, h, p, src_stride);
+      break;
+   case PIPE_FORMAT_A1R5G5B5_UNORM:
+      /*a1r5g5b5_put_tile_rgba((ushort *) packed, w, h, p, src_stride);*/
+      break;
+   case PIPE_FORMAT_R5G6B5_UNORM:
+      r5g5b5_put_tile_rgba((ushort *) packed, w, h, p, src_stride);
+      break;
+   case PIPE_FORMAT_R8G8B8A8_UNORM:
+      break;
+   case PIPE_FORMAT_U_L8:
+      /*l8_put_tile_rgba((ubyte *) packed, w, h, p, src_stride);*/
+      break;
+   case PIPE_FORMAT_U_A8:
+      /*a8_put_tile_rgba((ubyte *) packed, w, h, p, src_stride);*/
+      break;
+   case PIPE_FORMAT_U_I8:
+      /*i8_put_tile_rgba((ubyte *) packed, w, h, p, src_stride);*/
+      break;
+   case PIPE_FORMAT_U_A8_L8:
+      /*a8_l8_put_tile_rgba((ushort *) packed, w, h, p, src_stride);*/
+      break;
+   case PIPE_FORMAT_R16G16B16A16_SNORM:
+      r16g16b16a16_put_tile_rgba((short *) packed, w, h, p, src_stride);
+      break;
+   case PIPE_FORMAT_Z16_UNORM:
+      /*z16_put_tile_rgba((ushort *) packed, w, h, p, src_stride);*/
+      break;
+   case PIPE_FORMAT_Z32_UNORM:
+      /*z32_put_tile_rgba((unsigned *) packed, w, h, p, src_stride);*/
+      break;
+   case PIPE_FORMAT_S8Z24_UNORM:
+      /*s8z24_put_tile_rgba((unsigned *) packed, w, h, p, src_stride);*/
+      break;
+   case PIPE_FORMAT_Z24S8_UNORM:
+      /*z24s8_put_tile_rgba((unsigned *) packed, w, h, p, src_stride);*/
+      break;
+   default:
+      assert(0);
+   }
+
+   pipe_put_tile_raw(pipe, ps, x, y, w, h, packed, w * ps->cpp);
+
+   FREE(packed);
+}
diff --git a/src/gallium/aux/util/p_tile.h b/src/gallium/aux/util/p_tile.h
new file mode 100644
index 0000000000..cd8124bf11
--- /dev/null
+++ b/src/gallium/aux/util/p_tile.h
@@ -0,0 +1,81 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+#ifndef P_TILE_H
+#define P_TILE_H
+
+#include "pipe/p_compiler.h"
+
+struct pipe_context;
+struct pipe_surface;
+
+
+/**
+ * Clip tile against surface dims.
+ * \return TRUE if tile is totally clipped, FALSE otherwise
+ */
+static INLINE boolean
+pipe_clip_tile(uint x, uint y, uint *w, uint *h, const struct pipe_surface *ps)
+{
+   if (x >= ps->width)
+      return TRUE;
+   if (y >= ps->height)
+      return TRUE;
+   if (x + *w > ps->width)
+      *w = ps->width - x;
+   if (y + *h > ps->height)
+      *h = ps->height - y;
+   return FALSE;
+}
+
+
+extern void
+pipe_get_tile_raw(struct pipe_context *pipe,
+                  struct pipe_surface *ps,
+                  uint x, uint y, uint w, uint h,
+                  void *p, int dst_stride);
+
+extern void
+pipe_put_tile_raw(struct pipe_context *pipe,
+                  struct pipe_surface *ps,
+                  uint x, uint y, uint w, uint h,
+                  const void *p, int src_stride);
+
+
+extern void
+pipe_get_tile_rgba(struct pipe_context *pipe,
+                   struct pipe_surface *ps,
+                   uint x, uint y, uint w, uint h,
+                   float *p);
+
+extern void
+pipe_put_tile_rgba(struct pipe_context *pipe,
+                   struct pipe_surface *ps,
+                   uint x, uint y, uint w, uint h,
+                   const float *p);
+
+#endif
diff --git a/src/gallium/aux/util/p_util.c b/src/gallium/aux/util/p_util.c
new file mode 100644
index 0000000000..2a92f8e408
--- /dev/null
+++ b/src/gallium/aux/util/p_util.c
@@ -0,0 +1,73 @@
+/**************************************************************************
+ * 
+ * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+/**
+ * Miscellaneous utility functions.
+ */
+
+
+#include "pipe/p_defines.h"
+#include "pipe/p_util.h"
+
+
+/**
+ * Copy 2D rect from one place to another.
+ * Position and sizes are in pixels.
+ */
+void
+pipe_copy_rect(ubyte * dst,
+               unsigned cpp,
+               unsigned dst_pitch,
+               unsigned dst_x,
+               unsigned dst_y,
+               unsigned width,
+               unsigned height,
+               const ubyte * src,
+               int src_pitch,
+               unsigned src_x, 
+               int src_y)
+{
+   unsigned i;
+
+   dst_pitch *= cpp;
+   src_pitch *= cpp;
+   dst += dst_x * cpp;
+   src += src_x * cpp;
+   dst += dst_y * dst_pitch;
+   src += src_y * src_pitch;
+   width *= cpp;
+
+   if (width == dst_pitch && width == src_pitch)
+      memcpy(dst, src, height * width);
+   else {
+      for (i = 0; i < height; i++) {
+         memcpy(dst, src, width);
+         dst += dst_pitch;
+         src += src_pitch;
+      }
+   }
+}
diff --git a/src/gallium/drivers/cell/Makefile b/src/gallium/drivers/cell/Makefile
new file mode 100644
index 0000000000..47aef7b05f
--- /dev/null
+++ b/src/gallium/drivers/cell/Makefile
@@ -0,0 +1,12 @@
+# Cell Gallium driver Makefile
+
+
+default:
+	( cd spu ; make )
+	( cd ppu ; make )
+
+
+
+clean:
+	( cd spu ; make clean )
+	( cd ppu ; make clean )
diff --git a/src/gallium/drivers/cell/common.h b/src/gallium/drivers/cell/common.h
new file mode 100644
index 0000000000..4de514c358
--- /dev/null
+++ b/src/gallium/drivers/cell/common.h
@@ -0,0 +1,220 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+/**
+ * Types and tokens which are common to the SPU and PPU code.
+ */
+
+
+#ifndef CELL_COMMON_H
+#define CELL_COMMON_H
+
+#include "pipe/p_compiler.h"
+#include "pipe/p_util.h"
+#include "pipe/p_format.h"
+
+
+/** The standard assert macro doesn't seem to work reliably */
+#define ASSERT(x) \
+   if (!(x)) { \
+      ubyte *p = NULL; \
+      fprintf(stderr, "%s:%d: %s(): assertion %s failed.\n", \
+              __FILE__, __LINE__, __FUNCTION__, #x);             \
+      *p = 0; \
+      exit(1); \
+   }
+
+
+/** for sanity checking */
+#define ASSERT_ALIGN16(ptr) \
+  ASSERT((((unsigned long) (ptr)) & 0xf) == 0);
+
+
+/** round up value to next multiple of 4 */
+#define ROUNDUP4(k)  (((k) + 0x3) & ~0x3)
+
+/** round up value to next multiple of 8 */
+#define ROUNDUP8(k)  (((k) + 0x7) & ~0x7)
+
+/** round up value to next multiple of 16 */
+#define ROUNDUP16(k)  (((k) + 0xf) & ~0xf)
+
+
+#define CELL_MAX_SPUS 6
+
+#define TILE_SIZE 32
+
+
+/**
+ * The low byte of a mailbox word contains the command opcode.
+ * Remaining higher bytes are command specific.
+ */
+#define CELL_CMD_OPCODE_MASK 0xff
+
+#define CELL_CMD_EXIT                 1
+#define CELL_CMD_CLEAR_SURFACE        2
+#define CELL_CMD_FINISH               3
+#define CELL_CMD_RENDER               4
+#define CELL_CMD_BATCH                5
+#define CELL_CMD_RELEASE_VERTS        6
+#define CELL_CMD_STATE_FRAMEBUFFER   10
+#define CELL_CMD_STATE_DEPTH_STENCIL 11
+#define CELL_CMD_STATE_SAMPLER       12
+#define CELL_CMD_STATE_TEXTURE       13
+#define CELL_CMD_STATE_VERTEX_INFO   14
+#define CELL_CMD_STATE_VIEWPORT      15
+#define CELL_CMD_STATE_VS_ARRAY_INFO 16
+#define CELL_CMD_STATE_BLEND         17
+#define CELL_CMD_VS_EXECUTE          18
+
+
+#define CELL_NUM_BUFFERS 4
+#define CELL_BUFFER_SIZE (4*1024)  /**< 16KB would be the max */
+
+#define CELL_BUFFER_STATUS_FREE 10
+#define CELL_BUFFER_STATUS_USED 20
+
+
+
+/**
+ * Tell SPUs about the framebuffer size, location
+ */
+struct cell_command_framebuffer
+{
+   uint64_t opcode;     /**< CELL_CMD_FRAMEBUFFER */
+   int width, height;
+   void *color_start, *depth_start;
+   enum pipe_format color_format, depth_format;
+};
+
+
+/**
+ * Clear framebuffer to the given value/color.
+ */
+struct cell_command_clear_surface
+{
+   uint64_t opcode;     /**< CELL_CMD_CLEAR_SURFACE */
+   uint surface; /**< Temporary: 0=color, 1=Z */
+   uint value;
+};
+
+
+/**
+ * Array info used by the vertex shader's vertex puller.
+ */
+struct cell_array_info
+{
+    uint64_t base;      /**< Base address of the 0th element. */
+    uint attr;          /**< Attribute that this state is for. */
+    uint pitch;         /**< Byte pitch from one entry to the next. */
+    uint format;        /**< Pipe format of each entry. */
+} ALIGN16_ATTRIB;
+
+
+struct cell_shader_info
+{
+   unsigned num_outputs;
+
+   uint64_t declarations;
+   unsigned num_declarations;
+   uint64_t instructions;
+   unsigned num_instructions;
+   uint64_t uniforms;
+   uint64_t  immediates;
+   unsigned num_immediates;
+} ALIGN16_ATTRIB;
+
+
+#define SPU_VERTS_PER_BATCH 64
+struct cell_command_vs
+{
+   uint64_t opcode;       /**< CELL_CMD_VS_EXECUTE */
+   struct cell_shader_info   shader;
+   unsigned num_elts;
+   unsigned elts[SPU_VERTS_PER_BATCH];
+   uint64_t vOut[SPU_VERTS_PER_BATCH];
+   float plane[12][4];
+   unsigned nr_planes;
+   unsigned nr_attrs;
+} ALIGN16_ATTRIB;
+
+
+struct cell_command_render
+{
+   uint64_t opcode;   /**< CELL_CMD_RENDER */
+   uint prim_type;    /**< PIPE_PRIM_x */
+   uint num_verts;
+   uint vertex_size;  /**< bytes per vertex */
+   uint num_indexes;
+   uint vertex_buf;  /**< which cell->buffer[] contains the vertex data */
+   float xmin, ymin, xmax, ymax;  /* XXX another dummy field */
+   uint min_index;
+   boolean inline_verts;
+};
+
+
+struct cell_command_release_verts
+{
+   uint64_t opcode;         /**< CELL_CMD_RELEASE_VERTS */
+   uint vertex_buf;    /**< in [0, CELL_NUM_BUFFERS-1] */
+};
+
+
+struct cell_command_texture
+{
+   void *start;         /**< Address in main memory */
+   uint width, height;
+};
+
+
+/** XXX unions don't seem to work */
+/* XXX this should go away; all commands should be placed in batch buffers */
+struct cell_command
+{
+#if 0
+   struct cell_command_framebuffer fb;
+   struct cell_command_clear_surface clear;
+   struct cell_command_render render;
+#endif
+   struct cell_command_vs vs;
+} ALIGN16_ATTRIB;
+
+
+/** This is the object passed to spe_create_thread() */
+struct cell_init_info
+{
+   unsigned id;
+   unsigned num_spus;
+   struct cell_command *cmd;
+
+   /** Buffers for command batches, vertex/index data */
+   ubyte *buffers[CELL_NUM_BUFFERS];
+   uint *buffer_status;  /**< points at cell_context->buffer_status */
+} ALIGN16_ATTRIB;
+
+
+#endif /* CELL_COMMON_H */
diff --git a/src/gallium/drivers/cell/ppu/Makefile b/src/gallium/drivers/cell/ppu/Makefile
new file mode 100644
index 0000000000..50060f5cd3
--- /dev/null
+++ b/src/gallium/drivers/cell/ppu/Makefile
@@ -0,0 +1,76 @@
+# Gallium3D Cell driver: PPU code
+
+# This makefile builds the g3dcell.a library which gets pulled into
+# the main libGL.so library
+
+
+TOP = ../../../../..
+include $(TOP)/configs/linux-cell
+
+
+#PROG = gl4
+
+CELL_LIB = libcell.a
+
+SPU_CODE_MODULE = ../spu/g3d_spu.a
+
+
+SOURCES = \
+	cell_batch.c \
+	cell_clear.c \
+	cell_context.c \
+	cell_draw_arrays.c \
+	cell_flush.c \
+	cell_state_blend.c \
+	cell_state_clip.c \
+	cell_state_derived.c \
+	cell_state_emit.c \
+	cell_state_fs.c \
+	cell_state_rasterizer.c \
+	cell_state_sampler.c \
+	cell_state_surface.c \
+	cell_state_vertex.c \
+	cell_spu.c \
+	cell_surface.c \
+	cell_texture.c \
+	cell_vbuf.c \
+	cell_vertex_shader.c \
+	cell_winsys.c
+
+
+OBJECTS = $(SOURCES:.c=.o) \
+
+INCLUDE_DIRS = -I$(TOP)/src/mesa
+
+
+.c.o:
+	$(CC) -c $(INCLUDE_DIRS) $(CFLAGS) $< -o $@
+
+
+
+default: $(CELL_LIB)
+
+
+$(CELL_LIB): $(OBJECTS) $(SPU_CODE_MODULE)
+#	ar -ru $(CELL_LIB) $(OBJECTS) $(SPU_CODE_MODULE)
+	ar -ru $(CELL_LIB) $(OBJECTS)
+
+#$(PROG): $(PPU_OBJECTS)
+#	$(CC) -o $(PROG) $(PPU_OBJECTS) $(SPU_CODE_MODULE) $(PPU_LFLAGS)
+
+
+
+clean:
+	rm -f *.o *~ $(CELL_LIB)
+
+
+
+depend: $(SOURCES)
+	rm -f depend
+	touch depend
+	$(MKDEP) $(MKDEP_OPTIONS) $(INCLUDE_DIRS) $(SOURCES) 2> /dev/null
+
+include depend
+
+
+
diff --git a/src/gallium/drivers/cell/ppu/cell_batch.c b/src/gallium/drivers/cell/ppu/cell_batch.c
new file mode 100644
index 0000000000..f45e5f25b6
--- /dev/null
+++ b/src/gallium/drivers/cell/ppu/cell_batch.c
@@ -0,0 +1,217 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+
+#include "cell_context.h"
+#include "cell_batch.h"
+#include "cell_spu.h"
+
+
+
+uint
+cell_get_empty_buffer(struct cell_context *cell)
+{
+   uint buf = 0, tries = 0;
+
+   /* Find a buffer that's marked as free by all SPUs */
+   while (1) {
+      uint spu, num_free = 0;
+
+      for (spu = 0; spu < cell->num_spus; spu++) {
+         if (cell->buffer_status[spu][buf][0] == CELL_BUFFER_STATUS_FREE) {
+            num_free++;
+
+            if (num_free == cell->num_spus) {
+               /* found a free buffer, now mark status as used */
+               for (spu = 0; spu < cell->num_spus; spu++) {
+                  cell->buffer_status[spu][buf][0] = CELL_BUFFER_STATUS_USED;
+               }
+               /*
+               printf("PPU: ALLOC BUFFER %u\n", buf);
+               */
+               return buf;
+            }
+         }
+         else {
+            break;
+         }
+      }
+
+      /* try next buf */
+      buf = (buf + 1) % CELL_NUM_BUFFERS;
+
+      tries++;
+      if (tries == 100) {
+         /*
+         printf("PPU WAITING for buffer...\n");
+         */
+      }
+   }
+}
+
+
+void
+cell_batch_flush(struct cell_context *cell)
+{
+   static boolean flushing = FALSE;
+   uint batch = cell->cur_batch;
+   const uint size = cell->buffer_size[batch];
+   uint spu, cmd_word;
+
+   assert(!flushing);
+
+   if (size == 0)
+      return;
+
+   flushing = TRUE;
+
+   assert(batch < CELL_NUM_BUFFERS);
+
+   /*
+   printf("cell_batch_dispatch: buf %u at %p, size %u\n",
+          batch, &cell->batch_buffer[batch][0], size);
+   */
+     
+   /*
+    * Build "BATCH" command and sent to all SPUs.
+    */
+   cmd_word = CELL_CMD_BATCH | (batch << 8) | (size << 16);
+
+   for (spu = 0; spu < cell->num_spus; spu++) {
+      assert(cell->buffer_status[spu][batch][0] == CELL_BUFFER_STATUS_USED);
+      send_mbox_message(cell_global.spe_contexts[spu], cmd_word);
+   }
+
+   /* When the SPUs are done copying the buffer into their locals stores
+    * they'll write a BUFFER_STATUS_FREE message into the buffer_status[]
+    * array indicating that the PPU can re-use the buffer.
+    */
+
+   batch = cell_get_empty_buffer(cell);
+
+   cell->buffer_size[batch] = 0;  /* empty */
+   cell->cur_batch = batch;
+
+   flushing = FALSE;
+}
+
+
+uint
+cell_batch_free_space(const struct cell_context *cell)
+{
+   uint free = CELL_BUFFER_SIZE - cell->buffer_size[cell->cur_batch];
+   return free;
+}
+
+
+/**
+ * Append data to current batch.
+ */
+void
+cell_batch_append(struct cell_context *cell, const void *data, uint bytes)
+{
+   uint size;
+
+   ASSERT(bytes % 8 == 0);
+   ASSERT(bytes <= CELL_BUFFER_SIZE);
+   ASSERT(cell->cur_batch >= 0);
+
+#ifdef ASSERT
+   {
+      uint spu;
+      for (spu = 0; spu < cell->num_spus; spu++) {
+         ASSERT(cell->buffer_status[spu][cell->cur_batch][0]
+                 == CELL_BUFFER_STATUS_USED);
+      }
+   }
+#endif
+
+   size = cell->buffer_size[cell->cur_batch];
+
+   if (size + bytes > CELL_BUFFER_SIZE) {
+      cell_batch_flush(cell);
+      size = 0;
+   }
+
+   ASSERT(size + bytes <= CELL_BUFFER_SIZE);
+
+   memcpy(cell->buffer[cell->cur_batch] + size, data, bytes);
+
+   cell->buffer_size[cell->cur_batch] = size + bytes;
+}
+
+
+void *
+cell_batch_alloc(struct cell_context *cell, uint bytes)
+{
+   return cell_batch_alloc_aligned(cell, bytes, 1);
+}
+
+
+void *
+cell_batch_alloc_aligned(struct cell_context *cell, uint bytes,
+                         uint alignment)
+{
+   void *pos;
+   uint size, padbytes;
+
+   ASSERT(bytes % 8 == 0);
+   ASSERT(bytes <= CELL_BUFFER_SIZE);
+   ASSERT(alignment > 0);
+   ASSERT(cell->cur_batch >= 0);
+
+#ifdef ASSERT
+   {
+      uint spu;
+      for (spu = 0; spu < cell->num_spus; spu++) {
+         ASSERT(cell->buffer_status[spu][cell->cur_batch][0]
+                 == CELL_BUFFER_STATUS_USED);
+      }
+   }
+#endif
+
+   size = cell->buffer_size[cell->cur_batch];
+
+   padbytes = (alignment - (size % alignment)) % alignment;
+
+   if (padbytes + size + bytes > CELL_BUFFER_SIZE) {
+      cell_batch_flush(cell);
+      size = 0;
+   }
+   else {
+      size += padbytes;
+   }
+
+   ASSERT(size % alignment == 0);
+   ASSERT(size + bytes <= CELL_BUFFER_SIZE);
+
+   pos = (void *) (cell->buffer[cell->cur_batch] + size);
+
+   cell->buffer_size[cell->cur_batch] = size + bytes;
+
+   return pos;
+}
diff --git a/src/gallium/drivers/cell/ppu/cell_batch.h b/src/gallium/drivers/cell/ppu/cell_batch.h
new file mode 100644
index 0000000000..a6eee0a8b1
--- /dev/null
+++ b/src/gallium/drivers/cell/ppu/cell_batch.h
@@ -0,0 +1,58 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+
+#ifndef CELL_BATCH_H
+#define CELL_BATCH_H
+
+#include "pipe/p_compiler.h"
+
+
+struct cell_context;
+
+
+extern uint
+cell_get_empty_buffer(struct cell_context *cell);
+
+extern void
+cell_batch_flush(struct cell_context *cell);
+
+extern uint
+cell_batch_free_space(const struct cell_context *cell);
+
+extern void
+cell_batch_append(struct cell_context *cell, const void *data, uint bytes);
+
+extern void *
+cell_batch_alloc(struct cell_context *cell, uint bytes);
+
+extern void *
+cell_batch_alloc_aligned(struct cell_context *cell, uint bytes,
+                         uint alignment);
+
+
+#endif /* CELL_BATCH_H */
diff --git a/src/gallium/drivers/cell/ppu/cell_clear.c b/src/gallium/drivers/cell/ppu/cell_clear.c
new file mode 100644
index 0000000000..07b908eec5
--- /dev/null
+++ b/src/gallium/drivers/cell/ppu/cell_clear.c
@@ -0,0 +1,76 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+/**
+ * Authors
+ *  Brian Paul
+ */
+
+#include <stdio.h>
+#include <assert.h>
+#include <stdint.h>
+#include "pipe/p_inlines.h"
+#include "pipe/p_util.h"
+#include "pipe/cell/common.h"
+#include "cell_clear.h"
+#include "cell_context.h"
+#include "cell_batch.h"
+#include "cell_flush.h"
+#include "cell_spu.h"
+
+
+void
+cell_clear_surface(struct pipe_context *pipe, struct pipe_surface *ps,
+                   unsigned clearValue)
+{
+   struct cell_context *cell = cell_context(pipe);
+   uint surfIndex;
+
+   if (cell->dirty)
+      cell_update_derived(cell);
+
+
+   if (!cell->cbuf_map[0])
+      cell->cbuf_map[0] = pipe_surface_map(ps);
+
+   if (ps == cell->framebuffer.zsbuf) {
+      surfIndex = 1;
+   }
+   else {
+      surfIndex = 0;
+   }
+
+
+   {
+      struct cell_command_clear_surface *clr
+         = (struct cell_command_clear_surface *)
+         cell_batch_alloc(cell, sizeof(*clr));
+      clr->opcode = CELL_CMD_CLEAR_SURFACE;
+      clr->surface = surfIndex;
+      clr->value = clearValue;
+   }
+}
diff --git a/src/gallium/drivers/cell/ppu/cell_clear.h b/src/gallium/drivers/cell/ppu/cell_clear.h
new file mode 100644
index 0000000000..ff47d43f4c
--- /dev/null
+++ b/src/gallium/drivers/cell/ppu/cell_clear.h
@@ -0,0 +1,43 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+
+#ifndef CELL_CLEAR_H
+#define CELL_CLEAR_H
+
+
+struct pipe_context;
+struct pipe_surface;
+
+
+extern void
+cell_clear_surface(struct pipe_context *pipe, struct pipe_surface *ps,
+                   unsigned clearValue);
+
+
+
+#endif /* CELL_CLEAR_H */
diff --git a/src/gallium/drivers/cell/ppu/cell_context.c b/src/gallium/drivers/cell/ppu/cell_context.c
new file mode 100644
index 0000000000..bbe1fd7a11
--- /dev/null
+++ b/src/gallium/drivers/cell/ppu/cell_context.c
@@ -0,0 +1,287 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+/**
+ * Authors
+ *  Brian Paul
+ */
+
+
+#include <stdio.h>
+
+#include "pipe/p_defines.h"
+#include "pipe/p_format.h"
+#include "pipe/p_util.h"
+#include "pipe/p_winsys.h"
+#include "pipe/cell/common.h"
+#include "pipe/draw/draw_context.h"
+#include "pipe/draw/draw_private.h"
+#include "cell_clear.h"
+#include "cell_context.h"
+#include "cell_draw_arrays.h"
+#include "cell_flush.h"
+#include "cell_render.h"
+#include "cell_state.h"
+#include "cell_surface.h"
+#include "cell_spu.h"
+#include "cell_texture.h"
+#include "cell_vbuf.h"
+
+
+
+static boolean
+cell_is_format_supported( struct pipe_context *pipe,
+                          enum pipe_format format, uint type )
+{
+   /*struct cell_context *cell = cell_context( pipe );*/
+
+   switch (type) {
+   case PIPE_TEXTURE:
+      /* cell supports all texture formats, XXX for now anyway */
+      return TRUE;
+   case PIPE_SURFACE:
+      /* cell supports all (off-screen) surface formats, XXX for now */
+      return TRUE;
+   default:
+      assert(0);
+      return FALSE;
+   }
+}
+
+
+static int cell_get_param(struct pipe_context *pipe, int param)
+{
+   switch (param) {
+   case PIPE_CAP_MAX_TEXTURE_IMAGE_UNITS:
+      return 8;
+   case PIPE_CAP_NPOT_TEXTURES:
+      return 1;
+   case PIPE_CAP_TWO_SIDED_STENCIL:
+      return 1;
+   case PIPE_CAP_GLSL:
+      return 1;
+   case PIPE_CAP_S3TC:
+      return 0;
+   case PIPE_CAP_ANISOTROPIC_FILTER:
+      return 0;
+   case PIPE_CAP_POINT_SPRITE:
+      return 1;
+   case PIPE_CAP_MAX_RENDER_TARGETS:
+      return 1;
+   case PIPE_CAP_OCCLUSION_QUERY:
+      return 1;
+   case PIPE_CAP_TEXTURE_SHADOW_MAP:
+      return 1;
+   case PIPE_CAP_MAX_TEXTURE_2D_LEVELS:
+      return 12; /* max 2Kx2K */
+   case PIPE_CAP_MAX_TEXTURE_3D_LEVELS:
+      return 8;  /* max 128x128x128 */
+   case PIPE_CAP_MAX_TEXTURE_CUBE_LEVELS:
+      return 12; /* max 2Kx2K */
+   default:
+      return 0;
+   }
+}
+
+static float cell_get_paramf(struct pipe_context *pipe, int param)
+{
+   switch (param) {
+   case PIPE_CAP_MAX_LINE_WIDTH:
+      /* fall-through */
+   case PIPE_CAP_MAX_LINE_WIDTH_AA:
+      return 255.0; /* arbitrary */
+
+   case PIPE_CAP_MAX_POINT_WIDTH:
+      /* fall-through */
+   case PIPE_CAP_MAX_POINT_WIDTH_AA:
+      return 255.0; /* arbitrary */
+
+   case PIPE_CAP_MAX_TEXTURE_ANISOTROPY:
+      return 0.0;
+
+   case PIPE_CAP_MAX_TEXTURE_LOD_BIAS:
+      return 16.0; /* arbitrary */
+
+   default:
+      return 0;
+   }
+}
+
+
+static const char *
+cell_get_name( struct pipe_context *pipe )
+{
+   return "Cell";
+}
+
+static const char *
+cell_get_vendor( struct pipe_context *pipe )
+{
+   return "Tungsten Graphics, Inc.";
+}
+
+
+
+static void
+cell_destroy_context( struct pipe_context *pipe )
+{
+   struct cell_context *cell = cell_context(pipe);
+
+   cell_spu_exit(cell);
+
+   align_free(cell);
+}
+
+
+static struct draw_context *
+cell_draw_create(struct cell_context *cell)
+{
+   struct draw_context *draw = draw_create();
+
+   if (getenv("GALLIUM_CELL_VS")) {
+      /* plug in SPU-based vertex transformation code */
+      draw->shader_queue_flush = cell_vertex_shader_queue_flush;
+      draw->driver_private = cell;
+   }
+
+   return draw;
+}
+
+
+struct pipe_context *
+cell_create_context(struct pipe_winsys *winsys, struct cell_winsys *cws)
+{
+   struct cell_context *cell;
+   uint spu, buf;
+
+   /* some fields need to be 16-byte aligned, so align the whole object */
+   cell = (struct cell_context*) align_malloc(sizeof(struct cell_context), 16);
+   if (!cell)
+      return NULL;
+
+   memset(cell, 0, sizeof(*cell));
+
+   cell->winsys = cws;
+   cell->pipe.winsys = winsys;
+   cell->pipe.destroy = cell_destroy_context;
+
+   /* queries */
+   cell->pipe.is_format_supported = cell_is_format_supported;
+   cell->pipe.get_name = cell_get_name;
+   cell->pipe.get_vendor = cell_get_vendor;
+   cell->pipe.get_param = cell_get_param;
+   cell->pipe.get_paramf = cell_get_paramf;
+
+
+   /* state setters */
+   cell->pipe.create_blend_state = cell_create_blend_state;
+   cell->pipe.bind_blend_state   = cell_bind_blend_state;
+   cell->pipe.delete_blend_state = cell_delete_blend_state;
+
+   cell->pipe.create_sampler_state = cell_create_sampler_state;
+   cell->pipe.bind_sampler_state   = cell_bind_sampler_state;
+   cell->pipe.delete_sampler_state = cell_delete_sampler_state;
+
+   cell->pipe.create_depth_stencil_alpha_state = cell_create_depth_stencil_alpha_state;
+   cell->pipe.bind_depth_stencil_alpha_state   = cell_bind_depth_stencil_alpha_state;
+   cell->pipe.delete_depth_stencil_alpha_state = cell_delete_depth_stencil_alpha_state;
+
+   cell->pipe.create_rasterizer_state = cell_create_rasterizer_state;
+   cell->pipe.bind_rasterizer_state   = cell_bind_rasterizer_state;
+   cell->pipe.delete_rasterizer_state = cell_delete_rasterizer_state;
+
+   cell->pipe.create_fs_state = cell_create_fs_state;
+   cell->pipe.bind_fs_state   = cell_bind_fs_state;
+   cell->pipe.delete_fs_state = cell_delete_fs_state;
+
+   cell->pipe.create_vs_state = cell_create_vs_state;
+   cell->pipe.bind_vs_state   = cell_bind_vs_state;
+   cell->pipe.delete_vs_state = cell_delete_vs_state;
+
+   cell->pipe.set_blend_color = cell_set_blend_color;
+   cell->pipe.set_clip_state = cell_set_clip_state;
+   cell->pipe.set_constant_buffer = cell_set_constant_buffer;
+
+   cell->pipe.set_framebuffer_state = cell_set_framebuffer_state;
+
+   cell->pipe.set_polygon_stipple = cell_set_polygon_stipple;
+   cell->pipe.set_scissor_state = cell_set_scissor_state;
+   cell->pipe.set_viewport_state = cell_set_viewport_state;
+
+   cell->pipe.set_vertex_buffer = cell_set_vertex_buffer;
+   cell->pipe.set_vertex_element = cell_set_vertex_element;
+
+   cell->pipe.draw_arrays = cell_draw_arrays;
+   cell->pipe.draw_elements = cell_draw_elements;
+
+   cell->pipe.clear = cell_clear_surface;
+   cell->pipe.flush = cell_flush;
+
+   /* textures */
+   cell->pipe.texture_create = cell_texture_create;
+   cell->pipe.texture_release = cell_texture_release;
+   cell->pipe.get_tex_surface = cell_get_tex_surface;
+
+   cell->pipe.set_sampler_texture = cell_set_sampler_texture;
+
+#if 0
+   cell->pipe.begin_query = cell_begin_query;
+   cell->pipe.end_query = cell_end_query;
+   cell->pipe.wait_query = cell_wait_query;
+#endif
+
+   cell_init_surface_functions(cell);
+
+   cell->draw = cell_draw_create(cell);
+
+   cell_init_vbuf(cell);
+   draw_set_rasterize_stage(cell->draw, cell->vbuf);
+
+   /*
+    * SPU stuff
+    */
+   cell->num_spus = 6;
+
+   cell_start_spus(cell);
+
+   /* init command, vertex/index buffer info */
+   for (buf = 0; buf < CELL_NUM_BUFFERS; buf++) {
+      cell->buffer_size[buf] = 0;
+
+      /* init batch buffer status values,
+       * mark 0th buffer as used, rest as free.
+       */
+      for (spu = 0; spu < cell->num_spus; spu++) {
+         if (buf == 0)
+            cell->buffer_status[spu][buf][0] = CELL_BUFFER_STATUS_USED;
+         else
+            cell->buffer_status[spu][buf][0] = CELL_BUFFER_STATUS_FREE;
+      }
+   }
+
+   return &cell->pipe;
+}
diff --git a/src/gallium/drivers/cell/ppu/cell_context.h b/src/gallium/drivers/cell/ppu/cell_context.h
new file mode 100644
index 0000000000..3b63419b5e
--- /dev/null
+++ b/src/gallium/drivers/cell/ppu/cell_context.h
@@ -0,0 +1,135 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+
+#ifndef CELL_CONTEXT_H
+#define CELL_CONTEXT_H
+
+
+#include "pipe/p_context.h"
+#include "pipe/p_defines.h"
+#include "pipe/draw/draw_vertex.h"
+#include "pipe/draw/draw_vbuf.h"
+#include "cell_winsys.h"
+#include "pipe/cell/common.h"
+
+
+struct cell_vbuf_render;
+
+struct cell_vertex_shader_state
+{
+   struct pipe_shader_state shader;
+   void *draw_data;
+};
+
+
+struct cell_fragment_shader_state
+{
+   struct pipe_shader_state shader;
+   void *data;
+};
+
+
+struct cell_context
+{
+   struct pipe_context pipe;
+
+   struct cell_winsys *winsys;
+
+   const struct pipe_blend_state *blend;
+   const struct pipe_sampler_state *sampler[PIPE_MAX_SAMPLERS];
+   const struct pipe_depth_stencil_alpha_state   *depth_stencil;
+   const struct pipe_rasterizer_state *rasterizer;
+   const struct cell_vertex_shader_state *vs;
+   const struct cell_fragment_shader_state *fs;
+
+   struct pipe_blend_color blend_color;
+   struct pipe_clip_state clip;
+   struct pipe_constant_buffer constants[2];
+   struct pipe_framebuffer_state framebuffer;
+   struct pipe_poly_stipple poly_stipple;
+   struct pipe_scissor_state scissor;
+   struct cell_texture *texture[PIPE_MAX_SAMPLERS];
+   struct pipe_viewport_state viewport;
+   struct pipe_vertex_buffer vertex_buffer[PIPE_ATTRIB_MAX];
+   struct pipe_vertex_element vertex_element[PIPE_ATTRIB_MAX];
+
+   ubyte *cbuf_map[PIPE_MAX_COLOR_BUFS];
+   ubyte *zsbuf_map;
+
+   struct pipe_surface *tex_surf;
+   uint *tex_map;
+
+   uint dirty;
+
+   /** The primitive drawing context */
+   struct draw_context *draw;
+   struct draw_stage *render_stage;
+
+   /** For post-transformed vertex buffering: */
+   struct cell_vbuf_render *vbuf_render;
+   struct draw_stage *vbuf;
+
+   struct vertex_info vertex_info;
+
+   /** Mapped constant buffers */
+   void *mapped_constants[PIPE_SHADER_TYPES];
+
+
+   uint num_spus;
+
+   /** Buffers for command batches, vertex/index data */
+   uint buffer_size[CELL_NUM_BUFFERS];
+   ubyte buffer[CELL_NUM_BUFFERS][CELL_BUFFER_SIZE] ALIGN16_ATTRIB;
+
+   int cur_batch;  /**< which buffer is being filled w/ commands */
+
+   /** [4] to ensure 16-byte alignment for each status word */
+   uint buffer_status[CELL_MAX_SPUS][CELL_NUM_BUFFERS][4] ALIGN16_ATTRIB;
+
+};
+
+
+
+
+static INLINE struct cell_context *
+cell_context(struct pipe_context *pipe)
+{
+   return (struct cell_context *) pipe;
+}
+
+
+extern struct pipe_context *
+cell_create_context(struct pipe_winsys *ws, struct cell_winsys *cws);
+
+extern void
+cell_vertex_shader_queue_flush(struct draw_context *draw);
+
+
+
+
+#endif /* CELL_CONTEXT_H */
diff --git a/src/gallium/drivers/cell/ppu/cell_draw_arrays.c b/src/gallium/drivers/cell/ppu/cell_draw_arrays.c
new file mode 100644
index 0000000000..717cd8370f
--- /dev/null
+++ b/src/gallium/drivers/cell/ppu/cell_draw_arrays.c
@@ -0,0 +1,164 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+/* Author:
+ *    Brian Paul
+ *    Keith Whitwell
+ */
+
+
+#include "pipe/p_defines.h"
+#include "pipe/p_context.h"
+#include "pipe/p_winsys.h"
+
+#include "cell_context.h"
+#include "cell_draw_arrays.h"
+#include "cell_state.h"
+
+#include "pipe/draw/draw_context.h"
+
+
+
+static void
+cell_map_constant_buffers(struct cell_context *sp)
+{
+   struct pipe_winsys *ws = sp->pipe.winsys;
+   uint i;
+   for (i = 0; i < 2; i++) {
+      if (sp->constants[i].size)
+         sp->mapped_constants[i] = ws->buffer_map(ws, sp->constants[i].buffer,
+                                                  PIPE_BUFFER_USAGE_CPU_READ);
+   }
+
+   draw_set_mapped_constant_buffer(sp->draw,
+                                   sp->mapped_constants[PIPE_SHADER_VERTEX]);
+}
+
+static void
+cell_unmap_constant_buffers(struct cell_context *sp)
+{
+   struct pipe_winsys *ws = sp->pipe.winsys;
+   uint i;
+   for (i = 0; i < 2; i++) {
+      if (sp->constants[i].size)
+         ws->buffer_unmap(ws, sp->constants[i].buffer);
+      sp->mapped_constants[i] = NULL;
+   }
+}
+
+
+boolean
+cell_draw_arrays(struct pipe_context *pipe, unsigned mode,
+                     unsigned start, unsigned count)
+{
+   return cell_draw_elements(pipe, NULL, 0, mode, start, count);
+}
+
+
+
+/**
+ * Draw vertex arrays, with optional indexing.
+ * Basically, map the vertex buffers (and drawing surfaces), then hand off
+ * the drawing to the 'draw' module.
+ *
+ * XXX should the element buffer be specified/bound with a separate function?
+ */
+boolean
+cell_draw_elements(struct pipe_context *pipe,
+                       struct pipe_buffer *indexBuffer,
+                       unsigned indexSize,
+                       unsigned mode, unsigned start, unsigned count)
+{
+   struct cell_context *sp = cell_context(pipe);
+   struct draw_context *draw = sp->draw;
+   unsigned i;
+
+   /* first, check that the primitive is not malformed.  It is the
+    * state tracker's responsibility to do send only correctly formed
+    * primitives down.  It currently isn't doing that though...
+    */
+#if 1
+   count = draw_trim_prim( mode, count );
+#else
+   if (!draw_validate_prim( mode, count ))
+      assert(0);
+#endif
+
+   if (sp->dirty)
+      cell_update_derived( sp );
+
+#if 0
+   cell_map_surfaces(sp);
+#endif
+   cell_map_constant_buffers(sp);
+
+   /*
+    * Map vertex buffers
+    */
+   for (i = 0; i < PIPE_ATTRIB_MAX; i++) {
+      if (sp->vertex_buffer[i].buffer) {
+         void *buf = pipe->winsys->buffer_map(pipe->winsys,
+                                              sp->vertex_buffer[i].buffer,
+                                              PIPE_BUFFER_USAGE_CPU_READ);
+         draw_set_mapped_vertex_buffer(draw, i, buf);
+      }
+   }
+   /* Map index buffer, if present */
+   if (indexBuffer) {
+      void *mapped_indexes = pipe->winsys->buffer_map(pipe->winsys,
+                                                      indexBuffer,
+                                                      PIPE_BUFFER_USAGE_CPU_READ);
+      draw_set_mapped_element_buffer(draw, indexSize, mapped_indexes);
+   }
+   else {
+      /* no index/element buffer */
+      draw_set_mapped_element_buffer(draw, 0, NULL);
+   }
+
+
+   /* draw! */
+   draw_arrays(draw, mode, start, count);
+
+   /*
+    * unmap vertex/index buffers - will cause draw module to flush
+    */
+   for (i = 0; i < PIPE_ATTRIB_MAX; i++) {
+      if (sp->vertex_buffer[i].buffer) {
+         draw_set_mapped_vertex_buffer(draw, i, NULL);
+         pipe->winsys->buffer_unmap(pipe->winsys, sp->vertex_buffer[i].buffer);
+      }
+   }
+   if (indexBuffer) {
+      draw_set_mapped_element_buffer(draw, 0, NULL);
+      pipe->winsys->buffer_unmap(pipe->winsys, indexBuffer);
+   }
+
+   /* Note: leave drawing surfaces mapped */
+   cell_unmap_constant_buffers(sp);
+
+   return TRUE;
+}
diff --git a/src/gallium/drivers/cell/ppu/cell_draw_arrays.h b/src/gallium/drivers/cell/ppu/cell_draw_arrays.h
new file mode 100644
index 0000000000..d5df4aa05f
--- /dev/null
+++ b/src/gallium/drivers/cell/ppu/cell_draw_arrays.h
@@ -0,0 +1,42 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+#ifndef CELL_DRAW_ARRAYS_H
+#define CELL_DRAW_ARRAYS_H
+
+
+boolean cell_draw_arrays(struct pipe_context *pipe, unsigned mode,
+                         unsigned start, unsigned count);
+
+boolean cell_draw_elements(struct pipe_context *pipe,
+                           struct pipe_buffer *indexBuffer,
+                           unsigned indexSize,
+                           unsigned mode, unsigned start, unsigned count);
+
+
+
+#endif /* CELL_DRAW_ARRAYS_H */
diff --git a/src/gallium/drivers/cell/ppu/cell_flush.c b/src/gallium/drivers/cell/ppu/cell_flush.c
new file mode 100644
index 0000000000..f62bc4650c
--- /dev/null
+++ b/src/gallium/drivers/cell/ppu/cell_flush.c
@@ -0,0 +1,84 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+
+#include "cell_context.h"
+#include "cell_batch.h"
+#include "cell_flush.h"
+#include "cell_spu.h"
+#include "cell_render.h"
+#include "pipe/draw/draw_context.h"
+
+
+void
+cell_flush(struct pipe_context *pipe, unsigned flags)
+{
+   struct cell_context *cell = cell_context(pipe);
+
+   if (flags & PIPE_FLUSH_SWAPBUFFERS)
+      flags |= PIPE_FLUSH_WAIT;
+
+   draw_flush( cell->draw );
+   cell_flush_int(pipe, flags);
+}
+
+
+/** internal flush */
+void
+cell_flush_int(struct pipe_context *pipe, unsigned flags)
+{
+   static boolean flushing = FALSE;  /* recursion catcher */
+   struct cell_context *cell = cell_context(pipe);
+   uint i;
+
+   ASSERT(!flushing);
+   flushing = TRUE;
+
+   if (flags & PIPE_FLUSH_WAIT) {
+      uint64_t *cmd = (uint64_t *) cell_batch_alloc(cell, sizeof(uint64_t));
+      *cmd = CELL_CMD_FINISH;
+   }
+
+   cell_batch_flush(cell);
+
+#if 0
+   /* Send CMD_FINISH to all SPUs */
+   for (i = 0; i < cell->num_spus; i++) {
+      send_mbox_message(cell_global.spe_contexts[i], CELL_CMD_FINISH);
+   }
+#endif
+
+   if (flags & PIPE_FLUSH_WAIT) {
+      /* Wait for ack */
+      for (i = 0; i < cell->num_spus; i++) {
+         uint k = wait_mbox_message(cell_global.spe_contexts[i]);
+         assert(k == CELL_CMD_FINISH);
+      }
+   }
+
+   flushing = FALSE;
+}
diff --git a/src/gallium/drivers/cell/ppu/cell_flush.h b/src/gallium/drivers/cell/ppu/cell_flush.h
new file mode 100644
index 0000000000..eda351b1cb
--- /dev/null
+++ b/src/gallium/drivers/cell/ppu/cell_flush.h
@@ -0,0 +1,38 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+
+#ifndef CELL_FLUSH
+#define CELL_FLUSH
+
+extern void
+cell_flush(struct pipe_context *pipe, unsigned flags);
+
+extern void
+cell_flush_int(struct pipe_context *pipe, unsigned flags);
+
+#endif
diff --git a/src/gallium/drivers/cell/ppu/cell_render.c b/src/gallium/drivers/cell/ppu/cell_render.c
new file mode 100644
index 0000000000..4ab277a4b2
--- /dev/null
+++ b/src/gallium/drivers/cell/ppu/cell_render.c
@@ -0,0 +1,210 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+/**
+ * \brief  Last stage of 'draw' pipeline: send tris to SPUs.
+ * \author  Brian Paul
+ */
+
+#include "cell_context.h"
+#include "cell_render.h"
+#include "cell_spu.h"
+#include "pipe/p_util.h"
+#include "pipe/draw/draw_private.h"
+
+
+struct render_stage {
+   struct draw_stage stage; /**< This must be first (base class) */
+
+   struct cell_context *cell;
+};
+
+
+static INLINE struct render_stage *
+render_stage(struct draw_stage *stage)
+{
+   return (struct render_stage *) stage;
+}
+
+
+static void render_begin( struct draw_stage *stage )
+{
+#if 0
+   struct render_stage *render = render_stage(stage);
+   struct cell_context *sp = render->cell;
+   const struct pipe_shader_state *fs = &render->cell->fs->shader;
+   render->quad.nr_attrs = render->cell->nr_frag_attrs;
+
+   render->firstFpInput = fs->input_semantic_name[0];
+
+   sp->quad.first->begin(sp->quad.first);
+#endif
+}
+
+
+static void render_end( struct draw_stage *stage )
+{
+}
+
+
+static void reset_stipple_counter( struct draw_stage *stage )
+{
+   struct render_stage *render = render_stage(stage);
+   /*render->cell->line_stipple_counter = 0;*/
+}
+
+
+static void
+render_point(struct draw_stage *stage, struct prim_header *prim)
+{
+}
+
+
+static void
+render_line(struct draw_stage *stage, struct prim_header *prim)
+{
+}
+
+
+/** Write a vertex into the prim buffer */
+static void
+save_vertex(struct cell_prim_buffer *buf, uint pos,
+            const struct vertex_header *vert)
+{
+   uint attr, j;
+
+   for (attr = 0; attr < 2; attr++) {
+      for (j = 0; j < 4; j++) {
+         buf->vertex[pos][attr][j] = vert->data[attr][j];
+      }
+   }
+
+   /* update bounding box */
+   if (vert->data[0][0] < buf->xmin)
+      buf->xmin = vert->data[0][0];
+   if (vert->data[0][0] > buf->xmax)
+      buf->xmax = vert->data[0][0];
+   if (vert->data[0][1] < buf->ymin)
+      buf->ymin = vert->data[0][1];
+   if (vert->data[0][1] > buf->ymax)
+      buf->ymax = vert->data[0][1];
+}
+
+
+static void
+render_tri(struct draw_stage *stage, struct prim_header *prim)
+{
+   struct render_stage *rs = render_stage(stage);
+   struct cell_context *cell = rs->cell;
+   struct cell_prim_buffer *buf = &cell->prim_buffer;
+   uint i;
+
+   if (buf->num_verts + 3 > CELL_MAX_VERTS) {
+      cell_flush_prim_buffer(cell);
+   }
+
+   i = buf->num_verts;
+   assert(i+2 <= CELL_MAX_VERTS);
+   save_vertex(buf, i+0, prim->v[0]);
+   save_vertex(buf, i+1, prim->v[1]);
+   save_vertex(buf, i+2, prim->v[2]);
+   buf->num_verts += 3;
+}
+
+
+/**
+ * Send the a RENDER command to all SPUs to have them render the prims
+ * in the current prim_buffer.
+ */
+void
+cell_flush_prim_buffer(struct cell_context *cell)
+{
+   uint i;
+
+   if (cell->prim_buffer.num_verts == 0)
+      return;
+
+   for (i = 0; i < cell->num_spus; i++) {
+      struct cell_command_render *render = &cell_global.command[i].render;
+      render->prim_type = PIPE_PRIM_TRIANGLES;
+      render->num_verts = cell->prim_buffer.num_verts;
+      render->vertex_size = cell->vertex_info->size * 4;
+      render->xmin = cell->prim_buffer.xmin;
+      render->ymin = cell->prim_buffer.ymin;
+      render->xmax = cell->prim_buffer.xmax;
+      render->ymax = cell->prim_buffer.ymax;
+      render->vertex_data = &cell->prim_buffer.vertex;
+      ASSERT_ALIGN16(render->vertex_data);
+      send_mbox_message(cell_global.spe_contexts[i], CELL_CMD_RENDER);
+   }
+
+   cell->prim_buffer.num_verts = 0;
+
+   cell->prim_buffer.xmin = 1e100;
+   cell->prim_buffer.ymin = 1e100;
+   cell->prim_buffer.xmax = -1e100;
+   cell->prim_buffer.ymax = -1e100;
+
+   /* XXX temporary, need to double-buffer the prim buffer until we get
+    * a real command buffer/list system.
+    */
+   cell_flush(&cell->pipe, 0x0);
+}
+
+
+
+static void render_destroy( struct draw_stage *stage )
+{
+   FREE( stage );
+}
+
+
+/**
+ * Create a new draw/render stage.  This will be plugged into the
+ * draw module as the last pipeline stage.
+ */
+struct draw_stage *cell_draw_render_stage( struct cell_context *cell )
+{
+   struct render_stage *render = CALLOC_STRUCT(render_stage);
+
+   render->cell = cell;
+   render->stage.draw = cell->draw;
+   render->stage.begin = render_begin;
+   render->stage.point = render_point;
+   render->stage.line = render_line;
+   render->stage.tri = render_tri;
+   render->stage.end = render_end;
+   render->stage.reset_stipple_counter = reset_stipple_counter;
+   render->stage.destroy = render_destroy;
+
+   /*
+   render->quad.coef = render->coef;
+   render->quad.posCoef = &render->posCoef;
+   */
+
+   return &render->stage;
+}
diff --git a/src/gallium/drivers/cell/ppu/cell_render.h b/src/gallium/drivers/cell/ppu/cell_render.h
new file mode 100644
index 0000000000..826dcbafeb
--- /dev/null
+++ b/src/gallium/drivers/cell/ppu/cell_render.h
@@ -0,0 +1,39 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+#ifndef CELL_RENDER_H
+#define CELL_RENDER_H
+
+struct cell_context;
+struct draw_stage;
+
+extern void
+cell_flush_prim_buffer(struct cell_context *cell);
+
+extern struct draw_stage *cell_draw_render_stage( struct cell_context *cell );
+
+#endif /* CELL_RENDER_H */
diff --git a/src/gallium/drivers/cell/ppu/cell_spu.c b/src/gallium/drivers/cell/ppu/cell_spu.c
new file mode 100644
index 0000000000..7c83a47e57
--- /dev/null
+++ b/src/gallium/drivers/cell/ppu/cell_spu.c
@@ -0,0 +1,155 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+
+#include <pthread.h>
+
+#include "cell_spu.h"
+#include "pipe/p_format.h"
+#include "pipe/p_state.h"
+#include "pipe/cell/common.h"
+
+
+/*
+helpful headers:
+/opt/ibm/cell-sdk/prototype/src/include/ppu/cbe_mfc.h
+*/
+
+
+struct cell_global_info cell_global;
+
+
+/**
+ * Write a 1-word message to the given SPE mailbox.
+ */
+void
+send_mbox_message(spe_context_ptr_t ctx, unsigned int msg)
+{
+   spe_in_mbox_write(ctx, &msg, 1, SPE_MBOX_ALL_BLOCKING);
+}
+
+
+/**
+ * Wait for a 1-word message to arrive in given mailbox.
+ */
+uint
+wait_mbox_message(spe_context_ptr_t ctx)
+{
+   do {
+      unsigned data;
+      int count = spe_out_mbox_read(ctx, &data, 1);
+
+      if (count == 1) {
+	 return data;
+      }
+      
+      if (count < 0) {
+	 /* error */ ;
+      }
+   } while (1);
+}
+
+
+static void *cell_thread_function(void *arg)
+{
+   struct cell_init_info *init = (struct cell_init_info *) arg;
+   unsigned entry = SPE_DEFAULT_ENTRY;
+
+   ASSERT_ALIGN16(init);
+
+   if (spe_context_run(cell_global.spe_contexts[init->id], &entry, 0,
+                       init, NULL, NULL) < 0) {
+      fprintf(stderr, "spe_context_run() failed\n");
+      exit(1);
+   }
+
+   pthread_exit(NULL);
+}
+
+
+/**
+ * Create the SPU threads
+ */
+void
+cell_start_spus(struct cell_context *cell)
+{
+   uint i, j;
+
+   assert(cell->num_spus <= MAX_SPUS);
+
+   ASSERT_ALIGN16(&cell_global.command[0]);
+   ASSERT_ALIGN16(&cell_global.command[1]);
+
+   ASSERT_ALIGN16(&cell_global.inits[0]);
+   ASSERT_ALIGN16(&cell_global.inits[1]);
+
+   for (i = 0; i < cell->num_spus; i++) {
+      cell_global.inits[i].id = i;
+      cell_global.inits[i].num_spus = cell->num_spus;
+      cell_global.inits[i].cmd = &cell_global.command[i];
+      for (j = 0; j < CELL_NUM_BUFFERS; j++) {
+         cell_global.inits[i].buffers[j] = cell->buffer[j];
+      }
+      cell_global.inits[i].buffer_status = &cell->buffer_status[0][0][0];
+
+      cell_global.spe_contexts[i] = spe_context_create(0, NULL);
+      if (!cell_global.spe_contexts[i]) {
+         fprintf(stderr, "spe_context_create() failed\n");
+         exit(1);
+      }
+
+      if (spe_program_load(cell_global.spe_contexts[i], &g3d_spu)) {
+         fprintf(stderr, "spe_program_load() failed\n");
+         exit(1);
+      }
+      
+      pthread_create(&cell_global.spe_threads[i], NULL, &cell_thread_function,
+		     &cell_global.inits[i]);
+   }
+}
+
+
+/**
+ * Tell all the SPUs to stop/exit.
+ */
+void
+cell_spu_exit(struct cell_context *cell)
+{
+   uint i;
+
+   for (i = 0; i < cell->num_spus; i++) {
+      send_mbox_message(cell_global.spe_contexts[i], CELL_CMD_EXIT);
+   }
+
+   /* wait for threads to exit */
+   for (i = 0; i < cell->num_spus; i++) {
+      void *value;
+      pthread_join(cell_global.spe_threads[i], &value);
+      cell_global.spe_threads[i] = 0;
+      cell_global.spe_contexts[i] = 0;
+   }
+}
diff --git a/src/gallium/drivers/cell/ppu/cell_spu.h b/src/gallium/drivers/cell/ppu/cell_spu.h
new file mode 100644
index 0000000000..19eff94f96
--- /dev/null
+++ b/src/gallium/drivers/cell/ppu/cell_spu.h
@@ -0,0 +1,82 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+#ifndef CELL_SPU
+#define CELL_SPU
+
+
+#include <libspe2.h>
+#include <libmisc.h>
+#include "pipe/cell/common.h"
+
+#include "cell_context.h"
+
+
+#define MAX_SPUS 8
+
+/**
+ * Global vars, for now anyway.
+ */
+struct cell_global_info
+{
+   /**
+    * SPU/SPE handles, etc
+    */
+   spe_context_ptr_t spe_contexts[MAX_SPUS];
+   pthread_t spe_threads[MAX_SPUS];
+
+   /**
+    * Data sent to SPUs
+    */
+   struct cell_init_info inits[MAX_SPUS];
+   struct cell_command command[MAX_SPUS];
+};
+
+
+extern struct cell_global_info cell_global;
+
+
+/** This is the handle for the actual SPE code */
+extern spe_program_handle_t g3d_spu;
+
+
+extern void
+send_mbox_message(spe_context_ptr_t ctx, unsigned int msg);
+
+extern uint
+wait_mbox_message(spe_context_ptr_t ctx);
+
+
+extern void
+cell_start_spus(struct cell_context *cell);
+
+
+extern void
+cell_spu_exit(struct cell_context *cell);
+
+
+#endif /* CELL_SPU */
diff --git a/src/gallium/drivers/cell/ppu/cell_state.h b/src/gallium/drivers/cell/ppu/cell_state.h
new file mode 100644
index 0000000000..3a71ba14fa
--- /dev/null
+++ b/src/gallium/drivers/cell/ppu/cell_state.h
@@ -0,0 +1,115 @@
+
+
+#ifndef CELL_STATE_H
+#define CELL_STATE_H
+
+
+#define CELL_NEW_VIEWPORT      0x1
+#define CELL_NEW_RASTERIZER    0x2
+#define CELL_NEW_FS            0x4
+#define CELL_NEW_BLEND         0x8
+#define CELL_NEW_CLIP          0x10
+#define CELL_NEW_SCISSOR       0x20
+#define CELL_NEW_STIPPLE       0x40
+#define CELL_NEW_FRAMEBUFFER   0x80
+#define CELL_NEW_ALPHA_TEST    0x100
+#define CELL_NEW_DEPTH_STENCIL 0x200
+#define CELL_NEW_SAMPLER       0x400
+#define CELL_NEW_TEXTURE       0x800
+#define CELL_NEW_VERTEX        0x1000
+#define CELL_NEW_VS            0x2000
+#define CELL_NEW_CONSTANTS     0x4000
+#define CELL_NEW_VERTEX_INFO   0x8000
+
+
+
+extern void
+cell_set_framebuffer_state( struct pipe_context *,
+                            const struct pipe_framebuffer_state * );
+
+
+
+extern void *
+cell_create_blend_state(struct pipe_context *, const struct pipe_blend_state *);
+extern void cell_bind_blend_state(struct pipe_context *, void *);
+extern void cell_delete_blend_state(struct pipe_context *, void *);
+
+extern void cell_set_blend_color( struct pipe_context *pipe,
+                                  const struct pipe_blend_color *blend_color );
+
+
+void *
+cell_create_sampler_state(struct pipe_context *,
+                          const struct pipe_sampler_state *);
+
+extern void
+cell_bind_sampler_state(struct pipe_context *, unsigned, void *);
+
+extern void
+cell_delete_sampler_state(struct pipe_context *, void *);
+
+
+extern void *
+cell_create_depth_stencil_alpha_state(struct pipe_context *,
+                                const struct pipe_depth_stencil_alpha_state *);
+
+extern void
+cell_bind_depth_stencil_alpha_state(struct pipe_context *, void *);
+
+extern void
+cell_delete_depth_stencil_alpha_state(struct pipe_context *, void *);
+
+
+void *cell_create_fs_state(struct pipe_context *,
+                               const struct pipe_shader_state *);
+void cell_bind_fs_state(struct pipe_context *, void *);
+void cell_delete_fs_state(struct pipe_context *, void *);
+void *cell_create_vs_state(struct pipe_context *,
+                               const struct pipe_shader_state *);
+void cell_bind_vs_state(struct pipe_context *, void *);
+void cell_delete_vs_state(struct pipe_context *, void *);
+
+
+void *
+cell_create_rasterizer_state(struct pipe_context *,
+                             const struct pipe_rasterizer_state *);
+void cell_bind_rasterizer_state(struct pipe_context *, void *);
+void cell_delete_rasterizer_state(struct pipe_context *, void *);
+
+
+void cell_set_clip_state( struct pipe_context *,
+                          const struct pipe_clip_state * );
+
+void cell_set_constant_buffer(struct pipe_context *pipe,
+                              uint shader, uint index,
+                              const struct pipe_constant_buffer *buf);
+
+void cell_set_polygon_stipple( struct pipe_context *,
+                               const struct pipe_poly_stipple * );
+
+void
+cell_set_sampler_texture(struct pipe_context *pipe,
+                         unsigned sampler,
+                         struct pipe_texture *texture);
+
+void cell_set_scissor_state( struct pipe_context *,
+                             const struct pipe_scissor_state * );
+
+void cell_set_texture_state( struct pipe_context *,
+                             unsigned unit, struct pipe_texture * );
+
+void cell_set_vertex_element(struct pipe_context *,
+                             unsigned index,
+                             const struct pipe_vertex_element *);
+
+void cell_set_vertex_buffer(struct pipe_context *,
+                            unsigned index,
+                            const struct pipe_vertex_buffer *);
+
+void cell_set_viewport_state( struct pipe_context *,
+                              const struct pipe_viewport_state * );
+
+
+void cell_update_derived( struct cell_context *softpipe );
+
+#endif
diff --git a/src/gallium/drivers/cell/ppu/cell_state_blend.c b/src/gallium/drivers/cell/ppu/cell_state_blend.c
new file mode 100644
index 0000000000..4fc60548c8
--- /dev/null
+++ b/src/gallium/drivers/cell/ppu/cell_state_blend.c
@@ -0,0 +1,109 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+/* Authors:  Keith Whitwell <keith@tungstengraphics.com>
+ */
+
+#include "pipe/p_util.h"
+#include "pipe/draw/draw_context.h"
+#include "cell_context.h"
+#include "cell_state.h"
+
+
+
+void *
+cell_create_blend_state(struct pipe_context *pipe,
+                        const struct pipe_blend_state *blend)
+{
+   return mem_dup(blend, sizeof(*blend));
+}
+
+
+void
+cell_bind_blend_state(struct pipe_context *pipe, void *blend)
+{
+   struct cell_context *cell = cell_context(pipe);
+
+   draw_flush(cell->draw);
+
+   cell->blend = (const struct pipe_blend_state *)blend;
+
+   cell->dirty |= CELL_NEW_BLEND;
+}
+
+
+void
+cell_delete_blend_state(struct pipe_context *pipe, void *blend)
+{
+   FREE(blend);
+}
+
+
+void
+cell_set_blend_color(struct pipe_context *pipe,
+                     const struct pipe_blend_color *blend_color)
+{
+   struct cell_context *cell = cell_context(pipe);
+
+   draw_flush(cell->draw);
+
+   cell->blend_color = *blend_color;
+
+   cell->dirty |= CELL_NEW_BLEND;
+}
+
+
+
+
+void *
+cell_create_depth_stencil_alpha_state(struct pipe_context *pipe,
+                 const struct pipe_depth_stencil_alpha_state *depth_stencil)
+{
+   return mem_dup(depth_stencil, sizeof(*depth_stencil));
+}
+
+
+void
+cell_bind_depth_stencil_alpha_state(struct pipe_context *pipe,
+                                    void *depth_stencil)
+{
+   struct cell_context *cell = cell_context(pipe);
+
+   draw_flush(cell->draw);
+
+   cell->depth_stencil
+      = (const struct pipe_depth_stencil_alpha_state *) depth_stencil;
+
+   cell->dirty |= CELL_NEW_DEPTH_STENCIL;
+}
+
+
+void
+cell_delete_depth_stencil_alpha_state(struct pipe_context *pipe, void *depth)
+{
+   FREE(depth);
+}
diff --git a/src/gallium/drivers/cell/ppu/cell_state_clip.c b/src/gallium/drivers/cell/ppu/cell_state_clip.c
new file mode 100644
index 0000000000..4f43665941
--- /dev/null
+++ b/src/gallium/drivers/cell/ppu/cell_state_clip.c
@@ -0,0 +1,84 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+/* Authors:  Keith Whitwell <keith@tungstengraphics.com>
+ */
+
+#include "cell_context.h"
+#include "cell_state.h"
+#include "pipe/draw/draw_context.h"
+
+
+void cell_set_clip_state( struct pipe_context *pipe,
+			     const struct pipe_clip_state *clip )
+{
+   struct cell_context *cell = cell_context(pipe);
+
+   /* pass the clip state to the draw module */
+   draw_set_clip_state(cell->draw, clip);
+}
+
+
+
+/* Called when driver state tracker notices changes to the viewport
+ * matrix:
+ */
+void cell_set_viewport_state( struct pipe_context *pipe,
+                              const struct pipe_viewport_state *viewport )
+{
+   struct cell_context *cell = cell_context(pipe);
+
+   cell->viewport = *viewport; /* struct copy */
+   cell->dirty |= CELL_NEW_VIEWPORT;
+
+   /* pass the viewport info to the draw module */
+   draw_set_viewport_state(cell->draw, viewport);
+
+   /* Using tnl/ and vf/ modules is temporary while getting started.
+    * Full pipe will have vertex shader, vertex fetch of its own.
+    */
+}
+
+
+void cell_set_scissor_state( struct pipe_context *pipe,
+                             const struct pipe_scissor_state *scissor )
+{
+   struct cell_context *cell = cell_context(pipe);
+
+   memcpy( &cell->scissor, scissor, sizeof(*scissor) );
+   cell->dirty |= CELL_NEW_SCISSOR;
+}
+
+
+void cell_set_polygon_stipple( struct pipe_context *pipe,
+                                   const struct pipe_poly_stipple *stipple )
+{
+   struct cell_context *cell = cell_context(pipe);
+
+   memcpy( &cell->poly_stipple, stipple, sizeof(*stipple) );
+   cell->dirty |= CELL_NEW_STIPPLE;
+}
diff --git a/src/gallium/drivers/cell/ppu/cell_state_derived.c b/src/gallium/drivers/cell/ppu/cell_state_derived.c
new file mode 100644
index 0000000000..56daf5dfde
--- /dev/null
+++ b/src/gallium/drivers/cell/ppu/cell_state_derived.c
@@ -0,0 +1,192 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+#include "pipe/p_util.h"
+#include "pipe/p_shader_tokens.h"
+#include "pipe/draw/draw_context.h"
+#include "pipe/draw/draw_vertex.h"
+#include "cell_context.h"
+#include "cell_batch.h"
+#include "cell_state.h"
+#include "cell_state_emit.h"
+
+
+static int
+find_vs_output(const struct pipe_shader_state *vs,
+               uint semantic_name,
+               uint semantic_index)
+{
+   uint i;
+   for (i = 0; i < vs->num_outputs; i++) {
+      if (vs->output_semantic_name[i] == semantic_name &&
+          vs->output_semantic_index[i] == semantic_index)
+         return i;
+   }
+   return -1;
+}
+
+
+/**
+ * Determine how to map vertex program outputs to fragment program inputs.
+ * Basically, this will be used when computing the triangle interpolation
+ * coefficients from the post-transform vertex attributes.
+ */
+static void
+calculate_vertex_layout( struct cell_context *cell )
+{
+   const struct pipe_shader_state *vs = &cell->vs->shader;
+   const struct pipe_shader_state *fs = &cell->fs->shader;
+   const enum interp_mode colorInterp
+      = cell->rasterizer->flatshade ? INTERP_CONSTANT : INTERP_LINEAR;
+   struct vertex_info *vinfo = &cell->vertex_info;
+   uint i;
+   int src;
+
+#if 0
+   if (cell->vbuf) {
+      /* if using the post-transform vertex buffer, tell draw_vbuf to
+       * simply emit the whole post-xform vertex as-is:
+       */
+      struct vertex_info *vinfo_vbuf = &cell->vertex_info_vbuf;
+      vinfo_vbuf->num_attribs = 0;
+      draw_emit_vertex_attr(vinfo_vbuf, EMIT_ALL, INTERP_NONE, 0);
+      vinfo_vbuf->size = 4 * vs->num_outputs + sizeof(struct vertex_header)/4;
+   }
+#endif
+
+   /* reset vinfo */
+   vinfo->num_attribs = 0;
+
+   /* we always want to emit vertex pos */
+   src = find_vs_output(vs, TGSI_SEMANTIC_POSITION, 0);
+   assert(src >= 0);
+   draw_emit_vertex_attr(vinfo, EMIT_4F, INTERP_POS, src);
+
+
+   /*
+    * Loop over fragment shader inputs, searching for the matching output
+    * from the vertex shader.
+    */
+   for (i = 0; i < fs->num_inputs; i++) {
+      switch (fs->input_semantic_name[i]) {
+      case TGSI_SEMANTIC_POSITION:
+         /* already done above */
+         break;
+
+      case TGSI_SEMANTIC_COLOR:
+         src = find_vs_output(vs, TGSI_SEMANTIC_COLOR, 
+                              fs->input_semantic_index[i]);
+         assert(src >= 0);
+         draw_emit_vertex_attr(vinfo, EMIT_4F, colorInterp, src);
+         break;
+
+      case TGSI_SEMANTIC_FOG:
+         src = find_vs_output(vs, TGSI_SEMANTIC_FOG, 0);
+#if 1
+         if (src < 0) /* XXX temp hack, try demos/fogcoord.c with this */
+            src = 0;
+#endif
+         assert(src >= 0);
+         draw_emit_vertex_attr(vinfo, EMIT_1F, INTERP_PERSPECTIVE, src);
+         break;
+
+      case TGSI_SEMANTIC_GENERIC:
+         /* this includes texcoords and varying vars */
+         src = find_vs_output(vs, TGSI_SEMANTIC_GENERIC,
+                              fs->input_semantic_index[i]);
+         assert(src >= 0);
+         draw_emit_vertex_attr(vinfo, EMIT_4F, INTERP_PERSPECTIVE, src);
+         break;
+
+      default:
+         assert(0);
+      }
+   }
+
+   draw_compute_vertex_size(vinfo);
+
+   /* XXX only signal this if format really changes */
+   cell->dirty |= CELL_NEW_VERTEX_INFO;
+}
+
+
+#if 0
+/**
+ * Recompute cliprect from scissor bounds, scissor enable and surface size.
+ */
+static void
+compute_cliprect(struct cell_context *sp)
+{
+   unsigned surfWidth, surfHeight;
+
+   if (sp->framebuffer.num_cbufs > 0) {
+      surfWidth = sp->framebuffer.cbufs[0]->width;
+      surfHeight = sp->framebuffer.cbufs[0]->height;
+   }
+   else {
+      /* no surface? */
+      surfWidth = sp->scissor.maxx;
+      surfHeight = sp->scissor.maxy;
+   }
+
+   if (sp->rasterizer->scissor) {
+      /* clip to scissor rect */
+      sp->cliprect.minx = MAX2(sp->scissor.minx, 0);
+      sp->cliprect.miny = MAX2(sp->scissor.miny, 0);
+      sp->cliprect.maxx = MIN2(sp->scissor.maxx, surfWidth);
+      sp->cliprect.maxy = MIN2(sp->scissor.maxy, surfHeight);
+   }
+   else {
+      /* clip to surface bounds */
+      sp->cliprect.minx = 0;
+      sp->cliprect.miny = 0;
+      sp->cliprect.maxx = surfWidth;
+      sp->cliprect.maxy = surfHeight;
+   }
+}
+#endif
+
+
+
+void cell_update_derived( struct cell_context *cell )
+{
+   if (cell->dirty & (CELL_NEW_RASTERIZER |
+                      CELL_NEW_FS |
+                      CELL_NEW_VS))
+      calculate_vertex_layout( cell );
+
+#if 0
+   if (cell->dirty & (CELL_NEW_SCISSOR |
+                      CELL_NEW_DEPTH_STENCIL_ALPHA |
+                      CELL_NEW_FRAMEBUFFER))
+      compute_cliprect(cell);
+#endif
+
+   cell_emit_state(cell);
+
+   cell->dirty = 0;
+}
diff --git a/src/gallium/drivers/cell/ppu/cell_state_emit.c b/src/gallium/drivers/cell/ppu/cell_state_emit.c
new file mode 100644
index 0000000000..5d2a786449
--- /dev/null
+++ b/src/gallium/drivers/cell/ppu/cell_state_emit.c
@@ -0,0 +1,103 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+#include "pipe/p_util.h"
+#include "cell_context.h"
+#include "cell_state.h"
+#include "cell_state_emit.h"
+#include "cell_batch.h"
+#include "cell_texture.h"
+
+
+static void
+emit_state_cmd(struct cell_context *cell, uint cmd,
+               const void *state, uint state_size)
+{
+   uint64_t *dst = (uint64_t *) 
+       cell_batch_alloc(cell, ROUNDUP8(sizeof(uint64_t) + state_size));
+   *dst = cmd;
+   memcpy(dst + 1, state, state_size);
+}
+
+
+
+void
+cell_emit_state(struct cell_context *cell)
+{
+   if (cell->dirty & CELL_NEW_FRAMEBUFFER) {
+      struct pipe_surface *cbuf = cell->framebuffer.cbufs[0];
+      struct pipe_surface *zbuf = cell->framebuffer.zsbuf;
+      struct cell_command_framebuffer *fb
+         = cell_batch_alloc(cell, sizeof(*fb));
+      fb->opcode = CELL_CMD_STATE_FRAMEBUFFER;
+      fb->color_start = cell->cbuf_map[0];
+      fb->color_format = cbuf->format;
+      fb->depth_start = cell->zsbuf_map;
+      fb->depth_format = zbuf ? zbuf->format : PIPE_FORMAT_NONE;
+      fb->width = cell->framebuffer.cbufs[0]->width;
+      fb->height = cell->framebuffer.cbufs[0]->height;
+   }
+
+   if (cell->dirty & CELL_NEW_BLEND) {
+      emit_state_cmd(cell, CELL_CMD_STATE_BLEND,
+                     cell->blend,
+                     sizeof(struct pipe_blend_state));
+   }
+
+   if (cell->dirty & CELL_NEW_DEPTH_STENCIL) {
+      emit_state_cmd(cell, CELL_CMD_STATE_DEPTH_STENCIL,
+                     cell->depth_stencil,
+                     sizeof(struct pipe_depth_stencil_alpha_state));
+   }
+
+   if (cell->dirty & CELL_NEW_SAMPLER) {
+      emit_state_cmd(cell, CELL_CMD_STATE_SAMPLER,
+                     cell->sampler[0], sizeof(struct pipe_sampler_state));
+   }
+
+   if (cell->dirty & CELL_NEW_TEXTURE) {
+      struct cell_command_texture texture;
+      if (cell->texture[0]) {
+         texture.start = cell->texture[0]->tiled_data;
+         texture.width = cell->texture[0]->base.width[0];
+         texture.height = cell->texture[0]->base.height[0];
+      }
+      else {
+         texture.start = NULL;
+         texture.width = 0;
+         texture.height = 0;
+      }
+
+      emit_state_cmd(cell, CELL_CMD_STATE_TEXTURE,
+                     &texture, sizeof(struct cell_command_texture));
+   }
+
+   if (cell->dirty & CELL_NEW_VERTEX_INFO) {
+      emit_state_cmd(cell, CELL_CMD_STATE_VERTEX_INFO,
+                     &cell->vertex_info, sizeof(struct vertex_info));
+   }
+}
diff --git a/src/gallium/drivers/cell/ppu/cell_state_emit.h b/src/gallium/drivers/cell/ppu/cell_state_emit.h
new file mode 100644
index 0000000000..59f8affe8d
--- /dev/null
+++ b/src/gallium/drivers/cell/ppu/cell_state_emit.h
@@ -0,0 +1,36 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+#ifndef CELL_STATE_EMIT_H
+#define CELL_STATE_EMIT_H
+
+
+extern void
+cell_emit_state(struct cell_context *cell);
+
+
+#endif /* CELL_STATE_EMIT_H */
diff --git a/src/gallium/drivers/cell/ppu/cell_state_fs.c b/src/gallium/drivers/cell/ppu/cell_state_fs.c
new file mode 100644
index 0000000000..3f46a87d18
--- /dev/null
+++ b/src/gallium/drivers/cell/ppu/cell_state_fs.c
@@ -0,0 +1,171 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+#include "pipe/p_defines.h"
+#include "pipe/p_util.h"
+#include "pipe/p_inlines.h"
+#include "pipe/p_winsys.h"
+#include "pipe/draw/draw_context.h"
+#if 0
+#include "pipe/p_shader_tokens.h"
+#include "pipe/llvm/gallivm.h"
+#include "pipe/tgsi/util/tgsi_dump.h"
+#include "pipe/tgsi/exec/tgsi_sse2.h"
+#endif
+
+#include "cell_context.h"
+#include "cell_state.h"
+
+
+void *
+cell_create_fs_state(struct pipe_context *pipe,
+                     const struct pipe_shader_state *templ)
+{
+   /*struct cell_context *cell = cell_context(pipe);*/
+   struct cell_fragment_shader_state *state;
+
+   state = CALLOC_STRUCT(cell_fragment_shader_state);
+   if (!state)
+      return NULL;
+
+   state->shader = *templ;
+
+#if 0
+   if (cell->dump_fs) {
+      tgsi_dump(state->shader.tokens, 0);
+   }
+
+#if defined(__i386__) || defined(__386__)
+   if (cell->use_sse) {
+      x86_init_func( &state->sse2_program );
+      tgsi_emit_sse2_fs( state->shader.tokens, &state->sse2_program );
+   }
+#endif
+
+#ifdef MESA_LLVM
+   state->llvm_prog = 0;
+   if (!gallivm_global_cpu_engine()) {
+      gallivm_cpu_engine_create(state->llvm_prog);
+   }
+   else
+      gallivm_cpu_jit_compile(gallivm_global_cpu_engine(), state->llvm_prog);
+#endif
+#endif
+
+   return state;
+}
+
+
+void
+cell_bind_fs_state(struct pipe_context *pipe, void *fs)
+{
+   struct cell_context *cell = cell_context(pipe);
+
+   cell->fs = (struct cell_fragment_shader_state *) fs;
+
+   cell->dirty |= CELL_NEW_FS;
+}
+
+
+void
+cell_delete_fs_state(struct pipe_context *pipe, void *fs)
+{
+   struct cell_fragment_shader_state *state =
+      (struct cell_fragment_shader_state *) fs;
+
+   FREE( state );
+}
+
+
+void *
+cell_create_vs_state(struct pipe_context *pipe,
+                     const struct pipe_shader_state *templ)
+{
+   struct cell_context *cell = cell_context(pipe);
+   struct cell_vertex_shader_state *state;
+
+   state = CALLOC_STRUCT(cell_vertex_shader_state);
+   if (!state)
+      return NULL;
+
+   state->shader = *templ;
+
+   state->draw_data = draw_create_vertex_shader(cell->draw, &state->shader);
+   if (state->draw_data == NULL) {
+      FREE( state );
+      return NULL;
+   }
+
+   return state;
+}
+
+
+void
+cell_bind_vs_state(struct pipe_context *pipe, void *vs)
+{
+   struct cell_context *cell = cell_context(pipe);
+
+   cell->vs = (const struct cell_vertex_shader_state *) vs;
+
+   draw_bind_vertex_shader(cell->draw, cell->vs->draw_data);
+
+   cell->dirty |= CELL_NEW_VS;
+}
+
+
+void
+cell_delete_vs_state(struct pipe_context *pipe, void *vs)
+{
+   struct cell_context *cell = cell_context(pipe);
+
+   struct cell_vertex_shader_state *state =
+      (struct cell_vertex_shader_state *) vs;
+
+   draw_delete_vertex_shader(cell->draw, state->draw_data);
+   FREE( state );
+}
+
+
+void
+cell_set_constant_buffer(struct pipe_context *pipe,
+                         uint shader, uint index,
+                         const struct pipe_constant_buffer *buf)
+{
+   struct cell_context *cell = cell_context(pipe);
+   struct pipe_winsys *ws = pipe->winsys;
+
+   assert(shader < PIPE_SHADER_TYPES);
+   assert(index == 0);
+
+   /* note: reference counting */
+   pipe_buffer_reference(ws,
+                        &cell->constants[shader].buffer,
+                        buf->buffer);
+   cell->constants[shader].size = buf->size;
+
+   cell->dirty |= CELL_NEW_CONSTANTS;
+}
diff --git a/src/gallium/drivers/cell/ppu/cell_state_rasterizer.c b/src/gallium/drivers/cell/ppu/cell_state_rasterizer.c
new file mode 100644
index 0000000000..d8128ece54
--- /dev/null
+++ b/src/gallium/drivers/cell/ppu/cell_state_rasterizer.c
@@ -0,0 +1,106 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+#include "pipe/p_defines.h"
+#include "pipe/p_util.h"
+#include "pipe/draw/draw_context.h"
+#include "cell_context.h"
+#include "cell_state.h"
+
+
+
+struct spu_rasterizer_state
+{
+   unsigned flatshade:1;
+#if 0
+   unsigned light_twoside:1;
+   unsigned front_winding:2;  /**< PIPE_WINDING_x */
+   unsigned cull_mode:2;      /**< PIPE_WINDING_x */
+   unsigned fill_cw:2;        /**< PIPE_POLYGON_MODE_x */
+   unsigned fill_ccw:2;       /**< PIPE_POLYGON_MODE_x */
+   unsigned offset_cw:1;
+   unsigned offset_ccw:1;
+#endif
+   unsigned scissor:1;
+   unsigned poly_smooth:1;
+   unsigned poly_stipple_enable:1;
+   unsigned point_smooth:1;
+#if 0
+   unsigned point_sprite:1;
+   unsigned point_size_per_vertex:1; /**< size computed in vertex shader */
+#endif
+   unsigned multisample:1;         /* XXX maybe more ms state in future */
+   unsigned line_smooth:1;
+   unsigned line_stipple_enable:1;
+   unsigned line_stipple_factor:8;  /**< [1..256] actually */
+   unsigned line_stipple_pattern:16;
+#if 0
+   unsigned bypass_clipping:1;
+#endif
+   unsigned origin_lower_left:1;  /**< Is (0,0) the lower-left corner? */
+
+   float line_width;
+   float point_size;           /**< used when no per-vertex size */
+#if 0
+   float offset_units;
+   float offset_scale;
+   ubyte sprite_coord_mode[PIPE_MAX_SHADER_OUTPUTS]; /**< PIPE_SPRITE_COORD_ */
+#endif
+};
+
+
+
+void *
+cell_create_rasterizer_state(struct pipe_context *pipe,
+                             const struct pipe_rasterizer_state *setup)
+{
+   struct pipe_rasterizer_state *state
+      = MALLOC(sizeof(struct pipe_rasterizer_state));
+   memcpy(state, setup, sizeof(struct pipe_rasterizer_state));
+   return state;
+}
+
+
+void
+cell_bind_rasterizer_state(struct pipe_context *pipe, void *setup)
+{
+   struct cell_context *cell = cell_context(pipe);
+
+   /* pass-through to draw module */
+   draw_set_rasterizer_state(cell->draw, setup);
+
+   cell->rasterizer = (struct pipe_rasterizer_state *)setup;
+
+   cell->dirty |= CELL_NEW_RASTERIZER;
+}
+
+
+void
+cell_delete_rasterizer_state(struct pipe_context *pipe, void *rasterizer)
+{
+   FREE(rasterizer);
+}
diff --git a/src/gallium/drivers/cell/ppu/cell_state_sampler.c b/src/gallium/drivers/cell/ppu/cell_state_sampler.c
new file mode 100644
index 0000000000..ade6cc8338
--- /dev/null
+++ b/src/gallium/drivers/cell/ppu/cell_state_sampler.c
@@ -0,0 +1,84 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+/* Authors:
+ *  Brian Paul
+ */
+
+#include "pipe/p_util.h"
+#include "pipe/draw/draw_context.h"
+#include "cell_context.h"
+#include "cell_state.h"
+#include "cell_texture.h"
+
+
+void *
+cell_create_sampler_state(struct pipe_context *pipe,
+                          const struct pipe_sampler_state *sampler)
+{
+   return mem_dup(sampler, sizeof(*sampler));
+}
+
+void
+cell_bind_sampler_state(struct pipe_context *pipe,
+                            unsigned unit, void *sampler)
+{
+   struct cell_context *cell = cell_context(pipe);
+
+   draw_flush(cell->draw);
+
+   assert(unit < PIPE_MAX_SAMPLERS);
+   cell->sampler[unit] = (struct pipe_sampler_state *)sampler;
+
+   cell->dirty |= CELL_NEW_SAMPLER;
+}
+
+
+void
+cell_delete_sampler_state(struct pipe_context *pipe,
+                              void *sampler)
+{
+   FREE( sampler );
+}
+
+
+
+void
+cell_set_sampler_texture(struct pipe_context *pipe,
+                         unsigned sampler,
+                         struct pipe_texture *texture)
+{
+   struct cell_context *cell = cell_context(pipe);
+
+   draw_flush(cell->draw);
+
+   cell->texture[sampler] = texture;
+
+   cell_update_texture_mapping(cell);
+
+   cell->dirty |= CELL_NEW_TEXTURE;
+}
diff --git a/src/gallium/drivers/cell/ppu/cell_state_surface.c b/src/gallium/drivers/cell/ppu/cell_state_surface.c
new file mode 100644
index 0000000000..287610b76b
--- /dev/null
+++ b/src/gallium/drivers/cell/ppu/cell_state_surface.c
@@ -0,0 +1,71 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+
+#include "pipe/p_inlines.h"
+#include "cell_context.h"
+#include "cell_state.h"
+
+
+void
+cell_set_framebuffer_state(struct pipe_context *pipe,
+                           const struct pipe_framebuffer_state *fb)
+{
+   struct cell_context *cell = cell_context(pipe);
+
+   if (1 /*memcmp(&cell->framebuffer, fb, sizeof(*fb))*/) {
+      struct pipe_surface *csurf = fb->cbufs[0];
+      struct pipe_surface *zsurf = fb->zsbuf;
+      uint i;
+
+      /* unmap old surfaces */
+      for (i = 0; i < PIPE_MAX_COLOR_BUFS; i++) {
+         if (cell->framebuffer.cbufs[i] && cell->cbuf_map[i]) {
+            pipe_surface_unmap(cell->framebuffer.cbufs[i]);
+            cell->cbuf_map[i] = NULL;
+         }
+      }
+
+      if (cell->framebuffer.zsbuf && cell->zsbuf_map) {
+         pipe_surface_unmap(cell->framebuffer.zsbuf);
+         cell->zsbuf_map = NULL;
+      }
+
+      /* update my state */
+      cell->framebuffer = *fb;
+
+      /* map new surfaces */
+      if (csurf)
+         cell->cbuf_map[0] = pipe_surface_map(csurf);
+
+      if (zsurf)
+         cell->zsbuf_map = pipe_surface_map(zsurf);
+
+      cell->dirty |= CELL_NEW_FRAMEBUFFER;
+   }
+}
+
diff --git a/src/gallium/drivers/cell/ppu/cell_state_vertex.c b/src/gallium/drivers/cell/ppu/cell_state_vertex.c
new file mode 100644
index 0000000000..0f01e920f9
--- /dev/null
+++ b/src/gallium/drivers/cell/ppu/cell_state_vertex.c
@@ -0,0 +1,63 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+/* Authors:  Keith Whitwell <keith@tungstengraphics.com>
+ */
+
+
+#include "cell_context.h"
+#include "cell_state.h"
+
+#include "pipe/draw/draw_context.h"
+
+
+void
+cell_set_vertex_element(struct pipe_context *pipe,
+                            unsigned index,
+                            const struct pipe_vertex_element *attrib)
+{
+   struct cell_context *cell = cell_context(pipe);
+   assert(index < PIPE_ATTRIB_MAX);
+   cell->vertex_element[index] = *attrib; /* struct copy */
+   cell->dirty |= CELL_NEW_VERTEX;
+
+   draw_set_vertex_element(cell->draw, index, attrib);
+}
+
+
+void
+cell_set_vertex_buffer(struct pipe_context *pipe,
+                           unsigned index,
+                           const struct pipe_vertex_buffer *buffer)
+{
+   struct cell_context *cell = cell_context(pipe);
+   assert(index < PIPE_ATTRIB_MAX);
+   cell->vertex_buffer[index] = *buffer; /* struct copy */
+   cell->dirty |= CELL_NEW_VERTEX;
+
+   draw_set_vertex_buffer(cell->draw, index, buffer);
+}
diff --git a/src/gallium/drivers/cell/ppu/cell_surface.c b/src/gallium/drivers/cell/ppu/cell_surface.c
new file mode 100644
index 0000000000..fca93e4742
--- /dev/null
+++ b/src/gallium/drivers/cell/ppu/cell_surface.c
@@ -0,0 +1,179 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+#include "pipe/p_defines.h"
+#include "pipe/p_util.h"
+#include "pipe/p_inlines.h"
+#include "pipe/p_winsys.h"
+#include "pipe/util/p_tile.h"
+#include "cell_context.h"
+#include "cell_surface.h"
+
+
+/* Upload data to a rectangular sub-region.  Lots of choices how to do this:
+ *
+ * - memcpy by span to current destination
+ * - upload data as new buffer and blit
+ *
+ * Currently always memcpy.
+ */
+static void
+cell_surface_data(struct pipe_context *pipe,
+                  struct pipe_surface *dst,
+                  unsigned dstx, unsigned dsty,
+                  const void *src, unsigned src_pitch,
+                  unsigned srcx, unsigned srcy,
+                  unsigned width, unsigned height)
+{
+   pipe_copy_rect(pipe_surface_map(dst),
+                  dst->cpp,
+                  dst->pitch,
+                  dstx, dsty, width, height, src, src_pitch, srcx, srcy);
+
+   pipe_surface_unmap(dst);
+}
+
+
+static void
+cell_surface_copy(struct pipe_context *pipe,
+                  unsigned do_flip,
+                  struct pipe_surface *dst,
+                  unsigned dstx, unsigned dsty,
+                  struct pipe_surface *src,
+                  unsigned srcx, unsigned srcy,
+                  unsigned width, unsigned height)
+{
+   assert( dst->cpp == src->cpp );
+
+   pipe_copy_rect(pipe_surface_map(dst),
+                  dst->cpp,
+                  dst->pitch,
+                  dstx, dsty,
+                  width, height,
+                  pipe_surface_map(src),
+                  do_flip ? -src->pitch : src->pitch,
+                  srcx, do_flip ? 1 - srcy - height : srcy);
+
+   pipe_surface_unmap(src);
+   pipe_surface_unmap(dst);
+}
+
+
+static void *
+get_pointer(struct pipe_surface *dst, void *dst_map, unsigned x, unsigned y)
+{
+   return (char *)dst_map + (y * dst->pitch + x) * dst->cpp;
+}
+
+
+#define UBYTE_TO_USHORT(B) ((B) | ((B) << 8))
+
+
+/**
+ * Fill a rectangular sub-region.  Need better logic about when to
+ * push buffers into AGP - will currently do so whenever possible.
+ */
+static void
+cell_surface_fill(struct pipe_context *pipe,
+                  struct pipe_surface *dst,
+                  unsigned dstx, unsigned dsty,
+                  unsigned width, unsigned height, unsigned value)
+{
+   unsigned i, j;
+   void *dst_map = pipe_surface_map(dst);
+
+   assert(dst->pitch > 0);
+   assert(width <= dst->pitch);
+
+   switch (dst->cpp) {
+   case 1:
+      {
+	 ubyte *row = get_pointer(dst, dst_map, dstx, dsty);
+         for (i = 0; i < height; i++) {
+            memset(row, value, width);
+	 row += dst->pitch;
+         }
+      }
+      break;
+   case 2:
+      {
+         ushort *row = get_pointer(dst, dst_map, dstx, dsty);
+         for (i = 0; i < height; i++) {
+            for (j = 0; j < width; j++)
+               row[j] = (ushort) value;
+            row += dst->pitch;
+         }
+      }
+      break;
+   case 4:
+      {
+         unsigned *row = get_pointer(dst, dst_map, dstx, dsty);
+         for (i = 0; i < height; i++) {
+            for (j = 0; j < width; j++)
+               row[j] = value;
+            row += dst->pitch;
+         }
+      }
+      break;
+   case 8:
+      {
+         /* expand the 4-byte clear value to an 8-byte value */
+         ushort *row = (ushort *) get_pointer(dst, dst_map, dstx, dsty);
+         ushort val0 = UBYTE_TO_USHORT((value >>  0) & 0xff);
+         ushort val1 = UBYTE_TO_USHORT((value >>  8) & 0xff);
+         ushort val2 = UBYTE_TO_USHORT((value >> 16) & 0xff);
+         ushort val3 = UBYTE_TO_USHORT((value >> 24) & 0xff);
+         val0 = (val0 << 8) | val0;
+         val1 = (val1 << 8) | val1;
+         val2 = (val2 << 8) | val2;
+         val3 = (val3 << 8) | val3;
+         for (i = 0; i < height; i++) {
+            for (j = 0; j < width; j++) {
+               row[j*4+0] = val0;
+               row[j*4+1] = val1;
+               row[j*4+2] = val2;
+               row[j*4+3] = val3;
+            }
+            row += dst->pitch * 4;
+         }
+      }
+      break;
+   default:
+      assert(0);
+      break;
+   }
+
+   pipe_surface_unmap( dst );
+}
+
+
+void
+cell_init_surface_functions(struct cell_context *cell)
+{
+   cell->pipe.surface_copy = cell_surface_copy;
+   cell->pipe.surface_fill = cell_surface_fill;
+}
diff --git a/src/gallium/drivers/cell/ppu/cell_surface.h b/src/gallium/drivers/cell/ppu/cell_surface.h
new file mode 100644
index 0000000000..9e58f32944
--- /dev/null
+++ b/src/gallium/drivers/cell/ppu/cell_surface.h
@@ -0,0 +1,42 @@
+/**************************************************************************
+ * 
+ * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+/* Authors:  Keith Whitwell <keith@tungstengraphics.com>
+ */
+
+#ifndef CELL_SURFACE_H
+#define CELL_SURFACE_H
+
+
+struct cell_context;
+
+
+extern void
+cell_init_surface_functions(struct cell_context *cell);
+
+
+#endif /* SP_SURFACE_H */
diff --git a/src/gallium/drivers/cell/ppu/cell_texture.c b/src/gallium/drivers/cell/ppu/cell_texture.c
new file mode 100644
index 0000000000..c8ef36002f
--- /dev/null
+++ b/src/gallium/drivers/cell/ppu/cell_texture.c
@@ -0,0 +1,252 @@
+/**************************************************************************
+ * 
+ * Copyright 2006 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+ /*
+  * Authors:
+  *   Keith Whitwell <keith@tungstengraphics.com>
+  *   Michel Dänzer <michel@tungstengraphics.com>
+  */
+
+#include "pipe/p_context.h"
+#include "pipe/p_defines.h"
+#include "pipe/p_inlines.h"
+#include "pipe/p_util.h"
+#include "pipe/p_winsys.h"
+
+#include "cell_context.h"
+#include "cell_state.h"
+#include "cell_texture.h"
+
+
+/* Simple, maximally packed layout.
+ */
+
+static unsigned minify( unsigned d )
+{
+   return MAX2(1, d>>1);
+}
+
+
+static void
+cell_texture_layout(struct cell_texture * spt)
+{
+   struct pipe_texture *pt = &spt->base;
+   unsigned level;
+   unsigned width = pt->width[0];
+   unsigned height = pt->height[0];
+   unsigned depth = pt->depth[0];
+
+   spt->buffer_size = 0;
+
+   for ( level = 0 ; level <= pt->last_level ; level++ ) {
+      pt->width[level] = width;
+      pt->height[level] = height;
+      pt->depth[level] = depth;
+
+      spt->level_offset[level] = spt->buffer_size;
+
+      spt->buffer_size += ((pt->compressed) ? MAX2(1, height/4) : height) *
+			  ((pt->target == PIPE_TEXTURE_CUBE) ? 6 : depth) *
+			  width * pt->cpp;
+
+      width  = minify(width);
+      height = minify(height);
+      depth = minify(depth);
+   }
+}
+
+
+struct pipe_texture *
+cell_texture_create(struct pipe_context *pipe, const struct pipe_texture *templat)
+{
+   struct cell_texture *spt = CALLOC_STRUCT(cell_texture);
+   if (!spt)
+      return NULL;
+
+   spt->base = *templat;
+
+   cell_texture_layout(spt);
+
+   spt->buffer = pipe->winsys->buffer_create(pipe->winsys, 32,
+                                             PIPE_BUFFER_USAGE_PIXEL,
+                                             spt->buffer_size);
+
+   if (!spt->buffer) {
+      FREE(spt);
+      return NULL;
+   }
+
+   return &spt->base;
+}
+
+
+void
+cell_texture_release(struct pipe_context *pipe, struct pipe_texture **pt)
+{
+   if (!*pt)
+      return;
+
+   /*
+   DBG("%s %p refcount will be %d\n",
+       __FUNCTION__, (void *) *pt, (*pt)->refcount - 1);
+   */
+   if (--(*pt)->refcount <= 0) {
+      struct cell_texture *spt = cell_texture(*pt);
+
+      /*
+      DBG("%s deleting %p\n", __FUNCTION__, (void *) spt);
+      */
+
+      pipe_buffer_reference(pipe->winsys, &spt->buffer, NULL);
+
+      FREE(spt);
+   }
+   *pt = NULL;
+}
+
+
+/**
+ * Called via pipe->get_tex_surface()
+ */
+struct pipe_surface *
+cell_get_tex_surface(struct pipe_context *pipe,
+                         struct pipe_texture *pt,
+                         unsigned face, unsigned level, unsigned zslice)
+{
+   struct cell_texture *spt = cell_texture(pt);
+   struct pipe_surface *ps;
+
+   ps = pipe->winsys->surface_alloc(pipe->winsys);
+   if (ps) {
+      assert(ps->refcount);
+      assert(ps->winsys);
+      pipe_buffer_reference(pipe->winsys, &ps->buffer, spt->buffer);
+      ps->format = pt->format;
+      ps->cpp = pt->cpp;
+      ps->width = pt->width[level];
+      ps->height = pt->height[level];
+      ps->pitch = ps->width;
+      ps->offset = spt->level_offset[level];
+
+      if (pt->target == PIPE_TEXTURE_CUBE || pt->target == PIPE_TEXTURE_3D) {
+	 ps->offset += ((pt->target == PIPE_TEXTURE_CUBE) ? face : zslice) *
+		       (pt->compressed ? ps->height/4 : ps->height) *
+		       ps->width * ps->cpp;
+      } else {
+	 assert(face == 0);
+	 assert(zslice == 0);
+      }
+   }
+   return ps;
+}
+
+
+
+static void
+tile_copy_data(uint w, uint h, uint tile_size, uint *dst, const uint *src)
+{
+   const uint tile_size2 = tile_size * tile_size;
+   const uint h_t = h / tile_size, w_t = w / tile_size;
+
+   uint it, jt;  /* tile counters */
+   uint i, j;    /* intra-tile counters */
+
+   for (it = 0; it < h_t; it++) {
+      for (jt = 0; jt < w_t; jt++) {
+         /* fill in tile (i, j) */
+         uint *tdst = dst + (it * w_t + jt) * tile_size2;
+         for (i = 0; i < tile_size; i++) {
+            for (j = 0; j < tile_size; j++) {
+               const uint srci = it * tile_size + i;
+               const uint srcj = jt * tile_size + j;
+               *tdst++ = src[srci * h + srcj];
+            }
+         }
+      }
+   }
+}
+
+
+
+/**
+ * Convert linear texture image data to tiled format for SPU usage.
+ */
+static void
+cell_tile_texture(struct cell_context *cell,
+                  struct cell_texture *texture)
+{
+   uint face = 0, level = 0, zslice = 0;
+   struct pipe_surface *surf;
+   const uint w = texture->base.width[0], h = texture->base.height[0];
+   const uint *src;
+
+   /* temporary restrictions: */
+   assert(w >= TILE_SIZE);
+   assert(h >= TILE_SIZE);
+   assert(w % TILE_SIZE == 0);
+   assert(h % TILE_SIZE == 0);
+
+   surf = cell_get_tex_surface(&cell->pipe, &texture->base, face, level, zslice);
+   ASSERT(surf);
+
+   src = (const uint *) pipe_surface_map(surf);
+
+   if (texture->tiled_data) {
+      align_free(texture->tiled_data);
+   }
+   texture->tiled_data = align_malloc(w * h * 4, 16);
+
+   tile_copy_data(w, h, TILE_SIZE, texture->tiled_data, src);
+
+   pipe_surface_unmap(surf);
+
+   pipe_surface_reference(&surf, NULL);
+}
+
+
+
+void
+cell_update_texture_mapping(struct cell_context *cell)
+{
+   uint face = 0, level = 0, zslice = 0;
+
+   if (cell->texture[0])
+      cell_tile_texture(cell, cell->texture[0]);
+#if 0
+   if (cell->tex_surf && cell->tex_map) {
+      pipe_surface_unmap(cell->tex_surf);
+      cell->tex_map = NULL;
+   }
+
+   /* XXX free old surface */
+
+   cell->tex_surf = cell_get_tex_surface(&cell->pipe,
+                                         &cell->texture[0]->base,
+                                         face, level, zslice);
+
+   cell->tex_map = pipe_surface_map(cell->tex_surf);
+#endif
+}
diff --git a/src/gallium/drivers/cell/ppu/cell_texture.h b/src/gallium/drivers/cell/ppu/cell_texture.h
new file mode 100644
index 0000000000..0264fed88e
--- /dev/null
+++ b/src/gallium/drivers/cell/ppu/cell_texture.h
@@ -0,0 +1,80 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+#ifndef CELL_TEXTURE_H
+#define CELL_TEXTURE_H
+
+
+struct pipe_context;
+struct pipe_texture;
+
+
+/**
+ * Subclass of pipe_texture
+ */
+struct cell_texture
+{
+   struct pipe_texture base;
+
+   unsigned long level_offset[PIPE_MAX_TEXTURE_LEVELS];
+
+   /* The data is held here:
+    */
+   struct pipe_buffer *buffer;
+   unsigned long buffer_size;
+
+   void *tiled_data;  /* XXX this may be temporary */ /*ALIGN16*/
+};
+
+
+/** cast wrapper */
+static INLINE struct cell_texture *
+cell_texture(struct pipe_texture *pt)
+{
+   return (struct cell_texture *) pt;
+}
+
+
+
+extern struct pipe_texture *
+cell_texture_create(struct pipe_context *pipe,
+                    const struct pipe_texture *templat);
+
+extern void
+cell_texture_release(struct pipe_context *pipe, struct pipe_texture **pt);
+
+extern struct pipe_surface *
+cell_get_tex_surface(struct pipe_context *pipe,
+                     struct pipe_texture *pt,
+                     unsigned face, unsigned level, unsigned zslice);
+
+
+extern void
+cell_update_texture_mapping(struct cell_context *cell);
+
+
+#endif /* CELL_TEXTURE */
diff --git a/src/gallium/drivers/cell/ppu/cell_vbuf.c b/src/gallium/drivers/cell/ppu/cell_vbuf.c
new file mode 100644
index 0000000000..e9fafe492e
--- /dev/null
+++ b/src/gallium/drivers/cell/ppu/cell_vbuf.c
@@ -0,0 +1,294 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+/**
+ * Authors
+ *  Brian Paul
+ */
+
+
+#include "cell_batch.h"
+#include "cell_context.h"
+#include "cell_flush.h"
+#include "cell_spu.h"
+#include "cell_vbuf.h"
+#include "pipe/draw/draw_vbuf.h"
+
+
+/** Allow vertex data to be inlined after RENDER command */
+#define ALLOW_INLINE_VERTS 1
+
+
+/**
+ * Subclass of vbuf_render because we need a cell_context pointer in
+ * a few places.
+ */
+struct cell_vbuf_render
+{
+   struct vbuf_render base;
+   struct cell_context *cell;
+   uint prim;            /**< PIPE_PRIM_x */
+   uint vertex_size;     /**< in bytes */
+   void *vertex_buffer;  /**< just for debug, really */
+   uint vertex_buf;      /**< in [0, CELL_NUM_BUFFERS-1] */
+};
+
+
+/** cast wrapper */
+static struct cell_vbuf_render *
+cell_vbuf_render(struct vbuf_render *vbr)
+{
+   return (struct cell_vbuf_render *) vbr;
+}
+
+
+
+static const struct vertex_info *
+cell_vbuf_get_vertex_info(struct vbuf_render *vbr)
+{
+   struct cell_vbuf_render *cvbr = cell_vbuf_render(vbr);
+   return &cvbr->cell->vertex_info;
+}
+
+
+static void *
+cell_vbuf_allocate_vertices(struct vbuf_render *vbr,
+                            ushort vertex_size, ushort nr_vertices)
+{
+   struct cell_vbuf_render *cvbr = cell_vbuf_render(vbr);
+   /*printf("Alloc verts %u * %u\n", vertex_size, nr_vertices);*/
+
+   assert(cvbr->vertex_buf == ~0);
+   cvbr->vertex_buf = cell_get_empty_buffer(cvbr->cell);
+   cvbr->vertex_buffer = cvbr->cell->buffer[cvbr->vertex_buf];
+   cvbr->vertex_size = vertex_size;
+   return cvbr->vertex_buffer;
+}
+
+
+static void
+cell_vbuf_release_vertices(struct vbuf_render *vbr, void *vertices, 
+                           unsigned vertex_size, unsigned vertices_used)
+{
+   struct cell_vbuf_render *cvbr = cell_vbuf_render(vbr);
+   struct cell_context *cell = cvbr->cell;
+
+   /*
+   printf("%s vertex_buf = %u  count = %u\n",
+          __FUNCTION__, cvbr->vertex_buf, vertices_used);
+   */
+
+   /* Tell SPUs they can release the vert buf */
+   if (cvbr->vertex_buf != ~0U) {
+      struct cell_command_release_verts *release
+         = (struct cell_command_release_verts *)
+         cell_batch_alloc(cell, sizeof(struct cell_command_release_verts));
+      release->opcode = CELL_CMD_RELEASE_VERTS;
+      release->vertex_buf = cvbr->vertex_buf;
+   }
+
+   cvbr->vertex_buf = ~0;
+   cell_flush_int(&cell->pipe, 0x0);
+
+   assert(vertices == cvbr->vertex_buffer);
+   cvbr->vertex_buffer = NULL;
+}
+
+
+
+static void
+cell_vbuf_set_primitive(struct vbuf_render *vbr, unsigned prim)
+{
+   struct cell_vbuf_render *cvbr = cell_vbuf_render(vbr);
+   cvbr->prim = prim;
+   /*printf("cell_set_prim %u\n", prim);*/
+}
+
+
+static void
+cell_vbuf_draw(struct vbuf_render *vbr,
+	       const ushort *indices,
+               uint nr_indices)
+{
+   struct cell_vbuf_render *cvbr = cell_vbuf_render(vbr);
+   struct cell_context *cell = cvbr->cell;
+   float xmin, ymin, xmax, ymax;
+   uint i;
+   uint nr_vertices = 0, min_index = ~0;
+   const void *vertices = cvbr->vertex_buffer;
+   const uint vertex_size = cvbr->vertex_size;
+
+   for (i = 0; i < nr_indices; i++) {
+      if (indices[i] > nr_vertices)
+         nr_vertices = indices[i];
+      if (indices[i] < min_index)
+         min_index = indices[i];
+   }
+   nr_vertices++;
+
+#if 0
+   /*if (min_index > 0)*/
+      printf("%s min_index = %u\n", __FUNCTION__, min_index);
+#endif
+
+#if 0
+   printf("cell_vbuf_draw() nr_indices = %u nr_verts = %u\n",
+          nr_indices, nr_vertices);
+   printf("  ");
+   for (i = 0; i < nr_indices; i += 3) {
+      printf("%u %u %u, ", indices[i+0], indices[i+1], indices[i+2]);
+   }
+   printf("\n");
+#elif 0
+   printf("cell_vbuf_draw() nr_indices = %u nr_verts = %u  indexes = [%u %u %u ...]\n",
+          nr_indices, nr_vertices,
+          indices[0], indices[1], indices[2]);
+   printf("ind space = %u, vert space = %u, space = %u\n",
+          nr_indices * 2,
+          nr_vertices * 4 * cell->vertex_info.size,
+          cell_batch_free_space(cell));
+#endif
+
+   /* compute x/y bounding box */
+   xmin = ymin = 1e50;
+   xmax = ymax = -1e50;
+   for (i = min_index; i < nr_vertices; i++) {
+      const float *v = (float *) ((ubyte *) vertices + i * vertex_size);
+      if (v[0] < xmin)
+         xmin = v[0];
+      if (v[0] > xmax)
+         xmax = v[0];
+      if (v[1] < ymin)
+         ymin = v[1];
+      if (v[1] > ymax)
+         ymax = v[1];
+   }
+#if 0
+   printf("PPU Bounds %g, %g .. %g, %g\n", xmin, ymin, xmax, ymax);
+   fflush(stdout);
+#endif
+
+   if (cvbr->prim != PIPE_PRIM_TRIANGLES)
+      return; /* only render tris for now */
+
+   /* build/insert batch RENDER command */
+   {
+      const uint index_bytes = ROUNDUP8(nr_indices * 2);
+      const uint vertex_bytes = nr_vertices * 4 * cell->vertex_info.size;
+      const uint batch_size = sizeof(struct cell_command_render) + index_bytes;
+
+      struct cell_command_render *render
+         = (struct cell_command_render *)
+         cell_batch_alloc(cell, batch_size);
+
+      render->opcode = CELL_CMD_RENDER;
+      render->prim_type = cvbr->prim;
+
+      render->num_indexes = nr_indices;
+      render->min_index = min_index;
+
+      /* append indices after render command */
+      memcpy(render + 1, indices, nr_indices * 2);
+
+      /* if there's room, append vertices after the indices, else leave
+       * vertices in the original/separate buffer.
+       */
+      render->vertex_size = 4 * cell->vertex_info.size;
+      render->num_verts = nr_vertices;
+      if (ALLOW_INLINE_VERTS &&
+          min_index == 0 &&
+          vertex_bytes + 16 <= cell_batch_free_space(cell)) {
+         /* vertex data inlined, after indices, at 16-byte boundary */
+         void *dst = cell_batch_alloc_aligned(cell, vertex_bytes, 16);
+         memcpy(dst, vertices, vertex_bytes);
+         render->inline_verts = TRUE;
+         render->vertex_buf = ~0;
+      }
+      else {
+         /* vertex data in separate buffer */
+         render->inline_verts = FALSE;
+         ASSERT(cvbr->vertex_buf >= 0);
+         render->vertex_buf = cvbr->vertex_buf;
+      }
+
+      render->xmin = xmin;
+      render->ymin = ymin;
+      render->xmax = xmax;
+      render->ymax = ymax;
+   }
+
+#if 0
+   /* helpful for debug */
+   cell_flush_int(&cell->pipe, PIPE_FLUSH_WAIT);
+#endif
+}
+
+
+static void
+cell_vbuf_destroy(struct vbuf_render *vbr)
+{
+   struct cell_vbuf_render *cvbr = cell_vbuf_render(vbr);
+   cvbr->cell->vbuf_render = NULL;
+   FREE(cvbr);
+}
+
+
+/**
+ * Initialize the post-transform vertex buffer information for the given
+ * context.
+ */
+void
+cell_init_vbuf(struct cell_context *cell)
+{
+   assert(cell->draw);
+
+   cell->vbuf_render = CALLOC_STRUCT(cell_vbuf_render);
+
+   /* The max number of indexes is what can fix into a batch buffer,
+    * minus the render and release-verts commands.
+    */
+   cell->vbuf_render->base.max_indices
+      = (CELL_BUFFER_SIZE
+         - sizeof(struct cell_command_render)
+         - sizeof(struct cell_command_release_verts))
+      / sizeof(ushort);
+   cell->vbuf_render->base.max_vertex_buffer_bytes = CELL_BUFFER_SIZE;
+
+   cell->vbuf_render->base.get_vertex_info = cell_vbuf_get_vertex_info;
+   cell->vbuf_render->base.allocate_vertices = cell_vbuf_allocate_vertices;
+   cell->vbuf_render->base.set_primitive = cell_vbuf_set_primitive;
+   cell->vbuf_render->base.draw = cell_vbuf_draw;
+   cell->vbuf_render->base.release_vertices = cell_vbuf_release_vertices;
+   cell->vbuf_render->base.destroy = cell_vbuf_destroy;
+
+   cell->vbuf_render->cell = cell;
+#if 1
+   cell->vbuf_render->vertex_buf = ~0;
+#endif
+
+   cell->vbuf = draw_vbuf_stage(cell->draw, &cell->vbuf_render->base);
+}
diff --git a/src/gallium/drivers/cell/ppu/cell_vbuf.h b/src/gallium/drivers/cell/ppu/cell_vbuf.h
new file mode 100644
index 0000000000..d265cbf770
--- /dev/null
+++ b/src/gallium/drivers/cell/ppu/cell_vbuf.h
@@ -0,0 +1,38 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+#ifndef CELL_VBUF_H
+#define CELL_VBUF_H
+
+
+struct cell_context;
+
+extern void
+cell_init_vbuf(struct cell_context *cell);
+
+
+#endif /* CELL_VBUF_H */
diff --git a/src/gallium/drivers/cell/ppu/cell_vertex_shader.c b/src/gallium/drivers/cell/ppu/cell_vertex_shader.c
new file mode 100644
index 0000000000..80dd500b34
--- /dev/null
+++ b/src/gallium/drivers/cell/ppu/cell_vertex_shader.c
@@ -0,0 +1,120 @@
+/*
+ * (C) Copyright IBM Corporation 2008
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * on the rights to use, copy, modify, merge, publish, distribute, sub
+ * license, and/or sell copies of the Software, and to permit persons to whom
+ * the Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the next
+ * paragraph) shall be included in all copies or substantial portions of the
+ * Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.  IN NO EVENT SHALL
+ * AUTHORS, COPYRIGHT HOLDERS, AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
+ * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
+ * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
+ * USE OR OTHER DEALINGS IN THE SOFTWARE.
+ */
+
+/**
+ * \file cell_vertex_shader.c
+ * Vertex shader interface routines for Cell.
+ *
+ * \author Ian Romanick <idr@us.ibm.com>
+ */
+
+#include "pipe/p_defines.h"
+#include "pipe/p_context.h"
+#include "pipe/p_winsys.h"
+
+#include "cell_context.h"
+#include "cell_draw_arrays.h"
+#include "cell_spu.h"
+#include "cell_batch.h"
+
+#include "pipe/cell/common.h"
+#include "pipe/draw/draw_context.h"
+#include "pipe/draw/draw_private.h"
+
+/**
+ * Run the vertex shader on all vertices in the vertex queue.
+ * Called by the draw module when the vertx cache needs to be flushed.
+ */
+void
+cell_vertex_shader_queue_flush(struct draw_context *draw)
+{
+   struct cell_context *const cell =
+       (struct cell_context *) draw->driver_private;
+   struct cell_command_vs *const vs = &cell_global.command[0].vs;
+   uint64_t *batch;
+   struct cell_array_info *array_info;
+   unsigned i, j;
+
+   assert(draw->vs.queue_nr != 0);
+
+   /* XXX: do this on statechange: 
+    */
+   draw_update_vertex_fetch(draw);
+
+   for (i = 0; i < draw->vertex_fetch.nr_attrs; i++) {
+      batch = cell_batch_alloc(cell, sizeof(batch[0]) + sizeof(*array_info));
+
+      batch[0] = CELL_CMD_STATE_VS_ARRAY_INFO;
+
+      array_info = (struct cell_array_info *) &batch[1];
+      assert(draw->vertex_fetch.src_ptr[i] != NULL);
+      array_info->base = (uintptr_t) draw->vertex_fetch.src_ptr[i];
+      array_info->attr = i;
+      array_info->pitch = draw->vertex_fetch.pitch[i];
+      array_info->format = draw->vertex_element[i].src_format;
+   }
+
+   batch = cell_batch_alloc(cell, sizeof(batch[0])
+                            + sizeof(struct pipe_viewport_state));
+   batch[0] = CELL_CMD_STATE_VIEWPORT;
+   (void) memcpy(&batch[1], &draw->viewport,
+                 sizeof(struct pipe_viewport_state));
+
+   cell_batch_flush(cell);
+
+   vs->opcode = CELL_CMD_VS_EXECUTE;
+   vs->shader.num_outputs = draw->num_vs_outputs;
+   vs->shader.declarations = (uintptr_t) draw->machine.Declarations;
+   vs->shader.num_declarations = draw->machine.NumDeclarations;
+   vs->shader.instructions = (uintptr_t) draw->machine.Instructions;
+   vs->shader.num_instructions = draw->machine.NumInstructions;
+   vs->shader.uniforms = (uintptr_t) draw->user.constants;
+   vs->shader.immediates = (uintptr_t) draw->machine.Imms;
+   vs->shader.num_immediates = draw->machine.ImmLimit / 4;
+   vs->nr_attrs = draw->vertex_fetch.nr_attrs;
+
+   (void) memcpy(vs->plane, draw->plane, sizeof(draw->plane));
+   vs->nr_planes = draw->nr_planes;
+
+   for (i = 0; i < draw->vs.queue_nr; i += SPU_VERTS_PER_BATCH) {
+      const unsigned n = MIN2(SPU_VERTS_PER_BATCH, draw->vs.queue_nr - i);
+
+      for (j = 0; j < n; j++) {
+         vs->elts[j] = draw->vs.queue[i + j].elt;
+         vs->vOut[j] = (uintptr_t) draw->vs.queue[i + j].dest;
+      }
+
+      for (/* empty */; j < SPU_VERTS_PER_BATCH; j++) {
+         vs->elts[j] = vs->elts[0];
+         vs->vOut[j] = vs->vOut[0];
+      }
+
+      vs->num_elts = n;
+      send_mbox_message(cell_global.spe_contexts[0], CELL_CMD_VS_EXECUTE);
+
+      cell_flush_int(& cell->pipe, PIPE_FLUSH_WAIT);
+   }
+
+   draw->vs.queue_nr = 0;
+}
diff --git a/src/gallium/drivers/cell/ppu/cell_winsys.c b/src/gallium/drivers/cell/ppu/cell_winsys.c
new file mode 100644
index 0000000000..ebabce3c8f
--- /dev/null
+++ b/src/gallium/drivers/cell/ppu/cell_winsys.c
@@ -0,0 +1,40 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+
+#include "pipe/p_util.h"
+#include "cell_winsys.h"
+
+
+struct cell_winsys *
+cell_get_winsys(uint format)
+{
+   struct cell_winsys *cws = CALLOC_STRUCT(cell_winsys);
+   if (cws)
+      cws->preferredFormat = format;
+   return cws;
+}
diff --git a/src/gallium/drivers/cell/ppu/cell_winsys.h b/src/gallium/drivers/cell/ppu/cell_winsys.h
new file mode 100644
index 0000000000..ae2af5696b
--- /dev/null
+++ b/src/gallium/drivers/cell/ppu/cell_winsys.h
@@ -0,0 +1,50 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+
+#ifndef CELL_WINSYS_H
+#define CELL_WINSYS_H
+
+#include "pipe/p_compiler.h"
+
+
+/**
+ * Very simple winsys at this time.
+ * Will probably eventually add SPU control info.
+ */
+struct cell_winsys
+{
+   uint preferredFormat;
+};
+
+
+extern struct cell_winsys *
+cell_get_winsys(uint format);
+
+
+
+#endif
diff --git a/src/gallium/drivers/cell/spu/Makefile b/src/gallium/drivers/cell/spu/Makefile
new file mode 100644
index 0000000000..f202971d73
--- /dev/null
+++ b/src/gallium/drivers/cell/spu/Makefile
@@ -0,0 +1,72 @@
+# Gallium3D Cell driver: SPU code
+
+# This makefile builds the g3d_spu.a file that's linked into the
+# PPU code/library.
+
+
+TOP = ../../../../..
+include $(TOP)/configs/linux-cell
+
+
+PROG = g3d
+
+PROG_SPU = $(PROG)_spu
+PROG_SPU_A = $(PROG)_spu.a
+PROG_SPU_EMBED_O = $(PROG)_spu-embed.o
+
+
+SOURCES = \
+	spu_main.c \
+	spu_blend.c \
+	spu_render.c \
+	spu_texture.c \
+	spu_tile.c \
+	spu_tri.c \
+	spu_exec.c \
+	spu_util.c \
+	spu_vertex_fetch.c \
+	spu_vertex_shader.c
+
+SPU_OBJECTS = $(SOURCES:.c=.o) \
+
+SPU_ASM_OUT = $(SOURCES:.c=.s) \
+
+INCLUDE_DIRS = -I$(TOP)/src/mesa
+
+
+.c.o:
+	$(SPU_CC) $(SPU_CFLAGS) -c $<
+
+.c.s:
+	$(SPU_CC) $(SPU_CFLAGS) -S $<
+
+
+# The .a file will be linked into the main/PPU executable
+default: $(PROG_SPU_A)
+
+$(PROG_SPU_A): $(PROG_SPU_EMBED_O)
+	$(SPU_AR) $(SPU_AR_FLAGS) $(PROG_SPU_A) $(PROG_SPU_EMBED_O)
+
+$(PROG_SPU_EMBED_O): $(PROG_SPU)
+	$(SPU_EMBED) $(SPU_EMBED_FLAGS) $(PROG_SPU) $(PROG_SPU) $(PROG_SPU_EMBED_O)
+
+$(PROG_SPU): $(SPU_OBJECTS)
+	$(SPU_CC) -o $(PROG_SPU) $(SPU_OBJECTS) $(SPU_LFLAGS)
+
+
+
+asmfiles: $(SPU_ASM_OUT)
+
+
+clean:
+	rm -f *~ *.o *.a *.d *.s $(PROG_SPU)
+
+
+
+depend: $(SOURCES)
+	rm -f depend
+	touch depend
+	$(MKDEP) $(MKDEP_OPTIONS) $(INCLUDE_DIRS) $(SOURCES) 2> /dev/null
+
+include depend
+
diff --git a/src/gallium/drivers/cell/spu/spu_blend.c b/src/gallium/drivers/cell/spu/spu_blend.c
new file mode 100644
index 0000000000..23ec0eeb45
--- /dev/null
+++ b/src/gallium/drivers/cell/spu/spu_blend.c
@@ -0,0 +1,62 @@
+/**************************************************************************
+ * 
+ * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+
+#include "spu_main.h"
+#include "spu_blend.h"
+#include "spu_colorpack.h"
+
+
+void
+blend_quad(uint itx, uint ity, vector float colors[4])
+{
+   /* simple SRC_ALPHA, ONE_MINUS_SRC_ALPHA blending */
+   vector float fbc00 = spu_unpack_color(spu.ctile.ui[ity][itx]);
+   vector float fbc01 = spu_unpack_color(spu.ctile.ui[ity][itx+1]);
+   vector float fbc10 = spu_unpack_color(spu.ctile.ui[ity+1][itx]);
+   vector float fbc11 = spu_unpack_color(spu.ctile.ui[ity+1][itx+1]);
+
+   vector float alpha00 = spu_splats(spu_extract(colors[0], 3));
+   vector float alpha01 = spu_splats(spu_extract(colors[1], 3));
+   vector float alpha10 = spu_splats(spu_extract(colors[2], 3));
+   vector float alpha11 = spu_splats(spu_extract(colors[3], 3));
+
+   vector float one_minus_alpha00 = spu_sub(spu_splats(1.0f), alpha00);
+   vector float one_minus_alpha01 = spu_sub(spu_splats(1.0f), alpha01);
+   vector float one_minus_alpha10 = spu_sub(spu_splats(1.0f), alpha10);
+   vector float one_minus_alpha11 = spu_sub(spu_splats(1.0f), alpha11);
+
+   colors[0] = spu_add(spu_mul(colors[0], alpha00),
+                       spu_mul(fbc00, one_minus_alpha00));
+   colors[1] = spu_add(spu_mul(colors[1], alpha01),
+                       spu_mul(fbc01, one_minus_alpha01));
+   colors[2] = spu_add(spu_mul(colors[2], alpha10),
+                       spu_mul(fbc10, one_minus_alpha10));
+   colors[3] = spu_add(spu_mul(colors[3], alpha11),
+                       spu_mul(fbc11, one_minus_alpha11));
+}
+
diff --git a/src/gallium/drivers/cell/spu/spu_blend.h b/src/gallium/drivers/cell/spu/spu_blend.h
new file mode 100644
index 0000000000..2b594b578b
--- /dev/null
+++ b/src/gallium/drivers/cell/spu/spu_blend.h
@@ -0,0 +1,37 @@
+/**************************************************************************
+ * 
+ * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+
+#ifndef SPU_BLEND_H
+#define SPU_BLEND_H
+
+
+extern void
+blend_quad(uint itx, uint ity, vector float colors[4]);
+
+
+#endif /* SPU_BLEND_H */
diff --git a/src/gallium/drivers/cell/spu/spu_colorpack.h b/src/gallium/drivers/cell/spu/spu_colorpack.h
new file mode 100644
index 0000000000..e9fee8a3a6
--- /dev/null
+++ b/src/gallium/drivers/cell/spu/spu_colorpack.h
@@ -0,0 +1,110 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+
+
+#ifndef SPU_COLORPACK_H
+#define SPU_COLORPACK_H
+
+
+#include <spu_intrinsics.h>
+
+
+static INLINE unsigned int
+spu_pack_R8G8B8A8(vector float rgba)
+{
+  vector unsigned int out = spu_convtu(rgba, 32);
+
+  out = spu_shuffle(out, out, ((vector unsigned char) {
+                                  0, 4, 8, 12, 0, 0, 0, 0, 
+                                  0, 0, 0, 0, 0, 0, 0, 0 }) );
+
+  return spu_extract(out, 0);
+}
+
+
+static INLINE unsigned int
+spu_pack_A8R8G8B8(vector float rgba)
+{
+  vector unsigned int out = spu_convtu(rgba, 32);
+  out = spu_shuffle(out, out, ((vector unsigned char) {
+                                  12, 0, 4, 8, 0, 0, 0, 0, 
+                                  0, 0, 0, 0, 0, 0, 0, 0}) );
+  return spu_extract(out, 0);
+}
+
+
+static INLINE unsigned int
+spu_pack_B8G8R8A8(vector float rgba)
+{
+  vector unsigned int out = spu_convtu(rgba, 32);
+  out = spu_shuffle(out, out, ((vector unsigned char) {
+                                  8, 4, 0, 12, 0, 0, 0, 0, 
+                                  0, 0, 0, 0, 0, 0, 0, 0}) );
+  return spu_extract(out, 0);
+}
+
+
+static INLINE unsigned int
+spu_pack_color_shuffle(vector float rgba, vector unsigned char shuffle)
+{
+  vector unsigned int out = spu_convtu(rgba, 32);
+  out = spu_shuffle(out, out, shuffle);
+  return spu_extract(out, 0);
+}
+
+
+static INLINE vector float
+spu_unpack_color(uint color)
+{
+   vector unsigned int color_u4 = spu_splats(color);
+   color_u4 = spu_shuffle(color_u4, color_u4,
+                          ((vector unsigned char) {
+                             0, 0, 0, 0,
+                             5, 5, 5, 5,
+                             10, 10, 10, 10,
+                             15, 15, 15, 15}) );
+   return spu_convtf(color_u4, 32);
+}
+
+
+static INLINE vector float
+spu_unpack_A8R8G8B8(uint color)
+{
+   vector unsigned int color_u4 = spu_splats(color);
+   color_u4 = spu_shuffle(color_u4, color_u4,
+                          ((vector unsigned char) {
+                             5, 5, 5, 5,
+                             10, 10, 10, 10,
+                             15, 15, 15, 15,
+                             0, 0, 0, 0}) );
+
+   return spu_convtf(color_u4, 32);
+}
+
+
+#endif /* SPU_COLORPACK_H */
diff --git a/src/gallium/drivers/cell/spu/spu_exec.c b/src/gallium/drivers/cell/spu/spu_exec.c
new file mode 100644
index 0000000000..e51008b9b3
--- /dev/null
+++ b/src/gallium/drivers/cell/spu/spu_exec.c
@@ -0,0 +1,1948 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+/**
+ * TGSI interpretor/executor.
+ *
+ * Flow control information:
+ *
+ * Since we operate on 'quads' (4 pixels or 4 vertices in parallel)
+ * flow control statements (IF/ELSE/ENDIF, LOOP/ENDLOOP) require special
+ * care since a condition may be true for some quad components but false
+ * for other components.
+ *
+ * We basically execute all statements (even if they're in the part of
+ * an IF/ELSE clause that's "not taken") and use a special mask to
+ * control writing to destination registers.  This is the ExecMask.
+ * See store_dest().
+ *
+ * The ExecMask is computed from three other masks (CondMask, LoopMask and
+ * ContMask) which are controlled by the flow control instructions (namely:
+ * (IF/ELSE/ENDIF, LOOP/ENDLOOP and CONT).
+ *
+ *
+ * Authors:
+ *   Michal Krol
+ *   Brian Paul
+ */
+
+#include <libmisc.h>
+#include <spu_mfcio.h>
+#include <transpose_matrix4x4.h>
+#include <simdmath/ceilf4.h>
+#include <simdmath/cosf4.h>
+#include <simdmath/divf4.h>
+#include <simdmath/floorf4.h>
+#include <simdmath/log2f4.h>
+#include <simdmath/powf4.h>
+#include <simdmath/sinf4.h>
+#include <simdmath/sqrtf4.h>
+#include <simdmath/truncf4.h>
+
+#include "pipe/p_compiler.h"
+#include "pipe/p_state.h"
+#include "pipe/p_util.h"
+#include "pipe/p_shader_tokens.h"
+#include "pipe/tgsi/util/tgsi_parse.h"
+#include "pipe/tgsi/util/tgsi_util.h"
+#include "spu_exec.h"
+#include "spu_main.h"
+#include "spu_vertex_shader.h"
+
+#define TILE_TOP_LEFT     0
+#define TILE_TOP_RIGHT    1
+#define TILE_BOTTOM_LEFT  2
+#define TILE_BOTTOM_RIGHT 3
+
+/*
+ * Shorthand locations of various utility registers (_I = Index, _C = Channel)
+ */
+#define TEMP_0_I           TGSI_EXEC_TEMP_00000000_I
+#define TEMP_0_C           TGSI_EXEC_TEMP_00000000_C
+#define TEMP_7F_I          TGSI_EXEC_TEMP_7FFFFFFF_I
+#define TEMP_7F_C          TGSI_EXEC_TEMP_7FFFFFFF_C
+#define TEMP_80_I          TGSI_EXEC_TEMP_80000000_I
+#define TEMP_80_C          TGSI_EXEC_TEMP_80000000_C
+#define TEMP_FF_I          TGSI_EXEC_TEMP_FFFFFFFF_I
+#define TEMP_FF_C          TGSI_EXEC_TEMP_FFFFFFFF_C
+#define TEMP_1_I           TGSI_EXEC_TEMP_ONE_I
+#define TEMP_1_C           TGSI_EXEC_TEMP_ONE_C
+#define TEMP_2_I           TGSI_EXEC_TEMP_TWO_I
+#define TEMP_2_C           TGSI_EXEC_TEMP_TWO_C
+#define TEMP_128_I         TGSI_EXEC_TEMP_128_I
+#define TEMP_128_C         TGSI_EXEC_TEMP_128_C
+#define TEMP_M128_I        TGSI_EXEC_TEMP_MINUS_128_I
+#define TEMP_M128_C        TGSI_EXEC_TEMP_MINUS_128_C
+#define TEMP_KILMASK_I     TGSI_EXEC_TEMP_KILMASK_I
+#define TEMP_KILMASK_C     TGSI_EXEC_TEMP_KILMASK_C
+#define TEMP_OUTPUT_I      TGSI_EXEC_TEMP_OUTPUT_I
+#define TEMP_OUTPUT_C      TGSI_EXEC_TEMP_OUTPUT_C
+#define TEMP_PRIMITIVE_I   TGSI_EXEC_TEMP_PRIMITIVE_I
+#define TEMP_PRIMITIVE_C   TGSI_EXEC_TEMP_PRIMITIVE_C
+#define TEMP_R0            TGSI_EXEC_TEMP_R0
+
+#define FOR_EACH_CHANNEL(CHAN)\
+   for (CHAN = 0; CHAN < 4; CHAN++)
+
+#define IS_CHANNEL_ENABLED(INST, CHAN)\
+   ((INST).FullDstRegisters[0].DstRegister.WriteMask & (1 << (CHAN)))
+
+#define IS_CHANNEL_ENABLED2(INST, CHAN)\
+   ((INST).FullDstRegisters[1].DstRegister.WriteMask & (1 << (CHAN)))
+
+#define FOR_EACH_ENABLED_CHANNEL(INST, CHAN)\
+   FOR_EACH_CHANNEL( CHAN )\
+      if (IS_CHANNEL_ENABLED( INST, CHAN ))
+
+#define FOR_EACH_ENABLED_CHANNEL2(INST, CHAN)\
+   FOR_EACH_CHANNEL( CHAN )\
+      if (IS_CHANNEL_ENABLED2( INST, CHAN ))
+
+
+/** The execution mask depends on the conditional mask and the loop mask */
+#define UPDATE_EXEC_MASK(MACH) \
+      MACH->ExecMask = MACH->CondMask & MACH->LoopMask & MACH->ContMask & MACH->FuncMask
+
+
+#define CHAN_X  0
+#define CHAN_Y  1
+#define CHAN_Z  2
+#define CHAN_W  3
+
+
+
+/**
+ * Initialize machine state by expanding tokens to full instructions,
+ * allocating temporary storage, setting up constants, etc.
+ * After this, we can call spu_exec_machine_run() many times.
+ */
+void
+spu_exec_machine_init(struct spu_exec_machine *mach,
+                      uint numSamplers,
+                      struct spu_sampler *samplers,
+                      unsigned processor)
+{
+   qword zero;
+   qword not_zero;
+   uint i;
+
+   mach->Samplers = samplers;
+   mach->Processor = processor;
+   mach->Addrs = &mach->Temps[TGSI_EXEC_NUM_TEMPS];
+
+   zero = si_xor(zero, zero);
+   not_zero = si_xori(zero, 0xff);
+
+   /* Setup constants. */
+   mach->Temps[TEMP_0_I].xyzw[TEMP_0_C].q = zero;
+   mach->Temps[TEMP_FF_I].xyzw[TEMP_FF_C].q = not_zero;
+   mach->Temps[TEMP_7F_I].xyzw[TEMP_7F_C].q = si_shli(not_zero, -1);
+   mach->Temps[TEMP_80_I].xyzw[TEMP_80_C].q = si_shli(not_zero, 31);
+
+   mach->Temps[TEMP_1_I].xyzw[TEMP_1_C].q = (qword) spu_splats(1.0f);
+   mach->Temps[TEMP_2_I].xyzw[TEMP_2_C].q = (qword) spu_splats(2.0f);
+   mach->Temps[TEMP_128_I].xyzw[TEMP_128_C].q = (qword) spu_splats(128.0f);
+   mach->Temps[TEMP_M128_I].xyzw[TEMP_M128_C].q = (qword) spu_splats(-128.0f);
+}
+
+
+static INLINE qword
+micro_abs(qword src)
+{
+   return si_rotmi(si_shli(src, 1), -1);
+}
+
+static INLINE qword
+micro_ceil(qword src)
+{
+   return (qword) _ceilf4((vec_float4) src);
+}
+
+static INLINE qword
+micro_cos(qword src)
+{
+   return (qword) _cosf4((vec_float4) src);
+}
+
+static const qword br_shuf = {
+   TILE_BOTTOM_RIGHT + 0, TILE_BOTTOM_RIGHT + 1,
+   TILE_BOTTOM_RIGHT + 2, TILE_BOTTOM_RIGHT + 3,
+   TILE_BOTTOM_RIGHT + 0, TILE_BOTTOM_RIGHT + 1,
+   TILE_BOTTOM_RIGHT + 2, TILE_BOTTOM_RIGHT + 3,
+   TILE_BOTTOM_RIGHT + 0, TILE_BOTTOM_RIGHT + 1,
+   TILE_BOTTOM_RIGHT + 2, TILE_BOTTOM_RIGHT + 3,
+   TILE_BOTTOM_RIGHT + 0, TILE_BOTTOM_RIGHT + 1,
+   TILE_BOTTOM_RIGHT + 2, TILE_BOTTOM_RIGHT + 3,
+};
+
+static const qword bl_shuf = {
+   TILE_BOTTOM_LEFT + 0, TILE_BOTTOM_LEFT + 1,
+   TILE_BOTTOM_LEFT + 2, TILE_BOTTOM_LEFT + 3,
+   TILE_BOTTOM_LEFT + 0, TILE_BOTTOM_LEFT + 1,
+   TILE_BOTTOM_LEFT + 2, TILE_BOTTOM_LEFT + 3,
+   TILE_BOTTOM_LEFT + 0, TILE_BOTTOM_LEFT + 1,
+   TILE_BOTTOM_LEFT + 2, TILE_BOTTOM_LEFT + 3,
+   TILE_BOTTOM_LEFT + 0, TILE_BOTTOM_LEFT + 1,
+   TILE_BOTTOM_LEFT + 2, TILE_BOTTOM_LEFT + 3,
+};
+
+static const qword tl_shuf = {
+   TILE_TOP_LEFT + 0, TILE_TOP_LEFT + 1,
+   TILE_TOP_LEFT + 2, TILE_TOP_LEFT + 3,
+   TILE_TOP_LEFT + 0, TILE_TOP_LEFT + 1,
+   TILE_TOP_LEFT + 2, TILE_TOP_LEFT + 3,
+   TILE_TOP_LEFT + 0, TILE_TOP_LEFT + 1,
+   TILE_TOP_LEFT + 2, TILE_TOP_LEFT + 3,
+   TILE_TOP_LEFT + 0, TILE_TOP_LEFT + 1,
+   TILE_TOP_LEFT + 2, TILE_TOP_LEFT + 3,
+};
+
+static qword
+micro_ddx(qword src)
+{
+   qword bottom_right = si_shufb(src, src, br_shuf);
+   qword bottom_left = si_shufb(src, src, bl_shuf);
+
+   return si_fs(bottom_right, bottom_left);
+}
+
+static qword
+micro_ddy(qword src)
+{
+   qword top_left = si_shufb(src, src, tl_shuf);
+   qword bottom_left = si_shufb(src, src, bl_shuf);
+
+   return si_fs(top_left, bottom_left);
+}
+
+static INLINE qword
+micro_div(qword src0, qword src1)
+{
+   return (qword) _divf4((vec_float4) src0, (vec_float4) src1);
+}
+
+static qword
+micro_flr(qword src)
+{
+   return (qword) _floorf4((vec_float4) src);
+}
+
+static qword
+micro_frc(qword src)
+{
+   return si_fs(src, (qword) _floorf4((vec_float4) src));
+}
+
+static INLINE qword
+micro_ge(qword src0, qword src1)
+{
+   return si_or(si_fceq(src0, src1), si_fcgt(src0, src1));
+}
+
+static qword
+micro_lg2(qword src)
+{
+   return (qword) _log2f4((vec_float4) src);
+}
+
+static INLINE qword
+micro_lt(qword src0, qword src1)
+{
+   const qword tmp = si_or(si_fceq(src0, src1), si_fcgt(src0, src1));
+
+   return si_xori(tmp, 0xff);
+}
+
+static INLINE qword
+micro_max(qword src0, qword src1)
+{
+   return si_selb(src1, src0, si_fcgt(src0, src1));
+}
+
+static INLINE qword
+micro_min(qword src0, qword src1)
+{
+   return si_selb(src0, src1, si_fcgt(src0, src1));
+}
+
+static qword
+micro_neg(qword src)
+{
+   return si_xor(src, (qword) spu_splats(0x80000000));
+}
+
+static qword
+micro_set_sign(qword src)
+{
+   return si_or(src, (qword) spu_splats(0x80000000));
+}
+
+static qword
+micro_pow(qword src0, qword src1)
+{
+   return (qword) _powf4((vec_float4) src0, (vec_float4) src1);
+}
+
+static qword
+micro_rnd(qword src)
+{
+   const qword half = (qword) spu_splats(0.5f);
+
+   /* May be able to use _roundf4.  There may be some difference, though.
+    */
+   return (qword) _floorf4((vec_float4) si_fa(src, half));
+}
+
+static INLINE qword
+micro_ishr(qword src0, qword src1)
+{
+   return si_rotma(src0, si_sfi(src1, 0));
+}
+
+static qword
+micro_trunc(qword src)
+{
+   return (qword) _truncf4((vec_float4) src);
+}
+
+static qword
+micro_sin(qword src)
+{
+   return (qword) _sinf4((vec_float4) src);
+}
+
+static INLINE qword
+micro_sqrt(qword src)
+{
+   return (qword) _sqrtf4((vec_float4) src);
+}
+
+static void
+fetch_src_file_channel(
+   const struct spu_exec_machine *mach,
+   const uint file,
+   const uint swizzle,
+   const union spu_exec_channel *index,
+   union spu_exec_channel *chan )
+{
+   switch( swizzle ) {
+   case TGSI_EXTSWIZZLE_X:
+   case TGSI_EXTSWIZZLE_Y:
+   case TGSI_EXTSWIZZLE_Z:
+   case TGSI_EXTSWIZZLE_W:
+      switch( file ) {
+      case TGSI_FILE_CONSTANT: {
+         unsigned char buffer[32] ALIGN16_ATTRIB;
+         unsigned i;
+
+         for (i = 0; i < 4; i++) {
+            const float *ptr = mach->Consts[index->i[i]];
+            const uint64_t addr = (uint64_t)(uintptr_t) ptr;
+            const unsigned size = ((addr & 0x0f) == 0) ? 16 : 32;
+
+            mfc_get(buffer, addr & ~0x0f, size, TAG_VERTEX_BUFFER, 0, 0);
+            wait_on_mask(1 << TAG_VERTEX_BUFFER);
+
+            (void) memcpy(& chan->f[i], &buffer[(addr & 0x0f) 
+                + (sizeof(float) * swizzle)], sizeof(float));
+         }
+         break;
+      }
+
+      case TGSI_FILE_INPUT:
+         chan->u[0] = mach->Inputs[index->i[0]].xyzw[swizzle].u[0];
+         chan->u[1] = mach->Inputs[index->i[1]].xyzw[swizzle].u[1];
+         chan->u[2] = mach->Inputs[index->i[2]].xyzw[swizzle].u[2];
+         chan->u[3] = mach->Inputs[index->i[3]].xyzw[swizzle].u[3];
+         break;
+
+      case TGSI_FILE_TEMPORARY:
+         chan->u[0] = mach->Temps[index->i[0]].xyzw[swizzle].u[0];
+         chan->u[1] = mach->Temps[index->i[1]].xyzw[swizzle].u[1];
+         chan->u[2] = mach->Temps[index->i[2]].xyzw[swizzle].u[2];
+         chan->u[3] = mach->Temps[index->i[3]].xyzw[swizzle].u[3];
+         break;
+
+      case TGSI_FILE_IMMEDIATE:
+         assert( index->i[0] < (int) mach->ImmLimit );
+         assert( index->i[1] < (int) mach->ImmLimit );
+         assert( index->i[2] < (int) mach->ImmLimit );
+         assert( index->i[3] < (int) mach->ImmLimit );
+
+         chan->f[0] = mach->Imms[index->i[0]][swizzle];
+         chan->f[1] = mach->Imms[index->i[1]][swizzle];
+         chan->f[2] = mach->Imms[index->i[2]][swizzle];
+         chan->f[3] = mach->Imms[index->i[3]][swizzle];
+         break;
+
+      case TGSI_FILE_ADDRESS:
+         chan->u[0] = mach->Addrs[index->i[0]].xyzw[swizzle].u[0];
+         chan->u[1] = mach->Addrs[index->i[1]].xyzw[swizzle].u[1];
+         chan->u[2] = mach->Addrs[index->i[2]].xyzw[swizzle].u[2];
+         chan->u[3] = mach->Addrs[index->i[3]].xyzw[swizzle].u[3];
+         break;
+
+      case TGSI_FILE_OUTPUT:
+         /* vertex/fragment output vars can be read too */
+         chan->u[0] = mach->Outputs[index->i[0]].xyzw[swizzle].u[0];
+         chan->u[1] = mach->Outputs[index->i[1]].xyzw[swizzle].u[1];
+         chan->u[2] = mach->Outputs[index->i[2]].xyzw[swizzle].u[2];
+         chan->u[3] = mach->Outputs[index->i[3]].xyzw[swizzle].u[3];
+         break;
+
+      default:
+         assert( 0 );
+      }
+      break;
+
+   case TGSI_EXTSWIZZLE_ZERO:
+      *chan = mach->Temps[TEMP_0_I].xyzw[TEMP_0_C];
+      break;
+
+   case TGSI_EXTSWIZZLE_ONE:
+      *chan = mach->Temps[TEMP_1_I].xyzw[TEMP_1_C];
+      break;
+
+   default:
+      assert( 0 );
+   }
+}
+
+static void
+fetch_source(
+   const struct spu_exec_machine *mach,
+   union spu_exec_channel *chan,
+   const struct tgsi_full_src_register *reg,
+   const uint chan_index )
+{
+   union spu_exec_channel index;
+   uint swizzle;
+
+   index.i[0] =
+   index.i[1] =
+   index.i[2] =
+   index.i[3] = reg->SrcRegister.Index;
+
+   if (reg->SrcRegister.Indirect) {
+      union spu_exec_channel index2;
+      union spu_exec_channel indir_index;
+
+      index2.i[0] =
+      index2.i[1] =
+      index2.i[2] =
+      index2.i[3] = reg->SrcRegisterInd.Index;
+
+      swizzle = tgsi_util_get_src_register_swizzle(&reg->SrcRegisterInd,
+                                                   CHAN_X);
+      fetch_src_file_channel(
+         mach,
+         reg->SrcRegisterInd.File,
+         swizzle,
+         &index2,
+         &indir_index );
+
+      index.q = si_a(index.q, indir_index.q);
+   }
+
+   if( reg->SrcRegister.Dimension ) {
+      switch( reg->SrcRegister.File ) {
+      case TGSI_FILE_INPUT:
+         index.q = si_mpyi(index.q, 17);
+         break;
+      case TGSI_FILE_CONSTANT:
+         index.q = si_shli(index.q, 12);
+         break;
+      default:
+         assert( 0 );
+      }
+
+      index.i[0] += reg->SrcRegisterDim.Index;
+      index.i[1] += reg->SrcRegisterDim.Index;
+      index.i[2] += reg->SrcRegisterDim.Index;
+      index.i[3] += reg->SrcRegisterDim.Index;
+
+      if (reg->SrcRegisterDim.Indirect) {
+         union spu_exec_channel index2;
+         union spu_exec_channel indir_index;
+
+         index2.i[0] =
+         index2.i[1] =
+         index2.i[2] =
+         index2.i[3] = reg->SrcRegisterDimInd.Index;
+
+         swizzle = tgsi_util_get_src_register_swizzle( &reg->SrcRegisterDimInd, CHAN_X );
+         fetch_src_file_channel(
+            mach,
+            reg->SrcRegisterDimInd.File,
+            swizzle,
+            &index2,
+            &indir_index );
+
+         index.q = si_a(index.q, indir_index.q);
+      }
+   }
+
+   swizzle = tgsi_util_get_full_src_register_extswizzle( reg, chan_index );
+   fetch_src_file_channel(
+      mach,
+      reg->SrcRegister.File,
+      swizzle,
+      &index,
+      chan );
+
+   switch (tgsi_util_get_full_src_register_sign_mode( reg, chan_index )) {
+   case TGSI_UTIL_SIGN_CLEAR:
+      chan->q = micro_abs(chan->q);
+      break;
+
+   case TGSI_UTIL_SIGN_SET:
+      chan->q = micro_set_sign(chan->q);
+      break;
+
+   case TGSI_UTIL_SIGN_TOGGLE:
+      chan->q = micro_neg(chan->q);
+      break;
+
+   case TGSI_UTIL_SIGN_KEEP:
+      break;
+   }
+
+   if (reg->SrcRegisterExtMod.Complement) {
+      chan->q = si_fs(mach->Temps[TEMP_1_I].xyzw[TEMP_1_C].q, chan->q);
+   }
+}
+
+static void
+store_dest(
+   struct spu_exec_machine *mach,
+   const union spu_exec_channel *chan,
+   const struct tgsi_full_dst_register *reg,
+   const struct tgsi_full_instruction *inst,
+   uint chan_index )
+{
+   union spu_exec_channel *dst;
+
+   switch( reg->DstRegister.File ) {
+   case TGSI_FILE_NULL:
+      return;
+
+   case TGSI_FILE_OUTPUT:
+      dst = &mach->Outputs[mach->Temps[TEMP_OUTPUT_I].xyzw[TEMP_OUTPUT_C].u[0]
+                           + reg->DstRegister.Index].xyzw[chan_index];
+      break;
+
+   case TGSI_FILE_TEMPORARY:
+      dst = &mach->Temps[reg->DstRegister.Index].xyzw[chan_index];
+      break;
+
+   case TGSI_FILE_ADDRESS:
+      dst = &mach->Addrs[reg->DstRegister.Index].xyzw[chan_index];
+      break;
+
+   default:
+      assert( 0 );
+      return;
+   }
+
+   switch (inst->Instruction.Saturate)
+   {
+   case TGSI_SAT_NONE:
+      if (mach->ExecMask & 0x1)
+         dst->i[0] = chan->i[0];
+      if (mach->ExecMask & 0x2)
+         dst->i[1] = chan->i[1];
+      if (mach->ExecMask & 0x4)
+         dst->i[2] = chan->i[2];
+      if (mach->ExecMask & 0x8)
+         dst->i[3] = chan->i[3];
+      break;
+
+   case TGSI_SAT_ZERO_ONE:
+      /* XXX need to obey ExecMask here */
+      dst->q = micro_max(chan->q, mach->Temps[TEMP_0_I].xyzw[TEMP_0_C].q);
+      dst->q = micro_min(dst->q, mach->Temps[TEMP_1_I].xyzw[TEMP_1_C].q);
+      break;
+
+   case TGSI_SAT_MINUS_PLUS_ONE:
+      assert( 0 );
+      break;
+
+   default:
+      assert( 0 );
+   }
+}
+
+#define FETCH(VAL,INDEX,CHAN)\
+    fetch_source (mach, VAL, &inst->FullSrcRegisters[INDEX], CHAN)
+
+#define STORE(VAL,INDEX,CHAN)\
+    store_dest (mach, VAL, &inst->FullDstRegisters[INDEX], inst, CHAN )
+
+
+/**
+ * Execute ARB-style KIL which is predicated by a src register.
+ * Kill fragment if any of the four values is less than zero.
+ */
+static void
+exec_kilp(struct spu_exec_machine *mach,
+          const struct tgsi_full_instruction *inst)
+{
+   uint uniquemask;
+   uint chan_index;
+   uint kilmask = 0; /* bit 0 = pixel 0, bit 1 = pixel 1, etc */
+   union spu_exec_channel r[1];
+
+   /* This mask stores component bits that were already tested. Note that
+    * we test if the value is less than zero, so 1.0 and 0.0 need not to be
+    * tested. */
+   uniquemask = (1 << TGSI_EXTSWIZZLE_ZERO) | (1 << TGSI_EXTSWIZZLE_ONE);
+
+   for (chan_index = 0; chan_index < 4; chan_index++)
+   {
+      uint swizzle;
+      uint i;
+
+      /* unswizzle channel */
+      swizzle = tgsi_util_get_full_src_register_extswizzle (
+                        &inst->FullSrcRegisters[0],
+                        chan_index);
+
+      /* check if the component has not been already tested */
+      if (uniquemask & (1 << swizzle))
+         continue;
+      uniquemask |= 1 << swizzle;
+
+      FETCH(&r[0], 0, chan_index);
+      for (i = 0; i < 4; i++)
+         if (r[0].f[i] < 0.0f)
+            kilmask |= 1 << i;
+   }
+
+   mach->Temps[TEMP_KILMASK_I].xyzw[TEMP_KILMASK_C].u[0] |= kilmask;
+}
+
+
+/*
+ * Fetch a texel using STR texture coordinates.
+ */
+static void
+fetch_texel( struct spu_sampler *sampler,
+             const union spu_exec_channel *s,
+             const union spu_exec_channel *t,
+             const union spu_exec_channel *p,
+             float lodbias,  /* XXX should be float[4] */
+             union spu_exec_channel *r,
+             union spu_exec_channel *g,
+             union spu_exec_channel *b,
+             union spu_exec_channel *a )
+{
+   qword rgba[4];
+   qword out[4];
+
+   sampler->get_samples(sampler, s->f, t->f, p->f, lodbias, (float *) rgba);
+
+   _transpose_matrix4x4(out, rgba);
+   r->q = out[0];
+   g->q = out[1];
+   b->q = out[2];
+   a->q = out[3];
+}
+
+
+static void
+exec_tex(struct spu_exec_machine *mach,
+         const struct tgsi_full_instruction *inst,
+         boolean biasLod)
+{
+   const uint unit = inst->FullSrcRegisters[1].SrcRegister.Index;
+   union spu_exec_channel r[8];
+   uint chan_index;
+   float lodBias;
+
+   /*   printf("Sampler %u unit %u\n", sampler, unit); */
+
+   switch (inst->InstructionExtTexture.Texture) {
+   case TGSI_TEXTURE_1D:
+
+      FETCH(&r[0], 0, CHAN_X);
+
+      switch (inst->FullSrcRegisters[0].SrcRegisterExtSwz.ExtDivide) {
+      case TGSI_EXTSWIZZLE_W:
+         FETCH(&r[1], 0, CHAN_W);
+         r[0].q = micro_div(r[0].q, r[1].q);
+         break;
+
+      case TGSI_EXTSWIZZLE_ONE:
+         break;
+
+      default:
+         assert (0);
+      }
+
+      if (biasLod) {
+         FETCH(&r[1], 0, CHAN_W);
+         lodBias = r[2].f[0];
+      }
+      else
+         lodBias = 0.0;
+
+      fetch_texel(&mach->Samplers[unit],
+                  &r[0], NULL, NULL, lodBias,  /* S, T, P, BIAS */
+                  &r[0], &r[1], &r[2], &r[3]); /* R, G, B, A */
+      break;
+
+   case TGSI_TEXTURE_2D:
+   case TGSI_TEXTURE_RECT:
+
+      FETCH(&r[0], 0, CHAN_X);
+      FETCH(&r[1], 0, CHAN_Y);
+      FETCH(&r[2], 0, CHAN_Z);
+
+      switch (inst->FullSrcRegisters[0].SrcRegisterExtSwz.ExtDivide) {
+      case TGSI_EXTSWIZZLE_W:
+         FETCH(&r[3], 0, CHAN_W);
+         r[0].q = micro_div(r[0].q, r[3].q);
+         r[1].q = micro_div(r[1].q, r[3].q);
+         r[2].q = micro_div(r[2].q, r[3].q);
+         break;
+
+      case TGSI_EXTSWIZZLE_ONE:
+         break;
+
+      default:
+         assert (0);
+      }
+
+      if (biasLod) {
+         FETCH(&r[3], 0, CHAN_W);
+         lodBias = r[3].f[0];
+      }
+      else
+         lodBias = 0.0;
+
+      fetch_texel(&mach->Samplers[unit],
+                  &r[0], &r[1], &r[2], lodBias,  /* inputs */
+                  &r[0], &r[1], &r[2], &r[3]);  /* outputs */
+      break;
+
+   case TGSI_TEXTURE_3D:
+   case TGSI_TEXTURE_CUBE:
+
+      FETCH(&r[0], 0, CHAN_X);
+      FETCH(&r[1], 0, CHAN_Y);
+      FETCH(&r[2], 0, CHAN_Z);
+
+      switch (inst->FullSrcRegisters[0].SrcRegisterExtSwz.ExtDivide) {
+      case TGSI_EXTSWIZZLE_W:
+         FETCH(&r[3], 0, CHAN_W);
+         r[0].q = micro_div(r[0].q, r[3].q);
+         r[1].q = micro_div(r[1].q, r[3].q);
+         r[2].q = micro_div(r[2].q, r[3].q);
+         break;
+
+      case TGSI_EXTSWIZZLE_ONE:
+         break;
+
+      default:
+         assert (0);
+      }
+
+      if (biasLod) {
+         FETCH(&r[3], 0, CHAN_W);
+         lodBias = r[3].f[0];
+      }
+      else
+         lodBias = 0.0;
+
+      fetch_texel(&mach->Samplers[unit],
+                  &r[0], &r[1], &r[2], lodBias,
+                  &r[0], &r[1], &r[2], &r[3]);
+      break;
+
+   default:
+      assert (0);
+   }
+
+   FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
+      STORE( &r[chan_index], 0, chan_index );
+   }
+}
+
+
+
+static void
+constant_interpolation(
+   struct spu_exec_machine *mach,
+   unsigned attrib,
+   unsigned chan )
+{
+   unsigned i;
+
+   for( i = 0; i < QUAD_SIZE; i++ ) {
+      mach->Inputs[attrib].xyzw[chan].f[i] = mach->InterpCoefs[attrib].a0[chan];
+   }
+}
+
+static void
+linear_interpolation(
+   struct spu_exec_machine *mach,
+   unsigned attrib,
+   unsigned chan )
+{
+   const float x = mach->QuadPos.xyzw[0].f[0];
+   const float y = mach->QuadPos.xyzw[1].f[0];
+   const float dadx = mach->InterpCoefs[attrib].dadx[chan];
+   const float dady = mach->InterpCoefs[attrib].dady[chan];
+   const float a0 = mach->InterpCoefs[attrib].a0[chan] + dadx * x + dady * y;
+   mach->Inputs[attrib].xyzw[chan].f[0] = a0;
+   mach->Inputs[attrib].xyzw[chan].f[1] = a0 + dadx;
+   mach->Inputs[attrib].xyzw[chan].f[2] = a0 + dady;
+   mach->Inputs[attrib].xyzw[chan].f[3] = a0 + dadx + dady;
+}
+
+static void
+perspective_interpolation(
+   struct spu_exec_machine *mach,
+   unsigned attrib,
+   unsigned chan )
+{
+   const float x = mach->QuadPos.xyzw[0].f[0];
+   const float y = mach->QuadPos.xyzw[1].f[0];
+   const float dadx = mach->InterpCoefs[attrib].dadx[chan];
+   const float dady = mach->InterpCoefs[attrib].dady[chan];
+   const float a0 = mach->InterpCoefs[attrib].a0[chan] + dadx * x + dady * y;
+   const float *w = mach->QuadPos.xyzw[3].f;
+   /* divide by W here */
+   mach->Inputs[attrib].xyzw[chan].f[0] = a0 / w[0];
+   mach->Inputs[attrib].xyzw[chan].f[1] = (a0 + dadx) / w[1];
+   mach->Inputs[attrib].xyzw[chan].f[2] = (a0 + dady) / w[2];
+   mach->Inputs[attrib].xyzw[chan].f[3] = (a0 + dadx + dady) / w[3];
+}
+
+
+typedef void (* interpolation_func)(
+   struct spu_exec_machine *mach,
+   unsigned attrib,
+   unsigned chan );
+
+static void
+exec_declaration(struct spu_exec_machine *mach,
+                 const struct tgsi_full_declaration *decl)
+{
+   if( mach->Processor == TGSI_PROCESSOR_FRAGMENT ) {
+      if( decl->Declaration.File == TGSI_FILE_INPUT ) {
+         unsigned first, last, mask;
+         interpolation_func interp;
+
+         assert( decl->Declaration.Declare == TGSI_DECLARE_RANGE );
+
+         first = decl->u.DeclarationRange.First;
+         last = decl->u.DeclarationRange.Last;
+         mask = decl->Declaration.UsageMask;
+
+         switch( decl->Interpolation.Interpolate ) {
+         case TGSI_INTERPOLATE_CONSTANT:
+            interp = constant_interpolation;
+            break;
+
+         case TGSI_INTERPOLATE_LINEAR:
+            interp = linear_interpolation;
+            break;
+
+         case TGSI_INTERPOLATE_PERSPECTIVE:
+            interp = perspective_interpolation;
+            break;
+
+         default:
+            assert( 0 );
+         }
+
+         if( mask == TGSI_WRITEMASK_XYZW ) {
+            unsigned i, j;
+
+            for( i = first; i <= last; i++ ) {
+               for( j = 0; j < NUM_CHANNELS; j++ ) {
+                  interp( mach, i, j );
+               }
+            }
+         }
+         else {
+            unsigned i, j;
+
+            for( j = 0; j < NUM_CHANNELS; j++ ) {
+               if( mask & (1 << j) ) {
+                  for( i = first; i <= last; i++ ) {
+                     interp( mach, i, j );
+                  }
+               }
+            }
+         }
+      }
+   }
+}
+
+static void
+exec_instruction(
+   struct spu_exec_machine *mach,
+   const struct tgsi_full_instruction *inst,
+   int *pc )
+{
+   uint chan_index;
+   union spu_exec_channel r[8];
+
+   (*pc)++;
+
+   switch (inst->Instruction.Opcode) {
+   case TGSI_OPCODE_ARL:
+      FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
+	 FETCH( &r[0], 0, chan_index );
+         r[0].q = si_cflts(r[0].q, 0);
+	 STORE( &r[0], 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_MOV:
+   /* TGSI_OPCODE_SWZ */
+      FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
+         FETCH( &r[0], 0, chan_index );
+         STORE( &r[0], 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_LIT:
+      if (IS_CHANNEL_ENABLED( *inst, CHAN_X )) {
+	 STORE( &mach->Temps[TEMP_1_I].xyzw[TEMP_1_C], 0, CHAN_X );
+      }
+
+      if (IS_CHANNEL_ENABLED( *inst, CHAN_Y ) || IS_CHANNEL_ENABLED( *inst, CHAN_Z )) {
+	 FETCH( &r[0], 0, CHAN_X );
+         if (IS_CHANNEL_ENABLED( *inst, CHAN_Y )) {
+            r[0].q = micro_max(r[0].q, mach->Temps[TEMP_0_I].xyzw[TEMP_0_C].q);
+	    STORE( &r[0], 0, CHAN_Y );
+	 }
+
+         if (IS_CHANNEL_ENABLED( *inst, CHAN_Z )) {
+            FETCH( &r[1], 0, CHAN_Y );
+            r[1].q = micro_max(r[1].q, mach->Temps[TEMP_0_I].xyzw[TEMP_0_C].q);
+
+            FETCH( &r[2], 0, CHAN_W );
+            r[2].q = micro_min(r[2].q, mach->Temps[TEMP_128_I].xyzw[TEMP_128_C].q);
+            r[2].q = micro_max(r[2].q, mach->Temps[TEMP_M128_I].xyzw[TEMP_M128_C].q);
+            r[1].q = micro_pow(r[1].q, r[2].q);
+
+            /* r0 = (r0 > 0.0) ? r1 : 0.0
+             */
+            r[0].q = si_fcgt(r[0].q, mach->Temps[TEMP_0_I].xyzw[TEMP_0_C].q);
+            r[0].q = si_selb(mach->Temps[TEMP_0_I].xyzw[TEMP_0_C].q, r[1].q,
+                             r[0].q);
+            STORE( &r[0], 0, CHAN_Z );
+         }
+      }
+
+      if (IS_CHANNEL_ENABLED( *inst, CHAN_W )) {
+	 STORE( &mach->Temps[TEMP_1_I].xyzw[TEMP_1_C], 0, CHAN_W );
+      }
+      break;
+
+   case TGSI_OPCODE_RCP:
+   /* TGSI_OPCODE_RECIP */
+      FETCH( &r[0], 0, CHAN_X );
+      r[0].q = micro_div(mach->Temps[TEMP_1_I].xyzw[TEMP_1_C].q, r[0].q);
+      FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
+	 STORE( &r[0], 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_RSQ:
+   /* TGSI_OPCODE_RECIPSQRT */
+      FETCH( &r[0], 0, CHAN_X );
+      r[0].q = micro_sqrt(r[0].q);
+      r[0].q = micro_div(mach->Temps[TEMP_1_I].xyzw[TEMP_1_C].q, r[0].q);
+      FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
+	 STORE( &r[0], 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_EXP:
+      assert (0);
+      break;
+
+   case TGSI_OPCODE_LOG:
+      assert (0);
+      break;
+
+   case TGSI_OPCODE_MUL:
+      FOR_EACH_ENABLED_CHANNEL( *inst, chan_index )
+      {
+         FETCH(&r[0], 0, chan_index);
+         FETCH(&r[1], 1, chan_index);
+
+         r[0].q = si_fm(r[0].q, r[1].q);
+
+         STORE(&r[0], 0, chan_index);
+      }
+      break;
+
+   case TGSI_OPCODE_ADD:
+      FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
+         FETCH( &r[0], 0, chan_index );
+         FETCH( &r[1], 1, chan_index );
+         r[0].q = si_fa(r[0].q, r[1].q);
+         STORE( &r[0], 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_DP3:
+   /* TGSI_OPCODE_DOT3 */
+      FETCH( &r[0], 0, CHAN_X );
+      FETCH( &r[1], 1, CHAN_X );
+      r[0].q = si_fm(r[0].q, r[1].q);
+
+      FETCH( &r[1], 0, CHAN_Y );
+      FETCH( &r[2], 1, CHAN_Y );
+      r[0].q = si_fma(r[1].q, r[2].q, r[0].q);
+
+
+      FETCH( &r[1], 0, CHAN_Z );
+      FETCH( &r[2], 1, CHAN_Z );
+      r[0].q = si_fma(r[1].q, r[2].q, r[0].q);
+
+      FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
+         STORE( &r[0], 0, chan_index );
+      }
+      break;
+
+    case TGSI_OPCODE_DP4:
+    /* TGSI_OPCODE_DOT4 */
+       FETCH(&r[0], 0, CHAN_X);
+       FETCH(&r[1], 1, CHAN_X);
+
+      r[0].q = si_fm(r[0].q, r[1].q);
+
+       FETCH(&r[1], 0, CHAN_Y);
+       FETCH(&r[2], 1, CHAN_Y);
+
+      r[0].q = si_fma(r[1].q, r[2].q, r[0].q);
+
+       FETCH(&r[1], 0, CHAN_Z);
+       FETCH(&r[2], 1, CHAN_Z);
+
+      r[0].q = si_fma(r[1].q, r[2].q, r[0].q);
+
+       FETCH(&r[1], 0, CHAN_W);
+       FETCH(&r[2], 1, CHAN_W);
+
+      r[0].q = si_fma(r[1].q, r[2].q, r[0].q);
+
+      FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
+	 STORE( &r[0], 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_DST:
+      if (IS_CHANNEL_ENABLED( *inst, CHAN_X )) {
+	 STORE( &mach->Temps[TEMP_1_I].xyzw[TEMP_1_C], 0, CHAN_X );
+      }
+
+      if (IS_CHANNEL_ENABLED( *inst, CHAN_Y )) {
+	 FETCH( &r[0], 0, CHAN_Y );
+	 FETCH( &r[1], 1, CHAN_Y);
+      r[0].q = si_fm(r[0].q, r[1].q);
+	 STORE( &r[0], 0, CHAN_Y );
+      }
+
+      if (IS_CHANNEL_ENABLED( *inst, CHAN_Z )) {
+	 FETCH( &r[0], 0, CHAN_Z );
+	 STORE( &r[0], 0, CHAN_Z );
+      }
+
+      if (IS_CHANNEL_ENABLED( *inst, CHAN_W )) {
+	 FETCH( &r[0], 1, CHAN_W );
+	 STORE( &r[0], 0, CHAN_W );
+      }
+      break;
+
+   case TGSI_OPCODE_MIN:
+      FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
+         FETCH(&r[0], 0, chan_index);
+         FETCH(&r[1], 1, chan_index);
+
+         r[0].q = micro_min(r[0].q, r[1].q);
+
+         STORE(&r[0], 0, chan_index);
+      }
+      break;
+
+   case TGSI_OPCODE_MAX:
+      FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
+         FETCH(&r[0], 0, chan_index);
+         FETCH(&r[1], 1, chan_index);
+
+         r[0].q = micro_max(r[0].q, r[1].q);
+
+         STORE(&r[0], 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_SLT:
+   /* TGSI_OPCODE_SETLT */
+      FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
+         FETCH( &r[0], 0, chan_index );
+         FETCH( &r[1], 1, chan_index );
+
+         r[0].q = micro_ge(r[0].q, r[1].q);
+         r[0].q = si_xori(r[0].q, 0xff);
+
+         STORE( &r[0], 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_SGE:
+   /* TGSI_OPCODE_SETGE */
+      FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
+         FETCH( &r[0], 0, chan_index );
+         FETCH( &r[1], 1, chan_index );
+         r[0].q = micro_ge(r[0].q, r[1].q);
+         STORE( &r[0], 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_MAD:
+   /* TGSI_OPCODE_MADD */
+      FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
+         FETCH( &r[0], 0, chan_index );
+         FETCH( &r[1], 1, chan_index );
+         FETCH( &r[2], 2, chan_index );
+         r[0].q = si_fma(r[0].q, r[1].q, r[2].q);
+         STORE( &r[0], 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_SUB:
+      FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
+         FETCH(&r[0], 0, chan_index);
+         FETCH(&r[1], 1, chan_index);
+
+         r[0].q = si_fs(r[0].q, r[1].q);
+
+         STORE(&r[0], 0, chan_index);
+      }
+      break;
+
+   case TGSI_OPCODE_LERP:
+   /* TGSI_OPCODE_LRP */
+      FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
+         FETCH(&r[0], 0, chan_index);
+         FETCH(&r[1], 1, chan_index);
+         FETCH(&r[2], 2, chan_index);
+
+         r[1].q = si_fs(r[1].q, r[2].q);
+         r[0].q = si_fma(r[0].q, r[1].q, r[2].q);
+
+         STORE(&r[0], 0, chan_index);
+      }
+      break;
+
+   case TGSI_OPCODE_CND:
+      assert (0);
+      break;
+
+   case TGSI_OPCODE_CND0:
+      assert (0);
+      break;
+
+   case TGSI_OPCODE_DOT2ADD:
+      /* TGSI_OPCODE_DP2A */
+      assert (0);
+      break;
+
+   case TGSI_OPCODE_INDEX:
+      assert (0);
+      break;
+
+   case TGSI_OPCODE_NEGATE:
+      assert (0);
+      break;
+
+   case TGSI_OPCODE_FRAC:
+   /* TGSI_OPCODE_FRC */
+      FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
+         FETCH( &r[0], 0, chan_index );
+         r[0].q = micro_frc(r[0].q);
+         STORE( &r[0], 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_CLAMP:
+      assert (0);
+      break;
+
+   case TGSI_OPCODE_FLOOR:
+   /* TGSI_OPCODE_FLR */
+      FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
+         FETCH( &r[0], 0, chan_index );
+         r[0].q = micro_flr(r[0].q);
+         STORE( &r[0], 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_ROUND:
+      FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
+         FETCH( &r[0], 0, chan_index );
+         r[0].q = micro_rnd(r[0].q);
+         STORE( &r[0], 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_EXPBASE2:
+    /* TGSI_OPCODE_EX2 */
+      FETCH(&r[0], 0, CHAN_X);
+
+      r[0].q = micro_pow(mach->Temps[TEMP_2_I].xyzw[TEMP_2_C].q, r[0].q);
+
+      FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
+	 STORE( &r[0], 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_LOGBASE2:
+   /* TGSI_OPCODE_LG2 */
+      FETCH( &r[0], 0, CHAN_X );
+      r[0].q = micro_lg2(r[0].q);
+      FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
+         STORE( &r[0], 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_POWER:
+      /* TGSI_OPCODE_POW */
+      FETCH(&r[0], 0, CHAN_X);
+      FETCH(&r[1], 1, CHAN_X);
+
+      r[0].q = micro_pow(r[0].q, r[1].q);
+
+      FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
+	 STORE( &r[0], 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_CROSSPRODUCT:
+      /* TGSI_OPCODE_XPD */
+      FETCH(&r[0], 0, CHAN_Y);
+      FETCH(&r[1], 1, CHAN_Z);
+      FETCH(&r[3], 0, CHAN_Z);
+      FETCH(&r[4], 1, CHAN_Y);
+
+      /* r2 = (r0 * r1) - (r3 * r5)
+       */
+      r[2].q = si_fm(r[3].q, r[5].q);
+      r[2].q = si_fms(r[0].q, r[1].q, r[2].q);
+
+      if (IS_CHANNEL_ENABLED( *inst, CHAN_X )) {
+         STORE( &r[2], 0, CHAN_X );
+      }
+
+      FETCH(&r[2], 1, CHAN_X);
+      FETCH(&r[5], 0, CHAN_X);
+
+      /* r3 = (r3 * r2) - (r1 * r5)
+       */
+      r[1].q = si_fm(r[1].q, r[5].q);
+      r[3].q = si_fms(r[3].q, r[2].q, r[1].q);
+
+      if (IS_CHANNEL_ENABLED( *inst, CHAN_Y )) {
+         STORE( &r[3], 0, CHAN_Y );
+      }
+
+      /* r5 = (r5 * r4) - (r0 * r2)
+       */
+      r[0].q = si_fm(r[0].q, r[2].q);
+      r[5].q = si_fms(r[5].q, r[4].q, r[0].q);
+
+      if (IS_CHANNEL_ENABLED( *inst, CHAN_Z )) {
+         STORE( &r[5], 0, CHAN_Z );
+      }
+
+      if (IS_CHANNEL_ENABLED( *inst, CHAN_W )) {
+         STORE( &mach->Temps[TEMP_1_I].xyzw[TEMP_1_C], 0, CHAN_W );
+      }
+      break;
+
+    case TGSI_OPCODE_MULTIPLYMATRIX:
+       assert (0);
+       break;
+
+    case TGSI_OPCODE_ABS:
+       FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
+          FETCH(&r[0], 0, chan_index);
+
+          r[0].q = micro_abs(r[0].q);
+
+          STORE(&r[0], 0, chan_index);
+       }
+       break;
+
+   case TGSI_OPCODE_RCC:
+      assert (0);
+      break;
+
+   case TGSI_OPCODE_DPH:
+      FETCH(&r[0], 0, CHAN_X);
+      FETCH(&r[1], 1, CHAN_X);
+
+      r[0].q = si_fm(r[0].q, r[1].q);
+
+      FETCH(&r[1], 0, CHAN_Y);
+      FETCH(&r[2], 1, CHAN_Y);
+
+      r[0].q = si_fma(r[1].q, r[2].q, r[0].q);
+
+      FETCH(&r[1], 0, CHAN_Z);
+      FETCH(&r[2], 1, CHAN_Z);
+
+      r[0].q = si_fma(r[1].q, r[2].q, r[0].q);
+
+      FETCH(&r[1], 1, CHAN_W);
+
+      r[0].q = si_fa(r[0].q, r[1].q);
+
+      FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
+	 STORE( &r[0], 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_COS:
+      FETCH(&r[0], 0, CHAN_X);
+
+      r[0].q = micro_cos(r[0].q);
+
+      FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
+	 STORE( &r[0], 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_DDX:
+      FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
+         FETCH( &r[0], 0, chan_index );
+         r[0].q = micro_ddx(r[0].q);
+         STORE( &r[0], 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_DDY:
+      FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
+         FETCH( &r[0], 0, chan_index );
+         r[0].q = micro_ddy(r[0].q);
+         STORE( &r[0], 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_KILP:
+      exec_kilp (mach, inst);
+      break;
+
+   case TGSI_OPCODE_KIL:
+      /* for enabled ExecMask bits, set the killed bit */
+      mach->Temps[TEMP_KILMASK_I].xyzw[TEMP_KILMASK_C].u[0] |= mach->ExecMask;
+      break;
+
+   case TGSI_OPCODE_PK2H:
+      assert (0);
+      break;
+
+   case TGSI_OPCODE_PK2US:
+      assert (0);
+      break;
+
+   case TGSI_OPCODE_PK4B:
+      assert (0);
+      break;
+
+   case TGSI_OPCODE_PK4UB:
+      assert (0);
+      break;
+
+   case TGSI_OPCODE_RFL:
+      assert (0);
+      break;
+
+   case TGSI_OPCODE_SEQ:
+      FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
+         FETCH( &r[0], 0, chan_index );
+         FETCH( &r[1], 1, chan_index );
+
+         r[0].q = si_fceq(r[0].q, r[1].q);
+
+         STORE( &r[0], 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_SFL:
+      assert (0);
+      break;
+
+   case TGSI_OPCODE_SGT:
+      FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
+         FETCH( &r[0], 0, chan_index );
+         FETCH( &r[1], 1, chan_index );
+         r[0].q = si_fcgt(r[0].q, r[1].q);
+         STORE( &r[0], 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_SIN:
+      FETCH( &r[0], 0, CHAN_X );
+      r[0].q = micro_sin(r[0].q);
+      FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
+         STORE( &r[0], 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_SLE:
+      FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
+         FETCH( &r[0], 0, chan_index );
+         FETCH( &r[1], 1, chan_index );
+
+         r[0].q = si_fcgt(r[0].q, r[1].q);
+         r[0].q = si_xori(r[0].q, 0xff);
+
+         STORE( &r[0], 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_SNE:
+      FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
+         FETCH( &r[0], 0, chan_index );
+         FETCH( &r[1], 1, chan_index );
+
+         r[0].q = si_fceq(r[0].q, r[1].q);
+         r[0].q = si_xori(r[0].q, 0xff);
+
+         STORE( &r[0], 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_STR:
+      assert (0);
+      break;
+
+   case TGSI_OPCODE_TEX:
+      /* simple texture lookup */
+      /* src[0] = texcoord */
+      /* src[1] = sampler unit */
+      exec_tex(mach, inst, FALSE);
+      break;
+
+   case TGSI_OPCODE_TXB:
+      /* Texture lookup with lod bias */
+      /* src[0] = texcoord (src[0].w = load bias) */
+      /* src[1] = sampler unit */
+      exec_tex(mach, inst, TRUE);
+      break;
+
+   case TGSI_OPCODE_TXD:
+      /* Texture lookup with explict partial derivatives */
+      /* src[0] = texcoord */
+      /* src[1] = d[strq]/dx */
+      /* src[2] = d[strq]/dy */
+      /* src[3] = sampler unit */
+      assert (0);
+      break;
+
+   case TGSI_OPCODE_TXL:
+      /* Texture lookup with explit LOD */
+      /* src[0] = texcoord (src[0].w = load bias) */
+      /* src[1] = sampler unit */
+      exec_tex(mach, inst, TRUE);
+      break;
+
+   case TGSI_OPCODE_UP2H:
+      assert (0);
+      break;
+
+   case TGSI_OPCODE_UP2US:
+      assert (0);
+      break;
+
+   case TGSI_OPCODE_UP4B:
+      assert (0);
+      break;
+
+   case TGSI_OPCODE_UP4UB:
+      assert (0);
+      break;
+
+   case TGSI_OPCODE_X2D:
+      assert (0);
+      break;
+
+   case TGSI_OPCODE_ARA:
+      assert (0);
+      break;
+
+   case TGSI_OPCODE_ARR:
+      assert (0);
+      break;
+
+   case TGSI_OPCODE_BRA:
+      assert (0);
+      break;
+
+   case TGSI_OPCODE_CAL:
+      /* skip the call if no execution channels are enabled */
+      if (mach->ExecMask) {
+         /* do the call */
+
+         /* push the Cond, Loop, Cont stacks */
+         assert(mach->CondStackTop < TGSI_EXEC_MAX_COND_NESTING);
+         mach->CondStack[mach->CondStackTop++] = mach->CondMask;
+         assert(mach->LoopStackTop < TGSI_EXEC_MAX_LOOP_NESTING);
+         mach->LoopStack[mach->LoopStackTop++] = mach->LoopMask;
+         assert(mach->ContStackTop < TGSI_EXEC_MAX_LOOP_NESTING);
+         mach->ContStack[mach->ContStackTop++] = mach->ContMask;
+
+         assert(mach->FuncStackTop < TGSI_EXEC_MAX_CALL_NESTING);
+         mach->FuncStack[mach->FuncStackTop++] = mach->FuncMask;
+
+         /* note that PC was already incremented above */
+         mach->CallStack[mach->CallStackTop++] = *pc;
+         *pc = inst->InstructionExtLabel.Label;
+      }
+      break;
+
+   case TGSI_OPCODE_RET:
+      mach->FuncMask &= ~mach->ExecMask;
+      UPDATE_EXEC_MASK(mach);
+
+      if (mach->ExecMask == 0x0) {
+         /* really return now (otherwise, keep executing */
+
+         if (mach->CallStackTop == 0) {
+            /* returning from main() */
+            *pc = -1;
+            return;
+         }
+         *pc = mach->CallStack[--mach->CallStackTop];
+
+         /* pop the Cond, Loop, Cont stacks */
+         assert(mach->CondStackTop > 0);
+         mach->CondMask = mach->CondStack[--mach->CondStackTop];
+         assert(mach->LoopStackTop > 0);
+         mach->LoopMask = mach->LoopStack[--mach->LoopStackTop];
+         assert(mach->ContStackTop > 0);
+         mach->ContMask = mach->ContStack[--mach->ContStackTop];
+         assert(mach->FuncStackTop > 0);
+         mach->FuncMask = mach->FuncStack[--mach->FuncStackTop];
+
+         UPDATE_EXEC_MASK(mach);
+      }
+      break;
+
+   case TGSI_OPCODE_SSG:
+      assert (0);
+      break;
+
+   case TGSI_OPCODE_CMP:
+      FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
+         FETCH(&r[0], 0, chan_index);
+         FETCH(&r[1], 1, chan_index);
+         FETCH(&r[2], 2, chan_index);
+
+         /* r0 = (r0 < 0.0) ? r1 : r2
+          */
+         r[3].q = si_xor(r[3].q, r[3].q);
+         r[0].q = micro_lt(r[0].q, r[3].q);
+         r[0].q = si_selb(r[1].q, r[2].q, r[0].q);
+
+         STORE(&r[0], 0, chan_index);
+      }
+      break;
+
+   case TGSI_OPCODE_SCS:
+      if( IS_CHANNEL_ENABLED( *inst, CHAN_X ) || IS_CHANNEL_ENABLED( *inst, CHAN_Y ) ) {
+         FETCH( &r[0], 0, CHAN_X );
+      }
+      if( IS_CHANNEL_ENABLED( *inst, CHAN_X ) ) {
+         r[1].q = micro_cos(r[0].q);
+         STORE( &r[1], 0, CHAN_X );
+      }
+      if( IS_CHANNEL_ENABLED( *inst, CHAN_Y ) ) {
+         r[1].q = micro_sin(r[0].q);
+         STORE( &r[1], 0, CHAN_Y );
+      }
+      if( IS_CHANNEL_ENABLED( *inst, CHAN_Z ) ) {
+         STORE( &mach->Temps[TEMP_0_I].xyzw[TEMP_0_C], 0, CHAN_Z );
+      }
+      if( IS_CHANNEL_ENABLED( *inst, CHAN_W ) ) {
+         STORE( &mach->Temps[TEMP_1_I].xyzw[TEMP_1_C], 0, CHAN_W );
+      }
+      break;
+
+   case TGSI_OPCODE_NRM:
+      assert (0);
+      break;
+
+   case TGSI_OPCODE_DIV:
+      assert( 0 );
+      break;
+
+   case TGSI_OPCODE_DP2:
+      FETCH( &r[0], 0, CHAN_X );
+      FETCH( &r[1], 1, CHAN_X );
+      r[0].q = si_fm(r[0].q, r[1].q);
+
+      FETCH( &r[1], 0, CHAN_Y );
+      FETCH( &r[2], 1, CHAN_Y );
+      r[0].q = si_fma(r[1].q, r[2].q, r[0].q);
+
+      FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
+         STORE( &r[0], 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_IF:
+      /* push CondMask */
+      assert(mach->CondStackTop < TGSI_EXEC_MAX_COND_NESTING);
+      mach->CondStack[mach->CondStackTop++] = mach->CondMask;
+      FETCH( &r[0], 0, CHAN_X );
+      /* update CondMask */
+      if( ! r[0].u[0] ) {
+         mach->CondMask &= ~0x1;
+      }
+      if( ! r[0].u[1] ) {
+         mach->CondMask &= ~0x2;
+      }
+      if( ! r[0].u[2] ) {
+         mach->CondMask &= ~0x4;
+      }
+      if( ! r[0].u[3] ) {
+         mach->CondMask &= ~0x8;
+      }
+      UPDATE_EXEC_MASK(mach);
+      /* Todo: If CondMask==0, jump to ELSE */
+      break;
+
+   case TGSI_OPCODE_ELSE:
+      /* invert CondMask wrt previous mask */
+      {
+         uint prevMask;
+         assert(mach->CondStackTop > 0);
+         prevMask = mach->CondStack[mach->CondStackTop - 1];
+         mach->CondMask = ~mach->CondMask & prevMask;
+         UPDATE_EXEC_MASK(mach);
+         /* Todo: If CondMask==0, jump to ENDIF */
+      }
+      break;
+
+   case TGSI_OPCODE_ENDIF:
+      /* pop CondMask */
+      assert(mach->CondStackTop > 0);
+      mach->CondMask = mach->CondStack[--mach->CondStackTop];
+      UPDATE_EXEC_MASK(mach);
+      break;
+
+   case TGSI_OPCODE_END:
+      /* halt execution */
+      *pc = -1;
+      break;
+
+   case TGSI_OPCODE_REP:
+      assert (0);
+      break;
+
+   case TGSI_OPCODE_ENDREP:
+       assert (0);
+       break;
+
+   case TGSI_OPCODE_PUSHA:
+      assert (0);
+      break;
+
+   case TGSI_OPCODE_POPA:
+      assert (0);
+      break;
+
+   case TGSI_OPCODE_CEIL:
+      FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
+         FETCH( &r[0], 0, chan_index );
+         r[0].q = micro_ceil(r[0].q);
+         STORE( &r[0], 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_I2F:
+      FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
+         FETCH( &r[0], 0, chan_index );
+         r[0].q = si_csflt(r[0].q, 0);
+         STORE( &r[0], 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_NOT:
+      FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
+         FETCH( &r[0], 0, chan_index );
+         r[0].q = si_xorbi(r[0].q, 0xff);
+         STORE( &r[0], 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_TRUNC:
+      FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
+         FETCH( &r[0], 0, chan_index );
+         r[0].q = micro_trunc(r[0].q);
+         STORE( &r[0], 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_SHL:
+      FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
+         FETCH( &r[0], 0, chan_index );
+         FETCH( &r[1], 1, chan_index );
+
+         r[0].q = si_shl(r[0].q, r[1].q);
+
+         STORE( &r[0], 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_SHR:
+      FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
+         FETCH( &r[0], 0, chan_index );
+         FETCH( &r[1], 1, chan_index );
+         r[0].q = micro_ishr(r[0].q, r[1].q);
+         STORE( &r[0], 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_AND:
+      FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
+         FETCH( &r[0], 0, chan_index );
+         FETCH( &r[1], 1, chan_index );
+         r[0].q = si_and(r[0].q, r[1].q);
+         STORE( &r[0], 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_OR:
+      FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
+         FETCH( &r[0], 0, chan_index );
+         FETCH( &r[1], 1, chan_index );
+         r[0].q = si_or(r[0].q, r[1].q);
+         STORE( &r[0], 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_MOD:
+      assert (0);
+      break;
+
+   case TGSI_OPCODE_XOR:
+      FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
+         FETCH( &r[0], 0, chan_index );
+         FETCH( &r[1], 1, chan_index );
+         r[0].q = si_xor(r[0].q, r[1].q);
+         STORE( &r[0], 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_SAD:
+      assert (0);
+      break;
+
+   case TGSI_OPCODE_TXF:
+      assert (0);
+      break;
+
+   case TGSI_OPCODE_TXQ:
+      assert (0);
+      break;
+
+   case TGSI_OPCODE_EMIT:
+      mach->Temps[TEMP_OUTPUT_I].xyzw[TEMP_OUTPUT_C].u[0] += 16;
+      mach->Primitives[mach->Temps[TEMP_PRIMITIVE_I].xyzw[TEMP_PRIMITIVE_C].u[0]]++;
+      break;
+
+   case TGSI_OPCODE_ENDPRIM:
+      mach->Temps[TEMP_PRIMITIVE_I].xyzw[TEMP_PRIMITIVE_C].u[0]++;
+      mach->Primitives[mach->Temps[TEMP_PRIMITIVE_I].xyzw[TEMP_PRIMITIVE_C].u[0]] = 0;
+      break;
+
+   case TGSI_OPCODE_LOOP:
+      /* fall-through (for now) */
+   case TGSI_OPCODE_BGNLOOP2:
+      /* push LoopMask and ContMasks */
+      assert(mach->LoopStackTop < TGSI_EXEC_MAX_LOOP_NESTING);
+      mach->LoopStack[mach->LoopStackTop++] = mach->LoopMask;
+      assert(mach->ContStackTop < TGSI_EXEC_MAX_LOOP_NESTING);
+      mach->ContStack[mach->ContStackTop++] = mach->ContMask;
+      break;
+
+   case TGSI_OPCODE_ENDLOOP:
+      /* fall-through (for now at least) */
+   case TGSI_OPCODE_ENDLOOP2:
+      /* Restore ContMask, but don't pop */
+      assert(mach->ContStackTop > 0);
+      mach->ContMask = mach->ContStack[mach->ContStackTop - 1];
+      if (mach->LoopMask) {
+         /* repeat loop: jump to instruction just past BGNLOOP */
+         *pc = inst->InstructionExtLabel.Label + 1;
+      }
+      else {
+         /* exit loop: pop LoopMask */
+         assert(mach->LoopStackTop > 0);
+         mach->LoopMask = mach->LoopStack[--mach->LoopStackTop];
+         /* pop ContMask */
+         assert(mach->ContStackTop > 0);
+         mach->ContMask = mach->ContStack[--mach->ContStackTop];
+      }
+      UPDATE_EXEC_MASK(mach);
+      break;
+
+   case TGSI_OPCODE_BRK:
+      /* turn off loop channels for each enabled exec channel */
+      mach->LoopMask &= ~mach->ExecMask;
+      /* Todo: if mach->LoopMask == 0, jump to end of loop */
+      UPDATE_EXEC_MASK(mach);
+      break;
+
+   case TGSI_OPCODE_CONT:
+      /* turn off cont channels for each enabled exec channel */
+      mach->ContMask &= ~mach->ExecMask;
+      /* Todo: if mach->LoopMask == 0, jump to end of loop */
+      UPDATE_EXEC_MASK(mach);
+      break;
+
+   case TGSI_OPCODE_BGNSUB:
+      /* no-op */
+      break;
+
+   case TGSI_OPCODE_ENDSUB:
+      /* no-op */
+      break;
+
+   case TGSI_OPCODE_NOISE1:
+      assert( 0 );
+      break;
+
+   case TGSI_OPCODE_NOISE2:
+      assert( 0 );
+      break;
+
+   case TGSI_OPCODE_NOISE3:
+      assert( 0 );
+      break;
+
+   case TGSI_OPCODE_NOISE4:
+      assert( 0 );
+      break;
+
+   case TGSI_OPCODE_NOP:
+      break;
+
+   default:
+      assert( 0 );
+   }
+}
+
+
+/**
+ * Run TGSI interpreter.
+ * \return bitmask of "alive" quad components
+ */
+uint
+spu_exec_machine_run( struct spu_exec_machine *mach )
+{
+   uint i;
+   int pc = 0;
+
+   mach->CondMask = 0xf;
+   mach->LoopMask = 0xf;
+   mach->ContMask = 0xf;
+   mach->FuncMask = 0xf;
+   mach->ExecMask = 0xf;
+
+   mach->CondStackTop = 0; /* temporarily subvert this assertion */
+   assert(mach->CondStackTop == 0);
+   assert(mach->LoopStackTop == 0);
+   assert(mach->ContStackTop == 0);
+   assert(mach->CallStackTop == 0);
+
+   mach->Temps[TEMP_KILMASK_I].xyzw[TEMP_KILMASK_C].u[0] = 0;
+   mach->Temps[TEMP_OUTPUT_I].xyzw[TEMP_OUTPUT_C].u[0] = 0;
+
+   if( mach->Processor == TGSI_PROCESSOR_GEOMETRY ) {
+      mach->Temps[TEMP_PRIMITIVE_I].xyzw[TEMP_PRIMITIVE_C].u[0] = 0;
+      mach->Primitives[0] = 0;
+   }
+
+
+   /* execute declarations (interpolants) */
+   if( mach->Processor == TGSI_PROCESSOR_FRAGMENT ) {
+      for (i = 0; i < mach->NumDeclarations; i++) {
+	 uint8_t buffer[sizeof(struct tgsi_full_declaration) + 32] ALIGN16_ATTRIB;
+	 struct tgsi_full_declaration decl;
+	 unsigned long decl_addr = (unsigned long) (mach->Declarations+i);
+	 unsigned size = ((sizeof(decl) + (decl_addr & 0x0f) + 0x0f) & ~0x0f);
+
+	 mfc_get(buffer, decl_addr & ~0x0f, size, TAG_INSTRUCTION_FETCH, 0, 0);
+	 wait_on_mask(1 << TAG_INSTRUCTION_FETCH);
+
+	 memcpy(& decl, buffer + (decl_addr & 0x0f), sizeof(decl));
+	 exec_declaration( mach, &decl );
+      }
+   }
+
+   /* execute instructions, until pc is set to -1 */
+   while (pc != -1) {
+      uint8_t buffer[sizeof(struct tgsi_full_instruction) + 32] ALIGN16_ATTRIB;
+      struct tgsi_full_instruction inst;
+      unsigned long inst_addr = (unsigned long) (mach->Instructions + pc);
+      unsigned size = ((sizeof(inst) + (inst_addr & 0x0f) + 0x0f) & ~0x0f);
+
+      assert(pc < mach->NumInstructions);
+      mfc_get(buffer, inst_addr & ~0x0f, size, TAG_INSTRUCTION_FETCH, 0, 0);
+      wait_on_mask(1 << TAG_INSTRUCTION_FETCH);
+
+      memcpy(& inst, buffer + (inst_addr & 0x0f), sizeof(inst));
+      exec_instruction( mach, & inst, &pc );
+   }
+
+#if 0
+   /* we scale from floats in [0,1] to Zbuffer ints in sp_quad_depth_test.c */
+   if (mach->Processor == TGSI_PROCESSOR_FRAGMENT) {
+      /*
+       * Scale back depth component.
+       */
+      for (i = 0; i < 4; i++)
+         mach->Outputs[0].xyzw[2].f[i] *= ctx->DrawBuffer->_DepthMaxF;
+   }
+#endif
+
+   return ~mach->Temps[TEMP_KILMASK_I].xyzw[TEMP_KILMASK_C].u[0];
+}
+
+
diff --git a/src/gallium/drivers/cell/spu/spu_exec.h b/src/gallium/drivers/cell/spu/spu_exec.h
new file mode 100644
index 0000000000..b4c7661ef6
--- /dev/null
+++ b/src/gallium/drivers/cell/spu/spu_exec.h
@@ -0,0 +1,172 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+#if !defined SPU_EXEC_H
+#define SPU_EXEC_H
+
+#include "pipe/p_compiler.h"
+#include "pipe/tgsi/exec/tgsi_exec.h"
+
+#if defined __cplusplus
+extern "C" {
+#endif
+
+/**
+  * Registers may be treated as float, signed int or unsigned int.
+  */
+union spu_exec_channel
+{
+   float    f[QUAD_SIZE];
+   int      i[QUAD_SIZE];
+   unsigned u[QUAD_SIZE];
+   qword    q;
+};
+
+/**
+  * A vector[RGBA] of channels[4 pixels]
+  */
+struct spu_exec_vector
+{
+   union spu_exec_channel xyzw[NUM_CHANNELS];
+};
+
+/**
+ * For fragment programs, information for computing fragment input
+ * values from plane equation of the triangle/line.
+ */
+struct spu_interp_coef
+{
+   float a0[NUM_CHANNELS];	/* in an xyzw layout */
+   float dadx[NUM_CHANNELS];
+   float dady[NUM_CHANNELS];
+};
+
+
+struct softpipe_tile_cache;  /**< Opaque to TGSI */
+
+/**
+ * Information for sampling textures, which must be implemented
+ * by code outside the TGSI executor.
+ */
+struct spu_sampler
+{
+   const struct pipe_sampler_state *state;
+   struct pipe_texture *texture;
+   /** Get samples for four fragments in a quad */
+   void (*get_samples)(struct spu_sampler *sampler,
+                       const float s[QUAD_SIZE],
+                       const float t[QUAD_SIZE],
+                       const float p[QUAD_SIZE],
+                       float lodbias,
+                       float rgba[NUM_CHANNELS][QUAD_SIZE]);
+   void *pipe; /*XXX temporary*/
+   struct softpipe_tile_cache *cache;
+};
+
+
+/**
+ * Run-time virtual machine state for executing TGSI shader.
+ */
+struct spu_exec_machine
+{
+   /*
+    * 32 program temporaries
+    * 4  internal temporaries
+    * 1  address
+    */
+   struct spu_exec_vector       Temps[TGSI_EXEC_NUM_TEMPS 
+				      + TGSI_EXEC_NUM_ADDRS + 1]
+       ALIGN16_ATTRIB;
+
+   struct spu_exec_vector       *Addrs;
+
+   struct spu_sampler           *Samplers;
+
+   float                         Imms[TGSI_EXEC_NUM_IMMEDIATES][4];
+   unsigned                      ImmLimit;
+   float                         (*Consts)[4];
+   struct spu_exec_vector       *Inputs;
+   struct spu_exec_vector       *Outputs;
+   unsigned                      Processor;
+
+   /* GEOMETRY processor only. */
+   unsigned                      *Primitives;
+
+   /* FRAGMENT processor only. */
+   const struct spu_interp_coef *InterpCoefs;
+   struct spu_exec_vector       QuadPos;
+
+   /* Conditional execution masks */
+   uint CondMask;  /**< For IF/ELSE/ENDIF */
+   uint LoopMask;  /**< For BGNLOOP/ENDLOOP */
+   uint ContMask;  /**< For loop CONT statements */
+   uint FuncMask;  /**< For function calls */
+   uint ExecMask;  /**< = CondMask & LoopMask */
+
+   /** Condition mask stack (for nested conditionals) */
+   uint CondStack[TGSI_EXEC_MAX_COND_NESTING];
+   int CondStackTop;
+
+   /** Loop mask stack (for nested loops) */
+   uint LoopStack[TGSI_EXEC_MAX_LOOP_NESTING];
+   int LoopStackTop;
+
+   /** Loop continue mask stack (see comments in tgsi_exec.c) */
+   uint ContStack[TGSI_EXEC_MAX_LOOP_NESTING];
+   int ContStackTop;
+
+   /** Function execution mask stack (for executing subroutine code) */
+   uint FuncStack[TGSI_EXEC_MAX_CALL_NESTING];
+   int FuncStackTop;
+
+   /** Function call stack for saving/restoring the program counter */
+   uint CallStack[TGSI_EXEC_MAX_CALL_NESTING];
+   int CallStackTop;
+
+   struct tgsi_full_instruction *Instructions;
+   uint NumInstructions;
+
+   struct tgsi_full_declaration *Declarations;
+   uint NumDeclarations;
+};
+
+
+extern void
+spu_exec_machine_init(struct spu_exec_machine *mach,
+                      uint numSamplers,
+                      struct spu_sampler *samplers,
+                      unsigned processor);
+
+extern uint
+spu_exec_machine_run( struct spu_exec_machine *mach );
+
+
+#if defined __cplusplus
+} /* extern "C" */
+#endif
+
+#endif /* SPU_EXEC_H */
diff --git a/src/gallium/drivers/cell/spu/spu_main.c b/src/gallium/drivers/cell/spu/spu_main.c
new file mode 100644
index 0000000000..e375197fe6
--- /dev/null
+++ b/src/gallium/drivers/cell/spu/spu_main.c
@@ -0,0 +1,567 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+
+/* main() for Cell SPU code */
+
+
+#include <stdio.h>
+#include <libmisc.h>
+
+#include "spu_main.h"
+#include "spu_render.h"
+#include "spu_texture.h"
+#include "spu_tile.h"
+//#include "spu_test.h"
+#include "spu_vertex_shader.h"
+#include "pipe/cell/common.h"
+#include "pipe/p_defines.h"
+
+
+/*
+helpful headers:
+/usr/lib/gcc/spu/4.1.1/include/spu_mfcio.h
+/opt/ibm/cell-sdk/prototype/sysroot/usr/include/libmisc.h
+*/
+
+boolean Debug = FALSE;
+
+struct spu_global spu;
+
+struct spu_vs_context draw;
+
+/**
+ * Tell the PPU that this SPU has finished copying a buffer to
+ * local store and that it may be reused by the PPU.
+ * This is done by writting a 16-byte batch-buffer-status block back into
+ * main memory (in cell_context->buffer_status[]).
+ */
+static void
+release_buffer(uint buffer)
+{
+   /* Evidently, using less than a 16-byte status doesn't work reliably */
+   static const uint status[4] ALIGN16_ATTRIB
+      = {CELL_BUFFER_STATUS_FREE, 0, 0, 0};
+
+   const uint index = 4 * (spu.init.id * CELL_NUM_BUFFERS + buffer);
+   uint *dst = spu.init.buffer_status + index;
+
+   ASSERT(buffer < CELL_NUM_BUFFERS);
+
+   mfc_put((void *) &status,    /* src in local memory */
+           (unsigned int) dst,  /* dst in main memory */
+           sizeof(status),      /* size */
+           TAG_MISC,            /* tag is unimportant */
+           0, /* tid */
+           0  /* rid */);
+}
+
+
+/**
+ * For tiles whose status is TILE_STATUS_CLEAR, write solid-filled
+ * tiles back to the main framebuffer.
+ */
+static void
+really_clear_tiles(uint surfaceIndex)
+{
+   const uint num_tiles = spu.fb.width_tiles * spu.fb.height_tiles;
+   uint i;
+
+   if (surfaceIndex == 0) {
+      clear_c_tile(&spu.ctile);
+
+      for (i = spu.init.id; i < num_tiles; i += spu.init.num_spus) {
+         uint tx = i % spu.fb.width_tiles;
+         uint ty = i / spu.fb.width_tiles;
+         if (spu.ctile_status[ty][tx] == TILE_STATUS_CLEAR) {
+            put_tile(tx, ty, &spu.ctile, TAG_SURFACE_CLEAR, 0);
+         }
+      }
+   }
+   else {
+      clear_z_tile(&spu.ztile);
+
+      for (i = spu.init.id; i < num_tiles; i += spu.init.num_spus) {
+         uint tx = i % spu.fb.width_tiles;
+         uint ty = i / spu.fb.width_tiles;
+         if (spu.ztile_status[ty][tx] == TILE_STATUS_CLEAR)
+            put_tile(tx, ty, &spu.ctile, TAG_SURFACE_CLEAR, 1);
+      }
+   }
+
+#if 0
+   wait_on_mask(1 << TAG_SURFACE_CLEAR);
+#endif
+}
+
+
+static void
+cmd_clear_surface(const struct cell_command_clear_surface *clear)
+{
+   const uint num_tiles = spu.fb.width_tiles * spu.fb.height_tiles;
+   uint i;
+
+   if (Debug)
+      printf("SPU %u: CLEAR SURF %u to 0x%08x\n", spu.init.id,
+             clear->surface, clear->value);
+
+#define CLEAR_OPT 1
+#if CLEAR_OPT
+   /* set all tile's status to CLEAR */
+   if (clear->surface == 0) {
+      memset(spu.ctile_status, TILE_STATUS_CLEAR, sizeof(spu.ctile_status));
+      spu.fb.color_clear_value = clear->value;
+   }
+   else {
+      memset(spu.ztile_status, TILE_STATUS_CLEAR, sizeof(spu.ztile_status));
+      spu.fb.depth_clear_value = clear->value;
+   }
+   return;
+#endif
+
+   if (clear->surface == 0) {
+      spu.fb.color_clear_value = clear->value;
+      clear_c_tile(&spu.ctile);
+   }
+   else {
+      spu.fb.depth_clear_value = clear->value;
+      clear_z_tile(&spu.ztile);
+   }
+
+   /*
+   printf("SPU: %s num=%d w=%d h=%d\n",
+          __FUNCTION__, num_tiles, spu.fb.width_tiles, spu.fb.height_tiles);
+   */
+
+   for (i = spu.init.id; i < num_tiles; i += spu.init.num_spus) {
+      uint tx = i % spu.fb.width_tiles;
+      uint ty = i / spu.fb.width_tiles;
+      if (clear->surface == 0)
+         put_tile(tx, ty, &spu.ctile, TAG_SURFACE_CLEAR, 0);
+      else
+         put_tile(tx, ty, &spu.ztile, TAG_SURFACE_CLEAR, 1);
+      /* XXX we don't want this here, but it fixes bad tile results */
+   }
+
+#if 0
+   wait_on_mask(1 << TAG_SURFACE_CLEAR);
+#endif
+
+   if (Debug)
+      printf("SPU %u: CLEAR SURF done\n", spu.init.id);
+}
+
+
+static void
+cmd_release_verts(const struct cell_command_release_verts *release)
+{
+   if (Debug)
+      printf("SPU %u: RELEASE VERTS %u\n",
+             spu.init.id, release->vertex_buf);
+   ASSERT(release->vertex_buf != ~0U);
+   release_buffer(release->vertex_buf);
+}
+
+
+static void
+cmd_state_framebuffer(const struct cell_command_framebuffer *cmd)
+{
+   if (Debug)
+      printf("SPU %u: FRAMEBUFFER: %d x %d at %p, cformat 0x%x  zformat 0x%x\n",
+             spu.init.id,
+             cmd->width,
+             cmd->height,
+             cmd->color_start,
+             cmd->color_format,
+             cmd->depth_format);
+
+   ASSERT_ALIGN16(cmd->color_start);
+   ASSERT_ALIGN16(cmd->depth_start);
+
+   spu.fb.color_start = cmd->color_start;
+   spu.fb.depth_start = cmd->depth_start;
+   spu.fb.color_format = cmd->color_format;
+   spu.fb.depth_format = cmd->depth_format;
+   spu.fb.width = cmd->width;
+   spu.fb.height = cmd->height;
+   spu.fb.width_tiles = (spu.fb.width + TILE_SIZE - 1) / TILE_SIZE;
+   spu.fb.height_tiles = (spu.fb.height + TILE_SIZE - 1) / TILE_SIZE;
+
+   if (spu.fb.depth_format == PIPE_FORMAT_Z32_UNORM)
+      spu.fb.zsize = 4;
+   else if (spu.fb.depth_format == PIPE_FORMAT_Z16_UNORM)
+      spu.fb.zsize = 2;
+   else
+      spu.fb.zsize = 0;
+
+   if (spu.fb.color_format == PIPE_FORMAT_A8R8G8B8_UNORM)
+      spu.color_shuffle = ((vector unsigned char) {
+                              12, 0, 4, 8, 0, 0, 0, 0, 
+                              0, 0, 0, 0, 0, 0, 0, 0});
+   else if (spu.fb.color_format == PIPE_FORMAT_B8G8R8A8_UNORM)
+      spu.color_shuffle = ((vector unsigned char) {
+                              8, 4, 0, 12, 0, 0, 0, 0, 
+                              0, 0, 0, 0, 0, 0, 0, 0});
+   else
+      ASSERT(0);
+}
+
+
+static void
+cmd_state_blend(const struct pipe_blend_state *state)
+{
+   if (Debug)
+      printf("SPU %u: BLEND: enabled %d\n",
+             spu.init.id,
+             state->blend_enable);
+
+   memcpy(&spu.blend, state, sizeof(*state));
+}
+
+
+static void
+cmd_state_depth_stencil(const struct pipe_depth_stencil_alpha_state *state)
+{
+   if (Debug)
+      printf("SPU %u: DEPTH_STENCIL: ztest %d\n",
+             spu.init.id,
+             state->depth.enabled);
+
+   memcpy(&spu.depth_stencil, state, sizeof(*state));
+}
+
+
+static void
+cmd_state_sampler(const struct pipe_sampler_state *state)
+{
+   if (Debug)
+      printf("SPU %u: SAMPLER\n",
+             spu.init.id);
+
+   memcpy(&spu.sampler[0], state, sizeof(*state));
+   if (spu.sampler[0].min_img_filter == PIPE_TEX_FILTER_LINEAR)
+      spu.sample_texture = sample_texture_bilinear;
+   else
+      spu.sample_texture = sample_texture_nearest;
+}
+
+
+static void
+cmd_state_texture(const struct cell_command_texture *texture)
+{
+   if (Debug)
+      printf("SPU %u: TEXTURE at %p  size %u x %u\n",
+             spu.init.id, texture->start, texture->width, texture->height);
+
+   memcpy(&spu.texture, texture, sizeof(*texture));
+   spu.tex_size = (vector float)
+      { spu.texture.width, spu.texture.height, 0.0, 0.0};
+   spu.tex_size_mask = (vector unsigned int)
+      { spu.texture.width - 1, spu.texture.height - 1, 0, 0 };
+}
+
+
+static void
+cmd_state_vertex_info(const struct vertex_info *vinfo)
+{
+   if (Debug) {
+      printf("SPU %u: VERTEX_INFO num_attribs=%u\n", spu.init.id,
+             vinfo->num_attribs);
+   }
+   ASSERT(vinfo->num_attribs >= 1);
+   ASSERT(vinfo->num_attribs <= 8);
+   memcpy(&spu.vertex_info, vinfo, sizeof(*vinfo));
+}
+
+
+static void
+cmd_state_vs_array_info(const struct cell_array_info *vs_info)
+{
+   const unsigned attr = vs_info->attr;
+
+   ASSERT(attr < PIPE_ATTRIB_MAX);
+   draw.vertex_fetch.src_ptr[attr] = vs_info->base;
+   draw.vertex_fetch.pitch[attr] = vs_info->pitch;
+   draw.vertex_fetch.format[attr] = vs_info->format;
+   draw.vertex_fetch.dirty = 1;
+}
+
+
+static void
+cmd_finish(void)
+{
+   if (Debug)
+      printf("SPU %u: FINISH\n", spu.init.id);
+   really_clear_tiles(0);
+   /* wait for all outstanding DMAs to finish */
+   mfc_write_tag_mask(~0);
+   mfc_read_tag_status_all();
+   /* send mbox message to PPU */
+   spu_write_out_mbox(CELL_CMD_FINISH);
+}
+
+
+/**
+ * Execute a batch of commands
+ * The opcode param encodes the location of the buffer and its size.
+ */
+static void
+cmd_batch(uint opcode)
+{
+   const uint buf = (opcode >> 8) & 0xff;
+   uint size = (opcode >> 16);
+   uint64_t buffer[CELL_BUFFER_SIZE / 8] ALIGN16_ATTRIB;
+   const unsigned usize = size / sizeof(buffer[0]);
+   uint pos;
+
+   if (Debug)
+      printf("SPU %u: BATCH buffer %u, len %u, from %p\n",
+             spu.init.id, buf, size, spu.init.buffers[buf]);
+
+   ASSERT((opcode & CELL_CMD_OPCODE_MASK) == CELL_CMD_BATCH);
+
+   ASSERT_ALIGN16(spu.init.buffers[buf]);
+
+   size = ROUNDUP16(size);
+
+   ASSERT_ALIGN16(spu.init.buffers[buf]);
+
+   mfc_get(buffer,  /* dest */
+           (unsigned int) spu.init.buffers[buf],  /* src */
+           size,
+           TAG_BATCH_BUFFER,
+           0, /* tid */
+           0  /* rid */);
+   wait_on_mask(1 << TAG_BATCH_BUFFER);
+
+   /* Tell PPU we're done copying the buffer to local store */
+   if (Debug)
+      printf("SPU %u: release batch buf %u\n", spu.init.id, buf);
+   release_buffer(buf);
+
+   for (pos = 0; pos < usize; /* no incr */) {
+      switch (buffer[pos]) {
+      case CELL_CMD_STATE_FRAMEBUFFER:
+         {
+            struct cell_command_framebuffer *fb
+               = (struct cell_command_framebuffer *) &buffer[pos];
+            cmd_state_framebuffer(fb);
+            pos += sizeof(*fb) / 8;
+         }
+         break;
+      case CELL_CMD_CLEAR_SURFACE:
+         {
+            struct cell_command_clear_surface *clr
+               = (struct cell_command_clear_surface *) &buffer[pos];
+            cmd_clear_surface(clr);
+            pos += sizeof(*clr) / 8;
+         }
+         break;
+      case CELL_CMD_RENDER:
+         {
+            struct cell_command_render *render
+               = (struct cell_command_render *) &buffer[pos];
+            uint pos_incr;
+            cmd_render(render, &pos_incr);
+            pos += pos_incr;
+         }
+         break;
+      case CELL_CMD_RELEASE_VERTS:
+         {
+            struct cell_command_release_verts *release
+               = (struct cell_command_release_verts *) &buffer[pos];
+            cmd_release_verts(release);
+            pos += sizeof(*release) / 8;
+         }
+         break;
+      case CELL_CMD_FINISH:
+         cmd_finish();
+         pos += 1;
+         break;
+      case CELL_CMD_STATE_BLEND:
+         cmd_state_blend((struct pipe_blend_state *)
+                                 &buffer[pos+1]);
+         pos += (1 + ROUNDUP8(sizeof(struct pipe_blend_state)) / 8);
+         break;
+      case CELL_CMD_STATE_DEPTH_STENCIL:
+         cmd_state_depth_stencil((struct pipe_depth_stencil_alpha_state *)
+                                 &buffer[pos+1]);
+         pos += (1 + ROUNDUP8(sizeof(struct pipe_depth_stencil_alpha_state)) / 8);
+         break;
+      case CELL_CMD_STATE_SAMPLER:
+         cmd_state_sampler((struct pipe_sampler_state *) &buffer[pos+1]);
+         pos += (1 + ROUNDUP8(sizeof(struct pipe_sampler_state)) / 8);
+         break;
+      case CELL_CMD_STATE_TEXTURE:
+         cmd_state_texture((struct cell_command_texture *) &buffer[pos+1]);
+         pos += (1 + ROUNDUP8(sizeof(struct cell_command_texture)) / 8);
+         break;
+      case CELL_CMD_STATE_VERTEX_INFO:
+         cmd_state_vertex_info((struct vertex_info *) &buffer[pos+1]);
+         pos += (1 + ROUNDUP8(sizeof(struct vertex_info)) / 8);
+         break;
+      case CELL_CMD_STATE_VIEWPORT:
+         (void) memcpy(& draw.viewport, &buffer[pos+1],
+                       sizeof(struct pipe_viewport_state));
+         pos += (1 + ROUNDUP8(sizeof(struct pipe_viewport_state)) / 8);
+         break;
+      case CELL_CMD_STATE_VS_ARRAY_INFO:
+         cmd_state_vs_array_info((struct cell_array_info *) &buffer[pos+1]);
+         pos += (1 + ROUNDUP8(sizeof(struct cell_array_info)) / 8);
+         break;
+      default:
+         printf("SPU %u: bad opcode: 0x%llx\n", spu.init.id, buffer[pos]);
+         ASSERT(0);
+         break;
+      }
+   }
+
+   if (Debug)
+      printf("SPU %u: BATCH complete\n", spu.init.id);
+}
+
+
+/**
+ * Temporary/simple main loop for SPEs: Get a command, execute it, repeat.
+ */
+static void
+main_loop(void)
+{
+   struct cell_command cmd;
+   int exitFlag = 0;
+
+   if (Debug)
+      printf("SPU %u: Enter main loop\n", spu.init.id);
+
+   ASSERT((sizeof(struct cell_command) & 0xf) == 0);
+   ASSERT_ALIGN16(&cmd);
+
+   while (!exitFlag) {
+      unsigned opcode;
+      int tag = 0;
+
+      if (Debug)
+         printf("SPU %u: Wait for cmd...\n", spu.init.id);
+
+      /* read/wait from mailbox */
+      opcode = (unsigned int) spu_read_in_mbox();
+
+      if (Debug)
+         printf("SPU %u: got cmd 0x%x\n", spu.init.id, opcode);
+
+      /* command payload */
+      mfc_get(&cmd,  /* dest */
+              (unsigned int) spu.init.cmd, /* src */
+              sizeof(struct cell_command), /* bytes */
+              tag,
+              0, /* tid */
+              0  /* rid */);
+      wait_on_mask( 1 << tag );
+
+      /*
+       * NOTE: most commands should be contained in a batch buffer
+       */
+
+      switch (opcode & CELL_CMD_OPCODE_MASK) {
+      case CELL_CMD_EXIT:
+         if (Debug)
+            printf("SPU %u: EXIT\n", spu.init.id);
+         exitFlag = 1;
+         break;
+      case CELL_CMD_VS_EXECUTE:
+         spu_execute_vertex_shader(&draw, &cmd.vs);
+         break;
+      case CELL_CMD_BATCH:
+         cmd_batch(opcode);
+         break;
+      default:
+         printf("Bad opcode!\n");
+      }
+
+   }
+
+   if (Debug)
+      printf("SPU %u: Exit main loop\n", spu.init.id);
+}
+
+
+
+static void
+one_time_init(void)
+{
+   memset(spu.ctile_status, TILE_STATUS_DEFINED, sizeof(spu.ctile_status));
+   memset(spu.ztile_status, TILE_STATUS_DEFINED, sizeof(spu.ztile_status));
+   invalidate_tex_cache();
+}
+
+
+
+/* In some versions of the SDK the SPE main takes 'unsigned long' as a
+ * parameter.  In others it takes 'unsigned long long'.  Use a define to
+ * select between the two.
+ */
+#ifdef SPU_MAIN_PARAM_LONG_LONG
+typedef unsigned long long main_param_t;
+#else
+typedef unsigned long main_param_t;
+#endif
+
+/**
+ * SPE entrypoint.
+ */
+int
+main(main_param_t speid, main_param_t argp)
+{
+   int tag = 0;
+
+   (void) speid;
+
+   ASSERT(sizeof(tile_t) == TILE_SIZE * TILE_SIZE * 4);
+   ASSERT(sizeof(struct cell_command_render) % 8 == 0);
+
+   one_time_init();
+
+   if (Debug)
+      printf("SPU: main() speid=%lu\n", speid);
+
+   mfc_get(&spu.init,  /* dest */
+           (unsigned int) argp, /* src */
+           sizeof(struct cell_init_info), /* bytes */
+           tag,
+           0, /* tid */
+           0  /* rid */);
+   wait_on_mask( 1 << tag );
+
+#if 0
+   if (spu.init.id==0)
+      spu_test_misc();
+#endif
+
+   main_loop();
+
+   return 0;
+}
diff --git a/src/gallium/drivers/cell/spu/spu_main.h b/src/gallium/drivers/cell/spu/spu_main.h
new file mode 100644
index 0000000000..1710a17512
--- /dev/null
+++ b/src/gallium/drivers/cell/spu/spu_main.h
@@ -0,0 +1,177 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+#ifndef SPU_MAIN_H
+#define SPU_MAIN_H
+
+
+#include <spu_mfcio.h>
+
+#include "pipe/cell/common.h"
+#include "pipe/draw/draw_vertex.h"
+#include "pipe/p_state.h"
+
+
+
+#define MAX_WIDTH 1024
+#define MAX_HEIGHT 1024
+
+
+typedef union {
+   ushort us[TILE_SIZE][TILE_SIZE];
+   uint   ui[TILE_SIZE][TILE_SIZE];
+   vector unsigned short us8[TILE_SIZE/2][TILE_SIZE/4];
+   vector unsigned int ui4[TILE_SIZE/2][TILE_SIZE/2];
+} tile_t;
+
+
+#define TILE_STATUS_CLEAR   1
+#define TILE_STATUS_DEFINED 2  /**< defined in FB, but not in local store */
+#define TILE_STATUS_CLEAN   3  /**< in local store, but not changed */
+#define TILE_STATUS_DIRTY   4  /**< modified locally, but not put back yet */
+#define TILE_STATUS_GETTING 5  /**< mfc_get() called but not yet arrived */
+
+
+struct spu_framebuffer {
+   void *color_start;              /**< addr of color surface in main memory */
+   void *depth_start;              /**< addr of depth surface in main memory */
+   enum pipe_format color_format;
+   enum pipe_format depth_format;
+   uint width, height;             /**< size in pixels */
+   uint width_tiles, height_tiles; /**< width and height in tiles */
+
+   uint color_clear_value;
+   uint depth_clear_value;
+
+   uint zsize;                     /**< 0, 2 or 4 bytes per Z */
+} ALIGN16_ATTRIB;
+
+
+/**
+ * All SPU global/context state will be in singleton object of this type:
+ */
+struct spu_global
+{
+   struct cell_init_info init;
+
+   struct spu_framebuffer fb;
+   struct pipe_blend_state blend_stencil;
+   struct pipe_depth_stencil_alpha_state depth_stencil;
+   struct pipe_blend_state blend;
+   struct pipe_sampler_state sampler[PIPE_MAX_SAMPLERS];
+   struct cell_command_texture texture;
+
+   struct vertex_info vertex_info;
+
+   /* XXX more state to come */
+
+
+   /** current color and Z tiles */
+   tile_t ctile ALIGN16_ATTRIB;
+   tile_t ztile ALIGN16_ATTRIB;
+
+   /** Current tiles' status */
+   ubyte cur_ctile_status, cur_ztile_status;
+
+   /** Status of all tiles in framebuffer */
+   ubyte ctile_status[MAX_HEIGHT/TILE_SIZE][MAX_WIDTH/TILE_SIZE] ALIGN16_ATTRIB;
+   ubyte ztile_status[MAX_HEIGHT/TILE_SIZE][MAX_WIDTH/TILE_SIZE] ALIGN16_ATTRIB;
+
+
+   /** for converting RGBA to PIPE_FORMAT_x colors */
+   vector unsigned char color_shuffle;
+
+   vector float tex_size;
+   vector unsigned int tex_size_mask; /**< == int(size - 1) */
+
+   vector float (*sample_texture)(vector float texcoord);
+
+} ALIGN16_ATTRIB;
+
+
+extern struct spu_global spu;
+extern boolean Debug;
+
+
+
+
+/* DMA TAGS */
+
+#define TAG_SURFACE_CLEAR     10
+#define TAG_VERTEX_BUFFER     11
+#define TAG_READ_TILE_COLOR   12
+#define TAG_READ_TILE_Z       13
+#define TAG_WRITE_TILE_COLOR  14
+#define TAG_WRITE_TILE_Z      15
+#define TAG_INDEX_BUFFER      16
+#define TAG_BATCH_BUFFER      17
+#define TAG_MISC              18
+#define TAG_TEXTURE_TILE      19
+#define TAG_INSTRUCTION_FETCH 20
+
+
+
+static INLINE void
+wait_on_mask(unsigned tagMask)
+{
+   mfc_write_tag_mask( tagMask );
+   /* wait for completion of _any_ DMAs specified by tagMask */
+   mfc_read_tag_status_any();
+}
+
+
+static INLINE void
+wait_on_mask_all(unsigned tagMask)
+{
+   mfc_write_tag_mask( tagMask );
+   /* wait for completion of _any_ DMAs specified by tagMask */
+   mfc_read_tag_status_all();
+}
+
+
+
+
+
+static INLINE void
+memset16(ushort *d, ushort value, uint count)
+{
+   uint i;
+   for (i = 0; i < count; i++)
+      d[i] = value;
+}
+
+
+static INLINE void
+memset32(uint *d, uint value, uint count)
+{
+   uint i;
+   for (i = 0; i < count; i++)
+      d[i] = value;
+}
+
+
+#endif /* SPU_MAIN_H */
diff --git a/src/gallium/drivers/cell/spu/spu_render.c b/src/gallium/drivers/cell/spu/spu_render.c
new file mode 100644
index 0000000000..932fb500b3
--- /dev/null
+++ b/src/gallium/drivers/cell/spu/spu_render.c
@@ -0,0 +1,301 @@
+/**************************************************************************
+ * 
+ * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+
+#include <stdio.h>
+#include <libmisc.h>
+#include <spu_mfcio.h>
+
+#include "spu_main.h"
+#include "spu_render.h"
+#include "spu_tri.h"
+#include "spu_tile.h"
+#include "pipe/cell/common.h"
+
+
+
+/**
+ * Given a rendering command's bounding box (in pixels) compute the
+ * location of the corresponding screen tile bounding box.
+ */
+static INLINE void
+tile_bounding_box(const struct cell_command_render *render,
+                  uint *txmin, uint *tymin,
+                  uint *box_num_tiles, uint *box_width_tiles)
+{
+#if 0
+   /* Debug: full-window bounding box */
+   uint txmax = spu.fb.width_tiles - 1;
+   uint tymax = spu.fb.height_tiles - 1;
+   *txmin = 0;
+   *tymin = 0;
+   *box_num_tiles = spu.fb.width_tiles * spu.fb.height_tiles;
+   *box_width_tiles = spu.fb.width_tiles;
+   (void) render;
+   (void) txmax;
+   (void) tymax;
+#else
+   uint txmax, tymax, box_height_tiles;
+
+   *txmin = (uint) render->xmin / TILE_SIZE;
+   *tymin = (uint) render->ymin / TILE_SIZE;
+   txmax = (uint) render->xmax / TILE_SIZE;
+   tymax = (uint) render->ymax / TILE_SIZE;
+   if (txmax >= spu.fb.width_tiles)
+      txmax = spu.fb.width_tiles-1;
+   if (tymax >= spu.fb.height_tiles)
+      tymax = spu.fb.height_tiles-1;
+   *box_width_tiles = txmax - *txmin + 1;
+   box_height_tiles = tymax - *tymin + 1;
+   *box_num_tiles = *box_width_tiles * box_height_tiles;
+#endif
+#if 0
+   printf("SPU %u: bounds: %g, %g  ...  %g, %g\n", spu.init.id,
+          render->xmin, render->ymin, render->xmax, render->ymax);
+   printf("SPU %u: tiles:  %u, %u .. %u, %u\n",
+           spu.init.id, *txmin, *tymin, txmax, tymax);
+   ASSERT(render->xmin <= render->xmax);
+   ASSERT(render->ymin <= render->ymax);
+#endif
+}
+
+
+/** Check if the tile at (tx,ty) belongs to this SPU */
+static INLINE boolean
+my_tile(uint tx, uint ty)
+{
+   return (spu.fb.width_tiles * ty + tx) % spu.init.num_spus == spu.init.id;
+}
+
+
+/**
+ * Start fetching non-clear color/Z tiles from main memory
+ */
+static INLINE void
+get_cz_tiles(uint tx, uint ty)
+{
+   if (spu.depth_stencil.depth.enabled) {
+      if (spu.cur_ztile_status != TILE_STATUS_CLEAR) {
+         //printf("SPU %u: getting Z tile %u, %u\n", spu.init.id, tx, ty);
+         get_tile(tx, ty, &spu.ztile, TAG_READ_TILE_Z, 1);
+         spu.cur_ztile_status = TILE_STATUS_GETTING;
+      }
+   }
+
+   if (spu.cur_ctile_status != TILE_STATUS_CLEAR) {
+      //printf("SPU %u: getting C tile %u, %u\n", spu.init.id, tx, ty);
+      get_tile(tx, ty, &spu.ctile, TAG_READ_TILE_COLOR, 0);
+      spu.cur_ctile_status = TILE_STATUS_GETTING;
+   }
+}
+
+
+/**
+ * Start putting dirty color/Z tiles back to main memory
+ */
+static INLINE void
+put_cz_tiles(uint tx, uint ty)
+{
+   if (spu.cur_ztile_status == TILE_STATUS_DIRTY) {
+      /* tile was modified and needs to be written back */
+      //printf("SPU %u: put dirty Z tile %u, %u\n", spu.init.id, tx, ty);
+      put_tile(tx, ty, &spu.ztile, TAG_WRITE_TILE_Z, 1);
+      spu.cur_ztile_status = TILE_STATUS_DEFINED;
+   }
+   else if (spu.cur_ztile_status == TILE_STATUS_GETTING) {
+      /* tile was never used */
+      spu.cur_ztile_status = TILE_STATUS_DEFINED;
+      //printf("SPU %u: put getting Z tile %u, %u\n", spu.init.id, tx, ty);
+   }
+
+   if (spu.cur_ctile_status == TILE_STATUS_DIRTY) {
+      /* tile was modified and needs to be written back */
+      //printf("SPU %u: put dirty C tile %u, %u\n", spu.init.id, tx, ty);
+      put_tile(tx, ty, &spu.ctile, TAG_WRITE_TILE_COLOR, 0);
+      spu.cur_ctile_status = TILE_STATUS_DEFINED;
+   }
+   else if (spu.cur_ctile_status == TILE_STATUS_GETTING) {
+      /* tile was never used */
+      spu.cur_ctile_status = TILE_STATUS_DEFINED;
+      //printf("SPU %u: put getting C tile %u, %u\n", spu.init.id, tx, ty);
+   }
+}
+
+
+/**
+ * Wait for 'put' of color/z tiles to complete.
+ */
+static INLINE void
+wait_put_cz_tiles(void)
+{
+   wait_on_mask(1 << TAG_WRITE_TILE_COLOR);
+   if (spu.depth_stencil.depth.enabled) {
+      wait_on_mask(1 << TAG_WRITE_TILE_Z);
+   }
+}
+
+
+/**
+ * Render primitives
+ * \param pos_incr  returns value indicating how may words to skip after
+ *                  this command in the batch buffer
+ */
+void
+cmd_render(const struct cell_command_render *render, uint *pos_incr)
+{
+   /* we'll DMA into these buffers */
+   ubyte vertex_data[CELL_BUFFER_SIZE] ALIGN16_ATTRIB;
+   const uint vertex_size = render->vertex_size; /* in bytes */
+   /*const*/ uint total_vertex_bytes = render->num_verts * vertex_size;
+   uint index_bytes;
+   const ubyte *vertices;
+   const ushort *indexes;
+   uint i, j;
+
+
+   if (Debug) {
+      printf("SPU %u: RENDER prim %u, num_vert=%u  num_ind=%u  "
+             "inline_vert=%u\n",
+             spu.init.id,
+             render->prim_type,
+             render->num_verts,
+             render->num_indexes,
+             render->inline_verts);
+
+      /*
+      printf("       bound: %g, %g .. %g, %g\n",
+             render->xmin, render->ymin, render->xmax, render->ymax);
+      */
+   }
+
+   ASSERT(sizeof(*render) % 4 == 0);
+   ASSERT(total_vertex_bytes % 16 == 0);
+   ASSERT(render->prim_type == PIPE_PRIM_TRIANGLES);
+   ASSERT(render->num_indexes % 3 == 0);
+
+
+   /* indexes are right after the render command in the batch buffer */
+   indexes = (const ushort *) (render + 1);
+   index_bytes = ROUNDUP8(render->num_indexes * 2);
+   *pos_incr = index_bytes / 8 + sizeof(*render) / 8;
+
+
+   if (render->inline_verts) {
+      /* Vertices are after indexes in batch buffer at next 16-byte addr */
+      vertices = (const ubyte *) render + (*pos_incr * 8);
+      vertices = (const ubyte *) align_pointer((void *) vertices, 16);
+      ASSERT_ALIGN16(vertices);
+      *pos_incr = ((vertices + total_vertex_bytes) - (ubyte *) render) / 8;
+   }
+   else {
+      /* Begin DMA fetch of vertex buffer */
+      ubyte *src = spu.init.buffers[render->vertex_buf];
+      ubyte *dest = vertex_data;
+
+      /* skip vertex data we won't use */
+#if 01
+      src += render->min_index * vertex_size;
+      dest += render->min_index * vertex_size;
+      total_vertex_bytes -= render->min_index * vertex_size;
+#endif
+      ASSERT(total_vertex_bytes % 16 == 0);
+      ASSERT_ALIGN16(dest);
+      ASSERT_ALIGN16(src);
+
+      mfc_get(dest,   /* in vertex_data[] array */
+              (unsigned int) src,  /* src in main memory */
+              total_vertex_bytes,  /* size */
+              TAG_VERTEX_BUFFER,
+              0, /* tid */
+              0  /* rid */);
+
+      vertices = vertex_data;
+
+      wait_on_mask(1 << TAG_VERTEX_BUFFER);
+   }
+
+
+   /**
+    ** find tiles which intersect the prim bounding box
+    **/
+   uint txmin, tymin, box_width_tiles, box_num_tiles;
+   tile_bounding_box(render, &txmin, &tymin,
+                     &box_num_tiles, &box_width_tiles);
+
+
+   /* make sure any pending clears have completed */
+   wait_on_mask(1 << TAG_SURFACE_CLEAR); /* XXX temporary */
+
+
+   /**
+    ** loop over tiles, rendering tris
+    **/
+   for (i = 0; i < box_num_tiles; i++) {
+      const uint tx = txmin + i % box_width_tiles;
+      const uint ty = tymin + i / box_width_tiles;
+
+      ASSERT(tx < spu.fb.width_tiles);
+      ASSERT(ty < spu.fb.height_tiles);
+
+      if (!my_tile(tx, ty))
+         continue;
+
+      spu.cur_ctile_status = spu.ctile_status[ty][tx];
+      spu.cur_ztile_status = spu.ztile_status[ty][tx];
+
+      get_cz_tiles(tx, ty);
+
+      uint drawn = 0;
+
+      /* loop over tris */
+      for (j = 0; j < render->num_indexes; j += 3) {
+         const float *v0, *v1, *v2;
+
+         v0 = (const float *) (vertices + indexes[j+0] * vertex_size);
+         v1 = (const float *) (vertices + indexes[j+1] * vertex_size);
+         v2 = (const float *) (vertices + indexes[j+2] * vertex_size);
+
+         drawn += tri_draw(v0, v1, v2, tx, ty);
+      }
+
+      //printf("SPU %u: drew %u of %u\n", spu.init.id, drawn, render->num_indexes/3);
+
+      /* write color/z tiles back to main framebuffer, if dirtied */
+      put_cz_tiles(tx, ty);
+
+      wait_put_cz_tiles(); /* XXX seems unnecessary... */
+
+      spu.ctile_status[ty][tx] = spu.cur_ctile_status;
+      spu.ztile_status[ty][tx] = spu.cur_ztile_status;
+   }
+
+   if (Debug)
+      printf("SPU %u: RENDER done\n",
+             spu.init.id);
+}
+
+
diff --git a/src/gallium/drivers/cell/spu/spu_render.h b/src/gallium/drivers/cell/spu/spu_render.h
new file mode 100644
index 0000000000..fbcdc5ec31
--- /dev/null
+++ b/src/gallium/drivers/cell/spu/spu_render.h
@@ -0,0 +1,38 @@
+/**************************************************************************
+ * 
+ * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+
+#ifndef SPU_RENDER_H
+#define SPU_RENDER_H
+
+#include "pipe/cell/common.h"
+
+extern void
+cmd_render(const struct cell_command_render *render, uint *pos_incr);
+
+#endif /* SPU_RENDER_H */
+
diff --git a/src/gallium/drivers/cell/spu/spu_texture.c b/src/gallium/drivers/cell/spu/spu_texture.c
new file mode 100644
index 0000000000..3962aaa4a9
--- /dev/null
+++ b/src/gallium/drivers/cell/spu/spu_texture.c
@@ -0,0 +1,217 @@
+/**************************************************************************
+ * 
+ * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+
+#include "pipe/p_compiler.h"
+#include "spu_main.h"
+#include "spu_texture.h"
+#include "spu_tile.h"
+#include "spu_colorpack.h"
+
+
+/**
+ * Number of texture tiles to cache.
+ * Note that this will probably be the largest consumer of SPU local store/
+ * memory for this driver!
+ */
+#define CACHE_SIZE 16
+
+static tile_t tex_tiles[CACHE_SIZE]  ALIGN16_ATTRIB;
+
+static vector unsigned int tex_tile_xy[CACHE_SIZE];
+
+
+
+/**
+ * Mark all tex cache entries as invalid.
+ */
+void
+invalidate_tex_cache(void)
+{
+   /* XXX memset? */
+   uint i;
+   for (i = 0; i < CACHE_SIZE; i++) {
+      tex_tile_xy[i] = ((vector unsigned int) { ~0U, ~0U, ~0U, ~0U });
+   }
+}
+
+
+/**
+ * Return the cache pos/index which corresponds to tile (tx,ty)
+ */
+static INLINE uint
+cache_pos(vector unsigned int txty)
+{
+   uint pos = (spu_extract(txty,0) + spu_extract(txty,1) * 4) % CACHE_SIZE;
+   return pos;
+}
+
+
+/**
+ * Make sure the tile for texel (i,j) is present, return its position/index
+ * in the cache.
+ */
+static uint
+get_tex_tile(vector unsigned int ij)
+{
+   /* tile address: tx,ty */
+   const vector unsigned int txty = spu_rlmask(ij, -5);  /* divide by 32 */
+   const uint pos = cache_pos(txty);
+
+   if ((spu_extract(tex_tile_xy[pos], 0) != spu_extract(txty, 0)) ||
+       (spu_extract(tex_tile_xy[pos], 1) != spu_extract(txty, 1))) {
+
+      /* texture cache miss, fetch tile from main memory */
+      const uint tiles_per_row = spu.texture.width / TILE_SIZE;
+      const uint bytes_per_tile = sizeof(tile_t);
+      const void *src = (const ubyte *) spu.texture.start
+         + (spu_extract(txty,1) * tiles_per_row + spu_extract(txty,0)) * bytes_per_tile;
+
+      printf("SPU %u: tex cache miss at %d, %d  pos=%u  old=%d,%d\n",
+             spu.init.id,
+             spu_extract(txty,0),
+             spu_extract(txty,1),
+             pos,
+             spu_extract(tex_tile_xy[pos],0),
+             spu_extract(tex_tile_xy[pos],1));
+
+      ASSERT_ALIGN16(tex_tiles[pos].ui);
+      ASSERT_ALIGN16(src);
+
+      mfc_get(tex_tiles[pos].ui,  /* dest */
+              (unsigned int) src,
+              bytes_per_tile,      /* size */
+              TAG_TEXTURE_TILE,
+              0, /* tid */
+              0  /* rid */);
+
+      wait_on_mask(1 << TAG_TEXTURE_TILE);
+
+      tex_tile_xy[pos] = txty;
+   }
+   else {
+#if 0
+      printf("SPU %u: tex cache HIT at %d, %d\n",
+             spu.init.id, tx, ty);
+#endif
+   }
+
+   return pos;
+}
+
+
+/**
+ * Get texture sample at texcoord.
+ * XXX this is extremely primitive for now.
+ */
+vector float
+sample_texture_nearest(vector float texcoord)
+{
+   vector float tc = spu_mul(texcoord, spu.tex_size);
+   vector unsigned int itc = spu_convtu(tc, 0);  /* convert to int */
+   itc = spu_and(itc, spu.tex_size_mask);        /* mask (GL_REPEAT) */
+   vector unsigned int ij = spu_and(itc, TILE_SIZE-1); /* intra tile addr */
+   uint pos = get_tex_tile(itc);
+   uint texel = tex_tiles[pos].ui[spu_extract(ij, 1)][spu_extract(ij, 0)];
+   return spu_unpack_A8R8G8B8(texel);
+}
+
+
+vector float
+sample_texture_bilinear(vector float texcoord)
+{
+   static const vector unsigned int offset10 = {1, 0, 0, 0};
+   static const vector unsigned int offset01 = {0, 1, 0, 0};
+
+   vector float tc = spu_mul(texcoord, spu.tex_size);
+   tc = spu_add(tc, spu_splats(-0.5f));  /* half texel bias */
+
+   /* integer texcoords S,T: */
+   vector unsigned int itc00 = spu_convtu(tc, 0);  /* convert to int */
+   vector unsigned int itc01 = spu_add(itc00, offset01);
+   vector unsigned int itc10 = spu_add(itc00, offset10);
+   vector unsigned int itc11 = spu_add(itc10, offset01);
+
+   /* mask (GL_REPEAT) */
+   itc00 = spu_and(itc00, spu.tex_size_mask);
+   itc01 = spu_and(itc01, spu.tex_size_mask);
+   itc10 = spu_and(itc10, spu.tex_size_mask);
+   itc11 = spu_and(itc11, spu.tex_size_mask);
+
+   /* intra tile addr */
+   vector unsigned int ij00 = spu_and(itc00, TILE_SIZE-1);
+   vector unsigned int ij01 = spu_and(itc01, TILE_SIZE-1);
+   vector unsigned int ij10 = spu_and(itc10, TILE_SIZE-1);
+   vector unsigned int ij11 = spu_and(itc11, TILE_SIZE-1);
+
+   /* get tile cache positions */
+   uint pos00 = get_tex_tile(itc00);
+   uint pos01, pos10, pos11;
+   if ((spu_extract(ij00, 0) < TILE_SIZE-1) &&
+       (spu_extract(ij00, 1) < TILE_SIZE-1)) {
+      /* all texels are in the same tile */
+      pos01 = pos10 = pos11 = pos00;
+   }
+   else {
+      pos01 = get_tex_tile(itc01);
+      pos10 = get_tex_tile(itc10);
+      pos11 = get_tex_tile(itc11);
+   }
+
+   /* get texels from tiles and convert to float[4] */
+   vector float texel00 = spu_unpack_A8R8G8B8(tex_tiles[pos00].ui[spu_extract(ij00, 1)][spu_extract(ij00, 0)]);
+   vector float texel01 = spu_unpack_A8R8G8B8(tex_tiles[pos01].ui[spu_extract(ij01, 1)][spu_extract(ij01, 0)]);
+   vector float texel10 = spu_unpack_A8R8G8B8(tex_tiles[pos10].ui[spu_extract(ij10, 1)][spu_extract(ij10, 0)]);
+   vector float texel11 = spu_unpack_A8R8G8B8(tex_tiles[pos11].ui[spu_extract(ij11, 1)][spu_extract(ij11, 0)]);
+
+   /* Compute weighting factors in [0,1]
+    * Multiply texcoord by 1024, AND with 1023, convert back to float.
+    */
+   vector float tc1024 = spu_mul(tc, spu_splats(1024.0f));
+   vector signed int itc1024 = spu_convts(tc1024, 0);
+   itc1024 = spu_and(itc1024, spu_splats((1 << 10) - 1));
+   vector float weight = spu_convtf(itc1024, 10);
+
+   /* smeared frac and 1-frac */
+   vector float sfrac = spu_splats(spu_extract(weight, 0));
+   vector float tfrac = spu_splats(spu_extract(weight, 1));
+   vector float sfrac1 = spu_sub(spu_splats(1.0f), sfrac);
+   vector float tfrac1 = spu_sub(spu_splats(1.0f), tfrac);
+
+   /* multiply the samples (colors) by the S/T weights */
+   texel00 = spu_mul(spu_mul(texel00, sfrac1), tfrac1);
+   texel10 = spu_mul(spu_mul(texel10, sfrac ), tfrac1);
+   texel01 = spu_mul(spu_mul(texel01, sfrac1), tfrac );
+   texel11 = spu_mul(spu_mul(texel11, sfrac ), tfrac );
+
+   /* compute sum of weighted samples */
+   vector float texel_sum = spu_add(texel00, texel01);
+   texel_sum = spu_add(texel_sum, texel10);
+   texel_sum = spu_add(texel_sum, texel11);
+
+   return texel_sum;
+}
diff --git a/src/gallium/drivers/cell/spu/spu_texture.h b/src/gallium/drivers/cell/spu/spu_texture.h
new file mode 100644
index 0000000000..95eb87080f
--- /dev/null
+++ b/src/gallium/drivers/cell/spu/spu_texture.h
@@ -0,0 +1,47 @@
+/**************************************************************************
+ * 
+ * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+#ifndef SPU_TEXTURE_H
+#define SPU_TEXTURE_H
+
+
+#include "pipe/p_compiler.h"
+
+
+extern void
+invalidate_tex_cache(void);
+
+
+extern vector float
+sample_texture_nearest(vector float texcoord);
+
+
+extern vector float
+sample_texture_bilinear(vector float texcoord);
+
+
+#endif /* SPU_TEXTURE_H */
diff --git a/src/gallium/drivers/cell/spu/spu_tile.c b/src/gallium/drivers/cell/spu/spu_tile.c
new file mode 100644
index 0000000000..12dc246328
--- /dev/null
+++ b/src/gallium/drivers/cell/spu/spu_tile.c
@@ -0,0 +1,83 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+
+
+#include "spu_tile.h"
+#include "spu_main.h"
+
+
+void
+get_tile(uint tx, uint ty, tile_t *tile, int tag, int zBuf)
+{
+   const uint offset = ty * spu.fb.width_tiles + tx;
+   const uint bytesPerTile = TILE_SIZE * TILE_SIZE * (zBuf ? spu.fb.zsize : 4);
+   const ubyte *src = zBuf ? spu.fb.depth_start : spu.fb.color_start;
+
+   src += offset * bytesPerTile;
+
+   ASSERT(tx < spu.fb.width_tiles);
+   ASSERT(ty < spu.fb.height_tiles);
+   ASSERT_ALIGN16(tile);
+   /*
+   printf("get_tile:  dest: %p  src: 0x%x  size: %d\n",
+          tile, (unsigned int) src, bytesPerTile);
+   */
+   mfc_get(tile->ui,  /* dest in local memory */
+           (unsigned int) src, /* src in main memory */
+           bytesPerTile,
+           tag,
+           0, /* tid */
+           0  /* rid */);
+}
+
+
+void
+put_tile(uint tx, uint ty, const tile_t *tile, int tag, int zBuf)
+{
+   const uint offset = ty * spu.fb.width_tiles + tx;
+   const uint bytesPerTile = TILE_SIZE * TILE_SIZE * (zBuf ? spu.fb.zsize : 4);
+   ubyte *dst = zBuf ? spu.fb.depth_start : spu.fb.color_start;
+
+   dst += offset * bytesPerTile;
+
+   ASSERT(tx < spu.fb.width_tiles);
+   ASSERT(ty < spu.fb.height_tiles);
+   ASSERT_ALIGN16(tile);
+   /*
+   printf("SPU %u: put_tile:  src: %p  dst: 0x%x  size: %d\n",
+          spu.init.id,
+          tile, (unsigned int) dst, bytesPerTile);
+   */
+   mfc_put((void *) tile->ui,  /* src in local memory */
+           (unsigned int) dst,  /* dst in main memory */
+           bytesPerTile,
+           tag,
+           0, /* tid */
+           0  /* rid */);
+}
+
diff --git a/src/gallium/drivers/cell/spu/spu_tile.h b/src/gallium/drivers/cell/spu/spu_tile.h
new file mode 100644
index 0000000000..e53340a55a
--- /dev/null
+++ b/src/gallium/drivers/cell/spu/spu_tile.h
@@ -0,0 +1,73 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+#ifndef SPU_TILE_H
+#define SPU_TILE_H
+
+
+#include <libmisc.h>
+#include <spu_mfcio.h>
+#include "spu_main.h"
+#include "pipe/cell/common.h"
+
+
+
+void
+get_tile(uint tx, uint ty, tile_t *tile, int tag, int zBuf);
+
+void
+put_tile(uint tx, uint ty, const tile_t *tile, int tag, int zBuf);
+
+
+
+static INLINE void
+clear_c_tile(tile_t *ctile)
+{
+   memset32((uint*) ctile->ui,
+            spu.fb.color_clear_value,
+            TILE_SIZE * TILE_SIZE);
+}
+
+
+static INLINE void
+clear_z_tile(tile_t *ztile)
+{
+   if (spu.fb.depth_format == PIPE_FORMAT_Z16_UNORM) {
+      memset16((ushort*) ztile->us,
+               spu.fb.depth_clear_value,
+               TILE_SIZE * TILE_SIZE);
+   }
+   else {
+      ASSERT(spu.fb.depth_format == PIPE_FORMAT_Z32_UNORM);
+      memset32((uint*) ztile->ui,
+               spu.fb.depth_clear_value,
+               TILE_SIZE * TILE_SIZE);
+   }
+}
+
+
+#endif /* SPU_TILE_H */
diff --git a/src/gallium/drivers/cell/spu/spu_tri.c b/src/gallium/drivers/cell/spu/spu_tri.c
new file mode 100644
index 0000000000..be9624cf7d
--- /dev/null
+++ b/src/gallium/drivers/cell/spu/spu_tri.c
@@ -0,0 +1,926 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+/**
+ * Triangle rendering within a tile.
+ */
+
+#include "pipe/p_compiler.h"
+#include "pipe/p_format.h"
+#include "pipe/p_util.h"
+#include "spu_blend.h"
+#include "spu_colorpack.h"
+#include "spu_main.h"
+#include "spu_texture.h"
+#include "spu_tile.h"
+#include "spu_tri.h"
+
+#include "spu_ztest.h"
+
+
+/** Masks are uint[4] vectors with each element being 0 or 0xffffffff */
+typedef vector unsigned int mask_t;
+
+typedef union
+{
+   vector float v;
+   float f[4];
+} float4;
+
+
+/**
+ * Simplified types taken from other parts of Gallium
+ */
+struct vertex_header {
+   vector float data[1];
+};
+
+
+
+/* XXX fix this */
+#undef CEILF
+#define CEILF(X) ((float) (int) ((X) + 0.99999))
+
+
+#define QUAD_TOP_LEFT     0
+#define QUAD_TOP_RIGHT    1
+#define QUAD_BOTTOM_LEFT  2
+#define QUAD_BOTTOM_RIGHT 3
+#define MASK_TOP_LEFT     (1 << QUAD_TOP_LEFT)
+#define MASK_TOP_RIGHT    (1 << QUAD_TOP_RIGHT)
+#define MASK_BOTTOM_LEFT  (1 << QUAD_BOTTOM_LEFT)
+#define MASK_BOTTOM_RIGHT (1 << QUAD_BOTTOM_RIGHT)
+#define MASK_ALL          0xf
+
+
+#define DEBUG_VERTS 0
+
+/**
+ * Triangle edge info
+ */
+struct edge {
+   float dx;		/**< X(v1) - X(v0), used only during setup */
+   float dy;		/**< Y(v1) - Y(v0), used only during setup */
+   float dxdy;		/**< dx/dy */
+   float sx, sy;	/**< first sample point coord */
+   int lines;		/**< number of lines on this edge */
+};
+
+
+struct interp_coef
+{
+   float4 a0;
+   float4 dadx;
+   float4 dady;
+};
+
+
+/**
+ * Triangle setup info (derived from draw_stage).
+ * Also used for line drawing (taking some liberties).
+ */
+struct setup_stage {
+
+   /* Vertices are just an array of floats making up each attribute in
+    * turn.  Currently fixed at 4 floats, but should change in time.
+    * Codegen will help cope with this.
+    */
+   const struct vertex_header *vmax;
+   const struct vertex_header *vmid;
+   const struct vertex_header *vmin;
+   const struct vertex_header *vprovoke;
+
+   struct edge ebot;
+   struct edge etop;
+   struct edge emaj;
+
+   float oneoverarea;
+
+   uint tx, ty;
+
+   int cliprect_minx, cliprect_maxx, cliprect_miny, cliprect_maxy;
+
+#if 0
+   struct tgsi_interp_coef coef[PIPE_MAX_SHADER_INPUTS];
+#else
+   struct interp_coef coef[PIPE_MAX_SHADER_INPUTS];
+#endif
+
+#if 0
+   struct quad_header quad; 
+#endif
+
+   struct {
+      int left[2];   /**< [0] = row0, [1] = row1 */
+      int right[2];
+      int y;
+      unsigned y_flags;
+      unsigned mask;     /**< mask of MASK_BOTTOM/TOP_LEFT/RIGHT bits */
+   } span;
+};
+
+
+
+static struct setup_stage setup;
+
+
+
+
+#if 0
+/**
+ * Basically a cast wrapper.
+ */
+static INLINE struct setup_stage *setup_stage( struct draw_stage *stage )
+{
+   return (struct setup_stage *)stage;
+}
+#endif
+
+#if 0
+/**
+ * Clip setup.quad against the scissor/surface bounds.
+ */
+static INLINE void
+quad_clip(struct setup_stage *setup)
+{
+   const struct pipe_scissor_state *cliprect = &setup.softpipe->cliprect;
+   const int minx = (int) cliprect->minx;
+   const int maxx = (int) cliprect->maxx;
+   const int miny = (int) cliprect->miny;
+   const int maxy = (int) cliprect->maxy;
+
+   if (setup.quad.x0 >= maxx ||
+       setup.quad.y0 >= maxy ||
+       setup.quad.x0 + 1 < minx ||
+       setup.quad.y0 + 1 < miny) {
+      /* totally clipped */
+      setup.quad.mask = 0x0;
+      return;
+   }
+   if (setup.quad.x0 < minx)
+      setup.quad.mask &= (MASK_BOTTOM_RIGHT | MASK_TOP_RIGHT);
+   if (setup.quad.y0 < miny)
+      setup.quad.mask &= (MASK_BOTTOM_LEFT | MASK_BOTTOM_RIGHT);
+   if (setup.quad.x0 == maxx - 1)
+      setup.quad.mask &= (MASK_BOTTOM_LEFT | MASK_TOP_LEFT);
+   if (setup.quad.y0 == maxy - 1)
+      setup.quad.mask &= (MASK_TOP_LEFT | MASK_TOP_RIGHT);
+}
+#endif
+
+#if 0
+/**
+ * Emit a quad (pass to next stage) with clipping.
+ */
+static INLINE void
+clip_emit_quad(struct setup_stage *setup)
+{
+   quad_clip(setup);
+   if (setup.quad.mask) {
+      struct softpipe_context *sp = setup.softpipe;
+      sp->quad.first->run(sp->quad.first, &setup.quad);
+   }
+}
+#endif
+
+/**
+ * Evaluate attribute coefficients (plane equations) to compute
+ * attribute values for the four fragments in a quad.
+ * Eg: four colors will be compute.
+ */
+static INLINE void
+eval_coeff(uint slot, float x, float y, vector float result[4])
+{
+   switch (spu.vertex_info.interp_mode[slot]) {
+   case INTERP_CONSTANT:
+      result[QUAD_TOP_LEFT] =
+      result[QUAD_TOP_RIGHT] =
+      result[QUAD_BOTTOM_LEFT] =
+      result[QUAD_BOTTOM_RIGHT] = setup.coef[slot].a0.v;
+      break;
+
+   case INTERP_LINEAR:
+      /* fall-through, for now */
+   default:
+      {
+         register vector float dadx = setup.coef[slot].dadx.v;
+         register vector float dady = setup.coef[slot].dady.v;
+         register vector float topLeft
+            = spu_add(setup.coef[slot].a0.v,
+                      spu_add(spu_mul(spu_splats(x), dadx),
+                              spu_mul(spu_splats(y), dady)));
+
+         result[QUAD_TOP_LEFT] = topLeft;
+         result[QUAD_TOP_RIGHT] = spu_add(topLeft, dadx);
+         result[QUAD_BOTTOM_LEFT] = spu_add(topLeft, dady);
+         result[QUAD_BOTTOM_RIGHT] = spu_add(spu_add(topLeft, dadx), dady);
+      }
+   }
+}
+
+
+static INLINE vector float
+eval_z(float x, float y)
+{
+   const uint slot = 0;
+   const float dzdx = setup.coef[slot].dadx.f[2];
+   const float dzdy = setup.coef[slot].dady.f[2];
+   const float topLeft = setup.coef[slot].a0.f[2] + x * dzdx + y * dzdy;
+   const vector float topLeftv = spu_splats(topLeft);
+   const vector float derivs = (vector float) { 0.0, dzdx, dzdy, dzdx + dzdy };
+   return spu_add(topLeftv, derivs);
+}
+
+
+static INLINE mask_t
+do_depth_test(int x, int y, mask_t quadmask)
+{
+   float4 zvals;
+   mask_t mask;
+
+   zvals.v = eval_z((float) x, (float) y);
+
+   if (spu.fb.depth_format == PIPE_FORMAT_Z16_UNORM) {
+      int ix = (x - setup.cliprect_minx) / 4;
+      int iy = (y - setup.cliprect_miny) / 2;
+      mask = spu_z16_test_less(zvals.v, &spu.ztile.us8[iy][ix], x>>1, quadmask);
+   }
+   else {
+      int ix = (x - setup.cliprect_minx) / 2;
+      int iy = (y - setup.cliprect_miny) / 2;
+      mask = spu_z32_test_less(zvals.v, &spu.ztile.ui4[iy][ix], quadmask);
+   }
+
+   if (spu_extract(spu_orx(mask), 0))
+      spu.cur_ztile_status = TILE_STATUS_DIRTY;
+
+   return mask;
+}
+
+
+/**
+ * Emit a quad (pass to next stage).  No clipping is done.
+ * Note: about 1/5 to 1/7 of the time, mask is zero and this function
+ * should be skipped.  But adding the test for that slows things down
+ * overall.
+ */
+static INLINE void
+emit_quad( int x, int y, mask_t mask )
+{
+#if 0
+   struct softpipe_context *sp = setup.softpipe;
+   setup.quad.x0 = x;
+   setup.quad.y0 = y;
+   setup.quad.mask = mask;
+   sp->quad.first->run(sp->quad.first, &setup.quad);
+#else
+
+   if (spu.depth_stencil.depth.enabled) {
+      mask = do_depth_test(x, y, mask);
+   }
+
+   /* If any bits in mask are set... */
+   if (spu_extract(spu_orx(mask), 0)) {
+      const int ix = x - setup.cliprect_minx;
+      const int iy = y - setup.cliprect_miny;
+      const vector unsigned char shuffle = spu.color_shuffle;
+      vector float colors[4];
+
+      spu.cur_ctile_status = TILE_STATUS_DIRTY;
+
+      if (spu.texture.start) {
+         /* texture mapping */
+         vector float texcoords[4];
+         eval_coeff(2, (float) x, (float) y, texcoords);
+
+         if (spu_extract(mask, 0))
+            colors[0] = spu.sample_texture(texcoords[0]);
+         if (spu_extract(mask, 1))
+            colors[1] = spu.sample_texture(texcoords[1]);
+         if (spu_extract(mask, 2))
+            colors[2] = spu.sample_texture(texcoords[2]);
+         if (spu_extract(mask, 3))
+            colors[3] = spu.sample_texture(texcoords[3]);
+      }
+      else {
+         /* simple shading */
+         eval_coeff(1, (float) x, (float) y, colors);
+      }
+
+#if 1
+      if (spu.blend.blend_enable)
+         blend_quad(ix % TILE_SIZE, iy % TILE_SIZE, colors);
+#endif
+
+      if (spu_extract(mask, 0))
+         spu.ctile.ui[iy][ix] = spu_pack_color_shuffle(colors[0], shuffle);
+      if (spu_extract(mask, 1))
+         spu.ctile.ui[iy][ix+1] = spu_pack_color_shuffle(colors[1], shuffle);
+      if (spu_extract(mask, 2))
+         spu.ctile.ui[iy+1][ix] = spu_pack_color_shuffle(colors[2], shuffle);
+      if (spu_extract(mask, 3))
+         spu.ctile.ui[iy+1][ix+1] = spu_pack_color_shuffle(colors[3], shuffle);
+
+#if 0
+      /* SIMD_Z with swizzled color buffer (someday) */
+      vector unsigned int uicolors = *((vector unsigned int *) &colors);
+      spu.ctile.ui4[iy/2][ix/2] = spu_sel(spu.ctile.ui4[iy/2][ix/2], uicolors, mask);
+#endif
+   }
+
+#endif
+}
+
+
+/**
+ * Given an X or Y coordinate, return the block/quad coordinate that it
+ * belongs to.
+ */
+static INLINE int block( int x )
+{
+   return x & ~1;
+}
+
+
+/**
+ * Compute mask which indicates which pixels in the 2x2 quad are actually inside
+ * the triangle's bounds.
+ * The mask is a uint4 vector and each element will be 0 or 0xffffffff.
+ */
+static INLINE mask_t calculate_mask( int x )
+{
+   /* This is a little tricky.
+    * Use & instead of && to avoid branches.
+    * Use negation to convert true/false to ~0/0 values.
+    */
+   mask_t mask;
+   mask = spu_insert(-((x   >= setup.span.left[0]) & (x   < setup.span.right[0])), mask, 0);
+   mask = spu_insert(-((x+1 >= setup.span.left[0]) & (x+1 < setup.span.right[0])), mask, 1);
+   mask = spu_insert(-((x   >= setup.span.left[1]) & (x   < setup.span.right[1])), mask, 2);
+   mask = spu_insert(-((x+1 >= setup.span.left[1]) & (x+1 < setup.span.right[1])), mask, 3);
+   return mask;
+}
+
+
+/**
+ * Render a horizontal span of quads
+ */
+static void flush_spans( void )
+{
+   int minleft, maxright;
+   int x;
+
+   switch (setup.span.y_flags) {
+   case 0x3:
+      /* both odd and even lines written (both quad rows) */
+      minleft = MIN2(setup.span.left[0], setup.span.left[1]);
+      maxright = MAX2(setup.span.right[0], setup.span.right[1]);
+      break;
+
+   case 0x1:
+      /* only even line written (quad top row) */
+      minleft = setup.span.left[0];
+      maxright = setup.span.right[0];
+      break;
+
+   case 0x2:
+      /* only odd line written (quad bottom row) */
+      minleft = setup.span.left[1];
+      maxright = setup.span.right[1];
+      break;
+
+   default:
+      return;
+   }
+
+
+   /* OK, we're very likely to need the tile data now.
+    * clear or finish waiting if needed.
+    */
+   if (spu.cur_ctile_status == TILE_STATUS_GETTING) {
+      /* wait for mfc_get() to complete */
+      //printf("SPU: %u: waiting for ctile\n", spu.init.id);
+      wait_on_mask(1 << TAG_READ_TILE_COLOR);
+      spu.cur_ctile_status = TILE_STATUS_CLEAN;
+   }
+   else if (spu.cur_ctile_status == TILE_STATUS_CLEAR) {
+      //printf("SPU %u: clearing C tile %u, %u\n", spu.init.id, setup.tx, setup.ty);
+      clear_c_tile(&spu.ctile);
+      spu.cur_ctile_status = TILE_STATUS_DIRTY;
+   }
+   ASSERT(spu.cur_ctile_status != TILE_STATUS_DEFINED);
+
+   if (spu.depth_stencil.depth.enabled) {
+      if (spu.cur_ztile_status == TILE_STATUS_GETTING) {
+         /* wait for mfc_get() to complete */
+         //printf("SPU: %u: waiting for ztile\n", spu.init.id);
+         wait_on_mask(1 << TAG_READ_TILE_Z);
+         spu.cur_ztile_status = TILE_STATUS_CLEAN;
+      }
+      else if (spu.cur_ztile_status == TILE_STATUS_CLEAR) {
+         //printf("SPU %u: clearing Z tile %u, %u\n", spu.init.id, setup.tx, setup.ty);
+         clear_z_tile(&spu.ztile);
+         spu.cur_ztile_status = TILE_STATUS_DIRTY;
+      }
+      ASSERT(spu.cur_ztile_status != TILE_STATUS_DEFINED);
+   }
+
+   /* XXX this loop could be moved into the above switch cases and
+    * calculate_mask() could be simplified a bit...
+    */
+   for (x = block(minleft); x <= block(maxright); x += 2) {
+#if 1
+      emit_quad( x, setup.span.y, calculate_mask( x ) );
+#endif
+   }
+
+   setup.span.y = 0;
+   setup.span.y_flags = 0;
+   setup.span.right[0] = 0;
+   setup.span.right[1] = 0;
+}
+
+#if DEBUG_VERTS
+static void print_vertex(const struct vertex_header *v)
+{
+   int i;
+   fprintf(stderr, "Vertex: (%p)\n", v);
+   for (i = 0; i < setup.quad.nr_attrs; i++) {
+      fprintf(stderr, "  %d: %f %f %f %f\n",  i, 
+              v->data[i][0], v->data[i][1], v->data[i][2], v->data[i][3]);
+   }
+}
+#endif
+
+
+static boolean setup_sort_vertices(const struct vertex_header *v0,
+                                   const struct vertex_header *v1,
+                                   const struct vertex_header *v2)
+{
+
+#if DEBUG_VERTS
+   fprintf(stderr, "Triangle:\n");
+   print_vertex(v0);
+   print_vertex(v1);
+   print_vertex(v2);
+#endif
+
+   setup.vprovoke = v2;
+
+   /* determine bottom to top order of vertices */
+   {
+      float y0 = spu_extract(v0->data[0], 1);
+      float y1 = spu_extract(v1->data[0], 1);
+      float y2 = spu_extract(v2->data[0], 1);
+      if (y0 <= y1) {
+	 if (y1 <= y2) {
+	    /* y0<=y1<=y2 */
+	    setup.vmin = v0;   
+	    setup.vmid = v1;   
+	    setup.vmax = v2;
+	 }
+	 else if (y2 <= y0) {
+	    /* y2<=y0<=y1 */
+	    setup.vmin = v2;   
+	    setup.vmid = v0;   
+	    setup.vmax = v1;   
+	 }
+	 else {
+	    /* y0<=y2<=y1 */
+	    setup.vmin = v0;   
+	    setup.vmid = v2;   
+	    setup.vmax = v1;  
+	 }
+      }
+      else {
+	 if (y0 <= y2) {
+	    /* y1<=y0<=y2 */
+	    setup.vmin = v1;   
+	    setup.vmid = v0;   
+	    setup.vmax = v2;  
+	 }
+	 else if (y2 <= y1) {
+	    /* y2<=y1<=y0 */
+	    setup.vmin = v2;   
+	    setup.vmid = v1;   
+	    setup.vmax = v0;  
+	 }
+	 else {
+	    /* y1<=y2<=y0 */
+	    setup.vmin = v1;   
+	    setup.vmid = v2;   
+	    setup.vmax = v0;
+	 }
+      }
+   }
+
+   /* Check if triangle is completely outside the tile bounds */
+   if (spu_extract(setup.vmin->data[0], 1) > setup.cliprect_maxy)
+      return FALSE;
+   if (spu_extract(setup.vmax->data[0], 1) < setup.cliprect_miny)
+      return FALSE;
+   if (spu_extract(setup.vmin->data[0], 0) < setup.cliprect_minx &&
+       spu_extract(setup.vmid->data[0], 0) < setup.cliprect_minx &&
+       spu_extract(setup.vmax->data[0], 0) < setup.cliprect_minx)
+      return FALSE;
+   if (spu_extract(setup.vmin->data[0], 0) > setup.cliprect_maxx &&
+       spu_extract(setup.vmid->data[0], 0) > setup.cliprect_maxx &&
+       spu_extract(setup.vmax->data[0], 0) > setup.cliprect_maxx)
+      return FALSE;
+
+   setup.ebot.dx = spu_extract(setup.vmid->data[0], 0) - spu_extract(setup.vmin->data[0], 0);
+   setup.ebot.dy = spu_extract(setup.vmid->data[0], 1) - spu_extract(setup.vmin->data[0], 1);
+   setup.emaj.dx = spu_extract(setup.vmax->data[0], 0) - spu_extract(setup.vmin->data[0], 0);
+   setup.emaj.dy = spu_extract(setup.vmax->data[0], 1) - spu_extract(setup.vmin->data[0], 1);
+   setup.etop.dx = spu_extract(setup.vmax->data[0], 0) - spu_extract(setup.vmid->data[0], 0);
+   setup.etop.dy = spu_extract(setup.vmax->data[0], 1) - spu_extract(setup.vmid->data[0], 1);
+
+   /*
+    * Compute triangle's area.  Use 1/area to compute partial
+    * derivatives of attributes later.
+    *
+    * The area will be the same as prim->det, but the sign may be
+    * different depending on how the vertices get sorted above.
+    *
+    * To determine whether the primitive is front or back facing we
+    * use the prim->det value because its sign is correct.
+    */
+   {
+      const float area = (setup.emaj.dx * setup.ebot.dy - 
+			    setup.ebot.dx * setup.emaj.dy);
+
+      setup.oneoverarea = 1.0f / area;
+      /*
+      _mesa_printf("%s one-over-area %f  area %f  det %f\n",
+                   __FUNCTION__, setup.oneoverarea, area, prim->det );
+      */
+   }
+
+#if 0
+   /* We need to know if this is a front or back-facing triangle for:
+    *  - the GLSL gl_FrontFacing fragment attribute (bool)
+    *  - two-sided stencil test
+    */
+   setup.quad.facing = (prim->det > 0.0) ^ (setup.softpipe->rasterizer->front_winding == PIPE_WINDING_CW);
+#endif
+
+   return TRUE;
+}
+
+
+/**
+ * Compute a0 for a constant-valued coefficient (GL_FLAT shading).
+ * The value value comes from vertex->data[slot].
+ * The result will be put into setup.coef[slot].a0.
+ * \param slot  which attribute slot 
+ */
+static INLINE void
+const_coeff(uint slot)
+{
+   setup.coef[slot].dadx.v = (vector float) {0.0, 0.0, 0.0, 0.0};
+   setup.coef[slot].dady.v = (vector float) {0.0, 0.0, 0.0, 0.0};
+   setup.coef[slot].a0.v = setup.vprovoke->data[slot];
+}
+
+
+/**
+ * Compute a0, dadx and dady for a linearly interpolated coefficient,
+ * for a triangle.
+ */
+static INLINE void
+tri_linear_coeff(uint slot, uint firstComp, uint lastComp)
+{
+   uint i;
+   const float *vmin_d = (float *) &setup.vmin->data[slot];
+   const float *vmid_d = (float *) &setup.vmid->data[slot];
+   const float *vmax_d = (float *) &setup.vmax->data[slot];
+   const float x = spu_extract(setup.vmin->data[0], 0) - 0.5f;
+   const float y = spu_extract(setup.vmin->data[0], 1) - 0.5f;
+
+   for (i = firstComp; i < lastComp; i++) {
+      float botda = vmid_d[i] - vmin_d[i];
+      float majda = vmax_d[i] - vmin_d[i];
+      float a = setup.ebot.dy * majda - botda * setup.emaj.dy;
+      float b = setup.emaj.dx * botda - majda * setup.ebot.dx;
+   
+      ASSERT(slot < PIPE_MAX_SHADER_INPUTS);
+
+      setup.coef[slot].dadx.f[i] = a * setup.oneoverarea;
+      setup.coef[slot].dady.f[i] = b * setup.oneoverarea;
+
+      /* calculate a0 as the value which would be sampled for the
+       * fragment at (0,0), taking into account that we want to sample at
+       * pixel centers, in other words (0.5, 0.5).
+       *
+       * this is neat but unfortunately not a good way to do things for
+       * triangles with very large values of dadx or dady as it will
+       * result in the subtraction and re-addition from a0 of a very
+       * large number, which means we'll end up loosing a lot of the
+       * fractional bits and precision from a0.  the way to fix this is
+       * to define a0 as the sample at a pixel center somewhere near vmin
+       * instead - i'll switch to this later.
+       */
+      setup.coef[slot].a0.f[i] = (vmin_d[i] - 
+                                 (setup.coef[slot].dadx.f[i] * x + 
+                                  setup.coef[slot].dady.f[i] * y));
+   }
+
+   /*
+   _mesa_printf("attr[%d].%c: %f dx:%f dy:%f\n",
+		slot, "xyzw"[i], 
+		setup.coef[slot].a0[i],
+		setup.coef[slot].dadx.f[i],
+		setup.coef[slot].dady.f[i]);
+   */
+}
+
+
+/**
+ * As above, but interp setup all four vector components.
+ */
+static INLINE void
+tri_linear_coeff4(uint slot)
+{
+   const vector float vmin_d = setup.vmin->data[slot];
+   const vector float vmid_d = setup.vmid->data[slot];
+   const vector float vmax_d = setup.vmax->data[slot];
+   const vector float xxxx = spu_splats(spu_extract(setup.vmin->data[0], 0) - 0.5f);
+   const vector float yyyy = spu_splats(spu_extract(setup.vmin->data[0], 1) - 0.5f);
+
+   vector float botda = vmid_d - vmin_d;
+   vector float majda = vmax_d - vmin_d;
+
+   vector float a = spu_sub(spu_mul(spu_splats(setup.ebot.dy), majda),
+                            spu_mul(botda, spu_splats(setup.emaj.dy)));
+   vector float b = spu_sub(spu_mul(spu_splats(setup.emaj.dx), botda),
+                            spu_mul(majda, spu_splats(setup.ebot.dx)));
+
+   setup.coef[slot].dadx.v = spu_mul(a, spu_splats(setup.oneoverarea));
+   setup.coef[slot].dady.v = spu_mul(b, spu_splats(setup.oneoverarea));
+
+   vector float tempx = spu_mul(setup.coef[slot].dadx.v, xxxx);
+   vector float tempy = spu_mul(setup.coef[slot].dady.v, yyyy);
+                         
+   setup.coef[slot].a0.v = spu_sub(vmin_d, spu_add(tempx, tempy));
+}
+
+
+
+#if 0
+/**
+ * Compute a0, dadx and dady for a perspective-corrected interpolant,
+ * for a triangle.
+ * We basically multiply the vertex value by 1/w before computing
+ * the plane coefficients (a0, dadx, dady).
+ * Later, when we compute the value at a particular fragment position we'll
+ * divide the interpolated value by the interpolated W at that fragment.
+ */
+static void tri_persp_coeff( unsigned slot,
+                             unsigned i )
+{
+   /* premultiply by 1/w:
+    */
+   float mina = setup.vmin->data[slot][i] * setup.vmin->data[0][3];
+   float mida = setup.vmid->data[slot][i] * setup.vmid->data[0][3];
+   float maxa = setup.vmax->data[slot][i] * setup.vmax->data[0][3];
+
+   float botda = mida - mina;
+   float majda = maxa - mina;
+   float a = setup.ebot.dy * majda - botda * setup.emaj.dy;
+   float b = setup.emaj.dx * botda - majda * setup.ebot.dx;
+      
+   /*
+   printf("tri persp %d,%d: %f %f %f\n", slot, i,
+          setup.vmin->data[slot][i],
+          setup.vmid->data[slot][i],
+          setup.vmax->data[slot][i]
+          );
+   */
+
+   assert(slot < PIPE_MAX_SHADER_INPUTS);
+   assert(i <= 3);
+
+   setup.coef[slot].dadx.f[i] = a * setup.oneoverarea;
+   setup.coef[slot].dady.f[i] = b * setup.oneoverarea;
+   setup.coef[slot].a0.f[i] = (mina - 
+			    (setup.coef[slot].dadx.f[i] * (setup.vmin->data[0][0] - 0.5f) + 
+			     setup.coef[slot].dady.f[i] * (setup.vmin->data[0][1] - 0.5f)));
+}
+#endif
+
+
+/**
+ * Compute the setup.coef[] array dadx, dady, a0 values.
+ * Must be called after setup.vmin,vmid,vmax,vprovoke are initialized.
+ */
+static void setup_tri_coefficients(void)
+{
+#if 1
+   uint i;
+
+   for (i = 0; i < spu.vertex_info.num_attribs; i++) {
+      switch (spu.vertex_info.interp_mode[i]) {
+      case INTERP_NONE:
+         break;
+      case INTERP_POS:
+         /*tri_linear_coeff(i, 2, 3);*/
+         /* XXX interp W if PERSPECTIVE... */
+         tri_linear_coeff4(i);
+         break;
+      case INTERP_CONSTANT:
+         const_coeff(i);
+         break;
+      case INTERP_LINEAR:
+         tri_linear_coeff4(i);
+         break;
+      case INTERP_PERSPECTIVE:
+         tri_linear_coeff4(i);  /* temporary */
+         break;
+      default:
+         ASSERT(0);
+      }
+   }
+#else
+   ASSERT(spu.vertex_info.interp_mode[0] == INTERP_POS);
+   ASSERT(spu.vertex_info.interp_mode[1] == INTERP_LINEAR ||
+          spu.vertex_info.interp_mode[1] == INTERP_CONSTANT);
+   tri_linear_coeff(0, 2, 3);  /* slot 0, z */
+   tri_linear_coeff(1, 0, 4);  /* slot 1, color */
+#endif
+}
+
+
+static void setup_tri_edges(void)
+{
+   float vmin_x = spu_extract(setup.vmin->data[0], 0) + 0.5f;
+   float vmid_x = spu_extract(setup.vmid->data[0], 0) + 0.5f;
+
+   float vmin_y = spu_extract(setup.vmin->data[0], 1) - 0.5f;
+   float vmid_y = spu_extract(setup.vmid->data[0], 1) - 0.5f;
+   float vmax_y = spu_extract(setup.vmax->data[0], 1) - 0.5f;
+
+   setup.emaj.sy = CEILF(vmin_y);
+   setup.emaj.lines = (int) CEILF(vmax_y - setup.emaj.sy);
+   setup.emaj.dxdy = setup.emaj.dx / setup.emaj.dy;
+   setup.emaj.sx = vmin_x + (setup.emaj.sy - vmin_y) * setup.emaj.dxdy;
+
+   setup.etop.sy = CEILF(vmid_y);
+   setup.etop.lines = (int) CEILF(vmax_y - setup.etop.sy);
+   setup.etop.dxdy = setup.etop.dx / setup.etop.dy;
+   setup.etop.sx = vmid_x + (setup.etop.sy - vmid_y) * setup.etop.dxdy;
+
+   setup.ebot.sy = CEILF(vmin_y);
+   setup.ebot.lines = (int) CEILF(vmid_y - setup.ebot.sy);
+   setup.ebot.dxdy = setup.ebot.dx / setup.ebot.dy;
+   setup.ebot.sx = vmin_x + (setup.ebot.sy - vmin_y) * setup.ebot.dxdy;
+}
+
+
+/**
+ * Render the upper or lower half of a triangle.
+ * Scissoring/cliprect is applied here too.
+ */
+static void subtriangle( struct edge *eleft,
+			 struct edge *eright,
+			 unsigned lines )
+{
+   const int minx = setup.cliprect_minx;
+   const int maxx = setup.cliprect_maxx;
+   const int miny = setup.cliprect_miny;
+   const int maxy = setup.cliprect_maxy;
+   int y, start_y, finish_y;
+   int sy = (int)eleft->sy;
+
+   ASSERT((int)eleft->sy == (int) eright->sy);
+
+   /* clip top/bottom */
+   start_y = sy;
+   finish_y = sy + lines;
+
+   if (start_y < miny)
+      start_y = miny;
+
+   if (finish_y > maxy)
+      finish_y = maxy;
+
+   start_y -= sy;
+   finish_y -= sy;
+
+   /*
+   _mesa_printf("%s %d %d\n", __FUNCTION__, start_y, finish_y);  
+   */
+
+   for (y = start_y; y < finish_y; y++) {
+
+      /* avoid accumulating adds as floats don't have the precision to
+       * accurately iterate large triangle edges that way.  luckily we
+       * can just multiply these days.
+       *
+       * this is all drowned out by the attribute interpolation anyway.
+       */
+      int left = (int)(eleft->sx + y * eleft->dxdy);
+      int right = (int)(eright->sx + y * eright->dxdy);
+
+      /* clip left/right */
+      if (left < minx)
+         left = minx;
+      if (right > maxx)
+         right = maxx;
+
+      if (left < right) {
+         int _y = sy + y;
+         if (block(_y) != setup.span.y) {
+            flush_spans();
+            setup.span.y = block(_y);
+         }
+
+         setup.span.left[_y&1] = left;
+         setup.span.right[_y&1] = right;
+         setup.span.y_flags |= 1<<(_y&1);
+      }
+   }
+
+
+   /* save the values so that emaj can be restarted:
+    */
+   eleft->sx += lines * eleft->dxdy;
+   eright->sx += lines * eright->dxdy;
+   eleft->sy += lines;
+   eright->sy += lines;
+}
+
+
+/**
+ * Draw triangle into tile at (tx, ty) (tile coords)
+ * The tile data should have already been fetched.
+ */
+boolean
+tri_draw(const float *v0, const float *v1, const float *v2, uint tx, uint ty)
+{
+   setup.tx = tx;
+   setup.ty = ty;
+
+   /* set clipping bounds to tile bounds */
+   setup.cliprect_minx = tx * TILE_SIZE;
+   setup.cliprect_miny = ty * TILE_SIZE;
+   setup.cliprect_maxx = (tx + 1) * TILE_SIZE;
+   setup.cliprect_maxy = (ty + 1) * TILE_SIZE;
+
+   if (!setup_sort_vertices((struct vertex_header *) v0,
+                            (struct vertex_header *) v1,
+                            (struct vertex_header *) v2)) {
+      return FALSE; /* totally clipped */
+   }
+
+   setup_tri_coefficients();
+   setup_tri_edges();
+
+   setup.span.y = 0;
+   setup.span.y_flags = 0;
+   setup.span.right[0] = 0;
+   setup.span.right[1] = 0;
+   /*   setup.span.z_mode = tri_z_mode( setup.ctx ); */
+
+   /*   init_constant_attribs( setup ); */
+      
+   if (setup.oneoverarea < 0.0) {
+      /* emaj on left:
+       */
+      subtriangle( &setup.emaj, &setup.ebot, setup.ebot.lines );
+      subtriangle( &setup.emaj, &setup.etop, setup.etop.lines );
+   }
+   else {
+      /* emaj on right:
+       */
+      subtriangle( &setup.ebot, &setup.emaj, setup.ebot.lines );
+      subtriangle( &setup.etop, &setup.emaj, setup.etop.lines );
+   }
+
+   flush_spans();
+
+   return TRUE;
+}
diff --git a/src/gallium/drivers/cell/spu/spu_tri.h b/src/gallium/drivers/cell/spu/spu_tri.h
new file mode 100644
index 0000000000..aa694dd7c9
--- /dev/null
+++ b/src/gallium/drivers/cell/spu/spu_tri.h
@@ -0,0 +1,37 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+
+#ifndef SPU_TRI_H
+#define SPU_TRI_H
+
+
+extern boolean
+tri_draw(const float *v0, const float *v1, const float *v2, uint tx, uint ty);
+
+
+#endif /* SPU_TRI_H */
diff --git a/src/gallium/drivers/cell/spu/spu_util.c b/src/gallium/drivers/cell/spu/spu_util.c
new file mode 100644
index 0000000000..ac373240c1
--- /dev/null
+++ b/src/gallium/drivers/cell/spu/spu_util.c
@@ -0,0 +1,165 @@
+#include "pipe/p_util.h"
+#include "pipe/p_shader_tokens.h"
+#include "pipe/tgsi/util/tgsi_parse.h"
+//#include "tgsi_build.h"
+#include "pipe/tgsi/util/tgsi_util.h"
+
+unsigned
+tgsi_util_get_src_register_swizzle(
+   const struct tgsi_src_register *reg,
+   unsigned component )
+{
+   switch( component ) {
+   case 0:
+      return reg->SwizzleX;
+   case 1:
+      return reg->SwizzleY;
+   case 2:
+      return reg->SwizzleZ;
+   case 3:
+      return reg->SwizzleW;
+   default:
+      assert( 0 );
+   }
+   return 0;
+}
+
+unsigned
+tgsi_util_get_src_register_extswizzle(
+   const struct tgsi_src_register_ext_swz *reg,
+   unsigned component )
+{
+   switch( component ) {
+   case 0:
+      return reg->ExtSwizzleX;
+   case 1:
+      return reg->ExtSwizzleY;
+   case 2:
+      return reg->ExtSwizzleZ;
+   case 3:
+      return reg->ExtSwizzleW;
+   default:
+      assert( 0 );
+   }
+   return 0;
+}
+
+unsigned
+tgsi_util_get_full_src_register_extswizzle(
+   const struct tgsi_full_src_register  *reg,
+   unsigned component )
+{
+   unsigned swizzle;
+
+   /*
+    * First, calculate  the   extended swizzle for a given channel. This will give
+    * us either a channel index into the simple swizzle or  a constant 1 or   0.
+    */
+   swizzle = tgsi_util_get_src_register_extswizzle(
+      &reg->SrcRegisterExtSwz,
+      component );
+
+   assert (TGSI_SWIZZLE_X == TGSI_EXTSWIZZLE_X);
+   assert (TGSI_SWIZZLE_Y == TGSI_EXTSWIZZLE_Y);
+   assert (TGSI_SWIZZLE_Z == TGSI_EXTSWIZZLE_Z);
+   assert (TGSI_SWIZZLE_W == TGSI_EXTSWIZZLE_W);
+   assert (TGSI_EXTSWIZZLE_ZERO > TGSI_SWIZZLE_W);
+   assert (TGSI_EXTSWIZZLE_ONE > TGSI_SWIZZLE_W);
+
+   /*
+    * Second, calculate the simple  swizzle  for   the   unswizzled channel index.
+    * Leave the constants intact, they are   not   affected by the   simple swizzle.
+    */
+   if( swizzle <= TGSI_SWIZZLE_W ) {
+      swizzle = tgsi_util_get_src_register_swizzle(
+         &reg->SrcRegister,
+         component );
+   }
+
+   return swizzle;
+}
+
+unsigned
+tgsi_util_get_src_register_extnegate(
+   const  struct tgsi_src_register_ext_swz *reg,
+   unsigned component )
+{
+   switch( component ) {
+   case 0:
+      return reg->NegateX;
+   case 1:
+      return reg->NegateY;
+   case 2:
+      return reg->NegateZ;
+   case 3:
+      return reg->NegateW;
+   default:
+      assert( 0 );
+   }
+   return 0;
+}
+
+void
+tgsi_util_set_src_register_extnegate(
+   struct tgsi_src_register_ext_swz *reg,
+   unsigned negate,
+   unsigned component )
+{
+   switch( component ) {
+   case 0:
+      reg->NegateX = negate;
+      break;
+   case 1:
+      reg->NegateY = negate;
+      break;
+   case 2:
+      reg->NegateZ = negate;
+      break;
+   case 3:
+      reg->NegateW = negate;
+      break;
+   default:
+      assert( 0 );
+   }
+}
+
+unsigned
+tgsi_util_get_full_src_register_sign_mode(
+   const struct  tgsi_full_src_register *reg,
+   unsigned component )
+{
+   unsigned sign_mode;
+
+   if( reg->SrcRegisterExtMod.Absolute ) {
+      /* Consider only the post-abs negation. */
+
+      if( reg->SrcRegisterExtMod.Negate ) {
+         sign_mode = TGSI_UTIL_SIGN_SET;
+      }
+      else {
+         sign_mode = TGSI_UTIL_SIGN_CLEAR;
+      }
+   }
+   else {
+      /* Accumulate the three negations. */
+
+      unsigned negate;
+
+      negate = reg->SrcRegister.Negate;
+      if( tgsi_util_get_src_register_extnegate( &reg->SrcRegisterExtSwz, component ) ) {
+         negate = !negate;
+      }
+      if( reg->SrcRegisterExtMod.Negate ) {
+         negate = !negate;
+      }
+
+      if( negate ) {
+         sign_mode = TGSI_UTIL_SIGN_TOGGLE;
+      }
+      else {
+         sign_mode = TGSI_UTIL_SIGN_KEEP;
+      }
+   }
+
+   return sign_mode;
+}
diff --git a/src/gallium/drivers/cell/spu/spu_vertex_fetch.c b/src/gallium/drivers/cell/spu/spu_vertex_fetch.c
new file mode 100644
index 0000000000..45e3c26c00
--- /dev/null
+++ b/src/gallium/drivers/cell/spu/spu_vertex_fetch.c
@@ -0,0 +1,673 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+ /*
+  * Authors:
+  *   Keith Whitwell <keith@tungstengraphics.com>
+  */
+
+#include <spu_mfcio.h>
+#include <transpose_matrix4x4.h>
+
+#include "pipe/p_util.h"
+#include "pipe/p_state.h"
+#include "pipe/p_shader_tokens.h"
+#include "spu_exec.h"
+#include "spu_vertex_shader.h"
+#include "spu_main.h"
+
+#define CACHE_NAME            attribute
+#define CACHED_TYPE           qword
+#define CACHE_TYPE            CACHE_TYPE_RO
+#define CACHE_SET_TAGID(set)  TAG_VERTEX_BUFFER
+#define CACHE_LOG2NNWAY       2
+#define CACHE_LOG2NSETS       6
+#include <cache-api.h>
+
+/* Yes folks, this is ugly.
+ */
+#undef CACHE_NWAY
+#undef CACHE_NSETS
+#define CACHE_NAME            attribute
+#define CACHE_NWAY            4
+#define CACHE_NSETS           (1U << 6)
+
+
+#define DRAW_DBG 0
+
+static const qword fetch_shuffle_data[] = {
+   /* Shuffle used by CVT_64_FLOAT
+    */
+   {
+      0x00, 0x01, 0x02, 0x03, 0x10, 0x11, 0x12, 0x13,
+      0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80,
+   },
+
+   /* Shuffle used by CVT_8_USCALED and CVT_8_SSCALED
+    */
+   {
+      0x00, 0x80, 0x80, 0x80, 0x01, 0x80, 0x80, 0x80,
+      0x02, 0x80, 0x80, 0x80, 0x03, 0x80, 0x80, 0x80,
+   },
+   
+   /* Shuffle used by CVT_16_USCALED and CVT_16_SSCALED
+    */
+   {
+      0x00, 0x01, 0x80, 0x80, 0x02, 0x03, 0x80, 0x80,
+      0x04, 0x05, 0x80, 0x80, 0x06, 0x07, 0x80, 0x80,
+   },
+   
+   /* High value shuffle used by trans4x4.
+    */
+   {
+      0x00, 0x01, 0x02, 0x03, 0x10, 0x11, 0x12, 0x13,
+      0x04, 0x05, 0x06, 0x07, 0x14, 0x15, 0x16, 0x17
+   },
+
+   /* Low value shuffle used by trans4x4.
+    */
+   {
+      0x08, 0x09, 0x0A, 0x0B, 0x18, 0x19, 0x1A, 0x1B,
+      0x0C, 0x0D, 0x0E, 0x0F, 0x1C, 0x1D, 0x1E, 0x1F
+   }
+};
+
+
+static INLINE void
+trans4x4(qword row0, qword row1, qword row2, qword row3, qword *out,
+         const qword *shuffle)
+{
+   qword t1 = si_shufb(row0, row2, shuffle[3]);
+   qword t2 = si_shufb(row0, row2, shuffle[4]);
+   qword t3 = si_shufb(row1, row3, shuffle[3]);
+   qword t4 = si_shufb(row1, row3, shuffle[4]);
+
+   out[0] = si_shufb(t1, t3, shuffle[3]);
+   out[1] = si_shufb(t1, t3, shuffle[4]);
+   out[2] = si_shufb(t2, t4, shuffle[3]);
+   out[3] = si_shufb(t2, t4, shuffle[4]);
+}
+
+
+/**
+ * Fetch between 1 and 32 bytes from an unaligned address
+ */
+static INLINE void
+fetch_unaligned(qword *dst, unsigned ea, unsigned size)
+{
+   qword tmp[4];
+   const int shift = ea & 0x0f;
+   const unsigned aligned_start_ea = ea & ~0x0f;
+   const unsigned aligned_end_ea = (ea + size) & ~0x0f;
+   const unsigned num_entries = ((aligned_end_ea - aligned_start_ea) / 16) + 1;
+   unsigned i;
+
+
+   if (shift == 0) {
+      /* Data is already aligned.  Fetch directly into the destination buffer.
+       */
+      for (i = 0; i < num_entries; i++) {
+	 dst[i] = cache_rd(attribute, (ea & ~0x0f) + (i * 16));
+      }
+   } else {
+      /* Fetch data from the cache to the local buffer.
+       */
+      for (i = 0; i < num_entries; i++) {
+	 tmp[i] = cache_rd(attribute, (ea & ~0x0f) + (i * 16));
+      }
+
+
+      /* Fix the alignment of the data and write to the destination buffer.
+       */
+      for (i = 0; i < ((size + 15) / 16); i++) {
+	 dst[i] = si_or((qword) spu_slqwbyte(tmp[i], shift),
+			(qword) spu_rlmaskqwbyte(tmp[i + 1], shift - 16));
+      }
+   }
+}
+
+
+#define CVT_32_FLOAT(q, s)    (*(q))
+
+static INLINE qword
+CVT_64_FLOAT(const qword *qw, const qword *shuffle)
+{
+   qword a = si_frds(qw[0]);
+   qword b = si_frds(si_rotqbyi(qw[0], 8));
+   qword c = si_frds(qw[1]);
+   qword d = si_frds(si_rotqbyi(qw[1], 8));
+
+   qword ab = si_shufb(a, b, shuffle[0]);
+   qword cd = si_shufb(c, d, si_rotqbyi(shuffle[0], 8));
+   
+   return si_or(ab, cd);
+}
+
+
+static INLINE qword
+CVT_8_USCALED(const qword *qw, const qword *shuffle)
+{
+   return si_cuflt(si_shufb(*qw, *qw, shuffle[1]), 0);
+}
+
+
+static INLINE qword
+CVT_16_USCALED(const qword *qw, const qword *shuffle)
+{
+   return si_cuflt(si_shufb(*qw, *qw, shuffle[2]), 0);
+}
+
+
+static INLINE qword
+CVT_32_USCALED(const qword *qw, const qword *shuffle)
+{
+   (void) shuffle;
+   return si_cuflt(*qw, 0);
+}
+
+static INLINE qword
+CVT_8_SSCALED(const qword *qw, const qword *shuffle)
+{
+   return si_csflt(si_shufb(*qw, *qw, shuffle[1]), 0);
+}
+
+
+static INLINE qword
+CVT_16_SSCALED(const qword *qw, const qword *shuffle)
+{
+   return si_csflt(si_shufb(*qw, *qw, shuffle[2]), 0);
+}
+
+
+static INLINE qword
+CVT_32_SSCALED(const qword *qw, const qword *shuffle)
+{
+   (void) shuffle;
+   return si_csflt(*qw, 0);
+}
+
+
+static INLINE qword
+CVT_8_UNORM(const qword *qw, const qword *shuffle)
+{
+   const qword scale = (qword) spu_splats(1.0f / 255.0f);
+   return si_fm(CVT_8_USCALED(qw, shuffle), scale);
+}
+
+
+static INLINE qword
+CVT_16_UNORM(const qword *qw, const qword *shuffle)
+{
+   const qword scale = (qword) spu_splats(1.0f / 65535.0f);
+   return si_fm(CVT_16_USCALED(qw, shuffle), scale);
+}
+
+
+static INLINE qword
+CVT_32_UNORM(const qword *qw, const qword *shuffle)
+{
+   const qword scale = (qword) spu_splats(1.0f / 4294967295.0f);
+   return si_fm(CVT_32_USCALED(qw, shuffle), scale);
+}
+
+
+static INLINE qword
+CVT_8_SNORM(const qword *qw, const qword *shuffle)
+{
+   const qword scale = (qword) spu_splats(1.0f / 127.0f);
+   return si_fm(CVT_8_SSCALED(qw, shuffle), scale);
+}
+
+
+static INLINE qword
+CVT_16_SNORM(const qword *qw, const qword *shuffle)
+{
+   const qword scale = (qword) spu_splats(1.0f / 32767.0f);
+   return si_fm(CVT_16_SSCALED(qw, shuffle), scale);
+}
+
+
+static INLINE qword
+CVT_32_SNORM(const qword *qw, const qword *shuffle)
+{
+   const qword scale = (qword) spu_splats(1.0f / 2147483647.0f);
+   return si_fm(CVT_32_SSCALED(qw, shuffle), scale);
+}
+
+#define SZ_4 si_il(0U)
+#define SZ_3 si_fsmbi(0x000f)
+#define SZ_2 si_fsmbi(0x00ff)
+#define SZ_1 si_fsmbi(0x0fff)
+
+/**
+ * Fetch a float[4] vertex attribute from memory, doing format/type
+ * conversion as needed.
+ *
+ * This is probably needed/dupliocated elsewhere, eg format
+ * conversion, texture sampling etc.
+ */
+#define FETCH_ATTRIB( NAME, SZ, CVT, N )			\
+static void							\
+fetch_##NAME(qword *out, const qword *in, qword defaults, \
+                const qword *shuffle)	\
+{								\
+   qword tmp[4];						\
+								\
+   tmp[0] = si_selb(CVT(in + (0 * N), shuffle), defaults, SZ);		\
+   tmp[1] = si_selb(CVT(in + (1 * N), shuffle), defaults, SZ);		\
+   tmp[2] = si_selb(CVT(in + (2 * N), shuffle), defaults, SZ);		\
+   tmp[3] = si_selb(CVT(in + (3 * N), shuffle), defaults, SZ);		\
+   trans4x4(tmp[0], tmp[1], tmp[2], tmp[3], out, shuffle);		\
+}
+
+
+FETCH_ATTRIB( R64G64B64A64_FLOAT,   SZ_4, CVT_64_FLOAT, 2 )
+FETCH_ATTRIB( R64G64B64_FLOAT,      SZ_3, CVT_64_FLOAT, 2 )
+FETCH_ATTRIB( R64G64_FLOAT,         SZ_2, CVT_64_FLOAT, 2 )
+FETCH_ATTRIB( R64_FLOAT,            SZ_1, CVT_64_FLOAT, 2 )
+
+FETCH_ATTRIB( R32G32B32A32_FLOAT,   SZ_4, CVT_32_FLOAT, 1 )
+FETCH_ATTRIB( R32G32B32_FLOAT,      SZ_3, CVT_32_FLOAT, 1 )
+FETCH_ATTRIB( R32G32_FLOAT,         SZ_2, CVT_32_FLOAT, 1 )
+FETCH_ATTRIB( R32_FLOAT,            SZ_1, CVT_32_FLOAT, 1 )
+
+FETCH_ATTRIB( R32G32B32A32_USCALED, SZ_4, CVT_32_USCALED, 1 )
+FETCH_ATTRIB( R32G32B32_USCALED,    SZ_3, CVT_32_USCALED, 1 )
+FETCH_ATTRIB( R32G32_USCALED,       SZ_2, CVT_32_USCALED, 1 )
+FETCH_ATTRIB( R32_USCALED,          SZ_1, CVT_32_USCALED, 1 )
+
+FETCH_ATTRIB( R32G32B32A32_SSCALED, SZ_4, CVT_32_SSCALED, 1 )
+FETCH_ATTRIB( R32G32B32_SSCALED,    SZ_3, CVT_32_SSCALED, 1 )
+FETCH_ATTRIB( R32G32_SSCALED,       SZ_2, CVT_32_SSCALED, 1 )
+FETCH_ATTRIB( R32_SSCALED,          SZ_1, CVT_32_SSCALED, 1 )
+
+FETCH_ATTRIB( R32G32B32A32_UNORM, SZ_4, CVT_32_UNORM, 1 )
+FETCH_ATTRIB( R32G32B32_UNORM,    SZ_3, CVT_32_UNORM, 1 )
+FETCH_ATTRIB( R32G32_UNORM,       SZ_2, CVT_32_UNORM, 1 )
+FETCH_ATTRIB( R32_UNORM,          SZ_1, CVT_32_UNORM, 1 )
+
+FETCH_ATTRIB( R32G32B32A32_SNORM, SZ_4, CVT_32_SNORM, 1 )
+FETCH_ATTRIB( R32G32B32_SNORM,    SZ_3, CVT_32_SNORM, 1 )
+FETCH_ATTRIB( R32G32_SNORM,       SZ_2, CVT_32_SNORM, 1 )
+FETCH_ATTRIB( R32_SNORM,          SZ_1, CVT_32_SNORM, 1 )
+
+FETCH_ATTRIB( R16G16B16A16_USCALED, SZ_4, CVT_16_USCALED, 1 )
+FETCH_ATTRIB( R16G16B16_USCALED,    SZ_3, CVT_16_USCALED, 1 )
+FETCH_ATTRIB( R16G16_USCALED,       SZ_2, CVT_16_USCALED, 1 )
+FETCH_ATTRIB( R16_USCALED,          SZ_1, CVT_16_USCALED, 1 )
+
+FETCH_ATTRIB( R16G16B16A16_SSCALED, SZ_4, CVT_16_SSCALED, 1 )
+FETCH_ATTRIB( R16G16B16_SSCALED,    SZ_3, CVT_16_SSCALED, 1 )
+FETCH_ATTRIB( R16G16_SSCALED,       SZ_2, CVT_16_SSCALED, 1 )
+FETCH_ATTRIB( R16_SSCALED,          SZ_1, CVT_16_SSCALED, 1 )
+
+FETCH_ATTRIB( R16G16B16A16_UNORM, SZ_4, CVT_16_UNORM, 1 )
+FETCH_ATTRIB( R16G16B16_UNORM,    SZ_3, CVT_16_UNORM, 1 )
+FETCH_ATTRIB( R16G16_UNORM,       SZ_2, CVT_16_UNORM, 1 )
+FETCH_ATTRIB( R16_UNORM,          SZ_1, CVT_16_UNORM, 1 )
+
+FETCH_ATTRIB( R16G16B16A16_SNORM, SZ_4, CVT_16_SNORM, 1 )
+FETCH_ATTRIB( R16G16B16_SNORM,    SZ_3, CVT_16_SNORM, 1 )
+FETCH_ATTRIB( R16G16_SNORM,       SZ_2, CVT_16_SNORM, 1 )
+FETCH_ATTRIB( R16_SNORM,          SZ_1, CVT_16_SNORM, 1 )
+
+FETCH_ATTRIB( R8G8B8A8_USCALED,   SZ_4, CVT_8_USCALED, 1 )
+FETCH_ATTRIB( R8G8B8_USCALED,     SZ_3, CVT_8_USCALED, 1 )
+FETCH_ATTRIB( R8G8_USCALED,       SZ_2, CVT_8_USCALED, 1 )
+FETCH_ATTRIB( R8_USCALED,         SZ_1, CVT_8_USCALED, 1 )
+
+FETCH_ATTRIB( R8G8B8A8_SSCALED,  SZ_4, CVT_8_SSCALED, 1 )
+FETCH_ATTRIB( R8G8B8_SSCALED,    SZ_3, CVT_8_SSCALED, 1 )
+FETCH_ATTRIB( R8G8_SSCALED,      SZ_2, CVT_8_SSCALED, 1 )
+FETCH_ATTRIB( R8_SSCALED,        SZ_1, CVT_8_SSCALED, 1 )
+
+FETCH_ATTRIB( R8G8B8A8_UNORM,  SZ_4, CVT_8_UNORM, 1 )
+FETCH_ATTRIB( R8G8B8_UNORM,    SZ_3, CVT_8_UNORM, 1 )
+FETCH_ATTRIB( R8G8_UNORM,      SZ_2, CVT_8_UNORM, 1 )
+FETCH_ATTRIB( R8_UNORM,        SZ_1, CVT_8_UNORM, 1 )
+
+FETCH_ATTRIB( R8G8B8A8_SNORM,  SZ_4, CVT_8_SNORM, 1 )
+FETCH_ATTRIB( R8G8B8_SNORM,    SZ_3, CVT_8_SNORM, 1 )
+FETCH_ATTRIB( R8G8_SNORM,      SZ_2, CVT_8_SNORM, 1 )
+FETCH_ATTRIB( R8_SNORM,        SZ_1, CVT_8_SNORM, 1 )
+
+FETCH_ATTRIB( A8R8G8B8_UNORM,       SZ_4, CVT_8_UNORM, 1 )
+
+
+
+static spu_fetch_func get_fetch_func( enum pipe_format format )
+{
+   switch (format) {
+   case PIPE_FORMAT_R64_FLOAT:
+      return fetch_R64_FLOAT;
+   case PIPE_FORMAT_R64G64_FLOAT:
+      return fetch_R64G64_FLOAT;
+   case PIPE_FORMAT_R64G64B64_FLOAT:
+      return fetch_R64G64B64_FLOAT;
+   case PIPE_FORMAT_R64G64B64A64_FLOAT:
+      return fetch_R64G64B64A64_FLOAT;
+
+   case PIPE_FORMAT_R32_FLOAT:
+      return fetch_R32_FLOAT;
+   case PIPE_FORMAT_R32G32_FLOAT:
+      return fetch_R32G32_FLOAT;
+   case PIPE_FORMAT_R32G32B32_FLOAT:
+      return fetch_R32G32B32_FLOAT;
+   case PIPE_FORMAT_R32G32B32A32_FLOAT:
+      return fetch_R32G32B32A32_FLOAT;
+
+   case PIPE_FORMAT_R32_UNORM:
+      return fetch_R32_UNORM;
+   case PIPE_FORMAT_R32G32_UNORM:
+      return fetch_R32G32_UNORM;
+   case PIPE_FORMAT_R32G32B32_UNORM:
+      return fetch_R32G32B32_UNORM;
+   case PIPE_FORMAT_R32G32B32A32_UNORM:
+      return fetch_R32G32B32A32_UNORM;
+
+   case PIPE_FORMAT_R32_USCALED:
+      return fetch_R32_USCALED;
+   case PIPE_FORMAT_R32G32_USCALED:
+      return fetch_R32G32_USCALED;
+   case PIPE_FORMAT_R32G32B32_USCALED:
+      return fetch_R32G32B32_USCALED;
+   case PIPE_FORMAT_R32G32B32A32_USCALED:
+      return fetch_R32G32B32A32_USCALED;
+
+   case PIPE_FORMAT_R32_SNORM:
+      return fetch_R32_SNORM;
+   case PIPE_FORMAT_R32G32_SNORM:
+      return fetch_R32G32_SNORM;
+   case PIPE_FORMAT_R32G32B32_SNORM:
+      return fetch_R32G32B32_SNORM;
+   case PIPE_FORMAT_R32G32B32A32_SNORM:
+      return fetch_R32G32B32A32_SNORM;
+
+   case PIPE_FORMAT_R32_SSCALED:
+      return fetch_R32_SSCALED;
+   case PIPE_FORMAT_R32G32_SSCALED:
+      return fetch_R32G32_SSCALED;
+   case PIPE_FORMAT_R32G32B32_SSCALED:
+      return fetch_R32G32B32_SSCALED;
+   case PIPE_FORMAT_R32G32B32A32_SSCALED:
+      return fetch_R32G32B32A32_SSCALED;
+
+   case PIPE_FORMAT_R16_UNORM:
+      return fetch_R16_UNORM;
+   case PIPE_FORMAT_R16G16_UNORM:
+      return fetch_R16G16_UNORM;
+   case PIPE_FORMAT_R16G16B16_UNORM:
+      return fetch_R16G16B16_UNORM;
+   case PIPE_FORMAT_R16G16B16A16_UNORM:
+      return fetch_R16G16B16A16_UNORM;
+
+   case PIPE_FORMAT_R16_USCALED:
+      return fetch_R16_USCALED;
+   case PIPE_FORMAT_R16G16_USCALED:
+      return fetch_R16G16_USCALED;
+   case PIPE_FORMAT_R16G16B16_USCALED:
+      return fetch_R16G16B16_USCALED;
+   case PIPE_FORMAT_R16G16B16A16_USCALED:
+      return fetch_R16G16B16A16_USCALED;
+
+   case PIPE_FORMAT_R16_SNORM:
+      return fetch_R16_SNORM;
+   case PIPE_FORMAT_R16G16_SNORM:
+      return fetch_R16G16_SNORM;
+   case PIPE_FORMAT_R16G16B16_SNORM:
+      return fetch_R16G16B16_SNORM;
+   case PIPE_FORMAT_R16G16B16A16_SNORM:
+      return fetch_R16G16B16A16_SNORM;
+
+   case PIPE_FORMAT_R16_SSCALED:
+      return fetch_R16_SSCALED;
+   case PIPE_FORMAT_R16G16_SSCALED:
+      return fetch_R16G16_SSCALED;
+   case PIPE_FORMAT_R16G16B16_SSCALED:
+      return fetch_R16G16B16_SSCALED;
+   case PIPE_FORMAT_R16G16B16A16_SSCALED:
+      return fetch_R16G16B16A16_SSCALED;
+
+   case PIPE_FORMAT_R8_UNORM:
+      return fetch_R8_UNORM;
+   case PIPE_FORMAT_R8G8_UNORM:
+      return fetch_R8G8_UNORM;
+   case PIPE_FORMAT_R8G8B8_UNORM:
+      return fetch_R8G8B8_UNORM;
+   case PIPE_FORMAT_R8G8B8A8_UNORM:
+      return fetch_R8G8B8A8_UNORM;
+
+   case PIPE_FORMAT_R8_USCALED:
+      return fetch_R8_USCALED;
+   case PIPE_FORMAT_R8G8_USCALED:
+      return fetch_R8G8_USCALED;
+   case PIPE_FORMAT_R8G8B8_USCALED:
+      return fetch_R8G8B8_USCALED;
+   case PIPE_FORMAT_R8G8B8A8_USCALED:
+      return fetch_R8G8B8A8_USCALED;
+
+   case PIPE_FORMAT_R8_SNORM:
+      return fetch_R8_SNORM;
+   case PIPE_FORMAT_R8G8_SNORM:
+      return fetch_R8G8_SNORM;
+   case PIPE_FORMAT_R8G8B8_SNORM:
+      return fetch_R8G8B8_SNORM;
+   case PIPE_FORMAT_R8G8B8A8_SNORM:
+      return fetch_R8G8B8A8_SNORM;
+
+   case PIPE_FORMAT_R8_SSCALED:
+      return fetch_R8_SSCALED;
+   case PIPE_FORMAT_R8G8_SSCALED:
+      return fetch_R8G8_SSCALED;
+   case PIPE_FORMAT_R8G8B8_SSCALED:
+      return fetch_R8G8B8_SSCALED;
+   case PIPE_FORMAT_R8G8B8A8_SSCALED:
+      return fetch_R8G8B8A8_SSCALED;
+
+   case PIPE_FORMAT_A8R8G8B8_UNORM:
+      return fetch_A8R8G8B8_UNORM;
+
+   case 0:
+      return NULL;		/* not sure why this is needed */
+
+   default:
+      assert(0);
+      return NULL;
+   }
+}
+
+
+static unsigned get_vertex_size( enum pipe_format format )
+{
+   switch (format) {
+   case PIPE_FORMAT_R64_FLOAT:
+      return 8;
+   case PIPE_FORMAT_R64G64_FLOAT:
+      return 2 * 8;
+   case PIPE_FORMAT_R64G64B64_FLOAT:
+      return 3 * 8;
+   case PIPE_FORMAT_R64G64B64A64_FLOAT:
+      return 4 * 8;
+
+   case PIPE_FORMAT_R32_SSCALED:
+   case PIPE_FORMAT_R32_SNORM:
+   case PIPE_FORMAT_R32_USCALED:
+   case PIPE_FORMAT_R32_UNORM:
+   case PIPE_FORMAT_R32_FLOAT:
+      return 4;
+   case PIPE_FORMAT_R32G32_SSCALED:
+   case PIPE_FORMAT_R32G32_SNORM:
+   case PIPE_FORMAT_R32G32_USCALED:
+   case PIPE_FORMAT_R32G32_UNORM:
+   case PIPE_FORMAT_R32G32_FLOAT:
+      return 2 * 4;
+   case PIPE_FORMAT_R32G32B32_SSCALED:
+   case PIPE_FORMAT_R32G32B32_SNORM:
+   case PIPE_FORMAT_R32G32B32_USCALED:
+   case PIPE_FORMAT_R32G32B32_UNORM:
+   case PIPE_FORMAT_R32G32B32_FLOAT:
+      return 3 * 4;
+   case PIPE_FORMAT_R32G32B32A32_SSCALED:
+   case PIPE_FORMAT_R32G32B32A32_SNORM:
+   case PIPE_FORMAT_R32G32B32A32_USCALED:
+   case PIPE_FORMAT_R32G32B32A32_UNORM:
+   case PIPE_FORMAT_R32G32B32A32_FLOAT:
+      return 4 * 4;
+
+   case PIPE_FORMAT_R16_SSCALED:
+   case PIPE_FORMAT_R16_SNORM:
+   case PIPE_FORMAT_R16_UNORM:
+   case PIPE_FORMAT_R16_USCALED:
+      return 2;
+   case PIPE_FORMAT_R16G16_SSCALED:
+   case PIPE_FORMAT_R16G16_SNORM:
+   case PIPE_FORMAT_R16G16_USCALED:
+   case PIPE_FORMAT_R16G16_UNORM:
+      return 2 * 2;
+   case PIPE_FORMAT_R16G16B16_SSCALED:
+   case PIPE_FORMAT_R16G16B16_SNORM:
+   case PIPE_FORMAT_R16G16B16_USCALED:
+   case PIPE_FORMAT_R16G16B16_UNORM:
+      return 3 * 2;
+   case PIPE_FORMAT_R16G16B16A16_SSCALED:
+   case PIPE_FORMAT_R16G16B16A16_SNORM:
+   case PIPE_FORMAT_R16G16B16A16_USCALED:
+   case PIPE_FORMAT_R16G16B16A16_UNORM:
+      return 4 * 2;
+
+   case PIPE_FORMAT_R8_SSCALED:
+   case PIPE_FORMAT_R8_SNORM:
+   case PIPE_FORMAT_R8_USCALED:
+   case PIPE_FORMAT_R8_UNORM:
+      return 1;
+   case PIPE_FORMAT_R8G8_SSCALED:
+   case PIPE_FORMAT_R8G8_SNORM:
+   case PIPE_FORMAT_R8G8_USCALED:
+   case PIPE_FORMAT_R8G8_UNORM:
+      return 2 * 1;
+   case PIPE_FORMAT_R8G8B8_SSCALED:
+   case PIPE_FORMAT_R8G8B8_SNORM:
+   case PIPE_FORMAT_R8G8B8_USCALED:
+   case PIPE_FORMAT_R8G8B8_UNORM:
+      return 3 * 1;
+   case PIPE_FORMAT_A8R8G8B8_UNORM:
+   case PIPE_FORMAT_R8G8B8A8_SSCALED:
+   case PIPE_FORMAT_R8G8B8A8_SNORM:
+   case PIPE_FORMAT_R8G8B8A8_USCALED:
+   case PIPE_FORMAT_R8G8B8A8_UNORM:
+      return 4 * 1;
+
+   case 0:
+      return 0;		/* not sure why this is needed */
+
+   default:
+      assert(0);
+      return 0;
+   }
+}
+
+
+/**
+ * Fetch vertex attributes for 'count' vertices.
+ */
+static void generic_vertex_fetch(struct spu_vs_context *draw,
+                                 struct spu_exec_machine *machine,
+                                 const unsigned *elts,
+                                 unsigned count)
+{
+   unsigned nr_attrs = draw->vertex_fetch.nr_attrs;
+   unsigned attr;
+
+   assert(count <= 4);
+
+#if DRAW_DBG
+   printf("SPU: %s count = %u, nr_attrs = %u\n", 
+          __FUNCTION__, count, nr_attrs);
+#endif
+
+   /* loop over vertex attributes (vertex shader inputs)
+    */
+   for (attr = 0; attr < nr_attrs; attr++) {
+      const qword default_values = (qword)(vec_float4){ 0.0, 0.0, 0.0, 1.0 };
+      const unsigned pitch = draw->vertex_fetch.pitch[attr];
+      const uint64_t src = draw->vertex_fetch.src_ptr[attr];
+      const spu_fetch_func fetch = draw->vertex_fetch.fetch[attr];
+      unsigned i;
+      unsigned idx;
+      const unsigned bytes_per_entry = draw->vertex_fetch.size[attr];
+      const unsigned quads_per_entry = (bytes_per_entry + 15) / 16;
+      qword in[2 * 4];
+
+
+      /* Fetch four attributes for four vertices.  
+       */
+      idx = 0;
+      for (i = 0; i < count; i++) {
+         const uint64_t addr = src + (elts[i] * pitch);
+
+#if DRAW_DBG
+         printf("SPU: fetching = 0x%llx\n", addr);
+#endif
+
+         fetch_unaligned(& in[idx], addr, bytes_per_entry);
+         idx += quads_per_entry;
+      }
+
+      /* Be nice and zero out any missing vertices.
+       */
+      (void) memset(& in[idx], 0, (8 - idx) * sizeof(qword));
+
+
+      /* Convert all 4 vertices to vectors of float.
+       */
+      (*fetch)(&machine->Inputs[attr].xyzw[0].q, in, default_values,
+               fetch_shuffle_data);
+   }
+}
+
+
+void spu_update_vertex_fetch( struct spu_vs_context *draw )
+{
+   unsigned i;
+
+   
+   /* Invalidate the vertex cache.
+    */
+   for (i = 0; i < (CACHE_NWAY * CACHE_NSETS); i++) {
+      CACHELINE_CLEARVALID(i);
+   }
+
+
+   for (i = 0; i < draw->vertex_fetch.nr_attrs; i++) {
+      draw->vertex_fetch.fetch[i] =
+          get_fetch_func(draw->vertex_fetch.format[i]);
+      draw->vertex_fetch.size[i] =
+          get_vertex_size(draw->vertex_fetch.format[i]);
+   }
+
+   draw->vertex_fetch.fetch_func = generic_vertex_fetch;
+}
diff --git a/src/gallium/drivers/cell/spu/spu_vertex_shader.c b/src/gallium/drivers/cell/spu/spu_vertex_shader.c
new file mode 100644
index 0000000000..c1cbbb6d1e
--- /dev/null
+++ b/src/gallium/drivers/cell/spu/spu_vertex_shader.c
@@ -0,0 +1,231 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+ /*
+  * Authors:
+  *   Keith Whitwell <keith@tungstengraphics.com>
+  *   Brian Paul
+  *   Ian Romanick <idr@us.ibm.com>
+  */
+
+#include <spu_mfcio.h>
+
+#include "pipe/p_util.h"
+#include "pipe/p_state.h"
+#include "pipe/p_shader_tokens.h"
+#include "spu_vertex_shader.h"
+#include "spu_exec.h"
+#include "pipe/draw/draw_private.h"
+#include "pipe/draw/draw_context.h"
+#include "pipe/cell/common.h"
+#include "spu_main.h"
+
+static INLINE unsigned
+compute_clipmask(const float *clip, /*const*/ float plane[][4], unsigned nr)
+{
+   unsigned mask = 0;
+   unsigned i;
+
+   /* Do the hardwired planes first:
+    */
+   if (-clip[0] + clip[3] < 0) mask |= CLIP_RIGHT_BIT;
+   if ( clip[0] + clip[3] < 0) mask |= CLIP_LEFT_BIT;
+   if (-clip[1] + clip[3] < 0) mask |= CLIP_TOP_BIT;
+   if ( clip[1] + clip[3] < 0) mask |= CLIP_BOTTOM_BIT;
+   if (-clip[2] + clip[3] < 0) mask |= CLIP_FAR_BIT;
+   if ( clip[2] + clip[3] < 0) mask |= CLIP_NEAR_BIT;
+
+   /* Followed by any remaining ones:
+    */
+   for (i = 6; i < nr; i++) {
+      if (dot4(clip, plane[i]) < 0) 
+         mask |= (1<<i);
+   }
+
+   return mask;
+}
+
+
+/**
+ * Transform vertices with the current vertex program/shader
+ * Up to four vertices can be shaded at a time.
+ * \param vbuffer  the input vertex data
+ * \param elts  indexes of four input vertices
+ * \param count  number of vertices to shade [1..4]
+ * \param vOut  array of pointers to four output vertices
+ */
+static void
+run_vertex_program(struct spu_vs_context *draw,
+                   unsigned elts[4], unsigned count,
+                   const uint64_t *vOut)
+{
+   struct spu_exec_machine *machine = &draw->machine;
+   unsigned int j;
+
+   ALIGN16_DECL(struct spu_exec_vector, inputs, PIPE_ATTRIB_MAX);
+   ALIGN16_DECL(struct spu_exec_vector, outputs, PIPE_ATTRIB_MAX);
+   const float *scale = draw->viewport.scale;
+   const float *trans = draw->viewport.translate;
+
+   assert(count <= 4);
+
+   machine->Processor = TGSI_PROCESSOR_VERTEX;
+
+   ASSERT_ALIGN16(draw->constants);
+   machine->Consts = (float (*)[4]) draw->constants;
+
+   machine->Inputs = ALIGN16_ASSIGN(inputs);
+   machine->Outputs = ALIGN16_ASSIGN(outputs);
+
+   spu_vertex_fetch( draw, machine, elts, count );
+
+   /* run shader */
+   spu_exec_machine_run( machine );
+
+
+   /* store machine results */
+   for (j = 0; j < count; j++) {
+      unsigned slot;
+      float x, y, z, w;
+      unsigned char buffer[sizeof(struct vertex_header)
+          + MAX_VERTEX_SIZE] ALIGN16_ATTRIB;
+      struct vertex_header *const tmpOut =
+          (struct vertex_header *) buffer;
+      const unsigned vert_size = ROUNDUP16(sizeof(struct vertex_header)
+                                           + (sizeof(float) * 4 
+                                              * draw->num_vs_outputs));
+
+      mfc_get(tmpOut, vOut[j], vert_size, TAG_VERTEX_BUFFER, 0, 0);
+      wait_on_mask(1 << TAG_VERTEX_BUFFER);
+
+
+      /* Handle attr[0] (position) specially:
+       *
+       * XXX: Computing the clipmask should be done in the vertex
+       * program as a set of DP4 instructions appended to the
+       * user-provided code.
+       */
+      x = tmpOut->clip[0] = machine->Outputs[0].xyzw[0].f[j];
+      y = tmpOut->clip[1] = machine->Outputs[0].xyzw[1].f[j];
+      z = tmpOut->clip[2] = machine->Outputs[0].xyzw[2].f[j];
+      w = tmpOut->clip[3] = machine->Outputs[0].xyzw[3].f[j];
+
+      tmpOut->clipmask = compute_clipmask(tmpOut->clip, draw->plane,
+					   draw->nr_planes);
+      tmpOut->edgeflag = 1;
+
+      /* divide by w */
+      w = 1.0f / w;
+      x *= w;
+      y *= w;
+      z *= w;
+
+      /* Viewport mapping */
+      tmpOut->data[0][0] = x * scale[0] + trans[0];
+      tmpOut->data[0][1] = y * scale[1] + trans[1];
+      tmpOut->data[0][2] = z * scale[2] + trans[2];
+      tmpOut->data[0][3] = w;
+
+      /* Remaining attributes are packed into sequential post-transform
+       * vertex attrib slots.
+       */
+      for (slot = 1; slot < draw->num_vs_outputs; slot++) {
+         tmpOut->data[slot][0] = machine->Outputs[slot].xyzw[0].f[j];
+         tmpOut->data[slot][1] = machine->Outputs[slot].xyzw[1].f[j];
+         tmpOut->data[slot][2] = machine->Outputs[slot].xyzw[2].f[j];
+         tmpOut->data[slot][3] = machine->Outputs[slot].xyzw[3].f[j];
+      }
+
+      mfc_put(tmpOut, vOut[j], vert_size, TAG_VERTEX_BUFFER, 0, 0);
+   } /* loop over vertices */
+}
+
+
+static void
+spu_bind_vertex_shader(struct spu_vs_context *draw,
+		       void *uniforms,
+		       void *planes,
+		       unsigned nr_planes,
+		       unsigned num_outputs
+		       )
+{
+   draw->constants = (float (*)[4]) uniforms;
+
+   (void) memcpy(draw->plane, planes, sizeof(float) * 4 * nr_planes);
+   draw->nr_planes = nr_planes;
+   draw->num_vs_outputs = num_outputs;
+
+   /* specify the shader to interpret/execute */
+   spu_exec_machine_init(&draw->machine,
+			 PIPE_MAX_SAMPLERS,
+			 NULL /*samplers*/,
+			 PIPE_SHADER_VERTEX);
+}
+
+
+unsigned char immediates[(sizeof(float) * 4 * TGSI_EXEC_NUM_IMMEDIATES) + 32]
+    ALIGN16_ATTRIB;
+
+void
+spu_execute_vertex_shader(struct spu_vs_context *draw,
+                          const struct cell_command_vs *vs)
+{
+   unsigned i;
+
+   const uint64_t immediate_addr = vs->shader.immediates;
+   const unsigned immediate_size = 
+       ROUNDUP16((sizeof(float) * 4 * vs->shader.num_immediates)
+                 + (immediate_addr & 0x0f));
+
+   mfc_get(immediates, immediate_addr & ~0x0f, immediate_size,
+           TAG_VERTEX_BUFFER, 0, 0);
+
+   draw->machine.Instructions = (struct tgsi_full_instruction *)
+       vs->shader.instructions;
+   draw->machine.NumInstructions = vs->shader.num_instructions;
+
+   draw->machine.Declarations = (struct tgsi_full_declaration *)
+       vs->shader.declarations;
+   draw->machine.NumDeclarations = vs->shader.num_declarations;
+
+   draw->vertex_fetch.nr_attrs = vs->nr_attrs;
+
+   wait_on_mask(1 << TAG_VERTEX_BUFFER);
+
+   (void) memcpy(& draw->machine.Imms, &immediates[immediate_addr & 0x0f],
+                 sizeof(float) * 4 * vs->shader.num_immediates);
+
+   spu_bind_vertex_shader(draw, vs->shader.uniforms,
+                          vs->plane, vs->nr_planes,
+                          vs->shader.num_outputs);
+
+   for (i = 0; i < vs->num_elts; i += 4) {
+      const unsigned batch_size = MIN2(vs->num_elts - i, 4);
+
+      run_vertex_program(draw, & vs->elts[i], batch_size, &vs->vOut[i]);
+   }
+}
diff --git a/src/gallium/drivers/cell/spu/spu_vertex_shader.h b/src/gallium/drivers/cell/spu/spu_vertex_shader.h
new file mode 100644
index 0000000000..b5bf31e67d
--- /dev/null
+++ b/src/gallium/drivers/cell/spu/spu_vertex_shader.h
@@ -0,0 +1,63 @@
+#ifndef SPU_VERTEX_SHADER_H
+#define SPU_VERTEX_SHADER_H
+
+#include "pipe/p_format.h"
+#include "spu_exec.h"
+
+struct spu_vs_context;
+
+typedef void (*spu_fetch_func)(qword *out, const qword *in, qword defaults,
+			       const qword *shuffle_data);
+typedef void (*spu_full_fetch_func)( struct spu_vs_context *draw,
+				     struct spu_exec_machine *machine,
+				     const unsigned *elts,
+				     unsigned count );
+
+struct spu_vs_context {
+   struct pipe_viewport_state viewport;
+
+   struct {
+      uint64_t src_ptr[PIPE_ATTRIB_MAX];
+      unsigned pitch[PIPE_ATTRIB_MAX];
+      unsigned size[PIPE_ATTRIB_MAX];
+      enum pipe_format format[PIPE_ATTRIB_MAX];
+      unsigned nr_attrs;
+      boolean dirty;
+
+      spu_fetch_func fetch[PIPE_ATTRIB_MAX];
+      spu_full_fetch_func fetch_func;
+   } vertex_fetch;
+   
+   /* Clip derived state:
+    */
+   float plane[12][4];
+   unsigned nr_planes;
+
+   struct spu_exec_machine machine;
+   const float (*constants)[4];
+
+   unsigned num_vs_outputs;
+};
+
+extern void spu_update_vertex_fetch(struct spu_vs_context *draw);
+
+static INLINE void spu_vertex_fetch(struct spu_vs_context *draw,
+				    struct spu_exec_machine *machine,
+				    const unsigned *elts,
+				    unsigned count)
+{
+   if (draw->vertex_fetch.dirty) {
+      spu_update_vertex_fetch(draw);
+      draw->vertex_fetch.dirty = 0;
+   }
+   
+   (*draw->vertex_fetch.fetch_func)(draw, machine, elts, count);
+}
+
+struct cell_command_vs;
+
+extern void
+spu_execute_vertex_shader(struct spu_vs_context *draw,
+			  const struct cell_command_vs *vs);
+
+#endif /* SPU_VERTEX_SHADER_H */
diff --git a/src/gallium/drivers/cell/spu/spu_ztest.h b/src/gallium/drivers/cell/spu/spu_ztest.h
new file mode 100644
index 0000000000..ce8ad00339
--- /dev/null
+++ b/src/gallium/drivers/cell/spu/spu_ztest.h
@@ -0,0 +1,135 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+
+/**
+ * Zbuffer/depth test code.
+ */
+
+
+#ifndef SPU_ZTEST_H
+#define SPU_ZTEST_H
+
+
+#ifdef __SPU__
+#include <spu_intrinsics.h>
+#endif
+
+
+
+/**
+ * Perform Z testing for a 16-bit/value Z buffer.
+ *
+ * \param zvals  vector of four fragment zvalues as floats
+ * \param zbuf   ptr to vector of ushort[8] zbuffer values.  Note that this
+ *               contains the Z values for 2 quads, 8 pixels.
+ * \param x      x coordinate of quad (only lsbit is significant)
+ * \param inMask indicates which fragments in the quad are alive
+ * \return new mask indicating which fragments are alive after ztest
+ */
+static INLINE vector unsigned int
+spu_z16_test_less(vector float zvals, vector unsigned short *zbuf,
+                  uint x, vector unsigned int inMask)
+{
+#define ZERO 0x80
+   vector unsigned int zvals_ui4, zbuf_ui4, mask;
+
+   /* convert floats to uints in [0, 65535] */
+   zvals_ui4 = spu_convtu(zvals, 32); /* convert to [0, 2^32] */
+   zvals_ui4 = spu_rlmask(zvals_ui4, -16);  /* right shift 16 */
+
+   /* XXX this conditional could be removed with a bit of work */
+   if (x & 1) {
+      /* convert zbuffer values from ushorts to uints */
+      /* gather lower four ushorts */
+      zbuf_ui4 = spu_shuffle((vector unsigned int) *zbuf,
+                             (vector unsigned int) *zbuf,
+                             ((vector unsigned char) {
+                                ZERO, ZERO,  8,  9, ZERO, ZERO, 10, 11,
+                                ZERO, ZERO, 12, 13, ZERO, ZERO, 14, 15}));
+      /* mask = (zbuf_ui4 < zvals_ui4) ? ~0 : 0 */
+      mask = spu_cmpgt(zbuf_ui4, zvals_ui4);
+      /* mask &= inMask */
+      mask = spu_and(mask, inMask);
+      /* zbuf = mask ? zval : zbuf */
+      zbuf_ui4 = spu_sel(zbuf_ui4, zvals_ui4, mask);
+      /* convert zbuffer values from uints back to ushorts, preserve lower 4 */
+      *zbuf = (vector unsigned short)
+         spu_shuffle(zbuf_ui4, (vector unsigned int) *zbuf,
+                     ((vector unsigned char) {
+                        16, 17, 18, 19, 20, 21, 22, 23,
+                        2, 3, 6, 7, 10, 11, 14, 15}));
+   }
+   else {
+      /* convert zbuffer values from ushorts to uints */
+      /* gather upper four ushorts */
+      zbuf_ui4 = spu_shuffle((vector unsigned int) *zbuf,
+                             (vector unsigned int) *zbuf,
+                             ((vector unsigned char) {
+                                ZERO, ZERO, 0, 1, ZERO, ZERO, 2, 3,
+                                ZERO, ZERO, 4, 5, ZERO, ZERO, 6, 7}));
+      /* mask = (zbuf_ui4 < zvals_ui4) ? ~0 : 0 */
+      mask = spu_cmpgt(zbuf_ui4, zvals_ui4);
+      /* mask &= inMask */
+      mask = spu_and(mask, inMask);
+      /* zbuf = mask ? zval : zbuf */
+      zbuf_ui4 = spu_sel(zbuf_ui4, zvals_ui4, mask);
+      /* convert zbuffer values from uints back to ushorts, preserve upper 4 */
+      *zbuf = (vector unsigned short)
+         spu_shuffle(zbuf_ui4, (vector unsigned int) *zbuf,
+                     ((vector unsigned char) {
+                        2, 3, 6, 7, 10, 11, 14, 15,
+                        24, 25, 26, 27, 28, 29, 30, 31}));
+   }
+   return mask;
+#undef ZERO
+}
+
+
+/**
+ * As above, but Zbuffer values as 32-bit uints
+ */
+static INLINE vector unsigned int
+spu_z32_test_less(vector float zvals, vector unsigned int *zbuf_ptr,
+                  vector unsigned int inMask)
+{
+   vector unsigned int zvals_ui4, mask, zbuf = *zbuf_ptr;
+
+   /* convert floats to uints in [0, 0xffffffff] */
+   zvals_ui4 = spu_convtu(zvals, 32);
+   /* mask = (zbuf < zvals_ui4) ? ~0 : 0 */
+   mask = spu_cmpgt(zbuf, zvals_ui4);
+   /* mask &= inMask */
+   mask = spu_and(mask, inMask);
+   /* zbuf = mask ? zval : zbuf */
+   *zbuf_ptr = spu_sel(zbuf, zvals_ui4, mask);
+
+   return mask;
+}
+
+
+#endif /* SPU_ZTEST_H */
diff --git a/src/gallium/drivers/failover/Makefile b/src/gallium/drivers/failover/Makefile
new file mode 100644
index 0000000000..72d0895c74
--- /dev/null
+++ b/src/gallium/drivers/failover/Makefile
@@ -0,0 +1,21 @@
+
+TOP = ../../../..
+include $(TOP)/configs/current
+
+LIBNAME = failover
+
+DRIVER_SOURCES = \
+	fo_state.c \
+	fo_state_emit.c \
+	fo_context.c 
+
+C_SOURCES = \
+	$(COMMON_SOURCES) \
+	$(DRIVER_SOURCES)
+
+ASM_SOURCES = 
+
+include ../Makefile.template
+
+symlinks:
+
diff --git a/src/gallium/drivers/failover/fo_context.c b/src/gallium/drivers/failover/fo_context.c
new file mode 100644
index 0000000000..7ce4a7df17
--- /dev/null
+++ b/src/gallium/drivers/failover/fo_context.c
@@ -0,0 +1,155 @@
+/**************************************************************************
+ * 
+ * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+
+#include "pipe/p_defines.h"
+#include "pipe/p_winsys.h"
+#include "pipe/p_util.h"
+#include "pipe/p_context.h"
+
+#include "fo_context.h"
+#include "fo_winsys.h"
+
+
+
+static void failover_destroy( struct pipe_context *pipe )
+{
+   struct failover_context *failover = failover_context( pipe );
+
+   free( failover );
+}
+
+
+
+static boolean failover_draw_elements( struct pipe_context *pipe,
+				       struct pipe_buffer *indexBuffer,
+				       unsigned indexSize,
+				       unsigned prim, unsigned start, unsigned count)
+{
+   struct failover_context *failover = failover_context( pipe );
+
+   /* If there has been any statechange since last time, try hardware
+    * rendering again:
+    */
+   if (failover->dirty) {
+      failover->mode = FO_HW;
+   }
+
+   /* Try hardware:
+    */
+   if (failover->mode == FO_HW) {
+      if (!failover->hw->draw_elements( failover->hw, 
+					indexBuffer, 
+					indexSize, 
+					prim, 
+					start, 
+					count )) {
+
+	 failover->hw->flush( failover->hw, ~0 );
+	 failover->mode = FO_SW;
+      }
+   }
+
+   /* Possibly try software:
+    */
+   if (failover->mode == FO_SW) {
+
+      if (failover->dirty) 
+	 failover_state_emit( failover );
+
+      failover->sw->draw_elements( failover->sw, 
+				   indexBuffer, 
+				   indexSize, 
+				   prim, 
+				   start, 
+				   count );
+
+      /* Be ready to switch back to hardware rendering without an
+       * intervening flush.  Unlikely to be much performance impact to
+       * this:
+       */
+      failover->sw->flush( failover->sw, ~0 );
+   }
+
+   return TRUE;
+}
+
+
+static boolean failover_draw_arrays( struct pipe_context *pipe,
+				     unsigned prim, unsigned start, unsigned count)
+{
+   return failover_draw_elements(pipe, NULL, 0, prim, start, count);
+}
+
+
+
+struct pipe_context *failover_create( struct pipe_context *hw,
+				      struct pipe_context *sw )
+{
+   struct failover_context *failover = CALLOC_STRUCT(failover_context);
+   if (failover == NULL)
+      return NULL;
+
+   failover->hw = hw;
+   failover->sw = sw;
+   failover->pipe.winsys = hw->winsys;
+   failover->pipe.destroy = failover_destroy;
+   failover->pipe.is_format_supported = hw->is_format_supported;
+   failover->pipe.get_name = hw->get_name;
+   failover->pipe.get_vendor = hw->get_vendor;
+   failover->pipe.get_param = hw->get_param;
+   failover->pipe.get_paramf = hw->get_paramf;
+
+   failover->pipe.draw_arrays = failover_draw_arrays;
+   failover->pipe.draw_elements = failover_draw_elements;
+   failover->pipe.clear = hw->clear;
+
+   /* No software occlusion fallback (or other optional functionality)
+    * at this point - if the hardware doesn't support it, don't
+    * advertise it to the application.
+    */
+   failover->pipe.begin_query = hw->begin_query;
+   failover->pipe.end_query = hw->end_query;
+
+   failover_init_state_functions( failover );
+
+#if 0
+   failover->pipe.surface_alloc = hw->surface_alloc;
+#endif
+   failover->pipe.get_tex_surface = hw->get_tex_surface;
+
+   failover->pipe.surface_copy = hw->surface_copy;
+   failover->pipe.surface_fill = hw->surface_fill;
+   failover->pipe.texture_create = hw->texture_create;
+   failover->pipe.texture_release = hw->texture_release;
+   failover->pipe.flush = hw->flush;
+
+   failover->dirty = 0;
+
+   return &failover->pipe;
+}
+
diff --git a/src/gallium/drivers/failover/fo_context.h b/src/gallium/drivers/failover/fo_context.h
new file mode 100644
index 0000000000..1dc87291c9
--- /dev/null
+++ b/src/gallium/drivers/failover/fo_context.h
@@ -0,0 +1,114 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+/* Authors:  Keith Whitwell <keith@tungstengraphics.com>
+ */
+
+#ifndef FO_CONTEXT_H
+#define FO_CONTEXT_H
+
+#include "pipe/p_state.h"
+#include "pipe/p_context.h"
+
+
+
+#define FO_NEW_VIEWPORT        0x1
+#define FO_NEW_RASTERIZER      0x2
+#define FO_NEW_FRAGMENT_SHADER 0x4
+#define FO_NEW_BLEND           0x8
+#define FO_NEW_CLIP            0x10
+#define FO_NEW_SCISSOR         0x20
+#define FO_NEW_STIPPLE         0x40
+#define FO_NEW_FRAMEBUFFER     0x80
+#define FO_NEW_ALPHA_TEST      0x100
+#define FO_NEW_DEPTH_STENCIL   0x200
+#define FO_NEW_SAMPLER         0x400
+#define FO_NEW_TEXTURE         0x800
+#define FO_NEW_VERTEX          0x2000
+#define FO_NEW_VERTEX_SHADER   0x4000
+#define FO_NEW_BLEND_COLOR     0x8000
+#define FO_NEW_CLEAR_COLOR     0x10000
+#define FO_NEW_VERTEX_BUFFER   0x20000
+#define FO_NEW_VERTEX_ELEMENT  0x40000
+
+
+
+#define FO_HW 0
+#define FO_SW 1
+
+struct fo_state {
+   void *sw_state;
+   void *hw_state;
+};
+struct failover_context {
+   struct pipe_context pipe;  /**< base class */
+
+
+   /* The most recent drawing state as set by the driver:
+    */
+   const struct fo_state     *blend;
+   const struct fo_state     *sampler[PIPE_MAX_SAMPLERS];
+   const struct fo_state     *depth_stencil;
+   const struct fo_state     *rasterizer;
+   const struct fo_state     *fragment_shader;
+   const struct fo_state     *vertex_shader;
+
+   struct pipe_blend_color blend_color;
+   struct pipe_clip_state clip;
+   struct pipe_framebuffer_state framebuffer;
+   struct pipe_poly_stipple poly_stipple;
+   struct pipe_scissor_state scissor;
+   struct pipe_texture *texture[PIPE_MAX_SAMPLERS];
+   struct pipe_viewport_state viewport;
+   struct pipe_vertex_buffer vertex_buffer[PIPE_ATTRIB_MAX];
+   struct pipe_vertex_element vertex_element[PIPE_ATTRIB_MAX];
+
+   unsigned dirty;
+   unsigned dirty_sampler;
+   unsigned dirty_texture;
+   unsigned dirty_vertex_buffer;
+   unsigned dirty_vertex_element;
+
+
+   unsigned mode;
+   struct pipe_context *hw;
+   struct pipe_context *sw;
+};
+
+
+
+void failover_init_state_functions( struct failover_context *failover );
+void failover_state_emit( struct failover_context *failover );
+
+static INLINE struct failover_context *
+failover_context( struct pipe_context *pipe )
+{
+   return (struct failover_context *)pipe;
+}
+
+
+#endif /* FO_CONTEXT_H */
diff --git a/src/gallium/drivers/failover/fo_state.c b/src/gallium/drivers/failover/fo_state.c
new file mode 100644
index 0000000000..0fc5568da1
--- /dev/null
+++ b/src/gallium/drivers/failover/fo_state.c
@@ -0,0 +1,457 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+/* Authors:  Keith Whitwell <keith@tungstengraphics.com>
+ */
+
+#include "fo_context.h"
+
+
+/* This looks like a lot of work at the moment - we're keeping a
+ * duplicate copy of the state up-to-date.  
+ *
+ * This can change in two ways:
+ * - With constant state objects we would only need to save a pointer,
+ *     not the whole object.
+ * - By adding a callback in the state tracker to re-emit state.  The
+ *     state tracker knows the current state already and can re-emit it 
+ *     without additional complexity.
+ *
+ * This works as a proof-of-concept, but a final version will have
+ * lower overheads.
+ */
+
+
+
+static void *
+failover_create_blend_state( struct pipe_context *pipe,
+                             const struct pipe_blend_state *blend )
+{
+   struct fo_state *state = malloc(sizeof(struct fo_state));
+   struct failover_context *failover = failover_context(pipe);
+
+   state->sw_state = failover->sw->create_blend_state(failover->sw, blend);
+   state->hw_state = failover->hw->create_blend_state(failover->hw, blend);
+
+   return state;
+}
+
+static void
+failover_bind_blend_state( struct pipe_context *pipe,
+                           void *blend )
+{
+   struct failover_context *failover = failover_context(pipe);
+   struct fo_state *state = (struct fo_state *)blend;
+   failover->blend = state;
+   failover->dirty |= FO_NEW_BLEND;
+   failover->sw->bind_blend_state( failover->sw, state->sw_state );
+   failover->hw->bind_blend_state( failover->hw, state->hw_state );
+}
+
+static void
+failover_delete_blend_state( struct pipe_context *pipe,
+                             void *blend )
+{
+   struct fo_state *state = (struct fo_state*)blend;
+   struct failover_context *failover = failover_context(pipe);
+
+   failover->sw->delete_blend_state(failover->sw, state->sw_state);
+   failover->hw->delete_blend_state(failover->hw, state->hw_state);
+   state->sw_state = 0;
+   state->hw_state = 0;
+   free(state);
+}
+
+static void
+failover_set_blend_color( struct pipe_context *pipe,
+			  const struct pipe_blend_color *blend_color )
+{
+   struct failover_context *failover = failover_context(pipe);
+
+   failover->blend_color = *blend_color;
+   failover->dirty |= FO_NEW_BLEND_COLOR;
+   failover->sw->set_blend_color( failover->sw, blend_color );
+   failover->hw->set_blend_color( failover->hw, blend_color );
+}
+
+static void 
+failover_set_clip_state( struct pipe_context *pipe,
+			 const struct pipe_clip_state *clip )
+{
+   struct failover_context *failover = failover_context(pipe);
+
+   failover->clip = *clip;
+   failover->dirty |= FO_NEW_CLIP;
+   failover->sw->set_clip_state( failover->sw, clip );
+   failover->hw->set_clip_state( failover->hw, clip );
+}
+
+
+static void *
+failover_create_depth_stencil_state(struct pipe_context *pipe,
+                              const struct pipe_depth_stencil_alpha_state *templ)
+{
+   struct fo_state *state = malloc(sizeof(struct fo_state));
+   struct failover_context *failover = failover_context(pipe);
+
+   state->sw_state = failover->sw->create_depth_stencil_alpha_state(failover->sw, templ);
+   state->hw_state = failover->hw->create_depth_stencil_alpha_state(failover->hw, templ);
+
+   return state;
+}
+
+static void
+failover_bind_depth_stencil_state(struct pipe_context *pipe,
+                                  void *depth_stencil)
+{
+   struct failover_context *failover = failover_context(pipe);
+   struct fo_state *state = (struct fo_state *)depth_stencil;
+   failover->depth_stencil = state;
+   failover->dirty |= FO_NEW_DEPTH_STENCIL;
+   failover->sw->bind_depth_stencil_alpha_state(failover->sw, state->sw_state);
+   failover->hw->bind_depth_stencil_alpha_state(failover->hw, state->hw_state);
+}
+
+static void
+failover_delete_depth_stencil_state(struct pipe_context *pipe,
+                                    void *ds)
+{
+   struct fo_state *state = (struct fo_state*)ds;
+   struct failover_context *failover = failover_context(pipe);
+
+   failover->sw->delete_depth_stencil_alpha_state(failover->sw, state->sw_state);
+   failover->hw->delete_depth_stencil_alpha_state(failover->hw, state->hw_state);
+   state->sw_state = 0;
+   state->hw_state = 0;
+   free(state);
+}
+
+static void
+failover_set_framebuffer_state(struct pipe_context *pipe,
+			       const struct pipe_framebuffer_state *framebuffer)
+{
+   struct failover_context *failover = failover_context(pipe);
+
+   failover->framebuffer = *framebuffer;
+   failover->dirty |= FO_NEW_FRAMEBUFFER;
+   failover->sw->set_framebuffer_state( failover->sw, framebuffer );
+   failover->hw->set_framebuffer_state( failover->hw, framebuffer );
+}
+
+
+static void *
+failover_create_fs_state(struct pipe_context *pipe,
+                         const struct pipe_shader_state *templ)
+{
+   struct fo_state *state = malloc(sizeof(struct fo_state));
+   struct failover_context *failover = failover_context(pipe);
+
+   state->sw_state = failover->sw->create_fs_state(failover->sw, templ);
+   state->hw_state = failover->hw->create_fs_state(failover->hw, templ);
+
+   return state;
+}
+
+static void
+failover_bind_fs_state(struct pipe_context *pipe, void *fs)
+{
+   struct failover_context *failover = failover_context(pipe);
+   struct fo_state *state = (struct fo_state*)fs;
+   failover->fragment_shader = state;
+   failover->dirty |= FO_NEW_FRAGMENT_SHADER;
+   failover->sw->bind_fs_state(failover->sw, state->sw_state);
+   failover->hw->bind_fs_state(failover->hw, state->hw_state);
+}
+
+static void
+failover_delete_fs_state(struct pipe_context *pipe,
+                         void *fs)
+{
+   struct fo_state *state = (struct fo_state*)fs;
+   struct failover_context *failover = failover_context(pipe);
+
+   failover->sw->delete_fs_state(failover->sw, state->sw_state);
+   failover->hw->delete_fs_state(failover->hw, state->hw_state);
+   state->sw_state = 0;
+   state->hw_state = 0;
+   free(state);
+}
+
+static void *
+failover_create_vs_state(struct pipe_context *pipe,
+                         const struct pipe_shader_state *templ)
+{
+   struct fo_state *state = malloc(sizeof(struct fo_state));
+   struct failover_context *failover = failover_context(pipe);
+
+   state->sw_state = failover->sw->create_vs_state(failover->sw, templ);
+   state->hw_state = failover->hw->create_vs_state(failover->hw, templ);
+
+   return state;
+}
+
+static void
+failover_bind_vs_state(struct pipe_context *pipe,
+                       void *vs)
+{
+   struct failover_context *failover = failover_context(pipe);
+
+   struct fo_state *state = (struct fo_state*)vs;
+   failover->vertex_shader = state;
+   failover->dirty |= FO_NEW_VERTEX_SHADER;
+   failover->sw->bind_vs_state(failover->sw, state->sw_state);
+   failover->hw->bind_vs_state(failover->hw, state->hw_state);
+}
+
+static void
+failover_delete_vs_state(struct pipe_context *pipe,
+                         void *vs)
+{
+   struct fo_state *state = (struct fo_state*)vs;
+   struct failover_context *failover = failover_context(pipe);
+
+   failover->sw->delete_vs_state(failover->sw, state->sw_state);
+   failover->hw->delete_vs_state(failover->hw, state->hw_state);
+   state->sw_state = 0;
+   state->hw_state = 0;
+   free(state);
+}
+
+static void 
+failover_set_polygon_stipple( struct pipe_context *pipe,
+			      const struct pipe_poly_stipple *stipple )
+{
+   struct failover_context *failover = failover_context(pipe);
+
+   failover->poly_stipple = *stipple;
+   failover->dirty |= FO_NEW_STIPPLE;
+   failover->sw->set_polygon_stipple( failover->sw, stipple );
+   failover->hw->set_polygon_stipple( failover->hw, stipple );
+}
+
+
+static void *
+failover_create_rasterizer_state(struct pipe_context *pipe,
+                                 const struct pipe_rasterizer_state *templ)
+{
+   struct fo_state *state = malloc(sizeof(struct fo_state));
+   struct failover_context *failover = failover_context(pipe);
+
+   state->sw_state = failover->sw->create_rasterizer_state(failover->sw, templ);
+   state->hw_state = failover->hw->create_rasterizer_state(failover->hw, templ);
+
+   return state;
+}
+
+static void
+failover_bind_rasterizer_state(struct pipe_context *pipe,
+                               void *raster)
+{
+   struct failover_context *failover = failover_context(pipe);
+
+   struct fo_state *state = (struct fo_state*)raster;
+   failover->rasterizer = state;
+   failover->dirty |= FO_NEW_RASTERIZER;
+   failover->sw->bind_rasterizer_state(failover->sw, state->sw_state);
+   failover->hw->bind_rasterizer_state(failover->hw, state->hw_state);
+}
+
+static void
+failover_delete_rasterizer_state(struct pipe_context *pipe,
+                                 void *raster)
+{
+   struct fo_state *state = (struct fo_state*)raster;
+   struct failover_context *failover = failover_context(pipe);
+
+   failover->sw->delete_rasterizer_state(failover->sw, state->sw_state);
+   failover->hw->delete_rasterizer_state(failover->hw, state->hw_state);
+   state->sw_state = 0;
+   state->hw_state = 0;
+   free(state);
+}
+
+
+static void 
+failover_set_scissor_state( struct pipe_context *pipe,
+                                 const struct pipe_scissor_state *scissor )
+{
+   struct failover_context *failover = failover_context(pipe);
+
+   failover->scissor = *scissor;
+   failover->dirty |= FO_NEW_SCISSOR;
+   failover->sw->set_scissor_state( failover->sw, scissor );
+   failover->hw->set_scissor_state( failover->hw, scissor );
+}
+
+
+static void *
+failover_create_sampler_state(struct pipe_context *pipe,
+                              const struct pipe_sampler_state *templ)
+{
+   struct fo_state *state = malloc(sizeof(struct fo_state));
+   struct failover_context *failover = failover_context(pipe);
+
+   state->sw_state = failover->sw->create_sampler_state(failover->sw, templ);
+   state->hw_state = failover->hw->create_sampler_state(failover->hw, templ);
+
+   return state;
+}
+
+static void
+failover_bind_sampler_state(struct pipe_context *pipe,
+			   unsigned unit, void *sampler)
+{
+   struct failover_context *failover = failover_context(pipe);
+   struct fo_state *state = (struct fo_state*)sampler;
+   failover->sampler[unit] = state;
+   failover->dirty |= FO_NEW_SAMPLER;
+   failover->dirty_sampler |= (1<<unit);
+   failover->sw->bind_sampler_state(failover->sw, unit,
+                                    state->sw_state);
+   failover->hw->bind_sampler_state(failover->hw, unit,
+                                    state->hw_state);
+}
+
+static void
+failover_delete_sampler_state(struct pipe_context *pipe, void *sampler)
+{
+   struct fo_state *state = (struct fo_state*)sampler;
+   struct failover_context *failover = failover_context(pipe);
+
+   failover->sw->delete_sampler_state(failover->sw, state->sw_state);
+   failover->hw->delete_sampler_state(failover->hw, state->hw_state);
+   state->sw_state = 0;
+   state->hw_state = 0;
+   free(state);
+}
+
+
+static void
+failover_set_sampler_texture(struct pipe_context *pipe,
+			     unsigned unit,
+			     struct pipe_texture *texture)
+{
+   struct failover_context *failover = failover_context(pipe);
+
+   failover->texture[unit] = texture;
+   failover->dirty |= FO_NEW_TEXTURE;
+   failover->dirty_texture |= (1<<unit);
+   failover->sw->set_sampler_texture( failover->sw, unit, texture );
+   failover->hw->set_sampler_texture( failover->hw, unit, texture );
+}
+
+
+static void 
+failover_set_viewport_state( struct pipe_context *pipe,
+			     const struct pipe_viewport_state *viewport )
+{
+   struct failover_context *failover = failover_context(pipe);
+
+   failover->viewport = *viewport; 
+   failover->dirty |= FO_NEW_VIEWPORT;
+   failover->sw->set_viewport_state( failover->sw, viewport );
+   failover->hw->set_viewport_state( failover->hw, viewport );
+}
+
+
+static void
+failover_set_vertex_buffer(struct pipe_context *pipe,
+			   unsigned unit,
+                           const struct pipe_vertex_buffer *vertex_buffer)
+{
+   struct failover_context *failover = failover_context(pipe);
+
+   failover->vertex_buffer[unit] = *vertex_buffer;
+   failover->dirty |= FO_NEW_VERTEX_BUFFER;
+   failover->dirty_vertex_buffer |= (1<<unit);
+   failover->sw->set_vertex_buffer( failover->sw, unit, vertex_buffer );
+   failover->hw->set_vertex_buffer( failover->hw, unit, vertex_buffer );
+}
+
+
+static void
+failover_set_vertex_element(struct pipe_context *pipe,
+			    unsigned unit,
+			    const struct pipe_vertex_element *vertex_element)
+{
+   struct failover_context *failover = failover_context(pipe);
+
+   failover->vertex_element[unit] = *vertex_element;
+   failover->dirty |= FO_NEW_VERTEX_ELEMENT;
+   failover->dirty_vertex_element |= (1<<unit);
+   failover->sw->set_vertex_element( failover->sw, unit, vertex_element );
+   failover->hw->set_vertex_element( failover->hw, unit, vertex_element );
+}
+
+void
+failover_set_constant_buffer(struct pipe_context *pipe,
+                             uint shader, uint index,
+                             const struct pipe_constant_buffer *buf)
+{
+   struct failover_context *failover = failover_context(pipe);
+
+   assert(shader < PIPE_SHADER_TYPES);
+   assert(index == 0);
+
+   failover->sw->set_constant_buffer(failover->sw, shader, index, buf);
+   failover->hw->set_constant_buffer(failover->hw, shader, index, buf);
+}
+
+
+void
+failover_init_state_functions( struct failover_context *failover )
+{
+   failover->pipe.create_blend_state = failover_create_blend_state;
+   failover->pipe.bind_blend_state   = failover_bind_blend_state;
+   failover->pipe.delete_blend_state = failover_delete_blend_state;
+   failover->pipe.create_sampler_state = failover_create_sampler_state;
+   failover->pipe.bind_sampler_state   = failover_bind_sampler_state;
+   failover->pipe.delete_sampler_state = failover_delete_sampler_state;
+   failover->pipe.create_depth_stencil_alpha_state = failover_create_depth_stencil_state;
+   failover->pipe.bind_depth_stencil_alpha_state   = failover_bind_depth_stencil_state;
+   failover->pipe.delete_depth_stencil_alpha_state = failover_delete_depth_stencil_state;
+   failover->pipe.create_rasterizer_state = failover_create_rasterizer_state;
+   failover->pipe.bind_rasterizer_state = failover_bind_rasterizer_state;
+   failover->pipe.delete_rasterizer_state = failover_delete_rasterizer_state;
+   failover->pipe.create_fs_state = failover_create_fs_state;
+   failover->pipe.bind_fs_state   = failover_bind_fs_state;
+   failover->pipe.delete_fs_state = failover_delete_fs_state;
+   failover->pipe.create_vs_state = failover_create_vs_state;
+   failover->pipe.bind_vs_state   = failover_bind_vs_state;
+   failover->pipe.delete_vs_state = failover_delete_vs_state;
+
+   failover->pipe.set_blend_color = failover_set_blend_color;
+   failover->pipe.set_clip_state = failover_set_clip_state;
+   failover->pipe.set_framebuffer_state = failover_set_framebuffer_state;
+   failover->pipe.set_polygon_stipple = failover_set_polygon_stipple;
+   failover->pipe.set_scissor_state = failover_set_scissor_state;
+   failover->pipe.set_sampler_texture = failover_set_sampler_texture;
+   failover->pipe.set_viewport_state = failover_set_viewport_state;
+   failover->pipe.set_vertex_buffer = failover_set_vertex_buffer;
+   failover->pipe.set_vertex_element = failover_set_vertex_element;
+   failover->pipe.set_constant_buffer = failover_set_constant_buffer;
+}
diff --git a/src/gallium/drivers/failover/fo_state_emit.c b/src/gallium/drivers/failover/fo_state_emit.c
new file mode 100644
index 0000000000..c663dd4947
--- /dev/null
+++ b/src/gallium/drivers/failover/fo_state_emit.c
@@ -0,0 +1,137 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+/* Authors:  Keith Whitwell <keith@tungstengraphics.com>
+ */
+
+#include "fo_context.h"
+
+/* This looks like a lot of work at the moment - we're keeping a
+ * duplicate copy of the state up-to-date.  
+ *
+ * This can change in two ways:
+ * - With constant state objects we would only need to save a pointer,
+ *     not the whole object.
+ * - By adding a callback in the state tracker to re-emit state.  The
+ *     state tracker knows the current state already and can re-emit it 
+ *     without additional complexity.
+ *
+ * This works as a proof-of-concept, but a final version will have
+ * lower overheads.
+ */
+
+
+/* Bring the software pipe uptodate with current state.
+ * 
+ * With constant state objects we would probably just send all state
+ * to both rasterizers all the time???
+ */
+void
+failover_state_emit( struct failover_context *failover )
+{
+   unsigned i;
+
+   if (failover->dirty & FO_NEW_BLEND)
+      failover->sw->bind_blend_state( failover->sw,
+                                      failover->blend->sw_state );
+
+   if (failover->dirty & FO_NEW_BLEND_COLOR)
+      failover->sw->set_blend_color( failover->sw, &failover->blend_color );
+
+   if (failover->dirty & FO_NEW_CLIP)
+      failover->sw->set_clip_state( failover->sw, &failover->clip );
+
+   if (failover->dirty & FO_NEW_DEPTH_STENCIL)
+      failover->sw->bind_depth_stencil_alpha_state( failover->sw,
+						    failover->depth_stencil->sw_state );
+
+   if (failover->dirty & FO_NEW_FRAMEBUFFER)
+      failover->sw->set_framebuffer_state( failover->sw, &failover->framebuffer );
+
+   if (failover->dirty & FO_NEW_FRAGMENT_SHADER)
+      failover->sw->bind_fs_state( failover->sw,
+                                   failover->fragment_shader->sw_state );
+
+   if (failover->dirty & FO_NEW_VERTEX_SHADER)
+      failover->sw->bind_vs_state( failover->sw,
+                                   failover->vertex_shader->sw_state );
+
+   if (failover->dirty & FO_NEW_STIPPLE)
+      failover->sw->set_polygon_stipple( failover->sw, &failover->poly_stipple );
+
+   if (failover->dirty & FO_NEW_RASTERIZER)
+      failover->sw->bind_rasterizer_state( failover->sw,
+                                           failover->rasterizer->sw_state );
+
+   if (failover->dirty & FO_NEW_SCISSOR)
+      failover->sw->set_scissor_state( failover->sw, &failover->scissor );
+
+   if (failover->dirty & FO_NEW_VIEWPORT)
+      failover->sw->set_viewport_state( failover->sw, &failover->viewport );
+
+   if (failover->dirty & FO_NEW_SAMPLER) {
+      for (i = 0; i < PIPE_MAX_SAMPLERS; i++) {
+	 if (failover->dirty_sampler & (1<<i)) {
+	    failover->sw->bind_sampler_state( failover->sw, i,
+                                              failover->sampler[i]->sw_state );
+	 }
+      }
+   }
+
+   if (failover->dirty & FO_NEW_TEXTURE) {
+      for (i = 0; i < PIPE_MAX_SAMPLERS; i++) {
+	 if (failover->dirty_texture & (1<<i)) {
+	    failover->sw->set_sampler_texture( failover->sw, i, 
+					     failover->texture[i] );
+	 }
+      }
+   }
+
+   if (failover->dirty & FO_NEW_VERTEX_BUFFER) {
+      for (i = 0; i < PIPE_ATTRIB_MAX; i++) {
+	 if (failover->dirty_vertex_buffer & (1<<i)) {
+	    failover->sw->set_vertex_buffer( failover->sw, i, 
+					     &failover->vertex_buffer[i] );
+	 }
+      }
+   }
+
+   if (failover->dirty & FO_NEW_VERTEX_ELEMENT) {
+      for (i = 0; i < PIPE_ATTRIB_MAX; i++) {
+	 if (failover->dirty_vertex_element & (1<<i)) {
+	    failover->sw->set_vertex_element( failover->sw, i, 
+					      &failover->vertex_element[i] );
+	 }
+      }
+   }
+
+   failover->dirty = 0;
+   failover->dirty_vertex_element = 0;
+   failover->dirty_vertex_buffer = 0;
+   failover->dirty_texture = 0;
+   failover->dirty_sampler = 0;
+}
diff --git a/src/gallium/drivers/failover/fo_winsys.h b/src/gallium/drivers/failover/fo_winsys.h
new file mode 100644
index 0000000000..a8ce997a1f
--- /dev/null
+++ b/src/gallium/drivers/failover/fo_winsys.h
@@ -0,0 +1,45 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+#ifndef FO_WINSYS_H
+#define FO_WINSYS_H
+
+
+/* This is the interface that failover requires any window system
+ * hosting it to implement.  This is the only include file in failover
+ * which is public.
+ */
+
+
+struct pipe_context;
+
+
+struct pipe_context *failover_create( struct pipe_context *hw,
+				      struct pipe_context *sw );
+
+
+#endif /* FO_WINSYS_H */
diff --git a/src/gallium/drivers/i915simple/Makefile b/src/gallium/drivers/i915simple/Makefile
new file mode 100644
index 0000000000..2f91de3afc
--- /dev/null
+++ b/src/gallium/drivers/i915simple/Makefile
@@ -0,0 +1,38 @@
+
+TOP = ../../../..
+include $(TOP)/configs/current
+
+LIBNAME = i915simple
+
+DRIVER_SOURCES = \
+	i915_blit.c \
+	i915_clear.c \
+	i915_flush.c \
+	i915_context.c \
+	i915_context.c \
+	i915_debug.c \
+	i915_debug_fp.c \
+	i915_state.c \
+	i915_state_immediate.c \
+	i915_state_dynamic.c \
+	i915_state_derived.c \
+	i915_state_emit.c \
+	i915_state_sampler.c \
+	i915_strings.c \
+	i915_prim_emit.c \
+	i915_prim_vbuf.c \
+	i915_texture.c \
+	i915_fpc_emit.c \
+	i915_fpc_translate.c \
+	i915_surface.c 
+
+C_SOURCES = \
+	$(COMMON_SOURCES) \
+	$(DRIVER_SOURCES)
+
+ASM_SOURCES = 
+
+include ../Makefile.template
+
+symlinks:
+
diff --git a/src/gallium/drivers/i915simple/SConscript b/src/gallium/drivers/i915simple/SConscript
new file mode 100644
index 0000000000..f5fb96b995
--- /dev/null
+++ b/src/gallium/drivers/i915simple/SConscript
@@ -0,0 +1,29 @@
+Import('*')
+
+env = env.Clone()
+
+i915simple = env.ConvenienceLibrary(
+	target = 'i915simple',
+	source = [
+		'i915_blit.c',
+		'i915_clear.c',
+		'i915_context.c',
+		'i915_debug.c',
+		'i915_debug_fp.c',
+		'i915_flush.c',
+		'i915_fpc_emit.c',
+		'i915_fpc_translate.c',
+		'i915_prim_emit.c',
+		'i915_prim_vbuf.c',
+		'i915_state.c',
+		'i915_state_derived.c',
+		'i915_state_dynamic.c',
+		'i915_state_emit.c',
+		'i915_state_immediate.c',
+		'i915_state_sampler.c',
+		'i915_strings.c',
+		'i915_surface.c',
+		'i915_texture.c',
+	])
+
+Export('i915simple')
diff --git a/src/gallium/drivers/i915simple/i915_batch.h b/src/gallium/drivers/i915simple/i915_batch.h
new file mode 100644
index 0000000000..fb88cd6db0
--- /dev/null
+++ b/src/gallium/drivers/i915simple/i915_batch.h
@@ -0,0 +1,54 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+#ifndef I915_BATCH_H
+#define I915_BATCH_H
+
+#include "i915_winsys.h"
+#include "i915_debug.h"
+
+#define BATCH_LOCALS
+
+#define BEGIN_BATCH( dwords, relocs ) \
+   (i915->batch_start = i915->winsys->batch_start( i915->winsys, dwords, relocs ))
+
+#define OUT_BATCH( dword ) \
+   i915->winsys->batch_dword( i915->winsys, dword )
+
+#define OUT_RELOC( buf, flags, delta ) \
+   i915->winsys->batch_reloc( i915->winsys, buf, flags, delta )
+
+#define ADVANCE_BATCH()
+
+#define FLUSH_BATCH() do { 				\
+   if (0) i915_dump_batchbuffer( i915 );		\
+   i915->winsys->batch_flush( i915->winsys );		\
+   i915->batch_start = NULL;				\
+   i915->hardware_dirty = ~0;				\
+} while (0)
+
+#endif 
diff --git a/src/gallium/drivers/i915simple/i915_blit.c b/src/gallium/drivers/i915simple/i915_blit.c
new file mode 100644
index 0000000000..db4671ff55
--- /dev/null
+++ b/src/gallium/drivers/i915simple/i915_blit.c
@@ -0,0 +1,162 @@
+/**************************************************************************
+ * 
+ * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+
+#include "i915_context.h"
+#include "i915_winsys.h"
+#include "i915_blit.h"
+#include "i915_reg.h"
+#include "i915_batch.h"
+
+#define FILE_DEBUG_FLAG DEBUG_BLIT
+
+void
+i915_fill_blit(struct i915_context *i915,
+	       unsigned cpp,
+	       short dst_pitch,
+	       struct pipe_buffer *dst_buffer,
+	       unsigned dst_offset,
+	       short x, short y, 
+	       short w, short h, 
+	       unsigned color)
+{
+   unsigned BR13, CMD;
+   BATCH_LOCALS;
+
+   dst_pitch *= (short) cpp;
+
+   switch (cpp) {
+   case 1:
+   case 2:
+   case 3:
+      BR13 = dst_pitch | (0xF0 << 16) | (1 << 24);
+      CMD = XY_COLOR_BLT_CMD;
+      break;
+   case 4:
+      BR13 = dst_pitch | (0xF0 << 16) | (1 << 24) | (1 << 25);
+      CMD = (XY_COLOR_BLT_CMD | XY_COLOR_BLT_WRITE_ALPHA |
+             XY_COLOR_BLT_WRITE_RGB);
+      break;
+   default:
+      return;
+   }
+
+//   DBG("%s dst:buf(%p)/%d+%d %d,%d sz:%dx%d\n",
+//       __FUNCTION__, dst_buffer, dst_pitch, dst_offset, x, y, w, h);
+
+
+   if (!BEGIN_BATCH(6, 1)) {
+      FLUSH_BATCH();
+      assert(BEGIN_BATCH(6, 1));
+   }
+   OUT_BATCH(CMD);
+   OUT_BATCH(BR13);
+   OUT_BATCH((y << 16) | x);
+   OUT_BATCH(((y + h) << 16) | (x + w));
+   OUT_RELOC( dst_buffer, I915_BUFFER_ACCESS_WRITE, dst_offset);
+   OUT_BATCH(color);
+   ADVANCE_BATCH();
+}
+
+
+void
+i915_copy_blit( struct i915_context *i915,
+                  unsigned do_flip,
+                  unsigned cpp,
+                  short src_pitch,
+                  struct pipe_buffer *src_buffer,
+                  unsigned src_offset,
+                  short dst_pitch,
+                  struct pipe_buffer *dst_buffer,
+                  unsigned dst_offset,
+                  short src_x, short src_y,
+                  short dst_x, short dst_y, 
+		  short w, short h )
+{
+   unsigned CMD, BR13;
+   int dst_y2 = dst_y + h;
+   int dst_x2 = dst_x + w;
+   BATCH_LOCALS;
+
+
+   I915_DBG(i915,
+       "%s src:buf(%p)/%d+%d %d,%d dst:buf(%p)/%d+%d %d,%d sz:%dx%d\n",
+       __FUNCTION__,
+       src_buffer, src_pitch, src_offset, src_x, src_y,
+       dst_buffer, dst_pitch, dst_offset, dst_x, dst_y, w, h);
+
+   src_pitch *= (short) cpp;
+   dst_pitch *= (short) cpp;
+
+   switch (cpp) {
+   case 1:
+   case 2:
+   case 3:
+      BR13 = (((int) dst_pitch) & 0xffff) | 
+	 (0xCC << 16) | (1 << 24);
+      CMD = XY_SRC_COPY_BLT_CMD;
+      break;
+   case 4:
+      BR13 =
+         (((int) dst_pitch) & 0xffff) | 
+	 (0xCC << 16) | (1 << 24) | (1 << 25);
+      CMD =
+         (XY_SRC_COPY_BLT_CMD | XY_SRC_COPY_BLT_WRITE_ALPHA |
+          XY_SRC_COPY_BLT_WRITE_RGB);
+      break;
+   default:
+      return;
+   }
+
+   if (dst_y2 < dst_y || 
+       dst_x2 < dst_x) {
+      return;
+   }
+
+   /* Hardware can handle negative pitches but loses the ability to do
+    * proper overlapping blits in that case.  We don't really have a
+    * need for either at this stage.
+    */
+   assert (dst_pitch > 0 && src_pitch > 0);
+
+
+   if (!BEGIN_BATCH(8, 2)) {
+      FLUSH_BATCH();
+      assert(BEGIN_BATCH(8, 2));
+   }
+   OUT_BATCH(CMD);
+   OUT_BATCH(BR13);
+   OUT_BATCH((dst_y << 16) | dst_x);
+   OUT_BATCH((dst_y2 << 16) | dst_x2);
+   OUT_RELOC(dst_buffer, I915_BUFFER_ACCESS_WRITE, dst_offset);
+   OUT_BATCH((src_y << 16) | src_x);
+   OUT_BATCH(((int) src_pitch & 0xffff));
+   OUT_RELOC(src_buffer, I915_BUFFER_ACCESS_READ, src_offset);
+   ADVANCE_BATCH();
+}
+
+
diff --git a/src/gallium/drivers/i915simple/i915_blit.h b/src/gallium/drivers/i915simple/i915_blit.h
new file mode 100644
index 0000000000..6e5b44e124
--- /dev/null
+++ b/src/gallium/drivers/i915simple/i915_blit.h
@@ -0,0 +1,55 @@
+/**************************************************************************
+ * 
+ * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+#ifndef I915_BLIT_H
+#define I915_BLIT_H
+
+#include "i915_context.h"
+
+extern void i915_copy_blit(struct i915_context *i915,
+                           unsigned do_flip,
+			   unsigned cpp,
+			   short src_pitch,
+			   struct pipe_buffer *src_buffer,
+			   unsigned src_offset,
+			   short dst_pitch,
+			   struct pipe_buffer *dst_buffer,
+			   unsigned dst_offset,
+			   short srcx, short srcy,
+			   short dstx, short dsty,
+			   short w, short h );
+
+extern void i915_fill_blit(struct i915_context *i915,
+			   unsigned cpp,
+			   short dst_pitch,
+			   struct pipe_buffer *dst_buffer,
+			   unsigned dst_offset,
+			   short x, short y,
+			   short w, short h, unsigned color);
+
+
+#endif
diff --git a/src/gallium/drivers/i915simple/i915_clear.c b/src/gallium/drivers/i915simple/i915_clear.c
new file mode 100644
index 0000000000..cde69daacc
--- /dev/null
+++ b/src/gallium/drivers/i915simple/i915_clear.c
@@ -0,0 +1,47 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+/* Author:
+ *    Brian Paul
+ */
+
+
+#include "pipe/p_defines.h"
+#include "i915_context.h"
+#include "i915_state.h"
+
+
+/**
+ * Clear the given surface to the specified value.
+ * No masking, no scissor (clear entire buffer).
+ */
+void
+i915_clear(struct pipe_context *pipe, struct pipe_surface *ps,
+	   unsigned clearValue)
+{
+   pipe->surface_fill(pipe, ps, 0, 0, ps->width, ps->height, clearValue);
+}
diff --git a/src/gallium/drivers/i915simple/i915_context.c b/src/gallium/drivers/i915simple/i915_context.c
new file mode 100644
index 0000000000..497623a700
--- /dev/null
+++ b/src/gallium/drivers/i915simple/i915_context.c
@@ -0,0 +1,320 @@
+/**************************************************************************
+ * 
+ * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+#include "i915_context.h"
+#include "i915_winsys.h"
+#include "i915_state.h"
+#include "i915_batch.h"
+#include "i915_texture.h"
+#include "i915_reg.h"
+
+#include "pipe/draw/draw_context.h"
+#include "pipe/p_defines.h"
+#include "pipe/p_winsys.h"
+#include "pipe/p_util.h"
+
+
+/**
+ * Query format support for creating a texture, drawing surface, etc.
+ * \param format  the format to test
+ * \param type  one of PIPE_TEXTURE, PIPE_SURFACE
+ */
+static boolean
+i915_is_format_supported( struct pipe_context *pipe,
+                          enum pipe_format format, uint type )
+{
+   static const enum pipe_format tex_supported[] = {
+      PIPE_FORMAT_R8G8B8A8_UNORM,
+      PIPE_FORMAT_A8R8G8B8_UNORM,
+      PIPE_FORMAT_R5G6B5_UNORM,
+      PIPE_FORMAT_U_L8,
+      PIPE_FORMAT_U_A8,
+      PIPE_FORMAT_U_I8,
+      PIPE_FORMAT_U_A8_L8,
+      PIPE_FORMAT_YCBCR,
+      PIPE_FORMAT_YCBCR_REV,
+      PIPE_FORMAT_S8Z24_UNORM,
+      PIPE_FORMAT_NONE  /* list terminator */
+   };
+   static const enum pipe_format surface_supported[] = {
+      PIPE_FORMAT_A8R8G8B8_UNORM,
+      PIPE_FORMAT_R5G6B5_UNORM,
+      PIPE_FORMAT_S8Z24_UNORM,
+      /*PIPE_FORMAT_R16G16B16A16_SNORM,*/
+      PIPE_FORMAT_NONE  /* list terminator */
+   };
+   const enum pipe_format *list;
+   uint i;
+
+   switch (type) {
+   case PIPE_TEXTURE:
+      list = tex_supported;
+      break;
+   case PIPE_SURFACE:
+      list = surface_supported;
+      break;
+   default:
+      assert(0);
+   }
+
+   for (i = 0; list[i] != PIPE_FORMAT_NONE; i++) {
+      if (list[i] == format)
+         return TRUE;
+   }
+
+   return FALSE;
+}
+
+
+static int
+i915_get_param(struct pipe_context *pipe, int param)
+{
+   switch (param) {
+   case PIPE_CAP_MAX_TEXTURE_IMAGE_UNITS:
+      return 8;
+   case PIPE_CAP_NPOT_TEXTURES:
+      return 1;
+   case PIPE_CAP_TWO_SIDED_STENCIL:
+      return 1;
+   case PIPE_CAP_GLSL:
+      return 0;
+   case PIPE_CAP_S3TC:
+      return 0;
+   case PIPE_CAP_ANISOTROPIC_FILTER:
+      return 0;
+   case PIPE_CAP_POINT_SPRITE:
+      return 0;
+   case PIPE_CAP_MAX_RENDER_TARGETS:
+      return 1;
+   case PIPE_CAP_OCCLUSION_QUERY:
+      return 0;
+   case PIPE_CAP_TEXTURE_SHADOW_MAP:
+      return 1;
+   case PIPE_CAP_MAX_TEXTURE_2D_LEVELS:
+      return 11; /* max 1024x1024 */
+   case PIPE_CAP_MAX_TEXTURE_3D_LEVELS:
+      return 8;  /* max 128x128x128 */
+   case PIPE_CAP_MAX_TEXTURE_CUBE_LEVELS:
+      return 11; /* max 1024x1024 */
+   default:
+      return 0;
+   }
+}
+
+
+static float
+i915_get_paramf(struct pipe_context *pipe, int param)
+{
+   switch (param) {
+   case PIPE_CAP_MAX_LINE_WIDTH:
+      /* fall-through */
+   case PIPE_CAP_MAX_LINE_WIDTH_AA:
+      return 7.5;
+
+   case PIPE_CAP_MAX_POINT_WIDTH:
+      /* fall-through */
+   case PIPE_CAP_MAX_POINT_WIDTH_AA:
+      return 255.0;
+
+   case PIPE_CAP_MAX_TEXTURE_ANISOTROPY:
+      return 4.0;
+
+   case PIPE_CAP_MAX_TEXTURE_LOD_BIAS:
+      return 16.0;
+
+   default:
+      return 0;
+   }
+}
+
+
+static void i915_destroy( struct pipe_context *pipe )
+{
+   struct i915_context *i915 = i915_context( pipe );
+
+   draw_destroy( i915->draw );
+
+   FREE( i915 );
+}
+
+
+
+
+static boolean
+i915_draw_elements( struct pipe_context *pipe,
+                    struct pipe_buffer *indexBuffer,
+                    unsigned indexSize,
+                    unsigned prim, unsigned start, unsigned count)
+{
+   struct i915_context *i915 = i915_context( pipe );
+   struct draw_context *draw = i915->draw;
+   unsigned i;
+
+   if (i915->dirty)
+      i915_update_derived( i915 );
+
+   /*
+    * Map vertex buffers
+    */
+   for (i = 0; i < PIPE_ATTRIB_MAX; i++) {
+      if (i915->vertex_buffer[i].buffer) {
+         void *buf
+            = pipe->winsys->buffer_map(pipe->winsys,
+                                       i915->vertex_buffer[i].buffer,
+                                       PIPE_BUFFER_USAGE_CPU_READ);
+         draw_set_mapped_vertex_buffer(draw, i, buf);
+      }
+   }
+   /* Map index buffer, if present */
+   if (indexBuffer) {
+      void *mapped_indexes
+         = pipe->winsys->buffer_map(pipe->winsys, indexBuffer,
+                                    PIPE_BUFFER_USAGE_CPU_READ);
+      draw_set_mapped_element_buffer(draw, indexSize, mapped_indexes);
+   }
+   else {
+      /* no index/element buffer */
+      draw_set_mapped_element_buffer(draw, 0, NULL);
+   }
+
+
+   draw_set_mapped_constant_buffer(draw,
+                                i915->current.constants[PIPE_SHADER_VERTEX]);
+
+   /* draw! */
+   draw_arrays(i915->draw, prim, start, count);
+
+   /*
+    * unmap vertex/index buffers
+    */
+   for (i = 0; i < PIPE_ATTRIB_MAX; i++) {
+      if (i915->vertex_buffer[i].buffer) {
+         pipe->winsys->buffer_unmap(pipe->winsys, i915->vertex_buffer[i].buffer);
+         draw_set_mapped_vertex_buffer(draw, i, NULL);
+      }
+   }
+   if (indexBuffer) {
+      pipe->winsys->buffer_unmap(pipe->winsys, indexBuffer);
+      draw_set_mapped_element_buffer(draw, 0, NULL);
+   }
+
+   return TRUE;
+}
+
+
+static boolean i915_draw_arrays( struct pipe_context *pipe,
+				 unsigned prim, unsigned start, unsigned count)
+{
+   return i915_draw_elements(pipe, NULL, 0, prim, start, count);
+}
+
+
+
+struct pipe_context *i915_create( struct pipe_winsys *pipe_winsys,
+				  struct i915_winsys *i915_winsys,
+				  unsigned pci_id )
+{
+   struct i915_context *i915;
+   unsigned is_i945 = 0;
+
+   switch (pci_id) {
+   case PCI_CHIP_I915_G:
+   case PCI_CHIP_I915_GM:
+      break;
+
+   case PCI_CHIP_I945_G:
+   case PCI_CHIP_I945_GM:
+   case PCI_CHIP_I945_GME:
+   case PCI_CHIP_G33_G:
+   case PCI_CHIP_Q33_G:
+   case PCI_CHIP_Q35_G:
+      is_i945 = 1;
+      break;
+
+   default:
+      pipe_winsys->printf(pipe_winsys, 
+			  "%s: unknown pci id 0x%x, cannot create context\n", 
+			  __FUNCTION__, pci_id);
+      return NULL;
+   }
+
+   i915 = CALLOC_STRUCT(i915_context);
+   if (i915 == NULL)
+      return NULL;
+
+   i915->winsys = i915_winsys;
+   i915->pipe.winsys = pipe_winsys;
+
+   i915->pipe.destroy = i915_destroy;
+   i915->pipe.is_format_supported = i915_is_format_supported;
+   i915->pipe.get_param = i915_get_param;
+   i915->pipe.get_paramf = i915_get_paramf;
+
+   i915->pipe.clear = i915_clear;
+
+
+   i915->pipe.draw_arrays = i915_draw_arrays;
+   i915->pipe.draw_elements = i915_draw_elements;
+
+   /*
+    * Create drawing context and plug our rendering stage into it.
+    */
+   i915->draw = draw_create();
+   assert(i915->draw);
+   if (GETENV("I915_VBUF")) {
+      draw_set_rasterize_stage(i915->draw, i915_draw_vbuf_stage(i915));
+   }
+   else {
+      draw_set_rasterize_stage(i915->draw, i915_draw_render_stage(i915));
+   }
+
+   i915_init_surface_functions(i915);
+   i915_init_state_functions(i915);
+   i915_init_flush_functions(i915);
+   i915_init_string_functions(i915);
+
+   i915->pci_id = pci_id;
+   i915->flags.is_i945 = is_i945;
+
+   i915->pipe.texture_create = i915_texture_create;
+   i915->pipe.texture_release = i915_texture_release;
+
+   i915->dirty = ~0;
+   i915->hardware_dirty = ~0;
+
+   /* Batch stream debugging is a bit hacked up at the moment:
+    */
+   i915->batch_start = NULL;
+
+   /*
+    * XXX we could plug GL selection/feedback into the drawing pipeline
+    * by specifying a different setup/render stage.
+    */
+
+   return &i915->pipe;
+}
+
diff --git a/src/gallium/drivers/i915simple/i915_context.h b/src/gallium/drivers/i915simple/i915_context.h
new file mode 100644
index 0000000000..b4ea63c3e7
--- /dev/null
+++ b/src/gallium/drivers/i915simple/i915_context.h
@@ -0,0 +1,304 @@
+ /**************************************************************************
+ * 
+ * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+#ifndef I915_CONTEXT_H
+#define I915_CONTEXT_H
+
+
+#include "pipe/p_context.h"
+#include "pipe/p_defines.h"
+#include "pipe/p_state.h"
+
+#include "pipe/draw/draw_vertex.h"
+
+
+#define I915_TEX_UNITS 8
+
+#define I915_DYNAMIC_MODES4       0
+#define I915_DYNAMIC_DEPTHSCALE_0 1 /* just the header */
+#define I915_DYNAMIC_DEPTHSCALE_1 2 
+#define I915_DYNAMIC_IAB          3
+#define I915_DYNAMIC_BC_0         4 /* just the header */
+#define I915_DYNAMIC_BC_1         5
+#define I915_DYNAMIC_BFO_0        6 
+#define I915_DYNAMIC_BFO_1        7
+#define I915_DYNAMIC_STP_0        8 
+#define I915_DYNAMIC_STP_1        9 
+#define I915_DYNAMIC_SC_ENA_0     10 
+#define I915_DYNAMIC_SC_RECT_0    11 
+#define I915_DYNAMIC_SC_RECT_1    12 
+#define I915_DYNAMIC_SC_RECT_2    13 
+#define I915_MAX_DYNAMIC          14
+
+
+#define I915_IMMEDIATE_S0         0
+#define I915_IMMEDIATE_S1         1
+#define I915_IMMEDIATE_S2         2
+#define I915_IMMEDIATE_S3         3
+#define I915_IMMEDIATE_S4         4
+#define I915_IMMEDIATE_S5         5
+#define I915_IMMEDIATE_S6         6
+#define I915_IMMEDIATE_S7         7
+#define I915_MAX_IMMEDIATE        8
+
+/* These must mach the order of LI0_STATE_* bits, as they will be used
+ * to generate hardware packets:
+ */
+#define I915_CACHE_STATIC         0 
+#define I915_CACHE_DYNAMIC        1 /* handled specially */
+#define I915_CACHE_SAMPLER        2
+#define I915_CACHE_MAP            3
+#define I915_CACHE_PROGRAM        4
+#define I915_CACHE_CONSTANTS      5
+#define I915_MAX_CACHE            6
+
+#define I915_MAX_CONSTANT  32
+
+
+
+struct i915_cache_context;
+
+/* Use to calculate differences between state emitted to hardware and
+ * current driver-calculated state.  
+ */
+struct i915_state 
+{
+   unsigned immediate[I915_MAX_IMMEDIATE];
+   unsigned dynamic[I915_MAX_DYNAMIC];
+
+   float constants[PIPE_SHADER_TYPES][I915_MAX_CONSTANT][4];
+   /** number of constants passed in through a constant buffer */
+   uint num_user_constants[PIPE_SHADER_TYPES];
+   /** user constants, plus extra constants from shader translation */
+   uint num_constants[PIPE_SHADER_TYPES];
+
+   uint *program;
+   uint program_len;
+
+   /* texture sampler state */
+   unsigned sampler[I915_TEX_UNITS][3];
+   unsigned sampler_enable_flags;
+   unsigned sampler_enable_nr;
+
+   /* texture image buffers */
+   unsigned texbuffer[I915_TEX_UNITS][2];
+
+   /** Describes the current hardware vertex layout */
+   struct vertex_info vertex_info;
+   
+   unsigned id;			/* track lost context events */
+};
+
+struct i915_blend_state {
+   unsigned iab;
+   unsigned modes4;
+   unsigned LIS5;
+   unsigned LIS6;
+};
+
+struct i915_depth_stencil_state {
+   unsigned stencil_modes4;
+   unsigned bfo[2];
+   unsigned stencil_LIS5;
+   unsigned depth_LIS6;
+};
+
+struct i915_rasterizer_state {
+   int light_twoside : 1;
+   unsigned st;
+   enum interp_mode color_interp;
+
+   unsigned LIS4;
+   unsigned LIS7;
+   unsigned sc[1];
+
+   const struct pipe_rasterizer_state *templ;
+
+   union { float f; unsigned u; } ds[2];
+};
+
+struct i915_sampler_state {
+   unsigned state[3];
+   const struct pipe_sampler_state *templ;
+};
+
+
+struct i915_texture {
+   struct pipe_texture base;
+
+   /* Derived from the above:
+    */
+   unsigned pitch;
+   unsigned depth_pitch;          /* per-image on i945? */
+   unsigned total_height;
+
+   unsigned nr_images[PIPE_MAX_TEXTURE_LEVELS];
+
+   /* Explicitly store the offset of each image for each cube face or
+    * depth value.  Pretty much have to accept that hardware formats
+    * are going to be so diverse that there is no unified way to
+    * compute the offsets of depth/cube images within a mipmap level,
+    * so have to store them as a lookup table:
+    */
+   unsigned *image_offset[PIPE_MAX_TEXTURE_LEVELS];   /**< array [depth] of offsets */
+
+   /* Includes image offset tables:
+    */
+   unsigned level_offset[PIPE_MAX_TEXTURE_LEVELS];
+
+   /* The data is held here:
+    */
+   struct pipe_buffer *buffer;
+};
+
+struct i915_context
+{
+   struct pipe_context pipe;
+   struct i915_winsys *winsys;
+   struct draw_context *draw;
+
+   /* The most recent drawing state as set by the driver:
+    */
+   const struct i915_blend_state           *blend;
+   const struct i915_sampler_state         *sampler[PIPE_MAX_SAMPLERS];
+   const struct i915_depth_stencil_state   *depth_stencil;
+   const struct i915_rasterizer_state      *rasterizer;
+   const struct pipe_shader_state *fs;
+
+   struct pipe_blend_color blend_color;
+   struct pipe_clip_state clip;
+   struct pipe_constant_buffer constants[PIPE_SHADER_TYPES];
+   struct pipe_framebuffer_state framebuffer;
+   struct pipe_poly_stipple poly_stipple;
+   struct pipe_scissor_state scissor;
+   struct i915_texture *texture[PIPE_MAX_SAMPLERS];
+   struct pipe_viewport_state viewport;
+   struct pipe_vertex_buffer vertex_buffer[PIPE_ATTRIB_MAX];
+
+   unsigned dirty;
+
+   unsigned *batch_start;
+
+   /** Vertex buffer */
+   struct pipe_buffer *vbo;
+
+   struct i915_state current;
+   unsigned hardware_dirty;
+   
+   unsigned debug;
+   unsigned pci_id;
+
+   struct {
+      unsigned is_i945:1;
+   } flags;
+};
+
+/* A flag for each state_tracker state object:
+ */
+#define I915_NEW_VIEWPORT      0x1
+#define I915_NEW_RASTERIZER    0x2
+#define I915_NEW_FS            0x4
+#define I915_NEW_BLEND         0x8
+#define I915_NEW_CLIP          0x10
+#define I915_NEW_SCISSOR       0x20
+#define I915_NEW_STIPPLE       0x40
+#define I915_NEW_FRAMEBUFFER   0x80
+#define I915_NEW_ALPHA_TEST    0x100
+#define I915_NEW_DEPTH_STENCIL 0x200
+#define I915_NEW_SAMPLER       0x400
+#define I915_NEW_TEXTURE       0x800
+#define I915_NEW_CONSTANTS     0x1000
+#define I915_NEW_VBO           0x2000
+
+
+/* Driver's internally generated state flags:
+ */
+#define I915_NEW_VERTEX_FORMAT    0x10000
+
+
+/* Dirty flags for hardware emit
+ */
+#define I915_HW_STATIC            (1<<I915_CACHE_STATIC)
+#define I915_HW_DYNAMIC           (1<<I915_CACHE_DYNAMIC)
+#define I915_HW_SAMPLER           (1<<I915_CACHE_SAMPLER)
+#define I915_HW_MAP               (1<<I915_CACHE_MAP)
+#define I915_HW_PROGRAM           (1<<I915_CACHE_PROGRAM)
+#define I915_HW_CONSTANTS         (1<<I915_CACHE_CONSTANTS)
+#define I915_HW_IMMEDIATE         (1<<(I915_MAX_CACHE+0))
+#define I915_HW_INVARIENT         (1<<(I915_MAX_CACHE+1))
+
+
+/***********************************************************************
+ * i915_prim_emit.c: 
+ */
+struct draw_stage *i915_draw_render_stage( struct i915_context *i915 );
+
+
+/***********************************************************************
+ * i915_prim_vbuf.c: 
+ */
+struct draw_stage *i915_draw_vbuf_stage( struct i915_context *i915 );
+
+
+/***********************************************************************
+ * i915_state_emit.c: 
+ */
+void i915_emit_hardware_state(struct i915_context *i915 );
+
+
+
+/***********************************************************************
+ * i915_clear.c: 
+ */
+void i915_clear(struct pipe_context *pipe, struct pipe_surface *ps,
+		unsigned clearValue);
+
+
+/***********************************************************************
+ * i915_surface.c: 
+ */
+void i915_init_surface_functions( struct i915_context *i915 );
+
+void i915_init_state_functions( struct i915_context *i915 );
+void i915_init_flush_functions( struct i915_context *i915 );
+void i915_init_string_functions( struct i915_context *i915 );
+
+
+
+/***********************************************************************
+ * Inline conversion functions.  These are better-typed than the
+ * macros used previously:
+ */
+static INLINE struct i915_context *
+i915_context( struct pipe_context *pipe )
+{
+   return (struct i915_context *)pipe;
+}
+
+
+
+#endif
diff --git a/src/gallium/drivers/i915simple/i915_debug.c b/src/gallium/drivers/i915simple/i915_debug.c
new file mode 100644
index 0000000000..94db44e1aa
--- /dev/null
+++ b/src/gallium/drivers/i915simple/i915_debug.c
@@ -0,0 +1,901 @@
+/**************************************************************************
+ * 
+ * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+//#include "imports.h"
+
+#include "i915_reg.h"
+#include "i915_context.h"
+#include "i915_winsys.h"
+#include "i915_debug.h"
+#include "pipe/p_winsys.h"
+
+
+static void
+PRINTF(
+   struct debug_stream  *stream,
+   const char           *fmt,
+                        ... )
+{
+   va_list  args;
+   char     buffer[256];
+
+   va_start( args, fmt );
+   vsprintf( buffer, fmt, args );
+   stream->winsys->printf( stream->winsys, buffer );
+   va_end( args );
+}
+
+
+static boolean debug( struct debug_stream *stream, const char *name, unsigned len )
+{
+   unsigned i;
+   unsigned *ptr = (unsigned *)(stream->ptr + stream->offset);
+   
+   if (len == 0) {
+      PRINTF(stream, "Error - zero length packet (0x%08x)\n", stream->ptr[0]);
+      assert(0);
+      return FALSE;
+   }
+
+   if (stream->print_addresses)
+      PRINTF(stream, "%08x:  ", stream->offset);
+
+
+   PRINTF(stream, "%s (%d dwords):\n", name, len);
+   for (i = 0; i < len; i++)
+      PRINTF(stream, "\t0x%08x\n",  ptr[i]);   
+   PRINTF(stream, "\n");
+
+   stream->offset += len * sizeof(unsigned);
+   
+   return TRUE;
+}
+
+
+static const char *get_prim_name( unsigned val )
+{
+   switch (val & PRIM3D_MASK) {
+   case PRIM3D_TRILIST: return "TRILIST"; break;
+   case PRIM3D_TRISTRIP: return "TRISTRIP"; break;
+   case PRIM3D_TRISTRIP_RVRSE: return "TRISTRIP_RVRSE"; break;
+   case PRIM3D_TRIFAN: return "TRIFAN"; break;
+   case PRIM3D_POLY: return "POLY"; break;
+   case PRIM3D_LINELIST: return "LINELIST"; break;
+   case PRIM3D_LINESTRIP: return "LINESTRIP"; break;
+   case PRIM3D_RECTLIST: return "RECTLIST"; break;
+   case PRIM3D_POINTLIST: return "POINTLIST"; break;
+   case PRIM3D_DIB: return "DIB"; break;
+   case PRIM3D_CLEAR_RECT: return "CLEAR_RECT"; break;
+   case PRIM3D_ZONE_INIT: return "ZONE_INIT"; break;
+   default: return "????"; break;
+   }
+}
+
+static boolean debug_prim( struct debug_stream *stream, const char *name, 
+			     boolean dump_floats,
+			     unsigned len )
+{
+   unsigned *ptr = (unsigned *)(stream->ptr + stream->offset);
+   const char *prim = get_prim_name( ptr[0] );
+   unsigned i;
+   
+
+
+   PRINTF(stream, "%s %s (%d dwords):\n", name, prim, len);
+   PRINTF(stream, "\t0x%08x\n",  ptr[0]);   
+   for (i = 1; i < len; i++) {
+      if (dump_floats)
+	 PRINTF(stream, "\t0x%08x // %f\n",  ptr[i], *(float *)&ptr[i]);   
+      else
+	 PRINTF(stream, "\t0x%08x\n",  ptr[i]);   
+   }
+
+      
+   PRINTF(stream, "\n");
+
+   stream->offset += len * sizeof(unsigned);
+   
+   return TRUE;
+}
+   
+
+
+
+static boolean debug_program( struct debug_stream *stream, const char *name, unsigned len )
+{
+   unsigned *ptr = (unsigned *)(stream->ptr + stream->offset);
+
+   if (len == 0) {
+      PRINTF(stream, "Error - zero length packet (0x%08x)\n", stream->ptr[0]);
+      assert(0);
+      return FALSE;
+   }
+
+   if (stream->print_addresses)
+      PRINTF(stream, "%08x:  ", stream->offset);
+
+   PRINTF(stream, "%s (%d dwords):\n", name, len);
+   i915_disassemble_program( stream, ptr, len );
+
+   stream->offset += len * sizeof(unsigned);
+   return TRUE;
+}
+
+
+static boolean debug_chain( struct debug_stream *stream, const char *name, unsigned len )
+{
+   unsigned *ptr = (unsigned *)(stream->ptr + stream->offset);
+   unsigned old_offset = stream->offset + len * sizeof(unsigned);
+   unsigned i;
+
+   PRINTF(stream, "%s (%d dwords):\n", name, len);
+   for (i = 0; i < len; i++)
+      PRINTF(stream, "\t0x%08x\n",  ptr[i]);
+
+   stream->offset = ptr[1] & ~0x3;
+   
+   if (stream->offset < old_offset)
+      PRINTF(stream, "\n... skipping backwards from 0x%x --> 0x%x ...\n\n", 
+		   old_offset, stream->offset );
+   else
+      PRINTF(stream, "\n... skipping from 0x%x --> 0x%x ...\n\n", 
+		   old_offset, stream->offset );
+
+
+   return TRUE;
+}
+
+
+static boolean debug_variable_length_prim( struct debug_stream *stream )
+{
+   unsigned *ptr = (unsigned *)(stream->ptr + stream->offset);
+   const char *prim = get_prim_name( ptr[0] );
+   unsigned i, len;
+
+   ushort *idx = (ushort *)(ptr+1);
+   for (i = 0; idx[i] != 0xffff; i++)
+      ;
+
+   len = 1+(i+2)/2;
+
+   PRINTF(stream, "3DPRIM, %s variable length %d indicies (%d dwords):\n", prim, i, len);
+   for (i = 0; i < len; i++)
+      PRINTF(stream, "\t0x%08x\n",  ptr[i]);
+   PRINTF(stream, "\n");
+
+   stream->offset += len * sizeof(unsigned);
+   return TRUE;
+}
+
+
+static void
+BITS(
+   struct debug_stream  *stream,
+   unsigned             dw,
+   unsigned             hi,
+   unsigned             lo,
+   const char           *fmt,
+                        ... )
+{
+   va_list  args;
+   char     buffer[256];
+   unsigned himask = ~0UL >> (31 - (hi));
+
+   PRINTF(stream, "\t\t ");
+
+   va_start( args, fmt );
+   vsprintf( buffer, fmt, args );
+   stream->winsys->printf( stream->winsys, buffer );
+   va_end( args );
+
+   PRINTF(stream, ": 0x%x\n", ((dw) & himask) >> (lo));
+}
+
+#define MBZ( dw, hi, lo) do {							\
+   unsigned x = (dw) >> (lo);				\
+   unsigned lomask = (1 << (lo)) - 1;			\
+   unsigned himask;					\
+   himask = (1UL << (hi)) - 1;				\
+   assert ((x & himask & ~lomask) == 0);	\
+} while (0)
+
+static void
+FLAG(
+   struct debug_stream  *stream,
+   unsigned             dw,
+   unsigned             bit,
+   const char           *fmt,
+                        ... )
+{
+   if (((dw) >> (bit)) & 1) {
+      va_list  args;
+      char     buffer[256];
+
+      PRINTF(stream, "\t\t ");
+
+      va_start( args, fmt );
+      vsprintf( buffer, fmt, args );
+      stream->winsys->printf( stream->winsys, buffer );
+      va_end( args );
+
+      PRINTF(stream, "\n");
+   }
+}
+
+static boolean debug_load_immediate( struct debug_stream *stream,
+				       const char *name,
+				       unsigned len )
+{
+   unsigned *ptr = (unsigned *)(stream->ptr + stream->offset);
+   unsigned bits = (ptr[0] >> 4) & 0xff;
+   unsigned j = 0;
+   
+   PRINTF(stream, "%s (%d dwords, flags: %x):\n", name, len, bits);
+   PRINTF(stream, "\t0x%08x\n",  ptr[j++]);
+
+   if (bits & (1<<0)) {
+      PRINTF(stream, "\t  LIS0: 0x%08x\n", ptr[j]);
+      PRINTF(stream, "\t vb address: 0x%08x\n", (ptr[j] & ~0x3));
+      BITS(stream, ptr[j], 0, 0, "vb invalidate disable");
+      j++;
+   }
+   if (bits & (1<<1)) {
+      PRINTF(stream, "\t  LIS1: 0x%08x\n", ptr[j]);
+      BITS(stream, ptr[j], 29, 24, "vb dword width");
+      BITS(stream, ptr[j], 21, 16, "vb dword pitch");
+      BITS(stream, ptr[j], 15, 0, "vb max index");
+      j++;
+   }
+   if (bits & (1<<2)) {
+      int i;
+      PRINTF(stream, "\t  LIS2: 0x%08x\n", ptr[j]);
+      for (i = 0; i < 8; i++) {
+	 unsigned tc = (ptr[j] >> (i * 4)) & 0xf;
+	 if (tc != 0xf)
+	    BITS(stream, tc, 3, 0, "tex coord %d", i);
+      }
+      j++;
+   }
+   if (bits & (1<<3)) {
+      PRINTF(stream, "\t  LIS3: 0x%08x\n", ptr[j]);
+      j++;
+   }
+   if (bits & (1<<4)) {
+      PRINTF(stream, "\t  LIS4: 0x%08x\n", ptr[j]);
+      BITS(stream, ptr[j], 31, 23, "point width");
+      BITS(stream, ptr[j], 22, 19, "line width");
+      FLAG(stream, ptr[j], 18, "alpha flatshade");
+      FLAG(stream, ptr[j], 17, "fog flatshade");
+      FLAG(stream, ptr[j], 16, "spec flatshade");
+      FLAG(stream, ptr[j], 15, "rgb flatshade");
+      BITS(stream, ptr[j], 14, 13, "cull mode");
+      FLAG(stream, ptr[j], 12, "vfmt: point width");
+      FLAG(stream, ptr[j], 11, "vfmt: specular/fog");
+      FLAG(stream, ptr[j], 10, "vfmt: rgba");
+      FLAG(stream, ptr[j], 9, "vfmt: depth offset");
+      BITS(stream, ptr[j], 8, 6, "vfmt: position (2==xyzw)");
+      FLAG(stream, ptr[j], 5, "force dflt diffuse");
+      FLAG(stream, ptr[j], 4, "force dflt specular");
+      FLAG(stream, ptr[j], 3, "local depth offset enable");
+      FLAG(stream, ptr[j], 2, "vfmt: fp32 fog coord");
+      FLAG(stream, ptr[j], 1, "sprite point");
+      FLAG(stream, ptr[j], 0, "antialiasing");
+      j++;
+   }
+   if (bits & (1<<5)) {
+      PRINTF(stream, "\t  LIS5: 0x%08x\n", ptr[j]);
+      BITS(stream, ptr[j], 31, 28, "rgba write disables");
+      FLAG(stream, ptr[j], 27,     "force dflt point width");
+      FLAG(stream, ptr[j], 26,     "last pixel enable");
+      FLAG(stream, ptr[j], 25,     "global z offset enable");
+      FLAG(stream, ptr[j], 24,     "fog enable");
+      BITS(stream, ptr[j], 23, 16, "stencil ref");
+      BITS(stream, ptr[j], 15, 13, "stencil test");
+      BITS(stream, ptr[j], 12, 10, "stencil fail op");
+      BITS(stream, ptr[j], 9, 7,   "stencil pass z fail op");
+      BITS(stream, ptr[j], 6, 4,   "stencil pass z pass op");
+      FLAG(stream, ptr[j], 3,      "stencil write enable");
+      FLAG(stream, ptr[j], 2,      "stencil test enable");
+      FLAG(stream, ptr[j], 1,      "color dither enable");
+      FLAG(stream, ptr[j], 0,      "logiop enable");
+      j++;
+   }
+   if (bits & (1<<6)) {
+      PRINTF(stream, "\t  LIS6: 0x%08x\n", ptr[j]);
+      FLAG(stream, ptr[j], 31,      "alpha test enable");
+      BITS(stream, ptr[j], 30, 28,  "alpha func");
+      BITS(stream, ptr[j], 27, 20,  "alpha ref");
+      FLAG(stream, ptr[j], 19,      "depth test enable");
+      BITS(stream, ptr[j], 18, 16,  "depth func");
+      FLAG(stream, ptr[j], 15,      "blend enable");
+      BITS(stream, ptr[j], 14, 12,  "blend func");
+      BITS(stream, ptr[j], 11, 8,   "blend src factor");
+      BITS(stream, ptr[j], 7,  4,   "blend dst factor");
+      FLAG(stream, ptr[j], 3,       "depth write enable");
+      FLAG(stream, ptr[j], 2,       "color write enable");
+      BITS(stream, ptr[j], 1,  0,   "provoking vertex"); 
+      j++;
+   }
+
+
+   PRINTF(stream, "\n");
+
+   assert(j == len);
+
+   stream->offset += len * sizeof(unsigned);
+   
+   return TRUE;
+}
+ 
+
+
+static boolean debug_load_indirect( struct debug_stream *stream,
+				      const char *name,
+				      unsigned len )
+{
+   unsigned *ptr = (unsigned *)(stream->ptr + stream->offset);
+   unsigned bits = (ptr[0] >> 8) & 0x3f;
+   unsigned i, j = 0;
+   
+   PRINTF(stream, "%s (%d dwords):\n", name, len);
+   PRINTF(stream, "\t0x%08x\n",  ptr[j++]);
+
+   for (i = 0; i < 6; i++) {
+      if (bits & (1<<i)) {
+	 switch (1<<(8+i)) {
+	 case LI0_STATE_STATIC_INDIRECT:
+	    PRINTF(stream, "        STATIC: 0x%08x | %x\n", ptr[j]&~3, ptr[j]&3); j++;
+	    PRINTF(stream, "                0x%08x\n", ptr[j++]);
+	    break;
+	 case LI0_STATE_DYNAMIC_INDIRECT:
+	    PRINTF(stream, "       DYNAMIC: 0x%08x | %x\n", ptr[j]&~3, ptr[j]&3); j++;
+	    break;
+	 case LI0_STATE_SAMPLER:
+	    PRINTF(stream, "       SAMPLER: 0x%08x | %x\n", ptr[j]&~3, ptr[j]&3); j++;
+	    PRINTF(stream, "                0x%08x\n", ptr[j++]);
+	    break;
+	 case LI0_STATE_MAP:
+	    PRINTF(stream, "           MAP: 0x%08x | %x\n", ptr[j]&~3, ptr[j]&3); j++;
+	    PRINTF(stream, "                0x%08x\n", ptr[j++]);
+	    break;
+	 case LI0_STATE_PROGRAM:
+	    PRINTF(stream, "       PROGRAM: 0x%08x | %x\n", ptr[j]&~3, ptr[j]&3); j++;
+	    PRINTF(stream, "                0x%08x\n", ptr[j++]);
+	    break;
+	 case LI0_STATE_CONSTANTS:
+	    PRINTF(stream, "     CONSTANTS: 0x%08x | %x\n", ptr[j]&~3, ptr[j]&3); j++;
+	    PRINTF(stream, "                0x%08x\n", ptr[j++]);
+	    break;
+	 default:
+	    assert(0);
+	    break;
+	 }
+      }
+   }
+
+   if (bits == 0) {
+      PRINTF(stream, "\t  DUMMY: 0x%08x\n", ptr[j++]);
+   }
+
+   PRINTF(stream, "\n");
+
+
+   assert(j == len);
+
+   stream->offset += len * sizeof(unsigned);
+   
+   return TRUE;
+}
+ 	
+static void BR13( struct debug_stream *stream,
+		  unsigned val )
+{
+   PRINTF(stream, "\t0x%08x\n",  val);
+   FLAG(stream, val, 30, "clipping enable");
+   BITS(stream, val, 25, 24, "color depth (3==32bpp)");
+   BITS(stream, val, 23, 16, "raster op");
+   BITS(stream, val, 15, 0,  "dest pitch");
+}
+
+
+static void BR22( struct debug_stream *stream,
+		  unsigned val )
+{
+   PRINTF(stream, "\t0x%08x\n",  val);
+   BITS(stream, val, 31, 16, "dest y1");
+   BITS(stream, val, 15, 0,  "dest x1");
+}
+
+static void BR23( struct debug_stream *stream,
+		  unsigned val )
+{
+   PRINTF(stream, "\t0x%08x\n",  val);
+   BITS(stream, val, 31, 16, "dest y2");
+   BITS(stream, val, 15, 0,  "dest x2");
+}
+
+static void BR09( struct debug_stream *stream,
+		  unsigned val )
+{
+   PRINTF(stream, "\t0x%08x -- dest address\n",  val);
+}
+
+static void BR26( struct debug_stream *stream,
+		  unsigned val )
+{
+   PRINTF(stream, "\t0x%08x\n",  val);
+   BITS(stream, val, 31, 16, "src y1");
+   BITS(stream, val, 15, 0,  "src x1");
+}
+
+static void BR11( struct debug_stream *stream,
+		  unsigned val )
+{
+   PRINTF(stream, "\t0x%08x\n",  val);
+   BITS(stream, val, 15, 0,  "src pitch");
+}
+
+static void BR12( struct debug_stream *stream,
+		  unsigned val )
+{
+   PRINTF(stream, "\t0x%08x -- src address\n",  val);
+}
+
+static void BR16( struct debug_stream *stream,
+		  unsigned val )
+{
+   PRINTF(stream, "\t0x%08x -- color\n",  val);
+}
+   
+static boolean debug_copy_blit( struct debug_stream *stream,
+				  const char *name,
+				  unsigned len )
+{
+   unsigned *ptr = (unsigned *)(stream->ptr + stream->offset);
+   int j = 0;
+
+   PRINTF(stream, "%s (%d dwords):\n", name, len);
+   PRINTF(stream, "\t0x%08x\n",  ptr[j++]);
+   
+   BR13(stream, ptr[j++]);
+   BR22(stream, ptr[j++]);
+   BR23(stream, ptr[j++]);
+   BR09(stream, ptr[j++]);
+   BR26(stream, ptr[j++]);
+   BR11(stream, ptr[j++]);
+   BR12(stream, ptr[j++]);
+
+   stream->offset += len * sizeof(unsigned);
+   assert(j == len);
+   return TRUE;
+}
+
+static boolean debug_color_blit( struct debug_stream *stream,
+				  const char *name,
+				  unsigned len )
+{
+   unsigned *ptr = (unsigned *)(stream->ptr + stream->offset);
+   int j = 0;
+
+   PRINTF(stream, "%s (%d dwords):\n", name, len);
+   PRINTF(stream, "\t0x%08x\n",  ptr[j++]);
+
+   BR13(stream, ptr[j++]);
+   BR22(stream, ptr[j++]);
+   BR23(stream, ptr[j++]);
+   BR09(stream, ptr[j++]);
+   BR16(stream, ptr[j++]);
+
+   stream->offset += len * sizeof(unsigned);
+   assert(j == len);
+   return TRUE;
+}
+
+static boolean debug_modes4( struct debug_stream *stream,
+				  const char *name,
+				  unsigned len )
+{
+   unsigned *ptr = (unsigned *)(stream->ptr + stream->offset);
+   int j = 0;
+
+   PRINTF(stream, "%s (%d dwords):\n", name, len);
+   PRINTF(stream, "\t0x%08x\n",  ptr[j]);
+   BITS(stream, ptr[j], 21, 18, "logicop func");
+   FLAG(stream, ptr[j], 17, "stencil test mask modify-enable");
+   FLAG(stream, ptr[j], 16, "stencil write mask modify-enable");
+   BITS(stream, ptr[j], 15, 8, "stencil test mask");
+   BITS(stream, ptr[j], 7, 0,  "stencil write mask");
+   j++;
+
+   stream->offset += len * sizeof(unsigned);
+   assert(j == len);
+   return TRUE;
+}
+
+static boolean debug_map_state( struct debug_stream *stream,
+				  const char *name,
+				  unsigned len )
+{
+   unsigned *ptr = (unsigned *)(stream->ptr + stream->offset);
+   unsigned j = 0;
+
+   PRINTF(stream, "%s (%d dwords):\n", name, len);
+   PRINTF(stream, "\t0x%08x\n",  ptr[j++]);
+   
+   {
+      PRINTF(stream, "\t0x%08x\n",  ptr[j]);
+      BITS(stream, ptr[j], 15, 0,   "map mask");
+      j++;
+   }
+
+   while (j < len) {
+      {
+	 PRINTF(stream, "\t  TMn.0: 0x%08x\n", ptr[j]);
+	 PRINTF(stream, "\t map address: 0x%08x\n", (ptr[j] & ~0x3));
+	 FLAG(stream, ptr[j], 1, "vertical line stride");
+	 FLAG(stream, ptr[j], 0, "vertical line stride offset");
+	 j++;
+      }
+
+      {
+	 PRINTF(stream, "\t  TMn.1: 0x%08x\n", ptr[j]);
+	 BITS(stream, ptr[j], 31, 21, "height");
+	 BITS(stream, ptr[j], 20, 10, "width");
+	 BITS(stream, ptr[j], 9, 7, "surface format");
+	 BITS(stream, ptr[j], 6, 3, "texel format");
+	 FLAG(stream, ptr[j], 2, "use fence regs");
+	 FLAG(stream, ptr[j], 1, "tiled surface");
+	 FLAG(stream, ptr[j], 0, "tile walk ymajor");
+	 j++;
+      }
+      {
+	 PRINTF(stream, "\t  TMn.2: 0x%08x\n", ptr[j]);
+	 BITS(stream, ptr[j], 31, 21, "dword pitch");
+	 BITS(stream, ptr[j], 20, 15, "cube face enables");
+	 BITS(stream, ptr[j], 14, 9, "max lod");
+	 FLAG(stream, ptr[j], 8,     "mip layout right");
+	 BITS(stream, ptr[j], 7, 0, "depth");
+	 j++;
+      }
+   }
+
+   stream->offset += len * sizeof(unsigned);
+   assert(j == len);
+   return TRUE;
+}
+
+static boolean debug_sampler_state( struct debug_stream *stream,
+				  const char *name,
+				  unsigned len )
+{
+   unsigned *ptr = (unsigned *)(stream->ptr + stream->offset);
+   unsigned j = 0;
+
+   PRINTF(stream, "%s (%d dwords):\n", name, len);
+   PRINTF(stream, "\t0x%08x\n",  ptr[j++]);
+   
+   {
+      PRINTF(stream, "\t0x%08x\n",  ptr[j]);
+      BITS(stream, ptr[j], 15, 0,   "sampler mask");
+      j++;
+   }
+
+   while (j < len) {
+      {
+	 PRINTF(stream, "\t  TSn.0: 0x%08x\n", ptr[j]);
+	 FLAG(stream, ptr[j], 31, "reverse gamma");
+	 FLAG(stream, ptr[j], 30, "planar to packed");
+	 FLAG(stream, ptr[j], 29, "yuv->rgb");
+	 BITS(stream, ptr[j], 28, 27, "chromakey index");
+	 BITS(stream, ptr[j], 26, 22, "base mip level");
+	 BITS(stream, ptr[j], 21, 20, "mip mode filter");
+	 BITS(stream, ptr[j], 19, 17, "mag mode filter");
+	 BITS(stream, ptr[j], 16, 14, "min mode filter");
+	 BITS(stream, ptr[j], 13, 5,  "lod bias (s4.4)");
+	 FLAG(stream, ptr[j], 4,      "shadow enable");
+	 FLAG(stream, ptr[j], 3,      "max-aniso-4");
+	 BITS(stream, ptr[j], 2, 0,   "shadow func");
+	 j++;
+      }
+
+      {
+	 PRINTF(stream, "\t  TSn.1: 0x%08x\n", ptr[j]);
+	 BITS(stream, ptr[j], 31, 24, "min lod");
+	 MBZ( ptr[j], 23, 18 );
+	 FLAG(stream, ptr[j], 17,     "kill pixel enable");
+	 FLAG(stream, ptr[j], 16,     "keyed tex filter mode");
+	 FLAG(stream, ptr[j], 15,     "chromakey enable");
+	 BITS(stream, ptr[j], 14, 12, "tcx wrap mode");
+	 BITS(stream, ptr[j], 11, 9,  "tcy wrap mode");
+	 BITS(stream, ptr[j], 8,  6,  "tcz wrap mode");
+	 FLAG(stream, ptr[j], 5,      "normalized coords");
+	 BITS(stream, ptr[j], 4,  1,  "map (surface) index");
+	 FLAG(stream, ptr[j], 0,      "EAST deinterlacer enable");
+	 j++;
+      }
+      {
+	 PRINTF(stream, "\t  TSn.2: 0x%08x  (default color)\n", ptr[j]);
+	 j++;
+      }
+   }
+
+   stream->offset += len * sizeof(unsigned);
+   assert(j == len);
+   return TRUE;
+}
+
+static boolean debug_dest_vars( struct debug_stream *stream,
+				  const char *name,
+				  unsigned len )
+{
+   unsigned *ptr = (unsigned *)(stream->ptr + stream->offset);
+   int j = 0;
+
+   PRINTF(stream, "%s (%d dwords):\n", name, len);
+   PRINTF(stream, "\t0x%08x\n",  ptr[j++]);
+
+   {
+      PRINTF(stream, "\t0x%08x\n",  ptr[j]);
+      FLAG(stream, ptr[j], 31,     "early classic ztest");
+      FLAG(stream, ptr[j], 30,     "opengl tex default color");
+      FLAG(stream, ptr[j], 29,     "bypass iz");
+      FLAG(stream, ptr[j], 28,     "lod preclamp");
+      BITS(stream, ptr[j], 27, 26, "dither pattern");
+      FLAG(stream, ptr[j], 25,     "linear gamma blend");
+      FLAG(stream, ptr[j], 24,     "debug dither");
+      BITS(stream, ptr[j], 23, 20, "dstorg x");
+      BITS(stream, ptr[j], 19, 16, "dstorg y");
+      MBZ (ptr[j], 15, 15 );
+      BITS(stream, ptr[j], 14, 12, "422 write select");
+      BITS(stream, ptr[j], 11, 8,  "cbuf format");
+      BITS(stream, ptr[j], 3, 2,   "zbuf format");
+      FLAG(stream, ptr[j], 1,      "vert line stride");
+      FLAG(stream, ptr[j], 1,      "vert line stride offset");
+      j++;
+   }
+   
+   stream->offset += len * sizeof(unsigned);
+   assert(j == len);
+   return TRUE;
+}
+
+static boolean debug_buf_info( struct debug_stream *stream,
+				  const char *name,
+				  unsigned len )
+{
+   unsigned *ptr = (unsigned *)(stream->ptr + stream->offset);
+   int j = 0;
+
+   PRINTF(stream, "%s (%d dwords):\n", name, len);
+   PRINTF(stream, "\t0x%08x\n",  ptr[j++]);
+
+   {
+      PRINTF(stream, "\t0x%08x\n",  ptr[j]);
+      BITS(stream, ptr[j], 28, 28, "aux buffer id");
+      BITS(stream, ptr[j], 27, 24, "buffer id (7=depth, 3=back)");
+      FLAG(stream, ptr[j], 23,     "use fence regs");
+      FLAG(stream, ptr[j], 22,     "tiled surface");
+      FLAG(stream, ptr[j], 21,     "tile walk ymajor");
+      MBZ (ptr[j], 20, 14);
+      BITS(stream, ptr[j], 13, 2,  "dword pitch");
+      MBZ (ptr[j], 2,  0);
+      j++;
+   }
+   
+   PRINTF(stream, "\t0x%08x -- buffer base address\n",  ptr[j++]);
+
+   stream->offset += len * sizeof(unsigned);
+   assert(j == len);
+   return TRUE;
+}
+
+static boolean i915_debug_packet( struct debug_stream *stream )
+{
+   unsigned *ptr = (unsigned *)(stream->ptr + stream->offset);
+   unsigned cmd = *ptr;
+   
+   switch (((cmd >> 29) & 0x7)) {
+   case 0x0:
+      switch ((cmd >> 23) & 0x3f) {
+      case 0x0:
+	 return debug(stream, "MI_NOOP", 1);
+      case 0x3:
+	 return debug(stream, "MI_WAIT_FOR_EVENT", 1);
+      case 0x4:
+	 return debug(stream, "MI_FLUSH", 1);
+      case 0xA:
+	 debug(stream, "MI_BATCH_BUFFER_END", 1);
+	 return FALSE;
+      case 0x22:
+	 return debug(stream, "MI_LOAD_REGISTER_IMM", 3);
+      case 0x31:
+	 return debug_chain(stream, "MI_BATCH_BUFFER_START", 2);
+      default:
+         (void)debug(stream, "UNKNOWN 0x0 case!", 1);
+         assert(0);
+	 break;
+      }
+      break;
+   case 0x1:
+      (void) debug(stream, "UNKNOWN 0x1 case!", 1);
+      assert(0);
+      break;
+   case 0x2:
+      switch ((cmd >> 22) & 0xff) {	 
+      case 0x50:
+	 return debug_color_blit(stream, "XY_COLOR_BLT", (cmd & 0xff) + 2);
+      case 0x53:
+	 return debug_copy_blit(stream, "XY_SRC_COPY_BLT", (cmd & 0xff) + 2);
+      default:
+	 return debug(stream, "blit command", (cmd & 0xff) + 2);
+      }
+      break;
+   case 0x3:
+      switch ((cmd >> 24) & 0x1f) {	 
+      case 0x6:
+	 return debug(stream, "3DSTATE_ANTI_ALIASING", 1);
+      case 0x7:
+	 return debug(stream, "3DSTATE_RASTERIZATION_RULES", 1);
+      case 0x8:
+	 return debug(stream, "3DSTATE_BACKFACE_STENCIL_OPS", 2);
+      case 0x9:
+	 return debug(stream, "3DSTATE_BACKFACE_STENCIL_MASKS", 1);
+      case 0xb:
+	 return debug(stream, "3DSTATE_INDEPENDENT_ALPHA_BLEND", 1);
+      case 0xc:
+	 return debug(stream, "3DSTATE_MODES5", 1);	 
+      case 0xd:
+	 return debug_modes4(stream, "3DSTATE_MODES4", 1);
+      case 0x15:
+	 return debug(stream, "3DSTATE_FOG_COLOR", 1);
+      case 0x16:
+	 return debug(stream, "3DSTATE_COORD_SET_BINDINGS", 1);
+      case 0x1c:
+	 /* 3DState16NP */
+	 switch((cmd >> 19) & 0x1f) {
+	 case 0x10:
+	    return debug(stream, "3DSTATE_SCISSOR_ENABLE", 1);
+	 case 0x11:
+	    return debug(stream, "3DSTATE_DEPTH_SUBRECTANGLE_DISABLE", 1);
+	 default:
+            (void) debug(stream, "UNKNOWN 0x1c case!", 1);
+            assert(0);
+	    break;
+	 }
+	 break;
+      case 0x1d:
+	 /* 3DStateMW */
+	 switch ((cmd >> 16) & 0xff) {
+	 case 0x0:
+	    return debug_map_state(stream, "3DSTATE_MAP_STATE", (cmd & 0x1f) + 2);
+	 case 0x1:
+	    return debug_sampler_state(stream, "3DSTATE_SAMPLER_STATE", (cmd & 0x1f) + 2);
+	 case 0x4:
+	    return debug_load_immediate(stream, "3DSTATE_LOAD_STATE_IMMEDIATE", (cmd & 0xf) + 2);
+	 case 0x5:
+	    return debug_program(stream, "3DSTATE_PIXEL_SHADER_PROGRAM", (cmd & 0x1ff) + 2);
+	 case 0x6:
+	    return debug(stream, "3DSTATE_PIXEL_SHADER_CONSTANTS", (cmd & 0xff) + 2);
+	 case 0x7:
+	    return debug_load_indirect(stream, "3DSTATE_LOAD_INDIRECT", (cmd & 0xff) + 2);
+	 case 0x80:
+	    return debug(stream, "3DSTATE_DRAWING_RECTANGLE", (cmd & 0xffff) + 2);
+	 case 0x81:
+	    return debug(stream, "3DSTATE_SCISSOR_RECTANGLE", (cmd & 0xffff) + 2);
+	 case 0x83:
+	    return debug(stream, "3DSTATE_SPAN_STIPPLE", (cmd & 0xffff) + 2);
+	 case 0x85:
+	    return debug_dest_vars(stream, "3DSTATE_DEST_BUFFER_VARS", (cmd & 0xffff) + 2);
+	 case 0x88:
+	    return debug(stream, "3DSTATE_CONSTANT_BLEND_COLOR", (cmd & 0xffff) + 2);
+	 case 0x89:
+	    return debug(stream, "3DSTATE_FOG_MODE", (cmd & 0xffff) + 2);
+	 case 0x8e:
+	    return debug_buf_info(stream, "3DSTATE_BUFFER_INFO", (cmd & 0xffff) + 2);
+	 case 0x97:
+	    return debug(stream, "3DSTATE_DEPTH_OFFSET_SCALE", (cmd & 0xffff) + 2);
+	 case 0x98:
+	    return debug(stream, "3DSTATE_DEFAULT_Z", (cmd & 0xffff) + 2);
+	 case 0x99:
+	    return debug(stream, "3DSTATE_DEFAULT_DIFFUSE", (cmd & 0xffff) + 2);
+	 case 0x9a:
+	    return debug(stream, "3DSTATE_DEFAULT_SPECULAR", (cmd & 0xffff) + 2);
+	 case 0x9c:
+	    return debug(stream, "3DSTATE_CLEAR_PARAMETERS", (cmd & 0xffff) + 2);
+	 default:
+	    assert(0);
+	    return 0;
+	 }
+	 break;
+      case 0x1e:
+	 if (cmd & (1 << 23))
+	    return debug(stream, "???", (cmd & 0xffff) + 1);
+	 else
+	    return debug(stream, "", 1);
+	 break;
+      case 0x1f:
+	 if ((cmd & (1 << 23)) == 0)	
+	    return debug_prim(stream, "3DPRIM (inline)", 1, (cmd & 0x1ffff) + 2);
+	 else if (cmd & (1 << 17)) 
+	 {
+	    if ((cmd & 0xffff) == 0)
+	       return debug_variable_length_prim(stream);
+	    else
+	       return debug_prim(stream, "3DPRIM (indexed)", 0, (((cmd & 0xffff) + 1) / 2) + 1);
+	 }
+	 else
+	    return debug_prim(stream, "3DPRIM  (indirect sequential)", 0, 2); 
+	 break;
+      default:
+	 return debug(stream, "", 0);
+      }
+   default:
+      assert(0);
+      return 0;
+   }
+
+   assert(0);
+   return 0;
+}
+
+
+
+void
+i915_dump_batchbuffer( struct i915_context *i915 )
+{
+   struct debug_stream stream;
+   unsigned *start = i915->batch_start;
+   unsigned *end = i915->winsys->batch_start( i915->winsys, 0, 0 );
+   unsigned long bytes = (unsigned long) (end - start) * 4;
+   boolean done = FALSE;
+
+   stream.offset = 0;
+   stream.ptr = (char *)start;
+   stream.print_addresses = 0;
+   stream.winsys = i915->pipe.winsys;
+
+   if (!start || !end) {
+      stream.winsys->printf( stream.winsys, "\n\nBATCH: ???\n");
+      return;
+   }
+   
+   stream.winsys->printf( stream.winsys, "\n\nBATCH: (%d)\n", bytes / 4);
+
+   while (!done &&
+	  stream.offset < bytes)
+   {
+      if (!i915_debug_packet( &stream ))
+	 break;
+
+      assert(stream.offset <= bytes &&
+	     stream.offset >= 0);
+   }
+
+   stream.winsys->printf( stream.winsys, "END-BATCH\n\n\n");
+}
+
+
diff --git a/src/gallium/drivers/i915simple/i915_debug.h b/src/gallium/drivers/i915simple/i915_debug.h
new file mode 100644
index 0000000000..0bcd094233
--- /dev/null
+++ b/src/gallium/drivers/i915simple/i915_debug.h
@@ -0,0 +1,117 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+/* Authors:  Keith Whitwell <keith@tungstengraphics.com>
+ */
+
+#ifndef I915_DEBUG_H
+#define I915_DEBUG_H
+
+#include <stdarg.h>
+
+struct i915_context;
+
+struct debug_stream 
+{
+   unsigned offset;		/* current gtt offset */
+   char *ptr;		/* pointer to gtt offset zero */
+   char *end;		/* pointer to gtt offset zero */
+   unsigned print_addresses;
+   struct pipe_winsys *winsys;
+};
+
+
+/* Internal functions
+ */
+void i915_disassemble_program(struct debug_stream *stream, 
+			      const unsigned *program, unsigned sz);
+
+void i915_print_ureg(const char *msg, unsigned ureg);
+
+
+#define DEBUG_BATCH	 0x1
+#define DEBUG_BLIT       0x2
+#define DEBUG_BUFFER     0x4
+#define DEBUG_CONSTANTS  0x8
+#define DEBUG_CONTEXT    0x10
+#define DEBUG_DRAW	 0x20
+#define DEBUG_DYNAMIC	 0x40
+#define DEBUG_FLUSH      0x80
+#define DEBUG_MAP	 0x100
+#define DEBUG_PROGRAM	 0x200
+#define DEBUG_REGIONS    0x400
+#define DEBUG_SAMPLER	 0x800
+#define DEBUG_STATIC	 0x1000
+#define DEBUG_SURFACE    0x2000
+#define DEBUG_WINSYS     0x4000
+
+#include "pipe/p_compiler.h"
+
+#if defined(DEBUG) && defined(FILE_DEBUG_FLAG)
+
+#include "pipe/p_winsys.h"
+
+static INLINE void
+I915_DBG(
+   struct i915_context  *i915,
+   const char           *fmt,
+                        ... )
+{
+   if ((i915)->debug & FILE_DEBUG_FLAG) {
+      va_list  args;
+      char     buffer[256];
+
+      va_start( args, fmt );
+      vsprintf( buffer, fmt, args );
+      i915->pipe.winsys->printf( i915->pipe.winsys, buffer );
+      va_end( args );
+   }
+}
+
+#else
+
+static INLINE void
+I915_DBG(
+   struct i915_context  *i915,
+   const char           *fmt,
+                        ... )
+{
+   (void) i915;
+   (void) fmt;
+}
+
+#endif
+
+
+void i915_dump_batchbuffer( struct i915_context *i915 );
+
+
+
+void i915_debug_init( struct i915_context *i915 );
+
+
+#endif
diff --git a/src/gallium/drivers/i915simple/i915_debug_fp.c b/src/gallium/drivers/i915simple/i915_debug_fp.c
new file mode 100644
index 0000000000..ebfdb3d93c
--- /dev/null
+++ b/src/gallium/drivers/i915simple/i915_debug_fp.c
@@ -0,0 +1,366 @@
+/**************************************************************************
+ * 
+ * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+
+#include "i915_reg.h"
+#include "i915_debug.h"
+#include "pipe/p_winsys.h"
+#include "pipe/p_util.h"
+
+
+static void
+PRINTF(
+   struct debug_stream  *stream,
+   const char           *fmt,
+                        ... )
+{
+   va_list  args;
+   char     buffer[256];
+
+   va_start( args, fmt );
+   vsprintf( buffer, fmt, args );
+   stream->winsys->printf( stream->winsys, buffer );
+   va_end( args );
+}
+
+
+static const char *opcodes[0x20] = {
+   "NOP",
+   "ADD",
+   "MOV",
+   "MUL",
+   "MAD",
+   "DP2ADD",
+   "DP3",
+   "DP4",
+   "FRC",
+   "RCP",
+   "RSQ",
+   "EXP",
+   "LOG",
+   "CMP",
+   "MIN",
+   "MAX",
+   "FLR",
+   "MOD",
+   "TRC",
+   "SGE",
+   "SLT",
+   "TEXLD",
+   "TEXLDP",
+   "TEXLDB",
+   "TEXKILL",
+   "DCL",
+   "0x1a",
+   "0x1b",
+   "0x1c",
+   "0x1d",
+   "0x1e",
+   "0x1f",
+};
+
+
+static const int args[0x20] = {
+   0,                           /* 0 nop */
+   2,                           /* 1 add */
+   1,                           /* 2 mov */
+   2,                           /* 3 m ul */
+   3,                           /* 4 mad */
+   3,                           /* 5 dp2add */
+   2,                           /* 6 dp3 */
+   2,                           /* 7 dp4 */
+   1,                           /* 8 frc */
+   1,                           /* 9 rcp */
+   1,                           /* a rsq */
+   1,                           /* b exp */
+   1,                           /* c log */
+   3,                           /* d cmp */
+   2,                           /* e min */
+   2,                           /* f max */
+   1,                           /* 10 flr */
+   1,                           /* 11 mod */
+   1,                           /* 12 trc */
+   2,                           /* 13 sge */
+   2,                           /* 14 slt */
+   1,
+   1,
+   1,
+   1,
+   0,
+   0,
+   0,
+   0,
+   0,
+   0,
+   0,
+};
+
+
+static const char *regname[0x8] = {
+   "R",
+   "T",
+   "CONST",
+   "S",
+   "OC",
+   "OD",
+   "U",
+   "UNKNOWN",
+};
+
+static void
+print_reg_type_nr(struct debug_stream *stream, unsigned type, unsigned nr)
+{
+   switch (type) {
+   case REG_TYPE_T:
+      switch (nr) {
+      case T_DIFFUSE:
+         PRINTF(stream, "T_DIFFUSE");
+         return;
+      case T_SPECULAR:
+         PRINTF(stream, "T_SPECULAR");
+         return;
+      case T_FOG_W:
+         PRINTF(stream, "T_FOG_W");
+         return;
+      default:
+         PRINTF(stream, "T_TEX%d", nr);
+         return;
+      }
+   case REG_TYPE_OC:
+      if (nr == 0) {
+         PRINTF(stream, "oC");
+         return;
+      }
+      break;
+   case REG_TYPE_OD:
+      if (nr == 0) {
+         PRINTF(stream, "oD");
+         return;
+      }
+      break;
+   default:
+      break;
+   }
+
+   PRINTF(stream, "%s[%d]", regname[type], nr);
+}
+
+#define REG_SWIZZLE_MASK 0x7777
+#define REG_NEGATE_MASK 0x8888
+
+#define REG_SWIZZLE_XYZW ((SRC_X << A2_SRC2_CHANNEL_X_SHIFT) |	\
+		      (SRC_Y << A2_SRC2_CHANNEL_Y_SHIFT) |	\
+		      (SRC_Z << A2_SRC2_CHANNEL_Z_SHIFT) |	\
+		      (SRC_W << A2_SRC2_CHANNEL_W_SHIFT))
+
+
+static void
+print_reg_neg_swizzle(struct debug_stream *stream, unsigned reg)
+{
+   int i;
+
+   if ((reg & REG_SWIZZLE_MASK) == REG_SWIZZLE_XYZW &&
+       (reg & REG_NEGATE_MASK) == 0)
+      return;
+
+   PRINTF(stream, ".");
+
+   for (i = 3; i >= 0; i--) {
+      if (reg & (1 << ((i * 4) + 3)))
+         PRINTF(stream, "-");
+
+      switch ((reg >> (i * 4)) & 0x7) {
+      case 0:
+         PRINTF(stream, "x");
+         break;
+      case 1:
+         PRINTF(stream, "y");
+         break;
+      case 2:
+         PRINTF(stream, "z");
+         break;
+      case 3:
+         PRINTF(stream, "w");
+         break;
+      case 4:
+         PRINTF(stream, "0");
+         break;
+      case 5:
+         PRINTF(stream, "1");
+         break;
+      default:
+         PRINTF(stream, "?");
+         break;
+      }
+   }
+}
+
+
+static void
+print_src_reg(struct debug_stream *stream, unsigned dword)
+{
+   unsigned nr = (dword >> A2_SRC2_NR_SHIFT) & REG_NR_MASK;
+   unsigned type = (dword >> A2_SRC2_TYPE_SHIFT) & REG_TYPE_MASK;
+   print_reg_type_nr(stream, type, nr);
+   print_reg_neg_swizzle(stream, dword);
+}
+
+
+static void
+print_dest_reg(struct debug_stream *stream, unsigned dword)
+{
+   unsigned nr = (dword >> A0_DEST_NR_SHIFT) & REG_NR_MASK;
+   unsigned type = (dword >> A0_DEST_TYPE_SHIFT) & REG_TYPE_MASK;
+   print_reg_type_nr(stream, type, nr);
+   if ((dword & A0_DEST_CHANNEL_ALL) == A0_DEST_CHANNEL_ALL)
+      return;
+   PRINTF(stream, ".");
+   if (dword & A0_DEST_CHANNEL_X)
+      PRINTF(stream, "x");
+   if (dword & A0_DEST_CHANNEL_Y)
+      PRINTF(stream, "y");
+   if (dword & A0_DEST_CHANNEL_Z)
+      PRINTF(stream, "z");
+   if (dword & A0_DEST_CHANNEL_W)
+      PRINTF(stream, "w");
+}
+
+
+#define GET_SRC0_REG(r0, r1) ((r0<<14)|(r1>>A1_SRC0_CHANNEL_W_SHIFT))
+#define GET_SRC1_REG(r0, r1) ((r0<<8)|(r1>>A2_SRC1_CHANNEL_W_SHIFT))
+#define GET_SRC2_REG(r)      (r)
+
+
+static void
+print_arith_op(struct debug_stream *stream, 
+	       unsigned opcode, const unsigned * program)
+{
+   if (opcode != A0_NOP) {
+      print_dest_reg(stream, program[0]);
+      if (program[0] & A0_DEST_SATURATE)
+         PRINTF(stream, " = SATURATE ");
+      else
+         PRINTF(stream, " = ");
+   }
+
+   PRINTF(stream, "%s ", opcodes[opcode]);
+
+   print_src_reg(stream, GET_SRC0_REG(program[0], program[1]));
+   if (args[opcode] == 1) {
+      PRINTF(stream, "\n");
+      return;
+   }
+
+   PRINTF(stream, ", ");
+   print_src_reg(stream, GET_SRC1_REG(program[1], program[2]));
+   if (args[opcode] == 2) {
+      PRINTF(stream, "\n");
+      return;
+   }
+
+   PRINTF(stream, ", ");
+   print_src_reg(stream, GET_SRC2_REG(program[2]));
+   PRINTF(stream, "\n");
+   return;
+}
+
+
+static void
+print_tex_op(struct debug_stream *stream, 
+	     unsigned opcode, const unsigned * program)
+{
+   print_dest_reg(stream, program[0] | A0_DEST_CHANNEL_ALL);
+   PRINTF(stream, " = ");
+
+   PRINTF(stream, "%s ", opcodes[opcode]);
+
+   PRINTF(stream, "S[%d],", program[0] & T0_SAMPLER_NR_MASK);
+
+   print_reg_type_nr(stream, 
+		     (program[1] >> T1_ADDRESS_REG_TYPE_SHIFT) &
+                     REG_TYPE_MASK,
+                     (program[1] >> T1_ADDRESS_REG_NR_SHIFT) & REG_NR_MASK);
+   PRINTF(stream, "\n");
+}
+
+static void
+print_texkil_op(struct debug_stream *stream, 
+                unsigned opcode, const unsigned * program)
+{
+   PRINTF(stream, "TEXKIL ");
+
+   print_reg_type_nr(stream, 
+		     (program[1] >> T1_ADDRESS_REG_TYPE_SHIFT) &
+                     REG_TYPE_MASK,
+                     (program[1] >> T1_ADDRESS_REG_NR_SHIFT) & REG_NR_MASK);
+   PRINTF(stream, "\n");
+}
+
+static void
+print_dcl_op(struct debug_stream *stream, 
+	     unsigned opcode, const unsigned * program)
+{
+   PRINTF(stream, "%s ", opcodes[opcode]);
+   print_dest_reg(stream, 
+		  program[0] | A0_DEST_CHANNEL_ALL);
+   PRINTF(stream, "\n");
+}
+
+
+void
+i915_disassemble_program(struct debug_stream *stream, 
+			 const unsigned * program, unsigned sz)
+{
+   unsigned size = program[0] & 0x1ff;
+   unsigned i;
+
+   PRINTF(stream, "\t\tBEGIN\n");
+
+   assert(size + 2 == sz);
+
+   program++;
+   for (i = 1; i < sz; i += 3, program += 3) {
+      unsigned opcode = program[0] & (0x1f << 24);
+
+      PRINTF(stream, "\t\t");
+
+      if ((int) opcode >= A0_NOP && opcode <= A0_SLT)
+         print_arith_op(stream, opcode >> 24, program);
+      else if (opcode >= T0_TEXLD && opcode < T0_TEXKILL)
+         print_tex_op(stream, opcode >> 24, program);
+      else if (opcode == T0_TEXKILL)
+         print_texkil_op(stream, opcode >> 24, program);
+      else if (opcode == D0_DCL)
+         print_dcl_op(stream, opcode >> 24, program);
+      else
+         PRINTF(stream, "Unknown opcode 0x%x\n", opcode);
+   }
+
+   PRINTF(stream, "\t\tEND\n\n");
+}
+
+
diff --git a/src/gallium/drivers/i915simple/i915_flush.c b/src/gallium/drivers/i915simple/i915_flush.c
new file mode 100644
index 0000000000..3c2069b827
--- /dev/null
+++ b/src/gallium/drivers/i915simple/i915_flush.c
@@ -0,0 +1,81 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+/* Author:
+ *    Keith Whitwell <keith@tungstengraphics.com>
+ */
+
+
+#include "pipe/p_defines.h"
+#include "i915_context.h"
+#include "i915_reg.h"
+#include "i915_batch.h"
+
+
+/**
+ * In future we may want a fence-like interface instead of finish.
+ */
+static void i915_flush( struct pipe_context *pipe,
+			unsigned flags )
+{
+   struct i915_context *i915 = i915_context(pipe);
+
+   /* Do we need to emit an MI_FLUSH command to flush the hardware
+    * caches?
+    */
+   if (flags & (PIPE_FLUSH_RENDER_CACHE | PIPE_FLUSH_TEXTURE_CACHE)) {
+      unsigned flush = MI_FLUSH;
+      
+      if (!(flags & PIPE_FLUSH_RENDER_CACHE))
+	 flush |= INHIBIT_FLUSH_RENDER_CACHE;
+
+      if (flags & PIPE_FLUSH_TEXTURE_CACHE)
+	 flush |= FLUSH_MAP_CACHE;
+
+      if (!BEGIN_BATCH(1, 0)) {
+	 FLUSH_BATCH();
+	 assert(BEGIN_BATCH(1, 0));
+      }
+      OUT_BATCH( flush );
+      ADVANCE_BATCH();
+   }
+
+   /* If there are no flags, just flush pending commands to hardware:
+    */
+   FLUSH_BATCH();
+
+   if (flags & PIPE_FLUSH_WAIT) {
+      i915->winsys->batch_finish(i915->winsys);
+   }
+}
+
+
+
+void i915_init_flush_functions( struct i915_context *i915 )
+{
+   i915->pipe.flush = i915_flush;
+}
diff --git a/src/gallium/drivers/i915simple/i915_fpc.h b/src/gallium/drivers/i915simple/i915_fpc.h
new file mode 100644
index 0000000000..8c7b68aefb
--- /dev/null
+++ b/src/gallium/drivers/i915simple/i915_fpc.h
@@ -0,0 +1,213 @@
+/**************************************************************************
+ * 
+ * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+
+#ifndef I915_FPC_H
+#define I915_FPC_H
+
+#include "pipe/p_util.h"
+
+#include "i915_context.h"
+#include "i915_reg.h"
+
+
+
+#define I915_PROGRAM_SIZE 192
+
+
+
+/**
+ * Program translation state
+ */
+struct i915_fp_compile {
+   const struct pipe_shader_state *shader;
+
+   struct vertex_info *vertex_info;
+
+   uint declarations[I915_PROGRAM_SIZE];
+   uint program[I915_PROGRAM_SIZE];
+
+   uint input_semantic_name[PIPE_MAX_SHADER_INPUTS];
+   uint input_semantic_index[PIPE_MAX_SHADER_INPUTS];
+
+   uint output_semantic_name[PIPE_MAX_SHADER_OUTPUTS];
+   uint output_semantic_index[PIPE_MAX_SHADER_OUTPUTS];
+
+   /** points into the i915->current.constants array: */
+   float (*constants)[4];
+   uint num_constants;
+   uint constant_flags[I915_MAX_CONSTANT]; /**< status of each constant */
+
+   uint *csr;            /**< Cursor, points into program. */
+
+   uint *decl;           /**< Cursor, points into declarations. */
+
+   uint decl_s;          /**< flags for which s regs need to be decl'd */
+   uint decl_t;          /**< flags for which t regs need to be decl'd */
+
+   uint temp_flag;       /**< Tracks temporary regs which are in use */
+   uint utemp_flag;      /**< Tracks TYPE_U temporary regs which are in use */
+
+   uint nr_tex_indirect;
+   uint nr_tex_insn;
+   uint nr_alu_insn;
+   uint nr_decl_insn;
+
+   boolean error;      /**< Set if i915_program_error() is called */
+   uint wpos_tex;
+   uint NumNativeInstructions;
+   uint NumNativeAluInstructions;
+   uint NumNativeTexInstructions;
+   uint NumNativeTexIndirections;
+};
+
+
+/* Having zero and one in here makes the definition of swizzle a lot
+ * easier.
+ */
+#define UREG_TYPE_SHIFT               29
+#define UREG_NR_SHIFT                 24
+#define UREG_CHANNEL_X_NEGATE_SHIFT   23
+#define UREG_CHANNEL_X_SHIFT          20
+#define UREG_CHANNEL_Y_NEGATE_SHIFT   19
+#define UREG_CHANNEL_Y_SHIFT          16
+#define UREG_CHANNEL_Z_NEGATE_SHIFT   15
+#define UREG_CHANNEL_Z_SHIFT          12
+#define UREG_CHANNEL_W_NEGATE_SHIFT   11
+#define UREG_CHANNEL_W_SHIFT          8
+#define UREG_CHANNEL_ZERO_NEGATE_MBZ  5
+#define UREG_CHANNEL_ZERO_SHIFT       4
+#define UREG_CHANNEL_ONE_NEGATE_MBZ   1
+#define UREG_CHANNEL_ONE_SHIFT        0
+
+#define UREG_BAD          0xffffffff    /* not a valid ureg */
+
+#define X    SRC_X
+#define Y    SRC_Y
+#define Z    SRC_Z
+#define W    SRC_W
+#define ZERO SRC_ZERO
+#define ONE  SRC_ONE
+
+/* Construct a ureg:
+ */
+#define UREG( type, nr ) (((type)<< UREG_TYPE_SHIFT) |		\
+			  ((nr)  << UREG_NR_SHIFT) |		\
+			  (X     << UREG_CHANNEL_X_SHIFT) |	\
+			  (Y     << UREG_CHANNEL_Y_SHIFT) |	\
+			  (Z     << UREG_CHANNEL_Z_SHIFT) |	\
+			  (W     << UREG_CHANNEL_W_SHIFT) |	\
+			  (ZERO  << UREG_CHANNEL_ZERO_SHIFT) |	\
+			  (ONE   << UREG_CHANNEL_ONE_SHIFT))
+
+#define GET_CHANNEL_SRC( reg, channel ) ((reg<<(channel*4)) & (0xf<<20))
+#define CHANNEL_SRC( src, channel ) (src>>(channel*4))
+
+#define GET_UREG_TYPE(reg) (((reg)>>UREG_TYPE_SHIFT)&REG_TYPE_MASK)
+#define GET_UREG_NR(reg)   (((reg)>>UREG_NR_SHIFT)&REG_NR_MASK)
+
+
+
+#define UREG_XYZW_CHANNEL_MASK 0x00ffff00
+
+/* One neat thing about the UREG representation:  
+ */
+static INLINE int
+swizzle(int reg, uint x, uint y, uint z, uint w)
+{
+   assert(x <= SRC_ONE);
+   assert(y <= SRC_ONE);
+   assert(z <= SRC_ONE);
+   assert(w <= SRC_ONE);
+   return ((reg & ~UREG_XYZW_CHANNEL_MASK) |
+           CHANNEL_SRC(GET_CHANNEL_SRC(reg, x), 0) |
+           CHANNEL_SRC(GET_CHANNEL_SRC(reg, y), 1) |
+           CHANNEL_SRC(GET_CHANNEL_SRC(reg, z), 2) |
+           CHANNEL_SRC(GET_CHANNEL_SRC(reg, w), 3));
+}
+
+
+
+/***********************************************************************
+ * Public interface for the compiler
+ */
+extern void i915_translate_fragment_program( struct i915_context *i915 );
+
+
+
+extern uint i915_get_temp(struct i915_fp_compile *p);
+extern uint i915_get_utemp(struct i915_fp_compile *p);
+extern void i915_release_utemps(struct i915_fp_compile *p);
+
+
+extern uint i915_emit_texld(struct i915_fp_compile *p,
+                              uint dest,
+                              uint destmask,
+                              uint sampler, uint coord, uint op);
+
+extern uint i915_emit_arith(struct i915_fp_compile *p,
+                              uint op,
+                              uint dest,
+                              uint mask,
+                              uint saturate,
+                              uint src0, uint src1, uint src2);
+
+extern uint i915_emit_decl(struct i915_fp_compile *p,
+                             uint type, uint nr, uint d0_flags);
+
+
+extern uint i915_emit_const1f(struct i915_fp_compile *p, float c0);
+
+extern uint i915_emit_const2f(struct i915_fp_compile *p,
+                                float c0, float c1);
+
+extern uint i915_emit_const4fv(struct i915_fp_compile *p,
+                                 const float * c);
+
+extern uint i915_emit_const4f(struct i915_fp_compile *p,
+                                float c0, float c1,
+                                float c2, float c3);
+
+
+/*======================================================================
+ * i915_fpc_debug.c
+ */
+extern void i915_disassemble_program(const uint * program, uint sz);
+
+
+/*======================================================================
+ * i915_fpc_translate.c
+ */
+
+extern void
+i915_program_error(struct i915_fp_compile *p, const char *msg, ...);
+
+extern void
+i915_translate_fragment_program(struct i915_context *i915);
+
+
+#endif
diff --git a/src/gallium/drivers/i915simple/i915_fpc_emit.c b/src/gallium/drivers/i915simple/i915_fpc_emit.c
new file mode 100644
index 0000000000..74924ff0a1
--- /dev/null
+++ b/src/gallium/drivers/i915simple/i915_fpc_emit.c
@@ -0,0 +1,375 @@
+/**************************************************************************
+ * 
+ * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+#include "i915_reg.h"
+#include "i915_context.h"
+#include "i915_fpc.h"
+
+
+#define A0_DEST( reg ) (((reg)&UREG_TYPE_NR_MASK)>>UREG_A0_DEST_SHIFT_LEFT)
+#define D0_DEST( reg ) (((reg)&UREG_TYPE_NR_MASK)>>UREG_A0_DEST_SHIFT_LEFT)
+#define T0_DEST( reg ) (((reg)&UREG_TYPE_NR_MASK)>>UREG_A0_DEST_SHIFT_LEFT)
+#define A0_SRC0( reg ) (((reg)&UREG_MASK)>>UREG_A0_SRC0_SHIFT_LEFT)
+#define A1_SRC0( reg ) (((reg)&UREG_MASK)<<UREG_A1_SRC0_SHIFT_RIGHT)
+#define A1_SRC1( reg ) (((reg)&UREG_MASK)>>UREG_A1_SRC1_SHIFT_LEFT)
+#define A2_SRC1( reg ) (((reg)&UREG_MASK)<<UREG_A2_SRC1_SHIFT_RIGHT)
+#define A2_SRC2( reg ) (((reg)&UREG_MASK)>>UREG_A2_SRC2_SHIFT_LEFT)
+
+/* These are special, and don't have swizzle/negate bits.
+ */
+#define T0_SAMPLER( reg )     (GET_UREG_NR(reg)<<T0_SAMPLER_NR_SHIFT)
+#define T1_ADDRESS_REG( reg ) ((GET_UREG_NR(reg)<<T1_ADDRESS_REG_NR_SHIFT) | \
+			       (GET_UREG_TYPE(reg)<<T1_ADDRESS_REG_TYPE_SHIFT))
+
+
+/* Macros for translating UREG's into the various register fields used
+ * by the I915 programmable unit.
+ */
+#define UREG_A0_DEST_SHIFT_LEFT  (UREG_TYPE_SHIFT - A0_DEST_TYPE_SHIFT)
+#define UREG_A0_SRC0_SHIFT_LEFT  (UREG_TYPE_SHIFT - A0_SRC0_TYPE_SHIFT)
+#define UREG_A1_SRC0_SHIFT_RIGHT (A1_SRC0_CHANNEL_W_SHIFT - UREG_CHANNEL_W_SHIFT)
+#define UREG_A1_SRC1_SHIFT_LEFT  (UREG_TYPE_SHIFT - A1_SRC1_TYPE_SHIFT)
+#define UREG_A2_SRC1_SHIFT_RIGHT (A2_SRC1_CHANNEL_W_SHIFT - UREG_CHANNEL_W_SHIFT)
+#define UREG_A2_SRC2_SHIFT_LEFT  (UREG_TYPE_SHIFT - A2_SRC2_TYPE_SHIFT)
+
+#define UREG_MASK         0xffffff00
+#define UREG_TYPE_NR_MASK ((REG_TYPE_MASK << UREG_TYPE_SHIFT) | \
+  			   (REG_NR_MASK << UREG_NR_SHIFT))
+
+
+#define I915_CONSTFLAG_PARAM 0x1f
+
+uint
+i915_get_temp(struct i915_fp_compile *p)
+{
+   int bit = ffs(~p->temp_flag);
+   if (!bit) {
+      i915_program_error(p, "i915_get_temp: out of temporaries\n");
+      return 0;
+   }
+
+   p->temp_flag |= 1 << (bit - 1);
+   return UREG(REG_TYPE_R, (bit - 1));
+}
+
+
+uint
+i915_get_utemp(struct i915_fp_compile * p)
+{
+   int bit = ffs(~p->utemp_flag);
+   if (!bit) {
+      i915_program_error(p, "i915_get_utemp: out of temporaries\n");
+      return 0;
+   }
+
+   p->utemp_flag |= 1 << (bit - 1);
+   return UREG(REG_TYPE_U, (bit - 1));
+}
+
+void
+i915_release_utemps(struct i915_fp_compile *p)
+{
+   p->utemp_flag = ~0x7;
+}
+
+
+uint
+i915_emit_decl(struct i915_fp_compile *p,
+               uint type, uint nr, uint d0_flags)
+{
+   uint reg = UREG(type, nr);
+
+   if (type == REG_TYPE_T) {
+      if (p->decl_t & (1 << nr))
+         return reg;
+
+      p->decl_t |= (1 << nr);
+   }
+   else if (type == REG_TYPE_S) {
+      if (p->decl_s & (1 << nr))
+         return reg;
+
+      p->decl_s |= (1 << nr);
+   }
+   else
+      return reg;
+
+   *(p->decl++) = (D0_DCL | D0_DEST(reg) | d0_flags);
+   *(p->decl++) = D1_MBZ;
+   *(p->decl++) = D2_MBZ;
+
+   p->nr_decl_insn++;
+   return reg;
+}
+
+uint
+i915_emit_arith(struct i915_fp_compile * p,
+                uint op,
+                uint dest,
+                uint mask,
+                uint saturate, uint src0, uint src1, uint src2)
+{
+   uint c[3];
+   uint nr_const = 0;
+
+   assert(GET_UREG_TYPE(dest) != REG_TYPE_CONST);
+   dest = UREG(GET_UREG_TYPE(dest), GET_UREG_NR(dest));
+   assert(dest);
+
+   if (GET_UREG_TYPE(src0) == REG_TYPE_CONST)
+      c[nr_const++] = 0;
+   if (GET_UREG_TYPE(src1) == REG_TYPE_CONST)
+      c[nr_const++] = 1;
+   if (GET_UREG_TYPE(src2) == REG_TYPE_CONST)
+      c[nr_const++] = 2;
+
+   /* Recursively call this function to MOV additional const values
+    * into temporary registers.  Use utemp registers for this -
+    * currently shouldn't be possible to run out, but keep an eye on
+    * this.
+    */
+   if (nr_const > 1) {
+      uint s[3], first, i, old_utemp_flag;
+
+      s[0] = src0;
+      s[1] = src1;
+      s[2] = src2;
+      old_utemp_flag = p->utemp_flag;
+
+      first = GET_UREG_NR(s[c[0]]);
+      for (i = 1; i < nr_const; i++) {
+         if (GET_UREG_NR(s[c[i]]) != first) {
+            uint tmp = i915_get_utemp(p);
+
+            i915_emit_arith(p, A0_MOV, tmp, A0_DEST_CHANNEL_ALL, 0,
+                            s[c[i]], 0, 0);
+            s[c[i]] = tmp;
+         }
+      }
+
+      src0 = s[0];
+      src1 = s[1];
+      src2 = s[2];
+      p->utemp_flag = old_utemp_flag;   /* restore */
+   }
+
+   *(p->csr++) = (op | A0_DEST(dest) | mask | saturate | A0_SRC0(src0));
+   *(p->csr++) = (A1_SRC0(src0) | A1_SRC1(src1));
+   *(p->csr++) = (A2_SRC1(src1) | A2_SRC2(src2));
+
+   p->nr_alu_insn++;
+   return dest;
+}
+
+uint i915_emit_texld( struct i915_fp_compile *p,
+			uint dest,
+			uint destmask,
+			uint sampler,
+			uint coord,
+			uint op )
+{
+   uint k = UREG(GET_UREG_TYPE(coord), GET_UREG_NR(coord));
+   if (coord != k) {
+      /* No real way to work around this in the general case - need to
+       * allocate and declare a new temporary register (a utemp won't
+       * do).  Will fallback for now.
+       */
+      i915_program_error(p, "Can't (yet) swizzle TEX arguments");
+      assert(0);
+      return 0;
+   }
+
+   /* Don't worry about saturate as we only support  
+    */
+   if (destmask != A0_DEST_CHANNEL_ALL) {
+      uint tmp = i915_get_utemp(p);
+      i915_emit_texld( p, tmp, A0_DEST_CHANNEL_ALL, sampler, coord, op );
+      i915_emit_arith( p, A0_MOV, dest, destmask, 0, tmp, 0, 0 );
+      return dest;
+   }
+   else {
+      assert(GET_UREG_TYPE(dest) != REG_TYPE_CONST);
+      assert(dest = UREG(GET_UREG_TYPE(dest), GET_UREG_NR(dest)));
+
+      if (GET_UREG_TYPE(coord) != REG_TYPE_T) {
+	 p->nr_tex_indirect++;
+      }
+
+      *(p->csr++) = (op | 
+		     T0_DEST( dest ) |
+		     T0_SAMPLER( sampler ));
+
+      *(p->csr++) = T1_ADDRESS_REG( coord );
+      *(p->csr++) = T2_MBZ;
+
+      p->nr_tex_insn++;
+      return dest;
+   }
+}
+
+
+uint
+i915_emit_const1f(struct i915_fp_compile * p, float c0)
+{
+   unsigned reg, idx;
+
+   if (c0 == 0.0)
+      return swizzle(UREG(REG_TYPE_R, 0), ZERO, ZERO, ZERO, ZERO);
+   if (c0 == 1.0)
+      return swizzle(UREG(REG_TYPE_R, 0), ONE, ONE, ONE, ONE);
+
+   for (reg = 0; reg < I915_MAX_CONSTANT; reg++) {
+      if (p->constant_flags[reg] == I915_CONSTFLAG_PARAM)
+         continue;
+      for (idx = 0; idx < 4; idx++) {
+         if (!(p->constant_flags[reg] & (1 << idx)) ||
+             p->constants[reg][idx] == c0) {
+            p->constants[reg][idx] = c0;
+            p->constant_flags[reg] |= 1 << idx;
+            if (reg + 1 > p->num_constants)
+               p->num_constants = reg + 1;
+            return swizzle(UREG(REG_TYPE_CONST, reg), idx, ZERO, ZERO, ONE);
+         }
+      }
+   }
+
+   i915_program_error(p, "i915_emit_const1f: out of constants\n");
+   return 0;
+}
+
+uint
+i915_emit_const2f(struct i915_fp_compile * p, float c0, float c1)
+{
+   unsigned reg, idx;
+
+   if (c0 == 0.0)
+      return swizzle(i915_emit_const1f(p, c1), ZERO, X, Z, W);
+   if (c0 == 1.0)
+      return swizzle(i915_emit_const1f(p, c1), ONE, X, Z, W);
+
+   if (c1 == 0.0)
+      return swizzle(i915_emit_const1f(p, c0), X, ZERO, Z, W);
+   if (c1 == 1.0)
+      return swizzle(i915_emit_const1f(p, c0), X, ONE, Z, W);
+
+   for (reg = 0; reg < I915_MAX_CONSTANT; reg++) {
+      if (p->constant_flags[reg] == 0xf ||
+          p->constant_flags[reg] == I915_CONSTFLAG_PARAM)
+         continue;
+      for (idx = 0; idx < 3; idx++) {
+         if (!(p->constant_flags[reg] & (3 << idx))) {
+            p->constants[reg][idx + 0] = c0;
+            p->constants[reg][idx + 1] = c1;
+            p->constant_flags[reg] |= 3 << idx;
+            if (reg + 1 > p->num_constants)
+               p->num_constants = reg + 1;
+            return swizzle(UREG(REG_TYPE_CONST, reg), idx, idx + 1, ZERO, ONE);
+         }
+      }
+   }
+
+   i915_program_error(p, "i915_emit_const2f: out of constants\n");
+   return 0;
+}
+
+
+
+uint
+i915_emit_const4f(struct i915_fp_compile * p,
+                  float c0, float c1, float c2, float c3)
+{
+   unsigned reg;
+
+   for (reg = 0; reg < I915_MAX_CONSTANT; reg++) {
+      if (p->constant_flags[reg] == 0xf &&
+          p->constants[reg][0] == c0 &&
+          p->constants[reg][1] == c1 &&
+          p->constants[reg][2] == c2 &&
+          p->constants[reg][3] == c3) {
+         return UREG(REG_TYPE_CONST, reg);
+      }
+      else if (p->constant_flags[reg] == 0) {
+
+         p->constants[reg][0] = c0;
+         p->constants[reg][1] = c1;
+         p->constants[reg][2] = c2;
+         p->constants[reg][3] = c3;
+         p->constant_flags[reg] = 0xf;
+         if (reg + 1 > p->num_constants)
+            p->num_constants = reg + 1;
+         return UREG(REG_TYPE_CONST, reg);
+      }
+   }
+
+   i915_program_error(p, "i915_emit_const4f: out of constants\n");
+   return 0;
+}
+
+
+uint
+i915_emit_const4fv(struct i915_fp_compile * p, const float * c)
+{
+   return i915_emit_const4f(p, c[0], c[1], c[2], c[3]);
+}
+
+
+#if 00000/*UNUSED*/
+/* Reserve a slot in the constant file for a Mesa state parameter.
+ * These will later need to be tracked on statechanges, but that is
+ * done elsewhere.
+ */
+uint
+i915_emit_param4fv(struct i915_fp_compile * p, const float * values)
+{
+   struct i915_fragment_program *fp = p->fp;
+   int i;
+
+   for (i = 0; i < fp->nr_params; i++) {
+      if (fp->param[i].values == values)
+         return UREG(REG_TYPE_CONST, fp->param[i].reg);
+   }
+
+   if (p->constants->nr_constants == I915_MAX_CONSTANT ||
+       fp->nr_params == I915_MAX_CONSTANT) {
+      i915_program_error(p, "i915_emit_param4fv: out of constants\n");
+      return 0;
+   }
+
+   {
+      int reg = p->constants->nr_constants++;
+      int i = fp->nr_params++;
+
+      assert (p->constant_flags[reg] == 0);
+      p->constant_flags[reg] = I915_CONSTFLAG_PARAM;
+
+      fp->param[i].values = values;
+      fp->param[i].reg = reg;
+
+      return UREG(REG_TYPE_CONST, reg);
+   }
+}
+#endif
diff --git a/src/gallium/drivers/i915simple/i915_fpc_translate.c b/src/gallium/drivers/i915simple/i915_fpc_translate.c
new file mode 100644
index 0000000000..868f0c7e04
--- /dev/null
+++ b/src/gallium/drivers/i915simple/i915_fpc_translate.c
@@ -0,0 +1,1135 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+
+#include <stdarg.h>
+
+#include "i915_reg.h"
+#include "i915_context.h"
+#include "i915_fpc.h"
+
+#include "pipe/p_shader_tokens.h"
+#include "pipe/tgsi/util/tgsi_parse.h"
+
+#include "pipe/draw/draw_vertex.h"
+
+
+/**
+ * Simple pass-through fragment shader to use when we don't have
+ * a real shader (or it fails to compile for some reason).
+ */
+static unsigned passthrough[] = 
+{
+   _3DSTATE_PIXEL_SHADER_PROGRAM | ((2*3)-1),
+
+   /* declare input color:
+    */
+   (D0_DCL | 
+    (REG_TYPE_T << D0_TYPE_SHIFT) | 
+    (T_DIFFUSE << D0_NR_SHIFT) | 
+    D0_CHANNEL_ALL),
+   0,
+   0,
+
+   /* move to output color:
+    */
+   (A0_MOV | 
+    (REG_TYPE_OC << A0_DEST_TYPE_SHIFT) | 
+    A0_DEST_CHANNEL_ALL | 
+    (REG_TYPE_T << A0_SRC0_TYPE_SHIFT) |
+    (T_DIFFUSE << A0_SRC0_NR_SHIFT)),
+   0x01230000,			/* .xyzw */
+   0
+};
+
+
+/* 1, -1/3!, 1/5!, -1/7! */
+static const float sin_constants[4] = { 1.0,
+   -1.0f / (3 * 2 * 1),
+   1.0f / (5 * 4 * 3 * 2 * 1),
+   -1.0f / (7 * 6 * 5 * 4 * 3 * 2 * 1)
+};
+
+/* 1, -1/2!, 1/4!, -1/6! */
+static const float cos_constants[4] = { 1.0,
+   -1.0f / (2 * 1),
+   1.0f / (4 * 3 * 2 * 1),
+   -1.0f / (6 * 5 * 4 * 3 * 2 * 1)
+};
+
+
+
+/**
+ * component-wise negation of ureg
+ */
+static INLINE int
+negate(int reg, int x, int y, int z, int w)
+{
+   /* Another neat thing about the UREG representation */
+   return reg ^ (((x & 1) << UREG_CHANNEL_X_NEGATE_SHIFT) |
+                 ((y & 1) << UREG_CHANNEL_Y_NEGATE_SHIFT) |
+                 ((z & 1) << UREG_CHANNEL_Z_NEGATE_SHIFT) |
+                 ((w & 1) << UREG_CHANNEL_W_NEGATE_SHIFT));
+}
+
+
+static void
+i915_use_passthrough_shader(struct i915_context *i915)
+{
+   debug_printf("**** Using i915 pass-through fragment shader\n");
+
+   i915->current.program = (uint *) MALLOC(sizeof(passthrough));
+   if (i915->current.program) {
+      memcpy(i915->current.program, passthrough, sizeof(passthrough));
+      i915->current.program_len = Elements(passthrough);
+   }
+
+   i915->current.num_constants[PIPE_SHADER_FRAGMENT] = 0;
+   i915->current.num_user_constants[PIPE_SHADER_FRAGMENT] = 0;
+}
+
+
+void
+i915_program_error(struct i915_fp_compile *p, const char *msg, ...)
+{
+   va_list args;
+   char buffer[1024];
+
+   debug_printf("i915_program_error: ");
+   va_start( args, msg );  
+   vsprintf( buffer, msg, args );
+   va_end( args );
+   debug_printf(buffer);
+   debug_printf("\n");
+
+   p->error = 1;
+}
+
+
+
+/**
+ * Construct a ureg for the given source register.  Will emit
+ * constants, apply swizzling and negation as needed.
+ */
+static uint
+src_vector(struct i915_fp_compile *p,
+           const struct tgsi_full_src_register *source)
+{
+   uint index = source->SrcRegister.Index;
+   uint src, sem_name, sem_ind;
+
+   switch (source->SrcRegister.File) {
+   case TGSI_FILE_TEMPORARY:
+      if (source->SrcRegister.Index >= I915_MAX_TEMPORARY) {
+         i915_program_error(p, "Exceeded max temporary reg");
+         return 0;
+      }
+      src = UREG(REG_TYPE_R, index);
+      break;
+   case TGSI_FILE_INPUT:
+      /* XXX: Packing COL1, FOGC into a single attribute works for
+       * texenv programs, but will fail for real fragment programs
+       * that use these attributes and expect them to be a full 4
+       * components wide.  Could use a texcoord to pass these
+       * attributes if necessary, but that won't work in the general
+       * case.
+       * 
+       * We also use a texture coordinate to pass wpos when possible.
+       */
+
+      /* use vertex format info to map a slot number to a VF attrib */
+      assert(index < p->vertex_info->num_attribs);
+
+      sem_name = p->input_semantic_name[index];
+      sem_ind = p->input_semantic_index[index];
+
+      switch (sem_name) {
+      case TGSI_SEMANTIC_POSITION:
+         debug_printf("SKIP SEM POS\n");
+         /*
+         assert(p->wpos_tex != -1);
+         src = i915_emit_decl(p, REG_TYPE_T, p->wpos_tex, D0_CHANNEL_ALL);
+         */
+         break;
+      case TGSI_SEMANTIC_COLOR:
+         if (sem_ind == 0) {
+            src = i915_emit_decl(p, REG_TYPE_T, T_DIFFUSE, D0_CHANNEL_ALL);
+         }
+         else {
+            /* secondary color */
+            assert(sem_ind == 1);
+            src = i915_emit_decl(p, REG_TYPE_T, T_SPECULAR, D0_CHANNEL_XYZ);
+            src = swizzle(src, X, Y, Z, ONE);
+         }
+         break;
+      case TGSI_SEMANTIC_FOG:
+         src = i915_emit_decl(p, REG_TYPE_T, T_FOG_W, D0_CHANNEL_W);
+         src = swizzle(src, W, W, W, W);
+         break;
+      case TGSI_SEMANTIC_GENERIC:
+         /* usually a texcoord */
+         src = i915_emit_decl(p, REG_TYPE_T, T_TEX0 + sem_ind, D0_CHANNEL_ALL);
+         break;
+      default:
+         i915_program_error(p, "Bad source->Index");
+         return 0;
+      }
+      break;
+
+   case TGSI_FILE_IMMEDIATE:
+      /* XXX unfinished - need to append immediates onto const buffer */
+      /* fall-through */
+   case TGSI_FILE_CONSTANT:
+      src = UREG(REG_TYPE_CONST, index);
+      break;
+
+   default:
+      i915_program_error(p, "Bad source->File");
+      return 0;
+   }
+
+   if (source->SrcRegister.Extended) {
+      src = swizzle(src,
+                    source->SrcRegisterExtSwz.ExtSwizzleX,
+                    source->SrcRegisterExtSwz.ExtSwizzleY,
+                    source->SrcRegisterExtSwz.ExtSwizzleZ,
+                    source->SrcRegisterExtSwz.ExtSwizzleW);
+   }
+   else {
+      src = swizzle(src,
+                    source->SrcRegister.SwizzleX,
+                    source->SrcRegister.SwizzleY,
+                    source->SrcRegister.SwizzleZ,
+                    source->SrcRegister.SwizzleW);
+   }
+
+
+   /* There's both negate-all-components and per-component negation.
+    * Try to handle both here.
+    */
+   {
+      int nx = source->SrcRegisterExtSwz.NegateX;
+      int ny = source->SrcRegisterExtSwz.NegateY;
+      int nz = source->SrcRegisterExtSwz.NegateZ;
+      int nw = source->SrcRegisterExtSwz.NegateW;
+      if (source->SrcRegister.Negate) {
+         nx = !nx;
+         ny = !ny;
+         nz = !nz;
+         nw = !nw;
+      }
+      src = negate(src, nx, ny, nz, nw);
+   }
+
+   /* no abs() or post-abs negation */
+#if 0
+   /* XXX assertions disabled to allow arbfplight.c to run */
+   /* XXX enable these assertions, or fix things */
+   assert(!source->SrcRegisterExtMod.Absolute);
+   assert(!source->SrcRegisterExtMod.Negate);
+#endif
+   return src;
+}
+
+
+/**
+ * Construct a ureg for a destination register.
+ */
+static uint
+get_result_vector(struct i915_fp_compile *p,
+                  const struct tgsi_full_dst_register *dest)
+{
+   switch (dest->DstRegister.File) {
+   case TGSI_FILE_OUTPUT:
+      {
+         uint sem_name = p->output_semantic_name[dest->DstRegister.Index];
+         switch (sem_name) {
+         case TGSI_SEMANTIC_POSITION:
+            return UREG(REG_TYPE_OD, 0);
+         case TGSI_SEMANTIC_COLOR:
+            return UREG(REG_TYPE_OC, 0);
+         default:
+            i915_program_error(p, "Bad inst->DstReg.Index/semantics");
+            return 0;
+         }
+      }
+   case TGSI_FILE_TEMPORARY:
+      return UREG(REG_TYPE_R, dest->DstRegister.Index);
+   default:
+      i915_program_error(p, "Bad inst->DstReg.File");
+      return 0;
+   }
+}
+
+
+/**
+ * Compute flags for saturation and writemask.
+ */
+static uint
+get_result_flags(const struct tgsi_full_instruction *inst)
+{
+   const uint writeMask
+      = inst->FullDstRegisters[0].DstRegister.WriteMask;
+   uint flags = 0x0;
+
+   if (inst->Instruction.Saturate == TGSI_SAT_ZERO_ONE)
+      flags |= A0_DEST_SATURATE;
+
+   if (writeMask & TGSI_WRITEMASK_X)
+      flags |= A0_DEST_CHANNEL_X;
+   if (writeMask & TGSI_WRITEMASK_Y)
+      flags |= A0_DEST_CHANNEL_Y;
+   if (writeMask & TGSI_WRITEMASK_Z)
+      flags |= A0_DEST_CHANNEL_Z;
+   if (writeMask & TGSI_WRITEMASK_W)
+      flags |= A0_DEST_CHANNEL_W;
+
+   return flags;
+}
+
+
+/**
+ * Convert TGSI_TEXTURE_x token to DO_SAMPLE_TYPE_x token
+ */
+static uint
+translate_tex_src_target(struct i915_fp_compile *p, uint tex)
+{
+   switch (tex) {
+   case TGSI_TEXTURE_1D:
+      return D0_SAMPLE_TYPE_2D;
+   case TGSI_TEXTURE_2D:
+      return D0_SAMPLE_TYPE_2D;
+   case TGSI_TEXTURE_RECT:
+      return D0_SAMPLE_TYPE_2D;
+   case TGSI_TEXTURE_3D:
+      return D0_SAMPLE_TYPE_VOLUME;
+   case TGSI_TEXTURE_CUBE:
+      return D0_SAMPLE_TYPE_CUBE;
+   default:
+      i915_program_error(p, "TexSrc type");
+      return 0;
+   }
+}
+
+
+/**
+ * Generate texel lookup instruction.
+ */
+static void
+emit_tex(struct i915_fp_compile *p,
+         const struct tgsi_full_instruction *inst,
+         uint opcode)
+{
+   uint texture = inst->InstructionExtTexture.Texture;
+   uint unit = inst->FullSrcRegisters[1].SrcRegister.Index;
+   uint tex = translate_tex_src_target( p, texture );
+   uint sampler = i915_emit_decl(p, REG_TYPE_S, unit, tex);
+   uint coord = src_vector( p, &inst->FullSrcRegisters[0]);
+
+   i915_emit_texld( p,
+                    get_result_vector( p, &inst->FullDstRegisters[0] ),
+                    get_result_flags( inst ),
+                    sampler,
+                    coord,
+                    opcode);
+}
+
+
+/**
+ * Generate a simple arithmetic instruction
+ * \param opcode  the i915 opcode
+ * \param numArgs  the number of input/src arguments
+ */
+static void
+emit_simple_arith(struct i915_fp_compile *p,
+                  const struct tgsi_full_instruction *inst,
+                  uint opcode, uint numArgs)
+{
+   uint arg1, arg2, arg3;
+
+   assert(numArgs <= 3);
+
+   arg1 = (numArgs < 1) ? 0 : src_vector( p, &inst->FullSrcRegisters[0] );
+   arg2 = (numArgs < 2) ? 0 : src_vector( p, &inst->FullSrcRegisters[1] );
+   arg3 = (numArgs < 3) ? 0 : src_vector( p, &inst->FullSrcRegisters[2] );
+
+   i915_emit_arith( p,
+                    opcode,
+                    get_result_vector( p, &inst->FullDstRegisters[0]),
+                    get_result_flags( inst ), 0,
+                    arg1,
+                    arg2,
+                    arg3 );
+}
+
+#ifndef M_PI
+#define M_PI 3.14159265358979323846
+#endif
+
+/*
+ * Translate TGSI instruction to i915 instruction.
+ *
+ * Possible concerns:
+ *
+ * SIN, COS -- could use another taylor step?
+ * LIT      -- results seem a little different to sw mesa
+ * LOG      -- different to mesa on negative numbers, but this is conformant.
+ */ 
+static void
+i915_translate_instruction(struct i915_fp_compile *p,
+                           const struct tgsi_full_instruction *inst)
+{
+   uint writemask;
+   uint src0, src1, src2, flags;
+   uint tmp = 0;
+
+   switch (inst->Instruction.Opcode) {
+   case TGSI_OPCODE_ABS:
+      src0 = src_vector(p, &inst->FullSrcRegisters[0]);
+      i915_emit_arith(p,
+                      A0_MAX,
+                      get_result_vector(p, &inst->FullDstRegisters[0]),
+                      get_result_flags(inst), 0,
+                      src0, negate(src0, 1, 1, 1, 1), 0);
+      break;
+
+   case TGSI_OPCODE_ADD:
+      emit_simple_arith(p, inst, A0_ADD, 2);
+      break;
+
+   case TGSI_OPCODE_CMP:
+      src0 = src_vector(p, &inst->FullSrcRegisters[0]);
+      src1 = src_vector(p, &inst->FullSrcRegisters[1]);
+      src2 = src_vector(p, &inst->FullSrcRegisters[2]);
+      i915_emit_arith(p, A0_CMP, 
+                      get_result_vector(p, &inst->FullDstRegisters[0]),
+                      get_result_flags(inst), 
+                      0, src0, src2, src1);   /* NOTE: order of src2, src1 */
+      break;
+
+   case TGSI_OPCODE_COS:
+      src0 = src_vector(p, &inst->FullSrcRegisters[0]);
+      tmp = i915_get_utemp(p);
+
+      i915_emit_arith(p,
+                      A0_MUL,
+                      tmp, A0_DEST_CHANNEL_X, 0,
+                      src0, i915_emit_const1f(p, 1.0f / (float) (M_PI * 2.0)), 0);
+
+      i915_emit_arith(p, A0_MOD, tmp, A0_DEST_CHANNEL_X, 0, tmp, 0, 0);
+
+      /* By choosing different taylor constants, could get rid of this mul:
+       */
+      i915_emit_arith(p,
+                      A0_MUL,
+                      tmp, A0_DEST_CHANNEL_X, 0,
+                      tmp, i915_emit_const1f(p, (float) (M_PI * 2.0)), 0);
+
+      /* 
+       * t0.xy = MUL x.xx11, x.x1111  ; x^2, x, 1, 1
+       * t0 = MUL t0.xyxy t0.xx11 ; x^4, x^3, x^2, 1
+       * t0 = MUL t0.xxz1 t0.z111    ; x^6 x^4 x^2 1
+       * result = DP4 t0, cos_constants
+       */
+      i915_emit_arith(p,
+                      A0_MUL,
+                      tmp, A0_DEST_CHANNEL_XY, 0,
+                      swizzle(tmp, X, X, ONE, ONE),
+                      swizzle(tmp, X, ONE, ONE, ONE), 0);
+
+      i915_emit_arith(p,
+                      A0_MUL,
+                      tmp, A0_DEST_CHANNEL_XYZ, 0,
+                      swizzle(tmp, X, Y, X, ONE),
+                      swizzle(tmp, X, X, ONE, ONE), 0);
+
+      i915_emit_arith(p,
+                      A0_MUL,
+                      tmp, A0_DEST_CHANNEL_XYZ, 0,
+                      swizzle(tmp, X, X, Z, ONE),
+                      swizzle(tmp, Z, ONE, ONE, ONE), 0);
+
+      i915_emit_arith(p,
+                      A0_DP4,
+                      get_result_vector(p, &inst->FullDstRegisters[0]),
+                      get_result_flags(inst), 0,
+                      swizzle(tmp, ONE, Z, Y, X),
+                      i915_emit_const4fv(p, cos_constants), 0);
+      break;
+
+   case TGSI_OPCODE_DP3:
+      emit_simple_arith(p, inst, A0_DP3, 2);
+      break;
+
+   case TGSI_OPCODE_DP4:
+      emit_simple_arith(p, inst, A0_DP4, 2);
+      break;
+
+   case TGSI_OPCODE_DPH:
+      src0 = src_vector(p, &inst->FullSrcRegisters[0]);
+      src1 = src_vector(p, &inst->FullSrcRegisters[1]);
+
+      i915_emit_arith(p,
+                      A0_DP4,
+                      get_result_vector(p, &inst->FullDstRegisters[0]),
+                      get_result_flags(inst), 0,
+                      swizzle(src0, X, Y, Z, ONE), src1, 0);
+      break;
+
+   case TGSI_OPCODE_DST:
+      src0 = src_vector(p, &inst->FullSrcRegisters[0]);
+      src1 = src_vector(p, &inst->FullSrcRegisters[1]);
+
+      /* result[0] = 1    * 1;
+       * result[1] = a[1] * b[1];
+       * result[2] = a[2] * 1;
+       * result[3] = 1    * b[3];
+       */
+      i915_emit_arith(p,
+                      A0_MUL,
+                      get_result_vector(p, &inst->FullDstRegisters[0]),
+                      get_result_flags(inst), 0,
+                      swizzle(src0, ONE, Y, Z, ONE),
+                      swizzle(src1, ONE, Y, ONE, W), 0);
+      break;
+
+   case TGSI_OPCODE_END:
+      /* no-op */
+      break;
+
+   case TGSI_OPCODE_EX2:
+      src0 = src_vector(p, &inst->FullSrcRegisters[0]);
+
+      i915_emit_arith(p,
+                      A0_EXP,
+                      get_result_vector(p, &inst->FullDstRegisters[0]),
+                      get_result_flags(inst), 0,
+                      swizzle(src0, X, X, X, X), 0, 0);
+      break;
+
+   case TGSI_OPCODE_FLR:
+      emit_simple_arith(p, inst, A0_FLR, 1);
+      break;
+
+   case TGSI_OPCODE_FRC:
+      emit_simple_arith(p, inst, A0_FRC, 1);
+      break;
+
+   case TGSI_OPCODE_KIL:
+      /* unconditional kill */
+      assert(0); /* not tested yet */
+#if 0
+      src0 = src_vector(p, &inst->FullSrcRegisters[0]);
+      tmp = i915_get_utemp(p);
+
+      i915_emit_texld(p, tmp, A0_DEST_CHANNEL_ALL,   /* use a dummy dest reg */
+                      0, src0, T0_TEXKILL);
+#endif
+      break;
+
+   case TGSI_OPCODE_KILP:
+      /* kill if src[0].x < 0 || src[0].y < 0 ... */
+      src0 = src_vector(p, &inst->FullSrcRegisters[0]);
+      tmp = i915_get_utemp(p);
+
+      i915_emit_texld(p, tmp, A0_DEST_CHANNEL_ALL,   /* use a dummy dest reg */
+                      0, src0, T0_TEXKILL);
+      break;
+
+   case TGSI_OPCODE_LG2:
+      src0 = src_vector(p, &inst->FullSrcRegisters[0]);
+
+      i915_emit_arith(p,
+                      A0_LOG,
+                      get_result_vector(p, &inst->FullDstRegisters[0]),
+                      get_result_flags(inst), 0,
+                      swizzle(src0, X, X, X, X), 0, 0);
+      break;
+
+   case TGSI_OPCODE_LIT:
+      src0 = src_vector(p, &inst->FullSrcRegisters[0]);
+      tmp = i915_get_utemp(p);
+
+      /* tmp = max( a.xyzw, a.00zw )
+       * XXX: Clamp tmp.w to -128..128
+       * tmp.y = log(tmp.y)
+       * tmp.y = tmp.w * tmp.y
+       * tmp.y = exp(tmp.y)
+       * result = cmp (a.11-x1, a.1x01, a.1xy1 )
+       */
+      i915_emit_arith(p, A0_MAX, tmp, A0_DEST_CHANNEL_ALL, 0,
+                      src0, swizzle(src0, ZERO, ZERO, Z, W), 0);
+
+      i915_emit_arith(p, A0_LOG, tmp, A0_DEST_CHANNEL_Y, 0,
+                      swizzle(tmp, Y, Y, Y, Y), 0, 0);
+
+      i915_emit_arith(p, A0_MUL, tmp, A0_DEST_CHANNEL_Y, 0,
+                      swizzle(tmp, ZERO, Y, ZERO, ZERO),
+                      swizzle(tmp, ZERO, W, ZERO, ZERO), 0);
+
+      i915_emit_arith(p, A0_EXP, tmp, A0_DEST_CHANNEL_Y, 0,
+                      swizzle(tmp, Y, Y, Y, Y), 0, 0);
+
+      i915_emit_arith(p, A0_CMP,
+                      get_result_vector(p, &inst->FullDstRegisters[0]),
+                      get_result_flags(inst), 0,
+                      negate(swizzle(tmp, ONE, ONE, X, ONE), 0, 0, 1, 0),
+                      swizzle(tmp, ONE, X, ZERO, ONE),
+                      swizzle(tmp, ONE, X, Y, ONE));
+
+      break;
+
+   case TGSI_OPCODE_LRP:
+      src0 = src_vector(p, &inst->FullSrcRegisters[0]);
+      src1 = src_vector(p, &inst->FullSrcRegisters[1]);
+      src2 = src_vector(p, &inst->FullSrcRegisters[2]);
+      flags = get_result_flags(inst);
+      tmp = i915_get_utemp(p);
+
+      /* b*a + c*(1-a)
+       *
+       * b*a + c - ca 
+       *
+       * tmp = b*a + c, 
+       * result = (-c)*a + tmp 
+       */
+      i915_emit_arith(p, A0_MAD, tmp,
+                      flags & A0_DEST_CHANNEL_ALL, 0, src1, src0, src2);
+
+      i915_emit_arith(p, A0_MAD,
+                      get_result_vector(p, &inst->FullDstRegisters[0]),
+                      flags, 0, negate(src2, 1, 1, 1, 1), src0, tmp);
+      break;
+
+   case TGSI_OPCODE_MAD:
+      emit_simple_arith(p, inst, A0_MAD, 3);
+      break;
+
+   case TGSI_OPCODE_MAX:
+      emit_simple_arith(p, inst, A0_MAX, 2);
+      break;
+
+   case TGSI_OPCODE_MIN:
+      src0 = src_vector(p, &inst->FullSrcRegisters[0]);
+      src1 = src_vector(p, &inst->FullSrcRegisters[1]);
+      tmp = i915_get_utemp(p);
+      flags = get_result_flags(inst);
+
+      i915_emit_arith(p,
+                      A0_MAX,
+                      tmp, flags & A0_DEST_CHANNEL_ALL, 0,
+                      negate(src0, 1, 1, 1, 1),
+                      negate(src1, 1, 1, 1, 1), 0);
+
+      i915_emit_arith(p,
+                      A0_MOV,
+                      get_result_vector(p, &inst->FullDstRegisters[0]),
+                      flags, 0, negate(tmp, 1, 1, 1, 1), 0, 0);
+      break;
+
+   case TGSI_OPCODE_MOV:
+      /* aka TGSI_OPCODE_SWZ */
+      emit_simple_arith(p, inst, A0_MOV, 1);
+      break;
+
+   case TGSI_OPCODE_MUL:
+      emit_simple_arith(p, inst, A0_MUL, 2);
+      break;
+
+   case TGSI_OPCODE_POW:
+      src0 = src_vector(p, &inst->FullSrcRegisters[0]);
+      src1 = src_vector(p, &inst->FullSrcRegisters[1]);
+      tmp = i915_get_utemp(p);
+      flags = get_result_flags(inst);
+
+      /* XXX: masking on intermediate values, here and elsewhere.
+       */
+      i915_emit_arith(p,
+                      A0_LOG,
+                      tmp, A0_DEST_CHANNEL_X, 0,
+                      swizzle(src0, X, X, X, X), 0, 0);
+
+      i915_emit_arith(p, A0_MUL, tmp, A0_DEST_CHANNEL_X, 0, tmp, src1, 0);
+
+      i915_emit_arith(p,
+                      A0_EXP,
+                      get_result_vector(p, &inst->FullDstRegisters[0]),
+                      flags, 0, swizzle(tmp, X, X, X, X), 0, 0);
+      break;
+      
+   case TGSI_OPCODE_RET:
+      /* XXX: no-op? */
+      break;
+      
+   case TGSI_OPCODE_RCP:
+      src0 = src_vector(p, &inst->FullSrcRegisters[0]);
+
+      i915_emit_arith(p,
+                      A0_RCP,
+                      get_result_vector(p, &inst->FullDstRegisters[0]),
+                         get_result_flags(inst), 0,
+                      swizzle(src0, X, X, X, X), 0, 0);
+      break;
+
+   case TGSI_OPCODE_RSQ:
+      src0 = src_vector(p, &inst->FullSrcRegisters[0]);
+
+      i915_emit_arith(p,
+                      A0_RSQ,
+                      get_result_vector(p, &inst->FullDstRegisters[0]),
+                      get_result_flags(inst), 0,
+                      swizzle(src0, X, X, X, X), 0, 0);
+      break;
+
+   case TGSI_OPCODE_SCS:
+      src0 = src_vector(p, &inst->FullSrcRegisters[0]);
+      tmp = i915_get_utemp(p);
+
+      /* 
+       * t0.xy = MUL x.xx11, x.x1111  ; x^2, x, 1, 1
+       * t0 = MUL t0.xyxy t0.xx11 ; x^4, x^3, x^2, x
+       * t1 = MUL t0.xyyw t0.yz11    ; x^7 x^5 x^3 x
+       * scs.x = DP4 t1, sin_constants
+       * t1 = MUL t0.xxz1 t0.z111    ; x^6 x^4 x^2 1
+       * scs.y = DP4 t1, cos_constants
+       */
+      i915_emit_arith(p,
+                      A0_MUL,
+                      tmp, A0_DEST_CHANNEL_XY, 0,
+                      swizzle(src0, X, X, ONE, ONE),
+                      swizzle(src0, X, ONE, ONE, ONE), 0);
+
+      i915_emit_arith(p,
+                      A0_MUL,
+                      tmp, A0_DEST_CHANNEL_ALL, 0,
+                      swizzle(tmp, X, Y, X, Y),
+                      swizzle(tmp, X, X, ONE, ONE), 0);
+
+      writemask = inst->FullDstRegisters[0].DstRegister.WriteMask;
+
+      if (writemask & TGSI_WRITEMASK_Y) {
+         uint tmp1;
+
+         if (writemask & TGSI_WRITEMASK_X)
+            tmp1 = i915_get_utemp(p);
+         else
+            tmp1 = tmp;
+
+         i915_emit_arith(p,
+                         A0_MUL,
+                         tmp1, A0_DEST_CHANNEL_ALL, 0,
+                         swizzle(tmp, X, Y, Y, W),
+                         swizzle(tmp, X, Z, ONE, ONE), 0);
+
+         i915_emit_arith(p,
+                         A0_DP4,
+                         get_result_vector(p, &inst->FullDstRegisters[0]),
+                         A0_DEST_CHANNEL_Y, 0,
+                         swizzle(tmp1, W, Z, Y, X),
+                         i915_emit_const4fv(p, sin_constants), 0);
+      }
+
+      if (writemask & TGSI_WRITEMASK_X) {
+         i915_emit_arith(p,
+                         A0_MUL,
+                         tmp, A0_DEST_CHANNEL_XYZ, 0,
+                         swizzle(tmp, X, X, Z, ONE),
+                         swizzle(tmp, Z, ONE, ONE, ONE), 0);
+
+         i915_emit_arith(p,
+                         A0_DP4,
+                         get_result_vector(p, &inst->FullDstRegisters[0]),
+                         A0_DEST_CHANNEL_X, 0,
+                         swizzle(tmp, ONE, Z, Y, X),
+                         i915_emit_const4fv(p, cos_constants), 0);
+      }
+      break;
+
+   case TGSI_OPCODE_SGE:
+      emit_simple_arith(p, inst, A0_SGE, 2);
+      break;
+
+   case TGSI_OPCODE_SIN:
+      src0 = src_vector(p, &inst->FullSrcRegisters[0]);
+      tmp = i915_get_utemp(p);
+
+      i915_emit_arith(p,
+                      A0_MUL,
+                      tmp, A0_DEST_CHANNEL_X, 0,
+                      src0, i915_emit_const1f(p, 1.0f / (float) (M_PI * 2.0)), 0);
+
+      i915_emit_arith(p, A0_MOD, tmp, A0_DEST_CHANNEL_X, 0, tmp, 0, 0);
+
+      /* By choosing different taylor constants, could get rid of this mul:
+       */
+      i915_emit_arith(p,
+                      A0_MUL,
+                      tmp, A0_DEST_CHANNEL_X, 0,
+                      tmp, i915_emit_const1f(p, (float) (M_PI * 2.0)), 0);
+
+      /* 
+       * t0.xy = MUL x.xx11, x.x1111  ; x^2, x, 1, 1
+       * t0 = MUL t0.xyxy t0.xx11 ; x^4, x^3, x^2, x
+       * t1 = MUL t0.xyyw t0.yz11    ; x^7 x^5 x^3 x
+       * result = DP4 t1.wzyx, sin_constants
+       */
+      i915_emit_arith(p,
+                      A0_MUL,
+                      tmp, A0_DEST_CHANNEL_XY, 0,
+                      swizzle(tmp, X, X, ONE, ONE),
+                      swizzle(tmp, X, ONE, ONE, ONE), 0);
+
+      i915_emit_arith(p,
+                      A0_MUL,
+                      tmp, A0_DEST_CHANNEL_ALL, 0,
+                      swizzle(tmp, X, Y, X, Y),
+                      swizzle(tmp, X, X, ONE, ONE), 0);
+
+      i915_emit_arith(p,
+                      A0_MUL,
+                      tmp, A0_DEST_CHANNEL_ALL, 0,
+                      swizzle(tmp, X, Y, Y, W),
+                      swizzle(tmp, X, Z, ONE, ONE), 0);
+
+      i915_emit_arith(p,
+                      A0_DP4,
+                      get_result_vector(p, &inst->FullDstRegisters[0]),
+                      get_result_flags(inst), 0,
+                      swizzle(tmp, W, Z, Y, X),
+                      i915_emit_const4fv(p, sin_constants), 0);
+      break;
+
+   case TGSI_OPCODE_SLT:
+      emit_simple_arith(p, inst, A0_SLT, 2);
+      break;
+
+   case TGSI_OPCODE_SUB:
+      src0 = src_vector(p, &inst->FullSrcRegisters[0]);
+      src1 = src_vector(p, &inst->FullSrcRegisters[1]);
+
+      i915_emit_arith(p,
+                      A0_ADD,
+                      get_result_vector(p, &inst->FullDstRegisters[0]),
+                      get_result_flags(inst), 0,
+                      src0, negate(src1, 1, 1, 1, 1), 0);
+      break;
+
+   case TGSI_OPCODE_TEX:
+      if (inst->FullSrcRegisters[0].SrcRegisterExtSwz.ExtDivide
+          == TGSI_EXTSWIZZLE_W) {
+         emit_tex(p, inst, T0_TEXLDP);
+      }
+      else {
+         emit_tex(p, inst, T0_TEXLD);
+      }
+      break;
+
+   case TGSI_OPCODE_TXB:
+      emit_tex(p, inst, T0_TEXLDB);
+      break;
+
+   case TGSI_OPCODE_XPD:
+      /* Cross product:
+       *      result.x = src0.y * src1.z - src0.z * src1.y;
+       *      result.y = src0.z * src1.x - src0.x * src1.z;
+       *      result.z = src0.x * src1.y - src0.y * src1.x;
+       *      result.w = undef;
+       */
+      src0 = src_vector(p, &inst->FullSrcRegisters[0]);
+      src1 = src_vector(p, &inst->FullSrcRegisters[1]);
+      tmp = i915_get_utemp(p);
+
+      i915_emit_arith(p,
+                      A0_MUL,
+                      tmp, A0_DEST_CHANNEL_ALL, 0,
+                      swizzle(src0, Z, X, Y, ONE),
+                      swizzle(src1, Y, Z, X, ONE), 0);
+
+      i915_emit_arith(p,
+                      A0_MAD,
+                      get_result_vector(p, &inst->FullDstRegisters[0]),
+                      get_result_flags(inst), 0,
+                      swizzle(src0, Y, Z, X, ONE),
+                      swizzle(src1, Z, X, Y, ONE),
+                      negate(tmp, 1, 1, 1, 0));
+      break;
+
+   default:
+      i915_program_error(p, "bad opcode %d", inst->Instruction.Opcode);
+      return;
+   }
+
+   i915_release_utemps(p);
+}
+
+
+/**
+ * Translate TGSI fragment shader into i915 hardware instructions.
+ * \param p  the translation state
+ * \param tokens  the TGSI token array
+ */
+static void
+i915_translate_instructions(struct i915_fp_compile *p,
+                            const struct tgsi_token *tokens)
+{
+   struct tgsi_parse_context parse;
+
+   tgsi_parse_init( &parse, tokens );
+
+   while( !tgsi_parse_end_of_tokens( &parse ) ) {
+
+      tgsi_parse_token( &parse );
+
+      switch( parse.FullToken.Token.Type ) {
+      case TGSI_TOKEN_TYPE_DECLARATION:
+         if (parse.FullToken.FullDeclaration.Declaration.File
+             == TGSI_FILE_INPUT) {
+            /* save input register info for use in src_vector() */
+            uint ind, sem, semi;
+            ind = parse.FullToken.FullDeclaration.u.DeclarationRange.First;
+            sem = parse.FullToken.FullDeclaration.Semantic.SemanticName;
+            semi = parse.FullToken.FullDeclaration.Semantic.SemanticIndex;
+            /*debug_printf("FS Input DECL [%u] sem %u\n", ind, sem);*/
+            p->input_semantic_name[ind] = sem;
+            p->input_semantic_index[ind] = semi;
+         }
+         else if (parse.FullToken.FullDeclaration.Declaration.File
+             == TGSI_FILE_OUTPUT) {
+            /* save output register info for use in get_result_vector() */
+            uint ind, sem, semi;
+            ind = parse.FullToken.FullDeclaration.u.DeclarationRange.First;
+            sem = parse.FullToken.FullDeclaration.Semantic.SemanticName;
+            semi = parse.FullToken.FullDeclaration.Semantic.SemanticIndex;
+            /*debug_printf("FS Output DECL [%u] sem %u\n", ind, sem);*/
+            p->output_semantic_name[ind] = sem;
+            p->output_semantic_index[ind] = semi;
+         }
+         break;
+
+      case TGSI_TOKEN_TYPE_IMMEDIATE:
+         /* XXX append the immediate to the const buffer... */
+         break;
+
+      case TGSI_TOKEN_TYPE_INSTRUCTION:
+         i915_translate_instruction(p, &parse.FullToken.FullInstruction);
+         break;
+
+      default:
+         assert( 0 );
+      }
+
+   } /* while */
+
+   tgsi_parse_free (&parse);
+}
+
+
+static struct i915_fp_compile *
+i915_init_compile(struct i915_context *i915,
+                  const struct pipe_shader_state *fs)
+{
+   struct i915_fp_compile *p = CALLOC_STRUCT(i915_fp_compile);
+
+   p->shader = i915->fs;
+
+   p->vertex_info = &i915->current.vertex_info;
+
+   /* new constants found during translation get appended after the
+    * user-provided constants.
+    */
+   p->constants = i915->current.constants[PIPE_SHADER_FRAGMENT];
+   p->num_constants = i915->current.num_user_constants[PIPE_SHADER_FRAGMENT];
+
+   p->nr_tex_indirect = 1;      /* correct? */
+   p->nr_tex_insn = 0;
+   p->nr_alu_insn = 0;
+   p->nr_decl_insn = 0;
+
+   memset(p->constant_flags, 0, sizeof(p->constant_flags));
+
+   p->csr = p->program;
+   p->decl = p->declarations;
+   p->decl_s = 0;
+   p->decl_t = 0;
+   p->temp_flag = 0xffff000;
+   p->utemp_flag = ~0x7;
+
+   p->wpos_tex = -1;
+
+   /* initialize the first program word */
+   *(p->decl++) = _3DSTATE_PIXEL_SHADER_PROGRAM;
+
+   return p;
+}
+
+
+/* Copy compile results to the fragment program struct and destroy the
+ * compilation context.
+ */
+static void
+i915_fini_compile(struct i915_context *i915, struct i915_fp_compile *p)
+{
+   unsigned long program_size = (unsigned long) (p->csr - p->program);
+   unsigned long decl_size = (unsigned long) (p->decl - p->declarations);
+
+   if (p->nr_tex_indirect > I915_MAX_TEX_INDIRECT)
+      i915_program_error(p, "Exceeded max nr indirect texture lookups");
+
+   if (p->nr_tex_insn > I915_MAX_TEX_INSN)
+      i915_program_error(p, "Exceeded max TEX instructions");
+
+   if (p->nr_alu_insn > I915_MAX_ALU_INSN)
+      i915_program_error(p, "Exceeded max ALU instructions");
+
+   if (p->nr_decl_insn > I915_MAX_DECL_INSN)
+      i915_program_error(p, "Exceeded max DECL instructions");
+
+   /* free old program, if present */
+   if (i915->current.program) {
+      FREE(i915->current.program);
+      i915->current.program_len = 0;
+   }
+
+   if (p->error) {
+      p->NumNativeInstructions = 0;
+      p->NumNativeAluInstructions = 0;
+      p->NumNativeTexInstructions = 0;
+      p->NumNativeTexIndirections = 0;
+
+      i915_use_passthrough_shader(i915);
+   }
+   else {
+      p->NumNativeInstructions
+         = p->nr_alu_insn + p->nr_tex_insn + p->nr_decl_insn;
+      p->NumNativeAluInstructions = p->nr_alu_insn;
+      p->NumNativeTexInstructions = p->nr_tex_insn;
+      p->NumNativeTexIndirections = p->nr_tex_indirect;
+
+      /* patch in the program length */
+      p->declarations[0] |= program_size + decl_size - 2;
+
+      /* Copy compilation results to fragment program struct: 
+       */
+      i915->current.program
+         = (uint *) MALLOC((program_size + decl_size) * sizeof(uint));
+      if (i915->current.program) {
+         i915->current.program_len = program_size + decl_size;
+
+         memcpy(i915->current.program,
+                p->declarations, 
+                decl_size * sizeof(uint));
+
+         memcpy(i915->current.program + decl_size, 
+                p->program, 
+                program_size * sizeof(uint));
+      }
+
+      /* update number of constants */
+      i915->current.num_constants[PIPE_SHADER_FRAGMENT] = p->num_constants;
+      assert(i915->current.num_constants[PIPE_SHADER_FRAGMENT]
+             >= i915->current.num_user_constants[PIPE_SHADER_FRAGMENT]);
+   }
+
+   /* Release the compilation struct: 
+    */
+   FREE(p);
+}
+
+
+/**
+ * Find an unused texture coordinate slot to use for fragment WPOS.
+ * Update p->fp->wpos_tex with the result (-1 if no used texcoord slot is found).
+ */
+static void
+i915_find_wpos_space(struct i915_fp_compile *p)
+{
+#if 0
+   const uint inputs
+      = p->shader->inputs_read | (1 << TGSI_ATTRIB_POS); /*XXX hack*/
+   uint i;
+
+   p->wpos_tex = -1;
+
+   if (inputs & (1 << TGSI_ATTRIB_POS)) {
+      for (i = 0; i < I915_TEX_UNITS; i++) {
+	 if ((inputs & (1 << (TGSI_ATTRIB_TEX0 + i))) == 0) {
+	    p->wpos_tex = i;
+	    return;
+	 }
+      }
+
+      i915_program_error(p, "No free texcoord for wpos value");
+   }
+#else
+   if (p->shader->input_semantic_name[0] == TGSI_SEMANTIC_POSITION) {
+      /* frag shader using the fragment position input */
+#if 0
+      assert(0);
+#endif
+   }
+#endif
+}
+
+
+
+
+/**
+ * Rather than trying to intercept and jiggle depth writes during
+ * emit, just move the value into its correct position at the end of
+ * the program:
+ */
+static void
+i915_fixup_depth_write(struct i915_fp_compile *p)
+{
+   /* XXX assuming pos/depth is always in output[0] */
+   if (p->shader->output_semantic_name[0] == TGSI_SEMANTIC_POSITION) {
+      const uint depth = UREG(REG_TYPE_OD, 0);
+
+      i915_emit_arith(p,
+                      A0_MOV,                     /* opcode */
+                      depth,                      /* dest reg */
+                      A0_DEST_CHANNEL_W,          /* write mask */
+                      0,                          /* saturate? */
+                      swizzle(depth, X, Y, Z, Z), /* src0 */
+                      0, 0 /* src1, src2 */);
+   }
+}
+
+
+void
+i915_translate_fragment_program( struct i915_context *i915 )
+{
+   struct i915_fp_compile *p = i915_init_compile(i915, i915->fs);
+   const struct tgsi_token *tokens = i915->fs->tokens;
+
+   i915_find_wpos_space(p);
+
+   i915_translate_instructions(p, tokens);
+   i915_fixup_depth_write(p);
+
+   i915_fini_compile(i915, p);
+}
diff --git a/src/gallium/drivers/i915simple/i915_prim_emit.c b/src/gallium/drivers/i915simple/i915_prim_emit.c
new file mode 100644
index 0000000000..c4a706c37d
--- /dev/null
+++ b/src/gallium/drivers/i915simple/i915_prim_emit.c
@@ -0,0 +1,215 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+
+#include "pipe/draw/draw_private.h"
+#include "pipe/p_util.h"
+
+#include "i915_context.h"
+#include "i915_winsys.h"
+#include "i915_reg.h"
+#include "i915_state.h"
+#include "i915_batch.h"
+
+
+
+/**
+ * Primitive emit to hardware.  No support for vertex buffers or any
+ * nice fast paths.
+ */
+struct setup_stage {
+   struct draw_stage stage; /**< This must be first (base class) */
+
+   struct i915_context *i915;   
+};
+
+
+
+/**
+ * Basically a cast wrapper.
+ */
+static INLINE struct setup_stage *setup_stage( struct draw_stage *stage )
+{
+   return (struct setup_stage *)stage;
+}
+
+
+/**
+ * Extract the needed fields from vertex_header and emit i915 dwords.
+ * Recall that the vertices are constructed by the 'draw' module and
+ * have a couple of slots at the beginning (1-dword header, 4-dword
+ * clip pos) that we ignore here.
+ */
+static INLINE void
+emit_hw_vertex( struct i915_context *i915,
+                const struct vertex_header *vertex)
+{
+   const struct vertex_info *vinfo = &i915->current.vertex_info;
+   uint i;
+   uint count = 0;  /* for debug/sanity */
+
+   for (i = 0; i < vinfo->num_attribs; i++) {
+      switch (vinfo->emit[i]) {
+      case EMIT_OMIT:
+         /* no-op */
+         break;
+      case EMIT_1F:
+         OUT_BATCH( fui(vertex->data[i][0]) );
+         count++;
+         break;
+      case EMIT_2F:
+         OUT_BATCH( fui(vertex->data[i][0]) );
+         OUT_BATCH( fui(vertex->data[i][1]) );
+         count += 2;
+         break;
+      case EMIT_3F:
+         OUT_BATCH( fui(vertex->data[i][0]) );
+         OUT_BATCH( fui(vertex->data[i][1]) );
+         OUT_BATCH( fui(vertex->data[i][2]) );
+         count += 3;
+         break;
+      case EMIT_4F:
+         OUT_BATCH( fui(vertex->data[i][0]) );
+         OUT_BATCH( fui(vertex->data[i][1]) );
+         OUT_BATCH( fui(vertex->data[i][2]) );
+         OUT_BATCH( fui(vertex->data[i][3]) );
+         count += 4;
+         break;
+      case EMIT_4UB:
+         OUT_BATCH( pack_ub4(float_to_ubyte( vertex->data[i][2] ),
+                             float_to_ubyte( vertex->data[i][1] ),
+                             float_to_ubyte( vertex->data[i][0] ),
+                             float_to_ubyte( vertex->data[i][3] )) );
+         count += 1;
+         break;
+      default:
+         assert(0);
+      }
+   }
+   assert(count == vinfo->size);
+}
+
+
+
+static INLINE void 
+emit_prim( struct draw_stage *stage, 
+	   struct prim_header *prim,
+	   unsigned hwprim,
+	   unsigned nr )
+{
+   struct i915_context *i915 = setup_stage(stage)->i915;
+   unsigned vertex_size = i915->current.vertex_info.size * 4; /* in bytes */
+   unsigned i;
+
+   assert(vertex_size >= 12); /* never smaller than 12 bytes */
+
+   if (i915->dirty)
+      i915_update_derived( i915 );
+
+   if (i915->hardware_dirty)
+      i915_emit_hardware_state( i915 );
+
+   if (!BEGIN_BATCH( 1 + nr * vertex_size / 4, 0 )) {
+      FLUSH_BATCH();
+
+      /* Make sure state is re-emitted after a flush: 
+       */
+      i915_update_derived( i915 );
+      i915_emit_hardware_state( i915 );
+
+      if (!BEGIN_BATCH( 1 + nr * vertex_size / 4, 0 )) {
+	 assert(0);
+	 return;
+      }
+   }
+
+   /* Emit each triangle as a single primitive.  I told you this was
+    * simple.
+    */
+   OUT_BATCH(_3DPRIMITIVE | 
+	     hwprim |
+	     ((4 + vertex_size * nr)/4 - 2));
+
+   for (i = 0; i < nr; i++)
+      emit_hw_vertex(i915, prim->v[i]);
+}
+
+
+static void 
+setup_tri( struct draw_stage *stage, struct prim_header *prim )
+{
+   emit_prim( stage, prim, PRIM3D_TRILIST, 3 );
+}
+
+
+static void
+setup_line(struct draw_stage *stage, struct prim_header *prim)
+{
+   emit_prim( stage, prim, PRIM3D_LINELIST, 2 );
+}
+
+
+static void
+setup_point(struct draw_stage *stage, struct prim_header *prim)
+{
+   emit_prim( stage, prim, PRIM3D_POINTLIST, 1 );
+}
+
+
+static void setup_flush( struct draw_stage *stage, unsigned flags )
+{
+}
+
+static void reset_stipple_counter( struct draw_stage *stage )
+{
+}
+
+static void render_destroy( struct draw_stage *stage )
+{
+   FREE( stage );
+}
+
+
+/**
+ * Create a new primitive setup/render stage.  This gets plugged into
+ * the 'draw' module's pipeline.
+ */
+struct draw_stage *i915_draw_render_stage( struct i915_context *i915 )
+{
+   struct setup_stage *setup = CALLOC_STRUCT(setup_stage);
+
+   setup->i915 = i915;
+   setup->stage.draw = i915->draw;
+   setup->stage.point = setup_point;
+   setup->stage.line = setup_line;
+   setup->stage.tri = setup_tri;
+   setup->stage.flush = setup_flush;
+   setup->stage.reset_stipple_counter = reset_stipple_counter;
+   setup->stage.destroy = render_destroy;
+
+   return &setup->stage;
+}
diff --git a/src/gallium/drivers/i915simple/i915_prim_vbuf.c b/src/gallium/drivers/i915simple/i915_prim_vbuf.c
new file mode 100644
index 0000000000..e069773fd4
--- /dev/null
+++ b/src/gallium/drivers/i915simple/i915_prim_vbuf.c
@@ -0,0 +1,254 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+/**
+ * \file
+ * Build post-transformation, post-clipping vertex buffers and element
+ * lists by hooking into the end of the primitive pipeline and
+ * manipulating the vertex_id field in the vertex headers.
+ *
+ * XXX: work in progress 
+ * 
+ * \author José Fonseca <jrfonseca@tungstengraphics.com>
+ * \author Keith Whitwell <keith@tungstengraphics.com>
+ */
+
+
+#include "pipe/draw/draw_vbuf.h"
+#include "pipe/p_debug.h"
+#include "pipe/p_util.h"
+#include "pipe/p_inlines.h"
+#include "pipe/p_winsys.h"
+
+#include "i915_context.h"
+#include "i915_reg.h"
+#include "i915_winsys.h"
+#include "i915_batch.h"
+#include "i915_state.h"
+
+
+/**
+ * Primitive renderer for i915.
+ */
+struct i915_vbuf_render {
+   struct vbuf_render base;
+
+   struct i915_context *i915;   
+
+   /** Vertex size in bytes */
+   unsigned vertex_size;
+
+   /** Hardware primitive */
+   unsigned hwprim;
+};
+
+
+/**
+ * Basically a cast wrapper.
+ */
+static INLINE struct i915_vbuf_render *
+i915_vbuf_render( struct vbuf_render *render )
+{
+   assert(render);
+   return (struct i915_vbuf_render *)render;
+}
+
+
+static const struct vertex_info *
+i915_vbuf_render_get_vertex_info( struct vbuf_render *render )
+{
+   struct i915_vbuf_render *i915_render = i915_vbuf_render(render);
+   struct i915_context *i915 = i915_render->i915;
+   return &i915->current.vertex_info;
+}
+
+
+static void *
+i915_vbuf_render_allocate_vertices( struct vbuf_render *render,
+			            ushort vertex_size,
+				    ushort nr_vertices )
+{
+   struct i915_vbuf_render *i915_render = i915_vbuf_render(render);
+   struct i915_context *i915 = i915_render->i915;
+   struct pipe_winsys *winsys = i915->pipe.winsys;
+   size_t size = (size_t)vertex_size * (size_t)nr_vertices;
+
+   /* FIXME: handle failure */
+   assert(!i915->vbo);
+   i915->vbo = winsys->buffer_create(winsys, 64, I915_BUFFER_USAGE_LIT_VERTEX,
+                                     size);
+   
+   i915->dirty |= I915_NEW_VBO;
+   
+   return winsys->buffer_map(winsys, 
+                             i915->vbo, 
+                             PIPE_BUFFER_USAGE_CPU_WRITE);
+}
+
+
+static void 
+i915_vbuf_render_set_primitive( struct vbuf_render *render, 
+                                unsigned prim )
+{
+   struct i915_vbuf_render *i915_render = i915_vbuf_render(render);
+   
+   switch(prim) {
+   case PIPE_PRIM_POINTS:
+      i915_render->hwprim = PRIM3D_POINTLIST;
+      break;
+   case PIPE_PRIM_LINES:
+      i915_render->hwprim = PRIM3D_LINELIST;
+      break;
+   case PIPE_PRIM_TRIANGLES:
+      i915_render->hwprim = PRIM3D_TRILIST;
+      break;
+   default:
+      assert(0);
+   }
+}
+
+
+static void 
+i915_vbuf_render_draw( struct vbuf_render *render,
+                       const ushort *indices,
+                       uint nr_indices)
+{
+   struct i915_vbuf_render *i915_render = i915_vbuf_render(render);
+   struct i915_context *i915 = i915_render->i915;
+   unsigned i;
+
+   assert(nr_indices);
+
+   assert((i915->dirty & ~I915_NEW_VBO) == 0);
+   
+   if (i915->dirty)
+      i915_update_derived( i915 );
+
+   if (i915->hardware_dirty)
+      i915_emit_hardware_state( i915 );
+
+   if (!BEGIN_BATCH( 1 + (nr_indices + 1)/2, 1 )) {
+      FLUSH_BATCH();
+
+      /* Make sure state is re-emitted after a flush: 
+       */
+      i915_update_derived( i915 );
+      i915_emit_hardware_state( i915 );
+
+      if (!BEGIN_BATCH( 1 + (nr_indices + 1)/2, 1 )) {
+	 assert(0);
+	 return;
+      }
+   }
+
+   OUT_BATCH( _3DPRIMITIVE |
+              PRIM_INDIRECT |
+              i915_render->hwprim |
+   	      PRIM_INDIRECT_ELTS |
+   	      nr_indices );
+   for (i = 0; i + 1 < nr_indices; i += 2) {
+      OUT_BATCH( indices[i] |
+                 (indices[i + 1] << 16) );
+   }
+   if (i < nr_indices) {
+      OUT_BATCH( indices[i] );
+   }
+}
+
+
+static void
+i915_vbuf_render_release_vertices( struct vbuf_render *render,
+			           void *vertices, 
+			           unsigned vertex_size,
+			           unsigned vertices_used )
+{
+   struct i915_vbuf_render *i915_render = i915_vbuf_render(render);
+   struct i915_context *i915 = i915_render->i915;
+   struct pipe_winsys *winsys = i915->pipe.winsys;
+
+   assert(i915->vbo);
+   winsys->buffer_unmap(winsys, i915->vbo);
+   pipe_buffer_reference(winsys, &i915->vbo, NULL);
+}
+
+
+static void
+i915_vbuf_render_destroy( struct vbuf_render *render )
+{
+   struct i915_vbuf_render *i915_render = i915_vbuf_render(render);
+   FREE(i915_render);
+}
+
+
+/**
+ * Create a new primitive render.
+ */
+static struct vbuf_render *
+i915_vbuf_render_create( struct i915_context *i915 )
+{
+   struct i915_vbuf_render *i915_render = CALLOC_STRUCT(i915_vbuf_render);
+
+   i915_render->i915 = i915;
+   
+   i915_render->base.max_vertex_buffer_bytes = 128*1024;
+   
+   /* NOTE: it must be such that state and vertices indices fit in a single 
+    * batch buffer.
+    */
+   i915_render->base.max_indices = 16*1024;
+   
+   i915_render->base.get_vertex_info = i915_vbuf_render_get_vertex_info;
+   i915_render->base.allocate_vertices = i915_vbuf_render_allocate_vertices;
+   i915_render->base.set_primitive = i915_vbuf_render_set_primitive;
+   i915_render->base.draw = i915_vbuf_render_draw;
+   i915_render->base.release_vertices = i915_vbuf_render_release_vertices;
+   i915_render->base.destroy = i915_vbuf_render_destroy;
+   
+   return &i915_render->base;
+}
+
+
+/**
+ * Create a new primitive vbuf/render stage.
+ */
+struct draw_stage *i915_draw_vbuf_stage( struct i915_context *i915 )
+{
+   struct vbuf_render *render;
+   struct draw_stage *stage;
+   
+   render = i915_vbuf_render_create(i915);
+   if(!render)
+      return NULL;
+   
+   stage = draw_vbuf_stage( i915->draw, render );
+   if(!stage) {
+      render->destroy(render);
+      return NULL;
+   }
+    
+   return stage;
+}
diff --git a/src/gallium/drivers/i915simple/i915_reg.h b/src/gallium/drivers/i915simple/i915_reg.h
new file mode 100644
index 0000000000..04620fec68
--- /dev/null
+++ b/src/gallium/drivers/i915simple/i915_reg.h
@@ -0,0 +1,978 @@
+/**************************************************************************
+ * 
+ * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+
+#ifndef I915_REG_H
+#define I915_REG_H
+
+
+#define I915_SET_FIELD( var, mask, value ) (var &= ~(mask), var |= value)
+
+#define CMD_3D (0x3<<29)
+
+#define PRIM3D_INLINE		(CMD_3D | (0x1f<<24))
+#define PRIM3D_TRILIST		(0x0<<18)
+#define PRIM3D_TRISTRIP 	(0x1<<18)
+#define PRIM3D_TRISTRIP_RVRSE	(0x2<<18)
+#define PRIM3D_TRIFAN		(0x3<<18)
+#define PRIM3D_POLY		(0x4<<18)
+#define PRIM3D_LINELIST 	(0x5<<18)
+#define PRIM3D_LINESTRIP	(0x6<<18)
+#define PRIM3D_RECTLIST 	(0x7<<18)
+#define PRIM3D_POINTLIST	(0x8<<18)
+#define PRIM3D_DIB		(0x9<<18)
+#define PRIM3D_CLEAR_RECT	(0xa<<18)
+#define PRIM3D_ZONE_INIT	(0xd<<18)
+#define PRIM3D_MASK		(0x1f<<18)
+
+/* p137 */
+#define _3DSTATE_AA_CMD			(CMD_3D | (0x06<<24))
+#define AA_LINE_ECAAR_WIDTH_ENABLE	(1<<16)
+#define AA_LINE_ECAAR_WIDTH_0_5 	0
+#define AA_LINE_ECAAR_WIDTH_1_0		(1<<14)
+#define AA_LINE_ECAAR_WIDTH_2_0 	(2<<14)
+#define AA_LINE_ECAAR_WIDTH_4_0 	(3<<14)
+#define AA_LINE_REGION_WIDTH_ENABLE	(1<<8)
+#define AA_LINE_REGION_WIDTH_0_5	0
+#define AA_LINE_REGION_WIDTH_1_0	(1<<6)
+#define AA_LINE_REGION_WIDTH_2_0	(2<<6)
+#define AA_LINE_REGION_WIDTH_4_0	(3<<6)
+
+/* 3DSTATE_BACKFACE_STENCIL_OPS, p138*/
+#define _3DSTATE_BACKFACE_STENCIL_OPS    (CMD_3D | (0x8<<24))
+#define BFO_ENABLE_STENCIL_REF          (1<<23)
+#define BFO_STENCIL_REF_SHIFT           15
+#define BFO_STENCIL_REF_MASK            (0xff<<15)
+#define BFO_ENABLE_STENCIL_FUNCS        (1<<14)
+#define BFO_STENCIL_TEST_SHIFT          11
+#define BFO_STENCIL_TEST_MASK           (0x7<<11)
+#define BFO_STENCIL_FAIL_SHIFT          8
+#define BFO_STENCIL_FAIL_MASK           (0x7<<8)
+#define BFO_STENCIL_PASS_Z_FAIL_SHIFT   5
+#define BFO_STENCIL_PASS_Z_FAIL_MASK    (0x7<<5)
+#define BFO_STENCIL_PASS_Z_PASS_SHIFT   2
+#define BFO_STENCIL_PASS_Z_PASS_MASK    (0x7<<2)
+#define BFO_ENABLE_STENCIL_TWO_SIDE     (1<<1)
+#define BFO_STENCIL_TWO_SIDE            (1<<0)
+
+
+/* 3DSTATE_BACKFACE_STENCIL_MASKS, p140 */
+#define _3DSTATE_BACKFACE_STENCIL_MASKS    (CMD_3D | (0x9<<24))
+#define BFM_ENABLE_STENCIL_TEST_MASK      (1<<17)
+#define BFM_ENABLE_STENCIL_WRITE_MASK     (1<<16)
+#define BFM_STENCIL_TEST_MASK_SHIFT       8
+#define BFM_STENCIL_TEST_MASK_MASK        (0xff<<8)
+#define BFM_STENCIL_WRITE_MASK_SHIFT      0
+#define BFM_STENCIL_WRITE_MASK_MASK       (0xff<<0)
+
+
+
+/* 3DSTATE_BIN_CONTROL p141 */
+
+/* p143 */
+#define _3DSTATE_BUF_INFO_CMD	(CMD_3D | (0x1d<<24) | (0x8e<<16) | 1)
+/* Dword 1 */
+#define BUF_3D_ID_COLOR_BACK	(0x3<<24)
+#define BUF_3D_ID_DEPTH 	(0x7<<24)
+#define BUF_3D_USE_FENCE	(1<<23)
+#define BUF_3D_TILED_SURFACE	(1<<22)
+#define BUF_3D_TILE_WALK_X	0
+#define BUF_3D_TILE_WALK_Y	(1<<21)
+#define BUF_3D_PITCH(x)         (((x)/4)<<2)
+/* Dword 2 */
+#define BUF_3D_ADDR(x)		((x) & ~0x3)
+
+
+/* 3DSTATE_CHROMA_KEY */
+
+/* 3DSTATE_CLEAR_PARAMETERS, p150 */
+#define _3DSTATE_CLEAR_PARAMETERS	(CMD_3D | (0x1d<<24) | (0x9c<<16) | 5)
+/* Dword 1 */
+#define CLEARPARAM_CLEAR_RECT		(1 << 16)
+#define CLEARPARAM_ZONE_INIT		(0 << 16)
+#define CLEARPARAM_WRITE_COLOR		(1 << 2)
+#define CLEARPARAM_WRITE_DEPTH		(1 << 1)
+#define CLEARPARAM_WRITE_STENCIL	(1 << 0)
+
+/* 3DSTATE_CONSTANT_BLEND_COLOR, p153 */
+#define _3DSTATE_CONST_BLEND_COLOR_CMD	(CMD_3D | (0x1d<<24) | (0x88<<16))
+
+
+
+/* 3DSTATE_COORD_SET_BINDINGS, p154 */
+#define _3DSTATE_COORD_SET_BINDINGS      (CMD_3D | (0x16<<24))
+#define CSB_TCB(iunit, eunit)           ((eunit)<<(iunit*3))
+
+/* p156 */
+#define _3DSTATE_DFLT_DIFFUSE_CMD	(CMD_3D | (0x1d<<24) | (0x99<<16))
+
+/* p157 */
+#define _3DSTATE_DFLT_SPEC_CMD		(CMD_3D | (0x1d<<24) | (0x9a<<16))
+
+/* p158 */
+#define _3DSTATE_DFLT_Z_CMD		(CMD_3D | (0x1d<<24) | (0x98<<16))
+
+
+/* 3DSTATE_DEPTH_OFFSET_SCALE, p159 */
+#define _3DSTATE_DEPTH_OFFSET_SCALE       (CMD_3D | (0x1d<<24) | (0x97<<16))
+/* scale in dword 1 */
+
+
+/* 3DSTATE_DEPTH_SUBRECT_DISABLE, p160 */
+#define _3DSTATE_DEPTH_SUBRECT_DISABLE    (CMD_3D | (0x1c<<24) | (0x11<<19) | 0x2)
+
+/* p161 */
+#define _3DSTATE_DST_BUF_VARS_CMD	(CMD_3D | (0x1d<<24) | (0x85<<16))
+/* Dword 1 */
+#define TEX_DEFAULT_COLOR_OGL           (0<<30)
+#define TEX_DEFAULT_COLOR_D3D           (1<<30)
+#define ZR_EARLY_DEPTH                  (1<<29)
+#define LOD_PRECLAMP_OGL                (1<<28)
+#define LOD_PRECLAMP_D3D                (0<<28)
+#define DITHER_FULL_ALWAYS              (0<<26)
+#define DITHER_FULL_ON_FB_BLEND         (1<<26)
+#define DITHER_CLAMPED_ALWAYS           (2<<26)
+#define LINEAR_GAMMA_BLEND_32BPP        (1<<25)
+#define DEBUG_DISABLE_ENH_DITHER        (1<<24)
+#define DSTORG_HORT_BIAS(x)		((x)<<20)
+#define DSTORG_VERT_BIAS(x)		((x)<<16)
+#define COLOR_4_2_2_CHNL_WRT_ALL	0
+#define COLOR_4_2_2_CHNL_WRT_Y		(1<<12)
+#define COLOR_4_2_2_CHNL_WRT_CR		(2<<12)
+#define COLOR_4_2_2_CHNL_WRT_CB		(3<<12)
+#define COLOR_4_2_2_CHNL_WRT_CRCB	(4<<12)
+#define COLOR_BUF_8BIT			0
+#define COLOR_BUF_RGB555 		(1<<8)
+#define COLOR_BUF_RGB565 		(2<<8)
+#define COLOR_BUF_ARGB8888		(3<<8)
+#define DEPTH_FRMT_16_FIXED		0
+#define DEPTH_FRMT_16_FLOAT		(1<<2)
+#define DEPTH_FRMT_24_FIXED_8_OTHER	(2<<2)
+#define VERT_LINE_STRIDE_1		(1<<1)
+#define VERT_LINE_STRIDE_0		(0<<1)
+#define VERT_LINE_STRIDE_OFS_1		1
+#define VERT_LINE_STRIDE_OFS_0		0
+
+/* p166 */
+#define _3DSTATE_DRAW_RECT_CMD		(CMD_3D|(0x1d<<24)|(0x80<<16)|3)
+/* Dword 1 */
+#define DRAW_RECT_DIS_DEPTH_OFS 	(1<<30)
+#define DRAW_DITHER_OFS_X(x)		((x)<<26)
+#define DRAW_DITHER_OFS_Y(x)		((x)<<24)
+/* Dword 2 */
+#define DRAW_YMIN(x)			((x)<<16)
+#define DRAW_XMIN(x)			(x)
+/* Dword 3 */
+#define DRAW_YMAX(x)			((x)<<16)
+#define DRAW_XMAX(x)			(x)
+/* Dword 4 */
+#define DRAW_YORG(x)			((x)<<16)
+#define DRAW_XORG(x)			(x)
+
+
+/* 3DSTATE_FILTER_COEFFICIENTS_4X4, p170 */
+
+/* 3DSTATE_FILTER_COEFFICIENTS_6X5, p172 */
+
+
+/* _3DSTATE_FOG_COLOR, p173 */
+#define _3DSTATE_FOG_COLOR_CMD		(CMD_3D|(0x15<<24))
+#define FOG_COLOR_RED(x)		((x)<<16)
+#define FOG_COLOR_GREEN(x)		((x)<<8)
+#define FOG_COLOR_BLUE(x)		(x)
+
+/* _3DSTATE_FOG_MODE, p174 */
+#define _3DSTATE_FOG_MODE_CMD		(CMD_3D|(0x1d<<24)|(0x89<<16)|2)
+/* Dword 1 */
+#define FMC1_FOGFUNC_MODIFY_ENABLE	(1<<31)
+#define FMC1_FOGFUNC_VERTEX		(0<<28)
+#define FMC1_FOGFUNC_PIXEL_EXP		(1<<28)
+#define FMC1_FOGFUNC_PIXEL_EXP2		(2<<28)
+#define FMC1_FOGFUNC_PIXEL_LINEAR	(3<<28)
+#define FMC1_FOGFUNC_MASK		(3<<28)
+#define FMC1_FOGINDEX_MODIFY_ENABLE     (1<<27)
+#define FMC1_FOGINDEX_Z		        (0<<25)
+#define FMC1_FOGINDEX_W   		(1<<25)
+#define FMC1_C1_C2_MODIFY_ENABLE	(1<<24)
+#define FMC1_DENSITY_MODIFY_ENABLE	(1<<23)
+#define FMC1_C1_ONE      	        (1<<13)
+#define FMC1_C1_MASK		        (0xffff<<4)
+/* Dword 2 */
+#define FMC2_C2_ONE		        (1<<16)
+/* Dword 3 */
+#define FMC3_D_ONE      		(1<<16)
+
+
+
+/* _3DSTATE_INDEPENDENT_ALPHA_BLEND, p177 */
+#define _3DSTATE_INDEPENDENT_ALPHA_BLEND_CMD	(CMD_3D|(0x0b<<24))
+#define IAB_MODIFY_ENABLE	        (1<<23)
+#define IAB_ENABLE       	        (1<<22)
+#define IAB_MODIFY_FUNC         	(1<<21)
+#define IAB_FUNC_SHIFT          	16
+#define IAB_MODIFY_SRC_FACTOR   	(1<<11)
+#define IAB_SRC_FACTOR_SHIFT		6
+#define IAB_SRC_FACTOR_MASK		(BLENDFACT_MASK<<6)
+#define IAB_MODIFY_DST_FACTOR	        (1<<5)
+#define IAB_DST_FACTOR_SHIFT		0
+#define IAB_DST_FACTOR_MASK		(BLENDFACT_MASK<<0)
+
+
+#define BLENDFUNC_ADD			0x0
+#define BLENDFUNC_SUBTRACT		0x1
+#define BLENDFUNC_REVERSE_SUBTRACT	0x2
+#define BLENDFUNC_MIN			0x3
+#define BLENDFUNC_MAX			0x4
+#define BLENDFUNC_MASK			0x7
+
+/* 3DSTATE_LOAD_INDIRECT, p180 */
+
+#define _3DSTATE_LOAD_INDIRECT	        (CMD_3D|(0x1d<<24)|(0x7<<16))
+#define LI0_STATE_STATIC_INDIRECT       (0x01<<8)
+#define LI0_STATE_DYNAMIC_INDIRECT      (0x02<<8)
+#define LI0_STATE_SAMPLER               (0x04<<8)
+#define LI0_STATE_MAP                   (0x08<<8)
+#define LI0_STATE_PROGRAM               (0x10<<8)
+#define LI0_STATE_CONSTANTS             (0x20<<8)
+
+#define SIS0_BUFFER_ADDRESS(x)          ((x)&~0x3)
+#define SIS0_FORCE_LOAD                 (1<<1)
+#define SIS0_BUFFER_VALID               (1<<0)
+#define SIS1_BUFFER_LENGTH(x)           ((x)&0xff)
+
+#define DIS0_BUFFER_ADDRESS(x)          ((x)&~0x3)
+#define DIS0_BUFFER_RESET               (1<<1)
+#define DIS0_BUFFER_VALID               (1<<0)
+
+#define SSB0_BUFFER_ADDRESS(x)          ((x)&~0x3)
+#define SSB0_FORCE_LOAD                 (1<<1)
+#define SSB0_BUFFER_VALID               (1<<0)
+#define SSB1_BUFFER_LENGTH(x)           ((x)&0xff)
+
+#define MSB0_BUFFER_ADDRESS(x)          ((x)&~0x3)
+#define MSB0_FORCE_LOAD                 (1<<1)
+#define MSB0_BUFFER_VALID               (1<<0)
+#define MSB1_BUFFER_LENGTH(x)           ((x)&0xff)
+
+#define PSP0_BUFFER_ADDRESS(x)          ((x)&~0x3)
+#define PSP0_FORCE_LOAD                 (1<<1)
+#define PSP0_BUFFER_VALID               (1<<0)
+#define PSP1_BUFFER_LENGTH(x)           ((x)&0xff)
+
+#define PSC0_BUFFER_ADDRESS(x)          ((x)&~0x3)
+#define PSC0_FORCE_LOAD                 (1<<1)
+#define PSC0_BUFFER_VALID               (1<<0)
+#define PSC1_BUFFER_LENGTH(x)           ((x)&0xff)
+
+
+
+
+
+/* _3DSTATE_RASTERIZATION_RULES */
+#define _3DSTATE_RASTER_RULES_CMD	(CMD_3D|(0x07<<24))
+#define ENABLE_POINT_RASTER_RULE	(1<<15)
+#define OGL_POINT_RASTER_RULE		(1<<13)
+#define ENABLE_TEXKILL_3D_4D            (1<<10)
+#define TEXKILL_3D                      (0<<9)
+#define TEXKILL_4D                      (1<<9)
+#define ENABLE_LINE_STRIP_PROVOKE_VRTX	(1<<8)
+#define ENABLE_TRI_FAN_PROVOKE_VRTX	(1<<5)
+#define LINE_STRIP_PROVOKE_VRTX(x)	((x)<<6)
+#define TRI_FAN_PROVOKE_VRTX(x) 	((x)<<3)
+
+/* _3DSTATE_SCISSOR_ENABLE, p256 */
+#define _3DSTATE_SCISSOR_ENABLE_CMD	(CMD_3D|(0x1c<<24)|(0x10<<19))
+#define ENABLE_SCISSOR_RECT		((1<<1) | 1)
+#define DISABLE_SCISSOR_RECT		(1<<1)
+
+/* _3DSTATE_SCISSOR_RECTANGLE_0, p257 */
+#define _3DSTATE_SCISSOR_RECT_0_CMD	(CMD_3D|(0x1d<<24)|(0x81<<16)|1)
+/* Dword 1 */
+#define SCISSOR_RECT_0_YMIN(x)		((x)<<16)
+#define SCISSOR_RECT_0_XMIN(x)		(x)
+/* Dword 2 */
+#define SCISSOR_RECT_0_YMAX(x)		((x)<<16)
+#define SCISSOR_RECT_0_XMAX(x)		(x)
+
+/* p189 */
+#define _3DSTATE_LOAD_STATE_IMMEDIATE_1   ((0x3<<29)|(0x1d<<24)|(0x04<<16))
+#define I1_LOAD_S(n)                      (1<<(4+n))
+
+#define S0_VB_OFFSET_MASK              0xffffffc
+#define S0_AUTO_CACHE_INV_DISABLE      (1<<0)
+
+#define S1_VERTEX_WIDTH_SHIFT          24
+#define S1_VERTEX_WIDTH_MASK           (0x3f<<24)
+#define S1_VERTEX_PITCH_SHIFT          16
+#define S1_VERTEX_PITCH_MASK           (0x3f<<16)
+
+#define TEXCOORDFMT_2D                 0x0
+#define TEXCOORDFMT_3D                 0x1
+#define TEXCOORDFMT_4D                 0x2
+#define TEXCOORDFMT_1D                 0x3
+#define TEXCOORDFMT_2D_16              0x4
+#define TEXCOORDFMT_4D_16              0x5
+#define TEXCOORDFMT_NOT_PRESENT        0xf
+#define S2_TEXCOORD_FMT0_MASK            0xf
+#define S2_TEXCOORD_FMT1_SHIFT           4
+#define S2_TEXCOORD_FMT(unit, type)    ((type)<<(unit*4))
+#define S2_TEXCOORD_NONE               (~0)
+
+/* S3 not interesting */
+
+#define S4_POINT_WIDTH_SHIFT           23
+#define S4_POINT_WIDTH_MASK            (0x1ff<<23)
+#define S4_LINE_WIDTH_SHIFT            19
+#define S4_LINE_WIDTH_ONE              (0x2<<19)
+#define S4_LINE_WIDTH_MASK             (0xf<<19)
+#define S4_FLATSHADE_ALPHA             (1<<18)
+#define S4_FLATSHADE_FOG               (1<<17)
+#define S4_FLATSHADE_SPECULAR          (1<<16)
+#define S4_FLATSHADE_COLOR             (1<<15)
+#define S4_CULLMODE_BOTH	       (0<<13)
+#define S4_CULLMODE_NONE	       (1<<13)
+#define S4_CULLMODE_CW		       (2<<13)
+#define S4_CULLMODE_CCW		       (3<<13)
+#define S4_CULLMODE_MASK	       (3<<13)
+#define S4_VFMT_POINT_WIDTH            (1<<12)
+#define S4_VFMT_SPEC_FOG               (1<<11)
+#define S4_VFMT_COLOR                  (1<<10)
+#define S4_VFMT_DEPTH_OFFSET           (1<<9)
+#define S4_VFMT_XYZ     	       (1<<6)
+#define S4_VFMT_XYZW     	       (2<<6)
+#define S4_VFMT_XY     		       (3<<6)
+#define S4_VFMT_XYW     	       (4<<6)
+#define S4_VFMT_XYZW_MASK              (7<<6)
+#define S4_FORCE_DEFAULT_DIFFUSE       (1<<5)
+#define S4_FORCE_DEFAULT_SPECULAR      (1<<4)
+#define S4_LOCAL_DEPTH_OFFSET_ENABLE   (1<<3)
+#define S4_VFMT_FOG_PARAM              (1<<2)
+#define S4_SPRITE_POINT_ENABLE         (1<<1)
+#define S4_LINE_ANTIALIAS_ENABLE       (1<<0)
+
+#define S4_VFMT_MASK (S4_VFMT_POINT_WIDTH   | 	\
+		      S4_VFMT_SPEC_FOG      |	\
+		      S4_VFMT_COLOR         |	\
+		      S4_VFMT_DEPTH_OFFSET  |	\
+		      S4_VFMT_XYZW_MASK     |	\
+		      S4_VFMT_FOG_PARAM)
+
+
+#define S5_WRITEDISABLE_ALPHA          (1<<31)
+#define S5_WRITEDISABLE_RED            (1<<30)
+#define S5_WRITEDISABLE_GREEN          (1<<29)
+#define S5_WRITEDISABLE_BLUE           (1<<28)
+#define S5_WRITEDISABLE_MASK           (0xf<<28)
+#define S5_FORCE_DEFAULT_POINT_SIZE    (1<<27)
+#define S5_LAST_PIXEL_ENABLE           (1<<26)
+#define S5_GLOBAL_DEPTH_OFFSET_ENABLE  (1<<25)
+#define S5_FOG_ENABLE                  (1<<24)
+#define S5_STENCIL_REF_SHIFT           16
+#define S5_STENCIL_REF_MASK            (0xff<<16)
+#define S5_STENCIL_TEST_FUNC_SHIFT     13
+#define S5_STENCIL_TEST_FUNC_MASK      (0x7<<13)
+#define S5_STENCIL_FAIL_SHIFT          10
+#define S5_STENCIL_FAIL_MASK           (0x7<<10)
+#define S5_STENCIL_PASS_Z_FAIL_SHIFT   7
+#define S5_STENCIL_PASS_Z_FAIL_MASK    (0x7<<7)
+#define S5_STENCIL_PASS_Z_PASS_SHIFT   4
+#define S5_STENCIL_PASS_Z_PASS_MASK    (0x7<<4)
+#define S5_STENCIL_WRITE_ENABLE        (1<<3)
+#define S5_STENCIL_TEST_ENABLE         (1<<2)
+#define S5_COLOR_DITHER_ENABLE         (1<<1)
+#define S5_LOGICOP_ENABLE              (1<<0)
+
+
+#define S6_ALPHA_TEST_ENABLE           (1<<31)
+#define S6_ALPHA_TEST_FUNC_SHIFT       28
+#define S6_ALPHA_TEST_FUNC_MASK        (0x7<<28)
+#define S6_ALPHA_REF_SHIFT             20
+#define S6_ALPHA_REF_MASK              (0xff<<20)
+#define S6_DEPTH_TEST_ENABLE           (1<<19)
+#define S6_DEPTH_TEST_FUNC_SHIFT       16
+#define S6_DEPTH_TEST_FUNC_MASK        (0x7<<16)
+#define S6_CBUF_BLEND_ENABLE           (1<<15)
+#define S6_CBUF_BLEND_FUNC_SHIFT       12
+#define S6_CBUF_BLEND_FUNC_MASK        (0x7<<12)
+#define S6_CBUF_SRC_BLEND_FACT_SHIFT   8
+#define S6_CBUF_SRC_BLEND_FACT_MASK    (0xf<<8)
+#define S6_CBUF_DST_BLEND_FACT_SHIFT   4
+#define S6_CBUF_DST_BLEND_FACT_MASK    (0xf<<4)
+#define S6_DEPTH_WRITE_ENABLE          (1<<3)
+#define S6_COLOR_WRITE_ENABLE          (1<<2)
+#define S6_TRISTRIP_PV_SHIFT           0
+#define S6_TRISTRIP_PV_MASK            (0x3<<0)
+
+#define S7_DEPTH_OFFSET_CONST_MASK     ~0
+
+
+
+#define DST_BLND_FACT(f) ((f)<<S6_CBUF_DST_BLEND_FACT_SHIFT)
+#define SRC_BLND_FACT(f) ((f)<<S6_CBUF_SRC_BLEND_FACT_SHIFT)
+#define DST_ABLND_FACT(f) ((f)<<IAB_DST_FACTOR_SHIFT)
+#define SRC_ABLND_FACT(f) ((f)<<IAB_SRC_FACTOR_SHIFT)
+
+
+
+
+/* 3DSTATE_MAP_DEINTERLACER_PARAMETERS */
+
+/* 3DSTATE_MAP_PALETTE_LOAD_32, p206 */
+#define _3DSTATE_MAP_PALETTE_LOAD_32    (CMD_3D|(0x1d<<24)|(0x8f<<16))
+/* subsequent dwords up to length (max 16) are ARGB8888 color values */
+
+/* _3DSTATE_MODES_4, p218 */
+#define _3DSTATE_MODES_4_CMD		(CMD_3D|(0x0d<<24))
+#define ENABLE_LOGIC_OP_FUNC		(1<<23)
+#define LOGIC_OP_FUNC(x)		((x)<<18)
+#define LOGICOP_MASK			(0xf<<18)
+#define MODE4_ENABLE_STENCIL_TEST_MASK	((1<<17)|(0xff00))
+#define ENABLE_STENCIL_TEST_MASK	(1<<17)
+#define STENCIL_TEST_MASK(x)		(((x)&0xff)<<8)
+#define MODE4_ENABLE_STENCIL_WRITE_MASK	((1<<16)|(0x00ff))
+#define ENABLE_STENCIL_WRITE_MASK	(1<<16)
+#define STENCIL_WRITE_MASK(x)		((x)&0xff)
+
+/* _3DSTATE_MODES_5, p220 */
+#define _3DSTATE_MODES_5_CMD		(CMD_3D|(0x0c<<24))
+#define PIPELINE_FLUSH_RENDER_CACHE	(1<<18)
+#define PIPELINE_FLUSH_TEXTURE_CACHE	(1<<16)
+
+
+/* p221 */
+#define _3DSTATE_PIXEL_SHADER_CONSTANTS  (CMD_3D|(0x1d<<24)|(0x6<<16))
+#define PS1_REG(n)                      (1<<(n))
+#define PS2_CONST_X(n)                  (n)
+#define PS3_CONST_Y(n)                  (n)
+#define PS4_CONST_Z(n)                  (n)
+#define PS5_CONST_W(n)                  (n)
+
+/* p222 */
+
+
+#define I915_MAX_TEX_INDIRECT 4
+#define I915_MAX_TEX_INSN     32
+#define I915_MAX_ALU_INSN     64
+#define I915_MAX_DECL_INSN    27
+#define I915_MAX_TEMPORARY    16
+
+
+/* Each instruction is 3 dwords long, though most don't require all
+ * this space.  Maximum of 123 instructions.  Smaller maxes per insn
+ * type.
+ */
+#define _3DSTATE_PIXEL_SHADER_PROGRAM    (CMD_3D|(0x1d<<24)|(0x5<<16))
+
+#define REG_TYPE_R                 0    /* temporary regs, no need to
+                                         * dcl, must be written before
+                                         * read -- Preserved between
+                                         * phases. 
+                                         */
+#define REG_TYPE_T                 1    /* Interpolated values, must be
+                                         * dcl'ed before use.
+                                         *
+                                         * 0..7: texture coord,
+                                         * 8: diffuse spec,
+                                         * 9: specular color,
+                                         * 10: fog parameter in w.
+                                         */
+#define REG_TYPE_CONST             2    /* Restriction: only one const
+                                         * can be referenced per
+                                         * instruction, though it may be
+                                         * selected for multiple inputs.
+                                         * Constants not initialized
+                                         * default to zero.
+                                         */
+#define REG_TYPE_S                 3    /* sampler */
+#define REG_TYPE_OC                4    /* output color (rgba) */
+#define REG_TYPE_OD                5    /* output depth (w), xyz are
+                                         * temporaries.  If not written,
+                                         * interpolated depth is used?
+                                         */
+#define REG_TYPE_U                 6    /* unpreserved temporaries */
+#define REG_TYPE_MASK              0x7
+#define REG_NR_MASK                0xf
+
+
+/* REG_TYPE_T:
+ */
+#define T_TEX0     0
+#define T_TEX1     1
+#define T_TEX2     2
+#define T_TEX3     3
+#define T_TEX4     4
+#define T_TEX5     5
+#define T_TEX6     6
+#define T_TEX7     7
+#define T_DIFFUSE  8
+#define T_SPECULAR 9
+#define T_FOG_W    10           /* interpolated fog is in W coord */
+
+/* Arithmetic instructions */
+
+/* .replicate_swizzle == selection and replication of a particular
+ * scalar channel, ie., .xxxx, .yyyy, .zzzz or .wwww 
+ */
+#define A0_NOP    (0x0<<24)     /* no operation */
+#define A0_ADD    (0x1<<24)     /* dst = src0 + src1 */
+#define A0_MOV    (0x2<<24)     /* dst = src0 */
+#define A0_MUL    (0x3<<24)     /* dst = src0 * src1 */
+#define A0_MAD    (0x4<<24)     /* dst = src0 * src1 + src2 */
+#define A0_DP2ADD (0x5<<24)     /* dst.xyzw = src0.xy dot src1.xy + src2.replicate_swizzle */
+#define A0_DP3    (0x6<<24)     /* dst.xyzw = src0.xyz dot src1.xyz */
+#define A0_DP4    (0x7<<24)     /* dst.xyzw = src0.xyzw dot src1.xyzw */
+#define A0_FRC    (0x8<<24)     /* dst = src0 - floor(src0) */
+#define A0_RCP    (0x9<<24)     /* dst.xyzw = 1/(src0.replicate_swizzle) */
+#define A0_RSQ    (0xa<<24)     /* dst.xyzw = 1/(sqrt(abs(src0.replicate_swizzle))) */
+#define A0_EXP    (0xb<<24)     /* dst.xyzw = exp2(src0.replicate_swizzle) */
+#define A0_LOG    (0xc<<24)     /* dst.xyzw = log2(abs(src0.replicate_swizzle)) */
+#define A0_CMP    (0xd<<24)     /* dst = (src0 >= 0.0) ? src1 : src2 */
+#define A0_MIN    (0xe<<24)     /* dst = (src0 < src1) ? src0 : src1 */
+#define A0_MAX    (0xf<<24)     /* dst = (src0 >= src1) ? src0 : src1 */
+#define A0_FLR    (0x10<<24)    /* dst = floor(src0) */
+#define A0_MOD    (0x11<<24)    /* dst = src0 fmod 1.0 */
+#define A0_TRC    (0x12<<24)    /* dst = int(src0) */
+#define A0_SGE    (0x13<<24)    /* dst = src0 >= src1 ? 1.0 : 0.0 */
+#define A0_SLT    (0x14<<24)    /* dst = src0 < src1 ? 1.0 : 0.0 */
+#define A0_DEST_SATURATE                 (1<<22)
+#define A0_DEST_TYPE_SHIFT                19
+/* Allow: R, OC, OD, U */
+#define A0_DEST_NR_SHIFT                 14
+/* Allow R: 0..15, OC,OD: 0..0, U: 0..2 */
+#define A0_DEST_CHANNEL_X                (1<<10)
+#define A0_DEST_CHANNEL_Y                (2<<10)
+#define A0_DEST_CHANNEL_Z                (4<<10)
+#define A0_DEST_CHANNEL_W                (8<<10)
+#define A0_DEST_CHANNEL_ALL              (0xf<<10)
+#define A0_DEST_CHANNEL_SHIFT            10
+#define A0_SRC0_TYPE_SHIFT               7
+#define A0_SRC0_NR_SHIFT                 2
+
+#define A0_DEST_CHANNEL_XY              (A0_DEST_CHANNEL_X|A0_DEST_CHANNEL_Y)
+#define A0_DEST_CHANNEL_XYZ             (A0_DEST_CHANNEL_XY|A0_DEST_CHANNEL_Z)
+
+
+#define SRC_X        0
+#define SRC_Y        1
+#define SRC_Z        2
+#define SRC_W        3
+#define SRC_ZERO     4
+#define SRC_ONE      5
+
+#define A1_SRC0_CHANNEL_X_NEGATE         (1<<31)
+#define A1_SRC0_CHANNEL_X_SHIFT          28
+#define A1_SRC0_CHANNEL_Y_NEGATE         (1<<27)
+#define A1_SRC0_CHANNEL_Y_SHIFT          24
+#define A1_SRC0_CHANNEL_Z_NEGATE         (1<<23)
+#define A1_SRC0_CHANNEL_Z_SHIFT          20
+#define A1_SRC0_CHANNEL_W_NEGATE         (1<<19)
+#define A1_SRC0_CHANNEL_W_SHIFT          16
+#define A1_SRC1_TYPE_SHIFT               13
+#define A1_SRC1_NR_SHIFT                 8
+#define A1_SRC1_CHANNEL_X_NEGATE         (1<<7)
+#define A1_SRC1_CHANNEL_X_SHIFT          4
+#define A1_SRC1_CHANNEL_Y_NEGATE         (1<<3)
+#define A1_SRC1_CHANNEL_Y_SHIFT          0
+
+#define A2_SRC1_CHANNEL_Z_NEGATE         (1<<31)
+#define A2_SRC1_CHANNEL_Z_SHIFT          28
+#define A2_SRC1_CHANNEL_W_NEGATE         (1<<27)
+#define A2_SRC1_CHANNEL_W_SHIFT          24
+#define A2_SRC2_TYPE_SHIFT               21
+#define A2_SRC2_NR_SHIFT                 16
+#define A2_SRC2_CHANNEL_X_NEGATE         (1<<15)
+#define A2_SRC2_CHANNEL_X_SHIFT          12
+#define A2_SRC2_CHANNEL_Y_NEGATE         (1<<11)
+#define A2_SRC2_CHANNEL_Y_SHIFT          8
+#define A2_SRC2_CHANNEL_Z_NEGATE         (1<<7)
+#define A2_SRC2_CHANNEL_Z_SHIFT          4
+#define A2_SRC2_CHANNEL_W_NEGATE         (1<<3)
+#define A2_SRC2_CHANNEL_W_SHIFT          0
+
+
+
+/* Texture instructions */
+#define T0_TEXLD     (0x15<<24) /* Sample texture using predeclared
+                                 * sampler and address, and output
+                                 * filtered texel data to destination
+                                 * register */
+#define T0_TEXLDP    (0x16<<24) /* Same as texld but performs a
+                                 * perspective divide of the texture
+                                 * coordinate .xyz values by .w before
+                                 * sampling. */
+#define T0_TEXLDB    (0x17<<24) /* Same as texld but biases the
+                                 * computed LOD by w.  Only S4.6 two's
+                                 * comp is used.  This implies that a
+                                 * float to fixed conversion is
+                                 * done. */
+#define T0_TEXKILL   (0x18<<24) /* Does not perform a sampling
+                                 * operation.  Simply kills the pixel
+                                 * if any channel of the address
+                                 * register is < 0.0. */
+#define T0_DEST_TYPE_SHIFT                19
+/* Allow: R, OC, OD, U */
+/* Note: U (unpreserved) regs do not retain their values between
+ * phases (cannot be used for feedback) 
+ *
+ * Note: oC and OD registers can only be used as the destination of a
+ * texture instruction once per phase (this is an implementation
+ * restriction). 
+ */
+#define T0_DEST_NR_SHIFT                 14
+/* Allow R: 0..15, OC,OD: 0..0, U: 0..2 */
+#define T0_SAMPLER_NR_SHIFT              0      /* This field ignored for TEXKILL */
+#define T0_SAMPLER_NR_MASK               (0xf<<0)
+
+#define T1_ADDRESS_REG_TYPE_SHIFT        24     /* Reg to use as texture coord */
+/* Allow R, T, OC, OD -- R, OC, OD are 'dependent' reads, new program phase */
+#define T1_ADDRESS_REG_NR_SHIFT          17
+#define T2_MBZ                           0
+
+/* Declaration instructions */
+#define D0_DCL       (0x19<<24) /* Declare a t (interpolated attrib)
+                                 * register or an s (sampler)
+                                 * register. */
+#define D0_SAMPLE_TYPE_SHIFT              22
+#define D0_SAMPLE_TYPE_2D                 (0x0<<22)
+#define D0_SAMPLE_TYPE_CUBE               (0x1<<22)
+#define D0_SAMPLE_TYPE_VOLUME             (0x2<<22)
+#define D0_SAMPLE_TYPE_MASK               (0x3<<22)
+
+#define D0_TYPE_SHIFT                19
+/* Allow: T, S */
+#define D0_NR_SHIFT                  14
+/* Allow T: 0..10, S: 0..15 */
+#define D0_CHANNEL_X                (1<<10)
+#define D0_CHANNEL_Y                (2<<10)
+#define D0_CHANNEL_Z                (4<<10)
+#define D0_CHANNEL_W                (8<<10)
+#define D0_CHANNEL_ALL              (0xf<<10)
+#define D0_CHANNEL_NONE             (0<<10)
+
+#define D0_CHANNEL_XY               (D0_CHANNEL_X|D0_CHANNEL_Y)
+#define D0_CHANNEL_XYZ              (D0_CHANNEL_XY|D0_CHANNEL_Z)
+
+/* I915 Errata: Do not allow (xz), (xw), (xzw) combinations for diffuse
+ * or specular declarations. 
+ *
+ * For T dcls, only allow: (x), (xy), (xyz), (w), (xyzw) 
+ *
+ * Must be zero for S (sampler) dcls
+ */
+#define D1_MBZ                          0
+#define D2_MBZ                          0
+
+
+
+/* p207 */
+#define _3DSTATE_MAP_STATE               (CMD_3D|(0x1d<<24)|(0x0<<16))
+
+#define MS1_MAPMASK_SHIFT               0
+#define MS1_MAPMASK_MASK                (0x8fff<<0)
+
+#define MS2_UNTRUSTED_SURFACE           (1<<31)
+#define MS2_ADDRESS_MASK                0xfffffffc
+#define MS2_VERTICAL_LINE_STRIDE        (1<<1)
+#define MS2_VERTICAL_OFFSET             (1<<1)
+
+#define MS3_HEIGHT_SHIFT              21
+#define MS3_WIDTH_SHIFT               10
+#define MS3_PALETTE_SELECT            (1<<9)
+#define MS3_MAPSURF_FORMAT_SHIFT      7
+#define MS3_MAPSURF_FORMAT_MASK       (0x7<<7)
+#define    MAPSURF_8BIT		 	   (1<<7)
+#define    MAPSURF_16BIT		   (2<<7)
+#define    MAPSURF_32BIT		   (3<<7)
+#define    MAPSURF_422			   (5<<7)
+#define    MAPSURF_COMPRESSED		   (6<<7)
+#define    MAPSURF_4BIT_INDEXED		   (7<<7)
+#define MS3_MT_FORMAT_MASK         (0x7 << 3)
+#define MS3_MT_FORMAT_SHIFT        3
+#define    MT_4BIT_IDX_ARGB8888	           (7<<3)       /* SURFACE_4BIT_INDEXED */
+#define    MT_8BIT_I8		           (0<<3)       /* SURFACE_8BIT */
+#define    MT_8BIT_L8		           (1<<3)
+#define    MT_8BIT_A8		           (4<<3)
+#define    MT_8BIT_MONO8	           (5<<3)
+#define    MT_16BIT_RGB565 		   (0<<3)       /* SURFACE_16BIT */
+#define    MT_16BIT_ARGB1555		   (1<<3)
+#define    MT_16BIT_ARGB4444		   (2<<3)
+#define    MT_16BIT_AY88		   (3<<3)
+#define    MT_16BIT_88DVDU	           (5<<3)
+#define    MT_16BIT_BUMP_655LDVDU	   (6<<3)
+#define    MT_16BIT_I16	                   (7<<3)
+#define    MT_16BIT_L16	                   (8<<3)
+#define    MT_16BIT_A16	                   (9<<3)
+#define    MT_32BIT_ARGB8888		   (0<<3)       /* SURFACE_32BIT */
+#define    MT_32BIT_ABGR8888		   (1<<3)
+#define    MT_32BIT_XRGB8888		   (2<<3)
+#define    MT_32BIT_XBGR8888		   (3<<3)
+#define    MT_32BIT_QWVU8888		   (4<<3)
+#define    MT_32BIT_AXVU8888		   (5<<3)
+#define    MT_32BIT_LXVU8888	           (6<<3)
+#define    MT_32BIT_XLVU8888	           (7<<3)
+#define    MT_32BIT_ARGB2101010	           (8<<3)
+#define    MT_32BIT_ABGR2101010	           (9<<3)
+#define    MT_32BIT_AWVU2101010	           (0xA<<3)
+#define    MT_32BIT_GR1616	           (0xB<<3)
+#define    MT_32BIT_VU1616	           (0xC<<3)
+#define    MT_32BIT_xI824	           (0xD<<3)
+#define    MT_32BIT_xA824	           (0xE<<3)
+#define    MT_32BIT_xL824	           (0xF<<3)
+#define    MT_422_YCRCB_SWAPY	           (0<<3)       /* SURFACE_422 */
+#define    MT_422_YCRCB_NORMAL	           (1<<3)
+#define    MT_422_YCRCB_SWAPUV	           (2<<3)
+#define    MT_422_YCRCB_SWAPUVY	           (3<<3)
+#define    MT_COMPRESS_DXT1		   (0<<3)       /* SURFACE_COMPRESSED */
+#define    MT_COMPRESS_DXT2_3	           (1<<3)
+#define    MT_COMPRESS_DXT4_5	           (2<<3)
+#define    MT_COMPRESS_FXT1		   (3<<3)
+#define    MT_COMPRESS_DXT1_RGB		   (4<<3)
+#define MS3_USE_FENCE_REGS              (1<<2)
+#define MS3_TILED_SURFACE             (1<<1)
+#define MS3_TILE_WALK                 (1<<0)
+
+#define MS4_PITCH_SHIFT                 21
+#define MS4_CUBE_FACE_ENA_NEGX          (1<<20)
+#define MS4_CUBE_FACE_ENA_POSX          (1<<19)
+#define MS4_CUBE_FACE_ENA_NEGY          (1<<18)
+#define MS4_CUBE_FACE_ENA_POSY          (1<<17)
+#define MS4_CUBE_FACE_ENA_NEGZ          (1<<16)
+#define MS4_CUBE_FACE_ENA_POSZ          (1<<15)
+#define MS4_CUBE_FACE_ENA_MASK          (0x3f<<15)
+#define MS4_MAX_LOD_SHIFT		9
+#define MS4_MAX_LOD_MASK		(0x3f<<9)
+#define MS4_MIP_LAYOUT_LEGACY           (0<<8)
+#define MS4_MIP_LAYOUT_BELOW_LPT        (0<<8)
+#define MS4_MIP_LAYOUT_RIGHT_LPT        (1<<8)
+#define MS4_VOLUME_DEPTH_SHIFT          0
+#define MS4_VOLUME_DEPTH_MASK           (0xff<<0)
+
+/* p244 */
+#define _3DSTATE_SAMPLER_STATE         (CMD_3D|(0x1d<<24)|(0x1<<16))
+
+#define SS1_MAPMASK_SHIFT               0
+#define SS1_MAPMASK_MASK                (0x8fff<<0)
+
+#define SS2_REVERSE_GAMMA_ENABLE        (1<<31)
+#define SS2_PACKED_TO_PLANAR_ENABLE     (1<<30)
+#define SS2_COLORSPACE_CONVERSION       (1<<29)
+#define SS2_CHROMAKEY_SHIFT             27
+#define SS2_BASE_MIP_LEVEL_SHIFT        22
+#define SS2_BASE_MIP_LEVEL_MASK         (0x1f<<22)
+#define SS2_MIP_FILTER_SHIFT            20
+#define SS2_MIP_FILTER_MASK             (0x3<<20)
+#define   MIPFILTER_NONE       	0
+#define   MIPFILTER_NEAREST	1
+#define   MIPFILTER_LINEAR	3
+#define SS2_MAG_FILTER_SHIFT          17
+#define SS2_MAG_FILTER_MASK           (0x7<<17)
+#define   FILTER_NEAREST	0
+#define   FILTER_LINEAR		1
+#define   FILTER_ANISOTROPIC	2
+#define   FILTER_4X4_1    	3
+#define   FILTER_4X4_2    	4
+#define   FILTER_4X4_FLAT 	5
+#define   FILTER_6X5_MONO   	6       /* XXX - check */
+#define SS2_MIN_FILTER_SHIFT          14
+#define SS2_MIN_FILTER_MASK           (0x7<<14)
+#define SS2_LOD_BIAS_SHIFT            5
+#define SS2_LOD_BIAS_ONE              (0x10<<5)
+#define SS2_LOD_BIAS_MASK             (0x1ff<<5)
+/* Shadow requires:
+ *  MT_X8{I,L,A}24 or MT_{I,L,A}16 texture format
+ *  FILTER_4X4_x  MIN and MAG filters
+ */
+#define SS2_SHADOW_ENABLE             (1<<4)
+#define SS2_MAX_ANISO_MASK            (1<<3)
+#define SS2_MAX_ANISO_2               (0<<3)
+#define SS2_MAX_ANISO_4               (1<<3)
+#define SS2_SHADOW_FUNC_SHIFT         0
+#define SS2_SHADOW_FUNC_MASK          (0x7<<0)
+/* SS2_SHADOW_FUNC values: see COMPAREFUNC_* */
+
+#define SS3_MIN_LOD_SHIFT            24
+#define SS3_MIN_LOD_ONE              (0x10<<24)
+#define SS3_MIN_LOD_MASK             (0xff<<24)
+#define SS3_KILL_PIXEL_ENABLE        (1<<17)
+#define SS3_TCX_ADDR_MODE_SHIFT      12
+#define SS3_TCX_ADDR_MODE_MASK       (0x7<<12)
+#define   TEXCOORDMODE_WRAP		0
+#define   TEXCOORDMODE_MIRROR		1
+#define   TEXCOORDMODE_CLAMP_EDGE	2
+#define   TEXCOORDMODE_CUBE       	3
+#define   TEXCOORDMODE_CLAMP_BORDER	4
+#define   TEXCOORDMODE_MIRROR_ONCE      5
+#define SS3_TCY_ADDR_MODE_SHIFT      9
+#define SS3_TCY_ADDR_MODE_MASK       (0x7<<9)
+#define SS3_TCZ_ADDR_MODE_SHIFT      6
+#define SS3_TCZ_ADDR_MODE_MASK       (0x7<<6)
+#define SS3_NORMALIZED_COORDS        (1<<5)
+#define SS3_TEXTUREMAP_INDEX_SHIFT   1
+#define SS3_TEXTUREMAP_INDEX_MASK    (0xf<<1)
+#define SS3_DEINTERLACER_ENABLE      (1<<0)
+
+#define SS4_BORDER_COLOR_MASK        (~0)
+
+/* 3DSTATE_SPAN_STIPPLE, p258
+ */
+#define _3DSTATE_STIPPLE           ((0x3<<29)|(0x1d<<24)|(0x83<<16))
+#define ST1_ENABLE               (1<<16)
+#define ST1_MASK                 (0xffff)
+
+#define _3DSTATE_DEFAULT_Z          ((0x3<<29)|(0x1d<<24)|(0x98<<16))
+#define _3DSTATE_DEFAULT_DIFFUSE    ((0x3<<29)|(0x1d<<24)|(0x99<<16))
+#define _3DSTATE_DEFAULT_SPECULAR   ((0x3<<29)|(0x1d<<24)|(0x9a<<16))
+
+
+#define MI_FLUSH                   ((0<<29)|(4<<23))
+#define FLUSH_MAP_CACHE            (1<<0)
+#define INHIBIT_FLUSH_RENDER_CACHE (1<<2)
+
+
+#define CMD_3D (0x3<<29)
+
+
+#define _3DPRIMITIVE         ((0x3<<29)|(0x1f<<24))
+#define PRIM_INDIRECT            (1<<23)
+#define PRIM_INLINE              (0<<23)
+#define PRIM_INDIRECT_SEQUENTIAL (0<<17)
+#define PRIM_INDIRECT_ELTS       (1<<17)
+
+#define PRIM3D_TRILIST		(0x0<<18)
+#define PRIM3D_TRISTRIP 	(0x1<<18)
+#define PRIM3D_TRISTRIP_RVRSE	(0x2<<18)
+#define PRIM3D_TRIFAN		(0x3<<18)
+#define PRIM3D_POLY		(0x4<<18)
+#define PRIM3D_LINELIST 	(0x5<<18)
+#define PRIM3D_LINESTRIP	(0x6<<18)
+#define PRIM3D_RECTLIST 	(0x7<<18)
+#define PRIM3D_POINTLIST	(0x8<<18)
+#define PRIM3D_DIB		(0x9<<18)
+#define PRIM3D_MASK		(0x1f<<18)
+
+#define I915PACKCOLOR4444(r,g,b,a) \
+  ((((a) & 0xf0) << 8) | (((r) & 0xf0) << 4) | ((g) & 0xf0) | ((b) >> 4))
+
+#define I915PACKCOLOR1555(r,g,b,a) \
+  ((((r) & 0xf8) << 7) | (((g) & 0xf8) << 2) | (((b) & 0xf8) >> 3) | \
+    ((a) ? 0x8000 : 0))
+
+#define I915PACKCOLOR565(r,g,b) \
+  ((((r) & 0xf8) << 8) | (((g) & 0xfc) << 3) | (((b) & 0xf8) >> 3))
+
+#define I915PACKCOLOR8888(r,g,b,a) \
+  ((a<<24) | (r<<16) | (g<<8) | b)
+
+
+
+
+#define BR00_BITBLT_CLIENT   0x40000000
+#define BR00_OP_COLOR_BLT    0x10000000
+#define BR00_OP_SRC_COPY_BLT 0x10C00000
+#define BR13_SOLID_PATTERN   0x80000000
+
+#define XY_COLOR_BLT_CMD		((2<<29)|(0x50<<22)|0x4)
+#define XY_COLOR_BLT_WRITE_ALPHA	(1<<21)
+#define XY_COLOR_BLT_WRITE_RGB		(1<<20)
+
+#define XY_SRC_COPY_BLT_CMD             ((2<<29)|(0x53<<22)|6)
+#define XY_SRC_COPY_BLT_WRITE_ALPHA     (1<<21)
+#define XY_SRC_COPY_BLT_WRITE_RGB       (1<<20)
+
+#define MI_WAIT_FOR_EVENT               ((0x3<<23))
+#define MI_WAIT_FOR_PLANE_B_FLIP        (1<<6)
+#define MI_WAIT_FOR_PLANE_A_FLIP        (1<<2)
+
+#define MI_BATCH_BUFFER                 (0x30<<23)
+#define MI_BATCH_BUFFER_START           (0x31<<23)
+#define MI_BATCH_BUFFER_END             (0xa<<23)
+
+
+
+#define COMPAREFUNC_ALWAYS		0
+#define COMPAREFUNC_NEVER		0x1
+#define COMPAREFUNC_LESS		0x2
+#define COMPAREFUNC_EQUAL		0x3
+#define COMPAREFUNC_LEQUAL		0x4
+#define COMPAREFUNC_GREATER		0x5
+#define COMPAREFUNC_NOTEQUAL		0x6
+#define COMPAREFUNC_GEQUAL		0x7
+
+#define STENCILOP_KEEP			0
+#define STENCILOP_ZERO			0x1
+#define STENCILOP_REPLACE		0x2
+#define STENCILOP_INCRSAT		0x3
+#define STENCILOP_DECRSAT		0x4
+#define STENCILOP_INCR			0x5
+#define STENCILOP_DECR			0x6
+#define STENCILOP_INVERT		0x7
+
+#define LOGICOP_CLEAR			0
+#define LOGICOP_NOR			0x1
+#define LOGICOP_AND_INV 		0x2
+#define LOGICOP_COPY_INV		0x3
+#define LOGICOP_AND_RVRSE		0x4
+#define LOGICOP_INV			0x5
+#define LOGICOP_XOR			0x6
+#define LOGICOP_NAND			0x7
+#define LOGICOP_AND			0x8
+#define LOGICOP_EQUIV			0x9
+#define LOGICOP_NOOP			0xa
+#define LOGICOP_OR_INV			0xb
+#define LOGICOP_COPY			0xc
+#define LOGICOP_OR_RVRSE		0xd
+#define LOGICOP_OR			0xe
+#define LOGICOP_SET			0xf
+
+#define BLENDFACT_ZERO			0x01
+#define BLENDFACT_ONE			0x02
+#define BLENDFACT_SRC_COLR		0x03
+#define BLENDFACT_INV_SRC_COLR 		0x04
+#define BLENDFACT_SRC_ALPHA		0x05
+#define BLENDFACT_INV_SRC_ALPHA 	0x06
+#define BLENDFACT_DST_ALPHA		0x07
+#define BLENDFACT_INV_DST_ALPHA 	0x08
+#define BLENDFACT_DST_COLR		0x09
+#define BLENDFACT_INV_DST_COLR		0x0a
+#define BLENDFACT_SRC_ALPHA_SATURATE	0x0b
+#define BLENDFACT_CONST_COLOR		0x0c
+#define BLENDFACT_INV_CONST_COLOR	0x0d
+#define BLENDFACT_CONST_ALPHA		0x0e
+#define BLENDFACT_INV_CONST_ALPHA	0x0f
+#define BLENDFACT_MASK          	0x0f
+
+#define PCI_CHIP_I915_G			0x2582
+#define PCI_CHIP_I915_GM		0x2592
+#define PCI_CHIP_I945_G			0x2772
+#define PCI_CHIP_I945_GM		0x27A2
+#define PCI_CHIP_I945_GME		0x27AE
+#define PCI_CHIP_G33_G			0x29C2
+#define PCI_CHIP_Q35_G			0x29B2
+#define PCI_CHIP_Q33_G			0x29D2
+
+
+#endif
diff --git a/src/gallium/drivers/i915simple/i915_state.c b/src/gallium/drivers/i915simple/i915_state.c
new file mode 100644
index 0000000000..abd5571b88
--- /dev/null
+++ b/src/gallium/drivers/i915simple/i915_state.c
@@ -0,0 +1,694 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+/* Authors:  Keith Whitwell <keith@tungstengraphics.com>
+ */
+
+
+#include "pipe/draw/draw_context.h"
+#include "pipe/p_winsys.h"
+#include "pipe/p_util.h"
+
+#include "i915_context.h"
+#include "i915_reg.h"
+#include "i915_state.h"
+#include "i915_state_inlines.h"
+
+
+/* The i915 (and related graphics cores) do not support GL_CLAMP.  The
+ * Intel drivers for "other operating systems" implement GL_CLAMP as
+ * GL_CLAMP_TO_EDGE, so the same is done here.
+ */
+static unsigned
+translate_wrap_mode(unsigned wrap)
+{
+   switch (wrap) {
+   case PIPE_TEX_WRAP_REPEAT:
+      return TEXCOORDMODE_WRAP;
+   case PIPE_TEX_WRAP_CLAMP:
+      return TEXCOORDMODE_CLAMP_EDGE;   /* not quite correct */
+   case PIPE_TEX_WRAP_CLAMP_TO_EDGE:
+      return TEXCOORDMODE_CLAMP_EDGE;
+   case PIPE_TEX_WRAP_CLAMP_TO_BORDER:
+      return TEXCOORDMODE_CLAMP_BORDER;
+//   case PIPE_TEX_WRAP_MIRRORED_REPEAT:
+//      return TEXCOORDMODE_MIRROR;
+   default:
+      return TEXCOORDMODE_WRAP;
+   }
+}
+
+static unsigned translate_img_filter( unsigned filter )
+{
+   switch (filter) {
+   case PIPE_TEX_FILTER_NEAREST:
+      return FILTER_NEAREST;
+   case PIPE_TEX_FILTER_LINEAR:
+      return FILTER_LINEAR;
+   default:
+      assert(0);
+      return FILTER_NEAREST;
+   }
+}
+
+static unsigned translate_mip_filter( unsigned filter )
+{
+   switch (filter) {
+   case PIPE_TEX_MIPFILTER_NONE:
+      return MIPFILTER_NONE;
+   case PIPE_TEX_MIPFILTER_NEAREST:
+      return MIPFILTER_NEAREST;
+   case PIPE_TEX_FILTER_LINEAR:
+      return MIPFILTER_LINEAR;
+   default:
+      assert(0);
+      return MIPFILTER_NONE;
+   }
+}
+
+
+/* None of this state is actually used for anything yet.
+ */
+static void *
+i915_create_blend_state(struct pipe_context *pipe,
+                        const struct pipe_blend_state *blend)
+{
+   struct i915_blend_state *cso_data = CALLOC_STRUCT( i915_blend_state );
+
+   {
+      unsigned eqRGB  = blend->rgb_func;
+      unsigned srcRGB = blend->rgb_src_factor;
+      unsigned dstRGB = blend->rgb_dst_factor;
+
+      unsigned eqA    = blend->alpha_func;
+      unsigned srcA   = blend->alpha_src_factor;
+      unsigned dstA   = blend->alpha_dst_factor;
+
+      /* Special handling for MIN/MAX filter modes handled at
+       * state_tracker level.
+       */
+
+      if (srcA != srcRGB ||
+	  dstA != dstRGB ||
+	  eqA != eqRGB) {
+
+	 cso_data->iab = (_3DSTATE_INDEPENDENT_ALPHA_BLEND_CMD |
+                          IAB_MODIFY_ENABLE |
+                          IAB_ENABLE |
+                          IAB_MODIFY_FUNC |
+                          IAB_MODIFY_SRC_FACTOR |
+                          IAB_MODIFY_DST_FACTOR |
+                          SRC_ABLND_FACT(i915_translate_blend_factor(srcA)) |
+                          DST_ABLND_FACT(i915_translate_blend_factor(dstA)) |
+                          (i915_translate_blend_func(eqA) << IAB_FUNC_SHIFT));
+      }
+      else {
+	 cso_data->iab = (_3DSTATE_INDEPENDENT_ALPHA_BLEND_CMD |
+                          IAB_MODIFY_ENABLE |
+                          0);
+      }
+   }
+
+   cso_data->modes4 |= (_3DSTATE_MODES_4_CMD |
+                        ENABLE_LOGIC_OP_FUNC |
+                        LOGIC_OP_FUNC(i915_translate_logic_op(blend->logicop_func)));
+
+   if (blend->logicop_enable)
+      cso_data->LIS5 |= S5_LOGICOP_ENABLE;
+
+   if (blend->dither)
+      cso_data->LIS5 |= S5_COLOR_DITHER_ENABLE;
+
+   if ((blend->colormask & PIPE_MASK_R) == 0)
+      cso_data->LIS5 |= S5_WRITEDISABLE_RED;
+
+   if ((blend->colormask & PIPE_MASK_G) == 0)
+      cso_data->LIS5 |= S5_WRITEDISABLE_GREEN;
+
+   if ((blend->colormask & PIPE_MASK_B) == 0)
+      cso_data->LIS5 |= S5_WRITEDISABLE_BLUE;
+
+   if ((blend->colormask & PIPE_MASK_A) == 0)
+      cso_data->LIS5 |= S5_WRITEDISABLE_ALPHA;
+
+   if (blend->blend_enable) {
+      unsigned funcRGB = blend->rgb_func;
+      unsigned srcRGB  = blend->rgb_src_factor;
+      unsigned dstRGB  = blend->rgb_dst_factor;
+
+      cso_data->LIS6 |= (S6_CBUF_BLEND_ENABLE |
+                         SRC_BLND_FACT(i915_translate_blend_factor(srcRGB)) |
+                         DST_BLND_FACT(i915_translate_blend_factor(dstRGB)) |
+                         (i915_translate_blend_func(funcRGB) << S6_CBUF_BLEND_FUNC_SHIFT));
+   }
+
+   return cso_data;
+}
+
+static void i915_bind_blend_state(struct pipe_context *pipe,
+                                  void *blend)
+{
+   struct i915_context *i915 = i915_context(pipe);
+
+   i915->blend = (struct i915_blend_state*)blend;
+
+   i915->dirty |= I915_NEW_BLEND;
+}
+
+
+static void i915_delete_blend_state(struct pipe_context *pipe, void *blend)
+{
+   FREE(blend);
+}
+
+static void i915_set_blend_color( struct pipe_context *pipe,
+			     const struct pipe_blend_color *blend_color )
+{
+   struct i915_context *i915 = i915_context(pipe);
+
+   i915->blend_color = *blend_color;
+
+   i915->dirty |= I915_NEW_BLEND;
+}
+
+static void *
+i915_create_sampler_state(struct pipe_context *pipe,
+                          const struct pipe_sampler_state *sampler)
+{
+   struct i915_sampler_state *cso = CALLOC_STRUCT( i915_sampler_state );
+   const unsigned ws = sampler->wrap_s;
+   const unsigned wt = sampler->wrap_t;
+   const unsigned wr = sampler->wrap_r;
+   unsigned minFilt, magFilt;
+   unsigned mipFilt;
+
+   cso->templ = sampler;
+
+   mipFilt = translate_mip_filter(sampler->min_mip_filter);
+   if (sampler->max_anisotropy > 1.0) {
+      minFilt = FILTER_ANISOTROPIC;
+      magFilt = FILTER_ANISOTROPIC;
+      if (sampler->max_anisotropy > 2.0) {
+         cso->state[0] |= SS2_MAX_ANISO_4;
+      }
+   }
+   else {
+      minFilt = translate_img_filter( sampler->min_img_filter );
+      magFilt = translate_img_filter( sampler->mag_img_filter );
+   }
+
+   {
+      int b = (int) (sampler->lod_bias * 16.0);
+      b = CLAMP(b, -256, 255);
+      cso->state[0] |= ((b << SS2_LOD_BIAS_SHIFT) & SS2_LOD_BIAS_MASK);
+   }
+
+   /* Shadow:
+    */
+   if (sampler->compare_mode == PIPE_TEX_COMPARE_R_TO_TEXTURE) 
+   {
+      cso->state[0] |= (SS2_SHADOW_ENABLE |
+                        i915_translate_compare_func(sampler->compare_func));
+
+      minFilt = FILTER_4X4_FLAT;
+      magFilt = FILTER_4X4_FLAT;
+   }
+
+   cso->state[0] |= ((minFilt << SS2_MIN_FILTER_SHIFT) |
+                     (mipFilt << SS2_MIP_FILTER_SHIFT) |
+                     (magFilt << SS2_MAG_FILTER_SHIFT));
+
+   cso->state[1] |=
+      ((translate_wrap_mode(ws) << SS3_TCX_ADDR_MODE_SHIFT) |
+       (translate_wrap_mode(wt) << SS3_TCY_ADDR_MODE_SHIFT) |
+       (translate_wrap_mode(wr) << SS3_TCZ_ADDR_MODE_SHIFT));
+
+   if (sampler->normalized_coords)
+      cso->state[1] |= SS3_NORMALIZED_COORDS;
+
+   if (0) /* XXX not tested yet */
+   {
+      int minlod = (int) (16.0 * sampler->min_lod);
+      minlod = CLAMP(minlod, 0, 16 * 11);
+      cso->state[1] |= (minlod << SS3_MIN_LOD_SHIFT);
+   }
+
+   {
+      ubyte r = float_to_ubyte(sampler->border_color[0]);
+      ubyte g = float_to_ubyte(sampler->border_color[1]);
+      ubyte b = float_to_ubyte(sampler->border_color[2]);
+      ubyte a = float_to_ubyte(sampler->border_color[3]);
+      cso->state[2] = I915PACKCOLOR8888(r, g, b, a);
+   }
+   return cso;
+}
+
+static void i915_bind_sampler_state(struct pipe_context *pipe,
+                                    unsigned unit, void *sampler)
+{
+   struct i915_context *i915 = i915_context(pipe);
+
+   assert(unit < PIPE_MAX_SAMPLERS);
+   i915->sampler[unit] = (const struct i915_sampler_state*)sampler;
+
+   i915->dirty |= I915_NEW_SAMPLER;
+}
+
+static void i915_delete_sampler_state(struct pipe_context *pipe,
+                                      void *sampler)
+{
+   FREE(sampler);
+}
+
+
+/** XXX move someday?  Or consolidate all these simple state setters
+ * into one file.
+ */
+
+static void *
+i915_create_depth_stencil_state(struct pipe_context *pipe,
+				const struct pipe_depth_stencil_alpha_state *depth_stencil)
+{
+   struct i915_depth_stencil_state *cso = CALLOC_STRUCT( i915_depth_stencil_state );
+
+   {
+      int testmask = depth_stencil->stencil[0].value_mask & 0xff;
+      int writemask = depth_stencil->stencil[0].write_mask & 0xff;
+
+      cso->stencil_modes4 |= (_3DSTATE_MODES_4_CMD |
+                              ENABLE_STENCIL_TEST_MASK |
+                              STENCIL_TEST_MASK(testmask) |
+                              ENABLE_STENCIL_WRITE_MASK |
+                              STENCIL_WRITE_MASK(writemask));
+   }
+
+   if (depth_stencil->stencil[0].enabled) {
+      int test = i915_translate_compare_func(depth_stencil->stencil[0].func);
+      int fop  = i915_translate_stencil_op(depth_stencil->stencil[0].fail_op);
+      int dfop = i915_translate_stencil_op(depth_stencil->stencil[0].zfail_op);
+      int dpop = i915_translate_stencil_op(depth_stencil->stencil[0].zpass_op);
+      int ref  = depth_stencil->stencil[0].ref_value & 0xff;
+
+      cso->stencil_LIS5 |= (S5_STENCIL_TEST_ENABLE |
+                            S5_STENCIL_WRITE_ENABLE |
+                            (ref  << S5_STENCIL_REF_SHIFT) |
+                            (test << S5_STENCIL_TEST_FUNC_SHIFT) |
+                            (fop  << S5_STENCIL_FAIL_SHIFT) |
+                            (dfop << S5_STENCIL_PASS_Z_FAIL_SHIFT) |
+                            (dpop << S5_STENCIL_PASS_Z_PASS_SHIFT));
+   }
+
+   if (depth_stencil->stencil[1].enabled) {
+      int test  = i915_translate_compare_func(depth_stencil->stencil[1].func);
+      int fop   = i915_translate_stencil_op(depth_stencil->stencil[1].fail_op);
+      int dfop  = i915_translate_stencil_op(depth_stencil->stencil[1].zfail_op);
+      int dpop  = i915_translate_stencil_op(depth_stencil->stencil[1].zpass_op);
+      int ref   = depth_stencil->stencil[1].ref_value & 0xff;
+      int tmask = depth_stencil->stencil[1].value_mask & 0xff;
+      int wmask = depth_stencil->stencil[1].write_mask & 0xff;
+
+      cso->bfo[0] = (_3DSTATE_BACKFACE_STENCIL_OPS |
+                     BFO_ENABLE_STENCIL_FUNCS |
+                     BFO_ENABLE_STENCIL_TWO_SIDE |
+                     BFO_ENABLE_STENCIL_REF |
+                     BFO_STENCIL_TWO_SIDE |
+                     (ref  << BFO_STENCIL_REF_SHIFT) |
+                     (test << BFO_STENCIL_TEST_SHIFT) |
+                     (fop  << BFO_STENCIL_FAIL_SHIFT) |
+                     (dfop << BFO_STENCIL_PASS_Z_FAIL_SHIFT) |
+                     (dpop << BFO_STENCIL_PASS_Z_PASS_SHIFT));
+
+      cso->bfo[1] = (_3DSTATE_BACKFACE_STENCIL_MASKS |
+                     BFM_ENABLE_STENCIL_TEST_MASK |
+                     BFM_ENABLE_STENCIL_WRITE_MASK |
+                     (tmask << BFM_STENCIL_TEST_MASK_SHIFT) |
+                     (wmask << BFM_STENCIL_WRITE_MASK_SHIFT));
+   }
+   else {
+      /* This actually disables two-side stencil: The bit set is a
+       * modify-enable bit to indicate we are changing the two-side
+       * setting.  Then there is a symbolic zero to show that we are
+       * setting the flag to zero/off.
+       */
+      cso->bfo[0] = (_3DSTATE_BACKFACE_STENCIL_OPS |
+                     BFO_ENABLE_STENCIL_TWO_SIDE |
+                     0);
+      cso->bfo[1] = 0;
+   }
+
+   if (depth_stencil->depth.enabled) {
+      int func = i915_translate_compare_func(depth_stencil->depth.func);
+
+      cso->depth_LIS6 |= (S6_DEPTH_TEST_ENABLE |
+                          (func << S6_DEPTH_TEST_FUNC_SHIFT));
+
+      if (depth_stencil->depth.writemask)
+	 cso->depth_LIS6 |= S6_DEPTH_WRITE_ENABLE;
+   }
+
+   if (depth_stencil->alpha.enabled) {
+      int test = i915_translate_compare_func(depth_stencil->alpha.func);
+      ubyte refByte = float_to_ubyte(depth_stencil->alpha.ref);
+
+      cso->depth_LIS6 |= (S6_ALPHA_TEST_ENABLE |
+			  (test << S6_ALPHA_TEST_FUNC_SHIFT) |
+			  (((unsigned) refByte) << S6_ALPHA_REF_SHIFT));
+   }
+
+   return cso;
+}
+
+static void i915_bind_depth_stencil_state(struct pipe_context *pipe,
+                                          void *depth_stencil)
+{
+   struct i915_context *i915 = i915_context(pipe);
+
+   i915->depth_stencil = (const struct i915_depth_stencil_state *)depth_stencil;
+
+   i915->dirty |= I915_NEW_DEPTH_STENCIL;
+}
+
+static void i915_delete_depth_stencil_state(struct pipe_context *pipe,
+                                            void *depth_stencil)
+{
+   FREE(depth_stencil);
+}
+
+
+static void i915_set_scissor_state( struct pipe_context *pipe,
+                                 const struct pipe_scissor_state *scissor )
+{
+   struct i915_context *i915 = i915_context(pipe);
+
+   memcpy( &i915->scissor, scissor, sizeof(*scissor) );
+   i915->dirty |= I915_NEW_SCISSOR;
+}
+
+
+static void i915_set_polygon_stipple( struct pipe_context *pipe,
+                                   const struct pipe_poly_stipple *stipple )
+{
+}
+
+
+static void * i915_create_fs_state(struct pipe_context *pipe,
+                                   const struct pipe_shader_state *templ)
+{
+   return 0;
+}
+
+static void i915_bind_fs_state(struct pipe_context *pipe, void *fs)
+{
+   struct i915_context *i915 = i915_context(pipe);
+
+   i915->fs = (struct pipe_shader_state *)fs;
+
+   i915->dirty |= I915_NEW_FS;
+}
+
+static void i915_delete_fs_state(struct pipe_context *pipe, void *shader)
+{
+   /*do nothing*/
+}
+
+static void *
+i915_create_vs_state(struct pipe_context *pipe,
+                     const struct pipe_shader_state *templ)
+{
+   struct i915_context *i915 = i915_context(pipe);
+
+   /* just pass-through to draw module */
+   return draw_create_vertex_shader(i915->draw, templ);
+}
+
+static void i915_bind_vs_state(struct pipe_context *pipe, void *shader)
+{
+   struct i915_context *i915 = i915_context(pipe);
+
+   /* just pass-through to draw module */
+   draw_bind_vertex_shader(i915->draw, (struct draw_vertex_shader *) shader);
+}
+
+static void i915_delete_vs_state(struct pipe_context *pipe, void *shader)
+{
+   struct i915_context *i915 = i915_context(pipe);
+
+   /* just pass-through to draw module */
+   draw_delete_vertex_shader(i915->draw, (struct draw_vertex_shader *) shader);
+}
+
+static void i915_set_constant_buffer(struct pipe_context *pipe,
+                                     uint shader, uint index,
+                                     const struct pipe_constant_buffer *buf)
+{
+   struct i915_context *i915 = i915_context(pipe);
+   struct pipe_winsys *ws = pipe->winsys;
+
+   assert(shader < PIPE_SHADER_TYPES);
+   assert(index == 0);
+
+   /* Make a copy of shader constants.
+    * During fragment program translation we may add additional
+    * constants to the array.
+    *
+    * We want to consider the situation where some user constants
+    * (ex: a material color) may change frequently but the shader program
+    * stays the same.  In that case we should only be updating the first
+    * N constants, leaving any extras from shader translation alone.
+    */
+   {
+      void *mapped;
+      if (buf->size &&
+          (mapped = ws->buffer_map(ws, buf->buffer,
+                                   PIPE_BUFFER_USAGE_CPU_READ))) {
+         memcpy(i915->current.constants[shader], mapped, buf->size);
+         ws->buffer_unmap(ws, buf->buffer);
+         i915->current.num_user_constants[shader]
+            = buf->size / (4 * sizeof(float));
+      }
+      else {
+         i915->current.num_user_constants[shader] = 0;
+      }
+   }
+
+   i915->dirty |= I915_NEW_CONSTANTS;
+}
+
+
+static void i915_set_sampler_texture(struct pipe_context *pipe,
+				     unsigned sampler,
+				     struct pipe_texture *texture)
+{
+   struct i915_context *i915 = i915_context(pipe);
+
+   i915->texture[sampler] = (struct i915_texture*)texture;  /* ptr, not struct */
+
+   i915->dirty |= I915_NEW_TEXTURE;
+}
+
+
+
+static void i915_set_framebuffer_state(struct pipe_context *pipe,
+				       const struct pipe_framebuffer_state *fb)
+{
+   struct i915_context *i915 = i915_context(pipe);
+
+   i915->framebuffer = *fb; /* struct copy */
+
+   i915->dirty |= I915_NEW_FRAMEBUFFER;
+}
+
+
+
+static void i915_set_clip_state( struct pipe_context *pipe,
+			     const struct pipe_clip_state *clip )
+{
+   struct i915_context *i915 = i915_context(pipe);
+
+   draw_set_clip_state(i915->draw, clip);
+
+   i915->dirty |= I915_NEW_CLIP;
+}
+
+
+
+/* Called when driver state tracker notices changes to the viewport
+ * matrix:
+ */
+static void i915_set_viewport_state( struct pipe_context *pipe,
+				     const struct pipe_viewport_state *viewport )
+{
+   struct i915_context *i915 = i915_context(pipe);
+
+   i915->viewport = *viewport; /* struct copy */
+
+   /* pass the viewport info to the draw module */
+   draw_set_viewport_state(i915->draw, &i915->viewport);
+
+   i915->dirty |= I915_NEW_VIEWPORT;
+}
+
+
+static void *
+i915_create_rasterizer_state(struct pipe_context *pipe,
+                             const struct pipe_rasterizer_state *rasterizer)
+{
+   struct i915_rasterizer_state *cso = CALLOC_STRUCT( i915_rasterizer_state );
+
+   cso->templ = rasterizer;
+   cso->color_interp = rasterizer->flatshade ? INTERP_CONSTANT : INTERP_LINEAR;
+   cso->light_twoside = rasterizer->light_twoside;
+   cso->ds[0].u = _3DSTATE_DEPTH_OFFSET_SCALE;
+   cso->ds[1].f = rasterizer->offset_scale;
+   if (rasterizer->poly_stipple_enable) {
+      cso->st |= ST1_ENABLE;
+   }
+
+   if (rasterizer->scissor)
+      cso->sc[0] = _3DSTATE_SCISSOR_ENABLE_CMD | ENABLE_SCISSOR_RECT;
+   else
+      cso->sc[0] = _3DSTATE_SCISSOR_ENABLE_CMD | DISABLE_SCISSOR_RECT;
+
+   switch (rasterizer->cull_mode) {
+   case PIPE_WINDING_NONE:
+      cso->LIS4 |= S4_CULLMODE_NONE;
+      break;
+   case PIPE_WINDING_CW:
+      cso->LIS4 |= S4_CULLMODE_CW;
+      break;
+   case PIPE_WINDING_CCW:
+      cso->LIS4 |= S4_CULLMODE_CCW;
+      break;
+   case PIPE_WINDING_BOTH:
+      cso->LIS4 |= S4_CULLMODE_BOTH;
+      break;
+   }
+
+   {
+      int line_width = CLAMP((int)(rasterizer->line_width * 2), 1, 0xf);
+
+      cso->LIS4 |= line_width << S4_LINE_WIDTH_SHIFT;
+
+      if (rasterizer->line_smooth)
+	 cso->LIS4 |= S4_LINE_ANTIALIAS_ENABLE;
+   }
+
+   {
+      int point_size = CLAMP((int) rasterizer->point_size, 1, 0xff);
+
+      cso->LIS4 |= point_size << S4_POINT_WIDTH_SHIFT;
+   }
+
+   if (rasterizer->flatshade) {
+      cso->LIS4 |= (S4_FLATSHADE_ALPHA |
+                    S4_FLATSHADE_COLOR |
+                    S4_FLATSHADE_SPECULAR);
+   }
+
+   cso->LIS7 = fui( rasterizer->offset_units );
+
+
+   return cso;
+}
+
+static void i915_bind_rasterizer_state( struct pipe_context *pipe,
+                                        void *setup )
+{
+   struct i915_context *i915 = i915_context(pipe);
+
+   i915->rasterizer = (struct i915_rasterizer_state *)setup;
+
+   /* pass-through to draw module */
+   draw_set_rasterizer_state(i915->draw, i915->rasterizer->templ);
+
+   i915->dirty |= I915_NEW_RASTERIZER;
+}
+
+static void i915_delete_rasterizer_state(struct pipe_context *pipe,
+                                         void *setup)
+{
+   FREE(setup);
+}
+
+static void i915_set_vertex_buffer( struct pipe_context *pipe,
+                                    unsigned index,
+                                    const struct pipe_vertex_buffer *buffer )
+{
+   struct i915_context *i915 = i915_context(pipe);
+   i915->vertex_buffer[index] = *buffer;
+   /* pass-through to draw module */
+   draw_set_vertex_buffer(i915->draw, index, buffer);
+}
+
+static void i915_set_vertex_element( struct pipe_context *pipe,
+                                     unsigned index,
+                                     const struct pipe_vertex_element *element)
+{
+   struct i915_context *i915 = i915_context(pipe);
+   /* pass-through to draw module */
+   draw_set_vertex_element(i915->draw, index, element);
+}
+
+
+
+void
+i915_init_state_functions( struct i915_context *i915 )
+{
+   i915->pipe.create_blend_state = i915_create_blend_state;
+   i915->pipe.bind_blend_state = i915_bind_blend_state;
+   i915->pipe.delete_blend_state = i915_delete_blend_state;
+
+   i915->pipe.create_sampler_state = i915_create_sampler_state;
+   i915->pipe.bind_sampler_state = i915_bind_sampler_state;
+   i915->pipe.delete_sampler_state = i915_delete_sampler_state;
+
+   i915->pipe.create_depth_stencil_alpha_state = i915_create_depth_stencil_state;
+   i915->pipe.bind_depth_stencil_alpha_state = i915_bind_depth_stencil_state;
+   i915->pipe.delete_depth_stencil_alpha_state = i915_delete_depth_stencil_state;
+
+   i915->pipe.create_rasterizer_state = i915_create_rasterizer_state;
+   i915->pipe.bind_rasterizer_state = i915_bind_rasterizer_state;
+   i915->pipe.delete_rasterizer_state = i915_delete_rasterizer_state;
+   i915->pipe.create_fs_state = i915_create_fs_state;
+   i915->pipe.bind_fs_state = i915_bind_fs_state;
+   i915->pipe.delete_fs_state = i915_delete_fs_state;
+   i915->pipe.create_vs_state = i915_create_vs_state;
+   i915->pipe.bind_vs_state = i915_bind_vs_state;
+   i915->pipe.delete_vs_state = i915_delete_vs_state;
+
+   i915->pipe.set_blend_color = i915_set_blend_color;
+   i915->pipe.set_clip_state = i915_set_clip_state;
+   i915->pipe.set_constant_buffer = i915_set_constant_buffer;
+   i915->pipe.set_framebuffer_state = i915_set_framebuffer_state;
+
+   i915->pipe.set_polygon_stipple = i915_set_polygon_stipple;
+   i915->pipe.set_scissor_state = i915_set_scissor_state;
+   i915->pipe.set_sampler_texture = i915_set_sampler_texture;
+   i915->pipe.set_viewport_state = i915_set_viewport_state;
+   i915->pipe.set_vertex_buffer = i915_set_vertex_buffer;
+   i915->pipe.set_vertex_element = i915_set_vertex_element;
+}
diff --git a/src/gallium/drivers/i915simple/i915_state.h b/src/gallium/drivers/i915simple/i915_state.h
new file mode 100644
index 0000000000..86c6b0027d
--- /dev/null
+++ b/src/gallium/drivers/i915simple/i915_state.h
@@ -0,0 +1,50 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+/* Authors:  Keith Whitwell <keith@tungstengraphics.com>
+ */
+
+#ifndef I915_STATE_H
+#define I915_STATE_H
+
+struct i915_context;
+
+
+struct i915_tracked_state {
+   unsigned dirty;
+   void (*update)( struct i915_context * );
+};
+
+void i915_update_immediate( struct i915_context *i915 );
+void i915_update_dynamic( struct i915_context *i915 );
+void i915_update_derived( struct i915_context *i915 );
+void i915_update_samplers( struct i915_context *i915 );
+void i915_update_textures(struct i915_context *i915);
+
+void i915_emit_hardware_state( struct i915_context *i915 );
+
+#endif
diff --git a/src/gallium/drivers/i915simple/i915_state_derived.c b/src/gallium/drivers/i915simple/i915_state_derived.c
new file mode 100644
index 0000000000..653983e4a9
--- /dev/null
+++ b/src/gallium/drivers/i915simple/i915_state_derived.c
@@ -0,0 +1,177 @@
+/**************************************************************************
+ * 
+ * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+
+#include "pipe/p_util.h"
+#include "pipe/draw/draw_context.h"
+#include "pipe/draw/draw_vertex.h"
+#include "i915_context.h"
+#include "i915_state.h"
+#include "i915_reg.h"
+#include "i915_fpc.h"
+#include "pipe/p_shader_tokens.h"
+
+
+/**
+ * Determine which post-transform / pre-rasterization vertex attributes
+ * we need.
+ * Derived from:  fs, setup states.
+ */
+static void calculate_vertex_layout( struct i915_context *i915 )
+{
+   const struct pipe_shader_state *fs = i915->fs;
+   const enum interp_mode colorInterp = i915->rasterizer->color_interp;
+   struct vertex_info vinfo;
+   uint front0 = 0, back0 = 0, front1 = 0, back1 = 0;
+   boolean needW = 0;
+   uint i;
+   boolean texCoords[8];
+   uint src = 0;
+
+   memset(texCoords, 0, sizeof(texCoords));
+   memset(&vinfo, 0, sizeof(vinfo));
+
+   /* pos */
+   draw_emit_vertex_attr(&vinfo, EMIT_3F, INTERP_LINEAR, src++);
+   /* Note: we'll set the S4_VFMT_XYZ[W] bits below */
+
+   for (i = 0; i < fs->num_inputs; i++) {
+      switch (fs->input_semantic_name[i]) {
+      case TGSI_SEMANTIC_POSITION:
+         break;
+      case TGSI_SEMANTIC_COLOR:
+         if (fs->input_semantic_index[i] == 0) {
+            front0 = draw_emit_vertex_attr(&vinfo, EMIT_4UB, colorInterp, src++);
+            vinfo.hwfmt[0] |= S4_VFMT_COLOR;
+         }
+         else {
+            assert(fs->input_semantic_index[i] == 1);
+            front1 = draw_emit_vertex_attr(&vinfo, EMIT_4UB, colorInterp, src++);
+            vinfo.hwfmt[0] |= S4_VFMT_SPEC_FOG;
+         }
+         break;
+      case TGSI_SEMANTIC_GENERIC:
+         /* usually a texcoord */
+         {
+            const uint unit = fs->input_semantic_index[i];
+            uint hwtc;
+            texCoords[unit] = TRUE;
+            draw_emit_vertex_attr(&vinfo, EMIT_4F, INTERP_PERSPECTIVE, src++);
+            hwtc = TEXCOORDFMT_4D;
+            needW = TRUE;
+            vinfo.hwfmt[1] |= hwtc << (unit * 4);
+         }
+         break;
+      case TGSI_SEMANTIC_FOG:
+         debug_printf("i915 fogcoord not implemented yet\n");
+         draw_emit_vertex_attr(&vinfo, EMIT_1F, INTERP_PERSPECTIVE, src++);
+         break;
+      default:
+         assert(0);
+      }
+
+   }
+
+   /* finish up texcoord fields */
+   for (i = 0; i < 8; i++) {
+      if (!texCoords[i]) {
+         const uint hwtc = TEXCOORDFMT_NOT_PRESENT;
+         vinfo.hwfmt[1] |= hwtc << (i* 4);
+      }
+   }
+
+   /* go back and fill in the vertex position info now that we have needW */
+   if (needW) {
+      vinfo.hwfmt[0] |= S4_VFMT_XYZW;
+      vinfo.emit[0] = EMIT_4F;
+   }
+   else {
+      vinfo.hwfmt[0] |= S4_VFMT_XYZ;
+      vinfo.emit[0] = EMIT_3F;
+   }
+
+   /* Additional attributes required for setup: Just twosided
+    * lighting.  Edgeflag is dealt with specially by setting bits in
+    * the vertex header.
+    */
+   if (i915->rasterizer->light_twoside) {
+      if (front0) {
+         back0 = draw_emit_vertex_attr(&vinfo, EMIT_OMIT, colorInterp, src++);
+      }
+      if (back0) {
+         back1 = draw_emit_vertex_attr(&vinfo, EMIT_OMIT, colorInterp, src++);
+      }
+   }
+
+   draw_compute_vertex_size(&vinfo);
+
+   if (memcmp(&i915->current.vertex_info, &vinfo, sizeof(vinfo))) {
+      /* Need to set this flag so that the LIS2/4 registers get set.
+       * It also means the i915_update_immediate() function must be called
+       * after this one, in i915_update_derived().
+       */
+      i915->dirty |= I915_NEW_VERTEX_FORMAT;
+
+      memcpy(&i915->current.vertex_info, &vinfo, sizeof(vinfo));
+   }
+}
+
+
+
+
+/* Hopefully this will remain quite simple, otherwise need to pull in
+ * something like the state tracker mechanism.
+ */
+void i915_update_derived( struct i915_context *i915 )
+{
+   if (i915->dirty & (I915_NEW_RASTERIZER | I915_NEW_FS))
+      calculate_vertex_layout( i915 );
+
+   if (i915->dirty & (I915_NEW_SAMPLER | I915_NEW_TEXTURE))
+      i915_update_samplers(i915);
+
+   if (i915->dirty & I915_NEW_TEXTURE)
+      i915_update_textures(i915);
+
+   if (i915->dirty)
+      i915_update_immediate( i915 );
+
+   if (i915->dirty)
+      i915_update_dynamic( i915 );
+
+   if (i915->dirty & I915_NEW_FS) {
+      i915_translate_fragment_program(i915);
+      i915->hardware_dirty |= I915_HW_PROGRAM; /* XXX right? */
+   }
+
+   /* HW emit currently references framebuffer state directly:
+    */
+   if (i915->dirty & I915_NEW_FRAMEBUFFER)
+      i915->hardware_dirty |= I915_HW_STATIC;
+
+   i915->dirty = 0;
+}
diff --git a/src/gallium/drivers/i915simple/i915_state_dynamic.c b/src/gallium/drivers/i915simple/i915_state_dynamic.c
new file mode 100644
index 0000000000..8cfbdddd19
--- /dev/null
+++ b/src/gallium/drivers/i915simple/i915_state_dynamic.c
@@ -0,0 +1,308 @@
+/**************************************************************************
+ * 
+ * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+#include "i915_batch.h"
+#include "i915_state_inlines.h"
+#include "i915_context.h"
+#include "i915_reg.h"
+#include "i915_state.h"
+#include "pipe/p_util.h"
+
+#define FILE_DEBUG_FLAG DEBUG_STATE
+
+/* State that we have chosen to store in the DYNAMIC segment of the
+ * i915 indirect state mechanism.  
+ *
+ * Can't cache these in the way we do the static state, as there is no
+ * start/size in the command packet, instead an 'end' value that gets
+ * incremented.
+ *
+ * Additionally, there seems to be a requirement to re-issue the full
+ * (active) state every time a 4kb boundary is crossed.
+ */
+
+static INLINE void set_dynamic_indirect( struct i915_context *i915,
+					 unsigned offset,
+					 const unsigned *src,
+					 unsigned dwords )
+{
+   unsigned i;
+
+   for (i = 0; i < dwords; i++)
+      i915->current.dynamic[offset + i] = src[i];
+
+   i915->hardware_dirty |= I915_HW_DYNAMIC;
+}
+
+
+/***********************************************************************
+ * Modes4: stencil masks and logicop 
+ */
+static void upload_MODES4( struct i915_context *i915 )
+{
+   unsigned modes4 = 0;
+
+   /* I915_NEW_STENCIL */
+   modes4 |= i915->depth_stencil->stencil_modes4;
+   /* I915_NEW_BLEND */
+   modes4 |= i915->blend->modes4;
+
+   /* Always, so that we know when state is in-active: 
+    */
+   set_dynamic_indirect( i915, 
+			 I915_DYNAMIC_MODES4,
+			 &modes4,
+			 1 );
+}
+
+const struct i915_tracked_state i915_upload_MODES4 = {
+   I915_NEW_BLEND | I915_NEW_DEPTH_STENCIL,
+   upload_MODES4
+};
+
+
+
+
+/***********************************************************************
+ */
+
+static void upload_BFO( struct i915_context *i915 )
+{
+   set_dynamic_indirect( i915,
+			 I915_DYNAMIC_BFO_0,
+			 &(i915->depth_stencil->bfo[0]),
+			 2 );
+}
+
+const struct i915_tracked_state i915_upload_BFO = {
+   I915_NEW_DEPTH_STENCIL,
+   upload_BFO
+};
+
+
+/***********************************************************************
+ */
+
+
+static void upload_BLENDCOLOR( struct i915_context *i915 )
+{
+   unsigned bc[2];
+
+   memset( bc, 0, sizeof(bc) );
+
+   /* I915_NEW_BLEND {_COLOR} 
+    */
+   {
+      const float *color = i915->blend_color.color;
+
+      bc[0] = _3DSTATE_CONST_BLEND_COLOR_CMD;
+      bc[1] = pack_ui32_float4( color[0],
+				color[1],
+				color[2], 
+				color[3] );
+   }
+
+   set_dynamic_indirect( i915, 
+			 I915_DYNAMIC_BC_0,
+			 bc,
+			 2 );
+}
+
+const struct i915_tracked_state i915_upload_BLENDCOLOR = {
+   I915_NEW_BLEND,
+   upload_BLENDCOLOR
+};
+
+/***********************************************************************
+ */
+
+
+static void upload_IAB( struct i915_context *i915 )
+{
+   unsigned iab = i915->blend->iab;
+
+
+   set_dynamic_indirect( i915,
+			 I915_DYNAMIC_IAB,
+			 &iab,
+			 1 );
+}
+
+const struct i915_tracked_state i915_upload_IAB = {
+   I915_NEW_BLEND,
+   upload_IAB
+};
+
+
+/***********************************************************************
+ */
+
+
+
+static void upload_DEPTHSCALE( struct i915_context *i915 )
+{
+   set_dynamic_indirect( i915,
+			 I915_DYNAMIC_DEPTHSCALE_0,
+			 &(i915->rasterizer->ds[0].u),
+			 2 );
+}
+
+const struct i915_tracked_state i915_upload_DEPTHSCALE = {
+   I915_NEW_RASTERIZER,
+   upload_DEPTHSCALE
+};
+
+
+
+/***********************************************************************
+ * Polygon stipple
+ *
+ * The i915 supports a 4x4 stipple natively, GL wants 32x32.
+ * Fortunately stipple is usually a repeating pattern.
+ *
+ * XXX: does stipple pattern need to be adjusted according to
+ * the window position?
+ *
+ * XXX: possibly need workaround for conform paths test. 
+ */
+
+static void upload_STIPPLE( struct i915_context *i915 )
+{
+   unsigned st[2];
+
+   st[0] = _3DSTATE_STIPPLE;
+   st[1] = 0;
+
+   /* I915_NEW_RASTERIZER
+    */
+   st[1] |= i915->rasterizer->st;
+
+
+   /* I915_NEW_STIPPLE
+    */
+   {
+      const ubyte *mask = (const ubyte *)i915->poly_stipple.stipple;
+      ubyte p[4];
+
+      p[0] = mask[12] & 0xf;
+      p[1] = mask[8] & 0xf;
+      p[2] = mask[4] & 0xf;
+      p[3] = mask[0] & 0xf;
+
+      /* Not sure what to do about fallbacks, so for now just dont:
+       */
+      st[1] |= ((p[0] << 0) |
+		(p[1] << 4) |
+		(p[2] << 8) | 
+		(p[3] << 12));
+   }
+
+
+   set_dynamic_indirect( i915, 
+			 I915_DYNAMIC_STP_0,
+			 &st[0],
+			 2 );
+}
+
+
+const struct i915_tracked_state i915_upload_STIPPLE = {
+   I915_NEW_RASTERIZER | I915_NEW_STIPPLE,
+   upload_STIPPLE
+};
+
+
+
+/***********************************************************************
+ * Scissor.
+ */
+static void upload_SCISSOR_ENABLE( struct i915_context *i915 )
+{
+   set_dynamic_indirect( i915,
+			 I915_DYNAMIC_SC_ENA_0,
+			 &(i915->rasterizer->sc[0]),
+			 1 );
+}
+
+const struct i915_tracked_state i915_upload_SCISSOR_ENABLE = {
+   I915_NEW_RASTERIZER,
+   upload_SCISSOR_ENABLE
+};
+
+
+
+static void upload_SCISSOR_RECT( struct i915_context *i915 )
+{
+   unsigned x1 = i915->scissor.minx;
+   unsigned y1 = i915->scissor.miny;
+   unsigned x2 = i915->scissor.maxx;
+   unsigned y2 = i915->scissor.maxy;
+   unsigned sc[3];
+ 
+   sc[0] = _3DSTATE_SCISSOR_RECT_0_CMD;
+   sc[1] = (y1 << 16) | (x1 & 0xffff);
+   sc[2] = (y2 << 16) | (x2 & 0xffff);
+
+   set_dynamic_indirect( i915, 
+			 I915_DYNAMIC_SC_RECT_0,
+			 &sc[0],
+			 3 );
+}
+
+
+const struct i915_tracked_state i915_upload_SCISSOR_RECT = {
+   I915_NEW_SCISSOR,
+   upload_SCISSOR_RECT
+};
+
+
+
+
+
+
+static const struct i915_tracked_state *atoms[] = {
+   &i915_upload_MODES4,
+   &i915_upload_BFO,
+   &i915_upload_BLENDCOLOR,
+   &i915_upload_IAB,
+   &i915_upload_DEPTHSCALE,
+   &i915_upload_STIPPLE,
+   &i915_upload_SCISSOR_ENABLE,
+   &i915_upload_SCISSOR_RECT
+};
+
+/* These will be dynamic indirect state commands, but for now just end
+ * up on the batch buffer with everything else.
+ */
+void i915_update_dynamic( struct i915_context *i915 )
+{
+   int i;
+
+   for (i = 0; i < Elements(atoms); i++)
+      if (i915->dirty & atoms[i]->dirty)
+	 atoms[i]->update( i915 );
+}
+
diff --git a/src/gallium/drivers/i915simple/i915_state_emit.c b/src/gallium/drivers/i915simple/i915_state_emit.c
new file mode 100644
index 0000000000..3339287f49
--- /dev/null
+++ b/src/gallium/drivers/i915simple/i915_state_emit.c
@@ -0,0 +1,374 @@
+/**************************************************************************
+ * 
+ * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+
+#include "i915_reg.h"
+#include "i915_context.h"
+#include "i915_winsys.h"
+#include "i915_batch.h"
+#include "i915_reg.h"
+
+#include "pipe/p_context.h"
+#include "pipe/p_defines.h"
+
+static unsigned translate_format( enum pipe_format format )
+{
+   switch (format) {
+   case PIPE_FORMAT_A8R8G8B8_UNORM:
+      return COLOR_BUF_ARGB8888;
+   case PIPE_FORMAT_R5G6B5_UNORM:
+      return COLOR_BUF_RGB565;
+   default:
+      assert(0);
+      return 0;
+   }
+}
+
+static unsigned translate_depth_format( enum pipe_format zformat )
+{
+   switch (zformat) {
+   case PIPE_FORMAT_S8Z24_UNORM:
+      return DEPTH_FRMT_24_FIXED_8_OTHER;
+   case PIPE_FORMAT_Z16_UNORM:
+      return DEPTH_FRMT_16_FIXED;
+   default:
+      assert(0);
+      return 0;
+   }
+}
+
+
+/**
+ * Examine framebuffer state to determine width, height.
+ */
+static boolean
+framebuffer_size(const struct pipe_framebuffer_state *fb,
+                 uint *width, uint *height)
+{
+   if (fb->cbufs[0]) {
+      *width = fb->cbufs[0]->width;
+      *height = fb->cbufs[0]->height;
+      return TRUE;
+   }
+   else if (fb->zsbuf) {
+      *width = fb->zsbuf->width;
+      *height = fb->zsbuf->height;
+      return TRUE;
+   }
+   else {
+      *width = *height = 0;
+      return FALSE;
+   }
+}
+
+
+/* Push the state into the sarea and/or texture memory.
+ */
+void
+i915_emit_hardware_state(struct i915_context *i915 )
+{
+   /* XXX: there must be an easier way */
+   const unsigned dwords = ( 14 + 
+                             7 + 
+                             I915_MAX_DYNAMIC + 
+                             8 + 
+                             2 + I915_TEX_UNITS*3 + 
+                             2 + I915_TEX_UNITS*3 +
+                             2 + I915_MAX_CONSTANT*4 + 
+                             i915->current.program_len + 
+                             6 
+                           ) * 3/2; /* plus 50% margin */
+   const unsigned relocs = ( I915_TEX_UNITS +
+	                     3
+                           ) * 3/2; /* plus 50% margin */
+
+#if 0
+   debug_printf("i915_emit_hardware_state: %d dwords, %d relocs\n", dwords, relocs);
+#endif
+   
+   if(!BEGIN_BATCH(dwords, relocs)) {
+      FLUSH_BATCH();
+      assert(BEGIN_BATCH(dwords, relocs));
+   }
+
+   /* 14 dwords, 0 relocs */
+   if (i915->hardware_dirty & I915_HW_INVARIENT)
+   {
+      OUT_BATCH(_3DSTATE_AA_CMD |
+		AA_LINE_ECAAR_WIDTH_ENABLE |
+		AA_LINE_ECAAR_WIDTH_1_0 |
+		AA_LINE_REGION_WIDTH_ENABLE | AA_LINE_REGION_WIDTH_1_0);
+
+      OUT_BATCH(_3DSTATE_DFLT_DIFFUSE_CMD);
+      OUT_BATCH(0);
+
+      OUT_BATCH(_3DSTATE_DFLT_SPEC_CMD);
+      OUT_BATCH(0);
+      
+      OUT_BATCH(_3DSTATE_DFLT_Z_CMD);
+      OUT_BATCH(0);
+
+      OUT_BATCH(_3DSTATE_COORD_SET_BINDINGS |
+		CSB_TCB(0, 0) |
+		CSB_TCB(1, 1) |
+		CSB_TCB(2, 2) |
+		CSB_TCB(3, 3) |
+		CSB_TCB(4, 4) | 
+		CSB_TCB(5, 5) | 
+		CSB_TCB(6, 6) | 
+		CSB_TCB(7, 7));
+
+      OUT_BATCH(_3DSTATE_RASTER_RULES_CMD |
+		ENABLE_POINT_RASTER_RULE |
+		OGL_POINT_RASTER_RULE |
+		ENABLE_LINE_STRIP_PROVOKE_VRTX |
+		ENABLE_TRI_FAN_PROVOKE_VRTX |
+		LINE_STRIP_PROVOKE_VRTX(1) |
+		TRI_FAN_PROVOKE_VRTX(2) | 
+		ENABLE_TEXKILL_3D_4D | 
+		TEXKILL_4D);
+
+      /* Need to initialize this to zero.
+       */
+      OUT_BATCH(_3DSTATE_LOAD_STATE_IMMEDIATE_1 | I1_LOAD_S(3) | (0));
+      OUT_BATCH(0);
+
+      OUT_BATCH(_3DSTATE_DEPTH_SUBRECT_DISABLE);
+
+      /* disable indirect state for now
+       */
+      OUT_BATCH(_3DSTATE_LOAD_INDIRECT | 0);
+      OUT_BATCH(0);
+   }
+   
+   /* 7 dwords, 1 relocs */
+   if (i915->hardware_dirty & I915_HW_IMMEDIATE)
+   {
+      OUT_BATCH(_3DSTATE_LOAD_STATE_IMMEDIATE_1 | 
+		I1_LOAD_S(0) |
+		I1_LOAD_S(1) |
+		I1_LOAD_S(2) |
+		I1_LOAD_S(4) |
+		I1_LOAD_S(5) |
+		I1_LOAD_S(6) | 
+		(5));
+      
+      if(i915->vbo)
+         OUT_RELOC(i915->vbo,
+                   I915_BUFFER_ACCESS_READ,
+                   i915->current.immediate[I915_IMMEDIATE_S0]);
+      else
+	 /* FIXME: we should not do this */
+	 OUT_BATCH(0);
+      OUT_BATCH(i915->current.immediate[I915_IMMEDIATE_S1]);
+      OUT_BATCH(i915->current.immediate[I915_IMMEDIATE_S2]);
+      OUT_BATCH(i915->current.immediate[I915_IMMEDIATE_S4]);
+      OUT_BATCH(i915->current.immediate[I915_IMMEDIATE_S5]);
+      OUT_BATCH(i915->current.immediate[I915_IMMEDIATE_S6]);
+   } 
+   
+   /* I915_MAX_DYNAMIC dwords, 0 relocs */
+   if (i915->hardware_dirty & I915_HW_DYNAMIC) 
+   {
+      int i;
+      for (i = 0; i < I915_MAX_DYNAMIC; i++) {
+	 OUT_BATCH(i915->current.dynamic[i]);
+      }
+   }
+   
+   /* 8 dwords, 2 relocs */
+   if (i915->hardware_dirty & I915_HW_STATIC)
+   {
+      struct pipe_surface *cbuf_surface = i915->framebuffer.cbufs[0];
+      struct pipe_surface *depth_surface = i915->framebuffer.zsbuf;
+
+      if (cbuf_surface) {
+	 unsigned pitch = (cbuf_surface->pitch * cbuf_surface->cpp);
+
+	 OUT_BATCH(_3DSTATE_BUF_INFO_CMD);
+
+	 OUT_BATCH(BUF_3D_ID_COLOR_BACK | 
+		   BUF_3D_PITCH(pitch) |  /* pitch in bytes */
+		   BUF_3D_USE_FENCE);
+
+	 OUT_RELOC(cbuf_surface->buffer,
+		   I915_BUFFER_ACCESS_WRITE,
+		   0);
+      }
+
+      /* What happens if no zbuf??
+       */
+      if (depth_surface) {
+	 unsigned zpitch = (depth_surface->pitch * depth_surface->cpp);
+			 
+	 OUT_BATCH(_3DSTATE_BUF_INFO_CMD);
+
+	 OUT_BATCH(BUF_3D_ID_DEPTH |
+		   BUF_3D_PITCH(zpitch) |  /* pitch in bytes */
+		   BUF_3D_USE_FENCE);
+
+	 OUT_RELOC(depth_surface->buffer,
+		   I915_BUFFER_ACCESS_WRITE,
+		   0);
+      }
+   
+      {
+	 unsigned cformat, zformat = 0;
+      
+	 if (cbuf_surface)
+            cformat = cbuf_surface->format;
+         else
+            cformat = PIPE_FORMAT_A8R8G8B8_UNORM; /* arbitrary */
+         cformat = translate_format(cformat);
+
+	 if (depth_surface) 
+	    zformat = translate_depth_format( i915->framebuffer.zsbuf->format );
+
+	 OUT_BATCH(_3DSTATE_DST_BUF_VARS_CMD);
+	 OUT_BATCH(DSTORG_HORT_BIAS(0x8) | /* .5 */
+		   DSTORG_VERT_BIAS(0x8) | /* .5 */
+		   LOD_PRECLAMP_OGL |
+		   TEX_DEFAULT_COLOR_OGL |
+		   cformat |
+		   zformat );
+      }
+   }
+
+#if 01
+      /* texture images */
+      /* 2 + I915_TEX_UNITS*3 dwords, I915_TEX_UNITS relocs */
+      if (i915->hardware_dirty & (I915_HW_MAP | I915_HW_SAMPLER))
+      {
+	 /* XXX: we were refering to sampler state
+	  * (current.sampler_enable_nr) below, but only checking
+	  * I915_HW_MAP above.  Should probably calculate the enabled
+	  * flags separately - but there will be further rework of
+	  * state so perhaps not necessary yet.
+	  */
+         const uint nr = i915->current.sampler_enable_nr;
+         if (nr) {
+            const uint enabled = i915->current.sampler_enable_flags;
+            uint unit;
+            uint count = 0;
+            OUT_BATCH(_3DSTATE_MAP_STATE | (3 * nr));
+            OUT_BATCH(enabled);
+            for (unit = 0; unit < I915_TEX_UNITS; unit++) {
+               if (enabled & (1 << unit)) {
+                  struct pipe_buffer *buf =
+                     i915->texture[unit]->buffer;
+                  uint offset = 0;
+                  assert(buf);
+
+                  count++;
+
+                  OUT_RELOC(buf,
+                            I915_BUFFER_ACCESS_READ,
+                            offset);
+                  OUT_BATCH(i915->current.texbuffer[unit][0]); /* MS3 */
+                  OUT_BATCH(i915->current.texbuffer[unit][1]); /* MS4 */
+               }
+            }
+            assert(count == nr);
+         }
+      }
+#endif
+
+#if 01
+   /* samplers */
+   /* 2 + I915_TEX_UNITS*3 dwords, 0 relocs */
+   if (i915->hardware_dirty & I915_HW_SAMPLER) 
+   {
+      if (i915->current.sampler_enable_nr) {
+	 int i;
+	 
+	 OUT_BATCH( _3DSTATE_SAMPLER_STATE | 
+		    (3 * i915->current.sampler_enable_nr) );
+
+	 OUT_BATCH( i915->current.sampler_enable_flags );
+
+	 for (i = 0; i < I915_TEX_UNITS; i++) {
+	    if (i915->current.sampler_enable_flags & (1<<i)) {
+	       OUT_BATCH( i915->current.sampler[i][0] );
+	       OUT_BATCH( i915->current.sampler[i][1] );
+	       OUT_BATCH( i915->current.sampler[i][2] );
+	    }
+	 }
+      }
+   }
+#endif
+
+   /* constants */
+   /* 2 + I915_MAX_CONSTANT*4 dwords, 0 relocs */
+   if (i915->hardware_dirty & I915_HW_PROGRAM)
+   {
+      const uint nr = i915->current.num_constants[PIPE_SHADER_FRAGMENT];
+      assert(nr <= I915_MAX_CONSTANT);
+      if (nr > 0) {
+         const uint *c
+            = (const uint *) i915->current.constants[PIPE_SHADER_FRAGMENT];
+         uint i;
+         OUT_BATCH( _3DSTATE_PIXEL_SHADER_CONSTANTS | (nr * 4) );
+         OUT_BATCH( (1 << (nr - 1)) | ((1 << (nr - 1)) - 1) );
+         for (i = 0; i < nr; i++) {
+            OUT_BATCH(*c++);
+            OUT_BATCH(*c++);
+            OUT_BATCH(*c++);
+            OUT_BATCH(*c++);
+         }
+      }
+   }
+
+   /* Fragment program */
+   /* i915->current.program_len dwords, 0 relocs */
+   if (i915->hardware_dirty & I915_HW_PROGRAM)
+   {
+      uint i;
+      /* we should always have, at least, a pass-through program */
+      assert(i915->current.program_len > 0);
+      for (i = 0; i < i915->current.program_len; i++) {
+         OUT_BATCH(i915->current.program[i]);
+      }
+   }
+
+   /* drawing surface size */
+   /* 6 dwords, 0 relocs */
+   {
+      uint w, h;
+      boolean k = framebuffer_size(&i915->framebuffer, &w, &h);
+      assert(k);
+
+      OUT_BATCH(_3DSTATE_DRAW_RECT_CMD);
+      OUT_BATCH(0);
+      OUT_BATCH(0);
+      OUT_BATCH(((w - 1) & 0xffff) | ((h - 1) << 16));
+      OUT_BATCH(0);
+      OUT_BATCH(0);
+   }
+
+
+   i915->hardware_dirty = 0;
+}
diff --git a/src/gallium/drivers/i915simple/i915_state_immediate.c b/src/gallium/drivers/i915simple/i915_state_immediate.c
new file mode 100644
index 0000000000..07031fc6c5
--- /dev/null
+++ b/src/gallium/drivers/i915simple/i915_state_immediate.c
@@ -0,0 +1,221 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+ /*
+  * Authors:
+  *   Keith Whitwell <keith@tungstengraphics.com>
+  */
+ 
+#include "i915_state_inlines.h"
+#include "i915_context.h"
+#include "i915_state.h"
+#include "i915_reg.h"
+#include "p_util.h"
+
+
+/* All state expressable with the LOAD_STATE_IMMEDIATE_1 packet.
+ * Would like to opportunistically recombine all these fragments into
+ * a single packet containing only what has changed, but for now emit
+ * as multiple packets.
+ */
+
+
+
+
+/***********************************************************************
+ * S0,S1: Vertex buffer state.  
+ */
+static void upload_S0S1(struct i915_context *i915)
+{
+   unsigned LIS0, LIS1;
+
+   /* INTEL_NEW_VBO */
+   /* TODO: re-use vertex buffers here? */
+   LIS0 = 0;
+
+   /* INTEL_NEW_VERTEX_SIZE -- do this where the vertex size is calculated! 
+    */
+   {
+      unsigned vertex_size = i915->current.vertex_info.size;
+
+      LIS1 = ((vertex_size << 24) |
+	      (vertex_size << 16));
+   }
+
+   /* INTEL_NEW_VBO */
+   /* TODO: use a vertex generation number to track vbo changes */
+   if (1 ||
+       i915->current.immediate[I915_IMMEDIATE_S0] != LIS0 ||
+       i915->current.immediate[I915_IMMEDIATE_S1] != LIS1) 
+   {
+      i915->current.immediate[I915_IMMEDIATE_S0] = LIS0;
+      i915->current.immediate[I915_IMMEDIATE_S1] = LIS1;
+      i915->hardware_dirty |= I915_HW_IMMEDIATE;
+   }
+}
+
+const struct i915_tracked_state i915_upload_S0S1 = {
+   I915_NEW_VBO | I915_NEW_VERTEX_FORMAT,
+   upload_S0S1
+};
+
+
+
+
+/***********************************************************************
+ * S4: Vertex format, rasterization state
+ */
+static void upload_S2S4(struct i915_context *i915)
+{
+   unsigned LIS2, LIS4;
+
+   /* I915_NEW_VERTEX_FORMAT */
+   {
+      LIS2 = i915->current.vertex_info.hwfmt[1];
+      LIS4 = i915->current.vertex_info.hwfmt[0];
+      /*
+      debug_printf("LIS2: 0x%x  LIS4: 0x%x\n", LIS2, LIS4);
+      */
+      assert(LIS4); /* should never be zero? */
+   }
+
+   LIS4 |= i915->rasterizer->LIS4;
+
+   if (LIS2 != i915->current.immediate[I915_IMMEDIATE_S2] ||
+       LIS4 != i915->current.immediate[I915_IMMEDIATE_S4]) {
+
+      i915->current.immediate[I915_IMMEDIATE_S2] = LIS2;
+      i915->current.immediate[I915_IMMEDIATE_S4] = LIS4;
+      i915->hardware_dirty |= I915_HW_IMMEDIATE;
+   }
+}
+
+
+const struct i915_tracked_state i915_upload_S2S4 = {
+   I915_NEW_RASTERIZER | I915_NEW_VERTEX_FORMAT,
+   upload_S2S4
+};
+
+
+
+/***********************************************************************
+ * 
+ */
+static void upload_S5( struct i915_context *i915 )
+{
+   unsigned LIS5 = 0;
+
+   LIS5 |= i915->depth_stencil->stencil_LIS5;
+
+   LIS5 |= i915->blend->LIS5;
+
+#if 0
+   /* I915_NEW_RASTERIZER */
+   if (i915->state.Polygon->OffsetFill) {
+      LIS5 |= S5_GLOBAL_DEPTH_OFFSET_ENABLE;
+   }
+#endif
+
+
+   if (LIS5 != i915->current.immediate[I915_IMMEDIATE_S5]) {
+      i915->current.immediate[I915_IMMEDIATE_S5] = LIS5;
+      i915->hardware_dirty |= I915_HW_IMMEDIATE;
+   }
+}
+
+const struct i915_tracked_state i915_upload_S5 = {
+   (I915_NEW_DEPTH_STENCIL | I915_NEW_BLEND | I915_NEW_RASTERIZER),
+   upload_S5
+};
+
+
+/***********************************************************************
+ */
+static void upload_S6( struct i915_context *i915 )
+{
+   unsigned LIS6 = (S6_COLOR_WRITE_ENABLE |
+		  (2 << S6_TRISTRIP_PV_SHIFT));
+
+   /* I915_NEW_BLEND
+    */
+   LIS6 |= i915->blend->LIS6;
+
+   /* I915_NEW_DEPTH
+    */
+   LIS6 |= i915->depth_stencil->depth_LIS6;
+
+   if (LIS6 != i915->current.immediate[I915_IMMEDIATE_S6]) {
+      i915->current.immediate[I915_IMMEDIATE_S6] = LIS6;
+      i915->hardware_dirty |= I915_HW_IMMEDIATE;
+   }
+}
+
+const struct i915_tracked_state i915_upload_S6 = {
+   I915_NEW_BLEND | I915_NEW_DEPTH_STENCIL,
+   upload_S6
+};
+
+
+/***********************************************************************
+ */
+static void upload_S7( struct i915_context *i915 )
+{
+   unsigned LIS7;
+
+   /* I915_NEW_RASTERIZER
+    */
+   LIS7 = i915->rasterizer->LIS7;
+
+   if (LIS7 != i915->current.immediate[I915_IMMEDIATE_S7]) {
+      i915->current.immediate[I915_IMMEDIATE_S7] = LIS7;
+      i915->hardware_dirty |= I915_HW_IMMEDIATE;
+   }
+}
+
+const struct i915_tracked_state i915_upload_S7 = {
+   I915_NEW_RASTERIZER,
+   upload_S7
+};
+
+
+static const struct i915_tracked_state *atoms[] = {
+   &i915_upload_S0S1,
+   &i915_upload_S2S4,
+   &i915_upload_S5,
+   &i915_upload_S6,
+   &i915_upload_S7
+};
+
+/* 
+ */
+void i915_update_immediate( struct i915_context *i915 )
+{
+   int i;
+
+   for (i = 0; i < Elements(atoms); i++)
+      if (i915->dirty & atoms[i]->dirty)
+	 atoms[i]->update( i915 );
+}
diff --git a/src/gallium/drivers/i915simple/i915_state_inlines.h b/src/gallium/drivers/i915simple/i915_state_inlines.h
new file mode 100644
index 0000000000..0934ac79a4
--- /dev/null
+++ b/src/gallium/drivers/i915simple/i915_state_inlines.h
@@ -0,0 +1,230 @@
+/**************************************************************************
+ * 
+ * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+#ifndef I915_STATE_INLINES_H
+#define I915_STATE_INLINES_H
+
+#include "p_compiler.h"
+#include "p_defines.h"
+#include "i915_reg.h"
+
+
+static INLINE unsigned
+i915_translate_compare_func(unsigned func)
+{
+   switch (func) {
+   case PIPE_FUNC_NEVER:
+      return COMPAREFUNC_NEVER;
+   case PIPE_FUNC_LESS:
+      return COMPAREFUNC_LESS;
+   case PIPE_FUNC_LEQUAL:
+      return COMPAREFUNC_LEQUAL;
+   case PIPE_FUNC_GREATER:
+      return COMPAREFUNC_GREATER;
+   case PIPE_FUNC_GEQUAL:
+      return COMPAREFUNC_GEQUAL;
+   case PIPE_FUNC_NOTEQUAL:
+      return COMPAREFUNC_NOTEQUAL;
+   case PIPE_FUNC_EQUAL:
+      return COMPAREFUNC_EQUAL;
+   case PIPE_FUNC_ALWAYS:
+      return COMPAREFUNC_ALWAYS;
+   default:
+      return COMPAREFUNC_ALWAYS;
+   }
+}
+
+static INLINE unsigned
+i915_translate_stencil_op(unsigned op)
+{
+   switch (op) {
+   case PIPE_STENCIL_OP_KEEP:
+      return STENCILOP_KEEP;
+   case PIPE_STENCIL_OP_ZERO:
+      return STENCILOP_ZERO;
+   case PIPE_STENCIL_OP_REPLACE:
+      return STENCILOP_REPLACE;
+   case PIPE_STENCIL_OP_INCR:
+      return STENCILOP_INCRSAT;
+   case PIPE_STENCIL_OP_DECR:
+      return STENCILOP_DECRSAT;
+   case PIPE_STENCIL_OP_INCR_WRAP:
+      return STENCILOP_INCR;
+   case PIPE_STENCIL_OP_DECR_WRAP:
+      return STENCILOP_DECR;
+   case PIPE_STENCIL_OP_INVERT:
+      return STENCILOP_INVERT;
+   default:
+      return STENCILOP_ZERO;
+   }
+}
+
+static INLINE unsigned
+i915_translate_blend_factor(unsigned factor)
+{
+   switch (factor) {
+   case PIPE_BLENDFACTOR_ZERO:
+      return BLENDFACT_ZERO;
+   case PIPE_BLENDFACTOR_SRC_ALPHA:
+      return BLENDFACT_SRC_ALPHA;
+   case PIPE_BLENDFACTOR_ONE:
+      return BLENDFACT_ONE;
+   case PIPE_BLENDFACTOR_SRC_COLOR:
+      return BLENDFACT_SRC_COLR;
+   case PIPE_BLENDFACTOR_INV_SRC_COLOR:
+      return BLENDFACT_INV_SRC_COLR;
+   case PIPE_BLENDFACTOR_DST_COLOR:
+      return BLENDFACT_DST_COLR;
+   case PIPE_BLENDFACTOR_INV_DST_COLOR:
+      return BLENDFACT_INV_DST_COLR;
+   case PIPE_BLENDFACTOR_INV_SRC_ALPHA:
+      return BLENDFACT_INV_SRC_ALPHA;
+   case PIPE_BLENDFACTOR_DST_ALPHA:
+      return BLENDFACT_DST_ALPHA;
+   case PIPE_BLENDFACTOR_INV_DST_ALPHA:
+      return BLENDFACT_INV_DST_ALPHA;
+   case PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE:
+      return BLENDFACT_SRC_ALPHA_SATURATE;
+   case PIPE_BLENDFACTOR_CONST_COLOR:
+      return BLENDFACT_CONST_COLOR;
+   case PIPE_BLENDFACTOR_INV_CONST_COLOR:
+      return BLENDFACT_INV_CONST_COLOR;
+   case PIPE_BLENDFACTOR_CONST_ALPHA:
+      return BLENDFACT_CONST_ALPHA;
+   case PIPE_BLENDFACTOR_INV_CONST_ALPHA:
+      return BLENDFACT_INV_CONST_ALPHA;
+   default:
+      return BLENDFACT_ZERO;
+   }
+}
+
+static INLINE unsigned
+i915_translate_blend_func(unsigned mode)
+{
+   switch (mode) {
+   case PIPE_BLEND_ADD:
+      return BLENDFUNC_ADD;
+   case PIPE_BLEND_MIN:
+      return BLENDFUNC_MIN;
+   case PIPE_BLEND_MAX:
+      return BLENDFUNC_MAX;
+   case PIPE_BLEND_SUBTRACT:
+      return BLENDFUNC_SUBTRACT;
+   case PIPE_BLEND_REVERSE_SUBTRACT:
+      return BLENDFUNC_REVERSE_SUBTRACT;
+   default:
+      return 0;
+   }
+}
+
+
+static INLINE unsigned
+i915_translate_logic_op(unsigned opcode)
+{
+   switch (opcode) {
+   case PIPE_LOGICOP_CLEAR:
+      return LOGICOP_CLEAR;
+   case PIPE_LOGICOP_AND:
+      return LOGICOP_AND;
+   case PIPE_LOGICOP_AND_REVERSE:
+      return LOGICOP_AND_RVRSE;
+   case PIPE_LOGICOP_COPY:
+      return LOGICOP_COPY;
+   case PIPE_LOGICOP_COPY_INVERTED:
+      return LOGICOP_COPY_INV;
+   case PIPE_LOGICOP_AND_INVERTED:
+      return LOGICOP_AND_INV;
+   case PIPE_LOGICOP_NOOP:
+      return LOGICOP_NOOP;
+   case PIPE_LOGICOP_XOR:
+      return LOGICOP_XOR;
+   case PIPE_LOGICOP_OR:
+      return LOGICOP_OR;
+   case PIPE_LOGICOP_OR_INVERTED:
+      return LOGICOP_OR_INV;
+   case PIPE_LOGICOP_NOR:
+      return LOGICOP_NOR;
+   case PIPE_LOGICOP_EQUIV:
+      return LOGICOP_EQUIV;
+   case PIPE_LOGICOP_INVERT:
+      return LOGICOP_INV;
+   case PIPE_LOGICOP_OR_REVERSE:
+      return LOGICOP_OR_RVRSE;
+   case PIPE_LOGICOP_NAND:
+      return LOGICOP_NAND;
+   case PIPE_LOGICOP_SET:
+      return LOGICOP_SET;
+   default:
+      return LOGICOP_SET;
+   }
+}
+
+
+
+static INLINE boolean i915_validate_vertices( unsigned hw_prim, unsigned nr )
+{
+   boolean ok;
+
+   switch (hw_prim) {
+   case PRIM3D_POINTLIST:
+      ok = (nr >= 1);
+      assert(ok);
+      break;
+   case PRIM3D_LINELIST:
+      ok = (nr >= 2) && (nr % 2) == 0;
+      assert(ok);
+      break;
+   case PRIM3D_LINESTRIP:
+      ok = (nr >= 2);
+      assert(ok);
+      break;
+   case PRIM3D_TRILIST:
+      ok = (nr >= 3) && (nr % 3) == 0;
+      assert(ok);
+      break;
+   case PRIM3D_TRISTRIP:
+      ok = (nr >= 3);
+      assert(ok);
+      break;
+   case PRIM3D_TRIFAN:
+      ok = (nr >= 3);
+      assert(ok);
+      break;
+   case PRIM3D_POLY:
+      ok = (nr >= 3);
+      assert(ok);
+      break;
+   default:
+      assert(0);
+      ok = 0;
+      break;
+   }
+
+   return ok;
+}
+
+#endif
diff --git a/src/gallium/drivers/i915simple/i915_state_sampler.c b/src/gallium/drivers/i915simple/i915_state_sampler.c
new file mode 100644
index 0000000000..9c1a5bbbd6
--- /dev/null
+++ b/src/gallium/drivers/i915simple/i915_state_sampler.c
@@ -0,0 +1,231 @@
+/**************************************************************************
+ * 
+ * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+#include "pipe/p_context.h"
+#include "pipe/p_state.h"
+#include "pipe/p_util.h"
+
+#include "i915_state_inlines.h"
+#include "i915_context.h"
+#include "i915_reg.h"
+#include "i915_state.h"
+
+
+/**
+ * Compute i915 texture sampling state.
+ *
+ * Recalculate all state from scratch.  Perhaps not the most
+ * efficient, but this has gotten complex enough that we need
+ * something which is understandable and reliable.
+ * \param state  returns the 3 words of compute state
+ */
+static void update_sampler(struct i915_context *i915,
+                           uint unit,
+			   const struct i915_sampler_state *sampler,
+			   const struct i915_texture *tex,
+			   unsigned state[3] )
+{
+   const struct pipe_texture *pt = &tex->base;
+
+   /* Need to do this after updating the maps, which call the
+    * intel_finalize_mipmap_tree and hence can update firstLevel:
+    */
+   state[0] = sampler->state[0];
+   state[1] = sampler->state[1];
+   state[2] = sampler->state[2];
+
+   if (pt->format == PIPE_FORMAT_YCBCR ||
+       pt->format == PIPE_FORMAT_YCBCR_REV)
+      state[0] |= SS2_COLORSPACE_CONVERSION;
+
+   /* 3D textures don't seem to respect the border color.
+    * Fallback if there's ever a danger that they might refer to
+    * it.  
+    * 
+    * Effectively this means fallback on 3D clamp or
+    * clamp_to_border.
+    *
+    * XXX: Check if this is true on i945.  
+    * XXX: Check if this bug got fixed in release silicon.
+    */
+#if 0
+   {
+      const unsigned ws = sampler->templ->wrap_s;
+      const unsigned wt = sampler->templ->wrap_t;
+      const unsigned wr = sampler->templ->wrap_r;
+      if (pt->target == PIPE_TEXTURE_3D &&
+          (sampler->templ->min_img_filter != PIPE_TEX_FILTER_NEAREST ||
+           sampler->templ->mag_img_filter != PIPE_TEX_FILTER_NEAREST) &&
+          (ws == PIPE_TEX_WRAP_CLAMP ||
+           wt == PIPE_TEX_WRAP_CLAMP ||
+           wr == PIPE_TEX_WRAP_CLAMP ||
+           ws == PIPE_TEX_WRAP_CLAMP_TO_BORDER ||
+           wt == PIPE_TEX_WRAP_CLAMP_TO_BORDER || 
+           wr == PIPE_TEX_WRAP_CLAMP_TO_BORDER)) {
+         if (i915->strict_conformance) {
+            assert(0);
+            /* 	    sampler->fallback = true; */
+            /* TODO */
+         }
+      }
+   }
+#endif
+
+   state[1] |= (unit << SS3_TEXTUREMAP_INDEX_SHIFT);
+}
+
+
+void i915_update_samplers( struct i915_context *i915 )
+{
+   uint unit;
+
+   i915->current.sampler_enable_nr = 0;
+   i915->current.sampler_enable_flags = 0x0;
+
+   for (unit = 0; unit < I915_TEX_UNITS; unit++) {
+      /* determine unit enable/disable by looking for a bound texture */
+      /* could also examine the fragment program? */
+      if (i915->texture[unit]) {
+	 update_sampler( i915,
+                         unit,
+                         i915->sampler[unit],       /* sampler state */
+                         i915->texture[unit],        /* texture */
+			 i915->current.sampler[unit] /* the result */
+                         );
+
+         i915->current.sampler_enable_nr++;
+         i915->current.sampler_enable_flags |= (1 << unit);
+      }
+   }
+
+   i915->hardware_dirty |= I915_HW_SAMPLER;
+}
+
+
+static uint
+translate_texture_format(enum pipe_format pipeFormat)
+{
+   switch (pipeFormat) {
+   case PIPE_FORMAT_U_L8:
+      return MAPSURF_8BIT | MT_8BIT_L8;
+   case PIPE_FORMAT_U_I8:
+      return MAPSURF_8BIT | MT_8BIT_I8;
+   case PIPE_FORMAT_U_A8:
+      return MAPSURF_8BIT | MT_8BIT_A8;
+   case PIPE_FORMAT_U_A8_L8:
+      return MAPSURF_16BIT | MT_16BIT_AY88;
+   case PIPE_FORMAT_R5G6B5_UNORM:
+      return MAPSURF_16BIT | MT_16BIT_RGB565;
+   case PIPE_FORMAT_A1R5G5B5_UNORM:
+      return MAPSURF_16BIT | MT_16BIT_ARGB1555;
+   case PIPE_FORMAT_A4R4G4B4_UNORM:
+      return MAPSURF_16BIT | MT_16BIT_ARGB4444;
+   case PIPE_FORMAT_A8R8G8B8_UNORM:
+      return MAPSURF_32BIT | MT_32BIT_ARGB8888;
+   case PIPE_FORMAT_YCBCR_REV:
+      return (MAPSURF_422 | MT_422_YCRCB_NORMAL);
+   case PIPE_FORMAT_YCBCR:
+      return (MAPSURF_422 | MT_422_YCRCB_SWAPY);
+#if 0
+   case PIPE_FORMAT_RGB_FXT1:
+   case PIPE_FORMAT_RGBA_FXT1:
+      return (MAPSURF_COMPRESSED | MT_COMPRESS_FXT1);
+#endif
+   case PIPE_FORMAT_Z16_UNORM:
+      return (MAPSURF_16BIT | MT_16BIT_L16);
+#if 0
+   case PIPE_FORMAT_RGBA_DXT1:
+   case PIPE_FORMAT_RGB_DXT1:
+      return (MAPSURF_COMPRESSED | MT_COMPRESS_DXT1);
+   case PIPE_FORMAT_RGBA_DXT3:
+      return (MAPSURF_COMPRESSED | MT_COMPRESS_DXT2_3);
+   case PIPE_FORMAT_RGBA_DXT5:
+      return (MAPSURF_COMPRESSED | MT_COMPRESS_DXT4_5);
+#endif
+   case PIPE_FORMAT_S8Z24_UNORM:
+      return (MAPSURF_32BIT | MT_32BIT_xL824);
+   default:
+      debug_printf("i915: translate_texture_format() bad image format %x\n",
+              pipeFormat);
+      assert(0);
+      return 0;
+   }
+}
+
+
+static void
+i915_update_texture(struct i915_context *i915, uint unit,
+                    uint state[6])
+{
+   const struct i915_texture *tex = i915->texture[unit];
+   const struct pipe_texture *pt = &tex->base;
+   uint format, pitch;
+   const uint width = pt->width[0], height = pt->height[0], depth = pt->depth[0];
+   const uint num_levels = pt->last_level;
+
+   assert(tex);
+   assert(width);
+   assert(height);
+   assert(depth);
+
+   format = translate_texture_format(pt->format);
+   pitch = tex->pitch * pt->cpp;
+
+   assert(format);
+   assert(pitch);
+
+   /* MS3 state */
+   state[0] =
+      (((height - 1) << MS3_HEIGHT_SHIFT)
+       | ((width - 1) << MS3_WIDTH_SHIFT)
+       | format
+       | MS3_USE_FENCE_REGS);
+
+   /* MS4 state */
+   state[1] =
+      ((((pitch / 4) - 1) << MS4_PITCH_SHIFT)
+       | MS4_CUBE_FACE_ENA_MASK
+       | ((num_levels * 4) << MS4_MAX_LOD_SHIFT)
+       | ((depth - 1) << MS4_VOLUME_DEPTH_SHIFT));
+}
+
+
+void
+i915_update_textures(struct i915_context *i915)
+{
+   uint unit;
+
+   for (unit = 0; unit < I915_TEX_UNITS; unit++) {
+      /* determine unit enable/disable by looking for a bound texture */
+      /* could also examine the fragment program? */
+      if (i915->texture[unit]) {
+         i915_update_texture(i915, unit, i915->current.texbuffer[unit]);
+      }
+   }
+
+   i915->hardware_dirty |= I915_HW_MAP;
+}
diff --git a/src/gallium/drivers/i915simple/i915_strings.c b/src/gallium/drivers/i915simple/i915_strings.c
new file mode 100644
index 0000000000..c713bf7208
--- /dev/null
+++ b/src/gallium/drivers/i915simple/i915_strings.c
@@ -0,0 +1,83 @@
+/**************************************************************************
+ * 
+ * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+#include "i915_context.h"
+#include "i915_reg.h"
+
+
+static const char *i915_get_vendor( struct pipe_context *pipe )
+{
+   return "Tungsten Graphics, Inc.";
+}
+
+
+static const char *i915_get_name( struct pipe_context *pipe )
+{
+   static char buffer[128];
+   const char *chipset;
+
+   switch (i915_context(pipe)->pci_id) {
+   case PCI_CHIP_I915_G:
+      chipset = "915G";
+      break;
+   case PCI_CHIP_I915_GM:
+      chipset = "915GM";
+      break;
+   case PCI_CHIP_I945_G:
+      chipset = "945G";
+      break;
+   case PCI_CHIP_I945_GM:
+      chipset = "945GM";
+      break;
+   case PCI_CHIP_I945_GME:
+      chipset = "945GME";
+      break;
+   case PCI_CHIP_G33_G:
+      chipset = "G33";
+      break;
+   case PCI_CHIP_Q35_G:
+      chipset = "Q35";
+      break;
+   case PCI_CHIP_Q33_G:
+      chipset = "Q33";
+      break;
+   default:
+      chipset = "unknown";
+      break;
+   }
+
+   sprintf(buffer, "pipe/i915 (chipset: %s)", chipset);
+   return buffer;
+}
+
+
+void
+i915_init_string_functions(struct i915_context *i915)
+{
+   i915->pipe.get_name = i915_get_name;
+   i915->pipe.get_vendor = i915_get_vendor;
+}
diff --git a/src/gallium/drivers/i915simple/i915_surface.c b/src/gallium/drivers/i915simple/i915_surface.c
new file mode 100644
index 0000000000..de0cc5fe06
--- /dev/null
+++ b/src/gallium/drivers/i915simple/i915_surface.c
@@ -0,0 +1,191 @@
+/**************************************************************************
+ * 
+ * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+#include "i915_context.h"
+#include "i915_blit.h"
+#include "i915_state.h"
+#include "pipe/p_defines.h"
+#include "pipe/p_inlines.h"
+#include "pipe/p_util.h"
+#include "pipe/p_inlines.h"
+#include "pipe/p_winsys.h"
+#include "pipe/util/p_tile.h"
+
+
+/*
+ * XXX note: same as code in sp_surface.c
+ */
+static struct pipe_surface *
+i915_get_tex_surface(struct pipe_context *pipe,
+                     struct pipe_texture *pt,
+                     unsigned face, unsigned level, unsigned zslice)
+{
+   struct i915_texture *tex = (struct i915_texture *)pt;
+   struct pipe_surface *ps;
+   unsigned offset;  /* in bytes */
+
+   offset = tex->level_offset[level];
+
+   if (pt->target == PIPE_TEXTURE_CUBE) {
+      offset += tex->image_offset[level][face] * pt->cpp;
+   }
+   else if (pt->target == PIPE_TEXTURE_3D) {
+      offset += tex->image_offset[level][zslice] * pt->cpp;
+   }
+   else {
+      assert(face == 0);
+      assert(zslice == 0);
+   }
+
+   ps = pipe->winsys->surface_alloc(pipe->winsys);
+   if (ps) {
+      assert(ps->refcount);
+      assert(ps->winsys);
+      pipe_buffer_reference(pipe->winsys, &ps->buffer, tex->buffer);
+      ps->format = pt->format;
+      ps->cpp = pt->cpp;
+      ps->width = pt->width[level];
+      ps->height = pt->height[level];
+      ps->pitch = tex->pitch;
+      ps->offset = offset;
+   }
+   return ps;
+}
+
+
+
+/* Assumes all values are within bounds -- no checking at this level -
+ * do it higher up if required.
+ */
+static void
+i915_surface_copy(struct pipe_context *pipe,
+                  unsigned do_flip,
+		  struct pipe_surface *dst,
+		  unsigned dstx, unsigned dsty,
+		  struct pipe_surface *src,
+		  unsigned srcx, unsigned srcy, unsigned width, unsigned height)
+{
+   assert( dst != src );
+   assert( dst->cpp == src->cpp );
+
+   if (0) {
+      pipe_copy_rect(pipe_surface_map(dst),
+                     dst->cpp,
+                     dst->pitch,
+                     dstx, dsty, 
+                     width, height, 
+                     pipe_surface_map(src), 
+                     do_flip ? -(int) src->pitch : src->pitch, 
+                     srcx, do_flip ? 1 - srcy - height : srcy);
+
+      pipe_surface_unmap(src);
+      pipe_surface_unmap(dst);
+   }
+   else {
+      i915_copy_blit( i915_context(pipe),
+                      do_flip,
+		      dst->cpp,
+		      (short) src->pitch, src->buffer, src->offset,
+		      (short) dst->pitch, dst->buffer, dst->offset,
+		      (short) srcx, (short) srcy, (short) dstx, (short) dsty, (short) width, (short) height );
+   }
+}
+
+/* Fill a rectangular sub-region.  Need better logic about when to
+ * push buffers into AGP - will currently do so whenever possible.
+ */
+static void *
+get_pointer(struct pipe_surface *dst, void *dst_map, unsigned x, unsigned y)
+{
+   return (char *)dst_map + (y * dst->pitch + x) * dst->cpp;
+}
+
+
+static void
+i915_surface_fill(struct pipe_context *pipe,
+		  struct pipe_surface *dst,
+		  unsigned dstx, unsigned dsty,
+		  unsigned width, unsigned height, unsigned value)
+{
+   if (0) {
+      unsigned i, j;
+      void *dst_map = pipe_surface_map(dst);
+
+      switch (dst->cpp) {
+      case 1: {
+	 ubyte *row = get_pointer(dst, dst_map, dstx, dsty);
+	 for (i = 0; i < height; i++) {
+	    memset(row, value, width);
+	    row += dst->pitch;
+	 }
+      }
+	 break;
+      case 2: {
+	 ushort *row = get_pointer(dst, dst_map, dstx, dsty);
+	 for (i = 0; i < height; i++) {
+	    for (j = 0; j < width; j++)
+	       row[j] = (ushort) value;
+	    row += dst->pitch;
+	 }
+      }
+	 break;
+      case 4: {
+	 unsigned *row = get_pointer(dst, dst_map, dstx, dsty);
+	 for (i = 0; i < height; i++) {
+	    for (j = 0; j < width; j++)
+	       row[j] = value;
+	    row += dst->pitch;
+	 }
+      }
+	 break;
+      default:
+	 assert(0);
+	 break;
+      }
+
+      pipe_surface_unmap( dst );
+   }
+   else {
+      i915_fill_blit( i915_context(pipe),
+		      dst->cpp,
+		      (short) dst->pitch, 
+		      dst->buffer, dst->offset,
+		      (short) dstx, (short) dsty, 
+		      (short) width, (short) height, 
+		      value );
+   }
+}
+
+
+void
+i915_init_surface_functions(struct i915_context *i915)
+{
+   i915->pipe.get_tex_surface = i915_get_tex_surface;
+
+   i915->pipe.surface_copy = i915_surface_copy;
+   i915->pipe.surface_fill = i915_surface_fill;
+}
diff --git a/src/gallium/drivers/i915simple/i915_texture.c b/src/gallium/drivers/i915simple/i915_texture.c
new file mode 100644
index 0000000000..6d37ae3d74
--- /dev/null
+++ b/src/gallium/drivers/i915simple/i915_texture.c
@@ -0,0 +1,536 @@
+/**************************************************************************
+ * 
+ * Copyright 2006 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+ /*
+  * Authors:
+  *   Keith Whitwell <keith@tungstengraphics.com>
+  *   Michel Dänzer <michel@tungstengraphics.com>
+  */
+
+#include "pipe/p_state.h"
+#include "pipe/p_context.h"
+#include "pipe/p_defines.h"
+#include "pipe/p_inlines.h"
+#include "pipe/p_util.h"
+#include "pipe/p_winsys.h"
+
+#include "i915_context.h"
+#include "i915_texture.h"
+#include "i915_debug.h"
+
+
+static unsigned minify( unsigned d )
+{
+   return MAX2(1, d>>1);
+}
+
+
+
+static void
+i915_miptree_set_level_info(struct i915_texture *tex,
+                             unsigned level,
+                             unsigned nr_images,
+                             unsigned x, unsigned y, unsigned w, unsigned h, unsigned d)
+{
+   struct pipe_texture *pt = &tex->base;
+
+   assert(level < PIPE_MAX_TEXTURE_LEVELS);
+
+   pt->width[level] = w;
+   pt->height[level] = h;
+   pt->depth[level] = d;
+
+   tex->level_offset[level] = (x + y * tex->pitch) * pt->cpp;
+   tex->nr_images[level] = nr_images;
+
+   /*
+   DBG("%s level %d size: %d,%d,%d offset %d,%d (0x%x)\n", __FUNCTION__,
+       level, w, h, d, x, y, tex->level_offset[level]);
+   */
+
+   /* Not sure when this would happen, but anyway: 
+    */
+   if (tex->image_offset[level]) {
+      FREE(tex->image_offset[level]);
+      tex->image_offset[level] = NULL;
+   }
+
+   assert(nr_images);
+   assert(!tex->image_offset[level]);
+
+   tex->image_offset[level] = (unsigned *) MALLOC(nr_images * sizeof(unsigned));
+   tex->image_offset[level][0] = 0;
+}
+
+
+static void
+i915_miptree_set_image_offset(struct i915_texture *tex,
+			      unsigned level, unsigned img, unsigned x, unsigned y)
+{
+   if (img == 0 && level == 0)
+      assert(x == 0 && y == 0);
+
+   assert(img < tex->nr_images[level]);
+
+   tex->image_offset[level][img] = (x + y * tex->pitch);
+
+   /*
+   DBG("%s level %d img %d pos %d,%d image_offset %x\n",
+       __FUNCTION__, level, img, x, y, tex->image_offset[level][img]);
+   */
+}
+
+
+static void
+i945_miptree_layout_2d( struct i915_texture *tex )
+{
+   struct pipe_texture *pt = &tex->base;
+   int align_h = 2, align_w = 4;
+   unsigned level;
+   unsigned x = 0;
+   unsigned y = 0;
+   unsigned width = pt->width[0];
+   unsigned height = pt->height[0];
+
+   tex->pitch = pt->width[0];
+
+   /* May need to adjust pitch to accomodate the placement of
+    * the 2nd mipmap level.  This occurs when the alignment
+    * constraints of mipmap placement push the right edge of the
+    * 2nd mipmap level out past the width of its parent.
+    */
+   if (pt->last_level > 0) {
+      unsigned mip1_width = align_int(minify(pt->width[0]), align_w)
+			+ minify(minify(pt->width[0]));
+
+      if (mip1_width > pt->width[0])
+	 tex->pitch = mip1_width;
+   }
+
+   /* Pitch must be a whole number of dwords, even though we
+    * express it in texels.
+    */
+   tex->pitch = align_int(tex->pitch * pt->cpp, 4) / pt->cpp;
+   tex->total_height = 0;
+
+   for (level = 0; level <= pt->last_level; level++) {
+      unsigned img_height;
+
+      i915_miptree_set_level_info(tex, level, 1, x, y, width, height, 1);
+
+      if (pt->compressed)
+	 img_height = MAX2(1, height/4);
+      else
+	 img_height = align_int(height, align_h);
+
+
+      /* Because the images are packed better, the final offset
+       * might not be the maximal one:
+       */
+      tex->total_height = MAX2(tex->total_height, y + img_height);
+
+      /* Layout_below: step right after second mipmap level.
+       */
+      if (level == 1) {
+	 x += align_int(width, align_w);
+      }
+      else {
+	 y += img_height;
+      }
+
+      width  = minify(width);
+      height = minify(height);
+   }
+}
+
+
+static const int initial_offsets[6][2] = {
+   {0, 0},
+   {0, 2},
+   {1, 0},
+   {1, 2},
+   {1, 1},
+   {1, 3}
+};
+
+static const int step_offsets[6][2] = {
+   {0, 2},
+   {0, 2},
+   {-1, 2},
+   {-1, 2},
+   {-1, 1},
+   {-1, 1}
+};
+
+
+static boolean
+i915_miptree_layout(struct pipe_context *pipe, struct i915_texture * tex)
+{
+   struct pipe_texture *pt = &tex->base;
+   unsigned level;
+
+   switch (pt->target) {
+   case PIPE_TEXTURE_CUBE: {
+         const unsigned dim = pt->width[0];
+         unsigned face;
+         unsigned lvlWidth = pt->width[0], lvlHeight = pt->height[0];
+
+         assert(lvlWidth == lvlHeight); /* cubemap images are square */
+
+         /* double pitch for cube layouts */
+         tex->pitch = ((dim * pt->cpp * 2 + 3) & ~3) / pt->cpp;
+         tex->total_height = dim * 4;
+
+         for (level = 0; level <= pt->last_level; level++) {
+            i915_miptree_set_level_info(tex, level, 6,
+                                         0, 0,
+                                         /*OLD: tex->pitch, tex->total_height,*/
+                                         lvlWidth, lvlHeight,
+                                         1);
+            lvlWidth /= 2;
+            lvlHeight /= 2;
+         }
+
+         for (face = 0; face < 6; face++) {
+            unsigned x = initial_offsets[face][0] * dim;
+            unsigned y = initial_offsets[face][1] * dim;
+            unsigned d = dim;
+
+            for (level = 0; level <= pt->last_level; level++) {
+               i915_miptree_set_image_offset(tex, level, face, x, y);
+               d >>= 1;
+               x += step_offsets[face][0] * d;
+               y += step_offsets[face][1] * d;
+            }
+         }
+         break;
+      }
+   case PIPE_TEXTURE_3D:{
+         unsigned width = pt->width[0];
+         unsigned height = pt->height[0];
+         unsigned depth = pt->depth[0];
+         unsigned stack_height = 0;
+
+         /* Calculate the size of a single slice. 
+          */
+         tex->pitch = ((pt->width[0] * pt->cpp + 3) & ~3) / pt->cpp;
+
+         /* XXX: hardware expects/requires 9 levels at minimum.
+          */
+         for (level = 0; level <= MAX2(8, pt->last_level);
+              level++) {
+            i915_miptree_set_level_info(tex, level, depth, 0, tex->total_height,
+                                         width, height, depth);
+
+
+            stack_height += MAX2(2, height);
+
+            width = minify(width);
+            height = minify(height);
+            depth = minify(depth);
+         }
+
+         /* Fixup depth image_offsets: 
+          */
+         depth = pt->depth[0];
+         for (level = 0; level <= pt->last_level; level++) {
+            unsigned i;
+            for (i = 0; i < depth; i++) 
+               i915_miptree_set_image_offset(tex, level, i,
+                                              0, i * stack_height);
+
+            depth = minify(depth);
+         }
+
+
+         /* Multiply slice size by texture depth for total size.  It's
+          * remarkable how wasteful of memory the i915 texture layouts
+          * are.  They are largely fixed in the i945.
+          */
+         tex->total_height = stack_height * pt->depth[0];
+         break;
+      }
+
+   default:{
+         unsigned width = pt->width[0];
+         unsigned height = pt->height[0];
+	 unsigned img_height;
+
+         tex->pitch = ((pt->width[0] * pt->cpp + 3) & ~3) / pt->cpp;
+         tex->total_height = 0;
+
+         for (level = 0; level <= pt->last_level; level++) {
+            i915_miptree_set_level_info(tex, level, 1,
+                                         0, tex->total_height,
+                                         width, height, 1);
+
+            if (pt->compressed)
+               img_height = MAX2(1, height / 4);
+            else
+               img_height = (MAX2(2, height) + 1) & ~1;
+
+	    tex->total_height += img_height;
+
+            width = minify(width);
+            height = minify(height);
+         }
+         break;
+      }
+   }
+   /*
+   DBG("%s: %dx%dx%d - sz 0x%x\n", __FUNCTION__,
+       tex->pitch,
+       tex->total_height, pt->cpp, tex->pitch * tex->total_height * pt->cpp);
+   */
+
+   return TRUE;
+}
+
+
+static boolean
+i945_miptree_layout(struct pipe_context *pipe, struct i915_texture * tex)
+{
+   struct pipe_texture *pt = &tex->base;
+   unsigned level;
+
+   switch (pt->target) {
+   case PIPE_TEXTURE_CUBE:{
+         const unsigned dim = pt->width[0];
+         unsigned face;
+         unsigned lvlWidth = pt->width[0], lvlHeight = pt->height[0];
+
+         assert(lvlWidth == lvlHeight); /* cubemap images are square */
+
+         /* Depending on the size of the largest images, pitch can be
+          * determined either by the old-style packing of cubemap faces,
+          * or the final row of 4x4, 2x2 and 1x1 faces below this. 
+          */
+         if (dim > 32)
+            tex->pitch = ((dim * pt->cpp * 2 + 3) & ~3) / pt->cpp;
+         else
+            tex->pitch = 14 * 8;
+
+         tex->total_height = dim * 4 + 4;
+
+         /* Set all the levels to effectively occupy the whole rectangular region. 
+          */
+         for (level = 0; level <= pt->last_level; level++) {
+            i915_miptree_set_level_info(tex, level, 6,
+                                         0, 0,
+                                         lvlWidth, lvlHeight, 1);
+	    lvlWidth /= 2;
+	    lvlHeight /= 2;
+	 }
+
+
+         for (face = 0; face < 6; face++) {
+            unsigned x = initial_offsets[face][0] * dim;
+            unsigned y = initial_offsets[face][1] * dim;
+            unsigned d = dim;
+
+            if (dim == 4 && face >= 4) {
+               y = tex->total_height - 4;
+               x = (face - 4) * 8;
+            }
+            else if (dim < 4 && (face > 0)) {
+               y = tex->total_height - 4;
+               x = face * 8;
+            }
+
+            for (level = 0; level <= pt->last_level; level++) {
+               i915_miptree_set_image_offset(tex, level, face, x, y);
+
+               d >>= 1;
+
+               switch (d) {
+               case 4:
+                  switch (face) {
+                  case PIPE_TEX_FACE_POS_X:
+                  case PIPE_TEX_FACE_NEG_X:
+                     x += step_offsets[face][0] * d;
+                     y += step_offsets[face][1] * d;
+                     break;
+                  case PIPE_TEX_FACE_POS_Y:
+                  case PIPE_TEX_FACE_NEG_Y:
+                     y += 12;
+                     x -= 8;
+                     break;
+                  case PIPE_TEX_FACE_POS_Z:
+                  case PIPE_TEX_FACE_NEG_Z:
+                     y = tex->total_height - 4;
+                     x = (face - 4) * 8;
+                     break;
+                  }
+
+               case 2:
+                  y = tex->total_height - 4;
+                  x = 16 + face * 8;
+                  break;
+
+               case 1:
+                  x += 48;
+                  break;
+
+               default:
+                  x += step_offsets[face][0] * d;
+                  y += step_offsets[face][1] * d;
+                  break;
+               }
+            }
+         }
+         break;
+      }
+   case PIPE_TEXTURE_3D:{
+         unsigned width = pt->width[0];
+         unsigned height = pt->height[0];
+         unsigned depth = pt->depth[0];
+         unsigned pack_x_pitch, pack_x_nr;
+         unsigned pack_y_pitch;
+         unsigned level;
+
+         tex->pitch = ((pt->width[0] * pt->cpp + 3) & ~3) / pt->cpp;
+         tex->total_height = 0;
+
+         pack_y_pitch = MAX2(pt->height[0], 2);
+         pack_x_pitch = tex->pitch;
+         pack_x_nr = 1;
+
+         for (level = 0; level <= pt->last_level; level++) {
+            unsigned nr_images = pt->target == PIPE_TEXTURE_3D ? depth : 6;
+            int x = 0;
+            int y = 0;
+            unsigned q, j;
+
+            i915_miptree_set_level_info(tex, level, nr_images,
+                                         0, tex->total_height,
+                                         width, height, depth);
+
+            for (q = 0; q < nr_images;) {
+               for (j = 0; j < pack_x_nr && q < nr_images; j++, q++) {
+                  i915_miptree_set_image_offset(tex, level, q, x, y);
+                  x += pack_x_pitch;
+               }
+
+               x = 0;
+               y += pack_y_pitch;
+            }
+
+
+            tex->total_height += y;
+
+            if (pack_x_pitch > 4) {
+               pack_x_pitch >>= 1;
+               pack_x_nr <<= 1;
+               assert(pack_x_pitch * pack_x_nr <= tex->pitch);
+            }
+
+            if (pack_y_pitch > 2) {
+               pack_y_pitch >>= 1;
+            }
+
+            width = minify(width);
+            height = minify(height);
+            depth = minify(depth);
+         }
+         break;
+      }
+
+   case PIPE_TEXTURE_1D:
+   case PIPE_TEXTURE_2D:
+//   case PIPE_TEXTURE_RECTANGLE:
+         i945_miptree_layout_2d(tex);
+         break;
+   default:
+      assert(0);
+      return FALSE;
+   }
+
+   /*
+   DBG("%s: %dx%dx%d - sz 0x%x\n", __FUNCTION__,
+       tex->pitch,
+       tex->total_height, pt->cpp, tex->pitch * tex->total_height * pt->cpp);
+   */
+
+   return TRUE;
+}
+
+
+struct pipe_texture *
+i915_texture_create(struct pipe_context *pipe,
+                    const struct pipe_texture *templat)
+{
+   struct i915_texture *tex = CALLOC_STRUCT(i915_texture);
+
+   if (tex) {
+      struct i915_context *i915 = i915_context(pipe);
+
+      tex->base = *templat;
+
+      if (i915->flags.is_i945 ? i945_miptree_layout(pipe, tex) :
+	  i915_miptree_layout(pipe, tex))
+	 tex->buffer = pipe->winsys->buffer_create(pipe->winsys, 64,
+                                                   PIPE_BUFFER_USAGE_PIXEL,
+                                                   tex->pitch * tex->base.cpp *
+                                                   tex->total_height);
+
+      if (!tex->buffer) {
+	 FREE(tex);
+	 return NULL;
+      }
+   }
+
+   return &tex->base;
+}
+
+
+void
+i915_texture_release(struct pipe_context *pipe, struct pipe_texture **pt)
+{
+   if (!*pt)
+      return;
+
+   /*
+   DBG("%s %p refcount will be %d\n",
+       __FUNCTION__, (void *) *pt, (*pt)->refcount - 1);
+   */
+   if (--(*pt)->refcount <= 0) {
+      struct i915_texture *tex = (struct i915_texture *)*pt;
+      uint i;
+
+      /*
+      DBG("%s deleting %p\n", __FUNCTION__, (void *) tex);
+      */
+
+      pipe_buffer_reference(pipe->winsys, &tex->buffer, NULL);
+
+      for (i = 0; i < PIPE_MAX_TEXTURE_LEVELS; i++)
+         if (tex->image_offset[i])
+            FREE(tex->image_offset[i]);
+
+      FREE(tex);
+   }
+   *pt = NULL;
+}
diff --git a/src/gallium/drivers/i915simple/i915_texture.h b/src/gallium/drivers/i915simple/i915_texture.h
new file mode 100644
index 0000000000..330d111dc7
--- /dev/null
+++ b/src/gallium/drivers/i915simple/i915_texture.h
@@ -0,0 +1,17 @@
+
+#ifndef I915_TEXTURE_H
+#define I915_TEXTURE_H
+
+struct pipe_context;
+struct pipe_texture;
+
+
+struct pipe_texture *
+i915_texture_create(struct pipe_context *pipe,
+                    const struct pipe_texture *templat);
+
+extern void
+i915_texture_release(struct pipe_context *pipe, struct pipe_texture **pt);
+
+
+#endif /* I915_TEXTURE_H */
diff --git a/src/gallium/drivers/i915simple/i915_winsys.h b/src/gallium/drivers/i915simple/i915_winsys.h
new file mode 100644
index 0000000000..fe49710852
--- /dev/null
+++ b/src/gallium/drivers/i915simple/i915_winsys.h
@@ -0,0 +1,115 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+/**
+ * \file
+ * This is the interface that i915simple requires any window system
+ * hosting it to implement.  This is the only include file in i915simple
+ * which is public.
+ * 
+ */
+
+#ifndef I915_WINSYS_H
+#define I915_WINSYS_H
+
+
+#include "pipe/p_defines.h"
+
+
+/* Pipe drivers are (meant to be!) independent of both GL and the
+ * window system.  The window system provides a buffer manager and a
+ * set of additional hooks for things like command buffer submission,
+ * etc.
+ *
+ * There clearly has to be some agreement between the window system
+ * driver and the hardware driver about the format of command buffers,
+ * etc.
+ */
+
+struct pipe_buffer;
+struct pipe_winsys;
+
+
+/**
+ * Additional winsys interface for i915simple.
+ * 
+ * It is an over-simple batchbuffer mechanism.  Will want to improve the
+ * performance of this, perhaps based on the cmdstream stuff.  It
+ * would be pretty impossible to implement swz on top of this
+ * interface.
+ *
+ * Will also need additions/changes to implement static/dynamic
+ * indirect state.
+ */
+struct i915_winsys {
+
+   /**
+    * Reserve space on batch buffer. 
+    * 
+    * Returns a null pointer if there is insufficient space in the batch buffer 
+    * to hold the requested number of dwords and relocations.
+    * 
+    * The number of dwords should also include the number of relocations.
+    */
+   unsigned *(*batch_start)( struct i915_winsys *sws,
+			     unsigned dwords,
+			     unsigned relocs );
+   
+   void (*batch_dword)( struct i915_winsys *sws,
+			unsigned dword );
+   
+   /**
+    * Emit a relocation to a buffer.
+    * 
+    * Used not only when the buffer addresses are not pinned, but also to 
+    * ensure refered buffers will not be destroyed until the current batch 
+    * buffer execution is finished.
+    *
+    * The access flags is a combination of I915_BUFFER_ACCESS_WRITE and 
+    * I915_BUFFER_ACCESS_READ macros.
+    */
+   void (*batch_reloc)( struct i915_winsys *sws,
+			struct pipe_buffer *buf,
+			unsigned access_flags,
+			unsigned delta );
+   
+   void (*batch_flush)( struct i915_winsys *sws );
+   void (*batch_finish)( struct i915_winsys *sws );
+};
+
+#define I915_BUFFER_ACCESS_WRITE   0x1 
+#define I915_BUFFER_ACCESS_READ    0x2
+
+#define I915_BUFFER_USAGE_LIT_VERTEX  (PIPE_BUFFER_USAGE_CUSTOM << 0)
+
+
+struct pipe_context *i915_create( struct pipe_winsys *,
+				  struct i915_winsys *,
+				  unsigned pci_id );
+
+
+#endif 
diff --git a/src/gallium/drivers/i965simple/Makefile b/src/gallium/drivers/i965simple/Makefile
new file mode 100644
index 0000000000..48c00ab50b
--- /dev/null
+++ b/src/gallium/drivers/i965simple/Makefile
@@ -0,0 +1,66 @@
+
+TOP = ../../../..
+include $(TOP)/configs/current
+
+LIBNAME = i965simple
+
+DRIVER_SOURCES = \
+        brw_blit.c \
+        brw_flush.c \
+        brw_strings.c \
+        brw_surface.c \
+	brw_cc.c \
+	brw_clip.c \
+	brw_clip_line.c \
+	brw_clip_point.c \
+	brw_clip_state.c \
+	brw_clip_tri.c \
+	brw_clip_util.c \
+	brw_context.c \
+	brw_curbe.c \
+	brw_draw.c \
+	brw_draw_upload.c \
+	brw_eu.c \
+	brw_eu_debug.c \
+	brw_eu_emit.c \
+	brw_eu_util.c \
+	brw_gs.c \
+	brw_gs_emit.c \
+	brw_gs_state.c \
+	brw_misc_state.c \
+	brw_sf.c \
+	brw_sf_emit.c \
+	brw_sf_state.c \
+        brw_shader_info.c \
+        brw_state.c \
+	brw_state_batch.c \
+	brw_state_cache.c \
+	brw_state_pool.c \
+	brw_state_upload.c \
+	brw_tex_layout.c \
+	brw_urb.c \
+	brw_util.c \
+	brw_vs.c \
+	brw_vs_emit.c \
+	brw_vs_state.c \
+	brw_wm.c \
+	brw_wm_iz.c \
+	brw_wm_decl.c \
+	brw_wm_glsl.c \
+	brw_wm_sampler_state.c \
+	brw_wm_state.c \
+	brw_wm_surface_state.c
+
+C_SOURCES = \
+	$(COMMON_SOURCES) \
+	$(COMMON_BM_SOURCES) \
+	$(MINIGLX_SOURCES) \
+	$(DRIVER_SOURCES)
+
+ASM_SOURCES =
+
+DRIVER_DEFINES = -I.
+
+include ../Makefile.template
+
+symlinks:
diff --git a/src/gallium/drivers/i965simple/SConscript b/src/gallium/drivers/i965simple/SConscript
new file mode 100644
index 0000000000..74621de84c
--- /dev/null
+++ b/src/gallium/drivers/i965simple/SConscript
@@ -0,0 +1,55 @@
+Import('*')
+
+env = env.Clone()
+
+i965simple = env.ConvenienceLibrary(
+	target = 'i965simple',
+	source = [
+		'brw_blit.c',
+		'brw_cc.c',
+		'brw_clip.c',
+		'brw_clip_line.c',
+		'brw_clip_point.c',
+		'brw_clip_state.c',
+		'brw_clip_tri.c',
+		'brw_clip_util.c',
+		'brw_context.c',
+		'brw_curbe.c',
+		'brw_draw.c',
+		'brw_draw_upload.c',
+		'brw_eu.c',
+		'brw_eu_debug.c',
+		'brw_eu_emit.c',
+		'brw_eu_util.c',
+		'brw_flush.c',
+		'brw_gs.c',
+		'brw_gs_emit.c',
+		'brw_gs_state.c',
+		'brw_misc_state.c',
+		'brw_sf.c',
+		'brw_sf_emit.c',
+		'brw_sf_state.c',
+		'brw_shader_info.c',
+		'brw_state.c',
+		'brw_state_batch.c',
+		'brw_state_cache.c',
+		'brw_state_pool.c',
+		'brw_state_upload.c',
+		'brw_strings.c',
+		'brw_surface.c',
+		'brw_tex_layout.c',
+		'brw_urb.c',
+		'brw_util.c',
+		'brw_vs.c',
+		'brw_vs_emit.c',
+		'brw_vs_state.c',
+		'brw_wm.c',
+		'brw_wm_decl.c',
+		'brw_wm_glsl.c',
+		'brw_wm_iz.c',
+		'brw_wm_sampler_state.c',
+		'brw_wm_state.c',
+		'brw_wm_surface_state.c',
+	])
+
+Export('i965simple')
diff --git a/src/gallium/drivers/i965simple/brw_batch.h b/src/gallium/drivers/i965simple/brw_batch.h
new file mode 100644
index 0000000000..5f5932a488
--- /dev/null
+++ b/src/gallium/drivers/i965simple/brw_batch.h
@@ -0,0 +1,59 @@
+/**************************************************************************
+ *
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ **************************************************************************/
+
+#ifndef BRW_BATCH_H
+#define BRW_BATCH_H
+
+#include "brw_winsys.h"
+
+#define BATCH_LOCALS
+
+#define INTEL_BATCH_NO_CLIPRECTS 0x1
+#define INTEL_BATCH_CLIPRECTS    0x2
+
+#define BEGIN_BATCH( dwords, relocs ) \
+   brw->winsys->batch_start(brw->winsys, dwords, relocs)
+
+#define OUT_BATCH( dword ) \
+   brw->winsys->batch_dword(brw->winsys, dword)
+
+#define OUT_RELOC( buf, flags, delta ) \
+   brw->winsys->batch_reloc(brw->winsys, buf, flags, delta)
+
+#define ADVANCE_BATCH() \
+   brw->winsys->batch_end( brw->winsys )
+
+/* XXX: this is bogus - need proper handling for out-of-memory in batchbuffer.
+ */
+#define FLUSH_BATCH(fence) do {				\
+   brw->winsys->batch_flush(brw->winsys, fence);	\
+   brw->hardware_dirty = ~0;				\
+} while (0)
+
+#define BRW_BATCH_STRUCT(brw, s) brw_batchbuffer_data( brw->winsys, (s), sizeof(*(s)))
+
+#endif
diff --git a/src/gallium/drivers/i965simple/brw_blit.c b/src/gallium/drivers/i965simple/brw_blit.c
new file mode 100644
index 0000000000..8494f70493
--- /dev/null
+++ b/src/gallium/drivers/i965simple/brw_blit.c
@@ -0,0 +1,218 @@
+/**************************************************************************
+ *
+ * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ **************************************************************************/
+
+
+#include <stdio.h>
+#include <errno.h>
+
+#include "brw_batch.h"
+#include "brw_blit.h"
+#include "brw_context.h"
+#include "brw_reg.h"
+
+#include "pipe/p_context.h"
+#include "pipe/p_winsys.h"
+
+#define FILE_DEBUG_FLAG DEBUG_BLIT
+
+void brw_fill_blit(struct brw_context *brw,
+                   unsigned cpp,
+                   short dst_pitch,
+                   struct pipe_buffer *dst_buffer,
+                   unsigned dst_offset,
+                   boolean dst_tiled,
+                   short x, short y,
+                   short w, short h,
+                   unsigned color)
+{
+   unsigned BR13, CMD;
+   BATCH_LOCALS;
+
+   dst_pitch *= cpp;
+
+   switch(cpp) {
+   case 1:
+   case 2:
+   case 3:
+      BR13 = (0xF0 << 16) | (1<<24);
+      CMD = XY_COLOR_BLT_CMD;
+      break;
+   case 4:
+      BR13 = (0xF0 << 16) | (1<<24) | (1<<25);
+      CMD = XY_COLOR_BLT_CMD | XY_BLT_WRITE_ALPHA | XY_BLT_WRITE_RGB;
+      break;
+   default:
+      return;
+   }
+
+   if (dst_tiled) {
+      CMD |= XY_DST_TILED;
+      dst_pitch /= 4;
+   }
+
+   BEGIN_BATCH(6, INTEL_BATCH_NO_CLIPRECTS);
+   OUT_BATCH( CMD );
+   OUT_BATCH( dst_pitch | BR13 );
+   OUT_BATCH( (y << 16) | x );
+   OUT_BATCH( ((y+h) << 16) | (x+w) );
+   OUT_RELOC( dst_buffer, BRW_BUFFER_ACCESS_WRITE, dst_offset );
+   OUT_BATCH( color );
+   ADVANCE_BATCH();
+}
+
+static unsigned translate_raster_op(unsigned logicop)
+{
+   switch(logicop) {
+   case PIPE_LOGICOP_CLEAR: return 0x00;
+   case PIPE_LOGICOP_AND: return 0x88;
+   case PIPE_LOGICOP_AND_REVERSE: return 0x44;
+   case PIPE_LOGICOP_COPY: return 0xCC;
+   case PIPE_LOGICOP_AND_INVERTED: return 0x22;
+   case PIPE_LOGICOP_NOOP: return 0xAA;
+   case PIPE_LOGICOP_XOR: return 0x66;
+   case PIPE_LOGICOP_OR: return 0xEE;
+   case PIPE_LOGICOP_NOR: return 0x11;
+   case PIPE_LOGICOP_EQUIV: return 0x99;
+   case PIPE_LOGICOP_INVERT: return 0x55;
+   case PIPE_LOGICOP_OR_REVERSE: return 0xDD;
+   case PIPE_LOGICOP_COPY_INVERTED: return 0x33;
+   case PIPE_LOGICOP_OR_INVERTED: return 0xBB;
+   case PIPE_LOGICOP_NAND: return 0x77;
+   case PIPE_LOGICOP_SET: return 0xFF;
+   default: return 0;
+   }
+}
+
+
+/* Copy BitBlt
+ */
+void brw_copy_blit(struct brw_context *brw,
+                   unsigned do_flip,
+                   unsigned cpp,
+                   short src_pitch,
+                   struct pipe_buffer *src_buffer,
+                   unsigned  src_offset,
+                   boolean src_tiled,
+                   short dst_pitch,
+                   struct pipe_buffer *dst_buffer,
+                   unsigned  dst_offset,
+                   boolean dst_tiled,
+                   short src_x, short src_y,
+                   short dst_x, short dst_y,
+                   short w, short h,
+                   unsigned logic_op)
+{
+   unsigned CMD, BR13;
+   int dst_y2 = dst_y + h;
+   int dst_x2 = dst_x + w;
+   BATCH_LOCALS;
+
+
+   DBG("%s src:buf(%d)/%d %d,%d dst:buf(%d)/%d %d,%d sz:%dx%d op:%d\n",
+       __FUNCTION__,
+       src_buffer, src_pitch, src_x, src_y,
+       dst_buffer, dst_pitch, dst_x, dst_y,
+       w,h,logic_op);
+
+   assert( logic_op - PIPE_LOGICOP_CLEAR >= 0 );
+   assert( logic_op - PIPE_LOGICOP_CLEAR < 0x10 );
+
+   src_pitch *= cpp;
+   dst_pitch *= cpp;
+
+   switch(cpp) {
+   case 1:
+   case 2:
+   case 3:
+      BR13 = (translate_raster_op(logic_op) << 16) | (1<<24);
+      CMD = XY_SRC_COPY_BLT_CMD;
+      break;
+   case 4:
+      BR13 = (translate_raster_op(logic_op) << 16) | (1<<24) |
+	  (1<<25);
+      CMD = XY_SRC_COPY_BLT_CMD | XY_BLT_WRITE_ALPHA | XY_BLT_WRITE_RGB;
+      break;
+   default:
+      return;
+   }
+
+   if (src_tiled) {
+      CMD |= XY_SRC_TILED;
+      src_pitch /= 4;
+   }
+
+   if (dst_tiled) {
+      CMD |= XY_DST_TILED;
+      dst_pitch /= 4;
+   }
+
+   if (dst_y2 < dst_y ||
+       dst_x2 < dst_x) {
+      return;
+   }
+
+   dst_pitch &= 0xffff;
+   src_pitch &= 0xffff;
+
+   /* Initial y values don't seem to work with negative pitches.  If
+    * we adjust the offsets manually (below), it seems to work fine.
+    *
+    * On the other hand, if we always adjust, the hardware doesn't
+    * know which blit directions to use, so overlapping copypixels get
+    * the wrong result.
+    */
+   if (dst_pitch > 0 && src_pitch > 0) {
+      BEGIN_BATCH(8, INTEL_BATCH_NO_CLIPRECTS);
+      OUT_BATCH( CMD );
+      OUT_BATCH( dst_pitch | BR13 );
+      OUT_BATCH( (dst_y << 16) | dst_x );
+      OUT_BATCH( (dst_y2 << 16) | dst_x2 );
+      OUT_RELOC( dst_buffer, BRW_BUFFER_ACCESS_WRITE,
+		 dst_offset );
+      OUT_BATCH( (src_y << 16) | src_x );
+      OUT_BATCH( src_pitch );
+      OUT_RELOC( src_buffer, BRW_BUFFER_ACCESS_READ,
+		 src_offset );
+      ADVANCE_BATCH();
+   }
+   else {
+      BEGIN_BATCH(8, INTEL_BATCH_NO_CLIPRECTS);
+      OUT_BATCH( CMD );
+      OUT_BATCH( (dst_pitch & 0xffff) | BR13 );
+      OUT_BATCH( (0 << 16) | dst_x );
+      OUT_BATCH( (h << 16) | dst_x2 );
+      OUT_RELOC( dst_buffer, BRW_BUFFER_ACCESS_WRITE,
+		 dst_offset + dst_y * dst_pitch );
+      OUT_BATCH( (src_pitch & 0xffff) );
+      OUT_RELOC( src_buffer, BRW_BUFFER_ACCESS_READ,
+		 src_offset + src_y * src_pitch );
+      ADVANCE_BATCH();
+   }
+}
+
+
+
diff --git a/src/gallium/drivers/i965simple/brw_blit.h b/src/gallium/drivers/i965simple/brw_blit.h
new file mode 100644
index 0000000000..111c5d91d3
--- /dev/null
+++ b/src/gallium/drivers/i965simple/brw_blit.h
@@ -0,0 +1,33 @@
+#ifndef BRW_BLIT_H
+#define BRW_BLIT_H
+
+#include "pipe/p_compiler.h"
+
+struct pipe_buffer;
+struct brw_context;
+
+void brw_fill_blit(struct brw_context *intel,
+                   unsigned cpp,
+                   short dst_pitch,
+                   struct pipe_buffer *dst_buffer,
+                   unsigned dst_offset,
+                   boolean dst_tiled,
+                   short x, short y,
+                   short w, short h,
+                   unsigned color);
+void brw_copy_blit(struct brw_context *intel,
+                   unsigned do_flip,
+                   unsigned cpp,
+                   short src_pitch,
+                   struct pipe_buffer *src_buffer,
+                   unsigned  src_offset,
+                   boolean src_tiled,
+                   short dst_pitch,
+                   struct pipe_buffer *dst_buffer,
+                   unsigned  dst_offset,
+                   boolean dst_tiled,
+                   short src_x, short src_y,
+                   short dst_x, short dst_y,
+                   short w, short h,
+                   unsigned logic_op);
+#endif
diff --git a/src/gallium/drivers/i965simple/brw_cc.c b/src/gallium/drivers/i965simple/brw_cc.c
new file mode 100644
index 0000000000..337e4f95f6
--- /dev/null
+++ b/src/gallium/drivers/i965simple/brw_cc.c
@@ -0,0 +1,269 @@
+/*
+ Copyright (C) Intel Corp.  2006.  All Rights Reserved.
+ Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to
+ develop this 3D driver.
+
+ Permission is hereby granted, free of charge, to any person obtaining
+ a copy of this software and associated documentation files (the
+ "Software"), to deal in the Software without restriction, including
+ without limitation the rights to use, copy, modify, merge, publish,
+ distribute, sublicense, and/or sell copies of the Software, and to
+ permit persons to whom the Software is furnished to do so, subject to
+ the following conditions:
+
+ The above copyright notice and this permission notice (including the
+ next paragraph) shall be included in all copies or substantial
+ portions of the Software.
+
+ THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
+ IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
+ LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
+ OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+
+ **********************************************************************/
+ /*
+  * Authors:
+  *   Keith Whitwell <keith@tungstengraphics.com>
+  */
+
+#include "pipe/p_util.h"
+
+#include "brw_context.h"
+#include "brw_state.h"
+#include "brw_defines.h"
+#include "brw_util.h"
+
+
+static int brw_translate_compare_func(int func)
+{
+   switch(func) {
+   case PIPE_FUNC_NEVER:
+      return BRW_COMPAREFUNCTION_NEVER;
+   case PIPE_FUNC_LESS:
+      return BRW_COMPAREFUNCTION_LESS;
+   case PIPE_FUNC_LEQUAL:
+      return BRW_COMPAREFUNCTION_LEQUAL;
+   case PIPE_FUNC_GREATER:
+      return BRW_COMPAREFUNCTION_GREATER;
+   case PIPE_FUNC_GEQUAL:
+      return BRW_COMPAREFUNCTION_GEQUAL;
+   case PIPE_FUNC_NOTEQUAL:
+      return BRW_COMPAREFUNCTION_NOTEQUAL;
+   case PIPE_FUNC_EQUAL:
+      return BRW_COMPAREFUNCTION_EQUAL;
+   case PIPE_FUNC_ALWAYS:
+      return BRW_COMPAREFUNCTION_ALWAYS;
+   }
+
+   debug_printf("Unknown value in %s: %x\n", __FUNCTION__, func);
+   return BRW_COMPAREFUNCTION_ALWAYS;
+}
+
+static int brw_translate_stencil_op(int op)
+{
+   switch(op) {
+   case PIPE_STENCIL_OP_KEEP:
+      return BRW_STENCILOP_KEEP;
+   case PIPE_STENCIL_OP_ZERO:
+      return BRW_STENCILOP_ZERO;
+   case PIPE_STENCIL_OP_REPLACE:
+      return BRW_STENCILOP_REPLACE;
+   case PIPE_STENCIL_OP_INCR:
+      return BRW_STENCILOP_INCRSAT;
+   case PIPE_STENCIL_OP_DECR:
+      return BRW_STENCILOP_DECRSAT;
+   case PIPE_STENCIL_OP_INCR_WRAP:
+      return BRW_STENCILOP_INCR;
+   case PIPE_STENCIL_OP_DECR_WRAP:
+      return BRW_STENCILOP_DECR;
+   case PIPE_STENCIL_OP_INVERT:
+      return BRW_STENCILOP_INVERT;
+   default:
+      return BRW_STENCILOP_ZERO;
+   }
+}
+
+
+static int brw_translate_logic_op(int opcode)
+{
+   switch(opcode) {
+   case PIPE_LOGICOP_CLEAR:
+      return BRW_LOGICOPFUNCTION_CLEAR;
+   case PIPE_LOGICOP_AND:
+      return BRW_LOGICOPFUNCTION_AND;
+   case PIPE_LOGICOP_AND_REVERSE:
+      return BRW_LOGICOPFUNCTION_AND_REVERSE;
+   case PIPE_LOGICOP_COPY:
+      return BRW_LOGICOPFUNCTION_COPY;
+   case PIPE_LOGICOP_COPY_INVERTED:
+      return BRW_LOGICOPFUNCTION_COPY_INVERTED;
+   case PIPE_LOGICOP_AND_INVERTED:
+      return BRW_LOGICOPFUNCTION_AND_INVERTED;
+   case PIPE_LOGICOP_NOOP:
+      return BRW_LOGICOPFUNCTION_NOOP;
+   case PIPE_LOGICOP_XOR:
+      return BRW_LOGICOPFUNCTION_XOR;
+   case PIPE_LOGICOP_OR:
+      return BRW_LOGICOPFUNCTION_OR;
+   case PIPE_LOGICOP_OR_INVERTED:
+      return BRW_LOGICOPFUNCTION_OR_INVERTED;
+   case PIPE_LOGICOP_NOR:
+      return BRW_LOGICOPFUNCTION_NOR;
+   case PIPE_LOGICOP_EQUIV:
+      return BRW_LOGICOPFUNCTION_EQUIV;
+   case PIPE_LOGICOP_INVERT:
+      return BRW_LOGICOPFUNCTION_INVERT;
+   case PIPE_LOGICOP_OR_REVERSE:
+      return BRW_LOGICOPFUNCTION_OR_REVERSE;
+   case PIPE_LOGICOP_NAND:
+      return BRW_LOGICOPFUNCTION_NAND;
+   case PIPE_LOGICOP_SET:
+      return BRW_LOGICOPFUNCTION_SET;
+   default:
+      return BRW_LOGICOPFUNCTION_SET;
+   }
+}
+
+
+static void upload_cc_vp( struct brw_context *brw )
+{
+   struct brw_cc_viewport ccv;
+
+   memset(&ccv, 0, sizeof(ccv));
+
+   ccv.min_depth = 0.0;
+   ccv.max_depth = 1.0;
+
+   brw->cc.vp_gs_offset = brw_cache_data( &brw->cache[BRW_CC_VP], &ccv );
+}
+
+const struct brw_tracked_state brw_cc_vp = {
+   .dirty = {
+      .brw = BRW_NEW_SCENE,
+      .cache = 0
+   },
+   .update = upload_cc_vp
+};
+
+
+static void upload_cc_unit( struct brw_context *brw )
+{
+   struct brw_cc_unit_state cc;
+
+   memset(&cc, 0, sizeof(cc));
+
+   /* BRW_NEW_DEPTH_STENCIL */
+   if (brw->attribs.DepthStencil->stencil[0].enabled) {
+      cc.cc0.stencil_enable = brw->attribs.DepthStencil->stencil[0].enabled;
+      cc.cc0.stencil_func = brw_translate_compare_func(brw->attribs.DepthStencil->stencil[0].func);
+      cc.cc0.stencil_fail_op = brw_translate_stencil_op(brw->attribs.DepthStencil->stencil[0].fail_op);
+      cc.cc0.stencil_pass_depth_fail_op = brw_translate_stencil_op(
+         brw->attribs.DepthStencil->stencil[0].zfail_op);
+      cc.cc0.stencil_pass_depth_pass_op = brw_translate_stencil_op(
+         brw->attribs.DepthStencil->stencil[0].zpass_op);
+      cc.cc1.stencil_ref = brw->attribs.DepthStencil->stencil[0].ref_value;
+      cc.cc1.stencil_write_mask = brw->attribs.DepthStencil->stencil[0].write_mask;
+      cc.cc1.stencil_test_mask = brw->attribs.DepthStencil->stencil[0].value_mask;
+
+      if (brw->attribs.DepthStencil->stencil[1].enabled) {
+	 cc.cc0.bf_stencil_enable = brw->attribs.DepthStencil->stencil[1].enabled;
+	 cc.cc0.bf_stencil_func = brw_translate_compare_func(
+            brw->attribs.DepthStencil->stencil[1].func);
+	 cc.cc0.bf_stencil_fail_op = brw_translate_stencil_op(
+            brw->attribs.DepthStencil->stencil[1].fail_op);
+	 cc.cc0.bf_stencil_pass_depth_fail_op = brw_translate_stencil_op(
+            brw->attribs.DepthStencil->stencil[1].zfail_op);
+	 cc.cc0.bf_stencil_pass_depth_pass_op = brw_translate_stencil_op(
+            brw->attribs.DepthStencil->stencil[1].zpass_op);
+	 cc.cc1.bf_stencil_ref = brw->attribs.DepthStencil->stencil[1].ref_value;
+	 cc.cc2.bf_stencil_write_mask = brw->attribs.DepthStencil->stencil[1].write_mask;
+	 cc.cc2.bf_stencil_test_mask = brw->attribs.DepthStencil->stencil[1].value_mask;
+      }
+
+      /* Not really sure about this:
+       */
+      if (brw->attribs.DepthStencil->stencil[0].write_mask ||
+	  brw->attribs.DepthStencil->stencil[1].write_mask)
+	 cc.cc0.stencil_write_enable = 1;
+   }
+
+   /* BRW_NEW_BLEND */
+   if (brw->attribs.Blend->logicop_enable) {
+      cc.cc2.logicop_enable = 1;
+      cc.cc5.logicop_func = brw_translate_logic_op( brw->attribs.Blend->logicop_func );
+   }
+   else if (brw->attribs.Blend->blend_enable) {
+      int eqRGB = brw->attribs.Blend->rgb_func;
+      int eqA = brw->attribs.Blend->alpha_func;
+      int srcRGB = brw->attribs.Blend->rgb_src_factor;
+      int dstRGB = brw->attribs.Blend->rgb_dst_factor;
+      int srcA = brw->attribs.Blend->alpha_src_factor;
+      int dstA = brw->attribs.Blend->alpha_dst_factor;
+
+      if (eqRGB == PIPE_BLEND_MIN || eqRGB == PIPE_BLEND_MAX) {
+	 srcRGB = dstRGB = PIPE_BLENDFACTOR_ONE;
+      }
+
+      if (eqA == PIPE_BLEND_MIN || eqA == PIPE_BLEND_MAX) {
+	 srcA = dstA = PIPE_BLENDFACTOR_ONE;
+      }
+
+      cc.cc6.dest_blend_factor = brw_translate_blend_factor(dstRGB);
+      cc.cc6.src_blend_factor = brw_translate_blend_factor(srcRGB);
+      cc.cc6.blend_function = brw_translate_blend_equation( eqRGB );
+
+      cc.cc5.ia_dest_blend_factor = brw_translate_blend_factor(dstA);
+      cc.cc5.ia_src_blend_factor = brw_translate_blend_factor(srcA);
+      cc.cc5.ia_blend_function = brw_translate_blend_equation( eqA );
+
+      cc.cc3.blend_enable = 1;
+      cc.cc3.ia_blend_enable = (srcA != srcRGB ||
+				dstA != dstRGB ||
+				eqA != eqRGB);
+   }
+   
+   /* BRW_NEW_ALPHATEST
+    */
+   if (brw->attribs.DepthStencil->alpha.enabled) {
+      cc.cc3.alpha_test = 1;
+      cc.cc3.alpha_test_func = 
+	 brw_translate_compare_func(brw->attribs.DepthStencil->alpha.func);
+
+      UNCLAMPED_FLOAT_TO_UBYTE(cc.cc7.alpha_ref.ub[0], 
+			       brw->attribs.DepthStencil->alpha.ref);
+
+      cc.cc3.alpha_test_format = BRW_ALPHATEST_FORMAT_UNORM8;
+   }
+
+   if (brw->attribs.Blend->dither) {
+      cc.cc5.dither_enable = 1;
+      cc.cc6.y_dither_offset = 0;
+      cc.cc6.x_dither_offset = 0;
+   }
+
+   if (brw->attribs.DepthStencil->depth.enabled) {
+      cc.cc2.depth_test = brw->attribs.DepthStencil->depth.enabled;
+      cc.cc2.depth_test_function = brw_translate_compare_func(brw->attribs.DepthStencil->depth.func);
+      cc.cc2.depth_write_enable = brw->attribs.DepthStencil->depth.writemask;
+   }
+
+   /* CACHE_NEW_CC_VP */
+   cc.cc4.cc_viewport_state_offset =  brw->cc.vp_gs_offset >> 5;
+
+   if (BRW_DEBUG & DEBUG_STATS)
+      cc.cc5.statistics_enable = 1;
+
+   brw->cc.state_gs_offset = brw_cache_data( &brw->cache[BRW_CC_UNIT], &cc );
+}
+
+const struct brw_tracked_state brw_cc_unit = {
+   .dirty = {
+      .brw = BRW_NEW_DEPTH_STENCIL | BRW_NEW_BLEND | BRW_NEW_ALPHA_TEST,
+      .cache = CACHE_NEW_CC_VP
+   },
+   .update = upload_cc_unit
+};
+
diff --git a/src/gallium/drivers/i965simple/brw_clip.c b/src/gallium/drivers/i965simple/brw_clip.c
new file mode 100644
index 0000000000..268124cc53
--- /dev/null
+++ b/src/gallium/drivers/i965simple/brw_clip.c
@@ -0,0 +1,206 @@
+/*
+ Copyright (C) Intel Corp.  2006.  All Rights Reserved.
+ Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to
+ develop this 3D driver.
+
+ Permission is hereby granted, free of charge, to any person obtaining
+ a copy of this software and associated documentation files (the
+ "Software"), to deal in the Software without restriction, including
+ without limitation the rights to use, copy, modify, merge, publish,
+ distribute, sublicense, and/or sell copies of the Software, and to
+ permit persons to whom the Software is furnished to do so, subject to
+ the following conditions:
+
+ The above copyright notice and this permission notice (including the
+ next paragraph) shall be included in all copies or substantial
+ portions of the Software.
+
+ THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
+ IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
+ LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
+ OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+
+ **********************************************************************/
+ /*
+  * Authors:
+  *   Keith Whitwell <keith@tungstengraphics.com>
+  */
+
+#include "brw_defines.h"
+#include "brw_context.h"
+#include "brw_eu.h"
+#include "brw_util.h"
+#include "brw_state.h"
+#include "brw_clip.h"
+
+#define FRONT_UNFILLED_BIT  0x1
+#define BACK_UNFILLED_BIT   0x2
+
+
+static void compile_clip_prog( struct brw_context *brw,
+			     struct brw_clip_prog_key *key )
+{
+   struct brw_clip_compile c;
+   const unsigned *program;
+   unsigned program_size;
+   unsigned delta;
+   unsigned i;
+
+   memset(&c, 0, sizeof(c));
+
+   /* Begin the compilation:
+    */
+   brw_init_compile(&c.func);
+
+   c.func.single_program_flow = 1;
+
+   c.key = *key;
+
+
+   /* Need to locate the two positions present in vertex + header.
+    * These are currently hardcoded:
+    */
+   c.header_position_offset = ATTR_SIZE;
+
+   for (i = 0, delta = REG_SIZE; i < PIPE_MAX_SHADER_OUTPUTS; i++)
+      if (c.key.attrs & (1<<i)) {
+	 c.offset[i] = delta;
+	 delta += ATTR_SIZE;
+      }
+
+   c.nr_attrs = brw_count_bits(c.key.attrs);
+   c.nr_regs = (c.nr_attrs + 1) / 2 + 1;  /* are vertices packed, or reg-aligned? */
+   c.nr_bytes = c.nr_regs * REG_SIZE;
+
+   c.prog_data.clip_mode = c.key.clip_mode; /* XXX */
+
+   /* For some reason the thread is spawned with only 4 channels
+    * unmasked.
+    */
+   brw_set_mask_control(&c.func, BRW_MASK_DISABLE);
+
+
+   /* Would ideally have the option of producing a program which could
+    * do all three:
+    */
+   switch (key->primitive) {
+   case PIPE_PRIM_TRIANGLES:
+#if 0
+      if (key->do_unfilled)
+	 brw_emit_unfilled_clip( &c );
+      else
+#endif
+	 brw_emit_tri_clip( &c );
+      break;
+   case PIPE_PRIM_LINES:
+      brw_emit_line_clip( &c );
+      break;
+   case PIPE_PRIM_POINTS:
+      brw_emit_point_clip( &c );
+      break;
+   default:
+      assert(0);
+      return;
+   }
+
+
+
+   /* get the program
+    */
+   program = brw_get_program(&c.func, &program_size);
+
+   /* Upload
+    */
+   brw->clip.prog_gs_offset = brw_upload_cache( &brw->cache[BRW_CLIP_PROG],
+						&c.key,
+						sizeof(c.key),
+						program,
+						program_size,
+						&c.prog_data,
+						&brw->clip.prog_data );
+}
+
+
+static boolean search_cache( struct brw_context *brw,
+			       struct brw_clip_prog_key *key )
+{
+   return brw_search_cache(&brw->cache[BRW_CLIP_PROG],
+			   key, sizeof(*key),
+			   &brw->clip.prog_data,
+			   &brw->clip.prog_gs_offset);
+}
+
+
+
+
+/* Calculate interpolants for triangle and line rasterization.
+ */
+static void upload_clip_prog(struct brw_context *brw)
+{
+   struct brw_clip_prog_key key;
+
+   memset(&key, 0, sizeof(key));
+
+   /* Populate the key:
+    */
+   /* BRW_NEW_REDUCED_PRIMITIVE */
+   key.primitive = brw->reduced_primitive;
+   /* CACHE_NEW_VS_PROG */
+   key.attrs = brw->vs.prog_data->outputs_written;
+   /* BRW_NEW_RASTER */
+   key.do_flat_shading = (brw->attribs.Raster->flatshade);
+   /* BRW_NEW_CLIP */
+   key.nr_userclip = brw->attribs.Clip.nr; /* XXX */
+
+#if 0
+   key.clip_mode = BRW_CLIPMODE_NORMAL;
+
+   if (key.primitive == PIPE_PRIM_TRIANGLES) {
+      if (brw->attribs.Raster->cull_mode == PIPE_WINDING_BOTH)
+	 key.clip_mode = BRW_CLIPMODE_REJECT_ALL;
+      else {
+         if (brw->attribs.Raster->fill_cw != PIPE_POLYGON_MODE_FILL ||
+             brw->attribs.Raster->fill_ccw != PIPE_POLYGON_MODE_FILL)
+            key.do_unfilled = 1;
+
+	 /* Most cases the fixed function units will handle.  Cases where
+	  * one or more polygon faces are unfilled will require help:
+	  */
+	 if (key.do_unfilled) {
+	    key.clip_mode = BRW_CLIPMODE_CLIP_NON_REJECTED;
+
+	    if (brw->attribs.Raster->offset_cw ||
+                brw->attribs.Raster->offset_ccw) {
+	       key.offset_units = brw->attribs.Raster->offset_units;
+	       key.offset_factor = brw->attribs.Raster->offset_scale;
+	    }
+            key.fill_ccw = brw->attribs.Raster->fill_ccw;
+            key.fill_cw = brw->attribs.Raster->fill_cw;
+            key.offset_ccw = brw->attribs.Raster->offset_ccw;
+            key.offset_cw = brw->attribs.Raster->offset_cw;
+            if (brw->attribs.Raster->light_twoside &&
+                key.fill_cw != CLIP_CULL)
+               key.copy_bfc_cw = 1;
+	 }
+      }
+   }
+#else
+   key.clip_mode = BRW_CLIPMODE_ACCEPT_ALL;
+#endif
+
+   if (!search_cache(brw, &key))
+      compile_clip_prog( brw, &key );
+}
+
+const struct brw_tracked_state brw_clip_prog = {
+   .dirty = {
+      .brw   = (BRW_NEW_RASTERIZER |
+		BRW_NEW_CLIP |
+		BRW_NEW_REDUCED_PRIMITIVE),
+      .cache = CACHE_NEW_VS_PROG
+   },
+   .update = upload_clip_prog
+};
diff --git a/src/gallium/drivers/i965simple/brw_clip.h b/src/gallium/drivers/i965simple/brw_clip.h
new file mode 100644
index 0000000000..a89d08b791
--- /dev/null
+++ b/src/gallium/drivers/i965simple/brw_clip.h
@@ -0,0 +1,170 @@
+/*
+ Copyright (C) Intel Corp.  2006.  All Rights Reserved.
+ Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to
+ develop this 3D driver.
+ 
+ Permission is hereby granted, free of charge, to any person obtaining
+ a copy of this software and associated documentation files (the
+ "Software"), to deal in the Software without restriction, including
+ without limitation the rights to use, copy, modify, merge, publish,
+ distribute, sublicense, and/or sell copies of the Software, and to
+ permit persons to whom the Software is furnished to do so, subject to
+ the following conditions:
+ 
+ The above copyright notice and this permission notice (including the
+ next paragraph) shall be included in all copies or substantial
+ portions of the Software.
+ 
+ THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
+ IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
+ LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
+ OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ 
+ **********************************************************************/
+ /*
+  * Authors:
+  *   Keith Whitwell <keith@tungstengraphics.com>
+  */
+
+#ifndef BRW_CLIP_H
+#define BRW_CLIP_H
+
+
+#include "brw_context.h"
+#include "brw_eu.h"
+
+#define MAX_VERTS (3+6+6)	
+
+/* Note that if unfilled primitives are being emitted, we have to fix
+ * up polygon offset and flatshading at this point:
+ */
+struct brw_clip_prog_key {
+   unsigned attrs:32;		
+   unsigned primitive:4;
+   unsigned nr_userclip:3;
+   unsigned do_flat_shading:1;
+   unsigned do_unfilled:1;
+   unsigned fill_cw:2;		/* includes cull information */
+   unsigned fill_ccw:2;		/* includes cull information */
+   unsigned offset_cw:1;
+   unsigned offset_ccw:1;
+   unsigned pad0:17;
+
+   unsigned copy_bfc_cw:1;
+   unsigned copy_bfc_ccw:1;
+   unsigned clip_mode:3;
+   unsigned pad1:27;
+   
+   float offset_factor;
+   float offset_units;
+};
+
+
+#define CLIP_LINE   0
+#define CLIP_POINT  1
+#define CLIP_FILL   2
+#define CLIP_CULL   3
+
+
+#define PRIM_MASK  (0x1f)
+
+struct brw_clip_compile {
+   struct brw_compile func;
+   struct brw_clip_prog_key key;
+   struct brw_clip_prog_data prog_data;
+   
+   struct {
+      struct brw_reg R0;
+      struct brw_reg vertex[MAX_VERTS];
+
+      struct brw_reg t;
+      struct brw_reg t0, t1;
+      struct brw_reg dp0, dp1;
+
+      struct brw_reg dpPrev;
+      struct brw_reg dp;
+      struct brw_reg loopcount;
+      struct brw_reg nr_verts;
+      struct brw_reg planemask;
+
+      struct brw_reg inlist;
+      struct brw_reg outlist;
+      struct brw_reg freelist;
+
+      struct brw_reg dir;
+      struct brw_reg tmp0, tmp1;
+      struct brw_reg offset;
+      
+      struct brw_reg fixed_planes;
+      struct brw_reg plane_equation;
+   } reg;
+
+   /* 3 different ways of expressing vertex size:
+    */
+   unsigned nr_attrs;
+   unsigned nr_regs;
+   unsigned nr_bytes;
+
+   unsigned first_tmp;
+   unsigned last_tmp;
+
+   boolean need_direction;
+
+   unsigned last_mrf;
+
+   unsigned header_position_offset;
+   unsigned offset[PIPE_ATTRIB_MAX];
+};
+
+#define ATTR_SIZE  (4*4)
+
+/* Points are only culled, so no need for a clip routine, however it
+ * works out easier to have a dummy one.
+ */
+void brw_emit_unfilled_clip( struct brw_clip_compile *c );
+void brw_emit_tri_clip( struct brw_clip_compile *c );
+void brw_emit_line_clip( struct brw_clip_compile *c );
+void brw_emit_point_clip( struct brw_clip_compile *c );
+
+/* brw_clip_tri.c, for use by the unfilled clip routine:
+ */
+void brw_clip_tri_init_vertices( struct brw_clip_compile *c );
+void brw_clip_tri_flat_shade( struct brw_clip_compile *c );
+void brw_clip_tri( struct brw_clip_compile *c );
+void brw_clip_tri_emit_polygon( struct brw_clip_compile *c );
+void brw_clip_tri_alloc_regs( struct brw_clip_compile *c, 
+			      unsigned nr_verts );
+
+
+/* Utils:
+ */
+
+void brw_clip_interp_vertex( struct brw_clip_compile *c,
+			     struct brw_indirect dest_ptr,
+			     struct brw_indirect v0_ptr, /* from */
+			     struct brw_indirect v1_ptr, /* to */
+			     struct brw_reg t0,
+			     boolean force_edgeflag );
+
+void brw_clip_init_planes( struct brw_clip_compile *c );
+
+void brw_clip_emit_vue(struct brw_clip_compile *c, 
+		       struct brw_indirect vert,
+		       boolean allocate,
+		       boolean eot,
+		       unsigned header);
+
+void brw_clip_kill_thread(struct brw_clip_compile *c);
+
+struct brw_reg brw_clip_plane_stride( struct brw_clip_compile *c );
+struct brw_reg brw_clip_plane0_address( struct brw_clip_compile *c );
+
+void brw_clip_copy_colors( struct brw_clip_compile *c,
+			   unsigned to, unsigned from );
+
+void brw_clip_init_clipmask( struct brw_clip_compile *c );
+
+#endif
diff --git a/src/gallium/drivers/i965simple/brw_clip_line.c b/src/gallium/drivers/i965simple/brw_clip_line.c
new file mode 100644
index 0000000000..75d9e5fcda
--- /dev/null
+++ b/src/gallium/drivers/i965simple/brw_clip_line.c
@@ -0,0 +1,245 @@
+/*
+ Copyright (C) Intel Corp.  2006.  All Rights Reserved.
+ Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to
+ develop this 3D driver.
+
+ Permission is hereby granted, free of charge, to any person obtaining
+ a copy of this software and associated documentation files (the
+ "Software"), to deal in the Software without restriction, including
+ without limitation the rights to use, copy, modify, merge, publish,
+ distribute, sublicense, and/or sell copies of the Software, and to
+ permit persons to whom the Software is furnished to do so, subject to
+ the following conditions:
+
+ The above copyright notice and this permission notice (including the
+ next paragraph) shall be included in all copies or substantial
+ portions of the Software.
+
+ THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
+ IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
+ LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
+ OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+
+ **********************************************************************/
+ /*
+  * Authors:
+  *   Keith Whitwell <keith@tungstengraphics.com>
+  */
+
+#include "brw_defines.h"
+#include "brw_context.h"
+#include "brw_eu.h"
+#include "brw_util.h"
+#include "brw_clip.h"
+
+
+
+static void brw_clip_line_alloc_regs( struct brw_clip_compile *c )
+{
+   unsigned i = 0,j;
+
+   /* Register usage is static, precompute here:
+    */
+   c->reg.R0 = retype(brw_vec8_grf(i, 0), BRW_REGISTER_TYPE_UD); i++;
+
+   if (c->key.nr_userclip) {
+      c->reg.fixed_planes = brw_vec4_grf(i, 0);
+      i += (6 + c->key.nr_userclip + 1) / 2;
+
+      c->prog_data.curb_read_length = (6 + c->key.nr_userclip + 1) / 2;
+   }
+   else
+      c->prog_data.curb_read_length = 0;
+
+
+   /* Payload vertices plus space for more generated vertices:
+    */
+   for (j = 0; j < 4; j++) {
+      c->reg.vertex[j] = brw_vec4_grf(i, 0);
+      i += c->nr_regs;
+   }
+
+   c->reg.t           = brw_vec1_grf(i, 0);
+   c->reg.t0          = brw_vec1_grf(i, 1);
+   c->reg.t1          = brw_vec1_grf(i, 2);
+   c->reg.planemask   = retype(brw_vec1_grf(i, 3), BRW_REGISTER_TYPE_UD);
+   c->reg.plane_equation = brw_vec4_grf(i, 4);
+   i++;
+
+   c->reg.dp0         = brw_vec1_grf(i, 0); /* fixme - dp4 will clobber r.1,2,3 */
+   c->reg.dp1         = brw_vec1_grf(i, 4);
+   i++;
+
+   if (!c->key.nr_userclip) {
+      c->reg.fixed_planes = brw_vec8_grf(i, 0);
+      i++;
+   }
+
+
+   c->first_tmp = i;
+   c->last_tmp = i;
+
+   c->prog_data.urb_read_length = c->nr_regs; /* ? */
+   c->prog_data.total_grf = i;
+}
+
+
+
+/* Line clipping, more or less following the following algorithm:
+ *
+ *  for (p=0;p<MAX_PLANES;p++) {
+ *     if (clipmask & (1 << p)) {
+ *        float dp0 = DOTPROD( vtx0, plane[p] );
+ *        float dp1 = DOTPROD( vtx1, plane[p] );
+ *
+ *        if (IS_NEGATIVE(dp1)) {
+ *           float t = dp1 / (dp1 - dp0);
+ *           if (t > t1) t1 = t;
+ *        } else {
+ *           float t = dp0 / (dp0 - dp1);
+ *           if (t > t0) t0 = t;
+ *        }
+ *
+ *        if (t0 + t1 >= 1.0)
+ *           return;
+ *     }
+ *  }
+ *
+ *  interp( ctx, newvtx0, vtx0, vtx1, t0 );
+ *  interp( ctx, newvtx1, vtx1, vtx0, t1 );
+ *
+ */
+static void clip_and_emit_line( struct brw_clip_compile *c )
+{
+   struct brw_compile *p = &c->func;
+   struct brw_indirect vtx0     = brw_indirect(0, 0);
+   struct brw_indirect vtx1      = brw_indirect(1, 0);
+   struct brw_indirect newvtx0   = brw_indirect(2, 0);
+   struct brw_indirect newvtx1   = brw_indirect(3, 0);
+   struct brw_indirect plane_ptr = brw_indirect(4, 0);
+   struct brw_instruction *plane_loop;
+   struct brw_instruction *plane_active;
+   struct brw_instruction *is_negative;
+   struct brw_instruction *is_neg2;
+   struct brw_instruction *not_culled;
+   struct brw_reg v1_null_ud = retype(vec1(brw_null_reg()), BRW_REGISTER_TYPE_UD);
+
+   brw_MOV(p, get_addr_reg(vtx0),      brw_address(c->reg.vertex[0]));
+   brw_MOV(p, get_addr_reg(vtx1),      brw_address(c->reg.vertex[1]));
+   brw_MOV(p, get_addr_reg(newvtx0),   brw_address(c->reg.vertex[2]));
+   brw_MOV(p, get_addr_reg(newvtx1),   brw_address(c->reg.vertex[3]));
+   brw_MOV(p, get_addr_reg(plane_ptr), brw_clip_plane0_address(c));
+
+   /* Note: init t0, t1 together:
+    */
+   brw_MOV(p, vec2(c->reg.t0), brw_imm_f(0));
+
+   brw_clip_init_planes(c);
+   brw_clip_init_clipmask(c);
+
+   /* -ve rhw workaround */
+   brw_set_conditionalmod(p, BRW_CONDITIONAL_NZ);
+   brw_AND(p, brw_null_reg(), get_element_ud(c->reg.R0, 2),
+	   brw_imm_ud(1<<20));
+   brw_OR(p, c->reg.planemask, c->reg.planemask, brw_imm_ud(0x3f));
+   brw_set_predicate_control(p, BRW_PREDICATE_NONE);
+
+   plane_loop = brw_DO(p, BRW_EXECUTE_1);
+   {
+      /* if (planemask & 1)
+       */
+      brw_set_conditionalmod(p, BRW_CONDITIONAL_NZ);
+      brw_AND(p, v1_null_ud, c->reg.planemask, brw_imm_ud(1));
+
+      plane_active = brw_IF(p, BRW_EXECUTE_1);
+      {
+	 if (c->key.nr_userclip)
+	    brw_MOV(p, c->reg.plane_equation, deref_4f(plane_ptr, 0));
+	 else
+	    brw_MOV(p, c->reg.plane_equation, deref_4b(plane_ptr, 0));
+
+#if 0
+	 /* dp = DP4(vtx->position, plane)
+	  */
+	 brw_DP4(p, vec4(c->reg.dp0), deref_4f(vtx0, c->offset[VERT_RESULT_HPOS]), c->reg.plane_equation);
+
+	 /* if (IS_NEGATIVE(dp1))
+	  */
+	 brw_set_conditionalmod(p, BRW_CONDITIONAL_L);
+	 brw_DP4(p, vec4(c->reg.dp1), deref_4f(vtx1, c->offset[VERT_RESULT_HPOS]), c->reg.plane_equation);
+#else
+         #warning "disabled"
+#endif
+	 is_negative = brw_IF(p, BRW_EXECUTE_1);
+	 {
+	    brw_ADD(p, c->reg.t, c->reg.dp1, negate(c->reg.dp0));
+	    brw_math_invert(p, c->reg.t, c->reg.t);
+	    brw_MUL(p, c->reg.t, c->reg.t, c->reg.dp1);
+
+	    brw_CMP(p, vec1(brw_null_reg()), BRW_CONDITIONAL_G, c->reg.t, c->reg.t1 );
+	    brw_MOV(p, c->reg.t1, c->reg.t);
+	    brw_set_predicate_control(p, BRW_PREDICATE_NONE);
+	 }
+	 is_negative = brw_ELSE(p, is_negative);
+	 {
+	    /* Coming back in.  We know that both cannot be negative
+	     * because the line would have been culled in that case.
+	     */
+
+	    /* If both are positive, do nothing */
+             brw_CMP(p, vec1(brw_null_reg()), BRW_CONDITIONAL_L, c->reg.dp0, brw_imm_f(0.0));
+             is_neg2 = brw_IF(p, BRW_EXECUTE_1);
+             {
+		brw_ADD(p, c->reg.t, c->reg.dp0, negate(c->reg.dp1));
+		brw_math_invert(p, c->reg.t, c->reg.t);
+		brw_MUL(p, c->reg.t, c->reg.t, c->reg.dp0);
+
+		brw_CMP(p, vec1(brw_null_reg()), BRW_CONDITIONAL_G, c->reg.t, c->reg.t0 );
+		brw_MOV(p, c->reg.t0, c->reg.t);
+		brw_set_predicate_control(p, BRW_PREDICATE_NONE);
+	     }
+	     brw_ENDIF(p, is_neg2);
+	 }
+	 brw_ENDIF(p, is_negative);
+      }
+      brw_ENDIF(p, plane_active);
+
+      /* plane_ptr++;
+       */
+      brw_ADD(p, get_addr_reg(plane_ptr), get_addr_reg(plane_ptr), brw_clip_plane_stride(c));
+
+      /* while (planemask>>=1) != 0
+       */
+      brw_set_conditionalmod(p, BRW_CONDITIONAL_NZ);
+      brw_SHR(p, c->reg.planemask, c->reg.planemask, brw_imm_ud(1));
+   }
+   brw_WHILE(p, plane_loop);
+
+   brw_ADD(p, c->reg.t, c->reg.t0, c->reg.t1);
+   brw_CMP(p, vec1(brw_null_reg()), BRW_CONDITIONAL_L, c->reg.t, brw_imm_f(1.0));
+   not_culled = brw_IF(p, BRW_EXECUTE_1);
+   {
+      brw_clip_interp_vertex(c, newvtx0, vtx0, vtx1, c->reg.t0, FALSE);
+      brw_clip_interp_vertex(c, newvtx1, vtx1, vtx0, c->reg.t1, FALSE);
+
+      brw_clip_emit_vue(c, newvtx0, 1, 0, (_3DPRIM_LINESTRIP << 2) | R02_PRIM_START);
+      brw_clip_emit_vue(c, newvtx1, 0, 1, (_3DPRIM_LINESTRIP << 2) | R02_PRIM_END);
+   }
+   brw_ENDIF(p, not_culled);
+   brw_clip_kill_thread(c);
+}
+
+
+
+void brw_emit_line_clip( struct brw_clip_compile *c )
+{
+   brw_clip_line_alloc_regs(c);
+
+   if (c->key.do_flat_shading)
+      brw_clip_copy_colors(c, 0, 1);
+
+   clip_and_emit_line(c);
+}
diff --git a/src/gallium/drivers/i965simple/brw_clip_point.c b/src/gallium/drivers/i965simple/brw_clip_point.c
new file mode 100644
index 0000000000..6fce7210d1
--- /dev/null
+++ b/src/gallium/drivers/i965simple/brw_clip_point.c
@@ -0,0 +1,47 @@
+/*
+ Copyright (C) Intel Corp.  2006.  All Rights Reserved.
+ Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to
+ develop this 3D driver.
+ 
+ Permission is hereby granted, free of charge, to any person obtaining
+ a copy of this software and associated documentation files (the
+ "Software"), to deal in the Software without restriction, including
+ without limitation the rights to use, copy, modify, merge, publish,
+ distribute, sublicense, and/or sell copies of the Software, and to
+ permit persons to whom the Software is furnished to do so, subject to
+ the following conditions:
+ 
+ The above copyright notice and this permission notice (including the
+ next paragraph) shall be included in all copies or substantial
+ portions of the Software.
+ 
+ THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
+ IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
+ LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
+ OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ 
+ **********************************************************************/
+ /*
+  * Authors:
+  *   Keith Whitwell <keith@tungstengraphics.com>
+  */
+
+#include "brw_defines.h"
+#include "brw_context.h"
+#include "brw_eu.h"
+#include "brw_util.h"
+#include "brw_clip.h"
+
+
+/* Point clipping, nothing to do?
+ */
+void brw_emit_point_clip( struct brw_clip_compile *c )
+{
+   /* Send an empty message to kill the thread:
+    */
+   brw_clip_tri_alloc_regs(c, 0);
+   brw_clip_kill_thread(c);
+}
diff --git a/src/gallium/drivers/i965simple/brw_clip_state.c b/src/gallium/drivers/i965simple/brw_clip_state.c
new file mode 100644
index 0000000000..ea5c05a279
--- /dev/null
+++ b/src/gallium/drivers/i965simple/brw_clip_state.c
@@ -0,0 +1,92 @@
+/*
+ Copyright (C) Intel Corp.  2006.  All Rights Reserved.
+ Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to
+ develop this 3D driver.
+ 
+ Permission is hereby granted, free of charge, to any person obtaining
+ a copy of this software and associated documentation files (the
+ "Software"), to deal in the Software without restriction, including
+ without limitation the rights to use, copy, modify, merge, publish,
+ distribute, sublicense, and/or sell copies of the Software, and to
+ permit persons to whom the Software is furnished to do so, subject to
+ the following conditions:
+ 
+ The above copyright notice and this permission notice (including the
+ next paragraph) shall be included in all copies or substantial
+ portions of the Software.
+ 
+ THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
+ IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
+ LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
+ OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ 
+ **********************************************************************/
+ /*
+  * Authors:
+  *   Keith Whitwell <keith@tungstengraphics.com>
+  */
+
+#include "brw_context.h"
+#include "brw_state.h"
+#include "brw_defines.h"
+#include "pipe/p_util.h"
+
+
+static void upload_clip_unit( struct brw_context *brw )
+{
+   struct brw_clip_unit_state clip;
+
+   memset(&clip, 0, sizeof(clip));
+
+   /* CACHE_NEW_CLIP_PROG */
+   clip.thread0.grf_reg_count =
+      align(brw->clip.prog_data->total_grf, 16) / 16 - 1;
+   clip.thread0.kernel_start_pointer = brw->clip.prog_gs_offset >> 6;
+   clip.thread3.urb_entry_read_length = brw->clip.prog_data->urb_read_length;
+   clip.thread3.const_urb_entry_read_length = brw->clip.prog_data->curb_read_length;
+   clip.clip5.clip_mode = brw->clip.prog_data->clip_mode;
+
+   /* BRW_NEW_CURBE_OFFSETS */
+   clip.thread3.const_urb_entry_read_offset = brw->curbe.clip_start * 2;
+
+   /* BRW_NEW_URB_FENCE */
+   clip.thread4.nr_urb_entries = brw->urb.nr_clip_entries; 
+   clip.thread4.urb_entry_allocation_size = brw->urb.vsize - 1;
+   clip.thread4.max_threads = 1; /* 2 threads */
+
+   if (BRW_DEBUG & DEBUG_STATS)
+      clip.thread4.stats_enable = 1; 
+
+   /* CONSTANT */
+   clip.thread1.floating_point_mode = BRW_FLOATING_POINT_NON_IEEE_754;
+   clip.thread1.single_program_flow = 1;
+   clip.thread3.dispatch_grf_start_reg = 1;
+   clip.thread3.urb_entry_read_offset = 0;
+   clip.clip5.userclip_enable_flags = 0x7f;
+   clip.clip5.userclip_must_clip = 1;
+   clip.clip5.guard_band_enable = 0;
+   clip.clip5.viewport_z_clip_enable = 1;
+   clip.clip5.viewport_xy_clip_enable = 1;
+   clip.clip5.vertex_position_space = BRW_CLIP_NDCSPACE;
+   clip.clip5.api_mode = BRW_CLIP_API_OGL;   
+   clip.clip6.clipper_viewport_state_ptr = 0;
+   clip.viewport_xmin = -1;
+   clip.viewport_xmax = 1;
+   clip.viewport_ymin = -1;
+   clip.viewport_ymax = 1;
+
+   brw->clip.state_gs_offset = brw_cache_data( &brw->cache[BRW_CLIP_UNIT], &clip );
+}
+
+
+const struct brw_tracked_state brw_clip_unit = {
+   .dirty = {
+      .brw   = (BRW_NEW_CURBE_OFFSETS |
+		BRW_NEW_URB_FENCE),
+      .cache = CACHE_NEW_CLIP_PROG
+   },
+   .update = upload_clip_unit
+};
diff --git a/src/gallium/drivers/i965simple/brw_clip_tri.c b/src/gallium/drivers/i965simple/brw_clip_tri.c
new file mode 100644
index 0000000000..c5da7b825e
--- /dev/null
+++ b/src/gallium/drivers/i965simple/brw_clip_tri.c
@@ -0,0 +1,566 @@
+/*
+ Copyright (C) Intel Corp.  2006.  All Rights Reserved.
+ Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to
+ develop this 3D driver.
+
+ Permission is hereby granted, free of charge, to any person obtaining
+ a copy of this software and associated documentation files (the
+ "Software"), to deal in the Software without restriction, including
+ without limitation the rights to use, copy, modify, merge, publish,
+ distribute, sublicense, and/or sell copies of the Software, and to
+ permit persons to whom the Software is furnished to do so, subject to
+ the following conditions:
+
+ The above copyright notice and this permission notice (including the
+ next paragraph) shall be included in all copies or substantial
+ portions of the Software.
+
+ THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
+ IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
+ LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
+ OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+
+ **********************************************************************/
+ /*
+  * Authors:
+  *   Keith Whitwell <keith@tungstengraphics.com>
+  */
+
+#include "brw_defines.h"
+#include "brw_context.h"
+#include "brw_eu.h"
+#include "brw_util.h"
+#include "brw_clip.h"
+
+static struct brw_reg get_tmp( struct brw_clip_compile *c )
+{
+   struct brw_reg tmp = brw_vec4_grf(c->last_tmp, 0);
+
+   if (++c->last_tmp > c->prog_data.total_grf)
+      c->prog_data.total_grf = c->last_tmp;
+
+   return tmp;
+}
+
+static void release_tmps( struct brw_clip_compile *c )
+{
+   c->last_tmp = c->first_tmp;
+}
+
+
+void brw_clip_tri_alloc_regs( struct brw_clip_compile *c,
+			      unsigned nr_verts )
+{
+   unsigned i = 0,j;
+
+   /* Register usage is static, precompute here:
+    */
+   c->reg.R0 = retype(brw_vec8_grf(i, 0), BRW_REGISTER_TYPE_UD); i++;
+
+   if (c->key.nr_userclip) {
+      c->reg.fixed_planes = brw_vec4_grf(i, 0);
+      i += (6 + c->key.nr_userclip + 1) / 2;
+
+      c->prog_data.curb_read_length = (6 + c->key.nr_userclip + 1) / 2;
+   }
+   else
+      c->prog_data.curb_read_length = 0;
+
+
+   /* Payload vertices plus space for more generated vertices:
+    */
+   for (j = 0; j < nr_verts; j++) {
+      c->reg.vertex[j] = brw_vec4_grf(i, 0);
+      i += c->nr_regs;
+   }
+
+   if (c->nr_attrs & 1) {
+      for (j = 0; j < 3; j++) {
+	 unsigned delta = c->nr_attrs*16 + 32;
+	 brw_MOV(&c->func, byte_offset(c->reg.vertex[j], delta), brw_imm_f(0));
+      }
+   }
+
+   c->reg.t          = brw_vec1_grf(i, 0);
+   c->reg.loopcount  = retype(brw_vec1_grf(i, 1), BRW_REGISTER_TYPE_UD);
+   c->reg.nr_verts   = retype(brw_vec1_grf(i, 2), BRW_REGISTER_TYPE_UD);
+   c->reg.planemask  = retype(brw_vec1_grf(i, 3), BRW_REGISTER_TYPE_UD);
+   c->reg.plane_equation = brw_vec4_grf(i, 4);
+   i++;
+
+   c->reg.dpPrev     = brw_vec1_grf(i, 0); /* fixme - dp4 will clobber r.1,2,3 */
+   c->reg.dp         = brw_vec1_grf(i, 4);
+   i++;
+
+   c->reg.inlist     = brw_uw16_reg(BRW_GENERAL_REGISTER_FILE, i, 0);
+   i++;
+
+   c->reg.outlist    = brw_uw16_reg(BRW_GENERAL_REGISTER_FILE, i, 0);
+   i++;
+
+   c->reg.freelist   = brw_uw16_reg(BRW_GENERAL_REGISTER_FILE, i, 0);
+   i++;
+
+   if (!c->key.nr_userclip) {
+      c->reg.fixed_planes = brw_vec8_grf(i, 0);
+      i++;
+   }
+
+   if (c->key.do_unfilled) {
+      c->reg.dir     = brw_vec4_grf(i, 0);
+      c->reg.offset  = brw_vec4_grf(i, 4);
+      i++;
+      c->reg.tmp0    = brw_vec4_grf(i, 0);
+      c->reg.tmp1    = brw_vec4_grf(i, 4);
+      i++;
+   }
+
+   c->first_tmp = i;
+   c->last_tmp = i;
+
+   c->prog_data.urb_read_length = c->nr_regs; /* ? */
+   c->prog_data.total_grf = i;
+}
+
+
+
+void brw_clip_tri_init_vertices( struct brw_clip_compile *c )
+{
+   struct brw_compile *p = &c->func;
+   struct brw_reg tmp0 = c->reg.loopcount; /* handy temporary */
+   struct brw_instruction *is_rev;
+
+   /* Initial list of indices for incoming vertexes:
+    */
+   brw_AND(p, tmp0, get_element_ud(c->reg.R0, 2), brw_imm_ud(PRIM_MASK));
+   brw_CMP(p,
+	   vec1(brw_null_reg()),
+	   BRW_CONDITIONAL_EQ,
+	   tmp0,
+	   brw_imm_ud(_3DPRIM_TRISTRIP_REVERSE));
+
+   /* XXX: Is there an easier way to do this?  Need to reverse every
+    * second tristrip element:  Can ignore sometimes?
+    */
+   is_rev = brw_IF(p, BRW_EXECUTE_1);
+   {
+      brw_MOV(p, get_element(c->reg.inlist, 0),  brw_address(c->reg.vertex[1]) );
+      brw_MOV(p, get_element(c->reg.inlist, 1),  brw_address(c->reg.vertex[0]) );
+      if (c->need_direction)
+	 brw_MOV(p, c->reg.dir, brw_imm_f(-1));
+   }
+   is_rev = brw_ELSE(p, is_rev);
+   {
+      brw_MOV(p, get_element(c->reg.inlist, 0),  brw_address(c->reg.vertex[0]) );
+      brw_MOV(p, get_element(c->reg.inlist, 1),  brw_address(c->reg.vertex[1]) );
+      if (c->need_direction)
+	 brw_MOV(p, c->reg.dir, brw_imm_f(1));
+   }
+   brw_ENDIF(p, is_rev);
+
+   brw_MOV(p, get_element(c->reg.inlist, 2),  brw_address(c->reg.vertex[2]) );
+   brw_MOV(p, brw_vec8_grf(c->reg.outlist.nr, 0), brw_imm_f(0));
+   brw_MOV(p, c->reg.nr_verts, brw_imm_ud(3));
+}
+
+
+
+void brw_clip_tri_flat_shade( struct brw_clip_compile *c )
+{
+   struct brw_compile *p = &c->func;
+   struct brw_instruction *is_poly;
+   struct brw_reg tmp0 = c->reg.loopcount; /* handy temporary */
+
+   brw_AND(p, tmp0, get_element_ud(c->reg.R0, 2), brw_imm_ud(PRIM_MASK));
+   brw_CMP(p,
+	   vec1(brw_null_reg()),
+	   BRW_CONDITIONAL_EQ,
+	   tmp0,
+	   brw_imm_ud(_3DPRIM_POLYGON));
+
+   is_poly = brw_IF(p, BRW_EXECUTE_1);
+   {
+      brw_clip_copy_colors(c, 1, 0);
+      brw_clip_copy_colors(c, 2, 0);
+   }
+   is_poly = brw_ELSE(p, is_poly);
+   {
+      brw_clip_copy_colors(c, 0, 2);
+      brw_clip_copy_colors(c, 1, 2);
+   }
+   brw_ENDIF(p, is_poly);
+}
+
+
+
+/* Use mesa's clipping algorithms, translated to GEN4 assembly.
+ */
+void brw_clip_tri( struct brw_clip_compile *c )
+{
+#if 0
+   struct brw_compile *p = &c->func;
+   struct brw_indirect vtx = brw_indirect(0, 0);
+   struct brw_indirect vtxPrev = brw_indirect(1, 0);
+   struct brw_indirect vtxOut = brw_indirect(2, 0);
+   struct brw_indirect plane_ptr = brw_indirect(3, 0);
+   struct brw_indirect inlist_ptr = brw_indirect(4, 0);
+   struct brw_indirect outlist_ptr = brw_indirect(5, 0);
+   struct brw_indirect freelist_ptr = brw_indirect(6, 0);
+   struct brw_instruction *plane_loop;
+   struct brw_instruction *plane_active;
+   struct brw_instruction *vertex_loop;
+   struct brw_instruction *next_test;
+   struct brw_instruction *prev_test;
+
+   brw_MOV(p, get_addr_reg(vtxPrev),     brw_address(c->reg.vertex[2]) );
+   brw_MOV(p, get_addr_reg(plane_ptr),   brw_clip_plane0_address(c));
+   brw_MOV(p, get_addr_reg(inlist_ptr),  brw_address(c->reg.inlist));
+   brw_MOV(p, get_addr_reg(outlist_ptr), brw_address(c->reg.outlist));
+
+   brw_MOV(p, get_addr_reg(freelist_ptr), brw_address(c->reg.vertex[3]) );
+
+   plane_loop = brw_DO(p, BRW_EXECUTE_1);
+   {
+      /* if (planemask & 1)
+       */
+      brw_set_conditionalmod(p, BRW_CONDITIONAL_NZ);
+      brw_AND(p, vec1(brw_null_reg()), c->reg.planemask, brw_imm_ud(1));
+
+      plane_active = brw_IF(p, BRW_EXECUTE_1);
+      {
+	 /* vtxOut = freelist_ptr++
+	  */
+	 brw_MOV(p, get_addr_reg(vtxOut),       get_addr_reg(freelist_ptr) );
+	 brw_ADD(p, get_addr_reg(freelist_ptr), get_addr_reg(freelist_ptr), brw_imm_uw(c->nr_regs * REG_SIZE));
+
+	 if (c->key.nr_userclip)
+	    brw_MOV(p, c->reg.plane_equation, deref_4f(plane_ptr, 0));
+	 else
+	    brw_MOV(p, c->reg.plane_equation, deref_4b(plane_ptr, 0));
+
+	 brw_MOV(p, c->reg.loopcount, c->reg.nr_verts);
+	 brw_MOV(p, c->reg.nr_verts, brw_imm_ud(0));
+
+	 vertex_loop = brw_DO(p, BRW_EXECUTE_1);
+	 {
+	    /* vtx = *input_ptr;
+	     */
+	    brw_MOV(p, get_addr_reg(vtx), deref_1uw(inlist_ptr, 0));
+
+	    /* IS_NEGATIVE(prev) */
+	    brw_set_conditionalmod(p, BRW_CONDITIONAL_L);
+	    brw_DP4(p, vec4(c->reg.dpPrev), deref_4f(vtxPrev, c->offset[VERT_RESULT_HPOS]), c->reg.plane_equation);
+	    prev_test = brw_IF(p, BRW_EXECUTE_1);
+	    {
+	       /* IS_POSITIVE(next)
+		*/
+	       brw_set_conditionalmod(p, BRW_CONDITIONAL_GE);
+	       brw_DP4(p, vec4(c->reg.dp), deref_4f(vtx, c->offset[VERT_RESULT_HPOS]), c->reg.plane_equation);
+	       next_test = brw_IF(p, BRW_EXECUTE_1);
+	       {
+
+		  /* Coming back in.
+		   */
+		  brw_ADD(p, c->reg.t, c->reg.dpPrev, negate(c->reg.dp));
+		  brw_math_invert(p, c->reg.t, c->reg.t);
+		  brw_MUL(p, c->reg.t, c->reg.t, c->reg.dpPrev);
+
+		  /* If (vtxOut == 0) vtxOut = vtxPrev
+		   */
+		  brw_CMP(p, vec1(brw_null_reg()), BRW_CONDITIONAL_EQ, get_addr_reg(vtxOut), brw_imm_uw(0) );
+		  brw_MOV(p, get_addr_reg(vtxOut), get_addr_reg(vtxPrev) );
+		  brw_set_predicate_control(p, BRW_PREDICATE_NONE);
+
+		  brw_clip_interp_vertex(c, vtxOut, vtxPrev, vtx, c->reg.t, FALSE);
+
+		  /* *outlist_ptr++ = vtxOut;
+		   * nr_verts++;
+		   * vtxOut = 0;
+		   */
+		  brw_MOV(p, deref_1uw(outlist_ptr, 0), get_addr_reg(vtxOut));
+		  brw_ADD(p, get_addr_reg(outlist_ptr), get_addr_reg(outlist_ptr), brw_imm_uw(sizeof(short)));
+		  brw_ADD(p, c->reg.nr_verts, c->reg.nr_verts, brw_imm_ud(1));
+		  brw_MOV(p, get_addr_reg(vtxOut), brw_imm_uw(0) );
+	       }
+	       brw_ENDIF(p, next_test);
+
+	    }
+	    prev_test = brw_ELSE(p, prev_test);
+	    {
+	       /* *outlist_ptr++ = vtxPrev;
+		* nr_verts++;
+		*/
+	       brw_MOV(p, deref_1uw(outlist_ptr, 0), get_addr_reg(vtxPrev));
+	       brw_ADD(p, get_addr_reg(outlist_ptr), get_addr_reg(outlist_ptr), brw_imm_uw(sizeof(short)));
+	       brw_ADD(p, c->reg.nr_verts, c->reg.nr_verts, brw_imm_ud(1));
+
+	       /* IS_NEGATIVE(next)
+		*/
+	       brw_set_conditionalmod(p, BRW_CONDITIONAL_L);
+	       brw_DP4(p, vec4(c->reg.dp), deref_4f(vtx, c->offset[VERT_RESULT_HPOS]), c->reg.plane_equation);
+	       next_test = brw_IF(p, BRW_EXECUTE_1);
+	       {
+		  /* Going out of bounds.  Avoid division by zero as we
+		   * know dp != dpPrev from DIFFERENT_SIGNS, above.
+		   */
+		  brw_ADD(p, c->reg.t, c->reg.dp, negate(c->reg.dpPrev));
+		  brw_math_invert(p, c->reg.t, c->reg.t);
+		  brw_MUL(p, c->reg.t, c->reg.t, c->reg.dp);
+
+		  /* If (vtxOut == 0) vtxOut = vtx
+		   */
+		  brw_CMP(p, vec1(brw_null_reg()), BRW_CONDITIONAL_EQ, get_addr_reg(vtxOut), brw_imm_uw(0) );
+		  brw_MOV(p, get_addr_reg(vtxOut), get_addr_reg(vtx) );
+		  brw_set_predicate_control(p, BRW_PREDICATE_NONE);
+
+		  brw_clip_interp_vertex(c, vtxOut, vtx, vtxPrev, c->reg.t, TRUE);
+
+		  /* *outlist_ptr++ = vtxOut;
+		   * nr_verts++;
+		   * vtxOut = 0;
+		   */
+		  brw_MOV(p, deref_1uw(outlist_ptr, 0), get_addr_reg(vtxOut));
+		  brw_ADD(p, get_addr_reg(outlist_ptr), get_addr_reg(outlist_ptr), brw_imm_uw(sizeof(short)));
+		  brw_ADD(p, c->reg.nr_verts, c->reg.nr_verts, brw_imm_ud(1));
+		  brw_MOV(p, get_addr_reg(vtxOut), brw_imm_uw(0) );
+	       }
+	       brw_ENDIF(p, next_test);
+	    }
+	    brw_ENDIF(p, prev_test);
+
+	    /* vtxPrev = vtx;
+	     * inlist_ptr++;
+	     */
+	    brw_MOV(p, get_addr_reg(vtxPrev), get_addr_reg(vtx));
+	    brw_ADD(p, get_addr_reg(inlist_ptr), get_addr_reg(inlist_ptr), brw_imm_uw(sizeof(short)));
+
+	    /* while (--loopcount != 0)
+	     */
+	    brw_set_conditionalmod(p, BRW_CONDITIONAL_NZ);
+	    brw_ADD(p, c->reg.loopcount, c->reg.loopcount, brw_imm_d(-1));
+	 }
+	 brw_WHILE(p, vertex_loop);
+
+	 /* vtxPrev = *(outlist_ptr-1)  OR: outlist[nr_verts-1]
+	  * inlist = outlist
+	  * inlist_ptr = &inlist[0]
+	  * outlist_ptr = &outlist[0]
+	  */
+	 brw_ADD(p, get_addr_reg(outlist_ptr), get_addr_reg(outlist_ptr), brw_imm_w(-2));
+	 brw_MOV(p, get_addr_reg(vtxPrev), deref_1uw(outlist_ptr, 0));
+	 brw_MOV(p, brw_vec8_grf(c->reg.inlist.nr, 0), brw_vec8_grf(c->reg.outlist.nr, 0));
+	 brw_MOV(p, get_addr_reg(inlist_ptr), brw_address(c->reg.inlist));
+	 brw_MOV(p, get_addr_reg(outlist_ptr), brw_address(c->reg.outlist));
+      }
+      brw_ENDIF(p, plane_active);
+
+      /* plane_ptr++;
+       */
+      brw_ADD(p, get_addr_reg(plane_ptr), get_addr_reg(plane_ptr), brw_clip_plane_stride(c));
+
+      /* nr_verts >= 3
+       */
+      brw_CMP(p,
+	      vec1(brw_null_reg()),
+	      BRW_CONDITIONAL_GE,
+	      c->reg.nr_verts,
+	      brw_imm_ud(3));
+
+      /* && (planemask>>=1) != 0
+       */
+      brw_set_conditionalmod(p, BRW_CONDITIONAL_NZ);
+      brw_SHR(p, c->reg.planemask, c->reg.planemask, brw_imm_ud(1));
+   }
+   brw_WHILE(p, plane_loop);
+#else
+         #warning "disabled"
+#endif
+}
+
+
+
+void brw_clip_tri_emit_polygon(struct brw_clip_compile *c)
+{
+   struct brw_compile *p = &c->func;
+   struct brw_instruction *loop, *if_insn;
+
+   /* for (loopcount = nr_verts-2; loopcount > 0; loopcount--)
+    */
+   brw_set_conditionalmod(p, BRW_CONDITIONAL_G);
+   brw_ADD(p,
+	   c->reg.loopcount,
+	   c->reg.nr_verts,
+	   brw_imm_d(-2));
+
+   if_insn = brw_IF(p, BRW_EXECUTE_1);
+   {
+      struct brw_indirect v0 = brw_indirect(0, 0);
+      struct brw_indirect vptr = brw_indirect(1, 0);
+
+      brw_MOV(p, get_addr_reg(vptr), brw_address(c->reg.inlist));
+      brw_MOV(p, get_addr_reg(v0), deref_1uw(vptr, 0));
+
+      brw_clip_emit_vue(c, v0, 1, 0, ((_3DPRIM_TRIFAN << 2) | R02_PRIM_START));
+
+      brw_ADD(p, get_addr_reg(vptr), get_addr_reg(vptr), brw_imm_uw(2));
+      brw_MOV(p, get_addr_reg(v0), deref_1uw(vptr, 0));
+
+      loop = brw_DO(p, BRW_EXECUTE_1);
+      {
+	 brw_clip_emit_vue(c, v0, 1, 0, (_3DPRIM_TRIFAN << 2));
+
+	 brw_ADD(p, get_addr_reg(vptr), get_addr_reg(vptr), brw_imm_uw(2));
+	 brw_MOV(p, get_addr_reg(v0), deref_1uw(vptr, 0));
+
+	 brw_set_conditionalmod(p, BRW_CONDITIONAL_NZ);
+	 brw_ADD(p, c->reg.loopcount, c->reg.loopcount, brw_imm_d(-1));
+      }
+      brw_WHILE(p, loop);
+
+      brw_clip_emit_vue(c, v0, 0, 1, ((_3DPRIM_TRIFAN << 2) | R02_PRIM_END));
+   }
+   brw_ENDIF(p, if_insn);
+}
+
+static void do_clip_tri( struct brw_clip_compile *c )
+{
+   brw_clip_init_planes(c);
+
+   brw_clip_tri(c);
+}
+
+
+static void maybe_do_clip_tri( struct brw_clip_compile *c )
+{
+   struct brw_compile *p = &c->func;
+   struct brw_instruction *do_clip;
+
+   brw_CMP(p, vec1(brw_null_reg()), BRW_CONDITIONAL_NZ, c->reg.planemask, brw_imm_ud(0));
+   do_clip = brw_IF(p, BRW_EXECUTE_1);
+   {
+      do_clip_tri(c);
+   }
+   brw_ENDIF(p, do_clip);
+}
+
+static void brw_clip_test( struct brw_clip_compile *c )
+{
+#if 0
+    struct brw_reg t = retype(get_tmp(c), BRW_REGISTER_TYPE_UD);
+    struct brw_reg t1 = retype(get_tmp(c), BRW_REGISTER_TYPE_UD);
+    struct brw_reg t2 = retype(get_tmp(c), BRW_REGISTER_TYPE_UD);
+    struct brw_reg t3 = retype(get_tmp(c), BRW_REGISTER_TYPE_UD);
+
+    struct brw_reg v0 = get_tmp(c);
+    struct brw_reg v1 = get_tmp(c);
+    struct brw_reg v2 = get_tmp(c);
+
+    struct brw_indirect vt0 = brw_indirect(0, 0);
+    struct brw_indirect vt1 = brw_indirect(1, 0);
+    struct brw_indirect vt2 = brw_indirect(2, 0);
+
+    struct brw_compile *p = &c->func;
+
+    brw_MOV(p, get_addr_reg(vt0), brw_address(c->reg.vertex[0]));
+    brw_MOV(p, get_addr_reg(vt1), brw_address(c->reg.vertex[1]));
+    brw_MOV(p, get_addr_reg(vt2), brw_address(c->reg.vertex[2]));
+    brw_MOV(p, v0, deref_4f(vt0, c->offset[VERT_RESULT_HPOS]));
+    brw_MOV(p, v1, deref_4f(vt1, c->offset[VERT_RESULT_HPOS]));
+    brw_MOV(p, v2, deref_4f(vt2, c->offset[VERT_RESULT_HPOS]));
+
+    /* test nearz, xmin, ymin plane */
+    brw_CMP(p, t1, BRW_CONDITIONAL_LE, negate(v0), get_element(v0, 3));
+    brw_set_predicate_control(p, BRW_PREDICATE_NONE);
+    brw_CMP(p, t2, BRW_CONDITIONAL_LE, negate(v1), get_element(v1, 3));
+    brw_set_predicate_control(p, BRW_PREDICATE_NONE);
+    brw_CMP(p, t3, BRW_CONDITIONAL_LE, negate(v2), get_element(v2, 3));
+    brw_set_predicate_control(p, BRW_PREDICATE_NONE);
+    brw_XOR(p, t, t1, t2);
+    brw_XOR(p, t1, t2, t3);
+    brw_OR(p, t, t, t1);
+
+    brw_CMP(p, brw_null_reg(), BRW_CONDITIONAL_NZ,
+	    get_element(t, 0), brw_imm_ud(0));
+    brw_OR(p, c->reg.planemask, c->reg.planemask, brw_imm_ud((1<<5)));
+    brw_set_predicate_control(p, BRW_PREDICATE_NONE);
+    brw_CMP(p, brw_null_reg(), BRW_CONDITIONAL_NZ,
+	    get_element(t, 1), brw_imm_ud(0));
+    brw_OR(p, c->reg.planemask, c->reg.planemask, brw_imm_ud((1<<3)));
+    brw_set_predicate_control(p, BRW_PREDICATE_NONE);
+    brw_CMP(p, brw_null_reg(), BRW_CONDITIONAL_NZ,
+	    get_element(t, 2), brw_imm_ud(0));
+    brw_OR(p, c->reg.planemask, c->reg.planemask, brw_imm_ud((1<<1)));
+    brw_set_predicate_control(p, BRW_PREDICATE_NONE);
+
+    /* test farz, xmax, ymax plane */
+    brw_CMP(p, t1, BRW_CONDITIONAL_L, v0, get_element(v0, 3));
+    brw_set_predicate_control(p, BRW_PREDICATE_NONE);
+    brw_CMP(p, t2, BRW_CONDITIONAL_L, v1, get_element(v1, 3));
+    brw_set_predicate_control(p, BRW_PREDICATE_NONE);
+    brw_CMP(p, t3, BRW_CONDITIONAL_L, v2, get_element(v2, 3));
+    brw_set_predicate_control(p, BRW_PREDICATE_NONE);
+
+    brw_XOR(p, t, t1, t2);
+    brw_XOR(p, t1, t2, t3);
+    brw_OR(p, t, t, t1);
+
+    brw_CMP(p, brw_null_reg(), BRW_CONDITIONAL_NZ,
+	    get_element(t, 0), brw_imm_ud(0));
+    brw_OR(p, c->reg.planemask, c->reg.planemask, brw_imm_ud((1<<4)));
+    brw_set_predicate_control(p, BRW_PREDICATE_NONE);
+    brw_CMP(p, brw_null_reg(), BRW_CONDITIONAL_NZ,
+	    get_element(t, 1), brw_imm_ud(0));
+    brw_OR(p, c->reg.planemask, c->reg.planemask, brw_imm_ud((1<<2)));
+    brw_set_predicate_control(p, BRW_PREDICATE_NONE);
+    brw_CMP(p, brw_null_reg(), BRW_CONDITIONAL_NZ,
+	    get_element(t, 2), brw_imm_ud(0));
+    brw_OR(p, c->reg.planemask, c->reg.planemask, brw_imm_ud((1<<0)));
+    brw_set_predicate_control(p, BRW_PREDICATE_NONE);
+
+    release_tmps(c);
+#else
+         #warning "disabled"
+#endif
+}
+
+
+void brw_emit_tri_clip( struct brw_clip_compile *c )
+{
+   struct brw_instruction *neg_rhw;
+   struct brw_compile *p = &c->func;
+   brw_clip_tri_alloc_regs(c, 3 + c->key.nr_userclip + 6);
+   brw_clip_tri_init_vertices(c);
+   brw_clip_init_clipmask(c);
+
+   /* if -ve rhw workaround bit is set,
+      do cliptest */
+   brw_set_conditionalmod(p, BRW_CONDITIONAL_NZ);
+   brw_AND(p, brw_null_reg(), get_element_ud(c->reg.R0, 2),
+	   brw_imm_ud(1<<20));
+   neg_rhw = brw_IF(p, BRW_EXECUTE_1);
+   {
+       brw_clip_test(c);
+   }
+   brw_ENDIF(p, neg_rhw);
+
+   /* Can't push into do_clip_tri because with polygon (or quad)
+    * flatshading, need to apply the flatshade here because we don't
+    * respect the PV when converting to trifan for emit:
+    */
+   if (c->key.do_flat_shading)
+      brw_clip_tri_flat_shade(c);
+
+   if (c->key.clip_mode == BRW_CLIPMODE_NORMAL)
+      do_clip_tri(c);
+   else
+      maybe_do_clip_tri(c);
+
+   brw_clip_tri_emit_polygon(c);
+
+   /* Send an empty message to kill the thread:
+    */
+   brw_clip_kill_thread(c);
+}
diff --git a/src/gallium/drivers/i965simple/brw_clip_unfilled.c b/src/gallium/drivers/i965simple/brw_clip_unfilled.c
new file mode 100644
index 0000000000..b774a76dd6
--- /dev/null
+++ b/src/gallium/drivers/i965simple/brw_clip_unfilled.c
@@ -0,0 +1,477 @@
+/*
+ Copyright (C) Intel Corp.  2006.  All Rights Reserved.
+ Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to
+ develop this 3D driver.
+ 
+ Permission is hereby granted, free of charge, to any person obtaining
+ a copy of this software and associated documentation files (the
+ "Software"), to deal in the Software without restriction, including
+ without limitation the rights to use, copy, modify, merge, publish,
+ distribute, sublicense, and/or sell copies of the Software, and to
+ permit persons to whom the Software is furnished to do so, subject to
+ the following conditions:
+ 
+ The above copyright notice and this permission notice (including the
+ next paragraph) shall be included in all copies or substantial
+ portions of the Software.
+ 
+ THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
+ IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
+ LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
+ OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ 
+ **********************************************************************/
+ /*
+  * Authors:
+  *   Keith Whitwell <keith@tungstengraphics.com>
+  */
+
+#include "brw_defines.h"
+#include "brw_context.h"
+#include "brw_eu.h"
+#include "brw_util.h"
+#include "brw_clip.h"
+
+
+
+/* This is performed against the original triangles, so no indirection
+ * required:
+BZZZT!
+ */
+static void compute_tri_direction( struct brw_clip_compile *c )
+{
+   struct brw_compile *p = &c->func;
+   struct brw_reg e = c->reg.tmp0;
+   struct brw_reg f = c->reg.tmp1;
+   struct brw_reg v0 = byte_offset(c->reg.vertex[0], c->offset[VERT_RESULT_HPOS]); 
+   struct brw_reg v1 = byte_offset(c->reg.vertex[1], c->offset[VERT_RESULT_HPOS]); 
+   struct brw_reg v2 = byte_offset(c->reg.vertex[2], c->offset[VERT_RESULT_HPOS]); 
+
+
+   /* Calculate the vectors of two edges of the triangle:
+    */
+   brw_ADD(p, e, v0, negate(v2)); 
+   brw_ADD(p, f, v1, negate(v2)); 
+
+   /* Take their crossproduct:
+    */
+   brw_set_access_mode(p, BRW_ALIGN_16);
+   brw_MUL(p, vec4(brw_null_reg()), brw_swizzle(e, 1,2,0,3),  brw_swizzle(f,2,0,1,3));
+   brw_MAC(p, vec4(e),  negate(brw_swizzle(e, 2,0,1,3)), brw_swizzle(f,1,2,0,3));
+   brw_set_access_mode(p, BRW_ALIGN_1);
+
+   brw_MUL(p, c->reg.dir, c->reg.dir, vec4(e));
+}
+
+
+static void cull_direction( struct brw_clip_compile *c )
+{
+   struct brw_compile *p = &c->func;
+   struct brw_instruction *ccw;
+   unsigned conditional;
+
+   assert (!(c->key.fill_ccw == CLIP_CULL &&
+	     c->key.fill_cw == CLIP_CULL));
+
+   if (c->key.fill_ccw == CLIP_CULL)
+      conditional = BRW_CONDITIONAL_GE;
+   else
+      conditional = BRW_CONDITIONAL_L;
+
+   brw_CMP(p,
+	   vec1(brw_null_reg()),
+	   conditional,
+	   get_element(c->reg.dir, 2),
+	   brw_imm_f(0));
+   
+   ccw = brw_IF(p, BRW_EXECUTE_1);
+   {
+      brw_clip_kill_thread(c);
+   }
+   brw_ENDIF(p, ccw);
+}
+
+
+
+static void copy_bfc( struct brw_clip_compile *c )
+{
+   struct brw_compile *p = &c->func;
+   struct brw_instruction *ccw;
+   unsigned conditional;
+
+   /* Do we have any colors to copy? 
+    */
+   if (!(c->offset[VERT_RESULT_COL0] && c->offset[VERT_RESULT_BFC0]) &&
+       !(c->offset[VERT_RESULT_COL1] && c->offset[VERT_RESULT_BFC1]))
+      return;
+
+   /* In some wierd degnerate cases we can end up testing the
+    * direction twice, once for culling and once for bfc copying.  Oh
+    * well, that's what you get for setting wierd GL state.
+    */
+   if (c->key.copy_bfc_ccw)
+      conditional = BRW_CONDITIONAL_GE;
+   else
+      conditional = BRW_CONDITIONAL_L;
+
+   brw_CMP(p,
+	   vec1(brw_null_reg()),
+	   conditional,
+	   get_element(c->reg.dir, 2),
+	   brw_imm_f(0));
+   
+   ccw = brw_IF(p, BRW_EXECUTE_1);
+   {
+      unsigned i;
+
+      for (i = 0; i < 3; i++) {
+	 if (c->offset[VERT_RESULT_COL0] && c->offset[VERT_RESULT_BFC0])
+	    brw_MOV(p, 
+		    byte_offset(c->reg.vertex[i], c->offset[VERT_RESULT_COL0]),
+		    byte_offset(c->reg.vertex[i], c->offset[VERT_RESULT_BFC0]));
+
+	 if (c->offset[VERT_RESULT_COL1] && c->offset[VERT_RESULT_BFC1])
+	    brw_MOV(p, 
+		    byte_offset(c->reg.vertex[i], c->offset[VERT_RESULT_COL1]),
+		    byte_offset(c->reg.vertex[i], c->offset[VERT_RESULT_BFC1]));
+      }
+   }
+   brw_ENDIF(p, ccw);
+}
+
+
+
+
+/*
+  float iz	= 1.0 / dir.z;
+  float ac	= dir.x * iz;
+  float bc	= dir.y * iz;
+  offset = ctx->Polygon.OffsetUnits * DEPTH_SCALE;
+  offset += MAX2( abs(ac), abs(bc) ) * ctx->Polygon.OffsetFactor;
+  offset *= MRD;
+*/
+static void compute_offset( struct brw_clip_compile *c )
+{
+   struct brw_compile *p = &c->func;
+   struct brw_reg off = c->reg.offset;
+   struct brw_reg dir = c->reg.dir;
+   
+   brw_math_invert(p, get_element(off, 2), get_element(dir, 2));
+   brw_MUL(p, vec2(off), dir, get_element(off, 2));
+
+   brw_CMP(p, 
+	   vec1(brw_null_reg()), 
+	   BRW_CONDITIONAL_GE,
+	   brw_abs(get_element(off, 0)), 
+	   brw_abs(get_element(off, 1)));
+
+   brw_SEL(p, vec1(off), brw_abs(get_element(off, 0)), brw_abs(get_element(off, 1)));
+   brw_set_predicate_control(p, BRW_PREDICATE_NONE);
+
+   brw_MUL(p, vec1(off), off, brw_imm_f(c->key.offset_factor));
+   brw_ADD(p, vec1(off), off, brw_imm_f(c->key.offset_units));
+}
+
+
+static void merge_edgeflags( struct brw_clip_compile *c )
+{
+   struct brw_compile *p = &c->func;
+   struct brw_instruction *is_poly;
+   struct brw_reg tmp0 = get_element_ud(c->reg.tmp0, 0);
+
+   brw_AND(p, tmp0, get_element_ud(c->reg.R0, 2), brw_imm_ud(PRIM_MASK)); 
+   brw_CMP(p, 
+	   vec1(brw_null_reg()), 
+	   BRW_CONDITIONAL_EQ, 
+	   tmp0,
+	   brw_imm_ud(_3DPRIM_POLYGON));
+
+   /* Get away with using reg.vertex because we know that this is not
+    * a _3DPRIM_TRISTRIP_REVERSE:
+    */
+   is_poly = brw_IF(p, BRW_EXECUTE_1);
+   {   
+      brw_set_conditionalmod(p, BRW_CONDITIONAL_EQ);
+      brw_AND(p, vec1(brw_null_reg()), get_element_ud(c->reg.R0, 2), brw_imm_ud(1<<8));
+      brw_MOV(p, byte_offset(c->reg.vertex[0], c->offset[VERT_RESULT_EDGE]), brw_imm_f(0));
+      brw_set_predicate_control(p, BRW_PREDICATE_NONE);
+
+      brw_set_conditionalmod(p, BRW_CONDITIONAL_EQ);
+      brw_AND(p, vec1(brw_null_reg()), get_element_ud(c->reg.R0, 2), brw_imm_ud(1<<9));
+      brw_MOV(p, byte_offset(c->reg.vertex[2], c->offset[VERT_RESULT_EDGE]), brw_imm_f(0));
+      brw_set_predicate_control(p, BRW_PREDICATE_NONE);
+   }
+   brw_ENDIF(p, is_poly);
+}
+
+
+
+static void apply_one_offset( struct brw_clip_compile *c,
+			  struct brw_indirect vert )
+{
+   struct brw_compile *p = &c->func;
+   struct brw_reg pos = deref_4f(vert, c->offset[VERT_RESULT_HPOS]);
+   struct brw_reg z = get_element(pos, 2);
+
+   brw_ADD(p, z, z, vec1(c->reg.offset));
+}
+
+
+
+/***********************************************************************
+ * Output clipped polygon as an unfilled primitive:
+ */
+static void emit_lines(struct brw_clip_compile *c,
+		       boolean do_offset)
+{
+   struct brw_compile *p = &c->func;
+   struct brw_instruction *loop;
+   struct brw_instruction *draw_edge;
+   struct brw_indirect v0 = brw_indirect(0, 0);
+   struct brw_indirect v1 = brw_indirect(1, 0);
+   struct brw_indirect v0ptr = brw_indirect(2, 0);
+   struct brw_indirect v1ptr = brw_indirect(3, 0);
+
+   /* Need a seperate loop for offset:
+    */
+   if (do_offset) {
+      brw_MOV(p, c->reg.loopcount, c->reg.nr_verts);
+      brw_MOV(p, get_addr_reg(v0ptr), brw_address(c->reg.inlist));
+
+      loop = brw_DO(p, BRW_EXECUTE_1);
+      {
+	 brw_MOV(p, get_addr_reg(v0), deref_1uw(v0ptr, 0));
+	 brw_ADD(p, get_addr_reg(v0ptr), get_addr_reg(v0ptr), brw_imm_uw(2));
+	    
+	 apply_one_offset(c, v0);
+	    
+	 brw_set_conditionalmod(p, BRW_CONDITIONAL_G);
+	 brw_ADD(p, c->reg.loopcount, c->reg.loopcount, brw_imm_d(-1));
+      }
+      brw_WHILE(p, loop);
+   }
+
+   /* v1ptr = &inlist[nr_verts]
+    * *v1ptr = v0
+    */
+   brw_MOV(p, c->reg.loopcount, c->reg.nr_verts);
+   brw_MOV(p, get_addr_reg(v0ptr), brw_address(c->reg.inlist));
+   brw_ADD(p, get_addr_reg(v1ptr), get_addr_reg(v0ptr), retype(c->reg.nr_verts, BRW_REGISTER_TYPE_UW));
+   brw_ADD(p, get_addr_reg(v1ptr), get_addr_reg(v1ptr), retype(c->reg.nr_verts, BRW_REGISTER_TYPE_UW));
+   brw_MOV(p, deref_1uw(v1ptr, 0), deref_1uw(v0ptr, 0));
+
+   loop = brw_DO(p, BRW_EXECUTE_1);
+   {
+      brw_MOV(p, get_addr_reg(v0), deref_1uw(v0ptr, 0));
+      brw_MOV(p, get_addr_reg(v1), deref_1uw(v0ptr, 2));
+      brw_ADD(p, get_addr_reg(v0ptr), get_addr_reg(v0ptr), brw_imm_uw(2));
+
+      /* draw edge if edgeflag != 0 */
+      brw_CMP(p, 
+	      vec1(brw_null_reg()), BRW_CONDITIONAL_NZ, 
+	      deref_1f(v0, c->offset[VERT_RESULT_EDGE]),
+	      brw_imm_f(0));
+      draw_edge = brw_IF(p, BRW_EXECUTE_1);
+      {
+	 brw_clip_emit_vue(c, v0, 1, 0, (_3DPRIM_LINESTRIP << 2) | R02_PRIM_START);
+	 brw_clip_emit_vue(c, v1, 1, 0, (_3DPRIM_LINESTRIP << 2) | R02_PRIM_END);
+      }
+      brw_ENDIF(p, draw_edge);
+
+      brw_set_conditionalmod(p, BRW_CONDITIONAL_NZ);
+      brw_ADD(p, c->reg.loopcount, c->reg.loopcount, brw_imm_d(-1));
+   }
+   brw_WHILE(p, loop);
+}
+
+
+
+static void emit_points(struct brw_clip_compile *c,
+			boolean do_offset )
+{
+   struct brw_compile *p = &c->func;
+   struct brw_instruction *loop;
+   struct brw_instruction *draw_point;
+
+   struct brw_indirect v0 = brw_indirect(0, 0);
+   struct brw_indirect v0ptr = brw_indirect(2, 0);
+
+   brw_MOV(p, c->reg.loopcount, c->reg.nr_verts);
+   brw_MOV(p, get_addr_reg(v0ptr), brw_address(c->reg.inlist));
+
+   loop = brw_DO(p, BRW_EXECUTE_1);
+   {
+      brw_MOV(p, get_addr_reg(v0), deref_1uw(v0ptr, 0));
+      brw_ADD(p, get_addr_reg(v0ptr), get_addr_reg(v0ptr), brw_imm_uw(2));
+
+      /* draw if edgeflag != 0 
+       */
+      brw_CMP(p, 
+	      vec1(brw_null_reg()), BRW_CONDITIONAL_NZ, 
+	      deref_1f(v0, c->offset[VERT_RESULT_EDGE]),
+	      brw_imm_f(0));
+      draw_point = brw_IF(p, BRW_EXECUTE_1);
+      {
+	 if (do_offset)
+	    apply_one_offset(c, v0);
+
+	 brw_clip_emit_vue(c, v0, 1, 0, (_3DPRIM_POINTLIST << 2) | R02_PRIM_START | R02_PRIM_END);
+      }
+      brw_ENDIF(p, draw_point);
+
+      brw_set_conditionalmod(p, BRW_CONDITIONAL_NZ);
+      brw_ADD(p, c->reg.loopcount, c->reg.loopcount, brw_imm_d(-1));
+   }
+   brw_WHILE(p, loop);
+}
+
+
+
+
+
+
+
+static void emit_primitives( struct brw_clip_compile *c,
+			     unsigned mode, 
+			     boolean do_offset )
+{
+   switch (mode) {
+   case CLIP_FILL:
+      brw_clip_tri_emit_polygon(c);
+      break;
+
+   case CLIP_LINE:
+      emit_lines(c, do_offset);
+      break;
+
+   case CLIP_POINT:
+      emit_points(c, do_offset);
+      break;
+
+   case CLIP_CULL:
+      assert(0);
+      break;
+   }
+} 
+
+
+
+static void emit_unfilled_primitives( struct brw_clip_compile *c )
+{
+   struct brw_compile *p = &c->func;
+   struct brw_instruction *ccw;
+
+   /* Direction culling has already been done.
+    */
+   if (c->key.fill_ccw != c->key.fill_cw &&
+       c->key.fill_ccw != CLIP_CULL &&
+       c->key.fill_cw != CLIP_CULL)
+   {
+      brw_CMP(p,
+	      vec1(brw_null_reg()),
+	      BRW_CONDITIONAL_GE,
+	      get_element(c->reg.dir, 2),
+	      brw_imm_f(0));
+   
+      ccw = brw_IF(p, BRW_EXECUTE_1);
+      {
+	 emit_primitives(c, c->key.fill_ccw, c->key.offset_ccw);
+      }
+      ccw = brw_ELSE(p, ccw);
+      {
+	 emit_primitives(c, c->key.fill_cw, c->key.offset_cw);
+      }
+      brw_ENDIF(p, ccw);
+   }
+   else if (c->key.fill_cw != CLIP_CULL) {
+      emit_primitives(c, c->key.fill_cw, c->key.offset_cw);
+   }
+   else if (c->key.fill_ccw != CLIP_CULL) { 
+      emit_primitives(c, c->key.fill_ccw, c->key.offset_ccw);
+   }
+}
+
+
+
+
+static void check_nr_verts( struct brw_clip_compile *c )
+{
+   struct brw_compile *p = &c->func;
+   struct brw_instruction *if_insn;
+
+   brw_CMP(p, vec1(brw_null_reg()), BRW_CONDITIONAL_L, c->reg.nr_verts, brw_imm_d(3));      
+   if_insn = brw_IF(p, BRW_EXECUTE_1);
+   {
+      brw_clip_kill_thread(c);
+   }
+   brw_ENDIF(p, if_insn);
+}
+
+
+void brw_emit_unfilled_clip( struct brw_clip_compile *c )
+{
+   struct brw_compile *p = &c->func;
+   struct brw_instruction *do_clip;
+   
+
+   c->need_direction = ((c->key.offset_ccw || c->key.offset_cw) ||
+			(c->key.fill_ccw != c->key.fill_cw) ||
+			c->key.fill_ccw == CLIP_CULL ||
+			c->key.fill_cw == CLIP_CULL ||
+			c->key.copy_bfc_cw ||
+			c->key.copy_bfc_ccw);
+
+   brw_clip_tri_alloc_regs(c, 3 + c->key.nr_userclip + 6);
+   brw_clip_tri_init_vertices(c);
+
+   assert(c->offset[VERT_RESULT_EDGE]);
+
+   if (c->key.fill_ccw == CLIP_CULL &&
+       c->key.fill_cw == CLIP_CULL) {
+      brw_clip_kill_thread(c);
+      return;
+   }
+
+   merge_edgeflags(c);
+
+   /* Need to use the inlist indirection here: 
+    */
+   if (c->need_direction) 
+      compute_tri_direction(c);
+   
+   if (c->key.fill_ccw == CLIP_CULL ||
+       c->key.fill_cw == CLIP_CULL)
+      cull_direction(c);
+
+   if (c->key.offset_ccw ||
+       c->key.offset_cw)
+      compute_offset(c);
+
+   if (c->key.copy_bfc_ccw ||
+       c->key.copy_bfc_cw)
+      copy_bfc(c);
+
+   /* Need to do this whether we clip or not:
+    */
+   if (c->key.do_flat_shading)
+      brw_clip_tri_flat_shade(c);
+   
+   brw_clip_init_clipmask(c);
+   brw_CMP(p, vec1(brw_null_reg()), BRW_CONDITIONAL_NZ, c->reg.planemask, brw_imm_ud(0));
+   do_clip = brw_IF(p, BRW_EXECUTE_1);
+   {
+      brw_clip_init_planes(c);
+      brw_clip_tri(c);
+      check_nr_verts(c);
+   }
+   brw_ENDIF(p, do_clip);
+   
+   emit_unfilled_primitives(c);
+   brw_clip_kill_thread(c);
+}
+
+
+
diff --git a/src/gallium/drivers/i965simple/brw_clip_util.c b/src/gallium/drivers/i965simple/brw_clip_util.c
new file mode 100644
index 0000000000..6d58ceafff
--- /dev/null
+++ b/src/gallium/drivers/i965simple/brw_clip_util.c
@@ -0,0 +1,351 @@
+/*
+ Copyright (C) Intel Corp.  2006.  All Rights Reserved.
+ Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to
+ develop this 3D driver.
+
+ Permission is hereby granted, free of charge, to any person obtaining
+ a copy of this software and associated documentation files (the
+ "Software"), to deal in the Software without restriction, including
+ without limitation the rights to use, copy, modify, merge, publish,
+ distribute, sublicense, and/or sell copies of the Software, and to
+ permit persons to whom the Software is furnished to do so, subject to
+ the following conditions:
+
+ The above copyright notice and this permission notice (including the
+ next paragraph) shall be included in all copies or substantial
+ portions of the Software.
+
+ THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
+ IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
+ LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
+ OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+
+ **********************************************************************/
+ /*
+  * Authors:
+  *   Keith Whitwell <keith@tungstengraphics.com>
+  */
+
+
+#include "brw_defines.h"
+#include "brw_context.h"
+#include "brw_eu.h"
+#include "brw_util.h"
+#include "brw_clip.h"
+
+
+
+
+
+static struct brw_reg get_tmp( struct brw_clip_compile *c )
+{
+   struct brw_reg tmp = brw_vec4_grf(c->last_tmp, 0);
+
+   if (++c->last_tmp > c->prog_data.total_grf)
+      c->prog_data.total_grf = c->last_tmp;
+
+   return tmp;
+}
+
+static void release_tmp( struct brw_clip_compile *c, struct brw_reg tmp )
+{
+   if (tmp.nr == c->last_tmp-1)
+      c->last_tmp--;
+}
+
+
+static struct brw_reg make_plane_ud(unsigned x, unsigned y, unsigned z, unsigned w)
+{
+   return brw_imm_ud((w<<24) | (z<<16) | (y<<8) | x);
+}
+
+
+void brw_clip_init_planes( struct brw_clip_compile *c )
+{
+   struct brw_compile *p = &c->func;
+
+   if (!c->key.nr_userclip) {
+      brw_MOV(p, get_element_ud(c->reg.fixed_planes, 0), make_plane_ud( 0,    0, 0xff, 1));
+      brw_MOV(p, get_element_ud(c->reg.fixed_planes, 1), make_plane_ud( 0,    0,    1, 1));
+      brw_MOV(p, get_element_ud(c->reg.fixed_planes, 2), make_plane_ud( 0, 0xff,    0, 1));
+      brw_MOV(p, get_element_ud(c->reg.fixed_planes, 3), make_plane_ud( 0,    1,    0, 1));
+      brw_MOV(p, get_element_ud(c->reg.fixed_planes, 4), make_plane_ud(0xff,  0,    0, 1));
+      brw_MOV(p, get_element_ud(c->reg.fixed_planes, 5), make_plane_ud( 1,    0,    0, 1));
+   }
+}
+
+
+
+#define W 3
+
+/* Project 'pos' to screen space (or back again), overwrite with results:
+ */
+static void brw_clip_project_position(struct brw_clip_compile *c, struct brw_reg pos )
+{
+   struct brw_compile *p = &c->func;
+
+   /* calc rhw
+    */
+   brw_math_invert(p, get_element(pos, W), get_element(pos, W));
+
+   /* value.xyz *= value.rhw
+    */
+   brw_set_access_mode(p, BRW_ALIGN_16);
+   brw_MUL(p, brw_writemask(pos, TGSI_WRITEMASK_XYZ), pos, brw_swizzle1(pos, W));
+   brw_set_access_mode(p, BRW_ALIGN_1);
+}
+
+
+static void brw_clip_project_vertex( struct brw_clip_compile *c,
+				     struct brw_indirect vert_addr )
+{
+#if 0
+   struct brw_compile *p = &c->func;
+   struct brw_reg tmp = get_tmp(c);
+
+   /* Fixup position.  Extract from the original vertex and re-project
+    * to screen space:
+    */
+   brw_MOV(p, tmp, deref_4f(vert_addr, c->offset[VERT_RESULT_HPOS]));
+   brw_clip_project_position(c, tmp);
+   brw_MOV(p, deref_4f(vert_addr, c->header_position_offset), tmp);
+
+   release_tmp(c, tmp);
+#else
+         #warning "disabled"
+#endif
+}
+
+
+
+
+/* Interpolate between two vertices and put the result into a0.0.
+ * Increment a0.0 accordingly.
+ */
+void brw_clip_interp_vertex( struct brw_clip_compile *c,
+			     struct brw_indirect dest_ptr,
+			     struct brw_indirect v0_ptr, /* from */
+			     struct brw_indirect v1_ptr, /* to */
+			     struct brw_reg t0,
+			     boolean force_edgeflag)
+{
+#if 0
+   struct brw_compile *p = &c->func;
+   struct brw_reg tmp = get_tmp(c);
+   unsigned i;
+
+   /* Just copy the vertex header:
+    */
+   brw_copy_indirect_to_indirect(p, dest_ptr, v0_ptr, 1);
+
+   /* Iterate over each attribute (could be done in pairs?)
+    */
+   for (i = 0; i < c->nr_attrs; i++) {
+      unsigned delta = i*16 + 32;
+
+      if (delta == c->offset[VERT_RESULT_EDGE]) {
+	 if (force_edgeflag)
+	    brw_MOV(p, deref_4f(dest_ptr, delta), brw_imm_f(1));
+	 else
+	    brw_MOV(p, deref_4f(dest_ptr, delta), deref_4f(v0_ptr, delta));
+      }
+      else {
+	 /* Interpolate:
+	  *
+	  *        New = attr0 + t*attr1 - t*attr0
+	  */
+	 brw_MUL(p,
+		 vec4(brw_null_reg()),
+		 deref_4f(v1_ptr, delta),
+		 t0);
+
+	 brw_MAC(p,
+		 tmp,
+		 negate(deref_4f(v0_ptr, delta)),
+		 t0);
+
+	 brw_ADD(p,
+		 deref_4f(dest_ptr, delta),
+		 deref_4f(v0_ptr, delta),
+		 tmp);
+      }
+   }
+
+   if (i & 1) {
+      unsigned delta = i*16 + 32;
+      brw_MOV(p, deref_4f(dest_ptr, delta), brw_imm_f(0));
+   }
+
+   release_tmp(c, tmp);
+
+   /* Recreate the projected (NDC) coordinate in the new vertex
+    * header:
+    */
+   brw_clip_project_vertex(c, dest_ptr );
+#else
+         #warning "disabled"
+#endif
+}
+
+
+
+
+#define MAX_MRF 16
+
+void brw_clip_emit_vue(struct brw_clip_compile *c,
+		       struct brw_indirect vert,
+		       boolean allocate,
+		       boolean eot,
+		       unsigned header)
+{
+   struct brw_compile *p = &c->func;
+   unsigned start = c->last_mrf;
+
+   assert(!(allocate && eot));
+
+   /* Cycle through mrf regs - probably futile as we have to wait for
+    * the allocation response anyway.  Also, the order this function
+    * is invoked doesn't correspond to the order the instructions will
+    * be executed, so it won't have any effect in many cases.
+    */
+#if 0
+   if (start + c->nr_regs + 1 >= MAX_MRF)
+      start = 0;
+
+   c->last_mrf = start + c->nr_regs + 1;
+#endif
+
+   /* Copy the vertex from vertn into m1..mN+1:
+    */
+   brw_copy_from_indirect(p, brw_message_reg(start+1), vert, c->nr_regs);
+
+   /* Overwrite PrimType and PrimStart in the message header, for
+    * each vertex in turn:
+    */
+   brw_MOV(p, get_element_ud(c->reg.R0, 2), brw_imm_ud(header));
+
+
+   /* Send each vertex as a seperate write to the urb.  This
+    * is different to the concept in brw_sf_emit.c, where
+    * subsequent writes are used to build up a single urb
+    * entry.  Each of these writes instantiates a seperate
+    * urb entry - (I think... what about 'allocate'?)
+    */
+   brw_urb_WRITE(p,
+		 allocate ? c->reg.R0 : retype(brw_null_reg(), BRW_REGISTER_TYPE_UD),
+		 start,
+		 c->reg.R0,
+		 allocate,
+		 1,		/* used */
+		 c->nr_regs + 1, /* msg length */
+		 allocate ? 1 : 0, /* response_length */
+		 eot,		/* eot */
+		 1,		/* writes_complete */
+		 0,		/* urb offset */
+		 BRW_URB_SWIZZLE_NONE);
+}
+
+
+
+void brw_clip_kill_thread(struct brw_clip_compile *c)
+{
+   struct brw_compile *p = &c->func;
+
+   /* Send an empty message to kill the thread and release any
+    * allocated urb entry:
+    */
+   brw_urb_WRITE(p,
+		 retype(brw_null_reg(), BRW_REGISTER_TYPE_UD),
+		 0,
+		 c->reg.R0,
+		 0,		/* allocate */
+		 0,		/* used */
+		 0, 		/* msg len */
+		 0, 		/* response len */
+		 1, 		/* eot */
+		 1,		/* writes complete */
+		 0,
+		 BRW_URB_SWIZZLE_NONE);
+}
+
+
+
+
+struct brw_reg brw_clip_plane0_address( struct brw_clip_compile *c )
+{
+   return brw_address(c->reg.fixed_planes);
+}
+
+
+struct brw_reg brw_clip_plane_stride( struct brw_clip_compile *c )
+{
+   if (c->key.nr_userclip) {
+      return brw_imm_uw(16);
+   }
+   else {
+      return brw_imm_uw(4);
+   }
+}
+
+
+/* If flatshading, distribute color from provoking vertex prior to
+ * clipping.
+ */
+void brw_clip_copy_colors( struct brw_clip_compile *c,
+			   unsigned to, unsigned from )
+{
+#if 0
+   struct brw_compile *p = &c->func;
+
+   if (c->offset[VERT_RESULT_COL0])
+      brw_MOV(p,
+	      byte_offset(c->reg.vertex[to], c->offset[VERT_RESULT_COL0]),
+	      byte_offset(c->reg.vertex[from], c->offset[VERT_RESULT_COL0]));
+
+   if (c->offset[VERT_RESULT_COL1])
+      brw_MOV(p,
+	      byte_offset(c->reg.vertex[to], c->offset[VERT_RESULT_COL1]),
+	      byte_offset(c->reg.vertex[from], c->offset[VERT_RESULT_COL1]));
+
+   if (c->offset[VERT_RESULT_BFC0])
+      brw_MOV(p,
+	      byte_offset(c->reg.vertex[to], c->offset[VERT_RESULT_BFC0]),
+	      byte_offset(c->reg.vertex[from], c->offset[VERT_RESULT_BFC0]));
+
+   if (c->offset[VERT_RESULT_BFC1])
+      brw_MOV(p,
+	      byte_offset(c->reg.vertex[to], c->offset[VERT_RESULT_BFC1]),
+	      byte_offset(c->reg.vertex[from], c->offset[VERT_RESULT_BFC1]));
+#else
+         #warning "disabled"
+#endif
+}
+
+
+
+void brw_clip_init_clipmask( struct brw_clip_compile *c )
+{
+   struct brw_compile *p = &c->func;
+   struct brw_reg incoming = get_element_ud(c->reg.R0, 2);
+
+   /* Shift so that lowest outcode bit is rightmost:
+    */
+   brw_SHR(p, c->reg.planemask, incoming, brw_imm_ud(26));
+
+   if (c->key.nr_userclip) {
+      struct brw_reg tmp = retype(vec1(get_tmp(c)), BRW_REGISTER_TYPE_UD);
+
+      /* Rearrange userclip outcodes so that they come directly after
+       * the fixed plane bits.
+       */
+      brw_AND(p, tmp, incoming, brw_imm_ud(0x3f<<14));
+      brw_SHR(p, tmp, tmp, brw_imm_ud(8));
+      brw_OR(p, c->reg.planemask, c->reg.planemask, tmp);
+
+      release_tmp(c, tmp);
+   }
+}
+
diff --git a/src/gallium/drivers/i965simple/brw_context.c b/src/gallium/drivers/i965simple/brw_context.c
new file mode 100644
index 0000000000..5e58701e91
--- /dev/null
+++ b/src/gallium/drivers/i965simple/brw_context.c
@@ -0,0 +1,245 @@
+/*
+ Copyright (C) Intel Corp.  2006.  All Rights Reserved.
+ Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to
+ develop this 3D driver.
+
+ Permission is hereby granted, free of charge, to any person obtaining
+ a copy of this software and associated documentation files (the
+ "Software"), to deal in the Software without restriction, including
+ without limitation the rights to use, copy, modify, merge, publish,
+ distribute, sublicense, and/or sell copies of the Software, and to
+ permit persons to whom the Software is furnished to do so, subject to
+ the following conditions:
+
+ The above copyright notice and this permission notice (including the
+ next paragraph) shall be included in all copies or substantial
+ portions of the Software.
+
+ THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
+ IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
+ LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
+ OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+
+ **********************************************************************/
+ /*
+  * Authors:
+  *   Keith Whitwell <keith@tungstengraphics.com>
+  */
+
+
+#include "brw_context.h"
+#include "brw_defines.h"
+#include "brw_draw.h"
+#include "brw_vs.h"
+#include "brw_tex_layout.h"
+#include "brw_winsys.h"
+
+#include "pipe/p_winsys.h"
+#include "pipe/p_context.h"
+#include "pipe/p_util.h"
+
+/***************************************
+ * Mesa's Driver Functions
+ ***************************************/
+
+#ifndef BRW_DEBUG
+int BRW_DEBUG = (0);
+#endif
+
+static void brw_destroy(struct pipe_context *pipe)
+{
+   struct brw_context *brw = brw_context(pipe);
+
+   FREE(brw);
+}
+
+static void brw_clear(struct pipe_context *pipe, struct pipe_surface *ps,
+                      unsigned clearValue)
+{
+   int x, y, w, h;
+   /* FIXME: corny... */
+
+   x = 0;
+   y = 0;
+   w = ps->width;
+   h = ps->height;
+
+   pipe->surface_fill(pipe, ps, x, y, w, h, clearValue);
+}
+
+
+static int
+brw_get_param(struct pipe_context *pipe, int param)
+{
+   switch (param) {
+   case PIPE_CAP_MAX_TEXTURE_IMAGE_UNITS:
+      return 8;
+   case PIPE_CAP_NPOT_TEXTURES:
+      return 1;
+   case PIPE_CAP_TWO_SIDED_STENCIL:
+      return 1;
+   case PIPE_CAP_GLSL:
+      return 0;
+   case PIPE_CAP_S3TC:
+      return 0;
+   case PIPE_CAP_ANISOTROPIC_FILTER:
+      return 0;
+   case PIPE_CAP_POINT_SPRITE:
+      return 0;
+   case PIPE_CAP_MAX_RENDER_TARGETS:
+      return 1;
+   case PIPE_CAP_OCCLUSION_QUERY:
+      return 0;
+   case PIPE_CAP_TEXTURE_SHADOW_MAP:
+      return 1;
+   case PIPE_CAP_MAX_TEXTURE_2D_LEVELS:
+      return 11; /* max 1024x1024 */
+   case PIPE_CAP_MAX_TEXTURE_3D_LEVELS:
+      return 8;  /* max 128x128x128 */
+   case PIPE_CAP_MAX_TEXTURE_CUBE_LEVELS:
+      return 11; /* max 1024x1024 */
+   default:
+      return 0;
+   }
+}
+
+
+static float
+brw_get_paramf(struct pipe_context *pipe, int param)
+{
+   switch (param) {
+   case PIPE_CAP_MAX_LINE_WIDTH:
+      /* fall-through */
+   case PIPE_CAP_MAX_LINE_WIDTH_AA:
+      return 7.5;
+
+   case PIPE_CAP_MAX_POINT_WIDTH:
+      /* fall-through */
+   case PIPE_CAP_MAX_POINT_WIDTH_AA:
+      return 255.0;
+
+   case PIPE_CAP_MAX_TEXTURE_ANISOTROPY:
+      return 4.0;
+
+   case PIPE_CAP_MAX_TEXTURE_LOD_BIAS:
+      return 16.0;
+
+   default:
+      return 0;
+   }
+}
+
+static boolean
+brw_is_format_supported( struct pipe_context *pipe,
+                          enum pipe_format format, uint type )
+{
+#if 0
+   /* XXX: This is broken -- rewrite if still needed. */
+   static const unsigned tex_supported[] = {
+      PIPE_FORMAT_U_R8_G8_B8_A8,
+      PIPE_FORMAT_U_A8_R8_G8_B8,
+      PIPE_FORMAT_U_R5_G6_B5,
+      PIPE_FORMAT_U_L8,
+      PIPE_FORMAT_U_A8,
+      PIPE_FORMAT_U_I8,
+      PIPE_FORMAT_U_L8_A8,
+      PIPE_FORMAT_YCBCR,
+      PIPE_FORMAT_YCBCR_REV,
+      PIPE_FORMAT_S8_Z24,
+   };
+
+
+   /* Actually a lot more than this - add later:
+    */
+   static const unsigned render_supported[] = {
+      PIPE_FORMAT_U_A8_R8_G8_B8,
+      PIPE_FORMAT_U_R5_G6_B5,
+   };
+
+   /*
+    */
+   static const unsigned z_stencil_supported[] = {
+      PIPE_FORMAT_U_Z16,
+      PIPE_FORMAT_U_Z32,
+      PIPE_FORMAT_S8_Z24,
+   };
+
+   switch (type) {
+   case PIPE_RENDER_FORMAT:
+      *numFormats = Elements(render_supported);
+      return render_supported;
+
+   case PIPE_TEX_FORMAT:
+      *numFormats = Elements(tex_supported);
+      return render_supported;
+
+   case PIPE_Z_STENCIL_FORMAT:
+      *numFormats = Elements(render_supported);
+      return render_supported;
+
+   default:
+      *numFormats = 0;
+      return NULL;
+   }
+#else
+   switch (format) {
+   case PIPE_FORMAT_A8R8G8B8_UNORM:
+   case PIPE_FORMAT_R5G6B5_UNORM:
+   case PIPE_FORMAT_S8Z24_UNORM:
+      return TRUE;
+   default:
+      return FALSE;
+   };
+   return FALSE;
+#endif
+}
+
+
+
+
+struct pipe_context *brw_create(struct pipe_winsys *pipe_winsys,
+                                struct brw_winsys *brw_winsys,
+                                unsigned pci_id)
+{
+   struct brw_context *brw;
+
+   pipe_winsys->printf(pipe_winsys,
+                       "%s: creating brw_context with pci id 0x%x\n",
+                       __FUNCTION__, pci_id);
+
+   brw = CALLOC_STRUCT(brw_context);
+   if (brw == NULL)
+      return NULL;
+
+   brw->winsys = brw_winsys;
+   brw->pipe.winsys = pipe_winsys;
+
+   brw->pipe.destroy = brw_destroy;
+   brw->pipe.is_format_supported = brw_is_format_supported;
+   brw->pipe.get_param = brw_get_param;
+   brw->pipe.get_paramf = brw_get_paramf;
+   brw->pipe.clear = brw_clear;
+   brw->pipe.texture_create  = brw_texture_create;
+   brw->pipe.texture_release = brw_texture_release;
+
+   brw_init_surface_functions(brw);
+   brw_init_state_functions(brw);
+   brw_init_flush_functions(brw);
+   brw_init_string_functions(brw);
+   brw_init_draw_functions( brw );
+
+
+   brw_init_state( brw );
+
+   brw->pci_id = pci_id;
+   brw->dirty = ~0;
+   brw->hardware_dirty = ~0;
+
+   memset(&brw->wm.bind, ~0, sizeof(brw->wm.bind));
+
+   return &brw->pipe;
+}
+
diff --git a/src/gallium/drivers/i965simple/brw_context.h b/src/gallium/drivers/i965simple/brw_context.h
new file mode 100644
index 0000000000..65664d853d
--- /dev/null
+++ b/src/gallium/drivers/i965simple/brw_context.h
@@ -0,0 +1,690 @@
+/*
+ Copyright (C) Intel Corp.  2006.  All Rights Reserved.
+ Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to
+ develop this 3D driver.
+
+ Permission is hereby granted, free of charge, to any person obtaining
+ a copy of this software and associated documentation files (the
+ "Software"), to deal in the Software without restriction, including
+ without limitation the rights to use, copy, modify, merge, publish,
+ distribute, sublicense, and/or sell copies of the Software, and to
+ permit persons to whom the Software is furnished to do so, subject to
+ the following conditions:
+
+ The above copyright notice and this permission notice (including the
+ next paragraph) shall be included in all copies or substantial
+ portions of the Software.
+
+ THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
+ IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
+ LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
+ OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+
+ **********************************************************************/
+ /*
+  * Authors:
+  *   Keith Whitwell <keith@tungstengraphics.com>
+  */
+
+
+#ifndef BRWCONTEXT_INC
+#define BRWCONTEXT_INC
+
+
+#include "pipe/p_context.h"
+#include "pipe/p_defines.h"
+#include "pipe/p_state.h"
+
+#include "brw_structs.h"
+#include "brw_winsys.h"
+
+
+/* Glossary:
+ *
+ * URB - uniform resource buffer.  A mid-sized buffer which is
+ * partitioned between the fixed function units and used for passing
+ * values (vertices, primitives, constants) between them.
+ *
+ * CURBE - constant URB entry.  An urb region (entry) used to hold
+ * constant values which the fixed function units can be instructed to
+ * preload into the GRF when spawining a thread.
+ *
+ * VUE - vertex URB entry.  An urb entry holding a vertex and usually
+ * a vertex header.  The header contains control information and
+ * things like primitive type, Begin/end flags and clip codes.
+ *
+ * PUE - primitive URB entry.  An urb entry produced by the setup (SF)
+ * unit holding rasterization and interpolation parameters.
+ *
+ * GRF - general register file.  One of several register files
+ * addressable by programmed threads.  The inputs (r0, payload, curbe,
+ * urb) of the thread are preloaded to this area before the thread is
+ * spawned.  The registers are individually 8 dwords wide and suitable
+ * for general usage.  Registers holding thread input values are not
+ * special and may be overwritten.
+ *
+ * MRF - message register file.  Threads communicate (and terminate)
+ * by sending messages.  Message parameters are placed in contigous
+ * MRF registers.  All program output is via these messages.  URB
+ * entries are populated by sending a message to the shared URB
+ * function containing the new data, together with a control word,
+ * often an unmodified copy of R0.
+ *
+ * R0 - GRF register 0.  Typically holds control information used when
+ * sending messages to other threads.
+ *
+ * EU or GEN4 EU: The name of the programmable subsystem of the
+ * i965 hardware.  Threads are executed by the EU, the registers
+ * described above are part of the EU architecture.
+ *
+ * Fixed function units:
+ *
+ * CS - Command streamer.  Notional first unit, little software
+ * interaction.  Holds the URB entries used for constant data, ie the
+ * CURBEs.
+ *
+ * VF/VS - Vertex Fetch / Vertex Shader.  The fixed function part of
+ * this unit is responsible for pulling vertices out of vertex buffers
+ * in vram and injecting them into the processing pipe as VUEs.  If
+ * enabled, it first passes them to a VS thread which is a good place
+ * for the driver to implement any active vertex shader.
+ *
+ * GS - Geometry Shader.  This corresponds to a new DX10 concept.  If
+ * enabled, incoming strips etc are passed to GS threads in individual
+ * line/triangle/point units.  The GS thread may perform arbitary
+ * computation and emit whatever primtives with whatever vertices it
+ * chooses.  This makes GS an excellent place to implement GL's
+ * unfilled polygon modes, though of course it is capable of much
+ * more.  Additionally, GS is used to translate away primitives not
+ * handled by latter units, including Quads and Lineloops.
+ *
+ * CS - Clipper.  Mesa's clipping algorithms are imported to run on
+ * this unit.  The fixed function part performs cliptesting against
+ * the 6 fixed clipplanes and makes descisions on whether or not the
+ * incoming primitive needs to be passed to a thread for clipping.
+ * User clip planes are handled via cooperation with the VS thread.
+ *
+ * SF - Strips Fans or Setup: Triangles are prepared for
+ * rasterization.  Interpolation coefficients are calculated.
+ * Flatshading and two-side lighting usually performed here.
+ *
+ * WM - Windower.  Interpolation of vertex attributes performed here.
+ * Fragment shader implemented here.  SIMD aspects of EU taken full
+ * advantage of, as pixels are processed in blocks of 16.
+ *
+ * CC - Color Calculator.  No EU threads associated with this unit.
+ * Handles blending and (presumably) depth and stencil testing.
+ */
+
+#define BRW_MAX_CURBE                    (32*16)
+
+struct brw_context;
+struct brw_winsys;
+
+
+/* Raised when we receive new state across the pipe interface:
+ */
+#define BRW_NEW_VIEWPORT                0x1
+#define BRW_NEW_RASTERIZER              0x2
+#define BRW_NEW_FS                      0x4
+#define BRW_NEW_BLEND                   0x8
+#define BRW_NEW_CLIP                    0x10
+#define BRW_NEW_SCISSOR                 0x20
+#define BRW_NEW_STIPPLE                 0x40
+#define BRW_NEW_FRAMEBUFFER             0x80
+#define BRW_NEW_ALPHA_TEST              0x100
+#define BRW_NEW_DEPTH_STENCIL           0x200
+#define BRW_NEW_SAMPLER                 0x400
+#define BRW_NEW_TEXTURE                 0x800
+#define BRW_NEW_CONSTANTS               0x1000
+#define BRW_NEW_VBO                     0x2000
+#define BRW_NEW_VS                      0x4000
+
+/* Raised for other internal events:
+ */
+#define BRW_NEW_URB_FENCE               0x10000
+#define BRW_NEW_PSP                     0x20000
+#define BRW_NEW_CURBE_OFFSETS           0x40000
+#define BRW_NEW_REDUCED_PRIMITIVE       0x80000
+#define BRW_NEW_PRIMITIVE               0x100000
+#define BRW_NEW_SCENE                 0x200000
+#define BRW_NEW_SF_LINKAGE              0x400000
+
+extern int BRW_DEBUG;
+
+#define DEBUG_TEXTURE	0x1
+#define DEBUG_STATE	0x2
+#define DEBUG_IOCTL	0x4
+#define DEBUG_PRIMS	0x8
+#define DEBUG_VERTS	0x10
+#define DEBUG_FALLBACKS	0x20
+#define DEBUG_VERBOSE	0x40
+#define DEBUG_DRI       0x80
+#define DEBUG_DMA       0x100
+#define DEBUG_SANITY    0x200
+#define DEBUG_SYNC      0x400
+#define DEBUG_SLEEP     0x800
+#define DEBUG_PIXEL     0x1000
+#define DEBUG_STATS     0x2000
+#define DEBUG_TILE      0x4000
+#define DEBUG_SINGLE_THREAD   0x8000
+#define DEBUG_WM        0x10000
+#define DEBUG_URB       0x20000
+#define DEBUG_VS        0x40000
+#define DEBUG_BATCH	0x80000
+#define DEBUG_BUFMGR	0x100000
+#define DEBUG_BLIT	0x200000
+#define DEBUG_REGION	0x400000
+#define DEBUG_MIPTREE	0x800000
+
+#define DBG(...) do {						\
+	if (BRW_DEBUG & FILE_DEBUG_FLAG)			\
+	       brw->pipe.winsys->printf(brw->pipe.winsys, __VA_ARGS__);			\
+} while(0)
+
+#define PRINT(...) do {						\
+	       brw->pipe.winsys->printf(brw->pipe.winsys, __VA_ARGS__);			\
+} while(0)
+
+struct brw_state_flags {
+   unsigned cache;
+   unsigned brw;
+};
+
+
+struct brw_shader_info {
+   int nr_regs[8];		/* TGSI_FILE_* */
+};
+   
+
+
+struct brw_vertex_program {
+   struct pipe_shader_state program;
+   struct brw_shader_info info;
+   int id;
+};
+
+
+
+struct brw_fragment_program {
+   struct pipe_shader_state program;
+   struct brw_shader_info info;
+   
+   boolean UsesDepth;
+   boolean UsesKill;
+   boolean ComputesDepth;
+   int id;
+};
+
+
+
+
+struct pipe_setup_linkage {
+   struct {
+      unsigned vp_output:5;
+      unsigned interp_mode:4;
+      unsigned bf_vp_output:5;
+   } fp_input[PIPE_MAX_SHADER_INPUTS];
+
+   unsigned fp_input_count:5;
+   unsigned max_vp_output:5;
+};
+   
+
+
+struct brw_texture {
+   struct pipe_texture base;
+
+   /* Derived from the above:
+    */
+   unsigned pitch;
+   unsigned depth_pitch;          /* per-image on i945? */
+   unsigned total_height;
+
+   unsigned nr_images[PIPE_MAX_TEXTURE_LEVELS];
+
+   /* Explicitly store the offset of each image for each cube face or
+    * depth value.  Pretty much have to accept that hardware formats
+    * are going to be so diverse that there is no unified way to
+    * compute the offsets of depth/cube images within a mipmap level,
+    * so have to store them as a lookup table:
+    */
+   unsigned *image_offset[PIPE_MAX_TEXTURE_LEVELS];   /**< array [depth] of offsets */
+
+   /* Includes image offset tables:
+    */
+   unsigned level_offset[PIPE_MAX_TEXTURE_LEVELS];
+
+   /* The data is held here:
+    */
+   struct pipe_buffer *buffer;
+};
+
+/* Data about a particular attempt to compile a program.  Note that
+ * there can be many of these, each in a different GL state
+ * corresponding to a different brw_wm_prog_key struct, with different
+ * compiled programs:
+ */
+/* Data about a particular attempt to compile a program.  Note that
+ * there can be many of these, each in a different GL state
+ * corresponding to a different brw_wm_prog_key struct, with different
+ * compiled programs:
+ */
+
+struct brw_wm_prog_data {
+   unsigned curb_read_length;
+   unsigned urb_read_length;
+
+   unsigned first_curbe_grf;
+   unsigned total_grf;
+   unsigned total_scratch;
+
+   /* Internally generated constants for the CURBE.  These are loaded
+    * ahead of the data from the constant buffer.
+    */
+   const float internal_const[8];
+   unsigned nr_internal_consts;
+   unsigned max_const;
+
+   boolean error;
+};
+
+struct brw_sf_prog_data {
+   unsigned urb_read_length;
+   unsigned total_grf;
+
+   /* Each vertex may have upto 12 attributes, 4 components each,
+    * except WPOS which requires only 2.  (11*4 + 2) == 44 ==> 11
+    * rows.
+    *
+    * Actually we use 4 for each, so call it 12 rows.
+    */
+   unsigned urb_entry_size;
+};
+
+struct brw_clip_prog_data {
+   unsigned curb_read_length;	/* user planes? */
+   unsigned clip_mode;
+   unsigned urb_read_length;
+   unsigned total_grf;
+};
+
+struct brw_gs_prog_data {
+   unsigned urb_read_length;
+   unsigned total_grf;
+};
+
+struct brw_vs_prog_data {
+   unsigned curb_read_length;
+   unsigned urb_read_length;
+   unsigned total_grf;
+   unsigned outputs_written;
+
+   unsigned inputs_read;
+
+   unsigned max_const;
+
+   float    imm_buf[PIPE_MAX_CONSTANT][4];
+   unsigned num_imm;
+   unsigned num_consts;
+
+   /* Used for calculating urb partitions:
+    */
+   unsigned urb_entry_size;
+};
+
+
+#define BRW_MAX_TEX_UNIT 8
+#define BRW_WM_MAX_SURF BRW_MAX_TEX_UNIT + 1
+
+/* Create a fixed sized struct for caching binding tables:
+ */
+struct brw_surface_binding_table {
+   unsigned surf_ss_offset[BRW_WM_MAX_SURF];
+};
+
+
+struct brw_cache;
+
+struct brw_mem_pool {
+   struct pipe_buffer *buffer;
+
+   unsigned size;
+   unsigned offset;		/* offset of first free byte */
+
+   struct brw_context *brw;
+};
+
+struct brw_cache_item {
+   unsigned hash;
+   unsigned key_size;		/* for variable-sized keys */
+   const void *key;
+
+   unsigned offset;		/* offset within pool's buffer */
+   unsigned data_size;
+
+   struct brw_cache_item *next;
+};
+
+
+
+struct brw_cache {
+   unsigned id;
+
+   const char *name;
+
+   struct brw_context *brw;
+   struct brw_mem_pool *pool;
+
+   struct brw_cache_item **items;
+   unsigned size, n_items;
+
+   unsigned key_size;		/* for fixed-size keys */
+   unsigned aux_size;
+
+   unsigned last_addr;			/* offset of active item */
+};
+
+
+
+
+/* Considered adding a member to this struct to document which flags
+ * an update might raise so that ordering of the state atoms can be
+ * checked or derived at runtime.  Dropped the idea in favor of having
+ * a debug mode where the state is monitored for flags which are
+ * raised that have already been tested against.
+ */
+struct brw_tracked_state {
+   struct brw_state_flags dirty;
+   void (*update)( struct brw_context *brw );
+};
+
+
+/* Flags for brw->state.cache.
+ */
+#define CACHE_NEW_CC_VP                  (1<<BRW_CC_VP)
+#define CACHE_NEW_CC_UNIT                (1<<BRW_CC_UNIT)
+#define CACHE_NEW_WM_PROG                (1<<BRW_WM_PROG)
+#define CACHE_NEW_SAMPLER_DEFAULT_COLOR  (1<<BRW_SAMPLER_DEFAULT_COLOR)
+#define CACHE_NEW_SAMPLER                (1<<BRW_SAMPLER)
+#define CACHE_NEW_WM_UNIT                (1<<BRW_WM_UNIT)
+#define CACHE_NEW_SF_PROG                (1<<BRW_SF_PROG)
+#define CACHE_NEW_SF_VP                  (1<<BRW_SF_VP)
+#define CACHE_NEW_SF_UNIT                (1<<BRW_SF_UNIT)
+#define CACHE_NEW_VS_UNIT                (1<<BRW_VS_UNIT)
+#define CACHE_NEW_VS_PROG                (1<<BRW_VS_PROG)
+#define CACHE_NEW_GS_UNIT                (1<<BRW_GS_UNIT)
+#define CACHE_NEW_GS_PROG                (1<<BRW_GS_PROG)
+#define CACHE_NEW_CLIP_VP                (1<<BRW_CLIP_VP)
+#define CACHE_NEW_CLIP_UNIT              (1<<BRW_CLIP_UNIT)
+#define CACHE_NEW_CLIP_PROG              (1<<BRW_CLIP_PROG)
+#define CACHE_NEW_SURFACE                (1<<BRW_SS_SURFACE)
+#define CACHE_NEW_SURF_BIND              (1<<BRW_SS_SURF_BIND)
+
+
+
+
+enum brw_mempool_id {
+   BRW_GS_POOL,
+   BRW_SS_POOL,
+   BRW_MAX_POOL
+};
+
+
+struct brw_cached_batch_item {
+   struct header *header;
+   unsigned sz;
+   struct brw_cached_batch_item *next;
+};
+
+
+
+/* Protect against a future where PIPE_ATTRIB_MAX > 32.  Wouldn't life
+ * be easier if C allowed arrays of packed elements?
+ */
+#define ATTRIB_BIT_DWORDS  ((PIPE_ATTRIB_MAX+31)/32)
+
+
+
+
+struct brw_vertex_info {
+   unsigned varying;  /* varying:1[PIPE_ATTRIB_MAX] */
+   unsigned sizes[ATTRIB_BIT_DWORDS * 2]; /* sizes:2[PIPE_ATTRIB_MAX] */
+};
+
+
+
+
+
+struct brw_context
+{
+   struct pipe_context pipe;
+   struct brw_winsys *winsys;
+
+   unsigned primitive;
+   unsigned reduced_primitive;
+
+   boolean emit_state_always;
+
+   struct {
+      struct brw_state_flags dirty;
+   } state;
+
+
+   struct {
+      const struct pipe_blend_state         *Blend;
+      const struct pipe_depth_stencil_alpha_state *DepthStencil;
+      const struct pipe_poly_stipple        *PolygonStipple;
+      const struct pipe_rasterizer_state    *Raster;
+      const struct pipe_sampler_state       *Samplers[PIPE_MAX_SAMPLERS];
+      const struct brw_vertex_program       *VertexProgram;
+      const struct brw_fragment_program     *FragmentProgram;
+
+      struct pipe_clip_state          Clip;
+      struct pipe_blend_color         BlendColor;
+      struct pipe_scissor_state       Scissor;
+      struct pipe_viewport_state      Viewport;
+      struct pipe_framebuffer_state   FrameBuffer;
+
+      const struct pipe_constant_buffer *Constants[2];
+      const struct brw_texture          *Texture[PIPE_MAX_SAMPLERS];
+   } attribs;
+
+   struct brw_mem_pool pool[BRW_MAX_POOL];
+   struct brw_cache cache[BRW_MAX_CACHE];
+   struct brw_cached_batch_item *cached_batch_items;
+
+   struct {
+
+      /* Arrays with buffer objects to copy non-bufferobj arrays into
+       * for upload:
+       */
+      struct pipe_vertex_buffer *vbo_array[PIPE_ATTRIB_MAX];
+
+      struct brw_vertex_element_state inputs[PIPE_ATTRIB_MAX];
+
+#define BRW_NR_UPLOAD_BUFS 17
+#define BRW_UPLOAD_INIT_SIZE (128*1024)
+
+      /* Summary of size and varying of active arrays, so we can check
+       * for changes to this state:
+       */
+      struct brw_vertex_info info;
+   } vb;
+
+
+   unsigned hardware_dirty;
+   unsigned dirty;
+   unsigned pci_id;
+   /* BRW_NEW_URB_ALLOCATIONS:
+    */
+   struct {
+      unsigned vsize;		/* vertex size plus header in urb registers */
+      unsigned csize;		/* constant buffer size in urb registers */
+      unsigned sfsize;		/* setup data size in urb registers */
+
+      boolean constrained;
+
+      unsigned nr_vs_entries;
+      unsigned nr_gs_entries;
+      unsigned nr_clip_entries;
+      unsigned nr_sf_entries;
+      unsigned nr_cs_entries;
+
+/*       unsigned vs_size; */
+/*       unsigned gs_size; */
+/*       unsigned clip_size; */
+/*       unsigned sf_size; */
+/*       unsigned cs_size; */
+
+      unsigned vs_start;
+      unsigned gs_start;
+      unsigned clip_start;
+      unsigned sf_start;
+      unsigned cs_start;
+   } urb;
+
+
+   /* BRW_NEW_CURBE_OFFSETS:
+    */
+   struct {
+      unsigned wm_start;
+      unsigned wm_size;
+      unsigned clip_start;
+      unsigned clip_size;
+      unsigned vs_start;
+      unsigned vs_size;
+      unsigned total_size;
+
+      unsigned gs_offset;
+
+      float *last_buf;
+      unsigned last_bufsz;
+   } curbe;
+
+   struct {
+      struct brw_vs_prog_data *prog_data;
+
+      unsigned prog_gs_offset;
+      unsigned state_gs_offset;
+   } vs;
+
+   struct {
+      struct brw_gs_prog_data *prog_data;
+
+      boolean prog_active;
+      unsigned prog_gs_offset;
+      unsigned state_gs_offset;
+   } gs;
+
+   struct {
+      struct brw_clip_prog_data *prog_data;
+
+      unsigned prog_gs_offset;
+      unsigned vp_gs_offset;
+      unsigned state_gs_offset;
+   } clip;
+
+
+   struct {
+      struct brw_sf_prog_data *prog_data;
+
+      struct pipe_setup_linkage linkage;
+
+      unsigned prog_gs_offset;
+      unsigned vp_gs_offset;
+      unsigned state_gs_offset;
+   } sf;
+
+   struct {
+      struct brw_wm_prog_data *prog_data;
+
+//      struct brw_wm_compiler *compile_data;
+
+
+      /**
+       * Array of sampler state uploaded at sampler_gs_offset of BRW_SAMPLER
+       * cache
+       */
+      struct brw_sampler_state sampler[BRW_MAX_TEX_UNIT];
+
+      unsigned render_surf;
+      unsigned nr_surfaces;
+
+      unsigned max_threads;
+      struct pipe_buffer *scratch_buffer;
+      unsigned scratch_buffer_size;
+
+      unsigned sampler_count;
+      unsigned sampler_gs_offset;
+
+      struct brw_surface_binding_table bind;
+      unsigned bind_ss_offset;
+
+      unsigned prog_gs_offset;
+      unsigned state_gs_offset;
+   } wm;
+
+
+   struct {
+      unsigned vp_gs_offset;
+      unsigned state_gs_offset;
+   } cc;
+
+
+   /* Used to give every program string a unique id
+    */
+   unsigned program_id;
+};
+
+
+#define BRW_PACKCOLOR8888(r,g,b,a)  ((r<<24) | (g<<16) | (b<<8) | a)
+
+
+/*======================================================================
+ * brw_vtbl.c
+ */
+void brw_do_flush( struct brw_context *brw,
+		   unsigned flags );
+
+
+/*======================================================================
+ * brw_state.c
+ */
+void brw_validate_state(struct brw_context *brw);
+void brw_init_state(struct brw_context *brw);
+void brw_destroy_state(struct brw_context *brw);
+
+
+/*======================================================================
+ * brw_tex.c
+ */
+void brwUpdateTextureState( struct brw_context *brw );
+
+
+/* brw_urb.c
+ */
+void brw_upload_urb_fence(struct brw_context *brw);
+
+void brw_upload_constant_buffer_state(struct brw_context *brw);
+
+void brw_init_surface_functions(struct brw_context *brw);
+void brw_init_state_functions(struct brw_context *brw);
+void brw_init_flush_functions(struct brw_context *brw);
+void brw_init_string_functions(struct brw_context *brw);
+
+/*======================================================================
+ * Inline conversion functions.  These are better-typed than the
+ * macros used previously:
+ */
+static inline struct brw_context *
+brw_context( struct pipe_context *ctx )
+{
+   return (struct brw_context *)ctx;
+}
+
+#endif
+
diff --git a/src/gallium/drivers/i965simple/brw_curbe.c b/src/gallium/drivers/i965simple/brw_curbe.c
new file mode 100644
index 0000000000..52bbd525c1
--- /dev/null
+++ b/src/gallium/drivers/i965simple/brw_curbe.c
@@ -0,0 +1,368 @@
+/*
+ Copyright (C) Intel Corp.  2006.  All Rights Reserved.
+ Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to
+ develop this 3D driver.
+
+ Permission is hereby granted, free of charge, to any person obtaining
+ a copy of this software and associated documentation files (the
+ "Software"), to deal in the Software without restriction, including
+ without limitation the rights to use, copy, modify, merge, publish,
+ distribute, sublicense, and/or sell copies of the Software, and to
+ permit persons to whom the Software is furnished to do so, subject to
+ the following conditions:
+
+ The above copyright notice and this permission notice (including the
+ next paragraph) shall be included in all copies or substantial
+ portions of the Software.
+
+ THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
+ IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
+ LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
+ OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+
+ **********************************************************************/
+ /*
+  * Authors:
+  *   Keith Whitwell <keith@tungstengraphics.com>
+  */
+
+
+
+#include "brw_context.h"
+#include "brw_defines.h"
+#include "brw_state.h"
+#include "brw_batch.h"
+#include "brw_util.h"
+#include "brw_wm.h"
+#include "pipe/p_state.h"
+#include "pipe/p_winsys.h"
+#include "pipe/p_util.h"
+
+#define FILE_DEBUG_FLAG DEBUG_FALLBACKS
+
+/* Partition the CURBE between the various users of constant values:
+ */
+static void calculate_curbe_offsets( struct brw_context *brw )
+{
+   /* CACHE_NEW_WM_PROG */
+   unsigned nr_fp_regs = align(brw->wm.prog_data->max_const, 16);
+
+   /* BRW_NEW_VERTEX_PROGRAM */
+   unsigned nr_vp_regs = align(brw->vs.prog_data->max_const, 16);
+   unsigned nr_clip_regs = 0;
+   unsigned total_regs;
+
+#if 0
+   /* BRW_NEW_CLIP ? */
+   if (brw->attribs.Transform->ClipPlanesEnabled) {
+      unsigned nr_planes = 6 + brw_count_bits(brw->attribs.Transform->ClipPlanesEnabled);
+      nr_clip_regs = align(nr_planes * 4, 16);
+   }
+#endif
+
+
+   total_regs = nr_fp_regs + nr_vp_regs + nr_clip_regs;
+
+   /* This can happen - what to do?  Probably rather than falling
+    * back, the best thing to do is emit programs which code the
+    * constants as immediate values.  Could do this either as a static
+    * cap on WM and VS, or adaptively.
+    *
+    * Unfortunately, this is currently dependent on the results of the
+    * program generation process (in the case of wm), so this would
+    * introduce the need to re-generate programs in the event of a
+    * curbe allocation failure.
+    */
+   /* Max size is 32 - just large enough to
+    * hold the 128 parameters allowed by
+    * the fragment and vertex program
+    * api's.  It's not clear what happens
+    * when both VP and FP want to use 128
+    * parameters, though.
+    */
+   assert(total_regs <= 32);
+
+   /* Lazy resize:
+    */
+   if (nr_fp_regs > brw->curbe.wm_size ||
+       nr_vp_regs > brw->curbe.vs_size ||
+       nr_clip_regs != brw->curbe.clip_size ||
+       (total_regs < brw->curbe.total_size / 4 &&
+	brw->curbe.total_size > 16)) {
+
+      unsigned reg = 0;
+
+      /* Calculate a new layout:
+       */
+      reg = 0;
+      brw->curbe.wm_start = reg;
+      brw->curbe.wm_size = nr_fp_regs; reg += nr_fp_regs;
+      brw->curbe.clip_start = reg;
+      brw->curbe.clip_size = nr_clip_regs; reg += nr_clip_regs;
+      brw->curbe.vs_start = reg;
+      brw->curbe.vs_size = nr_vp_regs; reg += nr_vp_regs;
+      brw->curbe.total_size = reg;
+
+#if 0
+      if (0)
+	 DBG("curbe wm %d+%d clip %d+%d vs %d+%d\n",
+		      brw->curbe.wm_start,
+		      brw->curbe.wm_size,
+		      brw->curbe.clip_start,
+		      brw->curbe.clip_size,
+		      brw->curbe.vs_start,
+		      brw->curbe.vs_size );
+#endif
+
+      brw->state.dirty.brw |= BRW_NEW_CURBE_OFFSETS;
+   }
+}
+
+
+const struct brw_tracked_state brw_curbe_offsets = {
+   .dirty = {
+      .brw  = (BRW_NEW_CLIP |
+	       BRW_NEW_VS),
+      .cache = CACHE_NEW_WM_PROG
+   },
+   .update = calculate_curbe_offsets
+};
+
+
+
+/* Define the number of curbes within CS's urb allocation.  Multiple
+ * urb entries -> multiple curbes.  These will be used by
+ * fixed-function hardware in a double-buffering scheme to avoid a
+ * pipeline stall each time the contents of the curbe is changed.
+ */
+void brw_upload_constant_buffer_state(struct brw_context *brw)
+{
+   struct brw_constant_buffer_state cbs;
+   memset(&cbs, 0, sizeof(cbs));
+
+   /* It appears that this is the state packet for the CS unit, ie. the
+    * urb entries detailed here are housed in the CS range from the
+    * URB_FENCE command.
+    */
+   cbs.header.opcode = CMD_CONST_BUFFER_STATE;
+   cbs.header.length = sizeof(cbs)/4 - 2;
+
+   /* BRW_NEW_URB_FENCE */
+   cbs.bits0.nr_urb_entries = brw->urb.nr_cs_entries;
+   cbs.bits0.urb_entry_size = brw->urb.csize - 1;
+
+   assert(brw->urb.nr_cs_entries);
+   BRW_CACHED_BATCH_STRUCT(brw, &cbs);
+}
+
+
+static float fixed_plane[6][4] = {
+   { 0,    0,   -1, 1 },
+   { 0,    0,    1, 1 },
+   { 0,   -1,    0, 1 },
+   { 0,    1,    0, 1 },
+   {-1,    0,    0, 1 },
+   { 1,    0,    0, 1 }
+};
+
+/* Upload a new set of constants.  Too much variability to go into the
+ * cache mechanism, but maybe would benefit from a comparison against
+ * the current uploaded set of constants.
+ */
+static void upload_constant_buffer(struct brw_context *brw)
+{
+   struct brw_mem_pool *pool = &brw->pool[BRW_GS_POOL];
+   unsigned sz = brw->curbe.total_size;
+   unsigned bufsz = sz * sizeof(float);
+   float *buf;
+   unsigned i;
+
+
+   if (sz == 0) {
+      struct brw_constant_buffer cb;
+      cb.header.opcode = CMD_CONST_BUFFER;
+      cb.header.length = sizeof(cb)/4 - 2;
+      cb.header.valid = 0;
+      cb.bits0.buffer_length = 0;
+      cb.bits0.buffer_address = 0;
+      BRW_BATCH_STRUCT(brw, &cb);
+
+      if (brw->curbe.last_buf) {
+	 free(brw->curbe.last_buf);
+	 brw->curbe.last_buf = NULL;
+	 brw->curbe.last_bufsz  = 0;
+      }
+
+      return;
+   }
+
+   buf = (float *)malloc(bufsz);
+
+   memset(buf, 0, bufsz);
+
+   if (brw->curbe.wm_size) {
+      unsigned offset = brw->curbe.wm_start * 16;
+
+      /* First the constant buffer constants:
+       */
+      
+      /* Then any internally generated constants: 
+       */
+      for (i = 0; i < brw->wm.prog_data->nr_internal_consts; i++)
+	 buf[offset + i] = brw->wm.prog_data->internal_const[i];
+
+      assert(brw->wm.prog_data->max_const == 
+	     brw->wm.prog_data->nr_internal_consts);
+   }
+
+
+   /* The clipplanes are actually delivered to both CLIP and VS units.
+    * VS uses them to calculate the outcode bitmasks.
+    */
+   if (brw->curbe.clip_size) {
+      unsigned offset = brw->curbe.clip_start * 16;
+      unsigned j;
+
+      /* If any planes are going this way, send them all this way:
+       */
+      for (i = 0; i < 6; i++) {
+	 buf[offset + i * 4 + 0] = fixed_plane[i][0];
+	 buf[offset + i * 4 + 1] = fixed_plane[i][1];
+	 buf[offset + i * 4 + 2] = fixed_plane[i][2];
+	 buf[offset + i * 4 + 3] = fixed_plane[i][3];
+      }
+
+      /* Clip planes: BRW_NEW_CLIP:
+       */
+      for (j = 0; j < brw->attribs.Clip.nr; j++) {
+	 buf[offset + i * 4 + 0] = brw->attribs.Clip.ucp[j][0];
+	 buf[offset + i * 4 + 1] = brw->attribs.Clip.ucp[j][1];
+	 buf[offset + i * 4 + 2] = brw->attribs.Clip.ucp[j][2];
+	 buf[offset + i * 4 + 3] = brw->attribs.Clip.ucp[j][3];
+	 i++;
+      }
+   }
+
+
+   if (brw->curbe.vs_size) {
+      unsigned offset = brw->curbe.vs_start * 16;
+      /*unsigned nr = vp->max_const;*/
+      const struct pipe_constant_buffer *cbuffer = brw->attribs.Constants[0];
+      struct pipe_winsys *ws = brw->pipe.winsys;
+      /* FIXME: buffer size is num_consts + num_immediates */
+      if (brw->vs.prog_data->num_consts) {
+         /* map the vertex constant buffer and copy to curbe: */
+         void *data = ws->buffer_map(ws, cbuffer->buffer, 0);
+         /* FIXME: this is wrong. the cbuffer->size currently
+          * represents size of consts + immediates. so if we'll
+          * have both we'll copy over the end of the buffer
+          * with the subsequent memcpy */
+         memcpy(&buf[offset], data, cbuffer->size);
+         ws->buffer_unmap(ws, cbuffer->buffer);
+         offset += cbuffer->size;
+      }
+      /*immediates*/
+      if (brw->vs.prog_data->num_imm) {
+         memcpy(&buf[offset], brw->vs.prog_data->imm_buf,
+                brw->vs.prog_data->num_imm * 4 * sizeof(float));
+      }
+   }
+
+   if (1) {
+      for (i = 0; i < sz; i+=4)
+	 debug_printf("curbe %d.%d: %f %f %f %f\n", i/8, i&4,
+		      buf[i+0], buf[i+1], buf[i+2], buf[i+3]);
+
+      debug_printf("last_buf %p buf %p sz %d/%d cmp %d\n",
+		   brw->curbe.last_buf, buf,
+		   bufsz, brw->curbe.last_bufsz,
+		   brw->curbe.last_buf ? memcmp(buf, brw->curbe.last_buf, bufsz) : -1);
+   }
+
+   if (brw->curbe.last_buf &&
+       bufsz == brw->curbe.last_bufsz &&
+       memcmp(buf, brw->curbe.last_buf, bufsz) == 0) {
+      free(buf);
+/*       return; */
+   }
+   else {
+      if (brw->curbe.last_buf)
+	 free(brw->curbe.last_buf);
+      brw->curbe.last_buf = buf;
+      brw->curbe.last_bufsz = bufsz;
+
+
+      if (!brw_pool_alloc(pool,
+			  bufsz,
+			  1 << 6,
+			  &brw->curbe.gs_offset)) {
+	 debug_printf("out of GS memory for curbe\n");
+	 assert(0);
+	 return;
+      }
+
+
+      /* Copy data to the buffer:
+       */
+      brw->winsys->buffer_subdata_typed(brw->winsys,
+					pool->buffer, 
+					brw->curbe.gs_offset, 
+					bufsz, 
+					buf,
+					BRW_CONSTANT_BUFFER );
+   }
+
+   /* TODO: only emit the constant_buffer packet when necessary, ie:
+      - contents have changed
+      - offset has changed
+      - hw requirements due to other packets emitted.
+   */
+   {
+      struct brw_constant_buffer cb;
+
+      memset(&cb, 0, sizeof(cb));
+
+      cb.header.opcode = CMD_CONST_BUFFER;
+      cb.header.length = sizeof(cb)/4 - 2;
+      cb.header.valid = 1;
+      cb.bits0.buffer_length = sz - 1;
+      cb.bits0.buffer_address = brw->curbe.gs_offset >> 6;
+
+      /* Because this provokes an action (ie copy the constants into the
+       * URB), it shouldn't be shortcircuited if identical to the
+       * previous time - because eg. the urb destination may have
+       * changed, or the urb contents different to last time.
+       *
+       * Note that the data referred to is actually copied internally,
+       * not just used in place according to passed pointer.
+       *
+       * It appears that the CS unit takes care of using each available
+       * URB entry (Const URB Entry == CURBE) in turn, and issuing
+       * flushes as necessary when doublebuffering of CURBEs isn't
+       * possible.
+       */
+      BRW_BATCH_STRUCT(brw, &cb);
+   }
+}
+
+/* This tracked state is unique in that the state it monitors varies
+ * dynamically depending on the parameters tracked by the fragment and
+ * vertex programs.  This is the template used as a starting point,
+ * each context will maintain a copy of this internally and update as
+ * required.
+ */
+const struct brw_tracked_state brw_constant_buffer = {
+   .dirty = {
+      .brw  = (BRW_NEW_CLIP |
+	       BRW_NEW_CONSTANTS |
+	       BRW_NEW_URB_FENCE | /* Implicit - hardware requires this, not used above */
+	       BRW_NEW_PSP | /* Implicit - hardware requires this, not used above */
+	       BRW_NEW_CURBE_OFFSETS),
+      .cache = (CACHE_NEW_WM_PROG)
+   },
+   .update = upload_constant_buffer
+};
+
diff --git a/src/gallium/drivers/i965simple/brw_defines.h b/src/gallium/drivers/i965simple/brw_defines.h
new file mode 100644
index 0000000000..9379a397f6
--- /dev/null
+++ b/src/gallium/drivers/i965simple/brw_defines.h
@@ -0,0 +1,852 @@
+/*
+ Copyright (C) Intel Corp.  2006.  All Rights Reserved.
+ Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to
+ develop this 3D driver.
+
+ Permission is hereby granted, free of charge, to any person obtaining
+ a copy of this software and associated documentation files (the
+ "Software"), to deal in the Software without restriction, including
+ without limitation the rights to use, copy, modify, merge, publish,
+ distribute, sublicense, and/or sell copies of the Software, and to
+ permit persons to whom the Software is furnished to do so, subject to
+ the following conditions:
+
+ The above copyright notice and this permission notice (including the
+ next paragraph) shall be included in all copies or substantial
+ portions of the Software.
+
+ THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
+ IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
+ LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
+ OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+
+ **********************************************************************/
+ /*
+  * Authors:
+  *   Keith Whitwell <keith@tungstengraphics.com>
+  */
+
+
+#ifndef BRW_DEFINES_H
+#define BRW_DEFINES_H
+
+/*
+ */
+#define MI_NOOP                              0x00
+#define MI_USER_INTERRUPT                    0x02
+#define MI_WAIT_FOR_EVENT                    0x03
+#define MI_FLUSH                             0x04
+#define MI_REPORT_HEAD                       0x07
+#define MI_ARB_ON_OFF                        0x08
+#define MI_BATCH_BUFFER_END                  0x0A
+#define MI_OVERLAY_FLIP                      0x11
+#define MI_LOAD_SCAN_LINES_INCL              0x12
+#define MI_LOAD_SCAN_LINES_EXCL              0x13
+#define MI_DISPLAY_BUFFER_INFO               0x14
+#define MI_SET_CONTEXT                       0x18
+#define MI_STORE_DATA_IMM                    0x20
+#define MI_STORE_DATA_INDEX                  0x21
+#define MI_LOAD_REGISTER_IMM                 0x22
+#define MI_STORE_REGISTER_MEM                0x24
+#define MI_BATCH_BUFFER_START                0x31
+
+#define MI_SYNCHRONOUS_FLIP                  0x0
+#define MI_ASYNCHRONOUS_FLIP                 0x1
+
+#define MI_BUFFER_SECURE                     0x0
+#define MI_BUFFER_NONSECURE                  0x1
+
+#define MI_ARBITRATE_AT_CHAIN_POINTS         0x0
+#define MI_ARBITRATE_BETWEEN_INSTS           0x1
+#define MI_NO_ARBITRATION                    0x3
+
+#define MI_CONDITION_CODE_WAIT_DISABLED      0x0
+#define MI_CONDITION_CODE_WAIT_0             0x1
+#define MI_CONDITION_CODE_WAIT_1             0x2
+#define MI_CONDITION_CODE_WAIT_2             0x3
+#define MI_CONDITION_CODE_WAIT_3             0x4
+#define MI_CONDITION_CODE_WAIT_4             0x5
+
+#define MI_DISPLAY_PIPE_A                    0x0
+#define MI_DISPLAY_PIPE_B                    0x1
+
+#define MI_DISPLAY_PLANE_A                   0x0
+#define MI_DISPLAY_PLANE_B                   0x1
+#define MI_DISPLAY_PLANE_C                   0x2
+
+#define MI_STANDARD_FLIP                                 0x0
+#define MI_ENQUEUE_FLIP_PERFORM_BASE_FRAME_NUMBER_LOAD   0x1
+#define MI_ENQUEUE_FLIP_TARGET_FRAME_NUMBER_RELATIVE     0x2
+#define MI_ENQUEUE_FLIP_ABSOLUTE_TARGET_FRAME_NUMBER     0x3
+
+#define MI_PHYSICAL_ADDRESS                  0x0
+#define MI_VIRTUAL_ADDRESS                   0x1
+
+#define MI_BUFFER_MEMORY_MAIN                0x0
+#define MI_BUFFER_MEMORY_GTT                 0x2
+#define MI_BUFFER_MEMORY_PER_PROCESS_GTT     0x3
+
+#define MI_FLIP_CONTINUE                     0x0
+#define MI_FLIP_ON                           0x1
+#define MI_FLIP_OFF                          0x2
+
+#define MI_UNTRUSTED_REGISTER_SPACE          0x0
+#define MI_TRUSTED_REGISTER_SPACE            0x1
+
+/* 3D state:
+ */
+#define _3DOP_3DSTATE_PIPELINED       0x0
+#define _3DOP_3DSTATE_NONPIPELINED    0x1
+#define _3DOP_3DCONTROL               0x2
+#define _3DOP_3DPRIMITIVE             0x3
+
+#define _3DSTATE_PIPELINED_POINTERS       0x00
+#define _3DSTATE_BINDING_TABLE_POINTERS   0x01
+#define _3DSTATE_VERTEX_BUFFERS           0x08
+#define _3DSTATE_VERTEX_ELEMENTS          0x09
+#define _3DSTATE_INDEX_BUFFER             0x0A
+#define _3DSTATE_VF_STATISTICS            0x0B
+#define _3DSTATE_DRAWING_RECTANGLE            0x00
+#define _3DSTATE_CONSTANT_COLOR               0x01
+#define _3DSTATE_SAMPLER_PALETTE_LOAD         0x02
+#define _3DSTATE_CHROMA_KEY                   0x04
+#define _3DSTATE_DEPTH_BUFFER                 0x05
+#define _3DSTATE_POLY_STIPPLE_OFFSET          0x06
+#define _3DSTATE_POLY_STIPPLE_PATTERN         0x07
+#define _3DSTATE_LINE_STIPPLE                 0x08
+#define _3DSTATE_GLOBAL_DEPTH_OFFSET_CLAMP    0x09
+#define _3DCONTROL    0x00
+#define _3DPRIMITIVE  0x00
+
+#define PIPE_CONTROL_NOWRITE          0x00
+#define PIPE_CONTROL_WRITEIMMEDIATE   0x01
+#define PIPE_CONTROL_WRITEDEPTH       0x02
+#define PIPE_CONTROL_WRITETIMESTAMP   0x03
+
+#define PIPE_CONTROL_GTTWRITE_PROCESS_LOCAL 0x00
+#define PIPE_CONTROL_GTTWRITE_GLOBAL        0x01
+
+#define _3DPRIM_POINTLIST         0x01
+#define _3DPRIM_LINELIST          0x02
+#define _3DPRIM_LINESTRIP         0x03
+#define _3DPRIM_TRILIST           0x04
+#define _3DPRIM_TRISTRIP          0x05
+#define _3DPRIM_TRIFAN            0x06
+#define _3DPRIM_QUADLIST          0x07
+#define _3DPRIM_QUADSTRIP         0x08
+#define _3DPRIM_LINELIST_ADJ      0x09
+#define _3DPRIM_LINESTRIP_ADJ     0x0A
+#define _3DPRIM_TRILIST_ADJ       0x0B
+#define _3DPRIM_TRISTRIP_ADJ      0x0C
+#define _3DPRIM_TRISTRIP_REVERSE  0x0D
+#define _3DPRIM_POLYGON           0x0E
+#define _3DPRIM_RECTLIST          0x0F
+#define _3DPRIM_LINELOOP          0x10
+#define _3DPRIM_POINTLIST_BF      0x11
+#define _3DPRIM_LINESTRIP_CONT    0x12
+#define _3DPRIM_LINESTRIP_BF      0x13
+#define _3DPRIM_LINESTRIP_CONT_BF 0x14
+#define _3DPRIM_TRIFAN_NOSTIPPLE  0x15
+
+#define _3DPRIM_VERTEXBUFFER_ACCESS_SEQUENTIAL 0
+#define _3DPRIM_VERTEXBUFFER_ACCESS_RANDOM     1
+
+#define BRW_ANISORATIO_2     0
+#define BRW_ANISORATIO_4     1
+#define BRW_ANISORATIO_6     2
+#define BRW_ANISORATIO_8     3
+#define BRW_ANISORATIO_10    4
+#define BRW_ANISORATIO_12    5
+#define BRW_ANISORATIO_14    6
+#define BRW_ANISORATIO_16    7
+
+#define BRW_BLENDFACTOR_ONE                 0x1
+#define BRW_BLENDFACTOR_SRC_COLOR           0x2
+#define BRW_BLENDFACTOR_SRC_ALPHA           0x3
+#define BRW_BLENDFACTOR_DST_ALPHA           0x4
+#define BRW_BLENDFACTOR_DST_COLOR           0x5
+#define BRW_BLENDFACTOR_SRC_ALPHA_SATURATE  0x6
+#define BRW_BLENDFACTOR_CONST_COLOR         0x7
+#define BRW_BLENDFACTOR_CONST_ALPHA         0x8
+#define BRW_BLENDFACTOR_SRC1_COLOR          0x9
+#define BRW_BLENDFACTOR_SRC1_ALPHA          0x0A
+#define BRW_BLENDFACTOR_ZERO                0x11
+#define BRW_BLENDFACTOR_INV_SRC_COLOR       0x12
+#define BRW_BLENDFACTOR_INV_SRC_ALPHA       0x13
+#define BRW_BLENDFACTOR_INV_DST_ALPHA       0x14
+#define BRW_BLENDFACTOR_INV_DST_COLOR       0x15
+#define BRW_BLENDFACTOR_INV_CONST_COLOR     0x17
+#define BRW_BLENDFACTOR_INV_CONST_ALPHA     0x18
+#define BRW_BLENDFACTOR_INV_SRC1_COLOR      0x19
+#define BRW_BLENDFACTOR_INV_SRC1_ALPHA      0x1A
+
+#define BRW_BLENDFUNCTION_ADD               0
+#define BRW_BLENDFUNCTION_SUBTRACT          1
+#define BRW_BLENDFUNCTION_REVERSE_SUBTRACT  2
+#define BRW_BLENDFUNCTION_MIN               3
+#define BRW_BLENDFUNCTION_MAX               4
+
+#define BRW_ALPHATEST_FORMAT_UNORM8         0
+#define BRW_ALPHATEST_FORMAT_FLOAT32        1
+
+#define BRW_CHROMAKEY_KILL_ON_ANY_MATCH  0
+#define BRW_CHROMAKEY_REPLACE_BLACK      1
+
+#define BRW_CLIP_API_OGL     0
+#define BRW_CLIP_API_DX      1
+
+#define BRW_CLIPMODE_NORMAL              0
+#define BRW_CLIPMODE_CLIP_ALL            1
+#define BRW_CLIPMODE_CLIP_NON_REJECTED   2
+#define BRW_CLIPMODE_REJECT_ALL          3
+#define BRW_CLIPMODE_ACCEPT_ALL          4
+
+#define BRW_CLIP_NDCSPACE     0
+#define BRW_CLIP_SCREENSPACE  1
+
+#define BRW_COMPAREFUNCTION_ALWAYS       0
+#define BRW_COMPAREFUNCTION_NEVER        1
+#define BRW_COMPAREFUNCTION_LESS         2
+#define BRW_COMPAREFUNCTION_EQUAL        3
+#define BRW_COMPAREFUNCTION_LEQUAL       4
+#define BRW_COMPAREFUNCTION_GREATER      5
+#define BRW_COMPAREFUNCTION_NOTEQUAL     6
+#define BRW_COMPAREFUNCTION_GEQUAL       7
+
+#define BRW_COVERAGE_PIXELS_HALF     0
+#define BRW_COVERAGE_PIXELS_1        1
+#define BRW_COVERAGE_PIXELS_2        2
+#define BRW_COVERAGE_PIXELS_4        3
+
+#define BRW_CULLMODE_BOTH        0
+#define BRW_CULLMODE_NONE        1
+#define BRW_CULLMODE_FRONT       2
+#define BRW_CULLMODE_BACK        3
+
+#define BRW_DEFAULTCOLOR_R8G8B8A8_UNORM      0
+#define BRW_DEFAULTCOLOR_R32G32B32A32_FLOAT  1
+
+#define BRW_DEPTHFORMAT_D32_FLOAT_S8X24_UINT     0
+#define BRW_DEPTHFORMAT_D32_FLOAT                1
+#define BRW_DEPTHFORMAT_D24_UNORM_S8_UINT        2
+#define BRW_DEPTHFORMAT_D16_UNORM                5
+
+#define BRW_FLOATING_POINT_IEEE_754        0
+#define BRW_FLOATING_POINT_NON_IEEE_754    1
+
+#define BRW_FRONTWINDING_CW      0
+#define BRW_FRONTWINDING_CCW     1
+
+#define BRW_SPRITE_POINT_ENABLE  16
+
+#define BRW_INDEX_BYTE     0
+#define BRW_INDEX_WORD     1
+#define BRW_INDEX_DWORD    2
+
+#define BRW_LOGICOPFUNCTION_CLEAR            0
+#define BRW_LOGICOPFUNCTION_NOR              1
+#define BRW_LOGICOPFUNCTION_AND_INVERTED     2
+#define BRW_LOGICOPFUNCTION_COPY_INVERTED    3
+#define BRW_LOGICOPFUNCTION_AND_REVERSE      4
+#define BRW_LOGICOPFUNCTION_INVERT           5
+#define BRW_LOGICOPFUNCTION_XOR              6
+#define BRW_LOGICOPFUNCTION_NAND             7
+#define BRW_LOGICOPFUNCTION_AND              8
+#define BRW_LOGICOPFUNCTION_EQUIV            9
+#define BRW_LOGICOPFUNCTION_NOOP             10
+#define BRW_LOGICOPFUNCTION_OR_INVERTED      11
+#define BRW_LOGICOPFUNCTION_COPY             12
+#define BRW_LOGICOPFUNCTION_OR_REVERSE       13
+#define BRW_LOGICOPFUNCTION_OR               14
+#define BRW_LOGICOPFUNCTION_SET              15
+
+#define BRW_MAPFILTER_NEAREST        0x0
+#define BRW_MAPFILTER_LINEAR         0x1
+#define BRW_MAPFILTER_ANISOTROPIC    0x2
+
+#define BRW_MIPFILTER_NONE        0
+#define BRW_MIPFILTER_NEAREST     1
+#define BRW_MIPFILTER_LINEAR      3
+
+#define BRW_POLYGON_FRONT_FACING     0
+#define BRW_POLYGON_BACK_FACING      1
+
+#define BRW_PREFILTER_ALWAYS     0x0
+#define BRW_PREFILTER_NEVER      0x1
+#define BRW_PREFILTER_LESS       0x2
+#define BRW_PREFILTER_EQUAL      0x3
+#define BRW_PREFILTER_LEQUAL     0x4
+#define BRW_PREFILTER_GREATER    0x5
+#define BRW_PREFILTER_NOTEQUAL   0x6
+#define BRW_PREFILTER_GEQUAL     0x7
+
+#define BRW_PROVOKING_VERTEX_0    0
+#define BRW_PROVOKING_VERTEX_1    1
+#define BRW_PROVOKING_VERTEX_2    2
+
+#define BRW_RASTRULE_UPPER_LEFT  0
+#define BRW_RASTRULE_UPPER_RIGHT 1
+
+#define BRW_RENDERTARGET_CLAMPRANGE_UNORM    0
+#define BRW_RENDERTARGET_CLAMPRANGE_SNORM    1
+#define BRW_RENDERTARGET_CLAMPRANGE_FORMAT   2
+
+#define BRW_STENCILOP_KEEP               0
+#define BRW_STENCILOP_ZERO               1
+#define BRW_STENCILOP_REPLACE            2
+#define BRW_STENCILOP_INCRSAT            3
+#define BRW_STENCILOP_DECRSAT            4
+#define BRW_STENCILOP_INCR               5
+#define BRW_STENCILOP_DECR               6
+#define BRW_STENCILOP_INVERT             7
+
+#define BRW_SURFACE_MIPMAPLAYOUT_BELOW   0
+#define BRW_SURFACE_MIPMAPLAYOUT_RIGHT   1
+
+#define BRW_SURFACEFORMAT_R32G32B32A32_FLOAT             0x000
+#define BRW_SURFACEFORMAT_R32G32B32A32_SINT              0x001
+#define BRW_SURFACEFORMAT_R32G32B32A32_UINT              0x002
+#define BRW_SURFACEFORMAT_R32G32B32A32_UNORM             0x003
+#define BRW_SURFACEFORMAT_R32G32B32A32_SNORM             0x004
+#define BRW_SURFACEFORMAT_R64G64_FLOAT                   0x005
+#define BRW_SURFACEFORMAT_R32G32B32X32_FLOAT             0x006
+#define BRW_SURFACEFORMAT_R32G32B32A32_SSCALED           0x007
+#define BRW_SURFACEFORMAT_R32G32B32A32_USCALED           0x008
+#define BRW_SURFACEFORMAT_R32G32B32_FLOAT                0x040
+#define BRW_SURFACEFORMAT_R32G32B32_SINT                 0x041
+#define BRW_SURFACEFORMAT_R32G32B32_UINT                 0x042
+#define BRW_SURFACEFORMAT_R32G32B32_UNORM                0x043
+#define BRW_SURFACEFORMAT_R32G32B32_SNORM                0x044
+#define BRW_SURFACEFORMAT_R32G32B32_SSCALED              0x045
+#define BRW_SURFACEFORMAT_R32G32B32_USCALED              0x046
+#define BRW_SURFACEFORMAT_R16G16B16A16_UNORM             0x080
+#define BRW_SURFACEFORMAT_R16G16B16A16_SNORM             0x081
+#define BRW_SURFACEFORMAT_R16G16B16A16_SINT              0x082
+#define BRW_SURFACEFORMAT_R16G16B16A16_UINT              0x083
+#define BRW_SURFACEFORMAT_R16G16B16A16_FLOAT             0x084
+#define BRW_SURFACEFORMAT_R32G32_FLOAT                   0x085
+#define BRW_SURFACEFORMAT_R32G32_SINT                    0x086
+#define BRW_SURFACEFORMAT_R32G32_UINT                    0x087
+#define BRW_SURFACEFORMAT_R32_FLOAT_X8X24_TYPELESS       0x088
+#define BRW_SURFACEFORMAT_X32_TYPELESS_G8X24_UINT        0x089
+#define BRW_SURFACEFORMAT_L32A32_FLOAT                   0x08A
+#define BRW_SURFACEFORMAT_R32G32_UNORM                   0x08B
+#define BRW_SURFACEFORMAT_R32G32_SNORM                   0x08C
+#define BRW_SURFACEFORMAT_R64_FLOAT                      0x08D
+#define BRW_SURFACEFORMAT_R16G16B16X16_UNORM             0x08E
+#define BRW_SURFACEFORMAT_R16G16B16X16_FLOAT             0x08F
+#define BRW_SURFACEFORMAT_A32X32_FLOAT                   0x090
+#define BRW_SURFACEFORMAT_L32X32_FLOAT                   0x091
+#define BRW_SURFACEFORMAT_I32X32_FLOAT                   0x092
+#define BRW_SURFACEFORMAT_R16G16B16A16_SSCALED           0x093
+#define BRW_SURFACEFORMAT_R16G16B16A16_USCALED           0x094
+#define BRW_SURFACEFORMAT_R32G32_SSCALED                 0x095
+#define BRW_SURFACEFORMAT_R32G32_USCALED                 0x096
+#define BRW_SURFACEFORMAT_B8G8R8A8_UNORM                 0x0C0
+#define BRW_SURFACEFORMAT_B8G8R8A8_UNORM_SRGB            0x0C1
+#define BRW_SURFACEFORMAT_R10G10B10A2_UNORM              0x0C2
+#define BRW_SURFACEFORMAT_R10G10B10A2_UNORM_SRGB         0x0C3
+#define BRW_SURFACEFORMAT_R10G10B10A2_UINT               0x0C4
+#define BRW_SURFACEFORMAT_R10G10B10_SNORM_A2_UNORM       0x0C5
+#define BRW_SURFACEFORMAT_R8G8B8A8_UNORM                 0x0C7
+#define BRW_SURFACEFORMAT_R8G8B8A8_UNORM_SRGB            0x0C8
+#define BRW_SURFACEFORMAT_R8G8B8A8_SNORM                 0x0C9
+#define BRW_SURFACEFORMAT_R8G8B8A8_SINT                  0x0CA
+#define BRW_SURFACEFORMAT_R8G8B8A8_UINT                  0x0CB
+#define BRW_SURFACEFORMAT_R16G16_UNORM                   0x0CC
+#define BRW_SURFACEFORMAT_R16G16_SNORM                   0x0CD
+#define BRW_SURFACEFORMAT_R16G16_SINT                    0x0CE
+#define BRW_SURFACEFORMAT_R16G16_UINT                    0x0CF
+#define BRW_SURFACEFORMAT_R16G16_FLOAT                   0x0D0
+#define BRW_SURFACEFORMAT_B10G10R10A2_UNORM              0x0D1
+#define BRW_SURFACEFORMAT_B10G10R10A2_UNORM_SRGB         0x0D2
+#define BRW_SURFACEFORMAT_R11G11B10_FLOAT                0x0D3
+#define BRW_SURFACEFORMAT_R32_SINT                       0x0D6
+#define BRW_SURFACEFORMAT_R32_UINT                       0x0D7
+#define BRW_SURFACEFORMAT_R32_FLOAT                      0x0D8
+#define BRW_SURFACEFORMAT_R24_UNORM_X8_TYPELESS          0x0D9
+#define BRW_SURFACEFORMAT_X24_TYPELESS_G8_UINT           0x0DA
+#define BRW_SURFACEFORMAT_L16A16_UNORM                   0x0DF
+#define BRW_SURFACEFORMAT_I24X8_UNORM                    0x0E0
+#define BRW_SURFACEFORMAT_L24X8_UNORM                    0x0E1
+#define BRW_SURFACEFORMAT_A24X8_UNORM                    0x0E2
+#define BRW_SURFACEFORMAT_I32_FLOAT                      0x0E3
+#define BRW_SURFACEFORMAT_L32_FLOAT                      0x0E4
+#define BRW_SURFACEFORMAT_A32_FLOAT                      0x0E5
+#define BRW_SURFACEFORMAT_B8G8R8X8_UNORM                 0x0E9
+#define BRW_SURFACEFORMAT_B8G8R8X8_UNORM_SRGB            0x0EA
+#define BRW_SURFACEFORMAT_R8G8B8X8_UNORM                 0x0EB
+#define BRW_SURFACEFORMAT_R8G8B8X8_UNORM_SRGB            0x0EC
+#define BRW_SURFACEFORMAT_R9G9B9E5_SHAREDEXP             0x0ED
+#define BRW_SURFACEFORMAT_B10G10R10X2_UNORM              0x0EE
+#define BRW_SURFACEFORMAT_L16A16_FLOAT                   0x0F0
+#define BRW_SURFACEFORMAT_R32_UNORM                      0x0F1
+#define BRW_SURFACEFORMAT_R32_SNORM                      0x0F2
+#define BRW_SURFACEFORMAT_R10G10B10X2_USCALED            0x0F3
+#define BRW_SURFACEFORMAT_R8G8B8A8_SSCALED               0x0F4
+#define BRW_SURFACEFORMAT_R8G8B8A8_USCALED               0x0F5
+#define BRW_SURFACEFORMAT_R16G16_SSCALED                 0x0F6
+#define BRW_SURFACEFORMAT_R16G16_USCALED                 0x0F7
+#define BRW_SURFACEFORMAT_R32_SSCALED                    0x0F8
+#define BRW_SURFACEFORMAT_R32_USCALED                    0x0F9
+#define BRW_SURFACEFORMAT_B5G6R5_UNORM                   0x100
+#define BRW_SURFACEFORMAT_B5G6R5_UNORM_SRGB              0x101
+#define BRW_SURFACEFORMAT_B5G5R5A1_UNORM                 0x102
+#define BRW_SURFACEFORMAT_B5G5R5A1_UNORM_SRGB            0x103
+#define BRW_SURFACEFORMAT_B4G4R4A4_UNORM                 0x104
+#define BRW_SURFACEFORMAT_B4G4R4A4_UNORM_SRGB            0x105
+#define BRW_SURFACEFORMAT_R8G8_UNORM                     0x106
+#define BRW_SURFACEFORMAT_R8G8_SNORM                     0x107
+#define BRW_SURFACEFORMAT_R8G8_SINT                      0x108
+#define BRW_SURFACEFORMAT_R8G8_UINT                      0x109
+#define BRW_SURFACEFORMAT_R16_UNORM                      0x10A
+#define BRW_SURFACEFORMAT_R16_SNORM                      0x10B
+#define BRW_SURFACEFORMAT_R16_SINT                       0x10C
+#define BRW_SURFACEFORMAT_R16_UINT                       0x10D
+#define BRW_SURFACEFORMAT_R16_FLOAT                      0x10E
+#define BRW_SURFACEFORMAT_I16_UNORM                      0x111
+#define BRW_SURFACEFORMAT_L16_UNORM                      0x112
+#define BRW_SURFACEFORMAT_A16_UNORM                      0x113
+#define BRW_SURFACEFORMAT_L8A8_UNORM                     0x114
+#define BRW_SURFACEFORMAT_I16_FLOAT                      0x115
+#define BRW_SURFACEFORMAT_L16_FLOAT                      0x116
+#define BRW_SURFACEFORMAT_A16_FLOAT                      0x117
+#define BRW_SURFACEFORMAT_R5G5_SNORM_B6_UNORM            0x119
+#define BRW_SURFACEFORMAT_B5G5R5X1_UNORM                 0x11A
+#define BRW_SURFACEFORMAT_B5G5R5X1_UNORM_SRGB            0x11B
+#define BRW_SURFACEFORMAT_R8G8_SSCALED                   0x11C
+#define BRW_SURFACEFORMAT_R8G8_USCALED                   0x11D
+#define BRW_SURFACEFORMAT_R16_SSCALED                    0x11E
+#define BRW_SURFACEFORMAT_R16_USCALED                    0x11F
+#define BRW_SURFACEFORMAT_R8_UNORM                       0x140
+#define BRW_SURFACEFORMAT_R8_SNORM                       0x141
+#define BRW_SURFACEFORMAT_R8_SINT                        0x142
+#define BRW_SURFACEFORMAT_R8_UINT                        0x143
+#define BRW_SURFACEFORMAT_A8_UNORM                       0x144
+#define BRW_SURFACEFORMAT_I8_UNORM                       0x145
+#define BRW_SURFACEFORMAT_L8_UNORM                       0x146
+#define BRW_SURFACEFORMAT_P4A4_UNORM                     0x147
+#define BRW_SURFACEFORMAT_A4P4_UNORM                     0x148
+#define BRW_SURFACEFORMAT_R8_SSCALED                     0x149
+#define BRW_SURFACEFORMAT_R8_USCALED                     0x14A
+#define BRW_SURFACEFORMAT_R1_UINT                        0x181
+#define BRW_SURFACEFORMAT_YCRCB_NORMAL                   0x182
+#define BRW_SURFACEFORMAT_YCRCB_SWAPUVY                  0x183
+#define BRW_SURFACEFORMAT_BC1_UNORM                      0x186
+#define BRW_SURFACEFORMAT_BC2_UNORM                      0x187
+#define BRW_SURFACEFORMAT_BC3_UNORM                      0x188
+#define BRW_SURFACEFORMAT_BC4_UNORM                      0x189
+#define BRW_SURFACEFORMAT_BC5_UNORM                      0x18A
+#define BRW_SURFACEFORMAT_BC1_UNORM_SRGB                 0x18B
+#define BRW_SURFACEFORMAT_BC2_UNORM_SRGB                 0x18C
+#define BRW_SURFACEFORMAT_BC3_UNORM_SRGB                 0x18D
+#define BRW_SURFACEFORMAT_MONO8                          0x18E
+#define BRW_SURFACEFORMAT_YCRCB_SWAPUV                   0x18F
+#define BRW_SURFACEFORMAT_YCRCB_SWAPY                    0x190
+#define BRW_SURFACEFORMAT_DXT1_RGB                       0x191
+#define BRW_SURFACEFORMAT_FXT1                           0x192
+#define BRW_SURFACEFORMAT_R8G8B8_UNORM                   0x193
+#define BRW_SURFACEFORMAT_R8G8B8_SNORM                   0x194
+#define BRW_SURFACEFORMAT_R8G8B8_SSCALED                 0x195
+#define BRW_SURFACEFORMAT_R8G8B8_USCALED                 0x196
+#define BRW_SURFACEFORMAT_R64G64B64A64_FLOAT             0x197
+#define BRW_SURFACEFORMAT_R64G64B64_FLOAT                0x198
+#define BRW_SURFACEFORMAT_BC4_SNORM                      0x199
+#define BRW_SURFACEFORMAT_BC5_SNORM                      0x19A
+#define BRW_SURFACEFORMAT_R16G16B16_UNORM                0x19C
+#define BRW_SURFACEFORMAT_R16G16B16_SNORM                0x19D
+#define BRW_SURFACEFORMAT_R16G16B16_SSCALED              0x19E
+#define BRW_SURFACEFORMAT_R16G16B16_USCALED              0x19F
+
+#define BRW_SURFACERETURNFORMAT_FLOAT32  0
+#define BRW_SURFACERETURNFORMAT_S1       1
+
+#define BRW_SURFACE_1D      0
+#define BRW_SURFACE_2D      1
+#define BRW_SURFACE_3D      2
+#define BRW_SURFACE_CUBE    3
+#define BRW_SURFACE_BUFFER  4
+#define BRW_SURFACE_NULL    7
+
+#define BRW_TEXCOORDMODE_WRAP            0
+#define BRW_TEXCOORDMODE_MIRROR          1
+#define BRW_TEXCOORDMODE_CLAMP           2
+#define BRW_TEXCOORDMODE_CUBE            3
+#define BRW_TEXCOORDMODE_CLAMP_BORDER    4
+#define BRW_TEXCOORDMODE_MIRROR_ONCE     5
+
+#define BRW_THREAD_PRIORITY_NORMAL   0
+#define BRW_THREAD_PRIORITY_HIGH     1
+
+#define BRW_TILEWALK_XMAJOR                 0
+#define BRW_TILEWALK_YMAJOR                 1
+
+#define BRW_VERTEX_SUBPIXEL_PRECISION_8BITS  0
+#define BRW_VERTEX_SUBPIXEL_PRECISION_4BITS  1
+
+#define BRW_VERTEXBUFFER_ACCESS_VERTEXDATA     0
+#define BRW_VERTEXBUFFER_ACCESS_INSTANCEDATA   1
+
+#define BRW_VFCOMPONENT_NOSTORE      0
+#define BRW_VFCOMPONENT_STORE_SRC    1
+#define BRW_VFCOMPONENT_STORE_0      2
+#define BRW_VFCOMPONENT_STORE_1_FLT  3
+#define BRW_VFCOMPONENT_STORE_1_INT  4
+#define BRW_VFCOMPONENT_STORE_VID    5
+#define BRW_VFCOMPONENT_STORE_IID    6
+#define BRW_VFCOMPONENT_STORE_PID    7
+
+
+
+/* Execution Unit (EU) defines
+ */
+
+#define BRW_ALIGN_1   0
+#define BRW_ALIGN_16  1
+
+#define BRW_ADDRESS_DIRECT                        0
+#define BRW_ADDRESS_REGISTER_INDIRECT_REGISTER    1
+
+#define BRW_CHANNEL_X     0
+#define BRW_CHANNEL_Y     1
+#define BRW_CHANNEL_Z     2
+#define BRW_CHANNEL_W     3
+
+#define BRW_COMPRESSION_NONE          0
+#define BRW_COMPRESSION_2NDHALF       1
+#define BRW_COMPRESSION_COMPRESSED    2
+
+#define BRW_CONDITIONAL_NONE  0
+#define BRW_CONDITIONAL_Z     1
+#define BRW_CONDITIONAL_NZ    2
+#define BRW_CONDITIONAL_EQ    1	/* Z */
+#define BRW_CONDITIONAL_NEQ   2	/* NZ */
+#define BRW_CONDITIONAL_G     3
+#define BRW_CONDITIONAL_GE    4
+#define BRW_CONDITIONAL_L     5
+#define BRW_CONDITIONAL_LE    6
+#define BRW_CONDITIONAL_C     7
+#define BRW_CONDITIONAL_O     8
+
+#define BRW_DEBUG_NONE        0
+#define BRW_DEBUG_BREAKPOINT  1
+
+#define BRW_DEPENDENCY_NORMAL         0
+#define BRW_DEPENDENCY_NOTCLEARED     1
+#define BRW_DEPENDENCY_NOTCHECKED     2
+#define BRW_DEPENDENCY_DISABLE        3
+
+#define BRW_EXECUTE_1     0
+#define BRW_EXECUTE_2     1
+#define BRW_EXECUTE_4     2
+#define BRW_EXECUTE_8     3
+#define BRW_EXECUTE_16    4
+#define BRW_EXECUTE_32    5
+
+#define BRW_HORIZONTAL_STRIDE_0   0
+#define BRW_HORIZONTAL_STRIDE_1   1
+#define BRW_HORIZONTAL_STRIDE_2   2
+#define BRW_HORIZONTAL_STRIDE_4   3
+
+#define BRW_INSTRUCTION_NORMAL    0
+#define BRW_INSTRUCTION_SATURATE  1
+
+#define BRW_MASK_ENABLE   0
+#define BRW_MASK_DISABLE  1
+
+#define BRW_OPCODE_MOV        1
+#define BRW_OPCODE_SEL        2
+#define BRW_OPCODE_NOT        4
+#define BRW_OPCODE_AND        5
+#define BRW_OPCODE_OR         6
+#define BRW_OPCODE_XOR        7
+#define BRW_OPCODE_SHR        8
+#define BRW_OPCODE_SHL        9
+#define BRW_OPCODE_RSR        10
+#define BRW_OPCODE_RSL        11
+#define BRW_OPCODE_ASR        12
+#define BRW_OPCODE_CMP        16
+#define BRW_OPCODE_JMPI       32
+#define BRW_OPCODE_IF         34
+#define BRW_OPCODE_IFF        35
+#define BRW_OPCODE_ELSE       36
+#define BRW_OPCODE_ENDIF      37
+#define BRW_OPCODE_DO         38
+#define BRW_OPCODE_WHILE      39
+#define BRW_OPCODE_BREAK      40
+#define BRW_OPCODE_CONTINUE   41
+#define BRW_OPCODE_HALT       42
+#define BRW_OPCODE_MSAVE      44
+#define BRW_OPCODE_MRESTORE   45
+#define BRW_OPCODE_PUSH       46
+#define BRW_OPCODE_POP        47
+#define BRW_OPCODE_WAIT       48
+#define BRW_OPCODE_SEND       49
+#define BRW_OPCODE_ADD        64
+#define BRW_OPCODE_MUL        65
+#define BRW_OPCODE_AVG        66
+#define BRW_OPCODE_FRC        67
+#define BRW_OPCODE_RNDU       68
+#define BRW_OPCODE_RNDD       69
+#define BRW_OPCODE_RNDE       70
+#define BRW_OPCODE_RNDZ       71
+#define BRW_OPCODE_MAC        72
+#define BRW_OPCODE_MACH       73
+#define BRW_OPCODE_LZD        74
+#define BRW_OPCODE_SAD2       80
+#define BRW_OPCODE_SADA2      81
+#define BRW_OPCODE_DP4        84
+#define BRW_OPCODE_DPH        85
+#define BRW_OPCODE_DP3        86
+#define BRW_OPCODE_DP2        87
+#define BRW_OPCODE_DPA2       88
+#define BRW_OPCODE_LINE       89
+#define BRW_OPCODE_NOP        126
+
+#define BRW_PREDICATE_NONE             0
+#define BRW_PREDICATE_NORMAL           1
+#define BRW_PREDICATE_ALIGN1_ANYV             2
+#define BRW_PREDICATE_ALIGN1_ALLV             3
+#define BRW_PREDICATE_ALIGN1_ANY2H            4
+#define BRW_PREDICATE_ALIGN1_ALL2H            5
+#define BRW_PREDICATE_ALIGN1_ANY4H            6
+#define BRW_PREDICATE_ALIGN1_ALL4H            7
+#define BRW_PREDICATE_ALIGN1_ANY8H            8
+#define BRW_PREDICATE_ALIGN1_ALL8H            9
+#define BRW_PREDICATE_ALIGN1_ANY16H           10
+#define BRW_PREDICATE_ALIGN1_ALL16H           11
+#define BRW_PREDICATE_ALIGN16_REPLICATE_X     2
+#define BRW_PREDICATE_ALIGN16_REPLICATE_Y     3
+#define BRW_PREDICATE_ALIGN16_REPLICATE_Z     4
+#define BRW_PREDICATE_ALIGN16_REPLICATE_W     5
+#define BRW_PREDICATE_ALIGN16_ANY4H           6
+#define BRW_PREDICATE_ALIGN16_ALL4H           7
+
+#define BRW_ARCHITECTURE_REGISTER_FILE    0
+#define BRW_GENERAL_REGISTER_FILE         1
+#define BRW_MESSAGE_REGISTER_FILE         2
+#define BRW_IMMEDIATE_VALUE               3
+
+#define BRW_REGISTER_TYPE_UD  0
+#define BRW_REGISTER_TYPE_D   1
+#define BRW_REGISTER_TYPE_UW  2
+#define BRW_REGISTER_TYPE_W   3
+#define BRW_REGISTER_TYPE_UB  4
+#define BRW_REGISTER_TYPE_B   5
+#define BRW_REGISTER_TYPE_VF  5	/* packed float vector, immediates only? */
+#define BRW_REGISTER_TYPE_HF  6
+#define BRW_REGISTER_TYPE_V   6	/* packed int vector, immediates only, uword dest only */
+#define BRW_REGISTER_TYPE_F   7
+
+#define BRW_ARF_NULL                  0x00
+#define BRW_ARF_ADDRESS               0x10
+#define BRW_ARF_ACCUMULATOR           0x20
+#define BRW_ARF_FLAG                  0x30
+#define BRW_ARF_MASK                  0x40
+#define BRW_ARF_MASK_STACK            0x50
+#define BRW_ARF_MASK_STACK_DEPTH      0x60
+#define BRW_ARF_STATE                 0x70
+#define BRW_ARF_CONTROL               0x80
+#define BRW_ARF_NOTIFICATION_COUNT    0x90
+#define BRW_ARF_IP                    0xA0
+
+#define BRW_AMASK   0
+#define BRW_IMASK   1
+#define BRW_LMASK   2
+#define BRW_CMASK   3
+
+
+
+#define BRW_THREAD_NORMAL     0
+#define BRW_THREAD_ATOMIC     1
+#define BRW_THREAD_SWITCH     2
+
+#define BRW_VERTICAL_STRIDE_0                 0
+#define BRW_VERTICAL_STRIDE_1                 1
+#define BRW_VERTICAL_STRIDE_2                 2
+#define BRW_VERTICAL_STRIDE_4                 3
+#define BRW_VERTICAL_STRIDE_8                 4
+#define BRW_VERTICAL_STRIDE_16                5
+#define BRW_VERTICAL_STRIDE_32                6
+#define BRW_VERTICAL_STRIDE_64                7
+#define BRW_VERTICAL_STRIDE_128               8
+#define BRW_VERTICAL_STRIDE_256               9
+#define BRW_VERTICAL_STRIDE_ONE_DIMENSIONAL   0xF
+
+#define BRW_WIDTH_1       0
+#define BRW_WIDTH_2       1
+#define BRW_WIDTH_4       2
+#define BRW_WIDTH_8       3
+#define BRW_WIDTH_16      4
+
+#define BRW_STATELESS_BUFFER_BOUNDARY_1K      0
+#define BRW_STATELESS_BUFFER_BOUNDARY_2K      1
+#define BRW_STATELESS_BUFFER_BOUNDARY_4K      2
+#define BRW_STATELESS_BUFFER_BOUNDARY_8K      3
+#define BRW_STATELESS_BUFFER_BOUNDARY_16K     4
+#define BRW_STATELESS_BUFFER_BOUNDARY_32K     5
+#define BRW_STATELESS_BUFFER_BOUNDARY_64K     6
+#define BRW_STATELESS_BUFFER_BOUNDARY_128K    7
+#define BRW_STATELESS_BUFFER_BOUNDARY_256K    8
+#define BRW_STATELESS_BUFFER_BOUNDARY_512K    9
+#define BRW_STATELESS_BUFFER_BOUNDARY_1M      10
+#define BRW_STATELESS_BUFFER_BOUNDARY_2M      11
+
+#define BRW_POLYGON_FACING_FRONT      0
+#define BRW_POLYGON_FACING_BACK       1
+
+#define BRW_MESSAGE_TARGET_NULL               0
+#define BRW_MESSAGE_TARGET_MATH               1
+#define BRW_MESSAGE_TARGET_SAMPLER            2
+#define BRW_MESSAGE_TARGET_GATEWAY            3
+#define BRW_MESSAGE_TARGET_DATAPORT_READ      4
+#define BRW_MESSAGE_TARGET_DATAPORT_WRITE     5
+#define BRW_MESSAGE_TARGET_URB                6
+#define BRW_MESSAGE_TARGET_THREAD_SPAWNER     7
+
+#define BRW_SAMPLER_RETURN_FORMAT_FLOAT32     0
+#define BRW_SAMPLER_RETURN_FORMAT_UINT32      2
+#define BRW_SAMPLER_RETURN_FORMAT_SINT32      3
+
+#define BRW_SAMPLER_MESSAGE_SIMD8_SAMPLE              0
+#define BRW_SAMPLER_MESSAGE_SIMD16_SAMPLE             0
+#define BRW_SAMPLER_MESSAGE_SIMD16_SAMPLE_BIAS        0
+#define BRW_SAMPLER_MESSAGE_SIMD8_KILLPIX             1
+#define BRW_SAMPLER_MESSAGE_SIMD4X2_SAMPLE_LOD        1
+#define BRW_SAMPLER_MESSAGE_SIMD16_SAMPLE_LOD         1
+#define BRW_SAMPLER_MESSAGE_SIMD4X2_SAMPLE_GRADIENTS  2
+#define BRW_SAMPLER_MESSAGE_SIMD8_SAMPLE_GRADIENTS    2
+#define BRW_SAMPLER_MESSAGE_SIMD4X2_SAMPLE_COMPARE    0
+#define BRW_SAMPLER_MESSAGE_SIMD16_SAMPLE_COMPARE     2
+#define BRW_SAMPLER_MESSAGE_SIMD4X2_RESINFO           2
+#define BRW_SAMPLER_MESSAGE_SIMD8_RESINFO             2
+#define BRW_SAMPLER_MESSAGE_SIMD16_RESINFO            2
+#define BRW_SAMPLER_MESSAGE_SIMD4X2_LD                3
+#define BRW_SAMPLER_MESSAGE_SIMD8_LD                  3
+#define BRW_SAMPLER_MESSAGE_SIMD16_LD                 3
+
+#define BRW_DATAPORT_OWORD_BLOCK_1_OWORDLOW   0
+#define BRW_DATAPORT_OWORD_BLOCK_1_OWORDHIGH  1
+#define BRW_DATAPORT_OWORD_BLOCK_2_OWORDS     2
+#define BRW_DATAPORT_OWORD_BLOCK_4_OWORDS     3
+#define BRW_DATAPORT_OWORD_BLOCK_8_OWORDS     4
+
+#define BRW_DATAPORT_OWORD_DUAL_BLOCK_1OWORD     0
+#define BRW_DATAPORT_OWORD_DUAL_BLOCK_4OWORDS    2
+
+#define BRW_DATAPORT_DWORD_SCATTERED_BLOCK_8DWORDS   2
+#define BRW_DATAPORT_DWORD_SCATTERED_BLOCK_16DWORDS  3
+
+#define BRW_DATAPORT_READ_MESSAGE_OWORD_BLOCK_READ          0
+#define BRW_DATAPORT_READ_MESSAGE_OWORD_DUAL_BLOCK_READ     1
+#define BRW_DATAPORT_READ_MESSAGE_DWORD_BLOCK_READ          2
+#define BRW_DATAPORT_READ_MESSAGE_DWORD_SCATTERED_READ      3
+
+#define BRW_DATAPORT_READ_TARGET_DATA_CACHE      0
+#define BRW_DATAPORT_READ_TARGET_RENDER_CACHE    1
+#define BRW_DATAPORT_READ_TARGET_SAMPLER_CACHE   2
+
+#define BRW_DATAPORT_RENDER_TARGET_WRITE_SIMD16_SINGLE_SOURCE                0
+#define BRW_DATAPORT_RENDER_TARGET_WRITE_SIMD16_SINGLE_SOURCE_REPLICATED     1
+#define BRW_DATAPORT_RENDER_TARGET_WRITE_SIMD8_DUAL_SOURCE_SUBSPAN01         2
+#define BRW_DATAPORT_RENDER_TARGET_WRITE_SIMD8_DUAL_SOURCE_SUBSPAN23         3
+#define BRW_DATAPORT_RENDER_TARGET_WRITE_SIMD8_SINGLE_SOURCE_SUBSPAN01       4
+
+#define BRW_DATAPORT_WRITE_MESSAGE_OWORD_BLOCK_WRITE                0
+#define BRW_DATAPORT_WRITE_MESSAGE_OWORD_DUAL_BLOCK_WRITE           1
+#define BRW_DATAPORT_WRITE_MESSAGE_DWORD_BLOCK_WRITE                2
+#define BRW_DATAPORT_WRITE_MESSAGE_DWORD_SCATTERED_WRITE            3
+#define BRW_DATAPORT_WRITE_MESSAGE_RENDER_TARGET_WRITE              4
+#define BRW_DATAPORT_WRITE_MESSAGE_STREAMED_VERTEX_BUFFER_WRITE     5
+#define BRW_DATAPORT_WRITE_MESSAGE_FLUSH_RENDER_CACHE               7
+
+#define BRW_MATH_FUNCTION_INV                              1
+#define BRW_MATH_FUNCTION_LOG                              2
+#define BRW_MATH_FUNCTION_EXP                              3
+#define BRW_MATH_FUNCTION_SQRT                             4
+#define BRW_MATH_FUNCTION_RSQ                              5
+#define BRW_MATH_FUNCTION_SIN                              6 /* was 7 */
+#define BRW_MATH_FUNCTION_COS                              7 /* was 8 */
+#define BRW_MATH_FUNCTION_SINCOS                           8 /* was 6 */
+#define BRW_MATH_FUNCTION_TAN                              9
+#define BRW_MATH_FUNCTION_POW                              10
+#define BRW_MATH_FUNCTION_INT_DIV_QUOTIENT_AND_REMAINDER   11
+#define BRW_MATH_FUNCTION_INT_DIV_QUOTIENT                 12
+#define BRW_MATH_FUNCTION_INT_DIV_REMAINDER                13
+
+#define BRW_MATH_INTEGER_UNSIGNED     0
+#define BRW_MATH_INTEGER_SIGNED       1
+
+#define BRW_MATH_PRECISION_FULL        0
+#define BRW_MATH_PRECISION_PARTIAL     1
+
+#define BRW_MATH_SATURATE_NONE         0
+#define BRW_MATH_SATURATE_SATURATE     1
+
+#define BRW_MATH_DATA_VECTOR  0
+#define BRW_MATH_DATA_SCALAR  1
+
+#define BRW_URB_OPCODE_WRITE  0
+
+#define BRW_URB_SWIZZLE_NONE          0
+#define BRW_URB_SWIZZLE_INTERLEAVE    1
+#define BRW_URB_SWIZZLE_TRANSPOSE     2
+
+#define BRW_SCRATCH_SPACE_SIZE_1K     0
+#define BRW_SCRATCH_SPACE_SIZE_2K     1
+#define BRW_SCRATCH_SPACE_SIZE_4K     2
+#define BRW_SCRATCH_SPACE_SIZE_8K     3
+#define BRW_SCRATCH_SPACE_SIZE_16K    4
+#define BRW_SCRATCH_SPACE_SIZE_32K    5
+#define BRW_SCRATCH_SPACE_SIZE_64K    6
+#define BRW_SCRATCH_SPACE_SIZE_128K   7
+#define BRW_SCRATCH_SPACE_SIZE_256K   8
+#define BRW_SCRATCH_SPACE_SIZE_512K   9
+#define BRW_SCRATCH_SPACE_SIZE_1M     10
+#define BRW_SCRATCH_SPACE_SIZE_2M     11
+
+
+
+
+#define CMD_URB_FENCE                 0x6000
+#define CMD_CONST_BUFFER_STATE        0x6001
+#define CMD_CONST_BUFFER              0x6002
+
+#define CMD_STATE_BASE_ADDRESS        0x6101
+#define CMD_STATE_INSN_POINTER        0x6102
+#define CMD_PIPELINE_SELECT           0x6104
+
+#define CMD_PIPELINED_STATE_POINTERS  0x7800
+#define CMD_BINDING_TABLE_PTRS        0x7801
+#define CMD_VERTEX_BUFFER             0x7808
+#define CMD_VERTEX_ELEMENT            0x7809
+#define CMD_INDEX_BUFFER              0x780a
+#define CMD_VF_STATISTICS             0x780b
+
+#define CMD_DRAW_RECT                 0x7900
+#define CMD_BLEND_CONSTANT_COLOR      0x7901
+#define CMD_CHROMA_KEY                0x7904
+#define CMD_DEPTH_BUFFER              0x7905
+#define CMD_POLY_STIPPLE_OFFSET       0x7906
+#define CMD_POLY_STIPPLE_PATTERN      0x7907
+#define CMD_LINE_STIPPLE_PATTERN      0x7908
+#define CMD_GLOBAL_DEPTH_OFFSET_CLAMP 0x7909
+
+#define CMD_PIPE_CONTROL              0x7a00
+
+#define CMD_3D_PRIM                   0x7b00
+
+#define CMD_MI_FLUSH                  0x0200
+
+
+/* Various values from the R0 vertex header:
+ */
+#define R02_PRIM_END    0x1
+#define R02_PRIM_START  0x2
+
+
+
+#endif
diff --git a/src/gallium/drivers/i965simple/brw_draw.c b/src/gallium/drivers/i965simple/brw_draw.c
new file mode 100644
index 0000000000..7598e3dc8a
--- /dev/null
+++ b/src/gallium/drivers/i965simple/brw_draw.c
@@ -0,0 +1,239 @@
+/**************************************************************************
+ *
+ * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ **************************************************************************/
+
+#include <stdlib.h>
+
+#include "brw_batch.h"
+#include "brw_draw.h"
+#include "brw_defines.h"
+#include "brw_context.h"
+#include "brw_state.h"
+
+#include "pipe/p_context.h"
+#include "pipe/p_winsys.h"
+
+static unsigned hw_prim[PIPE_PRIM_POLYGON+1] = {
+   _3DPRIM_POINTLIST,
+   _3DPRIM_LINELIST,
+   _3DPRIM_LINELOOP,
+   _3DPRIM_LINESTRIP,
+   _3DPRIM_TRILIST,
+   _3DPRIM_TRISTRIP,
+   _3DPRIM_TRIFAN,
+   _3DPRIM_QUADLIST,
+   _3DPRIM_QUADSTRIP,
+   _3DPRIM_POLYGON
+};
+
+
+static const int reduced_prim[PIPE_PRIM_POLYGON+1] = {
+   PIPE_PRIM_POINTS,
+   PIPE_PRIM_LINES,
+   PIPE_PRIM_LINES,
+   PIPE_PRIM_LINES,
+   PIPE_PRIM_TRIANGLES,
+   PIPE_PRIM_TRIANGLES,
+   PIPE_PRIM_TRIANGLES,
+   PIPE_PRIM_TRIANGLES,
+   PIPE_PRIM_TRIANGLES,
+   PIPE_PRIM_TRIANGLES
+};
+
+
+/* When the primitive changes, set a state bit and re-validate.  Not
+ * the nicest and would rather deal with this by having all the
+ * programs be immune to the active primitive (ie. cope with all
+ * possibilities).  That may not be realistic however.
+ */
+static void brw_set_prim(struct brw_context *brw, int prim)
+{
+   PRINT("PRIM: %d\n", prim);
+
+   /* Slight optimization to avoid the GS program when not needed:
+    */
+   if (prim == PIPE_PRIM_QUAD_STRIP &&
+       brw->attribs.Raster->flatshade &&
+       brw->attribs.Raster->fill_cw == PIPE_POLYGON_MODE_FILL &&
+       brw->attribs.Raster->fill_ccw == PIPE_POLYGON_MODE_FILL)
+      prim = PIPE_PRIM_TRIANGLE_STRIP;
+
+   if (prim != brw->primitive) {
+      brw->primitive = prim;
+      brw->state.dirty.brw |= BRW_NEW_PRIMITIVE;
+
+      if (reduced_prim[prim] != brw->reduced_primitive) {
+	 brw->reduced_primitive = reduced_prim[prim];
+	 brw->state.dirty.brw |= BRW_NEW_REDUCED_PRIMITIVE;
+      }
+
+      brw_validate_state(brw);
+   }
+
+}
+
+
+static unsigned trim(int prim, unsigned length)
+{
+   if (prim == PIPE_PRIM_QUAD_STRIP)
+      return length > 3 ? (length - length % 2) : 0;
+   else if (prim == PIPE_PRIM_QUADS)
+      return length - length % 4;
+   else
+      return length;
+}
+
+
+
+static boolean brw_emit_prim( struct brw_context *brw,
+			      boolean indexed,
+			      unsigned start,
+			      unsigned count )
+
+{
+   struct brw_3d_primitive prim_packet;
+
+   if (BRW_DEBUG & DEBUG_PRIMS)
+      PRINT("PRIM: %d %d %d\n",  brw->primitive, start, count);
+
+   prim_packet.header.opcode = CMD_3D_PRIM;
+   prim_packet.header.length = sizeof(prim_packet)/4 - 2;
+   prim_packet.header.pad = 0;
+   prim_packet.header.topology = hw_prim[brw->primitive];
+   prim_packet.header.indexed = indexed;
+
+   prim_packet.verts_per_instance = trim(brw->primitive, count);
+   prim_packet.start_vert_location = start;
+   prim_packet.instance_count = 1;
+   prim_packet.start_instance_location = 0;
+   prim_packet.base_vert_location = 0;
+
+   if (prim_packet.verts_per_instance == 0)
+      return TRUE;
+
+   return brw_batchbuffer_data( brw->winsys,
+                                &prim_packet,
+                                sizeof(prim_packet) );
+}
+
+
+/* May fail if out of video memory for texture or vbo upload, or on
+ * fallback conditions.
+ */
+static boolean brw_try_draw_elements( struct pipe_context *pipe,
+				      struct pipe_buffer *index_buffer,
+				      unsigned index_size,
+				      unsigned mode,
+				      unsigned start,
+				      unsigned count )
+{
+   struct brw_context *brw = brw_context(pipe);
+
+   /* Set the first primitive ahead of validate_state:
+    */
+   brw_set_prim(brw, mode);
+
+   /* Upload index, vertex data:
+    */
+   if (index_buffer &&
+       !brw_upload_indices( brw, index_buffer, index_size, start, count ))
+      return FALSE;
+
+   if (!brw_upload_vertex_buffers(brw))
+      return FALSE;
+
+   if (!brw_upload_vertex_elements( brw ))
+      return FALSE;
+
+   /* XXX:  Need to separate validate and upload of state.
+    */
+   if (brw->state.dirty.brw)
+      brw_validate_state( brw );
+
+   if (!brw_emit_prim(brw, index_buffer != NULL,
+                      start, count))
+      return FALSE;
+
+   return TRUE;
+}
+
+
+
+static boolean brw_draw_elements( struct pipe_context *pipe,
+				  struct pipe_buffer *indexBuffer,
+				  unsigned indexSize,
+				  unsigned mode,
+				  unsigned start,
+				  unsigned count )
+{
+   if (!brw_try_draw_elements( pipe,
+			       indexBuffer,
+			       indexSize,
+			       mode, start, count ))
+   {
+      /* flush ? */
+
+      if (!brw_try_draw_elements( pipe,
+				  indexBuffer,
+				  indexSize,
+				  mode, start,
+				  count )) {
+	 assert(0);
+	 return FALSE;
+      }
+   }
+
+   return TRUE;
+}
+
+
+
+static boolean brw_draw_arrays( struct pipe_context *pipe,
+				    unsigned mode,
+				    unsigned start,
+				    unsigned count )
+{
+   if (!brw_try_draw_elements( pipe, NULL, 0, mode, start, count )) {
+      /* flush ? */
+
+      if (!brw_try_draw_elements( pipe, NULL, 0, mode, start, count )) {
+	 assert(0);
+	 return FALSE;
+      }
+   }
+   
+   return TRUE;
+}
+
+
+
+void brw_init_draw_functions( struct brw_context *brw )
+{
+   brw->pipe.draw_arrays = brw_draw_arrays;
+   brw->pipe.draw_elements = brw_draw_elements;
+}
+
+
diff --git a/src/gallium/drivers/i965simple/brw_draw.h b/src/gallium/drivers/i965simple/brw_draw.h
new file mode 100644
index 0000000000..62fe0d5d0e
--- /dev/null
+++ b/src/gallium/drivers/i965simple/brw_draw.h
@@ -0,0 +1,55 @@
+ /**************************************************************************
+ * 
+ * Copyright 2005 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+#ifndef BRW_DRAW_H
+#define BRW_DRAW_H
+
+#include "pipe/p_context.h"
+
+struct brw_context;
+
+
+
+void brw_init_draw_functions( struct brw_context *brw );
+
+
+boolean brw_upload_vertices( struct brw_context *brw,
+			       unsigned min_index,
+			       unsigned max_index );
+
+boolean brw_upload_indices(struct brw_context *brw,
+                           const struct pipe_buffer *index_buffer,
+                           int ib_size, int start, int count);
+
+boolean brw_upload_vertex_buffers( struct brw_context *brw );
+boolean brw_upload_vertex_elements( struct brw_context *brw );
+
+unsigned brw_translate_surface_format( unsigned id );
+
+
+
+#endif
diff --git a/src/gallium/drivers/i965simple/brw_draw_upload.c b/src/gallium/drivers/i965simple/brw_draw_upload.c
new file mode 100644
index 0000000000..aa85d93866
--- /dev/null
+++ b/src/gallium/drivers/i965simple/brw_draw_upload.c
@@ -0,0 +1,299 @@
+/**************************************************************************
+ *
+ * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ **************************************************************************/
+
+#include <stdlib.h>
+
+#include "brw_batch.h"
+#include "brw_draw.h"
+#include "brw_defines.h"
+#include "brw_context.h"
+#include "brw_state.h"
+
+struct brw_array_state {
+   union header_union header;
+
+   struct {
+      union {
+	 struct {
+	    unsigned pitch:11;
+	    unsigned pad:15;
+	    unsigned access_type:1;
+	    unsigned vb_index:5;
+	 } bits;
+	 unsigned dword;
+      } vb0;
+
+      struct pipe_buffer *buffer;
+      unsigned offset;
+
+      unsigned max_index;
+      unsigned instance_data_step_rate;
+
+   } vb[BRW_VBP_MAX];
+};
+
+
+
+unsigned brw_translate_surface_format( unsigned id )
+{
+   switch (id) {
+   case PIPE_FORMAT_R64_FLOAT:
+      return BRW_SURFACEFORMAT_R64_FLOAT;
+   case PIPE_FORMAT_R64G64_FLOAT:
+      return BRW_SURFACEFORMAT_R64G64_FLOAT;
+   case PIPE_FORMAT_R64G64B64_FLOAT:
+      return BRW_SURFACEFORMAT_R64G64B64_FLOAT;
+   case PIPE_FORMAT_R64G64B64A64_FLOAT:
+      return BRW_SURFACEFORMAT_R64G64B64A64_FLOAT;
+
+   case PIPE_FORMAT_R32_FLOAT:
+      return BRW_SURFACEFORMAT_R32_FLOAT;
+   case PIPE_FORMAT_R32G32_FLOAT:
+      return BRW_SURFACEFORMAT_R32G32_FLOAT;
+   case PIPE_FORMAT_R32G32B32_FLOAT:
+      return BRW_SURFACEFORMAT_R32G32B32_FLOAT;
+   case PIPE_FORMAT_R32G32B32A32_FLOAT:
+      return BRW_SURFACEFORMAT_R32G32B32A32_FLOAT;
+
+   case PIPE_FORMAT_R32_UNORM:
+      return BRW_SURFACEFORMAT_R32_UNORM;
+   case PIPE_FORMAT_R32G32_UNORM:
+      return BRW_SURFACEFORMAT_R32G32_UNORM;
+   case PIPE_FORMAT_R32G32B32_UNORM:
+      return BRW_SURFACEFORMAT_R32G32B32_UNORM;
+   case PIPE_FORMAT_R32G32B32A32_UNORM:
+      return BRW_SURFACEFORMAT_R32G32B32A32_UNORM;
+
+   case PIPE_FORMAT_R32_USCALED:
+      return BRW_SURFACEFORMAT_R32_USCALED;
+   case PIPE_FORMAT_R32G32_USCALED:
+      return BRW_SURFACEFORMAT_R32G32_USCALED;
+   case PIPE_FORMAT_R32G32B32_USCALED:
+      return BRW_SURFACEFORMAT_R32G32B32_USCALED;
+   case PIPE_FORMAT_R32G32B32A32_USCALED:
+      return BRW_SURFACEFORMAT_R32G32B32A32_USCALED;
+
+   case PIPE_FORMAT_R32_SNORM:
+      return BRW_SURFACEFORMAT_R32_SNORM;
+   case PIPE_FORMAT_R32G32_SNORM:
+      return BRW_SURFACEFORMAT_R32G32_SNORM;
+   case PIPE_FORMAT_R32G32B32_SNORM:
+      return BRW_SURFACEFORMAT_R32G32B32_SNORM;
+   case PIPE_FORMAT_R32G32B32A32_SNORM:
+      return BRW_SURFACEFORMAT_R32G32B32A32_SNORM;
+
+   case PIPE_FORMAT_R32_SSCALED:
+      return BRW_SURFACEFORMAT_R32_SSCALED;
+   case PIPE_FORMAT_R32G32_SSCALED:
+      return BRW_SURFACEFORMAT_R32G32_SSCALED;
+   case PIPE_FORMAT_R32G32B32_SSCALED:
+      return BRW_SURFACEFORMAT_R32G32B32_SSCALED;
+   case PIPE_FORMAT_R32G32B32A32_SSCALED:
+      return BRW_SURFACEFORMAT_R32G32B32A32_SSCALED;
+
+   case PIPE_FORMAT_R16_UNORM:
+      return BRW_SURFACEFORMAT_R16_UNORM;
+   case PIPE_FORMAT_R16G16_UNORM:
+      return BRW_SURFACEFORMAT_R16G16_UNORM;
+   case PIPE_FORMAT_R16G16B16_UNORM:
+      return BRW_SURFACEFORMAT_R16G16B16_UNORM;
+   case PIPE_FORMAT_R16G16B16A16_UNORM:
+      return BRW_SURFACEFORMAT_R16G16B16A16_UNORM;
+
+   case PIPE_FORMAT_R16_USCALED:
+      return BRW_SURFACEFORMAT_R16_USCALED;
+   case PIPE_FORMAT_R16G16_USCALED:
+      return BRW_SURFACEFORMAT_R16G16_USCALED;
+   case PIPE_FORMAT_R16G16B16_USCALED:
+      return BRW_SURFACEFORMAT_R16G16B16_USCALED;
+   case PIPE_FORMAT_R16G16B16A16_USCALED:
+      return BRW_SURFACEFORMAT_R16G16B16A16_USCALED;
+
+   case PIPE_FORMAT_R16_SNORM:
+      return BRW_SURFACEFORMAT_R16_SNORM;
+   case PIPE_FORMAT_R16G16_SNORM:
+      return BRW_SURFACEFORMAT_R16G16_SNORM;
+   case PIPE_FORMAT_R16G16B16_SNORM:
+      return BRW_SURFACEFORMAT_R16G16B16_SNORM;
+   case PIPE_FORMAT_R16G16B16A16_SNORM:
+      return BRW_SURFACEFORMAT_R16G16B16A16_SNORM;
+
+   case PIPE_FORMAT_R16_SSCALED:
+      return BRW_SURFACEFORMAT_R16_SSCALED;
+   case PIPE_FORMAT_R16G16_SSCALED:
+      return BRW_SURFACEFORMAT_R16G16_SSCALED;
+   case PIPE_FORMAT_R16G16B16_SSCALED:
+      return BRW_SURFACEFORMAT_R16G16B16_SSCALED;
+   case PIPE_FORMAT_R16G16B16A16_SSCALED:
+      return BRW_SURFACEFORMAT_R16G16B16A16_SSCALED;
+
+   case PIPE_FORMAT_R8_UNORM:
+      return BRW_SURFACEFORMAT_R8_UNORM;
+   case PIPE_FORMAT_R8G8_UNORM:
+      return BRW_SURFACEFORMAT_R8G8_UNORM;
+   case PIPE_FORMAT_R8G8B8_UNORM:
+      return BRW_SURFACEFORMAT_R8G8B8_UNORM;
+   case PIPE_FORMAT_R8G8B8A8_UNORM:
+      return BRW_SURFACEFORMAT_R8G8B8A8_UNORM;
+
+   case PIPE_FORMAT_R8_USCALED:
+      return BRW_SURFACEFORMAT_R8_USCALED;
+   case PIPE_FORMAT_R8G8_USCALED:
+      return BRW_SURFACEFORMAT_R8G8_USCALED;
+   case PIPE_FORMAT_R8G8B8_USCALED:
+      return BRW_SURFACEFORMAT_R8G8B8_USCALED;
+   case PIPE_FORMAT_R8G8B8A8_USCALED:
+      return BRW_SURFACEFORMAT_R8G8B8A8_USCALED;
+
+   case PIPE_FORMAT_R8_SNORM:
+      return BRW_SURFACEFORMAT_R8_SNORM;
+   case PIPE_FORMAT_R8G8_SNORM:
+      return BRW_SURFACEFORMAT_R8G8_SNORM;
+   case PIPE_FORMAT_R8G8B8_SNORM:
+      return BRW_SURFACEFORMAT_R8G8B8_SNORM;
+   case PIPE_FORMAT_R8G8B8A8_SNORM:
+      return BRW_SURFACEFORMAT_R8G8B8A8_SNORM;
+
+   case PIPE_FORMAT_R8_SSCALED:
+      return BRW_SURFACEFORMAT_R8_SSCALED;
+   case PIPE_FORMAT_R8G8_SSCALED:
+      return BRW_SURFACEFORMAT_R8G8_SSCALED;
+   case PIPE_FORMAT_R8G8B8_SSCALED:
+      return BRW_SURFACEFORMAT_R8G8B8_SSCALED;
+   case PIPE_FORMAT_R8G8B8A8_SSCALED:
+      return BRW_SURFACEFORMAT_R8G8B8A8_SSCALED;
+
+   default:
+      assert(0);
+      return 0;
+   }
+}
+
+static unsigned get_index_type(int type)
+{
+   switch (type) {
+   case 1: return BRW_INDEX_BYTE;
+   case 2: return BRW_INDEX_WORD;
+   case 4: return BRW_INDEX_DWORD;
+   default: assert(0); return 0;
+   }
+}
+
+
+boolean brw_upload_vertex_buffers( struct brw_context *brw )
+{
+   struct brw_array_state vbp;
+   unsigned nr_enabled = 0;
+   unsigned i;
+
+   memset(&vbp, 0, sizeof(vbp));
+
+   /* This is a hardware limit:
+    */
+
+   for (i = 0; i < BRW_VEP_MAX; i++)
+   {
+      if (brw->vb.vbo_array[i] == NULL) {
+	 nr_enabled = i;
+	 break;
+      }
+
+      vbp.vb[i].vb0.bits.pitch = brw->vb.vbo_array[i]->pitch;
+      vbp.vb[i].vb0.bits.pad = 0;
+      vbp.vb[i].vb0.bits.access_type = BRW_VERTEXBUFFER_ACCESS_VERTEXDATA;
+      vbp.vb[i].vb0.bits.vb_index = i;
+      vbp.vb[i].offset = brw->vb.vbo_array[i]->buffer_offset;
+      vbp.vb[i].buffer = brw->vb.vbo_array[i]->buffer;
+      vbp.vb[i].max_index = brw->vb.vbo_array[i]->max_index;
+   }
+
+
+   vbp.header.bits.length = (1 + nr_enabled * 4) - 2;
+   vbp.header.bits.opcode = CMD_VERTEX_BUFFER;
+
+   BEGIN_BATCH(vbp.header.bits.length+2, 0);
+   OUT_BATCH( vbp.header.dword );
+
+   for (i = 0; i < nr_enabled; i++) {
+      OUT_BATCH( vbp.vb[i].vb0.dword );
+      OUT_RELOC( vbp.vb[i].buffer,  PIPE_BUFFER_USAGE_GPU_READ,
+		 vbp.vb[i].offset);
+      OUT_BATCH( vbp.vb[i].max_index );
+      OUT_BATCH( vbp.vb[i].instance_data_step_rate );
+   }
+   ADVANCE_BATCH();
+   return TRUE;
+}
+
+
+
+boolean brw_upload_vertex_elements( struct brw_context *brw )
+{
+   struct brw_vertex_element_packet vep;
+
+   unsigned i;
+   unsigned nr_enabled = brw->attribs.VertexProgram->program.num_inputs;
+
+   memset(&vep, 0, sizeof(vep));
+
+   for (i = 0; i < nr_enabled; i++) 
+      vep.ve[i] = brw->vb.inputs[i];
+
+
+   vep.header.length = (1 + nr_enabled * sizeof(vep.ve[0])/4) - 2;
+   vep.header.opcode = CMD_VERTEX_ELEMENT;
+   brw_cached_batch_struct(brw, &vep, 4 + nr_enabled * sizeof(vep.ve[0]));
+
+   return TRUE;
+}
+
+boolean brw_upload_indices( struct brw_context *brw,
+                            const struct pipe_buffer *index_buffer,
+                            int ib_size, int start, int count)
+{
+   /* Emit the indexbuffer packet:
+    */
+   {
+      struct brw_indexbuffer ib;
+
+      memset(&ib, 0, sizeof(ib));
+
+      ib.header.bits.opcode = CMD_INDEX_BUFFER;
+      ib.header.bits.length = sizeof(ib)/4 - 2;
+      ib.header.bits.index_format = get_index_type(ib_size);
+      ib.header.bits.cut_index_enable = 0;
+
+
+      BEGIN_BATCH(4, 0);
+      OUT_BATCH( ib.header.dword );
+      OUT_RELOC( index_buffer, PIPE_BUFFER_USAGE_GPU_READ, start);
+      OUT_RELOC( index_buffer, PIPE_BUFFER_USAGE_GPU_READ, start + count);
+      OUT_BATCH( 0 );
+      ADVANCE_BATCH();
+   }
+   return TRUE;
+}
diff --git a/src/gallium/drivers/i965simple/brw_eu.c b/src/gallium/drivers/i965simple/brw_eu.c
new file mode 100644
index 0000000000..e2002d1821
--- /dev/null
+++ b/src/gallium/drivers/i965simple/brw_eu.c
@@ -0,0 +1,130 @@
+/*
+ Copyright (C) Intel Corp.  2006.  All Rights Reserved.
+ Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to
+ develop this 3D driver.
+ 
+ Permission is hereby granted, free of charge, to any person obtaining
+ a copy of this software and associated documentation files (the
+ "Software"), to deal in the Software without restriction, including
+ without limitation the rights to use, copy, modify, merge, publish,
+ distribute, sublicense, and/or sell copies of the Software, and to
+ permit persons to whom the Software is furnished to do so, subject to
+ the following conditions:
+ 
+ The above copyright notice and this permission notice (including the
+ next paragraph) shall be included in all copies or substantial
+ portions of the Software.
+ 
+ THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
+ IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
+ LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
+ OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ 
+ **********************************************************************/
+ /*
+  * Authors:
+  *   Keith Whitwell <keith@tungstengraphics.com>
+  */
+  
+
+#include "brw_context.h"
+#include "brw_defines.h"
+#include "brw_eu.h"
+
+
+
+/* How does predicate control work when execution_size != 8?  Do I
+ * need to test/set for 0xffff when execution_size is 16?
+ */
+void brw_set_predicate_control_flag_value( struct brw_compile *p, unsigned value )
+{
+   p->current->header.predicate_control = BRW_PREDICATE_NONE;
+
+   if (value != 0xff) {
+      if (value != p->flag_value) {
+	 brw_push_insn_state(p);
+	 brw_MOV(p, brw_flag_reg(), brw_imm_uw(value));
+	 p->flag_value = value;
+	 brw_pop_insn_state(p);
+      }
+
+      p->current->header.predicate_control = BRW_PREDICATE_NORMAL;
+   }   
+}
+
+void brw_set_predicate_control( struct brw_compile *p, unsigned pc )
+{
+   p->current->header.predicate_control = pc;
+}
+
+void brw_set_conditionalmod( struct brw_compile *p, unsigned conditional )
+{
+   p->current->header.destreg__conditonalmod = conditional;
+}
+
+void brw_set_access_mode( struct brw_compile *p, unsigned access_mode )
+{
+   p->current->header.access_mode = access_mode;
+}
+
+void brw_set_compression_control( struct brw_compile *p, boolean compression_control )
+{
+   p->current->header.compression_control = compression_control;
+}
+
+void brw_set_mask_control( struct brw_compile *p, unsigned value )
+{
+   p->current->header.mask_control = value;
+}
+
+void brw_set_saturate( struct brw_compile *p, unsigned value )
+{
+   p->current->header.saturate = value;
+}
+
+void brw_push_insn_state( struct brw_compile *p )
+{
+   assert(p->current != &p->stack[BRW_EU_MAX_INSN_STACK-1]);
+   memcpy(p->current+1, p->current, sizeof(struct brw_instruction));
+   p->current++;   
+}
+
+void brw_pop_insn_state( struct brw_compile *p )
+{
+   assert(p->current != p->stack);
+   p->current--;
+}
+
+
+/***********************************************************************
+ */
+void brw_init_compile( struct brw_compile *p )
+{
+   p->nr_insn = 0;
+   p->current = p->stack;
+   memset(p->current, 0, sizeof(p->current[0]));
+
+   /* Some defaults?
+    */
+   brw_set_mask_control(p, BRW_MASK_ENABLE); /* what does this do? */
+   brw_set_saturate(p, 0);
+   brw_set_compression_control(p, BRW_COMPRESSION_NONE);
+   brw_set_predicate_control_flag_value(p, 0xff); 
+}
+
+
+const unsigned *brw_get_program( struct brw_compile *p,
+			       unsigned *sz )
+{
+   unsigned i;
+
+   for (i = 0; i < 8; i++)
+      brw_NOP(p);
+
+   *sz = p->nr_insn * sizeof(struct brw_instruction);
+   return (const unsigned *)p->store;
+}
+
diff --git a/src/gallium/drivers/i965simple/brw_eu.h b/src/gallium/drivers/i965simple/brw_eu.h
new file mode 100644
index 0000000000..23151ae9ed
--- /dev/null
+++ b/src/gallium/drivers/i965simple/brw_eu.h
@@ -0,0 +1,888 @@
+/*
+ Copyright (C) Intel Corp.  2006.  All Rights Reserved.
+ Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to
+ develop this 3D driver.
+
+ Permission is hereby granted, free of charge, to any person obtaining
+ a copy of this software and associated documentation files (the
+ "Software"), to deal in the Software without restriction, including
+ without limitation the rights to use, copy, modify, merge, publish,
+ distribute, sublicense, and/or sell copies of the Software, and to
+ permit persons to whom the Software is furnished to do so, subject to
+ the following conditions:
+
+ The above copyright notice and this permission notice (including the
+ next paragraph) shall be included in all copies or substantial
+ portions of the Software.
+
+ THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
+ IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
+ LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
+ OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+
+ **********************************************************************/
+ /*
+  * Authors:
+  *   Keith Whitwell <keith@tungstengraphics.com>
+  */
+
+
+#ifndef BRW_EU_H
+#define BRW_EU_H
+
+#include "brw_structs.h"
+#include "brw_defines.h"
+
+#include "pipe/p_compiler.h"
+#include "pipe/p_shader_tokens.h"
+
+#define BRW_SWIZZLE4(a,b,c,d) (((a)<<0) | ((b)<<2) | ((c)<<4) | ((d)<<6))
+#define BRW_GET_SWZ(swz, idx) (((swz) >> ((idx)*2)) & 0x3)
+
+#define BRW_SWIZZLE_NOOP      BRW_SWIZZLE4(0,1,2,3)
+#define BRW_SWIZZLE_XYZW      BRW_SWIZZLE4(0,1,2,3)
+#define BRW_SWIZZLE_XXXX      BRW_SWIZZLE4(0,0,0,0)
+#define BRW_SWIZZLE_XYXY      BRW_SWIZZLE4(0,1,0,1)
+
+
+#define REG_SIZE (8*4)
+
+
+/* These aren't hardware structs, just something useful for us to pass around:
+ *
+ * Align1 operation has a lot of control over input ranges.  Used in
+ * WM programs to implement shaders decomposed into "channel serial"
+ * or "structure of array" form:
+ */
+struct brw_reg
+{
+   unsigned type:4;
+   unsigned file:2;
+   unsigned nr:8;
+   unsigned subnr:5;		/* :1 in align16 */
+   unsigned negate:1;		/* source only */
+   unsigned abs:1;		/* source only */
+   unsigned vstride:4;		/* source only */
+   unsigned width:3;		/* src only, align1 only */
+   unsigned hstride:2;   		/* src only, align1 only */
+   unsigned address_mode:1;	/* relative addressing, hopefully! */
+   unsigned pad0:1;
+
+   union {
+      struct {
+	 unsigned swizzle:8;		/* src only, align16 only */
+	 unsigned writemask:4;		/* dest only, align16 only */
+	 int  indirect_offset:10;	/* relative addressing offset */
+	 unsigned pad1:10;		/* two dwords total */
+      } bits;
+
+      float f;
+      int   d;
+      unsigned ud;
+   } dw1;
+};
+
+
+struct brw_indirect {
+   unsigned addr_subnr:4;
+   int addr_offset:10;
+   unsigned pad:18;
+};
+
+
+#define BRW_EU_MAX_INSN_STACK 5
+#define BRW_EU_MAX_INSN 1200
+
+struct brw_compile {
+   struct brw_instruction store[BRW_EU_MAX_INSN];
+   unsigned nr_insn;
+
+   /* Allow clients to push/pop instruction state:
+    */
+   struct brw_instruction stack[BRW_EU_MAX_INSN_STACK];
+   struct brw_instruction *current;
+
+   unsigned flag_value;
+   boolean single_program_flow;
+};
+
+
+
+static __inline int type_sz( unsigned type )
+{
+   switch( type ) {
+   case BRW_REGISTER_TYPE_UD:
+   case BRW_REGISTER_TYPE_D:
+   case BRW_REGISTER_TYPE_F:
+      return 4;
+   case BRW_REGISTER_TYPE_HF:
+   case BRW_REGISTER_TYPE_UW:
+   case BRW_REGISTER_TYPE_W:
+      return 2;
+   case BRW_REGISTER_TYPE_UB:
+   case BRW_REGISTER_TYPE_B:
+      return 1;
+   default:
+      return 0;
+   }
+}
+
+static __inline struct brw_reg brw_reg( unsigned file,
+					unsigned nr,
+					unsigned subnr,
+					unsigned type,
+					unsigned vstride,
+					unsigned width,
+					unsigned hstride,
+					unsigned swizzle,
+					unsigned writemask)
+{
+
+   struct brw_reg reg;
+   reg.type = type;
+   reg.file = file;
+   reg.nr = nr;
+   reg.subnr = subnr * type_sz(type);
+   reg.negate = 0;
+   reg.abs = 0;
+   reg.vstride = vstride;
+   reg.width = width;
+   reg.hstride = hstride;
+   reg.address_mode = BRW_ADDRESS_DIRECT;
+   reg.pad0 = 0;
+
+   /* Could do better: If the reg is r5.3<0;1,0>, we probably want to
+    * set swizzle and writemask to W, as the lower bits of subnr will
+    * be lost when converted to align16.  This is probably too much to
+    * keep track of as you'd want it adjusted by suboffset(), etc.
+    * Perhaps fix up when converting to align16?
+    */
+   reg.dw1.bits.swizzle = swizzle;
+   reg.dw1.bits.writemask = writemask;
+   reg.dw1.bits.indirect_offset = 0;
+   reg.dw1.bits.pad1 = 0;
+   return reg;
+}
+
+static __inline struct brw_reg brw_vec16_reg( unsigned file,
+					      unsigned nr,
+					      unsigned subnr )
+{
+   return brw_reg(file,
+		  nr,
+		  subnr,
+		  BRW_REGISTER_TYPE_F,
+		  BRW_VERTICAL_STRIDE_16,
+		  BRW_WIDTH_16,
+		  BRW_HORIZONTAL_STRIDE_1,
+		  BRW_SWIZZLE_XYZW,
+		  TGSI_WRITEMASK_XYZW);
+}
+
+static __inline struct brw_reg brw_vec8_reg( unsigned file,
+					     unsigned nr,
+					     unsigned subnr )
+{
+   return brw_reg(file,
+		  nr,
+		  subnr,
+		  BRW_REGISTER_TYPE_F,
+		  BRW_VERTICAL_STRIDE_8,
+		  BRW_WIDTH_8,
+		  BRW_HORIZONTAL_STRIDE_1,
+		  BRW_SWIZZLE_XYZW,
+		  TGSI_WRITEMASK_XYZW);
+}
+
+
+static __inline struct brw_reg brw_vec4_reg( unsigned file,
+					      unsigned nr,
+					      unsigned subnr )
+{
+   return brw_reg(file,
+		  nr,
+		  subnr,
+		  BRW_REGISTER_TYPE_F,
+		  BRW_VERTICAL_STRIDE_4,
+		  BRW_WIDTH_4,
+		  BRW_HORIZONTAL_STRIDE_1,
+		  BRW_SWIZZLE_XYZW,
+		  TGSI_WRITEMASK_XYZW);
+}
+
+
+static __inline struct brw_reg brw_vec2_reg( unsigned file,
+					      unsigned nr,
+					      unsigned subnr )
+{
+   return brw_reg(file,
+		  nr,
+		  subnr,
+		  BRW_REGISTER_TYPE_F,
+		  BRW_VERTICAL_STRIDE_2,
+		  BRW_WIDTH_2,
+		  BRW_HORIZONTAL_STRIDE_1,
+		  BRW_SWIZZLE_XYXY,
+		  TGSI_WRITEMASK_XY);
+}
+
+static __inline struct brw_reg brw_vec1_reg( unsigned file,
+					     unsigned nr,
+					     unsigned subnr )
+{
+   return brw_reg(file,
+		  nr,
+		  subnr,
+		  BRW_REGISTER_TYPE_F,
+		  BRW_VERTICAL_STRIDE_0,
+		  BRW_WIDTH_1,
+		  BRW_HORIZONTAL_STRIDE_0,
+		  BRW_SWIZZLE_XXXX,
+		  TGSI_WRITEMASK_X);
+}
+
+
+static __inline struct brw_reg retype( struct brw_reg reg,
+				       unsigned type )
+{
+   reg.type = type;
+   return reg;
+}
+
+static __inline struct brw_reg suboffset( struct brw_reg reg,
+					  unsigned delta )
+{
+   reg.subnr += delta * type_sz(reg.type);
+   return reg;
+}
+
+
+static __inline struct brw_reg offset( struct brw_reg reg,
+				       unsigned delta )
+{
+   reg.nr += delta;
+   return reg;
+}
+
+
+static __inline struct brw_reg byte_offset( struct brw_reg reg,
+					    unsigned bytes )
+{
+   unsigned newoffset = reg.nr * REG_SIZE + reg.subnr + bytes;
+   reg.nr = newoffset / REG_SIZE;
+   reg.subnr = newoffset % REG_SIZE;
+   return reg;
+}
+
+
+static __inline struct brw_reg brw_uw16_reg( unsigned file,
+					     unsigned nr,
+					     unsigned subnr )
+{
+   return suboffset(retype(brw_vec16_reg(file, nr, 0), BRW_REGISTER_TYPE_UW), subnr);
+}
+
+static __inline struct brw_reg brw_uw8_reg( unsigned file,
+					    unsigned nr,
+					    unsigned subnr )
+{
+   return suboffset(retype(brw_vec8_reg(file, nr, 0), BRW_REGISTER_TYPE_UW), subnr);
+}
+
+static __inline struct brw_reg brw_uw1_reg( unsigned file,
+					    unsigned nr,
+					    unsigned subnr )
+{
+   return suboffset(retype(brw_vec1_reg(file, nr, 0), BRW_REGISTER_TYPE_UW), subnr);
+}
+
+static __inline struct brw_reg brw_imm_reg( unsigned type )
+{
+   return brw_reg( BRW_IMMEDIATE_VALUE,
+		   0,
+		   0,
+		   type,
+		   BRW_VERTICAL_STRIDE_0,
+		   BRW_WIDTH_1,
+		   BRW_HORIZONTAL_STRIDE_0,
+		   0,
+		   0);
+}
+
+static __inline struct brw_reg brw_imm_f( float f )
+{
+   struct brw_reg imm = brw_imm_reg(BRW_REGISTER_TYPE_F);
+   imm.dw1.f = f;
+   return imm;
+}
+
+static __inline struct brw_reg brw_imm_d( int d )
+{
+   struct brw_reg imm = brw_imm_reg(BRW_REGISTER_TYPE_D);
+   imm.dw1.d = d;
+   return imm;
+}
+
+static __inline struct brw_reg brw_imm_ud( unsigned ud )
+{
+   struct brw_reg imm = brw_imm_reg(BRW_REGISTER_TYPE_UD);
+   imm.dw1.ud = ud;
+   return imm;
+}
+
+static __inline struct brw_reg brw_imm_uw( ushort uw )
+{
+   struct brw_reg imm = brw_imm_reg(BRW_REGISTER_TYPE_UW);
+   imm.dw1.ud = uw;
+   return imm;
+}
+
+static __inline struct brw_reg brw_imm_w( short w )
+{
+   struct brw_reg imm = brw_imm_reg(BRW_REGISTER_TYPE_W);
+   imm.dw1.d = w;
+   return imm;
+}
+
+/* brw_imm_b and brw_imm_ub aren't supported by hardware - the type
+ * numbers alias with _V and _VF below:
+ */
+
+/* Vector of eight signed half-byte values:
+ */
+static __inline struct brw_reg brw_imm_v( unsigned v )
+{
+   struct brw_reg imm = brw_imm_reg(BRW_REGISTER_TYPE_V);
+   imm.vstride = BRW_VERTICAL_STRIDE_0;
+   imm.width = BRW_WIDTH_8;
+   imm.hstride = BRW_HORIZONTAL_STRIDE_1;
+   imm.dw1.ud = v;
+   return imm;
+}
+
+/* Vector of four 8-bit float values:
+ */
+static __inline struct brw_reg brw_imm_vf( unsigned v )
+{
+   struct brw_reg imm = brw_imm_reg(BRW_REGISTER_TYPE_VF);
+   imm.vstride = BRW_VERTICAL_STRIDE_0;
+   imm.width = BRW_WIDTH_4;
+   imm.hstride = BRW_HORIZONTAL_STRIDE_1;
+   imm.dw1.ud = v;
+   return imm;
+}
+
+#define VF_ZERO 0x0
+#define VF_ONE  0x30
+#define VF_NEG  (1<<7)
+
+static __inline struct brw_reg brw_imm_vf4( unsigned v0,
+					    unsigned v1,
+					    unsigned v2,
+					    unsigned v3)
+{
+   struct brw_reg imm = brw_imm_reg(BRW_REGISTER_TYPE_VF);
+   imm.vstride = BRW_VERTICAL_STRIDE_0;
+   imm.width = BRW_WIDTH_4;
+   imm.hstride = BRW_HORIZONTAL_STRIDE_1;
+   imm.dw1.ud = ((v0 << 0) |
+		 (v1 << 8) |
+		 (v2 << 16) |
+		 (v3 << 24));
+   return imm;
+}
+
+
+static __inline struct brw_reg brw_address( struct brw_reg reg )
+{
+   return brw_imm_uw(reg.nr * REG_SIZE + reg.subnr);
+}
+
+
+static __inline struct brw_reg brw_vec1_grf( unsigned nr,
+					       unsigned subnr )
+{
+   return brw_vec1_reg(BRW_GENERAL_REGISTER_FILE, nr, subnr);
+}
+
+static __inline struct brw_reg brw_vec8_grf( unsigned nr,
+					     unsigned subnr )
+{
+   return brw_vec8_reg(BRW_GENERAL_REGISTER_FILE, nr, subnr);
+}
+
+static __inline struct brw_reg brw_vec4_grf( unsigned nr,
+					     unsigned subnr )
+{
+   return brw_vec4_reg(BRW_GENERAL_REGISTER_FILE, nr, subnr);
+}
+
+
+static __inline struct brw_reg brw_vec2_grf( unsigned nr,
+					     unsigned subnr )
+{
+   return brw_vec2_reg(BRW_GENERAL_REGISTER_FILE, nr, subnr);
+}
+
+static __inline struct brw_reg brw_uw8_grf( unsigned nr,
+					    unsigned subnr )
+{
+   return brw_uw8_reg(BRW_GENERAL_REGISTER_FILE, nr, subnr);
+}
+
+static __inline struct brw_reg brw_null_reg( void )
+{
+   return brw_vec8_reg(BRW_ARCHITECTURE_REGISTER_FILE,
+		       BRW_ARF_NULL,
+		       0);
+}
+
+static __inline struct brw_reg brw_address_reg( unsigned subnr )
+{
+   return brw_uw1_reg(BRW_ARCHITECTURE_REGISTER_FILE,
+		      BRW_ARF_ADDRESS,
+		      subnr);
+}
+
+/* If/else instructions break in align16 mode if writemask & swizzle
+ * aren't xyzw.  This goes against the convention for other scalar
+ * regs:
+ */
+static __inline struct brw_reg brw_ip_reg( void )
+{
+   return brw_reg(BRW_ARCHITECTURE_REGISTER_FILE,
+		  BRW_ARF_IP,
+		  0,
+		  BRW_REGISTER_TYPE_UD,
+		  BRW_VERTICAL_STRIDE_4, /* ? */
+		  BRW_WIDTH_1,
+		  BRW_HORIZONTAL_STRIDE_0,
+		  BRW_SWIZZLE_XYZW, /* NOTE! */
+		  TGSI_WRITEMASK_XYZW); /* NOTE! */
+}
+
+static __inline struct brw_reg brw_acc_reg( void )
+{
+   return brw_vec8_reg(BRW_ARCHITECTURE_REGISTER_FILE,
+		       BRW_ARF_ACCUMULATOR,
+		       0);
+}
+
+
+static __inline struct brw_reg brw_flag_reg( void )
+{
+   return brw_uw1_reg(BRW_ARCHITECTURE_REGISTER_FILE,
+		      BRW_ARF_FLAG,
+		      0);
+}
+
+
+static __inline struct brw_reg brw_mask_reg( unsigned subnr )
+{
+   return brw_uw1_reg(BRW_ARCHITECTURE_REGISTER_FILE,
+		      BRW_ARF_MASK,
+		      subnr);
+}
+
+static __inline struct brw_reg brw_message_reg( unsigned nr )
+{
+   return brw_vec8_reg(BRW_MESSAGE_REGISTER_FILE,
+		       nr,
+		       0);
+}
+
+
+
+
+/* This is almost always called with a numeric constant argument, so
+ * make things easy to evaluate at compile time:
+ */
+static __inline unsigned cvt( unsigned val )
+{
+   switch (val) {
+   case 0: return 0;
+   case 1: return 1;
+   case 2: return 2;
+   case 4: return 3;
+   case 8: return 4;
+   case 16: return 5;
+   case 32: return 6;
+   }
+   return 0;
+}
+
+static __inline struct brw_reg stride( struct brw_reg reg,
+				       unsigned vstride,
+				       unsigned width,
+				       unsigned hstride )
+{
+
+   reg.vstride = cvt(vstride);
+   reg.width = cvt(width) - 1;
+   reg.hstride = cvt(hstride);
+   return reg;
+}
+
+static __inline struct brw_reg vec16( struct brw_reg reg )
+{
+   return stride(reg, 16,16,1);
+}
+
+static __inline struct brw_reg vec8( struct brw_reg reg )
+{
+   return stride(reg, 8,8,1);
+}
+
+static __inline struct brw_reg vec4( struct brw_reg reg )
+{
+   return stride(reg, 4,4,1);
+}
+
+static __inline struct brw_reg vec2( struct brw_reg reg )
+{
+   return stride(reg, 2,2,1);
+}
+
+static __inline struct brw_reg vec1( struct brw_reg reg )
+{
+   return stride(reg, 0,1,0);
+}
+
+static __inline struct brw_reg get_element( struct brw_reg reg, unsigned elt )
+{
+   return vec1(suboffset(reg, elt));
+}
+
+static __inline struct brw_reg get_element_ud( struct brw_reg reg, unsigned elt )
+{
+   return vec1(suboffset(retype(reg, BRW_REGISTER_TYPE_UD), elt));
+}
+
+
+static __inline struct brw_reg brw_swizzle( struct brw_reg reg,
+					    unsigned x,
+					    unsigned y,
+					    unsigned z,
+					    unsigned w)
+{
+   reg.dw1.bits.swizzle = BRW_SWIZZLE4(BRW_GET_SWZ(reg.dw1.bits.swizzle, x),
+				       BRW_GET_SWZ(reg.dw1.bits.swizzle, y),
+				       BRW_GET_SWZ(reg.dw1.bits.swizzle, z),
+				       BRW_GET_SWZ(reg.dw1.bits.swizzle, w));
+   return reg;
+}
+
+
+static __inline struct brw_reg brw_swizzle1( struct brw_reg reg,
+					     unsigned x )
+{
+   return brw_swizzle(reg, x, x, x, x);
+}
+
+static __inline struct brw_reg brw_writemask( struct brw_reg reg,
+					      unsigned mask )
+{
+   reg.dw1.bits.writemask &= mask;
+   return reg;
+}
+
+static __inline struct brw_reg brw_set_writemask( struct brw_reg reg,
+						  unsigned mask )
+{
+   reg.dw1.bits.writemask = mask;
+   return reg;
+}
+
+static __inline struct brw_reg negate( struct brw_reg reg )
+{
+   reg.negate ^= 1;
+   return reg;
+}
+
+static __inline struct brw_reg brw_abs( struct brw_reg reg )
+{
+   reg.abs = 1;
+   return reg;
+}
+
+/***********************************************************************
+ */
+static __inline struct brw_reg brw_vec4_indirect( unsigned subnr,
+						  int offset )
+{
+   struct brw_reg reg =  brw_vec4_grf(0, 0);
+   reg.subnr = subnr;
+   reg.address_mode = BRW_ADDRESS_REGISTER_INDIRECT_REGISTER;
+   reg.dw1.bits.indirect_offset = offset;
+   return reg;
+}
+
+static __inline struct brw_reg brw_vec1_indirect( unsigned subnr,
+						  int offset )
+{
+   struct brw_reg reg =  brw_vec1_grf(0, 0);
+   reg.subnr = subnr;
+   reg.address_mode = BRW_ADDRESS_REGISTER_INDIRECT_REGISTER;
+   reg.dw1.bits.indirect_offset = offset;
+   return reg;
+}
+
+static __inline struct brw_reg deref_4f(struct brw_indirect ptr, int offset)
+{
+   return brw_vec4_indirect(ptr.addr_subnr, ptr.addr_offset + offset);
+}
+
+static __inline struct brw_reg deref_1f(struct brw_indirect ptr, int offset)
+{
+   return brw_vec1_indirect(ptr.addr_subnr, ptr.addr_offset + offset);
+}
+
+static __inline struct brw_reg deref_4b(struct brw_indirect ptr, int offset)
+{
+   return retype(deref_4f(ptr, offset), BRW_REGISTER_TYPE_B);
+}
+
+static __inline struct brw_reg deref_1uw(struct brw_indirect ptr, int offset)
+{
+   return retype(deref_1f(ptr, offset), BRW_REGISTER_TYPE_UW);
+}
+
+static __inline struct brw_reg deref_1ud(struct brw_indirect ptr, int offset)
+{
+   return retype(deref_1f(ptr, offset), BRW_REGISTER_TYPE_UD);
+}
+
+static __inline struct brw_reg get_addr_reg(struct brw_indirect ptr)
+{
+   return brw_address_reg(ptr.addr_subnr);
+}
+
+static __inline struct brw_indirect brw_indirect_offset( struct brw_indirect ptr, int offset )
+{
+   ptr.addr_offset += offset;
+   return ptr;
+}
+
+static __inline struct brw_indirect brw_indirect( unsigned addr_subnr, int offset )
+{
+   struct brw_indirect ptr;
+   ptr.addr_subnr = addr_subnr;
+   ptr.addr_offset = offset;
+   ptr.pad = 0;
+   return ptr;
+}
+
+static __inline struct brw_instruction *current_insn( struct brw_compile *p)
+{
+	return &p->store[p->nr_insn];
+}
+
+void brw_pop_insn_state( struct brw_compile *p );
+void brw_push_insn_state( struct brw_compile *p );
+void brw_set_mask_control( struct brw_compile *p, unsigned value );
+void brw_set_saturate( struct brw_compile *p, unsigned value );
+void brw_set_access_mode( struct brw_compile *p, unsigned access_mode );
+void brw_set_compression_control( struct brw_compile *p, boolean control );
+void brw_set_predicate_control_flag_value( struct brw_compile *p, unsigned value );
+void brw_set_predicate_control( struct brw_compile *p, unsigned pc );
+void brw_set_conditionalmod( struct brw_compile *p, unsigned conditional );
+
+void brw_init_compile( struct brw_compile *p );
+const unsigned *brw_get_program( struct brw_compile *p, unsigned *sz );
+
+
+struct brw_instruction *brw_alu1( struct brw_compile *p,
+				  unsigned opcode,
+				  struct brw_reg dest,
+				  struct brw_reg src );
+
+struct brw_instruction *brw_alu2(struct brw_compile *p,
+				 unsigned opcode,
+				 struct brw_reg dest,
+				 struct brw_reg src0,
+				 struct brw_reg src1 );
+
+/* Helpers for regular instructions:
+ */
+#define ALU1(OP)					\
+struct brw_instruction *brw_##OP(struct brw_compile *p,	\
+	      struct brw_reg dest,			\
+	      struct brw_reg src0);
+
+#define ALU2(OP)					\
+struct brw_instruction *brw_##OP(struct brw_compile *p,	\
+	      struct brw_reg dest,			\
+	      struct brw_reg src0,			\
+	      struct brw_reg src1);
+
+ALU1(MOV)
+ALU2(SEL)
+ALU1(NOT)
+ALU2(AND)
+ALU2(OR)
+ALU2(XOR)
+ALU2(SHR)
+ALU2(SHL)
+ALU2(RSR)
+ALU2(RSL)
+ALU2(ASR)
+ALU2(JMPI)
+ALU2(ADD)
+ALU2(MUL)
+ALU1(FRC)
+ALU1(RNDD)
+ALU2(MAC)
+ALU2(MACH)
+ALU1(LZD)
+ALU2(DP4)
+ALU2(DPH)
+ALU2(DP3)
+ALU2(DP2)
+ALU2(LINE)
+
+#undef ALU1
+#undef ALU2
+
+
+
+/* Helpers for SEND instruction:
+ */
+void brw_urb_WRITE(struct brw_compile *p,
+		   struct brw_reg dest,
+		   unsigned msg_reg_nr,
+		   struct brw_reg src0,
+		   boolean allocate,
+		   boolean used,
+		   unsigned msg_length,
+		   unsigned response_length,
+		   boolean eot,
+		   boolean writes_complete,
+		   unsigned offset,
+		   unsigned swizzle);
+
+void brw_fb_WRITE(struct brw_compile *p,
+		   struct brw_reg dest,
+		   unsigned msg_reg_nr,
+		   struct brw_reg src0,
+		   unsigned binding_table_index,
+		   unsigned msg_length,
+		   unsigned response_length,
+		   boolean eot);
+
+void brw_SAMPLE(struct brw_compile *p,
+		struct brw_reg dest,
+		unsigned msg_reg_nr,
+		struct brw_reg src0,
+		unsigned binding_table_index,
+		unsigned sampler,
+		unsigned writemask,
+		unsigned msg_type,
+		unsigned response_length,
+		unsigned msg_length,
+		boolean eot);
+
+void brw_math_16( struct brw_compile *p,
+		  struct brw_reg dest,
+		  unsigned function,
+		  unsigned saturate,
+		  unsigned msg_reg_nr,
+		  struct brw_reg src,
+		  unsigned precision );
+
+void brw_math( struct brw_compile *p,
+	       struct brw_reg dest,
+	       unsigned function,
+	       unsigned saturate,
+	       unsigned msg_reg_nr,
+	       struct brw_reg src,
+	       unsigned data_type,
+	       unsigned precision );
+
+void brw_dp_READ_16( struct brw_compile *p,
+		     struct brw_reg dest,
+		     unsigned msg_reg_nr,
+		     unsigned scratch_offset );
+
+void brw_dp_WRITE_16( struct brw_compile *p,
+		      struct brw_reg src,
+		      unsigned msg_reg_nr,
+		      unsigned scratch_offset );
+
+/* If/else/endif.  Works by manipulating the execution flags on each
+ * channel.
+ */
+struct brw_instruction *brw_IF(struct brw_compile *p,
+			       unsigned execute_size);
+
+struct brw_instruction *brw_ELSE(struct brw_compile *p,
+				 struct brw_instruction *if_insn);
+
+void brw_ENDIF(struct brw_compile *p,
+	       struct brw_instruction *if_or_else_insn);
+
+
+/* DO/WHILE loops:
+ */
+struct brw_instruction *brw_DO(struct brw_compile *p,
+			       unsigned execute_size);
+
+struct brw_instruction *brw_WHILE(struct brw_compile *p,
+	       struct brw_instruction *patch_insn);
+
+struct brw_instruction *brw_BREAK(struct brw_compile *p);
+struct brw_instruction *brw_CONT(struct brw_compile *p);
+/* Forward jumps:
+ */
+void brw_land_fwd_jump(struct brw_compile *p,
+		       struct brw_instruction *jmp_insn);
+
+
+
+void brw_NOP(struct brw_compile *p);
+
+/* Special case: there is never a destination, execution size will be
+ * taken from src0:
+ */
+void brw_CMP(struct brw_compile *p,
+	     struct brw_reg dest,
+	     unsigned conditional,
+	     struct brw_reg src0,
+	     struct brw_reg src1);
+
+void brw_print_reg( struct brw_reg reg );
+
+
+/***********************************************************************
+ * brw_eu_util.c:
+ */
+
+void brw_copy_indirect_to_indirect(struct brw_compile *p,
+				   struct brw_indirect dst_ptr,
+				   struct brw_indirect src_ptr,
+				   unsigned count);
+
+void brw_copy_from_indirect(struct brw_compile *p,
+			    struct brw_reg dst,
+			    struct brw_indirect ptr,
+			    unsigned count);
+
+void brw_copy4(struct brw_compile *p,
+	       struct brw_reg dst,
+	       struct brw_reg src,
+	       unsigned count);
+
+void brw_copy8(struct brw_compile *p,
+	       struct brw_reg dst,
+	       struct brw_reg src,
+	       unsigned count);
+
+void brw_math_invert( struct brw_compile *p,
+		      struct brw_reg dst,
+		      struct brw_reg src);
+
+void brw_set_src1( struct brw_instruction *insn,
+                          struct brw_reg reg );
+#endif
diff --git a/src/gallium/drivers/i965simple/brw_eu_debug.c b/src/gallium/drivers/i965simple/brw_eu_debug.c
new file mode 100644
index 0000000000..4a94ddefa6
--- /dev/null
+++ b/src/gallium/drivers/i965simple/brw_eu_debug.c
@@ -0,0 +1,90 @@
+/*
+ Copyright (C) Intel Corp.  2006.  All Rights Reserved.
+ Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to
+ develop this 3D driver.
+ 
+ Permission is hereby granted, free of charge, to any person obtaining
+ a copy of this software and associated documentation files (the
+ "Software"), to deal in the Software without restriction, including
+ without limitation the rights to use, copy, modify, merge, publish,
+ distribute, sublicense, and/or sell copies of the Software, and to
+ permit persons to whom the Software is furnished to do so, subject to
+ the following conditions:
+ 
+ The above copyright notice and this permission notice (including the
+ next paragraph) shall be included in all copies or substantial
+ portions of the Software.
+ 
+ THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
+ IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
+ LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
+ OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ 
+ **********************************************************************/
+ /*
+  * Authors:
+  *   Keith Whitwell <keith@tungstengraphics.com>
+  */
+    
+
+#include "pipe/p_debug.h"
+
+#include "brw_eu.h"
+
+void brw_print_reg( struct brw_reg hwreg )
+{
+   static const char *file[] = {
+      "arf",
+      "grf",
+      "msg",
+      "imm"
+   };
+
+   static const char *type[] = {
+      "ud",
+      "d",
+      "uw",
+      "w",
+      "ub",
+      "vf",
+      "hf",
+      "f"
+   };
+
+   debug_printf("%s%s", 
+		hwreg.abs ? "abs/" : "",
+		hwreg.negate ? "-" : "");
+     
+   if (hwreg.file == BRW_GENERAL_REGISTER_FILE &&
+       hwreg.nr % 2 == 0 &&
+       hwreg.subnr == 0 &&
+       hwreg.vstride == BRW_VERTICAL_STRIDE_8 &&
+       hwreg.width == BRW_WIDTH_8 &&
+       hwreg.hstride == BRW_HORIZONTAL_STRIDE_1 &&
+       hwreg.type == BRW_REGISTER_TYPE_F) {
+      debug_printf("vec%d", hwreg.nr);
+   }
+   else if (hwreg.file == BRW_GENERAL_REGISTER_FILE &&
+	    hwreg.vstride == BRW_VERTICAL_STRIDE_0 &&
+	    hwreg.width == BRW_WIDTH_1 &&
+	    hwreg.hstride == BRW_HORIZONTAL_STRIDE_0 &&
+	    hwreg.type == BRW_REGISTER_TYPE_F) {      
+      debug_printf("scl%d.%d", hwreg.nr, hwreg.subnr / 4);
+   }
+   else {
+      debug_printf("%s%d.%d<%d;%d,%d>:%s", 
+		   file[hwreg.file],
+		   hwreg.nr,
+		   hwreg.subnr / type_sz(hwreg.type),
+		   hwreg.vstride ? (1<<(hwreg.vstride-1)) : 0,
+		   1<<hwreg.width,
+		   hwreg.hstride ? (1<<(hwreg.hstride-1)) : 0,		
+		   type[hwreg.type]);
+   }
+}
+
+
+
diff --git a/src/gallium/drivers/i965simple/brw_eu_emit.c b/src/gallium/drivers/i965simple/brw_eu_emit.c
new file mode 100644
index 0000000000..400a80b6fb
--- /dev/null
+++ b/src/gallium/drivers/i965simple/brw_eu_emit.c
@@ -0,0 +1,1080 @@
+/*
+ Copyright (C) Intel Corp.  2006.  All Rights Reserved.
+ Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to
+ develop this 3D driver.
+
+ Permission is hereby granted, free of charge, to any person obtaining
+ a copy of this software and associated documentation files (the
+ "Software"), to deal in the Software without restriction, including
+ without limitation the rights to use, copy, modify, merge, publish,
+ distribute, sublicense, and/or sell copies of the Software, and to
+ permit persons to whom the Software is furnished to do so, subject to
+ the following conditions:
+
+ The above copyright notice and this permission notice (including the
+ next paragraph) shall be included in all copies or substantial
+ portions of the Software.
+
+ THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
+ IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
+ LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
+ OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+
+ **********************************************************************/
+ /*
+  * Authors:
+  *   Keith Whitwell <keith@tungstengraphics.com>
+  */
+
+
+#include "brw_context.h"
+#include "brw_defines.h"
+#include "brw_eu.h"
+
+
+
+
+/***********************************************************************
+ * Internal helper for constructing instructions
+ */
+
+static void guess_execution_size( struct brw_instruction *insn,
+				  struct brw_reg reg )
+{
+   if (reg.width == BRW_WIDTH_8 &&
+       insn->header.compression_control == BRW_COMPRESSION_COMPRESSED)
+      insn->header.execution_size = BRW_EXECUTE_16;
+   else
+      insn->header.execution_size = reg.width;	/* note - definitions are compatible */
+}
+
+
+static void brw_set_dest( struct brw_instruction *insn,
+			  struct brw_reg dest )
+{
+   insn->bits1.da1.dest_reg_file = dest.file;
+   insn->bits1.da1.dest_reg_type = dest.type;
+   insn->bits1.da1.dest_address_mode = dest.address_mode;
+
+   if (dest.address_mode == BRW_ADDRESS_DIRECT) {
+      insn->bits1.da1.dest_reg_nr = dest.nr;
+
+      if (insn->header.access_mode == BRW_ALIGN_1) {
+	 insn->bits1.da1.dest_subreg_nr = dest.subnr;
+	 insn->bits1.da1.dest_horiz_stride = BRW_HORIZONTAL_STRIDE_1;
+      }
+      else {
+	 insn->bits1.da16.dest_subreg_nr = dest.subnr / 16;
+	 insn->bits1.da16.dest_writemask = dest.dw1.bits.writemask;
+      }
+   }
+   else {
+      insn->bits1.ia1.dest_subreg_nr = dest.subnr;
+
+      /* These are different sizes in align1 vs align16:
+       */
+      if (insn->header.access_mode == BRW_ALIGN_1) {
+	 insn->bits1.ia1.dest_indirect_offset = dest.dw1.bits.indirect_offset;
+	 insn->bits1.ia1.dest_horiz_stride = BRW_HORIZONTAL_STRIDE_1;
+      }
+      else {
+	 insn->bits1.ia16.dest_indirect_offset = dest.dw1.bits.indirect_offset;
+      }
+   }
+
+   /* NEW: Set the execution size based on dest.width and
+    * insn->compression_control:
+    */
+   guess_execution_size(insn, dest);
+}
+
+static void brw_set_src0( struct brw_instruction *insn,
+		      struct brw_reg reg )
+{
+   assert(reg.file != BRW_MESSAGE_REGISTER_FILE);
+
+   insn->bits1.da1.src0_reg_file = reg.file;
+   insn->bits1.da1.src0_reg_type = reg.type;
+   insn->bits2.da1.src0_abs = reg.abs;
+   insn->bits2.da1.src0_negate = reg.negate;
+   insn->bits2.da1.src0_address_mode = reg.address_mode;
+
+   if (reg.file == BRW_IMMEDIATE_VALUE) {
+      insn->bits3.ud = reg.dw1.ud;
+
+      /* Required to set some fields in src1 as well:
+       */
+      insn->bits1.da1.src1_reg_file = 0; /* arf */
+      insn->bits1.da1.src1_reg_type = reg.type;
+   }
+   else
+   {
+      if (reg.address_mode == BRW_ADDRESS_DIRECT) {
+	 if (insn->header.access_mode == BRW_ALIGN_1) {
+	    insn->bits2.da1.src0_subreg_nr = reg.subnr;
+	    insn->bits2.da1.src0_reg_nr = reg.nr;
+	 }
+	 else {
+	    insn->bits2.da16.src0_subreg_nr = reg.subnr / 16;
+	    insn->bits2.da16.src0_reg_nr = reg.nr;
+	 }
+      }
+      else {
+	 insn->bits2.ia1.src0_subreg_nr = reg.subnr;
+
+	 if (insn->header.access_mode == BRW_ALIGN_1) {
+	    insn->bits2.ia1.src0_indirect_offset = reg.dw1.bits.indirect_offset;
+	 }
+	 else {
+	    insn->bits2.ia16.src0_subreg_nr = reg.dw1.bits.indirect_offset;
+	 }
+      }
+
+      if (insn->header.access_mode == BRW_ALIGN_1) {
+	 if (reg.width == BRW_WIDTH_1 &&
+	     insn->header.execution_size == BRW_EXECUTE_1) {
+	    insn->bits2.da1.src0_horiz_stride = BRW_HORIZONTAL_STRIDE_0;
+	    insn->bits2.da1.src0_width = BRW_WIDTH_1;
+	    insn->bits2.da1.src0_vert_stride = BRW_VERTICAL_STRIDE_0;
+	 }
+	 else {
+	    insn->bits2.da1.src0_horiz_stride = reg.hstride;
+	    insn->bits2.da1.src0_width = reg.width;
+	    insn->bits2.da1.src0_vert_stride = reg.vstride;
+	 }
+      }
+      else {
+	 insn->bits2.da16.src0_swz_x = BRW_GET_SWZ(reg.dw1.bits.swizzle, BRW_CHANNEL_X);
+	 insn->bits2.da16.src0_swz_y = BRW_GET_SWZ(reg.dw1.bits.swizzle, BRW_CHANNEL_Y);
+	 insn->bits2.da16.src0_swz_z = BRW_GET_SWZ(reg.dw1.bits.swizzle, BRW_CHANNEL_Z);
+	 insn->bits2.da16.src0_swz_w = BRW_GET_SWZ(reg.dw1.bits.swizzle, BRW_CHANNEL_W);
+
+	 /* This is an oddity of the fact we're using the same
+	  * descriptions for registers in align_16 as align_1:
+	  */
+	 if (reg.vstride == BRW_VERTICAL_STRIDE_8)
+	    insn->bits2.da16.src0_vert_stride = BRW_VERTICAL_STRIDE_4;
+	 else
+	    insn->bits2.da16.src0_vert_stride = reg.vstride;
+      }
+   }
+}
+
+
+void brw_set_src1( struct brw_instruction *insn,
+			  struct brw_reg reg )
+{
+   assert(reg.file != BRW_MESSAGE_REGISTER_FILE);
+
+   insn->bits1.da1.src1_reg_file = reg.file;
+   insn->bits1.da1.src1_reg_type = reg.type;
+   insn->bits3.da1.src1_abs = reg.abs;
+   insn->bits3.da1.src1_negate = reg.negate;
+
+   /* Only src1 can be immediate in two-argument instructions.
+    */
+   assert(insn->bits1.da1.src0_reg_file != BRW_IMMEDIATE_VALUE);
+
+   if (reg.file == BRW_IMMEDIATE_VALUE) {
+      insn->bits3.ud = reg.dw1.ud;
+   }
+   else {
+      /* This is a hardware restriction, which may or may not be lifted
+       * in the future:
+       */
+      assert (reg.address_mode == BRW_ADDRESS_DIRECT);
+      //assert (reg.file == BRW_GENERAL_REGISTER_FILE);
+
+      if (insn->header.access_mode == BRW_ALIGN_1) {
+	 insn->bits3.da1.src1_subreg_nr = reg.subnr;
+	 insn->bits3.da1.src1_reg_nr = reg.nr;
+      }
+      else {
+	 insn->bits3.da16.src1_subreg_nr = reg.subnr / 16;
+	 insn->bits3.da16.src1_reg_nr = reg.nr;
+      }
+
+      if (insn->header.access_mode == BRW_ALIGN_1) {
+	 if (reg.width == BRW_WIDTH_1 &&
+	     insn->header.execution_size == BRW_EXECUTE_1) {
+	    insn->bits3.da1.src1_horiz_stride = BRW_HORIZONTAL_STRIDE_0;
+	    insn->bits3.da1.src1_width = BRW_WIDTH_1;
+	    insn->bits3.da1.src1_vert_stride = BRW_VERTICAL_STRIDE_0;
+	 }
+	 else {
+	    insn->bits3.da1.src1_horiz_stride = reg.hstride;
+	    insn->bits3.da1.src1_width = reg.width;
+	    insn->bits3.da1.src1_vert_stride = reg.vstride;
+	 }
+      }
+      else {
+	 insn->bits3.da16.src1_swz_x = BRW_GET_SWZ(reg.dw1.bits.swizzle, BRW_CHANNEL_X);
+	 insn->bits3.da16.src1_swz_y = BRW_GET_SWZ(reg.dw1.bits.swizzle, BRW_CHANNEL_Y);
+	 insn->bits3.da16.src1_swz_z = BRW_GET_SWZ(reg.dw1.bits.swizzle, BRW_CHANNEL_Z);
+	 insn->bits3.da16.src1_swz_w = BRW_GET_SWZ(reg.dw1.bits.swizzle, BRW_CHANNEL_W);
+
+	 /* This is an oddity of the fact we're using the same
+	  * descriptions for registers in align_16 as align_1:
+	  */
+	 if (reg.vstride == BRW_VERTICAL_STRIDE_8)
+	    insn->bits3.da16.src1_vert_stride = BRW_VERTICAL_STRIDE_4;
+	 else
+	    insn->bits3.da16.src1_vert_stride = reg.vstride;
+      }
+   }
+}
+
+
+
+static void brw_set_math_message( struct brw_instruction *insn,
+				  unsigned msg_length,
+				  unsigned response_length,
+				  unsigned function,
+				  unsigned integer_type,
+				  boolean low_precision,
+				  boolean saturate,
+				  unsigned dataType )
+{
+   brw_set_src1(insn, brw_imm_d(0));
+
+   insn->bits3.math.function = function;
+   insn->bits3.math.int_type = integer_type;
+   insn->bits3.math.precision = low_precision;
+   insn->bits3.math.saturate = saturate;
+   insn->bits3.math.data_type = dataType;
+   insn->bits3.math.response_length = response_length;
+   insn->bits3.math.msg_length = msg_length;
+   insn->bits3.math.msg_target = BRW_MESSAGE_TARGET_MATH;
+   insn->bits3.math.end_of_thread = 0;
+}
+
+static void brw_set_urb_message( struct brw_instruction *insn,
+				 boolean allocate,
+				 boolean used,
+				 unsigned msg_length,
+				 unsigned response_length,
+				 boolean end_of_thread,
+				 boolean complete,
+				 unsigned offset,
+				 unsigned swizzle_control )
+{
+   brw_set_src1(insn, brw_imm_d(0));
+
+   insn->bits3.urb.opcode = 0;	/* ? */
+   insn->bits3.urb.offset = offset;
+   insn->bits3.urb.swizzle_control = swizzle_control;
+   insn->bits3.urb.allocate = allocate;
+   insn->bits3.urb.used = used;	/* ? */
+   insn->bits3.urb.complete = complete;
+   insn->bits3.urb.response_length = response_length;
+   insn->bits3.urb.msg_length = msg_length;
+   insn->bits3.urb.msg_target = BRW_MESSAGE_TARGET_URB;
+   insn->bits3.urb.end_of_thread = end_of_thread;
+}
+
+static void brw_set_dp_write_message( struct brw_instruction *insn,
+				      unsigned binding_table_index,
+				      unsigned msg_control,
+				      unsigned msg_type,
+				      unsigned msg_length,
+				      unsigned pixel_scoreboard_clear,
+				      unsigned response_length,
+				      unsigned end_of_thread )
+{
+   brw_set_src1(insn, brw_imm_d(0));
+
+   insn->bits3.dp_write.binding_table_index = binding_table_index;
+   insn->bits3.dp_write.msg_control = msg_control;
+   insn->bits3.dp_write.pixel_scoreboard_clear = pixel_scoreboard_clear;
+   insn->bits3.dp_write.msg_type = msg_type;
+   insn->bits3.dp_write.send_commit_msg = 0;
+   insn->bits3.dp_write.response_length = response_length;
+   insn->bits3.dp_write.msg_length = msg_length;
+   insn->bits3.dp_write.msg_target = BRW_MESSAGE_TARGET_DATAPORT_WRITE;
+   insn->bits3.urb.end_of_thread = end_of_thread;
+}
+
+static void brw_set_dp_read_message( struct brw_instruction *insn,
+				      unsigned binding_table_index,
+				      unsigned msg_control,
+				      unsigned msg_type,
+				      unsigned target_cache,
+				      unsigned msg_length,
+				      unsigned response_length,
+				      unsigned end_of_thread )
+{
+   brw_set_src1(insn, brw_imm_d(0));
+
+   insn->bits3.dp_read.binding_table_index = binding_table_index;
+   insn->bits3.dp_read.msg_control = msg_control;
+   insn->bits3.dp_read.msg_type = msg_type;
+   insn->bits3.dp_read.target_cache = target_cache;
+   insn->bits3.dp_read.response_length = response_length;
+   insn->bits3.dp_read.msg_length = msg_length;
+   insn->bits3.dp_read.msg_target = BRW_MESSAGE_TARGET_DATAPORT_READ;
+   insn->bits3.dp_read.end_of_thread = end_of_thread;
+}
+
+static void brw_set_sampler_message( struct brw_instruction *insn,
+				     unsigned binding_table_index,
+				     unsigned sampler,
+				     unsigned msg_type,
+				     unsigned response_length,
+				     unsigned msg_length,
+				     boolean eot)
+{
+   brw_set_src1(insn, brw_imm_d(0));
+
+   insn->bits3.sampler.binding_table_index = binding_table_index;
+   insn->bits3.sampler.sampler = sampler;
+   insn->bits3.sampler.msg_type = msg_type;
+   insn->bits3.sampler.return_format = BRW_SAMPLER_RETURN_FORMAT_FLOAT32;
+   insn->bits3.sampler.response_length = response_length;
+   insn->bits3.sampler.msg_length = msg_length;
+   insn->bits3.sampler.end_of_thread = eot;
+   insn->bits3.sampler.msg_target = BRW_MESSAGE_TARGET_SAMPLER;
+}
+
+
+
+static struct brw_instruction *next_insn( struct brw_compile *p,
+					  unsigned opcode )
+{
+   struct brw_instruction *insn;
+
+   assert(p->nr_insn + 1 < BRW_EU_MAX_INSN);
+
+   insn = &p->store[p->nr_insn++];
+   memcpy(insn, p->current, sizeof(*insn));
+
+   /* Reset this one-shot flag:
+    */
+
+   if (p->current->header.destreg__conditonalmod) {
+      p->current->header.destreg__conditonalmod = 0;
+      p->current->header.predicate_control = BRW_PREDICATE_NORMAL;
+   }
+
+   insn->header.opcode = opcode;
+   return insn;
+}
+
+
+struct brw_instruction *brw_alu1( struct brw_compile *p,
+				  unsigned opcode,
+				  struct brw_reg dest,
+				  struct brw_reg src )
+{
+   struct brw_instruction *insn = next_insn(p, opcode);
+   brw_set_dest(insn, dest);
+   brw_set_src0(insn, src);
+   return insn;
+}
+
+struct brw_instruction *brw_alu2(struct brw_compile *p,
+				 unsigned opcode,
+				 struct brw_reg dest,
+				 struct brw_reg src0,
+				 struct brw_reg src1 )
+{
+   struct brw_instruction *insn = next_insn(p, opcode);
+   brw_set_dest(insn, dest);
+   brw_set_src0(insn, src0);
+   brw_set_src1(insn, src1);
+   return insn;
+}
+
+
+/***********************************************************************
+ * Convenience routines.
+ */
+#define ALU1(OP)					\
+struct brw_instruction *brw_##OP(struct brw_compile *p,			\
+	      struct brw_reg dest,			\
+	      struct brw_reg src0)   			\
+{							\
+   return brw_alu1(p, BRW_OPCODE_##OP, dest, src0);    	\
+}
+
+#define ALU2(OP)					\
+struct brw_instruction *brw_##OP(struct brw_compile *p,			\
+	      struct brw_reg dest,			\
+	      struct brw_reg src0,			\
+	      struct brw_reg src1)   			\
+{							\
+   return brw_alu2(p, BRW_OPCODE_##OP, dest, src0, src1);	\
+}
+
+
+ALU1(MOV)
+ALU2(SEL)
+ALU1(NOT)
+ALU2(AND)
+ALU2(OR)
+ALU2(XOR)
+ALU2(SHR)
+ALU2(SHL)
+ALU2(RSR)
+ALU2(RSL)
+ALU2(ASR)
+ALU2(ADD)
+ALU2(MUL)
+ALU1(FRC)
+ALU1(RNDD)
+ALU2(MAC)
+ALU2(MACH)
+ALU1(LZD)
+ALU2(DP4)
+ALU2(DPH)
+ALU2(DP3)
+ALU2(DP2)
+ALU2(LINE)
+
+
+
+
+void brw_NOP(struct brw_compile *p)
+{
+   struct brw_instruction *insn = next_insn(p, BRW_OPCODE_NOP);
+   brw_set_dest(insn, retype(brw_vec4_grf(0,0), BRW_REGISTER_TYPE_UD));
+   brw_set_src0(insn, retype(brw_vec4_grf(0,0), BRW_REGISTER_TYPE_UD));
+   brw_set_src1(insn, brw_imm_ud(0x0));
+}
+
+
+
+
+
+/***********************************************************************
+ * Comparisons, if/else/endif
+ */
+
+struct brw_instruction *brw_JMPI(struct brw_compile *p,
+	      struct brw_reg dest,
+	      struct brw_reg src0,
+	      struct brw_reg src1)
+{
+   struct brw_instruction *insn = brw_alu2(p, BRW_OPCODE_JMPI, dest, src0, src1);
+
+   p->current->header.predicate_control = BRW_PREDICATE_NONE;
+
+   return insn;
+}
+
+/* EU takes the value from the flag register and pushes it onto some
+ * sort of a stack (presumably merging with any flag value already on
+ * the stack).  Within an if block, the flags at the top of the stack
+ * control execution on each channel of the unit, eg. on each of the
+ * 16 pixel values in our wm programs.
+ *
+ * When the matching 'else' instruction is reached (presumably by
+ * countdown of the instruction count patched in by our ELSE/ENDIF
+ * functions), the relevent flags are inverted.
+ *
+ * When the matching 'endif' instruction is reached, the flags are
+ * popped off.  If the stack is now empty, normal execution resumes.
+ *
+ * No attempt is made to deal with stack overflow (14 elements?).
+ */
+struct brw_instruction *brw_IF(struct brw_compile *p, unsigned execute_size)
+{
+   struct brw_instruction *insn;
+
+   if (p->single_program_flow) {
+      assert(execute_size == BRW_EXECUTE_1);
+
+      insn = next_insn(p, BRW_OPCODE_ADD);
+      insn->header.predicate_inverse = 1;
+   } else {
+      insn = next_insn(p, BRW_OPCODE_IF);
+   }
+
+   /* Override the defaults for this instruction:
+    */
+   brw_set_dest(insn, brw_ip_reg());
+   brw_set_src0(insn, brw_ip_reg());
+   brw_set_src1(insn, brw_imm_d(0x0));
+
+   insn->header.execution_size = execute_size;
+   insn->header.compression_control = BRW_COMPRESSION_NONE;
+   insn->header.predicate_control = BRW_PREDICATE_NORMAL;
+   insn->header.mask_control = BRW_MASK_ENABLE;
+
+   p->current->header.predicate_control = BRW_PREDICATE_NONE;
+
+   return insn;
+}
+
+
+struct brw_instruction *brw_ELSE(struct brw_compile *p,
+				 struct brw_instruction *if_insn)
+{
+   struct brw_instruction *insn;
+
+   if (p->single_program_flow) {
+      insn = next_insn(p, BRW_OPCODE_ADD);
+   } else {
+      insn = next_insn(p, BRW_OPCODE_ELSE);
+   }
+
+   brw_set_dest(insn, brw_ip_reg());
+   brw_set_src0(insn, brw_ip_reg());
+   brw_set_src1(insn, brw_imm_d(0x0));
+
+   insn->header.compression_control = BRW_COMPRESSION_NONE;
+   insn->header.execution_size = if_insn->header.execution_size;
+   insn->header.mask_control = BRW_MASK_ENABLE;
+
+   /* Patch the if instruction to point at this instruction.
+    */
+   if (p->single_program_flow) {
+      assert(if_insn->header.opcode == BRW_OPCODE_ADD);
+
+      if_insn->bits3.ud = (insn - if_insn + 1) * 16;
+   } else {
+      assert(if_insn->header.opcode == BRW_OPCODE_IF);
+
+      if_insn->bits3.if_else.jump_count = insn - if_insn;
+      if_insn->bits3.if_else.pop_count = 1;
+      if_insn->bits3.if_else.pad0 = 0;
+   }
+
+   return insn;
+}
+
+void brw_ENDIF(struct brw_compile *p,
+	       struct brw_instruction *patch_insn)
+{
+   if (p->single_program_flow) {
+      /* In single program flow mode, there's no need to execute an ENDIF,
+       * since we don't need to do any stack operations, and if we're executing
+       * currently, we want to just continue executing.
+       */
+      struct brw_instruction *next = &p->store[p->nr_insn];
+
+      assert(patch_insn->header.opcode == BRW_OPCODE_ADD);
+
+      patch_insn->bits3.ud = (next - patch_insn) * 16;
+   } else {
+      struct brw_instruction *insn = next_insn(p, BRW_OPCODE_ENDIF);
+
+      brw_set_dest(insn, retype(brw_vec4_grf(0,0), BRW_REGISTER_TYPE_UD));
+      brw_set_src0(insn, retype(brw_vec4_grf(0,0), BRW_REGISTER_TYPE_UD));
+      brw_set_src1(insn, brw_imm_d(0x0));
+
+      insn->header.compression_control = BRW_COMPRESSION_NONE;
+      insn->header.execution_size = patch_insn->header.execution_size;
+      insn->header.mask_control = BRW_MASK_ENABLE;
+
+      assert(patch_insn->bits3.if_else.jump_count == 0);
+
+      /* Patch the if or else instructions to point at this or the next
+       * instruction respectively.
+       */
+      if (patch_insn->header.opcode == BRW_OPCODE_IF) {
+	 /* Automagically turn it into an IFF:
+	  */
+	 patch_insn->header.opcode = BRW_OPCODE_IFF;
+	 patch_insn->bits3.if_else.jump_count = insn - patch_insn + 1;
+	 patch_insn->bits3.if_else.pop_count = 0;
+	 patch_insn->bits3.if_else.pad0 = 0;
+      } else if (patch_insn->header.opcode == BRW_OPCODE_ELSE) {
+	 patch_insn->bits3.if_else.jump_count = insn - patch_insn + 1;
+	 patch_insn->bits3.if_else.pop_count = 1;
+	 patch_insn->bits3.if_else.pad0 = 0;
+      } else {
+	 assert(0);
+      }
+
+      /* Also pop item off the stack in the endif instruction:
+       */
+      insn->bits3.if_else.jump_count = 0;
+      insn->bits3.if_else.pop_count = 1;
+      insn->bits3.if_else.pad0 = 0;
+   }
+}
+
+struct brw_instruction *brw_BREAK(struct brw_compile *p)
+{
+   struct brw_instruction *insn;
+   insn = next_insn(p, BRW_OPCODE_BREAK);
+   brw_set_dest(insn, brw_ip_reg());
+   brw_set_src0(insn, brw_ip_reg());
+   brw_set_src1(insn, brw_imm_d(0x0));
+   insn->header.compression_control = BRW_COMPRESSION_NONE;
+   insn->header.execution_size = BRW_EXECUTE_8;
+   insn->header.mask_control = BRW_MASK_DISABLE;
+   insn->bits3.if_else.pad0 = 0;
+   return insn;
+}
+
+struct brw_instruction *brw_CONT(struct brw_compile *p)
+{
+   struct brw_instruction *insn;
+   insn = next_insn(p, BRW_OPCODE_CONTINUE);
+   brw_set_dest(insn, brw_ip_reg());
+   brw_set_src0(insn, brw_ip_reg());
+   brw_set_src1(insn, brw_imm_d(0x0));
+   insn->header.compression_control = BRW_COMPRESSION_NONE;
+   insn->header.execution_size = BRW_EXECUTE_8;
+   insn->header.mask_control = BRW_MASK_DISABLE;
+   insn->bits3.if_else.pad0 = 0;
+   return insn;
+}
+
+/* DO/WHILE loop:
+ */
+struct brw_instruction *brw_DO(struct brw_compile *p, unsigned execute_size)
+{
+   if (p->single_program_flow) {
+      return &p->store[p->nr_insn];
+   } else {
+      struct brw_instruction *insn = next_insn(p, BRW_OPCODE_DO);
+
+      /* Override the defaults for this instruction:
+       */
+      brw_set_dest(insn, brw_null_reg());
+      brw_set_src0(insn, brw_null_reg());
+      brw_set_src1(insn, brw_null_reg());
+
+      insn->header.compression_control = BRW_COMPRESSION_NONE;
+      insn->header.execution_size = execute_size;
+      insn->header.predicate_control = BRW_PREDICATE_NONE;
+      /* insn->header.mask_control = BRW_MASK_ENABLE; */
+      insn->header.mask_control = BRW_MASK_DISABLE;
+
+      return insn;
+   }
+}
+
+
+
+struct brw_instruction *brw_WHILE(struct brw_compile *p,
+	       struct brw_instruction *do_insn)
+{
+   struct brw_instruction *insn;
+
+   if (p->single_program_flow)
+      insn = next_insn(p, BRW_OPCODE_ADD);
+   else
+      insn = next_insn(p, BRW_OPCODE_WHILE);
+
+   brw_set_dest(insn, brw_ip_reg());
+   brw_set_src0(insn, brw_ip_reg());
+   brw_set_src1(insn, brw_imm_d(0x0));
+
+   insn->header.compression_control = BRW_COMPRESSION_NONE;
+
+   if (p->single_program_flow) {
+      insn->header.execution_size = BRW_EXECUTE_1;
+
+      insn->bits3.d = (do_insn - insn) * 16;
+   } else {
+      insn->header.execution_size = do_insn->header.execution_size;
+
+      assert(do_insn->header.opcode == BRW_OPCODE_DO);
+      insn->bits3.if_else.jump_count = do_insn - insn;
+      insn->bits3.if_else.pop_count = 0;
+      insn->bits3.if_else.pad0 = 0;
+   }
+
+/*    insn->header.mask_control = BRW_MASK_ENABLE; */
+
+   insn->header.mask_control = BRW_MASK_DISABLE;
+   p->current->header.predicate_control = BRW_PREDICATE_NONE;
+   return insn;
+}
+
+
+/* FORWARD JUMPS:
+ */
+void brw_land_fwd_jump(struct brw_compile *p,
+		       struct brw_instruction *jmp_insn)
+{
+   struct brw_instruction *landing = &p->store[p->nr_insn];
+
+   assert(jmp_insn->header.opcode == BRW_OPCODE_JMPI);
+   assert(jmp_insn->bits1.da1.src1_reg_file = BRW_IMMEDIATE_VALUE);
+
+   jmp_insn->bits3.ud = (landing - jmp_insn) - 1;
+}
+
+
+
+/* To integrate with the above, it makes sense that the comparison
+ * instruction should populate the flag register.  It might be simpler
+ * just to use the flag reg for most WM tasks?
+ */
+void brw_CMP(struct brw_compile *p,
+	     struct brw_reg dest,
+	     unsigned conditional,
+	     struct brw_reg src0,
+	     struct brw_reg src1)
+{
+   struct brw_instruction *insn = next_insn(p, BRW_OPCODE_CMP);
+
+   insn->header.destreg__conditonalmod = conditional;
+   brw_set_dest(insn, dest);
+   brw_set_src0(insn, src0);
+   brw_set_src1(insn, src1);
+
+/*    guess_execution_size(insn, src0); */
+
+
+   /* Make it so that future instructions will use the computed flag
+    * value until brw_set_predicate_control_flag_value() is called
+    * again.
+    */
+   if (dest.file == BRW_ARCHITECTURE_REGISTER_FILE &&
+       dest.nr == 0) {
+      p->current->header.predicate_control = BRW_PREDICATE_NORMAL;
+      p->flag_value = 0xff;
+   }
+}
+
+
+
+/***********************************************************************
+ * Helpers for the various SEND message types:
+ */
+
+/* Invert 8 values
+ */
+void brw_math( struct brw_compile *p,
+	       struct brw_reg dest,
+	       unsigned function,
+	       unsigned saturate,
+	       unsigned msg_reg_nr,
+	       struct brw_reg src,
+	       unsigned data_type,
+	       unsigned precision )
+{
+   struct brw_instruction *insn = next_insn(p, BRW_OPCODE_SEND);
+   unsigned msg_length = (function == BRW_MATH_FUNCTION_POW) ? 2 : 1;
+   unsigned response_length = (function == BRW_MATH_FUNCTION_SINCOS) ? 2 : 1;
+
+   /* Example code doesn't set predicate_control for send
+    * instructions.
+    */
+   insn->header.predicate_control = 0;
+   insn->header.destreg__conditonalmod = msg_reg_nr;
+
+   brw_set_dest(insn, dest);
+   brw_set_src0(insn, src);
+   brw_set_math_message(insn,
+			msg_length, response_length,
+			function,
+			BRW_MATH_INTEGER_UNSIGNED,
+			precision,
+			saturate,
+			data_type);
+}
+
+/* Use 2 send instructions to invert 16 elements
+ */
+void brw_math_16( struct brw_compile *p,
+		  struct brw_reg dest,
+		  unsigned function,
+		  unsigned saturate,
+		  unsigned msg_reg_nr,
+		  struct brw_reg src,
+		  unsigned precision )
+{
+   struct brw_instruction *insn;
+   unsigned msg_length = (function == BRW_MATH_FUNCTION_POW) ? 2 : 1;
+   unsigned response_length = (function == BRW_MATH_FUNCTION_SINCOS) ? 2 : 1;
+
+   /* First instruction:
+    */
+   brw_push_insn_state(p);
+   brw_set_predicate_control_flag_value(p, 0xff);
+   brw_set_compression_control(p, BRW_COMPRESSION_NONE);
+
+   insn = next_insn(p, BRW_OPCODE_SEND);
+   insn->header.destreg__conditonalmod = msg_reg_nr;
+
+   brw_set_dest(insn, dest);
+   brw_set_src0(insn, src);
+   brw_set_math_message(insn,
+			msg_length, response_length,
+			function,
+			BRW_MATH_INTEGER_UNSIGNED,
+			precision,
+			saturate,
+			BRW_MATH_DATA_VECTOR);
+
+   /* Second instruction:
+    */
+   insn = next_insn(p, BRW_OPCODE_SEND);
+   insn->header.compression_control = BRW_COMPRESSION_2NDHALF;
+   insn->header.destreg__conditonalmod = msg_reg_nr+1;
+
+   brw_set_dest(insn, offset(dest,1));
+   brw_set_src0(insn, src);
+   brw_set_math_message(insn,
+			msg_length, response_length,
+			function,
+			BRW_MATH_INTEGER_UNSIGNED,
+			precision,
+			saturate,
+			BRW_MATH_DATA_VECTOR);
+
+   brw_pop_insn_state(p);
+}
+
+
+
+
+void brw_dp_WRITE_16( struct brw_compile *p,
+		      struct brw_reg src,
+		      unsigned msg_reg_nr,
+		      unsigned scratch_offset )
+{
+   {
+      brw_push_insn_state(p);
+      brw_set_mask_control(p, BRW_MASK_DISABLE);
+      brw_set_compression_control(p, BRW_COMPRESSION_NONE);
+
+      brw_MOV(p,
+	      retype(brw_vec1_grf(0, 2), BRW_REGISTER_TYPE_D),
+	      brw_imm_d(scratch_offset));
+
+      brw_pop_insn_state(p);
+   }
+
+   {
+      unsigned msg_length = 3;
+      struct brw_reg dest = retype(brw_null_reg(), BRW_REGISTER_TYPE_UW);
+      struct brw_instruction *insn = next_insn(p, BRW_OPCODE_SEND);
+
+      insn->header.predicate_control = 0; /* XXX */
+      insn->header.compression_control = BRW_COMPRESSION_NONE;
+      insn->header.destreg__conditonalmod = msg_reg_nr;
+
+      brw_set_dest(insn, dest);
+      brw_set_src0(insn, src);
+
+      brw_set_dp_write_message(insn,
+			       255, /* bti */
+			       BRW_DATAPORT_OWORD_BLOCK_4_OWORDS, /* msg_control */
+			       BRW_DATAPORT_WRITE_MESSAGE_OWORD_BLOCK_WRITE, /* msg_type */
+			       msg_length,
+			       0, /* pixel scoreboard */
+			       0, /* response_length */
+			       0); /* eot */
+   }
+
+}
+
+
+void brw_dp_READ_16( struct brw_compile *p,
+		      struct brw_reg dest,
+		      unsigned msg_reg_nr,
+		      unsigned scratch_offset )
+{
+   {
+      brw_push_insn_state(p);
+      brw_set_compression_control(p, BRW_COMPRESSION_NONE);
+      brw_set_mask_control(p, BRW_MASK_DISABLE);
+
+      brw_MOV(p,
+	      retype(brw_vec1_grf(0, 2), BRW_REGISTER_TYPE_D),
+	      brw_imm_d(scratch_offset));
+
+      brw_pop_insn_state(p);
+   }
+
+   {
+      struct brw_instruction *insn = next_insn(p, BRW_OPCODE_SEND);
+
+      insn->header.predicate_control = 0; /* XXX */
+      insn->header.compression_control = BRW_COMPRESSION_NONE;
+      insn->header.destreg__conditonalmod = msg_reg_nr;
+
+      brw_set_dest(insn, dest);	/* UW? */
+      brw_set_src0(insn, retype(brw_vec8_grf(0, 0), BRW_REGISTER_TYPE_UW));
+
+      brw_set_dp_read_message(insn,
+			      255, /* bti */
+			      3,  /* msg_control */
+			      BRW_DATAPORT_READ_MESSAGE_OWORD_BLOCK_READ, /* msg_type */
+			      1, /* target cache */
+			      1, /* msg_length */
+			      2, /* response_length */
+			      0); /* eot */
+   }
+}
+
+
+void brw_fb_WRITE(struct brw_compile *p,
+		   struct brw_reg dest,
+		   unsigned msg_reg_nr,
+		   struct brw_reg src0,
+		   unsigned binding_table_index,
+		   unsigned msg_length,
+		   unsigned response_length,
+		   boolean eot)
+{
+   struct brw_instruction *insn = next_insn(p, BRW_OPCODE_SEND);
+
+   insn->header.predicate_control = 0; /* XXX */
+   insn->header.compression_control = BRW_COMPRESSION_NONE;
+   insn->header.destreg__conditonalmod = msg_reg_nr;
+
+   brw_set_dest(insn, dest);
+   brw_set_src0(insn, src0);
+   brw_set_dp_write_message(insn,
+			    binding_table_index,
+			    BRW_DATAPORT_RENDER_TARGET_WRITE_SIMD16_SINGLE_SOURCE, /* msg_control */
+			    BRW_DATAPORT_WRITE_MESSAGE_RENDER_TARGET_WRITE, /* msg_type */
+			    msg_length,
+			    1,	/* pixel scoreboard */
+			    response_length,
+			    eot);
+}
+
+
+
+void brw_SAMPLE(struct brw_compile *p,
+		struct brw_reg dest,
+		unsigned msg_reg_nr,
+		struct brw_reg src0,
+		unsigned binding_table_index,
+		unsigned sampler,
+		unsigned writemask,
+		unsigned msg_type,
+		unsigned response_length,
+		unsigned msg_length,
+		boolean eot)
+{
+   boolean need_stall = 0;
+
+   if(writemask == 0) {
+/*       debug_printf("%s: zero writemask??\n", __FUNCTION__); */
+      return;
+   }
+
+   /* Hardware doesn't do destination dependency checking on send
+    * instructions properly.  Add a workaround which generates the
+    * dependency by other means.  In practice it seems like this bug
+    * only crops up for texture samples, and only where registers are
+    * written by the send and then written again later without being
+    * read in between.  Luckily for us, we already track that
+    * information and use it to modify the writemask for the
+    * instruction, so that is a guide for whether a workaround is
+    * needed.
+    */
+   if (writemask != TGSI_WRITEMASK_XYZW) {
+      unsigned dst_offset = 0;
+      unsigned i, newmask = 0, len = 0;
+
+      for (i = 0; i < 4; i++) {
+	 if (writemask & (1<<i))
+	    break;
+	 dst_offset += 2;
+      }
+      for (; i < 4; i++) {
+	 if (!(writemask & (1<<i)))
+	    break;
+	 newmask |= 1<<i;
+	 len++;
+      }
+
+      if (newmask != writemask) {
+	 need_stall = 1;
+/* 	 debug_printf("need stall %x %x\n", newmask , writemask); */
+      }
+      else {
+	 struct brw_reg m1 = brw_message_reg(msg_reg_nr);
+
+	 newmask = ~newmask & TGSI_WRITEMASK_XYZW;
+
+	 brw_push_insn_state(p);
+
+	 brw_set_compression_control(p, BRW_COMPRESSION_NONE);
+	 brw_set_mask_control(p, BRW_MASK_DISABLE);
+
+	 brw_MOV(p, m1, brw_vec8_grf(0,0));
+  	 brw_MOV(p, get_element_ud(m1, 2), brw_imm_ud(newmask << 12));
+
+	 brw_pop_insn_state(p);
+
+  	 src0 = retype(brw_null_reg(), BRW_REGISTER_TYPE_UW);
+	 dest = offset(dest, dst_offset);
+	 response_length = len * 2;
+      }
+   }
+
+   {
+      struct brw_instruction *insn = next_insn(p, BRW_OPCODE_SEND);
+
+      insn->header.predicate_control = 0; /* XXX */
+      insn->header.compression_control = BRW_COMPRESSION_NONE;
+      insn->header.destreg__conditonalmod = msg_reg_nr;
+
+      brw_set_dest(insn, dest);
+      brw_set_src0(insn, src0);
+      brw_set_sampler_message(insn,
+			      binding_table_index,
+			      sampler,
+			      msg_type,
+			      response_length,
+			      msg_length,
+			      eot);
+   }
+
+   if (need_stall)
+   {
+      struct brw_reg reg = vec8(offset(dest, response_length-1));
+
+      /*  mov (8) r9.0<1>:f    r9.0<8;8,1>:f    { Align1 }
+       */
+      brw_push_insn_state(p);
+      brw_set_compression_control(p, FALSE);
+      brw_MOV(p, reg, reg);
+      brw_pop_insn_state(p);
+   }
+
+}
+
+/* All these variables are pretty confusing - we might be better off
+ * using bitmasks and macros for this, in the old style.  Or perhaps
+ * just having the caller instantiate the fields in dword3 itself.
+ */
+void brw_urb_WRITE(struct brw_compile *p,
+		   struct brw_reg dest,
+		   unsigned msg_reg_nr,
+		   struct brw_reg src0,
+		   boolean allocate,
+		   boolean used,
+		   unsigned msg_length,
+		   unsigned response_length,
+		   boolean eot,
+		   boolean writes_complete,
+		   unsigned offset,
+		   unsigned swizzle)
+{
+   struct brw_instruction *insn = next_insn(p, BRW_OPCODE_SEND);
+
+   assert(msg_length < 16);
+
+   brw_set_dest(insn, dest);
+   brw_set_src0(insn, src0);
+   brw_set_src1(insn, brw_imm_d(0));
+
+   insn->header.destreg__conditonalmod = msg_reg_nr;
+
+   brw_set_urb_message(insn,
+		       allocate,
+		       used,
+		       msg_length,
+		       response_length,
+		       eot,
+		       writes_complete,
+		       offset,
+		       swizzle);
+}
+
diff --git a/src/gallium/drivers/i965simple/brw_eu_util.c b/src/gallium/drivers/i965simple/brw_eu_util.c
new file mode 100644
index 0000000000..3a65b141f0
--- /dev/null
+++ b/src/gallium/drivers/i965simple/brw_eu_util.c
@@ -0,0 +1,126 @@
+/*
+ Copyright (C) Intel Corp.  2006.  All Rights Reserved.
+ Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to
+ develop this 3D driver.
+ 
+ Permission is hereby granted, free of charge, to any person obtaining
+ a copy of this software and associated documentation files (the
+ "Software"), to deal in the Software without restriction, including
+ without limitation the rights to use, copy, modify, merge, publish,
+ distribute, sublicense, and/or sell copies of the Software, and to
+ permit persons to whom the Software is furnished to do so, subject to
+ the following conditions:
+ 
+ The above copyright notice and this permission notice (including the
+ next paragraph) shall be included in all copies or substantial
+ portions of the Software.
+ 
+ THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
+ IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
+ LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
+ OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ 
+ **********************************************************************/
+ /*
+  * Authors:
+  *   Keith Whitwell <keith@tungstengraphics.com>
+  */
+      
+
+#include "brw_context.h"
+#include "brw_defines.h"
+#include "brw_eu.h"
+
+
+void brw_math_invert( struct brw_compile *p, 
+			     struct brw_reg dst,
+			     struct brw_reg src)
+{
+   brw_math( p, 
+	     dst,
+	     BRW_MATH_FUNCTION_INV, 
+	     BRW_MATH_SATURATE_NONE,
+	     0,
+	     src,
+	     BRW_MATH_PRECISION_FULL, 
+	     BRW_MATH_DATA_VECTOR );
+}
+
+
+
+void brw_copy4(struct brw_compile *p,
+	       struct brw_reg dst,
+	       struct brw_reg src,
+	       unsigned count)
+{
+   unsigned i;
+
+   dst = vec4(dst);
+   src = vec4(src);
+
+   for (i = 0; i < count; i++)
+   {
+      unsigned delta = i*32;
+      brw_MOV(p, byte_offset(dst, delta),    byte_offset(src, delta));
+      brw_MOV(p, byte_offset(dst, delta+16), byte_offset(src, delta+16));
+   }
+}
+
+
+void brw_copy8(struct brw_compile *p,
+	       struct brw_reg dst,
+	       struct brw_reg src,
+	       unsigned count)
+{
+   unsigned i;
+
+   dst = vec8(dst);
+   src = vec8(src);
+
+   for (i = 0; i < count; i++)
+   {
+      unsigned delta = i*32;
+      brw_MOV(p, byte_offset(dst, delta),    byte_offset(src, delta));
+   }
+}
+
+
+void brw_copy_indirect_to_indirect(struct brw_compile *p,
+				   struct brw_indirect dst_ptr,
+				   struct brw_indirect src_ptr,
+				   unsigned count)
+{
+   unsigned i;
+
+   for (i = 0; i < count; i++)
+   {
+      unsigned delta = i*32;
+      brw_MOV(p, deref_4f(dst_ptr, delta),    deref_4f(src_ptr, delta));
+      brw_MOV(p, deref_4f(dst_ptr, delta+16), deref_4f(src_ptr, delta+16));
+   }
+}
+
+
+void brw_copy_from_indirect(struct brw_compile *p,
+			    struct brw_reg dst,
+			    struct brw_indirect ptr,
+			    unsigned count)
+{
+   unsigned i;
+
+   dst = vec4(dst);
+
+   for (i = 0; i < count; i++)
+   {
+      unsigned delta = i*32;
+      brw_MOV(p, byte_offset(dst, delta),    deref_4f(ptr, delta));
+      brw_MOV(p, byte_offset(dst, delta+16), deref_4f(ptr, delta+16));
+   }
+}
+
+
+
+
diff --git a/src/gallium/drivers/i965simple/brw_flush.c b/src/gallium/drivers/i965simple/brw_flush.c
new file mode 100644
index 0000000000..5216c680cf
--- /dev/null
+++ b/src/gallium/drivers/i965simple/brw_flush.c
@@ -0,0 +1,80 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+/* Author:
+ *    Keith Whitwell <keith@tungstengraphics.com>
+ */
+
+
+#include "pipe/p_defines.h"
+#include "brw_context.h"
+#include "brw_defines.h"
+#include "brw_batch.h"
+
+
+/**
+ * In future we may want a fence-like interface instead of finish.
+ */
+static void brw_flush( struct pipe_context *pipe,
+			unsigned flags )
+{
+   struct brw_context *brw = brw_context(pipe);
+   struct pipe_fence_handle *fence;
+
+   /* Do we need to emit an MI_FLUSH command to flush the hardware
+    * caches?
+    */
+   if (flags & (PIPE_FLUSH_RENDER_CACHE | PIPE_FLUSH_TEXTURE_CACHE)) {
+      struct brw_mi_flush flush;
+
+      memset(&flush, 0, sizeof(flush));      
+      flush.opcode = CMD_MI_FLUSH;
+
+      if (!(flags & PIPE_FLUSH_RENDER_CACHE))
+	 flush.flags |= BRW_INHIBIT_FLUSH_RENDER_CACHE;
+
+      if (flags & PIPE_FLUSH_TEXTURE_CACHE)
+	 flush.flags |= BRW_FLUSH_READ_CACHE;
+
+      BRW_BATCH_STRUCT(brw, &flush);
+   }
+
+   /* If there are no flags, just flush pending commands to hardware:
+    */
+   FLUSH_BATCH( &fence );
+
+   if (flags & PIPE_FLUSH_WAIT) {
+//      brw->winsys->wait_fence(brw->winsys, fence);
+   }
+}
+
+
+
+void brw_init_flush_functions( struct brw_context *brw )
+{
+   brw->pipe.flush = brw_flush;
+}
diff --git a/src/gallium/drivers/i965simple/brw_gs.c b/src/gallium/drivers/i965simple/brw_gs.c
new file mode 100644
index 0000000000..de60868ccc
--- /dev/null
+++ b/src/gallium/drivers/i965simple/brw_gs.c
@@ -0,0 +1,196 @@
+/*
+ Copyright (C) Intel Corp.  2006.  All Rights Reserved.
+ Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to
+ develop this 3D driver.
+
+ Permission is hereby granted, free of charge, to any person obtaining
+ a copy of this software and associated documentation files (the
+ "Software"), to deal in the Software without restriction, including
+ without limitation the rights to use, copy, modify, merge, publish,
+ distribute, sublicense, and/or sell copies of the Software, and to
+ permit persons to whom the Software is furnished to do so, subject to
+ the following conditions:
+
+ The above copyright notice and this permission notice (including the
+ next paragraph) shall be included in all copies or substantial
+ portions of the Software.
+
+ THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
+ IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
+ LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
+ OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+
+ **********************************************************************/
+ /*
+  * Authors:
+  *   Keith Whitwell <keith@tungstengraphics.com>
+  */
+
+#include "brw_defines.h"
+#include "brw_context.h"
+#include "brw_eu.h"
+#include "brw_util.h"
+#include "brw_state.h"
+#include "brw_gs.h"
+
+
+
+static void compile_gs_prog( struct brw_context *brw,
+			     struct brw_gs_prog_key *key )
+{
+   struct brw_gs_compile c;
+   const unsigned *program;
+   unsigned program_size;
+
+   memset(&c, 0, sizeof(c));
+
+   c.key = *key;
+
+   /* Need to locate the two positions present in vertex + header.
+    * These are currently hardcoded:
+    */
+   c.nr_attrs = brw_count_bits(c.key.attrs);
+   c.nr_regs = (c.nr_attrs + 1) / 2 + 1;  /* are vertices packed, or reg-aligned? */
+   c.nr_bytes = c.nr_regs * REG_SIZE;
+
+
+   /* Begin the compilation:
+    */
+   brw_init_compile(&c.func);
+
+   c.func.single_program_flow = 1;
+
+   /* For some reason the thread is spawned with only 4 channels
+    * unmasked.
+    */
+   brw_set_mask_control(&c.func, BRW_MASK_DISABLE);
+
+
+   /* Note that primitives which don't require a GS program have
+    * already been weeded out by this stage:
+    */
+   switch (key->primitive) {
+   case PIPE_PRIM_QUADS:
+      brw_gs_quads( &c );
+      break;
+   case PIPE_PRIM_QUAD_STRIP:
+      brw_gs_quad_strip( &c );
+      break;
+   case PIPE_PRIM_LINE_LOOP:
+      brw_gs_lines( &c );
+      break;
+   case PIPE_PRIM_LINES:
+      if (key->hint_gs_always)
+	 brw_gs_lines( &c );
+      else {
+	 return;
+      }
+      break;
+   case PIPE_PRIM_TRIANGLES:
+      if (key->hint_gs_always)
+	 brw_gs_tris( &c );
+      else {
+	 return;
+      }
+      break;
+   case PIPE_PRIM_POINTS:
+      if (key->hint_gs_always)
+	 brw_gs_points( &c );
+      else {
+	 return;
+      }
+      break;
+   default:
+      return;
+   }
+
+   /* get the program
+    */
+   program = brw_get_program(&c.func, &program_size);
+
+   /* Upload
+    */
+   brw->gs.prog_gs_offset = brw_upload_cache( &brw->cache[BRW_GS_PROG],
+					      &c.key,
+					      sizeof(c.key),
+					      program,
+					      program_size,
+					      &c.prog_data,
+					      &brw->gs.prog_data );
+}
+
+
+static boolean search_cache( struct brw_context *brw,
+			       struct brw_gs_prog_key *key )
+{
+   return brw_search_cache(&brw->cache[BRW_GS_PROG],
+			   key, sizeof(*key),
+			   &brw->gs.prog_data,
+			   &brw->gs.prog_gs_offset);
+}
+
+
+static const int gs_prim[PIPE_PRIM_POLYGON+1] = {
+   PIPE_PRIM_POINTS,
+   PIPE_PRIM_LINES,
+   PIPE_PRIM_LINE_LOOP,
+   PIPE_PRIM_LINES,
+   PIPE_PRIM_TRIANGLES,
+   PIPE_PRIM_TRIANGLES,
+   PIPE_PRIM_TRIANGLES,
+   PIPE_PRIM_QUADS,
+   PIPE_PRIM_QUAD_STRIP,
+   PIPE_PRIM_TRIANGLES
+};
+
+static void populate_key( struct brw_context *brw,
+			  struct brw_gs_prog_key *key )
+{
+   memset(key, 0, sizeof(*key));
+
+   /* CACHE_NEW_VS_PROG */
+   key->attrs = brw->vs.prog_data->outputs_written;
+
+   /* BRW_NEW_PRIMITIVE */
+   key->primitive = gs_prim[brw->primitive];
+
+   key->hint_gs_always = 0;	/* debug code? */
+
+   key->need_gs_prog = (key->hint_gs_always ||
+			brw->primitive == PIPE_PRIM_QUADS ||
+			brw->primitive == PIPE_PRIM_QUAD_STRIP ||
+			brw->primitive == PIPE_PRIM_LINE_LOOP);
+}
+
+/* Calculate interpolants for triangle and line rasterization.
+ */
+static void upload_gs_prog( struct brw_context *brw )
+{
+   struct brw_gs_prog_key key;
+
+   /* Populate the key:
+    */
+   populate_key(brw, &key);
+
+   if (brw->gs.prog_active != key.need_gs_prog) {
+      brw->state.dirty.cache |= CACHE_NEW_GS_PROG;
+      brw->gs.prog_active = key.need_gs_prog;
+   }
+
+   if (brw->gs.prog_active) {
+      if (!search_cache(brw, &key))
+	 compile_gs_prog( brw, &key );
+   }
+}
+
+
+const struct brw_tracked_state brw_gs_prog = {
+   .dirty = {
+      .brw   = BRW_NEW_PRIMITIVE,
+      .cache = CACHE_NEW_VS_PROG
+   },
+   .update = upload_gs_prog
+};
diff --git a/src/gallium/drivers/i965simple/brw_gs.h b/src/gallium/drivers/i965simple/brw_gs.h
new file mode 100644
index 0000000000..f09141c6aa
--- /dev/null
+++ b/src/gallium/drivers/i965simple/brw_gs.h
@@ -0,0 +1,75 @@
+/*
+ Copyright (C) Intel Corp.  2006.  All Rights Reserved.
+ Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to
+ develop this 3D driver.
+ 
+ Permission is hereby granted, free of charge, to any person obtaining
+ a copy of this software and associated documentation files (the
+ "Software"), to deal in the Software without restriction, including
+ without limitation the rights to use, copy, modify, merge, publish,
+ distribute, sublicense, and/or sell copies of the Software, and to
+ permit persons to whom the Software is furnished to do so, subject to
+ the following conditions:
+ 
+ The above copyright notice and this permission notice (including the
+ next paragraph) shall be included in all copies or substantial
+ portions of the Software.
+ 
+ THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
+ IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
+ LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
+ OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ 
+ **********************************************************************/
+ /*
+  * Authors:
+  *   Keith Whitwell <keith@tungstengraphics.com>
+  */
+ 
+
+#ifndef BRW_GS_H
+#define BRW_GS_H
+
+
+#include "brw_context.h"
+#include "brw_eu.h"
+
+#define MAX_GS_VERTS (4)	     
+
+struct brw_gs_prog_key {
+   unsigned attrs:32;
+   unsigned primitive:4;
+   unsigned hint_gs_always:1;
+   unsigned need_gs_prog:1;
+   unsigned pad:26;
+};
+
+struct brw_gs_compile {
+   struct brw_compile func;
+   struct brw_gs_prog_key key;
+   struct brw_gs_prog_data prog_data;
+   
+   struct {
+      struct brw_reg R0;
+      struct brw_reg vertex[MAX_GS_VERTS];
+   } reg;
+
+   /* 3 different ways of expressing vertex size:
+    */
+   unsigned nr_attrs;
+   unsigned nr_regs;
+   unsigned nr_bytes;
+};
+
+#define ATTR_SIZE  (4*4)
+
+void brw_gs_quads( struct brw_gs_compile *c );
+void brw_gs_quad_strip( struct brw_gs_compile *c );
+void brw_gs_tris( struct brw_gs_compile *c );
+void brw_gs_lines( struct brw_gs_compile *c );
+void brw_gs_points( struct brw_gs_compile *c );
+
+#endif
diff --git a/src/gallium/drivers/i965simple/brw_gs_emit.c b/src/gallium/drivers/i965simple/brw_gs_emit.c
new file mode 100644
index 0000000000..c3cc90b10f
--- /dev/null
+++ b/src/gallium/drivers/i965simple/brw_gs_emit.c
@@ -0,0 +1,148 @@
+/*
+ Copyright (C) Intel Corp.  2006.  All Rights Reserved.
+ Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to
+ develop this 3D driver.
+ 
+ Permission is hereby granted, free of charge, to any person obtaining
+ a copy of this software and associated documentation files (the
+ "Software"), to deal in the Software without restriction, including
+ without limitation the rights to use, copy, modify, merge, publish,
+ distribute, sublicense, and/or sell copies of the Software, and to
+ permit persons to whom the Software is furnished to do so, subject to
+ the following conditions:
+ 
+ The above copyright notice and this permission notice (including the
+ next paragraph) shall be included in all copies or substantial
+ portions of the Software.
+ 
+ THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
+ IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
+ LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
+ OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ 
+ **********************************************************************/
+ /*
+  * Authors:
+  *   Keith Whitwell <keith@tungstengraphics.com>
+  */
+
+#include "brw_defines.h"
+#include "brw_context.h"
+#include "brw_eu.h"
+#include "brw_util.h"
+#include "brw_gs.h"
+
+static void brw_gs_alloc_regs( struct brw_gs_compile *c,
+			       unsigned nr_verts )
+{
+   unsigned i = 0,j;
+
+   /* Register usage is static, precompute here:
+    */
+   c->reg.R0 = retype(brw_vec8_grf(i, 0), BRW_REGISTER_TYPE_UD); i++;
+
+   /* Payload vertices plus space for more generated vertices:
+    */
+   for (j = 0; j < nr_verts; j++) {
+      c->reg.vertex[j] = brw_vec4_grf(i, 0);
+      i += c->nr_regs;
+   }
+
+   c->prog_data.urb_read_length = c->nr_regs; 
+   c->prog_data.total_grf = i;
+}
+
+
+static void brw_gs_emit_vue(struct brw_gs_compile *c, 
+			    struct brw_reg vert,
+			    boolean last,
+			    unsigned header)
+{
+   struct brw_compile *p = &c->func;
+   boolean allocate = !last;
+
+   /* Overwrite PrimType and PrimStart in the message header, for
+    * each vertex in turn:
+    */
+   brw_MOV(p, get_element_ud(c->reg.R0, 2), brw_imm_ud(header));
+
+   /* Copy the vertex from vertn into m1..mN+1:
+    */
+   brw_copy8(p, brw_message_reg(1), vert, c->nr_regs);
+
+   /* Send each vertex as a seperate write to the urb.  This is
+    * different to the concept in brw_sf_emit.c, where subsequent
+    * writes are used to build up a single urb entry.  Each of these
+    * writes instantiates a seperate urb entry, and a new one must be
+    * allocated each time.
+    */
+   brw_urb_WRITE(p, 
+		 allocate ? c->reg.R0 : retype(brw_null_reg(), BRW_REGISTER_TYPE_UD),
+		 0,
+		 c->reg.R0,
+		 allocate,
+		 1,		/* used */
+		 c->nr_regs + 1, /* msg length */
+		 allocate ? 1 : 0, /* response length */
+		 allocate ? 0 : 1, /* eot */
+		 1,		/* writes_complete */
+		 0,		/* urb offset */
+		 BRW_URB_SWIZZLE_NONE);
+}
+
+
+
+void brw_gs_quads( struct brw_gs_compile *c )
+{
+   brw_gs_alloc_regs(c, 4);
+   
+   /* Use polygons for correct edgeflag behaviour. Note that vertex 3
+    * is the PV for quads, but vertex 0 for polygons:
+    */
+   brw_gs_emit_vue(c, c->reg.vertex[3], 0, ((_3DPRIM_POLYGON << 2) | R02_PRIM_START));
+   brw_gs_emit_vue(c, c->reg.vertex[0], 0, (_3DPRIM_POLYGON << 2));
+   brw_gs_emit_vue(c, c->reg.vertex[1], 0, (_3DPRIM_POLYGON << 2)); 
+   brw_gs_emit_vue(c, c->reg.vertex[2], 1, ((_3DPRIM_POLYGON << 2) | R02_PRIM_END));
+}
+
+void brw_gs_quad_strip( struct brw_gs_compile *c )
+{
+   brw_gs_alloc_regs(c, 4);
+   
+   brw_gs_emit_vue(c, c->reg.vertex[2], 0, ((_3DPRIM_POLYGON << 2) | R02_PRIM_START));
+   brw_gs_emit_vue(c, c->reg.vertex[3], 0, (_3DPRIM_POLYGON << 2));
+   brw_gs_emit_vue(c, c->reg.vertex[0], 0, (_3DPRIM_POLYGON << 2)); 
+   brw_gs_emit_vue(c, c->reg.vertex[1], 1, ((_3DPRIM_POLYGON << 2) | R02_PRIM_END));
+}
+
+void brw_gs_tris( struct brw_gs_compile *c )
+{
+   brw_gs_alloc_regs(c, 3);
+   brw_gs_emit_vue(c, c->reg.vertex[0], 0, ((_3DPRIM_TRILIST << 2) | R02_PRIM_START));
+   brw_gs_emit_vue(c, c->reg.vertex[1], 0, (_3DPRIM_TRILIST << 2));
+   brw_gs_emit_vue(c, c->reg.vertex[2], 1, ((_3DPRIM_TRILIST << 2) | R02_PRIM_END));
+}
+
+void brw_gs_lines( struct brw_gs_compile *c )
+{
+   brw_gs_alloc_regs(c, 2);
+   brw_gs_emit_vue(c, c->reg.vertex[0], 0, ((_3DPRIM_LINESTRIP << 2) | R02_PRIM_START));
+   brw_gs_emit_vue(c, c->reg.vertex[1], 1, ((_3DPRIM_LINESTRIP << 2) | R02_PRIM_END));
+}
+
+void brw_gs_points( struct brw_gs_compile *c )
+{
+   brw_gs_alloc_regs(c, 1);
+   brw_gs_emit_vue(c, c->reg.vertex[0], 1, ((_3DPRIM_POINTLIST << 2) | R02_PRIM_START | R02_PRIM_END));
+}
+
+
+
+
+
+
+
+
diff --git a/src/gallium/drivers/i965simple/brw_gs_state.c b/src/gallium/drivers/i965simple/brw_gs_state.c
new file mode 100644
index 0000000000..3932e9e939
--- /dev/null
+++ b/src/gallium/drivers/i965simple/brw_gs_state.c
@@ -0,0 +1,89 @@
+/*
+ Copyright (C) Intel Corp.  2006.  All Rights Reserved.
+ Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to
+ develop this 3D driver.
+
+ Permission is hereby granted, free of charge, to any person obtaining
+ a copy of this software and associated documentation files (the
+ "Software"), to deal in the Software without restriction, including
+ without limitation the rights to use, copy, modify, merge, publish,
+ distribute, sublicense, and/or sell copies of the Software, and to
+ permit persons to whom the Software is furnished to do so, subject to
+ the following conditions:
+
+ The above copyright notice and this permission notice (including the
+ next paragraph) shall be included in all copies or substantial
+ portions of the Software.
+
+ THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
+ IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
+ LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
+ OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+
+ **********************************************************************/
+ /*
+  * Authors:
+  *   Keith Whitwell <keith@tungstengraphics.com>
+  */
+
+
+
+#include "brw_context.h"
+#include "brw_state.h"
+#include "brw_defines.h"
+#include "pipe/p_util.h"
+
+
+
+static void upload_gs_unit( struct brw_context *brw )
+{
+   struct brw_gs_unit_state gs;
+
+   memset(&gs, 0, sizeof(gs));
+
+   /* CACHE_NEW_GS_PROG */
+   if (brw->gs.prog_active) {
+      gs.thread0.grf_reg_count =
+	 align(brw->gs.prog_data->total_grf, 16) / 16 - 1;
+      gs.thread0.kernel_start_pointer = brw->gs.prog_gs_offset >> 6;
+      gs.thread3.urb_entry_read_length = brw->gs.prog_data->urb_read_length;
+   }
+   else {
+      gs.thread0.grf_reg_count = 0;
+      gs.thread0.kernel_start_pointer = 0;
+      gs.thread3.urb_entry_read_length = 1;
+   }
+
+   /* BRW_NEW_URB_FENCE */
+   gs.thread4.nr_urb_entries = brw->urb.nr_gs_entries;
+   gs.thread4.urb_entry_allocation_size = brw->urb.vsize - 1;
+
+   gs.thread4.max_threads = 0; /* Hardware requirement */
+
+   if (BRW_DEBUG & DEBUG_STATS)
+      gs.thread4.stats_enable = 1;
+
+   /* CONSTANT */
+   gs.thread1.floating_point_mode = BRW_FLOATING_POINT_NON_IEEE_754;
+   gs.thread1.single_program_flow = 1;
+   gs.thread3.dispatch_grf_start_reg = 1;
+   gs.thread3.const_urb_entry_read_offset = 0;
+   gs.thread3.const_urb_entry_read_length = 0;
+   gs.thread3.urb_entry_read_offset = 0;
+
+
+   brw->gs.state_gs_offset = brw_cache_data( &brw->cache[BRW_GS_UNIT], &gs );
+}
+
+
+const struct brw_tracked_state brw_gs_unit = {
+   .dirty = {
+      .brw   = (BRW_NEW_CURBE_OFFSETS |
+		BRW_NEW_URB_FENCE),
+      .cache = CACHE_NEW_GS_PROG
+   },
+   .update = upload_gs_unit
+};
diff --git a/src/gallium/drivers/i965simple/brw_misc_state.c b/src/gallium/drivers/i965simple/brw_misc_state.c
new file mode 100644
index 0000000000..925049ecc1
--- /dev/null
+++ b/src/gallium/drivers/i965simple/brw_misc_state.c
@@ -0,0 +1,486 @@
+/*
+ Copyright (C) Intel Corp.  2006.  All Rights Reserved.
+ Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to
+ develop this 3D driver.
+
+ Permission is hereby granted, free of charge, to any person obtaining
+ a copy of this software and associated documentation files (the
+ "Software"), to deal in the Software without restriction, including
+ without limitation the rights to use, copy, modify, merge, publish,
+ distribute, sublicense, and/or sell copies of the Software, and to
+ permit persons to whom the Software is furnished to do so, subject to
+ the following conditions:
+
+ The above copyright notice and this permission notice (including the
+ next paragraph) shall be included in all copies or substantial
+ portions of the Software.
+
+ THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
+ IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
+ LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
+ OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+
+ **********************************************************************/
+ /*
+  * Authors:
+  *   Keith Whitwell <keith@tungstengraphics.com>
+  */
+
+#include "brw_batch.h"
+#include "brw_context.h"
+#include "brw_state.h"
+#include "brw_defines.h"
+
+
+
+
+
+/***********************************************************************
+ * Blend color
+ */
+
+static void upload_blend_constant_color(struct brw_context *brw)
+{
+   struct brw_blend_constant_color bcc;
+
+   memset(&bcc, 0, sizeof(bcc));
+   bcc.header.opcode = CMD_BLEND_CONSTANT_COLOR;
+   bcc.header.length = sizeof(bcc)/4-2;
+   bcc.blend_constant_color[0] = brw->attribs.BlendColor.color[0];
+   bcc.blend_constant_color[1] = brw->attribs.BlendColor.color[1];
+   bcc.blend_constant_color[2] = brw->attribs.BlendColor.color[2];
+   bcc.blend_constant_color[3] = brw->attribs.BlendColor.color[3];
+
+   BRW_CACHED_BATCH_STRUCT(brw, &bcc);
+}
+
+
+const struct brw_tracked_state brw_blend_constant_color = {
+   .dirty = {
+      .brw = BRW_NEW_BLEND,
+      .cache = 0
+   },
+   .update = upload_blend_constant_color
+};
+
+
+/***********************************************************************
+ * Drawing rectangle 
+ */
+static void upload_drawing_rect(struct brw_context *brw)
+{
+   struct brw_drawrect bdr;
+
+   memset(&bdr, 0, sizeof(bdr));
+   bdr.header.opcode = CMD_DRAW_RECT;
+   bdr.header.length = sizeof(bdr)/4 - 2;
+   bdr.xmin = 0;
+   bdr.ymin = 0;
+   bdr.xmax = brw->attribs.FrameBuffer.cbufs[0]->width;
+   bdr.ymax = brw->attribs.FrameBuffer.cbufs[0]->height;
+   bdr.xorg = 0;
+   bdr.yorg = 0;
+
+   /* Can't use BRW_CACHED_BATCH_STRUCT because this is also emitted
+    * uncached in brw_draw.c:
+    */
+   BRW_BATCH_STRUCT(brw, &bdr);
+}
+
+const struct brw_tracked_state brw_drawing_rect = {
+   .dirty = {
+      .brw = BRW_NEW_SCENE,
+      .cache = 0
+   },
+   .update = upload_drawing_rect
+};
+
+/**
+ * Upload the binding table pointers, which point each stage's array of surface
+ * state pointers.
+ *
+ * The binding table pointers are relative to the surface state base address,
+ * which is the BRW_SS_POOL cache buffer.
+ */
+static void upload_binding_table_pointers(struct brw_context *brw)
+{
+   struct brw_binding_table_pointers btp;
+   memset(&btp, 0, sizeof(btp));
+
+   btp.header.opcode = CMD_BINDING_TABLE_PTRS;
+   btp.header.length = sizeof(btp)/4 - 2;
+   btp.vs = 0;
+   btp.gs = 0;
+   btp.clp = 0;
+   btp.sf = 0;
+   btp.wm = brw->wm.bind_ss_offset;
+
+   BRW_CACHED_BATCH_STRUCT(brw, &btp);
+}
+
+const struct brw_tracked_state brw_binding_table_pointers = {
+   .dirty = {
+      .brw = 0,
+      .cache = CACHE_NEW_SURF_BIND
+   },
+   .update = upload_binding_table_pointers,
+};
+
+
+/**
+ * Upload pointers to the per-stage state.
+ *
+ * The state pointers in this packet are all relative to the general state
+ * base address set by CMD_STATE_BASE_ADDRESS, which is the BRW_GS_POOL buffer.
+ */
+static void upload_pipelined_state_pointers(struct brw_context *brw )
+{
+   struct brw_pipelined_state_pointers psp;
+   memset(&psp, 0, sizeof(psp));
+
+   psp.header.opcode = CMD_PIPELINED_STATE_POINTERS;
+   psp.header.length = sizeof(psp)/4 - 2;
+
+   psp.vs.offset = brw->vs.state_gs_offset >> 5;
+   psp.sf.offset = brw->sf.state_gs_offset >> 5;
+   psp.wm.offset = brw->wm.state_gs_offset >> 5;
+   psp.cc.offset = brw->cc.state_gs_offset >> 5;
+
+   /* GS gets turned on and off regularly.  Need to re-emit URB fence
+    * after this occurs.
+    */
+   if (brw->gs.prog_active) {
+      psp.gs.offset = brw->gs.state_gs_offset >> 5;
+      psp.gs.enable = 1;
+   }
+
+   if (0) {
+      psp.clp.offset = brw->clip.state_gs_offset >> 5;
+      psp.clp.enable = 1;
+   }
+
+
+   if (BRW_CACHED_BATCH_STRUCT(brw, &psp))
+      brw->state.dirty.brw |= BRW_NEW_PSP;
+}
+
+const struct brw_tracked_state brw_pipelined_state_pointers = {
+   .dirty = {
+      .brw = 0,
+      .cache = (CACHE_NEW_VS_UNIT |
+		CACHE_NEW_GS_UNIT |
+		CACHE_NEW_GS_PROG |
+		CACHE_NEW_CLIP_UNIT |
+		CACHE_NEW_SF_UNIT |
+		CACHE_NEW_WM_UNIT |
+		CACHE_NEW_CC_UNIT)
+   },
+   .update = upload_pipelined_state_pointers
+};
+
+static void upload_psp_urb_cbs(struct brw_context *brw )
+{
+   upload_pipelined_state_pointers(brw);
+   brw_upload_urb_fence(brw);
+   brw_upload_constant_buffer_state(brw);
+}
+
+
+const struct brw_tracked_state brw_psp_urb_cbs = {
+   .dirty = {
+      .brw = BRW_NEW_URB_FENCE,
+      .cache = (CACHE_NEW_VS_UNIT |
+		CACHE_NEW_GS_UNIT |
+		CACHE_NEW_GS_PROG |
+		CACHE_NEW_CLIP_UNIT |
+		CACHE_NEW_SF_UNIT |
+		CACHE_NEW_WM_UNIT |
+		CACHE_NEW_CC_UNIT)
+   },
+   .update = upload_psp_urb_cbs
+};
+
+/**
+ * Upload the depthbuffer offset and format.
+ *
+ * We have to do this per state validation as we need to emit the relocation
+ * in the batch buffer.
+ */
+static void upload_depthbuffer(struct brw_context *brw)
+{
+   struct pipe_surface *depth_surface = brw->attribs.FrameBuffer.zsbuf;
+
+   BEGIN_BATCH(5, INTEL_BATCH_NO_CLIPRECTS);
+   OUT_BATCH(CMD_DEPTH_BUFFER << 16 | (5 - 2));
+   if (depth_surface == NULL) {
+      OUT_BATCH((BRW_DEPTHFORMAT_D32_FLOAT << 18) |
+		(BRW_SURFACE_NULL << 29));
+      OUT_BATCH(0);
+      OUT_BATCH(0);
+      OUT_BATCH(0);
+   } else {
+      unsigned int format;
+
+      switch (depth_surface->cpp) {
+      case 2:
+	 format = BRW_DEPTHFORMAT_D16_UNORM;
+	 break;
+      case 4:
+	 if (depth_surface->format == PIPE_FORMAT_Z32_FLOAT)
+	    format = BRW_DEPTHFORMAT_D32_FLOAT;
+	 else
+	    format = BRW_DEPTHFORMAT_D24_UNORM_S8_UINT;
+	 break;
+      default:
+	 assert(0);
+	 return;
+      }
+
+      OUT_BATCH(((depth_surface->pitch * depth_surface->cpp) - 1) |
+		(format << 18) |
+		(BRW_TILEWALK_YMAJOR << 26) |
+//		(depth_surface->region->tiled << 27) |
+		(BRW_SURFACE_2D << 29));
+      OUT_RELOC(depth_surface->buffer,
+		PIPE_BUFFER_USAGE_GPU_READ | PIPE_BUFFER_USAGE_GPU_WRITE, 0);
+      OUT_BATCH((BRW_SURFACE_MIPMAPLAYOUT_BELOW << 1) |
+		((depth_surface->pitch - 1) << 6) |
+		((depth_surface->height - 1) << 19));
+      OUT_BATCH(0);
+   }
+   ADVANCE_BATCH();
+}
+
+const struct brw_tracked_state brw_depthbuffer = {
+   .dirty = {
+      .brw = BRW_NEW_SCENE,
+      .cache = 0
+   },
+   .update = upload_depthbuffer,
+};
+
+
+
+
+/***********************************************************************
+ * Polygon stipple packet
+ */
+
+static void upload_polygon_stipple(struct brw_context *brw)
+{
+   struct brw_polygon_stipple bps;
+   unsigned i;
+
+   memset(&bps, 0, sizeof(bps));
+   bps.header.opcode = CMD_POLY_STIPPLE_PATTERN;
+   bps.header.length = sizeof(bps)/4-2;
+
+   /* XXX: state tracker should send *all* state down initially!
+    */
+   if (brw->attribs.PolygonStipple)
+      for (i = 0; i < 32; i++)
+	 bps.stipple[i] = brw->attribs.PolygonStipple->stipple[31 - i]; /* invert */
+
+   BRW_CACHED_BATCH_STRUCT(brw, &bps);
+}
+
+const struct brw_tracked_state brw_polygon_stipple = {
+   .dirty = {
+      .brw = BRW_NEW_STIPPLE,
+      .cache = 0
+   },
+   .update = upload_polygon_stipple
+};
+
+
+/***********************************************************************
+ * Line stipple packet
+ */
+
+static void upload_line_stipple(struct brw_context *brw)
+{
+   struct brw_line_stipple bls;
+   float tmp;
+   int tmpi;
+
+   memset(&bls, 0, sizeof(bls));
+   bls.header.opcode = CMD_LINE_STIPPLE_PATTERN;
+   bls.header.length = sizeof(bls)/4 - 2;
+
+   bls.bits0.pattern = brw->attribs.Raster->line_stipple_pattern;
+   bls.bits1.repeat_count = brw->attribs.Raster->line_stipple_factor;
+
+   tmp = 1.0 / (float) brw->attribs.Raster->line_stipple_factor;
+   tmpi = tmp * (1<<13);
+
+
+   bls.bits1.inverse_repeat_count = tmpi;
+
+   BRW_CACHED_BATCH_STRUCT(brw, &bls);
+}
+
+const struct brw_tracked_state brw_line_stipple = {
+   .dirty = {
+      .brw = BRW_NEW_STIPPLE,
+      .cache = 0
+   },
+   .update = upload_line_stipple
+};
+
+
+/***********************************************************************
+ * Misc constant state packets
+ */
+
+static void upload_pipe_control(struct brw_context *brw)
+{
+   struct brw_pipe_control pc;
+
+   return;
+
+   memset(&pc, 0, sizeof(pc));
+
+   pc.header.opcode = CMD_PIPE_CONTROL;
+   pc.header.length = sizeof(pc)/4 - 2;
+   pc.header.post_sync_operation = PIPE_CONTROL_NOWRITE;
+
+   pc.header.instruction_state_cache_flush_enable = 1;
+
+   pc.bits1.dest_addr_type = PIPE_CONTROL_GTTWRITE_GLOBAL;
+
+   BRW_BATCH_STRUCT(brw, &pc);
+}
+
+const struct brw_tracked_state brw_pipe_control = {
+   .dirty = {
+      .brw = BRW_NEW_SCENE,
+      .cache = 0
+   },
+   .update = upload_pipe_control
+};
+
+
+/***********************************************************************
+ * Misc invarient state packets
+ */
+
+static void upload_invarient_state( struct brw_context *brw )
+{
+   {
+      struct brw_mi_flush flush;
+
+      memset(&flush, 0, sizeof(flush));      
+      flush.opcode = CMD_MI_FLUSH;
+      flush.flags = BRW_FLUSH_STATE_CACHE | BRW_FLUSH_READ_CACHE;
+      BRW_BATCH_STRUCT(brw, &flush);
+   }
+
+   {
+      /* 0x61040000  Pipeline Select */
+      /*     PipelineSelect            : 0 */
+      struct brw_pipeline_select ps;
+
+      memset(&ps, 0, sizeof(ps));
+      ps.header.opcode = CMD_PIPELINE_SELECT;
+      ps.header.pipeline_select = 0;
+      BRW_BATCH_STRUCT(brw, &ps);
+   }
+
+   {
+      struct brw_global_depth_offset_clamp gdo;
+      memset(&gdo, 0, sizeof(gdo));
+
+      /* Disable depth offset clamping.
+       */
+      gdo.header.opcode = CMD_GLOBAL_DEPTH_OFFSET_CLAMP;
+      gdo.header.length = sizeof(gdo)/4 - 2;
+      gdo.depth_offset_clamp = 0.0;
+
+      BRW_BATCH_STRUCT(brw, &gdo);
+   }
+
+
+   /* 0x61020000  State Instruction Pointer */
+   {
+      struct brw_system_instruction_pointer sip;
+      memset(&sip, 0, sizeof(sip));
+
+      sip.header.opcode = CMD_STATE_INSN_POINTER;
+      sip.header.length = 0;
+      sip.bits0.pad = 0;
+      sip.bits0.system_instruction_pointer = 0;
+      BRW_BATCH_STRUCT(brw, &sip);
+   }
+
+
+   {
+      struct brw_vf_statistics vfs;
+      memset(&vfs, 0, sizeof(vfs));
+
+      vfs.opcode = CMD_VF_STATISTICS;
+      if (BRW_DEBUG & DEBUG_STATS)
+	 vfs.statistics_enable = 1;
+
+      BRW_BATCH_STRUCT(brw, &vfs);
+   }
+
+   
+   {
+      struct brw_polygon_stipple_offset bpso;
+      
+      memset(&bpso, 0, sizeof(bpso));
+      bpso.header.opcode = CMD_POLY_STIPPLE_OFFSET;
+      bpso.header.length = sizeof(bpso)/4-2;      
+      bpso.bits0.x_offset = 0;
+      bpso.bits0.y_offset = 0;
+
+      BRW_BATCH_STRUCT(brw, &bpso);
+   }
+}
+
+const struct brw_tracked_state brw_invarient_state = {
+   .dirty = {
+      .brw = BRW_NEW_SCENE,
+      .cache = 0
+   },
+   .update = upload_invarient_state
+};
+
+/**
+ * Define the base addresses which some state is referenced from.
+ *
+ * This allows us to avoid having to emit relocations in many places for
+ * cached state, and instead emit pointers inside of large, mostly-static
+ * state pools.  This comes at the expense of memory, and more expensive cache
+ * misses.
+ */
+static void upload_state_base_address( struct brw_context *brw )
+{
+   /* Output the structure (brw_state_base_address) directly to the
+    * batchbuffer, so we can emit relocations inline.
+    */
+   BEGIN_BATCH(6, INTEL_BATCH_NO_CLIPRECTS);
+   OUT_BATCH(CMD_STATE_BASE_ADDRESS << 16 | (6 - 2));
+   OUT_RELOC(brw->pool[BRW_GS_POOL].buffer,
+	     PIPE_BUFFER_USAGE_GPU_READ,
+	     1); /* General state base address */
+   OUT_RELOC(brw->pool[BRW_SS_POOL].buffer,
+	     PIPE_BUFFER_USAGE_GPU_READ,
+	     1); /* Surface state base address */
+   OUT_BATCH(1); /* Indirect object base address */
+   OUT_BATCH(1); /* General state upper bound */
+   OUT_BATCH(1); /* Indirect object upper bound */
+   ADVANCE_BATCH();
+}
+
+
+const struct brw_tracked_state brw_state_base_address = {
+   .dirty = {
+      .brw = BRW_NEW_SCENE,
+      .cache = 0
+   },
+   .update = upload_state_base_address
+};
diff --git a/src/gallium/drivers/i965simple/brw_reg.h b/src/gallium/drivers/i965simple/brw_reg.h
new file mode 100644
index 0000000000..9e885c3b3b
--- /dev/null
+++ b/src/gallium/drivers/i965simple/brw_reg.h
@@ -0,0 +1,76 @@
+/**************************************************************************
+ * 
+ * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+#define CMD_MI				(0x0 << 29)
+#define CMD_2D				(0x2 << 29)
+#define CMD_3D				(0x3 << 29)
+
+#define MI_BATCH_BUFFER_END		(CMD_MI | 0xA << 23)
+
+/* Stalls command execution waiting for the given events to have occurred. */
+#define MI_WAIT_FOR_EVENT               (CMD_MI | (0x3 << 23))
+#define MI_WAIT_FOR_PLANE_B_FLIP        (1<<6)
+#define MI_WAIT_FOR_PLANE_A_FLIP        (1<<2)
+
+/* Primitive dispatch on 830-945 */
+#define _3DPRIMITIVE			(CMD_3D | (0x1f << 24))
+#define PRIM_INDIRECT            (1<<23)
+#define PRIM_INLINE              (0<<23)
+#define PRIM_INDIRECT_SEQUENTIAL (0<<17)
+#define PRIM_INDIRECT_ELTS       (1<<17)
+
+#define PRIM3D_TRILIST		(0x0<<18)
+#define PRIM3D_TRISTRIP 	(0x1<<18)
+#define PRIM3D_TRISTRIP_RVRSE	(0x2<<18)
+#define PRIM3D_TRIFAN		(0x3<<18)
+#define PRIM3D_POLY		(0x4<<18)
+#define PRIM3D_LINELIST 	(0x5<<18)
+#define PRIM3D_LINESTRIP	(0x6<<18)
+#define PRIM3D_RECTLIST 	(0x7<<18)
+#define PRIM3D_POINTLIST	(0x8<<18)
+#define PRIM3D_DIB		(0x9<<18)
+#define PRIM3D_MASK		(0x1f<<18)
+
+#define XY_SETUP_BLT_CMD		(CMD_2D | (0x01 << 22) | 6)
+
+#define XY_COLOR_BLT_CMD		(CMD_2D | (0x50 << 22) | 4)
+
+#define XY_SRC_COPY_BLT_CMD             (CMD_2D | (0x53 << 22) | 6)
+
+/* BR00 */
+#define XY_BLT_WRITE_ALPHA	(1 << 21)
+#define XY_BLT_WRITE_RGB	(1 << 20)
+#define XY_SRC_TILED		(1 << 15)
+#define XY_DST_TILED		(1 << 11)
+
+/* BR13 */
+#define BR13_565		(0x1 << 24)
+#define BR13_8888		(0x3 << 24)
+
+#define FENCE_LINEAR 0
+#define FENCE_XMAJOR 1
+#define FENCE_YMAJOR 2
diff --git a/src/gallium/drivers/i965simple/brw_sf.c b/src/gallium/drivers/i965simple/brw_sf.c
new file mode 100644
index 0000000000..7c83b81c85
--- /dev/null
+++ b/src/gallium/drivers/i965simple/brw_sf.c
@@ -0,0 +1,351 @@
+/*
+ Copyright (C) Intel Corp.  2006.  All Rights Reserved.
+ Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to
+ develop this 3D driver.
+
+ Permission is hereby granted, free of charge, to any person obtaining
+ a copy of this software and associated documentation files (the
+ "Software"), to deal in the Software without restriction, including
+ without limitation the rights to use, copy, modify, merge, publish,
+ distribute, sublicense, and/or sell copies of the Software, and to
+ permit persons to whom the Software is furnished to do so, subject to
+ the following conditions:
+
+ The above copyright notice and this permission notice (including the
+ next paragraph) shall be included in all copies or substantial
+ portions of the Software.
+
+ THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
+ IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
+ LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
+ OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+
+ **********************************************************************/
+ /*
+  * Authors:
+  *   Keith Whitwell <keith@tungstengraphics.com>
+  */
+
+
+#include "brw_defines.h"
+#include "brw_context.h"
+#include "brw_eu.h"
+#include "brw_util.h"
+#include "brw_sf.h"
+#include "brw_state.h"
+#include "tgsi/util/tgsi_parse.h"
+
+
+static void compile_sf_prog( struct brw_context *brw,
+			     struct brw_sf_prog_key *key )
+{
+   struct brw_sf_compile c;
+   const unsigned *program;
+   unsigned program_size;
+
+   memset(&c, 0, sizeof(c));
+
+   /* Begin the compilation:
+    */
+   brw_init_compile(&c.func);
+
+   c.key = *key;
+
+
+   c.nr_attrs = c.key.vp_output_count;
+   c.nr_attr_regs = (c.nr_attrs+1)/2;
+
+   c.nr_setup_attrs = c.key.fp_input_count + 1; /* +1 for position */
+   c.nr_setup_regs = (c.nr_setup_attrs+1)/2;
+
+   c.prog_data.urb_read_length = c.nr_attr_regs;
+   c.prog_data.urb_entry_size = c.nr_setup_regs * 2;
+
+
+   /* Which primitive?  Or all three?
+    */
+   switch (key->primitive) {
+   case SF_TRIANGLES:
+      c.nr_verts = 3;
+      brw_emit_tri_setup( &c );
+      break;
+   case SF_LINES:
+      c.nr_verts = 2;
+      brw_emit_line_setup( &c );
+      break;
+   case SF_POINTS:
+      c.nr_verts = 1;
+      brw_emit_point_setup( &c );
+      break;
+
+   case SF_UNFILLED_TRIS:
+   default:
+      assert(0);
+      return;
+   }
+
+
+
+   /* get the program
+    */
+   program = brw_get_program(&c.func, &program_size);
+
+   /* Upload
+    */
+   brw->sf.prog_gs_offset = brw_upload_cache( &brw->cache[BRW_SF_PROG],
+					      &c.key,
+					      sizeof(c.key),
+					      program,
+					      program_size,
+					      &c.prog_data,
+					      &brw->sf.prog_data );
+}
+
+
+static boolean search_cache( struct brw_context *brw,
+			       struct brw_sf_prog_key *key )
+{
+   return brw_search_cache(&brw->cache[BRW_SF_PROG],
+			   key, sizeof(*key),
+			   &brw->sf.prog_data,
+			   &brw->sf.prog_gs_offset);
+}
+
+
+/* Calculate interpolants for triangle and line rasterization.
+ */
+static void upload_sf_prog( struct brw_context *brw )
+{
+   const struct brw_fragment_program *fs = brw->attribs.FragmentProgram;
+   struct brw_sf_prog_key key;
+   struct tgsi_parse_context parse;
+   int i, done = 0;
+
+
+   memset(&key, 0, sizeof(key));
+
+   /* Populate the key, noting state dependencies:
+    */
+   /* CACHE_NEW_VS_PROG */
+   key.vp_output_count = brw->vs.prog_data->outputs_written;
+
+   /* BRW_NEW_FS */
+   key.fp_input_count = brw->attribs.FragmentProgram->info.nr_regs[TGSI_FILE_INPUT];
+
+
+   /* BRW_NEW_REDUCED_PRIMITIVE */
+   switch (brw->reduced_primitive) {
+   case PIPE_PRIM_TRIANGLES:
+//      if (key.attrs & (1<<VERT_RESULT_EDGE))
+//	 key.primitive = SF_UNFILLED_TRIS;
+//      else
+      key.primitive = SF_TRIANGLES;
+      break;
+   case PIPE_PRIM_LINES:
+      key.primitive = SF_LINES;
+      break;
+   case PIPE_PRIM_POINTS:
+      key.primitive = SF_POINTS;
+      break;
+   }
+
+
+
+   /* Scan fp inputs to figure out what interpolation modes are
+    * required for each incoming vp output.  There is an assumption
+    * that the state tracker makes sure there is a 1:1 linkage between
+    * these sets of attributes (XXX: position??)
+    */
+   tgsi_parse_init( &parse, fs->program.tokens );
+   while( !done &&
+	  !tgsi_parse_end_of_tokens( &parse ) ) 
+   {
+      tgsi_parse_token( &parse );
+
+      switch( parse.FullToken.Token.Type ) {
+      case TGSI_TOKEN_TYPE_DECLARATION:
+	 if (parse.FullToken.FullDeclaration.Declaration.File == TGSI_FILE_INPUT) 
+	 {
+	    int first = parse.FullToken.FullDeclaration.u.DeclarationRange.First;
+	    int last = parse.FullToken.FullDeclaration.u.DeclarationRange.Last;
+	    int interp_mode = parse.FullToken.FullDeclaration.Interpolation.Interpolate;
+	    //int semantic = parse.FullToken.FullDeclaration.Semantic.SemanticName;
+	    //int semantic_index = parse.FullToken.FullDeclaration.Semantic.SemanticIndex;
+
+	    debug_printf("fs input %d..%d interp mode %d\n", first, last, interp_mode);
+	    
+	    switch (interp_mode) {
+	    case TGSI_INTERPOLATE_CONSTANT:
+	       for (i = first; i <= last; i++) 
+		  key.const_mask |= (1 << i);
+	       break;
+	    case TGSI_INTERPOLATE_LINEAR:
+	       for (i = first; i <= last; i++) 
+		  key.linear_mask |= (1 << i);
+	       break;
+	    case TGSI_INTERPOLATE_PERSPECTIVE:
+	       for (i = first; i <= last; i++) 
+		  key.persp_mask |= (1 << i);
+	       break;
+	    default:
+	       break;
+	    }
+
+	    /* Also need stuff for flat shading, twosided color.
+	     */
+
+	 }
+	 break;
+      default:
+	 done = 1;
+	 break;
+      }
+   }
+
+   /* Hack: Adjust for position.  Optimize away when not required (ie
+    * for perspective interpolation).
+    */
+   key.persp_mask <<= 1;
+   key.linear_mask <<= 1; 
+   key.linear_mask |= 1;
+   key.const_mask <<= 1;
+
+   debug_printf("key.persp_mask: %x\n", key.persp_mask);
+   debug_printf("key.linear_mask: %x\n", key.linear_mask);
+   debug_printf("key.const_mask: %x\n", key.const_mask);
+
+
+//   key.do_point_sprite = brw->attribs.Point->PointSprite;
+//   key.SpriteOrigin = brw->attribs.Point->SpriteOrigin;
+
+//   key.do_flat_shading = (brw->attribs.Raster->flatshade);
+//   key.do_twoside_color = (brw->attribs.Light->Enabled && brw->attribs.Light->Model.TwoSide);
+
+//   if (key.do_twoside_color)
+//      key.frontface_ccw = (brw->attribs.Polygon->FrontFace == GL_CCW);
+
+
+   if (!search_cache(brw, &key))
+      compile_sf_prog( brw, &key );
+}
+
+
+const struct brw_tracked_state brw_sf_prog = {
+   .dirty = {
+      .brw   = (BRW_NEW_RASTERIZER |
+		BRW_NEW_REDUCED_PRIMITIVE |
+		BRW_NEW_VS |
+		BRW_NEW_FS),
+      .cache = 0,
+   },
+   .update = upload_sf_prog
+};
+
+
+
+#if 0
+/* Build a struct like the one we'd like the state tracker to pass to
+ * us.
+ */
+static void update_sf_linkage( struct brw_context *brw )
+{
+   const struct brw_vertex_program *vs = brw->attribs.VertexProgram;
+   const struct brw_fragment_program *fs = brw->attribs.FragmentProgram;
+   struct pipe_setup_linkage state;
+   struct tgsi_parse_context parse;
+
+   int i, j;
+   int nr_vp_outputs = 0;
+   int done = 0;
+
+   struct { 
+      unsigned semantic:8;
+      unsigned semantic_index:16;
+   } fp_semantic[32], vp_semantic[32];
+
+   memset(&state, 0, sizeof(state));
+
+   state.fp_input_count = 0;
+
+
+
+   
+
+
+   assert(state.fp_input_count == fs->program.num_inputs);
+
+      
+   /* Then scan vp outputs
+    */
+   done = 0;
+   tgsi_parse_init( &parse, vs->program.tokens );
+   while( !done &&
+	  !tgsi_parse_end_of_tokens( &parse ) ) 
+   {
+      tgsi_parse_token( &parse );
+
+      switch( parse.FullToken.Token.Type ) {
+      case TGSI_TOKEN_TYPE_DECLARATION:
+	 if (parse.FullToken.FullDeclaration.Declaration.File == TGSI_FILE_INPUT) 
+	 {
+	    int first = parse.FullToken.FullDeclaration.u.DeclarationRange.First;
+	    int last = parse.FullToken.FullDeclaration.u.DeclarationRange.Last;
+
+	    for (i = first; i < last; i++) {
+	       vp_semantic[i].semantic = 
+		  parse.FullToken.FullDeclaration.Semantic.SemanticName;
+	       vp_semantic[i].semantic_index = 
+		  parse.FullToken.FullDeclaration.Semantic.SemanticIndex;
+	    }
+	    
+	    assert(last > nr_vp_outputs);
+	    nr_vp_outputs = last;
+	 }
+	 break;
+      default:
+	 done = 1;
+	 break;
+      }
+   }
+
+
+   /* Now match based on semantic information.
+    */
+   for (i = 0; i< state.fp_input_count; i++) {
+      for (j = 0; j < nr_vp_outputs; j++) {
+	 if (fp_semantic[i].semantic == vp_semantic[j].semantic &&
+	     fp_semantic[i].semantic_index == vp_semantic[j].semantic_index) {
+	    state.fp_input[i].vp_output = j;
+	 }
+      }
+      if (fp_semantic[i].semantic == TGSI_SEMANTIC_COLOR) {
+	 for (j = 0; j < nr_vp_outputs; j++) {
+	    if (TGSI_SEMANTIC_BCOLOR == vp_semantic[j].semantic &&
+		fp_semantic[i].semantic_index == vp_semantic[j].semantic_index) {
+	       state.fp_input[i].bf_vp_output = j;
+	    }
+	 }
+      }
+   }
+
+   if (memcmp(&brw->sf.linkage, &state, sizeof(state)) != 0) {
+      brw->sf.linkage = state;
+      brw->state.dirty.brw |= BRW_NEW_SF_LINKAGE;
+   }
+}
+
+
+const struct brw_tracked_state brw_sf_linkage = {
+   .dirty = {
+      .brw   = (BRW_NEW_VS |
+		BRW_NEW_FS),
+      .cache = 0,
+   },
+   .update = update_sf_linkage
+};
+
+
+#endif
diff --git a/src/gallium/drivers/i965simple/brw_sf.h b/src/gallium/drivers/i965simple/brw_sf.h
new file mode 100644
index 0000000000..b7ada47560
--- /dev/null
+++ b/src/gallium/drivers/i965simple/brw_sf.h
@@ -0,0 +1,122 @@
+/*
+ Copyright (C) Intel Corp.  2006.  All Rights Reserved.
+ Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to
+ develop this 3D driver.
+
+ Permission is hereby granted, free of charge, to any person obtaining
+ a copy of this software and associated documentation files (the
+ "Software"), to deal in the Software without restriction, including
+ without limitation the rights to use, copy, modify, merge, publish,
+ distribute, sublicense, and/or sell copies of the Software, and to
+ permit persons to whom the Software is furnished to do so, subject to
+ the following conditions:
+
+ The above copyright notice and this permission notice (including the
+ next paragraph) shall be included in all copies or substantial
+ portions of the Software.
+
+ THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
+ IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
+ LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
+ OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+
+ **********************************************************************/
+ /*
+  * Authors:
+  *   Keith Whitwell <keith@tungstengraphics.com>
+  */
+
+
+#ifndef BRW_SF_H
+#define BRW_SF_H
+
+#include "brw_context.h"
+#include "brw_eu.h"
+
+
+#define SF_POINTS    0
+#define SF_LINES     1
+#define SF_TRIANGLES 2
+#define SF_UNFILLED_TRIS   3
+
+
+
+struct brw_sf_prog_key {
+   unsigned vp_output_count:5;
+   unsigned fp_input_count:5;
+
+   unsigned primitive:2;
+   unsigned do_twoside_color:1;
+   unsigned do_flat_shading:1;
+   unsigned frontface_ccw:1;
+   unsigned do_point_sprite:1;
+
+   /* Interpolation masks;
+    */
+   unsigned linear_mask;
+   unsigned persp_mask;
+   unsigned const_mask;
+
+
+//   int SpriteOrigin;
+};
+
+struct brw_sf_point_tex {
+	boolean CoordReplace;
+};
+
+struct brw_sf_compile {
+   struct brw_compile func;
+   struct brw_sf_prog_key key;
+   struct brw_sf_prog_data prog_data;
+
+   struct brw_reg pv;
+   struct brw_reg det;
+   struct brw_reg dx0;
+   struct brw_reg dx2;
+   struct brw_reg dy0;
+   struct brw_reg dy2;
+
+   /* z and 1/w passed in seperately:
+    */
+   struct brw_reg z[3];
+   struct brw_reg inv_w[3];
+
+   /* The vertices:
+    */
+   struct brw_reg vert[3];
+
+    /* Temporaries, allocated after last vertex reg.
+    */
+   struct brw_reg inv_det;
+   struct brw_reg a1_sub_a0;
+   struct brw_reg a2_sub_a0;
+   struct brw_reg tmp;
+
+   struct brw_reg m1Cx;
+   struct brw_reg m2Cy;
+   struct brw_reg m3C0;
+
+   unsigned nr_verts;
+   unsigned nr_attrs;
+   unsigned nr_attr_regs;
+   unsigned nr_setup_attrs;
+   unsigned nr_setup_regs;
+#if 0
+   ubyte attr_to_idx[VERT_RESULT_MAX];
+   ubyte idx_to_attr[VERT_RESULT_MAX];
+   struct brw_sf_point_tex point_attrs[VERT_RESULT_MAX];
+#endif
+};
+
+
+void brw_emit_tri_setup( struct brw_sf_compile *c );
+void brw_emit_line_setup( struct brw_sf_compile *c );
+void brw_emit_point_setup( struct brw_sf_compile *c );
+void brw_emit_point_sprite_setup( struct brw_sf_compile *c );
+void brw_emit_anyprim_setup( struct brw_sf_compile *c );
+
+#endif
diff --git a/src/gallium/drivers/i965simple/brw_sf_emit.c b/src/gallium/drivers/i965simple/brw_sf_emit.c
new file mode 100644
index 0000000000..78d6fa5e9e
--- /dev/null
+++ b/src/gallium/drivers/i965simple/brw_sf_emit.c
@@ -0,0 +1,382 @@
+/*
+ Copyright (C) Intel Corp.  2006.  All Rights Reserved.
+ Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to
+ develop this 3D driver.
+
+ Permission is hereby granted, free of charge, to any person obtaining
+ a copy of this software and associated documentation files (the
+ "Software"), to deal in the Software without restriction, including
+ without limitation the rights to use, copy, modify, merge, publish,
+ distribute, sublicense, and/or sell copies of the Software, and to
+ permit persons to whom the Software is furnished to do so, subject to
+ the following conditions:
+
+ The above copyright notice and this permission notice (including the
+ next paragraph) shall be included in all copies or substantial
+ portions of the Software.
+
+ THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
+ IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
+ LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
+ OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+
+ **********************************************************************/
+ /*
+  * Authors:
+  *   Keith Whitwell <keith@tungstengraphics.com>
+  */
+
+
+#include "brw_defines.h"
+#include "brw_context.h"
+#include "brw_eu.h"
+#include "brw_util.h"
+#include "brw_sf.h"
+
+
+
+/***********************************************************************
+ * Triangle setup.
+ */
+
+
+static void alloc_regs( struct brw_sf_compile *c )
+{
+   unsigned reg, i;
+
+   /* Values computed by fixed function unit:
+    */
+   c->pv  = retype(brw_vec1_grf(1, 1), BRW_REGISTER_TYPE_UD);
+   c->det = brw_vec1_grf(1, 2);
+   c->dx0 = brw_vec1_grf(1, 3);
+   c->dx2 = brw_vec1_grf(1, 4);
+   c->dy0 = brw_vec1_grf(1, 5);
+   c->dy2 = brw_vec1_grf(1, 6);
+
+   /* z and 1/w passed in seperately:
+    */
+   c->z[0]     = brw_vec1_grf(2, 0);
+   c->inv_w[0] = brw_vec1_grf(2, 1);
+   c->z[1]     = brw_vec1_grf(2, 2);
+   c->inv_w[1] = brw_vec1_grf(2, 3);
+   c->z[2]     = brw_vec1_grf(2, 4);
+   c->inv_w[2] = brw_vec1_grf(2, 5);
+
+   /* The vertices:
+    */
+   reg = 3;
+   for (i = 0; i < c->nr_verts; i++) {
+      c->vert[i] = brw_vec8_grf(reg, 0);
+      reg += c->nr_attr_regs;
+   }
+
+   /* Temporaries, allocated after last vertex reg.
+    */
+   c->inv_det = brw_vec1_grf(reg, 0);  reg++;
+   c->a1_sub_a0 = brw_vec8_grf(reg, 0);  reg++;
+   c->a2_sub_a0 = brw_vec8_grf(reg, 0);  reg++;
+   c->tmp = brw_vec8_grf(reg, 0);  reg++;
+
+   /* Note grf allocation:
+    */
+   c->prog_data.total_grf = reg;
+
+
+   /* Outputs of this program - interpolation coefficients for
+    * rasterization:
+    */
+   c->m1Cx = brw_vec8_reg(BRW_MESSAGE_REGISTER_FILE, 1, 0);
+   c->m2Cy = brw_vec8_reg(BRW_MESSAGE_REGISTER_FILE, 2, 0);
+   c->m3C0 = brw_vec8_reg(BRW_MESSAGE_REGISTER_FILE, 3, 0);
+}
+
+
+static void copy_z_inv_w( struct brw_sf_compile *c )
+{
+   struct brw_compile *p = &c->func;
+   unsigned i;
+
+   brw_push_insn_state(p);
+
+   /* Copy both scalars with a single MOV:
+    */
+   for (i = 0; i < c->nr_verts; i++)
+      brw_MOV(p, vec2(suboffset(c->vert[i], 2)), vec2(c->z[i]));
+
+   brw_pop_insn_state(p);
+}
+
+
+static void invert_det( struct brw_sf_compile *c)
+{
+   brw_math(&c->func,
+	    c->inv_det,
+	    BRW_MATH_FUNCTION_INV,
+	    BRW_MATH_SATURATE_NONE,
+	    0,
+	    c->det,
+	    BRW_MATH_DATA_SCALAR,
+	    BRW_MATH_PRECISION_FULL);
+
+}
+
+#define NON_PERPECTIVE_ATTRS  (FRAG_BIT_WPOS | \
+                               FRAG_BIT_COL0 | \
+			       FRAG_BIT_COL1)
+
+static boolean calculate_masks( struct brw_sf_compile *c,
+				  unsigned reg,
+				  ushort *pc,
+				  ushort *pc_persp,
+				  ushort *pc_linear)
+{
+   boolean is_last_attr = (reg == c->nr_setup_regs - 1);
+   unsigned persp_mask = c->key.persp_mask;
+   unsigned linear_mask = c->key.linear_mask;
+
+   debug_printf("persp_mask: %x\n", persp_mask);
+   debug_printf("linear_mask: %x\n", linear_mask);
+
+   *pc_persp = 0;
+   *pc_linear = 0;
+   *pc = 0xf;
+
+   if (persp_mask & (1 << (reg*2)))
+      *pc_persp = 0xf;
+
+   if (linear_mask & (1 << (reg*2)))
+      *pc_linear = 0xf;
+
+   /* Maybe only processs one attribute on the final round:
+    */
+   if (reg*2+1 < c->nr_setup_attrs) {
+      *pc |= 0xf0;
+
+      if (persp_mask & (1 << (reg*2+1)))
+	 *pc_persp |= 0xf0;
+
+      if (linear_mask & (1 << (reg*2+1)))
+	 *pc_linear |= 0xf0;
+   }
+
+   debug_printf("pc: %x\n", *pc);
+   debug_printf("pc_persp: %x\n", *pc_persp);
+   debug_printf("pc_linear: %x\n", *pc_linear);
+   
+
+   return is_last_attr;
+}
+
+
+
+void brw_emit_tri_setup( struct brw_sf_compile *c )
+{
+   struct brw_compile *p = &c->func;
+   unsigned i;
+
+   debug_printf("%s START ==============\n", __FUNCTION__);
+
+   c->nr_verts = 3;
+   alloc_regs(c);
+   invert_det(c);
+   copy_z_inv_w(c);
+
+
+   for (i = 0; i < c->nr_setup_regs; i++)
+   {
+      /* Pair of incoming attributes:
+       */
+      struct brw_reg a0 = offset(c->vert[0], i);
+      struct brw_reg a1 = offset(c->vert[1], i);
+      struct brw_reg a2 = offset(c->vert[2], i);
+      ushort pc = 0, pc_persp = 0, pc_linear = 0;
+      boolean last = calculate_masks(c, i, &pc, &pc_persp, &pc_linear);
+
+      if (pc_persp)
+      {
+	 brw_set_predicate_control_flag_value(p, pc_persp);
+	 brw_MUL(p, a0, a0, c->inv_w[0]);
+	 brw_MUL(p, a1, a1, c->inv_w[1]);
+	 brw_MUL(p, a2, a2, c->inv_w[2]);
+      }
+
+
+      /* Calculate coefficients for interpolated values:
+       */
+      if (pc_linear)
+      {
+	 brw_set_predicate_control_flag_value(p, pc_linear);
+
+	 brw_ADD(p, c->a1_sub_a0, a1, negate(a0));
+	 brw_ADD(p, c->a2_sub_a0, a2, negate(a0));
+
+	 /* calculate dA/dx
+	  */
+	 brw_MUL(p, brw_null_reg(), c->a1_sub_a0, c->dy2);
+	 brw_MAC(p, c->tmp, c->a2_sub_a0, negate(c->dy0));
+	 brw_MUL(p, c->m1Cx, c->tmp, c->inv_det);
+
+	 /* calculate dA/dy
+	  */
+	 brw_MUL(p, brw_null_reg(), c->a2_sub_a0, c->dx0);
+	 brw_MAC(p, c->tmp, c->a1_sub_a0, negate(c->dx2));
+	 brw_MUL(p, c->m2Cy, c->tmp, c->inv_det);
+      }
+
+      {
+	 brw_set_predicate_control_flag_value(p, pc);
+	 /* start point for interpolation
+	  */
+	 brw_MOV(p, c->m3C0, a0);
+
+	 /* Copy m0..m3 to URB.  m0 is implicitly copied from r0 in
+	  * the send instruction:
+	  */
+	 brw_urb_WRITE(p,
+		       brw_null_reg(),
+		       0,
+		       brw_vec8_grf(0, 0), /* r0, will be copied to m0 */
+		       0, 	/* allocate */
+		       1,	/* used */
+		       4, 	/* msg len */
+		       0,	/* response len */
+		       last,	/* eot */
+		       last, 	/* writes complete */
+		       i*4,	/* offset */
+		       BRW_URB_SWIZZLE_TRANSPOSE); /* XXX: Swizzle control "SF to windower" */
+      }
+   }
+
+   debug_printf("%s DONE ==============\n", __FUNCTION__);
+
+}
+
+
+
+void brw_emit_line_setup( struct brw_sf_compile *c )
+{
+   struct brw_compile *p = &c->func;
+   unsigned i;
+
+
+   c->nr_verts = 2;
+   alloc_regs(c);
+   invert_det(c);
+   copy_z_inv_w(c);
+
+   for (i = 0; i < c->nr_setup_regs; i++)
+   {
+      /* Pair of incoming attributes:
+       */
+      struct brw_reg a0 = offset(c->vert[0], i);
+      struct brw_reg a1 = offset(c->vert[1], i);
+      ushort pc, pc_persp, pc_linear;
+      boolean last = calculate_masks(c, i, &pc, &pc_persp, &pc_linear);
+
+      if (pc_persp)
+      {
+	 brw_set_predicate_control_flag_value(p, pc_persp);
+	 brw_MUL(p, a0, a0, c->inv_w[0]);
+	 brw_MUL(p, a1, a1, c->inv_w[1]);
+      }
+
+      /* Calculate coefficients for position, color:
+       */
+      if (pc_linear) {
+	 brw_set_predicate_control_flag_value(p, pc_linear);
+
+	 brw_ADD(p, c->a1_sub_a0, a1, negate(a0));
+
+ 	 brw_MUL(p, c->tmp, c->a1_sub_a0, c->dx0);
+	 brw_MUL(p, c->m1Cx, c->tmp, c->inv_det);
+
+	 brw_MUL(p, c->tmp, c->a1_sub_a0, c->dy0);
+	 brw_MUL(p, c->m2Cy, c->tmp, c->inv_det);
+      }
+
+      {
+	 brw_set_predicate_control_flag_value(p, pc);
+
+	 /* start point for interpolation
+	  */
+	 brw_MOV(p, c->m3C0, a0);
+
+	 /* Copy m0..m3 to URB.
+	  */
+	 brw_urb_WRITE(p,
+		       brw_null_reg(),
+		       0,
+		       brw_vec8_grf(0, 0),
+		       0, 	/* allocate */
+		       1, 	/* used */
+		       4, 	/* msg len */
+		       0,	/* response len */
+		       last, 	/* eot */
+		       last, 	/* writes complete */
+		       i*4,	/* urb destination offset */
+		       BRW_URB_SWIZZLE_TRANSPOSE);
+      }
+   }
+}
+
+
+/* Points setup - several simplifications as all attributes are
+ * constant across the face of the point (point sprites excluded!)
+ */
+void brw_emit_point_setup( struct brw_sf_compile *c )
+{
+   struct brw_compile *p = &c->func;
+   unsigned i;
+
+   c->nr_verts = 1;
+   alloc_regs(c);
+   copy_z_inv_w(c);
+
+   brw_MOV(p, c->m1Cx, brw_imm_ud(0)); /* zero - move out of loop */
+   brw_MOV(p, c->m2Cy, brw_imm_ud(0)); /* zero - move out of loop */
+
+   for (i = 0; i < c->nr_setup_regs; i++)
+   {
+      struct brw_reg a0 = offset(c->vert[0], i);
+      ushort pc, pc_persp, pc_linear;
+      boolean last = calculate_masks(c, i, &pc, &pc_persp, &pc_linear);
+
+      if (pc_persp)
+      {
+	 /* This seems odd as the values are all constant, but the
+	  * fragment shader will be expecting it:
+	  */
+	 brw_set_predicate_control_flag_value(p, pc_persp);
+	 brw_MUL(p, a0, a0, c->inv_w[0]);
+      }
+
+
+      /* The delta values are always zero, just send the starting
+       * coordinate.  Again, this is to fit in with the interpolation
+       * code in the fragment shader.
+       */
+      {
+	 brw_set_predicate_control_flag_value(p, pc);
+
+	 brw_MOV(p, c->m3C0, a0); /* constant value */
+
+	 /* Copy m0..m3 to URB.
+	  */
+	 brw_urb_WRITE(p,
+		       brw_null_reg(),
+		       0,
+		       brw_vec8_grf(0, 0),
+		       0, 	/* allocate */
+		       1,	/* used */
+		       4, 	/* msg len */
+		       0,	/* response len */
+		       last, 	/* eot */
+		       last, 	/* writes complete */
+		       i*4,	/* urb destination offset */
+		       BRW_URB_SWIZZLE_TRANSPOSE);
+      }
+   }
+}
diff --git a/src/gallium/drivers/i965simple/brw_sf_state.c b/src/gallium/drivers/i965simple/brw_sf_state.c
new file mode 100644
index 0000000000..9acd3ea61b
--- /dev/null
+++ b/src/gallium/drivers/i965simple/brw_sf_state.c
@@ -0,0 +1,180 @@
+/*
+ Copyright (C) Intel Corp.  2006.  All Rights Reserved.
+ Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to
+ develop this 3D driver.
+
+ Permission is hereby granted, free of charge, to any person obtaining
+ a copy of this software and associated documentation files (the
+ "Software"), to deal in the Software without restriction, including
+ without limitation the rights to use, copy, modify, merge, publish,
+ distribute, sublicense, and/or sell copies of the Software, and to
+ permit persons to whom the Software is furnished to do so, subject to
+ the following conditions:
+
+ The above copyright notice and this permission notice (including the
+ next paragraph) shall be included in all copies or substantial
+ portions of the Software.
+
+ THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
+ IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
+ LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
+ OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+
+ **********************************************************************/
+ /*
+  * Authors:
+  *   Keith Whitwell <keith@tungstengraphics.com>
+  */
+
+
+
+#include "brw_context.h"
+#include "brw_state.h"
+#include "brw_defines.h"
+#include "pipe/p_util.h"
+
+static void upload_sf_vp(struct brw_context *brw)
+{
+   struct brw_sf_viewport sfv;
+
+   memset(&sfv, 0, sizeof(sfv));
+
+
+   /* BRW_NEW_VIEWPORT */
+   {
+      const float *scale = brw->attribs.Viewport.scale;
+      const float *trans = brw->attribs.Viewport.translate;
+
+      sfv.viewport.m00 = scale[0];
+      sfv.viewport.m11 = scale[1];
+      sfv.viewport.m22 = scale[2]; 
+      sfv.viewport.m30 = trans[0];
+      sfv.viewport.m31 = trans[1];
+      sfv.viewport.m32 = trans[2];
+   }
+
+   /* _NEW_SCISSOR */
+   sfv.scissor.xmin = brw->attribs.Scissor.minx;
+   sfv.scissor.xmax = brw->attribs.Scissor.maxx - 1;
+   sfv.scissor.ymin = brw->attribs.Scissor.miny;
+   sfv.scissor.ymax = brw->attribs.Scissor.maxy - 1;
+
+   brw->sf.vp_gs_offset = brw_cache_data( &brw->cache[BRW_SF_VP], &sfv );
+}
+
+const struct brw_tracked_state brw_sf_vp = {
+   .dirty = {
+      .brw   = (BRW_NEW_SCISSOR |
+		BRW_NEW_VIEWPORT),
+      .cache = 0
+   },
+   .update = upload_sf_vp
+};
+
+static void upload_sf_unit( struct brw_context *brw )
+{
+   struct brw_sf_unit_state sf;
+   memset(&sf, 0, sizeof(sf));
+
+   /* CACHE_NEW_SF_PROG */
+   sf.thread0.grf_reg_count = align(brw->sf.prog_data->total_grf, 16) / 16 - 1;
+   sf.thread0.kernel_start_pointer = brw->sf.prog_gs_offset >> 6;
+   sf.thread3.urb_entry_read_length = brw->sf.prog_data->urb_read_length;
+
+   sf.thread1.floating_point_mode = BRW_FLOATING_POINT_NON_IEEE_754;
+   sf.thread3.dispatch_grf_start_reg = 3;
+   sf.thread3.urb_entry_read_offset = 1;
+
+   /* BRW_NEW_URB_FENCE */
+   sf.thread4.nr_urb_entries = brw->urb.nr_sf_entries;
+   sf.thread4.urb_entry_allocation_size = brw->urb.sfsize - 1;
+   sf.thread4.max_threads = MIN2(12, brw->urb.nr_sf_entries / 2) - 1;
+
+   if (BRW_DEBUG & DEBUG_SINGLE_THREAD)
+      sf.thread4.max_threads = 0;
+
+   if (BRW_DEBUG & DEBUG_STATS)
+      sf.thread4.stats_enable = 1;
+
+   /* CACHE_NEW_SF_VP */
+   sf.sf5.sf_viewport_state_offset = brw->sf.vp_gs_offset >> 5;
+   sf.sf5.viewport_transform = 1;
+
+   /* BRW_NEW_RASTER */
+   if (brw->attribs.Raster->scissor)
+      sf.sf6.scissor = 1;
+
+#if 0
+   if (brw->attribs.Polygon->FrontFace == GL_CCW)
+      sf.sf5.front_winding = BRW_FRONTWINDING_CCW;
+   else
+      sf.sf5.front_winding = BRW_FRONTWINDING_CW;
+
+
+   if (brw->attribs.Polygon->CullFlag) {
+      switch (brw->attribs.Polygon->CullFaceMode) {
+      case GL_FRONT:
+	 sf.sf6.cull_mode = BRW_CULLMODE_FRONT;
+	 break;
+      case GL_BACK:
+	 sf.sf6.cull_mode = BRW_CULLMODE_BACK;
+	 break;
+      case GL_FRONT_AND_BACK:
+	 sf.sf6.cull_mode = BRW_CULLMODE_BOTH;
+	 break;
+      default:
+	 assert(0);
+	 break;
+      }
+   }
+   else
+      sf.sf6.cull_mode = BRW_CULLMODE_NONE;
+#else
+   sf.sf5.front_winding = BRW_FRONTWINDING_CCW;
+   sf.sf6.cull_mode = BRW_CULLMODE_NONE;
+#endif
+
+   sf.sf6.line_width = CLAMP(brw->attribs.Raster->line_width, 1.0, 5.0) * (1<<1);
+
+   sf.sf6.line_endcap_aa_region_width = 1;
+   if (brw->attribs.Raster->line_smooth)
+      sf.sf6.aa_enable = 1;
+   else if (sf.sf6.line_width <= 0x2)
+       sf.sf6.line_width = 0;
+
+   sf.sf6.point_rast_rule = 1;	/* opengl conventions */
+
+   sf.sf7.sprite_point = brw->attribs.Raster->point_sprite;
+   sf.sf7.point_size = CLAMP(brw->attribs.Raster->line_width, 1.0, 255.0) * (1<<3);
+   sf.sf7.use_point_size_state = !brw->attribs.Raster->point_size_per_vertex;
+
+   /* might be BRW_NEW_PRIMITIVE if we have to adjust pv for polygons:
+    */
+   sf.sf7.trifan_pv = 2;
+   sf.sf7.linestrip_pv = 1;
+   sf.sf7.tristrip_pv = 2;
+   sf.sf7.line_last_pixel_enable = 0;
+
+   /* Set bias for OpenGL rasterization rules:
+    */
+   sf.sf6.dest_org_vbias = 0x8;
+   sf.sf6.dest_org_hbias = 0x8;
+
+   brw->sf.state_gs_offset = brw_cache_data( &brw->cache[BRW_SF_UNIT], &sf );
+}
+
+
+const struct brw_tracked_state brw_sf_unit = {
+   .dirty = {
+      .brw   = (BRW_NEW_RASTERIZER |
+		BRW_NEW_URB_FENCE),
+      .cache = (CACHE_NEW_SF_VP |
+		CACHE_NEW_SF_PROG)
+   },
+   .update = upload_sf_unit
+};
+
+
diff --git a/src/gallium/drivers/i965simple/brw_shader_info.c b/src/gallium/drivers/i965simple/brw_shader_info.c
new file mode 100644
index 0000000000..431b45466a
--- /dev/null
+++ b/src/gallium/drivers/i965simple/brw_shader_info.c
@@ -0,0 +1,49 @@
+
+#include "brw_context.h"
+#include "brw_state.h"
+#include "pipe/p_util.h"
+#include "pipe/p_shader_tokens.h"
+#include "pipe/tgsi/util/tgsi_parse.h"
+
+
+
+
+void brw_shader_info(const struct tgsi_token *tokens,
+		     struct brw_shader_info *info )
+{
+   struct tgsi_parse_context parse;
+   int done = 0;
+
+   tgsi_parse_init( &parse, tokens );
+
+   while( !done &&
+	  !tgsi_parse_end_of_tokens( &parse ) ) 
+   {
+      tgsi_parse_token( &parse );
+
+      switch( parse.FullToken.Token.Type ) {
+      case TGSI_TOKEN_TYPE_DECLARATION:
+      {
+	 const struct tgsi_full_declaration *decl = &parse.FullToken.FullDeclaration;
+	 unsigned last = decl->u.DeclarationRange.Last;
+	 
+	 assert( decl->Declaration.Declare == TGSI_DECLARE_RANGE );
+      
+	 // Broken by crazy wpos init:
+	 //assert( info->nr_regs[decl->Declaration.File] <= last);
+
+	 info->nr_regs[decl->Declaration.File] = MAX2(info->nr_regs[decl->Declaration.File],
+						      last+1);
+	 break;
+      }
+      case TGSI_TOKEN_TYPE_IMMEDIATE:
+      case TGSI_TOKEN_TYPE_INSTRUCTION:
+      default:
+	 done = 1;
+	 break;
+      }
+   }
+
+   tgsi_parse_free (&parse);
+   
+}
diff --git a/src/gallium/drivers/i965simple/brw_state.c b/src/gallium/drivers/i965simple/brw_state.c
new file mode 100644
index 0000000000..95dfce88e4
--- /dev/null
+++ b/src/gallium/drivers/i965simple/brw_state.c
@@ -0,0 +1,424 @@
+/**************************************************************************
+ *
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ **************************************************************************/
+
+/* Authors:  Zack Rusin <zack@tungstengraphics.com>
+ *           Keith Whitwell <keith@tungstengraphics.com>
+ */
+
+
+#include "pipe/p_winsys.h"
+#include "pipe/p_util.h"
+#include "pipe/p_shader_tokens.h"
+#include "pipe/tgsi/util/tgsi_dump.h"
+
+#include "brw_context.h"
+#include "brw_defines.h"
+#include "brw_state.h"
+#include "brw_draw.h"
+
+
+#define DUP( TYPE, VAL )                        \
+do {                                            \
+   struct TYPE *x = malloc(sizeof(*x));         \
+   memcpy(x, VAL, sizeof(*x) );                 \
+   return x;                                    \
+} while (0)
+
+/************************************************************************
+ * Blend 
+ */
+static void *
+brw_create_blend_state(struct pipe_context *pipe,
+                        const struct pipe_blend_state *blend)
+{   
+   DUP( pipe_blend_state, blend );
+}
+
+static void brw_bind_blend_state(struct pipe_context *pipe,
+                                 void *blend)
+{
+   struct brw_context *brw = brw_context(pipe);
+
+   brw->attribs.Blend = (struct pipe_blend_state*)blend;
+   brw->state.dirty.brw |= BRW_NEW_BLEND;
+}
+
+
+static void brw_delete_blend_state(struct pipe_context *pipe, void *blend)
+{
+   free(blend);
+}
+
+static void brw_set_blend_color( struct pipe_context *pipe,
+			     const struct pipe_blend_color *blend_color )
+{
+   struct brw_context *brw = brw_context(pipe);
+
+   brw->attribs.BlendColor = *blend_color;
+
+   brw->state.dirty.brw |= BRW_NEW_BLEND;
+}
+
+/************************************************************************
+ * Sampler 
+ */
+
+static void *
+brw_create_sampler_state(struct pipe_context *pipe,
+                          const struct pipe_sampler_state *sampler)
+{
+   DUP( pipe_sampler_state, sampler );
+}
+
+static void brw_bind_sampler_state(struct pipe_context *pipe,
+                                    unsigned unit, void *sampler)
+{
+   struct brw_context *brw = brw_context(pipe);
+
+   brw->attribs.Samplers[unit] = sampler;
+   brw->state.dirty.brw |= BRW_NEW_SAMPLER;
+}
+
+static void brw_delete_sampler_state(struct pipe_context *pipe,
+                                      void *sampler)
+{
+   free(sampler);
+}
+
+
+/************************************************************************
+ * Depth stencil 
+ */
+
+static void *
+brw_create_depth_stencil_state(struct pipe_context *pipe,
+                           const struct pipe_depth_stencil_alpha_state *depth_stencil)
+{
+   DUP( pipe_depth_stencil_alpha_state, depth_stencil );
+}
+
+static void brw_bind_depth_stencil_state(struct pipe_context *pipe,
+                                         void *depth_stencil)
+{
+   struct brw_context *brw = brw_context(pipe);
+
+   brw->attribs.DepthStencil = (const struct pipe_depth_stencil_alpha_state *)depth_stencil;
+
+   brw->state.dirty.brw |= BRW_NEW_DEPTH_STENCIL;
+}
+
+static void brw_delete_depth_stencil_state(struct pipe_context *pipe,
+                                           void *depth_stencil)
+{
+   free(depth_stencil);
+}
+
+/************************************************************************
+ * Scissor
+ */
+static void brw_set_scissor_state( struct pipe_context *pipe,
+                                 const struct pipe_scissor_state *scissor )
+{
+   struct brw_context *brw = brw_context(pipe);
+
+   memcpy( &brw->attribs.Scissor, scissor, sizeof(*scissor) );
+   brw->state.dirty.brw |= BRW_NEW_SCISSOR;
+}
+
+
+/************************************************************************
+ * Stipple
+ */
+
+static void brw_set_polygon_stipple( struct pipe_context *pipe,
+                                   const struct pipe_poly_stipple *stipple )
+{
+}
+
+
+/************************************************************************
+ * Fragment shader
+ */
+
+static void * brw_create_fs_state(struct pipe_context *pipe,
+                                   const struct pipe_shader_state *shader)
+{
+   struct brw_fragment_program *brw_fp = CALLOC_STRUCT(brw_fragment_program);
+
+   /* XXX: Do I have to duplicate the tokens as well??
+    */
+   brw_fp->program = *shader;
+   brw_fp->id = brw_context(pipe)->program_id++;
+
+   brw_shader_info(shader->tokens,
+		   &brw_fp->info);
+
+   tgsi_dump(shader->tokens, 0);
+
+
+   return (void *)brw_fp;
+}
+
+static void brw_bind_fs_state(struct pipe_context *pipe, void *shader)
+{
+   struct brw_context *brw = brw_context(pipe);
+
+   brw->attribs.FragmentProgram = (struct brw_fragment_program *)shader;
+   brw->state.dirty.brw |= BRW_NEW_FS;
+}
+
+static void brw_delete_fs_state(struct pipe_context *pipe, void *shader)
+{
+   FREE(shader);
+}
+
+
+/************************************************************************
+ * Vertex shader and other TNL state 
+ */
+
+static void *brw_create_vs_state(struct pipe_context *pipe,
+                                 const struct pipe_shader_state *shader)
+{
+   struct brw_vertex_program *brw_vp = CALLOC_STRUCT(brw_vertex_program);
+
+   /* XXX: Do I have to duplicate the tokens as well??
+    */
+   brw_vp->program = *shader;
+   brw_vp->id = brw_context(pipe)->program_id++;
+   brw_shader_info(shader->tokens,
+		   &brw_vp->info);
+
+   tgsi_dump(shader->tokens, 0);
+
+   return (void *)brw_vp;
+}
+
+static void brw_bind_vs_state(struct pipe_context *pipe, void *vs)
+{
+   struct brw_context *brw = brw_context(pipe);
+
+   brw->attribs.VertexProgram = (struct brw_vertex_program *)vs;
+   brw->state.dirty.brw |= BRW_NEW_VS;
+
+   debug_printf("YYYYYYYYYYYYY BINDING VERTEX SHADER\n");
+}
+
+static void brw_delete_vs_state(struct pipe_context *pipe, void *shader)
+{
+   FREE(shader);
+}
+
+
+static void brw_set_clip_state( struct pipe_context *pipe,
+                                const struct pipe_clip_state *clip )
+{
+   struct brw_context *brw = brw_context(pipe);
+
+   brw->attribs.Clip = *clip;
+}
+
+
+static void brw_set_viewport_state( struct pipe_context *pipe,
+				     const struct pipe_viewport_state *viewport )
+{
+   struct brw_context *brw = brw_context(pipe);
+
+   brw->attribs.Viewport = *viewport; /* struct copy */
+   brw->state.dirty.brw |= BRW_NEW_VIEWPORT;
+
+   /* pass the viewport info to the draw module */
+   //draw_set_viewport_state(brw->draw, viewport);
+}
+
+
+static void brw_set_vertex_buffer( struct pipe_context *pipe,
+				   unsigned index,
+				   const struct pipe_vertex_buffer *buffer )
+{
+   struct brw_context *brw = brw_context(pipe);
+   brw->vb.vbo_array[index] = buffer;
+}
+
+static void brw_set_vertex_element(struct pipe_context *pipe,
+                                   unsigned index,
+                                   const struct pipe_vertex_element *element)
+{
+   /* flush ? */
+   struct brw_context *brw = brw_context(pipe);
+
+   assert(index < PIPE_ATTRIB_MAX);
+   struct brw_vertex_element_state el;
+   memset(&el, 0, sizeof(el));
+
+   el.ve0.src_offset = element->src_offset;
+   el.ve0.src_format = brw_translate_surface_format(element->src_format);
+   el.ve0.valid = 1;
+   el.ve0.vertex_buffer_index = element->vertex_buffer_index;
+
+   el.ve1.dst_offset   = index * 4;
+
+   el.ve1.vfcomponent3 = BRW_VFCOMPONENT_STORE_SRC;
+   el.ve1.vfcomponent2 = BRW_VFCOMPONENT_STORE_SRC;
+   el.ve1.vfcomponent1 = BRW_VFCOMPONENT_STORE_SRC;
+   el.ve1.vfcomponent0 = BRW_VFCOMPONENT_STORE_SRC;
+
+   switch (element->nr_components) {
+   case 1: el.ve1.vfcomponent1 = BRW_VFCOMPONENT_STORE_0;
+   case 2: el.ve1.vfcomponent2 = BRW_VFCOMPONENT_STORE_0;
+   case 3: el.ve1.vfcomponent3 = BRW_VFCOMPONENT_STORE_1_FLT;
+      break;
+   }
+
+   brw->vb.inputs[index] = el;
+}
+
+
+
+/************************************************************************
+ * Constant buffers
+ */
+
+static void brw_set_constant_buffer(struct pipe_context *pipe,
+                                     uint shader, uint index,
+                                     const struct pipe_constant_buffer *buf)
+{
+   struct brw_context *brw = brw_context(pipe);
+
+   assert(buf == 0 || index == 0);
+
+   brw->attribs.Constants[shader] = buf;
+   brw->state.dirty.brw |= BRW_NEW_CONSTANTS;
+}
+
+
+/************************************************************************
+ * Texture surfaces
+ */
+
+
+static void brw_set_sampler_texture(struct pipe_context *pipe,
+                                  unsigned unit,
+                                  struct pipe_texture *texture)
+{
+   struct brw_context *brw = brw_context(pipe);
+
+   brw->attribs.Texture[unit] = (struct brw_texture*)texture;  /* ptr, not struct */
+
+   brw->state.dirty.brw |= BRW_NEW_TEXTURE;
+}
+
+
+/************************************************************************
+ * Render targets, etc
+ */
+
+static void brw_set_framebuffer_state(struct pipe_context *pipe,
+				       const struct pipe_framebuffer_state *fb)
+{
+   struct brw_context *brw = brw_context(pipe);
+
+   brw->attribs.FrameBuffer = *fb; /* struct copy */
+
+   brw->state.dirty.brw |= BRW_NEW_FRAMEBUFFER;
+}
+
+
+
+/************************************************************************
+ * Rasterizer state
+ */
+
+static void *
+brw_create_rasterizer_state(struct pipe_context *pipe,
+                             const struct pipe_rasterizer_state *rasterizer)
+{
+   DUP(pipe_rasterizer_state, rasterizer);
+}
+
+static void brw_bind_rasterizer_state( struct pipe_context *pipe,
+                                        void *setup )
+{
+   struct brw_context *brw = brw_context(pipe);
+
+   brw->attribs.Raster = (struct pipe_rasterizer_state *)setup;
+
+   /* Also pass-through to draw module:
+    */
+   //draw_set_rasterizer_state(brw->draw, setup);
+
+   brw->state.dirty.brw |= BRW_NEW_RASTERIZER;
+}
+
+static void brw_delete_rasterizer_state(struct pipe_context *pipe,
+                                         void *setup)
+{
+   free(setup);
+}
+
+
+
+void
+brw_init_state_functions( struct brw_context *brw )
+{
+   brw->pipe.create_blend_state = brw_create_blend_state;
+   brw->pipe.bind_blend_state = brw_bind_blend_state;
+   brw->pipe.delete_blend_state = brw_delete_blend_state;
+
+   brw->pipe.create_sampler_state = brw_create_sampler_state;
+   brw->pipe.bind_sampler_state = brw_bind_sampler_state;
+   brw->pipe.delete_sampler_state = brw_delete_sampler_state;
+
+   brw->pipe.create_depth_stencil_alpha_state = brw_create_depth_stencil_state;
+   brw->pipe.bind_depth_stencil_alpha_state = brw_bind_depth_stencil_state;
+   brw->pipe.delete_depth_stencil_alpha_state = brw_delete_depth_stencil_state;
+
+   brw->pipe.create_rasterizer_state = brw_create_rasterizer_state;
+   brw->pipe.bind_rasterizer_state = brw_bind_rasterizer_state;
+   brw->pipe.delete_rasterizer_state = brw_delete_rasterizer_state;
+   brw->pipe.create_fs_state = brw_create_fs_state;
+   brw->pipe.bind_fs_state = brw_bind_fs_state;
+   brw->pipe.delete_fs_state = brw_delete_fs_state;
+   brw->pipe.create_vs_state = brw_create_vs_state;
+   brw->pipe.bind_vs_state = brw_bind_vs_state;
+   brw->pipe.delete_vs_state = brw_delete_vs_state;
+
+   brw->pipe.set_blend_color = brw_set_blend_color;
+   brw->pipe.set_clip_state = brw_set_clip_state;
+   brw->pipe.set_constant_buffer = brw_set_constant_buffer;
+   brw->pipe.set_framebuffer_state = brw_set_framebuffer_state;
+
+//   brw->pipe.set_feedback_state = brw_set_feedback_state;
+//   brw->pipe.set_feedback_buffer = brw_set_feedback_buffer;
+
+   brw->pipe.set_polygon_stipple = brw_set_polygon_stipple;
+   brw->pipe.set_scissor_state = brw_set_scissor_state;
+   brw->pipe.set_sampler_texture = brw_set_sampler_texture;
+   brw->pipe.set_viewport_state = brw_set_viewport_state;
+   brw->pipe.set_vertex_buffer = brw_set_vertex_buffer;
+   brw->pipe.set_vertex_element = brw_set_vertex_element;
+}
diff --git a/src/gallium/drivers/i965simple/brw_state.h b/src/gallium/drivers/i965simple/brw_state.h
new file mode 100644
index 0000000000..258e9a556e
--- /dev/null
+++ b/src/gallium/drivers/i965simple/brw_state.h
@@ -0,0 +1,158 @@
+/*
+ Copyright (C) Intel Corp.  2006.  All Rights Reserved.
+ Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to
+ develop this 3D driver.
+ 
+ Permission is hereby granted, free of charge, to any person obtaining
+ a copy of this software and associated documentation files (the
+ "Software"), to deal in the Software without restriction, including
+ without limitation the rights to use, copy, modify, merge, publish,
+ distribute, sublicense, and/or sell copies of the Software, and to
+ permit persons to whom the Software is furnished to do so, subject to
+ the following conditions:
+ 
+ The above copyright notice and this permission notice (including the
+ next paragraph) shall be included in all copies or substantial
+ portions of the Software.
+ 
+ THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
+ IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
+ LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
+ OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ 
+ **********************************************************************/
+ /*
+  * Authors:
+  *   Keith Whitwell <keith@tungstengraphics.com>
+  */
+    
+
+#ifndef BRW_STATE_H
+#define BRW_STATE_H
+
+#include "brw_context.h"
+#include "brw_winsys.h"
+
+
+const struct brw_tracked_state brw_blend_constant_color;
+const struct brw_tracked_state brw_cc_unit;
+const struct brw_tracked_state brw_cc_vp;
+const struct brw_tracked_state brw_clip_prog;
+const struct brw_tracked_state brw_clip_unit;
+const struct brw_tracked_state brw_constant_buffer_state;
+const struct brw_tracked_state brw_constant_buffer;
+const struct brw_tracked_state brw_curbe_offsets;
+const struct brw_tracked_state brw_invarient_state;
+const struct brw_tracked_state brw_gs_prog;
+const struct brw_tracked_state brw_gs_unit;
+const struct brw_tracked_state brw_drawing_rect;
+const struct brw_tracked_state brw_line_stipple;
+const struct brw_tracked_state brw_pipelined_state_pointers;
+const struct brw_tracked_state brw_binding_table_pointers;
+const struct brw_tracked_state brw_depthbuffer;
+const struct brw_tracked_state brw_polygon_stipple_offset;
+const struct brw_tracked_state brw_polygon_stipple;
+const struct brw_tracked_state brw_program_parameters;
+const struct brw_tracked_state brw_recalculate_urb_fence;
+const struct brw_tracked_state brw_sf_prog;
+const struct brw_tracked_state brw_sf_unit;
+const struct brw_tracked_state brw_sf_vp;
+const struct brw_tracked_state brw_state_base_address;
+const struct brw_tracked_state brw_urb_fence;
+const struct brw_tracked_state brw_vertex_state;
+const struct brw_tracked_state brw_vs_prog;
+const struct brw_tracked_state brw_vs_unit;
+const struct brw_tracked_state brw_wm_prog;
+const struct brw_tracked_state brw_wm_samplers;
+const struct brw_tracked_state brw_wm_surfaces;
+const struct brw_tracked_state brw_wm_unit;
+
+const struct brw_tracked_state brw_psp_urb_cbs;
+
+const struct brw_tracked_state brw_active_vertprog;
+const struct brw_tracked_state brw_tnl_vertprog;
+const struct brw_tracked_state brw_pipe_control;
+
+const struct brw_tracked_state brw_clear_surface_cache;
+const struct brw_tracked_state brw_clear_batch_cache;
+
+/***********************************************************************
+ * brw_state_cache.c
+ */
+unsigned brw_cache_data(struct brw_cache *cache,
+		      const void *data );
+
+unsigned brw_cache_data_sz(struct brw_cache *cache,
+			 const void *data,
+			 unsigned data_sz);
+
+unsigned brw_upload_cache( struct brw_cache *cache,
+			 const void *key,
+			 unsigned key_sz,
+			 const void *data,
+			 unsigned data_sz,
+			 const void *aux,
+			 void *aux_return );
+
+boolean brw_search_cache( struct brw_cache *cache,
+			    const void *key,
+			    unsigned key_size,
+			    void *aux_return,
+			    unsigned *offset_return);
+
+void brw_init_caches( struct brw_context *brw );
+void brw_destroy_caches( struct brw_context *brw );
+
+static inline struct pipe_buffer *brw_cache_buffer(struct brw_context *brw,
+                                                          enum brw_cache_id id)
+{
+   return brw->cache[id].pool->buffer;
+}
+
+/***********************************************************************
+ * brw_state_batch.c
+ */
+#define BRW_CACHED_BATCH_STRUCT(brw, s) brw_cached_batch_struct( brw, (s), sizeof(*(s)) )
+
+boolean brw_cached_batch_struct( struct brw_context *brw,
+				   const void *data,
+				   unsigned sz );
+
+void brw_destroy_batch_cache( struct brw_context *brw );
+
+
+/***********************************************************************
+ * brw_state_pool.c
+ */
+void brw_init_pools( struct brw_context *brw );
+void brw_destroy_pools( struct brw_context *brw );
+
+boolean brw_pool_alloc( struct brw_mem_pool *pool,
+			  unsigned size,
+			  unsigned alignment,
+			  unsigned *offset_return);
+
+void brw_pool_fence( struct brw_context *brw,
+		     struct brw_mem_pool *pool,
+		     unsigned fence );
+
+
+void brw_pool_check_wrap( struct brw_context *brw,
+			  struct brw_mem_pool *pool );
+
+void brw_clear_all_caches( struct brw_context *brw );
+void brw_invalidate_pools( struct brw_context *brw );
+void brw_clear_batch_cache_flush( struct brw_context *brw );
+
+
+/* brw_shader_info.c
+ */
+
+void brw_shader_info(const struct tgsi_token *tokens,
+		     struct brw_shader_info *info );
+
+
+#endif
diff --git a/src/gallium/drivers/i965simple/brw_state_batch.c b/src/gallium/drivers/i965simple/brw_state_batch.c
new file mode 100644
index 0000000000..35db76b594
--- /dev/null
+++ b/src/gallium/drivers/i965simple/brw_state_batch.c
@@ -0,0 +1,113 @@
+/*
+ Copyright (C) Intel Corp.  2006.  All Rights Reserved.
+ Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to
+ develop this 3D driver.
+
+ Permission is hereby granted, free of charge, to any person obtaining
+ a copy of this software and associated documentation files (the
+ "Software"), to deal in the Software without restriction, including
+ without limitation the rights to use, copy, modify, merge, publish,
+ distribute, sublicense, and/or sell copies of the Software, and to
+ permit persons to whom the Software is furnished to do so, subject to
+ the following conditions:
+
+ The above copyright notice and this permission notice (including the
+ next paragraph) shall be included in all copies or substantial
+ portions of the Software.
+
+ THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
+ IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
+ LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
+ OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+
+ **********************************************************************/
+ /*
+  * Authors:
+  *   Keith Whitwell <keith@tungstengraphics.com>
+  */
+
+#include "brw_state.h"
+#include "brw_winsys.h"
+
+#include "pipe/p_util.h"
+
+/* A facility similar to the data caching code above, which aims to
+ * prevent identical commands being issued repeatedly.
+ */
+boolean brw_cached_batch_struct( struct brw_context *brw,
+                                 const void *data,
+                                 unsigned sz )
+{
+   struct brw_cached_batch_item *item = brw->cached_batch_items;
+   struct header *newheader = (struct header *)data;
+
+   if (brw->emit_state_always) {
+      brw_batchbuffer_data(brw->winsys, data, sz);
+      return TRUE;
+   }
+
+   while (item) {
+      if (item->header->opcode == newheader->opcode) {
+	 if (item->sz == sz && memcmp(item->header, newheader, sz) == 0)
+	    return FALSE;
+	 if (item->sz != sz) {
+	    FREE(item->header);
+	    item->header = MALLOC(sz);
+	    item->sz = sz;
+	 }
+	 goto emit;
+      }
+      item = item->next;
+   }
+
+   assert(!item);
+   item = CALLOC_STRUCT(brw_cached_batch_item);
+   item->header = MALLOC(sz);
+   item->sz = sz;
+   item->next = brw->cached_batch_items;
+   brw->cached_batch_items = item;
+
+emit:
+   memcpy(item->header, newheader, sz);
+   brw_batchbuffer_data(brw->winsys, data, sz);
+   return TRUE;
+}
+
+static void clear_batch_cache( struct brw_context *brw )
+{
+   struct brw_cached_batch_item *item = brw->cached_batch_items;
+
+   while (item) {
+      struct brw_cached_batch_item *next = item->next;
+      free((void *)item->header);
+      free(item);
+      item = next;
+   }
+
+   brw->cached_batch_items = NULL;
+
+
+   brw_clear_all_caches(brw);
+
+   brw_invalidate_pools(brw);
+}
+
+void brw_clear_batch_cache_flush( struct brw_context *brw )
+{
+   clear_batch_cache(brw);
+
+/*    brw_do_flush(brw, BRW_FLUSH_STATE_CACHE|BRW_FLUSH_READ_CACHE); */
+
+   brw->state.dirty.brw |= ~0;
+   brw->state.dirty.cache |= ~0;
+}
+
+
+
+void brw_destroy_batch_cache( struct brw_context *brw )
+{
+   clear_batch_cache(brw);
+}
diff --git a/src/gallium/drivers/i965simple/brw_state_cache.c b/src/gallium/drivers/i965simple/brw_state_cache.c
new file mode 100644
index 0000000000..b3a5124461
--- /dev/null
+++ b/src/gallium/drivers/i965simple/brw_state_cache.c
@@ -0,0 +1,443 @@
+/*
+ Copyright (C) Intel Corp.  2006.  All Rights Reserved.
+ Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to
+ develop this 3D driver.
+
+ Permission is hereby granted, free of charge, to any person obtaining
+ a copy of this software and associated documentation files (the
+ "Software"), to deal in the Software without restriction, including
+ without limitation the rights to use, copy, modify, merge, publish,
+ distribute, sublicense, and/or sell copies of the Software, and to
+ permit persons to whom the Software is furnished to do so, subject to
+ the following conditions:
+
+ The above copyright notice and this permission notice (including the
+ next paragraph) shall be included in all copies or substantial
+ portions of the Software.
+
+ THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
+ IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
+ LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
+ OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+
+ **********************************************************************/
+ /*
+  * Authors:
+  *   Keith Whitwell <keith@tungstengraphics.com>
+  */
+
+
+#include "brw_state.h"
+
+#include "brw_wm.h"
+#include "brw_vs.h"
+#include "brw_clip.h"
+#include "brw_sf.h"
+#include "brw_gs.h"
+
+#include "pipe/p_util.h"
+
+
+
+/***********************************************************************
+ * Check cache for uploaded version of struct, else upload new one.
+ * Fail when memory is exhausted.
+ *
+ * XXX: FIXME: Currently search is so slow it would be quicker to
+ * regenerate the data every time...
+ */
+
+static unsigned hash_key( const void *key, unsigned key_size )
+{
+   unsigned *ikey = (unsigned *)key;
+   unsigned hash = 0, i;
+
+   assert(key_size % 4 == 0);
+
+   /* I'm sure this can be improved on:
+    */
+   for (i = 0; i < key_size/4; i++)
+      hash ^= ikey[i];
+
+   return hash;
+}
+
+static struct brw_cache_item *search_cache( struct brw_cache *cache,
+					     unsigned hash,
+					     const void *key,
+					     unsigned key_size)
+{
+   struct brw_cache_item *c;
+
+   for (c = cache->items[hash % cache->size]; c; c = c->next) {
+      if (c->hash == hash &&
+	  c->key_size == key_size &&
+	  memcmp(c->key, key, key_size) == 0)
+	 return c;
+   }
+
+   return NULL;
+}
+
+
+static void rehash( struct brw_cache *cache )
+{
+   struct brw_cache_item **items;
+   struct brw_cache_item *c, *next;
+   unsigned size, i;
+
+   size = cache->size * 3;
+   items = (struct brw_cache_item**) MALLOC(size * sizeof(*items));
+   memset(items, 0, size * sizeof(*items));
+
+   for (i = 0; i < cache->size; i++)
+      for (c = cache->items[i]; c; c = next) {
+	 next = c->next;
+	 c->next = items[c->hash % size];
+	 items[c->hash % size] = c;
+      }
+
+   FREE(cache->items);
+   cache->items = items;
+   cache->size = size;
+}
+
+
+boolean brw_search_cache( struct brw_cache *cache,
+			    const void *key,
+			    unsigned key_size,
+			    void *aux_return,
+			    unsigned *offset_return)
+{
+   struct brw_cache_item *item;
+   unsigned addr = 0;
+   unsigned hash = hash_key(key, key_size);
+
+   item = search_cache(cache, hash, key, key_size);
+
+   if (item) {
+      if (aux_return)
+	 *(void **)aux_return = (void *)((char *)item->key + item->key_size);
+
+      *offset_return = addr = item->offset;
+   }
+
+   if (item == NULL || addr != cache->last_addr) {
+      cache->brw->state.dirty.cache |= 1<<cache->id;
+      cache->last_addr = addr;
+   }
+
+   return item != NULL;
+}
+
+unsigned brw_upload_cache( struct brw_cache *cache,
+			 const void *key,
+			 unsigned key_size,
+			 const void *data,
+			 unsigned data_size,
+			 const void *aux,
+			 void *aux_return )
+{
+   unsigned offset;
+   struct brw_cache_item *item = CALLOC_STRUCT(brw_cache_item);
+   unsigned hash = hash_key(key, key_size);
+   void *tmp = MALLOC(key_size + cache->aux_size);
+
+   if (!brw_pool_alloc(cache->pool, data_size, 1 << 6, &offset)) {
+      /* Should not be possible:
+       */
+      debug_printf("brw_pool_alloc failed\n");
+      exit(1);
+   }
+
+   memcpy(tmp, key, key_size);
+
+   if (cache->aux_size)
+      memcpy(tmp+key_size, aux, cache->aux_size);
+
+   item->key = tmp;
+   item->hash = hash;
+   item->key_size = key_size;
+   item->offset = offset;
+   item->data_size = data_size;
+
+   if (++cache->n_items > cache->size * 1.5)
+      rehash(cache);
+
+   hash %= cache->size;
+   item->next = cache->items[hash];
+   cache->items[hash] = item;
+
+   if (aux_return) {
+      assert(cache->aux_size);
+      *(void **)aux_return = (void *)((char *)item->key + item->key_size);
+   }
+
+   if (BRW_DEBUG & DEBUG_STATE)
+      debug_printf("upload %s: %d bytes to pool buffer %p offset %x\n",
+             cache->name, 
+	     data_size,
+             (void*)cache->pool->buffer,
+             offset);
+
+   /* Copy data to the buffer:
+    */
+   cache->brw->winsys->buffer_subdata_typed(cache->brw->winsys,
+					    cache->pool->buffer, 
+					    offset, 
+					    data_size, 
+					    data,
+					    cache->id);
+
+   cache->brw->state.dirty.cache |= 1<<cache->id;
+   cache->last_addr = offset;
+
+   return offset;
+}
+
+/* This doesn't really work with aux data.  Use search/upload instead
+ */
+unsigned brw_cache_data_sz(struct brw_cache *cache,
+			 const void *data,
+			 unsigned data_size)
+{
+   unsigned addr;
+
+   if (!brw_search_cache(cache, data, data_size, NULL, &addr)) {
+      addr = brw_upload_cache(cache,
+			      data, data_size,
+			      data, data_size,
+			      NULL, NULL);
+   }
+
+   return addr;
+}
+
+unsigned brw_cache_data(struct brw_cache *cache,
+		      const void *data)
+{
+   return brw_cache_data_sz(cache, data, cache->key_size);
+}
+
+enum pool_type {
+   DW_SURFACE_STATE,
+   DW_GENERAL_STATE
+};
+
+static void brw_init_cache( struct brw_context *brw,
+			    const char *name,
+			    unsigned id,
+			    unsigned key_size,
+			    unsigned aux_size,
+			    enum pool_type pool_type)
+{
+   struct brw_cache *cache = &brw->cache[id];
+   cache->brw = brw;
+   cache->id = id;
+   cache->name = name;
+   cache->items = NULL;
+
+   cache->size = 7;
+   cache->n_items = 0;
+   cache->items = (struct brw_cache_item **)
+      CALLOC(cache->size, sizeof(struct brw_cache_item));
+
+
+   cache->key_size = key_size;
+   cache->aux_size = aux_size;
+   switch (pool_type) {
+   case DW_GENERAL_STATE: cache->pool = &brw->pool[BRW_GS_POOL]; break;
+   case DW_SURFACE_STATE: cache->pool = &brw->pool[BRW_SS_POOL]; break;
+   default: assert(0); break;
+   }
+}
+
+void brw_init_caches( struct brw_context *brw )
+{
+
+   brw_init_cache(brw,
+		  "CC_VP",
+		  BRW_CC_VP,
+		  sizeof(struct brw_cc_viewport),
+		  0,
+		  DW_GENERAL_STATE);
+
+   brw_init_cache(brw,
+		  "CC_UNIT",
+		  BRW_CC_UNIT,
+		  sizeof(struct brw_cc_unit_state),
+		  0,
+		  DW_GENERAL_STATE);
+
+   brw_init_cache(brw,
+		  "WM_PROG",
+		  BRW_WM_PROG,
+		  sizeof(struct brw_wm_prog_key),
+		  sizeof(struct brw_wm_prog_data),
+		  DW_GENERAL_STATE);
+
+   brw_init_cache(brw,
+		  "SAMPLER_DEFAULT_COLOR",
+		  BRW_SAMPLER_DEFAULT_COLOR,
+		  sizeof(struct brw_sampler_default_color),
+		  0,
+		  DW_GENERAL_STATE);
+
+   brw_init_cache(brw,
+		  "SAMPLER",
+		  BRW_SAMPLER,
+		  0,		/* variable key/data size */
+		  0,
+		  DW_GENERAL_STATE);
+
+   brw_init_cache(brw,
+		  "WM_UNIT",
+		  BRW_WM_UNIT,
+		  sizeof(struct brw_wm_unit_state),
+		  0,
+		  DW_GENERAL_STATE);
+
+   brw_init_cache(brw,
+		  "SF_PROG",
+		  BRW_SF_PROG,
+		  sizeof(struct brw_sf_prog_key),
+		  sizeof(struct brw_sf_prog_data),
+		  DW_GENERAL_STATE);
+
+   brw_init_cache(brw,
+		  "SF_VP",
+		  BRW_SF_VP,
+		  sizeof(struct brw_sf_viewport),
+		  0,
+		  DW_GENERAL_STATE);
+
+   brw_init_cache(brw,
+		  "SF_UNIT",
+		  BRW_SF_UNIT,
+		  sizeof(struct brw_sf_unit_state),
+		  0,
+		  DW_GENERAL_STATE);
+
+   brw_init_cache(brw,
+		  "VS_UNIT",
+		  BRW_VS_UNIT,
+		  sizeof(struct brw_vs_unit_state),
+		  0,
+		  DW_GENERAL_STATE);
+
+   brw_init_cache(brw,
+		  "VS_PROG",
+		  BRW_VS_PROG,
+		  sizeof(struct brw_vs_prog_key),
+		  sizeof(struct brw_vs_prog_data),
+		  DW_GENERAL_STATE);
+
+   brw_init_cache(brw,
+		  "CLIP_UNIT",
+		  BRW_CLIP_UNIT,
+		  sizeof(struct brw_clip_unit_state),
+		  0,
+		  DW_GENERAL_STATE);
+
+   brw_init_cache(brw,
+		  "CLIP_PROG",
+		  BRW_CLIP_PROG,
+		  sizeof(struct brw_clip_prog_key),
+		  sizeof(struct brw_clip_prog_data),
+		  DW_GENERAL_STATE);
+
+   brw_init_cache(brw,
+		  "GS_UNIT",
+		  BRW_GS_UNIT,
+		  sizeof(struct brw_gs_unit_state),
+		  0,
+		  DW_GENERAL_STATE);
+
+   brw_init_cache(brw,
+		  "GS_PROG",
+		  BRW_GS_PROG,
+		  sizeof(struct brw_gs_prog_key),
+		  sizeof(struct brw_gs_prog_data),
+		  DW_GENERAL_STATE);
+
+   brw_init_cache(brw,
+		  "SS_SURFACE",
+		  BRW_SS_SURFACE,
+		  sizeof(struct brw_surface_state),
+		  0,
+		  DW_SURFACE_STATE);
+
+   brw_init_cache(brw,
+		  "SS_SURF_BIND",
+		  BRW_SS_SURF_BIND,
+		  sizeof(struct brw_surface_binding_table),
+		  0,
+		  DW_SURFACE_STATE);
+}
+
+
+/* When we lose hardware context, need to invalidate the surface cache
+ * as these structs must be explicitly re-uploaded.  They are subject
+ * to fixup by the memory manager as they contain absolute agp
+ * offsets, so we need to ensure there is a fresh version of the
+ * struct available to receive the fixup.
+ *
+ * XXX: Need to ensure that there aren't two versions of a surface or
+ * bufferobj with different backing data active in the same buffer at
+ * once?  Otherwise the cache could confuse them.  Maybe better not to
+ * cache at all?
+ *
+ * --> Isn't this the same as saying need to ensure batch is flushed
+ *         before new data is uploaded to an existing buffer?  We
+ *         already try to make sure of that.
+ */
+static void clear_cache( struct brw_cache *cache )
+{
+   struct brw_cache_item *c, *next;
+   unsigned i;
+
+   for (i = 0; i < cache->size; i++) {
+      for (c = cache->items[i]; c; c = next) {
+	 next = c->next;
+	 free((void *)c->key);
+	 free(c);
+      }
+      cache->items[i] = NULL;
+   }
+
+   cache->n_items = 0;
+}
+
+void brw_clear_all_caches( struct brw_context *brw )
+{
+   int i;
+
+   if (BRW_DEBUG & DEBUG_STATE)
+      debug_printf("%s\n", __FUNCTION__);
+
+   for (i = 0; i < BRW_MAX_CACHE; i++)
+      clear_cache(&brw->cache[i]);
+
+   if (brw->curbe.last_buf) {
+      FREE(brw->curbe.last_buf);
+      brw->curbe.last_buf = NULL;
+   }
+
+   brw->state.dirty.brw |= ~0;
+   brw->state.dirty.cache |= ~0;
+}
+
+
+
+
+
+void brw_destroy_caches( struct brw_context *brw )
+{
+   unsigned i;
+
+   for (i = 0; i < BRW_MAX_CACHE; i++)
+      clear_cache(&brw->cache[i]);
+}
diff --git a/src/gallium/drivers/i965simple/brw_state_pool.c b/src/gallium/drivers/i965simple/brw_state_pool.c
new file mode 100644
index 0000000000..f3174bfe0a
--- /dev/null
+++ b/src/gallium/drivers/i965simple/brw_state_pool.c
@@ -0,0 +1,137 @@
+/*
+ Copyright (C) Intel Corp.  2006.  All Rights Reserved.
+ Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to
+ develop this 3D driver.
+
+ Permission is hereby granted, free of charge, to any person obtaining
+ a copy of this software and associated documentation files (the
+ "Software"), to deal in the Software without restriction, including
+ without limitation the rights to use, copy, modify, merge, publish,
+ distribute, sublicense, and/or sell copies of the Software, and to
+ permit persons to whom the Software is furnished to do so, subject to
+ the following conditions:
+
+ The above copyright notice and this permission notice (including the
+ next paragraph) shall be included in all copies or substantial
+ portions of the Software.
+
+ THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
+ IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
+ LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
+ OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+
+ **********************************************************************/
+ /*
+  * Authors:
+  *   Keith Whitwell <keith@tungstengraphics.com>
+  */
+
+/** @file brw_state_pool.c
+ * Implements the state pool allocator.
+ *
+ * For the 965, we create two state pools for state cache entries.  Objects
+ * will be allocated into the pools depending on which state base address
+ * their pointer is relative to in other 965 state.
+ *
+ * The state pools are relatively simple: As objects are allocated, increment
+ * the offset to allocate space.  When the pool is "full" (rather, close to
+ * full), we reset the pool and reset the state cache entries that point into
+ * the pool.
+ */
+
+#include "pipe/p_winsys.h"
+#include "pipe/p_util.h"
+#include "pipe/p_inlines.h"
+#include "brw_context.h"
+#include "brw_state.h"
+
+boolean brw_pool_alloc( struct brw_mem_pool *pool,
+			  unsigned size,
+			  unsigned alignment,
+			  unsigned *offset_return)
+{
+   unsigned fixup = align(pool->offset, alignment) - pool->offset;
+
+   size = align(size, 4);
+
+   if (pool->offset + fixup + size >= pool->size) {
+      debug_printf("%s failed\n", __FUNCTION__);
+      assert(0);
+      exit(0);
+   }
+
+   pool->offset += fixup;
+   *offset_return = pool->offset;
+   pool->offset += size;
+
+   return TRUE;
+}
+
+static
+void brw_invalidate_pool( struct brw_mem_pool *pool )
+{
+   if (BRW_DEBUG & DEBUG_STATE)
+      debug_printf("\n\n\n %s \n\n\n", __FUNCTION__);
+
+   pool->offset = 0;
+
+   brw_clear_all_caches(pool->brw);
+}
+
+
+static void brw_init_pool( struct brw_context *brw,
+			   unsigned pool_id,
+			   unsigned size )
+{
+   struct brw_mem_pool *pool = &brw->pool[pool_id];
+
+   pool->size = size;
+   pool->brw = brw;
+
+   pool->buffer = brw->pipe.winsys->buffer_create(brw->pipe.winsys,
+						  4096,
+                                                  0 /*  DRM_BO_FLAG_MEM_TT */,
+                                                  size);
+}
+
+static void brw_destroy_pool( struct brw_context *brw,
+			      unsigned pool_id )
+{
+   struct brw_mem_pool *pool = &brw->pool[pool_id];
+
+   pipe_buffer_reference( pool->brw->pipe.winsys,
+			  &pool->buffer,
+			  NULL );
+}
+
+
+void brw_pool_check_wrap( struct brw_context *brw,
+			  struct brw_mem_pool *pool )
+{
+   if (pool->offset > (pool->size * 3) / 4) {
+      brw->state.dirty.brw |= BRW_NEW_SCENE;
+   }
+
+}
+
+void brw_init_pools( struct brw_context *brw )
+{
+   brw_init_pool(brw, BRW_GS_POOL, 0x80000);
+   brw_init_pool(brw, BRW_SS_POOL, 0x80000);
+}
+
+void brw_destroy_pools( struct brw_context *brw )
+{
+   brw_destroy_pool(brw, BRW_GS_POOL);
+   brw_destroy_pool(brw, BRW_SS_POOL);
+}
+
+
+void brw_invalidate_pools( struct brw_context *brw )
+{
+   brw_invalidate_pool(&brw->pool[BRW_GS_POOL]);
+   brw_invalidate_pool(&brw->pool[BRW_SS_POOL]);
+}
diff --git a/src/gallium/drivers/i965simple/brw_state_upload.c b/src/gallium/drivers/i965simple/brw_state_upload.c
new file mode 100644
index 0000000000..e727601e1e
--- /dev/null
+++ b/src/gallium/drivers/i965simple/brw_state_upload.c
@@ -0,0 +1,202 @@
+/*
+ Copyright (C) Intel Corp.  2006.  All Rights Reserved.
+ Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to
+ develop this 3D driver.
+
+ Permission is hereby granted, free of charge, to any person obtaining
+ a copy of this software and associated documentation files (the
+ "Software"), to deal in the Software without restriction, including
+ without limitation the rights to use, copy, modify, merge, publish,
+ distribute, sublicense, and/or sell copies of the Software, and to
+ permit persons to whom the Software is furnished to do so, subject to
+ the following conditions:
+
+ The above copyright notice and this permission notice (including the
+ next paragraph) shall be included in all copies or substantial
+ portions of the Software.
+
+ THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
+ IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
+ LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
+ OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+
+ **********************************************************************/
+ /*
+  * Authors:
+  *   Keith Whitwell <keith@tungstengraphics.com>
+  */
+
+
+#include "brw_context.h"
+#include "brw_state.h"
+
+#include "pipe/p_util.h"
+
+/* This is used to initialize brw->state.atoms[].  We could use this
+ * list directly except for a single atom, brw_constant_buffer, which
+ * has a .dirty value which changes according to the parameters of the
+ * current fragment and vertex programs, and so cannot be a static
+ * value.
+ */
+const struct brw_tracked_state *atoms[] =
+{
+   &brw_vs_prog,
+   &brw_gs_prog,
+   &brw_clip_prog,
+   &brw_sf_prog,
+   &brw_wm_prog,
+
+   /* Once all the programs are done, we know how large urb entry
+    * sizes need to be and can decide if we need to change the urb
+    * layout.
+    */
+   &brw_curbe_offsets,
+   &brw_recalculate_urb_fence,
+
+
+   &brw_cc_vp,
+   &brw_cc_unit,
+
+   &brw_wm_surfaces,		/* must do before samplers */
+   &brw_wm_samplers,
+
+   &brw_wm_unit,
+   &brw_sf_vp,
+   &brw_sf_unit,
+   &brw_vs_unit,		/* always required, enabled or not */
+   &brw_clip_unit,
+   &brw_gs_unit,
+
+   /* Command packets:
+    */
+   &brw_invarient_state,
+   &brw_state_base_address,
+   &brw_pipe_control,
+
+   &brw_binding_table_pointers,
+   &brw_blend_constant_color,
+
+   &brw_drawing_rect,
+   &brw_depthbuffer,
+
+   &brw_polygon_stipple,
+   &brw_line_stipple,
+
+   &brw_psp_urb_cbs,
+
+   &brw_constant_buffer
+};
+
+
+void brw_init_state( struct brw_context *brw )
+{
+   brw_init_pools(brw);
+   brw_init_caches(brw);
+
+   brw->state.dirty.brw = ~0;
+   brw->emit_state_always = 0;
+}
+
+
+void brw_destroy_state( struct brw_context *brw )
+{
+   brw_destroy_caches(brw);
+   brw_destroy_batch_cache(brw);
+   brw_destroy_pools(brw);
+}
+
+/***********************************************************************
+ */
+
+static boolean check_state( const struct brw_state_flags *a,
+			      const struct brw_state_flags *b )
+{
+   return ((a->brw & b->brw) ||
+	   (a->cache & b->cache));
+}
+
+static void accumulate_state( struct brw_state_flags *a,
+			      const struct brw_state_flags *b )
+{
+   a->brw |= b->brw;
+   a->cache |= b->cache;
+}
+
+
+static void xor_states( struct brw_state_flags *result,
+			     const struct brw_state_flags *a,
+			      const struct brw_state_flags *b )
+{
+   result->brw = a->brw ^ b->brw;
+   result->cache = a->cache ^ b->cache;
+}
+
+
+/***********************************************************************
+ * Emit all state:
+ */
+void brw_validate_state( struct brw_context *brw )
+{
+   struct brw_state_flags *state = &brw->state.dirty;
+   unsigned i;
+
+   if (brw->emit_state_always) 
+      state->brw |= ~0;
+
+   if (state->cache == 0 &&
+       state->brw == 0)
+      return;
+
+   if (brw->state.dirty.brw & BRW_NEW_SCENE)
+      brw_clear_batch_cache_flush(brw);
+
+   if (BRW_DEBUG) {
+      /* Debug version which enforces various sanity checks on the
+       * state flags which are generated and checked to help ensure
+       * state atoms are ordered correctly in the list.
+       */
+      struct brw_state_flags examined, prev;
+      memset(&examined, 0, sizeof(examined));
+      prev = *state;
+
+      for (i = 0; i < Elements(atoms); i++) {
+	 const struct brw_tracked_state *atom = atoms[i];
+	 struct brw_state_flags generated;
+
+	 assert(atom->dirty.brw ||
+		atom->dirty.cache);
+	 assert(atom->update);
+
+	 if (check_state(state, &atom->dirty)) {
+	    atom->update( brw );
+	 }
+
+	 accumulate_state(&examined, &atom->dirty);
+
+	 /* generated = (prev ^ state)
+	  * if (examined & generated)
+	  *     fail;
+	  */
+	 xor_states(&generated, &prev, state);
+	 assert(!check_state(&examined, &generated));
+	 prev = *state;
+      }
+   }
+   else {
+      for (i = 0; i < Elements(atoms); i++) {
+	 const struct brw_tracked_state *atom = atoms[i];
+
+	 assert(atom->dirty.brw ||
+		atom->dirty.cache);
+	 assert(atom->update);
+
+	 if (check_state(state, &atom->dirty))
+	    atom->update( brw );
+      }
+   }
+
+   memset(state, 0, sizeof(*state));
+}
diff --git a/src/gallium/drivers/i965simple/brw_strings.c b/src/gallium/drivers/i965simple/brw_strings.c
new file mode 100644
index 0000000000..29a41ed1e9
--- /dev/null
+++ b/src/gallium/drivers/i965simple/brw_strings.c
@@ -0,0 +1,72 @@
+/**************************************************************************
+ * 
+ * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+#include "brw_context.h"
+#include "brw_reg.h"
+
+
+static const char *brw_get_vendor( struct pipe_context *pipe )
+{
+   return "Tungsten Graphics, Inc.";
+}
+
+
+static const char *brw_get_name( struct pipe_context *pipe )
+{
+   static char buffer[128];
+   const char *chipset;
+
+   switch (brw_context(pipe)->pci_id) {
+   case PCI_CHIP_I965_Q:
+      chipset = "Intel(R) 965Q";
+      break;
+   case PCI_CHIP_I965_G:
+   case PCI_CHIP_I965_G_1:
+      chipset = "Intel(R) 965G";
+      break;
+   case PCI_CHIP_I965_GM:
+      chipset = "Intel(R) 965GM";
+      break;
+   case PCI_CHIP_I965_GME:
+      chipset = "Intel(R) 965GME/GLE";
+      break;
+   default:
+      chipset = "unknown";
+      break;
+   }
+
+   sprintf(buffer, "pipe/i965 (chipset: %s)", chipset);
+   return buffer;
+}
+
+
+void
+brw_init_string_functions(struct brw_context *brw)
+{
+   brw->pipe.get_name = brw_get_name;
+   brw->pipe.get_vendor = brw_get_vendor;
+}
diff --git a/src/gallium/drivers/i965simple/brw_structs.h b/src/gallium/drivers/i965simple/brw_structs.h
new file mode 100644
index 0000000000..bbb087e95d
--- /dev/null
+++ b/src/gallium/drivers/i965simple/brw_structs.h
@@ -0,0 +1,1348 @@
+/*
+ Copyright (C) Intel Corp.  2006.  All Rights Reserved.
+ Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to
+ develop this 3D driver.
+
+ Permission is hereby granted, free of charge, to any person obtaining
+ a copy of this software and associated documentation files (the
+ "Software"), to deal in the Software without restriction, including
+ without limitation the rights to use, copy, modify, merge, publish,
+ distribute, sublicense, and/or sell copies of the Software, and to
+ permit persons to whom the Software is furnished to do so, subject to
+ the following conditions:
+
+ The above copyright notice and this permission notice (including the
+ next paragraph) shall be included in all copies or substantial
+ portions of the Software.
+
+ THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
+ IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
+ LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
+ OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+
+ **********************************************************************/
+ /*
+  * Authors:
+  *   Keith Whitwell <keith@tungstengraphics.com>
+  */
+
+
+#ifndef BRW_STRUCTS_H
+#define BRW_STRUCTS_H
+
+#include "pipe/p_compiler.h"
+
+/* Command packets:
+ */
+struct header
+{
+   unsigned length:16;
+   unsigned opcode:16;
+};
+
+
+union header_union
+{
+   struct header bits;
+   unsigned dword;
+};
+
+struct brw_3d_control
+{
+   struct
+   {
+      unsigned length:8;
+      unsigned notify_enable:1;
+      unsigned pad:3;
+      unsigned wc_flush_enable:1;
+      unsigned depth_stall_enable:1;
+      unsigned operation:2;
+      unsigned opcode:16;
+   } header;
+
+   struct
+   {
+      unsigned pad:2;
+      unsigned dest_addr_type:1;
+      unsigned dest_addr:29;
+   } dest;
+
+   unsigned dword2;
+   unsigned dword3;
+};
+
+
+struct brw_3d_primitive
+{
+   struct
+   {
+      unsigned length:8;
+      unsigned pad:2;
+      unsigned topology:5;
+      unsigned indexed:1;
+      unsigned opcode:16;
+   } header;
+
+   unsigned verts_per_instance;
+   unsigned start_vert_location;
+   unsigned instance_count;
+   unsigned start_instance_location;
+   unsigned base_vert_location;
+};
+
+/* These seem to be passed around as function args, so it works out
+ * better to keep them as #defines:
+ */
+#define BRW_FLUSH_READ_CACHE           0x1
+#define BRW_FLUSH_STATE_CACHE          0x2
+#define BRW_INHIBIT_FLUSH_RENDER_CACHE 0x4
+#define BRW_FLUSH_SNAPSHOT_COUNTERS    0x8
+
+struct brw_mi_flush
+{
+   unsigned flags:4;
+   unsigned pad:12;
+   unsigned opcode:16;
+};
+
+struct brw_vf_statistics
+{
+   unsigned statistics_enable:1;
+   unsigned pad:15;
+   unsigned opcode:16;
+};
+
+
+
+struct brw_binding_table_pointers
+{
+   struct header header;
+   unsigned vs;
+   unsigned gs;
+   unsigned clp;
+   unsigned sf;
+   unsigned wm;
+};
+
+
+struct brw_blend_constant_color
+{
+   struct header header;
+   float blend_constant_color[4];
+};
+
+
+struct brw_depthbuffer
+{
+   union header_union header;
+
+   union {
+      struct {
+	 unsigned pitch:18;
+	 unsigned format:3;
+	 unsigned pad:4;
+	 unsigned depth_offset_disable:1;
+	 unsigned tile_walk:1;
+	 unsigned tiled_surface:1;
+	 unsigned pad2:1;
+	 unsigned surface_type:3;
+      } bits;
+      unsigned dword;
+   } dword1;
+
+   unsigned dword2_base_addr;
+
+   union {
+      struct {
+	 unsigned pad:1;
+	 unsigned mipmap_layout:1;
+	 unsigned lod:4;
+	 unsigned width:13;
+	 unsigned height:13;
+      } bits;
+      unsigned dword;
+   } dword3;
+
+   union {
+      struct {
+	 unsigned pad:12;
+	 unsigned min_array_element:9;
+	 unsigned depth:11;
+      } bits;
+      unsigned dword;
+   } dword4;
+};
+
+struct brw_drawrect
+{
+   struct header header;
+   unsigned xmin:16;
+   unsigned ymin:16;
+   unsigned xmax:16;
+   unsigned ymax:16;
+   unsigned xorg:16;
+   unsigned yorg:16;
+};
+
+
+
+
+struct brw_global_depth_offset_clamp
+{
+   struct header header;
+   float depth_offset_clamp;
+};
+
+struct brw_indexbuffer
+{
+   union {
+      struct
+      {
+	 unsigned length:8;
+	 unsigned index_format:2;
+	 unsigned cut_index_enable:1;
+	 unsigned pad:5;
+	 unsigned opcode:16;
+      } bits;
+      unsigned dword;
+
+   } header;
+
+   unsigned buffer_start;
+   unsigned buffer_end;
+};
+
+
+struct brw_line_stipple
+{
+   struct header header;
+
+   struct
+   {
+      unsigned pattern:16;
+      unsigned pad:16;
+   } bits0;
+
+   struct
+   {
+      unsigned repeat_count:9;
+      unsigned pad:7;
+      unsigned inverse_repeat_count:16;
+   } bits1;
+};
+
+
+struct brw_pipelined_state_pointers
+{
+   struct header header;
+
+   struct {
+      unsigned pad:5;
+      unsigned offset:27;
+   } vs;
+
+   struct
+   {
+      unsigned enable:1;
+      unsigned pad:4;
+      unsigned offset:27;
+   } gs;
+
+   struct
+   {
+      unsigned enable:1;
+      unsigned pad:4;
+      unsigned offset:27;
+   } clp;
+
+   struct
+   {
+      unsigned pad:5;
+      unsigned offset:27;
+   } sf;
+
+   struct
+   {
+      unsigned pad:5;
+      unsigned offset:27;
+   } wm;
+
+   struct
+   {
+      unsigned pad:5;
+      unsigned offset:27; /* KW: check me! */
+   } cc;
+};
+
+
+struct brw_polygon_stipple_offset
+{
+   struct header header;
+
+   struct {
+      unsigned y_offset:5;
+      unsigned pad:3;
+      unsigned x_offset:5;
+      unsigned pad0:19;
+   } bits0;
+};
+
+
+
+struct brw_polygon_stipple
+{
+   struct header header;
+   unsigned stipple[32];
+};
+
+
+
+struct brw_pipeline_select
+{
+   struct
+   {
+      unsigned pipeline_select:1;
+      unsigned pad:15;
+      unsigned opcode:16;
+   } header;
+};
+
+
+struct brw_pipe_control
+{
+   struct
+   {
+      unsigned length:8;
+      unsigned notify_enable:1;
+      unsigned pad:2;
+      unsigned instruction_state_cache_flush_enable:1;
+      unsigned write_cache_flush_enable:1;
+      unsigned depth_stall_enable:1;
+      unsigned post_sync_operation:2;
+
+      unsigned opcode:16;
+   } header;
+
+   struct
+   {
+      unsigned pad:2;
+      unsigned dest_addr_type:1;
+      unsigned dest_addr:29;
+   } bits1;
+
+   unsigned data0;
+   unsigned data1;
+};
+
+
+struct brw_urb_fence
+{
+   struct
+   {
+      unsigned length:8;
+      unsigned vs_realloc:1;
+      unsigned gs_realloc:1;
+      unsigned clp_realloc:1;
+      unsigned sf_realloc:1;
+      unsigned vfe_realloc:1;
+      unsigned cs_realloc:1;
+      unsigned pad:2;
+      unsigned opcode:16;
+   } header;
+
+   struct
+   {
+      unsigned vs_fence:10;
+      unsigned gs_fence:10;
+      unsigned clp_fence:10;
+      unsigned pad:2;
+   } bits0;
+
+   struct
+   {
+      unsigned sf_fence:10;
+      unsigned vf_fence:10;
+      unsigned cs_fence:10;
+      unsigned pad:2;
+   } bits1;
+};
+
+struct brw_constant_buffer_state /* previously brw_command_streamer */
+{
+   struct header header;
+
+   struct
+   {
+      unsigned nr_urb_entries:3;
+      unsigned pad:1;
+      unsigned urb_entry_size:5;
+      unsigned pad0:23;
+   } bits0;
+};
+
+struct brw_constant_buffer
+{
+   struct
+   {
+      unsigned length:8;
+      unsigned valid:1;
+      unsigned pad:7;
+      unsigned opcode:16;
+   } header;
+
+   struct
+   {
+      unsigned buffer_length:6;
+      unsigned buffer_address:26;
+   } bits0;
+};
+
+struct brw_state_base_address
+{
+   struct header header;
+
+   struct
+   {
+      unsigned modify_enable:1;
+      unsigned pad:4;
+      unsigned general_state_address:27;
+   } bits0;
+
+   struct
+   {
+      unsigned modify_enable:1;
+      unsigned pad:4;
+      unsigned surface_state_address:27;
+   } bits1;
+
+   struct
+   {
+      unsigned modify_enable:1;
+      unsigned pad:4;
+      unsigned indirect_object_state_address:27;
+   } bits2;
+
+   struct
+   {
+      unsigned modify_enable:1;
+      unsigned pad:11;
+      unsigned general_state_upper_bound:20;
+   } bits3;
+
+   struct
+   {
+      unsigned modify_enable:1;
+      unsigned pad:11;
+      unsigned indirect_object_state_upper_bound:20;
+   } bits4;
+};
+
+struct brw_state_prefetch
+{
+   struct header header;
+
+   struct
+   {
+      unsigned prefetch_count:3;
+      unsigned pad:3;
+      unsigned prefetch_pointer:26;
+   } bits0;
+};
+
+struct brw_system_instruction_pointer
+{
+   struct header header;
+
+   struct
+   {
+      unsigned pad:4;
+      unsigned system_instruction_pointer:28;
+   } bits0;
+};
+
+
+
+
+/* State structs for the various fixed function units:
+ */
+
+
+struct thread0
+{
+   unsigned pad0:1;
+   unsigned grf_reg_count:3;
+   unsigned pad1:2;
+   unsigned kernel_start_pointer:26;
+};
+
+struct thread1
+{
+   unsigned ext_halt_exception_enable:1;
+   unsigned sw_exception_enable:1;
+   unsigned mask_stack_exception_enable:1;
+   unsigned timeout_exception_enable:1;
+   unsigned illegal_op_exception_enable:1;
+   unsigned pad0:3;
+   unsigned depth_coef_urb_read_offset:6;	/* WM only */
+   unsigned pad1:2;
+   unsigned floating_point_mode:1;
+   unsigned thread_priority:1;
+   unsigned binding_table_entry_count:8;
+   unsigned pad3:5;
+   unsigned single_program_flow:1;
+};
+
+struct thread2
+{
+   unsigned per_thread_scratch_space:4;
+   unsigned pad0:6;
+   unsigned scratch_space_base_pointer:22;
+};
+
+
+struct thread3
+{
+   unsigned dispatch_grf_start_reg:4;
+   unsigned urb_entry_read_offset:6;
+   unsigned pad0:1;
+   unsigned urb_entry_read_length:6;
+   unsigned pad1:1;
+   unsigned const_urb_entry_read_offset:6;
+   unsigned pad2:1;
+   unsigned const_urb_entry_read_length:6;
+   unsigned pad3:1;
+};
+
+
+
+struct brw_clip_unit_state
+{
+   struct thread0 thread0;
+   struct
+   {
+      unsigned pad0:7;
+      unsigned sw_exception_enable:1;
+      unsigned pad1:3;
+      unsigned mask_stack_exception_enable:1;
+      unsigned pad2:1;
+      unsigned illegal_op_exception_enable:1;
+      unsigned pad3:2;
+      unsigned floating_point_mode:1;
+      unsigned thread_priority:1;
+      unsigned binding_table_entry_count:8;
+      unsigned pad4:5;
+      unsigned single_program_flow:1;
+   } thread1;
+
+   struct thread2 thread2;
+   struct thread3 thread3;
+
+   struct
+   {
+      unsigned pad0:9;
+      unsigned gs_output_stats:1; /* not always */
+      unsigned stats_enable:1;
+      unsigned nr_urb_entries:7;
+      unsigned pad1:1;
+      unsigned urb_entry_allocation_size:5;
+      unsigned pad2:1;
+      unsigned max_threads:1; 	/* may be less */
+      unsigned pad3:6;
+   } thread4;
+
+   struct
+   {
+      unsigned pad0:13;
+      unsigned clip_mode:3;
+      unsigned userclip_enable_flags:8;
+      unsigned userclip_must_clip:1;
+      unsigned pad1:1;
+      unsigned guard_band_enable:1;
+      unsigned viewport_z_clip_enable:1;
+      unsigned viewport_xy_clip_enable:1;
+      unsigned vertex_position_space:1;
+      unsigned api_mode:1;
+      unsigned pad2:1;
+   } clip5;
+
+   struct
+   {
+      unsigned pad0:5;
+      unsigned clipper_viewport_state_ptr:27;
+   } clip6;
+
+
+   float viewport_xmin;
+   float viewport_xmax;
+   float viewport_ymin;
+   float viewport_ymax;
+};
+
+
+
+struct brw_cc_unit_state
+{
+   struct
+   {
+      unsigned pad0:3;
+      unsigned bf_stencil_pass_depth_pass_op:3;
+      unsigned bf_stencil_pass_depth_fail_op:3;
+      unsigned bf_stencil_fail_op:3;
+      unsigned bf_stencil_func:3;
+      unsigned bf_stencil_enable:1;
+      unsigned pad1:2;
+      unsigned stencil_write_enable:1;
+      unsigned stencil_pass_depth_pass_op:3;
+      unsigned stencil_pass_depth_fail_op:3;
+      unsigned stencil_fail_op:3;
+      unsigned stencil_func:3;
+      unsigned stencil_enable:1;
+   } cc0;
+
+
+   struct
+   {
+      unsigned bf_stencil_ref:8;
+      unsigned stencil_write_mask:8;
+      unsigned stencil_test_mask:8;
+      unsigned stencil_ref:8;
+   } cc1;
+
+
+   struct
+   {
+      unsigned logicop_enable:1;
+      unsigned pad0:10;
+      unsigned depth_write_enable:1;
+      unsigned depth_test_function:3;
+      unsigned depth_test:1;
+      unsigned bf_stencil_write_mask:8;
+      unsigned bf_stencil_test_mask:8;
+   } cc2;
+
+
+   struct
+   {
+      unsigned pad0:8;
+      unsigned alpha_test_func:3;
+      unsigned alpha_test:1;
+      unsigned blend_enable:1;
+      unsigned ia_blend_enable:1;
+      unsigned pad1:1;
+      unsigned alpha_test_format:1;
+      unsigned pad2:16;
+   } cc3;
+
+   struct
+   {
+      unsigned pad0:5;
+      unsigned cc_viewport_state_offset:27;
+   } cc4;
+
+   struct
+   {
+      unsigned pad0:2;
+      unsigned ia_dest_blend_factor:5;
+      unsigned ia_src_blend_factor:5;
+      unsigned ia_blend_function:3;
+      unsigned statistics_enable:1;
+      unsigned logicop_func:4;
+      unsigned pad1:11;
+      unsigned dither_enable:1;
+   } cc5;
+
+   struct
+   {
+      unsigned clamp_post_alpha_blend:1;
+      unsigned clamp_pre_alpha_blend:1;
+      unsigned clamp_range:2;
+      unsigned pad0:11;
+      unsigned y_dither_offset:2;
+      unsigned x_dither_offset:2;
+      unsigned dest_blend_factor:5;
+      unsigned src_blend_factor:5;
+      unsigned blend_function:3;
+   } cc6;
+
+   struct {
+      union {
+	 float f;
+	 ubyte ub[4];
+      } alpha_ref;
+   } cc7;
+};
+
+
+
+struct brw_sf_unit_state
+{
+   struct thread0 thread0;
+   struct thread1 thread1;
+   struct thread2 thread2;
+   struct thread3 thread3;
+
+   struct
+   {
+      unsigned pad0:10;
+      unsigned stats_enable:1;
+      unsigned nr_urb_entries:7;
+      unsigned pad1:1;
+      unsigned urb_entry_allocation_size:5;
+      unsigned pad2:1;
+      unsigned max_threads:6;
+      unsigned pad3:1;
+   } thread4;
+
+   struct
+   {
+      unsigned front_winding:1;
+      unsigned viewport_transform:1;
+      unsigned pad0:3;
+      unsigned sf_viewport_state_offset:27;
+   } sf5;
+
+   struct
+   {
+      unsigned pad0:9;
+      unsigned dest_org_vbias:4;
+      unsigned dest_org_hbias:4;
+      unsigned scissor:1;
+      unsigned disable_2x2_trifilter:1;
+      unsigned disable_zero_pix_trifilter:1;
+      unsigned point_rast_rule:2;
+      unsigned line_endcap_aa_region_width:2;
+      unsigned line_width:4;
+      unsigned fast_scissor_disable:1;
+      unsigned cull_mode:2;
+      unsigned aa_enable:1;
+   } sf6;
+
+   struct
+   {
+      unsigned point_size:11;
+      unsigned use_point_size_state:1;
+      unsigned subpixel_precision:1;
+      unsigned sprite_point:1;
+      unsigned pad0:11;
+      unsigned trifan_pv:2;
+      unsigned linestrip_pv:2;
+      unsigned tristrip_pv:2;
+      unsigned line_last_pixel_enable:1;
+   } sf7;
+
+};
+
+
+struct brw_gs_unit_state
+{
+   struct thread0 thread0;
+   struct thread1 thread1;
+   struct thread2 thread2;
+   struct thread3 thread3;
+
+   struct
+   {
+      unsigned pad0:10;
+      unsigned stats_enable:1;
+      unsigned nr_urb_entries:7;
+      unsigned pad1:1;
+      unsigned urb_entry_allocation_size:5;
+      unsigned pad2:1;
+      unsigned max_threads:1;
+      unsigned pad3:6;
+   } thread4;
+
+   struct
+   {
+      unsigned sampler_count:3;
+      unsigned pad0:2;
+      unsigned sampler_state_pointer:27;
+   } gs5;
+
+
+   struct
+   {
+      unsigned max_vp_index:4;
+      unsigned pad0:26;
+      unsigned reorder_enable:1;
+      unsigned pad1:1;
+   } gs6;
+};
+
+
+struct brw_vs_unit_state
+{
+   struct thread0 thread0;
+   struct thread1 thread1;
+   struct thread2 thread2;
+   struct thread3 thread3;
+
+   struct
+   {
+      unsigned pad0:10;
+      unsigned stats_enable:1;
+      unsigned nr_urb_entries:7;
+      unsigned pad1:1;
+      unsigned urb_entry_allocation_size:5;
+      unsigned pad2:1;
+      unsigned max_threads:4;
+      unsigned pad3:3;
+   } thread4;
+
+   struct
+   {
+      unsigned sampler_count:3;
+      unsigned pad0:2;
+      unsigned sampler_state_pointer:27;
+   } vs5;
+
+   struct
+   {
+      unsigned vs_enable:1;
+      unsigned vert_cache_disable:1;
+      unsigned pad0:30;
+   } vs6;
+};
+
+
+struct brw_wm_unit_state
+{
+   struct thread0 thread0;
+   struct thread1 thread1;
+   struct thread2 thread2;
+   struct thread3 thread3;
+
+   struct {
+      unsigned stats_enable:1;
+      unsigned pad0:1;
+      unsigned sampler_count:3;
+      unsigned sampler_state_pointer:27;
+   } wm4;
+
+   struct
+   {
+      unsigned enable_8_pix:1;
+      unsigned enable_16_pix:1;
+      unsigned enable_32_pix:1;
+      unsigned pad0:7;
+      unsigned legacy_global_depth_bias:1;
+      unsigned line_stipple:1;
+      unsigned depth_offset:1;
+      unsigned polygon_stipple:1;
+      unsigned line_aa_region_width:2;
+      unsigned line_endcap_aa_region_width:2;
+      unsigned early_depth_test:1;
+      unsigned thread_dispatch_enable:1;
+      unsigned program_uses_depth:1;
+      unsigned program_computes_depth:1;
+      unsigned program_uses_killpixel:1;
+      unsigned legacy_line_rast: 1;
+      unsigned pad1:1;
+      unsigned max_threads:6;
+      unsigned pad2:1;
+   } wm5;
+
+   float global_depth_offset_constant;
+   float global_depth_offset_scale;
+};
+
+struct brw_sampler_default_color {
+   float color[4];
+};
+
+struct brw_sampler_state
+{
+
+   struct
+   {
+      unsigned shadow_function:3;
+      unsigned lod_bias:11;
+      unsigned min_filter:3;
+      unsigned mag_filter:3;
+      unsigned mip_filter:2;
+      unsigned base_level:5;
+      unsigned pad:1;
+      unsigned lod_preclamp:1;
+      unsigned default_color_mode:1;
+      unsigned pad0:1;
+      unsigned disable:1;
+   } ss0;
+
+   struct
+   {
+      unsigned r_wrap_mode:3;
+      unsigned t_wrap_mode:3;
+      unsigned s_wrap_mode:3;
+      unsigned pad:3;
+      unsigned max_lod:10;
+      unsigned min_lod:10;
+   } ss1;
+
+
+   struct
+   {
+      unsigned pad:5;
+      unsigned default_color_pointer:27;
+   } ss2;
+
+   struct
+   {
+      unsigned pad:19;
+      unsigned max_aniso:3;
+      unsigned chroma_key_mode:1;
+      unsigned chroma_key_index:2;
+      unsigned chroma_key_enable:1;
+      unsigned monochrome_filter_width:3;
+      unsigned monochrome_filter_height:3;
+   } ss3;
+};
+
+
+struct brw_clipper_viewport
+{
+   float xmin;
+   float xmax;
+   float ymin;
+   float ymax;
+};
+
+struct brw_cc_viewport
+{
+   float min_depth;
+   float max_depth;
+};
+
+struct brw_sf_viewport
+{
+   struct {
+      float m00;
+      float m11;
+      float m22;
+      float m30;
+      float m31;
+      float m32;
+   } viewport;
+
+   struct {
+      short xmin;
+      short ymin;
+      short xmax;
+      short ymax;
+   } scissor;
+};
+
+/* Documented in the subsystem/shared-functions/sampler chapter...
+ */
+struct brw_surface_state
+{
+   struct {
+      unsigned cube_pos_z:1;
+      unsigned cube_neg_z:1;
+      unsigned cube_pos_y:1;
+      unsigned cube_neg_y:1;
+      unsigned cube_pos_x:1;
+      unsigned cube_neg_x:1;
+      unsigned pad:4;
+      unsigned mipmap_layout_mode:1;
+      unsigned vert_line_stride_ofs:1;
+      unsigned vert_line_stride:1;
+      unsigned color_blend:1;
+      unsigned writedisable_blue:1;
+      unsigned writedisable_green:1;
+      unsigned writedisable_red:1;
+      unsigned writedisable_alpha:1;
+      unsigned surface_format:9;
+      unsigned data_return_format:1;
+      unsigned pad0:1;
+      unsigned surface_type:3;
+   } ss0;
+
+   struct {
+      unsigned base_addr;
+   } ss1;
+
+   struct {
+      unsigned pad:2;
+      unsigned mip_count:4;
+      unsigned width:13;
+      unsigned height:13;
+   } ss2;
+
+   struct {
+      unsigned tile_walk:1;
+      unsigned tiled_surface:1;
+      unsigned pad:1;
+      unsigned pitch:18;
+      unsigned depth:11;
+   } ss3;
+
+   struct {
+      unsigned pad:19;
+      unsigned min_array_elt:9;
+      unsigned min_lod:4;
+   } ss4;
+};
+
+
+
+struct brw_vertex_buffer_state
+{
+   struct {
+      unsigned pitch:11;
+      unsigned pad:15;
+      unsigned access_type:1;
+      unsigned vb_index:5;
+   } vb0;
+
+   unsigned start_addr;
+   unsigned max_index;
+#if 1
+   unsigned instance_data_step_rate; /* not included for sequential/random vertices? */
+#endif
+};
+
+#define BRW_VBP_MAX 17
+
+struct brw_vb_array_state {
+   struct header header;
+   struct brw_vertex_buffer_state vb[BRW_VBP_MAX];
+};
+
+
+struct brw_vertex_element_state
+{
+   struct
+   {
+      unsigned src_offset:11;
+      unsigned pad:5;
+      unsigned src_format:9;
+      unsigned pad0:1;
+      unsigned valid:1;
+      unsigned vertex_buffer_index:5;
+   } ve0;
+
+   struct
+   {
+      unsigned dst_offset:8;
+      unsigned pad:8;
+      unsigned vfcomponent3:4;
+      unsigned vfcomponent2:4;
+      unsigned vfcomponent1:4;
+      unsigned vfcomponent0:4;
+   } ve1;
+};
+
+#define BRW_VEP_MAX 18
+
+struct brw_vertex_element_packet {
+   struct header header;
+   struct brw_vertex_element_state ve[BRW_VEP_MAX]; /* note: less than _TNL_ATTRIB_MAX */
+};
+
+
+struct brw_urb_immediate {
+   unsigned opcode:4;
+   unsigned offset:6;
+   unsigned swizzle_control:2;
+   unsigned pad:1;
+   unsigned allocate:1;
+   unsigned used:1;
+   unsigned complete:1;
+   unsigned response_length:4;
+   unsigned msg_length:4;
+   unsigned msg_target:4;
+   unsigned pad1:3;
+   unsigned end_of_thread:1;
+};
+
+/* Instruction format for the execution units:
+ */
+
+struct brw_instruction
+{
+   struct
+   {
+      unsigned opcode:7;
+      unsigned pad:1;
+      unsigned access_mode:1;
+      unsigned mask_control:1;
+      unsigned dependency_control:2;
+      unsigned compression_control:2;
+      unsigned thread_control:2;
+      unsigned predicate_control:4;
+      unsigned predicate_inverse:1;
+      unsigned execution_size:3;
+      unsigned destreg__conditonalmod:4; /* destreg - send, conditionalmod - others */
+      unsigned pad0:2;
+      unsigned debug_control:1;
+      unsigned saturate:1;
+   } header;
+
+   union {
+      struct
+      {
+	 unsigned dest_reg_file:2;
+	 unsigned dest_reg_type:3;
+	 unsigned src0_reg_file:2;
+	 unsigned src0_reg_type:3;
+	 unsigned src1_reg_file:2;
+	 unsigned src1_reg_type:3;
+	 unsigned pad:1;
+	 unsigned dest_subreg_nr:5;
+	 unsigned dest_reg_nr:8;
+	 unsigned dest_horiz_stride:2;
+	 unsigned dest_address_mode:1;
+      } da1;
+
+      struct
+      {
+	 unsigned dest_reg_file:2;
+	 unsigned dest_reg_type:3;
+	 unsigned src0_reg_file:2;
+	 unsigned src0_reg_type:3;
+	 unsigned pad:6;
+	 int dest_indirect_offset:10;	/* offset against the deref'd address reg */
+	 unsigned dest_subreg_nr:3; /* subnr for the address reg a0.x */
+	 unsigned dest_horiz_stride:2;
+	 unsigned dest_address_mode:1;
+      } ia1;
+
+      struct
+      {
+	 unsigned dest_reg_file:2;
+	 unsigned dest_reg_type:3;
+	 unsigned src0_reg_file:2;
+	 unsigned src0_reg_type:3;
+	 unsigned src1_reg_file:2;
+	 unsigned src1_reg_type:3;
+	 unsigned pad0:1;
+	 unsigned dest_writemask:4;
+	 unsigned dest_subreg_nr:1;
+	 unsigned dest_reg_nr:8;
+	 unsigned pad1:2;
+	 unsigned dest_address_mode:1;
+      } da16;
+
+      struct
+      {
+	 unsigned dest_reg_file:2;
+	 unsigned dest_reg_type:3;
+	 unsigned src0_reg_file:2;
+	 unsigned src0_reg_type:3;
+	 unsigned pad0:6;
+	 unsigned dest_writemask:4;
+	 int dest_indirect_offset:6;
+	 unsigned dest_subreg_nr:3;
+	 unsigned pad1:2;
+	 unsigned dest_address_mode:1;
+      } ia16;
+   } bits1;
+
+
+   union {
+      struct
+      {
+	 unsigned src0_subreg_nr:5;
+	 unsigned src0_reg_nr:8;
+	 unsigned src0_abs:1;
+	 unsigned src0_negate:1;
+	 unsigned src0_address_mode:1;
+	 unsigned src0_horiz_stride:2;
+	 unsigned src0_width:3;
+	 unsigned src0_vert_stride:4;
+	 unsigned flag_reg_nr:1;
+	 unsigned pad:6;
+      } da1;
+
+      struct
+      {
+	 int src0_indirect_offset:10;
+	 unsigned src0_subreg_nr:3;
+	 unsigned src0_abs:1;
+	 unsigned src0_negate:1;
+	 unsigned src0_address_mode:1;
+	 unsigned src0_horiz_stride:2;
+	 unsigned src0_width:3;
+	 unsigned src0_vert_stride:4;
+	 unsigned flag_reg_nr:1;
+	 unsigned pad:6;
+      } ia1;
+
+      struct
+      {
+	 unsigned src0_swz_x:2;
+	 unsigned src0_swz_y:2;
+	 unsigned src0_subreg_nr:1;
+	 unsigned src0_reg_nr:8;
+	 unsigned src0_abs:1;
+	 unsigned src0_negate:1;
+	 unsigned src0_address_mode:1;
+	 unsigned src0_swz_z:2;
+	 unsigned src0_swz_w:2;
+	 unsigned pad0:1;
+	 unsigned src0_vert_stride:4;
+	 unsigned flag_reg_nr:1;
+	 unsigned pad1:6;
+      } da16;
+
+      struct
+      {
+	 unsigned src0_swz_x:2;
+	 unsigned src0_swz_y:2;
+	 int src0_indirect_offset:6;
+	 unsigned src0_subreg_nr:3;
+	 unsigned src0_abs:1;
+	 unsigned src0_negate:1;
+	 unsigned src0_address_mode:1;
+	 unsigned src0_swz_z:2;
+	 unsigned src0_swz_w:2;
+	 unsigned pad0:1;
+	 unsigned src0_vert_stride:4;
+	 unsigned flag_reg_nr:1;
+	 unsigned pad1:6;
+      } ia16;
+
+   } bits2;
+
+   union
+   {
+      struct
+      {
+	 unsigned src1_subreg_nr:5;
+	 unsigned src1_reg_nr:8;
+	 unsigned src1_abs:1;
+	 unsigned src1_negate:1;
+	 unsigned pad:1;
+	 unsigned src1_horiz_stride:2;
+	 unsigned src1_width:3;
+	 unsigned src1_vert_stride:4;
+	 unsigned pad0:7;
+      } da1;
+
+      struct
+      {
+	 unsigned src1_swz_x:2;
+	 unsigned src1_swz_y:2;
+	 unsigned src1_subreg_nr:1;
+	 unsigned src1_reg_nr:8;
+	 unsigned src1_abs:1;
+	 unsigned src1_negate:1;
+	 unsigned pad0:1;
+	 unsigned src1_swz_z:2;
+	 unsigned src1_swz_w:2;
+	 unsigned pad1:1;
+	 unsigned src1_vert_stride:4;
+	 unsigned pad2:7;
+      } da16;
+
+      struct
+      {
+	 int  src1_indirect_offset:10;
+	 unsigned src1_subreg_nr:3;
+	 unsigned src1_abs:1;
+	 unsigned src1_negate:1;
+	 unsigned pad0:1;
+	 unsigned src1_horiz_stride:2;
+	 unsigned src1_width:3;
+	 unsigned src1_vert_stride:4;
+	 unsigned flag_reg_nr:1;
+	 unsigned pad1:6;
+      } ia1;
+
+      struct
+      {
+	 unsigned src1_swz_x:2;
+	 unsigned src1_swz_y:2;
+	 int  src1_indirect_offset:6;
+	 unsigned src1_subreg_nr:3;
+	 unsigned src1_abs:1;
+	 unsigned src1_negate:1;
+	 unsigned pad0:1;
+	 unsigned src1_swz_z:2;
+	 unsigned src1_swz_w:2;
+	 unsigned pad1:1;
+	 unsigned src1_vert_stride:4;
+	 unsigned flag_reg_nr:1;
+	 unsigned pad2:6;
+      } ia16;
+
+
+      struct
+      {
+	 int  jump_count:16;	/* note: signed */
+	 unsigned  pop_count:4;
+	 unsigned  pad0:12;
+      } if_else;
+
+      struct {
+	 unsigned function:4;
+	 unsigned int_type:1;
+	 unsigned precision:1;
+	 unsigned saturate:1;
+	 unsigned data_type:1;
+	 unsigned pad0:8;
+	 unsigned response_length:4;
+	 unsigned msg_length:4;
+	 unsigned msg_target:4;
+	 unsigned pad1:3;
+	 unsigned end_of_thread:1;
+      } math;
+
+      struct {
+	 unsigned binding_table_index:8;
+	 unsigned sampler:4;
+	 unsigned return_format:2;
+	 unsigned msg_type:2;
+	 unsigned response_length:4;
+	 unsigned msg_length:4;
+	 unsigned msg_target:4;
+	 unsigned pad1:3;
+	 unsigned end_of_thread:1;
+      } sampler;
+
+      struct brw_urb_immediate urb;
+
+      struct {
+	 unsigned binding_table_index:8;
+	 unsigned msg_control:4;
+	 unsigned msg_type:2;
+	 unsigned target_cache:2;
+	 unsigned response_length:4;
+	 unsigned msg_length:4;
+	 unsigned msg_target:4;
+	 unsigned pad1:3;
+	 unsigned end_of_thread:1;
+      } dp_read;
+
+      struct {
+	 unsigned binding_table_index:8;
+	 unsigned msg_control:3;
+	 unsigned pixel_scoreboard_clear:1;
+	 unsigned msg_type:3;
+	 unsigned send_commit_msg:1;
+	 unsigned response_length:4;
+	 unsigned msg_length:4;
+	 unsigned msg_target:4;
+	 unsigned pad1:3;
+	 unsigned end_of_thread:1;
+      } dp_write;
+
+      struct {
+	 unsigned pad:16;
+	 unsigned response_length:4;
+	 unsigned msg_length:4;
+	 unsigned msg_target:4;
+	 unsigned pad1:3;
+	 unsigned end_of_thread:1;
+      } generic;
+
+      int d;
+      unsigned ud;
+   } bits3;
+};
+
+
+#endif
diff --git a/src/gallium/drivers/i965simple/brw_surface.c b/src/gallium/drivers/i965simple/brw_surface.c
new file mode 100644
index 0000000000..518845e4b2
--- /dev/null
+++ b/src/gallium/drivers/i965simple/brw_surface.c
@@ -0,0 +1,210 @@
+/**************************************************************************
+ * 
+ * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+#include "brw_blit.h"
+#include "brw_context.h"
+#include "brw_state.h"
+#include "pipe/p_defines.h"
+#include "pipe/p_util.h"
+#include "pipe/p_inlines.h"
+#include "pipe/p_winsys.h"
+#include "pipe/util/p_tile.h"
+
+
+/*
+ * XXX note: same as code in sp_surface.c
+ */
+static struct pipe_surface *
+brw_get_tex_surface(struct pipe_context *pipe,
+                     struct pipe_texture *pt,
+                     unsigned face, unsigned level, unsigned zslice)
+{
+   struct brw_texture *tex = (struct brw_texture *)pt;
+   struct pipe_surface *ps;
+   unsigned offset;  /* in bytes */
+
+   offset = tex->level_offset[level];
+
+   if (pt->target == PIPE_TEXTURE_CUBE) {
+      offset += tex->image_offset[level][face] * pt->cpp;
+   }
+   else if (pt->target == PIPE_TEXTURE_3D) {
+      offset += tex->image_offset[level][zslice] * pt->cpp;
+   }
+   else {
+      assert(face == 0);
+      assert(zslice == 0);
+   }
+
+   ps = pipe->winsys->surface_alloc(pipe->winsys);
+   if (ps) {
+      assert(ps->format);
+      assert(ps->refcount);
+      pipe_buffer_reference(pipe->winsys, &ps->buffer, tex->buffer);
+      ps->format = pt->format;
+      ps->cpp = pt->cpp;
+      ps->width = pt->width[level];
+      ps->height = pt->height[level];
+      ps->pitch = tex->pitch;
+      ps->offset = offset;
+   }
+   return ps;
+}
+
+
+/* Upload data to a rectangular sub-region.  Lots of choices how to do this:
+ *
+ * - memcpy by span to current destination
+ * - upload data as new buffer and blit
+ *
+ * Currently always memcpy.
+ */
+static void
+brw_surface_data(struct pipe_context *pipe,
+                 struct pipe_surface *dst,
+                 unsigned dstx, unsigned dsty,
+                 const void *src, unsigned src_pitch,
+                 unsigned srcx, unsigned srcy, unsigned width, unsigned height)
+{
+   pipe_copy_rect(pipe_surface_map(dst) + dst->offset,
+                  dst->cpp, dst->pitch,
+                  dstx, dsty, width, height, src, src_pitch, srcx, srcy);
+
+   pipe_surface_unmap(dst);
+}
+
+
+/* Assumes all values are within bounds -- no checking at this level -
+ * do it higher up if required.
+ */
+static void
+brw_surface_copy(struct pipe_context *pipe,
+                 unsigned do_flip,
+                 struct pipe_surface *dst,
+                 unsigned dstx, unsigned dsty,
+                 struct pipe_surface *src,
+                 unsigned srcx, unsigned srcy, unsigned width, unsigned height)
+{
+   assert(dst != src);
+   assert(dst->cpp == src->cpp);
+
+   if (0) {
+      pipe_copy_rect(pipe_surface_map(dst) + dst->offset,
+                     dst->cpp,
+                     dst->pitch,
+                     dstx, dsty,
+                     width, height,
+                     pipe_surface_map(src) + src->offset,
+                     do_flip ? -src->pitch : src->pitch,
+                     srcx, do_flip ? 1 - srcy - height : srcy);
+
+      pipe_surface_unmap(src);
+      pipe_surface_unmap(dst);
+   }
+   else {
+      brw_copy_blit(brw_context(pipe),
+                    do_flip,
+                    dst->cpp,
+                    (short) src->pitch, src->buffer, src->offset, FALSE,
+                    (short) dst->pitch, dst->buffer, dst->offset, FALSE,
+                    (short) srcx, (short) srcy, (short) dstx, (short) dsty,
+                    (short) width, (short) height, PIPE_LOGICOP_COPY);
+   }
+}
+
+/* Fill a rectangular sub-region.  Need better logic about when to
+ * push buffers into AGP - will currently do so whenever possible.
+ */
+static void *
+get_pointer(struct pipe_surface *dst, void *dst_map, unsigned x, unsigned y)
+{
+   return (char *)dst_map + (y * dst->pitch + x) * dst->cpp;
+}
+
+
+static void
+brw_surface_fill(struct pipe_context *pipe,
+                 struct pipe_surface *dst,
+                 unsigned dstx, unsigned dsty,
+                 unsigned width, unsigned height, unsigned value)
+{
+   if (0) {
+      unsigned i, j;
+      void *dst_map = pipe_surface_map(dst);
+
+      switch (dst->cpp) {
+      case 1: {
+	 ubyte *row = get_pointer(dst, dst_map, dstx, dsty);
+	 for (i = 0; i < height; i++) {
+	    memset(row, value, width);
+	    row += dst->pitch;
+	 }
+      }
+	 break;
+      case 2: {
+	 ushort *row = get_pointer(dst, dst_map, dstx, dsty);
+	 for (i = 0; i < height; i++) {
+	    for (j = 0; j < width; j++)
+	       row[j] = (ushort) value;
+	    row += dst->pitch;
+	 }
+      }
+	 break;
+      case 4: {
+	 unsigned *row = get_pointer(dst, dst_map, dstx, dsty);
+	 for (i = 0; i < height; i++) {
+	    for (j = 0; j < width; j++)
+	       row[j] = value;
+	    row += dst->pitch;
+	 }
+      }
+	 break;
+      default:
+	 assert(0);
+	 break;
+      }
+
+      pipe_surface_unmap( dst );
+   }
+   else {
+      brw_fill_blit(brw_context(pipe),
+                    dst->cpp,
+                    (short) dst->pitch,
+                    dst->buffer, dst->offset, FALSE,
+                    (short) dstx, (short) dsty,
+                    (short) width, (short) height,
+                    value);
+   }
+}
+
+void
+brw_init_surface_functions(struct brw_context *brw)
+{
+   brw->pipe.get_tex_surface = brw_get_tex_surface;
+   brw->pipe.surface_copy  = brw_surface_copy;
+   brw->pipe.surface_fill  = brw_surface_fill;
+}
diff --git a/src/gallium/drivers/i965simple/brw_tex_layout.c b/src/gallium/drivers/i965simple/brw_tex_layout.c
new file mode 100644
index 0000000000..90561f1307
--- /dev/null
+++ b/src/gallium/drivers/i965simple/brw_tex_layout.c
@@ -0,0 +1,353 @@
+/*
+ Copyright (C) Intel Corp.  2006.  All Rights Reserved.
+ Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to
+ develop this 3D driver.
+
+ Permission is hereby granted, free of charge, to any person obtaining
+ a copy of this software and associated documentation files (the
+ "Software"), to deal in the Software without restriction, including
+ without limitation the rights to use, copy, modify, merge, publish,
+ distribute, sublicense, and/or sell copies of the Software, and to
+ permit persons to whom the Software is furnished to do so, subject to
+ the following conditions:
+
+ The above copyright notice and this permission notice (including the
+ next paragraph) shall be included in all copies or substantial
+ portions of the Software.
+
+ THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
+ IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
+ LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
+ OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+
+ **********************************************************************/
+ /*
+  * Authors:
+  *   Keith Whitwell <keith@tungstengraphics.com>
+  */
+
+
+/* Code to layout images in a mipmap tree for i965.
+ */
+
+#include "brw_tex_layout.h"
+
+#include "pipe/p_state.h"
+#include "pipe/p_context.h"
+#include "pipe/p_defines.h"
+#include "pipe/p_util.h"
+#include "pipe/p_inlines.h"
+#include "pipe/p_winsys.h"
+
+#include "brw_context.h"
+
+#define FILE_DEBUG_FLAG DEBUG_TEXTURE
+
+#if 0
+unsigned intel_compressed_alignment(unsigned internalFormat)
+{
+    unsigned alignment = 4;
+
+    switch (internalFormat) {
+    case GL_COMPRESSED_RGB_FXT1_3DFX:
+    case GL_COMPRESSED_RGBA_FXT1_3DFX:
+        alignment = 8;
+        break;
+
+    default:
+        break;
+    }
+
+    return alignment;
+}
+#endif
+
+static unsigned minify( unsigned d )
+{
+   return MAX2(1, d>>1);
+}
+
+
+static boolean brw_miptree_layout(struct pipe_context *, struct brw_texture *);
+
+static void intel_miptree_set_image_offset(struct brw_texture *tex,
+                                           unsigned level,
+                                           unsigned img,
+                                           unsigned x, unsigned y)
+{
+   struct pipe_texture *pt = &tex->base;
+   if (img == 0 && level == 0)
+      assert(x == 0 && y == 0);
+   assert(img < tex->nr_images[level]);
+
+   tex->image_offset[level][img] = (x + y * tex->pitch) * pt->cpp;
+}
+
+static void intel_miptree_set_level_info(struct brw_texture *tex,
+                                         unsigned level,
+                                         unsigned nr_images,
+                                         unsigned x, unsigned y,
+                                         unsigned w, unsigned h, unsigned d)
+{
+   struct pipe_texture *pt = &tex->base;
+
+   assert(level < PIPE_MAX_TEXTURE_LEVELS);
+
+   pt->width[level] = w;
+   pt->height[level] = h;
+   pt->depth[level] = d;
+
+   tex->level_offset[level] = (x + y * tex->pitch) * pt->cpp;
+   tex->nr_images[level] = nr_images;
+
+   /*
+   DBG("%s level %d size: %d,%d,%d offset %d,%d (0x%x)\n", __FUNCTION__,
+       level, w, h, d, x, y, tex->level_offset[level]);
+   */
+
+   /* Not sure when this would happen, but anyway: 
+    */
+   if (tex->image_offset[level]) {
+      FREE(tex->image_offset[level]);
+      tex->image_offset[level] = NULL;
+   }
+
+   assert(nr_images);
+   assert(!tex->image_offset[level]);
+
+   tex->image_offset[level] = (unsigned *) MALLOC(nr_images * sizeof(unsigned));
+   tex->image_offset[level][0] = 0;
+}
+
+static void i945_miptree_layout_2d(struct brw_texture *tex)
+{
+   struct pipe_texture *pt = &tex->base;
+   unsigned align_h = 2, align_w = 4;
+   unsigned level;
+   unsigned x = 0;
+   unsigned y = 0;
+   unsigned width = pt->width[0];
+   unsigned height = pt->height[0];
+
+   tex->pitch = pt->width[0];
+
+#if 0
+   if (pt->compressed) {
+      align_w = intel_compressed_alignment(pt->internal_format);
+      tex->pitch = ALIGN(pt->width[0], align_w);
+   }
+#endif
+
+   /* May need to adjust pitch to accomodate the placement of
+    * the 2nd mipmap.  This occurs when the alignment
+    * constraints of mipmap placement push the right edge of the
+    * 2nd mipmap out past the width of its parent.
+    */
+   if (pt->last_level > 0) {
+      unsigned mip1_width;
+
+      if (pt->compressed) {
+         mip1_width = align(minify(pt->width[0]), align_w)
+                      + align(minify(minify(pt->width[0])), align_w);
+      } else {
+         mip1_width = align(minify(pt->width[0]), align_w)
+                      + minify(minify(pt->width[0]));
+      }
+
+      if (mip1_width > tex->pitch) {
+         tex->pitch = mip1_width;
+      }
+   }
+
+   /* Pitch must be a whole number of dwords, even though we
+    * express it in texels.
+    */
+   tex->pitch = align(tex->pitch * pt->cpp, 4) / pt->cpp;
+   tex->total_height = 0;
+
+   for (level = 0; level <= pt->last_level; level++) {
+      unsigned img_height;
+
+      intel_miptree_set_level_info(tex, level, 1, x, y, width,
+				   height, 1);
+
+      if (pt->compressed)
+	 img_height = MAX2(1, height/4);
+      else
+	 img_height = align(height, align_h);
+
+
+      /* Because the images are packed better, the final offset
+       * might not be the maximal one:
+       */
+      tex->total_height = MAX2(tex->total_height, y + img_height);
+
+      /* Layout_below: step right after second mipmap.
+       */
+      if (level == 1) {
+	 x += align(width, align_w);
+      }
+      else {
+	 y += img_height;
+      }
+
+      width  = minify(width);
+      height = minify(height);
+   }
+}
+
+static boolean brw_miptree_layout(struct pipe_context *pipe, struct brw_texture *tex)
+{
+   struct pipe_texture *pt = &tex->base;
+   /* XXX: these vary depending on image format:
+    */
+/*    int align_w = 4; */
+
+   switch (pt->target) {
+   case PIPE_TEXTURE_CUBE:
+   case PIPE_TEXTURE_3D: {
+      unsigned width  = pt->width[0];
+      unsigned height = pt->height[0];
+      unsigned depth = pt->depth[0];
+      unsigned pack_x_pitch, pack_x_nr;
+      unsigned pack_y_pitch;
+      unsigned level;
+      unsigned align_h = 2;
+      unsigned align_w = 4;
+
+      tex->total_height = 0;
+#if 0
+      if (pt->compressed) {
+         align_w = intel_compressed_alignment(pt->internal_format);
+         pt->pitch = align(width, align_w);
+         pack_y_pitch = (height + 3) / 4;
+      } else
+#endif
+      {
+         tex->pitch = align(pt->width[0] * pt->cpp, 4) / pt->cpp;
+         pack_y_pitch = align(pt->height[0], align_h);
+      }
+
+      pack_x_pitch = tex->pitch;
+      pack_x_nr = 1;
+
+      for (level = 0; level <= pt->last_level; level++) {
+	 unsigned nr_images = pt->target == PIPE_TEXTURE_3D ? depth : 6;
+	 int x = 0;
+	 int y = 0;
+	 uint q, j;
+
+	 intel_miptree_set_level_info(tex, level, nr_images,
+				      0, tex->total_height,
+				      width, height, depth);
+
+	 for (q = 0; q < nr_images;) {
+	    for (j = 0; j < pack_x_nr && q < nr_images; j++, q++) {
+	       intel_miptree_set_image_offset(tex, level, q, x, y);
+	       x += pack_x_pitch;
+	    }
+
+	    x = 0;
+	    y += pack_y_pitch;
+	 }
+
+
+	 tex->total_height += y;
+	 width  = minify(width);
+	 height = minify(height);
+	 depth  = minify(depth);
+
+         if (pt->compressed) {
+            pack_y_pitch = (height + 3) / 4;
+
+            if (pack_x_pitch > align(width, align_w)) {
+               pack_x_pitch = align(width, align_w);
+               pack_x_nr <<= 1;
+            }
+         } else {
+            if (pack_x_pitch > 4) {
+               pack_x_pitch >>= 1;
+               pack_x_nr <<= 1;
+               assert(pack_x_pitch * pack_x_nr <= tex->pitch);
+            }
+
+            if (pack_y_pitch > 2) {
+               pack_y_pitch >>= 1;
+               pack_y_pitch = align(pack_y_pitch, align_h);
+            }
+         }
+
+      }
+      break;
+   }
+
+   default:
+      i945_miptree_layout_2d(tex);
+      break;
+   }
+#if 0
+   PRINT("%s: %dx%dx%d - sz 0x%x\n", __FUNCTION__,
+       pt->pitch,
+       pt->total_height,
+       pt->cpp,
+       pt->pitch * pt->total_height * pt->cpp );
+#endif
+
+   return TRUE;
+}
+
+
+struct pipe_texture *
+brw_texture_create(struct pipe_context *pipe, const struct pipe_texture *templat)
+{
+   struct brw_texture *tex = CALLOC_STRUCT(brw_texture);
+
+   if (tex) {
+      tex->base = *templat;
+
+      if (brw_miptree_layout(pipe, tex))
+	 tex->buffer = pipe->winsys->buffer_create(pipe->winsys, 64,
+                                                   PIPE_BUFFER_USAGE_PIXEL,
+                                                   tex->pitch * tex->base.cpp *
+                                                   tex->total_height);
+
+      if (!tex->buffer) {
+	 FREE(tex);
+         return NULL;
+      }
+   }
+
+   return &tex->base;
+}
+
+void
+brw_texture_release(struct pipe_context *pipe, struct pipe_texture **pt)
+{
+   if (!*pt)
+      return;
+
+   /*
+   DBG("%s %p refcount will be %d\n",
+       __FUNCTION__, (void *) *pt, (*pt)->refcount - 1);
+   */
+   if (--(*pt)->refcount <= 0) {
+      struct brw_texture *tex = (struct brw_texture *)*pt;
+      uint i;
+
+      /*
+      DBG("%s deleting %p\n", __FUNCTION__, (void *) tex);
+      */
+
+      pipe_buffer_reference(pipe->winsys, &tex->buffer, NULL);
+
+      for (i = 0; i < PIPE_MAX_TEXTURE_LEVELS; i++)
+         if (tex->image_offset[i])
+            free(tex->image_offset[i]);
+
+      free(tex);
+   }
+   *pt = NULL;
+}
diff --git a/src/gallium/drivers/i965simple/brw_tex_layout.h b/src/gallium/drivers/i965simple/brw_tex_layout.h
new file mode 100644
index 0000000000..cfd6b1ef3a
--- /dev/null
+++ b/src/gallium/drivers/i965simple/brw_tex_layout.h
@@ -0,0 +1,15 @@
+#ifndef BRW_TEX_LAYOUT_H
+#define BRW_TEX_LAYOUT_H
+
+#include "pipe/p_compiler.h"
+
+struct pipe_context;
+struct pipe_texture;
+
+extern struct pipe_texture *
+brw_texture_create(struct pipe_context *pipe, const struct pipe_texture *templat);
+
+extern void
+brw_texture_release(struct pipe_context *pipe, struct pipe_texture **pt);
+
+#endif
diff --git a/src/gallium/drivers/i965simple/brw_urb.c b/src/gallium/drivers/i965simple/brw_urb.c
new file mode 100644
index 0000000000..101a4367b9
--- /dev/null
+++ b/src/gallium/drivers/i965simple/brw_urb.c
@@ -0,0 +1,186 @@
+/*
+ Copyright (C) Intel Corp.  2006.  All Rights Reserved.
+ Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to
+ develop this 3D driver.
+
+ Permission is hereby granted, free of charge, to any person obtaining
+ a copy of this software and associated documentation files (the
+ "Software"), to deal in the Software without restriction, including
+ without limitation the rights to use, copy, modify, merge, publish,
+ distribute, sublicense, and/or sell copies of the Software, and to
+ permit persons to whom the Software is furnished to do so, subject to
+ the following conditions:
+
+ The above copyright notice and this permission notice (including the
+ next paragraph) shall be included in all copies or substantial
+ portions of the Software.
+
+ THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
+ IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
+ LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
+ OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+
+ **********************************************************************/
+ /*
+  * Authors:
+  *   Keith Whitwell <keith@tungstengraphics.com>
+  */
+
+
+#include "brw_context.h"
+//#include "brw_state.h"
+#include "brw_batch.h"
+#include "brw_defines.h"
+
+#define VS 0
+#define GS 1
+#define CLP 2
+#define SF 3
+#define CS 4
+
+/* XXX: Are the min_entry_size numbers useful?
+ * XXX: Verify min_nr_entries, esp for VS.
+ * XXX: Verify SF min_entry_size.
+ */
+static const struct {
+   unsigned min_nr_entries;
+   unsigned preferred_nr_entries;
+   unsigned min_entry_size;
+   unsigned max_entry_size;
+} limits[CS+1] = {
+   { 8, 32, 1, 5 },			/* vs */
+   { 4, 8,  1, 5 },			/* gs */
+   { 6, 8,  1, 5 },			/* clp */
+   { 1, 8,  1, 12 },		        /* sf */
+   { 1, 4,  1, 32 }			/* cs */
+};
+
+
+static boolean check_urb_layout( struct brw_context *brw )
+{
+   brw->urb.vs_start = 0;
+   brw->urb.gs_start = brw->urb.nr_vs_entries * brw->urb.vsize;
+   brw->urb.clip_start = brw->urb.gs_start + brw->urb.nr_gs_entries * brw->urb.vsize;
+   brw->urb.sf_start = brw->urb.clip_start + brw->urb.nr_clip_entries * brw->urb.vsize;
+   brw->urb.cs_start = brw->urb.sf_start + brw->urb.nr_sf_entries * brw->urb.sfsize;
+
+   return brw->urb.cs_start + brw->urb.nr_cs_entries * brw->urb.csize <= 256;
+}
+
+/* Most minimal update, forces re-emit of URB fence packet after GS
+ * unit turned on/off.
+ */
+static void recalculate_urb_fence( struct brw_context *brw )
+{
+   unsigned csize = brw->curbe.total_size;
+   unsigned vsize = brw->vs.prog_data->urb_entry_size;
+   unsigned sfsize = brw->sf.prog_data->urb_entry_size;
+
+   if (csize < limits[CS].min_entry_size)
+      csize = limits[CS].min_entry_size;
+
+   if (vsize < limits[VS].min_entry_size)
+      vsize = limits[VS].min_entry_size;
+
+   if (sfsize < limits[SF].min_entry_size)
+      sfsize = limits[SF].min_entry_size;
+
+   if (brw->urb.vsize < vsize ||
+       brw->urb.sfsize < sfsize ||
+       brw->urb.csize < csize ||
+       (brw->urb.constrained && (brw->urb.vsize > brw->urb.vsize ||
+				 brw->urb.sfsize > brw->urb.sfsize ||
+				 brw->urb.csize > brw->urb.csize))) {
+
+
+      brw->urb.csize = csize;
+      brw->urb.sfsize = sfsize;
+      brw->urb.vsize = vsize;
+
+      brw->urb.nr_vs_entries = limits[VS].preferred_nr_entries;
+      brw->urb.nr_gs_entries = limits[GS].preferred_nr_entries;
+      brw->urb.nr_clip_entries = limits[CLP].preferred_nr_entries;
+      brw->urb.nr_sf_entries = limits[SF].preferred_nr_entries;
+      brw->urb.nr_cs_entries = limits[CS].preferred_nr_entries;
+
+      if (!check_urb_layout(brw)) {
+	 brw->urb.nr_vs_entries = limits[VS].min_nr_entries;
+	 brw->urb.nr_gs_entries = limits[GS].min_nr_entries;
+	 brw->urb.nr_clip_entries = limits[CLP].min_nr_entries;
+	 brw->urb.nr_sf_entries = limits[SF].min_nr_entries;
+	 brw->urb.nr_cs_entries = limits[CS].min_nr_entries;
+
+	 brw->urb.constrained = 1;
+
+	 if (!check_urb_layout(brw)) {
+	    /* This is impossible, given the maximal sizes of urb
+	     * entries and the values for minimum nr of entries
+	     * provided above.
+	     */
+	    debug_printf("couldn't calculate URB layout!\n");
+	    exit(1);
+	 }
+
+	 if (BRW_DEBUG & (DEBUG_URB|DEBUG_FALLBACKS))
+	    debug_printf("URB CONSTRAINED\n");
+      }
+      else
+	 brw->urb.constrained = 0;
+
+      if (BRW_DEBUG & DEBUG_URB)
+	 debug_printf("URB fence: %d ..VS.. %d ..GS.. %d ..CLP.. %d ..SF.. %d ..CS.. %d\n",
+		      brw->urb.vs_start,
+		      brw->urb.gs_start,
+		      brw->urb.clip_start,
+		      brw->urb.sf_start,
+		      brw->urb.cs_start,
+		      256);
+
+      brw->state.dirty.brw |= BRW_NEW_URB_FENCE;
+   }
+}
+
+
+const struct brw_tracked_state brw_recalculate_urb_fence = {
+   .dirty = {
+      .brw = BRW_NEW_CURBE_OFFSETS,
+      .cache = (CACHE_NEW_VS_PROG |
+		CACHE_NEW_SF_PROG)
+   },
+   .update = recalculate_urb_fence
+};
+
+
+
+
+
+void brw_upload_urb_fence(struct brw_context *brw)
+{
+   struct brw_urb_fence uf;
+   memset(&uf, 0, sizeof(uf));
+
+   uf.header.opcode = CMD_URB_FENCE;
+   uf.header.length = sizeof(uf)/4-2;
+   uf.header.vs_realloc = 1;
+   uf.header.gs_realloc = 1;
+   uf.header.clp_realloc = 1;
+   uf.header.sf_realloc = 1;
+   uf.header.vfe_realloc = 1;
+   uf.header.cs_realloc = 1;
+
+   /* The ordering below is correct, not the layout in the
+    * instruction.
+    *
+    * There are 256 urb reg pairs in total.
+    */
+   uf.bits0.vs_fence  = brw->urb.gs_start;
+   uf.bits0.gs_fence  = brw->urb.clip_start;
+   uf.bits0.clp_fence = brw->urb.sf_start;
+   uf.bits1.sf_fence  = brw->urb.cs_start;
+   uf.bits1.cs_fence  = 256;
+
+   BRW_BATCH_STRUCT(brw, &uf);
+}
diff --git a/src/gallium/drivers/i965simple/brw_util.c b/src/gallium/drivers/i965simple/brw_util.c
new file mode 100644
index 0000000000..42391d7c8c
--- /dev/null
+++ b/src/gallium/drivers/i965simple/brw_util.c
@@ -0,0 +1,104 @@
+/*
+ Copyright (C) Intel Corp.  2006.  All Rights Reserved.
+ Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to
+ develop this 3D driver.
+
+ Permission is hereby granted, free of charge, to any person obtaining
+ a copy of this software and associated documentation files (the
+ "Software"), to deal in the Software without restriction, including
+ without limitation the rights to use, copy, modify, merge, publish,
+ distribute, sublicense, and/or sell copies of the Software, and to
+ permit persons to whom the Software is furnished to do so, subject to
+ the following conditions:
+
+ The above copyright notice and this permission notice (including the
+ next paragraph) shall be included in all copies or substantial
+ portions of the Software.
+
+ THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
+ IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
+ LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
+ OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+
+ **********************************************************************/
+ /*
+  * Authors:
+  *   Keith Whitwell <keith@tungstengraphics.com>
+  */
+
+
+#include "brw_util.h"
+#include "brw_defines.h"
+
+#include "pipe/p_defines.h"
+
+unsigned brw_count_bits( unsigned val )
+{
+   unsigned i;
+   for (i = 0; val ; val >>= 1)
+      if (val & 1)
+	 i++;
+   return i;
+}
+
+
+unsigned brw_translate_blend_equation( int mode )
+{
+   switch (mode) {
+   case PIPE_BLEND_ADD:
+      return BRW_BLENDFUNCTION_ADD;
+   case PIPE_BLEND_MIN:
+      return BRW_BLENDFUNCTION_MIN;
+   case PIPE_BLEND_MAX:
+      return BRW_BLENDFUNCTION_MAX;
+   case PIPE_BLEND_SUBTRACT:
+      return BRW_BLENDFUNCTION_SUBTRACT;
+   case PIPE_BLEND_REVERSE_SUBTRACT:
+      return BRW_BLENDFUNCTION_REVERSE_SUBTRACT;
+   default:
+      assert(0);
+      return BRW_BLENDFUNCTION_ADD;
+   }
+}
+
+unsigned brw_translate_blend_factor( int factor )
+{
+   switch(factor) {
+   case PIPE_BLENDFACTOR_ZERO:
+      return BRW_BLENDFACTOR_ZERO;
+   case PIPE_BLENDFACTOR_SRC_ALPHA:
+      return BRW_BLENDFACTOR_SRC_ALPHA;
+   case PIPE_BLENDFACTOR_ONE:
+      return BRW_BLENDFACTOR_ONE;
+   case PIPE_BLENDFACTOR_SRC_COLOR:
+      return BRW_BLENDFACTOR_SRC_COLOR;
+   case PIPE_BLENDFACTOR_INV_SRC_COLOR:
+      return BRW_BLENDFACTOR_INV_SRC_COLOR;
+   case PIPE_BLENDFACTOR_DST_COLOR:
+      return BRW_BLENDFACTOR_DST_COLOR;
+   case PIPE_BLENDFACTOR_INV_DST_COLOR:
+      return BRW_BLENDFACTOR_INV_DST_COLOR;
+   case PIPE_BLENDFACTOR_INV_SRC_ALPHA:
+      return BRW_BLENDFACTOR_INV_SRC_ALPHA;
+   case PIPE_BLENDFACTOR_DST_ALPHA:
+      return BRW_BLENDFACTOR_DST_ALPHA;
+   case PIPE_BLENDFACTOR_INV_DST_ALPHA:
+      return BRW_BLENDFACTOR_INV_DST_ALPHA;
+   case PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE:
+      return BRW_BLENDFACTOR_SRC_ALPHA_SATURATE;
+   case PIPE_BLENDFACTOR_CONST_COLOR:
+      return BRW_BLENDFACTOR_CONST_COLOR;
+   case PIPE_BLENDFACTOR_INV_CONST_COLOR:
+      return BRW_BLENDFACTOR_INV_CONST_COLOR;
+   case PIPE_BLENDFACTOR_CONST_ALPHA:
+      return BRW_BLENDFACTOR_CONST_ALPHA;
+   case PIPE_BLENDFACTOR_INV_CONST_ALPHA:
+      return BRW_BLENDFACTOR_INV_CONST_ALPHA;
+   default:
+      assert(0);
+      return BRW_BLENDFACTOR_ZERO;
+   }
+}
diff --git a/src/gallium/drivers/i965simple/brw_util.h b/src/gallium/drivers/i965simple/brw_util.h
new file mode 100644
index 0000000000..d60e5934db
--- /dev/null
+++ b/src/gallium/drivers/i965simple/brw_util.h
@@ -0,0 +1,43 @@
+/*
+ Copyright (C) Intel Corp.  2006.  All Rights Reserved.
+ Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to
+ develop this 3D driver.
+ 
+ Permission is hereby granted, free of charge, to any person obtaining
+ a copy of this software and associated documentation files (the
+ "Software"), to deal in the Software without restriction, including
+ without limitation the rights to use, copy, modify, merge, publish,
+ distribute, sublicense, and/or sell copies of the Software, and to
+ permit persons to whom the Software is furnished to do so, subject to
+ the following conditions:
+ 
+ The above copyright notice and this permission notice (including the
+ next paragraph) shall be included in all copies or substantial
+ portions of the Software.
+ 
+ THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
+ IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
+ LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
+ OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ 
+ **********************************************************************/
+ /*
+  * Authors:
+  *   Keith Whitwell <keith@tungstengraphics.com>
+  */
+          
+
+#ifndef BRW_UTIL_H
+#define BRW_UTIL_H
+
+#include "pipe/p_state.h"
+
+extern unsigned brw_count_bits( unsigned val );
+extern unsigned brw_translate_blend_factor( int factor );
+extern unsigned brw_translate_blend_equation( int mode );
+
+
+#endif
diff --git a/src/gallium/drivers/i965simple/brw_vs.c b/src/gallium/drivers/i965simple/brw_vs.c
new file mode 100644
index 0000000000..738c6346d5
--- /dev/null
+++ b/src/gallium/drivers/i965simple/brw_vs.c
@@ -0,0 +1,120 @@
+/*
+ Copyright (C) Intel Corp.  2006.  All Rights Reserved.
+ Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to
+ develop this 3D driver.
+ 
+ Permission is hereby granted, free of charge, to any person obtaining
+ a copy of this software and associated documentation files (the
+ "Software"), to deal in the Software without restriction, including
+ without limitation the rights to use, copy, modify, merge, publish,
+ distribute, sublicense, and/or sell copies of the Software, and to
+ permit persons to whom the Software is furnished to do so, subject to
+ the following conditions:
+ 
+ The above copyright notice and this permission notice (including the
+ next paragraph) shall be included in all copies or substantial
+ portions of the Software.
+ 
+ THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
+ IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
+ LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
+ OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ 
+ **********************************************************************/
+ /*
+  * Authors:
+  *   Keith Whitwell <keith@tungstengraphics.com>
+  */
+
+
+#include "brw_context.h"
+#include "brw_vs.h"
+#include "brw_util.h"
+#include "brw_state.h"
+
+
+static void do_vs_prog( struct brw_context *brw,
+			const struct brw_vertex_program *vp,
+			struct brw_vs_prog_key *key )
+{
+   unsigned program_size;
+   const unsigned *program;
+   struct brw_vs_compile c;
+
+   memset(&c, 0, sizeof(c));
+   memcpy(&c.key, key, sizeof(*key));
+
+   brw_init_compile(&c.func);
+   c.vp = vp;
+
+   c.prog_data.outputs_written = vp->program.num_outputs;
+   c.prog_data.inputs_read = vp->program.num_inputs;
+
+#if 0
+   if (c.key.copy_edgeflag) {
+      c.prog_data.outputs_written |= 1<<VERT_RESULT_EDGE;
+      c.prog_data.inputs_read |= 1<<VERT_ATTRIB_EDGEFLAG;
+   }
+#endif
+
+   /* Emit GEN4 code.
+    */
+   brw_vs_emit(&c);
+
+   /* get the program
+    */
+   program = brw_get_program(&c.func, &program_size);
+
+   /*
+    */
+   brw->vs.prog_gs_offset = brw_upload_cache( &brw->cache[BRW_VS_PROG],
+					      &c.key,
+					      sizeof(c.key),
+					      program,
+					      program_size,
+					      &c.prog_data,
+					      &brw->vs.prog_data);
+}
+
+
+static void brw_upload_vs_prog( struct brw_context *brw )
+{
+   struct brw_vs_prog_key key;
+   const struct brw_vertex_program *vp = brw->attribs.VertexProgram;
+
+   assert(vp);
+
+   memset(&key, 0, sizeof(key));
+
+   /* Just upload the program verbatim for now.  Always send it all
+    * the inputs it asks for, whether they are varying or not.
+    */
+   key.program_string_id = vp->id;
+   key.nr_userclip = brw->attribs.Clip.nr;
+   key.copy_edgeflag = (brw->attribs.Raster->fill_cw != PIPE_POLYGON_MODE_FILL ||
+			brw->attribs.Raster->fill_ccw != PIPE_POLYGON_MODE_FILL);
+
+   /* Make an early check for the key.
+    */
+   if (brw_search_cache(&brw->cache[BRW_VS_PROG],
+			&key, sizeof(key),
+			&brw->vs.prog_data,
+			&brw->vs.prog_gs_offset))
+       return;
+
+   do_vs_prog(brw, vp, &key);
+}
+
+
+/* See brw_vs.c:
+ */
+const struct brw_tracked_state brw_vs_prog = {
+   .dirty = {
+      .brw   = BRW_NEW_VS,
+      .cache = 0
+   },
+   .update = brw_upload_vs_prog
+};
diff --git a/src/gallium/drivers/i965simple/brw_vs.h b/src/gallium/drivers/i965simple/brw_vs.h
new file mode 100644
index 0000000000..0e58f043b0
--- /dev/null
+++ b/src/gallium/drivers/i965simple/brw_vs.h
@@ -0,0 +1,82 @@
+/*
+ Copyright (C) Intel Corp.  2006.  All Rights Reserved.
+ Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to
+ develop this 3D driver.
+
+ Permission is hereby granted, free of charge, to any person obtaining
+ a copy of this software and associated documentation files (the
+ "Software"), to deal in the Software without restriction, including
+ without limitation the rights to use, copy, modify, merge, publish,
+ distribute, sublicense, and/or sell copies of the Software, and to
+ permit persons to whom the Software is furnished to do so, subject to
+ the following conditions:
+
+ The above copyright notice and this permission notice (including the
+ next paragraph) shall be included in all copies or substantial
+ portions of the Software.
+
+ THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
+ IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
+ LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
+ OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+
+ **********************************************************************/
+ /*
+  * Authors:
+  *   Keith Whitwell <keith@tungstengraphics.com>
+  */
+
+
+#ifndef BRW_VS_H
+#define BRW_VS_H
+
+
+#include "brw_context.h"
+#include "brw_eu.h"
+
+
+struct brw_vs_prog_key {
+   unsigned program_string_id;
+   unsigned nr_userclip:4;
+   unsigned copy_edgeflag:1;
+   unsigned know_w_is_one:1;
+   unsigned pad:26;
+};
+
+
+struct brw_vs_compile {
+   struct brw_compile func;
+   struct brw_vs_prog_key key;
+   struct brw_vs_prog_data prog_data;
+
+   struct brw_vertex_program *vp;
+
+   unsigned nr_inputs;
+
+   unsigned first_output;
+   unsigned nr_outputs;
+
+   unsigned first_tmp;
+   unsigned last_tmp;
+
+   struct brw_reg r0;
+   struct brw_reg r1;
+   struct brw_reg regs[12][128];
+   struct brw_reg tmp;
+   struct brw_reg stack;
+
+   struct {
+       boolean used_in_src;
+       struct brw_reg reg;
+   } output_regs[128];
+
+   struct brw_reg userplane[6];
+
+};
+
+void brw_vs_emit( struct brw_vs_compile *c );
+
+#endif
diff --git a/src/gallium/drivers/i965simple/brw_vs_emit.c b/src/gallium/drivers/i965simple/brw_vs_emit.c
new file mode 100644
index 0000000000..98915ba101
--- /dev/null
+++ b/src/gallium/drivers/i965simple/brw_vs_emit.c
@@ -0,0 +1,1332 @@
+/*
+ Copyright (C) Intel Corp.  2006.  All Rights Reserved.
+ Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to
+ develop this 3D driver.
+
+ Permission is hereby granted, free of charge, to any person obtaining
+ a copy of this software and associated documentation files (the
+ "Software"), to deal in the Software without restriction, including
+ without limitation the rights to use, copy, modify, merge, publish,
+ distribute, sublicense, and/or sell copies of the Software, and to
+ permit persons to whom the Software is furnished to do so, subject to
+ the following conditions:
+
+ The above copyright notice and this permission notice (including the
+ next paragraph) shall be included in all copies or substantial
+ portions of the Software.
+
+ THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
+ IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
+ LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
+ OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+
+ **********************************************************************/
+ /*
+  * Authors:
+  *   Keith Whitwell <keith@tungstengraphics.com>
+  */
+
+#include "brw_context.h"
+#include "brw_vs.h"
+
+#include "pipe/p_shader_tokens.h"
+#include "pipe/tgsi/util/tgsi_parse.h"
+
+struct brw_prog_info {
+   unsigned num_temps;
+   unsigned num_addrs;
+   unsigned num_consts;
+
+   unsigned writes_psize;
+
+   unsigned pos_idx;
+   unsigned result_edge_idx;
+   unsigned edge_flag_idx;
+   unsigned psize_idx;
+};
+
+/* Do things as simply as possible.  Allocate and populate all regs
+ * ahead of time.
+ */
+static void brw_vs_alloc_regs( struct brw_vs_compile *c,
+                               struct brw_prog_info *info )
+{
+   unsigned i, reg = 0, mrf;
+   unsigned nr_params;
+
+   /* r0 -- reserved as usual
+    */
+   c->r0 = brw_vec8_grf(reg, 0); reg++;
+
+   /* User clip planes from curbe:
+    */
+   if (c->key.nr_userclip) {
+      for (i = 0; i < c->key.nr_userclip; i++) {
+	 c->userplane[i] = stride( brw_vec4_grf(reg+3+i/2, (i%2) * 4), 0, 4, 1);
+      }
+
+      /* Deal with curbe alignment:
+       */
+      reg += ((6+c->key.nr_userclip+3)/4)*2;
+   }
+
+   /* Vertex program parameters from curbe:
+    */
+   nr_params = c->prog_data.max_const;
+   for (i = 0; i < nr_params; i++) {
+      c->regs[TGSI_FILE_CONSTANT][i] = stride(brw_vec4_grf(reg+i/2, (i%2) * 4), 0, 4, 1);
+   }
+   reg += (nr_params+1)/2;
+   c->prog_data.curb_read_length = reg - 1;
+
+
+
+   /* Allocate input regs:
+    */
+   c->nr_inputs = c->vp->program.num_inputs;
+   for (i = 0; i < c->nr_inputs; i++) {
+	 c->regs[TGSI_FILE_INPUT][i] = brw_vec8_grf(reg, 0);
+	 reg++;
+   }
+
+
+   /* Allocate outputs: TODO: could organize the non-position outputs
+    * to go straight into message regs.
+    */
+   c->nr_outputs = 0;
+   c->first_output = reg;
+   mrf = 4;
+   for (i = 0; i < c->vp->program.num_outputs; i++) {
+      c->nr_outputs++;
+#if 0
+      if (i == VERT_RESULT_HPOS) {
+         c->regs[TGSI_FILE_OUTPUT][i] = brw_vec8_grf(reg, 0);
+         reg++;
+      }
+      else if (i == VERT_RESULT_PSIZ) {
+         c->regs[TGSI_FILE_OUTPUT][i] = brw_vec8_grf(reg, 0);
+         reg++;
+         mrf++;		/* just a placeholder?  XXX fix later stages & remove this */
+      }
+      else {
+         c->regs[TGSI_FILE_OUTPUT][i] = brw_message_reg(mrf);
+         mrf++;
+      }
+#else
+      /*treat pos differently for now */
+      if (i == info->pos_idx) {
+         c->regs[TGSI_FILE_OUTPUT][i] = brw_vec8_grf(reg, 0);
+         reg++;
+      } else {
+         c->regs[TGSI_FILE_OUTPUT][i] = brw_message_reg(mrf);
+         mrf++;
+      }
+#endif
+   }
+
+   /* Allocate program temporaries:
+    */
+   for (i = 0; i < info->num_temps; i++) {
+      c->regs[TGSI_FILE_TEMPORARY][i] = brw_vec8_grf(reg, 0);
+      reg++;
+   }
+
+   /* Address reg(s).  Don't try to use the internal address reg until
+    * deref time.
+    */
+   for (i = 0; i < info->num_addrs; i++) {
+      c->regs[TGSI_FILE_ADDRESS][i] =  brw_reg(BRW_GENERAL_REGISTER_FILE,
+                                               reg,
+                                               0,
+                                               BRW_REGISTER_TYPE_D,
+                                               BRW_VERTICAL_STRIDE_8,
+                                               BRW_WIDTH_8,
+                                               BRW_HORIZONTAL_STRIDE_1,
+                                               BRW_SWIZZLE_XXXX,
+                                               TGSI_WRITEMASK_X);
+      reg++;
+   }
+
+   for (i = 0; i < 128; i++) {
+      if (c->output_regs[i].used_in_src) {
+         c->output_regs[i].reg = brw_vec8_grf(reg, 0);
+         reg++;
+      }
+   }
+
+   c->stack =  brw_uw16_reg(BRW_GENERAL_REGISTER_FILE, reg, 0);
+   reg += 2;
+
+
+   /* Some opcodes need an internal temporary:
+    */
+   c->first_tmp = reg;
+   c->last_tmp = reg;		/* for allocation purposes */
+
+   /* Each input reg holds data from two vertices.  The
+    * urb_read_length is the number of registers read from *each*
+    * vertex urb, so is half the amount:
+    */
+   c->prog_data.urb_read_length = (c->nr_inputs+1)/2;
+
+   c->prog_data.urb_entry_size = (c->nr_outputs+2+3)/4;
+   c->prog_data.total_grf = reg;
+}
+
+
+static struct brw_reg get_tmp( struct brw_vs_compile *c )
+{
+   struct brw_reg tmp = brw_vec8_grf(c->last_tmp, 0);
+
+   if (++c->last_tmp > c->prog_data.total_grf)
+      c->prog_data.total_grf = c->last_tmp;
+
+   return tmp;
+}
+
+static void release_tmp( struct brw_vs_compile *c, struct brw_reg tmp )
+{
+   if (tmp.nr == c->last_tmp-1)
+      c->last_tmp--;
+}
+
+static void release_tmps( struct brw_vs_compile *c )
+{
+   c->last_tmp = c->first_tmp;
+}
+
+
+static void unalias1( struct brw_vs_compile *c,
+		      struct brw_reg dst,
+		      struct brw_reg arg0,
+		      void (*func)( struct brw_vs_compile *,
+				    struct brw_reg,
+				    struct brw_reg ))
+{
+   if (dst.file == arg0.file && dst.nr == arg0.nr) {
+      struct brw_compile *p = &c->func;
+      struct brw_reg tmp = brw_writemask(get_tmp(c), dst.dw1.bits.writemask);
+      func(c, tmp, arg0);
+      brw_MOV(p, dst, tmp);
+   }
+   else {
+      func(c, dst, arg0);
+   }
+}
+
+static void unalias2( struct brw_vs_compile *c,
+		      struct brw_reg dst,
+		      struct brw_reg arg0,
+		      struct brw_reg arg1,
+		      void (*func)( struct brw_vs_compile *,
+				    struct brw_reg,
+				    struct brw_reg,
+				    struct brw_reg ))
+{
+   if ((dst.file == arg0.file && dst.nr == arg0.nr) ||
+       (dst.file == arg1.file && dst.nr == arg1.nr)) {
+      struct brw_compile *p = &c->func;
+      struct brw_reg tmp = brw_writemask(get_tmp(c), dst.dw1.bits.writemask);
+      func(c, tmp, arg0, arg1);
+      brw_MOV(p, dst, tmp);
+   }
+   else {
+      func(c, dst, arg0, arg1);
+   }
+}
+
+static void emit_sop( struct brw_compile *p,
+                      struct brw_reg dst,
+                      struct brw_reg arg0,
+                      struct brw_reg arg1,
+		      unsigned cond)
+{
+   brw_push_insn_state(p);
+   brw_CMP(p, brw_null_reg(), cond, arg0, arg1);
+   brw_set_predicate_control(p, BRW_PREDICATE_NONE);
+   brw_MOV(p, dst, brw_imm_f(1.0f));
+   brw_set_predicate_control(p, BRW_PREDICATE_NORMAL);
+   brw_MOV(p, dst, brw_imm_f(0.0f));
+   brw_pop_insn_state(p);
+}
+
+static void emit_seq( struct brw_compile *p,
+                      struct brw_reg dst,
+                      struct brw_reg arg0,
+                      struct brw_reg arg1 )
+{
+   emit_sop(p, dst, arg0, arg1, BRW_CONDITIONAL_EQ);
+}
+
+static void emit_sne( struct brw_compile *p,
+                      struct brw_reg dst,
+                      struct brw_reg arg0,
+                      struct brw_reg arg1 )
+{
+   emit_sop(p, dst, arg0, arg1, BRW_CONDITIONAL_NEQ);
+}
+static void emit_slt( struct brw_compile *p,
+		      struct brw_reg dst,
+		      struct brw_reg arg0,
+		      struct brw_reg arg1 )
+{
+   emit_sop(p, dst, arg0, arg1, BRW_CONDITIONAL_L);
+}
+
+static void emit_sle( struct brw_compile *p,
+		      struct brw_reg dst,
+		      struct brw_reg arg0,
+		      struct brw_reg arg1 )
+{
+   emit_sop(p, dst, arg0, arg1, BRW_CONDITIONAL_LE);
+}
+
+static void emit_sgt( struct brw_compile *p,
+		      struct brw_reg dst,
+		      struct brw_reg arg0,
+		      struct brw_reg arg1 )
+{
+   emit_sop(p, dst, arg0, arg1, BRW_CONDITIONAL_G);
+}
+
+static void emit_sge( struct brw_compile *p,
+		      struct brw_reg dst,
+		      struct brw_reg arg0,
+		      struct brw_reg arg1 )
+{
+  emit_sop(p, dst, arg0, arg1, BRW_CONDITIONAL_GE);
+}
+
+static void emit_max( struct brw_compile *p,
+		      struct brw_reg dst,
+		      struct brw_reg arg0,
+		      struct brw_reg arg1 )
+{
+   brw_CMP(p, brw_null_reg(), BRW_CONDITIONAL_L, arg0, arg1);
+   brw_SEL(p, dst, arg1, arg0);
+   brw_set_predicate_control(p, BRW_PREDICATE_NONE);
+}
+
+static void emit_min( struct brw_compile *p,
+		      struct brw_reg dst,
+		      struct brw_reg arg0,
+		      struct brw_reg arg1 )
+{
+   brw_CMP(p, brw_null_reg(), BRW_CONDITIONAL_L, arg0, arg1);
+   brw_SEL(p, dst, arg0, arg1);
+   brw_set_predicate_control(p, BRW_PREDICATE_NONE);
+}
+
+
+static void emit_math1( struct brw_vs_compile *c,
+			unsigned function,
+			struct brw_reg dst,
+			struct brw_reg arg0,
+			unsigned precision)
+{
+   /* There are various odd behaviours with SEND on the simulator.  In
+    * addition there are documented issues with the fact that the GEN4
+    * processor doesn't do dependency control properly on SEND
+    * results.  So, on balance, this kludge to get around failures
+    * with writemasked math results looks like it might be necessary
+    * whether that turns out to be a simulator bug or not:
+    */
+   struct brw_compile *p = &c->func;
+   struct brw_reg tmp = dst;
+   boolean need_tmp = (dst.dw1.bits.writemask != 0xf ||
+			 dst.file != BRW_GENERAL_REGISTER_FILE);
+
+   if (need_tmp)
+      tmp = get_tmp(c);
+
+   brw_math(p,
+	    tmp,
+	    function,
+	    BRW_MATH_SATURATE_NONE,
+	    2,
+	    arg0,
+	    BRW_MATH_DATA_SCALAR,
+	    precision);
+
+   if (need_tmp) {
+      brw_MOV(p, dst, tmp);
+      release_tmp(c, tmp);
+   }
+}
+
+static void emit_math2( struct brw_vs_compile *c,
+			unsigned function,
+			struct brw_reg dst,
+			struct brw_reg arg0,
+			struct brw_reg arg1,
+			unsigned precision)
+{
+   struct brw_compile *p = &c->func;
+   struct brw_reg tmp = dst;
+   boolean need_tmp = (dst.dw1.bits.writemask != 0xf ||
+			 dst.file != BRW_GENERAL_REGISTER_FILE);
+
+   if (need_tmp)
+      tmp = get_tmp(c);
+
+   brw_MOV(p, brw_message_reg(3), arg1);
+
+   brw_math(p,
+	    tmp,
+	    function,
+	    BRW_MATH_SATURATE_NONE,
+	    2,
+ 	    arg0,
+	    BRW_MATH_DATA_SCALAR,
+	    precision);
+
+   if (need_tmp) {
+      brw_MOV(p, dst, tmp);
+      release_tmp(c, tmp);
+   }
+}
+
+
+
+static void emit_exp_noalias( struct brw_vs_compile *c,
+			      struct brw_reg dst,
+			      struct brw_reg arg0 )
+{
+   struct brw_compile *p = &c->func;
+
+
+   if (dst.dw1.bits.writemask & TGSI_WRITEMASK_X) {
+      struct brw_reg tmp = get_tmp(c);
+      struct brw_reg tmp_d = retype(tmp, BRW_REGISTER_TYPE_D);
+
+      /* tmp_d = floor(arg0.x) */
+      brw_RNDD(p, tmp_d, brw_swizzle1(arg0, 0));
+
+      /* result[0] = 2.0 ^ tmp */
+
+      /* Adjust exponent for floating point:
+       * exp += 127
+       */
+      brw_ADD(p, brw_writemask(tmp_d, TGSI_WRITEMASK_X), tmp_d, brw_imm_d(127));
+
+      /* Install exponent and sign.
+       * Excess drops off the edge:
+       */
+      brw_SHL(p, brw_writemask(retype(dst, BRW_REGISTER_TYPE_D), TGSI_WRITEMASK_X),
+	      tmp_d, brw_imm_d(23));
+
+      release_tmp(c, tmp);
+   }
+
+   if (dst.dw1.bits.writemask & TGSI_WRITEMASK_Y) {
+      /* result[1] = arg0.x - floor(arg0.x) */
+      brw_FRC(p, brw_writemask(dst, TGSI_WRITEMASK_Y), brw_swizzle1(arg0, 0));
+   }
+
+   if (dst.dw1.bits.writemask & TGSI_WRITEMASK_Z) {
+      /* As with the LOG instruction, we might be better off just
+       * doing a taylor expansion here, seeing as we have to do all
+       * the prep work.
+       *
+       * If mathbox partial precision is too low, consider also:
+       * result[3] = result[0] * EXP(result[1])
+       */
+      emit_math1(c,
+		 BRW_MATH_FUNCTION_EXP,
+		 brw_writemask(dst, TGSI_WRITEMASK_Z),
+		 brw_swizzle1(arg0, 0),
+		 BRW_MATH_PRECISION_PARTIAL);
+   }
+
+   if (dst.dw1.bits.writemask & TGSI_WRITEMASK_W) {
+      /* result[3] = 1.0; */
+      brw_MOV(p, brw_writemask(dst, TGSI_WRITEMASK_W), brw_imm_f(1));
+   }
+}
+
+
+static void emit_log_noalias( struct brw_vs_compile *c,
+			      struct brw_reg dst,
+			      struct brw_reg arg0 )
+{
+   struct brw_compile *p = &c->func;
+   struct brw_reg tmp = dst;
+   struct brw_reg tmp_ud = retype(tmp, BRW_REGISTER_TYPE_UD);
+   struct brw_reg arg0_ud = retype(arg0, BRW_REGISTER_TYPE_UD);
+   boolean need_tmp = (dst.dw1.bits.writemask != 0xf ||
+			 dst.file != BRW_GENERAL_REGISTER_FILE);
+
+   if (need_tmp) {
+      tmp = get_tmp(c);
+      tmp_ud = retype(tmp, BRW_REGISTER_TYPE_UD);
+   }
+
+   /* Perform mant = frexpf(fabsf(x), &exp), adjust exp and mnt
+    * according to spec:
+    *
+    * These almost look likey they could be joined up, but not really
+    * practical:
+    *
+    * result[0].f = (x.i & ((1<<31)-1) >> 23) - 127
+    * result[1].i = (x.i & ((1<<23)-1)        + (127<<23)
+    */
+   if (dst.dw1.bits.writemask & TGSI_WRITEMASK_XZ) {
+      brw_AND(p,
+	      brw_writemask(tmp_ud, TGSI_WRITEMASK_X),
+	      brw_swizzle1(arg0_ud, 0),
+	      brw_imm_ud((1U<<31)-1));
+
+      brw_SHR(p,
+	      brw_writemask(tmp_ud, TGSI_WRITEMASK_X),
+	      tmp_ud,
+	      brw_imm_ud(23));
+
+      brw_ADD(p,
+	      brw_writemask(tmp, TGSI_WRITEMASK_X),
+	      retype(tmp_ud, BRW_REGISTER_TYPE_D),	/* does it matter? */
+	      brw_imm_d(-127));
+   }
+
+   if (dst.dw1.bits.writemask & TGSI_WRITEMASK_YZ) {
+      brw_AND(p,
+	      brw_writemask(tmp_ud, TGSI_WRITEMASK_Y),
+	      brw_swizzle1(arg0_ud, 0),
+	      brw_imm_ud((1<<23)-1));
+
+      brw_OR(p,
+	     brw_writemask(tmp_ud, TGSI_WRITEMASK_Y),
+	     tmp_ud,
+	     brw_imm_ud(127<<23));
+   }
+
+   if (dst.dw1.bits.writemask & TGSI_WRITEMASK_Z) {
+      /* result[2] = result[0] + LOG2(result[1]); */
+
+      /* Why bother?  The above is just a hint how to do this with a
+       * taylor series.  Maybe we *should* use a taylor series as by
+       * the time all the above has been done it's almost certainly
+       * quicker than calling the mathbox, even with low precision.
+       *
+       * Options are:
+       *    - result[0] + mathbox.LOG2(result[1])
+       *    - mathbox.LOG2(arg0.x)
+       *    - result[0] + inline_taylor_approx(result[1])
+       */
+      emit_math1(c,
+		 BRW_MATH_FUNCTION_LOG,
+		 brw_writemask(tmp, TGSI_WRITEMASK_Z),
+		 brw_swizzle1(tmp, 1),
+		 BRW_MATH_PRECISION_FULL);
+
+      brw_ADD(p,
+	      brw_writemask(tmp, TGSI_WRITEMASK_Z),
+	      brw_swizzle1(tmp, 2),
+	      brw_swizzle1(tmp, 0));
+   }
+
+   if (dst.dw1.bits.writemask & TGSI_WRITEMASK_W) {
+      /* result[3] = 1.0; */
+      brw_MOV(p, brw_writemask(tmp, TGSI_WRITEMASK_W), brw_imm_f(1));
+   }
+
+   if (need_tmp) {
+      brw_MOV(p, dst, tmp);
+      release_tmp(c, tmp);
+   }
+}
+
+
+
+
+/* Need to unalias - consider swizzles:   r0 = DST r0.xxxx r1
+ */
+static void emit_dst_noalias( struct brw_vs_compile *c,
+			      struct brw_reg dst,
+			      struct brw_reg arg0,
+			      struct brw_reg arg1)
+{
+   struct brw_compile *p = &c->func;
+
+   /* There must be a better way to do this:
+    */
+   if (dst.dw1.bits.writemask & TGSI_WRITEMASK_X)
+      brw_MOV(p, brw_writemask(dst, TGSI_WRITEMASK_X), brw_imm_f(1.0));
+   if (dst.dw1.bits.writemask & TGSI_WRITEMASK_Y)
+      brw_MUL(p, brw_writemask(dst, TGSI_WRITEMASK_Y), arg0, arg1);
+   if (dst.dw1.bits.writemask & TGSI_WRITEMASK_Z)
+      brw_MOV(p, brw_writemask(dst, TGSI_WRITEMASK_Z), arg0);
+   if (dst.dw1.bits.writemask & TGSI_WRITEMASK_W)
+      brw_MOV(p, brw_writemask(dst, TGSI_WRITEMASK_W), arg1);
+}
+
+static void emit_xpd( struct brw_compile *p,
+		      struct brw_reg dst,
+		      struct brw_reg t,
+		      struct brw_reg u)
+{
+   brw_MUL(p, brw_null_reg(), brw_swizzle(t, 1,2,0,3),  brw_swizzle(u,2,0,1,3));
+   brw_MAC(p, dst,     negate(brw_swizzle(t, 2,0,1,3)), brw_swizzle(u,1,2,0,3));
+}
+
+
+
+static void emit_lit_noalias( struct brw_vs_compile *c,
+			      struct brw_reg dst,
+			      struct brw_reg arg0 )
+{
+   struct brw_compile *p = &c->func;
+   struct brw_instruction *if_insn;
+   struct brw_reg tmp = dst;
+   boolean need_tmp = (dst.file != BRW_GENERAL_REGISTER_FILE);
+
+   if (need_tmp)
+      tmp = get_tmp(c);
+
+   brw_MOV(p, brw_writemask(dst, TGSI_WRITEMASK_YZ), brw_imm_f(0));
+   brw_MOV(p, brw_writemask(dst, TGSI_WRITEMASK_XW), brw_imm_f(1));
+
+   /* Need to use BRW_EXECUTE_8 and also do an 8-wide compare in order
+    * to get all channels active inside the IF.  In the clipping code
+    * we run with NoMask, so it's not an option and we can use
+    * BRW_EXECUTE_1 for all comparisions.
+    */
+   brw_CMP(p, brw_null_reg(), BRW_CONDITIONAL_G, brw_swizzle1(arg0,0), brw_imm_f(0));
+   if_insn = brw_IF(p, BRW_EXECUTE_8);
+   {
+      brw_MOV(p, brw_writemask(dst, TGSI_WRITEMASK_Y), brw_swizzle1(arg0,0));
+
+      brw_CMP(p, brw_null_reg(), BRW_CONDITIONAL_G, brw_swizzle1(arg0,1), brw_imm_f(0));
+      brw_MOV(p, brw_writemask(tmp, TGSI_WRITEMASK_Z),  brw_swizzle1(arg0,1));
+      brw_set_predicate_control(p, BRW_PREDICATE_NONE);
+
+      emit_math2(c,
+		 BRW_MATH_FUNCTION_POW,
+		 brw_writemask(dst, TGSI_WRITEMASK_Z),
+		 brw_swizzle1(tmp, 2),
+		 brw_swizzle1(arg0, 3),
+		 BRW_MATH_PRECISION_PARTIAL);
+   }
+
+   brw_ENDIF(p, if_insn);
+}
+
+
+
+
+
+/* TODO: relative addressing!
+ */
+static struct brw_reg get_reg( struct brw_vs_compile *c,
+			       unsigned file,
+			       unsigned index )
+{
+   switch (file) {
+   case TGSI_FILE_TEMPORARY:
+   case TGSI_FILE_INPUT:
+   case TGSI_FILE_OUTPUT:
+      assert(c->regs[file][index].nr != 0);
+      return c->regs[file][index];
+   case TGSI_FILE_CONSTANT:
+      assert(c->regs[TGSI_FILE_CONSTANT][index + c->prog_data.num_imm].nr != 0);
+      return c->regs[TGSI_FILE_CONSTANT][index + c->prog_data.num_imm];
+   case TGSI_FILE_IMMEDIATE:
+      assert(c->regs[TGSI_FILE_CONSTANT][index].nr != 0);
+      return c->regs[TGSI_FILE_CONSTANT][index];
+   case TGSI_FILE_ADDRESS:
+      assert(index == 0);
+      return c->regs[file][index];
+
+   case TGSI_FILE_NULL:			/* undef values */
+      return brw_null_reg();
+
+   default:
+      assert(0);
+      return brw_null_reg();
+   }
+}
+
+
+
+static struct brw_reg deref( struct brw_vs_compile *c,
+			     struct brw_reg arg,
+			     int offset)
+{
+   struct brw_compile *p = &c->func;
+   struct brw_reg tmp = vec4(get_tmp(c));
+   struct brw_reg vp_address = retype(vec1(get_reg(c, TGSI_FILE_ADDRESS, 0)), BRW_REGISTER_TYPE_UW);
+   unsigned byte_offset = arg.nr * 32 + arg.subnr + offset * 16;
+   struct brw_reg indirect = brw_vec4_indirect(0,0);
+
+   {
+      brw_push_insn_state(p);
+      brw_set_access_mode(p, BRW_ALIGN_1);
+
+      /* This is pretty clunky - load the address register twice and
+       * fetch each 4-dword value in turn.  There must be a way to do
+       * this in a single pass, but I couldn't get it to work.
+       */
+      brw_ADD(p, brw_address_reg(0), vp_address, brw_imm_d(byte_offset));
+      brw_MOV(p, tmp, indirect);
+
+      brw_ADD(p, brw_address_reg(0), suboffset(vp_address, 8), brw_imm_d(byte_offset));
+      brw_MOV(p, suboffset(tmp, 4), indirect);
+
+      brw_pop_insn_state(p);
+   }
+
+   return vec8(tmp);
+}
+
+
+static void emit_arl( struct brw_vs_compile *c,
+		      struct brw_reg dst,
+		      struct brw_reg arg0 )
+{
+   struct brw_compile *p = &c->func;
+   struct brw_reg tmp = dst;
+   boolean need_tmp = (dst.file != BRW_GENERAL_REGISTER_FILE);
+
+   if (need_tmp)
+      tmp = get_tmp(c);
+
+   brw_RNDD(p, tmp, arg0);
+   brw_MUL(p, dst, tmp, brw_imm_d(16));
+
+   if (need_tmp)
+      release_tmp(c, tmp);
+}
+
+
+/* Will return mangled results for SWZ op.  The emit_swz() function
+ * ignores this result and recalculates taking extended swizzles into
+ * account.
+ */
+static struct brw_reg get_arg( struct brw_vs_compile *c,
+			       struct tgsi_src_register *src )
+{
+   struct brw_reg reg;
+
+   if (src->File == TGSI_FILE_NULL)
+      return brw_null_reg();
+
+#if 0
+   if (src->RelAddr)
+      reg = deref(c, c->regs[PROGRAM_STATE_VAR][0], src->Index);
+   else
+#endif
+      reg = get_reg(c, src->File, src->Index);
+
+   /* Convert 3-bit swizzle to 2-bit.
+    */
+   reg.dw1.bits.swizzle = BRW_SWIZZLE4(src->SwizzleX,
+				       src->SwizzleY,
+				       src->SwizzleZ,
+				       src->SwizzleW);
+
+   /* Note this is ok for non-swizzle instructions:
+    */
+   reg.negate = src->Negate ? 1 : 0;
+
+   return reg;
+}
+
+
+static struct brw_reg get_dst( struct brw_vs_compile *c,
+			       const struct tgsi_dst_register *dst )
+{
+   struct brw_reg reg = get_reg(c, dst->File, dst->Index);
+
+   reg.dw1.bits.writemask = dst->WriteMask;
+
+   return reg;
+}
+
+
+
+
+static void emit_swz( struct brw_vs_compile *c,
+		      struct brw_reg dst,
+		      struct tgsi_src_register src )
+{
+   struct brw_compile *p = &c->func;
+   unsigned zeros_mask = 0;
+   unsigned ones_mask = 0;
+   unsigned src_mask = 0;
+   ubyte src_swz[4];
+   boolean need_tmp = (src.Negate &&
+			 dst.file != BRW_GENERAL_REGISTER_FILE);
+   struct brw_reg tmp = dst;
+   unsigned i;
+
+   if (need_tmp)
+      tmp = get_tmp(c);
+
+   for (i = 0; i < 4; i++) {
+      if (dst.dw1.bits.writemask & (1<<i)) {
+	 ubyte s = 0;
+         switch(i) {
+         case 0:
+            s = src.SwizzleX;
+            break;
+            s = src.SwizzleY;
+         case 1:
+            break;
+            s = src.SwizzleZ;
+         case 2:
+            break;
+            s = src.SwizzleW;
+         case 3:
+            break;
+         }
+	 switch (s) {
+	 case TGSI_SWIZZLE_X:
+	 case TGSI_SWIZZLE_Y:
+	 case TGSI_SWIZZLE_Z:
+	 case TGSI_SWIZZLE_W:
+	    src_mask |= 1<<i;
+	    src_swz[i] = s;
+	    break;
+	 case TGSI_EXTSWIZZLE_ZERO:
+	    zeros_mask |= 1<<i;
+	    break;
+	 case TGSI_EXTSWIZZLE_ONE:
+	    ones_mask |= 1<<i;
+	    break;
+	 }
+      }
+   }
+
+   /* Do src first, in case dst aliases src:
+    */
+   if (src_mask) {
+      struct brw_reg arg0;
+
+#if 0
+      if (src.RelAddr)
+	 arg0 = deref(c, c->regs[PROGRAM_STATE_VAR][0], src.Index);
+      else
+#endif
+	 arg0 = get_reg(c, src.File, src.Index);
+
+      arg0 = brw_swizzle(arg0,
+			 src_swz[0], src_swz[1],
+			 src_swz[2], src_swz[3]);
+
+      brw_MOV(p, brw_writemask(tmp, src_mask), arg0);
+   }
+
+   if (zeros_mask)
+      brw_MOV(p, brw_writemask(tmp, zeros_mask), brw_imm_f(0));
+
+   if (ones_mask)
+      brw_MOV(p, brw_writemask(tmp, ones_mask), brw_imm_f(1));
+
+   if (src.Negate)
+      brw_MOV(p, brw_writemask(tmp, src.Negate), negate(tmp));
+
+   if (need_tmp) {
+      brw_MOV(p, dst, tmp);
+      release_tmp(c, tmp);
+   }
+}
+
+
+
+/* Post-vertex-program processing.  Send the results to the URB.
+ */
+static void emit_vertex_write( struct brw_vs_compile *c, struct brw_prog_info *info)
+{
+   struct brw_compile *p = &c->func;
+   struct brw_reg m0 = brw_message_reg(0);
+   struct brw_reg pos = c->regs[TGSI_FILE_OUTPUT][info->pos_idx];
+   struct brw_reg ndc;
+
+   if (c->key.copy_edgeflag) {
+      brw_MOV(p,
+	      get_reg(c, TGSI_FILE_OUTPUT, info->result_edge_idx),
+	      get_reg(c, TGSI_FILE_INPUT, info->edge_flag_idx));
+   }
+
+
+   /* Build ndc coords?   TODO: Shortcircuit when w is known to be one.
+    */
+   if (!c->key.know_w_is_one) {
+      ndc = get_tmp(c);
+      emit_math1(c, BRW_MATH_FUNCTION_INV, ndc, brw_swizzle1(pos, 3), BRW_MATH_PRECISION_FULL);
+      brw_MUL(p, brw_writemask(ndc, TGSI_WRITEMASK_XYZ), pos, ndc);
+   }
+   else {
+      ndc = pos;
+   }
+
+   /* This includes the workaround for -ve rhw, so is no longer an
+    * optional step:
+    */
+   if (info->writes_psize ||
+       c->key.nr_userclip ||
+       !c->key.know_w_is_one)
+   {
+      struct brw_reg header1 = retype(get_tmp(c), BRW_REGISTER_TYPE_UD);
+      unsigned i;
+
+      brw_MOV(p, header1, brw_imm_ud(0));
+
+      brw_set_access_mode(p, BRW_ALIGN_16);
+
+      if (info->writes_psize) {
+	 struct brw_reg psiz = c->regs[TGSI_FILE_OUTPUT][info->psize_idx];
+	 brw_MUL(p, brw_writemask(header1, TGSI_WRITEMASK_W),
+                 brw_swizzle1(psiz, 0), brw_imm_f(1<<11));
+	 brw_AND(p, brw_writemask(header1, TGSI_WRITEMASK_W), header1,
+                 brw_imm_ud(0x7ff<<8));
+      }
+
+
+      for (i = 0; i < c->key.nr_userclip; i++) {
+	 brw_set_conditionalmod(p, BRW_CONDITIONAL_L);
+	 brw_DP4(p, brw_null_reg(), pos, c->userplane[i]);
+	 brw_OR(p, brw_writemask(header1, TGSI_WRITEMASK_W), header1, brw_imm_ud(1<<i));
+	 brw_set_predicate_control(p, BRW_PREDICATE_NONE);
+      }
+
+
+      /* i965 clipping workaround:
+       * 1) Test for -ve rhw
+       * 2) If set,
+       *      set ndc = (0,0,0,0)
+       *      set ucp[6] = 1
+       *
+       * Later, clipping will detect ucp[6] and ensure the primitive is
+       * clipped against all fixed planes.
+       */
+      if (!c->key.know_w_is_one) {
+	 brw_CMP(p,
+		 vec8(brw_null_reg()),
+		 BRW_CONDITIONAL_L,
+		 brw_swizzle1(ndc, 3),
+		 brw_imm_f(0));
+
+	 brw_OR(p, brw_writemask(header1, TGSI_WRITEMASK_W), header1, brw_imm_ud(1<<6));
+	 brw_MOV(p, ndc, brw_imm_f(0));
+	 brw_set_predicate_control(p, BRW_PREDICATE_NONE);
+      }
+
+      brw_set_access_mode(p, BRW_ALIGN_1);	/* why? */
+      brw_MOV(p, retype(brw_message_reg(1), BRW_REGISTER_TYPE_UD), header1);
+      brw_set_access_mode(p, BRW_ALIGN_16);
+
+      release_tmp(c, header1);
+   }
+   else {
+      brw_MOV(p, retype(brw_message_reg(1), BRW_REGISTER_TYPE_UD), brw_imm_ud(0));
+   }
+
+
+   /* Emit the (interleaved) headers for the two vertices - an 8-reg
+    * of zeros followed by two sets of NDC coordinates:
+    */
+   brw_set_access_mode(p, BRW_ALIGN_1);
+   brw_MOV(p, offset(m0, 2), ndc);
+   brw_MOV(p, offset(m0, 3), pos);
+
+
+   brw_urb_WRITE(p,
+		 brw_null_reg(), /* dest */
+		 0,		/* starting mrf reg nr */
+		 c->r0,		/* src */
+		 0,		/* allocate */
+		 1,		/* used */
+		 c->nr_outputs + 3, /* msg len */
+		 0,		/* response len */
+		 1, 		/* eot */
+		 1, 		/* writes complete */
+		 0, 		/* urb destination offset */
+		 BRW_URB_SWIZZLE_INTERLEAVE);
+
+}
+
+static void
+post_vs_emit( struct brw_vs_compile *c, struct brw_instruction *end_inst )
+{
+   struct tgsi_parse_context parse;
+   const struct tgsi_token *tokens = c->vp->program.tokens;
+   tgsi_parse_init(&parse, tokens);
+   while (!tgsi_parse_end_of_tokens(&parse)) {
+      tgsi_parse_token(&parse);
+      if (parse.FullToken.Token.Type == TGSI_TOKEN_TYPE_INSTRUCTION) {
+#if 0
+         struct brw_instruction *brw_inst1, *brw_inst2;
+         const struct tgsi_full_instruction *inst1, *inst2;
+         int offset;
+         inst1 = &parse.FullToken.FullInstruction;
+         brw_inst1 = inst1->Data;
+         switch (inst1->Opcode) {
+	 case TGSI_OPCODE_CAL:
+	 case TGSI_OPCODE_BRA:
+	    target_insn = inst1->BranchTarget;
+	    inst2 = &c->vp->program.Base.Instructions[target_insn];
+	    brw_inst2 = inst2->Data;
+	    offset = brw_inst2 - brw_inst1;
+	    brw_set_src1(brw_inst1, brw_imm_d(offset*16));
+	    break;
+	 case TGSI_OPCODE_END:
+	    offset = end_inst - brw_inst1;
+	    brw_set_src1(brw_inst1, brw_imm_d(offset*16));
+	    break;
+	 default:
+	    break;
+         }
+#endif
+      }
+   }
+   tgsi_parse_free(&parse);
+}
+
+static void process_declaration(const struct tgsi_full_declaration *decl,
+                                struct brw_prog_info *info)
+{
+   int first = decl->u.DeclarationRange.First;
+   int last = decl->u.DeclarationRange.Last;
+
+   assert (decl->Declaration.Declare != TGSI_DECLARE_MASK);
+   
+   switch(decl->Declaration.File) {
+   case TGSI_FILE_CONSTANT: 
+      info->num_consts += last - first + 1;
+      break;
+   case TGSI_FILE_INPUT: {
+   }
+      break;
+   case TGSI_FILE_OUTPUT: {
+      assert(last == first);	/* for now */
+      if (decl->Declaration.Semantic) {
+         switch (decl->Semantic.SemanticName) {
+         case TGSI_SEMANTIC_POSITION: {
+            info->pos_idx = first;
+         }
+            break;
+         case TGSI_SEMANTIC_COLOR:
+            break;
+         case TGSI_SEMANTIC_BCOLOR:
+            break;
+         case TGSI_SEMANTIC_FOG:
+            break;
+         case TGSI_SEMANTIC_PSIZE: {
+            info->writes_psize = TRUE;
+            info->psize_idx = first;
+         }
+            break;
+         case TGSI_SEMANTIC_GENERIC:
+            break;
+         }
+      }
+   }
+      break;
+   case TGSI_FILE_TEMPORARY: {
+      info->num_temps += (last - first) + 1;
+   }
+      break;
+   case TGSI_FILE_SAMPLER: {
+   }
+      break;
+   case TGSI_FILE_ADDRESS: {
+      info->num_addrs += (last - first) + 1;
+   }
+      break;
+   case TGSI_FILE_IMMEDIATE: {
+   }
+      break;
+   case TGSI_FILE_NULL: {
+   }
+      break;
+   }
+}
+
+static void process_instruction(struct brw_vs_compile *c,
+                                struct tgsi_full_instruction *inst,
+                                struct brw_prog_info *info)
+{
+   struct brw_reg args[3], dst;
+   struct brw_compile *p = &c->func;
+   struct brw_indirect stack_index = brw_indirect(0, 0);
+   unsigned i;
+   unsigned index;
+   unsigned file;
+   /*FIXME: might not be the only one*/
+   const struct tgsi_dst_register *dst_reg = &inst->FullDstRegisters[0].DstRegister;
+   /*
+   struct brw_instruction *if_inst[MAX_IFSN];
+   unsigned insn, if_insn = 0;
+   */
+
+   for (i = 0; i < 3; i++) {
+      struct tgsi_full_src_register *src = &inst->FullSrcRegisters[i];
+      index = src->SrcRegister.Index;
+      file = src->SrcRegister.File;
+      if (file == TGSI_FILE_OUTPUT && c->output_regs[index].used_in_src)
+         args[i] = c->output_regs[index].reg;
+      else
+         args[i] = get_arg(c, &src->SrcRegister);
+   }
+
+   /* Get dest regs.  Note that it is possible for a reg to be both
+    * dst and arg, given the static allocation of registers.  So
+    * care needs to be taken emitting multi-operation instructions.
+    */
+   index = dst_reg->Index;
+   file = dst_reg->File;
+   if (file == TGSI_FILE_OUTPUT && c->output_regs[index].used_in_src)
+      dst = c->output_regs[index].reg;
+   else
+      dst = get_dst(c, dst_reg);
+
+   switch (inst->Instruction.Opcode) {
+   case TGSI_OPCODE_ABS:
+      brw_MOV(p, dst, brw_abs(args[0]));
+      break;
+   case TGSI_OPCODE_ADD:
+      brw_ADD(p, dst, args[0], args[1]);
+      break;
+   case TGSI_OPCODE_DP3:
+      brw_DP3(p, dst, args[0], args[1]);
+      break;
+   case TGSI_OPCODE_DP4:
+      brw_DP4(p, dst, args[0], args[1]);
+      break;
+   case TGSI_OPCODE_DPH:
+      brw_DPH(p, dst, args[0], args[1]);
+      break;
+   case TGSI_OPCODE_DST:
+      unalias2(c, dst, args[0], args[1], emit_dst_noalias);
+      break;
+   case TGSI_OPCODE_EXP:
+      unalias1(c, dst, args[0], emit_exp_noalias);
+      break;
+   case TGSI_OPCODE_EX2:
+      emit_math1(c, BRW_MATH_FUNCTION_EXP, dst, args[0], BRW_MATH_PRECISION_FULL);
+      break;
+   case TGSI_OPCODE_ARL:
+      emit_arl(c, dst, args[0]);
+      break;
+   case TGSI_OPCODE_FLR:
+      brw_RNDD(p, dst, args[0]);
+      break;
+   case TGSI_OPCODE_FRC:
+      brw_FRC(p, dst, args[0]);
+      break;
+   case TGSI_OPCODE_LOG:
+      unalias1(c, dst, args[0], emit_log_noalias);
+      break;
+   case TGSI_OPCODE_LG2:
+      emit_math1(c, BRW_MATH_FUNCTION_LOG, dst, args[0], BRW_MATH_PRECISION_FULL);
+      break;
+   case TGSI_OPCODE_LIT:
+      unalias1(c, dst, args[0], emit_lit_noalias);
+      break;
+   case TGSI_OPCODE_MAD:
+      brw_MOV(p, brw_acc_reg(), args[2]);
+      brw_MAC(p, dst, args[0], args[1]);
+      break;
+   case TGSI_OPCODE_MAX:
+      emit_max(p, dst, args[0], args[1]);
+      break;
+   case TGSI_OPCODE_MIN:
+      emit_min(p, dst, args[0], args[1]);
+      break;
+   case TGSI_OPCODE_MOV:
+#if 0
+   case TGSI_OPCODE_SWZ:
+      /* The args[0] value can't be used here as it won't have
+       * correctly encoded the full swizzle:
+       */
+      emit_swz(c, dst, inst->SrcReg[0] );
+#endif
+      brw_MOV(p, dst, args[0]);
+      break;
+   case TGSI_OPCODE_MUL:
+      brw_MUL(p, dst, args[0], args[1]);
+      break;
+   case TGSI_OPCODE_POW:
+      emit_math2(c, BRW_MATH_FUNCTION_POW, dst, args[0], args[1], BRW_MATH_PRECISION_FULL);
+      break;
+   case TGSI_OPCODE_RCP:
+      emit_math1(c, BRW_MATH_FUNCTION_INV, dst, args[0], BRW_MATH_PRECISION_FULL);
+      break;
+   case TGSI_OPCODE_RSQ:
+      emit_math1(c, BRW_MATH_FUNCTION_RSQ, dst, args[0], BRW_MATH_PRECISION_FULL);
+      break;
+
+   case TGSI_OPCODE_SEQ:
+      emit_seq(p, dst, args[0], args[1]);
+      break;
+   case TGSI_OPCODE_SNE:
+      emit_sne(p, dst, args[0], args[1]);
+      break;
+   case TGSI_OPCODE_SGE:
+      emit_sge(p, dst, args[0], args[1]);
+      break;
+   case TGSI_OPCODE_SGT:
+      emit_sgt(p, dst, args[0], args[1]);
+      break;
+   case TGSI_OPCODE_SLT:
+      emit_slt(p, dst, args[0], args[1]);
+      break;
+   case TGSI_OPCODE_SLE:
+      emit_sle(p, dst, args[0], args[1]);
+      break;
+   case TGSI_OPCODE_SUB:
+      brw_ADD(p, dst, args[0], negate(args[1]));
+      break;
+   case TGSI_OPCODE_XPD:
+      emit_xpd(p, dst, args[0], args[1]);
+      break;
+#if 0
+   case TGSI_OPCODE_IF:
+      assert(if_insn < MAX_IFSN);
+      if_inst[if_insn++] = brw_IF(p, BRW_EXECUTE_8);
+      break;
+   case TGSI_OPCODE_ELSE:
+      if_inst[if_insn-1] = brw_ELSE(p, if_inst[if_insn-1]);
+      break;
+   case TGSI_OPCODE_ENDIF:
+      assert(if_insn > 0);
+      brw_ENDIF(p, if_inst[--if_insn]);
+      break;
+   case TGSI_OPCODE_BRA:
+      brw_set_predicate_control(p, BRW_PREDICATE_NORMAL);
+      brw_ADD(p, brw_ip_reg(), brw_ip_reg(), brw_imm_d(1*16));
+      brw_set_predicate_control_flag_value(p, 0xff);
+      break;
+   case TGSI_OPCODE_CAL:
+      brw_set_access_mode(p, BRW_ALIGN_1);
+      brw_ADD(p, deref_1uw(stack_index, 0), brw_ip_reg(), brw_imm_d(3*16));
+      brw_set_access_mode(p, BRW_ALIGN_16);
+      brw_ADD(p, get_addr_reg(stack_index),
+              get_addr_reg(stack_index), brw_imm_d(4));
+      inst->Data = &p->store[p->nr_insn];
+      brw_ADD(p, brw_ip_reg(), brw_ip_reg(), brw_imm_d(1*16));
+      break;
+#endif
+   case TGSI_OPCODE_RET:
+#if 0
+      brw_ADD(p, get_addr_reg(stack_index),
+              get_addr_reg(stack_index), brw_imm_d(-4));
+      brw_set_access_mode(p, BRW_ALIGN_1);
+      brw_MOV(p, brw_ip_reg(), deref_1uw(stack_index, 0));
+      brw_set_access_mode(p, BRW_ALIGN_16);
+#else
+      /*brw_ADD(p, brw_ip_reg(), brw_ip_reg(), brw_imm_d(1*16));*/
+#endif
+      break;
+   case TGSI_OPCODE_END:
+      brw_ADD(p, brw_ip_reg(), brw_ip_reg(), brw_imm_d(1*16));
+      break;
+   case TGSI_OPCODE_BGNSUB:
+   case TGSI_OPCODE_ENDSUB:
+      break;
+   default:
+      debug_printf("Unsupport opcode %d in vertex shader\n", inst->Instruction.Opcode);
+      break;
+   }
+
+   if (dst_reg->File == TGSI_FILE_OUTPUT
+       && dst_reg->Index != info->pos_idx
+       && c->output_regs[dst_reg->Index].used_in_src)
+      brw_MOV(p, get_dst(c, dst_reg), dst);
+
+   release_tmps(c);
+}
+
+/* Emit the fragment program instructions here.
+ */
+void brw_vs_emit(struct brw_vs_compile *c)
+{
+#define MAX_IFSN 32
+   struct brw_compile *p = &c->func;
+   struct brw_instruction *end_inst;
+   struct tgsi_parse_context parse;
+   struct brw_indirect stack_index = brw_indirect(0, 0);
+   const struct tgsi_token *tokens = c->vp->program.tokens;
+   struct brw_prog_info prog_info;
+   unsigned allocated_registers = 0;
+   memset(&prog_info, 0, sizeof(struct brw_prog_info));
+
+   brw_set_compression_control(p, BRW_COMPRESSION_NONE);
+   brw_set_access_mode(p, BRW_ALIGN_16);
+
+   tgsi_parse_init(&parse, tokens);
+   /* Message registers can't be read, so copy the output into GRF register
+      if they are used in source registers */
+   while (!tgsi_parse_end_of_tokens(&parse)) {
+      tgsi_parse_token(&parse);
+      unsigned i;
+      switch (parse.FullToken.Token.Type) {
+      case TGSI_TOKEN_TYPE_INSTRUCTION: {
+         const struct tgsi_full_instruction *inst = &parse.FullToken.FullInstruction;
+         for (i = 0; i < 3; ++i) {
+            const struct tgsi_src_register *src = &inst->FullSrcRegisters[i].SrcRegister;
+            unsigned index = src->Index;
+            unsigned file = src->File;
+            if (file == TGSI_FILE_OUTPUT)
+               c->output_regs[index].used_in_src = TRUE;
+         }
+      }
+         break;
+      default:
+         /* nothing */
+         break;
+      }
+   }
+   tgsi_parse_free(&parse);
+
+   tgsi_parse_init(&parse, tokens);
+
+   while (!tgsi_parse_end_of_tokens(&parse)) {
+      tgsi_parse_token(&parse);
+
+      switch (parse.FullToken.Token.Type) {
+      case TGSI_TOKEN_TYPE_DECLARATION: {
+         struct tgsi_full_declaration *decl = &parse.FullToken.FullDeclaration;
+         process_declaration(decl, &prog_info);
+      }
+         break;
+      case TGSI_TOKEN_TYPE_IMMEDIATE: {
+         struct tgsi_full_immediate *imm = &parse.FullToken.FullImmediate;
+         /*assert(imm->Immediate.Size == 4);*/
+         c->prog_data.imm_buf[c->prog_data.num_imm][0] = imm->u.ImmediateFloat32[0].Float;
+         c->prog_data.imm_buf[c->prog_data.num_imm][1] = imm->u.ImmediateFloat32[1].Float;
+         c->prog_data.imm_buf[c->prog_data.num_imm][2] = imm->u.ImmediateFloat32[2].Float;
+         c->prog_data.imm_buf[c->prog_data.num_imm][3] = imm->u.ImmediateFloat32[3].Float;
+         c->prog_data.num_imm++;
+      }
+         break;
+      case TGSI_TOKEN_TYPE_INSTRUCTION: {
+         struct tgsi_full_instruction *inst = &parse.FullToken.FullInstruction;
+         if (!allocated_registers) {
+            /* first instruction (declerations finished).
+             * now that we know what vars are being used allocate
+             * registers for them.*/
+            c->prog_data.num_consts = prog_info.num_consts;
+            c->prog_data.max_const = prog_info.num_consts + c->prog_data.num_imm;
+            brw_vs_alloc_regs(c, &prog_info);
+
+	    brw_set_access_mode(p, BRW_ALIGN_1);
+            brw_MOV(p, get_addr_reg(stack_index), brw_address(c->stack));
+	    brw_set_access_mode(p, BRW_ALIGN_16);
+            allocated_registers = 1;
+         }
+         process_instruction(c, inst, &prog_info);
+      }
+         break;
+      }
+   }
+
+   end_inst = &p->store[p->nr_insn];
+   emit_vertex_write(c, &prog_info);
+   post_vs_emit(c, end_inst);
+   tgsi_parse_free(&parse);
+
+}
diff --git a/src/gallium/drivers/i965simple/brw_vs_state.c b/src/gallium/drivers/i965simple/brw_vs_state.c
new file mode 100644
index 0000000000..c73469929c
--- /dev/null
+++ b/src/gallium/drivers/i965simple/brw_vs_state.c
@@ -0,0 +1,102 @@
+/*
+ Copyright (C) Intel Corp.  2006.  All Rights Reserved.
+ Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to
+ develop this 3D driver.
+
+ Permission is hereby granted, free of charge, to any person obtaining
+ a copy of this software and associated documentation files (the
+ "Software"), to deal in the Software without restriction, including
+ without limitation the rights to use, copy, modify, merge, publish,
+ distribute, sublicense, and/or sell copies of the Software, and to
+ permit persons to whom the Software is furnished to do so, subject to
+ the following conditions:
+
+ The above copyright notice and this permission notice (including the
+ next paragraph) shall be included in all copies or substantial
+ portions of the Software.
+
+ THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
+ IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
+ LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
+ OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+
+ **********************************************************************/
+ /*
+  * Authors:
+  *   Keith Whitwell <keith@tungstengraphics.com>
+  */
+
+
+#include "brw_context.h"
+#include "brw_state.h"
+#include "brw_defines.h"
+
+#include "pipe/p_util.h"
+
+static void upload_vs_unit( struct brw_context *brw )
+{
+   struct brw_vs_unit_state vs;
+
+   memset(&vs, 0, sizeof(vs));
+
+   /* CACHE_NEW_VS_PROG */
+   vs.thread0.kernel_start_pointer = brw->vs.prog_gs_offset >> 6;
+   vs.thread0.grf_reg_count = align(brw->vs.prog_data->total_grf, 16) / 16 - 1;
+   vs.thread3.urb_entry_read_length = brw->vs.prog_data->urb_read_length;
+   vs.thread3.const_urb_entry_read_length = brw->vs.prog_data->curb_read_length;
+   vs.thread3.dispatch_grf_start_reg = 1;
+
+
+   /* BRW_NEW_URB_FENCE  */
+   vs.thread4.nr_urb_entries = brw->urb.nr_vs_entries;
+   vs.thread4.urb_entry_allocation_size = brw->urb.vsize - 1;
+   vs.thread4.max_threads = MIN2(
+      MAX2(0, (brw->urb.nr_vs_entries - 6) / 2 - 1),
+      15);
+
+
+
+   if (BRW_DEBUG & DEBUG_SINGLE_THREAD)
+      vs.thread4.max_threads = 0;
+
+   /* BRW_NEW_CURBE_OFFSETS, _NEW_TRANSFORM */
+   if (0 /*brw->attribs.Clip->ClipPlanesEnabled*/) {
+      /* Note that we read in the userclip planes as well, hence
+       * clip_start:
+       */
+      vs.thread3.const_urb_entry_read_offset = brw->curbe.clip_start * 2;
+   }
+   else {
+      vs.thread3.const_urb_entry_read_offset = brw->curbe.vs_start * 2;
+   }
+
+   vs.thread1.floating_point_mode = BRW_FLOATING_POINT_NON_IEEE_754;
+   vs.thread3.urb_entry_read_offset = 0;
+
+   /* No samplers for ARB_vp programs:
+    */
+   vs.vs5.sampler_count = 0;
+
+   if (BRW_DEBUG & DEBUG_STATS)
+      vs.thread4.stats_enable = 1;
+
+   /* Vertex program always enabled:
+    */
+   vs.vs6.vs_enable = 1;
+
+   brw->vs.state_gs_offset = brw_cache_data( &brw->cache[BRW_VS_UNIT], &vs );
+}
+
+
+const struct brw_tracked_state brw_vs_unit = {
+   .dirty = {
+      .brw   = (BRW_NEW_CLIP |
+		BRW_NEW_CURBE_OFFSETS |
+		BRW_NEW_URB_FENCE),
+      .cache = CACHE_NEW_VS_PROG
+   },
+   .update = upload_vs_unit
+};
diff --git a/src/gallium/drivers/i965simple/brw_winsys.h b/src/gallium/drivers/i965simple/brw_winsys.h
new file mode 100644
index 0000000000..3523a58614
--- /dev/null
+++ b/src/gallium/drivers/i965simple/brw_winsys.h
@@ -0,0 +1,205 @@
+/**************************************************************************
+ *
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ **************************************************************************/
+
+/**
+ * \file
+ * This is the interface that i965simple requires any window system
+ * hosting it to implement.  This is the only include file in i965simple
+ * which is public.
+ *
+ */
+
+#ifndef BRW_WINSYS_H
+#define BRW_WINSYS_H
+
+
+#include "pipe/p_defines.h"
+
+
+/* Pipe drivers are (meant to be!) independent of both GL and the
+ * window system.  The window system provides a buffer manager and a
+ * set of additional hooks for things like command buffer submission,
+ * etc.
+ *
+ * There clearly has to be some agreement between the window system
+ * driver and the hardware driver about the format of command buffers,
+ * etc.
+ */
+
+struct pipe_buffer;
+struct pipe_fence_handle;
+struct pipe_winsys;
+
+/* The pipe driver currently understands the following chipsets:
+ */
+#define PCI_CHIP_I965_G			0x29A2
+#define PCI_CHIP_I965_Q			0x2992
+#define PCI_CHIP_I965_G_1		0x2982
+#define PCI_CHIP_I965_GM                0x2A02
+#define PCI_CHIP_I965_GME               0x2A12
+
+
+/* These are the names of all the state caches managed by the driver.
+ * 
+ * When data is uploaded to a buffer with buffer_subdata, we use the
+ * special version of that function below so that information about
+ * what type of data this is can be passed to the winsys backend.
+ * That in turn allows the correct flags to be set in the aub file
+ * dump to allow human-readable file dumps later on.
+ */
+
+enum brw_cache_id {
+   BRW_CC_VP,
+   BRW_CC_UNIT,
+   BRW_WM_PROG,
+   BRW_SAMPLER_DEFAULT_COLOR,
+   BRW_SAMPLER,
+   BRW_WM_UNIT,
+   BRW_SF_PROG,
+   BRW_SF_VP,
+   BRW_SF_UNIT,
+   BRW_VS_UNIT,
+   BRW_VS_PROG,
+   BRW_GS_UNIT,
+   BRW_GS_PROG,
+   BRW_CLIP_VP,
+   BRW_CLIP_UNIT,
+   BRW_CLIP_PROG,
+   BRW_SS_SURFACE,
+   BRW_SS_SURF_BIND,
+
+   BRW_MAX_CACHE
+};
+
+#define BRW_CONSTANT_BUFFER BRW_MAX_CACHE
+
+/**
+ * Additional winsys interface for i965simple.
+ *
+ * It is an over-simple batchbuffer mechanism.  Will want to improve the
+ * performance of this, perhaps based on the cmdstream stuff.  It
+ * would be pretty impossible to implement swz on top of this
+ * interface.
+ *
+ * Will also need additions/changes to implement static/dynamic
+ * indirect state.
+ */
+struct brw_winsys {
+
+   /**
+    * Reserve space on batch buffer.
+    *
+    * Returns a null pointer if there is insufficient space in the batch buffer
+    * to hold the requested number of dwords and relocations.
+    *
+    * The number of dwords should also include the number of relocations.
+    */
+   unsigned *(*batch_start)(struct brw_winsys *sws,
+                            unsigned dwords,
+                            unsigned relocs);
+
+   void (*batch_dword)(struct brw_winsys *sws,
+                       unsigned dword);
+
+   /**
+    * Emit a relocation to a buffer.
+    *
+    * Used not only when the buffer addresses are not pinned, but also to
+    * ensure refered buffers will not be destroyed until the current batch
+    * buffer execution is finished.
+    *
+    * The access flags is a combination of I915_BUFFER_ACCESS_WRITE and
+    * I915_BUFFER_ACCESS_READ macros.
+    */
+   void (*batch_reloc)(struct brw_winsys *sws,
+                       struct pipe_buffer *buf,
+                       unsigned access_flags,
+                       unsigned delta);
+
+
+   /* Not used yet, but really want this:
+    */
+   void (*batch_end)( struct brw_winsys *sws );
+
+   /**
+    * Flush the batch buffer.
+    *
+    * Fence argument must point to NULL or to a previous fence, and the caller
+    * must call fence_reference when done with the fence.
+    */
+   void (*batch_flush)(struct brw_winsys *sws,
+                       struct pipe_fence_handle **fence);
+
+
+   /* A version of buffer_subdata that includes information for the
+    * simulator:
+    */
+   void (*buffer_subdata_typed)(struct brw_winsys *sws, 
+				struct pipe_buffer *buf,
+				unsigned long offset, 
+				unsigned long size, 
+				const void *data,
+				unsigned data_type);
+   
+
+   /* A cheat so we don't have to think about relocations in a couple
+    * of places yet:
+    */
+   unsigned (*get_buffer_offset)( struct brw_winsys *sws,
+				  struct pipe_buffer *buf,
+				  unsigned flags );
+
+};
+
+#define BRW_BUFFER_ACCESS_WRITE   0x1
+#define BRW_BUFFER_ACCESS_READ    0x2
+
+#define BRW_BUFFER_USAGE_LIT_VERTEX  (PIPE_BUFFER_USAGE_CUSTOM << 0)
+
+
+struct pipe_context *brw_create(struct pipe_winsys *,
+                                struct brw_winsys *,
+                                unsigned pci_id);
+
+static inline boolean brw_batchbuffer_data(struct brw_winsys *winsys,
+                                           const void *data,
+                                           unsigned bytes)
+{
+   static const unsigned incr = sizeof(unsigned);
+   uint i;
+   const unsigned *udata = (const unsigned*)(data);
+   unsigned size = bytes/incr;
+
+   winsys->batch_start(winsys, size, 0);
+   for (i = 0; i < size; ++i) {
+      winsys->batch_dword(winsys, udata[i]);
+   }
+   winsys->batch_end(winsys);
+
+   return (i == size);
+}
+#endif
diff --git a/src/gallium/drivers/i965simple/brw_wm.c b/src/gallium/drivers/i965simple/brw_wm.c
new file mode 100644
index 0000000000..539b170744
--- /dev/null
+++ b/src/gallium/drivers/i965simple/brw_wm.c
@@ -0,0 +1,210 @@
+/*
+ Copyright (C) Intel Corp.  2006.  All Rights Reserved.
+ Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to
+ develop this 3D driver.
+
+ Permission is hereby granted, free of charge, to any person obtaining
+ a copy of this software and associated documentation files (the
+ "Software"), to deal in the Software without restriction, including
+ without limitation the rights to use, copy, modify, merge, publish,
+ distribute, sublicense, and/or sell copies of the Software, and to
+ permit persons to whom the Software is furnished to do so, subject to
+ the following conditions:
+
+ The above copyright notice and this permission notice (including the
+ next paragraph) shall be included in all copies or substantial
+ portions of the Software.
+
+ THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
+ IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
+ LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
+ OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+
+ **********************************************************************/
+ /*
+  * Authors:
+  *   Keith Whitwell <keith@tungstengraphics.com>
+  */
+
+
+#include "brw_context.h"
+#include "brw_util.h"
+#include "brw_wm.h"
+#include "brw_eu.h"
+#include "brw_state.h"
+#include "pipe/p_util.h"
+
+
+
+static void do_wm_prog( struct brw_context *brw,
+			struct brw_fragment_program *fp,
+			struct brw_wm_prog_key *key)
+{
+   struct brw_wm_compile *c = CALLOC_STRUCT(brw_wm_compile);
+   const unsigned *program;
+   unsigned program_size;
+
+   c->key = *key;
+   c->fp = fp;
+   
+   c->delta_xy[0] = brw_null_reg();
+   c->delta_xy[1] = brw_null_reg();
+   c->pixel_xy[0] = brw_null_reg();
+   c->pixel_xy[1] = brw_null_reg();
+   c->pixel_w = brw_null_reg();
+
+
+   debug_printf("XXXXXXXX FP\n");
+   
+   brw_wm_glsl_emit(c);
+
+   /* get the program
+    */
+   program = brw_get_program(&c->func, &program_size);
+
+   /*
+    */
+   brw->wm.prog_gs_offset = brw_upload_cache( &brw->cache[BRW_WM_PROG],
+					      &c->key,
+					      sizeof(c->key),
+					      program,
+					      program_size,
+					      &c->prog_data,
+					      &brw->wm.prog_data );
+
+   FREE(c);
+}
+
+
+
+static void brw_wm_populate_key( struct brw_context *brw,
+				 struct brw_wm_prog_key *key )
+{
+   /* BRW_NEW_FRAGMENT_PROGRAM */
+   struct brw_fragment_program *fp =
+      (struct brw_fragment_program *)brw->attribs.FragmentProgram;
+   unsigned lookup = 0;
+   unsigned line_aa;
+   
+   memset(key, 0, sizeof(*key));
+
+   /* Build the index for table lookup
+    */
+   /* BRW_NEW_DEPTH_STENCIL */
+   if (fp->UsesKill ||
+       brw->attribs.DepthStencil->alpha.enabled)
+      lookup |= IZ_PS_KILL_ALPHATEST_BIT;
+
+   if (fp->ComputesDepth)
+      lookup |= IZ_PS_COMPUTES_DEPTH_BIT;
+
+   if (brw->attribs.DepthStencil->depth.enabled)
+      lookup |= IZ_DEPTH_TEST_ENABLE_BIT;
+
+   if (brw->attribs.DepthStencil->depth.enabled &&
+       brw->attribs.DepthStencil->depth.writemask) /* ?? */
+      lookup |= IZ_DEPTH_WRITE_ENABLE_BIT;
+
+   if (brw->attribs.DepthStencil->stencil[0].enabled) {
+      lookup |= IZ_STENCIL_TEST_ENABLE_BIT;
+
+      if (brw->attribs.DepthStencil->stencil[0].write_mask ||
+	  brw->attribs.DepthStencil->stencil[1].write_mask)
+	 lookup |= IZ_STENCIL_WRITE_ENABLE_BIT;
+   }
+
+   /* XXX: when should this be disabled?
+    */
+   if (1)
+      lookup |= IZ_EARLY_DEPTH_TEST_BIT;
+
+
+   line_aa = AA_NEVER;
+
+   /* _NEW_LINE, _NEW_POLYGON, BRW_NEW_REDUCED_PRIMITIVE */
+   if (brw->attribs.Raster->line_smooth) {
+      if (brw->reduced_primitive == PIPE_PRIM_LINES) {
+	 line_aa = AA_ALWAYS;
+      }
+      else if (brw->reduced_primitive == PIPE_PRIM_TRIANGLES) {
+	 if (brw->attribs.Raster->fill_ccw == PIPE_POLYGON_MODE_LINE) {
+	    line_aa = AA_SOMETIMES;
+
+	    if (brw->attribs.Raster->fill_cw == PIPE_POLYGON_MODE_LINE ||
+		(brw->attribs.Raster->cull_mode == PIPE_WINDING_CW))
+	       line_aa = AA_ALWAYS;
+	 }
+	 else if (brw->attribs.Raster->fill_cw == PIPE_POLYGON_MODE_LINE) {
+	    line_aa = AA_SOMETIMES;
+
+	    if (brw->attribs.Raster->cull_mode == PIPE_WINDING_CCW)
+	       line_aa = AA_ALWAYS;
+	 }
+      }
+   }
+
+   brw_wm_lookup_iz(line_aa,
+		    lookup,
+		    key);
+
+
+#if 0
+   /* BRW_NEW_SAMPLER 
+    *
+    * Not doing any of this at the moment:
+    */
+   for (i = 0; i < BRW_MAX_TEX_UNIT; i++) {
+      const struct pipe_sampler_state *unit = brw->attribs.Samplers[i];
+
+      if (unit) {
+
+	 if (unit->compare &&
+             unit->compare_mode == PIPE_TEX_COMPARE_R_TO_TEXTURE) {
+	    key->shadowtex_mask |= 1<<i;
+	 }
+	 if (t->Image[0][t->BaseLevel]->InternalFormat == GL_YCBCR_MESA)
+	    key->yuvtex_mask |= 1<<i;
+      }
+   }
+#endif
+
+
+   /* Extra info:
+    */
+   key->program_string_id = fp->id;
+
+}
+
+
+static void brw_upload_wm_prog( struct brw_context *brw )
+{
+   struct brw_wm_prog_key key;
+   struct brw_fragment_program *fp = (struct brw_fragment_program *)
+      brw->attribs.FragmentProgram;
+
+   brw_wm_populate_key(brw, &key);
+
+   /* Make an early check for the key.
+    */
+   if (brw_search_cache(&brw->cache[BRW_WM_PROG],
+			&key, sizeof(key),
+			&brw->wm.prog_data,
+			&brw->wm.prog_gs_offset))
+      return;
+
+   do_wm_prog(brw, fp, &key);
+}
+
+
+const struct brw_tracked_state brw_wm_prog = {
+   .dirty = {
+      .brw   = (BRW_NEW_FS |
+		BRW_NEW_REDUCED_PRIMITIVE),
+      .cache = 0
+   },
+   .update = brw_upload_wm_prog
+};
+
diff --git a/src/gallium/drivers/i965simple/brw_wm.h b/src/gallium/drivers/i965simple/brw_wm.h
new file mode 100644
index 0000000000..a1ac0f504a
--- /dev/null
+++ b/src/gallium/drivers/i965simple/brw_wm.h
@@ -0,0 +1,142 @@
+/*
+ Copyright (C) Intel Corp.  2006.  All Rights Reserved.
+ Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to
+ develop this 3D driver.
+ 
+ Permission is hereby granted, free of charge, to any person obtaining
+ a copy of this software and associated documentation files (the
+ "Software"), to deal in the Software without restriction, including
+ without limitation the rights to use, copy, modify, merge, publish,
+ distribute, sublicense, and/or sell copies of the Software, and to
+ permit persons to whom the Software is furnished to do so, subject to
+ the following conditions:
+ 
+ The above copyright notice and this permission notice (including the
+ next paragraph) shall be included in all copies or substantial
+ portions of the Software.
+ 
+ THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
+ IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
+ LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
+ OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ 
+ **********************************************************************/
+ /*
+  * Authors:
+  *   Keith Whitwell <keith@tungstengraphics.com>
+  */
+              
+
+#ifndef BRW_WM_H
+#define BRW_WM_H
+
+
+#include "brw_context.h"
+#include "brw_eu.h"
+
+/* A big lookup table is used to figure out which and how many
+ * additional regs will inserted before the main payload in the WM
+ * program execution.  These mainly relate to depth and stencil
+ * processing and the early-depth-test optimization.
+ */
+#define IZ_PS_KILL_ALPHATEST_BIT    0x1
+#define IZ_PS_COMPUTES_DEPTH_BIT    0x2
+#define IZ_DEPTH_WRITE_ENABLE_BIT   0x4
+#define IZ_DEPTH_TEST_ENABLE_BIT    0x8
+#define IZ_STENCIL_WRITE_ENABLE_BIT 0x10
+#define IZ_STENCIL_TEST_ENABLE_BIT  0x20
+#define IZ_EARLY_DEPTH_TEST_BIT     0x40
+#define IZ_BIT_MAX                  0x80
+
+#define AA_NEVER     0
+#define AA_SOMETIMES 1
+#define AA_ALWAYS    2
+
+struct brw_wm_prog_key {
+   unsigned source_depth_reg:3;
+   unsigned aa_dest_stencil_reg:3;
+   unsigned dest_depth_reg:3;
+   unsigned nr_depth_regs:3;
+   unsigned shadowtex_mask:8;
+   unsigned computes_depth:1;	/* could be derived from program string */
+   unsigned source_depth_to_render_target:1;
+   unsigned runtime_check_aads_emit:1;
+
+   unsigned yuvtex_mask:8;
+
+   unsigned program_string_id;
+};
+
+
+
+
+
+#define PROGRAM_INTERNAL_PARAM
+#define MAX_NV_FRAGMENT_PROGRAM_INSTRUCTIONS 1024 /* 72 for GL_ARB_f_p */
+#define BRW_WM_MAX_INSN  (MAX_NV_FRAGMENT_PROGRAM_INSTRUCTIONS*3 + PIPE_ATTRIB_MAX + 3)
+#define BRW_WM_MAX_GRF   128		/* hardware limit */
+#define BRW_WM_MAX_VREG  (BRW_WM_MAX_INSN * 4)
+#define BRW_WM_MAX_REF   (BRW_WM_MAX_INSN * 12)
+#define BRW_WM_MAX_PARAM 256
+#define BRW_WM_MAX_CONST 256
+#define BRW_WM_MAX_KILLS MAX_NV_FRAGMENT_PROGRAM_INSTRUCTIONS
+
+#define PAYLOAD_DEPTH     (PIPE_ATTRIB_MAX)
+
+#define MAX_IFSN 32
+#define MAX_LOOP_DEPTH 32
+
+struct brw_wm_compile {
+   struct brw_compile func;
+   struct brw_wm_prog_key key;
+   struct brw_wm_prog_data prog_data; /* result */
+
+   struct brw_fragment_program *fp;
+
+   unsigned grf_limit;
+   unsigned max_wm_grf;
+
+
+   struct brw_reg pixel_xy[2];
+   struct brw_reg delta_xy[2];
+   struct brw_reg pixel_w;
+
+
+   struct brw_reg wm_regs[8][32][4];
+
+   struct brw_reg payload_depth[4];
+   struct brw_reg payload_coef[16];
+
+   struct brw_reg emit_mask_reg;
+
+   struct brw_instruction *if_inst[MAX_IFSN];
+   int if_insn;
+
+   struct brw_instruction *loop_inst[MAX_LOOP_DEPTH];
+   int loop_insn;
+
+   struct brw_instruction *inst0;
+   struct brw_instruction *inst1;
+
+   struct brw_reg stack;
+   struct brw_indirect stack_index;
+
+   unsigned reg_index;
+
+   unsigned tmp_start;
+   unsigned tmp_index;
+};
+
+
+
+void brw_wm_lookup_iz( unsigned line_aa,
+		       unsigned lookup,
+		       struct brw_wm_prog_key *key );
+
+void brw_wm_glsl_emit(struct brw_wm_compile *c);
+void brw_wm_emit_decls(struct brw_wm_compile *c);
+
+#endif
diff --git a/src/gallium/drivers/i965simple/brw_wm_decl.c b/src/gallium/drivers/i965simple/brw_wm_decl.c
new file mode 100644
index 0000000000..b45a333a2e
--- /dev/null
+++ b/src/gallium/drivers/i965simple/brw_wm_decl.c
@@ -0,0 +1,383 @@
+
+#include "brw_context.h"
+#include "brw_eu.h"
+#include "brw_wm.h"
+#include "pipe/p_util.h"
+#include "pipe/p_shader_tokens.h"
+#include "pipe/tgsi/util/tgsi_parse.h"
+
+static struct brw_reg alloc_tmp(struct brw_wm_compile *c)
+{
+   c->tmp_index++;
+   c->reg_index = MAX2(c->reg_index, c->tmp_start + c->tmp_index);
+   return brw_vec8_grf(c->tmp_start + c->tmp_index, 0);
+}
+
+static void release_tmps(struct brw_wm_compile *c)
+{
+   c->tmp_index = 0;
+}
+
+
+
+static int is_null( struct brw_reg reg )
+{
+   return (reg.file == BRW_ARCHITECTURE_REGISTER_FILE &&
+	   reg.nr == BRW_ARF_NULL);
+}
+
+static void emit_pixel_xy( struct brw_wm_compile *c )
+{
+   if (is_null(c->pixel_xy[0])) {
+
+      struct brw_compile *p = &c->func;
+      struct brw_reg r1_uw = retype(brw_vec1_grf(1, 0), BRW_REGISTER_TYPE_UW);
+
+      c->pixel_xy[0] = vec8(retype(alloc_tmp(c), BRW_REGISTER_TYPE_UW));
+      c->pixel_xy[1] = vec8(retype(alloc_tmp(c), BRW_REGISTER_TYPE_UW));
+
+      /* Calculate pixel centers by adding 1 or 0 to each of the
+       * micro-tile coordinates passed in r1.
+       */
+      brw_ADD(p,
+	      c->pixel_xy[0],
+	      stride(suboffset(r1_uw, 4), 2, 4, 0),
+	      brw_imm_v(0x10101010));
+
+      brw_ADD(p,
+	      c->pixel_xy[1],
+	      stride(suboffset(r1_uw, 5), 2, 4, 0),
+	      brw_imm_v(0x11001100));
+   }
+}
+
+
+
+
+
+
+static void emit_delta_xy( struct brw_wm_compile *c )
+{
+   if (is_null(c->delta_xy[0])) {
+      struct brw_compile *p = &c->func;
+      struct brw_reg r1 = brw_vec1_grf(1, 0);
+
+      emit_pixel_xy(c);
+
+      c->delta_xy[0] = alloc_tmp(c);
+      c->delta_xy[1] = alloc_tmp(c);
+
+      /* Calc delta X,Y by subtracting origin in r1 from the pixel
+       * centers.
+       */
+      brw_ADD(p,
+	      c->delta_xy[0],
+	      retype(c->pixel_xy[0], BRW_REGISTER_TYPE_UW),
+	      negate(r1));
+
+      brw_ADD(p,
+	      c->delta_xy[1],
+	      retype(c->pixel_xy[1], BRW_REGISTER_TYPE_UW),
+	      negate(suboffset(r1,1)));
+   }
+}
+
+
+
+#if 0
+static void emit_pixel_w( struct brw_wm_compile *c )
+{
+   if (is_null(c->pixel_w)) {
+      struct brw_compile *p = &c->func;
+
+      struct brw_reg interp_wpos = c->coef_wpos;
+      
+      c->pixel_w = alloc_tmp(c);
+
+      emit_delta_xy(c);
+
+      /* Calc 1/w - just linterp wpos[3] optimized by putting the
+       * result straight into a message reg.
+       */
+      struct brw_reg interp3 = brw_vec1_grf(interp_wpos.nr+1, 4);
+      brw_LINE(p, brw_null_reg(), interp3, c->delta_xy[0]);
+      brw_MAC(p, brw_message_reg(2), suboffset(interp3, 1), c->delta_xy[1]);
+
+      /* Calc w */
+      brw_math_16( p, 
+		   c->pixel_w,
+		   BRW_MATH_FUNCTION_INV,
+		   BRW_MATH_SATURATE_NONE,
+		   2, 
+		   brw_null_reg(),
+		   BRW_MATH_PRECISION_FULL);
+   }
+}
+#endif
+
+
+static void emit_cinterp(struct brw_wm_compile *c,
+			 int idx,
+			 int mask )
+{
+   struct brw_compile *p = &c->func;
+   struct brw_reg interp[4];
+   struct brw_reg coef = c->payload_coef[idx];
+   int i;
+
+   interp[0] = brw_vec1_grf(coef.nr, 0);
+   interp[1] = brw_vec1_grf(coef.nr, 4);
+   interp[2] = brw_vec1_grf(coef.nr+1, 0);
+   interp[3] = brw_vec1_grf(coef.nr+1, 4);
+
+   for(i = 0; i < 4; i++ ) {
+      if (mask & (1<<i)) {
+	 struct brw_reg dst = c->wm_regs[TGSI_FILE_INPUT][idx][i];
+	 brw_MOV(p, dst, suboffset(interp[i],3));
+      }
+   }
+}
+
+static void emit_linterp(struct brw_wm_compile *c,
+			 int idx,
+			 int mask )
+{
+   struct brw_compile *p = &c->func;
+   struct brw_reg interp[4];
+   struct brw_reg coef = c->payload_coef[idx];
+   int i;
+
+   emit_delta_xy(c);
+
+   interp[0] = brw_vec1_grf(coef.nr, 0);
+   interp[1] = brw_vec1_grf(coef.nr, 4);
+   interp[2] = brw_vec1_grf(coef.nr+1, 0);
+   interp[3] = brw_vec1_grf(coef.nr+1, 4);
+
+   for(i = 0; i < 4; i++ ) {
+      if (mask & (1<<i)) {
+	 struct brw_reg dst = c->wm_regs[TGSI_FILE_INPUT][idx][i];
+	 brw_LINE(p, brw_null_reg(), interp[i], c->delta_xy[0]);
+	 brw_MAC(p, dst, suboffset(interp[i],1), c->delta_xy[1]);
+      }
+   }
+}
+
+#if 0
+static void emit_pinterp(struct brw_wm_compile *c,
+			 int idx,
+			 int mask )
+{
+   struct brw_compile *p = &c->func;
+   struct brw_reg interp[4];
+   struct brw_reg coef = c->payload_coef[idx];
+   int i;
+
+   get_delta_xy(c);
+   get_pixel_w(c);
+
+   interp[0] = brw_vec1_grf(coef.nr, 0);
+   interp[1] = brw_vec1_grf(coef.nr, 4);
+   interp[2] = brw_vec1_grf(coef.nr+1, 0);
+   interp[3] = brw_vec1_grf(coef.nr+1, 4);
+
+   for(i = 0; i < 4; i++ ) {
+      if (mask & (1<<i)) {
+	 struct brw_reg dst = allocate_reg(c, TGSI_FILE_INPUT, idx, i);
+	 brw_LINE(p, brw_null_reg(), interp[i], c->delta_xy[0]);
+	 brw_MAC(p, dst, suboffset(interp[i],1), c->delta_xy[1]);
+	 brw_MUL(p, dst, dst, c->pixel_w);
+      }
+   }
+}
+#endif
+
+
+
+#if 0
+static void emit_wpos( )
+{ 
+   struct prog_dst_register dst = dst_reg(PROGRAM_INPUT, idx);
+   struct tgsi_full_src_register interp = src_reg(PROGRAM_PAYLOAD, idx);
+   struct tgsi_full_src_register deltas = get_delta_xy(c);
+   struct tgsi_full_src_register arg2;
+   unsigned opcode;
+
+   opcode = WM_LINTERP;
+   arg2 = src_undef();
+
+   /* Have to treat wpos.xy specially:
+    */
+   emit_op(c,
+	   WM_WPOSXY,
+	   dst_mask(dst, WRITEMASK_XY),
+	   0, 0, 0,
+	   get_pixel_xy(c),
+	   src_undef(),
+	   src_undef());
+      
+   dst = dst_mask(dst, WRITEMASK_ZW);
+
+   /* PROGRAM_INPUT.attr.xyzw = INTERP payload.interp[attr].x, deltas.xyw
+    */
+   emit_op(c,
+	   WM_LINTERP,
+	   dst,
+	   0, 0, 0,
+	   interp,
+	   deltas,
+	   arg2);
+}
+#endif
+
+
+
+
+/* Perform register allocation:
+ * 
+ *  -- r0???
+ *  -- passthrough depth regs (and stencil/aa??)
+ *  -- curbe ??
+ *  -- inputs (coefficients)
+ *
+ * Use a totally static register allocation.  This will perform poorly
+ * but is an easy way to get started (again).
+ */
+static void prealloc_reg(struct brw_wm_compile *c)
+{
+   int i, j;
+   int nr_curbe_regs = 0;
+
+   /* R0, then some depth related regs:
+    */
+   for (i = 0; i < c->key.nr_depth_regs; i++) {
+      c->payload_depth[i] =  brw_vec8_grf(i*2, 0);
+      c->reg_index += 2;
+   }
+
+
+   /* Then a copy of our part of the CURBE entry:
+    */
+   {
+      int nr_constants = c->fp->info.nr_regs[TGSI_FILE_CONSTANT];
+      int index = 0;
+
+      c->prog_data.max_const = 4*nr_constants;
+      for (i = 0; i < nr_constants; i++) {
+	 for (j = 0; j < 4; j++, index++) 
+	    c->wm_regs[TGSI_FILE_CONSTANT][i][j] = brw_vec1_grf(c->reg_index + index/8,
+								index%8);
+      }
+
+      nr_curbe_regs = 2*((4*nr_constants+15)/16);
+      c->reg_index += nr_curbe_regs;
+   }
+
+   /* Adjust for parameter coefficients for position, which are
+    * currently always provided.
+    */
+//   c->position_coef[i] = brw_vec8_grf(c->reg_index, 0);
+   c->reg_index += 2;
+
+   /* Next we receive the plane coefficients for parameter
+    * interpolation:
+    */
+   for (i = 0; i < c->fp->info.nr_regs[TGSI_FILE_INPUT]; i++) {
+      c->payload_coef[i] = brw_vec8_grf(c->reg_index, 0);
+      c->reg_index += 2;
+   }
+
+   c->prog_data.first_curbe_grf = c->key.nr_depth_regs * 2;
+   c->prog_data.urb_read_length = (c->fp->program.num_inputs + 1) * 2;
+   c->prog_data.curb_read_length = nr_curbe_regs;
+
+   /* That's the end of the payload, now we can start allocating registers.
+    */
+   c->emit_mask_reg = brw_uw1_reg(BRW_GENERAL_REGISTER_FILE, c->reg_index, 0);
+   c->reg_index++;
+
+   c->stack = brw_uw16_reg(BRW_GENERAL_REGISTER_FILE, c->reg_index, 0);
+   c->reg_index += 2;
+
+   /* Now allocate room for the interpolated inputs and staging
+    * registers for the outputs:
+    */
+   for (i = 0; i < c->fp->info.nr_regs[TGSI_FILE_INPUT]; i++) 
+      for (j = 0; j < 4; j++)
+	 c->wm_regs[TGSI_FILE_INPUT][i][j] = brw_vec8_grf( c->reg_index++, 0 );
+
+   for (i = 0; i < c->fp->info.nr_regs[TGSI_FILE_OUTPUT]; i++) 
+      for (j = 0; j < 4; j++)
+	 c->wm_regs[TGSI_FILE_OUTPUT][i][j] = brw_vec8_grf( c->reg_index++, 0 );
+
+   /* Beyond this we should only need registers for internal temporaries:
+    */
+   c->tmp_start = c->reg_index;
+}
+
+
+
+
+
+/* Need to interpolate fragment program inputs in as a preamble to the
+ * shader.  A more sophisticated compiler would do this on demand, but
+ * we'll do it up front:
+ */
+void brw_wm_emit_decls(struct brw_wm_compile *c)
+{
+   struct tgsi_parse_context parse;
+   int done = 0;
+
+   prealloc_reg(c);
+
+   tgsi_parse_init( &parse, c->fp->program.tokens );
+
+   while( !done &&
+	  !tgsi_parse_end_of_tokens( &parse ) ) 
+   {
+      tgsi_parse_token( &parse );
+
+      switch( parse.FullToken.Token.Type ) {
+      case TGSI_TOKEN_TYPE_DECLARATION:
+      {
+	 const struct tgsi_full_declaration *decl = &parse.FullToken.FullDeclaration;
+	 unsigned first = decl->u.DeclarationRange.First;
+	 unsigned last = decl->u.DeclarationRange.Last;
+	 unsigned mask = decl->Declaration.UsageMask; /* ? */
+	 unsigned i;
+
+	 if (decl->Declaration.File != TGSI_FILE_INPUT)
+	    break;
+
+	 assert(decl->Declaration.Interpolate);
+
+	 for( i = first; i <= last; i++ ) {
+	    switch (decl->Interpolation.Interpolate) {
+	    case TGSI_INTERPOLATE_CONSTANT:
+	       emit_cinterp(c, i, mask);
+	       break;
+
+	    case TGSI_INTERPOLATE_LINEAR:
+	       emit_linterp(c, i, mask);
+	       break;
+
+	    case TGSI_INTERPOLATE_PERSPECTIVE:
+	       //emit_pinterp(c, i, mask);
+	       emit_linterp(c, i, mask);
+	       break;
+	    }
+	 }
+	 break;
+      }
+      case TGSI_TOKEN_TYPE_IMMEDIATE:
+      case TGSI_TOKEN_TYPE_INSTRUCTION:
+      default:
+         done = 1;
+	 break;
+      }
+   }
+
+   tgsi_parse_free (&parse);
+   
+   release_tmps(c);
+}
diff --git a/src/gallium/drivers/i965simple/brw_wm_glsl.c b/src/gallium/drivers/i965simple/brw_wm_glsl.c
new file mode 100644
index 0000000000..d95645d108
--- /dev/null
+++ b/src/gallium/drivers/i965simple/brw_wm_glsl.c
@@ -0,0 +1,1079 @@
+
+#include "brw_context.h"
+#include "brw_eu.h"
+#include "brw_wm.h"
+#include "pipe/p_util.h"
+#include "pipe/p_shader_tokens.h"
+#include "pipe/tgsi/util/tgsi_parse.h"
+
+
+
+static int get_scalar_dst_index(struct tgsi_full_instruction *inst)
+{
+   struct tgsi_dst_register dst = inst->FullDstRegisters[0].DstRegister;
+   int i;
+   for (i = 0; i < 4; i++)
+      if (dst.WriteMask & (1<<i))
+	 break;
+   return i;
+}
+
+static struct brw_reg alloc_tmp(struct brw_wm_compile *c)
+{
+   c->tmp_index++;
+   c->reg_index = MAX2(c->reg_index, c->tmp_index);
+   return brw_vec8_grf(c->tmp_start + c->tmp_index, 0);
+}
+
+static void release_tmps(struct brw_wm_compile *c)
+{
+   c->tmp_index = 0;
+}
+
+
+static struct brw_reg
+get_reg(struct brw_wm_compile *c, int file, int index, int component )
+{
+   switch (file) {
+   case TGSI_FILE_NULL:
+      return brw_null_reg();
+
+   case TGSI_FILE_SAMPLER:
+      /* Should never get here:
+       */
+      assert (0);	       
+      return brw_null_reg();
+
+   case TGSI_FILE_IMMEDIATE:
+      /* These need a different path:
+       */
+      assert(0);
+      return brw_null_reg();
+
+       
+   case TGSI_FILE_CONSTANT:
+   case TGSI_FILE_INPUT:
+   case TGSI_FILE_OUTPUT:
+   case TGSI_FILE_TEMPORARY:
+   case TGSI_FILE_ADDRESS:
+      return c->wm_regs[file][index][component];
+
+   default:
+      assert(0);
+      return brw_null_reg();
+   }
+}
+
+
+static struct brw_reg get_dst_reg(struct brw_wm_compile *c,
+				  struct tgsi_full_instruction *inst, 
+				  int component)
+{
+   return get_reg(c, 
+		  inst->FullDstRegisters[0].DstRegister.File, 
+		  inst->FullDstRegisters[0].DstRegister.Index,
+		  component);
+}
+
+static int get_swz( struct tgsi_src_register src, int index )
+{
+   switch (index & 3) {
+   case 0: return src.SwizzleX;
+   case 1: return src.SwizzleY;
+   case 2: return src.SwizzleZ;
+   case 3: return src.SwizzleW;
+   default: return 0;
+   }
+}
+
+static int get_ext_swz( struct tgsi_src_register_ext_swz src, int index )
+{
+   switch (index & 3) {
+   case 0: return src.ExtSwizzleX;
+   case 1: return src.ExtSwizzleY;
+   case 2: return src.ExtSwizzleZ;
+   case 3: return src.ExtSwizzleW;
+   default: return 0;
+   }
+}
+
+static struct brw_reg get_src_reg(struct brw_wm_compile *c,
+				  struct tgsi_full_src_register *src, 
+				  int index)
+{
+   struct brw_reg reg;
+   int component = index;
+   int neg = 0;
+   int abs = 0;
+
+   if (src->SrcRegister.Negate)
+      neg = 1;
+
+   component = get_swz(src->SrcRegister, component);
+
+   /* Yes, there are multiple negates:
+    */
+   switch (component & 3) {
+   case 0: neg ^= src->SrcRegisterExtSwz.NegateX; break;
+   case 1: neg ^= src->SrcRegisterExtSwz.NegateY; break;
+   case 2: neg ^= src->SrcRegisterExtSwz.NegateZ; break;
+   case 3: neg ^= src->SrcRegisterExtSwz.NegateW; break;
+   }
+
+   /* And multiple swizzles, fun isn't it:
+    */
+   component = get_ext_swz(src->SrcRegisterExtSwz, component);
+
+   /* Can't handle this, don't know if we need to:
+    */
+   assert(src->SrcRegisterExtSwz.ExtDivide == TGSI_EXTSWIZZLE_ONE);
+       
+   /* Not handling indirect lookups yet:
+    */
+   assert(src->SrcRegister.Indirect == 0);
+
+   /* Don't know what dimension means:
+    */
+   assert(src->SrcRegister.Dimension == 0);
+
+   /* Will never handle any of this stuff: 
+    */
+   assert(src->SrcRegisterExtMod.Complement == 0);
+   assert(src->SrcRegisterExtMod.Bias == 0);
+   assert(src->SrcRegisterExtMod.Scale2X == 0);
+
+   if (src->SrcRegisterExtMod.Absolute)
+      abs = 1;
+
+   /* Another negate!  This is a post-absolute negate, which we
+    * can't do.  Need to clean the crap out of tgsi somehow.
+    */
+   assert(src->SrcRegisterExtMod.Negate == 0);
+
+   switch( component ) {
+   case TGSI_EXTSWIZZLE_X:
+   case TGSI_EXTSWIZZLE_Y:
+   case TGSI_EXTSWIZZLE_Z:
+   case TGSI_EXTSWIZZLE_W:
+      reg = get_reg(c, 
+		    src->SrcRegister.File, 
+		    src->SrcRegister.Index, 
+		    component );
+
+      if (neg) 
+	 reg = negate(reg);
+   
+      if (abs)
+	 reg = brw_abs(reg);
+
+      break;
+
+      /* XXX: this won't really work in the general case, but we know
+       * that the extended swizzle is only allowed in the SWZ
+       * instruction (right??), in which case using an immediate
+       * directly will work.
+       */
+   case TGSI_EXTSWIZZLE_ZERO:
+      reg = brw_imm_f(0);
+      break;
+
+   case TGSI_EXTSWIZZLE_ONE:
+      if (neg && !abs)
+	 reg = brw_imm_f(-1.0);
+      else
+	 reg = brw_imm_f(1.0);
+      break;
+
+   default:
+      assert(0);
+      break;
+   }
+
+    
+   return reg;
+}
+
+static void emit_abs( struct brw_wm_compile *c,
+		      struct tgsi_full_instruction *inst)
+{
+   unsigned mask = inst->FullDstRegisters[0].DstRegister.WriteMask;
+
+   int i;
+   struct brw_compile *p = &c->func;
+   brw_set_saturate(p, inst->Instruction.Saturate != TGSI_SAT_NONE);
+   for (i = 0; i < 4; i++) {
+      if (mask & (1<<i)) {
+	 struct brw_reg src, dst;
+	 dst = get_dst_reg(c, inst, i);
+	 src = get_src_reg(c, &inst->FullSrcRegisters[0], i);
+	 brw_MOV(p, dst, brw_abs(src)); /* NOTE */
+      }
+   }
+   brw_set_saturate(p, 0);
+}
+
+
+static void emit_xpd(struct brw_wm_compile *c,
+		     struct tgsi_full_instruction *inst)
+{
+   int i;
+   struct brw_compile *p = &c->func;
+   unsigned mask = inst->FullDstRegisters[0].DstRegister.WriteMask;
+   for (i = 0; i < 4; i++) {
+      unsigned i2 = (i+2)%3;
+      unsigned i1 = (i+1)%3;
+      if (mask & (1<<i)) {
+	 struct brw_reg src0, src1, dst;
+	 dst = get_dst_reg(c, inst, i);
+	 src0 = negate(get_src_reg(c, &inst->FullSrcRegisters[0], i2));
+	 src1 = get_src_reg(c, &inst->FullSrcRegisters[1], i1);
+	 brw_MUL(p, brw_null_reg(), src0, src1);
+	 src0 = get_src_reg(c, &inst->FullSrcRegisters[0], i1);
+	 src1 = get_src_reg(c, &inst->FullSrcRegisters[1], i2);
+	 brw_set_saturate(p, inst->Instruction.Saturate != TGSI_SAT_NONE);
+	 brw_MAC(p, dst, src0, src1);
+	 brw_set_saturate(p, 0);
+      }
+   }
+   brw_set_saturate(p, 0);
+}
+
+static void emit_dp3(struct brw_wm_compile *c,
+		     struct tgsi_full_instruction *inst)
+{
+   struct brw_reg src0[3], src1[3], dst;
+   int i;
+   struct brw_compile *p = &c->func;
+   for (i = 0; i < 3; i++) {
+      src0[i] = get_src_reg(c, &inst->FullSrcRegisters[0], i);
+      src1[i] = get_src_reg(c, &inst->FullSrcRegisters[1], i);
+   }
+
+   dst = get_dst_reg(c, inst, get_scalar_dst_index(inst));
+   brw_MUL(p, brw_null_reg(), src0[0], src1[0]);
+   brw_MAC(p, brw_null_reg(), src0[1], src1[1]);
+   brw_set_saturate(p, (inst->Instruction.Saturate != TGSI_SAT_NONE) ? 1 : 0);
+   brw_MAC(p, dst, src0[2], src1[2]);
+   brw_set_saturate(p, 0);
+}
+
+static void emit_dp4(struct brw_wm_compile *c,
+		     struct tgsi_full_instruction *inst)
+{
+   struct brw_reg src0[4], src1[4], dst;
+   int i;
+   struct brw_compile *p = &c->func;
+   for (i = 0; i < 4; i++) {
+      src0[i] = get_src_reg(c, &inst->FullSrcRegisters[0], i);
+      src1[i] = get_src_reg(c, &inst->FullSrcRegisters[1], i);
+   }
+   dst = get_dst_reg(c, inst, get_scalar_dst_index(inst));
+   brw_MUL(p, brw_null_reg(), src0[0], src1[0]);
+   brw_MAC(p, brw_null_reg(), src0[1], src1[1]);
+   brw_MAC(p, brw_null_reg(), src0[2], src1[2]);
+   brw_set_saturate(p, (inst->Instruction.Saturate != TGSI_SAT_NONE) ? 1 : 0);
+   brw_MAC(p, dst, src0[3], src1[3]);
+   brw_set_saturate(p, 0);
+}
+
+static void emit_dph(struct brw_wm_compile *c,
+		     struct tgsi_full_instruction *inst)
+{
+   struct brw_reg src0[4], src1[4], dst;
+   int i;
+   struct brw_compile *p = &c->func;
+   for (i = 0; i < 4; i++) {
+      src0[i] = get_src_reg(c, &inst->FullSrcRegisters[0], i);
+      src1[i] = get_src_reg(c, &inst->FullSrcRegisters[1], i);
+   }
+   dst = get_dst_reg(c, inst, get_scalar_dst_index(inst));
+   brw_MUL(p, brw_null_reg(), src0[0], src1[0]);
+   brw_MAC(p, brw_null_reg(), src0[1], src1[1]);
+   brw_MAC(p, dst, src0[2], src1[2]);
+   brw_set_saturate(p, (inst->Instruction.Saturate != TGSI_SAT_NONE) ? 1 : 0);
+   brw_ADD(p, dst, src0[3], src1[3]);
+   brw_set_saturate(p, 0);
+}
+
+static void emit_math1(struct brw_wm_compile *c,
+		       struct tgsi_full_instruction *inst, unsigned func)
+{
+   struct brw_compile *p = &c->func;
+   struct brw_reg src0, dst;
+
+   src0 = get_src_reg(c, &inst->FullSrcRegisters[0], 0);
+   dst = get_dst_reg(c, inst, get_scalar_dst_index(inst));
+   brw_MOV(p, brw_message_reg(2), src0);
+   brw_math(p,
+	    dst,
+	    func,
+	    ((inst->Instruction.Saturate != TGSI_SAT_NONE) 
+	     ? BRW_MATH_SATURATE_SATURATE 
+	     : BRW_MATH_SATURATE_NONE),
+	    2,
+	    brw_null_reg(),
+	    BRW_MATH_DATA_VECTOR,
+	    BRW_MATH_PRECISION_FULL);
+}
+
+
+static void emit_alu2(struct brw_wm_compile *c,		      
+		      struct tgsi_full_instruction *inst,
+		      unsigned opcode)
+{
+   struct brw_compile *p = &c->func;
+   struct brw_reg src0, src1, dst;
+   unsigned mask = inst->FullDstRegisters[0].DstRegister.WriteMask;
+   int i;
+   brw_set_saturate(p, (inst->Instruction.Saturate != TGSI_SAT_NONE) ? 1 : 0);
+   for (i = 0 ; i < 4; i++) {
+      if (mask & (1<<i)) {
+	 dst = get_dst_reg(c, inst, i);
+	 src0 = get_src_reg(c, &inst->FullSrcRegisters[0], i);
+	 src1 = get_src_reg(c, &inst->FullSrcRegisters[1], i);
+	 brw_alu2(p, opcode, dst, src0, src1);
+      }
+   }
+   brw_set_saturate(p, 0);
+}
+
+
+static void emit_alu1(struct brw_wm_compile *c,
+		      struct tgsi_full_instruction *inst,
+		      unsigned opcode)
+{
+   struct brw_compile *p = &c->func;
+   struct brw_reg src0, dst;
+   unsigned mask = inst->FullDstRegisters[0].DstRegister.WriteMask;
+   int i;
+   brw_set_saturate(p, (inst->Instruction.Saturate != TGSI_SAT_NONE) ? 1 : 0);
+   for (i = 0 ; i < 4; i++) {
+      if (mask & (1<<i)) {
+	 dst = get_dst_reg(c, inst, i);
+	 src0 = get_src_reg(c, &inst->FullSrcRegisters[0], i);
+	 brw_alu1(p, opcode, dst, src0);
+      }
+   }
+   if (inst->Instruction.Saturate != TGSI_SAT_NONE)
+      brw_set_saturate(p, 0);
+}
+
+
+static void emit_max(struct brw_wm_compile *c,
+		     struct tgsi_full_instruction *inst)
+{
+   struct brw_compile *p = &c->func;
+   unsigned mask = inst->FullDstRegisters[0].DstRegister.WriteMask;
+   struct brw_reg src0, src1, dst;
+   int i;
+   brw_push_insn_state(p);
+   for (i = 0; i < 4; i++) {
+      if (mask & (1<<i)) {
+	 dst = get_dst_reg(c, inst, i);
+	 src0 = get_src_reg(c, &inst->FullSrcRegisters[0], i);
+	 src1 = get_src_reg(c, &inst->FullSrcRegisters[1], i);
+	 brw_set_saturate(p, (inst->Instruction.Saturate != TGSI_SAT_NONE) ? 1 : 0);
+	 brw_MOV(p, dst, src0);
+	 brw_set_saturate(p, 0);
+
+	 brw_CMP(p, brw_null_reg(), BRW_CONDITIONAL_L, src0, src1);
+	 brw_set_saturate(p, (inst->Instruction.Saturate != TGSI_SAT_NONE) ? 1 : 0);
+	 brw_set_predicate_control(p, BRW_PREDICATE_NORMAL);
+	 brw_MOV(p, dst, src1);
+	 brw_set_saturate(p, 0);
+	 brw_set_predicate_control_flag_value(p, 0xff);
+      }
+   }
+   brw_pop_insn_state(p);
+}
+
+static void emit_min(struct brw_wm_compile *c,
+		     struct tgsi_full_instruction *inst)
+{
+   struct brw_compile *p = &c->func;
+   unsigned mask = inst->FullDstRegisters[0].DstRegister.WriteMask;
+   struct brw_reg src0, src1, dst;
+   int i;
+   brw_push_insn_state(p);
+   for (i = 0; i < 4; i++) {
+      if (mask & (1<<i)) {
+	 dst = get_dst_reg(c, inst, i);
+	 src0 = get_src_reg(c, &inst->FullSrcRegisters[0], i);
+	 src1 = get_src_reg(c, &inst->FullSrcRegisters[1], i);
+	 brw_set_saturate(p, (inst->Instruction.Saturate != TGSI_SAT_NONE) ? 1 : 0);
+	 brw_MOV(p, dst, src0);
+	 brw_set_saturate(p, 0);
+
+	 brw_CMP(p, brw_null_reg(), BRW_CONDITIONAL_L, src1, src0);
+	 brw_set_saturate(p, (inst->Instruction.Saturate != TGSI_SAT_NONE) ? 1 : 0);
+	 brw_set_predicate_control(p, BRW_PREDICATE_NORMAL);
+	 brw_MOV(p, dst, src1);
+	 brw_set_saturate(p, 0);
+	 brw_set_predicate_control_flag_value(p, 0xff);
+      }
+   }
+   brw_pop_insn_state(p);
+}
+
+static void emit_pow(struct brw_wm_compile *c,
+		     struct tgsi_full_instruction *inst)
+{
+   struct brw_compile *p = &c->func;
+   struct brw_reg dst, src0, src1;
+   dst = get_dst_reg(c, inst, get_scalar_dst_index(inst));
+   src0 = get_src_reg(c, &inst->FullSrcRegisters[0], 0);
+   src1 = get_src_reg(c, &inst->FullSrcRegisters[1], 0);
+
+   brw_MOV(p, brw_message_reg(2), src0);
+   brw_MOV(p, brw_message_reg(3), src1);
+
+   brw_math(p,
+	    dst,
+	    BRW_MATH_FUNCTION_POW,
+	    (inst->Instruction.Saturate != TGSI_SAT_NONE 
+	     ? BRW_MATH_SATURATE_SATURATE 
+	     : BRW_MATH_SATURATE_NONE),
+	    2,
+	    brw_null_reg(),
+	    BRW_MATH_DATA_VECTOR,
+	    BRW_MATH_PRECISION_FULL);
+}
+
+static void emit_lrp(struct brw_wm_compile *c,
+		     struct tgsi_full_instruction *inst)
+{
+   struct brw_compile *p = &c->func;
+   unsigned mask = inst->FullDstRegisters[0].DstRegister.WriteMask;
+   struct brw_reg dst, tmp1, tmp2, src0, src1, src2;
+   int i;
+   for (i = 0; i < 4; i++) {
+      if (mask & (1<<i)) {
+	 dst = get_dst_reg(c, inst, i);
+	 src0 = get_src_reg(c, &inst->FullSrcRegisters[0], i);
+
+	 src1 = get_src_reg(c, &inst->FullSrcRegisters[1], i);
+
+	 if (src1.nr == dst.nr) {
+	    tmp1 = alloc_tmp(c);
+	    brw_MOV(p, tmp1, src1);
+	 } else
+	    tmp1 = src1;
+
+	 src2 = get_src_reg(c, &inst->FullSrcRegisters[2], i);
+	 if (src2.nr == dst.nr) {
+	    tmp2 = alloc_tmp(c);
+	    brw_MOV(p, tmp2, src2);
+	 } else
+	    tmp2 = src2;
+
+	 brw_ADD(p, dst, negate(src0), brw_imm_f(1.0));
+	 brw_MUL(p, brw_null_reg(), dst, tmp2);
+	 brw_set_saturate(p, (inst->Instruction.Saturate != TGSI_SAT_NONE) ? 1 : 0);
+	 brw_MAC(p, dst, src0, tmp1);
+	 brw_set_saturate(p, 0);
+      }
+      release_tmps(c);
+   }
+}
+
+static void emit_kil(struct brw_wm_compile *c)
+{
+   struct brw_compile *p = &c->func;
+   struct brw_reg depth = retype(brw_vec1_grf(0, 0), BRW_REGISTER_TYPE_UW);
+   brw_push_insn_state(p);
+   brw_set_mask_control(p, BRW_MASK_DISABLE);
+   brw_NOT(p, c->emit_mask_reg, brw_mask_reg(1)); //IMASK
+   brw_AND(p, depth, c->emit_mask_reg, depth);
+   brw_pop_insn_state(p);
+}
+
+static void emit_mad(struct brw_wm_compile *c,
+		     struct tgsi_full_instruction *inst)
+{
+   struct brw_compile *p = &c->func;
+   unsigned mask = inst->FullDstRegisters[0].DstRegister.WriteMask;
+   struct brw_reg dst, src0, src1, src2;
+   int i;
+
+   for (i = 0; i < 4; i++) {
+      if (mask & (1<<i)) {
+	 dst = get_dst_reg(c, inst, i);
+	 src0 = get_src_reg(c, &inst->FullSrcRegisters[0], i);
+	 src1 = get_src_reg(c, &inst->FullSrcRegisters[1], i);
+	 src2 = get_src_reg(c, &inst->FullSrcRegisters[2], i);
+	 brw_MUL(p, dst, src0, src1);
+
+	 brw_set_saturate(p, (inst->Instruction.Saturate != TGSI_SAT_NONE) ? 1 : 0);
+	 brw_ADD(p, dst, dst, src2);
+	 brw_set_saturate(p, 0);
+      }
+   }
+}
+
+static void emit_sop(struct brw_wm_compile *c,
+		     struct tgsi_full_instruction *inst, unsigned cond)
+{
+   struct brw_compile *p = &c->func;
+   unsigned mask = inst->FullDstRegisters[0].DstRegister.WriteMask;
+   struct brw_reg dst, src0, src1;
+   int i;
+
+   brw_push_insn_state(p);
+   for (i = 0; i < 4; i++) {
+      if (mask & (1<<i)) {
+	 dst = get_dst_reg(c, inst, i);
+	 src0 = get_src_reg(c, &inst->FullSrcRegisters[0], i);
+	 src1 = get_src_reg(c, &inst->FullSrcRegisters[1], i);
+	 brw_CMP(p, brw_null_reg(), cond, src0, src1);
+	 brw_set_predicate_control(p, BRW_PREDICATE_NONE);
+	 brw_MOV(p, dst, brw_imm_f(0.0));
+	 brw_set_predicate_control(p, BRW_PREDICATE_NORMAL);
+	 brw_MOV(p, dst, brw_imm_f(1.0));
+      }
+   }
+   brw_pop_insn_state(p);
+}
+
+
+static void emit_ddx(struct brw_wm_compile *c,
+		     struct tgsi_full_instruction *inst)
+{
+   struct brw_compile *p = &c->func;
+   unsigned mask = inst->FullDstRegisters[0].DstRegister.WriteMask;
+   struct brw_reg interp[4];
+   struct brw_reg dst;
+   struct brw_reg src0, w;
+   unsigned nr, i;
+   src0 = get_src_reg(c, &inst->FullSrcRegisters[0], 0);
+   w = get_src_reg(c, &inst->FullSrcRegisters[1], 3);
+   nr = src0.nr;
+   interp[0] = brw_vec1_grf(nr, 0);
+   interp[1] = brw_vec1_grf(nr, 4);
+   interp[2] = brw_vec1_grf(nr+1, 0);
+   interp[3] = brw_vec1_grf(nr+1, 4);
+   brw_set_saturate(p, inst->Instruction.Saturate != TGSI_SAT_NONE);
+   for(i = 0; i < 4; i++ ) {
+      if (mask & (1<<i)) {
+	 dst = get_dst_reg(c, inst, i);
+	 brw_MOV(p, dst, interp[i]);
+	 brw_MUL(p, dst, dst, w);
+      }
+   }
+   brw_set_saturate(p, 0);
+}
+
+static void emit_ddy(struct brw_wm_compile *c,
+		     struct tgsi_full_instruction *inst)
+{
+   struct brw_compile *p = &c->func;
+   unsigned mask = inst->FullDstRegisters[0].DstRegister.WriteMask;
+   struct brw_reg interp[4];
+   struct brw_reg dst;
+   struct brw_reg src0, w;
+   unsigned nr, i;
+
+   src0 = get_src_reg(c, &inst->FullSrcRegisters[0], 0);
+   nr = src0.nr;
+   w = get_src_reg(c, &inst->FullSrcRegisters[1], 3);
+   interp[0] = brw_vec1_grf(nr, 0);
+   interp[1] = brw_vec1_grf(nr, 4);
+   interp[2] = brw_vec1_grf(nr+1, 0);
+   interp[3] = brw_vec1_grf(nr+1, 4);
+   brw_set_saturate(p, inst->Instruction.Saturate != TGSI_SAT_NONE);
+   for(i = 0; i < 4; i++ ) {
+      if (mask & (1<<i)) {
+	 dst = get_dst_reg(c, inst, i);
+	 brw_MOV(p, dst, suboffset(interp[i], 1));
+	 brw_MUL(p, dst, dst, w);
+      }
+   }
+   brw_set_saturate(p, 0);
+}
+
+/* TODO
+   BIAS on SIMD8 not workind yet...
+*/
+static void emit_txb(struct brw_wm_compile *c,
+		     struct tgsi_full_instruction *inst)
+{
+#if 0
+   struct brw_compile *p = &c->func;
+   struct brw_reg payload_reg = c->payload_depth[0];
+   struct brw_reg dst[4], src[4];
+   unsigned i;
+   for (i = 0; i < 4; i++)
+      dst[i] = get_dst_reg(c, inst, i);
+   for (i = 0; i < 4; i++)
+      src[i] = get_src_reg(c, &inst->FullSrcRegisters[0], i);
+
+#if 0
+   switch (inst->TexSrcTarget) {
+   case TEXTURE_1D_INDEX:
+      brw_MOV(p, brw_message_reg(2), src[0]);
+      brw_MOV(p, brw_message_reg(3), brw_imm_f(0));
+      brw_MOV(p, brw_message_reg(4), brw_imm_f(0));
+      break;
+   case TEXTURE_2D_INDEX:
+   case TEXTURE_RECT_INDEX:
+      brw_MOV(p, brw_message_reg(2), src[0]);
+      brw_MOV(p, brw_message_reg(3), src[1]);
+      brw_MOV(p, brw_message_reg(4), brw_imm_f(0));
+      break;
+   default:
+      brw_MOV(p, brw_message_reg(2), src[0]);
+      brw_MOV(p, brw_message_reg(3), src[1]);
+      brw_MOV(p, brw_message_reg(4), src[2]);
+      break;
+   }
+#else
+   brw_MOV(p, brw_message_reg(2), src[0]);
+   brw_MOV(p, brw_message_reg(3), src[1]);
+   brw_MOV(p, brw_message_reg(4), brw_imm_f(0));
+#endif
+
+   brw_MOV(p, brw_message_reg(5), src[3]);
+   brw_MOV(p, brw_message_reg(6), brw_imm_f(0));
+   brw_SAMPLE(p,
+	      retype(vec8(dst[0]), BRW_REGISTER_TYPE_UW),
+	      1,
+	      retype(payload_reg, BRW_REGISTER_TYPE_UW),
+	      inst->TexSrcUnit + 1, /* surface */
+	      inst->TexSrcUnit,     /* sampler */
+	      inst->FullDstRegisters[0].DstRegister.WriteMask,
+	      BRW_SAMPLER_MESSAGE_SIMD16_SAMPLE_BIAS,
+	      4,
+	      4,
+	      0);
+#endif
+}
+
+static void emit_tex(struct brw_wm_compile *c,
+		     struct tgsi_full_instruction *inst)
+{
+#if 0
+   struct brw_compile *p = &c->func;
+   struct brw_reg payload_reg = c->payload_depth[0];
+   struct brw_reg dst[4], src[4];
+   unsigned msg_len;
+   unsigned i, nr;
+   unsigned emit;
+   boolean shadow = (c->key.shadowtex_mask & (1<<inst->TexSrcUnit)) ? 1 : 0;
+
+   for (i = 0; i < 4; i++)
+      dst[i] = get_dst_reg(c, inst, i);
+   for (i = 0; i < 4; i++)
+      src[i] = get_src_reg(c, &inst->FullSrcRegisters[0], i);
+
+#if 0
+   switch (inst->TexSrcTarget) {
+   case TEXTURE_1D_INDEX:
+      emit = WRITEMASK_X;
+      nr = 1;
+      break;
+   case TEXTURE_2D_INDEX:
+   case TEXTURE_RECT_INDEX:
+      emit = WRITEMASK_XY;
+      nr = 2;
+      break;
+   default:
+      emit = WRITEMASK_XYZ;
+      nr = 3;
+      break;
+   }
+#else
+   emit = WRITEMASK_XY;
+   nr = 2;
+#endif
+
+   msg_len = 1;
+
+   for (i = 0; i < nr; i++) {
+      static const unsigned swz[4] = {0,1,2,2};
+      if (emit & (1<<i))
+	 brw_MOV(p, brw_message_reg(msg_len+1), src[swz[i]]);
+      else
+	 brw_MOV(p, brw_message_reg(msg_len+1), brw_imm_f(0));
+      msg_len += 1;
+   }
+
+   if (shadow) {
+      brw_MOV(p, brw_message_reg(5), brw_imm_f(0));
+      brw_MOV(p, brw_message_reg(6), src[2]);
+   }
+
+   brw_SAMPLE(p,
+	      retype(vec8(dst[0]), BRW_REGISTER_TYPE_UW),
+	      1,
+	      retype(payload_reg, BRW_REGISTER_TYPE_UW),
+	      inst->TexSrcUnit + 1, /* surface */
+	      inst->TexSrcUnit,     /* sampler */
+	      inst->FullDstRegisters[0].DstRegister.WriteMask,
+	      BRW_SAMPLER_MESSAGE_SIMD8_SAMPLE,
+	      4,
+	      shadow ? 6 : 4,
+	      0);
+
+   if (shadow)
+      brw_MOV(p, dst[3], brw_imm_f(1.0));
+#endif
+}
+
+
+
+
+
+
+
+
+static void emit_fb_write(struct brw_wm_compile *c,
+			  struct tgsi_full_instruction *inst)
+{
+   struct brw_compile *p = &c->func;
+   int nr = 2;
+   int channel;
+   int base_reg = 0;
+
+   // src0 = output color
+   // src1 = payload_depth[0]
+   // src2 = output depth
+   // dst = ???
+
+
+
+   /* Reserve a space for AA - may not be needed:
+    */
+   if (c->key.aa_dest_stencil_reg)
+      nr += 1;
+
+   {
+      brw_push_insn_state(p);
+      for (channel = 0; channel < 4; channel++) {
+	 struct brw_reg src0 = c->wm_regs[TGSI_FILE_OUTPUT][0][channel];
+
+	 /*  mov (8) m2.0<1>:ud   r28.0<8;8,1>:ud  { Align1 } */
+	 /*  mov (8) m6.0<1>:ud   r29.0<8;8,1>:ud  { Align1 SecHalf } */
+	 brw_MOV(p, brw_message_reg(nr + channel), src0);
+      }
+      /* skip over the regs populated above: */
+      nr += 8;
+      brw_pop_insn_state(p);
+   }
+    
+
+   /* Pass through control information:
+    */
+   /*  mov (8) m1.0<1>:ud   r1.0<8;8,1>:ud   { Align1 NoMask } */
+   {
+      brw_push_insn_state(p);
+      brw_set_mask_control(p, BRW_MASK_DISABLE); /* ? */
+      brw_MOV(p,
+	      brw_message_reg(base_reg + 1),
+	      brw_vec8_grf(1, 0));
+      brw_pop_insn_state(p);
+   }
+
+   /* Send framebuffer write message: */
+   brw_fb_WRITE(p,
+		retype(vec8(brw_null_reg()), BRW_REGISTER_TYPE_UW),
+		base_reg,
+		retype(brw_vec8_grf(0, 0), BRW_REGISTER_TYPE_UW),
+		0,              /* render surface always 0 */
+		nr,
+		0,
+		1);
+
+}
+
+
+static void brw_wm_emit_instruction( struct brw_wm_compile *c,
+				     struct tgsi_full_instruction *inst )
+{
+   struct brw_compile *p = &c->func;
+
+#if 0   
+   if (inst->CondUpdate)
+      brw_set_conditionalmod(p, BRW_CONDITIONAL_NZ);
+   else
+      brw_set_conditionalmod(p, BRW_CONDITIONAL_NONE);
+#else
+   brw_set_conditionalmod(p, BRW_CONDITIONAL_NONE);
+#endif
+
+   switch (inst->Instruction.Opcode) {
+   case TGSI_OPCODE_ABS:
+      emit_abs(c, inst);
+      break;
+   case TGSI_OPCODE_ADD:
+      emit_alu2(c, inst, BRW_OPCODE_ADD);
+      break;
+   case TGSI_OPCODE_SUB:
+      assert(0);
+//      emit_alu2(c, inst, BRW_OPCODE_SUB);
+      break;
+   case TGSI_OPCODE_FRC:
+      emit_alu1(c, inst, BRW_OPCODE_FRC);
+      break;
+   case TGSI_OPCODE_FLR:
+      assert(0);
+//      emit_alu1(c, inst, BRW_OPCODE_FLR);
+      break;
+   case TGSI_OPCODE_LRP:
+      emit_lrp(c, inst);
+      break;
+   case TGSI_OPCODE_INT:
+      emit_alu1(c, inst, BRW_OPCODE_RNDD);
+      break;
+   case TGSI_OPCODE_MOV:
+      emit_alu1(c, inst, BRW_OPCODE_MOV);
+      break;
+   case TGSI_OPCODE_DP3:
+      emit_dp3(c, inst);
+      break;
+   case TGSI_OPCODE_DP4:
+      emit_dp4(c, inst);
+      break;
+   case TGSI_OPCODE_XPD:
+      emit_xpd(c, inst);
+      break;
+   case TGSI_OPCODE_DPH:
+      emit_dph(c, inst);
+      break;
+   case TGSI_OPCODE_RCP:
+      emit_math1(c, inst, BRW_MATH_FUNCTION_INV);
+      break;
+   case TGSI_OPCODE_RSQ:
+      emit_math1(c, inst, BRW_MATH_FUNCTION_RSQ);
+      break;
+   case TGSI_OPCODE_SIN:
+      emit_math1(c, inst, BRW_MATH_FUNCTION_SIN);
+      break;
+   case TGSI_OPCODE_COS:
+      emit_math1(c, inst, BRW_MATH_FUNCTION_COS);
+      break;
+   case TGSI_OPCODE_EX2:
+      emit_math1(c, inst, BRW_MATH_FUNCTION_EXP);
+      break;
+   case TGSI_OPCODE_LG2:
+      emit_math1(c, inst, BRW_MATH_FUNCTION_LOG);
+      break;
+   case TGSI_OPCODE_MAX:
+      emit_max(c, inst);
+      break;
+   case TGSI_OPCODE_MIN:
+      emit_min(c, inst);
+      break;
+   case TGSI_OPCODE_DDX:
+      emit_ddx(c, inst);
+      break;
+   case TGSI_OPCODE_DDY:
+      emit_ddy(c, inst);
+      break;
+   case TGSI_OPCODE_SLT:
+      emit_sop(c, inst, BRW_CONDITIONAL_L);
+      break;
+   case TGSI_OPCODE_SLE:
+      emit_sop(c, inst, BRW_CONDITIONAL_LE);
+      break;
+   case TGSI_OPCODE_SGT:
+      emit_sop(c, inst, BRW_CONDITIONAL_G);
+      break;
+   case TGSI_OPCODE_SGE:
+      emit_sop(c, inst, BRW_CONDITIONAL_GE);
+      break;
+   case TGSI_OPCODE_SEQ:
+      emit_sop(c, inst, BRW_CONDITIONAL_EQ);
+      break;
+   case TGSI_OPCODE_SNE:
+      emit_sop(c, inst, BRW_CONDITIONAL_NEQ);
+      break;
+   case TGSI_OPCODE_MUL:
+      emit_alu2(c, inst, BRW_OPCODE_MUL);
+      break;
+   case TGSI_OPCODE_POW:
+      emit_pow(c, inst);
+      break;
+   case TGSI_OPCODE_MAD:
+      emit_mad(c, inst);
+      break;
+   case TGSI_OPCODE_TEX:
+      emit_tex(c, inst);
+      break;
+   case TGSI_OPCODE_TXB:
+      emit_txb(c, inst);
+      break;
+   case TGSI_OPCODE_TEXKILL:
+      emit_kil(c);
+      break;
+   case TGSI_OPCODE_IF:
+      assert(c->if_insn < MAX_IFSN);
+      c->if_inst[c->if_insn++] = brw_IF(p, BRW_EXECUTE_8);
+      break;
+   case TGSI_OPCODE_ELSE:
+      c->if_inst[c->if_insn-1]  = brw_ELSE(p, c->if_inst[c->if_insn-1]);
+      break;
+   case TGSI_OPCODE_ENDIF:
+      assert(c->if_insn > 0);
+      brw_ENDIF(p, c->if_inst[--c->if_insn]);
+      break;
+   case TGSI_OPCODE_BGNSUB:
+   case TGSI_OPCODE_ENDSUB:
+      break;
+   case TGSI_OPCODE_CAL:
+      brw_push_insn_state(p);
+      brw_set_mask_control(p, BRW_MASK_DISABLE);
+      brw_set_access_mode(p, BRW_ALIGN_1);
+      brw_ADD(p, deref_1ud(c->stack_index, 0), brw_ip_reg(), brw_imm_d(3*16));
+      brw_set_access_mode(p, BRW_ALIGN_16);
+      brw_ADD(p, 
+	      get_addr_reg(c->stack_index),
+	      get_addr_reg(c->stack_index), brw_imm_d(4));
+//      orig_inst = inst->Data;
+//      orig_inst->Data = &p->store[p->nr_insn];
+      assert(0);
+      brw_ADD(p, brw_ip_reg(), brw_ip_reg(), brw_imm_d(1*16));
+      brw_pop_insn_state(p);
+      break;
+
+   case TGSI_OPCODE_RET:
+#if 0
+      brw_push_insn_state(p);
+      brw_set_mask_control(p, BRW_MASK_DISABLE);
+      brw_ADD(p, 
+	      get_addr_reg(c->stack_index),
+	      get_addr_reg(c->stack_index), brw_imm_d(-4));
+      brw_set_access_mode(p, BRW_ALIGN_1);
+      brw_MOV(p, brw_ip_reg(), deref_1ud(c->stack_index, 0));
+      brw_set_access_mode(p, BRW_ALIGN_16);
+      brw_pop_insn_state(p);
+#else
+      emit_fb_write(c, inst);
+#endif
+
+      break;
+   case TGSI_OPCODE_LOOP:
+      c->loop_inst[c->loop_insn++] = brw_DO(p, BRW_EXECUTE_8);
+      break;
+   case TGSI_OPCODE_BRK:
+      brw_BREAK(p);
+      brw_set_predicate_control(p, BRW_PREDICATE_NONE);
+      break;
+   case TGSI_OPCODE_CONT:
+      brw_CONT(p);
+      brw_set_predicate_control(p, BRW_PREDICATE_NONE);
+      break;
+   case TGSI_OPCODE_ENDLOOP:
+      c->loop_insn--;
+      c->inst0 = c->inst1 = brw_WHILE(p, c->loop_inst[c->loop_insn]);
+      /* patch all the BREAK instructions from
+	 last BEGINLOOP */
+      while (c->inst0 > c->loop_inst[c->loop_insn]) {
+	 c->inst0--;
+	 if (c->inst0->header.opcode == BRW_OPCODE_BREAK) {
+	    c->inst0->bits3.if_else.jump_count = c->inst1 - c->inst0 + 1;
+	    c->inst0->bits3.if_else.pop_count = 0;
+	 } else if (c->inst0->header.opcode == BRW_OPCODE_CONTINUE) {
+	    c->inst0->bits3.if_else.jump_count = c->inst1 - c->inst0;
+	    c->inst0->bits3.if_else.pop_count = 0;
+	 }
+      }
+      break;
+   case TGSI_OPCODE_END:
+      emit_fb_write(c, inst);
+      break;
+
+   default:
+      debug_printf("unsupported IR in fragment shader %d\n",
+		   inst->Instruction.Opcode);
+   }
+#if 0
+   if (inst->CondUpdate)
+      brw_set_predicate_control(p, BRW_PREDICATE_NORMAL);
+   else
+      brw_set_predicate_control(p, BRW_PREDICATE_NONE);
+#endif
+}
+
+
+
+
+
+
+void brw_wm_glsl_emit(struct brw_wm_compile *c)
+{
+   struct tgsi_parse_context parse;
+   struct brw_compile *p = &c->func;
+
+   brw_init_compile(&c->func);
+   brw_set_compression_control(p, BRW_COMPRESSION_NONE);
+
+   c->reg_index = 0;
+   c->if_insn = 0;
+   c->loop_insn = 0;
+   c->stack_index = brw_indirect(0,0);
+
+   /* Do static register allocation and parameter interpolation:
+    */
+   brw_wm_emit_decls( c );
+
+   /* Emit the actual program.  All done with very direct translation,
+    * hopefully we can improve on this shortly...
+    */
+   brw_MOV(p, get_addr_reg(c->stack_index), brw_address(c->stack));
+
+   tgsi_parse_init( &parse, c->fp->program.tokens );
+
+   while( !tgsi_parse_end_of_tokens( &parse ) ) 
+   {
+      tgsi_parse_token( &parse );
+
+      switch( parse.FullToken.Token.Type ) {
+      case TGSI_TOKEN_TYPE_DECLARATION:
+	 /* already done */
+	 break;
+
+      case TGSI_TOKEN_TYPE_IMMEDIATE:
+         /* not handled yet */
+	 assert(0);
+         break;
+
+      case TGSI_TOKEN_TYPE_INSTRUCTION:
+         brw_wm_emit_instruction(c, &parse.FullToken.FullInstruction);
+         break;
+
+      default:
+         assert( 0 );
+      }
+   }
+
+   tgsi_parse_free (&parse);
+   
+   /* Fix up call targets:
+    */
+#if 0
+   {
+      unsigned nr_insns = c->fp->program.Base.NumInstructions;
+      unsigned insn, target_insn;
+      struct tgsi_full_instruction *inst1, *inst2;
+      struct brw_instruction *brw_inst1, *brw_inst2;
+      int offset;
+      for (insn = 0; insn < nr_insns; insn++) {
+	 inst1 = &c->fp->program.Base.Instructions[insn];
+	 brw_inst1 = inst1->Data;
+	 switch (inst1->Opcode) {
+	 case TGSI_OPCODE_CAL:
+	    target_insn = inst1->BranchTarget;
+	    inst2 = &c->fp->program.Base.Instructions[target_insn];
+	    brw_inst2 = inst2->Data;
+	    offset = brw_inst2 - brw_inst1;
+	    brw_set_src1(brw_inst1, brw_imm_d(offset*16));
+	    break;
+	 default:
+	    break;
+	 }
+      }
+   }
+#endif
+
+   c->prog_data.total_grf = c->reg_index;
+   c->prog_data.total_scratch = 0;
+}
diff --git a/src/gallium/drivers/i965simple/brw_wm_iz.c b/src/gallium/drivers/i965simple/brw_wm_iz.c
new file mode 100644
index 0000000000..6c5f25bf39
--- /dev/null
+++ b/src/gallium/drivers/i965simple/brw_wm_iz.c
@@ -0,0 +1,214 @@
+/*
+ Copyright (C) Intel Corp.  2006.  All Rights Reserved.
+ Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to
+ develop this 3D driver.
+ 
+ Permission is hereby granted, free of charge, to any person obtaining
+ a copy of this software and associated documentation files (the
+ "Software"), to deal in the Software without restriction, including
+ without limitation the rights to use, copy, modify, merge, publish,
+ distribute, sublicense, and/or sell copies of the Software, and to
+ permit persons to whom the Software is furnished to do so, subject to
+ the following conditions:
+ 
+ The above copyright notice and this permission notice (including the
+ next paragraph) shall be included in all copies or substantial
+ portions of the Software.
+ 
+ THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
+ IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
+ LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
+ OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ 
+ **********************************************************************/
+ /*
+  * Authors:
+  *   Keith Whitwell <keith@tungstengraphics.com>
+  */
+
+#include "brw_wm.h"
+
+
+#undef P			/* prompted depth */
+#undef C			/* computed */
+#undef N			/* non-promoted? */
+
+#define P 0
+#define C 1
+#define N 2
+
+const struct {
+   unsigned mode:2;
+   unsigned sd_present:1;
+   unsigned sd_to_rt:1;
+   unsigned dd_present:1;
+   unsigned ds_present:1;
+} wm_iz_table[IZ_BIT_MAX] =
+{
+ { P, 0, 0, 0, 0 }, 
+ { P, 0, 0, 0, 0 }, 
+ { C, 0, 1, 0, 0 }, 
+ { C, 0, 1, 0, 0 }, 
+ { C, 1, 1, 0, 0 }, 
+ { C, 1, 1, 0, 0 }, 
+ { C, 0, 1, 0, 0 }, 
+ { C, 0, 1, 0, 0 }, 
+ { C, 1, 1, 1, 0 }, 
+ { C, 1, 1, 1, 0 }, 
+ { C, 0, 1, 1, 0 }, 
+ { C, 0, 1, 1, 0 }, 
+ { C, 1, 1, 1, 0 }, 
+ { C, 1, 1, 1, 0 }, 
+ { C, 0, 1, 1, 0 }, 
+ { C, 0, 1, 1, 0 }, 
+ { P, 0, 0, 0, 0 }, 
+ { P, 0, 0, 0, 0 }, 
+ { C, 0, 1, 0, 0 }, 
+ { C, 0, 1, 0, 0 }, 
+ { C, 1, 1, 0, 0 }, 
+ { C, 1, 1, 0, 0 }, 
+ { C, 0, 1, 0, 0 }, 
+ { C, 0, 1, 0, 0 }, 
+ { C, 1, 1, 1, 0 }, 
+ { C, 1, 1, 1, 0 }, 
+ { C, 0, 1, 1, 0 }, 
+ { C, 0, 1, 1, 0 }, 
+ { C, 1, 1, 1, 0 }, 
+ { C, 1, 1, 1, 0 }, 
+ { C, 0, 1, 1, 0 }, 
+ { C, 0, 1, 1, 0 }, 
+ { C, 0, 0, 0, 1 }, 
+ { C, 0, 0, 0, 1 }, 
+ { C, 0, 1, 0, 1 }, 
+ { C, 0, 1, 0, 1 }, 
+ { C, 1, 1, 0, 1 }, 
+ { C, 1, 1, 0, 1 }, 
+ { C, 0, 1, 0, 1 }, 
+ { C, 0, 1, 0, 1 }, 
+ { C, 1, 1, 1, 1 }, 
+ { C, 1, 1, 1, 1 }, 
+ { C, 0, 1, 1, 1 }, 
+ { C, 0, 1, 1, 1 }, 
+ { C, 1, 1, 1, 1 }, 
+ { C, 1, 1, 1, 1 }, 
+ { C, 0, 1, 1, 1 }, 
+ { C, 0, 1, 1, 1 }, 
+ { C, 0, 0, 0, 1 }, 
+ { C, 0, 0, 0, 1 }, 
+ { C, 0, 1, 0, 1 }, 
+ { C, 0, 1, 0, 1 }, 
+ { C, 1, 1, 0, 1 }, 
+ { C, 1, 1, 0, 1 }, 
+ { C, 0, 1, 0, 1 }, 
+ { C, 0, 1, 0, 1 }, 
+ { C, 1, 1, 1, 1 }, 
+ { C, 1, 1, 1, 1 }, 
+ { C, 0, 1, 1, 1 }, 
+ { C, 0, 1, 1, 1 }, 
+ { C, 1, 1, 1, 1 }, 
+ { C, 1, 1, 1, 1 }, 
+ { C, 0, 1, 1, 1 }, 
+ { C, 0, 1, 1, 1 }, 
+ { P, 0, 0, 0, 0 }, 
+ { P, 0, 0, 0, 0 }, 
+ { P, 0, 0, 0, 0 }, 
+ { P, 0, 0, 0, 0 }, 
+ { P, 0, 0, 0, 0 }, 
+ { N, 1, 1, 0, 0 }, 
+ { N, 0, 1, 0, 0 }, 
+ { N, 0, 1, 0, 0 }, 
+ { P, 0, 0, 0, 0 }, 
+ { P, 0, 0, 0, 0 }, 
+ { C, 0, 1, 1, 0 }, 
+ { C, 0, 1, 1, 0 }, 
+ { P, 0, 0, 0, 0 }, 
+ { N, 1, 1, 0, 0 }, 
+ { C, 0, 1, 1, 0 }, 
+ { C, 0, 1, 1, 0 }, 
+ { P, 0, 0, 0, 0 }, 
+ { P, 0, 0, 0, 0 }, 
+ { P, 0, 0, 0, 0 }, 
+ { P, 0, 0, 0, 0 }, 
+ { P, 0, 0, 0, 0 }, 
+ { N, 1, 1, 0, 0 }, 
+ { N, 0, 1, 0, 0 }, 
+ { N, 0, 1, 0, 0 }, 
+ { P, 0, 0, 0, 0 }, 
+ { P, 0, 0, 0, 0 }, 
+ { C, 0, 1, 1, 0 }, 
+ { C, 0, 1, 1, 0 }, 
+ { P, 0, 0, 0, 0 }, 
+ { N, 1, 1, 0, 0 }, 
+ { C, 0, 1, 1, 0 }, 
+ { C, 0, 1, 1, 0 }, 
+ { P, 0, 0, 0, 0 }, 
+ { P, 0, 0, 0, 0 }, 
+ { P, 0, 0, 0, 0 }, 
+ { P, 0, 0, 0, 0 }, 
+ { P, 0, 0, 0, 0 }, 
+ { N, 1, 1, 0, 1 }, 
+ { N, 0, 1, 0, 1 }, 
+ { N, 0, 1, 0, 1 }, 
+ { P, 0, 0, 0, 0 }, 
+ { P, 0, 0, 0, 0 }, 
+ { C, 0, 1, 1, 1 }, 
+ { C, 0, 1, 1, 1 }, 
+ { P, 0, 0, 0, 0 }, 
+ { N, 1, 1, 0, 1 }, 
+ { C, 0, 1, 1, 1 }, 
+ { C, 0, 1, 1, 1 }, 
+ { P, 0, 0, 0, 0 }, 
+ { C, 0, 0, 0, 1 }, 
+ { P, 0, 0, 0, 0 }, 
+ { C, 0, 1, 0, 1 }, 
+ { P, 0, 0, 0, 0 }, 
+ { C, 1, 1, 0, 1 }, 
+ { C, 0, 1, 0, 1 }, 
+ { C, 0, 1, 0, 1 }, 
+ { P, 0, 0, 0, 0 }, 
+ { C, 1, 1, 1, 1 }, 
+ { C, 0, 1, 1, 1 }, 
+ { C, 0, 1, 1, 1 }, 
+ { P, 0, 0, 0, 0 }, 
+ { C, 1, 1, 1, 1 }, 
+ { C, 0, 1, 1, 1 }, 
+ { C, 0, 1, 1, 1 } 
+};
+
+void brw_wm_lookup_iz( unsigned line_aa,
+		       unsigned lookup,
+		       struct brw_wm_prog_key *key )
+{
+   unsigned reg = 2;
+
+   assert (lookup < IZ_BIT_MAX);
+      
+   if (lookup & IZ_PS_COMPUTES_DEPTH_BIT)
+      key->computes_depth = 1;
+
+   if (wm_iz_table[lookup].sd_present) {
+      key->source_depth_reg = reg;
+      reg += 2;
+   }
+
+   if (wm_iz_table[lookup].sd_to_rt)
+      key->source_depth_to_render_target = 1;
+
+   if (wm_iz_table[lookup].ds_present || line_aa != AA_NEVER) {
+      key->aa_dest_stencil_reg = reg;
+      key->runtime_check_aads_emit = (!wm_iz_table[lookup].ds_present &&
+				      line_aa == AA_SOMETIMES);
+      reg++;
+   }
+
+   if (wm_iz_table[lookup].dd_present) {
+      key->dest_depth_reg = reg;
+      reg+=2;
+   }
+
+   key->nr_depth_regs = (reg+1)/2;
+}
+
diff --git a/src/gallium/drivers/i965simple/brw_wm_sampler_state.c b/src/gallium/drivers/i965simple/brw_wm_sampler_state.c
new file mode 100644
index 0000000000..de42ffc5b1
--- /dev/null
+++ b/src/gallium/drivers/i965simple/brw_wm_sampler_state.c
@@ -0,0 +1,273 @@
+/*
+ Copyright (C) Intel Corp.  2006.  All Rights Reserved.
+ Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to
+ develop this 3D driver.
+
+ Permission is hereby granted, free of charge, to any person obtaining
+ a copy of this software and associated documentation files (the
+ "Software"), to deal in the Software without restriction, including
+ without limitation the rights to use, copy, modify, merge, publish,
+ distribute, sublicense, and/or sell copies of the Software, and to
+ permit persons to whom the Software is furnished to do so, subject to
+ the following conditions:
+
+ The above copyright notice and this permission notice (including the
+ next paragraph) shall be included in all copies or substantial
+ portions of the Software.
+
+ THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
+ IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
+ LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
+ OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+
+ **********************************************************************/
+ /*
+  * Authors:
+  *   Keith Whitwell <keith@tungstengraphics.com>
+  */
+
+
+#include "brw_context.h"
+#include "brw_state.h"
+#include "brw_defines.h"
+
+#include "pipe/p_util.h"
+
+
+#define COMPAREFUNC_ALWAYS		0
+#define COMPAREFUNC_NEVER		0x1
+#define COMPAREFUNC_LESS		0x2
+#define COMPAREFUNC_EQUAL		0x3
+#define COMPAREFUNC_LEQUAL		0x4
+#define COMPAREFUNC_GREATER		0x5
+#define COMPAREFUNC_NOTEQUAL		0x6
+#define COMPAREFUNC_GEQUAL		0x7
+
+/* Samplers aren't strictly wm state from the hardware's perspective,
+ * but that is the only situation in which we use them in this driver.
+ */
+
+static int intel_translate_shadow_compare_func(unsigned func)
+{
+   switch(func) {
+   case PIPE_FUNC_NEVER:
+       return COMPAREFUNC_ALWAYS;
+   case PIPE_FUNC_LESS:
+       return COMPAREFUNC_LEQUAL;
+   case PIPE_FUNC_LEQUAL:
+       return COMPAREFUNC_LESS;
+   case PIPE_FUNC_GREATER:
+       return COMPAREFUNC_GEQUAL;
+   case PIPE_FUNC_GEQUAL:
+      return COMPAREFUNC_GREATER;
+   case PIPE_FUNC_NOTEQUAL:
+      return COMPAREFUNC_EQUAL;
+   case PIPE_FUNC_EQUAL:
+      return COMPAREFUNC_NOTEQUAL;
+   case PIPE_FUNC_ALWAYS:
+       return COMPAREFUNC_NEVER;
+   }
+
+   debug_printf("Unknown value in %s: %x\n", __FUNCTION__, func);
+   return COMPAREFUNC_NEVER;
+}
+
+/* The brw (and related graphics cores) do not support GL_CLAMP.  The
+ * Intel drivers for "other operating systems" implement GL_CLAMP as
+ * GL_CLAMP_TO_EDGE, so the same is done here.
+ */
+static unsigned translate_wrap_mode( int wrap )
+{
+   switch( wrap ) {
+   case PIPE_TEX_WRAP_REPEAT:
+      return BRW_TEXCOORDMODE_WRAP;
+   case PIPE_TEX_WRAP_CLAMP:
+      return BRW_TEXCOORDMODE_CLAMP;
+   case PIPE_TEX_WRAP_CLAMP_TO_EDGE:
+      return BRW_TEXCOORDMODE_CLAMP; /* conform likes it this way */
+   case PIPE_TEX_WRAP_CLAMP_TO_BORDER:
+      return BRW_TEXCOORDMODE_CLAMP_BORDER;
+   case PIPE_TEX_WRAP_MIRROR_REPEAT:
+      return BRW_TEXCOORDMODE_MIRROR;
+   default:
+      return BRW_TEXCOORDMODE_WRAP;
+   }
+}
+
+
+static unsigned U_FIXED(float value, unsigned frac_bits)
+{
+   value *= (1<<frac_bits);
+   return value < 0 ? 0 : value;
+}
+
+static int S_FIXED(float value, unsigned frac_bits)
+{
+   return value * (1<<frac_bits);
+}
+
+
+static unsigned upload_default_color( struct brw_context *brw,
+                                      const float *color )
+{
+   struct brw_sampler_default_color sdc;
+
+   COPY_4V(sdc.color, color);
+
+   return brw_cache_data( &brw->cache[BRW_SAMPLER_DEFAULT_COLOR], &sdc );
+}
+
+
+/*
+ */
+static void brw_update_sampler_state( const struct pipe_sampler_state *pipe_sampler,
+				      unsigned sdc_gs_offset,
+				      struct brw_sampler_state *sampler)
+{
+   memset(sampler, 0, sizeof(*sampler));
+
+   switch (pipe_sampler->min_mip_filter) {
+   case PIPE_TEX_FILTER_NEAREST:
+      sampler->ss0.min_filter = BRW_MAPFILTER_NEAREST;
+      break;
+   case PIPE_TEX_FILTER_LINEAR:
+      sampler->ss0.min_filter = BRW_MAPFILTER_LINEAR;
+      break;
+   default:
+      break;
+   }
+
+   switch (pipe_sampler->min_mip_filter) {
+   case PIPE_TEX_MIPFILTER_NEAREST:
+      sampler->ss0.mip_filter = BRW_MIPFILTER_NEAREST;
+      break;
+   case PIPE_TEX_MIPFILTER_LINEAR:
+      sampler->ss0.mip_filter = BRW_MIPFILTER_LINEAR;
+      break;
+   case PIPE_TEX_MIPFILTER_NONE:
+      sampler->ss0.mip_filter = BRW_MIPFILTER_NONE;
+      break;
+   default:
+      break;
+   }
+   /* Set Anisotropy:
+    */
+   if (pipe_sampler->max_anisotropy > 1.0) {
+      sampler->ss0.min_filter = BRW_MAPFILTER_ANISOTROPIC;
+      sampler->ss0.mag_filter = BRW_MAPFILTER_ANISOTROPIC;
+
+      if (pipe_sampler->max_anisotropy > 2.0) {
+	 sampler->ss3.max_aniso = MAX2((pipe_sampler->max_anisotropy - 2) / 2,
+				       BRW_ANISORATIO_16);
+      }
+   }
+   else {
+      switch (pipe_sampler->mag_img_filter) {
+      case PIPE_TEX_FILTER_NEAREST:
+	 sampler->ss0.mag_filter = BRW_MAPFILTER_NEAREST;
+	 break;
+      case PIPE_TEX_FILTER_LINEAR:
+	 sampler->ss0.mag_filter = BRW_MAPFILTER_LINEAR;
+	 break;
+      default:
+	 break;
+      }
+   }
+
+   sampler->ss1.s_wrap_mode = translate_wrap_mode(pipe_sampler->wrap_s);
+   sampler->ss1.r_wrap_mode = translate_wrap_mode(pipe_sampler->wrap_r);
+   sampler->ss1.t_wrap_mode = translate_wrap_mode(pipe_sampler->wrap_t);
+
+   /* Fulsim complains if I don't do this.  Hardware doesn't mind:
+    */
+#if 0
+   if (texObj->Target == GL_TEXTURE_CUBE_MAP_ARB) {
+      sampler->ss1.r_wrap_mode = BRW_TEXCOORDMODE_CUBE;
+      sampler->ss1.s_wrap_mode = BRW_TEXCOORDMODE_CUBE;
+      sampler->ss1.t_wrap_mode = BRW_TEXCOORDMODE_CUBE;
+   }
+#endif
+
+   /* Set shadow function:
+    */
+   if (pipe_sampler->compare_mode == PIPE_TEX_COMPARE_R_TO_TEXTURE) {
+      /* Shadowing is "enabled" by emitting a particular sampler
+       * message (sample_c).  So need to recompile WM program when
+       * shadow comparison is enabled on each/any texture unit.
+       */
+      sampler->ss0.shadow_function = intel_translate_shadow_compare_func(pipe_sampler->compare_func);
+   }
+
+   /* Set LOD bias:
+    */
+   sampler->ss0.lod_bias = S_FIXED(CLAMP(pipe_sampler->lod_bias, -16, 15), 6);
+
+   sampler->ss0.lod_preclamp = 1; /* OpenGL mode */
+   sampler->ss0.default_color_mode = 0; /* OpenGL/DX10 mode */
+
+   /* Set BaseMipLevel, MaxLOD, MinLOD:
+    *
+    * XXX: I don't think that using firstLevel, lastLevel works,
+    * because we always setup the surface state as if firstLevel ==
+    * level zero.  Probably have to subtract firstLevel from each of
+    * these:
+    */
+   sampler->ss0.base_level = U_FIXED(0, 1);
+
+   sampler->ss1.max_lod = U_FIXED(MIN2(MAX2(pipe_sampler->max_lod, 0), 13), 6);
+   sampler->ss1.min_lod = U_FIXED(MIN2(MAX2(pipe_sampler->min_lod, 0), 13), 6);
+
+   sampler->ss2.default_color_pointer = sdc_gs_offset >> 5;
+}
+
+
+
+/* All samplers must be uploaded in a single contiguous array, which
+ * complicates various things.  However, this is still too confusing -
+ * FIXME: simplify all the different new texture state flags.
+ */
+static void upload_wm_samplers(struct brw_context *brw)
+{
+   unsigned unit;
+   unsigned sampler_count = 0;
+
+   /* BRW_NEW_SAMPLER */
+   for (unit = 0; unit < BRW_MAX_TEX_UNIT; unit++) {
+      /* determine unit enable/disable by looking for a bound texture */
+      if (brw->attribs.Texture[unit]) {
+         const struct pipe_sampler_state *sampler = brw->attribs.Samplers[unit];
+	 unsigned sdc_gs_offset = upload_default_color(brw, sampler->border_color);
+
+	 brw_update_sampler_state(sampler,
+				  sdc_gs_offset,
+				  &brw->wm.sampler[unit]);
+
+	 sampler_count = unit + 1;
+      }
+   }
+
+   if (brw->wm.sampler_count != sampler_count) {
+      brw->wm.sampler_count = sampler_count;
+      brw->state.dirty.cache |= CACHE_NEW_SAMPLER;
+   }
+
+   brw->wm.sampler_gs_offset = 0;
+
+   if (brw->wm.sampler_count)
+      brw->wm.sampler_gs_offset =
+	 brw_cache_data_sz(&brw->cache[BRW_SAMPLER],
+			   brw->wm.sampler,
+			   sizeof(struct brw_sampler_state) * brw->wm.sampler_count);
+}
+
+const struct brw_tracked_state brw_wm_samplers = {
+   .dirty = {
+      .brw = BRW_NEW_SAMPLER,
+      .cache = 0
+   },
+   .update = upload_wm_samplers
+};
+
diff --git a/src/gallium/drivers/i965simple/brw_wm_state.c b/src/gallium/drivers/i965simple/brw_wm_state.c
new file mode 100644
index 0000000000..5ccd488842
--- /dev/null
+++ b/src/gallium/drivers/i965simple/brw_wm_state.c
@@ -0,0 +1,194 @@
+/*
+ Copyright (C) Intel Corp.  2006.  All Rights Reserved.
+ Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to
+ develop this 3D driver.
+
+ Permission is hereby granted, free of charge, to any person obtaining
+ a copy of this software and associated documentation files (the
+ "Software"), to deal in the Software without restriction, including
+ without limitation the rights to use, copy, modify, merge, publish,
+ distribute, sublicense, and/or sell copies of the Software, and to
+ permit persons to whom the Software is furnished to do so, subject to
+ the following conditions:
+
+ The above copyright notice and this permission notice (including the
+ next paragraph) shall be included in all copies or substantial
+ portions of the Software.
+
+ THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
+ IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
+ LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
+ OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+
+ **********************************************************************/
+ /*
+  * Authors:
+  *   Keith Whitwell <keith@tungstengraphics.com>
+  */
+
+
+#include "brw_context.h"
+#include "brw_state.h"
+#include "brw_defines.h"
+#include "brw_wm.h"
+#include "pipe/p_util.h"
+
+/***********************************************************************
+ * WM unit - fragment programs and rasterization
+ */
+static void upload_wm_unit(struct brw_context *brw )
+{
+   struct brw_wm_unit_state wm;
+   unsigned max_threads;
+   unsigned per_thread;
+
+   if (BRW_DEBUG & DEBUG_SINGLE_THREAD)
+      max_threads = 0;
+   else
+      max_threads = 31;
+
+
+   memset(&wm, 0, sizeof(wm));
+
+   /* CACHE_NEW_WM_PROG */
+   wm.thread0.grf_reg_count = align(brw->wm.prog_data->total_grf, 16) / 16 - 1;
+   wm.thread0.kernel_start_pointer = brw->wm.prog_gs_offset >> 6;
+   wm.thread3.dispatch_grf_start_reg = brw->wm.prog_data->first_curbe_grf;
+   wm.thread3.urb_entry_read_length = brw->wm.prog_data->urb_read_length;
+   wm.thread3.const_urb_entry_read_length = brw->wm.prog_data->curb_read_length;
+
+   wm.wm5.max_threads = max_threads;
+
+   per_thread = align(brw->wm.prog_data->total_scratch, 1024);
+   assert(per_thread <= 12 * 1024);
+
+#if 0
+   if (brw->wm.prog_data->total_scratch) {
+      unsigned total = per_thread * (max_threads + 1);
+
+      /* Scratch space -- just have to make sure there is sufficient
+       * allocated for the active program and current number of threads.
+       */
+      brw->wm.scratch_buffer_size = total;
+      if (brw->wm.scratch_buffer &&
+	  brw->wm.scratch_buffer_size > brw->wm.scratch_buffer->size) {
+	 dri_bo_unreference(brw->wm.scratch_buffer);
+	 brw->wm.scratch_buffer = NULL;
+      }
+      if (!brw->wm.scratch_buffer) {
+	 brw->wm.scratch_buffer = dri_bo_alloc(intel->intelScreen->bufmgr,
+					       "wm scratch",
+					       brw->wm.scratch_buffer_size,
+					       4096, DRM_BO_FLAG_MEM_TT);
+      }
+   }
+   /* XXX: Scratch buffers are not implemented correectly.
+    *
+    * The scratch offset to be programmed into wm is relative to the general
+    * state base address.  However, using dri_bo_alloc/dri_bo_emit_reloc (or
+    * the previous bmGenBuffers scheme), we get an offset relative to the
+    * start of framebuffer.  Even before then, it was broken in other ways,
+    * so just fail for now if we hit that path.
+    */
+   assert(brw->wm.prog_data->total_scratch == 0);
+#endif
+
+   /* CACHE_NEW_SURFACE */
+   wm.thread1.binding_table_entry_count = brw->wm.nr_surfaces;
+
+   /* BRW_NEW_CURBE_OFFSETS */
+   wm.thread3.const_urb_entry_read_offset = brw->curbe.wm_start * 2;
+
+   wm.thread3.urb_entry_read_offset = 0;
+   wm.thread1.depth_coef_urb_read_offset = 1;
+   wm.thread1.floating_point_mode = BRW_FLOATING_POINT_NON_IEEE_754;
+
+   /* CACHE_NEW_SAMPLER */
+   wm.wm4.sampler_count = (brw->wm.sampler_count + 1) / 4;
+   wm.wm4.sampler_state_pointer = brw->wm.sampler_gs_offset >> 5;
+
+   /* BRW_NEW_FRAGMENT_PROGRAM */
+   {
+      const struct brw_fragment_program *fp = brw->attribs.FragmentProgram;
+
+      if (fp->UsesDepth)
+	 wm.wm5.program_uses_depth = 1; /* as far as we can tell */
+
+      if (fp->ComputesDepth)
+	 wm.wm5.program_computes_depth = 1;
+
+      /* BRW_NEW_ALPHA_TEST */
+      if (fp->UsesKill ||
+	  brw->attribs.DepthStencil->alpha.enabled)
+	 wm.wm5.program_uses_killpixel = 1;
+
+      wm.wm5.enable_8_pix = 1;
+   }
+
+   wm.wm5.thread_dispatch_enable = 1;	/* AKA: color_write */
+   wm.wm5.legacy_line_rast = 0;
+   wm.wm5.legacy_global_depth_bias = 0;
+   wm.wm5.early_depth_test = 1;	        /* never need to disable */
+   wm.wm5.line_aa_region_width = 0;
+   wm.wm5.line_endcap_aa_region_width = 1;
+
+   /* BRW_NEW_RASTERIZER */
+   if (brw->attribs.Raster->poly_stipple_enable)
+      wm.wm5.polygon_stipple = 1;
+
+#if 0
+   if (brw->attribs.Polygon->OffsetFill) {
+      wm.wm5.depth_offset = 1;
+      /* Something wierd going on with legacy_global_depth_bias,
+       * offset_constant, scaling and MRD.  This value passes glean
+       * but gives some odd results elsewere (eg. the
+       * quad-offset-units test).
+       */
+      wm.global_depth_offset_constant = brw->attribs.Polygon->OffsetUnits * 2;
+
+      /* This is the only value that passes glean:
+       */
+      wm.global_depth_offset_scale = brw->attribs.Polygon->OffsetFactor;
+   }
+#endif
+
+   if (brw->attribs.Raster->line_stipple_enable) {
+      wm.wm5.line_stipple = 1;
+   }
+
+   if (BRW_DEBUG & DEBUG_STATS)
+      wm.wm4.stats_enable = 1;
+
+   brw->wm.state_gs_offset = brw_cache_data( &brw->cache[BRW_WM_UNIT], &wm );
+
+   if (brw->wm.prog_data->total_scratch) {
+      /*
+      dri_emit_reloc(brw->cache[BRW_WM_UNIT].pool->buffer,
+		     DRM_BO_FLAG_MEM_TT | DRM_BO_FLAG_READ | DRM_BO_FLAG_WRITE,
+		     (per_thread / 1024) - 1,
+		     brw->wm.state_gs_offset +
+		     ((char *)&wm.thread2 - (char *)&wm),
+		     brw->wm.scratch_buffer);
+      */
+   } else {
+      wm.thread2.scratch_space_base_pointer = 0;
+   }
+}
+
+const struct brw_tracked_state brw_wm_unit = {
+   .dirty = {
+      .brw = (BRW_NEW_RASTERIZER |
+	      BRW_NEW_ALPHA_TEST |
+	      BRW_NEW_FS |
+	      BRW_NEW_CURBE_OFFSETS),
+
+      .cache = (CACHE_NEW_SURFACE |
+		CACHE_NEW_WM_PROG |
+		CACHE_NEW_SAMPLER)
+   },
+   .update = upload_wm_unit
+};
+
diff --git a/src/gallium/drivers/i965simple/brw_wm_surface_state.c b/src/gallium/drivers/i965simple/brw_wm_surface_state.c
new file mode 100644
index 0000000000..d16d919bce
--- /dev/null
+++ b/src/gallium/drivers/i965simple/brw_wm_surface_state.c
@@ -0,0 +1,304 @@
+/*
+ Copyright (C) Intel Corp.  2006.  All Rights Reserved.
+ Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to
+ develop this 3D driver.
+
+ Permission is hereby granted, free of charge, to any person obtaining
+ a copy of this software and associated documentation files (the
+ "Software"), to deal in the Software without restriction, including
+ without limitation the rights to use, copy, modify, merge, publish,
+ distribute, sublicense, and/or sell copies of the Software, and to
+ permit persons to whom the Software is furnished to do so, subject to
+ the following conditions:
+
+ The above copyright notice and this permission notice (including the
+ next paragraph) shall be included in all copies or substantial
+ portions of the Software.
+
+ THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
+ IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
+ LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
+ OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+
+ **********************************************************************/
+ /*
+  * Authors:
+  *   Keith Whitwell <keith@tungstengraphics.com>
+  */
+
+#include "brw_context.h"
+#include "brw_state.h"
+#include "brw_defines.h"
+
+static unsigned translate_tex_target( enum pipe_texture_target target )
+{
+   switch (target) {
+   case PIPE_TEXTURE_1D:
+      return BRW_SURFACE_1D;
+
+   case PIPE_TEXTURE_2D:
+      return BRW_SURFACE_2D;
+
+   case PIPE_TEXTURE_3D:
+      return BRW_SURFACE_3D;
+
+   case PIPE_TEXTURE_CUBE:
+      return BRW_SURFACE_CUBE;
+
+   default:
+      assert(0);
+      return 0;
+   }
+}
+
+static unsigned translate_tex_format( enum pipe_format pipe_format )
+{
+   switch( pipe_format ) {
+   case PIPE_FORMAT_U_L8:
+      return BRW_SURFACEFORMAT_L8_UNORM;
+
+   case PIPE_FORMAT_U_I8:
+      return BRW_SURFACEFORMAT_I8_UNORM;
+
+   case PIPE_FORMAT_U_A8:
+      return BRW_SURFACEFORMAT_A8_UNORM;
+
+   case PIPE_FORMAT_U_A8_L8:
+      return BRW_SURFACEFORMAT_L8A8_UNORM;
+
+   case PIPE_FORMAT_R8G8B8_UNORM:
+      assert(0);		/* not supported for sampling */
+      return BRW_SURFACEFORMAT_R8G8B8_UNORM;
+
+   case PIPE_FORMAT_B8G8R8A8_UNORM:
+      return BRW_SURFACEFORMAT_B8G8R8A8_UNORM;
+
+   case PIPE_FORMAT_R8G8B8A8_UNORM:
+      return BRW_SURFACEFORMAT_R8G8B8A8_UNORM;
+
+   case PIPE_FORMAT_R5G6B5_UNORM:
+      return BRW_SURFACEFORMAT_B5G6R5_UNORM;
+
+   case PIPE_FORMAT_A1R5G5B5_UNORM:
+      return BRW_SURFACEFORMAT_B5G5R5A1_UNORM;
+
+   case PIPE_FORMAT_A4R4G4B4_UNORM:
+      return BRW_SURFACEFORMAT_B4G4R4A4_UNORM;
+
+   case PIPE_FORMAT_YCBCR_REV:
+      return BRW_SURFACEFORMAT_YCRCB_NORMAL;
+
+   case PIPE_FORMAT_YCBCR:
+      return BRW_SURFACEFORMAT_YCRCB_SWAPUVY;
+#if 0
+   case PIPE_FORMAT_RGB_FXT1:
+   case PIPE_FORMAT_RGBA_FXT1:
+      return BRW_SURFACEFORMAT_FXT1;
+#endif
+
+   case PIPE_FORMAT_Z16_UNORM:
+      return BRW_SURFACEFORMAT_I16_UNORM;
+#if 0
+   case PIPE_FORMAT_RGB_DXT1:
+       return BRW_SURFACEFORMAT_DXT1_RGB;
+
+   case PIPE_FORMAT_RGBA_DXT1:
+       return BRW_SURFACEFORMAT_BC1_UNORM;
+
+   case PIPE_FORMAT_RGBA_DXT3:
+       return BRW_SURFACEFORMAT_BC2_UNORM;
+
+   case PIPE_FORMAT_RGBA_DXT5:
+       return BRW_SURFACEFORMAT_BC3_UNORM;
+
+   case PIPE_FORMAT_SRGBA8:
+      return BRW_SURFACEFORMAT_R8G8B8A8_UNORM_SRGB;
+   case PIPE_FORMAT_SRGB_DXT1:
+      return BRW_SURFACEFORMAT_BC1_UNORM_SRGB;
+#endif
+
+   default:
+      assert(0);
+      return 0;
+   }
+}
+
+static unsigned brw_buffer_offset(struct brw_context *brw,
+                                  struct pipe_buffer *buffer)
+{
+   return brw->winsys->get_buffer_offset(brw->winsys,
+                                         buffer,
+                                         0);
+}
+
+static
+void brw_update_texture_surface( struct brw_context *brw,
+				 unsigned unit )
+{
+   const struct brw_texture *tObj = brw->attribs.Texture[unit];
+   struct brw_surface_state surf;
+
+   memset(&surf, 0, sizeof(surf));
+
+   surf.ss0.mipmap_layout_mode = BRW_SURFACE_MIPMAPLAYOUT_BELOW;
+   surf.ss0.surface_type = translate_tex_target(tObj->base.target);
+   surf.ss0.surface_format = translate_tex_format(tObj->base.format);
+
+   /* This is ok for all textures with channel width 8bit or less:
+    */
+/*    surf.ss0.data_return_format = BRW_SURFACERETURNFORMAT_S1; */
+
+   /* Updated in emit_reloc */
+   surf.ss1.base_addr = brw_buffer_offset( brw, tObj->buffer );
+
+   surf.ss2.mip_count = tObj->base.last_level;
+   surf.ss2.width = tObj->base.width[0] - 1;
+   surf.ss2.height = tObj->base.height[0] - 1;
+
+   surf.ss3.tile_walk = BRW_TILEWALK_XMAJOR;
+   surf.ss3.tiled_surface = 0; /* always zero */
+   surf.ss3.pitch = tObj->pitch - 1;
+   surf.ss3.depth = tObj->base.depth[0] - 1;
+
+   surf.ss4.min_lod = 0;
+
+   if (tObj->base.target == PIPE_TEXTURE_CUBE) {
+      surf.ss0.cube_pos_x = 1;
+      surf.ss0.cube_pos_y = 1;
+      surf.ss0.cube_pos_z = 1;
+      surf.ss0.cube_neg_x = 1;
+      surf.ss0.cube_neg_y = 1;
+      surf.ss0.cube_neg_z = 1;
+   }
+
+   brw->wm.bind.surf_ss_offset[unit + 1] =
+      brw_cache_data( &brw->cache[BRW_SS_SURFACE], &surf );
+}
+
+
+
+#define OFFSET(TYPE, FIELD) ( (unsigned)&(((TYPE *)0)->FIELD) )
+
+
+static void upload_wm_surfaces(struct brw_context *brw )
+{
+   unsigned i;
+
+   {
+      struct brw_surface_state surf;
+
+      /* BRW_NEW_FRAMEBUFFER
+       */
+      struct pipe_surface *pipe_surface = brw->attribs.FrameBuffer.cbufs[0];/*fixme*/
+
+      memset(&surf, 0, sizeof(surf));
+
+      if (pipe_surface != NULL) {
+	 if (pipe_surface->cpp == 4)
+	    surf.ss0.surface_format = BRW_SURFACEFORMAT_B8G8R8A8_UNORM;
+	 else
+	    surf.ss0.surface_format = BRW_SURFACEFORMAT_B5G6R5_UNORM;
+
+	 surf.ss0.surface_type = BRW_SURFACE_2D;
+
+	 surf.ss1.base_addr = brw_buffer_offset( brw, pipe_surface->buffer );
+
+	 surf.ss2.width = pipe_surface->width - 1;
+	 surf.ss2.height = pipe_surface->height - 1;
+	 surf.ss3.tile_walk = BRW_TILEWALK_XMAJOR;
+	 surf.ss3.tiled_surface = 0;
+	 surf.ss3.pitch = (pipe_surface->pitch * pipe_surface->cpp) - 1;
+      } else {
+	 surf.ss0.surface_format = BRW_SURFACEFORMAT_B8G8R8A8_UNORM;
+	 surf.ss0.surface_type = BRW_SURFACE_NULL;
+      }
+
+      /* BRW_NEW_BLEND */
+      surf.ss0.color_blend = (!brw->attribs.Blend->logicop_enable &&
+			      brw->attribs.Blend->blend_enable);
+
+
+      surf.ss0.writedisable_red =   !(brw->attribs.Blend->colormask & PIPE_MASK_R);
+      surf.ss0.writedisable_green = !(brw->attribs.Blend->colormask & PIPE_MASK_G);
+      surf.ss0.writedisable_blue =  !(brw->attribs.Blend->colormask & PIPE_MASK_B);
+      surf.ss0.writedisable_alpha = !(brw->attribs.Blend->colormask & PIPE_MASK_A);
+
+
+
+
+      brw->wm.bind.surf_ss_offset[0] = brw_cache_data( &brw->cache[BRW_SS_SURFACE], &surf );
+
+      brw->wm.nr_surfaces = 1;
+   }
+
+
+   /* BRW_NEW_TEXTURE
+    */
+   for (i = 0; i < BRW_MAX_TEX_UNIT; i++) {
+      const struct brw_texture *texUnit = brw->attribs.Texture[i];
+
+      if (texUnit &&
+	  texUnit->base.refcount/*(texUnit->refcount > 0) == really used */) {
+
+	 brw_update_texture_surface(brw, i);
+
+	 brw->wm.nr_surfaces = i+2;
+      }
+      else {
+	 brw->wm.bind.surf_ss_offset[i+1] = 0;
+      }
+   }
+
+   brw->wm.bind_ss_offset = brw_cache_data( &brw->cache[BRW_SS_SURF_BIND],
+					    &brw->wm.bind );
+}
+
+
+/* KW: Will find a different way to acheive this, see for example the
+ * state caches with relocs in the i915 swz driver.
+ */
+#if 0
+static void emit_reloc_wm_surfaces(struct brw_context *brw)
+{
+   int unit;
+
+   if (brw->state.draw_region != NULL) {
+      /* Emit framebuffer relocation */
+      dri_emit_reloc(brw_cache_buffer(brw, BRW_SS_SURFACE),
+		     DRM_BO_FLAG_MEM_TT | DRM_BO_FLAG_READ | DRM_BO_FLAG_WRITE,
+		     0,
+		     brw->wm.bind.surf_ss_offset[0] +
+		     offsetof(struct brw_surface_state, ss1),
+		     brw->state.draw_region->buffer);
+   }
+
+   /* Emit relocations for texture buffers */
+   for (unit = 0; unit < BRW_MAX_TEX_UNIT; unit++) {
+      struct gl_texture_unit *texUnit = &brw->attribs.Texture->Unit[unit];
+      struct gl_texture_object *tObj = texUnit->_Current;
+      struct intel_texture_object *intelObj = intel_texture_object(tObj);
+
+      if (texUnit->_ReallyEnabled && intelObj->mt != NULL) {
+	 dri_emit_reloc(brw_cache_buffer(brw, BRW_SS_SURFACE),
+			DRM_BO_FLAG_MEM_TT | DRM_BO_FLAG_READ,
+			0,
+			brw->wm.bind.surf_ss_offset[unit + 1] +
+			offsetof(struct brw_surface_state, ss1),
+			intelObj->mt->region->buffer);
+      }
+   }
+}
+#endif
+
+const struct brw_tracked_state brw_wm_surfaces = {
+   .dirty = {
+      .brw = (BRW_NEW_FRAMEBUFFER |
+	      BRW_NEW_BLEND |
+	      BRW_NEW_TEXTURE),
+      .cache = 0
+   },
+   .update = upload_wm_surfaces,
+};
diff --git a/src/gallium/drivers/softpipe/Makefile b/src/gallium/drivers/softpipe/Makefile
new file mode 100644
index 0000000000..31438a882e
--- /dev/null
+++ b/src/gallium/drivers/softpipe/Makefile
@@ -0,0 +1,50 @@
+
+TOP = ../../../..
+include $(TOP)/configs/current
+
+LIBNAME = softpipe
+
+DRIVER_SOURCES = \
+	sp_clear.c \
+	sp_flush.c \
+	sp_query.c \
+	sp_context.c \
+	sp_draw_arrays.c \
+	sp_prim_setup.c \
+	sp_prim_vbuf.c \
+	sp_quad.c \
+	sp_quad_alpha_test.c \
+	sp_quad_blend.c \
+	sp_quad_bufloop.c \
+	sp_quad_colormask.c \
+	sp_quad_coverage.c \
+	sp_quad_depth_test.c \
+	sp_quad_earlyz.c \
+	sp_quad_fs.c \
+	sp_quad_occlusion.c \
+	sp_quad_output.c \
+	sp_quad_stencil.c \
+	sp_quad_stipple.c \
+	sp_state_blend.c \
+	sp_state_clip.c \
+	sp_state_derived.c \
+	sp_state_fs.c \
+	sp_state_sampler.c \
+	sp_state_rasterizer.c \
+	sp_state_surface.c \
+	sp_state_vertex.c \
+	sp_texture.c \
+	sp_tex_sample.c \
+	sp_tile_cache.c \
+	sp_surface.c 
+
+C_SOURCES = \
+	$(COMMON_SOURCES) \
+	$(DRIVER_SOURCES)
+
+ASM_SOURCES = 
+
+include ../Makefile.template
+
+symlinks:
+
diff --git a/src/gallium/drivers/softpipe/SConscript b/src/gallium/drivers/softpipe/SConscript
new file mode 100644
index 0000000000..d581ee8d3c
--- /dev/null
+++ b/src/gallium/drivers/softpipe/SConscript
@@ -0,0 +1,42 @@
+Import('*')
+
+env = env.Clone()
+
+softpipe = env.ConvenienceLibrary(
+	target = 'softpipe',
+	source = [
+		'sp_clear.c',
+		'sp_context.c',
+		'sp_draw_arrays.c',
+		'sp_flush.c',
+		'sp_prim_setup.c',
+		'sp_prim_vbuf.c',
+		'sp_quad_alpha_test.c',
+		'sp_quad_blend.c',
+		'sp_quad_bufloop.c',
+		'sp_quad.c',
+		'sp_quad_colormask.c',
+		'sp_quad_coverage.c',
+		'sp_quad_depth_test.c',
+		'sp_quad_earlyz.c',
+		'sp_quad_fs.c',
+		'sp_quad_occlusion.c',
+		'sp_quad_output.c',
+		'sp_quad_stencil.c',
+		'sp_quad_stipple.c',
+		'sp_query.c',
+		'sp_state_blend.c',
+		'sp_state_clip.c',
+		'sp_state_derived.c',
+		'sp_state_fs.c',
+		'sp_state_rasterizer.c',
+		'sp_state_sampler.c',
+		'sp_state_surface.c',
+		'sp_state_vertex.c',
+		'sp_surface.c',
+		'sp_tex_sample.c',
+		'sp_texture.c',
+		'sp_tile_cache.c',
+	])
+
+Export('softpipe')
\ No newline at end of file
diff --git a/src/gallium/drivers/softpipe/sp_clear.c b/src/gallium/drivers/softpipe/sp_clear.c
new file mode 100644
index 0000000000..8d295a30ca
--- /dev/null
+++ b/src/gallium/drivers/softpipe/sp_clear.c
@@ -0,0 +1,73 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+/* Author:
+ *    Brian Paul
+ */
+
+
+#include "pipe/p_defines.h"
+#include "sp_clear.h"
+#include "sp_context.h"
+#include "sp_surface.h"
+#include "sp_state.h"
+#include "sp_tile_cache.h"
+
+
+/**
+ * Clear the given surface to the specified value.
+ * No masking, no scissor (clear entire buffer).
+ */
+void
+softpipe_clear(struct pipe_context *pipe, struct pipe_surface *ps,
+               unsigned clearValue)
+{
+   struct softpipe_context *softpipe = softpipe_context(pipe);
+   uint i;
+
+#if 0
+   softpipe_update_derived(softpipe); /* not needed?? */
+#endif
+
+   if (ps == sp_tile_cache_get_surface(softpipe->zsbuf_cache)) {
+      sp_tile_cache_clear(softpipe->zsbuf_cache, clearValue);
+#if TILE_CLEAR_OPTIMIZATION
+      return;
+#endif
+   }
+
+   for (i = 0; i < softpipe->framebuffer.num_cbufs; i++) {
+      if (ps == sp_tile_cache_get_surface(softpipe->cbuf_cache[i])) {
+         sp_tile_cache_clear(softpipe->cbuf_cache[i], clearValue);
+      }
+   }
+
+#if !TILE_CLEAR_OPTIMIZATION
+   /* non-cached surface */
+   pipe->surface_fill(pipe, ps, 0, 0, ps->width, ps->height, clearValue);
+#endif
+}
diff --git a/src/gallium/drivers/softpipe/sp_clear.h b/src/gallium/drivers/softpipe/sp_clear.h
new file mode 100644
index 0000000000..a8ed1c4ecc
--- /dev/null
+++ b/src/gallium/drivers/softpipe/sp_clear.h
@@ -0,0 +1,43 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+/* Author:
+ *    Brian Paul
+ */
+
+#ifndef SP_CLEAR_H
+#define SP_CLEAR_H
+
+#include "pipe/p_state.h"
+struct pipe_context;
+
+extern void
+softpipe_clear(struct pipe_context *pipe, struct pipe_surface *ps,
+               unsigned clearValue);
+
+
+#endif /* SP_CLEAR_H */
diff --git a/src/gallium/drivers/softpipe/sp_context.c b/src/gallium/drivers/softpipe/sp_context.c
new file mode 100644
index 0000000000..cea6b90104
--- /dev/null
+++ b/src/gallium/drivers/softpipe/sp_context.c
@@ -0,0 +1,333 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+/* Author:
+ *    Keith Whitwell <keith@tungstengraphics.com>
+ */
+
+#include "pipe/draw/draw_context.h"
+#include "pipe/p_defines.h"
+#include "pipe/p_inlines.h"
+#include "pipe/p_util.h"
+#include "sp_clear.h"
+#include "sp_context.h"
+#include "sp_flush.h"
+#include "sp_prim_setup.h"
+#include "sp_prim_vbuf.h"
+#include "sp_state.h"
+#include "sp_surface.h"
+#include "sp_tile_cache.h"
+#include "sp_texture.h"
+#include "sp_winsys.h"
+#include "sp_query.h"
+
+
+
+/**
+ * Query format support for creating a texture, drawing surface, etc.
+ * \param format  the format to test
+ * \param type  one of PIPE_TEXTURE, PIPE_SURFACE
+ */
+static boolean
+softpipe_is_format_supported( struct pipe_context *pipe,
+                              enum pipe_format format, uint type )
+{
+   switch (type) {
+   case PIPE_TEXTURE:
+      /* softpipe supports all texture formats */
+      return TRUE;
+   case PIPE_SURFACE:
+      /* softpipe supports all (off-screen) surface formats */
+      return TRUE;
+   default:
+      assert(0);
+      return FALSE;
+   }
+}
+
+
+/**
+ * Map any drawing surfaces which aren't already mapped
+ */
+void
+softpipe_map_surfaces(struct softpipe_context *sp)
+{
+   unsigned i;
+
+   for (i = 0; i < sp->framebuffer.num_cbufs; i++) {
+      sp_tile_cache_map_surfaces(sp->cbuf_cache[i]);
+   }
+
+   sp_tile_cache_map_surfaces(sp->zsbuf_cache);
+}
+
+
+/**
+ * Unmap any mapped drawing surfaces
+ */
+void
+softpipe_unmap_surfaces(struct softpipe_context *sp)
+{
+   uint i;
+
+   for (i = 0; i < sp->framebuffer.num_cbufs; i++)
+      sp_flush_tile_cache(sp, sp->cbuf_cache[i]);
+   sp_flush_tile_cache(sp, sp->zsbuf_cache);
+
+   for (i = 0; i < sp->framebuffer.num_cbufs; i++) {
+      sp_tile_cache_unmap_surfaces(sp->cbuf_cache[i]);
+   }
+   sp_tile_cache_unmap_surfaces(sp->zsbuf_cache);
+}
+
+
+static void softpipe_destroy( struct pipe_context *pipe )
+{
+   struct softpipe_context *softpipe = softpipe_context( pipe );
+   struct pipe_winsys *ws = pipe->winsys;
+   uint i;
+
+   draw_destroy( softpipe->draw );
+
+   softpipe->quad.polygon_stipple->destroy( softpipe->quad.polygon_stipple );
+   softpipe->quad.earlyz->destroy( softpipe->quad.earlyz );
+   softpipe->quad.shade->destroy( softpipe->quad.shade );
+   softpipe->quad.alpha_test->destroy( softpipe->quad.alpha_test );
+   softpipe->quad.depth_test->destroy( softpipe->quad.depth_test );
+   softpipe->quad.stencil_test->destroy( softpipe->quad.stencil_test );
+   softpipe->quad.occlusion->destroy( softpipe->quad.occlusion );
+   softpipe->quad.coverage->destroy( softpipe->quad.coverage );
+   softpipe->quad.bufloop->destroy( softpipe->quad.bufloop );
+   softpipe->quad.blend->destroy( softpipe->quad.blend );
+   softpipe->quad.colormask->destroy( softpipe->quad.colormask );
+   softpipe->quad.output->destroy( softpipe->quad.output );
+
+   for (i = 0; i < PIPE_MAX_COLOR_BUFS; i++)
+      sp_destroy_tile_cache(softpipe->cbuf_cache[i]);
+   sp_destroy_tile_cache(softpipe->zsbuf_cache);
+
+   for (i = 0; i < PIPE_MAX_SAMPLERS; i++)
+      sp_destroy_tile_cache(softpipe->tex_cache[i]);
+
+   for (i = 0; i < Elements(softpipe->constants); i++) {
+      if (softpipe->constants[i].buffer) {
+         pipe_buffer_reference(ws, &softpipe->constants[i].buffer, NULL);
+      }
+   }
+
+   FREE( softpipe );
+}
+
+
+static const char *softpipe_get_name( struct pipe_context *pipe )
+{
+   return "softpipe";
+}
+
+static const char *softpipe_get_vendor( struct pipe_context *pipe )
+{
+   return "Tungsten Graphics, Inc.";
+}
+
+static int softpipe_get_param(struct pipe_context *pipe, int param)
+{
+   switch (param) {
+   case PIPE_CAP_MAX_TEXTURE_IMAGE_UNITS:
+      return 8;
+   case PIPE_CAP_NPOT_TEXTURES:
+      return 1;
+   case PIPE_CAP_TWO_SIDED_STENCIL:
+      return 1;
+   case PIPE_CAP_GLSL:
+      return 1;
+   case PIPE_CAP_S3TC:
+      return 0;
+   case PIPE_CAP_ANISOTROPIC_FILTER:
+      return 0;
+   case PIPE_CAP_POINT_SPRITE:
+      return 1;
+   case PIPE_CAP_MAX_RENDER_TARGETS:
+      return 1;
+   case PIPE_CAP_OCCLUSION_QUERY:
+      return 1;
+   case PIPE_CAP_TEXTURE_SHADOW_MAP:
+      return 1;
+   case PIPE_CAP_MAX_TEXTURE_2D_LEVELS:
+      return 12; /* max 2Kx2K */
+   case PIPE_CAP_MAX_TEXTURE_3D_LEVELS:
+      return 8;  /* max 128x128x128 */
+   case PIPE_CAP_MAX_TEXTURE_CUBE_LEVELS:
+      return 12; /* max 2Kx2K */
+   default:
+      return 0;
+   }
+}
+
+static float softpipe_get_paramf(struct pipe_context *pipe, int param)
+{
+   switch (param) {
+   case PIPE_CAP_MAX_LINE_WIDTH:
+      /* fall-through */
+   case PIPE_CAP_MAX_LINE_WIDTH_AA:
+      return 255.0; /* arbitrary */
+
+   case PIPE_CAP_MAX_POINT_WIDTH:
+      /* fall-through */
+   case PIPE_CAP_MAX_POINT_WIDTH_AA:
+      return 255.0; /* arbitrary */
+
+   case PIPE_CAP_MAX_TEXTURE_ANISOTROPY:
+      return 0.0;
+
+   case PIPE_CAP_MAX_TEXTURE_LOD_BIAS:
+      return 16.0; /* arbitrary */
+
+   default:
+      return 0;
+   }
+}
+
+struct pipe_context *softpipe_create( struct pipe_winsys *pipe_winsys,
+				      struct softpipe_winsys *softpipe_winsys )
+{
+   struct softpipe_context *softpipe = CALLOC_STRUCT(softpipe_context);
+   uint i;
+
+#if defined(__i386__) || defined(__386__)
+   softpipe->use_sse = GETENV( "GALLIUM_NOSSE" ) == NULL;
+#else
+   softpipe->use_sse = FALSE;
+#endif
+
+   softpipe->dump_fs = GETENV( "GALLIUM_DUMP_FS" ) != NULL;
+
+   softpipe->pipe.winsys = pipe_winsys;
+   softpipe->pipe.destroy = softpipe_destroy;
+
+   /* queries */
+   softpipe->pipe.is_format_supported = softpipe_is_format_supported;
+   softpipe->pipe.get_name = softpipe_get_name;
+   softpipe->pipe.get_vendor = softpipe_get_vendor;
+   softpipe->pipe.get_param = softpipe_get_param;
+   softpipe->pipe.get_paramf = softpipe_get_paramf;
+
+   /* state setters */
+   softpipe->pipe.create_blend_state = softpipe_create_blend_state;
+   softpipe->pipe.bind_blend_state   = softpipe_bind_blend_state;
+   softpipe->pipe.delete_blend_state = softpipe_delete_blend_state;
+
+   softpipe->pipe.create_sampler_state = softpipe_create_sampler_state;
+   softpipe->pipe.bind_sampler_state   = softpipe_bind_sampler_state;
+   softpipe->pipe.delete_sampler_state = softpipe_delete_sampler_state;
+
+   softpipe->pipe.create_depth_stencil_alpha_state = softpipe_create_depth_stencil_state;
+   softpipe->pipe.bind_depth_stencil_alpha_state   = softpipe_bind_depth_stencil_state;
+   softpipe->pipe.delete_depth_stencil_alpha_state = softpipe_delete_depth_stencil_state;
+
+   softpipe->pipe.create_rasterizer_state = softpipe_create_rasterizer_state;
+   softpipe->pipe.bind_rasterizer_state   = softpipe_bind_rasterizer_state;
+   softpipe->pipe.delete_rasterizer_state = softpipe_delete_rasterizer_state;
+
+   softpipe->pipe.create_fs_state = softpipe_create_fs_state;
+   softpipe->pipe.bind_fs_state   = softpipe_bind_fs_state;
+   softpipe->pipe.delete_fs_state = softpipe_delete_fs_state;
+
+   softpipe->pipe.create_vs_state = softpipe_create_vs_state;
+   softpipe->pipe.bind_vs_state   = softpipe_bind_vs_state;
+   softpipe->pipe.delete_vs_state = softpipe_delete_vs_state;
+
+   softpipe->pipe.set_blend_color = softpipe_set_blend_color;
+   softpipe->pipe.set_clip_state = softpipe_set_clip_state;
+   softpipe->pipe.set_constant_buffer = softpipe_set_constant_buffer;
+   softpipe->pipe.set_framebuffer_state = softpipe_set_framebuffer_state;
+   softpipe->pipe.set_polygon_stipple = softpipe_set_polygon_stipple;
+   softpipe->pipe.set_scissor_state = softpipe_set_scissor_state;
+   softpipe->pipe.set_sampler_texture = softpipe_set_sampler_texture;
+   softpipe->pipe.set_viewport_state = softpipe_set_viewport_state;
+
+   softpipe->pipe.set_vertex_buffer = softpipe_set_vertex_buffer;
+   softpipe->pipe.set_vertex_element = softpipe_set_vertex_element;
+
+   softpipe->pipe.draw_arrays = softpipe_draw_arrays;
+   softpipe->pipe.draw_elements = softpipe_draw_elements;
+
+   softpipe->pipe.clear = softpipe_clear;
+   softpipe->pipe.flush = softpipe_flush;
+
+   softpipe_init_query_funcs( softpipe );
+
+   /* textures */
+   softpipe->pipe.texture_create = softpipe_texture_create;
+   softpipe->pipe.texture_release = softpipe_texture_release;
+   softpipe->pipe.get_tex_surface = softpipe_get_tex_surface;
+
+   /*
+    * Alloc caches for accessing drawing surfaces and textures.
+    * Must be before quad stage setup!
+    */
+   for (i = 0; i < PIPE_MAX_COLOR_BUFS; i++)
+      softpipe->cbuf_cache[i] = sp_create_tile_cache();
+   softpipe->zsbuf_cache = sp_create_tile_cache();
+
+   for (i = 0; i < PIPE_MAX_SAMPLERS; i++)
+      softpipe->tex_cache[i] = sp_create_tile_cache();
+
+
+   /* setup quad rendering stages */
+   softpipe->quad.polygon_stipple = sp_quad_polygon_stipple_stage(softpipe);
+   softpipe->quad.earlyz = sp_quad_earlyz_stage(softpipe);
+   softpipe->quad.shade = sp_quad_shade_stage(softpipe);
+   softpipe->quad.alpha_test = sp_quad_alpha_test_stage(softpipe);
+   softpipe->quad.depth_test = sp_quad_depth_test_stage(softpipe);
+   softpipe->quad.stencil_test = sp_quad_stencil_test_stage(softpipe);
+   softpipe->quad.occlusion = sp_quad_occlusion_stage(softpipe);
+   softpipe->quad.coverage = sp_quad_coverage_stage(softpipe);
+   softpipe->quad.bufloop = sp_quad_bufloop_stage(softpipe);
+   softpipe->quad.blend = sp_quad_blend_stage(softpipe);
+   softpipe->quad.colormask = sp_quad_colormask_stage(softpipe);
+   softpipe->quad.output = sp_quad_output_stage(softpipe);
+
+   softpipe->winsys = softpipe_winsys;
+
+   /*
+    * Create drawing context and plug our rendering stage into it.
+    */
+   softpipe->draw = draw_create();
+   assert(softpipe->draw);
+   softpipe->setup = sp_draw_render_stage(softpipe);
+
+   if (GETENV( "SP_VBUF" ) != NULL) {
+      sp_init_vbuf(softpipe);
+   }
+   else {
+      draw_set_rasterize_stage(softpipe->draw, softpipe->setup);
+   }
+
+   sp_init_surface_functions(softpipe);
+
+   return &softpipe->pipe;
+}
diff --git a/src/gallium/drivers/softpipe/sp_context.h b/src/gallium/drivers/softpipe/sp_context.h
new file mode 100644
index 0000000000..aff8c2cc5d
--- /dev/null
+++ b/src/gallium/drivers/softpipe/sp_context.h
@@ -0,0 +1,152 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+/* Authors:  Keith Whitwell <keith@tungstengraphics.com>
+ */
+
+#ifndef SP_CONTEXT_H
+#define SP_CONTEXT_H
+
+#include "pipe/p_context.h"
+#include "pipe/p_defines.h"
+
+#include "pipe/draw/draw_vertex.h"
+
+#include "sp_quad.h"
+
+
+struct softpipe_winsys;
+struct softpipe_vbuf_render;
+struct draw_context;
+struct draw_stage;
+struct softpipe_tile_cache;
+struct sp_fragment_shader_state;
+struct sp_vertex_shader_state;
+
+
+struct softpipe_context {
+   struct pipe_context pipe;  /**< base class */
+   struct softpipe_winsys *winsys;	/**< window system interface */
+
+
+   /* The most recent drawing state as set by the driver:
+    */
+   const struct pipe_blend_state   *blend;
+   const struct pipe_sampler_state *sampler[PIPE_MAX_SAMPLERS];
+   const struct pipe_depth_stencil_alpha_state   *depth_stencil;
+   const struct pipe_rasterizer_state *rasterizer;
+   const struct sp_fragment_shader_state *fs;
+   const struct sp_vertex_shader_state *vs;
+
+   struct pipe_blend_color blend_color;
+   struct pipe_clip_state clip;
+   struct pipe_constant_buffer constants[2];
+   struct pipe_framebuffer_state framebuffer;
+   struct pipe_poly_stipple poly_stipple;
+   struct pipe_scissor_state scissor;
+   struct softpipe_texture *texture[PIPE_MAX_SAMPLERS];
+   struct pipe_viewport_state viewport;
+   struct pipe_vertex_buffer vertex_buffer[PIPE_ATTRIB_MAX];
+   struct pipe_vertex_element vertex_element[PIPE_ATTRIB_MAX];
+   unsigned dirty;
+
+   /* Counter for occlusion queries.  Note this supports overlapping
+    * queries.
+    */
+   uint64 occlusion_count;
+
+   /*
+    * Mapped vertex buffers
+    */
+   ubyte *mapped_vbuffer[PIPE_ATTRIB_MAX];
+   
+   /** Mapped constant buffers */
+   void *mapped_constants[PIPE_SHADER_TYPES];
+
+   /** Vertex format */
+   struct vertex_info vertex_info;
+   struct vertex_info vertex_info_vbuf;
+
+   int psize_slot;
+
+#if 0
+   /* Stipple derived state:
+    */
+   ubyte stipple_masks[16][16];
+#endif
+
+   /** Derived from scissor and surface bounds: */
+   struct pipe_scissor_state cliprect;
+
+   unsigned line_stipple_counter;
+
+   /** Software quad rendering pipeline */
+   struct {
+      struct quad_stage *polygon_stipple;
+      struct quad_stage *earlyz;
+      struct quad_stage *shade;
+      struct quad_stage *alpha_test;
+      struct quad_stage *stencil_test;
+      struct quad_stage *depth_test;
+      struct quad_stage *occlusion;
+      struct quad_stage *coverage;
+      struct quad_stage *bufloop;
+      struct quad_stage *blend;
+      struct quad_stage *colormask;
+      struct quad_stage *output;
+
+      struct quad_stage *first; /**< points to one of the above stages */
+   } quad;
+
+   /** The primitive drawing context */
+   struct draw_context *draw;
+   struct draw_stage *setup;
+   struct draw_stage *vbuf;
+   struct softpipe_vbuf_render *vbuf_render;
+
+   uint current_cbuf;      /**< current color buffer being written to */
+
+   struct softpipe_tile_cache *cbuf_cache[PIPE_MAX_COLOR_BUFS];
+   struct softpipe_tile_cache *zsbuf_cache;
+
+   struct softpipe_tile_cache *tex_cache[PIPE_MAX_SAMPLERS];
+
+   int use_sse : 1;
+   int dump_fs : 1;
+};
+
+
+
+
+static INLINE struct softpipe_context *
+softpipe_context( struct pipe_context *pipe )
+{
+   return (struct softpipe_context *)pipe;
+}
+
+
+#endif /* SP_CONTEXT_H */
diff --git a/src/gallium/drivers/softpipe/sp_draw_arrays.c b/src/gallium/drivers/softpipe/sp_draw_arrays.c
new file mode 100644
index 0000000000..71a303a8b5
--- /dev/null
+++ b/src/gallium/drivers/softpipe/sp_draw_arrays.c
@@ -0,0 +1,164 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+/* Author:
+ *    Brian Paul
+ *    Keith Whitwell
+ */
+
+
+#include "pipe/p_defines.h"
+#include "pipe/p_context.h"
+#include "pipe/p_winsys.h"
+
+#include "sp_context.h"
+#include "sp_state.h"
+
+#include "pipe/draw/draw_context.h"
+
+
+
+static void
+softpipe_map_constant_buffers(struct softpipe_context *sp)
+{
+   struct pipe_winsys *ws = sp->pipe.winsys;
+   uint i;
+   for (i = 0; i < 2; i++) {
+      if (sp->constants[i].size)
+         sp->mapped_constants[i] = ws->buffer_map(ws, sp->constants[i].buffer,
+                                                  PIPE_BUFFER_USAGE_CPU_READ);
+   }
+
+   draw_set_mapped_constant_buffer(sp->draw,
+                                   sp->mapped_constants[PIPE_SHADER_VERTEX]);
+}
+
+static void
+softpipe_unmap_constant_buffers(struct softpipe_context *sp)
+{
+   struct pipe_winsys *ws = sp->pipe.winsys;
+   uint i;
+   for (i = 0; i < 2; i++) {
+      if (sp->constants[i].size)
+         ws->buffer_unmap(ws, sp->constants[i].buffer);
+      sp->mapped_constants[i] = NULL;
+   }
+}
+
+
+boolean
+softpipe_draw_arrays(struct pipe_context *pipe, unsigned mode,
+                     unsigned start, unsigned count)
+{
+   return softpipe_draw_elements(pipe, NULL, 0, mode, start, count);
+}
+
+
+
+/**
+ * Draw vertex arrays, with optional indexing.
+ * Basically, map the vertex buffers (and drawing surfaces), then hand off
+ * the drawing to the 'draw' module.
+ *
+ * XXX should the element buffer be specified/bound with a separate function?
+ */
+boolean
+softpipe_draw_elements(struct pipe_context *pipe,
+                       struct pipe_buffer *indexBuffer,
+                       unsigned indexSize,
+                       unsigned mode, unsigned start, unsigned count)
+{
+   struct softpipe_context *sp = softpipe_context(pipe);
+   struct draw_context *draw = sp->draw;
+   unsigned i;
+
+   /* first, check that the primitive is not malformed.  It is the
+    * state tracker's responsibility to do send only correctly formed
+    * primitives down.  It currently isn't doing that though...
+    */
+#if 1
+   count = draw_trim_prim( mode, count );
+#else
+   if (!draw_validate_prim( mode, count ))
+      assert(0);
+#endif
+
+
+   if (sp->dirty)
+      softpipe_update_derived( sp );
+
+   softpipe_map_surfaces(sp);
+   softpipe_map_constant_buffers(sp);
+
+   /*
+    * Map vertex buffers
+    */
+   for (i = 0; i < PIPE_ATTRIB_MAX; i++) {
+      if (sp->vertex_buffer[i].buffer) {
+         void *buf
+            = pipe->winsys->buffer_map(pipe->winsys,
+                                       sp->vertex_buffer[i].buffer,
+                                       PIPE_BUFFER_USAGE_CPU_READ);
+         draw_set_mapped_vertex_buffer(draw, i, buf);
+      }
+   }
+   /* Map index buffer, if present */
+   if (indexBuffer) {
+      void *mapped_indexes
+         = pipe->winsys->buffer_map(pipe->winsys, indexBuffer,
+                                    PIPE_BUFFER_USAGE_CPU_READ);
+      draw_set_mapped_element_buffer(draw, indexSize, mapped_indexes);
+   }
+   else {
+      /* no index/element buffer */
+      draw_set_mapped_element_buffer(draw, 0, NULL);
+   }
+
+
+   /* draw! */
+   draw_arrays(draw, mode, start, count);
+
+   /*
+    * unmap vertex/index buffers - will cause draw module to flush
+    */
+   for (i = 0; i < PIPE_ATTRIB_MAX; i++) {
+      if (sp->vertex_buffer[i].buffer) {
+         draw_set_mapped_vertex_buffer(draw, i, NULL);
+         pipe->winsys->buffer_unmap(pipe->winsys, sp->vertex_buffer[i].buffer);
+      }
+   }
+   if (indexBuffer) {
+      draw_set_mapped_element_buffer(draw, 0, NULL);
+      pipe->winsys->buffer_unmap(pipe->winsys, indexBuffer);
+   }
+
+
+   /* Note: leave drawing surfaces mapped */
+   softpipe_unmap_constant_buffers(sp);
+
+   return TRUE;
+}
diff --git a/src/gallium/drivers/softpipe/sp_flush.c b/src/gallium/drivers/softpipe/sp_flush.c
new file mode 100644
index 0000000000..ced0d5d098
--- /dev/null
+++ b/src/gallium/drivers/softpipe/sp_flush.c
@@ -0,0 +1,76 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+/* Author:
+ *    Keith Whitwell <keith@tungstengraphics.com>
+ */
+
+
+#include "pipe/p_defines.h"
+#include "pipe/draw/draw_context.h"
+#include "sp_flush.h"
+#include "sp_context.h"
+#include "sp_surface.h"
+#include "sp_state.h"
+#include "sp_tile_cache.h"
+#include "sp_winsys.h"
+
+
+/* There will be actual work to do here.  In future we may want a
+ * fence-like interface instead of finish, and perhaps flush will take
+ * flags to indicate what type of flush is required.
+ */
+void
+softpipe_flush( struct pipe_context *pipe,
+		unsigned flags )
+{
+   struct softpipe_context *softpipe = softpipe_context(pipe);
+   uint i;
+
+   draw_flush(softpipe->draw);
+
+   /* - flush the quad pipeline
+    * - flush the texture cache
+    * - flush the render cache
+    */
+
+   for (i = 0; i < softpipe->framebuffer.num_cbufs; i++)
+      if (softpipe->cbuf_cache[i])
+         sp_flush_tile_cache(softpipe, softpipe->cbuf_cache[i]);
+
+   if (softpipe->zsbuf_cache)
+      sp_flush_tile_cache(softpipe, softpipe->zsbuf_cache);
+
+   /* Need this call for hardware buffers before swapbuffers.
+    *
+    * there should probably be another/different flush-type function
+    * that's called before swapbuffers because we don't always want
+    * to unmap surfaces when flushing.
+    */
+   softpipe_unmap_surfaces(softpipe);
+}
+
diff --git a/src/gallium/drivers/softpipe/sp_flush.h b/src/gallium/drivers/softpipe/sp_flush.h
new file mode 100644
index 0000000000..34ec617866
--- /dev/null
+++ b/src/gallium/drivers/softpipe/sp_flush.h
@@ -0,0 +1,35 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+#ifndef SP_FLUSH_H
+#define SP_FLUSH_H
+
+struct pipe_context;
+
+void softpipe_flush(struct pipe_context *pipe, unsigned flags );
+
+#endif
diff --git a/src/gallium/drivers/softpipe/sp_headers.h b/src/gallium/drivers/softpipe/sp_headers.h
new file mode 100644
index 0000000000..0ae31d8796
--- /dev/null
+++ b/src/gallium/drivers/softpipe/sp_headers.h
@@ -0,0 +1,82 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+/* Authors:  Keith Whitwell <keith@tungstengraphics.com>
+ */
+
+#ifndef SP_HEADERS_H
+#define SP_HEADERS_H
+
+#include "pipe/tgsi/exec/tgsi_exec.h"
+
+#define PRIM_POINT 1
+#define PRIM_LINE  2
+#define PRIM_TRI   3
+
+
+/* The rasterizer generates 2x2 quads of fragment and feeds them to
+ * the current fp_machine (see below).
+ * Remember that Y=0=top with Y increasing down the window.
+ */
+#define QUAD_TOP_LEFT     0
+#define QUAD_TOP_RIGHT    1
+#define QUAD_BOTTOM_LEFT  2
+#define QUAD_BOTTOM_RIGHT 3
+
+#define MASK_TOP_LEFT     (1 << QUAD_TOP_LEFT)
+#define MASK_TOP_RIGHT    (1 << QUAD_TOP_RIGHT)
+#define MASK_BOTTOM_LEFT  (1 << QUAD_BOTTOM_LEFT)
+#define MASK_BOTTOM_RIGHT (1 << QUAD_BOTTOM_RIGHT)
+#define MASK_ALL          0xf
+
+
+/**
+ * Encodes everything we need to know about a 2x2 pixel block.  Uses
+ * "Channel-Serial" or "SoA" layout.  
+ */
+struct quad_header {
+   int x0;
+   int y0;
+   unsigned mask:4;
+   unsigned facing:1;   /**< Front (0) or back (1) facing? */
+   unsigned prim:2;     /**< PRIM_POINT, LINE, TRI */
+
+   struct {
+      float color[NUM_CHANNELS][QUAD_SIZE];	/* rrrr, gggg, bbbb, aaaa */
+      float depth[QUAD_SIZE];
+   } outputs;
+
+   float coverage[QUAD_SIZE];    /** fragment coverage for antialiasing */
+
+   const struct tgsi_interp_coef *coef;
+   const struct tgsi_interp_coef *posCoef;
+
+   unsigned nr_attrs;
+};
+
+
+#endif /* SP_HEADERS_H */
diff --git a/src/gallium/drivers/softpipe/sp_prim_setup.c b/src/gallium/drivers/softpipe/sp_prim_setup.c
new file mode 100644
index 0000000000..2772048661
--- /dev/null
+++ b/src/gallium/drivers/softpipe/sp_prim_setup.c
@@ -0,0 +1,1247 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+/**
+ * \brief  Primitive rasterization/rendering (points, lines, triangles)
+ *
+ * \author  Keith Whitwell <keith@tungstengraphics.com>
+ * \author  Brian Paul
+ */
+
+
+#include "sp_context.h"
+#include "sp_headers.h"
+#include "sp_quad.h"
+#include "sp_state.h"
+#include "sp_prim_setup.h"
+#include "pipe/draw/draw_private.h"
+#include "pipe/draw/draw_vertex.h"
+#include "pipe/p_util.h"
+#include "pipe/p_shader_tokens.h"
+
+#define DEBUG_VERTS 0
+
+/**
+ * Triangle edge info
+ */
+struct edge {
+   float dx;		/**< X(v1) - X(v0), used only during setup */
+   float dy;		/**< Y(v1) - Y(v0), used only during setup */
+   float dxdy;		/**< dx/dy */
+   float sx, sy;	/**< first sample point coord */
+   int lines;		/**< number of lines on this edge */
+};
+
+
+/**
+ * Triangle setup info (derived from draw_stage).
+ * Also used for line drawing (taking some liberties).
+ */
+struct setup_stage {
+   struct draw_stage stage; /**< This must be first (base class) */
+
+   struct softpipe_context *softpipe;
+
+   /* Vertices are just an array of floats making up each attribute in
+    * turn.  Currently fixed at 4 floats, but should change in time.
+    * Codegen will help cope with this.
+    */
+   const struct vertex_header *vmax;
+   const struct vertex_header *vmid;
+   const struct vertex_header *vmin;
+   const struct vertex_header *vprovoke;
+
+   struct edge ebot;
+   struct edge etop;
+   struct edge emaj;
+
+   float oneoverarea;
+
+   struct tgsi_interp_coef coef[PIPE_MAX_SHADER_INPUTS];
+   struct tgsi_interp_coef posCoef;  /* For Z, W */
+   struct quad_header quad; 
+
+   struct {
+      int left[2];   /**< [0] = row0, [1] = row1 */
+      int right[2];
+      int y;
+      unsigned y_flags;
+      unsigned mask;     /**< mask of MASK_BOTTOM/TOP_LEFT/RIGHT bits */
+   } span;
+};
+
+
+
+/**
+ * Basically a cast wrapper.
+ */
+static INLINE struct setup_stage *setup_stage( struct draw_stage *stage )
+{
+   return (struct setup_stage *)stage;
+}
+
+
+/**
+ * Clip setup->quad against the scissor/surface bounds.
+ */
+static INLINE void
+quad_clip(struct setup_stage *setup)
+{
+   const struct pipe_scissor_state *cliprect = &setup->softpipe->cliprect;
+   const int minx = (int) cliprect->minx;
+   const int maxx = (int) cliprect->maxx;
+   const int miny = (int) cliprect->miny;
+   const int maxy = (int) cliprect->maxy;
+
+   if (setup->quad.x0 >= maxx ||
+       setup->quad.y0 >= maxy ||
+       setup->quad.x0 + 1 < minx ||
+       setup->quad.y0 + 1 < miny) {
+      /* totally clipped */
+      setup->quad.mask = 0x0;
+      return;
+   }
+   if (setup->quad.x0 < minx)
+      setup->quad.mask &= (MASK_BOTTOM_RIGHT | MASK_TOP_RIGHT);
+   if (setup->quad.y0 < miny)
+      setup->quad.mask &= (MASK_BOTTOM_LEFT | MASK_BOTTOM_RIGHT);
+   if (setup->quad.x0 == maxx - 1)
+      setup->quad.mask &= (MASK_BOTTOM_LEFT | MASK_TOP_LEFT);
+   if (setup->quad.y0 == maxy - 1)
+      setup->quad.mask &= (MASK_TOP_LEFT | MASK_TOP_RIGHT);
+}
+
+
+/**
+ * Emit a quad (pass to next stage) with clipping.
+ */
+static INLINE void
+clip_emit_quad(struct setup_stage *setup)
+{
+   quad_clip(setup);
+   if (setup->quad.mask) {
+      struct softpipe_context *sp = setup->softpipe;
+      sp->quad.first->run(sp->quad.first, &setup->quad);
+   }
+}
+
+
+/**
+ * Emit a quad (pass to next stage).  No clipping is done.
+ */
+static INLINE void
+emit_quad( struct setup_stage *setup, int x, int y, unsigned mask )
+{
+   struct softpipe_context *sp = setup->softpipe;
+   setup->quad.x0 = x;
+   setup->quad.y0 = y;
+   setup->quad.mask = mask;
+   sp->quad.first->run(sp->quad.first, &setup->quad);
+}
+
+
+/**
+ * Given an X or Y coordinate, return the block/quad coordinate that it
+ * belongs to.
+ */
+static INLINE int block( int x )
+{
+   return x & ~1;
+}
+
+
+/**
+ * Compute mask which indicates which pixels in the 2x2 quad are actually inside
+ * the triangle's bounds.
+ *
+ * this is pretty nasty...  may need to rework flush_spans again to
+ * fix it, if possible.
+ */
+static unsigned calculate_mask( struct setup_stage *setup, int x )
+{
+   unsigned mask = 0x0;
+
+   if (x >= setup->span.left[0] && x < setup->span.right[0]) 
+      mask |= MASK_TOP_LEFT;
+
+   if (x >= setup->span.left[1] && x < setup->span.right[1]) 
+      mask |= MASK_BOTTOM_LEFT;
+      
+   if (x+1 >= setup->span.left[0] && x+1 < setup->span.right[0]) 
+      mask |= MASK_TOP_RIGHT;
+
+   if (x+1 >= setup->span.left[1] && x+1 < setup->span.right[1]) 
+      mask |= MASK_BOTTOM_RIGHT;
+
+   return mask;
+}
+
+
+/**
+ * Render a horizontal span of quads
+ */
+static void flush_spans( struct setup_stage *setup )
+{
+   int minleft, maxright;
+   int x;
+
+   switch (setup->span.y_flags) {
+   case 0x3:
+      /* both odd and even lines written (both quad rows) */
+      minleft = MIN2(setup->span.left[0], setup->span.left[1]);
+      maxright = MAX2(setup->span.right[0], setup->span.right[1]);
+      break;
+
+   case 0x1:
+      /* only even line written (quad top row) */
+      minleft = setup->span.left[0];
+      maxright = setup->span.right[0];
+      break;
+
+   case 0x2:
+      /* only odd line written (quad bottom row) */
+      minleft = setup->span.left[1];
+      maxright = setup->span.right[1];
+      break;
+
+   default:
+      return;
+   }
+
+   /* XXX this loop could be moved into the above switch cases and
+    * calculate_mask() could be simplified a bit...
+    */
+   for (x = block(minleft); x <= block(maxright); x += 2) {
+      emit_quad( setup, x, setup->span.y, 
+                 calculate_mask( setup, x ) );
+   }
+
+   setup->span.y = 0;
+   setup->span.y_flags = 0;
+   setup->span.right[0] = 0;
+   setup->span.right[1] = 0;
+}
+
+#if DEBUG_VERTS
+static void print_vertex(const struct setup_stage *setup,
+                         const struct vertex_header *v)
+{
+   int i;
+   debug_printf("Vertex: (%p)\n", v);
+   for (i = 0; i < setup->quad.nr_attrs; i++) {
+      debug_printf("  %d: %f %f %f %f\n",  i, 
+              v->data[i][0], v->data[i][1], v->data[i][2], v->data[i][3]);
+   }
+}
+#endif
+
+static boolean setup_sort_vertices( struct setup_stage *setup,
+				      const struct prim_header *prim )
+{
+   const struct vertex_header *v0 = prim->v[0];
+   const struct vertex_header *v1 = prim->v[1];
+   const struct vertex_header *v2 = prim->v[2];
+
+#if DEBUG_VERTS
+   debug_printf("Triangle:\n");
+   print_vertex(setup, v0);
+   print_vertex(setup, v1);
+   print_vertex(setup, v2);
+#endif
+
+   setup->vprovoke = v2;
+
+   /* determine bottom to top order of vertices */
+   {
+      float y0 = v0->data[0][1];
+      float y1 = v1->data[0][1];
+      float y2 = v2->data[0][1];
+      if (y0 <= y1) {
+	 if (y1 <= y2) {
+	    /* y0<=y1<=y2 */
+	    setup->vmin = v0;   
+	    setup->vmid = v1;   
+	    setup->vmax = v2;
+	 }
+	 else if (y2 <= y0) {
+	    /* y2<=y0<=y1 */
+	    setup->vmin = v2;   
+	    setup->vmid = v0;   
+	    setup->vmax = v1;   
+	 }
+	 else {
+	    /* y0<=y2<=y1 */
+	    setup->vmin = v0;   
+	    setup->vmid = v2;   
+	    setup->vmax = v1;  
+	 }
+      }
+      else {
+	 if (y0 <= y2) {
+	    /* y1<=y0<=y2 */
+	    setup->vmin = v1;   
+	    setup->vmid = v0;   
+	    setup->vmax = v2;  
+	 }
+	 else if (y2 <= y1) {
+	    /* y2<=y1<=y0 */
+	    setup->vmin = v2;   
+	    setup->vmid = v1;   
+	    setup->vmax = v0;  
+	 }
+	 else {
+	    /* y1<=y2<=y0 */
+	    setup->vmin = v1;   
+	    setup->vmid = v2;   
+	    setup->vmax = v0;
+	 }
+      }
+   }
+
+   setup->ebot.dx = setup->vmid->data[0][0] - setup->vmin->data[0][0];
+   setup->ebot.dy = setup->vmid->data[0][1] - setup->vmin->data[0][1];
+   setup->emaj.dx = setup->vmax->data[0][0] - setup->vmin->data[0][0];
+   setup->emaj.dy = setup->vmax->data[0][1] - setup->vmin->data[0][1];
+   setup->etop.dx = setup->vmax->data[0][0] - setup->vmid->data[0][0];
+   setup->etop.dy = setup->vmax->data[0][1] - setup->vmid->data[0][1];
+
+   /*
+    * Compute triangle's area.  Use 1/area to compute partial
+    * derivatives of attributes later.
+    *
+    * The area will be the same as prim->det, but the sign may be
+    * different depending on how the vertices get sorted above.
+    *
+    * To determine whether the primitive is front or back facing we
+    * use the prim->det value because its sign is correct.
+    */
+   {
+      const float area = (setup->emaj.dx * setup->ebot.dy - 
+			    setup->ebot.dx * setup->emaj.dy);
+
+      setup->oneoverarea = 1.0f / area;
+      /*
+      debug_printf("%s one-over-area %f  area %f  det %f\n",
+                   __FUNCTION__, setup->oneoverarea, area, prim->det );
+      */
+   }
+
+   /* We need to know if this is a front or back-facing triangle for:
+    *  - the GLSL gl_FrontFacing fragment attribute (bool)
+    *  - two-sided stencil test
+    */
+   setup->quad.facing = (prim->det > 0.0) ^ (setup->softpipe->rasterizer->front_winding == PIPE_WINDING_CW);
+
+   return TRUE;
+}
+
+
+/**
+ * Compute a0 for a constant-valued coefficient (GL_FLAT shading).
+ * The value value comes from vertex->data[slot][i].
+ * The result will be put into setup->coef[slot].a0[i].
+ * \param slot  which attribute slot 
+ * \param i  which component of the slot (0..3)
+ */
+static void const_coeff( struct setup_stage *setup,
+                         struct tgsi_interp_coef *coef,
+                         uint vertSlot, uint i)
+{
+   assert(i <= 3);
+
+   coef->dadx[i] = 0;
+   coef->dady[i] = 0;
+
+   /* need provoking vertex info!
+    */
+   coef->a0[i] = setup->vprovoke->data[vertSlot][i];
+}
+
+
+/**
+ * Compute a0, dadx and dady for a linearly interpolated coefficient,
+ * for a triangle.
+ */
+static void tri_linear_coeff( struct setup_stage *setup,
+                              struct tgsi_interp_coef *coef,
+                              uint vertSlot, uint i)
+{
+   float botda = setup->vmid->data[vertSlot][i] - setup->vmin->data[vertSlot][i];
+   float majda = setup->vmax->data[vertSlot][i] - setup->vmin->data[vertSlot][i];
+   float a = setup->ebot.dy * majda - botda * setup->emaj.dy;
+   float b = setup->emaj.dx * botda - majda * setup->ebot.dx;
+   float dadx = a * setup->oneoverarea;
+   float dady = b * setup->oneoverarea;
+
+   assert(i <= 3);
+
+   coef->dadx[i] = dadx;
+   coef->dady[i] = dady;
+
+   /* calculate a0 as the value which would be sampled for the
+    * fragment at (0,0), taking into account that we want to sample at
+    * pixel centers, in other words (0.5, 0.5).
+    *
+    * this is neat but unfortunately not a good way to do things for
+    * triangles with very large values of dadx or dady as it will
+    * result in the subtraction and re-addition from a0 of a very
+    * large number, which means we'll end up loosing a lot of the
+    * fractional bits and precision from a0.  the way to fix this is
+    * to define a0 as the sample at a pixel center somewhere near vmin
+    * instead - i'll switch to this later.
+    */
+   coef->a0[i] = (setup->vmin->data[vertSlot][i] - 
+                  (dadx * (setup->vmin->data[0][0] - 0.5f) + 
+                   dady * (setup->vmin->data[0][1] - 0.5f)));
+
+   /*
+   debug_printf("attr[%d].%c: %f dx:%f dy:%f\n",
+		slot, "xyzw"[i], 
+		setup->coef[slot].a0[i],
+		setup->coef[slot].dadx[i],
+		setup->coef[slot].dady[i]);
+   */
+}
+
+
+/**
+ * Compute a0, dadx and dady for a perspective-corrected interpolant,
+ * for a triangle.
+ * We basically multiply the vertex value by 1/w before computing
+ * the plane coefficients (a0, dadx, dady).
+ * Later, when we compute the value at a particular fragment position we'll
+ * divide the interpolated value by the interpolated W at that fragment.
+ */
+static void tri_persp_coeff( struct setup_stage *setup,
+                             struct tgsi_interp_coef *coef,
+                             uint vertSlot, uint i)
+{
+   /* premultiply by 1/w  (v->data[0][3] is always W):
+    */
+   float mina = setup->vmin->data[vertSlot][i] * setup->vmin->data[0][3];
+   float mida = setup->vmid->data[vertSlot][i] * setup->vmid->data[0][3];
+   float maxa = setup->vmax->data[vertSlot][i] * setup->vmax->data[0][3];
+   float botda = mida - mina;
+   float majda = maxa - mina;
+   float a = setup->ebot.dy * majda - botda * setup->emaj.dy;
+   float b = setup->emaj.dx * botda - majda * setup->ebot.dx;
+   float dadx = a * setup->oneoverarea;
+   float dady = b * setup->oneoverarea;
+      
+   /*
+   debug_printf("tri persp %d,%d: %f %f %f\n", vertSlot, i,
+          	setup->vmin->data[vertSlot][i],
+          	setup->vmid->data[vertSlot][i],
+       		setup->vmax->data[vertSlot][i]
+          );
+   */
+   assert(i <= 3);
+
+   coef->dadx[i] = dadx;
+   coef->dady[i] = dady;
+   coef->a0[i] = (mina - 
+                  (dadx * (setup->vmin->data[0][0] - 0.5f) + 
+                   dady * (setup->vmin->data[0][1] - 0.5f)));
+}
+
+
+/**
+ * Special coefficient setup for gl_FragCoord.
+ * X and Y are trivial, though Y has to be inverted for OpenGL.
+ * Z and W are copied from posCoef which should have already been computed.
+ * We could do a bit less work if we'd examine gl_FragCoord's swizzle mask.
+ */
+static void
+setup_fragcoord_coeff(struct setup_stage *setup)
+{
+   /*X*/
+   setup->coef[0].a0[0] = 0;
+   setup->coef[0].dadx[0] = 1.0;
+   setup->coef[0].dady[0] = 0.0;
+   /*Y*/
+   if (setup->softpipe->rasterizer->origin_lower_left) {
+      /* y=0=bottom */
+      const int winHeight = setup->softpipe->framebuffer.cbufs[0]->height;
+      setup->coef[0].a0[1] = (float) (winHeight - 1);
+      setup->coef[0].dady[1] = -1.0;
+   }
+   else {
+      /* y=0=top */
+      setup->coef[0].a0[1] = 0.0;
+      setup->coef[0].dady[1] = 1.0;
+   }
+   setup->coef[0].dadx[1] = 0.0;
+   /*Z*/
+   setup->coef[0].a0[2] = setup->posCoef.a0[2];
+   setup->coef[0].dadx[2] = setup->posCoef.dadx[2];
+   setup->coef[0].dady[2] = setup->posCoef.dady[2];
+   /*W*/
+   setup->coef[0].a0[3] = setup->posCoef.a0[3];
+   setup->coef[0].dadx[3] = setup->posCoef.dadx[3];
+   setup->coef[0].dady[3] = setup->posCoef.dady[3];
+}
+
+
+
+/**
+ * Compute the setup->coef[] array dadx, dady, a0 values.
+ * Must be called after setup->vmin,vmid,vmax,vprovoke are initialized.
+ */
+static void setup_tri_coefficients( struct setup_stage *setup )
+{
+   struct softpipe_context *softpipe = setup->softpipe;
+   const struct pipe_shader_state *fs = &softpipe->fs->shader;
+   const struct vertex_info *vinfo = softpipe_get_vertex_info(softpipe);
+   uint fragSlot;
+
+   /* z and w are done by linear interpolation:
+    */
+   tri_linear_coeff(setup, &setup->posCoef, 0, 2);
+   tri_linear_coeff(setup, &setup->posCoef, 0, 3);
+
+   /* setup interpolation for all the remaining attributes:
+    */
+   for (fragSlot = 0; fragSlot < fs->num_inputs; fragSlot++) {
+      const uint vertSlot = vinfo->src_index[fragSlot];
+      uint j;
+
+      switch (vinfo->interp_mode[fragSlot]) {
+      case INTERP_CONSTANT:
+         for (j = 0; j < NUM_CHANNELS; j++)
+            const_coeff(setup, &setup->coef[fragSlot], vertSlot, j);
+         break;
+      case INTERP_LINEAR:
+         for (j = 0; j < NUM_CHANNELS; j++)
+            tri_linear_coeff(setup, &setup->coef[fragSlot], vertSlot, j);
+         break;
+      case INTERP_PERSPECTIVE:
+         for (j = 0; j < NUM_CHANNELS; j++)
+            tri_persp_coeff(setup, &setup->coef[fragSlot], vertSlot, j);
+         break;
+      case INTERP_POS:
+         assert(fragSlot == 0);
+         setup_fragcoord_coeff(setup);
+         break;
+      default:
+         assert(0);
+      }
+
+      if (fs->input_semantic_name[fragSlot] == TGSI_SEMANTIC_FOG) {
+         /* FOG.y = front/back facing  XXX fix this */
+         setup->coef[fragSlot].a0[1] = 1.0f - setup->quad.facing;
+         setup->coef[fragSlot].dadx[1] = 0.0;
+         setup->coef[fragSlot].dady[1] = 0.0;
+      }
+   }
+}
+
+
+
+static void setup_tri_edges( struct setup_stage *setup )
+{
+   float vmin_x = setup->vmin->data[0][0] + 0.5f;
+   float vmid_x = setup->vmid->data[0][0] + 0.5f;
+
+   float vmin_y = setup->vmin->data[0][1] - 0.5f;
+   float vmid_y = setup->vmid->data[0][1] - 0.5f;
+   float vmax_y = setup->vmax->data[0][1] - 0.5f;
+
+   setup->emaj.sy = CEILF(vmin_y);
+   setup->emaj.lines = (int) CEILF(vmax_y - setup->emaj.sy);
+   setup->emaj.dxdy = setup->emaj.dx / setup->emaj.dy;
+   setup->emaj.sx = vmin_x + (setup->emaj.sy - vmin_y) * setup->emaj.dxdy;
+
+   setup->etop.sy = CEILF(vmid_y);
+   setup->etop.lines = (int) CEILF(vmax_y - setup->etop.sy);
+   setup->etop.dxdy = setup->etop.dx / setup->etop.dy;
+   setup->etop.sx = vmid_x + (setup->etop.sy - vmid_y) * setup->etop.dxdy;
+
+   setup->ebot.sy = CEILF(vmin_y);
+   setup->ebot.lines = (int) CEILF(vmid_y - setup->ebot.sy);
+   setup->ebot.dxdy = setup->ebot.dx / setup->ebot.dy;
+   setup->ebot.sx = vmin_x + (setup->ebot.sy - vmin_y) * setup->ebot.dxdy;
+}
+
+
+/**
+ * Render the upper or lower half of a triangle.
+ * Scissoring/cliprect is applied here too.
+ */
+static void subtriangle( struct setup_stage *setup,
+			 struct edge *eleft,
+			 struct edge *eright,
+			 unsigned lines )
+{
+   const struct pipe_scissor_state *cliprect = &setup->softpipe->cliprect;
+   const int minx = (int) cliprect->minx;
+   const int maxx = (int) cliprect->maxx;
+   const int miny = (int) cliprect->miny;
+   const int maxy = (int) cliprect->maxy;
+   int y, start_y, finish_y;
+   int sy = (int)eleft->sy;
+
+   assert((int)eleft->sy == (int) eright->sy);
+
+   /* clip top/bottom */
+   start_y = sy;
+   finish_y = sy + lines;
+
+   if (start_y < miny)
+      start_y = miny;
+
+   if (finish_y > maxy)
+      finish_y = maxy;
+
+   start_y -= sy;
+   finish_y -= sy;
+
+   /*
+   debug_printf("%s %d %d\n", __FUNCTION__, start_y, finish_y);  
+   */
+
+   for (y = start_y; y < finish_y; y++) {
+
+      /* avoid accumulating adds as floats don't have the precision to
+       * accurately iterate large triangle edges that way.  luckily we
+       * can just multiply these days.
+       *
+       * this is all drowned out by the attribute interpolation anyway.
+       */
+      int left = (int)(eleft->sx + y * eleft->dxdy);
+      int right = (int)(eright->sx + y * eright->dxdy);
+
+      /* clip left/right */
+      if (left < minx)
+         left = minx;
+      if (right > maxx)
+         right = maxx;
+
+      if (left < right) {
+         int _y = sy + y;
+         if (block(_y) != setup->span.y) {
+            flush_spans(setup);
+            setup->span.y = block(_y);
+         }
+
+         setup->span.left[_y&1] = left;
+         setup->span.right[_y&1] = right;
+         setup->span.y_flags |= 1<<(_y&1);
+      }
+   }
+
+
+   /* save the values so that emaj can be restarted:
+    */
+   eleft->sx += lines * eleft->dxdy;
+   eright->sx += lines * eright->dxdy;
+   eleft->sy += lines;
+   eright->sy += lines;
+}
+
+
+/**
+ * Do setup for triangle rasterization, then render the triangle.
+ */
+static void setup_tri( struct draw_stage *stage,
+		       struct prim_header *prim )
+{
+   struct setup_stage *setup = setup_stage( stage );
+
+   /*
+   debug_printf("%s\n", __FUNCTION__ );
+   */
+
+   setup_sort_vertices( setup, prim );
+   setup_tri_coefficients( setup );
+   setup_tri_edges( setup );
+
+   setup->quad.prim = PRIM_TRI;
+
+   setup->span.y = 0;
+   setup->span.y_flags = 0;
+   setup->span.right[0] = 0;
+   setup->span.right[1] = 0;
+   /*   setup->span.z_mode = tri_z_mode( setup->ctx ); */
+
+   /*   init_constant_attribs( setup ); */
+      
+   if (setup->oneoverarea < 0.0) {
+      /* emaj on left:
+       */
+      subtriangle( setup, &setup->emaj, &setup->ebot, setup->ebot.lines );
+      subtriangle( setup, &setup->emaj, &setup->etop, setup->etop.lines );
+   }
+   else {
+      /* emaj on right:
+       */
+      subtriangle( setup, &setup->ebot, &setup->emaj, setup->ebot.lines );
+      subtriangle( setup, &setup->etop, &setup->emaj, setup->etop.lines );
+   }
+
+   flush_spans( setup );
+}
+
+
+
+/**
+ * Compute a0, dadx and dady for a linearly interpolated coefficient,
+ * for a line.
+ */
+static void
+line_linear_coeff(struct setup_stage *setup,
+                  struct tgsi_interp_coef *coef,
+                  uint vertSlot, uint i)
+{
+   const float da = setup->vmax->data[vertSlot][i] - setup->vmin->data[vertSlot][i];
+   const float dadx = da * setup->emaj.dx * setup->oneoverarea;
+   const float dady = da * setup->emaj.dy * setup->oneoverarea;
+   coef->dadx[i] = dadx;
+   coef->dady[i] = dady;
+   coef->a0[i] = (setup->vmin->data[vertSlot][i] - 
+                  (dadx * (setup->vmin->data[0][0] - 0.5f) + 
+                   dady * (setup->vmin->data[0][1] - 0.5f)));
+}
+
+
+/**
+ * Compute a0, dadx and dady for a perspective-corrected interpolant,
+ * for a line.
+ */
+static void
+line_persp_coeff(struct setup_stage *setup,
+                  struct tgsi_interp_coef *coef,
+                  uint vertSlot, uint i)
+{
+   /* XXX double-check/verify this arithmetic */
+   const float a0 = setup->vmin->data[vertSlot][i] * setup->vmin->data[0][3];
+   const float a1 = setup->vmax->data[vertSlot][i] * setup->vmax->data[0][3];
+   const float da = a1 - a0;
+   const float dadx = da * setup->emaj.dx * setup->oneoverarea;
+   const float dady = da * setup->emaj.dy * setup->oneoverarea;
+   coef->dadx[i] = dadx;
+   coef->dady[i] = dady;
+   coef->a0[i] = (setup->vmin->data[vertSlot][i] - 
+                  (dadx * (setup->vmin->data[0][0] - 0.5f) + 
+                   dady * (setup->vmin->data[0][1] - 0.5f)));
+}
+
+
+/**
+ * Compute the setup->coef[] array dadx, dady, a0 values.
+ * Must be called after setup->vmin,vmax are initialized.
+ */
+static INLINE void
+setup_line_coefficients(struct setup_stage *setup, struct prim_header *prim)
+{
+   struct softpipe_context *softpipe = setup->softpipe;
+   const struct pipe_shader_state *fs = &setup->softpipe->fs->shader;
+   const struct vertex_info *vinfo = softpipe_get_vertex_info(softpipe);
+   uint fragSlot;
+
+   /* use setup->vmin, vmax to point to vertices */
+   setup->vprovoke = prim->v[1];
+   setup->vmin = prim->v[0];
+   setup->vmax = prim->v[1];
+
+   setup->emaj.dx = setup->vmax->data[0][0] - setup->vmin->data[0][0];
+   setup->emaj.dy = setup->vmax->data[0][1] - setup->vmin->data[0][1];
+   /* NOTE: this is not really 1/area */
+   setup->oneoverarea = 1.0f / (setup->emaj.dx * setup->emaj.dx +
+                                setup->emaj.dy * setup->emaj.dy);
+
+   /* z and w are done by linear interpolation:
+    */
+   line_linear_coeff(setup, &setup->posCoef, 0, 2);
+   line_linear_coeff(setup, &setup->posCoef, 0, 3);
+
+   /* setup interpolation for all the remaining attributes:
+    */
+   for (fragSlot = 0; fragSlot < fs->num_inputs; fragSlot++) {
+      const uint vertSlot = vinfo->src_index[fragSlot];
+      uint j;
+
+      switch (vinfo->interp_mode[fragSlot]) {
+      case INTERP_CONSTANT:
+         for (j = 0; j < NUM_CHANNELS; j++)
+            const_coeff(setup, &setup->coef[fragSlot], vertSlot, j);
+         break;
+      case INTERP_LINEAR:
+         for (j = 0; j < NUM_CHANNELS; j++)
+            line_linear_coeff(setup, &setup->coef[fragSlot], vertSlot, j);
+         break;
+      case INTERP_PERSPECTIVE:
+         for (j = 0; j < NUM_CHANNELS; j++)
+            line_persp_coeff(setup, &setup->coef[fragSlot], vertSlot, j);
+         break;
+      case INTERP_POS:
+         assert(fragSlot == 0);
+         assert(0); /* XXX fix this: */
+         setup_fragcoord_coeff(setup);
+         break;
+      default:
+         assert(0);
+      }
+
+      if (fs->input_semantic_name[fragSlot] == TGSI_SEMANTIC_FOG) {
+         /* FOG.y = front/back facing  XXX fix this */
+         setup->coef[fragSlot].a0[1] = 1.0f - setup->quad.facing;
+         setup->coef[fragSlot].dadx[1] = 0.0;
+         setup->coef[fragSlot].dady[1] = 0.0;
+      }
+   }
+}
+
+
+/**
+ * Plot a pixel in a line segment.
+ */
+static INLINE void
+plot(struct setup_stage *setup, int x, int y)
+{
+   const int iy = y & 1;
+   const int ix = x & 1;
+   const int quadX = x - ix;
+   const int quadY = y - iy;
+   const int mask = (1 << ix) << (2 * iy);
+
+   if (quadX != setup->quad.x0 || 
+       quadY != setup->quad.y0) 
+   {
+      /* flush prev quad, start new quad */
+
+      if (setup->quad.x0 != -1)
+         clip_emit_quad(setup);
+
+      setup->quad.x0 = quadX;
+      setup->quad.y0 = quadY;
+      setup->quad.mask = 0x0;
+   }
+
+   setup->quad.mask |= mask;
+}
+
+
+/**
+ * Do setup for line rasterization, then render the line.
+ * Single-pixel width, no stipple, etc.  We rely on the 'draw' module
+ * to handle stippling and wide lines.
+ */
+static void
+setup_line(struct draw_stage *stage, struct prim_header *prim)
+{
+   const struct vertex_header *v0 = prim->v[0];
+   const struct vertex_header *v1 = prim->v[1];
+   struct setup_stage *setup = setup_stage( stage );
+   int x0 = (int) v0->data[0][0];
+   int x1 = (int) v1->data[0][0];
+   int y0 = (int) v0->data[0][1];
+   int y1 = (int) v1->data[0][1];
+   int dx = x1 - x0;
+   int dy = y1 - y0;
+   int xstep, ystep;
+
+   if (dx == 0 && dy == 0)
+      return;
+
+   setup_line_coefficients(setup, prim);
+
+   if (dx < 0) {
+      dx = -dx;   /* make positive */
+      xstep = -1;
+   }
+   else {
+      xstep = 1;
+   }
+
+   if (dy < 0) {
+      dy = -dy;   /* make positive */
+      ystep = -1;
+   }
+   else {
+      ystep = 1;
+   }
+
+   assert(dx >= 0);
+   assert(dy >= 0);
+
+   setup->quad.x0 = setup->quad.y0 = -1;
+   setup->quad.mask = 0x0;
+   setup->quad.prim = PRIM_LINE;
+   /* XXX temporary: set coverage to 1.0 so the line appears
+    * if AA mode happens to be enabled.
+    */
+   setup->quad.coverage[0] =
+   setup->quad.coverage[1] =
+   setup->quad.coverage[2] =
+   setup->quad.coverage[3] = 1.0;
+
+   if (dx > dy) {
+      /*** X-major line ***/
+      int i;
+      const int errorInc = dy + dy;
+      int error = errorInc - dx;
+      const int errorDec = error - dx;
+
+      for (i = 0; i < dx; i++) {
+         plot(setup, x0, y0);
+
+         x0 += xstep;
+         if (error < 0) {
+            error += errorInc;
+         }
+         else {
+            error += errorDec;
+            y0 += ystep;
+         }
+      }
+   }
+   else {
+      /*** Y-major line ***/
+      int i;
+      const int errorInc = dx + dx;
+      int error = errorInc - dy;
+      const int errorDec = error - dy;
+
+      for (i = 0; i < dy; i++) {
+         plot(setup, x0, y0);
+
+         y0 += ystep;
+         if (error < 0) {
+            error += errorInc;
+         }
+         else {
+            error += errorDec;
+            x0 += xstep;
+         }
+      }
+   }
+
+   /* draw final quad */
+   if (setup->quad.mask) {
+      clip_emit_quad(setup);
+   }
+}
+
+
+static void
+point_persp_coeff(struct setup_stage *setup,
+                  const struct vertex_header *vert,
+                  struct tgsi_interp_coef *coef,
+                  uint vertSlot, uint i)
+{
+   assert(i <= 3);
+   coef->dadx[i] = 0.0F;
+   coef->dady[i] = 0.0F;
+   coef->a0[i] = vert->data[vertSlot][i] * vert->data[0][3];
+}
+
+
+/**
+ * Do setup for point rasterization, then render the point.
+ * Round or square points...
+ * XXX could optimize a lot for 1-pixel points.
+ */
+static void
+setup_point(struct draw_stage *stage, struct prim_header *prim)
+{
+   struct setup_stage *setup = setup_stage( stage );
+   struct softpipe_context *softpipe = setup->softpipe;
+   const struct pipe_shader_state *fs = &softpipe->fs->shader;
+   const struct vertex_header *v0 = prim->v[0];
+   const int sizeAttr = setup->softpipe->psize_slot;
+   const float size
+      = sizeAttr > 0 ? v0->data[sizeAttr][0]
+      : setup->softpipe->rasterizer->point_size;
+   const float halfSize = 0.5F * size;
+   const boolean round = (boolean) setup->softpipe->rasterizer->point_smooth;
+   const float x = v0->data[0][0];  /* Note: data[0] is always position */
+   const float y = v0->data[0][1];
+   const struct vertex_info *vinfo = softpipe_get_vertex_info(softpipe);
+   uint fragSlot;
+
+   /* For points, all interpolants are constant-valued.
+    * However, for point sprites, we'll need to setup texcoords appropriately.
+    * XXX: which coefficients are the texcoords???
+    * We may do point sprites as textured quads...
+    *
+    * KW: We don't know which coefficients are texcoords - ultimately
+    * the choice of what interpolation mode to use for each attribute
+    * should be determined by the fragment program, using
+    * per-attribute declaration statements that include interpolation
+    * mode as a parameter.  So either the fragment program will have
+    * to be adjusted for pointsprite vs normal point behaviour, or
+    * otherwise a special interpolation mode will have to be defined
+    * which matches the required behaviour for point sprites.  But -
+    * the latter is not a feature of normal hardware, and as such
+    * probably should be ruled out on that basis.
+    */
+   setup->vprovoke = prim->v[0];
+
+   /* setup Z, W */
+   const_coeff(setup, &setup->posCoef, 0, 2);
+   const_coeff(setup, &setup->posCoef, 0, 3);
+
+   for (fragSlot = 0; fragSlot < fs->num_inputs; fragSlot++) {
+      const uint vertSlot = vinfo->src_index[fragSlot];
+      uint j;
+
+      switch (vinfo->interp_mode[fragSlot]) {
+      case INTERP_CONSTANT:
+         /* fall-through */
+      case INTERP_LINEAR:
+         for (j = 0; j < NUM_CHANNELS; j++)
+            const_coeff(setup, &setup->coef[fragSlot], vertSlot, j);
+         break;
+      case INTERP_PERSPECTIVE:
+         for (j = 0; j < NUM_CHANNELS; j++)
+            point_persp_coeff(setup, setup->vprovoke,
+                              &setup->coef[fragSlot], vertSlot, j);
+         break;
+      case INTERP_POS:
+         assert(fragSlot == 0);
+         assert(0); /* XXX fix this: */
+         setup_fragcoord_coeff(setup);
+         break;
+      default:
+         assert(0);
+      }
+
+      if (fs->input_semantic_name[fragSlot] == TGSI_SEMANTIC_FOG) {
+         /* FOG.y = front/back facing  XXX fix this */
+         setup->coef[fragSlot].a0[1] = 1.0f - setup->quad.facing;
+         setup->coef[fragSlot].dadx[1] = 0.0;
+         setup->coef[fragSlot].dady[1] = 0.0;
+      }
+   }
+
+   setup->quad.prim = PRIM_POINT;
+
+   if (halfSize <= 0.5 && !round) {
+      /* special case for 1-pixel points */
+      const int ix = ((int) x) & 1;
+      const int iy = ((int) y) & 1;
+      setup->quad.x0 = (int) x - ix;
+      setup->quad.y0 = (int) y - iy;
+      setup->quad.mask = (1 << ix) << (2 * iy);
+      clip_emit_quad(setup);
+   }
+   else {
+      if (round) {
+         /* rounded points */
+         const int ixmin = block((int) (x - halfSize));
+         const int ixmax = block((int) (x + halfSize));
+         const int iymin = block((int) (y - halfSize));
+         const int iymax = block((int) (y + halfSize));
+         const float rmin = halfSize - 0.7071F;  /* 0.7071 = sqrt(2)/2 */
+         const float rmax = halfSize + 0.7071F;
+         const float rmin2 = MAX2(0.0F, rmin * rmin);
+         const float rmax2 = rmax * rmax;
+         const float cscale = 1.0F / (rmax2 - rmin2);
+         int ix, iy;
+
+         for (iy = iymin; iy <= iymax; iy += 2) {
+            for (ix = ixmin; ix <= ixmax; ix += 2) {
+               float dx, dy, dist2, cover;
+
+               setup->quad.mask = 0x0;
+
+               dx = (ix + 0.5f) - x;
+               dy = (iy + 0.5f) - y;
+               dist2 = dx * dx + dy * dy;
+               if (dist2 <= rmax2) {
+                  cover = 1.0F - (dist2 - rmin2) * cscale;
+                  setup->quad.coverage[QUAD_TOP_LEFT] = MIN2(cover, 1.0f);
+                  setup->quad.mask |= MASK_TOP_LEFT;
+               }
+
+               dx = (ix + 1.5f) - x;
+               dy = (iy + 0.5f) - y;
+               dist2 = dx * dx + dy * dy;
+               if (dist2 <= rmax2) {
+                  cover = 1.0F - (dist2 - rmin2) * cscale;
+                  setup->quad.coverage[QUAD_TOP_RIGHT] = MIN2(cover, 1.0f);
+                  setup->quad.mask |= MASK_TOP_RIGHT;
+               }
+
+               dx = (ix + 0.5f) - x;
+               dy = (iy + 1.5f) - y;
+               dist2 = dx * dx + dy * dy;
+               if (dist2 <= rmax2) {
+                  cover = 1.0F - (dist2 - rmin2) * cscale;
+                  setup->quad.coverage[QUAD_BOTTOM_LEFT] = MIN2(cover, 1.0f);
+                  setup->quad.mask |= MASK_BOTTOM_LEFT;
+               }
+
+               dx = (ix + 1.5f) - x;
+               dy = (iy + 1.5f) - y;
+               dist2 = dx * dx + dy * dy;
+               if (dist2 <= rmax2) {
+                  cover = 1.0F - (dist2 - rmin2) * cscale;
+                  setup->quad.coverage[QUAD_BOTTOM_RIGHT] = MIN2(cover, 1.0f);
+                  setup->quad.mask |= MASK_BOTTOM_RIGHT;
+               }
+
+               if (setup->quad.mask) {
+                  setup->quad.x0 = ix;
+                  setup->quad.y0 = iy;
+                  clip_emit_quad(setup);
+               }
+            }
+         }
+      }
+      else {
+         /* square points */
+         const int xmin = (int) (x + 0.75 - halfSize);
+         const int ymin = (int) (y + 0.25 - halfSize);
+         const int xmax = xmin + (int) size;
+         const int ymax = ymin + (int) size;
+         /* XXX could apply scissor to xmin,ymin,xmax,ymax now */
+         const int ixmin = block(xmin);
+         const int ixmax = block(xmax - 1);
+         const int iymin = block(ymin);
+         const int iymax = block(ymax - 1);
+         int ix, iy;
+
+         /*
+         debug_printf("(%f, %f) -> X:%d..%d Y:%d..%d\n", x, y, xmin, xmax,ymin,ymax);
+         */
+         for (iy = iymin; iy <= iymax; iy += 2) {
+            uint rowMask = 0xf;
+            if (iy < ymin) {
+               /* above the top edge */
+               rowMask &= (MASK_BOTTOM_LEFT | MASK_BOTTOM_RIGHT);
+            }
+            if (iy + 1 >= ymax) {
+               /* below the bottom edge */
+               rowMask &= (MASK_TOP_LEFT | MASK_TOP_RIGHT);
+            }
+
+            for (ix = ixmin; ix <= ixmax; ix += 2) {
+               uint mask = rowMask;
+
+               if (ix < xmin) {
+                  /* fragment is past left edge of point, turn off left bits */
+                  mask &= (MASK_BOTTOM_RIGHT | MASK_TOP_RIGHT);
+               }
+               if (ix + 1 >= xmax) {
+                  /* past the right edge */
+                  mask &= (MASK_BOTTOM_LEFT | MASK_TOP_LEFT);
+               }
+                  
+               setup->quad.mask = mask;
+               setup->quad.x0 = ix;
+               setup->quad.y0 = iy;
+               clip_emit_quad(setup);
+            }
+         }
+      }
+   }
+}
+
+
+
+static void setup_begin( struct draw_stage *stage )
+{
+   struct setup_stage *setup = setup_stage(stage);
+   struct softpipe_context *sp = setup->softpipe;
+   const struct pipe_shader_state *fs = &setup->softpipe->fs->shader;
+
+   setup->quad.nr_attrs = fs->num_inputs;
+
+   sp->quad.first->begin(sp->quad.first);
+
+   stage->point = setup_point;
+   stage->line = setup_line;
+   stage->tri = setup_tri;
+}
+
+
+static void setup_first_point( struct draw_stage *stage,
+			       struct prim_header *header )
+{
+   setup_begin(stage);
+   stage->point( stage, header );
+}
+
+static void setup_first_line( struct draw_stage *stage,
+			       struct prim_header *header )
+{
+   setup_begin(stage);
+   stage->line( stage, header );
+}
+
+
+static void setup_first_tri( struct draw_stage *stage,
+			       struct prim_header *header )
+{
+   setup_begin(stage);
+   stage->tri( stage, header );
+}
+
+
+
+static void setup_flush( struct draw_stage *stage,
+			 unsigned flags )
+{
+   stage->point = setup_first_point;
+   stage->line = setup_first_line;
+   stage->tri = setup_first_tri;
+}
+
+
+static void reset_stipple_counter( struct draw_stage *stage )
+{
+}
+
+
+static void render_destroy( struct draw_stage *stage )
+{
+   FREE( stage );
+}
+
+
+/**
+ * Create a new primitive setup/render stage.
+ */
+struct draw_stage *sp_draw_render_stage( struct softpipe_context *softpipe )
+{
+   struct setup_stage *setup = CALLOC_STRUCT(setup_stage);
+
+   setup->softpipe = softpipe;
+   setup->stage.draw = softpipe->draw;
+   setup->stage.point = setup_first_point;
+   setup->stage.line = setup_first_line;
+   setup->stage.tri = setup_first_tri;
+   setup->stage.flush = setup_flush;
+   setup->stage.reset_stipple_counter = reset_stipple_counter;
+   setup->stage.destroy = render_destroy;
+
+   setup->quad.coef = setup->coef;
+   setup->quad.posCoef = &setup->posCoef;
+
+   return &setup->stage;
+}
diff --git a/src/gallium/drivers/softpipe/sp_prim_setup.h b/src/gallium/drivers/softpipe/sp_prim_setup.h
new file mode 100644
index 0000000000..f3e8a79dd9
--- /dev/null
+++ b/src/gallium/drivers/softpipe/sp_prim_setup.h
@@ -0,0 +1,79 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+
+#ifndef SP_PRIM_SETUP_H
+#define SP_PRIM_SETUP_H
+
+
+/**
+ * vbuf is a special stage to gather the stream of triangles, lines, points
+ * together and reconstruct vertex buffers for hardware upload.
+ *
+ * First attempt, work in progress.
+ * 
+ * TODO:
+ *    - separate out vertex buffer building and primitive emit, ie >1 draw per vb.
+ *    - tell vbuf stage how to build hw vertices directly
+ *    - pass vbuf stage a buffer pointer for direct emit to agp/vram.
+ *
+ *
+ *
+ * Vertices are just an array of floats, with all the attributes
+ * packed.  We currently assume a layout like:
+ *
+ * attr[0][0..3] - window position
+ * attr[1..n][0..3] - remaining attributes.
+ *
+ * Attributes are assumed to be 4 floats wide but are packed so that
+ * all the enabled attributes run contiguously.
+ */
+
+
+struct draw_stage;
+struct softpipe_context;
+
+
+typedef void (*vbuf_draw_func)( struct pipe_context *pipe,
+                                unsigned prim,
+                                const ushort *elements,
+                                unsigned nr_elements,
+                                const void *vertex_buffer,
+                                unsigned nr_vertices );
+
+
+extern struct draw_stage *
+sp_draw_render_stage( struct softpipe_context *softpipe );
+
+
+extern struct draw_stage *
+sp_draw_vbuf_stage( struct draw_context *draw_context,
+                    struct pipe_context *pipe,
+                    vbuf_draw_func draw );
+
+
+#endif /* SP_PRIM_SETUP_H */
diff --git a/src/gallium/drivers/softpipe/sp_prim_vbuf.c b/src/gallium/drivers/softpipe/sp_prim_vbuf.c
new file mode 100644
index 0000000000..7f71fdb6a9
--- /dev/null
+++ b/src/gallium/drivers/softpipe/sp_prim_vbuf.c
@@ -0,0 +1,221 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+/**
+ * Post-transform vertex buffering.  This is an optional part of the
+ * softpipe rendering pipeline.
+ * Probably not desired in general, but useful for testing/debuggin.
+ * Enabled/Disabled with SP_VBUF env var.
+ * 
+ * Authors
+ *  Brian Paul
+ */
+
+
+#include "sp_context.h"
+#include "sp_state.h"
+#include "sp_prim_vbuf.h"
+#include "pipe/draw/draw_context.h"
+#include "pipe/draw/draw_private.h"
+#include "pipe/draw/draw_vbuf.h"
+
+
+#define SP_MAX_VBUF_INDEXES 1024
+#define SP_MAX_VBUF_SIZE    4096
+
+
+/**
+ * Subclass of vbuf_render.
+ */
+struct softpipe_vbuf_render
+{
+   struct vbuf_render base;
+   struct softpipe_context *softpipe;
+   uint prim;
+   uint vertex_size;
+   void *vertex_buffer;
+};
+
+
+/** cast wrapper */
+static struct softpipe_vbuf_render *
+softpipe_vbuf_render(struct vbuf_render *vbr)
+{
+   return (struct softpipe_vbuf_render *) vbr;
+}
+
+
+
+static const struct vertex_info *
+sp_vbuf_get_vertex_info(struct vbuf_render *vbr)
+{
+   struct softpipe_vbuf_render *cvbr = softpipe_vbuf_render(vbr);
+   return softpipe_get_vbuf_vertex_info(cvbr->softpipe);
+}
+
+
+static void *
+sp_vbuf_allocate_vertices(struct vbuf_render *vbr,
+                            ushort vertex_size, ushort nr_vertices)
+{
+   struct softpipe_vbuf_render *cvbr = softpipe_vbuf_render(vbr);
+   assert(!cvbr->vertex_buffer);
+   cvbr->vertex_buffer = align_malloc(vertex_size * nr_vertices, 16);
+   cvbr->vertex_size = vertex_size;
+   return cvbr->vertex_buffer;
+}
+
+
+static void
+sp_vbuf_release_vertices(struct vbuf_render *vbr, void *vertices, 
+                           unsigned vertex_size, unsigned vertices_used)
+{
+   struct softpipe_vbuf_render *cvbr = softpipe_vbuf_render(vbr);
+   align_free(vertices);
+   assert(vertices == cvbr->vertex_buffer);
+   cvbr->vertex_buffer = NULL;
+}
+
+
+static void
+sp_vbuf_set_primitive(struct vbuf_render *vbr, unsigned prim)
+{
+   struct softpipe_vbuf_render *cvbr = softpipe_vbuf_render(vbr);
+   cvbr->prim = prim;
+}
+
+
+/**
+ * Recalculate prim's determinant.
+ * XXX is this needed?
+ */
+static void
+calc_det(struct prim_header *header)
+{
+   /* Window coords: */
+   const float *v0 = header->v[0]->data[0];
+   const float *v1 = header->v[1]->data[0];
+   const float *v2 = header->v[2]->data[0];
+
+   /* edge vectors e = v0 - v2, f = v1 - v2 */
+   const float ex = v0[0] - v2[0];
+   const float ey = v0[1] - v2[1];
+   const float fx = v1[0] - v2[0];
+   const float fy = v1[1] - v2[1];
+   
+   /* det = cross(e,f).z */
+   header->det = ex * fy - ey * fx;
+}
+
+
+static void
+sp_vbuf_draw(struct vbuf_render *vbr, const ushort *indices, uint nr_indices)
+{
+   struct softpipe_vbuf_render *cvbr = softpipe_vbuf_render(vbr);
+   struct softpipe_context *softpipe = cvbr->softpipe;
+   struct draw_stage *setup = softpipe->setup;
+   struct prim_header prim;
+   unsigned vertex_size = softpipe->vertex_info_vbuf.size * sizeof(float);
+   unsigned i, j;
+   void *vertex_buffer = cvbr->vertex_buffer;
+
+   prim.det = 0;
+   prim.reset_line_stipple = 0;
+   prim.edgeflags = 0;
+   prim.pad = 0;
+
+   switch (cvbr->prim) {
+   case PIPE_PRIM_TRIANGLES:
+      for (i = 0; i < nr_indices; i += 3) {
+         for (j = 0; j < 3; j++) 
+            prim.v[j] = (struct vertex_header *)((char *)vertex_buffer + 
+                                                 indices[i+j] * vertex_size);
+         
+         calc_det(&prim);
+         setup->tri( setup, &prim );
+      }
+      break;
+
+   case PIPE_PRIM_LINES:
+      for (i = 0; i < nr_indices; i += 2) {
+         for (j = 0; j < 2; j++) 
+            prim.v[j] = (struct vertex_header *)((char *)vertex_buffer + 
+                                                 indices[i+j] * vertex_size);
+         
+         setup->line( setup, &prim );
+      }
+      break;
+
+   case PIPE_PRIM_POINTS:
+      for (i = 0; i < nr_indices; i++) {
+         prim.v[0] = (struct vertex_header *)((char *)vertex_buffer + 
+                                              indices[i] * vertex_size);         
+         setup->point( setup, &prim );
+      }
+      break;
+   }
+
+   setup->flush( setup, 0 );
+}
+
+
+static void
+sp_vbuf_destroy(struct vbuf_render *vbr)
+{
+   struct softpipe_vbuf_render *cvbr = softpipe_vbuf_render(vbr);
+   cvbr->softpipe->vbuf_render = NULL;
+   FREE(cvbr);
+}
+
+
+/**
+ * Initialize the post-transform vertex buffer information for the given
+ * context.
+ */
+void
+sp_init_vbuf(struct softpipe_context *sp)
+{
+   assert(sp->draw);
+
+   sp->vbuf_render = CALLOC_STRUCT(softpipe_vbuf_render);
+
+   sp->vbuf_render->base.max_indices = SP_MAX_VBUF_INDEXES;
+   sp->vbuf_render->base.max_vertex_buffer_bytes = SP_MAX_VBUF_SIZE;
+
+   sp->vbuf_render->base.get_vertex_info = sp_vbuf_get_vertex_info;
+   sp->vbuf_render->base.allocate_vertices = sp_vbuf_allocate_vertices;
+   sp->vbuf_render->base.set_primitive = sp_vbuf_set_primitive;
+   sp->vbuf_render->base.draw = sp_vbuf_draw;
+   sp->vbuf_render->base.release_vertices = sp_vbuf_release_vertices;
+   sp->vbuf_render->base.destroy = sp_vbuf_destroy;
+
+   sp->vbuf_render->softpipe = sp;
+
+   sp->vbuf = draw_vbuf_stage(sp->draw, &sp->vbuf_render->base);
+
+   draw_set_rasterize_stage(sp->draw, sp->vbuf);
+}
diff --git a/src/gallium/drivers/softpipe/sp_prim_vbuf.h b/src/gallium/drivers/softpipe/sp_prim_vbuf.h
new file mode 100644
index 0000000000..1de9cc2a89
--- /dev/null
+++ b/src/gallium/drivers/softpipe/sp_prim_vbuf.h
@@ -0,0 +1,38 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+#ifndef SP_VBUF_H
+#define SP_VBUF_H
+
+
+struct softpipe_context;
+
+extern void
+sp_init_vbuf(struct softpipe_context *softpipe);
+
+
+#endif /* SP_VBUF_H */
diff --git a/src/gallium/drivers/softpipe/sp_quad.c b/src/gallium/drivers/softpipe/sp_quad.c
new file mode 100644
index 0000000000..6bd468a51c
--- /dev/null
+++ b/src/gallium/drivers/softpipe/sp_quad.c
@@ -0,0 +1,118 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+
+#include "sp_context.h"
+#include "sp_state.h"
+#include "pipe/p_shader_tokens.h"
+
+static void
+sp_push_quad_first(
+   struct softpipe_context *sp,
+   struct quad_stage       *quad )
+{
+   quad->next = sp->quad.first;
+   sp->quad.first = quad;
+}
+
+static void
+sp_build_depth_stencil(
+   struct softpipe_context *sp )
+{
+   if (sp->depth_stencil->stencil[0].enabled ||
+       sp->depth_stencil->stencil[1].enabled) {
+      sp_push_quad_first( sp, sp->quad.stencil_test );
+   }
+   else if (sp->depth_stencil->depth.enabled &&
+            sp->framebuffer.zsbuf) {
+      sp_push_quad_first( sp, sp->quad.depth_test );
+   }
+}
+
+void
+sp_build_quad_pipeline(struct softpipe_context *sp)
+{
+   boolean  early_depth_test =
+               sp->depth_stencil->depth.enabled &&
+               sp->framebuffer.zsbuf &&
+               !sp->depth_stencil->alpha.enabled &&
+               sp->fs->shader.output_semantic_name[0] != TGSI_SEMANTIC_POSITION;
+
+   /* build up the pipeline in reverse order... */
+
+   sp->quad.first = sp->quad.output;
+
+   if (sp->blend->colormask != 0xf) {
+      sp_push_quad_first( sp, sp->quad.colormask );
+   }
+
+   if (sp->blend->blend_enable ||
+       sp->blend->logicop_enable) {
+      sp_push_quad_first( sp, sp->quad.blend );
+   }
+
+   if (sp->framebuffer.num_cbufs == 1) {
+      /* the usual case: write to exactly one colorbuf */
+      sp->current_cbuf = 0;
+   }
+   else {
+      /* insert bufloop stage */
+      sp_push_quad_first( sp, sp->quad.bufloop );
+   }
+
+   if (sp->depth_stencil->depth.occlusion_count) {
+      sp_push_quad_first( sp, sp->quad.occlusion );
+   }
+
+   if (sp->rasterizer->poly_smooth ||
+       sp->rasterizer->line_smooth ||
+       sp->rasterizer->point_smooth) {
+      sp_push_quad_first( sp, sp->quad.coverage );
+   }
+
+   if (!early_depth_test) {
+      sp_build_depth_stencil( sp );
+   }
+
+   if (sp->depth_stencil->alpha.enabled) {
+      sp_push_quad_first( sp, sp->quad.alpha_test );
+   }
+
+   /* XXX always enable shader? */
+   if (1) {
+      sp_push_quad_first( sp, sp->quad.shade );
+   }
+
+   if (early_depth_test) {
+      sp_build_depth_stencil( sp );
+      sp_push_quad_first( sp, sp->quad.earlyz );
+   }
+
+   if (sp->rasterizer->poly_stipple_enable) {
+      sp_push_quad_first( sp, sp->quad.polygon_stipple );
+   }
+}
diff --git a/src/gallium/drivers/softpipe/sp_quad.h b/src/gallium/drivers/softpipe/sp_quad.h
new file mode 100644
index 0000000000..f1e0281764
--- /dev/null
+++ b/src/gallium/drivers/softpipe/sp_quad.h
@@ -0,0 +1,70 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+/* Authors:  Keith Whitwell <keith@tungstengraphics.com>
+ */
+
+#ifndef SP_QUAD_H
+#define SP_QUAD_H
+
+
+struct softpipe_context;
+struct quad_header;
+
+
+struct quad_stage {
+   struct softpipe_context *softpipe;
+
+   struct quad_stage *next;
+
+   void (*begin)(struct quad_stage *qs);
+
+   /** the stage action */
+   void (*run)(struct quad_stage *qs, struct quad_header *quad);
+
+   void (*destroy)(struct quad_stage *qs);
+};
+
+
+struct quad_stage *sp_quad_polygon_stipple_stage( struct softpipe_context *softpipe );
+struct quad_stage *sp_quad_earlyz_stage( struct softpipe_context *softpipe );
+struct quad_stage *sp_quad_shade_stage( struct softpipe_context *softpipe );
+struct quad_stage *sp_quad_alpha_test_stage( struct softpipe_context *softpipe );
+struct quad_stage *sp_quad_stencil_test_stage( struct softpipe_context *softpipe );
+struct quad_stage *sp_quad_depth_test_stage( struct softpipe_context *softpipe );
+struct quad_stage *sp_quad_occlusion_stage( struct softpipe_context *softpipe );
+struct quad_stage *sp_quad_coverage_stage( struct softpipe_context *softpipe );
+struct quad_stage *sp_quad_bufloop_stage( struct softpipe_context *softpipe );
+struct quad_stage *sp_quad_blend_stage( struct softpipe_context *softpipe );
+struct quad_stage *sp_quad_colormask_stage( struct softpipe_context *softpipe );
+struct quad_stage *sp_quad_output_stage( struct softpipe_context *softpipe );
+
+void sp_build_quad_pipeline(struct softpipe_context *sp);
+
+void sp_depth_test_quad(struct quad_stage *qs, struct quad_header *quad);
+
+#endif /* SP_QUAD_H */
diff --git a/src/gallium/drivers/softpipe/sp_quad_alpha_test.c b/src/gallium/drivers/softpipe/sp_quad_alpha_test.c
new file mode 100644
index 0000000000..4ffeac35e1
--- /dev/null
+++ b/src/gallium/drivers/softpipe/sp_quad_alpha_test.c
@@ -0,0 +1,108 @@
+
+/**
+ * quad alpha test
+ */
+
+#include "sp_context.h"
+#include "sp_headers.h"
+#include "sp_quad.h"
+#include "pipe/p_defines.h"
+#include "pipe/p_util.h"
+
+
+static void
+alpha_test_quad(struct quad_stage *qs, struct quad_header *quad)
+{
+   struct softpipe_context *softpipe = qs->softpipe;
+   const float ref = softpipe->depth_stencil->alpha.ref;
+   unsigned passMask = 0x0, j;
+   const float *aaaa = quad->outputs.color[3];
+
+   switch (softpipe->depth_stencil->alpha.func) {
+   case PIPE_FUNC_NEVER:
+      quad->mask = 0x0;
+      break;
+   case PIPE_FUNC_LESS:
+      /*
+       * If mask were an array [4] we could do this SIMD-style:
+       * passMask = (quad->outputs.color[3] <= vec4(ref));
+       */
+      for (j = 0; j < QUAD_SIZE; j++) {
+         if (aaaa[j] < ref) {
+            passMask |= (1 << j);
+         }
+      }
+      break;
+   case PIPE_FUNC_EQUAL:
+      for (j = 0; j < QUAD_SIZE; j++) {
+         if (aaaa[j] == ref) {
+            passMask |= (1 << j);
+         }
+      }
+      break;
+   case PIPE_FUNC_LEQUAL:
+      for (j = 0; j < QUAD_SIZE; j++) {
+         if (aaaa[j] <= ref) {
+            passMask |= (1 << j);
+         }
+      }
+      break;
+   case PIPE_FUNC_GREATER:
+      for (j = 0; j < QUAD_SIZE; j++) {
+         if (aaaa[j] > ref) {
+            passMask |= (1 << j);
+         }
+      }
+      break;
+   case PIPE_FUNC_NOTEQUAL:
+      for (j = 0; j < QUAD_SIZE; j++) {
+         if (aaaa[j] != ref) {
+            passMask |= (1 << j);
+         }
+      }
+      break;
+   case PIPE_FUNC_GEQUAL:
+      for (j = 0; j < QUAD_SIZE; j++) {
+         if (aaaa[j] >= ref) {
+            passMask |= (1 << j);
+         }
+      }
+      break;
+   case PIPE_FUNC_ALWAYS:
+      passMask = MASK_ALL;
+      break;
+   default:
+      abort();
+   }
+
+   quad->mask &= passMask;
+
+   if (quad->mask)
+      qs->next->run(qs->next, quad);
+}
+
+
+static void alpha_test_begin(struct quad_stage *qs)
+{
+   qs->next->begin(qs->next);
+}
+
+
+static void alpha_test_destroy(struct quad_stage *qs)
+{
+   FREE( qs );
+}
+
+
+struct quad_stage *
+sp_quad_alpha_test_stage( struct softpipe_context *softpipe )
+{
+   struct quad_stage *stage = CALLOC_STRUCT(quad_stage);
+
+   stage->softpipe = softpipe;
+   stage->begin = alpha_test_begin;
+   stage->run = alpha_test_quad;
+   stage->destroy = alpha_test_destroy;
+
+   return stage;
+}
diff --git a/src/gallium/drivers/softpipe/sp_quad_blend.c b/src/gallium/drivers/softpipe/sp_quad_blend.c
new file mode 100644
index 0000000000..17f3ecd0b8
--- /dev/null
+++ b/src/gallium/drivers/softpipe/sp_quad_blend.c
@@ -0,0 +1,749 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+/**
+ * quad blending
+ * \author Brian Paul
+ */
+
+#include "pipe/p_defines.h"
+#include "pipe/p_util.h"
+#include "sp_context.h"
+#include "sp_headers.h"
+#include "sp_surface.h"
+#include "sp_tile_cache.h"
+#include "sp_quad.h"
+
+
+#define VEC4_COPY(DST, SRC) \
+do { \
+    DST[0] = SRC[0]; \
+    DST[1] = SRC[1]; \
+    DST[2] = SRC[2]; \
+    DST[3] = SRC[3]; \
+} while(0)
+
+#define VEC4_SCALAR(DST, SRC) \
+do { \
+    DST[0] = SRC; \
+    DST[1] = SRC; \
+    DST[2] = SRC; \
+    DST[3] = SRC; \
+} while(0)
+
+#define VEC4_ADD(R, A, B) \
+do { \
+   R[0] = A[0] + B[0]; \
+   R[1] = A[1] + B[1]; \
+   R[2] = A[2] + B[2]; \
+   R[3] = A[3] + B[3]; \
+} while (0)
+
+#define VEC4_SUB(R, A, B) \
+do { \
+   R[0] = A[0] - B[0]; \
+   R[1] = A[1] - B[1]; \
+   R[2] = A[2] - B[2]; \
+   R[3] = A[3] - B[3]; \
+} while (0)
+
+#define VEC4_MUL(R, A, B) \
+do { \
+   R[0] = A[0] * B[0]; \
+   R[1] = A[1] * B[1]; \
+   R[2] = A[2] * B[2]; \
+   R[3] = A[3] * B[3]; \
+} while (0)
+
+#define VEC4_MIN(R, A, B) \
+do { \
+   R[0] = (A[0] < B[0]) ? A[0] : B[0]; \
+   R[1] = (A[1] < B[1]) ? A[1] : B[1]; \
+   R[2] = (A[2] < B[2]) ? A[2] : B[2]; \
+   R[3] = (A[3] < B[3]) ? A[3] : B[3]; \
+} while (0)
+
+#define VEC4_MAX(R, A, B) \
+do { \
+   R[0] = (A[0] > B[0]) ? A[0] : B[0]; \
+   R[1] = (A[1] > B[1]) ? A[1] : B[1]; \
+   R[2] = (A[2] > B[2]) ? A[2] : B[2]; \
+   R[3] = (A[3] > B[3]) ? A[3] : B[3]; \
+} while (0)
+
+
+
+static void
+logicop_quad(struct quad_stage *qs, struct quad_header *quad)
+{
+   struct softpipe_context *softpipe = qs->softpipe;
+   float dest[4][QUAD_SIZE];
+   ubyte src[4][4], dst[4][4], res[4][4];
+   uint *src4 = (uint *) src;
+   uint *dst4 = (uint *) dst;
+   uint *res4 = (uint *) res;
+   struct softpipe_cached_tile *
+      tile = sp_get_cached_tile(softpipe,
+                                softpipe->cbuf_cache[softpipe->current_cbuf],
+                                quad->x0, quad->y0);
+   float (*quadColor)[4] = quad->outputs.color;
+   uint i, j;
+
+   /* get/swizzle dest colors */
+   for (j = 0; j < QUAD_SIZE; j++) {
+      int x = (quad->x0 & (TILE_SIZE-1)) + (j & 1);
+      int y = (quad->y0 & (TILE_SIZE-1)) + (j >> 1);
+      for (i = 0; i < 4; i++) {
+         dest[i][j] = tile->data.color[y][x][i];
+      }
+   }
+
+   /* convert to ubyte */
+   for (j = 0; j < 4; j++) { /* loop over R,G,B,A channels */
+      UNCLAMPED_FLOAT_TO_UBYTE(dst[j][0], dest[j][0]); /* P0 */
+      UNCLAMPED_FLOAT_TO_UBYTE(dst[j][1], dest[j][1]); /* P1 */
+      UNCLAMPED_FLOAT_TO_UBYTE(dst[j][2], dest[j][2]); /* P2 */
+      UNCLAMPED_FLOAT_TO_UBYTE(dst[j][3], dest[j][3]); /* P3 */
+
+      UNCLAMPED_FLOAT_TO_UBYTE(src[j][0], quadColor[j][0]); /* P0 */
+      UNCLAMPED_FLOAT_TO_UBYTE(src[j][1], quadColor[j][1]); /* P1 */
+      UNCLAMPED_FLOAT_TO_UBYTE(src[j][2], quadColor[j][2]); /* P2 */
+      UNCLAMPED_FLOAT_TO_UBYTE(src[j][3], quadColor[j][3]); /* P3 */
+   }
+
+   switch (softpipe->blend->logicop_func) {
+   case PIPE_LOGICOP_CLEAR:
+      for (j = 0; j < 4; j++)
+         res4[j] = 0;
+      break;
+   case PIPE_LOGICOP_NOR:
+      for (j = 0; j < 4; j++)
+         res4[j] = ~(src4[j] | dst4[j]);
+      break;
+   case PIPE_LOGICOP_AND_INVERTED:
+      for (j = 0; j < 4; j++)
+         res4[j] = ~src4[j] & dst4[j];
+      break;
+   case PIPE_LOGICOP_COPY_INVERTED:
+      for (j = 0; j < 4; j++)
+         res4[j] = ~src4[j];
+      break;
+   case PIPE_LOGICOP_AND_REVERSE:
+      for (j = 0; j < 4; j++)
+         res4[j] = src4[j] & ~dst4[j];
+      break;
+   case PIPE_LOGICOP_INVERT:
+      for (j = 0; j < 4; j++)
+         res4[j] = ~dst4[j];
+      break;
+   case PIPE_LOGICOP_XOR:
+      for (j = 0; j < 4; j++)
+         res4[j] = dst4[j] ^ src4[j];
+      break;
+   case PIPE_LOGICOP_NAND:
+      for (j = 0; j < 4; j++)
+         res4[j] = ~(src4[j] & dst4[j]);
+      break;
+   case PIPE_LOGICOP_AND:
+      for (j = 0; j < 4; j++)
+         res4[j] = src4[j] & dst4[j];
+      break;
+   case PIPE_LOGICOP_EQUIV:
+      for (j = 0; j < 4; j++)
+         res4[j] = ~(src4[j] ^ dst4[j]);
+      break;
+   case PIPE_LOGICOP_NOOP:
+      for (j = 0; j < 4; j++)
+         res4[j] = dst4[j];
+      break;
+   case PIPE_LOGICOP_OR_INVERTED:
+      for (j = 0; j < 4; j++)
+         res4[j] = ~src4[j] | dst4[j];
+      break;
+   case PIPE_LOGICOP_COPY:
+      for (j = 0; j < 4; j++)
+         res4[j] = src4[j];
+      break;
+   case PIPE_LOGICOP_OR_REVERSE:
+      for (j = 0; j < 4; j++)
+         res4[j] = src4[j] | ~dst4[j];
+      break;
+   case PIPE_LOGICOP_OR:
+      for (j = 0; j < 4; j++)
+         res4[j] = src4[j] | dst4[j];
+      break;
+   case PIPE_LOGICOP_SET:
+      for (j = 0; j < 4; j++)
+         res4[j] = ~0;
+      break;
+   default:
+      assert(0);
+   }
+
+   for (j = 0; j < 4; j++) {
+      quadColor[j][0] = UBYTE_TO_FLOAT(res[j][0]);
+      quadColor[j][1] = UBYTE_TO_FLOAT(res[j][1]);
+      quadColor[j][2] = UBYTE_TO_FLOAT(res[j][2]);
+      quadColor[j][3] = UBYTE_TO_FLOAT(res[j][3]);
+   }
+
+   /* pass quad to next stage */
+   qs->next->run(qs->next, quad);
+}
+
+
+
+
+static void
+blend_quad(struct quad_stage *qs, struct quad_header *quad)
+{
+   struct softpipe_context *softpipe = qs->softpipe;
+   static const float zero[4] = { 0, 0, 0, 0 };
+   static const float one[4] = { 1, 1, 1, 1 };
+   float source[4][QUAD_SIZE], dest[4][QUAD_SIZE];
+   struct softpipe_cached_tile *tile
+      = sp_get_cached_tile(softpipe,
+                           softpipe->cbuf_cache[softpipe->current_cbuf],
+                           quad->x0, quad->y0);
+   float (*quadColor)[4] = quad->outputs.color;
+   uint i, j;
+
+   if (softpipe->blend->logicop_enable) {
+      logicop_quad(qs, quad);
+      return;
+   }
+
+   /* get/swizzle dest colors */
+   for (j = 0; j < QUAD_SIZE; j++) {
+      int x = (quad->x0 & (TILE_SIZE-1)) + (j & 1);
+      int y = (quad->y0 & (TILE_SIZE-1)) + (j >> 1);
+      for (i = 0; i < 4; i++) {
+         dest[i][j] = tile->data.color[y][x][i];
+      }
+   }
+
+   /*
+    * Compute src/first term RGB
+    */
+   switch (softpipe->blend->rgb_src_factor) {
+   case PIPE_BLENDFACTOR_ONE:
+      VEC4_COPY(source[0], quadColor[0]); /* R */
+      VEC4_COPY(source[1], quadColor[1]); /* G */
+      VEC4_COPY(source[2], quadColor[2]); /* B */
+      break;
+   case PIPE_BLENDFACTOR_SRC_COLOR:
+      VEC4_MUL(source[0], quadColor[0], quadColor[0]); /* R */
+      VEC4_MUL(source[1], quadColor[1], quadColor[1]); /* G */
+      VEC4_MUL(source[2], quadColor[2], quadColor[2]); /* B */
+      break;
+   case PIPE_BLENDFACTOR_SRC_ALPHA:
+      {
+         const float *alpha = quadColor[3];
+         VEC4_MUL(source[0], quadColor[0], alpha); /* R */
+         VEC4_MUL(source[1], quadColor[1], alpha); /* G */
+         VEC4_MUL(source[2], quadColor[2], alpha); /* B */
+      }
+      break;
+   case PIPE_BLENDFACTOR_DST_COLOR:
+      VEC4_MUL(source[0], quadColor[0], dest[0]); /* R */
+      VEC4_MUL(source[1], quadColor[1], dest[1]); /* G */
+      VEC4_MUL(source[2], quadColor[2], dest[2]); /* B */
+      break;
+   case PIPE_BLENDFACTOR_DST_ALPHA:
+      {
+         const float *alpha = dest[3];
+         VEC4_MUL(source[0], quadColor[0], alpha); /* R */
+         VEC4_MUL(source[1], quadColor[1], alpha); /* G */
+         VEC4_MUL(source[2], quadColor[2], alpha); /* B */
+      }
+      break;
+   case PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE:
+      {
+         const float *alpha = quadColor[3];
+         float diff[4];
+         VEC4_SUB(diff, one, dest[3]);
+         VEC4_MIN(source[0], alpha, diff); /* R */
+         VEC4_MIN(source[1], alpha, diff); /* G */
+         VEC4_MIN(source[2], alpha, diff); /* B */
+      }
+      break;
+   case PIPE_BLENDFACTOR_CONST_COLOR:
+      {
+         float comp[4];
+         VEC4_SCALAR(comp, softpipe->blend_color.color[0]); /* R */
+         VEC4_MUL(source[0], quadColor[0], comp); /* R */
+         VEC4_SCALAR(comp, softpipe->blend_color.color[1]); /* G */
+         VEC4_MUL(source[1], quadColor[1], comp); /* G */
+         VEC4_SCALAR(comp, softpipe->blend_color.color[2]); /* B */
+         VEC4_MUL(source[2], quadColor[2], comp); /* B */
+      }
+      break;
+   case PIPE_BLENDFACTOR_CONST_ALPHA:
+      {
+         float alpha[4];
+         VEC4_SCALAR(alpha, softpipe->blend_color.color[3]);
+         VEC4_MUL(source[0], quadColor[0], alpha); /* R */
+         VEC4_MUL(source[1], quadColor[1], alpha); /* G */
+         VEC4_MUL(source[2], quadColor[2], alpha); /* B */
+      }
+      break;
+   case PIPE_BLENDFACTOR_SRC1_COLOR:
+      assert(0); /* to do */
+      break;
+   case PIPE_BLENDFACTOR_SRC1_ALPHA:
+      assert(0); /* to do */
+      break;
+   case PIPE_BLENDFACTOR_ZERO:
+      VEC4_COPY(source[0], zero); /* R */
+      VEC4_COPY(source[1], zero); /* G */
+      VEC4_COPY(source[2], zero); /* B */
+      break;
+   case PIPE_BLENDFACTOR_INV_SRC_COLOR:
+      {
+         float inv_comp[4];
+         VEC4_SUB(inv_comp, one, quadColor[0]); /* R */
+         VEC4_MUL(source[0], quadColor[0], inv_comp); /* R */
+         VEC4_SUB(inv_comp, one, quadColor[1]); /* G */
+         VEC4_MUL(source[1], quadColor[1], inv_comp); /* G */
+         VEC4_SUB(inv_comp, one, quadColor[2]); /* B */
+         VEC4_MUL(source[2], quadColor[2], inv_comp); /* B */
+      }
+      break;
+   case PIPE_BLENDFACTOR_INV_SRC_ALPHA:
+      {
+         float inv_alpha[4];
+         VEC4_SUB(inv_alpha, one, quadColor[3]);
+         VEC4_MUL(source[0], quadColor[0], inv_alpha); /* R */
+         VEC4_MUL(source[1], quadColor[1], inv_alpha); /* G */
+         VEC4_MUL(source[2], quadColor[2], inv_alpha); /* B */
+      }
+      break;
+   case PIPE_BLENDFACTOR_INV_DST_ALPHA:
+      {
+         float inv_alpha[4];
+         VEC4_SUB(inv_alpha, one, dest[3]);
+         VEC4_MUL(source[0], quadColor[0], inv_alpha); /* R */
+         VEC4_MUL(source[1], quadColor[1], inv_alpha); /* G */
+         VEC4_MUL(source[2], quadColor[2], inv_alpha); /* B */
+      }
+      break;
+   case PIPE_BLENDFACTOR_INV_DST_COLOR:
+      {
+         float inv_comp[4];
+         VEC4_SUB(inv_comp, one, dest[0]); /* R */
+         VEC4_MUL(source[0], quadColor[0], inv_comp); /* R */
+         VEC4_SUB(inv_comp, one, dest[1]); /* G */
+         VEC4_MUL(source[1], quadColor[1], inv_comp); /* G */
+         VEC4_SUB(inv_comp, one, dest[2]); /* B */
+         VEC4_MUL(source[2], quadColor[2], inv_comp); /* B */
+      }
+      break;
+   case PIPE_BLENDFACTOR_INV_CONST_COLOR:
+      {
+         float inv_comp[4];
+         /* R */
+         VEC4_SCALAR(inv_comp, 1.0f - softpipe->blend_color.color[0]);
+         VEC4_MUL(source[0], quadColor[0], inv_comp);
+         /* G */
+         VEC4_SCALAR(inv_comp, 1.0f - softpipe->blend_color.color[1]);
+         VEC4_MUL(source[1], quadColor[1], inv_comp);
+         /* B */
+         VEC4_SCALAR(inv_comp, 1.0f - softpipe->blend_color.color[2]);
+         VEC4_MUL(source[2], quadColor[2], inv_comp);
+      }
+      break;
+   case PIPE_BLENDFACTOR_INV_CONST_ALPHA:
+      {
+         float inv_alpha[4];
+         VEC4_SCALAR(inv_alpha, 1.0f - softpipe->blend_color.color[3]);
+         VEC4_MUL(source[0], quadColor[0], inv_alpha); /* R */
+         VEC4_MUL(source[1], quadColor[1], inv_alpha); /* G */
+         VEC4_MUL(source[2], quadColor[2], inv_alpha); /* B */
+      }
+      break;
+   case PIPE_BLENDFACTOR_INV_SRC1_COLOR:
+      assert(0); /* to do */
+      break;
+   case PIPE_BLENDFACTOR_INV_SRC1_ALPHA:
+      assert(0); /* to do */
+      break;
+   default:
+      abort();
+   }
+   
+   /*
+    * Compute src/first term A
+    */
+   switch (softpipe->blend->alpha_src_factor) {
+   case PIPE_BLENDFACTOR_ONE:
+      VEC4_COPY(source[3], quadColor[3]); /* A */
+      break;
+   case PIPE_BLENDFACTOR_SRC_COLOR:
+      /* fall-through */
+   case PIPE_BLENDFACTOR_SRC_ALPHA:
+      {
+         const float *alpha = quadColor[3];
+         VEC4_MUL(source[3], quadColor[3], alpha); /* A */
+      }
+      break;
+   case PIPE_BLENDFACTOR_DST_COLOR:
+      /* fall-through */
+   case PIPE_BLENDFACTOR_DST_ALPHA:
+      VEC4_MUL(source[3], quadColor[3], dest[3]); /* A */
+      break;
+   case PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE:
+      {
+         const float *alpha = quadColor[3];
+         float diff[4];
+         VEC4_SUB(diff, one, dest[3]);
+         VEC4_MIN(source[3], alpha, diff); /* A */
+      }
+      break;
+   case PIPE_BLENDFACTOR_CONST_COLOR:
+      /* fall-through */
+   case PIPE_BLENDFACTOR_CONST_ALPHA:
+      {
+         float comp[4];
+         VEC4_SCALAR(comp, softpipe->blend_color.color[3]); /* A */
+         VEC4_MUL(source[3], quadColor[3], comp); /* A */
+      }
+      break;
+   case PIPE_BLENDFACTOR_ZERO:
+      VEC4_COPY(source[3], zero); /* A */
+      break;
+   case PIPE_BLENDFACTOR_INV_SRC_COLOR:
+      /* fall-through */
+   case PIPE_BLENDFACTOR_INV_SRC_ALPHA:
+      {
+         float inv_alpha[4];
+         VEC4_SUB(inv_alpha, one, quadColor[3]);
+         VEC4_MUL(source[3], quadColor[3], inv_alpha); /* A */
+      }
+      break;
+   case PIPE_BLENDFACTOR_INV_DST_COLOR:
+      /* fall-through */
+   case PIPE_BLENDFACTOR_INV_DST_ALPHA:
+      {
+         float inv_alpha[4];
+         VEC4_SUB(inv_alpha, one, dest[3]);
+         VEC4_MUL(source[3], quadColor[3], inv_alpha); /* A */
+      }
+      break;
+   case PIPE_BLENDFACTOR_INV_CONST_COLOR:
+      /* fall-through */
+   case PIPE_BLENDFACTOR_INV_CONST_ALPHA:
+      {
+         float inv_comp[4];
+         /* A */
+         VEC4_SCALAR(inv_comp, 1.0f - softpipe->blend_color.color[3]);
+         VEC4_MUL(source[3], quadColor[3], inv_comp);
+      }
+      break;
+   default:
+      abort();
+   }
+   
+   
+   /*
+    * Compute dest/second term RGB
+    */
+   switch (softpipe->blend->rgb_dst_factor) {
+   case PIPE_BLENDFACTOR_ONE:
+      /* dest = dest * 1   NO-OP, leave dest as-is */
+      break;
+   case PIPE_BLENDFACTOR_SRC_COLOR:
+      VEC4_MUL(dest[0], dest[0], quadColor[0]); /* R */
+      VEC4_MUL(dest[1], dest[1], quadColor[1]); /* G */
+      VEC4_MUL(dest[2], dest[2], quadColor[2]); /* B */
+      break;
+   case PIPE_BLENDFACTOR_SRC_ALPHA:
+      VEC4_MUL(dest[0], dest[0], quadColor[3]); /* R * A */
+      VEC4_MUL(dest[1], dest[1], quadColor[3]); /* G * A */
+      VEC4_MUL(dest[2], dest[2], quadColor[3]); /* B * A */
+      break;
+   case PIPE_BLENDFACTOR_DST_ALPHA:
+      VEC4_MUL(dest[0], dest[0], dest[3]); /* R * A */
+      VEC4_MUL(dest[1], dest[1], dest[3]); /* G * A */
+      VEC4_MUL(dest[2], dest[2], dest[3]); /* B * A */
+      break;
+   case PIPE_BLENDFACTOR_DST_COLOR:
+      VEC4_MUL(dest[0], dest[0], dest[0]); /* R */
+      VEC4_MUL(dest[1], dest[1], dest[1]); /* G */
+      VEC4_MUL(dest[2], dest[2], dest[2]); /* B */
+      break;
+   case PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE:
+      assert(0); /* illegal */
+      break;
+   case PIPE_BLENDFACTOR_CONST_COLOR:
+      {
+         float comp[4];
+         VEC4_SCALAR(comp, softpipe->blend_color.color[0]); /* R */
+         VEC4_MUL(dest[0], dest[0], comp); /* R */
+         VEC4_SCALAR(comp, softpipe->blend_color.color[1]); /* G */
+         VEC4_MUL(dest[1], dest[1], comp); /* G */
+         VEC4_SCALAR(comp, softpipe->blend_color.color[2]); /* B */
+         VEC4_MUL(dest[2], dest[2], comp); /* B */
+      }
+      break;
+   case PIPE_BLENDFACTOR_CONST_ALPHA:
+      {
+         float comp[4];
+         VEC4_SCALAR(comp, softpipe->blend_color.color[3]); /* A */
+         VEC4_MUL(dest[0], dest[0], comp); /* R */
+         VEC4_MUL(dest[1], dest[1], comp); /* G */
+         VEC4_MUL(dest[2], dest[2], comp); /* B */
+      }
+      break;
+   case PIPE_BLENDFACTOR_ZERO:
+      VEC4_COPY(dest[0], zero); /* R */
+      VEC4_COPY(dest[1], zero); /* G */
+      VEC4_COPY(dest[2], zero); /* B */
+      break;
+   case PIPE_BLENDFACTOR_SRC1_COLOR:
+   case PIPE_BLENDFACTOR_SRC1_ALPHA:
+      /* XXX what are these? */
+      assert(0);
+      break;
+   case PIPE_BLENDFACTOR_INV_SRC_COLOR:
+      {
+         float inv_comp[4];
+         VEC4_SUB(inv_comp, one, quadColor[0]); /* R */
+         VEC4_MUL(dest[0], inv_comp, dest[0]); /* R */
+         VEC4_SUB(inv_comp, one, quadColor[1]); /* G */
+         VEC4_MUL(dest[1], inv_comp, dest[1]); /* G */
+         VEC4_SUB(inv_comp, one, quadColor[2]); /* B */
+         VEC4_MUL(dest[2], inv_comp, dest[2]); /* B */
+      }
+      break;
+   case PIPE_BLENDFACTOR_INV_SRC_ALPHA:
+      {
+         float one_minus_alpha[QUAD_SIZE];
+         VEC4_SUB(one_minus_alpha, one, quadColor[3]);
+         VEC4_MUL(dest[0], dest[0], one_minus_alpha); /* R */
+         VEC4_MUL(dest[1], dest[1], one_minus_alpha); /* G */
+         VEC4_MUL(dest[2], dest[2], one_minus_alpha); /* B */
+      }
+      break;
+   case PIPE_BLENDFACTOR_INV_DST_ALPHA:
+      {
+         float inv_comp[4];
+         VEC4_SUB(inv_comp, one, quadColor[3]); /* A */
+         VEC4_MUL(dest[0], inv_comp, dest[0]); /* R */
+         VEC4_MUL(dest[1], inv_comp, dest[1]); /* G */
+         VEC4_MUL(dest[2], inv_comp, dest[2]); /* B */
+      }
+      break;
+   case PIPE_BLENDFACTOR_INV_DST_COLOR:
+      {
+         float inv_comp[4];
+         VEC4_SUB(inv_comp, one, dest[0]); /* R */
+         VEC4_MUL(dest[0], dest[0], inv_comp); /* R */
+         VEC4_SUB(inv_comp, one, dest[1]); /* G */
+         VEC4_MUL(dest[1], dest[1], inv_comp); /* G */
+         VEC4_SUB(inv_comp, one, dest[2]); /* B */
+         VEC4_MUL(dest[2], dest[2], inv_comp); /* B */
+      }
+      break;
+   case PIPE_BLENDFACTOR_INV_CONST_COLOR:
+      {
+         float inv_comp[4];
+         /* R */
+         VEC4_SCALAR(inv_comp, 1.0f - softpipe->blend_color.color[0]);
+         VEC4_MUL(dest[0], dest[0], inv_comp);
+         /* G */
+         VEC4_SCALAR(inv_comp, 1.0f - softpipe->blend_color.color[1]);
+         VEC4_MUL(dest[1], dest[1], inv_comp);
+         /* B */
+         VEC4_SCALAR(inv_comp, 1.0f - softpipe->blend_color.color[2]);
+         VEC4_MUL(dest[2], dest[2], inv_comp);
+      }
+      break;
+   case PIPE_BLENDFACTOR_INV_CONST_ALPHA:
+      {
+         float inv_comp[4];
+         VEC4_SCALAR(inv_comp, 1.0f - softpipe->blend_color.color[3]);
+         VEC4_MUL(dest[0], dest[0], inv_comp);
+         VEC4_MUL(dest[1], dest[1], inv_comp);
+         VEC4_MUL(dest[2], dest[2], inv_comp);
+      }
+      break;
+   case PIPE_BLENDFACTOR_INV_SRC1_COLOR:
+   case PIPE_BLENDFACTOR_INV_SRC1_ALPHA:
+      /* XXX what are these? */
+      assert(0);
+      break;
+   default:
+      assert(0);
+   }
+   
+   /*
+    * Compute dest/second term A
+    */
+   switch (softpipe->blend->alpha_dst_factor) {
+   case PIPE_BLENDFACTOR_ONE:
+      /* dest = dest * 1   NO-OP, leave dest as-is */
+      break;
+   case PIPE_BLENDFACTOR_SRC_COLOR:
+      /* fall-through */
+   case PIPE_BLENDFACTOR_SRC_ALPHA:
+      VEC4_MUL(dest[3], dest[3], quadColor[3]); /* A * A */
+      break;
+   case PIPE_BLENDFACTOR_DST_COLOR:
+      /* fall-through */
+   case PIPE_BLENDFACTOR_DST_ALPHA:
+      VEC4_MUL(dest[3], dest[3], dest[3]); /* A */
+      break;
+   case PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE:
+      assert(0); /* illegal */
+      break;
+   case PIPE_BLENDFACTOR_CONST_COLOR:
+      /* fall-through */
+   case PIPE_BLENDFACTOR_CONST_ALPHA:
+      {
+         float comp[4];
+         VEC4_SCALAR(comp, softpipe->blend_color.color[3]); /* A */
+         VEC4_MUL(dest[3], dest[3], comp); /* A */
+      }
+      break;
+   case PIPE_BLENDFACTOR_ZERO:
+      VEC4_COPY(dest[3], zero); /* A */
+      break;
+   case PIPE_BLENDFACTOR_INV_SRC_COLOR:
+      /* fall-through */
+   case PIPE_BLENDFACTOR_INV_SRC_ALPHA:
+      {
+         float one_minus_alpha[QUAD_SIZE];
+         VEC4_SUB(one_minus_alpha, one, quadColor[3]);
+         VEC4_MUL(dest[3], dest[3], one_minus_alpha); /* A */
+      }
+      break;
+   case PIPE_BLENDFACTOR_INV_DST_COLOR:
+      /* fall-through */
+   case PIPE_BLENDFACTOR_INV_DST_ALPHA:
+      {
+         float inv_comp[4];
+         VEC4_SUB(inv_comp, one, dest[3]); /* A */
+         VEC4_MUL(dest[3], inv_comp, dest[3]); /* A */
+      }
+      break;
+   case PIPE_BLENDFACTOR_INV_CONST_COLOR:
+      /* fall-through */
+   case PIPE_BLENDFACTOR_INV_CONST_ALPHA:
+      {
+         float inv_comp[4];
+         VEC4_SCALAR(inv_comp, 1.0f - softpipe->blend_color.color[3]);
+         VEC4_MUL(dest[3], dest[3], inv_comp);
+      }
+      break;
+   default:
+      assert(0);
+   }
+   
+   /*
+    * Combine RGB terms
+    */
+   switch (softpipe->blend->rgb_func) {
+   case PIPE_BLEND_ADD:
+      VEC4_ADD(quadColor[0], source[0], dest[0]); /* R */
+      VEC4_ADD(quadColor[1], source[1], dest[1]); /* G */
+      VEC4_ADD(quadColor[2], source[2], dest[2]); /* B */
+      break;
+   case PIPE_BLEND_SUBTRACT:
+      VEC4_SUB(quadColor[0], source[0], dest[0]); /* R */
+      VEC4_SUB(quadColor[1], source[1], dest[1]); /* G */
+      VEC4_SUB(quadColor[2], source[2], dest[2]); /* B */
+      break;
+   case PIPE_BLEND_REVERSE_SUBTRACT:
+      VEC4_SUB(quadColor[0], dest[0], source[0]); /* R */
+      VEC4_SUB(quadColor[1], dest[1], source[1]); /* G */
+      VEC4_SUB(quadColor[2], dest[2], source[2]); /* B */
+      break;
+   case PIPE_BLEND_MIN:
+      VEC4_MIN(quadColor[0], source[0], dest[0]); /* R */
+      VEC4_MIN(quadColor[1], source[1], dest[1]); /* G */
+      VEC4_MIN(quadColor[2], source[2], dest[2]); /* B */
+      break;
+   case PIPE_BLEND_MAX:
+      VEC4_MAX(quadColor[0], source[0], dest[0]); /* R */
+      VEC4_MAX(quadColor[1], source[1], dest[1]); /* G */
+      VEC4_MAX(quadColor[2], source[2], dest[2]); /* B */
+      break;
+   default:
+      assert(0);
+   }
+   
+   /*
+    * Combine A terms
+    */
+   switch (softpipe->blend->alpha_func) {
+   case PIPE_BLEND_ADD:
+      VEC4_ADD(quadColor[3], source[3], dest[3]); /* A */
+      break;
+   case PIPE_BLEND_SUBTRACT:
+      VEC4_SUB(quadColor[3], source[3], dest[3]); /* A */
+      break;
+   case PIPE_BLEND_REVERSE_SUBTRACT:
+      VEC4_SUB(quadColor[3], dest[3], source[3]); /* A */
+      break;
+   case PIPE_BLEND_MIN:
+      VEC4_MIN(quadColor[3], source[3], dest[3]); /* A */
+      break;
+   case PIPE_BLEND_MAX:
+      VEC4_MAX(quadColor[3], source[3], dest[3]); /* A */
+      break;
+   default:
+      abort();
+   }
+   
+   /* pass blended quad to next stage */
+   qs->next->run(qs->next, quad);
+}
+
+
+static void blend_begin(struct quad_stage *qs)
+{
+   qs->next->begin(qs->next);
+}
+
+
+static void blend_destroy(struct quad_stage *qs)
+{
+   FREE( qs );
+}
+
+
+struct quad_stage *sp_quad_blend_stage( struct softpipe_context *softpipe )
+{
+   struct quad_stage *stage = CALLOC_STRUCT(quad_stage);
+
+   stage->softpipe = softpipe;
+   stage->begin = blend_begin;
+   stage->run = blend_quad;
+   stage->destroy = blend_destroy;
+
+   return stage;
+}
diff --git a/src/gallium/drivers/softpipe/sp_quad_bufloop.c b/src/gallium/drivers/softpipe/sp_quad_bufloop.c
new file mode 100644
index 0000000000..2ae4e22a7d
--- /dev/null
+++ b/src/gallium/drivers/softpipe/sp_quad_bufloop.c
@@ -0,0 +1,72 @@
+
+#include "pipe/p_util.h"
+#include "sp_context.h"
+#include "sp_headers.h"
+#include "sp_surface.h"
+#include "sp_quad.h"
+
+
+/**
+ * Loop over colorbuffers, passing quad to next stage each time.
+ */
+static void
+cbuf_loop_quad(struct quad_stage *qs, struct quad_header *quad)
+{
+   struct softpipe_context *softpipe = qs->softpipe;
+   float tmp[4][QUAD_SIZE];
+   unsigned i;
+
+   assert(sizeof(quad->outputs.color) == sizeof(tmp));
+   assert(softpipe->framebuffer.num_cbufs <= PIPE_MAX_COLOR_BUFS);
+
+   /* make copy of original colors since they can get modified
+    * by blending and masking.
+    * XXX we won't have to do this if the fragment program actually emits
+    * N separate colors and we're drawing to N color buffers (MRT).
+    * But if we emitted one color and glDrawBuffer(GL_FRONT_AND_BACK) is
+    * in effect, we need to save/restore colors like this.
+    */
+   memcpy(tmp, quad->outputs.color, sizeof(tmp));
+
+   for (i = 0; i < softpipe->framebuffer.num_cbufs; i++) {
+      /* set current cbuffer */
+      softpipe->current_cbuf = i;
+
+      /* pass blended quad to next stage */
+      qs->next->run(qs->next, quad);
+
+      /* restore quad's colors for next buffer */
+      memcpy(quad->outputs.color, tmp, sizeof(tmp));
+   }
+}
+
+
+static void cbuf_loop_begin(struct quad_stage *qs)
+{
+   qs->next->begin(qs->next);
+}
+
+
+static void cbuf_loop_destroy(struct quad_stage *qs)
+{
+   FREE( qs );
+}
+
+
+/**
+ * Create the colorbuffer loop stage.
+ * This is used to implement multiple render targets and GL_FRONT_AND_BACK
+ * rendering.
+ */
+struct quad_stage *sp_quad_bufloop_stage( struct softpipe_context *softpipe )
+{
+   struct quad_stage *stage = CALLOC_STRUCT(quad_stage);
+
+   stage->softpipe = softpipe;
+   stage->begin = cbuf_loop_begin;
+   stage->run = cbuf_loop_quad;
+   stage->destroy = cbuf_loop_destroy;
+
+   return stage;
+}
+
diff --git a/src/gallium/drivers/softpipe/sp_quad_colormask.c b/src/gallium/drivers/softpipe/sp_quad_colormask.c
new file mode 100644
index 0000000000..1f09d900ca
--- /dev/null
+++ b/src/gallium/drivers/softpipe/sp_quad_colormask.c
@@ -0,0 +1,110 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+/**
+ * \brief  quad colormask stage
+ * \author Brian Paul
+ */
+
+#include "pipe/p_defines.h"
+#include "pipe/p_util.h"
+#include "sp_context.h"
+#include "sp_headers.h"
+#include "sp_surface.h"
+#include "sp_quad.h"
+#include "sp_tile_cache.h"
+
+
+
+/**
+ * XXX colormask could be rolled into blending...
+ */
+static void
+colormask_quad(struct quad_stage *qs, struct quad_header *quad)
+{
+   struct softpipe_context *softpipe = qs->softpipe;
+   float dest[4][QUAD_SIZE];
+   struct softpipe_cached_tile *tile
+      = sp_get_cached_tile(softpipe,
+                           softpipe->cbuf_cache[softpipe->current_cbuf],
+                           quad->x0, quad->y0);
+   float (*quadColor)[4] = quad->outputs.color;
+   uint i, j;
+
+   /* get/swizzle dest colors */
+   for (j = 0; j < QUAD_SIZE; j++) {
+      int x = (quad->x0 & (TILE_SIZE-1)) + (j & 1);
+      int y = (quad->y0 & (TILE_SIZE-1)) + (j >> 1);
+      for (i = 0; i < 4; i++) {
+         dest[i][j] = tile->data.color[y][x][i];
+      }
+   }
+
+   /* R */
+   if (!(softpipe->blend->colormask & PIPE_MASK_R))
+       COPY_4V(quadColor[0], dest[0]);
+
+   /* G */
+   if (!(softpipe->blend->colormask & PIPE_MASK_G))
+       COPY_4V(quadColor[1], dest[1]);
+
+   /* B */
+   if (!(softpipe->blend->colormask & PIPE_MASK_B))
+       COPY_4V(quadColor[2], dest[2]);
+
+   /* A */
+   if (!(softpipe->blend->colormask & PIPE_MASK_A))
+       COPY_4V(quadColor[3], dest[3]);
+
+   /* pass quad to next stage */
+   qs->next->run(qs->next, quad);
+}
+
+
+static void colormask_begin(struct quad_stage *qs)
+{
+   qs->next->begin(qs->next);
+}
+
+
+static void colormask_destroy(struct quad_stage *qs)
+{
+   FREE( qs );
+}
+
+
+struct quad_stage *sp_quad_colormask_stage( struct softpipe_context *softpipe )
+{
+   struct quad_stage *stage = CALLOC_STRUCT(quad_stage);
+
+   stage->softpipe = softpipe;
+   stage->begin = colormask_begin;
+   stage->run = colormask_quad;
+   stage->destroy = colormask_destroy;
+
+   return stage;
+}
diff --git a/src/gallium/drivers/softpipe/sp_quad_coverage.c b/src/gallium/drivers/softpipe/sp_quad_coverage.c
new file mode 100644
index 0000000000..b3d3fae22f
--- /dev/null
+++ b/src/gallium/drivers/softpipe/sp_quad_coverage.c
@@ -0,0 +1,88 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+
+/**
+ * \brief  Apply AA coverage to quad alpha valus
+ * \author  Brian Paul
+ */
+
+
+#include "pipe/p_defines.h"
+#include "pipe/p_util.h"
+#include "sp_context.h"
+#include "sp_headers.h"
+#include "sp_quad.h"
+
+
+/**
+ * Multiply quad's alpha values by the fragment coverage.
+ */
+static void
+coverage_quad(struct quad_stage *qs, struct quad_header *quad)
+{
+   struct softpipe_context *softpipe = qs->softpipe;
+
+   if ((softpipe->rasterizer->poly_smooth && quad->prim == PRIM_TRI) ||
+       (softpipe->rasterizer->line_smooth && quad->prim == PRIM_LINE) ||
+       (softpipe->rasterizer->point_smooth && quad->prim == PRIM_POINT)) {
+      float (*quadColor)[4] = quad->outputs.color;
+      unsigned j;
+      for (j = 0; j < QUAD_SIZE; j++) {
+         assert(quad->coverage[j] >= 0.0);
+         assert(quad->coverage[j] <= 1.0);
+         quadColor[3][j] *= quad->coverage[j];
+      }
+   }
+
+   qs->next->run(qs->next, quad);
+}
+
+
+static void coverage_begin(struct quad_stage *qs)
+{
+   qs->next->begin(qs->next);
+}
+
+
+static void coverage_destroy(struct quad_stage *qs)
+{
+   FREE( qs );
+}
+
+
+struct quad_stage *sp_quad_coverage_stage( struct softpipe_context *softpipe )
+{
+   struct quad_stage *stage = CALLOC_STRUCT(quad_stage);
+
+   stage->softpipe = softpipe;
+   stage->begin = coverage_begin;
+   stage->run = coverage_quad;
+   stage->destroy = coverage_destroy;
+
+   return stage;
+}
diff --git a/src/gallium/drivers/softpipe/sp_quad_depth_test.c b/src/gallium/drivers/softpipe/sp_quad_depth_test.c
new file mode 100644
index 0000000000..a9a0754f27
--- /dev/null
+++ b/src/gallium/drivers/softpipe/sp_quad_depth_test.c
@@ -0,0 +1,276 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+/**
+ * \brief  Quad depth testing
+ */
+
+#include "pipe/p_defines.h"
+#include "pipe/p_util.h"
+#include "sp_context.h"
+#include "sp_headers.h"
+#include "sp_surface.h"
+#include "sp_quad.h"
+#include "sp_tile_cache.h"
+
+
+/**
+ * Do depth testing for a quad.
+ * Not static since it's used by the stencil code.
+ */
+
+/*
+ * To increase efficiency, we should probably have multiple versions
+ * of this function that are specifically for Z16, Z32 and FP Z buffers.
+ * Try to effectively do that with codegen...
+ */
+
+void
+sp_depth_test_quad(struct quad_stage *qs, struct quad_header *quad)
+{
+   struct softpipe_context *softpipe = qs->softpipe;
+   struct pipe_surface *ps = softpipe->framebuffer.zsbuf;
+   const enum pipe_format format = ps->format;
+   unsigned bzzzz[QUAD_SIZE];  /**< Z values fetched from depth buffer */
+   unsigned qzzzz[QUAD_SIZE];  /**< Z values from the quad */
+   unsigned zmask = 0;
+   unsigned j;
+   struct softpipe_cached_tile *tile
+      = sp_get_cached_tile(softpipe, softpipe->zsbuf_cache, quad->x0, quad->y0);
+
+   assert(ps); /* shouldn't get here if there's no zbuffer */
+
+   /*
+    * Convert quad's float depth values to int depth values (qzzzz).
+    * If the Z buffer stores integer values, we _have_ to do the depth
+    * compares with integers (not floats).  Otherwise, the float->int->float
+    * conversion of Z values (which isn't an identity function) will cause
+    * Z-fighting errors.
+    *
+    * Also, get the zbuffer values (bzzzz) from the cached tile.
+    */
+   switch (format) {
+   case PIPE_FORMAT_Z16_UNORM:
+      {
+         float scale = 65535.0;
+
+         for (j = 0; j < QUAD_SIZE; j++) {
+            qzzzz[j] = (unsigned) (quad->outputs.depth[j] * scale);
+         }
+
+         for (j = 0; j < QUAD_SIZE; j++) {
+            int x = quad->x0 % TILE_SIZE + (j & 1);
+            int y = quad->y0 % TILE_SIZE + (j >> 1);
+            bzzzz[j] = tile->data.depth16[y][x];
+         }
+      }
+      break;
+   case PIPE_FORMAT_Z32_UNORM:
+      {
+         double scale = (double) (uint) ~0UL;
+
+         for (j = 0; j < QUAD_SIZE; j++) {
+            qzzzz[j] = (unsigned) (quad->outputs.depth[j] * scale);
+         }
+
+         for (j = 0; j < QUAD_SIZE; j++) {
+            int x = quad->x0 % TILE_SIZE + (j & 1);
+            int y = quad->y0 % TILE_SIZE + (j >> 1);
+            bzzzz[j] = tile->data.depth32[y][x];
+         }
+      }
+      break;
+   case PIPE_FORMAT_S8Z24_UNORM:
+      {
+         float scale = (float) ((1 << 24) - 1);
+
+         for (j = 0; j < QUAD_SIZE; j++) {
+            qzzzz[j] = (unsigned) (quad->outputs.depth[j] * scale);
+         }
+
+         for (j = 0; j < QUAD_SIZE; j++) {
+            int x = quad->x0 % TILE_SIZE + (j & 1);
+            int y = quad->y0 % TILE_SIZE + (j >> 1);
+            bzzzz[j] = tile->data.depth32[y][x] & 0xffffff;
+         }
+      }
+      break;
+   case PIPE_FORMAT_Z24S8_UNORM:
+      {
+         float scale = (float) ((1 << 24) - 1);
+
+         for (j = 0; j < QUAD_SIZE; j++) {
+            qzzzz[j] = (unsigned) (quad->outputs.depth[j] * scale);
+         }
+
+         for (j = 0; j < QUAD_SIZE; j++) {
+            int x = quad->x0 % TILE_SIZE + (j & 1);
+            int y = quad->y0 % TILE_SIZE + (j >> 1);
+            bzzzz[j] = tile->data.depth32[y][x] >> 8;
+         }
+      }
+      break;
+   default:
+      assert(0);
+   }
+
+   switch (softpipe->depth_stencil->depth.func) {
+   case PIPE_FUNC_NEVER:
+      /* zmask = 0 */
+      break;
+   case PIPE_FUNC_LESS:
+      /* Note this is pretty much a single sse or cell instruction.  
+       * Like this:  quad->mask &= (quad->outputs.depth < zzzz);
+       */
+      for (j = 0; j < QUAD_SIZE; j++) {
+	 if (qzzzz[j] < bzzzz[j]) 
+	    zmask |= 1 << j;
+      }
+      break;
+   case PIPE_FUNC_EQUAL:
+      for (j = 0; j < QUAD_SIZE; j++) {
+	 if (qzzzz[j] == bzzzz[j]) 
+	    zmask |= 1 << j;
+      }
+      break;
+   case PIPE_FUNC_LEQUAL:
+      for (j = 0; j < QUAD_SIZE; j++) {
+	 if (qzzzz[j] <= bzzzz[j]) 
+	    zmask |= (1 << j);
+      }
+      break;
+   case PIPE_FUNC_GREATER:
+      for (j = 0; j < QUAD_SIZE; j++) {
+	 if (qzzzz[j] > bzzzz[j]) 
+	    zmask |= (1 << j);
+      }
+      break;
+   case PIPE_FUNC_NOTEQUAL:
+      for (j = 0; j < QUAD_SIZE; j++) {
+	 if (qzzzz[j] != bzzzz[j]) 
+	    zmask |= (1 << j);
+      }
+      break;
+   case PIPE_FUNC_GEQUAL:
+      for (j = 0; j < QUAD_SIZE; j++) {
+	 if (qzzzz[j] >= bzzzz[j]) 
+	    zmask |= (1 << j);
+      }
+      break;
+   case PIPE_FUNC_ALWAYS:
+      zmask = MASK_ALL;
+      break;
+   default:
+      abort();
+   }
+
+   quad->mask &= zmask;
+
+   if (softpipe->depth_stencil->depth.writemask) {
+      
+      /* This is also efficient with sse / spe instructions: 
+       */
+      for (j = 0; j < QUAD_SIZE; j++) {
+	 if (quad->mask & (1 << j)) {
+	    bzzzz[j] = qzzzz[j];
+	 }
+      }
+
+      /* put updated Z values back into cached tile */
+      switch (format) {
+      case PIPE_FORMAT_Z16_UNORM:
+         for (j = 0; j < QUAD_SIZE; j++) {
+            int x = quad->x0 % TILE_SIZE + (j & 1);
+            int y = quad->y0 % TILE_SIZE + (j >> 1);
+            tile->data.depth16[y][x] = (ushort) bzzzz[j];
+         }
+         break;
+      case PIPE_FORMAT_Z32_UNORM:
+         for (j = 0; j < QUAD_SIZE; j++) {
+            int x = quad->x0 % TILE_SIZE + (j & 1);
+            int y = quad->y0 % TILE_SIZE + (j >> 1);
+            tile->data.depth32[y][x] = bzzzz[j];
+         }
+         break;
+      case PIPE_FORMAT_S8Z24_UNORM:
+         for (j = 0; j < QUAD_SIZE; j++) {
+            int x = quad->x0 % TILE_SIZE + (j & 1);
+            int y = quad->y0 % TILE_SIZE + (j >> 1);
+            uint s8z24 = tile->data.depth32[y][x];
+            s8z24 = (s8z24 & 0xff000000) | bzzzz[j];
+            tile->data.depth32[y][x] = s8z24;
+         }
+         break;
+      case PIPE_FORMAT_Z24S8_UNORM:
+         for (j = 0; j < QUAD_SIZE; j++) {
+            int x = quad->x0 % TILE_SIZE + (j & 1);
+            int y = quad->y0 % TILE_SIZE + (j >> 1);
+            uint z24s8 = tile->data.depth32[y][x];
+            z24s8 = (z24s8 & 0xff) | (bzzzz[j] << 24);
+            tile->data.depth32[y][x] = z24s8;
+         }
+         break;
+      default:
+         assert(0);
+      }
+   }
+}
+
+
+static void
+depth_test_quad(struct quad_stage *qs, struct quad_header *quad)
+{
+   sp_depth_test_quad(qs, quad);
+
+   if (quad->mask)
+      qs->next->run(qs->next, quad);
+}
+
+
+static void depth_test_begin(struct quad_stage *qs)
+{
+   qs->next->begin(qs->next);
+}
+
+
+static void depth_test_destroy(struct quad_stage *qs)
+{
+   FREE( qs );
+}
+
+
+struct quad_stage *sp_quad_depth_test_stage( struct softpipe_context *softpipe )
+{
+   struct quad_stage *stage = CALLOC_STRUCT(quad_stage);
+
+   stage->softpipe = softpipe;
+   stage->begin = depth_test_begin;
+   stage->run = depth_test_quad;
+   stage->destroy = depth_test_destroy;
+
+   return stage;
+}
diff --git a/src/gallium/drivers/softpipe/sp_quad_earlyz.c b/src/gallium/drivers/softpipe/sp_quad_earlyz.c
new file mode 100644
index 0000000000..22ea99049f
--- /dev/null
+++ b/src/gallium/drivers/softpipe/sp_quad_earlyz.c
@@ -0,0 +1,88 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+/**
+ * \brief  Quad early-z testing
+ */
+
+#include "pipe/p_defines.h"
+#include "pipe/p_util.h"
+#include "sp_headers.h"
+#include "sp_quad.h"
+
+
+/**
+ * All this stage does is compute the quad's Z values (which is normally
+ * done by the shading stage).
+ * The next stage will do the actual depth test.
+ */
+static void
+earlyz_quad(
+   struct quad_stage    *qs,
+   struct quad_header   *quad )
+{
+   const float fx = (float) quad->x0;
+   const float fy = (float) quad->y0;
+   const float dzdx = quad->posCoef->dadx[2];
+   const float dzdy = quad->posCoef->dady[2];
+   const float z0 = quad->posCoef->a0[2] + dzdx * fx + dzdy * fy;
+
+   quad->outputs.depth[0] = z0;
+   quad->outputs.depth[1] = z0 + dzdx;
+   quad->outputs.depth[2] = z0 + dzdy;
+   quad->outputs.depth[3] = z0 + dzdx + dzdy;
+
+   qs->next->run( qs->next, quad );
+}
+
+static void
+earlyz_begin(
+   struct quad_stage *qs )
+{
+   qs->next->begin( qs->next );
+}
+
+static void
+earlyz_destroy(
+   struct quad_stage *qs )
+{
+   FREE( qs );
+}
+
+struct quad_stage *
+sp_quad_earlyz_stage(
+   struct softpipe_context *softpipe )
+{
+   struct quad_stage *stage = CALLOC_STRUCT( quad_stage );
+
+   stage->softpipe = softpipe;
+   stage->begin = earlyz_begin;
+   stage->run = earlyz_quad;
+   stage->destroy = earlyz_destroy;
+
+   return stage;
+}
diff --git a/src/gallium/drivers/softpipe/sp_quad_fs.c b/src/gallium/drivers/softpipe/sp_quad_fs.c
new file mode 100644
index 0000000000..3316858413
--- /dev/null
+++ b/src/gallium/drivers/softpipe/sp_quad_fs.c
@@ -0,0 +1,390 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+/* Vertices are just an array of floats, with all the attributes
+ * packed.  We currently assume a layout like:
+ *
+ * attr[0][0..3] - window position
+ * attr[1..n][0..3] - remaining attributes.
+ *
+ * Attributes are assumed to be 4 floats wide but are packed so that
+ * all the enabled attributes run contiguously.
+ */
+
+#include "pipe/p_util.h"
+#include "pipe/p_defines.h"
+#include "pipe/p_shader_tokens.h"
+
+#include "x86/rtasm/x86sse.h"
+
+#ifdef MESA_LLVM
+#include "pipe/llvm/gallivm.h"
+#endif
+
+#include "sp_context.h"
+#include "sp_state.h"
+#include "sp_headers.h"
+#include "sp_quad.h"
+#include "sp_texture.h"
+#include "sp_tex_sample.h"
+
+
+struct quad_shade_stage
+{
+   struct quad_stage stage;
+   struct tgsi_sampler samplers[PIPE_MAX_SAMPLERS];
+   struct tgsi_exec_machine machine;
+   struct tgsi_exec_vector *inputs, *outputs;
+   int colorOutSlot, depthOutSlot;
+#ifdef MESA_LLVM
+   struct gallivm_prog *llvm_prog;
+#endif
+};
+
+
+/** cast wrapper */
+static INLINE struct quad_shade_stage *
+quad_shade_stage(struct quad_stage *qs)
+{
+   return (struct quad_shade_stage *) qs;
+}
+
+
+/**
+ * Compute quad X,Y,Z,W for the four fragments in a quad.
+ * Note that we only need to "compute" X and Y for the upper-left fragment.
+ * We could do less work if we're not depth testing, or there's no
+ * perspective-corrected attributes, but that's seldom.
+ */
+static void
+setup_pos_vector(const struct tgsi_interp_coef *coef,
+                 float x, float y,
+                 struct tgsi_exec_vector *quadpos)
+{
+   uint chan;
+   /* do X */
+   quadpos->xyzw[0].f[0] = x;
+   /* do Y */
+   quadpos->xyzw[1].f[0] = y;
+   /* do Z and W for all fragments in the quad */
+   for (chan = 2; chan < 4; chan++) {
+      const float dadx = coef->dadx[chan];
+      const float dady = coef->dady[chan];
+      const float a0 = coef->a0[chan] + dadx * x + dady * y;
+      quadpos->xyzw[chan].f[0] = a0;
+      quadpos->xyzw[chan].f[1] = a0 + dadx;
+      quadpos->xyzw[chan].f[2] = a0 + dady;
+      quadpos->xyzw[chan].f[3] = a0 + dadx + dady;
+   }
+}
+
+
+typedef void (XSTDCALL *codegen_function)(
+   const struct tgsi_exec_vector *input,
+   struct tgsi_exec_vector *output,
+   float (*constant)[4],
+   struct tgsi_exec_vector *temporary,
+   const struct tgsi_interp_coef *coef
+#if 0
+   ,const struct tgsi_exec_vector *quadPos
+#endif
+ );
+
+
+/**
+ * Execute fragment shader for the four fragments in the quad.
+ */
+static void
+shade_quad(
+   struct quad_stage *qs,
+   struct quad_header *quad )
+{
+   struct quad_shade_stage *qss = quad_shade_stage( qs );
+   struct softpipe_context *softpipe = qs->softpipe;
+   struct tgsi_exec_machine *machine = &qss->machine;
+
+   /* Consts do not require 16 byte alignment. */
+   machine->Consts = softpipe->mapped_constants[PIPE_SHADER_FRAGMENT];
+
+   machine->InterpCoefs = quad->coef;
+
+   /* Compute X, Y, Z, W vals for this quad */
+   setup_pos_vector(quad->posCoef, (float) quad->x0, (float) quad->y0, &machine->QuadPos);
+
+   /* run shader */
+#if defined(__i386__) || defined(__386__)
+   if( softpipe->use_sse ) {
+      codegen_function func = (codegen_function) x86_get_func( &softpipe->fs->sse2_program );
+      func(
+         machine->Inputs,
+         machine->Outputs,
+         machine->Consts,
+         machine->Temps,
+         machine->InterpCoefs
+#if 0
+         ,machine->QuadPos
+#endif
+           );
+      quad->mask &= ~(machine->Temps[TGSI_EXEC_TEMP_KILMASK_I].xyzw[TGSI_EXEC_TEMP_KILMASK_C].u[0]);
+   }
+   else
+#endif
+   {
+      quad->mask &= tgsi_exec_machine_run( machine );
+   }
+
+   /* store result color */
+   if (qss->colorOutSlot >= 0) {
+      /* XXX need to handle multiple color outputs someday */
+      assert(qss->stage.softpipe->fs->shader.output_semantic_name[qss->colorOutSlot]
+             == TGSI_SEMANTIC_COLOR);
+      memcpy(
+             quad->outputs.color,
+             &machine->Outputs[qss->colorOutSlot].xyzw[0].f[0],
+             sizeof( quad->outputs.color ) );
+   }
+
+   /*
+    * XXX the following code for updating quad->outputs.depth
+    * isn't really needed if we did early z testing.
+    */
+
+   /* store result Z */
+   if (qss->depthOutSlot >= 0) {
+      /* output[slot] is new Z */
+      uint i;
+      for (i = 0; i < 4; i++) {
+         quad->outputs.depth[i] = machine->Outputs[0].xyzw[2].f[i];
+      }
+   }
+   else {
+      /* copy input Z (which was interpolated by the executor) to output Z */
+      uint i;
+      for (i = 0; i < 4; i++) {
+         quad->outputs.depth[i] = machine->Inputs[0].xyzw[2].f[i];
+         /* XXX not sure the above line is always correct.  The following
+          * might be better:
+         quad->outputs.depth[i] = machine->QuadPos.xyzw[2].f[i];
+          */
+      }
+   }
+
+   /* shader may cull fragments */
+   if( quad->mask ) {
+      qs->next->run( qs->next, quad );
+   }
+}
+
+#if 0
+#ifdef MESA_LLVM
+#define DLLVM 0
+static void
+shade_quad_llvm(struct quad_stage *qs,
+                struct quad_header *quad)
+{
+   struct quad_shade_stage *qss = quad_shade_stage(qs);
+   struct softpipe_context *softpipe = qs->softpipe;
+   float dests[4][16][4] ALIGN16_ATTRIB;
+   float inputs[4][16][4] ALIGN16_ATTRIB;
+   const float fx = (float) quad->x0;
+   const float fy = (float) quad->y0;
+   struct gallivm_prog *llvm = qss->llvm_prog;
+
+   inputs[0][0][0] = fx;
+   inputs[1][0][0] = fx + 1.0f;
+   inputs[2][0][0] = fx;
+   inputs[3][0][0] = fx + 1.0f;
+
+   inputs[0][0][1] = fy;
+   inputs[1][0][1] = fy;
+   inputs[2][0][1] = fy + 1.0f;
+   inputs[3][0][1] = fy + 1.0f;
+#if DLLVM
+   debug_printf("MASK = %d\n", quad->mask);
+#endif
+   gallivm_prog_inputs_interpolate(llvm, inputs, quad->coef);
+#if DLLVM
+   for (int i = 0; i < 4; ++i) {
+      for (int j = 0; j < 2; ++j) {
+         debug_printf("IN(%d,%d) [%f %f %f %f]\n", i, j, 
+                inputs[i][j][0], inputs[i][j][1], inputs[i][j][2], inputs[i][j][3]);
+      }
+   }
+#endif
+
+   quad->mask &=
+      gallivm_fragment_shader_exec(llvm, fx, fy, dests, inputs,
+                                   softpipe->mapped_constants[PIPE_SHADER_FRAGMENT],
+                                   qss->samplers);
+#if DLLVM
+   debug_printf("OUT LLVM = 1[%f %f %f %f], 2[%f %f %f %f]\n",
+          dests[0][0][0], dests[0][0][1], dests[0][0][2], dests[0][0][3], 
+          dests[0][1][0], dests[0][1][1], dests[0][1][2], dests[0][1][3]);
+#endif
+
+   /* store result color */
+   if (qss->colorOutSlot >= 0) {
+      unsigned i;
+      /* XXX need to handle multiple color outputs someday */
+      assert(qss->stage.softpipe->fs->shader.output_semantic_name[qss->colorOutSlot]
+             == TGSI_SEMANTIC_COLOR);
+      for (i = 0; i < QUAD_SIZE; ++i) {
+         quad->outputs.color[0][i] = dests[i][qss->colorOutSlot][0];
+         quad->outputs.color[1][i] = dests[i][qss->colorOutSlot][1];
+         quad->outputs.color[2][i] = dests[i][qss->colorOutSlot][2];
+         quad->outputs.color[3][i] = dests[i][qss->colorOutSlot][3];
+      }
+   }
+#if DLLVM
+   for (int i = 0; i < QUAD_SIZE; ++i) {
+      debug_printf("QLLVM%d(%d) [%f, %f, %f, %f]\n", i, qss->colorOutSlot,
+             quad->outputs.color[0][i],
+             quad->outputs.color[1][i],
+             quad->outputs.color[2][i],
+             quad->outputs.color[3][i]);
+   }
+#endif
+
+   /* store result Z */
+   if (qss->depthOutSlot >= 0) {
+      /* output[slot] is new Z */
+      uint i;
+      for (i = 0; i < 4; i++) {
+         quad->outputs.depth[i] = dests[i][0][2];
+      }
+   }
+   else {
+      /* copy input Z (which was interpolated by the executor) to output Z */
+      uint i;
+      for (i = 0; i < 4; i++) {
+         quad->outputs.depth[i] = inputs[i][0][2];
+      }
+   }
+#if DLLVM
+   debug_printf("D [%f, %f, %f, %f] mask = %d\n",
+             quad->outputs.depth[0],
+             quad->outputs.depth[1],
+             quad->outputs.depth[2],
+             quad->outputs.depth[3], quad->mask);
+#endif
+
+   /* shader may cull fragments */
+   if( quad->mask ) {
+      qs->next->run( qs->next, quad );
+   }
+}
+#endif /*MESA_LLVM*/
+#endif
+
+/**
+ * Per-primitive (or per-begin?) setup
+ */
+static void shade_begin(struct quad_stage *qs)
+{
+   struct quad_shade_stage *qss = quad_shade_stage(qs);
+   struct softpipe_context *softpipe = qs->softpipe;
+   unsigned i;
+
+   /* set TGSI sampler state that varies */
+   for (i = 0; i < PIPE_MAX_SAMPLERS; i++) {
+      qss->samplers[i].state = softpipe->sampler[i];
+      qss->samplers[i].texture = &softpipe->texture[i]->base;
+   }
+
+#ifdef MESA_LLVM
+   qss->llvm_prog = softpipe->fs->llvm_prog;
+#endif
+   /* XXX only do this if the fragment shader changes... */
+   tgsi_exec_machine_init(&qss->machine,
+                          softpipe->fs->shader.tokens,
+                          PIPE_MAX_SAMPLERS,
+                          qss->samplers );
+
+   /* find output slots for depth, color */
+   qss->colorOutSlot = -1;
+   qss->depthOutSlot = -1;
+   for (i = 0; i < qss->stage.softpipe->fs->shader.num_outputs; i++) {
+      switch (qss->stage.softpipe->fs->shader.output_semantic_name[i]) {
+      case TGSI_SEMANTIC_POSITION:
+         qss->depthOutSlot = i;
+         break;
+      case TGSI_SEMANTIC_COLOR:
+         qss->colorOutSlot = i;
+         break;
+      }
+   }
+
+   qs->next->begin(qs->next);
+}
+
+
+static void shade_destroy(struct quad_stage *qs)
+{
+   struct quad_shade_stage *qss = (struct quad_shade_stage *) qs;
+
+   tgsi_exec_machine_free_data(&qss->machine);
+   FREE( qss->inputs );
+   FREE( qss->outputs );
+   FREE( qs );
+}
+
+
+struct quad_stage *sp_quad_shade_stage( struct softpipe_context *softpipe )
+{
+   struct quad_shade_stage *qss = CALLOC_STRUCT(quad_shade_stage);
+   uint i;
+
+   /* allocate storage for program inputs/outputs, aligned to 16 bytes */
+   qss->inputs = MALLOC(PIPE_ATTRIB_MAX * sizeof(*qss->inputs) + 16);
+   qss->outputs = MALLOC(PIPE_ATTRIB_MAX * sizeof(*qss->outputs) + 16);
+   qss->machine.Inputs = align16(qss->inputs);
+   qss->machine.Outputs = align16(qss->outputs);
+
+   qss->stage.softpipe = softpipe;
+   qss->stage.begin = shade_begin;
+#ifdef MESA_LLVM
+   /* disable until ported to accept
+    * x/y and soa layout
+   qss->stage.run = shade_quad_llvm;
+   */
+   softpipe->use_sse = FALSE;
+   qss->stage.run = shade_quad;
+#else
+   qss->stage.run = shade_quad;
+#endif
+   qss->stage.destroy = shade_destroy;
+
+   /* set TGSI sampler state that's constant */
+   for (i = 0; i < PIPE_MAX_SAMPLERS; i++) {
+      assert(softpipe->tex_cache[i]);
+      qss->samplers[i].get_samples = sp_get_samples;
+      qss->samplers[i].pipe = &softpipe->pipe;
+      qss->samplers[i].cache = softpipe->tex_cache[i];
+   }
+
+   return &qss->stage;
+}
diff --git a/src/gallium/drivers/softpipe/sp_quad_occlusion.c b/src/gallium/drivers/softpipe/sp_quad_occlusion.c
new file mode 100644
index 0000000000..54254df1f1
--- /dev/null
+++ b/src/gallium/drivers/softpipe/sp_quad_occlusion.c
@@ -0,0 +1,85 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+
+/**
+ * \brief  Quad occlusion counter stage
+ * \author  Brian Paul
+ */
+
+
+#include "pipe/p_defines.h"
+#include "pipe/p_util.h"
+#include "sp_context.h"
+#include "sp_headers.h"
+#include "sp_surface.h"
+#include "sp_quad.h"
+
+static unsigned count_bits( unsigned val )
+{
+   unsigned i;
+
+   for (i = 0; val ; val >>= 1)
+      i += (val & 1);
+
+   return i;
+}
+
+static void
+occlusion_count_quad(struct quad_stage *qs, struct quad_header *quad)
+{
+   struct softpipe_context *softpipe = qs->softpipe;
+
+   softpipe->occlusion_count += count_bits(quad->mask);
+
+   qs->next->run(qs->next, quad);
+}
+
+
+static void occlusion_begin(struct quad_stage *qs)
+{
+   qs->next->begin(qs->next);
+}
+
+
+static void occlusion_destroy(struct quad_stage *qs)
+{
+   FREE( qs );
+}
+
+
+struct quad_stage *sp_quad_occlusion_stage( struct softpipe_context *softpipe )
+{
+   struct quad_stage *stage = CALLOC_STRUCT(quad_stage);
+
+   stage->softpipe = softpipe;
+   stage->begin = occlusion_begin;
+   stage->run = occlusion_count_quad;
+   stage->destroy = occlusion_destroy;
+
+   return stage;
+}
diff --git a/src/gallium/drivers/softpipe/sp_quad_output.c b/src/gallium/drivers/softpipe/sp_quad_output.c
new file mode 100644
index 0000000000..cfe8f11808
--- /dev/null
+++ b/src/gallium/drivers/softpipe/sp_quad_output.c
@@ -0,0 +1,90 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+#include "pipe/p_util.h"
+#include "sp_context.h"
+#include "sp_headers.h"
+#include "sp_surface.h"
+#include "sp_quad.h"
+#include "sp_tile_cache.h"
+
+
+/**
+ * Write quad to framebuffer, taking mask into account.
+ *
+ * Note that surfaces support only full quad reads and writes.
+ */
+static void
+output_quad(struct quad_stage *qs, struct quad_header *quad)
+{
+   struct softpipe_context *softpipe = qs->softpipe;
+   struct softpipe_cached_tile *tile
+      = sp_get_cached_tile(softpipe,
+                           softpipe->cbuf_cache[softpipe->current_cbuf],
+                           quad->x0, quad->y0);
+   /* in-tile pos: */
+   const int itx = quad->x0 % TILE_SIZE;
+   const int ity = quad->y0 % TILE_SIZE;
+   float (*quadColor)[4] = quad->outputs.color;
+   int i, j;
+
+   /* get/swizzle dest colors */
+   for (j = 0; j < QUAD_SIZE; j++) {
+      if (quad->mask & (1 << j)) {
+         int x = itx + (j & 1);
+         int y = ity + (j >> 1);
+         for (i = 0; i < 4; i++) { /* loop over color chans */
+            tile->data.color[y][x][i] = quadColor[i][j];
+         }
+      }
+   }
+}
+
+
+static void output_begin(struct quad_stage *qs)
+{
+   assert(qs->next == NULL);
+}
+
+
+static void output_destroy(struct quad_stage *qs)
+{
+   FREE( qs );
+}
+
+
+struct quad_stage *sp_quad_output_stage( struct softpipe_context *softpipe )
+{
+   struct quad_stage *stage = CALLOC_STRUCT(quad_stage);
+
+   stage->softpipe = softpipe;
+   stage->begin = output_begin;
+   stage->run = output_quad;
+   stage->destroy = output_destroy;
+
+   return stage;
+}
diff --git a/src/gallium/drivers/softpipe/sp_quad_stencil.c b/src/gallium/drivers/softpipe/sp_quad_stencil.c
new file mode 100644
index 0000000000..92a0da0083
--- /dev/null
+++ b/src/gallium/drivers/softpipe/sp_quad_stencil.c
@@ -0,0 +1,352 @@
+
+/**
+ * \brief Quad stencil testing
+ */
+
+
+#include "sp_context.h"
+#include "sp_headers.h"
+#include "sp_surface.h"
+#include "sp_tile_cache.h"
+#include "sp_quad.h"
+#include "pipe/p_defines.h"
+#include "pipe/p_util.h"
+
+
+/** Only 8-bit stencil supported */
+#define STENCIL_MAX 0xff
+
+
+/**
+ * Do the basic stencil test (compare stencil buffer values against the
+ * reference value.
+ *
+ * \param stencilVals  the stencil values from the stencil buffer
+ * \param func  the stencil func (PIPE_FUNC_x)
+ * \param ref  the stencil reference value
+ * \param valMask  the stencil value mask indicating which bits of the stencil
+ *                 values and ref value are to be used.
+ * \return mask indicating which pixels passed the stencil test
+ */
+static unsigned
+do_stencil_test(const ubyte stencilVals[QUAD_SIZE], unsigned func,
+                unsigned ref, unsigned valMask)
+{
+   unsigned passMask = 0x0;
+   unsigned j;
+
+   ref &= valMask;
+
+   switch (func) {
+   case PIPE_FUNC_NEVER:
+      /* passMask = 0x0 */
+      break;
+   case PIPE_FUNC_LESS:
+      for (j = 0; j < QUAD_SIZE; j++) {
+         if (ref < (stencilVals[j] & valMask)) {
+            passMask |= (1 << j);
+         }
+      }
+      break;
+   case PIPE_FUNC_EQUAL:
+      for (j = 0; j < QUAD_SIZE; j++) {
+         if (ref == (stencilVals[j] & valMask)) {
+            passMask |= (1 << j);
+         }
+      }
+      break;
+   case PIPE_FUNC_LEQUAL:
+      for (j = 0; j < QUAD_SIZE; j++) {
+         if (ref <= (stencilVals[j] & valMask)) {
+            passMask |= (1 << j);
+         }
+      }
+      break;
+   case PIPE_FUNC_GREATER:
+      for (j = 0; j < QUAD_SIZE; j++) {
+         if (ref > (stencilVals[j] & valMask)) {
+            passMask |= (1 << j);
+         }
+      }
+      break;
+   case PIPE_FUNC_NOTEQUAL:
+      for (j = 0; j < QUAD_SIZE; j++) {
+         if (ref != (stencilVals[j] & valMask)) {
+            passMask |= (1 << j);
+         }
+      }
+      break;
+   case PIPE_FUNC_GEQUAL:
+      for (j = 0; j < QUAD_SIZE; j++) {
+         if (ref >= (stencilVals[j] & valMask)) {
+            passMask |= (1 << j);
+         }
+      }
+      break;
+   case PIPE_FUNC_ALWAYS:
+      passMask = MASK_ALL;
+      break;
+   default:
+      assert(0);
+   }
+
+   return passMask;
+}
+
+
+/**
+ * Apply the stencil operator to stencil values.
+ *
+ * \param stencilVals  the stencil buffer values (read and written)
+ * \param mask  indicates which pixels to update
+ * \param op  the stencil operator (PIPE_STENCIL_OP_x)
+ * \param ref  the stencil reference value
+ * \param wrtMask  writemask controlling which bits are changed in the
+ *                 stencil values
+ */
+static void
+apply_stencil_op(ubyte stencilVals[QUAD_SIZE],
+                 unsigned mask, unsigned op, ubyte ref, ubyte wrtMask)
+{
+   unsigned j;
+   ubyte newstencil[QUAD_SIZE];
+
+   for (j = 0; j < QUAD_SIZE; j++) {
+      newstencil[j] = stencilVals[j];
+   }
+
+   switch (op) {
+   case PIPE_STENCIL_OP_KEEP:
+      /* no-op */
+      break;
+   case PIPE_STENCIL_OP_ZERO:
+      for (j = 0; j < QUAD_SIZE; j++) {
+         if (mask & (1 << j)) {
+            newstencil[j] = 0;
+         }
+      }
+      break;
+   case PIPE_STENCIL_OP_REPLACE:
+      for (j = 0; j < QUAD_SIZE; j++) {
+         if (mask & (1 << j)) {
+            newstencil[j] = ref;
+         }
+      }
+      break;
+   case PIPE_STENCIL_OP_INCR:
+      for (j = 0; j < QUAD_SIZE; j++) {
+         if (mask & (1 << j)) {
+            if (stencilVals[j] < STENCIL_MAX) {
+               newstencil[j] = stencilVals[j] + 1;
+            }
+         }
+      }
+      break;
+   case PIPE_STENCIL_OP_DECR:
+      for (j = 0; j < QUAD_SIZE; j++) {
+         if (mask & (1 << j)) {
+            if (stencilVals[j] > 0) {
+               newstencil[j] = stencilVals[j] - 1;
+            }
+         }
+      }
+      break;
+   case PIPE_STENCIL_OP_INCR_WRAP:
+      for (j = 0; j < QUAD_SIZE; j++) {
+         if (mask & (1 << j)) {
+            newstencil[j] = stencilVals[j] + 1;
+         }
+      }
+      break;
+   case PIPE_STENCIL_OP_DECR_WRAP:
+      for (j = 0; j < QUAD_SIZE; j++) {
+         if (mask & (1 << j)) {
+            newstencil[j] = stencilVals[j] - 1;
+         }
+      }
+      break;
+   case PIPE_STENCIL_OP_INVERT:
+      for (j = 0; j < QUAD_SIZE; j++) {
+         if (mask & (1 << j)) {
+            newstencil[j] = ~stencilVals[j];
+         }
+      }
+      break;
+   default:
+      assert(0);
+   }
+
+   /*
+    * update the stencil values
+    */
+   if (wrtMask != STENCIL_MAX) {
+      /* apply bit-wise stencil buffer writemask */
+      for (j = 0; j < QUAD_SIZE; j++) {
+         stencilVals[j] = (wrtMask & newstencil[j]) | (~wrtMask & stencilVals[j]);
+      }
+   }
+   else {
+      for (j = 0; j < QUAD_SIZE; j++) {
+         stencilVals[j] = newstencil[j];
+      }
+   }
+}
+
+
+/**
+ * Do stencil (and depth) testing.  Stenciling depends on the outcome of
+ * depth testing.
+ */
+static void
+stencil_test_quad(struct quad_stage *qs, struct quad_header *quad)
+{
+   struct softpipe_context *softpipe = qs->softpipe;
+   struct pipe_surface *ps = softpipe->framebuffer.zsbuf;
+   unsigned func, zFailOp, zPassOp, failOp;
+   ubyte ref, wrtMask, valMask;
+   ubyte stencilVals[QUAD_SIZE];
+   struct softpipe_cached_tile *tile
+      = sp_get_cached_tile(softpipe, softpipe->zsbuf_cache, quad->x0, quad->y0);
+   uint j;
+   uint face = quad->facing;
+
+   if (!softpipe->depth_stencil->stencil[1].enabled) {
+      /* single-sided stencil test, use front (face=0) state */
+      face = 0;
+   }
+
+   /* choose front or back face function, operator, etc */
+   /* XXX we could do these initializations once per primitive */
+   func    = softpipe->depth_stencil->stencil[face].func;
+   failOp  = softpipe->depth_stencil->stencil[face].fail_op;
+   zFailOp = softpipe->depth_stencil->stencil[face].zfail_op;
+   zPassOp = softpipe->depth_stencil->stencil[face].zpass_op;
+   ref     = softpipe->depth_stencil->stencil[face].ref_value;
+   wrtMask = softpipe->depth_stencil->stencil[face].write_mask;
+   valMask = softpipe->depth_stencil->stencil[face].value_mask;
+
+   assert(ps); /* shouldn't get here if there's no stencil buffer */
+
+   /* get stencil values from cached tile */
+   switch (ps->format) {
+   case PIPE_FORMAT_S8Z24_UNORM:
+      for (j = 0; j < QUAD_SIZE; j++) {
+         int x = quad->x0 % TILE_SIZE + (j & 1);
+         int y = quad->y0 % TILE_SIZE + (j >> 1);
+         stencilVals[j] = tile->data.depth32[y][x] >> 24;
+      }
+      break;
+   case PIPE_FORMAT_Z24S8_UNORM:
+      for (j = 0; j < QUAD_SIZE; j++) {
+         int x = quad->x0 % TILE_SIZE + (j & 1);
+         int y = quad->y0 % TILE_SIZE + (j >> 1);
+         stencilVals[j] = tile->data.depth32[y][x] & 0xff;
+      }
+      break;
+   case PIPE_FORMAT_U_S8:
+      for (j = 0; j < QUAD_SIZE; j++) {
+         int x = quad->x0 % TILE_SIZE + (j & 1);
+         int y = quad->y0 % TILE_SIZE + (j >> 1);
+         stencilVals[j] = tile->data.stencil8[y][x];
+      }
+      break;
+   default:
+      assert(0);
+   }
+
+   /* do the stencil test first */
+   {
+      unsigned passMask, failMask;
+      passMask = do_stencil_test(stencilVals, func, ref, valMask);
+      failMask = quad->mask & ~passMask;
+      quad->mask &= passMask;
+
+      if (failOp != PIPE_STENCIL_OP_KEEP) {
+         apply_stencil_op(stencilVals, failMask, failOp, ref, wrtMask);
+      }
+   }
+
+   if (quad->mask) {
+      /* now the pixels that passed the stencil test are depth tested */
+      if (softpipe->depth_stencil->depth.enabled) {
+         const unsigned origMask = quad->mask;
+
+         sp_depth_test_quad(qs, quad);  /* quad->mask is updated */
+
+         /* update stencil buffer values according to z pass/fail result */
+         if (zFailOp != PIPE_STENCIL_OP_KEEP) {
+            const unsigned failMask = origMask & ~quad->mask;
+            apply_stencil_op(stencilVals, failMask, zFailOp, ref, wrtMask);
+         }
+
+         if (zPassOp != PIPE_STENCIL_OP_KEEP) {
+            const unsigned passMask = origMask & quad->mask;
+            apply_stencil_op(stencilVals, passMask, zPassOp, ref, wrtMask);
+         }
+      }
+      else {
+         /* no depth test, apply Zpass operator to stencil buffer values */
+         apply_stencil_op(stencilVals, quad->mask, zPassOp, ref, wrtMask);
+      }
+
+   }
+
+   /* put new stencil values into cached tile */
+   switch (ps->format) {
+   case PIPE_FORMAT_S8Z24_UNORM:
+      for (j = 0; j < QUAD_SIZE; j++) {
+         int x = quad->x0 % TILE_SIZE + (j & 1);
+         int y = quad->y0 % TILE_SIZE + (j >> 1);
+         uint s8z24 = tile->data.depth32[y][x];
+         s8z24 = (stencilVals[j] << 24) | (s8z24 & 0xffffff);
+         tile->data.depth32[y][x] = s8z24;
+      }
+      break;
+   case PIPE_FORMAT_Z24S8_UNORM:
+      for (j = 0; j < QUAD_SIZE; j++) {
+         int x = quad->x0 % TILE_SIZE + (j & 1);
+         int y = quad->y0 % TILE_SIZE + (j >> 1);
+         uint z24s8 = tile->data.depth32[y][x];
+         z24s8 = (z24s8 & 0xffffff00) | stencilVals[j];
+         tile->data.depth32[y][x] = z24s8;
+      }
+      break;
+   case PIPE_FORMAT_U_S8:
+      for (j = 0; j < QUAD_SIZE; j++) {
+         int x = quad->x0 % TILE_SIZE + (j & 1);
+         int y = quad->y0 % TILE_SIZE + (j >> 1);
+         tile->data.stencil8[y][x] = stencilVals[j];
+      }
+      break;
+   default:
+      assert(0);
+   }
+
+   if (quad->mask)
+      qs->next->run(qs->next, quad);
+}
+
+
+static void stencil_begin(struct quad_stage *qs)
+{
+   qs->next->begin(qs->next);
+}
+
+
+static void stencil_destroy(struct quad_stage *qs)
+{
+   FREE( qs );
+}
+
+
+struct quad_stage *sp_quad_stencil_test_stage( struct softpipe_context *softpipe )
+{
+   struct quad_stage *stage = CALLOC_STRUCT(quad_stage);
+
+   stage->softpipe = softpipe;
+   stage->begin = stencil_begin;
+   stage->run = stencil_test_quad;
+   stage->destroy = stencil_destroy;
+
+   return stage;
+}
diff --git a/src/gallium/drivers/softpipe/sp_quad_stipple.c b/src/gallium/drivers/softpipe/sp_quad_stipple.c
new file mode 100644
index 0000000000..8660432259
--- /dev/null
+++ b/src/gallium/drivers/softpipe/sp_quad_stipple.c
@@ -0,0 +1,94 @@
+
+/**
+ * quad polygon stipple stage
+ */
+
+#include "sp_context.h"
+#include "sp_headers.h"
+#include "sp_quad.h"
+#include "pipe/p_defines.h"
+#include "pipe/p_util.h"
+
+
+/**
+ * Apply polygon stipple to quads produced by triangle rasterization
+ */
+static void
+stipple_quad(struct quad_stage *qs, struct quad_header *quad)
+{
+   static const uint bit31 = 1 << 31;
+   static const uint bit30 = 1 << 30;
+
+   if (quad->prim == PRIM_TRI) {
+      struct softpipe_context *softpipe = qs->softpipe;
+      /* need to invert Y to index into OpenGL's stipple pattern */
+      int y0, y1;
+      uint stipple0, stipple1;
+      if (softpipe->rasterizer->origin_lower_left) {
+         y0 = softpipe->framebuffer.cbufs[0]->height - 1 - quad->y0;
+         y1 = y0 - 1;
+      }
+      else {
+         y0 = quad->y0;
+         y1 = y0 + 1;
+      }
+      stipple0 = softpipe->poly_stipple.stipple[y0 % 32];
+      stipple1 = softpipe->poly_stipple.stipple[y1 % 32];
+
+#if 1
+      {
+      const int col0 = quad->x0 % 32;
+      if ((stipple0 & (bit31 >> col0)) == 0)
+         quad->mask &= ~MASK_TOP_LEFT;
+
+      if ((stipple0 & (bit30 >> col0)) == 0)
+         quad->mask &= ~MASK_TOP_RIGHT;
+
+      if ((stipple1 & (bit31 >> col0)) == 0)
+         quad->mask &= ~MASK_BOTTOM_LEFT;
+
+      if ((stipple1 & (bit30 >> col0)) == 0)
+         quad->mask &= ~MASK_BOTTOM_RIGHT;
+      }
+#else
+      /* We'd like to use this code, but we'd need to redefine
+       * MASK_TOP_LEFT to be (1 << 1) and MASK_TOP_RIGHT to be (1 << 0),
+       * and similarly for the BOTTOM bits.  But that may have undesirable
+       * side effects elsewhere.
+       */
+      const int col0 = 30 - (quad->x0 % 32);
+      quad->mask &= (((stipple0 >> col0) & 0x3) | 
+                     (((stipple1 >> col0) & 0x3) << 2));
+#endif
+      if (!quad->mask)
+         return;
+   }
+
+   qs->next->run(qs->next, quad);
+}
+
+
+static void stipple_begin(struct quad_stage *qs)
+{
+   qs->next->begin(qs->next);
+}
+
+
+static void stipple_destroy(struct quad_stage *qs)
+{
+   FREE( qs );
+}
+
+
+struct quad_stage *
+sp_quad_polygon_stipple_stage( struct softpipe_context *softpipe )
+{
+   struct quad_stage *stage = CALLOC_STRUCT(quad_stage);
+
+   stage->softpipe = softpipe;
+   stage->begin = stipple_begin;
+   stage->run = stipple_quad;
+   stage->destroy = stipple_destroy;
+
+   return stage;
+}
diff --git a/src/gallium/drivers/softpipe/sp_query.c b/src/gallium/drivers/softpipe/sp_query.c
new file mode 100644
index 0000000000..6a8a43aeda
--- /dev/null
+++ b/src/gallium/drivers/softpipe/sp_query.c
@@ -0,0 +1,107 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+/* Author:
+ *    Keith Whitwell <keith@tungstengraphics.com>
+ */
+
+#include "pipe/draw/draw_context.h"
+#include "pipe/p_defines.h"
+#include "pipe/p_inlines.h"
+#include "pipe/p_util.h"
+#include "sp_context.h"
+#include "sp_query.h"
+
+struct softpipe_query {
+   uint64 start;
+   uint64 end;
+};
+
+
+static struct softpipe_query *softpipe_query( struct pipe_query *p )
+{
+   return (struct softpipe_query *)p;
+}
+
+static struct pipe_query *
+softpipe_create_query(struct pipe_context *pipe, 
+		      unsigned type)
+{
+   assert(type == PIPE_QUERY_OCCLUSION_COUNTER);
+   return (struct pipe_query *)CALLOC_STRUCT( softpipe_query );
+}
+
+
+static void
+softpipe_destroy_query(struct pipe_context *pipe, struct pipe_query *q)
+{
+   FREE(q);
+}
+
+
+static void
+softpipe_begin_query(struct pipe_context *pipe, struct pipe_query *q)
+{
+   struct softpipe_context *softpipe = softpipe_context( pipe );
+   struct softpipe_query *sq = softpipe_query(q);
+   
+   sq->start = softpipe->occlusion_count;
+}
+
+
+static void
+softpipe_end_query(struct pipe_context *pipe, struct pipe_query *q)
+{
+   struct softpipe_context *softpipe = softpipe_context( pipe );
+   struct softpipe_query *sq = softpipe_query(q);
+
+   sq->end = softpipe->occlusion_count;
+}
+
+
+static boolean
+softpipe_get_query_result(struct pipe_context *pipe, 
+			  struct pipe_query *q,
+			  boolean wait,
+			  uint64 *result )
+{
+   struct softpipe_query *sq = softpipe_query(q);
+   *result = sq->end - sq->start;
+   return TRUE;
+}
+
+
+void softpipe_init_query_funcs(struct softpipe_context *softpipe )
+{
+   softpipe->pipe.create_query = softpipe_create_query;
+   softpipe->pipe.destroy_query = softpipe_destroy_query;
+   softpipe->pipe.begin_query = softpipe_begin_query;
+   softpipe->pipe.end_query = softpipe_end_query;
+   softpipe->pipe.get_query_result = softpipe_get_query_result;
+}
+
+
diff --git a/src/gallium/drivers/softpipe/sp_query.h b/src/gallium/drivers/softpipe/sp_query.h
new file mode 100644
index 0000000000..05060a4575
--- /dev/null
+++ b/src/gallium/drivers/softpipe/sp_query.h
@@ -0,0 +1,39 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+/* Author:
+ *    Keith Whitwell
+ */
+
+#ifndef SP_QUERY_H
+#define SP_QUERY_H
+
+struct softpipe_context;
+extern void softpipe_init_query_funcs(struct softpipe_context * );
+
+
+#endif /* SP_QUERY_H */
diff --git a/src/gallium/drivers/softpipe/sp_state.h b/src/gallium/drivers/softpipe/sp_state.h
new file mode 100644
index 0000000000..b79db0d1f1
--- /dev/null
+++ b/src/gallium/drivers/softpipe/sp_state.h
@@ -0,0 +1,187 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+/* Authors:  Keith Whitwell <keith@tungstengraphics.com>
+ */
+
+#ifndef SP_STATE_H
+#define SP_STATE_H
+
+#include "pipe/p_state.h"
+
+#include "x86/rtasm/x86sse.h"
+
+
+#define SP_NEW_VIEWPORT      0x1
+#define SP_NEW_RASTERIZER    0x2
+#define SP_NEW_FS            0x4
+#define SP_NEW_BLEND         0x8
+#define SP_NEW_CLIP          0x10
+#define SP_NEW_SCISSOR       0x20
+#define SP_NEW_STIPPLE       0x40
+#define SP_NEW_FRAMEBUFFER   0x80
+#define SP_NEW_DEPTH_STENCIL_ALPHA 0x100
+#define SP_NEW_CONSTANTS     0x200
+#define SP_NEW_SAMPLER       0x400
+#define SP_NEW_TEXTURE       0x800
+#define SP_NEW_VERTEX        0x1000
+#define SP_NEW_VS            0x2000
+#define SP_NEW_QUERY         0x4000
+
+
+
+#ifdef MESA_LLVM
+struct gallivm_prog;
+#endif
+
+
+struct vertex_info;
+
+
+/** Subclass of pipe_shader_state */
+struct sp_fragment_shader_state {
+   struct pipe_shader_state   shader;
+#if defined(__i386__) || defined(__386__)
+   struct x86_function        sse2_program;
+#endif
+#ifdef MESA_LLVM
+   struct gallivm_prog *llvm_prog;
+#endif
+};
+
+
+/** Subclass of pipe_shader_state */
+struct sp_vertex_shader_state {
+   struct pipe_shader_state shader;
+   struct draw_vertex_shader *draw_data;
+};
+
+
+
+void *
+softpipe_create_blend_state(struct pipe_context *,
+                            const struct pipe_blend_state *);
+void softpipe_bind_blend_state(struct pipe_context *,
+                               void *);
+void softpipe_delete_blend_state(struct pipe_context *,
+                                 void *);
+
+void *
+softpipe_create_sampler_state(struct pipe_context *,
+                              const struct pipe_sampler_state *);
+void softpipe_bind_sampler_state(struct pipe_context *, unsigned, void *);
+void softpipe_delete_sampler_state(struct pipe_context *, void *);
+
+void *
+softpipe_create_depth_stencil_state(struct pipe_context *,
+                                    const struct pipe_depth_stencil_alpha_state *);
+void softpipe_bind_depth_stencil_state(struct pipe_context *, void *);
+void softpipe_delete_depth_stencil_state(struct pipe_context *, void *);
+
+void *
+softpipe_create_rasterizer_state(struct pipe_context *,
+                                 const struct pipe_rasterizer_state *);
+void softpipe_bind_rasterizer_state(struct pipe_context *, void *);
+void softpipe_delete_rasterizer_state(struct pipe_context *, void *);
+
+void softpipe_set_framebuffer_state( struct pipe_context *,
+			     const struct pipe_framebuffer_state * );
+
+void softpipe_set_blend_color( struct pipe_context *pipe,
+                               const struct pipe_blend_color *blend_color );
+
+void softpipe_set_clip_state( struct pipe_context *,
+			     const struct pipe_clip_state * );
+
+void softpipe_set_constant_buffer(struct pipe_context *,
+                                  uint shader, uint index,
+                                  const struct pipe_constant_buffer *buf);
+
+void *softpipe_create_fs_state(struct pipe_context *,
+                               const struct pipe_shader_state *);
+void softpipe_bind_fs_state(struct pipe_context *, void *);
+void softpipe_delete_fs_state(struct pipe_context *, void *);
+void *softpipe_create_vs_state(struct pipe_context *,
+                               const struct pipe_shader_state *);
+void softpipe_bind_vs_state(struct pipe_context *, void *);
+void softpipe_delete_vs_state(struct pipe_context *, void *);
+
+void softpipe_set_polygon_stipple( struct pipe_context *,
+				  const struct pipe_poly_stipple * );
+
+void softpipe_set_scissor_state( struct pipe_context *,
+                                 const struct pipe_scissor_state * );
+
+void softpipe_set_sampler_texture( struct pipe_context *,
+                                 unsigned unit,
+                                 struct pipe_texture * );
+
+void softpipe_set_viewport_state( struct pipe_context *,
+                                  const struct pipe_viewport_state * );
+
+void softpipe_set_vertex_element(struct pipe_context *,
+                                 unsigned index,
+                                 const struct pipe_vertex_element *);
+
+void softpipe_set_vertex_buffer(struct pipe_context *,
+                                unsigned index,
+                                const struct pipe_vertex_buffer *);
+
+
+void softpipe_update_derived( struct softpipe_context *softpipe );
+
+
+boolean softpipe_draw_arrays(struct pipe_context *pipe, unsigned mode,
+			     unsigned start, unsigned count);
+
+boolean softpipe_draw_elements(struct pipe_context *pipe,
+			       struct pipe_buffer *indexBuffer,
+			       unsigned indexSize,
+			       unsigned mode, unsigned start, unsigned count);
+
+
+void
+softpipe_map_surfaces(struct softpipe_context *sp);
+
+void
+softpipe_unmap_surfaces(struct softpipe_context *sp);
+
+void
+softpipe_map_texture_surfaces(struct softpipe_context *sp);
+
+void
+softpipe_unmap_texture_surfaces(struct softpipe_context *sp);
+
+
+struct vertex_info *
+softpipe_get_vertex_info(struct softpipe_context *softpipe);
+
+struct vertex_info *
+softpipe_get_vbuf_vertex_info(struct softpipe_context *softpipe);
+
+
+#endif
diff --git a/src/gallium/drivers/softpipe/sp_state_blend.c b/src/gallium/drivers/softpipe/sp_state_blend.c
new file mode 100644
index 0000000000..2d40d6bd8f
--- /dev/null
+++ b/src/gallium/drivers/softpipe/sp_state_blend.c
@@ -0,0 +1,98 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+/* Authors:  Keith Whitwell <keith@tungstengraphics.com>
+ */
+
+#include "pipe/p_util.h"
+#include "sp_context.h"
+#include "sp_state.h"
+
+
+void *
+softpipe_create_blend_state(struct pipe_context *pipe,
+                            const struct pipe_blend_state *blend)
+{
+   return mem_dup(blend, sizeof(*blend));
+}
+
+void softpipe_bind_blend_state( struct pipe_context *pipe,
+                                void *blend )
+{
+   struct softpipe_context *softpipe = softpipe_context(pipe);
+
+   softpipe->blend = (const struct pipe_blend_state *)blend;
+
+   softpipe->dirty |= SP_NEW_BLEND;
+}
+
+void softpipe_delete_blend_state(struct pipe_context *pipe,
+                                 void *blend)
+{
+   FREE( blend );
+}
+
+
+void softpipe_set_blend_color( struct pipe_context *pipe,
+			     const struct pipe_blend_color *blend_color )
+{
+   struct softpipe_context *softpipe = softpipe_context(pipe);
+
+   softpipe->blend_color = *blend_color;
+
+   softpipe->dirty |= SP_NEW_BLEND;
+}
+
+
+/** XXX move someday?  Or consolidate all these simple state setters
+ * into one file.
+ */
+
+
+void *
+softpipe_create_depth_stencil_state(struct pipe_context *pipe,
+				    const struct pipe_depth_stencil_alpha_state *depth_stencil)
+{
+   return mem_dup(depth_stencil, sizeof(*depth_stencil));
+}
+
+void
+softpipe_bind_depth_stencil_state(struct pipe_context *pipe,
+                                  void *depth_stencil)
+{
+   struct softpipe_context *softpipe = softpipe_context(pipe);
+
+   softpipe->depth_stencil = (const struct pipe_depth_stencil_alpha_state *)depth_stencil;
+
+   softpipe->dirty |= SP_NEW_DEPTH_STENCIL_ALPHA;
+}
+
+void
+softpipe_delete_depth_stencil_state(struct pipe_context *pipe, void *depth)
+{
+   FREE( depth );
+}
diff --git a/src/gallium/drivers/softpipe/sp_state_clip.c b/src/gallium/drivers/softpipe/sp_state_clip.c
new file mode 100644
index 0000000000..08c5f06d05
--- /dev/null
+++ b/src/gallium/drivers/softpipe/sp_state_clip.c
@@ -0,0 +1,83 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+/* Authors:  Keith Whitwell <keith@tungstengraphics.com>
+ */
+#include "sp_context.h"
+#include "sp_state.h"
+#include "pipe/draw/draw_context.h"
+
+
+void softpipe_set_clip_state( struct pipe_context *pipe,
+			     const struct pipe_clip_state *clip )
+{
+   struct softpipe_context *softpipe = softpipe_context(pipe);
+
+   /* pass the clip state to the draw module */
+   draw_set_clip_state(softpipe->draw, clip);
+}
+
+
+
+/* Called when driver state tracker notices changes to the viewport
+ * matrix:
+ */
+void softpipe_set_viewport_state( struct pipe_context *pipe,
+                                  const struct pipe_viewport_state *viewport )
+{
+   struct softpipe_context *softpipe = softpipe_context(pipe);
+
+   softpipe->viewport = *viewport; /* struct copy */
+   softpipe->dirty |= SP_NEW_VIEWPORT;
+
+   /* pass the viewport info to the draw module */
+   draw_set_viewport_state(softpipe->draw, viewport);
+
+   /* Using tnl/ and vf/ modules is temporary while getting started.
+    * Full pipe will have vertex shader, vertex fetch of its own.
+    */
+}
+
+
+void softpipe_set_scissor_state( struct pipe_context *pipe,
+                                 const struct pipe_scissor_state *scissor )
+{
+   struct softpipe_context *softpipe = softpipe_context(pipe);
+
+   memcpy( &softpipe->scissor, scissor, sizeof(*scissor) );
+   softpipe->dirty |= SP_NEW_SCISSOR;
+}
+
+
+void softpipe_set_polygon_stipple( struct pipe_context *pipe,
+                                   const struct pipe_poly_stipple *stipple )
+{
+   struct softpipe_context *softpipe = softpipe_context(pipe);
+
+   memcpy( &softpipe->poly_stipple, stipple, sizeof(*stipple) );
+   softpipe->dirty |= SP_NEW_STIPPLE;
+}
diff --git a/src/gallium/drivers/softpipe/sp_state_derived.c b/src/gallium/drivers/softpipe/sp_state_derived.c
new file mode 100644
index 0000000000..372597869f
--- /dev/null
+++ b/src/gallium/drivers/softpipe/sp_state_derived.c
@@ -0,0 +1,235 @@
+/**************************************************************************
+ * 
+ * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+#include "pipe/p_util.h"
+#include "pipe/p_shader_tokens.h"
+#include "pipe/draw/draw_context.h"
+#include "pipe/draw/draw_vertex.h"
+#include "pipe/draw/draw_private.h"
+#include "sp_context.h"
+#include "sp_state.h"
+
+
+/**
+ * Search vertex program's outputs to find a match for the given
+ * semantic name/index.  Return the index of the output slot.
+ *
+ * Return 0 if not found.  This will cause the fragment program to use
+ * vertex attrib 0 (position) in the cases where the fragment program
+ * attempts to use a missing vertex program output.  This is an undefined
+ * condition that users shouldn't hit anyway.
+ */
+static int
+find_vs_output(const struct pipe_shader_state *vs,
+               uint semantic_name,
+               uint semantic_index)
+{
+   uint i;
+   for (i = 0; i < vs->num_outputs; i++) {
+      if (vs->output_semantic_name[i] == semantic_name &&
+          vs->output_semantic_index[i] == semantic_index)
+         return i;
+   }
+   return 0;
+}
+
+
+/**
+ * Mark the current vertex layout as "invalid".
+ * We'll validate the vertex layout later, when we start to actually
+ * render a point or line or tri.
+ */
+static void
+invalidate_vertex_layout(struct softpipe_context *softpipe)
+{
+   softpipe->vertex_info.num_attribs =  0;
+}
+
+
+/**
+ * The vertex info describes how to convert the post-transformed vertices
+ * (simple float[][4]) used by the 'draw' module into vertices for
+ * rasterization.
+ *
+ * This function validates the vertex layout and returns a pointer to a
+ * vertex_info object.
+ */
+struct vertex_info *
+softpipe_get_vertex_info(struct softpipe_context *softpipe)
+{
+   struct vertex_info *vinfo = &softpipe->vertex_info;
+
+   if (vinfo->num_attribs == 0) {
+      /* compute vertex layout now */
+      const struct pipe_shader_state *vs = &softpipe->vs->shader;
+      const struct pipe_shader_state *fs = &softpipe->fs->shader;
+      const enum interp_mode colorInterp
+         = softpipe->rasterizer->flatshade ? INTERP_CONSTANT : INTERP_LINEAR;
+      uint i;
+
+      if (softpipe->vbuf) {
+         /* if using the post-transform vertex buffer, tell draw_vbuf to
+          * simply emit the whole post-xform vertex as-is:
+          */
+         struct vertex_info *vinfo_vbuf = &softpipe->vertex_info_vbuf;
+         vinfo_vbuf->num_attribs = 0;
+         draw_emit_vertex_attr(vinfo_vbuf, EMIT_ALL, INTERP_NONE, 0);
+         vinfo_vbuf->size = 4 * vs->num_outputs
+                          + sizeof(struct vertex_header) / 4;
+      }
+
+      /*
+       * Loop over fragment shader inputs, searching for the matching output
+       * from the vertex shader.
+       */
+      vinfo->num_attribs = 0;
+      for (i = 0; i < fs->num_inputs; i++) {
+         int src;
+         switch (fs->input_semantic_name[i]) {
+         case TGSI_SEMANTIC_POSITION:
+            src = find_vs_output(vs, TGSI_SEMANTIC_POSITION, 0);
+            draw_emit_vertex_attr(vinfo, EMIT_4F, INTERP_POS, src);
+            break;
+
+         case TGSI_SEMANTIC_COLOR:
+            src = find_vs_output(vs, TGSI_SEMANTIC_COLOR, 
+                                 fs->input_semantic_index[i]);
+            draw_emit_vertex_attr(vinfo, EMIT_4F, colorInterp, src);
+            break;
+
+         case TGSI_SEMANTIC_FOG:
+            src = find_vs_output(vs, TGSI_SEMANTIC_FOG, 0);
+            draw_emit_vertex_attr(vinfo, EMIT_4F, INTERP_PERSPECTIVE, src);
+            break;
+
+         case TGSI_SEMANTIC_GENERIC:
+            /* this includes texcoords and varying vars */
+            src = find_vs_output(vs, TGSI_SEMANTIC_GENERIC,
+                                 fs->input_semantic_index[i]);
+            draw_emit_vertex_attr(vinfo, EMIT_4F, INTERP_PERSPECTIVE, src);
+            break;
+
+         default:
+            assert(0);
+         }
+      }
+
+      softpipe->psize_slot = find_vs_output(vs, TGSI_SEMANTIC_PSIZE, 0);
+      if (softpipe->psize_slot > 0) {
+         draw_emit_vertex_attr(vinfo, EMIT_4F, INTERP_CONSTANT,
+                               softpipe->psize_slot);
+      }
+
+      draw_compute_vertex_size(vinfo);
+   }
+
+   return vinfo;
+}
+
+
+/**
+ * Called from vbuf module.
+ *
+ * Note that there's actually two different vertex layouts in softpipe.
+ *
+ * The normal one is computed in softpipe_get_vertex_info() above and is
+ * used by the point/line/tri "setup" code.
+ *
+ * The other one (this one) is only used by the vbuf module (which is
+ * not normally used by default but used in testing).  For the vbuf module,
+ * we basically want to pass-through the draw module's vertex layout as-is.
+ * When the softpipe vbuf code begins drawing, the normal vertex layout
+ * will come into play again.
+ */
+struct vertex_info *
+softpipe_get_vbuf_vertex_info(struct softpipe_context *softpipe)
+{
+   (void) softpipe_get_vertex_info(softpipe);
+   return &softpipe->vertex_info_vbuf;
+}
+
+
+/**
+ * Recompute cliprect from scissor bounds, scissor enable and surface size.
+ */
+static void
+compute_cliprect(struct softpipe_context *sp)
+{
+   unsigned surfWidth, surfHeight;
+
+   if (sp->framebuffer.num_cbufs > 0) {
+      surfWidth = sp->framebuffer.cbufs[0]->width;
+      surfHeight = sp->framebuffer.cbufs[0]->height;
+   }
+   else {
+      /* no surface? */
+      surfWidth = sp->scissor.maxx;
+      surfHeight = sp->scissor.maxy;
+   }
+
+   if (sp->rasterizer->scissor) {
+      /* clip to scissor rect */
+      sp->cliprect.minx = MAX2(sp->scissor.minx, 0);
+      sp->cliprect.miny = MAX2(sp->scissor.miny, 0);
+      sp->cliprect.maxx = MIN2(sp->scissor.maxx, surfWidth);
+      sp->cliprect.maxy = MIN2(sp->scissor.maxy, surfHeight);
+   }
+   else {
+      /* clip to surface bounds */
+      sp->cliprect.minx = 0;
+      sp->cliprect.miny = 0;
+      sp->cliprect.maxx = surfWidth;
+      sp->cliprect.maxy = surfHeight;
+   }
+}
+
+
+/* Hopefully this will remain quite simple, otherwise need to pull in
+ * something like the state tracker mechanism.
+ */
+void softpipe_update_derived( struct softpipe_context *softpipe )
+{
+   if (softpipe->dirty & (SP_NEW_RASTERIZER |
+                          SP_NEW_FS |
+                          SP_NEW_VS))
+      invalidate_vertex_layout( softpipe );
+
+   if (softpipe->dirty & (SP_NEW_SCISSOR |
+                          SP_NEW_DEPTH_STENCIL_ALPHA |
+                          SP_NEW_FRAMEBUFFER))
+      compute_cliprect(softpipe);
+
+   if (softpipe->dirty & (SP_NEW_BLEND |
+                          SP_NEW_DEPTH_STENCIL_ALPHA |
+                          SP_NEW_FRAMEBUFFER |
+                          SP_NEW_RASTERIZER |
+                          SP_NEW_FS | 
+			  SP_NEW_QUERY))
+      sp_build_quad_pipeline(softpipe);
+
+   softpipe->dirty = 0;
+}
diff --git a/src/gallium/drivers/softpipe/sp_state_fs.c b/src/gallium/drivers/softpipe/sp_state_fs.c
new file mode 100644
index 0000000000..0b814fc284
--- /dev/null
+++ b/src/gallium/drivers/softpipe/sp_state_fs.c
@@ -0,0 +1,179 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+#include "sp_context.h"
+#include "sp_state.h"
+
+#include "pipe/p_defines.h"
+#include "pipe/p_util.h"
+#include "pipe/p_inlines.h"
+#include "pipe/p_winsys.h"
+#include "pipe/draw/draw_context.h"
+#include "pipe/p_shader_tokens.h"
+#include "pipe/llvm/gallivm.h"
+#include "pipe/tgsi/util/tgsi_dump.h"
+#include "pipe/tgsi/exec/tgsi_sse2.h"
+
+
+void *
+softpipe_create_fs_state(struct pipe_context *pipe,
+                         const struct pipe_shader_state *templ)
+{
+   struct softpipe_context *softpipe = softpipe_context(pipe);
+   struct sp_fragment_shader_state *state;
+
+   /* Decide whether we'll be codegenerating this shader and if so do
+    * that now.
+    */
+
+   state = CALLOC_STRUCT(sp_fragment_shader_state);
+   if (!state)
+      return NULL;
+
+   state->shader = *templ;
+
+   if (softpipe->dump_fs) {
+      tgsi_dump(state->shader.tokens, 0);
+   }
+
+#ifdef MESA_LLVM
+   state->llvm_prog = 0;
+
+#if 0
+   if (!gallivm_global_cpu_engine()) {
+      gallivm_cpu_engine_create(state->llvm_prog);
+   }
+   else
+      gallivm_cpu_jit_compile(gallivm_global_cpu_engine(), state->llvm_prog);
+#endif
+
+#elif defined(__i386__) || defined(__386__)
+   if (softpipe->use_sse) {
+      x86_init_func( &state->sse2_program );
+      tgsi_emit_sse2_fs( state->shader.tokens, &state->sse2_program );
+   }
+#endif
+
+   return state;
+}
+
+
+void
+softpipe_bind_fs_state(struct pipe_context *pipe, void *fs)
+{
+   struct softpipe_context *softpipe = softpipe_context(pipe);
+
+   softpipe->fs = (struct sp_fragment_shader_state *) fs;
+
+   softpipe->dirty |= SP_NEW_FS;
+}
+
+
+void
+softpipe_delete_fs_state(struct pipe_context *pipe, void *fs)
+{
+   struct sp_fragment_shader_state *state = fs;
+
+#if defined(__i386__) || defined(__386__)
+   x86_release_func( &state->sse2_program );
+#endif
+
+   FREE( state );
+}
+
+
+void *
+softpipe_create_vs_state(struct pipe_context *pipe,
+                         const struct pipe_shader_state *templ)
+{
+   struct softpipe_context *softpipe = softpipe_context(pipe);
+   struct sp_vertex_shader_state *state;
+
+   state = CALLOC_STRUCT(sp_vertex_shader_state);
+   if (state == NULL ) {
+      return NULL;
+   }
+
+   state->shader = *templ;
+
+   state->draw_data = draw_create_vertex_shader(softpipe->draw,
+                                                &state->shader);
+   if (state->draw_data == NULL) {
+      FREE( state );
+      return NULL;
+   }
+
+   return state;
+}
+
+
+void
+softpipe_bind_vs_state(struct pipe_context *pipe, void *vs)
+{
+   struct softpipe_context *softpipe = softpipe_context(pipe);
+
+   softpipe->vs = (const struct sp_vertex_shader_state *)vs;
+
+   draw_bind_vertex_shader(softpipe->draw, softpipe->vs->draw_data);
+
+   softpipe->dirty |= SP_NEW_VS;
+}
+
+
+void
+softpipe_delete_vs_state(struct pipe_context *pipe, void *vs)
+{
+   struct softpipe_context *softpipe = softpipe_context(pipe);
+
+   struct sp_vertex_shader_state *state =
+      (struct sp_vertex_shader_state *)vs;
+
+   draw_delete_vertex_shader(softpipe->draw, state->draw_data);
+   FREE( state );
+}
+
+
+
+void
+softpipe_set_constant_buffer(struct pipe_context *pipe,
+                             uint shader, uint index,
+                             const struct pipe_constant_buffer *buf)
+{
+   struct softpipe_context *softpipe = softpipe_context(pipe);
+   struct pipe_winsys *ws = pipe->winsys;
+
+   assert(shader < PIPE_SHADER_TYPES);
+   assert(index == 0);
+
+   /* note: reference counting */
+   pipe_buffer_reference(ws,
+			 &softpipe->constants[shader].buffer,
+			 buf->buffer);
+   softpipe->constants[shader].size = buf->size;
+
+   softpipe->dirty |= SP_NEW_CONSTANTS;
+}
diff --git a/src/gallium/drivers/softpipe/sp_state_rasterizer.c b/src/gallium/drivers/softpipe/sp_state_rasterizer.c
new file mode 100644
index 0000000000..53755099dd
--- /dev/null
+++ b/src/gallium/drivers/softpipe/sp_state_rasterizer.c
@@ -0,0 +1,62 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+#include "pipe/p_defines.h"
+#include "pipe/p_util.h"
+#include "sp_context.h"
+#include "sp_state.h"
+#include "pipe/draw/draw_context.h"
+
+
+
+void *
+softpipe_create_rasterizer_state(struct pipe_context *pipe,
+                                 const struct pipe_rasterizer_state *rast)
+{
+   return mem_dup(rast, sizeof(*rast));
+}
+
+void softpipe_bind_rasterizer_state(struct pipe_context *pipe,
+                                    void *setup)
+{
+   struct softpipe_context *softpipe = softpipe_context(pipe);
+
+   /* pass-through to draw module */
+   draw_set_rasterizer_state(softpipe->draw, setup);
+
+   softpipe->rasterizer = (struct pipe_rasterizer_state *)setup;
+
+   softpipe->dirty |= SP_NEW_RASTERIZER;
+}
+
+void softpipe_delete_rasterizer_state(struct pipe_context *pipe,
+                                      void *rasterizer)
+{
+   FREE( rasterizer );
+}
+
+
diff --git a/src/gallium/drivers/softpipe/sp_state_sampler.c b/src/gallium/drivers/softpipe/sp_state_sampler.c
new file mode 100644
index 0000000000..ea348c7e95
--- /dev/null
+++ b/src/gallium/drivers/softpipe/sp_state_sampler.c
@@ -0,0 +1,93 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+/* Authors:
+ *  Brian Paul
+ */
+
+#include "pipe/p_util.h"
+
+#include "pipe/draw/draw_context.h"
+
+#include "sp_context.h"
+#include "sp_context.h"
+#include "sp_state.h"
+#include "sp_texture.h"
+#include "sp_tile_cache.h"
+#include "pipe/draw/draw_context.h"
+
+
+
+void *
+softpipe_create_sampler_state(struct pipe_context *pipe,
+                              const struct pipe_sampler_state *sampler)
+{
+   return mem_dup(sampler, sizeof(*sampler));
+}
+
+void
+softpipe_bind_sampler_state(struct pipe_context *pipe,
+                            unsigned unit, void *sampler)
+{
+   struct softpipe_context *softpipe = softpipe_context(pipe);
+
+   draw_flush(softpipe->draw);
+
+   assert(unit < PIPE_MAX_SAMPLERS);
+   softpipe->sampler[unit] = (struct pipe_sampler_state *)sampler;
+
+   softpipe->dirty |= SP_NEW_SAMPLER;
+}
+
+
+void
+softpipe_delete_sampler_state(struct pipe_context *pipe,
+                              void *sampler)
+{
+   FREE( sampler );
+}
+
+
+void
+softpipe_set_sampler_texture(struct pipe_context *pipe,
+			     unsigned unit,
+			     struct pipe_texture *texture)
+{
+   struct softpipe_context *softpipe = softpipe_context(pipe);
+
+   draw_flush(softpipe->draw);
+
+   assert(unit < PIPE_MAX_SAMPLERS);
+   softpipe->texture[unit] = softpipe_texture(texture);  /* ptr, not struct */
+
+   sp_tile_cache_set_texture(softpipe->tex_cache[unit], texture);
+
+   softpipe->dirty |= SP_NEW_TEXTURE;
+}
+
+
+
diff --git a/src/gallium/drivers/softpipe/sp_state_surface.c b/src/gallium/drivers/softpipe/sp_state_surface.c
new file mode 100644
index 0000000000..e2c6893e9f
--- /dev/null
+++ b/src/gallium/drivers/softpipe/sp_state_surface.c
@@ -0,0 +1,109 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+/* Authors:  Keith Whitwell <keith@tungstengraphics.com>
+ */
+#include "p_inlines.h"
+
+#include "sp_context.h"
+#include "sp_state.h"
+#include "sp_surface.h"
+#include "sp_tile_cache.h"
+
+
+/**
+ * XXX this might get moved someday
+ * Set the framebuffer surface info: color buffers, zbuffer, stencil buffer.
+ * Here, we flush the old surfaces and update the tile cache to point to the new
+ * surfaces.
+ */
+void
+softpipe_set_framebuffer_state(struct pipe_context *pipe,
+                               const struct pipe_framebuffer_state *fb)
+{
+   struct softpipe_context *sp = softpipe_context(pipe);
+   uint i;
+
+   for (i = 0; i < PIPE_MAX_COLOR_BUFS; i++) {
+      /* check if changing cbuf */
+      if (sp->framebuffer.cbufs[i] != fb->cbufs[i]) {
+         /* flush old */
+         sp_flush_tile_cache(sp, sp->cbuf_cache[i]);
+
+         /* assign new */
+         sp->framebuffer.cbufs[i] = fb->cbufs[i];
+
+         /* update cache */
+         sp_tile_cache_set_surface(sp->cbuf_cache[i], fb->cbufs[i]);
+      }
+   }
+
+   sp->framebuffer.num_cbufs = fb->num_cbufs;
+
+   /* zbuf changing? */
+   if (sp->framebuffer.zsbuf != fb->zsbuf) {
+      /* flush old */
+      sp_flush_tile_cache(sp, sp->zsbuf_cache);
+
+      /* assign new */
+      sp->framebuffer.zsbuf = fb->zsbuf;
+
+      /* update cache */
+      sp_tile_cache_set_surface(sp->zsbuf_cache, fb->zsbuf);
+   }
+
+#if 0
+   /* XXX combined depth/stencil here */
+
+   /* sbuf changing? */
+   if (sp->framebuffer.sbuf != fb->sbuf) {
+      /* flush old */
+      sp_flush_tile_cache(sp, sp->sbuf_cache_sep);
+
+      /* assign new */
+      sp->framebuffer.sbuf = fb->sbuf;
+
+      /* update cache */
+      if (fb->sbuf != fb->zbuf) {
+         /* separate stencil buf */
+         sp->sbuf_cache = sp->sbuf_cache_sep;
+         sp_tile_cache_set_surface(sp->sbuf_cache, fb->sbuf);
+      }
+      else {
+         /* combined depth/stencil */
+         sp->sbuf_cache = sp->zbuf_cache;
+         sp_tile_cache_set_surface(sp->sbuf_cache, fb->sbuf);
+      }
+   }
+#endif
+
+   sp->dirty |= SP_NEW_FRAMEBUFFER;
+}
+
+
+
+
diff --git a/src/gallium/drivers/softpipe/sp_state_vertex.c b/src/gallium/drivers/softpipe/sp_state_vertex.c
new file mode 100644
index 0000000000..09ff540ccf
--- /dev/null
+++ b/src/gallium/drivers/softpipe/sp_state_vertex.c
@@ -0,0 +1,64 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+/* Authors:  Keith Whitwell <keith@tungstengraphics.com>
+ */
+
+
+#include "sp_context.h"
+#include "sp_state.h"
+#include "sp_surface.h"
+
+#include "pipe/draw/draw_context.h"
+
+
+void
+softpipe_set_vertex_element(struct pipe_context *pipe,
+                            unsigned index,
+                            const struct pipe_vertex_element *attrib)
+{
+   struct softpipe_context *softpipe = softpipe_context(pipe);
+   assert(index < PIPE_ATTRIB_MAX);
+   softpipe->vertex_element[index] = *attrib; /* struct copy */
+   softpipe->dirty |= SP_NEW_VERTEX;
+
+   draw_set_vertex_element(softpipe->draw, index, attrib);
+}
+
+
+void
+softpipe_set_vertex_buffer(struct pipe_context *pipe,
+                           unsigned index,
+                           const struct pipe_vertex_buffer *buffer)
+{
+   struct softpipe_context *softpipe = softpipe_context(pipe);
+   assert(index < PIPE_ATTRIB_MAX);
+   softpipe->vertex_buffer[index] = *buffer; /* struct copy */
+   softpipe->dirty |= SP_NEW_VERTEX;
+
+   draw_set_vertex_buffer(softpipe->draw, index, buffer);
+}
diff --git a/src/gallium/drivers/softpipe/sp_surface.c b/src/gallium/drivers/softpipe/sp_surface.c
new file mode 100644
index 0000000000..8802ced187
--- /dev/null
+++ b/src/gallium/drivers/softpipe/sp_surface.c
@@ -0,0 +1,159 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+#include "pipe/p_defines.h"
+#include "pipe/p_util.h"
+#include "pipe/p_inlines.h"
+#include "pipe/p_winsys.h"
+#include "pipe/util/p_tile.h"
+#include "sp_context.h"
+#include "sp_surface.h"
+
+
+
+/* Assumes all values are within bounds -- no checking at this level -
+ * do it higher up if required.
+ */
+static void
+sp_surface_copy(struct pipe_context *pipe,
+                unsigned do_flip,
+		struct pipe_surface *dst,
+		unsigned dstx, unsigned dsty,
+		struct pipe_surface *src,
+		unsigned srcx, unsigned srcy, unsigned width, unsigned height)
+{
+   assert( dst->cpp == src->cpp );
+
+   pipe_copy_rect(pipe_surface_map(dst),
+                  dst->cpp,
+                  dst->pitch,
+                  dstx, dsty,
+                  width, height,
+                  pipe_surface_map(src),
+                  do_flip ? -(int) src->pitch : src->pitch,
+                  srcx, do_flip ? 1 - srcy - height : srcy);
+
+   pipe_surface_unmap(src);
+   pipe_surface_unmap(dst);
+}
+
+
+static void *
+get_pointer(struct pipe_surface *dst, void *dst_map, unsigned x, unsigned y)
+{
+   return (char *)dst_map + (y * dst->pitch + x) * dst->cpp;
+}
+
+
+#define UBYTE_TO_USHORT(B) ((B) | ((B) << 8))
+
+
+/**
+ * Fill a rectangular sub-region.  Need better logic about when to
+ * push buffers into AGP - will currently do so whenever possible.
+ */
+static void
+sp_surface_fill(struct pipe_context *pipe,
+		struct pipe_surface *dst,
+		unsigned dstx, unsigned dsty,
+		unsigned width, unsigned height, unsigned value)
+{
+   unsigned i, j;
+   void *dst_map = pipe_surface_map(dst);
+
+   assert(dst->pitch > 0);
+   assert(width <= dst->pitch);
+
+
+   switch (dst->cpp) {
+   case 1:
+      {
+	 ubyte *row = get_pointer(dst, dst_map, dstx, dsty);
+         for (i = 0; i < height; i++) {
+            memset(row, value, width);
+	 row += dst->pitch;
+         }
+      }
+      break;
+   case 2:
+      {
+         ushort *row = get_pointer(dst, dst_map, dstx, dsty);
+         for (i = 0; i < height; i++) {
+            for (j = 0; j < width; j++)
+               row[j] = (ushort) value;
+            row += dst->pitch;
+         }
+      }
+      break;
+   case 4:
+      {
+         unsigned *row = get_pointer(dst, dst_map, dstx, dsty);
+         for (i = 0; i < height; i++) {
+            for (j = 0; j < width; j++)
+               row[j] = value;
+            row += dst->pitch;
+         }
+      }
+      break;
+   case 8:
+      {
+         /* expand the 4-byte clear value to an 8-byte value */
+         ushort *row = (ushort *) get_pointer(dst, dst_map, dstx, dsty);
+         ushort val0 = UBYTE_TO_USHORT((value >>  0) & 0xff);
+         ushort val1 = UBYTE_TO_USHORT((value >>  8) & 0xff);
+         ushort val2 = UBYTE_TO_USHORT((value >> 16) & 0xff);
+         ushort val3 = UBYTE_TO_USHORT((value >> 24) & 0xff);
+         val0 = (val0 << 8) | val0;
+         val1 = (val1 << 8) | val1;
+         val2 = (val2 << 8) | val2;
+         val3 = (val3 << 8) | val3;
+         for (i = 0; i < height; i++) {
+            for (j = 0; j < width; j++) {
+               row[j*4+0] = val0;
+               row[j*4+1] = val1;
+               row[j*4+2] = val2;
+               row[j*4+3] = val3;
+            }
+            row += dst->pitch * 4;
+         }
+      }
+      break;
+   default:
+      assert(0);
+      break;
+   }
+
+   pipe_surface_unmap( dst );
+}
+
+
+void
+sp_init_surface_functions(struct softpipe_context *sp)
+{
+   sp->pipe.surface_copy = sp_surface_copy;
+   sp->pipe.surface_fill = sp_surface_fill;
+}
diff --git a/src/gallium/drivers/softpipe/sp_surface.h b/src/gallium/drivers/softpipe/sp_surface.h
new file mode 100644
index 0000000000..22de3ba43f
--- /dev/null
+++ b/src/gallium/drivers/softpipe/sp_surface.h
@@ -0,0 +1,42 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+/* Authors:  Keith Whitwell <keith@tungstengraphics.com>
+ */
+
+#ifndef SP_SURFACE_H
+#define SP_SURFACE_H
+
+
+struct softpipe_context;
+
+
+extern void
+sp_init_surface_functions(struct softpipe_context *sp);
+
+
+#endif /* SP_SURFACE_H */
diff --git a/src/gallium/drivers/softpipe/sp_tex_sample.c b/src/gallium/drivers/softpipe/sp_tex_sample.c
new file mode 100644
index 0000000000..325bdb86da
--- /dev/null
+++ b/src/gallium/drivers/softpipe/sp_tex_sample.c
@@ -0,0 +1,916 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+/**
+ * Texture sampling
+ *
+ * Authors:
+ *   Brian Paul
+ */
+
+#include "sp_context.h"
+#include "sp_headers.h"
+#include "sp_surface.h"
+#include "sp_tex_sample.h"
+#include "sp_tile_cache.h"
+#include "pipe/p_context.h"
+#include "pipe/p_defines.h"
+#include "pipe/p_util.h"
+#include "pipe/tgsi/exec/tgsi_exec.h"
+
+
+/*
+ * Note, the FRAC macro has to work perfectly.  Otherwise you'll sometimes
+ * see 1-pixel bands of improperly weighted linear-filtered textures.
+ * The tests/texwrap.c demo is a good test.
+ * Also note, FRAC(x) doesn't truly return the fractional part of x for x < 0.
+ * Instead, if x < 0 then FRAC(x) = 1 - true_frac(x).
+ */
+#define FRAC(f)  ((f) - ifloor(f))
+
+
+/**
+ * Linear interpolation macro
+ */
+#define LERP(T, A, B)  ( (A) + (T) * ((B) - (A)) )
+
+
+/**
+ * Do 2D/biliner interpolation of float values.
+ * v00, v10, v01 and v11 are typically four texture samples in a square/box.
+ * a and b are the horizontal and vertical interpolants.
+ * It's important that this function is inlined when compiled with
+ * optimization!  If we find that's not true on some systems, convert
+ * to a macro.
+ */
+static INLINE float
+lerp_2d(float a, float b,
+        float v00, float v10, float v01, float v11)
+{
+   const float temp0 = LERP(a, v00, v10);
+   const float temp1 = LERP(a, v01, v11);
+   return LERP(b, temp0, temp1);
+}
+
+
+/**
+ * If A is a signed integer, A % B doesn't give the right value for A < 0
+ * (in terms of texture repeat).  Just casting to unsigned fixes that.
+ */
+#define REMAINDER(A, B) ((unsigned) (A) % (unsigned) (B))
+
+
+/**
+ * Apply texture coord wrapping mode and return integer texture index.
+ * \param wrapMode  PIPE_TEX_WRAP_x
+ * \param s  the texcoord
+ * \param size  the texture image size
+ * \return  integer texture index
+ */
+static INLINE int
+nearest_texcoord(unsigned wrapMode, float s, unsigned size)
+{
+   int i;
+   switch (wrapMode) {
+   case PIPE_TEX_WRAP_REPEAT:
+      /* s limited to [0,1) */
+      /* i limited to [0,size-1] */
+      i = ifloor(s * size);
+      i = REMAINDER(i, size);
+      return i;
+   case PIPE_TEX_WRAP_CLAMP:
+      /* s limited to [0,1] */
+      /* i limited to [0,size-1] */
+      if (s <= 0.0F)
+         i = 0;
+      else if (s >= 1.0F)
+         i = size - 1;
+      else
+         i = ifloor(s * size);
+      return i;
+   case PIPE_TEX_WRAP_CLAMP_TO_EDGE:
+      {
+         /* s limited to [min,max] */
+         /* i limited to [0, size-1] */
+         const float min = 1.0F / (2.0F * size);
+         const float max = 1.0F - min;
+         if (s < min)
+            i = 0;
+         else if (s > max)
+            i = size - 1;
+         else
+            i = ifloor(s * size);
+      }
+      return i;
+   case PIPE_TEX_WRAP_CLAMP_TO_BORDER:
+      {
+         /* s limited to [min,max] */
+         /* i limited to [-1, size] */
+         const float min = -1.0F / (2.0F * size);
+         const float max = 1.0F - min;
+         if (s <= min)
+            i = -1;
+         else if (s >= max)
+            i = size;
+         else
+            i = ifloor(s * size);
+      }
+      return i;
+   case PIPE_TEX_WRAP_MIRROR_REPEAT:
+      {
+         const float min = 1.0F / (2.0F * size);
+         const float max = 1.0F - min;
+         const int flr = ifloor(s);
+         float u;
+         if (flr & 1)
+            u = 1.0F - (s - (float) flr);
+         else
+            u = s - (float) flr;
+         if (u < min)
+            i = 0;
+         else if (u > max)
+            i = size - 1;
+         else
+            i = ifloor(u * size);
+      }
+      return i;
+   case PIPE_TEX_WRAP_MIRROR_CLAMP:
+      {
+         /* s limited to [0,1] */
+         /* i limited to [0,size-1] */
+         const float u = FABSF(s);
+         if (u <= 0.0F)
+            i = 0;
+         else if (u >= 1.0F)
+            i = size - 1;
+         else
+            i = ifloor(u * size);
+      }
+      return i;
+   case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_EDGE:
+      {
+         /* s limited to [min,max] */
+         /* i limited to [0, size-1] */
+         const float min = 1.0F / (2.0F * size);
+         const float max = 1.0F - min;
+         const float u = FABSF(s);
+         if (u < min)
+            i = 0;
+         else if (u > max)
+            i = size - 1;
+         else
+            i = ifloor(u * size);
+      }
+      return i;
+   case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_BORDER:
+      {
+         /* s limited to [min,max] */
+         /* i limited to [0, size-1] */
+         const float min = -1.0F / (2.0F * size);
+         const float max = 1.0F - min;
+         const float u = FABSF(s);
+         if (u < min)
+            i = -1;
+         else if (u > max)
+            i = size;
+         else
+            i = ifloor(u * size);
+      }
+      return i;
+   default:
+      assert(0);
+      return 0;
+   }
+}
+
+
+/**
+ * Used to compute texel locations for linear sampling.
+ * \param wrapMode  PIPE_TEX_WRAP_x
+ * \param s  the texcoord
+ * \param size  the texture image size
+ * \param i0  returns first texture index
+ * \param i1  returns second texture index (usually *i0 + 1)
+ * \param a  returns blend factor/weight between texture indexes
+ */
+static INLINE void
+linear_texcoord(unsigned wrapMode, float s, unsigned size,
+                int *i0, int *i1, float *a)
+{
+   float u;
+   switch (wrapMode) {
+   case PIPE_TEX_WRAP_REPEAT:
+      u = s * size - 0.5F;
+      *i0 = REMAINDER(ifloor(u), size);
+      *i1 = REMAINDER(*i0 + 1, size);
+      break;
+   case PIPE_TEX_WRAP_CLAMP:
+      if (s <= 0.0F)
+         u = 0.0F;
+      else if (s >= 1.0F)
+         u = (float) size;
+      else
+         u = s * size;
+      u -= 0.5F;
+      *i0 = ifloor(u);
+      *i1 = *i0 + 1;
+      break;
+   case PIPE_TEX_WRAP_CLAMP_TO_EDGE:
+      if (s <= 0.0F)
+         u = 0.0F;
+      else if (s >= 1.0F)
+         u = (float) size;
+      else
+         u = s * size;
+      u -= 0.5F;
+      *i0 = ifloor(u);
+      *i1 = *i0 + 1;
+      if (*i0 < 0)
+         *i0 = 0;
+      if (*i1 >= (int) size)
+         *i1 = size - 1;
+      break;
+   case PIPE_TEX_WRAP_CLAMP_TO_BORDER:
+      {
+         const float min = -1.0F / (2.0F * size);
+         const float max = 1.0F - min;
+         if (s <= min)
+            u = min * size;
+         else if (s >= max)
+            u = max * size;
+         else
+            u = s * size;
+         u -= 0.5F;
+         *i0 = ifloor(u);
+         *i1 = *i0 + 1;
+      }
+      break;
+   case PIPE_TEX_WRAP_MIRROR_REPEAT:
+      {
+         const int flr = ifloor(s);
+         if (flr & 1)
+            u = 1.0F - (s - (float) flr);
+         else
+            u = s - (float) flr;
+         u = (u * size) - 0.5F;
+         *i0 = ifloor(u);
+         *i1 = *i0 + 1;
+         if (*i0 < 0)
+            *i0 = 0;
+         if (*i1 >= (int) size)
+            *i1 = size - 1;
+      }
+      break;
+   case PIPE_TEX_WRAP_MIRROR_CLAMP:
+      u = FABSF(s);
+      if (u >= 1.0F)
+         u = (float) size;
+      else
+         u *= size;
+      u -= 0.5F;
+      *i0 = ifloor(u);
+      *i1 = *i0 + 1;
+      break;
+   case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_EDGE:
+      u = FABSF(s);
+      if (u >= 1.0F)
+         u = (float) size;
+      else
+         u *= size;
+      u -= 0.5F;
+      *i0 = ifloor(u);
+      *i1 = *i0 + 1;
+      if (*i0 < 0)
+         *i0 = 0;
+      if (*i1 >= (int) size)
+         *i1 = size - 1;
+      break;
+   case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_BORDER:
+      {
+         const float min = -1.0F / (2.0F * size);
+         const float max = 1.0F - min;
+         u = FABSF(s);
+         if (u <= min)
+            u = min * size;
+         else if (u >= max)
+            u = max * size;
+         else
+            u *= size;
+         u -= 0.5F;
+         *i0 = ifloor(u);
+         *i1 = *i0 + 1;
+      }
+      break;
+   default:
+      assert(0);
+   }
+   *a = FRAC(u);
+}
+
+
+static unsigned
+choose_cube_face(float rx, float ry, float rz, float *newS, float *newT)
+{
+   /*
+      major axis
+      direction     target                             sc     tc    ma
+      ----------    -------------------------------    ---    ---   ---
+       +rx          TEXTURE_CUBE_MAP_POSITIVE_X_EXT    -rz    -ry   rx
+       -rx          TEXTURE_CUBE_MAP_NEGATIVE_X_EXT    +rz    -ry   rx
+       +ry          TEXTURE_CUBE_MAP_POSITIVE_Y_EXT    +rx    +rz   ry
+       -ry          TEXTURE_CUBE_MAP_NEGATIVE_Y_EXT    +rx    -rz   ry
+       +rz          TEXTURE_CUBE_MAP_POSITIVE_Z_EXT    +rx    -ry   rz
+       -rz          TEXTURE_CUBE_MAP_NEGATIVE_Z_EXT    -rx    -ry   rz
+   */
+   const float arx = FABSF(rx), ary = FABSF(ry), arz = FABSF(rz);
+   unsigned face;
+   float sc, tc, ma;
+
+   if (arx > ary && arx > arz) {
+      if (rx >= 0.0F) {
+         face = PIPE_TEX_FACE_POS_X;
+         sc = -rz;
+         tc = -ry;
+         ma = arx;
+      }
+      else {
+         face = PIPE_TEX_FACE_NEG_X;
+         sc = rz;
+         tc = -ry;
+         ma = arx;
+      }
+   }
+   else if (ary > arx && ary > arz) {
+      if (ry >= 0.0F) {
+         face = PIPE_TEX_FACE_POS_Y;
+         sc = rx;
+         tc = rz;
+         ma = ary;
+      }
+      else {
+         face = PIPE_TEX_FACE_NEG_Y;
+         sc = rx;
+         tc = -rz;
+         ma = ary;
+      }
+   }
+   else {
+      if (rz > 0.0F) {
+         face = PIPE_TEX_FACE_POS_Z;
+         sc = rx;
+         tc = -ry;
+         ma = arz;
+      }
+      else {
+         face = PIPE_TEX_FACE_NEG_Z;
+         sc = -rx;
+         tc = -ry;
+         ma = arz;
+      }
+   }
+
+   *newS = ( sc / ma + 1.0F ) * 0.5F;
+   *newT = ( tc / ma + 1.0F ) * 0.5F;
+
+   return face;
+}
+
+
+/**
+ * Examine the quad's texture coordinates to compute the partial
+ * derivatives w.r.t X and Y, then compute lambda (level of detail).
+ *
+ * This is only done for fragment shaders, not vertex shaders.
+ */
+static float
+compute_lambda(struct tgsi_sampler *sampler,
+               const float s[QUAD_SIZE],
+               const float t[QUAD_SIZE],
+               const float p[QUAD_SIZE],
+               float lodbias)
+{
+   float rho, lambda;
+
+   assert(s);
+   {
+      float dsdx = s[QUAD_BOTTOM_RIGHT] - s[QUAD_BOTTOM_LEFT];
+      float dsdy = s[QUAD_TOP_LEFT]     - s[QUAD_BOTTOM_LEFT];
+      dsdx = FABSF(dsdx);
+      dsdy = FABSF(dsdy);
+      rho = MAX2(dsdx, dsdy);
+      if (sampler->state->normalized_coords)
+         rho *= sampler->texture->width[0];
+   }
+   if (t) {
+      float dtdx = t[QUAD_BOTTOM_RIGHT] - t[QUAD_BOTTOM_LEFT];
+      float dtdy = t[QUAD_TOP_LEFT]     - t[QUAD_BOTTOM_LEFT];
+      float max;
+      dtdx = FABSF(dtdx);
+      dtdy = FABSF(dtdy);
+      max = MAX2(dtdx, dtdy);
+      if (sampler->state->normalized_coords)
+         max *= sampler->texture->height[0];
+      rho = MAX2(rho, max);
+   }
+   if (p) {
+      float dpdx = p[QUAD_BOTTOM_RIGHT] - p[QUAD_BOTTOM_LEFT];
+      float dpdy = p[QUAD_TOP_LEFT]     - p[QUAD_BOTTOM_LEFT];
+      float max;
+      dpdx = FABSF(dpdx);
+      dpdy = FABSF(dpdy);
+      max = MAX2(dpdx, dpdy);
+      if (sampler->state->normalized_coords)
+         max *= sampler->texture->depth[0];
+      rho = MAX2(rho, max);
+   }
+
+   lambda = LOG2(rho);
+   lambda += lodbias + sampler->state->lod_bias;
+   lambda = CLAMP(lambda, sampler->state->min_lod, sampler->state->max_lod);
+
+   return lambda;
+}
+
+
+/**
+ * Do several things here:
+ * 1. Compute lambda from the texcoords, if needed
+ * 2. Determine if we're minifying or magnifying
+ * 3. If minifying, choose mipmap levels
+ * 4. Return image filter to use within mipmap images
+ */
+static void
+choose_mipmap_levels(struct tgsi_sampler *sampler,
+                     const float s[QUAD_SIZE],
+                     const float t[QUAD_SIZE],
+                     const float p[QUAD_SIZE],
+                     float lodbias,
+                     unsigned *level0, unsigned *level1, float *levelBlend,
+                     unsigned *imgFilter)
+{
+   if (sampler->state->min_mip_filter == PIPE_TEX_MIPFILTER_NONE) {
+      /* no mipmap selection needed */
+      *imgFilter = sampler->state->mag_img_filter;
+      *level0 = *level1 = (int) sampler->state->min_lod;
+   }
+   else {
+      float lambda;
+
+      if (1)
+         /* fragment shader */
+         lambda = compute_lambda(sampler, s, t, p, lodbias);
+      else
+         /* vertex shader */
+         lambda = lodbias; /* not really a bias, but absolute LOD */
+
+      if (lambda < 0.0) { /* XXX threshold depends on the filter */
+         /* magnifying */
+         *imgFilter = sampler->state->mag_img_filter;
+         *level0 = *level1 = 0;
+      }
+      else {
+         /* minifying */
+         *imgFilter = sampler->state->min_img_filter;
+
+         /* choose mipmap level(s) and compute the blend factor between them */
+         if (sampler->state->min_mip_filter == PIPE_TEX_MIPFILTER_NEAREST) {
+            /* Nearest mipmap level */
+            const int lvl = (int) (lambda + 0.5);
+            *level0 =
+            *level1 = CLAMP(lvl, 0, (int) sampler->texture->last_level);
+         }
+         else {
+            /* Linear interpolation between mipmap levels */
+            const int lvl = (int) lambda;
+            *level0 = CLAMP(lvl,     0, (int) sampler->texture->last_level);
+            *level1 = CLAMP(lvl + 1, 0, (int) sampler->texture->last_level);
+            *levelBlend = FRAC(lambda);  /* blending weight between levels */
+         }
+      }
+   }
+}
+
+
+/**
+ * Get a texel from a texture, using the texture tile cache.
+ *
+ * \param face  the cube face in 0..5
+ * \param level  the mipmap level
+ * \param x  the x coord of texel within 2D image
+ * \param y  the y coord of texel within 2D image
+ * \param z  which slice of a 3D texture
+ * \param rgba  the quad to put the texel/color into
+ * \param j  which element of the rgba quad to write to
+ *
+ * XXX maybe move this into sp_tile_cache.c and merge with the
+ * sp_get_cached_tile_tex() function.  Also, get 4 texels instead of 1...
+ */
+static void
+get_texel(struct tgsi_sampler *sampler,
+          unsigned face, unsigned level, int x, int y, int z,
+          float rgba[NUM_CHANNELS][QUAD_SIZE], unsigned j)
+{
+   const int tx = x % TILE_SIZE;
+   const int ty = y % TILE_SIZE;
+   const struct softpipe_cached_tile *tile
+      = sp_get_cached_tile_tex(sampler->pipe, sampler->cache,
+                               x, y, z, face, level);
+   rgba[0][j] = tile->data.color[ty][tx][0];
+   rgba[1][j] = tile->data.color[ty][tx][1];
+   rgba[2][j] = tile->data.color[ty][tx][2];
+   rgba[3][j] = tile->data.color[ty][tx][3];
+}
+
+
+/**
+ * Compare texcoord 'p' (aka R) against texture value 'rgba[0]'
+ * When we sampled the depth texture, the depth value was put into all
+ * RGBA channels.  We look at the red channel here.
+ */
+static INLINE void
+shadow_compare(uint compare_func,
+               float rgba[NUM_CHANNELS][QUAD_SIZE],
+               const float p[QUAD_SIZE],
+               uint j)
+{
+   int k;
+   switch (compare_func) {
+   case PIPE_FUNC_LESS:
+      k = p[j] < rgba[0][j];
+      break;
+   case PIPE_FUNC_LEQUAL:
+      k = p[j] <= rgba[0][j];
+      break;
+   case PIPE_FUNC_GREATER:
+      k = p[j] > rgba[0][j];
+      break;
+   case PIPE_FUNC_GEQUAL:
+      k = p[j] >= rgba[0][j];
+      break;
+   case PIPE_FUNC_EQUAL:
+      k = p[j] == rgba[0][j];
+      break;
+   case PIPE_FUNC_NOTEQUAL:
+      k = p[j] != rgba[0][j];
+      break;
+   case PIPE_FUNC_ALWAYS:
+      k = 1;
+      break;
+   case PIPE_FUNC_NEVER:
+      k = 0;
+      break;
+   default:
+      assert(0);
+   }
+
+   rgba[0][j] = rgba[1][j] = rgba[2][j] = (float) k;
+}
+
+
+/**
+ * Common code for sampling 1D/2D/cube textures.
+ * Could probably extend for 3D...
+ */
+static void
+sp_get_samples_2d_common(struct tgsi_sampler *sampler,
+                         const float s[QUAD_SIZE],
+                         const float t[QUAD_SIZE],
+                         const float p[QUAD_SIZE],
+                         float lodbias,
+                         float rgba[NUM_CHANNELS][QUAD_SIZE],
+                         const unsigned faces[4])
+{
+   const uint compare_func = sampler->state->compare_func;
+   unsigned level0, level1, j, imgFilter;
+   int width, height;
+   float levelBlend;
+
+   choose_mipmap_levels(sampler, s, t, p, lodbias,
+                        &level0, &level1, &levelBlend, &imgFilter);
+
+   if (sampler->state->normalized_coords) {
+      width = sampler->texture->width[level0];
+      height = sampler->texture->height[level0];
+   }
+   else {
+      width = height = 1;
+   }
+
+   assert(width > 0);
+
+   switch (imgFilter) {
+   case PIPE_TEX_FILTER_NEAREST:
+      for (j = 0; j < QUAD_SIZE; j++) {
+         int x = nearest_texcoord(sampler->state->wrap_s, s[j], width);
+         int y = nearest_texcoord(sampler->state->wrap_t, t[j], height);
+         get_texel(sampler, faces[j], level0, x, y, 0, rgba, j);
+         if (sampler->state->compare_mode == PIPE_TEX_COMPARE_R_TO_TEXTURE) {
+            shadow_compare(compare_func, rgba, p, j);
+         }
+
+         if (level0 != level1) {
+            /* get texels from second mipmap level and blend */
+            float rgba2[4][4];
+            unsigned c;
+            x = x / 2;
+            y = y / 2;
+            get_texel(sampler, faces[j], level1, x, y, 0, rgba2, j);
+            if (sampler->state->compare_mode == PIPE_TEX_COMPARE_R_TO_TEXTURE){
+               shadow_compare(compare_func, rgba2, p, j);
+            }
+
+            for (c = 0; c < NUM_CHANNELS; c++) {
+               rgba[c][j] = LERP(levelBlend, rgba[c][j], rgba2[c][j]);
+            }
+         }
+      }
+      break;
+   case PIPE_TEX_FILTER_LINEAR:
+      for (j = 0; j < QUAD_SIZE; j++) {
+         float tx[4][4], a, b;
+         int x0, y0, x1, y1, c;
+         linear_texcoord(sampler->state->wrap_s, s[j], width,  &x0, &x1, &a);
+         linear_texcoord(sampler->state->wrap_t, t[j], height, &y0, &y1, &b);
+         get_texel(sampler, faces[j], level0, x0, y0, 0, tx, 0);
+         get_texel(sampler, faces[j], level0, x1, y0, 0, tx, 1);
+         get_texel(sampler, faces[j], level0, x0, y1, 0, tx, 2);
+         get_texel(sampler, faces[j], level0, x1, y1, 0, tx, 3);
+         if (sampler->state->compare_mode == PIPE_TEX_COMPARE_R_TO_TEXTURE) {
+            shadow_compare(compare_func, tx, p, 0);
+            shadow_compare(compare_func, tx, p, 1);
+            shadow_compare(compare_func, tx, p, 2);
+            shadow_compare(compare_func, tx, p, 3);
+         }
+
+         for (c = 0; c < 4; c++) {
+            rgba[c][j] = lerp_2d(a, b, tx[c][0], tx[c][1], tx[c][2], tx[c][3]);
+         }
+
+         if (level0 != level1) {
+            /* get texels from second mipmap level and blend */
+            float rgba2[4][4];
+            x0 = x0 / 2;
+            y0 = y0 / 2;
+            x1 = x1 / 2;
+            y1 = y1 / 2;
+            get_texel(sampler, faces[j], level1, x0, y0, 0, tx, 0);
+            get_texel(sampler, faces[j], level1, x1, y0, 0, tx, 1);
+            get_texel(sampler, faces[j], level1, x0, y1, 0, tx, 2);
+            get_texel(sampler, faces[j], level1, x1, y1, 0, tx, 3);
+            if (sampler->state->compare_mode == PIPE_TEX_COMPARE_R_TO_TEXTURE){
+               shadow_compare(compare_func, tx, p, 0);
+               shadow_compare(compare_func, tx, p, 1);
+               shadow_compare(compare_func, tx, p, 2);
+               shadow_compare(compare_func, tx, p, 3);
+            }
+
+            for (c = 0; c < 4; c++) {
+               rgba2[c][j] = lerp_2d(a, b,
+                                     tx[c][0], tx[c][1], tx[c][2], tx[c][3]);
+            }
+
+            for (c = 0; c < NUM_CHANNELS; c++) {
+               rgba[c][j] = LERP(levelBlend, rgba[c][j], rgba2[c][j]);
+            }
+         }
+      }
+      break;
+   default:
+      assert(0);
+   }
+}
+
+
+static void
+sp_get_samples_1d(struct tgsi_sampler *sampler,
+                  const float s[QUAD_SIZE],
+                  const float t[QUAD_SIZE],
+                  const float p[QUAD_SIZE],
+                  float lodbias,
+                  float rgba[NUM_CHANNELS][QUAD_SIZE])
+{
+   static const unsigned faces[4] = {0, 0, 0, 0};
+   static const float tzero[4] = {0, 0, 0, 0};
+   sp_get_samples_2d_common(sampler, s, tzero, NULL, lodbias, rgba, faces);
+}
+
+
+static void
+sp_get_samples_2d(struct tgsi_sampler *sampler,
+                  const float s[QUAD_SIZE],
+                  const float t[QUAD_SIZE],
+                  const float p[QUAD_SIZE],
+                  float lodbias,
+                  float rgba[NUM_CHANNELS][QUAD_SIZE])
+{
+   static const unsigned faces[4] = {0, 0, 0, 0};
+   sp_get_samples_2d_common(sampler, s, t, p, lodbias, rgba, faces);
+}
+
+
+static void
+sp_get_samples_3d(struct tgsi_sampler *sampler,
+                  const float s[QUAD_SIZE],
+                  const float t[QUAD_SIZE],
+                  const float p[QUAD_SIZE],
+                  float lodbias,
+                  float rgba[NUM_CHANNELS][QUAD_SIZE])
+{
+   /* get/map pipe_surfaces corresponding to 3D tex slices */
+   unsigned level0, level1, j, imgFilter;
+   int width, height, depth;
+   float levelBlend;
+   const uint face = 0;
+
+   choose_mipmap_levels(sampler, s, t, p, lodbias,
+                        &level0, &level1, &levelBlend, &imgFilter);
+
+   if (sampler->state->normalized_coords) {
+      width = sampler->texture->width[level0];
+      height = sampler->texture->height[level0];
+      depth = sampler->texture->depth[level0];
+   }
+   else {
+      width = height = depth = 1;
+   }
+
+   assert(width > 0);
+   assert(height > 0);
+   assert(depth > 0);
+
+   switch (imgFilter) {
+   case PIPE_TEX_FILTER_NEAREST:
+      for (j = 0; j < QUAD_SIZE; j++) {
+         int x = nearest_texcoord(sampler->state->wrap_s, s[j], width);
+         int y = nearest_texcoord(sampler->state->wrap_t, t[j], height);
+         int z = nearest_texcoord(sampler->state->wrap_r, p[j], depth);
+         get_texel(sampler, face, level0, x, y, z, rgba, j);
+
+         if (level0 != level1) {
+            /* get texels from second mipmap level and blend */
+            float rgba2[4][4];
+            unsigned c;
+            x /= 2;
+            y /= 2;
+            z /= 2;
+            get_texel(sampler, face, level1, x, y, z, rgba2, j);
+            for (c = 0; c < NUM_CHANNELS; c++) {
+               rgba[c][j] = LERP(levelBlend, rgba2[c][j], rgba[c][j]);
+            }
+         }
+      }
+      break;
+   case PIPE_TEX_FILTER_LINEAR:
+      for (j = 0; j < QUAD_SIZE; j++) {
+         float texel0[4][4], texel1[4][4];
+         float xw, yw, zw; /* interpolation weights */
+         int x0, x1, y0, y1, z0, z1, c;
+         linear_texcoord(sampler->state->wrap_s, s[j], width,  &x0, &x1, &xw);
+         linear_texcoord(sampler->state->wrap_t, t[j], height, &y0, &y1, &yw);
+         linear_texcoord(sampler->state->wrap_r, p[j], depth,  &z0, &z1, &zw);
+         get_texel(sampler, face, level0, x0, y0, z0, texel0, 0);
+         get_texel(sampler, face, level0, x1, y0, z0, texel0, 1);
+         get_texel(sampler, face, level0, x0, y1, z0, texel0, 2);
+         get_texel(sampler, face, level0, x1, y1, z0, texel0, 3);
+         get_texel(sampler, face, level0, x0, y0, z1, texel1, 0);
+         get_texel(sampler, face, level0, x1, y0, z1, texel1, 1);
+         get_texel(sampler, face, level0, x0, y1, z1, texel1, 2);
+         get_texel(sampler, face, level0, x1, y1, z1, texel1, 3);
+
+         /* 3D lerp */
+         for (c = 0; c < 4; c++) {
+            float ctemp0[4][4], ctemp1[4][4];
+            ctemp0[c][j] = lerp_2d(xw, yw,
+                                   texel0[c][0], texel0[c][1],
+                                   texel0[c][2], texel0[c][3]);
+            ctemp1[c][j] = lerp_2d(xw, yw,
+                                   texel1[c][0], texel1[c][1],
+                                   texel1[c][2], texel1[c][3]);
+            rgba[c][j] = LERP(zw, ctemp0[c][j], ctemp1[c][j]);
+         }
+
+         if (level0 != level1) {
+            /* get texels from second mipmap level and blend */
+            float rgba2[4][4];
+            x0 /= 2;
+            y0 /= 2;
+            z0 /= 2;
+            x1 /= 2;
+            y1 /= 2;
+            z1 /= 2;
+            get_texel(sampler, face, level1, x0, y0, z0, texel0, 0);
+            get_texel(sampler, face, level1, x1, y0, z0, texel0, 1);
+            get_texel(sampler, face, level1, x0, y1, z0, texel0, 2);
+            get_texel(sampler, face, level1, x1, y1, z0, texel0, 3);
+            get_texel(sampler, face, level1, x0, y0, z1, texel1, 0);
+            get_texel(sampler, face, level1, x1, y0, z1, texel1, 1);
+            get_texel(sampler, face, level1, x0, y1, z1, texel1, 2);
+            get_texel(sampler, face, level1, x1, y1, z1, texel1, 3);
+
+            /* 3D lerp */
+            for (c = 0; c < 4; c++) {
+               float ctemp0[4][4], ctemp1[4][4];
+               ctemp0[c][j] = lerp_2d(xw, yw,
+                                      texel0[c][0], texel0[c][1],
+                                      texel0[c][2], texel0[c][3]);
+               ctemp1[c][j] = lerp_2d(xw, yw,
+                                      texel1[c][0], texel1[c][1],
+                                      texel1[c][2], texel1[c][3]);
+               rgba2[c][j] = LERP(zw, ctemp0[c][j], ctemp1[c][j]);
+            }
+
+            /* blend mipmap levels */
+            for (c = 0; c < NUM_CHANNELS; c++) {
+               rgba[c][j] = LERP(levelBlend, rgba[c][j], rgba2[c][j]);
+            }
+         }
+      }
+      break;
+   default:
+      assert(0);
+   }
+}
+
+
+static void
+sp_get_samples_cube(struct tgsi_sampler *sampler,
+                    const float s[QUAD_SIZE],
+                    const float t[QUAD_SIZE],
+                    const float p[QUAD_SIZE],
+                    float lodbias,
+                    float rgba[NUM_CHANNELS][QUAD_SIZE])
+{
+   unsigned faces[QUAD_SIZE], j;
+   float ssss[4], tttt[4];
+   for (j = 0; j < QUAD_SIZE; j++) {
+      faces[j] = choose_cube_face(s[j], t[j], p[j], ssss + j, tttt + j);
+   }
+   sp_get_samples_2d_common(sampler, ssss, tttt, NULL, lodbias, rgba, faces);
+}
+
+
+/**
+ * Called via tgsi_sampler::get_samples()
+ * Use the sampler's state setting to get a filtered RGBA value
+ * from the sampler's texture.
+ *
+ * XXX we can implement many versions of this function, each
+ * tightly coded for a specific combination of sampler state
+ * (nearest + repeat), (bilinear mipmap + clamp), etc.
+ *
+ * The update_samplers() function in st_atom_sampler.c could create
+ * a new tgsi_sampler object for each state combo it finds....
+ */
+void
+sp_get_samples(struct tgsi_sampler *sampler,
+               const float s[QUAD_SIZE],
+               const float t[QUAD_SIZE],
+               const float p[QUAD_SIZE],
+               float lodbias,
+               float rgba[NUM_CHANNELS][QUAD_SIZE])
+{
+   if (!sampler->texture)
+      return;
+
+   switch (sampler->texture->target) {
+   case PIPE_TEXTURE_1D:
+      sp_get_samples_1d(sampler, s, t, p, lodbias, rgba);
+      break;
+   case PIPE_TEXTURE_2D:
+      sp_get_samples_2d(sampler, s, t, p, lodbias, rgba);
+      break;
+   case PIPE_TEXTURE_3D:
+      sp_get_samples_3d(sampler, s, t, p, lodbias, rgba);
+      break;
+   case PIPE_TEXTURE_CUBE:
+      sp_get_samples_cube(sampler, s, t, p, lodbias, rgba);
+      break;
+   default:
+      assert(0);
+   }
+}
+
diff --git a/src/gallium/drivers/softpipe/sp_tex_sample.h b/src/gallium/drivers/softpipe/sp_tex_sample.h
new file mode 100644
index 0000000000..404bfd0c36
--- /dev/null
+++ b/src/gallium/drivers/softpipe/sp_tex_sample.h
@@ -0,0 +1,17 @@
+#ifndef SP_TEX_SAMPLE_H
+#define SP_TEX_SAMPLE_H
+
+
+struct tgsi_sampler;
+
+
+extern void
+sp_get_samples(struct tgsi_sampler *sampler,
+               const float s[QUAD_SIZE],
+               const float t[QUAD_SIZE],
+               const float p[QUAD_SIZE],
+               float lodbias,
+               float rgba[NUM_CHANNELS][QUAD_SIZE]);
+
+
+#endif /* SP_TEX_SAMPLE_H */
diff --git a/src/gallium/drivers/softpipe/sp_texture.c b/src/gallium/drivers/softpipe/sp_texture.c
new file mode 100644
index 0000000000..6de7a9b543
--- /dev/null
+++ b/src/gallium/drivers/softpipe/sp_texture.c
@@ -0,0 +1,166 @@
+/**************************************************************************
+ * 
+ * Copyright 2006 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+ /*
+  * Authors:
+  *   Keith Whitwell <keith@tungstengraphics.com>
+  *   Michel Dänzer <michel@tungstengraphics.com>
+  */
+
+#include "pipe/p_context.h"
+#include "pipe/p_defines.h"
+#include "pipe/p_inlines.h"
+#include "pipe/p_util.h"
+#include "pipe/p_winsys.h"
+
+#include "sp_context.h"
+#include "sp_state.h"
+#include "sp_texture.h"
+
+
+/* Simple, maximally packed layout.
+ */
+
+static unsigned minify( unsigned d )
+{
+   return MAX2(1, d>>1);
+}
+
+
+static void
+softpipe_texture_layout(struct softpipe_texture * spt)
+{
+   struct pipe_texture *pt = &spt->base;
+   unsigned level;
+   unsigned width = pt->width[0];
+   unsigned height = pt->height[0];
+   unsigned depth = pt->depth[0];
+
+   spt->buffer_size = 0;
+
+   for (level = 0; level <= pt->last_level; level++) {
+      pt->width[level] = width;
+      pt->height[level] = height;
+      pt->depth[level] = depth;
+
+      spt->level_offset[level] = spt->buffer_size;
+
+      spt->buffer_size += ((pt->compressed) ? MAX2(1, height/4) : height) *
+			  ((pt->target == PIPE_TEXTURE_CUBE) ? 6 : depth) *
+			  width * pt->cpp;
+
+      width  = minify(width);
+      height = minify(height);
+      depth = minify(depth);
+   }
+}
+
+
+struct pipe_texture *
+softpipe_texture_create(struct pipe_context *pipe,
+                        const struct pipe_texture *templat)
+{
+   struct softpipe_texture *spt = CALLOC_STRUCT(softpipe_texture);
+   if (!spt)
+      return NULL;
+
+   spt->base = *templat;
+
+   softpipe_texture_layout(spt);
+
+   spt->buffer = pipe->winsys->buffer_create(pipe->winsys, 32,
+                                             PIPE_BUFFER_USAGE_PIXEL,
+                                             spt->buffer_size);
+   if (!spt->buffer) {
+      FREE(spt);
+      return NULL;
+   }
+
+   return &spt->base;
+}
+
+
+void
+softpipe_texture_release(struct pipe_context *pipe, struct pipe_texture **pt)
+{
+   if (!*pt)
+      return;
+
+   /*
+   DBG("%s %p refcount will be %d\n",
+       __FUNCTION__, (void *) *pt, (*pt)->refcount - 1);
+   */
+   if (--(*pt)->refcount <= 0) {
+      struct softpipe_texture *spt = softpipe_texture(*pt);
+
+      /*
+      DBG("%s deleting %p\n", __FUNCTION__, (void *) spt);
+      */
+
+      pipe_buffer_reference(pipe->winsys, &spt->buffer, NULL);
+
+      FREE(spt);
+   }
+   *pt = NULL;
+}
+
+
+/**
+ * Called via pipe->get_tex_surface()
+ */
+struct pipe_surface *
+softpipe_get_tex_surface(struct pipe_context *pipe,
+                         struct pipe_texture *pt,
+                         unsigned face, unsigned level, unsigned zslice)
+{
+   struct softpipe_texture *spt = softpipe_texture(pt);
+   struct pipe_surface *ps;
+
+   assert(level <= pt->last_level);
+
+   ps = pipe->winsys->surface_alloc(pipe->winsys);
+   if (ps) {
+      assert(ps->refcount);
+      assert(ps->winsys);
+      pipe_buffer_reference(pipe->winsys, &ps->buffer, spt->buffer);
+      ps->format = pt->format;
+      ps->cpp = pt->cpp;
+      ps->width = pt->width[level];
+      ps->height = pt->height[level];
+      ps->pitch = ps->width;
+      ps->offset = spt->level_offset[level];
+
+      if (pt->target == PIPE_TEXTURE_CUBE || pt->target == PIPE_TEXTURE_3D) {
+	 ps->offset += ((pt->target == PIPE_TEXTURE_CUBE) ? face : zslice) *
+		       (pt->compressed ? ps->height/4 : ps->height) *
+		       ps->width * ps->cpp;
+      } else {
+	 assert(face == 0);
+	 assert(zslice == 0);
+      }
+   }
+   return ps;
+}
diff --git a/src/gallium/drivers/softpipe/sp_texture.h b/src/gallium/drivers/softpipe/sp_texture.h
new file mode 100644
index 0000000000..fa646c0de9
--- /dev/null
+++ b/src/gallium/drivers/softpipe/sp_texture.h
@@ -0,0 +1,71 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+#ifndef SP_TEXTURE_H
+#define SP_TEXTURE_H
+
+
+struct pipe_context;
+struct pipe_texture;
+
+
+struct softpipe_texture
+{
+   struct pipe_texture base;
+
+   unsigned long level_offset[PIPE_MAX_TEXTURE_LEVELS];
+
+   /* The data is held here:
+    */
+   struct pipe_buffer *buffer;
+   unsigned long buffer_size;
+};
+
+
+/** cast wrapper */
+static INLINE struct softpipe_texture *
+softpipe_texture(struct pipe_texture *pt)
+{
+   return (struct softpipe_texture *) pt;
+}
+
+
+
+extern struct pipe_texture *
+softpipe_texture_create(struct pipe_context *pipe,
+                        const struct pipe_texture *templat);
+
+extern void
+softpipe_texture_release(struct pipe_context *pipe, struct pipe_texture **pt);
+
+extern struct pipe_surface *
+softpipe_get_tex_surface(struct pipe_context *pipe,
+                         struct pipe_texture *pt,
+                         unsigned face, unsigned level, unsigned zslice);
+
+
+#endif /* SP_TEXTURE */
diff --git a/src/gallium/drivers/softpipe/sp_tile_cache.c b/src/gallium/drivers/softpipe/sp_tile_cache.c
new file mode 100644
index 0000000000..1597361b82
--- /dev/null
+++ b/src/gallium/drivers/softpipe/sp_tile_cache.c
@@ -0,0 +1,585 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+/**
+ * Framebuffer/surface tile caching.
+ *
+ * Author:
+ *    Brian Paul
+ */
+
+#include "pipe/p_util.h"
+#include "pipe/p_inlines.h"
+#include "pipe/util/p_tile.h"
+#include "sp_context.h"
+#include "sp_surface.h"
+#include "sp_tile_cache.h"
+
+#define NUM_ENTRIES 30
+
+
+/** XXX move these */
+#define MAX_WIDTH 2048
+#define MAX_HEIGHT 2048
+
+
+struct softpipe_tile_cache
+{
+   struct pipe_surface *surface;  /**< the surface we're caching */
+   void *surface_map;
+   struct pipe_texture *texture;  /**< if caching a texture */
+   struct softpipe_cached_tile entries[NUM_ENTRIES];
+   uint clear_flags[(MAX_WIDTH / TILE_SIZE) * (MAX_HEIGHT / TILE_SIZE) / 32];
+   float clear_color[4];
+   uint clear_val;
+   boolean depth_stencil; /** Is the surface a depth/stencil format? */
+
+   struct pipe_surface *tex_surf;
+   void *tex_surf_map;
+   int tex_face, tex_level, tex_z;
+
+   struct softpipe_cached_tile tile;  /**< scratch tile for clears */
+};
+
+
+/**
+ * Return the position in the cache for the tile that contains win pos (x,y).
+ * We currently use a direct mapped cache so this is like a hack key.
+ * At some point we should investige something more sophisticated, like
+ * a LRU replacement policy.
+ */
+#define CACHE_POS(x, y) \
+   (((x) / TILE_SIZE + ((y) / TILE_SIZE) * 5) % NUM_ENTRIES)
+
+
+
+/**
+ * Is the tile at (x,y) in cleared state?
+ */
+static INLINE uint
+is_clear_flag_set(const uint *bitvec, int x, int y)
+{
+   int pos, bit;
+   x /= TILE_SIZE;
+   y /= TILE_SIZE;
+   pos = y * (MAX_WIDTH / TILE_SIZE) + x;
+   assert(pos / 32 < (MAX_WIDTH / TILE_SIZE) * (MAX_HEIGHT / TILE_SIZE) / 32);
+   bit = bitvec[pos / 32] & (1 << (pos & 31));
+   return bit;
+}
+   
+
+/**
+ * Mark the tile at (x,y) as not cleared.
+ */
+static INLINE void
+clear_clear_flag(uint *bitvec, int x, int y)
+{
+   int pos;
+   x /= TILE_SIZE;
+   y /= TILE_SIZE;
+   pos = y * (MAX_WIDTH / TILE_SIZE) + x;
+   assert(pos / 32 < (MAX_WIDTH / TILE_SIZE) * (MAX_HEIGHT / TILE_SIZE) / 32);
+   bitvec[pos / 32] &= ~(1 << (pos & 31));
+}
+   
+
+struct softpipe_tile_cache *
+sp_create_tile_cache(void)
+{
+   struct softpipe_tile_cache *tc;
+   uint pos;
+
+   tc = CALLOC_STRUCT( softpipe_tile_cache );
+   if (tc) {
+      for (pos = 0; pos < NUM_ENTRIES; pos++) {
+         tc->entries[pos].x =
+         tc->entries[pos].y = -1;
+      }
+   }
+   return tc;
+}
+
+
+void
+sp_destroy_tile_cache(struct softpipe_tile_cache *tc)
+{
+   uint pos;
+
+   for (pos = 0; pos < NUM_ENTRIES; pos++) {
+      //assert(tc->entries[pos].x < 0);
+   }
+   if (tc->surface) {
+      pipe_surface_reference(&tc->surface, NULL);
+   }
+   if (tc->tex_surf) {
+      pipe_surface_reference(&tc->tex_surf, NULL);
+   }
+
+   FREE( tc );
+}
+
+
+/**
+ * Specify the surface to cache.
+ */
+void
+sp_tile_cache_set_surface(struct softpipe_tile_cache *tc,
+                          struct pipe_surface *ps)
+{
+   assert(!tc->texture);
+
+   if (tc->surface_map) {
+      /*assert(tc->surface != ps);*/
+      pipe_surface_unmap(tc->surface);
+   }
+
+   pipe_surface_reference(&tc->surface, ps);
+
+   if (ps) {
+      if (tc->surface_map)
+	 tc->surface_map = pipe_surface_map(ps);
+
+      tc->depth_stencil = (ps->format == PIPE_FORMAT_S8Z24_UNORM ||
+                           ps->format == PIPE_FORMAT_Z16_UNORM ||
+                           ps->format == PIPE_FORMAT_Z32_UNORM ||
+                           ps->format == PIPE_FORMAT_U_S8);
+   }
+}
+
+
+/**
+ * Return the surface being cached.
+ */
+struct pipe_surface *
+sp_tile_cache_get_surface(struct softpipe_tile_cache *tc)
+{
+   return tc->surface;
+}
+
+
+void
+sp_tile_cache_map_surfaces(struct softpipe_tile_cache *tc)
+{
+   if (tc->surface && !tc->surface_map)
+      tc->surface_map = pipe_surface_map(tc->surface);
+
+   if (tc->tex_surf && !tc->tex_surf_map)
+      tc->tex_surf_map = pipe_surface_map(tc->tex_surf);
+}
+
+
+void
+sp_tile_cache_unmap_surfaces(struct softpipe_tile_cache *tc)
+{
+   if (tc->surface_map) {
+      pipe_surface_unmap(tc->surface);
+      tc->surface_map = NULL;
+   }
+
+   if (tc->tex_surf_map) {
+      pipe_surface_unmap(tc->tex_surf);
+      tc->tex_surf_map = NULL;
+   }
+}
+
+
+/**
+ * Specify the texture to cache.
+ */
+void
+sp_tile_cache_set_texture(struct softpipe_tile_cache *tc,
+                          struct pipe_texture *texture)
+{
+   uint i;
+
+   assert(!tc->surface);
+
+   tc->texture = texture;
+
+   if (tc->tex_surf_map) {
+      pipe_surface_unmap(tc->tex_surf);
+      tc->tex_surf_map = NULL;
+   }
+   pipe_surface_reference(&tc->tex_surf, NULL);
+
+   /* mark as entries as invalid/empty */
+   /* XXX we should try to avoid this when the teximage hasn't changed */
+   for (i = 0; i < NUM_ENTRIES; i++) {
+      tc->entries[i].x = -1;
+   }
+
+   tc->tex_face = -1; /* any invalid value here */
+}
+
+
+/**
+ * Set pixels in a tile to the given clear color/value, float.
+ */
+static void
+clear_tile_rgba(struct softpipe_cached_tile *tile,
+                enum pipe_format format,
+                const float clear_value[4])
+{
+   if (clear_value[0] == 0.0 &&
+       clear_value[1] == 0.0 &&
+       clear_value[2] == 0.0 &&
+       clear_value[3] == 0.0) {
+      memset(tile->data.color, 0, sizeof(tile->data.color));
+   }
+   else {
+      uint i, j;
+      for (i = 0; i < TILE_SIZE; i++) {
+         for (j = 0; j < TILE_SIZE; j++) {
+            tile->data.color[i][j][0] = clear_value[0];
+            tile->data.color[i][j][1] = clear_value[1];
+            tile->data.color[i][j][2] = clear_value[2];
+            tile->data.color[i][j][3] = clear_value[3];
+         }
+      }
+   }
+}
+
+
+/**
+ * Set a tile to a solid value/color.
+ */
+static void
+clear_tile(struct softpipe_cached_tile *tile,
+           enum pipe_format format,
+           uint clear_value)
+{
+   uint i, j;
+
+   switch (pf_get_size(format)) {
+   case 1:
+      memset(tile->data.any, 0, TILE_SIZE * TILE_SIZE);
+      break;
+   case 2:
+      if (clear_value == 0) {
+         memset(tile->data.any, 0, 2 * TILE_SIZE * TILE_SIZE);
+      }
+      else {
+         for (i = 0; i < TILE_SIZE; i++) {
+            for (j = 0; j < TILE_SIZE; j++) {
+               tile->data.depth16[i][j] = (ushort) clear_value;
+            }
+         }
+      }
+      break;
+   case 4:
+      if (clear_value == 0) {
+         memset(tile->data.any, 0, 4 * TILE_SIZE * TILE_SIZE);
+      }
+      else {
+         for (i = 0; i < TILE_SIZE; i++) {
+            for (j = 0; j < TILE_SIZE; j++) {
+               tile->data.color32[i][j] = clear_value;
+            }
+         }
+      }
+      break;
+   default:
+      assert(0);
+   }
+}
+
+
+/**
+ * Actually clear the tiles which were flagged as being in a clear state.
+ */
+static void
+sp_tile_cache_flush_clear(struct pipe_context *pipe,
+                          struct softpipe_tile_cache *tc)
+{
+   struct pipe_surface *ps = tc->surface;
+   const uint w = tc->surface->width;
+   const uint h = tc->surface->height;
+   uint x, y;
+   uint numCleared = 0;
+
+   /* clear the scratch tile to the clear value */
+   clear_tile(&tc->tile, ps->format, tc->clear_val);
+
+   /* push the tile to all positions marked as clear */
+   for (y = 0; y < h; y += TILE_SIZE) {
+      for (x = 0; x < w; x += TILE_SIZE) {
+         if (is_clear_flag_set(tc->clear_flags, x, y)) {
+            pipe_put_tile_raw(pipe, ps,
+                           x, y, TILE_SIZE, TILE_SIZE,
+                           tc->tile.data.color32, 0/*STRIDE*/);
+
+            /* do this? */
+            clear_clear_flag(tc->clear_flags, x, y);
+
+            numCleared++;
+         }
+      }
+   }
+#if 0
+   debug_printf("num cleared: %u\n", numCleared);
+#endif
+}
+
+
+/**
+ * Flush the tile cache: write all dirty tiles back to the surface.
+ * any tiles "flagged" as cleared will be "really" cleared.
+ */
+void
+sp_flush_tile_cache(struct softpipe_context *softpipe,
+                    struct softpipe_tile_cache *tc)
+{
+   struct pipe_context *pipe = &softpipe->pipe;
+   struct pipe_surface *ps = tc->surface;
+   int inuse = 0, pos;
+
+   if (!ps || !ps->buffer)
+      return;
+
+   for (pos = 0; pos < NUM_ENTRIES; pos++) {
+      struct softpipe_cached_tile *tile = tc->entries + pos;
+      if (tile->x >= 0) {
+         if (tc->depth_stencil) {
+            pipe_put_tile_raw(pipe, ps,
+                           tile->x, tile->y, TILE_SIZE, TILE_SIZE,
+                           tile->data.depth32, 0/*STRIDE*/);
+         }
+         else {
+            pipe_put_tile_rgba(pipe, ps,
+                               tile->x, tile->y, TILE_SIZE, TILE_SIZE,
+                               (float *) tile->data.color);
+         }
+         tile->x = tile->y = -1;  /* mark as empty */
+         inuse++;
+      }
+   }
+
+#if TILE_CLEAR_OPTIMIZATION
+   sp_tile_cache_flush_clear(&softpipe->pipe, tc);
+#endif
+
+#if 0
+   debug_printf("flushed tiles in use: %d\n", inuse);
+#endif
+}
+
+
+/**
+ * Get a tile from the cache.
+ * \param x, y  position of tile, in pixels
+ */
+struct softpipe_cached_tile *
+sp_get_cached_tile(struct softpipe_context *softpipe,
+                   struct softpipe_tile_cache *tc, int x, int y)
+{
+   struct pipe_context *pipe = &softpipe->pipe;
+   struct pipe_surface *ps = tc->surface;
+
+   /* tile pos in framebuffer: */
+   const int tile_x = x & ~(TILE_SIZE - 1);
+   const int tile_y = y & ~(TILE_SIZE - 1);
+
+   /* cache pos/entry: */
+   const int pos = CACHE_POS(x, y);
+   struct softpipe_cached_tile *tile = tc->entries + pos;
+
+   if (tile_x != tile->x ||
+       tile_y != tile->y) {
+
+      if (tile->x != -1) {
+         /* put dirty tile back in framebuffer */
+         if (tc->depth_stencil) {
+            pipe_put_tile_raw(pipe, ps,
+                              tile->x, tile->y, TILE_SIZE, TILE_SIZE,
+                              tile->data.depth32, 0/*STRIDE*/);
+         }
+         else {
+            pipe_put_tile_rgba(pipe, ps,
+                               tile->x, tile->y, TILE_SIZE, TILE_SIZE,
+                               (float *) tile->data.color);
+         }
+      }
+
+      tile->x = tile_x;
+      tile->y = tile_y;
+
+      if (is_clear_flag_set(tc->clear_flags, x, y)) {
+         /* don't get tile from framebuffer, just clear it */
+         if (tc->depth_stencil) {
+            clear_tile(tile, ps->format, tc->clear_val);
+         }
+         else {
+            clear_tile_rgba(tile, ps->format, tc->clear_color);
+         }
+         clear_clear_flag(tc->clear_flags, x, y);
+      }
+      else {
+         /* get new tile data from surface */
+         if (tc->depth_stencil) {
+            pipe_get_tile_raw(pipe, ps,
+                              tile->x, tile->y, TILE_SIZE, TILE_SIZE,
+                              tile->data.depth32, 0/*STRIDE*/);
+         }
+         else {
+            pipe_get_tile_rgba(pipe, ps,
+                               tile->x, tile->y, TILE_SIZE, TILE_SIZE,
+                               (float *) tile->data.color);
+         }
+      }
+   }
+
+   return tile;
+}
+
+
+/**
+ * Given the texture face, level, zslice, x and y values, compute
+ * the cache entry position/index where we'd hope to find the
+ * cached texture tile.
+ * This is basically a direct-map cache.
+ * XXX There's probably lots of ways in which we can improve this.
+ */
+static INLINE uint
+tex_cache_pos(int x, int y, int z, int face, int level)
+{
+   uint entry = x + y * 5 + z * 4 + face + level;
+   return entry % NUM_ENTRIES;
+}
+
+
+/**
+ * Similar to sp_get_cached_tile() but for textures.
+ * Tiles are read-only and indexed with more params.
+ */
+const struct softpipe_cached_tile *
+sp_get_cached_tile_tex(struct pipe_context *pipe,
+                       struct softpipe_tile_cache *tc, int x, int y, int z,
+                       int face, int level)
+{
+   /* tile pos in framebuffer: */
+   const int tile_x = x & ~(TILE_SIZE - 1);
+   const int tile_y = y & ~(TILE_SIZE - 1);
+   /* cache pos/entry: */
+   const uint pos = tex_cache_pos(x / TILE_SIZE, y / TILE_SIZE, z,
+                                  face, level);
+   struct softpipe_cached_tile *tile = tc->entries + pos;
+
+   if (tile_x != tile->x ||
+       tile_y != tile->y ||
+       z != tile->z ||
+       face != tile->face ||
+       level != tile->level) {
+      /* cache miss */
+
+      /* check if we need to get a new surface */
+      if (!tc->tex_surf ||
+          tc->tex_face != face ||
+          tc->tex_level != level ||
+          tc->tex_z != z) {
+         /* get new surface (view into texture) */
+
+	 if (tc->tex_surf_map)
+            pipe_surface_unmap(tc->tex_surf);
+
+         tc->tex_surf = pipe->get_tex_surface(pipe, tc->texture, face, level, z);
+         tc->tex_surf_map = pipe_surface_map(tc->tex_surf);
+
+         tc->tex_face = face;
+         tc->tex_level = level;
+         tc->tex_z = z;
+      }
+
+      /* get tile from the surface (view into texture) */
+      pipe_get_tile_rgba(pipe, tc->tex_surf,
+                         tile_x, tile_y, TILE_SIZE, TILE_SIZE,
+                         (float *) tile->data.color);
+      tile->x = tile_x;
+      tile->y = tile_y;
+      tile->z = z;
+      tile->face = face;
+      tile->level = level;
+   }
+
+   return tile;
+}
+
+
+/**
+ * When a whole surface is being cleared to a value we can avoid
+ * fetching tiles above.
+ * Save the color and set a 'clearflag' for each tile of the screen.
+ */
+void
+sp_tile_cache_clear(struct softpipe_tile_cache *tc, uint clearValue)
+{
+   uint r, g, b, a;
+   uint pos;
+
+   tc->clear_val = clearValue;
+
+   switch (tc->surface->format) {
+   case PIPE_FORMAT_R8G8B8A8_UNORM:
+      r = (clearValue >> 24) & 0xff;
+      g = (clearValue >> 16) & 0xff;
+      b = (clearValue >>  8) & 0xff;
+      a = (clearValue      ) & 0xff;
+      break;
+   case PIPE_FORMAT_A8R8G8B8_UNORM:
+      r = (clearValue >> 16) & 0xff;
+      g = (clearValue >>  8) & 0xff;
+      b = (clearValue      ) & 0xff;
+      a = (clearValue >> 24) & 0xff;
+      break;
+   case PIPE_FORMAT_B8G8R8A8_UNORM:
+      r = (clearValue >>  8) & 0xff;
+      g = (clearValue >> 16) & 0xff;
+      b = (clearValue >> 24) & 0xff;
+      a = (clearValue      ) & 0xff;
+      break;
+   default:
+      r = g = b = a = 0;
+   }
+
+   tc->clear_color[0] = r / 255.0f;
+   tc->clear_color[1] = g / 255.0f;
+   tc->clear_color[2] = b / 255.0f;
+   tc->clear_color[3] = a / 255.0f;
+
+#if TILE_CLEAR_OPTIMIZATION
+   /* set flags to indicate all the tiles are cleared */
+   memset(tc->clear_flags, 255, sizeof(tc->clear_flags));
+#else
+   /* disable the optimization */
+   memset(tc->clear_flags, 0, sizeof(tc->clear_flags));
+#endif
+
+   for (pos = 0; pos < NUM_ENTRIES; pos++) {
+      struct softpipe_cached_tile *tile = tc->entries + pos;
+      tile->x = tile->y = -1;
+   }
+}
diff --git a/src/gallium/drivers/softpipe/sp_tile_cache.h b/src/gallium/drivers/softpipe/sp_tile_cache.h
new file mode 100644
index 0000000000..7fd1081286
--- /dev/null
+++ b/src/gallium/drivers/softpipe/sp_tile_cache.h
@@ -0,0 +1,104 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+#ifndef SP_TILE_CACHE_H
+#define SP_TILE_CACHE_H
+
+#define TILE_CLEAR_OPTIMIZATION 1
+
+
+#include "pipe/p_compiler.h"
+
+
+struct softpipe_context;
+struct softpipe_tile_cache;
+
+
+/**
+ * Cache tile size (width and height). This needs to be a power of two.
+ */
+#define TILE_SIZE 64
+
+
+
+struct softpipe_cached_tile
+{
+   int x, y;           /**< pos of tile in window coords */
+   int z, face, level; /**< Extra texture indexes */
+   union {
+      float color[TILE_SIZE][TILE_SIZE][4];
+      uint color32[TILE_SIZE][TILE_SIZE];
+      uint depth32[TILE_SIZE][TILE_SIZE];
+      ushort depth16[TILE_SIZE][TILE_SIZE];
+      ubyte stencil8[TILE_SIZE][TILE_SIZE];
+      ubyte any[1];
+   } data;
+};
+
+
+extern struct softpipe_tile_cache *
+sp_create_tile_cache(void);
+
+extern void
+sp_destroy_tile_cache(struct softpipe_tile_cache *tc);
+
+extern void
+sp_tile_cache_set_surface(struct softpipe_tile_cache *tc,
+                          struct pipe_surface *sps);
+
+extern struct pipe_surface *
+sp_tile_cache_get_surface(struct softpipe_tile_cache *tc);
+
+extern void
+sp_tile_cache_map_surfaces(struct softpipe_tile_cache *tc);
+
+extern void
+sp_tile_cache_unmap_surfaces(struct softpipe_tile_cache *tc);
+
+extern void
+sp_tile_cache_set_texture(struct softpipe_tile_cache *tc,
+                          struct pipe_texture *texture);
+
+extern void
+sp_flush_tile_cache(struct softpipe_context *softpipe,
+                    struct softpipe_tile_cache *tc);
+
+extern void
+sp_tile_cache_clear(struct softpipe_tile_cache *tc, uint clearValue);
+
+extern struct softpipe_cached_tile *
+sp_get_cached_tile(struct softpipe_context *softpipe,
+                   struct softpipe_tile_cache *tc, int x, int y);
+
+extern const struct softpipe_cached_tile *
+sp_get_cached_tile_tex(struct pipe_context *pipe,
+                       struct softpipe_tile_cache *tc, int x, int y, int z,
+                       int face, int level);
+
+
+#endif /* SP_TILE_CACHE_H */
+
diff --git a/src/gallium/drivers/softpipe/sp_winsys.h b/src/gallium/drivers/softpipe/sp_winsys.h
new file mode 100644
index 0000000000..d6b379f58c
--- /dev/null
+++ b/src/gallium/drivers/softpipe/sp_winsys.h
@@ -0,0 +1,57 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+/* This is the interface that softpipe requires any window system
+ * hosting it to implement.  This is the only include file in softpipe
+ * which is public.
+ */
+
+
+#ifndef SP_WINSYS_H
+#define SP_WINSYS_H
+
+
+#include "pipe/p_compiler.h" /* for boolean */
+
+enum pipe_format;
+
+struct softpipe_winsys {
+   /** test if the given format is supported for front/back color bufs */
+   boolean (*is_format_supported)( struct softpipe_winsys *sws,
+                                   enum pipe_format format );
+
+};
+
+struct pipe_winsys;
+struct pipe_context;
+
+
+struct pipe_context *softpipe_create( struct pipe_winsys *,
+				      struct softpipe_winsys * );
+
+
+#endif /* SP_WINSYS_H */
diff --git a/src/gallium/include/pipe/p_compiler.h b/src/gallium/include/pipe/p_compiler.h
new file mode 100644
index 0000000000..30cd729c56
--- /dev/null
+++ b/src/gallium/include/pipe/p_compiler.h
@@ -0,0 +1,116 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+#ifndef P_COMPILER_H
+#define P_COMPILER_H
+
+
+#include <stdlib.h>
+#include <string.h>
+
+
+#if defined(_WIN32) && !defined(__WIN32__)
+#define __WIN32__
+#endif
+
+#if defined(_MSC_VER) && !defined(__MSC__)
+#define __MSC__
+#endif
+
+
+typedef unsigned int       uint;
+typedef unsigned char      ubyte;
+typedef unsigned char      boolean;
+typedef unsigned short     ushort;
+typedef unsigned long long uint64;
+
+
+#if defined(__MSC__)
+
+typedef char               int8_t;
+typedef unsigned char      uint8_t;
+typedef short              int16_t;
+typedef unsigned short     uint16_t;
+typedef long               int32_t;
+typedef unsigned long      uint32_t;
+typedef long long          int64_t;
+typedef unsigned long long uint64_t;
+
+#if defined(_WIN64)
+typedef unsigned __int64   uintptr_t;
+#else
+typedef unsigned int       uintptr_t;
+#endif
+
+#else
+#include <stdint.h>
+#endif
+
+
+#define TRUE  1
+#define FALSE 0
+
+
+/* Function inlining */
+#if defined(__GNUC__)
+#  define INLINE __inline__
+#elif defined(__MSC__)
+#  define INLINE __inline
+#elif defined(__ICL)
+#  define INLINE __inline
+#elif defined(__INTEL_COMPILER)
+#  define INLINE inline
+#elif defined(__WATCOMC__) && (__WATCOMC__ >= 1100)
+#  define INLINE __inline
+#else
+#  define INLINE
+#endif
+
+
+#if defined __GNUC__
+#define ALIGN16_DECL(TYPE, NAME, SIZE)  TYPE NAME##___aligned[SIZE] __attribute__(( aligned( 16 ) ))
+#define ALIGN16_ASSIGN(NAME) NAME##___aligned
+#define ALIGN16_ATTRIB  __attribute__(( aligned( 16 ) ))
+#else
+#define ALIGN16_DECL(TYPE, NAME, SIZE)  TYPE NAME##___unaligned[SIZE + 1]
+#define ALIGN16_ASSIGN(NAME) align16(NAME##___unaligned)
+#define ALIGN16_ATTRIB
+#endif
+
+
+
+/** For calling code-gen'd functions */
+#if !defined(XSTDCALL) 
+#if defined(WIN32)
+#define XSTDCALL __stdcall
+#else
+#define XSTDCALL
+#endif
+#endif
+
+
+#endif /* P_COMPILER_H */
diff --git a/src/gallium/include/pipe/p_context.h b/src/gallium/include/pipe/p_context.h
new file mode 100644
index 0000000000..39f95695fb
--- /dev/null
+++ b/src/gallium/include/pipe/p_context.h
@@ -0,0 +1,221 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+#ifndef PIPE_CONTEXT_H
+#define PIPE_CONTEXT_H
+
+#include "p_state.h"
+
+
+struct pipe_state_cache;
+
+/* Opaque driver handles:
+ */
+struct pipe_query;
+
+/**
+ * Gallium rendering context.  Basically:
+ *  - state setting functions
+ *  - VBO drawing functions
+ *  - surface functions
+ *  - device queries
+ */
+struct pipe_context {
+   struct pipe_winsys *winsys;
+
+   void *priv;  /** context private data (for DRI for example) */
+
+   void (*destroy)( struct pipe_context * );
+
+   /*
+    * Queries
+    */
+   /** type is one of PIPE_SURFACE, PIPE_TEXTURE, etc. */
+   boolean (*is_format_supported)( struct pipe_context *pipe,
+                                   enum pipe_format format, uint type );
+
+   const char *(*get_name)( struct pipe_context *pipe );
+
+   const char *(*get_vendor)( struct pipe_context *pipe );
+
+   int (*get_param)( struct pipe_context *pipe, int param );
+   float (*get_paramf)( struct pipe_context *pipe, int param );
+
+
+   /*
+    * Drawing.  
+    * Return false on fallbacks (temporary??)
+    */
+   boolean (*draw_arrays)( struct pipe_context *pipe,
+			   unsigned mode, unsigned start, unsigned count);
+
+   boolean (*draw_elements)( struct pipe_context *pipe,
+			     struct pipe_buffer *indexBuffer,
+			     unsigned indexSize,
+			     unsigned mode, unsigned start, unsigned count);
+
+
+   /**
+    * Query objects
+    */
+   struct pipe_query *(*create_query)( struct pipe_context *pipe,
+                                              unsigned query_type );
+
+   void (*destroy_query)(struct pipe_context *pipe,
+                         struct pipe_query *q);
+
+   void (*begin_query)(struct pipe_context *pipe, struct pipe_query *q);
+   void (*end_query)(struct pipe_context *pipe, struct pipe_query *q);
+
+   boolean (*get_query_result)(struct pipe_context *pipe, 
+                               struct pipe_query *q,
+                               boolean wait,
+                               uint64 *result);
+
+   /*
+    * State functions
+    */
+   void * (*create_blend_state)(struct pipe_context *,
+                                const struct pipe_blend_state *);
+   void   (*bind_blend_state)(struct pipe_context *, void *);
+   void   (*delete_blend_state)(struct pipe_context *, void  *);
+
+   void * (*create_sampler_state)(struct pipe_context *,
+                                  const struct pipe_sampler_state *);
+   void   (*bind_sampler_state)(struct pipe_context *, unsigned unit, void *);
+   void   (*delete_sampler_state)(struct pipe_context *, void *);
+
+   void * (*create_rasterizer_state)(struct pipe_context *,
+                                     const struct pipe_rasterizer_state *);
+   void   (*bind_rasterizer_state)(struct pipe_context *, void *);
+   void   (*delete_rasterizer_state)(struct pipe_context *, void *);
+
+   void * (*create_depth_stencil_alpha_state)(struct pipe_context *,
+                                        const struct pipe_depth_stencil_alpha_state *);
+   void   (*bind_depth_stencil_alpha_state)(struct pipe_context *, void *);
+   void   (*delete_depth_stencil_alpha_state)(struct pipe_context *, void *);
+
+   void * (*create_fs_state)(struct pipe_context *,
+                             const struct pipe_shader_state *);
+   void   (*bind_fs_state)(struct pipe_context *, void *);
+   void   (*delete_fs_state)(struct pipe_context *, void *);
+
+   void * (*create_vs_state)(struct pipe_context *,
+                             const struct pipe_shader_state *);
+   void   (*bind_vs_state)(struct pipe_context *, void *);
+   void   (*delete_vs_state)(struct pipe_context *, void *);
+
+   /* The following look more properties than states.
+    * maybe combine a few of them into states or pass them
+    * in the bind calls to the state */
+   void (*set_blend_color)( struct pipe_context *,
+                            const struct pipe_blend_color * );
+
+   void (*set_clip_state)( struct pipe_context *,
+			   const struct pipe_clip_state * );
+
+   void (*set_constant_buffer)( struct pipe_context *,
+                                uint shader, uint index,
+                                const struct pipe_constant_buffer *buf );
+
+   void (*set_framebuffer_state)( struct pipe_context *,
+                                  const struct pipe_framebuffer_state * );
+
+   void (*set_polygon_stipple)( struct pipe_context *,
+				const struct pipe_poly_stipple * );
+
+   void (*set_scissor_state)( struct pipe_context *,
+                              const struct pipe_scissor_state * );
+
+
+   /* Currently a sampler is constrained to sample from a single texture:
+    */
+   void (*set_sampler_texture)( struct pipe_context *,
+				unsigned sampler,
+				struct pipe_texture * );
+
+   void (*set_viewport_state)( struct pipe_context *,
+                               const struct pipe_viewport_state * );
+
+   /*
+    * Vertex arrays
+    */
+   void (*set_vertex_buffer)( struct pipe_context *,
+                              unsigned index,
+                              const struct pipe_vertex_buffer * );
+
+   void (*set_vertex_element)( struct pipe_context *,
+			       unsigned index,
+			       const struct pipe_vertex_element * );
+
+
+   /*
+    * Surface functions
+    */
+
+   void (*surface_copy)(struct pipe_context *pipe,
+                        unsigned do_flip,	/*<< flip surface contents vertically */
+			struct pipe_surface *dest,
+			unsigned destx, unsigned desty,
+			struct pipe_surface *src, /* don't make this const - 
+						     need to map/unmap */
+			unsigned srcx, unsigned srcy,
+			unsigned width, unsigned height);
+
+   void (*surface_fill)(struct pipe_context *pipe,
+			struct pipe_surface *dst,
+			unsigned dstx, unsigned dsty,
+			unsigned width, unsigned height,
+			unsigned value);
+
+   void (*clear)(struct pipe_context *pipe, 
+		 struct pipe_surface *ps,
+		 unsigned clearValue);
+
+
+   /*
+    * Texture functions
+    */
+   struct pipe_texture * (*texture_create)(struct pipe_context *pipe,
+                                           const struct pipe_texture *templat);
+
+   void (*texture_release)(struct pipe_context *pipe,
+			   struct pipe_texture **pt);
+
+   /** Get a surface which is a "view" into a texture */
+   struct pipe_surface *(*get_tex_surface)(struct pipe_context *pipe,
+                                           struct pipe_texture *texture,
+                                           unsigned face, unsigned level,
+                                           unsigned zslice);
+
+   /* Flush rendering:
+    */
+   void (*flush)( struct pipe_context *pipe,
+		  unsigned flags );
+};
+
+#endif /* PIPE_CONTEXT_H */
diff --git a/src/gallium/include/pipe/p_debug.h b/src/gallium/include/pipe/p_debug.h
new file mode 100644
index 0000000000..2a11627b36
--- /dev/null
+++ b/src/gallium/include/pipe/p_debug.h
@@ -0,0 +1,86 @@
+/**************************************************************************
+ * 
+ * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+/**
+ * @file
+ * Cross-platform debugging helpers.
+ * 
+ * For now it just has assert and printf replacements, but it might be extended 
+ * with stack trace reports and more advanced logging in the near future. 
+ * 
+ * @author Jose Fonseca <jrfonseca@tungstengraphics.com>
+ */
+
+#ifndef P_DEBUG_H_
+#define P_DEBUG_H_
+
+
+#include <stdarg.h>
+
+
+#ifdef	__cplusplus
+extern "C" {
+#endif
+
+
+#ifdef DBG
+#ifndef DEBUG
+#define DEBUG 1
+#endif
+#else
+#ifndef NDEBUG
+#define NDEBUG 1
+#endif
+#endif
+
+
+void debug_printf(const char *format, ...);
+
+void debug_vprintf(const char *format, va_list ap);
+
+void debug_assert_fail(const char *expr, const char *file, unsigned line);
+
+
+/** Assert macro */
+#ifdef DEBUG
+#define debug_assert(expr) ((expr) ? (void)0 : debug_assert_fail(#expr, __FILE__, __LINE__))
+#else
+#define debug_assert(expr) ((void)0)
+#endif
+
+
+#ifdef assert
+#undef assert
+#endif
+#define assert(expr) debug_assert(expr)
+
+
+#ifdef	__cplusplus
+}
+#endif
+
+#endif /* P_DEBUG_H_ */
diff --git a/src/gallium/include/pipe/p_defines.h b/src/gallium/include/pipe/p_defines.h
new file mode 100644
index 0000000000..0bf53ecb79
--- /dev/null
+++ b/src/gallium/include/pipe/p_defines.h
@@ -0,0 +1,270 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+#ifndef PIPE_DEFINES_H
+#define PIPE_DEFINES_H
+
+#include "p_format.h"
+
+#define PIPE_BLENDFACTOR_ONE                 0x1
+#define PIPE_BLENDFACTOR_SRC_COLOR           0x2
+#define PIPE_BLENDFACTOR_SRC_ALPHA           0x3
+#define PIPE_BLENDFACTOR_DST_ALPHA           0x4
+#define PIPE_BLENDFACTOR_DST_COLOR           0x5
+#define PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE  0x6
+#define PIPE_BLENDFACTOR_CONST_COLOR         0x7
+#define PIPE_BLENDFACTOR_CONST_ALPHA         0x8
+#define PIPE_BLENDFACTOR_SRC1_COLOR          0x9
+#define PIPE_BLENDFACTOR_SRC1_ALPHA          0x0A
+#define PIPE_BLENDFACTOR_ZERO                0x11
+#define PIPE_BLENDFACTOR_INV_SRC_COLOR       0x12
+#define PIPE_BLENDFACTOR_INV_SRC_ALPHA       0x13
+#define PIPE_BLENDFACTOR_INV_DST_ALPHA       0x14
+#define PIPE_BLENDFACTOR_INV_DST_COLOR       0x15
+#define PIPE_BLENDFACTOR_INV_CONST_COLOR     0x17
+#define PIPE_BLENDFACTOR_INV_CONST_ALPHA     0x18
+#define PIPE_BLENDFACTOR_INV_SRC1_COLOR      0x19
+#define PIPE_BLENDFACTOR_INV_SRC1_ALPHA      0x1A
+
+#define PIPE_BLEND_ADD               0
+#define PIPE_BLEND_SUBTRACT          1
+#define PIPE_BLEND_REVERSE_SUBTRACT  2
+#define PIPE_BLEND_MIN               3
+#define PIPE_BLEND_MAX               4
+
+#define PIPE_LOGICOP_CLEAR            0
+#define PIPE_LOGICOP_NOR              1
+#define PIPE_LOGICOP_AND_INVERTED     2
+#define PIPE_LOGICOP_COPY_INVERTED    3
+#define PIPE_LOGICOP_AND_REVERSE      4
+#define PIPE_LOGICOP_INVERT           5
+#define PIPE_LOGICOP_XOR              6
+#define PIPE_LOGICOP_NAND             7
+#define PIPE_LOGICOP_AND              8
+#define PIPE_LOGICOP_EQUIV            9
+#define PIPE_LOGICOP_NOOP             10
+#define PIPE_LOGICOP_OR_INVERTED      11
+#define PIPE_LOGICOP_COPY             12
+#define PIPE_LOGICOP_OR_REVERSE       13
+#define PIPE_LOGICOP_OR               14
+#define PIPE_LOGICOP_SET              15  
+
+#define PIPE_MASK_R  0x1
+#define PIPE_MASK_G  0x2
+#define PIPE_MASK_B  0x4
+#define PIPE_MASK_A  0x8
+#define PIPE_MASK_RGBA 0xf
+
+
+/**
+ * Inequality functions.  Used for depth test, stencil compare, alpha
+ * test, shadow compare, etc.
+ */
+#define PIPE_FUNC_NEVER    0
+#define PIPE_FUNC_LESS     1
+#define PIPE_FUNC_EQUAL    2
+#define PIPE_FUNC_LEQUAL   3
+#define PIPE_FUNC_GREATER  4
+#define PIPE_FUNC_NOTEQUAL 5
+#define PIPE_FUNC_GEQUAL   6
+#define PIPE_FUNC_ALWAYS   7
+
+/** Polygon fill mode */
+#define PIPE_POLYGON_MODE_FILL  0
+#define PIPE_POLYGON_MODE_LINE  1
+#define PIPE_POLYGON_MODE_POINT 2
+
+/** Polygon front/back window, also for culling */
+#define PIPE_WINDING_NONE 0
+#define PIPE_WINDING_CW   1
+#define PIPE_WINDING_CCW  2
+#define PIPE_WINDING_BOTH (PIPE_WINDING_CW | PIPE_WINDING_CCW)
+
+/** Stencil ops */
+#define PIPE_STENCIL_OP_KEEP       0
+#define PIPE_STENCIL_OP_ZERO       1
+#define PIPE_STENCIL_OP_REPLACE    2
+#define PIPE_STENCIL_OP_INCR       3
+#define PIPE_STENCIL_OP_DECR       4
+#define PIPE_STENCIL_OP_INCR_WRAP  5
+#define PIPE_STENCIL_OP_DECR_WRAP  6
+#define PIPE_STENCIL_OP_INVERT     7
+
+/** Texture types */
+enum pipe_texture_target {
+   PIPE_TEXTURE_1D   = 0,
+   PIPE_TEXTURE_2D   = 1,
+   PIPE_TEXTURE_3D   = 2,
+   PIPE_TEXTURE_CUBE = 3
+};
+
+#define PIPE_TEX_FACE_POS_X 0
+#define PIPE_TEX_FACE_NEG_X 1
+#define PIPE_TEX_FACE_POS_Y 2
+#define PIPE_TEX_FACE_NEG_Y 3
+#define PIPE_TEX_FACE_POS_Z 4
+#define PIPE_TEX_FACE_NEG_Z 5
+
+#define PIPE_TEX_WRAP_REPEAT                   0
+#define PIPE_TEX_WRAP_CLAMP                    1
+#define PIPE_TEX_WRAP_CLAMP_TO_EDGE            2
+#define PIPE_TEX_WRAP_CLAMP_TO_BORDER          3
+#define PIPE_TEX_WRAP_MIRROR_REPEAT            4
+#define PIPE_TEX_WRAP_MIRROR_CLAMP             5
+#define PIPE_TEX_WRAP_MIRROR_CLAMP_TO_EDGE     6
+#define PIPE_TEX_WRAP_MIRROR_CLAMP_TO_BORDER   7
+
+/* Between mipmaps, ie mipfilter
+ */
+#define PIPE_TEX_MIPFILTER_NEAREST  0
+#define PIPE_TEX_MIPFILTER_LINEAR   1
+#define PIPE_TEX_MIPFILTER_NONE     2
+
+/* Within a mipmap, ie min/mag filter 
+ */
+#define PIPE_TEX_FILTER_NEAREST      0
+#define PIPE_TEX_FILTER_LINEAR       1
+//#define PIPE_TEX_FILTER_ANISO        2
+
+
+#define PIPE_TEX_COMPARE_NONE          0
+#define PIPE_TEX_COMPARE_R_TO_TEXTURE  1
+
+#define PIPE_TEX_FACE_POS_X   0
+#define PIPE_TEX_FACE_NEG_X   1
+#define PIPE_TEX_FACE_POS_Y   2
+#define PIPE_TEX_FACE_NEG_Y   3
+#define PIPE_TEX_FACE_POS_Z   4
+#define PIPE_TEX_FACE_NEG_Z   5
+#define PIPE_TEX_FACE_MAX     6
+
+/**
+ * Surfaces, textures, etc. (others may be added)
+ */
+#define PIPE_TEXTURE        1
+#define PIPE_SURFACE        2  /**< user-created surfaces */
+
+
+/**
+ * Surface status
+ */
+#define PIPE_SURFACE_STATUS_UNDEFINED  0
+#define PIPE_SURFACE_STATUS_DEFINED    1
+#define PIPE_SURFACE_STATUS_CLEAR      2
+
+
+/**
+ * Buffer usage flags
+ */
+#define PIPE_BUFFER_USAGE_CPU_READ  (1 << 0)
+#define PIPE_BUFFER_USAGE_CPU_WRITE (1 << 1)
+#define PIPE_BUFFER_USAGE_GPU_READ  (1 << 2)
+#define PIPE_BUFFER_USAGE_GPU_WRITE (1 << 3)
+#define PIPE_BUFFER_USAGE_PIXEL     (1 << 4)
+#define PIPE_BUFFER_USAGE_VERTEX    (1 << 5)
+#define PIPE_BUFFER_USAGE_INDEX     (1 << 6)
+#define PIPE_BUFFER_USAGE_CONSTANT  (1 << 7)
+/** Pipe driver custam usage flags should be greater or equal to this value */
+#define PIPE_BUFFER_USAGE_CUSTOM    (1 << 16)
+
+
+/** 
+ * Flush types:
+ */
+#define PIPE_FLUSH_RENDER_CACHE   0x1
+#define PIPE_FLUSH_TEXTURE_CACHE  0x2
+#define PIPE_FLUSH_WAIT           0x4
+#define PIPE_FLUSH_SWAPBUFFERS    0x8 
+
+
+/**
+ * Shaders
+ */
+#define PIPE_SHADER_VERTEX   0
+#define PIPE_SHADER_FRAGMENT 1
+#define PIPE_SHADER_TYPES    2
+
+
+/**
+ * Primitive types:
+ */
+#define PIPE_PRIM_POINTS          0
+#define PIPE_PRIM_LINES           1
+#define PIPE_PRIM_LINE_LOOP       2
+#define PIPE_PRIM_LINE_STRIP      3
+#define PIPE_PRIM_TRIANGLES       4
+#define PIPE_PRIM_TRIANGLE_STRIP  5
+#define PIPE_PRIM_TRIANGLE_FAN    6
+#define PIPE_PRIM_QUADS           7
+#define PIPE_PRIM_QUAD_STRIP      8
+#define PIPE_PRIM_POLYGON         9
+
+
+/**
+ * Query object types
+ */
+#define PIPE_QUERY_OCCLUSION_COUNTER     0
+#define PIPE_QUERY_PRIMITIVES_GENERATED  1
+#define PIPE_QUERY_PRIMITIVES_EMITTED    2
+#define PIPE_QUERY_TYPES                 3
+
+
+/**
+ * Point sprite coord modes
+ */
+#define PIPE_SPRITE_COORD_NONE       0
+#define PIPE_SPRITE_COORD_UPPER_LEFT 1
+#define PIPE_SPRITE_COORD_LOWER_LEFT 2
+
+
+/**
+ * Implementation capabilities/limits
+ * Passed to pipe->get_param()
+ * XXX this will need some fine tuning...
+ */
+#define PIPE_CAP_MAX_TEXTURE_IMAGE_UNITS 1
+#define PIPE_CAP_NPOT_TEXTURES           2
+#define PIPE_CAP_TWO_SIDED_STENCIL       3
+#define PIPE_CAP_GLSL                    4  /* XXX need something better */
+#define PIPE_CAP_S3TC                    5
+#define PIPE_CAP_ANISOTROPIC_FILTER      6
+#define PIPE_CAP_POINT_SPRITE            7
+#define PIPE_CAP_MAX_RENDER_TARGETS      8
+#define PIPE_CAP_OCCLUSION_QUERY         9
+#define PIPE_CAP_TEXTURE_SHADOW_MAP      10
+#define PIPE_CAP_MAX_TEXTURE_2D_LEVELS   11
+#define PIPE_CAP_MAX_TEXTURE_3D_LEVELS   12
+#define PIPE_CAP_MAX_TEXTURE_CUBE_LEVELS 13
+#define PIPE_CAP_MAX_LINE_WIDTH          14
+#define PIPE_CAP_MAX_LINE_WIDTH_AA       15
+#define PIPE_CAP_MAX_POINT_WIDTH         16
+#define PIPE_CAP_MAX_POINT_WIDTH_AA      17
+#define PIPE_CAP_MAX_TEXTURE_ANISOTROPY  18
+#define PIPE_CAP_MAX_TEXTURE_LOD_BIAS    19
+#define PIPE_CAP_BITMAP_TEXCOORD_BIAS    20
+
+#endif
diff --git a/src/gallium/include/pipe/p_format.h b/src/gallium/include/pipe/p_format.h
new file mode 100644
index 0000000000..c9ad324315
--- /dev/null
+++ b/src/gallium/include/pipe/p_format.h
@@ -0,0 +1,421 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+#ifndef PIPE_FORMAT_H
+#define PIPE_FORMAT_H
+
+#include <stdio.h> // for sprintf
+
+#include "p_compiler.h"
+#include "p_debug.h"
+
+/**
+ * The PIPE_FORMAT is a 32-bit wide bitfield that encodes all the information
+ * needed to uniquely describe a pixel format.
+ */
+
+/**
+ * Possible format layouts -- occupy first 2 bits. The interpretation of
+ * the remaining 30 bits depends on a particual format layout.
+ */
+#define PIPE_FORMAT_LAYOUT_RGBAZS   0
+#define PIPE_FORMAT_LAYOUT_YCBCR    1
+
+static INLINE uint pf_layout(uint f)  /**< PIPE_FORMAT_LAYOUT_ */
+{
+   return f & 0x3;
+}
+
+/**
+ * RGBAZS Format Layout.
+ */
+
+/**
+ * Format component selectors for RGBAZS layout.
+ */
+#define PIPE_FORMAT_COMP_R    0
+#define PIPE_FORMAT_COMP_G    1
+#define PIPE_FORMAT_COMP_B    2
+#define PIPE_FORMAT_COMP_A    3
+#define PIPE_FORMAT_COMP_0    4
+#define PIPE_FORMAT_COMP_1    5
+#define PIPE_FORMAT_COMP_Z    6
+#define PIPE_FORMAT_COMP_S    7
+
+/**
+ * Format types for RGBAZS layout.
+ */
+#define PIPE_FORMAT_TYPE_UNKNOWN 0
+#define PIPE_FORMAT_TYPE_FLOAT   1
+#define PIPE_FORMAT_TYPE_UNORM   2
+#define PIPE_FORMAT_TYPE_SNORM   3
+#define PIPE_FORMAT_TYPE_USCALED 4
+#define PIPE_FORMAT_TYPE_SSCALED 5
+
+/**
+ * Because the destination vector is assumed to be RGBA FLOAT, we
+ * need to know how to swizzle and expand components from the source
+ * vector.
+ * Let's take U_A1_R5_G5_B5 as an example. X swizzle is A, X size
+ * is 1 bit and type is UNORM. So we take the most significant bit
+ * from source vector, convert 0 to 0.0 and 1 to 1.0 and save it
+ * in the last component of the destination RGBA component.
+ * Next, Y swizzle is R, Y size is 5 and type is UNORM. We normalize
+ * those 5 bits into [0.0; 1.0] range and put it into second
+ * component of the destination vector. Rinse and repeat for
+ * components Z and W.
+ * If any of size fields is zero, it means the source format contains
+ * less than four components.
+ * If any swizzle is 0 or 1, the corresponding destination component
+ * should be filled with 0.0 and 1.0, respectively.
+ */
+typedef uint pipe_format_rgbazs_t;
+
+static INLINE uint pf_get(pipe_format_rgbazs_t f, uint shift, uint mask)
+{
+   return (f >> shift) & mask;
+}
+
+/* XXX: The bit layout needs to be revised, can't currently encode 10-bit components. */
+
+#define pf_swizzle_x(f)       pf_get(f, 2, 0x7)  /**< PIPE_FORMAT_COMP_ */
+#define pf_swizzle_y(f)       pf_get(f, 5, 0x7)  /**< PIPE_FORMAT_COMP_ */
+#define pf_swizzle_z(f)       pf_get(f, 8, 0x7)  /**< PIPE_FORMAT_COMP_ */
+#define pf_swizzle_w(f)       pf_get(f, 11, 0x7) /**< PIPE_FORMAT_COMP_ */
+#define pf_swizzle_xyzw(f,i)  pf_get(f, 2+((i)*3), 0x7)
+#define pf_size_x(f)          pf_get(f, 14, 0x7) /**< Size of X */
+#define pf_size_y(f)          pf_get(f, 17, 0x7) /**< Size of Y */
+#define pf_size_z(f)          pf_get(f, 20, 0x7) /**< Size of Z */
+#define pf_size_w(f)          pf_get(f, 23, 0x7) /**< Size of W */
+#define pf_size_xyzw(f,i)     pf_get(f, 14+((i)*3), 0x7)
+#define pf_exp8(f)            pf_get(f, 26, 0x3) /**< Scale size by 8 ^ exp8 */
+#define pf_type(f)            pf_get(f, 28, 0xf) /**< PIPE_FORMAT_TYPE_ */
+
+/**
+ * Helper macro to encode the above structure into a 32-bit value.
+ */
+#define _PIPE_FORMAT_RGBAZS( SWZ, SIZEX, SIZEY, SIZEZ, SIZEW, EXP8, TYPE ) (\
+   (PIPE_FORMAT_LAYOUT_RGBAZS << 0) |\
+   ((SWZ) << 2) |\
+   ((SIZEX) << 14) |\
+   ((SIZEY) << 17) |\
+   ((SIZEZ) << 20) |\
+   ((SIZEW) << 23) |\
+   ((EXP8) << 26) |\
+   ((TYPE) << 28) )
+
+/**
+ * Helper macro to encode the swizzle part of the structure above.
+ */
+#define _PIPE_FORMAT_SWZ( SWZX, SWZY, SWZZ, SWZW ) (((SWZX) << 0) | ((SWZY) << 3) | ((SWZZ) << 6) | ((SWZW) << 9))
+
+/**
+ * Shorthand macro for RGBAZS layout with component sizes in 1-bit units.
+ */
+#define _PIPE_FORMAT_RGBAZS_1( SWZ, SIZEX, SIZEY, SIZEZ, SIZEW, TYPE )\
+   _PIPE_FORMAT_RGBAZS( SWZ, SIZEX, SIZEY, SIZEZ, SIZEW, 0, TYPE )
+
+/**
+ * Shorthand macro for RGBAZS layout with component sizes in 8-bit units.
+ */
+#define _PIPE_FORMAT_RGBAZS_8( SWZ, SIZEX, SIZEY, SIZEZ, SIZEW, TYPE )\
+   _PIPE_FORMAT_RGBAZS( SWZ, SIZEX, SIZEY, SIZEZ, SIZEW, 1, TYPE )
+
+/**
+ * Shorthand macro for RGBAZS layout with component sizes in 64-bit units.
+ */
+#define _PIPE_FORMAT_RGBAZS_64( SWZ, SIZEX, SIZEY, SIZEZ, SIZEW, TYPE )\
+   _PIPE_FORMAT_RGBAZS( SWZ, SIZEX, SIZEY, SIZEZ, SIZEW, 2, TYPE )
+
+/**
+ * Shorthand macro for common format swizzles.
+ */
+#define _PIPE_FORMAT_R000 _PIPE_FORMAT_SWZ( PIPE_FORMAT_COMP_R, PIPE_FORMAT_COMP_0, PIPE_FORMAT_COMP_0, PIPE_FORMAT_COMP_0 )
+#define _PIPE_FORMAT_RG00 _PIPE_FORMAT_SWZ( PIPE_FORMAT_COMP_R, PIPE_FORMAT_COMP_G, PIPE_FORMAT_COMP_0, PIPE_FORMAT_COMP_0 )
+#define _PIPE_FORMAT_RGB0 _PIPE_FORMAT_SWZ( PIPE_FORMAT_COMP_R, PIPE_FORMAT_COMP_G, PIPE_FORMAT_COMP_B, PIPE_FORMAT_COMP_0 )
+#define _PIPE_FORMAT_RGBA _PIPE_FORMAT_SWZ( PIPE_FORMAT_COMP_R, PIPE_FORMAT_COMP_G, PIPE_FORMAT_COMP_B, PIPE_FORMAT_COMP_A )
+#define _PIPE_FORMAT_ARGB _PIPE_FORMAT_SWZ( PIPE_FORMAT_COMP_A, PIPE_FORMAT_COMP_R, PIPE_FORMAT_COMP_G, PIPE_FORMAT_COMP_B )
+#define _PIPE_FORMAT_BGRA _PIPE_FORMAT_SWZ( PIPE_FORMAT_COMP_B, PIPE_FORMAT_COMP_G, PIPE_FORMAT_COMP_R, PIPE_FORMAT_COMP_A )
+#define _PIPE_FORMAT_0000 _PIPE_FORMAT_SWZ( PIPE_FORMAT_COMP_0, PIPE_FORMAT_COMP_0, PIPE_FORMAT_COMP_0, PIPE_FORMAT_COMP_0 )
+#define _PIPE_FORMAT_000R _PIPE_FORMAT_SWZ( PIPE_FORMAT_COMP_0, PIPE_FORMAT_COMP_0, PIPE_FORMAT_COMP_0, PIPE_FORMAT_COMP_R )
+#define _PIPE_FORMAT_RRR1 _PIPE_FORMAT_SWZ( PIPE_FORMAT_COMP_R, PIPE_FORMAT_COMP_R, PIPE_FORMAT_COMP_R, PIPE_FORMAT_COMP_1 )
+#define _PIPE_FORMAT_RRRR _PIPE_FORMAT_SWZ( PIPE_FORMAT_COMP_R, PIPE_FORMAT_COMP_R, PIPE_FORMAT_COMP_R, PIPE_FORMAT_COMP_R )
+#define _PIPE_FORMAT_RRRG _PIPE_FORMAT_SWZ( PIPE_FORMAT_COMP_R, PIPE_FORMAT_COMP_R, PIPE_FORMAT_COMP_R, PIPE_FORMAT_COMP_G )
+#define _PIPE_FORMAT_Z000 _PIPE_FORMAT_SWZ( PIPE_FORMAT_COMP_Z, PIPE_FORMAT_COMP_0, PIPE_FORMAT_COMP_0, PIPE_FORMAT_COMP_0 )
+#define _PIPE_FORMAT_SZ00 _PIPE_FORMAT_SWZ( PIPE_FORMAT_COMP_S, PIPE_FORMAT_COMP_Z, PIPE_FORMAT_COMP_0, PIPE_FORMAT_COMP_0 )
+#define _PIPE_FORMAT_ZS00 _PIPE_FORMAT_SWZ( PIPE_FORMAT_COMP_Z, PIPE_FORMAT_COMP_S, PIPE_FORMAT_COMP_0, PIPE_FORMAT_COMP_0 )
+#define _PIPE_FORMAT_S000 _PIPE_FORMAT_SWZ( PIPE_FORMAT_COMP_S, PIPE_FORMAT_COMP_0, PIPE_FORMAT_COMP_0, PIPE_FORMAT_COMP_0 )
+
+/**
+ * YCBCR Format Layout.
+ */
+
+/**
+ * This only contains a flag that indicates whether the format is reversed or
+ * not.
+ */
+typedef uint pipe_format_ycbcr_t;
+
+/**
+ * Helper macro to encode the above structure into a 32-bit value.
+ */
+#define _PIPE_FORMAT_YCBCR( REV ) (\
+   (PIPE_FORMAT_LAYOUT_YCBCR << 0) |\
+   ((REV) << 2) )
+
+static INLINE uint pf_rev(pipe_format_ycbcr_t f)
+{
+   return (f >> 2) & 0x1;
+}
+
+/**
+ * Texture/surface image formats (preliminary)
+ */
+
+/* KW: Added lots of surface formats to support vertex element layout
+ * definitions, and eventually render-to-vertex-buffer.  Could
+ * consider making float/int/uint/scaled/normalized a separate
+ * parameter, but on the other hand there are special cases like
+ * z24s8, compressed textures, ycbcr, etc that won't fit that model.
+ */
+
+enum pipe_format {
+   PIPE_FORMAT_NONE                  = _PIPE_FORMAT_RGBAZS_1 ( _PIPE_FORMAT_0000, 0, 0, 0, 0, PIPE_FORMAT_TYPE_UNKNOWN ),
+   PIPE_FORMAT_A8R8G8B8_UNORM        = _PIPE_FORMAT_RGBAZS_8 ( _PIPE_FORMAT_ARGB, 1, 1, 1, 1, PIPE_FORMAT_TYPE_UNORM ),
+   PIPE_FORMAT_B8G8R8A8_UNORM        = _PIPE_FORMAT_RGBAZS_8 ( _PIPE_FORMAT_BGRA, 1, 1, 1, 1, PIPE_FORMAT_TYPE_UNORM ),
+   PIPE_FORMAT_A1R5G5B5_UNORM        = _PIPE_FORMAT_RGBAZS_1 ( _PIPE_FORMAT_ARGB, 1, 5, 5, 5, PIPE_FORMAT_TYPE_UNORM ),
+   PIPE_FORMAT_A4R4G4B4_UNORM        = _PIPE_FORMAT_RGBAZS_1 ( _PIPE_FORMAT_ARGB, 4, 4, 4, 4, PIPE_FORMAT_TYPE_UNORM ),
+   PIPE_FORMAT_R5G6B5_UNORM          = _PIPE_FORMAT_RGBAZS_1 ( _PIPE_FORMAT_RGB0, 5, 6, 5, 0, PIPE_FORMAT_TYPE_UNORM ),
+   PIPE_FORMAT_U_L8                  = _PIPE_FORMAT_RGBAZS_8 ( _PIPE_FORMAT_RRR1, 1, 1, 1, 1, PIPE_FORMAT_TYPE_UNORM ), /**< ubyte luminance */
+   PIPE_FORMAT_U_A8                  = _PIPE_FORMAT_RGBAZS_8 ( _PIPE_FORMAT_000R, 1, 1, 1, 1, PIPE_FORMAT_TYPE_UNORM ), /**< ubyte alpha */
+   PIPE_FORMAT_U_I8                  = _PIPE_FORMAT_RGBAZS_8 ( _PIPE_FORMAT_RRRR, 1, 1, 1, 1, PIPE_FORMAT_TYPE_UNORM ), /**< ubyte intensity */
+   PIPE_FORMAT_U_A8_L8               = _PIPE_FORMAT_RGBAZS_8 ( _PIPE_FORMAT_RRRG, 1, 1, 1, 1, PIPE_FORMAT_TYPE_UNORM ), /**< ubyte alpha, luminance */
+   PIPE_FORMAT_YCBCR                 = _PIPE_FORMAT_YCBCR( 0 ),
+   PIPE_FORMAT_YCBCR_REV             = _PIPE_FORMAT_YCBCR( 1 ),
+   PIPE_FORMAT_Z16_UNORM             = _PIPE_FORMAT_RGBAZS_8 ( _PIPE_FORMAT_Z000, 2, 0, 0, 0, PIPE_FORMAT_TYPE_UNORM ),
+   PIPE_FORMAT_Z32_UNORM             = _PIPE_FORMAT_RGBAZS_8 ( _PIPE_FORMAT_Z000, 4, 0, 0, 0, PIPE_FORMAT_TYPE_UNORM ),
+   PIPE_FORMAT_Z32_FLOAT             = _PIPE_FORMAT_RGBAZS_8 ( _PIPE_FORMAT_Z000, 4, 0, 0, 0, PIPE_FORMAT_TYPE_FLOAT ),
+   PIPE_FORMAT_S8Z24_UNORM           = _PIPE_FORMAT_RGBAZS_8 ( _PIPE_FORMAT_SZ00, 1, 3, 0, 0, PIPE_FORMAT_TYPE_UNORM ),
+   PIPE_FORMAT_Z24S8_UNORM           = _PIPE_FORMAT_RGBAZS_8 ( _PIPE_FORMAT_ZS00, 3, 1, 0, 0, PIPE_FORMAT_TYPE_UNORM ),
+   PIPE_FORMAT_S8_UNORM              = _PIPE_FORMAT_RGBAZS_8 ( _PIPE_FORMAT_S000, 1, 0, 0, 0, PIPE_FORMAT_TYPE_UNORM ), /**< ubyte stencil */
+   PIPE_FORMAT_R64_FLOAT             = _PIPE_FORMAT_RGBAZS_64( _PIPE_FORMAT_R000, 1, 0, 0, 0, PIPE_FORMAT_TYPE_FLOAT ),
+   PIPE_FORMAT_R64G64_FLOAT          = _PIPE_FORMAT_RGBAZS_64( _PIPE_FORMAT_RG00, 1, 1, 0, 0, PIPE_FORMAT_TYPE_FLOAT ),
+   PIPE_FORMAT_R64G64B64_FLOAT       = _PIPE_FORMAT_RGBAZS_64( _PIPE_FORMAT_RGB0, 1, 1, 1, 0, PIPE_FORMAT_TYPE_FLOAT ),
+   PIPE_FORMAT_R64G64B64A64_FLOAT    = _PIPE_FORMAT_RGBAZS_64( _PIPE_FORMAT_RGBA, 1, 1, 1, 1, PIPE_FORMAT_TYPE_FLOAT ),
+   PIPE_FORMAT_R32_FLOAT             = _PIPE_FORMAT_RGBAZS_8 ( _PIPE_FORMAT_R000, 4, 0, 0, 0, PIPE_FORMAT_TYPE_FLOAT ),
+   PIPE_FORMAT_R32G32_FLOAT          = _PIPE_FORMAT_RGBAZS_8 ( _PIPE_FORMAT_RG00, 4, 4, 0, 0, PIPE_FORMAT_TYPE_FLOAT ),
+   PIPE_FORMAT_R32G32B32_FLOAT       = _PIPE_FORMAT_RGBAZS_8 ( _PIPE_FORMAT_RGB0, 4, 4, 4, 0, PIPE_FORMAT_TYPE_FLOAT ),
+   PIPE_FORMAT_R32G32B32A32_FLOAT    = _PIPE_FORMAT_RGBAZS_8 ( _PIPE_FORMAT_RGBA, 4, 4, 4, 4, PIPE_FORMAT_TYPE_FLOAT ),
+   PIPE_FORMAT_R32_UNORM             = _PIPE_FORMAT_RGBAZS_8 ( _PIPE_FORMAT_R000, 4, 0, 0, 0, PIPE_FORMAT_TYPE_UNORM ),
+   PIPE_FORMAT_R32G32_UNORM          = _PIPE_FORMAT_RGBAZS_8 ( _PIPE_FORMAT_RG00, 4, 4, 0, 0, PIPE_FORMAT_TYPE_UNORM ),
+   PIPE_FORMAT_R32G32B32_UNORM       = _PIPE_FORMAT_RGBAZS_8 ( _PIPE_FORMAT_RGB0, 4, 4, 4, 0, PIPE_FORMAT_TYPE_UNORM ),
+   PIPE_FORMAT_R32G32B32A32_UNORM    = _PIPE_FORMAT_RGBAZS_8 ( _PIPE_FORMAT_RGBA, 4, 4, 4, 4, PIPE_FORMAT_TYPE_UNORM ),
+   PIPE_FORMAT_R32_USCALED           = _PIPE_FORMAT_RGBAZS_8 ( _PIPE_FORMAT_R000, 4, 0, 0, 0, PIPE_FORMAT_TYPE_USCALED ),
+   PIPE_FORMAT_R32G32_USCALED        = _PIPE_FORMAT_RGBAZS_8 ( _PIPE_FORMAT_RG00, 4, 4, 0, 0, PIPE_FORMAT_TYPE_USCALED ),
+   PIPE_FORMAT_R32G32B32_USCALED     = _PIPE_FORMAT_RGBAZS_8 ( _PIPE_FORMAT_RGB0, 4, 4, 4, 0, PIPE_FORMAT_TYPE_USCALED ),
+   PIPE_FORMAT_R32G32B32A32_USCALED  = _PIPE_FORMAT_RGBAZS_8 ( _PIPE_FORMAT_RGBA, 4, 4, 4, 4, PIPE_FORMAT_TYPE_USCALED ),
+   PIPE_FORMAT_R32_SNORM             = _PIPE_FORMAT_RGBAZS_8 ( _PIPE_FORMAT_R000, 4, 0, 0, 0, PIPE_FORMAT_TYPE_SNORM ),
+   PIPE_FORMAT_R32G32_SNORM          = _PIPE_FORMAT_RGBAZS_8 ( _PIPE_FORMAT_RG00, 4, 4, 0, 0, PIPE_FORMAT_TYPE_SNORM ),
+   PIPE_FORMAT_R32G32B32_SNORM       = _PIPE_FORMAT_RGBAZS_8 ( _PIPE_FORMAT_RGB0, 4, 4, 4, 0, PIPE_FORMAT_TYPE_SNORM ),
+   PIPE_FORMAT_R32G32B32A32_SNORM    = _PIPE_FORMAT_RGBAZS_8 ( _PIPE_FORMAT_RGBA, 4, 4, 4, 4, PIPE_FORMAT_TYPE_SNORM ),
+   PIPE_FORMAT_R32_SSCALED           = _PIPE_FORMAT_RGBAZS_8 ( _PIPE_FORMAT_R000, 4, 0, 0, 0, PIPE_FORMAT_TYPE_SSCALED ),
+   PIPE_FORMAT_R32G32_SSCALED        = _PIPE_FORMAT_RGBAZS_8 ( _PIPE_FORMAT_RG00, 4, 4, 0, 0, PIPE_FORMAT_TYPE_SSCALED ),
+   PIPE_FORMAT_R32G32B32_SSCALED     = _PIPE_FORMAT_RGBAZS_8 ( _PIPE_FORMAT_RGB0, 4, 4, 4, 0, PIPE_FORMAT_TYPE_SSCALED ),
+   PIPE_FORMAT_R32G32B32A32_SSCALED  = _PIPE_FORMAT_RGBAZS_8 ( _PIPE_FORMAT_RGBA, 4, 4, 4, 4, PIPE_FORMAT_TYPE_SSCALED ),
+   PIPE_FORMAT_R16_UNORM             = _PIPE_FORMAT_RGBAZS_8 ( _PIPE_FORMAT_R000, 2, 0, 0, 0, PIPE_FORMAT_TYPE_UNORM ),
+   PIPE_FORMAT_R16G16_UNORM          = _PIPE_FORMAT_RGBAZS_8 ( _PIPE_FORMAT_RG00, 2, 2, 0, 0, PIPE_FORMAT_TYPE_UNORM ),
+   PIPE_FORMAT_R16G16B16_UNORM       = _PIPE_FORMAT_RGBAZS_8 ( _PIPE_FORMAT_RGB0, 2, 2, 2, 0, PIPE_FORMAT_TYPE_UNORM ),
+   PIPE_FORMAT_R16G16B16A16_UNORM    = _PIPE_FORMAT_RGBAZS_8 ( _PIPE_FORMAT_RGBA, 2, 2, 2, 2, PIPE_FORMAT_TYPE_UNORM ),
+   PIPE_FORMAT_R16_USCALED           = _PIPE_FORMAT_RGBAZS_8 ( _PIPE_FORMAT_R000, 2, 0, 0, 0, PIPE_FORMAT_TYPE_USCALED ),
+   PIPE_FORMAT_R16G16_USCALED        = _PIPE_FORMAT_RGBAZS_8 ( _PIPE_FORMAT_RG00, 2, 2, 0, 0, PIPE_FORMAT_TYPE_USCALED ),
+   PIPE_FORMAT_R16G16B16_USCALED     = _PIPE_FORMAT_RGBAZS_8 ( _PIPE_FORMAT_RGB0, 2, 2, 2, 0, PIPE_FORMAT_TYPE_USCALED ),
+   PIPE_FORMAT_R16G16B16A16_USCALED  = _PIPE_FORMAT_RGBAZS_8 ( _PIPE_FORMAT_RGBA, 2, 2, 2, 2, PIPE_FORMAT_TYPE_USCALED ),
+   PIPE_FORMAT_R16_SNORM             = _PIPE_FORMAT_RGBAZS_8 ( _PIPE_FORMAT_R000, 2, 0, 0, 0, PIPE_FORMAT_TYPE_SNORM ),
+   PIPE_FORMAT_R16G16_SNORM          = _PIPE_FORMAT_RGBAZS_8 ( _PIPE_FORMAT_RG00, 2, 2, 0, 0, PIPE_FORMAT_TYPE_SNORM ),
+   PIPE_FORMAT_R16G16B16_SNORM       = _PIPE_FORMAT_RGBAZS_8 ( _PIPE_FORMAT_RGB0, 2, 2, 2, 0, PIPE_FORMAT_TYPE_SNORM ),
+   PIPE_FORMAT_R16G16B16A16_SNORM    = _PIPE_FORMAT_RGBAZS_8 ( _PIPE_FORMAT_RGBA, 2, 2, 2, 2, PIPE_FORMAT_TYPE_SNORM ),
+   PIPE_FORMAT_R16_SSCALED           = _PIPE_FORMAT_RGBAZS_8 ( _PIPE_FORMAT_R000, 2, 0, 0, 0, PIPE_FORMAT_TYPE_SSCALED ),
+   PIPE_FORMAT_R16G16_SSCALED        = _PIPE_FORMAT_RGBAZS_8 ( _PIPE_FORMAT_RG00, 2, 2, 0, 0, PIPE_FORMAT_TYPE_SSCALED ),
+   PIPE_FORMAT_R16G16B16_SSCALED     = _PIPE_FORMAT_RGBAZS_8 ( _PIPE_FORMAT_RGB0, 2, 2, 2, 0, PIPE_FORMAT_TYPE_SSCALED ),
+   PIPE_FORMAT_R16G16B16A16_SSCALED  = _PIPE_FORMAT_RGBAZS_8 ( _PIPE_FORMAT_RGBA, 2, 2, 2, 2, PIPE_FORMAT_TYPE_SSCALED ),
+   PIPE_FORMAT_R8_UNORM              = _PIPE_FORMAT_RGBAZS_8 ( _PIPE_FORMAT_R000, 1, 0, 0, 0, PIPE_FORMAT_TYPE_UNORM ),
+   PIPE_FORMAT_R8G8_UNORM            = _PIPE_FORMAT_RGBAZS_8 ( _PIPE_FORMAT_RG00, 1, 1, 0, 0, PIPE_FORMAT_TYPE_UNORM ),
+   PIPE_FORMAT_R8G8B8_UNORM          = _PIPE_FORMAT_RGBAZS_8 ( _PIPE_FORMAT_RGB0, 1, 1, 1, 0, PIPE_FORMAT_TYPE_UNORM ),
+   PIPE_FORMAT_R8G8B8A8_UNORM        = _PIPE_FORMAT_RGBAZS_8 ( _PIPE_FORMAT_RGBA, 1, 1, 1, 1, PIPE_FORMAT_TYPE_UNORM ),
+   PIPE_FORMAT_R8_USCALED            = _PIPE_FORMAT_RGBAZS_8 ( _PIPE_FORMAT_R000, 1, 0, 0, 0, PIPE_FORMAT_TYPE_USCALED ),
+   PIPE_FORMAT_R8G8_USCALED          = _PIPE_FORMAT_RGBAZS_8 ( _PIPE_FORMAT_RG00, 1, 1, 0, 0, PIPE_FORMAT_TYPE_USCALED ),
+   PIPE_FORMAT_R8G8B8_USCALED        = _PIPE_FORMAT_RGBAZS_8 ( _PIPE_FORMAT_RGB0, 1, 1, 1, 0, PIPE_FORMAT_TYPE_USCALED ),
+   PIPE_FORMAT_R8G8B8A8_USCALED      = _PIPE_FORMAT_RGBAZS_8 ( _PIPE_FORMAT_RGBA, 1, 1, 1, 1, PIPE_FORMAT_TYPE_USCALED ),
+   PIPE_FORMAT_R8_SNORM              = _PIPE_FORMAT_RGBAZS_8 ( _PIPE_FORMAT_R000, 1, 0, 0, 0, PIPE_FORMAT_TYPE_SNORM ),
+   PIPE_FORMAT_R8G8_SNORM            = _PIPE_FORMAT_RGBAZS_8 ( _PIPE_FORMAT_RG00, 1, 1, 0, 0, PIPE_FORMAT_TYPE_SNORM ),
+   PIPE_FORMAT_R8G8B8_SNORM          = _PIPE_FORMAT_RGBAZS_8 ( _PIPE_FORMAT_RGB0, 1, 1, 1, 0, PIPE_FORMAT_TYPE_SNORM ),
+   PIPE_FORMAT_R8G8B8A8_SNORM        = _PIPE_FORMAT_RGBAZS_8 ( _PIPE_FORMAT_RGBA, 1, 1, 1, 1, PIPE_FORMAT_TYPE_SNORM ),
+   PIPE_FORMAT_R8_SSCALED            = _PIPE_FORMAT_RGBAZS_8 ( _PIPE_FORMAT_R000, 1, 0, 0, 0, PIPE_FORMAT_TYPE_SSCALED ),
+   PIPE_FORMAT_R8G8_SSCALED          = _PIPE_FORMAT_RGBAZS_8 ( _PIPE_FORMAT_RG00, 1, 1, 0, 0, PIPE_FORMAT_TYPE_SSCALED ),
+   PIPE_FORMAT_R8G8B8_SSCALED        = _PIPE_FORMAT_RGBAZS_8 ( _PIPE_FORMAT_RGB0, 1, 1, 1, 0, PIPE_FORMAT_TYPE_SSCALED ),
+   PIPE_FORMAT_R8G8B8A8_SSCALED      = _PIPE_FORMAT_RGBAZS_8 ( _PIPE_FORMAT_RGBA, 1, 1, 1, 1, PIPE_FORMAT_TYPE_SSCALED )
+};
+
+
+/**
+ * XXX should remove this, but S8_UNORM is a poor name
+ */
+#define PIPE_FORMAT_U_S8                  PIPE_FORMAT_S8_UNORM
+
+
+/**
+ * Builds pipe format name from format token.
+ */
+static INLINE char *pf_sprint_name( char *str, uint format )
+{
+   strcpy( str, "PIPE_FORMAT_" );
+   switch (pf_layout( format )) {
+   case PIPE_FORMAT_LAYOUT_RGBAZS: {
+         pipe_format_rgbazs_t rgbazs = (pipe_format_rgbazs_t) format;
+         uint                 i;
+         uint                 scale = 1 << (pf_exp8( rgbazs ) * 3);
+
+         for (i = 0; i < 4; i++) {
+            uint  size = pf_size_xyzw( rgbazs, i );
+
+            if (size == 0) {
+               break;
+            }
+            switch (pf_swizzle_xyzw( rgbazs, i )) {
+            case PIPE_FORMAT_COMP_R:
+               strcat( str, "R" );
+               break;
+            case PIPE_FORMAT_COMP_G:
+               strcat( str, "G" );
+               break;
+            case PIPE_FORMAT_COMP_B:
+               strcat( str, "B" );
+               break;
+            case PIPE_FORMAT_COMP_A:
+               strcat( str, "A" );
+               break;
+            case PIPE_FORMAT_COMP_0:
+               strcat( str, "0" );
+               break;
+            case PIPE_FORMAT_COMP_1:
+               strcat( str, "1" );
+               break;
+            case PIPE_FORMAT_COMP_Z:
+               strcat( str, "Z" );
+               break;
+            case PIPE_FORMAT_COMP_S:
+               strcat( str, "S" );
+               break;
+            }
+            sprintf( &str[strlen( str )], "%u", size * scale );
+         }
+         if (i != 0) {
+            strcat( str, "_" );
+         }
+         switch (pf_type( rgbazs )) {
+         case PIPE_FORMAT_TYPE_UNKNOWN:
+            strcat( str, "NONE" );
+            break;
+         case PIPE_FORMAT_TYPE_FLOAT:
+            strcat( str, "FLOAT" );
+            break;
+         case PIPE_FORMAT_TYPE_UNORM:
+            strcat( str, "UNORM" );
+            break;
+         case PIPE_FORMAT_TYPE_SNORM:
+            strcat( str, "SNORM" );
+            break;
+         case PIPE_FORMAT_TYPE_USCALED:
+            strcat( str, "USCALED" );
+            break;
+         case PIPE_FORMAT_TYPE_SSCALED:
+            strcat( str, "SSCALED" );
+            break;
+         }
+      }
+      break;
+   case PIPE_FORMAT_LAYOUT_YCBCR: {
+         pipe_format_ycbcr_t  ycbcr = (pipe_format_ycbcr_t) format;
+
+         strcat( str, "YCBCR" );
+         if (pf_rev( ycbcr )) {
+            strcat( str, "_REV" );
+         }
+      }
+      break;
+   }
+   return str;
+}
+
+static INLINE uint pf_get_component_bits( enum pipe_format format, uint comp )
+{
+   uint size;
+
+   if (pf_swizzle_x(format) == comp) {
+      size = pf_size_x(format);
+   }
+   else if (pf_swizzle_y(format) == comp) {
+      size = pf_size_y(format);
+   }
+   else if (pf_swizzle_z(format) == comp) {
+      size = pf_size_z(format);
+   }
+   else if (pf_swizzle_w(format) == comp) {
+      size = pf_size_w(format);
+   }
+   else {
+      size = 0;
+   }
+   return size << (pf_exp8(format) * 3);
+}
+
+static INLINE uint pf_get_bits( enum pipe_format format )
+{
+   if (pf_layout(format) == PIPE_FORMAT_LAYOUT_RGBAZS) {
+      return
+         pf_get_component_bits( format, PIPE_FORMAT_COMP_R ) +
+         pf_get_component_bits( format, PIPE_FORMAT_COMP_G ) +
+         pf_get_component_bits( format, PIPE_FORMAT_COMP_B ) +
+         pf_get_component_bits( format, PIPE_FORMAT_COMP_A ) +
+         pf_get_component_bits( format, PIPE_FORMAT_COMP_Z ) +
+         pf_get_component_bits( format, PIPE_FORMAT_COMP_S );
+   }
+   else {
+      assert( pf_layout(format) == PIPE_FORMAT_LAYOUT_YCBCR );
+
+      /* TODO */
+      assert( 0 );
+      return 0;
+   }
+}
+
+static INLINE uint pf_get_size( enum pipe_format format ) {
+   assert(pf_get_bits(format) % 8 == 0);
+   return pf_get_bits(format) / 8;
+}
+
+#endif
diff --git a/src/gallium/include/pipe/p_inlines.h b/src/gallium/include/pipe/p_inlines.h
new file mode 100644
index 0000000000..ebf6ed86bc
--- /dev/null
+++ b/src/gallium/include/pipe/p_inlines.h
@@ -0,0 +1,112 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+#ifndef P_INLINES_H
+#define P_INLINES_H
+
+#include "p_context.h"
+#include "p_defines.h"
+#include "p_winsys.h"
+
+
+static INLINE void *
+pipe_surface_map(struct pipe_surface *surface)
+{
+   return (char *)surface->winsys->buffer_map( surface->winsys, surface->buffer,
+					       PIPE_BUFFER_USAGE_CPU_WRITE |
+					       PIPE_BUFFER_USAGE_CPU_READ )
+      + surface->offset;
+}
+
+static INLINE void
+pipe_surface_unmap(struct pipe_surface *surface)
+{
+   surface->winsys->buffer_unmap( surface->winsys, surface->buffer );
+}
+
+/**
+ * Set 'ptr' to point to 'surf' and update reference counting.
+ * The old thing pointed to, if any, will be unreferenced first.
+ * 'surf' may be NULL.
+ */
+static INLINE void
+pipe_surface_reference(struct pipe_surface **ptr, struct pipe_surface *surf)
+{
+   /* bump the refcount first */
+   if (surf) 
+      surf->refcount++;
+
+   if (*ptr /* && --(*ptr)->refcount == 0 */) {
+      struct pipe_winsys *winsys = (*ptr)->winsys;
+      winsys->surface_release(winsys, ptr);
+      assert(!*ptr);
+   }
+
+   *ptr = surf;
+}
+
+
+/* XXX: thread safety issues!
+ */
+static INLINE void
+pipe_buffer_reference(struct pipe_winsys *winsys,
+		      struct pipe_buffer **ptr,
+		      struct pipe_buffer *buf)
+{
+   if (buf) 
+      buf->refcount++;
+
+   if (*ptr && --(*ptr)->refcount == 0)
+      winsys->buffer_destroy( winsys, *ptr );
+
+   *ptr = buf;
+}
+
+
+
+/**
+ * \sa pipe_surface_reference
+ */
+static INLINE void
+pipe_texture_reference(struct pipe_context *pipe, struct pipe_texture **ptr,
+		       struct pipe_texture *pt)
+{
+   assert(ptr);
+
+   if (pt) 
+      pt->refcount++;
+
+   if (*ptr) {
+      pipe->texture_release(pipe, ptr);
+      assert(!*ptr);
+   }
+
+   *ptr = pt;
+}
+
+
+#endif /* P_INLINES_H */
diff --git a/src/gallium/include/pipe/p_shader_tokens.h b/src/gallium/include/pipe/p_shader_tokens.h
new file mode 100644
index 0000000000..3ce35310f6
--- /dev/null
+++ b/src/gallium/include/pipe/p_shader_tokens.h
@@ -0,0 +1,806 @@
+#if !defined TGSI_TOKEN_H
+#define TGSI_TOKEN_H
+
+#if defined __cplusplus
+extern "C" {
+#endif // defined __cplusplus
+
+struct tgsi_version
+{
+   unsigned MajorVersion  : 8;
+   unsigned MinorVersion  : 8;
+   unsigned Padding       : 16;
+};
+
+struct tgsi_header
+{
+   unsigned HeaderSize : 8;
+   unsigned BodySize   : 24;
+};
+
+#define TGSI_PROCESSOR_FRAGMENT  0
+#define TGSI_PROCESSOR_VERTEX    1
+#define TGSI_PROCESSOR_GEOMETRY  2
+
+struct tgsi_processor
+{
+   unsigned Processor  : 4;  /* TGSI_PROCESSOR_ */
+   unsigned Padding    : 28;
+};
+
+#define TGSI_TOKEN_TYPE_DECLARATION    0
+#define TGSI_TOKEN_TYPE_IMMEDIATE      1
+#define TGSI_TOKEN_TYPE_INSTRUCTION    2
+
+struct tgsi_token
+{
+   unsigned Type       : 4;  /* TGSI_TOKEN_TYPE_ */
+   unsigned Size       : 8;  /* UINT */
+   unsigned Padding    : 19;
+   unsigned Extended   : 1;  /* BOOL */
+};
+
+#define TGSI_FILE_NULL        0
+#define TGSI_FILE_CONSTANT    1
+#define TGSI_FILE_INPUT       2
+#define TGSI_FILE_OUTPUT      3
+#define TGSI_FILE_TEMPORARY   4
+#define TGSI_FILE_SAMPLER     5
+#define TGSI_FILE_ADDRESS     6
+#define TGSI_FILE_IMMEDIATE   7
+
+#define TGSI_DECLARE_RANGE    0
+#define TGSI_DECLARE_MASK     1
+
+#define TGSI_WRITEMASK_NONE     0x00
+#define TGSI_WRITEMASK_X        0x01
+#define TGSI_WRITEMASK_Y        0x02
+#define TGSI_WRITEMASK_XY       0x03
+#define TGSI_WRITEMASK_Z        0x04
+#define TGSI_WRITEMASK_XZ       0x05
+#define TGSI_WRITEMASK_YZ       0x06
+#define TGSI_WRITEMASK_XYZ      0x07
+#define TGSI_WRITEMASK_W        0x08
+#define TGSI_WRITEMASK_XW       0x09
+#define TGSI_WRITEMASK_YW       0x0A
+#define TGSI_WRITEMASK_XYW      0x0B
+#define TGSI_WRITEMASK_ZW       0x0C
+#define TGSI_WRITEMASK_XZW      0x0D
+#define TGSI_WRITEMASK_YZW      0x0E
+#define TGSI_WRITEMASK_XYZW     0x0F
+
+struct tgsi_declaration
+{
+   unsigned Type        : 4;  /* TGSI_TOKEN_TYPE_DECLARATION */
+   unsigned Size        : 8;  /* UINT */
+   unsigned File        : 4;  /* one of TGSI_FILE_x */
+   unsigned Declare     : 4;  /* one of TGSI_DECLARE_x */
+   unsigned UsageMask   : 4;  /* bitmask of TGSI_WRITEMASK_x flags */
+   unsigned Interpolate : 1;  /* BOOL, any interpolation info? */
+   unsigned Semantic    : 1;  /* BOOL, any semantic info? */
+   unsigned Padding     : 5;
+   unsigned Extended    : 1;  /* BOOL */
+};
+
+struct tgsi_declaration_range
+{
+   unsigned First   : 16; /* UINT */
+   unsigned Last    : 16; /* UINT */
+};
+
+struct tgsi_declaration_mask
+{
+   unsigned Mask : 32; /* UINT */
+};
+
+#define TGSI_INTERPOLATE_CONSTANT      0
+#define TGSI_INTERPOLATE_LINEAR        1
+#define TGSI_INTERPOLATE_PERSPECTIVE   2
+
+struct tgsi_declaration_interpolation
+{
+   unsigned Interpolate   : 4;  /* TGSI_INTERPOLATE_ */
+   unsigned Padding       : 28;
+};
+
+#define TGSI_SEMANTIC_POSITION 0
+#define TGSI_SEMANTIC_COLOR    1
+#define TGSI_SEMANTIC_BCOLOR   2 /**< back-face color */
+#define TGSI_SEMANTIC_FOG      3
+#define TGSI_SEMANTIC_PSIZE    4
+#define TGSI_SEMANTIC_GENERIC  5
+#define TGSI_SEMANTIC_NORMAL   6
+#define TGSI_SEMANTIC_COUNT    7 /**< number of semantic values */
+
+struct tgsi_declaration_semantic
+{
+   unsigned SemanticName   : 8;  /* one of TGSI_SEMANTIC_ */
+   unsigned SemanticIndex  : 16; /* UINT */
+   unsigned Padding        : 8;
+};
+
+#define TGSI_IMM_FLOAT32   0
+
+struct tgsi_immediate
+{
+   unsigned Type       : 4;  /* TGSI_TOKEN_TYPE_IMMEDIATE */
+   unsigned Size       : 8;  /* UINT */
+   unsigned DataType   : 4;  /* TGSI_IMM_ */
+   unsigned Padding    : 15;
+   unsigned Extended   : 1;  /* BOOL */
+};
+
+struct tgsi_immediate_float32
+{
+   float Float;
+};
+
+/*
+ * GL_NV_vertex_program
+ */
+#define TGSI_OPCODE_ARL                 0
+#define TGSI_OPCODE_MOV                 1
+#define TGSI_OPCODE_LIT                 2
+#define TGSI_OPCODE_RCP                 3
+#define TGSI_OPCODE_RSQ                 4
+#define TGSI_OPCODE_EXP                 5
+#define TGSI_OPCODE_LOG                 6
+#define TGSI_OPCODE_MUL                 7
+#define TGSI_OPCODE_ADD                 8
+#define TGSI_OPCODE_DP3                 9
+#define TGSI_OPCODE_DP4                 10
+#define TGSI_OPCODE_DST                 11
+#define TGSI_OPCODE_MIN                 12
+#define TGSI_OPCODE_MAX                 13
+#define TGSI_OPCODE_SLT                 14
+#define TGSI_OPCODE_SGE                 15
+#define TGSI_OPCODE_MAD                 16
+
+/*
+ * GL_ATI_fragment_shader
+ */
+#define TGSI_OPCODE_SUB                 17
+#define TGSI_OPCODE_DOT3                TGSI_OPCODE_DP3
+#define TGSI_OPCODE_DOT4                TGSI_OPCODE_DP4
+#define TGSI_OPCODE_LERP                18
+#define TGSI_OPCODE_CND                 19
+#define TGSI_OPCODE_CND0                20
+#define TGSI_OPCODE_DOT2ADD             21
+
+/*
+ * GL_EXT_vertex_shader
+ */
+#define TGSI_OPCODE_INDEX               22
+#define TGSI_OPCODE_NEGATE              23
+#define TGSI_OPCODE_MADD                TGSI_OPCODE_MAD
+#define TGSI_OPCODE_FRAC                24
+#define TGSI_OPCODE_SETGE               TGSI_OPCODE_SGE
+#define TGSI_OPCODE_SETLT               TGSI_OPCODE_SLT
+#define TGSI_OPCODE_CLAMP               25
+#define TGSI_OPCODE_FLOOR               26
+#define TGSI_OPCODE_ROUND               27
+#define TGSI_OPCODE_EXPBASE2            28
+#define TGSI_OPCODE_LOGBASE2            29
+#define TGSI_OPCODE_POWER               30
+#define TGSI_OPCODE_RECIP               TGSI_OPCODE_RCP
+#define TGSI_OPCODE_RECIPSQRT           TGSI_OPCODE_RSQ
+#define TGSI_OPCODE_CROSSPRODUCT        31
+#define TGSI_OPCODE_MULTIPLYMATRIX      32
+
+/*
+ * GL_NV_vertex_program1_1
+ */
+#define TGSI_OPCODE_ABS                 33
+#define TGSI_OPCODE_RCC                 34
+#define TGSI_OPCODE_DPH                 35
+
+/*
+ * GL_NV_fragment_program
+ */
+#define TGSI_OPCODE_COS                 36
+#define TGSI_OPCODE_DDX                 37
+#define TGSI_OPCODE_DDY                 38
+#define TGSI_OPCODE_EX2                 TGSI_OPCODE_EXPBASE2
+#define TGSI_OPCODE_FLR                 TGSI_OPCODE_FLOOR
+#define TGSI_OPCODE_FRC                 TGSI_OPCODE_FRAC
+#define TGSI_OPCODE_KILP                39  /* predicated kill */
+#define TGSI_OPCODE_LG2                 TGSI_OPCODE_LOGBASE2
+#define TGSI_OPCODE_LRP                 TGSI_OPCODE_LERP
+#define TGSI_OPCODE_PK2H                40
+#define TGSI_OPCODE_PK2US               41
+#define TGSI_OPCODE_PK4B                42
+#define TGSI_OPCODE_PK4UB               43
+#define TGSI_OPCODE_POW                 TGSI_OPCODE_POWER
+#define TGSI_OPCODE_RFL                 44
+#define TGSI_OPCODE_SEQ                 45
+#define TGSI_OPCODE_SFL                 46
+#define TGSI_OPCODE_SGT                 47
+#define TGSI_OPCODE_SIN                 48
+#define TGSI_OPCODE_SLE                 49
+#define TGSI_OPCODE_SNE                 50
+#define TGSI_OPCODE_STR                 51
+#define TGSI_OPCODE_TEX                 52
+#define TGSI_OPCODE_TXD                 53
+#define TGSI_OPCODE_UP2H                54
+#define TGSI_OPCODE_UP2US               55
+#define TGSI_OPCODE_UP4B                56
+#define TGSI_OPCODE_UP4UB               57
+#define TGSI_OPCODE_X2D                 58
+
+/*
+ * GL_NV_vertex_program2
+ */
+#define TGSI_OPCODE_ARA                 59
+#define TGSI_OPCODE_ARR                 60
+#define TGSI_OPCODE_BRA                 61
+#define TGSI_OPCODE_CAL                 62
+#define TGSI_OPCODE_RET                 63
+#define TGSI_OPCODE_SSG                 64
+
+/*
+ * GL_ARB_vertex_program
+ */
+#define TGSI_OPCODE_SWZ                 TGSI_OPCODE_MOV
+#define TGSI_OPCODE_XPD                 TGSI_OPCODE_CROSSPRODUCT
+
+/*
+ * GL_ARB_fragment_program
+ */
+#define TGSI_OPCODE_CMP                 65
+#define TGSI_OPCODE_SCS                 66
+#define TGSI_OPCODE_TXB                 67
+
+/*
+ * GL_NV_fragment_program_option
+ */
+/* No new opcode */
+
+/*
+ * GL_NV_fragment_program2
+ */
+#define TGSI_OPCODE_NRM                 68
+#define TGSI_OPCODE_DIV                 69
+#define TGSI_OPCODE_DP2                 70
+#define TGSI_OPCODE_DP2A                TGSI_OPCODE_DOT2ADD
+#define TGSI_OPCODE_TXL                 71
+#define TGSI_OPCODE_BRK                 72
+#define TGSI_OPCODE_IF                  73
+#define TGSI_OPCODE_LOOP                74
+#define TGSI_OPCODE_REP                 75
+#define TGSI_OPCODE_ELSE                76
+#define TGSI_OPCODE_ENDIF               77
+#define TGSI_OPCODE_ENDLOOP             78
+#define TGSI_OPCODE_ENDREP              79
+
+/*
+ * GL_NV_vertex_program2_option
+ */
+
+/*
+ * GL_NV_vertex_program3
+ */
+#define TGSI_OPCODE_PUSHA               80
+#define TGSI_OPCODE_POPA                81
+
+/*
+ * GL_NV_gpu_program4
+ */
+#define TGSI_OPCODE_CEIL                82
+#define TGSI_OPCODE_I2F                 83
+#define TGSI_OPCODE_NOT                 84
+#define TGSI_OPCODE_TRUNC               85
+#define TGSI_OPCODE_SHL                 86
+#define TGSI_OPCODE_SHR                 87
+#define TGSI_OPCODE_AND                 88
+#define TGSI_OPCODE_OR                  89
+#define TGSI_OPCODE_MOD                 90
+#define TGSI_OPCODE_XOR                 91
+#define TGSI_OPCODE_SAD                 92
+#define TGSI_OPCODE_TXF                 93
+#define TGSI_OPCODE_TXQ                 94
+#define TGSI_OPCODE_CONT                95
+
+/*
+ * GL_NV_vertex_program4
+ */
+/* Same as GL_NV_gpu_program4 */
+
+/*
+ * GL_NV_fragment_program4
+ */
+/* Same as GL_NV_gpu_program4 */
+
+/*
+ * GL_NV_geometry_program4
+ */
+/* Same as GL_NV_gpu_program4 */
+#define TGSI_OPCODE_EMIT                96
+#define TGSI_OPCODE_ENDPRIM             97
+
+/*
+ * GLSL
+ */
+#define TGSI_OPCODE_BGNLOOP2            98
+#define TGSI_OPCODE_BGNSUB              99
+#define TGSI_OPCODE_ENDLOOP2            100
+#define TGSI_OPCODE_ENDSUB              101
+#define TGSI_OPCODE_INT                 TGSI_OPCODE_TRUNC
+#define TGSI_OPCODE_NOISE1              102
+#define TGSI_OPCODE_NOISE2              103
+#define TGSI_OPCODE_NOISE3              104
+#define TGSI_OPCODE_NOISE4              105
+#define TGSI_OPCODE_NOP                 106
+
+/*
+ * ps_1_1
+ */
+#define TGSI_OPCODE_TEXCOORD            TGSI_OPCODE_NOP
+#define TGSI_OPCODE_TEXKILL             TGSI_OPCODE_KIL
+#define TGSI_OPCODE_TEXBEM              107
+#define TGSI_OPCODE_TEXBEML             108
+#define TGSI_OPCODE_TEXREG2AR           109
+#define TGSI_OPCODE_TEXM3X2PAD          110
+#define TGSI_OPCODE_TEXM3X2TEX          111
+#define TGSI_OPCODE_TEXM3X3PAD          112
+#define TGSI_OPCODE_TEXM3X3TEX          113
+#define TGSI_OPCODE_TEXM3X3SPEC         114
+#define TGSI_OPCODE_TEXM3X3VSPEC        115
+
+/*
+ * ps_1_2
+ */
+#define TGSI_OPCODE_TEXREG2GB           116
+#define TGSI_OPCODE_TEXREG2RGB          117
+#define TGSI_OPCODE_TEXDP3TEX           118
+#define TGSI_OPCODE_TEXDP3              119
+#define TGSI_OPCODE_TEXM3X3             120
+/* CMP - use TGSI_OPCODE_CND0 */
+
+/*
+ * ps_1_3
+ */
+#define TGSI_OPCODE_TEXM3X2DEPTH        121
+/* CMP - use TGSI_OPCODE_CND0 */
+
+/*
+ * ps_1_4
+ */
+#define TGSI_OPCODE_TEXCRD              TGSI_OPCODE_TEXCOORD
+#define TGSI_OPCODE_TEXLD               TGSI_OPCODE_TEX
+#define TGSI_OPCODE_TEXDEPTH            122
+#define TGSI_OPCODE_BEM                 123
+
+/*
+ * ps_2_0
+ */
+#define TGSI_OPCODE_M4X4                TGSI_OPCODE_MULTIPLYMATRIX
+#define TGSI_OPCODE_M4X3                124
+#define TGSI_OPCODE_M3X4                125
+#define TGSI_OPCODE_M3X3                126
+#define TGSI_OPCODE_M3X2                127
+#define TGSI_OPCODE_CRS                 TGSI_OPCODE_XPD
+#define TGSI_OPCODE_NRM4                128
+#define TGSI_OPCODE_SINCOS              TGSI_OPCODE_SCS
+#define TGSI_OPCODE_TEXLDB              TGSI_OPCODE_TXB
+#define TGSI_OPCODE_DP2ADD              TGSI_OPCODE_DP2A
+
+/*
+ * ps_2_x
+ */
+#define TGSI_OPCODE_CALL                TGSI_OPCODE_CAL
+#define TGSI_OPCODE_CALLNZ              129
+#define TGSI_OPCODE_IFC                 130
+#define TGSI_OPCODE_BREAK               TGSI_OPCODE_BRK
+#define TGSI_OPCODE_BREAKC              131
+#define TGSI_OPCODE_DSX                 TGSI_OPCODE_DDX
+#define TGSI_OPCODE_DSY                 TGSI_OPCODE_DDY
+#define TGSI_OPCODE_TEXLDD              TGSI_OPCODE_TXD
+
+/*
+ * vs_1_1
+ */
+#define TGSI_OPCODE_EXPP                TGSI_OPCODE_EXP
+#define TGSI_OPCODE_LOGP                TGSI_OPCODE_LG2
+
+/*
+ * vs_2_0
+ */
+#define TGSI_OPCODE_SGN                 TGSI_OPCODE_SSG
+#define TGSI_OPCODE_MOVA                TGSI_OPCODE_ARR
+
+/*
+ * vs_2_x
+ */
+
+#define TGSI_OPCODE_KIL                 132  /* unpredicated kill */
+#define TGSI_OPCODE_END                 133  /* aka HALT */
+
+#define TGSI_OPCODE_LAST                134
+
+#define TGSI_SAT_NONE            0  /* do not saturate */
+#define TGSI_SAT_ZERO_ONE        1  /* clamp to [0,1] */
+#define TGSI_SAT_MINUS_PLUS_ONE  2  /* clamp to [-1,1] */
+
+/*
+ * Opcode is the operation code to execute. A given operation defines the
+ * semantics how the source registers (if any) are interpreted and what is
+ * written to the destination registers (if any) as a result of execution.
+ *
+ * NumDstRegs and NumSrcRegs is the number of destination and source registers,
+ * respectively. For a given operation code, those numbers are fixed and are
+ * present here only for convenience.
+ *
+ * If Extended is TRUE, it is now executed.
+ *
+ * Saturate controls how are final results in destination registers modified.
+ */
+
+struct tgsi_instruction
+{
+   unsigned Type       : 4;  /* TGSI_TOKEN_TYPE_INSTRUCTION */
+   unsigned Size       : 8;  /* UINT */
+   unsigned Opcode     : 8;  /* TGSI_OPCODE_ */
+   unsigned Saturate   : 2;  /* TGSI_SAT_ */
+   unsigned NumDstRegs : 2;  /* UINT */
+   unsigned NumSrcRegs : 4;  /* UINT */
+   unsigned Padding    : 3;
+   unsigned Extended   : 1;  /* BOOL */
+};
+
+/*
+ * If tgsi_instruction::Extended is TRUE, tgsi_instruction_ext follows.
+ * 
+ * Then, tgsi_instruction::NumDstRegs of tgsi_dst_register follow.
+ * 
+ * Then, tgsi_instruction::NumSrcRegs of tgsi_src_register follow.
+ *
+ * tgsi_instruction::Size contains the total number of words that make the
+ * instruction, including the instruction word.
+ */
+
+#define TGSI_INSTRUCTION_EXT_TYPE_NV        0
+#define TGSI_INSTRUCTION_EXT_TYPE_LABEL     1
+#define TGSI_INSTRUCTION_EXT_TYPE_TEXTURE   2
+#define TGSI_INSTRUCTION_EXT_TYPE_PREDICATE 3
+
+struct tgsi_instruction_ext
+{
+   unsigned Type       : 4;  /* TGSI_INSTRUCTION_EXT_TYPE_ */
+   unsigned Padding    : 27;
+   unsigned Extended   : 1;  /* BOOL */
+};
+
+/*
+ * If tgsi_instruction_ext::Type is TGSI_INSTRUCTION_EXT_TYPE_NV, it should
+ * be cast to tgsi_instruction_ext_nv.
+ * 
+ * If tgsi_instruction_ext::Type is TGSI_INSTRUCTION_EXT_TYPE_LABEL, it
+ * should be cast to tgsi_instruction_ext_label.
+ * 
+ * If tgsi_instruction_ext::Type is TGSI_INSTRUCTION_EXT_TYPE_TEXTURE, it
+ * should be cast to tgsi_instruction_ext_texture.
+ * 
+ * If tgsi_instruction_ext::Type is TGSI_INSTRUCTION_EXT_TYPE_PREDICATE, it
+ * should be cast to tgsi_instruction_ext_predicate.
+ * 
+ * If tgsi_instruction_ext::Extended is TRUE, another tgsi_instruction_ext
+ * follows.
+ */
+
+#define TGSI_PRECISION_DEFAULT      0
+#define TGSI_PRECISION_FLOAT32      1
+#define TGSI_PRECISION_FLOAT16      2
+#define TGSI_PRECISION_FIXED12      3
+
+#define TGSI_CC_GT      0
+#define TGSI_CC_EQ      1
+#define TGSI_CC_LT      2
+#define TGSI_CC_UN      3
+#define TGSI_CC_GE      4
+#define TGSI_CC_LE      5
+#define TGSI_CC_NE      6
+#define TGSI_CC_TR      7
+#define TGSI_CC_FL      8
+
+#define TGSI_SWIZZLE_X      0
+#define TGSI_SWIZZLE_Y      1
+#define TGSI_SWIZZLE_Z      2
+#define TGSI_SWIZZLE_W      3
+
+/*
+ * Precision controls the precision at which the operation should be executed.
+ *
+ * CondDstUpdate enables condition code register writes. When this field is
+ * TRUE, CondDstIndex specifies the index of the condition code register to
+ * update.
+ *
+ * CondFlowEnable enables conditional execution of the operation. When this
+ * field is TRUE, CondFlowIndex specifies the index of the condition code
+ * register to test against CondMask with component swizzle controled by
+ * CondSwizzleX, CondSwizzleY, CondSwizzleZ and CondSwizzleW. If the test fails,
+ * the operation is not executed.
+ */
+
+struct tgsi_instruction_ext_nv
+{
+   unsigned Type             : 4;    /* TGSI_INSTRUCTION_EXT_TYPE_NV */
+   unsigned Precision        : 4;    /* TGSI_PRECISION_ */
+   unsigned CondDstIndex     : 4;    /* UINT */
+   unsigned CondFlowIndex    : 4;    /* UINT */
+   unsigned CondMask         : 4;    /* TGSI_CC_ */
+   unsigned CondSwizzleX     : 2;    /* TGSI_SWIZZLE_ */
+   unsigned CondSwizzleY     : 2;    /* TGSI_SWIZZLE_ */
+   unsigned CondSwizzleZ     : 2;    /* TGSI_SWIZZLE_ */
+   unsigned CondSwizzleW     : 2;    /* TGSI_SWIZZLE_ */
+   unsigned CondDstUpdate    : 1;    /* BOOL */
+   unsigned CondFlowEnable   : 1;    /* BOOL */
+   unsigned Padding          : 1;
+   unsigned Extended         : 1;    /* BOOL */
+};
+
+struct tgsi_instruction_ext_label
+{
+   unsigned Type     : 4;    /* TGSI_INSTRUCTION_EXT_TYPE_LABEL */
+   unsigned Label    : 24;   /* UINT */
+   unsigned Padding  : 3;
+   unsigned Extended : 1;    /* BOOL */
+};
+
+#define TGSI_TEXTURE_UNKNOWN        0
+#define TGSI_TEXTURE_1D             1
+#define TGSI_TEXTURE_2D             2
+#define TGSI_TEXTURE_3D             3
+#define TGSI_TEXTURE_CUBE           4
+#define TGSI_TEXTURE_RECT           5
+#define TGSI_TEXTURE_SHADOW1D       6
+#define TGSI_TEXTURE_SHADOW2D       7
+#define TGSI_TEXTURE_SHADOWRECT     8
+
+struct tgsi_instruction_ext_texture
+{
+   unsigned Type     : 4;    /* TGSI_INSTRUCTION_EXT_TYPE_TEXTURE */
+   unsigned Texture  : 8;    /* TGSI_TEXTURE_ */
+   unsigned Padding  : 19;
+   unsigned Extended : 1;    /* BOOL */
+};
+
+struct tgsi_instruction_ext_predicate
+{
+   unsigned Type             : 4;    /* TGSI_INSTRUCTION_EXT_TYPE_PREDICATE */
+   unsigned PredDstIndex     : 4;    /* UINT */
+   unsigned PredWriteMask    : 4;    /* TGSI_WRITEMASK_ */
+   unsigned Padding          : 19;
+   unsigned Extended         : 1;    /* BOOL */
+};
+
+/*
+ * File specifies the register array to access.
+ *
+ * Index specifies the element number of a register in the register file.
+ *
+ * If Indirect is TRUE, Index should be offset by the X component of a source
+ * register that follows. The register can be now fetched into local storage
+ * for further processing.
+ *
+ * If Negate is TRUE, all components of the fetched register are negated.
+ *
+ * The fetched register components are swizzled according to SwizzleX, SwizzleY,
+ * SwizzleZ and SwizzleW.
+ *
+ * If Extended is TRUE, any further modifications to the source register are
+ * made to this temporary storage.
+ */
+
+struct tgsi_src_register
+{
+   unsigned File        : 4;  /* TGSI_FILE_ */
+   unsigned SwizzleX    : 2;  /* TGSI_SWIZZLE_ */
+   unsigned SwizzleY    : 2;  /* TGSI_SWIZZLE_ */
+   unsigned SwizzleZ    : 2;  /* TGSI_SWIZZLE_ */
+   unsigned SwizzleW    : 2;  /* TGSI_SWIZZLE_ */
+   unsigned Negate      : 1;  /* BOOL */
+   unsigned Indirect    : 1;  /* BOOL */
+   unsigned Dimension   : 1;  /* BOOL */
+   int      Index       : 16; /* SINT */
+   unsigned Extended    : 1;  /* BOOL */
+};
+
+/*
+ * If tgsi_src_register::Extended is TRUE, tgsi_src_register_ext follows.
+ * 
+ * Then, if tgsi_src_register::Indirect is TRUE, another tgsi_src_register
+ * follows.
+ *
+ * Then, if tgsi_src_register::Dimension is TRUE, tgsi_dimension follows.
+ */
+
+#define TGSI_SRC_REGISTER_EXT_TYPE_SWZ      0
+#define TGSI_SRC_REGISTER_EXT_TYPE_MOD      1
+
+struct tgsi_src_register_ext
+{
+   unsigned Type     : 4;    /* TGSI_SRC_REGISTER_EXT_TYPE_ */
+   unsigned Padding  : 27;
+   unsigned Extended : 1;    /* BOOL */
+};
+
+/*
+ * If tgsi_src_register_ext::Type is TGSI_SRC_REGISTER_EXT_TYPE_SWZ,
+ * it should be cast to tgsi_src_register_ext_swz.
+ * 
+ * If tgsi_src_register_ext::Type is TGSI_SRC_REGISTER_EXT_TYPE_MOD,
+ * it should be cast to tgsi_src_register_ext_mod.
+ * 
+ * If tgsi_dst_register_ext::Extended is TRUE, another tgsi_dst_register_ext
+ * follows.
+ */
+
+#define TGSI_EXTSWIZZLE_X       TGSI_SWIZZLE_X
+#define TGSI_EXTSWIZZLE_Y       TGSI_SWIZZLE_Y
+#define TGSI_EXTSWIZZLE_Z       TGSI_SWIZZLE_Z
+#define TGSI_EXTSWIZZLE_W       TGSI_SWIZZLE_W
+#define TGSI_EXTSWIZZLE_ZERO    4
+#define TGSI_EXTSWIZZLE_ONE     5
+
+/*
+ * ExtSwizzleX, ExtSwizzleY, ExtSwizzleZ and ExtSwizzleW swizzle the source
+ * register in an extended manner.
+ *
+ * NegateX, NegateY, NegateZ and NegateW negate individual components of the
+ * source register.
+ *
+ * ExtDivide specifies which component is used to divide all components of the
+ * source register.
+ */
+
+struct tgsi_src_register_ext_swz
+{
+   unsigned Type         : 4;    /* TGSI_SRC_REGISTER_EXT_TYPE_SWZ */
+   unsigned ExtSwizzleX  : 4;    /* TGSI_EXTSWIZZLE_ */
+   unsigned ExtSwizzleY  : 4;    /* TGSI_EXTSWIZZLE_ */
+   unsigned ExtSwizzleZ  : 4;    /* TGSI_EXTSWIZZLE_ */
+   unsigned ExtSwizzleW  : 4;    /* TGSI_EXTSWIZZLE_ */
+   unsigned NegateX      : 1;    /* BOOL */
+   unsigned NegateY      : 1;    /* BOOL */
+   unsigned NegateZ      : 1;    /* BOOL */
+   unsigned NegateW      : 1;    /* BOOL */
+   unsigned ExtDivide    : 4;    /* TGSI_EXTSWIZZLE_ */
+   unsigned Padding      : 3;
+   unsigned Extended     : 1;    /* BOOL */
+};
+
+/**
+ * Extra src register modifiers
+ *
+ * If Complement is TRUE, the source register is modified by subtracting it
+ * from 1.0.
+ *
+ * If Bias is TRUE, the source register is modified by subtracting 0.5 from it.
+ *
+ * If Scale2X is TRUE, the source register is modified by multiplying it by 2.0.
+ *
+ * If Absolute is TRUE, the source register is modified by removing the sign.
+ *
+ * If Negate is TRUE, the source register is modified by negating it.
+ */
+
+struct tgsi_src_register_ext_mod
+{
+   unsigned Type         : 4;    /* TGSI_SRC_REGISTER_EXT_TYPE_MOD */
+   unsigned Complement   : 1;    /* BOOL */
+   unsigned Bias         : 1;    /* BOOL */
+   unsigned Scale2X      : 1;    /* BOOL */
+   unsigned Absolute     : 1;    /* BOOL */
+   unsigned Negate       : 1;    /* BOOL */
+   unsigned Padding      : 22;
+   unsigned Extended     : 1;    /* BOOL */
+};
+
+struct tgsi_dimension
+{
+   unsigned Indirect    : 1;  /* BOOL */
+   unsigned Dimension   : 1;  /* BOOL */
+   unsigned Padding     : 13;
+   int      Index       : 16; /* SINT */
+   unsigned Extended    : 1;  /* BOOL */
+};
+
+struct tgsi_dst_register
+{
+   unsigned File        : 4;  /* TGSI_FILE_ */
+   unsigned WriteMask   : 4;  /* TGSI_WRITEMASK_ */
+   unsigned Indirect    : 1;  /* BOOL */
+   unsigned Dimension   : 1;  /* BOOL */
+   int      Index       : 16; /* SINT */
+   unsigned Padding     : 5;
+   unsigned Extended    : 1;  /* BOOL */
+};
+
+/*
+ * If tgsi_dst_register::Extended is TRUE, tgsi_dst_register_ext follows.
+ * 
+ * Then, if tgsi_dst_register::Indirect is TRUE, tgsi_src_register follows.
+ */
+
+#define TGSI_DST_REGISTER_EXT_TYPE_CONDCODE     0
+#define TGSI_DST_REGISTER_EXT_TYPE_MODULATE     1
+#define TGSI_DST_REGISTER_EXT_TYPE_PREDICATE    2
+
+struct tgsi_dst_register_ext
+{
+   unsigned Type     : 4;    /* TGSI_DST_REGISTER_EXT_TYPE_ */
+   unsigned Padding  : 27;
+   unsigned Extended : 1;    /* BOOL */
+};
+
+/**
+ * Extra destination register modifiers
+ *
+ * If tgsi_dst_register_ext::Type is TGSI_DST_REGISTER_EXT_TYPE_CONDCODE,
+ * it should be cast to tgsi_dst_register_ext_condcode.
+ * 
+ * If tgsi_dst_register_ext::Type is TGSI_DST_REGISTER_EXT_TYPE_MODULATE,
+ * it should be cast to tgsi_dst_register_ext_modulate.
+ * 
+ * If tgsi_dst_register_ext::Type is TGSI_DST_REGISTER_EXT_TYPE_PREDICATE,
+ * it should be cast to tgsi_dst_register_ext_predicate.
+ * 
+ * If tgsi_dst_register_ext::Extended is TRUE, another tgsi_dst_register_ext
+ * follows.
+ */
+struct tgsi_dst_register_ext_concode
+{
+   unsigned Type         : 4;    /* TGSI_DST_REGISTER_EXT_TYPE_CONDCODE */
+   unsigned CondMask     : 4;    /* TGSI_CC_ */
+   unsigned CondSwizzleX : 2;    /* TGSI_SWIZZLE_ */
+   unsigned CondSwizzleY : 2;    /* TGSI_SWIZZLE_ */
+   unsigned CondSwizzleZ : 2;    /* TGSI_SWIZZLE_ */
+   unsigned CondSwizzleW : 2;    /* TGSI_SWIZZLE_ */
+   unsigned CondSrcIndex : 4;    /* UINT */
+   unsigned Padding      : 11;
+   unsigned Extended     : 1;    /* BOOL */
+};
+
+#define TGSI_MODULATE_1X        0
+#define TGSI_MODULATE_2X        1
+#define TGSI_MODULATE_4X        2
+#define TGSI_MODULATE_8X        3
+#define TGSI_MODULATE_HALF      4
+#define TGSI_MODULATE_QUARTER   5
+#define TGSI_MODULATE_EIGHTH    6
+
+struct tgsi_dst_register_ext_modulate
+{
+   unsigned Type     : 4;    /* TGSI_DST_REGISTER_EXT_TYPE_MODULATE */
+   unsigned Modulate : 4;    /* TGSI_MODULATE_ */
+   unsigned Padding  : 23;
+   unsigned Extended : 1;    /* BOOL */
+};
+
+/*
+ * Currently, the following constraints apply.
+ *
+ * - PredSwizzleXYZW is either set to identity or replicate.
+ * - PredSrcIndex is 0.
+ */
+
+struct tgsi_dst_register_ext_predicate
+{
+   unsigned Type         : 4;    /* TGSI_DST_REGISTER_EXT_TYPE_PREDICATE */
+   unsigned PredSwizzleX : 2;    /* TGSI_SWIZZLE_ */
+   unsigned PredSwizzleY : 2;    /* TGSI_SWIZZLE_ */
+   unsigned PredSwizzleZ : 2;    /* TGSI_SWIZZLE_ */
+   unsigned PredSwizzleW : 2;    /* TGSI_SWIZZLE_ */
+   unsigned PredSrcIndex : 4;    /* UINT */
+   unsigned Negate       : 1;    /* BOOL */
+   unsigned Padding      : 14;
+   unsigned Extended     : 1;    /* BOOL */
+};
+
+
+#if defined __cplusplus
+} // extern "C"
+#endif // defined __cplusplus
+
+#endif // !defined TGSI_TOKEN_H
+
diff --git a/src/gallium/include/pipe/p_state.h b/src/gallium/include/pipe/p_state.h
new file mode 100644
index 0000000000..4d3a6b2f41
--- /dev/null
+++ b/src/gallium/include/pipe/p_state.h
@@ -0,0 +1,322 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+
+/**
+ * Abstract graphics pipe state objects.
+ *
+ * Basic notes:
+ *   1. Want compact representations, so we use bitfields.
+ *   2. Put bitfields before other (GLfloat) fields.
+ */
+
+
+#ifndef PIPE_STATE_H
+#define PIPE_STATE_H
+
+#include "p_compiler.h"
+#include "p_defines.h"
+#include "p_format.h"
+
+/**
+ * Implementation limits
+ */
+#define PIPE_MAX_SAMPLERS     8
+#define PIPE_MAX_CLIP_PLANES  6
+#define PIPE_MAX_CONSTANT    32
+#define PIPE_ATTRIB_MAX      32
+#define PIPE_MAX_COLOR_BUFS   8
+#define PIPE_MAX_TEXTURE_LEVELS  16
+#define PIPE_MAX_FEEDBACK_ATTRIBS 16
+#define PIPE_MAX_SHADER_INPUTS 16
+#define PIPE_MAX_SHADER_OUTPUTS 16
+
+
+/* fwd decls */
+struct pipe_surface;
+struct pipe_winsys;
+
+
+
+/**
+ * The driver will certainly subclass this to include actual memory
+ * management information.
+ */
+struct pipe_buffer {
+   unsigned alignment;
+   unsigned usage;
+   unsigned size;
+
+   /** Reference count */
+   unsigned refcount;
+};
+
+
+
+
+/**
+ * Primitive (point/line/tri) rasterization info
+ */
+struct pipe_rasterizer_state
+{
+   unsigned flatshade:1;
+   unsigned light_twoside:1;
+   unsigned front_winding:2;  /**< PIPE_WINDING_x */
+   unsigned cull_mode:2;      /**< PIPE_WINDING_x */
+   unsigned fill_cw:2;        /**< PIPE_POLYGON_MODE_x */
+   unsigned fill_ccw:2;       /**< PIPE_POLYGON_MODE_x */
+   unsigned offset_cw:1;
+   unsigned offset_ccw:1;
+   unsigned scissor:1;
+   unsigned poly_smooth:1;
+   unsigned poly_stipple_enable:1;
+   unsigned point_smooth:1;
+   unsigned point_sprite:1;
+   unsigned point_size_per_vertex:1; /**< size computed in vertex shader */
+   unsigned multisample:1;         /* XXX maybe more ms state in future */
+   unsigned line_smooth:1;
+   unsigned line_stipple_enable:1;
+   unsigned line_stipple_factor:8;  /**< [1..256] actually */
+   unsigned line_stipple_pattern:16;
+   unsigned bypass_clipping:1;
+   unsigned origin_lower_left:1;  /**< Is (0,0) the lower-left corner? */
+
+   float line_width;
+   float point_size;           /**< used when no per-vertex size */
+   float offset_units;
+   float offset_scale;
+   ubyte sprite_coord_mode[PIPE_MAX_SHADER_OUTPUTS]; /**< PIPE_SPRITE_COORD_ */
+};
+
+
+struct pipe_poly_stipple {
+   unsigned stipple[32];
+};
+
+
+struct pipe_viewport_state {
+   float scale[4];
+   float translate[4];
+};
+
+
+struct pipe_scissor_state {
+   unsigned minx:16;
+   unsigned miny:16;
+   unsigned maxx:16;
+   unsigned maxy:16;
+};
+
+
+struct pipe_clip_state {
+   float ucp[PIPE_MAX_CLIP_PLANES][4];
+   unsigned nr;
+};
+
+
+/**
+ * Constants for vertex/fragment shaders
+ */
+struct pipe_constant_buffer {
+   struct pipe_buffer *buffer;
+   unsigned size;    /** in bytes */
+};
+
+
+struct pipe_shader_state {
+   const struct tgsi_token *tokens;
+   ubyte num_inputs;
+   ubyte num_outputs;
+   ubyte input_map[PIPE_MAX_SHADER_INPUTS]; /* XXX this may be temporary */
+   ubyte input_semantic_name[PIPE_MAX_SHADER_INPUTS]; /**< TGSI_SEMANTIC_x */
+   ubyte input_semantic_index[PIPE_MAX_SHADER_INPUTS];
+   ubyte output_semantic_name[PIPE_MAX_SHADER_OUTPUTS]; /**< TGSI_SEMANTIC_x */
+   ubyte output_semantic_index[PIPE_MAX_SHADER_OUTPUTS];
+};
+
+
+struct pipe_depth_stencil_alpha_state
+{
+   struct {
+      unsigned enabled:1;         /**< depth test enabled? */
+      unsigned writemask:1;       /**< allow depth buffer writes? */
+      unsigned func:3;            /**< depth test func (PIPE_FUNC_x) */
+      unsigned occlusion_count:1; /**< do occlusion counting? */
+   } depth;
+   struct {
+      unsigned enabled:1;
+      unsigned func:3;     /**< PIPE_FUNC_x */
+      unsigned fail_op:3;  /**< PIPE_STENCIL_OP_x */
+      unsigned zpass_op:3; /**< PIPE_STENCIL_OP_x */
+      unsigned zfail_op:3; /**< PIPE_STENCIL_OP_x */
+      ubyte ref_value;    
+      ubyte value_mask;
+      ubyte write_mask;
+   } stencil[2];           /**< [0] = front, [1] = back */
+   struct {
+      unsigned enabled:1;
+      unsigned func:3;     /**< PIPE_FUNC_x */
+      float ref;           /**< reference value */
+   } alpha;
+};
+
+
+struct pipe_blend_state {
+   unsigned blend_enable:1;
+
+   unsigned rgb_func:3;          /**< PIPE_BLEND_x */
+   unsigned rgb_src_factor:5;    /**< PIPE_BLENDFACTOR_x */
+   unsigned rgb_dst_factor:5;    /**< PIPE_BLENDFACTOR_x */
+
+   unsigned alpha_func:3;        /**< PIPE_BLEND_x */
+   unsigned alpha_src_factor:5;  /**< PIPE_BLENDFACTOR_x */
+   unsigned alpha_dst_factor:5;  /**< PIPE_BLENDFACTOR_x */
+
+   unsigned logicop_enable:1;
+   unsigned logicop_func:4;      /**< PIPE_LOGICOP_x */
+
+   unsigned colormask:4;         /**< bitmask of PIPE_MASK_R/G/B/A */
+   unsigned dither:1;
+};
+
+
+struct pipe_blend_color {
+   float color[4];
+};
+
+
+struct pipe_framebuffer_state
+{
+   /** multiple colorbuffers for multiple render targets */
+   unsigned num_cbufs;
+   struct pipe_surface *cbufs[PIPE_MAX_COLOR_BUFS];
+
+   struct pipe_surface *zsbuf;      /**< Z/stencil buffer */
+};
+
+
+/**
+ * Texture sampler state.
+ */
+struct pipe_sampler_state
+{
+   unsigned wrap_s:3;        /**< PIPE_TEX_WRAP_x */
+   unsigned wrap_t:3;        /**< PIPE_TEX_WRAP_x */
+   unsigned wrap_r:3;        /**< PIPE_TEX_WRAP_x */
+   unsigned min_img_filter:2;    /**< PIPE_TEX_FILTER_x */
+   unsigned min_mip_filter:2;    /**< PIPE_TEX_MIPFILTER_x */
+   unsigned mag_img_filter:2;    /**< PIPE_TEX_FILTER_x */
+   unsigned compare:1;       /**< shadow/depth compare enabled? */
+   unsigned compare_mode:1;  /**< PIPE_TEX_COMPARE_x */
+   unsigned compare_func:3;  /**< PIPE_FUNC_x */
+   unsigned normalized_coords:1;  /**< Are coords normalized to [0,1]? */
+   float shadow_ambient;          /**< shadow test fail color/intensity */
+   float lod_bias;                /**< LOD/lambda bias */
+   float min_lod, max_lod;        /**< LOD clamp range, after bias */
+   float border_color[4];
+   float max_anisotropy;
+};
+
+
+/**
+ * 2D surface.  This is basically a view into a memory buffer.
+ * May be a renderbuffer, texture mipmap level, etc.
+ */
+struct pipe_surface
+{
+   struct pipe_buffer *buffer; /**< driver private buffer handle */
+   enum pipe_format format;      /**< PIPE_FORMAT_x */
+   unsigned status;              /**< PIPE_SURFACE_STATUS_x */
+   unsigned clear_value;         /**< may be temporary */
+   unsigned cpp;                 /**< bytes per pixel */
+   unsigned width, height;
+   unsigned pitch;               /**< in pixels */
+   unsigned offset;              /**< offset from start of buffer, in bytes */
+   unsigned refcount;
+   struct pipe_winsys *winsys;   /**< winsys which owns/created the surface */
+};
+
+
+/**
+ * Texture. Represents one or several texture images on one or several mipmap
+ * levels.
+ */
+struct pipe_texture
+{ 
+   /* Effectively the key:
+    */
+   enum pipe_texture_target target; /**< PIPE_TEXTURE_x */
+   enum pipe_format format;         /**< PIPE_FORMAT_x */
+
+   unsigned last_level;    /**< Index of last mipmap level present/defined */
+
+   unsigned width[PIPE_MAX_TEXTURE_LEVELS];
+   unsigned height[PIPE_MAX_TEXTURE_LEVELS];
+   unsigned depth[PIPE_MAX_TEXTURE_LEVELS];
+   unsigned cpp;
+
+   unsigned compressed:1;
+
+   /* These are also refcounted:
+    */
+   unsigned refcount;
+};
+
+
+/**
+ * A vertex buffer.  Typically, all the vertex data/attributes for
+ * drawing something will be in one buffer.  But it's also possible, for
+ * example, to put colors in one buffer and texcoords in another.
+ */
+struct pipe_vertex_buffer
+{
+   unsigned pitch:11;    /**< stride to same attrib in next vertex, in bytes */
+   unsigned max_index;   /**< number of vertices in this buffer */
+   unsigned buffer_offset;  /**< offset to start of data in buffer, in bytes */
+   struct pipe_buffer *buffer;  /**< the actual buffer */
+};
+
+
+/**
+ * Information to describe a vertex attribute (position, color, etc)
+ */
+struct pipe_vertex_element
+{
+   /** Offset of this attribute, in bytes, from the start of the vertex */
+   unsigned src_offset:11;
+
+   /** Which vertex_buffer (as given to pipe->set_vertex_buffer()) does
+    * this attribute live in?
+    */
+   unsigned vertex_buffer_index:5;
+   unsigned nr_components:3;
+ 
+   enum pipe_format src_format; 	   /**< PIPE_FORMAT_* */
+};
+
+
+#endif
diff --git a/src/gallium/include/pipe/p_thread.h b/src/gallium/include/pipe/p_thread.h
new file mode 100644
index 0000000000..cd432c547c
--- /dev/null
+++ b/src/gallium/include/pipe/p_thread.h
@@ -0,0 +1,54 @@
+/**************************************************************************
+ * 
+ * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+#ifndef P_THREAD_H
+#define P_THREAD_H
+
+#include "p_compiler.h"
+
+/*
+ * XXX: We should come up with a system-independent thread definitions.
+ * XXX: Patching glthread defs for now.
+ */
+
+#ifndef __MSC__
+
+#include "glapi/glthread.h"
+
+#else /* __MSC__ */
+
+typedef int _glthread_Mutex;
+
+#define _glthread_INIT_MUTEX( M )   ((void) (M))
+#define _glthread_LOCK_MUTEX( M )   ((void) (M))
+#define _glthread_UNLOCK_MUTEX( M ) ((void) (M))
+
+#define sched_yield() ((void) 0)
+
+#endif /* __MSC__ */
+
+#endif /* P_THREAD_H */
diff --git a/src/gallium/include/pipe/p_util.h b/src/gallium/include/pipe/p_util.h
new file mode 100644
index 0000000000..d7da2801c9
--- /dev/null
+++ b/src/gallium/include/pipe/p_util.h
@@ -0,0 +1,408 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+#ifndef P_UTIL_H
+#define P_UTIL_H
+
+#include "p_compiler.h"
+#include "p_debug.h"
+#include <math.h>
+
+
+#ifdef WIN32
+
+#ifdef __cplusplus
+extern "C"
+{
+#endif
+
+void * __stdcall
+EngAllocMem(
+    unsigned long Flags,
+    unsigned long MemSize,
+    unsigned long Tag );
+
+void __stdcall
+EngFreeMem(
+    void *Mem );
+
+#ifdef __cplusplus
+}
+#endif
+
+static INLINE void *
+MALLOC( unsigned size )
+{
+   return EngAllocMem( 0, size, 'D3AG' );
+}
+
+static INLINE void *
+CALLOC( unsigned count, unsigned size )
+{
+   void *ptr = MALLOC( count * size );
+   if( ptr ) {
+      memset( ptr, 0, count * size );
+   }
+   return ptr;
+}
+
+static INLINE void
+FREE( void *ptr )
+{
+   if( ptr ) {
+      EngFreeMem( ptr );
+   }
+}
+
+static INLINE void *
+REALLOC( void *old_ptr, unsigned old_size, unsigned new_size )
+{
+   void *new_ptr;
+   if( new_size <= old_size ) {
+      return old_ptr;
+   }
+   new_ptr = MALLOC( new_size );
+   if( new_ptr ) {
+      memcpy( new_ptr, old_ptr, old_size );
+   }
+   FREE( old_ptr );
+   return new_ptr;
+}
+
+#define GETENV( X )  NULL
+
+#else /* WIN32 */
+
+#define MALLOC( SIZE )  malloc( SIZE )
+
+#define CALLOC( COUNT, SIZE )   calloc( COUNT, SIZE )
+
+#define FREE( PTR )  free( PTR )
+
+#define REALLOC( OLDPTR, OLDSIZE, NEWSIZE )  realloc( OLDPTR, NEWSIZE )
+
+#define GETENV( X )  getenv( X )
+
+#endif /* WIN32 */
+
+#define MALLOC_STRUCT(T)   (struct T *) MALLOC(sizeof(struct T))
+
+#define CALLOC_STRUCT(T)   (struct T *) CALLOC(1, sizeof(struct T))
+
+
+/**
+ * Return a pointer aligned to next multiple of N bytes.
+ */
+static INLINE void *
+align_pointer( void *unaligned, uint alignment )
+{
+   if (sizeof(void *) == 64) {
+      union {
+         void *p;
+         uint64 u;
+      } pu;
+      pu.p = unaligned;
+      pu.u = (pu.u + alignment - 1) & ~(uint64) (alignment - 1);
+      return pu.p;
+   }
+   else {
+      /* 32-bit pointers */
+      union {
+         void *p;
+         uint u;
+      } pu;
+      pu.p = unaligned;
+      pu.u = (pu.u + alignment - 1) & ~(alignment - 1);
+      return pu.p;
+   }
+}
+
+/**
+ * Return memory on given byte alignment
+ */
+static INLINE void *
+align_malloc(size_t bytes, uint alignment)
+{
+#if defined(HAVE_POSIX_MEMALIGN)
+   void *mem;
+   (void) posix_memalign(& mem, alignment, bytes);
+   return mem;
+#else
+   char *ptr, *buf;
+
+   assert( alignment > 0 );
+
+   ptr = (char *) MALLOC(bytes + alignment + sizeof(void *));
+   if (!ptr)
+      return NULL;
+
+   buf = (char *) align_pointer( ptr + sizeof(void *), alignment );
+   *(char **)(buf - sizeof(void *)) = ptr;
+
+   return buf;
+#endif /* defined(HAVE_POSIX_MEMALIGN) */
+}
+
+/**
+ * Free memory returned by align_malloc().
+ */
+static INLINE void
+align_free(void *ptr)
+{
+#if defined(HAVE_POSIX_MEMALIGN)
+   FREE(ptr);
+#else
+   void **cubbyHole = (void **) ((char *) ptr - sizeof(void *));
+   void *realAddr = *cubbyHole;
+   FREE(realAddr);
+#endif /* defined(HAVE_POSIX_MEMALIGN) */
+}
+
+
+
+/**
+ * Duplicate a block of memory.
+ */
+static INLINE void *
+mem_dup(const void *src, uint size)
+{
+   void *dup = MALLOC(size);
+   if (dup)
+      memcpy(dup, src, size);
+   return dup;
+}
+
+
+
+#define CLAMP( X, MIN, MAX )  ( (X)<(MIN) ? (MIN) : ((X)>(MAX) ? (MAX) : (X)) )
+#define MIN2( A, B )   ( (A)<(B) ? (A) : (B) )
+#define MAX2( A, B )   ( (A)>(B) ? (A) : (B) )
+
+#define Elements(x) sizeof(x)/sizeof(*(x))
+#define Offset(TYPE, MEMBER) ((unsigned)&(((TYPE *)NULL)->MEMBER))
+
+/**
+ * Return a pointer aligned to next multiple of 16 bytes.
+ */
+static INLINE void *
+align16( void *unaligned )
+{
+   return align_pointer( unaligned, 16 );
+}
+
+
+static INLINE int align_int(int x, int align)
+{
+   return (x + align - 1) & ~(align - 1);
+}
+
+
+
+#if defined(__MSC__) && defined(__WIN32__)
+static INLINE unsigned ffs( unsigned u )
+{
+   unsigned i;
+
+   if( u == 0 ) {
+      return 0;
+   }
+
+   __asm bsf eax, [u]
+   __asm inc eax
+   __asm mov [i], eax
+
+   return i;
+}
+#endif
+
+union fi {
+   float f;
+   int i;
+   unsigned ui;
+};
+
+#define UBYTE_TO_FLOAT( ub ) ((float)(ub) / 255.0F)
+
+#define IEEE_0996 0x3f7f0000	/* 0.996 or so */
+
+/* This function/macro is sensitive to precision.  Test very carefully
+ * if you change it!
+ */
+#define UNCLAMPED_FLOAT_TO_UBYTE(UB, F)					\
+        do {								\
+           union fi __tmp;						\
+           __tmp.f = (F);						\
+           if (__tmp.i < 0)						\
+              UB = (ubyte) 0;						\
+           else if (__tmp.i >= IEEE_0996)				\
+              UB = (ubyte) 255;					\
+           else {							\
+              __tmp.f = __tmp.f * (255.0f/256.0f) + 32768.0f;		\
+              UB = (ubyte) __tmp.i;					\
+           }								\
+        } while (0)
+
+
+
+static INLINE unsigned pack_ub4( unsigned char b0,
+				 unsigned char b1,
+				 unsigned char b2,
+				 unsigned char b3 )
+{
+   return ((((unsigned int)b0) << 0) |
+	   (((unsigned int)b1) << 8) |
+	   (((unsigned int)b2) << 16) |
+	   (((unsigned int)b3) << 24));
+}
+
+static INLINE unsigned fui( float f )
+{
+   union fi fi;
+   fi.f = f;
+   return fi.ui;
+}
+
+static INLINE unsigned char float_to_ubyte( float f )
+{
+   unsigned char ub;
+   UNCLAMPED_FLOAT_TO_UBYTE(ub, f);
+   return ub;
+}
+
+static INLINE unsigned pack_ui32_float4( float a,
+					 float b, 
+					 float c, 
+					 float d )
+{
+   return pack_ub4( float_to_ubyte(a),
+		    float_to_ubyte(b),
+		    float_to_ubyte(c),
+		    float_to_ubyte(d) );
+}
+
+#define COPY_4V( DST, SRC )         \
+do {                                \
+   (DST)[0] = (SRC)[0];             \
+   (DST)[1] = (SRC)[1];             \
+   (DST)[2] = (SRC)[2];             \
+   (DST)[3] = (SRC)[3];             \
+} while (0)
+
+
+#define COPY_4FV( DST, SRC )  COPY_4V(DST, SRC)
+
+
+#define ASSIGN_4V( DST, V0, V1, V2, V3 ) \
+do {                                     \
+   (DST)[0] = (V0);                      \
+   (DST)[1] = (V1);                      \
+   (DST)[2] = (V2);                      \
+   (DST)[3] = (V3);                      \
+} while (0)
+
+
+static INLINE int ifloor(float f)
+{
+   int ai, bi;
+   double af, bf;
+   union fi u;
+
+   af = (3 << 22) + 0.5 + (double)f;
+   bf = (3 << 22) + 0.5 - (double)f;
+   u.f = (float) af;  ai = u.i;
+   u.f = (float) bf;  bi = u.i;
+   return (ai - bi) >> 1;
+}
+
+
+#if defined(__GNUC__) && defined(__i386__) 
+static INLINE int iround(float f)
+{
+   int r;
+   __asm__ ("fistpl %0" : "=m" (r) : "t" (f) : "st");
+   return r;
+}
+#elif defined(__MSC__) && defined(__WIN32__)
+static INLINE int iround(float f)
+{
+   int r;
+   _asm {
+	 fld f
+	 fistp r
+	}
+   return r;
+}
+#else
+#define IROUND(f)  ((int) (((f) >= 0.0F) ? ((f) + 0.5F) : ((f) - 0.5F)))
+#endif
+
+
+/* Could maybe have an inline version of this?
+ */
+#if defined(__GNUC__)
+#define FABSF(x)   fabsf(x)
+#else
+#define FABSF(x)   ((float) fabs(x))
+#endif
+
+/* Pretty fast, and accurate.
+ * Based on code from http://www.flipcode.com/totd/
+ */
+static INLINE float LOG2(float val)
+{
+   union fi num;
+   int log_2;
+
+   num.f = val;
+   log_2 = ((num.i >> 23) & 255) - 128;
+   num.i &= ~(255 << 23);
+   num.i += 127 << 23;
+   num.f = ((-1.0f/3) * num.f + 2) * num.f - 2.0f/3;
+   return num.f + log_2;
+}
+
+#if defined(__GNUC__)
+#define CEILF(x)   ceilf(x)
+#else
+#define CEILF(x)   ((float) ceil(x))
+#endif
+
+static INLINE int align(int value, int alignment)
+{
+   return (value + alignment - 1) & ~(alignment - 1);
+}
+
+
+/* util/p_util.c
+ */
+extern void pipe_copy_rect(ubyte * dst, unsigned cpp, unsigned dst_pitch,
+                           unsigned dst_x, unsigned dst_y, unsigned width,
+                           unsigned height, const ubyte * src,
+                           int src_pitch, unsigned src_x, int src_y);
+
+
+#endif
diff --git a/src/gallium/include/pipe/p_winsys.h b/src/gallium/include/pipe/p_winsys.h
new file mode 100644
index 0000000000..1e81eebd78
--- /dev/null
+++ b/src/gallium/include/pipe/p_winsys.h
@@ -0,0 +1,160 @@
+ /**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+#ifndef P_WINSYS_H
+#define P_WINSYS_H
+
+
+#include "p_format.h"
+
+/**
+ * \file
+ * This is the interface that Gallium3D requires any window system
+ * hosting it to implement.  This is the only include file in Gallium3D
+ * which is public.
+ */
+
+
+/** Opaque type */
+struct pipe_fence_handle;
+
+struct pipe_surface;
+
+
+/**
+ * Gallium3D drivers are (meant to be!) independent of both GL and the
+ * window system.  The window system provides a buffer manager and a
+ * set of additional hooks for things like command buffer submission,
+ * etc.
+ *
+ * There clearly has to be some agreement between the window system
+ * driver and the hardware driver about the format of command buffers,
+ * etc.
+ */
+struct pipe_winsys
+{
+   /** Returns name of this winsys interface */
+   const char *(*get_name)( struct pipe_winsys *sws );
+
+   /**
+    * Do any special operations to ensure frontbuffer contents are
+    * displayed, eg copy fake frontbuffer.
+    */
+   void (*flush_frontbuffer)( struct pipe_winsys *sws,
+                              struct pipe_surface *surf,
+                              void *context_private );
+
+   /** Debug output */
+   void (*printf)( struct pipe_winsys *sws,
+		   const char *, ... );	
+
+
+   /** allocate a new surface (no context dependency) */
+   struct pipe_surface *(*surface_alloc)(struct pipe_winsys *ws);
+
+   /**
+    * Allocate storage for a pipe_surface.
+    * Returns 0 if succeeds.
+    */
+   int (*surface_alloc_storage)(struct pipe_winsys *ws,
+                                struct pipe_surface *surf,
+                                unsigned width, unsigned height,
+                                enum pipe_format format,
+                                unsigned flags);
+   
+   void (*surface_release)(struct pipe_winsys *ws, struct pipe_surface **s);
+
+   
+   /**
+    * Buffer management. Buffer attributes are mostly fixed over its lifetime.
+    *
+    * Remember that gallium gets to choose the interface it needs, and the
+    * window systems must then implement that interface (rather than the
+    * other way around...).
+    *
+    * usage is a bitmask of PIPE_BUFFER_USAGE_PIXEL/VERTEX/INDEX/CONSTANT. This
+    * usage argument is only an optimization hint, not a guarantee, therefore 
+    * proper behavior must be observed in all circumstances.
+    */
+   struct pipe_buffer *(*buffer_create)( struct pipe_winsys *sws, 
+					        unsigned alignment,
+                                                unsigned usage,
+                                                unsigned size );
+
+   /** Create a buffer that wraps user-space data */
+   struct pipe_buffer *(*user_buffer_create)(struct pipe_winsys *sws, 
+                                                    void *ptr,
+                                                    unsigned bytes);
+
+   /** 
+    * Map the entire data store of a buffer object into the client's address.
+    * flags is bitmask of PIPE_BUFFER_USAGE_CPU_READ/WRITE flags. 
+    */
+   void *(*buffer_map)( struct pipe_winsys *sws, 
+			struct pipe_buffer *buf,
+			unsigned usage );
+   
+   void (*buffer_unmap)( struct pipe_winsys *sws, 
+			 struct pipe_buffer *buf );
+
+   void (*buffer_destroy)( struct pipe_winsys *sws,
+			   struct pipe_buffer *buf );
+
+
+   /** Set ptr = fence, with reference counting */
+   void (*fence_reference)( struct pipe_winsys *sws,
+                            struct pipe_fence_handle **ptr,
+                            struct pipe_fence_handle *fence );
+
+   /**
+    * Checks whether the fence has been signalled.
+    *  
+    * The meaning of flag is pipe-driver specific.
+    *
+    * Returns zero if it has.
+    */
+   int (*fence_signalled)( struct pipe_winsys *sws,
+                           struct pipe_fence_handle *fence,
+                           unsigned flag );
+
+   /**
+    * Wait for the fence to finish.
+    * 
+    * The meaning of flag is pipe-driver specific.
+    * 
+    * Returns zero on success.
+    */
+   int (*fence_finish)( struct pipe_winsys *sws,
+                        struct pipe_fence_handle *fence,
+                        unsigned flag );
+
+
+};
+
+
+
+#endif /* P_WINSYS_H */
diff --git a/src/gallium/winsys/dri/intel/Makefile b/src/gallium/winsys/dri/intel/Makefile
new file mode 100644
index 0000000000..9ae0f01325
--- /dev/null
+++ b/src/gallium/winsys/dri/intel/Makefile
@@ -0,0 +1,38 @@
+
+TOP = ../../../../..
+include $(TOP)/configs/current
+
+LIBNAME = i915tex_dri.so
+
+MINIGLX_SOURCES = server/intel_dri.c
+
+PIPE_DRIVERS = \
+	$(TOP)/src/mesa/pipe/softpipe/libsoftpipe.a \
+	$(TOP)/src/mesa/pipe/i915simple/libi915simple.a
+
+DRIVER_SOURCES = \
+	intel_winsys_pipe.c \
+	intel_winsys_softpipe.c \
+	intel_winsys_i915.c \
+	intel_batchbuffer.c \
+	intel_swapbuffers.c \
+	intel_context.c \
+	intel_lock.c \
+	intel_screen.c \
+	intel_batchpool.c
+
+C_SOURCES = \
+	$(COMMON_GALLIUM_SOURCES) \
+	$(COMMON_BM_SOURCES) \
+	$(DRIVER_SOURCES)
+
+ASM_SOURCES = 
+
+DRIVER_DEFINES = -I../intel $(shell pkg-config libdrm --atleast-version=2.3.1 \
+				&& echo "-DDRM_VBLANK_FLIP=DRM_VBLANK_FLIP")
+
+include ../Makefile.template
+
+intel_tex_layout.o: ../intel/intel_tex_layout.c
+
+symlinks:
diff --git a/src/gallium/winsys/dri/intel/SConscript b/src/gallium/winsys/dri/intel/SConscript
new file mode 100644
index 0000000000..a7cc10450e
--- /dev/null
+++ b/src/gallium/winsys/dri/intel/SConscript
@@ -0,0 +1,41 @@
+Import('*')
+
+env = drienv.Clone()
+
+env.Append(CPPPATH = [
+	'../intel',
+	'server'
+])
+
+#MINIGLX_SOURCES = server/intel_dri.c
+
+pipe_drivers = [
+	softpipe,
+	i915simple
+]
+
+DRIVER_SOURCES = [
+	'intel_winsys_pipe.c',
+	'intel_winsys_softpipe.c',
+	'intel_winsys_i915.c',
+	'intel_batchbuffer.c',
+	'intel_swapbuffers.c',
+	'intel_context.c',
+	'intel_lock.c',
+	'intel_screen.c',
+	'intel_batchpool.c',
+]
+
+sources = \
+	COMMON_GALLIUM_SOURCES + \
+	COMMON_BM_SOURCES + \
+	DRIVER_SOURCES
+
+# DRIVER_DEFINES = -I../intel $(shell pkg-config libdrm --atleast-version=2.3.1 \
+#				&& echo "-DDRM_VBLANK_FLIP=DRM_VBLANK_FLIP")
+
+env.SharedLibrary(
+	target ='i915tex_dri.so',
+	source = sources,
+	LIBS = pipe_drivers + env['LIBS'],
+)
\ No newline at end of file
diff --git a/src/gallium/winsys/dri/intel/intel_batchbuffer.c b/src/gallium/winsys/dri/intel/intel_batchbuffer.c
new file mode 100644
index 0000000000..49e04d81ec
--- /dev/null
+++ b/src/gallium/winsys/dri/intel/intel_batchbuffer.c
@@ -0,0 +1,357 @@
+/**************************************************************************
+ * 
+ * Copyright 2006 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+#include <errno.h>
+#include "intel_batchbuffer.h"
+#include "intel_context.h"
+#include "intel_screen.h"
+#include "intel_reg.h"
+#include "drm.h"
+
+/* Relocations in kernel space:
+ *    - pass dma buffer seperately
+ *    - memory manager knows how to patch
+ *    - pass list of dependent buffers
+ *    - pass relocation list
+ *
+ * Either:
+ *    - get back an offset for buffer to fire
+ *    - memory manager knows how to fire buffer
+ *
+ * Really want the buffer to be AGP and pinned.
+ *
+ */
+
+/* Cliprect fence: The highest fence protecting a dma buffer
+ * containing explicit cliprect information.  Like the old drawable
+ * lock but irq-driven.  X server must wait for this fence to expire
+ * before changing cliprects [and then doing sw rendering?].  For
+ * other dma buffers, the scheduler will grab current cliprect info
+ * and mix into buffer.  X server must hold the lock while changing
+ * cliprects???  Make per-drawable.  Need cliprects in shared memory
+ * -- beats storing them with every cmd buffer in the queue.
+ *
+ * ==> X server must wait for this fence to expire before touching the
+ * framebuffer with new cliprects.
+ *
+ * ==> Cliprect-dependent buffers associated with a
+ * cliprect-timestamp.  All of the buffers associated with a timestamp
+ * must go to hardware before any buffer with a newer timestamp.
+ *
+ * ==> Dma should be queued per-drawable for correct X/GL
+ * synchronization.  Or can fences be used for this?
+ *
+ * Applies to: Blit operations, metaops, X server operations -- X
+ * server automatically waits on its own dma to complete before
+ * modifying cliprects ???
+ */
+
+static void
+intel_dump_batchbuffer(uint offset, uint * ptr, uint count)
+{
+   int i;
+   printf("\n\n\nSTART BATCH (%d dwords):\n", count / 4);
+   for (i = 0; i < count / 4; i += 1)
+      printf("\t0x%08x\n", ptr[i]);
+   printf("END BATCH\n\n\n");
+}
+
+
+void
+intel_batchbuffer_reset(struct intel_batchbuffer *batch)
+{
+   int i;
+
+   if (batch->map) {
+      driBOUnmap(batch->buffer);
+      batch->map = NULL;
+   }
+
+   /*
+    * Get a new, free batchbuffer.
+    */
+   batch->size =  BATCH_SZ;
+   driBOData(batch->buffer, batch->size, NULL, 0);
+
+   driBOResetList(&batch->list);
+
+   /*
+    * Unreference buffers previously on the relocation list.
+    */
+   for (i = 0; i < batch->nr_relocs; i++) {
+      struct buffer_reloc *r = &batch->reloc[i];
+      driBOUnReference(r->buf);
+   }
+
+   batch->list_count = 0;
+   batch->nr_relocs = 0;
+   batch->flags = 0;
+
+   /*
+    * We don't refcount the batchbuffer itself since we can't destroy it
+    * while it's on the list.
+    */
+
+   driBOAddListItem(&batch->list, batch->buffer,
+                    DRM_BO_FLAG_MEM_TT | DRM_BO_FLAG_EXE,
+                    DRM_BO_MASK_MEM | DRM_BO_FLAG_EXE);
+
+
+   batch->map = driBOMap(batch->buffer, DRM_BO_FLAG_WRITE, 0);
+   batch->ptr = batch->map;
+}
+
+
+/*======================================================================
+ * Public functions
+ */
+struct intel_batchbuffer *
+intel_batchbuffer_alloc(struct intel_context *intel)
+{
+   struct intel_batchbuffer *batch = calloc(sizeof(*batch), 1);
+
+   batch->intel = intel;
+
+   driGenBuffers(intel->intelScreen->batchPool, "batchbuffer", 1,
+                 &batch->buffer, 4096,
+                 DRM_BO_FLAG_MEM_TT | DRM_BO_FLAG_EXE, 0);
+   batch->last_fence = NULL;
+   driBOCreateList(20, &batch->list);
+   intel_batchbuffer_reset(batch);
+   return batch;
+}
+
+
+void
+intel_batchbuffer_free(struct intel_batchbuffer *batch)
+{
+   if (batch->last_fence) {
+      driFenceFinish(batch->last_fence,
+                     DRM_FENCE_TYPE_EXE | DRM_I915_FENCE_TYPE_RW,
+                     GL_FALSE);
+      driFenceUnReference(batch->last_fence);
+      batch->last_fence = NULL;
+   }
+   if (batch->map) {
+      driBOUnmap(batch->buffer);
+      batch->map = NULL;
+   }
+   driBOUnReference(batch->buffer);
+   batch->buffer = NULL;
+   free(batch);
+}
+
+
+static void
+intel_batch_ioctl(struct intel_context *intel,
+                  uint start_offset, uint used, boolean allow_unlock)
+{
+   drmI830BatchBuffer batch;
+
+   batch.start = start_offset;
+   batch.used = used;
+   batch.cliprects = NULL; /* unused */
+   batch.num_cliprects = 0;
+   batch.DR1 = 0;
+   batch.DR4 = 0; /* still need this ? */
+
+   DBG(IOCTL, "%s: 0x%x..0x%x DR4: %x cliprects: %d\n",
+       __FUNCTION__,
+       batch.start,
+       batch.start + batch.used * 4, batch.DR4, batch.num_cliprects);
+
+   if (drmCommandWrite(intel->driFd, DRM_I830_BATCHBUFFER, &batch,
+                       sizeof(batch))) {
+      printf("DRM_I830_BATCHBUFFER: %d\n", -errno);
+      UNLOCK_HARDWARE(intel);
+      exit(1);
+   }
+}
+
+
+/* TODO: Push this whole function into bufmgr.
+ */
+static void
+do_flush_locked(struct intel_batchbuffer *batch,
+                uint used, boolean allow_unlock)
+{
+   uint *ptr;
+   uint i, fenceFlags;
+   struct _DriFenceObject *fo;
+
+   driBOValidateList(batch->intel->driFd, &batch->list);
+
+   /* Apply the relocations.  This nasty map indicates to me that the
+    * whole task should be done internally by the memory manager, and
+    * that dma buffers probably need to be pinned within agp space.
+    */
+   ptr = (uint *) driBOMap(batch->buffer, DRM_BO_FLAG_WRITE,
+                             DRM_BO_HINT_ALLOW_UNFENCED_MAP);
+
+   for (i = 0; i < batch->nr_relocs; i++) {
+      struct buffer_reloc *r = &batch->reloc[i];
+
+      ptr[r->offset / 4] = driBOOffset(r->buf) + r->delta;
+   }
+
+   if (0)
+      intel_dump_batchbuffer(0, ptr, used);
+
+   driBOUnmap(batch->buffer);
+   batch->map = NULL;
+
+   intel_batch_ioctl(batch->intel,
+                     driBOOffset(batch->buffer),
+                     used, allow_unlock);
+
+   /*
+    * Kernel fencing. The flags tells the kernel that we've 
+    * programmed an MI_FLUSH.
+    */
+   fenceFlags = DRM_I915_FENCE_FLAG_FLUSHED;
+   fo = driFenceBuffers(batch->intel->driFd, "Batch fence", fenceFlags);
+
+   /*
+    * User space fencing.
+    */
+   driBOFence(batch->buffer, fo);
+
+   if (driFenceType(fo) == DRM_FENCE_TYPE_EXE) {
+     /*
+      * Oops. We only validated a batch buffer. This means we
+      * didn't do any proper rendering. Discard this fence object.
+      */
+      driFenceUnReference(fo);
+   }
+   else {
+      driFenceUnReference(batch->last_fence);
+      batch->last_fence = fo;
+      for (i = 0; i < batch->nr_relocs; i++) {
+	struct buffer_reloc *r = &batch->reloc[i];
+	driBOFence(r->buf, fo);
+      }
+   }
+}
+
+
+struct _DriFenceObject *
+intel_batchbuffer_flush(struct intel_batchbuffer *batch)
+{
+   struct intel_context *intel = batch->intel;
+   uint used = batch->ptr - batch->map;
+   const boolean was_locked = intel->locked;
+
+   if (used == 0)
+      return batch->last_fence;
+
+#define MI_FLUSH ((0 << 29) | (4 << 23))
+
+   /* Add the MI_BATCH_BUFFER_END.  Always add an MI_FLUSH - this is a
+    * performance drain that we would like to avoid.
+    */
+   if (used & 4) {
+      ((int *) batch->ptr)[0] = MI_FLUSH;
+      ((int *) batch->ptr)[1] = 0;
+      ((int *) batch->ptr)[2] = MI_BATCH_BUFFER_END;
+      used += 12;
+   }
+   else {
+      ((int *) batch->ptr)[0] = MI_FLUSH;
+      ((int *) batch->ptr)[1] = MI_BATCH_BUFFER_END;
+      used += 8;
+   }
+
+   driBOUnmap(batch->buffer);
+   batch->ptr = NULL;
+   batch->map = NULL;
+
+   /* TODO: Just pass the relocation list and dma buffer up to the
+    * kernel.
+    */
+   if (!was_locked)
+      LOCK_HARDWARE(intel);
+
+   do_flush_locked(batch, used, GL_FALSE);
+
+   if (!was_locked)
+      UNLOCK_HARDWARE(intel);
+
+   /* Reset the buffer:
+    */
+   intel_batchbuffer_reset(batch);
+   return batch->last_fence;
+}
+
+
+void
+intel_batchbuffer_finish(struct intel_batchbuffer *batch)
+{
+   struct _DriFenceObject *fence = intel_batchbuffer_flush(batch);
+   if (fence) {
+      driFenceReference(fence);
+      driFenceFinish(fence,
+                     DRM_FENCE_TYPE_EXE | DRM_I915_FENCE_TYPE_RW,
+                     GL_FALSE);
+      driFenceUnReference(fence);
+   }
+}
+
+
+/*  This is the only way buffers get added to the validate list.
+ */
+boolean
+intel_batchbuffer_emit_reloc(struct intel_batchbuffer *batch,
+                             struct _DriBufferObject *buffer,
+                             uint flags, uint mask, uint delta)
+{
+   assert(batch->nr_relocs < MAX_RELOCS);
+
+   driBOAddListItem(&batch->list, buffer, flags, mask);
+
+   {
+      struct buffer_reloc *r = &batch->reloc[batch->nr_relocs++];
+      driBOReference(buffer);
+      r->buf = buffer;
+      r->offset = batch->ptr - batch->map;
+      r->delta = delta;
+      *(uint *) batch->ptr = 0x12345678;
+   }
+
+   batch->ptr += 4;
+   return GL_TRUE;
+}
+
+
+void
+intel_batchbuffer_data(struct intel_batchbuffer *batch,
+                       const void *data, uint bytes, uint flags)
+{
+   assert((bytes & 3) == 0);
+   intel_batchbuffer_require_space(batch, bytes, flags);
+   memcpy(batch->ptr, data, bytes);
+   batch->ptr += bytes;
+}
diff --git a/src/gallium/winsys/dri/intel/intel_batchbuffer.h b/src/gallium/winsys/dri/intel/intel_batchbuffer.h
new file mode 100644
index 0000000000..82feafa21f
--- /dev/null
+++ b/src/gallium/winsys/dri/intel/intel_batchbuffer.h
@@ -0,0 +1,149 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+#ifndef INTEL_BATCHBUFFER_H
+#define INTEL_BATCHBUFFER_H
+
+#include "pipe/p_compiler.h"
+#include "dri_bufmgr.h"
+
+struct intel_context;
+
+#define BATCH_SZ 16384
+#define BATCH_RESERVED 16
+
+#define MAX_RELOCS 4096
+
+#define INTEL_BATCH_NO_CLIPRECTS 0x1
+#define INTEL_BATCH_CLIPRECTS    0x2
+
+struct buffer_reloc
+{
+   struct _DriBufferObject *buf;
+   uint offset;
+   uint delta;                /* not needed? */
+};
+
+struct intel_batchbuffer
+{
+   struct bufmgr *bm;
+   struct intel_context *intel;
+
+   struct _DriBufferObject *buffer;
+   struct _DriFenceObject *last_fence;
+   uint flags;
+
+   drmBOList list;
+   uint list_count;
+   ubyte *map;
+   ubyte *ptr;
+
+   struct buffer_reloc reloc[MAX_RELOCS];
+   uint nr_relocs;
+   uint size;
+};
+
+struct intel_batchbuffer *intel_batchbuffer_alloc(struct intel_context *intel);
+
+void intel_batchbuffer_free(struct intel_batchbuffer *batch);
+
+
+void intel_batchbuffer_finish(struct intel_batchbuffer *batch);
+
+struct _DriFenceObject *intel_batchbuffer_flush(struct intel_batchbuffer
+                                                *batch);
+
+void intel_batchbuffer_reset(struct intel_batchbuffer *batch);
+
+
+/* Unlike bmBufferData, this currently requires the buffer be mapped.
+ * Consider it a convenience function wrapping multiple
+ * intel_buffer_dword() calls.
+ */
+void intel_batchbuffer_data(struct intel_batchbuffer *batch,
+                            const void *data, uint bytes, uint flags);
+
+void intel_batchbuffer_release_space(struct intel_batchbuffer *batch,
+                                     uint bytes);
+
+boolean intel_batchbuffer_emit_reloc(struct intel_batchbuffer *batch,
+                                     struct _DriBufferObject *buffer,
+                                     uint flags,
+                                     uint mask, uint offset);
+
+/* Inline functions - might actually be better off with these
+ * non-inlined.  Certainly better off switching all command packets to
+ * be passed as structs rather than dwords, but that's a little bit of
+ * work...
+ */
+static INLINE uint
+intel_batchbuffer_space(struct intel_batchbuffer *batch)
+{
+   return (batch->size - BATCH_RESERVED) - (batch->ptr - batch->map);
+}
+
+
+static INLINE void
+intel_batchbuffer_emit_dword(struct intel_batchbuffer *batch, uint dword)
+{
+   assert(batch->map);
+   assert(intel_batchbuffer_space(batch) >= 4);
+   *(uint *) (batch->ptr) = dword;
+   batch->ptr += 4;
+}
+
+static INLINE void
+intel_batchbuffer_require_space(struct intel_batchbuffer *batch,
+                                uint sz, uint flags)
+{
+   assert(sz < batch->size - 8);
+   if (intel_batchbuffer_space(batch) < sz ||
+       (batch->flags != 0 && flags != 0 && batch->flags != flags))
+      intel_batchbuffer_flush(batch);
+
+   batch->flags |= flags;
+}
+
+/* Here are the crusty old macros, to be removed:
+ */
+#define BATCH_LOCALS
+
+#define BEGIN_BATCH(n, flags) do {				\
+   intel_batchbuffer_require_space(intel->batch, (n)*4, flags);	\
+} while (0)
+
+#define OUT_BATCH(d)  intel_batchbuffer_emit_dword(intel->batch, d)
+
+#define OUT_RELOC(buf,flags,mask,delta) do { 				\
+   assert((delta) >= 0);						\
+   intel_batchbuffer_emit_reloc(intel->batch, buf, flags, mask, delta);	\
+} while (0)
+
+#define ADVANCE_BATCH() do { } while(0)
+
+
+#endif
diff --git a/src/gallium/winsys/dri/intel/intel_batchpool.c b/src/gallium/winsys/dri/intel/intel_batchpool.c
new file mode 100644
index 0000000000..33b56817f6
--- /dev/null
+++ b/src/gallium/winsys/dri/intel/intel_batchpool.c
@@ -0,0 +1,424 @@
+/**************************************************************************
+ * 
+ * Copyright 2006 Tungsten Graphics, Inc., Bismarck, ND., USA
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDERS, AUTHORS AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM,
+ * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR 
+ * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE 
+ * USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * 
+ **************************************************************************/
+/*
+ * Authors: Thomas Hellström <thomas-at-tungstengraphics-dot-com>
+ */
+
+/**
+ * XXX NOTE: there are no intel dependencies in this file.
+ * Rename to dri_batchpool.c?
+ */
+
+#include <xf86drm.h>
+#include <stdlib.h>
+#include <errno.h>
+#include "imports.h"
+#include "glthread.h"
+#include "dri_bufpool.h"
+#include "dri_bufmgr.h"
+#include "intel_batchpool.h"
+
+
+typedef struct
+{
+   drmMMListHead head;
+   struct _BPool *parent;
+   struct _DriFenceObject *fence;
+   unsigned long start;
+   int unfenced;
+   int mapped;
+} BBuf;
+
+typedef struct _BPool
+{
+   _glthread_Mutex mutex;
+   unsigned long bufSize;
+   unsigned poolSize;
+   unsigned numFree;
+   unsigned numTot;
+   unsigned numDelayed;
+   unsigned checkDelayed;
+   drmMMListHead free;
+   drmMMListHead delayed;
+   drmMMListHead head;
+   drmBO kernelBO;
+   void *virtual;
+   BBuf *bufs;
+} BPool;
+
+
+static BPool *
+createBPool(int fd, unsigned long bufSize, unsigned numBufs, unsigned flags,
+            unsigned checkDelayed)
+{
+   BPool *p = (BPool *) malloc(sizeof(*p));
+   BBuf *buf;
+   int i;
+
+   if (!p)
+      return NULL;
+
+   p->bufs = (BBuf *) malloc(numBufs * sizeof(*p->bufs));
+   if (!p->bufs) {
+      free(p);
+      return NULL;
+   }
+
+   DRMINITLISTHEAD(&p->free);
+   DRMINITLISTHEAD(&p->head);
+   DRMINITLISTHEAD(&p->delayed);
+
+   p->numTot = numBufs;
+   p->numFree = numBufs;
+   p->bufSize = bufSize;
+   p->numDelayed = 0;
+   p->checkDelayed = checkDelayed;
+
+   _glthread_INIT_MUTEX(p->mutex);
+
+   if (drmBOCreate(fd, 0, numBufs * bufSize, 0, NULL, drm_bo_type_dc,
+                   flags, DRM_BO_HINT_DONT_FENCE, &p->kernelBO)) {
+      free(p->bufs);
+      free(p);
+      return NULL;
+   }
+   if (drmBOMap(fd, &p->kernelBO, DRM_BO_FLAG_READ | DRM_BO_FLAG_WRITE, 0,
+                &p->virtual)) {
+      drmBODestroy(fd, &p->kernelBO);
+      free(p->bufs);
+      free(p);
+      return NULL;
+   }
+
+   /*
+    * We unmap the buffer so that we can validate it later. Note that this is
+    * just a synchronizing operation. The buffer will have a virtual mapping
+    * until it is destroyed.
+    */
+
+   drmBOUnmap(fd, &p->kernelBO);
+
+   buf = p->bufs;
+   for (i = 0; i < numBufs; ++i) {
+      buf->parent = p;
+      buf->fence = NULL;
+      buf->start = i * bufSize;
+      buf->mapped = 0;
+      buf->unfenced = 0;
+      DRMLISTADDTAIL(&buf->head, &p->free);
+      buf++;
+   }
+
+   return p;
+}
+
+
+static void
+pool_checkFree(BPool * p, int wait)
+{
+   drmMMListHead *list, *prev;
+   BBuf *buf;
+   int signaled = 0;
+   int i;
+
+   list = p->delayed.next;
+
+   if (p->numDelayed > 3) {
+      for (i = 0; i < p->numDelayed; i += 3) {
+         list = list->next;
+      }
+   }
+
+   prev = list->prev;
+   for (; list != &p->delayed; list = prev, prev = list->prev) {
+
+      buf = DRMLISTENTRY(BBuf, list, head);
+
+      if (!signaled) {
+         if (wait) {
+            driFenceFinish(buf->fence, DRM_FENCE_TYPE_EXE, 1);
+            signaled = 1;
+         }
+         else {
+            signaled = driFenceSignaled(buf->fence, DRM_FENCE_TYPE_EXE);
+         }
+      }
+
+      if (!signaled)
+         break;
+
+      driFenceUnReference(buf->fence);
+      buf->fence = NULL;
+      DRMLISTDEL(list);
+      p->numDelayed--;
+      DRMLISTADD(list, &p->free);
+      p->numFree++;
+   }
+}
+
+static void *
+pool_create(struct _DriBufferPool *pool,
+            unsigned long size, unsigned flags, unsigned hint,
+            unsigned alignment)
+{
+   BPool *p = (BPool *) pool->data;
+
+   drmMMListHead *item;
+
+   if (alignment && (alignment != 4096))
+      return NULL;
+
+   _glthread_LOCK_MUTEX(p->mutex);
+
+   if (p->numFree == 0)
+      pool_checkFree(p, GL_TRUE);
+
+   if (p->numFree == 0) {
+      fprintf(stderr, "Out of fixed size buffer objects\n");
+      BM_CKFATAL(-ENOMEM);
+   }
+
+   item = p->free.next;
+
+   if (item == &p->free) {
+      fprintf(stderr, "Fixed size buffer pool corruption\n");
+   }
+
+   DRMLISTDEL(item);
+   --p->numFree;
+
+   _glthread_UNLOCK_MUTEX(p->mutex);
+   return (void *) DRMLISTENTRY(BBuf, item, head);
+}
+
+
+static int
+pool_destroy(struct _DriBufferPool *pool, void *private)
+{
+   BBuf *buf = (BBuf *) private;
+   BPool *p = buf->parent;
+
+   _glthread_LOCK_MUTEX(p->mutex);
+
+   if (buf->fence) {
+      DRMLISTADDTAIL(&buf->head, &p->delayed);
+      p->numDelayed++;
+   }
+   else {
+      buf->unfenced = 0;
+      DRMLISTADD(&buf->head, &p->free);
+      p->numFree++;
+   }
+
+   if ((p->numDelayed % p->checkDelayed) == 0)
+      pool_checkFree(p, 0);
+
+   _glthread_UNLOCK_MUTEX(p->mutex);
+   return 0;
+}
+
+
+static int
+pool_map(struct _DriBufferPool *pool, void *private, unsigned flags,
+         int hint, void **virtual)
+{
+
+   BBuf *buf = (BBuf *) private;
+   BPool *p = buf->parent;
+
+   _glthread_LOCK_MUTEX(p->mutex);
+
+   /*
+    * Currently Mesa doesn't have any condition variables to resolve this
+    * cleanly in a multithreading environment.
+    * We bail out instead.
+    */
+
+   if (buf->mapped) {
+      fprintf(stderr, "Trying to map already mapped buffer object\n");
+      BM_CKFATAL(-EINVAL);
+   }
+
+#if 0
+   if (buf->unfenced && !(hint & DRM_BO_HINT_ALLOW_UNFENCED_MAP)) {
+      fprintf(stderr, "Trying to map an unfenced buffer object 0x%08x"
+              " 0x%08x %d\n", hint, flags, buf->start);
+      BM_CKFATAL(-EINVAL);
+   }
+
+#endif
+
+   if (buf->fence) {
+      _glthread_UNLOCK_MUTEX(p->mutex);
+      return -EBUSY;
+   }
+
+   buf->mapped = GL_TRUE;
+   *virtual = (unsigned char *) p->virtual + buf->start;
+   _glthread_UNLOCK_MUTEX(p->mutex);
+   return 0;
+}
+
+static int
+pool_waitIdle(struct _DriBufferPool *pool, void *private, int lazy)
+{
+   BBuf *buf = (BBuf *) private;
+   driFenceFinish(buf->fence, 0x0, lazy);
+   return 0;
+}
+
+static int
+pool_unmap(struct _DriBufferPool *pool, void *private)
+{
+   BBuf *buf = (BBuf *) private;
+
+   buf->mapped = 0;
+   return 0;
+}
+
+static unsigned long
+pool_offset(struct _DriBufferPool *pool, void *private)
+{
+   BBuf *buf = (BBuf *) private;
+   BPool *p = buf->parent;
+
+   return p->kernelBO.offset + buf->start;
+}
+
+static unsigned
+pool_flags(struct _DriBufferPool *pool, void *private)
+{
+   BPool *p = (BPool *) pool->data;
+
+   return p->kernelBO.flags;
+}
+
+static unsigned long
+pool_size(struct _DriBufferPool *pool, void *private)
+{
+   BPool *p = (BPool *) pool->data;
+
+   return p->bufSize;
+}
+
+
+static int
+pool_fence(struct _DriBufferPool *pool, void *private,
+           struct _DriFenceObject *fence)
+{
+   BBuf *buf = (BBuf *) private;
+   BPool *p = buf->parent;
+
+   _glthread_LOCK_MUTEX(p->mutex);
+   if (buf->fence) {
+      driFenceUnReference(buf->fence);
+   }
+   buf->fence = fence;
+   buf->unfenced = 0;
+   driFenceReference(buf->fence);
+   _glthread_UNLOCK_MUTEX(p->mutex);
+
+   return 0;
+}
+
+static drmBO *
+pool_kernel(struct _DriBufferPool *pool, void *private)
+{
+   BBuf *buf = (BBuf *) private;
+   BPool *p = buf->parent;
+
+   return &p->kernelBO;
+}
+
+static int
+pool_validate(struct _DriBufferPool *pool, void *private)
+{
+   BBuf *buf = (BBuf *) private;
+   BPool *p = buf->parent;
+   _glthread_LOCK_MUTEX(p->mutex);
+   buf->unfenced = GL_TRUE;
+   _glthread_UNLOCK_MUTEX(p->mutex);
+   return 0;
+}
+
+static void
+pool_takedown(struct _DriBufferPool *pool)
+{
+   BPool *p = (BPool *) pool->data;
+
+   /*
+    * Wait on outstanding fences. 
+    */
+
+   _glthread_LOCK_MUTEX(p->mutex);
+   while ((p->numFree < p->numTot) && p->numDelayed) {
+      _glthread_UNLOCK_MUTEX(p->mutex);
+      sched_yield();
+      pool_checkFree(p, GL_TRUE);
+      _glthread_LOCK_MUTEX(p->mutex);
+   }
+
+   drmBODestroy(pool->fd, &p->kernelBO);
+   free(p->bufs);
+   _glthread_UNLOCK_MUTEX(p->mutex);
+   free(p);
+   free(pool);
+}
+
+
+struct _DriBufferPool *
+driBatchPoolInit(int fd, unsigned flags,
+                 unsigned long bufSize,
+                 unsigned numBufs, unsigned checkDelayed)
+{
+   struct _DriBufferPool *pool;
+
+   pool = (struct _DriBufferPool *) malloc(sizeof(*pool));
+   if (!pool)
+      return NULL;
+
+   pool->data = createBPool(fd, bufSize, numBufs, flags, checkDelayed);
+   if (!pool->data)
+      return NULL;
+
+   pool->fd = fd;
+   pool->map = &pool_map;
+   pool->unmap = &pool_unmap;
+   pool->destroy = &pool_destroy;
+   pool->offset = &pool_offset;
+   pool->flags = &pool_flags;
+   pool->size = &pool_size;
+   pool->create = &pool_create;
+   pool->fence = &pool_fence;
+   pool->kernel = &pool_kernel;
+   pool->validate = &pool_validate;
+   pool->waitIdle = &pool_waitIdle;
+   pool->setstatic = NULL;
+   pool->takeDown = &pool_takedown;
+   return pool;
+}
diff --git a/src/gallium/winsys/dri/intel/intel_batchpool.h b/src/gallium/winsys/dri/intel/intel_batchpool.h
new file mode 100644
index 0000000000..f6a95723bc
--- /dev/null
+++ b/src/gallium/winsys/dri/intel/intel_batchpool.h
@@ -0,0 +1,37 @@
+/**************************************************************************
+ * 
+ * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+#ifndef INTEL_BATCHPOOL_H
+#define INTEL_BATCHPOOL_H
+
+extern struct _DriBufferPool *driBatchPoolInit(int fd, unsigned flags,
+                                               unsigned long bufSize,
+                                               unsigned numBufs,
+                                               unsigned checkDelayed);
+
+
+#endif /* INTEL_BATCHPOOL_H */
diff --git a/src/gallium/winsys/dri/intel/intel_context.c b/src/gallium/winsys/dri/intel/intel_context.c
new file mode 100644
index 0000000000..c033f2a592
--- /dev/null
+++ b/src/gallium/winsys/dri/intel/intel_context.c
@@ -0,0 +1,304 @@
+/**************************************************************************
+ * 
+ * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+
+#include "i830_dri.h"
+
+#include "intel_screen.h"
+#include "intel_context.h"
+#include "intel_swapbuffers.h"
+#include "intel_winsys.h"
+#include "intel_batchbuffer.h"
+
+#include "state_tracker/st_public.h"
+#include "pipe/p_defines.h"
+#include "pipe/p_context.h"
+
+#include "utils.h"
+
+
+#ifdef DEBUG
+int __intel_debug = 0;
+#endif
+
+
+#define need_GL_ARB_multisample
+#define need_GL_ARB_point_parameters
+#define need_GL_ARB_texture_compression
+#define need_GL_ARB_vertex_buffer_object
+#define need_GL_ARB_vertex_program
+#define need_GL_ARB_window_pos
+#define need_GL_EXT_blend_color
+#define need_GL_EXT_blend_equation_separate
+#define need_GL_EXT_blend_func_separate
+#define need_GL_EXT_blend_minmax
+#define need_GL_EXT_cull_vertex
+#define need_GL_EXT_fog_coord
+#define need_GL_EXT_framebuffer_object
+#define need_GL_EXT_multi_draw_arrays
+#define need_GL_EXT_secondary_color
+#define need_GL_NV_vertex_program
+#include "extension_helper.h"
+
+
+/**
+ * Extension strings exported by the intel driver.
+ *
+ * \note
+ * It appears that ARB_texture_env_crossbar has "disappeared" compared to the
+ * old i830-specific driver.
+ */
+const struct dri_extension card_extensions[] = {
+   {"GL_ARB_multisample", GL_ARB_multisample_functions},
+   {"GL_ARB_multitexture", NULL},
+   {"GL_ARB_point_parameters", GL_ARB_point_parameters_functions},
+   {"GL_ARB_texture_border_clamp", NULL},
+   {"GL_ARB_texture_compression", GL_ARB_texture_compression_functions},
+   {"GL_ARB_texture_cube_map", NULL},
+   {"GL_ARB_texture_env_add", NULL},
+   {"GL_ARB_texture_env_combine", NULL},
+   {"GL_ARB_texture_env_dot3", NULL},
+   {"GL_ARB_texture_mirrored_repeat", NULL},
+   {"GL_ARB_texture_rectangle", NULL},
+   {"GL_ARB_vertex_buffer_object", GL_ARB_vertex_buffer_object_functions},
+   {"GL_ARB_pixel_buffer_object", NULL},
+   {"GL_ARB_vertex_program", GL_ARB_vertex_program_functions},
+   {"GL_ARB_window_pos", GL_ARB_window_pos_functions},
+   {"GL_EXT_blend_color", GL_EXT_blend_color_functions},
+   {"GL_EXT_blend_equation_separate", GL_EXT_blend_equation_separate_functions},
+   {"GL_EXT_blend_func_separate", GL_EXT_blend_func_separate_functions},
+   {"GL_EXT_blend_minmax", GL_EXT_blend_minmax_functions},
+   {"GL_EXT_blend_subtract", NULL},
+   {"GL_EXT_cull_vertex", GL_EXT_cull_vertex_functions},
+   {"GL_EXT_fog_coord", GL_EXT_fog_coord_functions},
+   {"GL_EXT_framebuffer_object", GL_EXT_framebuffer_object_functions},
+   {"GL_EXT_multi_draw_arrays", GL_EXT_multi_draw_arrays_functions},
+   {"GL_EXT_packed_depth_stencil", NULL},
+   {"GL_EXT_pixel_buffer_object", NULL},
+   {"GL_EXT_secondary_color", GL_EXT_secondary_color_functions},
+   {"GL_EXT_stencil_wrap", NULL},
+   {"GL_EXT_texture_edge_clamp", NULL},
+   {"GL_EXT_texture_env_combine", NULL},
+   {"GL_EXT_texture_env_dot3", NULL},
+   {"GL_EXT_texture_filter_anisotropic", NULL},
+   {"GL_EXT_texture_lod_bias", NULL},
+   {"GL_3DFX_texture_compression_FXT1", NULL},
+   {"GL_APPLE_client_storage", NULL},
+   {"GL_MESA_pack_invert", NULL},
+   {"GL_MESA_ycbcr_texture", NULL},
+   {"GL_NV_blend_square", NULL},
+   {"GL_NV_vertex_program", GL_NV_vertex_program_functions},
+   {"GL_NV_vertex_program1_1", NULL},
+   {"GL_SGIS_generate_mipmap", NULL },
+   {NULL, NULL}
+};
+
+
+
+#ifdef DEBUG
+static const struct dri_debug_control debug_control[] = {
+   {"ioctl", DEBUG_IOCTL},
+   {"bat", DEBUG_BATCH},
+   {"lock", DEBUG_LOCK},
+   {"swap", DEBUG_SWAP},
+   {NULL, 0}
+};
+#endif
+
+
+
+GLboolean
+intelCreateContext(const __GLcontextModes * visual,
+                   __DRIcontextPrivate * driContextPriv,
+                   void *sharedContextPrivate)
+{
+   struct intel_context *intel = CALLOC_STRUCT(intel_context);
+   __DRIscreenPrivate *sPriv = driContextPriv->driScreenPriv;
+   struct intel_screen *intelScreen = intel_screen(sPriv);
+   drmI830Sarea *saPriv = intelScreen->sarea;
+   int fthrottle_mode;
+   GLboolean havePools;
+   struct pipe_context *pipe;
+   struct st_context *st_share = NULL;
+
+   if (sharedContextPrivate) {
+      st_share = ((struct intel_context *) sharedContextPrivate)->st;
+   }
+
+   driContextPriv->driverPrivate = intel;
+   intel->intelScreen = intelScreen;
+   intel->driScreen = sPriv;
+   intel->sarea = saPriv;
+
+   driParseConfigFiles(&intel->optionCache, &intelScreen->optionCache,
+                       intel->driScreen->myNum, "i915");
+
+
+   /*
+    * memory pools
+    */
+   DRM_LIGHT_LOCK(sPriv->fd, &sPriv->pSAREA->lock, driContextPriv->hHWContext);
+   havePools = intelCreatePools(sPriv);
+   DRM_UNLOCK(sPriv->fd, &sPriv->pSAREA->lock, driContextPriv->hHWContext);
+   if (!havePools)
+      return GL_FALSE;
+
+
+   /* Dri stuff */
+   intel->hHWContext = driContextPriv->hHWContext;
+   intel->driFd = sPriv->fd;
+   intel->driHwLock = (drmLock *) & sPriv->pSAREA->lock;
+
+   fthrottle_mode = driQueryOptioni(&intel->optionCache, "fthrottle_mode");
+   intel->iw.irq_seq = -1;
+   intel->irqsEmitted = 0;
+
+   intel->batch = intel_batchbuffer_alloc(intel);
+   intel->last_swap_fence = NULL;
+   intel->first_swap_fence = NULL;
+
+#ifdef DEBUG
+   __intel_debug = driParseDebugString(getenv("INTEL_DEBUG"), debug_control);
+#endif
+
+   /*
+    * Pipe-related setup
+    */
+   if (!getenv("INTEL_HW")) {
+      pipe = intel_create_softpipe( intel, intelScreen->winsys );
+   }
+   else {
+      switch (intel->intelScreen->deviceID) {
+      case PCI_CHIP_I945_G:
+      case PCI_CHIP_I945_GM:
+      case PCI_CHIP_I945_GME:
+      case PCI_CHIP_G33_G:
+      case PCI_CHIP_Q33_G:
+      case PCI_CHIP_Q35_G:
+      case PCI_CHIP_I915_G:
+      case PCI_CHIP_I915_GM:
+	 pipe = intel_create_i915simple( intel, intelScreen->winsys );
+	 break;
+      default:
+	 fprintf(stderr, "Unknown PCIID %x in %s, using software driver\n", 
+                 intel->intelScreen->deviceID, __FUNCTION__);
+
+	 pipe = intel_create_softpipe( intel, intelScreen->winsys );
+	 break;
+      }
+   }
+
+   pipe->priv = intel;
+
+   intel->st = st_create_context(pipe, visual, st_share);
+
+   return GL_TRUE;
+}
+
+
+void
+intelDestroyContext(__DRIcontextPrivate * driContextPriv)
+{
+   struct intel_context *intel = intel_context(driContextPriv);
+
+   assert(intel);               /* should never be null */
+   if (intel) {
+      st_flush(intel->st, PIPE_FLUSH_WAIT);
+
+      intel_batchbuffer_free(intel->batch);
+
+      if (intel->last_swap_fence) {
+	 driFenceFinish(intel->last_swap_fence, DRM_FENCE_TYPE_EXE, GL_TRUE);
+	 driFenceUnReference(intel->last_swap_fence);
+	 intel->last_swap_fence = NULL;
+      }
+      if (intel->first_swap_fence) {
+	 driFenceFinish(intel->first_swap_fence, DRM_FENCE_TYPE_EXE, GL_TRUE);
+	 driFenceUnReference(intel->first_swap_fence);
+	 intel->first_swap_fence = NULL;
+      }
+
+      if (intel->intelScreen->dummyContext == intel)
+         intel->intelScreen->dummyContext = NULL;
+
+      st_destroy_context(intel->st);
+      free(intel);
+   }
+}
+
+
+GLboolean
+intelUnbindContext(__DRIcontextPrivate * driContextPriv)
+{
+   struct intel_context *intel = intel_context(driContextPriv);
+   st_flush(intel->st, 0x0);
+   /* XXX make_current(NULL)? */
+   return GL_TRUE;
+}
+
+
+GLboolean
+intelMakeCurrent(__DRIcontextPrivate * driContextPriv,
+                 __DRIdrawablePrivate * driDrawPriv,
+                 __DRIdrawablePrivate * driReadPriv)
+{
+   if (driContextPriv) {
+      struct intel_context *intel = intel_context(driContextPriv);
+      struct intel_framebuffer *draw_fb = intel_framebuffer(driDrawPriv);
+      struct intel_framebuffer *read_fb = intel_framebuffer(driReadPriv);
+
+      assert(draw_fb->stfb);
+      assert(read_fb->stfb);
+
+      /* This is for situations in which we need a rendering context but
+       * there may not be any currently bound.
+       */
+      intel->intelScreen->dummyContext = intel;
+
+      st_make_current(intel->st, draw_fb->stfb, read_fb->stfb);
+
+      if ((intel->driDrawable != driDrawPriv) ||
+	  (intel->lastStamp != driDrawPriv->lastStamp)) {
+         intel->driDrawable = driDrawPriv;
+         intelUpdateWindowSize(driDrawPriv);
+         intel->lastStamp = driDrawPriv->lastStamp;
+      }
+
+      /* The size of the draw buffer will have been updated above.
+       * If the readbuffer is a different window, check/update its size now.
+       */
+      if (driReadPriv != driDrawPriv) {
+         intelUpdateWindowSize(driReadPriv);
+      }
+
+   }
+   else {
+      st_make_current(NULL, NULL, NULL);
+   }
+
+   return GL_TRUE;
+}
diff --git a/src/gallium/winsys/dri/intel/intel_context.h b/src/gallium/winsys/dri/intel/intel_context.h
new file mode 100644
index 0000000000..b01370c049
--- /dev/null
+++ b/src/gallium/winsys/dri/intel/intel_context.h
@@ -0,0 +1,158 @@
+/**************************************************************************
+ * 
+ * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+#ifndef INTEL_CONTEXT_H
+#define INTEL_CONTEXT_H
+
+
+#include "drm.h"
+#include "intel_screen.h"
+#include "i915_drm.h"
+
+
+struct pipe_context;
+struct intel_context;
+struct _DriBufferObject;
+struct st_context;
+
+
+#define INTEL_MAX_FIXUP 64
+
+/**
+ * Intel rendering context, contains a state tracker and intel-specific info.
+ */
+struct intel_context
+{
+   struct st_context *st;
+
+   struct _DriFenceObject *last_swap_fence;
+   struct _DriFenceObject *first_swap_fence;
+
+   struct intel_batchbuffer *batch;
+
+   boolean locked;
+   char *prevLockFile;
+   int prevLockLine;
+
+   uint irqsEmitted;
+   drm_i915_irq_wait_t iw;
+
+   drm_context_t hHWContext;
+   drmLock *driHwLock;
+   int driFd;
+
+   __DRIdrawablePrivate *driDrawable;
+   __DRIscreenPrivate *driScreen;
+   struct intel_screen *intelScreen;
+   drmI830Sarea *sarea;
+
+   uint lastStamp;
+
+   /**
+    * Configuration cache
+    */
+   driOptionCache optionCache;
+};
+
+
+
+/**
+ * Intel framebuffer.
+ */
+struct intel_framebuffer
+{
+   struct st_framebuffer *stfb;
+
+   /* other fields TBD */
+   int other;
+};
+
+
+
+
+/* These are functions now:
+ */
+void LOCK_HARDWARE( struct intel_context *intel );
+void UNLOCK_HARDWARE( struct intel_context *intel );
+
+extern char *__progname;
+
+
+
+/* ================================================================
+ * Debugging:
+ */
+#ifdef DEBUG
+extern int __intel_debug;
+
+#define DEBUG_SWAP	0x1
+#define DEBUG_LOCK      0x2
+#define DEBUG_IOCTL	0x4
+#define DEBUG_BATCH     0x8
+
+#define DBG(flag, ...)  do { 			\
+   if (__intel_debug & (DEBUG_##flag)) 		\
+      printf(__VA_ARGS__); 		\
+} while(0)
+
+#else
+#define DBG(flag, ...) 
+#endif
+
+
+
+#define PCI_CHIP_845_G			0x2562
+#define PCI_CHIP_I830_M			0x3577
+#define PCI_CHIP_I855_GM		0x3582
+#define PCI_CHIP_I865_G			0x2572
+#define PCI_CHIP_I915_G			0x2582
+#define PCI_CHIP_I915_GM		0x2592
+#define PCI_CHIP_I945_G			0x2772
+#define PCI_CHIP_I945_GM		0x27A2
+#define PCI_CHIP_I945_GME		0x27AE
+#define PCI_CHIP_G33_G			0x29C2
+#define PCI_CHIP_Q35_G			0x29B2
+#define PCI_CHIP_Q33_G			0x29D2
+
+
+/** Cast wrapper */
+static INLINE struct intel_context *
+intel_context(__DRIcontextPrivate *driContextPriv)
+{
+   return (struct intel_context *) driContextPriv->driverPrivate;
+}
+
+
+/** Cast wrapper */
+static INLINE struct intel_framebuffer *
+intel_framebuffer(__DRIdrawablePrivate * driDrawPriv)
+{
+   return (struct intel_framebuffer *) driDrawPriv->driverPrivate;
+}
+
+
+#endif
diff --git a/src/gallium/winsys/dri/intel/intel_lock.c b/src/gallium/winsys/dri/intel/intel_lock.c
new file mode 100644
index 0000000000..70aa7ea5f4
--- /dev/null
+++ b/src/gallium/winsys/dri/intel/intel_lock.c
@@ -0,0 +1,102 @@
+/**************************************************************************
+ * 
+ * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+
+#include "main/glheader.h"
+#include "glapi/glthread.h"
+#include <GL/internal/glcore.h>
+#include "state_tracker/st_public.h"
+#include "intel_context.h"
+#include "i830_dri.h"
+
+
+
+_glthread_DECLARE_STATIC_MUTEX( lockMutex );
+
+
+static void
+intelContendedLock(struct intel_context *intel, uint flags)
+{
+   __DRIdrawablePrivate *dPriv = intel->driDrawable;
+   __DRIscreenPrivate *sPriv = intel->driScreen;
+   struct intel_screen *intelScreen = intel_screen(sPriv);
+   drmI830Sarea *sarea = intel->sarea;
+
+   drmGetLock(intel->driFd, intel->hHWContext, flags);
+
+   DBG(LOCK, "%s - got contended lock\n", __progname);
+
+   /* If the window moved, may need to set a new cliprect now.
+    *
+    * NOTE: This releases and regains the hw lock, so all state
+    * checking must be done *after* this call:
+    */
+   if (dPriv)
+      DRI_VALIDATE_DRAWABLE_INFO(sPriv, dPriv);
+
+   if (sarea->width != intelScreen->front.width ||
+       sarea->height != intelScreen->front.height) {
+
+      intelUpdateScreenRotation(sPriv, sarea);
+   }
+}
+
+
+/* Lock the hardware and validate our state.
+ */
+void LOCK_HARDWARE( struct intel_context *intel )
+{
+    char __ret = 0;
+
+    _glthread_LOCK_MUTEX(lockMutex);
+    assert(!intel->locked);
+
+    DRM_CAS(intel->driHwLock, intel->hHWContext,
+            (DRM_LOCK_HELD|intel->hHWContext), __ret);
+
+    if (__ret)
+       intelContendedLock( intel, 0 );
+
+    DBG(LOCK, "%s - locked\n", __progname);
+
+    intel->locked = 1;
+}
+
+
+/* Unlock the hardware using the global current context 
+ */
+void UNLOCK_HARDWARE( struct intel_context *intel )
+{
+   assert(intel->locked);
+   intel->locked = 0;
+
+   DRM_UNLOCK(intel->driFd, intel->driHwLock, intel->hHWContext);
+
+   _glthread_UNLOCK_MUTEX(lockMutex);
+
+   DBG(LOCK, "%s - unlocked\n", __progname);
+} 
diff --git a/src/gallium/winsys/dri/intel/intel_reg.h b/src/gallium/winsys/dri/intel/intel_reg.h
new file mode 100644
index 0000000000..f37c24fda9
--- /dev/null
+++ b/src/gallium/winsys/dri/intel/intel_reg.h
@@ -0,0 +1,53 @@
+/**************************************************************************
+ * 
+ * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+
+#ifndef _INTEL_REG_H_
+#define _INTEL_REG_H_
+
+
+#define BR00_BITBLT_CLIENT   0x40000000
+#define BR00_OP_COLOR_BLT    0x10000000
+#define BR00_OP_SRC_COPY_BLT 0x10C00000
+#define BR13_SOLID_PATTERN   0x80000000
+
+#define XY_COLOR_BLT_CMD		((2<<29)|(0x50<<22)|0x4)
+#define XY_COLOR_BLT_WRITE_ALPHA	(1<<21)
+#define XY_COLOR_BLT_WRITE_RGB		(1<<20)
+
+#define XY_SRC_COPY_BLT_CMD             ((2<<29)|(0x53<<22)|6)
+#define XY_SRC_COPY_BLT_WRITE_ALPHA     (1<<21)
+#define XY_SRC_COPY_BLT_WRITE_RGB       (1<<20)
+
+#define MI_WAIT_FOR_EVENT               ((0x3<<23))
+#define MI_WAIT_FOR_PLANE_B_FLIP        (1<<6)
+#define MI_WAIT_FOR_PLANE_A_FLIP        (1<<2)
+
+#define MI_BATCH_BUFFER_END 	        (0xA<<23)
+
+
+#endif
diff --git a/src/gallium/winsys/dri/intel/intel_screen.c b/src/gallium/winsys/dri/intel/intel_screen.c
new file mode 100644
index 0000000000..9e31c013a9
--- /dev/null
+++ b/src/gallium/winsys/dri/intel/intel_screen.c
@@ -0,0 +1,537 @@
+/**************************************************************************
+ * 
+ * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+#include "utils.h"
+#include "vblank.h"
+#include "xmlpool.h"
+
+#include "intel_context.h"
+#include "intel_screen.h"
+#include "intel_batchbuffer.h"
+#include "intel_batchpool.h"
+#include "intel_swapbuffers.h"
+#include "intel_winsys.h"
+
+#include "i830_dri.h"
+#include "dri_bufpool.h"
+
+#include "pipe/p_context.h"
+#include "state_tracker/st_public.h"
+#include "state_tracker/st_cb_fbo.h"
+
+
+
+PUBLIC const char __driConfigOptions[] =
+   DRI_CONF_BEGIN DRI_CONF_SECTION_PERFORMANCE
+   DRI_CONF_FTHROTTLE_MODE(DRI_CONF_FTHROTTLE_IRQS)
+   DRI_CONF_VBLANK_MODE(DRI_CONF_VBLANK_DEF_INTERVAL_0)
+   DRI_CONF_SECTION_END DRI_CONF_SECTION_QUALITY
+   DRI_CONF_FORCE_S3TC_ENABLE(false)
+   DRI_CONF_ALLOW_LARGE_TEXTURES(1)
+   DRI_CONF_SECTION_END DRI_CONF_END;
+
+const uint __driNConfigOptions = 4;
+
+#ifdef USE_NEW_INTERFACE
+static PFNGLXCREATECONTEXTMODES create_context_modes = NULL;
+#endif /*USE_NEW_INTERFACE */
+
+extern const struct dri_extension card_extensions[];
+
+
+
+
+static void
+intelPrintDRIInfo(struct intel_screen * intelScreen,
+                  __DRIscreenPrivate * sPriv, I830DRIPtr gDRIPriv)
+{
+   fprintf(stderr, "*** Front size:   0x%x  offset: 0x%x  pitch: %d\n",
+           intelScreen->front.size, intelScreen->front.offset,
+           intelScreen->front.pitch);
+   fprintf(stderr, "*** Memory : 0x%x\n", gDRIPriv->mem);
+}
+
+
+#if 0
+static void
+intelPrintSAREA(const drmI830Sarea * sarea)
+{
+   fprintf(stderr, "SAREA: sarea width %d  height %d\n", sarea->width,
+           sarea->height);
+   fprintf(stderr, "SAREA: pitch: %d\n", sarea->pitch);
+   fprintf(stderr,
+           "SAREA: front offset: 0x%08x  size: 0x%x  handle: 0x%x\n",
+           sarea->front_offset, sarea->front_size,
+           (unsigned) sarea->front_handle);
+   fprintf(stderr,
+           "SAREA: back  offset: 0x%08x  size: 0x%x  handle: 0x%x\n",
+           sarea->back_offset, sarea->back_size,
+           (unsigned) sarea->back_handle);
+   fprintf(stderr, "SAREA: depth offset: 0x%08x  size: 0x%x  handle: 0x%x\n",
+           sarea->depth_offset, sarea->depth_size,
+           (unsigned) sarea->depth_handle);
+   fprintf(stderr, "SAREA: tex   offset: 0x%08x  size: 0x%x  handle: 0x%x\n",
+           sarea->tex_offset, sarea->tex_size, (unsigned) sarea->tex_handle);
+   fprintf(stderr, "SAREA: rotation: %d\n", sarea->rotation);
+   fprintf(stderr,
+           "SAREA: rotated offset: 0x%08x  size: 0x%x\n",
+           sarea->rotated_offset, sarea->rotated_size);
+   fprintf(stderr, "SAREA: rotated pitch: %d\n", sarea->rotated_pitch);
+}
+#endif
+
+
+/**
+ * Use the information in the sarea to update the screen parameters
+ * related to screen rotation. Needs to be called locked.
+ */
+void
+intelUpdateScreenRotation(__DRIscreenPrivate * sPriv, drmI830Sarea * sarea)
+{
+   struct intel_screen *intelScreen = intel_screen(sPriv);
+
+   if (intelScreen->front.map) {
+      drmUnmap(intelScreen->front.map, intelScreen->front.size);
+      intelScreen->front.map = NULL;
+   }
+
+   if (intelScreen->front.buffer)
+      driDeleteBuffers(1, &intelScreen->front.buffer);
+
+   intelScreen->front.width = sarea->width;
+   intelScreen->front.height = sarea->height;
+   intelScreen->front.offset = sarea->front_offset;
+   intelScreen->front.pitch = sarea->pitch * intelScreen->front.cpp;
+   intelScreen->front.size = sarea->front_size;
+   intelScreen->front.handle = sarea->front_handle;
+
+   assert( sarea->front_size >=
+	   intelScreen->front.pitch * intelScreen->front.height );
+
+   if (!sarea->front_handle)
+      return;
+
+   if (drmMap(sPriv->fd,
+	      sarea->front_handle,
+	      intelScreen->front.size,
+	      (drmAddress *) & intelScreen->front.map) != 0) {
+      fprintf(stderr, "drmMap(frontbuffer) failed!\n");
+      return;
+   }
+
+   if (intelScreen->staticPool) {
+      driGenBuffers(intelScreen->staticPool, "static region", 1,
+		    &intelScreen->front.buffer, 64,
+		    DRM_BO_FLAG_MEM_TT | DRM_BO_FLAG_NO_MOVE |
+		    DRM_BO_FLAG_READ | DRM_BO_FLAG_WRITE, 0);
+      
+      driBOSetStatic(intelScreen->front.buffer, 
+		     intelScreen->front.offset, 		  
+		     intelScreen->front.pitch * intelScreen->front.height, 
+		     intelScreen->front.map, 0);
+   }
+}
+
+
+boolean
+intelCreatePools(__DRIscreenPrivate * sPriv)
+{
+   unsigned batchPoolSize = 1024*1024;
+   struct intel_screen *intelScreen = intel_screen(sPriv);
+
+   if (intelScreen->havePools)
+      return GL_TRUE;
+
+   intelScreen->staticPool = driDRMStaticPoolInit(sPriv->fd);
+   if (!intelScreen->staticPool)
+      return GL_FALSE;
+
+   batchPoolSize /= BATCH_SZ;
+   intelScreen->batchPool = driBatchPoolInit(sPriv->fd,
+                                             DRM_BO_FLAG_EXE |
+                                             DRM_BO_FLAG_MEM_TT |
+                                             DRM_BO_FLAG_MEM_LOCAL,
+                                             BATCH_SZ, 
+					     batchPoolSize, 5);
+   if (!intelScreen->batchPool) {
+      fprintf(stderr, "Failed to initialize batch pool - possible incorrect agpgart installed\n");
+      return GL_FALSE;
+   }
+   
+   intelScreen->havePools = GL_TRUE;
+
+   intelUpdateScreenRotation(sPriv, intelScreen->sarea);
+
+   return GL_TRUE;
+}
+
+
+static boolean
+intelInitDriver(__DRIscreenPrivate * sPriv)
+{
+   struct intel_screen *intelScreen;
+   I830DRIPtr gDRIPriv = (I830DRIPtr) sPriv->pDevPriv;
+
+   PFNGLXSCRENABLEEXTENSIONPROC glx_enable_extension =
+      (PFNGLXSCRENABLEEXTENSIONPROC) (*dri_interface->
+                                      getProcAddress("glxEnableExtension"));
+   void *const psc = sPriv->psc->screenConfigs;
+
+   if (sPriv->devPrivSize != sizeof(I830DRIRec)) {
+      fprintf(stderr,
+              "\nERROR!  sizeof(I830DRIRec) does not match passed size from device driver\n");
+      return GL_FALSE;
+   }
+
+   /* Allocate the private area */
+   intelScreen = CALLOC_STRUCT(intel_screen);
+   if (!intelScreen) 
+      return GL_FALSE;
+
+   /* parse information in __driConfigOptions */
+   driParseOptionInfo(&intelScreen->optionCache,
+                      __driConfigOptions, __driNConfigOptions);
+
+   sPriv->private = (void *) intelScreen;
+
+   intelScreen->sarea = (drmI830Sarea *) (((GLubyte *) sPriv->pSAREA) +
+					  gDRIPriv->sarea_priv_offset);
+   intelScreen->deviceID = gDRIPriv->deviceID;
+   intelScreen->front.cpp = gDRIPriv->cpp;
+   intelScreen->drmMinor = sPriv->drmMinor;
+
+   assert(gDRIPriv->bitsPerPixel == 16 ||
+	  gDRIPriv->bitsPerPixel == 32);
+
+   intelUpdateScreenRotation(sPriv, intelScreen->sarea);
+
+   if (0)
+      intelPrintDRIInfo(intelScreen, sPriv, gDRIPriv);
+
+   if (glx_enable_extension != NULL) {
+      (*glx_enable_extension) (psc, "GLX_SGI_swap_control");
+      (*glx_enable_extension) (psc, "GLX_SGI_video_sync");
+      (*glx_enable_extension) (psc, "GLX_MESA_swap_control");
+      (*glx_enable_extension) (psc, "GLX_MESA_swap_frame_usage");
+      (*glx_enable_extension) (psc, "GLX_SGI_make_current_read");
+   }
+
+   intelScreen->winsys = intel_create_pipe_winsys(sPriv->fd);
+
+   return GL_TRUE;
+}
+
+
+static void
+intelDestroyScreen(__DRIscreenPrivate * sPriv)
+{
+   struct intel_screen *intelScreen = intel_screen(sPriv);
+
+   /*  intelUnmapScreenRegions(intelScreen); */
+
+   if (intelScreen->havePools) {
+      driPoolTakeDown(intelScreen->staticPool);
+      driPoolTakeDown(intelScreen->batchPool);
+   }
+   FREE(intelScreen);
+   sPriv->private = NULL;
+}
+
+
+/**
+ * This is called when we need to set up GL rendering to a new X window.
+ */
+static boolean
+intelCreateBuffer(__DRIscreenPrivate * driScrnPriv,
+                  __DRIdrawablePrivate * driDrawPriv,
+                  const __GLcontextModes * visual, boolean isPixmap)
+{
+   if (isPixmap) {
+      return GL_FALSE;          /* not implemented */
+   }
+   else {
+      enum pipe_format colorFormat, depthFormat, stencilFormat;
+      struct intel_framebuffer *intelfb = CALLOC_STRUCT(intel_framebuffer);
+
+      if (!intelfb)
+         return GL_FALSE;
+
+      if (visual->redBits == 5)
+         colorFormat = PIPE_FORMAT_R5G6B5_UNORM;
+      else
+         colorFormat = PIPE_FORMAT_A8R8G8B8_UNORM;
+
+      if (visual->depthBits == 16)
+         depthFormat = PIPE_FORMAT_Z16_UNORM;
+      else if (visual->depthBits == 24)
+         depthFormat = PIPE_FORMAT_S8Z24_UNORM;
+      else
+         depthFormat = PIPE_FORMAT_NONE;
+
+      if (visual->stencilBits == 8)
+         stencilFormat = PIPE_FORMAT_S8Z24_UNORM;
+      else
+         stencilFormat = PIPE_FORMAT_NONE;
+
+      intelfb->stfb = st_create_framebuffer(visual,
+                                            colorFormat,
+                                            depthFormat,
+                                            stencilFormat,
+                                            driDrawPriv->w,
+                                            driDrawPriv->h,
+                                            (void*) intelfb);
+      if (!intelfb->stfb) {
+         free(intelfb);
+         return GL_FALSE;
+      }
+
+      driDrawPriv->driverPrivate = (void *) intelfb;
+      return GL_TRUE;
+   }
+}
+
+static void
+intelDestroyBuffer(__DRIdrawablePrivate * driDrawPriv)
+{
+   struct intel_framebuffer *intelfb = intel_framebuffer(driDrawPriv);
+   assert(intelfb->stfb);
+   st_unreference_framebuffer(&intelfb->stfb);
+   free(intelfb);
+}
+
+
+/**
+ * Get information about previous buffer swaps.
+ */
+static int
+intelGetSwapInfo(__DRIdrawablePrivate * dPriv, __DRIswapInfo * sInfo)
+{
+   if ((dPriv == NULL) || (dPriv->driverPrivate == NULL)
+       || (sInfo == NULL)) {
+      return -1;
+   }
+
+   return 0;
+}
+
+
+static void
+intelSetTexOffset(__DRIcontext *pDRICtx, int texname,
+		  unsigned long long offset, int depth, uint pitch)
+{
+   abort();
+#if 0
+   struct intel_context *intel = (struct intel_context*)
+      ((__DRIcontextPrivate*)pDRICtx->private)->driverPrivate;
+   struct gl_texture_object *tObj = _mesa_lookup_texture(&intel->ctx, texname);
+   struct st_texture_object *stObj = st_texture_object(tObj);
+
+   if (!stObj)
+      return;
+
+   if (stObj->pt)
+      st->pipe->texture_release(intel->st->pipe, &stObj->pt);
+
+   stObj->imageOverride = GL_TRUE;
+   stObj->depthOverride = depth;
+   stObj->pitchOverride = pitch;
+
+   if (offset)
+      stObj->textureOffset = offset;
+#endif
+}
+
+
+static const struct __DriverAPIRec intelAPI = {
+   .InitDriver = intelInitDriver,
+   .DestroyScreen = intelDestroyScreen,
+   .CreateContext = intelCreateContext,
+   .DestroyContext = intelDestroyContext,
+   .CreateBuffer = intelCreateBuffer,
+   .DestroyBuffer = intelDestroyBuffer,
+   .SwapBuffers = intelSwapBuffers,
+   .MakeCurrent = intelMakeCurrent,
+   .UnbindContext = intelUnbindContext,
+   .GetSwapInfo = intelGetSwapInfo,
+   .GetMSC = driGetMSC32,
+   .WaitForMSC = driWaitForMSC32,
+   .WaitForSBC = NULL,
+   .SwapBuffersMSC = NULL,
+   .CopySubBuffer = intelCopySubBuffer,
+   .setTexOffset = intelSetTexOffset,
+};
+
+
+static __GLcontextModes *
+intelFillInModes(unsigned pixel_bits, unsigned depth_bits,
+                 unsigned stencil_bits, boolean have_back_buffer)
+{
+   __GLcontextModes *modes;
+   __GLcontextModes *m;
+   unsigned num_modes;
+   unsigned depth_buffer_factor;
+   unsigned back_buffer_factor;
+   GLenum fb_format;
+   GLenum fb_type;
+
+   /* GLX_SWAP_COPY_OML is only supported because the Intel driver doesn't
+    * support pageflipping at all.
+    */
+   static const GLenum back_buffer_modes[] = {
+      GLX_NONE, GLX_SWAP_UNDEFINED_OML, GLX_SWAP_COPY_OML
+   };
+
+   u_int8_t depth_bits_array[3];
+   u_int8_t stencil_bits_array[3];
+
+
+   depth_bits_array[0] = 0;
+   depth_bits_array[1] = depth_bits;
+   depth_bits_array[2] = depth_bits;
+
+   /* Just like with the accumulation buffer, always provide some modes
+    * with a stencil buffer.  It will be a sw fallback, but some apps won't
+    * care about that.
+    */
+   stencil_bits_array[0] = 0;
+   stencil_bits_array[1] = 0;
+   if (depth_bits == 24)
+      stencil_bits_array[1] = (stencil_bits == 0) ? 8 : stencil_bits;
+
+   stencil_bits_array[2] = (stencil_bits == 0) ? 8 : stencil_bits;
+
+   depth_buffer_factor = ((depth_bits != 0) || (stencil_bits != 0)) ? 3 : 1;
+   back_buffer_factor = (have_back_buffer) ? 3 : 1;
+
+   num_modes = depth_buffer_factor * back_buffer_factor * 4;
+
+   if (pixel_bits == 16) {
+      fb_format = GL_RGB;
+      fb_type = GL_UNSIGNED_SHORT_5_6_5;
+   }
+   else {
+      fb_format = GL_BGRA;
+      fb_type = GL_UNSIGNED_INT_8_8_8_8_REV;
+   }
+
+   modes =
+      (*dri_interface->createContextModes) (num_modes,
+                                            sizeof(__GLcontextModes));
+   m = modes;
+   if (!driFillInModes(&m, fb_format, fb_type,
+                       depth_bits_array, stencil_bits_array,
+                       depth_buffer_factor, back_buffer_modes,
+                       back_buffer_factor, GLX_TRUE_COLOR)) {
+      fprintf(stderr, "[%s:%u] Error creating FBConfig!\n", __func__,
+              __LINE__);
+      return NULL;
+   }
+   if (!driFillInModes(&m, fb_format, fb_type,
+                       depth_bits_array, stencil_bits_array,
+                       depth_buffer_factor, back_buffer_modes,
+                       back_buffer_factor, GLX_DIRECT_COLOR)) {
+      fprintf(stderr, "[%s:%u] Error creating FBConfig!\n", __func__,
+              __LINE__);
+      return NULL;
+   }
+
+   /* Mark the visual as slow if there are "fake" stencil bits.
+    */
+   for (m = modes; m != NULL; m = m->next) {
+      if ((m->stencilBits != 0) && (m->stencilBits != stencil_bits)) {
+         m->visualRating = GLX_SLOW_CONFIG;
+      }
+   }
+
+   return modes;
+}
+
+
+/**
+ * This is the bootstrap function for the driver.  libGL supplies all of the
+ * requisite information about the system, and the driver initializes itself.
+ * This routine also fills in the linked list pointed to by \c driver_modes
+ * with the \c __GLcontextModes that the driver can support for windows or
+ * pbuffers.
+ * 
+ * \return A pointer to a \c __DRIscreenPrivate on success, or \c NULL on 
+ *         failure.
+ */
+PUBLIC void *
+__driCreateNewScreen_20050727(__DRInativeDisplay * dpy, int scrn,
+                              __DRIscreen * psc,
+                              const __GLcontextModes * modes,
+                              const __DRIversion * ddx_version,
+                              const __DRIversion * dri_version,
+                              const __DRIversion * drm_version,
+                              const __DRIframebuffer * frame_buffer,
+                              drmAddress pSAREA, int fd,
+                              int internal_api_version,
+                              const __DRIinterfaceMethods * interface,
+                              __GLcontextModes ** driver_modes)
+{
+   __DRIscreenPrivate *psp;
+   static const __DRIversion ddx_expected = { 1, 7, 0 };
+   static const __DRIversion dri_expected = { 4, 0, 0 };
+   static const __DRIversion drm_expected = { 1, 7, 0 };
+
+   dri_interface = interface;
+
+   if (!driCheckDriDdxDrmVersions2("i915",
+                                   dri_version, &dri_expected,
+                                   ddx_version, &ddx_expected,
+                                   drm_version, &drm_expected)) {
+      return NULL;
+   }
+
+   psp = __driUtilCreateNewScreen(dpy, scrn, psc, NULL,
+                                  ddx_version, dri_version, drm_version,
+                                  frame_buffer, pSAREA, fd,
+                                  internal_api_version, &intelAPI);
+
+   if (psp != NULL) {
+      I830DRIPtr dri_priv = (I830DRIPtr) psp->pDevPriv;
+      *driver_modes = intelFillInModes(dri_priv->cpp * 8,
+                                       (dri_priv->cpp == 2) ? 16 : 24,
+                                       (dri_priv->cpp == 2) ? 0 : 8, 1);
+
+      /* Calling driInitExtensions here, with a NULL context pointer,
+       * does not actually enable the extensions.  It just makes sure
+       * that all the dispatch offsets for all the extensions that
+       * *might* be enables are known.  This is needed because the
+       * dispatch offsets need to be known when _mesa_context_create
+       * is called, but we can't enable the extensions until we have a
+       * context pointer.
+       *
+       * Hello chicken.  Hello egg.  How are you two today?
+       */
+      driInitExtensions(NULL, card_extensions, GL_FALSE);
+   }
+
+   return (void *) psp;
+}
+
diff --git a/src/gallium/winsys/dri/intel/intel_screen.h b/src/gallium/winsys/dri/intel/intel_screen.h
new file mode 100644
index 0000000000..3396f9e564
--- /dev/null
+++ b/src/gallium/winsys/dri/intel/intel_screen.h
@@ -0,0 +1,113 @@
+/**************************************************************************
+ * 
+ * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+#ifndef _INTEL_SCREEN_H_
+#define _INTEL_SCREEN_H_
+
+#include "dri_util.h"
+#include "i830_common.h"
+#include "xmlconfig.h"
+#include "dri_bufpool.h"
+
+#include "pipe/p_compiler.h"
+
+
+struct intel_screen
+{
+   struct {
+      drm_handle_t handle;
+
+      /* We create a static dri buffer for the frontbuffer.
+       */
+      struct _DriBufferObject *buffer;
+
+      char *map;                   /* memory map */
+      int offset;                  /* from start of video mem, in bytes */
+      int pitch;                   /* row stride, in bytes */
+      int width;
+      int height;
+      int size;
+      int cpp;                     /* for front and back buffers */   
+   } front;
+
+   int deviceID;
+   int drmMinor;
+
+   drmI830Sarea *sarea;
+
+   /**
+   * Configuration cache with default values for all contexts
+   */
+   driOptionCache optionCache;
+
+   struct _DriBufferPool *batchPool;
+   struct _DriBufferPool *staticPool; /** for the X screen/framebuffer */
+   boolean havePools;
+
+   /**
+    * Temporary(?) context to use for SwapBuffers or other situations in
+    * which we need a rendering context, but none is currently bound.
+    */
+   struct intel_context *dummyContext;
+
+   struct pipe_winsys *winsys;
+};
+
+
+
+/** cast wrapper */
+static INLINE struct intel_screen *
+intel_screen(__DRIscreenPrivate *sPriv)
+{
+   return (struct intel_screen *) sPriv->private;
+}
+
+
+extern void
+intelUpdateScreenRotation(__DRIscreenPrivate * sPriv, drmI830Sarea * sarea);
+
+
+extern void intelDestroyContext(__DRIcontextPrivate * driContextPriv);
+
+extern boolean intelUnbindContext(__DRIcontextPrivate * driContextPriv);
+
+extern boolean
+intelMakeCurrent(__DRIcontextPrivate * driContextPriv,
+                 __DRIdrawablePrivate * driDrawPriv,
+                 __DRIdrawablePrivate * driReadPriv);
+
+
+extern boolean
+intelCreatePools(__DRIscreenPrivate *sPriv);
+
+extern boolean
+intelCreateContext(const __GLcontextModes * visual,
+                   __DRIcontextPrivate * driContextPriv,
+                   void *sharedContextPrivate);
+
+
+#endif
diff --git a/src/gallium/winsys/dri/intel/intel_swapbuffers.c b/src/gallium/winsys/dri/intel/intel_swapbuffers.c
new file mode 100644
index 0000000000..56b86d6a63
--- /dev/null
+++ b/src/gallium/winsys/dri/intel/intel_swapbuffers.c
@@ -0,0 +1,253 @@
+/**************************************************************************
+ * 
+ * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+#include "intel_screen.h"
+#include "intel_context.h"
+#include "intel_swapbuffers.h"
+#include "intel_batchbuffer.h"
+#include "intel_reg.h"
+#include "intel_winsys.h"
+
+#include "pipe/p_context.h"
+#include "state_tracker/st_public.h"
+#include "state_tracker/st_context.h"
+#include "state_tracker/st_cb_fbo.h"
+
+
+/**
+ * Display a colorbuffer surface in an X window.
+ * Used for SwapBuffers and flushing front buffer rendering.
+ *
+ * \param dPriv  the window/drawable to display into
+ * \param surf  the surface to display
+ * \param rect  optional subrect of surface to display (may be NULL).
+ */
+void
+intelDisplaySurface(__DRIdrawablePrivate *dPriv,
+                    struct pipe_surface *surf,
+                    const drm_clip_rect_t *rect)
+{
+   struct intel_screen *intelScreen = intel_screen(dPriv->driScreenPriv);
+   struct intel_context *intel = intelScreen->dummyContext;
+
+   DBG(SWAP, "%s\n", __FUNCTION__);
+
+   if (!intel) {
+      /* XXX this is where some kind of extra/meta context could be useful */
+      return;
+   }
+
+   if (intel->last_swap_fence) {
+      driFenceFinish(intel->last_swap_fence, DRM_FENCE_TYPE_EXE, TRUE);
+      driFenceUnReference(intel->last_swap_fence);
+      intel->last_swap_fence = NULL;
+   }
+   intel->last_swap_fence = intel->first_swap_fence;
+   intel->first_swap_fence = NULL;
+
+   /* The LOCK_HARDWARE is required for the cliprects.  Buffer offsets
+    * should work regardless.
+    */
+   LOCK_HARDWARE(intel);
+   /* if this drawable isn't currently bound the LOCK_HARDWARE done on the
+    * current context (which is what intelScreenContext should return) might
+    * not get a contended lock and thus cliprects not updated (tests/manywin)
+    */
+   if (intel_context(dPriv->driContextPriv) != intel)
+      DRI_VALIDATE_DRAWABLE_INFO(intel->driScreen, dPriv);
+
+
+   if (dPriv && dPriv->numClipRects) {
+      const int srcWidth = surf->width;
+      const int srcHeight = surf->height;
+      const int nbox = dPriv->numClipRects;
+      const drm_clip_rect_t *pbox = dPriv->pClipRects;
+      const int pitch = intelScreen->front.pitch / intelScreen->front.cpp;
+      const int cpp = intelScreen->front.cpp;
+      const int srcpitch = surf->pitch;
+      int BR13, CMD;
+      int i;
+
+      ASSERT(surf->buffer);
+      ASSERT(surf->cpp == cpp);
+
+      DBG(SWAP, "screen pitch %d  src surface pitch %d\n",
+	  pitch, surf->pitch);
+
+      if (cpp == 2) {
+	 BR13 = (pitch * cpp) | (0xCC << 16) | (1 << 24);
+	 CMD = XY_SRC_COPY_BLT_CMD;
+      }
+      else {
+	 BR13 = (pitch * cpp) | (0xCC << 16) | (1 << 24) | (1 << 25);
+	 CMD = (XY_SRC_COPY_BLT_CMD | XY_SRC_COPY_BLT_WRITE_ALPHA |
+		XY_SRC_COPY_BLT_WRITE_RGB);
+      }
+
+      for (i = 0; i < nbox; i++, pbox++) {
+	 drm_clip_rect_t box;
+	 drm_clip_rect_t sbox;
+
+	 if (pbox->x1 > pbox->x2 ||
+	     pbox->y1 > pbox->y2 ||
+	     pbox->x2 > intelScreen->front.width || 
+	     pbox->y2 > intelScreen->front.height) {
+            /* invalid cliprect, skip it */
+	    continue;
+         }
+
+	 box = *pbox;
+
+	 if (rect) {
+            /* intersect cliprect with user-provided src rect */
+	    drm_clip_rect_t rrect;
+
+	    rrect.x1 = dPriv->x + rect->x1;
+	    rrect.y1 = (dPriv->h - rect->y1 - rect->y2) + dPriv->y;
+	    rrect.x2 = rect->x2 + rrect.x1;
+	    rrect.y2 = rect->y2 + rrect.y1;
+	    if (rrect.x1 > box.x1)
+	       box.x1 = rrect.x1;
+	    if (rrect.y1 > box.y1)
+	       box.y1 = rrect.y1;
+	    if (rrect.x2 < box.x2)
+	       box.x2 = rrect.x2;
+	    if (rrect.y2 < box.y2)
+	       box.y2 = rrect.y2;
+
+	    if (box.x1 > box.x2 || box.y1 > box.y2)
+	       continue;
+	 }
+
+	 /* restrict blit to size of actually rendered area */
+	 if (box.x2 - box.x1 > srcWidth)
+	    box.x2 = srcWidth + box.x1;
+	 if (box.y2 - box.y1 > srcHeight)
+	    box.y2 = srcHeight + box.y1;
+
+	 DBG(SWAP, "box x1 x2 y1 y2 %d %d %d %d\n",
+	     box.x1, box.x2, box.y1, box.y2);
+
+	 sbox.x1 = box.x1 - dPriv->x;
+	 sbox.y1 = box.y1 - dPriv->y;
+
+         assert(box.x1 < box.x2);
+         assert(box.y1 < box.y2);
+
+         /* XXX this could be done with pipe->surface_copy() */
+	 BEGIN_BATCH(8, INTEL_BATCH_NO_CLIPRECTS);
+	 OUT_BATCH(CMD);
+	 OUT_BATCH(BR13);
+	 OUT_BATCH((box.y1 << 16) | box.x1);
+	 OUT_BATCH((box.y2 << 16) | box.x2);
+
+	 OUT_RELOC(intelScreen->front.buffer, 
+		   DRM_BO_FLAG_MEM_TT | DRM_BO_FLAG_WRITE,
+		   DRM_BO_MASK_MEM | DRM_BO_FLAG_WRITE, 0);
+	 OUT_BATCH((sbox.y1 << 16) | sbox.x1);
+	 OUT_BATCH((srcpitch * cpp) & 0xffff);
+	 OUT_RELOC(dri_bo(surf->buffer),
+                   DRM_BO_FLAG_MEM_TT | DRM_BO_FLAG_READ,
+		   DRM_BO_MASK_MEM | DRM_BO_FLAG_READ, 0);
+
+	 ADVANCE_BATCH();
+      }
+
+      if (intel->first_swap_fence)
+	 driFenceUnReference(intel->first_swap_fence);
+      intel->first_swap_fence = intel_batchbuffer_flush(intel->batch);
+      driFenceReference(intel->first_swap_fence);
+   }
+
+   UNLOCK_HARDWARE(intel);
+
+   if (intel->lastStamp != dPriv->lastStamp) {
+      intelUpdateWindowSize(dPriv);
+      intel->lastStamp = dPriv->lastStamp;
+   }
+}
+
+
+
+/**
+ * This will be called whenever the currently bound window is moved/resized.
+ */
+void
+intelUpdateWindowSize(__DRIdrawablePrivate *dPriv)
+{
+   struct intel_framebuffer *intelfb = intel_framebuffer(dPriv);
+   assert(intelfb->stfb);
+   st_resize_framebuffer(intelfb->stfb, dPriv->w, dPriv->h);
+}
+
+
+
+void
+intelSwapBuffers(__DRIdrawablePrivate * dPriv)
+{
+   struct intel_framebuffer *intel_fb = intel_framebuffer(dPriv);
+   struct pipe_surface *back_surf;
+
+   assert(intel_fb);
+   assert(intel_fb->stfb);
+
+   back_surf = st_get_framebuffer_surface(intel_fb->stfb,
+                                          ST_SURFACE_BACK_LEFT);
+   if (back_surf) {
+      st_notify_swapbuffers(intel_fb->stfb);
+      intelDisplaySurface(dPriv, back_surf, NULL);
+      st_notify_swapbuffers_complete(intel_fb->stfb);
+   }
+}
+
+
+/**
+ * Called via glXCopySubBufferMESA() to copy a subrect of the back
+ * buffer to the front buffer/screen.
+ */
+void
+intelCopySubBuffer(__DRIdrawablePrivate * dPriv, int x, int y, int w, int h)
+{
+   struct intel_framebuffer *intel_fb = intel_framebuffer(dPriv);
+   struct pipe_surface *back_surf;
+
+   assert(intel_fb);
+   assert(intel_fb->stfb);
+
+   back_surf = st_get_framebuffer_surface(intel_fb->stfb,
+                                          ST_SURFACE_BACK_LEFT);
+   if (back_surf) {
+      drm_clip_rect_t rect;
+      rect.x1 = x;
+      rect.y1 = y;
+      rect.x2 = w;
+      rect.y2 = h;
+
+      st_notify_swapbuffers(intel_fb->stfb);
+      intelDisplaySurface(dPriv, back_surf, &rect);
+   }
+}
diff --git a/src/gallium/winsys/dri/intel/intel_swapbuffers.h b/src/gallium/winsys/dri/intel/intel_swapbuffers.h
new file mode 100644
index 0000000000..7ae5fd15a5
--- /dev/null
+++ b/src/gallium/winsys/dri/intel/intel_swapbuffers.h
@@ -0,0 +1,47 @@
+/**************************************************************************
+ * 
+ * Copyright 2006 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+#ifndef INTEL_SWAPBUFFERS_H
+#define INTEL_SWAPBUFFERS_H
+
+
+struct pipe_surface;
+
+
+extern void intelDisplaySurface(__DRIdrawablePrivate * dPriv,
+                                struct pipe_surface *surf,
+                                const drm_clip_rect_t * rect);
+
+extern void intelSwapBuffers(__DRIdrawablePrivate * dPriv);
+
+extern void intelCopySubBuffer(__DRIdrawablePrivate * dPriv,
+                               int x, int y, int w, int h);
+
+extern void intelUpdateWindowSize(__DRIdrawablePrivate *dPriv);
+
+
+#endif /* INTEL_SWAPBUFFERS_H */
diff --git a/src/gallium/winsys/dri/intel/intel_winsys.h b/src/gallium/winsys/dri/intel/intel_winsys.h
new file mode 100644
index 0000000000..ffc40782be
--- /dev/null
+++ b/src/gallium/winsys/dri/intel/intel_winsys.h
@@ -0,0 +1,73 @@
+/**************************************************************************
+ * 
+ * Copyright 2006 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+#ifndef INTEL_WINSYS_H
+#define INTEL_WINSYS_H
+
+#include "pipe/p_state.h"
+
+struct intel_context;
+struct pipe_context;
+struct pipe_winsys;
+struct pipe_buffer;
+struct _DriBufferObject;
+
+struct pipe_winsys *
+intel_create_pipe_winsys( int fd );
+
+void
+intel_destroy_pipe_winsys( struct pipe_winsys *winsys );
+
+struct pipe_context *
+intel_create_softpipe( struct intel_context *intel,
+                       struct pipe_winsys *winsys );
+
+struct pipe_context *
+intel_create_i915simple( struct intel_context *intel,
+                       struct pipe_winsys *winsys );
+
+
+struct intel_buffer {
+   struct pipe_buffer base;
+   struct _DriBufferObject *driBO;
+};
+
+static INLINE struct intel_buffer *
+intel_buffer( struct pipe_buffer *buf )
+{
+   return (struct intel_buffer *)buf;
+}
+
+static INLINE struct _DriBufferObject *
+dri_bo( struct pipe_buffer *buf )
+{
+   return intel_buffer(buf)->driBO;
+}
+
+
+
+#endif
diff --git a/src/gallium/winsys/dri/intel/intel_winsys_i915.c b/src/gallium/winsys/dri/intel/intel_winsys_i915.c
new file mode 100644
index 0000000000..1ba6a9e1b2
--- /dev/null
+++ b/src/gallium/winsys/dri/intel/intel_winsys_i915.c
@@ -0,0 +1,154 @@
+/**************************************************************************
+ * 
+ * Copyright 2006 Tungsten Graphics, Inc., Bismarck, ND., USA
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDERS, AUTHORS AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM,
+ * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR 
+ * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE 
+ * USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * 
+ **************************************************************************/
+/*
+ * Authors: Keith Whitwell <keithw-at-tungstengraphics-dot-com>
+ */
+
+#include <stdlib.h>
+#include <xf86drm.h>
+#include "dri_bufpool.h"
+#include "dri_bufmgr.h"
+
+#include "intel_context.h"
+#include "intel_batchbuffer.h"
+#include "intel_winsys.h"
+
+#include "pipe/p_util.h"
+#include "pipe/i915simple/i915_winsys.h"
+
+
+struct intel_i915_winsys {
+   struct i915_winsys winsys;   /**< batch buffer funcs */
+   struct intel_context *intel;
+};
+
+
+/* Turn a i915simple winsys into an intel/i915simple winsys:
+ */
+static inline struct intel_i915_winsys *
+intel_i915_winsys( struct i915_winsys *sws )
+{
+   return (struct intel_i915_winsys *)sws;
+}
+
+
+/* Simple batchbuffer interface:
+ */
+
+static unsigned *intel_i915_batch_start( struct i915_winsys *sws,
+					 unsigned dwords,
+					 unsigned relocs )
+{
+   struct intel_context *intel = intel_i915_winsys(sws)->intel;
+
+   /* XXX: check relocs. 
+    */
+   if (intel_batchbuffer_space( intel->batch ) >= dwords * 4) {
+      /* XXX: Hmm, the driver can't really do much with this pointer: 
+       */
+      return (unsigned *)intel->batch->ptr;	
+   }
+   else 
+      return NULL;
+}
+
+static void intel_i915_batch_dword( struct i915_winsys *sws,
+				    unsigned dword )
+{
+   struct intel_context *intel = intel_i915_winsys(sws)->intel;
+   intel_batchbuffer_emit_dword( intel->batch, dword );
+}
+
+static void intel_i915_batch_reloc( struct i915_winsys *sws,
+			     struct pipe_buffer *buf,
+			     unsigned access_flags,
+			     unsigned delta )
+{
+   struct intel_context *intel = intel_i915_winsys(sws)->intel;
+   unsigned flags = DRM_BO_FLAG_MEM_TT;
+   unsigned mask = DRM_BO_MASK_MEM;
+
+   if (access_flags & I915_BUFFER_ACCESS_WRITE) {
+      flags |= DRM_BO_FLAG_WRITE;
+      mask |= DRM_BO_FLAG_WRITE;
+   }
+
+   if (access_flags & I915_BUFFER_ACCESS_READ) {
+      flags |= DRM_BO_FLAG_READ;
+      mask |= DRM_BO_FLAG_READ;
+   }
+
+   intel_batchbuffer_emit_reloc( intel->batch, 
+				 dri_bo( buf ),
+				 flags, mask, 
+				 delta );
+}
+
+
+
+static void intel_i915_batch_flush( struct i915_winsys *sws )
+{
+   struct intel_context *intel = intel_i915_winsys(sws)->intel;
+
+   intel_batchbuffer_flush( intel->batch );
+//   if (0) intel_i915_batch_wait_idle( sws );
+}
+
+
+static void intel_i915_batch_finish( struct i915_winsys *sws )
+{
+   struct intel_context *intel = intel_i915_winsys(sws)->intel;
+   intel_batchbuffer_finish( intel->batch );
+}
+
+
+/**
+ * Create i915 hardware rendering context.
+ */
+struct pipe_context *
+intel_create_i915simple( struct intel_context *intel,
+                         struct pipe_winsys *winsys )
+{
+   struct intel_i915_winsys *iws = CALLOC_STRUCT( intel_i915_winsys );
+   
+   /* Fill in this struct with callbacks that i915simple will need to
+    * communicate with the window system, buffer manager, etc. 
+    */
+   iws->winsys.batch_start = intel_i915_batch_start;
+   iws->winsys.batch_dword = intel_i915_batch_dword;
+   iws->winsys.batch_reloc = intel_i915_batch_reloc;
+   iws->winsys.batch_flush = intel_i915_batch_flush;
+   iws->winsys.batch_finish = intel_i915_batch_finish;
+   iws->intel = intel;
+
+   /* Create the i915simple context:
+    */
+   return i915_create( winsys,
+		       &iws->winsys,
+		       intel->intelScreen->deviceID );
+}
diff --git a/src/gallium/winsys/dri/intel/intel_winsys_pipe.c b/src/gallium/winsys/dri/intel/intel_winsys_pipe.c
new file mode 100644
index 0000000000..789a386500
--- /dev/null
+++ b/src/gallium/winsys/dri/intel/intel_winsys_pipe.c
@@ -0,0 +1,302 @@
+/**************************************************************************
+ * 
+ * Copyright 2006 Tungsten Graphics, Inc., Bismarck, ND., USA
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDERS, AUTHORS AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM,
+ * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR 
+ * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE 
+ * USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * 
+ **************************************************************************/
+/*
+ * Authors: Keith Whitwell <keithw-at-tungstengraphics-dot-com>
+ */
+
+#include <stdlib.h>
+#include <xf86drm.h>
+#include "dri_bufpool.h"
+#include "dri_bufmgr.h"
+
+#include "intel_context.h"
+#include "intel_winsys.h"
+#include "intel_swapbuffers.h"
+#include "intel_batchbuffer.h"
+
+#include "pipe/p_winsys.h"
+#include "pipe/p_defines.h"
+#include "pipe/p_state.h"
+#include "pipe/p_util.h"
+#include "pipe/p_inlines.h"
+
+
+
+struct intel_pipe_winsys {
+   struct pipe_winsys winsys;
+   struct _DriBufferPool *regionPool;
+};
+
+
+
+/* Turn a pipe winsys into an intel/pipe winsys:
+ */
+static inline struct intel_pipe_winsys *
+intel_pipe_winsys( struct pipe_winsys *winsys )
+{
+   return (struct intel_pipe_winsys *)winsys;
+}
+
+
+/* Most callbacks map direcly onto dri_bufmgr operations:
+ */
+static void *intel_buffer_map(struct pipe_winsys *winsys, 
+			      struct pipe_buffer *buf,
+			      unsigned flags )
+{
+   unsigned drm_flags = 0;
+   
+   if (flags & PIPE_BUFFER_USAGE_CPU_WRITE)
+      drm_flags |= DRM_BO_FLAG_WRITE;
+
+   if (flags & PIPE_BUFFER_USAGE_CPU_READ)
+      drm_flags |= DRM_BO_FLAG_READ;
+
+   return driBOMap( dri_bo(buf), drm_flags, 0 );
+}
+
+static void intel_buffer_unmap(struct pipe_winsys *winsys, 
+			       struct pipe_buffer *buf)
+{
+   driBOUnmap( dri_bo(buf) );
+}
+
+
+static void
+intel_buffer_destroy(struct pipe_winsys *winsys,
+		     struct pipe_buffer *buf)
+{
+   driBOUnReference( dri_bo(buf) );
+}
+
+
+/* Pipe has no concept of pools.  We choose the tex/region pool
+ * for all buffers.
+ * Grabs the hardware lock!
+ */
+static struct pipe_buffer *
+intel_buffer_create(struct pipe_winsys *winsys, 
+                    unsigned alignment, 
+                    unsigned usage, 
+                    unsigned size )
+{
+   struct intel_buffer *buffer = CALLOC_STRUCT( intel_buffer );
+   struct intel_pipe_winsys *iws = intel_pipe_winsys(winsys);
+   unsigned flags = 0;
+
+   buffer->base.refcount = 1;
+   buffer->base.alignment = alignment;
+   buffer->base.usage = usage;
+   buffer->base.size = size;
+
+   if (usage & (PIPE_BUFFER_USAGE_VERTEX /*| IWS_BUFFER_USAGE_LOCAL*/)) {
+      flags |= DRM_BO_FLAG_MEM_LOCAL | DRM_BO_FLAG_CACHED;
+   } else {
+      flags |= DRM_BO_FLAG_MEM_VRAM | DRM_BO_FLAG_MEM_TT;
+   }
+
+   if (usage & PIPE_BUFFER_USAGE_GPU_READ)
+      flags |= DRM_BO_FLAG_READ;
+
+   if (usage & PIPE_BUFFER_USAGE_GPU_WRITE)
+      flags |= DRM_BO_FLAG_WRITE;
+
+   /* drm complains if we don't set any read/write flags.
+    */
+   if ((flags & (DRM_BO_FLAG_READ | DRM_BO_FLAG_WRITE)) == 0)
+      flags |= DRM_BO_FLAG_READ | DRM_BO_FLAG_WRITE;
+
+#if 0
+   if (flags & IWS_BUFFER_USAGE_EXE)
+      flags |= DRM_BO_FLAG_EXE;
+
+   if (usage & IWS_BUFFER_USAGE_CACHED)
+      flags |= DRM_BO_FLAG_CACHED;
+#endif
+
+   driGenBuffers( iws->regionPool, 
+		  "pipe buffer", 1, &buffer->driBO, alignment, flags, 0 );
+
+   driBOData( buffer->driBO, size, NULL, 0 );
+
+   return &buffer->base;
+}
+
+
+static struct pipe_buffer *
+intel_user_buffer_create(struct pipe_winsys *winsys, void *ptr, unsigned bytes)
+{
+   struct intel_buffer *buffer = CALLOC_STRUCT( intel_buffer );
+   struct intel_pipe_winsys *iws = intel_pipe_winsys(winsys);
+
+   driGenUserBuffer( iws->regionPool, 
+                     "pipe user buffer", &buffer->driBO, ptr, bytes);
+
+   return &buffer->base;
+}
+
+
+/* The state tracker (should!) keep track of whether the fake
+ * frontbuffer has been touched by any rendering since the last time
+ * we copied its contents to the real frontbuffer.  Our task is easy:
+ */
+static void
+intel_flush_frontbuffer( struct pipe_winsys *winsys,
+                         struct pipe_surface *surf,
+                         void *context_private)
+{
+   struct intel_context *intel = (struct intel_context *) context_private;
+   __DRIdrawablePrivate *dPriv = intel->driDrawable;
+
+   intelDisplaySurface(dPriv, surf, NULL);
+}
+
+
+static struct pipe_surface *
+intel_i915_surface_alloc(struct pipe_winsys *winsys)
+{
+   struct pipe_surface *surf = CALLOC_STRUCT(pipe_surface);
+   if (surf) {
+      surf->refcount = 1;
+      surf->winsys = winsys;
+   }
+   return surf;
+}
+
+
+/**
+ * Round n up to next multiple.
+ */
+static INLINE unsigned
+round_up(unsigned n, unsigned multiple)
+{
+   return (n + multiple - 1) & ~(multiple - 1);
+}
+
+/**
+ * Copied from xm_winsys.c
+ */
+static int
+intel_i915_surface_alloc_storage(struct pipe_winsys *winsys,
+                                 struct pipe_surface *surf,
+                                 unsigned width, unsigned height,
+                                 enum pipe_format format, 
+                                 unsigned flags)
+{
+   const unsigned alignment = 64;
+   int ret;
+
+   surf->width = width;
+   surf->height = height;
+   surf->format = format;
+   surf->cpp = pf_get_size(format);
+   surf->pitch = round_up(width, alignment / surf->cpp);
+
+   assert(!surf->buffer);
+   surf->buffer = winsys->buffer_create(winsys, alignment,
+                                        PIPE_BUFFER_USAGE_PIXEL,
+                                        surf->pitch * surf->cpp * height);
+   if(!surf->buffer)
+      return -1;
+
+   return 0;
+}
+
+
+static void
+intel_i915_surface_release(struct pipe_winsys *winsys, struct pipe_surface **s)
+{
+   struct pipe_surface *surf = *s;
+   surf->refcount--;
+   if (surf->refcount == 0) {
+      if (surf->buffer)
+	 pipe_buffer_reference(winsys, &surf->buffer, NULL);
+      free(surf);
+   }
+   *s = NULL;
+}
+
+
+
+static void
+intel_printf( struct pipe_winsys *winsys, const char *fmtString, ... )
+{
+   va_list args;
+   va_start( args, fmtString );  
+   vfprintf(stderr, fmtString, args);
+   va_end( args );
+}
+
+static const char *
+intel_get_name( struct pipe_winsys *winsys )
+{
+   return "Intel/DRI/ttm";
+}
+
+
+struct pipe_winsys *
+intel_create_pipe_winsys( int fd )
+{
+   struct intel_pipe_winsys *iws = CALLOC_STRUCT( intel_pipe_winsys );
+   
+   /* Fill in this struct with callbacks that pipe will need to
+    * communicate with the window system, buffer manager, etc. 
+    *
+    * Pipe would be happy with a malloc based memory manager, but
+    * the SwapBuffers implementation in this winsys driver requires
+    * that rendering be done to an appropriate _DriBufferObject.  
+    */
+   iws->winsys.buffer_create = intel_buffer_create;
+   iws->winsys.user_buffer_create = intel_user_buffer_create;
+   iws->winsys.buffer_map = intel_buffer_map;
+   iws->winsys.buffer_unmap = intel_buffer_unmap;
+   iws->winsys.buffer_destroy = intel_buffer_destroy;
+   iws->winsys.flush_frontbuffer = intel_flush_frontbuffer;
+   iws->winsys.printf = intel_printf;
+   iws->winsys.get_name = intel_get_name;
+   iws->winsys.surface_alloc = intel_i915_surface_alloc;
+   iws->winsys.surface_alloc_storage = intel_i915_surface_alloc_storage;
+   iws->winsys.surface_release = intel_i915_surface_release;
+
+   if (fd)
+      iws->regionPool = driDRMPoolInit(fd);
+
+   return &iws->winsys;
+}
+
+
+void
+intel_destroy_pipe_winsys( struct pipe_winsys *winsys )
+{
+   struct intel_pipe_winsys *iws = intel_pipe_winsys(winsys);
+   if (iws->regionPool) {
+      driPoolTakeDown(iws->regionPool);
+   }
+   free(iws);
+}
+
diff --git a/src/gallium/winsys/dri/intel/intel_winsys_softpipe.c b/src/gallium/winsys/dri/intel/intel_winsys_softpipe.c
new file mode 100644
index 0000000000..cec3437831
--- /dev/null
+++ b/src/gallium/winsys/dri/intel/intel_winsys_softpipe.c
@@ -0,0 +1,81 @@
+/**************************************************************************
+ * 
+ * Copyright 2006 Tungsten Graphics, Inc., Bismarck, ND., USA
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDERS, AUTHORS AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM,
+ * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR 
+ * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE 
+ * USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * 
+ **************************************************************************/
+/*
+ * Authors: Keith Whitwell <keithw-at-tungstengraphics-dot-com>
+ */
+
+#include "intel_context.h"
+#include "intel_winsys.h"
+#include "pipe/p_defines.h"
+#include "pipe/p_util.h"
+#include "pipe/p_format.h"
+#include "pipe/softpipe/sp_winsys.h"
+
+
+struct intel_softpipe_winsys {
+   struct softpipe_winsys sws;
+   struct intel_context *intel;
+};
+
+/**
+ * Return list of surface formats supported by this driver.
+ */
+static boolean
+intel_is_format_supported(struct softpipe_winsys *sws,
+                          enum pipe_format format)
+{
+   switch(format) {
+   case PIPE_FORMAT_A8R8G8B8_UNORM:
+   case PIPE_FORMAT_R5G6B5_UNORM:
+   case PIPE_FORMAT_S8Z24_UNORM:
+      return TRUE;
+   default:
+      return FALSE;
+   }
+}
+
+
+/**
+ * Create rendering context which uses software rendering.
+ */
+struct pipe_context *
+intel_create_softpipe( struct intel_context *intel,
+                       struct pipe_winsys *winsys )
+{
+   struct intel_softpipe_winsys *isws = CALLOC_STRUCT( intel_softpipe_winsys );
+   
+   /* Fill in this struct with callbacks that softpipe will need to
+    * communicate with the window system, buffer manager, etc. 
+    */
+   isws->sws.is_format_supported = intel_is_format_supported;
+   isws->intel = intel;
+
+   /* Create the softpipe context:
+    */
+   return softpipe_create( winsys, &isws->sws );
+}
diff --git a/src/gallium/winsys/dri/intel/server/i830_common.h b/src/gallium/winsys/dri/intel/server/i830_common.h
new file mode 100644
index 0000000000..d4d58886ce
--- /dev/null
+++ b/src/gallium/winsys/dri/intel/server/i830_common.h
@@ -0,0 +1,226 @@
+/**************************************************************************
+
+Copyright 2001 VA Linux Systems Inc., Fremont, California.
+Copyright 2002 Tungsten Graphics Inc., Cedar Park, Texas.
+
+All Rights Reserved.
+
+Permission is hereby granted, free of charge, to any person obtaining a
+copy of this software and associated documentation files (the "Software"),
+to deal in the Software without restriction, including without limitation
+on the rights to use, copy, modify, merge, publish, distribute, sub
+license, and/or sell copies of the Software, and to permit persons to whom
+the Software is furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice (including the next
+paragraph) shall be included in all copies or substantial portions of the
+Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
+ATI, VA LINUX SYSTEMS AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
+DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
+OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
+USE OR OTHER DEALINGS IN THE SOFTWARE.
+
+**************************************************************************/
+
+/* $XFree86: xc/programs/Xserver/hw/xfree86/drivers/i810/i830_common.h,v 1.1 2002/09/11 00:29:32 dawes Exp $ */
+
+#ifndef _I830_COMMON_H_
+#define _I830_COMMON_H_
+
+
+#define I830_NR_TEX_REGIONS 255	/* maximum due to use of chars for next/prev */
+#define I830_LOG_MIN_TEX_REGION_SIZE 14
+
+
+/* Driver specific DRM command indices
+ * NOTE: these are not OS specific, but they are driver specific
+ */
+#define DRM_I830_INIT                     0x00
+#define DRM_I830_FLUSH                    0x01
+#define DRM_I830_FLIP                     0x02
+#define DRM_I830_BATCHBUFFER              0x03
+#define DRM_I830_IRQ_EMIT                 0x04
+#define DRM_I830_IRQ_WAIT                 0x05
+#define DRM_I830_GETPARAM                 0x06
+#define DRM_I830_SETPARAM                 0x07
+#define DRM_I830_ALLOC                    0x08
+#define DRM_I830_FREE                     0x09
+#define DRM_I830_INIT_HEAP                0x0a
+#define DRM_I830_CMDBUFFER                0x0b
+#define DRM_I830_DESTROY_HEAP             0x0c
+#define DRM_I830_SET_VBLANK_PIPE          0x0d
+#define DRM_I830_GET_VBLANK_PIPE          0x0e
+
+typedef struct {
+   enum {
+      I830_INIT_DMA = 0x01,
+      I830_CLEANUP_DMA = 0x02,
+      I830_RESUME_DMA = 0x03
+   } func;
+   unsigned int mmio_offset;
+   int sarea_priv_offset;
+   unsigned int ring_start;
+   unsigned int ring_end;
+   unsigned int ring_size;
+   unsigned int front_offset;
+   unsigned int back_offset;
+   unsigned int depth_offset;
+   unsigned int w;
+   unsigned int h;
+   unsigned int pitch;
+   unsigned int pitch_bits;
+   unsigned int back_pitch;
+   unsigned int depth_pitch;
+   unsigned int cpp;
+   unsigned int chipset;
+} drmI830Init;
+
+typedef struct {
+	drmTextureRegion texList[I830_NR_TEX_REGIONS+1];
+        int last_upload;	/* last time texture was uploaded */
+        int last_enqueue;	/* last time a buffer was enqueued */
+	int last_dispatch;	/* age of the most recently dispatched buffer */
+	int ctxOwner;		/* last context to upload state */
+	int texAge;
+        int pf_enabled;		/* is pageflipping allowed? */
+        int pf_active;               
+        int pf_current_page;	/* which buffer is being displayed? */
+        int perf_boxes;	        /* performance boxes to be displayed */   
+	int width, height;      /* screen size in pixels */
+
+	drm_handle_t front_handle;
+	int front_offset;
+	int front_size;
+
+	drm_handle_t back_handle;
+	int back_offset;
+	int back_size;
+
+	drm_handle_t depth_handle;
+	int depth_offset;
+	int depth_size;
+
+	drm_handle_t tex_handle;
+	int tex_offset;
+	int tex_size;
+	int log_tex_granularity;
+	int pitch;
+	int rotation;           /* 0, 90, 180 or 270 */
+	int rotated_offset;
+	int rotated_size;
+	int rotated_pitch;
+	int virtualX, virtualY;
+
+	unsigned int front_tiled;
+	unsigned int back_tiled;
+	unsigned int depth_tiled;
+	unsigned int rotated_tiled;
+	unsigned int rotated2_tiled;
+
+	int pipeA_x;
+	int pipeA_y;
+	int pipeA_w;
+	int pipeA_h;
+	int pipeB_x;
+	int pipeB_y;
+	int pipeB_w;
+	int pipeB_h;
+
+	/* Triple buffering */
+	drm_handle_t third_handle;
+	int third_offset;
+	int third_size;
+	unsigned int third_tiled;
+} drmI830Sarea;
+
+/* Flags for perf_boxes
+ */
+#define I830_BOX_RING_EMPTY    0x1 /* populated by kernel */
+#define I830_BOX_FLIP          0x2 /* populated by kernel */
+#define I830_BOX_WAIT          0x4 /* populated by kernel & client */
+#define I830_BOX_TEXTURE_LOAD  0x8 /* populated by kernel */
+#define I830_BOX_LOST_CONTEXT  0x10 /* populated by client */
+
+
+typedef struct {
+   	int start;		/* agp offset */
+	int used;		/* nr bytes in use */
+	int DR1;		/* hw flags for GFX_OP_DRAWRECT_INFO */
+        int DR4;		/* window origin for GFX_OP_DRAWRECT_INFO*/
+	int num_cliprects;	/* mulitpass with multiple cliprects? */
+        drm_clip_rect_t *cliprects; /* pointer to userspace cliprects */
+} drmI830BatchBuffer;
+
+typedef struct {
+   	char *buf;		/* agp offset */
+	int sz; 		/* nr bytes in use */
+	int DR1;		/* hw flags for GFX_OP_DRAWRECT_INFO */
+        int DR4;		/* window origin for GFX_OP_DRAWRECT_INFO*/
+	int num_cliprects;	/* mulitpass with multiple cliprects? */
+        drm_clip_rect_t *cliprects; /* pointer to userspace cliprects */
+} drmI830CmdBuffer;
+ 
+typedef struct {
+	int *irq_seq;
+} drmI830IrqEmit;
+
+typedef struct {
+	int irq_seq;
+} drmI830IrqWait;
+
+typedef struct {
+	int param;
+	int *value;
+} drmI830GetParam;
+
+#define I830_PARAM_IRQ_ACTIVE     1
+#define I830_PARAM_ALLOW_BATCHBUFFER   2 
+
+typedef struct {
+	int param;
+	int value;
+} drmI830SetParam;
+
+#define I830_SETPARAM_USE_MI_BATCHBUFFER_START  1
+#define I830_SETPARAM_TEX_LRU_LOG_GRANULARITY   2
+#define I830_SETPARAM_ALLOW_BATCHBUFFER         3
+
+
+/* A memory manager for regions of shared memory:
+ */
+#define I830_MEM_REGION_AGP 1
+
+typedef struct {
+	int region;
+	int alignment;
+	int size;
+	int *region_offset;	/* offset from start of fb or agp */
+} drmI830MemAlloc;
+
+typedef struct {
+	int region;
+	int region_offset;
+} drmI830MemFree;
+
+typedef struct {
+	int region;
+	int size;
+	int start;	
+} drmI830MemInitHeap;
+
+typedef struct {
+	int region;
+} drmI830MemDestroyHeap;
+
+#define DRM_I830_VBLANK_PIPE_A  1
+#define DRM_I830_VBLANK_PIPE_B  2
+
+typedef struct {
+        int pipe;
+} drmI830VBlankPipe;
+
+#endif /* _I830_DRM_H_ */
diff --git a/src/gallium/winsys/dri/intel/server/i830_dri.h b/src/gallium/winsys/dri/intel/server/i830_dri.h
new file mode 100644
index 0000000000..c2a3af8cbf
--- /dev/null
+++ b/src/gallium/winsys/dri/intel/server/i830_dri.h
@@ -0,0 +1,63 @@
+/* $XFree86: xc/programs/Xserver/hw/xfree86/drivers/i810/i830_dri.h,v 1.4 2002/10/30 12:52:18 alanh Exp $ */
+
+#ifndef _I830_DRI_H
+#define _I830_DRI_H
+
+#include "xf86drm.h"
+#include "i830_common.h"
+
+#define I830_MAX_DRAWABLES 256
+
+#define I830_MAJOR_VERSION 1
+#define I830_MINOR_VERSION 7
+#define I830_PATCHLEVEL 2
+
+#define I830_REG_SIZE 0x80000
+
+typedef struct _I830DRIRec {
+   drm_handle_t regs;
+   drmSize regsSize;
+
+   drmSize unused1; /* backbufferSize */
+   drm_handle_t unused2; /* backbuffer */
+
+   drmSize unused3; /* depthbufferSize */
+   drm_handle_t unused4; /* depthbuffer */
+
+   drmSize unused5; /* rotatedSize */
+   drm_handle_t unused6; /* rotatedbuffer */
+
+   drm_handle_t unused7; /* textures */
+   int unused8; /* textureSize */
+
+   drm_handle_t unused9; /* agp_buffers */
+   drmSize unused10; /* agp_buf_size */
+
+   int deviceID;
+   int width;
+   int height;
+   int mem;
+   int cpp;
+   int bitsPerPixel;
+
+   int unused11[8]; /* was front/back/depth/rotated offset/pitch */
+   
+   int unused12; /* logTextureGranularity */
+   int unused13; /* textureOffset */
+
+   int irq;
+   int sarea_priv_offset;
+} I830DRIRec, *I830DRIPtr;
+
+typedef struct {
+   /* Nothing here yet */
+   int dummy;
+} I830ConfigPrivRec, *I830ConfigPrivPtr;
+
+typedef struct {
+   /* Nothing here yet */
+   int dummy;
+} I830DRIContextRec, *I830DRIContextPtr;
+
+
+#endif
diff --git a/src/gallium/winsys/dri/intel/server/intel.h b/src/gallium/winsys/dri/intel/server/intel.h
new file mode 100644
index 0000000000..6ea72499c1
--- /dev/null
+++ b/src/gallium/winsys/dri/intel/server/intel.h
@@ -0,0 +1,331 @@
+#ifndef _INTEL_H_
+#define _INTEL_H_
+
+#include "xf86drm.h"		/* drm_handle_t, etc */
+
+/* Intel */
+#ifndef PCI_CHIP_I810
+#define PCI_CHIP_I810              0x7121
+#define PCI_CHIP_I810_DC100        0x7123
+#define PCI_CHIP_I810_E            0x7125
+#define PCI_CHIP_I815              0x1132
+#define PCI_CHIP_I810_BRIDGE       0x7120
+#define PCI_CHIP_I810_DC100_BRIDGE 0x7122
+#define PCI_CHIP_I810_E_BRIDGE     0x7124
+#define PCI_CHIP_I815_BRIDGE       0x1130
+#endif
+
+#define PCI_CHIP_845_G			0x2562
+#define PCI_CHIP_I830_M			0x3577
+
+#ifndef PCI_CHIP_I855_GM
+#define PCI_CHIP_I855_GM	   0x3582
+#define PCI_CHIP_I855_GM_BRIDGE	   0x3580
+#endif
+
+#ifndef PCI_CHIP_I865_G
+#define PCI_CHIP_I865_G		   0x2572
+#define PCI_CHIP_I865_G_BRIDGE	   0x2570
+#endif
+
+#ifndef PCI_CHIP_I915_G
+#define PCI_CHIP_I915_G		   0x2582
+#define PCI_CHIP_I915_G_BRIDGE	   0x2580
+#endif
+
+#ifndef PCI_CHIP_I915_GM
+#define PCI_CHIP_I915_GM	   0x2592
+#define PCI_CHIP_I915_GM_BRIDGE	   0x2590
+#endif
+
+#ifndef PCI_CHIP_E7221_G
+#define PCI_CHIP_E7221_G	   0x258A
+/* Same as I915_G_BRIDGE */
+#define PCI_CHIP_E7221_G_BRIDGE	   0x2580
+#endif
+
+#ifndef PCI_CHIP_I945_G
+#define PCI_CHIP_I945_G        0x2772
+#define PCI_CHIP_I945_G_BRIDGE 0x2770
+#endif
+
+#ifndef PCI_CHIP_I945_GM
+#define PCI_CHIP_I945_GM        0x27A2
+#define PCI_CHIP_I945_GM_BRIDGE 0x27A0
+#endif
+
+#define IS_I810(pI810) (pI810->Chipset == PCI_CHIP_I810 ||	\
+			pI810->Chipset == PCI_CHIP_I810_DC100 || \
+			pI810->Chipset == PCI_CHIP_I810_E)
+#define IS_I815(pI810) (pI810->Chipset == PCI_CHIP_I815)
+#define IS_I830(pI810) (pI810->Chipset == PCI_CHIP_I830_M)
+#define IS_845G(pI810) (pI810->Chipset == PCI_CHIP_845_G)
+#define IS_I85X(pI810)  (pI810->Chipset == PCI_CHIP_I855_GM)
+#define IS_I852(pI810)  (pI810->Chipset == PCI_CHIP_I855_GM && (pI810->variant == I852_GM || pI810->variant == I852_GME))
+#define IS_I855(pI810)  (pI810->Chipset == PCI_CHIP_I855_GM && (pI810->variant == I855_GM || pI810->variant == I855_GME))
+#define IS_I865G(pI810) (pI810->Chipset == PCI_CHIP_I865_G)
+
+#define IS_I915G(pI810) (pI810->Chipset == PCI_CHIP_I915_G || pI810->Chipset == PCI_CHIP_E7221_G)
+#define IS_I915GM(pI810) (pI810->Chipset == PCI_CHIP_I915_GM)
+#define IS_I945G(pI810) (pI810->Chipset == PCI_CHIP_I945_G)
+#define IS_I945GM(pI810) (pI810->Chipset == PCI_CHIP_I945_GM)
+#define IS_I9XX(pI810) (IS_I915G(pI810) || IS_I915GM(pI810) || IS_I945G(pI810) || IS_I945GM(pI810))
+
+#define IS_MOBILE(pI810) (IS_I830(pI810) || IS_I85X(pI810) || IS_I915GM(pI810) || IS_I945GM(pI810))
+
+#define I830_GMCH_CTRL		0x52
+
+#define I830_GMCH_MEM_MASK      0x1
+#define I830_GMCH_MEM_64M       0x1
+#define I830_GMCH_MEM_128M      0
+
+#define I830_GMCH_GMS_MASK			0x70
+#define I830_GMCH_GMS_DISABLED		0x00
+#define I830_GMCH_GMS_LOCAL			0x10
+#define I830_GMCH_GMS_STOLEN_512	0x20
+#define I830_GMCH_GMS_STOLEN_1024	0x30
+#define I830_GMCH_GMS_STOLEN_8192	0x40
+
+#define I855_GMCH_GMS_MASK			(0x7 << 4)
+#define I855_GMCH_GMS_DISABLED			0x00
+#define I855_GMCH_GMS_STOLEN_1M			(0x1 << 4)
+#define I855_GMCH_GMS_STOLEN_4M			(0x2 << 4)
+#define I855_GMCH_GMS_STOLEN_8M			(0x3 << 4)
+#define I855_GMCH_GMS_STOLEN_16M		(0x4 << 4)
+#define I855_GMCH_GMS_STOLEN_32M		(0x5 << 4)
+#define I915G_GMCH_GMS_STOLEN_48M		(0x6 << 4)
+#define I915G_GMCH_GMS_STOLEN_64M		(0x7 << 4)
+
+typedef unsigned char Bool;
+#define TRUE 1
+#define FALSE 0
+
+#define PIPE_NONE	0<<0
+#define PIPE_CRT	1<<0
+#define PIPE_TV		1<<1
+#define PIPE_DFP	1<<2
+#define PIPE_LFP	1<<3
+#define PIPE_CRT2	1<<4
+#define PIPE_TV2	1<<5
+#define PIPE_DFP2	1<<6
+#define PIPE_LFP2	1<<7
+
+typedef struct _I830MemPool *I830MemPoolPtr;
+typedef struct _I830MemRange *I830MemRangePtr;
+typedef struct _I830MemRange {
+   long Start;
+   long End;
+   long Size;
+   unsigned long Physical;
+   unsigned long Offset;		/* Offset of AGP-allocated portion */
+   unsigned long Alignment;
+   drm_handle_t Key;
+   unsigned long Pitch; // add pitch
+   I830MemPoolPtr Pool;
+} I830MemRange;
+
+typedef struct _I830MemPool {
+   I830MemRange Total;
+   I830MemRange Free;
+   I830MemRange Fixed;
+   I830MemRange Allocated;
+} I830MemPool;
+
+typedef struct {
+   int tail_mask;
+   I830MemRange mem;
+   unsigned char *virtual_start;
+   int head;
+   int tail;
+   int space;
+} I830RingBuffer;
+
+typedef struct _I830Rec {
+   unsigned char *MMIOBase;
+   unsigned char *FbBase;
+   int cpp;
+   uint32_t aper_size;
+   unsigned int bios_version;
+
+   /* These are set in PreInit and never changed. */
+   long FbMapSize;
+   long TotalVideoRam;
+   I830MemRange StolenMemory;		/* pre-allocated memory */
+   long BIOSMemorySize;			/* min stolen pool size */
+   int BIOSMemSizeLoc;
+
+   /* These change according to what has been allocated. */
+   long FreeMemory;
+   I830MemRange MemoryAperture;
+   I830MemPool StolenPool;
+   long allocatedMemory;
+
+   /* Regions allocated either from the above pools, or from agpgart. */
+   /* for single and dual head configurations */
+   I830MemRange FrontBuffer;
+   I830MemRange FrontBuffer2;
+   I830MemRange Scratch;
+   I830MemRange Scratch2;
+
+   I830RingBuffer *LpRing;
+
+   I830MemRange BackBuffer;
+   I830MemRange DepthBuffer;
+   I830MemRange TexMem;
+   int TexGranularity;
+   I830MemRange ContextMem;
+   int drmMinor;
+   Bool have3DWindows;
+
+   Bool NeedRingBufferLow;
+   Bool allowPageFlip;
+   Bool disableTiling;
+
+   int Chipset;
+   unsigned long LinearAddr;
+   unsigned long MMIOAddr;
+
+   drmSize           registerSize;     /**< \brief MMIO register map size */
+   drm_handle_t         registerHandle;   /**< \brief MMIO register map handle */
+  //   IOADDRESS ioBase;
+   int               irq;              /**< \brief IRQ number */
+   int GttBound;
+
+   drm_handle_t ring_map;
+   unsigned int Fence[8];
+
+} I830Rec;
+
+/*
+ * 12288 is set as the maximum, chosen because it is enough for
+ * 1920x1440@32bpp with a 2048 pixel line pitch with some to spare.
+ */
+#define I830_MAXIMUM_VBIOS_MEM		12288
+#define I830_DEFAULT_VIDEOMEM_2D	(MB(32) / 1024)
+#define I830_DEFAULT_VIDEOMEM_3D	(MB(64) / 1024)
+
+/* Flags for memory allocation function */
+#define FROM_ANYWHERE			0x00000000
+#define FROM_POOL_ONLY			0x00000001
+#define FROM_NEW_ONLY			0x00000002
+#define FROM_MASK			0x0000000f
+
+#define ALLOCATE_AT_TOP			0x00000010
+#define ALLOCATE_AT_BOTTOM		0x00000020
+#define FORCE_GAPS			0x00000040
+
+#define NEED_PHYSICAL_ADDR		0x00000100
+#define ALIGN_BOTH_ENDS			0x00000200
+#define FORCE_LOW			0x00000400
+
+#define ALLOC_NO_TILING			0x00001000
+#define ALLOC_INITIAL			0x00002000
+
+#define ALLOCATE_DRY_RUN		0x80000000
+
+/* Chipset registers for VIDEO BIOS memory RW access */
+#define _855_DRAM_RW_CONTROL 0x58
+#define _845_DRAM_RW_CONTROL 0x90
+#define DRAM_WRITE    0x33330000
+
+#define KB(x) ((x) * 1024)
+#define MB(x) ((x) * KB(1024))
+
+#define GTT_PAGE_SIZE			KB(4)
+#define ROUND_TO(x, y)			(((x) + (y) - 1) / (y) * (y))
+#define ROUND_DOWN_TO(x, y)		((x) / (y) * (y))
+#define ROUND_TO_PAGE(x)		ROUND_TO((x), GTT_PAGE_SIZE)
+#define ROUND_TO_MB(x)			ROUND_TO((x), MB(1))
+#define PRIMARY_RINGBUFFER_SIZE		KB(128)
+
+
+/* Ring buffer registers, p277, overview p19
+ */
+#define LP_RING     0x2030
+#define HP_RING     0x2040
+
+#define RING_TAIL      0x00
+#define TAIL_ADDR           0x000FFFF8
+#define I830_TAIL_MASK	    0x001FFFF8
+
+#define RING_HEAD      0x04
+#define HEAD_WRAP_COUNT     0xFFE00000
+#define HEAD_WRAP_ONE       0x00200000
+#define HEAD_ADDR           0x001FFFFC
+#define I830_HEAD_MASK      0x001FFFFC
+
+#define RING_START     0x08
+#define START_ADDR          0x03FFFFF8
+#define I830_RING_START_MASK	0xFFFFF000
+
+#define RING_LEN       0x0C
+#define RING_NR_PAGES       0x001FF000 
+#define I830_RING_NR_PAGES	0x001FF000
+#define RING_REPORT_MASK    0x00000006
+#define RING_REPORT_64K     0x00000002
+#define RING_REPORT_128K    0x00000004
+#define RING_NO_REPORT      0x00000000
+#define RING_VALID_MASK     0x00000001
+#define RING_VALID          0x00000001
+#define RING_INVALID        0x00000000
+
+
+/* Fence/Tiling ranges [0..7]
+ */
+#define FENCE            0x2000
+#define FENCE_NR         8
+
+#define I915G_FENCE_START_MASK	0x0ff00000
+
+#define I830_FENCE_START_MASK	0x07f80000
+
+#define FENCE_START_MASK    0x03F80000
+#define FENCE_X_MAJOR       0x00000000
+#define FENCE_Y_MAJOR       0x00001000
+#define FENCE_SIZE_MASK     0x00000700
+#define FENCE_SIZE_512K     0x00000000
+#define FENCE_SIZE_1M       0x00000100
+#define FENCE_SIZE_2M       0x00000200
+#define FENCE_SIZE_4M       0x00000300
+#define FENCE_SIZE_8M       0x00000400
+#define FENCE_SIZE_16M      0x00000500
+#define FENCE_SIZE_32M      0x00000600
+#define FENCE_SIZE_64M	    0x00000700
+#define I915G_FENCE_SIZE_1M       0x00000000
+#define I915G_FENCE_SIZE_2M       0x00000100
+#define I915G_FENCE_SIZE_4M       0x00000200
+#define I915G_FENCE_SIZE_8M       0x00000300
+#define I915G_FENCE_SIZE_16M      0x00000400
+#define I915G_FENCE_SIZE_32M      0x00000500
+#define I915G_FENCE_SIZE_64M	0x00000600
+#define I915G_FENCE_SIZE_128M	0x00000700
+#define FENCE_PITCH_1       0x00000000
+#define FENCE_PITCH_2       0x00000010
+#define FENCE_PITCH_4       0x00000020
+#define FENCE_PITCH_8       0x00000030
+#define FENCE_PITCH_16      0x00000040
+#define FENCE_PITCH_32      0x00000050
+#define FENCE_PITCH_64	    0x00000060
+#define FENCE_VALID         0x00000001
+
+#include <mmio.h>
+
+#  define MMIO_IN8(base, offset) \
+	*(volatile unsigned char *)(((unsigned char*)(base)) + (offset))
+#  define MMIO_IN32(base, offset) \
+	read_MMIO_LE32(base, offset)
+#  define MMIO_OUT8(base, offset, val) \
+	*(volatile unsigned char *)(((unsigned char*)(base)) + (offset)) = (val)
+#  define MMIO_OUT32(base, offset, val) \
+	*(volatile unsigned int *)(void *)(((unsigned char*)(base)) + (offset)) = CPU_TO_LE32(val)
+
+
+				/* Memory mapped register access macros */
+#define INREG8(addr)        MMIO_IN8(MMIO, addr)
+#define INREG(addr)         MMIO_IN32(MMIO, addr)
+#define OUTREG8(addr, val)  MMIO_OUT8(MMIO, addr, val)
+#define OUTREG(addr, val)   MMIO_OUT32(MMIO, addr, val)
+
+#define DSPABASE		0x70184
+
+#endif
diff --git a/src/gallium/winsys/dri/intel/server/intel_dri.c b/src/gallium/winsys/dri/intel/server/intel_dri.c
new file mode 100644
index 0000000000..e49c4214ad
--- /dev/null
+++ b/src/gallium/winsys/dri/intel/server/intel_dri.c
@@ -0,0 +1,1306 @@
+/**
+ * \file server/intel_dri.c
+ * \brief File to perform the device-specific initialization tasks typically
+ * done in the X server.
+ *
+ * Here they are converted to run in the client (or perhaps a standalone
+ * process), and to work with the frame buffer device rather than the X
+ * server infrastructure.
+ * 
+ * Copyright (C) 2006 Dave Airlie (airlied@linux.ie)
+
+ Permission is hereby granted, free of charge, to any person obtaining a
+ copy of this software and associated documentation files (the
+ "Software"), to deal in the Software without restriction, including
+ without limitation the rights to use, copy, modify, merge, publish,
+ distribute, sub license, and/or sell copies of the Software, and to
+ permit persons to whom the Software is furnished to do so, subject to
+ the following conditions:
+ 
+ The above copyright notice and this permission notice (including the
+ next paragraph) shall be included in all copies or substantial portions
+ of the Software.
+ 
+ THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ IN NO EVENT SHALL THE COPYRIGHT HOLDERS AND/OR THEIR SUPPLIERS BE LIABLE FOR
+ ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+*/
+
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include <errno.h>
+#include <unistd.h>
+
+#include "driver.h"
+#include "drm.h"
+
+#include "intel.h"
+#include "i830_dri.h"
+
+#include "memops.h"
+#include "pciaccess.h"
+
+static size_t drm_page_size;
+static int nextTile = 0;
+#define xf86DrvMsg(...) do {} while(0)
+
+static const int pitches[] = {
+  128 * 8,
+  128 * 16,
+  128 * 32,
+  128 * 64,
+  0
+};
+
+static Bool I830DRIDoMappings(DRIDriverContext *ctx, I830Rec *pI830, drmI830Sarea *sarea);
+
+static unsigned long
+GetBestTileAlignment(unsigned long size)
+{
+   unsigned long i;
+
+   for (i = KB(512); i < size; i <<= 1)
+      ;
+
+   if (i > MB(64))
+      i = MB(64);
+
+   return i;
+}
+
+static void SetFenceRegs(const DRIDriverContext *ctx, I830Rec *pI830)
+{
+  int i;
+  unsigned char *MMIO = ctx->MMIOAddress;
+
+  for (i = 0; i < 8; i++) {
+    OUTREG(FENCE + i * 4, pI830->Fence[i]);
+    //    if (I810_DEBUG & DEBUG_VERBOSE_VGA)
+    fprintf(stderr,"Fence Register : %x\n", pI830->Fence[i]);
+  }
+}
+
+/* Tiled memory is good... really, really good...
+ *
+ * Need to make it less likely that we miss out on this - probably
+ * need to move the frontbuffer away from the 'guarenteed' alignment
+ * of the first memory segment, or perhaps allocate a discontigous
+ * framebuffer to get more alignment 'sweet spots'.
+ */
+static void
+SetFence(const DRIDriverContext *ctx, I830Rec *pI830,
+	 int nr, unsigned int start, unsigned int pitch,
+         unsigned int size)
+{
+   unsigned int val;
+   unsigned int fence_mask = 0;
+   unsigned int fence_pitch;
+
+   if (nr < 0 || nr > 7) {
+      fprintf(stderr,
+		 "SetFence: fence %d out of range\n",nr);
+      return;
+   }
+
+   pI830->Fence[nr] = 0;
+
+   if (IS_I9XX(pI830))
+   	fence_mask = ~I915G_FENCE_START_MASK;
+   else
+   	fence_mask = ~I830_FENCE_START_MASK;
+
+   if (start & fence_mask) {
+      fprintf(stderr,
+		 "SetFence: %d: start (0x%08x) is not %s aligned\n",
+		 nr, start, (IS_I9XX(pI830)) ? "1MB" : "512k");
+      return;
+   }
+
+   if (start % size) {
+      fprintf(stderr,
+		 "SetFence: %d: start (0x%08x) is not size (%dk) aligned\n",
+		 nr, start, size / 1024);
+      return;
+   }
+
+   if (pitch & 127) {
+      fprintf(stderr,
+		 "SetFence: %d: pitch (%d) not a multiple of 128 bytes\n",
+		 nr, pitch);
+      return;
+   }
+
+   val = (start | FENCE_X_MAJOR | FENCE_VALID);
+
+   if (IS_I9XX(pI830)) {
+   	switch (size) {
+	   case MB(1):
+      		val |= I915G_FENCE_SIZE_1M;
+      		break;
+   	   case MB(2):
+      		val |= I915G_FENCE_SIZE_2M;
+      		break;
+   	   case MB(4):
+      		val |= I915G_FENCE_SIZE_4M;
+      		break;
+   	   case MB(8):
+      		val |= I915G_FENCE_SIZE_8M;
+      		break;
+   	   case MB(16):
+      		val |= I915G_FENCE_SIZE_16M;
+      		break;
+   	   case MB(32):
+      		val |= I915G_FENCE_SIZE_32M;
+      		break;
+   	   case MB(64):
+      		val |= I915G_FENCE_SIZE_64M;
+      		break;
+   	   default:
+      		fprintf(stderr,
+		 "SetFence: %d: illegal size (%d kByte)\n", nr, size / 1024);
+      		return;
+   	}
+    } else {
+   	switch (size) {
+	   case KB(512):
+      		val |= FENCE_SIZE_512K;
+      		break;
+	   case MB(1):
+      		val |= FENCE_SIZE_1M;
+      		break;
+   	   case MB(2):
+      		val |= FENCE_SIZE_2M;
+      		break;
+   	   case MB(4):
+      		val |= FENCE_SIZE_4M;
+      		break;
+   	   case MB(8):
+      		val |= FENCE_SIZE_8M;
+      		break;
+   	   case MB(16):
+      		val |= FENCE_SIZE_16M;
+      		break;
+   	   case MB(32):
+      		val |= FENCE_SIZE_32M;
+      		break;
+   	   case MB(64):
+      		val |= FENCE_SIZE_64M;
+      		break;
+   	   default:
+      		fprintf(stderr,
+		 "SetFence: %d: illegal size (%d kByte)\n", nr, size / 1024);
+      		return;
+   	}
+   }
+
+   if (IS_I9XX(pI830))
+	fence_pitch = pitch / 512;
+   else
+	fence_pitch = pitch / 128;
+
+   switch (fence_pitch) {
+   case 1:
+      val |= FENCE_PITCH_1;
+      break;
+   case 2:
+      val |= FENCE_PITCH_2;
+      break;
+   case 4:
+      val |= FENCE_PITCH_4;
+      break;
+   case 8:
+      val |= FENCE_PITCH_8;
+      break;
+   case 16:
+      val |= FENCE_PITCH_16;
+      break;
+   case 32:
+      val |= FENCE_PITCH_32;
+      break;
+   case 64:
+      val |= FENCE_PITCH_64;
+      break;
+   default:
+      fprintf(stderr,
+		 "SetFence: %d: illegal pitch (%d)\n", nr, pitch);
+      return;
+   }
+
+   pI830->Fence[nr] = val;
+}
+
+static Bool
+MakeTiles(const DRIDriverContext *ctx, I830Rec *pI830, I830MemRange *pMem)
+{
+   int pitch, ntiles, i;
+
+   pitch = pMem->Pitch * ctx->cpp;
+   /*
+    * Simply try to break the region up into at most four pieces of size
+    * equal to the alignment.
+    */
+   ntiles = ROUND_TO(pMem->Size, pMem->Alignment) / pMem->Alignment;
+   if (ntiles >= 4) {
+      return FALSE;
+   }
+
+   for (i = 0; i < ntiles; i++, nextTile++) {
+     SetFence(ctx, pI830, nextTile, pMem->Start + i * pMem->Alignment,
+	       pitch, pMem->Alignment);
+   }
+   return TRUE;
+}
+
+static void I830SetupMemoryTiling(const DRIDriverContext *ctx, I830Rec *pI830)
+{
+  int i;
+
+  /* Clear out */
+  for (i = 0; i < 8; i++)
+    pI830->Fence[i] = 0;
+  
+  nextTile = 0;
+
+  if (pI830->BackBuffer.Alignment >= KB(512)) {
+    if (MakeTiles(ctx, pI830, &(pI830->BackBuffer))) {
+      fprintf(stderr,
+		 "Activating tiled memory for the back buffer.\n");
+    } else {
+      fprintf(stderr,
+		 "MakeTiles failed for the back buffer.\n");
+      pI830->allowPageFlip = FALSE;
+    }
+  }
+  
+  if (pI830->DepthBuffer.Alignment >= KB(512)) {
+    if (MakeTiles(ctx, pI830, &(pI830->DepthBuffer))) {
+      fprintf(stderr,
+		 "Activating tiled memory for the depth buffer.\n");
+    } else {
+      fprintf(stderr,
+		 "MakeTiles failed for the depth buffer.\n");
+    }
+  }
+
+  return;
+}
+
+static int I830DetectMemory(const DRIDriverContext *ctx, I830Rec *pI830)
+{
+  struct pci_device host_bridge, ig_dev;
+  uint32_t gmch_ctrl;
+  int memsize = 0;
+  int range;
+  uint32_t aper_size;
+  uint32_t membase2 = 0;
+      
+  memset(&host_bridge, 0, sizeof(host_bridge));
+  memset(&ig_dev, 0, sizeof(ig_dev));
+
+  ig_dev.dev = 2;
+
+  pci_device_cfg_read_u32(&host_bridge, &gmch_ctrl, I830_GMCH_CTRL);
+
+  if (IS_I830(pI830) || IS_845G(pI830)) {
+    if ((gmch_ctrl & I830_GMCH_MEM_MASK) == I830_GMCH_MEM_128M) {
+      aper_size = 0x80000000;
+    } else {
+      aper_size = 0x40000000;
+    }
+  } else {
+    if (IS_I9XX(pI830)) {
+      int ret;
+      ret = pci_device_cfg_read_u32(&ig_dev, &membase2, 0x18);
+      if (membase2 & 0x08000000)
+	aper_size = 0x8000000;
+      else
+	aper_size = 0x10000000;
+
+      fprintf(stderr,"aper size is %08X %08x %d\n", aper_size, membase2, ret);
+    } else
+      aper_size = 0x8000000;
+  }
+
+  pI830->aper_size = aper_size;
+
+
+  /* We need to reduce the stolen size, by the GTT and the popup.
+   * The GTT varying according the the FbMapSize and the popup is 4KB */
+  range = (ctx->shared.fbSize / (1024*1024)) + 4;
+
+   if (IS_I85X(pI830) || IS_I865G(pI830) || IS_I9XX(pI830)) {
+      switch (gmch_ctrl & I830_GMCH_GMS_MASK) {
+      case I855_GMCH_GMS_STOLEN_1M:
+	 memsize = MB(1) - KB(range);
+	 break;
+      case I855_GMCH_GMS_STOLEN_4M:
+	 memsize = MB(4) - KB(range);
+	 break;
+      case I855_GMCH_GMS_STOLEN_8M:
+	 memsize = MB(8) - KB(range);
+	 break;
+      case I855_GMCH_GMS_STOLEN_16M:
+	 memsize = MB(16) - KB(range);
+	 break;
+      case I855_GMCH_GMS_STOLEN_32M:
+	 memsize = MB(32) - KB(range);
+	 break;
+      case I915G_GMCH_GMS_STOLEN_48M:
+	 if (IS_I9XX(pI830))
+	    memsize = MB(48) - KB(range);
+	 break;
+      case I915G_GMCH_GMS_STOLEN_64M:
+	 if (IS_I9XX(pI830))
+	    memsize = MB(64) - KB(range);
+	 break;
+      }
+   } else {
+      switch (gmch_ctrl & I830_GMCH_GMS_MASK) {
+      case I830_GMCH_GMS_STOLEN_512:
+	 memsize = KB(512) - KB(range);
+	 break;
+      case I830_GMCH_GMS_STOLEN_1024:
+	 memsize = MB(1) - KB(range);
+	 break;
+      case I830_GMCH_GMS_STOLEN_8192:
+	 memsize = MB(8) - KB(range);
+	 break;
+      case I830_GMCH_GMS_LOCAL:
+	 memsize = 0;
+	 xf86DrvMsg(pScrn->scrnIndex, X_WARNING,
+		    "Local memory found, but won't be used.\n");
+	 break;
+      }
+   }
+   if (memsize > 0) {
+     fprintf(stderr,
+		 "detected %d kB stolen memory.\n", memsize / 1024);
+   } else {
+     fprintf(stderr,
+		 "no video memory detected.\n");
+   }
+   return memsize;
+}
+
+static int AgpInit(const DRIDriverContext *ctx, I830Rec *info)
+{
+  unsigned long mode = 0x4;
+
+  if (drmAgpAcquire(ctx->drmFD) < 0) {
+    fprintf(stderr, "[gart] AGP not available\n");
+    return 0;
+  }
+  
+  if (drmAgpEnable(ctx->drmFD, mode) < 0) {
+    fprintf(stderr, "[gart] AGP not enabled\n");
+    drmAgpRelease(ctx->drmFD);
+    return 0;
+  }
+  else
+    fprintf(stderr, "[gart] AGP enabled at %dx\n", ctx->agpmode);
+
+  return 1;
+}
+
+/*
+ * Allocate memory from the given pool.  Grow the pool if needed and if
+ * possible.
+ */
+static unsigned long
+AllocFromPool(const DRIDriverContext *ctx, I830Rec *pI830, 
+	      I830MemRange *result, I830MemPool *pool,
+	      long size, unsigned long alignment, int flags)
+{
+   long needed, start, end;
+
+   if (!result || !pool || !size)
+      return 0;
+
+   /* Calculate how much space is needed. */
+   if (alignment <= GTT_PAGE_SIZE)
+      needed = size;
+   else {
+	 start = ROUND_TO(pool->Free.Start, alignment);
+	 end = ROUND_TO(start + size, alignment);
+	 needed = end - pool->Free.Start;
+   }
+   if (needed > pool->Free.Size) {
+     return 0;
+   }
+
+   result->Start = ROUND_TO(pool->Free.Start, alignment);
+   pool->Free.Start += needed;
+   result->End = pool->Free.Start;
+
+   pool->Free.Size = pool->Free.End - pool->Free.Start;
+   result->Size = result->End - result->Start;
+   result->Pool = pool;
+   result->Alignment = alignment;
+   return needed;
+}
+
+static unsigned long AllocFromAGP(const DRIDriverContext *ctx, I830Rec *pI830, long size, unsigned long alignment, I830MemRange  *result)
+{
+   unsigned long start, end;
+   unsigned long newApStart, newApEnd;
+   int ret;
+   if (!result || !size)
+      return 0;
+   
+   if (!alignment)
+     alignment = 4;
+
+   start = ROUND_TO(pI830->MemoryAperture.Start, alignment);
+   end = ROUND_TO(start + size, alignment);
+   newApStart = end;
+   newApEnd = pI830->MemoryAperture.End;
+
+   ret=drmAgpAlloc(ctx->drmFD, size, 0, &(result->Physical), (drm_handle_t *)&(result->Key));
+   
+   if (ret)
+   {
+     fprintf(stderr,"drmAgpAlloc failed %d\n", ret);
+     return 0;
+   }
+   pI830->allocatedMemory += size;
+   pI830->MemoryAperture.Start = newApStart;
+   pI830->MemoryAperture.End = newApEnd;
+   pI830->MemoryAperture.Size = newApEnd - newApStart;
+   //   pI830->FreeMemory -= size;
+   result->Start = start;
+   result->End = start + size;
+   result->Size = size;
+   result->Offset = start;
+   result->Alignment = alignment;
+   result->Pool = NULL;
+  
+   return size;
+}
+
+unsigned long
+I830AllocVidMem(const DRIDriverContext *ctx, I830Rec *pI830,
+                I830MemRange *result, I830MemPool *pool, long size,
+                unsigned long alignment, int flags)
+{
+   unsigned long ret;
+
+   if (!result)
+      return 0;
+
+   /* Make sure these are initialised. */
+   result->Size = 0;
+   result->Key = -1;
+
+   if (!size) {
+      return 0;
+   }
+
+   if (pool->Free.Size < size) {
+      ret = AllocFromAGP(ctx, pI830, size, alignment, result);
+   }
+   else {
+      ret = AllocFromPool(ctx, pI830, result, pool, size, alignment, flags);
+      if (ret == 0)
+         ret = AllocFromAGP(ctx, pI830, size, alignment, result);
+   }
+   return ret;
+}
+
+static Bool BindAgpRange(const DRIDriverContext *ctx, I830MemRange *mem)
+{
+  if (!mem)
+    return FALSE;
+  
+  if (mem->Key == -1)
+    return TRUE;
+
+  return !drmAgpBind(ctx->drmFD, mem->Key, mem->Offset);
+}
+
+/* simple memory allocation routines needed */
+/* put ring buffer in low memory */
+/* need to allocate front, back, depth buffers aligned correctly,
+   allocate ring buffer, 
+*/
+
+/* */
+static Bool
+I830AllocateMemory(const DRIDriverContext *ctx, I830Rec *pI830)
+{
+  unsigned long size, ret;
+  unsigned long lines, lineSize, align;
+
+  /* allocate ring buffer */
+  memset(pI830->LpRing, 0, sizeof(I830RingBuffer));
+  pI830->LpRing->mem.Key = -1;
+
+  size = PRIMARY_RINGBUFFER_SIZE;
+  
+  ret = I830AllocVidMem(ctx, pI830, &pI830->LpRing->mem, &pI830->StolenPool, size, 0x1000, 0);
+  
+  if (ret != size)
+  {
+    fprintf(stderr,"unable to allocate ring buffer %ld\n", ret);
+    return FALSE;
+  }
+
+  pI830->LpRing->tail_mask = pI830->LpRing->mem.Size - 1;
+
+  
+  /* allocate front buffer */
+  memset(&(pI830->FrontBuffer), 0, sizeof(pI830->FrontBuffer));
+  pI830->FrontBuffer.Key = -1;
+  pI830->FrontBuffer.Pitch = ctx->shared.virtualWidth;
+
+  align = KB(512);  
+
+  lineSize = ctx->shared.virtualWidth * ctx->cpp;
+  lines = (ctx->shared.virtualHeight + 15) / 16 * 16;
+  size = lineSize * lines;
+  size = ROUND_TO_PAGE(size);
+
+  align = GetBestTileAlignment(size);
+
+  ret = I830AllocVidMem(ctx, pI830, &pI830->FrontBuffer, &pI830->StolenPool, size, align, 0);
+  if (ret < size)
+  {
+    fprintf(stderr,"unable to allocate front buffer %ld\n", ret);
+    return FALSE;
+  }
+
+  memset(&(pI830->BackBuffer), 0, sizeof(pI830->BackBuffer));
+  pI830->BackBuffer.Key = -1;
+  pI830->BackBuffer.Pitch = ctx->shared.virtualWidth;
+
+  ret = I830AllocVidMem(ctx, pI830, &pI830->BackBuffer, &pI830->StolenPool, size, align, 0);
+  if (ret < size)
+  {
+    fprintf(stderr,"unable to allocate back buffer %ld\n", ret);
+    return FALSE;
+  }
+  
+  memset(&(pI830->DepthBuffer), 0, sizeof(pI830->DepthBuffer));
+  pI830->DepthBuffer.Key = -1;
+  pI830->DepthBuffer.Pitch = ctx->shared.virtualWidth;
+
+  ret = I830AllocVidMem(ctx, pI830, &pI830->DepthBuffer, &pI830->StolenPool, size, align, 0);
+  if (ret < size)
+  {
+    fprintf(stderr,"unable to allocate depth buffer %ld\n", ret);
+    return FALSE;
+  }
+
+  memset(&(pI830->ContextMem), 0, sizeof(pI830->ContextMem));
+  pI830->ContextMem.Key = -1;
+  size = KB(32);
+
+  ret = I830AllocVidMem(ctx, pI830, &pI830->ContextMem, &pI830->StolenPool, size, align, 0);
+  if (ret < size)
+  {
+    fprintf(stderr,"unable to allocate context buffer %ld\n", ret);
+    return FALSE;
+  }
+
+#if 0  
+  memset(&(pI830->TexMem), 0, sizeof(pI830->TexMem));
+  pI830->TexMem.Key = -1;
+
+  size = 32768 * 1024;
+  ret = AllocFromAGP(ctx, pI830, size, align, &pI830->TexMem);
+  if (ret < size)
+  {
+    fprintf(stderr,"unable to allocate texture memory %ld\n", ret);
+    return FALSE;
+  }
+#endif
+
+  return TRUE;
+}
+
+static Bool
+I830BindMemory(const DRIDriverContext *ctx, I830Rec *pI830)
+{
+  if (!BindAgpRange(ctx, &pI830->LpRing->mem))
+    return FALSE;
+  if (!BindAgpRange(ctx, &pI830->FrontBuffer))
+    return FALSE;
+  if (!BindAgpRange(ctx, &pI830->BackBuffer))
+    return FALSE;
+  if (!BindAgpRange(ctx, &pI830->DepthBuffer))
+    return FALSE;
+  if (!BindAgpRange(ctx, &pI830->ContextMem))
+    return FALSE;
+#if 0
+  if (!BindAgpRange(ctx, &pI830->TexMem))
+    return FALSE;
+#endif
+  return TRUE;
+}
+
+static void SetupDRIMM(const DRIDriverContext *ctx, I830Rec *pI830)
+{
+  unsigned long aperEnd = ROUND_DOWN_TO(pI830->aper_size, GTT_PAGE_SIZE) / GTT_PAGE_SIZE;
+  unsigned long aperStart = ROUND_TO(pI830->aper_size - KB(32768), GTT_PAGE_SIZE) / GTT_PAGE_SIZE;
+
+  fprintf(stderr, "aper size is %08X\n", ctx->shared.fbSize);
+  if (drmMMInit(ctx->drmFD, aperStart, aperEnd - aperStart, DRM_BO_MEM_TT)) {
+      fprintf(stderr,
+	      "DRM MM Initialization Failed\n");
+  } else {
+    fprintf(stderr,
+	    "DRM MM Initialized at offset 0x%lx length %d page\n", aperStart, aperEnd-aperStart);
+  }
+
+}
+
+static Bool
+I830CleanupDma(const DRIDriverContext *ctx)
+{
+   drmI830Init info;
+
+   memset(&info, 0, sizeof(drmI830Init));
+   info.func = I830_CLEANUP_DMA;
+
+   if (drmCommandWrite(ctx->drmFD, DRM_I830_INIT,
+		       &info, sizeof(drmI830Init))) {
+     fprintf(stderr, "I830 Dma Cleanup Failed\n");
+      return FALSE;
+   }
+
+   return TRUE;
+}
+
+static Bool
+I830InitDma(const DRIDriverContext *ctx, I830Rec *pI830)
+{
+   I830RingBuffer *ring = pI830->LpRing;
+   drmI830Init info;
+
+   memset(&info, 0, sizeof(drmI830Init));
+   info.func = I830_INIT_DMA;
+
+   info.ring_start = ring->mem.Start + pI830->LinearAddr;
+   info.ring_end = ring->mem.End + pI830->LinearAddr;
+   info.ring_size = ring->mem.Size;
+
+   info.mmio_offset = (unsigned int)ctx->MMIOStart;
+
+   info.sarea_priv_offset = sizeof(drm_sarea_t);
+
+   info.front_offset = pI830->FrontBuffer.Start;
+   info.back_offset = pI830->BackBuffer.Start;
+   info.depth_offset = pI830->DepthBuffer.Start;
+   info.w = ctx->shared.virtualWidth;
+   info.h = ctx->shared.virtualHeight;
+   info.pitch = ctx->shared.virtualWidth;
+   info.back_pitch = pI830->BackBuffer.Pitch;
+   info.depth_pitch = pI830->DepthBuffer.Pitch;
+   info.cpp = ctx->cpp;
+
+   if (drmCommandWrite(ctx->drmFD, DRM_I830_INIT,
+		       &info, sizeof(drmI830Init))) {
+      fprintf(stderr,
+		 "I830 Dma Initialization Failed\n");
+      return FALSE;
+   }
+
+   return TRUE;
+}
+
+static int I830CheckDRMVersion( const DRIDriverContext *ctx,
+				  I830Rec *pI830 )
+{
+   drmVersionPtr  version;
+
+   version = drmGetVersion(ctx->drmFD);
+
+   if (version) {
+     int req_minor, req_patch;
+
+     req_minor = 4;
+     req_patch = 0;	
+
+     if (version->version_major != 1 ||
+	 version->version_minor < req_minor ||
+	 (version->version_minor == req_minor && 
+	  version->version_patchlevel < req_patch)) {
+       /* Incompatible drm version */
+       fprintf(stderr,
+	       "[dri] I830DRIScreenInit failed because of a version "
+	       "mismatch.\n"
+	       "[dri] i915.o kernel module version is %d.%d.%d "
+	       "but version 1.%d.%d or newer is needed.\n"
+	       "[dri] Disabling DRI.\n",
+	       version->version_major,
+	       version->version_minor,
+	       version->version_patchlevel,
+	       req_minor,
+	       req_patch);
+       drmFreeVersion(version);
+       return 0;
+     }
+     
+     pI830->drmMinor = version->version_minor;
+     drmFreeVersion(version);
+   }
+   return 1;
+}
+
+static void
+I830SetRingRegs(const DRIDriverContext *ctx, I830Rec *pI830)
+{
+  unsigned int itemp;
+  unsigned char *MMIO = ctx->MMIOAddress;
+
+   OUTREG(LP_RING + RING_LEN, 0);
+   OUTREG(LP_RING + RING_TAIL, 0);
+   OUTREG(LP_RING + RING_HEAD, 0);
+
+   if ((long)(pI830->LpRing->mem.Start & I830_RING_START_MASK) !=
+       pI830->LpRing->mem.Start) {
+      fprintf(stderr,
+		 "I830SetRingRegs: Ring buffer start (%lx) violates its "
+		 "mask (%x)\n", pI830->LpRing->mem.Start, I830_RING_START_MASK);
+   }
+   /* Don't care about the old value.  Reserved bits must be zero anyway. */
+   itemp = pI830->LpRing->mem.Start & I830_RING_START_MASK;
+   OUTREG(LP_RING + RING_START, itemp);
+
+   if (((pI830->LpRing->mem.Size - 4096) & I830_RING_NR_PAGES) !=
+       pI830->LpRing->mem.Size - 4096) {
+      fprintf(stderr,
+		 "I830SetRingRegs: Ring buffer size - 4096 (%lx) violates its "
+		 "mask (%x)\n", pI830->LpRing->mem.Size - 4096,
+		 I830_RING_NR_PAGES);
+   }
+   /* Don't care about the old value.  Reserved bits must be zero anyway. */
+   itemp = (pI830->LpRing->mem.Size - 4096) & I830_RING_NR_PAGES;
+   itemp |= (RING_NO_REPORT | RING_VALID);
+   OUTREG(LP_RING + RING_LEN, itemp);
+
+   pI830->LpRing->head = INREG(LP_RING + RING_HEAD) & I830_HEAD_MASK;
+   pI830->LpRing->tail = INREG(LP_RING + RING_TAIL);
+   pI830->LpRing->space = pI830->LpRing->head - (pI830->LpRing->tail + 8);
+   if (pI830->LpRing->space < 0)
+      pI830->LpRing->space += pI830->LpRing->mem.Size;
+
+   SetFenceRegs(ctx, pI830);
+   
+   /* RESET THE DISPLAY PIPE TO POINT TO THE FRONTBUFFER - hacky
+      hacky hacky */
+   OUTREG(DSPABASE, pI830->FrontBuffer.Start + pI830->LinearAddr);
+
+}
+
+static Bool
+I830SetParam(const DRIDriverContext *ctx, int param, int value)
+{
+   drmI830SetParam sp;
+
+   memset(&sp, 0, sizeof(sp));
+   sp.param = param;
+   sp.value = value;
+
+   if (drmCommandWrite(ctx->drmFD, DRM_I830_SETPARAM, &sp, sizeof(sp))) {
+      fprintf(stderr, "I830 SetParam Failed\n");
+      return FALSE;
+   }
+
+   return TRUE;
+}
+
+static Bool
+I830DRIMapScreenRegions(DRIDriverContext *ctx, I830Rec *pI830, drmI830Sarea *sarea)
+{
+   fprintf(stderr,
+              "[drm] Mapping front buffer\n");
+
+   if (drmAddMap(ctx->drmFD,
+                 (drm_handle_t)(sarea->front_offset + pI830->LinearAddr),
+                 sarea->front_size,
+                 DRM_FRAME_BUFFER,  /*DRM_AGP,*/
+                 0,
+                 &sarea->front_handle) < 0) {
+     fprintf(stderr,
+	     "[drm] drmAddMap(front_handle) failed. Disabling DRI\n");
+      return FALSE;
+   }
+   ctx->shared.hFrameBuffer = sarea->front_handle;
+   ctx->shared.fbSize = sarea->front_size;
+   fprintf(stderr, "[drm] Front Buffer = 0x%08x\n",
+	   sarea->front_handle);
+
+   if (drmAddMap(ctx->drmFD,
+                 (drm_handle_t)(sarea->back_offset),
+                 sarea->back_size, DRM_AGP, 0,
+                 &sarea->back_handle) < 0) {
+      fprintf(stderr,
+                 "[drm] drmAddMap(back_handle) failed. Disabling DRI\n");
+      return FALSE;
+   }
+   fprintf(stderr, "[drm] Back Buffer = 0x%08x\n",
+              sarea->back_handle);
+
+   if (drmAddMap(ctx->drmFD,
+                 (drm_handle_t)sarea->depth_offset,
+                 sarea->depth_size, DRM_AGP, 0,
+                 &sarea->depth_handle) < 0) {
+      fprintf(stderr,
+                 "[drm] drmAddMap(depth_handle) failed. Disabling DRI\n");
+      return FALSE;
+   }
+   fprintf(stderr, "[drm] Depth Buffer = 0x%08x\n",
+              sarea->depth_handle);
+
+#if 0
+   if (drmAddMap(ctx->drmFD,
+		 (drm_handle_t)sarea->tex_offset,
+		 sarea->tex_size, DRM_AGP, 0,
+		 &sarea->tex_handle) < 0) {
+      fprintf(stderr,
+		 "[drm] drmAddMap(tex_handle) failed. Disabling DRI\n");
+      return FALSE;
+   }
+   fprintf(stderr, "[drm] textures = 0x%08x\n",
+	      sarea->tex_handle);
+#endif
+   return TRUE;
+}
+
+
+static void
+I830DRIUnmapScreenRegions(const DRIDriverContext *ctx, I830Rec *pI830, drmI830Sarea *sarea)
+{
+#if 1
+   if (sarea->front_handle) {
+      drmRmMap(ctx->drmFD, sarea->front_handle);
+      sarea->front_handle = 0;
+   }
+#endif
+   if (sarea->back_handle) {
+      drmRmMap(ctx->drmFD, sarea->back_handle);
+      sarea->back_handle = 0;
+   }
+   if (sarea->depth_handle) {
+      drmRmMap(ctx->drmFD, sarea->depth_handle);
+      sarea->depth_handle = 0;
+   }
+   if (sarea->tex_handle) {
+      drmRmMap(ctx->drmFD, sarea->tex_handle);
+      sarea->tex_handle = 0;
+   }
+}
+
+static Bool
+I830DRIDoMappings(DRIDriverContext *ctx, I830Rec *pI830, drmI830Sarea *sarea)
+{
+  if (drmAddMap(ctx->drmFD,
+		(drm_handle_t)pI830->LpRing->mem.Start,
+		pI830->LpRing->mem.Size, DRM_AGP, 0,
+		&pI830->ring_map) < 0) {
+    fprintf(stderr,
+	    "[drm] drmAddMap(ring_map) failed. Disabling DRI\n");
+    return FALSE;
+  }
+  fprintf(stderr, "[drm] ring buffer = 0x%08x\n",
+	  pI830->ring_map);
+
+  if (I830InitDma(ctx, pI830) == FALSE) {
+    return FALSE;
+  }
+  
+   /* init to zero to be safe */
+
+  I830DRIMapScreenRegions(ctx, pI830, sarea);
+  SetupDRIMM(ctx, pI830);
+
+   if (ctx->pciDevice != PCI_CHIP_845_G &&
+       ctx->pciDevice != PCI_CHIP_I830_M) {
+      I830SetParam(ctx, I830_SETPARAM_USE_MI_BATCHBUFFER_START, 1 );
+   }
+
+   /* Okay now initialize the dma engine */
+   {
+      pI830->irq = drmGetInterruptFromBusID(ctx->drmFD,
+					    ctx->pciBus,
+					    ctx->pciDevice,
+					    ctx->pciFunc);
+
+      if (drmCtlInstHandler(ctx->drmFD, pI830->irq)) {
+	 fprintf(stderr,
+		    "[drm] failure adding irq handler\n");
+	 pI830->irq = 0;
+	 return FALSE;
+      }
+      else
+	 fprintf(stderr,
+		    "[drm] dma control initialized, using IRQ %d\n",
+		    pI830->irq);
+   }
+
+   fprintf(stderr, "[dri] visual configs initialized\n");
+
+   return TRUE;
+}
+
+static Bool
+I830ClearScreen(DRIDriverContext *ctx, I830Rec *pI830, drmI830Sarea *sarea)
+{
+  /* need to drmMap front and back buffers and zero them */
+  drmAddress map_addr;
+  int ret;
+
+  ret = drmMap(ctx->drmFD,
+	       sarea->front_handle,
+	       sarea->front_size,
+	       &map_addr);
+
+  if (ret)
+  {
+    fprintf(stderr, "Unable to map front buffer\n");
+    return FALSE;
+  }
+
+  drimemsetio((char *)map_addr,
+	      0,
+	      sarea->front_size);
+  drmUnmap(map_addr, sarea->front_size);
+
+
+  ret = drmMap(ctx->drmFD,
+	       sarea->back_handle,
+	       sarea->back_size,
+	       &map_addr);
+
+  if (ret)
+  {
+    fprintf(stderr, "Unable to map back buffer\n");
+    return FALSE;
+  }
+
+  drimemsetio((char *)map_addr,
+	      0,
+	      sarea->back_size);
+  drmUnmap(map_addr, sarea->back_size);
+
+  return TRUE;
+}
+
+static Bool
+I830ScreenInit(DRIDriverContext *ctx, I830Rec *pI830)
+		  
+{
+   I830DRIPtr pI830DRI;
+   drmI830Sarea *pSAREAPriv;
+   int err;
+      
+   drm_page_size = getpagesize();   
+
+   pI830->registerSize = ctx->MMIOSize;
+   /* This is a hack for now.  We have to have more than a 4k page here
+    * because of the size of the state.  However, the state should be
+    * in a per-context mapping.  This will be added in the Mesa 3.5 port
+    * of the I830 driver.
+    */
+   ctx->shared.SAREASize = SAREA_MAX;
+
+   /* Note that drmOpen will try to load the kernel module, if needed. */
+   ctx->drmFD = drmOpen("i915", NULL );
+   if (ctx->drmFD < 0) {
+      fprintf(stderr, "[drm] drmOpen failed\n");
+      return 0;
+   }
+
+   if ((err = drmSetBusid(ctx->drmFD, ctx->pciBusID)) < 0) {
+      fprintf(stderr, "[drm] drmSetBusid failed (%d, %s), %s\n",
+	      ctx->drmFD, ctx->pciBusID, strerror(-err));
+      return 0;
+   }
+
+   if (drmAddMap( ctx->drmFD,
+		  0,
+		  ctx->shared.SAREASize,
+		  DRM_SHM,
+		  DRM_CONTAINS_LOCK,
+		  &ctx->shared.hSAREA) < 0)
+   {
+     fprintf(stderr, "[drm] drmAddMap failed\n");
+     return 0;
+   }
+
+   fprintf(stderr, "[drm] added %d byte SAREA at 0x%08x\n",
+	   ctx->shared.SAREASize, ctx->shared.hSAREA);
+   
+   if (drmMap( ctx->drmFD,
+	       ctx->shared.hSAREA,
+	       ctx->shared.SAREASize,
+	       (drmAddressPtr)(&ctx->pSAREA)) < 0)
+   {
+      fprintf(stderr, "[drm] drmMap failed\n");
+      return 0;
+   
+   }
+   
+   memset(ctx->pSAREA, 0, ctx->shared.SAREASize);
+   fprintf(stderr, "[drm] mapped SAREA 0x%08x to %p, size %d\n",
+	   ctx->shared.hSAREA, ctx->pSAREA, ctx->shared.SAREASize);
+   
+
+   if (drmAddMap(ctx->drmFD, 
+		 ctx->MMIOStart,
+		 ctx->MMIOSize,
+		 DRM_REGISTERS, 
+		 DRM_READ_ONLY, 
+		 &pI830->registerHandle) < 0) {
+      fprintf(stderr, "[drm] drmAddMap mmio failed\n");	
+      return 0;
+   }
+   fprintf(stderr,
+	   "[drm] register handle = 0x%08x\n", pI830->registerHandle);
+
+
+   if (!I830CheckDRMVersion(ctx, pI830)) {
+     return FALSE;
+   }
+
+   /* Create a 'server' context so we can grab the lock for
+    * initialization ioctls.
+    */
+   if ((err = drmCreateContext(ctx->drmFD, &ctx->serverContext)) != 0) {
+      fprintf(stderr, "%s: drmCreateContext failed %d\n", __FUNCTION__, err);
+      return 0;
+   }
+
+   DRM_LOCK(ctx->drmFD, ctx->pSAREA, ctx->serverContext, 0); 
+
+   /* Initialize the SAREA private data structure */
+   pSAREAPriv = (drmI830Sarea *)(((char*)ctx->pSAREA) + 
+				 sizeof(drm_sarea_t));
+   memset(pSAREAPriv, 0, sizeof(*pSAREAPriv));
+
+   pI830->StolenMemory.Size = I830DetectMemory(ctx, pI830);
+   pI830->StolenMemory.Start = 0;
+   pI830->StolenMemory.End = pI830->StolenMemory.Size;
+
+   pI830->MemoryAperture.Start = pI830->StolenMemory.End;
+   pI830->MemoryAperture.End = KB(40000);
+   pI830->MemoryAperture.Size = pI830->MemoryAperture.End - pI830->MemoryAperture.Start;
+
+   pI830->StolenPool.Fixed = pI830->StolenMemory;
+   pI830->StolenPool.Total = pI830->StolenMemory;
+   pI830->StolenPool.Free = pI830->StolenPool.Total;
+   pI830->FreeMemory = pI830->StolenPool.Total.Size;
+
+   if (!AgpInit(ctx, pI830))
+     return FALSE;
+
+   if (I830AllocateMemory(ctx, pI830) == FALSE)
+   {
+     return FALSE;
+   }
+
+   if (I830BindMemory(ctx, pI830) == FALSE)
+   {
+     return FALSE;
+   }
+
+   pSAREAPriv->rotated_offset = -1;
+   pSAREAPriv->rotated_size = 0;
+   pSAREAPriv->rotated_pitch = ctx->shared.virtualWidth;
+
+   pSAREAPriv->front_offset = pI830->FrontBuffer.Start;
+   pSAREAPriv->front_size = pI830->FrontBuffer.Size;
+   pSAREAPriv->width = ctx->shared.virtualWidth;
+   pSAREAPriv->height = ctx->shared.virtualHeight;
+   pSAREAPriv->pitch = ctx->shared.virtualWidth;
+   pSAREAPriv->virtualX = ctx->shared.virtualWidth;
+   pSAREAPriv->virtualY = ctx->shared.virtualHeight;
+   pSAREAPriv->back_offset = pI830->BackBuffer.Start;
+   pSAREAPriv->back_size = pI830->BackBuffer.Size;
+   pSAREAPriv->depth_offset = pI830->DepthBuffer.Start;
+   pSAREAPriv->depth_size = pI830->DepthBuffer.Size;
+#if 0
+   pSAREAPriv->tex_offset = pI830->TexMem.Start;
+   pSAREAPriv->tex_size = pI830->TexMem.Size;
+#endif
+   pSAREAPriv->log_tex_granularity = pI830->TexGranularity;
+
+   ctx->driverClientMsg = malloc(sizeof(I830DRIRec));
+   ctx->driverClientMsgSize = sizeof(I830DRIRec);
+   pI830DRI = (I830DRIPtr)ctx->driverClientMsg;
+   pI830DRI->deviceID = pI830->Chipset;
+   pI830DRI->regsSize = I830_REG_SIZE;
+   pI830DRI->width = ctx->shared.virtualWidth;
+   pI830DRI->height = ctx->shared.virtualHeight;
+   pI830DRI->mem = ctx->shared.fbSize;
+   pI830DRI->cpp = ctx->cpp;
+
+   pI830DRI->bitsPerPixel = ctx->bpp;
+   pI830DRI->sarea_priv_offset = sizeof(drm_sarea_t);
+   
+   err = I830DRIDoMappings(ctx, pI830, pSAREAPriv);
+   if (err == FALSE)
+       return FALSE;
+
+   I830SetupMemoryTiling(ctx, pI830);
+
+   /* Quick hack to clear the front & back buffers.  Could also use
+    * the clear ioctl to do this, but would need to setup hw state
+    * first.
+    */
+   I830ClearScreen(ctx, pI830, pSAREAPriv);
+
+   I830SetRingRegs(ctx, pI830);
+
+   return TRUE;
+}
+
+
+/**
+ * \brief Validate the fbdev mode.
+ * 
+ * \param ctx display handle.
+ *
+ * \return one on success, or zero on failure.
+ *
+ * Saves some registers and returns 1.
+ *
+ * \sa radeonValidateMode().
+ */
+static int i830ValidateMode( const DRIDriverContext *ctx )
+{
+  return 1;
+}
+
+/**
+ * \brief Examine mode returned by fbdev.
+ * 
+ * \param ctx display handle.
+ *
+ * \return one on success, or zero on failure.
+ *
+ * Restores registers that fbdev has clobbered and returns 1.
+ *
+ * \sa i810ValidateMode().
+ */
+static int i830PostValidateMode( const DRIDriverContext *ctx )
+{
+  I830Rec *pI830 = ctx->driverPrivate;
+
+  I830SetRingRegs(ctx, pI830);
+  return 1;
+}
+
+
+/**
+ * \brief Initialize the framebuffer device mode
+ *
+ * \param ctx display handle.
+ *
+ * \return one on success, or zero on failure.
+ *
+ * Fills in \p info with some default values and some information from \p ctx
+ * and then calls I810ScreenInit() for the screen initialization.
+ * 
+ * Before exiting clears the framebuffer memory accessing it directly.
+ */
+static int i830InitFBDev( DRIDriverContext *ctx )
+{
+  I830Rec *pI830 = calloc(1, sizeof(I830Rec));
+  int i;
+
+   {
+      int  dummy = ctx->shared.virtualWidth;
+
+      switch (ctx->bpp / 8) {
+      case 1: dummy = (ctx->shared.virtualWidth + 127) & ~127; break;
+      case 2: dummy = (ctx->shared.virtualWidth +  31) &  ~31; break;
+      case 3:
+      case 4: dummy = (ctx->shared.virtualWidth +  15) &  ~15; break;
+      }
+
+      ctx->shared.virtualWidth = dummy;
+      ctx->shared.Width = ctx->shared.virtualWidth;
+   }
+
+
+   for (i = 0; pitches[i] != 0; i++) {
+     if (pitches[i] >= ctx->shared.virtualWidth) {
+       ctx->shared.virtualWidth = pitches[i];
+       break;
+     }
+   }
+
+   ctx->driverPrivate = (void *)pI830;
+   
+   pI830->LpRing = calloc(1, sizeof(I830RingBuffer));
+   pI830->Chipset = ctx->chipset;
+   pI830->LinearAddr = ctx->FBStart;
+
+   if (!I830ScreenInit( ctx, pI830 ))
+      return 0;
+
+   
+   return 1;
+}
+
+
+/**
+ * \brief The screen is being closed, so clean up any state and free any
+ * resources used by the DRI.
+ *
+ * \param ctx display handle.
+ *
+ * Unmaps the SAREA, closes the DRM device file descriptor and frees the driver
+ * private data.
+ */
+static void i830HaltFBDev( DRIDriverContext *ctx )
+{
+  drmI830Sarea *pSAREAPriv;
+  I830Rec *pI830 = ctx->driverPrivate;
+
+   if (pI830->irq) {
+       drmCtlUninstHandler(ctx->drmFD);
+       pI830->irq = 0;   }
+
+   I830CleanupDma(ctx);
+
+  pSAREAPriv = (drmI830Sarea *)(((char*)ctx->pSAREA) + 
+				sizeof(drm_sarea_t));
+
+  I830DRIUnmapScreenRegions(ctx, pI830, pSAREAPriv);
+  drmUnmap( ctx->pSAREA, ctx->shared.SAREASize );
+  drmClose(ctx->drmFD);
+  
+  if (ctx->driverPrivate) {
+    free(ctx->driverPrivate);
+    ctx->driverPrivate = 0;
+  }
+}
+
+
+extern void i810NotifyFocus( int );
+
+/**
+ * \brief Exported driver interface for Mini GLX.
+ *
+ * \sa DRIDriverRec.
+ */
+const struct DRIDriverRec __driDriver = {
+   i830ValidateMode,
+   i830PostValidateMode,
+   i830InitFBDev,
+   i830HaltFBDev,
+   NULL,//I830EngineShutdown,
+   NULL, //I830EngineRestore,  
+#ifndef _EMBEDDED
+   0,
+#else
+   i810NotifyFocus, 
+#endif
+};
diff --git a/src/gallium/winsys/xlib/brw_aub.c b/src/gallium/winsys/xlib/brw_aub.c
new file mode 100644
index 0000000000..541d50c6e4
--- /dev/null
+++ b/src/gallium/winsys/xlib/brw_aub.c
@@ -0,0 +1,392 @@
+/*
+ Copyright (C) Intel Corp.  2006.  All Rights Reserved.
+ Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to
+ develop this 3D driver.
+ 
+ Permission is hereby granted, free of charge, to any person obtaining
+ a copy of this software and associated documentation files (the
+ "Software"), to deal in the Software without restriction, including
+ without limitation the rights to use, copy, modify, merge, publish,
+ distribute, sublicense, and/or sell copies of the Software, and to
+ permit persons to whom the Software is furnished to do so, subject to
+ the following conditions:
+ 
+ The above copyright notice and this permission notice (including the
+ next paragraph) shall be included in all copies or substantial
+ portions of the Software.
+ 
+ THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
+ IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
+ LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
+ OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ 
+ **********************************************************************/
+ /*
+  * Authors:
+  *   Keith Whitwell <keith@tungstengraphics.com>
+  */
+
+#include "brw_aub.h"
+#include "pipe/p_context.h"
+#include "pipe/p_state.h"
+#include "imports.h"
+//#include "intel_winsys.h"
+
+
+struct brw_aubfile {
+   FILE *file;
+   unsigned next_free_page;
+};
+
+
+extern char *__progname;
+
+
+struct aub_file_header {
+   unsigned int instruction_type;
+   unsigned int pad0:16;
+   unsigned int minor:8;
+   unsigned int major:8;
+   unsigned char application[8*4];
+   unsigned int day:8;
+   unsigned int month:8;
+   unsigned int year:16;
+   unsigned int timezone:8;
+   unsigned int second:8;
+   unsigned int minute:8;
+   unsigned int hour:8;
+   unsigned int comment_length:16;   
+   unsigned int pad1:16;
+};
+
+struct aub_block_header {
+   unsigned int instruction_type;
+   unsigned int operation:8;
+   unsigned int type:8;
+   unsigned int address_space:8;
+   unsigned int pad0:8;
+   unsigned int general_state_type:8;
+   unsigned int surface_state_type:8;
+   unsigned int pad1:16;
+   unsigned int address;
+   unsigned int length;
+};
+
+struct aub_dump_bmp {
+   unsigned int instruction_type;
+   unsigned int xmin:16;
+   unsigned int ymin:16;
+   unsigned int pitch:16;
+   unsigned int bpp:8;
+   unsigned int format:8;
+   unsigned int xsize:16;
+   unsigned int ysize:16;
+   unsigned int addr;
+   unsigned int unknown;
+};
+
+enum bh_operation {
+   BH_COMMENT,
+   BH_DATA_WRITE,
+   BH_COMMAND_WRITE,
+   BH_MMI0_WRITE32,
+   BH_END_SCENE,
+   BH_CONFIG_MEMORY_MAP,
+   BH_MAX_OPERATION
+};
+
+enum command_write_type {
+   CW_HWB_RING = 1,
+   CW_PRIMARY_RING_A,
+   CW_PRIMARY_RING_B,		/* XXX - disagreement with listaub! */
+   CW_PRIMARY_RING_C,
+   CW_MAX_TYPE
+};
+
+enum memory_map_type {
+   MM_DEFAULT,
+   MM_DYNAMIC,
+   MM_MAX_TYPE
+};
+
+enum address_space {
+   ADDR_GTT,
+   ADDR_LOCAL,
+   ADDR_MAIN,
+   ADDR_MAX
+};
+
+
+#define AUB_FILE_HEADER 0xe085000b
+#define AUB_BLOCK_HEADER 0xe0c10003
+#define AUB_DUMP_BMP 0xe09e0004
+
+/* Registers to control page table
+ */
+#define PGETBL_CTL       0x2020
+#define PGETBL_ENABLED   0x1
+
+#define NR_GTT_ENTRIES  65536	/* 256 mb */
+
+#define FAIL										\
+do {											\
+   fprintf(stderr, "failed to write aub data at %s/%d\n", __FUNCTION__, __LINE__);	\
+   exit(1);										\
+} while (0)
+
+
+/* Emit the headers at the top of each aubfile.  Initialize the GTT.
+ */
+static void init_aubfile( FILE *aub_file )
+{   
+   struct aub_file_header fh;
+   struct aub_block_header bh;
+   unsigned int data;
+
+   static int nr;
+   
+   nr++;
+
+   /* Emit the aub header:
+    */
+   memset(&fh, 0, sizeof(fh));
+
+   fh.instruction_type = AUB_FILE_HEADER;
+   fh.minor = 0x0;
+   fh.major = 0x7;
+   memcpy(fh.application, __progname, sizeof(fh.application));
+   fh.day = (nr>>24) & 0xff;
+   fh.month = 0x0;
+   fh.year = 0x0;
+   fh.timezone = 0x0;
+   fh.second = nr & 0xff;
+   fh.minute = (nr>>8) & 0xff;
+   fh.hour = (nr>>16) & 0xff;
+   fh.comment_length = 0x0;   
+
+   if (fwrite(&fh, sizeof(fh), 1, aub_file) < 0) 
+      FAIL;
+         
+   /* Setup the GTT starting at main memory address zero (!):
+    */
+   memset(&bh, 0, sizeof(bh));
+   
+   bh.instruction_type = AUB_BLOCK_HEADER;
+   bh.operation = BH_MMI0_WRITE32;
+   bh.type = 0x0;
+   bh.address_space = ADDR_GTT;	/* ??? */
+   bh.general_state_type = 0x0;
+   bh.surface_state_type = 0x0;
+   bh.address = PGETBL_CTL;
+   bh.length = 0x4;
+
+   if (fwrite(&bh, sizeof(bh), 1, aub_file) < 0) 
+      FAIL;
+
+   data = 0x0 | PGETBL_ENABLED;
+
+   if (fwrite(&data, sizeof(data), 1, aub_file) < 0) 
+      FAIL;
+}
+
+
+static void init_aub_gtt( struct brw_aubfile *aubfile,
+			  unsigned start_offset, 
+			  unsigned size )
+{
+   FILE *aub_file = aubfile->file;
+   struct aub_block_header bh;
+   unsigned int i;
+
+   assert(start_offset + size < NR_GTT_ENTRIES * 4096);
+
+
+   memset(&bh, 0, sizeof(bh));
+   
+   bh.instruction_type = AUB_BLOCK_HEADER;
+   bh.operation = BH_DATA_WRITE;
+   bh.type = 0x0;
+   bh.address_space = ADDR_MAIN;
+   bh.general_state_type = 0x0;
+   bh.surface_state_type = 0x0;
+   bh.address =  start_offset / 4096 * 4;
+   bh.length = size / 4096 * 4;
+
+   if (fwrite(&bh, sizeof(bh), 1, aub_file) < 0) 
+      FAIL;
+
+   for (i = 0; i < size / 4096; i++) {
+      unsigned data = aubfile->next_free_page | 1;
+
+      aubfile->next_free_page += 4096;
+
+      if (fwrite(&data, sizeof(data), 1, aub_file) < 0) 
+	 FAIL;
+   }
+
+}
+
+static void write_block_header( FILE *aub_file,
+				struct aub_block_header *bh,
+				const unsigned *data,
+				unsigned sz )
+{
+   sz = (sz + 3) & ~3;
+
+   if (fwrite(bh, sizeof(*bh), 1, aub_file) < 0) 
+      FAIL;
+
+   if (fwrite(data, sz, 1, aub_file) < 0) 
+      FAIL;
+
+   fflush(aub_file);
+}
+
+
+static void write_dump_bmp( FILE *aub_file,
+			    struct aub_dump_bmp *db )
+{
+   if (fwrite(db, sizeof(*db), 1, aub_file) < 0) 
+      FAIL;
+
+   fflush(aub_file);
+}
+
+
+
+void brw_aub_gtt_data( struct brw_aubfile *aubfile,
+		       unsigned offset,
+		       const void *data,
+		       unsigned sz,
+		       unsigned type,
+		       unsigned state_type )
+{
+   struct aub_block_header bh;
+
+   bh.instruction_type = AUB_BLOCK_HEADER;
+   bh.operation = BH_DATA_WRITE;
+   bh.type = type;
+   bh.address_space = ADDR_GTT;
+   bh.pad0 = 0;
+
+   if (type == DW_GENERAL_STATE) {
+      bh.general_state_type = state_type;
+      bh.surface_state_type = 0;
+   }
+   else {
+      bh.general_state_type = 0;
+      bh.surface_state_type = state_type;
+   }
+
+   bh.pad1 = 0;
+   bh.address = offset;
+   bh.length = sz;
+
+   write_block_header(aubfile->file, &bh, data, sz);
+}
+
+
+
+void brw_aub_gtt_cmds( struct brw_aubfile *aubfile,
+		       unsigned offset,
+		       const void *data,
+		       unsigned sz )
+{
+   struct aub_block_header bh;   
+   unsigned type = CW_PRIMARY_RING_A;
+   
+
+   bh.instruction_type = AUB_BLOCK_HEADER;
+   bh.operation = BH_COMMAND_WRITE;
+   bh.type = type;
+   bh.address_space = ADDR_GTT;
+   bh.pad0 = 0;
+   bh.general_state_type = 0;
+   bh.surface_state_type = 0;
+   bh.pad1 = 0;
+   bh.address = offset;
+   bh.length = sz;
+
+   write_block_header(aubfile->file, &bh, data, sz);
+}
+
+void brw_aub_dump_bmp( struct brw_aubfile *aubfile,
+		       struct pipe_surface *surface,
+		       unsigned gtt_offset )
+{
+   struct aub_dump_bmp db;
+   unsigned format;
+
+   if (surface->cpp == 4)
+      format = 0x7;
+   else
+      format = 0x3;
+
+   db.instruction_type = AUB_DUMP_BMP;
+   db.xmin = 0;
+   db.ymin = 0;
+   db.format = format;
+   db.bpp = surface->cpp * 8;
+   db.pitch = surface->pitch;
+   db.xsize = surface->width;
+   db.ysize = surface->height;
+   db.addr = gtt_offset;
+   db.unknown = /* surface->tiled ? 0x4 : */ 0x0;
+
+   write_dump_bmp(aubfile->file, &db);
+}
+
+
+
+struct brw_aubfile *brw_aubfile_create( void )
+{
+   struct brw_aubfile *aubfile = CALLOC_STRUCT(brw_aubfile);
+   char filename[80];
+   int val;
+   static int i = 0;
+
+   i++;
+
+   if (_mesa_getenv("INTEL_AUBFILE")) {
+      val = snprintf(filename, sizeof(filename), "%s%d.aub", _mesa_getenv("INTEL_AUBFILE"), i%4);
+      _mesa_printf("--> Aub file: %s\n", filename);
+      aubfile->file = fopen(filename, "w");
+   }
+   else {
+      val = snprintf(filename, sizeof(filename), "%s.aub", __progname);
+      if (val < 0 || val > sizeof(filename)) 
+	 strcpy(filename, "default.aub");   
+   
+      _mesa_printf("--> Aub file: %s\n", filename);
+      aubfile->file = fopen(filename, "w");
+   }
+
+   if (!aubfile->file) {
+      _mesa_printf("couldn't open aubfile\n");
+      exit(1);
+   }
+
+   init_aubfile(aubfile->file);
+
+   /* The GTT is located starting address zero in main memory.  Pages
+    * to populate the gtt start after this point.
+    */
+   aubfile->next_free_page = (NR_GTT_ENTRIES * 4 + 4095) & ~4095;
+
+   /* More or less correspond with all the agp regions mapped by the
+    * driver:
+    */
+   init_aub_gtt(aubfile, 0, 4096*4);
+   init_aub_gtt(aubfile, AUB_BUF_START, AUB_BUF_SIZE);
+
+   return aubfile;
+}
+
+void brw_aub_destroy( struct brw_aubfile *aubfile )
+{
+   fclose(aubfile->file);
+   FREE(aubfile);
+}
diff --git a/src/gallium/winsys/xlib/brw_aub.h b/src/gallium/winsys/xlib/brw_aub.h
new file mode 100644
index 0000000000..f5c60c7be2
--- /dev/null
+++ b/src/gallium/winsys/xlib/brw_aub.h
@@ -0,0 +1,114 @@
+/*
+ Copyright (C) Intel Corp.  2006.  All Rights Reserved.
+ Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to
+ develop this 3D driver.
+ 
+ Permission is hereby granted, free of charge, to any person obtaining
+ a copy of this software and associated documentation files (the
+ "Software"), to deal in the Software without restriction, including
+ without limitation the rights to use, copy, modify, merge, publish,
+ distribute, sublicense, and/or sell copies of the Software, and to
+ permit persons to whom the Software is furnished to do so, subject to
+ the following conditions:
+ 
+ The above copyright notice and this permission notice (including the
+ next paragraph) shall be included in all copies or substantial
+ portions of the Software.
+ 
+ THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
+ IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
+ LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
+ OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ 
+ **********************************************************************/
+ /*
+  * Authors:
+  *   Keith Whitwell <keith@tungstengraphics.com>
+  */
+
+#ifndef BRW_AUB_H
+#define BRW_AUB_H
+
+/* We set up this region, buffers may be allocated here:
+ */
+#define AUB_BUF_START (4096*4)
+#define AUB_BUF_SIZE  (8*1024*1024)
+
+struct intel_context;
+struct pipe_surface;
+
+struct brw_aubfile *brw_aubfile_create( void );
+
+void brw_aub_destroy( struct brw_aubfile *aubfile );
+
+void brw_aub_gtt_data( struct brw_aubfile *aubfile,
+		       unsigned offset,
+		       const void *data,
+		       unsigned sz,
+		       unsigned type,
+		       unsigned state_type );
+
+void brw_aub_gtt_cmds( struct brw_aubfile *aubfile,
+		       unsigned offset,
+		       const void *data,
+		       unsigned sz );
+
+void brw_aub_dump_bmp( struct brw_aubfile *aubfile,
+		       struct pipe_surface *surface,
+		       unsigned gtt_offset );
+
+
+enum data_write_type {
+   DW_NOTYPE,
+   DW_BATCH_BUFFER,
+   DW_BIN_BUFFER,
+   DW_BIN_POINTER_LIST,
+   DW_SLOW_STATE_BUFFER,
+   DW_VERTEX_BUFFER,
+   DW_2D_MAP,
+   DW_CUBE_MAP,
+   DW_INDIRECT_STATE_BUFFER,
+   DW_VOLUME_MAP,
+   DW_1D_MAP,
+   DW_CONSTANT_BUFFER,
+   DW_CONSTANT_URB_ENTRY,
+   DW_INDEX_BUFFER,
+   DW_GENERAL_STATE,
+   DW_SURFACE_STATE,
+   DW_MEDIA_OBJECT_INDIRECT_DATA,
+   DW_MAX_TYPE
+};
+
+enum data_write_general_state_type {
+   DWGS_NOTYPE,
+   DWGS_VERTEX_SHADER_STATE,
+   DWGS_GEOMETRY_SHADER_STATE ,
+   DWGS_CLIPPER_STATE,
+   DWGS_STRIPS_FANS_STATE,
+   DWGS_WINDOWER_IZ_STATE,
+   DWGS_COLOR_CALC_STATE,
+   DWGS_CLIPPER_VIEWPORT_STATE,	/* was 0x7 */
+   DWGS_STRIPS_FANS_VIEWPORT_STATE,
+   DWGS_COLOR_CALC_VIEWPORT_STATE, /* was 0x9 */
+   DWGS_SAMPLER_STATE,
+   DWGS_KERNEL_INSTRUCTIONS,
+   DWGS_SCRATCH_SPACE,
+   DWGS_SAMPLER_DEFAULT_COLOR,
+   DWGS_INTERFACE_DESCRIPTOR,
+   DWGS_VLD_STATE,
+   DWGS_VFE_STATE,
+   DWGS_MAX_TYPE
+};
+
+enum data_write_surface_state_type {
+   DWSS_NOTYPE,
+   DWSS_BINDING_TABLE_STATE,
+   DWSS_SURFACE_STATE,
+   DWSS_MAX_TYPE
+};
+
+
+#endif
diff --git a/src/gallium/winsys/xlib/fakeglx.c b/src/gallium/winsys/xlib/fakeglx.c
new file mode 100644
index 0000000000..902a755075
--- /dev/null
+++ b/src/gallium/winsys/xlib/fakeglx.c
@@ -0,0 +1,3188 @@
+/*
+ * Mesa 3-D graphics library
+ * Version:  7.1
+ *
+ * Copyright (C) 1999-2007  Brian Paul   All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included
+ * in all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * BRIAN PAUL BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN
+ * AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ */
+
+
+/*
+ * This is an emulation of the GLX API which allows Mesa/GLX-based programs
+ * to run on X servers which do not have the real GLX extension.
+ *
+ * Thanks to the contributors:
+ *
+ * Initial version:  Philip Brown (phil@bolthole.com)
+ * Better glXGetConfig() support: Armin Liebchen (liebchen@asylum.cs.utah.edu)
+ * Further visual-handling refinements: Wolfram Gloger
+ *    (wmglo@Dent.MED.Uni-Muenchen.DE).
+ *
+ * Notes:
+ *   Don't be fooled, stereo isn't supported yet.
+ */
+
+
+
+#include "glxheader.h"
+#include "glxapi.h"
+#include "GL/xmesa.h"
+#include "context.h"
+#include "config.h"
+#include "macros.h"
+#include "imports.h"
+#include "mtypes.h"
+#include "version.h"
+#include "xfonts.h"
+#include "xmesaP.h"
+#include "state_tracker/st_context.h"
+#include "state_tracker/st_public.h"
+
+
+#ifdef __VMS
+#define _mesa_sprintf sprintf
+#endif
+
+/* This indicates the client-side GLX API and GLX encoder version. */
+#define CLIENT_MAJOR_VERSION 1
+#define CLIENT_MINOR_VERSION 4  /* but don't have 1.3's pbuffers, etc yet */
+
+/* This indicates the server-side GLX decoder version.
+ * GLX 1.4 indicates OpenGL 1.3 support
+ */
+#define SERVER_MAJOR_VERSION 1
+#define SERVER_MINOR_VERSION 4
+
+/* This is appended onto the glXGetClient/ServerString version strings. */
+#define MESA_GLX_VERSION "Mesa " MESA_VERSION_STRING
+
+/* Who implemented this GLX? */
+#define VENDOR "Brian Paul"
+
+#define EXTENSIONS \
+   "GLX_MESA_set_3dfx_mode " \
+   "GLX_MESA_copy_sub_buffer " \
+   "GLX_MESA_pixmap_colormap " \
+   "GLX_MESA_release_buffers " \
+   "GLX_ARB_get_proc_address " \
+   "GLX_EXT_texture_from_pixmap " \
+   "GLX_EXT_visual_info " \
+   "GLX_EXT_visual_rating " \
+   /*"GLX_SGI_video_sync "*/ \
+   "GLX_SGIX_fbconfig " \
+   "GLX_SGIX_pbuffer "
+
+/*
+ * Our fake GLX context will contain a "real" GLX context and an XMesa context.
+ *
+ * Note that a pointer to a __GLXcontext is a pointer to a fake_glx_context,
+ * and vice versa.
+ *
+ * We really just need this structure in order to make the libGL functions
+ * glXGetCurrentContext(), glXGetCurrentDrawable() and glXGetCurrentDisplay()
+ * work correctly.
+ */
+struct fake_glx_context {
+   __GLXcontext glxContext;   /* this MUST be first! */
+   XMesaContext xmesaContext;
+};
+
+
+
+/**********************************************************************/
+/***                       GLX Visual Code                          ***/
+/**********************************************************************/
+
+#define DONT_CARE -1
+
+
+static XMesaVisual *VisualTable = NULL;
+static int NumVisuals = 0;
+
+
+/*
+ * This struct and some code fragments borrowed
+ * from Mark Kilgard's GLUT library.
+ */
+typedef struct _OverlayInfo {
+  /* Avoid 64-bit portability problems by being careful to use
+     longs due to the way XGetWindowProperty is specified. Note
+     that these parameters are passed as CARD32s over X
+     protocol. */
+  unsigned long overlay_visual;
+  long transparent_type;
+  long value;
+  long layer;
+} OverlayInfo;
+
+
+
+/* Macro to handle c_class vs class field name in XVisualInfo struct */
+#if defined(__cplusplus) || defined(c_plusplus)
+#define CLASS c_class
+#else
+#define CLASS class
+#endif
+
+
+
+/*
+ * Test if the given XVisualInfo is usable for Mesa rendering.
+ */
+static GLboolean
+is_usable_visual( XVisualInfo *vinfo )
+{
+   switch (vinfo->CLASS) {
+      case StaticGray:
+      case GrayScale:
+         /* Any StaticGray/GrayScale visual works in RGB or CI mode */
+         return GL_TRUE;
+      case StaticColor:
+      case PseudoColor:
+	 /* Any StaticColor/PseudoColor visual of at least 4 bits */
+	 if (vinfo->depth>=4) {
+	    return GL_TRUE;
+	 }
+	 else {
+	    return GL_FALSE;
+	 }
+      case TrueColor:
+      case DirectColor:
+	 /* Any depth of TrueColor or DirectColor works in RGB mode */
+	 return GL_TRUE;
+      default:
+	 /* This should never happen */
+	 return GL_FALSE;
+   }
+}
+
+
+
+/**
+ * Get an array OverlayInfo records for specified screen.
+ * \param dpy  the display
+ * \param screen  screen number
+ * \param numOverlays  returns numver of OverlayInfo records
+ * \return  pointer to OverlayInfo array, free with XFree()
+ */
+static OverlayInfo *
+GetOverlayInfo(Display *dpy, int screen, int *numOverlays)
+{
+   Atom overlayVisualsAtom;
+   Atom actualType;
+   Status status;
+   unsigned char *ovInfo;
+   unsigned long sizeData, bytesLeft;
+   int actualFormat;
+
+   /*
+    * The SERVER_OVERLAY_VISUALS property on the root window contains
+    * a list of overlay visuals.  Get that list now.
+    */
+   overlayVisualsAtom = XInternAtom(dpy,"SERVER_OVERLAY_VISUALS", True);
+   if (overlayVisualsAtom == None) {
+      return 0;
+   }
+
+   status = XGetWindowProperty(dpy, RootWindow(dpy, screen),
+                               overlayVisualsAtom, 0L, (long) 10000, False,
+                               overlayVisualsAtom, &actualType, &actualFormat,
+                               &sizeData, &bytesLeft,
+                               &ovInfo);
+
+   if (status != Success || actualType != overlayVisualsAtom ||
+       actualFormat != 32 || sizeData < 4) {
+      /* something went wrong */
+      XFree((void *) ovInfo);
+      *numOverlays = 0;
+      return NULL;
+   }
+
+   *numOverlays = sizeData / 4;
+   return (OverlayInfo *) ovInfo;
+}
+
+
+
+/**
+ * Return the level (overlay, normal, underlay) of a given XVisualInfo.
+ * Input:  dpy - the X display
+ *         vinfo - the XVisualInfo to test
+ * Return:  level of the visual:
+ *             0 = normal planes
+ *            >0 = overlay planes
+ *            <0 = underlay planes
+ */
+static int
+level_of_visual( Display *dpy, XVisualInfo *vinfo )
+{
+   OverlayInfo *overlay_info;
+   int numOverlaysPerScreen, i;
+
+   overlay_info = GetOverlayInfo(dpy, vinfo->screen, &numOverlaysPerScreen);
+   if (!overlay_info) {
+      return 0;
+   }
+
+   /* search the overlay visual list for the visual ID of interest */
+   for (i = 0; i < numOverlaysPerScreen; i++) {
+      const OverlayInfo *ov = overlay_info + i;
+      if (ov->overlay_visual == vinfo->visualid) {
+         /* found the visual */
+         if (/*ov->transparent_type==1 &&*/ ov->layer!=0) {
+            int level = ov->layer;
+            XFree((void *) overlay_info);
+            return level;
+         }
+         else {
+            XFree((void *) overlay_info);
+            return 0;
+         }
+      }
+   }
+
+   /* The visual ID was not found in the overlay list. */
+   XFree((void *) overlay_info);
+   return 0;
+}
+
+
+
+
+/*
+ * Given an XVisualInfo and RGB, Double, and Depth buffer flags, save the
+ * configuration in our list of GLX visuals.
+ */
+static XMesaVisual
+save_glx_visual( Display *dpy, XVisualInfo *vinfo,
+                 GLboolean rgbFlag, GLboolean alphaFlag, GLboolean dbFlag,
+                 GLboolean stereoFlag,
+                 GLint depth_size, GLint stencil_size,
+                 GLint accumRedSize, GLint accumGreenSize,
+                 GLint accumBlueSize, GLint accumAlphaSize,
+                 GLint level, GLint numAuxBuffers )
+{
+   GLboolean ximageFlag = GL_TRUE;
+   XMesaVisual xmvis;
+   GLint i;
+   GLboolean comparePointers;
+
+   if (dbFlag) {
+      /* Check if the MESA_BACK_BUFFER env var is set */
+      char *backbuffer = _mesa_getenv("MESA_BACK_BUFFER");
+      if (backbuffer) {
+         if (backbuffer[0]=='p' || backbuffer[0]=='P') {
+            ximageFlag = GL_FALSE;
+         }
+         else if (backbuffer[0]=='x' || backbuffer[0]=='X') {
+            ximageFlag = GL_TRUE;
+         }
+         else {
+            _mesa_warning(NULL, "Mesa: invalid value for MESA_BACK_BUFFER environment variable, using an XImage.");
+         }
+      }
+   }
+
+   if (stereoFlag) {
+      /* stereo not supported */
+      return NULL;
+   }
+
+   /* Comparing IDs uses less memory but sometimes fails. */
+   /* XXX revisit this after 3.0 is finished. */
+   if (_mesa_getenv("MESA_GLX_VISUAL_HACK"))
+      comparePointers = GL_TRUE;
+   else
+      comparePointers = GL_FALSE;
+
+   /* Force the visual to have an alpha channel */
+   if (rgbFlag && _mesa_getenv("MESA_GLX_FORCE_ALPHA"))
+      alphaFlag = GL_TRUE;
+
+   /* First check if a matching visual is already in the list */
+   for (i=0; i<NumVisuals; i++) {
+      XMesaVisual v = VisualTable[i];
+      if (v->display == dpy
+          && v->mesa_visual.level == level
+          && v->mesa_visual.numAuxBuffers == numAuxBuffers
+          && v->ximage_flag == ximageFlag
+          && v->mesa_visual.rgbMode == rgbFlag
+          && v->mesa_visual.doubleBufferMode == dbFlag
+          && v->mesa_visual.stereoMode == stereoFlag
+          && (v->mesa_visual.alphaBits > 0) == alphaFlag
+          && (v->mesa_visual.depthBits >= depth_size || depth_size == 0)
+          && (v->mesa_visual.stencilBits >= stencil_size || stencil_size == 0)
+          && (v->mesa_visual.accumRedBits >= accumRedSize || accumRedSize == 0)
+          && (v->mesa_visual.accumGreenBits >= accumGreenSize || accumGreenSize == 0)
+          && (v->mesa_visual.accumBlueBits >= accumBlueSize || accumBlueSize == 0)
+          && (v->mesa_visual.accumAlphaBits >= accumAlphaSize || accumAlphaSize == 0)) {
+         /* now either compare XVisualInfo pointers or visual IDs */
+         if ((!comparePointers && v->visinfo->visualid == vinfo->visualid)
+             || (comparePointers && v->vishandle == vinfo)) {
+            return v;
+         }
+      }
+   }
+
+   /* Create a new visual and add it to the list. */
+
+   xmvis = XMesaCreateVisual( dpy, vinfo, rgbFlag, alphaFlag, dbFlag,
+                              stereoFlag, ximageFlag,
+                              depth_size, stencil_size,
+                              accumRedSize, accumBlueSize,
+                              accumBlueSize, accumAlphaSize, 0, level,
+                              GLX_NONE_EXT );
+   if (xmvis) {
+      /* Save a copy of the pointer now so we can find this visual again
+       * if we need to search for it in find_glx_visual().
+       */
+      xmvis->vishandle = vinfo;
+      /* Allocate more space for additional visual */
+      VisualTable = (XMesaVisual *) _mesa_realloc( VisualTable, 
+                                   sizeof(XMesaVisual) * NumVisuals, 
+                                   sizeof(XMesaVisual) * (NumVisuals + 1));
+      /* add xmvis to the list */
+      VisualTable[NumVisuals] = xmvis;
+      NumVisuals++;
+      /* XXX minor hack, because XMesaCreateVisual doesn't support an
+       * aux buffers parameter.
+       */
+      xmvis->mesa_visual.numAuxBuffers = numAuxBuffers;
+   }
+   return xmvis;
+}
+
+
+/**
+ * Return the default number of bits for the Z buffer.
+ * If defined, use the MESA_GLX_DEPTH_BITS env var value.
+ * Otherwise, use the DEFAULT_SOFTWARE_DEPTH_BITS constant.
+ * XXX probably do the same thing for stencil, accum, etc.
+ */
+static GLint
+default_depth_bits(void)
+{
+   int zBits;
+   const char *zEnv = _mesa_getenv("MESA_GLX_DEPTH_BITS");
+   if (zEnv)
+      zBits = _mesa_atoi(zEnv);
+   else
+      zBits = DEFAULT_SOFTWARE_DEPTH_BITS;
+   return zBits;
+}
+
+static GLint
+default_alpha_bits(void)
+{
+   int aBits;
+   const char *aEnv = _mesa_getenv("MESA_GLX_ALPHA_BITS");
+   if (aEnv)
+      aBits = _mesa_atoi(aEnv);
+   else
+      aBits = 0;
+   return aBits;
+}
+
+static GLint
+default_accum_bits(void)
+{
+   return 16;
+}
+
+
+
+/*
+ * Create a GLX visual from a regular XVisualInfo.
+ * This is called when Fake GLX is given an XVisualInfo which wasn't
+ * returned by glXChooseVisual.  Since this is the first time we're
+ * considering this visual we'll take a guess at reasonable values
+ * for depth buffer size, stencil size, accum size, etc.
+ * This is the best we can do with a client-side emulation of GLX.
+ */
+static XMesaVisual
+create_glx_visual( Display *dpy, XVisualInfo *visinfo )
+{
+   int vislevel;
+   GLint zBits = 24; /*default_depth_bits();*/
+   GLint accBits = default_accum_bits();
+   GLboolean alphaFlag = default_alpha_bits() > 0;
+
+   vislevel = level_of_visual( dpy, visinfo );
+   if (vislevel) {
+      /* Configure this visual as a CI, single-buffered overlay */
+      return save_glx_visual( dpy, visinfo,
+                              GL_FALSE,  /* rgb */
+                              GL_FALSE,  /* alpha */
+                              GL_FALSE,  /* double */
+                              GL_FALSE,  /* stereo */
+                              0,         /* depth bits */
+                              0,         /* stencil bits */
+                              0,0,0,0,   /* accum bits */
+                              vislevel,  /* level */
+                              0          /* numAux */
+                            );
+   }
+   else if (is_usable_visual( visinfo )) {
+      if (_mesa_getenv("MESA_GLX_FORCE_CI")) {
+         /* Configure this visual as a COLOR INDEX visual. */
+         return save_glx_visual( dpy, visinfo,
+                                 GL_FALSE,   /* rgb */
+                                 GL_FALSE,  /* alpha */
+                                 GL_TRUE,   /* double */
+                                 GL_FALSE,  /* stereo */
+                                 zBits,
+                                 STENCIL_BITS,
+                                 0, 0, 0, 0, /* accum bits */
+                                 0,         /* level */
+                                 0          /* numAux */
+                               );
+      }
+      else {
+         /* Configure this visual as RGB, double-buffered, depth-buffered. */
+         /* This is surely wrong for some people's needs but what else */
+         /* can be done?  They should use glXChooseVisual(). */
+         return save_glx_visual( dpy, visinfo,
+                                 GL_TRUE,   /* rgb */
+                                 alphaFlag, /* alpha */
+                                 GL_TRUE,   /* double */
+                                 GL_FALSE,  /* stereo */
+                                 zBits,
+                                 STENCIL_BITS,
+                                 accBits, /* r */
+                                 accBits, /* g */
+                                 accBits, /* b */
+                                 accBits, /* a */
+                                 0,         /* level */
+                                 0          /* numAux */
+                               );
+      }
+   }
+   else {
+      _mesa_warning(NULL, "Mesa: error in glXCreateContext: bad visual\n");
+      return NULL;
+   }
+}
+
+
+
+/*
+ * Find the GLX visual associated with an XVisualInfo.
+ */
+static XMesaVisual
+find_glx_visual( Display *dpy, XVisualInfo *vinfo )
+{
+   int i;
+
+   /* try to match visual id */
+   for (i=0;i<NumVisuals;i++) {
+      if (VisualTable[i]->display==dpy
+          && VisualTable[i]->visinfo->visualid == vinfo->visualid) {
+         return VisualTable[i];
+      }
+   }
+
+   /* if that fails, try to match pointers */
+   for (i=0;i<NumVisuals;i++) {
+      if (VisualTable[i]->display==dpy && VisualTable[i]->vishandle==vinfo) {
+         return VisualTable[i];
+      }
+   }
+
+   return NULL;
+}
+
+
+
+/**
+ * Return the transparent pixel value for a GLX visual.
+ * Input:  glxvis - the glx_visual
+ * Return:  a pixel value or -1 if no transparent pixel
+ */
+static int
+transparent_pixel( XMesaVisual glxvis )
+{
+   Display *dpy = glxvis->display;
+   XVisualInfo *vinfo = glxvis->visinfo;
+   OverlayInfo *overlay_info;
+   int numOverlaysPerScreen, i;
+
+   overlay_info = GetOverlayInfo(dpy, vinfo->screen, &numOverlaysPerScreen);
+   if (!overlay_info) {
+      return -1;
+   }
+
+   for (i = 0; i < numOverlaysPerScreen; i++) {
+      const OverlayInfo *ov = overlay_info + i;
+      if (ov->overlay_visual == vinfo->visualid) {
+         /* found it! */
+         if (ov->transparent_type == 0) {
+            /* type 0 indicates no transparency */
+            XFree((void *) overlay_info);
+            return -1;
+         }
+         else {
+            /* ov->value is the transparent pixel */
+            XFree((void *) overlay_info);
+            return ov->value;
+         }
+      }
+   }
+
+   /* The visual ID was not found in the overlay list. */
+   XFree((void *) overlay_info);
+   return -1;
+}
+
+
+
+/**
+ * Try to get an X visual which matches the given arguments.
+ */
+static XVisualInfo *
+get_visual( Display *dpy, int scr, unsigned int depth, int xclass )
+{
+   XVisualInfo temp, *vis;
+   long mask;
+   int n;
+   unsigned int default_depth;
+   int default_class;
+
+   mask = VisualScreenMask | VisualDepthMask | VisualClassMask;
+   temp.screen = scr;
+   temp.depth = depth;
+   temp.CLASS = xclass;
+
+   default_depth = DefaultDepth(dpy,scr);
+   default_class = DefaultVisual(dpy,scr)->CLASS;
+
+   if (depth==default_depth && xclass==default_class) {
+      /* try to get root window's visual */
+      temp.visualid = DefaultVisual(dpy,scr)->visualid;
+      mask |= VisualIDMask;
+   }
+
+   vis = XGetVisualInfo( dpy, mask, &temp, &n );
+
+   /* In case bits/pixel > 24, make sure color channels are still <=8 bits.
+    * An SGI Infinite Reality system, for example, can have 30bpp pixels:
+    * 10 bits per color channel.  Mesa's limited to a max of 8 bits/channel.
+    */
+   if (vis && depth > 24 && (xclass==TrueColor || xclass==DirectColor)) {
+      if (_mesa_bitcount((GLuint) vis->red_mask  ) <= 8 &&
+          _mesa_bitcount((GLuint) vis->green_mask) <= 8 &&
+          _mesa_bitcount((GLuint) vis->blue_mask ) <= 8) {
+         return vis;
+      }
+      else {
+         XFree((void *) vis);
+         return NULL;
+      }
+   }
+
+   return vis;
+}
+
+
+
+/*
+ * Retrieve the value of the given environment variable and find
+ * the X visual which matches it.
+ * Input:  dpy - the display
+ *         screen - the screen number
+ *         varname - the name of the environment variable
+ * Return:  an XVisualInfo pointer to NULL if error.
+ */
+static XVisualInfo *
+get_env_visual(Display *dpy, int scr, const char *varname)
+{
+   char value[100], type[100];
+   int depth, xclass = -1;
+   XVisualInfo *vis;
+
+   if (!_mesa_getenv( varname )) {
+      return NULL;
+   }
+
+   _mesa_strncpy( value, _mesa_getenv(varname), 100 );
+   value[99] = 0;
+
+   sscanf( value, "%s %d", type, &depth );
+
+   if (_mesa_strcmp(type,"TrueColor")==0)          xclass = TrueColor;
+   else if (_mesa_strcmp(type,"DirectColor")==0)   xclass = DirectColor;
+   else if (_mesa_strcmp(type,"PseudoColor")==0)   xclass = PseudoColor;
+   else if (_mesa_strcmp(type,"StaticColor")==0)   xclass = StaticColor;
+   else if (_mesa_strcmp(type,"GrayScale")==0)     xclass = GrayScale;
+   else if (_mesa_strcmp(type,"StaticGray")==0)    xclass = StaticGray;
+
+   if (xclass>-1 && depth>0) {
+      vis = get_visual( dpy, scr, depth, xclass );
+      if (vis) {
+	 return vis;
+      }
+   }
+
+   _mesa_warning(NULL, "GLX unable to find visual class=%s, depth=%d.",
+                 type, depth);
+
+   return NULL;
+}
+
+
+
+/*
+ * Select an X visual which satisfies the RGBA/CI flag and minimum depth.
+ * Input:  dpy, screen - X display and screen number
+ *         rgba - GL_TRUE = RGBA mode, GL_FALSE = CI mode
+ *         min_depth - minimum visual depth
+ *         preferred_class - preferred GLX visual class or DONT_CARE
+ * Return:  pointer to an XVisualInfo or NULL.
+ */
+static XVisualInfo *
+choose_x_visual( Display *dpy, int screen, GLboolean rgba, int min_depth,
+                 int preferred_class )
+{
+   XVisualInfo *vis;
+   int xclass, visclass = 0;
+   int depth;
+
+   if (rgba) {
+      Atom hp_cr_maps = XInternAtom(dpy, "_HP_RGB_SMOOTH_MAP_LIST", True);
+      /* First see if the MESA_RGB_VISUAL env var is defined */
+      vis = get_env_visual( dpy, screen, "MESA_RGB_VISUAL" );
+      if (vis) {
+	 return vis;
+      }
+      /* Otherwise, search for a suitable visual */
+      if (preferred_class==DONT_CARE) {
+         for (xclass=0;xclass<6;xclass++) {
+            switch (xclass) {
+               case 0:  visclass = TrueColor;    break;
+               case 1:  visclass = DirectColor;  break;
+               case 2:  visclass = PseudoColor;  break;
+               case 3:  visclass = StaticColor;  break;
+               case 4:  visclass = GrayScale;    break;
+               case 5:  visclass = StaticGray;   break;
+            }
+            if (min_depth==0) {
+               /* start with shallowest */
+               for (depth=0;depth<=32;depth++) {
+                  if (visclass==TrueColor && depth==8 && !hp_cr_maps) {
+                     /* Special case:  try to get 8-bit PseudoColor before */
+                     /* 8-bit TrueColor */
+                     vis = get_visual( dpy, screen, 8, PseudoColor );
+                     if (vis) {
+                        return vis;
+                     }
+                  }
+                  vis = get_visual( dpy, screen, depth, visclass );
+                  if (vis) {
+                     return vis;
+                  }
+               }
+            }
+            else {
+               /* start with deepest */
+               for (depth=32;depth>=min_depth;depth--) {
+                  if (visclass==TrueColor && depth==8 && !hp_cr_maps) {
+                     /* Special case:  try to get 8-bit PseudoColor before */
+                     /* 8-bit TrueColor */
+                     vis = get_visual( dpy, screen, 8, PseudoColor );
+                     if (vis) {
+                        return vis;
+                     }
+                  }
+                  vis = get_visual( dpy, screen, depth, visclass );
+                  if (vis) {
+                     return vis;
+                  }
+               }
+            }
+         }
+      }
+      else {
+         /* search for a specific visual class */
+         switch (preferred_class) {
+            case GLX_TRUE_COLOR_EXT:    visclass = TrueColor;    break;
+            case GLX_DIRECT_COLOR_EXT:  visclass = DirectColor;  break;
+            case GLX_PSEUDO_COLOR_EXT:  visclass = PseudoColor;  break;
+            case GLX_STATIC_COLOR_EXT:  visclass = StaticColor;  break;
+            case GLX_GRAY_SCALE_EXT:    visclass = GrayScale;    break;
+            case GLX_STATIC_GRAY_EXT:   visclass = StaticGray;   break;
+            default:   return NULL;
+         }
+         if (min_depth==0) {
+            /* start with shallowest */
+            for (depth=0;depth<=32;depth++) {
+               vis = get_visual( dpy, screen, depth, visclass );
+               if (vis) {
+                  return vis;
+               }
+            }
+         }
+         else {
+            /* start with deepest */
+            for (depth=32;depth>=min_depth;depth--) {
+               vis = get_visual( dpy, screen, depth, visclass );
+               if (vis) {
+                  return vis;
+               }
+            }
+         }
+      }
+   }
+   else {
+      /* First see if the MESA_CI_VISUAL env var is defined */
+      vis = get_env_visual( dpy, screen, "MESA_CI_VISUAL" );
+      if (vis) {
+	 return vis;
+      }
+      /* Otherwise, search for a suitable visual, starting with shallowest */
+      if (preferred_class==DONT_CARE) {
+         for (xclass=0;xclass<4;xclass++) {
+            switch (xclass) {
+               case 0:  visclass = PseudoColor;  break;
+               case 1:  visclass = StaticColor;  break;
+               case 2:  visclass = GrayScale;    break;
+               case 3:  visclass = StaticGray;   break;
+            }
+            /* try 8-bit up through 16-bit */
+            for (depth=8;depth<=16;depth++) {
+               vis = get_visual( dpy, screen, depth, visclass );
+               if (vis) {
+                  return vis;
+               }
+            }
+            /* try min_depth up to 8-bit */
+            for (depth=min_depth;depth<8;depth++) {
+               vis = get_visual( dpy, screen, depth, visclass );
+               if (vis) {
+                  return vis;
+               }
+            }
+         }
+      }
+      else {
+         /* search for a specific visual class */
+         switch (preferred_class) {
+            case GLX_TRUE_COLOR_EXT:    visclass = TrueColor;    break;
+            case GLX_DIRECT_COLOR_EXT:  visclass = DirectColor;  break;
+            case GLX_PSEUDO_COLOR_EXT:  visclass = PseudoColor;  break;
+            case GLX_STATIC_COLOR_EXT:  visclass = StaticColor;  break;
+            case GLX_GRAY_SCALE_EXT:    visclass = GrayScale;    break;
+            case GLX_STATIC_GRAY_EXT:   visclass = StaticGray;   break;
+            default:   return NULL;
+         }
+         /* try 8-bit up through 16-bit */
+         for (depth=8;depth<=16;depth++) {
+            vis = get_visual( dpy, screen, depth, visclass );
+            if (vis) {
+               return vis;
+            }
+         }
+         /* try min_depth up to 8-bit */
+         for (depth=min_depth;depth<8;depth++) {
+            vis = get_visual( dpy, screen, depth, visclass );
+            if (vis) {
+               return vis;
+            }
+         }
+      }
+   }
+
+   /* didn't find a visual */
+   return NULL;
+}
+
+
+
+/*
+ * Find the deepest X over/underlay visual of at least min_depth.
+ * Input:  dpy, screen - X display and screen number
+ *         level - the over/underlay level
+ *         trans_type - transparent pixel type: GLX_NONE_EXT,
+ *                      GLX_TRANSPARENT_RGB_EXT, GLX_TRANSPARENT_INDEX_EXT,
+ *                      or DONT_CARE
+ *         trans_value - transparent pixel value or DONT_CARE
+ *         min_depth - minimum visual depth
+ *         preferred_class - preferred GLX visual class or DONT_CARE
+ * Return:  pointer to an XVisualInfo or NULL.
+ */
+static XVisualInfo *
+choose_x_overlay_visual( Display *dpy, int scr, GLboolean rgbFlag,
+                         int level, int trans_type, int trans_value,
+                         int min_depth, int preferred_class )
+{
+   OverlayInfo *overlay_info;
+   int numOverlaysPerScreen;
+   int i;
+   XVisualInfo *deepvis;
+   int deepest;
+
+   /*DEBUG int tt, tv; */
+
+   switch (preferred_class) {
+      case GLX_TRUE_COLOR_EXT:    preferred_class = TrueColor;    break;
+      case GLX_DIRECT_COLOR_EXT:  preferred_class = DirectColor;  break;
+      case GLX_PSEUDO_COLOR_EXT:  preferred_class = PseudoColor;  break;
+      case GLX_STATIC_COLOR_EXT:  preferred_class = StaticColor;  break;
+      case GLX_GRAY_SCALE_EXT:    preferred_class = GrayScale;    break;
+      case GLX_STATIC_GRAY_EXT:   preferred_class = StaticGray;   break;
+      default:                    preferred_class = DONT_CARE;
+   }
+
+   overlay_info = GetOverlayInfo(dpy, scr, &numOverlaysPerScreen);
+   if (!overlay_info) {
+      return NULL;
+   }
+
+   /* Search for the deepest overlay which satisifies all criteria. */
+   deepest = min_depth;
+   deepvis = NULL;
+
+   for (i = 0; i < numOverlaysPerScreen; i++) {
+      const OverlayInfo *ov = overlay_info + i;
+      XVisualInfo *vislist, vistemplate;
+      int count;
+
+      if (ov->layer!=level) {
+         /* failed overlay level criteria */
+         continue;
+      }
+      if (!(trans_type==DONT_CARE
+            || (trans_type==GLX_TRANSPARENT_INDEX_EXT
+                && ov->transparent_type>0)
+            || (trans_type==GLX_NONE_EXT && ov->transparent_type==0))) {
+         /* failed transparent pixel type criteria */
+         continue;
+      }
+      if (trans_value!=DONT_CARE && trans_value!=ov->value) {
+         /* failed transparent pixel value criteria */
+         continue;
+      }
+
+      /* get XVisualInfo and check the depth */
+      vistemplate.visualid = ov->overlay_visual;
+      vistemplate.screen = scr;
+      vislist = XGetVisualInfo( dpy, VisualIDMask | VisualScreenMask,
+                                &vistemplate, &count );
+
+      if (count!=1) {
+         /* something went wrong */
+         continue;
+      }
+      if (preferred_class!=DONT_CARE && preferred_class!=vislist->CLASS) {
+         /* wrong visual class */
+         continue;
+      }
+
+      /* if RGB was requested, make sure we have True/DirectColor */
+      if (rgbFlag && vislist->CLASS != TrueColor
+          && vislist->CLASS != DirectColor)
+         continue;
+
+      /* if CI was requested, make sure we have a color indexed visual */
+      if (!rgbFlag
+          && (vislist->CLASS == TrueColor || vislist->CLASS == DirectColor))
+         continue;
+
+      if (deepvis==NULL || vislist->depth > deepest) {
+         /* YES!  found a satisfactory visual */
+         if (deepvis) {
+            XFree( deepvis );
+         }
+         deepest = vislist->depth;
+         deepvis = vislist;
+         /* DEBUG  tt = ov->transparent_type;*/
+         /* DEBUG  tv = ov->value; */
+      }
+   }
+
+/*DEBUG
+   if (deepvis) {
+      printf("chose 0x%x:  layer=%d depth=%d trans_type=%d trans_value=%d\n",
+             deepvis->visualid, level, deepvis->depth, tt, tv );
+   }
+*/
+   return deepvis;
+}
+
+
+/**********************************************************************/
+/***             Display-related functions                          ***/
+/**********************************************************************/
+
+
+/**
+ * Free all XMesaVisuals which are associated with the given display.
+ */
+static void
+destroy_visuals_on_display(Display *dpy)
+{
+   int i;
+   for (i = 0; i < NumVisuals; i++) {
+      if (VisualTable[i]->display == dpy) {
+         /* remove this visual */
+         int j;
+         free(VisualTable[i]);
+         for (j = i; j < NumVisuals - 1; j++)
+            VisualTable[j] = VisualTable[j + 1];
+         NumVisuals--;
+      }
+   }
+}
+
+
+/**
+ * Called from XCloseDisplay() to let us free our display-related data.
+ */
+static int
+close_display_callback(Display *dpy, XExtCodes *codes)
+{
+   destroy_visuals_on_display(dpy);
+   xmesa_destroy_buffers_on_display(dpy);
+   return 0;
+}
+
+
+/**
+ * Look for the named extension on given display and return a pointer
+ * to the _XExtension data, or NULL if extension not found.
+ */
+static _XExtension *
+lookup_extension(Display *dpy, const char *extName)
+{
+   _XExtension *ext;
+   for (ext = dpy->ext_procs; ext; ext = ext->next) {
+      if (ext->name && strcmp(ext->name, extName) == 0) {
+         return ext;
+      }
+   }
+   return NULL;
+}
+
+
+/**
+ * Whenever we're given a new Display pointer, call this function to
+ * register our close_display_callback function.
+ */
+static void
+register_with_display(Display *dpy)
+{
+   const char *extName = "MesaGLX";
+   _XExtension *ext;
+
+   ext = lookup_extension(dpy, extName);
+   if (!ext) {
+      XExtCodes *c = XAddExtension(dpy);
+      ext = dpy->ext_procs;  /* new extension is at head of list */
+      assert(c->extension == ext->codes.extension);
+      ext->name = _mesa_strdup(extName);
+      ext->close_display = close_display_callback;
+   }
+}
+
+
+/**********************************************************************/
+/***                  Begin Fake GLX API Functions                  ***/
+/**********************************************************************/
+
+
+/**
+ * Helper used by glXChooseVisual and glXChooseFBConfig.
+ * The fbConfig parameter must be GL_FALSE for the former and GL_TRUE for
+ * the later.
+ * In either case, the attribute list is terminated with the value 'None'.
+ */
+static XMesaVisual
+choose_visual( Display *dpy, int screen, const int *list, GLboolean fbConfig )
+{
+   const GLboolean rgbModeDefault = fbConfig;
+   const int *parselist;
+   XVisualInfo *vis;
+   int min_ci = 0;
+   int min_red=0, min_green=0, min_blue=0;
+   GLboolean rgb_flag = rgbModeDefault;
+   GLboolean alpha_flag = GL_FALSE;
+   GLboolean double_flag = GL_FALSE;
+   GLboolean stereo_flag = GL_FALSE;
+   GLint depth_size = 0;
+   GLint stencil_size = 0;
+   GLint accumRedSize = 0;
+   GLint accumGreenSize = 0;
+   GLint accumBlueSize = 0;
+   GLint accumAlphaSize = 0;
+   int level = 0;
+   int visual_type = DONT_CARE;
+   int trans_type = DONT_CARE;
+   int trans_value = DONT_CARE;
+   GLint caveat = DONT_CARE;
+   XMesaVisual xmvis = NULL;
+   int desiredVisualID = -1;
+   int numAux = 0;
+
+   parselist = list;
+
+   while (*parselist) {
+
+      switch (*parselist) {
+	 case GLX_USE_GL:
+            if (fbConfig) {
+               /* invalid token */
+               return NULL;
+            }
+            else {
+               /* skip */
+               parselist++;
+            }
+	    break;
+	 case GLX_BUFFER_SIZE:
+	    parselist++;
+	    min_ci = *parselist++;
+	    break;
+	 case GLX_LEVEL:
+	    parselist++;
+            level = *parselist++;
+	    break;
+	 case GLX_RGBA:
+            if (fbConfig) {
+               /* invalid token */
+               return NULL;
+            }
+            else {
+               rgb_flag = GL_TRUE;
+               parselist++;
+            }
+	    break;
+	 case GLX_DOUBLEBUFFER:
+            parselist++;
+            if (fbConfig) {
+               double_flag = *parselist++;
+            }
+            else {
+               double_flag = GL_TRUE;
+            }
+	    break;
+	 case GLX_STEREO:
+            parselist++;
+            if (fbConfig) {
+               stereo_flag = *parselist++;
+            }
+            else {
+               stereo_flag = GL_TRUE;
+            }
+            break;
+	 case GLX_AUX_BUFFERS:
+	    parselist++;
+            numAux = *parselist++;
+            if (numAux > MAX_AUX_BUFFERS)
+               return NULL;
+	    break;
+	 case GLX_RED_SIZE:
+	    parselist++;
+	    min_red = *parselist++;
+	    break;
+	 case GLX_GREEN_SIZE:
+	    parselist++;
+	    min_green = *parselist++;
+	    break;
+	 case GLX_BLUE_SIZE:
+	    parselist++;
+	    min_blue = *parselist++;
+	    break;
+	 case GLX_ALPHA_SIZE:
+	    parselist++;
+            {
+               GLint size = *parselist++;
+               alpha_flag = size ? GL_TRUE : GL_FALSE;
+            }
+	    break;
+	 case GLX_DEPTH_SIZE:
+	    parselist++;
+	    depth_size = *parselist++;
+	    break;
+	 case GLX_STENCIL_SIZE:
+	    parselist++;
+	    stencil_size = *parselist++;
+	    break;
+	 case GLX_ACCUM_RED_SIZE:
+	    parselist++;
+            {
+               GLint size = *parselist++;
+               accumRedSize = MAX2( accumRedSize, size );
+            }
+            break;
+	 case GLX_ACCUM_GREEN_SIZE:
+	    parselist++;
+            {
+               GLint size = *parselist++;
+               accumGreenSize = MAX2( accumGreenSize, size );
+            }
+            break;
+	 case GLX_ACCUM_BLUE_SIZE:
+	    parselist++;
+            {
+               GLint size = *parselist++;
+               accumBlueSize = MAX2( accumBlueSize, size );
+            }
+            break;
+	 case GLX_ACCUM_ALPHA_SIZE:
+	    parselist++;
+            {
+               GLint size = *parselist++;
+               accumAlphaSize = MAX2( accumAlphaSize, size );
+            }
+	    break;
+
+         /*
+          * GLX_EXT_visual_info extension
+          */
+         case GLX_X_VISUAL_TYPE_EXT:
+            parselist++;
+            visual_type = *parselist++;
+            break;
+         case GLX_TRANSPARENT_TYPE_EXT:
+            parselist++;
+            trans_type = *parselist++;
+            break;
+         case GLX_TRANSPARENT_INDEX_VALUE_EXT:
+            parselist++;
+            trans_value = *parselist++;
+            break;
+         case GLX_TRANSPARENT_RED_VALUE_EXT:
+         case GLX_TRANSPARENT_GREEN_VALUE_EXT:
+         case GLX_TRANSPARENT_BLUE_VALUE_EXT:
+         case GLX_TRANSPARENT_ALPHA_VALUE_EXT:
+	    /* ignore */
+	    parselist++;
+	    parselist++;
+	    break;
+
+         /*
+          * GLX_EXT_visual_info extension
+          */
+         case GLX_VISUAL_CAVEAT_EXT:
+            parselist++;
+            caveat = *parselist++; /* ignored for now */
+            break;
+
+         /*
+          * GLX_ARB_multisample
+          */
+         case GLX_SAMPLE_BUFFERS_ARB:
+            /* ms not supported */
+            return NULL;
+         case GLX_SAMPLES_ARB:
+            /* ms not supported */
+            return NULL;
+
+         /*
+          * FBConfig attribs.
+          */
+         case GLX_RENDER_TYPE:
+            if (!fbConfig)
+               return NULL;
+            parselist++;
+            if (*parselist == GLX_RGBA_BIT) {
+               rgb_flag = GL_TRUE;
+            }
+            else if (*parselist == GLX_COLOR_INDEX_BIT) {
+               rgb_flag = GL_FALSE;
+            }
+            else if (*parselist == 0) {
+               rgb_flag = GL_TRUE;
+            }
+            parselist++;
+            break;
+         case GLX_DRAWABLE_TYPE:
+            if (!fbConfig)
+               return NULL;
+            parselist++;
+            if (*parselist & ~(GLX_WINDOW_BIT | GLX_PIXMAP_BIT | GLX_PBUFFER_BIT)) {
+               return NULL; /* bad bit */
+            }
+            parselist++;
+            break;
+         case GLX_FBCONFIG_ID:
+            if (!fbConfig)
+               return NULL;
+            parselist++;
+            desiredVisualID = *parselist++;
+            break;
+         case GLX_X_RENDERABLE:
+            if (!fbConfig)
+               return NULL;
+            parselist += 2;
+            /* ignore */
+            break;
+
+#ifdef GLX_EXT_texture_from_pixmap
+         case GLX_BIND_TO_TEXTURE_RGB_EXT:
+            parselist++; /*skip*/
+            break;
+         case GLX_BIND_TO_TEXTURE_RGBA_EXT:
+            parselist++; /*skip*/
+            break;
+         case GLX_BIND_TO_MIPMAP_TEXTURE_EXT:
+            parselist++; /*skip*/
+            break;
+         case GLX_BIND_TO_TEXTURE_TARGETS_EXT:
+            parselist++;
+            if (*parselist & ~(GLX_TEXTURE_1D_BIT_EXT |
+                               GLX_TEXTURE_2D_BIT_EXT |
+                               GLX_TEXTURE_RECTANGLE_BIT_EXT)) {
+               /* invalid bit */
+               return NULL;
+            }
+            break;
+         case GLX_Y_INVERTED_EXT:
+            parselist++; /*skip*/
+            break;
+#endif
+
+	 case None:
+            /* end of list */
+	    break;
+
+	 default:
+	    /* undefined attribute */
+            _mesa_warning(NULL, "unexpected attrib 0x%x in choose_visual()",
+                          *parselist);
+	    return NULL;
+      }
+   }
+
+   (void) caveat;
+
+   /*
+    * Since we're only simulating the GLX extension this function will never
+    * find any real GL visuals.  Instead, all we can do is try to find an RGB
+    * or CI visual of appropriate depth.  Other requested attributes such as
+    * double buffering, depth buffer, etc. will be associated with the X
+    * visual and stored in the VisualTable[].
+    */
+   if (desiredVisualID != -1) {
+      /* try to get a specific visual, by visualID */
+      XVisualInfo temp;
+      int n;
+      temp.visualid = desiredVisualID;
+      temp.screen = screen;
+      vis = XGetVisualInfo(dpy, VisualIDMask | VisualScreenMask, &temp, &n);
+      if (vis) {
+         /* give the visual some useful GLX attributes */
+         double_flag = GL_TRUE;
+         if (vis->depth > 8)
+            rgb_flag = GL_TRUE;
+         depth_size = 24; /*default_depth_bits();*/
+         stencil_size = STENCIL_BITS;
+         /* XXX accum??? */
+      }
+   }
+   else if (level==0) {
+      /* normal color planes */
+      if (rgb_flag) {
+         /* Get an RGB visual */
+         int min_rgb = min_red + min_green + min_blue;
+         if (min_rgb>1 && min_rgb<8) {
+            /* a special case to be sure we can get a monochrome visual */
+            min_rgb = 1;
+         }
+         vis = choose_x_visual( dpy, screen, rgb_flag, min_rgb, visual_type );
+      }
+      else {
+         /* Get a color index visual */
+         vis = choose_x_visual( dpy, screen, rgb_flag, min_ci, visual_type );
+         accumRedSize = accumGreenSize = accumBlueSize = accumAlphaSize = 0;
+      }
+   }
+   else {
+      /* over/underlay planes */
+      if (rgb_flag) {
+         /* rgba overlay */
+         int min_rgb = min_red + min_green + min_blue;
+         if (min_rgb>1 && min_rgb<8) {
+            /* a special case to be sure we can get a monochrome visual */
+            min_rgb = 1;
+         }
+         vis = choose_x_overlay_visual( dpy, screen, rgb_flag, level,
+                              trans_type, trans_value, min_rgb, visual_type );
+      }
+      else {
+         /* color index overlay */
+         vis = choose_x_overlay_visual( dpy, screen, rgb_flag, level,
+                              trans_type, trans_value, min_ci, visual_type );
+      }
+   }
+
+   if (vis) {
+      /* Note: we're not exactly obeying the glXChooseVisual rules here.
+       * When GLX_DEPTH_SIZE = 1 is specified we're supposed to choose the
+       * largest depth buffer size, which is 32bits/value.  Instead, we
+       * return 16 to maintain performance with earlier versions of Mesa.
+       */
+      if (stencil_size > 0)
+         depth_size = 24;  /* if Z and stencil, always use 24+8 format */
+      else if (depth_size > 24)
+         depth_size = 32;
+      else if (depth_size > 16)
+         depth_size = 24;
+      else if (depth_size > 0) {
+         depth_size = default_depth_bits();
+      }
+
+      if (!alpha_flag) {
+         alpha_flag = default_alpha_bits() > 0;
+      }
+
+      /* we only support one size of stencil and accum buffers. */
+      if (stencil_size > 0)
+         stencil_size = STENCIL_BITS;
+      if (accumRedSize > 0 || accumGreenSize > 0 || accumBlueSize > 0 ||
+          accumAlphaSize > 0) {
+         accumRedSize = 
+         accumGreenSize = 
+         accumBlueSize = default_accum_bits();
+         accumAlphaSize = alpha_flag ? accumRedSize : 0;
+      }
+
+      xmvis = save_glx_visual( dpy, vis, rgb_flag, alpha_flag, double_flag,
+                               stereo_flag, depth_size, stencil_size,
+                               accumRedSize, accumGreenSize,
+                               accumBlueSize, accumAlphaSize, level, numAux );
+   }
+
+   return xmvis;
+}
+
+
+static XVisualInfo *
+Fake_glXChooseVisual( Display *dpy, int screen, int *list )
+{
+   XMesaVisual xmvis;
+
+   /* register ourselves as an extension on this display */
+   register_with_display(dpy);
+
+   xmvis = choose_visual(dpy, screen, list, GL_FALSE);
+   if (xmvis) {
+#if 0
+      return xmvis->vishandle;
+#else
+      /* create a new vishandle - the cached one may be stale */
+      xmvis->vishandle = (XVisualInfo *) _mesa_malloc(sizeof(XVisualInfo));
+      if (xmvis->vishandle) {
+         _mesa_memcpy(xmvis->vishandle, xmvis->visinfo, sizeof(XVisualInfo));
+      }
+      return xmvis->vishandle;
+#endif
+   }
+   else
+      return NULL;
+}
+
+
+static GLXContext
+Fake_glXCreateContext( Display *dpy, XVisualInfo *visinfo,
+                       GLXContext share_list, Bool direct )
+{
+   XMesaVisual xmvis;
+   struct fake_glx_context *glxCtx;
+   struct fake_glx_context *shareCtx = (struct fake_glx_context *) share_list;
+
+   if (!dpy || !visinfo)
+      return 0;
+
+   glxCtx = CALLOC_STRUCT(fake_glx_context);
+   if (!glxCtx)
+      return 0;
+
+   /* deallocate unused windows/buffers */
+#if 0
+   XMesaGarbageCollect();
+#endif
+
+   xmvis = find_glx_visual( dpy, visinfo );
+   if (!xmvis) {
+      /* This visual wasn't found with glXChooseVisual() */
+      xmvis = create_glx_visual( dpy, visinfo );
+      if (!xmvis) {
+         /* unusable visual */
+         _mesa_free(glxCtx);
+         return NULL;
+      }
+   }
+
+   glxCtx->xmesaContext = XMesaCreateContext(xmvis,
+                                   shareCtx ? shareCtx->xmesaContext : NULL);
+   if (!glxCtx->xmesaContext) {
+      _mesa_free(glxCtx);
+      return NULL;
+   }
+
+   glxCtx->glxContext.isDirect = GL_FALSE;
+   glxCtx->glxContext.currentDpy = dpy;
+   glxCtx->glxContext.xid = (XID) glxCtx;  /* self pointer */
+
+   assert((void *) glxCtx == (void *) &(glxCtx->glxContext));
+
+   return (GLXContext) glxCtx;
+}
+
+
+/* XXX these may have to be removed due to thread-safety issues. */
+static GLXContext MakeCurrent_PrevContext = 0;
+static GLXDrawable MakeCurrent_PrevDrawable = 0;
+static GLXDrawable MakeCurrent_PrevReadable = 0;
+static XMesaBuffer MakeCurrent_PrevDrawBuffer = 0;
+static XMesaBuffer MakeCurrent_PrevReadBuffer = 0;
+
+
+/* GLX 1.3 and later */
+static Bool
+Fake_glXMakeContextCurrent( Display *dpy, GLXDrawable draw,
+                            GLXDrawable read, GLXContext ctx )
+{
+   struct fake_glx_context *glxCtx = (struct fake_glx_context *) ctx;
+
+   if (ctx && draw && read) {
+      XMesaBuffer drawBuffer, readBuffer;
+      XMesaContext xmctx = glxCtx->xmesaContext;
+
+      /* Find the XMesaBuffer which corresponds to the GLXDrawable 'draw' */
+      if (ctx == MakeCurrent_PrevContext
+          && draw == MakeCurrent_PrevDrawable) {
+         drawBuffer = MakeCurrent_PrevDrawBuffer;
+      }
+      else {
+         drawBuffer = XMesaFindBuffer( dpy, draw );
+      }
+      if (!drawBuffer) {
+         /* drawable must be a new window! */
+         drawBuffer = XMesaCreateWindowBuffer( xmctx->xm_visual, draw );
+         if (!drawBuffer) {
+            /* Out of memory, or context/drawable depth mismatch */
+            return False;
+         }
+#ifdef FX
+         FXcreateContext( xmctx->xm_visual, draw, xmctx, drawBuffer );
+#endif
+      }
+
+      /* Find the XMesaBuffer which corresponds to the GLXDrawable 'read' */
+      if (ctx == MakeCurrent_PrevContext
+          && read == MakeCurrent_PrevReadable) {
+         readBuffer = MakeCurrent_PrevReadBuffer;
+      }
+      else {
+         readBuffer = XMesaFindBuffer( dpy, read );
+      }
+      if (!readBuffer) {
+         /* drawable must be a new window! */
+         readBuffer = XMesaCreateWindowBuffer( xmctx->xm_visual, read );
+         if (!readBuffer) {
+            /* Out of memory, or context/drawable depth mismatch */
+            return False;
+         }
+#ifdef FX
+         FXcreateContext( xmctx->xm_visual, read, xmctx, readBuffer );
+#endif
+      }
+
+      MakeCurrent_PrevContext = ctx;
+      MakeCurrent_PrevDrawable = draw;
+      MakeCurrent_PrevReadable = read;
+      MakeCurrent_PrevDrawBuffer = drawBuffer;
+      MakeCurrent_PrevReadBuffer = readBuffer;
+
+      /* Now make current! */
+      if (XMesaMakeCurrent2(xmctx, drawBuffer, readBuffer)) {
+         ((__GLXcontext *) ctx)->currentDpy = dpy;
+         ((__GLXcontext *) ctx)->currentDrawable = draw;
+         ((__GLXcontext *) ctx)->currentReadable = read;
+         return True;
+      }
+      else {
+         return False;
+      }
+   }
+   else if (!ctx && !draw && !read) {
+      /* release current context w/out assigning new one. */
+      XMesaMakeCurrent( NULL, NULL );
+      MakeCurrent_PrevContext = 0;
+      MakeCurrent_PrevDrawable = 0;
+      MakeCurrent_PrevReadable = 0;
+      MakeCurrent_PrevDrawBuffer = 0;
+      MakeCurrent_PrevReadBuffer = 0;
+      return True;
+   }
+   else {
+      /* The args must either all be non-zero or all zero.
+       * This is an error.
+       */
+      return False;
+   }
+}
+
+
+static Bool
+Fake_glXMakeCurrent( Display *dpy, GLXDrawable drawable, GLXContext ctx )
+{
+   return Fake_glXMakeContextCurrent( dpy, drawable, drawable, ctx );
+}
+
+
+static GLXPixmap
+Fake_glXCreateGLXPixmap( Display *dpy, XVisualInfo *visinfo, Pixmap pixmap )
+{
+   XMesaVisual v;
+   XMesaBuffer b;
+
+   v = find_glx_visual( dpy, visinfo );
+   if (!v) {
+      v = create_glx_visual( dpy, visinfo );
+      if (!v) {
+         /* unusable visual */
+         return 0;
+      }
+   }
+
+   b = XMesaCreatePixmapBuffer( v, pixmap, 0 );
+   if (!b) {
+      return 0;
+   }
+   return b->drawable;
+}
+
+
+/*** GLX_MESA_pixmap_colormap ***/
+
+static GLXPixmap
+Fake_glXCreateGLXPixmapMESA( Display *dpy, XVisualInfo *visinfo,
+                             Pixmap pixmap, Colormap cmap )
+{
+   XMesaVisual v;
+   XMesaBuffer b;
+
+   v = find_glx_visual( dpy, visinfo );
+   if (!v) {
+      v = create_glx_visual( dpy, visinfo );
+      if (!v) {
+         /* unusable visual */
+         return 0;
+      }
+   }
+
+   b = XMesaCreatePixmapBuffer( v, pixmap, cmap );
+   if (!b) {
+      return 0;
+   }
+   return b->drawable;
+}
+
+
+static void
+Fake_glXDestroyGLXPixmap( Display *dpy, GLXPixmap pixmap )
+{
+   XMesaBuffer b = XMesaFindBuffer(dpy, pixmap);
+   if (b) {
+      XMesaDestroyBuffer(b);
+   }
+   else if (_mesa_getenv("MESA_DEBUG")) {
+      _mesa_warning(NULL, "Mesa: glXDestroyGLXPixmap: invalid pixmap\n");
+   }
+}
+
+
+static void
+Fake_glXCopyContext( Display *dpy, GLXContext src, GLXContext dst,
+                     unsigned long mask )
+{
+   struct fake_glx_context *fakeSrc = (struct fake_glx_context *) src;
+   struct fake_glx_context *fakeDst = (struct fake_glx_context *) dst;
+   XMesaContext xm_src = fakeSrc->xmesaContext;
+   XMesaContext xm_dst = fakeDst->xmesaContext;
+   (void) dpy;
+   if (MakeCurrent_PrevContext == src) {
+      _mesa_Flush();
+   }
+   st_copy_context_state( xm_src->st, xm_dst->st, (GLuint) mask );
+}
+
+
+static Bool
+Fake_glXQueryExtension( Display *dpy, int *errorb, int *event )
+{
+   /* Mesa's GLX isn't really an X extension but we try to act like one. */
+   (void) dpy;
+   (void) errorb;
+   (void) event;
+   return True;
+}
+
+
+extern void _kw_ungrab_all( Display *dpy );
+void _kw_ungrab_all( Display *dpy )
+{
+   XUngrabPointer( dpy, CurrentTime );
+   XUngrabKeyboard( dpy, CurrentTime );
+}
+
+
+static void
+Fake_glXDestroyContext( Display *dpy, GLXContext ctx )
+{
+   struct fake_glx_context *glxCtx = (struct fake_glx_context *) ctx;
+   (void) dpy;
+   MakeCurrent_PrevContext = 0;
+   MakeCurrent_PrevDrawable = 0;
+   MakeCurrent_PrevReadable = 0;
+   MakeCurrent_PrevDrawBuffer = 0;
+   MakeCurrent_PrevReadBuffer = 0;
+   XMesaDestroyContext( glxCtx->xmesaContext );
+   XMesaGarbageCollect();
+   _mesa_free(glxCtx);
+}
+
+
+static Bool
+Fake_glXIsDirect( Display *dpy, GLXContext ctx )
+{
+   (void) dpy;
+   (void) ctx;
+   return False;
+}
+
+
+
+static void
+Fake_glXSwapBuffers( Display *dpy, GLXDrawable drawable )
+{
+   XMesaBuffer buffer = XMesaFindBuffer( dpy, drawable );
+
+   if (buffer) {
+      XMesaSwapBuffers(buffer);
+   }
+   else if (_mesa_getenv("MESA_DEBUG")) {
+      _mesa_warning(NULL, "glXSwapBuffers: invalid drawable 0x%x\n",
+                    (int) drawable);
+   }
+}
+
+
+
+/*** GLX_MESA_copy_sub_buffer ***/
+
+static void
+Fake_glXCopySubBufferMESA( Display *dpy, GLXDrawable drawable,
+                           int x, int y, int width, int height )
+{
+   XMesaBuffer buffer = XMesaFindBuffer( dpy, drawable );
+   if (buffer) {
+      XMesaCopySubBuffer(buffer, x, y, width, height);
+   }
+   else if (_mesa_getenv("MESA_DEBUG")) {
+      _mesa_warning(NULL, "Mesa: glXCopySubBufferMESA: invalid drawable\n");
+   }
+}
+
+
+static Bool
+Fake_glXQueryVersion( Display *dpy, int *maj, int *min )
+{
+   (void) dpy;
+   /* Return GLX version, not Mesa version */
+   assert(CLIENT_MAJOR_VERSION == SERVER_MAJOR_VERSION);
+   *maj = CLIENT_MAJOR_VERSION;
+   *min = MIN2( CLIENT_MINOR_VERSION, SERVER_MINOR_VERSION );
+   return True;
+}
+
+
+/*
+ * Query the GLX attributes of the given XVisualInfo.
+ */
+static int
+get_config( XMesaVisual xmvis, int attrib, int *value, GLboolean fbconfig )
+{
+   ASSERT(xmvis);
+   switch(attrib) {
+      case GLX_USE_GL:
+         if (fbconfig)
+            return GLX_BAD_ATTRIBUTE;
+         *value = (int) True;
+	 return 0;
+      case GLX_BUFFER_SIZE:
+	 *value = xmvis->visinfo->depth;
+	 return 0;
+      case GLX_LEVEL:
+	 *value = xmvis->mesa_visual.level;
+	 return 0;
+      case GLX_RGBA:
+         if (fbconfig)
+            return GLX_BAD_ATTRIBUTE;
+	 if (xmvis->mesa_visual.rgbMode) {
+	    *value = True;
+	 }
+	 else {
+	    *value = False;
+	 }
+	 return 0;
+      case GLX_DOUBLEBUFFER:
+	 *value = (int) xmvis->mesa_visual.doubleBufferMode;
+	 return 0;
+      case GLX_STEREO:
+	 *value = (int) xmvis->mesa_visual.stereoMode;
+	 return 0;
+      case GLX_AUX_BUFFERS:
+	 *value = xmvis->mesa_visual.numAuxBuffers;
+	 return 0;
+      case GLX_RED_SIZE:
+         *value = xmvis->mesa_visual.redBits;
+	 return 0;
+      case GLX_GREEN_SIZE:
+         *value = xmvis->mesa_visual.greenBits;
+	 return 0;
+      case GLX_BLUE_SIZE:
+         *value = xmvis->mesa_visual.blueBits;
+	 return 0;
+      case GLX_ALPHA_SIZE:
+         *value = xmvis->mesa_visual.alphaBits;
+	 return 0;
+      case GLX_DEPTH_SIZE:
+         *value = xmvis->mesa_visual.depthBits;
+	 return 0;
+      case GLX_STENCIL_SIZE:
+	 *value = xmvis->mesa_visual.stencilBits;
+	 return 0;
+      case GLX_ACCUM_RED_SIZE:
+	 *value = xmvis->mesa_visual.accumRedBits;
+	 return 0;
+      case GLX_ACCUM_GREEN_SIZE:
+	 *value = xmvis->mesa_visual.accumGreenBits;
+	 return 0;
+      case GLX_ACCUM_BLUE_SIZE:
+	 *value = xmvis->mesa_visual.accumBlueBits;
+	 return 0;
+      case GLX_ACCUM_ALPHA_SIZE:
+         *value = xmvis->mesa_visual.accumAlphaBits;
+	 return 0;
+
+      /*
+       * GLX_EXT_visual_info extension
+       */
+      case GLX_X_VISUAL_TYPE_EXT:
+         switch (xmvis->visinfo->CLASS) {
+            case StaticGray:   *value = GLX_STATIC_GRAY_EXT;   return 0;
+            case GrayScale:    *value = GLX_GRAY_SCALE_EXT;    return 0;
+            case StaticColor:  *value = GLX_STATIC_GRAY_EXT;   return 0;
+            case PseudoColor:  *value = GLX_PSEUDO_COLOR_EXT;  return 0;
+            case TrueColor:    *value = GLX_TRUE_COLOR_EXT;    return 0;
+            case DirectColor:  *value = GLX_DIRECT_COLOR_EXT;  return 0;
+         }
+         return 0;
+      case GLX_TRANSPARENT_TYPE_EXT:
+         if (xmvis->mesa_visual.level==0) {
+            /* normal planes */
+            *value = GLX_NONE_EXT;
+         }
+         else if (xmvis->mesa_visual.level>0) {
+            /* overlay */
+            if (xmvis->mesa_visual.rgbMode) {
+               *value = GLX_TRANSPARENT_RGB_EXT;
+            }
+            else {
+               *value = GLX_TRANSPARENT_INDEX_EXT;
+            }
+         }
+         else if (xmvis->mesa_visual.level<0) {
+            /* underlay */
+            *value = GLX_NONE_EXT;
+         }
+         return 0;
+      case GLX_TRANSPARENT_INDEX_VALUE_EXT:
+         {
+            int pixel = transparent_pixel( xmvis );
+            if (pixel>=0) {
+               *value = pixel;
+            }
+            /* else undefined */
+         }
+         return 0;
+      case GLX_TRANSPARENT_RED_VALUE_EXT:
+         /* undefined */
+         return 0;
+      case GLX_TRANSPARENT_GREEN_VALUE_EXT:
+         /* undefined */
+         return 0;
+      case GLX_TRANSPARENT_BLUE_VALUE_EXT:
+         /* undefined */
+         return 0;
+      case GLX_TRANSPARENT_ALPHA_VALUE_EXT:
+         /* undefined */
+         return 0;
+
+      /*
+       * GLX_EXT_visual_info extension
+       */
+      case GLX_VISUAL_CAVEAT_EXT:
+         /* test for zero, just in case */
+         if (xmvis->mesa_visual.visualRating > 0)
+            *value = xmvis->mesa_visual.visualRating;
+         else
+            *value = GLX_NONE_EXT;
+         return 0;
+
+      /*
+       * GLX_ARB_multisample
+       */
+      case GLX_SAMPLE_BUFFERS_ARB:
+         *value = 0;
+         return 0;
+      case GLX_SAMPLES_ARB:
+         *value = 0;
+         return 0;
+
+      /*
+       * For FBConfigs:
+       */
+      case GLX_SCREEN_EXT:
+         if (!fbconfig)
+            return GLX_BAD_ATTRIBUTE;
+         *value = xmvis->visinfo->screen;
+         break;
+      case GLX_DRAWABLE_TYPE: /*SGIX too */
+         if (!fbconfig)
+            return GLX_BAD_ATTRIBUTE;
+         *value = GLX_WINDOW_BIT | GLX_PIXMAP_BIT | GLX_PBUFFER_BIT;
+         break;
+      case GLX_RENDER_TYPE_SGIX:
+         if (!fbconfig)
+            return GLX_BAD_ATTRIBUTE;
+         if (xmvis->mesa_visual.rgbMode)
+            *value = GLX_RGBA_BIT;
+         else
+            *value = GLX_COLOR_INDEX_BIT;
+         break;
+      case GLX_X_RENDERABLE_SGIX:
+         if (!fbconfig)
+            return GLX_BAD_ATTRIBUTE;
+         *value = True; /* XXX really? */
+         break;
+      case GLX_FBCONFIG_ID_SGIX:
+         if (!fbconfig)
+            return GLX_BAD_ATTRIBUTE;
+         *value = xmvis->visinfo->visualid;
+         break;
+      case GLX_MAX_PBUFFER_WIDTH:
+         if (!fbconfig)
+            return GLX_BAD_ATTRIBUTE;
+         /* XXX or MAX_WIDTH? */
+         *value = DisplayWidth(xmvis->display, xmvis->visinfo->screen);
+         break;
+      case GLX_MAX_PBUFFER_HEIGHT:
+         if (!fbconfig)
+            return GLX_BAD_ATTRIBUTE;
+         *value = DisplayHeight(xmvis->display, xmvis->visinfo->screen);
+         break;
+      case GLX_MAX_PBUFFER_PIXELS:
+         if (!fbconfig)
+            return GLX_BAD_ATTRIBUTE;
+         *value = DisplayWidth(xmvis->display, xmvis->visinfo->screen) *
+                  DisplayHeight(xmvis->display, xmvis->visinfo->screen);
+         break;
+      case GLX_VISUAL_ID:
+         if (!fbconfig)
+            return GLX_BAD_ATTRIBUTE;
+         *value = xmvis->visinfo->visualid;
+         break;
+
+#ifdef GLX_EXT_texture_from_pixmap
+      case GLX_BIND_TO_TEXTURE_RGB_EXT:
+         *value = True; /*XXX*/
+         break;
+      case GLX_BIND_TO_TEXTURE_RGBA_EXT:
+         /* XXX review */
+         *value = xmvis->mesa_visual.alphaBits > 0 ? True : False;
+         break;
+      case GLX_BIND_TO_MIPMAP_TEXTURE_EXT:
+         *value = True; /*XXX*/
+         break;
+      case GLX_BIND_TO_TEXTURE_TARGETS_EXT:
+         *value = (GLX_TEXTURE_1D_BIT_EXT |
+                   GLX_TEXTURE_2D_BIT_EXT |
+                   GLX_TEXTURE_RECTANGLE_BIT_EXT); /*XXX*/
+         break;
+      case GLX_Y_INVERTED_EXT:
+         *value = True; /*XXX*/
+         break;
+#endif
+
+      default:
+	 return GLX_BAD_ATTRIBUTE;
+   }
+   return Success;
+}
+
+
+static int
+Fake_glXGetConfig( Display *dpy, XVisualInfo *visinfo,
+                   int attrib, int *value )
+{
+   XMesaVisual xmvis;
+   int k;
+   if (!dpy || !visinfo)
+      return GLX_BAD_ATTRIBUTE;
+
+   xmvis = find_glx_visual( dpy, visinfo );
+   if (!xmvis) {
+      /* this visual wasn't obtained with glXChooseVisual */
+      xmvis = create_glx_visual( dpy, visinfo );
+      if (!xmvis) {
+	 /* this visual can't be used for GL rendering */
+	 if (attrib==GLX_USE_GL) {
+	    *value = (int) False;
+	    return 0;
+	 }
+	 else {
+	    return GLX_BAD_VISUAL;
+	 }
+      }
+   }
+
+   k = get_config(xmvis, attrib, value, GL_FALSE);
+   return k;
+}
+
+
+static void
+Fake_glXWaitGL( void )
+{
+   XMesaContext xmesa = XMesaGetCurrentContext();
+   XMesaFlush( xmesa );
+}
+
+
+
+static void
+Fake_glXWaitX( void )
+{
+   XMesaContext xmesa = XMesaGetCurrentContext();
+   XMesaFlush( xmesa );
+}
+
+
+static const char *
+get_extensions( void )
+{
+#ifdef FX
+   const char *fx = _mesa_getenv("MESA_GLX_FX");
+   if (fx && fx[0] != 'd') {
+      return EXTENSIONS;
+   }
+#endif
+   return EXTENSIONS + 23; /* skip "GLX_MESA_set_3dfx_mode" */
+}
+
+
+
+/* GLX 1.1 and later */
+static const char *
+Fake_glXQueryExtensionsString( Display *dpy, int screen )
+{
+   (void) dpy;
+   (void) screen;
+   return get_extensions();
+}
+
+
+
+/* GLX 1.1 and later */
+static const char *
+Fake_glXQueryServerString( Display *dpy, int screen, int name )
+{
+   static char version[1000];
+   _mesa_sprintf(version, "%d.%d %s",
+                 SERVER_MAJOR_VERSION, SERVER_MINOR_VERSION, MESA_GLX_VERSION);
+
+   (void) dpy;
+   (void) screen;
+
+   switch (name) {
+      case GLX_EXTENSIONS:
+         return get_extensions();
+      case GLX_VENDOR:
+	 return VENDOR;
+      case GLX_VERSION:
+	 return version;
+      default:
+         return NULL;
+   }
+}
+
+
+
+/* GLX 1.1 and later */
+static const char *
+Fake_glXGetClientString( Display *dpy, int name )
+{
+   static char version[1000];
+   _mesa_sprintf(version, "%d.%d %s", CLIENT_MAJOR_VERSION,
+                 CLIENT_MINOR_VERSION, MESA_GLX_VERSION);
+
+   (void) dpy;
+
+   switch (name) {
+      case GLX_EXTENSIONS:
+         return get_extensions();
+      case GLX_VENDOR:
+	 return VENDOR;
+      case GLX_VERSION:
+	 return version;
+      default:
+         return NULL;
+   }
+}
+
+
+
+/*
+ * GLX 1.3 and later
+ */
+
+
+static int
+Fake_glXGetFBConfigAttrib( Display *dpy, GLXFBConfig config,
+                           int attribute, int *value )
+{
+   XMesaVisual v = (XMesaVisual) config;
+   (void) dpy;
+   (void) config;
+
+   if (!dpy || !config || !value)
+      return -1;
+
+   return get_config(v, attribute, value, GL_TRUE);
+}
+
+
+static GLXFBConfig *
+Fake_glXGetFBConfigs( Display *dpy, int screen, int *nelements )
+{
+   XVisualInfo *visuals, visTemplate;
+   const long visMask = VisualScreenMask;
+   int i;
+
+   /* Get list of all X visuals */
+   visTemplate.screen = screen;
+   visuals = XGetVisualInfo(dpy, visMask, &visTemplate, nelements);
+   if (*nelements > 0) {
+      XMesaVisual *results;
+      results = (XMesaVisual *) _mesa_malloc(*nelements * sizeof(XMesaVisual));
+      if (!results) {
+         *nelements = 0;
+         return NULL;
+      }
+      for (i = 0; i < *nelements; i++) {
+         results[i] = create_glx_visual(dpy, visuals + i);
+      }
+      return (GLXFBConfig *) results;
+   }
+   return NULL;
+}
+
+
+static GLXFBConfig *
+Fake_glXChooseFBConfig( Display *dpy, int screen,
+                        const int *attribList, int *nitems )
+{
+   XMesaVisual xmvis;
+
+   if (!attribList || !attribList[0]) {
+      /* return list of all configs (per GLX_SGIX_fbconfig spec) */
+      return Fake_glXGetFBConfigs(dpy, screen, nitems);
+   }
+
+   xmvis = choose_visual(dpy, screen, attribList, GL_TRUE);
+   if (xmvis) {
+      GLXFBConfig *config = (GLXFBConfig *) _mesa_malloc(sizeof(XMesaVisual));
+      if (!config) {
+         *nitems = 0;
+         return NULL;
+      }
+      *nitems = 1;
+      config[0] = (GLXFBConfig) xmvis;
+      return (GLXFBConfig *) config;
+   }
+   else {
+      *nitems = 0;
+      return NULL;
+   }
+}
+
+
+static XVisualInfo *
+Fake_glXGetVisualFromFBConfig( Display *dpy, GLXFBConfig config )
+{
+   if (dpy && config) {
+      XMesaVisual xmvis = (XMesaVisual) config;
+#if 0      
+      return xmvis->vishandle;
+#else
+      /* create a new vishandle - the cached one may be stale */
+      xmvis->vishandle = (XVisualInfo *) _mesa_malloc(sizeof(XVisualInfo));
+      if (xmvis->vishandle) {
+         _mesa_memcpy(xmvis->vishandle, xmvis->visinfo, sizeof(XVisualInfo));
+      }
+      return xmvis->vishandle;
+#endif
+   }
+   else {
+      return NULL;
+   }
+}
+
+
+static GLXWindow
+Fake_glXCreateWindow( Display *dpy, GLXFBConfig config, Window win,
+                      const int *attribList )
+{
+   XMesaVisual xmvis = (XMesaVisual) config;
+   XMesaBuffer xmbuf;
+   if (!xmvis)
+      return 0;
+
+   xmbuf = XMesaCreateWindowBuffer(xmvis, win);
+   if (!xmbuf)
+      return 0;
+
+#ifdef FX
+   /* XXX this will segfault if actually called */
+   FXcreateContext(xmvis, win, NULL, xmbuf);
+#endif
+
+   (void) dpy;
+   (void) attribList;  /* Ignored in GLX 1.3 */
+
+   return win;  /* A hack for now */
+}
+
+
+static void
+Fake_glXDestroyWindow( Display *dpy, GLXWindow window )
+{
+   XMesaBuffer b = XMesaFindBuffer(dpy, (XMesaDrawable) window);
+   if (b)
+      XMesaDestroyBuffer(b);
+   /* don't destroy X window */
+}
+
+
+/* XXX untested */
+static GLXPixmap
+Fake_glXCreatePixmap( Display *dpy, GLXFBConfig config, Pixmap pixmap,
+                      const int *attribList )
+{
+   XMesaVisual v = (XMesaVisual) config;
+   XMesaBuffer b;
+   const int *attr;
+   int target = 0, format = 0, mipmap = 0;
+   int value;
+
+   if (!dpy || !config || !pixmap)
+      return 0;
+
+   for (attr = attribList; *attr; attr++) {
+      switch (*attr) {
+      case GLX_TEXTURE_FORMAT_EXT:
+         attr++;
+         switch (*attr) {
+         case GLX_TEXTURE_FORMAT_NONE_EXT:
+         case GLX_TEXTURE_FORMAT_RGB_EXT:
+         case GLX_TEXTURE_FORMAT_RGBA_EXT:
+            format = *attr;
+            break;
+         default:
+            /* error */
+            return 0;
+         }
+         break;
+      case GLX_TEXTURE_TARGET_EXT:
+         attr++;
+         switch (*attr) {
+         case GLX_TEXTURE_1D_EXT:
+         case GLX_TEXTURE_2D_EXT:
+         case GLX_TEXTURE_RECTANGLE_EXT:
+            target = *attr;
+            break;
+         default:
+            /* error */
+            return 0;
+         }
+         break;
+      case GLX_MIPMAP_TEXTURE_EXT:
+         attr++;
+         if (*attr)
+            mipmap = 1;
+         break;
+      default:
+         /* error */
+         return 0;
+      }
+   }
+
+   if (format == GLX_TEXTURE_FORMAT_RGB_EXT) {
+      if (get_config(v, GLX_BIND_TO_TEXTURE_RGB_EXT,
+                     &value, GL_TRUE) != Success
+          || !value) {
+         return 0; /* error! */
+      }
+   }
+   else if (format == GLX_TEXTURE_FORMAT_RGBA_EXT) {
+      if (get_config(v, GLX_BIND_TO_TEXTURE_RGBA_EXT,
+                     &value, GL_TRUE) != Success
+          || !value) {
+         return 0; /* error! */
+      }
+   }
+   if (mipmap) {
+      if (get_config(v, GLX_BIND_TO_MIPMAP_TEXTURE_EXT,
+                     &value, GL_TRUE) != Success
+          || !value) {
+         return 0; /* error! */
+      }
+   }
+   if (target == GLX_TEXTURE_1D_EXT) {
+      if (get_config(v, GLX_BIND_TO_TEXTURE_TARGETS_EXT,
+                     &value, GL_TRUE) != Success
+          || (value & GLX_TEXTURE_1D_BIT_EXT) == 0) {
+         return 0; /* error! */
+      }
+   }
+   else if (target == GLX_TEXTURE_2D_EXT) {
+      if (get_config(v, GLX_BIND_TO_TEXTURE_TARGETS_EXT,
+                     &value, GL_TRUE) != Success
+          || (value & GLX_TEXTURE_2D_BIT_EXT) == 0) {
+         return 0; /* error! */
+      }
+   }
+   if (target == GLX_TEXTURE_RECTANGLE_EXT) {
+      if (get_config(v, GLX_BIND_TO_TEXTURE_TARGETS_EXT,
+                     &value, GL_TRUE) != Success
+          || (value & GLX_TEXTURE_RECTANGLE_BIT_EXT) == 0) {
+         return 0; /* error! */
+      }
+   }
+
+   if (format || target || mipmap) {
+      /* texture from pixmap */
+      b = XMesaCreatePixmapTextureBuffer(v, pixmap, 0, format, target, mipmap);
+   }
+   else {
+      b = XMesaCreatePixmapBuffer( v, pixmap, 0 );
+   }
+   if (!b) {
+      return 0;
+   }
+
+   return pixmap;
+}
+
+
+static void
+Fake_glXDestroyPixmap( Display *dpy, GLXPixmap pixmap )
+{
+   XMesaBuffer b = XMesaFindBuffer(dpy, (XMesaDrawable)pixmap);
+   if (b)
+      XMesaDestroyBuffer(b);
+   /* don't destroy X pixmap */
+}
+
+
+static GLXPbuffer
+Fake_glXCreatePbuffer( Display *dpy, GLXFBConfig config,
+                       const int *attribList )
+{
+   XMesaVisual xmvis = (XMesaVisual) config;
+   XMesaBuffer xmbuf;
+   const int *attrib;
+   int width = 0, height = 0;
+   GLboolean useLargest = GL_FALSE, preserveContents = GL_FALSE;
+
+   (void) dpy;
+
+   for (attrib = attribList; *attrib; attrib++) {
+      switch (*attrib) {
+         case GLX_PBUFFER_WIDTH:
+            attrib++;
+            width = *attrib;
+            break;
+         case GLX_PBUFFER_HEIGHT:
+            attrib++;
+            height = *attrib;
+            break;
+         case GLX_PRESERVED_CONTENTS:
+            attrib++;
+            preserveContents = *attrib; /* ignored */
+            break;
+         case GLX_LARGEST_PBUFFER:
+            attrib++;
+            useLargest = *attrib; /* ignored */
+            break;
+         default:
+            return 0;
+      }
+   }
+
+   /* not used at this time */
+   (void) useLargest;
+   (void) preserveContents;
+
+   if (width == 0 || height == 0)
+      return 0;
+
+   xmbuf = XMesaCreatePBuffer( xmvis, 0, width, height);
+   /* A GLXPbuffer handle must be an X Drawable because that's what
+    * glXMakeCurrent takes.
+    */
+   if (xmbuf)
+      return (GLXPbuffer) xmbuf->drawable;
+   else
+      return 0;
+}
+
+
+static void
+Fake_glXDestroyPbuffer( Display *dpy, GLXPbuffer pbuf )
+{
+   XMesaBuffer b = XMesaFindBuffer(dpy, pbuf);
+   if (b) {
+      XMesaDestroyBuffer(b);
+   }
+}
+
+
+static void
+Fake_glXQueryDrawable( Display *dpy, GLXDrawable draw, int attribute,
+                       unsigned int *value )
+{
+   XMesaBuffer xmbuf = XMesaFindBuffer(dpy, draw);
+   if (!xmbuf)
+      return;
+
+   switch (attribute) {
+      case GLX_WIDTH:
+         *value = xmesa_buffer_width(xmbuf);
+         break;
+      case GLX_HEIGHT:
+         *value = xmesa_buffer_width(xmbuf);
+         break;
+      case GLX_PRESERVED_CONTENTS:
+         *value = True;
+         break;
+      case GLX_LARGEST_PBUFFER:
+         *value = xmesa_buffer_width(xmbuf) * xmesa_buffer_height(xmbuf);
+         break;
+      case GLX_FBCONFIG_ID:
+         *value = xmbuf->xm_visual->visinfo->visualid;
+         return;
+#ifdef GLX_EXT_texture_from_pixmap
+      case GLX_TEXTURE_FORMAT_EXT:
+         *value = xmbuf->TextureFormat;
+         break;
+      case GLX_TEXTURE_TARGET_EXT:
+         *value = xmbuf->TextureTarget;
+         break;
+      case GLX_MIPMAP_TEXTURE_EXT:
+         *value = xmbuf->TextureMipmap;
+         break;
+#endif
+
+      default:
+         return; /* raise BadValue error */
+   }
+}
+
+
+static GLXContext
+Fake_glXCreateNewContext( Display *dpy, GLXFBConfig config,
+                          int renderType, GLXContext shareList, Bool direct )
+{
+   struct fake_glx_context *glxCtx;
+   struct fake_glx_context *shareCtx = (struct fake_glx_context *) shareList;
+   XMesaVisual xmvis = (XMesaVisual) config;
+
+   if (!dpy || !config ||
+       (renderType != GLX_RGBA_TYPE && renderType != GLX_COLOR_INDEX_TYPE))
+      return 0;
+
+   glxCtx = CALLOC_STRUCT(fake_glx_context);
+   if (!glxCtx)
+      return 0;
+
+   /* deallocate unused windows/buffers */
+   XMesaGarbageCollect();
+
+   glxCtx->xmesaContext = XMesaCreateContext(xmvis,
+                                   shareCtx ? shareCtx->xmesaContext : NULL);
+   if (!glxCtx->xmesaContext) {
+      _mesa_free(glxCtx);
+      return NULL;
+   }
+
+   glxCtx->glxContext.isDirect = GL_FALSE;
+   glxCtx->glxContext.currentDpy = dpy;
+   glxCtx->glxContext.xid = (XID) glxCtx;  /* self pointer */
+
+   assert((void *) glxCtx == (void *) &(glxCtx->glxContext));
+
+   return (GLXContext) glxCtx;
+}
+
+
+static int
+Fake_glXQueryContext( Display *dpy, GLXContext ctx, int attribute, int *value )
+{
+   struct fake_glx_context *glxCtx = (struct fake_glx_context *) ctx;
+   XMesaContext xmctx = glxCtx->xmesaContext;
+
+   (void) dpy;
+   (void) ctx;
+
+   switch (attribute) {
+   case GLX_FBCONFIG_ID:
+      *value = xmctx->xm_visual->visinfo->visualid;
+      break;
+   case GLX_RENDER_TYPE:
+      if (xmctx->xm_visual->mesa_visual.rgbMode)
+         *value = GLX_RGBA_BIT;
+      else
+         *value = GLX_COLOR_INDEX_BIT;
+      break;
+   case GLX_SCREEN:
+      *value = 0;
+      return Success;
+   default:
+      return GLX_BAD_ATTRIBUTE;
+   }
+   return 0;
+}
+
+
+static void
+Fake_glXSelectEvent( Display *dpy, GLXDrawable drawable, unsigned long mask )
+{
+   XMesaBuffer xmbuf = XMesaFindBuffer(dpy, drawable);
+   if (xmbuf)
+      xmbuf->selectedEvents = mask;
+}
+
+
+static void
+Fake_glXGetSelectedEvent( Display *dpy, GLXDrawable drawable,
+                          unsigned long *mask )
+{
+   XMesaBuffer xmbuf = XMesaFindBuffer(dpy, drawable);
+   if (xmbuf)
+      *mask = xmbuf->selectedEvents;
+   else
+      *mask = 0;
+}
+
+
+
+/*** GLX_SGI_swap_control ***/
+
+static int
+Fake_glXSwapIntervalSGI(int interval)
+{
+   (void) interval;
+   return 0;
+}
+
+
+
+/*** GLX_SGI_video_sync ***/
+
+static unsigned int FrameCounter = 0;
+
+static int
+Fake_glXGetVideoSyncSGI(unsigned int *count)
+{
+   /* this is a bogus implementation */
+   *count = FrameCounter++;
+   return 0;
+}
+
+static int
+Fake_glXWaitVideoSyncSGI(int divisor, int remainder, unsigned int *count)
+{
+   if (divisor <= 0 || remainder < 0)
+      return GLX_BAD_VALUE;
+   /* this is a bogus implementation */
+   FrameCounter++;
+   while (FrameCounter % divisor != remainder)
+      FrameCounter++;
+   *count = FrameCounter;
+   return 0;
+}
+
+
+
+/*** GLX_SGI_make_current_read ***/
+
+static Bool
+Fake_glXMakeCurrentReadSGI(Display *dpy, GLXDrawable draw, GLXDrawable read, GLXContext ctx)
+{
+   return Fake_glXMakeContextCurrent( dpy, draw, read, ctx );
+}
+
+/* not used
+static GLXDrawable
+Fake_glXGetCurrentReadDrawableSGI(void)
+{
+   return 0;
+}
+*/
+
+
+/*** GLX_SGIX_video_source ***/
+#if defined(_VL_H)
+
+static GLXVideoSourceSGIX
+Fake_glXCreateGLXVideoSourceSGIX(Display *dpy, int screen, VLServer server, VLPath path, int nodeClass, VLNode drainNode)
+{
+   (void) dpy;
+   (void) screen;
+   (void) server;
+   (void) path;
+   (void) nodeClass;
+   (void) drainNode;
+   return 0;
+}
+
+static void
+Fake_glXDestroyGLXVideoSourceSGIX(Display *dpy, GLXVideoSourceSGIX src)
+{
+   (void) dpy;
+   (void) src;
+}
+
+#endif
+
+
+/*** GLX_EXT_import_context ***/
+
+static void
+Fake_glXFreeContextEXT(Display *dpy, GLXContext context)
+{
+   (void) dpy;
+   (void) context;
+}
+
+static GLXContextID
+Fake_glXGetContextIDEXT(const GLXContext context)
+{
+   (void) context;
+   return 0;
+}
+
+static GLXContext
+Fake_glXImportContextEXT(Display *dpy, GLXContextID contextID)
+{
+   (void) dpy;
+   (void) contextID;
+   return 0;
+}
+
+static int
+Fake_glXQueryContextInfoEXT(Display *dpy, GLXContext context, int attribute, int *value)
+{
+   (void) dpy;
+   (void) context;
+   (void) attribute;
+   (void) value;
+   return 0;
+}
+
+
+
+/*** GLX_SGIX_fbconfig ***/
+
+static int
+Fake_glXGetFBConfigAttribSGIX(Display *dpy, GLXFBConfigSGIX config, int attribute, int *value)
+{
+   return Fake_glXGetFBConfigAttrib(dpy, config, attribute, value);
+}
+
+static GLXFBConfigSGIX *
+Fake_glXChooseFBConfigSGIX(Display *dpy, int screen, int *attrib_list, int *nelements)
+{
+   return (GLXFBConfig *) Fake_glXChooseFBConfig(dpy, screen, attrib_list, nelements);
+}
+
+
+static GLXPixmap
+Fake_glXCreateGLXPixmapWithConfigSGIX(Display *dpy, GLXFBConfigSGIX config, Pixmap pixmap)
+{
+   XMesaVisual xmvis = (XMesaVisual) config;
+   XMesaBuffer xmbuf = XMesaCreatePixmapBuffer(xmvis, pixmap, 0);
+   return xmbuf->drawable; /* need to return an X ID */
+}
+
+
+static GLXContext
+Fake_glXCreateContextWithConfigSGIX(Display *dpy, GLXFBConfigSGIX config, int render_type, GLXContext share_list, Bool direct)
+{
+   XMesaVisual xmvis = (XMesaVisual) config;
+   struct fake_glx_context *glxCtx;
+   struct fake_glx_context *shareCtx = (struct fake_glx_context *) share_list;
+
+   glxCtx = CALLOC_STRUCT(fake_glx_context);
+   if (!glxCtx)
+      return 0;
+
+   /* deallocate unused windows/buffers */
+   XMesaGarbageCollect();
+
+   glxCtx->xmesaContext = XMesaCreateContext(xmvis,
+                                   shareCtx ? shareCtx->xmesaContext : NULL);
+   if (!glxCtx->xmesaContext) {
+      _mesa_free(glxCtx);
+      return NULL;
+   }
+
+   glxCtx->glxContext.isDirect = GL_FALSE;
+   glxCtx->glxContext.currentDpy = dpy;
+   glxCtx->glxContext.xid = (XID) glxCtx;  /* self pointer */
+
+   assert((void *) glxCtx == (void *) &(glxCtx->glxContext));
+
+   return (GLXContext) glxCtx;
+}
+
+
+static XVisualInfo *
+Fake_glXGetVisualFromFBConfigSGIX(Display *dpy, GLXFBConfigSGIX config)
+{
+   return Fake_glXGetVisualFromFBConfig(dpy, config);
+}
+
+
+static GLXFBConfigSGIX
+Fake_glXGetFBConfigFromVisualSGIX(Display *dpy, XVisualInfo *vis)
+{
+   XMesaVisual xmvis = find_glx_visual(dpy, vis);
+   if (!xmvis) {
+      /* This visual wasn't found with glXChooseVisual() */
+      xmvis = create_glx_visual(dpy, vis);
+   }
+
+   return (GLXFBConfigSGIX) xmvis;
+}
+
+
+
+/*** GLX_SGIX_pbuffer ***/
+
+static GLXPbufferSGIX
+Fake_glXCreateGLXPbufferSGIX(Display *dpy, GLXFBConfigSGIX config,
+                             unsigned int width, unsigned int height,
+                             int *attribList)
+{
+   XMesaVisual xmvis = (XMesaVisual) config;
+   XMesaBuffer xmbuf;
+   const int *attrib;
+   GLboolean useLargest = GL_FALSE, preserveContents = GL_FALSE;
+
+   (void) dpy;
+
+   for (attrib = attribList; attrib && *attrib; attrib++) {
+      switch (*attrib) {
+         case GLX_PRESERVED_CONTENTS_SGIX:
+            attrib++;
+            preserveContents = *attrib; /* ignored */
+            break;
+         case GLX_LARGEST_PBUFFER_SGIX:
+            attrib++;
+            useLargest = *attrib; /* ignored */
+            break;
+         default:
+            return 0;
+      }
+   }
+
+   /* not used at this time */
+   (void) useLargest;
+   (void) preserveContents;
+
+   xmbuf = XMesaCreatePBuffer( xmvis, 0, width, height);
+   /* A GLXPbuffer handle must be an X Drawable because that's what
+    * glXMakeCurrent takes.
+    */
+   return (GLXPbuffer) xmbuf->drawable;
+}
+
+
+static void
+Fake_glXDestroyGLXPbufferSGIX(Display *dpy, GLXPbufferSGIX pbuf)
+{
+   XMesaBuffer xmbuf = XMesaFindBuffer(dpy, pbuf);
+   if (xmbuf) {
+      XMesaDestroyBuffer(xmbuf);
+   }
+}
+
+
+static int
+Fake_glXQueryGLXPbufferSGIX(Display *dpy, GLXPbufferSGIX pbuf, int attribute, unsigned int *value)
+{
+   const XMesaBuffer xmbuf = XMesaFindBuffer(dpy, pbuf);
+
+   if (!xmbuf) {
+      /* Generate GLXBadPbufferSGIX for bad pbuffer */
+      return 0;
+   }
+
+   switch (attribute) {
+      case GLX_PRESERVED_CONTENTS_SGIX:
+         *value = True;
+         break;
+      case GLX_LARGEST_PBUFFER_SGIX:
+         *value = xmesa_buffer_width(xmbuf) * xmesa_buffer_height(xmbuf);
+         break;
+      case GLX_WIDTH_SGIX:
+         *value = xmesa_buffer_width(xmbuf);
+         break;
+      case GLX_HEIGHT_SGIX:
+         *value = xmesa_buffer_height(xmbuf);
+         break;
+      case GLX_EVENT_MASK_SGIX:
+         *value = 0;  /* XXX might be wrong */
+         break;
+      default:
+         *value = 0;
+   }
+   return 0;
+}
+
+
+static void
+Fake_glXSelectEventSGIX(Display *dpy, GLXDrawable drawable, unsigned long mask)
+{
+   XMesaBuffer xmbuf = XMesaFindBuffer(dpy, drawable);
+   if (xmbuf) {
+      /* Note: we'll never generate clobber events */
+      xmbuf->selectedEvents = mask;
+   }
+}
+
+
+static void
+Fake_glXGetSelectedEventSGIX(Display *dpy, GLXDrawable drawable, unsigned long *mask)
+{
+   XMesaBuffer xmbuf = XMesaFindBuffer(dpy, drawable);
+   if (xmbuf) {
+      *mask = xmbuf->selectedEvents;
+   }
+   else {
+      *mask = 0;
+   }
+}
+
+
+
+/*** GLX_SGI_cushion ***/
+
+static void
+Fake_glXCushionSGI(Display *dpy, Window win, float cushion)
+{
+   (void) dpy;
+   (void) win;
+   (void) cushion;
+}
+
+
+
+/*** GLX_SGIX_video_resize ***/
+
+static int
+Fake_glXBindChannelToWindowSGIX(Display *dpy, int screen, int channel , Window window)
+{
+   (void) dpy;
+   (void) screen;
+   (void) channel;
+   (void) window;
+   return 0;
+}
+
+static int
+Fake_glXChannelRectSGIX(Display *dpy, int screen, int channel, int x, int y, int w, int h)
+{
+   (void) dpy;
+   (void) screen;
+   (void) channel;
+   (void) x;
+   (void) y;
+   (void) w;
+   (void) h;
+   return 0;
+}
+
+static int
+Fake_glXQueryChannelRectSGIX(Display *dpy, int screen, int channel, int *x, int *y, int *w, int *h)
+{
+   (void) dpy;
+   (void) screen;
+   (void) channel;
+   (void) x;
+   (void) y;
+   (void) w;
+   (void) h;
+   return 0;
+}
+
+static int
+Fake_glXQueryChannelDeltasSGIX(Display *dpy, int screen, int channel, int *dx, int *dy, int *dw, int *dh)
+{
+   (void) dpy;
+   (void) screen;
+   (void) channel;
+   (void) dx;
+   (void) dy;
+   (void) dw;
+   (void) dh;
+   return 0;
+}
+
+static int
+Fake_glXChannelRectSyncSGIX(Display *dpy, int screen, int channel, GLenum synctype)
+{
+   (void) dpy;
+   (void) screen;
+   (void) channel;
+   (void) synctype;
+   return 0;
+}
+
+
+
+/*** GLX_SGIX_dmbuffer **/
+
+#if defined(_DM_BUFFER_H_)
+static Bool
+Fake_glXAssociateDMPbufferSGIX(Display *dpy, GLXPbufferSGIX pbuffer, DMparams *params, DMbuffer dmbuffer)
+{
+   (void) dpy;
+   (void) pbuffer;
+   (void) params;
+   (void) dmbuffer;
+   return False;
+}
+#endif
+
+
+/*** GLX_SGIX_swap_group ***/
+
+static void
+Fake_glXJoinSwapGroupSGIX(Display *dpy, GLXDrawable drawable, GLXDrawable member)
+{
+   (void) dpy;
+   (void) drawable;
+   (void) member;
+}
+
+
+
+/*** GLX_SGIX_swap_barrier ***/
+
+static void
+Fake_glXBindSwapBarrierSGIX(Display *dpy, GLXDrawable drawable, int barrier)
+{
+   (void) dpy;
+   (void) drawable;
+   (void) barrier;
+}
+
+static Bool
+Fake_glXQueryMaxSwapBarriersSGIX(Display *dpy, int screen, int *max)
+{
+   (void) dpy;
+   (void) screen;
+   (void) max;
+   return False;
+}
+
+
+
+/*** GLX_SUN_get_transparent_index ***/
+
+static Status
+Fake_glXGetTransparentIndexSUN(Display *dpy, Window overlay, Window underlay, long *pTransparent)
+{
+   (void) dpy;
+   (void) overlay;
+   (void) underlay;
+   (void) pTransparent;
+   return 0;
+}
+
+
+
+/*** GLX_MESA_release_buffers ***/
+
+/*
+ * Release the depth, stencil, accum buffers attached to a GLXDrawable
+ * (a window or pixmap) prior to destroying the GLXDrawable.
+ */
+static Bool
+Fake_glXReleaseBuffersMESA( Display *dpy, GLXDrawable d )
+{
+   XMesaBuffer b = XMesaFindBuffer(dpy, d);
+   if (b) {
+      XMesaDestroyBuffer(b);
+      return True;
+   }
+   return False;
+}
+
+
+
+/*** GLX_MESA_set_3dfx_mode ***/
+
+static Bool
+Fake_glXSet3DfxModeMESA( int mode )
+{
+   return XMesaSetFXmode( mode );
+}
+
+
+
+/*** GLX_NV_vertex_array range ***/
+static void *
+Fake_glXAllocateMemoryNV( GLsizei size,
+                          GLfloat readFrequency,
+                          GLfloat writeFrequency,
+                          GLfloat priority )
+{
+   (void) size;
+   (void) readFrequency;
+   (void) writeFrequency;
+   (void) priority;
+   return NULL;
+}
+
+
+static void 
+Fake_glXFreeMemoryNV( GLvoid *pointer )
+{
+   (void) pointer;
+}
+
+
+/*** GLX_MESA_agp_offset ***/
+
+static GLuint
+Fake_glXGetAGPOffsetMESA( const GLvoid *pointer )
+{
+   (void) pointer;
+   return ~0;
+}
+
+
+/*** GLX_EXT_texture_from_pixmap ***/
+
+static void
+Fake_glXBindTexImageEXT(Display *dpy, GLXDrawable drawable, int buffer,
+                        const int *attrib_list)
+{
+   XMesaBuffer b = XMesaFindBuffer(dpy, drawable);
+   if (b)
+      XMesaBindTexImage(dpy, b, buffer, attrib_list);
+}
+
+static void
+Fake_glXReleaseTexImageEXT(Display *dpy, GLXDrawable drawable, int buffer)
+{
+   XMesaBuffer b = XMesaFindBuffer(dpy, drawable);
+   if (b)
+      XMesaReleaseTexImage(dpy, b, buffer);
+}
+
+
+/* silence warning */
+extern struct _glxapi_table *_mesa_GetGLXDispatchTable(void);
+
+
+/**
+ * Create a new GLX API dispatch table with its function pointers
+ * initialized to point to Mesa's "fake" GLX API functions.
+ * Note: there's a similar function (_real_GetGLXDispatchTable) that
+ * returns a new dispatch table with all pointers initalized to point
+ * to "real" GLX functions (which understand GLX wire protocol, etc).
+ */
+struct _glxapi_table *
+_mesa_GetGLXDispatchTable(void)
+{
+   static struct _glxapi_table glx;
+
+   /* be sure our dispatch table size <= libGL's table */
+   {
+      GLuint size = sizeof(struct _glxapi_table) / sizeof(void *);
+      (void) size;
+      assert(_glxapi_get_dispatch_table_size() >= size);
+   }
+
+   /* initialize the whole table to no-ops */
+   _glxapi_set_no_op_table(&glx);
+
+   /* now initialize the table with the functions I implement */
+   glx.ChooseVisual = Fake_glXChooseVisual;
+   glx.CopyContext = Fake_glXCopyContext;
+   glx.CreateContext = Fake_glXCreateContext;
+   glx.CreateGLXPixmap = Fake_glXCreateGLXPixmap;
+   glx.DestroyContext = Fake_glXDestroyContext;
+   glx.DestroyGLXPixmap = Fake_glXDestroyGLXPixmap;
+   glx.GetConfig = Fake_glXGetConfig;
+   /*glx.GetCurrentContext = Fake_glXGetCurrentContext;*/
+   /*glx.GetCurrentDrawable = Fake_glXGetCurrentDrawable;*/
+   glx.IsDirect = Fake_glXIsDirect;
+   glx.MakeCurrent = Fake_glXMakeCurrent;
+   glx.QueryExtension = Fake_glXQueryExtension;
+   glx.QueryVersion = Fake_glXQueryVersion;
+   glx.SwapBuffers = Fake_glXSwapBuffers;
+   glx.UseXFont = Fake_glXUseXFont;
+   glx.WaitGL = Fake_glXWaitGL;
+   glx.WaitX = Fake_glXWaitX;
+
+   /*** GLX_VERSION_1_1 ***/
+   glx.GetClientString = Fake_glXGetClientString;
+   glx.QueryExtensionsString = Fake_glXQueryExtensionsString;
+   glx.QueryServerString = Fake_glXQueryServerString;
+
+   /*** GLX_VERSION_1_2 ***/
+   /*glx.GetCurrentDisplay = Fake_glXGetCurrentDisplay;*/
+
+   /*** GLX_VERSION_1_3 ***/
+   glx.ChooseFBConfig = Fake_glXChooseFBConfig;
+   glx.CreateNewContext = Fake_glXCreateNewContext;
+   glx.CreatePbuffer = Fake_glXCreatePbuffer;
+   glx.CreatePixmap = Fake_glXCreatePixmap;
+   glx.CreateWindow = Fake_glXCreateWindow;
+   glx.DestroyPbuffer = Fake_glXDestroyPbuffer;
+   glx.DestroyPixmap = Fake_glXDestroyPixmap;
+   glx.DestroyWindow = Fake_glXDestroyWindow;
+   /*glx.GetCurrentReadDrawable = Fake_glXGetCurrentReadDrawable;*/
+   glx.GetFBConfigAttrib = Fake_glXGetFBConfigAttrib;
+   glx.GetFBConfigs = Fake_glXGetFBConfigs;
+   glx.GetSelectedEvent = Fake_glXGetSelectedEvent;
+   glx.GetVisualFromFBConfig = Fake_glXGetVisualFromFBConfig;
+   glx.MakeContextCurrent = Fake_glXMakeContextCurrent;
+   glx.QueryContext = Fake_glXQueryContext;
+   glx.QueryDrawable = Fake_glXQueryDrawable;
+   glx.SelectEvent = Fake_glXSelectEvent;
+
+   /*** GLX_SGI_swap_control ***/
+   glx.SwapIntervalSGI = Fake_glXSwapIntervalSGI;
+
+   /*** GLX_SGI_video_sync ***/
+   glx.GetVideoSyncSGI = Fake_glXGetVideoSyncSGI;
+   glx.WaitVideoSyncSGI = Fake_glXWaitVideoSyncSGI;
+
+   /*** GLX_SGI_make_current_read ***/
+   glx.MakeCurrentReadSGI = Fake_glXMakeCurrentReadSGI;
+   /*glx.GetCurrentReadDrawableSGI = Fake_glXGetCurrentReadDrawableSGI;*/
+
+/*** GLX_SGIX_video_source ***/
+#if defined(_VL_H)
+   glx.CreateGLXVideoSourceSGIX = Fake_glXCreateGLXVideoSourceSGIX;
+   glx.DestroyGLXVideoSourceSGIX = Fake_glXDestroyGLXVideoSourceSGIX;
+#endif
+
+   /*** GLX_EXT_import_context ***/
+   glx.FreeContextEXT = Fake_glXFreeContextEXT;
+   glx.GetContextIDEXT = Fake_glXGetContextIDEXT;
+   /*glx.GetCurrentDisplayEXT = Fake_glXGetCurrentDisplayEXT;*/
+   glx.ImportContextEXT = Fake_glXImportContextEXT;
+   glx.QueryContextInfoEXT = Fake_glXQueryContextInfoEXT;
+
+   /*** GLX_SGIX_fbconfig ***/
+   glx.GetFBConfigAttribSGIX = Fake_glXGetFBConfigAttribSGIX;
+   glx.ChooseFBConfigSGIX = Fake_glXChooseFBConfigSGIX;
+   glx.CreateGLXPixmapWithConfigSGIX = Fake_glXCreateGLXPixmapWithConfigSGIX;
+   glx.CreateContextWithConfigSGIX = Fake_glXCreateContextWithConfigSGIX;
+   glx.GetVisualFromFBConfigSGIX = Fake_glXGetVisualFromFBConfigSGIX;
+   glx.GetFBConfigFromVisualSGIX = Fake_glXGetFBConfigFromVisualSGIX;
+
+   /*** GLX_SGIX_pbuffer ***/
+   glx.CreateGLXPbufferSGIX = Fake_glXCreateGLXPbufferSGIX;
+   glx.DestroyGLXPbufferSGIX = Fake_glXDestroyGLXPbufferSGIX;
+   glx.QueryGLXPbufferSGIX = Fake_glXQueryGLXPbufferSGIX;
+   glx.SelectEventSGIX = Fake_glXSelectEventSGIX;
+   glx.GetSelectedEventSGIX = Fake_glXGetSelectedEventSGIX;
+
+   /*** GLX_SGI_cushion ***/
+   glx.CushionSGI = Fake_glXCushionSGI;
+
+   /*** GLX_SGIX_video_resize ***/
+   glx.BindChannelToWindowSGIX = Fake_glXBindChannelToWindowSGIX;
+   glx.ChannelRectSGIX = Fake_glXChannelRectSGIX;
+   glx.QueryChannelRectSGIX = Fake_glXQueryChannelRectSGIX;
+   glx.QueryChannelDeltasSGIX = Fake_glXQueryChannelDeltasSGIX;
+   glx.ChannelRectSyncSGIX = Fake_glXChannelRectSyncSGIX;
+
+   /*** GLX_SGIX_dmbuffer **/
+#if defined(_DM_BUFFER_H_)
+   glx.AssociateDMPbufferSGIX = NULL;
+#endif
+
+   /*** GLX_SGIX_swap_group ***/
+   glx.JoinSwapGroupSGIX = Fake_glXJoinSwapGroupSGIX;
+
+   /*** GLX_SGIX_swap_barrier ***/
+   glx.BindSwapBarrierSGIX = Fake_glXBindSwapBarrierSGIX;
+   glx.QueryMaxSwapBarriersSGIX = Fake_glXQueryMaxSwapBarriersSGIX;
+
+   /*** GLX_SUN_get_transparent_index ***/
+   glx.GetTransparentIndexSUN = Fake_glXGetTransparentIndexSUN;
+
+   /*** GLX_MESA_copy_sub_buffer ***/
+   glx.CopySubBufferMESA = Fake_glXCopySubBufferMESA;
+
+   /*** GLX_MESA_release_buffers ***/
+   glx.ReleaseBuffersMESA = Fake_glXReleaseBuffersMESA;
+
+   /*** GLX_MESA_pixmap_colormap ***/
+   glx.CreateGLXPixmapMESA = Fake_glXCreateGLXPixmapMESA;
+
+   /*** GLX_MESA_set_3dfx_mode ***/
+   glx.Set3DfxModeMESA = Fake_glXSet3DfxModeMESA;
+
+   /*** GLX_NV_vertex_array_range ***/
+   glx.AllocateMemoryNV = Fake_glXAllocateMemoryNV;
+   glx.FreeMemoryNV = Fake_glXFreeMemoryNV;
+
+   /*** GLX_MESA_agp_offset ***/
+   glx.GetAGPOffsetMESA = Fake_glXGetAGPOffsetMESA;
+
+   /*** GLX_EXT_texture_from_pixmap ***/
+   glx.BindTexImageEXT = Fake_glXBindTexImageEXT;
+   glx.ReleaseTexImageEXT = Fake_glXReleaseTexImageEXT;
+
+   return &glx;
+}
diff --git a/src/gallium/winsys/xlib/glxapi.c b/src/gallium/winsys/xlib/glxapi.c
new file mode 100644
index 0000000000..c2ccce6f52
--- /dev/null
+++ b/src/gallium/winsys/xlib/glxapi.c
@@ -0,0 +1,1408 @@
+/*
+ * Mesa 3-D graphics library
+ * Version:  7.1
+ * 
+ * Copyright (C) 1999-2007  Brian Paul   All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ * 
+ * The above copyright notice and this permission notice shall be included
+ * in all copies or substantial portions of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * BRIAN PAUL BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN
+ * AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ */
+
+
+/*
+ * This is the GLX API dispatcher.  Calls to the glX* functions are
+ * either routed to the real GLX encoders or to Mesa's pseudo-GLX functions.
+ * See the glxapi.h file for more details.
+ */
+
+
+#include <assert.h>
+#include <stdlib.h>
+#include <stdio.h>
+#include <string.h>
+#include "main/glheader.h"
+#include "glapi/glapi.h"
+#include "glxapi.h"
+
+
+extern struct _glxapi_table *_real_GetGLXDispatchTable(void);
+extern struct _glxapi_table *_mesa_GetGLXDispatchTable(void);
+
+
+struct display_dispatch {
+   Display *Dpy;
+   struct _glxapi_table *Table;
+   struct display_dispatch *Next;
+};
+
+static struct display_dispatch *DispatchList = NULL;
+
+
+/* Display -> Dispatch caching */
+static Display *prevDisplay = NULL;
+static struct _glxapi_table *prevTable = NULL;
+
+
+static struct _glxapi_table *
+get_dispatch(Display *dpy)
+{
+   if (!dpy)
+      return NULL;
+
+   /* search list of display/dispatch pairs for this display */
+   {
+      const struct display_dispatch *d = DispatchList;
+      while (d) {
+         if (d->Dpy == dpy) {
+            prevDisplay = dpy;
+            prevTable = d->Table;
+            return d->Table;  /* done! */
+         }
+         d = d->Next;
+      }
+   }
+
+   /* A new display, determine if we should use real GLX
+    * or Mesa's pseudo-GLX.
+    */
+   {
+      struct _glxapi_table *t = _mesa_GetGLXDispatchTable();
+
+      if (t) {
+         struct display_dispatch *d;
+         d = (struct display_dispatch *) malloc(sizeof(struct display_dispatch));
+         if (d) {
+            d->Dpy = dpy;
+            d->Table = t;
+            /* insert at head of list */
+            d->Next = DispatchList;
+            DispatchList = d;
+            /* update cache */
+            prevDisplay = dpy;
+            prevTable = t;
+            return t;
+         }
+      }
+   }
+
+   /* If we get here that means we can't use real GLX on this display
+    * and the Mesa pseudo-GLX software renderer wasn't compiled in.
+    * Or, we ran out of memory!
+    */
+   return NULL;
+}
+
+
+/* Don't use the GET_DISPATCH defined in glthread.h */
+#undef GET_DISPATCH
+
+#define GET_DISPATCH(DPY, TABLE)	\
+   if (DPY == prevDisplay) {		\
+      TABLE = prevTable;		\
+   }					\
+   else if (!DPY) {			\
+      TABLE = NULL;			\
+   }					\
+   else {				\
+      TABLE = get_dispatch(DPY);	\
+   }
+
+   
+
+
+/**
+ * GLX API current context.
+ */
+#if defined(GLX_USE_TLS)
+PUBLIC __thread void * CurrentContext
+    __attribute__((tls_model("initial-exec")));
+#elif defined(THREADS)
+static _glthread_TSD ContextTSD;         /**< Per-thread context pointer */
+#else
+static GLXContext CurrentContext = 0;
+#endif
+
+
+static void
+SetCurrentContext(GLXContext c)
+{
+#if defined(GLX_USE_TLS)
+   CurrentContext = c;
+#elif defined(THREADS)
+   _glthread_SetTSD(&ContextTSD, c);
+#else
+   CurrentContext = c;
+#endif
+}
+
+
+/*
+ * GLX API entrypoints
+ */
+
+/*** GLX_VERSION_1_0 ***/
+
+XVisualInfo PUBLIC *
+glXChooseVisual(Display *dpy, int screen, int *list)
+{
+   struct _glxapi_table *t;
+   GET_DISPATCH(dpy, t);
+   if (!t)
+      return NULL;
+   return (t->ChooseVisual)(dpy, screen, list);
+}
+
+
+void PUBLIC
+glXCopyContext(Display *dpy, GLXContext src, GLXContext dst, unsigned long mask)
+{
+   struct _glxapi_table *t;
+   GET_DISPATCH(dpy, t);
+   if (!t)
+      return;
+   (t->CopyContext)(dpy, src, dst, mask);
+}
+
+
+GLXContext PUBLIC
+glXCreateContext(Display *dpy, XVisualInfo *visinfo, GLXContext shareList, Bool direct)
+{
+   struct _glxapi_table *t;
+   GET_DISPATCH(dpy, t);
+   if (!t)
+      return 0;
+   return (t->CreateContext)(dpy, visinfo, shareList, direct);
+}
+
+
+GLXPixmap PUBLIC
+glXCreateGLXPixmap(Display *dpy, XVisualInfo *visinfo, Pixmap pixmap)
+{
+   struct _glxapi_table *t;
+   GET_DISPATCH(dpy, t);
+   if (!t)
+      return 0;
+   return (t->CreateGLXPixmap)(dpy, visinfo, pixmap);
+}
+
+
+void PUBLIC
+glXDestroyContext(Display *dpy, GLXContext ctx)
+{
+   struct _glxapi_table *t;
+   GET_DISPATCH(dpy, t);
+   if (!t)
+      return;
+   if (glXGetCurrentContext() == ctx)
+      SetCurrentContext(NULL);
+   (t->DestroyContext)(dpy, ctx);
+}
+
+
+void PUBLIC
+glXDestroyGLXPixmap(Display *dpy, GLXPixmap pixmap)
+{
+   struct _glxapi_table *t;
+   GET_DISPATCH(dpy, t);
+   if (!t)
+      return;
+   (t->DestroyGLXPixmap)(dpy, pixmap);
+}
+
+
+int PUBLIC
+glXGetConfig(Display *dpy, XVisualInfo *visinfo, int attrib, int *value)
+{
+   struct _glxapi_table *t;
+   GET_DISPATCH(dpy, t);
+   if (!t)
+      return GLX_NO_EXTENSION;
+   return (t->GetConfig)(dpy, visinfo, attrib, value);
+}
+
+
+GLXContext PUBLIC
+glXGetCurrentContext(void)
+{
+#if defined(GLX_USE_TLS)
+   return CurrentContext;
+#elif defined(THREADS)
+   return (GLXContext) _glthread_GetTSD(&ContextTSD);
+#else
+   return CurrentContext;
+#endif
+}
+
+
+GLXDrawable PUBLIC
+glXGetCurrentDrawable(void)
+{
+   __GLXcontext *gc = (__GLXcontext *) glXGetCurrentContext();
+   return gc ? gc->currentDrawable : 0;
+}
+
+
+Bool PUBLIC
+glXIsDirect(Display *dpy, GLXContext ctx)
+{
+   struct _glxapi_table *t;
+   GET_DISPATCH(dpy, t);
+   if (!t)
+      return False;
+   return (t->IsDirect)(dpy, ctx);
+}
+
+
+Bool PUBLIC
+glXMakeCurrent(Display *dpy, GLXDrawable drawable, GLXContext ctx)
+{
+   Bool b;
+   struct _glxapi_table *t;
+   GET_DISPATCH(dpy, t);
+   if (!t) {
+      return False;
+   }
+   b = (*t->MakeCurrent)(dpy, drawable, ctx);
+   if (b) {
+      SetCurrentContext(ctx);
+   }
+   return b;
+}
+
+
+Bool PUBLIC
+glXQueryExtension(Display *dpy, int *errorb, int *event)
+{
+   struct _glxapi_table *t;
+   GET_DISPATCH(dpy, t);
+   if (!t)
+      return False;
+   return (t->QueryExtension)(dpy, errorb, event);
+}
+
+
+Bool PUBLIC
+glXQueryVersion(Display *dpy, int *maj, int *min)
+{
+   struct _glxapi_table *t;
+   GET_DISPATCH(dpy, t);
+   if (!t)
+      return False;
+   return (t->QueryVersion)(dpy, maj, min);
+}
+
+
+void PUBLIC
+glXSwapBuffers(Display *dpy, GLXDrawable drawable)
+{
+   struct _glxapi_table *t;
+   GET_DISPATCH(dpy, t);
+   if (!t)
+      return;
+   (t->SwapBuffers)(dpy, drawable);
+}
+
+
+void PUBLIC
+glXUseXFont(Font font, int first, int count, int listBase)
+{
+   struct _glxapi_table *t;
+   Display *dpy = glXGetCurrentDisplay();
+   GET_DISPATCH(dpy, t);
+   if (!t)
+      return;
+   (t->UseXFont)(font, first, count, listBase);
+}
+
+
+void PUBLIC
+glXWaitGL(void)
+{
+   struct _glxapi_table *t;
+   Display *dpy = glXGetCurrentDisplay();
+   GET_DISPATCH(dpy, t);
+   if (!t)
+      return;
+   (t->WaitGL)();
+}
+
+
+void PUBLIC
+glXWaitX(void)
+{
+   struct _glxapi_table *t;
+   Display *dpy = glXGetCurrentDisplay();
+   GET_DISPATCH(dpy, t);
+   if (!t)
+      return;
+   (t->WaitX)();
+}
+
+
+
+/*** GLX_VERSION_1_1 ***/
+
+const char PUBLIC *
+glXGetClientString(Display *dpy, int name)
+{
+   struct _glxapi_table *t;
+   GET_DISPATCH(dpy, t);
+   if (!t)
+      return NULL;
+   return (t->GetClientString)(dpy, name);
+}
+
+
+const char PUBLIC *
+glXQueryExtensionsString(Display *dpy, int screen)
+{
+   struct _glxapi_table *t;
+   GET_DISPATCH(dpy, t);
+   if (!t)
+      return NULL;
+   return (t->QueryExtensionsString)(dpy, screen);
+}
+
+
+const char PUBLIC *
+glXQueryServerString(Display *dpy, int screen, int name)
+{
+   struct _glxapi_table *t;
+   GET_DISPATCH(dpy, t);
+   if (!t)
+      return NULL;
+   return (t->QueryServerString)(dpy, screen, name);
+}
+
+
+/*** GLX_VERSION_1_2 ***/
+
+Display PUBLIC *
+glXGetCurrentDisplay(void)
+{
+   /* Same code as in libGL's glxext.c */
+   __GLXcontext *gc = (__GLXcontext *) glXGetCurrentContext();
+   if (NULL == gc) return NULL;
+   return gc->currentDpy;
+}
+
+
+
+/*** GLX_VERSION_1_3 ***/
+
+GLXFBConfig PUBLIC *
+glXChooseFBConfig(Display *dpy, int screen, const int *attribList, int *nitems)
+{
+   struct _glxapi_table *t;
+   GET_DISPATCH(dpy, t);
+   if (!t)
+      return 0;
+   return (t->ChooseFBConfig)(dpy, screen, attribList, nitems);
+}
+
+
+GLXContext PUBLIC
+glXCreateNewContext(Display *dpy, GLXFBConfig config, int renderType, GLXContext shareList, Bool direct)
+{
+   struct _glxapi_table *t;
+   GET_DISPATCH(dpy, t);
+   if (!t)
+      return 0;
+   return (t->CreateNewContext)(dpy, config, renderType, shareList, direct);
+}
+
+
+GLXPbuffer PUBLIC
+glXCreatePbuffer(Display *dpy, GLXFBConfig config, const int *attribList)
+{
+   struct _glxapi_table *t;
+   GET_DISPATCH(dpy, t);
+   if (!t)
+      return 0;
+   return (t->CreatePbuffer)(dpy, config, attribList);
+}
+
+
+GLXPixmap PUBLIC
+glXCreatePixmap(Display *dpy, GLXFBConfig config, Pixmap pixmap, const int *attribList)
+{
+   struct _glxapi_table *t;
+   GET_DISPATCH(dpy, t);
+   if (!t)
+      return 0;
+   return (t->CreatePixmap)(dpy, config, pixmap, attribList);
+}
+
+
+GLXWindow PUBLIC
+glXCreateWindow(Display *dpy, GLXFBConfig config, Window win, const int *attribList)
+{
+   struct _glxapi_table *t;
+   GET_DISPATCH(dpy, t);
+   if (!t)
+      return 0;
+   return (t->CreateWindow)(dpy, config, win, attribList);
+}
+
+
+void PUBLIC
+glXDestroyPbuffer(Display *dpy, GLXPbuffer pbuf)
+{
+   struct _glxapi_table *t;
+   GET_DISPATCH(dpy, t);
+   if (!t)
+      return;
+   (t->DestroyPbuffer)(dpy, pbuf);
+}
+
+
+void PUBLIC
+glXDestroyPixmap(Display *dpy, GLXPixmap pixmap)
+{
+   struct _glxapi_table *t;
+   GET_DISPATCH(dpy, t);
+   if (!t)
+      return;
+   (t->DestroyPixmap)(dpy, pixmap);
+}
+
+
+void PUBLIC
+glXDestroyWindow(Display *dpy, GLXWindow window)
+{
+   struct _glxapi_table *t;
+   GET_DISPATCH(dpy, t);
+   if (!t)
+      return;
+   (t->DestroyWindow)(dpy, window);
+}
+
+
+GLXDrawable PUBLIC
+glXGetCurrentReadDrawable(void)
+{
+   __GLXcontext *gc = (__GLXcontext *) glXGetCurrentContext();
+   return gc ? gc->currentReadable : 0;
+}
+
+
+int PUBLIC
+glXGetFBConfigAttrib(Display *dpy, GLXFBConfig config, int attribute, int *value)
+{
+   struct _glxapi_table *t;
+   GET_DISPATCH(dpy, t);
+   if (!t)
+      return GLX_NO_EXTENSION;
+   return (t->GetFBConfigAttrib)(dpy, config, attribute, value);
+}
+
+
+GLXFBConfig PUBLIC *
+glXGetFBConfigs(Display *dpy, int screen, int *nelements)
+{
+   struct _glxapi_table *t;
+   GET_DISPATCH(dpy, t);
+   if (!t)
+      return 0;
+   return (t->GetFBConfigs)(dpy, screen, nelements);
+}
+
+void PUBLIC
+glXGetSelectedEvent(Display *dpy, GLXDrawable drawable, unsigned long *mask)
+{
+   struct _glxapi_table *t;
+   GET_DISPATCH(dpy, t);
+   if (!t)
+      return;
+   (t->GetSelectedEvent)(dpy, drawable, mask);
+}
+
+
+XVisualInfo PUBLIC *
+glXGetVisualFromFBConfig(Display *dpy, GLXFBConfig config)
+{
+   struct _glxapi_table *t;
+   GET_DISPATCH(dpy, t);
+   if (!t)
+      return NULL;
+   return (t->GetVisualFromFBConfig)(dpy, config);
+}
+
+
+Bool PUBLIC
+glXMakeContextCurrent(Display *dpy, GLXDrawable draw, GLXDrawable read, GLXContext ctx)
+{
+   Bool b;
+   struct _glxapi_table *t;
+   GET_DISPATCH(dpy, t);
+   if (!t)
+      return False;
+   b = (t->MakeContextCurrent)(dpy, draw, read, ctx);
+   if (b) {
+      SetCurrentContext(ctx);
+   }
+   return b;
+}
+
+
+int PUBLIC
+glXQueryContext(Display *dpy, GLXContext ctx, int attribute, int *value)
+{
+   struct _glxapi_table *t;
+   GET_DISPATCH(dpy, t);
+   assert(t);
+   if (!t)
+      return 0; /* XXX correct? */
+   return (t->QueryContext)(dpy, ctx, attribute, value);
+}
+
+
+void PUBLIC
+glXQueryDrawable(Display *dpy, GLXDrawable draw, int attribute, unsigned int *value)
+{
+   struct _glxapi_table *t;
+   GET_DISPATCH(dpy, t);
+   if (!t)
+      return;
+   (t->QueryDrawable)(dpy, draw, attribute, value);
+}
+
+
+void PUBLIC
+glXSelectEvent(Display *dpy, GLXDrawable drawable, unsigned long mask)
+{
+   struct _glxapi_table *t;
+   GET_DISPATCH(dpy, t);
+   if (!t)
+      return;
+   (t->SelectEvent)(dpy, drawable, mask);
+}
+
+
+
+/*** GLX_SGI_swap_control ***/
+
+int PUBLIC
+glXSwapIntervalSGI(int interval)
+{
+   struct _glxapi_table *t;
+   Display *dpy = glXGetCurrentDisplay();
+   GET_DISPATCH(dpy, t);
+   if (!t)
+      return 0;
+   return (t->SwapIntervalSGI)(interval);
+}
+
+
+
+/*** GLX_SGI_video_sync ***/
+
+int PUBLIC
+glXGetVideoSyncSGI(unsigned int *count)
+{
+   struct _glxapi_table *t;
+   Display *dpy = glXGetCurrentDisplay();
+   GET_DISPATCH(dpy, t);
+   if (!t || !glXGetCurrentContext())
+      return GLX_BAD_CONTEXT;
+   return (t->GetVideoSyncSGI)(count);
+}
+
+int PUBLIC
+glXWaitVideoSyncSGI(int divisor, int remainder, unsigned int *count)
+{
+   struct _glxapi_table *t;
+   Display *dpy = glXGetCurrentDisplay();
+   GET_DISPATCH(dpy, t);
+   if (!t || !glXGetCurrentContext())
+      return GLX_BAD_CONTEXT;
+   return (t->WaitVideoSyncSGI)(divisor, remainder, count);
+}
+
+
+
+/*** GLX_SGI_make_current_read ***/
+
+Bool PUBLIC
+glXMakeCurrentReadSGI(Display *dpy, GLXDrawable draw, GLXDrawable read, GLXContext ctx)
+{
+   struct _glxapi_table *t;
+   GET_DISPATCH(dpy, t);
+   if (!t)
+      return False;
+   return (t->MakeCurrentReadSGI)(dpy, draw, read, ctx);
+}
+
+GLXDrawable PUBLIC
+glXGetCurrentReadDrawableSGI(void)
+{
+   return glXGetCurrentReadDrawable();
+}
+
+
+#if defined(_VL_H)
+
+GLXVideoSourceSGIX PUBLIC
+glXCreateGLXVideoSourceSGIX(Display *dpy, int screen, VLServer server, VLPath path, int nodeClass, VLNode drainNode)
+{
+   struct _glxapi_table *t;
+   GET_DISPATCH(dpy, t);
+   if (!t)
+      return 0;
+   return (t->CreateGLXVideoSourceSGIX)(dpy, screen, server, path, nodeClass, drainNode);
+}
+
+void PUBLIC
+glXDestroyGLXVideoSourceSGIX(Display *dpy, GLXVideoSourceSGIX src)
+{
+   struct _glxapi_table *t;
+   GET_DISPATCH(dpy, t);
+   if (!t)
+      return 0;
+   return (t->DestroyGLXVideoSourceSGIX)(dpy, src);
+}
+
+#endif
+
+
+/*** GLX_EXT_import_context ***/
+
+void PUBLIC
+glXFreeContextEXT(Display *dpy, GLXContext context)
+{
+   struct _glxapi_table *t;
+   GET_DISPATCH(dpy, t);
+   if (!t)
+      return;
+   (t->FreeContextEXT)(dpy, context);
+}
+
+GLXContextID PUBLIC
+glXGetContextIDEXT(const GLXContext context)
+{
+   return ((__GLXcontext *) context)->xid;
+}
+
+Display PUBLIC *
+glXGetCurrentDisplayEXT(void)
+{
+   return glXGetCurrentDisplay();
+}
+
+GLXContext PUBLIC
+glXImportContextEXT(Display *dpy, GLXContextID contextID)
+{
+   struct _glxapi_table *t;
+   GET_DISPATCH(dpy, t);
+   if (!t)
+      return 0;
+   return (t->ImportContextEXT)(dpy, contextID);
+}
+
+int PUBLIC
+glXQueryContextInfoEXT(Display *dpy, GLXContext context, int attribute,int *value)
+{
+   struct _glxapi_table *t;
+   GET_DISPATCH(dpy, t);
+   if (!t)
+      return 0;  /* XXX ok? */
+   return (t->QueryContextInfoEXT)(dpy, context, attribute, value);
+}
+
+
+
+/*** GLX_SGIX_fbconfig ***/
+
+int PUBLIC
+glXGetFBConfigAttribSGIX(Display *dpy, GLXFBConfigSGIX config, int attribute, int *value)
+{
+   struct _glxapi_table *t;
+   GET_DISPATCH(dpy, t);
+   if (!t)
+      return 0;
+   return (t->GetFBConfigAttribSGIX)(dpy, config, attribute, value);
+}
+
+GLXFBConfigSGIX PUBLIC *
+glXChooseFBConfigSGIX(Display *dpy, int screen, int *attrib_list, int *nelements)
+{
+   struct _glxapi_table *t;
+   GET_DISPATCH(dpy, t);
+   if (!t)
+      return 0;
+   return (t->ChooseFBConfigSGIX)(dpy, screen, attrib_list, nelements);
+}
+
+GLXPixmap PUBLIC
+glXCreateGLXPixmapWithConfigSGIX(Display *dpy, GLXFBConfigSGIX config, Pixmap pixmap)
+{
+   struct _glxapi_table *t;
+   GET_DISPATCH(dpy, t);
+   if (!t)
+      return 0;
+   return (t->CreateGLXPixmapWithConfigSGIX)(dpy, config, pixmap);
+}
+
+GLXContext PUBLIC
+glXCreateContextWithConfigSGIX(Display *dpy, GLXFBConfigSGIX config, int render_type, GLXContext share_list, Bool direct)
+{
+   struct _glxapi_table *t;
+   GET_DISPATCH(dpy, t);
+   if (!t)
+      return 0;
+   return (t->CreateContextWithConfigSGIX)(dpy, config, render_type, share_list, direct);
+}
+
+XVisualInfo PUBLIC *
+glXGetVisualFromFBConfigSGIX(Display *dpy, GLXFBConfigSGIX config)
+{
+   struct _glxapi_table *t;
+   GET_DISPATCH(dpy, t);
+   if (!t)
+      return 0;
+   return (t->GetVisualFromFBConfigSGIX)(dpy, config);
+}
+
+GLXFBConfigSGIX PUBLIC
+glXGetFBConfigFromVisualSGIX(Display *dpy, XVisualInfo *vis)
+{
+   struct _glxapi_table *t;
+   GET_DISPATCH(dpy, t);
+   if (!t)
+      return 0;
+   return (t->GetFBConfigFromVisualSGIX)(dpy, vis);
+}
+
+
+
+/*** GLX_SGIX_pbuffer ***/
+
+GLXPbufferSGIX PUBLIC
+glXCreateGLXPbufferSGIX(Display *dpy, GLXFBConfigSGIX config, unsigned int width, unsigned int height, int *attrib_list)
+{
+   struct _glxapi_table *t;
+   GET_DISPATCH(dpy, t);
+   if (!t)
+      return 0;
+   return (t->CreateGLXPbufferSGIX)(dpy, config, width, height, attrib_list);
+}
+
+void PUBLIC
+glXDestroyGLXPbufferSGIX(Display *dpy, GLXPbufferSGIX pbuf)
+{
+   struct _glxapi_table *t;
+   GET_DISPATCH(dpy, t);
+   if (!t)
+      return;
+   (t->DestroyGLXPbufferSGIX)(dpy, pbuf);
+}
+
+int PUBLIC
+glXQueryGLXPbufferSGIX(Display *dpy, GLXPbufferSGIX pbuf, int attribute, unsigned int *value)
+{
+   struct _glxapi_table *t;
+   GET_DISPATCH(dpy, t);
+   if (!t)
+      return 0;
+   return (t->QueryGLXPbufferSGIX)(dpy, pbuf, attribute, value);
+}
+
+void PUBLIC
+glXSelectEventSGIX(Display *dpy, GLXDrawable drawable, unsigned long mask)
+{
+   struct _glxapi_table *t;
+   GET_DISPATCH(dpy, t);
+   if (!t)
+      return;
+   (t->SelectEventSGIX)(dpy, drawable, mask);
+}
+
+void PUBLIC
+glXGetSelectedEventSGIX(Display *dpy, GLXDrawable drawable, unsigned long *mask)
+{
+   struct _glxapi_table *t;
+   GET_DISPATCH(dpy, t);
+   if (!t)
+      return;
+   (t->GetSelectedEventSGIX)(dpy, drawable, mask);
+}
+
+
+
+/*** GLX_SGI_cushion ***/
+
+void PUBLIC
+glXCushionSGI(Display *dpy, Window win, float cushion)
+{
+   struct _glxapi_table *t;
+   GET_DISPATCH(dpy, t);
+   if (!t)
+      return;
+   (t->CushionSGI)(dpy, win, cushion);
+}
+
+
+
+/*** GLX_SGIX_video_resize ***/
+
+int PUBLIC
+glXBindChannelToWindowSGIX(Display *dpy, int screen, int channel , Window window)
+{
+   struct _glxapi_table *t;
+   GET_DISPATCH(dpy, t);
+   if (!t)
+      return 0;
+   return (t->BindChannelToWindowSGIX)(dpy, screen, channel, window);
+}
+
+int PUBLIC
+glXChannelRectSGIX(Display *dpy, int screen, int channel, int x, int y, int w, int h)
+{
+   struct _glxapi_table *t;
+   GET_DISPATCH(dpy, t);
+   if (!t)
+      return 0;
+   return (t->ChannelRectSGIX)(dpy, screen, channel, x, y, w, h);
+}
+
+int PUBLIC
+glXQueryChannelRectSGIX(Display *dpy, int screen, int channel, int *x, int *y, int *w, int *h)
+{
+   struct _glxapi_table *t;
+   GET_DISPATCH(dpy, t);
+   if (!t)
+      return 0;
+   return (t->QueryChannelRectSGIX)(dpy, screen, channel, x, y, w, h);
+}
+
+int PUBLIC
+glXQueryChannelDeltasSGIX(Display *dpy, int screen, int channel, int *dx, int *dy, int *dw, int *dh)
+{
+   struct _glxapi_table *t;
+   GET_DISPATCH(dpy, t);
+   if (!t)
+      return 0;
+   return (t->QueryChannelDeltasSGIX)(dpy, screen, channel, dx, dy, dw, dh);
+}
+
+int PUBLIC
+glXChannelRectSyncSGIX(Display *dpy, int screen, int channel, GLenum synctype)
+{
+   struct _glxapi_table *t;
+   GET_DISPATCH(dpy, t);
+   if (!t)
+      return 0;
+   return (t->ChannelRectSyncSGIX)(dpy, screen, channel, synctype);
+}
+
+
+
+#if defined(_DM_BUFFER_H_)
+
+Bool PUBLIC
+glXAssociateDMPbufferSGIX(Display *dpy, GLXPbufferSGIX pbuffer, DMparams *params, DMbuffer dmbuffer)
+{
+   struct _glxapi_table *t;
+   GET_DISPATCH(dpy, t);
+   if (!t)
+      return False;
+   return (t->AssociateDMPbufferSGIX)(dpy, pbuffer, params, dmbuffer);
+}
+
+#endif
+
+
+/*** GLX_SGIX_swap_group ***/
+
+void PUBLIC
+glXJoinSwapGroupSGIX(Display *dpy, GLXDrawable drawable, GLXDrawable member)
+{
+   struct _glxapi_table *t;
+   GET_DISPATCH(dpy, t);
+   if (!t)
+      return;
+   (*t->JoinSwapGroupSGIX)(dpy, drawable, member);
+}
+
+
+/*** GLX_SGIX_swap_barrier ***/
+
+void PUBLIC
+glXBindSwapBarrierSGIX(Display *dpy, GLXDrawable drawable, int barrier)
+{
+   struct _glxapi_table *t;
+   GET_DISPATCH(dpy, t);
+   if (!t)
+      return;
+   (*t->BindSwapBarrierSGIX)(dpy, drawable, barrier);
+}
+
+Bool PUBLIC
+glXQueryMaxSwapBarriersSGIX(Display *dpy, int screen, int *max)
+{
+   struct _glxapi_table *t;
+   GET_DISPATCH(dpy, t);
+   if (!t)
+      return False;
+   return (*t->QueryMaxSwapBarriersSGIX)(dpy, screen, max);
+}
+
+
+
+/*** GLX_SUN_get_transparent_index ***/
+
+Status PUBLIC
+glXGetTransparentIndexSUN(Display *dpy, Window overlay, Window underlay, long *pTransparent)
+{
+   struct _glxapi_table *t;
+   GET_DISPATCH(dpy, t);
+   if (!t)
+      return False;
+   return (*t->GetTransparentIndexSUN)(dpy, overlay, underlay, pTransparent);
+}
+
+
+
+/*** GLX_MESA_copy_sub_buffer ***/
+
+void PUBLIC
+glXCopySubBufferMESA(Display *dpy, GLXDrawable drawable, int x, int y, int width, int height)
+{
+   struct _glxapi_table *t;
+   GET_DISPATCH(dpy, t);
+   if (!t)
+      return;
+   (t->CopySubBufferMESA)(dpy, drawable, x, y, width, height);
+}
+
+
+
+/*** GLX_MESA_release_buffers ***/
+
+Bool PUBLIC
+glXReleaseBuffersMESA(Display *dpy, Window w)
+{
+   struct _glxapi_table *t;
+   GET_DISPATCH(dpy, t);
+   if (!t)
+      return False;
+   return (t->ReleaseBuffersMESA)(dpy, w);
+}
+
+
+
+/*** GLX_MESA_pixmap_colormap ***/
+
+GLXPixmap PUBLIC
+glXCreateGLXPixmapMESA(Display *dpy, XVisualInfo *visinfo, Pixmap pixmap, Colormap cmap)
+{
+   struct _glxapi_table *t;
+   GET_DISPATCH(dpy, t);
+   if (!t)
+      return 0;
+   return (t->CreateGLXPixmapMESA)(dpy, visinfo, pixmap, cmap);
+}
+
+
+
+/*** GLX_MESA_set_3dfx_mode ***/
+
+Bool PUBLIC
+glXSet3DfxModeMESA(int mode)
+{
+   struct _glxapi_table *t;
+   Display *dpy = glXGetCurrentDisplay();
+   GET_DISPATCH(dpy, t);
+   if (!t)
+      return False;
+   return (t->Set3DfxModeMESA)(mode);
+}
+
+
+
+/*** GLX_NV_vertex_array_range ***/
+
+void PUBLIC *
+glXAllocateMemoryNV( GLsizei size,
+                     GLfloat readFrequency,
+                     GLfloat writeFrequency,
+                     GLfloat priority )
+{
+   struct _glxapi_table *t;
+   Display *dpy = glXGetCurrentDisplay();
+   GET_DISPATCH(dpy, t);
+   if (!t)
+      return NULL;
+   return (t->AllocateMemoryNV)(size, readFrequency, writeFrequency, priority);
+}
+
+
+void PUBLIC
+glXFreeMemoryNV( GLvoid *pointer )
+{
+   struct _glxapi_table *t;
+   Display *dpy = glXGetCurrentDisplay();
+   GET_DISPATCH(dpy, t);
+   if (!t)
+      return;
+   (t->FreeMemoryNV)(pointer);
+}
+
+
+
+
+/*** GLX_MESA_agp_offset */
+
+GLuint PUBLIC
+glXGetAGPOffsetMESA( const GLvoid *pointer )
+{
+   struct _glxapi_table *t;
+   Display *dpy = glXGetCurrentDisplay();
+   GET_DISPATCH(dpy, t);
+   if (!t)
+      return ~0;
+   return (t->GetAGPOffsetMESA)(pointer);
+}
+
+
+/*** GLX_MESA_allocate_memory */
+
+void *
+glXAllocateMemoryMESA(Display *dpy, int scrn, size_t size,
+                      float readfreq, float writefreq, float priority)
+{
+   /* dummy */
+   return NULL;
+}
+
+void
+glXFreeMemoryMESA(Display *dpy, int scrn, void *pointer)
+{
+   /* dummy */
+}
+
+
+GLuint
+glXGetMemoryOffsetMESA(Display *dpy, int scrn, const void *pointer)
+{
+   /* dummy */
+   return 0;
+}
+
+
+/*** GLX_EXT_texture_from_pixmap */
+
+void
+glXBindTexImageEXT(Display *dpy, GLXDrawable drawable, int buffer,
+                   const int *attrib_list)
+{
+   struct _glxapi_table *t;
+   GET_DISPATCH(dpy, t);
+   if (t)
+      t->BindTexImageEXT(dpy, drawable, buffer, attrib_list);
+}
+
+void
+glXReleaseTexImageEXT(Display *dpy, GLXDrawable drawable, int buffer)
+{
+   struct _glxapi_table *t;
+   GET_DISPATCH(dpy, t);
+   if (t)
+      t->ReleaseTexImageEXT(dpy, drawable, buffer);
+}
+
+
+/**********************************************************************/
+/* GLX API management functions                                       */
+/**********************************************************************/
+
+
+const char *
+_glxapi_get_version(void)
+{
+   return "1.3";
+}
+
+
+/*
+ * Return array of extension strings.
+ */
+const char **
+_glxapi_get_extensions(void)
+{
+   static const char *extensions[] = {
+#ifdef GLX_EXT_import_context
+      "GLX_EXT_import_context",
+#endif
+#ifdef GLX_SGI_video_sync
+      "GLX_SGI_video_sync",
+#endif
+#ifdef GLX_MESA_copy_sub_buffer
+      "GLX_MESA_copy_sub_buffer",
+#endif
+#ifdef GLX_MESA_release_buffers
+      "GLX_MESA_release_buffers",
+#endif
+#ifdef GLX_MESA_pixmap_colormap
+      "GLX_MESA_pixmap_colormap",
+#endif
+#ifdef GLX_MESA_set_3dfx_mode
+      "GLX_MESA_set_3dfx_mode",
+#endif
+#ifdef GLX_SGIX_fbconfig
+      "GLX_SGIX_fbconfig",
+#endif
+#ifdef GLX_SGIX_pbuffer
+      "GLX_SGIX_pbuffer",
+#endif
+#ifdef GLX_EXT_texture_from_pixmap
+      "GLX_EXT_texture_from_pixmap",
+#endif
+      NULL
+   };
+   return extensions;
+}
+
+
+/*
+ * Return size of the GLX dispatch table, in entries, not bytes.
+ */
+GLuint
+_glxapi_get_dispatch_table_size(void)
+{
+   return sizeof(struct _glxapi_table) / sizeof(void *);
+}
+
+
+static int
+generic_no_op_func(void)
+{
+   return 0;
+}
+
+
+/*
+ * Initialize all functions in given dispatch table to be no-ops
+ */
+void
+_glxapi_set_no_op_table(struct _glxapi_table *t)
+{
+   typedef int (*nop_func)(void);
+   nop_func *dispatch = (nop_func *) t;
+   GLuint n = _glxapi_get_dispatch_table_size();
+   GLuint i;
+   for (i = 0; i < n; i++) {
+      dispatch[i] = generic_no_op_func;
+   }
+}
+
+
+struct name_address_pair {
+   const char *Name;
+   __GLXextFuncPtr Address;
+};
+
+static struct name_address_pair GLX_functions[] = {
+   /*** GLX_VERSION_1_0 ***/
+   { "glXChooseVisual", (__GLXextFuncPtr) glXChooseVisual },
+   { "glXCopyContext", (__GLXextFuncPtr) glXCopyContext },
+   { "glXCreateContext", (__GLXextFuncPtr) glXCreateContext },
+   { "glXCreateGLXPixmap", (__GLXextFuncPtr) glXCreateGLXPixmap },
+   { "glXDestroyContext", (__GLXextFuncPtr) glXDestroyContext },
+   { "glXDestroyGLXPixmap", (__GLXextFuncPtr) glXDestroyGLXPixmap },
+   { "glXGetConfig", (__GLXextFuncPtr) glXGetConfig },
+   { "glXGetCurrentContext", (__GLXextFuncPtr) glXGetCurrentContext },
+   { "glXGetCurrentDrawable", (__GLXextFuncPtr) glXGetCurrentDrawable },
+   { "glXIsDirect", (__GLXextFuncPtr) glXIsDirect },
+   { "glXMakeCurrent", (__GLXextFuncPtr) glXMakeCurrent },
+   { "glXQueryExtension", (__GLXextFuncPtr) glXQueryExtension },
+   { "glXQueryVersion", (__GLXextFuncPtr) glXQueryVersion },
+   { "glXSwapBuffers", (__GLXextFuncPtr) glXSwapBuffers },
+   { "glXUseXFont", (__GLXextFuncPtr) glXUseXFont },
+   { "glXWaitGL", (__GLXextFuncPtr) glXWaitGL },
+   { "glXWaitX", (__GLXextFuncPtr) glXWaitX },
+
+   /*** GLX_VERSION_1_1 ***/
+   { "glXGetClientString", (__GLXextFuncPtr) glXGetClientString },
+   { "glXQueryExtensionsString", (__GLXextFuncPtr) glXQueryExtensionsString },
+   { "glXQueryServerString", (__GLXextFuncPtr) glXQueryServerString },
+
+   /*** GLX_VERSION_1_2 ***/
+   { "glXGetCurrentDisplay", (__GLXextFuncPtr) glXGetCurrentDisplay },
+
+   /*** GLX_VERSION_1_3 ***/
+   { "glXChooseFBConfig", (__GLXextFuncPtr) glXChooseFBConfig },
+   { "glXCreateNewContext", (__GLXextFuncPtr) glXCreateNewContext },
+   { "glXCreatePbuffer", (__GLXextFuncPtr) glXCreatePbuffer },
+   { "glXCreatePixmap", (__GLXextFuncPtr) glXCreatePixmap },
+   { "glXCreateWindow", (__GLXextFuncPtr) glXCreateWindow },
+   { "glXDestroyPbuffer", (__GLXextFuncPtr) glXDestroyPbuffer },
+   { "glXDestroyPixmap", (__GLXextFuncPtr) glXDestroyPixmap },
+   { "glXDestroyWindow", (__GLXextFuncPtr) glXDestroyWindow },
+   { "glXGetCurrentReadDrawable", (__GLXextFuncPtr) glXGetCurrentReadDrawable },
+   { "glXGetFBConfigAttrib", (__GLXextFuncPtr) glXGetFBConfigAttrib },
+   { "glXGetFBConfigs", (__GLXextFuncPtr) glXGetFBConfigs },
+   { "glXGetSelectedEvent", (__GLXextFuncPtr) glXGetSelectedEvent },
+   { "glXGetVisualFromFBConfig", (__GLXextFuncPtr) glXGetVisualFromFBConfig },
+   { "glXMakeContextCurrent", (__GLXextFuncPtr) glXMakeContextCurrent },
+   { "glXQueryContext", (__GLXextFuncPtr) glXQueryContext },
+   { "glXQueryDrawable", (__GLXextFuncPtr) glXQueryDrawable },
+   { "glXSelectEvent", (__GLXextFuncPtr) glXSelectEvent },
+
+   /*** GLX_VERSION_1_4 ***/
+   { "glXGetProcAddress", (__GLXextFuncPtr) glXGetProcAddress },
+
+   /*** GLX_SGI_swap_control ***/
+   { "glXSwapIntervalSGI", (__GLXextFuncPtr) glXSwapIntervalSGI },
+
+   /*** GLX_SGI_video_sync ***/
+   { "glXGetVideoSyncSGI", (__GLXextFuncPtr) glXGetVideoSyncSGI },
+   { "glXWaitVideoSyncSGI", (__GLXextFuncPtr) glXWaitVideoSyncSGI },
+
+   /*** GLX_SGI_make_current_read ***/
+   { "glXMakeCurrentReadSGI", (__GLXextFuncPtr) glXMakeCurrentReadSGI },
+   { "glXGetCurrentReadDrawableSGI", (__GLXextFuncPtr) glXGetCurrentReadDrawableSGI },
+
+   /*** GLX_SGIX_video_source ***/
+#if defined(_VL_H)
+   { "glXCreateGLXVideoSourceSGIX", (__GLXextFuncPtr) glXCreateGLXVideoSourceSGIX },
+   { "glXDestroyGLXVideoSourceSGIX", (__GLXextFuncPtr) glXDestroyGLXVideoSourceSGIX },
+#endif
+
+   /*** GLX_EXT_import_context ***/
+   { "glXFreeContextEXT", (__GLXextFuncPtr) glXFreeContextEXT },
+   { "glXGetContextIDEXT", (__GLXextFuncPtr) glXGetContextIDEXT },
+   { "glXGetCurrentDisplayEXT", (__GLXextFuncPtr) glXGetCurrentDisplayEXT },
+   { "glXImportContextEXT", (__GLXextFuncPtr) glXImportContextEXT },
+   { "glXQueryContextInfoEXT", (__GLXextFuncPtr) glXQueryContextInfoEXT },
+
+   /*** GLX_SGIX_fbconfig ***/
+   { "glXGetFBConfigAttribSGIX", (__GLXextFuncPtr) glXGetFBConfigAttribSGIX },
+   { "glXChooseFBConfigSGIX", (__GLXextFuncPtr) glXChooseFBConfigSGIX },
+   { "glXCreateGLXPixmapWithConfigSGIX", (__GLXextFuncPtr) glXCreateGLXPixmapWithConfigSGIX },
+   { "glXCreateContextWithConfigSGIX", (__GLXextFuncPtr) glXCreateContextWithConfigSGIX },
+   { "glXGetVisualFromFBConfigSGIX", (__GLXextFuncPtr) glXGetVisualFromFBConfigSGIX },
+   { "glXGetFBConfigFromVisualSGIX", (__GLXextFuncPtr) glXGetFBConfigFromVisualSGIX },
+
+   /*** GLX_SGIX_pbuffer ***/
+   { "glXCreateGLXPbufferSGIX", (__GLXextFuncPtr) glXCreateGLXPbufferSGIX },
+   { "glXDestroyGLXPbufferSGIX", (__GLXextFuncPtr) glXDestroyGLXPbufferSGIX },
+   { "glXQueryGLXPbufferSGIX", (__GLXextFuncPtr) glXQueryGLXPbufferSGIX },
+   { "glXSelectEventSGIX", (__GLXextFuncPtr) glXSelectEventSGIX },
+   { "glXGetSelectedEventSGIX", (__GLXextFuncPtr) glXGetSelectedEventSGIX },
+
+   /*** GLX_SGI_cushion ***/
+   { "glXCushionSGI", (__GLXextFuncPtr) glXCushionSGI },
+
+   /*** GLX_SGIX_video_resize ***/
+   { "glXBindChannelToWindowSGIX", (__GLXextFuncPtr) glXBindChannelToWindowSGIX },
+   { "glXChannelRectSGIX", (__GLXextFuncPtr) glXChannelRectSGIX },
+   { "glXQueryChannelRectSGIX", (__GLXextFuncPtr) glXQueryChannelRectSGIX },
+   { "glXQueryChannelDeltasSGIX", (__GLXextFuncPtr) glXQueryChannelDeltasSGIX },
+   { "glXChannelRectSyncSGIX", (__GLXextFuncPtr) glXChannelRectSyncSGIX },
+
+   /*** GLX_SGIX_dmbuffer **/
+#if defined(_DM_BUFFER_H_)
+   { "glXAssociateDMPbufferSGIX", (__GLXextFuncPtr) glXAssociateDMPbufferSGIX },
+#endif
+
+   /*** GLX_SGIX_swap_group ***/
+   { "glXJoinSwapGroupSGIX", (__GLXextFuncPtr) glXJoinSwapGroupSGIX },
+
+   /*** GLX_SGIX_swap_barrier ***/
+   { "glXBindSwapBarrierSGIX", (__GLXextFuncPtr) glXBindSwapBarrierSGIX },
+   { "glXQueryMaxSwapBarriersSGIX", (__GLXextFuncPtr) glXQueryMaxSwapBarriersSGIX },
+
+   /*** GLX_SUN_get_transparent_index ***/
+   { "glXGetTransparentIndexSUN", (__GLXextFuncPtr) glXGetTransparentIndexSUN },
+
+   /*** GLX_MESA_copy_sub_buffer ***/
+   { "glXCopySubBufferMESA", (__GLXextFuncPtr) glXCopySubBufferMESA },
+
+   /*** GLX_MESA_pixmap_colormap ***/
+   { "glXCreateGLXPixmapMESA", (__GLXextFuncPtr) glXCreateGLXPixmapMESA },
+
+   /*** GLX_MESA_release_buffers ***/
+   { "glXReleaseBuffersMESA", (__GLXextFuncPtr) glXReleaseBuffersMESA },
+
+   /*** GLX_MESA_set_3dfx_mode ***/
+   { "glXSet3DfxModeMESA", (__GLXextFuncPtr) glXSet3DfxModeMESA },
+
+   /*** GLX_ARB_get_proc_address ***/
+   { "glXGetProcAddressARB", (__GLXextFuncPtr) glXGetProcAddressARB },
+
+   /*** GLX_NV_vertex_array_range ***/
+   { "glXAllocateMemoryNV", (__GLXextFuncPtr) glXAllocateMemoryNV },
+   { "glXFreeMemoryNV", (__GLXextFuncPtr) glXFreeMemoryNV },
+
+   /*** GLX_MESA_agp_offset ***/
+   { "glXGetAGPOffsetMESA", (__GLXextFuncPtr) glXGetAGPOffsetMESA },
+
+   /*** GLX_MESA_allocate_memory ***/
+   { "glXAllocateMemoryMESA", (__GLXextFuncPtr) glXAllocateMemoryMESA },
+   { "glXFreeMemoryMESA", (__GLXextFuncPtr) glXFreeMemoryMESA },
+   { "glXGetMemoryOffsetMESA", (__GLXextFuncPtr) glXGetMemoryOffsetMESA },
+
+   /*** GLX_EXT_texture_from_pixmap ***/
+   { "glXBindTexImageEXT", (__GLXextFuncPtr) glXBindTexImageEXT },
+   { "glXReleaseTexImageEXT", (__GLXextFuncPtr) glXReleaseTexImageEXT },
+
+   { NULL, NULL }   /* end of list */
+};
+
+
+
+/*
+ * Return address of named glX function, or NULL if not found.
+ */
+__GLXextFuncPtr
+_glxapi_get_proc_address(const char *funcName)
+{
+   GLuint i;
+   for (i = 0; GLX_functions[i].Name; i++) {
+      if (strcmp(GLX_functions[i].Name, funcName) == 0)
+         return GLX_functions[i].Address;
+   }
+   return NULL;
+}
+
+
+
+/*
+ * This function does not get dispatched through the dispatch table
+ * since it's really a "meta" function.
+ */
+__GLXextFuncPtr
+glXGetProcAddressARB(const GLubyte *procName)
+{
+   __GLXextFuncPtr f;
+
+   f = _glxapi_get_proc_address((const char *) procName);
+   if (f) {
+      return f;
+   }
+
+   f = (__GLXextFuncPtr) _glapi_get_proc_address((const char *) procName);
+   return f;
+}
+
+
+/* GLX 1.4 */
+void (*glXGetProcAddress(const GLubyte *procName))()
+{
+   return glXGetProcAddressARB(procName);
+}
diff --git a/src/gallium/winsys/xlib/glxapi.h b/src/gallium/winsys/xlib/glxapi.h
new file mode 100644
index 0000000000..37de81e55a
--- /dev/null
+++ b/src/gallium/winsys/xlib/glxapi.h
@@ -0,0 +1,228 @@
+/*
+ * Mesa 3-D graphics library
+ * Version:  6.3
+ * 
+ * Copyright (C) 1999-2004  Brian Paul   All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ * 
+ * The above copyright notice and this permission notice shall be included
+ * in all copies or substantial portions of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * BRIAN PAUL BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN
+ * AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ */
+
+
+#ifndef _glxapi_h_
+#define _glxapi_h_
+
+
+#define GLX_GLXEXT_PROTOTYPES
+#include "GL/glx.h"
+
+
+/* The GLX API dispatcher (i.e. this code) is being built into stand-alone
+ * Mesa.  We don't know anything about XFree86 or real GLX so we define a
+ * minimal __GLXContextRec here so some of the functions in this file can
+ * work properly.
+ */
+typedef struct __GLXcontextRec {
+   Display *currentDpy;
+   GLboolean isDirect;
+   GLXDrawable currentDrawable;
+   GLXDrawable currentReadable;
+   XID xid;
+} __GLXcontext;
+
+
+/*
+ * Almost all the GLX API functions get routed through this dispatch table.
+ * The exceptions are the glXGetCurrentXXX() functions.
+ *
+ * This dispatch table allows multiple GLX client-side modules to coexist.
+ * Specifically, a real GLX library (like SGI's or the Utah GLX) and Mesa's
+ * pseudo-GLX can be present at the same time.  The former being used on
+ * GLX-enabled X servers and the later on non-GLX X servers.
+ *
+ * Red Hat has been using this since Red Hat Linux 7.0 (I think).
+ * This'll be a standard feature in XFree86 4.3.  It basically allows one
+ * libGL to do both DRI-rendering and "fake GLX" rendering to X displays
+ * that lack the GLX extension.
+ */
+struct _glxapi_table {
+   /*** GLX_VERSION_1_0 ***/
+   XVisualInfo *(*ChooseVisual)(Display *dpy, int screen, int *list);
+   void (*CopyContext)(Display *dpy, GLXContext src, GLXContext dst, unsigned long mask);
+   GLXContext (*CreateContext)(Display *dpy, XVisualInfo *visinfo, GLXContext shareList, Bool direct);
+   GLXPixmap (*CreateGLXPixmap)(Display *dpy, XVisualInfo *visinfo, Pixmap pixmap);
+   void (*DestroyContext)(Display *dpy, GLXContext ctx);
+   void (*DestroyGLXPixmap)(Display *dpy, GLXPixmap pixmap);
+   int (*GetConfig)(Display *dpy, XVisualInfo *visinfo, int attrib, int *value);
+   /*GLXContext (*GetCurrentContext)(void);*/
+   /*GLXDrawable (*GetCurrentDrawable)(void);*/
+   Bool (*IsDirect)(Display *dpy, GLXContext ctx);
+   Bool (*MakeCurrent)(Display *dpy, GLXDrawable drawable, GLXContext ctx);
+   Bool (*QueryExtension)(Display *dpy, int *errorb, int *event);
+   Bool (*QueryVersion)(Display *dpy, int *maj, int *min);
+   void (*SwapBuffers)(Display *dpy, GLXDrawable drawable);
+   void (*UseXFont)(Font font, int first, int count, int listBase);
+   void (*WaitGL)(void);
+   void (*WaitX)(void);
+
+   /*** GLX_VERSION_1_1 ***/
+   const char *(*GetClientString)(Display *dpy, int name);
+   const char *(*QueryExtensionsString)(Display *dpy, int screen);
+   const char *(*QueryServerString)(Display *dpy, int screen, int name);
+
+   /*** GLX_VERSION_1_2 ***/
+   /*Display *(*GetCurrentDisplay)(void);*/
+
+   /*** GLX_VERSION_1_3 ***/
+   GLXFBConfig *(*ChooseFBConfig)(Display *dpy, int screen, const int *attribList, int *nitems);
+   GLXContext (*CreateNewContext)(Display *dpy, GLXFBConfig config, int renderType, GLXContext shareList, Bool direct);
+   GLXPbuffer (*CreatePbuffer)(Display *dpy, GLXFBConfig config, const int *attribList);
+   GLXPixmap (*CreatePixmap)(Display *dpy, GLXFBConfig config, Pixmap pixmap, const int *attribList);
+   GLXWindow (*CreateWindow)(Display *dpy, GLXFBConfig config, Window win, const int *attribList);
+   void (*DestroyPbuffer)(Display *dpy, GLXPbuffer pbuf);
+   void (*DestroyPixmap)(Display *dpy, GLXPixmap pixmap);
+   void (*DestroyWindow)(Display *dpy, GLXWindow window);
+   /*GLXDrawable (*GetCurrentReadDrawable)(void);*/
+   int (*GetFBConfigAttrib)(Display *dpy, GLXFBConfig config, int attribute, int *value);
+   GLXFBConfig *(*GetFBConfigs)(Display *dpy, int screen, int *nelements);
+   void (*GetSelectedEvent)(Display *dpy, GLXDrawable drawable, unsigned long *mask);
+   XVisualInfo *(*GetVisualFromFBConfig)(Display *dpy, GLXFBConfig config);
+   Bool (*MakeContextCurrent)(Display *dpy, GLXDrawable draw, GLXDrawable read, GLXContext ctx);
+   int (*QueryContext)(Display *dpy, GLXContext ctx, int attribute, int *value);
+   void (*QueryDrawable)(Display *dpy, GLXDrawable draw, int attribute, unsigned int *value);
+   void (*SelectEvent)(Display *dpy, GLXDrawable drawable, unsigned long mask);
+
+   /*** GLX_SGI_swap_control ***/
+   int (*SwapIntervalSGI)(int);
+
+   /*** GLX_SGI_video_sync ***/
+   int (*GetVideoSyncSGI)(unsigned int *count);
+   int (*WaitVideoSyncSGI)(int divisor, int remainder, unsigned int *count);
+
+   /*** GLX_SGI_make_current_read ***/
+   Bool (*MakeCurrentReadSGI)(Display *, GLXDrawable, GLXDrawable, GLXContext);
+   /*GLXDrawable (*GetCurrentReadDrawableSGI)(void);*/
+
+   /*** GLX_SGIX_video_source (needs video library) ***/
+#if defined(_VL_H_)
+   GLXVideoSourceSGIX (*CreateGLXVideoSourceSGIX)(Display *, int, VLServer, VLPath, int, VLNode);
+   void (*DestroyGLXVideoSourceSGIX)(Display *, GLXVideoSourceSGIX);
+#else
+   void *CreateGLXVideoSourceSGIX;
+   void *DestroyGLXVideoSourceSGIX;
+#endif
+
+   /*** GLX_EXT_import_context ***/
+   void (*FreeContextEXT)(Display *dpy, GLXContext context);
+   GLXContextID (*GetContextIDEXT)(const GLXContext context);
+   /*Display *(*GetCurrentDisplayEXT)(void);*/
+   GLXContext (*ImportContextEXT)(Display *dpy, GLXContextID contextID);
+   int (*QueryContextInfoEXT)(Display *dpy, GLXContext context, int attribute,int *value);
+
+   /*** GLX_SGIX_fbconfig ***/
+   int (*GetFBConfigAttribSGIX)(Display *, GLXFBConfigSGIX, int, int *);
+   GLXFBConfigSGIX * (*ChooseFBConfigSGIX)(Display *, int, int *, int *);
+   GLXPixmap (*CreateGLXPixmapWithConfigSGIX)(Display *, GLXFBConfigSGIX, Pixmap);
+   GLXContext (*CreateContextWithConfigSGIX)(Display *, GLXFBConfigSGIX, int, GLXContext, Bool);
+   XVisualInfo * (*GetVisualFromFBConfigSGIX)(Display *, GLXFBConfigSGIX);
+   GLXFBConfigSGIX (*GetFBConfigFromVisualSGIX)(Display *, XVisualInfo *);
+
+   /*** GLX_SGIX_pbuffer ***/
+   GLXPbufferSGIX (*CreateGLXPbufferSGIX)(Display *, GLXFBConfigSGIX, unsigned int, unsigned int, int *);
+   void (*DestroyGLXPbufferSGIX)(Display *, GLXPbufferSGIX);
+   int (*QueryGLXPbufferSGIX)(Display *, GLXPbufferSGIX, int, unsigned int *);
+   void (*SelectEventSGIX)(Display *, GLXDrawable, unsigned long);
+   void (*GetSelectedEventSGIX)(Display *, GLXDrawable, unsigned long *);
+
+   /*** GLX_SGI_cushion ***/
+   void (*CushionSGI)(Display *, Window, float);
+
+   /*** GLX_SGIX_video_resize ***/
+   int (*BindChannelToWindowSGIX)(Display *, int, int, Window);
+   int (*ChannelRectSGIX)(Display *, int, int, int, int, int, int);
+   int (*QueryChannelRectSGIX)(Display *, int, int, int *, int *, int *, int *);
+   int (*QueryChannelDeltasSGIX)(Display *, int, int, int *, int *, int *, int *);
+   int (*ChannelRectSyncSGIX)(Display *, int, int, GLenum);
+
+   /*** GLX_SGIX_dmbuffer (needs dmedia library) ***/
+#if defined (_DM_BUFFER_H_)
+   Bool (*AssociateDMPbufferSGIX)(Display *, GLXPbufferSGIX, DMparams *, DMbuffer);
+#else
+   void *AssociciateDMPbufferSGIX;
+#endif
+
+   /*** GLX_SGIX_swap_group ***/
+   void (*JoinSwapGroupSGIX)(Display *, GLXDrawable, GLXDrawable);
+
+   /*** GLX_SGIX_swap_barrier ***/
+   void (*BindSwapBarrierSGIX)(Display *, GLXDrawable, int);
+   Bool (*QueryMaxSwapBarriersSGIX)(Display *, int, int *);
+
+   /*** GLX_SUN_get_transparent_index ***/
+   Status (*GetTransparentIndexSUN)(Display *, Window, Window, long *);
+
+   /*** GLX_MESA_copy_sub_buffer ***/
+   void (*CopySubBufferMESA)(Display *dpy, GLXDrawable drawable, int x, int y, int width, int height);
+
+   /*** GLX_MESA_release_buffers ***/
+   Bool (*ReleaseBuffersMESA)(Display *dpy, Window w);
+
+   /*** GLX_MESA_pixmap_colormap ***/
+   GLXPixmap (*CreateGLXPixmapMESA)(Display *dpy, XVisualInfo *visinfo, Pixmap pixmap, Colormap cmap);
+
+   /*** GLX_MESA_set_3dfx_mode ***/
+   Bool (*Set3DfxModeMESA)(int mode);
+
+   /*** GLX_NV_vertex_array_range ***/
+   void * (*AllocateMemoryNV)( GLsizei size,
+                               GLfloat readFrequency,
+                               GLfloat writeFrequency,
+                               GLfloat priority );
+   void (*FreeMemoryNV)( GLvoid *pointer );
+
+   /*** GLX_MESA_agp_offset ***/
+   GLuint (*GetAGPOffsetMESA)( const GLvoid *pointer );
+
+   /*** GLX_EXT_texture_from_pixmap ***/
+   void (*BindTexImageEXT)(Display *dpy, GLXDrawable drawable, int buffer,
+                           const int *attrib_list);
+   void (*ReleaseTexImageEXT)(Display *dpy, GLXDrawable drawable, int buffer);
+};
+
+
+
+extern const char *
+_glxapi_get_version(void);
+
+
+extern const char **
+_glxapi_get_extensions(void);
+
+
+extern GLuint
+_glxapi_get_dispatch_table_size(void);
+
+
+extern void
+_glxapi_set_no_op_table(struct _glxapi_table *t);
+
+
+extern __GLXextFuncPtr
+_glxapi_get_proc_address(const char *funcName);
+
+
+#endif
diff --git a/src/gallium/winsys/xlib/glxheader.h b/src/gallium/winsys/xlib/glxheader.h
new file mode 100644
index 0000000000..a402191f13
--- /dev/null
+++ b/src/gallium/winsys/xlib/glxheader.h
@@ -0,0 +1,62 @@
+/*
+ * Mesa 3-D graphics library
+ * Version:  6.5.1
+ * 
+ * Copyright (C) 1999-2006  Brian Paul   All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ * 
+ * The above copyright notice and this permission notice shall be included
+ * in all copies or substantial portions of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * BRIAN PAUL BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN
+ * AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ */
+
+
+#ifndef GLX_HEADER_H
+#define GLX_HEADER_H
+
+#ifdef __VMS
+#include <GL/vms_x_fix.h>
+#endif
+
+#include "glheader.h"
+
+#ifdef XFree86Server
+
+# include "resource.h"
+# include "windowstr.h"
+
+#else
+
+# include <X11/Xlib.h>
+# include <X11/Xlibint.h>
+# include <X11/Xutil.h>
+# ifdef USE_XSHM  /* was SHM */
+#  include <sys/ipc.h>
+#  include <sys/shm.h>
+#  include <X11/extensions/XShm.h>
+# endif
+# include <GL/glx.h>
+# include <sys/time.h>
+
+#endif
+
+
+
+/* this silences a compiler warning on several systems */
+struct timespec;
+struct itimerspec;
+
+
+#endif /*GLX_HEADER*/
diff --git a/src/gallium/winsys/xlib/realglx.c b/src/gallium/winsys/xlib/realglx.c
new file mode 100644
index 0000000000..30adb7465b
--- /dev/null
+++ b/src/gallium/winsys/xlib/realglx.c
@@ -0,0 +1,180 @@
+
+/*
+ * Mesa 3-D graphics library
+ * Version:  5.1
+ * 
+ * Copyright (C) 1999-2002  Brian Paul   All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ * 
+ * The above copyright notice and this permission notice shall be included
+ * in all copies or substantial portions of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * BRIAN PAUL BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN
+ * AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ */
+
+
+#include <assert.h>
+#include <GL/glx.h>
+#include "realglx.h"
+#include "glxapi.h"
+
+
+struct _glxapi_table *
+_real_GetGLXDispatchTable(void)
+{
+   static struct _glxapi_table glx;
+
+   /* be sure our dispatch table size <= libGL's table */
+   {
+      GLuint size = sizeof(struct _glxapi_table) / sizeof(void *);
+      (void) size;
+      assert(_glxapi_get_dispatch_table_size() >= size);
+   }
+
+   /* initialize the whole table to no-ops */
+   _glxapi_set_no_op_table(&glx);
+
+   /* now initialize the table with the functions I implement */
+
+   /*** GLX_VERSION_1_0 ***/
+   glx.ChooseVisual = _real_glXChooseVisual;
+   glx.CopyContext = _real_glXCopyContext;
+   glx.CreateContext = _real_glXCreateContext;
+   glx.CreateGLXPixmap = _real_glXCreateGLXPixmap;
+   glx.DestroyContext = _real_glXDestroyContext;
+   glx.DestroyGLXPixmap = _real_glXDestroyGLXPixmap;
+   glx.GetConfig = _real_glXGetConfig;
+   /*glx.GetCurrentContext = _real_glXGetCurrentContext;*/
+   /*glx.GetCurrentDrawable = _real_glXGetCurrentDrawable;*/
+   glx.IsDirect = _real_glXIsDirect;
+   glx.MakeCurrent = _real_glXMakeCurrent;
+   glx.QueryExtension = _real_glXQueryExtension;
+   glx.QueryVersion = _real_glXQueryVersion;
+   glx.SwapBuffers = _real_glXSwapBuffers;
+   glx.UseXFont = _real_glXUseXFont;
+   glx.WaitGL = _real_glXWaitGL;
+   glx.WaitX = _real_glXWaitX;
+
+   /*** GLX_VERSION_1_1 ***/
+   glx.GetClientString = _real_glXGetClientString;
+   glx.QueryExtensionsString = _real_glXQueryExtensionsString;
+   glx.QueryServerString = _real_glXQueryServerString;
+
+   /*** GLX_VERSION_1_2 ***/
+   /*glx.GetCurrentDisplay = _real_glXGetCurrentDisplay;*/
+
+   /*** GLX_VERSION_1_3 ***/
+   glx.ChooseFBConfig = _real_glXChooseFBConfig;
+   glx.CreateNewContext = _real_glXCreateNewContext;
+   glx.CreatePbuffer = _real_glXCreatePbuffer;
+   glx.CreatePixmap = _real_glXCreatePixmap;
+   glx.CreateWindow = _real_glXCreateWindow;
+   glx.DestroyPbuffer = _real_glXDestroyPbuffer;
+   glx.DestroyPixmap = _real_glXDestroyPixmap;
+   glx.DestroyWindow = _real_glXDestroyWindow;
+   /*glx.GetCurrentReadDrawable = _real_glXGetCurrentReadDrawable;*/
+   glx.GetFBConfigAttrib = _real_glXGetFBConfigAttrib;
+   glx.GetFBConfigs = _real_glXGetFBConfigs;
+   glx.GetSelectedEvent = _real_glXGetSelectedEvent;
+   glx.GetVisualFromFBConfig = _real_glXGetVisualFromFBConfig;
+   glx.MakeContextCurrent = _real_glXMakeContextCurrent;
+   glx.QueryContext = _real_glXQueryContext;
+   glx.QueryDrawable = _real_glXQueryDrawable;
+   glx.SelectEvent = _real_glXSelectEvent;
+
+   /*** GLX_SGI_swap_control ***/
+   glx.SwapIntervalSGI = _real_glXSwapIntervalSGI;
+
+   /*** GLX_SGI_video_sync ***/
+   glx.GetVideoSyncSGI = _real_glXGetVideoSyncSGI;
+   glx.WaitVideoSyncSGI = _real_glXWaitVideoSyncSGI;
+
+   /*** GLX_SGI_make_current_read ***/
+   glx.MakeCurrentReadSGI = _real_glXMakeCurrentReadSGI;
+   /*glx.GetCurrentReadDrawableSGI = _real_glXGetCurrentReadDrawableSGI;*/
+
+#if defined(_VL_H)
+   /*** GLX_SGIX_video_source ***/
+   glx.CreateGLXVideoSourceSGIX = _real_glXCreateGLXVideoSourceSGIX;
+   glx.DestroyGLXVideoSourceSGIX = _real_glXDestroyGLXVideoSourceSGIX;
+#endif
+
+   /*** GLX_EXT_import_context ***/
+   glx.FreeContextEXT = _real_glXFreeContextEXT;
+   /*glx.GetContextIDEXT = _real_glXGetContextIDEXT;*/
+   /*glx.GetCurrentDisplayEXT = _real_glXGetCurrentDisplayEXT;*/
+   glx.ImportContextEXT = _real_glXImportContextEXT;
+   glx.QueryContextInfoEXT = _real_glXQueryContextInfoEXT;
+
+   /*** GLX_SGIX_fbconfig ***/
+   glx.GetFBConfigAttribSGIX = _real_glXGetFBConfigAttribSGIX;
+   glx.ChooseFBConfigSGIX = _real_glXChooseFBConfigSGIX;
+   glx.CreateGLXPixmapWithConfigSGIX = _real_glXCreateGLXPixmapWithConfigSGIX;
+   glx.CreateContextWithConfigSGIX = _real_glXCreateContextWithConfigSGIX;
+   glx.GetVisualFromFBConfigSGIX = _real_glXGetVisualFromFBConfigSGIX;
+   glx.GetFBConfigFromVisualSGIX = _real_glXGetFBConfigFromVisualSGIX;
+
+   /*** GLX_SGIX_pbuffer ***/
+   glx.CreateGLXPbufferSGIX = _real_glXCreateGLXPbufferSGIX;
+   glx.DestroyGLXPbufferSGIX = _real_glXDestroyGLXPbufferSGIX;
+   glx.QueryGLXPbufferSGIX = _real_glXQueryGLXPbufferSGIX;
+   glx.SelectEventSGIX = _real_glXSelectEventSGIX;
+   glx.GetSelectedEventSGIX = _real_glXGetSelectedEventSGIX;
+
+   /*** GLX_SGI_cushion ***/
+   glx.CushionSGI = _real_glXCushionSGI;
+
+   /*** GLX_SGIX_video_resize ***/
+   glx.BindChannelToWindowSGIX = _real_glXBindChannelToWindowSGIX;
+   glx.ChannelRectSGIX = _real_glXChannelRectSGIX;
+   glx.QueryChannelRectSGIX = _real_glXQueryChannelRectSGIX;
+   glx.QueryChannelDeltasSGIX = _real_glXQueryChannelDeltasSGIX;
+   glx.ChannelRectSyncSGIX = _real_glXChannelRectSyncSGIX;
+
+#if defined(_DM_BUFFER_H_)
+   /*** (GLX_SGIX_dmbuffer ***/
+   glx.AssociateDMPbufferSGIX = NULL;
+#endif
+
+   /*** GLX_SGIX_swap_group ***/
+   glx.JoinSwapGroupSGIX = _real_glXJoinSwapGroupSGIX;
+
+   /*** GLX_SGIX_swap_barrier ***/
+   glx.BindSwapBarrierSGIX = _real_glXBindSwapBarrierSGIX;
+   glx.QueryMaxSwapBarriersSGIX = _real_glXQueryMaxSwapBarriersSGIX;
+
+   /*** GLX_SUN_get_transparent_index ***/
+   glx.GetTransparentIndexSUN = _real_glXGetTransparentIndexSUN;
+
+   /*** GLX_MESA_copy_sub_buffer ***/
+   glx.CopySubBufferMESA = _real_glXCopySubBufferMESA;
+
+   /*** GLX_MESA_release_buffers ***/
+   glx.ReleaseBuffersMESA = _real_glXReleaseBuffersMESA;
+
+   /*** GLX_MESA_pixmap_colormap ***/
+   glx.CreateGLXPixmapMESA = _real_glXCreateGLXPixmapMESA;
+
+   /*** GLX_MESA_set_3dfx_mode ***/
+   glx.Set3DfxModeMESA = _real_glXSet3DfxModeMESA;
+
+   /*** GLX_NV_vertex_array_range ***/
+   glx.AllocateMemoryNV = _real_glXAllocateMemoryNV;
+   glx.FreeMemoryNV = _real_glXFreeMemoryNV;
+
+   /*** GLX_MESA_agp_offset ***/
+   glx.GetAGPOffsetMESA = _real_glXGetAGPOffsetMESA;
+
+   return &glx;
+}
diff --git a/src/gallium/winsys/xlib/realglx.h b/src/gallium/winsys/xlib/realglx.h
new file mode 100644
index 0000000000..150129db68
--- /dev/null
+++ b/src/gallium/winsys/xlib/realglx.h
@@ -0,0 +1,326 @@
+
+/*
+ * Mesa 3-D graphics library
+ * Version:  3.5
+ * 
+ * Copyright (C) 1999-2001  Brian Paul   All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ * 
+ * The above copyright notice and this permission notice shall be included
+ * in all copies or substantial portions of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * BRIAN PAUL BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN
+ * AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ */
+
+
+#ifndef REALGLX_H
+#define REALGLX_H
+
+
+extern struct _glxapi_table *
+_real_GetGLXDispatchTable(void);
+
+
+/*
+ * Basically just need these to prevent compiler warnings.
+ */
+
+
+extern XVisualInfo *
+_real_glXChooseVisual( Display *dpy, int screen, int *list );
+
+extern GLXContext
+_real_glXCreateContext( Display *dpy, XVisualInfo *visinfo,
+                        GLXContext share_list, Bool direct );
+
+extern GLXPixmap
+_real_glXCreateGLXPixmap( Display *dpy, XVisualInfo *visinfo, Pixmap pixmap );
+
+extern GLXPixmap
+_real_glXCreateGLXPixmapMESA( Display *dpy, XVisualInfo *visinfo,
+                              Pixmap pixmap, Colormap cmap );
+
+extern void
+_real_glXDestroyGLXPixmap( Display *dpy, GLXPixmap pixmap );
+
+extern void
+_real_glXCopyContext( Display *dpy, GLXContext src, GLXContext dst,
+                      unsigned long mask );
+
+extern Bool
+_real_glXMakeCurrent( Display *dpy, GLXDrawable drawable, GLXContext ctx );
+
+extern Bool
+_real_glXQueryExtension( Display *dpy, int *errorb, int *event );
+
+extern void
+_real_glXDestroyContext( Display *dpy, GLXContext ctx );
+
+extern Bool
+_real_glXIsDirect( Display *dpy, GLXContext ctx );
+
+extern void
+_real_glXSwapBuffers( Display *dpy, GLXDrawable drawable );
+
+extern void
+_real_glXUseXFont( Font font, int first, int count, int listbase );
+
+extern Bool
+_real_glXQueryVersion( Display *dpy, int *maj, int *min );
+
+extern int
+_real_glXGetConfig( Display *dpy, XVisualInfo *visinfo,
+                    int attrib, int *value );
+
+extern void
+_real_glXWaitGL( void );
+
+
+extern void
+_real_glXWaitX( void );
+
+/* GLX 1.1 and later */
+extern const char *
+_real_glXQueryExtensionsString( Display *dpy, int screen );
+
+/* GLX 1.1 and later */
+extern const char *
+_real_glXQueryServerString( Display *dpy, int screen, int name );
+
+/* GLX 1.1 and later */
+extern const char *
+_real_glXGetClientString( Display *dpy, int name );
+
+
+/*
+ * GLX 1.3 and later
+ */
+
+extern GLXFBConfig *
+_real_glXChooseFBConfig( Display *dpy, int screen,
+                         const int *attribList, int *nitems );
+
+extern int
+_real_glXGetFBConfigAttrib( Display *dpy, GLXFBConfig config,
+                            int attribute, int *value );
+
+extern GLXFBConfig *
+_real_glXGetFBConfigs( Display *dpy, int screen, int *nelements );
+
+extern XVisualInfo *
+_real_glXGetVisualFromFBConfig( Display *dpy, GLXFBConfig config );
+
+extern GLXWindow
+_real_glXCreateWindow( Display *dpy, GLXFBConfig config, Window win,
+                       const int *attribList );
+
+extern void
+_real_glXDestroyWindow( Display *dpy, GLXWindow window );
+
+extern GLXPixmap
+_real_glXCreatePixmap( Display *dpy, GLXFBConfig config, Pixmap pixmap,
+                       const int *attribList );
+
+extern void
+_real_glXDestroyPixmap( Display *dpy, GLXPixmap pixmap );
+
+extern GLXPbuffer
+_real_glXCreatePbuffer( Display *dpy, GLXFBConfig config,
+                        const int *attribList );
+
+extern void
+_real_glXDestroyPbuffer( Display *dpy, GLXPbuffer pbuf );
+
+extern void
+_real_glXQueryDrawable( Display *dpy, GLXDrawable draw, int attribute,
+                        unsigned int *value );
+
+extern GLXContext
+_real_glXCreateNewContext( Display *dpy, GLXFBConfig config,
+                           int renderType, GLXContext shareList, Bool direct );
+
+
+extern Bool
+_real_glXMakeContextCurrent( Display *dpy, GLXDrawable draw,
+                             GLXDrawable read, GLXContext ctx );
+
+extern int
+_real_glXQueryContext( Display *dpy, GLXContext ctx, int attribute, int *value );
+
+extern void
+_real_glXSelectEvent( Display *dpy, GLXDrawable drawable, unsigned long mask );
+
+extern void
+_real_glXGetSelectedEvent( Display *dpy, GLXDrawable drawable,
+                           unsigned long *mask );
+
+#ifdef GLX_SGI_swap_control
+extern int
+_real_glXSwapIntervalSGI(int interval);
+#endif
+
+
+#ifdef GLX_SGI_video_sync
+extern int
+_real_glXGetVideoSyncSGI(unsigned int *count);
+
+extern int
+_real_glXWaitVideoSyncSGI(int divisor, int remainder, unsigned int *count);
+#endif
+
+
+#ifdef GLX_SGI_make_current_read
+extern Bool
+_real_glXMakeCurrentReadSGI(Display *dpy, GLXDrawable draw, GLXDrawable read, GLXContext ctx);
+
+extern GLXDrawable
+_real_glXGetCurrentReadDrawableSGI(void);
+#endif
+
+#if defined(_VL_H) && defined(GLX_SGIX_video_source)
+extern GLXVideoSourceSGIX
+_real_glXCreateGLXVideoSourceSGIX(Display *dpy, int screen, VLServer server, VLPath path, int nodeClass, VLNode drainNode);
+
+extern void
+_real_glXDestroyGLXVideoSourceSGIX(Display *dpy, GLXVideoSourceSGIX src);
+#endif
+
+#ifdef GLX_EXT_import_context
+extern void
+_real_glXFreeContextEXT(Display *dpy, GLXContext context);
+
+extern GLXContextID
+_real_glXGetContextIDEXT(const GLXContext context);
+
+extern Display *
+_real_glXGetCurrentDisplayEXT(void);
+
+extern GLXContext
+_real_glXImportContextEXT(Display *dpy, GLXContextID contextID);
+
+extern int
+_real_glXQueryContextInfoEXT(Display *dpy, GLXContext context, int attribute, int *value);
+#endif
+
+#ifdef GLX_SGIX_fbconfig
+extern int
+_real_glXGetFBConfigAttribSGIX(Display *dpy, GLXFBConfigSGIX config, int attribute, int *value);
+
+extern GLXFBConfigSGIX *
+_real_glXChooseFBConfigSGIX(Display *dpy, int screen, int *attrib_list, int *nelements);
+
+extern GLXPixmap
+_real_glXCreateGLXPixmapWithConfigSGIX(Display *dpy, GLXFBConfigSGIX config, Pixmap pixmap);
+
+extern GLXContext
+_real_glXCreateContextWithConfigSGIX(Display *dpy, GLXFBConfigSGIX config, int render_type, GLXContext share_list, Bool direct);
+
+extern XVisualInfo *
+_real_glXGetVisualFromFBConfigSGIX(Display *dpy, GLXFBConfigSGIX config);
+
+extern GLXFBConfigSGIX
+_real_glXGetFBConfigFromVisualSGIX(Display *dpy, XVisualInfo *vis);
+#endif
+
+#ifdef GLX_SGIX_pbuffer
+extern GLXPbufferSGIX
+_real_glXCreateGLXPbufferSGIX(Display *dpy, GLXFBConfigSGIX config, unsigned int width, unsigned int height, int *attrib_list);
+
+extern void
+_real_glXDestroyGLXPbufferSGIX(Display *dpy, GLXPbufferSGIX pbuf);
+
+extern int
+_real_glXQueryGLXPbufferSGIX(Display *dpy, GLXPbufferSGIX pbuf, int attribute, unsigned int *value);
+
+extern void
+_real_glXSelectEventSGIX(Display *dpy, GLXDrawable drawable, unsigned long mask);
+
+extern void
+_real_glXGetSelectedEventSGIX(Display *dpy, GLXDrawable drawable, unsigned long *mask);
+#endif
+
+#ifdef GLX_SGI_cushion
+extern void
+_real_glXCushionSGI(Display *dpy, Window win, float cushion);
+#endif
+
+#ifdef GLX_SGIX_video_resize
+extern int
+_real_glXBindChannelToWindowSGIX(Display *dpy, int screen, int channel , Window window);
+
+extern int
+_real_glXChannelRectSGIX(Display *dpy, int screen, int channel, int x, int y, int w, int h);
+
+extern int
+_real_glXQueryChannelRectSGIX(Display *dpy, int screen, int channel, int *x, int *y, int *w, int *h);
+
+extern int
+_real_glXQueryChannelDeltasSGIX(Display *dpy, int screen, int channel, int *dx, int *dy, int *dw, int *dh);
+
+extern int
+_real_glXChannelRectSyncSGIX(Display *dpy, int screen, int channel, GLenum synctype);
+#endif
+
+#if defined(_DM_BUFFER_H_) && defined(GLX_SGIX_dmbuffer)
+extern Bool
+_real_glXAssociateDMPbufferSGIX(Display *dpy, GLXPbufferSGIX pbuffer, DMparams *params, DMbuffer dmbuffer);
+#endif
+
+#ifdef GLX_SGIX_swap_group
+extern void
+_real_glXJoinSwapGroupSGIX(Display *dpy, GLXDrawable drawable, GLXDrawable member);
+#endif
+
+#ifdef GLX_SGIX_swap_barrier
+extern void
+_real_glXBindSwapBarrierSGIX(Display *dpy, GLXDrawable drawable, int barrier);
+
+extern Bool
+_real_glXQueryMaxSwapBarriersSGIX(Display *dpy, int screen, int *max);
+#endif
+
+#ifdef GLX_SUN_get_transparent_index
+extern Status
+_real_glXGetTransparentIndexSUN(Display *dpy, Window overlay, Window underlay, long *pTransparent);
+#endif
+
+#ifdef GLX_MESA_release_buffers
+extern Bool
+_real_glXReleaseBuffersMESA( Display *dpy, GLXDrawable d );
+#endif
+
+#ifdef GLX_MESA_set_3dfx_mode
+extern Bool
+_real_glXSet3DfxModeMESA( int mode );
+#endif
+
+#ifdef GLX_NV_vertex_array_range
+extern void *
+_real_glXAllocateMemoryNV(GLsizei size, GLfloat readfreq, GLfloat writefreq, GLfloat priority);
+extern void
+_real_glXFreeMemoryNV(GLvoid *pointer);
+#endif
+
+#ifdef GLX_MESA_agp_offset
+extern GLuint
+_real_glXGetAGPOffsetMESA(const GLvoid *pointer);
+#endif
+
+#ifdef GLX_MESA_copy_sub_buffer
+extern void
+_real_glXCopySubBufferMESA( Display *dpy, GLXDrawable drawable,
+                            int x, int y, int width, int height );
+#endif
+
+#endif /* REALGLX_H */
diff --git a/src/gallium/winsys/xlib/xfonts.c b/src/gallium/winsys/xlib/xfonts.c
new file mode 100644
index 0000000000..d72c600bd1
--- /dev/null
+++ b/src/gallium/winsys/xlib/xfonts.c
@@ -0,0 +1,377 @@
+
+/*
+ * Mesa 3-D graphics library
+ * Version:  3.5
+ *
+ * Copyright (C) 1999-2000  Brian Paul   All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included
+ * in all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * BRIAN PAUL BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN
+ * AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ */
+
+
+/* xfonts.c -- glXUseXFont() for Mesa written by
+ * Copyright (C) 1995 Thorsten.Ohl @ Physik.TH-Darmstadt.de
+ */
+
+#ifdef __VMS
+#include <GL/vms_x_fix.h>
+#endif
+
+#include "glxheader.h"
+#include "context.h"
+#include "imports.h"
+#include "xfonts.h"
+
+
+/* Some debugging info.  */
+
+#ifdef DEBUG
+#undef _R
+#undef _G
+#undef _B
+#include <ctype.h>
+
+int debug_xfonts = 0;
+
+static void
+dump_char_struct(XCharStruct * ch, char *prefix)
+{
+   printf("%slbearing = %d, rbearing = %d, width = %d\n",
+	  prefix, ch->lbearing, ch->rbearing, ch->width);
+   printf("%sascent = %d, descent = %d, attributes = %u\n",
+	  prefix, ch->ascent, ch->descent, (unsigned int) ch->attributes);
+}
+
+static void
+dump_font_struct(XFontStruct * font)
+{
+   printf("ascent = %d, descent = %d\n", font->ascent, font->descent);
+   printf("char_or_byte2 = (%u,%u)\n",
+	  font->min_char_or_byte2, font->max_char_or_byte2);
+   printf("byte1 = (%u,%u)\n", font->min_byte1, font->max_byte1);
+   printf("all_chars_exist = %s\n", font->all_chars_exist ? "True" : "False");
+   printf("default_char = %c (\\%03o)\n",
+	  (char) (isprint(font->default_char) ? font->default_char : ' '),
+	  font->default_char);
+   dump_char_struct(&font->min_bounds, "min> ");
+   dump_char_struct(&font->max_bounds, "max> ");
+#if 0
+   for (c = font->min_char_or_byte2; c <= font->max_char_or_byte2; c++) {
+      char prefix[8];
+      sprintf(prefix, "%d> ", c);
+      dump_char_struct(&font->per_char[c], prefix);
+   }
+#endif
+}
+
+static void
+dump_bitmap(unsigned int width, unsigned int height, GLubyte * bitmap)
+{
+   unsigned int x, y;
+
+   printf("    ");
+   for (x = 0; x < 8 * width; x++)
+      printf("%o", 7 - (x % 8));
+   putchar('\n');
+   for (y = 0; y < height; y++) {
+      printf("%3o:", y);
+      for (x = 0; x < 8 * width; x++)
+	 putchar((bitmap[width * (height - y - 1) + x / 8] & (1 << (7 - (x %
+									 8))))
+		 ? '*' : '.');
+      printf("   ");
+      for (x = 0; x < width; x++)
+	 printf("0x%02x, ", bitmap[width * (height - y - 1) + x]);
+      putchar('\n');
+   }
+}
+#endif /* DEBUG */
+
+
+/* Implementation.  */
+
+/* Fill a BITMAP with a character C from thew current font
+   in the graphics context GC.  WIDTH is the width in bytes
+   and HEIGHT is the height in bits.
+
+   Note that the generated bitmaps must be used with
+
+        glPixelStorei (GL_UNPACK_SWAP_BYTES, GL_FALSE);
+        glPixelStorei (GL_UNPACK_LSB_FIRST, GL_FALSE);
+        glPixelStorei (GL_UNPACK_ROW_LENGTH, 0);
+        glPixelStorei (GL_UNPACK_SKIP_ROWS, 0);
+        glPixelStorei (GL_UNPACK_SKIP_PIXELS, 0);
+        glPixelStorei (GL_UNPACK_ALIGNMENT, 1);
+
+   Possible optimizations:
+
+     * use only one reusable pixmap with the maximum dimensions.
+     * draw the entire font into a single pixmap (careful with
+       proportional fonts!).
+*/
+
+
+/*
+ * Generate OpenGL-compatible bitmap.
+ */
+static void
+fill_bitmap(Display * dpy, Window win, GC gc,
+	    unsigned int width, unsigned int height,
+	    int x0, int y0, unsigned int c, GLubyte * bitmap)
+{
+   XImage *image;
+   unsigned int x, y;
+   Pixmap pixmap;
+   XChar2b char2b;
+
+   pixmap = XCreatePixmap(dpy, win, 8 * width, height, 1);
+   XSetForeground(dpy, gc, 0);
+   XFillRectangle(dpy, pixmap, gc, 0, 0, 8 * width, height);
+   XSetForeground(dpy, gc, 1);
+
+   char2b.byte1 = (c >> 8) & 0xff;
+   char2b.byte2 = (c & 0xff);
+
+   XDrawString16(dpy, pixmap, gc, x0, y0, &char2b, 1);
+
+   image = XGetImage(dpy, pixmap, 0, 0, 8 * width, height, 1, XYPixmap);
+   if (image) {
+      /* Fill the bitmap (X11 and OpenGL are upside down wrt each other).  */
+      for (y = 0; y < height; y++)
+	 for (x = 0; x < 8 * width; x++)
+	    if (XGetPixel(image, x, y))
+	       bitmap[width * (height - y - 1) + x / 8] |=
+		  (1 << (7 - (x % 8)));
+      XDestroyImage(image);
+   }
+
+   XFreePixmap(dpy, pixmap);
+}
+
+/*
+ * determine if a given glyph is valid and return the
+ * corresponding XCharStruct.
+ */
+static XCharStruct *
+isvalid(XFontStruct * fs, unsigned int which)
+{
+   unsigned int rows, pages;
+   unsigned int byte1 = 0, byte2 = 0;
+   int i, valid = 1;
+
+   rows = fs->max_byte1 - fs->min_byte1 + 1;
+   pages = fs->max_char_or_byte2 - fs->min_char_or_byte2 + 1;
+
+   if (rows == 1) {
+      /* "linear" fonts */
+      if ((fs->min_char_or_byte2 > which) || (fs->max_char_or_byte2 < which))
+	 valid = 0;
+   }
+   else {
+      /* "matrix" fonts */
+      byte2 = which & 0xff;
+      byte1 = which >> 8;
+      if ((fs->min_char_or_byte2 > byte2) ||
+	  (fs->max_char_or_byte2 < byte2) ||
+	  (fs->min_byte1 > byte1) || (fs->max_byte1 < byte1))
+	 valid = 0;
+   }
+
+   if (valid) {
+      if (fs->per_char) {
+	 if (rows == 1) {
+	    /* "linear" fonts */
+	    return (fs->per_char + (which - fs->min_char_or_byte2));
+	 }
+	 else {
+	    /* "matrix" fonts */
+	    i = ((byte1 - fs->min_byte1) * pages) +
+	       (byte2 - fs->min_char_or_byte2);
+	    return (fs->per_char + i);
+	 }
+      }
+      else {
+	 return (&fs->min_bounds);
+      }
+   }
+   return (NULL);
+}
+
+
+void
+Fake_glXUseXFont(Font font, int first, int count, int listbase)
+{
+   Display *dpy;
+   Window win;
+   Pixmap pixmap;
+   GC gc;
+   XGCValues values;
+   unsigned long valuemask;
+   XFontStruct *fs;
+   GLint swapbytes, lsbfirst, rowlength;
+   GLint skiprows, skippixels, alignment;
+   unsigned int max_width, max_height, max_bm_width, max_bm_height;
+   GLubyte *bm;
+   int i;
+
+   dpy = glXGetCurrentDisplay();
+   if (!dpy)
+      return;			/* I guess glXMakeCurrent wasn't called */
+   win = RootWindow(dpy, DefaultScreen(dpy));
+
+   fs = XQueryFont(dpy, font);
+   if (!fs) {
+      _mesa_error(NULL, GL_INVALID_VALUE,
+		  "Couldn't get font structure information");
+      return;
+   }
+
+   /* Allocate a bitmap that can fit all characters.  */
+   max_width = fs->max_bounds.rbearing - fs->min_bounds.lbearing;
+   max_height = fs->max_bounds.ascent + fs->max_bounds.descent;
+   max_bm_width = (max_width + 7) / 8;
+   max_bm_height = max_height;
+
+   bm = (GLubyte *) MALLOC((max_bm_width * max_bm_height) * sizeof(GLubyte));
+   if (!bm) {
+      XFreeFontInfo(NULL, fs, 1);
+      _mesa_error(NULL, GL_OUT_OF_MEMORY,
+		  "Couldn't allocate bitmap in glXUseXFont()");
+      return;
+   }
+
+#if 0
+   /* get the page info */
+   pages = fs->max_char_or_byte2 - fs->min_char_or_byte2 + 1;
+   firstchar = (fs->min_byte1 << 8) + fs->min_char_or_byte2;
+   lastchar = (fs->max_byte1 << 8) + fs->max_char_or_byte2;
+   rows = fs->max_byte1 - fs->min_byte1 + 1;
+   unsigned int first_char, last_char, pages, rows;
+#endif
+
+   /* Save the current packing mode for bitmaps.  */
+   glGetIntegerv(GL_UNPACK_SWAP_BYTES, &swapbytes);
+   glGetIntegerv(GL_UNPACK_LSB_FIRST, &lsbfirst);
+   glGetIntegerv(GL_UNPACK_ROW_LENGTH, &rowlength);
+   glGetIntegerv(GL_UNPACK_SKIP_ROWS, &skiprows);
+   glGetIntegerv(GL_UNPACK_SKIP_PIXELS, &skippixels);
+   glGetIntegerv(GL_UNPACK_ALIGNMENT, &alignment);
+
+   /* Enforce a standard packing mode which is compatible with
+      fill_bitmap() from above.  This is actually the default mode,
+      except for the (non)alignment.  */
+   glPixelStorei(GL_UNPACK_SWAP_BYTES, GL_FALSE);
+   glPixelStorei(GL_UNPACK_LSB_FIRST, GL_FALSE);
+   glPixelStorei(GL_UNPACK_ROW_LENGTH, 0);
+   glPixelStorei(GL_UNPACK_SKIP_ROWS, 0);
+   glPixelStorei(GL_UNPACK_SKIP_PIXELS, 0);
+   glPixelStorei(GL_UNPACK_ALIGNMENT, 1);
+
+   pixmap = XCreatePixmap(dpy, win, 10, 10, 1);
+   values.foreground = BlackPixel(dpy, DefaultScreen(dpy));
+   values.background = WhitePixel(dpy, DefaultScreen(dpy));
+   values.font = fs->fid;
+   valuemask = GCForeground | GCBackground | GCFont;
+   gc = XCreateGC(dpy, pixmap, valuemask, &values);
+   XFreePixmap(dpy, pixmap);
+
+#ifdef DEBUG
+   if (debug_xfonts)
+      dump_font_struct(fs);
+#endif
+
+   for (i = 0; i < count; i++) {
+      unsigned int width, height, bm_width, bm_height;
+      GLfloat x0, y0, dx, dy;
+      XCharStruct *ch;
+      int x, y;
+      unsigned int c = first + i;
+      int list = listbase + i;
+      int valid;
+
+      /* check on index validity and get the bounds */
+      ch = isvalid(fs, c);
+      if (!ch) {
+	 ch = &fs->max_bounds;
+	 valid = 0;
+      }
+      else {
+	 valid = 1;
+      }
+
+#ifdef DEBUG
+      if (debug_xfonts) {
+	 char s[7];
+	 sprintf(s, isprint(c) ? "%c> " : "\\%03o> ", c);
+	 dump_char_struct(ch, s);
+      }
+#endif
+
+      /* glBitmap()' parameters:
+         straight from the glXUseXFont(3) manpage.  */
+      width = ch->rbearing - ch->lbearing;
+      height = ch->ascent + ch->descent;
+      x0 = -ch->lbearing;
+      y0 = ch->descent - 0;	/* XXX used to subtract 1 here */
+      /* but that caused a conformace failure */
+      dx = ch->width;
+      dy = 0;
+
+      /* X11's starting point.  */
+      x = -ch->lbearing;
+      y = ch->ascent;
+
+      /* Round the width to a multiple of eight.  We will use this also
+         for the pixmap for capturing the X11 font.  This is slightly
+         inefficient, but it makes the OpenGL part real easy.  */
+      bm_width = (width + 7) / 8;
+      bm_height = height;
+
+      glNewList(list, GL_COMPILE);
+      if (valid && (bm_width > 0) && (bm_height > 0)) {
+
+	 MEMSET(bm, '\0', bm_width * bm_height);
+	 fill_bitmap(dpy, win, gc, bm_width, bm_height, x, y, c, bm);
+
+	 glBitmap(width, height, x0, y0, dx, dy, bm);
+#ifdef DEBUG
+	 if (debug_xfonts) {
+	    printf("width/height = %u/%u\n", width, height);
+	    printf("bm_width/bm_height = %u/%u\n", bm_width, bm_height);
+	    dump_bitmap(bm_width, bm_height, bm);
+	 }
+#endif
+      }
+      else {
+	 glBitmap(0, 0, 0.0, 0.0, dx, dy, NULL);
+      }
+      glEndList();
+   }
+
+   FREE(bm);
+   XFreeFontInfo(NULL, fs, 1);
+   XFreeGC(dpy, gc);
+
+   /* Restore saved packing modes.  */
+   glPixelStorei(GL_UNPACK_SWAP_BYTES, swapbytes);
+   glPixelStorei(GL_UNPACK_LSB_FIRST, lsbfirst);
+   glPixelStorei(GL_UNPACK_ROW_LENGTH, rowlength);
+   glPixelStorei(GL_UNPACK_SKIP_ROWS, skiprows);
+   glPixelStorei(GL_UNPACK_SKIP_PIXELS, skippixels);
+   glPixelStorei(GL_UNPACK_ALIGNMENT, alignment);
+}
diff --git a/src/gallium/winsys/xlib/xfonts.h b/src/gallium/winsys/xlib/xfonts.h
new file mode 100644
index 0000000000..e36f42f817
--- /dev/null
+++ b/src/gallium/winsys/xlib/xfonts.h
@@ -0,0 +1,41 @@
+
+/*
+ * Mesa 3-D graphics library
+ * Version:  3.5
+ *
+ * Copyright (C) 1999-2000  Brian Paul   All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included
+ * in all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * BRIAN PAUL BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN
+ * AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ */
+
+
+#ifndef XFONTS_H
+#define XFONTS_H
+
+#ifdef __VMS
+#include <GL/vms_x_fix.h>
+#endif
+
+#include <X11/Xlib.h>
+
+
+extern void Fake_glXUseXFont( Font font, int first, int count, int listbase );
+
+
+#endif
+
diff --git a/src/gallium/winsys/xlib/xm_api.c b/src/gallium/winsys/xlib/xm_api.c
new file mode 100644
index 0000000000..e5fef1d7a8
--- /dev/null
+++ b/src/gallium/winsys/xlib/xm_api.c
@@ -0,0 +1,1380 @@
+/*
+ * Mesa 3-D graphics library
+ * Version:  7.1
+ *
+ * Copyright (C) 1999-2007  Brian Paul   All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included
+ * in all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * BRIAN PAUL BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN
+ * AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ */
+
+/**
+ * \file xm_api.c
+ *
+ * All the XMesa* API functions.
+ *
+ *
+ * NOTES:
+ *
+ * The window coordinate system origin (0,0) is in the lower-left corner
+ * of the window.  X11's window coordinate origin is in the upper-left
+ * corner of the window.  Therefore, most drawing functions in this
+ * file have to flip Y coordinates.
+ *
+ * Define USE_XSHM in the Makefile with -DUSE_XSHM if you want to compile
+ * in support for the MIT Shared Memory extension.  If enabled, when you
+ * use an Ximage for the back buffer in double buffered mode, the "swap"
+ * operation will be faster.  You must also link with -lXext.
+ *
+ * Byte swapping:  If the Mesa host and the X display use a different
+ * byte order then there's some trickiness to be aware of when using
+ * XImages.  The byte ordering used for the XImage is that of the X
+ * display, not the Mesa host.
+ * The color-to-pixel encoding for True/DirectColor must be done
+ * according to the display's visual red_mask, green_mask, and blue_mask.
+ * If XPutPixel is used to put a pixel into an XImage then XPutPixel will
+ * do byte swapping if needed.  If one wants to directly "poke" the pixel
+ * into the XImage's buffer then the pixel must be byte swapped first.
+ *
+ */
+
+#ifdef __CYGWIN__
+#undef WIN32
+#undef __WIN32__
+#endif
+
+#include "glxheader.h"
+#include "GL/xmesa.h"
+#include "xmesaP.h"
+#include "main/context.h"
+#include "main/framebuffer.h"
+#include "glapi/glthread.h"
+
+#include "state_tracker/st_public.h"
+#include "state_tracker/st_context.h"
+#include "pipe/p_defines.h"
+#include "pipe/p_context.h"
+
+#include "xm_winsys_aub.h"
+
+/**
+ * Global X driver lock
+ */
+_glthread_Mutex _xmesa_lock;
+
+
+int xmesa_mode;
+
+
+/**********************************************************************/
+/*****                     X Utility Functions                    *****/
+/**********************************************************************/
+
+
+/**
+ * Return the host's byte order as LSBFirst or MSBFirst ala X.
+ */
+#ifndef XFree86Server
+static int host_byte_order( void )
+{
+   int i = 1;
+   char *cptr = (char *) &i;
+   return (*cptr==1) ? LSBFirst : MSBFirst;
+}
+#endif
+
+
+/**
+ * Check if the X Shared Memory extension is available.
+ * Return:  0 = not available
+ *          1 = shared XImage support available
+ *          2 = shared Pixmap support available also
+ */
+static int check_for_xshm( XMesaDisplay *display )
+{
+#if defined(USE_XSHM) && !defined(XFree86Server)
+   int major, minor, ignore;
+   Bool pixmaps;
+
+   if (XQueryExtension( display, "MIT-SHM", &ignore, &ignore, &ignore )) {
+      if (XShmQueryVersion( display, &major, &minor, &pixmaps )==True) {
+	 return (pixmaps==True) ? 2 : 1;
+      }
+      else {
+	 return 0;
+      }
+   }
+   else {
+      return 0;
+   }
+#else
+   /* No  XSHM support */
+   return 0;
+#endif
+}
+
+
+/**
+ * Return the true number of bits per pixel for XImages.
+ * For example, if we request a 24-bit deep visual we may actually need/get
+ * 32bpp XImages.  This function returns the appropriate bpp.
+ * Input:  dpy - the X display
+ *         visinfo - desribes the visual to be used for XImages
+ * Return:  true number of bits per pixel for XImages
+ */
+static int
+bits_per_pixel( XMesaVisual xmv )
+{
+#ifdef XFree86Server
+   const int depth = xmv->nplanes;
+   int i;
+   assert(depth > 0);
+   for (i = 0; i < screenInfo.numPixmapFormats; i++) {
+      if (screenInfo.formats[i].depth == depth)
+         return screenInfo.formats[i].bitsPerPixel;
+   }
+   return depth;  /* should never get here, but this should be safe */
+#else
+   XMesaDisplay *dpy = xmv->display;
+   XMesaVisualInfo visinfo = xmv->visinfo;
+   XMesaImage *img;
+   int bitsPerPixel;
+   /* Create a temporary XImage */
+   img = XCreateImage( dpy, visinfo->visual, visinfo->depth,
+		       ZPixmap, 0,           /*format, offset*/
+		       (char*) MALLOC(8),    /*data*/
+		       1, 1,                 /*width, height*/
+		       32,                   /*bitmap_pad*/
+		       0                     /*bytes_per_line*/
+                     );
+   assert(img);
+   /* grab the bits/pixel value */
+   bitsPerPixel = img->bits_per_pixel;
+   /* free the XImage */
+   _mesa_free( img->data );
+   img->data = NULL;
+   XMesaDestroyImage( img );
+   return bitsPerPixel;
+#endif
+}
+
+
+
+/*
+ * Determine if a given X window ID is valid (window exists).
+ * Do this by calling XGetWindowAttributes() for the window and
+ * checking if we catch an X error.
+ * Input:  dpy - the display
+ *         win - the window to check for existance
+ * Return:  GL_TRUE - window exists
+ *          GL_FALSE - window doesn't exist
+ */
+#ifndef XFree86Server
+static GLboolean WindowExistsFlag;
+
+static int window_exists_err_handler( XMesaDisplay* dpy, XErrorEvent* xerr )
+{
+   (void) dpy;
+   if (xerr->error_code == BadWindow) {
+      WindowExistsFlag = GL_FALSE;
+   }
+   return 0;
+}
+
+static GLboolean window_exists( XMesaDisplay *dpy, Window win )
+{
+   XWindowAttributes wa;
+   int (*old_handler)( XMesaDisplay*, XErrorEvent* );
+   WindowExistsFlag = GL_TRUE;
+   old_handler = XSetErrorHandler(window_exists_err_handler);
+   XGetWindowAttributes( dpy, win, &wa ); /* dummy request */
+   XSetErrorHandler(old_handler);
+   return WindowExistsFlag;
+}
+
+static Status
+get_drawable_size( XMesaDisplay *dpy, Drawable d, uint *width, uint *height )
+{
+   Window root;
+   Status stat;
+   int xpos, ypos;
+   unsigned int w, h, bw, depth;
+   stat = XGetGeometry(dpy, d, &root, &xpos, &ypos, &w, &h, &bw, &depth);
+   *width = w;
+   *height = h;
+   return stat;
+}
+#endif
+
+
+/**
+ * Return the size of the window (or pixmap) that corresponds to the
+ * given XMesaBuffer.
+ * \param width  returns width in pixels
+ * \param height  returns height in pixels
+ */
+static void
+xmesa_get_window_size(XMesaDisplay *dpy, XMesaBuffer b,
+                      GLuint *width, GLuint *height)
+{
+#ifdef XFree86Server
+   *width = MIN2(b->drawable->width, MAX_WIDTH);
+   *height = MIN2(b->drawable->height, MAX_HEIGHT);
+#else
+   Status stat;
+
+   _glthread_LOCK_MUTEX(_xmesa_lock);
+   XSync(b->xm_visual->display, 0); /* added for Chromium */
+   stat = get_drawable_size(dpy, b->drawable, width, height);
+   _glthread_UNLOCK_MUTEX(_xmesa_lock);
+
+   if (!stat) {
+      /* probably querying a window that's recently been destroyed */
+      _mesa_warning(NULL, "XGetGeometry failed!\n");
+      *width = *height = 1;
+   }
+#endif
+}
+
+
+/**
+ * Choose the pixel format for the given visual.
+ * This will tell the gallium driver how to pack pixel data into
+ * drawing surfaces.
+ */
+static GLuint
+choose_pixel_format(XMesaVisual v)
+{
+   if (   GET_REDMASK(v)   == 0x0000ff
+       && GET_GREENMASK(v) == 0x00ff00
+       && GET_BLUEMASK(v)  == 0xff0000
+       && v->BitsPerPixel == 32) {
+      if (CHECK_BYTE_ORDER(v)) {
+         /* no byteswapping needed */
+         return 0 /* PIXEL_FORMAT_U_A8_B8_G8_R8 */;
+      }
+      else {
+         return PIPE_FORMAT_R8G8B8A8_UNORM;
+      }
+   }
+   else if (   GET_REDMASK(v)   == 0xff0000
+            && GET_GREENMASK(v) == 0x00ff00
+            && GET_BLUEMASK(v)  == 0x0000ff
+            && v->BitsPerPixel == 32) {
+      if (CHECK_BYTE_ORDER(v)) {
+         /* no byteswapping needed */
+         return PIPE_FORMAT_A8R8G8B8_UNORM;
+      }
+      else {
+         return PIPE_FORMAT_B8G8R8A8_UNORM;
+      }
+   }
+   else if (   GET_REDMASK(v)   == 0xf800
+            && GET_GREENMASK(v) == 0x07e0
+            && GET_BLUEMASK(v)  == 0x001f
+            && CHECK_BYTE_ORDER(v)
+            && v->BitsPerPixel == 16) {
+      /* 5-6-5 RGB */
+      return PIPE_FORMAT_R5G6B5_UNORM;
+   }
+
+   assert(0);
+   return 0;
+}
+
+
+
+/**********************************************************************/
+/*****                Linked list of XMesaBuffers                 *****/
+/**********************************************************************/
+
+XMesaBuffer XMesaBufferList = NULL;
+
+
+/**
+ * Allocate a new XMesaBuffer object which corresponds to the given drawable.
+ * Note that XMesaBuffer is derived from GLframebuffer.
+ * The new XMesaBuffer will not have any size (Width=Height=0).
+ *
+ * \param d  the corresponding X drawable (window or pixmap)
+ * \param type  either WINDOW, PIXMAP or PBUFFER, describing d
+ * \param vis  the buffer's visual
+ * \param cmap  the window's colormap, if known.
+ * \return new XMesaBuffer or NULL if any problem
+ */
+static XMesaBuffer
+create_xmesa_buffer(XMesaDrawable d, BufferType type,
+                    XMesaVisual vis, XMesaColormap cmap)
+{
+   XMesaBuffer b;
+   GLframebuffer *fb;
+   enum pipe_format colorFormat, depthFormat, stencilFormat;
+   uint width, height;
+
+   ASSERT(type == WINDOW || type == PIXMAP || type == PBUFFER);
+
+   b = (XMesaBuffer) CALLOC_STRUCT(xmesa_buffer);
+   if (!b)
+      return NULL;
+
+   b->drawable = d;
+
+   b->xm_visual = vis;
+   b->type = type;
+   b->cmap = cmap;
+
+   /* determine PIPE_FORMATs for buffers */
+   colorFormat = choose_pixel_format(vis);
+
+   if (vis->mesa_visual.depthBits == 0)
+      depthFormat = PIPE_FORMAT_NONE;
+   else if (vis->mesa_visual.depthBits <= 16)
+      depthFormat = PIPE_FORMAT_Z16_UNORM;
+   else if (vis->mesa_visual.depthBits <= 24)
+      depthFormat = PIPE_FORMAT_S8Z24_UNORM;
+   else
+      depthFormat = PIPE_FORMAT_Z32_UNORM;
+
+   if (vis->mesa_visual.stencilBits == 8) {
+      if (depthFormat == PIPE_FORMAT_S8Z24_UNORM)
+         stencilFormat = depthFormat;
+      else
+         stencilFormat = PIPE_FORMAT_S8_UNORM;
+   }
+   else {
+      stencilFormat = PIPE_FORMAT_NONE;
+   }
+
+
+   get_drawable_size(vis->display, d, &width, &height);
+
+   /*
+    * Create framebuffer, but we'll plug in our own renderbuffers below.
+    */
+   b->stfb = st_create_framebuffer(&vis->mesa_visual,
+                                   colorFormat, depthFormat, stencilFormat,
+                                   width, height,
+                                   (void *) b);
+   fb = &b->stfb->Base;
+
+   /*
+    * Create scratch XImage for xmesa_display_surface()
+    */
+   b->tempImage = XCreateImage(vis->display,
+                               vis->visinfo->visual,
+                               vis->visinfo->depth,
+                               ZPixmap, 0,   /* format, offset */
+                               NULL,         /* data */
+                               0, 0,         /* size */
+                               32,           /* bitmap_pad */
+                               0);           /* bytes_per_line */
+
+   /* GLX_EXT_texture_from_pixmap */
+   b->TextureTarget = 0;
+   b->TextureFormat = GLX_TEXTURE_FORMAT_NONE_EXT;
+   b->TextureMipmap = 0;
+
+   /* insert buffer into linked list */
+   b->Next = XMesaBufferList;
+   XMesaBufferList = b;
+
+   return b;
+}
+
+
+/**
+ * Find an XMesaBuffer by matching X display and colormap but NOT matching
+ * the notThis buffer.
+ */
+XMesaBuffer
+xmesa_find_buffer(XMesaDisplay *dpy, XMesaColormap cmap, XMesaBuffer notThis)
+{
+   XMesaBuffer b;
+   for (b = XMesaBufferList; b; b = b->Next) {
+      if (b->xm_visual->display == dpy &&
+          b->cmap == cmap &&
+          b != notThis) {
+         return b;
+      }
+   }
+   return NULL;
+}
+
+
+/**
+ * Remove buffer from linked list, delete if no longer referenced.
+ */
+static void
+xmesa_free_buffer(XMesaBuffer buffer)
+{
+   XMesaBuffer prev = NULL, b;
+
+   for (b = XMesaBufferList; b; b = b->Next) {
+      if (b == buffer) {
+         struct gl_framebuffer *fb = &buffer->stfb->Base;
+
+         /* unlink buffer from list */
+         if (prev)
+            prev->Next = buffer->Next;
+         else
+            XMesaBufferList = buffer->Next;
+
+         /* mark as delete pending */
+         fb->DeletePending = GL_TRUE;
+
+         /* Since the X window for the XMesaBuffer is going away, we don't
+          * want to dereference this pointer in the future.
+          */
+         b->drawable = 0;
+
+         buffer->tempImage->data = NULL;
+         XDestroyImage(buffer->tempImage);
+
+         /* Unreference.  If count = zero we'll really delete the buffer */
+         _mesa_unreference_framebuffer(&fb);
+
+         XFreeGC(b->xm_visual->display, b->gc);
+
+         free(buffer);
+
+         return;
+      }
+      /* continue search */
+      prev = b;
+   }
+   /* buffer not found in XMesaBufferList */
+   _mesa_problem(NULL,"xmesa_free_buffer() - buffer not found\n");
+}
+
+
+
+/**********************************************************************/
+/*****                   Misc Private Functions                   *****/
+/**********************************************************************/
+
+
+/**
+ * When a context is bound for the first time, we can finally finish
+ * initializing the context's visual and buffer information.
+ * \param v  the XMesaVisual to initialize
+ * \param b  the XMesaBuffer to initialize (may be NULL)
+ * \param rgb_flag  TRUE = RGBA mode, FALSE = color index mode
+ * \param window  the window/pixmap we're rendering into
+ * \param cmap  the colormap associated with the window/pixmap
+ * \return GL_TRUE=success, GL_FALSE=failure
+ */
+static GLboolean
+initialize_visual_and_buffer(XMesaVisual v, XMesaBuffer b,
+                             GLboolean rgb_flag, XMesaDrawable window,
+                             XMesaColormap cmap)
+{
+#ifdef XFree86Server
+   int client = (window) ? CLIENT_ID(window->id) : 0;
+#endif
+
+   ASSERT(!b || b->xm_visual == v);
+
+   /* Save true bits/pixel */
+   v->BitsPerPixel = bits_per_pixel(v);
+   assert(v->BitsPerPixel > 0);
+
+   if (rgb_flag == GL_FALSE) {
+      /* COLOR-INDEXED WINDOW: not supported*/
+      return GL_FALSE;
+   }
+   else {
+      /* RGB WINDOW:
+       * We support RGB rendering into almost any kind of visual.
+       */
+      const int xclass = v->mesa_visual.visualType;
+      if (xclass != GLX_TRUE_COLOR && xclass == !GLX_DIRECT_COLOR) {
+	 _mesa_warning(NULL,
+            "XMesa: RGB mode rendering not supported in given visual.\n");
+	 return GL_FALSE;
+      }
+      v->mesa_visual.indexBits = 0;
+   }
+
+   /*
+    * If MESA_INFO env var is set print out some debugging info
+    * which can help Brian figure out what's going on when a user
+    * reports bugs.
+    */
+   if (_mesa_getenv("MESA_INFO")) {
+      _mesa_printf("X/Mesa visual = %p\n", (void *) v);
+      _mesa_printf("X/Mesa level = %d\n", v->mesa_visual.level);
+      _mesa_printf("X/Mesa depth = %d\n", GET_VISUAL_DEPTH(v));
+      _mesa_printf("X/Mesa bits per pixel = %d\n", v->BitsPerPixel);
+   }
+
+   if (b && window) {
+      /* these should have been set in create_xmesa_buffer */
+      ASSERT(b->drawable == window);
+
+      /* Setup for single/double buffering */
+      if (v->mesa_visual.doubleBufferMode) {
+         /* Double buffered */
+         b->shm = check_for_xshm( v->display );
+      }
+
+      /* X11 graphics context */
+#ifdef XFree86Server
+      b->gc = CreateScratchGC(v->display, window->depth);
+#else
+      b->gc = XCreateGC( v->display, window, 0, NULL );
+#endif
+      XMesaSetFunction( v->display, b->gc, GXcopy );
+   }
+
+   return GL_TRUE;
+}
+
+
+
+#define NUM_VISUAL_TYPES   6
+
+/**
+ * Convert an X visual type to a GLX visual type.
+ * 
+ * \param visualType X visual type (i.e., \c TrueColor, \c StaticGray, etc.)
+ *        to be converted.
+ * \return If \c visualType is a valid X visual type, a GLX visual type will
+ *         be returned.  Otherwise \c GLX_NONE will be returned.
+ * 
+ * \note
+ * This code was lifted directly from lib/GL/glx/glcontextmodes.c in the
+ * DRI CVS tree.
+ */
+static GLint
+xmesa_convert_from_x_visual_type( int visualType )
+{
+    static const int glx_visual_types[ NUM_VISUAL_TYPES ] = {
+	GLX_STATIC_GRAY,  GLX_GRAY_SCALE,
+	GLX_STATIC_COLOR, GLX_PSEUDO_COLOR,
+	GLX_TRUE_COLOR,   GLX_DIRECT_COLOR
+    };
+
+    return ( (unsigned) visualType < NUM_VISUAL_TYPES )
+	? glx_visual_types[ visualType ] : GLX_NONE;
+}
+
+
+/**********************************************************************/
+/*****                       Public Functions                     *****/
+/**********************************************************************/
+
+
+/*
+ * Create a new X/Mesa visual.
+ * Input:  display - X11 display
+ *         visinfo - an XVisualInfo pointer
+ *         rgb_flag - GL_TRUE = RGB mode,
+ *                    GL_FALSE = color index mode
+ *         alpha_flag - alpha buffer requested?
+ *         db_flag - GL_TRUE = double-buffered,
+ *                   GL_FALSE = single buffered
+ *         stereo_flag - stereo visual?
+ *         ximage_flag - GL_TRUE = use an XImage for back buffer,
+ *                       GL_FALSE = use an off-screen pixmap for back buffer
+ *         depth_size - requested bits/depth values, or zero
+ *         stencil_size - requested bits/stencil values, or zero
+ *         accum_red_size - requested bits/red accum values, or zero
+ *         accum_green_size - requested bits/green accum values, or zero
+ *         accum_blue_size - requested bits/blue accum values, or zero
+ *         accum_alpha_size - requested bits/alpha accum values, or zero
+ *         num_samples - number of samples/pixel if multisampling, or zero
+ *         level - visual level, usually 0
+ *         visualCaveat - ala the GLX extension, usually GLX_NONE
+ * Return;  a new XMesaVisual or 0 if error.
+ */
+PUBLIC
+XMesaVisual XMesaCreateVisual( XMesaDisplay *display,
+                               XMesaVisualInfo visinfo,
+                               GLboolean rgb_flag,
+                               GLboolean alpha_flag,
+                               GLboolean db_flag,
+                               GLboolean stereo_flag,
+                               GLboolean ximage_flag,
+                               GLint depth_size,
+                               GLint stencil_size,
+                               GLint accum_red_size,
+                               GLint accum_green_size,
+                               GLint accum_blue_size,
+                               GLint accum_alpha_size,
+                               GLint num_samples,
+                               GLint level,
+                               GLint visualCaveat )
+{
+   XMesaVisual v;
+   GLint red_bits, green_bits, blue_bits, alpha_bits;
+
+#ifndef XFree86Server
+   /* For debugging only */
+   if (_mesa_getenv("MESA_XSYNC")) {
+      /* This makes debugging X easier.
+       * In your debugger, set a breakpoint on _XError to stop when an
+       * X protocol error is generated.
+       */
+      XSynchronize( display, 1 );
+   }
+#endif
+
+   v = (XMesaVisual) CALLOC_STRUCT(xmesa_visual);
+   if (!v) {
+      return NULL;
+   }
+
+   v->display = display;
+
+   /* Save a copy of the XVisualInfo struct because the user may X_mesa_free()
+    * the struct but we may need some of the information contained in it
+    * at a later time.
+    */
+#ifndef XFree86Server
+   v->visinfo = (XVisualInfo *) MALLOC(sizeof(*visinfo));
+   if(!v->visinfo) {
+      _mesa_free(v);
+      return NULL;
+   }
+   MEMCPY(v->visinfo, visinfo, sizeof(*visinfo));
+#endif
+
+   v->ximage_flag = ximage_flag;
+
+#ifdef XFree86Server
+   /* We could calculate these values by ourselves.  nplanes is either the sum
+    * of the red, green, and blue bits or the number index bits.
+    * ColormapEntries is either (1U << index_bits) or
+    * (1U << max(redBits, greenBits, blueBits)).
+    */
+   assert(visinfo->nplanes > 0);
+   v->nplanes = visinfo->nplanes;
+   v->ColormapEntries = visinfo->ColormapEntries;
+
+   v->mesa_visual.redMask = visinfo->redMask;
+   v->mesa_visual.greenMask = visinfo->greenMask;
+   v->mesa_visual.blueMask = visinfo->blueMask;
+   v->mesa_visual.visualID = visinfo->vid;
+   v->mesa_visual.screen = 0; /* FIXME: What should be done here? */
+#else
+   v->mesa_visual.redMask = visinfo->red_mask;
+   v->mesa_visual.greenMask = visinfo->green_mask;
+   v->mesa_visual.blueMask = visinfo->blue_mask;
+   v->mesa_visual.visualID = visinfo->visualid;
+   v->mesa_visual.screen = visinfo->screen;
+#endif
+
+#if defined(XFree86Server) || !(defined(__cplusplus) || defined(c_plusplus))
+   v->mesa_visual.visualType = xmesa_convert_from_x_visual_type(visinfo->class);
+#else
+   v->mesa_visual.visualType = xmesa_convert_from_x_visual_type(visinfo->c_class);
+#endif
+
+   v->mesa_visual.visualRating = visualCaveat;
+
+   if (alpha_flag)
+      v->mesa_visual.alphaBits = 8;
+
+   (void) initialize_visual_and_buffer( v, NULL, rgb_flag, 0, 0 );
+
+   {
+      const int xclass = v->mesa_visual.visualType;
+      if (xclass == GLX_TRUE_COLOR || xclass == GLX_DIRECT_COLOR) {
+         red_bits   = _mesa_bitcount(GET_REDMASK(v));
+         green_bits = _mesa_bitcount(GET_GREENMASK(v));
+         blue_bits  = _mesa_bitcount(GET_BLUEMASK(v));
+      }
+      else {
+         /* this is an approximation */
+         int depth;
+         depth = GET_VISUAL_DEPTH(v);
+         red_bits = depth / 3;
+         depth -= red_bits;
+         green_bits = depth / 2;
+         depth -= green_bits;
+         blue_bits = depth;
+         alpha_bits = 0;
+         assert( red_bits + green_bits + blue_bits == GET_VISUAL_DEPTH(v) );
+      }
+      alpha_bits = v->mesa_visual.alphaBits;
+   }
+
+   _mesa_initialize_visual( &v->mesa_visual,
+                            rgb_flag, db_flag, stereo_flag,
+                            red_bits, green_bits,
+                            blue_bits, alpha_bits,
+                            v->mesa_visual.indexBits,
+                            depth_size,
+                            stencil_size,
+                            accum_red_size, accum_green_size,
+                            accum_blue_size, accum_alpha_size,
+                            0 );
+
+   /* XXX minor hack */
+   v->mesa_visual.level = level;
+   return v;
+}
+
+
+PUBLIC
+void XMesaDestroyVisual( XMesaVisual v )
+{
+#ifndef XFree86Server
+   _mesa_free(v->visinfo);
+#endif
+   _mesa_free(v);
+}
+
+
+
+/**
+ * Create a new XMesaContext.
+ * \param v  the XMesaVisual
+ * \param share_list  another XMesaContext with which to share display
+ *                    lists or NULL if no sharing is wanted.
+ * \return an XMesaContext or NULL if error.
+ */
+PUBLIC
+XMesaContext XMesaCreateContext( XMesaVisual v, XMesaContext share_list )
+{
+   static GLboolean firstTime = GL_TRUE;
+   struct pipe_context *pipe;
+   XMesaContext c;
+   GLcontext *mesaCtx;
+   uint pf;
+
+   if (firstTime) {
+      _glthread_INIT_MUTEX(_xmesa_lock);
+      firstTime = GL_FALSE;
+   }
+
+   /* Note: the XMesaContext contains a Mesa GLcontext struct (inheritance) */
+   c = (XMesaContext) CALLOC_STRUCT(xmesa_context);
+   if (!c)
+      return NULL;
+
+   pf = choose_pixel_format(v);
+   assert(pf);
+
+   if (!getenv("XM_AUB")) {
+      xmesa_mode = XMESA_SOFTPIPE;
+      pipe = xmesa_create_pipe_context( c, pf );
+   }
+   else {
+      xmesa_mode = XMESA_AUB;
+      pipe = xmesa_create_i965simple( xmesa_get_pipe_winsys_aub() );
+   }
+
+   c->st = st_create_context(pipe, &v->mesa_visual,
+                             share_list ? share_list->st : NULL);
+   mesaCtx = c->st->ctx;
+   c->st->ctx->DriverCtx = c;
+
+#if 00
+   _mesa_enable_sw_extensions(mesaCtx);
+   _mesa_enable_1_3_extensions(mesaCtx);
+   _mesa_enable_1_4_extensions(mesaCtx);
+   _mesa_enable_1_5_extensions(mesaCtx);
+   _mesa_enable_2_0_extensions(mesaCtx);
+#endif
+
+#ifdef XFree86Server
+   /* If we're running in the X server, do bounds checking to prevent
+    * segfaults and server crashes!
+    */
+   mesaCtx->Const.CheckArrayBounds = GL_TRUE;
+#endif
+
+   /* finish up xmesa context initializations */
+   c->xm_visual = v;
+   c->xm_buffer = NULL;   /* set later by XMesaMakeCurrent */
+
+   c->st->haveFramebufferSurfaces = GL_TRUE;
+
+   return c;
+}
+
+
+
+PUBLIC
+void XMesaDestroyContext( XMesaContext c )
+{
+   st_destroy_context(c->st);
+   _mesa_free(c);
+}
+
+
+
+/**
+ * Private function for creating an XMesaBuffer which corresponds to an
+ * X window or pixmap.
+ * \param v  the window's XMesaVisual
+ * \param w  the window we're wrapping
+ * \return  new XMesaBuffer or NULL if error
+ */
+PUBLIC XMesaBuffer
+XMesaCreateWindowBuffer(XMesaVisual v, XMesaWindow w)
+{
+#ifndef XFree86Server
+   XWindowAttributes attr;
+#endif
+   XMesaBuffer b;
+   XMesaColormap cmap;
+   int depth;
+
+   assert(v);
+   assert(w);
+
+   /* Check that window depth matches visual depth */
+#ifdef XFree86Server
+   depth = ((XMesaDrawable)w)->depth;
+#else
+   XGetWindowAttributes( v->display, w, &attr );
+   depth = attr.depth;
+#endif
+   if (GET_VISUAL_DEPTH(v) != depth) {
+      _mesa_warning(NULL, "XMesaCreateWindowBuffer: depth mismatch between visual (%d) and window (%d)!\n",
+                    GET_VISUAL_DEPTH(v), depth);
+      return NULL;
+   }
+
+   /* Find colormap */
+#ifdef XFree86Server
+   cmap = (ColormapPtr)LookupIDByType(wColormap(w), RT_COLORMAP);
+#else
+   if (attr.colormap) {
+      cmap = attr.colormap;
+   }
+   else {
+      _mesa_warning(NULL, "Window %u has no colormap!\n", (unsigned int) w);
+      /* this is weird, a window w/out a colormap!? */
+      /* OK, let's just allocate a new one and hope for the best */
+      cmap = XCreateColormap(v->display, w, attr.visual, AllocNone);
+   }
+#endif
+
+   b = create_xmesa_buffer((XMesaDrawable) w, WINDOW, v, cmap);
+   if (!b)
+      return NULL;
+
+   if (!initialize_visual_and_buffer( v, b, v->mesa_visual.rgbMode,
+                                      (XMesaDrawable) w, cmap )) {
+      xmesa_free_buffer(b);
+      return NULL;
+   }
+
+   return b;
+}
+
+
+
+/**
+ * Create a new XMesaBuffer from an X pixmap.
+ *
+ * \param v    the XMesaVisual
+ * \param p    the pixmap
+ * \param cmap the colormap, may be 0 if using a \c GLX_TRUE_COLOR or
+ *             \c GLX_DIRECT_COLOR visual for the pixmap
+ * \returns new XMesaBuffer or NULL if error
+ */
+PUBLIC XMesaBuffer
+XMesaCreatePixmapBuffer(XMesaVisual v, XMesaPixmap p, XMesaColormap cmap)
+{
+   XMesaBuffer b;
+
+   assert(v);
+
+   b = create_xmesa_buffer((XMesaDrawable) p, PIXMAP, v, cmap);
+   if (!b)
+      return NULL;
+
+   if (!initialize_visual_and_buffer(v, b, v->mesa_visual.rgbMode,
+				     (XMesaDrawable) p, cmap)) {
+      xmesa_free_buffer(b);
+      return NULL;
+   }
+
+   return b;
+}
+
+
+/**
+ * For GLX_EXT_texture_from_pixmap
+ */
+XMesaBuffer
+XMesaCreatePixmapTextureBuffer(XMesaVisual v, XMesaPixmap p,
+                               XMesaColormap cmap,
+                               int format, int target, int mipmap)
+{
+   GET_CURRENT_CONTEXT(ctx);
+   XMesaBuffer b;
+   GLuint width, height;
+
+   assert(v);
+
+   b = create_xmesa_buffer((XMesaDrawable) p, PIXMAP, v, cmap);
+   if (!b)
+      return NULL;
+
+   /* get pixmap size, update framebuffer/renderbuffer dims */
+   xmesa_get_window_size(v->display, b, &width, &height);
+   _mesa_resize_framebuffer(NULL, &(b->stfb->Base), width, height);
+
+   if (target == 0) {
+      /* examine dims */
+      if (ctx->Extensions.ARB_texture_non_power_of_two) {
+         target = GLX_TEXTURE_2D_EXT;
+      }
+      else if (   _mesa_bitcount(width)  == 1
+               && _mesa_bitcount(height) == 1) {
+         /* power of two size */
+         if (height == 1) {
+            target = GLX_TEXTURE_1D_EXT;
+         }
+         else {
+            target = GLX_TEXTURE_2D_EXT;
+         }
+      }
+      else if (ctx->Extensions.NV_texture_rectangle) {
+         target = GLX_TEXTURE_RECTANGLE_EXT;
+      }
+      else {
+         /* non power of two textures not supported */
+         XMesaDestroyBuffer(b);
+         return 0;
+      }
+   }
+
+   b->TextureTarget = target;
+   b->TextureFormat = format;
+   b->TextureMipmap = mipmap;
+
+   if (!initialize_visual_and_buffer(v, b, v->mesa_visual.rgbMode,
+				     (XMesaDrawable) p, cmap)) {
+      xmesa_free_buffer(b);
+      return NULL;
+   }
+
+   return b;
+}
+
+
+
+XMesaBuffer
+XMesaCreatePBuffer(XMesaVisual v, XMesaColormap cmap,
+                   unsigned int width, unsigned int height)
+{
+#ifndef XFree86Server
+   XMesaWindow root;
+   XMesaDrawable drawable;  /* X Pixmap Drawable */
+   XMesaBuffer b;
+
+   /* allocate pixmap for front buffer */
+   root = RootWindow( v->display, v->visinfo->screen );
+   drawable = XCreatePixmap(v->display, root, width, height,
+                            v->visinfo->depth);
+   if (!drawable)
+      return NULL;
+
+   b = create_xmesa_buffer(drawable, PBUFFER, v, cmap);
+   if (!b)
+      return NULL;
+
+   if (!initialize_visual_and_buffer(v, b, v->mesa_visual.rgbMode,
+				     drawable, cmap)) {
+      xmesa_free_buffer(b);
+      return NULL;
+   }
+
+   return b;
+#else
+   return 0;
+#endif
+}
+
+
+
+/*
+ * Deallocate an XMesaBuffer structure and all related info.
+ */
+PUBLIC void
+XMesaDestroyBuffer(XMesaBuffer b)
+{
+   xmesa_free_buffer(b);
+}
+
+
+/**
+ * Query the current window size and update the corresponding GLframebuffer
+ * and all attached renderbuffers.
+ * Called when:
+ *  1. the first time a buffer is bound to a context.
+ *  2. from the XMesaResizeBuffers() API function.
+ *  3. SwapBuffers.  XXX probabaly from xm_flush_frontbuffer() too...
+ * Note: it's possible (and legal) for xmctx to be NULL.  That can happen
+ * when resizing a buffer when no rendering context is bound.
+ */
+void
+xmesa_check_and_update_buffer_size(XMesaContext xmctx, XMesaBuffer drawBuffer)
+{
+   GLuint width, height;
+   xmesa_get_window_size(drawBuffer->xm_visual->display, drawBuffer, &width, &height);
+   st_resize_framebuffer(drawBuffer->stfb, width, height);
+}
+
+
+/*
+ * Bind buffer b to context c and make c the current rendering context.
+ */
+GLboolean XMesaMakeCurrent( XMesaContext c, XMesaBuffer b )
+{
+   return XMesaMakeCurrent2( c, b, b );
+}
+
+
+/*
+ * Bind buffer b to context c and make c the current rendering context.
+ */
+PUBLIC
+GLboolean XMesaMakeCurrent2( XMesaContext c, XMesaBuffer drawBuffer,
+                             XMesaBuffer readBuffer )
+{
+   if (c) {
+      if (!drawBuffer || !readBuffer)
+         return GL_FALSE;  /* must specify buffers! */
+
+#if 0
+      /* XXX restore this optimization */
+      if (&(c->mesa) == _mesa_get_current_context()
+          && c->mesa.DrawBuffer == &drawBuffer->mesa_buffer
+          && c->mesa.ReadBuffer == &readBuffer->mesa_buffer
+          && xmesa_buffer(c->mesa.DrawBuffer)->wasCurrent) {
+         /* same context and buffer, do nothing */
+         return GL_TRUE;
+      }
+#endif
+
+      c->xm_buffer = drawBuffer;
+
+      /* Call this periodically to detect when the user has begun using
+       * GL rendering from multiple threads.
+       */
+      _glapi_check_multithread();
+
+      st_make_current(c->st, drawBuffer->stfb, readBuffer->stfb);
+
+      xmesa_check_and_update_buffer_size(c, drawBuffer);
+      if (readBuffer != drawBuffer)
+         xmesa_check_and_update_buffer_size(c, readBuffer);
+
+      /* Solution to Stephane Rehel's problem with glXReleaseBuffersMESA(): */
+      drawBuffer->wasCurrent = GL_TRUE;
+   }
+   else {
+      /* Detach */
+      st_make_current( NULL, NULL, NULL );
+   }
+   return GL_TRUE;
+}
+
+
+/*
+ * Unbind the context c from its buffer.
+ */
+GLboolean XMesaUnbindContext( XMesaContext c )
+{
+   /* A no-op for XFree86 integration purposes */
+   return GL_TRUE;
+}
+
+
+XMesaContext XMesaGetCurrentContext( void )
+{
+   GET_CURRENT_CONTEXT(ctx);
+   if (ctx) {
+      XMesaContext xmesa = xmesa_context(ctx);
+      return xmesa;
+   }
+   else {
+      return 0;
+   }
+}
+
+
+XMesaBuffer XMesaGetCurrentBuffer( void )
+{
+   GET_CURRENT_CONTEXT(ctx);
+   if (ctx) {
+      XMesaBuffer xmbuf = xmesa_buffer(ctx->DrawBuffer);
+      return xmbuf;
+   }
+   else {
+      return 0;
+   }
+}
+
+
+/* New in Mesa 3.1 */
+XMesaBuffer XMesaGetCurrentReadBuffer( void )
+{
+   GET_CURRENT_CONTEXT(ctx);
+   if (ctx) {
+      return xmesa_buffer(ctx->ReadBuffer);
+   }
+   else {
+      return 0;
+   }
+}
+
+
+#ifdef XFree86Server
+PUBLIC
+GLboolean XMesaForceCurrent(XMesaContext c)
+{
+   if (c) {
+      _glapi_set_dispatch(c->mesa.CurrentDispatch);
+
+      if (&(c->mesa) != _mesa_get_current_context()) {
+	 _mesa_make_current(&c->mesa, c->mesa.DrawBuffer, c->mesa.ReadBuffer);
+      }
+   }
+   else {
+      _mesa_make_current(NULL, NULL, NULL);
+   }
+   return GL_TRUE;
+}
+
+
+PUBLIC
+GLboolean XMesaLoseCurrent(XMesaContext c)
+{
+   (void) c;
+   _mesa_make_current(NULL, NULL, NULL);
+   return GL_TRUE;
+}
+
+
+PUBLIC
+GLboolean XMesaCopyContext( XMesaContext xm_src, XMesaContext xm_dst, GLuint mask )
+{
+   _mesa_copy_context(&xm_src->mesa, &xm_dst->mesa, mask);
+   return GL_TRUE;
+}
+#endif /* XFree86Server */
+
+
+#ifndef FX
+GLboolean XMesaSetFXmode( GLint mode )
+{
+   (void) mode;
+   return GL_FALSE;
+}
+#endif
+
+
+
+/*
+ * Copy the back buffer to the front buffer.  If there's no back buffer
+ * this is a no-op.
+ */
+PUBLIC
+void XMesaSwapBuffers( XMesaBuffer b )
+{
+   struct pipe_surface *surf;
+
+   /* If we're swapping the buffer associated with the current context
+    * we have to flush any pending rendering commands first.
+    */
+   st_notify_swapbuffers(b->stfb);
+
+   surf = st_get_framebuffer_surface(b->stfb, ST_SURFACE_BACK_LEFT);
+   if (surf) {
+      if (xmesa_mode == XMESA_AUB)
+         xmesa_display_aub( surf );
+      else
+	 xmesa_display_surface(b, surf);
+   }
+
+   xmesa_check_and_update_buffer_size(NULL, b);
+}
+
+
+
+/*
+ * Copy sub-region of back buffer to front buffer
+ */
+void XMesaCopySubBuffer( XMesaBuffer b, int x, int y, int width, int height )
+{
+   struct pipe_surface *surf_front
+      = st_get_framebuffer_surface(b->stfb, ST_SURFACE_FRONT_LEFT);
+   struct pipe_surface *surf_back
+      = st_get_framebuffer_surface(b->stfb, ST_SURFACE_BACK_LEFT);
+   struct pipe_context *pipe = NULL; /* XXX fix */
+
+   if (!surf_front || !surf_back)
+      return;
+
+   pipe->surface_copy(pipe,
+                      FALSE,
+                      surf_front, x, y,  /* dest */
+                      surf_back, x, y,   /* src */
+                      width, height);
+}
+
+
+
+/*
+ * Return the depth buffer associated with an XMesaBuffer.
+ * Input:  b - the XMesa buffer handle
+ * Output:  width, height - size of buffer in pixels
+ *          bytesPerValue - bytes per depth value (2 or 4)
+ *          buffer - pointer to depth buffer values
+ * Return:  GL_TRUE or GL_FALSE to indicate success or failure.
+ */
+GLboolean XMesaGetDepthBuffer( XMesaBuffer b, GLint *width, GLint *height,
+                               GLint *bytesPerValue, void **buffer )
+{
+   *width = 0;
+   *height = 0;
+   *bytesPerValue = 0;
+   *buffer = 0;
+   return GL_FALSE;
+}
+
+
+void XMesaFlush( XMesaContext c )
+{
+   if (c && c->xm_visual->display) {
+#ifdef XFree86Server
+      /* NOT_NEEDED */
+#else
+      XSync( c->xm_visual->display, False );
+#endif
+   }
+}
+
+
+
+const char *XMesaGetString( XMesaContext c, int name )
+{
+   (void) c;
+   if (name==XMESA_VERSION) {
+      return "5.0";
+   }
+   else if (name==XMESA_EXTENSIONS) {
+      return "";
+   }
+   else {
+      return NULL;
+   }
+}
+
+
+
+XMesaBuffer XMesaFindBuffer( XMesaDisplay *dpy, XMesaDrawable d )
+{
+   XMesaBuffer b;
+   for (b=XMesaBufferList; b; b=b->Next) {
+      if (b->drawable == d && b->xm_visual->display == dpy) {
+         return b;
+      }
+   }
+   return NULL;
+}
+
+
+/**
+ * Free/destroy all XMesaBuffers associated with given display.
+ */
+void xmesa_destroy_buffers_on_display(XMesaDisplay *dpy)
+{
+   XMesaBuffer b, next;
+   for (b = XMesaBufferList; b; b = next) {
+      next = b->Next;
+      if (b->xm_visual->display == dpy) {
+         xmesa_free_buffer(b);
+      }
+   }
+}
+
+
+/*
+ * Look for XMesaBuffers whose X window has been destroyed.
+ * Deallocate any such XMesaBuffers.
+ */
+void XMesaGarbageCollect( void )
+{
+   XMesaBuffer b, next;
+   for (b=XMesaBufferList; b; b=next) {
+      next = b->Next;
+      if (b->xm_visual &&
+          b->xm_visual->display &&
+          b->drawable &&
+          b->type == WINDOW) {
+#ifdef XFree86Server
+	 /* NOT_NEEDED */
+#else
+         XSync(b->xm_visual->display, False);
+         if (!window_exists( b->xm_visual->display, b->drawable )) {
+            /* found a dead window, free the ancillary info */
+            XMesaDestroyBuffer( b );
+         }
+#endif
+      }
+   }
+}
+
+
+unsigned long XMesaDitherColor( XMesaContext xmesa, GLint x, GLint y,
+                                GLfloat red, GLfloat green,
+                                GLfloat blue, GLfloat alpha )
+{
+   /* no longer supported */
+   return 0;
+}
+
+
+/*
+ * This is typically called when the window size changes and we need
+ * to reallocate the buffer's back/depth/stencil/accum buffers.
+ */
+PUBLIC void
+XMesaResizeBuffers( XMesaBuffer b )
+{
+   GET_CURRENT_CONTEXT(ctx);
+   XMesaContext xmctx = xmesa_context(ctx);
+   if (!xmctx)
+      return;
+   xmesa_check_and_update_buffer_size(xmctx, b);
+}
+
+
+
+
+PUBLIC void
+XMesaBindTexImage(XMesaDisplay *dpy, XMesaBuffer drawable, int buffer,
+                  const int *attrib_list)
+{
+}
+
+
+
+PUBLIC void
+XMesaReleaseTexImage(XMesaDisplay *dpy, XMesaBuffer drawable, int buffer)
+{
+}
+
diff --git a/src/gallium/winsys/xlib/xm_image.c b/src/gallium/winsys/xlib/xm_image.c
new file mode 100644
index 0000000000..087b4e4c3a
--- /dev/null
+++ b/src/gallium/winsys/xlib/xm_image.c
@@ -0,0 +1,133 @@
+/**************************************************************************
+
+Copyright 1998-1999 Precision Insight, Inc., Cedar Park, Texas.
+All Rights Reserved.
+
+Permission is hereby granted, free of charge, to any person obtaining a
+copy of this software and associated documentation files (the
+"Software"), to deal in the Software without restriction, including
+without limitation the rights to use, copy, modify, merge, publish,
+distribute, sub license, and/or sell copies of the Software, and to
+permit persons to whom the Software is furnished to do so, subject to
+the following conditions:
+
+The above copyright notice and this permission notice (including the
+next paragraph) shall be included in all copies or substantial portions
+of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+IN NO EVENT SHALL PRECISION INSIGHT AND/OR ITS SUPPLIERS BE LIABLE FOR
+ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+
+**************************************************************************/
+
+/*
+ * Authors:
+ *   Kevin E. Martin <kevin@precisioninsight.com>
+ *   Brian Paul <brian@precisioninsight.com>
+ */
+
+#include <stdlib.h>
+#include <X11/Xmd.h>
+
+#include "glxheader.h"
+#include "xmesaP.h"
+
+#ifdef XFree86Server
+
+#ifdef ROUNDUP
+#undef ROUNDUP
+#endif
+
+#define ROUNDUP(nbytes, pad) ((((nbytes) + ((pad)-1)) / (pad)) * ((pad)>>3))
+
+XMesaImage *XMesaCreateImage(int bitsPerPixel, int width, int height, char *data)
+{
+    XMesaImage *image;
+
+    image = (XMesaImage *)xalloc(sizeof(XMesaImage));
+
+    if (image) {
+	image->width = width;
+	image->height = height;
+	image->data = data;
+	/* Always pad to 32 bits */
+	image->bytes_per_line = ROUNDUP((bitsPerPixel * width), 32);
+	image->bits_per_pixel = bitsPerPixel;
+    }
+
+    return image;
+}
+
+void XMesaDestroyImage(XMesaImage *image)
+{
+    if (image->data)
+	free(image->data);
+    xfree(image);
+}
+
+unsigned long XMesaGetPixel(XMesaImage *image, int x, int y)
+{
+    CARD8  *row = (CARD8 *)(image->data + y*image->bytes_per_line);
+    CARD8  *i8;
+    CARD16 *i16;
+    CARD32 *i32;
+    switch (image->bits_per_pixel) {
+    case 8:
+	i8 = (CARD8 *)row;
+	return i8[x];
+	break;
+    case 15:
+    case 16:
+	i16 = (CARD16 *)row;
+	return i16[x];
+	break;
+    case 24: /* WARNING: architecture specific code */
+	i8 = (CARD8 *)row;
+	return (((CARD32)i8[x*3]) |
+		(((CARD32)i8[x*3+1])<<8) |
+		(((CARD32)i8[x*3+2])<<16));
+	break;
+    case 32:
+	i32 = (CARD32 *)row;
+	return i32[x];
+	break;
+    }
+    return 0;
+}
+
+#ifndef XMESA_USE_PUTPIXEL_MACRO
+void XMesaPutPixel(XMesaImage *image, int x, int y, unsigned long pixel)
+{
+    CARD8  *row = (CARD8 *)(image->data + y*image->bytes_per_line);
+    CARD8  *i8;
+    CARD16 *i16;
+    CARD32 *i32;
+    switch (image->bits_per_pixel) {
+    case 8:
+	i8 = (CARD8 *)row;
+	i8[x] = (CARD8)pixel;
+	break;
+    case 15:
+    case 16:
+	i16 = (CARD16 *)row;
+	i16[x] = (CARD16)pixel;
+	break;
+    case 24: /* WARNING: architecture specific code */
+	i8 = (CARD8 *)__row;
+	i8[x*3]   = (CARD8)(p);
+	i8[x*3+1] = (CARD8)(p>>8);
+	i8[x*3+2] = (CARD8)(p>>16);
+    case 32:
+	i32 = (CARD32 *)row;
+	i32[x] = (CARD32)pixel;
+	break;
+    }
+}
+#endif
+
+#endif /* XFree86Server */
diff --git a/src/gallium/winsys/xlib/xm_image.h b/src/gallium/winsys/xlib/xm_image.h
new file mode 100644
index 0000000000..2a5e0f3777
--- /dev/null
+++ b/src/gallium/winsys/xlib/xm_image.h
@@ -0,0 +1,77 @@
+/**************************************************************************
+
+Copyright 1998-1999 Precision Insight, Inc., Cedar Park, Texas.
+All Rights Reserved.
+
+Permission is hereby granted, free of charge, to any person obtaining a
+copy of this software and associated documentation files (the
+"Software"), to deal in the Software without restriction, including
+without limitation the rights to use, copy, modify, merge, publish,
+distribute, sub license, and/or sell copies of the Software, and to
+permit persons to whom the Software is furnished to do so, subject to
+the following conditions:
+
+The above copyright notice and this permission notice (including the
+next paragraph) shall be included in all copies or substantial portions
+of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+IN NO EVENT SHALL PRECISION INSIGHT AND/OR ITS SUPPLIERS BE LIABLE FOR
+ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+
+**************************************************************************/
+
+/*
+ * Authors:
+ *   Kevin E. Martin <kevin@precisioninsight.com>
+ *   Brian Paul <brian@precisioninsight.com>
+ */
+
+#ifndef _XM_IMAGE_H_
+#define _XM_IMAGE_H_
+
+#define XMESA_USE_PUTPIXEL_MACRO
+
+extern XMesaImage *XMesaCreateImage(int bitsPerPixel, int width, int height,
+				    char *data);
+extern void XMesaDestroyImage(XMesaImage *image);
+extern unsigned long XMesaGetPixel(XMesaImage *image, int x, int y);
+#ifdef XMESA_USE_PUTPIXEL_MACRO
+#define XMesaPutPixel(__i,__x,__y,__p) \
+{ \
+    CARD8  *__row = (CARD8 *)(__i->data + __y*__i->bytes_per_line); \
+    CARD8  *__i8; \
+    CARD16 *__i16; \
+    CARD32 *__i32; \
+    switch (__i->bits_per_pixel) { \
+    case 8: \
+	__i8 = (CARD8 *)__row; \
+	__i8[__x] = (CARD8)__p; \
+	break; \
+    case 15: \
+    case 16: \
+	__i16 = (CARD16 *)__row; \
+	__i16[__x] = (CARD16)__p; \
+	break; \
+    case 24: /* WARNING: architecture specific code */ \
+	__i8 = (CARD8 *)__row; \
+	__i8[__x*3]   = (CARD8)(__p); \
+	__i8[__x*3+1] = (CARD8)(__p>>8); \
+	__i8[__x*3+2] = (CARD8)(__p>>16); \
+	break; \
+    case 32: \
+	__i32 = (CARD32 *)__row; \
+	__i32[__x] = (CARD32)__p; \
+	break; \
+    } \
+}
+#else
+extern void XMesaPutPixel(XMesaImage *image, int x, int y,
+			  unsigned long pixel);
+#endif
+
+#endif /* _XM_IMAGE_H_ */
diff --git a/src/gallium/winsys/xlib/xm_winsys.c b/src/gallium/winsys/xlib/xm_winsys.c
new file mode 100644
index 0000000000..c3cd22eea3
--- /dev/null
+++ b/src/gallium/winsys/xlib/xm_winsys.c
@@ -0,0 +1,466 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Bismarck, ND., USA
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDERS, AUTHORS AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM,
+ * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR 
+ * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE 
+ * USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * 
+ **************************************************************************/
+
+/*
+ * Authors:
+ *   Keith Whitwell
+ *   Brian Paul
+ */
+
+
+#include "glxheader.h"
+#include "xmesaP.h"
+
+#include "pipe/p_winsys.h"
+#include "pipe/p_format.h"
+#include "pipe/p_context.h"
+#include "pipe/p_util.h"
+#include "pipe/p_inlines.h"
+#include "pipe/softpipe/sp_winsys.h"
+
+#ifdef GALLIUM_CELL
+#include "pipe/cell/ppu/cell_context.h"
+#include "pipe/cell/ppu/cell_winsys.h"
+#else
+#define TILE_SIZE 32  /* avoid compilation errors */
+#endif
+
+#include "xm_winsys_aub.h"
+
+
+/**
+ * Low-level OS/window system memory buffer
+ */
+struct xm_buffer
+{
+   struct pipe_buffer base;
+   boolean userBuffer;  /** Is this a user-space buffer? */
+   void *data;
+   void *mapped;
+};
+
+
+struct xmesa_surface
+{
+   struct pipe_surface surface;
+
+   int tileSize;
+};
+
+
+/**
+ * Derived from softpipe_winsys.
+ * We just need one extra field which indicates the pixel format to use for
+ * drawing surfaces so that we're compatible with the XVisual/window format.
+ */
+struct xmesa_softpipe_winsys
+{
+   struct softpipe_winsys spws;
+   enum pipe_format pixelformat;
+};
+
+
+
+/** Cast wrapper */
+static INLINE struct xmesa_surface *
+xmesa_surface(struct pipe_surface *ps)
+{
+//   assert(0);
+   return (struct xmesa_surface *) ps;
+}
+
+/** cast wrapper */
+static INLINE struct xmesa_softpipe_winsys *
+xmesa_softpipe_winsys(struct softpipe_winsys *spws)
+{
+   return (struct xmesa_softpipe_winsys *) spws;
+}
+
+/**
+ * Turn the softpipe opaque buffer pointer into a dri_bufmgr opaque
+ * buffer pointer...
+ */
+static INLINE struct xm_buffer *
+xm_buffer( struct pipe_buffer *buf )
+{
+   return (struct xm_buffer *)buf;
+}
+
+
+
+/* Most callbacks map direcly onto dri_bufmgr operations:
+ */
+static void *
+xm_buffer_map(struct pipe_winsys *pws, struct pipe_buffer *buf,
+              unsigned flags)
+{
+   struct xm_buffer *xm_buf = xm_buffer(buf);
+   xm_buf->mapped = xm_buf->data;
+   return xm_buf->mapped;
+}
+
+static void
+xm_buffer_unmap(struct pipe_winsys *pws, struct pipe_buffer *buf)
+{
+   struct xm_buffer *xm_buf = xm_buffer(buf);
+   xm_buf->mapped = NULL;
+}
+
+static void
+xm_buffer_destroy(struct pipe_winsys *pws,
+		  struct pipe_buffer *buf)
+{
+   struct xm_buffer *oldBuf = xm_buffer(buf);
+
+   if (oldBuf->data) {
+      if (!oldBuf->userBuffer)
+	 align_free(oldBuf->data);
+      oldBuf->data = NULL;
+   }
+
+   free(oldBuf);
+}
+
+
+/**
+ * Display a surface that's in a tiled configuration.  That is, all the
+ * pixels for a TILE_SIZExTILE_SIZE block are contiguous in memory.
+ */
+static void
+xmesa_display_surface_tiled(XMesaBuffer b, const struct pipe_surface *surf)
+{
+   XImage *ximage = b->tempImage;
+   struct xm_buffer *xm_buf = xm_buffer(surf->buffer);
+   const uint tilesPerRow = (surf->width + TILE_SIZE - 1) / TILE_SIZE;
+   uint x, y;
+
+   /* check that the XImage has been previously initialized */
+   assert(ximage->format);
+   assert(ximage->bitmap_unit);
+
+   /* update XImage's fields */
+   ximage->width = TILE_SIZE;
+   ximage->height = TILE_SIZE;
+   ximage->bytes_per_line = TILE_SIZE * 4;
+
+   for (y = 0; y < surf->height; y += TILE_SIZE) {
+      for (x = 0; x < surf->width; x += TILE_SIZE) {
+         int dx = x;
+         int dy = y;
+         int tx = x / TILE_SIZE;
+         int ty = y / TILE_SIZE;
+         int offset = ty * tilesPerRow + tx;
+
+         offset *= 4 * TILE_SIZE * TILE_SIZE;
+
+         ximage->data = (char *) xm_buf->data + offset;
+
+         XPutImage(b->xm_visual->display, b->drawable, b->gc,
+                   ximage, 0, 0, dx, dy, TILE_SIZE, TILE_SIZE);
+      }
+   }
+}
+
+
+/**
+ * Display/copy the image in the surface into the X window specified
+ * by the XMesaBuffer.
+ */
+void
+xmesa_display_surface(XMesaBuffer b, const struct pipe_surface *surf)
+{
+   XImage *ximage = b->tempImage;
+   struct xm_buffer *xm_buf = xm_buffer(surf->buffer);
+   const struct xmesa_surface *xm_surf
+      = xmesa_surface((struct pipe_surface *) surf);
+
+   if (xm_surf->tileSize) {
+      xmesa_display_surface_tiled(b, surf);
+      return;
+   }
+
+   /* check that the XImage has been previously initialized */
+   assert(ximage->format);
+   assert(ximage->bitmap_unit);
+
+   /* update XImage's fields */
+   ximage->width = surf->width;
+   ximage->height = surf->height;
+   ximage->bytes_per_line = surf->pitch * (ximage->bits_per_pixel / 8);
+   ximage->data = xm_buf->data;
+
+   /* display image in Window */
+   XPutImage(b->xm_visual->display, b->drawable, b->gc,
+             ximage, 0, 0, 0, 0, surf->width, surf->height);
+}
+
+
+static void
+xm_flush_frontbuffer(struct pipe_winsys *pws,
+                     struct pipe_surface *surf,
+                     void *context_private)
+{
+   /* The Xlib driver's front color surfaces are actually X Windows so
+    * this flush is a no-op.
+    * If we instead did front buffer rendering to a temporary XImage,
+    * this would be the place to copy the Ximage to the on-screen Window.
+    */
+   XMesaContext xmctx = (XMesaContext) context_private;
+   xmesa_display_surface(xmctx->xm_buffer, surf);
+}
+
+
+
+static void
+xm_printf(struct pipe_winsys *pws, const char *fmtString, ...)
+{
+   va_list args;
+   va_start( args, fmtString );  
+   vfprintf(stderr, fmtString, args);
+   va_end( args );
+}
+
+
+static const char *
+xm_get_name(struct pipe_winsys *pws)
+{
+   return "Xlib";
+}
+
+
+static struct pipe_buffer *
+xm_buffer_create(struct pipe_winsys *pws, 
+                 unsigned alignment, 
+                 unsigned usage,
+                 unsigned size)
+{
+   struct xm_buffer *buffer = CALLOC_STRUCT(xm_buffer);
+   buffer->base.refcount = 1;
+   buffer->base.alignment = alignment;
+   buffer->base.usage = usage;
+   buffer->base.size = size;
+
+   /* align to 16-byte multiple for Cell */
+   buffer->data = align_malloc(size, max(alignment, 16));
+
+   return &buffer->base;
+}
+
+
+/**
+ * Create buffer which wraps user-space data.
+ */
+static struct pipe_buffer *
+xm_user_buffer_create(struct pipe_winsys *pws, void *ptr, unsigned bytes)
+{
+   struct xm_buffer *buffer = CALLOC_STRUCT(xm_buffer);
+   buffer->base.refcount = 1;
+   buffer->base.size = bytes;
+   buffer->userBuffer = TRUE;
+   buffer->data = ptr;
+
+   return &buffer->base;
+}
+
+
+
+/**
+ * Round n up to next multiple.
+ */
+static INLINE unsigned
+round_up(unsigned n, unsigned multiple)
+{
+   return (n + multiple - 1) & ~(multiple - 1);
+}
+
+static int
+xm_surface_alloc_storage(struct pipe_winsys *winsys,
+                         struct pipe_surface *surf,
+                         unsigned width, unsigned height,
+                         enum pipe_format format, 
+                         unsigned flags)
+{
+   const unsigned alignment = 64;
+
+   surf->width = width;
+   surf->height = height;
+   surf->format = format;
+   surf->cpp = pf_get_size(format);
+   surf->pitch = round_up(width, alignment / surf->cpp);
+
+#ifdef GALLIUM_CELL /* XXX a bit of a hack */
+   height = round_up(height, TILE_SIZE);
+#endif
+
+   assert(!surf->buffer);
+   surf->buffer = winsys->buffer_create(winsys, alignment,
+                                        PIPE_BUFFER_USAGE_PIXEL,
+                                        surf->pitch * surf->cpp * height);
+   if(!surf->buffer)
+      return -1;
+   
+   return 0;
+}
+
+
+/**
+ * Called via pipe->surface_alloc() to create new surfaces (textures,
+ * renderbuffers, etc.
+ */
+static struct pipe_surface *
+xm_surface_alloc(struct pipe_winsys *ws)
+{
+   struct xmesa_surface *xms = CALLOC_STRUCT(xmesa_surface);
+
+   assert(ws);
+
+   xms->surface.refcount = 1;
+   xms->surface.winsys = ws;
+
+#ifdef GALLIUM_CELL
+   if (!getenv("GALLIUM_NOCELL")) {
+      xms->tileSize = 32; /** probably temporary */
+   }
+#endif
+
+   return &xms->surface;
+}
+
+
+
+static void
+xm_surface_release(struct pipe_winsys *winsys, struct pipe_surface **s)
+{
+   struct pipe_surface *surf = *s;
+   surf->refcount--;
+   if (surf->refcount == 0) {
+      if (surf->buffer)
+	pipe_buffer_reference(winsys, &surf->buffer, NULL);
+      free(surf);
+   }
+   *s = NULL;
+}
+
+
+
+/**
+ * Return pointer to a pipe_winsys object.
+ * For Xlib, this is a singleton object.
+ * Nothing special for the Xlib driver so no subclassing or anything.
+ */
+struct pipe_winsys *
+xmesa_get_pipe_winsys_aub(void)
+{
+   static struct pipe_winsys *ws = NULL;
+
+   if (!ws && getenv("XM_AUB")) {
+      ws = xmesa_create_pipe_winsys_aub();
+   }
+   else if (!ws) {
+      ws = CALLOC_STRUCT(pipe_winsys);
+   
+      /* Fill in this struct with callbacks that pipe will need to
+       * communicate with the window system, buffer manager, etc. 
+       */
+      ws->buffer_create = xm_buffer_create;
+      ws->user_buffer_create = xm_user_buffer_create;
+      ws->buffer_map = xm_buffer_map;
+      ws->buffer_unmap = xm_buffer_unmap;
+      ws->buffer_destroy = xm_buffer_destroy;
+
+      ws->surface_alloc = xm_surface_alloc;
+      ws->surface_alloc_storage = xm_surface_alloc_storage;
+      ws->surface_release = xm_surface_release;
+
+      ws->flush_frontbuffer = xm_flush_frontbuffer;
+      ws->printf = xm_printf;
+      ws->get_name = xm_get_name;
+   }
+
+   return ws;
+}
+
+
+/**
+ * Called via softpipe_winsys->is_format_supported().
+ * This function is only called to test formats for front/back color surfaces.
+ * The winsys being queried will have been created at glXCreateContext
+ * time, with a pixel format corresponding to the context's visual.
+ */
+static boolean
+xmesa_is_format_supported(struct softpipe_winsys *sws,
+                          enum pipe_format format)
+{
+   struct xmesa_softpipe_winsys *xmws = xmesa_softpipe_winsys(sws);
+   return (format == xmws->pixelformat);
+}
+
+
+/**
+ * Return pointer to a softpipe_winsys object.
+ */
+static struct softpipe_winsys *
+xmesa_get_softpipe_winsys(uint pixelformat)
+{
+   struct xmesa_softpipe_winsys *xmws
+      = CALLOC_STRUCT(xmesa_softpipe_winsys);
+   if (!xmws)
+      return NULL;
+
+   xmws->spws.is_format_supported = xmesa_is_format_supported;
+   xmws->pixelformat = pixelformat;
+
+   return &xmws->spws;
+}
+
+
+struct pipe_context *
+xmesa_create_pipe_context(XMesaContext xmesa, uint pixelformat)
+{
+   struct pipe_winsys *pws = xmesa_get_pipe_winsys_aub();
+   struct pipe_context *pipe;
+   
+#ifdef GALLIUM_CELL
+   if (!getenv("GALLIUM_NOCELL")) {
+      struct cell_winsys *cws = cell_get_winsys(pixelformat);
+      pipe = cell_create_context(pws, cws);
+      if (pipe)
+         pipe->priv = xmesa;
+      return pipe;
+   }
+   else
+#endif
+   {
+      struct softpipe_winsys *spws = xmesa_get_softpipe_winsys(pixelformat);
+      pipe = softpipe_create( pws, spws );
+      if (pipe)
+         pipe->priv = xmesa;
+
+      return pipe;
+   }
+}
diff --git a/src/gallium/winsys/xlib/xm_winsys_aub.c b/src/gallium/winsys/xlib/xm_winsys_aub.c
new file mode 100644
index 0000000000..bf41570257
--- /dev/null
+++ b/src/gallium/winsys/xlib/xm_winsys_aub.c
@@ -0,0 +1,589 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Bismarck, ND., USA
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDERS, AUTHORS AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM,
+ * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR 
+ * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE 
+ * USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * 
+ **************************************************************************/
+
+/*
+ * Authors:
+ *   Keith Whitwell
+ *   Brian Paul
+ */
+
+
+#include "glxheader.h"
+#include "xmesaP.h"
+
+#include "pipe/p_winsys.h"
+#include "pipe/p_util.h"
+#include "pipe/p_inlines.h"
+#include "pipe/i965simple/brw_winsys.h"
+#include "brw_aub.h"
+#include "xm_winsys_aub.h"
+
+
+
+struct aub_buffer {
+   char *data;
+   unsigned offset;
+   unsigned size;
+   unsigned refcount;
+   unsigned map_count;
+   boolean dump_on_unmap;
+};
+
+
+
+struct aub_pipe_winsys {
+   struct pipe_winsys winsys;
+
+   struct brw_aubfile *aubfile;
+
+   /* This is simple, isn't it:
+    */
+   char *pool;
+   unsigned size;
+   unsigned used;
+};
+
+
+/* Turn a pipe winsys into an aub/pipe winsys:
+ */
+static inline struct aub_pipe_winsys *
+aub_pipe_winsys( struct pipe_winsys *winsys )
+{
+   return (struct aub_pipe_winsys *)winsys;
+}
+
+
+
+static INLINE struct aub_buffer *
+aub_bo( struct pipe_buffer *bo )
+{
+   return (struct aub_buffer *)bo;
+}
+
+static INLINE struct pipe_buffer *
+pipe_bo( struct aub_buffer *bo )
+{
+   return (struct pipe_buffer *)bo;
+}
+
+
+
+
+static void *aub_buffer_map(struct pipe_winsys *winsys, 
+			      struct pipe_buffer *buf,
+			      unsigned flags )
+{
+   struct aub_buffer *sbo = aub_bo(buf);
+
+   assert(sbo->data);
+
+   if (flags & PIPE_BUFFER_USAGE_CPU_WRITE)
+      sbo->dump_on_unmap = 1;
+
+   sbo->map_count++;
+   return sbo->data;
+}
+
+static void aub_buffer_unmap(struct pipe_winsys *winsys, 
+			       struct pipe_buffer *buf)
+{
+   struct aub_pipe_winsys *iws = aub_pipe_winsys(winsys);
+   struct aub_buffer *sbo = aub_bo(buf);
+
+   sbo->map_count--;
+
+   if (sbo->map_count == 0 &&
+       sbo->dump_on_unmap) {
+
+      sbo->dump_on_unmap = 0;
+
+      brw_aub_gtt_data( iws->aubfile, 
+			sbo->offset,
+			sbo->data,
+			sbo->size,
+			0,
+			0);
+   }
+}
+
+
+static void
+aub_buffer_destroy(struct pipe_winsys *winsys,
+		   struct pipe_buffer *buf)
+{
+   free(buf);
+}
+
+
+void xmesa_buffer_subdata_aub(struct pipe_winsys *winsys, 
+			      struct pipe_buffer *buf,
+			      unsigned long offset, 
+			      unsigned long size, 
+			      const void *data,
+			      unsigned aub_type,
+			      unsigned aub_sub_type)
+{
+   struct aub_pipe_winsys *iws = aub_pipe_winsys(winsys);
+   struct aub_buffer *sbo = aub_bo(buf);
+
+   assert(sbo->size > offset + size);
+   memcpy(sbo->data + offset, data, size);
+
+   brw_aub_gtt_data( iws->aubfile, 
+		     sbo->offset + offset,
+		     sbo->data + offset,
+		     size,
+		     aub_type,
+		     aub_sub_type );
+}
+
+void xmesa_commands_aub(struct pipe_winsys *winsys,
+			unsigned *cmds,
+			unsigned nr_dwords)
+{
+   struct aub_pipe_winsys *iws = aub_pipe_winsys(winsys);
+   unsigned size = nr_dwords * 4;
+
+   assert(iws->used + size < iws->size);
+
+   brw_aub_gtt_cmds( iws->aubfile, 
+		     AUB_BUF_START + iws->used,
+		     cmds,
+		     nr_dwords * sizeof(int) );
+
+   iws->used += align(size, 4096);
+}
+
+
+static struct aub_pipe_winsys *global_winsys = NULL;
+
+void xmesa_display_aub( /* struct pipe_winsys *winsys, */
+		       struct pipe_surface *surface )
+{
+//   struct aub_pipe_winsys *iws = aub_pipe_winsys(winsys);
+   brw_aub_dump_bmp( global_winsys->aubfile, 
+		     surface,
+		     aub_bo(surface->buffer)->offset );
+}
+
+
+
+/* Pipe has no concept of pools.  We choose the tex/region pool
+ * for all buffers.
+ */
+static struct pipe_buffer *
+aub_buffer_create(struct pipe_winsys *winsys,
+                  unsigned alignment,
+                  unsigned usage,
+                  unsigned size)
+{
+   struct aub_pipe_winsys *iws = aub_pipe_winsys(winsys);
+   struct aub_buffer *sbo = CALLOC_STRUCT(aub_buffer);
+
+   sbo->refcount = 1;
+
+   /* Could reuse buffers that are not referenced in current
+    * batchbuffer.  Can't do that atm, so always reallocate:
+    */
+   assert(iws->used + size < iws->size);
+   sbo->data = iws->pool + iws->used;
+   sbo->offset = AUB_BUF_START + iws->used;
+   iws->used += align(size, 4096);
+
+   sbo->size = size;
+
+   return pipe_bo(sbo);
+}
+
+
+static struct pipe_buffer *
+aub_user_buffer_create(struct pipe_winsys *winsys, void *ptr, unsigned bytes)
+{
+   struct aub_buffer *sbo;
+
+   /* Lets hope this is meant for upload, not as a result!  
+    */
+   sbo = aub_bo(aub_buffer_create( winsys, 0, 0, 0 ));
+
+   sbo->data = ptr;
+   sbo->size = bytes;
+
+   return pipe_bo(sbo);
+}
+
+
+/* The state tracker (should!) keep track of whether the fake
+ * frontbuffer has been touched by any rendering since the last time
+ * we copied its contents to the real frontbuffer.  Our task is easy:
+ */
+static void
+aub_flush_frontbuffer( struct pipe_winsys *winsys,
+                         struct pipe_surface *surf,
+                         void *context_private)
+{
+   xmesa_display_aub( surf );
+}
+
+static struct pipe_surface *
+aub_i915_surface_alloc(struct pipe_winsys *winsys)
+{
+   struct pipe_surface *surf = CALLOC_STRUCT(pipe_surface);
+   if (surf) {
+      surf->refcount = 1;
+      surf->winsys = winsys;
+   }
+   return surf;
+}
+
+
+/**
+ * Round n up to next multiple.
+ */
+static INLINE unsigned
+round_up(unsigned n, unsigned multiple)
+{
+   return (n + multiple - 1) & ~(multiple - 1);
+}
+
+static int
+aub_i915_surface_alloc_storage(struct pipe_winsys *winsys,
+                               struct pipe_surface *surf,
+                               unsigned width, unsigned height,
+                               enum pipe_format format,
+                               unsigned flags)
+{
+    const unsigned alignment = 64;
+
+    surf->width = width;
+    surf->height = height;
+    surf->format = format;
+    surf->cpp = pf_get_size(format);
+    surf->pitch = round_up(width, alignment / surf->cpp);
+
+    assert(!surf->buffer);
+    surf->buffer = winsys->buffer_create(winsys, alignment,
+                                         PIPE_BUFFER_USAGE_PIXEL,
+                                         surf->pitch * surf->cpp * height);
+    if(!surf->buffer)
+        return -1;
+
+    return 0;
+}
+
+static void
+aub_i915_surface_release(struct pipe_winsys *winsys, struct pipe_surface **s)
+{
+   struct pipe_surface *surf = *s;
+   surf->refcount--;
+   if (surf->refcount == 0) {
+      if (surf->buffer)
+         pipe_buffer_reference(winsys, &surf->buffer, NULL);
+      free(surf);
+   }
+   *s = NULL;
+}
+
+
+
+static void
+aub_printf( struct pipe_winsys *winsys, const char *fmtString, ... )
+{
+   va_list args;
+   va_start( args, fmtString );  
+   vfprintf(stderr, fmtString, args);
+   va_end( args );
+}
+
+static const char *
+aub_get_name( struct pipe_winsys *winsys )
+{
+   return "Aub/xlib";
+}
+
+struct pipe_winsys *
+xmesa_create_pipe_winsys_aub( void )
+{
+   struct aub_pipe_winsys *iws = CALLOC_STRUCT( aub_pipe_winsys );
+   
+   /* Fill in this struct with callbacks that pipe will need to
+    * communicate with the window system, buffer manager, etc. 
+    *
+    * Pipe would be happy with a malloc based memory manager, but
+    * the SwapBuffers implementation in this winsys driver requires
+    * that rendering be done to an appropriate _DriBufferObject.  
+    */
+   iws->winsys.buffer_create = aub_buffer_create;
+   iws->winsys.user_buffer_create = aub_user_buffer_create;
+   iws->winsys.buffer_map = aub_buffer_map;
+   iws->winsys.buffer_unmap = aub_buffer_unmap;
+   iws->winsys.buffer_destroy = aub_buffer_destroy;
+   iws->winsys.flush_frontbuffer = aub_flush_frontbuffer;
+   iws->winsys.printf = aub_printf;
+   iws->winsys.get_name = aub_get_name;
+
+   iws->winsys.surface_alloc = aub_i915_surface_alloc;
+   iws->winsys.surface_alloc_storage = aub_i915_surface_alloc_storage;
+   iws->winsys.surface_release = aub_i915_surface_release;
+
+   iws->aubfile = brw_aubfile_create();
+   iws->size = AUB_BUF_SIZE;
+   iws->pool = malloc(AUB_BUF_SIZE);
+
+   /* HACK: static copy of this pointer:
+    */
+   assert(global_winsys == NULL);
+   global_winsys = iws;
+
+   return &iws->winsys;
+}
+
+
+void
+xmesa_destroy_pipe_winsys_aub( struct pipe_winsys *winsys )
+
+{
+   struct aub_pipe_winsys *iws = aub_pipe_winsys(winsys);
+   brw_aub_destroy(iws->aubfile);
+   free(iws->pool);
+   free(iws);
+}
+
+
+
+
+
+
+
+#define IWS_BATCHBUFFER_SIZE 1024
+
+struct aub_brw_winsys {
+   struct brw_winsys winsys;   /**< batch buffer funcs */
+   struct aub_context *aub;
+                         
+   struct pipe_winsys *pipe_winsys;
+
+   unsigned batch_data[IWS_BATCHBUFFER_SIZE];
+   unsigned batch_nr;
+   unsigned batch_size;
+   unsigned batch_alloc;
+};
+
+
+/* Turn a i965simple winsys into an aub/i965simple winsys:
+ */
+static inline struct aub_brw_winsys *
+aub_brw_winsys( struct brw_winsys *sws )
+{
+   return (struct aub_brw_winsys *)sws;
+}
+
+
+/* Simple batchbuffer interface:
+ */
+
+static unsigned *aub_i965_batch_start( struct brw_winsys *sws,
+					 unsigned dwords,
+					 unsigned relocs )
+{
+   struct aub_brw_winsys *iws = aub_brw_winsys(sws);
+
+   if (iws->batch_size < iws->batch_nr + dwords)
+      return NULL;
+
+   iws->batch_alloc = iws->batch_nr + dwords;
+   return (void *)1;			/* not a valid pointer! */
+}
+
+static void aub_i965_batch_dword( struct brw_winsys *sws,
+				    unsigned dword )
+{
+   struct aub_brw_winsys *iws = aub_brw_winsys(sws);
+
+   assert(iws->batch_nr < iws->batch_alloc);
+   iws->batch_data[iws->batch_nr++] = dword;
+}
+
+static void aub_i965_batch_reloc( struct brw_winsys *sws,
+			     struct pipe_buffer *buf,
+			     unsigned access_flags,
+			     unsigned delta )
+{
+   struct aub_brw_winsys *iws = aub_brw_winsys(sws);
+
+   assert(iws->batch_nr < iws->batch_alloc);
+   iws->batch_data[iws->batch_nr++] = aub_bo(buf)->offset + delta;
+}
+
+static unsigned aub_i965_get_buffer_offset( struct brw_winsys *sws,
+					    struct pipe_buffer *buf,
+					    unsigned access_flags )
+{
+   return aub_bo(buf)->offset;
+}
+
+static void aub_i965_batch_end( struct brw_winsys *sws )
+{
+   struct aub_brw_winsys *iws = aub_brw_winsys(sws);
+
+   assert(iws->batch_nr <= iws->batch_alloc);
+   iws->batch_alloc = 0;
+}
+
+static void aub_i965_batch_flush( struct brw_winsys *sws,
+				    struct pipe_fence_handle **fence )
+{
+   struct aub_brw_winsys *iws = aub_brw_winsys(sws);
+   assert(iws->batch_nr <= iws->batch_size);
+
+   if (iws->batch_nr) {
+      xmesa_commands_aub( iws->pipe_winsys,
+			  iws->batch_data,
+			  iws->batch_nr );
+   }
+
+   iws->batch_nr = 0;
+}
+
+
+
+static void aub_i965_buffer_subdata_typed(struct brw_winsys *winsys, 
+					    struct pipe_buffer *buf,
+					    unsigned long offset, 
+					    unsigned long size, 
+					    const void *data,
+					    unsigned data_type)
+{
+   struct aub_brw_winsys *iws = aub_brw_winsys(winsys);
+   unsigned aub_type = DW_GENERAL_STATE;
+   unsigned aub_sub_type;
+
+   switch (data_type) {
+   case BRW_CC_VP:
+      aub_sub_type = DWGS_COLOR_CALC_VIEWPORT_STATE;
+      break;
+   case BRW_CC_UNIT:
+      aub_sub_type = DWGS_COLOR_CALC_STATE;
+      break;
+   case BRW_WM_PROG:
+      aub_sub_type = DWGS_KERNEL_INSTRUCTIONS;
+      break;
+   case BRW_SAMPLER_DEFAULT_COLOR:
+      aub_sub_type = DWGS_SAMPLER_DEFAULT_COLOR;
+      break;
+   case BRW_SAMPLER:
+      aub_sub_type = DWGS_SAMPLER_STATE;
+      break;
+   case BRW_WM_UNIT:
+      aub_sub_type = DWGS_WINDOWER_IZ_STATE;
+      break;
+   case BRW_SF_PROG:
+      aub_sub_type = DWGS_KERNEL_INSTRUCTIONS;
+      break;
+   case BRW_SF_VP:
+      aub_sub_type = DWGS_STRIPS_FANS_VIEWPORT_STATE;
+      break;
+   case BRW_SF_UNIT:
+      aub_sub_type = DWGS_STRIPS_FANS_STATE;
+      break;
+   case BRW_VS_UNIT:
+      aub_sub_type = DWGS_VERTEX_SHADER_STATE;
+      break;
+   case BRW_VS_PROG:
+      aub_sub_type = DWGS_KERNEL_INSTRUCTIONS;
+      break;
+   case BRW_GS_UNIT:
+      aub_sub_type = DWGS_GEOMETRY_SHADER_STATE;
+      break;
+   case BRW_GS_PROG:
+      aub_sub_type = DWGS_KERNEL_INSTRUCTIONS;
+      break;
+   case BRW_CLIP_VP:
+      aub_sub_type = DWGS_CLIPPER_VIEWPORT_STATE;
+      break;
+   case BRW_CLIP_UNIT:
+      aub_sub_type = DWGS_CLIPPER_STATE;
+      break;
+   case BRW_CLIP_PROG:
+      aub_sub_type = DWGS_KERNEL_INSTRUCTIONS;
+      break;
+   case BRW_SS_SURFACE:
+      aub_type = DW_SURFACE_STATE;
+      aub_sub_type = DWSS_SURFACE_STATE; 
+      break;
+   case BRW_SS_SURF_BIND:
+      aub_type = DW_SURFACE_STATE;
+      aub_sub_type = DWSS_BINDING_TABLE_STATE; 
+      break;
+   case BRW_CONSTANT_BUFFER:
+      aub_type = DW_CONSTANT_URB_ENTRY;
+      aub_sub_type = 0; 
+      break;
+
+   default:
+      assert(0);
+      break;
+   }
+
+   xmesa_buffer_subdata_aub( iws->pipe_winsys,
+			     buf,
+			     offset,
+			     size,
+			     data,
+			     aub_type,
+			     aub_sub_type );
+}
+   
+/**
+ * Create i965 hardware rendering context.
+ */
+struct pipe_context *
+xmesa_create_i965simple( struct pipe_winsys *winsys )
+{
+   struct aub_brw_winsys *iws = CALLOC_STRUCT( aub_brw_winsys );
+   
+   /* Fill in this struct with callbacks that i965simple will need to
+    * communicate with the window system, buffer manager, etc. 
+    */
+   iws->winsys.batch_start = aub_i965_batch_start;
+   iws->winsys.batch_dword = aub_i965_batch_dword;
+   iws->winsys.batch_reloc = aub_i965_batch_reloc;
+   iws->winsys.batch_end = aub_i965_batch_end;
+   iws->winsys.batch_flush = aub_i965_batch_flush;
+   iws->winsys.buffer_subdata_typed = aub_i965_buffer_subdata_typed;
+   iws->winsys.get_buffer_offset = aub_i965_get_buffer_offset;
+
+   iws->pipe_winsys = winsys;
+
+   iws->batch_size = IWS_BATCHBUFFER_SIZE;
+
+   /* Create the i965simple context:
+    */
+   return brw_create( winsys,
+		      &iws->winsys,
+		      0 );
+}
diff --git a/src/gallium/winsys/xlib/xm_winsys_aub.h b/src/gallium/winsys/xlib/xm_winsys_aub.h
new file mode 100644
index 0000000000..7bee199116
--- /dev/null
+++ b/src/gallium/winsys/xlib/xm_winsys_aub.h
@@ -0,0 +1,67 @@
+/**************************************************************************
+ * 
+ * Copyright 2006 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+#ifndef AUB_WINSYS_H
+#define AUB_WINSYS_H
+
+struct pipe_context;
+struct pipe_winsys;
+struct pipe_buffer;
+struct pipe_surface;
+
+struct pipe_winsys *
+xmesa_create_pipe_winsys_aub( void );
+
+void
+xmesa_destroy_pipe_winsys_aub( struct pipe_winsys *winsys );
+
+
+
+struct pipe_context *
+xmesa_create_i965simple( struct pipe_winsys *winsys );
+
+
+
+void xmesa_buffer_subdata_aub(struct pipe_winsys *winsys, 
+			      struct pipe_buffer *buf,
+			      unsigned long offset, 
+			      unsigned long size, 
+			      const void *data,
+			      unsigned aub_type,
+			      unsigned aub_sub_type);
+
+void xmesa_commands_aub(struct pipe_winsys *winsys,
+			unsigned *cmds,
+			unsigned nr_dwords);
+
+
+void xmesa_display_aub( /* struct pipe_winsys *winsys, */
+   struct pipe_surface *surface );
+
+struct pipe_winsys *xmesa_get_pipe_winsys_aub(void);
+
+#endif
diff --git a/src/gallium/winsys/xlib/xmesaP.h b/src/gallium/winsys/xlib/xmesaP.h
new file mode 100644
index 0000000000..fa8d1f14b9
--- /dev/null
+++ b/src/gallium/winsys/xlib/xmesaP.h
@@ -0,0 +1,176 @@
+/*
+ * Mesa 3-D graphics library
+ * Version:  7.1
+ *
+ * Copyright (C) 1999-2007  Brian Paul   All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included
+ * in all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * BRIAN PAUL BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN
+ * AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ */
+
+
+#ifndef XMESAP_H
+#define XMESAP_H
+
+
+#include "GL/xmesa.h"
+#include "mtypes.h"
+#ifdef XFree86Server
+#include "xm_image.h"
+#endif
+
+#include "state_tracker/st_context.h"
+#include "state_tracker/st_public.h"
+
+
+extern _glthread_Mutex _xmesa_lock;
+
+extern XMesaBuffer XMesaBufferList;
+
+/*
+ */
+#define XMESA_SOFTPIPE 1
+#define XMESA_AUB      2
+extern int xmesa_mode;
+
+
+/**
+ * Visual inforation, derived from GLvisual.
+ * Basically corresponds to an XVisualInfo.
+ */
+struct xmesa_visual {
+   GLvisual mesa_visual;	/* Device independent visual parameters */
+   XMesaDisplay *display;	/* The X11 display */
+#ifdef XFree86Server
+   GLint ColormapEntries;
+   GLint nplanes;
+#else
+   XMesaVisualInfo visinfo;	/* X's visual info (pointer to private copy) */
+   XVisualInfo *vishandle;	/* Only used in fakeglx.c */
+#endif
+   GLint BitsPerPixel;		/* True bits per pixel for XImages */
+
+   GLboolean ximage_flag;	/* Use XImage for back buffer (not pixmap)? */
+};
+
+
+/**
+ * Context info, derived from st_context.
+ * Basically corresponds to a GLXContext.
+ */
+struct xmesa_context {
+   struct st_context *st;
+   XMesaVisual xm_visual;	/** pixel format info */
+   XMesaBuffer xm_buffer;	/** current drawbuffer */
+};
+
+
+/**
+ * Types of X/GLX drawables we might render into.
+ */
+typedef enum {
+   WINDOW,          /* An X window */
+   GLXWINDOW,       /* GLX window */
+   PIXMAP,          /* GLX pixmap */
+   PBUFFER          /* GLX Pbuffer */
+} BufferType;
+
+
+/**
+ * Framebuffer information, derived from.
+ * Basically corresponds to a GLXDrawable.
+ */
+struct xmesa_buffer {
+   struct st_framebuffer *stfb;
+
+   GLboolean wasCurrent;	/* was ever the current buffer? */
+   XMesaVisual xm_visual;	/* the X/Mesa visual */
+   XMesaDrawable drawable;	/* Usually the X window ID */
+   XMesaColormap cmap;		/* the X colormap */
+   BufferType type;             /* window, pixmap, pbuffer or glxwindow */
+
+   XMesaImage *tempImage;
+   unsigned long selectedEvents;/* for pbuffers only */
+
+   GLuint shm;			/* X Shared Memory extension status:	*/
+				/*    0 = not available			*/
+				/*    1 = XImage support available	*/
+				/*    2 = Pixmap support available too	*/
+#if defined(USE_XSHM) && !defined(XFree86Server)
+   XShmSegmentInfo shminfo;
+#endif
+
+   XMesaGC gc;			/* scratch GC for span, line, tri drawing */
+
+   /* GLX_EXT_texture_from_pixmap */
+   GLint TextureTarget; /** GLX_TEXTURE_1D_EXT, for example */
+   GLint TextureFormat; /** GLX_TEXTURE_FORMAT_RGB_EXT, for example */
+   GLint TextureMipmap; /** 0 or 1 */
+
+   struct xmesa_buffer *Next;	/* Linked list pointer: */
+};
+
+
+
+/** cast wrapper */
+static INLINE XMesaContext
+xmesa_context(GLcontext *ctx)
+{
+   return (XMesaContext) ctx->DriverCtx;
+}
+
+
+/** cast wrapper */
+static INLINE XMesaBuffer
+xmesa_buffer(GLframebuffer *fb)
+{
+   struct st_framebuffer *stfb = (struct st_framebuffer *) fb;
+   return (XMesaBuffer) st_framebuffer_private(stfb);
+}
+
+
+extern void
+xmesa_delete_framebuffer(struct gl_framebuffer *fb);
+
+extern XMesaBuffer
+xmesa_find_buffer(XMesaDisplay *dpy, XMesaColormap cmap, XMesaBuffer notThis);
+
+extern void
+xmesa_check_and_update_buffer_size(XMesaContext xmctx, XMesaBuffer drawBuffer);
+
+extern void
+xmesa_destroy_buffers_on_display(XMesaDisplay *dpy);
+
+extern struct pipe_context *
+xmesa_create_pipe_context(XMesaContext xm, uint pixelformat);
+
+static INLINE GLuint
+xmesa_buffer_width(XMesaBuffer b)
+{
+   return b->stfb->Base.Width;
+}
+
+static INLINE GLuint
+xmesa_buffer_height(XMesaBuffer b)
+{
+   return b->stfb->Base.Height;
+}
+
+extern void
+xmesa_display_surface(XMesaBuffer b, const struct pipe_surface *surf);
+
+#endif