Merge remote branch 'upstream/gallium-0.1' into nouveau-gallium-0.1

123 files changed, 7757 insertions, 3220 deletions

diff --git a/SConstruct b/SConstruct
index 8013e7c3b2..4785e39ec5 100644
--- a/SConstruct
+++ b/SConstruct
@@ -33,18 +33,16 @@ if common.default_platform in ('linux', 'freebsd', 'darwin'):
 	default_statetrackers = 'mesa'
 	default_drivers = 'softpipe,failover,i915simple,i965simple'
 	default_winsys = 'xlib'
-	default_dri = 'yes'
 elif common.default_platform in ('winddk',):
 	default_statetrackers = 'none'
 	default_drivers = 'softpipe,i915simple'
 	default_winsys = 'none'
-	default_dri = 'no'
 else:
 	default_drivers = 'all'
 	default_winsys = 'all'
-	default_dri = 'no'
 
-opts = common.Options()
+opts = Options('config.py')
+common.AddOptions(opts)
 opts.Add(ListOption('statetrackers', 'state_trackers to build', default_statetrackers,
                      ['mesa']))
 opts.Add(ListOption('drivers', 'pipe drivers to build', default_drivers,
@@ -57,9 +55,6 @@ env = Environment(
 	ENV = os.environ)
 Help(opts.GenerateHelpText(env))
 
-# for debugging
-#print env.Dump()
-
 # replicate options values in local variables
 debug = env['debug']
 dri = env['dri']
@@ -89,6 +84,7 @@ Export([
 # TODO: put the compiler specific settings in separate files
 # TODO: auto-detect as much as possible
 
+common.generate(env)
 
 if platform == 'winddk':
 	env.Tool('winddk', ['.'])
@@ -117,17 +113,9 @@ if gcc:
 	env.Append(CXXFLAGS = '-fmessage-length=0')
 
 if msvc:
-	env.Append(CFLAGS = '/W3')
-	if debug:
-		cflags = [
-			'/Od', # disable optimizations
-			'/Oy-', # disable frame pointer omission
-		]
-	else:
-		cflags = [
-			'/Ox', # maximum optimizations
-			'/Os', # favor code space
-		]
+	cflags = [
+		#'/Wp64', # enable 64 bit porting warnings
+	]
 	env.Append(CFLAGS = cflags)
 	env.Append(CXXFLAGS = cflags)
 	# Put debugging information in a separate .pdb file for each object file as
@@ -136,14 +124,7 @@ if msvc:
 
 # Defines
 if debug:
-	if gcc:
-		env.Append(CPPDEFINES = ['DEBUG'])
-	if msvc:
-		env.Append(CPPDEFINES = [
-			('DBG', '1'),
-			('DEBUG', '1'),
-			('_DEBUG', '1'),
-		])
+	env.Append(CPPDEFINES = ['DEBUG'])
 else:
 	env.Append(CPPDEFINES = ['NDEBUG'])
 
@@ -208,7 +189,8 @@ if llvm:
 	env.ParseConfig('llvm-config --cflags --ldflags --libs')
 	env.Append(CPPDEFINES = ['MESA_LLVM'])
 	env.Append(CXXFLAGS = ['-Wno-long-long'])
-	
+        # Force C++ linkage
+	env['LINK'] = env['CXX']
 
 # libGL
 if platform not in ('winddk',):
@@ -220,8 +202,8 @@ if platform not in ('winddk',):
 		'Xfixes',
 	])
 
-# Convenience library support
-common.createConvenienceLibBuilder(env)
+# for debugging
+#print env.Dump()
 
 Export('env')
 
diff --git a/common.py b/common.py
index b65ed7eaaa..b836a8b41d 100644
--- a/common.py
+++ b/common.py
@@ -43,19 +43,30 @@ else:
 #######################################################################
 # Common options
 
-def Options():
-	from SCons.Options import Options
+def AddOptions(opts):
 	from SCons.Options.BoolOption import BoolOption
 	from SCons.Options.EnumOption import EnumOption
-	opts = Options('config.py')
 	opts.Add(BoolOption('debug', 'build debug version', 'no'))
+	#opts.Add(BoolOption('quiet', 'quiet command lines', 'no'))
 	opts.Add(EnumOption('machine', 'use machine-specific assembly code', default_machine,
 											 allowed_values=('generic', 'x86', 'x86_64')))
 	opts.Add(EnumOption('platform', 'target platform', default_platform,
 											 allowed_values=('linux', 'cell', 'winddk')))
 	opts.Add(BoolOption('llvm', 'use LLVM', 'no'))
 	opts.Add(BoolOption('dri', 'build DRI drivers', default_dri))
-	return opts
+
+
+#######################################################################
+# Quiet command lines
+#
+# See also http://www.scons.org/wiki/HidingCommandLinesInOutput
+
+def quietCommandLines(env):
+	env['CCCOMSTR'] = "Compiling $SOURCE ..."
+	env['CXXCOMSTR'] = "Compiling $SOURCE ..."
+	env['ARCOMSTR'] = "Archiving $TARGET ..."
+	env['RANLIBCOMSTR'] = ""
+	env['LINKCOMSTR'] = "Linking $TARGET ..."
 
 
 #######################################################################
@@ -114,3 +125,16 @@ def make_build_dir(env):
 	env.SConsignFile(os.path.join(build_dir, '.sconsign'))
 	return build_dir
 
+
+#######################################################################
+# Common environment generation code
+
+def generate(env):
+	# FIXME: this is already too late
+	#if env.get('quiet', False):
+	#	quietCommandLines(env)
+	createConvenienceLibBuilder(env)
+
+	# for debugging
+	#print env.Dump()
+
diff --git a/configs/default b/configs/default
index af9ff6f426..49d3737ec7 100644
--- a/configs/default
+++ b/configs/default
@@ -68,7 +68,7 @@ PROGRAM_DIRS = demos redbook samples glsl xdemos
 
 
 # Gallium directories and 
-GALLIUM_AUXILIARY_DIRS = draw cso_cache pipebuffer tgsi rtasm util
+GALLIUM_AUXILIARY_DIRS = draw cso_cache pipebuffer tgsi rtasm util sct
 GALLIUM_AUXILIARIES = $(foreach DIR,$(GALLIUM_AUXILIARY_DIRS),$(TOP)/src/gallium/auxiliary/$(DIR)/lib$(DIR).a)
 GALLIUM_DRIVER_DIRS = softpipe i915simple i965simple nv30 nv40 nv50 failover
 GALLIUM_DRIVERS = $(foreach DIR,$(GALLIUM_DRIVER_DIRS),$(TOP)/src/gallium/drivers/$(DIR)/lib$(DIR).a)
diff --git a/src/gallium/Makefile b/src/gallium/Makefile
index 568747e157..aa77021daf 100644
--- a/src/gallium/Makefile
+++ b/src/gallium/Makefile
@@ -19,3 +19,7 @@ subdirs:
 clean:
 	rm -f `find . -name \*.[oa]`
 	rm -f `find . -name depend`
+
+
+# Dummy install target
+install:
diff --git a/src/gallium/README.portability b/src/gallium/README.portability
index c70ca774da..ab0c197847 100644
--- a/src/gallium/README.portability
+++ b/src/gallium/README.portability
@@ -39,5 +39,6 @@ portable way.
 
 == Debugging ==
 
-TODO
+* Use the functions/macros in p_debug.h.
 
+* Don't include assert.h, call abort, printf, etc.
diff --git a/src/gallium/SConscript b/src/gallium/SConscript
index fa4833cbcf..f09778ce99 100644
--- a/src/gallium/SConscript
+++ b/src/gallium/SConscript
@@ -9,6 +9,9 @@ auxiliaries = []
 Export('auxiliaries')
 
 
+if llvm:
+	SConscript(['auxiliary/gallivm/SConscript'])
+
 SConscript([
 	# NOTE: order matters!
 	'auxiliary/util/SConscript',
@@ -19,8 +22,5 @@ SConscript([
 	'auxiliary/pipebuffer/SConscript',
 ])
 
-if llvm:
-	SConscript(['auxiliary/gallivm/SConscript'])
-
 for driver in env['drivers']:
 	SConscript(os.path.join('drivers', driver, 'SConscript'))
diff --git a/src/gallium/auxiliary/cso_cache/Makefile b/src/gallium/auxiliary/cso_cache/Makefile
index 3e49266163..6bd6602088 100644
--- a/src/gallium/auxiliary/cso_cache/Makefile
+++ b/src/gallium/auxiliary/cso_cache/Makefile
@@ -4,6 +4,7 @@ include $(TOP)/configs/current
 LIBNAME = cso_cache
 
 C_SOURCES = \
+	cso_context.c \
 	cso_cache.c \
 	cso_hash.c
 
diff --git a/src/gallium/auxiliary/cso_cache/SConscript b/src/gallium/auxiliary/cso_cache/SConscript
index 9751881613..651e68a191 100644
--- a/src/gallium/auxiliary/cso_cache/SConscript
+++ b/src/gallium/auxiliary/cso_cache/SConscript
@@ -3,6 +3,7 @@ Import('*')
 cso_cache = env.ConvenienceLibrary(
 	target = 'cso_cache',
 	source = [
+		'cso_context.c',
 		'cso_cache.c',
 		'cso_hash.c',
 	])
diff --git a/src/gallium/auxiliary/cso_cache/cso_cache.c b/src/gallium/auxiliary/cso_cache/cso_cache.c
index b427b509f8..a2764b4265 100644
--- a/src/gallium/auxiliary/cso_cache/cso_cache.c
+++ b/src/gallium/auxiliary/cso_cache/cso_cache.c
@@ -207,8 +207,11 @@ static INLINE void sanitize_hash(struct cso_hash *hash, enum cso_cache_type type
 {
    /* if we're approach the maximum size, remove fourth of the entries
     * otherwise every subsequent call will go through the same */
-   int max_entries = (max_size > cso_hash_size(hash)) ? max_size : cso_hash_size(hash);
+   int hash_size = cso_hash_size(hash);
+   int max_entries = (max_size > hash_size) ? max_size : hash_size;
    int to_remove =  (max_size < max_entries) * max_entries/4;
+   if (hash_size > max_size)
+      to_remove += hash_size - max_size;
    while (to_remove) {
       /*remove elements until we're good */
       /*fixme: currently we pick the nodes to remove at random*/
diff --git a/src/gallium/auxiliary/cso_cache/cso_cache.h b/src/gallium/auxiliary/cso_cache/cso_cache.h
index 44ee128a4a..e5edbbb556 100644
--- a/src/gallium/auxiliary/cso_cache/cso_cache.h
+++ b/src/gallium/auxiliary/cso_cache/cso_cache.h
@@ -84,47 +84,49 @@
 extern "C" {
 #endif
 
+typedef void (*cso_state_callback)(void *ctx, void *obj);
+
 struct cso_cache;
 
 struct cso_blend {
    struct pipe_blend_state state;
    void *data;
-   void (*delete_state)(struct pipe_context *, void *);
+   cso_state_callback delete_state;
    struct pipe_context *context;
 };
 
 struct cso_depth_stencil_alpha {
    struct pipe_depth_stencil_alpha_state state;
    void *data;
-   void (*delete_state)(struct pipe_context *, void *);
+   cso_state_callback delete_state;
    struct pipe_context *context;
 };
 
 struct cso_rasterizer {
    struct pipe_rasterizer_state state;
    void *data;
-   void (*delete_state)(struct pipe_context *, void *);
+   cso_state_callback delete_state;
    struct pipe_context *context;
 };
 
 struct cso_fragment_shader {
    struct pipe_shader_state state;
    void *data;
-   void (*delete_state)(struct pipe_context *, void *);
+   cso_state_callback delete_state;
    struct pipe_context *context;
 };
 
 struct cso_vertex_shader {
    struct pipe_shader_state state;
    void *data;
-   void (*delete_state)(struct pipe_context *, void *);
+   cso_state_callback delete_state;
    struct pipe_context *context;
 };
 
 struct cso_sampler {
    struct pipe_sampler_state state;
    void *data;
-   void (*delete_state)(struct pipe_context *, void *);
+   cso_state_callback delete_state;
    struct pipe_context *context;
 };
 
@@ -138,8 +140,6 @@ enum cso_cache_type {
    CSO_VERTEX_SHADER
 };
 
-typedef void (*cso_state_callback)(void *, void *);
-
 unsigned cso_construct_key(void *item, int item_size);
 
 struct cso_cache *cso_cache_create(void);
diff --git a/src/gallium/auxiliary/cso_cache/cso_context.c b/src/gallium/auxiliary/cso_cache/cso_context.c
new file mode 100644
index 0000000000..f7f4aebb16
--- /dev/null
+++ b/src/gallium/auxiliary/cso_cache/cso_context.c
@@ -0,0 +1,354 @@
+/**************************************************************************
+ *
+ * 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.
+ *
+ **************************************************************************/
+
+ /* Wrap the cso cache & hash mechanisms in a simplified
+  * pipe-driver-specific interface.
+  *
+  * Authors:
+  *   Zack Rusin <zack@tungstengraphics.com>
+  *   Keith Whitwell <keith@tungstengraphics.com>
+  */
+
+#include "pipe/p_state.h"
+#include "pipe/p_util.h"
+
+#include "cso_cache/cso_context.h"
+#include "cso_cache/cso_cache.h"
+#include "cso_cache/cso_hash.h"
+
+struct cso_context {
+   struct pipe_context *pipe;
+   struct cso_cache *cache;
+
+   struct {
+      void *samplers[PIPE_MAX_SAMPLERS];
+      unsigned nr_samplers;
+   } hw;
+
+   void *samplers[PIPE_MAX_SAMPLERS];
+   unsigned nr_samplers;
+
+   void *blend;
+   void *depth_stencil;
+   void *rasterizer;
+   void *fragment_shader;
+   void *vertex_shader;
+};
+
+
+struct cso_context *cso_create_context( struct pipe_context *pipe )
+{
+   struct cso_context *ctx = CALLOC_STRUCT(cso_context);
+   if (ctx == NULL)
+      goto out;
+
+   ctx->cache = cso_cache_create();
+   if (ctx->cache == NULL)
+      goto out;
+
+   ctx->pipe = pipe;
+
+   /* Enable for testing: */
+   if (0) cso_set_maximum_cache_size( ctx->cache, 4 );
+
+   return ctx;
+
+out:
+   cso_destroy_context( ctx );      
+   return NULL;
+}
+
+static void cso_release_all( struct cso_context *ctx )
+{
+   if (ctx->pipe) {
+      ctx->pipe->bind_blend_state( ctx->pipe, NULL );
+      ctx->pipe->bind_rasterizer_state( ctx->pipe, NULL );
+      ctx->pipe->bind_sampler_states( ctx->pipe, 0, NULL );
+      ctx->pipe->bind_depth_stencil_alpha_state( ctx->pipe, NULL );
+      ctx->pipe->bind_fs_state( ctx->pipe, NULL );
+      ctx->pipe->bind_vs_state( ctx->pipe, NULL );
+   }
+
+   if (ctx->cache) {
+      cso_cache_delete( ctx->cache );
+      ctx->cache = NULL;
+   }
+}
+
+
+void cso_destroy_context( struct cso_context *ctx )
+{
+   debug_printf("%s\n", __FUNCTION__);
+
+   if (ctx)
+      cso_release_all( ctx );
+
+   FREE( ctx );
+}
+
+
+/* Those function will either find the state of the given template
+ * in the cache or they will create a new state from the given
+ * template, insert it in the cache and return it.
+ */
+
+/*
+ * If the driver returns 0 from the create method then they will assign
+ * the data member of the cso to be the template itself.
+ */
+
+void cso_set_blend(struct cso_context *ctx,
+                   const struct pipe_blend_state *templ)
+{
+   unsigned hash_key = cso_construct_key((void*)templ, sizeof(struct pipe_blend_state));
+   struct cso_hash_iter iter = cso_find_state_template(ctx->cache,
+                                                       hash_key, CSO_BLEND,
+                                                       (void*)templ);
+   void *handle;
+
+   if (cso_hash_iter_is_null(iter)) {
+      struct cso_blend *cso = MALLOC(sizeof(struct cso_blend));
+
+      cso->state = *templ;
+      cso->data = ctx->pipe->create_blend_state(ctx->pipe, &cso->state);
+      cso->delete_state = (cso_state_callback)ctx->pipe->delete_blend_state;
+      cso->context = ctx->pipe;
+
+      iter = cso_insert_state(ctx->cache, hash_key, CSO_BLEND, cso);
+      handle = cso->data;
+   }
+   else {
+      handle = ((struct cso_blend *)cso_hash_iter_data(iter))->data;
+   }
+
+   if (ctx->blend != handle) {
+      ctx->blend = handle;
+      ctx->pipe->bind_blend_state(ctx->pipe, handle);
+   }
+}
+
+void cso_single_sampler(struct cso_context *ctx,
+                        unsigned idx,
+                        const struct pipe_sampler_state *templ)
+{
+   void *handle = NULL;
+   
+   if (templ != NULL) {
+      unsigned hash_key = cso_construct_key((void*)templ, sizeof(struct pipe_sampler_state));
+      struct cso_hash_iter iter = cso_find_state_template(ctx->cache,
+                                                          hash_key, CSO_SAMPLER,
+                                                          (void*)templ);
+
+      if (cso_hash_iter_is_null(iter)) {
+         struct cso_sampler *cso = MALLOC(sizeof(struct cso_sampler));
+         
+         cso->state = *templ;
+         cso->data = ctx->pipe->create_sampler_state(ctx->pipe, &cso->state);
+         cso->delete_state = (cso_state_callback)ctx->pipe->delete_sampler_state;
+         cso->context = ctx->pipe;
+
+         iter = cso_insert_state(ctx->cache, hash_key, CSO_SAMPLER, cso);
+         handle = cso->data;
+      }
+      else {
+         handle = ((struct cso_sampler *)cso_hash_iter_data(iter))->data;
+      }
+   }
+
+   ctx->samplers[idx] = handle;
+}
+
+void cso_single_sampler_done( struct cso_context *ctx )
+{
+   unsigned i; 
+
+   for (i = 0; i < 8; i++)
+      if (ctx->samplers[i] == NULL)
+         break;
+
+   ctx->nr_samplers = i;
+
+   if (ctx->hw.nr_samplers != ctx->nr_samplers ||
+       memcmp(ctx->hw.samplers, 
+              ctx->samplers, 
+              ctx->nr_samplers * sizeof(void *)) != 0) 
+   {
+      memcpy(ctx->hw.samplers, ctx->samplers, ctx->nr_samplers * sizeof(void *));
+      ctx->hw.nr_samplers = ctx->nr_samplers;
+
+      ctx->pipe->bind_sampler_states(ctx->pipe, ctx->nr_samplers, ctx->samplers);
+   }
+}
+
+void cso_set_samplers( struct cso_context *ctx,
+                       unsigned nr,
+                       const struct pipe_sampler_state **templates )
+{
+   unsigned i;
+   
+   /* TODO: fastpath
+    */
+
+   for (i = 0; i < nr; i++)
+      cso_single_sampler( ctx, i, templates[i] );
+
+   for ( ; i < ctx->nr_samplers; i++)
+      cso_single_sampler( ctx, i, NULL );
+   
+   cso_single_sampler_done( ctx );
+}
+
+void cso_set_depth_stencil_alpha(struct cso_context *ctx,
+                                 const struct pipe_depth_stencil_alpha_state *templ)
+{
+   unsigned hash_key = cso_construct_key((void*)templ,
+                                         sizeof(struct pipe_depth_stencil_alpha_state));
+   struct cso_hash_iter iter = cso_find_state_template(ctx->cache,
+                                                       hash_key, 
+						       CSO_DEPTH_STENCIL_ALPHA,
+                                                       (void*)templ);
+   void *handle;
+
+   if (cso_hash_iter_is_null(iter)) {
+      struct cso_depth_stencil_alpha *cso = MALLOC(sizeof(struct cso_depth_stencil_alpha));
+
+      cso->state = *templ;
+      cso->data = ctx->pipe->create_depth_stencil_alpha_state(ctx->pipe, &cso->state);
+      cso->delete_state = (cso_state_callback)ctx->pipe->delete_depth_stencil_alpha_state;
+      cso->context = ctx->pipe;
+
+      cso_insert_state(ctx->cache, hash_key, CSO_DEPTH_STENCIL_ALPHA, cso);
+      handle = cso->data;
+   }
+   else {
+      handle = ((struct cso_depth_stencil_alpha *)cso_hash_iter_data(iter))->data;
+   }
+
+   if (ctx->depth_stencil != handle) {
+      ctx->depth_stencil = handle;
+      ctx->pipe->bind_depth_stencil_alpha_state(ctx->pipe, handle);
+   }
+}
+
+
+
+void cso_set_rasterizer(struct cso_context *ctx,
+                              const struct pipe_rasterizer_state *templ)
+{
+   unsigned hash_key = cso_construct_key((void*)templ,
+                                         sizeof(struct pipe_rasterizer_state));
+   struct cso_hash_iter iter = cso_find_state_template(ctx->cache,
+                                                       hash_key, CSO_RASTERIZER,
+                                                       (void*)templ);
+   void *handle = NULL;
+
+   if (cso_hash_iter_is_null(iter)) {
+      struct cso_rasterizer *cso = MALLOC(sizeof(struct cso_rasterizer));
+
+      cso->state = *templ;
+      cso->data = ctx->pipe->create_rasterizer_state(ctx->pipe, &cso->state);
+      cso->delete_state = (cso_state_callback)ctx->pipe->delete_rasterizer_state;
+      cso->context = ctx->pipe;
+
+      cso_insert_state(ctx->cache, hash_key, CSO_RASTERIZER, cso);
+      handle = cso->data;
+   }
+   else {
+      handle = ((struct cso_rasterizer *)cso_hash_iter_data(iter))->data;
+   }
+
+   if (ctx->rasterizer != handle) {
+      ctx->rasterizer = handle;
+      ctx->pipe->bind_rasterizer_state(ctx->pipe, handle);
+   }
+}
+
+
+
+
+
+void cso_set_fragment_shader(struct cso_context *ctx,
+                             const struct pipe_shader_state *templ)
+{
+   unsigned hash_key = cso_construct_key((void*)templ,
+                                         sizeof(struct pipe_shader_state));
+   struct cso_hash_iter iter = cso_find_state_template(ctx->cache,
+                                                       hash_key, CSO_FRAGMENT_SHADER,
+                                                       (void*)templ);
+   void *handle = NULL;
+
+   if (cso_hash_iter_is_null(iter)) {
+      struct cso_fragment_shader *cso = MALLOC(sizeof(struct cso_fragment_shader));
+
+      cso->state = *templ;
+      cso->data = ctx->pipe->create_fs_state(ctx->pipe, &cso->state);
+      cso->delete_state = (cso_state_callback)ctx->pipe->delete_fs_state;
+      cso->context = ctx->pipe;
+
+      iter = cso_insert_state(ctx->cache, hash_key, CSO_FRAGMENT_SHADER, cso);
+      handle = cso->data;
+   }
+   else {
+      handle = ((struct cso_fragment_shader *)cso_hash_iter_data(iter))->data;
+   }
+
+   if (ctx->fragment_shader != handle) {
+      ctx->fragment_shader = handle;
+      ctx->pipe->bind_fs_state(ctx->pipe, handle);
+   }
+}
+
+void cso_set_vertex_shader(struct cso_context *ctx,
+                           const struct pipe_shader_state *templ)
+{
+   unsigned hash_key = cso_construct_key((void*)templ,
+                                         sizeof(struct pipe_shader_state));
+   struct cso_hash_iter iter = cso_find_state_template(ctx->cache,
+                                                       hash_key, CSO_VERTEX_SHADER,
+                                                       (void*)templ);
+   void *handle = NULL;
+
+   if (cso_hash_iter_is_null(iter)) {
+      struct cso_vertex_shader *cso = MALLOC(sizeof(struct cso_vertex_shader));
+
+      cso->state = *templ;
+      cso->data = ctx->pipe->create_vs_state(ctx->pipe, &cso->state);
+      cso->delete_state = (cso_state_callback)ctx->pipe->delete_vs_state;
+      cso->context = ctx->pipe;
+
+      iter = cso_insert_state(ctx->cache, hash_key, CSO_VERTEX_SHADER, cso);
+      handle = cso->data;
+   }
+   else {
+      handle = ((struct cso_vertex_shader *)cso_hash_iter_data(iter))->data;
+   }
+
+   if (ctx->vertex_shader != handle) {
+      ctx->vertex_shader = handle;
+      ctx->pipe->bind_vs_state(ctx->pipe, handle);
+   }
+}
diff --git a/src/gallium/auxiliary/cso_cache/cso_context.h b/src/gallium/auxiliary/cso_cache/cso_context.h
new file mode 100644
index 0000000000..1f2a630804
--- /dev/null
+++ b/src/gallium/auxiliary/cso_cache/cso_context.h
@@ -0,0 +1,85 @@
+/**************************************************************************
+ *
+ * Copyright 2007-2008 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 CSO_CONTEXT_H
+#define CSO_CONTEXT_H
+
+#include "pipe/p_context.h"
+#include "pipe/p_state.h"
+
+
+#ifdef	__cplusplus
+extern "C" {
+#endif
+
+struct cso_context;
+
+struct cso_context *cso_create_context( struct pipe_context *pipe );
+
+void cso_set_blend( struct cso_context *cso,
+                    const struct pipe_blend_state *blend );
+
+void cso_set_depth_stencil_alpha( struct cso_context *cso,
+                                  const struct pipe_depth_stencil_alpha_state *dsa );
+
+void cso_set_rasterizer( struct cso_context *cso,
+                         const struct pipe_rasterizer_state *rasterizer );
+
+void cso_set_samplers( struct cso_context *cso,
+                       unsigned count,
+                       const struct pipe_sampler_state **states );
+
+/* Alternate interface to support state trackers that like to modify
+ * samplers one at a time:
+ */
+void cso_single_sampler( struct cso_context *cso,
+                         unsigned nr,
+                         const struct pipe_sampler_state *states );
+
+void cso_single_sampler_done( struct cso_context *cso );
+
+
+/* These aren't really sensible -- most of the time the api provides
+ * object semantics for shaders anyway, and the cases where it doesn't
+ * (eg mesa's internall-generated texenv programs), it will be up to
+ * the state tracker to implement their own specialized caching.
+ */
+void cso_set_fragment_shader( struct cso_context *cso,
+                              const struct pipe_shader_state *shader );
+
+void cso_set_vertex_shader( struct cso_context *cso,
+                            const struct pipe_shader_state *shader );
+
+void cso_destroy_context( struct cso_context *cso );
+
+
+#ifdef	__cplusplus
+}
+#endif
+
+#endif
diff --git a/src/gallium/auxiliary/cso_cache/cso_hash.c b/src/gallium/auxiliary/cso_cache/cso_hash.c
index b3b4d667d2..5cad5d3be7 100644
--- a/src/gallium/auxiliary/cso_cache/cso_hash.c
+++ b/src/gallium/auxiliary/cso_cache/cso_hash.c
@@ -101,13 +101,6 @@ static void *cso_data_allocate_node(struct cso_hash_data *hash)
 
 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);
 }
 
diff --git a/src/gallium/auxiliary/cso_cache/cso_hash.h b/src/gallium/auxiliary/cso_cache/cso_hash.h
index d5bca9d591..84b45a5963 100644
--- a/src/gallium/auxiliary/cso_cache/cso_hash.h
+++ b/src/gallium/auxiliary/cso_cache/cso_hash.h
@@ -25,6 +25,16 @@
  *
  **************************************************************************/
 
+/**
+ This file provides a hash implementation that is capable of dealing
+ with collisions. It stores colliding entries in linked list. All
+ functions operating on the hash return an iterator. The iterator
+ itself points to the collision list. If there wasn't any collision
+ the list will have just one entry, otherwise client code should
+ iterate over the entries to find the exact entry among ones that
+ had the same key (e.g. memcmp could be used on the data to check
+ that)
+*/
  /*
   * Authors:
   *   Zack Rusin <zack@tungstengraphics.com>
diff --git a/src/gallium/auxiliary/draw/SConscript b/src/gallium/auxiliary/draw/SConscript
index c18dcb2927..5cb7664c85 100644
--- a/src/gallium/auxiliary/draw/SConscript
+++ b/src/gallium/auxiliary/draw/SConscript
@@ -16,6 +16,7 @@ draw = env.ConvenienceLibrary(
 		'draw_offset.c',
 		'draw_prim.c',
 		'draw_pstipple.c',
+		'draw_passthrough.c',
 		'draw_stipple.c',
 		'draw_twoside.c',
 		'draw_unfilled.c',
diff --git a/src/gallium/auxiliary/draw/draw_aaline.c b/src/gallium/auxiliary/draw/draw_aaline.c
index 7660e56fe6..6b1e640ae9 100644
--- a/src/gallium/auxiliary/draw/draw_aaline.c
+++ b/src/gallium/auxiliary/draw/draw_aaline.c
@@ -78,7 +78,8 @@ struct aaline_stage
 
    void *sampler_cso;
    struct pipe_texture *texture;
-   uint sampler_unit;
+   uint num_samplers;
+   uint num_textures;
 
 
    /*
@@ -98,11 +99,10 @@ struct aaline_stage
    void (*driver_bind_fs_state)(struct pipe_context *, void *);
    void (*driver_delete_fs_state)(struct pipe_context *, void *);
 
-   void (*driver_bind_sampler_state)(struct pipe_context *, unsigned, void *);
-
-   void (*driver_set_sampler_texture)(struct pipe_context *,
-                                      unsigned sampler,
-                                      struct pipe_texture *);
+   void (*driver_bind_sampler_states)(struct pipe_context *, unsigned,
+                                      void **);
+   void (*driver_set_sampler_textures)(struct pipe_context *, unsigned,
+                                       struct pipe_texture **);
 
    struct pipe_context *pipe;
 };
@@ -607,6 +607,7 @@ aaline_first_line(struct draw_stage *stage, struct prim_header *header)
    auto struct aaline_stage *aaline = aaline_stage(stage);
    struct draw_context *draw = stage->draw;
    struct pipe_context *pipe = aaline->pipe;
+   uint num = MAX2(aaline->num_textures, aaline->num_samplers);
 
    assert(draw->rasterizer->line_smooth);
 
@@ -624,8 +625,11 @@ aaline_first_line(struct draw_stage *stage, struct prim_header *header)
     */
    bind_aaline_fragment_shader(aaline);
 
-   aaline->driver_bind_sampler_state(pipe, aaline->sampler_unit, aaline->sampler_cso);
-   aaline->driver_set_sampler_texture(pipe, aaline->sampler_unit, aaline->texture);
+   aaline->state.sampler[num] = aaline->sampler_cso;
+   aaline->state.texture[num] = aaline->texture;
+
+   aaline->driver_bind_sampler_states(pipe, num + 1, aaline->state.sampler);
+   aaline->driver_set_sampler_textures(pipe, num + 1, aaline->state.texture);
 
    /* now really draw first line */
    stage->line = aaline_line;
@@ -647,10 +651,10 @@ aaline_flush(struct draw_stage *stage, unsigned flags)
    aaline->driver_bind_fs_state(pipe, aaline->fs->driver_fs);
 
    /* XXX restore original texture, sampler state */
-   aaline->driver_bind_sampler_state(pipe, aaline->sampler_unit,
-                                 aaline->state.sampler[aaline->sampler_unit]);
-   aaline->driver_set_sampler_texture(pipe, aaline->sampler_unit,
-                                 aaline->state.texture[aaline->sampler_unit]);
+   aaline->driver_bind_sampler_states(pipe, aaline->num_samplers,
+                                      aaline->state.sampler);
+   aaline->driver_set_sampler_textures(pipe, aaline->num_textures,
+                                       aaline->state.texture);
 
    draw->extra_vp_outputs.slot = 0;
 }
@@ -729,7 +733,8 @@ aaline_bind_fs_state(struct pipe_context *pipe, void *fs)
    /* save current */
    aaline->fs = aafs;
    /* pass-through */
-   aaline->driver_bind_fs_state(aaline->pipe, aafs->driver_fs);
+   aaline->driver_bind_fs_state(aaline->pipe,
+                                (aafs ? aafs->driver_fs : NULL));
 }
 
 
@@ -745,26 +750,28 @@ aaline_delete_fs_state(struct pipe_context *pipe, void *fs)
 
 
 static void
-aaline_bind_sampler_state(struct pipe_context *pipe,
-                          unsigned unit, void *sampler)
+aaline_bind_sampler_states(struct pipe_context *pipe,
+                           unsigned num, void **sampler)
 {
    struct aaline_stage *aaline = aaline_stage_from_pipe(pipe);
    /* save current */
-   aaline->state.sampler[unit] = sampler;
+   memcpy(aaline->state.sampler, sampler, num * sizeof(void *));
+   aaline->num_samplers = num;
    /* pass-through */
-   aaline->driver_bind_sampler_state(aaline->pipe, unit, sampler);
+   aaline->driver_bind_sampler_states(aaline->pipe, num, sampler);
 }
 
 
 static void
-aaline_set_sampler_texture(struct pipe_context *pipe,
-                           unsigned sampler, struct pipe_texture *texture)
+aaline_set_sampler_textures(struct pipe_context *pipe,
+                            unsigned num, struct pipe_texture **texture)
 {
    struct aaline_stage *aaline = aaline_stage_from_pipe(pipe);
    /* save current */
-   aaline->state.texture[sampler] = texture;
+   memcpy(aaline->state.texture, texture, num * sizeof(struct pipe_texture *));
+   aaline->num_textures = num;
    /* pass-through */
-   aaline->driver_set_sampler_texture(aaline->pipe, sampler, texture);
+   aaline->driver_set_sampler_textures(aaline->pipe, num, texture);
 }
 
 
@@ -798,14 +805,14 @@ draw_install_aaline_stage(struct draw_context *draw, struct pipe_context *pipe)
    aaline->driver_bind_fs_state = pipe->bind_fs_state;
    aaline->driver_delete_fs_state = pipe->delete_fs_state;
 
-   aaline->driver_bind_sampler_state = pipe->bind_sampler_state;
-   aaline->driver_set_sampler_texture = pipe->set_sampler_texture;
+   aaline->driver_bind_sampler_states = pipe->bind_sampler_states;
+   aaline->driver_set_sampler_textures = pipe->set_sampler_textures;
 
    /* override the driver's functions */
    pipe->create_fs_state = aaline_create_fs_state;
    pipe->bind_fs_state = aaline_bind_fs_state;
    pipe->delete_fs_state = aaline_delete_fs_state;
 
-   pipe->bind_sampler_state = aaline_bind_sampler_state;
-   pipe->set_sampler_texture = aaline_set_sampler_texture;
+   pipe->bind_sampler_states = aaline_bind_sampler_states;
+   pipe->set_sampler_textures = aaline_set_sampler_textures;
 }
diff --git a/src/gallium/auxiliary/draw/draw_aapoint.c b/src/gallium/auxiliary/draw/draw_aapoint.c
index 70f696475f..99e9e9fe34 100644
--- a/src/gallium/auxiliary/draw/draw_aapoint.c
+++ b/src/gallium/auxiliary/draw/draw_aapoint.c
@@ -800,7 +800,8 @@ aapoint_bind_fs_state(struct pipe_context *pipe, void *fs)
    /* save current */
    aapoint->fs = aafs;
    /* pass-through */
-   aapoint->driver_bind_fs_state(aapoint->pipe, aafs->driver_fs);
+   aapoint->driver_bind_fs_state(aapoint->pipe,
+                                 (aafs ? aafs->driver_fs : NULL));
 }
 
 
diff --git a/src/gallium/auxiliary/draw/draw_context.c b/src/gallium/auxiliary/draw/draw_context.c
index 428b6209e0..fed2b6e759 100644
--- a/src/gallium/auxiliary/draw/draw_context.c
+++ b/src/gallium/auxiliary/draw/draw_context.c
@@ -34,6 +34,7 @@
 #include "pipe/p_util.h"
 #include "draw_context.h"
 #include "draw_private.h"
+#include "draw_vbuf.h"
 
 
 
@@ -114,6 +115,13 @@ void draw_destroy( struct draw_context *draw )
       draw->pipeline.rasterize->destroy( draw->pipeline.rasterize );
    tgsi_exec_machine_free_data(&draw->machine);
    align_free( draw->vs.queue[0].vertex ); /* Frees all the vertices. */
+
+   /* Not so fast -- we're just borrowing this at the moment.
+    * 
+   if (draw->render)
+      draw->render->destroy( draw->render );
+   */
+
    FREE( draw );
 }
 
@@ -349,3 +357,10 @@ void draw_reset_vertex_ids(struct draw_context *draw)
 
    draw_vertex_cache_reset_vertex_ids(draw);
 }
+
+
+void draw_set_render( struct draw_context *draw, 
+		      struct vbuf_render *render )
+{
+   draw->render = render;
+}
diff --git a/src/gallium/auxiliary/draw/draw_context.h b/src/gallium/auxiliary/draw/draw_context.h
index ab87b4127c..df63e91a22 100644
--- a/src/gallium/auxiliary/draw/draw_context.h
+++ b/src/gallium/auxiliary/draw/draw_context.h
@@ -168,4 +168,9 @@ unsigned draw_trim_prim( unsigned mode, unsigned count );
 
 
 
+
+struct vbuf_render;
+void draw_set_render( struct draw_context *draw, 
+		      struct vbuf_render *render );
+
 #endif /* DRAW_CONTEXT_H */
diff --git a/src/gallium/auxiliary/draw/draw_passthrough.c b/src/gallium/auxiliary/draw/draw_passthrough.c
new file mode 100644
index 0000000000..a51fa0ab23
--- /dev/null
+++ b/src/gallium/auxiliary/draw/draw_passthrough.c
@@ -0,0 +1,222 @@
+/**************************************************************************
+ * 
+ * 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>
+  */
+
+
+/* This code is a prototype of what a passhthrough vertex shader might
+ * look like.
+ *
+ * Probably the best approach for us is to do:
+ *    - vertex fetch
+ *    - vertex shader
+ *    - cliptest / viewport transform
+ *
+ * in one step, then examine the clipOrMask & choose between two paths:
+ *
+ * Either:
+ *    - build primitive headers
+ *    - clip and the primitive path
+ *    - build clipped vertex buffers,
+ *    - vertex-emit to vbuf buffers
+ *
+ * Or, if no clipping:
+ *    - vertex-emit directly to vbuf buffers
+ *
+ * But when bypass clipping is enabled, we just take the latter
+ * choice.  If (some new) passthrough-vertex-shader flag is also set,
+ * the pipeline degenerates to:
+ *
+ *    - vertex fetch
+ *    - vertex emit to vbuf buffers
+ *
+ * Which is what is prototyped here.
+ */
+#include "pipe/p_util.h"
+#include "draw/draw_context.h"
+#include "draw/draw_private.h"
+#include "draw/draw_vbuf.h"
+#include "draw/draw_vertex.h"
+
+
+
+/* Example of a fetch/emit passthrough shader which could be
+ * generated when bypass_clipping is enabled on a passthrough vertex
+ * shader.
+ */
+static void fetch_xyz_rgb_st( struct draw_context *draw,
+			      float *out,
+			      unsigned start,
+			      unsigned count )
+{
+   const unsigned *pitch   = draw->vertex_fetch.pitch;
+   const ubyte **src       = draw->vertex_fetch.src_ptr;
+   unsigned i;
+
+   const ubyte *xyzw = src[0] + start * pitch[0];
+   const ubyte *rgba = src[1] + start * pitch[1];
+   const ubyte *st = src[2] + start * pitch[2];
+
+   /* loop over vertex attributes (vertex shader inputs)
+    */
+   for (i = 0; i < count; i++) {
+      {
+	 const float *in = (const float *)xyzw; xyzw += pitch[0];
+         /* decode input, encode output.  Assume both are float[4] */
+	 out[0] = in[0];
+	 out[1] = in[1];
+	 out[2] = in[2];
+	 out[3] = in[3];
+      }
+
+      {
+	 const float *in = (const float *)rgba; rgba += pitch[1];
+         /* decode input, encode output.  Assume both are float[4] */
+	 out[4] = in[0];
+	 out[5] = in[1];
+	 out[6] = in[2];
+ 	 out[7] = in[3];
+      }
+
+      {
+	 const float *in = (const float *)st; st += pitch[2];
+         /* decode input, encode output.  Assume both are float[2] */
+	 out[8] = in[0];
+	 out[9] = in[1];
+      }
+
+      out += 10;
+   }
+}
+
+			       
+static boolean update_shader( struct draw_context *draw )
+{
+   const struct vertex_info *vinfo = draw->render->get_vertex_info(draw->render);
+
+   unsigned nr_attrs = vinfo->num_attribs;
+   unsigned i;
+
+   for (i = 0; i < nr_attrs; i++) {
+      unsigned buf = draw->vertex_element[i].vertex_buffer_index;
+
+      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] = NULL;
+   }
+
+   draw->vertex_fetch.nr_attrs = nr_attrs;
+   draw->vertex_fetch.fetch_func = NULL;
+   draw->vertex_fetch.pt_fetch = NULL;
+
+   draw->pt.hw_vertex_size = vinfo->size * 4;
+
+   /* Just trying to figure out how this would work:
+    */
+   if (nr_attrs == 3 &&
+       0 /* some other tests */)
+   {
+      draw->vertex_fetch.pt_fetch = fetch_xyz_rgb_st;
+      assert(vinfo->size == 10);
+      return TRUE;
+   }
+   
+   return FALSE;
+}
+
+
+
+static boolean set_prim( struct draw_context *draw,
+		      unsigned prim )
+{
+   assert(!draw->user.elts);   
+
+   draw->pt.prim = prim;
+
+   switch (prim) { 
+   case PIPE_PRIM_LINE_LOOP:
+   case PIPE_PRIM_QUADS:
+   case PIPE_PRIM_QUAD_STRIP:
+      return FALSE;
+   default:
+      draw->render->set_primitive( draw->render, prim );
+      return TRUE;
+   }
+}
+
+
+
+boolean
+draw_passthrough_arrays(struct draw_context *draw, 
+                        unsigned prim,
+                        unsigned start, 
+                        unsigned count)
+{
+   float *hw_verts;
+
+   if (!set_prim(draw, prim))
+      return FALSE;
+
+   if (!update_shader( draw ))
+      return FALSE;
+
+   hw_verts = draw->render->allocate_vertices( draw->render,
+                                               draw->pt.hw_vertex_size,
+                                               count );
+
+   if (!hw_verts)
+      return FALSE;
+					
+   /* Single routine to fetch vertices, run shader and emit HW verts.
+    * Clipping and viewport transformation are done on hardware.
+    */
+   draw->vertex_fetch.pt_fetch( draw, 
+				hw_verts,
+				start, count );
+
+   /* Draw arrays path to avoid re-emitting index list again and
+    * again.
+    */
+   draw->render->draw_arrays( draw->render,
+                              start,
+                              count );
+   
+
+   draw->render->release_vertices( draw->render, 
+				   hw_verts, 
+				   draw->pt.hw_vertex_size, 
+				   count );
+
+   return TRUE;
+}
+
diff --git a/src/gallium/auxiliary/draw/draw_private.h b/src/gallium/auxiliary/draw/draw_private.h
index c732d723a7..4147472d45 100644
--- a/src/gallium/auxiliary/draw/draw_private.h
+++ b/src/gallium/auxiliary/draw/draw_private.h
@@ -133,7 +133,7 @@ struct draw_vertex_shader {
 
    /* This member will disappear shortly:
     */
-   const struct pipe_shader_state   *state;
+   struct pipe_shader_state   state;
 
    struct tgsi_shader_info info;
 
@@ -162,8 +162,14 @@ typedef void (*full_fetch_func)( struct draw_context *draw,
 				 const unsigned *elts,
 				 unsigned count );
 
+typedef void (*pt_fetch_func)( struct draw_context *draw,
+			      float *out,
+			      unsigned start,
+			       unsigned count );
 
 
+struct vbuf_render;
+
 /**
  * Private context for the drawing module.
  */
@@ -191,6 +197,17 @@ struct draw_context
       struct draw_stage *rasterize;
    } pipeline;
 
+
+   struct vbuf_render *render;
+
+   /* Support prototype passthrough path:
+    */
+   struct {
+      unsigned prim;
+      unsigned hw_vertex_size;
+   } pt;
+
+
    /* pipe state that we need: */
    const struct pipe_rasterizer_state *rasterizer;
    struct pipe_viewport_state viewport;
@@ -244,6 +261,7 @@ struct draw_context
       fetch_func fetch[PIPE_ATTRIB_MAX];
       unsigned nr_attrs;
       full_fetch_func fetch_func;
+      pt_fetch_func pt_fetch;
    } vertex_fetch;
 
    /* Post-tnl vertex cache:
@@ -331,6 +349,15 @@ struct tgsi_exec_machine;
 extern void draw_update_vertex_fetch( struct draw_context *draw );
 
 
+/* Prototype/hack
+ */
+boolean
+draw_passthrough_arrays(struct draw_context *draw, 
+                        unsigned prim,
+                        unsigned start, 
+                        unsigned count);
+
+
 #define DRAW_FLUSH_SHADER_QUEUE              0x1 /* sized not to overflow, never raised */
 #define DRAW_FLUSH_PRIM_QUEUE                0x2
 #define DRAW_FLUSH_VERTEX_CACHE              0x4
diff --git a/src/gallium/auxiliary/draw/draw_pstipple.c b/src/gallium/auxiliary/draw/draw_pstipple.c
index 2cfeb813b3..8b3e84a9a0 100644
--- a/src/gallium/auxiliary/draw/draw_pstipple.c
+++ b/src/gallium/auxiliary/draw/draw_pstipple.c
@@ -67,16 +67,18 @@ struct pstip_stage
    struct draw_stage stage;
 
    void *sampler_cso;
-   struct pipe_texture *texture;
    uint sampler_unit;
+   struct pipe_texture *texture;
+   uint num_samplers;
+   uint num_textures;
 
    /*
     * Currently bound state
     */
    struct pstip_fragment_shader *fs;
    struct {
-      void *sampler[PIPE_MAX_SAMPLERS];
-      struct pipe_texture *texture[PIPE_MAX_SAMPLERS];
+      void *samplers[PIPE_MAX_SAMPLERS];
+      struct pipe_texture *textures[PIPE_MAX_SAMPLERS];
       const struct pipe_poly_stipple *stipple;
    } state;
 
@@ -88,11 +90,10 @@ struct pstip_stage
    void (*driver_bind_fs_state)(struct pipe_context *, void *);
    void (*driver_delete_fs_state)(struct pipe_context *, void *);
 
-   void (*driver_bind_sampler_state)(struct pipe_context *, unsigned, void *);
+   void (*driver_bind_sampler_states)(struct pipe_context *, unsigned, void **);
 
-   void (*driver_set_sampler_texture)(struct pipe_context *,
-                                      unsigned sampler,
-                                      struct pipe_texture *);
+   void (*driver_set_sampler_textures)(struct pipe_context *, unsigned,
+                                       struct pipe_texture **);
 
    void (*driver_set_polygon_stipple)(struct pipe_context *,
                                       const struct pipe_poly_stipple *);
@@ -484,18 +485,25 @@ static void
 pstip_first_tri(struct draw_stage *stage, struct prim_header *header)
 {
    struct pstip_stage *pstip = pstip_stage(stage);
-   struct draw_context *draw = stage->draw;
    struct pipe_context *pipe = pstip->pipe;
+   uint num_samplers;
 
-   assert(draw->rasterizer->poly_stipple_enable);
+   /* how many samplers? */
+   /* we'll use sampler/texture[pstip->sampler_unit] for the stipple */
+   num_samplers = MAX2(pstip->num_textures, pstip->num_samplers);
+   num_samplers = MAX2(num_samplers, pstip->sampler_unit + 1);
 
-   /*
-    * Bind our fragprog, sampler and texture
-    */
+   assert(stage->draw->rasterizer->poly_stipple_enable);
+
+   /* bind our fragprog */
    bind_pstip_fragment_shader(pstip);
 
-   pstip->driver_bind_sampler_state(pipe, pstip->sampler_unit, pstip->sampler_cso);
-   pstip->driver_set_sampler_texture(pipe, pstip->sampler_unit, pstip->texture);
+   /* plug in our sampler, texture */
+   pstip->state.samplers[pstip->sampler_unit] = pstip->sampler_cso;
+   pstip->state.textures[pstip->sampler_unit] = pstip->texture;
+
+   pstip->driver_bind_sampler_states(pipe, num_samplers, pstip->state.samplers);
+   pstip->driver_set_sampler_textures(pipe, num_samplers, pstip->state.textures);
 
    /* now really draw first line */
    stage->tri = passthrough_tri;
@@ -517,10 +525,10 @@ pstip_flush(struct draw_stage *stage, unsigned flags)
    pstip->driver_bind_fs_state(pipe, pstip->fs->driver_fs);
 
    /* XXX restore original texture, sampler state */
-   pstip->driver_bind_sampler_state(pipe, pstip->sampler_unit,
-                                 pstip->state.sampler[pstip->sampler_unit]);
-   pstip->driver_set_sampler_texture(pipe, pstip->sampler_unit,
-                                 pstip->state.texture[pstip->sampler_unit]);
+   pstip->driver_bind_sampler_states(pipe, pstip->num_samplers,
+                                     pstip->state.samplers);
+   pstip->driver_set_sampler_textures(pipe, pstip->num_textures,
+                                      pstip->state.textures);
 }
 
 
@@ -597,7 +605,8 @@ pstip_bind_fs_state(struct pipe_context *pipe, void *fs)
    /* save current */
    pstip->fs = aafs;
    /* pass-through */
-   pstip->driver_bind_fs_state(pstip->pipe, aafs->driver_fs);
+   pstip->driver_bind_fs_state(pstip->pipe,
+                               (aafs ? aafs->driver_fs : NULL));
 }
 
 
@@ -613,26 +622,28 @@ pstip_delete_fs_state(struct pipe_context *pipe, void *fs)
 
 
 static void
-pstip_bind_sampler_state(struct pipe_context *pipe,
-                         unsigned unit, void *sampler)
+pstip_bind_sampler_states(struct pipe_context *pipe,
+                          unsigned num, void **sampler)
 {
    struct pstip_stage *pstip = pstip_stage_from_pipe(pipe);
    /* save current */
-   pstip->state.sampler[unit] = sampler;
+   memcpy(pstip->state.samplers, sampler, num * sizeof(void *));
+   pstip->num_samplers = num;
    /* pass-through */
-   pstip->driver_bind_sampler_state(pstip->pipe, unit, sampler);
+   pstip->driver_bind_sampler_states(pstip->pipe, num, sampler);
 }
 
 
 static void
-pstip_set_sampler_texture(struct pipe_context *pipe,
-                          unsigned sampler, struct pipe_texture *texture)
+pstip_set_sampler_textures(struct pipe_context *pipe,
+                           unsigned num, struct pipe_texture **texture)
 {
    struct pstip_stage *pstip = pstip_stage_from_pipe(pipe);
    /* save current */
-   pstip->state.texture[sampler] = texture;
+   memcpy(pstip->state.textures, texture, num * sizeof(struct pipe_texture *));
+   pstip->num_textures = num;
    /* pass-through */
-   pstip->driver_set_sampler_texture(pstip->pipe, sampler, texture);
+   pstip->driver_set_sampler_textures(pstip->pipe, num, texture);
 }
 
 
@@ -682,8 +693,8 @@ draw_install_pstipple_stage(struct draw_context *draw,
    pstip->driver_bind_fs_state = pipe->bind_fs_state;
    pstip->driver_delete_fs_state = pipe->delete_fs_state;
 
-   pstip->driver_bind_sampler_state = pipe->bind_sampler_state;
-   pstip->driver_set_sampler_texture = pipe->set_sampler_texture;
+   pstip->driver_bind_sampler_states = pipe->bind_sampler_states;
+   pstip->driver_set_sampler_textures = pipe->set_sampler_textures;
    pstip->driver_set_polygon_stipple = pipe->set_polygon_stipple;
 
    /* override the driver's functions */
@@ -691,7 +702,7 @@ draw_install_pstipple_stage(struct draw_context *draw,
    pipe->bind_fs_state = pstip_bind_fs_state;
    pipe->delete_fs_state = pstip_delete_fs_state;
 
-   pipe->bind_sampler_state = pstip_bind_sampler_state;
-   pipe->set_sampler_texture = pstip_set_sampler_texture;
+   pipe->bind_sampler_states = pstip_bind_sampler_states;
+   pipe->set_sampler_textures = pstip_set_sampler_textures;
    pipe->set_polygon_stipple = pstip_set_polygon_stipple;
 }
diff --git a/src/gallium/auxiliary/draw/draw_unfilled.c b/src/gallium/auxiliary/draw/draw_unfilled.c
index 4d718d514c..b07860cd9e 100644
--- a/src/gallium/auxiliary/draw/draw_unfilled.c
+++ b/src/gallium/auxiliary/draw/draw_unfilled.c
@@ -129,7 +129,7 @@ static void unfilled_tri( struct draw_stage *stage,
       points( stage, header );
       break;
    default:
-      abort();
+      assert(0);
    }   
 }
 
diff --git a/src/gallium/auxiliary/draw/draw_vbuf.h b/src/gallium/auxiliary/draw/draw_vbuf.h
index cfd2b9820c..5e7de905c1 100644
--- a/src/gallium/auxiliary/draw/draw_vbuf.h
+++ b/src/gallium/auxiliary/draw/draw_vbuf.h
@@ -85,6 +85,12 @@ struct vbuf_render {
 		 const ushort *indices,
 		 uint nr_indices );
 
+   /* Draw Arrays path too.
+    */
+   void (*draw_arrays)( struct vbuf_render *,
+			unsigned start,
+			uint nr );
+
    /**
     * Called when vbuf is done with this set of vertices:
     */
diff --git a/src/gallium/auxiliary/draw/draw_vertex_cache.c b/src/gallium/auxiliary/draw/draw_vertex_cache.c
index 53f8bbec44..161b247d4e 100644
--- a/src/gallium/auxiliary/draw/draw_vertex_cache.c
+++ b/src/gallium/auxiliary/draw/draw_vertex_cache.c
@@ -41,7 +41,11 @@ void draw_vertex_cache_invalidate( struct draw_context *draw )
    assert(draw->vs.queue_nr == 0);
    assert(draw->vcache.referenced == 0);
 
-//   memset(draw->vcache.idx, ~0, sizeof(draw->vcache.idx));
+   /* There's an error somewhere in the vcache code that requires this
+    * memset.  The bug is exposed in q3demo demo001, but probably
+    * elsewhere as well.  Will track it down later.
+    */
+   memset(draw->vcache.idx, ~0, sizeof(draw->vcache.idx));
 }
 
 
diff --git a/src/gallium/auxiliary/draw/draw_vertex_fetch.c b/src/gallium/auxiliary/draw/draw_vertex_fetch.c
index cb8cdd04a3..b56d85396d 100644
--- a/src/gallium/auxiliary/draw/draw_vertex_fetch.c
+++ b/src/gallium/auxiliary/draw/draw_vertex_fetch.c
@@ -54,7 +54,7 @@ fetch_##NAME(const void *ptr, float *attrib)		\
    int i;						\
 							\
    for (i = 0; i < SZ; i++) {				\
-      attrib[i] = CVT;					\
+      attrib[i] = CVT(i);                              \
    }							\
 							\
    for (; i < 4; i++) {					\
@@ -62,24 +62,24 @@ fetch_##NAME(const void *ptr, float *attrib)		\
    }							\
 }
 
-#define CVT_64_FLOAT   (float) ((double *) ptr)[i]
-#define CVT_32_FLOAT   ((float *) ptr)[i]
+#define CVT_64_FLOAT(i)   (float) ((double *) ptr)[i]
+#define CVT_32_FLOAT(i)   ((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_USCALED(i)  (float) ((unsigned char *) ptr)[i]
+#define CVT_16_USCALED(i) (float) ((unsigned short *) ptr)[i]
+#define CVT_32_USCALED(i) (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_SSCALED(i)  (float) ((char *) ptr)[i]
+#define CVT_16_SSCALED(i) (float) ((short *) ptr)[i]
+#define CVT_32_SSCALED(i) (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_UNORM(i)    (float) ((unsigned char *) ptr)[i] / 255.0f
+#define CVT_16_UNORM(i)   (float) ((unsigned short *) ptr)[i] / 65535.0f
+#define CVT_32_UNORM(i)   (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
+#define CVT_8_SNORM(i)    (float) ((char *) ptr)[i] / 127.0f
+#define CVT_16_SNORM(i)   (float) ((short *) ptr)[i] / 32767.0f
+#define CVT_32_SNORM(i)   (float) ((int *) ptr)[i] / 2147483647.0f
 
 FETCH_ATTRIB( R64G64B64A64_FLOAT,   4, CVT_64_FLOAT )
 FETCH_ATTRIB( R64G64B64_FLOAT,      3, CVT_64_FLOAT )
@@ -156,6 +156,16 @@ FETCH_ATTRIB( A8R8G8B8_UNORM,       4, CVT_8_UNORM )
 
 
 
+static void
+fetch_B8G8R8A8_UNORM(const void *ptr, float *attrib)
+{
+   attrib[2] = CVT_8_UNORM(0);
+   attrib[1] = CVT_8_UNORM(1);
+   attrib[0] = CVT_8_UNORM(2);
+   attrib[3] = CVT_8_UNORM(3);
+}
+
+
 static fetch_func get_fetch_func( enum pipe_format format )
 {
 #if 0
@@ -296,6 +306,10 @@ static fetch_func get_fetch_func( enum pipe_format format )
    case PIPE_FORMAT_A8R8G8B8_UNORM:
       return fetch_A8R8G8B8_UNORM;
 
+
+   case PIPE_FORMAT_B8G8R8A8_UNORM:
+      return fetch_B8G8R8A8_UNORM;
+
    case 0:
       return NULL;		/* not sure why this is needed */
 
diff --git a/src/gallium/auxiliary/draw/draw_vertex_shader.c b/src/gallium/auxiliary/draw/draw_vertex_shader.c
index 1e95355555..133418baca 100644
--- a/src/gallium/auxiliary/draw/draw_vertex_shader.c
+++ b/src/gallium/auxiliary/draw/draw_vertex_shader.c
@@ -110,13 +110,20 @@ draw_bind_vertex_shader(struct draw_context *draw,
                         struct draw_vertex_shader *dvs)
 {
    draw_do_flush( draw, DRAW_FLUSH_STATE_CHANGE );
+   
+   if (dvs) 
+   {
+      draw->vertex_shader = dvs;
+      draw->num_vs_outputs = dvs->info.num_outputs;
 
-   draw->vertex_shader = dvs;
-   draw->num_vs_outputs = dvs->info.num_outputs;
+      tgsi_exec_machine_init(&draw->machine);
 
-   tgsi_exec_machine_init(&draw->machine);
-
-   dvs->prepare( dvs, draw );
+      dvs->prepare( dvs, draw );
+   }
+   else {
+      draw->vertex_shader = NULL;
+      draw->num_vs_outputs = 0;
+   }
 }
 
 
diff --git a/src/gallium/auxiliary/draw/draw_vs_exec.c b/src/gallium/auxiliary/draw/draw_vs_exec.c
index 583812aadd..55bec14116 100644
--- a/src/gallium/auxiliary/draw/draw_vs_exec.c
+++ b/src/gallium/auxiliary/draw/draw_vs_exec.c
@@ -71,7 +71,7 @@ vs_exec_prepare( struct draw_vertex_shader *shader,
 {
    /* specify the vertex program to interpret/execute */
    tgsi_exec_machine_bind_shader(&draw->machine,
-				 shader->state->tokens,
+				 shader->state.tokens,
 				 PIPE_MAX_SAMPLERS,
 				 NULL /*samplers*/ );
 
@@ -132,20 +132,30 @@ vs_exec_run( struct draw_vertex_shader *shader,
       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 (!draw->rasterizer->bypass_clipping) {
+         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;
+      }
+      else {
+         vOut[j]->clipmask = 0;
+         vOut[j]->edgeflag = 1;
+         vOut[j]->data[0][0] = x;
+         vOut[j]->data[0][1] = y;
+         vOut[j]->data[0][2] = z;
+         vOut[j]->data[0][3] = w;
+      }
 
       /* Remaining attributes are packed into sequential post-transform
        * vertex attrib slots.
@@ -177,7 +187,7 @@ draw_create_vs_exec(struct draw_context *draw,
    if (vs == NULL) 
       return NULL;
 
-   vs->state = state;
+   vs->state = *state;
    vs->prepare = vs_exec_prepare;
    vs->run = vs_exec_run;
    vs->delete = vs_exec_delete;
diff --git a/src/gallium/auxiliary/draw/draw_vs_llvm.c b/src/gallium/auxiliary/draw/draw_vs_llvm.c
index 0fd557d667..53c260be53 100644
--- a/src/gallium/auxiliary/draw/draw_vs_llvm.c
+++ b/src/gallium/auxiliary/draw/draw_vs_llvm.c
@@ -135,25 +135,35 @@ vs_llvm_run( struct draw_vertex_shader *base,
       unsigned slot;
       float x, y, z, w;
 
-      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 (!draw->rasterizer->bypass_clipping) {
+         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;
+      }
+      else {
+         vOut[j]->clipmask = 0;
+         vOut[j]->edgeflag = 1;
+         vOut[j]->data[0][0] = x;
+         vOut[j]->data[0][1] = y;
+         vOut[j]->data[0][2] = z;
+         vOut[j]->data[0][3] = w;
+      }
 
       /* Remaining attributes are packed into sequential post-transform
        * vertex attrib slots.
diff --git a/src/gallium/auxiliary/draw/draw_vs_sse.c b/src/gallium/auxiliary/draw/draw_vs_sse.c
index 0b8bc2bf14..5ee2adb344 100644
--- a/src/gallium/auxiliary/draw/draw_vs_sse.c
+++ b/src/gallium/auxiliary/draw/draw_vs_sse.c
@@ -158,20 +158,30 @@ vs_sse_run( struct draw_vertex_shader *base,
       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 (!draw->rasterizer->bypass_clipping) {
+         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;
+      }
+      else {
+         vOut[j]->clipmask = 0;
+         vOut[j]->edgeflag = 1;
+         vOut[j]->data[0][0] = x;
+         vOut[j]->data[0][1] = y;
+         vOut[j]->data[0][2] = z;
+         vOut[j]->data[0][3] = w;
+      }
 
       /* Remaining attributes are packed into sequential post-transform
        * vertex attrib slots.
@@ -211,14 +221,14 @@ draw_create_vs_sse(struct draw_context *draw,
    if (vs == NULL) 
       return NULL;
 
-   vs->base.state = templ;
+   vs->base.state = *templ;
    vs->base.prepare = vs_sse_prepare;
    vs->base.run = vs_sse_run;
    vs->base.delete = vs_sse_delete;
    
    x86_init_func( &vs->sse2_program );
 
-   if (!tgsi_emit_sse2( (struct tgsi_token *) vs->base.state->tokens,
+   if (!tgsi_emit_sse2( (struct tgsi_token *) vs->base.state.tokens,
 			&vs->sse2_program )) 
       goto fail;
       
diff --git a/src/gallium/auxiliary/gallivm/Makefile b/src/gallium/auxiliary/gallivm/Makefile
index 39fac6ea4a..c24e19e062 100644
--- a/src/gallium/auxiliary/gallivm/Makefile
+++ b/src/gallium/auxiliary/gallivm/Makefile
@@ -15,7 +15,7 @@ GALLIVM_SOURCES = \
         storagesoa.cpp \
         instructionssoa.cpp
 
-INC_SOURCES = gallivm_builtins.cpp
+INC_SOURCES = gallivm_builtins.cpp gallivmsoabuiltins.cpp
 
 CPP_SOURCES = \
 	$(GALLIVM_SOURCES)
@@ -65,8 +65,10 @@ depend: $(C_SOURCES) $(CPP_SOURCES) $(ASM_SOURCES) $(INC_SOURCES)
 
 
 gallivm_builtins.cpp: llvm_builtins.c
-	clang --emit-llvm $< |llvm-as|opt -std-compile-opts|llvm2cpp -gen-contents -o=$@ -f -for=shader -funcname=createGallivmBuiltins
+	clang --emit-llvm < $< |llvm-as|opt -std-compile-opts|llvm2cpp -gen-contents -o=$@ -f -for=shader -funcname=createGallivmBuiltins
 
+gallivmsoabuiltins.cpp: soabuiltins.c
+	clang --emit-llvm < $< |llvm-as|opt -std-compile-opts|llvm2cpp -gen-module -o=$@ -f -for=shader -funcname=createSoaBuiltins
 
 # Emacs tags
 tags:
@@ -78,6 +80,7 @@ clean:
 	-rm -f *.o */*.o *~ *.so *~ server/*.o
 	-rm -f depend depend.bak
 	-rm -f gallivm_builtins.cpp
+	-rm -f gallivmsoabuiltins.cpp
 
 symlinks:
 
diff --git a/src/gallium/auxiliary/gallivm/gallivm.cpp b/src/gallium/auxiliary/gallivm/gallivm.cpp
index d14bb3b99a..b6f641a3f8 100644
--- a/src/gallium/auxiliary/gallivm/gallivm.cpp
+++ b/src/gallium/auxiliary/gallivm/gallivm.cpp
@@ -306,11 +306,19 @@ struct gallivm_prog * gallivm_ir_compile(struct gallivm_ir *ir)
 {
    struct gallivm_prog *prog =
       (struct gallivm_prog *)calloc(1, sizeof(struct gallivm_prog));
+   
+   std::cout << "Before optimizations:"<<std::endl;
+   ir->module->dump();
+   std::cout<<"-------------------------------"<<std::endl;
+   
+   PassManager veri;
+   veri.add(createVerifierPass());
+   veri.run(*ir->module);
    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));
diff --git a/src/gallium/auxiliary/gallivm/gallivm.h b/src/gallium/auxiliary/gallivm/gallivm.h
index 57912a952f..b4d6555d2f 100644
--- a/src/gallium/auxiliary/gallivm/gallivm.h
+++ b/src/gallium/auxiliary/gallivm/gallivm.h
@@ -33,6 +33,16 @@
 #ifndef GALLIVM_H
 #define GALLIVM_H
 
+/*
+  LLVM representation consists of two stages - layout independent
+  intermediate representation gallivm_ir and driver specific
+  gallivm_prog. TGSI is first being translated into gallivm_ir
+  after that driver can set number of options on gallivm_ir and
+  have it compiled into gallivm_prog. gallivm_prog can be either
+  executed (assuming there's LLVM JIT backend for the current
+  target) or machine code generation can be done (assuming there's
+  a LLVM code generator for thecurrent target)
+ */
 #if defined __cplusplus
 extern "C" {
 #endif
diff --git a/src/gallium/auxiliary/gallivm/instructions.cpp b/src/gallium/auxiliary/gallivm/instructions.cpp
index 55d39fa5f1..8919491792 100644
--- a/src/gallium/auxiliary/gallivm/instructions.cpp
+++ b/src/gallium/auxiliary/gallivm/instructions.cpp
@@ -42,6 +42,7 @@
 #include <llvm/InstrTypes.h>
 #include <llvm/Instructions.h>
 #include <llvm/ParameterAttributes.h>
+#include <llvm/ParamAttrsList.h>
 
 #include <sstream>
 #include <fstream>
@@ -51,6 +52,15 @@ using namespace llvm;
 
 #include "gallivm_builtins.cpp"
 
+#if 0
+
+llvm::Value *arrayFromChannels(std::vector<llvm::Value*> &vals)
+{
+   VectorType *vectorType = VectorType::get(Type::FloatTy, 4);
+   ArrayType *vectorArray = ArrayType::get(vectorType, 4);
+}
+#endif
+
 static inline std::string createFuncName(int label)
 {
    std::ostringstream stream;
diff --git a/src/gallium/auxiliary/gallivm/instructionssoa.cpp b/src/gallium/auxiliary/gallivm/instructionssoa.cpp
index a4d5046637..89d513afd0 100644
--- a/src/gallium/auxiliary/gallivm/instructionssoa.cpp
+++ b/src/gallium/auxiliary/gallivm/instructionssoa.cpp
@@ -2,9 +2,28 @@
 
 #include "storagesoa.h"
 
+#include "pipe/p_shader_tokens.h"
+
+#include <llvm/CallingConv.h>
 #include <llvm/Constants.h>
+#include <llvm/Module.h>
+#include <llvm/Function.h>
+#include <llvm/Instructions.h>
+#include <llvm/Transforms/Utils/Cloning.h>
+#include <llvm/ParamAttrsList.h>
+
+#include <iostream>
 
+
+/* disable some warnings. this file is autogenerated */
+#if defined(__GNUC__)
+#pragma GCC diagnostic ignored "-Wunused-variable"
+#endif
 using namespace llvm;
+#include "gallivmsoabuiltins.cpp"
+#if defined(__GNUC__)
+#pragma GCC diagnostic warning "-Wunused-variable"
+#endif
 
 InstructionsSoa::InstructionsSoa(llvm::Module *mod, llvm::Function *func,
                                  llvm::BasicBlock *block, StorageSoa *storage)
@@ -12,6 +31,8 @@ InstructionsSoa::InstructionsSoa(llvm::Module *mod, llvm::Function *func,
      m_storage(storage),
      m_idx(0)
 {
+   createFunctionMap();
+   createBuiltins();
 }
 
 const char * InstructionsSoa::name(const char *prefix) const
@@ -119,3 +140,167 @@ std::vector<llvm::Value*> InstructionsSoa::extractVector(llvm::Value *vector)
 
    return res;
 }
+
+void InstructionsSoa::createFunctionMap()
+{
+   m_functionsMap[TGSI_OPCODE_DP3] = "dp3";
+   m_functionsMap[TGSI_OPCODE_DP4] = "dp4";
+}
+
+llvm::Function * InstructionsSoa::function(int op)
+{
+    if (m_functions.find(op) != m_functions.end())
+       return m_functions[op];
+
+    std::string name = m_functionsMap[op];
+
+    llvm::Function *originalFunc = m_builtins->getFunction(name);
+    llvm::Function *func = CloneFunction(originalFunc);
+    currentModule()->getFunctionList().push_back(func);
+    std::cout << "Func parent is "<<func->getParent()
+              <<", cur is "<<currentModule() <<std::endl;
+    func->dump();
+       //func->setParent(currentModule());
+    m_functions[op] = func;
+    return func;
+}
+
+llvm::Module * InstructionsSoa::currentModule() const
+{
+   BasicBlock *block = m_builder.GetInsertBlock();
+   if (!block || !block->getParent())
+      return 0;
+
+   return block->getParent()->getParent();
+}
+
+void InstructionsSoa::createBuiltins()
+{
+   m_builtins = createSoaBuiltins();
+}
+
+std::vector<llvm::Value*> InstructionsSoa::dp3(const std::vector<llvm::Value*> in1,
+                                               const std::vector<llvm::Value*> in2)
+{
+   llvm::Function *func = function(TGSI_OPCODE_DP3);
+   return callBuiltin(func, in1, in2);
+}
+
+llvm::Value * InstructionsSoa::allocaTemp()
+{
+   VectorType *vector   = VectorType::get(Type::FloatTy, 4);
+   ArrayType  *vecArray = ArrayType::get(vector, 4);
+   AllocaInst *alloca = new AllocaInst(vecArray, name("tmpRes"),
+                                       m_builder.GetInsertBlock());
+
+   std::vector<Value*> indices;
+   indices.push_back(m_storage->constantInt(0));
+   indices.push_back(m_storage->constantInt(0));
+   GetElementPtrInst *getElem = new GetElementPtrInst(alloca,
+                                                      indices.begin(),
+                                                      indices.end(),
+                                                      name("allocaPtr"),
+                                                      m_builder.GetInsertBlock());
+   return getElem;
+}
+
+std::vector<llvm::Value*> InstructionsSoa::allocaToResult(llvm::Value *allocaPtr)
+{
+   GetElementPtrInst *xElemPtr =  new GetElementPtrInst(allocaPtr,
+                                                        m_storage->constantInt(0),
+                                                        name("xPtr"),
+                                                        m_builder.GetInsertBlock());
+   GetElementPtrInst *yElemPtr =  new GetElementPtrInst(allocaPtr,
+                                                        m_storage->constantInt(1),
+                                                        name("yPtr"),
+                                                        m_builder.GetInsertBlock());
+   GetElementPtrInst *zElemPtr =  new GetElementPtrInst(allocaPtr,
+                                                        m_storage->constantInt(2),
+                                                        name("zPtr"),
+                                                        m_builder.GetInsertBlock());
+   GetElementPtrInst *wElemPtr =  new GetElementPtrInst(allocaPtr,
+                                                        m_storage->constantInt(3),
+                                                        name("wPtr"),
+                                                        m_builder.GetInsertBlock());
+
+   std::vector<llvm::Value*> res(4);
+   res[0] = new LoadInst(xElemPtr, name("xRes"), false, m_builder.GetInsertBlock());
+   res[1] = new LoadInst(yElemPtr, name("yRes"), false, m_builder.GetInsertBlock());
+   res[2] = new LoadInst(zElemPtr, name("zRes"), false, m_builder.GetInsertBlock());
+   res[3] = new LoadInst(wElemPtr, name("wRes"), false, m_builder.GetInsertBlock());
+
+   return res;
+}
+
+std::vector<llvm::Value*> InstructionsSoa::dp4(const std::vector<llvm::Value*> in1,
+                                               const std::vector<llvm::Value*> in2)
+{
+   llvm::Function *func = function(TGSI_OPCODE_DP4);
+   return callBuiltin(func, in1, in2);
+}
+
+std::vector<Value*> InstructionsSoa::callBuiltin(llvm::Function *func, const std::vector<llvm::Value*> in1)
+{
+   std::vector<Value*> params;
+
+   llvm::Value *allocaPtr = allocaTemp();
+   params.push_back(allocaPtr);
+   params.push_back(in1[0]);
+   params.push_back(in1[1]);
+   params.push_back(in1[2]);
+   params.push_back(in1[3]);
+   CallInst *call = m_builder.CreateCall(func, params.begin(), params.end());
+   call->setCallingConv(CallingConv::C);
+   call->setTailCall(false);
+
+   return allocaToResult(allocaPtr);
+}
+
+std::vector<Value*> InstructionsSoa::callBuiltin(llvm::Function *func, const std::vector<llvm::Value*> in1,
+                                                 const std::vector<llvm::Value*> in2)
+{
+   std::vector<Value*> params;
+
+   llvm::Value *allocaPtr = allocaTemp();
+   params.push_back(allocaPtr);
+   params.push_back(in1[0]);
+   params.push_back(in1[1]);
+   params.push_back(in1[2]);
+   params.push_back(in1[3]);
+   params.push_back(in2[0]);
+   params.push_back(in2[1]);
+   params.push_back(in2[2]);
+   params.push_back(in2[3]);
+   CallInst *call = m_builder.CreateCall(func, params.begin(), params.end());
+   call->setCallingConv(CallingConv::C);
+   call->setTailCall(false);
+
+   return allocaToResult(allocaPtr);
+}
+
+std::vector<Value*> InstructionsSoa::callBuiltin(llvm::Function *func, const std::vector<llvm::Value*> in1,
+                                                 const std::vector<llvm::Value*> in2,
+                                                 const std::vector<llvm::Value*> in3)
+{
+   std::vector<Value*> params;
+
+   llvm::Value *allocaPtr = allocaTemp();
+   params.push_back(allocaPtr);
+   params.push_back(in1[0]);
+   params.push_back(in1[1]);
+   params.push_back(in1[2]);
+   params.push_back(in1[3]);
+   params.push_back(in2[0]);
+   params.push_back(in2[1]);
+   params.push_back(in2[2]);
+   params.push_back(in2[3]);
+   params.push_back(in3[0]);
+   params.push_back(in3[1]);
+   params.push_back(in3[2]);
+   params.push_back(in3[3]);
+   CallInst *call = m_builder.CreateCall(func, params.begin(), params.end());
+   call->setCallingConv(CallingConv::C);
+   call->setTailCall(false);
+
+   return allocaToResult(allocaPtr);
+}
diff --git a/src/gallium/auxiliary/gallivm/instructionssoa.h b/src/gallium/auxiliary/gallivm/instructionssoa.h
index 4169dcbb2e..3ef51dcaff 100644
--- a/src/gallium/auxiliary/gallivm/instructionssoa.h
+++ b/src/gallium/auxiliary/gallivm/instructionssoa.h
@@ -30,6 +30,7 @@
 
 #include <llvm/Support/LLVMBuilder.h>
 
+#include <map>
 #include <vector>
 
 namespace llvm {
@@ -47,9 +48,12 @@ public:
                    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*> dp3(const std::vector<llvm::Value*> in1,
+                                 const std::vector<llvm::Value*> in2);
+   std::vector<llvm::Value*> dp4(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);
@@ -62,9 +66,29 @@ private:
    const char * name(const char *prefix) const;
    llvm::Value *vectorFromVals(llvm::Value *x, llvm::Value *y,
                                llvm::Value *z, llvm::Value *w);
+   void createFunctionMap();
+   void createBuiltins();
+   llvm::Function *function(int);
+   llvm::Module *currentModule() const;
+   llvm::Value *allocaTemp();
+   std::vector<llvm::Value*> allocaToResult(llvm::Value *allocaPtr);
+   std::vector<llvm::Value*> callBuiltin(llvm::Function *func,
+                                         const std::vector<llvm::Value*> in1);
+   std::vector<llvm::Value*> callBuiltin(llvm::Function *func,
+                                         const std::vector<llvm::Value*> in1,
+                                         const std::vector<llvm::Value*> in2);
+   std::vector<llvm::Value*> callBuiltin(llvm::Function *func,
+                                         const std::vector<llvm::Value*> in1,
+                                         const std::vector<llvm::Value*> in2,
+                                         const std::vector<llvm::Value*> in3);
 private:
    llvm::LLVMFoldingBuilder  m_builder;
    StorageSoa *m_storage;
+
+   std::map<int, std::string> m_functionsMap;
+   std::map<int, llvm::Function*> m_functions;
+   llvm::Module *m_builtins;
+
 private:
    mutable int  m_idx;
    mutable char m_name[32];
diff --git a/src/gallium/auxiliary/gallivm/soabuiltins.c b/src/gallium/auxiliary/gallivm/soabuiltins.c
new file mode 100644
index 0000000000..24c14e1b69
--- /dev/null
+++ b/src/gallium/auxiliary/gallivm/soabuiltins.c
@@ -0,0 +1,72 @@
+/**************************************************************************
+ *
+ * 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 file is compiled with clang into the LLVM bitcode
+  *
+  * Authors:
+  *   Zack Rusin zack@tungstengraphics.com
+  */
+typedef __attribute__(( ocu_vector_type(4) )) float float4;
+
+void dp3(float4 *res,
+         float4 tmp0x, float4 tmp0y, float4 tmp0z, float4 tmp0w,
+         float4 tmp1x, float4 tmp1y, float4 tmp1z, float4 tmp1w)
+{
+   float4 dot = (tmp0x * tmp1x) + (tmp0y * tmp1y) +
+                (tmp0z * tmp1z);
+
+   res[0] = dot;
+   res[1] = dot;
+   res[2] = dot;
+   res[3] = dot;
+}
+
+
+void dp4(float4 *res,
+         float4 tmp0x, float4 tmp0y, float4 tmp0z, float4 tmp0w,
+         float4 tmp1x, float4 tmp1y, float4 tmp1z, float4 tmp1w)
+{
+   float4 dot = (tmp0x * tmp1x) + (tmp0y * tmp1y) +
+                (tmp0z * tmp1z) + (tmp0w * tmp1w);
+
+   res[0] = dot;
+   res[1] = dot;
+   res[2] = dot;
+   res[3] = dot;
+}
+
+#if 0
+void yo(float4 *out, float4 *in)
+{
+   float4 res[4];
+
+   dp3(res, in[0], in[1], in[2], in[3],
+       in[4], in[5], in[6], in[7]);
+   out[1] = res[1];
+}
+#endif
diff --git a/src/gallium/auxiliary/gallivm/storagesoa.cpp b/src/gallium/auxiliary/gallivm/storagesoa.cpp
index ed0674a96f..bb6fe3d7e1 100644
--- a/src/gallium/auxiliary/gallivm/storagesoa.cpp
+++ b/src/gallium/auxiliary/gallivm/storagesoa.cpp
@@ -277,7 +277,7 @@ llvm::Constant * StorageSoa::createConstGlobalVector(const std::vector<float> &v
    return constVector;
 }
 
-std::vector<llvm::Value*> StorageSoa::load(Argument type, int idx, int swizzle,
+std::vector<llvm::Value*> StorageSoa::load(enum tgsi_file_type type, int idx, int swizzle,
                                            llvm::Value *indIdx)
 {
    std::vector<llvm::Value*> val(4);
@@ -292,25 +292,29 @@ std::vector<llvm::Value*> StorageSoa::load(Argument type, int idx, int swizzle,
    debug_printf("XXXXXXXXX realIdx = %p, indIdx = %p\n", realIndex, indIdx);
 
    switch(type) {
-   case Input:
+   case TGSI_FILE_INPUT:
       val = inputElement(realIndex);
       break;
-   case Output:
+   case TGSI_FILE_OUTPUT:
       val = outputElement(realIndex);
       break;
-   case Temp:
+   case TGSI_FILE_TEMPORARY:
       val = tempElement(realIndex);
       break;
-   case Const:
+   case TGSI_FILE_CONSTANT:
       val = constElement(realIndex);
       break;
-   case Immediate:
+   case TGSI_FILE_IMMEDIATE:
       val = immediateElement(realIndex);
       break;
-   case Address:
+   case TGSI_FILE_ADDRESS:
       debug_printf("Address not handled in the load phase!\n");
       assert(0);
       break;
+   default:
+      debug_printf("Unknown load!\n");
+      assert(0);
+      break;
    }
    if (!gallivm_is_swizzle(swizzle))
       return val;
@@ -324,21 +328,21 @@ std::vector<llvm::Value*> StorageSoa::load(Argument type, int idx, int swizzle,
    return res;
 }
 
-void StorageSoa::store(Argument type, int idx, const std::vector<llvm::Value*> &val,
+void StorageSoa::store(enum tgsi_file_type type, int idx, const std::vector<llvm::Value*> &val,
                        int mask)
 {
    llvm::Value *out = 0;
    switch(type) {
-   case Output:
+   case TGSI_FILE_OUTPUT:
       out = m_output;
       break;
-   case Temp:
+   case TGSI_FILE_TEMPORARY:
       out = m_temps;
       break;
-   case Input:
+   case TGSI_FILE_INPUT:
       out = m_input;
       break;
-   case Address: {
+   case TGSI_FILE_ADDRESS: {
       llvm::Value *addr = m_addresses[idx];
       if (!addr) {
          addAddress(idx);
diff --git a/src/gallium/auxiliary/gallivm/storagesoa.h b/src/gallium/auxiliary/gallivm/storagesoa.h
index 6443351f27..ae2fc7c6ae 100644
--- a/src/gallium/auxiliary/gallivm/storagesoa.h
+++ b/src/gallium/auxiliary/gallivm/storagesoa.h
@@ -28,6 +28,8 @@
 #ifndef STORAGESOA_H
 #define STORAGESOA_H
 
+#include <pipe/p_shader_tokens.h>
+
 #include <vector>
 #include <list>
 #include <map>
@@ -46,15 +48,6 @@ namespace llvm {
 class StorageSoa
 {
 public:
-   enum Argument {
-      Input,
-      Output,
-      Temp,
-      Const,
-      Immediate,
-      Address
-   };
-public:
    StorageSoa(llvm::BasicBlock *block,
               llvm::Value *input,
               llvm::Value *output,
@@ -62,9 +55,9 @@ public:
               llvm::Value *temps);
 
 
-   std::vector<llvm::Value*> load(Argument type, int idx, int swizzle, 
+   std::vector<llvm::Value*> load(enum tgsi_file_type type, int idx, int swizzle, 
                                   llvm::Value *indIdx =0);
-   void store(Argument type, int idx, const std::vector<llvm::Value*> &val,
+   void store(enum tgsi_file_type type, int idx, const std::vector<llvm::Value*> &val,
               int mask);
 
    void addImmediate(float *vec);
diff --git a/src/gallium/auxiliary/gallivm/tgsitollvm.cpp b/src/gallium/auxiliary/gallivm/tgsitollvm.cpp
index 2cb4acce32..3f65865a5a 100644
--- a/src/gallium/auxiliary/gallivm/tgsitollvm.cpp
+++ b/src/gallium/auxiliary/gallivm/tgsitollvm.cpp
@@ -708,25 +708,9 @@ translate_instructionir(llvm::Module *module,
       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;
-      }
+
+      val = storage->load((enum tgsi_file_type)src->SrcRegister.File,
+                          src->SrcRegister.Index, swizzle, indIdx);
 
       inputs[i] = val;
    }
@@ -763,9 +747,11 @@ translate_instructionir(llvm::Module *module,
    }
       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: {
@@ -1067,19 +1053,8 @@ translate_instructionir(llvm::Module *module,
    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");
-      }
+      storage->store((enum tgsi_file_type)dst->DstRegister.File,
+                     dst->DstRegister.Index, out, dst->DstRegister.WriteMask);
    }
 }
 
diff --git a/src/gallium/auxiliary/rtasm/rtasm_ppc_spe.c b/src/gallium/auxiliary/rtasm/rtasm_ppc_spe.c
index 95a2d6fcbb..a996218ce7 100644
--- a/src/gallium/auxiliary/rtasm/rtasm_ppc_spe.c
+++ b/src/gallium/auxiliary/rtasm/rtasm_ppc_spe.c
@@ -306,6 +306,11 @@ void spe_init_func(struct spe_function *p, unsigned code_size)
 {
     p->store = align_malloc(code_size, 16);
     p->csr = p->store;
+    
+    /* Conservatively treat R0 - R2 and R80 - R127 as non-volatile.
+     */
+    p->regs[0] = ~7;
+    p->regs[1] = (1U << (80 - 64)) - 1;
 }
 
 
@@ -317,6 +322,48 @@ void spe_release_func(struct spe_function *p)
 }
 
 
+int spe_allocate_available_register(struct spe_function *p)
+{
+   unsigned i;
+   for (i = 0; i < 128; i++) {
+      const uint64_t mask = (1ULL << (i % 128));
+      const unsigned idx = i / 128;
+
+      if ((p->regs[idx] & mask) != 0) {
+         p->regs[idx] &= ~mask;
+         return i;
+      }
+   }
+
+   return -1;
+}
+
+
+int spe_allocate_register(struct spe_function *p, int reg)
+{
+   const unsigned idx = reg / 128;
+   const unsigned bit = reg % 128;
+
+   assert((p->regs[idx] & (1ULL << bit)) != 0);
+
+   p->regs[idx] &= ~(1ULL << bit);
+   return reg;
+}
+
+
+void spe_release_register(struct spe_function *p, int reg)
+{
+   const unsigned idx = reg / 128;
+   const unsigned bit = reg % 128;
+
+   assert((p->regs[idx] & (1ULL << bit)) == 0);
+
+   p->regs[idx] |= (1ULL << bit);
+}
+
+
+
+
 void spe_bi(struct spe_function *p, unsigned rA, int d, int e)
 {
     emit_RI7(p, 0x1a8, 0, rA, (d << 5) | (e << 4));
diff --git a/src/gallium/auxiliary/rtasm/rtasm_ppc_spe.h b/src/gallium/auxiliary/rtasm/rtasm_ppc_spe.h
index 10ce44b3a0..5a1eb1ed8d 100644
--- a/src/gallium/auxiliary/rtasm/rtasm_ppc_spe.h
+++ b/src/gallium/auxiliary/rtasm/rtasm_ppc_spe.h
@@ -39,11 +39,27 @@ struct spe_function {
     uint32_t *store;
     uint32_t *csr;
     const char *fn;
+
+    /**
+     * Mask of used / unused registers
+     *
+     * Each set bit corresponds to an available register.  Each cleared bit
+     * corresponds to an allocated register.
+     *
+     * \sa
+     * spe_allocate_register, spe_allocate_available_register,
+     * spe_release_register
+     */
+    uint64_t regs[2];
 };
 
 extern void spe_init_func(struct spe_function *p, unsigned code_size);
 extern void spe_release_func(struct spe_function *p);
 
+extern int spe_allocate_available_register(struct spe_function *p);
+extern int spe_allocate_register(struct spe_function *p, int reg);
+extern void spe_release_register(struct spe_function *p, int reg);
+
 #endif /* RTASM_PPC_SPE_H */
 
 #ifndef EMIT_
diff --git a/src/gallium/auxiliary/sct/Makefile b/src/gallium/auxiliary/sct/Makefile
new file mode 100644
index 0000000000..516d1756cf
--- /dev/null
+++ b/src/gallium/auxiliary/sct/Makefile
@@ -0,0 +1,12 @@
+TOP = ../../../..
+include $(TOP)/configs/current
+
+LIBNAME = sct
+
+C_SOURCES = \
+	sct.c
+
+include ../../Makefile.template
+
+symlinks:
+
diff --git a/src/gallium/auxiliary/sct/SConscript b/src/gallium/auxiliary/sct/SConscript
new file mode 100644
index 0000000000..76927d973f
--- /dev/null
+++ b/src/gallium/auxiliary/sct/SConscript
@@ -0,0 +1,9 @@
+Import('*')
+
+sct = env.ConvenienceLibrary(
+	target = 'sct',
+	source = [
+		'sct.c'
+	])
+
+auxiliaries.insert(0, sct)
diff --git a/src/gallium/auxiliary/sct/sct.c b/src/gallium/auxiliary/sct/sct.c
new file mode 100644
index 0000000000..97ee5882a1
--- /dev/null
+++ b/src/gallium/auxiliary/sct/sct.c
@@ -0,0 +1,453 @@
+/**************************************************************************
+ * 
+ * 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_util.h"
+#include "pipe/p_state.h"
+#include "pipe/p_inlines.h"
+#include "sct.h"
+
+
+struct texture_list
+{
+   struct pipe_texture *texture;
+   struct texture_list *next;
+};
+
+
+
+#define MAX_SURFACES  ((PIPE_MAX_COLOR_BUFS) + 1)
+
+struct sct_context
+{
+   const struct pipe_context *context;
+
+   /** surfaces the context is drawing into */
+   struct pipe_surface *surfaces[MAX_SURFACES];
+
+   /** currently bound textures */
+   struct pipe_texture *textures[PIPE_MAX_SAMPLERS];
+
+   /** previously bound textures, used but not flushed */
+   struct texture_list *textures_used;
+
+   boolean needs_flush;
+
+   struct sct_context *next;
+};
+
+
+
+struct sct_surface
+{
+   const struct pipe_surface *surface;
+
+   /** list of contexts drawing to this surface */
+   struct sct_context_list *contexts;
+
+   struct sct_surface *next;
+};
+
+
+
+/**
+ * Find the surface_info for the given pipe_surface
+ */
+static struct sct_surface *
+find_surface_info(struct surface_context_tracker *sct,
+                  const struct pipe_surface *surface)
+{
+   struct sct_surface *si;
+   for (si = sct->surfaces; si; si = si->next)
+      if (si->surface == surface)
+         return si;
+   return NULL;
+}
+
+
+/**
+ * As above, but create new surface_info if surface is new.
+ */
+static struct sct_surface *
+find_create_surface_info(struct surface_context_tracker *sct,
+                         const struct pipe_surface *surface)
+{
+   struct sct_surface *si = find_surface_info(sct, surface);
+   if (si)
+      return si;
+
+   /* alloc new */
+   si = CALLOC_STRUCT(sct_surface);
+   if (si) {
+      si->surface = surface;
+
+      /* insert at head */
+      si->next = sct->surfaces;
+      sct->surfaces = si;
+   }
+
+   return si;
+}
+
+
+/**
+ * Find a context_info for the given context.
+ */
+static struct sct_context *
+find_context_info(struct surface_context_tracker *sct,
+                  const struct pipe_context *context)
+{
+   struct sct_context *ci;
+   for (ci = sct->contexts; ci; ci = ci->next)
+      if (ci->context == context)
+         return ci;
+   return NULL;
+}
+
+
+/**
+ * As above, but create new context_info if context is new.
+ */
+static struct sct_context *
+find_create_context_info(struct surface_context_tracker *sct,
+                         const struct pipe_context *context)
+{
+   struct sct_context *ci = find_context_info(sct, context);
+   if (ci)
+      return ci;
+
+   /* alloc new */
+   ci = CALLOC_STRUCT(sct_context);
+   if (ci) {
+      ci->context = context;
+
+      /* insert at head */
+      ci->next = sct->contexts;
+      sct->contexts = ci;
+   }
+
+   return ci;
+}
+
+
+/**
+ * Is the context already bound to the surface?
+ */
+static boolean
+find_surface_context(const struct sct_surface *si,
+                     const struct pipe_context *context)
+{
+   const struct sct_context_list *cl;
+   for (cl = si->contexts; cl; cl = cl->next) {
+      if (cl->context == context) {
+         return TRUE;
+      }
+   }
+   return FALSE;
+}
+
+
+/**
+ * Add a context to the list of contexts associated with a surface.
+ */
+static void
+add_context_to_surface(struct sct_surface *si,
+                       const struct pipe_context *context)
+{
+   struct sct_context_list *cl = CALLOC_STRUCT(sct_context_list);
+   if (cl) {
+      cl->context = context;
+      /* insert at head of list of contexts */
+      cl->next = si->contexts;
+      si->contexts = cl;
+   }
+}
+
+
+/**
+ * Remove a context from the list of contexts associated with a surface.
+ */
+static void
+remove_context_from_surface(struct sct_surface *si,
+                            const struct pipe_context *context)
+{
+   struct sct_context_list *prev = NULL, *curr, *next;
+
+   for (curr = si->contexts; curr; curr = next) {
+      if (curr->context == context) {
+         /* remove */
+         if (prev)
+            prev->next = curr->next;
+         else
+            si->contexts = curr->next;
+         next = curr->next;
+         FREE(curr);
+      }
+      else {
+         prev = curr;
+      }
+   }
+}
+
+
+/**
+ * Unbind context from surface.
+ */
+static void
+unbind_context_surface(struct surface_context_tracker *sct,
+                       struct pipe_context *context,
+                       struct pipe_surface *surface)
+{
+   struct sct_surface *si = find_surface_info(sct, surface);
+   if (si) {
+      remove_context_from_surface(si, context);
+   }
+}
+
+
+/**
+ * Bind context to a set of surfaces (color + Z).
+ * Like MakeCurrent().
+ */
+void
+sct_bind_surfaces(struct surface_context_tracker *sct,
+                  struct pipe_context *context,
+                  uint num_surf,
+                  struct pipe_surface **surfaces)
+{
+   struct sct_context *ci = find_create_context_info(sct, context);
+   uint i;
+
+   if (!ci) {
+      return; /* out of memory */
+   }
+
+   /* unbind currently bound surfaces */
+   for (i = 0; i < MAX_SURFACES; i++) {
+      if (ci->surfaces[i]) {
+         unbind_context_surface(sct, context, ci->surfaces[i]);
+      }
+   }
+
+   /* bind new surfaces */
+   for (i = 0; i < num_surf; i++) {
+      struct sct_surface *si = find_create_surface_info(sct, surfaces[i]);
+      if (!find_surface_context(si, context)) {
+         add_context_to_surface(si, context);
+      }
+   }
+}
+
+
+/**
+ * Return list of contexts bound to a surface.
+ */
+const struct sct_context_list *
+sct_get_surface_contexts(struct surface_context_tracker *sct,
+                         const struct pipe_surface *surface)
+{
+   const struct sct_surface *si = find_surface_info(sct, surface);
+   return si->contexts;
+}
+
+
+
+static boolean
+find_texture(const struct sct_context *ci,
+             const struct pipe_texture *texture)
+{
+   const struct texture_list *tl;
+
+   for (tl = ci->textures_used; tl; tl = tl->next) {
+      if (tl->texture == texture) {
+         return TRUE;
+      }
+   }
+   return FALSE;
+}
+
+
+/**
+ * Add the given texture to the context's list of used textures.
+ */
+static void
+add_texture_used(struct sct_context *ci,
+                 struct pipe_texture *texture)
+{
+   if (!find_texture(ci, texture)) {
+      /* add to list */
+      struct texture_list *tl = CALLOC_STRUCT(texture_list);
+      if (tl) {
+         pipe_texture_reference(&tl->texture, texture);
+         /* insert at head */
+         tl->next = ci->textures_used;
+         ci->textures_used = tl;
+      }
+   }
+}
+
+
+/**
+ * Bind a texture to a rendering context.
+ */
+void
+sct_bind_texture(struct surface_context_tracker *sct,
+                 struct pipe_context *context,
+                 uint unit,
+                 struct pipe_texture *tex)
+{
+   struct sct_context *ci = find_context_info(sct, context);
+
+   if (ci->textures[unit] != tex) {
+      /* put texture on the 'used' list */
+      add_texture_used(ci, tex);
+      /* bind new */
+      pipe_texture_reference(&ci->textures[unit], tex);
+   }
+}
+
+
+/**
+ * Check if the given texture has been used by the rendering context
+ * since the last call to sct_flush_textures().
+ */
+boolean
+sct_is_texture_used(struct surface_context_tracker *sct,
+                    const struct pipe_context *context,
+                    const struct pipe_texture *texture)
+{
+   const struct sct_context *ci = find_context_info(sct, context);
+   return find_texture(ci, texture);
+}
+
+
+/**
+ * To be called when the image contents of a texture are changed, such
+ * as for gl[Copy]TexSubImage().
+ * XXX this may not be needed
+ */
+void
+sct_update_texture(struct pipe_texture *tex)
+{
+
+}
+
+
+/**
+ * When a scene is flushed/rendered we can release the list of
+ * used textures.
+ */
+void
+sct_flush_textures(struct surface_context_tracker *sct,
+                   struct pipe_context *context)
+{
+   struct sct_context *ci = find_context_info(sct, context);
+   struct texture_list *tl, *next;
+   uint i;
+
+   for (tl = ci->textures_used; tl; tl = next) {
+      next = tl->next;
+      pipe_texture_release(&tl->texture);
+      FREE(tl);
+   }
+   ci->textures_used = NULL;
+
+   /* put the currently bound textures on the 'used' list */
+   for (i = 0; i < PIPE_MAX_SAMPLERS; i++) {
+      add_texture_used(ci, ci->textures[i]);
+   }
+}
+
+
+
+void
+sct_destroy_context(struct surface_context_tracker *sct,
+                    struct pipe_context *context)
+{
+   /* XXX should we require an unbinding first? */
+   {
+      struct sct_surface *si;
+      for (si = sct->surfaces; si; si = si->next) {
+         remove_context_from_surface(si, context);
+      }
+   }
+
+   /* remove context from context_info list */
+   {
+      struct sct_context *ci, *next, *prev = NULL;
+      for (ci = sct->contexts; ci; ci = next) {
+         next = ci->next;
+         if (ci->context == context) {
+            if (prev)
+               prev->next = ci->next;
+            else
+               sct->contexts = ci->next;
+            FREE(ci);
+         }
+         else {
+            prev = ci;
+         }
+      }
+   }
+
+}
+
+
+void
+sct_destroy_surface(struct surface_context_tracker *sct,
+                    struct pipe_surface *surface)
+{
+   if (1) {
+      /* debug/sanity: no context should be bound to surface */
+      struct sct_context *ci;
+      uint i;
+      for (ci = sct->contexts; ci; ci = ci->next) {
+         for (i = 0; i < MAX_SURFACES; i++) {
+            assert(ci->surfaces[i] != surface);
+         }
+      }
+   }
+
+   /* remove surface from sct_surface list */
+   {
+      struct sct_surface *si, *next, *prev = NULL;
+      for (si = sct->surfaces; si; si = next) {
+         next = si->next;
+         if (si->surface == surface) {
+            /* unlink */
+            if (prev)
+               prev->next = si->next;
+            else
+               sct->surfaces = si->next;
+            FREE(si);
+         }
+         else {
+            prev = si;
+         }
+      }
+   }
+}
diff --git a/src/gallium/auxiliary/sct/sct.h b/src/gallium/auxiliary/sct/sct.h
new file mode 100644
index 0000000000..cf7c4d3bdf
--- /dev/null
+++ b/src/gallium/auxiliary/sct/sct.h
@@ -0,0 +1,123 @@
+/**************************************************************************
+ * 
+ * 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.
+ * 
+ **************************************************************************/
+
+/**
+ * Surface/Context Tracking
+ *
+ * For some drivers, we need to monitor the binding between contexts and
+ * surfaces/textures.
+ * This code may evolve quite a bit...
+ */
+
+
+#ifndef SCT_H
+#define SCT_H
+
+
+#ifdef	__cplusplus
+extern "C" {
+#endif
+
+
+struct pipe_context;
+struct pipe_surface;
+
+struct sct_context;
+struct sct_surface;
+
+
+/**
+ * Per-device info, basically
+ */
+struct surface_context_tracker
+{
+   struct sct_context *contexts;
+   struct sct_surface *surfaces;
+};
+
+
+
+/**
+ * Simple linked list of contexts
+ */
+struct sct_context_list
+{
+   const struct pipe_context *context;
+   struct sct_context_list *next;
+};
+
+
+
+extern void
+sct_bind_surfaces(struct surface_context_tracker *sct,
+                  struct pipe_context *context,
+                  uint num_surf,
+                  struct pipe_surface **surfaces);
+
+
+extern void
+sct_bind_texture(struct surface_context_tracker *sct,
+                 struct pipe_context *context,
+                 uint unit,
+                 struct pipe_texture *texture);
+
+
+extern void
+sct_update_texture(struct pipe_texture *tex);
+
+
+extern boolean
+sct_is_texture_used(struct surface_context_tracker *sct,
+                    const struct pipe_context *context,
+                    const struct pipe_texture *texture);
+
+extern void
+sct_flush_textures(struct surface_context_tracker *sct,
+                   struct pipe_context *context);
+
+
+extern const struct sct_context_list *
+sct_get_surface_contexts(struct surface_context_tracker *sct,
+                         const struct pipe_surface *surf);
+
+
+extern void
+sct_destroy_context(struct surface_context_tracker *sct,
+                    struct pipe_context *context);
+
+
+extern void
+sct_destroy_surface(struct surface_context_tracker *sct,
+                    struct pipe_surface *surface);
+
+
+
+#ifdef	__cplusplus
+}
+#endif
+
+#endif /* SCT_H */
diff --git a/src/gallium/auxiliary/sct/usage.c b/src/gallium/auxiliary/sct/usage.c
new file mode 100644
index 0000000000..6227f19962
--- /dev/null
+++ b/src/gallium/auxiliary/sct/usage.c
@@ -0,0 +1,61 @@
+/* surface / context tracking */
+
+
+/*
+
+context A:
+  render to texture T
+
+context B:
+  texture from T
+
+-----------------------
+
+flush surface:
+  which contexts are bound to the surface?
+
+-----------------------
+
+glTexSubImage():
+  which contexts need to be flushed?
+
+ */
+
+
+/*
+
+in MakeCurrent():
+
+  call sct_bind_surfaces(context, list of surfaces) to update the
+  dependencies between context and surfaces
+
+
+in SurfaceFlush(), or whatever it is in D3D:
+
+  call sct_get_surface_contexts(surface) to get a list of contexts
+  which are currently bound to the surface.
+
+
+
+in BindTexture():
+
+  call sct_bind_texture(context, texture) to indicate that the texture
+  is used in the scene.
+
+
+in glTexSubImage() or RenderToTexture():
+
+  call sct_is_texture_used(context, texture) to determine if the texture
+  has been used in the scene, but the scene's not flushed.  If TRUE is
+  returned it means the scene has to be rendered/flushed before the contents
+  of the texture can be changed.
+
+
+in psb_scene_flush/terminate():
+
+  call sct_flush_textures(context) to tell the SCT that the textures which
+  were used in the scene can be released.
+
+
+
+*/
diff --git a/src/gallium/auxiliary/tgsi/exec/tgsi_exec.c b/src/gallium/auxiliary/tgsi/exec/tgsi_exec.c
index ac52441400..f2ed9e0353 100644
--- a/src/gallium/auxiliary/tgsi/exec/tgsi_exec.c
+++ b/src/gallium/auxiliary/tgsi/exec/tgsi_exec.c
@@ -1220,7 +1220,8 @@ fetch_texel( struct tgsi_sampler *sampler,
 static void
 exec_tex(struct tgsi_exec_machine *mach,
          const struct tgsi_full_instruction *inst,
-         boolean biasLod)
+         boolean biasLod,
+         boolean projected)
 {
    const uint unit = inst->FullSrcRegisters[1].SrcRegister.Index;
    union tgsi_exec_channel r[8];
@@ -1234,17 +1235,9 @@ exec_tex(struct tgsi_exec_machine *mach,
 
       FETCH(&r[0], 0, CHAN_X);
 
-      switch (inst->FullSrcRegisters[0].SrcRegisterExtSwz.ExtDivide) {
-      case TGSI_EXTSWIZZLE_W:
+      if (projected) {
          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) {
@@ -1266,19 +1259,11 @@ exec_tex(struct tgsi_exec_machine *mach,
       FETCH(&r[1], 0, CHAN_Y);
       FETCH(&r[2], 0, CHAN_Z);
 
-      switch (inst->FullSrcRegisters[0].SrcRegisterExtSwz.ExtDivide) {
-      case TGSI_EXTSWIZZLE_W:
+      if (projected) {
          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) {
@@ -1300,19 +1285,11 @@ exec_tex(struct tgsi_exec_machine *mach,
       FETCH(&r[1], 0, CHAN_Y);
       FETCH(&r[2], 0, CHAN_Z);
 
-      switch (inst->FullSrcRegisters[0].SrcRegisterExtSwz.ExtDivide) {
-      case TGSI_EXTSWIZZLE_W:
+      if (projected) {
          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) {
@@ -2007,14 +1984,14 @@ exec_instruction(
       /* simple texture lookup */
       /* src[0] = texcoord */
       /* src[1] = sampler unit */
-      exec_tex(mach, inst, FALSE);
+      exec_tex(mach, inst, FALSE, 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);
+      exec_tex(mach, inst, TRUE, FALSE);
       break;
 
    case TGSI_OPCODE_TXD:
@@ -2030,7 +2007,14 @@ exec_instruction(
       /* Texture lookup with explit LOD */
       /* src[0] = texcoord (src[0].w = LOD) */
       /* src[1] = sampler unit */
-      exec_tex(mach, inst, TRUE);
+      exec_tex(mach, inst, TRUE, FALSE);
+      break;
+
+   case TGSI_OPCODE_TXP:
+      /* Texture lookup with projection */
+      /* src[0] = texcoord (src[0].w = projection) */
+      /* src[1] = sampler unit */
+      exec_tex(mach, inst, FALSE, TRUE);
       break;
 
    case TGSI_OPCODE_UP2H:
diff --git a/src/gallium/auxiliary/tgsi/util/tgsi_build.c b/src/gallium/auxiliary/tgsi/util/tgsi_build.c
index a00ff1c2a5..9c883ab704 100644
--- a/src/gallium/auxiliary/tgsi/util/tgsi_build.c
+++ b/src/gallium/auxiliary/tgsi/util/tgsi_build.c
@@ -719,7 +719,6 @@ tgsi_build_full_instruction(
             reg->SrcRegisterExtSwz.NegateY,
             reg->SrcRegisterExtSwz.NegateZ,
             reg->SrcRegisterExtSwz.NegateW,
-            reg->SrcRegisterExtSwz.ExtDivide,
             prev_token,
             instruction,
             header );
@@ -1057,7 +1056,6 @@ tgsi_default_src_register_ext_swz( void )
    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;
 
@@ -1084,7 +1082,6 @@ tgsi_build_src_register_ext_swz(
    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 )
@@ -1099,7 +1096,6 @@ tgsi_build_src_register_ext_swz(
    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;
@@ -1110,7 +1106,6 @@ tgsi_build_src_register_ext_swz(
    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 );
diff --git a/src/gallium/auxiliary/tgsi/util/tgsi_build.h b/src/gallium/auxiliary/tgsi/util/tgsi_build.h
index 607860e7fc..80bffc4ae7 100644
--- a/src/gallium/auxiliary/tgsi/util/tgsi_build.h
+++ b/src/gallium/auxiliary/tgsi/util/tgsi_build.h
@@ -229,7 +229,6 @@ tgsi_build_src_register_ext_swz(
    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 );
diff --git a/src/gallium/auxiliary/tgsi/util/tgsi_dump.c b/src/gallium/auxiliary/tgsi/util/tgsi_dump.c
index 59be14a748..ceb407b884 100644
--- a/src/gallium/auxiliary/tgsi/util/tgsi_dump.c
+++ b/src/gallium/auxiliary/tgsi/util/tgsi_dump.c
@@ -459,7 +459,8 @@ static const char *TGSI_OPCODES[] =
    "OPCODE_IFC",
    "OPCODE_BREAKC",
    "OPCODE_KIL",
-   "OPCODE_END"
+   "OPCODE_END",
+   "OPCODE_TXP"
 };
 
 static const char *TGSI_OPCODES_SHORT[] =
@@ -597,7 +598,8 @@ static const char *TGSI_OPCODES_SHORT[] =
    "IFC",
    "BREAKC",
    "KIL",
-   "END"
+   "END",
+   "TXP"
 };
 
 static const char *TGSI_SATS[] =
@@ -1361,10 +1363,6 @@ dump_instruction_verbose(
             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 );
diff --git a/src/gallium/auxiliary/util/Makefile b/src/gallium/auxiliary/util/Makefile
index 906a46d6b4..2abbe9500e 100644
--- a/src/gallium/auxiliary/util/Makefile
+++ b/src/gallium/auxiliary/util/Makefile
@@ -7,7 +7,9 @@ C_SOURCES = \
 	p_debug.c \
 	p_tile.c \
 	p_util.c \
-	u_mm.c
+	u_handle_table.c \
+	u_mm.c \
+	u_snprintf.c
 
 include ../../Makefile.template
 
diff --git a/src/gallium/auxiliary/util/SConscript b/src/gallium/auxiliary/util/SConscript
index 4717941434..2030214aa7 100644
--- a/src/gallium/auxiliary/util/SConscript
+++ b/src/gallium/auxiliary/util/SConscript
@@ -6,7 +6,9 @@ util = env.ConvenienceLibrary(
 		'p_debug.c',
 		'p_tile.c',
 		'p_util.c',
+		'u_handle_table.c',
 		'u_mm.c',
+		'u_snprintf.c',
 	])
 
 auxiliaries.insert(0, util)
diff --git a/src/gallium/auxiliary/util/p_debug.c b/src/gallium/auxiliary/util/p_debug.c
index b9607a6ba7..09cabdae25 100644
--- a/src/gallium/auxiliary/util/p_debug.c
+++ b/src/gallium/auxiliary/util/p_debug.c
@@ -36,14 +36,37 @@
 #include <stdlib.h>
 #endif
 
-#include "pipe/p_debug.h" 
 #include "pipe/p_compiler.h" 
+#include "pipe/p_util.h" 
+#include "pipe/p_debug.h" 
+
+
+#ifdef WIN32
+static INLINE void 
+rpl_EngDebugPrint(const char *format, ...)
+{
+   va_list ap;
+   va_start(ap, format);
+   EngDebugPrint("", (PCHAR)format, ap);
+   va_end(ap);
+}
+
+int rpl_vsnprintf(char *, size_t, const char *, va_list);
+#endif
 
 
 void debug_vprintf(const char *format, va_list ap)
 {
 #ifdef WIN32
-   EngDebugPrint("Gallium3D: ", (PCHAR)format, ap);
+#ifndef WINCE
+   /* EngDebugPrint does not handle float point arguments, so we need to use
+    * our own vsnprintf implementation */
+   char buf[512 + 1];
+   rpl_vsnprintf(buf, sizeof(buf), format, ap);
+   rpl_EngDebugPrint("%s", buf);
+#else
+   /* TODO: Implement debug print for WINCE */
+#endif
 #else
    vfprintf(stderr, format, ap);
 #endif
@@ -59,18 +82,92 @@ void debug_printf(const char *format, ...)
 }
 
 
-static INLINE void debug_abort(void) 
+/* TODO: implement a debug_abort that calls EngBugCheckEx on WIN32 */
+
+
+static INLINE void debug_break(void) 
 {
-#ifdef WIN32
+#if (defined(__i386__) || defined(__386__)) && defined(__GNUC__)
+   __asm("int3");
+#elif (defined(__i386__) || defined(__386__)) && defined(__MSC__)
+   _asm {int 3};
+#elif defined(WIN32) && !defined(WINCE)
    EngDebugBreak();
 #else
    abort();
 #endif
 }
 
+#if defined(WIN32)
+ULONG_PTR debug_config_file = 0;
+void *mapped_config_file = 0;
+
+enum {
+	eAssertAbortEn = 0x1,
+};
+
+/* Check for aborts enabled. */
+static unsigned abort_en()
+{
+	if (!mapped_config_file)
+	{
+		/* Open an 8 byte file for configuration data. */
+		mapped_config_file = EngMapFile(L"\\??\\c:\\gaDebug.cfg", 8, &debug_config_file);
+	}
+	/* An value of "0" (ascii) in the configuration file will clear the first 8 bits in the test byte. */
+	/* An value of "1" (ascii) in the configuration file will set the first bit in the test byte. */
+	/* An value of "2" (ascii) in the configuration file will set the second bit in the test byte. */
+	return ((((char *)mapped_config_file)[0]) - 0x30) & eAssertAbortEn;
+}
+#else /* WIN32 */
+static unsigned abort_en()
+{
+	return !GETENV("GALLIUM_ABORT_ON_ASSERT");
+}
+#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();
+   if (abort_en())
+   {
+      debug_break();
+   } else
+   {
+      debug_printf("continuing...\n");
+   }
+}
+
+
+#define DEBUG_MASK_TABLE_SIZE 256
+
+
+/**
+ * Mask hash table.
+ * 
+ * For now we just take the lower bits of the key, and do no attempt to solve
+ * collisions. Use a proper hash table when we have dozens of drivers. 
+ */
+static uint32_t debug_mask_table[DEBUG_MASK_TABLE_SIZE];
+
+
+void debug_mask_set(uint32_t uuid, uint32_t mask) 
+{
+   unsigned hash = uuid & (DEBUG_MASK_TABLE_SIZE - 1);
+   debug_mask_table[hash] = mask;
+}
+
+
+uint32_t debug_mask_get(uint32_t uuid)
+{
+   unsigned hash = uuid & (DEBUG_MASK_TABLE_SIZE - 1);
+   return debug_mask_table[hash];
+}
+
+
+void debug_mask_vprintf(uint32_t uuid, uint32_t what, const char *format, va_list ap)
+{
+   uint32_t mask = debug_mask_get(uuid);
+   if(mask & what)
+      debug_vprintf(format, ap);
 }
diff --git a/src/gallium/auxiliary/util/p_tile.h b/src/gallium/auxiliary/util/p_tile.h
index 318b6d11a6..fdc80a13b3 100644
--- a/src/gallium/auxiliary/util/p_tile.h
+++ b/src/gallium/auxiliary/util/p_tile.h
@@ -52,44 +52,50 @@ pipe_clip_tile(uint x, uint y, uint *w, uint *h, const struct pipe_surface *ps)
    return FALSE;
 }
 
+#ifdef __cplusplus
+extern "C" {
+#endif
 
-extern void
+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
+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
+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
+void
 pipe_put_tile_rgba(struct pipe_context *pipe,
                    struct pipe_surface *ps,
                    uint x, uint y, uint w, uint h,
                    const float *p);
 
 
-extern void
+void
 pipe_get_tile_z(struct pipe_context *pipe,
                 struct pipe_surface *ps,
                 uint x, uint y, uint w, uint h,
                 uint *z);
 
-extern void
+void
 pipe_put_tile_z(struct pipe_context *pipe,
                 struct pipe_surface *ps,
                 uint x, uint y, uint w, uint h,
                 const uint *z);
 
+#ifdef __cplusplus
+}
+#endif
 
 #endif
diff --git a/src/gallium/auxiliary/util/u_cpu_detect.c b/src/gallium/auxiliary/util/u_cpu_detect.c
new file mode 100644
index 0000000000..d9f2f8fc28
--- /dev/null
+++ b/src/gallium/auxiliary/util/u_cpu_detect.c
@@ -0,0 +1,506 @@
+/**************************************************************************
+ * 
+ * Copyright 2008 Dennis Smit
+ * 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.
+ * 
+ **************************************************************************/
+
+/*
+ * Based on the work of Eric Anholt <anholt@FreeBSD.org>
+ */
+
+/* FIXME: clean this entire file up */
+
+#include "u_cpu_detect.h"
+
+#ifdef __linux__
+#define OS_LINUX
+#endif
+#ifdef WIN32
+#define OS_WIN32
+#endif
+
+#if defined(ARCH_POWERPC)
+#if defined(OS_DARWIN)
+#include <sys/sysctl.h>
+#else
+#include <signal.h>
+#include <setjmp.h>
+#endif
+#endif
+
+#if defined(OS_NETBSD) || defined(OS_OPENBSD)
+#include <sys/param.h>
+#include <sys/sysctl.h>
+#include <machine/cpu.h>
+#endif
+
+#if defined(OS_FREEBSD)
+#include <sys/types.h>
+#include <sys/sysctl.h>
+#endif
+
+#if defined(OS_LINUX)
+#include <signal.h>
+#endif
+
+#if defined(OS_WIN32)
+#include <windows.h>
+#endif
+
+#include <stdio.h>
+#include <stdlib.h>
+#include <unistd.h>
+#include <string.h>
+
+
+static struct cpu_detect_caps __cpu_detect_caps;
+static int __cpu_detect_initialized = 0;
+
+static int has_cpuid(void);
+static int cpuid(unsigned int ax, unsigned int *p);
+
+/* The sigill handlers */
+#if defined(ARCH_X86) /* x86 (linux katmai handler check thing) */
+#if defined(OS_LINUX) && defined(_POSIX_SOURCE) && defined(X86_FXSR_MAGIC)
+static void sigill_handler_sse(int signal, struct sigcontext sc)
+{
+	/* Both the "xorps %%xmm0,%%xmm0" and "divps %xmm0,%%xmm1"
+	 * instructions are 3 bytes long.  We must increment the instruction
+	 * pointer manually to avoid repeated execution of the offending
+	 * instruction.
+	 *
+	 * If the SIGILL is caused by a divide-by-zero when unmasked
+	 * exceptions aren't supported, the SIMD FPU status and control
+	 * word will be restored at the end of the test, so we don't need
+	 * to worry about doing it here.  Besides, we may not be able to...
+	 */
+	sc.eip += 3;
+
+	__cpu_detect_caps.hasSSE=0;
+}
+
+static void sigfpe_handler_sse(int signal, struct sigcontext sc)
+{
+	if (sc.fpstate->magic != 0xffff) {
+		/* Our signal context has the extended FPU state, so reset the
+		 * divide-by-zero exception mask and clear the divide-by-zero
+		 * exception bit.
+		 */
+		sc.fpstate->mxcsr |= 0x00000200;
+		sc.fpstate->mxcsr &= 0xfffffffb;
+	} else {
+		/* If we ever get here, we're completely hosed.
+		*/
+	}
+}
+#endif
+#endif /* OS_LINUX && _POSIX_SOURCE && X86_FXSR_MAGIC */
+
+#if defined(OS_WIN32)
+LONG CALLBACK win32_sig_handler_sse(EXCEPTION_POINTERS* ep)
+{
+	if(ep->ExceptionRecord->ExceptionCode==EXCEPTION_ILLEGAL_INSTRUCTION){
+		ep->ContextRecord->Eip +=3;
+		__cpu_detect_caps.hasSSE=0;
+		return EXCEPTION_CONTINUE_EXECUTION;
+	}
+	return EXCEPTION_CONTINUE_SEARCH;
+}
+#endif /* OS_WIN32 */
+
+
+#if defined(ARCH_POWERPC) && !defined(OS_DARWIN)
+static sigjmp_buf __lv_powerpc_jmpbuf;
+static volatile sig_atomic_t __lv_powerpc_canjump = 0;
+
+static void sigill_handler (int sig);
+
+static void sigill_handler (int sig)
+{
+	if (!__lv_powerpc_canjump) {
+		signal (sig, SIG_DFL);
+		raise (sig);
+	}
+
+	__lv_powerpc_canjump = 0;
+	siglongjmp(__lv_powerpc_jmpbuf, 1);
+}
+
+static void check_os_altivec_support(void)
+{
+#if defined(OS_DARWIN)
+	int sels[2] = {CTL_HW, HW_VECTORUNIT};
+	int has_vu = 0;
+	int len = sizeof (has_vu);
+	int err;
+
+	err = sysctl(sels, 2, &has_vu, &len, NULL, 0);
+
+	if (err == 0) {
+		if (has_vu != 0) {
+			__cpu_detect_caps.hasAltiVec = 1;
+		}
+	}
+#else /* !OS_DARWIN */
+	/* no Darwin, do it the brute-force way */
+	/* this is borrowed from the libmpeg2 library */
+	signal(SIGILL, sigill_handler);
+	if (sigsetjmp(__lv_powerpc_jmpbuf, 1)) {
+		signal(SIGILL, SIG_DFL);
+	} else {
+		__lv_powerpc_canjump = 1;
+
+		__asm __volatile
+			("mtspr 256, %0\n\t"
+			 "vand %%v0, %%v0, %%v0"
+			 :
+			 : "r" (-1));
+
+		signal(SIGILL, SIG_DFL);
+		__cpu_detect_caps.hasAltiVec = 1;
+	}
+#endif
+}
+#endif
+
+/* If we're running on a processor that can do SSE, let's see if we
+ * are allowed to or not.  This will catch 2.4.0 or later kernels that
+ * haven't been configured for a Pentium III but are running on one,
+ * and RedHat patched 2.2 kernels that have broken exception handling
+ * support for user space apps that do SSE.
+ */
+static void check_os_katmai_support(void)
+{
+#if defined(ARCH_X86)
+#if defined(OS_FREEBSD)
+	int has_sse=0, ret;
+	int len = sizeof (has_sse);
+
+	ret = sysctlbyname("hw.instruction_sse", &has_sse, &len, NULL, 0);
+	if (ret || !has_sse)
+		__cpu_detect_caps.hasSSE=0;
+
+#elif defined(OS_NETBSD) || defined(OS_OPENBSD)
+	int has_sse, has_sse2, ret, mib[2];
+	int varlen;
+
+	mib[0] = CTL_MACHDEP;
+	mib[1] = CPU_SSE;
+	varlen = sizeof (has_sse);
+
+	ret = sysctl(mib, 2, &has_sse, &varlen, NULL, 0);
+	if (ret < 0 || !has_sse) {
+		__cpu_detect_caps.hasSSE = 0;
+	} else {
+		__cpu_detect_caps.hasSSE = 1;
+	}
+
+	mib[1] = CPU_SSE2;
+	varlen = sizeof (has_sse2);
+	ret = sysctl(mib, 2, &has_sse2, &varlen, NULL, 0);
+	if (ret < 0 || !has_sse2) {
+		__cpu_detect_caps.hasSSE2 = 0;
+	} else {
+		__cpu_detect_caps.hasSSE2 = 1;
+	}
+	__cpu_detect_caps.hasSSE = 0; /* FIXME ?!?!? */
+
+#elif defined(OS_WIN32)
+	LPTOP_LEVEL_EXCEPTION_FILTER exc_fil;
+	if (__cpu_detect_caps.hasSSE) {
+		exc_fil = SetUnhandledExceptionFilter(win32_sig_handler_sse);
+		__asm __volatile ("xorps %xmm0, %xmm0");
+		SetUnhandledExceptionFilter(exc_fil);
+	}
+#elif defined(OS_LINUX)
+	struct sigaction saved_sigill;
+	struct sigaction saved_sigfpe;
+
+	/* Save the original signal handlers.
+	*/
+	sigaction(SIGILL, NULL, &saved_sigill);
+	sigaction(SIGFPE, NULL, &saved_sigfpe);
+
+	signal(SIGILL, (void (*)(int))sigill_handler_sse);
+	signal(SIGFPE, (void (*)(int))sigfpe_handler_sse);
+
+	/* Emulate test for OSFXSR in CR4.  The OS will set this bit if it
+	 * supports the extended FPU save and restore required for SSE.  If
+	 * we execute an SSE instruction on a PIII and get a SIGILL, the OS
+	 * doesn't support Streaming SIMD Exceptions, even if the processor
+	 * does.
+	 */
+	if (__cpu_detect_caps.hasSSE) {
+		__asm __volatile ("xorps %xmm1, %xmm0");
+	}
+
+	/* Emulate test for OSXMMEXCPT in CR4.  The OS will set this bit if
+	 * it supports unmasked SIMD FPU exceptions.  If we unmask the
+	 * exceptions, do a SIMD divide-by-zero and get a SIGILL, the OS
+	 * doesn't support unmasked SIMD FPU exceptions.  If we get a SIGFPE
+	 * as expected, we're okay but we need to clean up after it.
+	 *
+	 * Are we being too stringent in our requirement that the OS support
+	 * unmasked exceptions?  Certain RedHat 2.2 kernels enable SSE by
+	 * setting CR4.OSFXSR but don't support unmasked exceptions.  Win98
+	 * doesn't even support them.  We at least know the user-space SSE
+	 * support is good in kernels that do support unmasked exceptions,
+	 * and therefore to be safe I'm going to leave this test in here.
+	 */
+	if (__cpu_detect_caps.hasSSE) {
+		//      test_os_katmai_exception_support();
+	}
+
+	/* Restore the original signal handlers.
+	*/
+	sigaction(SIGILL, &saved_sigill, NULL);
+	sigaction(SIGFPE, &saved_sigfpe, NULL);
+
+#else
+	/* We can't use POSIX signal handling to test the availability of
+	 * SSE, so we disable it by default.
+	 */
+	__cpu_detect_caps.hasSSE = 0;
+#endif /* __linux__ */
+#endif
+}
+
+
+static int has_cpuid(void)
+{
+#if defined(ARCH_X86)
+	int a, c;
+
+	__asm __volatile
+		("pushf\n"
+		 "popl %0\n"
+		 "movl %0, %1\n"
+		 "xorl $0x200000, %0\n"
+		 "push %0\n"
+		 "popf\n"
+		 "pushf\n"
+		 "popl %0\n"
+		 : "=a" (a), "=c" (c)
+		 :
+		 : "cc");
+
+	return a != c;
+#else
+	return 0;
+#endif
+}
+
+static int cpuid(unsigned int ax, unsigned int *p)
+{
+#if defined(ARCH_X86)
+	unsigned int flags;
+
+	__asm __volatile
+		("movl %%ebx, %%esi\n\t"
+		 "cpuid\n\t"
+		 "xchgl %%ebx, %%esi"
+		 : "=a" (p[0]), "=S" (p[1]),
+		 "=c" (p[2]), "=d" (p[3])
+		 : "0" (ax));
+
+	return 0;
+#else
+	return -1;
+#endif
+}
+
+void cpu_detect_initialize()
+{
+	unsigned int regs[4];
+	unsigned int regs2[4];
+
+	int mib[2], ncpu;
+	int len;
+
+	memset(&__cpu_detect_caps, 0, sizeof (struct cpu_detect_caps));
+
+	/* Check for arch type */
+#if defined(ARCH_MIPS)
+	__cpu_detect_caps.type = CPU_DETECT_TYPE_MIPS;
+#elif defined(ARCH_ALPHA)
+	__cpu_detect_caps.type = CPU_DETECT_TYPE_ALPHA;
+#elif defined(ARCH_SPARC)
+	__cpu_detect_caps.type = CPU_DETECT_TYPE_SPARC;
+#elif defined(ARCH_X86)
+	__cpu_detect_caps.type = CPU_DETECT_TYPE_X86;
+#elif defined(ARCH_POWERPC)
+	__cpu_detect_caps.type = CPU_DETECT_TYPE_POWERPC;
+#else
+	__cpu_detect_caps.type = CPU_DETECT_TYPE_OTHER;
+#endif
+
+	/* Count the number of CPUs in system */
+#if !defined(OS_WIN32) && !defined(OS_UNKNOWN) && defined(_SC_NPROCESSORS_ONLN)
+	__cpu_detect_caps.nrcpu = sysconf(_SC_NPROCESSORS_ONLN);
+	if (__cpu_detect_caps.nrcpu == -1)
+		__cpu_detect_caps.nrcpu = 1;
+
+#elif defined(OS_NETBSD) || defined(OS_FREEBSD) || defined(OS_OPENBSD)
+
+	mib[0] = CTL_HW;
+	mib[1] = HW_NCPU;
+
+	len = sizeof (ncpu);
+	sysctl(mib, 2, &ncpu, &len, NULL, 0);
+	__cpu_detect_caps.nrcpu = ncpu;
+
+#else
+	__cpu_detect_caps.nrcpu = 1;
+#endif
+
+#if defined(ARCH_X86)
+	/* No cpuid, old 486 or lower */
+	if (has_cpuid() == 0)
+		return;
+
+	__cpu_detect_caps.cacheline = 32;
+
+	/* Get max cpuid level */
+	cpuid(0x00000000, regs);
+
+	if (regs[0] >= 0x00000001) {
+		unsigned int cacheline;
+
+		cpuid (0x00000001, regs2);
+
+		__cpu_detect_caps.x86cpuType = (regs2[0] >> 8) & 0xf;
+		if (__cpu_detect_caps.x86cpuType == 0xf)
+		    __cpu_detect_caps.x86cpuType = 8 + ((regs2[0] >> 20) & 255); /* use extended family (P4, IA64) */
+
+		/* general feature flags */
+		__cpu_detect_caps.hasTSC  = (regs2[3] & (1 << 8  )) >>  8; /* 0x0000010 */
+		__cpu_detect_caps.hasMMX  = (regs2[3] & (1 << 23 )) >> 23; /* 0x0800000 */
+		__cpu_detect_caps.hasSSE  = (regs2[3] & (1 << 25 )) >> 25; /* 0x2000000 */
+		__cpu_detect_caps.hasSSE2 = (regs2[3] & (1 << 26 )) >> 26; /* 0x4000000 */
+		__cpu_detect_caps.hasSSE3 = (regs2[2] & (1));	       /* 0x0000001 */
+		__cpu_detect_caps.hasSSSE3 = (regs2[2] & (1 << 9 )) >> 9;   /* 0x0000020 */
+		__cpu_detect_caps.hasMMX2 = __cpu_detect_caps.hasSSE; /* SSE cpus supports mmxext too */
+
+		cacheline = ((regs2[1] >> 8) & 0xFF) * 8;
+		if (cacheline > 0)
+			__cpu_detect_caps.cacheline = cacheline;
+	}
+
+	cpuid(0x80000000, regs);
+
+	if (regs[0] >= 0x80000001) {
+
+		cpuid(0x80000001, regs2);
+
+		__cpu_detect_caps.hasMMX  |= (regs2[3] & (1 << 23 )) >> 23; /* 0x0800000 */
+		__cpu_detect_caps.hasMMX2 |= (regs2[3] & (1 << 22 )) >> 22; /* 0x400000 */
+		__cpu_detect_caps.has3DNow    = (regs2[3] & (1 << 31 )) >> 31; /* 0x80000000 */
+		__cpu_detect_caps.has3DNowExt = (regs2[3] & (1 << 30 )) >> 30;
+	}
+
+	if (regs[0] >= 0x80000006) {
+		cpuid(0x80000006, regs2);
+		__cpu_detect_caps.cacheline = regs2[2] & 0xFF;
+	}
+
+
+#if defined(OS_LINUX) || defined(OS_FREEBSD) || defined(OS_NETBSD) || defined(OS_CYGWIN) || defined(OS_OPENBSD)
+	if (__cpu_detect_caps.hasSSE)
+		check_os_katmai_support();
+
+	if (!__cpu_detect_caps.hasSSE) {
+		__cpu_detect_caps.hasSSE2 = 0;
+		__cpu_detect_caps.hasSSE3 = 0;
+		__cpu_detect_caps.hasSSSE3 = 0;
+	}
+#else
+	__cpu_detect_caps.hasSSE = 0;
+	__cpu_detect_caps.hasSSE2 = 0;
+	__cpu_detect_caps.hasSSE3 = 0;
+	__cpu_detect_caps.hasSSSE3 = 0;
+#endif
+#endif /* ARCH_X86 */
+
+#if defined(ARCH_POWERPC)
+	check_os_altivec_support();
+#endif /* ARCH_POWERPC */
+
+	__cpu_detect_initialized = 1;
+}
+
+struct cpu_detect_caps *cpu_detect_get_caps()
+{
+	return &__cpu_detect_caps;
+}
+
+/* The getters and setters for feature flags */
+int cpu_detect_get_tsc()
+{
+	return __cpu_detect_caps.hasTSC;
+}
+
+int cpu_detect_get_mmx()
+{
+	return __cpu_detect_caps.hasMMX;
+}
+
+int cpu_detect_get_mmx2()
+{
+	return __cpu_detect_caps.hasMMX2;
+}
+
+int cpu_detect_get_sse()
+{
+	return __cpu_detect_caps.hasSSE;
+}
+
+int cpu_detect_get_sse2()
+{
+	return __cpu_detect_caps.hasSSE2;
+}
+
+int cpu_detect_get_sse3()
+{
+	return __cpu_detect_caps.hasSSE3;
+}
+
+int cpu_detect_get_ssse3()
+{
+	return __cpu_detect_caps.hasSSSE3;
+}
+
+int cpu_detect_get_3dnow()
+{
+	return __cpu_detect_caps.has3DNow;
+}
+
+int cpu_detect_get_3dnow2()
+{
+	return __cpu_detect_caps.has3DNowExt;
+}
+
+int cpu_detect_get_altivec()
+{
+	return __cpu_detect_caps.hasAltiVec;
+}
+
diff --git a/src/gallium/auxiliary/util/u_cpu_detect.h b/src/gallium/auxiliary/util/u_cpu_detect.h
new file mode 100644
index 0000000000..1612d49286
--- /dev/null
+++ b/src/gallium/auxiliary/util/u_cpu_detect.h
@@ -0,0 +1,78 @@
+/**************************************************************************
+ * 
+ * Copyright 2008 Dennis Smit
+ * 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.
+ *
+ ***************************************************************************/
+
+/*
+ * Based on the work of Eric Anholt <anholt@FreeBSD.org>
+ */
+
+#ifndef _CPU_DETECT_H
+#define _CPU_DETECT_H
+
+typedef enum {
+	CPU_DETECT_TYPE_MIPS,
+	CPU_DETECT_TYPE_ALPHA,
+	CPU_DETECT_TYPE_SPARC,
+	CPU_DETECT_TYPE_X86,
+	CPU_DETECT_TYPE_POWERPC,
+	CPU_DETECT_TYPE_OTHER
+} cpu_detect_type;
+
+struct cpu_detect_caps {
+	cpu_detect_type	type;
+	int		nrcpu;
+
+	/* Feature flags */
+	int		x86cpuType;
+	int		cacheline;
+
+	int		hasTSC;
+	int		hasMMX;
+	int		hasMMX2;
+	int		hasSSE;
+	int		hasSSE2;
+	int		hasSSE3;
+	int		hasSSSE3;
+	int		has3DNow;
+	int		has3DNowExt;
+	int		hasAltiVec;
+};
+
+/* prototypes */
+void cpu_detect_initialize(void);
+struct cpu_detect_caps *cpu_detect_get_caps(void);
+
+int cpu_detect_get_tsc(void);
+int cpu_detect_get_mmx(void);
+int cpu_detect_get_mmx2(void);
+int cpu_detect_get_sse(void);
+int cpu_detect_get_sse2(void);
+int cpu_detect_get_sse3(void);
+int cpu_detect_get_ssse3(void);
+int cpu_detect_get_3dnow(void);
+int cpu_detect_get_3dnow2(void);
+int cpu_detect_get_altivec(void);
+
+#endif /* _CPU_DETECT_H */
diff --git a/src/gallium/auxiliary/util/u_handle_table.c b/src/gallium/auxiliary/util/u_handle_table.c
new file mode 100644
index 0000000000..8a298f7c41
--- /dev/null
+++ b/src/gallium/auxiliary/util/u_handle_table.c
@@ -0,0 +1,207 @@
+/**************************************************************************
+ *
+ * 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.
+ *
+ **************************************************************************/
+
+/**
+ * Generic handle table implementation.
+ *  
+ * @author José Fonseca <jrfonseca@tungstengraphics.com>
+ */
+
+
+#include "pipe/p_compiler.h"
+#include "pipe/p_debug.h"
+#include "pipe/p_util.h"
+
+#include "u_handle_table.h"
+
+
+#define HANDLE_TABLE_INITIAL_SIZE 16  
+
+
+struct handle_table
+{
+   /** Object array. Empty handles have a null object */
+   void **objects;
+   
+   /** Number of objects the handle can currently hold */
+   unsigned size;
+   /** Number of consecutive objects allocated at the start of the table */
+   unsigned filled;
+   
+   /** Optional object destructor */
+   void (*destroy)(void *object);
+};
+
+
+struct handle_table *
+handle_table_create(void)
+{
+   struct handle_table *ht;
+   
+   ht = MALLOC_STRUCT(handle_table);
+   if(!ht)
+      return NULL;
+   
+   ht->objects = (void **)CALLOC(HANDLE_TABLE_INITIAL_SIZE, sizeof(void *));
+   if(!ht->objects) {
+      FREE(ht);
+      return NULL;
+   }
+   
+   ht->size = HANDLE_TABLE_INITIAL_SIZE;
+   ht->filled = 0;
+   
+   ht->destroy = NULL;
+   
+   return ht;
+}
+
+
+void
+handle_table_set_destroy(struct handle_table *ht,
+                         void (*destroy)(void *object))
+{
+   assert(ht);
+   ht->destroy = destroy;
+}
+
+
+unsigned
+handle_table_add(struct handle_table *ht, 
+                 void *object)
+{
+   unsigned index;
+   unsigned handle;
+   
+   assert(ht);
+   assert(object);
+   if(!object)
+      return 0;
+
+   /* linear search for an empty handle */
+   while(ht->filled < ht->size) {
+      if(!ht->objects[ht->filled])
+	 break;
+      ++ht->filled;
+   }
+  
+   /* grow the table */
+   if(ht->filled == ht->size) {
+      unsigned new_size;
+      void **new_objects;
+      
+      new_size = ht->size*2;
+      assert(new_size);
+      
+      new_objects = (void **)REALLOC((void *)ht->objects,
+                                     ht->size*sizeof(void *),
+                                     new_size*sizeof(void *));
+      if(!new_objects)
+	 return 0;
+      
+      memset(new_objects + ht->size, 0, (new_size - ht->size)*sizeof(void *));
+      
+      ht->size = new_size;
+      ht->objects = new_objects;
+   }
+
+   index = ht->filled;
+   
+   handle = index + 1;
+   
+   /* check integer overflow */
+   if(!handle)
+      return 0;
+   
+   assert(!ht->objects[index]);
+   ht->objects[index] = object;
+   ++ht->filled;
+   
+   return handle;
+}
+
+
+void *
+handle_table_get(struct handle_table *ht, 
+                 unsigned handle)
+{
+   void *object;
+   
+   assert(ht);
+   assert(handle > 0);
+   assert(handle <= ht->size);
+   if(!handle || handle > ht->size)
+      return NULL;
+
+   object = ht->objects[handle - 1];
+   assert(object);
+   
+   return object;
+}
+
+
+void
+handle_table_remove(struct handle_table *ht, 
+                    unsigned handle)
+{
+   void *object;
+   unsigned index;
+   
+   assert(ht);
+   assert(handle > 0);
+   assert(handle <= ht->size);
+   if(!handle || handle > ht->size)
+      return;
+
+   index = handle - 1;
+   object = ht->objects[index];
+   assert(object);
+   
+   if(object && ht->destroy)
+      ht->destroy(object);
+
+   ht->objects[index] = NULL;
+   if(index < ht->filled)
+      ht->filled = index;
+}
+
+
+void
+handle_table_destroy(struct handle_table *ht)
+{
+   unsigned index;
+   assert(ht);
+
+   if(ht->destroy)
+      for(index = 0; index < ht->size; ++index)
+	 if(ht->objects[index])
+	    ht->destroy(ht->objects[index]);
+   
+   FREE(ht->objects);
+   FREE(ht);
+}
+
diff --git a/src/gallium/auxiliary/util/u_handle_table.h b/src/gallium/auxiliary/util/u_handle_table.h
new file mode 100644
index 0000000000..51fc273865
--- /dev/null
+++ b/src/gallium/auxiliary/util/u_handle_table.h
@@ -0,0 +1,96 @@
+/**************************************************************************
+ *
+ * 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.
+ *
+ **************************************************************************/
+
+/**
+ * Generic handle table.
+ *  
+ * @author José Fonseca <jrfonseca@tungstengraphics.com>
+ */
+
+#ifndef U_HANDLE_TABLE_H_
+#define U_HANDLE_TABLE_H_
+
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+   
+/**
+ * Abstract data type to map integer handles to objects.
+ */
+struct handle_table;
+
+
+struct handle_table *
+handle_table_create(void);
+
+
+/**
+ * Set an optional destructor callback.
+ * 
+ * If set, it will be called during handle_table_remove and 
+ * handle_table_destroy calls.
+ */
+void
+handle_table_set_destroy(struct handle_table *ht,
+                         void (*destroy)(void *object));
+
+
+/**
+ * Add a new object.
+ * 
+ * Returns a zero handle on failure (out of memory).
+ */
+unsigned
+handle_table_add(struct handle_table *ht, 
+                 void *object);
+
+/**
+ * Fetch an existing object.
+ * 
+ * Returns NULL for an invalid handle.
+ */
+void *
+handle_table_get(struct handle_table *ht, 
+                 unsigned handle);
+
+
+void
+handle_table_remove(struct handle_table *ht, 
+                    unsigned handle);
+
+
+void
+handle_table_destroy(struct handle_table *ht);
+
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif /* U_HANDLE_TABLE_H_ */
diff --git a/src/gallium/auxiliary/util/u_snprintf.c b/src/gallium/auxiliary/util/u_snprintf.c
new file mode 100644
index 0000000000..61c20b48f7
--- /dev/null
+++ b/src/gallium/auxiliary/util/u_snprintf.c
@@ -0,0 +1,1478 @@
+/*
+ * Copyright (c) 1995 Patrick Powell.
+ *
+ * This code is based on code written by Patrick Powell <papowell@astart.com>.
+ * It may be used for any purpose as long as this notice remains intact on all
+ * source code distributions.
+ */
+
+/*
+ * Copyright (c) 2008 Holger Weiss.
+ *
+ * This version of the code is maintained by Holger Weiss <holger@jhweiss.de>.
+ * My changes to the code may freely be used, modified and/or redistributed for
+ * any purpose.  It would be nice if additions and fixes to this file (including
+ * trivial code cleanups) would be sent back in order to let me include them in
+ * the version available at <http://www.jhweiss.de/software/snprintf.html>.
+ * However, this is not a requirement for using or redistributing (possibly
+ * modified) versions of this file, nor is leaving this notice intact mandatory.
+ */
+
+/*
+ * History
+ *
+ * 2008-01-20 Holger Weiss <holger@jhweiss.de> for C99-snprintf 1.1:
+ *
+ * 	Fixed the detection of infinite floating point values on IRIX (and
+ * 	possibly other systems) and applied another few minor cleanups.
+ *
+ * 2008-01-06 Holger Weiss <holger@jhweiss.de> for C99-snprintf 1.0:
+ *
+ * 	Added a lot of new features, fixed many bugs, and incorporated various
+ * 	improvements done by Andrew Tridgell <tridge@samba.org>, Russ Allbery
+ * 	<rra@stanford.edu>, Hrvoje Niksic <hniksic@xemacs.org>, Damien Miller
+ * 	<djm@mindrot.org>, and others for the Samba, INN, Wget, and OpenSSH
+ * 	projects.  The additions include: support the "e", "E", "g", "G", and
+ * 	"F" conversion specifiers (and use conversion style "f" or "F" for the
+ * 	still unsupported "a" and "A" specifiers); support the "hh", "ll", "j",
+ * 	"t", and "z" length modifiers; support the "#" flag and the (non-C99)
+ * 	"'" flag; use localeconv(3) (if available) to get both the current
+ * 	locale's decimal point character and the separator between groups of
+ * 	digits; fix the handling of various corner cases of field width and
+ * 	precision specifications; fix various floating point conversion bugs;
+ * 	handle infinite and NaN floating point values; don't attempt to write to
+ * 	the output buffer (which may be NULL) if a size of zero was specified;
+ * 	check for integer overflow of the field width, precision, and return
+ * 	values and during the floating point conversion; use the OUTCHAR() macro
+ * 	instead of a function for better performance; provide asprintf(3) and
+ * 	vasprintf(3) functions; add new test cases.  The replacement functions
+ * 	have been renamed to use an "rpl_" prefix, the function calls in the
+ * 	main project (and in this file) must be redefined accordingly for each
+ * 	replacement function which is needed (by using Autoconf or other means).
+ * 	Various other minor improvements have been applied and the coding style
+ * 	was cleaned up for consistency.
+ *
+ * 2007-07-23 Holger Weiss <holger@jhweiss.de> for Mutt 1.5.13:
+ *
+ * 	C99 compliant snprintf(3) and vsnprintf(3) functions return the number
+ * 	of characters that would have been written to a sufficiently sized
+ * 	buffer (excluding the '\0').  The original code simply returned the
+ * 	length of the resulting output string, so that's been fixed.
+ *
+ * 1998-03-05 Michael Elkins <me@mutt.org> for Mutt 0.90.8:
+ *
+ * 	The original code assumed that both snprintf(3) and vsnprintf(3) were
+ * 	missing.  Some systems only have snprintf(3) but not vsnprintf(3), so
+ * 	the code is now broken down under HAVE_SNPRINTF and HAVE_VSNPRINTF.
+ *
+ * 1998-01-27 Thomas Roessler <roessler@does-not-exist.org> for Mutt 0.89i:
+ *
+ * 	The PGP code was using unsigned hexadecimal formats.  Unfortunately,
+ * 	unsigned formats simply didn't work.
+ *
+ * 1997-10-22 Brandon Long <blong@fiction.net> for Mutt 0.87.1:
+ *
+ * 	Ok, added some minimal floating point support, which means this probably
+ * 	requires libm on most operating systems.  Don't yet support the exponent
+ * 	(e,E) and sigfig (g,G).  Also, fmtint() was pretty badly broken, it just
+ * 	wasn't being exercised in ways which showed it, so that's been fixed.
+ * 	Also, formatted the code to Mutt conventions, and removed dead code left
+ * 	over from the original.  Also, there is now a builtin-test, run with:
+ * 	gcc -DTEST_SNPRINTF -o snprintf snprintf.c -lm && ./snprintf
+ *
+ * 2996-09-15 Brandon Long <blong@fiction.net> for Mutt 0.43:
+ *
+ * 	This was ugly.  It is still ugly.  I opted out of floating point
+ * 	numbers, but the formatter understands just about everything from the
+ * 	normal C string format, at least as far as I can tell from the Solaris
+ * 	2.5 printf(3S) man page.
+ */
+
+/*
+ * ToDo
+ *
+ * - Add wide character support.
+ * - Add support for "%a" and "%A" conversions.
+ * - Create test routines which predefine the expected results.  Our test cases
+ *   usually expose bugs in system implementations rather than in ours :-)
+ */
+
+/*
+ * Usage
+ *
+ * 1) The following preprocessor macros should be defined to 1 if the feature or
+ *    file in question is available on the target system (by using Autoconf or
+ *    other means), though basic functionality should be available as long as
+ *    HAVE_STDARG_H and HAVE_STDLIB_H are defined correctly:
+ *
+ *    	HAVE_VSNPRINTF
+ *    	HAVE_SNPRINTF
+ *    	HAVE_VASPRINTF
+ *    	HAVE_ASPRINTF
+ *    	HAVE_STDARG_H
+ *    	HAVE_STDDEF_H
+ *    	HAVE_STDINT_H
+ *    	HAVE_STDLIB_H
+ *    	HAVE_INTTYPES_H
+ *    	HAVE_LOCALE_H
+ *    	HAVE_LOCALECONV
+ *    	HAVE_LCONV_DECIMAL_POINT
+ *    	HAVE_LCONV_THOUSANDS_SEP
+ *    	HAVE_LONG_DOUBLE
+ *    	HAVE_LONG_LONG_INT
+ *    	HAVE_UNSIGNED_LONG_LONG_INT
+ *    	HAVE_INTMAX_T
+ *    	HAVE_UINTMAX_T
+ *    	HAVE_UINTPTR_T
+ *    	HAVE_PTRDIFF_T
+ *    	HAVE_VA_COPY
+ *    	HAVE___VA_COPY
+ *
+ * 2) The calls to the functions which should be replaced must be redefined
+ *    throughout the project files (by using Autoconf or other means):
+ *
+ *    	#define vsnprintf rpl_vsnprintf
+ *    	#define snprintf rpl_snprintf
+ *    	#define vasprintf rpl_vasprintf
+ *    	#define asprintf rpl_asprintf
+ *
+ * 3) The required replacement functions should be declared in some header file
+ *    included throughout the project files:
+ *
+ *    	#if HAVE_CONFIG_H
+ *    	#include <config.h>
+ *    	#endif
+ *    	#if HAVE_STDARG_H
+ *    	#include <stdarg.h>
+ *    	#if !HAVE_VSNPRINTF
+ *    	int rpl_vsnprintf(char *, size_t, const char *, va_list);
+ *    	#endif
+ *    	#if !HAVE_SNPRINTF
+ *    	int rpl_snprintf(char *, size_t, const char *, ...);
+ *    	#endif
+ *    	#if !HAVE_VASPRINTF
+ *    	int rpl_vasprintf(char **, const char *, va_list);
+ *    	#endif
+ *    	#if !HAVE_ASPRINTF
+ *    	int rpl_asprintf(char **, const char *, ...);
+ *    	#endif
+ *    	#endif
+ *
+ * Autoconf macros for handling step 1 and step 2 are available at
+ * <http://www.jhweiss.de/software/snprintf.html>.
+ */
+
+#if HAVE_CONFIG_H
+#include <config.h>
+#else
+#ifdef WIN32
+#define vsnprintf rpl_vsnprintf
+#define snprintf rpl_snprintf
+#define HAVE_VSNPRINTF 0
+#define HAVE_SNPRINTF 0
+#define HAVE_VASPRINTF 1 /* not needed */
+#define HAVE_ASPRINTF 1 /* not needed */
+#define HAVE_STDARG_H 1
+#define HAVE_STDDEF_H 1
+#define HAVE_STDINT_H 0
+#define HAVE_STDLIB_H 1
+#define HAVE_INTTYPES_H 0
+#define HAVE_LOCALE_H 0
+#define HAVE_LOCALECONV 0
+#define HAVE_LCONV_DECIMAL_POINT 0
+#define HAVE_LCONV_THOUSANDS_SEP 0
+#define HAVE_LONG_DOUBLE 0
+#define HAVE_LONG_LONG_INT 1
+#define HAVE_UNSIGNED_LONG_LONG_INT 1
+#define HAVE_INTMAX_T 0
+#define HAVE_UINTMAX_T 0
+#define HAVE_UINTPTR_T 1
+#define HAVE_PTRDIFF_T 1
+#define HAVE_VA_COPY 0
+#define HAVE___VA_COPY 0
+#else
+#define HAVE_VSNPRINTF 1
+#define HAVE_SNPRINTF 1
+#define HAVE_VASPRINTF 1
+#define HAVE_ASPRINTF 1
+#endif
+#endif	/* HAVE_CONFIG_H */
+
+#if !HAVE_SNPRINTF || !HAVE_VSNPRINTF || !HAVE_ASPRINTF || !HAVE_VASPRINTF
+#include <stdio.h>	/* For NULL, size_t, vsnprintf(3), and vasprintf(3). */
+#ifdef VA_START
+#undef VA_START
+#endif	/* defined(VA_START) */
+#ifdef VA_SHIFT
+#undef VA_SHIFT
+#endif	/* defined(VA_SHIFT) */
+#if HAVE_STDARG_H
+#include <stdarg.h>
+#define VA_START(ap, last) va_start(ap, last)
+#define VA_SHIFT(ap, value, type) /* No-op for ANSI C. */
+#else	/* Assume <varargs.h> is available. */
+#include <varargs.h>
+#define VA_START(ap, last) va_start(ap)	/* "last" is ignored. */
+#define VA_SHIFT(ap, value, type) value = va_arg(ap, type)
+#endif	/* HAVE_STDARG_H */
+
+#if !HAVE_VASPRINTF
+#if HAVE_STDLIB_H
+#include <stdlib.h>	/* For malloc(3). */
+#endif	/* HAVE_STDLIB_H */
+#ifdef VA_COPY
+#undef VA_COPY
+#endif	/* defined(VA_COPY) */
+#ifdef VA_END_COPY
+#undef VA_END_COPY
+#endif	/* defined(VA_END_COPY) */
+#if HAVE_VA_COPY
+#define VA_COPY(dest, src) va_copy(dest, src)
+#define VA_END_COPY(ap) va_end(ap)
+#elif HAVE___VA_COPY
+#define VA_COPY(dest, src) __va_copy(dest, src)
+#define VA_END_COPY(ap) va_end(ap)
+#else
+#define VA_COPY(dest, src) (void)mymemcpy(&dest, &src, sizeof(va_list))
+#define VA_END_COPY(ap) /* No-op. */
+#define NEED_MYMEMCPY 1
+static void *mymemcpy(void *, void *, size_t);
+#endif	/* HAVE_VA_COPY */
+#endif	/* !HAVE_VASPRINTF */
+
+#if !HAVE_VSNPRINTF
+#include <limits.h>	/* For *_MAX. */
+#if HAVE_INTTYPES_H
+#include <inttypes.h>	/* For intmax_t (if not defined in <stdint.h>). */
+#endif	/* HAVE_INTTYPES_H */
+#if HAVE_LOCALE_H
+#include <locale.h>	/* For localeconv(3). */
+#endif	/* HAVE_LOCALE_H */
+#if HAVE_STDDEF_H
+#include <stddef.h>	/* For ptrdiff_t. */
+#endif	/* HAVE_STDDEF_H */
+#if HAVE_STDINT_H
+#include <stdint.h>	/* For intmax_t. */
+#endif	/* HAVE_STDINT_H */
+
+/* Support for unsigned long long int.  We may also need ULLONG_MAX. */
+#ifndef ULONG_MAX	/* We may need ULONG_MAX as a fallback. */
+#ifdef UINT_MAX
+#define ULONG_MAX UINT_MAX
+#else
+#define ULONG_MAX INT_MAX
+#endif	/* defined(UINT_MAX) */
+#endif	/* !defined(ULONG_MAX) */
+#ifdef ULLONG
+#undef ULLONG
+#endif	/* defined(ULLONG) */
+#if HAVE_UNSIGNED_LONG_LONG_INT
+#define ULLONG unsigned long long int
+#ifndef ULLONG_MAX
+#define ULLONG_MAX ULONG_MAX
+#endif	/* !defined(ULLONG_MAX) */
+#else
+#define ULLONG unsigned long int
+#ifdef ULLONG_MAX
+#undef ULLONG_MAX
+#endif	/* defined(ULLONG_MAX) */
+#define ULLONG_MAX ULONG_MAX
+#endif	/* HAVE_LONG_LONG_INT */
+
+/* Support for uintmax_t.  We also need UINTMAX_MAX. */
+#ifdef UINTMAX_T
+#undef UINTMAX_T
+#endif	/* defined(UINTMAX_T) */
+#if HAVE_UINTMAX_T || defined(uintmax_t)
+#define UINTMAX_T uintmax_t
+#ifndef UINTMAX_MAX
+#define UINTMAX_MAX ULLONG_MAX
+#endif	/* !defined(UINTMAX_MAX) */
+#else
+#define UINTMAX_T ULLONG
+#ifdef UINTMAX_MAX
+#undef UINTMAX_MAX
+#endif	/* defined(UINTMAX_MAX) */
+#define UINTMAX_MAX ULLONG_MAX
+#endif	/* HAVE_UINTMAX_T || defined(uintmax_t) */
+
+/* Support for long double. */
+#ifndef LDOUBLE
+#if HAVE_LONG_DOUBLE
+#define LDOUBLE long double
+#else
+#define LDOUBLE double
+#endif	/* HAVE_LONG_DOUBLE */
+#endif	/* !defined(LDOUBLE) */
+
+/* Support for long long int. */
+#ifndef LLONG
+#if HAVE_LONG_LONG_INT
+#define LLONG long long int
+#else
+#define LLONG long int
+#endif	/* HAVE_LONG_LONG_INT */
+#endif	/* !defined(LLONG) */
+
+/* Support for intmax_t. */
+#ifndef INTMAX_T
+#if HAVE_INTMAX_T || defined(intmax_t)
+#define INTMAX_T intmax_t
+#else
+#define INTMAX_T LLONG
+#endif	/* HAVE_INTMAX_T || defined(intmax_t) */
+#endif	/* !defined(INTMAX_T) */
+
+/* Support for uintptr_t. */
+#ifndef UINTPTR_T
+#if HAVE_UINTPTR_T || defined(uintptr_t)
+#define UINTPTR_T uintptr_t
+#else
+#define UINTPTR_T unsigned long int
+#endif	/* HAVE_UINTPTR_T || defined(uintptr_t) */
+#endif	/* !defined(UINTPTR_T) */
+
+/* Support for ptrdiff_t. */
+#ifndef PTRDIFF_T
+#if HAVE_PTRDIFF_T || defined(ptrdiff_t)
+#define PTRDIFF_T ptrdiff_t
+#else
+#define PTRDIFF_T long int
+#endif	/* HAVE_PTRDIFF_T || defined(ptrdiff_t) */
+#endif	/* !defined(PTRDIFF_T) */
+
+/*
+ * We need an unsigned integer type corresponding to ptrdiff_t (cf. C99:
+ * 7.19.6.1, 7).  However, we'll simply use PTRDIFF_T and convert it to an
+ * unsigned type if necessary.  This should work just fine in practice.
+ */
+#ifndef UPTRDIFF_T
+#define UPTRDIFF_T PTRDIFF_T
+#endif	/* !defined(UPTRDIFF_T) */
+
+/*
+ * We need a signed integer type corresponding to size_t (cf. C99: 7.19.6.1, 7).
+ * However, we'll simply use size_t and convert it to a signed type if
+ * necessary.  This should work just fine in practice.
+ */
+#ifndef SSIZE_T
+#define SSIZE_T size_t
+#endif	/* !defined(SSIZE_T) */
+
+/* Either ERANGE or E2BIG should be available everywhere. */
+#ifndef ERANGE
+#define ERANGE E2BIG
+#endif	/* !defined(ERANGE) */
+#ifndef EOVERFLOW
+#define EOVERFLOW ERANGE
+#endif	/* !defined(EOVERFLOW) */
+
+/*
+ * Buffer size to hold the octal string representation of UINT128_MAX without
+ * nul-termination ("3777777777777777777777777777777777777777777").
+ */
+#ifdef MAX_CONVERT_LENGTH
+#undef MAX_CONVERT_LENGTH
+#endif	/* defined(MAX_CONVERT_LENGTH) */
+#define MAX_CONVERT_LENGTH      43
+
+/* Format read states. */
+#define PRINT_S_DEFAULT         0
+#define PRINT_S_FLAGS           1
+#define PRINT_S_WIDTH           2
+#define PRINT_S_DOT             3
+#define PRINT_S_PRECISION       4
+#define PRINT_S_MOD             5
+#define PRINT_S_CONV            6
+
+/* Format flags. */
+#define PRINT_F_MINUS           (1 << 0)
+#define PRINT_F_PLUS            (1 << 1)
+#define PRINT_F_SPACE           (1 << 2)
+#define PRINT_F_NUM             (1 << 3)
+#define PRINT_F_ZERO            (1 << 4)
+#define PRINT_F_QUOTE           (1 << 5)
+#define PRINT_F_UP              (1 << 6)
+#define PRINT_F_UNSIGNED        (1 << 7)
+#define PRINT_F_TYPE_G          (1 << 8)
+#define PRINT_F_TYPE_E          (1 << 9)
+
+/* Conversion flags. */
+#define PRINT_C_CHAR            1
+#define PRINT_C_SHORT           2
+#define PRINT_C_LONG            3
+#define PRINT_C_LLONG           4
+#define PRINT_C_LDOUBLE         5
+#define PRINT_C_SIZE            6
+#define PRINT_C_PTRDIFF         7
+#define PRINT_C_INTMAX          8
+
+#ifndef MAX
+#define MAX(x, y) ((x >= y) ? x : y)
+#endif	/* !defined(MAX) */
+#ifndef CHARTOINT
+#define CHARTOINT(ch) (ch - '0')
+#endif	/* !defined(CHARTOINT) */
+#ifndef ISDIGIT
+#define ISDIGIT(ch) ('0' <= (unsigned char)ch && (unsigned char)ch <= '9')
+#endif	/* !defined(ISDIGIT) */
+#ifndef ISNAN
+#define ISNAN(x) (x != x)
+#endif	/* !defined(ISNAN) */
+#ifndef ISINF
+#define ISINF(x) (x != 0.0 && x + x == x)
+#endif	/* !defined(ISINF) */
+
+#ifdef OUTCHAR
+#undef OUTCHAR
+#endif	/* defined(OUTCHAR) */
+#define OUTCHAR(str, len, size, ch)                                          \
+do {                                                                         \
+	if (len + 1 < size)                                                  \
+		str[len] = ch;                                               \
+	(len)++;                                                             \
+} while (/* CONSTCOND */ 0)
+
+static void fmtstr(char *, size_t *, size_t, const char *, int, int, int);
+static void fmtint(char *, size_t *, size_t, INTMAX_T, int, int, int, int);
+static void fmtflt(char *, size_t *, size_t, LDOUBLE, int, int, int, int *);
+static void printsep(char *, size_t *, size_t);
+static int getnumsep(int);
+static int getexponent(LDOUBLE);
+static int convert(UINTMAX_T, char *, size_t, int, int);
+static UINTMAX_T cast(LDOUBLE);
+static UINTMAX_T myround(LDOUBLE);
+static LDOUBLE mypow10(int);
+
+int
+rpl_vsnprintf(char *str, size_t size, const char *format, va_list args)
+{
+	LDOUBLE fvalue;
+	INTMAX_T value;
+	unsigned char cvalue;
+	const char *strvalue;
+	INTMAX_T *intmaxptr;
+	PTRDIFF_T *ptrdiffptr;
+	SSIZE_T *sizeptr;
+	LLONG *llongptr;
+	long int *longptr;
+	int *intptr;
+	short int *shortptr;
+	signed char *charptr;
+	size_t len = 0;
+	int overflow = 0;
+	int base = 0;
+	int cflags = 0;
+	int flags = 0;
+	int width = 0;
+	int precision = -1;
+	int state = PRINT_S_DEFAULT;
+	char ch = *format++;
+
+	/*
+	 * C99 says: "If `n' is zero, nothing is written, and `s' may be a null
+	 * pointer." (7.19.6.5, 2)  We're forgiving and allow a NULL pointer
+	 * even if a size larger than zero was specified.  At least NetBSD's
+	 * snprintf(3) does the same, as well as other versions of this file.
+	 * (Though some of these versions will write to a non-NULL buffer even
+	 * if a size of zero was specified, which violates the standard.)
+	 */
+	if (str == NULL && size != 0)
+		size = 0;
+
+	while (ch != '\0')
+		switch (state) {
+		case PRINT_S_DEFAULT:
+			if (ch == '%')
+				state = PRINT_S_FLAGS;
+			else
+				OUTCHAR(str, len, size, ch);
+			ch = *format++;
+			break;
+		case PRINT_S_FLAGS:
+			switch (ch) {
+			case '-':
+				flags |= PRINT_F_MINUS;
+				ch = *format++;
+				break;
+			case '+':
+				flags |= PRINT_F_PLUS;
+				ch = *format++;
+				break;
+			case ' ':
+				flags |= PRINT_F_SPACE;
+				ch = *format++;
+				break;
+			case '#':
+				flags |= PRINT_F_NUM;
+				ch = *format++;
+				break;
+			case '0':
+				flags |= PRINT_F_ZERO;
+				ch = *format++;
+				break;
+			case '\'':	/* SUSv2 flag (not in C99). */
+				flags |= PRINT_F_QUOTE;
+				ch = *format++;
+				break;
+			default:
+				state = PRINT_S_WIDTH;
+				break;
+			}
+			break;
+		case PRINT_S_WIDTH:
+			if (ISDIGIT(ch)) {
+				ch = CHARTOINT(ch);
+				if (width > (INT_MAX - ch) / 10) {
+					overflow = 1;
+					goto out;
+				}
+				width = 10 * width + ch;
+				ch = *format++;
+			} else if (ch == '*') {
+				/*
+				 * C99 says: "A negative field width argument is
+				 * taken as a `-' flag followed by a positive
+				 * field width." (7.19.6.1, 5)
+				 */
+				if ((width = va_arg(args, int)) < 0) {
+					flags |= PRINT_F_MINUS;
+					width = -width;
+				}
+				ch = *format++;
+				state = PRINT_S_DOT;
+			} else
+				state = PRINT_S_DOT;
+			break;
+		case PRINT_S_DOT:
+			if (ch == '.') {
+				state = PRINT_S_PRECISION;
+				ch = *format++;
+			} else
+				state = PRINT_S_MOD;
+			break;
+		case PRINT_S_PRECISION:
+			if (precision == -1)
+				precision = 0;
+			if (ISDIGIT(ch)) {
+				ch = CHARTOINT(ch);
+				if (precision > (INT_MAX - ch) / 10) {
+					overflow = 1;
+					goto out;
+				}
+				precision = 10 * precision + ch;
+				ch = *format++;
+			} else if (ch == '*') {
+				/*
+				 * C99 says: "A negative precision argument is
+				 * taken as if the precision were omitted."
+				 * (7.19.6.1, 5)
+				 */
+				if ((precision = va_arg(args, int)) < 0)
+					precision = -1;
+				ch = *format++;
+				state = PRINT_S_MOD;
+			} else
+				state = PRINT_S_MOD;
+			break;
+		case PRINT_S_MOD:
+			switch (ch) {
+			case 'h':
+				ch = *format++;
+				if (ch == 'h') {	/* It's a char. */
+					ch = *format++;
+					cflags = PRINT_C_CHAR;
+				} else
+					cflags = PRINT_C_SHORT;
+				break;
+			case 'l':
+				ch = *format++;
+				if (ch == 'l') {	/* It's a long long. */
+					ch = *format++;
+					cflags = PRINT_C_LLONG;
+				} else
+					cflags = PRINT_C_LONG;
+				break;
+			case 'L':
+				cflags = PRINT_C_LDOUBLE;
+				ch = *format++;
+				break;
+			case 'j':
+				cflags = PRINT_C_INTMAX;
+				ch = *format++;
+				break;
+			case 't':
+				cflags = PRINT_C_PTRDIFF;
+				ch = *format++;
+				break;
+			case 'z':
+				cflags = PRINT_C_SIZE;
+				ch = *format++;
+				break;
+			}
+			state = PRINT_S_CONV;
+			break;
+		case PRINT_S_CONV:
+			switch (ch) {
+			case 'd':
+				/* FALLTHROUGH */
+			case 'i':
+				switch (cflags) {
+				case PRINT_C_CHAR:
+					value = (signed char)va_arg(args, int);
+					break;
+				case PRINT_C_SHORT:
+					value = (short int)va_arg(args, int);
+					break;
+				case PRINT_C_LONG:
+					value = va_arg(args, long int);
+					break;
+				case PRINT_C_LLONG:
+					value = va_arg(args, LLONG);
+					break;
+				case PRINT_C_SIZE:
+					value = va_arg(args, SSIZE_T);
+					break;
+				case PRINT_C_INTMAX:
+					value = va_arg(args, INTMAX_T);
+					break;
+				case PRINT_C_PTRDIFF:
+					value = va_arg(args, PTRDIFF_T);
+					break;
+				default:
+					value = va_arg(args, int);
+					break;
+				}
+				fmtint(str, &len, size, value, 10, width,
+				    precision, flags);
+				break;
+			case 'X':
+				flags |= PRINT_F_UP;
+				/* FALLTHROUGH */
+			case 'x':
+				base = 16;
+				/* FALLTHROUGH */
+			case 'o':
+				if (base == 0)
+					base = 8;
+				/* FALLTHROUGH */
+			case 'u':
+				if (base == 0)
+					base = 10;
+				flags |= PRINT_F_UNSIGNED;
+				switch (cflags) {
+				case PRINT_C_CHAR:
+					value = (unsigned char)va_arg(args,
+					    unsigned int);
+					break;
+				case PRINT_C_SHORT:
+					value = (unsigned short int)va_arg(args,
+					    unsigned int);
+					break;
+				case PRINT_C_LONG:
+					value = va_arg(args, unsigned long int);
+					break;
+				case PRINT_C_LLONG:
+					value = va_arg(args, ULLONG);
+					break;
+				case PRINT_C_SIZE:
+					value = va_arg(args, size_t);
+					break;
+				case PRINT_C_INTMAX:
+					value = va_arg(args, UINTMAX_T);
+					break;
+				case PRINT_C_PTRDIFF:
+					value = va_arg(args, UPTRDIFF_T);
+					break;
+				default:
+					value = va_arg(args, unsigned int);
+					break;
+				}
+				fmtint(str, &len, size, value, base, width,
+				    precision, flags);
+				break;
+			case 'A':
+				/* Not yet supported, we'll use "%F". */
+				/* FALLTHROUGH */
+			case 'F':
+				flags |= PRINT_F_UP;
+			case 'a':
+				/* Not yet supported, we'll use "%f". */
+				/* FALLTHROUGH */
+			case 'f':
+				if (cflags == PRINT_C_LDOUBLE)
+					fvalue = va_arg(args, LDOUBLE);
+				else
+					fvalue = va_arg(args, double);
+				fmtflt(str, &len, size, fvalue, width,
+				    precision, flags, &overflow);
+				if (overflow)
+					goto out;
+				break;
+			case 'E':
+				flags |= PRINT_F_UP;
+				/* FALLTHROUGH */
+			case 'e':
+				flags |= PRINT_F_TYPE_E;
+				if (cflags == PRINT_C_LDOUBLE)
+					fvalue = va_arg(args, LDOUBLE);
+				else
+					fvalue = va_arg(args, double);
+				fmtflt(str, &len, size, fvalue, width,
+				    precision, flags, &overflow);
+				if (overflow)
+					goto out;
+				break;
+			case 'G':
+				flags |= PRINT_F_UP;
+				/* FALLTHROUGH */
+			case 'g':
+				flags |= PRINT_F_TYPE_G;
+				if (cflags == PRINT_C_LDOUBLE)
+					fvalue = va_arg(args, LDOUBLE);
+				else
+					fvalue = va_arg(args, double);
+				/*
+				 * If the precision is zero, it is treated as
+				 * one (cf. C99: 7.19.6.1, 8).
+				 */
+				if (precision == 0)
+					precision = 1;
+				fmtflt(str, &len, size, fvalue, width,
+				    precision, flags, &overflow);
+				if (overflow)
+					goto out;
+				break;
+			case 'c':
+				cvalue = (unsigned char)va_arg(args, int);
+				OUTCHAR(str, len, size, cvalue);
+				break;
+			case 's':
+				strvalue = va_arg(args, char *);
+				fmtstr(str, &len, size, strvalue, width,
+				    precision, flags);
+				break;
+			case 'p':
+				/*
+				 * C99 says: "The value of the pointer is
+				 * converted to a sequence of printing
+				 * characters, in an implementation-defined
+				 * manner." (C99: 7.19.6.1, 8)
+				 */
+				if ((strvalue = va_arg(args, void *)) == NULL)
+					/*
+					 * We use the glibc format.  BSD prints
+					 * "0x0", SysV "0".
+					 */
+					fmtstr(str, &len, size, "(nil)", width,
+					    -1, flags);
+				else {
+					/*
+					 * We use the BSD/glibc format.  SysV
+					 * omits the "0x" prefix (which we emit
+					 * using the PRINT_F_NUM flag).
+					 */
+					flags |= PRINT_F_NUM;
+					flags |= PRINT_F_UNSIGNED;
+					fmtint(str, &len, size,
+					    (UINTPTR_T)strvalue, 16, width,
+					    precision, flags);
+				}
+				break;
+			case 'n':
+				switch (cflags) {
+				case PRINT_C_CHAR:
+					charptr = va_arg(args, signed char *);
+					*charptr = len;
+					break;
+				case PRINT_C_SHORT:
+					shortptr = va_arg(args, short int *);
+					*shortptr = len;
+					break;
+				case PRINT_C_LONG:
+					longptr = va_arg(args, long int *);
+					*longptr = len;
+					break;
+				case PRINT_C_LLONG:
+					llongptr = va_arg(args, LLONG *);
+					*llongptr = len;
+					break;
+				case PRINT_C_SIZE:
+					/*
+					 * C99 says that with the "z" length
+					 * modifier, "a following `n' conversion
+					 * specifier applies to a pointer to a
+					 * signed integer type corresponding to
+					 * size_t argument." (7.19.6.1, 7)
+					 */
+					sizeptr = va_arg(args, SSIZE_T *);
+					*sizeptr = len;
+					break;
+				case PRINT_C_INTMAX:
+					intmaxptr = va_arg(args, INTMAX_T *);
+					*intmaxptr = len;
+					break;
+				case PRINT_C_PTRDIFF:
+					ptrdiffptr = va_arg(args, PTRDIFF_T *);
+					*ptrdiffptr = len;
+					break;
+				default:
+					intptr = va_arg(args, int *);
+					*intptr = len;
+					break;
+				}
+				break;
+			case '%':	/* Print a "%" character verbatim. */
+				OUTCHAR(str, len, size, ch);
+				break;
+			default:	/* Skip other characters. */
+				break;
+			}
+			ch = *format++;
+			state = PRINT_S_DEFAULT;
+			base = cflags = flags = width = 0;
+			precision = -1;
+			break;
+		}
+out:
+	if (len < size)
+		str[len] = '\0';
+	else if (size > 0)
+		str[size - 1] = '\0';
+
+	if (overflow || len >= INT_MAX) {
+		return -1;
+	}
+	return (int)len;
+}
+
+static void
+fmtstr(char *str, size_t *len, size_t size, const char *value, int width,
+       int precision, int flags)
+{
+	int padlen, strln;	/* Amount to pad. */
+	int noprecision = (precision == -1);
+
+	if (value == NULL)	/* We're forgiving. */
+		value = "(null)";
+
+	/* If a precision was specified, don't read the string past it. */
+	for (strln = 0; value[strln] != '\0' &&
+	    (noprecision || strln < precision); strln++)
+		continue;
+
+	if ((padlen = width - strln) < 0)
+		padlen = 0;
+	if (flags & PRINT_F_MINUS)	/* Left justify. */
+		padlen = -padlen;
+
+	while (padlen > 0) {	/* Leading spaces. */
+		OUTCHAR(str, *len, size, ' ');
+		padlen--;
+	}
+	while (*value != '\0' && (noprecision || precision-- > 0)) {
+		OUTCHAR(str, *len, size, *value);
+		value++;
+	}
+	while (padlen < 0) {	/* Trailing spaces. */
+		OUTCHAR(str, *len, size, ' ');
+		padlen++;
+	}
+}
+
+static void
+fmtint(char *str, size_t *len, size_t size, INTMAX_T value, int base, int width,
+       int precision, int flags)
+{
+	UINTMAX_T uvalue;
+	char iconvert[MAX_CONVERT_LENGTH];
+	char sign = 0;
+	char hexprefix = 0;
+	int spadlen = 0;	/* Amount to space pad. */
+	int zpadlen = 0;	/* Amount to zero pad. */
+	int pos;
+	int separators = (flags & PRINT_F_QUOTE);
+	int noprecision = (precision == -1);
+
+	if (flags & PRINT_F_UNSIGNED)
+		uvalue = value;
+	else {
+		uvalue = (value >= 0) ? value : -value;
+		if (value < 0)
+			sign = '-';
+		else if (flags & PRINT_F_PLUS)	/* Do a sign. */
+			sign = '+';
+		else if (flags & PRINT_F_SPACE)
+			sign = ' ';
+	}
+
+	pos = convert(uvalue, iconvert, sizeof(iconvert), base,
+	    flags & PRINT_F_UP);
+
+	if (flags & PRINT_F_NUM && uvalue != 0) {
+		/*
+		 * C99 says: "The result is converted to an `alternative form'.
+		 * For `o' conversion, it increases the precision, if and only
+		 * if necessary, to force the first digit of the result to be a
+		 * zero (if the value and precision are both 0, a single 0 is
+		 * printed).  For `x' (or `X') conversion, a nonzero result has
+		 * `0x' (or `0X') prefixed to it." (7.19.6.1, 6)
+		 */
+		switch (base) {
+		case 8:
+			if (precision <= pos)
+				precision = pos + 1;
+			break;
+		case 16:
+			hexprefix = (flags & PRINT_F_UP) ? 'X' : 'x';
+			break;
+		}
+	}
+
+	if (separators)	/* Get the number of group separators we'll print. */
+		separators = getnumsep(pos);
+
+	zpadlen = precision - pos - separators;
+	spadlen = width                         /* Minimum field width. */
+	    - separators                        /* Number of separators. */
+	    - MAX(precision, pos)               /* Number of integer digits. */
+	    - ((sign != 0) ? 1 : 0)             /* Will we print a sign? */
+	    - ((hexprefix != 0) ? 2 : 0);       /* Will we print a prefix? */
+
+	if (zpadlen < 0)
+		zpadlen = 0;
+	if (spadlen < 0)
+		spadlen = 0;
+
+	/*
+	 * C99 says: "If the `0' and `-' flags both appear, the `0' flag is
+	 * ignored.  For `d', `i', `o', `u', `x', and `X' conversions, if a
+	 * precision is specified, the `0' flag is ignored." (7.19.6.1, 6)
+	 */
+	if (flags & PRINT_F_MINUS)	/* Left justify. */
+		spadlen = -spadlen;
+	else if (flags & PRINT_F_ZERO && noprecision) {
+		zpadlen += spadlen;
+		spadlen = 0;
+	}
+	while (spadlen > 0) {	/* Leading spaces. */
+		OUTCHAR(str, *len, size, ' ');
+		spadlen--;
+	}
+	if (sign != 0)	/* Sign. */
+		OUTCHAR(str, *len, size, sign);
+	if (hexprefix != 0) {	/* A "0x" or "0X" prefix. */
+		OUTCHAR(str, *len, size, '0');
+		OUTCHAR(str, *len, size, hexprefix);
+	}
+	while (zpadlen > 0) {	/* Leading zeros. */
+		OUTCHAR(str, *len, size, '0');
+		zpadlen--;
+	}
+	while (pos > 0) {	/* The actual digits. */
+		pos--;
+		OUTCHAR(str, *len, size, iconvert[pos]);
+		if (separators > 0 && pos > 0 && pos % 3 == 0)
+			printsep(str, len, size);
+	}
+	while (spadlen < 0) {	/* Trailing spaces. */
+		OUTCHAR(str, *len, size, ' ');
+		spadlen++;
+	}
+}
+
+static void
+fmtflt(char *str, size_t *len, size_t size, LDOUBLE fvalue, int width,
+       int precision, int flags, int *overflow)
+{
+	LDOUBLE ufvalue;
+	UINTMAX_T intpart;
+	UINTMAX_T fracpart;
+	UINTMAX_T mask;
+	const char *infnan = NULL;
+	char iconvert[MAX_CONVERT_LENGTH];
+	char fconvert[MAX_CONVERT_LENGTH];
+	char econvert[4];	/* "e-12" (without nul-termination). */
+	char esign = 0;
+	char sign = 0;
+	int leadfraczeros = 0;
+	int exponent = 0;
+	int emitpoint = 0;
+	int omitzeros = 0;
+	int omitcount = 0;
+	int padlen = 0;
+	int epos = 0;
+	int fpos = 0;
+	int ipos = 0;
+	int separators = (flags & PRINT_F_QUOTE);
+	int estyle = (flags & PRINT_F_TYPE_E);
+#if HAVE_LOCALECONV && HAVE_LCONV_DECIMAL_POINT
+	struct lconv *lc = localeconv();
+#endif	/* HAVE_LOCALECONV && HAVE_LCONV_DECIMAL_POINT */
+
+	/*
+	 * AIX' man page says the default is 0, but C99 and at least Solaris'
+	 * and NetBSD's man pages say the default is 6, and sprintf(3) on AIX
+	 * defaults to 6.
+	 */
+	if (precision == -1)
+		precision = 6;
+
+	if (fvalue < 0.0)
+		sign = '-';
+	else if (flags & PRINT_F_PLUS)	/* Do a sign. */
+		sign = '+';
+	else if (flags & PRINT_F_SPACE)
+		sign = ' ';
+
+	if (ISNAN(fvalue))
+		infnan = (flags & PRINT_F_UP) ? "NAN" : "nan";
+	else if (ISINF(fvalue))
+		infnan = (flags & PRINT_F_UP) ? "INF" : "inf";
+
+	if (infnan != NULL) {
+		if (sign != 0)
+			iconvert[ipos++] = sign;
+		while (*infnan != '\0')
+			iconvert[ipos++] = *infnan++;
+		fmtstr(str, len, size, iconvert, width, ipos, flags);
+		return;
+	}
+
+	/* "%e" (or "%E") or "%g" (or "%G") conversion. */
+	if (flags & PRINT_F_TYPE_E || flags & PRINT_F_TYPE_G) {
+		if (flags & PRINT_F_TYPE_G) {
+			/*
+			 * For "%g" (and "%G") conversions, the precision
+			 * specifies the number of significant digits, which
+			 * includes the digits in the integer part.  The
+			 * conversion will or will not be using "e-style" (like
+			 * "%e" or "%E" conversions) depending on the precision
+			 * and on the exponent.  However, the exponent can be
+			 * affected by rounding the converted value, so we'll
+			 * leave this decision for later.  Until then, we'll
+			 * assume that we're going to do an "e-style" conversion
+			 * (in order to get the exponent calculated).  For
+			 * "e-style", the precision must be decremented by one.
+			 */
+			precision--;
+			/*
+			 * For "%g" (and "%G") conversions, trailing zeros are
+			 * removed from the fractional portion of the result
+			 * unless the "#" flag was specified.
+			 */
+			if (!(flags & PRINT_F_NUM))
+				omitzeros = 1;
+		}
+		exponent = getexponent(fvalue);
+		estyle = 1;
+	}
+
+again:
+	/*
+	 * Sorry, we only support 9, 19, or 38 digits (that is, the number of
+	 * digits of the 32-bit, the 64-bit, or the 128-bit UINTMAX_MAX value
+	 * minus one) past the decimal point due to our conversion method.
+	 */
+	switch (sizeof(UINTMAX_T)) {
+	case 16:
+		if (precision > 38)
+			precision = 38;
+		break;
+	case 8:
+		if (precision > 19)
+			precision = 19;
+		break;
+	default:
+		if (precision > 9)
+			precision = 9;
+		break;
+	}
+
+	ufvalue = (fvalue >= 0.0) ? fvalue : -fvalue;
+	if (estyle)	/* We want exactly one integer digit. */
+		ufvalue /= mypow10(exponent);
+
+	if ((intpart = cast(ufvalue)) == UINTMAX_MAX) {
+		*overflow = 1;
+		return;
+	}
+
+	/*
+	 * Factor of ten with the number of digits needed for the fractional
+	 * part.  For example, if the precision is 3, the mask will be 1000.
+	 */
+	mask = (UINTMAX_T)mypow10(precision);
+	/*
+	 * We "cheat" by converting the fractional part to integer by
+	 * multiplying by a factor of ten.
+	 */
+	if ((fracpart = myround(mask * (ufvalue - intpart))) >= mask) {
+		/*
+		 * For example, ufvalue = 2.99962, intpart = 2, and mask = 1000
+		 * (because precision = 3).  Now, myround(1000 * 0.99962) will
+		 * return 1000.  So, the integer part must be incremented by one
+		 * and the fractional part must be set to zero.
+		 */
+		intpart++;
+		fracpart = 0;
+		if (estyle && intpart == 10) {
+			/*
+			 * The value was rounded up to ten, but we only want one
+			 * integer digit if using "e-style".  So, the integer
+			 * part must be set to one and the exponent must be
+			 * incremented by one.
+			 */
+			intpart = 1;
+			exponent++;
+		}
+	}
+
+	/*
+	 * Now that we know the real exponent, we can check whether or not to
+	 * use "e-style" for "%g" (and "%G") conversions.  If we don't need
+	 * "e-style", the precision must be adjusted and the integer and
+	 * fractional parts must be recalculated from the original value.
+	 *
+	 * C99 says: "Let P equal the precision if nonzero, 6 if the precision
+	 * is omitted, or 1 if the precision is zero.  Then, if a conversion
+	 * with style `E' would have an exponent of X:
+	 *
+	 * - if P > X >= -4, the conversion is with style `f' (or `F') and
+	 *   precision P - (X + 1).
+	 *
+	 * - otherwise, the conversion is with style `e' (or `E') and precision
+	 *   P - 1." (7.19.6.1, 8)
+	 *
+	 * Note that we had decremented the precision by one.
+	 */
+	if (flags & PRINT_F_TYPE_G && estyle &&
+	    precision + 1 > exponent && exponent >= -4) {
+		precision -= exponent;
+		estyle = 0;
+		goto again;
+	}
+
+	if (estyle) {
+		if (exponent < 0) {
+			exponent = -exponent;
+			esign = '-';
+		} else
+			esign = '+';
+
+		/*
+		 * Convert the exponent.  The sizeof(econvert) is 4.  So, the
+		 * econvert buffer can hold e.g. "e+99" and "e-99".  We don't
+		 * support an exponent which contains more than two digits.
+		 * Therefore, the following stores are safe.
+		 */
+		epos = convert(exponent, econvert, 2, 10, 0);
+		/*
+		 * C99 says: "The exponent always contains at least two digits,
+		 * and only as many more digits as necessary to represent the
+		 * exponent." (7.19.6.1, 8)
+		 */
+		if (epos == 1)
+			econvert[epos++] = '0';
+		econvert[epos++] = esign;
+		econvert[epos++] = (flags & PRINT_F_UP) ? 'E' : 'e';
+	}
+
+	/* Convert the integer part and the fractional part. */
+	ipos = convert(intpart, iconvert, sizeof(iconvert), 10, 0);
+	if (fracpart != 0)	/* convert() would return 1 if fracpart == 0. */
+		fpos = convert(fracpart, fconvert, sizeof(fconvert), 10, 0);
+
+	leadfraczeros = precision - fpos;
+
+	if (omitzeros) {
+		if (fpos > 0)	/* Omit trailing fractional part zeros. */
+			while (omitcount < fpos && fconvert[omitcount] == '0')
+				omitcount++;
+		else {	/* The fractional part is zero, omit it completely. */
+			omitcount = precision;
+			leadfraczeros = 0;
+		}
+		precision -= omitcount;
+	}
+
+	/*
+	 * Print a decimal point if either the fractional part is non-zero
+	 * and/or the "#" flag was specified.
+	 */
+	if (precision > 0 || flags & PRINT_F_NUM)
+		emitpoint = 1;
+	if (separators)	/* Get the number of group separators we'll print. */
+		separators = getnumsep(ipos);
+
+	padlen = width                  /* Minimum field width. */
+	    - ipos                      /* Number of integer digits. */
+	    - epos                      /* Number of exponent characters. */
+	    - precision                 /* Number of fractional digits. */
+	    - separators                /* Number of group separators. */
+	    - (emitpoint ? 1 : 0)       /* Will we print a decimal point? */
+	    - ((sign != 0) ? 1 : 0);    /* Will we print a sign character? */
+
+	if (padlen < 0)
+		padlen = 0;
+
+	/*
+	 * C99 says: "If the `0' and `-' flags both appear, the `0' flag is
+	 * ignored." (7.19.6.1, 6)
+	 */
+	if (flags & PRINT_F_MINUS)	/* Left justifty. */
+		padlen = -padlen;
+	else if (flags & PRINT_F_ZERO && padlen > 0) {
+		if (sign != 0) {	/* Sign. */
+			OUTCHAR(str, *len, size, sign);
+			sign = 0;
+		}
+		while (padlen > 0) {	/* Leading zeros. */
+			OUTCHAR(str, *len, size, '0');
+			padlen--;
+		}
+	}
+	while (padlen > 0) {	/* Leading spaces. */
+		OUTCHAR(str, *len, size, ' ');
+		padlen--;
+	}
+	if (sign != 0)	/* Sign. */
+		OUTCHAR(str, *len, size, sign);
+	while (ipos > 0) {	/* Integer part. */
+		ipos--;
+		OUTCHAR(str, *len, size, iconvert[ipos]);
+		if (separators > 0 && ipos > 0 && ipos % 3 == 0)
+			printsep(str, len, size);
+	}
+	if (emitpoint) {	/* Decimal point. */
+#if HAVE_LOCALECONV && HAVE_LCONV_DECIMAL_POINT
+		if (lc->decimal_point != NULL && *lc->decimal_point != '\0')
+			OUTCHAR(str, *len, size, *lc->decimal_point);
+		else	/* We'll always print some decimal point character. */
+#endif	/* HAVE_LOCALECONV && HAVE_LCONV_DECIMAL_POINT */
+			OUTCHAR(str, *len, size, '.');
+	}
+	while (leadfraczeros > 0) {	/* Leading fractional part zeros. */
+		OUTCHAR(str, *len, size, '0');
+		leadfraczeros--;
+	}
+	while (fpos > omitcount) {	/* The remaining fractional part. */
+		fpos--;
+		OUTCHAR(str, *len, size, fconvert[fpos]);
+	}
+	while (epos > 0) {	/* Exponent. */
+		epos--;
+		OUTCHAR(str, *len, size, econvert[epos]);
+	}
+	while (padlen < 0) {	/* Trailing spaces. */
+		OUTCHAR(str, *len, size, ' ');
+		padlen++;
+	}
+}
+
+static void
+printsep(char *str, size_t *len, size_t size)
+{
+#if HAVE_LOCALECONV && HAVE_LCONV_THOUSANDS_SEP
+	struct lconv *lc = localeconv();
+	int i;
+
+	if (lc->thousands_sep != NULL)
+		for (i = 0; lc->thousands_sep[i] != '\0'; i++)
+			OUTCHAR(str, *len, size, lc->thousands_sep[i]);
+	else
+#endif	/* HAVE_LOCALECONV && HAVE_LCONV_THOUSANDS_SEP */
+		OUTCHAR(str, *len, size, ',');
+}
+
+static int
+getnumsep(int digits)
+{
+	int separators = (digits - ((digits % 3 == 0) ? 1 : 0)) / 3;
+#if HAVE_LOCALECONV && HAVE_LCONV_THOUSANDS_SEP
+	int strln;
+	struct lconv *lc = localeconv();
+
+	/* We support an arbitrary separator length (including zero). */
+	if (lc->thousands_sep != NULL) {
+		for (strln = 0; lc->thousands_sep[strln] != '\0'; strln++)
+			continue;
+		separators *= strln;
+	}
+#endif	/* HAVE_LOCALECONV && HAVE_LCONV_THOUSANDS_SEP */
+	return separators;
+}
+
+static int
+getexponent(LDOUBLE value)
+{
+	LDOUBLE tmp = (value >= 0.0) ? value : -value;
+	int exponent = 0;
+
+	/*
+	 * We check for 99 > exponent > -99 in order to work around possible
+	 * endless loops which could happen (at least) in the second loop (at
+	 * least) if we're called with an infinite value.  However, we checked
+	 * for infinity before calling this function using our ISINF() macro, so
+	 * this might be somewhat paranoid.
+	 */
+	while (tmp < 1.0 && tmp > 0.0 && --exponent > -99)
+		tmp *= 10;
+	while (tmp >= 10.0 && ++exponent < 99)
+		tmp /= 10;
+
+	return exponent;
+}
+
+static int
+convert(UINTMAX_T value, char *buf, size_t size, int base, int caps)
+{
+	const char *digits = caps ? "0123456789ABCDEF" : "0123456789abcdef";
+	size_t pos = 0;
+
+	/* We return an unterminated buffer with the digits in reverse order. */
+	do {
+		buf[pos++] = digits[value % base];
+		value /= base;
+	} while (value != 0 && pos < size);
+
+	return (int)pos;
+}
+
+static UINTMAX_T
+cast(LDOUBLE value)
+{
+	UINTMAX_T result;
+
+	/*
+	 * We check for ">=" and not for ">" because if UINTMAX_MAX cannot be
+	 * represented exactly as an LDOUBLE value (but is less than LDBL_MAX),
+	 * it may be increased to the nearest higher representable value for the
+	 * comparison (cf. C99: 6.3.1.4, 2).  It might then equal the LDOUBLE
+	 * value although converting the latter to UINTMAX_T would overflow.
+	 */
+	if (value >= UINTMAX_MAX)
+		return UINTMAX_MAX;
+
+	result = (UINTMAX_T)value;
+	/*
+	 * At least on NetBSD/sparc64 3.0.2 and 4.99.30, casting long double to
+	 * an integer type converts e.g. 1.9 to 2 instead of 1 (which violates
+	 * the standard).  Sigh.
+	 */
+	return (result <= value) ? result : result - 1;
+}
+
+static UINTMAX_T
+myround(LDOUBLE value)
+{
+	UINTMAX_T intpart = cast(value);
+
+	return ((value -= intpart) < 0.5) ? intpart : intpart + 1;
+}
+
+static LDOUBLE
+mypow10(int exponent)
+{
+	LDOUBLE result = 1;
+
+	while (exponent > 0) {
+		result *= 10;
+		exponent--;
+	}
+	while (exponent < 0) {
+		result /= 10;
+		exponent++;
+	}
+	return result;
+}
+#endif	/* !HAVE_VSNPRINTF */
+
+#if !HAVE_VASPRINTF
+#if NEED_MYMEMCPY
+void *
+mymemcpy(void *dst, void *src, size_t len)
+{
+	const char *from = src;
+	char *to = dst;
+
+	/* No need for optimization, we use this only to replace va_copy(3). */
+	while (len-- > 0)
+		*to++ = *from++;
+	return dst;
+}
+#endif	/* NEED_MYMEMCPY */
+
+int
+rpl_vasprintf(char **ret, const char *format, va_list ap)
+{
+	size_t size;
+	int len;
+	va_list aq;
+
+	VA_COPY(aq, ap);
+	len = vsnprintf(NULL, 0, format, aq);
+	VA_END_COPY(aq);
+	if (len < 0 || (*ret = malloc(size = len + 1)) == NULL)
+		return -1;
+	return vsnprintf(*ret, size, format, ap);
+}
+#endif	/* !HAVE_VASPRINTF */
+
+#if !HAVE_SNPRINTF
+#if HAVE_STDARG_H
+int
+rpl_snprintf(char *str, size_t size, const char *format, ...)
+#else
+int
+rpl_snprintf(va_alist) va_dcl
+#endif	/* HAVE_STDARG_H */
+{
+#if !HAVE_STDARG_H
+	char *str;
+	size_t size;
+	char *format;
+#endif	/* HAVE_STDARG_H */
+	va_list ap;
+	int len;
+
+	VA_START(ap, format);
+	VA_SHIFT(ap, str, char *);
+	VA_SHIFT(ap, size, size_t);
+	VA_SHIFT(ap, format, const char *);
+	len = vsnprintf(str, size, format, ap);
+	va_end(ap);
+	return len;
+}
+#endif	/* !HAVE_SNPRINTF */
+
+#if !HAVE_ASPRINTF
+#if HAVE_STDARG_H
+int
+rpl_asprintf(char **ret, const char *format, ...)
+#else
+int
+rpl_asprintf(va_alist) va_dcl
+#endif	/* HAVE_STDARG_H */
+{
+#if !HAVE_STDARG_H
+	char **ret;
+	char *format;
+#endif	/* HAVE_STDARG_H */
+	va_list ap;
+	int len;
+
+	VA_START(ap, format);
+	VA_SHIFT(ap, ret, char **);
+	VA_SHIFT(ap, format, const char *);
+	len = vasprintf(ret, format, ap);
+	va_end(ap);
+	return len;
+}
+#endif	/* !HAVE_ASPRINTF */
+#else	/* Dummy declaration to avoid empty translation unit warnings. */
+int main(void);
+#endif	/* !HAVE_SNPRINTF || !HAVE_VSNPRINTF || !HAVE_ASPRINTF || [...] */
+
+
+/* vim: set joinspaces textwidth=80: */
diff --git a/src/gallium/drivers/cell/ppu/cell_context.h b/src/gallium/drivers/cell/ppu/cell_context.h
index bf27289f3f..c568922cbd 100644
--- a/src/gallium/drivers/cell/ppu/cell_context.h
+++ b/src/gallium/drivers/cell/ppu/cell_context.h
@@ -63,6 +63,7 @@ struct cell_context
 
    const struct pipe_blend_state *blend;
    const struct pipe_sampler_state *sampler[PIPE_MAX_SAMPLERS];
+   uint num_samplers;
    const struct pipe_depth_stencil_alpha_state   *depth_stencil;
    const struct pipe_rasterizer_state *rasterizer;
    const struct cell_vertex_shader_state *vs;
@@ -75,6 +76,7 @@ struct cell_context
    struct pipe_poly_stipple poly_stipple;
    struct pipe_scissor_state scissor;
    struct cell_texture *texture[PIPE_MAX_SAMPLERS];
+   uint num_textures;
    struct pipe_viewport_state viewport;
    struct pipe_vertex_buffer vertex_buffer[PIPE_ATTRIB_MAX];
    struct pipe_vertex_element vertex_element[PIPE_ATTRIB_MAX];
diff --git a/src/gallium/drivers/cell/ppu/cell_pipe_state.c b/src/gallium/drivers/cell/ppu/cell_pipe_state.c
index 075e0a0c47..025ed3bbbf 100644
--- a/src/gallium/drivers/cell/ppu/cell_pipe_state.c
+++ b/src/gallium/drivers/cell/ppu/cell_pipe_state.c
@@ -210,15 +210,19 @@ cell_create_sampler_state(struct pipe_context *pipe,
 
 
 static void
-cell_bind_sampler_state(struct pipe_context *pipe,
-                            unsigned unit, void *sampler)
+cell_bind_sampler_states(struct pipe_context *pipe,
+                         unsigned num, void **samplers)
 {
    struct cell_context *cell = cell_context(pipe);
 
    draw_flush(cell->draw);
 
    assert(unit < PIPE_MAX_SAMPLERS);
-   cell->sampler[unit] = (struct pipe_sampler_state *)sampler;
+
+   memcpy(cell->sampler, samplers, num * sizeof(void *));
+   memset(&cell->sampler[num], 0, (PIPE_MAX_SAMPLERS - num) *
+          sizeof(void *));
+   cell->num_samplers = num;
 
    cell->dirty |= CELL_NEW_SAMPLER;
 }
@@ -234,16 +238,24 @@ cell_delete_sampler_state(struct pipe_context *pipe,
 
 
 static void
-cell_set_sampler_texture(struct pipe_context *pipe,
-                         unsigned sampler,
-                         struct pipe_texture *texture)
+cell_set_sampler_textures(struct pipe_context *pipe,
+                          unsigned num, struct pipe_texture **texture)
 {
    struct cell_context *cell = cell_context(pipe);
+   uint i;
+
+   /* Check for no-op */
+   if (num == cell->num_textures &&
+       !memcmp(cell->texture, texture, num * sizeof(struct pipe_texture *)))
+      return;
 
    draw_flush(cell->draw);
 
-   pipe_texture_reference((struct pipe_texture **) &cell->texture[sampler],
-                          texture);
+   for (i = 0; i < PIPE_MAX_SAMPLERS; i++) {
+      struct pipe_texture *tex = i < num ? texture[i] : NULL;
+
+      pipe_texture_reference((struct pipe_texture **) &cell->texture[i], tex);
+   }
 
    cell_update_texture_mapping(cell);
 
@@ -300,10 +312,10 @@ cell_init_state_functions(struct cell_context *cell)
    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.bind_sampler_states = cell_bind_sampler_states;
    cell->pipe.delete_sampler_state = cell_delete_sampler_state;
 
-   cell->pipe.set_sampler_texture = cell_set_sampler_texture;
+   cell->pipe.set_sampler_textures = cell_set_sampler_textures;
 
    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;
diff --git a/src/gallium/drivers/cell/ppu/cell_state_emit.c b/src/gallium/drivers/cell/ppu/cell_state_emit.c
index 49c0d130c5..670eb26bdd 100644
--- a/src/gallium/drivers/cell/ppu/cell_state_emit.c
+++ b/src/gallium/drivers/cell/ppu/cell_state_emit.c
@@ -77,8 +77,10 @@ cell_emit_state(struct cell_context *cell)
    }
 
    if (cell->dirty & CELL_NEW_SAMPLER) {
-      emit_state_cmd(cell, CELL_CMD_STATE_SAMPLER,
-                     cell->sampler[0], sizeof(struct pipe_sampler_state));
+      if (cell->sampler[0]) {
+         emit_state_cmd(cell, CELL_CMD_STATE_SAMPLER,
+                        cell->sampler[0], sizeof(struct pipe_sampler_state));
+      }
    }
 
    if (cell->dirty & CELL_NEW_TEXTURE) {
diff --git a/src/gallium/drivers/cell/ppu/cell_state_shader.c b/src/gallium/drivers/cell/ppu/cell_state_shader.c
index 935501441b..269a5c15ba 100644
--- a/src/gallium/drivers/cell/ppu/cell_state_shader.c
+++ b/src/gallium/drivers/cell/ppu/cell_state_shader.c
@@ -131,7 +131,8 @@ cell_bind_vs_state(struct pipe_context *pipe, void *vs)
 
    cell->vs = (const struct cell_vertex_shader_state *) vs;
 
-   draw_bind_vertex_shader(cell->draw, cell->vs->draw_data);
+   draw_bind_vertex_shader(cell->draw,
+                           (cell->vs ? cell->vs->draw_data : NULL));
 
    cell->dirty |= CELL_NEW_VS;
 }
diff --git a/src/gallium/drivers/cell/ppu/cell_vertex_fetch.c b/src/gallium/drivers/cell/ppu/cell_vertex_fetch.c
index 9cf74bab47..4828a8023b 100644
--- a/src/gallium/drivers/cell/ppu/cell_vertex_fetch.c
+++ b/src/gallium/drivers/cell/ppu/cell_vertex_fetch.c
@@ -33,46 +33,11 @@
 #include "cell_context.h"
 #include "rtasm/rtasm_ppc_spe.h"
 
-typedef uint64_t register_mask;
-
-int allocate_available_register(register_mask *m)
-{
-   unsigned i;
-   for (i = 0; i < 64; i++) {
-      const uint64_t mask = (1ULL << i);
-
-      if ((m[0] & mask) != 0) {
-	 m[0] &= ~mask;
-	 return i;
-      }
-   }
-
-   return -1;
-}
-
-
-int allocate_register(register_mask *m, unsigned reg)
-{
-   assert((m[0] & (1ULL << reg)) != 0);
-
-   m[0] &= ~(1ULL << reg);
-   return reg;
-}
-
-
-void release_register(register_mask *m, unsigned reg)
-{
-   assert((m[0] & (1ULL << reg)) == 0);
-
-   m[0] |= (1ULL << reg);
-}
-
 
 /**
  * Emit a 4x4 matrix transpose operation
  *
  * \param p         Function that the transpose operation is to be appended to
- * \param m         Live register mask
  * \param row0      Register containing row 0 of the source matrix
  * \param row1      Register containing row 1 of the source matrix
  * \param row2      Register containing row 2 of the source matrix
@@ -91,15 +56,15 @@ void release_register(register_mask *m, unsigned reg)
  * This function requires that four temporary are available on entry.
  */
 static void
-emit_matrix_transpose(struct spe_function *p, register_mask *m,
+emit_matrix_transpose(struct spe_function *p,
 		      unsigned row0, unsigned row1, unsigned row2,
 		      unsigned row3, unsigned dest_ptr,
 		      unsigned shuf_ptr, unsigned count)
 {
-   int shuf_hi = allocate_available_register(m);
-   int shuf_lo = allocate_available_register(m);
-   int t1 = allocate_available_register(m);
-   int t2 = allocate_available_register(m);
+   int shuf_hi = spe_allocate_available_register(p);
+   int shuf_lo = spe_allocate_available_register(p);
+   int t1 = spe_allocate_available_register(p);
+   int t2 = spe_allocate_available_register(p);
    int t3;
    int t4;
    int col0;
@@ -169,19 +134,19 @@ emit_matrix_transpose(struct spe_function *p, register_mask *m,
 
    /* Release all of the temporary registers used.
     */
-   release_register(m, col0);
-   release_register(m, col1);
-   release_register(m, col2);
-   release_register(m, col3);
-   release_register(m, shuf_hi);
-   release_register(m, shuf_lo);
-   release_register(m, t2);
-   release_register(m, t4);
+   spe_release_register(p, col0);
+   spe_release_register(p, col1);
+   spe_release_register(p, col2);
+   spe_release_register(p, col3);
+   spe_release_register(p, shuf_hi);
+   spe_release_register(p, shuf_lo);
+   spe_release_register(p, t2);
+   spe_release_register(p, t4);
 }
 
 
 static void
-emit_fetch(struct spe_function *p, register_mask *m,
+emit_fetch(struct spe_function *p,
 	   unsigned in_ptr, unsigned *offset,
 	   unsigned out_ptr, unsigned shuf_ptr,
 	   enum pipe_format format)
@@ -191,11 +156,11 @@ emit_fetch(struct spe_function *p, register_mask *m,
    const unsigned type = pf_type(format);
    const unsigned bytes = pf_size_x(format);
 
-   int v0 = allocate_available_register(m);
-   int v1 = allocate_available_register(m);
-   int v2 = allocate_available_register(m);
-   int v3 = allocate_available_register(m);
-   int tmp = allocate_available_register(m);
+   int v0 = spe_allocate_available_register(p);
+   int v1 = spe_allocate_available_register(p);
+   int v2 = spe_allocate_available_register(p);
+   int v3 = spe_allocate_available_register(p);
+   int tmp = spe_allocate_available_register(p);
    int float_zero = -1;
    int float_one = -1;
    float scale_signed = 0.0;
@@ -260,19 +225,19 @@ emit_fetch(struct spe_function *p, register_mask *m,
 
 
    if (count < 4) {
-      float_one = allocate_available_register(m);
+      float_one = spe_allocate_available_register(p);
       spe_il(p, float_one, 1);
       spe_cuflt(p, float_one, float_one, 0);
       
       if (count < 3) {
-	 float_zero = allocate_available_register(m);
+	 float_zero = spe_allocate_available_register(p);
 	 spe_il(p, float_zero, 0);
       }
    }
 
-   release_register(m, tmp);
+   spe_release_register(p, tmp);
 
-   emit_matrix_transpose(p, m, v0, v1, v2, v3, out_ptr, shuf_ptr, count);
+   emit_matrix_transpose(p, v0, v1, v2, v3, out_ptr, shuf_ptr, count);
 
    switch (count) {
    case 1:
@@ -284,11 +249,11 @@ emit_fetch(struct spe_function *p, register_mask *m,
    }
 
    if (float_zero != -1) {
-      release_register(m, float_zero);
+      spe_release_register(p, float_zero);
    }
 
    if (float_one != -1) {
-      release_register(m, float_one);
+      spe_release_register(p, float_one);
    }
 }
 
@@ -297,7 +262,6 @@ void cell_update_vertex_fetch(struct draw_context *draw)
 {
    struct cell_context *const cell =
        (struct cell_context *) draw->driver_private;
-   register_mask m = ~0;
    struct spe_function *p = &cell->attrib_fetch;
    unsigned function_index[PIPE_ATTRIB_MAX];
    unsigned unique_attr_formats;
@@ -338,18 +302,11 @@ void cell_update_vertex_fetch(struct draw_context *draw)
    spe_init_func(p, 136 * unique_attr_formats);
 
 
-   /* Registers 0, 1, and 2 are reserved by the ABI.
-    */
-   allocate_register(&m, 0);
-   allocate_register(&m, 1);
-   allocate_register(&m, 2);
-
-
    /* Allocate registers for the function's input parameters.
     */
-   out_ptr = allocate_register(&m, 3);
-   in_ptr = allocate_register(&m, 4);
-   shuf_ptr = allocate_register(&m, 5);
+   out_ptr = spe_allocate_register(p, 3);
+   in_ptr = spe_allocate_register(p, 4);
+   shuf_ptr = spe_allocate_register(p, 5);
 
 
    /* Generate code for the individual attribute fetch functions.
@@ -362,7 +319,7 @@ void cell_update_vertex_fetch(struct draw_context *draw)
 						     - (void *) p->store);
 
 	 offset = 0;
-	 emit_fetch(p, & m, in_ptr, &offset, out_ptr, shuf_ptr,
+	 emit_fetch(p, in_ptr, &offset, out_ptr, shuf_ptr,
 		    draw->vertex_element[i].src_format);
 	 spe_bi(p, 0, 0, 0);
 
diff --git a/src/gallium/drivers/failover/fo_context.h b/src/gallium/drivers/failover/fo_context.h
index 1dc87291c9..8f3ad3ee79 100644
--- a/src/gallium/drivers/failover/fo_context.h
+++ b/src/gallium/drivers/failover/fo_context.h
@@ -87,12 +87,15 @@ struct failover_context {
    struct pipe_vertex_buffer vertex_buffer[PIPE_ATTRIB_MAX];
    struct pipe_vertex_element vertex_element[PIPE_ATTRIB_MAX];
 
+   void *sw_sampler_state[PIPE_MAX_SAMPLERS];
+   void *hw_sampler_state[PIPE_MAX_SAMPLERS];
+
    unsigned dirty;
-   unsigned dirty_sampler;
-   unsigned dirty_texture;
    unsigned dirty_vertex_buffer;
    unsigned dirty_vertex_element;
 
+   unsigned num_samplers;
+   unsigned num_textures;
 
    unsigned mode;
    struct pipe_context *hw;
diff --git a/src/gallium/drivers/failover/fo_state.c b/src/gallium/drivers/failover/fo_state.c
index 0fc5568da1..11eec2714e 100644
--- a/src/gallium/drivers/failover/fo_state.c
+++ b/src/gallium/drivers/failover/fo_state.c
@@ -28,6 +28,8 @@
 /* Authors:  Keith Whitwell <keith@tungstengraphics.com>
  */
 
+#include "pipe/p_inlines.h"
+
 #include "fo_context.h"
 
 
@@ -322,18 +324,27 @@ failover_create_sampler_state(struct pipe_context *pipe,
 }
 
 static void
-failover_bind_sampler_state(struct pipe_context *pipe,
-			   unsigned unit, void *sampler)
+failover_bind_sampler_states(struct pipe_context *pipe,
+                             unsigned num, void **sampler)
 {
    struct failover_context *failover = failover_context(pipe);
    struct fo_state *state = (struct fo_state*)sampler;
-   failover->sampler[unit] = state;
+   uint i;
+   assert(num <= PIPE_MAX_SAMPLERS);
+   /* Check for no-op */
+   if (num == failover->num_samplers &&
+       !memcmp(failover->sampler, sampler, num * sizeof(void *)))
+      return;
+   for (i = 0; i < PIPE_MAX_SAMPLERS; i++) {
+      failover->sw_sampler_state[i] = i < num ? state[i].sw_state : NULL;
+      failover->hw_sampler_state[i] = i < num ? state[i].hw_state : NULL;
+   }
    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);
+   failover->num_samplers = num;
+   failover->sw->bind_sampler_states(failover->sw, num,
+                                     failover->sw_sampler_state);
+   failover->hw->bind_sampler_states(failover->hw, num,
+                                     failover->hw_sampler_state);
 }
 
 static void
@@ -351,17 +362,29 @@ failover_delete_sampler_state(struct pipe_context *pipe, void *sampler)
 
 
 static void
-failover_set_sampler_texture(struct pipe_context *pipe,
-			     unsigned unit,
-			     struct pipe_texture *texture)
+failover_set_sampler_textures(struct pipe_context *pipe,
+                              unsigned num,
+                              struct pipe_texture **texture)
 {
    struct failover_context *failover = failover_context(pipe);
-
-   failover->texture[unit] = texture;
+   uint i;
+
+   assert(num <= PIPE_MAX_SAMPLERS);
+
+   /* Check for no-op */
+   if (num == failover->num_textures &&
+       !memcmp(failover->texture, texture, num * sizeof(struct pipe_texture *)))
+      return;
+   for (i = 0; i < num; i++)
+      pipe_texture_reference((struct pipe_texture **) &failover->texture[i],
+                             texture[i]);
+   for (i = num; i < failover->num_textures; i++)
+      pipe_texture_reference((struct pipe_texture **) &failover->texture[i],
+                             NULL);
    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 );
+   failover->num_textures = num;
+   failover->sw->set_sampler_textures( failover->sw, num, texture );
+   failover->hw->set_sampler_textures( failover->hw, num, texture );
 }
 
 
@@ -429,7 +452,7 @@ failover_init_state_functions( struct failover_context *failover )
    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.bind_sampler_states  = failover_bind_sampler_states;
    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;
@@ -449,7 +472,7 @@ failover_init_state_functions( struct failover_context *failover )
    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_sampler_textures = failover_set_sampler_textures;
    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;
diff --git a/src/gallium/drivers/failover/fo_state_emit.c b/src/gallium/drivers/failover/fo_state_emit.c
index c663dd4947..3de931e04e 100644
--- a/src/gallium/drivers/failover/fo_state_emit.c
+++ b/src/gallium/drivers/failover/fo_state_emit.c
@@ -94,21 +94,13 @@ failover_state_emit( struct failover_context *failover )
       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 );
-	 }
-      }
+      failover->sw->bind_sampler_states( failover->sw, failover->num_samplers,
+                                         failover->sw_sampler_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] );
-	 }
-      }
+      failover->sw->set_sampler_textures( failover->sw, failover->num_textures, 
+                                          failover->texture );
    }
 
    if (failover->dirty & FO_NEW_VERTEX_BUFFER) {
@@ -132,6 +124,4 @@ failover_state_emit( struct failover_context *failover )
    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/i915simple/i915_context.h b/src/gallium/drivers/i915simple/i915_context.h
index 6401112f83..746f18ba38 100644
--- a/src/gallium/drivers/i915simple/i915_context.h
+++ b/src/gallium/drivers/i915simple/i915_context.h
@@ -1,336 +1,339 @@
- /**************************************************************************
- * 
- * 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 "draw/draw_vertex.h"
-
-#include "tgsi/util/tgsi_scan.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
-
-
-/** See constant_flags[] below */
-#define I915_CONSTFLAG_USER 0x1f
-
-
-/**
- * Subclass of pipe_shader_state
- */
-struct i915_fragment_shader
-{
-   struct pipe_shader_state state;
-
-   struct tgsi_shader_info info;
-
-   uint *program;
-   uint program_len;
-
-   /**
-    * constants introduced during translation.
-    * These are placed at the end of the constant buffer and grow toward
-    * the beginning (eg: slot 31, 30 29, ...)
-    * User-provided constants start at 0.
-    * This allows both types of constants to co-exist (until there's too many)
-    * and doesn't require regenerating/changing the fragment program to
-    * shuffle constants around.
-    */
-   uint num_constants;
-   float constants[I915_MAX_CONSTANT][4];
-
-   /**
-    * Status of each constant
-    * if I915_CONSTFLAG_PARAM, the value must be taken from the corresponding
-    * slot of the user's constant buffer. (set by pipe->set_constant_buffer())
-    * Else, the bitmask indicates which components are occupied by immediates.
-    */
-   ubyte constant_flags[I915_MAX_CONSTANT];
-};
-
-
-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];
-
-   /* 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;
-
-   struct i915_fragment_shader *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;
-};
-
-/* 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
-#define I915_NEW_VS            0x4000
-
-
-/* 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
+ /**************************************************************************

+ * 

+ * 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 "draw/draw_vertex.h"

+

+#include "tgsi/util/tgsi_scan.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

+

+

+/** See constant_flags[] below */

+#define I915_CONSTFLAG_USER 0x1f

+

+

+/**

+ * Subclass of pipe_shader_state

+ */

+struct i915_fragment_shader

+{

+   struct pipe_shader_state state;

+

+   struct tgsi_shader_info info;

+

+   uint *program;

+   uint program_len;

+

+   /**

+    * constants introduced during translation.

+    * These are placed at the end of the constant buffer and grow toward

+    * the beginning (eg: slot 31, 30 29, ...)

+    * User-provided constants start at 0.

+    * This allows both types of constants to co-exist (until there's too many)

+    * and doesn't require regenerating/changing the fragment program to

+    * shuffle constants around.

+    */

+   uint num_constants;

+   float constants[I915_MAX_CONSTANT][4];

+

+   /**

+    * Status of each constant

+    * if I915_CONSTFLAG_PARAM, the value must be taken from the corresponding

+    * slot of the user's constant buffer. (set by pipe->set_constant_buffer())

+    * Else, the bitmask indicates which components are occupied by immediates.

+    */

+   ubyte constant_flags[I915_MAX_CONSTANT];

+};

+

+

+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];

+

+   /* 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;

+

+   struct i915_fragment_shader *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 num_samplers;

+   unsigned num_textures;

+

+   unsigned *batch_start;

+

+   /** Vertex buffer */

+   struct pipe_buffer *vbo;

+

+   struct i915_state current;

+   unsigned hardware_dirty;

+   

+   unsigned debug;

+};

+

+/* 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

+#define I915_NEW_VS            0x4000

+

+

+/* 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_fpc_translate.c b/src/gallium/drivers/i915simple/i915_fpc_translate.c
index c2d9a93c6c..7b4fca5db1 100644
--- a/src/gallium/drivers/i915simple/i915_fpc_translate.c
+++ b/src/gallium/drivers/i915simple/i915_fpc_translate.c
@@ -872,19 +872,17 @@ i915_translate_instruction(struct i915_fp_compile *p,
       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);
-      }
+      emit_tex(p, inst, T0_TEXLD);
       break;
 
    case TGSI_OPCODE_TXB:
       emit_tex(p, inst, T0_TEXLDB);
       break;
 
+   case TGSI_OPCODE_TXP:
+      emit_tex(p, inst, T0_TEXLDP);
+      break;
+
    case TGSI_OPCODE_XPD:
       /* Cross product:
        *      result.x = src0.y * src1.z - src0.z * src1.y;
diff --git a/src/gallium/drivers/i915simple/i915_state.c b/src/gallium/drivers/i915simple/i915_state.c
index 27af46bea0..503c092400 100644
--- a/src/gallium/drivers/i915simple/i915_state.c
+++ b/src/gallium/drivers/i915simple/i915_state.c
@@ -71,6 +71,8 @@ static unsigned translate_img_filter( unsigned filter )
       return FILTER_NEAREST;
    case PIPE_TEX_FILTER_LINEAR:
       return FILTER_LINEAR;
+   case PIPE_TEX_FILTER_ANISO:
+      return FILTER_ANISOTROPIC;
    default:
       assert(0);
       return FILTER_NEAREST;
@@ -84,7 +86,7 @@ static unsigned translate_mip_filter( unsigned filter )
       return MIPFILTER_NONE;
    case PIPE_TEX_MIPFILTER_NEAREST:
       return MIPFILTER_NEAREST;
-   case PIPE_TEX_FILTER_LINEAR:
+   case PIPE_TEX_MIPFILTER_LINEAR:
       return MIPFILTER_LINEAR;
    default:
       assert(0);
@@ -211,16 +213,11 @@ i915_create_sampler_state(struct pipe_context *pipe,
    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 );
+   minFilt = translate_img_filter( sampler->min_img_filter );
+   magFilt = translate_img_filter( sampler->mag_img_filter );
+   
+   if (sampler->max_anisotropy > 2.0) {
+      cso->state[0] |= SS2_MAX_ANISO_4;
    }
 
    {
@@ -269,13 +266,25 @@ i915_create_sampler_state(struct pipe_context *pipe,
    return cso;
 }
 
-static void i915_bind_sampler_state(struct pipe_context *pipe,
-                                    unsigned unit, void *sampler)
+static void i915_bind_sampler_states(struct pipe_context *pipe,
+                                     unsigned num, void **sampler)
 {
    struct i915_context *i915 = i915_context(pipe);
+   unsigned i;
+
+   assert(num <= PIPE_MAX_SAMPLERS);
+
+   /* Check for no-op */
+   if (num == i915->num_samplers &&
+       !memcmp(i915->sampler, sampler, num * sizeof(void *)))
+      return;
 
-   assert(unit < PIPE_MAX_SAMPLERS);
-   i915->sampler[unit] = (const struct i915_sampler_state*)sampler;
+   for (i = 0; i < num; ++i)
+      i915->sampler[i] = sampler[i];
+   for (i = num; i < PIPE_MAX_SAMPLERS; ++i)
+      i915->sampler[i] = NULL;
+
+   i915->num_samplers = num;
 
    i915->dirty |= I915_NEW_SAMPLER;
 }
@@ -526,14 +535,29 @@ static void i915_set_constant_buffer(struct pipe_context *pipe,
 }
 
 
-static void i915_set_sampler_texture(struct pipe_context *pipe,
-				     unsigned sampler,
-				     struct pipe_texture *texture)
+static void i915_set_sampler_textures(struct pipe_context *pipe,
+                                      unsigned num,
+                                      struct pipe_texture **texture)
 {
    struct i915_context *i915 = i915_context(pipe);
+   uint i;
+
+   assert(num <= PIPE_MAX_SAMPLERS);
+
+   /* Check for no-op */
+   if (num == i915->num_textures &&
+       !memcmp(i915->texture, texture, num * sizeof(struct pipe_texture *)))
+      return;
+
+   for (i = 0; i < num; i++)
+      pipe_texture_reference((struct pipe_texture **) &i915->texture[i],
+                             texture[i]);
+
+   for (i = num; i < i915->num_textures; i++)
+      pipe_texture_reference((struct pipe_texture **) &i915->texture[i],
+                             NULL);
 
-   pipe_texture_reference((struct pipe_texture **) &i915->texture[sampler],
-                          texture);
+   i915->num_textures = num;
 
    i915->dirty |= I915_NEW_TEXTURE;
 }
@@ -644,22 +668,23 @@ i915_create_rasterizer_state(struct pipe_context *pipe,
 }
 
 static void i915_bind_rasterizer_state( struct pipe_context *pipe,
-                                        void *setup )
+                                        void *raster )
 {
    struct i915_context *i915 = i915_context(pipe);
 
-   i915->rasterizer = (struct i915_rasterizer_state *)setup;
+   i915->rasterizer = (struct i915_rasterizer_state *)raster;
 
    /* pass-through to draw module */
-   draw_set_rasterizer_state(i915->draw, i915->rasterizer->templ);
+   draw_set_rasterizer_state(i915->draw,
+                          (i915->rasterizer ? i915->rasterizer->templ : NULL));
 
    i915->dirty |= I915_NEW_RASTERIZER;
 }
 
 static void i915_delete_rasterizer_state(struct pipe_context *pipe,
-                                         void *setup)
+                                         void *raster)
 {
-   FREE(setup);
+   FREE(raster);
 }
 
 static void i915_set_vertex_buffer( struct pipe_context *pipe,
@@ -691,7 +716,7 @@ i915_init_state_functions( struct i915_context *i915 )
    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.bind_sampler_states = i915_bind_sampler_states;
    i915->pipe.delete_sampler_state = i915_delete_sampler_state;
 
    i915->pipe.create_depth_stencil_alpha_state = i915_create_depth_stencil_state;
@@ -715,7 +740,7 @@ i915_init_state_functions( struct i915_context *i915 )
 
    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_sampler_textures = i915_set_sampler_textures;
    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_emit.c b/src/gallium/drivers/i915simple/i915_state_emit.c
index 6bbaac4e34..a7498d22b7 100644
--- a/src/gallium/drivers/i915simple/i915_state_emit.c
+++ b/src/gallium/drivers/i915simple/i915_state_emit.c
@@ -267,12 +267,6 @@ i915_emit_hardware_state(struct i915_context *i915 )
       /* 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;
diff --git a/src/gallium/drivers/i915simple/i915_state_sampler.c b/src/gallium/drivers/i915simple/i915_state_sampler.c
index 9c1a5bbbd6..9dbb1b1b23 100644
--- a/src/gallium/drivers/i915simple/i915_state_sampler.c
+++ b/src/gallium/drivers/i915simple/i915_state_sampler.c
@@ -106,7 +106,8 @@ void i915_update_samplers( struct i915_context *i915 )
    i915->current.sampler_enable_nr = 0;
    i915->current.sampler_enable_flags = 0x0;
 
-   for (unit = 0; unit < I915_TEX_UNITS; unit++) {
+   for (unit = 0; unit < i915->num_textures && unit < i915->num_samplers;
+        unit++) {
       /* determine unit enable/disable by looking for a bound texture */
       /* could also examine the fragment program? */
       if (i915->texture[unit]) {
@@ -219,7 +220,8 @@ i915_update_textures(struct i915_context *i915)
 {
    uint unit;
 
-   for (unit = 0; unit < I915_TEX_UNITS; unit++) {
+   for (unit = 0; unit < i915->num_textures && unit < i915->num_samplers;
+        unit++) {
       /* determine unit enable/disable by looking for a bound texture */
       /* could also examine the fragment program? */
       if (i915->texture[unit]) {
diff --git a/src/gallium/drivers/i965simple/brw_context.h b/src/gallium/drivers/i965simple/brw_context.h
index 4da3a8cffc..b83a13c3b6 100644
--- a/src/gallium/drivers/i965simple/brw_context.h
+++ b/src/gallium/drivers/i965simple/brw_context.h
@@ -1,681 +1,684 @@
-/*
- 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 "tgsi/util/tgsi_scan.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_vertex_program {
-   struct pipe_shader_state program;
-   struct tgsi_shader_info info;
-   int id;
-};
-
-
-struct brw_fragment_program {
-   struct pipe_shader_state program;
-   struct tgsi_shader_info info;
-   
-   boolean UsesDepth; /* XXX add this to tgsi_shader_info? */
-   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:
-       */
-      const 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
-
+/*

+ 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 "tgsi/util/tgsi_scan.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_vertex_program {

+   struct pipe_shader_state program;

+   struct tgsi_shader_info info;

+   int id;

+};

+

+

+struct brw_fragment_program {

+   struct pipe_shader_state program;

+   struct tgsi_shader_info info;

+   

+   boolean UsesDepth; /* XXX add this to tgsi_shader_info? */

+   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;

+

+   unsigned num_samplers;

+   unsigned num_textures;

+

+   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:

+       */

+      const 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_state.c b/src/gallium/drivers/i965simple/brw_state.c
index 6744a8aa4f..f5efe9fc06 100644
--- a/src/gallium/drivers/i965simple/brw_state.c
+++ b/src/gallium/drivers/i965simple/brw_state.c
@@ -1,434 +1,462 @@
-/**************************************************************************
- *
- * 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_inlines.h"
-#include "pipe/p_shader_tokens.h"
-#include "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++;
-
-   tgsi_scan_shader(shader->tokens, &brw_fp->info);
-
-#if 0
-   brw_shader_info(shader->tokens,
-		   &brw_fp->info2);
-#endif
-
-   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++;
-
-   tgsi_scan_shader(shader->tokens, &brw_vp->info);
-
-#if 0
-   brw_shader_info(shader->tokens,
-		   &brw_vp->info2);
-#endif
-   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);
-
-   pipe_texture_reference((struct pipe_texture **) &brw->attribs.Texture[unit],
-                          texture);
-
-   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;
-}
+/**************************************************************************

+ *

+ * 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_inlines.h"

+#include "pipe/p_shader_tokens.h"

+#include "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_states(struct pipe_context *pipe,

+                                    unsigned num, void **sampler)

+{

+   struct brw_context *brw = brw_context(pipe);

+

+   assert(num <= PIPE_MAX_SAMPLERS);

+

+   /* Check for no-op */

+   if (num == brw->num_samplers &&

+       !memcmp(brw->attribs.Samplers, sampler, num * sizeof(void *)))

+      return;

+

+   memcpy(brw->attribs.Samplers, sampler, num * sizeof(void *));

+   memset(&brw->attribs.Samplers[num], 0, (PIPE_MAX_SAMPLERS - num) *

+          sizeof(void *));

+

+   brw->num_samplers = num;

+

+   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++;

+

+   tgsi_scan_shader(shader->tokens, &brw_fp->info);

+

+#if 0

+   brw_shader_info(shader->tokens,

+		   &brw_fp->info2);

+#endif

+

+   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++;

+

+   tgsi_scan_shader(shader->tokens, &brw_vp->info);

+

+#if 0

+   brw_shader_info(shader->tokens,

+		   &brw_vp->info2);

+#endif

+   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_textures(struct pipe_context *pipe,

+                                     unsigned num,

+                                     struct pipe_texture **texture)

+{

+   struct brw_context *brw = brw_context(pipe);

+   uint i;

+

+   assert(num <= PIPE_MAX_SAMPLERS);

+

+   /* Check for no-op */

+   if (num == brw->num_textures &&

+       !memcmp(brw->attribs.Texture, texture, num *

+               sizeof(struct pipe_texture *)))

+      return;

+

+   for (i = 0; i < num; i++)

+      pipe_texture_reference((struct pipe_texture **) &brw->attribs.Texture[i],

+                             texture[i]);

+

+   for (i = num; i < brw->num_textures; i++)

+      pipe_texture_reference((struct pipe_texture **) &brw->attribs.Texture[i],

+                             NULL);

+

+   brw->num_textures = num;

+

+   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_states = brw_bind_sampler_states;

+   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_textures = brw_set_sampler_textures;

+   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_wm_glsl.c b/src/gallium/drivers/i965simple/brw_wm_glsl.c
index 44f946ea74..5c90583824 100644
--- a/src/gallium/drivers/i965simple/brw_wm_glsl.c
+++ b/src/gallium/drivers/i965simple/brw_wm_glsl.c
@@ -124,10 +124,6 @@ static struct brw_reg get_src_reg(struct brw_wm_compile *c,
     */
    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);
diff --git a/src/gallium/drivers/i965simple/brw_wm_sampler_state.c b/src/gallium/drivers/i965simple/brw_wm_sampler_state.c
index de42ffc5b1..b9eaee56ee 100644
--- a/src/gallium/drivers/i965simple/brw_wm_sampler_state.c
+++ b/src/gallium/drivers/i965simple/brw_wm_sampler_state.c
@@ -136,6 +136,9 @@ static void brw_update_sampler_state( const struct pipe_sampler_state *pipe_samp
    case PIPE_TEX_FILTER_LINEAR:
       sampler->ss0.min_filter = BRW_MAPFILTER_LINEAR;
       break;
+   case PIPE_TEX_FILTER_ANISO:
+      sampler->ss0.min_filter = BRW_MAPFILTER_ANISOTROPIC;
+      break;
    default:
       break;
    }
@@ -155,26 +158,23 @@ static void brw_update_sampler_state( const struct pipe_sampler_state *pipe_samp
    }
    /* 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);
-      }
+   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;
+   case PIPE_TEX_FILTER_ANISO:
+      sampler->ss0.mag_filter = BRW_MAPFILTER_LINEAR;
+      break;
+   default:
+      break;
    }
-   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;
-      }
+
+   if (pipe_sampler->max_anisotropy > 2.0) {
+      sampler->ss3.max_aniso = MAX2((pipe_sampler->max_anisotropy - 2) / 2,
+                                    BRW_ANISORATIO_16);
    }
 
    sampler->ss1.s_wrap_mode = translate_wrap_mode(pipe_sampler->wrap_s);
@@ -235,7 +235,8 @@ static void upload_wm_samplers(struct brw_context *brw)
    unsigned sampler_count = 0;
 
    /* BRW_NEW_SAMPLER */
-   for (unit = 0; unit < BRW_MAX_TEX_UNIT; unit++) {
+   for (unit = 0; unit < brw->num_textures && unit < brw->num_samplers;
+        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];
diff --git a/src/gallium/drivers/i965simple/brw_wm_surface_state.c b/src/gallium/drivers/i965simple/brw_wm_surface_state.c
index d16d919bce..853c743ccf 100644
--- a/src/gallium/drivers/i965simple/brw_wm_surface_state.c
+++ b/src/gallium/drivers/i965simple/brw_wm_surface_state.c
@@ -237,7 +237,7 @@ static void upload_wm_surfaces(struct brw_context *brw )
 
    /* BRW_NEW_TEXTURE
     */
-   for (i = 0; i < BRW_MAX_TEX_UNIT; i++) {
+   for (i = 0; i < brw->num_textures && i < brw->num_samplers; i++) {
       const struct brw_texture *texUnit = brw->attribs.Texture[i];
 
       if (texUnit &&
diff --git a/src/gallium/drivers/softpipe/sp_context.c b/src/gallium/drivers/softpipe/sp_context.c
index fa16ed94e8..316ae552b8 100644
--- a/src/gallium/drivers/softpipe/sp_context.c
+++ b/src/gallium/drivers/softpipe/sp_context.c
@@ -1,243 +1,243 @@
-/**************************************************************************
- * 
- * 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 "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"
-
-
-
-/**
- * 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 );
-}
-
-
-struct pipe_context *
-softpipe_create( struct pipe_screen *screen,
-                 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.screen = screen;
-   softpipe->pipe.destroy = softpipe_destroy;
-
-   /* 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 );
-   softpipe_init_texture_funcs( softpipe );
-
-   /*
-    * 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);
-   }
-
-   /* plug in AA line/point stages */
-   draw_install_aaline_stage(softpipe->draw, &softpipe->pipe);
-   draw_install_aapoint_stage(softpipe->draw, &softpipe->pipe);
-
-#if USE_DRAW_STAGE_PSTIPPLE
-   /* Do polygon stipple w/ texture map + frag prog? */
-   draw_install_pstipple_stage(softpipe->draw, &softpipe->pipe);
-#endif
-
-   sp_init_surface_functions(softpipe);
-
-   return &softpipe->pipe;
-}
+/**************************************************************************

+ * 

+ * 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 "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"

+

+

+

+/**

+ * 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 );

+}

+

+

+struct pipe_context *

+softpipe_create( struct pipe_screen *screen,

+                 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.screen = screen;

+   softpipe->pipe.destroy = softpipe_destroy;

+

+   /* 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_states  = softpipe_bind_sampler_states;

+   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_textures = softpipe_set_sampler_textures;

+   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 );

+   softpipe_init_texture_funcs( softpipe );

+

+   /*

+    * 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);

+   }

+

+   /* plug in AA line/point stages */

+   draw_install_aaline_stage(softpipe->draw, &softpipe->pipe);

+   draw_install_aapoint_stage(softpipe->draw, &softpipe->pipe);

+

+#if USE_DRAW_STAGE_PSTIPPLE

+   /* Do polygon stipple w/ texture map + frag prog? */

+   draw_install_pstipple_stage(softpipe->draw, &softpipe->pipe);

+#endif

+

+   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
index feeafc7084..19e6cfaf02 100644
--- a/src/gallium/drivers/softpipe/sp_context.h
+++ b/src/gallium/drivers/softpipe/sp_context.h
@@ -81,6 +81,9 @@ struct softpipe_context {
    struct pipe_vertex_element vertex_element[PIPE_ATTRIB_MAX];
    unsigned dirty;
 
+   unsigned num_samplers;
+   unsigned num_textures;
+
    /* Counter for occlusion queries.  Note this supports overlapping
     * queries.
     */
diff --git a/src/gallium/drivers/softpipe/sp_quad_alpha_test.c b/src/gallium/drivers/softpipe/sp_quad_alpha_test.c
index 4ffeac35e1..318916fe30 100644
--- a/src/gallium/drivers/softpipe/sp_quad_alpha_test.c
+++ b/src/gallium/drivers/softpipe/sp_quad_alpha_test.c
@@ -72,7 +72,7 @@ alpha_test_quad(struct quad_stage *qs, struct quad_header *quad)
       passMask = MASK_ALL;
       break;
    default:
-      abort();
+      assert(0);
    }
 
    quad->mask &= passMask;
diff --git a/src/gallium/drivers/softpipe/sp_quad_blend.c b/src/gallium/drivers/softpipe/sp_quad_blend.c
index 17f3ecd0b8..0b79cfa1ed 100644
--- a/src/gallium/drivers/softpipe/sp_quad_blend.c
+++ b/src/gallium/drivers/softpipe/sp_quad_blend.c
@@ -392,7 +392,7 @@ blend_quad(struct quad_stage *qs, struct quad_header *quad)
       assert(0); /* to do */
       break;
    default:
-      abort();
+      assert(0);
    }
    
    /*
@@ -464,7 +464,7 @@ blend_quad(struct quad_stage *qs, struct quad_header *quad)
       }
       break;
    default:
-      abort();
+      assert(0);
    }
    
    
@@ -716,7 +716,7 @@ blend_quad(struct quad_stage *qs, struct quad_header *quad)
       VEC4_MAX(quadColor[3], source[3], dest[3]); /* A */
       break;
    default:
-      abort();
+      assert(0);
    }
    
    /* pass blended quad to next stage */
diff --git a/src/gallium/drivers/softpipe/sp_quad_depth_test.c b/src/gallium/drivers/softpipe/sp_quad_depth_test.c
index a9a0754f27..a1859f9883 100644
--- a/src/gallium/drivers/softpipe/sp_quad_depth_test.c
+++ b/src/gallium/drivers/softpipe/sp_quad_depth_test.c
@@ -185,7 +185,7 @@ sp_depth_test_quad(struct quad_stage *qs, struct quad_header *quad)
       zmask = MASK_ALL;
       break;
    default:
-      abort();
+      assert(0);
    }
 
    quad->mask &= zmask;
diff --git a/src/gallium/drivers/softpipe/sp_quad_fs.c b/src/gallium/drivers/softpipe/sp_quad_fs.c
index 1fbb2e38c4..9198198db3 100644
--- a/src/gallium/drivers/softpipe/sp_quad_fs.c
+++ b/src/gallium/drivers/softpipe/sp_quad_fs.c
@@ -1,208 +1,209 @@
-/**************************************************************************
- * 
- * 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 "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;
-};
-
-
-/** cast wrapper */
-static INLINE struct quad_shade_stage *
-quad_shade_stage(struct quad_stage *qs)
-{
-   return (struct quad_shade_stage *) qs;
-}
-
-
-
-/**
- * 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;
-
-   /* run shader */
-   quad->mask &= softpipe->fs->run( softpipe->fs, 
-				    &qss->machine,
-				    quad );
-
-   /* store result color */
-   if (qss->colorOutSlot >= 0) {
-      /* XXX need to handle multiple color outputs someday */
-      assert(qss->stage.softpipe->fs->info.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 );
-   }
-}
-
-/**
- * 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];
-   }
-
-   /* find output slots for depth, color */
-   qss->colorOutSlot = -1;
-   qss->depthOutSlot = -1;
-   for (i = 0; i < qss->stage.softpipe->fs->info.num_outputs; i++) {
-      switch (qss->stage.softpipe->fs->info.output_semantic_name[i]) {
-      case TGSI_SEMANTIC_POSITION:
-         qss->depthOutSlot = i;
-         break;
-      case TGSI_SEMANTIC_COLOR:
-         qss->colorOutSlot = i;
-         break;
-      }
-   }
-   
-   softpipe->fs->prepare( softpipe->fs, 
-			  &qss->machine,
-			  qss->samplers );
-
-   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;
-   qss->stage.run = shade_quad;
-   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];
-   }
-
-   tgsi_exec_machine_init( &qss->machine );
-
-   return &qss->stage;
-}
+/**************************************************************************

+ * 

+ * 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 "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;

+};

+

+

+/** cast wrapper */

+static INLINE struct quad_shade_stage *

+quad_shade_stage(struct quad_stage *qs)

+{

+   return (struct quad_shade_stage *) qs;

+}

+

+

+

+/**

+ * 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;

+

+   /* run shader */

+   quad->mask &= softpipe->fs->run( softpipe->fs, 

+				    &qss->machine,

+				    quad );

+

+   /* store result color */

+   if (qss->colorOutSlot >= 0) {

+      /* XXX need to handle multiple color outputs someday */

+      assert(qss->stage.softpipe->fs->info.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 );

+   }

+}

+

+/**

+ * 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;

+   unsigned num = MAX2(softpipe->num_textures, softpipe->num_samplers);

+

+   /* set TGSI sampler state that varies */

+   for (i = 0; i < num; i++) {

+      qss->samplers[i].state = softpipe->sampler[i];

+      qss->samplers[i].texture = softpipe->texture[i];

+   }

+

+   /* find output slots for depth, color */

+   qss->colorOutSlot = -1;

+   qss->depthOutSlot = -1;

+   for (i = 0; i < qss->stage.softpipe->fs->info.num_outputs; i++) {

+      switch (qss->stage.softpipe->fs->info.output_semantic_name[i]) {

+      case TGSI_SEMANTIC_POSITION:

+         qss->depthOutSlot = i;

+         break;

+      case TGSI_SEMANTIC_COLOR:

+         qss->colorOutSlot = i;

+         break;

+      }

+   }

+   

+   softpipe->fs->prepare( softpipe->fs, 

+			  &qss->machine,

+			  qss->samplers );

+

+   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;

+   qss->stage.run = shade_quad;

+   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];

+   }

+

+   tgsi_exec_machine_init( &qss->machine );

+

+   return &qss->stage;

+}

diff --git a/src/gallium/drivers/softpipe/sp_state.h b/src/gallium/drivers/softpipe/sp_state.h
index 3943d4ed2b..45d159143f 100644
--- a/src/gallium/drivers/softpipe/sp_state.h
+++ b/src/gallium/drivers/softpipe/sp_state.h
@@ -1,195 +1,195 @@
-/**************************************************************************
- * 
- * 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 "tgsi/util/tgsi_scan.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
-
-
-struct tgsi_sampler;
-struct tgsi_exec_machine;
-
-
-/** Subclass of pipe_shader_state (though it doesn't really need to be).
- *
- * This is starting to look an awful lot like a quad pipeline stage...
- */
-struct sp_fragment_shader {
-   struct pipe_shader_state shader;
-
-   struct tgsi_shader_info info;
-
-   void (*prepare)( const struct sp_fragment_shader *shader,
-		    struct tgsi_exec_machine *machine,
-		    struct tgsi_sampler *samplers);
-
-   /* Run the shader - this interface will get cleaned up in the
-    * future:
-    */
-   unsigned (*run)( const struct sp_fragment_shader *shader,
-		    struct tgsi_exec_machine *machine,
-		    struct quad_header *quad );
-
-
-   void (*delete)( struct sp_fragment_shader * );
-};
-
-struct vertex_info;
-
-/** Subclass of pipe_shader_state */
-struct sp_vertex_shader {
-   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
+/**************************************************************************

+ * 

+ * 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 "tgsi/util/tgsi_scan.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

+

+

+struct tgsi_sampler;

+struct tgsi_exec_machine;

+

+

+/** Subclass of pipe_shader_state (though it doesn't really need to be).

+ *

+ * This is starting to look an awful lot like a quad pipeline stage...

+ */

+struct sp_fragment_shader {

+   struct pipe_shader_state shader;

+

+   struct tgsi_shader_info info;

+

+   void (*prepare)( const struct sp_fragment_shader *shader,

+		    struct tgsi_exec_machine *machine,

+		    struct tgsi_sampler *samplers);

+

+   /* Run the shader - this interface will get cleaned up in the

+    * future:

+    */

+   unsigned (*run)( const struct sp_fragment_shader *shader,

+		    struct tgsi_exec_machine *machine,

+		    struct quad_header *quad );

+

+

+   void (*delete)( struct sp_fragment_shader * );

+};

+

+struct vertex_info;

+

+/** Subclass of pipe_shader_state */

+struct sp_vertex_shader {

+   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_states(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_textures( struct pipe_context *,

+                                    unsigned num,

+                                    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_fs.c b/src/gallium/drivers/softpipe/sp_state_fs.c
index eb641ed321..4eefd1d61f 100644
--- a/src/gallium/drivers/softpipe/sp_state_fs.c
+++ b/src/gallium/drivers/softpipe/sp_state_fs.c
@@ -120,7 +120,8 @@ softpipe_bind_vs_state(struct pipe_context *pipe, void *vs)
 
    softpipe->vs = (const struct sp_vertex_shader *)vs;
 
-   draw_bind_vertex_shader(softpipe->draw, softpipe->vs->draw_data);
+   draw_bind_vertex_shader(softpipe->draw,
+                           (softpipe->vs ? softpipe->vs->draw_data : NULL));
 
    softpipe->dirty |= SP_NEW_VS;
 }
diff --git a/src/gallium/drivers/softpipe/sp_state_sampler.c b/src/gallium/drivers/softpipe/sp_state_sampler.c
index 1d6dd17d1d..033288a0aa 100644
--- a/src/gallium/drivers/softpipe/sp_state_sampler.c
+++ b/src/gallium/drivers/softpipe/sp_state_sampler.c
@@ -50,45 +50,69 @@ softpipe_create_sampler_state(struct pipe_context *pipe,
    return mem_dup(sampler, sizeof(*sampler));
 }
 
+
 void
-softpipe_bind_sampler_state(struct pipe_context *pipe,
-                            unsigned unit, void *sampler)
+softpipe_bind_sampler_states(struct pipe_context *pipe,
+                             unsigned num, void **sampler)
 {
    struct softpipe_context *softpipe = softpipe_context(pipe);
+   unsigned i;
+
+   assert(num <= PIPE_MAX_SAMPLERS);
+
+   /* Check for no-op */
+   if (num == softpipe->num_samplers &&
+       !memcmp(softpipe->sampler, sampler, num * sizeof(void *)))
+      return;
 
    draw_flush(softpipe->draw);
 
-   assert(unit < PIPE_MAX_SAMPLERS);
-   softpipe->sampler[unit] = (struct pipe_sampler_state *)sampler;
+   for (i = 0; i < num; ++i)
+      softpipe->sampler[i] = sampler[i];
+   for (i = num; i < PIPE_MAX_SAMPLERS; ++i)
+      softpipe->sampler[i] = NULL;
+
+   softpipe->num_samplers = num;
 
    softpipe->dirty |= SP_NEW_SAMPLER;
 }
 
 
 void
-softpipe_delete_sampler_state(struct pipe_context *pipe,
-                              void *sampler)
+softpipe_set_sampler_textures(struct pipe_context *pipe,
+                              unsigned num, struct pipe_texture **texture)
 {
-   FREE( sampler );
-}
+   struct softpipe_context *softpipe = softpipe_context(pipe);
+   uint i;
 
+   assert(num <= PIPE_MAX_SAMPLERS);
 
-void
-softpipe_set_sampler_texture(struct pipe_context *pipe,
-			     unsigned unit,
-			     struct pipe_texture *texture)
-{
-   struct softpipe_context *softpipe = softpipe_context(pipe);
+   /* Check for no-op */
+   if (num == softpipe->num_textures &&
+       !memcmp(softpipe->texture, texture, num * sizeof(struct pipe_texture *)))
+      return;
 
    draw_flush(softpipe->draw);
 
-   assert(unit < PIPE_MAX_SAMPLERS);
-   pipe_texture_reference(&softpipe->texture[unit], texture);
+   for (i = 0; i < PIPE_MAX_SAMPLERS; i++) {
+      struct pipe_texture *tex = i < num ? texture[i] : NULL;
+
+      pipe_texture_reference(&softpipe->texture[i], tex);
+      sp_tile_cache_set_texture(pipe, softpipe->tex_cache[i], tex);
+   }
 
-   sp_tile_cache_set_texture(pipe, softpipe->tex_cache[unit], texture);
+   softpipe->num_textures = num;
 
    softpipe->dirty |= SP_NEW_TEXTURE;
 }
 
 
+void
+softpipe_delete_sampler_state(struct pipe_context *pipe,
+                              void *sampler)
+{
+   FREE( sampler );
+}
+
+
 
diff --git a/src/gallium/drivers/softpipe/sp_tex_sample.c b/src/gallium/drivers/softpipe/sp_tex_sample.c
index 0ced585c7f..34da6356d7 100644
--- a/src/gallium/drivers/softpipe/sp_tex_sample.c
+++ b/src/gallium/drivers/softpipe/sp_tex_sample.c
@@ -714,6 +714,7 @@ sp_get_samples_2d_common(struct tgsi_sampler *sampler,
       }
       break;
    case PIPE_TEX_FILTER_LINEAR:
+   case PIPE_TEX_FILTER_ANISO:
       for (j = 0; j < QUAD_SIZE; j++) {
          float tx[4][4], a, b;
          int x0, y0, x1, y1, c;
@@ -846,6 +847,7 @@ sp_get_samples_3d(struct tgsi_sampler *sampler,
       }
       break;
    case PIPE_TEX_FILTER_LINEAR:
+   case PIPE_TEX_FILTER_ANISO:
       for (j = 0; j < QUAD_SIZE; j++) {
          float texel0[4][4], texel1[4][4];
          float xw, yw, zw; /* interpolation weights */
@@ -972,6 +974,7 @@ sp_get_samples_rect(struct tgsi_sampler *sampler,
       }
       break;
    case PIPE_TEX_FILTER_LINEAR:
+   case PIPE_TEX_FILTER_ANISO:
       for (j = 0; j < QUAD_SIZE; j++) {
          float tx[4][4], a, b;
          int x0, y0, x1, y1, c;
diff --git a/src/gallium/drivers/softpipe/sp_texture.c b/src/gallium/drivers/softpipe/sp_texture.c
index c605ed925b..bbf2d80308 100644
--- a/src/gallium/drivers/softpipe/sp_texture.c
+++ b/src/gallium/drivers/softpipe/sp_texture.c
@@ -1,202 +1,202 @@
-/**************************************************************************
- * 
- * 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"
-#include "sp_tile_cache.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);
-   }
-}
-
-
-static struct pipe_texture *
-softpipe_texture_create_screen(struct pipe_screen *screen,
-                               const struct pipe_texture *templat)
-{
-   struct pipe_winsys *ws = screen->winsys;
-   struct softpipe_texture *spt = CALLOC_STRUCT(softpipe_texture);
-   if (!spt)
-      return NULL;
-
-   spt->base = *templat;
-   spt->base.refcount = 1;
-   spt->base.screen = screen;
-
-   softpipe_texture_layout(spt);
-
-   spt->buffer = ws->buffer_create(ws, 32,
-                                   PIPE_BUFFER_USAGE_PIXEL,
-                                   spt->buffer_size);
-   if (!spt->buffer) {
-      FREE(spt);
-      return NULL;
-   }
-
-   assert(spt->base.refcount == 1);
-
-   return &spt->base;
-}
-
-
-static void
-softpipe_texture_release_screen(struct pipe_screen *screen,
-                                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(screen->winsys, &spt->buffer, NULL);
-
-      FREE(spt);
-   }
-   *pt = NULL;
-}
-
-
-static struct pipe_surface *
-softpipe_get_tex_surface_screen(struct pipe_screen *screen,
-                                struct pipe_texture *pt,
-                                unsigned face, unsigned level, unsigned zslice)
-{
-   struct pipe_winsys *ws = screen->winsys;
-   struct softpipe_texture *spt = softpipe_texture(pt);
-   struct pipe_surface *ps;
-
-   assert(level <= pt->last_level);
-
-   ps = ws->surface_alloc(ws);
-   if (ps) {
-      assert(ps->refcount);
-      assert(ps->winsys);
-      pipe_buffer_reference(ws, &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
-softpipe_texture_update(struct pipe_context *pipe,
-                        struct pipe_texture *texture)
-{
-   struct softpipe_context *softpipe = softpipe_context(pipe);
-   uint unit;
-   for (unit = 0; unit < PIPE_MAX_SAMPLERS; unit++) {
-      if (softpipe->texture[unit] == texture) {
-         sp_flush_tile_cache(softpipe, softpipe->tex_cache[unit]);
-      }
-   }
-}
-
-
-void
-softpipe_init_texture_funcs( struct softpipe_context *softpipe )
-{
-   softpipe->pipe.texture_update = softpipe_texture_update;
-}
-
-
-void
-softpipe_init_screen_texture_funcs(struct pipe_screen *screen)
-{
-   screen->texture_create = softpipe_texture_create_screen;
-   screen->texture_release = softpipe_texture_release_screen;
-   screen->get_tex_surface = softpipe_get_tex_surface_screen;
-}
+/**************************************************************************

+ * 

+ * 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"

+#include "sp_tile_cache.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);

+   }

+}

+

+

+static struct pipe_texture *

+softpipe_texture_create_screen(struct pipe_screen *screen,

+                               const struct pipe_texture *templat)

+{

+   struct pipe_winsys *ws = screen->winsys;

+   struct softpipe_texture *spt = CALLOC_STRUCT(softpipe_texture);

+   if (!spt)

+      return NULL;

+

+   spt->base = *templat;

+   spt->base.refcount = 1;

+   spt->base.screen = screen;

+

+   softpipe_texture_layout(spt);

+

+   spt->buffer = ws->buffer_create(ws, 32,

+                                   PIPE_BUFFER_USAGE_PIXEL,

+                                   spt->buffer_size);

+   if (!spt->buffer) {

+      FREE(spt);

+      return NULL;

+   }

+

+   assert(spt->base.refcount == 1);

+

+   return &spt->base;

+}

+

+

+static void

+softpipe_texture_release_screen(struct pipe_screen *screen,

+                                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(screen->winsys, &spt->buffer, NULL);

+

+      FREE(spt);

+   }

+   *pt = NULL;

+}

+

+

+static struct pipe_surface *

+softpipe_get_tex_surface_screen(struct pipe_screen *screen,

+                                struct pipe_texture *pt,

+                                unsigned face, unsigned level, unsigned zslice)

+{

+   struct pipe_winsys *ws = screen->winsys;

+   struct softpipe_texture *spt = softpipe_texture(pt);

+   struct pipe_surface *ps;

+

+   assert(level <= pt->last_level);

+

+   ps = ws->surface_alloc(ws);

+   if (ps) {

+      assert(ps->refcount);

+      assert(ps->winsys);

+      pipe_buffer_reference(ws, &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

+softpipe_texture_update(struct pipe_context *pipe,

+                        struct pipe_texture *texture)

+{

+   struct softpipe_context *softpipe = softpipe_context(pipe);

+   uint unit;

+   for (unit = 0; unit < softpipe->num_textures; unit++) {

+      if (softpipe->texture[unit] == texture) {

+         sp_flush_tile_cache(softpipe, softpipe->tex_cache[unit]);

+      }

+   }

+}

+

+

+void

+softpipe_init_texture_funcs( struct softpipe_context *softpipe )

+{

+   softpipe->pipe.texture_update = softpipe_texture_update;

+}

+

+

+void

+softpipe_init_screen_texture_funcs(struct pipe_screen *screen)

+{

+   screen->texture_create = softpipe_texture_create_screen;

+   screen->texture_release = softpipe_texture_release_screen;

+   screen->get_tex_surface = softpipe_get_tex_surface_screen;

+}

diff --git a/src/gallium/include/pipe/p_context.h b/src/gallium/include/pipe/p_context.h
index 1501b52f3e..b64948f0f3 100644
--- a/src/gallium/include/pipe/p_context.h
+++ b/src/gallium/include/pipe/p_context.h
@@ -100,7 +100,7 @@ struct pipe_context {
 
    void * (*create_sampler_state)(struct pipe_context *,
                                   const struct pipe_sampler_state *);
-   void   (*bind_sampler_state)(struct pipe_context *, unsigned unit, void *);
+   void   (*bind_sampler_states)(struct pipe_context *, unsigned num, void **);
    void   (*delete_sampler_state)(struct pipe_context *, void *);
 
    void * (*create_rasterizer_state)(struct pipe_context *,
@@ -148,9 +148,9 @@ struct pipe_context {
 
    /* 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_sampler_textures)( struct pipe_context *,
+                                 unsigned num,
+                                 struct pipe_texture ** );
 
    void (*set_viewport_state)( struct pipe_context *,
                                const struct pipe_viewport_state * );
diff --git a/src/gallium/include/pipe/p_debug.h b/src/gallium/include/pipe/p_debug.h
index 2a11627b36..f45363f355 100644
--- a/src/gallium/include/pipe/p_debug.h
+++ b/src/gallium/include/pipe/p_debug.h
@@ -41,6 +41,8 @@
 
 #include <stdarg.h>
 
+#include "p_compiler.h"
+
 
 #ifdef	__cplusplus
 extern "C" {
@@ -58,8 +60,22 @@ extern "C" {
 #endif
 
 
+/**
+ * Print debug messages.
+ *
+ * A debug message will be printed regardless of the DEBUG/NDEBUG macros.
+ *
+ * The actual channel used to output debug message is platform specific. To 
+ * avoid misformating or truncation, follow these rules of thumb:   
+ * - output whole lines
+ * - avoid outputing large strings (512 bytes is the current maximum length 
+ * that is guaranteed to be printed in all platforms)
+ */
 void debug_printf(const char *format, ...);
 
+/**
+ * @sa debug_printf 
+ */
 void debug_vprintf(const char *format, va_list ap);
 
 void debug_assert_fail(const char *expr, const char *file, unsigned line);
@@ -79,6 +95,68 @@ void debug_assert_fail(const char *expr, const char *file, unsigned line);
 #define assert(expr) debug_assert(expr)
 
 
+/**
+ * Set a channel's debug mask.
+ * 
+ * uuid is just a random 32 bit integer that uniquely identifies the debugging 
+ * channel. 
+ * 
+ * @note Due to current implementation issues, make sure the lower 8 bits of 
+ * UUID are unique.
+ */
+void debug_mask_set(uint32_t uuid, uint32_t mask);
+
+
+uint32_t debug_mask_get(uint32_t uuid);
+
+
+/**
+ * Conditional debug output. 
+ * 
+ * This is just a generalization of the debug filtering mechanism used 
+ * throughout Gallium.
+ * 
+ * You use this function as:
+ * 
+ * @code
+ * #define MYDRIVER_UUID 0x12345678 // random 32 bit identifier
+ * 
+ * static inline mydriver_debug(uint32_t what, const char *format, ...)
+ * {
+ * #ifdef DEBUG
+ *    va_list ap;
+ *    va_start(ap, format);
+ *    debug_mask_vprintf(MYDRIVER_UUID, what, format, ap);
+ *    va_end(ap);
+ * #endif
+ * }
+ * 
+ * ...
+ * 
+ *    debug_mask_set(MYDRIVER_UUID, 
+ *                   MYDRIVER_DEBUG_THIS | 
+ *                   MYDRIVER_DEBUG_THAT |
+ *                   ... );
+ * 
+ * ...
+ * 
+ *    mydriver_debug(MYDRIVER_DEBUG_THIS,
+ *                   "this and this happened\n");
+ *
+ *    mydriver_debug(MYDRIVER_DEBUG_THAT,
+ *                   "that = %f\n", that);
+ * ...
+ * @endcode     
+ * 
+ * You can also define several variants of mydriver_debug, with hardcoded what. 
+ * Note that although macros with variable number of arguments would accomplish 
+ * more in less code, they are not portable. 
+ */
+void debug_mask_vprintf(uint32_t uuid, 
+                        uint32_t what, 
+                        const char *format, 
+                        va_list ap);
+
 #ifdef	__cplusplus
 }
 #endif
diff --git a/src/gallium/include/pipe/p_defines.h b/src/gallium/include/pipe/p_defines.h
index 0c662d6517..bc938ba253 100644
--- a/src/gallium/include/pipe/p_defines.h
+++ b/src/gallium/include/pipe/p_defines.h
@@ -152,7 +152,7 @@ enum pipe_texture_target {
  */
 #define PIPE_TEX_FILTER_NEAREST      0
 #define PIPE_TEX_FILTER_LINEAR       1
-/* #define PIPE_TEX_FILTER_ANISO        2 */
+#define PIPE_TEX_FILTER_ANISO        2 
 
 
 #define PIPE_TEX_COMPARE_NONE          0
diff --git a/src/gallium/include/pipe/p_shader_tokens.h b/src/gallium/include/pipe/p_shader_tokens.h
index 0a6145a6bf..210169f54f 100644
--- a/src/gallium/include/pipe/p_shader_tokens.h
+++ b/src/gallium/include/pipe/p_shader_tokens.h
@@ -42,15 +42,17 @@ struct tgsi_token
    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_FILE_COUNT       8  /**< how many TGSI_FILE_ types */
+enum tgsi_file_type {
+   TGSI_FILE_NULL        =0,
+   TGSI_FILE_CONSTANT    =1,
+   TGSI_FILE_INPUT       =2,
+   TGSI_FILE_OUTPUT      =3,
+   TGSI_FILE_TEMPORARY   =4,
+   TGSI_FILE_SAMPLER     =5,
+   TGSI_FILE_ADDRESS     =6,
+   TGSI_FILE_IMMEDIATE   =7,
+   TGSI_FILE_COUNT      /**< how many TGSI_FILE_ types */
+};
 
 
 #define TGSI_DECLARE_RANGE    0
@@ -225,6 +227,7 @@ struct tgsi_immediate_float32
 #define TGSI_OPCODE_STR                 51
 #define TGSI_OPCODE_TEX                 52
 #define TGSI_OPCODE_TXD                 53
+#define TGSI_OPCODE_TXP                 134
 #define TGSI_OPCODE_UP2H                54
 #define TGSI_OPCODE_UP2US               55
 #define TGSI_OPCODE_UP4B                56
@@ -339,7 +342,7 @@ struct tgsi_immediate_float32
  * ps_1_1
  */
 #define TGSI_OPCODE_TEXCOORD            TGSI_OPCODE_NOP
-#define TGSI_OPCODE_TEXKILL             TGSI_OPCODE_KIL
+#define TGSI_OPCODE_TEXKILL             TGSI_OPCODE_KILP
 #define TGSI_OPCODE_TEXBEM              107
 #define TGSI_OPCODE_TEXBEML             108
 #define TGSI_OPCODE_TEXREG2AR           109
@@ -419,7 +422,7 @@ struct tgsi_immediate_float32
 #define TGSI_OPCODE_KIL                 132  /* unpredicated kill */
 #define TGSI_OPCODE_END                 133  /* aka HALT */
 
-#define TGSI_OPCODE_LAST                134
+#define TGSI_OPCODE_LAST                135
 
 #define TGSI_SAT_NONE            0  /* do not saturate */
 #define TGSI_SAT_ZERO_ONE        1  /* clamp to [0,1] */
@@ -652,9 +655,6 @@ struct tgsi_src_register_ext
  *
  * 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
@@ -668,8 +668,7 @@ struct tgsi_src_register_ext_swz
    unsigned NegateY      : 1;    /* BOOL */
    unsigned NegateZ      : 1;    /* BOOL */
    unsigned NegateW      : 1;    /* BOOL */
-   unsigned ExtDivide    : 4;    /* TGSI_EXTSWIZZLE_ */
-   unsigned Padding      : 3;
+   unsigned Padding      : 7;
    unsigned Extended     : 1;    /* BOOL */
 };
 
diff --git a/src/gallium/include/pipe/p_state.h b/src/gallium/include/pipe/p_state.h
index 5fab41acbd..e338a27383 100644
--- a/src/gallium/include/pipe/p_state.h
+++ b/src/gallium/include/pipe/p_state.h
@@ -108,6 +108,7 @@ struct pipe_rasterizer_state
    unsigned line_stipple_enable:1;
    unsigned line_stipple_factor:8;  /**< [1..256] actually */
    unsigned line_stipple_pattern:16;
+   unsigned line_last_pixel:1;
    unsigned bypass_clipping:1;
    unsigned origin_lower_left:1;  /**< Is (0,0) the lower-left corner? */
 
@@ -154,7 +155,7 @@ struct pipe_clip_state
 struct pipe_constant_buffer
 {
    struct pipe_buffer *buffer;
-   unsigned size;    /** in bytes */
+   unsigned size;    /** in bytes (XXX: redundant!) */
 };
 
 
@@ -248,6 +249,7 @@ struct pipe_sampler_state
    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]? */
+   unsigned prefilter:4;         /**< Wierd sampling state exposed by some api's */
    float shadow_ambient;         /**< shadow test fail color/intensity */
    float lod_bias;               /**< LOD/lambda bias */
    float min_lod, max_lod;       /**< LOD clamp range, after bias */
@@ -281,8 +283,6 @@ struct pipe_surface
  */
 struct pipe_texture
 { 
-   /* Effectively the key:
-    */
    enum pipe_texture_target target; /**< PIPE_TEXTURE_x */
    enum pipe_format format;         /**< PIPE_FORMAT_x */
 
diff --git a/src/gallium/include/pipe/p_util.h b/src/gallium/include/pipe/p_util.h
index 3b32ba1d24..ef36ce75f7 100644
--- a/src/gallium/include/pipe/p_util.h
+++ b/src/gallium/include/pipe/p_util.h
@@ -54,7 +54,12 @@ EngFreeMem(
 static INLINE void *
 MALLOC( unsigned size )
 {
+#ifdef WINCE
+   /* TODO: Need to abstract this */
+   return malloc( size );
+#else
    return EngAllocMem( 0, size, 'D3AG' );
+#endif
 }
 
 static INLINE void *
@@ -71,7 +76,12 @@ static INLINE void
 FREE( void *ptr )
 {
    if( ptr ) {
+#ifdef WINCE
+      /* TODO: Need to abstract this */
+      free( ptr );
+#else
       EngFreeMem( ptr );
+#endif
    }
 }
 
diff --git a/src/gallium/include/pipe/p_winsys.h b/src/gallium/include/pipe/p_winsys.h
index e784c92491..1383bd0544 100644
--- a/src/gallium/include/pipe/p_winsys.h
+++ b/src/gallium/include/pipe/p_winsys.h
@@ -104,13 +104,36 @@ struct pipe_winsys
     * 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.
+    *
+    * alignment indicates the client's alignment requirements, eg for
+    * SSE instructions.
     */
    struct pipe_buffer *(*buffer_create)( struct pipe_winsys *sws, 
-					        unsigned alignment,
-                                                unsigned usage,
-                                                unsigned size );
+                                         unsigned alignment, 
+                                         unsigned usage,
+                                         unsigned size );
 
-   /** Create a buffer that wraps user-space data */
+   /** 
+    * Create a buffer that wraps user-space data.
+    *
+    * Effectively this schedules a delayed call to buffer_create
+    * followed by an upload of the data at *some point in the future*,
+    * or perhaps never.  Basically the allocate/upload is delayed
+    * until the buffer is actually passed to hardware.
+    *
+    * The intention is to provide a quick way to turn regular data
+    * into a buffer, and secondly to avoid a copy operation if that
+    * data subsequently turns out to be only accessed by the CPU.  
+    *
+    * Common example is OpenGL vertex buffers that are subsequently
+    * processed either by software TNL in the driver or by passing to
+    * hardware.
+    *
+    * XXX: What happens if the delayed call to buffer_create() fails?
+    *
+    * Note that ptr may be accessed at any time upto the time when the
+    * buffer is destroyed, so the data must not be freed before then.
+    */
    struct pipe_buffer *(*user_buffer_create)(struct pipe_winsys *sws, 
                                                     void *ptr,
                                                     unsigned bytes);
diff --git a/src/gallium/winsys/SConscript b/src/gallium/winsys/SConscript
index 635a68eea2..f18d3bd2f8 100644
--- a/src/gallium/winsys/SConscript
+++ b/src/gallium/winsys/SConscript
@@ -5,7 +5,7 @@ if dri:
 		'dri/SConscript',
 	])
 	
-if 'xlib' in env['drivers'] and not dri:
+if 'xlib' in env['winsys'] and not dri:
 	SConscript([
 		'xlib/SConscript',
 	])
diff --git a/src/mesa/SConscript b/src/mesa/SConscript
index a77c3de6c7..678e4ad0e8 100644
--- a/src/mesa/SConscript
+++ b/src/mesa/SConscript
@@ -161,6 +161,7 @@ STATETRACKER_SOURCES = [
 	'state_tracker/st_extensions.c',
 	'state_tracker/st_format.c',
 	'state_tracker/st_framebuffer.c',
+	'state_tracker/st_gen_mipmap.c',
 	'state_tracker/st_mesa_to_tgsi.c',
 	'state_tracker/st_program.c',
 	'state_tracker/st_texture.c',
diff --git a/src/mesa/main/teximage.c b/src/mesa/main/teximage.c
index 5c96be9216..f15e404527 100644
--- a/src/mesa/main/teximage.c
+++ b/src/mesa/main/teximage.c
@@ -1214,19 +1214,30 @@ _mesa_init_teximage_fields(GLcontext *ctx, GLenum target,
    img->Width = width;
    img->Height = height;
    img->Depth = depth;
+
    img->Width2 = width - 2 * border;   /* == 1 << img->WidthLog2; */
-   img->Height2 = height - 2 * border; /* == 1 << img->HeightLog2; */
-   img->Depth2 = depth - 2 * border;   /* == 1 << img->DepthLog2; */
    img->WidthLog2 = logbase2(img->Width2);
-   if (height == 1)  /* 1-D texture */
+
+   if (height == 1) { /* 1-D texture */
+      img->Height2 = 1;
       img->HeightLog2 = 0;
-   else
+   }
+   else {
+      img->Height2 = height - 2 * border; /* == 1 << img->HeightLog2; */
       img->HeightLog2 = logbase2(img->Height2);
-   if (depth == 1)   /* 2-D texture */
+   }
+
+   if (depth == 1) {  /* 2-D texture */
+      img->Depth2 = 1;
       img->DepthLog2 = 0;
-   else
+   }
+   else {
+      img->Depth2 = depth - 2 * border;   /* == 1 << img->DepthLog2; */
       img->DepthLog2 = logbase2(img->Depth2);
+   }
+
    img->MaxLog2 = MAX2(img->WidthLog2, img->HeightLog2);
+
    img->IsCompressed = GL_FALSE;
    img->CompressedSize = 0;
 
diff --git a/src/mesa/shader/arbprogparse.c b/src/mesa/shader/arbprogparse.c
index b6b3c88b14..e385b9d997 100644
--- a/src/mesa/shader/arbprogparse.c
+++ b/src/mesa/shader/arbprogparse.c
@@ -3910,8 +3910,11 @@ _mesa_parse_arb_fragment_program(GLcontext* ctx, GLenum target,
    program->Base.NumNativeTexIndirections = ap.Base.NumTexIndirections;
    program->Base.InputsRead      = ap.Base.InputsRead;
    program->Base.OutputsWritten  = ap.Base.OutputsWritten;
-   for (i = 0; i < MAX_TEXTURE_IMAGE_UNITS; i++)
+   for (i = 0; i < MAX_TEXTURE_IMAGE_UNITS; i++) {
       program->Base.TexturesUsed[i] = ap.TexturesUsed[i];
+      if (ap.TexturesUsed[i])
+         program->Base.SamplersUsed |= (1 << i);
+   }
    program->Base.ShadowSamplers = ap.ShadowSamplers;
    program->FogOption          = ap.FogOption;
    program->UsesKill          = ap.UsesKill;
diff --git a/src/mesa/sources b/src/mesa/sources
index f0bf7b31fb..e3d5f22849 100644
--- a/src/mesa/sources
+++ b/src/mesa/sources
@@ -184,7 +184,6 @@ STATETRACKER_SOURCES = \
 	state_tracker/st_cb_readpixels.c \
 	state_tracker/st_cb_strings.c \
 	state_tracker/st_cb_texture.c \
-	state_tracker/st_cache.c \
 	state_tracker/st_context.c \
 	state_tracker/st_debug.c \
 	state_tracker/st_draw.c \
diff --git a/src/mesa/state_tracker/st_atom_blend.c b/src/mesa/state_tracker/st_atom_blend.c
index 2a9d209153..6c13fc8141 100644
--- a/src/mesa/state_tracker/st_atom_blend.c
+++ b/src/mesa/state_tracker/st_atom_blend.c
@@ -33,11 +33,11 @@
  
 
 #include "st_context.h"
-#include "st_cache.h"
 #include "st_atom.h"
 
 #include "pipe/p_context.h"
 #include "pipe/p_defines.h"
+#include "cso_cache/cso_context.h"
 
 
 /**
@@ -155,44 +155,43 @@ translate_logicop(GLenum logicop)
 static void 
 update_blend( struct st_context *st )
 {
-   struct pipe_blend_state blend;
-   const struct cso_blend *cso;
+   struct pipe_blend_state *blend = &st->state.blend;
 
-   memset(&blend, 0, sizeof(blend));
+   memset(blend, 0, sizeof(*blend));
 
    if (st->ctx->Color.ColorLogicOpEnabled ||
        (st->ctx->Color.BlendEnabled &&
         st->ctx->Color.BlendEquationRGB == GL_LOGIC_OP)) {
       /* logicop enabled */
-      blend.logicop_enable = 1;
-      blend.logicop_func = translate_logicop(st->ctx->Color.LogicOp);
+      blend->logicop_enable = 1;
+      blend->logicop_func = translate_logicop(st->ctx->Color.LogicOp);
    }
    else if (st->ctx->Color.BlendEnabled) {
       /* blending enabled */
-      blend.blend_enable = 1;
+      blend->blend_enable = 1;
 
-      blend.rgb_func = translate_blend(st->ctx->Color.BlendEquationRGB);
+      blend->rgb_func = translate_blend(st->ctx->Color.BlendEquationRGB);
       if (st->ctx->Color.BlendEquationRGB == GL_MIN ||
           st->ctx->Color.BlendEquationRGB == GL_MAX) {
          /* Min/max are special */
-         blend.rgb_src_factor = PIPE_BLENDFACTOR_ONE;
-         blend.rgb_dst_factor = PIPE_BLENDFACTOR_ONE;
+         blend->rgb_src_factor = PIPE_BLENDFACTOR_ONE;
+         blend->rgb_dst_factor = PIPE_BLENDFACTOR_ONE;
       }
       else {
-         blend.rgb_src_factor = translate_blend(st->ctx->Color.BlendSrcRGB);
-         blend.rgb_dst_factor = translate_blend(st->ctx->Color.BlendDstRGB);
+         blend->rgb_src_factor = translate_blend(st->ctx->Color.BlendSrcRGB);
+         blend->rgb_dst_factor = translate_blend(st->ctx->Color.BlendDstRGB);
       }
 
-      blend.alpha_func = translate_blend(st->ctx->Color.BlendEquationA);
+      blend->alpha_func = translate_blend(st->ctx->Color.BlendEquationA);
       if (st->ctx->Color.BlendEquationA == GL_MIN ||
           st->ctx->Color.BlendEquationA == GL_MAX) {
          /* Min/max are special */
-         blend.alpha_src_factor = PIPE_BLENDFACTOR_ONE;
-         blend.alpha_dst_factor = PIPE_BLENDFACTOR_ONE;
+         blend->alpha_src_factor = PIPE_BLENDFACTOR_ONE;
+         blend->alpha_dst_factor = PIPE_BLENDFACTOR_ONE;
       }
       else {
-         blend.alpha_src_factor = translate_blend(st->ctx->Color.BlendSrcA);
-         blend.alpha_dst_factor = translate_blend(st->ctx->Color.BlendDstA);
+         blend->alpha_src_factor = translate_blend(st->ctx->Color.BlendSrcA);
+         blend->alpha_dst_factor = translate_blend(st->ctx->Color.BlendDstA);
       }
    }
    else {
@@ -201,25 +200,18 @@ update_blend( struct st_context *st )
 
    /* Colormask - maybe reverse these bits? */
    if (st->ctx->Color.ColorMask[0])
-      blend.colormask |= PIPE_MASK_R;
+      blend->colormask |= PIPE_MASK_R;
    if (st->ctx->Color.ColorMask[1])
-      blend.colormask |= PIPE_MASK_G;
+      blend->colormask |= PIPE_MASK_G;
    if (st->ctx->Color.ColorMask[2])
-      blend.colormask |= PIPE_MASK_B;
+      blend->colormask |= PIPE_MASK_B;
    if (st->ctx->Color.ColorMask[3])
-      blend.colormask |= PIPE_MASK_A;
+      blend->colormask |= PIPE_MASK_A;
 
    if (st->ctx->Color.DitherFlag)
-      blend.dither = 1;
+      blend->dither = 1;
 
-   cso = st_cached_blend_state(st, &blend);
-
-   if (st->state.blend != cso) {
-      /* state has changed */
-      st->state.blend = cso;
-      /* bind new state */
-      st->pipe->bind_blend_state(st->pipe, cso->data);
-   }
+   cso_set_blend(st->cso_context, blend);
 
    if (memcmp(st->ctx->Color.BlendColor, &st->state.blend_color, 4 * sizeof(GLfloat)) != 0) {
       /* state has changed */
diff --git a/src/mesa/state_tracker/st_atom_depth.c b/src/mesa/state_tracker/st_atom_depth.c
index 7aecdbfbcc..827ad3b548 100644
--- a/src/mesa/state_tracker/st_atom_depth.c
+++ b/src/mesa/state_tracker/st_atom_depth.c
@@ -34,10 +34,10 @@
  
 
 #include "st_context.h"
-#include "st_cache.h"
 #include "st_atom.h"
 #include "pipe/p_context.h"
 #include "pipe/p_defines.h"
+#include "cso_cache/cso_context.h"
 
 
 /**
@@ -93,53 +93,47 @@ gl_stencil_op_to_pipe(GLenum func)
 static void
 update_depth_stencil_alpha(struct st_context *st)
 {
-   struct pipe_depth_stencil_alpha_state depth_stencil;
-   const struct cso_depth_stencil_alpha *cso;
+   struct pipe_depth_stencil_alpha_state *dsa = &st->state.depth_stencil;
 
-   memset(&depth_stencil, 0, sizeof(depth_stencil));
+   memset(dsa, 0, sizeof(*dsa));
 
-   depth_stencil.depth.enabled = st->ctx->Depth.Test;
-   depth_stencil.depth.writemask = st->ctx->Depth.Mask;
-   depth_stencil.depth.func = st_compare_func_to_pipe(st->ctx->Depth.Func);
+   dsa->depth.enabled = st->ctx->Depth.Test;
+   dsa->depth.writemask = st->ctx->Depth.Mask;
+   dsa->depth.func = st_compare_func_to_pipe(st->ctx->Depth.Func);
 
    if (st->ctx->Query.CurrentOcclusionObject &&
        st->ctx->Query.CurrentOcclusionObject->Active)
-      depth_stencil.depth.occlusion_count = 1;
+      dsa->depth.occlusion_count = 1;
 
    if (st->ctx->Stencil.Enabled) {
-      depth_stencil.stencil[0].enabled = 1;
-      depth_stencil.stencil[0].func = st_compare_func_to_pipe(st->ctx->Stencil.Function[0]);
-      depth_stencil.stencil[0].fail_op = gl_stencil_op_to_pipe(st->ctx->Stencil.FailFunc[0]);
-      depth_stencil.stencil[0].zfail_op = gl_stencil_op_to_pipe(st->ctx->Stencil.ZFailFunc[0]);
-      depth_stencil.stencil[0].zpass_op = gl_stencil_op_to_pipe(st->ctx->Stencil.ZPassFunc[0]);
-      depth_stencil.stencil[0].ref_value = st->ctx->Stencil.Ref[0] & 0xff;
-      depth_stencil.stencil[0].value_mask = st->ctx->Stencil.ValueMask[0] & 0xff;
-      depth_stencil.stencil[0].write_mask = st->ctx->Stencil.WriteMask[0] & 0xff;
+      dsa->stencil[0].enabled = 1;
+      dsa->stencil[0].func = st_compare_func_to_pipe(st->ctx->Stencil.Function[0]);
+      dsa->stencil[0].fail_op = gl_stencil_op_to_pipe(st->ctx->Stencil.FailFunc[0]);
+      dsa->stencil[0].zfail_op = gl_stencil_op_to_pipe(st->ctx->Stencil.ZFailFunc[0]);
+      dsa->stencil[0].zpass_op = gl_stencil_op_to_pipe(st->ctx->Stencil.ZPassFunc[0]);
+      dsa->stencil[0].ref_value = st->ctx->Stencil.Ref[0] & 0xff;
+      dsa->stencil[0].value_mask = st->ctx->Stencil.ValueMask[0] & 0xff;
+      dsa->stencil[0].write_mask = st->ctx->Stencil.WriteMask[0] & 0xff;
 
       if (st->ctx->Stencil.TestTwoSide) {
-         depth_stencil.stencil[1].enabled = 1;
-         depth_stencil.stencil[1].func = st_compare_func_to_pipe(st->ctx->Stencil.Function[1]);
-         depth_stencil.stencil[1].fail_op = gl_stencil_op_to_pipe(st->ctx->Stencil.FailFunc[1]);
-         depth_stencil.stencil[1].zfail_op = gl_stencil_op_to_pipe(st->ctx->Stencil.ZFailFunc[1]);
-         depth_stencil.stencil[1].zpass_op = gl_stencil_op_to_pipe(st->ctx->Stencil.ZPassFunc[1]);
-         depth_stencil.stencil[1].ref_value = st->ctx->Stencil.Ref[1] & 0xff;
-         depth_stencil.stencil[1].value_mask = st->ctx->Stencil.ValueMask[1] & 0xff;
-         depth_stencil.stencil[1].write_mask = st->ctx->Stencil.WriteMask[1] & 0xff;
+         dsa->stencil[1].enabled = 1;
+         dsa->stencil[1].func = st_compare_func_to_pipe(st->ctx->Stencil.Function[1]);
+         dsa->stencil[1].fail_op = gl_stencil_op_to_pipe(st->ctx->Stencil.FailFunc[1]);
+         dsa->stencil[1].zfail_op = gl_stencil_op_to_pipe(st->ctx->Stencil.ZFailFunc[1]);
+         dsa->stencil[1].zpass_op = gl_stencil_op_to_pipe(st->ctx->Stencil.ZPassFunc[1]);
+         dsa->stencil[1].ref_value = st->ctx->Stencil.Ref[1] & 0xff;
+         dsa->stencil[1].value_mask = st->ctx->Stencil.ValueMask[1] & 0xff;
+         dsa->stencil[1].write_mask = st->ctx->Stencil.WriteMask[1] & 0xff;
       }
    }
 
    if (st->ctx->Color.AlphaEnabled) {
-      depth_stencil.alpha.enabled = 1;
-      depth_stencil.alpha.func = st_compare_func_to_pipe(st->ctx->Color.AlphaFunc);
-      depth_stencil.alpha.ref = st->ctx->Color.AlphaRef;
+      dsa->alpha.enabled = 1;
+      dsa->alpha.func = st_compare_func_to_pipe(st->ctx->Color.AlphaFunc);
+      dsa->alpha.ref = st->ctx->Color.AlphaRef;
    }
 
-   cso = st_cached_depth_stencil_alpha_state(st, &depth_stencil);
-   if (st->state.depth_stencil != cso) {
-      /* state has changed */
-      st->state.depth_stencil = cso;
-      st->pipe->bind_depth_stencil_alpha_state(st->pipe, cso->data); /* bind new state */
-   }
+   cso_set_depth_stencil_alpha(st->cso_context, dsa);
 }
 
 
diff --git a/src/mesa/state_tracker/st_atom_rasterizer.c b/src/mesa/state_tracker/st_atom_rasterizer.c
index 229839d8b2..77cef9236b 100644
--- a/src/mesa/state_tracker/st_atom_rasterizer.c
+++ b/src/mesa/state_tracker/st_atom_rasterizer.c
@@ -32,10 +32,11 @@
  
 #include "main/macros.h"
 #include "st_context.h"
-#include "st_cache.h"
+#include "st_atom.h"
 #include "pipe/p_context.h"
 #include "pipe/p_defines.h"
-#include "st_atom.h"
+#include "cso_cache/cso_context.h"
+
 
 static GLuint translate_fill( GLenum mode )
 {
@@ -72,22 +73,21 @@ static GLboolean get_offset_flag( GLuint fill_mode,
 static void update_raster_state( struct st_context *st )
 {
    GLcontext *ctx = st->ctx;
-   struct pipe_rasterizer_state raster;
-   const struct cso_rasterizer *cso;
+   struct pipe_rasterizer_state *raster = &st->state.rasterizer;
    const struct gl_vertex_program *vertProg = ctx->VertexProgram._Current;
    uint i;
 
-   memset(&raster, 0, sizeof(raster));
+   memset(raster, 0, sizeof(*raster));
 
-   raster.origin_lower_left = 1; /* Always true for OpenGL */
+   raster->origin_lower_left = 1; /* Always true for OpenGL */
    
    /* _NEW_POLYGON, _NEW_BUFFERS
     */
    {
       if (ctx->Polygon.FrontFace == GL_CCW)
-         raster.front_winding = PIPE_WINDING_CCW;
+         raster->front_winding = PIPE_WINDING_CCW;
       else
-         raster.front_winding = PIPE_WINDING_CW;
+         raster->front_winding = PIPE_WINDING_CW;
 
       /* XXX
        * I think the intention here is that user-created framebuffer objects
@@ -96,13 +96,13 @@ static void update_raster_state( struct st_context *st )
        * But this is an implementation/driver-specific artifact - remove...
        */
       if (ctx->DrawBuffer && ctx->DrawBuffer->Name != 0)
-         raster.front_winding ^= PIPE_WINDING_BOTH;
+         raster->front_winding ^= PIPE_WINDING_BOTH;
    }
 
    /* _NEW_LIGHT
     */
    if (ctx->Light.ShadeModel == GL_FLAT)
-      raster.flatshade = 1;
+      raster->flatshade = 1;
 
    /* _NEW_LIGHT | _NEW_PROGRAM
     *
@@ -113,28 +113,28 @@ static void update_raster_state( struct st_context *st )
    if (ctx->VertexProgram._Current) {
       if (ctx->VertexProgram._Enabled) {
          /* user-defined program */
-         raster.light_twoside = ctx->VertexProgram.TwoSideEnabled;
+         raster->light_twoside = ctx->VertexProgram.TwoSideEnabled;
       }
       else {
          /* TNL-generated program */
-         raster.light_twoside = ctx->Light.Enabled && ctx->Light.Model.TwoSide;
+         raster->light_twoside = ctx->Light.Enabled && ctx->Light.Model.TwoSide;
       }
    }
    else if (ctx->Light.Enabled && ctx->Light.Model.TwoSide) {
-      raster.light_twoside = 1;
+      raster->light_twoside = 1;
    }
 
    /* _NEW_POLYGON
     */
    if (ctx->Polygon.CullFlag) {
       if (ctx->Polygon.CullFaceMode == GL_FRONT_AND_BACK) {
-	 raster.cull_mode = PIPE_WINDING_BOTH;
+	 raster->cull_mode = PIPE_WINDING_BOTH;
       }
       else if (ctx->Polygon.CullFaceMode == GL_FRONT) {
-	 raster.cull_mode = raster.front_winding;
+	 raster->cull_mode = raster->front_winding;
       }
       else {
-	 raster.cull_mode = raster.front_winding ^ PIPE_WINDING_BOTH;
+	 raster->cull_mode = raster->front_winding ^ PIPE_WINDING_BOTH;
       }
    }
 
@@ -144,23 +144,23 @@ static void update_raster_state( struct st_context *st )
       GLuint fill_front = translate_fill( ctx->Polygon.FrontMode );
       GLuint fill_back = translate_fill( ctx->Polygon.BackMode );
       
-      if (raster.front_winding == PIPE_WINDING_CW) {
-	 raster.fill_cw = fill_front;
-	 raster.fill_ccw = fill_back;
+      if (raster->front_winding == PIPE_WINDING_CW) {
+	 raster->fill_cw = fill_front;
+	 raster->fill_ccw = fill_back;
       }
       else {
-	 raster.fill_cw = fill_back;
-	 raster.fill_ccw = fill_front;
+	 raster->fill_cw = fill_back;
+	 raster->fill_ccw = fill_front;
       }
 
       /* Simplify when culling is active:
        */
-      if (raster.cull_mode & PIPE_WINDING_CW) {
-	 raster.fill_cw = raster.fill_ccw;
+      if (raster->cull_mode & PIPE_WINDING_CW) {
+	 raster->fill_cw = raster->fill_ccw;
       }
       
-      if (raster.cull_mode & PIPE_WINDING_CCW) {
-	 raster.fill_ccw = raster.fill_cw;
+      if (raster->cull_mode & PIPE_WINDING_CCW) {
+	 raster->fill_ccw = raster->fill_cw;
       }
    }
 
@@ -168,95 +168,91 @@ static void update_raster_state( struct st_context *st )
     */
    if (ctx->Polygon.OffsetUnits != 0.0 ||
        ctx->Polygon.OffsetFactor != 0.0) {
-      raster.offset_cw = get_offset_flag( raster.fill_cw, &ctx->Polygon );
-      raster.offset_ccw = get_offset_flag( raster.fill_ccw, &ctx->Polygon );
-      raster.offset_units = ctx->Polygon.OffsetUnits;
-      raster.offset_scale = ctx->Polygon.OffsetFactor;
+      raster->offset_cw = get_offset_flag( raster->fill_cw, &ctx->Polygon );
+      raster->offset_ccw = get_offset_flag( raster->fill_ccw, &ctx->Polygon );
+      raster->offset_units = ctx->Polygon.OffsetUnits;
+      raster->offset_scale = ctx->Polygon.OffsetFactor;
    }
 
    if (ctx->Polygon.SmoothFlag)
-      raster.poly_smooth = 1;
+      raster->poly_smooth = 1;
 
    if (ctx->Polygon.StippleFlag)
-      raster.poly_stipple_enable = 1;
+      raster->poly_stipple_enable = 1;
 
 
    /* _NEW_BUFFERS, _NEW_POLYGON
     */
-   if (raster.fill_cw != PIPE_POLYGON_MODE_FILL ||
-       raster.fill_ccw != PIPE_POLYGON_MODE_FILL)
+   if (raster->fill_cw != PIPE_POLYGON_MODE_FILL ||
+       raster->fill_ccw != PIPE_POLYGON_MODE_FILL)
    {
       GLfloat mrd = (ctx->DrawBuffer ? 
 		     ctx->DrawBuffer->_MRD : 
 		     1.0);
 
-      raster.offset_units = ctx->Polygon.OffsetFactor * mrd;
-      raster.offset_scale = (ctx->Polygon.OffsetUnits * mrd *
+      raster->offset_units = ctx->Polygon.OffsetFactor * mrd;
+      raster->offset_scale = (ctx->Polygon.OffsetUnits * mrd *
 			    st->polygon_offset_scale);
    }
       
    /* _NEW_POINT
     */
-   raster.point_size = ctx->Point.Size;
-   raster.point_smooth = ctx->Point.SmoothFlag;
-   raster.point_sprite = ctx->Point.PointSprite;
+   raster->point_size = ctx->Point.Size;
+   raster->point_smooth = ctx->Point.SmoothFlag;
+   raster->point_sprite = ctx->Point.PointSprite;
    for (i = 0; i < MAX_TEXTURE_COORD_UNITS; i++) {
       if (ctx->Point.CoordReplace[i]) {
          if (ctx->Point.SpriteOrigin == GL_UPPER_LEFT)
-            raster.sprite_coord_mode[i] = PIPE_SPRITE_COORD_UPPER_LEFT;
+            raster->sprite_coord_mode[i] = PIPE_SPRITE_COORD_UPPER_LEFT;
          else 
-            raster.sprite_coord_mode[i] = PIPE_SPRITE_COORD_LOWER_LEFT;
+            raster->sprite_coord_mode[i] = PIPE_SPRITE_COORD_LOWER_LEFT;
       }
       else {
-         raster.sprite_coord_mode[i] = PIPE_SPRITE_COORD_NONE;
+         raster->sprite_coord_mode[i] = PIPE_SPRITE_COORD_NONE;
       }
    }
    if (vertProg) {
       if (vertProg->Base.Id == 0) {
          if (vertProg->Base.OutputsWritten & (1 << VERT_RESULT_PSIZ)) {
             /* generated program which emits point size */
-            raster.point_size_per_vertex = TRUE;
+            raster->point_size_per_vertex = TRUE;
          }
       }
       else if (ctx->VertexProgram.PointSizeEnabled) {
          /* user-defined program and GL_VERTEX_PROGRAM_POINT_SIZE set */
-         raster.point_size_per_vertex = ctx->VertexProgram.PointSizeEnabled;
+         raster->point_size_per_vertex = ctx->VertexProgram.PointSizeEnabled;
       }
    }
 
 
    /* _NEW_LINE
     */
-   raster.line_smooth = ctx->Line.SmoothFlag;
+   raster->line_smooth = ctx->Line.SmoothFlag;
    if (ctx->Line.SmoothFlag) {
-      raster.line_width = CLAMP(ctx->Line.Width,
+      raster->line_width = CLAMP(ctx->Line.Width,
                                 ctx->Const.MinLineWidthAA,
                                 ctx->Const.MaxLineWidthAA);
    }
    else {
-      raster.line_width = CLAMP(ctx->Line.Width,
+      raster->line_width = CLAMP(ctx->Line.Width,
                                 ctx->Const.MinLineWidth,
                                 ctx->Const.MaxLineWidth);
    }
 
-   raster.line_stipple_enable = ctx->Line.StippleFlag;
-   raster.line_stipple_pattern = ctx->Line.StipplePattern;
+   raster->line_stipple_enable = ctx->Line.StippleFlag;
+   raster->line_stipple_pattern = ctx->Line.StipplePattern;
    /* GL stipple factor is in [1,256], remap to [0, 255] here */
-   raster.line_stipple_factor = ctx->Line.StippleFactor - 1;
+   raster->line_stipple_factor = ctx->Line.StippleFactor - 1;
 
    /* _NEW_MULTISAMPLE */
    if (ctx->Multisample.Enabled)
-      raster.multisample = 1;
+      raster->multisample = 1;
 
    /* _NEW_SCISSOR */
    if (ctx->Scissor.Enabled)
-      raster.scissor = 1;
+      raster->scissor = 1;
 
-   cso = st_cached_rasterizer_state(st, &raster);
-   if (st->state.rasterizer != cso) {
-      st->state.rasterizer = cso;
-      st->pipe->bind_rasterizer_state(st->pipe, cso->data);
-   }
+   cso_set_rasterizer(st->cso_context, raster);
 }
 
 const struct st_tracked_state st_update_rasterizer = {
diff --git a/src/mesa/state_tracker/st_atom_sampler.c b/src/mesa/state_tracker/st_atom_sampler.c
index 92263cb688..5787a7492c 100644
--- a/src/mesa/state_tracker/st_atom_sampler.c
+++ b/src/mesa/state_tracker/st_atom_sampler.c
@@ -33,11 +33,11 @@
  
 
 #include "st_context.h"
-#include "st_cache.h"
 #include "st_atom.h"
 #include "st_program.h"
 #include "pipe/p_context.h"
 #include "pipe/p_defines.h"
+#include "cso_cache/cso_context.h"
 
 
 /**
@@ -120,12 +120,13 @@ update_samplers(struct st_context *st)
    const struct st_fragment_program *fs = st->fp;
    GLuint su;
 
+   st->state.num_samplers = 0;
+
    /* loop over sampler units (aka tex image units) */
    for (su = 0; su < st->ctx->Const.MaxTextureImageUnits; su++) {
-      struct pipe_sampler_state sampler;
-      const struct cso_sampler *cso;
+      struct pipe_sampler_state *sampler = st->state.samplers + su;
 
-      memset(&sampler, 0, sizeof(sampler));
+      memset(sampler, 0, sizeof(*sampler));
 
       if (fs->Base.Base.SamplersUsed & (1 << su)) {
          GLuint texUnit = fs->Base.Base.SamplerUnits[su];
@@ -134,51 +135,56 @@ update_samplers(struct st_context *st)
 
          assert(texobj);
 
-         sampler.wrap_s = gl_wrap_to_sp(texobj->WrapS);
-         sampler.wrap_t = gl_wrap_to_sp(texobj->WrapT);
-         sampler.wrap_r = gl_wrap_to_sp(texobj->WrapR);
+         sampler->wrap_s = gl_wrap_to_sp(texobj->WrapS);
+         sampler->wrap_t = gl_wrap_to_sp(texobj->WrapT);
+         sampler->wrap_r = gl_wrap_to_sp(texobj->WrapR);
 
-         sampler.min_img_filter = gl_filter_to_img_filter(texobj->MinFilter);
-         sampler.min_mip_filter = gl_filter_to_mip_filter(texobj->MinFilter);
-         sampler.mag_img_filter = gl_filter_to_img_filter(texobj->MagFilter);
+         sampler->min_img_filter = gl_filter_to_img_filter(texobj->MinFilter);
+         sampler->min_mip_filter = gl_filter_to_mip_filter(texobj->MinFilter);
+         sampler->mag_img_filter = gl_filter_to_img_filter(texobj->MagFilter);
 
          if (texobj->Target != GL_TEXTURE_RECTANGLE_ARB)
-            sampler.normalized_coords = 1;
+            sampler->normalized_coords = 1;
 
-         sampler.lod_bias = st->ctx->Texture.Unit[su].LodBias;
+         sampler->lod_bias = st->ctx->Texture.Unit[su].LodBias;
 #if 1
-         sampler.min_lod = (texobj->MinLod) < 0.0 ? 0.0 : texobj->MinLod;
-         sampler.max_lod = texobj->MaxLod;
+         sampler->min_lod = (texobj->MinLod) < 0.0 ? 0.0 : texobj->MinLod;
+         sampler->max_lod = texobj->MaxLod;
 #else
          /* min/max lod should really be as follows (untested).
           * Also, calculate_first_last_level() needs to be overhauled
           * since today's hardware had real support for LOD clamping.
           */
-         sampler.min_lod = MAX2(texobj->BaseLevel, texobj->MinLod);
-         sampler.max_lod = MIN2(texobj->MaxLevel, texobj->MaxLod);
+         sampler->min_lod = MAX2(texobj->BaseLevel, texobj->MinLod);
+         sampler->max_lod = MIN2(texobj->MaxLevel, texobj->MaxLod);
 #endif
 
-	 sampler.max_anisotropy = texobj->MaxAnisotropy;
+	 sampler->max_anisotropy = texobj->MaxAnisotropy;
+         if (sampler->max_anisotropy > 1.0) {
+            sampler->min_img_filter = PIPE_TEX_FILTER_ANISO;
+            sampler->mag_img_filter = PIPE_TEX_FILTER_ANISO;
+         }
 
          /* only care about ARB_shadow, not SGI shadow */
          if (texobj->CompareMode == GL_COMPARE_R_TO_TEXTURE) {
-            sampler.compare = 1;
-            sampler.compare_mode = PIPE_TEX_COMPARE_R_TO_TEXTURE;
-            sampler.compare_func
+            sampler->compare = 1;
+            sampler->compare_mode = PIPE_TEX_COMPARE_R_TO_TEXTURE;
+            sampler->compare_func
                = st_compare_func_to_pipe(texobj->CompareFunc);
          }
 
-         /* XXX more sampler state here */
-      }
+         st->state.num_samplers = su + 1;
 
-      cso = st_cached_sampler_state(st, &sampler);
+         /* XXX more sampler state here */
 
-      if (cso != st->state.sampler[su]) {
-         /* state has changed */
-         st->state.sampler[su] = cso;
-         st->pipe->bind_sampler_state(st->pipe, su, cso->data);
+         cso_single_sampler(st->cso_context, su, sampler);
+      }
+      else {
+         cso_single_sampler(st->cso_context, su, NULL);
       }
    }
+
+   cso_single_sampler_done(st->cso_context);
 }
 
 
diff --git a/src/mesa/state_tracker/st_atom_shader.c b/src/mesa/state_tracker/st_atom_shader.c
index 10c131d554..0726688493 100644
--- a/src/mesa/state_tracker/st_atom_shader.c
+++ b/src/mesa/state_tracker/st_atom_shader.c
@@ -43,10 +43,9 @@
 #include "pipe/p_context.h"
 #include "pipe/p_shader_tokens.h"
 
-#include "cso_cache/cso_cache.h"
+#include "cso_cache/cso_context.h"
 
 #include "st_context.h"
-#include "st_cache.h"
 #include "st_atom.h"
 #include "st_program.h"
 #include "st_atom_shader.h"
@@ -70,9 +69,6 @@ struct translated_vertex_program
    /** Maps VERT_RESULT_x to slot */
    GLuint output_to_slot[VERT_RESULT_MAX];
 
-   /** The program in TGSI format */
-   struct tgsi_token tokens[ST_MAX_SHADER_TOKENS];
-
    /** Pointer to the translated vertex program */
    struct st_vertex_program *vp;
 
@@ -158,7 +154,7 @@ find_translated_vp(struct st_context *st,
    /*
     * Translate fragment program if needed.
     */
-   if (!stfp->cso) {
+   if (!stfp->state.tokens) {
       GLuint inAttr, numIn = 0;
 
       for (inAttr = 0; inAttr < FRAG_ATTRIB_MAX; inAttr++) {
@@ -175,11 +171,7 @@ find_translated_vp(struct st_context *st,
 
       assert(stfp->Base.Base.NumInstructions > 1);
 
-      (void) st_translate_fragment_program(st, stfp,
-                                           stfp->input_to_slot,
-                                           stfp->tokens,
-                                           ST_MAX_SHADER_TOKENS);
-      assert(stfp->cso);
+      st_translate_fragment_program(st, stfp, stfp->input_to_slot);
    }
 
 
@@ -261,12 +253,8 @@ find_translated_vp(struct st_context *st,
 
       assert(stvp->Base.Base.NumInstructions > 1);
 
-      st_translate_vertex_program(st, stvp,
-                                  xvp->output_to_slot,
-                                  xvp->tokens,
-                                  ST_MAX_SHADER_TOKENS);
+      st_translate_vertex_program(st, stvp, xvp->output_to_slot);
 
-      assert(stvp->cso);
       xvp->vp = stvp;
 
       /* translated VP is up to date now */
@@ -296,12 +284,10 @@ update_linkage( struct st_context *st )
    xvp = find_translated_vp(st, stvp, stfp);
 
    st->vp = stvp;
-   st->state.vs = xvp->vp;
-   st->pipe->bind_vs_state(st->pipe, st->state.vs->cso->data);
-
    st->fp = stfp;
-   st->state.fs = stfp->cso;
-   st->pipe->bind_fs_state(st->pipe, st->state.fs->data);
+
+   st->pipe->bind_vs_state(st->pipe, stvp->driver_shader);
+   st->pipe->bind_fs_state(st->pipe, stfp->driver_shader);
 
    st->vertex_result_to_slot = xvp->output_to_slot;
 }
diff --git a/src/mesa/state_tracker/st_atom_texture.c b/src/mesa/state_tracker/st_atom_texture.c
index 697d2cdfb4..e53a897637 100644
--- a/src/mesa/state_tracker/st_atom_texture.c
+++ b/src/mesa/state_tracker/st_atom_texture.c
@@ -37,6 +37,7 @@
 #include "st_texture.h"
 #include "st_cb_texture.h"
 #include "pipe/p_context.h"
+#include "pipe/p_inlines.h"
 
 
 /**
@@ -51,6 +52,8 @@ update_textures(struct st_context *st)
    struct gl_fragment_program *fprog = st->ctx->FragmentProgram._Current;
    GLuint unit;
 
+   st->state.num_textures = 0;
+
    for (unit = 0; unit < st->ctx->Const.MaxTextureCoordUnits; unit++) {
       const GLuint su = fprog->Base.SamplerUnits[unit];
       struct gl_texture_object *texObj = st->ctx->Texture.Unit[su]._Current;
@@ -62,6 +65,8 @@ update_textures(struct st_context *st)
 
          retval = st_finalize_texture(st->ctx, st->pipe, texObj, &flush);
          /* XXX retval indicates whether there's a texture border */
+
+         st->state.num_textures = unit + 1;
       }
 
       /* XXX: need to ensure that textures are unbound/removed from
@@ -70,18 +75,16 @@ update_textures(struct st_context *st)
        */
 
       pt = st_get_stobj_texture(stObj);
-
-      if (st->state.sampler_texture[unit] != pt) {
-         st->state.sampler_texture[unit] = pt;
-         st->pipe->set_sampler_texture(st->pipe, unit, pt);
-      }
+      pipe_texture_reference(&st->state.sampler_texture[unit], pt);
 
       if (stObj && stObj->dirtyData) {
          st->pipe->texture_update(st->pipe, pt);
          stObj->dirtyData = GL_FALSE;
       }
-
    }
+
+   st->pipe->set_sampler_textures(st->pipe, st->state.num_textures,
+                                  st->state.sampler_texture);
 }
 
 
diff --git a/src/mesa/state_tracker/st_cb_accum.c b/src/mesa/state_tracker/st_cb_accum.c
index 663c4f205d..f1fddc4e02 100644
--- a/src/mesa/state_tracker/st_cb_accum.c
+++ b/src/mesa/state_tracker/st_cb_accum.c
@@ -35,7 +35,6 @@
 #include "main/macros.h"
 
 #include "st_context.h"
-#include "st_cache.h"
 #include "st_cb_accum.h"
 #include "st_cb_fbo.h"
 #include "st_draw.h"
diff --git a/src/mesa/state_tracker/st_cb_clear.c b/src/mesa/state_tracker/st_cb_clear.c
index e712fd84cd..cc8a136292 100644
--- a/src/mesa/state_tracker/st_cb_clear.c
+++ b/src/mesa/state_tracker/st_cb_clear.c
@@ -35,7 +35,6 @@
 #include "main/macros.h"
 #include "shader/prog_instruction.h"
 #include "st_atom.h"
-#include "st_cache.h"
 #include "st_context.h"
 #include "st_cb_accum.h"
 #include "st_cb_clear.h"
@@ -50,6 +49,8 @@
 #include "pipe/p_defines.h"
 #include "pipe/p_winsys.h"
 
+#include "cso_cache/cso_context.h"
+
 
 
 
@@ -157,8 +158,7 @@ make_frag_shader(struct st_context *st)
    p->OutputsWritten = (1 << FRAG_RESULT_COLR);
 
    stfp = (struct st_fragment_program *) p;
-   st_translate_fragment_program(st, stfp, NULL,
-                                 stfp->tokens, ST_MAX_SHADER_TOKENS);
+   st_translate_fragment_program(st, stfp, NULL);
 
    return stfp;
 }
@@ -206,9 +206,10 @@ make_vertex_shader(struct st_context *st)
                         (1 << VERT_RESULT_HPOS));
 
    stvp = (struct st_vertex_program *) p;
-   st_translate_vertex_program(st, stvp, NULL,
-                               stvp->tokens, ST_MAX_SHADER_TOKENS);
+   st_translate_vertex_program(st, stvp, NULL);
+#if 0
    assert(stvp->cso);
+#endif
 
    return stvp;
 }
@@ -284,7 +285,6 @@ clear_with_quad(GLcontext *ctx,
    /* blend state: RGBA masking */
    {
       struct pipe_blend_state blend;
-      const struct cso_blend *cso;
       memset(&blend, 0, sizeof(blend));
       if (color) {
          if (ctx->Color.ColorMask[0])
@@ -298,14 +298,12 @@ clear_with_quad(GLcontext *ctx,
          if (st->ctx->Color.DitherFlag)
             blend.dither = 1;
       }
-      cso = st_cached_blend_state(st, &blend);
-      pipe->bind_blend_state(pipe, cso->data);
+      cso_set_blend(st->cso_context, &blend);
    }
 
    /* depth_stencil state: always pass/set to ref value */
    {
       struct pipe_depth_stencil_alpha_state depth_stencil;
-      const struct cso_depth_stencil_alpha *cso;
       memset(&depth_stencil, 0, sizeof(depth_stencil));
       if (depth) {
          depth_stencil.depth.enabled = 1;
@@ -323,14 +321,13 @@ clear_with_quad(GLcontext *ctx,
          depth_stencil.stencil[0].value_mask = 0xff;
          depth_stencil.stencil[0].write_mask = ctx->Stencil.WriteMask[0] & 0xff;
       }
-      cso = st_cached_depth_stencil_alpha_state(st, &depth_stencil);
-      pipe->bind_depth_stencil_alpha_state(pipe, cso->data);
+
+      cso_set_depth_stencil_alpha(st->cso_context, &depth_stencil);
    }
 
    /* rasterizer state: nothing */
    {
       struct pipe_rasterizer_state raster;
-      const struct cso_rasterizer *cso;
       memset(&raster, 0, sizeof(raster));
 #if 0
       /* don't do per-pixel scissor; we'll just draw a PIPE_PRIM_QUAD
@@ -339,8 +336,7 @@ clear_with_quad(GLcontext *ctx,
       if (ctx->Scissor.Enabled)
          raster.scissor = 1;
 #endif
-      cso = st_cached_rasterizer_state(st, &raster);
-      pipe->bind_rasterizer_state(pipe, cso->data);
+      cso_set_rasterizer(st->cso_context, &raster);
    }
 
    /* fragment shader state: color pass-through program */
@@ -349,7 +345,7 @@ clear_with_quad(GLcontext *ctx,
       if (!stfp) {
          stfp = make_frag_shader(st);
       }
-      pipe->bind_fs_state(pipe, stfp->cso->data);
+      pipe->bind_fs_state(pipe, stfp->driver_shader);
    }
 
    /* vertex shader state: color/position pass-through */
@@ -358,7 +354,7 @@ clear_with_quad(GLcontext *ctx,
       if (!stvp) {
          stvp = make_vertex_shader(st);
       }
-      pipe->bind_vs_state(pipe, stvp->cso->data);
+      pipe->bind_vs_state(pipe, stvp->driver_shader);
    }
 
    /* viewport state: viewport matching window dims */
@@ -380,16 +376,19 @@ clear_with_quad(GLcontext *ctx,
    /* draw quad matching scissor rect (XXX verify coord round-off) */
    draw_quad(ctx, x0, y0, x1, y1, ctx->Depth.Clear, ctx->Color.ClearColor);
 
+#if 0
+   /* Can't depend on old state objects still existing -- may have
+    * been deleted to make room in the hash, etc.  (Should get
+    * fixed...)
+    */
+   st_invalidate_state(ctx, _NEW_COLOR | _NEW_DEPTH | _NEW_STENCIL);
+#else
    /* Restore pipe state */
-   pipe->bind_blend_state(pipe, st->state.blend->data);
-   pipe->bind_depth_stencil_alpha_state(pipe, st->state.depth_stencil->data);
-   pipe->bind_fs_state(pipe, st->state.fs->data);
-   pipe->bind_vs_state(pipe, st->state.vs->cso->data);
-   pipe->bind_rasterizer_state(pipe, st->state.rasterizer->data);
+   cso_set_rasterizer(st->cso_context, &st->state.rasterizer);
+   pipe->bind_fs_state(pipe, st->fp->driver_shader);
+   pipe->bind_vs_state(pipe, st->vp->driver_shader);
+#endif
    pipe->set_viewport_state(pipe, &st->state.viewport);
-   /* OR:
-   st_invalidate_state(ctx, _NEW_COLOR | _NEW_DEPTH | _NEW_STENCIL);
-   */
 }
 
 
@@ -545,6 +544,15 @@ clear_stencil_buffer(GLcontext *ctx, struct gl_renderbuffer *rb)
 
       /* simple clear of whole buffer */
       GLuint clearValue = ctx->Stencil.Clear;
+
+      switch (strb->surface->format) {
+      case PIPE_FORMAT_S8Z24_UNORM:
+         clearValue <<= 24;
+         break;
+      default:
+         ; /* no-op, stencil value is in least significant bits */
+      }  
+
       ctx->st->pipe->clear(ctx->st->pipe, strb->surface, clearValue);
    }
 }
diff --git a/src/mesa/state_tracker/st_cb_drawpixels.c b/src/mesa/state_tracker/st_cb_drawpixels.c
index 65c9fda9cb..18ec9645c4 100644
--- a/src/mesa/state_tracker/st_cb_drawpixels.c
+++ b/src/mesa/state_tracker/st_cb_drawpixels.c
@@ -41,7 +41,6 @@
 #include "st_context.h"
 #include "st_atom.h"
 #include "st_atom_constbuf.h"
-#include "st_cache.h"
 #include "st_draw.h"
 #include "st_program.h"
 #include "st_cb_drawpixels.h"
@@ -58,6 +57,7 @@
 #include "pipe/p_winsys.h"
 #include "util/p_tile.h"
 #include "shader/prog_instruction.h"
+#include "cso_cache/cso_context.h"
 
 
 /**
@@ -159,8 +159,7 @@ make_bitmap_fragment_program(GLcontext *ctx)
 
    stfp = (struct st_fragment_program *) p;
    stfp->Base.UsesKill = GL_TRUE;
-   st_translate_fragment_program(ctx->st, stfp, NULL,
-                                 stfp->tokens, ST_MAX_SHADER_TOKENS);
+   st_translate_fragment_program(ctx->st, stfp, NULL);
 
    return stfp;
 }
@@ -204,8 +203,7 @@ combined_bitmap_fragment_program(GLcontext *ctx)
 #endif
 
       /* translate to TGSI tokens */
-      st_translate_fragment_program(st, stfp, NULL,
-                                    stfp->tokens, ST_MAX_SHADER_TOKENS);
+      st_translate_fragment_program(st, stfp, NULL);
 
       /* save new program, update serial numbers */
       st->bitmap.user_prog_sn = st->fp->serialNo;
@@ -266,8 +264,7 @@ combined_drawpix_fragment_program(GLcontext *ctx)
 #endif
 
       /* translate to TGSI tokens */
-      st_translate_fragment_program(st, stfp, NULL,
-                                    stfp->tokens, ST_MAX_SHADER_TOKENS);
+      st_translate_fragment_program(st, stfp, NULL);
 
       /* save new program, update serial numbers */
       st->pixel_xfer.xfer_prog_sn = st->pixel_xfer.program->serialNo;
@@ -343,8 +340,7 @@ make_fragment_shader_z(struct st_context *st)
    p->OutputsWritten = (1 << FRAG_RESULT_COLR) | (1 << FRAG_RESULT_DEPR);
 
    stfp = (struct st_fragment_program *) p;
-   st_translate_fragment_program(st, stfp, NULL,
-                                 stfp->tokens, ST_MAX_SHADER_TOKENS);
+   st_translate_fragment_program(st, stfp, NULL);
 
    return stfp;
 }
@@ -421,8 +417,7 @@ st_make_passthrough_vertex_shader(struct st_context *st, GLboolean passColor)
    }
 
    stvp = (struct st_vertex_program *) p;
-   st_translate_vertex_program(st, stvp, NULL,
-                               stvp->tokens, ST_MAX_SHADER_TOKENS);
+   st_translate_vertex_program(st, stvp, NULL);
 
    progs[passColor] = stvp;
 
@@ -641,8 +636,9 @@ draw_textured_quad(GLcontext *ctx, GLint x, GLint y, GLfloat z,
                    const GLfloat *color,
                    GLboolean invertTex)
 {
-   const GLuint unit = 0;
+   struct st_context *st = ctx->st;
    struct pipe_context *pipe = ctx->st->pipe;
+   struct cso_context *cso = ctx->st->cso_context;
    GLfloat x0, y0, x1, y1;
    GLuint maxSize;
 
@@ -657,24 +653,23 @@ draw_textured_quad(GLcontext *ctx, GLint x, GLint y, GLfloat z,
    /* setup state: just scissor */
    {
       struct pipe_rasterizer_state  setup;
-      const struct cso_rasterizer  *cso;
       memset(&setup, 0, sizeof(setup));
       if (ctx->Scissor.Enabled)
          setup.scissor = 1;
-      cso = st_cached_rasterizer_state(ctx->st, &setup);
-      pipe->bind_rasterizer_state(pipe, cso->data);
+
+      cso_set_rasterizer(cso, &setup);
    }
 
    /* fragment shader state: TEX lookup program */
-   pipe->bind_fs_state(pipe, stfp->cso->data);
+   pipe->bind_fs_state(pipe, stfp->driver_shader);
 
    /* vertex shader state: position + texcoord pass-through */
-   pipe->bind_vs_state(pipe, stvp->cso->data);
+   pipe->bind_vs_state(pipe, stvp->driver_shader);
+
 
    /* texture sampling state: */
    {
       struct pipe_sampler_state sampler;
-      const struct cso_sampler *cso;
       memset(&sampler, 0, sizeof(sampler));
       sampler.wrap_s = PIPE_TEX_WRAP_CLAMP;
       sampler.wrap_t = PIPE_TEX_WRAP_CLAMP;
@@ -683,8 +678,9 @@ draw_textured_quad(GLcontext *ctx, GLint x, GLint y, GLfloat z,
       sampler.min_mip_filter = PIPE_TEX_MIPFILTER_NONE;
       sampler.mag_img_filter = PIPE_TEX_FILTER_NEAREST;
       sampler.normalized_coords = 1;
-      cso = st_cached_sampler_state(ctx->st, &sampler);
-      pipe->bind_sampler_state(pipe, unit, cso->data);
+
+      cso_single_sampler(cso, 0, &sampler);
+      cso_single_sampler_done(cso);
    }
 
    /* viewport state: viewport matching window dims */
@@ -705,7 +701,7 @@ draw_textured_quad(GLcontext *ctx, GLint x, GLint y, GLfloat z,
 
    /* texture state: */
    {
-      pipe->set_sampler_texture(pipe, unit, pt);
+      pipe->set_sampler_textures(pipe, 1, &pt);
    }
 
    /* Compute window coords (y=0=bottom) with pixel zoom.
@@ -724,12 +720,25 @@ draw_textured_quad(GLcontext *ctx, GLint x, GLint y, GLfloat z,
       draw_quad(ctx, x0, y0, z, x1, y1, invertTex);
 
    /* restore GL state */
-   pipe->bind_rasterizer_state(pipe, ctx->st->state.rasterizer->data);
-   pipe->bind_fs_state(pipe, ctx->st->state.fs->data);
-   pipe->bind_vs_state(pipe, ctx->st->state.vs->cso->data);
-   pipe->set_sampler_texture(pipe, unit, ctx->st->state.sampler_texture[unit]);
-   pipe->bind_sampler_state(pipe, unit, ctx->st->state.sampler[unit]->data);
+   pipe->set_sampler_textures(pipe, ctx->st->state.num_textures,
+                              ctx->st->state.sampler_texture);
+
    pipe->set_viewport_state(pipe, &ctx->st->state.viewport);
+
+#if 0
+   /* Can't depend on old state objects still existing -- may have
+    * been deleted to make room in the hash, etc.  (Should get
+    * fixed...)
+    */
+   st_invalidate_state(ctx, _NEW_COLOR | _NEW_TEXTURE);
+#else
+   /* restore state */
+   pipe->bind_fs_state(pipe, st->fp->driver_shader);
+   pipe->bind_vs_state(pipe, st->vp->driver_shader);
+   cso_set_rasterizer(cso, &st->state.rasterizer);
+   cso_set_samplers(cso, PIPE_MAX_SAMPLERS,
+                 (const struct pipe_sampler_state **) st->state.sampler_list);
+#endif
 }
 
 
@@ -797,10 +806,10 @@ compatible_formats(GLenum format, GLenum type, enum pipe_format pipeFormat)
 static GLboolean
 any_fragment_ops(const struct st_context *st)
 {
-   if (st->state.depth_stencil->state.alpha.enabled ||
-       st->state.blend->state.blend_enable ||
-       st->state.blend->state.logicop_enable ||
-       st->state.depth_stencil->state.depth.enabled)
+   if (st->state.depth_stencil.alpha.enabled ||
+       st->state.depth_stencil.depth.enabled ||
+       st->state.blend.blend_enable ||
+       st->state.blend.logicop_enable)
       /* XXX more checks */
       return GL_TRUE;
    else
diff --git a/src/mesa/state_tracker/st_cb_program.c b/src/mesa/state_tracker/st_cb_program.c
index 61d4f4c41c..4dc76f19b1 100644
--- a/src/mesa/state_tracker/st_cb_program.c
+++ b/src/mesa/state_tracker/st_cb_program.c
@@ -168,11 +168,13 @@ static void st_program_string_notify( GLcontext *ctx,
 
       stfp->serialNo++;
 
+#if 0
       if (stfp->cso) {
          /* free the TGSI code */
          // cso_delete(stfp->vs);
          stfp->cso = NULL;
       }
+#endif
 
       stfp->param_state = stfp->Base.Base.Parameters->StateFlags;
 
@@ -184,13 +186,14 @@ static void st_program_string_notify( GLcontext *ctx,
 
       stvp->serialNo++;
 
+#if 0
       if (stvp->cso) {
          /* free the CSO data */
          st->pipe->delete_vs_state(st->pipe, stvp->cso->data);
          FREE((void *) stvp->cso);
          stvp->cso = NULL;
       }
-
+#endif
       if (stvp->draw_shader) {
          draw_delete_vertex_shader(st->draw, stvp->draw_shader);
          stvp->draw_shader = NULL;
diff --git a/src/mesa/state_tracker/st_cb_texture.c b/src/mesa/state_tracker/st_cb_texture.c
index 1ba3173312..d4731c7737 100644
--- a/src/mesa/state_tracker/st_cb_texture.c
+++ b/src/mesa/state_tracker/st_cb_texture.c
@@ -529,7 +529,8 @@ st_TexImage(GLcontext * ctx,
 	 texImage->RowStride = postConvWidth;
       }
       
-      assert(texImage->RowStride == postConvWidth);
+      /* we'll set RowStride elsewhere when the texture is a "mapped" state */
+      /*assert(texImage->RowStride == postConvWidth);*/
    }
 
    /* Release the reference to a potentially orphaned buffer.   
diff --git a/src/mesa/state_tracker/st_context.c b/src/mesa/state_tracker/st_context.c
index 09e389f9dc..5458ab420e 100644
--- a/src/mesa/state_tracker/st_context.c
+++ b/src/mesa/state_tracker/st_context.c
@@ -56,6 +56,7 @@
 #include "pipe/p_inlines.h"
 #include "draw/draw_context.h"
 #include "cso_cache/cso_cache.h"
+#include "cso_cache/cso_context.h"
 
 
 /**
@@ -78,6 +79,7 @@ void st_invalidate_state(GLcontext * ctx, GLuint new_state)
 static struct st_context *
 st_create_context_priv( GLcontext *ctx, struct pipe_context *pipe )
 {
+   uint i;
    struct st_context *st = CALLOC_STRUCT( st_context );
    
    ctx->st = st;
@@ -93,12 +95,15 @@ st_create_context_priv( GLcontext *ctx, struct pipe_context *pipe )
    st->dirty.mesa = ~0;
    st->dirty.st = ~0;
 
-   st->cache = cso_cache_create();
+   st->cso_context = cso_create_context(pipe);
 
    st_init_atoms( st );
    st_init_draw( st );
    st_init_generate_mipmap(st);
 
+   for (i = 0; i < PIPE_MAX_SAMPLERS; i++)
+      st->state.sampler_list[i] = &st->state.samplers[i];
+
    /* we want all vertex data to be placed in buffer objects */
    vbo_use_buffer_objects(ctx);
 
@@ -149,7 +154,7 @@ static void st_destroy_context_priv( struct st_context *st )
 
    _vbo_DestroyContext(st->ctx);
 
-   cso_cache_delete( st->cache );
+   cso_destroy_context(st->cso_context);
 
    _mesa_delete_program_cache(st->ctx, st->pixel_xfer.cache);
 
diff --git a/src/mesa/state_tracker/st_context.h b/src/mesa/state_tracker/st_context.h
index 1fbf9721e7..e81aebba3d 100644
--- a/src/mesa/state_tracker/st_context.h
+++ b/src/mesa/state_tracker/st_context.h
@@ -74,14 +74,11 @@ struct st_context
     * Other state is just parameter values.
     */
    struct {
-      const struct cso_alpha_test      *alpha_test;
-      const struct cso_blend           *blend;
-      const struct cso_sampler         *sampler[PIPE_MAX_SAMPLERS];
-      const struct cso_depth_stencil_alpha   *depth_stencil;
-      const struct cso_rasterizer      *rasterizer;
-      const struct cso_fragment_shader *fs;
-      struct st_vertex_program   *vs;
-
+      struct pipe_blend_state               blend;
+      struct pipe_depth_stencil_alpha_state depth_stencil;
+      struct pipe_rasterizer_state          rasterizer;
+      struct pipe_sampler_state             samplers[PIPE_MAX_SAMPLERS];
+      struct pipe_sampler_state             *sampler_list[PIPE_MAX_SAMPLERS];
       struct pipe_blend_color  blend_color;
       struct pipe_clip_state clip;
       struct pipe_constant_buffer constants[2];
@@ -90,6 +87,9 @@ struct st_context
       struct pipe_poly_stipple poly_stipple;
       struct pipe_scissor_state scissor;
       struct pipe_viewport_state viewport;
+
+      GLuint num_samplers;
+      GLuint num_textures;
    } state;
 
    struct {
@@ -148,6 +148,10 @@ struct st_context
 
    /** For gen/render mipmap feature */
    struct {
+      struct pipe_blend_state blend;
+      struct pipe_depth_stencil_alpha_state depthstencil;
+      struct pipe_rasterizer_state rasterizer;
+
       void *blend_cso;
       void *depthstencil_cso;
       void *rasterizer_cso;
@@ -155,7 +159,7 @@ struct st_context
       struct st_vertex_program *stvp;
    } gen_mipmap;
 
-   struct cso_cache *cache;
+   struct cso_context *cso_context;
 };
 
 
diff --git a/src/mesa/state_tracker/st_debug.c b/src/mesa/state_tracker/st_debug.c
index 5888bcb98a..9c13010da8 100644
--- a/src/mesa/state_tracker/st_debug.c
+++ b/src/mesa/state_tracker/st_debug.c
@@ -53,15 +53,15 @@ st_print_current(void)
    int i;
 
    printf("Vertex Transform Inputs:\n");
-   for (i = 0; i < st->state.vs->cso->state.num_inputs; i++) {
+   for (i = 0; i < st->vp->state.num_inputs; i++) {
       printf("  Slot %d:  VERT_ATTRIB_%d\n", i, st->vp->index_to_input[i]);
    }
 
-   tgsi_dump( st->state.vs->cso->state.tokens, 0 );
+   tgsi_dump( st->vp->state.tokens, 0 );
    if (st->vp->Base.Base.Parameters)
       _mesa_print_parameter_list(st->vp->Base.Base.Parameters);
 
-   tgsi_dump( st->state.fs->state.tokens, 0 );
+   tgsi_dump( st->fp->state.tokens, 0 );
    if (st->fp->Base.Base.Parameters)
       _mesa_print_parameter_list(st->fp->Base.Base.Parameters);
 }
diff --git a/src/mesa/state_tracker/st_draw.c b/src/mesa/state_tracker/st_draw.c
index 1c0fa8c6aa..98504e46c1 100644
--- a/src/mesa/state_tracker/st_draw.c
+++ b/src/mesa/state_tracker/st_draw.c
@@ -36,7 +36,6 @@
 #include "vbo/vbo.h"
 
 #include "st_atom.h"
-#include "st_cache.h"
 #include "st_context.h"
 #include "st_cb_bufferobjects.h"
 #include "st_draw.h"
@@ -219,7 +218,7 @@ st_draw_vbo(GLcontext *ctx,
 
    /* must get these after state validation! */
    vp = ctx->st->vp;
-   vs = &ctx->st->state.vs->cso->state;
+   vs = &ctx->st->vp->state;
 
    /* loop over TGSI shader inputs to determine vertex buffer
     * and attribute info
@@ -481,10 +480,10 @@ st_feedback_draw_vbo(GLcontext *ctx,
 
    /* must get these after state validation! */
    vp = ctx->st->vp;
-   vs = &ctx->st->state.vs->cso->state;
+   vs = &st->vp->state;
 
-   if (!st->state.vs->draw_shader) {
-      st->state.vs->draw_shader = draw_create_vertex_shader(draw, vs);
+   if (!st->vp->draw_shader) {
+      st->vp->draw_shader = draw_create_vertex_shader(draw, vs);
    }
 
    /*
@@ -496,8 +495,8 @@ st_feedback_draw_vbo(GLcontext *ctx,
    assert(draw);
    draw_set_viewport_state(draw, &st->state.viewport);
    draw_set_clip_state(draw, &st->state.clip);
-   draw_set_rasterizer_state(draw, &st->state.rasterizer->state);
-   draw_bind_vertex_shader(draw, st->state.vs->draw_shader);
+   draw_set_rasterizer_state(draw, &st->state.rasterizer);
+   draw_bind_vertex_shader(draw, st->vp->draw_shader);
    set_feedback_vertex_format(ctx);
 
    /* loop over TGSI shader inputs to determine vertex buffer
diff --git a/src/mesa/state_tracker/st_gen_mipmap.c b/src/mesa/state_tracker/st_gen_mipmap.c
index 841d77abbc..9c4e1032ef 100644
--- a/src/mesa/state_tracker/st_gen_mipmap.c
+++ b/src/mesa/state_tracker/st_gen_mipmap.c
@@ -38,6 +38,7 @@
 #include "pipe/p_inlines.h"
 #include "pipe/p_winsys.h"
 #include "cso_cache/cso_cache.h"
+#include "cso_cache/cso_context.h"
 
 #include "st_context.h"
 #include "st_draw.h"
@@ -89,8 +90,7 @@ make_tex_fragment_program(GLcontext *ctx)
 
    stfp = (struct st_fragment_program *) p;
 
-   st_translate_fragment_program(ctx->st, stfp, NULL,
-                                 stfp->tokens, ST_MAX_SHADER_TOKENS);
+   st_translate_fragment_program(ctx->st, stfp, NULL);
 
    return stfp;
 }
@@ -118,6 +118,7 @@ st_init_generate_mipmap(struct st_context *st)
    blend.rgb_dst_factor = PIPE_BLENDFACTOR_ZERO;
    blend.alpha_dst_factor = PIPE_BLENDFACTOR_ZERO;
    blend.colormask = PIPE_MASK_RGBA;
+   st->gen_mipmap.blend = blend;
    st->gen_mipmap.blend_cso = pipe->create_blend_state(pipe, &blend);
 
    memset(&depthstencil, 0, sizeof(depthstencil));
@@ -257,14 +258,14 @@ st_render_mipmap(struct st_context *st,
    sampler.normalized_coords = 1;
 
 
-   /* bind CSOs */
-   pipe->bind_blend_state(pipe, st->gen_mipmap.blend_cso);
-   pipe->bind_depth_stencil_alpha_state(pipe, st->gen_mipmap.depthstencil_cso);
-   pipe->bind_rasterizer_state(pipe, st->gen_mipmap.rasterizer_cso);
+   /* bind state */
+   cso_set_blend(st->cso_context, &st->gen_mipmap.blend);
+   cso_set_depth_stencil_alpha(st->cso_context, &st->gen_mipmap.depthstencil);
+   cso_set_rasterizer(st->cso_context, &st->gen_mipmap.rasterizer);
 
    /* bind shaders */
-   pipe->bind_fs_state(pipe, st->gen_mipmap.stfp->cso->data);
-   pipe->bind_vs_state(pipe, st->gen_mipmap.stvp->cso->data);
+   pipe->bind_fs_state(pipe, st->gen_mipmap.stfp->driver_shader);
+   pipe->bind_vs_state(pipe, st->gen_mipmap.stvp->driver_shader);
 
    /*
     * XXX for small mipmap levels, it may be faster to use the software
@@ -284,7 +285,7 @@ st_render_mipmap(struct st_context *st,
        */
       sampler.min_lod = sampler.max_lod = srcLevel;
       sampler_cso = pipe->create_sampler_state(pipe, &sampler);
-      pipe->bind_sampler_state(pipe, 0, sampler_cso);
+      pipe->bind_sampler_states(pipe, 1, &sampler_cso);
 
       simple_viewport(pipe, pt->width[dstLevel], pt->height[dstLevel]);
 
@@ -293,7 +294,7 @@ st_render_mipmap(struct st_context *st,
        * the right mipmap level.
        */
       /*pt->first_level = srcLevel;*/
-      pipe->set_sampler_texture(pipe, 0, pt);
+      pipe->set_sampler_textures(pipe, 1, &pt);
 
       draw_quad(st->ctx);
 
@@ -304,16 +305,18 @@ st_render_mipmap(struct st_context *st,
    /*pt->first_level = first_level_save;*/
 
    /* restore pipe state */
-   if (st->state.rasterizer)
-      pipe->bind_rasterizer_state(pipe, st->state.rasterizer->data);
-   if (st->state.fs)
-      pipe->bind_fs_state(pipe, st->state.fs->data);
-   if (st->state.vs)
-      pipe->bind_vs_state(pipe, st->state.vs->cso->data);
-   if (st->state.sampler[0])
-      pipe->bind_sampler_state(pipe, 0, st->state.sampler[0]->data);
-   pipe->set_sampler_texture(pipe, 0, st->state.sampler_texture[0]);
+#if 0
+   cso_set_rasterizer(st->cso_context, &st->state.rasterizer);
+   cso_set_samplers(st->cso_context, st->state.samplers_list);
+   pipe->bind_fs_state(pipe, st->fp->shader_program);
+   pipe->bind_vs_state(pipe, st->vp->shader_program);
+   pipe->set_sampler_textures(pipe, st->state.num_textures,
+                              st->state.sampler_texture);
    pipe->set_viewport_state(pipe, &st->state.viewport);
+#else
+   /* XXX is this sufficient? */
+   st_invalidate_state(st->ctx, _NEW_COLOR | _NEW_TEXTURE);
+#endif
 
    return TRUE;
 }
diff --git a/src/mesa/state_tracker/st_mesa_to_tgsi.c b/src/mesa/state_tracker/st_mesa_to_tgsi.c
index 97206752af..f3cfda0bfb 100644
--- a/src/mesa/state_tracker/st_mesa_to_tgsi.c
+++ b/src/mesa/state_tracker/st_mesa_to_tgsi.c
@@ -513,10 +513,9 @@ compile_instruction(
    case OPCODE_TXP:
       /* texture lookup with divide by Q component */
       /* convert to TEX w/ special flag for division */
-      fullinst->Instruction.Opcode = TGSI_OPCODE_TEX;
+      fullinst->Instruction.Opcode = TGSI_OPCODE_TXP;
       fullinst->Instruction.NumSrcRegs = 2;
       fullinst->InstructionExtTexture.Texture = map_texture_target( inst->TexSrcTarget );
-      fullinst->FullSrcRegisters[0].SrcRegisterExtSwz.ExtDivide = TGSI_EXTSWIZZLE_W;
       fullinst->FullSrcRegisters[1].SrcRegister.File = TGSI_FILE_SAMPLER;
       fullinst->FullSrcRegisters[1].SrcRegister.Index = inst->TexSrcUnit;
       break;
@@ -683,8 +682,9 @@ find_temporaries(const struct gl_program *program,
  * \param tokens  array to store translated tokens in
  * \param maxTokens  size of the tokens array
  *
+ * \return number of tokens placed in 'tokens' buffer, or zero if error
  */
-GLboolean
+GLuint
 tgsi_translate_mesa_program(
    uint procType,
    const struct gl_program *program,
@@ -918,6 +918,6 @@ tgsi_translate_mesa_program(
          maxTokens - ti );
    }
 
-   return GL_TRUE;
+   return ti;
 }
 
diff --git a/src/mesa/state_tracker/st_mesa_to_tgsi.h b/src/mesa/state_tracker/st_mesa_to_tgsi.h
index 3ababf1339..63fc855b53 100644
--- a/src/mesa/state_tracker/st_mesa_to_tgsi.h
+++ b/src/mesa/state_tracker/st_mesa_to_tgsi.h
@@ -39,7 +39,7 @@ extern "C" {
 struct tgsi_token;
 struct gl_program;
 
-GLboolean
+GLuint
 tgsi_translate_mesa_program(
    uint procType,
    const struct gl_program *program,
diff --git a/src/mesa/state_tracker/st_program.c b/src/mesa/state_tracker/st_program.c
index aa252c845a..0f8784e132 100644
--- a/src/mesa/state_tracker/st_program.c
+++ b/src/mesa/state_tracker/st_program.c
@@ -42,15 +42,29 @@
 #include "tgsi/util/tgsi_dump.h"
 
 #include "st_context.h"
-#include "st_cache.h"
 #include "st_atom.h"
 #include "st_program.h"
 #include "st_mesa_to_tgsi.h"
+#include "cso_cache/cso_context.h"
 
 
 #define TGSI_DEBUG 0
 
 
+/** XXX we should use the version of this from p_util.h but including
+ * that header causes symbol collisions.
+ */
+static INLINE void *
+mem_dup(const void *src, uint size)
+{
+   void *dup = MALLOC(size);
+   if (dup)
+      memcpy(dup, src, size);
+   return dup;
+}
+
+
+
 /**
  * Translate a Mesa vertex shader into a TGSI shader.
  * \param outputMapping  to map vertex program output registers to TGSI
@@ -61,15 +75,15 @@
 void
 st_translate_vertex_program(struct st_context *st,
                             struct st_vertex_program *stvp,
-                            const GLuint outputMapping[],
-                            struct tgsi_token *tokensOut,
-                            GLuint maxTokens)
+                            const GLuint outputMapping[])
 {
+   struct pipe_context *pipe = st->pipe;
+   struct tgsi_token tokens[ST_MAX_SHADER_TOKENS];
    GLuint defaultOutputMapping[VERT_RESULT_MAX];
    struct pipe_shader_state vs;
-   const struct cso_vertex_shader *cso;
    GLuint attr, i;
    GLuint num_generic = 0;
+   GLuint num_tokens;
 
    memset(&vs, 0, sizeof(vs));
 
@@ -240,7 +254,7 @@ st_translate_vertex_program(struct st_context *st,
 
    /* XXX: fix static allocation of tokens:
     */
-   tgsi_translate_mesa_program( TGSI_PROCESSOR_VERTEX,
+   num_tokens = tgsi_translate_mesa_program( TGSI_PROCESSOR_VERTEX,
                                 &stvp->Base.Base,
                                 /* inputs */
                                 vs.num_inputs,
@@ -252,20 +266,21 @@ st_translate_vertex_program(struct st_context *st,
                                 vs.num_outputs,
                                 outputMapping,
                                 vs.output_semantic_name,
-                                 vs.output_semantic_index,
+                                vs.output_semantic_index,
                                 /* tokenized result */
-                                tokensOut, maxTokens);
+                                tokens, ST_MAX_SHADER_TOKENS);
 
-   vs.tokens = tokensOut;
+   vs.tokens = (struct tgsi_token *)
+      mem_dup(tokens, num_tokens * sizeof(tokens[0]));
 
-   cso = st_cached_vs_state(st, &vs);
-   stvp->cso = cso;
+   stvp->state = vs; /* struct copy */
+   stvp->driver_shader = pipe->create_vs_state(pipe, &vs);
 
    if (0)
       _mesa_print_program(&stvp->Base.Base);
 
    if (TGSI_DEBUG)
-      tgsi_dump( tokensOut, 0 );
+      tgsi_dump( vs.tokens, 0 );
 }
 
 
@@ -280,10 +295,10 @@ st_translate_vertex_program(struct st_context *st,
 const struct cso_fragment_shader *
 st_translate_fragment_program(struct st_context *st,
                               struct st_fragment_program *stfp,
-                              const GLuint inputMapping[],
-                              struct tgsi_token *tokensOut,
-                              GLuint maxTokens)
+                              const GLuint inputMapping[])
 {
+   struct pipe_context *pipe = st->pipe;
+   struct tgsi_token tokens[ST_MAX_SHADER_TOKENS];
    GLuint outputMapping[FRAG_RESULT_MAX];
    GLuint defaultInputMapping[FRAG_ATTRIB_MAX];
    struct pipe_shader_state fs;
@@ -293,6 +308,7 @@ st_translate_fragment_program(struct st_context *st,
    const GLbitfield inputsRead = stfp->Base.Base.InputsRead;
    GLuint vslot = 0;
    GLuint num_generic = 0;
+   GLuint num_tokens;
 
    memset(&fs, 0, sizeof(fs));
 
@@ -401,7 +417,7 @@ st_translate_fragment_program(struct st_context *st,
 
    /* XXX: fix static allocation of tokens:
     */
-   tgsi_translate_mesa_program( TGSI_PROCESSOR_FRAGMENT,
+   num_tokens = tgsi_translate_mesa_program( TGSI_PROCESSOR_FRAGMENT,
                                 &stfp->Base.Base,
                                 /* inputs */
                                 fs.num_inputs,
@@ -415,18 +431,19 @@ st_translate_fragment_program(struct st_context *st,
                                 fs.output_semantic_name,
                                 fs.output_semantic_index,
                                 /* tokenized result */
-                                tokensOut, maxTokens);
+                                tokens, ST_MAX_SHADER_TOKENS);
 
-   fs.tokens = tokensOut;
+   fs.tokens = (struct tgsi_token *)
+      mem_dup(tokens, num_tokens * sizeof(tokens[0]));
 
-   cso = st_cached_fs_state(st, &fs);
-   stfp->cso = cso;
+   stfp->state = fs; /* struct copy */
+   stfp->driver_shader = pipe->create_fs_state(pipe, &fs);
 
    if (0)
       _mesa_print_program(&stfp->Base.Base);
 
    if (TGSI_DEBUG)
-      tgsi_dump( tokensOut, 0/*TGSI_DUMP_VERBOSE*/ );
+      tgsi_dump( fs.tokens, 0/*TGSI_DUMP_VERBOSE*/ );
 
    return cso;
 }
diff --git a/src/mesa/state_tracker/st_program.h b/src/mesa/state_tracker/st_program.h
index 31558af6ce..78786dcbb6 100644
--- a/src/mesa/state_tracker/st_program.h
+++ b/src/mesa/state_tracker/st_program.h
@@ -60,11 +60,8 @@ struct st_fragment_program
    /** map FP input back to VP output */
    GLuint input_map[PIPE_MAX_SHADER_INPUTS];
 
-   /** The program in TGSI format */
-   struct tgsi_token tokens[ST_MAX_SHADER_TOKENS];
-
-   /** Pointer to the corresponding cached shader */
-   const struct cso_fragment_shader *cso;
+   struct pipe_shader_state state;
+   struct pipe_shader_state *driver_shader;
 
    GLuint param_state;
 
@@ -88,11 +85,8 @@ struct st_vertex_program
    /** maps a TGSI input index back to a Mesa VERT_ATTRIB_x */
    GLuint index_to_input[PIPE_MAX_SHADER_INPUTS];
 
-   /** The program in TGSI format */
-   struct tgsi_token tokens[ST_MAX_SHADER_TOKENS];
-
-   /** Pointer to the corresponding cached shader */
-   const struct cso_vertex_shader *cso;
+   struct pipe_shader_state state;
+   struct pipe_shader_state *driver_shader;
 
    /** For using our private draw module (glRasterPos) */
    struct draw_vertex_shader *draw_shader;
@@ -122,16 +116,12 @@ st_vertex_program( struct gl_vertex_program *vp )
 extern const struct cso_fragment_shader *
 st_translate_fragment_program(struct st_context *st,
                               struct st_fragment_program *fp,
-                              const GLuint inputMapping[],
-                              struct tgsi_token *tokens,
-                              GLuint maxTokens);
+                              const GLuint inputMapping[]);
 
 
 extern void
 st_translate_vertex_program(struct st_context *st,
                             struct st_vertex_program *vp,
-                            const GLuint vert_output_to_slot[],
-                            struct tgsi_token *tokens,
-                            GLuint maxTokens);
+                            const GLuint vert_output_to_slot[]);
 
 #endif
diff --git a/src/mesa/vbo/vbo_exec_api.c b/src/mesa/vbo/vbo_exec_api.c
index b7f4d8a307..35dc0e768f 100644
--- a/src/mesa/vbo/vbo_exec_api.c
+++ b/src/mesa/vbo/vbo_exec_api.c
@@ -309,16 +309,23 @@ static void vbo_exec_fixup_vertex( GLcontext *ctx,
 				   GLuint attr, GLuint sz )
 {
    struct vbo_exec_context *exec = &vbo_context(ctx)->exec;
+   static const GLfloat id[4] = { 0, 0, 0, 1 };
    int i;
 
-   if (sz > exec->vtx.attrsz[attr]) {
+   if (exec->vtx.prim_count == 0) {
+      GLfloat *current = (GLfloat *)vbo_context(ctx)->currval[attr].Ptr;
+      exec->vtx.attrptr[attr] = current;
+      memcpy(current, id, sizeof(id));
+      ctx->Driver.NeedFlush |= FLUSH_UPDATE_CURRENT;
+      return;
+   }
+   else if (sz > exec->vtx.attrsz[attr]) {
       /* New size is larger.  Need to flush existing vertices and get
        * an enlarged vertex format.
        */
       vbo_exec_wrap_upgrade_vertex( exec, attr, sz );
    }
    else if (sz < exec->vtx.active_sz[attr]) {
-      static const GLfloat id[4] = { 0, 0, 0, 1 };
 
       /* New size is smaller - just need to fill in some
        * zeros.  Don't need to flush or wrap.
diff --git a/winddk.py b/winddk.py
index 845cf90a9a..c112329cd6 100644
--- a/winddk.py
+++ b/winddk.py
@@ -64,24 +64,23 @@ def get_winddk_paths(env, version=None):
     if 'BASEDIR' in os.environ:
         WINDDKdir = os.environ['BASEDIR']
     else:
-        #WINDDKdir = "C:\\WINDDK\\3790.1830"
-        WINDDKdir = "C:/WINDDK/3790.1830"
+        WINDDKdir = "C:\\WINDDK\\3790.1830"
 
     exe_paths.append( os.path.join(WINDDKdir, 'bin') )
-    exe_paths.append( os.path.join(WINDDKdir, 'bin/x86') )
-    include_paths.append( os.path.join(WINDDKdir, 'inc/wxp') )
+    exe_paths.append( os.path.join(WINDDKdir, 'bin', 'x86') )
+    include_paths.append( os.path.join(WINDDKdir, 'inc', 'wxp') )
     lib_paths.append( os.path.join(WINDDKdir, 'lib') )
 
     target_os = 'wxp'
     target_cpu = 'i386'
     
     env['SDK_INC_PATH'] = os.path.join(WINDDKdir, 'inc', target_os) 
-    env['CRT_INC_PATH'] = os.path.join(WINDDKdir, 'inc/crt') 
-    env['DDK_INC_PATH'] = os.path.join(WINDDKdir, 'inc/ddk', target_os) 
-    env['WDM_INC_PATH'] = os.path.join(WINDDKdir, 'inc/ddk/wdm', target_os) 
+    env['CRT_INC_PATH'] = os.path.join(WINDDKdir, 'inc', 'crt') 
+    env['DDK_INC_PATH'] = os.path.join(WINDDKdir, 'inc', 'ddk', target_os) 
+    env['WDM_INC_PATH'] = os.path.join(WINDDKdir, 'inc', 'ddk', 'wdm', target_os) 
 
     env['SDK_LIB_PATH'] = os.path.join(WINDDKdir, 'lib', target_os, target_cpu) 
-    env['CRT_LIB_PATH'] = os.path.join(WINDDKdir, 'lib/crt', target_cpu) 
+    env['CRT_LIB_PATH'] = os.path.join(WINDDKdir, 'lib', 'crt', target_cpu) 
     env['DDK_LIB_PATH'] = os.path.join(WINDDKdir, 'lib', target_os, target_cpu)
     env['WDM_LIB_PATH'] = os.path.join(WINDDKdir, 'lib', target_os, target_cpu)
                                      
@@ -90,6 +89,105 @@ def get_winddk_paths(env, version=None):
     exe_path = string.join(exe_paths, os.pathsep )
     return (include_path, lib_path, exe_path)
 
+def set_winddk_flags(env):
+    """Mimic WINDDK's builtin flags.
+
+    See also:
+    - WINDDK's bin/makefile.new i386mk.inc for more info.
+    - buildchk_wxp_x86.log files, generated by the WINDDK's build
+    - http://alter.org.ua/docs/nt_kernel/vc8_proj/
+    """
+    
+    cppdefines = [
+        ('_X86_', '1'), 
+        ('i386', '1'), 
+        'STD_CALL', 
+        ('CONDITION_HANDLING', '1'),
+        ('NT_INST', '0'), 
+        ('WIN32', '100'),
+        ('_NT1X_', '100'),
+        ('WINNT', '1'),
+        ('_WIN32_WINNT', '0x0501'), # minimum required OS version
+        ('WINVER', '0x0501'),
+        ('_WIN32_IE', '0x0603'),
+        ('WIN32_LEAN_AND_MEAN', '1'),
+        ('DEVL', '1'),
+        ('__BUILDMACHINE__', 'WinDDK'),
+        ('FPO', '0'),
+    ]
+    if env.get('DEBUG', False):
+         cppdefines += [
+              ('DBG', 1),
+         ]
+    env.Append(CPPDEFINES = cppdefines)
+
+    # See also:
+    # - http://msdn2.microsoft.com/en-us/library/y0zzbyt4.aspx
+    # - cl /?
+    cflags = [
+        '/Zl', # omit default library name in .OBJ
+        '/Zp8', # 8bytes struct member alignment
+        '/Gy', # separate functions for linker
+        '/Gm-', # disable minimal rebuild
+        '/W3', # warning level
+        '/WX', # treat warnings as errors
+        '/Gz', # __stdcall Calling convention
+        '/GX-', # disable C++ EH
+        '/GR-', # disable C++ RTTI
+        '/GF', # enable read-only string pooling
+        '/GS', # enable security checks
+        '/G6', # optimize for PPro, P-II, P-III
+        '/Ze', # enable extensions
+        #'/Gi-', # ???
+        '/QIfdiv-', # disable Pentium FDIV fix
+        #'/hotpatch', # ???
+        #'/Z7', #enable old-style debug info
+    ]
+    if env.get('debug', False):
+        cflags += [
+          '/Od', # disable optimizations
+          '/Oi', # enable intrinsic functions
+          '/Oy-', # disable frame pointer omission
+        ]
+    else:
+        cflags += [
+          '/Ox', # maximum optimizations
+          '/Oi', # enable intrinsic functions
+          '/Os', # favor code space
+        ]
+    env.Append(CFLAGS = cflags)
+    env.Append(CXXFLAGS = cflags)
+    
+    # See also:
+    # - http://msdn2.microsoft.com/en-us/library/y0zzbyt4.aspx
+    env.Append(LINKFLAGS = [
+        '/merge:_PAGE=PAGE',
+        '/merge:_TEXT=.text',
+        '/section:INIT,d',
+        '/opt:ref',
+        '/opt:icf',
+        '/ignore:4198,4010,4037,4039,4065,4070,4078,4087,4089,4221',
+        '/incremental:no',
+        '/fullbuild',
+        '/release',
+        '/nodefaultlib',
+        '/wx',
+        '/debug',
+        '/debugtype:cv',
+        '/version:5.1',
+        '/osversion:5.1',
+        '/functionpadmin:5',
+        '/safeseh',
+        '/pdbcompress',
+        '/stack:0x40000,0x1000',
+        '/driver', 
+        '/align:0x80',
+        '/subsystem:native,5.01',
+        '/base:0x10000',
+        
+        '/entry:DrvEnableDriver',
+    ])
+
 def validate_vars(env):
     """Validate the PCH and PCHSTOP construction variables."""
     if env.has_key('PCH') and env['PCH']:
@@ -201,17 +299,6 @@ def generate(env):
     env['SHOBJPREFIX']    = '$OBJPREFIX'
     env['SHOBJSUFFIX']    = '$OBJSUFFIX'
 
-    try:
-        include_path, lib_path, exe_path = get_winddk_paths(env)
-
-        # since other tools can set these, we just make sure that the
-        # relevant stuff from MSVS is in there somewhere.
-        env.PrependENVPath('INCLUDE', include_path)
-        env.PrependENVPath('LIB', lib_path)
-        env.PrependENVPath('PATH', exe_path)
-    except (SCons.Util.RegError, SCons.Errors.InternalError):
-        pass
-
     env['CFILESUFFIX'] = '.c'
     env['CXXFILESUFFIX'] = '.cc'
 
@@ -227,53 +314,24 @@ def generate(env):
 
     SCons.Tool.mslink.generate(env)
 
-    # See also:
-    # - WINDDK's bin/makefile.new i386mk.inc for more info.
-    # - http://alter.org.ua/docs/nt_kernel/vc8_proj/
-    env.Append(CPPDEFINES = [
-        'WIN32', 
-        '_WINDOWS', 
-        ('i386', '1'), 
-        ('_X86_', '1'), 
-        'STD_CALL', 
-        ('CONDITION_HANDLING', '1'),
-        ('NT_INST', '0'), 
-        ('_NT1X_', '100'),
-        ('WINNT', '1'),
-        ('_WIN32_WINNT', '0x0500'), # minimum required OS version
-        ('WIN32_LEAN_AND_MEAN', '1'),
-        ('DEVL', '1'),
-        ('FPO', '1'),
-    ])
-    cflags = [
-        '/GF', # Enable String Pooling
-        '/GX-', # Disable C++ Exceptions
-        '/Zp8', # 8bytes struct member alignment
-        #'/GS-', # No Buffer Security Check
-        '/GR-', # Disable Run-Time Type Info
-        '/Gz', # __stdcall Calling convention
-    ]
-    env.Append(CFLAGS = cflags)
-    env.Append(CXXFLAGS = cflags)
-    
-    env.Append(LINKFLAGS = [
-        '/DEBUG',
-        '/NODEFAULTLIB',
-        '/SUBSYSTEM:NATIVE',
-        '/INCREMENTAL:NO',
-        #'/DRIVER', 
-    
-        #'-subsystem:native,4.00',
-        '-base:0x10000',
-        
-        '-entry:DrvEnableDriver',
-    ])
+    set_winddk_flags(env)
     
     if not env.has_key('ENV'):
         env['ENV'] = {}
-    if not env['ENV'].has_key('SystemRoot'):    # required for dlls in the winsxs folders
-        env['ENV']['SystemRoot'] = SCons.Platform.win32.get_system_root()
+    
+    try:
+        include_path, lib_path, exe_path = get_winddk_paths(env)
+
+        # since other tools can set these, we just make sure that the
+        # relevant stuff from WINDDK is in there somewhere.
+        env.PrependENVPath('INCLUDE', include_path)
+        env.PrependENVPath('LIB', lib_path)
+        env.PrependENVPath('PATH', exe_path)
+    except (SCons.Util.RegError, SCons.Errors.InternalError):
+        pass
+
 
 def exists(env):
     return env.Detect('cl')
 
+# vim:set sw=4 et: