Merge remote branch 'origin/master' into nvc0-new

467 files changed, 14933 insertions, 10890 deletions

diff --git a/src/gallium/auxiliary/Makefile b/src/gallium/auxiliary/Makefile
index f37d59e9a3..ff355c4783 100644
--- a/src/gallium/auxiliary/Makefile
+++ b/src/gallium/auxiliary/Makefile
@@ -128,12 +128,12 @@ C_SOURCES = \
 	util/u_linkage.c \
 	util/u_network.c \
 	util/u_math.c \
-	util/u_mempool.c \
 	util/u_mm.c \
 	util/u_rect.c \
 	util/u_ringbuffer.c \
 	util/u_sampler.c \
 	util/u_simple_shaders.c \
+	util/u_slab.c \
 	util/u_snprintf.c \
 	util/u_staging.c \
 	util/u_surface.c \
@@ -142,8 +142,7 @@ C_SOURCES = \
 	util/u_tile.c \
 	util/u_transfer.c \
 	util/u_resource.c \
-	util/u_upload_mgr.c \
-	target-helpers/wrap_screen.c
+	util/u_upload_mgr.c
 
 	# Disabling until pipe-video branch gets merged in
 	#vl/vl_bitstream_parser.c \
@@ -186,7 +185,7 @@ GALLIVM_SOURCES = \
         draw/draw_pt_fetch_shade_pipeline_llvm.c
 
 GALLIVM_CPP_SOURCES = \
-    gallivm/lp_bld_misc.cpp
+	gallivm/lp_bld_misc.cpp
 
 GENERATED_SOURCES = \
 	indices/u_indices_gen.c \
@@ -204,9 +203,6 @@ CPP_SOURCES += \
 endif
 
 
-LIBRARY_DEFINES += -D__STDC_CONSTANT_MACROS
-
-
 include ../Makefile.template
 
 
diff --git a/src/gallium/auxiliary/SConscript b/src/gallium/auxiliary/SConscript
index 0e5da1374f..fca7e5fd11 100644
--- a/src/gallium/auxiliary/SConscript
+++ b/src/gallium/auxiliary/SConscript
@@ -175,13 +175,13 @@ source = [
     'util/u_linkage.c',
     'util/u_network.c',
     'util/u_math.c',
-    'util/u_mempool.c',
     'util/u_mm.c',
     'util/u_rect.c',
     'util/u_resource.c',
     'util/u_ringbuffer.c',
     'util/u_sampler.c',
     'util/u_simple_shaders.c',
+    'util/u_slab.c',
     'util/u_snprintf.c',
     'util/u_staging.c',
     'util/u_surface.c',
@@ -196,7 +196,6 @@ source = [
     #'vl/vl_compositor.c',
     #'vl/vl_csc.c',
     #'vl/vl_shader_build.c',
-    'target-helpers/wrap_screen.c',
 ]
 
 if env['llvm']:
diff --git a/src/gallium/auxiliary/draw/draw_context.c b/src/gallium/auxiliary/draw/draw_context.c
index 39d82f3289..73d5b6e403 100644
--- a/src/gallium/auxiliary/draw/draw_context.c
+++ b/src/gallium/auxiliary/draw/draw_context.c
@@ -63,19 +63,32 @@ draw_get_option_use_llvm(void)
 }
 #endif
 
-struct draw_context *draw_create( struct pipe_context *pipe )
+
+
+/**
+ * Create new draw module context.
+ */
+struct draw_context *
+draw_create(struct pipe_context *pipe)
+{
+   return draw_create_gallivm(pipe, NULL);
+}
+
+
+
+/**
+ * Create new draw module context with gallivm state for LLVM JIT.
+ */
+struct draw_context *
+draw_create_gallivm(struct pipe_context *pipe, struct gallivm_state *gallivm)
 {
    struct draw_context *draw = CALLOC_STRUCT( draw_context );
    if (draw == NULL)
       goto fail;
 
 #if HAVE_LLVM
-   if(draw_get_option_use_llvm())
-   {
-      lp_build_init();
-      assert(lp_build_engine);
-      draw->engine = lp_build_engine;
-      draw->llvm = draw_llvm_create(draw);
+   if (draw_get_option_use_llvm() && gallivm) {
+      draw->llvm = draw_llvm_create(draw, gallivm);
    }
 #endif
 
@@ -91,6 +104,8 @@ fail:
    return NULL;
 }
 
+
+
 boolean draw_init(struct draw_context *draw)
 {
    /*
diff --git a/src/gallium/auxiliary/draw/draw_context.h b/src/gallium/auxiliary/draw/draw_context.h
index ff4f753604..a0b217e4d3 100644
--- a/src/gallium/auxiliary/draw/draw_context.h
+++ b/src/gallium/auxiliary/draw/draw_context.h
@@ -48,10 +48,15 @@ struct draw_vertex_shader;
 struct draw_geometry_shader;
 struct draw_fragment_shader;
 struct tgsi_sampler;
+struct gallivm_state;
+
 
 
 struct draw_context *draw_create( struct pipe_context *pipe );
 
+struct draw_context *
+draw_create_gallivm(struct pipe_context *pipe, struct gallivm_state *gallivm);
+
 void draw_destroy( struct draw_context *draw );
 
 void draw_flush(struct draw_context *draw);
diff --git a/src/gallium/auxiliary/draw/draw_llvm.c b/src/gallium/auxiliary/draw/draw_llvm.c
index 2b5f01cda7..943ec44bcd 100644
--- a/src/gallium/auxiliary/draw/draw_llvm.c
+++ b/src/gallium/auxiliary/draw/draw_llvm.c
@@ -42,6 +42,7 @@
 #include "gallivm/lp_bld_printf.h"
 #include "gallivm/lp_bld_intr.h"
 #include "gallivm/lp_bld_init.h"
+#include "gallivm/lp_bld_type.h"
 
 #include "tgsi/tgsi_exec.h"
 #include "tgsi/tgsi_dump.h"
@@ -49,259 +50,352 @@
 #include "util/u_math.h"
 #include "util/u_pointer.h"
 #include "util/u_string.h"
+#include "util/u_simple_list.h"
 
-#include <llvm-c/Transforms/Scalar.h>
 
 #define DEBUG_STORE 0
 
-/* generates the draw jit function */
+
+/**
+ * This function is called by the gallivm "garbage collector" when
+ * the LLVM global data structures are freed.  We must free all LLVM-related
+ * data.  Specifically, all JIT'd shader variants.
+ */
+static void
+draw_llvm_garbage_collect_callback(void *cb_data)
+{
+   struct draw_llvm *llvm = (struct draw_llvm *) cb_data;
+   struct draw_llvm_variant_list_item *li;
+
+   /* free all shader variants */
+   li = first_elem(&llvm->vs_variants_list);
+   while (!at_end(&llvm->vs_variants_list, li)) {
+      struct draw_llvm_variant_list_item *next = next_elem(li);
+      draw_llvm_destroy_variant(li->base);
+      li = next;
+   }
+
+   /* Null-out these pointers so they get remade next time they're needed.
+    * See the accessor functions below.
+    */
+   llvm->context_ptr_type = NULL;
+   llvm->buffer_ptr_type = NULL;
+   llvm->vb_ptr_type = NULL;
+   llvm->vertex_header_ptr_type = NULL;
+}
+
+
 static void
 draw_llvm_generate(struct draw_llvm *llvm, struct draw_llvm_variant *var);
+
 static void
 draw_llvm_generate_elts(struct draw_llvm *llvm, struct draw_llvm_variant *var);
 
-static void
-init_globals(struct draw_llvm *llvm)
+
+/**
+ * Create LLVM type for struct draw_jit_texture
+ */
+static LLVMTypeRef
+create_jit_texture_type(struct gallivm_state *gallivm)
 {
+   LLVMTargetDataRef target = gallivm->target;
    LLVMTypeRef texture_type;
+   LLVMTypeRef elem_types[DRAW_JIT_TEXTURE_NUM_FIELDS];
+   LLVMTypeRef int32_type = LLVMInt32TypeInContext(gallivm->context);
+
+   elem_types[DRAW_JIT_TEXTURE_WIDTH]  =
+   elem_types[DRAW_JIT_TEXTURE_HEIGHT] =
+   elem_types[DRAW_JIT_TEXTURE_DEPTH] =
+   elem_types[DRAW_JIT_TEXTURE_LAST_LEVEL] = int32_type;
+   elem_types[DRAW_JIT_TEXTURE_ROW_STRIDE] =
+   elem_types[DRAW_JIT_TEXTURE_IMG_STRIDE] =
+      LLVMArrayType(int32_type, PIPE_MAX_TEXTURE_LEVELS);
+   elem_types[DRAW_JIT_TEXTURE_DATA] =
+      LLVMArrayType(LLVMPointerType(LLVMInt8TypeInContext(gallivm->context), 0),
+                    PIPE_MAX_TEXTURE_LEVELS);
+   elem_types[DRAW_JIT_TEXTURE_MIN_LOD] =
+   elem_types[DRAW_JIT_TEXTURE_MAX_LOD] =
+   elem_types[DRAW_JIT_TEXTURE_LOD_BIAS] = LLVMFloatTypeInContext(gallivm->context);
+   elem_types[DRAW_JIT_TEXTURE_BORDER_COLOR] = 
+      LLVMArrayType(LLVMFloatTypeInContext(gallivm->context), 4);
+
+   texture_type = LLVMStructTypeInContext(gallivm->context, elem_types,
+                                          Elements(elem_types), 0);
+
+   /* Make sure the target's struct layout cache doesn't return
+    * stale/invalid data.
+    */
+   LLVMInvalidateStructLayout(gallivm->target, texture_type);
+
+   LP_CHECK_MEMBER_OFFSET(struct draw_jit_texture, width,
+                          target, texture_type,
+                          DRAW_JIT_TEXTURE_WIDTH);
+   LP_CHECK_MEMBER_OFFSET(struct draw_jit_texture, height,
+                          target, texture_type,
+                          DRAW_JIT_TEXTURE_HEIGHT);
+   LP_CHECK_MEMBER_OFFSET(struct draw_jit_texture, depth,
+                          target, texture_type,
+                          DRAW_JIT_TEXTURE_DEPTH);
+   LP_CHECK_MEMBER_OFFSET(struct draw_jit_texture, last_level,
+                          target, texture_type,
+                          DRAW_JIT_TEXTURE_LAST_LEVEL);
+   LP_CHECK_MEMBER_OFFSET(struct draw_jit_texture, row_stride,
+                          target, texture_type,
+                          DRAW_JIT_TEXTURE_ROW_STRIDE);
+   LP_CHECK_MEMBER_OFFSET(struct draw_jit_texture, img_stride,
+                          target, texture_type,
+                          DRAW_JIT_TEXTURE_IMG_STRIDE);
+   LP_CHECK_MEMBER_OFFSET(struct draw_jit_texture, data,
+                          target, texture_type,
+                          DRAW_JIT_TEXTURE_DATA);
+   LP_CHECK_MEMBER_OFFSET(struct draw_jit_texture, min_lod,
+                          target, texture_type,
+                          DRAW_JIT_TEXTURE_MIN_LOD);
+   LP_CHECK_MEMBER_OFFSET(struct draw_jit_texture, max_lod,
+                          target, texture_type,
+                          DRAW_JIT_TEXTURE_MAX_LOD);
+   LP_CHECK_MEMBER_OFFSET(struct draw_jit_texture, lod_bias,
+                          target, texture_type,
+                          DRAW_JIT_TEXTURE_LOD_BIAS);
+   LP_CHECK_MEMBER_OFFSET(struct draw_jit_texture, border_color,
+                          target, texture_type,
+                          DRAW_JIT_TEXTURE_BORDER_COLOR);
+
+   LP_CHECK_STRUCT_SIZE(struct draw_jit_texture, target, texture_type);
+
+   return texture_type;
+}
 
-   /* struct draw_jit_texture */
-   {
-      LLVMTypeRef elem_types[DRAW_JIT_TEXTURE_NUM_FIELDS];
-
-      elem_types[DRAW_JIT_TEXTURE_WIDTH]  = LLVMInt32Type();
-      elem_types[DRAW_JIT_TEXTURE_HEIGHT] = LLVMInt32Type();
-      elem_types[DRAW_JIT_TEXTURE_DEPTH] = LLVMInt32Type();
-      elem_types[DRAW_JIT_TEXTURE_LAST_LEVEL] = LLVMInt32Type();
-      elem_types[DRAW_JIT_TEXTURE_ROW_STRIDE] =
-         LLVMArrayType(LLVMInt32Type(), PIPE_MAX_TEXTURE_LEVELS);
-      elem_types[DRAW_JIT_TEXTURE_IMG_STRIDE] =
-         LLVMArrayType(LLVMInt32Type(), PIPE_MAX_TEXTURE_LEVELS);
-      elem_types[DRAW_JIT_TEXTURE_DATA] =
-         LLVMArrayType(LLVMPointerType(LLVMInt8Type(), 0),
-                       PIPE_MAX_TEXTURE_LEVELS);
-      elem_types[DRAW_JIT_TEXTURE_MIN_LOD] = LLVMFloatType();
-      elem_types[DRAW_JIT_TEXTURE_MAX_LOD] = LLVMFloatType();
-      elem_types[DRAW_JIT_TEXTURE_LOD_BIAS] = LLVMFloatType();
-      elem_types[DRAW_JIT_TEXTURE_BORDER_COLOR] = 
-         LLVMArrayType(LLVMFloatType(), 4);
-
-      texture_type = LLVMStructType(elem_types, Elements(elem_types), 0);
-
-      LP_CHECK_MEMBER_OFFSET(struct draw_jit_texture, width,
-                             llvm->target, texture_type,
-                             DRAW_JIT_TEXTURE_WIDTH);
-      LP_CHECK_MEMBER_OFFSET(struct draw_jit_texture, height,
-                             llvm->target, texture_type,
-                             DRAW_JIT_TEXTURE_HEIGHT);
-      LP_CHECK_MEMBER_OFFSET(struct draw_jit_texture, depth,
-                             llvm->target, texture_type,
-                             DRAW_JIT_TEXTURE_DEPTH);
-      LP_CHECK_MEMBER_OFFSET(struct draw_jit_texture, last_level,
-                             llvm->target, texture_type,
-                             DRAW_JIT_TEXTURE_LAST_LEVEL);
-      LP_CHECK_MEMBER_OFFSET(struct draw_jit_texture, row_stride,
-                             llvm->target, texture_type,
-                             DRAW_JIT_TEXTURE_ROW_STRIDE);
-      LP_CHECK_MEMBER_OFFSET(struct draw_jit_texture, img_stride,
-                             llvm->target, texture_type,
-                             DRAW_JIT_TEXTURE_IMG_STRIDE);
-      LP_CHECK_MEMBER_OFFSET(struct draw_jit_texture, data,
-                             llvm->target, texture_type,
-                             DRAW_JIT_TEXTURE_DATA);
-      LP_CHECK_MEMBER_OFFSET(struct draw_jit_texture, min_lod,
-                             llvm->target, texture_type,
-                             DRAW_JIT_TEXTURE_MIN_LOD);
-      LP_CHECK_MEMBER_OFFSET(struct draw_jit_texture, max_lod,
-                             llvm->target, texture_type,
-                             DRAW_JIT_TEXTURE_MAX_LOD);
-      LP_CHECK_MEMBER_OFFSET(struct draw_jit_texture, lod_bias,
-                             llvm->target, texture_type,
-                             DRAW_JIT_TEXTURE_LOD_BIAS);
-      LP_CHECK_MEMBER_OFFSET(struct draw_jit_texture, border_color,
-                             llvm->target, texture_type,
-                             DRAW_JIT_TEXTURE_BORDER_COLOR);
-      LP_CHECK_STRUCT_SIZE(struct draw_jit_texture,
-                           llvm->target, texture_type);
-
-      LLVMAddTypeName(llvm->module, "texture", texture_type);
-   }
 
+/**
+ * Create LLVM type for struct draw_jit_texture
+ */
+static LLVMTypeRef
+create_jit_context_type(struct gallivm_state *gallivm,
+                        LLVMTypeRef texture_type)
+{
+   LLVMTargetDataRef target = gallivm->target;
+   LLVMTypeRef float_type = LLVMFloatTypeInContext(gallivm->context);
+   LLVMTypeRef elem_types[5];
+   LLVMTypeRef context_type;
+
+   elem_types[0] = LLVMPointerType(float_type, 0); /* vs_constants */
+   elem_types[1] = LLVMPointerType(float_type, 0); /* gs_constants */
+   elem_types[2] = LLVMPointerType(LLVMArrayType(LLVMArrayType(float_type, 4), 12), 0); /* planes */
+   elem_types[3] = LLVMPointerType(float_type, 0); /* viewport */
+   elem_types[4] = LLVMArrayType(texture_type,
+                                 PIPE_MAX_VERTEX_SAMPLERS); /* textures */
+
+   context_type = LLVMStructTypeInContext(gallivm->context, elem_types,
+                                          Elements(elem_types), 0);
+
+   LLVMInvalidateStructLayout(gallivm->target, context_type);
+
+   LP_CHECK_MEMBER_OFFSET(struct draw_jit_context, vs_constants,
+                          target, context_type, 0);
+   LP_CHECK_MEMBER_OFFSET(struct draw_jit_context, gs_constants,
+                          target, context_type, 1);
+   LP_CHECK_MEMBER_OFFSET(struct draw_jit_context, planes,
+                          target, context_type, 2);
+   LP_CHECK_MEMBER_OFFSET(struct draw_jit_context, textures,
+                          target, context_type,
+                          DRAW_JIT_CTX_TEXTURES);
+   LP_CHECK_STRUCT_SIZE(struct draw_jit_context,
+                        target, context_type);
+
+   return context_type;
+}
 
-   /* struct draw_jit_context */
-   {
-      LLVMTypeRef elem_types[5];
-      LLVMTypeRef context_type;
-
-      elem_types[0] = LLVMPointerType(LLVMFloatType(), 0); /* vs_constants */
-      elem_types[1] = LLVMPointerType(LLVMFloatType(), 0); /* gs_constants */
-      elem_types[2] = LLVMPointerType(LLVMArrayType(LLVMArrayType(LLVMFloatType(), 4), 12), 0); /* planes */
-      elem_types[3] = LLVMPointerType(LLVMFloatType(), 0); /* viewport */
-      elem_types[4] = LLVMArrayType(texture_type,
-                                    PIPE_MAX_VERTEX_SAMPLERS); /* textures */
-
-      context_type = LLVMStructType(elem_types, Elements(elem_types), 0);
-
-      LP_CHECK_MEMBER_OFFSET(struct draw_jit_context, vs_constants,
-                             llvm->target, context_type, 0);
-      LP_CHECK_MEMBER_OFFSET(struct draw_jit_context, gs_constants,
-                             llvm->target, context_type, 1);
-      LP_CHECK_MEMBER_OFFSET(struct draw_jit_context, planes,
-                             llvm->target, context_type, 2);
-      LP_CHECK_MEMBER_OFFSET(struct draw_jit_context, textures,
-                             llvm->target, context_type,
-                             DRAW_JIT_CTX_TEXTURES);
-      LP_CHECK_STRUCT_SIZE(struct draw_jit_context,
-                           llvm->target, context_type);
-
-      LLVMAddTypeName(llvm->module, "draw_jit_context", context_type);
-
-      llvm->context_ptr_type = LLVMPointerType(context_type, 0);
-   }
-   {
-      LLVMTypeRef buffer_ptr = LLVMPointerType(LLVMIntType(8), 0);
-      llvm->buffer_ptr_type = LLVMPointerType(buffer_ptr, 0);
-   }
-   /* struct pipe_vertex_buffer */
-   {
-      LLVMTypeRef elem_types[4];
-      LLVMTypeRef vb_type;
 
-      elem_types[0] = LLVMInt32Type();
-      elem_types[1] = LLVMInt32Type();
-      elem_types[2] = LLVMInt32Type();
-      elem_types[3] = LLVMPointerType(LLVMOpaqueType(), 0); /* vs_constants */
+/**
+ * Create LLVM type for struct pipe_vertex_buffer
+ */
+static LLVMTypeRef
+create_jit_vertex_buffer_type(struct gallivm_state *gallivm)
+{
+   LLVMTargetDataRef target = gallivm->target;
+   LLVMTypeRef elem_types[4];
+   LLVMTypeRef vb_type;
 
-      vb_type = LLVMStructType(elem_types, Elements(elem_types), 0);
+   elem_types[0] =
+   elem_types[1] =
+   elem_types[2] = LLVMInt32TypeInContext(gallivm->context);
+   elem_types[3] = LLVMPointerType(LLVMOpaqueTypeInContext(gallivm->context), 0); /* vs_constants */
 
-      LP_CHECK_MEMBER_OFFSET(struct pipe_vertex_buffer, stride,
-                             llvm->target, vb_type, 0);
-      LP_CHECK_MEMBER_OFFSET(struct pipe_vertex_buffer, buffer_offset,
-                             llvm->target, vb_type, 2);
-      LP_CHECK_STRUCT_SIZE(struct pipe_vertex_buffer,
-                           llvm->target, vb_type);
+   vb_type = LLVMStructTypeInContext(gallivm->context, elem_types,
+                                     Elements(elem_types), 0);
 
-      LLVMAddTypeName(llvm->module, "pipe_vertex_buffer", vb_type);
+   LLVMInvalidateStructLayout(gallivm->target, vb_type);
 
-      llvm->vb_ptr_type = LLVMPointerType(vb_type, 0);
-   }
+   LP_CHECK_MEMBER_OFFSET(struct pipe_vertex_buffer, stride,
+                          target, vb_type, 0);
+   LP_CHECK_MEMBER_OFFSET(struct pipe_vertex_buffer, buffer_offset,
+                          target, vb_type, 2);
+
+   LP_CHECK_STRUCT_SIZE(struct pipe_vertex_buffer, target, vb_type);
+
+   return vb_type;
 }
 
+
+/**
+ * Create LLVM type for struct vertex_header;
+ */
 static LLVMTypeRef
-create_vertex_header(struct draw_llvm *llvm, int data_elems)
+create_jit_vertex_header(struct gallivm_state *gallivm, int data_elems)
 {
-   /* struct vertex_header */
+   LLVMTargetDataRef target = gallivm->target;
    LLVMTypeRef elem_types[3];
    LLVMTypeRef vertex_header;
    char struct_name[24];
 
    util_snprintf(struct_name, 23, "vertex_header%d", data_elems);
 
-   elem_types[0]  = LLVMIntType(32);
-   elem_types[1]  = LLVMArrayType(LLVMFloatType(), 4);
+   elem_types[0]  = LLVMIntTypeInContext(gallivm->context, 32);
+   elem_types[1]  = LLVMArrayType(LLVMFloatTypeInContext(gallivm->context), 4);
    elem_types[2]  = LLVMArrayType(elem_types[1], data_elems);
 
-   vertex_header = LLVMStructType(elem_types, Elements(elem_types), 0);
+   vertex_header = LLVMStructTypeInContext(gallivm->context, elem_types,
+                                           Elements(elem_types), 0);
+
+   LLVMInvalidateStructLayout(gallivm->target, vertex_header);
 
    /* these are bit-fields and we can't take address of them
       LP_CHECK_MEMBER_OFFSET(struct vertex_header, clipmask,
-      llvm->target, vertex_header,
+      target, vertex_header,
       DRAW_JIT_VERTEX_CLIPMASK);
       LP_CHECK_MEMBER_OFFSET(struct vertex_header, edgeflag,
-      llvm->target, vertex_header,
+      target, vertex_header,
       DRAW_JIT_VERTEX_EDGEFLAG);
       LP_CHECK_MEMBER_OFFSET(struct vertex_header, pad,
-      llvm->target, vertex_header,
+      target, vertex_header,
       DRAW_JIT_VERTEX_PAD);
       LP_CHECK_MEMBER_OFFSET(struct vertex_header, vertex_id,
-      llvm->target, vertex_header,
+      target, vertex_header,
       DRAW_JIT_VERTEX_VERTEX_ID);
    */
    LP_CHECK_MEMBER_OFFSET(struct vertex_header, clip,
-                          llvm->target, vertex_header,
+                          target, vertex_header,
                           DRAW_JIT_VERTEX_CLIP);
    LP_CHECK_MEMBER_OFFSET(struct vertex_header, data,
-                          llvm->target, vertex_header,
+                          target, vertex_header,
                           DRAW_JIT_VERTEX_DATA);
 
-   LLVMAddTypeName(llvm->module, struct_name, vertex_header);
+   LLVMAddTypeName(gallivm->module, struct_name, vertex_header);
 
-   return LLVMPointerType(vertex_header, 0);
+   return vertex_header;
 }
 
-struct draw_llvm *
-draw_llvm_create(struct draw_context *draw)
+
+/**
+ * Create LLVM types for various structures.
+ */
+static void
+create_jit_types(struct draw_llvm *llvm)
 {
-   struct draw_llvm *llvm;
+   struct gallivm_state *gallivm = llvm->gallivm;
+   LLVMTypeRef texture_type, context_type, buffer_type, vb_type;
 
-   llvm = CALLOC_STRUCT( draw_llvm );
-   if (!llvm)
-      return NULL;
+   texture_type = create_jit_texture_type(gallivm);
+   LLVMAddTypeName(gallivm->module, "texture", texture_type);
 
-   llvm->draw = draw;
-   llvm->engine = draw->engine;
+   context_type = create_jit_context_type(gallivm, texture_type);
+   LLVMAddTypeName(gallivm->module, "draw_jit_context", context_type);
+   llvm->context_ptr_type = LLVMPointerType(context_type, 0);
 
-   debug_assert(llvm->engine);
+   buffer_type = LLVMPointerType(LLVMIntTypeInContext(gallivm->context, 8), 0);
+   LLVMAddTypeName(gallivm->module, "buffer", buffer_type);
+   llvm->buffer_ptr_type = LLVMPointerType(buffer_type, 0);
 
-   llvm->module = LLVMModuleCreateWithName("draw_llvm");
-   llvm->provider = LLVMCreateModuleProviderForExistingModule(llvm->module);
+   vb_type = create_jit_vertex_buffer_type(gallivm);
+   LLVMAddTypeName(gallivm->module, "pipe_vertex_buffer", vb_type);
+   llvm->vb_ptr_type = LLVMPointerType(vb_type, 0);
+}
 
-   LLVMAddModuleProvider(llvm->engine, llvm->provider);
 
-   llvm->target = LLVMGetExecutionEngineTargetData(llvm->engine);
+static LLVMTypeRef
+get_context_ptr_type(struct draw_llvm *llvm)
+{
+   if (!llvm->context_ptr_type)
+      create_jit_types(llvm);
+   return llvm->context_ptr_type;
+}
 
-   llvm->pass = LLVMCreateFunctionPassManager(llvm->provider);
-   LLVMAddTargetData(llvm->target, llvm->pass);
 
-   if ((gallivm_debug & GALLIVM_DEBUG_NO_OPT) == 0) {
-      /* These are the passes currently listed in llvm-c/Transforms/Scalar.h,
-       * but there are more on SVN. */
-      /* TODO: Add more passes */
+static LLVMTypeRef
+get_buffer_ptr_type(struct draw_llvm *llvm)
+{
+   if (!llvm->buffer_ptr_type)
+      create_jit_types(llvm);
+   return llvm->buffer_ptr_type;
+}
 
-      LLVMAddCFGSimplificationPass(llvm->pass);
 
-      if (HAVE_LLVM >= 0x207 && sizeof(void*) == 4) {
-         /* For LLVM >= 2.7 and 32-bit build, use this order of passes to
-          * avoid generating bad code.
-          * Test with piglit glsl-vs-sqrt-zero test.
-          */
-         LLVMAddConstantPropagationPass(llvm->pass);
-         LLVMAddPromoteMemoryToRegisterPass(llvm->pass);
-      }
-      else {
-         LLVMAddPromoteMemoryToRegisterPass(llvm->pass);
-         LLVMAddConstantPropagationPass(llvm->pass);
-      }
+static LLVMTypeRef
+get_vb_ptr_type(struct draw_llvm *llvm)
+{
+   if (!llvm->vb_ptr_type)
+      create_jit_types(llvm);
+   return llvm->vb_ptr_type;
+}
 
-      LLVMAddInstructionCombiningPass(llvm->pass);
-      LLVMAddGVNPass(llvm->pass);
-   } else {
-      /* We need at least this pass to prevent the backends to fail in
-       * unexpected ways.
-       */
-      LLVMAddPromoteMemoryToRegisterPass(llvm->pass);
-   }
+static LLVMTypeRef
+get_vertex_header_ptr_type(struct draw_llvm *llvm)
+{
+   if (!llvm->vertex_header_ptr_type)
+      create_jit_types(llvm);
+   return llvm->vertex_header_ptr_type;
+}
 
-   init_globals(llvm);
+
+/**
+ * Create per-context LLVM info.
+ */
+struct draw_llvm *
+draw_llvm_create(struct draw_context *draw, struct gallivm_state *gallivm)
+{
+   struct draw_llvm *llvm;
+
+   llvm = CALLOC_STRUCT( draw_llvm );
+   if (!llvm)
+      return NULL;
+
+   lp_build_init();
+
+   llvm->draw = draw;
+   llvm->gallivm = gallivm;
 
    if (gallivm_debug & GALLIVM_DEBUG_IR) {
-      LLVMDumpModule(llvm->module);
+      LLVMDumpModule(llvm->gallivm->module);
    }
 
    llvm->nr_variants = 0;
    make_empty_list(&llvm->vs_variants_list);
 
+   gallivm_register_garbage_collector_callback(
+                              draw_llvm_garbage_collect_callback, llvm);
+
    return llvm;
 }
 
+
+/**
+ * Free per-context LLVM info.
+ */
 void
 draw_llvm_destroy(struct draw_llvm *llvm)
 {
-   LLVMDisposePassManager(llvm->pass);
+   gallivm_remove_garbage_collector_callback(
+                              draw_llvm_garbage_collect_callback, llvm);
 
+   /* XXX free other draw_llvm data? */
    FREE(llvm);
 }
 
+
+/**
+ * Create LLVM-generated code for a vertex shader.
+ */
 struct draw_llvm_variant *
 draw_llvm_create_variant(struct draw_llvm *llvm,
 			 unsigned num_inputs,
@@ -310,6 +404,7 @@ draw_llvm_create_variant(struct draw_llvm *llvm,
    struct draw_llvm_variant *variant;
    struct llvm_vertex_shader *shader =
       llvm_vertex_shader(llvm->draw->vs.vertex_shader);
+   LLVMTypeRef vertex_header;
 
    variant = MALLOC(sizeof *variant +
 		    shader->variant_key_size -
@@ -321,7 +416,9 @@ draw_llvm_create_variant(struct draw_llvm *llvm,
 
    memcpy(&variant->key, key, shader->variant_key_size);
 
-   llvm->vertex_header_ptr_type = create_vertex_header(llvm, num_inputs);
+   vertex_header = create_jit_vertex_header(llvm->gallivm, num_inputs);
+
+   llvm->vertex_header_ptr_type = LLVMPointerType(vertex_header, 0);
 
    draw_llvm_generate(llvm, variant);
    draw_llvm_generate_elts(llvm, variant);
@@ -345,7 +442,7 @@ generate_vs(struct draw_llvm *llvm,
 {
    const struct tgsi_token *tokens = llvm->draw->vs.vertex_shader->state.tokens;
    struct lp_type vs_type;
-   LLVMValueRef consts_ptr = draw_jit_context_vs_constants(builder, context_ptr);
+   LLVMValueRef consts_ptr = draw_jit_context_vs_constants(llvm->gallivm, context_ptr);
    struct lp_build_sampler_soa *sampler = 0;
 
    memset(&vs_type, 0, sizeof vs_type);
@@ -366,7 +463,7 @@ generate_vs(struct draw_llvm *llvm,
        llvm->draw->num_samplers)
       sampler = draw_sampler;
 
-   lp_build_tgsi_soa(builder,
+   lp_build_tgsi_soa(llvm->gallivm,
                      tokens,
                      vs_type,
                      NULL /*struct lp_build_mask_context *mask*/,
@@ -384,20 +481,20 @@ static void print_vectorf(LLVMBuilderRef builder,
 {
    LLVMValueRef val[4];
    val[0] = LLVMBuildExtractElement(builder, vec,
-                                    LLVMConstInt(LLVMInt32Type(), 0, 0), "");
+                                    lp_build_const_int32(gallivm, 0), "");
    val[1] = LLVMBuildExtractElement(builder, vec,
-                                    LLVMConstInt(LLVMInt32Type(), 1, 0), "");
+                                    lp_build_const_int32(gallivm, 1), "");
    val[2] = LLVMBuildExtractElement(builder, vec,
-                                    LLVMConstInt(LLVMInt32Type(), 2, 0), "");
+                                    lp_build_const_int32(gallivm, 2), "");
    val[3] = LLVMBuildExtractElement(builder, vec,
-                                    LLVMConstInt(LLVMInt32Type(), 3, 0), "");
+                                    lp_build_const_int32(gallivm, 3), "");
    lp_build_printf(builder, "vector = [%f, %f, %f, %f]\n",
                    val[0], val[1], val[2], val[3]);
 }
 #endif
 
 static void
-generate_fetch(LLVMBuilderRef builder,
+generate_fetch(struct gallivm_state *gallivm,
                LLVMValueRef vbuffers_ptr,
                LLVMValueRef *res,
                struct pipe_vertex_element *velem,
@@ -405,19 +502,22 @@ generate_fetch(LLVMBuilderRef builder,
                LLVMValueRef index,
                LLVMValueRef instance_id)
 {
-   LLVMValueRef indices = LLVMConstInt(LLVMInt64Type(), velem->vertex_buffer_index, 0);
+   LLVMBuilderRef builder = gallivm->builder;
+   LLVMValueRef indices =
+      LLVMConstInt(LLVMInt64TypeInContext(gallivm->context),
+                   velem->vertex_buffer_index, 0);
    LLVMValueRef vbuffer_ptr = LLVMBuildGEP(builder, vbuffers_ptr,
                                            &indices, 1, "");
-   LLVMValueRef vb_stride = draw_jit_vbuffer_stride(builder, vbuf);
-   LLVMValueRef vb_max_index = draw_jit_vbuffer_max_index(builder, vbuf);
-   LLVMValueRef vb_buffer_offset = draw_jit_vbuffer_offset(builder, vbuf);
+   LLVMValueRef vb_stride = draw_jit_vbuffer_stride(gallivm, vbuf);
+   LLVMValueRef vb_max_index = draw_jit_vbuffer_max_index(gallivm, vbuf);
+   LLVMValueRef vb_buffer_offset = draw_jit_vbuffer_offset(gallivm, vbuf);
    LLVMValueRef cond;
    LLVMValueRef stride;
 
    if (velem->instance_divisor) {
       /* array index = instance_id / instance_divisor */
       index = LLVMBuildUDiv(builder, instance_id,
-                            LLVMConstInt(LLVMInt32Type(), velem->instance_divisor, 0),
+                            lp_build_const_int32(gallivm, velem->instance_divisor),
                             "instance_divisor");
    }
 
@@ -433,23 +533,24 @@ generate_fetch(LLVMBuilderRef builder,
                          vb_buffer_offset,
                          "");
    stride = LLVMBuildAdd(builder, stride,
-                         LLVMConstInt(LLVMInt32Type(), velem->src_offset, 0),
+                         lp_build_const_int32(gallivm, velem->src_offset),
                          "");
 
    /*lp_build_printf(builder, "vbuf index = %d, stride is %d\n", indices, stride);*/
    vbuffer_ptr = LLVMBuildGEP(builder, vbuffer_ptr, &stride, 1, "");
 
-   *res = draw_llvm_translate_from(builder, vbuffer_ptr, velem->src_format);
+   *res = draw_llvm_translate_from(gallivm, vbuffer_ptr, velem->src_format);
 }
 
 static LLVMValueRef
-aos_to_soa(LLVMBuilderRef builder,
+aos_to_soa(struct gallivm_state *gallivm,
            LLVMValueRef val0,
            LLVMValueRef val1,
            LLVMValueRef val2,
            LLVMValueRef val3,
            LLVMValueRef channel)
 {
+   LLVMBuilderRef builder = gallivm->builder;
    LLVMValueRef ex, res;
 
    ex = LLVMBuildExtractElement(builder, val0,
@@ -457,38 +558,39 @@ aos_to_soa(LLVMBuilderRef builder,
    res = LLVMBuildInsertElement(builder,
                                 LLVMConstNull(LLVMTypeOf(val0)),
                                 ex,
-                                LLVMConstInt(LLVMInt32Type(), 0, 0),
+                                lp_build_const_int32(gallivm, 0),
                                 "");
 
    ex = LLVMBuildExtractElement(builder, val1,
                                 channel, "");
    res = LLVMBuildInsertElement(builder,
                                 res, ex,
-                                LLVMConstInt(LLVMInt32Type(), 1, 0),
+                                lp_build_const_int32(gallivm, 1),
                                 "");
 
    ex = LLVMBuildExtractElement(builder, val2,
                                 channel, "");
    res = LLVMBuildInsertElement(builder,
                                 res, ex,
-                                LLVMConstInt(LLVMInt32Type(), 2, 0),
+                                lp_build_const_int32(gallivm, 2),
                                 "");
 
    ex = LLVMBuildExtractElement(builder, val3,
                                 channel, "");
    res = LLVMBuildInsertElement(builder,
                                 res, ex,
-                                LLVMConstInt(LLVMInt32Type(), 3, 0),
+                                lp_build_const_int32(gallivm, 3),
                                 "");
 
    return res;
 }
 
 static void
-soa_to_aos(LLVMBuilderRef builder,
+soa_to_aos(struct gallivm_state *gallivm,
            LLVMValueRef soa[NUM_CHANNELS],
            LLVMValueRef aos[NUM_CHANNELS])
 {
+   LLVMBuilderRef builder = gallivm->builder;
    LLVMValueRef comp;
    int i = 0;
 
@@ -498,29 +600,29 @@ soa_to_aos(LLVMBuilderRef builder,
    aos[1] = aos[2] = aos[3] = aos[0];
 
    for (i = 0; i < NUM_CHANNELS; ++i) {
-      LLVMValueRef channel = LLVMConstInt(LLVMInt32Type(), i, 0);
+      LLVMValueRef channel = lp_build_const_int32(gallivm, i);
 
       comp = LLVMBuildExtractElement(builder, soa[i],
-                                     LLVMConstInt(LLVMInt32Type(), 0, 0), "");
+                                     lp_build_const_int32(gallivm, 0), "");
       aos[0] = LLVMBuildInsertElement(builder, aos[0], comp, channel, "");
 
       comp = LLVMBuildExtractElement(builder, soa[i],
-                                     LLVMConstInt(LLVMInt32Type(), 1, 0), "");
+                                     lp_build_const_int32(gallivm, 1), "");
       aos[1] = LLVMBuildInsertElement(builder, aos[1], comp, channel, "");
 
       comp = LLVMBuildExtractElement(builder, soa[i],
-                                     LLVMConstInt(LLVMInt32Type(), 2, 0), "");
+                                     lp_build_const_int32(gallivm, 2), "");
       aos[2] = LLVMBuildInsertElement(builder, aos[2], comp, channel, "");
 
       comp = LLVMBuildExtractElement(builder, soa[i],
-                                     LLVMConstInt(LLVMInt32Type(), 3, 0), "");
+                                     lp_build_const_int32(gallivm, 3), "");
       aos[3] = LLVMBuildInsertElement(builder, aos[3], comp, channel, "");
 
    }
 }
 
 static void
-convert_to_soa(LLVMBuilderRef builder,
+convert_to_soa(struct gallivm_state *gallivm,
                LLVMValueRef (*aos)[NUM_CHANNELS],
                LLVMValueRef (*soa)[NUM_CHANNELS],
                int num_attribs)
@@ -535,36 +637,37 @@ convert_to_soa(LLVMBuilderRef builder,
       LLVMValueRef val2 = aos[i][2];
       LLVMValueRef val3 = aos[i][3];
 
-      soa[i][0] = aos_to_soa(builder, val0, val1, val2, val3,
-                             LLVMConstInt(LLVMInt32Type(), 0, 0));
-      soa[i][1] = aos_to_soa(builder, val0, val1, val2, val3,
-                             LLVMConstInt(LLVMInt32Type(), 1, 0));
-      soa[i][2] = aos_to_soa(builder, val0, val1, val2, val3,
-                             LLVMConstInt(LLVMInt32Type(), 2, 0));
-      soa[i][3] = aos_to_soa(builder, val0, val1, val2, val3,
-                             LLVMConstInt(LLVMInt32Type(), 3, 0));
+      soa[i][0] = aos_to_soa(gallivm, val0, val1, val2, val3,
+                             lp_build_const_int32(gallivm, 0));
+      soa[i][1] = aos_to_soa(gallivm, val0, val1, val2, val3,
+                             lp_build_const_int32(gallivm, 1));
+      soa[i][2] = aos_to_soa(gallivm, val0, val1, val2, val3,
+                             lp_build_const_int32(gallivm, 2));
+      soa[i][3] = aos_to_soa(gallivm, val0, val1, val2, val3,
+                             lp_build_const_int32(gallivm, 3));
    }
 }
 
 static void
-store_aos(LLVMBuilderRef builder,
+store_aos(struct gallivm_state *gallivm,
           LLVMValueRef io_ptr,
           LLVMValueRef index,
           LLVMValueRef value,
           LLVMValueRef clipmask)
 {
-   LLVMValueRef id_ptr = draw_jit_header_id(builder, io_ptr);
-   LLVMValueRef data_ptr = draw_jit_header_data(builder, io_ptr);
+   LLVMBuilderRef builder = gallivm->builder;
+   LLVMValueRef id_ptr = draw_jit_header_id(gallivm, io_ptr);
+   LLVMValueRef data_ptr = draw_jit_header_data(gallivm, io_ptr);
    LLVMValueRef indices[3];
    LLVMValueRef val, shift;
 
-   indices[0] = LLVMConstInt(LLVMInt32Type(), 0, 0);
+   indices[0] = lp_build_const_int32(gallivm, 0);
    indices[1] = index;
-   indices[2] = LLVMConstInt(LLVMInt32Type(), 0, 0);
+   indices[2] = lp_build_const_int32(gallivm, 0);
 
    /* initialize vertex id:16 = 0xffff, pad:3 = 0, edgeflag:1 = 1 */
-   val = LLVMConstInt(LLVMInt32Type(), 0xffff1, 0); 
-   shift  = LLVMConstInt(LLVMInt32Type(), 12, 0);          
+   val = lp_build_const_int32(gallivm, 0xffff1);
+   shift = lp_build_const_int32(gallivm, 12);
    val = LLVMBuildShl(builder, val, shift, "");
    /* add clipmask:12 */   
    val = LLVMBuildOr(builder, val, clipmask, "");               
@@ -580,7 +683,7 @@ store_aos(LLVMBuilderRef builder,
    /*lp_build_printf(builder, " ---- %p storing at %d (%p)  ", io_ptr, index, data_ptr);
      print_vectorf(builder, value);*/
    data_ptr = LLVMBuildBitCast(builder, data_ptr,
-                               LLVMPointerType(LLVMArrayType(LLVMVectorType(LLVMFloatType(), 4), 0), 0),
+                               LLVMPointerType(LLVMArrayType(LLVMVectorType(LLVMFloatTypeInContext(gallivm->context), 4), 0), 0),
                                "datavec");
    data_ptr = LLVMBuildGEP(builder, data_ptr, indices, 2, "");
 
@@ -592,10 +695,10 @@ store_aos(LLVMBuilderRef builder,
       LLVMValueRef gep0, gep1, gep2, gep3;
       data_ptr = LLVMBuildGEP(builder, data_ptr, indices, 3, "");
 
-      idx0 = LLVMConstInt(LLVMInt32Type(), 0, 0);
-      idx1 = LLVMConstInt(LLVMInt32Type(), 1, 0);
-      idx2 = LLVMConstInt(LLVMInt32Type(), 2, 0);
-      idx3 = LLVMConstInt(LLVMInt32Type(), 3, 0);
+      idx0 = lp_build_const_int32(gallivm, 0);
+      idx1 = lp_build_const_int32(gallivm, 1);
+      idx2 = lp_build_const_int32(gallivm, 2);
+      idx3 = lp_build_const_int32(gallivm, 3);
 
       x = LLVMBuildExtractElement(builder, value,
                                   idx0, "");
@@ -622,18 +725,19 @@ store_aos(LLVMBuilderRef builder,
 }
 
 static void
-store_aos_array(LLVMBuilderRef builder,
+store_aos_array(struct gallivm_state *gallivm,
                 LLVMValueRef io_ptr,
                 LLVMValueRef aos[NUM_CHANNELS],
                 int attrib,
                 int num_outputs,
                 LLVMValueRef clipmask)
 {
-   LLVMValueRef attr_index = LLVMConstInt(LLVMInt32Type(), attrib, 0);
-   LLVMValueRef ind0 = LLVMConstInt(LLVMInt32Type(), 0, 0);
-   LLVMValueRef ind1 = LLVMConstInt(LLVMInt32Type(), 1, 0);
-   LLVMValueRef ind2 = LLVMConstInt(LLVMInt32Type(), 2, 0);
-   LLVMValueRef ind3 = LLVMConstInt(LLVMInt32Type(), 3, 0);
+   LLVMBuilderRef builder = gallivm->builder;
+   LLVMValueRef attr_index = lp_build_const_int32(gallivm, attrib);
+   LLVMValueRef ind0 = lp_build_const_int32(gallivm, 0);
+   LLVMValueRef ind1 = lp_build_const_int32(gallivm, 1);
+   LLVMValueRef ind2 = lp_build_const_int32(gallivm, 2);
+   LLVMValueRef ind3 = lp_build_const_int32(gallivm, 3);
    LLVMValueRef io0_ptr, io1_ptr, io2_ptr, io3_ptr;
    LLVMValueRef clipmask0, clipmask1, clipmask2, clipmask3;
    
@@ -662,20 +766,21 @@ store_aos_array(LLVMBuilderRef builder,
                    io_ptr, ind0, ind1, ind2, ind3, clipmask0, clipmask1, clipmask2, clipmask3);
 #endif
    /* store for each of the 4 vertices */
-   store_aos(builder, io0_ptr, attr_index, aos[0], clipmask0);
-   store_aos(builder, io1_ptr, attr_index, aos[1], clipmask1);
-   store_aos(builder, io2_ptr, attr_index, aos[2], clipmask2);
-   store_aos(builder, io3_ptr, attr_index, aos[3], clipmask3);
+   store_aos(gallivm, io0_ptr, attr_index, aos[0], clipmask0);
+   store_aos(gallivm, io1_ptr, attr_index, aos[1], clipmask1);
+   store_aos(gallivm, io2_ptr, attr_index, aos[2], clipmask2);
+   store_aos(gallivm, io3_ptr, attr_index, aos[3], clipmask3);
 }
 
 static void
-convert_to_aos(LLVMBuilderRef builder,
+convert_to_aos(struct gallivm_state *gallivm,
                LLVMValueRef io,
                LLVMValueRef (*outputs)[NUM_CHANNELS],
                LLVMValueRef clipmask,
                int num_outputs,
                int max_vertices)
 {
+   LLVMBuilderRef builder = gallivm->builder;
    unsigned chan, attrib;
 
 #if DEBUG_STORE
@@ -696,8 +801,8 @@ convert_to_aos(LLVMBuilderRef builder,
          } else
             soa[chan] = 0;
       }
-      soa_to_aos(builder, soa, aos);
-      store_aos_array(builder,
+      soa_to_aos(gallivm, soa, aos);
+      store_aos_array(gallivm,
                       io,
                       aos,
                       attrib,
@@ -715,10 +820,11 @@ convert_to_aos(LLVMBuilderRef builder,
  * rather than extracting each element one by one.
  */
 static void
-store_clip(LLVMBuilderRef builder,
+store_clip(struct gallivm_state *gallivm,
            LLVMValueRef io_ptr,           
            LLVMValueRef (*outputs)[NUM_CHANNELS])
 {
+   LLVMBuilderRef builder = gallivm->builder;
    LLVMValueRef out[4];
    LLVMValueRef indices[2]; 
    LLVMValueRef io0_ptr, io1_ptr, io2_ptr, io3_ptr;
@@ -727,13 +833,13 @@ store_clip(LLVMBuilderRef builder,
    LLVMValueRef out0elem, out1elem, out2elem, out3elem;
    int i;
 
-   LLVMValueRef ind0 = LLVMConstInt(LLVMInt32Type(), 0, 0);
-   LLVMValueRef ind1 = LLVMConstInt(LLVMInt32Type(), 1, 0);
-   LLVMValueRef ind2 = LLVMConstInt(LLVMInt32Type(), 2, 0);
-   LLVMValueRef ind3 = LLVMConstInt(LLVMInt32Type(), 3, 0);
+   LLVMValueRef ind0 = lp_build_const_int32(gallivm, 0);
+   LLVMValueRef ind1 = lp_build_const_int32(gallivm, 1);
+   LLVMValueRef ind2 = lp_build_const_int32(gallivm, 2);
+   LLVMValueRef ind3 = lp_build_const_int32(gallivm, 3);
    
-   indices[0] = LLVMConstInt(LLVMInt32Type(), 0, 0);
-   indices[1] = LLVMConstInt(LLVMInt32Type(), 0, 0);
+   indices[0] =
+   indices[1] = lp_build_const_int32(gallivm, 0);
    
    out[0] = LLVMBuildLoad(builder, outputs[0][0], ""); /*x0 x1 x2 x3*/
    out[1] = LLVMBuildLoad(builder, outputs[0][1], ""); /*y0 y1 y2 y3*/
@@ -745,29 +851,21 @@ store_clip(LLVMBuilderRef builder,
    io2_ptr = LLVMBuildGEP(builder, io_ptr, &ind2, 1, "");
    io3_ptr = LLVMBuildGEP(builder, io_ptr, &ind3, 1, "");
 
-   clip_ptr0 = draw_jit_header_clip(builder, io0_ptr);
-   clip_ptr1 = draw_jit_header_clip(builder, io1_ptr);
-   clip_ptr2 = draw_jit_header_clip(builder, io2_ptr);
-   clip_ptr3 = draw_jit_header_clip(builder, io3_ptr);
+   clip_ptr0 = draw_jit_header_clip(gallivm, io0_ptr);
+   clip_ptr1 = draw_jit_header_clip(gallivm, io1_ptr);
+   clip_ptr2 = draw_jit_header_clip(gallivm, io2_ptr);
+   clip_ptr3 = draw_jit_header_clip(gallivm, io3_ptr);
 
    for (i = 0; i<4; i++){
-      clip0_ptr = LLVMBuildGEP(builder, clip_ptr0,
-                               indices, 2, ""); //x0
-      clip1_ptr = LLVMBuildGEP(builder, clip_ptr1,
-                               indices, 2, ""); //x1
-      clip2_ptr = LLVMBuildGEP(builder, clip_ptr2,
-                               indices, 2, ""); //x2
-      clip3_ptr = LLVMBuildGEP(builder, clip_ptr3,
-                               indices, 2, ""); //x3
-
-      out0elem = LLVMBuildExtractElement(builder, out[i],
-                                         ind0, ""); //x0
-      out1elem = LLVMBuildExtractElement(builder, out[i],
-                                         ind1, ""); //x1
-      out2elem = LLVMBuildExtractElement(builder, out[i],
-                                         ind2, ""); //x2
-      out3elem = LLVMBuildExtractElement(builder, out[i],
-                                         ind3, ""); //x3
+      clip0_ptr = LLVMBuildGEP(builder, clip_ptr0, indices, 2, ""); /* x0 */
+      clip1_ptr = LLVMBuildGEP(builder, clip_ptr1, indices, 2, ""); /* x1 */
+      clip2_ptr = LLVMBuildGEP(builder, clip_ptr2, indices, 2, ""); /* x2 */
+      clip3_ptr = LLVMBuildGEP(builder, clip_ptr3, indices, 2, ""); /* x3 */
+
+      out0elem = LLVMBuildExtractElement(builder, out[i], ind0, ""); /* x0 */
+      out1elem = LLVMBuildExtractElement(builder, out[i], ind1, ""); /* x1 */
+      out2elem = LLVMBuildExtractElement(builder, out[i], ind2, ""); /* x2 */
+      out3elem = LLVMBuildExtractElement(builder, out[i], ind3, ""); /* x3 */
   
       LLVMBuildStore(builder, out0elem, clip0_ptr);
       LLVMBuildStore(builder, out1elem, clip1_ptr);
@@ -781,16 +879,19 @@ store_clip(LLVMBuilderRef builder,
 
 /* Equivalent of _mm_set1_ps(a)
  */
-static LLVMValueRef vec4f_from_scalar(LLVMBuilderRef bld,
-				      LLVMValueRef a,
-				      const char *name)
+static LLVMValueRef
+vec4f_from_scalar(struct gallivm_state *gallivm,
+                  LLVMValueRef a,
+                  const char *name)
 {
-   LLVMValueRef res = LLVMGetUndef(LLVMVectorType(LLVMFloatType(), 4));
+   LLVMTypeRef float_type = LLVMFloatTypeInContext(gallivm->context);
+   LLVMValueRef res = LLVMGetUndef(LLVMVectorType(float_type, 4));
    int i;
 
    for(i = 0; i < 4; ++i) {
-      LLVMValueRef index = LLVMConstInt(LLVMInt32Type(), i, 0);
-      res = LLVMBuildInsertElement(bld, res, a, index, i == 3 ? name : "");
+      LLVMValueRef index = lp_build_const_int32(gallivm, i);
+      res = LLVMBuildInsertElement(gallivm->builder, res, a,
+                                   index, i == 3 ? name : "");
    }
 
    return res;
@@ -806,10 +907,11 @@ generate_viewport(struct draw_llvm *llvm,
                   LLVMValueRef context_ptr)
 {
    int i;
+   struct gallivm_state *gallivm = llvm->gallivm;
    struct lp_type f32_type = lp_type_float_vec(32);
    LLVMValueRef out3 = LLVMBuildLoad(builder, outputs[0][3], ""); /*w0 w1 w2 w3*/   
-   LLVMValueRef const1 = lp_build_const_vec(f32_type, 1.0);       /*1.0 1.0 1.0 1.0*/ 
-   LLVMValueRef vp_ptr = draw_jit_context_viewport(builder, context_ptr);
+   LLVMValueRef const1 = lp_build_const_vec(gallivm, f32_type, 1.0);       /*1.0 1.0 1.0 1.0*/ 
+   LLVMValueRef vp_ptr = draw_jit_context_viewport(gallivm, context_ptr);
 
    /* for 1/w convention*/
    out3 = LLVMBuildFDiv(builder, const1, out3, "");
@@ -824,14 +926,14 @@ generate_viewport(struct draw_llvm *llvm,
       LLVMValueRef trans_i;
       LLVMValueRef index;
       
-      index = LLVMConstInt(LLVMInt32Type(), i, 0);
+      index = lp_build_const_int32(gallivm, i);
       scale_i = LLVMBuildGEP(builder, vp_ptr, &index, 1, "");
 
-      index = LLVMConstInt(LLVMInt32Type(), i+4, 0);
+      index = lp_build_const_int32(gallivm, i+4);
       trans_i = LLVMBuildGEP(builder, vp_ptr, &index, 1, "");
 
-      scale = vec4f_from_scalar(builder, LLVMBuildLoad(builder, scale_i, ""), "scale");
-      trans = vec4f_from_scalar(builder, LLVMBuildLoad(builder, trans_i, ""), "trans");
+      scale = vec4f_from_scalar(gallivm, LLVMBuildLoad(builder, scale_i, ""), "scale");
+      trans = vec4f_from_scalar(gallivm, LLVMBuildLoad(builder, trans_i, ""), "trans");
 
       /* divide by w */
       out = LLVMBuildFMul(builder, out, out3, "");
@@ -851,7 +953,7 @@ generate_viewport(struct draw_llvm *llvm,
  * Returns clipmask as 4xi32 bitmask for the 4 vertices
  */
 static LLVMValueRef 
-generate_clipmask(LLVMBuilderRef builder,
+generate_clipmask(struct gallivm_state *gallivm,
                   LLVMValueRef (*outputs)[NUM_CHANNELS],
                   boolean clip_xy,
                   boolean clip_z,
@@ -860,6 +962,7 @@ generate_clipmask(LLVMBuilderRef builder,
                   unsigned nr,
                   LLVMValueRef context_ptr)
 {
+   LLVMBuilderRef builder = gallivm->builder;
    LLVMValueRef mask; /* stores the <4xi32> clipmasks */     
    LLVMValueRef test, temp; 
    LLVMValueRef zero, shift;
@@ -870,10 +973,10 @@ generate_clipmask(LLVMBuilderRef builder,
 
    struct lp_type f32_type = lp_type_float_vec(32); 
 
-   mask = lp_build_const_int_vec(lp_type_int_vec(32), 0);
-   temp = lp_build_const_int_vec(lp_type_int_vec(32), 0);
-   zero = lp_build_const_vec(f32_type, 0);                    /* 0.0f 0.0f 0.0f 0.0f */
-   shift = lp_build_const_int_vec(lp_type_int_vec(32), 1);    /* 1 1 1 1 */
+   mask = lp_build_const_int_vec(gallivm, lp_type_int_vec(32), 0);
+   temp = lp_build_const_int_vec(gallivm, lp_type_int_vec(32), 0);
+   zero = lp_build_const_vec(gallivm, f32_type, 0);                    /* 0.0f 0.0f 0.0f 0.0f */
+   shift = lp_build_const_int_vec(gallivm, lp_type_int_vec(32), 1);    /* 1 1 1 1 */
 
    /* Assuming position stored at output[0] */
    pos_x = LLVMBuildLoad(builder, outputs[0][0], ""); /*x0 x1 x2 x3*/
@@ -884,92 +987,92 @@ generate_clipmask(LLVMBuilderRef builder,
    /* Cliptest, for hardwired planes */
    if (clip_xy){
       /* plane 1 */
-      test = lp_build_compare(builder, f32_type, PIPE_FUNC_GREATER, pos_x , pos_w);
+      test = lp_build_compare(gallivm, f32_type, PIPE_FUNC_GREATER, pos_x , pos_w);
       temp = shift;
       test = LLVMBuildAnd(builder, test, temp, ""); 
       mask = test;
    
       /* plane 2 */
       test = LLVMBuildFAdd(builder, pos_x, pos_w, "");
-      test = lp_build_compare(builder, f32_type, PIPE_FUNC_GREATER, zero, test);
+      test = lp_build_compare(gallivm, f32_type, PIPE_FUNC_GREATER, zero, test);
       temp = LLVMBuildShl(builder, temp, shift, "");
       test = LLVMBuildAnd(builder, test, temp, ""); 
       mask = LLVMBuildOr(builder, mask, test, "");
    
       /* plane 3 */
-      test = lp_build_compare(builder, f32_type, PIPE_FUNC_GREATER, pos_y, pos_w);
+      test = lp_build_compare(gallivm, f32_type, PIPE_FUNC_GREATER, pos_y, pos_w);
       temp = LLVMBuildShl(builder, temp, shift, "");
       test = LLVMBuildAnd(builder, test, temp, ""); 
       mask = LLVMBuildOr(builder, mask, test, "");
 
       /* plane 4 */
       test = LLVMBuildFAdd(builder, pos_y, pos_w, "");
-      test = lp_build_compare(builder, f32_type, PIPE_FUNC_GREATER, zero, test);
+      test = lp_build_compare(gallivm, f32_type, PIPE_FUNC_GREATER, zero, test);
       temp = LLVMBuildShl(builder, temp, shift, "");
       test = LLVMBuildAnd(builder, test, temp, ""); 
       mask = LLVMBuildOr(builder, mask, test, "");
    }
 
    if (clip_z){
-      temp = lp_build_const_int_vec(lp_type_int_vec(32), 16);
+      temp = lp_build_const_int_vec(gallivm, lp_type_int_vec(32), 16);
       if (clip_halfz){
          /* plane 5 */
-         test = lp_build_compare(builder, f32_type, PIPE_FUNC_GREATER, zero, pos_z);
+         test = lp_build_compare(gallivm, f32_type, PIPE_FUNC_GREATER, zero, pos_z);
          test = LLVMBuildAnd(builder, test, temp, ""); 
          mask = LLVMBuildOr(builder, mask, test, "");
       }  
       else{
          /* plane 5 */
          test = LLVMBuildFAdd(builder, pos_z, pos_w, "");
-         test = lp_build_compare(builder, f32_type, PIPE_FUNC_GREATER, zero, test);
+         test = lp_build_compare(gallivm, f32_type, PIPE_FUNC_GREATER, zero, test);
          test = LLVMBuildAnd(builder, test, temp, ""); 
          mask = LLVMBuildOr(builder, mask, test, "");
       }
       /* plane 6 */
-      test = lp_build_compare(builder, f32_type, PIPE_FUNC_GREATER, pos_z, pos_w);
+      test = lp_build_compare(gallivm, f32_type, PIPE_FUNC_GREATER, pos_z, pos_w);
       temp = LLVMBuildShl(builder, temp, shift, "");
       test = LLVMBuildAnd(builder, test, temp, ""); 
       mask = LLVMBuildOr(builder, mask, test, "");
    }   
 
    if (clip_user){
-      LLVMValueRef planes_ptr = draw_jit_context_planes(builder, context_ptr);
+      LLVMValueRef planes_ptr = draw_jit_context_planes(gallivm, context_ptr);
       LLVMValueRef indices[3];
-      temp = lp_build_const_int_vec(lp_type_int_vec(32), 32);
+      temp = lp_build_const_int_vec(gallivm, lp_type_int_vec(32), 32);
 
       /* userclip planes */
       for (i = 6; i < nr; i++) {
-         indices[0] = LLVMConstInt(LLVMInt32Type(), 0, 0);
-         indices[1] = LLVMConstInt(LLVMInt32Type(), i, 0);
+         indices[0] = lp_build_const_int32(gallivm, 0);
+         indices[1] = lp_build_const_int32(gallivm, i);
 
-         indices[2] = LLVMConstInt(LLVMInt32Type(), 0, 0);
+         indices[2] = lp_build_const_int32(gallivm, 0);
          plane_ptr = LLVMBuildGEP(builder, planes_ptr, indices, 3, "");
          plane1 = LLVMBuildLoad(builder, plane_ptr, "plane_x");
-         planes = vec4f_from_scalar(builder, plane1, "plane4_x");
+         planes = vec4f_from_scalar(gallivm, plane1, "plane4_x");
          sum = LLVMBuildFMul(builder, planes, pos_x, "");
 
-         indices[2] = LLVMConstInt(LLVMInt32Type(), 1, 0);
+         indices[2] = lp_build_const_int32(gallivm, 1);
          plane_ptr = LLVMBuildGEP(builder, planes_ptr, indices, 3, "");
          plane1 = LLVMBuildLoad(builder, plane_ptr, "plane_y"); 
-         planes = vec4f_from_scalar(builder, plane1, "plane4_y");
+         planes = vec4f_from_scalar(gallivm, plane1, "plane4_y");
          test = LLVMBuildFMul(builder, planes, pos_y, "");
          sum = LLVMBuildFAdd(builder, sum, test, "");
          
-         indices[2] = LLVMConstInt(LLVMInt32Type(), 2, 0);
+         indices[2] = lp_build_const_int32(gallivm, 2);
          plane_ptr = LLVMBuildGEP(builder, planes_ptr, indices, 3, "");
          plane1 = LLVMBuildLoad(builder, plane_ptr, "plane_z"); 
-         planes = vec4f_from_scalar(builder, plane1, "plane4_z");
+         planes = vec4f_from_scalar(gallivm, plane1, "plane4_z");
          test = LLVMBuildFMul(builder, planes, pos_z, "");
          sum = LLVMBuildFAdd(builder, sum, test, "");
 
-         indices[2] = LLVMConstInt(LLVMInt32Type(), 3, 0);
+         indices[2] = lp_build_const_int32(gallivm, 3);
          plane_ptr = LLVMBuildGEP(builder, planes_ptr, indices, 3, "");
          plane1 = LLVMBuildLoad(builder, plane_ptr, "plane_w"); 
-         planes = vec4f_from_scalar(builder, plane1, "plane4_w");
+         planes = vec4f_from_scalar(gallivm, plane1, "plane4_w");
          test = LLVMBuildFMul(builder, planes, pos_w, "");
          sum = LLVMBuildFAdd(builder, sum, test, "");
 
-         test = lp_build_compare(builder, f32_type, PIPE_FUNC_GREATER, zero, sum);
+         test = lp_build_compare(gallivm, f32_type, PIPE_FUNC_GREATER, zero, sum);
          temp = LLVMBuildShl(builder, temp, shift, "");
          test = LLVMBuildAnd(builder, test, temp, ""); 
          mask = LLVMBuildOr(builder, mask, test, "");
@@ -983,17 +1086,18 @@ generate_clipmask(LLVMBuilderRef builder,
  * Used zero/non-zero i32 value to represent boolean 
  */
 static void
-clipmask_bool(LLVMBuilderRef builder, 
+clipmask_bool(struct gallivm_state *gallivm,
               LLVMValueRef clipmask,
               LLVMValueRef ret_ptr)
 {
+   LLVMBuilderRef builder = gallivm->builder;
    LLVMValueRef ret = LLVMBuildLoad(builder, ret_ptr, "");   
    LLVMValueRef temp;
    int i;
 
    for (i=0; i<4; i++){   
       temp = LLVMBuildExtractElement(builder, clipmask,
-                                     LLVMConstInt(LLVMInt32Type(), i, 0) , "");
+                                     lp_build_const_int32(gallivm, i) , "");
       ret = LLVMBuildOr(builder, ret, temp, "");
    }
    
@@ -1003,6 +1107,9 @@ clipmask_bool(LLVMBuilderRef builder,
 static void
 draw_llvm_generate(struct draw_llvm *llvm, struct draw_llvm_variant *variant)
 {
+   struct gallivm_state *gallivm = llvm->gallivm;
+   LLVMContextRef context = gallivm->context;
+   LLVMTypeRef int32_type = LLVMInt32TypeInContext(context);
    LLVMTypeRef arg_types[8];
    LLVMTypeRef func_type;
    LLVMValueRef context_ptr;
@@ -1025,18 +1132,19 @@ draw_llvm_generate(struct draw_llvm *llvm, struct draw_llvm_variant *variant)
                              variant->key.clip_z  ||
                              variant->key.clip_user;
    
-   arg_types[0] = llvm->context_ptr_type;           /* context */
-   arg_types[1] = llvm->vertex_header_ptr_type;     /* vertex_header */
-   arg_types[2] = llvm->buffer_ptr_type;            /* vbuffers */
-   arg_types[3] = LLVMInt32Type();                  /* start */
-   arg_types[4] = LLVMInt32Type();                  /* count */
-   arg_types[5] = LLVMInt32Type();                  /* stride */
-   arg_types[6] = llvm->vb_ptr_type;                /* pipe_vertex_buffer's */
-   arg_types[7] = LLVMInt32Type();                  /* instance_id */
-
-   func_type = LLVMFunctionType(LLVMInt32Type(), arg_types, Elements(arg_types), 0);
-
-   variant->function = LLVMAddFunction(llvm->module, "draw_llvm_shader", func_type);
+   arg_types[0] = get_context_ptr_type(llvm);       /* context */
+   arg_types[1] = get_vertex_header_ptr_type(llvm); /* vertex_header */
+   arg_types[2] = get_buffer_ptr_type(llvm);        /* vbuffers */
+   arg_types[3] = int32_type;                       /* start */
+   arg_types[4] = int32_type;                       /* count */
+   arg_types[5] = int32_type;                       /* stride */
+   arg_types[6] = get_vb_ptr_type(llvm);            /* pipe_vertex_buffer's */
+   arg_types[7] = int32_type;                       /* instance_id */
+
+   func_type = LLVMFunctionType(int32_type, arg_types, Elements(arg_types), 0);
+
+   variant->function = LLVMAddFunction(gallivm->module, "draw_llvm_shader",
+                                       func_type);
    LLVMSetFunctionCallConv(variant->function, LLVMCCallConv);
    for(i = 0; i < Elements(arg_types); ++i)
       if(LLVMGetTypeKind(arg_types[i]) == LLVMPointerTypeKind)
@@ -1064,19 +1172,20 @@ draw_llvm_generate(struct draw_llvm *llvm, struct draw_llvm_variant *variant)
     * Function body
     */
 
-   block = LLVMAppendBasicBlock(variant->function, "entry");
-   builder = LLVMCreateBuilder();
+   block = LLVMAppendBasicBlockInContext(gallivm->context, variant->function, "entry");
+   builder = gallivm->builder;
+   assert(builder);
    LLVMPositionBuilderAtEnd(builder, block);
 
-   lp_build_context_init(&bld, builder, lp_type_int(32));
+   lp_build_context_init(&bld, llvm->gallivm, lp_type_int(32));
 
    end = lp_build_add(&bld, start, count);
 
-   step = LLVMConstInt(LLVMInt32Type(), max_vertices, 0);
+   step = lp_build_const_int32(gallivm, max_vertices);
 
    /* function will return non-zero i32 value if any clipped vertices */     
-   ret_ptr = lp_build_alloca(builder, LLVMInt32Type(), "");   
-   LLVMBuildStore(builder, LLVMConstInt(LLVMInt32Type(), 0, 0), ret_ptr);
+   ret_ptr = lp_build_alloca(gallivm, int32_type, "");   
+   LLVMBuildStore(builder, lp_build_const_int32(gallivm, 0), ret_ptr);
 
    /* code generated texture sampling */
    sampler = draw_llvm_sampler_soa_create(
@@ -1087,7 +1196,7 @@ draw_llvm_generate(struct draw_llvm *llvm, struct draw_llvm_variant *variant)
    lp_build_printf(builder, "start = %d, end = %d, step = %d\n",
                    start, end, step);
 #endif
-   lp_build_loop_begin(builder, start, &lp_loop);
+   lp_build_loop_begin(&lp_loop, llvm->gallivm, start);
    {
       LLVMValueRef inputs[PIPE_MAX_SHADER_INPUTS][NUM_CHANNELS];
       LLVMValueRef aos_attribs[PIPE_MAX_SHADER_INPUTS][NUM_CHANNELS] = { { 0 } };
@@ -1105,20 +1214,18 @@ draw_llvm_generate(struct draw_llvm *llvm, struct draw_llvm_variant *variant)
          LLVMValueRef true_index = LLVMBuildAdd(
             builder,
             lp_loop.counter,
-            LLVMConstInt(LLVMInt32Type(), i, 0), "");
+            lp_build_const_int32(gallivm, i), "");
          for (j = 0; j < draw->pt.nr_vertex_elements; ++j) {
             struct pipe_vertex_element *velem = &draw->pt.vertex_element[j];
-            LLVMValueRef vb_index = LLVMConstInt(LLVMInt32Type(),
-                                                 velem->vertex_buffer_index,
-                                                 0);
+            LLVMValueRef vb_index = lp_build_const_int32(gallivm, velem->vertex_buffer_index);
             LLVMValueRef vb = LLVMBuildGEP(builder, vb_ptr,
                                            &vb_index, 1, "");
-            generate_fetch(builder, vbuffers_ptr,
+            generate_fetch(llvm->gallivm, vbuffers_ptr,
                            &aos_attribs[j][i], velem, vb, true_index,
                            instance_id);
          }
       }
-      convert_to_soa(builder, aos_attribs, inputs,
+      convert_to_soa(gallivm, aos_attribs, inputs,
                      draw->pt.nr_vertex_elements);
 
       ptr_aos = (const LLVMValueRef (*)[NUM_CHANNELS]) inputs;
@@ -1130,12 +1237,12 @@ draw_llvm_generate(struct draw_llvm *llvm, struct draw_llvm_variant *variant)
                   sampler);
 
       /* store original positions in clip before further manipulation */
-      store_clip(builder, io, outputs);
+      store_clip(gallivm, io, outputs);
 
       /* do cliptest */
       if (enable_cliptest){
          /* allocate clipmask, assign it integer type */
-         clipmask = generate_clipmask(builder, outputs,
+         clipmask = generate_clipmask(gallivm, outputs,
                                       variant->key.clip_xy,
                                       variant->key.clip_z, 
                                       variant->key.clip_user,
@@ -1143,10 +1250,10 @@ draw_llvm_generate(struct draw_llvm *llvm, struct draw_llvm_variant *variant)
                                       variant->key.nr_planes,
                                       context_ptr);
          /* return clipping boolean value for function */
-         clipmask_bool(builder, clipmask, ret_ptr);
+         clipmask_bool(gallivm, clipmask, ret_ptr);
       }
       else{
-         clipmask = lp_build_const_int_vec(lp_type_int_vec(32), 0);    
+         clipmask = lp_build_const_int_vec(gallivm, lp_type_int_vec(32), 0);    
       }
       
       /* do viewport mapping */
@@ -1155,20 +1262,18 @@ draw_llvm_generate(struct draw_llvm *llvm, struct draw_llvm_variant *variant)
       }
 
       /* store clipmask in vertex header and positions in data */
-      convert_to_aos(builder, io, outputs, clipmask,
+      convert_to_aos(gallivm, io, outputs, clipmask,
                      draw->vs.vertex_shader->info.num_outputs,
                      max_vertices);
    }
 
-   lp_build_loop_end_cond(builder, end, step, LLVMIntUGE, &lp_loop);
+   lp_build_loop_end_cond(&lp_loop, end, step, LLVMIntUGE);
 
    sampler->destroy(sampler);
 
    ret = LLVMBuildLoad(builder, ret_ptr,"");
    LLVMBuildRet(builder, ret);
       
-   LLVMDisposeBuilder(builder);
-
    /*
     * Translate the LLVM IR into machine code.
     */
@@ -1179,14 +1284,14 @@ draw_llvm_generate(struct draw_llvm *llvm, struct draw_llvm_variant *variant)
    }
 #endif
 
-   LLVMRunFunctionPassManager(llvm->pass, variant->function);
+   LLVMRunFunctionPassManager(gallivm->passmgr, variant->function);
 
    if (gallivm_debug & GALLIVM_DEBUG_IR) {
       lp_debug_dump_value(variant->function);
       debug_printf("\n");
    }
 
-   code = LLVMGetPointerToGlobal(llvm->draw->engine, variant->function);
+   code = LLVMGetPointerToGlobal(gallivm->engine, variant->function);
    variant->jit_func = (draw_jit_vert_func)pointer_to_func(code);
 
    if (gallivm_debug & GALLIVM_DEBUG_ASM) {
@@ -1199,6 +1304,9 @@ draw_llvm_generate(struct draw_llvm *llvm, struct draw_llvm_variant *variant)
 static void
 draw_llvm_generate_elts(struct draw_llvm *llvm, struct draw_llvm_variant *variant)
 {
+   struct gallivm_state *gallivm = llvm->gallivm;
+   LLVMContextRef context = gallivm->context;
+   LLVMTypeRef int32_type = LLVMInt32TypeInContext(context);
    LLVMTypeRef arg_types[8];
    LLVMTypeRef func_type;
    LLVMValueRef context_ptr;
@@ -1222,18 +1330,18 @@ draw_llvm_generate_elts(struct draw_llvm *llvm, struct draw_llvm_variant *varian
                              variant->key.clip_z  ||
                              variant->key.clip_user;
    
-   arg_types[0] = llvm->context_ptr_type;               /* context */
-   arg_types[1] = llvm->vertex_header_ptr_type;         /* vertex_header */
-   arg_types[2] = llvm->buffer_ptr_type;                /* vbuffers */
-   arg_types[3] = LLVMPointerType(LLVMInt32Type(), 0);  /* fetch_elts * */
-   arg_types[4] = LLVMInt32Type();                      /* fetch_count */
-   arg_types[5] = LLVMInt32Type();                      /* stride */
-   arg_types[6] = llvm->vb_ptr_type;                    /* pipe_vertex_buffer's */
-   arg_types[7] = LLVMInt32Type();                      /* instance_id */
-
-   func_type = LLVMFunctionType(LLVMInt32Type(), arg_types, Elements(arg_types), 0);
-
-   variant->function_elts = LLVMAddFunction(llvm->module, "draw_llvm_shader_elts", func_type);
+   arg_types[0] = get_context_ptr_type(llvm);           /* context */
+   arg_types[1] = get_vertex_header_ptr_type(llvm);     /* vertex_header */
+   arg_types[2] = get_buffer_ptr_type(llvm);            /* vbuffers */
+   arg_types[3] = LLVMPointerType(int32_type, 0);       /* fetch_elts * */
+   arg_types[4] = int32_type;                           /* fetch_count */
+   arg_types[5] = int32_type;                           /* stride */
+   arg_types[6] = get_vb_ptr_type(llvm);                /* pipe_vertex_buffer's */
+   arg_types[7] = int32_type;                           /* instance_id */
+
+   func_type = LLVMFunctionType(int32_type, arg_types, Elements(arg_types), 0);
+
+   variant->function_elts = LLVMAddFunction(gallivm->module, "draw_llvm_shader_elts", func_type);
    LLVMSetFunctionCallConv(variant->function_elts, LLVMCCallConv);
    for(i = 0; i < Elements(arg_types); ++i)
       if(LLVMGetTypeKind(arg_types[i]) == LLVMPointerTypeKind)
@@ -1262,13 +1370,13 @@ draw_llvm_generate_elts(struct draw_llvm *llvm, struct draw_llvm_variant *varian
     * Function body
     */
 
-   block = LLVMAppendBasicBlock(variant->function_elts, "entry");
-   builder = LLVMCreateBuilder();
+   block = LLVMAppendBasicBlockInContext(gallivm->context, variant->function_elts, "entry");
+   builder = gallivm->builder;
    LLVMPositionBuilderAtEnd(builder, block);
 
-   lp_build_context_init(&bld, builder, lp_type_int(32));
+   lp_build_context_init(&bld, gallivm, lp_type_int(32));
 
-   step = LLVMConstInt(LLVMInt32Type(), max_vertices, 0);
+   step = lp_build_const_int32(gallivm, max_vertices);
 
    /* code generated texture sampling */
    sampler = draw_llvm_sampler_soa_create(
@@ -1276,14 +1384,14 @@ draw_llvm_generate_elts(struct draw_llvm *llvm, struct draw_llvm_variant *varian
       context_ptr);
 
    fetch_max = LLVMBuildSub(builder, fetch_count,
-                            LLVMConstInt(LLVMInt32Type(), 1, 0),
+                            lp_build_const_int32(gallivm, 1),
                             "fetch_max");
 
    /* function returns non-zero i32 value if any clipped vertices */
-   ret_ptr = lp_build_alloca(builder, LLVMInt32Type(), ""); 
-   LLVMBuildStore(builder, LLVMConstInt(LLVMInt32Type(), 0, 0), ret_ptr);
+   ret_ptr = lp_build_alloca(gallivm, int32_type, ""); 
+   LLVMBuildStore(builder, lp_build_const_int32(gallivm, 0), ret_ptr);
 
-   lp_build_loop_begin(builder, LLVMConstInt(LLVMInt32Type(), 0, 0), &lp_loop);
+   lp_build_loop_begin(&lp_loop, gallivm, lp_build_const_int32(gallivm, 0));
    {
       LLVMValueRef inputs[PIPE_MAX_SHADER_INPUTS][NUM_CHANNELS];
       LLVMValueRef aos_attribs[PIPE_MAX_SHADER_INPUTS][NUM_CHANNELS] = { { 0 } };
@@ -1301,7 +1409,7 @@ draw_llvm_generate_elts(struct draw_llvm *llvm, struct draw_llvm_variant *varian
          LLVMValueRef true_index = LLVMBuildAdd(
             builder,
             lp_loop.counter,
-            LLVMConstInt(LLVMInt32Type(), i, 0), "");
+            lp_build_const_int32(gallivm, i), "");
          LLVMValueRef fetch_ptr;
 
          /* make sure we're not out of bounds which can happen
@@ -1314,17 +1422,15 @@ draw_llvm_generate_elts(struct draw_llvm *llvm, struct draw_llvm_variant *varian
          true_index = LLVMBuildLoad(builder, fetch_ptr, "fetch_elt");
          for (j = 0; j < draw->pt.nr_vertex_elements; ++j) {
             struct pipe_vertex_element *velem = &draw->pt.vertex_element[j];
-            LLVMValueRef vb_index = LLVMConstInt(LLVMInt32Type(),
-                                                 velem->vertex_buffer_index,
-                                                 0);
+            LLVMValueRef vb_index = lp_build_const_int32(gallivm, velem->vertex_buffer_index);
             LLVMValueRef vb = LLVMBuildGEP(builder, vb_ptr,
                                            &vb_index, 1, "");
-            generate_fetch(builder, vbuffers_ptr,
+            generate_fetch(gallivm, vbuffers_ptr,
                            &aos_attribs[j][i], velem, vb, true_index,
                            instance_id);
          }
       }
-      convert_to_soa(builder, aos_attribs, inputs,
+      convert_to_soa(gallivm, aos_attribs, inputs,
                      draw->pt.nr_vertex_elements);
 
       ptr_aos = (const LLVMValueRef (*)[NUM_CHANNELS]) inputs;
@@ -1336,12 +1442,12 @@ draw_llvm_generate_elts(struct draw_llvm *llvm, struct draw_llvm_variant *varian
                   sampler);
 
       /* store original positions in clip before further manipulation */
-      store_clip(builder, io, outputs);
+      store_clip(gallivm, io, outputs);
 
       /* do cliptest */
       if (enable_cliptest){
          /* allocate clipmask, assign it integer type */
-         clipmask = generate_clipmask(builder, outputs,
+         clipmask = generate_clipmask(gallivm, outputs,
                                       variant->key.clip_xy,
                                       variant->key.clip_z, 
                                       variant->key.clip_user,
@@ -1349,10 +1455,10 @@ draw_llvm_generate_elts(struct draw_llvm *llvm, struct draw_llvm_variant *varian
                                       variant->key.nr_planes,
                                       context_ptr);
          /* return clipping boolean value for function */
-         clipmask_bool(builder, clipmask, ret_ptr);
+         clipmask_bool(gallivm, clipmask, ret_ptr);
       }
       else{
-         clipmask = lp_build_const_int_vec(lp_type_int_vec(32), 0);
+         clipmask = lp_build_const_int_vec(gallivm, lp_type_int_vec(32), 0);
       }
       
       /* do viewport mapping */
@@ -1364,20 +1470,18 @@ draw_llvm_generate_elts(struct draw_llvm *llvm, struct draw_llvm_variant *varian
        * original positions in clip 
        * and transformed positions in data 
        */   
-      convert_to_aos(builder, io, outputs, clipmask,
+      convert_to_aos(gallivm, io, outputs, clipmask,
                      draw->vs.vertex_shader->info.num_outputs,
                      max_vertices);
    }
 
-   lp_build_loop_end_cond(builder, fetch_count, step, LLVMIntUGE, &lp_loop);
+   lp_build_loop_end_cond(&lp_loop, fetch_count, step, LLVMIntUGE);
 
    sampler->destroy(sampler);
 
    ret = LLVMBuildLoad(builder, ret_ptr,"");   
    LLVMBuildRet(builder, ret);
    
-   LLVMDisposeBuilder(builder);
-
    /*
     * Translate the LLVM IR into machine code.
     */
@@ -1388,14 +1492,14 @@ draw_llvm_generate_elts(struct draw_llvm *llvm, struct draw_llvm_variant *varian
    }
 #endif
 
-   LLVMRunFunctionPassManager(llvm->pass, variant->function_elts);
+   LLVMRunFunctionPassManager(gallivm->passmgr, variant->function_elts);
 
    if (gallivm_debug & GALLIVM_DEBUG_IR) {
       lp_debug_dump_value(variant->function_elts);
       debug_printf("\n");
    }
 
-   code = LLVMGetPointerToGlobal(llvm->draw->engine, variant->function_elts);
+   code = LLVMGetPointerToGlobal(gallivm->engine, variant->function_elts);
    variant->jit_func_elts = (draw_jit_vert_func_elts)pointer_to_func(code);
 
    if (gallivm_debug & GALLIVM_DEBUG_ASM) {
@@ -1504,19 +1608,16 @@ void
 draw_llvm_destroy_variant(struct draw_llvm_variant *variant)
 {
    struct draw_llvm *llvm = variant->llvm;
-   struct draw_context *draw = llvm->draw;
 
    if (variant->function_elts) {
-      if (variant->function_elts)
-         LLVMFreeMachineCodeForFunction(draw->engine,
-                                        variant->function_elts);
+      LLVMFreeMachineCodeForFunction(llvm->gallivm->engine,
+                                     variant->function_elts);
       LLVMDeleteFunction(variant->function_elts);
    }
 
    if (variant->function) {
-      if (variant->function)
-         LLVMFreeMachineCodeForFunction(draw->engine,
-                                        variant->function);
+      LLVMFreeMachineCodeForFunction(llvm->gallivm->engine,
+                                     variant->function);
       LLVMDeleteFunction(variant->function);
    }
 
diff --git a/src/gallium/auxiliary/draw/draw_llvm.h b/src/gallium/auxiliary/draw/draw_llvm.h
index c3c30c07c6..9f038f1f04 100644
--- a/src/gallium/auxiliary/draw/draw_llvm.h
+++ b/src/gallium/auxiliary/draw/draw_llvm.h
@@ -103,41 +103,41 @@ struct draw_jit_context
 };
 
 
-#define draw_jit_context_vs_constants(_builder, _ptr) \
-   lp_build_struct_get(_builder, _ptr, 0, "vs_constants")
+#define draw_jit_context_vs_constants(_gallivm, _ptr) \
+   lp_build_struct_get(_gallivm, _ptr, 0, "vs_constants")
 
-#define draw_jit_context_gs_constants(_builder, _ptr) \
-   lp_build_struct_get(_builder, _ptr, 1, "gs_constants")
+#define draw_jit_context_gs_constants(_gallivm, _ptr) \
+   lp_build_struct_get(_gallivm, _ptr, 1, "gs_constants")
 
-#define draw_jit_context_planes(_builder, _ptr) \
-   lp_build_struct_get(_builder, _ptr, 2, "planes")
+#define draw_jit_context_planes(_gallivm, _ptr) \
+   lp_build_struct_get(_gallivm, _ptr, 2, "planes")
 
-#define draw_jit_context_viewport(_builder, _ptr) \
-   lp_build_struct_get(_builder, _ptr, 3, "viewport")
+#define draw_jit_context_viewport(_gallivm, _ptr) \
+   lp_build_struct_get(_gallivm, _ptr, 3, "viewport")
 
 #define DRAW_JIT_CTX_TEXTURES 4
 
-#define draw_jit_context_textures(_builder, _ptr) \
-   lp_build_struct_get_ptr(_builder, _ptr, DRAW_JIT_CTX_TEXTURES, "textures")
+#define draw_jit_context_textures(_gallivm, _ptr) \
+   lp_build_struct_get_ptr(_gallivm, _ptr, DRAW_JIT_CTX_TEXTURES, "textures")
 
-#define draw_jit_header_id(_builder, _ptr)              \
-   lp_build_struct_get_ptr(_builder, _ptr, 0, "id")
+#define draw_jit_header_id(_gallivm, _ptr)              \
+   lp_build_struct_get_ptr(_gallivm, _ptr, 0, "id")
 
-#define draw_jit_header_clip(_builder, _ptr) \
-   lp_build_struct_get_ptr(_builder, _ptr, 1, "clip")
+#define draw_jit_header_clip(_gallivm, _ptr) \
+   lp_build_struct_get_ptr(_gallivm, _ptr, 1, "clip")
 
-#define draw_jit_header_data(_builder, _ptr)            \
-   lp_build_struct_get_ptr(_builder, _ptr, 2, "data")
+#define draw_jit_header_data(_gallivm, _ptr)            \
+   lp_build_struct_get_ptr(_gallivm, _ptr, 2, "data")
 
 
-#define draw_jit_vbuffer_stride(_builder, _ptr)         \
-   lp_build_struct_get(_builder, _ptr, 0, "stride")
+#define draw_jit_vbuffer_stride(_gallivm, _ptr)         \
+   lp_build_struct_get(_gallivm, _ptr, 0, "stride")
 
-#define draw_jit_vbuffer_max_index(_builder, _ptr)      \
-   lp_build_struct_get(_builder, _ptr, 1, "max_index")
+#define draw_jit_vbuffer_max_index(_gallivm, _ptr)      \
+   lp_build_struct_get(_gallivm, _ptr, 1, "max_index")
 
-#define draw_jit_vbuffer_offset(_builder, _ptr)         \
-   lp_build_struct_get(_builder, _ptr, 2, "buffer_offset")
+#define draw_jit_vbuffer_offset(_gallivm, _ptr)         \
+   lp_build_struct_get(_gallivm, _ptr, 2, "buffer_offset")
 
 
 typedef int
@@ -229,7 +229,6 @@ struct draw_llvm_variant
 
    /* key is variable-sized, must be last */
    struct draw_llvm_variant_key key;
-   /* key is variable-sized, must be last */
 };
 
 struct llvm_vertex_shader {
@@ -246,21 +245,19 @@ struct draw_llvm {
 
    struct draw_jit_context jit_context;
 
+   struct gallivm_state *gallivm;
+
    struct draw_llvm_variant_list_item vs_variants_list;
    int nr_variants;
 
-   LLVMModuleRef module;
-   LLVMExecutionEngineRef engine;
-   LLVMModuleProviderRef provider;
-   LLVMTargetDataRef target;
-   LLVMPassManagerRef pass;
-
+   /* LLVM JIT builder types */
    LLVMTypeRef context_ptr_type;
-   LLVMTypeRef vertex_header_ptr_type;
    LLVMTypeRef buffer_ptr_type;
    LLVMTypeRef vb_ptr_type;
+   LLVMTypeRef vertex_header_ptr_type;
 };
 
+
 static INLINE struct llvm_vertex_shader *
 llvm_vertex_shader(struct draw_vertex_shader *vs)
 {
@@ -269,7 +266,7 @@ llvm_vertex_shader(struct draw_vertex_shader *vs)
 
 
 struct draw_llvm *
-draw_llvm_create(struct draw_context *draw);
+draw_llvm_create(struct draw_context *draw, struct gallivm_state *gallivm);
 
 void
 draw_llvm_destroy(struct draw_llvm *llvm);
@@ -286,7 +283,7 @@ struct draw_llvm_variant_key *
 draw_llvm_make_variant_key(struct draw_llvm *llvm, char *store);
 
 LLVMValueRef
-draw_llvm_translate_from(LLVMBuilderRef builder,
+draw_llvm_translate_from(struct gallivm_state *gallivm,
                          LLVMValueRef vbuffer,
                          enum pipe_format from_format);
 
diff --git a/src/gallium/auxiliary/draw/draw_llvm_sample.c b/src/gallium/auxiliary/draw/draw_llvm_sample.c
index ac1fbb179c..574c7cc452 100644
--- a/src/gallium/auxiliary/draw/draw_llvm_sample.c
+++ b/src/gallium/auxiliary/draw/draw_llvm_sample.c
@@ -32,6 +32,7 @@
 
 #include "pipe/p_defines.h"
 #include "pipe/p_shader_tokens.h"
+#include "gallivm/lp_bld_const.h"
 #include "gallivm/lp_bld_debug.h"
 #include "gallivm/lp_bld_type.h"
 #include "gallivm/lp_bld_sample.h"
@@ -84,12 +85,13 @@ struct draw_llvm_sampler_soa
  */
 static LLVMValueRef
 draw_llvm_texture_member(const struct lp_sampler_dynamic_state *base,
-                         LLVMBuilderRef builder,
+                         struct gallivm_state *gallivm,
                          unsigned unit,
                          unsigned member_index,
                          const char *member_name,
                          boolean emit_load)
 {
+   LLVMBuilderRef builder = gallivm->builder;
    struct draw_llvm_sampler_dynamic_state *state =
       (struct draw_llvm_sampler_dynamic_state *)base;
    LLVMValueRef indices[4];
@@ -99,13 +101,13 @@ draw_llvm_texture_member(const struct lp_sampler_dynamic_state *base,
    debug_assert(unit < PIPE_MAX_VERTEX_SAMPLERS);
 
    /* context[0] */
-   indices[0] = LLVMConstInt(LLVMInt32Type(), 0, 0);
+   indices[0] = lp_build_const_int32(gallivm, 0);
    /* context[0].textures */
-   indices[1] = LLVMConstInt(LLVMInt32Type(), DRAW_JIT_CTX_TEXTURES, 0);
+   indices[1] = lp_build_const_int32(gallivm, DRAW_JIT_CTX_TEXTURES);
    /* context[0].textures[unit] */
-   indices[2] = LLVMConstInt(LLVMInt32Type(), unit, 0);
+   indices[2] = lp_build_const_int32(gallivm, unit);
    /* context[0].textures[unit].member */
-   indices[3] = LLVMConstInt(LLVMInt32Type(), member_index, 0);
+   indices[3] = lp_build_const_int32(gallivm, member_index);
 
    ptr = LLVMBuildGEP(builder, state->context_ptr, indices, Elements(indices), "");
 
@@ -132,10 +134,10 @@ draw_llvm_texture_member(const struct lp_sampler_dynamic_state *base,
 #define DRAW_LLVM_TEXTURE_MEMBER(_name, _index, _emit_load)  \
    static LLVMValueRef \
    draw_llvm_texture_##_name( const struct lp_sampler_dynamic_state *base, \
-                              LLVMBuilderRef builder,                   \
+                              struct gallivm_state *gallivm,               \
                               unsigned unit)                            \
    { \
-      return draw_llvm_texture_member(base, builder, unit, _index, #_name, _emit_load ); \
+      return draw_llvm_texture_member(base, gallivm, unit, _index, #_name, _emit_load ); \
    }
 
 
@@ -165,7 +167,7 @@ draw_llvm_sampler_soa_destroy(struct lp_build_sampler_soa *sampler)
  */
 static void
 draw_llvm_sampler_soa_emit_fetch_texel(const struct lp_build_sampler_soa *base,
-                                       LLVMBuilderRef builder,
+                                       struct gallivm_state *gallivm,
                                        struct lp_type type,
                                        unsigned unit,
                                        unsigned num_coords,
@@ -180,7 +182,7 @@ draw_llvm_sampler_soa_emit_fetch_texel(const struct lp_build_sampler_soa *base,
 
    assert(unit < PIPE_MAX_VERTEX_SAMPLERS);
 
-   lp_build_sample_soa(builder,
+   lp_build_sample_soa(gallivm,
                        &sampler->dynamic_state.static_state[unit],
                        &sampler->dynamic_state.base,
                        type,
diff --git a/src/gallium/auxiliary/draw/draw_llvm_translate.c b/src/gallium/auxiliary/draw/draw_llvm_translate.c
index 5171327ce2..77d0af7473 100644
--- a/src/gallium/auxiliary/draw/draw_llvm_translate.c
+++ b/src/gallium/auxiliary/draw/draw_llvm_translate.c
@@ -3,6 +3,7 @@
 
 #include "draw_llvm.h"
 
+#include "gallivm/lp_bld_const.h"
 #include "gallivm/lp_bld_struct.h"
 #include "gallivm/lp_bld_format.h"
 #include "gallivm/lp_bld_debug.h"
@@ -16,272 +17,279 @@
 #define DRAW_DBG 0
 
 static  LLVMValueRef
-from_64_float(LLVMBuilderRef builder, LLVMValueRef val)
+from_64_float(struct gallivm_state *gallivm, LLVMValueRef val)
 {
-   LLVMValueRef bc = LLVMBuildBitCast(builder, val,
-                                      LLVMPointerType(LLVMDoubleType(), 0) , "");
-   LLVMValueRef l = LLVMBuildLoad(builder, bc, "");
-   return LLVMBuildFPTrunc(builder, l, LLVMFloatType(), "");
+   LLVMValueRef bc = LLVMBuildBitCast(gallivm->builder, val,
+                                      LLVMPointerType(LLVMDoubleTypeInContext(gallivm->context), 0) , "");
+   LLVMValueRef l = LLVMBuildLoad(gallivm->builder, bc, "");
+   return LLVMBuildFPTrunc(gallivm->builder, l, LLVMFloatTypeInContext(gallivm->context), "");
 }
 
 static LLVMValueRef
-from_32_float(LLVMBuilderRef builder, LLVMValueRef val)
+from_32_float(struct gallivm_state *gallivm, LLVMValueRef val)
 {
-   LLVMValueRef bc = LLVMBuildBitCast(builder, val,
-                                      LLVMPointerType(LLVMFloatType(), 0) , "");
-   return LLVMBuildLoad(builder, bc, "");
+   LLVMValueRef bc = LLVMBuildBitCast(gallivm->builder, val,
+                                      LLVMPointerType(LLVMFloatTypeInContext(gallivm->context), 0) , "");
+   return LLVMBuildLoad(gallivm->builder, bc, "");
 }
 
 static INLINE LLVMValueRef
-from_8_uscaled(LLVMBuilderRef builder, LLVMValueRef val)
+from_8_uscaled(struct gallivm_state *gallivm, LLVMValueRef val)
 {
-   LLVMValueRef l = LLVMBuildLoad(builder, val, "");
-   return LLVMBuildUIToFP(builder, l, LLVMFloatType(), "");
+   LLVMValueRef l = LLVMBuildLoad(gallivm->builder, val, "");
+   return LLVMBuildUIToFP(gallivm->builder, l, LLVMFloatTypeInContext(gallivm->context), "");
 }
 
 static INLINE LLVMValueRef
-from_16_uscaled(LLVMBuilderRef builder, LLVMValueRef val)
+from_16_uscaled(struct gallivm_state *gallivm, LLVMValueRef val)
 {
-   LLVMValueRef bc = LLVMBuildBitCast(builder, val,
-                                      LLVMPointerType(LLVMIntType(16), 0) , "");
-   LLVMValueRef l = LLVMBuildLoad(builder, bc, "");
-   return LLVMBuildUIToFP(builder, l, LLVMFloatType(), "");
+   LLVMValueRef bc = LLVMBuildBitCast(gallivm->builder, val,
+                                      LLVMPointerType(LLVMIntTypeInContext(gallivm->context, 16), 0) , "");
+   LLVMValueRef l = LLVMBuildLoad(gallivm->builder, bc, "");
+   return LLVMBuildUIToFP(gallivm->builder, l, LLVMFloatTypeInContext(gallivm->context), "");
 }
 
 static INLINE LLVMValueRef
-from_32_uscaled(LLVMBuilderRef builder, LLVMValueRef val)
+from_32_uscaled(struct gallivm_state *gallivm, LLVMValueRef val)
 {
-   LLVMValueRef bc = LLVMBuildBitCast(builder, val,
-                                      LLVMPointerType(LLVMIntType(32), 0) , "");
-   LLVMValueRef l = LLVMBuildLoad(builder, bc, "");
-   return LLVMBuildUIToFP(builder, l, LLVMFloatType(), "");
+   LLVMValueRef bc = LLVMBuildBitCast(gallivm->builder, val,
+                                      LLVMPointerType(LLVMIntTypeInContext(gallivm->context, 32), 0) , "");
+   LLVMValueRef l = LLVMBuildLoad(gallivm->builder, bc, "");
+   return LLVMBuildUIToFP(gallivm->builder, l, LLVMFloatTypeInContext(gallivm->context), "");
 }
 
 static INLINE LLVMValueRef
-from_8_sscaled(LLVMBuilderRef builder, LLVMValueRef val)
+from_8_sscaled(struct gallivm_state *gallivm, LLVMValueRef val)
 {
-   LLVMValueRef l = LLVMBuildLoad(builder, val, "");
-   return LLVMBuildSIToFP(builder, l, LLVMFloatType(), "");
+   LLVMValueRef l = LLVMBuildLoad(gallivm->builder, val, "");
+   return LLVMBuildSIToFP(gallivm->builder, l, LLVMFloatTypeInContext(gallivm->context), "");
 }
 
 static INLINE LLVMValueRef
-from_16_sscaled(LLVMBuilderRef builder, LLVMValueRef val)
+from_16_sscaled(struct gallivm_state *gallivm, LLVMValueRef val)
 {
-   LLVMValueRef bc = LLVMBuildBitCast(builder, val,
-                                      LLVMPointerType(LLVMIntType(16), 0) , "");
-   LLVMValueRef l = LLVMBuildLoad(builder, bc, "");
-   return LLVMBuildSIToFP(builder, l, LLVMFloatType(), "");
+   LLVMValueRef bc = LLVMBuildBitCast(gallivm->builder, val,
+                                      LLVMPointerType(LLVMIntTypeInContext(gallivm->context, 16), 0) , "");
+   LLVMValueRef l = LLVMBuildLoad(gallivm->builder, bc, "");
+   return LLVMBuildSIToFP(gallivm->builder, l, LLVMFloatTypeInContext(gallivm->context), "");
 }
 
 static INLINE LLVMValueRef
-from_32_sscaled(LLVMBuilderRef builder, LLVMValueRef val)
+from_32_sscaled(struct gallivm_state *gallivm, LLVMValueRef val)
 {
-   LLVMValueRef bc = LLVMBuildBitCast(builder, val,
-                                      LLVMPointerType(LLVMIntType(32), 0) , "");
-   LLVMValueRef l = LLVMBuildLoad(builder, bc, "");
-   return LLVMBuildSIToFP(builder, l, LLVMFloatType(), "");
+   LLVMValueRef bc = LLVMBuildBitCast(gallivm->builder, val,
+                                      LLVMPointerType(LLVMIntTypeInContext(gallivm->context, 32), 0) , "");
+   LLVMValueRef l = LLVMBuildLoad(gallivm->builder, bc, "");
+   return LLVMBuildSIToFP(gallivm->builder, l, LLVMFloatTypeInContext(gallivm->context), "");
 }
 
 
 static INLINE LLVMValueRef
-from_8_unorm(LLVMBuilderRef builder, LLVMValueRef val)
+from_8_unorm(struct gallivm_state *gallivm, LLVMValueRef val)
 {
-   LLVMValueRef l = LLVMBuildLoad(builder, val, "");
-   LLVMValueRef uscaled = LLVMBuildUIToFP(builder, l, LLVMFloatType(), "");
-   return LLVMBuildFDiv(builder, uscaled,
-                        LLVMConstReal(LLVMFloatType(), 255.), "");
+   LLVMValueRef l = LLVMBuildLoad(gallivm->builder, val, "");
+   LLVMValueRef uscaled = LLVMBuildUIToFP(gallivm->builder, l, LLVMFloatTypeInContext(gallivm->context), "");
+   return LLVMBuildFDiv(gallivm->builder, uscaled,
+                        lp_build_const_float(gallivm, 255.), "");
 }
 
 static INLINE LLVMValueRef
-from_16_unorm(LLVMBuilderRef builder, LLVMValueRef val)
+from_16_unorm(struct gallivm_state *gallivm, LLVMValueRef val)
 {
-   LLVMValueRef bc = LLVMBuildBitCast(builder, val,
-                                      LLVMPointerType(LLVMIntType(16), 0) , "");
-   LLVMValueRef l = LLVMBuildLoad(builder, bc, "");
-   LLVMValueRef uscaled = LLVMBuildUIToFP(builder, l, LLVMFloatType(), "");
-   return LLVMBuildFDiv(builder, uscaled,
-                        LLVMConstReal(LLVMFloatType(), 65535.), "");
+   LLVMValueRef bc = LLVMBuildBitCast(gallivm->builder, val,
+                                      LLVMPointerType(LLVMIntTypeInContext(gallivm->context, 16), 0) , "");
+   LLVMValueRef l = LLVMBuildLoad(gallivm->builder, bc, "");
+   LLVMValueRef uscaled = LLVMBuildUIToFP(gallivm->builder, l, LLVMFloatTypeInContext(gallivm->context), "");
+   return LLVMBuildFDiv(gallivm->builder, uscaled,
+                        lp_build_const_float(gallivm, 65535.), "");
 }
 
 static INLINE LLVMValueRef
-from_32_unorm(LLVMBuilderRef builder, LLVMValueRef val)
+from_32_unorm(struct gallivm_state *gallivm, LLVMValueRef val)
 {
-   LLVMValueRef bc = LLVMBuildBitCast(builder, val,
-                                      LLVMPointerType(LLVMIntType(32), 0) , "");
-   LLVMValueRef l = LLVMBuildLoad(builder, bc, "");
-   LLVMValueRef uscaled = LLVMBuildUIToFP(builder, l, LLVMFloatType(), "");
+   LLVMValueRef bc = LLVMBuildBitCast(gallivm->builder, val,
+                                      LLVMPointerType(LLVMIntTypeInContext(gallivm->context, 32), 0) , "");
+   LLVMValueRef l = LLVMBuildLoad(gallivm->builder, bc, "");
+   LLVMValueRef uscaled = LLVMBuildUIToFP(gallivm->builder, l, LLVMFloatTypeInContext(gallivm->context), "");
 
-   return LLVMBuildFDiv(builder, uscaled,
-                        LLVMConstReal(LLVMFloatType(), 4294967295.), "");
+   return LLVMBuildFDiv(gallivm->builder, uscaled,
+                        lp_build_const_float(gallivm, 4294967295.), "");
 }
 
 static INLINE LLVMValueRef
-from_8_snorm(LLVMBuilderRef builder, LLVMValueRef val)
+from_8_snorm(struct gallivm_state *gallivm, LLVMValueRef val)
 {
-   LLVMValueRef l = LLVMBuildLoad(builder, val, "");
-   LLVMValueRef uscaled = LLVMBuildSIToFP(builder, l, LLVMFloatType(), "");
-   return LLVMBuildFDiv(builder, uscaled,
-                        LLVMConstReal(LLVMFloatType(), 127.0), "");
+   LLVMValueRef l = LLVMBuildLoad(gallivm->builder, val, "");
+   LLVMValueRef uscaled = LLVMBuildSIToFP(gallivm->builder, l, LLVMFloatTypeInContext(gallivm->context), "");
+   return LLVMBuildFDiv(gallivm->builder, uscaled,
+                        lp_build_const_float(gallivm, 127.0), "");
 }
 
 static INLINE LLVMValueRef
-from_16_snorm(LLVMBuilderRef builder, LLVMValueRef val)
+from_16_snorm(struct gallivm_state *gallivm, LLVMValueRef val)
 {
-   LLVMValueRef bc = LLVMBuildBitCast(builder, val,
-                                      LLVMPointerType(LLVMIntType(16), 0) , "");
-   LLVMValueRef l = LLVMBuildLoad(builder, bc, "");
-   LLVMValueRef uscaled = LLVMBuildSIToFP(builder, l, LLVMFloatType(), "");
-   return LLVMBuildFDiv(builder, uscaled,
-                        LLVMConstReal(LLVMFloatType(), 32767.0f), "");
+   LLVMValueRef bc = LLVMBuildBitCast(gallivm->builder, val,
+                                      LLVMPointerType(LLVMIntTypeInContext(gallivm->context, 16), 0) , "");
+   LLVMValueRef l = LLVMBuildLoad(gallivm->builder, bc, "");
+   LLVMValueRef uscaled = LLVMBuildSIToFP(gallivm->builder, l, LLVMFloatTypeInContext(gallivm->context), "");
+   return LLVMBuildFDiv(gallivm->builder, uscaled,
+                        lp_build_const_float(gallivm, 32767.0f), "");
 }
 
 static INLINE LLVMValueRef
-from_32_snorm(LLVMBuilderRef builder, LLVMValueRef val)
+from_32_snorm(struct gallivm_state *gallivm, LLVMValueRef val)
 {
-   LLVMValueRef bc = LLVMBuildBitCast(builder, val,
-                                      LLVMPointerType(LLVMIntType(32), 0) , "");
-   LLVMValueRef l = LLVMBuildLoad(builder, bc, "");
-   LLVMValueRef uscaled = LLVMBuildSIToFP(builder, l, LLVMFloatType(), "");
+   LLVMValueRef bc = LLVMBuildBitCast(gallivm->builder, val,
+                                      LLVMPointerType(LLVMIntTypeInContext(gallivm->context, 32), 0) , "");
+   LLVMValueRef l = LLVMBuildLoad(gallivm->builder, bc, "");
+   LLVMValueRef uscaled = LLVMBuildSIToFP(gallivm->builder, l, LLVMFloatTypeInContext(gallivm->context), "");
 
-   return LLVMBuildFDiv(builder, uscaled,
-                        LLVMConstReal(LLVMFloatType(), 2147483647.0), "");
+   return LLVMBuildFDiv(gallivm->builder, uscaled,
+                        lp_build_const_float(gallivm, 2147483647.0), "");
 }
 
 static INLINE LLVMValueRef
-from_32_fixed(LLVMBuilderRef builder, LLVMValueRef val)
+from_32_fixed(struct gallivm_state *gallivm, LLVMValueRef val)
 {
-   LLVMValueRef bc = LLVMBuildBitCast(builder, val,
-                                      LLVMPointerType(LLVMIntType(32), 0) , "");
-   LLVMValueRef l = LLVMBuildLoad(builder, bc, "");
-   LLVMValueRef uscaled = LLVMBuildSIToFP(builder, l, LLVMFloatType(), "");
+   LLVMValueRef bc = LLVMBuildBitCast(gallivm->builder, val,
+                                      LLVMPointerType(LLVMIntTypeInContext(gallivm->context, 32), 0) , "");
+   LLVMValueRef l = LLVMBuildLoad(gallivm->builder, bc, "");
+   LLVMValueRef uscaled = LLVMBuildSIToFP(gallivm->builder, l, LLVMFloatTypeInContext(gallivm->context), "");
 
-   return LLVMBuildFDiv(builder, uscaled,
-                        LLVMConstReal(LLVMFloatType(), 65536.0), "");
+   return LLVMBuildFDiv(gallivm->builder, uscaled,
+                        lp_build_const_float(gallivm, 65536.0), "");
 }
 
 static LLVMValueRef
-to_64_float(LLVMBuilderRef builder, LLVMValueRef fp)
+to_64_float(struct gallivm_state *gallivm, LLVMValueRef fp)
 {
-   LLVMValueRef l = LLVMBuildLoad(builder, fp, "");
-   return LLVMBuildFPExt(builder, l, LLVMDoubleType(), "");
+   LLVMValueRef l = LLVMBuildLoad(gallivm->builder, fp, "");
+   return LLVMBuildFPExt(gallivm->builder, l, LLVMDoubleTypeInContext(gallivm->context), "");
 }
 
 static LLVMValueRef
-to_32_float(LLVMBuilderRef builder, LLVMValueRef fp)
+to_32_float(struct gallivm_state *gallivm, LLVMValueRef fp)
 {
-   return LLVMBuildLoad(builder, fp, "");
+   return LLVMBuildLoad(gallivm->builder, fp, "");
 }
 
 static INLINE LLVMValueRef
-to_8_uscaled(LLVMBuilderRef builder, LLVMValueRef fp)
+to_8_uscaled(struct gallivm_state *gallivm, LLVMValueRef fp)
 {
-   LLVMValueRef l = LLVMBuildLoad(builder, fp, "");
-   return LLVMBuildFPToUI(builder, l, LLVMIntType(8), "");
+   LLVMValueRef l = LLVMBuildLoad(gallivm->builder, fp, "");
+   return LLVMBuildFPToUI(gallivm->builder, l, LLVMIntTypeInContext(gallivm->context, 8), "");
 }
 
 static INLINE LLVMValueRef
-to_16_uscaled(LLVMBuilderRef builder, LLVMValueRef fp)
+to_16_uscaled(struct gallivm_state *gallivm, LLVMValueRef fp)
 {
-   LLVMValueRef l = LLVMBuildLoad(builder, fp, "");
-   return LLVMBuildFPToUI(builder, l, LLVMIntType(16), "");
+   LLVMValueRef l = LLVMBuildLoad(gallivm->builder, fp, "");
+   return LLVMBuildFPToUI(gallivm->builder, l, LLVMIntTypeInContext(gallivm->context, 16), "");
 }
 
 static INLINE LLVMValueRef
-to_32_uscaled(LLVMBuilderRef builder, LLVMValueRef fp)
+to_32_uscaled(struct gallivm_state *gallivm, LLVMValueRef fp)
 {
-   LLVMValueRef l = LLVMBuildLoad(builder, fp, "");
-   return LLVMBuildFPToUI(builder, l, LLVMIntType(32), "");
+   LLVMValueRef l = LLVMBuildLoad(gallivm->builder, fp, "");
+   return LLVMBuildFPToUI(gallivm->builder, l, LLVMIntTypeInContext(gallivm->context, 32), "");
 }
 
 static INLINE LLVMValueRef
-to_8_sscaled(LLVMBuilderRef builder, LLVMValueRef fp)
+to_8_sscaled(struct gallivm_state *gallivm, LLVMValueRef fp)
 {
-   LLVMValueRef l = LLVMBuildLoad(builder, fp, "");
-   return LLVMBuildFPToSI(builder, l, LLVMIntType(8), "");
+   LLVMValueRef l = LLVMBuildLoad(gallivm->builder, fp, "");
+   return LLVMBuildFPToSI(gallivm->builder, l, LLVMIntTypeInContext(gallivm->context, 8), "");
 }
 
 static INLINE LLVMValueRef
-to_16_sscaled(LLVMBuilderRef builder, LLVMValueRef fp)
+to_16_sscaled(struct gallivm_state *gallivm, LLVMValueRef fp)
 {
-   LLVMValueRef l = LLVMBuildLoad(builder, fp, "");
-   return LLVMBuildFPToSI(builder, l, LLVMIntType(16), "");
+   LLVMValueRef l = LLVMBuildLoad(gallivm->builder, fp, "");
+   return LLVMBuildFPToSI(gallivm->builder, l, LLVMIntTypeInContext(gallivm->context, 16), "");
 }
 
 static INLINE LLVMValueRef
-to_32_sscaled(LLVMBuilderRef builder, LLVMValueRef fp)
+to_32_sscaled(struct gallivm_state *gallivm, LLVMValueRef fp)
 {
-   LLVMValueRef l = LLVMBuildLoad(builder, fp, "");
-   return LLVMBuildFPToSI(builder, l, LLVMIntType(32), "");
+   LLVMValueRef l = LLVMBuildLoad(gallivm->builder, fp, "");
+   return LLVMBuildFPToSI(gallivm->builder, l, LLVMIntTypeInContext(gallivm->context, 32), "");
 }
 
 static INLINE LLVMValueRef
-to_8_unorm(LLVMBuilderRef builder, LLVMValueRef fp)
+to_8_unorm(struct gallivm_state *gallivm, LLVMValueRef fp)
 {
-   LLVMValueRef l = LLVMBuildLoad(builder, fp, "");
-   LLVMValueRef uscaled = LLVMBuildFPToUI(builder, l, LLVMIntType(8), "");
-   return LLVMBuildFMul(builder, uscaled,
-                        LLVMConstReal(LLVMFloatType(), 255.), "");
+   LLVMValueRef l = LLVMBuildLoad(gallivm->builder, fp, "");
+   LLVMValueRef uscaled = LLVMBuildFPToUI(gallivm->builder, l,
+                                          LLVMIntTypeInContext(gallivm->context, 8), "");
+   return LLVMBuildFMul(gallivm->builder, uscaled,
+                        lp_build_const_float(gallivm, 255.), "");
 }
 
 static INLINE LLVMValueRef
-to_16_unorm(LLVMBuilderRef builder, LLVMValueRef fp)
+to_16_unorm(struct gallivm_state *gallivm, LLVMValueRef fp)
 {
-   LLVMValueRef l = LLVMBuildLoad(builder, fp, "");
-   LLVMValueRef uscaled = LLVMBuildFPToUI(builder, l, LLVMIntType(32), "");
-   return LLVMBuildFMul(builder, uscaled,
-                        LLVMConstReal(LLVMFloatType(), 65535.), "");
+   LLVMValueRef l = LLVMBuildLoad(gallivm->builder, fp, "");
+   LLVMValueRef uscaled = LLVMBuildFPToUI(gallivm->builder, l,
+                                          LLVMIntTypeInContext(gallivm->context, 32), "");
+   return LLVMBuildFMul(gallivm->builder, uscaled,
+                        lp_build_const_float(gallivm, 65535.), "");
 }
 
 static INLINE LLVMValueRef
-to_32_unorm(LLVMBuilderRef builder, LLVMValueRef fp)
+to_32_unorm(struct gallivm_state *gallivm, LLVMValueRef fp)
 {
-   LLVMValueRef l = LLVMBuildLoad(builder, fp, "");
-   LLVMValueRef uscaled = LLVMBuildFPToUI(builder, l, LLVMIntType(32), "");
+   LLVMValueRef l = LLVMBuildLoad(gallivm->builder, fp, "");
+   LLVMValueRef uscaled = LLVMBuildFPToUI(gallivm->builder, l,
+                                          LLVMIntTypeInContext(gallivm->context, 32), "");
 
-   return LLVMBuildFMul(builder, uscaled,
-                        LLVMConstReal(LLVMFloatType(), 4294967295.), "");
+   return LLVMBuildFMul(gallivm->builder, uscaled,
+                        lp_build_const_float(gallivm, 4294967295.), "");
 }
 
 static INLINE LLVMValueRef
-to_8_snorm(LLVMBuilderRef builder, LLVMValueRef val)
+to_8_snorm(struct gallivm_state *gallivm, LLVMValueRef val)
 {
-   LLVMValueRef l = LLVMBuildLoad(builder, val, "");
-   LLVMValueRef uscaled = LLVMBuildFPToSI(builder, l, LLVMIntType(8), "");
-   return LLVMBuildFMul(builder, uscaled,
-                        LLVMConstReal(LLVMFloatType(), 127.0), "");
+   LLVMValueRef l = LLVMBuildLoad(gallivm->builder, val, "");
+   LLVMValueRef uscaled = LLVMBuildFPToSI(gallivm->builder, l,
+                                          LLVMIntTypeInContext(gallivm->context, 8), "");
+   return LLVMBuildFMul(gallivm->builder, uscaled,
+                        lp_build_const_float(gallivm, 127.0), "");
 }
 
 static INLINE LLVMValueRef
-to_16_snorm(LLVMBuilderRef builder, LLVMValueRef fp)
+to_16_snorm(struct gallivm_state *gallivm, LLVMValueRef fp)
 {
-   LLVMValueRef l = LLVMBuildLoad(builder, fp, "");
-   LLVMValueRef uscaled = LLVMBuildFPToSI(builder, l, LLVMIntType(16), "");
-   return LLVMBuildFMul(builder, uscaled,
-                        LLVMConstReal(LLVMFloatType(), 32767.0f), "");
+   LLVMValueRef l = LLVMBuildLoad(gallivm->builder, fp, "");
+   LLVMValueRef uscaled = LLVMBuildFPToSI(gallivm->builder, l,
+                                          LLVMIntTypeInContext(gallivm->context, 16), "");
+   return LLVMBuildFMul(gallivm->builder, uscaled,
+                        lp_build_const_float(gallivm, 32767.0f), "");
 }
 
 static INLINE LLVMValueRef
-to_32_snorm(LLVMBuilderRef builder, LLVMValueRef fp)
+to_32_snorm(struct gallivm_state *gallivm, LLVMValueRef fp)
 {
-   LLVMValueRef l = LLVMBuildLoad(builder, fp, "");
-   LLVMValueRef uscaled = LLVMBuildFPToSI(builder, l, LLVMIntType(32), "");
+   LLVMValueRef l = LLVMBuildLoad(gallivm->builder, fp, "");
+   LLVMValueRef uscaled = LLVMBuildFPToSI(gallivm->builder, l,
+                                          LLVMIntTypeInContext(gallivm->context, 32), "");
 
-   return LLVMBuildFMul(builder, uscaled,
-                        LLVMConstReal(LLVMFloatType(), 2147483647.0), "");
+   return LLVMBuildFMul(gallivm->builder, uscaled,
+                        lp_build_const_float(gallivm, 2147483647.0), "");
 }
 
 static INLINE LLVMValueRef
-to_32_fixed(LLVMBuilderRef builder, LLVMValueRef fp)
+to_32_fixed(struct gallivm_state *gallivm, LLVMValueRef fp)
 {
-   LLVMValueRef l = LLVMBuildLoad(builder, fp, "");
-   LLVMValueRef uscaled = LLVMBuildFPToSI(builder, l, LLVMIntType(32), "");
+   LLVMValueRef l = LLVMBuildLoad(gallivm->builder, fp, "");
+   LLVMValueRef uscaled = LLVMBuildFPToSI(gallivm->builder, l,
+                                          LLVMIntTypeInContext(gallivm->context, 32), "");
 
-   return LLVMBuildFMul(builder, uscaled,
-                        LLVMConstReal(LLVMFloatType(), 65536.0), "");
+   return LLVMBuildFMul(gallivm->builder, uscaled,
+                        lp_build_const_float(gallivm, 65536.0), "");
 }
 
-typedef LLVMValueRef (*from_func)(LLVMBuilderRef, LLVMValueRef);
-typedef  LLVMValueRef (*to_func)(LLVMBuilderRef, LLVMValueRef);
+typedef LLVMValueRef (*from_func)(struct gallivm_state *, LLVMValueRef);
+typedef  LLVMValueRef (*to_func)(struct gallivm_state *, LLVMValueRef);
 
 /* so that underneath can avoid function calls which are prohibited
  * for static initialization we need this conversion */
@@ -294,21 +302,21 @@ enum ll_type {
 };
 
 static INLINE LLVMTypeRef
-ll_type_to_llvm(enum ll_type type)
+ll_type_to_llvm(struct gallivm_state *gallivm, enum ll_type type)
 {
    switch (type) {
    case LL_Double:
-      return LLVMDoubleType();
+      return LLVMDoubleTypeInContext(gallivm->context);
    case LL_Float:
-      return LLVMFloatType();
+      return LLVMFloatTypeInContext(gallivm->context);
    case LL_Int32:
-      return LLVMInt32Type();
+      return LLVMInt32TypeInContext(gallivm->context);
    case LL_Int16:
-      return LLVMIntType(16);
+      return LLVMIntTypeInContext(gallivm->context, 16);
    case LL_Int8:
-      return LLVMIntType(8);
+      return LLVMIntTypeInContext(gallivm->context, 8);
    }
-   return LLVMIntType(8);
+   return LLVMIntTypeInContext(gallivm->context, 8);
 }
 
 static INLINE int
@@ -414,42 +422,42 @@ struct draw_llvm_translate {
 
 
 static LLVMValueRef
-fetch(LLVMBuilderRef builder,
+fetch(struct gallivm_state *gallivm,
       LLVMValueRef ptr, int val_size, int nr_components,
       from_func func)
 {
    int i;
    int offset = 0;
-   LLVMValueRef res = LLVMConstNull(
-      LLVMVectorType(LLVMFloatType(), 4));
+   LLVMValueRef res =
+      LLVMConstNull(LLVMVectorType(LLVMFloatTypeInContext(gallivm->context), 4));
    LLVMValueRef defaults[4];
 
-   defaults[0] = LLVMConstReal(LLVMFloatType(), 0);
-   defaults[1] = LLVMConstReal(LLVMFloatType(), 0);
-   defaults[2] = LLVMConstReal(LLVMFloatType(), 0);
-   defaults[3] = LLVMConstReal(LLVMFloatType(), 1);
+   defaults[0] =
+   defaults[1] =
+   defaults[2] = lp_build_const_float(gallivm, 0.0);
+   defaults[3] = lp_build_const_float(gallivm, 1.0);
 
    for (i = 0; i < nr_components; ++i) {
-      LLVMValueRef src_index = LLVMConstInt(LLVMInt32Type(), offset, 0);
-      LLVMValueRef dst_index = LLVMConstInt(LLVMInt32Type(), i, 0);
+      LLVMValueRef src_index = lp_build_const_int32(gallivm, offset);
+      LLVMValueRef dst_index = lp_build_const_int32(gallivm, i);
       LLVMValueRef src_tmp;
       LLVMValueRef component;
 
-      src_tmp = LLVMBuildGEP(builder, ptr, &src_index, 1, "src_tmp");
+      src_tmp = LLVMBuildGEP(gallivm->builder, ptr, &src_index, 1, "src_tmp");
 
       /* convert src_tmp to float */
-      component = func(builder, src_tmp);
+      component = func(gallivm, src_tmp);
 
       /* vec.comp = component */
-      res = LLVMBuildInsertElement(builder,
+      res = LLVMBuildInsertElement(gallivm->builder,
                                    res,
                                    component,
                                    dst_index, "");
       offset += val_size;
    }
    for (; i < 4; ++i) {
-      LLVMValueRef dst_index = LLVMConstInt(LLVMInt32Type(), i, 0);
-      res = LLVMBuildInsertElement(builder,
+      LLVMValueRef dst_index = lp_build_const_int32(gallivm, i);
+      res = LLVMBuildInsertElement(gallivm->builder,
                                    res,
                                    defaults[i],
                                    dst_index, "");
@@ -459,7 +467,7 @@ fetch(LLVMBuilderRef builder,
 
 
 LLVMValueRef
-draw_llvm_translate_from(LLVMBuilderRef builder,
+draw_llvm_translate_from(struct gallivm_state *gallivm,
                          LLVMValueRef vbuffer,
                          enum pipe_format from_format)
 {
@@ -476,7 +484,7 @@ draw_llvm_translate_from(LLVMBuilderRef builder,
    for (i = 0; i < Elements(translates); ++i) {
       if (translates[i].format == from_format) {
          /*LLVMTypeRef type = ll_type_to_llvm(translates[i].type);*/
-         return fetch(builder,
+         return fetch(gallivm,
                       vbuffer,
                       ll_type_size(translates[i].type),
                       translates[i].num_components,
@@ -493,6 +501,6 @@ draw_llvm_translate_from(LLVMBuilderRef builder,
     */
 
    format_desc = util_format_description(from_format);
-   zero = LLVMConstNull(LLVMInt32Type());
-   return lp_build_fetch_rgba_aos(builder, format_desc, type, vbuffer, zero, zero, zero);
+   zero = LLVMConstNull(LLVMInt32TypeInContext(gallivm->context));
+   return lp_build_fetch_rgba_aos(gallivm, format_desc, type, vbuffer, zero, zero, zero);
 }
diff --git a/src/gallium/auxiliary/draw/draw_pipe_aaline.c b/src/gallium/auxiliary/draw/draw_pipe_aaline.c
index d1aba76309..0851b9acc0 100644
--- a/src/gallium/auxiliary/draw/draw_pipe_aaline.c
+++ b/src/gallium/auxiliary/draw/draw_pipe_aaline.c
@@ -406,6 +406,7 @@ aaline_create_texture(struct aaline_stage *aaline)
    texTemp.width0 = 1 << MAX_TEXTURE_LEVEL;
    texTemp.height0 = 1 << MAX_TEXTURE_LEVEL;
    texTemp.depth0 = 1;
+   texTemp.array_size = 1;
    texTemp.bind = PIPE_BIND_SAMPLER_VIEW;
 
    aaline->texture = screen->resource_create(screen, &texTemp);
@@ -441,10 +442,10 @@ aaline_create_texture(struct aaline_stage *aaline)
       /* This texture is new, no need to flush. 
        */
       transfer = pipe->get_transfer(pipe,
-				    aaline->texture,
-				    u_subresource(0, level), 
-				    PIPE_TRANSFER_WRITE,
-				    &box);
+                                    aaline->texture,
+                                    level,
+                                    PIPE_TRANSFER_WRITE,
+                                    &box);
 
       data = pipe->transfer_map(pipe, transfer);
       if (data == NULL)
diff --git a/src/gallium/auxiliary/draw/draw_pipe_pstipple.c b/src/gallium/auxiliary/draw/draw_pipe_pstipple.c
index ed9a53e154..f5515c1df7 100644
--- a/src/gallium/auxiliary/draw/draw_pipe_pstipple.c
+++ b/src/gallium/auxiliary/draw/draw_pipe_pstipple.c
@@ -393,8 +393,8 @@ pstip_update_texture(struct pstip_stage *pstip)
     */
    pipe->flush( pipe, PIPE_FLUSH_TEXTURE_CACHE, NULL );
 
-   transfer = pipe_get_transfer(pipe, pstip->texture, 0, 0, 0,
-				    PIPE_TRANSFER_WRITE, 0, 0, 32, 32);
+   transfer = pipe_get_transfer(pipe, pstip->texture, 0, 0,
+                                PIPE_TRANSFER_WRITE, 0, 0, 32, 32);
    data = pipe->transfer_map(pipe, transfer);
 
    /*
@@ -440,6 +440,7 @@ pstip_create_texture(struct pstip_stage *pstip)
    texTemp.width0 = 32;
    texTemp.height0 = 32;
    texTemp.depth0 = 1;
+   texTemp.array_size = 1;
    texTemp.bind = PIPE_BIND_SAMPLER_VIEW;
 
    pstip->texture = screen->resource_create(screen, &texTemp);
diff --git a/src/gallium/auxiliary/draw/draw_private.h b/src/gallium/auxiliary/draw/draw_private.h
index 54163d7f9e..06ed4d60ef 100644
--- a/src/gallium/auxiliary/draw/draw_private.h
+++ b/src/gallium/auxiliary/draw/draw_private.h
@@ -286,7 +286,6 @@ struct draw_context
 
 #ifdef HAVE_LLVM
    struct draw_llvm *llvm;
-   LLVMExecutionEngineRef engine;
 #endif
 
    struct pipe_sampler_view *sampler_views[PIPE_MAX_VERTEX_SAMPLERS];
diff --git a/src/gallium/auxiliary/draw/draw_pt_fetch_shade_emit.c b/src/gallium/auxiliary/draw/draw_pt_fetch_shade_emit.c
index 7c198c6026..c98fb3d520 100644
--- a/src/gallium/auxiliary/draw/draw_pt_fetch_shade_emit.c
+++ b/src/gallium/auxiliary/draw/draw_pt_fetch_shade_emit.c
@@ -58,8 +58,8 @@ struct fetch_shade_emit {
    const ubyte *src[PIPE_MAX_ATTRIBS];
    unsigned prim;
 
-   struct draw_vs_varient_key key;
-   struct draw_vs_varient *active;
+   struct draw_vs_variant_key key;
+   struct draw_vs_variant *active;
 
 
    const struct vertex_info *vinfo;
@@ -150,7 +150,7 @@ static void fse_prepare( struct draw_pt_middle_end *middle,
    }
 
    
-   fse->active = draw_vs_lookup_varient( draw->vs.vertex_shader, 
+   fse->active = draw_vs_lookup_variant( draw->vs.vertex_shader, 
                                          &fse->key );
 
    if (!fse->active) {
diff --git a/src/gallium/auxiliary/draw/draw_pt_fetch_shade_pipeline_llvm.c b/src/gallium/auxiliary/draw/draw_pt_fetch_shade_pipeline_llvm.c
index a53a768d02..2e3afb22c4 100644
--- a/src/gallium/auxiliary/draw/draw_pt_fetch_shade_pipeline_llvm.c
+++ b/src/gallium/auxiliary/draw/draw_pt_fetch_shade_pipeline_llvm.c
@@ -34,6 +34,7 @@
 #include "draw/draw_pt.h"
 #include "draw/draw_vs.h"
 #include "draw/draw_llvm.h"
+#include "gallivm/lp_bld_init.h"
 
 
 struct llvm_middle_end {
@@ -72,19 +73,18 @@ llvm_middle_end_prepare( struct draw_pt_middle_end *middle,
    struct draw_llvm_variant_list_item *li;
    unsigned i;
    unsigned instance_id_index = ~0;
-
-
-   unsigned out_prim = (draw->gs.geometry_shader ? 
-                        draw->gs.geometry_shader->output_primitive :
-                        in_prim);
+   const unsigned out_prim = (draw->gs.geometry_shader ? 
+                              draw->gs.geometry_shader->output_primitive :
+                              in_prim);
 
    /* Add one to num_outputs because the pipeline occasionally tags on
     * an additional texcoord, eg for AA lines.
     */
-   unsigned nr = MAX2( shader->base.info.num_inputs,
-		       shader->base.info.num_outputs + 1 );
+   const unsigned nr = MAX2( shader->base.info.num_inputs,
+                             shader->base.info.num_outputs + 1 );
 
    /* Scan for instanceID system value.
+    * XXX but we never use instance_id_index?!
     */
    for (i = 0; i < shader->base.info.num_inputs; i++) {
       if (shader->base.info.input_semantic_name[i] == TGSI_SEMANTIC_INSTANCEID) {
@@ -133,9 +133,10 @@ llvm_middle_end_prepare( struct draw_pt_middle_end *middle,
    
    key = draw_llvm_make_variant_key(fpme->llvm, store);
 
+   /* Search shader's list of variants for the key */
    li = first_elem(&shader->variants);
-   while(!at_end(&shader->variants, li)) {
-      if(memcmp(&li->base->key, key, shader->variant_key_size) == 0) {
+   while (!at_end(&shader->variants, li)) {
+      if (memcmp(&li->base->key, key, shader->variant_key_size) == 0) {
          variant = li->base;
          break;
       }
@@ -143,10 +144,16 @@ llvm_middle_end_prepare( struct draw_pt_middle_end *middle,
    }
 
    if (variant) {
+      /* found the variant, move to head of global list (for LRU) */
       move_to_head(&fpme->llvm->vs_variants_list, &variant->list_item_global);
    }
    else {
+      /* Need to create new variant */
       unsigned i;
+
+      /* First check if we've created too many variants.  If so, free
+       * 25% of the LRU to avoid using too much memory.
+       */
       if (fpme->llvm->nr_variants >= DRAW_MAX_SHADER_VARIANTS) {
          /*
           * XXX: should we flush here ?
@@ -422,7 +429,7 @@ draw_pt_fetch_pipeline_or_emit_llvm(struct draw_context *draw)
 {
    struct llvm_middle_end *fpme = 0;
 
-   if (!draw->engine)
+   if (!draw->llvm->gallivm->engine)
       return NULL;
 
    fpme = CALLOC_STRUCT( llvm_middle_end );
diff --git a/src/gallium/auxiliary/draw/draw_vs.c b/src/gallium/auxiliary/draw/draw_vs.c
index fb665b08ff..7caad6f005 100644
--- a/src/gallium/auxiliary/draw/draw_vs.c
+++ b/src/gallium/auxiliary/draw/draw_vs.c
@@ -165,10 +165,10 @@ draw_delete_vertex_shader(struct draw_context *draw,
 {
    unsigned i;
 
-   for (i = 0; i < dvs->nr_varients; i++) 
-      dvs->varient[i]->destroy( dvs->varient[i] );
+   for (i = 0; i < dvs->nr_variants; i++) 
+      dvs->variant[i]->destroy( dvs->variant[i] );
 
-   dvs->nr_varients = 0;
+   dvs->nr_variants = 0;
 
    dvs->delete( dvs );
 }
@@ -225,40 +225,40 @@ draw_vs_destroy( struct draw_context *draw )
 }
 
 
-struct draw_vs_varient *
-draw_vs_lookup_varient( struct draw_vertex_shader *vs,
-                        const struct draw_vs_varient_key *key )
+struct draw_vs_variant *
+draw_vs_lookup_variant( struct draw_vertex_shader *vs,
+                        const struct draw_vs_variant_key *key )
 {
-   struct draw_vs_varient *varient;
+   struct draw_vs_variant *variant;
    unsigned i;
 
-   /* Lookup existing varient: 
+   /* Lookup existing variant: 
     */
-   for (i = 0; i < vs->nr_varients; i++)
-      if (draw_vs_varient_key_compare(key, &vs->varient[i]->key) == 0)
-         return vs->varient[i];
+   for (i = 0; i < vs->nr_variants; i++)
+      if (draw_vs_variant_key_compare(key, &vs->variant[i]->key) == 0)
+         return vs->variant[i];
    
    /* Else have to create a new one: 
     */
-   varient = vs->create_varient( vs, key );
-   if (varient == NULL)
+   variant = vs->create_variant( vs, key );
+   if (variant == NULL)
       return NULL;
 
    /* Add it to our list, could be smarter: 
     */
-   if (vs->nr_varients < Elements(vs->varient)) {
-      vs->varient[vs->nr_varients++] = varient;
+   if (vs->nr_variants < Elements(vs->variant)) {
+      vs->variant[vs->nr_variants++] = variant;
    }
    else {
-      vs->last_varient++;
-      vs->last_varient %= Elements(vs->varient);
-      vs->varient[vs->last_varient]->destroy(vs->varient[vs->last_varient]);
-      vs->varient[vs->last_varient] = varient;
+      vs->last_variant++;
+      vs->last_variant %= Elements(vs->variant);
+      vs->variant[vs->last_variant]->destroy(vs->variant[vs->last_variant]);
+      vs->variant[vs->last_variant] = variant;
    }
 
    /* Done 
     */
-   return varient;
+   return variant;
 }
 
 
diff --git a/src/gallium/auxiliary/draw/draw_vs.h b/src/gallium/auxiliary/draw/draw_vs.h
index f9a038788f..bfb72d50ef 100644
--- a/src/gallium/auxiliary/draw/draw_vs.h
+++ b/src/gallium/auxiliary/draw/draw_vs.h
@@ -38,7 +38,7 @@
 struct draw_context;
 struct pipe_shader_state;
 
-struct draw_varient_input 
+struct draw_variant_input 
 {
    enum pipe_format format;
    unsigned buffer;
@@ -46,19 +46,19 @@ struct draw_varient_input
    unsigned instance_divisor;
 };
 
-struct draw_varient_output
+struct draw_variant_output
 {
    enum pipe_format format;     /* output format */
    unsigned vs_output:8;        /* which vertex shader output is this? */
    unsigned offset:24;          /* offset into output vertex */
 };
 
-struct draw_varient_element {
-   struct draw_varient_input in;
-   struct draw_varient_output out;
+struct draw_variant_element {
+   struct draw_variant_input in;
+   struct draw_variant_output out;
 };
 
-struct draw_vs_varient_key {
+struct draw_vs_variant_key {
    unsigned output_stride;
    unsigned nr_elements:8;      /* max2(nr_inputs, nr_outputs) */
    unsigned nr_inputs:8;
@@ -66,34 +66,34 @@ struct draw_vs_varient_key {
    unsigned viewport:1;
    unsigned clip:1;
    unsigned const_vbuffers:5;
-   struct draw_varient_element element[PIPE_MAX_ATTRIBS];
+   struct draw_variant_element element[PIPE_MAX_ATTRIBS];
 };
 
-struct draw_vs_varient;
+struct draw_vs_variant;
 
 
-struct draw_vs_varient {
-   struct draw_vs_varient_key key;
+struct draw_vs_variant {
+   struct draw_vs_variant_key key;
 
    struct draw_vertex_shader *vs;
 
-   void (*set_buffer)( struct draw_vs_varient *,
+   void (*set_buffer)( struct draw_vs_variant *,
                       unsigned i,
                       const void *ptr,
                       unsigned stride,
                       unsigned max_stride );
 
-   void (PIPE_CDECL *run_linear)( struct draw_vs_varient *shader,
+   void (PIPE_CDECL *run_linear)( struct draw_vs_variant *shader,
                                   unsigned start,
                                   unsigned count,
                                   void *output_buffer );
 
-   void (PIPE_CDECL *run_elts)( struct draw_vs_varient *shader,
+   void (PIPE_CDECL *run_elts)( struct draw_vs_variant *shader,
                                 const unsigned *elts,
                                 unsigned count,
                                 void *output_buffer );
 
-   void (*destroy)( struct draw_vs_varient * );
+   void (*destroy)( struct draw_vs_variant * );
 };
 
 
@@ -117,11 +117,11 @@ struct draw_vertex_shader {
 
    /* 
     */
-   struct draw_vs_varient *varient[16];
-   unsigned nr_varients;
-   unsigned last_varient;
-   struct draw_vs_varient *(*create_varient)( struct draw_vertex_shader *shader,
-                                              const struct draw_vs_varient_key *key );
+   struct draw_vs_variant *variant[16];
+   unsigned nr_variants;
+   unsigned last_variant;
+   struct draw_vs_variant *(*create_variant)( struct draw_vertex_shader *shader,
+                                              const struct draw_vs_variant_key *key );
 
 
    void (*prepare)( struct draw_vertex_shader *shader,
@@ -144,9 +144,9 @@ struct draw_vertex_shader {
 };
 
 
-struct draw_vs_varient *
-draw_vs_lookup_varient( struct draw_vertex_shader *base,
-                        const struct draw_vs_varient_key *key );
+struct draw_vs_variant *
+draw_vs_lookup_variant( struct draw_vertex_shader *base,
+                        const struct draw_vs_variant_key *key );
 
 
 /********************************************************************************
@@ -166,12 +166,12 @@ draw_create_vs_ppc(struct draw_context *draw,
 		   const struct pipe_shader_state *templ);
 
 
-struct draw_vs_varient_key;
+struct draw_vs_variant_key;
 struct draw_vertex_shader;
 
-struct draw_vs_varient *
-draw_vs_create_varient_aos_sse( struct draw_vertex_shader *vs,
-                                const struct draw_vs_varient_key *key );
+struct draw_vs_variant *
+draw_vs_create_variant_aos_sse( struct draw_vertex_shader *vs,
+                                const struct draw_vs_variant_key *key );
 
 #if HAVE_LLVM
 struct draw_vertex_shader *
@@ -181,7 +181,7 @@ draw_create_vs_llvm(struct draw_context *draw,
 
 
 /********************************************************************************
- * Helpers for vs implementations that don't do their own fetch/emit varients.
+ * Helpers for vs implementations that don't do their own fetch/emit variants.
  * Means these can be shared between shaders.
  */
 struct translate;
@@ -194,21 +194,21 @@ struct translate *draw_vs_get_fetch( struct draw_context *draw,
 struct translate *draw_vs_get_emit( struct draw_context *draw,
                                     struct translate_key *key );
 
-struct draw_vs_varient *
-draw_vs_create_varient_generic( struct draw_vertex_shader *vs,
-                                const struct draw_vs_varient_key *key );
+struct draw_vs_variant *
+draw_vs_create_variant_generic( struct draw_vertex_shader *vs,
+                                const struct draw_vs_variant_key *key );
 
 
 
-static INLINE int draw_vs_varient_keysize( const struct draw_vs_varient_key *key )
+static INLINE int draw_vs_variant_keysize( const struct draw_vs_variant_key *key )
 {
-   return 2 * sizeof(int) + key->nr_elements * sizeof(struct draw_varient_element);
+   return 2 * sizeof(int) + key->nr_elements * sizeof(struct draw_variant_element);
 }
 
-static INLINE int draw_vs_varient_key_compare( const struct draw_vs_varient_key *a,
-                                         const struct draw_vs_varient_key *b )
+static INLINE int draw_vs_variant_key_compare( const struct draw_vs_variant_key *a,
+                                         const struct draw_vs_variant_key *b )
 {
-   int keysize = draw_vs_varient_keysize(a);
+   int keysize = draw_vs_variant_keysize(a);
    return memcmp(a, b, keysize);
 }
 
diff --git a/src/gallium/auxiliary/draw/draw_vs_aos.c b/src/gallium/auxiliary/draw/draw_vs_aos.c
index 19f49e34c8..7b90dba0cd 100644
--- a/src/gallium/auxiliary/draw/draw_vs_aos.c
+++ b/src/gallium/auxiliary/draw/draw_vs_aos.c
@@ -1918,7 +1918,7 @@ static void find_last_write_outputs( struct aos_compilation *cp )
 #define ARG_OUTBUF     4
 
 
-static boolean build_vertex_program( struct draw_vs_varient_aos_sse *varient,
+static boolean build_vertex_program( struct draw_vs_variant_aos_sse *variant,
                                      boolean linear )
 { 
    struct tgsi_parse_context parse;
@@ -1927,14 +1927,14 @@ static boolean build_vertex_program( struct draw_vs_varient_aos_sse *varient,
 
    util_init_math();
 
-   tgsi_parse_init( &parse, varient->base.vs->state.tokens );
+   tgsi_parse_init( &parse, variant->base.vs->state.tokens );
 
    memset(&cp, 0, sizeof(cp));
 
    cp.insn_counter = 1;
-   cp.vaos = varient;
+   cp.vaos = variant;
    cp.have_sse2 = 1;
-   cp.func = &varient->func[ linear ? 0 : 1 ];
+   cp.func = &variant->func[ linear ? 0 : 1 ];
 
    cp.tmp_EAX       = x86_make_reg(file_REG32, reg_AX);
    cp.idx_EBX      = x86_make_reg(file_REG32, reg_BX);
@@ -2090,20 +2090,20 @@ static boolean build_vertex_program( struct draw_vs_varient_aos_sse *varient,
 
 
 /** cast wrapper */
-static INLINE struct draw_vs_varient_aos_sse *
-draw_vs_varient_aos_sse(struct draw_vs_varient *varient)
+static INLINE struct draw_vs_variant_aos_sse *
+draw_vs_variant_aos_sse(struct draw_vs_variant *variant)
 {
-   return (struct draw_vs_varient_aos_sse *) varient;
+   return (struct draw_vs_variant_aos_sse *) variant;
 }
 
 
-static void vaos_set_buffer( struct draw_vs_varient *varient,
+static void vaos_set_buffer( struct draw_vs_variant *variant,
                              unsigned buf,
                              const void *ptr,
                              unsigned stride,
                              unsigned max_stride)
 {
-   struct draw_vs_varient_aos_sse *vaos = draw_vs_varient_aos_sse(varient);
+   struct draw_vs_variant_aos_sse *vaos = draw_vs_variant_aos_sse(variant);
 
    if (buf < vaos->nr_vb) {
       vaos->buffer[buf].base_ptr = (char *)ptr;
@@ -2115,12 +2115,12 @@ static void vaos_set_buffer( struct draw_vs_varient *varient,
 
 
 
-static void PIPE_CDECL vaos_run_elts( struct draw_vs_varient *varient,
+static void PIPE_CDECL vaos_run_elts( struct draw_vs_variant *variant,
                                       const unsigned *elts,
                                       unsigned count,
                                       void *output_buffer )
 {
-   struct draw_vs_varient_aos_sse *vaos = draw_vs_varient_aos_sse(varient);
+   struct draw_vs_variant_aos_sse *vaos = draw_vs_variant_aos_sse(variant);
    struct aos_machine *machine = vaos->draw->vs.aos_machine;
    unsigned i;
 
@@ -2139,12 +2139,12 @@ static void PIPE_CDECL vaos_run_elts( struct draw_vs_varient *varient,
                        output_buffer );
 }
 
-static void PIPE_CDECL vaos_run_linear( struct draw_vs_varient *varient,
+static void PIPE_CDECL vaos_run_linear( struct draw_vs_variant *variant,
                                         unsigned start,
                                         unsigned count,
                                         void *output_buffer )
 {
-   struct draw_vs_varient_aos_sse *vaos = draw_vs_varient_aos_sse(varient);
+   struct draw_vs_variant_aos_sse *vaos = draw_vs_variant_aos_sse(variant);
    struct aos_machine *machine = vaos->draw->vs.aos_machine;
    unsigned i;
 
@@ -2171,9 +2171,9 @@ static void PIPE_CDECL vaos_run_linear( struct draw_vs_varient *varient,
 
 
 
-static void vaos_destroy( struct draw_vs_varient *varient )
+static void vaos_destroy( struct draw_vs_variant *variant )
 {
-   struct draw_vs_varient_aos_sse *vaos = draw_vs_varient_aos_sse(varient);
+   struct draw_vs_variant_aos_sse *vaos = draw_vs_variant_aos_sse(variant);
 
    FREE( vaos->buffer );
 
@@ -2185,11 +2185,11 @@ static void vaos_destroy( struct draw_vs_varient *varient )
 
 
 
-static struct draw_vs_varient *varient_aos_sse( struct draw_vertex_shader *vs,
-                                                 const struct draw_vs_varient_key *key )
+static struct draw_vs_variant *variant_aos_sse( struct draw_vertex_shader *vs,
+                                                 const struct draw_vs_variant_key *key )
 {
    unsigned i;
-   struct draw_vs_varient_aos_sse *vaos = CALLOC_STRUCT(draw_vs_varient_aos_sse);
+   struct draw_vs_variant_aos_sse *vaos = CALLOC_STRUCT(draw_vs_variant_aos_sse);
 
    if (!vaos)
       goto fail;
@@ -2249,17 +2249,17 @@ static struct draw_vs_varient *varient_aos_sse( struct draw_vertex_shader *vs,
 }
 
 
-struct draw_vs_varient *
-draw_vs_create_varient_aos_sse( struct draw_vertex_shader *vs,
-                                const struct draw_vs_varient_key *key )
+struct draw_vs_variant *
+draw_vs_create_variant_aos_sse( struct draw_vertex_shader *vs,
+                                const struct draw_vs_variant_key *key )
 {
-   struct draw_vs_varient *varient = varient_aos_sse( vs, key );
+   struct draw_vs_variant *variant = variant_aos_sse( vs, key );
 
-   if (varient == NULL) {
-      varient = draw_vs_create_varient_generic( vs, key );
+   if (variant == NULL) {
+      variant = draw_vs_create_variant_generic( vs, key );
    }
 
-   return varient;
+   return variant;
 }
 
 
diff --git a/src/gallium/auxiliary/draw/draw_vs_aos.h b/src/gallium/auxiliary/draw/draw_vs_aos.h
index 68e8295b5e..55e63d8b9f 100644
--- a/src/gallium/auxiliary/draw/draw_vs_aos.h
+++ b/src/gallium/auxiliary/draw/draw_vs_aos.h
@@ -98,9 +98,9 @@ struct aos_buffer {
 
 
 
-/* This is the temporary storage used by all the aos_sse vs varients.
+/* This is the temporary storage used by all the aos_sse vs variants.
  * Create one per context and reuse by passing a pointer in at
- * vs_varient creation??
+ * vs_variant creation??
  */
 struct aos_machine {
    float input    [MAX_INPUTS    ][4];
@@ -134,7 +134,7 @@ struct aos_machine {
 
 struct aos_compilation {
    struct x86_function *func;
-   struct draw_vs_varient_aos_sse *vaos;
+   struct draw_vs_variant_aos_sse *vaos;
 
    unsigned insn_counter;
    unsigned num_immediates;
@@ -234,8 +234,8 @@ typedef void (PIPE_CDECL *vaos_run_linear_func)( struct aos_machine *,
                                                 void *output_buffer);
 
 
-struct draw_vs_varient_aos_sse {
-   struct draw_vs_varient base;
+struct draw_vs_variant_aos_sse {
+   struct draw_vs_variant base;
    struct draw_context *draw;
 
    struct aos_buffer *buffer;
diff --git a/src/gallium/auxiliary/draw/draw_vs_exec.c b/src/gallium/auxiliary/draw/draw_vs_exec.c
index dab3eb1ca8..667eb50785 100644
--- a/src/gallium/auxiliary/draw/draw_vs_exec.c
+++ b/src/gallium/auxiliary/draw/draw_vs_exec.c
@@ -203,7 +203,7 @@ draw_create_vs_exec(struct draw_context *draw,
    vs->base.prepare = vs_exec_prepare;
    vs->base.run_linear = vs_exec_run_linear;
    vs->base.delete = vs_exec_delete;
-   vs->base.create_varient = draw_vs_create_varient_generic;
+   vs->base.create_variant = draw_vs_create_variant_generic;
    vs->machine = draw->vs.machine;
 
    return &vs->base;
diff --git a/src/gallium/auxiliary/draw/draw_vs_llvm.c b/src/gallium/auxiliary/draw/draw_vs_llvm.c
index fa9992db78..54a5574f73 100644
--- a/src/gallium/auxiliary/draw/draw_vs_llvm.c
+++ b/src/gallium/auxiliary/draw/draw_vs_llvm.c
@@ -65,19 +65,7 @@ static void
 vs_llvm_delete( struct draw_vertex_shader *dvs )
 {
    struct llvm_vertex_shader *shader = llvm_vertex_shader(dvs);
-   struct pipe_fence_handle *fence = NULL;
    struct draw_llvm_variant_list_item *li;
-   struct pipe_context *pipe = dvs->draw->pipe;
-
-   /*
-    * XXX: This might be not neccessary at all.
-    */
-   pipe->flush(pipe, 0, &fence);
-   if (fence) {
-      pipe->screen->fence_finish(pipe->screen, fence, 0);
-      pipe->screen->fence_reference(pipe->screen, &fence, NULL);
-   }
-
 
    li = first_elem(&shader->variants);
    while(!at_end(&shader->variants, li)) {
@@ -119,7 +107,7 @@ draw_create_vs_llvm(struct draw_context *draw,
    vs->base.prepare = vs_llvm_prepare;
    vs->base.run_linear = vs_llvm_run_linear;
    vs->base.delete = vs_llvm_delete;
-   vs->base.create_varient = draw_vs_create_varient_generic;
+   vs->base.create_variant = draw_vs_create_variant_generic;
 
    make_empty_list(&vs->variants);
 
diff --git a/src/gallium/auxiliary/draw/draw_vs_ppc.c b/src/gallium/auxiliary/draw/draw_vs_ppc.c
index 5df84916c5..cf894bbe8a 100644
--- a/src/gallium/auxiliary/draw/draw_vs_ppc.c
+++ b/src/gallium/auxiliary/draw/draw_vs_ppc.c
@@ -187,10 +187,10 @@ draw_create_vs_ppc(struct draw_context *draw,
    vs->base.draw = draw;
 #if 0
    if (1)
-      vs->base.create_varient = draw_vs_varient_aos_ppc;
+      vs->base.create_variant = draw_vs_variant_aos_ppc;
    else
 #endif
-      vs->base.create_varient = draw_vs_create_varient_generic;
+      vs->base.create_variant = draw_vs_create_variant_generic;
    vs->base.prepare = vs_ppc_prepare;
    vs->base.run_linear = vs_ppc_run_linear;
    vs->base.delete = vs_ppc_delete;
diff --git a/src/gallium/auxiliary/draw/draw_vs_sse.c b/src/gallium/auxiliary/draw/draw_vs_sse.c
index 0b0c6077c6..dee7c0da9b 100644
--- a/src/gallium/auxiliary/draw/draw_vs_sse.c
+++ b/src/gallium/auxiliary/draw/draw_vs_sse.c
@@ -166,9 +166,9 @@ draw_create_vs_sse(struct draw_context *draw,
 
    vs->base.draw = draw;
    if (1)
-      vs->base.create_varient = draw_vs_create_varient_aos_sse;
+      vs->base.create_variant = draw_vs_create_variant_aos_sse;
    else
-      vs->base.create_varient = draw_vs_create_varient_generic;
+      vs->base.create_variant = draw_vs_create_variant_generic;
    vs->base.prepare = vs_sse_prepare;
    vs->base.run_linear = vs_sse_run_linear;
    vs->base.delete = vs_sse_delete;
diff --git a/src/gallium/auxiliary/draw/draw_vs_varient.c b/src/gallium/auxiliary/draw/draw_vs_varient.c
index eacd160187..d8f030f61e 100644
--- a/src/gallium/auxiliary/draw/draw_vs_varient.c
+++ b/src/gallium/auxiliary/draw/draw_vs_varient.c
@@ -41,8 +41,8 @@
 
 /* A first pass at incorporating vertex fetch/emit functionality into 
  */
-struct draw_vs_varient_generic {
-   struct draw_vs_varient base;
+struct draw_vs_variant_generic {
+   struct draw_vs_variant base;
 
    struct draw_vertex_shader *shader;
    struct draw_context *draw;
@@ -63,13 +63,13 @@ struct draw_vs_varient_generic {
 
 
 
-static void vsvg_set_buffer( struct draw_vs_varient *varient,
+static void vsvg_set_buffer( struct draw_vs_variant *variant,
                              unsigned buffer,
                              const void *ptr,
                              unsigned stride,
                              unsigned max_index )
 {
-   struct draw_vs_varient_generic *vsvg = (struct draw_vs_varient_generic *)varient;
+   struct draw_vs_variant_generic *vsvg = (struct draw_vs_variant_generic *)variant;
 
    vsvg->fetch->set_buffer(vsvg->fetch, 
                            buffer, 
@@ -81,7 +81,7 @@ static void vsvg_set_buffer( struct draw_vs_varient *varient,
 
 /* Mainly for debug at this stage:
  */
-static void do_rhw_viewport( struct draw_vs_varient_generic *vsvg,
+static void do_rhw_viewport( struct draw_vs_variant_generic *vsvg,
                              unsigned count,
                              void *output_buffer )
 {
@@ -104,7 +104,7 @@ static void do_rhw_viewport( struct draw_vs_varient_generic *vsvg,
    }
 }
 
-static void do_viewport( struct draw_vs_varient_generic *vsvg,
+static void do_viewport( struct draw_vs_variant_generic *vsvg,
                          unsigned count,
                          void *output_buffer )
 {
@@ -126,12 +126,12 @@ static void do_viewport( struct draw_vs_varient_generic *vsvg,
 }
                          
 
-static void PIPE_CDECL vsvg_run_elts( struct draw_vs_varient *varient,
+static void PIPE_CDECL vsvg_run_elts( struct draw_vs_variant *variant,
                                       const unsigned *elts,
                                       unsigned count,
                                       void *output_buffer)
 {
-   struct draw_vs_varient_generic *vsvg = (struct draw_vs_varient_generic *)varient;
+   struct draw_vs_variant_generic *vsvg = (struct draw_vs_variant_generic *)variant;
    unsigned temp_vertex_stride = vsvg->temp_vertex_stride;
    void *temp_buffer = MALLOC( align(count,4) * temp_vertex_stride );
    
@@ -193,12 +193,12 @@ static void PIPE_CDECL vsvg_run_elts( struct draw_vs_varient *varient,
 }
 
 
-static void PIPE_CDECL vsvg_run_linear( struct draw_vs_varient *varient,
+static void PIPE_CDECL vsvg_run_linear( struct draw_vs_variant *variant,
                                         unsigned start,
                                         unsigned count,
                                         void *output_buffer )
 {
-   struct draw_vs_varient_generic *vsvg = (struct draw_vs_varient_generic *)varient;
+   struct draw_vs_variant_generic *vsvg = (struct draw_vs_variant_generic *)variant;
    unsigned temp_vertex_stride = vsvg->temp_vertex_stride;
    void *temp_buffer = MALLOC( align(count,4) * temp_vertex_stride );
 	
@@ -259,20 +259,20 @@ static void PIPE_CDECL vsvg_run_linear( struct draw_vs_varient *varient,
 
 
 
-static void vsvg_destroy( struct draw_vs_varient *varient )
+static void vsvg_destroy( struct draw_vs_variant *variant )
 {
-   FREE(varient);
+   FREE(variant);
 }
 
 
-struct draw_vs_varient *
-draw_vs_create_varient_generic( struct draw_vertex_shader *vs,
-                                const struct draw_vs_varient_key *key )
+struct draw_vs_variant *
+draw_vs_create_variant_generic( struct draw_vertex_shader *vs,
+                                const struct draw_vs_variant_key *key )
 {
    unsigned i;
    struct translate_key fetch, emit;
 
-   struct draw_vs_varient_generic *vsvg = CALLOC_STRUCT( draw_vs_varient_generic );
+   struct draw_vs_variant_generic *vsvg = CALLOC_STRUCT( draw_vs_variant_generic );
    if (vsvg == NULL)
       return NULL;
 
diff --git a/src/gallium/auxiliary/gallivm/lp_bld.h b/src/gallium/auxiliary/gallivm/lp_bld.h
index 8103bc917f..ee05c6ba01 100644
--- a/src/gallium/auxiliary/gallivm/lp_bld.h
+++ b/src/gallium/auxiliary/gallivm/lp_bld.h
@@ -55,4 +55,33 @@
 #endif
 
 
+/**
+ * Redefine these LLVM entrypoints as invalid macros to make sure we
+ * don't accidentally use them.  We need to use the functions which
+ * take an explicit LLVMContextRef parameter.
+ */
+#define LLVMInt1Type ILLEGAL_LLVM_FUNCTION
+#define LLVMInt8Type ILLEGAL_LLVM_FUNCTION
+#define LLVMInt16Type ILLEGAL_LLVM_FUNCTION
+#define LLVMInt32Type ILLEGAL_LLVM_FUNCTION
+#define LLVMInt64Type ILLEGAL_LLVM_FUNCTION
+#define LLVMIntType ILLEGAL_LLVM_FUNCTION
+#define LLVMFloatType ILLEGAL_LLVM_FUNCTION
+#define LLVMDoubleType ILLEGAL_LLVM_FUNCTION
+#define LLVMX86FP80Type ILLEGAL_LLVM_FUNCTION
+#define LLVMFP128Type ILLEGAL_LLVM_FUNCTION
+#define LLVMPPCFP128Type ILLEGAL_LLVM_FUNCTION
+#define LLVMStructType ILLEGAL_LLVM_FUNCTION
+#define LLVMVoidType ILLEGAL_LLVM_FUNCTION
+#define LLVMLabelType ILLEGAL_LLVM_FUNCTION
+#define LLVMOpaqueType ILLEGAL_LLVM_FUNCTION
+#define LLVMUnionType ILLEGAL_LLVM_FUNCTION
+#define LLVMMDString ILLEGAL_LLVM_FUNCTION
+#define LLVMMDNode ILLEGAL_LLVM_FUNCTION
+#define LLVMConstString ILLEGAL_LLVM_FUNCTION
+#define LLVMConstStruct ILLEGAL_LLVM_FUNCTION
+#define LLVMAppendBasicBlock ILLEGAL_LLVM_FUNCTION
+#define LLVMInsertBasicBlock ILLEGAL_LLVM_FUNCTION
+#define LLVMCreateBuilder ILLEGAL_LLVM_FUNCTION
+
 #endif /* LP_BLD_H */
diff --git a/src/gallium/auxiliary/gallivm/lp_bld_arit.c b/src/gallium/auxiliary/gallivm/lp_bld_arit.c
index f9a12a41a1..02b3bde789 100644
--- a/src/gallium/auxiliary/gallivm/lp_bld_arit.c
+++ b/src/gallium/auxiliary/gallivm/lp_bld_arit.c
@@ -53,6 +53,7 @@
 
 #include "lp_bld_type.h"
 #include "lp_bld_const.h"
+#include "lp_bld_init.h"
 #include "lp_bld_intr.h"
 #include "lp_bld_logic.h"
 #include "lp_bld_pack.h"
@@ -74,6 +75,7 @@ lp_build_min_simple(struct lp_build_context *bld,
                     LLVMValueRef a,
                     LLVMValueRef b)
 {
+   LLVMBuilderRef builder = bld->gallivm->builder;
    const struct lp_type type = bld->type;
    const char *intrinsic = NULL;
    LLVMValueRef cond;
@@ -107,7 +109,7 @@ lp_build_min_simple(struct lp_build_context *bld,
    }
 
    if(intrinsic)
-      return lp_build_intrinsic_binary(bld->builder, intrinsic, lp_build_vec_type(bld->type), a, b);
+      return lp_build_intrinsic_binary(builder, intrinsic, lp_build_vec_type(bld->gallivm, bld->type), a, b);
 
    cond = lp_build_cmp(bld, PIPE_FUNC_LESS, a, b);
    return lp_build_select(bld, cond, a, b);
@@ -123,6 +125,7 @@ lp_build_max_simple(struct lp_build_context *bld,
                     LLVMValueRef a,
                     LLVMValueRef b)
 {
+   LLVMBuilderRef builder = bld->gallivm->builder;
    const struct lp_type type = bld->type;
    const char *intrinsic = NULL;
    LLVMValueRef cond;
@@ -156,7 +159,7 @@ lp_build_max_simple(struct lp_build_context *bld,
    }
 
    if(intrinsic)
-      return lp_build_intrinsic_binary(bld->builder, intrinsic, lp_build_vec_type(bld->type), a, b);
+      return lp_build_intrinsic_binary(builder, intrinsic, lp_build_vec_type(bld->gallivm, bld->type), a, b);
 
    cond = lp_build_cmp(bld, PIPE_FUNC_GREATER, a, b);
    return lp_build_select(bld, cond, a, b);
@@ -170,6 +173,7 @@ LLVMValueRef
 lp_build_comp(struct lp_build_context *bld,
               LLVMValueRef a)
 {
+   LLVMBuilderRef builder = bld->gallivm->builder;
    const struct lp_type type = bld->type;
 
    assert(lp_check_value(type, a));
@@ -183,7 +187,7 @@ lp_build_comp(struct lp_build_context *bld,
       if(LLVMIsConstant(a))
          return LLVMConstNot(a);
       else
-         return LLVMBuildNot(bld->builder, a, "");
+         return LLVMBuildNot(builder, a, "");
    }
 
    if(LLVMIsConstant(a))
@@ -193,9 +197,9 @@ lp_build_comp(struct lp_build_context *bld,
           return LLVMConstSub(bld->one, a);
    else
       if (type.floating)
-         return LLVMBuildFSub(bld->builder, bld->one, a, "");
+         return LLVMBuildFSub(builder, bld->one, a, "");
       else
-         return LLVMBuildSub(bld->builder, bld->one, a, "");
+         return LLVMBuildSub(builder, bld->one, a, "");
 }
 
 
@@ -207,6 +211,7 @@ lp_build_add(struct lp_build_context *bld,
              LLVMValueRef a,
              LLVMValueRef b)
 {
+   LLVMBuilderRef builder = bld->gallivm->builder;
    const struct lp_type type = bld->type;
    LLVMValueRef res;
 
@@ -236,7 +241,7 @@ lp_build_add(struct lp_build_context *bld,
       }
    
       if(intrinsic)
-         return lp_build_intrinsic_binary(bld->builder, intrinsic, lp_build_vec_type(bld->type), a, b);
+         return lp_build_intrinsic_binary(builder, intrinsic, lp_build_vec_type(bld->gallivm, bld->type), a, b);
    }
 
    if(LLVMIsConstant(a) && LLVMIsConstant(b))
@@ -246,9 +251,9 @@ lp_build_add(struct lp_build_context *bld,
          res = LLVMConstAdd(a, b);
    else
       if (type.floating)
-         res = LLVMBuildFAdd(bld->builder, a, b, "");
+         res = LLVMBuildFAdd(builder, a, b, "");
       else
-         res = LLVMBuildAdd(bld->builder, a, b, "");
+         res = LLVMBuildAdd(builder, a, b, "");
 
    /* clamp to ceiling of 1.0 */
    if(bld->type.norm && (bld->type.floating || bld->type.fixed))
@@ -265,6 +270,7 @@ LLVMValueRef
 lp_build_sum_vector(struct lp_build_context *bld,
                     LLVMValueRef a)
 {
+   LLVMBuilderRef builder = bld->gallivm->builder;
    const struct lp_type type = bld->type;
    LLVMValueRef index, res;
    unsigned i;
@@ -277,19 +283,19 @@ lp_build_sum_vector(struct lp_build_context *bld,
 
    assert(!bld->type.norm);
 
-   index = LLVMConstInt(LLVMInt32Type(), 0, 0);
-   res = LLVMBuildExtractElement(bld->builder, a, index, "");
+   index = lp_build_const_int32(bld->gallivm, 0);
+   res = LLVMBuildExtractElement(builder, a, index, "");
 
    for (i = 1; i < type.length; i++) {
-      index = LLVMConstInt(LLVMInt32Type(), i, 0);
+      index = lp_build_const_int32(bld->gallivm, i);
       if (type.floating)
-         res = LLVMBuildFAdd(bld->builder, res,
-                            LLVMBuildExtractElement(bld->builder,
+         res = LLVMBuildFAdd(builder, res,
+                            LLVMBuildExtractElement(builder,
                                                     a, index, ""),
                             "");
       else
-         res = LLVMBuildAdd(bld->builder, res,
-                            LLVMBuildExtractElement(bld->builder,
+         res = LLVMBuildAdd(builder, res,
+                            LLVMBuildExtractElement(builder,
                                                     a, index, ""),
                             "");
    }
@@ -306,6 +312,7 @@ lp_build_sub(struct lp_build_context *bld,
              LLVMValueRef a,
              LLVMValueRef b)
 {
+   LLVMBuilderRef builder = bld->gallivm->builder;
    const struct lp_type type = bld->type;
    LLVMValueRef res;
 
@@ -335,7 +342,7 @@ lp_build_sub(struct lp_build_context *bld,
       }
    
       if(intrinsic)
-         return lp_build_intrinsic_binary(bld->builder, intrinsic, lp_build_vec_type(bld->type), a, b);
+         return lp_build_intrinsic_binary(builder, intrinsic, lp_build_vec_type(bld->gallivm, bld->type), a, b);
    }
 
    if(LLVMIsConstant(a) && LLVMIsConstant(b))
@@ -345,9 +352,9 @@ lp_build_sub(struct lp_build_context *bld,
          res = LLVMConstSub(a, b);
    else
       if (type.floating)
-         res = LLVMBuildFSub(bld->builder, a, b, "");
+         res = LLVMBuildFSub(builder, a, b, "");
       else
-         res = LLVMBuildSub(bld->builder, a, b, "");
+         res = LLVMBuildSub(builder, a, b, "");
 
    if(bld->type.norm && (bld->type.floating || bld->type.fixed))
       res = lp_build_max_simple(bld, res, bld->zero);
@@ -398,10 +405,11 @@ lp_build_sub(struct lp_build_context *bld,
  *     http://www.stereopsis.com/doubleblend.html
  */
 static LLVMValueRef
-lp_build_mul_u8n(LLVMBuilderRef builder,
+lp_build_mul_u8n(struct gallivm_state *gallivm,
                  struct lp_type i16_type,
                  LLVMValueRef a, LLVMValueRef b)
 {
+   LLVMBuilderRef builder = gallivm->builder;
    LLVMValueRef c8;
    LLVMValueRef ab;
 
@@ -409,12 +417,12 @@ lp_build_mul_u8n(LLVMBuilderRef builder,
    assert(lp_check_value(i16_type, a));
    assert(lp_check_value(i16_type, b));
 
-   c8 = lp_build_const_int_vec(i16_type, 8);
+   c8 = lp_build_const_int_vec(gallivm, i16_type, 8);
    
 #if 0
    
    /* a*b/255 ~= (a*(b + 1)) >> 256 */
-   b = LLVMBuildAdd(builder, b, lp_build_const_int_vec(i16_type, 1), "");
+   b = LLVMBuildAdd(builder, b, lp_build_const_int_vec(gallium, i16_type, 1), "");
    ab = LLVMBuildMul(builder, a, b, "");
 
 #else
@@ -422,7 +430,7 @@ lp_build_mul_u8n(LLVMBuilderRef builder,
    /* ab/255 ~= (ab + (ab >> 8) + 0x80) >> 8 */
    ab = LLVMBuildMul(builder, a, b, "");
    ab = LLVMBuildAdd(builder, ab, LLVMBuildLShr(builder, ab, c8, ""), "");
-   ab = LLVMBuildAdd(builder, ab, lp_build_const_int_vec(i16_type, 0x80), "");
+   ab = LLVMBuildAdd(builder, ab, lp_build_const_int_vec(gallivm, i16_type, 0x80), "");
 
 #endif
    
@@ -440,6 +448,7 @@ lp_build_mul(struct lp_build_context *bld,
              LLVMValueRef a,
              LLVMValueRef b)
 {
+   LLVMBuilderRef builder = bld->gallivm->builder;
    const struct lp_type type = bld->type;
    LLVMValueRef shift;
    LLVMValueRef res;
@@ -463,14 +472,14 @@ lp_build_mul(struct lp_build_context *bld,
          struct lp_type i16_type = lp_wider_type(type);
          LLVMValueRef al, ah, bl, bh, abl, abh, ab;
 
-         lp_build_unpack2(bld->builder, type, i16_type, a, &al, &ah);
-         lp_build_unpack2(bld->builder, type, i16_type, b, &bl, &bh);
+         lp_build_unpack2(bld->gallivm, type, i16_type, a, &al, &ah);
+         lp_build_unpack2(bld->gallivm, type, i16_type, b, &bl, &bh);
 
          /* PMULLW, PSRLW, PADDW */
-         abl = lp_build_mul_u8n(bld->builder, i16_type, al, bl);
-         abh = lp_build_mul_u8n(bld->builder, i16_type, ah, bh);
+         abl = lp_build_mul_u8n(bld->gallivm, i16_type, al, bl);
+         abh = lp_build_mul_u8n(bld->gallivm, i16_type, ah, bh);
 
-         ab = lp_build_pack2(bld->builder, i16_type, type, abl, abh);
+         ab = lp_build_pack2(bld->gallivm, i16_type, type, abl, abh);
          
          return ab;
       }
@@ -480,7 +489,7 @@ lp_build_mul(struct lp_build_context *bld,
    }
 
    if(type.fixed)
-      shift = lp_build_const_int_vec(type, type.width/2);
+      shift = lp_build_const_int_vec(bld->gallivm, type, type.width/2);
    else
       shift = NULL;
 
@@ -498,14 +507,14 @@ lp_build_mul(struct lp_build_context *bld,
    }
    else {
       if (type.floating)
-         res = LLVMBuildFMul(bld->builder, a, b, "");
+         res = LLVMBuildFMul(builder, a, b, "");
       else
-         res = LLVMBuildMul(bld->builder, a, b, "");
+         res = LLVMBuildMul(builder, a, b, "");
       if(shift) {
          if(type.sign)
-            res = LLVMBuildAShr(bld->builder, res, shift, "");
+            res = LLVMBuildAShr(builder, res, shift, "");
          else
-            res = LLVMBuildLShr(bld->builder, res, shift, "");
+            res = LLVMBuildLShr(builder, res, shift, "");
       }
    }
 
@@ -521,6 +530,7 @@ lp_build_mul_imm(struct lp_build_context *bld,
                  LLVMValueRef a,
                  int b)
 {
+   LLVMBuilderRef builder = bld->gallivm->builder;
    LLVMValueRef factor;
 
    assert(lp_check_value(bld->type, a));
@@ -550,20 +560,20 @@ lp_build_mul_imm(struct lp_build_context *bld,
           * for Inf and NaN.
           */
          unsigned mantissa = lp_mantissa(bld->type);
-         factor = lp_build_const_int_vec(bld->type, (unsigned long long)shift << mantissa);
-         a = LLVMBuildBitCast(bld->builder, a, lp_build_int_vec_type(bld->type), "");
-         a = LLVMBuildAdd(bld->builder, a, factor, "");
-         a = LLVMBuildBitCast(bld->builder, a, lp_build_vec_type(bld->type), "");
+         factor = lp_build_const_int_vec(bld->gallivm, bld->type, (unsigned long long)shift << mantissa);
+         a = LLVMBuildBitCast(builder, a, lp_build_int_vec_type(bld->type), "");
+         a = LLVMBuildAdd(builder, a, factor, "");
+         a = LLVMBuildBitCast(builder, a, lp_build_vec_type(bld->gallivm, bld->type), "");
          return a;
 #endif
       }
       else {
-         factor = lp_build_const_vec(bld->type, shift);
-         return LLVMBuildShl(bld->builder, a, factor, "");
+         factor = lp_build_const_vec(bld->gallivm, bld->type, shift);
+         return LLVMBuildShl(builder, a, factor, "");
       }
    }
 
-   factor = lp_build_const_vec(bld->type, (double)b);
+   factor = lp_build_const_vec(bld->gallivm, bld->type, (double)b);
    return lp_build_mul(bld, a, factor);
 }
 
@@ -576,6 +586,7 @@ lp_build_div(struct lp_build_context *bld,
              LLVMValueRef a,
              LLVMValueRef b)
 {
+   LLVMBuilderRef builder = bld->gallivm->builder;
    const struct lp_type type = bld->type;
 
    assert(lp_check_value(type, a));
@@ -605,11 +616,11 @@ lp_build_div(struct lp_build_context *bld,
       return lp_build_mul(bld, a, lp_build_rcp(bld, b));
 
    if (type.floating)
-      return LLVMBuildFDiv(bld->builder, a, b, "");
+      return LLVMBuildFDiv(builder, a, b, "");
    else if (type.sign)
-      return LLVMBuildSDiv(bld->builder, a, b, "");
+      return LLVMBuildSDiv(builder, a, b, "");
    else
-      return LLVMBuildUDiv(bld->builder, a, b, "");
+      return LLVMBuildUDiv(builder, a, b, "");
 }
 
 
@@ -624,6 +635,7 @@ lp_build_lerp_simple(struct lp_build_context *bld,
                      LLVMValueRef v0,
                      LLVMValueRef v1)
 {
+   LLVMBuilderRef builder = bld->gallivm->builder;
    LLVMValueRef delta;
    LLVMValueRef res;
 
@@ -642,7 +654,7 @@ lp_build_lerp_simple(struct lp_build_context *bld,
        * but it will be wrong for other uses. Basically we need a more
        * powerful lp_type, capable of further distinguishing the values
        * interpretation from the value storage. */
-      res = LLVMBuildAnd(bld->builder, res, lp_build_const_int_vec(bld->type, (1 << bld->type.width/2) - 1), "");
+      res = LLVMBuildAnd(builder, res, lp_build_const_int_vec(bld->gallivm, bld->type, (1 << bld->type.width/2) - 1), "");
    }
 
    return res;
@@ -658,6 +670,7 @@ lp_build_lerp(struct lp_build_context *bld,
               LLVMValueRef v0,
               LLVMValueRef v1)
 {
+   LLVMBuilderRef builder = bld->gallivm->builder;
    const struct lp_type type = bld->type;
    LLVMValueRef res;
 
@@ -683,22 +696,22 @@ lp_build_lerp(struct lp_build_context *bld,
       wide_type.width  = type.width*2;
       wide_type.length = type.length/2;
 
-      lp_build_context_init(&wide_bld, bld->builder, wide_type);
+      lp_build_context_init(&wide_bld, bld->gallivm, wide_type);
 
-      lp_build_unpack2(bld->builder, type, wide_type, x,  &xl,  &xh);
-      lp_build_unpack2(bld->builder, type, wide_type, v0, &v0l, &v0h);
-      lp_build_unpack2(bld->builder, type, wide_type, v1, &v1l, &v1h);
+      lp_build_unpack2(bld->gallivm, type, wide_type, x,  &xl,  &xh);
+      lp_build_unpack2(bld->gallivm, type, wide_type, v0, &v0l, &v0h);
+      lp_build_unpack2(bld->gallivm, type, wide_type, v1, &v1l, &v1h);
 
       /*
        * Scale x from [0, 255] to [0, 256]
        */
 
-      shift = lp_build_const_int_vec(wide_type, type.width - 1);
+      shift = lp_build_const_int_vec(bld->gallivm, wide_type, type.width - 1);
 
       xl = lp_build_add(&wide_bld, xl,
-                        LLVMBuildAShr(bld->builder, xl, shift, ""));
+                        LLVMBuildAShr(builder, xl, shift, ""));
       xh = lp_build_add(&wide_bld, xh,
-                        LLVMBuildAShr(bld->builder, xh, shift, ""));
+                        LLVMBuildAShr(builder, xh, shift, ""));
 
       /*
        * Lerp both halves.
@@ -707,7 +720,7 @@ lp_build_lerp(struct lp_build_context *bld,
       resl = lp_build_lerp_simple(&wide_bld, xl, v0l, v1l);
       resh = lp_build_lerp_simple(&wide_bld, xh, v0h, v1h);
 
-      res = lp_build_pack2(bld->builder, wide_type, type, resl, resh);
+      res = lp_build_pack2(bld->gallivm, wide_type, type, resl, resh);
    } else {
       res = lp_build_lerp_simple(bld, x, v0, v1);
    }
@@ -820,8 +833,9 @@ LLVMValueRef
 lp_build_abs(struct lp_build_context *bld,
              LLVMValueRef a)
 {
+   LLVMBuilderRef builder = bld->gallivm->builder;
    const struct lp_type type = bld->type;
-   LLVMTypeRef vec_type = lp_build_vec_type(type);
+   LLVMTypeRef vec_type = lp_build_vec_type(bld->gallivm, type);
 
    assert(lp_check_value(type, a));
 
@@ -830,27 +844,27 @@ lp_build_abs(struct lp_build_context *bld,
 
    if(type.floating) {
       /* Mask out the sign bit */
-      LLVMTypeRef int_vec_type = lp_build_int_vec_type(type);
+      LLVMTypeRef int_vec_type = lp_build_int_vec_type(bld->gallivm, type);
       unsigned long long absMask = ~(1ULL << (type.width - 1));
-      LLVMValueRef mask = lp_build_const_int_vec(type, ((unsigned long long) absMask));
-      a = LLVMBuildBitCast(bld->builder, a, int_vec_type, "");
-      a = LLVMBuildAnd(bld->builder, a, mask, "");
-      a = LLVMBuildBitCast(bld->builder, a, vec_type, "");
+      LLVMValueRef mask = lp_build_const_int_vec(bld->gallivm, type, ((unsigned long long) absMask));
+      a = LLVMBuildBitCast(builder, a, int_vec_type, "");
+      a = LLVMBuildAnd(builder, a, mask, "");
+      a = LLVMBuildBitCast(builder, a, vec_type, "");
       return a;
    }
 
    if(type.width*type.length == 128 && util_cpu_caps.has_ssse3) {
       switch(type.width) {
       case 8:
-         return lp_build_intrinsic_unary(bld->builder, "llvm.x86.ssse3.pabs.b.128", vec_type, a);
+         return lp_build_intrinsic_unary(builder, "llvm.x86.ssse3.pabs.b.128", vec_type, a);
       case 16:
-         return lp_build_intrinsic_unary(bld->builder, "llvm.x86.ssse3.pabs.w.128", vec_type, a);
+         return lp_build_intrinsic_unary(builder, "llvm.x86.ssse3.pabs.w.128", vec_type, a);
       case 32:
-         return lp_build_intrinsic_unary(bld->builder, "llvm.x86.ssse3.pabs.d.128", vec_type, a);
+         return lp_build_intrinsic_unary(builder, "llvm.x86.ssse3.pabs.d.128", vec_type, a);
       }
    }
 
-   return lp_build_max(bld, a, LLVMBuildNeg(bld->builder, a, ""));
+   return lp_build_max(bld, a, LLVMBuildNeg(builder, a, ""));
 }
 
 
@@ -858,14 +872,16 @@ LLVMValueRef
 lp_build_negate(struct lp_build_context *bld,
                 LLVMValueRef a)
 {
+   LLVMBuilderRef builder = bld->gallivm->builder;
+
    assert(lp_check_value(bld->type, a));
 
 #if HAVE_LLVM >= 0x0207
    if (bld->type.floating)
-      a = LLVMBuildFNeg(bld->builder, a, "");
+      a = LLVMBuildFNeg(builder, a, "");
    else
 #endif
-      a = LLVMBuildNeg(bld->builder, a, "");
+      a = LLVMBuildNeg(builder, a, "");
 
    return a;
 }
@@ -876,6 +892,7 @@ LLVMValueRef
 lp_build_sgn(struct lp_build_context *bld,
              LLVMValueRef a)
 {
+   LLVMBuilderRef builder = bld->gallivm->builder;
    const struct lp_type type = bld->type;
    LLVMValueRef cond;
    LLVMValueRef res;
@@ -895,20 +912,20 @@ lp_build_sgn(struct lp_build_context *bld,
       LLVMValueRef one;
       unsigned long long maskBit = (unsigned long long)1 << (type.width - 1);
 
-      int_type = lp_build_int_vec_type(type);
-      vec_type = lp_build_vec_type(type);
-      mask = lp_build_const_int_vec(type, maskBit);
+      int_type = lp_build_int_vec_type(bld->gallivm, type);
+      vec_type = lp_build_vec_type(bld->gallivm, type);
+      mask = lp_build_const_int_vec(bld->gallivm, type, maskBit);
 
       /* Take the sign bit and add it to 1 constant */
-      sign = LLVMBuildBitCast(bld->builder, a, int_type, "");
-      sign = LLVMBuildAnd(bld->builder, sign, mask, "");
+      sign = LLVMBuildBitCast(builder, a, int_type, "");
+      sign = LLVMBuildAnd(builder, sign, mask, "");
       one = LLVMConstBitCast(bld->one, int_type);
-      res = LLVMBuildOr(bld->builder, sign, one, "");
-      res = LLVMBuildBitCast(bld->builder, res, vec_type, "");
+      res = LLVMBuildOr(builder, sign, one, "");
+      res = LLVMBuildBitCast(builder, res, vec_type, "");
    }
    else
    {
-      LLVMValueRef minus_one = lp_build_const_vec(type, -1.0);
+      LLVMValueRef minus_one = lp_build_const_vec(bld->gallivm, type, -1.0);
       cond = lp_build_cmp(bld, PIPE_FUNC_GREATER, a, bld->zero);
       res = lp_build_select(bld, cond, bld->one, minus_one);
    }
@@ -931,11 +948,12 @@ LLVMValueRef
 lp_build_set_sign(struct lp_build_context *bld,
                   LLVMValueRef a, LLVMValueRef sign)
 {
+   LLVMBuilderRef builder = bld->gallivm->builder;
    const struct lp_type type = bld->type;
-   LLVMTypeRef int_vec_type = lp_build_int_vec_type(type);
-   LLVMTypeRef vec_type = lp_build_vec_type(type);
-   LLVMValueRef shift = lp_build_const_int_vec(type, type.width - 1);
-   LLVMValueRef mask = lp_build_const_int_vec(type,
+   LLVMTypeRef int_vec_type = lp_build_int_vec_type(bld->gallivm, type);
+   LLVMTypeRef vec_type = lp_build_vec_type(bld->gallivm, type);
+   LLVMValueRef shift = lp_build_const_int_vec(bld->gallivm, type, type.width - 1);
+   LLVMValueRef mask = lp_build_const_int_vec(bld->gallivm, type,
                              ~((unsigned long long) 1 << (type.width - 1)));
    LLVMValueRef val, res;
 
@@ -943,15 +961,15 @@ lp_build_set_sign(struct lp_build_context *bld,
    assert(lp_check_value(type, a));
 
    /* val = reinterpret_cast<int>(a) */
-   val = LLVMBuildBitCast(bld->builder, a, int_vec_type, "");
+   val = LLVMBuildBitCast(builder, a, int_vec_type, "");
    /* val = val & mask */
-   val = LLVMBuildAnd(bld->builder, val, mask, "");
+   val = LLVMBuildAnd(builder, val, mask, "");
    /* sign = sign << shift */
-   sign = LLVMBuildShl(bld->builder, sign, shift, "");
+   sign = LLVMBuildShl(builder, sign, shift, "");
    /* res = val | sign */
-   res = LLVMBuildOr(bld->builder, val, sign, "");
+   res = LLVMBuildOr(builder, val, sign, "");
    /* res = reinterpret_cast<float>(res) */
-   res = LLVMBuildBitCast(bld->builder, res, vec_type, "");
+   res = LLVMBuildBitCast(builder, res, vec_type, "");
 
    return res;
 }
@@ -964,12 +982,13 @@ LLVMValueRef
 lp_build_int_to_float(struct lp_build_context *bld,
                       LLVMValueRef a)
 {
+   LLVMBuilderRef builder = bld->gallivm->builder;
    const struct lp_type type = bld->type;
-   LLVMTypeRef vec_type = lp_build_vec_type(type);
+   LLVMTypeRef vec_type = lp_build_vec_type(bld->gallivm, type);
 
    assert(type.floating);
 
-   return LLVMBuildSIToFP(bld->builder, a, vec_type, "");
+   return LLVMBuildSIToFP(builder, a, vec_type, "");
 }
 
 
@@ -994,8 +1013,9 @@ lp_build_round_sse41(struct lp_build_context *bld,
                      LLVMValueRef a,
                      enum lp_build_round_sse41_mode mode)
 {
+   LLVMBuilderRef builder = bld->gallivm->builder;
    const struct lp_type type = bld->type;
-   LLVMTypeRef i32t = LLVMInt32Type();
+   LLVMTypeRef i32t = LLVMInt32TypeInContext(bld->gallivm->context);
    const char *intrinsic;
    LLVMValueRef res;
 
@@ -1027,13 +1047,13 @@ lp_build_round_sse41(struct lp_build_context *bld,
       undef = LLVMGetUndef(vec_type);
 
       args[0] = undef;
-      args[1] = LLVMBuildInsertElement(bld->builder, undef, a, index0, "");
+      args[1] = LLVMBuildInsertElement(builder, undef, a, index0, "");
       args[2] = LLVMConstInt(i32t, mode, 0);
 
-      res = lp_build_intrinsic(bld->builder, intrinsic,
+      res = lp_build_intrinsic(builder, intrinsic,
                                vec_type, args, Elements(args));
 
-      res = LLVMBuildExtractElement(bld->builder, res, index0, "");
+      res = LLVMBuildExtractElement(builder, res, index0, "");
    }
    else {
       assert(type.width*type.length == 128);
@@ -1050,7 +1070,7 @@ lp_build_round_sse41(struct lp_build_context *bld,
          return bld->undef;
       }
 
-      res = lp_build_intrinsic_binary(bld->builder, intrinsic,
+      res = lp_build_intrinsic_binary(builder, intrinsic,
                                       bld->vec_type, a,
                                       LLVMConstInt(i32t, mode, 0));
    }
@@ -1063,9 +1083,10 @@ static INLINE LLVMValueRef
 lp_build_iround_nearest_sse2(struct lp_build_context *bld,
                              LLVMValueRef a)
 {
+   LLVMBuilderRef builder = bld->gallivm->builder;
    const struct lp_type type = bld->type;
-   LLVMTypeRef i32t = LLVMInt32Type();
-   LLVMTypeRef ret_type = lp_build_int_vec_type(type);
+   LLVMTypeRef i32t = LLVMInt32TypeInContext(bld->gallivm->context);
+   LLVMTypeRef ret_type = lp_build_int_vec_type(bld->gallivm, type);
    const char *intrinsic;
    LLVMValueRef res;
 
@@ -1089,9 +1110,9 @@ lp_build_iround_nearest_sse2(struct lp_build_context *bld,
 
       undef = LLVMGetUndef(vec_type);
 
-      arg = LLVMBuildInsertElement(bld->builder, undef, a, index0, "");
+      arg = LLVMBuildInsertElement(builder, undef, a, index0, "");
 
-      res = lp_build_intrinsic_unary(bld->builder, intrinsic,
+      res = lp_build_intrinsic_unary(builder, intrinsic,
                                      ret_type, arg);
    }
    else {
@@ -1099,7 +1120,7 @@ lp_build_iround_nearest_sse2(struct lp_build_context *bld,
 
       intrinsic = "llvm.x86.sse2.cvtps2dq";
 
-      res = lp_build_intrinsic_unary(bld->builder, intrinsic,
+      res = lp_build_intrinsic_unary(builder, intrinsic,
                                      ret_type, a);
    }
 
@@ -1116,6 +1137,7 @@ LLVMValueRef
 lp_build_trunc(struct lp_build_context *bld,
                LLVMValueRef a)
 {
+   LLVMBuilderRef builder = bld->gallivm->builder;
    const struct lp_type type = bld->type;
 
    assert(type.floating);
@@ -1126,11 +1148,11 @@ lp_build_trunc(struct lp_build_context *bld,
       return lp_build_round_sse41(bld, a, LP_BUILD_ROUND_SSE41_TRUNCATE);
    }
    else {
-      LLVMTypeRef vec_type = lp_build_vec_type(type);
-      LLVMTypeRef int_vec_type = lp_build_int_vec_type(type);
+      LLVMTypeRef vec_type = lp_build_vec_type(bld->gallivm, type);
+      LLVMTypeRef int_vec_type = lp_build_int_vec_type(bld->gallivm, type);
       LLVMValueRef res;
-      res = LLVMBuildFPToSI(bld->builder, a, int_vec_type, "");
-      res = LLVMBuildSIToFP(bld->builder, res, vec_type, "");
+      res = LLVMBuildFPToSI(builder, a, int_vec_type, "");
+      res = LLVMBuildSIToFP(builder, res, vec_type, "");
       return res;
    }
 }
@@ -1146,6 +1168,7 @@ LLVMValueRef
 lp_build_round(struct lp_build_context *bld,
                LLVMValueRef a)
 {
+   LLVMBuilderRef builder = bld->gallivm->builder;
    const struct lp_type type = bld->type;
 
    assert(type.floating);
@@ -1156,10 +1179,10 @@ lp_build_round(struct lp_build_context *bld,
       return lp_build_round_sse41(bld, a, LP_BUILD_ROUND_SSE41_NEAREST);
    }
    else {
-      LLVMTypeRef vec_type = lp_build_vec_type(type);
+      LLVMTypeRef vec_type = lp_build_vec_type(bld->gallivm, type);
       LLVMValueRef res;
       res = lp_build_iround(bld, a);
-      res = LLVMBuildSIToFP(bld->builder, res, vec_type, "");
+      res = LLVMBuildSIToFP(builder, res, vec_type, "");
       return res;
    }
 }
@@ -1174,6 +1197,7 @@ LLVMValueRef
 lp_build_floor(struct lp_build_context *bld,
                LLVMValueRef a)
 {
+   LLVMBuilderRef builder = bld->gallivm->builder;
    const struct lp_type type = bld->type;
 
    assert(type.floating);
@@ -1184,10 +1208,10 @@ lp_build_floor(struct lp_build_context *bld,
       return lp_build_round_sse41(bld, a, LP_BUILD_ROUND_SSE41_FLOOR);
    }
    else {
-      LLVMTypeRef vec_type = lp_build_vec_type(type);
+      LLVMTypeRef vec_type = lp_build_vec_type(bld->gallivm, type);
       LLVMValueRef res;
       res = lp_build_ifloor(bld, a);
-      res = LLVMBuildSIToFP(bld->builder, res, vec_type, "");
+      res = LLVMBuildSIToFP(builder, res, vec_type, "");
       return res;
    }
 }
@@ -1202,6 +1226,7 @@ LLVMValueRef
 lp_build_ceil(struct lp_build_context *bld,
               LLVMValueRef a)
 {
+   LLVMBuilderRef builder = bld->gallivm->builder;
    const struct lp_type type = bld->type;
 
    assert(type.floating);
@@ -1212,10 +1237,10 @@ lp_build_ceil(struct lp_build_context *bld,
       return lp_build_round_sse41(bld, a, LP_BUILD_ROUND_SSE41_CEIL);
    }
    else {
-      LLVMTypeRef vec_type = lp_build_vec_type(type);
+      LLVMTypeRef vec_type = lp_build_vec_type(bld->gallivm, type);
       LLVMValueRef res;
       res = lp_build_iceil(bld, a);
-      res = LLVMBuildSIToFP(bld->builder, res, vec_type, "");
+      res = LLVMBuildSIToFP(builder, res, vec_type, "");
       return res;
    }
 }
@@ -1243,13 +1268,14 @@ LLVMValueRef
 lp_build_itrunc(struct lp_build_context *bld,
                 LLVMValueRef a)
 {
+   LLVMBuilderRef builder = bld->gallivm->builder;
    const struct lp_type type = bld->type;
-   LLVMTypeRef int_vec_type = lp_build_int_vec_type(type);
+   LLVMTypeRef int_vec_type = lp_build_int_vec_type(bld->gallivm, type);
 
    assert(type.floating);
    assert(lp_check_value(type, a));
 
-   return LLVMBuildFPToSI(bld->builder, a, int_vec_type, "");
+   return LLVMBuildFPToSI(builder, a, int_vec_type, "");
 }
 
 
@@ -1263,6 +1289,7 @@ LLVMValueRef
 lp_build_iround(struct lp_build_context *bld,
                 LLVMValueRef a)
 {
+   LLVMBuilderRef builder = bld->gallivm->builder;
    const struct lp_type type = bld->type;
    LLVMTypeRef int_vec_type = bld->int_vec_type;
    LLVMValueRef res;
@@ -1282,27 +1309,28 @@ lp_build_iround(struct lp_build_context *bld,
    else {
       LLVMValueRef half;
 
-      half = lp_build_const_vec(type, 0.5);
+      half = lp_build_const_vec(bld->gallivm, type, 0.5);
 
       if (type.sign) {
          LLVMTypeRef vec_type = bld->vec_type;
-         LLVMValueRef mask = lp_build_const_int_vec(type, (unsigned long long)1 << (type.width - 1));
+         LLVMValueRef mask = lp_build_const_int_vec(bld->gallivm, type,
+                                    (unsigned long long)1 << (type.width - 1));
          LLVMValueRef sign;
 
          /* get sign bit */
-         sign = LLVMBuildBitCast(bld->builder, a, int_vec_type, "");
-         sign = LLVMBuildAnd(bld->builder, sign, mask, "");
+         sign = LLVMBuildBitCast(builder, a, int_vec_type, "");
+         sign = LLVMBuildAnd(builder, sign, mask, "");
 
          /* sign * 0.5 */
-         half = LLVMBuildBitCast(bld->builder, half, int_vec_type, "");
-         half = LLVMBuildOr(bld->builder, sign, half, "");
-         half = LLVMBuildBitCast(bld->builder, half, vec_type, "");
+         half = LLVMBuildBitCast(builder, half, int_vec_type, "");
+         half = LLVMBuildOr(builder, sign, half, "");
+         half = LLVMBuildBitCast(builder, half, vec_type, "");
       }
 
-      res = LLVMBuildFAdd(bld->builder, a, half, "");
+      res = LLVMBuildFAdd(builder, a, half, "");
    }
 
-   res = LLVMBuildFPToSI(bld->builder, res, int_vec_type, "");
+   res = LLVMBuildFPToSI(builder, res, int_vec_type, "");
 
    return res;
 }
@@ -1317,6 +1345,7 @@ LLVMValueRef
 lp_build_ifloor(struct lp_build_context *bld,
                 LLVMValueRef a)
 {
+   LLVMBuilderRef builder = bld->gallivm->builder;
    const struct lp_type type = bld->type;
    LLVMTypeRef int_vec_type = bld->int_vec_type;
    LLVMValueRef res;
@@ -1335,29 +1364,34 @@ lp_build_ifloor(struct lp_build_context *bld,
          /* Take the sign bit and add it to 1 constant */
          LLVMTypeRef vec_type = bld->vec_type;
          unsigned mantissa = lp_mantissa(type);
-         LLVMValueRef mask = lp_build_const_int_vec(type, (unsigned long long)1 << (type.width - 1));
+         LLVMValueRef mask = lp_build_const_int_vec(bld->gallivm, type,
+                                  (unsigned long long)1 << (type.width - 1));
          LLVMValueRef sign;
          LLVMValueRef offset;
 
          /* sign = a < 0 ? ~0 : 0 */
-         sign = LLVMBuildBitCast(bld->builder, a, int_vec_type, "");
-         sign = LLVMBuildAnd(bld->builder, sign, mask, "");
-         sign = LLVMBuildAShr(bld->builder, sign, lp_build_const_int_vec(type, type.width - 1), "ifloor.sign");
+         sign = LLVMBuildBitCast(builder, a, int_vec_type, "");
+         sign = LLVMBuildAnd(builder, sign, mask, "");
+         sign = LLVMBuildAShr(builder, sign,
+                              lp_build_const_int_vec(bld->gallivm, type,
+                                                     type.width - 1),
+                              "ifloor.sign");
 
          /* offset = -0.99999(9)f */
-         offset = lp_build_const_vec(type, -(double)(((unsigned long long)1 << mantissa) - 10)/((unsigned long long)1 << mantissa));
+         offset = lp_build_const_vec(bld->gallivm, type,
+                                     -(double)(((unsigned long long)1 << mantissa) - 10)/((unsigned long long)1 << mantissa));
          offset = LLVMConstBitCast(offset, int_vec_type);
 
          /* offset = a < 0 ? offset : 0.0f */
-         offset = LLVMBuildAnd(bld->builder, offset, sign, "");
-         offset = LLVMBuildBitCast(bld->builder, offset, vec_type, "ifloor.offset");
+         offset = LLVMBuildAnd(builder, offset, sign, "");
+         offset = LLVMBuildBitCast(builder, offset, vec_type, "ifloor.offset");
 
-         res = LLVMBuildFAdd(bld->builder, res, offset, "ifloor.res");
+         res = LLVMBuildFAdd(builder, res, offset, "ifloor.res");
       }
    }
 
    /* round to nearest (toward zero) */
-   res = LLVMBuildFPToSI(bld->builder, res, int_vec_type, "ifloor.res");
+   res = LLVMBuildFPToSI(builder, res, int_vec_type, "ifloor.res");
 
    return res;
 }
@@ -1372,6 +1406,7 @@ LLVMValueRef
 lp_build_iceil(struct lp_build_context *bld,
                LLVMValueRef a)
 {
+   LLVMBuilderRef builder = bld->gallivm->builder;
    const struct lp_type type = bld->type;
    LLVMTypeRef int_vec_type = bld->int_vec_type;
    LLVMValueRef res;
@@ -1389,29 +1424,34 @@ lp_build_iceil(struct lp_build_context *bld,
       LLVMValueRef offset;
 
       /* offset = 0.99999(9)f */
-      offset = lp_build_const_vec(type, (double)(((unsigned long long)1 << mantissa) - 10)/((unsigned long long)1 << mantissa));
+      offset = lp_build_const_vec(bld->gallivm, type,
+                                  (double)(((unsigned long long)1 << mantissa) - 10)/((unsigned long long)1 << mantissa));
 
       if (type.sign) {
-         LLVMValueRef mask = lp_build_const_int_vec(type, (unsigned long long)1 << (type.width - 1));
+         LLVMValueRef mask = lp_build_const_int_vec(bld->gallivm, type,
+                                (unsigned long long)1 << (type.width - 1));
          LLVMValueRef sign;
 
          /* sign = a < 0 ? 0 : ~0 */
-         sign = LLVMBuildBitCast(bld->builder, a, int_vec_type, "");
-         sign = LLVMBuildAnd(bld->builder, sign, mask, "");
-         sign = LLVMBuildAShr(bld->builder, sign, lp_build_const_int_vec(type, type.width - 1), "iceil.sign");
-         sign = LLVMBuildNot(bld->builder, sign, "iceil.not");
+         sign = LLVMBuildBitCast(builder, a, int_vec_type, "");
+         sign = LLVMBuildAnd(builder, sign, mask, "");
+         sign = LLVMBuildAShr(builder, sign,
+                              lp_build_const_int_vec(bld->gallivm, type,
+                                                     type.width - 1),
+                              "iceil.sign");
+         sign = LLVMBuildNot(builder, sign, "iceil.not");
 
          /* offset = a < 0 ? 0.0 : offset */
          offset = LLVMConstBitCast(offset, int_vec_type);
-         offset = LLVMBuildAnd(bld->builder, offset, sign, "");
-         offset = LLVMBuildBitCast(bld->builder, offset, vec_type, "iceil.offset");
+         offset = LLVMBuildAnd(builder, offset, sign, "");
+         offset = LLVMBuildBitCast(builder, offset, vec_type, "iceil.offset");
       }
 
-      res = LLVMBuildFAdd(bld->builder, a, offset, "iceil.res");
+      res = LLVMBuildFAdd(builder, a, offset, "iceil.res");
    }
 
    /* round to nearest (toward zero) */
-   res = LLVMBuildFPToSI(bld->builder, res, int_vec_type, "iceil.res");
+   res = LLVMBuildFPToSI(builder, res, int_vec_type, "iceil.res");
 
    return res;
 }
@@ -1429,6 +1469,7 @@ lp_build_ifloor_fract(struct lp_build_context *bld,
                       LLVMValueRef *out_ipart,
                       LLVMValueRef *out_fpart)
 {
+   LLVMBuilderRef builder = bld->gallivm->builder;
    const struct lp_type type = bld->type;
    LLVMValueRef ipart;
 
@@ -1442,8 +1483,8 @@ lp_build_ifloor_fract(struct lp_build_context *bld,
        */
 
       ipart = lp_build_floor(bld, a);
-      *out_fpart = LLVMBuildFSub(bld->builder, a, ipart, "fpart");
-      *out_ipart = LLVMBuildFPToSI(bld->builder, ipart, bld->int_vec_type, "ipart");
+      *out_fpart = LLVMBuildFSub(builder, a, ipart, "fpart");
+      *out_ipart = LLVMBuildFPToSI(builder, ipart, bld->int_vec_type, "ipart");
    }
    else {
       /*
@@ -1451,8 +1492,8 @@ lp_build_ifloor_fract(struct lp_build_context *bld,
        */
 
       *out_ipart = lp_build_ifloor(bld, a);
-      ipart = LLVMBuildSIToFP(bld->builder, *out_ipart, bld->vec_type, "ipart");
-      *out_fpart = LLVMBuildFSub(bld->builder, a, ipart, "fpart");
+      ipart = LLVMBuildSIToFP(builder, *out_ipart, bld->vec_type, "ipart");
+      *out_fpart = LLVMBuildFSub(builder, a, ipart, "fpart");
    }
 }
 
@@ -1461,8 +1502,9 @@ LLVMValueRef
 lp_build_sqrt(struct lp_build_context *bld,
               LLVMValueRef a)
 {
+   LLVMBuilderRef builder = bld->gallivm->builder;
    const struct lp_type type = bld->type;
-   LLVMTypeRef vec_type = lp_build_vec_type(type);
+   LLVMTypeRef vec_type = lp_build_vec_type(bld->gallivm, type);
    char intrinsic[32];
 
    assert(lp_check_value(type, a));
@@ -1473,7 +1515,7 @@ lp_build_sqrt(struct lp_build_context *bld,
    assert(type.floating);
    util_snprintf(intrinsic, sizeof intrinsic, "llvm.sqrt.v%uf%u", type.length, type.width);
 
-   return lp_build_intrinsic_unary(bld->builder, intrinsic, vec_type, a);
+   return lp_build_intrinsic_unary(builder, intrinsic, vec_type, a);
 }
 
 
@@ -1496,12 +1538,13 @@ lp_build_rcp_refine(struct lp_build_context *bld,
                     LLVMValueRef a,
                     LLVMValueRef rcp_a)
 {
-   LLVMValueRef two = lp_build_const_vec(bld->type, 2.0);
+   LLVMBuilderRef builder = bld->gallivm->builder;
+   LLVMValueRef two = lp_build_const_vec(bld->gallivm, bld->type, 2.0);
    LLVMValueRef res;
 
-   res = LLVMBuildFMul(bld->builder, a, rcp_a, "");
-   res = LLVMBuildFSub(bld->builder, two, res, "");
-   res = LLVMBuildFMul(bld->builder, rcp_a, res, "");
+   res = LLVMBuildFMul(builder, a, rcp_a, "");
+   res = LLVMBuildFSub(builder, two, res, "");
+   res = LLVMBuildFMul(builder, rcp_a, res, "");
 
    return res;
 }
@@ -1511,6 +1554,7 @@ LLVMValueRef
 lp_build_rcp(struct lp_build_context *bld,
              LLVMValueRef a)
 {
+   LLVMBuilderRef builder = bld->gallivm->builder;
    const struct lp_type type = bld->type;
 
    assert(lp_check_value(type, a));
@@ -1545,7 +1589,7 @@ lp_build_rcp(struct lp_build_context *bld,
       LLVMValueRef res;
       unsigned i;
 
-      res = lp_build_intrinsic_unary(bld->builder, "llvm.x86.sse.rcp.ps", bld->vec_type, a);
+      res = lp_build_intrinsic_unary(builder, "llvm.x86.sse.rcp.ps", bld->vec_type, a);
 
       for (i = 0; i < num_iterations; ++i) {
          res = lp_build_rcp_refine(bld, a, res);
@@ -1554,7 +1598,7 @@ lp_build_rcp(struct lp_build_context *bld,
       return res;
    }
 
-   return LLVMBuildFDiv(bld->builder, bld->one, a, "");
+   return LLVMBuildFDiv(builder, bld->one, a, "");
 }
 
 
@@ -1571,15 +1615,16 @@ lp_build_rsqrt_refine(struct lp_build_context *bld,
                       LLVMValueRef a,
                       LLVMValueRef rsqrt_a)
 {
-   LLVMValueRef half = lp_build_const_vec(bld->type, 0.5);
-   LLVMValueRef three = lp_build_const_vec(bld->type, 3.0);
+   LLVMBuilderRef builder = bld->gallivm->builder;
+   LLVMValueRef half = lp_build_const_vec(bld->gallivm, bld->type, 0.5);
+   LLVMValueRef three = lp_build_const_vec(bld->gallivm, bld->type, 3.0);
    LLVMValueRef res;
 
-   res = LLVMBuildFMul(bld->builder, rsqrt_a, rsqrt_a, "");
-   res = LLVMBuildFMul(bld->builder, a, res, "");
-   res = LLVMBuildFSub(bld->builder, three, res, "");
-   res = LLVMBuildFMul(bld->builder, rsqrt_a, res, "");
-   res = LLVMBuildFMul(bld->builder, half, res, "");
+   res = LLVMBuildFMul(builder, rsqrt_a, rsqrt_a, "");
+   res = LLVMBuildFMul(builder, a, res, "");
+   res = LLVMBuildFSub(builder, three, res, "");
+   res = LLVMBuildFMul(builder, rsqrt_a, res, "");
+   res = LLVMBuildFMul(builder, half, res, "");
 
    return res;
 }
@@ -1592,6 +1637,7 @@ LLVMValueRef
 lp_build_rsqrt(struct lp_build_context *bld,
                LLVMValueRef a)
 {
+   LLVMBuilderRef builder = bld->gallivm->builder;
    const struct lp_type type = bld->type;
 
    assert(lp_check_value(type, a));
@@ -1603,7 +1649,7 @@ lp_build_rsqrt(struct lp_build_context *bld,
       LLVMValueRef res;
       unsigned i;
 
-      res = lp_build_intrinsic_unary(bld->builder, "llvm.x86.sse.rsqrt.ps", bld->vec_type, a);
+      res = lp_build_intrinsic_unary(builder, "llvm.x86.sse.rsqrt.ps", bld->vec_type, a);
 
       for (i = 0; i < num_iterations; ++i) {
          res = lp_build_rsqrt_refine(bld, a, res);
@@ -1617,17 +1663,17 @@ lp_build_rsqrt(struct lp_build_context *bld,
 
 
 static inline LLVMValueRef
-lp_build_const_v4si(unsigned long value)
+lp_build_const_v4si(struct gallivm_state *gallivm, unsigned long value)
 {
-   LLVMValueRef element = LLVMConstInt(LLVMInt32Type(), value, 0);
+   LLVMValueRef element = lp_build_const_int32(gallivm, value);
    LLVMValueRef elements[4] = { element, element, element, element };
    return LLVMConstVector(elements, 4);
 }
 
 static inline LLVMValueRef
-lp_build_const_v4sf(float value)
+lp_build_const_v4sf(struct gallivm_state *gallivm, float value)
 {
-   LLVMValueRef element = LLVMConstReal(LLVMFloatType(), value);
+   LLVMValueRef element = lp_build_const_float(gallivm, value);
    LLVMValueRef elements[4] = { element, element, element, element };
    return LLVMConstVector(elements, 4);
 }
@@ -1640,17 +1686,19 @@ LLVMValueRef
 lp_build_sin(struct lp_build_context *bld,
              LLVMValueRef a)
 {
+   LLVMBuilderRef builder = bld->gallivm->builder;
+   struct gallivm_state *gallivm = bld->gallivm;
    struct lp_type int_type = lp_int_type(bld->type);
-   LLVMBuilderRef b = bld->builder;
-   LLVMTypeRef v4sf = LLVMVectorType(LLVMFloatType(), 4);
-   LLVMTypeRef v4si = LLVMVectorType(LLVMInt32Type(), 4);
+   LLVMBuilderRef b = builder;
+   LLVMTypeRef v4sf = LLVMVectorType(LLVMFloatTypeInContext(bld->gallivm->context), 4);
+   LLVMTypeRef v4si = LLVMVectorType(LLVMInt32TypeInContext(bld->gallivm->context), 4);
 
    /*
     *  take the absolute value,
     *  x = _mm_and_ps(x, *(v4sf*)_ps_inv_sign_mask);
     */
 
-   LLVMValueRef inv_sig_mask = lp_build_const_v4si(~0x80000000);
+   LLVMValueRef inv_sig_mask = lp_build_const_v4si(bld->gallivm, ~0x80000000);
    LLVMValueRef a_v4si = LLVMBuildBitCast(b, a, v4si, "a_v4si");
 
    LLVMValueRef absi = LLVMBuildAnd(b, a_v4si, inv_sig_mask, "absi");
@@ -1660,7 +1708,7 @@ lp_build_sin(struct lp_build_context *bld,
     * extract the sign bit (upper one)
     * sign_bit = _mm_and_ps(sign_bit, *(v4sf*)_ps_sign_mask);
     */
-   LLVMValueRef sig_mask = lp_build_const_v4si(0x80000000);
+   LLVMValueRef sig_mask = lp_build_const_v4si(bld->gallivm, 0x80000000);
    LLVMValueRef sign_bit_i = LLVMBuildAnd(b, a_v4si, sig_mask, "sign_bit_i");
 
    /*
@@ -1668,7 +1716,7 @@ lp_build_sin(struct lp_build_context *bld,
     * y = _mm_mul_ps(x, *(v4sf*)_ps_cephes_FOPI);
     */
    
-   LLVMValueRef FOPi = lp_build_const_v4sf(1.27323954473516);
+   LLVMValueRef FOPi = lp_build_const_v4sf(gallivm, 1.27323954473516);
    LLVMValueRef scale_y = LLVMBuildFMul(b, x_abs, FOPi, "scale_y");
 
    /*
@@ -1683,12 +1731,12 @@ lp_build_sin(struct lp_build_context *bld,
     * emm2 = _mm_add_epi32(emm2, *(v4si*)_pi32_1);
     */
 
-   LLVMValueRef all_one = lp_build_const_v4si(1);
+   LLVMValueRef all_one = lp_build_const_v4si(bld->gallivm, 1);
    LLVMValueRef emm2_add =  LLVMBuildAdd(b, emm2_i, all_one, "emm2_add");
    /*
     * emm2 = _mm_and_si128(emm2, *(v4si*)_pi32_inv1);
     */
-   LLVMValueRef inv_one = lp_build_const_v4si(~1);
+   LLVMValueRef inv_one = lp_build_const_v4si(bld->gallivm, ~1);
    LLVMValueRef emm2_and =  LLVMBuildAnd(b, emm2_add, inv_one, "emm2_and");
 
    /*
@@ -1699,13 +1747,13 @@ lp_build_sin(struct lp_build_context *bld,
    /* get the swap sign flag
     * emm0 = _mm_and_si128(emm2, *(v4si*)_pi32_4);
     */
-   LLVMValueRef pi32_4 = lp_build_const_v4si(4);
+   LLVMValueRef pi32_4 = lp_build_const_v4si(bld->gallivm, 4);
    LLVMValueRef emm0_and =  LLVMBuildAnd(b, emm2_add, pi32_4, "emm0_and");
    
    /*
     * emm2 = _mm_slli_epi32(emm0, 29);
     */  
-   LLVMValueRef const_29 = lp_build_const_v4si(29);
+   LLVMValueRef const_29 = lp_build_const_v4si(bld->gallivm, 29);
    LLVMValueRef swap_sign_bit = LLVMBuildShl(b, emm0_and, const_29, "swap_sign_bit");
 
    /*
@@ -1718,10 +1766,11 @@ lp_build_sin(struct lp_build_context *bld,
     * emm2 = _mm_cmpeq_epi32(emm2, _mm_setzero_si128());
     */
 
-   LLVMValueRef pi32_2 = lp_build_const_v4si(2);
+   LLVMValueRef pi32_2 = lp_build_const_v4si(bld->gallivm, 2);
    LLVMValueRef emm2_3 =  LLVMBuildAnd(b, emm2_and, pi32_2, "emm2_3");
-   LLVMValueRef poly_mask = lp_build_compare(b, int_type, PIPE_FUNC_EQUAL,
-                                             emm2_3, lp_build_const_v4si(0));
+   LLVMValueRef poly_mask = lp_build_compare(bld->gallivm,
+                                             int_type, PIPE_FUNC_EQUAL,
+                                             emm2_3, lp_build_const_v4si(bld->gallivm, 0));
    /*
     *   sign_bit = _mm_xor_ps(sign_bit, swap_sign_bit);
     */
@@ -1732,9 +1781,9 @@ lp_build_sin(struct lp_build_context *bld,
     * _PS_CONST(minus_cephes_DP2, -2.4187564849853515625e-4);
     * _PS_CONST(minus_cephes_DP3, -3.77489497744594108e-8);
     */
-   LLVMValueRef DP1 = lp_build_const_v4sf(-0.78515625);
-   LLVMValueRef DP2 = lp_build_const_v4sf(-2.4187564849853515625e-4);
-   LLVMValueRef DP3 = lp_build_const_v4sf(-3.77489497744594108e-8);
+   LLVMValueRef DP1 = lp_build_const_v4sf(gallivm, -0.78515625);
+   LLVMValueRef DP2 = lp_build_const_v4sf(gallivm, -2.4187564849853515625e-4);
+   LLVMValueRef DP3 = lp_build_const_v4sf(gallivm, -3.77489497744594108e-8);
 
    /*
     * The magic pass: "Extended precision modular arithmetic" 
@@ -1769,9 +1818,9 @@ lp_build_sin(struct lp_build_context *bld,
     * _PS_CONST(coscof_p1, -1.388731625493765E-003);
     * _PS_CONST(coscof_p2,  4.166664568298827E-002);
     */
-   LLVMValueRef coscof_p0 = lp_build_const_v4sf(2.443315711809948E-005);
-   LLVMValueRef coscof_p1 = lp_build_const_v4sf(-1.388731625493765E-003);
-   LLVMValueRef coscof_p2 = lp_build_const_v4sf(4.166664568298827E-002);
+   LLVMValueRef coscof_p0 = lp_build_const_v4sf(gallivm, 2.443315711809948E-005);
+   LLVMValueRef coscof_p1 = lp_build_const_v4sf(gallivm, -1.388731625493765E-003);
+   LLVMValueRef coscof_p2 = lp_build_const_v4sf(gallivm, 4.166664568298827E-002);
 
    /*
     * y = *(v4sf*)_ps_coscof_p0;
@@ -1790,10 +1839,10 @@ lp_build_sin(struct lp_build_context *bld,
     * y = _mm_sub_ps(y, tmp);
     * y = _mm_add_ps(y, *(v4sf*)_ps_1);
     */ 
-   LLVMValueRef half = lp_build_const_v4sf(0.5);
+   LLVMValueRef half = lp_build_const_v4sf(gallivm, 0.5);
    LLVMValueRef tmp = LLVMBuildFMul(b, z, half, "tmp");
    LLVMValueRef y_9 = LLVMBuildFSub(b, y_8, tmp, "y_8");
-   LLVMValueRef one = lp_build_const_v4sf(1.0);
+   LLVMValueRef one = lp_build_const_v4sf(gallivm, 1.0);
    LLVMValueRef y_10 = LLVMBuildFAdd(b, y_9, one, "y_9");
 
    /*
@@ -1801,9 +1850,9 @@ lp_build_sin(struct lp_build_context *bld,
     * _PS_CONST(sincof_p1,  8.3321608736E-3);
     * _PS_CONST(sincof_p2, -1.6666654611E-1);
     */
-   LLVMValueRef sincof_p0 = lp_build_const_v4sf(-1.9515295891E-4);
-   LLVMValueRef sincof_p1 = lp_build_const_v4sf(8.3321608736E-3);
-   LLVMValueRef sincof_p2 = lp_build_const_v4sf(-1.6666654611E-1);
+   LLVMValueRef sincof_p0 = lp_build_const_v4sf(gallivm, -1.9515295891E-4);
+   LLVMValueRef sincof_p1 = lp_build_const_v4sf(gallivm, 8.3321608736E-3);
+   LLVMValueRef sincof_p2 = lp_build_const_v4sf(gallivm, -1.6666654611E-1);
 
    /*
     * Evaluate the second polynom  (Pi/4 <= x <= 0)
@@ -1836,7 +1885,7 @@ lp_build_sin(struct lp_build_context *bld,
    LLVMValueRef y2_i = LLVMBuildBitCast(b, y2_9, v4si, "y2_i");
    LLVMValueRef y_i = LLVMBuildBitCast(b, y_10, v4si, "y_i");
    LLVMValueRef y2_and = LLVMBuildAnd(b, y2_i, poly_mask, "y2_and");
-   LLVMValueRef inv = lp_build_const_v4si(~0);
+   LLVMValueRef inv = lp_build_const_v4si(bld->gallivm, ~0);
    LLVMValueRef poly_mask_inv = LLVMBuildXor(b, poly_mask, inv, "poly_mask_inv");
    LLVMValueRef y_and = LLVMBuildAnd(b, y_i, poly_mask_inv, "y_and");
    LLVMValueRef y_combine = LLVMBuildAdd(b, y_and, y2_and, "y_combine");
@@ -1858,17 +1907,19 @@ LLVMValueRef
 lp_build_cos(struct lp_build_context *bld,
              LLVMValueRef a)
 {
+   LLVMBuilderRef builder = bld->gallivm->builder;
+   struct gallivm_state *gallivm = bld->gallivm;
    struct lp_type int_type = lp_int_type(bld->type);
-   LLVMBuilderRef b = bld->builder;
-   LLVMTypeRef v4sf = LLVMVectorType(LLVMFloatType(), 4);
-   LLVMTypeRef v4si = LLVMVectorType(LLVMInt32Type(), 4);
+   LLVMBuilderRef b = builder;
+   LLVMTypeRef v4sf = LLVMVectorType(LLVMFloatTypeInContext(bld->gallivm->context), 4);
+   LLVMTypeRef v4si = LLVMVectorType(LLVMInt32TypeInContext(bld->gallivm->context), 4);
 
    /*
     *  take the absolute value,
     *  x = _mm_and_ps(x, *(v4sf*)_ps_inv_sign_mask);
     */
 
-   LLVMValueRef inv_sig_mask = lp_build_const_v4si(~0x80000000);
+   LLVMValueRef inv_sig_mask = lp_build_const_v4si(bld->gallivm, ~0x80000000);
    LLVMValueRef a_v4si = LLVMBuildBitCast(b, a, v4si, "a_v4si");
 
    LLVMValueRef absi = LLVMBuildAnd(b, a_v4si, inv_sig_mask, "absi");
@@ -1879,7 +1930,7 @@ lp_build_cos(struct lp_build_context *bld,
     * y = _mm_mul_ps(x, *(v4sf*)_ps_cephes_FOPI);
     */
    
-   LLVMValueRef FOPi = lp_build_const_v4sf(1.27323954473516);
+   LLVMValueRef FOPi = lp_build_const_v4sf(gallivm, 1.27323954473516);
    LLVMValueRef scale_y = LLVMBuildFMul(b, x_abs, FOPi, "scale_y");
 
    /*
@@ -1894,12 +1945,12 @@ lp_build_cos(struct lp_build_context *bld,
     * emm2 = _mm_add_epi32(emm2, *(v4si*)_pi32_1);
     */
 
-   LLVMValueRef all_one = lp_build_const_v4si(1);
+   LLVMValueRef all_one = lp_build_const_v4si(bld->gallivm, 1);
    LLVMValueRef emm2_add =  LLVMBuildAdd(b, emm2_i, all_one, "emm2_add");
    /*
     * emm2 = _mm_and_si128(emm2, *(v4si*)_pi32_inv1);
     */
-   LLVMValueRef inv_one = lp_build_const_v4si(~1);
+   LLVMValueRef inv_one = lp_build_const_v4si(bld->gallivm, ~1);
    LLVMValueRef emm2_and =  LLVMBuildAnd(b, emm2_add, inv_one, "emm2_and");
 
    /*
@@ -1911,22 +1962,22 @@ lp_build_cos(struct lp_build_context *bld,
    /*
     * emm2 = _mm_sub_epi32(emm2, *(v4si*)_pi32_2);
     */
-   LLVMValueRef const_2 = lp_build_const_v4si(2);
+   LLVMValueRef const_2 = lp_build_const_v4si(bld->gallivm, 2);
    LLVMValueRef emm2_2 = LLVMBuildSub(b, emm2_and, const_2, "emm2_2");
 
 
    /* get the swap sign flag
     * emm0 = _mm_andnot_si128(emm2, *(v4si*)_pi32_4);
     */
-   LLVMValueRef inv = lp_build_const_v4si(~0);
+   LLVMValueRef inv = lp_build_const_v4si(bld->gallivm, ~0);
    LLVMValueRef emm0_not = LLVMBuildXor(b, emm2_2, inv, "emm0_not");
-   LLVMValueRef pi32_4 = lp_build_const_v4si(4);
+   LLVMValueRef pi32_4 = lp_build_const_v4si(bld->gallivm, 4);
    LLVMValueRef emm0_and =  LLVMBuildAnd(b, emm0_not, pi32_4, "emm0_and");
    
    /*
     * emm2 = _mm_slli_epi32(emm0, 29);
     */  
-   LLVMValueRef const_29 = lp_build_const_v4si(29);
+   LLVMValueRef const_29 = lp_build_const_v4si(bld->gallivm, 29);
    LLVMValueRef sign_bit = LLVMBuildShl(b, emm0_and, const_29, "sign_bit");
 
    /*
@@ -1939,19 +1990,20 @@ lp_build_cos(struct lp_build_context *bld,
     * emm2 = _mm_cmpeq_epi32(emm2, _mm_setzero_si128());
     */
 
-   LLVMValueRef pi32_2 = lp_build_const_v4si(2);
+   LLVMValueRef pi32_2 = lp_build_const_v4si(bld->gallivm, 2);
    LLVMValueRef emm2_3 =  LLVMBuildAnd(b, emm2_2, pi32_2, "emm2_3");
-   LLVMValueRef poly_mask = lp_build_compare(b, int_type, PIPE_FUNC_EQUAL,
-   				             emm2_3, lp_build_const_v4si(0));
+   LLVMValueRef poly_mask = lp_build_compare(bld->gallivm,
+                                             int_type, PIPE_FUNC_EQUAL,
+   				             emm2_3, lp_build_const_v4si(bld->gallivm, 0));
 
    /*
     * _PS_CONST(minus_cephes_DP1, -0.78515625);
     * _PS_CONST(minus_cephes_DP2, -2.4187564849853515625e-4);
     * _PS_CONST(minus_cephes_DP3, -3.77489497744594108e-8);
     */
-   LLVMValueRef DP1 = lp_build_const_v4sf(-0.78515625);
-   LLVMValueRef DP2 = lp_build_const_v4sf(-2.4187564849853515625e-4);
-   LLVMValueRef DP3 = lp_build_const_v4sf(-3.77489497744594108e-8);
+   LLVMValueRef DP1 = lp_build_const_v4sf(gallivm, -0.78515625);
+   LLVMValueRef DP2 = lp_build_const_v4sf(gallivm, -2.4187564849853515625e-4);
+   LLVMValueRef DP3 = lp_build_const_v4sf(gallivm, -3.77489497744594108e-8);
 
    /*
     * The magic pass: "Extended precision modular arithmetic" 
@@ -1986,9 +2038,9 @@ lp_build_cos(struct lp_build_context *bld,
     * _PS_CONST(coscof_p1, -1.388731625493765E-003);
     * _PS_CONST(coscof_p2,  4.166664568298827E-002);
     */
-   LLVMValueRef coscof_p0 = lp_build_const_v4sf(2.443315711809948E-005);
-   LLVMValueRef coscof_p1 = lp_build_const_v4sf(-1.388731625493765E-003);
-   LLVMValueRef coscof_p2 = lp_build_const_v4sf(4.166664568298827E-002);
+   LLVMValueRef coscof_p0 = lp_build_const_v4sf(gallivm, 2.443315711809948E-005);
+   LLVMValueRef coscof_p1 = lp_build_const_v4sf(gallivm, -1.388731625493765E-003);
+   LLVMValueRef coscof_p2 = lp_build_const_v4sf(gallivm, 4.166664568298827E-002);
 
    /*
     * y = *(v4sf*)_ps_coscof_p0;
@@ -2007,10 +2059,10 @@ lp_build_cos(struct lp_build_context *bld,
     * y = _mm_sub_ps(y, tmp);
     * y = _mm_add_ps(y, *(v4sf*)_ps_1);
     */ 
-   LLVMValueRef half = lp_build_const_v4sf(0.5);
+   LLVMValueRef half = lp_build_const_v4sf(gallivm, 0.5);
    LLVMValueRef tmp = LLVMBuildFMul(b, z, half, "tmp");
    LLVMValueRef y_9 = LLVMBuildFSub(b, y_8, tmp, "y_8");
-   LLVMValueRef one = lp_build_const_v4sf(1.0);
+   LLVMValueRef one = lp_build_const_v4sf(gallivm, 1.0);
    LLVMValueRef y_10 = LLVMBuildFAdd(b, y_9, one, "y_9");
 
    /*
@@ -2018,9 +2070,9 @@ lp_build_cos(struct lp_build_context *bld,
     * _PS_CONST(sincof_p1,  8.3321608736E-3);
     * _PS_CONST(sincof_p2, -1.6666654611E-1);
     */
-   LLVMValueRef sincof_p0 = lp_build_const_v4sf(-1.9515295891E-4);
-   LLVMValueRef sincof_p1 = lp_build_const_v4sf(8.3321608736E-3);
-   LLVMValueRef sincof_p2 = lp_build_const_v4sf(-1.6666654611E-1);
+   LLVMValueRef sincof_p0 = lp_build_const_v4sf(gallivm, -1.9515295891E-4);
+   LLVMValueRef sincof_p1 = lp_build_const_v4sf(gallivm, 8.3321608736E-3);
+   LLVMValueRef sincof_p2 = lp_build_const_v4sf(gallivm, -1.6666654611E-1);
 
    /*
     * Evaluate the second polynom  (Pi/4 <= x <= 0)
@@ -2094,7 +2146,8 @@ lp_build_exp(struct lp_build_context *bld,
              LLVMValueRef x)
 {
    /* log2(e) = 1/log(2) */
-   LLVMValueRef log2e = lp_build_const_vec(bld->type, 1.4426950408889634);
+   LLVMValueRef log2e = lp_build_const_vec(bld->gallivm, bld->type,
+                                           1.4426950408889634);
 
    assert(lp_check_value(bld->type, x));
 
@@ -2110,7 +2163,8 @@ lp_build_log(struct lp_build_context *bld,
              LLVMValueRef x)
 {
    /* log(2) */
-   LLVMValueRef log2 = lp_build_const_vec(bld->type, 0.69314718055994529);
+   LLVMValueRef log2 = lp_build_const_vec(bld->gallivm, bld->type,
+                                          0.69314718055994529);
 
    assert(lp_check_value(bld->type, x));
 
@@ -2144,7 +2198,7 @@ lp_build_polynomial(struct lp_build_context *bld,
    for (i = num_coeffs; i--; ) {
       LLVMValueRef coeff;
 
-      coeff = lp_build_const_vec(type, coeffs[i]);
+      coeff = lp_build_const_vec(bld->gallivm, type, coeffs[i]);
 
       if(res)
          res = lp_build_add(bld, coeff, lp_build_mul(bld, x, res));
@@ -2198,9 +2252,10 @@ lp_build_exp2_approx(struct lp_build_context *bld,
                      LLVMValueRef *p_frac_part,
                      LLVMValueRef *p_exp2)
 {
+   LLVMBuilderRef builder = bld->gallivm->builder;
    const struct lp_type type = bld->type;
-   LLVMTypeRef vec_type = lp_build_vec_type(type);
-   LLVMTypeRef int_vec_type = lp_build_int_vec_type(type);
+   LLVMTypeRef vec_type = lp_build_vec_type(bld->gallivm, type);
+   LLVMTypeRef int_vec_type = lp_build_int_vec_type(bld->gallivm, type);
    LLVMValueRef ipart = NULL;
    LLVMValueRef fpart = NULL;
    LLVMValueRef expipart = NULL;
@@ -2219,29 +2274,31 @@ lp_build_exp2_approx(struct lp_build_context *bld,
 
       assert(type.floating && type.width == 32);
 
-      x = lp_build_min(bld, x, lp_build_const_vec(type,  129.0));
-      x = lp_build_max(bld, x, lp_build_const_vec(type, -126.99999));
+      x = lp_build_min(bld, x, lp_build_const_vec(bld->gallivm, type,  129.0));
+      x = lp_build_max(bld, x, lp_build_const_vec(bld->gallivm, type, -126.99999));
 
       /* ipart = floor(x) */
       ipart = lp_build_floor(bld, x);
 
       /* fpart = x - ipart */
-      fpart = LLVMBuildFSub(bld->builder, x, ipart, "");
+      fpart = LLVMBuildFSub(builder, x, ipart, "");
    }
 
    if(p_exp2_int_part || p_exp2) {
       /* expipart = (float) (1 << ipart) */
-      ipart = LLVMBuildFPToSI(bld->builder, ipart, int_vec_type, "");
-      expipart = LLVMBuildAdd(bld->builder, ipart, lp_build_const_int_vec(type, 127), "");
-      expipart = LLVMBuildShl(bld->builder, expipart, lp_build_const_int_vec(type, 23), "");
-      expipart = LLVMBuildBitCast(bld->builder, expipart, vec_type, "");
+      ipart = LLVMBuildFPToSI(builder, ipart, int_vec_type, "");
+      expipart = LLVMBuildAdd(builder, ipart,
+                              lp_build_const_int_vec(bld->gallivm, type, 127), "");
+      expipart = LLVMBuildShl(builder, expipart,
+                              lp_build_const_int_vec(bld->gallivm, type, 23), "");
+      expipart = LLVMBuildBitCast(builder, expipart, vec_type, "");
    }
 
    if(p_exp2) {
       expfpart = lp_build_polynomial(bld, fpart, lp_build_exp2_polynomial,
                                      Elements(lp_build_exp2_polynomial));
 
-      res = LLVMBuildFMul(bld->builder, expipart, expfpart, "");
+      res = LLVMBuildFMul(builder, expipart, expfpart, "");
    }
 
    if(p_exp2_int_part)
@@ -2279,6 +2336,7 @@ lp_build_extract_exponent(struct lp_build_context *bld,
                           LLVMValueRef x,
                           int bias)
 {
+   LLVMBuilderRef builder = bld->gallivm->builder;
    const struct lp_type type = bld->type;
    unsigned mantissa = lp_mantissa(type);
    LLVMValueRef res;
@@ -2287,11 +2345,14 @@ lp_build_extract_exponent(struct lp_build_context *bld,
 
    assert(lp_check_value(bld->type, x));
 
-   x = LLVMBuildBitCast(bld->builder, x, bld->int_vec_type, "");
+   x = LLVMBuildBitCast(builder, x, bld->int_vec_type, "");
 
-   res = LLVMBuildLShr(bld->builder, x, lp_build_const_int_vec(type, mantissa), "");
-   res = LLVMBuildAnd(bld->builder, res, lp_build_const_int_vec(type, 255), "");
-   res = LLVMBuildSub(bld->builder, res, lp_build_const_int_vec(type, 127 - bias), "");
+   res = LLVMBuildLShr(builder, x,
+                       lp_build_const_int_vec(bld->gallivm, type, mantissa), "");
+   res = LLVMBuildAnd(builder, res,
+                      lp_build_const_int_vec(bld->gallivm, type, 255), "");
+   res = LLVMBuildSub(builder, res,
+                      lp_build_const_int_vec(bld->gallivm, type, 127 - bias), "");
 
    return res;
 }
@@ -2308,9 +2369,11 @@ LLVMValueRef
 lp_build_extract_mantissa(struct lp_build_context *bld,
                           LLVMValueRef x)
 {
+   LLVMBuilderRef builder = bld->gallivm->builder;
    const struct lp_type type = bld->type;
    unsigned mantissa = lp_mantissa(type);
-   LLVMValueRef mantmask = lp_build_const_int_vec(type, (1ULL << mantissa) - 1);
+   LLVMValueRef mantmask = lp_build_const_int_vec(bld->gallivm, type,
+                                                  (1ULL << mantissa) - 1);
    LLVMValueRef one = LLVMConstBitCast(bld->one, bld->int_vec_type);
    LLVMValueRef res;
 
@@ -2318,12 +2381,12 @@ lp_build_extract_mantissa(struct lp_build_context *bld,
 
    assert(type.floating);
 
-   x = LLVMBuildBitCast(bld->builder, x, bld->int_vec_type, "");
+   x = LLVMBuildBitCast(builder, x, bld->int_vec_type, "");
 
    /* res = x / 2**ipart */
-   res = LLVMBuildAnd(bld->builder, x, mantmask, "");
-   res = LLVMBuildOr(bld->builder, res, one, "");
-   res = LLVMBuildBitCast(bld->builder, res, bld->vec_type, "");
+   res = LLVMBuildAnd(builder, x, mantmask, "");
+   res = LLVMBuildOr(builder, res, one, "");
+   res = LLVMBuildBitCast(builder, res, bld->vec_type, "");
 
    return res;
 }
@@ -2374,12 +2437,13 @@ lp_build_log2_approx(struct lp_build_context *bld,
                      LLVMValueRef *p_floor_log2,
                      LLVMValueRef *p_log2)
 {
+   LLVMBuilderRef builder = bld->gallivm->builder;
    const struct lp_type type = bld->type;
-   LLVMTypeRef vec_type = lp_build_vec_type(type);
-   LLVMTypeRef int_vec_type = lp_build_int_vec_type(type);
+   LLVMTypeRef vec_type = lp_build_vec_type(bld->gallivm, type);
+   LLVMTypeRef int_vec_type = lp_build_int_vec_type(bld->gallivm, type);
 
-   LLVMValueRef expmask = lp_build_const_int_vec(type, 0x7f800000);
-   LLVMValueRef mantmask = lp_build_const_int_vec(type, 0x007fffff);
+   LLVMValueRef expmask = lp_build_const_int_vec(bld->gallivm, type, 0x7f800000);
+   LLVMValueRef mantmask = lp_build_const_int_vec(bld->gallivm, type, 0x007fffff);
    LLVMValueRef one = LLVMConstBitCast(bld->one, int_vec_type);
 
    LLVMValueRef i = NULL;
@@ -2401,35 +2465,35 @@ lp_build_log2_approx(struct lp_build_context *bld,
 
       assert(type.floating && type.width == 32);
 
-      i = LLVMBuildBitCast(bld->builder, x, int_vec_type, "");
+      i = LLVMBuildBitCast(builder, x, int_vec_type, "");
 
       /* exp = (float) exponent(x) */
-      exp = LLVMBuildAnd(bld->builder, i, expmask, "");
+      exp = LLVMBuildAnd(builder, i, expmask, "");
    }
 
    if(p_floor_log2 || p_log2) {
-      logexp = LLVMBuildLShr(bld->builder, exp, lp_build_const_int_vec(type, 23), "");
-      logexp = LLVMBuildSub(bld->builder, logexp, lp_build_const_int_vec(type, 127), "");
-      logexp = LLVMBuildSIToFP(bld->builder, logexp, vec_type, "");
+      logexp = LLVMBuildLShr(builder, exp, lp_build_const_int_vec(bld->gallivm, type, 23), "");
+      logexp = LLVMBuildSub(builder, logexp, lp_build_const_int_vec(bld->gallivm, type, 127), "");
+      logexp = LLVMBuildSIToFP(builder, logexp, vec_type, "");
    }
 
    if(p_log2) {
       /* mant = (float) mantissa(x) */
-      mant = LLVMBuildAnd(bld->builder, i, mantmask, "");
-      mant = LLVMBuildOr(bld->builder, mant, one, "");
-      mant = LLVMBuildBitCast(bld->builder, mant, vec_type, "");
+      mant = LLVMBuildAnd(builder, i, mantmask, "");
+      mant = LLVMBuildOr(builder, mant, one, "");
+      mant = LLVMBuildBitCast(builder, mant, vec_type, "");
 
       logmant = lp_build_polynomial(bld, mant, lp_build_log2_polynomial,
                                     Elements(lp_build_log2_polynomial));
 
       /* This effectively increases the polynomial degree by one, but ensures that log2(1) == 0*/
-      logmant = LLVMBuildFMul(bld->builder, logmant, LLVMBuildFSub(bld->builder, mant, bld->one, ""), "");
+      logmant = LLVMBuildFMul(builder, logmant, LLVMBuildFSub(builder, mant, bld->one, ""), "");
 
-      res = LLVMBuildFAdd(bld->builder, logmant, logexp, "");
+      res = LLVMBuildFAdd(builder, logmant, logexp, "");
    }
 
    if(p_exp) {
-      exp = LLVMBuildBitCast(bld->builder, exp, vec_type, "");
+      exp = LLVMBuildBitCast(builder, exp, vec_type, "");
       *p_exp = exp;
    }
 
@@ -2465,6 +2529,7 @@ LLVMValueRef
 lp_build_fast_log2(struct lp_build_context *bld,
                    LLVMValueRef x)
 {
+   LLVMBuilderRef builder = bld->gallivm->builder;
    LLVMValueRef ipart;
    LLVMValueRef fpart;
 
@@ -2474,13 +2539,13 @@ lp_build_fast_log2(struct lp_build_context *bld,
 
    /* ipart = floor(log2(x)) - 1 */
    ipart = lp_build_extract_exponent(bld, x, -1);
-   ipart = LLVMBuildSIToFP(bld->builder, ipart, bld->vec_type, "");
+   ipart = LLVMBuildSIToFP(builder, ipart, bld->vec_type, "");
 
    /* fpart = x / 2**ipart */
    fpart = lp_build_extract_mantissa(bld, x);
 
    /* ipart + fpart */
-   return LLVMBuildFAdd(bld->builder, ipart, fpart, "");
+   return LLVMBuildFAdd(builder, ipart, fpart, "");
 }
 
 
@@ -2493,7 +2558,8 @@ LLVMValueRef
 lp_build_ilog2(struct lp_build_context *bld,
                LLVMValueRef x)
 {
-   LLVMValueRef sqrt2 = lp_build_const_vec(bld->type, M_SQRT2);
+   LLVMBuilderRef builder = bld->gallivm->builder;
+   LLVMValueRef sqrt2 = lp_build_const_vec(bld->gallivm, bld->type, M_SQRT2);
    LLVMValueRef ipart;
 
    assert(bld->type.floating);
@@ -2501,7 +2567,7 @@ lp_build_ilog2(struct lp_build_context *bld,
    assert(lp_check_value(bld->type, x));
 
    /* x * 2^(0.5)   i.e., add 0.5 to the log2(x) */
-   x = LLVMBuildFMul(bld->builder, x, sqrt2, "");
+   x = LLVMBuildFMul(builder, x, sqrt2, "");
 
    /* ipart = floor(log2(x) + 0.5)  */
    ipart = lp_build_extract_exponent(bld, x, 0);
diff --git a/src/gallium/auxiliary/gallivm/lp_bld_assert.c b/src/gallium/auxiliary/gallivm/lp_bld_assert.c
index f2ebd868a8..9de5e8e7b5 100644
--- a/src/gallium/auxiliary/gallivm/lp_bld_assert.c
+++ b/src/gallium/auxiliary/gallivm/lp_bld_assert.c
@@ -56,20 +56,21 @@ lp_assert(int condition, const char *msg)
  * \param msg  a string to print if the assertion fails.
  */
 LLVMValueRef
-lp_build_assert(LLVMBuilderRef builder, LLVMValueRef condition,
+lp_build_assert(struct gallivm_state *gallivm,
+                LLVMValueRef condition,
                 const char *msg)
 {
-   LLVMModuleRef module;
+   LLVMBuilderRef builder = gallivm->builder;
+   LLVMContextRef context = gallivm->context;
+   LLVMModuleRef module = gallivm->module;
    LLVMTypeRef arg_types[2];
    LLVMValueRef msg_string, assert_func, params[2], r;
 
-   module = LLVMGetGlobalParent(LLVMGetBasicBlockParent(
-                            LLVMGetInsertBlock(builder)));
+   msg_string = lp_build_const_string_variable(module, context,
+                                               msg, strlen(msg) + 1);
 
-   msg_string = lp_build_const_string_variable(module, msg, strlen(msg) + 1);
-
-   arg_types[0] = LLVMInt32Type();
-   arg_types[1] = LLVMPointerType(LLVMInt8Type(), 0);
+   arg_types[0] = LLVMInt32TypeInContext(context);
+   arg_types[1] = LLVMPointerType(LLVMInt8TypeInContext(context), 0);
 
    /* lookup the lp_assert function */
    assert_func = LLVMGetNamedFunction(module, "lp_assert");
@@ -77,12 +78,12 @@ lp_build_assert(LLVMBuilderRef builder, LLVMValueRef condition,
    /* Create the assertion function if not found */
    if (!assert_func) {
       LLVMTypeRef func_type =
-         LLVMFunctionType(LLVMVoidType(), arg_types, 2, 0);
+         LLVMFunctionType(LLVMVoidTypeInContext(context), arg_types, 2, 0);
 
       assert_func = LLVMAddFunction(module, "lp_assert", func_type);
       LLVMSetFunctionCallConv(assert_func, LLVMCCallConv);
       LLVMSetLinkage(assert_func, LLVMExternalLinkage);
-      LLVMAddGlobalMapping(lp_build_engine, assert_func,
+      LLVMAddGlobalMapping(gallivm->engine, assert_func,
                            func_to_pointer((func_pointer)lp_assert));
    }
    assert(assert_func);
diff --git a/src/gallium/auxiliary/gallivm/lp_bld_assert.h b/src/gallium/auxiliary/gallivm/lp_bld_assert.h
index ddd879dc2c..1d2baab30a 100644
--- a/src/gallium/auxiliary/gallivm/lp_bld_assert.h
+++ b/src/gallium/auxiliary/gallivm/lp_bld_assert.h
@@ -30,10 +30,12 @@
 
 
 #include "lp_bld.h"
+#include "lp_bld_init.h"
 
 
 LLVMValueRef
-lp_build_assert(LLVMBuilderRef builder, LLVMValueRef condition,
+lp_build_assert(struct gallivm_state *gallivm,
+                LLVMValueRef condition,
                 const char *msg);
 
 
diff --git a/src/gallium/auxiliary/gallivm/lp_bld_bitarit.c b/src/gallium/auxiliary/gallivm/lp_bld_bitarit.c
index 706479b4d5..a9c57d682f 100644
--- a/src/gallium/auxiliary/gallivm/lp_bld_bitarit.c
+++ b/src/gallium/auxiliary/gallivm/lp_bld_bitarit.c
@@ -40,6 +40,7 @@
 LLVMValueRef
 lp_build_or(struct lp_build_context *bld, LLVMValueRef a, LLVMValueRef b)
 {
+   LLVMBuilderRef builder = bld->gallivm->builder;
    const struct lp_type type = bld->type;
    LLVMValueRef res;
 
@@ -48,14 +49,14 @@ lp_build_or(struct lp_build_context *bld, LLVMValueRef a, LLVMValueRef b)
 
    /* can't do bitwise ops on floating-point values */
    if (type.floating) {
-      a = LLVMBuildBitCast(bld->builder, a, bld->int_vec_type, "");
-      b = LLVMBuildBitCast(bld->builder, b, bld->int_vec_type, "");
+      a = LLVMBuildBitCast(builder, a, bld->int_vec_type, "");
+      b = LLVMBuildBitCast(builder, b, bld->int_vec_type, "");
    }
 
-   res = LLVMBuildOr(bld->builder, a, b, "");
+   res = LLVMBuildOr(builder, a, b, "");
 
    if (type.floating) {
-      res = LLVMBuildBitCast(bld->builder, res, bld->vec_type, "");
+      res = LLVMBuildBitCast(builder, res, bld->vec_type, "");
    }
 
    return res;
@@ -68,6 +69,7 @@ lp_build_or(struct lp_build_context *bld, LLVMValueRef a, LLVMValueRef b)
 LLVMValueRef
 lp_build_and(struct lp_build_context *bld, LLVMValueRef a, LLVMValueRef b)
 {
+   LLVMBuilderRef builder = bld->gallivm->builder;
    const struct lp_type type = bld->type;
    LLVMValueRef res;
 
@@ -76,14 +78,14 @@ lp_build_and(struct lp_build_context *bld, LLVMValueRef a, LLVMValueRef b)
 
    /* can't do bitwise ops on floating-point values */
    if (type.floating) {
-      a = LLVMBuildBitCast(bld->builder, a, bld->int_vec_type, "");
-      b = LLVMBuildBitCast(bld->builder, b, bld->int_vec_type, "");
+      a = LLVMBuildBitCast(builder, a, bld->int_vec_type, "");
+      b = LLVMBuildBitCast(builder, b, bld->int_vec_type, "");
    }
 
-   res = LLVMBuildAnd(bld->builder, a, b, "");
+   res = LLVMBuildAnd(builder, a, b, "");
 
    if (type.floating) {
-      res = LLVMBuildBitCast(bld->builder, res, bld->vec_type, "");
+      res = LLVMBuildBitCast(builder, res, bld->vec_type, "");
    }
 
    return res;
@@ -96,6 +98,7 @@ lp_build_and(struct lp_build_context *bld, LLVMValueRef a, LLVMValueRef b)
 LLVMValueRef
 lp_build_andnot(struct lp_build_context *bld, LLVMValueRef a, LLVMValueRef b)
 {
+   LLVMBuilderRef builder = bld->gallivm->builder;
    const struct lp_type type = bld->type;
    LLVMValueRef res;
 
@@ -104,15 +107,15 @@ lp_build_andnot(struct lp_build_context *bld, LLVMValueRef a, LLVMValueRef b)
 
    /* can't do bitwise ops on floating-point values */
    if (type.floating) {
-      a = LLVMBuildBitCast(bld->builder, a, bld->int_vec_type, "");
-      b = LLVMBuildBitCast(bld->builder, b, bld->int_vec_type, "");
+      a = LLVMBuildBitCast(builder, a, bld->int_vec_type, "");
+      b = LLVMBuildBitCast(builder, b, bld->int_vec_type, "");
    }
 
-   res = LLVMBuildNot(bld->builder, b, "");
-   res = LLVMBuildAnd(bld->builder, a, res, "");
+   res = LLVMBuildNot(builder, b, "");
+   res = LLVMBuildAnd(builder, a, res, "");
 
    if (type.floating) {
-      res = LLVMBuildBitCast(bld->builder, res, bld->vec_type, "");
+      res = LLVMBuildBitCast(builder, res, bld->vec_type, "");
    }
 
    return res;
@@ -125,6 +128,7 @@ lp_build_andnot(struct lp_build_context *bld, LLVMValueRef a, LLVMValueRef b)
 LLVMValueRef
 lp_build_shl(struct lp_build_context *bld, LLVMValueRef a, LLVMValueRef b)
 {
+   LLVMBuilderRef builder = bld->gallivm->builder;
    const struct lp_type type = bld->type;
    LLVMValueRef res;
 
@@ -133,7 +137,7 @@ lp_build_shl(struct lp_build_context *bld, LLVMValueRef a, LLVMValueRef b)
    assert(lp_check_value(type, a));
    assert(lp_check_value(type, b));
 
-   res = LLVMBuildShl(bld->builder, a, b, "");
+   res = LLVMBuildShl(builder, a, b, "");
 
    return res;
 }
@@ -145,6 +149,7 @@ lp_build_shl(struct lp_build_context *bld, LLVMValueRef a, LLVMValueRef b)
 LLVMValueRef
 lp_build_shr(struct lp_build_context *bld, LLVMValueRef a, LLVMValueRef b)
 {
+   LLVMBuilderRef builder = bld->gallivm->builder;
    const struct lp_type type = bld->type;
    LLVMValueRef res;
 
@@ -154,9 +159,9 @@ lp_build_shr(struct lp_build_context *bld, LLVMValueRef a, LLVMValueRef b)
    assert(lp_check_value(type, b));
 
    if (type.sign) {
-      res = LLVMBuildAShr(bld->builder, a, b, "");
+      res = LLVMBuildAShr(builder, a, b, "");
    } else {
-      res = LLVMBuildLShr(bld->builder, a, b, "");
+      res = LLVMBuildLShr(builder, a, b, "");
    }
 
    return res;
@@ -169,7 +174,7 @@ lp_build_shr(struct lp_build_context *bld, LLVMValueRef a, LLVMValueRef b)
 LLVMValueRef
 lp_build_shl_imm(struct lp_build_context *bld, LLVMValueRef a, unsigned imm)
 {
-   LLVMValueRef b = lp_build_const_int_vec(bld->type, imm);
+   LLVMValueRef b = lp_build_const_int_vec(bld->gallivm, bld->type, imm);
    assert(imm <= bld->type.width);
    return lp_build_shl(bld, a, b);
 }
@@ -181,7 +186,7 @@ lp_build_shl_imm(struct lp_build_context *bld, LLVMValueRef a, unsigned imm)
 LLVMValueRef
 lp_build_shr_imm(struct lp_build_context *bld, LLVMValueRef a, unsigned imm)
 {
-   LLVMValueRef b = lp_build_const_int_vec(bld->type, imm);
+   LLVMValueRef b = lp_build_const_int_vec(bld->gallivm, bld->type, imm);
    assert(imm <= bld->type.width);
    return lp_build_shr(bld, a, b);
 }
diff --git a/src/gallium/auxiliary/gallivm/lp_bld_const.c b/src/gallium/auxiliary/gallivm/lp_bld_const.c
index dd839c0bea..6d8b7c26fc 100644
--- a/src/gallium/auxiliary/gallivm/lp_bld_const.c
+++ b/src/gallium/auxiliary/gallivm/lp_bld_const.c
@@ -39,6 +39,7 @@
 
 #include "lp_bld_type.h"
 #include "lp_bld_const.h"
+#include "lp_bld_init.h"
 
 
 unsigned
@@ -211,31 +212,31 @@ lp_const_eps(struct lp_type type)
 
 
 LLVMValueRef
-lp_build_undef(struct lp_type type)
+lp_build_undef(struct gallivm_state *gallivm, struct lp_type type)
 {
-   LLVMTypeRef vec_type = lp_build_vec_type(type);
+   LLVMTypeRef vec_type = lp_build_vec_type(gallivm, type);
    return LLVMGetUndef(vec_type);
 }
                
 
 LLVMValueRef
-lp_build_zero(struct lp_type type)
+lp_build_zero(struct gallivm_state *gallivm, struct lp_type type)
 {
    if (type.length == 1) {
       if (type.floating)
-         return LLVMConstReal(LLVMFloatType(), 0.0);
+         return lp_build_const_float(gallivm, 0.0);
       else
-         return LLVMConstInt(LLVMIntType(type.width), 0, 0);
+         return LLVMConstInt(LLVMIntTypeInContext(gallivm->context, type.width), 0, 0);
    }
    else {
-      LLVMTypeRef vec_type = lp_build_vec_type(type);
+      LLVMTypeRef vec_type = lp_build_vec_type(gallivm, type);
       return LLVMConstNull(vec_type);
    }
 }
                
 
 LLVMValueRef
-lp_build_one(struct lp_type type)
+lp_build_one(struct gallivm_state *gallivm, struct lp_type type)
 {
    LLVMTypeRef elem_type;
    LLVMValueRef elems[LP_MAX_VECTOR_LENGTH];
@@ -243,7 +244,7 @@ lp_build_one(struct lp_type type)
 
    assert(type.length <= LP_MAX_VECTOR_LENGTH);
 
-   elem_type = lp_build_elem_type(type);
+   elem_type = lp_build_elem_type(gallivm, type);
 
    if(type.floating)
       elems[0] = LLVMConstReal(elem_type, 1.0);
@@ -283,10 +284,11 @@ lp_build_one(struct lp_type type)
  * Build constant-valued element from a scalar value.
  */
 LLVMValueRef
-lp_build_const_elem(struct lp_type type,
+lp_build_const_elem(struct gallivm_state *gallivm,
+                    struct lp_type type,
                     double val)
 {
-   LLVMTypeRef elem_type = lp_build_elem_type(type);
+   LLVMTypeRef elem_type = lp_build_elem_type(gallivm, type);
    LLVMValueRef elem;
 
    if(type.floating) {
@@ -306,15 +308,15 @@ lp_build_const_elem(struct lp_type type,
  * Build constant-valued vector from a scalar value.
  */
 LLVMValueRef
-lp_build_const_vec(struct lp_type type,
+lp_build_const_vec(struct gallivm_state *gallivm, struct lp_type type,
                    double val)
 {
    if (type.length == 1) {
-      return lp_build_const_elem(type, val);
+      return lp_build_const_elem(gallivm, type, val);
    } else {
       LLVMValueRef elems[LP_MAX_VECTOR_LENGTH];
       unsigned i;
-      elems[0] = lp_build_const_elem(type, val);
+      elems[0] = lp_build_const_elem(gallivm, type, val);
       for(i = 1; i < type.length; ++i)
          elems[i] = elems[0];
       return LLVMConstVector(elems, type.length);
@@ -323,10 +325,10 @@ lp_build_const_vec(struct lp_type type,
 
 
 LLVMValueRef
-lp_build_const_int_vec(struct lp_type type,
-                          long long val)
+lp_build_const_int_vec(struct gallivm_state *gallivm, struct lp_type type,
+                       long long val)
 {
-   LLVMTypeRef elem_type = lp_build_int_elem_type(type);
+   LLVMTypeRef elem_type = lp_build_int_elem_type(gallivm, type);
    LLVMValueRef elems[LP_MAX_VECTOR_LENGTH];
    unsigned i;
 
@@ -343,7 +345,8 @@ lp_build_const_int_vec(struct lp_type type,
 
 
 LLVMValueRef
-lp_build_const_aos(struct lp_type type, 
+lp_build_const_aos(struct gallivm_state *gallivm,
+                   struct lp_type type, 
                    double r, double g, double b, double a, 
                    const unsigned char *swizzle)
 {
@@ -355,7 +358,7 @@ lp_build_const_aos(struct lp_type type,
    assert(type.length % 4 == 0);
    assert(type.length <= LP_MAX_VECTOR_LENGTH);
 
-   elem_type = lp_build_elem_type(type);
+   elem_type = lp_build_elem_type(gallivm, type);
 
    if(swizzle == NULL)
       swizzle = default_swizzle;
@@ -386,10 +389,11 @@ lp_build_const_aos(struct lp_type type,
  * @param mask TGSI_WRITEMASK_xxx
  */
 LLVMValueRef
-lp_build_const_mask_aos(struct lp_type type,
+lp_build_const_mask_aos(struct gallivm_state *gallivm,
+                        struct lp_type type,
                         unsigned mask)
 {
-   LLVMTypeRef elem_type = LLVMIntType(type.width);
+   LLVMTypeRef elem_type = LLVMIntTypeInContext(gallivm->context, type.width);
    LLVMValueRef masks[LP_MAX_VECTOR_LENGTH];
    unsigned i, j;
 
diff --git a/src/gallium/auxiliary/gallivm/lp_bld_const.h b/src/gallium/auxiliary/gallivm/lp_bld_const.h
index 6b1fc590c1..680211fbbd 100644
--- a/src/gallium/auxiliary/gallivm/lp_bld_const.h
+++ b/src/gallium/auxiliary/gallivm/lp_bld_const.h
@@ -39,6 +39,7 @@
 
 #include "pipe/p_compiler.h"
 #include "gallivm/lp_bld.h"
+#include "gallivm/lp_bld_init.h"
 
 
 
@@ -73,44 +74,71 @@ lp_const_eps(struct lp_type type);
 
 
 LLVMValueRef
-lp_build_undef(struct lp_type type);
+lp_build_undef(struct gallivm_state *gallivm, struct lp_type type);
 
 
 LLVMValueRef
-lp_build_zero(struct lp_type type);
+lp_build_zero(struct gallivm_state *gallivm, struct lp_type type);
 
 
 LLVMValueRef
-lp_build_one(struct lp_type type);
+lp_build_one(struct gallivm_state *gallivm, struct lp_type type);
 
 
 LLVMValueRef
-lp_build_const_elem(struct lp_type type,
+lp_build_const_elem(struct gallivm_state *gallivm, struct lp_type type,
                     double val);
 
 LLVMValueRef
-lp_build_const_vec(struct lp_type type, double val);
+lp_build_const_vec(struct gallivm_state *gallivm, struct lp_type type,
+                   double val);
 
 
 LLVMValueRef
-lp_build_const_int_vec(struct lp_type type, long long val);
+lp_build_const_int_vec(struct gallivm_state *gallivm,
+                       struct lp_type type, long long val);
 
 
 LLVMValueRef
-lp_build_const_aos(struct lp_type type, 
+lp_build_const_aos(struct gallivm_state *gallivm, struct lp_type type, 
                    double r, double g, double b, double a, 
                    const unsigned char *swizzle);
 
 
 LLVMValueRef
-lp_build_const_mask_aos(struct lp_type type,
+lp_build_const_mask_aos(struct gallivm_state *gallivm,
+                        struct lp_type type,
                         unsigned mask);
 
 
 static INLINE LLVMValueRef
-lp_build_const_int32(int i)
+lp_build_const_int32(struct gallivm_state *gallivm, int i)
 {
-   return LLVMConstInt(LLVMInt32Type(), i, 0);
+   return LLVMConstInt(LLVMInt32TypeInContext(gallivm->context), i, 0);
+}
+
+
+static INLINE LLVMValueRef
+lp_build_const_float(struct gallivm_state *gallivm, float x)
+{
+   return LLVMConstReal(LLVMFloatTypeInContext(gallivm->context), x);
+}
+
+
+/** Return constant-valued pointer to int */
+static INLINE LLVMValueRef
+lp_build_const_int_pointer(struct gallivm_state *gallivm, const void *ptr)
+{
+   LLVMTypeRef int_type;
+   LLVMValueRef v;
+
+   /* int type large enough to hold a pointer */
+   int_type = LLVMIntTypeInContext(gallivm->context, 8 * sizeof(void *));
+   v = LLVMConstInt(int_type, (unsigned long long) ptr, 0);
+   v = LLVMBuildIntToPtr(gallivm->builder, v,
+                         LLVMPointerType(int_type, 0),
+                         "cast int to ptr");
+   return v;
 }
 
 
diff --git a/src/gallium/auxiliary/gallivm/lp_bld_conv.c b/src/gallium/auxiliary/gallivm/lp_bld_conv.c
index 6967dd2622..c43ee8ac63 100644
--- a/src/gallium/auxiliary/gallivm/lp_bld_conv.c
+++ b/src/gallium/auxiliary/gallivm/lp_bld_conv.c
@@ -89,12 +89,13 @@
  * return { i32, i32, i32, i32 } where each value is in [0, 2^dst_width-1].
  */
 LLVMValueRef
-lp_build_clamped_float_to_unsigned_norm(LLVMBuilderRef builder,
+lp_build_clamped_float_to_unsigned_norm(struct gallivm_state *gallivm,
                                         struct lp_type src_type,
                                         unsigned dst_width,
                                         LLVMValueRef src)
 {
-   LLVMTypeRef int_vec_type = lp_build_int_vec_type(src_type);
+   LLVMBuilderRef builder = gallivm->builder;
+   LLVMTypeRef int_vec_type = lp_build_int_vec_type(gallivm, src_type);
    LLVMValueRef res;
    unsigned mantissa;
 
@@ -122,10 +123,11 @@ lp_build_clamped_float_to_unsigned_norm(LLVMBuilderRef builder,
       scale = (double)mask/ubound;
       bias = (double)(1ULL << (mantissa - dst_width));
 
-      res = LLVMBuildFMul(builder, src, lp_build_const_vec(src_type, scale), "");
-      res = LLVMBuildFAdd(builder, res, lp_build_const_vec(src_type, bias), "");
+      res = LLVMBuildFMul(builder, src, lp_build_const_vec(gallivm, src_type, scale), "");
+      res = LLVMBuildFAdd(builder, res, lp_build_const_vec(gallivm, src_type, bias), "");
       res = LLVMBuildBitCast(builder, res, int_vec_type, "");
-      res = LLVMBuildAnd(builder, res, lp_build_const_int_vec(src_type, mask), "");
+      res = LLVMBuildAnd(builder, res,
+                         lp_build_const_int_vec(gallivm, src_type, mask), "");
    }
    else if (dst_width == (mantissa + 1)) {
       /*
@@ -138,7 +140,8 @@ lp_build_clamped_float_to_unsigned_norm(LLVMBuilderRef builder,
 
       scale = (double)((1ULL << dst_width) - 1);
 
-      res = LLVMBuildFMul(builder, src, lp_build_const_vec(src_type, scale), "");
+      res = LLVMBuildFMul(builder, src,
+                          lp_build_const_vec(gallivm, src_type, scale), "");
       res = LLVMBuildFPToSI(builder, res, int_vec_type, "");
    }
    else {
@@ -166,7 +169,8 @@ lp_build_clamped_float_to_unsigned_norm(LLVMBuilderRef builder,
       LLVMValueRef lshifted;
       LLVMValueRef rshifted;
 
-      res = LLVMBuildFMul(builder, src, lp_build_const_vec(src_type, scale), "");
+      res = LLVMBuildFMul(builder, src,
+                          lp_build_const_vec(gallivm, src_type, scale), "");
       res = LLVMBuildFPToSI(builder, res, int_vec_type, "");
 
       /*
@@ -177,7 +181,8 @@ lp_build_clamped_float_to_unsigned_norm(LLVMBuilderRef builder,
        */
       if (lshift) {
          lshifted = LLVMBuildShl(builder, res,
-                                 lp_build_const_int_vec(src_type, lshift), "");
+                                 lp_build_const_int_vec(gallivm, src_type,
+                                                        lshift), "");
       } else {
          lshifted = res;
       }
@@ -186,7 +191,8 @@ lp_build_clamped_float_to_unsigned_norm(LLVMBuilderRef builder,
        * Align the most significant bit to the right.
        */
       rshifted =  LLVMBuildAShr(builder, res,
-                                lp_build_const_int_vec(src_type, rshift), "");
+                                lp_build_const_int_vec(gallivm, src_type, rshift),
+                                "");
 
       /*
        * Subtract the MSB to the LSB, therefore re-scaling from
@@ -206,13 +212,14 @@ lp_build_clamped_float_to_unsigned_norm(LLVMBuilderRef builder,
  * return {float, float, float, float} with values in range [0, 1].
  */
 LLVMValueRef
-lp_build_unsigned_norm_to_float(LLVMBuilderRef builder,
+lp_build_unsigned_norm_to_float(struct gallivm_state *gallivm,
                                 unsigned src_width,
                                 struct lp_type dst_type,
                                 LLVMValueRef src)
 {
-   LLVMTypeRef vec_type = lp_build_vec_type(dst_type);
-   LLVMTypeRef int_vec_type = lp_build_int_vec_type(dst_type);
+   LLVMBuilderRef builder = gallivm->builder;
+   LLVMTypeRef vec_type = lp_build_vec_type(gallivm, dst_type);
+   LLVMTypeRef int_vec_type = lp_build_int_vec_type(gallivm, dst_type);
    LLVMValueRef bias_;
    LLVMValueRef res;
    unsigned mantissa;
@@ -230,7 +237,8 @@ lp_build_unsigned_norm_to_float(LLVMBuilderRef builder,
    if (src_width == 8) {
       scale = 1.0/255.0;
       res = LLVMBuildSIToFP(builder, src, vec_type, "");
-      res = LLVMBuildFMul(builder, res, lp_build_const_vec(dst_type, scale), "");
+      res = LLVMBuildFMul(builder, res,
+                          lp_build_const_vec(gallivm, dst_type, scale), "");
       return res;
    }
 
@@ -247,10 +255,11 @@ lp_build_unsigned_norm_to_float(LLVMBuilderRef builder,
 
    if(src_width > mantissa) {
       int shift = src_width - mantissa;
-      res = LLVMBuildLShr(builder, res, lp_build_const_int_vec(dst_type, shift), "");
+      res = LLVMBuildLShr(builder, res,
+                          lp_build_const_int_vec(gallivm, dst_type, shift), "");
    }
 
-   bias_ = lp_build_const_vec(dst_type, bias);
+   bias_ = lp_build_const_vec(gallivm, dst_type, bias);
 
    res = LLVMBuildOr(builder,
                      res,
@@ -259,7 +268,7 @@ lp_build_unsigned_norm_to_float(LLVMBuilderRef builder,
    res = LLVMBuildBitCast(builder, res, vec_type, "");
 
    res = LLVMBuildFSub(builder, res, bias_, "");
-   res = LLVMBuildFMul(builder, res, lp_build_const_vec(dst_type, scale), "");
+   res = LLVMBuildFMul(builder, res, lp_build_const_vec(gallivm, dst_type, scale), "");
 
    return res;
 }
@@ -272,12 +281,13 @@ lp_build_unsigned_norm_to_float(LLVMBuilderRef builder,
  * to the lp_type union.
  */
 void
-lp_build_conv(LLVMBuilderRef builder,
+lp_build_conv(struct gallivm_state *gallivm,
               struct lp_type src_type,
               struct lp_type dst_type,
               const LLVMValueRef *src, unsigned num_srcs,
               LLVMValueRef *dst, unsigned num_dsts)
 {
+   LLVMBuilderRef builder = gallivm->builder;
    struct lp_type tmp_type;
    LLVMValueRef tmp[LP_MAX_VECTOR_LENGTH];
    unsigned num_tmps;
@@ -342,12 +352,12 @@ lp_build_conv(LLVMBuilderRef builder,
          int32_type.length /= 4;
          int32_type.sign = 1;
 
-         src_vec_type   = lp_build_vec_type(src_type);
-         dst_vec_type   = lp_build_vec_type(dst_type);
-         int16_vec_type = lp_build_vec_type(int16_type);
-         int32_vec_type = lp_build_vec_type(int32_type);
+         src_vec_type   = lp_build_vec_type(gallivm, src_type);
+         dst_vec_type   = lp_build_vec_type(gallivm, dst_type);
+         int16_vec_type = lp_build_vec_type(gallivm, int16_type);
+         int32_vec_type = lp_build_vec_type(gallivm, int32_type);
 
-         const_255f = lp_build_const_vec(src_type, 255.0f);
+         const_255f = lp_build_const_vec(gallivm, src_type, 255.0f);
 
          a = LLVMBuildFMul(builder, src[0], const_255f, "");
          b = LLVMBuildFMul(builder, src[1], const_255f, "");
@@ -357,14 +367,14 @@ lp_build_conv(LLVMBuilderRef builder,
          {
             struct lp_build_context bld;
 
-            bld.builder = builder;
+            bld.gallivm = gallivm;
             bld.type = src_type;
             bld.vec_type = src_vec_type;
-            bld.int_elem_type = lp_build_elem_type(int32_type);
+            bld.int_elem_type = lp_build_elem_type(gallivm, int32_type);
             bld.int_vec_type = int32_vec_type;
-            bld.undef = lp_build_undef(src_type);
-            bld.zero = lp_build_zero(src_type);
-            bld.one = lp_build_one(src_type);
+            bld.undef = lp_build_undef(gallivm, src_type);
+            bld.zero = lp_build_zero(gallivm, src_type);
+            bld.one = lp_build_one(gallivm, src_type);
 
             src_int0 = lp_build_iround(&bld, a);
             src_int1 = lp_build_iround(&bld, b);
@@ -372,9 +382,9 @@ lp_build_conv(LLVMBuilderRef builder,
             src_int3 = lp_build_iround(&bld, d);
          }
          /* relying on clamping behavior of sse2 intrinsics here */
-         lo = lp_build_pack2(builder, int32_type, int16_type, src_int0, src_int1);
-         hi = lp_build_pack2(builder, int32_type, int16_type, src_int2, src_int3);
-         dst[i] = lp_build_pack2(builder, int16_type, dst_type, lo, hi);
+         lo = lp_build_pack2(gallivm, int32_type, int16_type, src_int0, src_int1);
+         hi = lp_build_pack2(gallivm, int32_type, int16_type, src_int2, src_int3);
+         dst[i] = lp_build_pack2(gallivm, int16_type, dst_type, lo, hi);
       }
       return; 
    }
@@ -391,13 +401,13 @@ lp_build_conv(LLVMBuilderRef builder,
       double dst_max = lp_const_max(dst_type);
       LLVMValueRef thres;
 
-      lp_build_context_init(&bld, builder, tmp_type);
+      lp_build_context_init(&bld, gallivm, tmp_type);
 
       if(src_min < dst_min) {
          if(dst_min == 0.0)
             thres = bld.zero;
          else
-            thres = lp_build_const_vec(src_type, dst_min);
+            thres = lp_build_const_vec(gallivm, src_type, dst_min);
          for(i = 0; i < num_tmps; ++i)
             tmp[i] = lp_build_max(&bld, tmp[i], thres);
       }
@@ -406,7 +416,7 @@ lp_build_conv(LLVMBuilderRef builder,
          if(dst_max == 1.0)
             thres = bld.one;
          else
-            thres = lp_build_const_vec(src_type, dst_max);
+            thres = lp_build_const_vec(gallivm, src_type, dst_max);
          for(i = 0; i < num_tmps; ++i)
             tmp[i] = lp_build_min(&bld, tmp[i], thres);
       }
@@ -422,7 +432,7 @@ lp_build_conv(LLVMBuilderRef builder,
    else if(tmp_type.floating) {
       if(!dst_type.fixed && !dst_type.sign && dst_type.norm) {
          for(i = 0; i < num_tmps; ++i) {
-            tmp[i] = lp_build_clamped_float_to_unsigned_norm(builder,
+            tmp[i] = lp_build_clamped_float_to_unsigned_norm(gallivm,
                                                              tmp_type,
                                                              dst_type.width,
                                                              tmp[i]);
@@ -434,14 +444,14 @@ lp_build_conv(LLVMBuilderRef builder,
          LLVMTypeRef tmp_vec_type;
 
          if (dst_scale != 1.0) {
-            LLVMValueRef scale = lp_build_const_vec(tmp_type, dst_scale);
+            LLVMValueRef scale = lp_build_const_vec(gallivm, tmp_type, dst_scale);
             for(i = 0; i < num_tmps; ++i)
                tmp[i] = LLVMBuildFMul(builder, tmp[i], scale, "");
          }
 
          /* Use an equally sized integer for intermediate computations */
          tmp_type.floating = FALSE;
-         tmp_vec_type = lp_build_vec_type(tmp_type);
+         tmp_vec_type = lp_build_vec_type(gallivm, tmp_type);
          for(i = 0; i < num_tmps; ++i) {
 #if 0
             if(dst_type.sign)
@@ -461,7 +471,8 @@ lp_build_conv(LLVMBuilderRef builder,
 
       /* FIXME: compensate different offsets too */
       if(src_shift > dst_shift) {
-         LLVMValueRef shift = lp_build_const_int_vec(tmp_type, src_shift - dst_shift);
+         LLVMValueRef shift = lp_build_const_int_vec(gallivm, tmp_type,
+                                                     src_shift - dst_shift);
          for(i = 0; i < num_tmps; ++i)
             if(src_type.sign)
                tmp[i] = LLVMBuildAShr(builder, tmp[i], shift, "");
@@ -485,7 +496,7 @@ lp_build_conv(LLVMBuilderRef builder,
       new_type.width  = dst_type.width;
       new_type.length = dst_type.length;
 
-      lp_build_resize(builder, tmp_type, new_type, tmp, num_srcs, tmp, num_dsts);
+      lp_build_resize(gallivm, tmp_type, new_type, tmp, num_srcs, tmp, num_dsts);
 
       tmp_type = new_type;
       num_tmps = num_dsts;
@@ -501,7 +512,7 @@ lp_build_conv(LLVMBuilderRef builder,
    else if(!src_type.floating && dst_type.floating) {
       if(!src_type.fixed && !src_type.sign && src_type.norm) {
          for(i = 0; i < num_tmps; ++i) {
-            tmp[i] = lp_build_unsigned_norm_to_float(builder,
+            tmp[i] = lp_build_unsigned_norm_to_float(gallivm,
                                                      src_type.width,
                                                      dst_type,
                                                      tmp[i]);
@@ -515,7 +526,7 @@ lp_build_conv(LLVMBuilderRef builder,
          /* Use an equally sized integer for intermediate computations */
          tmp_type.floating = TRUE;
          tmp_type.sign = TRUE;
-         tmp_vec_type = lp_build_vec_type(tmp_type);
+         tmp_vec_type = lp_build_vec_type(gallivm, tmp_type);
          for(i = 0; i < num_tmps; ++i) {
 #if 0
             if(dst_type.sign)
@@ -529,7 +540,7 @@ lp_build_conv(LLVMBuilderRef builder,
           }
 
           if (src_scale != 1.0) {
-             LLVMValueRef scale = lp_build_const_vec(tmp_type, 1.0/src_scale);
+             LLVMValueRef scale = lp_build_const_vec(gallivm, tmp_type, 1.0/src_scale);
              for(i = 0; i < num_tmps; ++i)
                 tmp[i] = LLVMBuildFMul(builder, tmp[i], scale, "");
           }
@@ -541,7 +552,7 @@ lp_build_conv(LLVMBuilderRef builder,
 
        /* FIXME: compensate different offsets too */
        if(src_shift < dst_shift) {
-          LLVMValueRef shift = lp_build_const_int_vec(tmp_type, dst_shift - src_shift);
+          LLVMValueRef shift = lp_build_const_int_vec(gallivm, tmp_type, dst_shift - src_shift);
           for(i = 0; i < num_tmps; ++i)
              tmp[i] = LLVMBuildShl(builder, tmp[i], shift, "");
        }
@@ -565,7 +576,7 @@ lp_build_conv(LLVMBuilderRef builder,
  * This is basically a very trimmed down version of lp_build_conv.
  */
 void
-lp_build_conv_mask(LLVMBuilderRef builder,
+lp_build_conv_mask(struct gallivm_state *gallivm,
                    struct lp_type src_type,
                    struct lp_type dst_type,
                    const LLVMValueRef *src, unsigned num_srcs,
@@ -599,11 +610,11 @@ lp_build_conv_mask(LLVMBuilderRef builder,
 
    if(src_type.width > dst_type.width) {
       assert(num_dsts == 1);
-      dst[0] = lp_build_pack(builder, src_type, dst_type, TRUE, src, num_srcs);
+      dst[0] = lp_build_pack(gallivm, src_type, dst_type, TRUE, src, num_srcs);
    }
    else if(src_type.width < dst_type.width) {
       assert(num_srcs == 1);
-      lp_build_unpack(builder, src_type, dst_type, src[0], dst, num_dsts);
+      lp_build_unpack(gallivm, src_type, dst_type, src[0], dst, num_dsts);
    }
    else {
       assert(num_srcs == num_dsts);
diff --git a/src/gallium/auxiliary/gallivm/lp_bld_conv.h b/src/gallium/auxiliary/gallivm/lp_bld_conv.h
index 628831c3ad..cec655980f 100644
--- a/src/gallium/auxiliary/gallivm/lp_bld_conv.h
+++ b/src/gallium/auxiliary/gallivm/lp_bld_conv.h
@@ -44,27 +44,27 @@ struct lp_type;
 
 
 LLVMValueRef
-lp_build_clamped_float_to_unsigned_norm(LLVMBuilderRef builder,
+lp_build_clamped_float_to_unsigned_norm(struct gallivm_state *gallivm,
                                         struct lp_type src_type,
                                         unsigned dst_width,
                                         LLVMValueRef src);
 
 LLVMValueRef
-lp_build_unsigned_norm_to_float(LLVMBuilderRef builder,
+lp_build_unsigned_norm_to_float(struct gallivm_state *gallivm,
                                 unsigned src_width,
                                 struct lp_type dst_type,
                                 LLVMValueRef src);
 
 
 void
-lp_build_conv(LLVMBuilderRef builder,
+lp_build_conv(struct gallivm_state *gallivm,
               struct lp_type src_type,
               struct lp_type dst_type,
               const LLVMValueRef *srcs, unsigned num_srcs,
               LLVMValueRef *dsts, unsigned num_dsts);
 
 void
-lp_build_conv_mask(LLVMBuilderRef builder,
+lp_build_conv_mask(struct gallivm_state *gallivm,
                    struct lp_type src_type,
                    struct lp_type dst_type,
                    const LLVMValueRef *src, unsigned num_srcs,
diff --git a/src/gallium/auxiliary/gallivm/lp_bld_debug.h b/src/gallium/auxiliary/gallivm/lp_bld_debug.h
index eb11dcd4ef..8a58f95b78 100644
--- a/src/gallium/auxiliary/gallivm/lp_bld_debug.h
+++ b/src/gallium/auxiliary/gallivm/lp_bld_debug.h
@@ -42,6 +42,7 @@
 #define GALLIVM_DEBUG_NO_OPT        (1 << 3)
 #define GALLIVM_DEBUG_PERF          (1 << 4)
 #define GALLIVM_DEBUG_NO_BRILINEAR  (1 << 5)
+#define GALLIVM_DEBUG_GC            (1 << 6)
 
 
 #ifdef DEBUG
diff --git a/src/gallium/auxiliary/gallivm/lp_bld_flow.c b/src/gallium/auxiliary/gallivm/lp_bld_flow.c
index a2cee199a0..a9c9c7af10 100644
--- a/src/gallium/auxiliary/gallivm/lp_bld_flow.c
+++ b/src/gallium/auxiliary/gallivm/lp_bld_flow.c
@@ -34,6 +34,7 @@
 #include "util/u_debug.h"
 #include "util/u_memory.h"
 
+#include "lp_bld_init.h"
 #include "lp_bld_type.h"
 #include "lp_bld_flow.h"
 
@@ -51,25 +52,25 @@
  * be used elsewhere.
  */
 LLVMBasicBlockRef
-lp_build_insert_new_block(LLVMBuilderRef builder, const char *name)
+lp_build_insert_new_block(struct gallivm_state *gallivm, const char *name)
 {
    LLVMBasicBlockRef current_block;
    LLVMBasicBlockRef next_block;
    LLVMBasicBlockRef new_block;
 
    /* get current basic block */
-   current_block = LLVMGetInsertBlock(builder);
+   current_block = LLVMGetInsertBlock(gallivm->builder);
 
    /* check if there's another block after this one */
    next_block = LLVMGetNextBasicBlock(current_block);
    if (next_block) {
       /* insert the new block before the next block */
-      new_block = LLVMInsertBasicBlock(next_block, name);
+      new_block = LLVMInsertBasicBlockInContext(gallivm->context, next_block, name);
    }
    else {
       /* append new block after current block */
       LLVMValueRef function = LLVMGetBasicBlockParent(current_block);
-      new_block = LLVMAppendBasicBlock(function, name);
+      new_block = LLVMAppendBasicBlockInContext(gallivm->context, function, name);
    }
 
    return new_block;
@@ -82,12 +83,11 @@ lp_build_insert_new_block(LLVMBuilderRef builder, const char *name)
  */
 void
 lp_build_flow_skip_begin(struct lp_build_skip_context *skip,
-                         LLVMBuilderRef builder)
+                         struct gallivm_state *gallivm)
 {
-   skip->builder = builder;
-
+   skip->gallivm = gallivm;
    /* create new basic block */
-   skip->block = lp_build_insert_new_block(skip->builder, "skip");
+   skip->block = lp_build_insert_new_block(gallivm, "skip");
 }
 
 
@@ -101,12 +101,12 @@ lp_build_flow_skip_cond_break(struct lp_build_skip_context *skip,
 {
    LLVMBasicBlockRef new_block;
 
-   new_block = lp_build_insert_new_block(skip->builder, "");
+   new_block = lp_build_insert_new_block(skip->gallivm, "");
 
    /* if cond is true, goto skip->block, else goto new_block */
-   LLVMBuildCondBr(skip->builder, cond, skip->block, new_block);
+   LLVMBuildCondBr(skip->gallivm->builder, cond, skip->block, new_block);
 
-   LLVMPositionBuilderAtEnd(skip->builder, new_block);
+   LLVMPositionBuilderAtEnd(skip->gallivm->builder, new_block);
 }
 
 
@@ -114,8 +114,8 @@ void
 lp_build_flow_skip_end(struct lp_build_skip_context *skip)
 {
    /* goto block */
-   LLVMBuildBr(skip->builder, skip->block);
-   LLVMPositionBuilderAtEnd(skip->builder, skip->block);
+   LLVMBuildBr(skip->gallivm->builder, skip->block);
+   LLVMPositionBuilderAtEnd(skip->gallivm->builder, skip->block);
 }
 
 
@@ -125,7 +125,7 @@ lp_build_flow_skip_end(struct lp_build_skip_context *skip)
 void
 lp_build_mask_check(struct lp_build_mask_context *mask)
 {
-   LLVMBuilderRef builder = mask->skip.builder;
+   LLVMBuilderRef builder = mask->skip.gallivm->builder;
    LLVMValueRef value;
    LLVMValueRef cond;
 
@@ -152,27 +152,27 @@ lp_build_mask_check(struct lp_build_mask_context *mask)
  */
 void
 lp_build_mask_begin(struct lp_build_mask_context *mask,
-                    LLVMBuilderRef builder,
+                    struct gallivm_state *gallivm,
                     struct lp_type type,
                     LLVMValueRef value)
 {
    memset(mask, 0, sizeof *mask);
 
-   mask->reg_type = LLVMIntType(type.width * type.length);
-   mask->var = lp_build_alloca(builder,
-                               lp_build_int_vec_type(type),
+   mask->reg_type = LLVMIntTypeInContext(gallivm->context, type.width * type.length);
+   mask->var = lp_build_alloca(gallivm,
+                               lp_build_int_vec_type(gallivm, type),
                                "execution_mask");
 
-   LLVMBuildStore(builder, value, mask->var);
+   LLVMBuildStore(gallivm->builder, value, mask->var);
 
-   lp_build_flow_skip_begin(&mask->skip, builder);
+   lp_build_flow_skip_begin(&mask->skip, gallivm);
 }
 
 
 LLVMValueRef
 lp_build_mask_value(struct lp_build_mask_context *mask)
 {
-   return LLVMBuildLoad(mask->skip.builder, mask->var, "");
+   return LLVMBuildLoad(mask->skip.gallivm->builder, mask->var, "");
 }
 
 
@@ -185,10 +185,10 @@ void
 lp_build_mask_update(struct lp_build_mask_context *mask,
                      LLVMValueRef value)
 {
-   value = LLVMBuildAnd(mask->skip.builder,
+   value = LLVMBuildAnd(mask->skip.gallivm->builder,
                         lp_build_mask_value(mask),
                         value, "");
-   LLVMBuildStore(mask->skip.builder, value, mask->var);
+   LLVMBuildStore(mask->skip.gallivm->builder, value, mask->var);
 }
 
 
@@ -205,13 +205,17 @@ lp_build_mask_end(struct lp_build_mask_context *mask)
 
 
 void
-lp_build_loop_begin(LLVMBuilderRef builder,
-                    LLVMValueRef start,
-                    struct lp_build_loop_state *state)
+lp_build_loop_begin(struct lp_build_loop_state *state,
+                    struct gallivm_state *gallivm,
+                    LLVMValueRef start)
+                    
 {
-   state->block = lp_build_insert_new_block(builder, "loop_begin");
+   LLVMBuilderRef builder = gallivm->builder;
+
+   state->block = lp_build_insert_new_block(gallivm, "loop_begin");
 
-   state->counter_var = lp_build_alloca(builder, LLVMTypeOf(start), "loop_counter");
+   state->counter_var = lp_build_alloca(gallivm, LLVMTypeOf(start), "loop_counter");
+   state->gallivm = gallivm;
 
    LLVMBuildStore(builder, start, state->counter_var);
 
@@ -224,12 +228,12 @@ lp_build_loop_begin(LLVMBuilderRef builder,
 
 
 void
-lp_build_loop_end_cond(LLVMBuilderRef builder,
+lp_build_loop_end_cond(struct lp_build_loop_state *state,
                        LLVMValueRef end,
                        LLVMValueRef step,
-                       LLVMIntPredicate llvm_cond,
-                       struct lp_build_loop_state *state)
+                       LLVMIntPredicate llvm_cond)
 {
+   LLVMBuilderRef builder = state->gallivm->builder;
    LLVMValueRef next;
    LLVMValueRef cond;
    LLVMBasicBlockRef after_block;
@@ -243,7 +247,7 @@ lp_build_loop_end_cond(LLVMBuilderRef builder,
 
    cond = LLVMBuildICmp(builder, llvm_cond, next, end, "");
 
-   after_block = lp_build_insert_new_block(builder, "loop_end");
+   after_block = lp_build_insert_new_block(state->gallivm, "loop_end");
 
    LLVMBuildCondBr(builder, cond, after_block, state->block);
 
@@ -254,12 +258,11 @@ lp_build_loop_end_cond(LLVMBuilderRef builder,
 
 
 void
-lp_build_loop_end(LLVMBuilderRef builder,
+lp_build_loop_end(struct lp_build_loop_state *state,
                   LLVMValueRef end,
-                  LLVMValueRef step,
-                  struct lp_build_loop_state *state)
+                  LLVMValueRef step)
 {
-   lp_build_loop_end_cond(builder, end, step, LLVMIntNE, state);
+   lp_build_loop_end_cond(state, end, step, LLVMIntNE);
 }
 
 
@@ -296,24 +299,27 @@ lp_build_loop_end(LLVMBuilderRef builder,
  */
 void
 lp_build_if(struct lp_build_if_state *ifthen,
-            LLVMBuilderRef builder,
+            struct gallivm_state *gallivm,
             LLVMValueRef condition)
 {
-   LLVMBasicBlockRef block = LLVMGetInsertBlock(builder);
+   LLVMBasicBlockRef block = LLVMGetInsertBlock(gallivm->builder);
 
    memset(ifthen, 0, sizeof *ifthen);
-   ifthen->builder = builder;
+   ifthen->gallivm = gallivm;
    ifthen->condition = condition;
    ifthen->entry_block = block;
 
    /* create endif/merge basic block for the phi functions */
-   ifthen->merge_block = lp_build_insert_new_block(builder, "endif-block");
+   ifthen->merge_block = lp_build_insert_new_block(gallivm, "endif-block");
 
    /* create/insert true_block before merge_block */
-   ifthen->true_block = LLVMInsertBasicBlock(ifthen->merge_block, "if-true-block");
+   ifthen->true_block =
+      LLVMInsertBasicBlockInContext(gallivm->context,
+                                    ifthen->merge_block,
+                                    "if-true-block");
 
    /* successive code goes into the true block */
-   LLVMPositionBuilderAtEnd(builder, ifthen->true_block);
+   LLVMPositionBuilderAtEnd(gallivm->builder, ifthen->true_block);
 }
 
 
@@ -323,14 +329,19 @@ lp_build_if(struct lp_build_if_state *ifthen,
 void
 lp_build_else(struct lp_build_if_state *ifthen)
 {
+   LLVMBuilderRef builder = ifthen->gallivm->builder;
+
    /* Append an unconditional Br(anch) instruction on the true_block */
-   LLVMBuildBr(ifthen->builder, ifthen->merge_block);
+   LLVMBuildBr(builder, ifthen->merge_block);
 
    /* create/insert false_block before the merge block */
-   ifthen->false_block = LLVMInsertBasicBlock(ifthen->merge_block, "if-false-block");
+   ifthen->false_block =
+      LLVMInsertBasicBlockInContext(ifthen->gallivm->context,
+                                    ifthen->merge_block,
+                                    "if-false-block");
 
    /* successive code goes into the else block */
-   LLVMPositionBuilderAtEnd(ifthen->builder, ifthen->false_block);
+   LLVMPositionBuilderAtEnd(builder, ifthen->false_block);
 }
 
 
@@ -340,28 +351,30 @@ lp_build_else(struct lp_build_if_state *ifthen)
 void
 lp_build_endif(struct lp_build_if_state *ifthen)
 {
+   LLVMBuilderRef builder = ifthen->gallivm->builder;
+
    /* Insert branch to the merge block from current block */
-   LLVMBuildBr(ifthen->builder, ifthen->merge_block);
+   LLVMBuildBr(builder, ifthen->merge_block);
 
    /*
     * Now patch in the various branch instructions.
     */
 
    /* Insert the conditional branch instruction at the end of entry_block */
-   LLVMPositionBuilderAtEnd(ifthen->builder, ifthen->entry_block);
+   LLVMPositionBuilderAtEnd(builder, ifthen->entry_block);
    if (ifthen->false_block) {
       /* we have an else clause */
-      LLVMBuildCondBr(ifthen->builder, ifthen->condition,
+      LLVMBuildCondBr(builder, ifthen->condition,
                       ifthen->true_block, ifthen->false_block);
    }
    else {
       /* no else clause */
-      LLVMBuildCondBr(ifthen->builder, ifthen->condition,
+      LLVMBuildCondBr(builder, ifthen->condition,
                       ifthen->true_block, ifthen->merge_block);
    }
 
    /* Resume building code at end of the ifthen->merge_block */
-   LLVMPositionBuilderAtEnd(ifthen->builder, ifthen->merge_block);
+   LLVMPositionBuilderAtEnd(builder, ifthen->merge_block);
 }
 
 
@@ -381,15 +394,16 @@ lp_build_endif(struct lp_build_if_state *ifthen)
  * - http://www.llvm.org/docs/tutorial/OCamlLangImpl7.html#memory
  */
 LLVMValueRef
-lp_build_alloca(LLVMBuilderRef builder,
+lp_build_alloca(struct gallivm_state *gallivm,
                 LLVMTypeRef type,
                 const char *name)
 {
+   LLVMBuilderRef builder = gallivm->builder;
    LLVMBasicBlockRef current_block = LLVMGetInsertBlock(builder);
    LLVMValueRef function = LLVMGetBasicBlockParent(current_block);
    LLVMBasicBlockRef first_block = LLVMGetEntryBasicBlock(function);
    LLVMValueRef first_instr = LLVMGetFirstInstruction(first_block);
-   LLVMBuilderRef first_builder = LLVMCreateBuilder();
+   LLVMBuilderRef first_builder = LLVMCreateBuilderInContext(gallivm->context);
    LLVMValueRef res;
 
    if (first_instr) {
@@ -422,16 +436,17 @@ lp_build_alloca(LLVMBuilderRef builder,
  * - http://www.llvm.org/docs/tutorial/OCamlLangImpl7.html#memory
  */
 LLVMValueRef
-lp_build_array_alloca(LLVMBuilderRef builder,
+lp_build_array_alloca(struct gallivm_state *gallivm,
                       LLVMTypeRef type,
                       LLVMValueRef count,
                       const char *name)
 {
+   LLVMBuilderRef builder = gallivm->builder;
    LLVMBasicBlockRef current_block = LLVMGetInsertBlock(builder);
    LLVMValueRef function = LLVMGetBasicBlockParent(current_block);
    LLVMBasicBlockRef first_block = LLVMGetEntryBasicBlock(function);
    LLVMValueRef first_instr = LLVMGetFirstInstruction(first_block);
-   LLVMBuilderRef first_builder = LLVMCreateBuilder();
+   LLVMBuilderRef first_builder = LLVMCreateBuilderInContext(gallivm->context);
    LLVMValueRef res;
 
    if (first_instr) {
diff --git a/src/gallium/auxiliary/gallivm/lp_bld_flow.h b/src/gallium/auxiliary/gallivm/lp_bld_flow.h
index e729ee6eaa..3cd5a9f42a 100644
--- a/src/gallium/auxiliary/gallivm/lp_bld_flow.h
+++ b/src/gallium/auxiliary/gallivm/lp_bld_flow.h
@@ -47,7 +47,7 @@ struct lp_type;
  */
 struct lp_build_skip_context
 {
-   LLVMBuilderRef builder;
+   struct gallivm_state *gallivm;
 
    /** Block to skip to */
    LLVMBasicBlockRef block;
@@ -55,7 +55,7 @@ struct lp_build_skip_context
 
 void
 lp_build_flow_skip_begin(struct lp_build_skip_context *ctx,
-                         LLVMBuilderRef builder);
+                         struct gallivm_state *gallivm);
 
 void
 lp_build_flow_skip_cond_break(struct lp_build_skip_context *ctx,
@@ -77,7 +77,7 @@ struct lp_build_mask_context
 
 void
 lp_build_mask_begin(struct lp_build_mask_context *mask,
-                    LLVMBuilderRef builder,
+                    struct gallivm_state *gallivm,
                     struct lp_type type,
                     LLVMValueRef value);
 
@@ -107,31 +107,28 @@ lp_build_mask_end(struct lp_build_mask_context *mask);
  */
 struct lp_build_loop_state
 {
-  LLVMBasicBlockRef block;
-  LLVMValueRef counter_var;
-  LLVMValueRef counter;
+   LLVMBasicBlockRef block;
+   LLVMValueRef counter_var;
+   LLVMValueRef counter;
+   struct gallivm_state *gallivm;
 };
 
 
 void
-lp_build_loop_begin(LLVMBuilderRef builder,
-                    LLVMValueRef start,
-                    struct lp_build_loop_state *state);
-
+lp_build_loop_begin(struct lp_build_loop_state *state,
+                    struct gallivm_state *gallivm,
+                    LLVMValueRef start);
 
 void
-lp_build_loop_end(LLVMBuilderRef builder,
+lp_build_loop_end(struct lp_build_loop_state *state,
                   LLVMValueRef end,
-                  LLVMValueRef step,
-                  struct lp_build_loop_state *state);
+                  LLVMValueRef step);
 
 void
-lp_build_loop_end_cond(LLVMBuilderRef builder,
+lp_build_loop_end_cond(struct lp_build_loop_state *state,
                        LLVMValueRef end,
                        LLVMValueRef step,
-                       LLVMIntPredicate cond,
-                       struct lp_build_loop_state *state);
-
+                       LLVMIntPredicate cond);
 
 
 
@@ -140,7 +137,7 @@ lp_build_loop_end_cond(LLVMBuilderRef builder,
  */
 struct lp_build_if_state
 {
-   LLVMBuilderRef builder;
+   struct gallivm_state *gallivm;
    LLVMValueRef condition;
    LLVMBasicBlockRef entry_block;
    LLVMBasicBlockRef true_block;
@@ -151,7 +148,7 @@ struct lp_build_if_state
 
 void
 lp_build_if(struct lp_build_if_state *ctx,
-            LLVMBuilderRef builder,
+            struct gallivm_state *gallivm,
             LLVMValueRef condition);
 
 void
@@ -161,15 +158,15 @@ void
 lp_build_endif(struct lp_build_if_state *ctx);
 
 LLVMBasicBlockRef
-lp_build_insert_new_block(LLVMBuilderRef builder, const char *name);
+lp_build_insert_new_block(struct gallivm_state *gallivm, const char *name);
 
 LLVMValueRef
-lp_build_alloca(LLVMBuilderRef builder,
+lp_build_alloca(struct gallivm_state *gallivm,
                 LLVMTypeRef type,
                 const char *name);
 
 LLVMValueRef
-lp_build_array_alloca(LLVMBuilderRef builder,
+lp_build_array_alloca(struct gallivm_state *gallivm,
                       LLVMTypeRef type,
                       LLVMValueRef count,
                       const char *name);
diff --git a/src/gallium/auxiliary/gallivm/lp_bld_format.h b/src/gallium/auxiliary/gallivm/lp_bld_format.h
index 60e22d727a..04142d905b 100644
--- a/src/gallium/auxiliary/gallivm/lp_bld_format.h
+++ b/src/gallium/auxiliary/gallivm/lp_bld_format.h
@@ -35,6 +35,7 @@
  */
 
 #include "gallivm/lp_bld.h"
+#include "gallivm/lp_bld_init.h"
 
 #include "pipe/p_format.h"
 
@@ -53,12 +54,12 @@ lp_build_format_swizzle_aos(const struct util_format_description *desc,
                             LLVMValueRef unswizzled);
 
 LLVMValueRef
-lp_build_pack_rgba_aos(LLVMBuilderRef builder,
+lp_build_pack_rgba_aos(struct gallivm_state *gallivm,
                        const struct util_format_description *desc,
                        LLVMValueRef rgba);
 
 LLVMValueRef
-lp_build_fetch_rgba_aos(LLVMBuilderRef builder,
+lp_build_fetch_rgba_aos(struct gallivm_state *gallivm,
                         const struct util_format_description *format_desc,
                         struct lp_type type,
                         LLVMValueRef base_ptr,
@@ -78,20 +79,20 @@ lp_build_format_swizzle_soa(const struct util_format_description *format_desc,
                             LLVMValueRef swizzled_out[4]);
 
 void
-lp_build_unpack_rgba_soa(LLVMBuilderRef builder,
+lp_build_unpack_rgba_soa(struct gallivm_state *gallivm,
                          const struct util_format_description *format_desc,
                          struct lp_type type,
                          LLVMValueRef packed,
                          LLVMValueRef rgba_out[4]);
 
 void
-lp_build_rgba8_to_f32_soa(LLVMBuilderRef builder,
+lp_build_rgba8_to_f32_soa(struct gallivm_state *gallivm,
                           struct lp_type dst_type,
                           LLVMValueRef packed,
                           LLVMValueRef *rgba);
 
 void
-lp_build_fetch_rgba_soa(LLVMBuilderRef builder,
+lp_build_fetch_rgba_soa(struct gallivm_state *gallivm,
                         const struct util_format_description *format_desc,
                         struct lp_type type,
                         LLVMValueRef base_ptr,
@@ -106,7 +107,7 @@ lp_build_fetch_rgba_soa(LLVMBuilderRef builder,
 
 
 LLVMValueRef
-lp_build_fetch_subsampled_rgba_aos(LLVMBuilderRef builder,
+lp_build_fetch_subsampled_rgba_aos(struct gallivm_state *gallivm,
                                    const struct util_format_description *format_desc,
                                    unsigned n,
                                    LLVMValueRef base_ptr,
diff --git a/src/gallium/auxiliary/gallivm/lp_bld_format_aos.c b/src/gallium/auxiliary/gallivm/lp_bld_format_aos.c
index 6b9189e1da..82ab19eda1 100644
--- a/src/gallium/auxiliary/gallivm/lp_bld_format_aos.c
+++ b/src/gallium/auxiliary/gallivm/lp_bld_format_aos.c
@@ -36,6 +36,7 @@
 #include "util/u_format.h"
 #include "util/u_memory.h"
 #include "util/u_math.h"
+#include "util/u_pointer.h"
 #include "util/u_string.h"
 
 #include "lp_bld_arit.h"
@@ -145,10 +146,11 @@ format_matches_type(const struct util_format_description *desc,
  * @return RGBA in a float[4] or ubyte[4] or ushort[4] vector.
  */
 static INLINE LLVMValueRef
-lp_build_unpack_arith_rgba_aos(LLVMBuilderRef builder,
+lp_build_unpack_arith_rgba_aos(struct gallivm_state *gallivm,
                                const struct util_format_description *desc,
                                LLVMValueRef packed)
 {
+   LLVMBuilderRef builder = gallivm->builder;
    LLVMValueRef shifted, casted, scaled, masked;
    LLVMValueRef shifts[4];
    LLVMValueRef masks[4];
@@ -167,21 +169,21 @@ lp_build_unpack_arith_rgba_aos(LLVMBuilderRef builder,
 
    /* Do the intermediate integer computations with 32bit integers since it
     * matches floating point size */
-   assert (LLVMTypeOf(packed) == LLVMInt32Type());
+   assert (LLVMTypeOf(packed) == LLVMInt32TypeInContext(gallivm->context));
 
    /* Broadcast the packed value to all four channels
     * before: packed = BGRA
     * after: packed = {BGRA, BGRA, BGRA, BGRA}
     */
    packed = LLVMBuildInsertElement(builder,
-                                   LLVMGetUndef(LLVMVectorType(LLVMInt32Type(), 4)),
+                                   LLVMGetUndef(LLVMVectorType(LLVMInt32TypeInContext(gallivm->context), 4)),
                                    packed,
-                                   LLVMConstNull(LLVMInt32Type()),
+                                   LLVMConstNull(LLVMInt32TypeInContext(gallivm->context)),
                                    "");
    packed = LLVMBuildShuffleVector(builder,
                                    packed,
-                                   LLVMGetUndef(LLVMVectorType(LLVMInt32Type(), 4)),
-                                   LLVMConstNull(LLVMVectorType(LLVMInt32Type(), 4)),
+                                   LLVMGetUndef(LLVMVectorType(LLVMInt32TypeInContext(gallivm->context), 4)),
+                                   LLVMConstNull(LLVMVectorType(LLVMInt32TypeInContext(gallivm->context), 4)),
                                    "");
 
    /* Initialize vector constants */
@@ -194,9 +196,9 @@ lp_build_unpack_arith_rgba_aos(LLVMBuilderRef builder,
       unsigned bits = desc->channel[i].size;
 
       if (desc->channel[i].type == UTIL_FORMAT_TYPE_VOID) {
-         shifts[i] = LLVMGetUndef(LLVMInt32Type());
-         masks[i] = LLVMConstNull(LLVMInt32Type());
-         scales[i] =  LLVMConstNull(LLVMFloatType());
+         shifts[i] = LLVMGetUndef(LLVMInt32TypeInContext(gallivm->context));
+         masks[i] = LLVMConstNull(LLVMInt32TypeInContext(gallivm->context));
+         scales[i] =  LLVMConstNull(LLVMFloatTypeInContext(gallivm->context));
       }
       else {
          unsigned long long mask = (1ULL << bits) - 1;
@@ -207,15 +209,15 @@ lp_build_unpack_arith_rgba_aos(LLVMBuilderRef builder,
             needs_uitofp = TRUE;
          }
 
-         shifts[i] = LLVMConstInt(LLVMInt32Type(), shift, 0);
-         masks[i] = LLVMConstInt(LLVMInt32Type(), mask, 0);
+         shifts[i] = lp_build_const_int32(gallivm, shift);
+         masks[i] = lp_build_const_int32(gallivm, mask);
 
          if (desc->channel[i].normalized) {
-            scales[i] = LLVMConstReal(LLVMFloatType(), 1.0/mask);
+            scales[i] = lp_build_const_float(gallivm, 1.0 / mask);
             normalized = TRUE;
          }
          else
-            scales[i] =  LLVMConstReal(LLVMFloatType(), 1.0);
+            scales[i] =  lp_build_const_float(gallivm, 1.0);
       }
 
       shift += bits;
@@ -230,9 +232,9 @@ lp_build_unpack_arith_rgba_aos(LLVMBuilderRef builder,
 
    if (!needs_uitofp) {
       /* UIToFP can't be expressed in SSE2 */
-      casted = LLVMBuildSIToFP(builder, masked, LLVMVectorType(LLVMFloatType(), 4), "");
+      casted = LLVMBuildSIToFP(builder, masked, LLVMVectorType(LLVMFloatTypeInContext(gallivm->context), 4), "");
    } else {
-      casted = LLVMBuildUIToFP(builder, masked, LLVMVectorType(LLVMFloatType(), 4), "");
+      casted = LLVMBuildUIToFP(builder, masked, LLVMVectorType(LLVMFloatTypeInContext(gallivm->context), 4), "");
    }
 
    /* At this point 'casted' may be a vector of floats such as
@@ -258,10 +260,11 @@ lp_build_unpack_arith_rgba_aos(LLVMBuilderRef builder,
  * a time is rarely if ever needed.
  */
 LLVMValueRef
-lp_build_pack_rgba_aos(LLVMBuilderRef builder,
+lp_build_pack_rgba_aos(struct gallivm_state *gallivm,
                        const struct util_format_description *desc,
                        LLVMValueRef rgba)
 {
+   LLVMBuilderRef builder = gallivm->builder;
    LLVMTypeRef type;
    LLVMValueRef packed = NULL;
    LLVMValueRef swizzles[4];
@@ -276,7 +279,7 @@ lp_build_pack_rgba_aos(LLVMBuilderRef builder,
    assert(desc->block.width == 1);
    assert(desc->block.height == 1);
 
-   type = LLVMIntType(desc->block.bits);
+   type = LLVMIntTypeInContext(gallivm->context, desc->block.bits);
 
    /* Unswizzle the color components into the source vector. */
    for (i = 0; i < 4; ++i) {
@@ -285,13 +288,13 @@ lp_build_pack_rgba_aos(LLVMBuilderRef builder,
             break;
       }
       if (j < 4)
-         swizzles[i] = LLVMConstInt(LLVMInt32Type(), j, 0);
+         swizzles[i] = lp_build_const_int32(gallivm, j);
       else
-         swizzles[i] = LLVMGetUndef(LLVMInt32Type());
+         swizzles[i] = LLVMGetUndef(LLVMInt32TypeInContext(gallivm->context));
    }
 
    unswizzled = LLVMBuildShuffleVector(builder, rgba,
-                                       LLVMGetUndef(LLVMVectorType(LLVMFloatType(), 4)),
+                                       LLVMGetUndef(LLVMVectorType(LLVMFloatTypeInContext(gallivm->context), 4)),
                                        LLVMConstVector(swizzles, 4), "");
 
    normalized = FALSE;
@@ -300,8 +303,8 @@ lp_build_pack_rgba_aos(LLVMBuilderRef builder,
       unsigned bits = desc->channel[i].size;
 
       if (desc->channel[i].type == UTIL_FORMAT_TYPE_VOID) {
-         shifts[i] = LLVMGetUndef(LLVMInt32Type());
-         scales[i] =  LLVMGetUndef(LLVMFloatType());
+         shifts[i] = LLVMGetUndef(LLVMInt32TypeInContext(gallivm->context));
+         scales[i] =  LLVMGetUndef(LLVMFloatTypeInContext(gallivm->context));
       }
       else {
          unsigned mask = (1 << bits) - 1;
@@ -309,14 +312,14 @@ lp_build_pack_rgba_aos(LLVMBuilderRef builder,
          assert(desc->channel[i].type == UTIL_FORMAT_TYPE_UNSIGNED);
          assert(bits < 32);
 
-         shifts[i] = LLVMConstInt(LLVMInt32Type(), shift, 0);
+         shifts[i] = lp_build_const_int32(gallivm, shift);
 
          if (desc->channel[i].normalized) {
-            scales[i] = LLVMConstReal(LLVMFloatType(), mask);
+            scales[i] = lp_build_const_float(gallivm, mask);
             normalized = TRUE;
          }
          else
-            scales[i] =  LLVMConstReal(LLVMFloatType(), 1.0);
+            scales[i] = lp_build_const_float(gallivm, 1.0);
       }
 
       shift += bits;
@@ -327,14 +330,15 @@ lp_build_pack_rgba_aos(LLVMBuilderRef builder,
    else
       scaled = unswizzled;
 
-   casted = LLVMBuildFPToSI(builder, scaled, LLVMVectorType(LLVMInt32Type(), 4), "");
+   casted = LLVMBuildFPToSI(builder, scaled, LLVMVectorType(LLVMInt32TypeInContext(gallivm->context), 4), "");
 
    shifted = LLVMBuildShl(builder, casted, LLVMConstVector(shifts, 4), "");
    
    /* Bitwise or all components */
    for (i = 0; i < 4; ++i) {
       if (desc->channel[i].type == UTIL_FORMAT_TYPE_UNSIGNED) {
-         LLVMValueRef component = LLVMBuildExtractElement(builder, shifted, LLVMConstInt(LLVMInt32Type(), i, 0), "");
+         LLVMValueRef component = LLVMBuildExtractElement(builder, shifted,
+                                               lp_build_const_int32(gallivm, i), "");
          if (packed)
             packed = LLVMBuildOr(builder, packed, component, "");
          else
@@ -343,7 +347,7 @@ lp_build_pack_rgba_aos(LLVMBuilderRef builder,
    }
 
    if (!packed)
-      packed = LLVMGetUndef(LLVMInt32Type());
+      packed = LLVMGetUndef(LLVMInt32TypeInContext(gallivm->context));
 
    if (desc->block.bits < 32)
       packed = LLVMBuildTrunc(builder, packed, type, "");
@@ -364,7 +368,7 @@ lp_build_pack_rgba_aos(LLVMBuilderRef builder,
  * \return  a 4 element vector with the pixel's RGBA values.
  */
 LLVMValueRef
-lp_build_fetch_rgba_aos(LLVMBuilderRef builder,
+lp_build_fetch_rgba_aos(struct gallivm_state *gallivm,
                         const struct util_format_description *format_desc,
                         struct lp_type type,
                         LLVMValueRef base_ptr,
@@ -372,13 +376,14 @@ lp_build_fetch_rgba_aos(LLVMBuilderRef builder,
                         LLVMValueRef i,
                         LLVMValueRef j)
 {
+   LLVMBuilderRef builder = gallivm->builder;
    unsigned num_pixels = type.length / 4;
    struct lp_build_context bld;
 
    assert(type.length <= LP_MAX_VECTOR_LENGTH);
    assert(type.length % 4 == 0);
 
-   lp_build_context_init(&bld, builder, type);
+   lp_build_context_init(&bld, gallivm, type);
 
    /*
     * Trivial case
@@ -397,13 +402,14 @@ lp_build_fetch_rgba_aos(LLVMBuilderRef builder,
        * scaling or converting.
        */
 
-      packed = lp_build_gather(builder, type.length/4,
+      packed = lp_build_gather(gallivm, type.length/4,
                                format_desc->block.bits, type.width*4,
                                base_ptr, offset);
 
       assert(format_desc->block.bits <= type.width * type.length);
 
-      packed = LLVMBuildBitCast(builder, packed, lp_build_vec_type(type), "");
+      packed = LLVMBuildBitCast(gallivm->builder, packed,
+                                lp_build_vec_type(gallivm, type), "");
 
       return lp_build_format_swizzle_aos(format_desc, &bld, packed);
    }
@@ -435,11 +441,12 @@ lp_build_fetch_rgba_aos(LLVMBuilderRef builder,
       for (k = 0; k < num_pixels; ++k) {
          LLVMValueRef packed;
 
-         packed = lp_build_gather_elem(builder, num_pixels,
+         packed = lp_build_gather_elem(gallivm, num_pixels,
                                        format_desc->block.bits, 32,
                                        base_ptr, offset, k);
 
-         tmps[k] = lp_build_unpack_arith_rgba_aos(builder, format_desc,
+         tmps[k] = lp_build_unpack_arith_rgba_aos(gallivm,
+                                                  format_desc,
                                                   packed);
       }
 
@@ -455,7 +462,7 @@ lp_build_fetch_rgba_aos(LLVMBuilderRef builder,
                       __FUNCTION__, format_desc->short_name);
       }
 
-      lp_build_conv(builder,
+      lp_build_conv(gallivm,
                     lp_float32_vec4_type(),
                     type,
                     tmps, num_pixels, &res, 1);
@@ -476,14 +483,14 @@ lp_build_fetch_rgba_aos(LLVMBuilderRef builder,
       tmp_type.length = num_pixels * 4;
       tmp_type.norm = TRUE;
 
-      tmp = lp_build_fetch_subsampled_rgba_aos(builder,
+      tmp = lp_build_fetch_subsampled_rgba_aos(gallivm,
                                                format_desc,
                                                num_pixels,
                                                base_ptr,
                                                offset,
                                                i, j);
 
-      lp_build_conv(builder,
+      lp_build_conv(gallivm,
                     tmp_type, type,
                     &tmp, 1, &tmp, 1);
 
@@ -505,51 +512,52 @@ lp_build_fetch_rgba_aos(LLVMBuilderRef builder,
        * or incentive to optimize.
        */
 
-      LLVMModuleRef module = LLVMGetGlobalParent(LLVMGetBasicBlockParent(LLVMGetInsertBlock(builder)));
-      char name[256];
-      LLVMTypeRef i8t = LLVMInt8Type();
+      LLVMTypeRef i8t = LLVMInt8TypeInContext(gallivm->context);
       LLVMTypeRef pi8t = LLVMPointerType(i8t, 0);
-      LLVMTypeRef i32t = LLVMInt32Type();
+      LLVMTypeRef i32t = LLVMInt32TypeInContext(gallivm->context);
       LLVMValueRef function;
       LLVMValueRef tmp_ptr;
       LLVMValueRef tmp;
       LLVMValueRef res;
       unsigned k;
 
-      util_snprintf(name, sizeof name, "util_format_%s_fetch_rgba_8unorm",
-                    format_desc->short_name);
-
       if (gallivm_debug & GALLIVM_DEBUG_PERF) {
-         debug_printf("%s: falling back to %s\n", __FUNCTION__, name);
+         debug_printf("%s: falling back to util_format_%s_fetch_rgba_8unorm\n",
+                      __FUNCTION__, format_desc->short_name);
       }
 
       /*
        * Declare and bind format_desc->fetch_rgba_8unorm().
        */
 
-      function = LLVMGetNamedFunction(module, name);
-      if (!function) {
+      {
+         /*
+          * Function to call looks like:
+          *   fetch(uint8_t *dst, const uint8_t *src, unsigned i, unsigned j)
+          */
          LLVMTypeRef ret_type;
          LLVMTypeRef arg_types[4];
          LLVMTypeRef function_type;
 
-         ret_type = LLVMVoidType();
+         ret_type = LLVMVoidTypeInContext(gallivm->context);
          arg_types[0] = pi8t;
          arg_types[1] = pi8t;
-         arg_types[3] = arg_types[2] = LLVMIntType(sizeof(unsigned) * 8);
-         function_type = LLVMFunctionType(ret_type, arg_types, Elements(arg_types), 0);
-         function = LLVMAddFunction(module, name, function_type);
-
-         LLVMSetFunctionCallConv(function, LLVMCCallConv);
-         LLVMSetLinkage(function, LLVMExternalLinkage);
-
-         assert(LLVMIsDeclaration(function));
-
-         LLVMAddGlobalMapping(lp_build_engine, function,
-                              func_to_pointer((func_pointer)format_desc->fetch_rgba_8unorm));
+         arg_types[2] = i32t;
+         arg_types[3] = i32t;
+         function_type = LLVMFunctionType(ret_type, arg_types,
+                                          Elements(arg_types), 0);
+
+         /* make const pointer for the C fetch_rgba_8unorm function */
+         function = lp_build_const_int_pointer(gallivm,
+            func_to_pointer((func_pointer) format_desc->fetch_rgba_8unorm));
+
+         /* cast the callee pointer to the function's type */
+         function = LLVMBuildBitCast(builder, function,
+                                     LLVMPointerType(function_type, 0),
+                                     "cast callee");
       }
 
-      tmp_ptr = lp_build_alloca(builder, i32t, "");
+      tmp_ptr = lp_build_alloca(gallivm, i32t, "");
 
       res = LLVMGetUndef(LLVMVectorType(i32t, num_pixels));
 
@@ -559,11 +567,11 @@ lp_build_fetch_rgba_aos(LLVMBuilderRef builder,
        */
 
       for (k = 0; k < num_pixels; ++k) {
-         LLVMValueRef index = LLVMConstInt(LLVMInt32Type(), k, 0);
+         LLVMValueRef index = lp_build_const_int32(gallivm, k);
          LLVMValueRef args[4];
 
          args[0] = LLVMBuildBitCast(builder, tmp_ptr, pi8t, "");
-         args[1] = lp_build_gather_elem_ptr(builder, num_pixels,
+         args[1] = lp_build_gather_elem_ptr(gallivm, num_pixels,
                                             base_ptr, offset, k);
 
          if (num_pixels == 1) {
@@ -608,51 +616,58 @@ lp_build_fetch_rgba_aos(LLVMBuilderRef builder,
        * or incentive to optimize.
        */
 
-      LLVMModuleRef module = LLVMGetGlobalParent(LLVMGetBasicBlockParent(LLVMGetInsertBlock(builder)));
-      char name[256];
-      LLVMTypeRef f32t = LLVMFloatType();
+      LLVMTypeRef f32t = LLVMFloatTypeInContext(gallivm->context);
       LLVMTypeRef f32x4t = LLVMVectorType(f32t, 4);
       LLVMTypeRef pf32t = LLVMPointerType(f32t, 0);
+      LLVMTypeRef pi8t = LLVMPointerType(LLVMInt8TypeInContext(gallivm->context), 0);
+      LLVMTypeRef i32t = LLVMInt32TypeInContext(gallivm->context);
       LLVMValueRef function;
       LLVMValueRef tmp_ptr;
       LLVMValueRef tmps[LP_MAX_VECTOR_LENGTH/4];
       LLVMValueRef res;
       unsigned k;
 
-      util_snprintf(name, sizeof name, "util_format_%s_fetch_rgba_float",
-                    format_desc->short_name);
-
       if (gallivm_debug & GALLIVM_DEBUG_PERF) {
-         debug_printf("%s: falling back to %s\n", __FUNCTION__, name);
+         debug_printf("%s: falling back to util_format_%s_fetch_rgba_float\n",
+                      __FUNCTION__, format_desc->short_name);
       }
 
       /*
        * Declare and bind format_desc->fetch_rgba_float().
        */
 
-      function = LLVMGetNamedFunction(module, name);
-      if (!function) {
+      {
+         /*
+          * Function to call looks like:
+          *   fetch(float *dst, const uint8_t *src, unsigned i, unsigned j)
+          */
          LLVMTypeRef ret_type;
          LLVMTypeRef arg_types[4];
          LLVMTypeRef function_type;
 
-         ret_type = LLVMVoidType();
+         ret_type = LLVMVoidTypeInContext(gallivm->context);
          arg_types[0] = pf32t;
-         arg_types[1] = LLVMPointerType(LLVMInt8Type(), 0);
-         arg_types[3] = arg_types[2] = LLVMIntType(sizeof(unsigned) * 8);
-         function_type = LLVMFunctionType(ret_type, arg_types, Elements(arg_types), 0);
-         function = LLVMAddFunction(module, name, function_type);
+         arg_types[1] = pi8t;
+         arg_types[2] = i32t;
+         arg_types[3] = i32t;
+         function_type = LLVMFunctionType(ret_type, arg_types,
+                                          Elements(arg_types), 0);
 
-         LLVMSetFunctionCallConv(function, LLVMCCallConv);
-         LLVMSetLinkage(function, LLVMExternalLinkage);
+         /* Note: we're using this casting here instead of LLVMAddGlobalMapping()
+          * to work around a bug in LLVM 2.6, and for efficiency/simplicity.
+          */
 
-         assert(LLVMIsDeclaration(function));
+         /* make const pointer for the C fetch_rgba_float function */
+         function = lp_build_const_int_pointer(gallivm,
+            func_to_pointer((func_pointer) format_desc->fetch_rgba_float));
 
-         LLVMAddGlobalMapping(lp_build_engine, function,
-                              func_to_pointer((func_pointer)format_desc->fetch_rgba_float));
+         /* cast the callee pointer to the function's type */
+         function = LLVMBuildBitCast(builder, function,
+                                     LLVMPointerType(function_type, 0),
+                                     "cast callee");
       }
 
-      tmp_ptr = lp_build_alloca(builder, f32x4t, "");
+      tmp_ptr = lp_build_alloca(gallivm, f32x4t, "");
 
       /*
        * Invoke format_desc->fetch_rgba_float() for each pixel and insert the result
@@ -663,7 +678,7 @@ lp_build_fetch_rgba_aos(LLVMBuilderRef builder,
          LLVMValueRef args[4];
 
          args[0] = LLVMBuildBitCast(builder, tmp_ptr, pf32t, "");
-         args[1] = lp_build_gather_elem_ptr(builder, num_pixels,
+         args[1] = lp_build_gather_elem_ptr(gallivm, num_pixels,
                                             base_ptr, offset, k);
 
          if (num_pixels == 1) {
@@ -671,7 +686,7 @@ lp_build_fetch_rgba_aos(LLVMBuilderRef builder,
             args[3] = j;
          }
          else {
-            LLVMValueRef index = LLVMConstInt(LLVMInt32Type(), k, 0);
+            LLVMValueRef index = lp_build_const_int32(gallivm, k);
             args[2] = LLVMBuildExtractElement(builder, i, index, "");
             args[3] = LLVMBuildExtractElement(builder, j, index, "");
          }
@@ -681,7 +696,7 @@ lp_build_fetch_rgba_aos(LLVMBuilderRef builder,
          tmps[k] = LLVMBuildLoad(builder, tmp_ptr, "");
       }
 
-      lp_build_conv(builder,
+      lp_build_conv(gallivm,
                     lp_float32_vec4_type(),
                     type,
                     tmps, num_pixels, &res, 1);
@@ -690,5 +705,5 @@ lp_build_fetch_rgba_aos(LLVMBuilderRef builder,
    }
 
    assert(0);
-   return lp_build_undef(type);
+   return lp_build_undef(gallivm, type);
 }
diff --git a/src/gallium/auxiliary/gallivm/lp_bld_format_soa.c b/src/gallium/auxiliary/gallivm/lp_bld_format_soa.c
index ce7e54afc7..0a57b3ce79 100644
--- a/src/gallium/auxiliary/gallivm/lp_bld_format_soa.c
+++ b/src/gallium/auxiliary/gallivm/lp_bld_format_soa.c
@@ -97,12 +97,13 @@ lp_build_format_swizzle_soa(const struct util_format_description *format_desc,
  * \param rgba_out  returns the SoA R,G,B,A vectors
  */
 void
-lp_build_unpack_rgba_soa(LLVMBuilderRef builder,
+lp_build_unpack_rgba_soa(struct gallivm_state *gallivm,
                          const struct util_format_description *format_desc,
                          struct lp_type type,
                          LLVMValueRef packed,
                          LLVMValueRef rgba_out[4])
 {
+   LLVMBuilderRef builder = gallivm->builder;
    struct lp_build_context bld;
    LLVMValueRef inputs[4];
    unsigned start;
@@ -116,7 +117,7 @@ lp_build_unpack_rgba_soa(LLVMBuilderRef builder,
    assert(type.floating);
    assert(type.width == 32);
 
-   lp_build_context_init(&bld, builder, type);
+   lp_build_context_init(&bld, gallivm, type);
 
    /* Decode the input vector components */
    start = 0;
@@ -129,7 +130,7 @@ lp_build_unpack_rgba_soa(LLVMBuilderRef builder,
 
       switch(format_desc->channel[chan].type) {
       case UTIL_FORMAT_TYPE_VOID:
-         input = lp_build_undef(type);
+         input = lp_build_undef(gallivm, type);
          break;
 
       case UTIL_FORMAT_TYPE_UNSIGNED:
@@ -138,7 +139,7 @@ lp_build_unpack_rgba_soa(LLVMBuilderRef builder,
           */
 
          if (start) {
-            input = LLVMBuildLShr(builder, input, lp_build_const_int_vec(type, start), "");
+            input = LLVMBuildLShr(builder, input, lp_build_const_int_vec(gallivm, type, start), "");
          }
 
          /*
@@ -147,7 +148,7 @@ lp_build_unpack_rgba_soa(LLVMBuilderRef builder,
 
          if (stop < format_desc->block.bits) {
             unsigned mask = ((unsigned long long)1 << width) - 1;
-            input = LLVMBuildAnd(builder, input, lp_build_const_int_vec(type, mask), "");
+            input = LLVMBuildAnd(builder, input, lp_build_const_int_vec(gallivm, type, mask), "");
          }
 
          /*
@@ -156,14 +157,15 @@ lp_build_unpack_rgba_soa(LLVMBuilderRef builder,
 
          if (type.floating) {
             if(format_desc->channel[chan].normalized)
-               input = lp_build_unsigned_norm_to_float(builder, width, type, input);
+               input = lp_build_unsigned_norm_to_float(gallivm, width, type, input);
             else
-               input = LLVMBuildSIToFP(builder, input, lp_build_vec_type(type), "");
+               input = LLVMBuildSIToFP(builder, input,
+                                       lp_build_vec_type(gallivm, type), "");
          }
          else {
             /* FIXME */
             assert(0);
-            input = lp_build_undef(type);
+            input = lp_build_undef(gallivm, type);
          }
 
          break;
@@ -175,7 +177,7 @@ lp_build_unpack_rgba_soa(LLVMBuilderRef builder,
 
          if (stop < type.width) {
             unsigned bits = type.width - stop;
-            LLVMValueRef bits_val = lp_build_const_int_vec(type, bits);
+            LLVMValueRef bits_val = lp_build_const_int_vec(gallivm, type, bits);
             input = LLVMBuildShl(builder, input, bits_val, "");
          }
 
@@ -185,7 +187,7 @@ lp_build_unpack_rgba_soa(LLVMBuilderRef builder,
 
          if (format_desc->channel[chan].size < type.width) {
             unsigned bits = type.width - format_desc->channel[chan].size;
-            LLVMValueRef bits_val = lp_build_const_int_vec(type, bits);
+            LLVMValueRef bits_val = lp_build_const_int_vec(gallivm, type, bits);
             input = LLVMBuildAShr(builder, input, bits_val, "");
          }
 
@@ -194,17 +196,17 @@ lp_build_unpack_rgba_soa(LLVMBuilderRef builder,
           */
 
          if (type.floating) {
-            input = LLVMBuildSIToFP(builder, input, lp_build_vec_type(type), "");
+            input = LLVMBuildSIToFP(builder, input, lp_build_vec_type(gallivm, type), "");
             if (format_desc->channel[chan].normalized) {
                double scale = 1.0 / ((1 << (format_desc->channel[chan].size - 1)) - 1);
-               LLVMValueRef scale_val = lp_build_const_vec(type, scale);
+               LLVMValueRef scale_val = lp_build_const_vec(gallivm, type, scale);
                input = LLVMBuildFMul(builder, input, scale_val, "");
             }
          }
          else {
             /* FIXME */
             assert(0);
-            input = lp_build_undef(type);
+            input = lp_build_undef(gallivm, type);
          }
 
          break;
@@ -214,32 +216,32 @@ lp_build_unpack_rgba_soa(LLVMBuilderRef builder,
             assert(start == 0);
             assert(stop == 32);
             assert(type.width == 32);
-            input = LLVMBuildBitCast(builder, input, lp_build_vec_type(type), "");
+            input = LLVMBuildBitCast(builder, input, lp_build_vec_type(gallivm, type), "");
          }
          else {
             /* FIXME */
             assert(0);
-            input = lp_build_undef(type);
+            input = lp_build_undef(gallivm, type);
          }
          break;
 
       case UTIL_FORMAT_TYPE_FIXED:
          if (type.floating) {
             double scale = 1.0 / ((1 << (format_desc->channel[chan].size/2)) - 1);
-            LLVMValueRef scale_val = lp_build_const_vec(type, scale);
-            input = LLVMBuildSIToFP(builder, input, lp_build_vec_type(type), "");
+            LLVMValueRef scale_val = lp_build_const_vec(gallivm, type, scale);
+            input = LLVMBuildSIToFP(builder, input, lp_build_vec_type(gallivm, type), "");
             input = LLVMBuildFMul(builder, input, scale_val, "");
          }
          else {
             /* FIXME */
             assert(0);
-            input = lp_build_undef(type);
+            input = lp_build_undef(gallivm, type);
          }
          break;
 
       default:
          assert(0);
-         input = lp_build_undef(type);
+         input = lp_build_undef(gallivm, type);
          break;
       }
 
@@ -253,16 +255,17 @@ lp_build_unpack_rgba_soa(LLVMBuilderRef builder,
 
 
 void
-lp_build_rgba8_to_f32_soa(LLVMBuilderRef builder,
+lp_build_rgba8_to_f32_soa(struct gallivm_state *gallivm,
                           struct lp_type dst_type,
                           LLVMValueRef packed,
                           LLVMValueRef *rgba)
 {
-   LLVMValueRef mask = lp_build_const_int_vec(dst_type, 0xff);
+   LLVMBuilderRef builder = gallivm->builder;
+   LLVMValueRef mask = lp_build_const_int_vec(gallivm, dst_type, 0xff);
    unsigned chan;
 
    packed = LLVMBuildBitCast(builder, packed,
-                             lp_build_int_vec_type(dst_type), "");
+                             lp_build_int_vec_type(gallivm, dst_type), "");
 
    /* Decode the input vector components */
    for (chan = 0; chan < 4; ++chan) {
@@ -274,12 +277,12 @@ lp_build_rgba8_to_f32_soa(LLVMBuilderRef builder,
 
       if (start)
          input = LLVMBuildLShr(builder, input,
-                               lp_build_const_int_vec(dst_type, start), "");
+                               lp_build_const_int_vec(gallivm, dst_type, start), "");
 
       if (stop < 32)
          input = LLVMBuildAnd(builder, input, mask, "");
 
-      input = lp_build_unsigned_norm_to_float(builder, 8, dst_type, input);
+      input = lp_build_unsigned_norm_to_float(gallivm, 8, dst_type, input);
 
       rgba[chan] = input;
    }
@@ -303,7 +306,7 @@ lp_build_rgba8_to_f32_soa(LLVMBuilderRef builder,
  *              be in [0, block_width-1] and j will be in [0, block_height-1].
  */
 void
-lp_build_fetch_rgba_soa(LLVMBuilderRef builder,
+lp_build_fetch_rgba_soa(struct gallivm_state *gallivm,
                         const struct util_format_description *format_desc,
                         struct lp_type type,
                         LLVMValueRef base_ptr,
@@ -312,6 +315,7 @@ lp_build_fetch_rgba_soa(LLVMBuilderRef builder,
                         LLVMValueRef j,
                         LLVMValueRef rgba_out[4])
 {
+   LLVMBuilderRef builder = gallivm->builder;
 
    if (format_desc->layout == UTIL_FORMAT_LAYOUT_PLAIN &&
        (format_desc->colorspace == UTIL_FORMAT_COLORSPACE_RGB ||
@@ -334,7 +338,7 @@ lp_build_fetch_rgba_soa(LLVMBuilderRef builder,
        * gather the texels from the texture
        * Ex: packed = {BGRA, BGRA, BGRA, BGRA}.
        */
-      packed = lp_build_gather(builder,
+      packed = lp_build_gather(gallivm,
                                type.length,
                                format_desc->block.bits,
                                type.width,
@@ -343,7 +347,7 @@ lp_build_fetch_rgba_soa(LLVMBuilderRef builder,
       /*
        * convert texels to float rgba
        */
-      lp_build_unpack_rgba_soa(builder,
+      lp_build_unpack_rgba_soa(gallivm,
                                format_desc,
                                type,
                                packed, rgba_out);
@@ -364,10 +368,10 @@ lp_build_fetch_rgba_soa(LLVMBuilderRef builder,
       tmp_type.length = type.length * 4;
       tmp_type.norm = TRUE;
 
-      tmp = lp_build_fetch_rgba_aos(builder, format_desc, tmp_type,
+      tmp = lp_build_fetch_rgba_aos(gallivm, format_desc, tmp_type,
                                     base_ptr, offset, i, j);
 
-      lp_build_rgba8_to_f32_soa(builder,
+      lp_build_rgba8_to_f32_soa(gallivm,
                                 type,
                                 tmp,
                                 rgba_out);
@@ -397,23 +401,24 @@ lp_build_fetch_rgba_soa(LLVMBuilderRef builder,
       tmp_type.length = 4;
 
       for (chan = 0; chan < 4; ++chan) {
-         rgba_out[chan] = lp_build_undef(type);
+         rgba_out[chan] = lp_build_undef(gallivm, type);
       }
 
       /* loop over number of pixels */
       for(k = 0; k < type.length; ++k) {
-         LLVMValueRef index = LLVMConstInt(LLVMInt32Type(), k, 0);
+         LLVMValueRef index = lp_build_const_int32(gallivm, k);
          LLVMValueRef offset_elem;
          LLVMValueRef i_elem, j_elem;
          LLVMValueRef tmp;
 
-         offset_elem = LLVMBuildExtractElement(builder, offset, index, "");
+         offset_elem = LLVMBuildExtractElement(builder, offset,
+                                               index, "");
 
          i_elem = LLVMBuildExtractElement(builder, i, index, "");
          j_elem = LLVMBuildExtractElement(builder, j, index, "");
 
          /* Get a single float[4]={R,G,B,A} pixel */
-         tmp = lp_build_fetch_rgba_aos(builder, format_desc, tmp_type,
+         tmp = lp_build_fetch_rgba_aos(gallivm, format_desc, tmp_type,
                                        base_ptr, offset_elem,
                                        i_elem, j_elem);
 
@@ -422,7 +427,7 @@ lp_build_fetch_rgba_soa(LLVMBuilderRef builder,
           * position = 'index'.
           */
          for (chan = 0; chan < 4; ++chan) {
-            LLVMValueRef chan_val = LLVMConstInt(LLVMInt32Type(), chan, 0),
+            LLVMValueRef chan_val = lp_build_const_int32(gallivm, chan),
             tmp_chan = LLVMBuildExtractElement(builder, tmp, chan_val, "");
             rgba_out[chan] = LLVMBuildInsertElement(builder, rgba_out[chan],
                                                     tmp_chan, index, "");
diff --git a/src/gallium/auxiliary/gallivm/lp_bld_format_yuv.c b/src/gallium/auxiliary/gallivm/lp_bld_format_yuv.c
index 2bce289555..cdf1956c09 100644
--- a/src/gallium/auxiliary/gallivm/lp_bld_format_yuv.c
+++ b/src/gallium/auxiliary/gallivm/lp_bld_format_yuv.c
@@ -43,6 +43,7 @@
 #include "lp_bld_conv.h"
 #include "lp_bld_gather.h"
 #include "lp_bld_format.h"
+#include "lp_bld_init.h"
 #include "lp_bld_logic.h"
 
 /**
@@ -51,7 +52,7 @@
  * @param i  is a <n x i32> vector with the x pixel coordinate (0 or 1)
  */
 static void
-uyvy_to_yuv_soa(LLVMBuilderRef builder,
+uyvy_to_yuv_soa(struct gallivm_state *gallivm,
                 unsigned n,
                 LLVMValueRef packed,
                 LLVMValueRef i,
@@ -59,6 +60,7 @@ uyvy_to_yuv_soa(LLVMBuilderRef builder,
                 LLVMValueRef *u,
                 LLVMValueRef *v)
 {
+   LLVMBuilderRef builder = gallivm->builder;
    struct lp_type type;
    LLVMValueRef mask;
 
@@ -86,25 +88,25 @@ uyvy_to_yuv_soa(LLVMBuilderRef builder,
       LLVMValueRef sel, tmp, tmp2;
       struct lp_build_context bld32;
 
-      lp_build_context_init(&bld32, builder, type);
+      lp_build_context_init(&bld32, gallivm, type);
 
-      tmp = LLVMBuildLShr(builder, packed, lp_build_const_int_vec(type, 8), "");
-      tmp2 = LLVMBuildLShr(builder, tmp, lp_build_const_int_vec(type, 16), "");
-      sel = lp_build_compare(builder, type, PIPE_FUNC_EQUAL, i, lp_build_const_int_vec(type, 0));
+      tmp = LLVMBuildLShr(builder, packed, lp_build_const_int_vec(gallivm, type, 8), "");
+      tmp2 = LLVMBuildLShr(builder, tmp, lp_build_const_int_vec(gallivm, type, 16), "");
+      sel = lp_build_compare(gallivm, type, PIPE_FUNC_EQUAL, i, lp_build_const_int_vec(gallivm, type, 0));
       *y = lp_build_select(&bld32, sel, tmp, tmp2);
    } else
 #endif
    {
       LLVMValueRef shift;
-      shift = LLVMBuildMul(builder, i, lp_build_const_int_vec(type, 16), "");
-      shift = LLVMBuildAdd(builder, shift, lp_build_const_int_vec(type, 8), "");
+      shift = LLVMBuildMul(builder, i, lp_build_const_int_vec(gallivm, type, 16), "");
+      shift = LLVMBuildAdd(builder, shift, lp_build_const_int_vec(gallivm, type, 8), "");
       *y = LLVMBuildLShr(builder, packed, shift, "");
    }
 
    *u = packed;
-   *v = LLVMBuildLShr(builder, packed, lp_build_const_int_vec(type, 16), "");
+   *v = LLVMBuildLShr(builder, packed, lp_build_const_int_vec(gallivm, type, 16), "");
 
-   mask = lp_build_const_int_vec(type, 0xff);
+   mask = lp_build_const_int_vec(gallivm, type, 0xff);
 
    *y = LLVMBuildAnd(builder, *y, mask, "y");
    *u = LLVMBuildAnd(builder, *u, mask, "u");
@@ -118,7 +120,7 @@ uyvy_to_yuv_soa(LLVMBuilderRef builder,
  * @param i  is a <n x i32> vector with the x pixel coordinate (0 or 1)
  */
 static void
-yuyv_to_yuv_soa(LLVMBuilderRef builder,
+yuyv_to_yuv_soa(struct gallivm_state *gallivm,
                 unsigned n,
                 LLVMValueRef packed,
                 LLVMValueRef i,
@@ -126,6 +128,7 @@ yuyv_to_yuv_soa(LLVMBuilderRef builder,
                 LLVMValueRef *u,
                 LLVMValueRef *v)
 {
+   LLVMBuilderRef builder = gallivm->builder;
    struct lp_type type;
    LLVMValueRef mask;
 
@@ -153,23 +156,23 @@ yuyv_to_yuv_soa(LLVMBuilderRef builder,
       LLVMValueRef sel, tmp;
       struct lp_build_context bld32;
 
-      lp_build_context_init(&bld32, builder, type);
+      lp_build_context_init(&bld32, gallivm, type);
 
-      tmp = LLVMBuildLShr(builder, packed, lp_build_const_int_vec(type, 16), "");
-      sel = lp_build_compare(builder, type, PIPE_FUNC_EQUAL, i, lp_build_const_int_vec(type, 0));
+      tmp = LLVMBuildLShr(builder, packed, lp_build_const_int_vec(gallivm, type, 16), "");
+      sel = lp_build_compare(gallivm, type, PIPE_FUNC_EQUAL, i, lp_build_const_int_vec(gallivm, type, 0));
        *y = lp_build_select(&bld32, sel, packed, tmp);
    } else
 #endif
    {
       LLVMValueRef shift;
-      shift = LLVMBuildMul(builder, i, lp_build_const_int_vec(type, 16), "");
+      shift = LLVMBuildMul(builder, i, lp_build_const_int_vec(gallivm, type, 16), "");
       *y = LLVMBuildLShr(builder, packed, shift, "");
    }
 
-   *u = LLVMBuildLShr(builder, packed, lp_build_const_int_vec(type, 8), "");
-   *v = LLVMBuildLShr(builder, packed, lp_build_const_int_vec(type, 24), "");
+   *u = LLVMBuildLShr(builder, packed, lp_build_const_int_vec(gallivm, type, 8), "");
+   *v = LLVMBuildLShr(builder, packed, lp_build_const_int_vec(gallivm, type, 24), "");
 
-   mask = lp_build_const_int_vec(type, 0xff);
+   mask = lp_build_const_int_vec(gallivm, type, 0xff);
 
    *y = LLVMBuildAnd(builder, *y, mask, "y");
    *u = LLVMBuildAnd(builder, *u, mask, "u");
@@ -178,11 +181,12 @@ yuyv_to_yuv_soa(LLVMBuilderRef builder,
 
 
 static INLINE void
-yuv_to_rgb_soa(LLVMBuilderRef builder,
+yuv_to_rgb_soa(struct gallivm_state *gallivm,
                unsigned n,
                LLVMValueRef y, LLVMValueRef u, LLVMValueRef v,
                LLVMValueRef *r, LLVMValueRef *g, LLVMValueRef *b)
 {
+   LLVMBuilderRef builder = gallivm->builder;
    struct lp_type type;
    struct lp_build_context bld;
 
@@ -203,7 +207,7 @@ yuv_to_rgb_soa(LLVMBuilderRef builder,
    type.width = 32;
    type.length = n;
 
-   lp_build_context_init(&bld, builder, type);
+   lp_build_context_init(&bld, gallivm, type);
 
    assert(lp_check_value(type, y));
    assert(lp_check_value(type, u));
@@ -213,17 +217,17 @@ yuv_to_rgb_soa(LLVMBuilderRef builder,
     * Constants
     */
 
-   c0   = lp_build_const_int_vec(type,   0);
-   c8   = lp_build_const_int_vec(type,   8);
-   c16  = lp_build_const_int_vec(type,  16);
-   c128 = lp_build_const_int_vec(type, 128);
-   c255 = lp_build_const_int_vec(type, 255);
+   c0   = lp_build_const_int_vec(gallivm, type,   0);
+   c8   = lp_build_const_int_vec(gallivm, type,   8);
+   c16  = lp_build_const_int_vec(gallivm, type,  16);
+   c128 = lp_build_const_int_vec(gallivm, type, 128);
+   c255 = lp_build_const_int_vec(gallivm, type, 255);
 
-   cy  = lp_build_const_int_vec(type,  298);
-   cug = lp_build_const_int_vec(type, -100);
-   cub = lp_build_const_int_vec(type,  516);
-   cvr = lp_build_const_int_vec(type,  409);
-   cvg = lp_build_const_int_vec(type, -208);
+   cy  = lp_build_const_int_vec(gallivm, type,  298);
+   cug = lp_build_const_int_vec(gallivm, type, -100);
+   cub = lp_build_const_int_vec(gallivm, type,  516);
+   cvr = lp_build_const_int_vec(gallivm, type,  409);
+   cvg = lp_build_const_int_vec(gallivm, type, -208);
 
    /*
     *  y -= 16;
@@ -276,10 +280,11 @@ yuv_to_rgb_soa(LLVMBuilderRef builder,
 
 
 static LLVMValueRef
-rgb_to_rgba_aos(LLVMBuilderRef builder,
+rgb_to_rgba_aos(struct gallivm_state *gallivm,
                 unsigned n,
                 LLVMValueRef r, LLVMValueRef g, LLVMValueRef b)
 {
+   LLVMBuilderRef builder = gallivm->builder;
    struct lp_type type;
    LLVMValueRef a;
    LLVMValueRef rgba;
@@ -298,9 +303,9 @@ rgb_to_rgba_aos(LLVMBuilderRef builder,
     */
 
    r = r;
-   g = LLVMBuildShl(builder, g, lp_build_const_int_vec(type, 8), "");
-   b = LLVMBuildShl(builder, b, lp_build_const_int_vec(type, 16), "");
-   a = lp_build_const_int_vec(type, 0xff000000);
+   g = LLVMBuildShl(builder, g, lp_build_const_int_vec(gallivm, type, 8), "");
+   b = LLVMBuildShl(builder, b, lp_build_const_int_vec(gallivm, type, 16), "");
+   a = lp_build_const_int_vec(gallivm, type, 0xff000000);
 
    rgba = r;
    rgba = LLVMBuildOr(builder, rgba, g, "");
@@ -308,7 +313,7 @@ rgb_to_rgba_aos(LLVMBuilderRef builder,
    rgba = LLVMBuildOr(builder, rgba, a, "");
 
    rgba = LLVMBuildBitCast(builder, rgba,
-                           LLVMVectorType(LLVMInt8Type(), 4*n), "");
+                           LLVMVectorType(LLVMInt8TypeInContext(gallivm->context), 4*n), "");
 
    return rgba;
 }
@@ -318,7 +323,7 @@ rgb_to_rgba_aos(LLVMBuilderRef builder,
  * Convert from <n x i32> packed UYVY to <4n x i8> RGBA AoS
  */
 static LLVMValueRef
-uyvy_to_rgba_aos(LLVMBuilderRef builder,
+uyvy_to_rgba_aos(struct gallivm_state *gallivm,
                  unsigned n,
                  LLVMValueRef packed,
                  LLVMValueRef i)
@@ -327,9 +332,9 @@ uyvy_to_rgba_aos(LLVMBuilderRef builder,
    LLVMValueRef r, g, b;
    LLVMValueRef rgba;
 
-   uyvy_to_yuv_soa(builder, n, packed, i, &y, &u, &v);
-   yuv_to_rgb_soa(builder, n, y, u, v, &r, &g, &b);
-   rgba = rgb_to_rgba_aos(builder, n, r, g, b);
+   uyvy_to_yuv_soa(gallivm, n, packed, i, &y, &u, &v);
+   yuv_to_rgb_soa(gallivm, n, y, u, v, &r, &g, &b);
+   rgba = rgb_to_rgba_aos(gallivm, n, r, g, b);
 
    return rgba;
 }
@@ -339,7 +344,7 @@ uyvy_to_rgba_aos(LLVMBuilderRef builder,
  * Convert from <n x i32> packed YUYV to <4n x i8> RGBA AoS
  */
 static LLVMValueRef
-yuyv_to_rgba_aos(LLVMBuilderRef builder,
+yuyv_to_rgba_aos(struct gallivm_state *gallivm,
                  unsigned n,
                  LLVMValueRef packed,
                  LLVMValueRef i)
@@ -348,9 +353,9 @@ yuyv_to_rgba_aos(LLVMBuilderRef builder,
    LLVMValueRef r, g, b;
    LLVMValueRef rgba;
 
-   yuyv_to_yuv_soa(builder, n, packed, i, &y, &u, &v);
-   yuv_to_rgb_soa(builder, n, y, u, v, &r, &g, &b);
-   rgba = rgb_to_rgba_aos(builder, n, r, g, b);
+   yuyv_to_yuv_soa(gallivm, n, packed, i, &y, &u, &v);
+   yuv_to_rgb_soa(gallivm, n, y, u, v, &r, &g, &b);
+   rgba = rgb_to_rgba_aos(gallivm, n, r, g, b);
 
    return rgba;
 }
@@ -360,7 +365,7 @@ yuyv_to_rgba_aos(LLVMBuilderRef builder,
  * Convert from <n x i32> packed RG_BG to <4n x i8> RGBA AoS
  */
 static LLVMValueRef
-rgbg_to_rgba_aos(LLVMBuilderRef builder,
+rgbg_to_rgba_aos(struct gallivm_state *gallivm,
                  unsigned n,
                  LLVMValueRef packed,
                  LLVMValueRef i)
@@ -368,8 +373,8 @@ rgbg_to_rgba_aos(LLVMBuilderRef builder,
    LLVMValueRef r, g, b;
    LLVMValueRef rgba;
 
-   uyvy_to_yuv_soa(builder, n, packed, i, &g, &r, &b);
-   rgba = rgb_to_rgba_aos(builder, n, r, g, b);
+   uyvy_to_yuv_soa(gallivm, n, packed, i, &g, &r, &b);
+   rgba = rgb_to_rgba_aos(gallivm, n, r, g, b);
 
    return rgba;
 }
@@ -379,7 +384,7 @@ rgbg_to_rgba_aos(LLVMBuilderRef builder,
  * Convert from <n x i32> packed GR_GB to <4n x i8> RGBA AoS
  */
 static LLVMValueRef
-grgb_to_rgba_aos(LLVMBuilderRef builder,
+grgb_to_rgba_aos(struct gallivm_state *gallivm,
                  unsigned n,
                  LLVMValueRef packed,
                  LLVMValueRef i)
@@ -387,8 +392,8 @@ grgb_to_rgba_aos(LLVMBuilderRef builder,
    LLVMValueRef r, g, b;
    LLVMValueRef rgba;
 
-   yuyv_to_yuv_soa(builder, n, packed, i, &g, &r, &b);
-   rgba = rgb_to_rgba_aos(builder, n, r, g, b);
+   yuyv_to_yuv_soa(gallivm, n, packed, i, &g, &r, &b);
+   rgba = rgb_to_rgba_aos(gallivm, n, r, g, b);
 
    return rgba;
 }
@@ -401,7 +406,7 @@ grgb_to_rgba_aos(LLVMBuilderRef builder,
  * @return  a <4*n x i8> vector with the pixel RGBA values in AoS
  */
 LLVMValueRef
-lp_build_fetch_subsampled_rgba_aos(LLVMBuilderRef builder,
+lp_build_fetch_subsampled_rgba_aos(struct gallivm_state *gallivm,
                                    const struct util_format_description *format_desc,
                                    unsigned n,
                                    LLVMValueRef base_ptr,
@@ -417,26 +422,26 @@ lp_build_fetch_subsampled_rgba_aos(LLVMBuilderRef builder,
    assert(format_desc->block.width == 2);
    assert(format_desc->block.height == 1);
 
-   packed = lp_build_gather(builder, n, 32, 32, base_ptr, offset);
+   packed = lp_build_gather(gallivm, n, 32, 32, base_ptr, offset);
 
    (void)j;
 
    switch (format_desc->format) {
    case PIPE_FORMAT_UYVY:
-      rgba = uyvy_to_rgba_aos(builder, n, packed, i);
+      rgba = uyvy_to_rgba_aos(gallivm, n, packed, i);
       break;
    case PIPE_FORMAT_YUYV:
-      rgba = yuyv_to_rgba_aos(builder, n, packed, i);
+      rgba = yuyv_to_rgba_aos(gallivm, n, packed, i);
       break;
    case PIPE_FORMAT_R8G8_B8G8_UNORM:
-      rgba = rgbg_to_rgba_aos(builder, n, packed, i);
+      rgba = rgbg_to_rgba_aos(gallivm, n, packed, i);
       break;
    case PIPE_FORMAT_G8R8_G8B8_UNORM:
-      rgba = grgb_to_rgba_aos(builder, n, packed, i);
+      rgba = grgb_to_rgba_aos(gallivm, n, packed, i);
       break;
    default:
       assert(0);
-      rgba =  LLVMGetUndef(LLVMVectorType(LLVMInt8Type(), 4*n));
+      rgba =  LLVMGetUndef(LLVMVectorType(LLVMInt8TypeInContext(gallivm->context), 4*n));
       break;
    }
 
diff --git a/src/gallium/auxiliary/gallivm/lp_bld_gather.c b/src/gallium/auxiliary/gallivm/lp_bld_gather.c
index d60472e065..0dc81b1abb 100644
--- a/src/gallium/auxiliary/gallivm/lp_bld_gather.c
+++ b/src/gallium/auxiliary/gallivm/lp_bld_gather.c
@@ -31,6 +31,7 @@
 #include "lp_bld_const.h"
 #include "lp_bld_format.h"
 #include "lp_bld_gather.h"
+#include "lp_bld_init.h"
 
 
 /**
@@ -39,7 +40,7 @@
  * @sa lp_build_gather()
  */
 LLVMValueRef
-lp_build_gather_elem_ptr(LLVMBuilderRef builder,
+lp_build_gather_elem_ptr(struct gallivm_state *gallivm,
                          unsigned length,
                          LLVMValueRef base_ptr,
                          LLVMValueRef offsets,
@@ -48,17 +49,17 @@ lp_build_gather_elem_ptr(LLVMBuilderRef builder,
    LLVMValueRef offset;
    LLVMValueRef ptr;
 
-   assert(LLVMTypeOf(base_ptr) == LLVMPointerType(LLVMInt8Type(), 0));
+   assert(LLVMTypeOf(base_ptr) == LLVMPointerType(LLVMInt8TypeInContext(gallivm->context), 0));
 
    if (length == 1) {
       assert(i == 0);
       offset = offsets;
    } else {
-      LLVMValueRef index = LLVMConstInt(LLVMInt32Type(), i, 0);
-      offset = LLVMBuildExtractElement(builder, offsets, index, "");
+      LLVMValueRef index = lp_build_const_int32(gallivm, i);
+      offset = LLVMBuildExtractElement(gallivm->builder, offsets, index, "");
    }
 
-   ptr = LLVMBuildGEP(builder, base_ptr, &offset, 1, "");
+   ptr = LLVMBuildGEP(gallivm->builder, base_ptr, &offset, 1, "");
 
    return ptr;
 }
@@ -70,7 +71,7 @@ lp_build_gather_elem_ptr(LLVMBuilderRef builder,
  * @sa lp_build_gather()
  */
 LLVMValueRef
-lp_build_gather_elem(LLVMBuilderRef builder,
+lp_build_gather_elem(struct gallivm_state *gallivm,
                      unsigned length,
                      unsigned src_width,
                      unsigned dst_width,
@@ -78,23 +79,23 @@ lp_build_gather_elem(LLVMBuilderRef builder,
                      LLVMValueRef offsets,
                      unsigned i)
 {
-   LLVMTypeRef src_type = LLVMIntType(src_width);
+   LLVMTypeRef src_type = LLVMIntTypeInContext(gallivm->context, src_width);
    LLVMTypeRef src_ptr_type = LLVMPointerType(src_type, 0);
-   LLVMTypeRef dst_elem_type = LLVMIntType(dst_width);
+   LLVMTypeRef dst_elem_type = LLVMIntTypeInContext(gallivm->context, dst_width);
    LLVMValueRef ptr;
    LLVMValueRef res;
 
-   assert(LLVMTypeOf(base_ptr) == LLVMPointerType(LLVMInt8Type(), 0));
+   assert(LLVMTypeOf(base_ptr) == LLVMPointerType(LLVMInt8TypeInContext(gallivm->context), 0));
 
-   ptr = lp_build_gather_elem_ptr(builder, length, base_ptr, offsets, i);
-   ptr = LLVMBuildBitCast(builder, ptr, src_ptr_type, "");
-   res = LLVMBuildLoad(builder, ptr, "");
+   ptr = lp_build_gather_elem_ptr(gallivm, length, base_ptr, offsets, i);
+   ptr = LLVMBuildBitCast(gallivm->builder, ptr, src_ptr_type, "");
+   res = LLVMBuildLoad(gallivm->builder, ptr, "");
 
    assert(src_width <= dst_width);
    if (src_width > dst_width)
-      res = LLVMBuildTrunc(builder, res, dst_elem_type, "");
+      res = LLVMBuildTrunc(gallivm->builder, res, dst_elem_type, "");
    if (src_width < dst_width)
-      res = LLVMBuildZExt(builder, res, dst_elem_type, "");
+      res = LLVMBuildZExt(gallivm->builder, res, dst_elem_type, "");
 
    return res;
 }
@@ -112,7 +113,7 @@ lp_build_gather_elem(LLVMBuilderRef builder,
  * @param offsets vector with offsets
  */
 LLVMValueRef
-lp_build_gather(LLVMBuilderRef builder,
+lp_build_gather(struct gallivm_state *gallivm,
                 unsigned length,
                 unsigned src_width,
                 unsigned dst_width,
@@ -123,24 +124,24 @@ lp_build_gather(LLVMBuilderRef builder,
 
    if (length == 1) {
       /* Scalar */
-      return lp_build_gather_elem(builder, length,
+      return lp_build_gather_elem(gallivm, length,
                                   src_width, dst_width,
                                   base_ptr, offsets, 0);
    } else {
       /* Vector */
 
-      LLVMTypeRef dst_elem_type = LLVMIntType(dst_width);
+      LLVMTypeRef dst_elem_type = LLVMIntTypeInContext(gallivm->context, dst_width);
       LLVMTypeRef dst_vec_type = LLVMVectorType(dst_elem_type, length);
       unsigned i;
 
       res = LLVMGetUndef(dst_vec_type);
       for (i = 0; i < length; ++i) {
-         LLVMValueRef index = LLVMConstInt(LLVMInt32Type(), i, 0);
+         LLVMValueRef index = lp_build_const_int32(gallivm, i);
          LLVMValueRef elem;
-         elem = lp_build_gather_elem(builder, length,
+         elem = lp_build_gather_elem(gallivm, length,
                                      src_width, dst_width,
                                      base_ptr, offsets, i);
-         res = LLVMBuildInsertElement(builder, res, elem, index, "");
+         res = LLVMBuildInsertElement(gallivm->builder, res, elem, index, "");
       }
    }
 
diff --git a/src/gallium/auxiliary/gallivm/lp_bld_gather.h b/src/gallium/auxiliary/gallivm/lp_bld_gather.h
index 131af8ea07..5b04131730 100644
--- a/src/gallium/auxiliary/gallivm/lp_bld_gather.h
+++ b/src/gallium/auxiliary/gallivm/lp_bld_gather.h
@@ -34,14 +34,14 @@
 
 
 LLVMValueRef
-lp_build_gather_elem_ptr(LLVMBuilderRef builder,
+lp_build_gather_elem_ptr(struct gallivm_state *gallivm,
                          unsigned length,
                          LLVMValueRef base_ptr,
                          LLVMValueRef offsets,
                          unsigned i);
 
 LLVMValueRef
-lp_build_gather_elem(LLVMBuilderRef builder,
+lp_build_gather_elem(struct gallivm_state *gallivm,
                      unsigned length,
                      unsigned src_width,
                      unsigned dst_width,
@@ -50,7 +50,7 @@ lp_build_gather_elem(LLVMBuilderRef builder,
                      unsigned i);
 
 LLVMValueRef
-lp_build_gather(LLVMBuilderRef builder,
+lp_build_gather(struct gallivm_state *gallivm,
                 unsigned length,
                 unsigned src_width,
                 unsigned dst_width,
diff --git a/src/gallium/auxiliary/gallivm/lp_bld_init.c b/src/gallium/auxiliary/gallivm/lp_bld_init.c
index 0b9a6f745f..7504cb5cf2 100644
--- a/src/gallium/auxiliary/gallivm/lp_bld_init.c
+++ b/src/gallium/auxiliary/gallivm/lp_bld_init.c
@@ -29,6 +29,8 @@
 #include "pipe/p_compiler.h"
 #include "util/u_cpu_detect.h"
 #include "util/u_debug.h"
+#include "util/u_memory.h"
+#include "util/u_simple_list.h"
 #include "lp_bld_debug.h"
 #include "lp_bld_init.h"
 
@@ -45,6 +47,7 @@ static const struct debug_named_value lp_bld_debug_flags[] = {
    { "nopt",   GALLIVM_DEBUG_NO_OPT, NULL },
    { "perf",   GALLIVM_DEBUG_PERF, NULL },
    { "no_brilinear", GALLIVM_DEBUG_NO_BRILINEAR, NULL },
+   { "gc",     GALLIVM_DEBUG_GC, NULL },
    DEBUG_NAMED_VALUE_END
 };
 
@@ -52,11 +55,7 @@ DEBUG_GET_ONCE_FLAGS_OPTION(gallivm_debug, "GALLIVM_DEBUG", lp_bld_debug_flags,
 #endif
 
 
-LLVMModuleRef lp_build_module = NULL;
-LLVMExecutionEngineRef lp_build_engine = NULL;
-LLVMModuleProviderRef lp_build_provider = NULL;
-LLVMTargetDataRef lp_build_target = NULL;
-LLVMPassManagerRef lp_build_pass = NULL;
+static boolean gallivm_initialized = FALSE;
 
 
 /*
@@ -82,6 +81,19 @@ enum LLVM_CodeGenOpt_Level {
 };
 
 
+/**
+ * LLVM 2.6 permits only one ExecutionEngine to be created.  This is it.
+ */
+static LLVMExecutionEngineRef GlobalEngine = NULL;
+
+/**
+ * Same gallivm state shared by all contexts.
+ */
+static struct gallivm_state *GlobalGallivm = NULL;
+
+
+
+
 extern void
 lp_register_oprofile_jit_event_listener(LLVMExecutionEngineRef EE);
 
@@ -89,26 +101,148 @@ extern void
 lp_set_target_options(void);
 
 
-void
-lp_build_init(void)
+
+/**
+ * Create the LLVM (optimization) pass manager and install
+ * relevant optimization passes.
+ * \return  TRUE for success, FALSE for failure
+ */
+static boolean
+create_pass_manager(struct gallivm_state *gallivm)
 {
-#ifdef DEBUG
-   gallivm_debug = debug_get_option_gallivm_debug();
+   assert(!gallivm->passmgr);
+
+   gallivm->passmgr = LLVMCreateFunctionPassManager(gallivm->provider);
+   if (!gallivm->passmgr)
+      return FALSE;
+
+   LLVMAddTargetData(gallivm->target, gallivm->passmgr);
+
+   if ((gallivm_debug & GALLIVM_DEBUG_NO_OPT) == 0) {
+      /* These are the passes currently listed in llvm-c/Transforms/Scalar.h,
+       * but there are more on SVN.
+       * TODO: Add more passes.
+       */
+      LLVMAddCFGSimplificationPass(gallivm->passmgr);
+
+      if (HAVE_LLVM >= 0x207 && sizeof(void*) == 4) {
+         /* For LLVM >= 2.7 and 32-bit build, use this order of passes to
+          * avoid generating bad code.
+          * Test with piglit glsl-vs-sqrt-zero test.
+          */
+         LLVMAddConstantPropagationPass(gallivm->passmgr);
+         LLVMAddPromoteMemoryToRegisterPass(gallivm->passmgr);
+      }
+      else {
+         LLVMAddPromoteMemoryToRegisterPass(gallivm->passmgr);
+         LLVMAddConstantPropagationPass(gallivm->passmgr);
+      }
+
+      if (util_cpu_caps.has_sse4_1) {
+         /* FIXME: There is a bug in this pass, whereby the combination
+          * of fptosi and sitofp (necessary for trunc/floor/ceil/round
+          * implementation) somehow becomes invalid code.
+          */
+         LLVMAddInstructionCombiningPass(gallivm->passmgr);
+      }
+      LLVMAddGVNPass(gallivm->passmgr);
+   }
+   else {
+      /* We need at least this pass to prevent the backends to fail in
+       * unexpected ways.
+       */
+      LLVMAddPromoteMemoryToRegisterPass(gallivm->passmgr);
+   }
+
+   return TRUE;
+}
+
+
+/**
+ * Free gallivm object's LLVM allocations, but not the gallivm object itself.
+ */
+static void
+free_gallivm_state(struct gallivm_state *gallivm)
+{
+#if HAVE_LLVM >= 0x207 /* XXX or 0x208? */
+   /* This leads to crashes w/ some versions of LLVM */
+   LLVMModuleRef mod;
+   char *error;
+
+   if (gallivm->engine && gallivm->provider)
+      LLVMRemoveModuleProvider(gallivm->engine, gallivm->provider,
+                               &mod, &error);
 #endif
 
-   lp_set_target_options();
+#if 0
+   /* XXX this seems to crash with all versions of LLVM */
+   if (gallivm->provider)
+      LLVMDisposeModuleProvider(gallivm->provider);
+#endif
 
-   LLVMInitializeNativeTarget();
+   if (gallivm->passmgr)
+      LLVMDisposePassManager(gallivm->passmgr);
 
-   LLVMLinkInJIT();
+#if HAVE_LLVM >= 0x207
+   if (gallivm->module)
+      LLVMDisposeModule(gallivm->module);
+#endif
+
+#if 0
+   /* Don't free the exec engine, it's a global/singleton */
+   if (gallivm->engine)
+      LLVMDisposeExecutionEngine(gallivm->engine);
+#endif
+
+#if 0
+   /* Don't free the TargetData, it's owned by the exec engine */
+   LLVMDisposeTargetData(gallivm->target);
+#endif
+
+   if (gallivm->context)
+      LLVMContextDispose(gallivm->context);
 
-   if (!lp_build_module)
-      lp_build_module = LLVMModuleCreateWithName("gallivm");
+   if (gallivm->builder)
+      LLVMDisposeBuilder(gallivm->builder);
+
+   gallivm->engine = NULL;
+   gallivm->target = NULL;
+   gallivm->module = NULL;
+   gallivm->provider = NULL;
+   gallivm->passmgr = NULL;
+   gallivm->context = NULL;
+   gallivm->builder = NULL;
+}
 
-   if (!lp_build_provider)
-      lp_build_provider = LLVMCreateModuleProviderForExistingModule(lp_build_module);
 
-   if (!lp_build_engine) {
+/**
+ * Allocate gallivm LLVM objects.
+ * \return  TRUE for success, FALSE for failure
+ */
+static boolean
+init_gallivm_state(struct gallivm_state *gallivm)
+{
+   assert(gallivm_initialized);
+   assert(!gallivm->context);
+   assert(!gallivm->module);
+   assert(!gallivm->provider);
+
+   gallivm->context = LLVMContextCreate();
+   if (!gallivm->context)
+      goto fail;
+
+   gallivm->module = LLVMModuleCreateWithNameInContext("gallivm",
+                                                       gallivm->context);
+   if (!gallivm->module)
+      goto fail;
+
+   gallivm->provider =
+      LLVMCreateModuleProviderForExistingModule(gallivm->module);
+   if (!gallivm->provider)
+      goto fail;
+
+   if (!GlobalEngine) {
+      /* We can only create one LLVMExecutionEngine (w/ LLVM 2.6 anyway) */
       enum LLVM_CodeGenOpt_Level optlevel;
       char *error = NULL;
 
@@ -119,43 +253,153 @@ lp_build_init(void)
          optlevel = Default;
       }
 
-      if (LLVMCreateJITCompiler(&lp_build_engine, lp_build_provider,
-                                (unsigned)optlevel, &error)) {
+      if (LLVMCreateJITCompiler(&GlobalEngine, gallivm->provider,
+                                (unsigned) optlevel, &error)) {
          _debug_printf("%s\n", error);
          LLVMDisposeMessage(error);
-         assert(0);
+         goto fail;
       }
 
 #if defined(DEBUG) || defined(PROFILE)
-      lp_register_oprofile_jit_event_listener(lp_build_engine);
+      lp_register_oprofile_jit_event_listener(GlobalEngine);
 #endif
    }
 
-   if (!lp_build_target)
-      lp_build_target = LLVMGetExecutionEngineTargetData(lp_build_engine);
-
-   if (!lp_build_pass) {
-      lp_build_pass = LLVMCreateFunctionPassManager(lp_build_provider);
-      LLVMAddTargetData(lp_build_target, lp_build_pass);
-
-      if ((gallivm_debug & GALLIVM_DEBUG_NO_OPT) == 0) {
-         /* These are the passes currently listed in llvm-c/Transforms/Scalar.h,
-          * but there are more on SVN. */
-         /* TODO: Add more passes */
-         LLVMAddCFGSimplificationPass(lp_build_pass);
-         LLVMAddPromoteMemoryToRegisterPass(lp_build_pass);
-         LLVMAddConstantPropagationPass(lp_build_pass);
-         LLVMAddInstructionCombiningPass(lp_build_pass);
-         LLVMAddGVNPass(lp_build_pass);
-      } else {
-         /* We need at least this pass to prevent the backends to fail in
-          * unexpected ways.
-          */
-         LLVMAddPromoteMemoryToRegisterPass(lp_build_pass);
+   gallivm->engine = GlobalEngine;
+
+   LLVMAddModuleProvider(gallivm->engine, gallivm->provider);//new
+
+   gallivm->target = LLVMGetExecutionEngineTargetData(gallivm->engine);
+   if (!gallivm->target)
+      goto fail;
+
+   if (!create_pass_manager(gallivm))
+      goto fail;
+
+   gallivm->builder = LLVMCreateBuilderInContext(gallivm->context);
+   if (!gallivm->builder)
+      goto fail;
+
+   return TRUE;
+
+fail:
+   free_gallivm_state(gallivm);
+   return FALSE;
+}
+
+
+struct callback
+{
+   garbage_collect_callback_func func;
+   void *cb_data;
+   struct callback *prev, *next;
+};
+
+
+/** list of all garbage collector callbacks */
+static struct callback callback_list = {NULL, NULL, NULL, NULL};
+
+
+/**
+ * Register a function with gallivm which will be called when we
+ * do garbage collection.
+ */
+void
+gallivm_register_garbage_collector_callback(garbage_collect_callback_func func,
+                                            void *cb_data)
+{
+   struct callback *cb;
+
+   if (!callback_list.prev) {
+      make_empty_list(&callback_list);
+   }
+
+   /* see if already in list */
+   foreach(cb, &callback_list) {
+      if (cb->func == func && cb->cb_data == cb_data)
+         return;
+   }
+
+   /* add to list */
+   cb = CALLOC_STRUCT(callback);
+   if (cb) {
+      cb->func = func;
+      cb->cb_data = cb_data;
+      insert_at_head(&callback_list, cb);
+   }
+}
+
+
+/**
+ * Remove a callback.
+ */
+void
+gallivm_remove_garbage_collector_callback(garbage_collect_callback_func func,
+                                          void *cb_data)
+{
+   struct callback *cb;
+
+   /* search list */
+   foreach(cb, &callback_list) {
+      if (cb->func == func && cb->cb_data == cb_data) {
+         /* found, remove it */
+         remove_from_list(cb);
+         return;
       }
    }
+}
+
+
+/**
+ * Call the callback functions (which are typically in the
+ * draw module and llvmpipe driver.
+ */
+static void
+call_garbage_collector_callbacks(void)
+{
+   struct callback *cb;
+   foreach(cb, &callback_list) {
+      cb->func(cb->cb_data);
+   }
+}
+
+
+
+/**
+ * Other gallium components using gallivm should call this periodically
+ * to let us do garbage collection (or at least try to free memory
+ * accumulated by the LLVM libraries).
+ */
+void
+gallivm_garbage_collect(struct gallivm_state *gallivm)
+{
+   if (gallivm->context) {
+      if (gallivm_debug & GALLIVM_DEBUG_GC)
+         debug_printf("***** Doing LLVM garbage collection\n");
+
+      call_garbage_collector_callbacks();
+      free_gallivm_state(gallivm);
+      init_gallivm_state(gallivm);
+   }
+}
+
+
+void
+lp_build_init(void)
+{
+#ifdef DEBUG
+   gallivm_debug = debug_get_option_gallivm_debug();
+#endif
+
+   lp_set_target_options();
+
+   LLVMInitializeNativeTarget();
+
+   LLVMLinkInJIT();
 
    util_cpu_detect();
+ 
+   gallivm_initialized = TRUE;
 
 #if 0
    /* For simulating less capable machines */
@@ -166,6 +410,39 @@ lp_build_init(void)
 }
 
 
+
+/**
+ * Create a new gallivm_state object.
+ * Note that we return a singleton.
+ */
+struct gallivm_state *
+gallivm_create(void)
+{
+   if (!GlobalGallivm) {
+      GlobalGallivm = CALLOC_STRUCT(gallivm_state);
+      if (GlobalGallivm) {
+         if (!init_gallivm_state(GlobalGallivm)) {
+            FREE(GlobalGallivm);
+            GlobalGallivm = NULL;
+         }
+      }
+   }
+   return GlobalGallivm;
+}
+
+
+/**
+ * Destroy a gallivm_state object.
+ */
+void
+gallivm_destroy(struct gallivm_state *gallivm)
+{
+   /* No-op: don't destroy the singleton */
+   (void) gallivm;
+}
+
+
+
 /* 
  * Hack to allow the linking of release LLVM static libraries on a debug build.
  *
diff --git a/src/gallium/auxiliary/gallivm/lp_bld_init.h b/src/gallium/auxiliary/gallivm/lp_bld_init.h
index 0b4b1ca7d1..f68bf75a85 100644
--- a/src/gallium/auxiliary/gallivm/lp_bld_init.h
+++ b/src/gallium/auxiliary/gallivm/lp_bld_init.h
@@ -30,24 +30,53 @@
 #define LP_BLD_INIT_H
 
 
+#include "pipe/p_compiler.h"
 #include "lp_bld.h"
 #include <llvm-c/ExecutionEngine.h>
 
 
-extern LLVMModuleRef lp_build_module;
-extern LLVMExecutionEngineRef lp_build_engine;
-extern LLVMModuleProviderRef lp_build_provider;
-extern LLVMTargetDataRef lp_build_target;
-extern LLVMPassManagerRef lp_build_pass;
+struct gallivm_state
+{
+   LLVMModuleRef module;
+   LLVMExecutionEngineRef engine;
+   LLVMModuleProviderRef provider;
+   LLVMTargetDataRef target;
+   LLVMPassManagerRef passmgr;
+   LLVMContextRef context;
+   LLVMBuilderRef builder;
+};
 
 
 void
 lp_build_init(void);
 
+
 extern void
 lp_func_delete_body(LLVMValueRef func);
 
 
+void
+gallivm_garbage_collect(struct gallivm_state *gallivm);
+
+
+typedef void (*garbage_collect_callback_func)(void *cb_data);
+
+void
+gallivm_register_garbage_collector_callback(garbage_collect_callback_func func,
+                                            void *cb_data);
+
+void
+gallivm_remove_garbage_collector_callback(garbage_collect_callback_func func,
+                                          void *cb_data);
+
+
+struct gallivm_state *
+gallivm_create(void);
+
+void
+gallivm_destroy(struct gallivm_state *gallivm);
+
+
 extern LLVMValueRef
 lp_build_load_volatile(LLVMBuilderRef B, LLVMValueRef PointerVal,
                        const char *Name);
diff --git a/src/gallium/auxiliary/gallivm/lp_bld_intr.c b/src/gallium/auxiliary/gallivm/lp_bld_intr.c
index 9895749d56..518a01fdb9 100644
--- a/src/gallium/auxiliary/gallivm/lp_bld_intr.c
+++ b/src/gallium/auxiliary/gallivm/lp_bld_intr.c
@@ -46,6 +46,7 @@
 
 #include "util/u_debug.h"
 
+#include "lp_bld_const.h"
 #include "lp_bld_intr.h"
 
 
@@ -136,12 +137,13 @@ lp_build_intrinsic_binary(LLVMBuilderRef builder,
 
 
 LLVMValueRef
-lp_build_intrinsic_map(LLVMBuilderRef builder,
+lp_build_intrinsic_map(struct gallivm_state *gallivm,
                        const char *name,
                        LLVMTypeRef ret_type,
                        LLVMValueRef *args,
                        unsigned num_args)
 {
+   LLVMBuilderRef builder = gallivm->builder;
    LLVMTypeRef ret_elem_type = LLVMGetElementType(ret_type);
    unsigned n = LLVMGetVectorSize(ret_type);
    unsigned i, j;
@@ -151,7 +153,7 @@ lp_build_intrinsic_map(LLVMBuilderRef builder,
 
    res = LLVMGetUndef(ret_type);
    for(i = 0; i < n; ++i) {
-      LLVMValueRef index = LLVMConstInt(LLVMInt32Type(), i, 0);
+      LLVMValueRef index = lp_build_const_int32(gallivm, i);
       LLVMValueRef arg_elems[LP_MAX_FUNC_ARGS];
       LLVMValueRef res_elem;
       for(j = 0; j < num_args; ++j)
@@ -165,17 +167,17 @@ lp_build_intrinsic_map(LLVMBuilderRef builder,
 
 
 LLVMValueRef
-lp_build_intrinsic_map_unary(LLVMBuilderRef builder,
+lp_build_intrinsic_map_unary(struct gallivm_state *gallivm,
                              const char *name,
                              LLVMTypeRef ret_type,
                              LLVMValueRef a)
 {
-   return lp_build_intrinsic_map(builder, name, ret_type, &a, 1);
+   return lp_build_intrinsic_map(gallivm, name, ret_type, &a, 1);
 }
 
 
 LLVMValueRef
-lp_build_intrinsic_map_binary(LLVMBuilderRef builder,
+lp_build_intrinsic_map_binary(struct gallivm_state *gallivm,
                               const char *name,
                               LLVMTypeRef ret_type,
                               LLVMValueRef a,
@@ -186,7 +188,7 @@ lp_build_intrinsic_map_binary(LLVMBuilderRef builder,
    args[0] = a;
    args[1] = b;
 
-   return lp_build_intrinsic_map(builder, name, ret_type, args, 2);
+   return lp_build_intrinsic_map(gallivm, name, ret_type, args, 2);
 }
 
 
diff --git a/src/gallium/auxiliary/gallivm/lp_bld_intr.h b/src/gallium/auxiliary/gallivm/lp_bld_intr.h
index 977f767322..b73dd70036 100644
--- a/src/gallium/auxiliary/gallivm/lp_bld_intr.h
+++ b/src/gallium/auxiliary/gallivm/lp_bld_intr.h
@@ -38,6 +38,7 @@
 
 
 #include "gallivm/lp_bld.h"
+#include "gallivm/lp_bld_init.h"
 
 
 /**
@@ -77,7 +78,7 @@ lp_build_intrinsic_binary(LLVMBuilderRef builder,
 
 
 LLVMValueRef
-lp_build_intrinsic_map(LLVMBuilderRef builder,
+lp_build_intrinsic_map(struct gallivm_state *gallivm,
                        const char *name,
                        LLVMTypeRef ret_type,
                        LLVMValueRef *args,
@@ -85,14 +86,14 @@ lp_build_intrinsic_map(LLVMBuilderRef builder,
 
 
 LLVMValueRef
-lp_build_intrinsic_map_unary(LLVMBuilderRef builder,
+lp_build_intrinsic_map_unary(struct gallivm_state *gallivm,
                              const char *name,
                              LLVMTypeRef ret_type,
                              LLVMValueRef a);
 
 
 LLVMValueRef
-lp_build_intrinsic_map_binary(LLVMBuilderRef builder,
+lp_build_intrinsic_map_binary(struct gallivm_state *gallivm,
                               const char *name,
                               LLVMTypeRef ret_type,
                               LLVMValueRef a,
diff --git a/src/gallium/auxiliary/gallivm/lp_bld_logic.c b/src/gallium/auxiliary/gallivm/lp_bld_logic.c
index 026b60ac36..f7e6fbaff1 100644
--- a/src/gallium/auxiliary/gallivm/lp_bld_logic.c
+++ b/src/gallium/auxiliary/gallivm/lp_bld_logic.c
@@ -39,6 +39,7 @@
 
 #include "lp_bld_type.h"
 #include "lp_bld_const.h"
+#include "lp_bld_init.h"
 #include "lp_bld_intr.h"
 #include "lp_bld_debug.h"
 #include "lp_bld_logic.h"
@@ -70,13 +71,14 @@
  * The result values will be 0 for false or ~0 for true.
  */
 LLVMValueRef
-lp_build_compare(LLVMBuilderRef builder,
+lp_build_compare(struct gallivm_state *gallivm,
                  const struct lp_type type,
                  unsigned func,
                  LLVMValueRef a,
                  LLVMValueRef b)
 {
-   LLVMTypeRef int_vec_type = lp_build_int_vec_type(type);
+   LLVMBuilderRef builder = gallivm->builder;
+   LLVMTypeRef int_vec_type = lp_build_int_vec_type(gallivm, type);
    LLVMValueRef zeros = LLVMConstNull(int_vec_type);
    LLVMValueRef ones = LLVMConstAllOnes(int_vec_type);
    LLVMValueRef cond;
@@ -115,7 +117,7 @@ lp_build_compare(LLVMBuilderRef builder,
    if(type.width * type.length == 128) {
       if(type.floating && util_cpu_caps.has_sse) {
          /* float[4] comparison */
-         LLVMTypeRef vec_type = lp_build_vec_type(type);
+         LLVMTypeRef vec_type = lp_build_vec_type(gallivm, type);
          LLVMValueRef args[3];
          unsigned cc;
          boolean swap;
@@ -144,7 +146,7 @@ lp_build_compare(LLVMBuilderRef builder,
             break;
          default:
             assert(0);
-            return lp_build_undef(type);
+            return lp_build_undef(gallivm, type);
          }
 
          if(swap) {
@@ -156,7 +158,7 @@ lp_build_compare(LLVMBuilderRef builder,
             args[1] = b;
          }
 
-         args[2] = LLVMConstInt(LLVMInt8Type(), cc, 0);
+         args[2] = LLVMConstInt(LLVMInt8TypeInContext(gallivm->context), cc, 0);
          res = lp_build_intrinsic(builder,
                                   "llvm.x86.sse.cmp.ps",
                                   vec_type,
@@ -185,7 +187,7 @@ lp_build_compare(LLVMBuilderRef builder,
          const char *pcmpgt;
          LLVMValueRef args[2];
          LLVMValueRef res;
-         LLVMTypeRef vec_type = lp_build_vec_type(type);
+         LLVMTypeRef vec_type = lp_build_vec_type(gallivm, type);
 
          switch (type.width) {
          case 8:
@@ -202,14 +204,14 @@ lp_build_compare(LLVMBuilderRef builder,
             break;
          default:
             assert(0);
-            return lp_build_undef(type);
+            return lp_build_undef(gallivm, type);
          }
 
          /* There are no unsigned comparison instructions. So flip the sign bit
           * so that the results match.
           */
          if (table[func].gt && !type.sign) {
-            LLVMValueRef msb = lp_build_const_int_vec(type, (unsigned long long)1 << (type.width - 1));
+            LLVMValueRef msb = lp_build_const_int_vec(gallivm, type, (unsigned long long)1 << (type.width - 1));
             a = LLVMBuildXor(builder, a, msb, "");
             b = LLVMBuildXor(builder, b, msb, "");
          }
@@ -270,7 +272,7 @@ lp_build_compare(LLVMBuilderRef builder,
          break;
       default:
          assert(0);
-         return lp_build_undef(type);
+         return lp_build_undef(gallivm, type);
       }
 
 #if HAVE_LLVM >= 0x0207
@@ -289,7 +291,7 @@ lp_build_compare(LLVMBuilderRef builder,
          debug_printf("%s: warning: using slow element-wise float"
                       " vector comparison\n", __FUNCTION__);
          for (i = 0; i < type.length; ++i) {
-            LLVMValueRef index = LLVMConstInt(LLVMInt32Type(), i, 0);
+            LLVMValueRef index = lp_build_const_int32(gallivm, i);
             cond = LLVMBuildFCmp(builder, op,
                                  LLVMBuildExtractElement(builder, a, index, ""),
                                  LLVMBuildExtractElement(builder, b, index, ""),
@@ -326,7 +328,7 @@ lp_build_compare(LLVMBuilderRef builder,
          break;
       default:
          assert(0);
-         return lp_build_undef(type);
+         return lp_build_undef(gallivm, type);
       }
 
 #if HAVE_LLVM >= 0x0207
@@ -348,7 +350,7 @@ lp_build_compare(LLVMBuilderRef builder,
          }
 
          for(i = 0; i < type.length; ++i) {
-            LLVMValueRef index = LLVMConstInt(LLVMInt32Type(), i, 0);
+            LLVMValueRef index = lp_build_const_int32(gallivm, i);
             cond = LLVMBuildICmp(builder, op,
                                  LLVMBuildExtractElement(builder, a, index, ""),
                                  LLVMBuildExtractElement(builder, b, index, ""),
@@ -379,7 +381,7 @@ lp_build_cmp(struct lp_build_context *bld,
              LLVMValueRef a,
              LLVMValueRef b)
 {
-   return lp_build_compare(bld->builder, bld->type, func, a, b);
+   return lp_build_compare(bld->gallivm, bld->type, func, a, b);
 }
 
 
@@ -392,6 +394,7 @@ lp_build_select_bitwise(struct lp_build_context *bld,
                         LLVMValueRef a,
                         LLVMValueRef b)
 {
+   LLVMBuilderRef builder = bld->gallivm->builder;
    struct lp_type type = bld->type;
    LLVMValueRef res;
 
@@ -403,25 +406,25 @@ lp_build_select_bitwise(struct lp_build_context *bld,
    }
 
    if(type.floating) {
-      LLVMTypeRef int_vec_type = lp_build_int_vec_type(type);
-      a = LLVMBuildBitCast(bld->builder, a, int_vec_type, "");
-      b = LLVMBuildBitCast(bld->builder, b, int_vec_type, "");
+      LLVMTypeRef int_vec_type = lp_build_int_vec_type(bld->gallivm, type);
+      a = LLVMBuildBitCast(builder, a, int_vec_type, "");
+      b = LLVMBuildBitCast(builder, b, int_vec_type, "");
    }
 
-   a = LLVMBuildAnd(bld->builder, a, mask, "");
+   a = LLVMBuildAnd(builder, a, mask, "");
 
    /* This often gets translated to PANDN, but sometimes the NOT is
     * pre-computed and stored in another constant. The best strategy depends
     * on available registers, so it is not a big deal -- hopefully LLVM does
     * the right decision attending the rest of the program.
     */
-   b = LLVMBuildAnd(bld->builder, b, LLVMBuildNot(bld->builder, mask, ""), "");
+   b = LLVMBuildAnd(builder, b, LLVMBuildNot(builder, mask, ""), "");
 
-   res = LLVMBuildOr(bld->builder, a, b, "");
+   res = LLVMBuildOr(builder, a, b, "");
 
    if(type.floating) {
-      LLVMTypeRef vec_type = lp_build_vec_type(type);
-      res = LLVMBuildBitCast(bld->builder, res, vec_type, "");
+      LLVMTypeRef vec_type = lp_build_vec_type(bld->gallivm, type);
+      res = LLVMBuildBitCast(builder, res, vec_type, "");
    }
 
    return res;
@@ -440,6 +443,8 @@ lp_build_select(struct lp_build_context *bld,
                 LLVMValueRef a,
                 LLVMValueRef b)
 {
+   LLVMBuilderRef builder = bld->gallivm->builder;
+   LLVMContextRef lc = bld->gallivm->context;
    struct lp_type type = bld->type;
    LLVMValueRef res;
 
@@ -450,8 +455,8 @@ lp_build_select(struct lp_build_context *bld,
       return a;
 
    if (type.length == 1) {
-      mask = LLVMBuildTrunc(bld->builder, mask, LLVMInt1Type(), "");
-      res = LLVMBuildSelect(bld->builder, mask, a, b, "");
+      mask = LLVMBuildTrunc(builder, mask, LLVMInt1TypeInContext(lc), "");
+      res = LLVMBuildSelect(builder, mask, a, b, "");
    }
    else if (util_cpu_caps.has_sse4_1 &&
             type.width * type.length == 128 &&
@@ -465,34 +470,34 @@ lp_build_select(struct lp_build_context *bld,
       if (type.floating &&
           type.width == 64) {
          intrinsic = "llvm.x86.sse41.blendvpd";
-         arg_type = LLVMVectorType(LLVMDoubleType(), 2);
+         arg_type = LLVMVectorType(LLVMDoubleTypeInContext(lc), 2);
       } else if (type.floating &&
                  type.width == 32) {
          intrinsic = "llvm.x86.sse41.blendvps";
-         arg_type = LLVMVectorType(LLVMFloatType(), 4);
+         arg_type = LLVMVectorType(LLVMFloatTypeInContext(lc), 4);
       } else {
          intrinsic = "llvm.x86.sse41.pblendvb";
-         arg_type = LLVMVectorType(LLVMInt8Type(), 16);
+         arg_type = LLVMVectorType(LLVMInt8TypeInContext(lc), 16);
       }
 
       if (arg_type != bld->int_vec_type) {
-         mask = LLVMBuildBitCast(bld->builder, mask, arg_type, "");
+         mask = LLVMBuildBitCast(builder, mask, arg_type, "");
       }
 
       if (arg_type != bld->vec_type) {
-         a = LLVMBuildBitCast(bld->builder, a, arg_type, "");
-         b = LLVMBuildBitCast(bld->builder, b, arg_type, "");
+         a = LLVMBuildBitCast(builder, a, arg_type, "");
+         b = LLVMBuildBitCast(builder, b, arg_type, "");
       }
 
       args[0] = b;
       args[1] = a;
       args[2] = mask;
 
-      res = lp_build_intrinsic(bld->builder, intrinsic,
+      res = lp_build_intrinsic(builder, intrinsic,
                                arg_type, args, Elements(args));
 
       if (arg_type != bld->vec_type) {
-         res = LLVMBuildBitCast(bld->builder, res, bld->vec_type, "");
+         res = LLVMBuildBitCast(builder, res, bld->vec_type, "");
       }
    }
    else {
@@ -514,6 +519,7 @@ lp_build_select_aos(struct lp_build_context *bld,
                     LLVMValueRef a,
                     LLVMValueRef b)
 {
+   LLVMBuilderRef builder = bld->gallivm->builder;
    const struct lp_type type = bld->type;
    const unsigned n = type.length;
    unsigned i, j;
@@ -544,7 +550,7 @@ lp_build_select_aos(struct lp_build_context *bld,
       /*
        * Shuffle.
        */
-      LLVMTypeRef elem_type = LLVMInt32Type();
+      LLVMTypeRef elem_type = LLVMInt32TypeInContext(bld->gallivm->context);
       LLVMValueRef shuffles[LP_MAX_VECTOR_LENGTH];
 
       for(j = 0; j < n; j += 4)
@@ -553,7 +559,7 @@ lp_build_select_aos(struct lp_build_context *bld,
                                            (mask & (1 << i) ? 0 : n) + j + i,
                                            0);
 
-      return LLVMBuildShuffleVector(bld->builder, a, b, LLVMConstVector(shuffles, n), "");
+      return LLVMBuildShuffleVector(builder, a, b, LLVMConstVector(shuffles, n), "");
    }
    else {
 #if 0
@@ -567,9 +573,9 @@ lp_build_select_aos(struct lp_build_context *bld,
             cond_vec[j + i] = LLVMConstInt(elem_type,
                                            mask & (1 << i) ? 1 : 0, 0);
 
-      return LLVMBuildSelect(bld->builder, LLVMConstVector(cond_vec, n), a, b, "");
+      return LLVMBuildSelect(builder, LLVMConstVector(cond_vec, n), a, b, "");
 #else
-      LLVMValueRef mask_vec = lp_build_const_mask_aos(type, mask);
+      LLVMValueRef mask_vec = lp_build_const_mask_aos(bld->gallivm, type, mask);
       return lp_build_select(bld, mask_vec, a, b);
 #endif
    }
diff --git a/src/gallium/auxiliary/gallivm/lp_bld_logic.h b/src/gallium/auxiliary/gallivm/lp_bld_logic.h
index 141fb92058..ef33a65368 100644
--- a/src/gallium/auxiliary/gallivm/lp_bld_logic.h
+++ b/src/gallium/auxiliary/gallivm/lp_bld_logic.h
@@ -47,7 +47,7 @@ struct lp_build_context;
 
 
 LLVMValueRef
-lp_build_compare(LLVMBuilderRef builder,
+lp_build_compare(struct gallivm_state *gallivm,
                  const struct lp_type type,
                  unsigned func,
                  LLVMValueRef a,
diff --git a/src/gallium/auxiliary/gallivm/lp_bld_pack.c b/src/gallium/auxiliary/gallivm/lp_bld_pack.c
index f7eb7148ab..fde6bb594f 100644
--- a/src/gallium/auxiliary/gallivm/lp_bld_pack.c
+++ b/src/gallium/auxiliary/gallivm/lp_bld_pack.c
@@ -72,6 +72,7 @@
 
 #include "lp_bld_type.h"
 #include "lp_bld_const.h"
+#include "lp_bld_init.h"
 #include "lp_bld_intr.h"
 #include "lp_bld_arit.h"
 #include "lp_bld_pack.h"
@@ -81,7 +82,8 @@
  * Build shuffle vectors that match PUNPCKLxx and PUNPCKHxx instructions.
  */
 static LLVMValueRef
-lp_build_const_unpack_shuffle(unsigned n, unsigned lo_hi)
+lp_build_const_unpack_shuffle(struct gallivm_state *gallivm,
+                              unsigned n, unsigned lo_hi)
 {
    LLVMValueRef elems[LP_MAX_VECTOR_LENGTH];
    unsigned i, j;
@@ -92,8 +94,8 @@ lp_build_const_unpack_shuffle(unsigned n, unsigned lo_hi)
    /* TODO: cache results in a static table */
 
    for(i = 0, j = lo_hi*n/2; i < n; i += 2, ++j) {
-      elems[i + 0] = LLVMConstInt(LLVMInt32Type(), 0 + j, 0);
-      elems[i + 1] = LLVMConstInt(LLVMInt32Type(), n + j, 0);
+      elems[i + 0] = lp_build_const_int32(gallivm, 0 + j);
+      elems[i + 1] = lp_build_const_int32(gallivm, n + j);
    }
 
    return LLVMConstVector(elems, n);
@@ -104,7 +106,7 @@ lp_build_const_unpack_shuffle(unsigned n, unsigned lo_hi)
  * Build shuffle vectors that match PACKxx instructions.
  */
 static LLVMValueRef
-lp_build_const_pack_shuffle(unsigned n)
+lp_build_const_pack_shuffle(struct gallivm_state *gallivm, unsigned n)
 {
    LLVMValueRef elems[LP_MAX_VECTOR_LENGTH];
    unsigned i;
@@ -112,7 +114,7 @@ lp_build_const_pack_shuffle(unsigned n)
    assert(n <= LP_MAX_VECTOR_LENGTH);
 
    for(i = 0; i < n; ++i)
-      elems[i] = LLVMConstInt(LLVMInt32Type(), 2*i, 0);
+      elems[i] = lp_build_const_int32(gallivm, 2*i);
 
    return LLVMConstVector(elems, n);
 }
@@ -124,7 +126,7 @@ lp_build_const_pack_shuffle(unsigned n)
  * Matches the PUNPCKLxx and PUNPCKHxx SSE instructions.
  */
 LLVMValueRef
-lp_build_interleave2(LLVMBuilderRef builder,
+lp_build_interleave2(struct gallivm_state *gallivm,
                      struct lp_type type,
                      LLVMValueRef a,
                      LLVMValueRef b,
@@ -132,9 +134,9 @@ lp_build_interleave2(LLVMBuilderRef builder,
 {
    LLVMValueRef shuffle;
 
-   shuffle = lp_build_const_unpack_shuffle(type.length, lo_hi);
+   shuffle = lp_build_const_unpack_shuffle(gallivm, type.length, lo_hi);
 
-   return LLVMBuildShuffleVector(builder, a, b, shuffle, "");
+   return LLVMBuildShuffleVector(gallivm->builder, a, b, shuffle, "");
 }
 
 
@@ -145,13 +147,14 @@ lp_build_interleave2(LLVMBuilderRef builder,
  * values themselves.
  */
 void
-lp_build_unpack2(LLVMBuilderRef builder,
+lp_build_unpack2(struct gallivm_state *gallivm,
                  struct lp_type src_type,
                  struct lp_type dst_type,
                  LLVMValueRef src,
                  LLVMValueRef *dst_lo,
                  LLVMValueRef *dst_hi)
 {
+   LLVMBuilderRef builder = gallivm->builder;
    LLVMValueRef msb;
    LLVMTypeRef dst_vec_type;
 
@@ -162,24 +165,24 @@ lp_build_unpack2(LLVMBuilderRef builder,
 
    if(dst_type.sign && src_type.sign) {
       /* Replicate the sign bit in the most significant bits */
-      msb = LLVMBuildAShr(builder, src, lp_build_const_int_vec(src_type, src_type.width - 1), "");
+      msb = LLVMBuildAShr(builder, src, lp_build_const_int_vec(gallivm, src_type, src_type.width - 1), "");
    }
    else
       /* Most significant bits always zero */
-      msb = lp_build_zero(src_type);
+      msb = lp_build_zero(gallivm, src_type);
 
    /* Interleave bits */
 #ifdef PIPE_ARCH_LITTLE_ENDIAN
-      *dst_lo = lp_build_interleave2(builder, src_type, src, msb, 0);
-      *dst_hi = lp_build_interleave2(builder, src_type, src, msb, 1);
+   *dst_lo = lp_build_interleave2(gallivm, src_type, src, msb, 0);
+   *dst_hi = lp_build_interleave2(gallivm, src_type, src, msb, 1);
 #else
-      *dst_lo = lp_build_interleave2(builder, src_type, msb, src, 0);
-      *dst_hi = lp_build_interleave2(builder, src_type, msb, src, 1);
+   *dst_lo = lp_build_interleave2(gallivm, src_type, msb, src, 0);
+   *dst_hi = lp_build_interleave2(gallivm, src_type, msb, src, 1);
 #endif
 
    /* Cast the result into the new type (twice as wide) */
 
-   dst_vec_type = lp_build_vec_type(dst_type);
+   dst_vec_type = lp_build_vec_type(gallivm, dst_type);
 
    *dst_lo = LLVMBuildBitCast(builder, *dst_lo, dst_vec_type, "");
    *dst_hi = LLVMBuildBitCast(builder, *dst_hi, dst_vec_type, "");
@@ -193,7 +196,7 @@ lp_build_unpack2(LLVMBuilderRef builder,
  * values themselves.
  */
 void
-lp_build_unpack(LLVMBuilderRef builder,
+lp_build_unpack(struct gallivm_state *gallivm,
                 struct lp_type src_type,
                 struct lp_type dst_type,
                 LLVMValueRef src,
@@ -218,7 +221,7 @@ lp_build_unpack(LLVMBuilderRef builder,
       tmp_type.length /= 2;
 
       for(i = num_tmps; i--; ) {
-         lp_build_unpack2(builder, src_type, tmp_type, dst[i], &dst[2*i + 0], &dst[2*i + 1]);
+         lp_build_unpack2(gallivm, src_type, tmp_type, dst[i], &dst[2*i + 0], &dst[2*i + 1]);
       }
 
       src_type = tmp_type;
@@ -247,16 +250,17 @@ lp_build_unpack(LLVMBuilderRef builder,
  * lp_build_packs2 instead.
  */
 LLVMValueRef
-lp_build_pack2(LLVMBuilderRef builder,
+lp_build_pack2(struct gallivm_state *gallivm,
                struct lp_type src_type,
                struct lp_type dst_type,
                LLVMValueRef lo,
                LLVMValueRef hi)
 {
+   LLVMBuilderRef builder = gallivm->builder;
 #if HAVE_LLVM < 0x0207
-   LLVMTypeRef src_vec_type = lp_build_vec_type(src_type);
+   LLVMTypeRef src_vec_type = lp_build_vec_type(gallivm, src_type);
 #endif
-   LLVMTypeRef dst_vec_type = lp_build_vec_type(dst_type);
+   LLVMTypeRef dst_vec_type = lp_build_vec_type(gallivm, dst_type);
    LLVMValueRef shuffle;
    LLVMValueRef res = NULL;
 
@@ -318,7 +322,7 @@ lp_build_pack2(LLVMBuilderRef builder,
    lo = LLVMBuildBitCast(builder, lo, dst_vec_type, "");
    hi = LLVMBuildBitCast(builder, hi, dst_vec_type, "");
 
-   shuffle = lp_build_const_pack_shuffle(dst_type.length);
+   shuffle = lp_build_const_pack_shuffle(gallivm, dst_type.length);
 
    res = LLVMBuildShuffleVector(builder, lo, hi, shuffle, "");
 
@@ -334,7 +338,7 @@ lp_build_pack2(LLVMBuilderRef builder,
  * destination type.
  */
 LLVMValueRef
-lp_build_packs2(LLVMBuilderRef builder,
+lp_build_packs2(struct gallivm_state *gallivm,
                 struct lp_type src_type,
                 struct lp_type dst_type,
                 LLVMValueRef lo,
@@ -360,14 +364,14 @@ lp_build_packs2(LLVMBuilderRef builder,
    if(clamp) {
       struct lp_build_context bld;
       unsigned dst_bits = dst_type.sign ? dst_type.width - 1 : dst_type.width;
-      LLVMValueRef dst_max = lp_build_const_int_vec(src_type, ((unsigned long long)1 << dst_bits) - 1);
-      lp_build_context_init(&bld, builder, src_type);
+      LLVMValueRef dst_max = lp_build_const_int_vec(gallivm, src_type, ((unsigned long long)1 << dst_bits) - 1);
+      lp_build_context_init(&bld, gallivm, src_type);
       lo = lp_build_min(&bld, lo, dst_max);
       hi = lp_build_min(&bld, hi, dst_max);
       /* FIXME: What about lower bound? */
    }
 
-   return lp_build_pack2(builder, src_type, dst_type, lo, hi);
+   return lp_build_pack2(gallivm, src_type, dst_type, lo, hi);
 }
 
 
@@ -377,13 +381,13 @@ lp_build_packs2(LLVMBuilderRef builder,
  * TODO: Handle saturation consistently.
  */
 LLVMValueRef
-lp_build_pack(LLVMBuilderRef builder,
+lp_build_pack(struct gallivm_state *gallivm,
               struct lp_type src_type,
               struct lp_type dst_type,
               boolean clamped,
               const LLVMValueRef *src, unsigned num_srcs)
 {
-   LLVMValueRef (*pack2)(LLVMBuilderRef builder,
+   LLVMValueRef (*pack2)(struct gallivm_state *gallivm,
                          struct lp_type src_type,
                          struct lp_type dst_type,
                          LLVMValueRef lo,
@@ -419,7 +423,8 @@ lp_build_pack(LLVMBuilderRef builder,
       num_srcs /= 2;
 
       for(i = 0; i < num_srcs; ++i)
-         tmp[i] = pack2(builder, src_type, tmp_type, tmp[2*i + 0], tmp[2*i + 1]);
+         tmp[i] = pack2(gallivm, src_type, tmp_type,
+                        tmp[2*i + 0], tmp[2*i + 1]);
 
       src_type = tmp_type;
    }
@@ -437,12 +442,13 @@ lp_build_pack(LLVMBuilderRef builder,
  * intrinsics that do saturation.
  */
 void
-lp_build_resize(LLVMBuilderRef builder,
+lp_build_resize(struct gallivm_state *gallivm,
                 struct lp_type src_type,
                 struct lp_type dst_type,
                 const LLVMValueRef *src, unsigned num_srcs,
                 LLVMValueRef *dst, unsigned num_dsts)
 {
+   LLVMBuilderRef builder = gallivm->builder;
    LLVMValueRef tmp[LP_MAX_VECTOR_LENGTH];
    unsigned i;
 
@@ -482,7 +488,7 @@ lp_build_resize(LLVMBuilderRef builder,
          * Register width remains constant -- use vector packing intrinsics
          */
 
-         tmp[0] = lp_build_pack(builder, src_type, dst_type, TRUE, src, num_srcs);
+         tmp[0] = lp_build_pack(gallivm, src_type, dst_type, TRUE, src, num_srcs);
       }
       else {
          /*
@@ -490,11 +496,11 @@ lp_build_resize(LLVMBuilderRef builder,
           */
 
          assert(src_type.length == dst_type.length);
-         tmp[0] = lp_build_undef(dst_type);
+         tmp[0] = lp_build_undef(gallivm, dst_type);
          for (i = 0; i < dst_type.length; ++i) {
-            LLVMValueRef index = LLVMConstInt(LLVMInt32Type(), i, 0);
+            LLVMValueRef index = lp_build_const_int32(gallivm, i);
             LLVMValueRef val = LLVMBuildExtractElement(builder, src[0], index, "");
-            val = LLVMBuildTrunc(builder, val, lp_build_elem_type(dst_type), "");
+            val = LLVMBuildTrunc(builder, val, lp_build_elem_type(gallivm, dst_type), "");
             tmp[0] = LLVMBuildInsertElement(builder, tmp[0], val, index, "");
          }
       }
@@ -510,7 +516,7 @@ lp_build_resize(LLVMBuilderRef builder,
          /*
           * Register width remains constant -- use vector unpack intrinsics
           */
-         lp_build_unpack(builder, src_type, dst_type, src[0], tmp, num_dsts);
+         lp_build_unpack(gallivm, src_type, dst_type, src[0], tmp, num_dsts);
       }
       else {
          /*
@@ -518,15 +524,15 @@ lp_build_resize(LLVMBuilderRef builder,
           */
 
          assert(src_type.length == dst_type.length);
-         tmp[0] = lp_build_undef(dst_type);
+         tmp[0] = lp_build_undef(gallivm, dst_type);
          for (i = 0; i < dst_type.length; ++i) {
-            LLVMValueRef index = LLVMConstInt(LLVMInt32Type(), i, 0);
+            LLVMValueRef index = lp_build_const_int32(gallivm, i);
             LLVMValueRef val = LLVMBuildExtractElement(builder, src[0], index, "");
 
             if (src_type.sign && dst_type.sign) {
-               val = LLVMBuildSExt(builder, val, lp_build_elem_type(dst_type), "");
+               val = LLVMBuildSExt(builder, val, lp_build_elem_type(gallivm, dst_type), "");
             } else {
-               val = LLVMBuildZExt(builder, val, lp_build_elem_type(dst_type), "");
+               val = LLVMBuildZExt(builder, val, lp_build_elem_type(gallivm, dst_type), "");
             }
             tmp[0] = LLVMBuildInsertElement(builder, tmp[0], val, index, "");
          }
diff --git a/src/gallium/auxiliary/gallivm/lp_bld_pack.h b/src/gallium/auxiliary/gallivm/lp_bld_pack.h
index e947b90d16..d58da4f01b 100644
--- a/src/gallium/auxiliary/gallivm/lp_bld_pack.h
+++ b/src/gallium/auxiliary/gallivm/lp_bld_pack.h
@@ -46,7 +46,7 @@ struct lp_type;
 
 
 LLVMValueRef
-lp_build_interleave2(LLVMBuilderRef builder,
+lp_build_interleave2(struct gallivm_state *gallivm,
                      struct lp_type type,
                      LLVMValueRef a,
                      LLVMValueRef b,
@@ -54,7 +54,7 @@ lp_build_interleave2(LLVMBuilderRef builder,
 
 
 void
-lp_build_unpack2(LLVMBuilderRef builder,
+lp_build_unpack2(struct gallivm_state *gallivm,
                  struct lp_type src_type,
                  struct lp_type dst_type,
                  LLVMValueRef src,
@@ -63,7 +63,7 @@ lp_build_unpack2(LLVMBuilderRef builder,
 
 
 void
-lp_build_unpack(LLVMBuilderRef builder,
+lp_build_unpack(struct gallivm_state *gallivm,
                 struct lp_type src_type,
                 struct lp_type dst_type,
                 LLVMValueRef src,
@@ -71,7 +71,7 @@ lp_build_unpack(LLVMBuilderRef builder,
 
 
 LLVMValueRef
-lp_build_packs2(LLVMBuilderRef builder,
+lp_build_packs2(struct gallivm_state *gallivm,
                 struct lp_type src_type,
                 struct lp_type dst_type,
                 LLVMValueRef lo,
@@ -79,7 +79,7 @@ lp_build_packs2(LLVMBuilderRef builder,
 
 
 LLVMValueRef
-lp_build_pack2(LLVMBuilderRef builder,
+lp_build_pack2(struct gallivm_state *gallivm,
                struct lp_type src_type,
                struct lp_type dst_type,
                LLVMValueRef lo,
@@ -87,7 +87,7 @@ lp_build_pack2(LLVMBuilderRef builder,
 
 
 LLVMValueRef
-lp_build_pack(LLVMBuilderRef builder,
+lp_build_pack(struct gallivm_state *gallivm,
               struct lp_type src_type,
               struct lp_type dst_type,
               boolean clamped,
@@ -95,7 +95,7 @@ lp_build_pack(LLVMBuilderRef builder,
 
 
 void
-lp_build_resize(LLVMBuilderRef builder,
+lp_build_resize(struct gallivm_state *gallivm,
                 struct lp_type src_type,
                 struct lp_type dst_type,
                 const LLVMValueRef *src, unsigned num_srcs,
diff --git a/src/gallium/auxiliary/gallivm/lp_bld_printf.c b/src/gallium/auxiliary/gallivm/lp_bld_printf.c
index f418e96aff..60cc6094f5 100644
--- a/src/gallium/auxiliary/gallivm/lp_bld_printf.c
+++ b/src/gallium/auxiliary/gallivm/lp_bld_printf.c
@@ -31,6 +31,8 @@
 #include "util/u_memory.h"
 #include "util/u_string.h"
 #include "lp_bld_const.h"
+#include "lp_bld_init.h"
+#include "lp_bld_const.h"
 #include "lp_bld_printf.h"
 
 
@@ -65,12 +67,14 @@ lp_get_printf_arg_count(const char *fmt)
 }
 
 LLVMValueRef 
-lp_build_const_string_variable(LLVMModuleRef module, const char *str, int len)
+lp_build_const_string_variable(LLVMModuleRef module,
+                               LLVMContextRef context,
+                               const char *str, int len)
 {
-   LLVMValueRef string = LLVMAddGlobal(module, LLVMArrayType(LLVMInt8Type(), len + 1), "");
+   LLVMValueRef string = LLVMAddGlobal(module, LLVMArrayType(LLVMInt8TypeInContext(context), len + 1), "");
    LLVMSetGlobalConstant(string, TRUE);
    LLVMSetLinkage(string, LLVMInternalLinkage);
-   LLVMSetInitializer(string, LLVMConstString(str, len + 1, TRUE));
+   LLVMSetInitializer(string, LLVMConstStringInContext(context, str, len + 1, TRUE));
    return string;
 }
  
@@ -83,15 +87,18 @@ lp_build_const_string_variable(LLVMModuleRef module, const char *str, int len)
  * LLVMValueRef.
  */
 LLVMValueRef
-lp_build_printf(LLVMBuilderRef builder, const char *fmt, ...)
+lp_build_printf(struct gallivm_state *gallivm, const char *fmt, ...)
 {
    va_list arglist;
    int i = 0;
    int argcount = lp_get_printf_arg_count(fmt);
-   LLVMModuleRef module = LLVMGetGlobalParent(LLVMGetBasicBlockParent(LLVMGetInsertBlock(builder)));
+   LLVMBuilderRef builder = gallivm->builder;
+   LLVMContextRef context = gallivm->context;
+   LLVMModuleRef module = gallivm->module;
    LLVMValueRef params[50];
-   LLVMValueRef fmtarg = lp_build_const_string_variable(module, fmt, strlen(fmt) + 1);
-   LLVMValueRef int0 = LLVMConstInt(LLVMInt32Type(), 0, 0);
+   LLVMValueRef fmtarg = lp_build_const_string_variable(module, context,
+                                                        fmt, strlen(fmt) + 1);
+   LLVMValueRef int0 = lp_build_const_int32(gallivm, 0);
    LLVMValueRef index[2];
    LLVMValueRef func_printf = LLVMGetNamedFunction(module, "printf");
 
@@ -100,7 +107,7 @@ lp_build_printf(LLVMBuilderRef builder, const char *fmt, ...)
    index[0] = index[1] = int0;
 
    if (!func_printf) {
-      LLVMTypeRef printf_type = LLVMFunctionType(LLVMIntType(32), NULL, 0, 1);
+      LLVMTypeRef printf_type = LLVMFunctionType(LLVMIntTypeInContext(context, 32), NULL, 0, 1);
       func_printf = LLVMAddFunction(module, "printf", printf_type);
    }
 
@@ -113,7 +120,7 @@ lp_build_printf(LLVMBuilderRef builder, const char *fmt, ...)
       /* printf wants doubles, so lets convert so that
        * we can actually print them */
       if (LLVMGetTypeKind(type) == LLVMFloatTypeKind)
-         val = LLVMBuildFPExt(builder, val, LLVMDoubleType(), "");
+         val = LLVMBuildFPExt(builder, val, LLVMDoubleTypeInContext(context), "");
       params[i] = val;
    }
    va_end(arglist);
@@ -127,16 +134,18 @@ lp_build_printf(LLVMBuilderRef builder, const char *fmt, ...)
  * Print a float[4] vector.
  */
 LLVMValueRef
-lp_build_print_vec4(LLVMBuilderRef builder, const char *msg, LLVMValueRef vec)
+lp_build_print_vec4(struct gallivm_state *gallivm,
+                    const char *msg, LLVMValueRef vec)
 {
+   LLVMBuilderRef builder = gallivm->builder;
    char format[1000];
    LLVMValueRef x, y, z, w;
 
-   x = LLVMBuildExtractElement(builder, vec, lp_build_const_int32(0), "");
-   y = LLVMBuildExtractElement(builder, vec, lp_build_const_int32(1), "");
-   z = LLVMBuildExtractElement(builder, vec, lp_build_const_int32(2), "");
-   w = LLVMBuildExtractElement(builder, vec, lp_build_const_int32(3), "");
+   x = LLVMBuildExtractElement(builder, vec, lp_build_const_int32(gallivm, 0), "");
+   y = LLVMBuildExtractElement(builder, vec, lp_build_const_int32(gallivm, 1), "");
+   z = LLVMBuildExtractElement(builder, vec, lp_build_const_int32(gallivm, 2), "");
+   w = LLVMBuildExtractElement(builder, vec, lp_build_const_int32(gallivm, 3), "");
 
    util_snprintf(format, sizeof(format), "%s %%f %%f %%f %%f\n", msg);
-   return lp_build_printf(builder, format, x, y, z, w);
+   return lp_build_printf(gallivm, format, x, y, z, w);
 }
diff --git a/src/gallium/auxiliary/gallivm/lp_bld_printf.h b/src/gallium/auxiliary/gallivm/lp_bld_printf.h
index b6222c62eb..f6bb834869 100644
--- a/src/gallium/auxiliary/gallivm/lp_bld_printf.h
+++ b/src/gallium/auxiliary/gallivm/lp_bld_printf.h
@@ -31,12 +31,19 @@
 
 #include "pipe/p_compiler.h"
 #include "lp_bld.h"
+#include "lp_bld_init.h"
 
-LLVMValueRef lp_build_const_string_variable(LLVMModuleRef module, const char *str, int len);
-LLVMValueRef lp_build_printf(LLVMBuilderRef builder, const char *fmt, ...);
+
+LLVMValueRef lp_build_const_string_variable(LLVMModuleRef module,
+                                            LLVMContextRef context,
+                                            const char *str, int len);
+
+LLVMValueRef lp_build_printf(struct gallivm_state *gallivm,
+                             const char *fmt, ...);
 
 LLVMValueRef
-lp_build_print_vec4(LLVMBuilderRef builder, const char *msg, LLVMValueRef vec);
+lp_build_print_vec4(struct gallivm_state *gallivm,
+                    const char *msg, LLVMValueRef vec);
 
 
 #endif
diff --git a/src/gallium/auxiliary/gallivm/lp_bld_quad.c b/src/gallium/auxiliary/gallivm/lp_bld_quad.c
index c18c8b4710..b0a5bc0267 100644
--- a/src/gallium/auxiliary/gallivm/lp_bld_quad.c
+++ b/src/gallium/auxiliary/gallivm/lp_bld_quad.c
@@ -28,6 +28,7 @@
 
 #include "lp_bld_type.h"
 #include "lp_bld_arit.h"
+#include "lp_bld_const.h"
 #include "lp_bld_swizzle.h"
 #include "lp_bld_quad.h"
 
@@ -81,15 +82,15 @@ LLVMValueRef
 lp_build_scalar_ddx(struct lp_build_context *bld,
                     LLVMValueRef a)
 {
-   LLVMTypeRef i32t = LLVMInt32Type();
-   LLVMValueRef idx_left  = LLVMConstInt(i32t, LP_BLD_QUAD_TOP_LEFT, 0);
-   LLVMValueRef idx_right = LLVMConstInt(i32t, LP_BLD_QUAD_TOP_RIGHT, 0);
-   LLVMValueRef a_left  = LLVMBuildExtractElement(bld->builder, a, idx_left, "left");
-   LLVMValueRef a_right = LLVMBuildExtractElement(bld->builder, a, idx_right, "right");
+   LLVMBuilderRef builder = bld->gallivm->builder;
+   LLVMValueRef idx_left  = lp_build_const_int32(bld->gallivm, LP_BLD_QUAD_TOP_LEFT);
+   LLVMValueRef idx_right = lp_build_const_int32(bld->gallivm, LP_BLD_QUAD_TOP_RIGHT);
+   LLVMValueRef a_left  = LLVMBuildExtractElement(builder, a, idx_left, "left");
+   LLVMValueRef a_right = LLVMBuildExtractElement(builder, a, idx_right, "right");
    if (bld->type.floating)
-      return LLVMBuildFSub(bld->builder, a_right, a_left, "ddx");
+      return LLVMBuildFSub(builder, a_right, a_left, "ddx");
    else
-      return LLVMBuildSub(bld->builder, a_right, a_left, "ddx");
+      return LLVMBuildSub(builder, a_right, a_left, "ddx");
 }
 
 
@@ -97,13 +98,13 @@ LLVMValueRef
 lp_build_scalar_ddy(struct lp_build_context *bld,
                     LLVMValueRef a)
 {
-   LLVMTypeRef i32t = LLVMInt32Type();
-   LLVMValueRef idx_top    = LLVMConstInt(i32t, LP_BLD_QUAD_TOP_LEFT, 0);
-   LLVMValueRef idx_bottom = LLVMConstInt(i32t, LP_BLD_QUAD_BOTTOM_LEFT, 0);
-   LLVMValueRef a_top    = LLVMBuildExtractElement(bld->builder, a, idx_top, "top");
-   LLVMValueRef a_bottom = LLVMBuildExtractElement(bld->builder, a, idx_bottom, "bottom");
+   LLVMBuilderRef builder = bld->gallivm->builder;
+   LLVMValueRef idx_top    = lp_build_const_int32(bld->gallivm, LP_BLD_QUAD_TOP_LEFT);
+   LLVMValueRef idx_bottom = lp_build_const_int32(bld->gallivm, LP_BLD_QUAD_BOTTOM_LEFT);
+   LLVMValueRef a_top    = LLVMBuildExtractElement(builder, a, idx_top, "top");
+   LLVMValueRef a_bottom = LLVMBuildExtractElement(builder, a, idx_bottom, "bottom");
    if (bld->type.floating)
-      return LLVMBuildFSub(bld->builder, a_bottom, a_top, "ddy");
+      return LLVMBuildFSub(builder, a_bottom, a_top, "ddy");
    else
-      return LLVMBuildSub(bld->builder, a_bottom, a_top, "ddy");
+      return LLVMBuildSub(builder, a_bottom, a_top, "ddy");
 }
diff --git a/src/gallium/auxiliary/gallivm/lp_bld_sample.c b/src/gallium/auxiliary/gallivm/lp_bld_sample.c
index 844d1d935b..8ad34598a9 100644
--- a/src/gallium/auxiliary/gallivm/lp_bld_sample.c
+++ b/src/gallium/auxiliary/gallivm/lp_bld_sample.c
@@ -134,7 +134,7 @@ lp_sampler_static_state(struct lp_sampler_static_state *state,
    state->min_img_filter    = sampler->min_img_filter;
    state->mag_img_filter    = sampler->mag_img_filter;
 
-   if (view->last_level && sampler->max_lod > 0.0f) {
+   if (view->u.tex.last_level && sampler->max_lod > 0.0f) {
       state->min_mip_filter = sampler->min_mip_filter;
    } else {
       state->min_mip_filter = PIPE_TEX_MIPFILTER_NONE;
@@ -155,7 +155,7 @@ lp_sampler_static_state(struct lp_sampler_static_state *state,
             state->apply_min_lod = 1;
          }
 
-         if (sampler->max_lod < (float)view->last_level) {
+         if (sampler->max_lod < (float)view->u.tex.last_level) {
             state->apply_max_lod = 1;
          }
       }
@@ -190,7 +190,8 @@ lp_build_rho(struct lp_build_sample_context *bld,
    struct lp_build_context *float_size_bld = &bld->float_size_bld;
    struct lp_build_context *float_bld = &bld->float_bld;
    const unsigned dims = bld->dims;
-   LLVMTypeRef i32t = LLVMInt32Type();
+   LLVMBuilderRef builder = bld->gallivm->builder;
+   LLVMTypeRef i32t = LLVMInt32TypeInContext(bld->gallivm->context);
    LLVMValueRef index0 = LLVMConstInt(i32t, 0, 0);
    LLVMValueRef index1 = LLVMConstInt(i32t, 1, 0);
    LLVMValueRef index2 = LLVMConstInt(i32t, 2, 0);
@@ -211,21 +212,21 @@ lp_build_rho(struct lp_build_sample_context *bld,
       rho_x = float_size_bld->undef;
       rho_y = float_size_bld->undef;
 
-      rho_x = LLVMBuildInsertElement(bld->builder, rho_x, dsdx, index0, "");
-      rho_y = LLVMBuildInsertElement(bld->builder, rho_y, dsdy, index0, "");
+      rho_x = LLVMBuildInsertElement(builder, rho_x, dsdx, index0, "");
+      rho_y = LLVMBuildInsertElement(builder, rho_y, dsdy, index0, "");
 
       dtdx = ddx[1];
       dtdy = ddy[1];
 
-      rho_x = LLVMBuildInsertElement(bld->builder, rho_x, dtdx, index1, "");
-      rho_y = LLVMBuildInsertElement(bld->builder, rho_y, dtdy, index1, "");
+      rho_x = LLVMBuildInsertElement(builder, rho_x, dtdx, index1, "");
+      rho_y = LLVMBuildInsertElement(builder, rho_y, dtdy, index1, "");
 
       if (dims >= 3) {
          drdx = ddx[2];
          drdy = ddy[2];
 
-         rho_x = LLVMBuildInsertElement(bld->builder, rho_x, drdx, index2, "");
-         rho_y = LLVMBuildInsertElement(bld->builder, rho_y, drdy, index2, "");
+         rho_x = LLVMBuildInsertElement(builder, rho_x, drdx, index2, "");
+         rho_y = LLVMBuildInsertElement(builder, rho_y, drdy, index2, "");
       }
    }
 
@@ -245,13 +246,13 @@ lp_build_rho(struct lp_build_sample_context *bld,
       if (dims >= 2) {
          LLVMValueRef rho_s, rho_t, rho_r;
 
-         rho_s = LLVMBuildExtractElement(bld->builder, rho_vec, index0, "");
-         rho_t = LLVMBuildExtractElement(bld->builder, rho_vec, index1, "");
+         rho_s = LLVMBuildExtractElement(builder, rho_vec, index0, "");
+         rho_t = LLVMBuildExtractElement(builder, rho_vec, index1, "");
 
          rho = lp_build_max(float_bld, rho_s, rho_t);
 
          if (dims >= 3) {
-            rho_r = LLVMBuildExtractElement(bld->builder, rho_vec, index0, "");
+            rho_r = LLVMBuildExtractElement(builder, rho_vec, index0, "");
             rho = lp_build_max(float_bld, rho, rho_r);
          }
       }
@@ -304,19 +305,19 @@ lp_build_brilinear_lod(struct lp_build_context *bld,
    double post_offset = 1 - factor;
 
    if (0) {
-      lp_build_printf(bld->builder, "lod = %f\n", lod);
+      lp_build_printf(bld->gallivm, "lod = %f\n", lod);
    }
 
    lod = lp_build_add(bld, lod,
-                      lp_build_const_vec(bld->type, pre_offset));
+                      lp_build_const_vec(bld->gallivm, bld->type, pre_offset));
 
    lp_build_ifloor_fract(bld, lod, out_lod_ipart, &lod_fpart);
 
    lod_fpart = lp_build_mul(bld, lod_fpart,
-                            lp_build_const_vec(bld->type, factor));
+                            lp_build_const_vec(bld->gallivm, bld->type, factor));
 
    lod_fpart = lp_build_add(bld, lod_fpart,
-                            lp_build_const_vec(bld->type, post_offset));
+                            lp_build_const_vec(bld->gallivm, bld->type, post_offset));
 
    /*
     * It's not necessary to clamp lod_fpart since:
@@ -327,8 +328,8 @@ lp_build_brilinear_lod(struct lp_build_context *bld,
    *out_lod_fpart = lod_fpart;
 
    if (0) {
-      lp_build_printf(bld->builder, "lod_ipart = %i\n", *out_lod_ipart);
-      lp_build_printf(bld->builder, "lod_fpart = %f\n\n", *out_lod_fpart);
+      lp_build_printf(bld->gallivm, "lod_ipart = %i\n", *out_lod_ipart);
+      lp_build_printf(bld->gallivm, "lod_fpart = %f\n\n", *out_lod_fpart);
    }
 }
 
@@ -363,7 +364,7 @@ lp_build_brilinear_rho(struct lp_build_context *bld,
     * part will not need any post adjustments.
     */
    rho = lp_build_mul(bld, rho,
-                      lp_build_const_vec(bld->type, pre_factor));
+                      lp_build_const_vec(bld->gallivm, bld->type, pre_factor));
 
    /* ipart = ifloor(log2(rho)) */
    lod_ipart = lp_build_extract_exponent(bld, rho, 0);
@@ -372,10 +373,10 @@ lp_build_brilinear_rho(struct lp_build_context *bld,
    lod_fpart = lp_build_extract_mantissa(bld, rho);
 
    lod_fpart = lp_build_mul(bld, lod_fpart,
-                            lp_build_const_vec(bld->type, factor));
+                            lp_build_const_vec(bld->gallivm, bld->type, factor));
 
    lod_fpart = lp_build_add(bld, lod_fpart,
-                            lp_build_const_vec(bld->type, post_offset));
+                            lp_build_const_vec(bld->gallivm, bld->type, post_offset));
 
    /*
     * Like lp_build_brilinear_lod, it's not necessary to clamp lod_fpart since:
@@ -413,6 +414,7 @@ lp_build_lod_selector(struct lp_build_sample_context *bld,
                       LLVMValueRef *out_lod_fpart)
 
 {
+   LLVMBuilderRef builder = bld->gallivm->builder;
    struct lp_build_context *float_bld = &bld->float_bld;
    LLVMValueRef lod;
 
@@ -424,17 +426,17 @@ lp_build_lod_selector(struct lp_build_sample_context *bld,
        * This is hit during mipmap generation.
        */
       LLVMValueRef min_lod =
-         bld->dynamic_state->min_lod(bld->dynamic_state, bld->builder, unit);
+         bld->dynamic_state->min_lod(bld->dynamic_state, bld->gallivm, unit);
 
       lod = min_lod;
    }
    else {
       LLVMValueRef sampler_lod_bias =
-         bld->dynamic_state->lod_bias(bld->dynamic_state, bld->builder, unit);
-      LLVMValueRef index0 = LLVMConstInt(LLVMInt32Type(), 0, 0);
+         bld->dynamic_state->lod_bias(bld->dynamic_state, bld->gallivm, unit);
+      LLVMValueRef index0 = lp_build_const_int32(bld->gallivm, 0);
 
       if (explicit_lod) {
-         lod = LLVMBuildExtractElement(bld->builder, explicit_lod,
+         lod = LLVMBuildExtractElement(builder, explicit_lod,
                                        index0, "");
       }
       else {
@@ -479,27 +481,27 @@ lp_build_lod_selector(struct lp_build_sample_context *bld,
 
          /* add shader lod bias */
          if (lod_bias) {
-            lod_bias = LLVMBuildExtractElement(bld->builder, lod_bias,
+            lod_bias = LLVMBuildExtractElement(builder, lod_bias,
                                                index0, "");
-            lod = LLVMBuildFAdd(bld->builder, lod, lod_bias, "shader_lod_bias");
+            lod = LLVMBuildFAdd(builder, lod, lod_bias, "shader_lod_bias");
          }
       }
 
       /* add sampler lod bias */
       if (bld->static_state->lod_bias_non_zero)
-         lod = LLVMBuildFAdd(bld->builder, lod, sampler_lod_bias, "sampler_lod_bias");
+         lod = LLVMBuildFAdd(builder, lod, sampler_lod_bias, "sampler_lod_bias");
 
 
       /* clamp lod */
       if (bld->static_state->apply_max_lod) {
          LLVMValueRef max_lod =
-            bld->dynamic_state->max_lod(bld->dynamic_state, bld->builder, unit);
+            bld->dynamic_state->max_lod(bld->dynamic_state, bld->gallivm, unit);
 
          lod = lp_build_min(float_bld, lod, max_lod);
       }
       if (bld->static_state->apply_min_lod) {
          LLVMValueRef min_lod =
-            bld->dynamic_state->min_lod(bld->dynamic_state, bld->builder, unit);
+            bld->dynamic_state->min_lod(bld->dynamic_state, bld->gallivm, unit);
 
          lod = lp_build_max(float_bld, lod, min_lod);
       }
@@ -542,10 +544,10 @@ lp_build_nearest_mip_level(struct lp_build_sample_context *bld,
    struct lp_build_context *int_bld = &bld->int_bld;
    LLVMValueRef last_level, level;
 
-   LLVMValueRef zero = LLVMConstInt(LLVMInt32Type(), 0, 0);
+   LLVMValueRef zero = lp_build_const_int32(bld->gallivm, 0);
 
    last_level = bld->dynamic_state->last_level(bld->dynamic_state,
-                                               bld->builder, unit);
+                                               bld->gallivm, unit);
 
    /* convert float lod to integer */
    level = lod_ipart;
@@ -568,7 +570,7 @@ lp_build_linear_mip_levels(struct lp_build_sample_context *bld,
                            LLVMValueRef *level0_out,
                            LLVMValueRef *level1_out)
 {
-   LLVMBuilderRef builder = bld->builder;
+   LLVMBuilderRef builder = bld->gallivm->builder;
    struct lp_build_context *int_bld = &bld->int_bld;
    struct lp_build_context *float_bld = &bld->float_bld;
    LLVMValueRef last_level;
@@ -579,7 +581,7 @@ lp_build_linear_mip_levels(struct lp_build_sample_context *bld,
    *level1_out = lp_build_add(int_bld, lod_ipart, int_bld->one);
 
    last_level = bld->dynamic_state->last_level(bld->dynamic_state,
-                                               bld->builder, unit);
+                                               bld->gallivm, unit);
 
    /*
     * Clamp both lod_ipart and lod_ipart + 1 to [0, last_level], with the
@@ -630,11 +632,13 @@ LLVMValueRef
 lp_build_get_mipmap_level(struct lp_build_sample_context *bld,
                           LLVMValueRef level)
 {
+   LLVMBuilderRef builder = bld->gallivm->builder;
    LLVMValueRef indexes[2], data_ptr;
-   indexes[0] = LLVMConstInt(LLVMInt32Type(), 0, 0);
+
+   indexes[0] = lp_build_const_int32(bld->gallivm, 0);
    indexes[1] = level;
-   data_ptr = LLVMBuildGEP(bld->builder, bld->data_array, indexes, 2, "");
-   data_ptr = LLVMBuildLoad(bld->builder, data_ptr, "");
+   data_ptr = LLVMBuildGEP(builder, bld->data_array, indexes, 2, "");
+   data_ptr = LLVMBuildLoad(builder, data_ptr, "");
    return data_ptr;
 }
 
@@ -643,7 +647,7 @@ LLVMValueRef
 lp_build_get_const_mipmap_level(struct lp_build_sample_context *bld,
                                 int level)
 {
-   LLVMValueRef lvl = LLVMConstInt(LLVMInt32Type(), level, 0);
+   LLVMValueRef lvl = lp_build_const_int32(bld->gallivm, level);
    return lp_build_get_mipmap_level(bld, lvl);
 }
 
@@ -657,6 +661,7 @@ lp_build_minify(struct lp_build_context *bld,
                 LLVMValueRef base_size,
                 LLVMValueRef level)
 {
+   LLVMBuilderRef builder = bld->gallivm->builder;
    assert(lp_check_value(bld->type, base_size));
    assert(lp_check_value(bld->type, level));
 
@@ -666,7 +671,7 @@ lp_build_minify(struct lp_build_context *bld,
    }
    else {
       LLVMValueRef size =
-         LLVMBuildLShr(bld->builder, base_size, level, "minify");
+         LLVMBuildLShr(builder, base_size, level, "minify");
       assert(bld->type.sign);
       size = lp_build_max(bld, size, bld->one);
       return size;
@@ -682,11 +687,12 @@ static LLVMValueRef
 lp_build_get_level_stride_vec(struct lp_build_sample_context *bld,
                               LLVMValueRef stride_array, LLVMValueRef level)
 {
+   LLVMBuilderRef builder = bld->gallivm->builder;
    LLVMValueRef indexes[2], stride;
-   indexes[0] = LLVMConstInt(LLVMInt32Type(), 0, 0);
+   indexes[0] = lp_build_const_int32(bld->gallivm, 0);
    indexes[1] = level;
-   stride = LLVMBuildGEP(bld->builder, stride_array, indexes, 2, "");
-   stride = LLVMBuildLoad(bld->builder, stride, "");
+   stride = LLVMBuildGEP(builder, stride_array, indexes, 2, "");
+   stride = LLVMBuildLoad(builder, stride, "");
    stride = lp_build_broadcast_scalar(&bld->int_coord_bld, stride);
    return stride;
 }
@@ -747,21 +753,21 @@ lp_build_extract_image_sizes(struct lp_build_sample_context *bld,
                              LLVMValueRef *out_depth)
 {
    const unsigned dims = bld->dims;
-   LLVMTypeRef i32t = LLVMInt32Type();
+   LLVMTypeRef i32t = LLVMInt32TypeInContext(bld->gallivm->context);
 
-   *out_width = lp_build_extract_broadcast(bld->builder,
+   *out_width = lp_build_extract_broadcast(bld->gallivm,
                                            size_type,
                                            coord_type,
                                            size,
                                            LLVMConstInt(i32t, 0, 0));
    if (dims >= 2) {
-      *out_height = lp_build_extract_broadcast(bld->builder,
+      *out_height = lp_build_extract_broadcast(bld->gallivm,
                                                size_type,
                                                coord_type,
                                                size,
                                                LLVMConstInt(i32t, 1, 0));
       if (dims == 3) {
-         *out_depth = lp_build_extract_broadcast(bld->builder,
+         *out_depth = lp_build_extract_broadcast(bld->gallivm,
                                                  size_type,
                                                  coord_type,
                                                  size,
@@ -812,7 +818,7 @@ static LLVMValueRef
 lp_build_cube_ima(struct lp_build_context *coord_bld, LLVMValueRef coord)
 {
    /* ima = -0.5 / abs(coord); */
-   LLVMValueRef negHalf = lp_build_const_vec(coord_bld->type, -0.5);
+   LLVMValueRef negHalf = lp_build_const_vec(coord_bld->gallivm, coord_bld->type, -0.5);
    LLVMValueRef absCoord = lp_build_abs(coord_bld, coord);
    LLVMValueRef ima = lp_build_div(coord_bld, negHalf, absCoord);
    return ima;
@@ -831,7 +837,7 @@ lp_build_cube_coord(struct lp_build_context *coord_bld,
                     LLVMValueRef coord, LLVMValueRef ima)
 {
    /* return negate(coord) * ima * sign + 0.5; */
-   LLVMValueRef half = lp_build_const_vec(coord_bld->type, 0.5);
+   LLVMValueRef half = lp_build_const_vec(coord_bld->gallivm, coord_bld->type, 0.5);
    LLVMValueRef res;
 
    assert(negate_coord == +1 || negate_coord == -1);
@@ -859,12 +865,14 @@ lp_build_cube_face(struct lp_build_sample_context *bld,
                    LLVMValueRef major_coord,
                    unsigned pos_face, unsigned neg_face)
 {
-   LLVMValueRef cmp = LLVMBuildFCmp(bld->builder, LLVMRealUGE,
+   struct gallivm_state *gallivm = bld->gallivm;
+   LLVMBuilderRef builder = gallivm->builder;
+   LLVMValueRef cmp = LLVMBuildFCmp(builder, LLVMRealUGE,
                                     major_coord,
                                     bld->float_bld.zero, "");
-   LLVMValueRef pos = LLVMConstInt(LLVMInt32Type(), pos_face, 0);
-   LLVMValueRef neg = LLVMConstInt(LLVMInt32Type(), neg_face, 0);
-   LLVMValueRef res = LLVMBuildSelect(bld->builder, cmp, pos, neg, "");
+   LLVMValueRef pos = lp_build_const_int32(gallivm, pos_face);
+   LLVMValueRef neg = lp_build_const_int32(gallivm, neg_face);
+   LLVMValueRef res = LLVMBuildSelect(builder, cmp, pos, neg, "");
    return res;
 }
 
@@ -884,9 +892,10 @@ lp_build_cube_lookup(struct lp_build_sample_context *bld,
 {
    struct lp_build_context *float_bld = &bld->float_bld;
    struct lp_build_context *coord_bld = &bld->coord_bld;
+   LLVMBuilderRef builder = bld->gallivm->builder;
    LLVMValueRef rx, ry, rz;
    LLVMValueRef arx, ary, arz;
-   LLVMValueRef c25 = LLVMConstReal(LLVMFloatType(), 0.25);
+   LLVMValueRef c25 = lp_build_const_float(bld->gallivm, 0.25);
    LLVMValueRef arx_ge_ary, arx_ge_arz;
    LLVMValueRef ary_ge_arx, ary_ge_arz;
    LLVMValueRef arx_ge_ary_arz, ary_ge_arx_arz;
@@ -911,17 +920,17 @@ lp_build_cube_lookup(struct lp_build_sample_context *bld,
    /*
     * Compare sign/magnitude of rx,ry,rz to determine face
     */
-   arx_ge_ary = LLVMBuildFCmp(bld->builder, LLVMRealUGE, arx, ary, "");
-   arx_ge_arz = LLVMBuildFCmp(bld->builder, LLVMRealUGE, arx, arz, "");
-   ary_ge_arx = LLVMBuildFCmp(bld->builder, LLVMRealUGE, ary, arx, "");
-   ary_ge_arz = LLVMBuildFCmp(bld->builder, LLVMRealUGE, ary, arz, "");
+   arx_ge_ary = LLVMBuildFCmp(builder, LLVMRealUGE, arx, ary, "");
+   arx_ge_arz = LLVMBuildFCmp(builder, LLVMRealUGE, arx, arz, "");
+   ary_ge_arx = LLVMBuildFCmp(builder, LLVMRealUGE, ary, arx, "");
+   ary_ge_arz = LLVMBuildFCmp(builder, LLVMRealUGE, ary, arz, "");
 
-   arx_ge_ary_arz = LLVMBuildAnd(bld->builder, arx_ge_ary, arx_ge_arz, "");
-   ary_ge_arx_arz = LLVMBuildAnd(bld->builder, ary_ge_arx, ary_ge_arz, "");
+   arx_ge_ary_arz = LLVMBuildAnd(builder, arx_ge_ary, arx_ge_arz, "");
+   ary_ge_arx_arz = LLVMBuildAnd(builder, ary_ge_arx, ary_ge_arz, "");
 
-   rx_pos = LLVMBuildFCmp(bld->builder, LLVMRealUGE, rx, float_bld->zero, "");
-   ry_pos = LLVMBuildFCmp(bld->builder, LLVMRealUGE, ry, float_bld->zero, "");
-   rz_pos = LLVMBuildFCmp(bld->builder, LLVMRealUGE, rz, float_bld->zero, "");
+   rx_pos = LLVMBuildFCmp(builder, LLVMRealUGE, rx, float_bld->zero, "");
+   ry_pos = LLVMBuildFCmp(builder, LLVMRealUGE, ry, float_bld->zero, "");
+   rz_pos = LLVMBuildFCmp(builder, LLVMRealUGE, rz, float_bld->zero, "");
 
    {
       struct lp_build_if_state if_ctx;
@@ -929,11 +938,11 @@ lp_build_cube_lookup(struct lp_build_sample_context *bld,
       LLVMValueRef face_t_var;
       LLVMValueRef face_var;
 
-      face_s_var = lp_build_alloca(bld->builder, bld->coord_bld.vec_type, "face_s_var");
-      face_t_var = lp_build_alloca(bld->builder, bld->coord_bld.vec_type, "face_t_var");
-      face_var = lp_build_alloca(bld->builder, bld->int_bld.vec_type, "face_var");
+      face_s_var = lp_build_alloca(bld->gallivm, bld->coord_bld.vec_type, "face_s_var");
+      face_t_var = lp_build_alloca(bld->gallivm, bld->coord_bld.vec_type, "face_t_var");
+      face_var = lp_build_alloca(bld->gallivm, bld->int_bld.vec_type, "face_var");
 
-      lp_build_if(&if_ctx, bld->builder, arx_ge_ary_arz);
+      lp_build_if(&if_ctx, bld->gallivm, arx_ge_ary_arz);
       {
          /* +/- X face */
          LLVMValueRef sign = lp_build_sgn(float_bld, rx);
@@ -943,17 +952,17 @@ lp_build_cube_lookup(struct lp_build_sample_context *bld,
          *face = lp_build_cube_face(bld, rx,
                                     PIPE_TEX_FACE_POS_X,
                                     PIPE_TEX_FACE_NEG_X);
-         LLVMBuildStore(bld->builder, *face_s, face_s_var);
-         LLVMBuildStore(bld->builder, *face_t, face_t_var);
-         LLVMBuildStore(bld->builder, *face, face_var);
+         LLVMBuildStore(builder, *face_s, face_s_var);
+         LLVMBuildStore(builder, *face_t, face_t_var);
+         LLVMBuildStore(builder, *face, face_var);
       }
       lp_build_else(&if_ctx);
       {
          struct lp_build_if_state if_ctx2;
 
-         ary_ge_arx_arz = LLVMBuildAnd(bld->builder, ary_ge_arx, ary_ge_arz, "");
+         ary_ge_arx_arz = LLVMBuildAnd(builder, ary_ge_arx, ary_ge_arz, "");
 
-         lp_build_if(&if_ctx2, bld->builder, ary_ge_arx_arz);
+         lp_build_if(&if_ctx2, bld->gallivm, ary_ge_arx_arz);
          {
             /* +/- Y face */
             LLVMValueRef sign = lp_build_sgn(float_bld, ry);
@@ -963,9 +972,9 @@ lp_build_cube_lookup(struct lp_build_sample_context *bld,
             *face = lp_build_cube_face(bld, ry,
                                        PIPE_TEX_FACE_POS_Y,
                                        PIPE_TEX_FACE_NEG_Y);
-            LLVMBuildStore(bld->builder, *face_s, face_s_var);
-            LLVMBuildStore(bld->builder, *face_t, face_t_var);
-            LLVMBuildStore(bld->builder, *face, face_var);
+            LLVMBuildStore(builder, *face_s, face_s_var);
+            LLVMBuildStore(builder, *face_t, face_t_var);
+            LLVMBuildStore(builder, *face, face_var);
          }
          lp_build_else(&if_ctx2);
          {
@@ -977,18 +986,18 @@ lp_build_cube_lookup(struct lp_build_sample_context *bld,
             *face = lp_build_cube_face(bld, rz,
                                        PIPE_TEX_FACE_POS_Z,
                                        PIPE_TEX_FACE_NEG_Z);
-            LLVMBuildStore(bld->builder, *face_s, face_s_var);
-            LLVMBuildStore(bld->builder, *face_t, face_t_var);
-            LLVMBuildStore(bld->builder, *face, face_var);
+            LLVMBuildStore(builder, *face_s, face_s_var);
+            LLVMBuildStore(builder, *face_t, face_t_var);
+            LLVMBuildStore(builder, *face, face_var);
          }
          lp_build_endif(&if_ctx2);
       }
 
       lp_build_endif(&if_ctx);
 
-      *face_s = LLVMBuildLoad(bld->builder, face_s_var, "face_s");
-      *face_t = LLVMBuildLoad(bld->builder, face_t_var, "face_t");
-      *face   = LLVMBuildLoad(bld->builder, face_var, "face");
+      *face_s = LLVMBuildLoad(builder, face_s_var, "face_s");
+      *face_t = LLVMBuildLoad(builder, face_t_var, "face_t");
+      *face   = LLVMBuildLoad(builder, face_var, "face");
    }
 }
 
@@ -1011,6 +1020,7 @@ lp_build_sample_partial_offset(struct lp_build_context *bld,
                                LLVMValueRef *out_offset,
                                LLVMValueRef *out_subcoord)
 {
+   LLVMBuilderRef builder = bld->gallivm->builder;
    LLVMValueRef offset;
    LLVMValueRef subcoord;
 
@@ -1028,14 +1038,14 @@ lp_build_sample_partial_offset(struct lp_build_context *bld,
        */
 #if 0
       LLVMValueRef block_width = lp_build_const_int_vec(bld->type, block_length);
-      subcoord = LLVMBuildURem(bld->builder, coord, block_width, "");
-      coord    = LLVMBuildUDiv(bld->builder, coord, block_width, "");
+      subcoord = LLVMBuildURem(builder, coord, block_width, "");
+      coord    = LLVMBuildUDiv(builder, coord, block_width, "");
 #else
       unsigned logbase2 = util_unsigned_logbase2(block_length);
-      LLVMValueRef block_shift = lp_build_const_int_vec(bld->type, logbase2);
-      LLVMValueRef block_mask = lp_build_const_int_vec(bld->type, block_length - 1);
-      subcoord = LLVMBuildAnd(bld->builder, coord, block_mask, "");
-      coord = LLVMBuildLShr(bld->builder, coord, block_shift, "");
+      LLVMValueRef block_shift = lp_build_const_int_vec(bld->gallivm, bld->type, logbase2);
+      LLVMValueRef block_mask = lp_build_const_int_vec(bld->gallivm, bld->type, block_length - 1);
+      subcoord = LLVMBuildAnd(builder, coord, block_mask, "");
+      coord = LLVMBuildLShr(builder, coord, block_shift, "");
 #endif
    }
 
@@ -1071,7 +1081,8 @@ lp_build_sample_offset(struct lp_build_context *bld,
    LLVMValueRef x_stride;
    LLVMValueRef offset;
 
-   x_stride = lp_build_const_vec(bld->type, format_desc->block.bits/8);
+   x_stride = lp_build_const_vec(bld->gallivm, bld->type,
+                                 format_desc->block.bits/8);
 
    lp_build_sample_partial_offset(bld,
                                   format_desc->block.width,
diff --git a/src/gallium/auxiliary/gallivm/lp_bld_sample.h b/src/gallium/auxiliary/gallivm/lp_bld_sample.h
index ffed27cee8..8c9d5df4e4 100644
--- a/src/gallium/auxiliary/gallivm/lp_bld_sample.h
+++ b/src/gallium/auxiliary/gallivm/lp_bld_sample.h
@@ -105,64 +105,64 @@ struct lp_sampler_dynamic_state
    /** Obtain the base texture width (returns int32) */
    LLVMValueRef
    (*width)( const struct lp_sampler_dynamic_state *state,
-             LLVMBuilderRef builder,
+             struct gallivm_state *gallivm,
              unsigned unit);
 
    /** Obtain the base texture height (returns int32) */
    LLVMValueRef
    (*height)( const struct lp_sampler_dynamic_state *state,
-              LLVMBuilderRef builder,
+              struct gallivm_state *gallivm,
               unsigned unit);
 
    /** Obtain the base texture depth (returns int32) */
    LLVMValueRef
    (*depth)( const struct lp_sampler_dynamic_state *state,
-             LLVMBuilderRef builder,
+             struct gallivm_state *gallivm,
              unsigned unit);
 
    /** Obtain the number of mipmap levels minus one (returns int32) */
    LLVMValueRef
    (*last_level)( const struct lp_sampler_dynamic_state *state,
-                  LLVMBuilderRef builder,
+                  struct gallivm_state *gallivm,
                   unsigned unit);
 
    /** Obtain stride in bytes between image rows/blocks (returns int32) */
    LLVMValueRef
    (*row_stride)( const struct lp_sampler_dynamic_state *state,
-                  LLVMBuilderRef builder,
+                  struct gallivm_state *gallivm,
                   unsigned unit);
 
    /** Obtain stride in bytes between image slices (returns int32) */
    LLVMValueRef
    (*img_stride)( const struct lp_sampler_dynamic_state *state,
-                  LLVMBuilderRef builder,
+                  struct gallivm_state *gallivm,
                   unsigned unit);
 
    /** Obtain pointer to array of pointers to mimpap levels */
    LLVMValueRef
    (*data_ptr)( const struct lp_sampler_dynamic_state *state,
-                LLVMBuilderRef builder,
+                struct gallivm_state *gallivm,
                 unsigned unit);
 
    /** Obtain texture min lod (returns float) */
    LLVMValueRef
    (*min_lod)(const struct lp_sampler_dynamic_state *state,
-              LLVMBuilderRef builder, unsigned unit);
+              struct gallivm_state *gallivm, unsigned unit);
 
    /** Obtain texture max lod (returns float) */
    LLVMValueRef
    (*max_lod)(const struct lp_sampler_dynamic_state *state,
-              LLVMBuilderRef builder, unsigned unit);
+              struct gallivm_state *gallivm, unsigned unit);
 
    /** Obtain texture lod bias (returns float) */
    LLVMValueRef
    (*lod_bias)(const struct lp_sampler_dynamic_state *state,
-               LLVMBuilderRef builder, unsigned unit);
+               struct gallivm_state *gallivm, unsigned unit);
 
    /** Obtain texture border color (returns ptr to float[4]) */
    LLVMValueRef
    (*border_color)(const struct lp_sampler_dynamic_state *state,
-                   LLVMBuilderRef builder, unsigned unit);
+                   struct gallivm_state *gallivm, unsigned unit);
 };
 
 
@@ -171,7 +171,7 @@ struct lp_sampler_dynamic_state
  */
 struct lp_build_sample_context
 {
-   LLVMBuilderRef builder;
+   struct gallivm_state *gallivm;
 
    const struct lp_sampler_static_state *static_state;
 
@@ -385,7 +385,7 @@ lp_build_sample_offset(struct lp_build_context *bld,
 
 
 void
-lp_build_sample_soa(LLVMBuilderRef builder,
+lp_build_sample_soa(struct gallivm_state *gallivm,
                     const struct lp_sampler_static_state *static_state,
                     struct lp_sampler_dynamic_state *dynamic_state,
                     struct lp_type fp_type,
@@ -399,7 +399,7 @@ lp_build_sample_soa(LLVMBuilderRef builder,
                     LLVMValueRef texel_out[4]);
 
 void
-lp_build_sample_nop(struct lp_type type,
+lp_build_sample_nop(struct gallivm_state *gallivm, struct lp_type type,
                     LLVMValueRef texel_out[4]);
 
 
diff --git a/src/gallium/auxiliary/gallivm/lp_bld_sample_aos.c b/src/gallium/auxiliary/gallivm/lp_bld_sample_aos.c
index d6831a580b..e61cf9541e 100644
--- a/src/gallium/auxiliary/gallivm/lp_bld_sample_aos.c
+++ b/src/gallium/auxiliary/gallivm/lp_bld_sample_aos.c
@@ -52,6 +52,7 @@
 #include "lp_bld_flow.h"
 #include "lp_bld_gather.h"
 #include "lp_bld_format.h"
+#include "lp_bld_init.h"
 #include "lp_bld_sample.h"
 #include "lp_bld_sample_aos.h"
 #include "lp_bld_quad.h"
@@ -82,6 +83,7 @@ lp_build_sample_wrap_nearest_int(struct lp_build_sample_context *bld,
                                  LLVMValueRef *out_i)
 {
    struct lp_build_context *int_coord_bld = &bld->int_coord_bld;
+   LLVMBuilderRef builder = bld->gallivm->builder;
    LLVMValueRef length_minus_one;
 
    length_minus_one = lp_build_sub(int_coord_bld, length, int_coord_bld->one);
@@ -89,12 +91,12 @@ lp_build_sample_wrap_nearest_int(struct lp_build_sample_context *bld,
    switch(wrap_mode) {
    case PIPE_TEX_WRAP_REPEAT:
       if(is_pot)
-         coord = LLVMBuildAnd(bld->builder, coord, length_minus_one, "");
+         coord = LLVMBuildAnd(builder, coord, length_minus_one, "");
       else {
          /* Add a bias to the texcoord to handle negative coords */
          LLVMValueRef bias = lp_build_mul_imm(int_coord_bld, length, 1024);
-         coord = LLVMBuildAdd(bld->builder, coord, bias, "");
-         coord = LLVMBuildURem(bld->builder, coord, length, "");
+         coord = LLVMBuildAdd(builder, coord, bias, "");
+         coord = LLVMBuildURem(builder, coord, length, "");
       }
       break;
 
@@ -147,6 +149,7 @@ lp_build_sample_wrap_linear_int(struct lp_build_sample_context *bld,
                                 LLVMValueRef *i1)
 {
    struct lp_build_context *int_coord_bld = &bld->int_coord_bld;
+   LLVMBuilderRef builder = bld->gallivm->builder;
    LLVMValueRef length_minus_one;
    LLVMValueRef lmask, umask, mask;
 
@@ -195,39 +198,39 @@ lp_build_sample_wrap_linear_int(struct lp_build_sample_context *bld,
    switch(wrap_mode) {
    case PIPE_TEX_WRAP_REPEAT:
       if (is_pot) {
-         coord0 = LLVMBuildAnd(bld->builder, coord0, length_minus_one, "");
+         coord0 = LLVMBuildAnd(builder, coord0, length_minus_one, "");
       }
       else {
          /* Add a bias to the texcoord to handle negative coords */
          LLVMValueRef bias = lp_build_mul_imm(int_coord_bld, length, 1024);
-         coord0 = LLVMBuildAdd(bld->builder, coord0, bias, "");
-         coord0 = LLVMBuildURem(bld->builder, coord0, length, "");
+         coord0 = LLVMBuildAdd(builder, coord0, bias, "");
+         coord0 = LLVMBuildURem(builder, coord0, length, "");
       }
 
-      mask = lp_build_compare(bld->builder, int_coord_bld->type,
+      mask = lp_build_compare(bld->gallivm, int_coord_bld->type,
                               PIPE_FUNC_NOTEQUAL, coord0, length_minus_one);
 
       *offset0 = lp_build_mul(int_coord_bld, coord0, stride);
-      *offset1 = LLVMBuildAnd(bld->builder,
+      *offset1 = LLVMBuildAnd(builder,
                               lp_build_add(int_coord_bld, *offset0, stride),
                               mask, "");
       break;
 
    case PIPE_TEX_WRAP_CLAMP_TO_EDGE:
-      lmask = lp_build_compare(int_coord_bld->builder, int_coord_bld->type,
+      lmask = lp_build_compare(int_coord_bld->gallivm, int_coord_bld->type,
                                PIPE_FUNC_GEQUAL, coord0, int_coord_bld->zero);
-      umask = lp_build_compare(int_coord_bld->builder, int_coord_bld->type,
+      umask = lp_build_compare(int_coord_bld->gallivm, int_coord_bld->type,
                                PIPE_FUNC_LESS, coord0, length_minus_one);
 
       coord0 = lp_build_select(int_coord_bld, lmask, coord0, int_coord_bld->zero);
       coord0 = lp_build_select(int_coord_bld, umask, coord0, length_minus_one);
 
-      mask = LLVMBuildAnd(bld->builder, lmask, umask, "");
+      mask = LLVMBuildAnd(builder, lmask, umask, "");
 
       *offset0 = lp_build_mul(int_coord_bld, coord0, stride);
       *offset1 = lp_build_add(int_coord_bld,
                               *offset0,
-                              LLVMBuildAnd(bld->builder, stride, mask, ""));
+                              LLVMBuildAnd(builder, stride, mask, ""));
       break;
 
    case PIPE_TEX_WRAP_CLAMP:
@@ -263,7 +266,7 @@ lp_build_sample_image_nearest(struct lp_build_sample_context *bld,
                               LLVMValueRef *colors_hi)
 {
    const unsigned dims = bld->dims;
-   LLVMBuilderRef builder = bld->builder;
+   LLVMBuilderRef builder = bld->gallivm->builder;
    struct lp_build_context i32, h16, u8n;
    LLVMTypeRef i32_vec_type, h16_vec_type, u8n_vec_type;
    LLVMValueRef i32_c8;
@@ -273,13 +276,13 @@ lp_build_sample_image_nearest(struct lp_build_sample_context *bld,
    LLVMValueRef x_offset, offset;
    LLVMValueRef x_subcoord, y_subcoord, z_subcoord;
 
-   lp_build_context_init(&i32, builder, lp_type_int_vec(32));
-   lp_build_context_init(&h16, builder, lp_type_ufixed(16));
-   lp_build_context_init(&u8n, builder, lp_type_unorm(8));
+   lp_build_context_init(&i32, bld->gallivm, lp_type_int_vec(32));
+   lp_build_context_init(&h16, bld->gallivm, lp_type_ufixed(16));
+   lp_build_context_init(&u8n, bld->gallivm, lp_type_unorm(8));
 
-   i32_vec_type = lp_build_vec_type(i32.type);
-   h16_vec_type = lp_build_vec_type(h16.type);
-   u8n_vec_type = lp_build_vec_type(u8n.type);
+   i32_vec_type = lp_build_vec_type(bld->gallivm, i32.type);
+   h16_vec_type = lp_build_vec_type(bld->gallivm, h16.type);
+   u8n_vec_type = lp_build_vec_type(bld->gallivm, u8n.type);
 
    lp_build_extract_image_sizes(bld,
                                 bld->int_size_type,
@@ -317,7 +320,7 @@ lp_build_sample_image_nearest(struct lp_build_sample_context *bld,
       r = LLVMBuildFPToSI(builder, r, i32_vec_type, "");
 
    /* compute floor (shift right 8) */
-   i32_c8 = lp_build_const_int_vec(i32.type, 8);
+   i32_c8 = lp_build_const_int_vec(bld->gallivm, i32.type, 8);
    s_ipart = LLVMBuildAShr(builder, s, i32_c8, "");
    if (dims >= 2)
       t_ipart = LLVMBuildAShr(builder, t, i32_c8, "");
@@ -325,7 +328,8 @@ lp_build_sample_image_nearest(struct lp_build_sample_context *bld,
       r_ipart = LLVMBuildAShr(builder, r, i32_c8, "");
 
    /* get pixel, row, image strides */
-   x_stride = lp_build_const_vec(bld->int_coord_bld.type,
+   x_stride = lp_build_const_vec(bld->gallivm,
+                                 bld->int_coord_bld.type,
                                  bld->format_desc->block.bits/8);
 
    /* Do texcoord wrapping, compute texel offset */
@@ -387,7 +391,7 @@ lp_build_sample_image_nearest(struct lp_build_sample_context *bld,
           * Given the format is a rgba8, just read the pixels as is,
           * without any swizzling. Swizzling will be done later.
           */
-         rgba8 = lp_build_gather(bld->builder,
+         rgba8 = lp_build_gather(bld->gallivm,
                                  bld->texel_type.length,
                                  bld->format_desc->block.bits,
                                  bld->texel_type.width,
@@ -396,7 +400,7 @@ lp_build_sample_image_nearest(struct lp_build_sample_context *bld,
          rgba8 = LLVMBuildBitCast(builder, rgba8, u8n_vec_type, "");
       }
       else {
-         rgba8 = lp_build_fetch_rgba_aos(bld->builder,
+         rgba8 = lp_build_fetch_rgba_aos(bld->gallivm,
                                          bld->format_desc,
                                          u8n.type,
                                          data_ptr, offset,
@@ -405,7 +409,7 @@ lp_build_sample_image_nearest(struct lp_build_sample_context *bld,
       }
 
       /* Expand one 4*rgba8 to two 2*rgba16 */
-      lp_build_unpack2(builder, u8n.type, h16.type,
+      lp_build_unpack2(bld->gallivm, u8n.type, h16.type,
                        rgba8,
                        colors_lo, colors_hi);
    }
@@ -429,7 +433,7 @@ lp_build_sample_image_linear(struct lp_build_sample_context *bld,
                              LLVMValueRef *colors_hi)
 {
    const unsigned dims = bld->dims;
-   LLVMBuilderRef builder = bld->builder;
+   LLVMBuilderRef builder = bld->gallivm->builder;
    struct lp_build_context i32, h16, u8n;
    LLVMTypeRef i32_vec_type, h16_vec_type, u8n_vec_type;
    LLVMValueRef i32_c8, i32_c128, i32_c255;
@@ -450,13 +454,13 @@ lp_build_sample_image_linear(struct lp_build_sample_context *bld,
    unsigned i, j, k;
    unsigned numj, numk;
 
-   lp_build_context_init(&i32, builder, lp_type_int_vec(32));
-   lp_build_context_init(&h16, builder, lp_type_ufixed(16));
-   lp_build_context_init(&u8n, builder, lp_type_unorm(8));
+   lp_build_context_init(&i32, bld->gallivm, lp_type_int_vec(32));
+   lp_build_context_init(&h16, bld->gallivm, lp_type_ufixed(16));
+   lp_build_context_init(&u8n, bld->gallivm, lp_type_unorm(8));
 
-   i32_vec_type = lp_build_vec_type(i32.type);
-   h16_vec_type = lp_build_vec_type(h16.type);
-   u8n_vec_type = lp_build_vec_type(u8n.type);
+   i32_vec_type = lp_build_vec_type(bld->gallivm, i32.type);
+   h16_vec_type = lp_build_vec_type(bld->gallivm, h16.type);
+   u8n_vec_type = lp_build_vec_type(bld->gallivm, u8n.type);
 
    lp_build_extract_image_sizes(bld,
                                 bld->int_size_type,
@@ -494,7 +498,7 @@ lp_build_sample_image_linear(struct lp_build_sample_context *bld,
       r = LLVMBuildFPToSI(builder, r, i32_vec_type, "");
 
    /* subtract 0.5 (add -128) */
-   i32_c128 = lp_build_const_int_vec(i32.type, -128);
+   i32_c128 = lp_build_const_int_vec(bld->gallivm, i32.type, -128);
    s = LLVMBuildAdd(builder, s, i32_c128, "");
    if (dims >= 2) {
       t = LLVMBuildAdd(builder, t, i32_c128, "");
@@ -504,7 +508,7 @@ lp_build_sample_image_linear(struct lp_build_sample_context *bld,
    }
 
    /* compute floor (shift right 8) */
-   i32_c8 = lp_build_const_int_vec(i32.type, 8);
+   i32_c8 = lp_build_const_int_vec(bld->gallivm, i32.type, 8);
    s_ipart = LLVMBuildAShr(builder, s, i32_c8, "");
    if (dims >= 2)
       t_ipart = LLVMBuildAShr(builder, t, i32_c8, "");
@@ -512,7 +516,7 @@ lp_build_sample_image_linear(struct lp_build_sample_context *bld,
       r_ipart = LLVMBuildAShr(builder, r, i32_c8, "");
 
    /* compute fractional part (AND with 0xff) */
-   i32_c255 = lp_build_const_int_vec(i32.type, 255);
+   i32_c255 = lp_build_const_int_vec(bld->gallivm, i32.type, 255);
    s_fpart = LLVMBuildAnd(builder, s, i32_c255, "");
    if (dims >= 2)
       t_fpart = LLVMBuildAnd(builder, t, i32_c255, "");
@@ -520,7 +524,7 @@ lp_build_sample_image_linear(struct lp_build_sample_context *bld,
       r_fpart = LLVMBuildAnd(builder, r, i32_c255, "");
 
    /* get pixel, row and image strides */
-   x_stride = lp_build_const_vec(bld->int_coord_bld.type,
+   x_stride = lp_build_const_vec(bld->gallivm, bld->int_coord_bld.type,
                                  bld->format_desc->block.bits/8);
    y_stride = row_stride_vec;
    z_stride = img_stride_vec;
@@ -612,7 +616,7 @@ lp_build_sample_image_linear(struct lp_build_sample_context *bld,
       r_fpart = LLVMBuildBitCast(builder, r_fpart, h16_vec_type, "");
 
    {
-      LLVMTypeRef elem_type = LLVMInt32Type();
+      LLVMTypeRef elem_type = LLVMInt32TypeInContext(bld->gallivm->context);
       LLVMValueRef shuffles_lo[LP_MAX_VECTOR_LENGTH];
       LLVMValueRef shuffles_hi[LP_MAX_VECTOR_LENGTH];
       LLVMValueRef shuffle_lo;
@@ -685,7 +689,7 @@ lp_build_sample_image_linear(struct lp_build_sample_context *bld,
                 * Given the format is a rgba8, just read the pixels as is,
                 * without any swizzling. Swizzling will be done later.
                 */
-               rgba8 = lp_build_gather(bld->builder,
+               rgba8 = lp_build_gather(bld->gallivm,
                                        bld->texel_type.length,
                                        bld->format_desc->block.bits,
                                        bld->texel_type.width,
@@ -694,7 +698,7 @@ lp_build_sample_image_linear(struct lp_build_sample_context *bld,
                rgba8 = LLVMBuildBitCast(builder, rgba8, u8n_vec_type, "");
             }
             else {
-               rgba8 = lp_build_fetch_rgba_aos(bld->builder,
+               rgba8 = lp_build_fetch_rgba_aos(bld->gallivm,
                                                bld->format_desc,
                                                u8n.type,
                                                data_ptr, offset[k][j][i],
@@ -703,7 +707,7 @@ lp_build_sample_image_linear(struct lp_build_sample_context *bld,
             }
 
             /* Expand one 4*rgba8 to two 2*rgba16 */
-            lp_build_unpack2(builder, u8n.type, h16.type,
+            lp_build_unpack2(bld->gallivm, u8n.type, h16.type,
                              rgba8,
                              &neighbors_lo[k][j][i], &neighbors_hi[k][j][i]);
          }
@@ -790,7 +794,7 @@ lp_build_sample_mipmap(struct lp_build_sample_context *bld,
                        LLVMValueRef colors_lo_var,
                        LLVMValueRef colors_hi_var)
 {
-   LLVMBuilderRef builder = bld->builder;
+   LLVMBuilderRef builder = bld->gallivm->builder;
    LLVMValueRef size0;
    LLVMValueRef size1;
    LLVMValueRef row_stride0_vec;
@@ -802,7 +806,6 @@ lp_build_sample_mipmap(struct lp_build_sample_context *bld,
    LLVMValueRef colors0_lo, colors0_hi;
    LLVMValueRef colors1_lo, colors1_hi;
 
-
    /* sample the first mipmap level */
    lp_build_mipmap_level_sizes(bld, ilevel0,
                                &size0,
@@ -829,8 +832,8 @@ lp_build_sample_mipmap(struct lp_build_sample_context *bld,
    LLVMBuildStore(builder, colors0_hi, colors_hi_var);
 
    if (mip_filter == PIPE_TEX_MIPFILTER_LINEAR) {
-      LLVMValueRef h16_scale = LLVMConstReal(LLVMFloatType(), 256.0);
-      LLVMTypeRef i32_type = LLVMIntType(32);
+      LLVMValueRef h16_scale = lp_build_const_float(bld->gallivm, 256.0);
+      LLVMTypeRef i32_type = LLVMIntTypeInContext(bld->gallivm->context, 32);
       struct lp_build_if_state if_ctx;
       LLVMValueRef need_lerp;
 
@@ -842,11 +845,11 @@ lp_build_sample_mipmap(struct lp_build_sample_context *bld,
                                 lod_fpart, LLVMConstNull(i32_type),
                                 "need_lerp");
 
-      lp_build_if(&if_ctx, builder, need_lerp);
+      lp_build_if(&if_ctx, bld->gallivm, need_lerp);
       {
          struct lp_build_context h16_bld;
 
-         lp_build_context_init(&h16_bld, builder, lp_type_ufixed(16));
+         lp_build_context_init(&h16_bld, bld->gallivm, lp_type_ufixed(16));
 
          /* sample the second mipmap level */
          lp_build_mipmap_level_sizes(bld, ilevel1,
@@ -885,7 +888,7 @@ lp_build_sample_mipmap(struct lp_build_sample_context *bld,
             int i;
             assert(h16_bld.type.length <= Elements(shuffles));
             for (i = 0; i < h16_bld.type.length; i++)
-               shuffles[i] = lp_build_const_int32(2 * (i & 1));
+               shuffles[i] = lp_build_const_int32(bld->gallivm, 2 * (i & 1));
             shuffle = LLVMConstVector(shuffles, h16_bld.type.length);
             lod_fpart = LLVMBuildShuffleVector(builder,
                                                lod_fpart, lod_fpart,
@@ -925,7 +928,7 @@ lp_build_sample_aos(struct lp_build_sample_context *bld,
                     LLVMValueRef texel_out[4])
 {
    struct lp_build_context *int_bld = &bld->int_bld;
-   LLVMBuilderRef builder = bld->builder;
+   LLVMBuilderRef builder = bld->gallivm->builder;
    const unsigned mip_filter = bld->static_state->min_mip_filter;
    const unsigned min_filter = bld->static_state->min_img_filter;
    const unsigned mag_filter = bld->static_state->mag_img_filter;
@@ -936,8 +939,7 @@ lp_build_sample_aos(struct lp_build_sample_context *bld,
    LLVMValueRef unswizzled[4];
    LLVMValueRef face_ddx[4], face_ddy[4];
    struct lp_build_context h16_bld;
-   LLVMTypeRef i32t = LLVMInt32Type();
-   LLVMValueRef i32t_zero = LLVMConstInt(i32t, 0, 0);
+   LLVMValueRef i32t_zero = lp_build_const_int32(bld->gallivm, 0);
 
    /* we only support the common/simple wrap modes at this time */
    assert(lp_is_simple_wrap_mode(bld->static_state->wrap_s));
@@ -948,7 +950,7 @@ lp_build_sample_aos(struct lp_build_sample_context *bld,
 
 
    /* make 16-bit fixed-pt builder context */
-   lp_build_context_init(&h16_bld, builder, lp_type_ufixed(16));
+   lp_build_context_init(&h16_bld, bld->gallivm, lp_type_ufixed(16));
 
    /* cube face selection, compute pre-face coords, etc. */
    if (bld->static_state->target == PIPE_TEXTURE_CUBE) {
@@ -1026,8 +1028,8 @@ lp_build_sample_aos(struct lp_build_sample_context *bld,
     * Get/interpolate texture colors.
     */
 
-   packed_lo = lp_build_alloca(builder, h16_bld.vec_type, "packed_lo");
-   packed_hi = lp_build_alloca(builder, h16_bld.vec_type, "packed_hi");
+   packed_lo = lp_build_alloca(bld->gallivm, h16_bld.vec_type, "packed_lo");
+   packed_hi = lp_build_alloca(bld->gallivm, h16_bld.vec_type, "packed_hi");
 
    if (min_filter == mag_filter) {
       /* no need to distinquish between minification and magnification */
@@ -1048,7 +1050,7 @@ lp_build_sample_aos(struct lp_build_sample_context *bld,
       minify = LLVMBuildICmp(builder, LLVMIntSGE,
                              lod_ipart, int_bld->zero, "");
 
-      lp_build_if(&if_ctx, builder, minify);
+      lp_build_if(&if_ctx, bld->gallivm, minify);
       {
          /* Use the minification filter */
          lp_build_sample_mipmap(bld,
@@ -1073,7 +1075,7 @@ lp_build_sample_aos(struct lp_build_sample_context *bld,
     * combine the values stored in 'packed_lo' and 'packed_hi' variables
     * into 'packed'
     */
-   packed = lp_build_pack2(builder,
+   packed = lp_build_pack2(bld->gallivm,
                            h16_bld.type, lp_type_unorm(8),
                            LLVMBuildLoad(builder, packed_lo, ""),
                            LLVMBuildLoad(builder, packed_hi, ""));
@@ -1081,7 +1083,7 @@ lp_build_sample_aos(struct lp_build_sample_context *bld,
    /*
     * Convert to SoA and swizzle.
     */
-   lp_build_rgba8_to_f32_soa(builder,
+   lp_build_rgba8_to_f32_soa(bld->gallivm,
                              bld->texel_type,
                              packed, unswizzled);
 
@@ -1096,6 +1098,4 @@ lp_build_sample_aos(struct lp_build_sample_context *bld,
       texel_out[2] = unswizzled[2];
       texel_out[3] = unswizzled[3];
    }
-
-   apply_sampler_swizzle(bld, texel_out);
 }
diff --git a/src/gallium/auxiliary/gallivm/lp_bld_sample_soa.c b/src/gallium/auxiliary/gallivm/lp_bld_sample_soa.c
index 53cc0c5f34..e685f4b73f 100644
--- a/src/gallium/auxiliary/gallivm/lp_bld_sample_soa.c
+++ b/src/gallium/auxiliary/gallivm/lp_bld_sample_soa.c
@@ -84,6 +84,7 @@ lp_build_sample_texel_soa(struct lp_build_sample_context *bld,
    const struct lp_sampler_static_state *static_state = bld->static_state;
    const unsigned dims = bld->dims;
    struct lp_build_context *int_coord_bld = &bld->int_coord_bld;
+   LLVMBuilderRef builder = bld->gallivm->builder;
    LLVMValueRef offset;
    LLVMValueRef i, j;
    LLVMValueRef use_border = NULL;
@@ -95,7 +96,7 @@ lp_build_sample_texel_soa(struct lp_build_sample_context *bld,
       LLVMValueRef b1, b2;
       b1 = lp_build_cmp(int_coord_bld, PIPE_FUNC_LESS, x, int_coord_bld->zero);
       b2 = lp_build_cmp(int_coord_bld, PIPE_FUNC_GEQUAL, x, width);
-      use_border = LLVMBuildOr(bld->builder, b1, b2, "b1_or_b2");
+      use_border = LLVMBuildOr(builder, b1, b2, "b1_or_b2");
    }
 
    if (dims >= 2 &&
@@ -106,11 +107,11 @@ lp_build_sample_texel_soa(struct lp_build_sample_context *bld,
       b1 = lp_build_cmp(int_coord_bld, PIPE_FUNC_LESS, y, int_coord_bld->zero);
       b2 = lp_build_cmp(int_coord_bld, PIPE_FUNC_GEQUAL, y, height);
       if (use_border) {
-         use_border = LLVMBuildOr(bld->builder, use_border, b1, "ub_or_b1");
-         use_border = LLVMBuildOr(bld->builder, use_border, b2, "ub_or_b2");
+         use_border = LLVMBuildOr(builder, use_border, b1, "ub_or_b1");
+         use_border = LLVMBuildOr(builder, use_border, b2, "ub_or_b2");
       }
       else {
-         use_border = LLVMBuildOr(bld->builder, b1, b2, "b1_or_b2");
+         use_border = LLVMBuildOr(builder, b1, b2, "b1_or_b2");
       }
    }
 
@@ -122,11 +123,11 @@ lp_build_sample_texel_soa(struct lp_build_sample_context *bld,
       b1 = lp_build_cmp(int_coord_bld, PIPE_FUNC_LESS, z, int_coord_bld->zero);
       b2 = lp_build_cmp(int_coord_bld, PIPE_FUNC_GEQUAL, z, depth);
       if (use_border) {
-         use_border = LLVMBuildOr(bld->builder, use_border, b1, "ub_or_b1");
-         use_border = LLVMBuildOr(bld->builder, use_border, b2, "ub_or_b2");
+         use_border = LLVMBuildOr(builder, use_border, b1, "ub_or_b1");
+         use_border = LLVMBuildOr(builder, use_border, b2, "ub_or_b2");
       }
       else {
-         use_border = LLVMBuildOr(bld->builder, b1, b2, "b1_or_b2");
+         use_border = LLVMBuildOr(builder, b1, b2, "b1_or_b2");
       }
    }
 
@@ -148,7 +149,7 @@ lp_build_sample_texel_soa(struct lp_build_sample_context *bld,
       offset = lp_build_andnot(&bld->int_coord_bld, offset, use_border);
    }
 
-   lp_build_fetch_rgba_soa(bld->builder,
+   lp_build_fetch_rgba_soa(bld->gallivm,
                            bld->format_desc,
                            bld->texel_type,
                            data_ptr, offset,
@@ -174,20 +175,18 @@ lp_build_sample_texel_soa(struct lp_build_sample_context *bld,
       /* select texel color or border color depending on use_border */
       LLVMValueRef border_color_ptr = 
          bld->dynamic_state->border_color(bld->dynamic_state,
-                                          bld->builder, unit);
+                                          bld->gallivm, unit);
       int chan;
       for (chan = 0; chan < 4; chan++) {
          LLVMValueRef border_chan =
-            lp_build_array_get(bld->builder, border_color_ptr,
-                               lp_build_const_int32(chan));
+            lp_build_array_get(bld->gallivm, border_color_ptr,
+                               lp_build_const_int32(bld->gallivm, chan));
          LLVMValueRef border_chan_vec =
             lp_build_broadcast_scalar(&bld->float_vec_bld, border_chan);
          texel_out[chan] = lp_build_select(&bld->texel_bld, use_border,
                                            border_chan_vec, texel_out[chan]);
       }
    }
-
-   apply_sampler_swizzle(bld, texel_out);
 }
 
 
@@ -205,7 +204,7 @@ lp_build_coord_mirror(struct lp_build_sample_context *bld,
    lp_build_ifloor_fract(coord_bld, coord, &flr, &fract);
 
    /* isOdd = flr & 1 */
-   isOdd = LLVMBuildAnd(bld->builder, flr, int_coord_bld->one, "");
+   isOdd = LLVMBuildAnd(bld->gallivm->builder, flr, int_coord_bld->one, "");
 
    /* make coord positive or negative depending on isOdd */
    coord = lp_build_set_sign(coord_bld, fract, isOdd);
@@ -239,7 +238,8 @@ lp_build_sample_wrap_linear(struct lp_build_sample_context *bld,
 {
    struct lp_build_context *coord_bld = &bld->coord_bld;
    struct lp_build_context *int_coord_bld = &bld->int_coord_bld;
-   LLVMValueRef half = lp_build_const_vec(coord_bld->type, 0.5);
+   LLVMBuilderRef builder = bld->gallivm->builder;
+   LLVMValueRef half = lp_build_const_vec(bld->gallivm, coord_bld->type, 0.5);
    LLVMValueRef length_minus_one = lp_build_sub(int_coord_bld, length, int_coord_bld->one);
    LLVMValueRef coord0, coord1, weight;
 
@@ -253,18 +253,18 @@ lp_build_sample_wrap_linear(struct lp_build_sample_context *bld,
       /* repeat wrap */
       if (is_pot) {
          coord1 = lp_build_add(int_coord_bld, coord0, int_coord_bld->one);
-         coord0 = LLVMBuildAnd(bld->builder, coord0, length_minus_one, "");
-         coord1 = LLVMBuildAnd(bld->builder, coord1, length_minus_one, "");
+         coord0 = LLVMBuildAnd(builder, coord0, length_minus_one, "");
+         coord1 = LLVMBuildAnd(builder, coord1, length_minus_one, "");
       }
       else {
          /* Add a bias to the texcoord to handle negative coords */
          LLVMValueRef bias = lp_build_mul_imm(int_coord_bld, length, 1024);
          LLVMValueRef mask;
-         coord0 = LLVMBuildAdd(bld->builder, coord0, bias, "");
-         coord0 = LLVMBuildURem(bld->builder, coord0, length, "");
-         mask = lp_build_compare(bld->builder, int_coord_bld->type,
+         coord0 = LLVMBuildAdd(builder, coord0, bias, "");
+         coord0 = LLVMBuildURem(builder, coord0, length, "");
+         mask = lp_build_compare(bld->gallivm, int_coord_bld->type,
                                  PIPE_FUNC_NOTEQUAL, coord0, length_minus_one);
-         coord1 = LLVMBuildAnd(bld->builder,
+         coord1 = LLVMBuildAnd(builder,
                               lp_build_add(int_coord_bld, coord0, int_coord_bld->one),
                               mask, "");
       }
@@ -318,7 +318,7 @@ lp_build_sample_wrap_linear(struct lp_build_sample_context *bld,
          }
          /* was: clamp to [-0.5, length + 0.5], then sub 0.5 */
          coord = lp_build_sub(coord_bld, coord, half);
-         min = lp_build_const_vec(coord_bld->type, -1.0F);
+         min = lp_build_const_vec(bld->gallivm, coord_bld->type, -1.0F);
          coord = lp_build_clamp(coord_bld, coord, min, length_f);
          /* convert to int, compute lerp weight */
          lp_build_ifloor_fract(coord_bld, coord, &coord0, &weight);
@@ -437,6 +437,7 @@ lp_build_sample_wrap_nearest(struct lp_build_sample_context *bld,
 {
    struct lp_build_context *coord_bld = &bld->coord_bld;
    struct lp_build_context *int_coord_bld = &bld->int_coord_bld;
+   LLVMBuilderRef builder = bld->gallivm->builder;
    LLVMValueRef length_minus_one = lp_build_sub(int_coord_bld, length, int_coord_bld->one);
    LLVMValueRef icoord;
    
@@ -445,12 +446,12 @@ lp_build_sample_wrap_nearest(struct lp_build_sample_context *bld,
       coord = lp_build_mul(coord_bld, coord, length_f);
       icoord = lp_build_ifloor(coord_bld, coord);
       if (is_pot)
-         icoord = LLVMBuildAnd(bld->builder, icoord, length_minus_one, "");
+         icoord = LLVMBuildAnd(builder, icoord, length_minus_one, "");
       else {
          /* Add a bias to the texcoord to handle negative coords */
          LLVMValueRef bias = lp_build_mul_imm(int_coord_bld, length, 1024);
-         icoord = LLVMBuildAdd(bld->builder, icoord, bias, "");
-         icoord = LLVMBuildURem(bld->builder, icoord, length, "");
+         icoord = LLVMBuildAdd(builder, icoord, bias, "");
+         icoord = LLVMBuildURem(builder, icoord, length, "");
       }
       break;
 
@@ -830,7 +831,7 @@ lp_build_sample_mipmap(struct lp_build_sample_context *bld,
                        LLVMValueRef lod_fpart,
                        LLVMValueRef *colors_out)
 {
-   LLVMBuilderRef builder = bld->builder;
+   LLVMBuilderRef builder = bld->gallivm->builder;
    LLVMValueRef size0;
    LLVMValueRef size1;
    LLVMValueRef row_stride0_vec;
@@ -878,7 +879,7 @@ lp_build_sample_mipmap(struct lp_build_sample_context *bld,
                                 bld->float_bld.zero,
                                 "need_lerp");
 
-      lp_build_if(&if_ctx, builder, need_lerp);
+      lp_build_if(&if_ctx, bld->gallivm, need_lerp);
       {
          /* sample the second mipmap level */
          lp_build_mipmap_level_sizes(bld, ilevel1,
@@ -934,7 +935,7 @@ lp_build_sample_general(struct lp_build_sample_context *bld,
                         LLVMValueRef *colors_out)
 {
    struct lp_build_context *int_bld = &bld->int_bld;
-   LLVMBuilderRef builder = bld->builder;
+   LLVMBuilderRef builder = bld->gallivm->builder;
    const unsigned mip_filter = bld->static_state->min_mip_filter;
    const unsigned min_filter = bld->static_state->min_img_filter;
    const unsigned mag_filter = bld->static_state->mag_img_filter;
@@ -942,8 +943,7 @@ lp_build_sample_general(struct lp_build_sample_context *bld,
    LLVMValueRef ilevel0, ilevel1 = NULL;
    LLVMValueRef face_ddx[4], face_ddy[4];
    LLVMValueRef texels[4];
-   LLVMTypeRef i32t = LLVMInt32Type();
-   LLVMValueRef i32t_zero = LLVMConstInt(i32t, 0, 0);
+   LLVMValueRef i32t_zero = lp_build_const_int32(bld->gallivm, 0);
    unsigned chan;
 
    /*
@@ -1030,7 +1030,7 @@ lp_build_sample_general(struct lp_build_sample_context *bld,
     */
 
    for (chan = 0; chan < 4; ++chan) {
-     texels[chan] = lp_build_alloca(builder, bld->texel_bld.vec_type, "");
+     texels[chan] = lp_build_alloca(bld->gallivm, bld->texel_bld.vec_type, "");
      lp_build_name(texels[chan], "sampler%u_texel_%c_var", unit, "xyzw"[chan]);
    }
 
@@ -1053,7 +1053,7 @@ lp_build_sample_general(struct lp_build_sample_context *bld,
       minify = LLVMBuildICmp(builder, LLVMIntSGE,
                              lod_ipart, int_bld->zero, "");
 
-      lp_build_if(&if_ctx, builder, minify);
+      lp_build_if(&if_ctx, bld->gallivm, minify);
       {
          /* Use the minification filter */
          lp_build_sample_mipmap(bld, unit,
@@ -1092,6 +1092,7 @@ lp_build_sample_compare(struct lp_build_sample_context *bld,
                         LLVMValueRef texel[4])
 {
    struct lp_build_context *texel_bld = &bld->texel_bld;
+   LLVMBuilderRef builder = bld->gallivm->builder;
    LLVMValueRef res;
    const unsigned chan = 0;
 
@@ -1100,11 +1101,10 @@ lp_build_sample_compare(struct lp_build_sample_context *bld,
 
    /* debug code */
    if (0) {
-      LLVMValueRef indx = lp_build_const_int32(0);
-      LLVMValueRef coord = LLVMBuildExtractElement(bld->builder, p, indx, "");
-      LLVMValueRef tex = LLVMBuildExtractElement(bld->builder,
-                                                 texel[chan], indx, "");
-      lp_build_printf(bld->builder, "shadow compare coord %f to texture %f\n",
+      LLVMValueRef indx = lp_build_const_int32(bld->gallivm, 0);
+      LLVMValueRef coord = LLVMBuildExtractElement(builder, p, indx, "");
+      LLVMValueRef tex = LLVMBuildExtractElement(builder, texel[chan], indx, "");
+      lp_build_printf(bld->gallivm, "shadow compare coord %f to texture %f\n",
                       coord, tex);
    }
 
@@ -1126,10 +1126,10 @@ lp_build_sample_compare(struct lp_build_sample_context *bld,
  * For debugging.
  */
 void
-lp_build_sample_nop(struct lp_type type,
+lp_build_sample_nop(struct gallivm_state *gallivm, struct lp_type type,
                     LLVMValueRef texel_out[4])
 {
-   LLVMValueRef one = lp_build_one(type);
+   LLVMValueRef one = lp_build_one(gallivm, type);
    unsigned chan;
 
    for (chan = 0; chan < 4; chan++) {
@@ -1147,7 +1147,7 @@ lp_build_sample_nop(struct lp_type type,
  * \param ddy  partial derivatives of (s,t,r,q) with respect to y
  */
 void
-lp_build_sample_soa(LLVMBuilderRef builder,
+lp_build_sample_soa(struct gallivm_state *gallivm,
                     const struct lp_sampler_static_state *static_state,
                     struct lp_sampler_dynamic_state *dynamic_state,
                     struct lp_type type,
@@ -1162,8 +1162,8 @@ lp_build_sample_soa(LLVMBuilderRef builder,
 {
    unsigned dims = texture_dims(static_state->target);
    struct lp_build_sample_context bld;
-   LLVMTypeRef i32t = LLVMInt32Type();
-
+   LLVMTypeRef i32t = LLVMInt32TypeInContext(gallivm->context);
+   LLVMBuilderRef builder = gallivm->builder;
    LLVMValueRef s;
    LLVMValueRef t;
    LLVMValueRef r;
@@ -1178,7 +1178,7 @@ lp_build_sample_soa(LLVMBuilderRef builder,
 
    /* Setup our build context */
    memset(&bld, 0, sizeof bld);
-   bld.builder = builder;
+   bld.gallivm = gallivm;
    bld.static_state = static_state;
    bld.dynamic_state = dynamic_state;
    bld.format_desc = util_format_description(static_state->format);
@@ -1195,22 +1195,22 @@ lp_build_sample_soa(LLVMBuilderRef builder,
 
    float_vec_type = lp_type_float_vec(32);
 
-   lp_build_context_init(&bld.float_bld, builder, bld.float_type);
-   lp_build_context_init(&bld.float_vec_bld, builder, float_vec_type);
-   lp_build_context_init(&bld.int_bld, builder, bld.int_type);
-   lp_build_context_init(&bld.coord_bld, builder, bld.coord_type);
-   lp_build_context_init(&bld.int_coord_bld, builder, bld.int_coord_type);
-   lp_build_context_init(&bld.int_size_bld, builder, bld.int_size_type);
-   lp_build_context_init(&bld.float_size_bld, builder, bld.float_size_type);
-   lp_build_context_init(&bld.texel_bld, builder, bld.texel_type);
+   lp_build_context_init(&bld.float_bld, gallivm, bld.float_type);
+   lp_build_context_init(&bld.float_vec_bld, gallivm, float_vec_type);
+   lp_build_context_init(&bld.int_bld, gallivm, bld.int_type);
+   lp_build_context_init(&bld.coord_bld, gallivm, bld.coord_type);
+   lp_build_context_init(&bld.int_coord_bld, gallivm, bld.int_coord_type);
+   lp_build_context_init(&bld.int_size_bld, gallivm, bld.int_size_type);
+   lp_build_context_init(&bld.float_size_bld, gallivm, bld.float_size_type);
+   lp_build_context_init(&bld.texel_bld, gallivm, bld.texel_type);
 
    /* Get the dynamic state */
-   bld.width = dynamic_state->width(dynamic_state, builder, unit);
-   bld.height = dynamic_state->height(dynamic_state, builder, unit);
-   bld.depth = dynamic_state->depth(dynamic_state, builder, unit);
-   bld.row_stride_array = dynamic_state->row_stride(dynamic_state, builder, unit);
-   bld.img_stride_array = dynamic_state->img_stride(dynamic_state, builder, unit);
-   bld.data_array = dynamic_state->data_ptr(dynamic_state, builder, unit);
+   bld.width = dynamic_state->width(dynamic_state, gallivm, unit);
+   bld.height = dynamic_state->height(dynamic_state, gallivm, unit);
+   bld.depth = dynamic_state->depth(dynamic_state, gallivm, unit);
+   bld.row_stride_array = dynamic_state->row_stride(dynamic_state, gallivm, unit);
+   bld.img_stride_array = dynamic_state->img_stride(dynamic_state, gallivm, unit);
+   bld.data_array = dynamic_state->data_ptr(dynamic_state, gallivm, unit);
    /* Note that data_array is an array[level] of pointers to texture images */
 
    s = coords[0];
@@ -1236,7 +1236,7 @@ lp_build_sample_soa(LLVMBuilderRef builder,
 
    if (0) {
       /* For debug: no-op texture sampling */
-      lp_build_sample_nop(bld.texel_type, texel_out);
+      lp_build_sample_nop(gallivm, bld.texel_type, texel_out);
    }
    else if (util_format_fits_8unorm(bld.format_desc) &&
             lp_is_simple_wrap_mode(static_state->wrap_s) &&
@@ -1266,4 +1266,6 @@ lp_build_sample_soa(LLVMBuilderRef builder,
    }
 
    lp_build_sample_compare(&bld, r, texel_out);
+
+   apply_sampler_swizzle(&bld, texel_out);
 }
diff --git a/src/gallium/auxiliary/gallivm/lp_bld_struct.c b/src/gallium/auxiliary/gallivm/lp_bld_struct.c
index 4693c2de6f..0dc2f24d10 100644
--- a/src/gallium/auxiliary/gallivm/lp_bld_struct.c
+++ b/src/gallium/auxiliary/gallivm/lp_bld_struct.c
@@ -37,12 +37,13 @@
 #include "util/u_debug.h"
 #include "util/u_memory.h"
 
+#include "lp_bld_const.h"
 #include "lp_bld_debug.h"
 #include "lp_bld_struct.h"
 
 
 LLVMValueRef
-lp_build_struct_get_ptr(LLVMBuilderRef builder,
+lp_build_struct_get_ptr(struct gallivm_state *gallivm,
                         LLVMValueRef ptr,
                         unsigned member,
                         const char *name)
@@ -51,16 +52,16 @@ lp_build_struct_get_ptr(LLVMBuilderRef builder,
    LLVMValueRef member_ptr;
    assert(LLVMGetTypeKind(LLVMTypeOf(ptr)) == LLVMPointerTypeKind);
    assert(LLVMGetTypeKind(LLVMGetElementType(LLVMTypeOf(ptr))) == LLVMStructTypeKind);
-   indices[0] = LLVMConstInt(LLVMInt32Type(), 0, 0);
-   indices[1] = LLVMConstInt(LLVMInt32Type(), member, 0);
-   member_ptr = LLVMBuildGEP(builder, ptr, indices, Elements(indices), "");
+   indices[0] = lp_build_const_int32(gallivm, 0);
+   indices[1] = lp_build_const_int32(gallivm, member);
+   member_ptr = LLVMBuildGEP(gallivm->builder, ptr, indices, Elements(indices), "");
    lp_build_name(member_ptr, "%s.%s_ptr", LLVMGetValueName(ptr), name);
    return member_ptr;
 }
 
 
 LLVMValueRef
-lp_build_struct_get(LLVMBuilderRef builder,
+lp_build_struct_get(struct gallivm_state *gallivm,
                     LLVMValueRef ptr,
                     unsigned member,
                     const char *name)
@@ -69,15 +70,15 @@ lp_build_struct_get(LLVMBuilderRef builder,
    LLVMValueRef res;
    assert(LLVMGetTypeKind(LLVMTypeOf(ptr)) == LLVMPointerTypeKind);
    assert(LLVMGetTypeKind(LLVMGetElementType(LLVMTypeOf(ptr))) == LLVMStructTypeKind);
-   member_ptr = lp_build_struct_get_ptr(builder, ptr, member, name);
-   res = LLVMBuildLoad(builder, member_ptr, "");
+   member_ptr = lp_build_struct_get_ptr(gallivm, ptr, member, name);
+   res = LLVMBuildLoad(gallivm->builder, member_ptr, "");
    lp_build_name(res, "%s.%s", LLVMGetValueName(ptr), name);
    return res;
 }
 
 
 LLVMValueRef
-lp_build_array_get_ptr(LLVMBuilderRef builder,
+lp_build_array_get_ptr(struct gallivm_state *gallivm,
                        LLVMValueRef ptr,
                        LLVMValueRef index)
 {
@@ -85,9 +86,9 @@ lp_build_array_get_ptr(LLVMBuilderRef builder,
    LLVMValueRef element_ptr;
    assert(LLVMGetTypeKind(LLVMTypeOf(ptr)) == LLVMPointerTypeKind);
    assert(LLVMGetTypeKind(LLVMGetElementType(LLVMTypeOf(ptr))) == LLVMArrayTypeKind);
-   indices[0] = LLVMConstInt(LLVMInt32Type(), 0, 0);
+   indices[0] = lp_build_const_int32(gallivm, 0);
    indices[1] = index;
-   element_ptr = LLVMBuildGEP(builder, ptr, indices, Elements(indices), "");
+   element_ptr = LLVMBuildGEP(gallivm->builder, ptr, indices, Elements(indices), "");
 #ifdef DEBUG
    lp_build_name(element_ptr, "&%s[%s]",
                  LLVMGetValueName(ptr), LLVMGetValueName(index));
@@ -97,7 +98,7 @@ lp_build_array_get_ptr(LLVMBuilderRef builder,
 
 
 LLVMValueRef
-lp_build_array_get(LLVMBuilderRef builder,
+lp_build_array_get(struct gallivm_state *gallivm,
                    LLVMValueRef ptr,
                    LLVMValueRef index)
 {
@@ -105,8 +106,8 @@ lp_build_array_get(LLVMBuilderRef builder,
    LLVMValueRef res;
    assert(LLVMGetTypeKind(LLVMTypeOf(ptr)) == LLVMPointerTypeKind);
    assert(LLVMGetTypeKind(LLVMGetElementType(LLVMTypeOf(ptr))) == LLVMArrayTypeKind);
-   element_ptr = lp_build_array_get_ptr(builder, ptr, index);
-   res = LLVMBuildLoad(builder, element_ptr, "");
+   element_ptr = lp_build_array_get_ptr(gallivm, ptr, index);
+   res = LLVMBuildLoad(gallivm->builder, element_ptr, "");
 #ifdef DEBUG
    lp_build_name(res, "%s[%s]", LLVMGetValueName(ptr), LLVMGetValueName(index));
 #endif
@@ -115,7 +116,7 @@ lp_build_array_get(LLVMBuilderRef builder,
 
 
 void
-lp_build_array_set(LLVMBuilderRef builder,
+lp_build_array_set(struct gallivm_state *gallivm,
                    LLVMValueRef ptr,
                    LLVMValueRef index,
                    LLVMValueRef value)
@@ -123,8 +124,8 @@ lp_build_array_set(LLVMBuilderRef builder,
    LLVMValueRef element_ptr;
    assert(LLVMGetTypeKind(LLVMTypeOf(ptr)) == LLVMPointerTypeKind);
    assert(LLVMGetTypeKind(LLVMGetElementType(LLVMTypeOf(ptr))) == LLVMArrayTypeKind);
-   element_ptr = lp_build_array_get_ptr(builder, ptr, index);
-   LLVMBuildStore(builder, value, element_ptr);
+   element_ptr = lp_build_array_get_ptr(gallivm, ptr, index);
+   LLVMBuildStore(gallivm->builder, value, element_ptr);
 }
 
 
diff --git a/src/gallium/auxiliary/gallivm/lp_bld_struct.h b/src/gallium/auxiliary/gallivm/lp_bld_struct.h
index eb87a8eee9..11605c685f 100644
--- a/src/gallium/auxiliary/gallivm/lp_bld_struct.h
+++ b/src/gallium/auxiliary/gallivm/lp_bld_struct.h
@@ -38,7 +38,7 @@
 
 
 #include "gallivm/lp_bld.h"
-#include <llvm-c/Target.h>
+#include "gallivm/lp_bld_init.h"
 
 #include "util/u_debug.h"
 #include "util/u_memory.h"
@@ -57,7 +57,7 @@
  * Get value pointer to a structure member.
  */
 LLVMValueRef
-lp_build_struct_get_ptr(LLVMBuilderRef builder,
+lp_build_struct_get_ptr(struct gallivm_state *gallivm,
                         LLVMValueRef ptr,
                         unsigned member,
                         const char *name);
@@ -66,7 +66,7 @@ lp_build_struct_get_ptr(LLVMBuilderRef builder,
  * Get the value of a structure member.
  */
 LLVMValueRef
-lp_build_struct_get(LLVMBuilderRef builder,
+lp_build_struct_get(struct gallivm_state *gallivm,
                     LLVMValueRef ptr,
                     unsigned member,
                     const char *name);
@@ -75,7 +75,7 @@ lp_build_struct_get(LLVMBuilderRef builder,
  * Get value pointer to an array element.
  */
 LLVMValueRef
-lp_build_array_get_ptr(LLVMBuilderRef builder,
+lp_build_array_get_ptr(struct gallivm_state *gallivm,
                        LLVMValueRef ptr,
                        LLVMValueRef index);
 
@@ -83,7 +83,7 @@ lp_build_array_get_ptr(LLVMBuilderRef builder,
  * Get the value of an array element.
  */
 LLVMValueRef
-lp_build_array_get(LLVMBuilderRef builder,
+lp_build_array_get(struct gallivm_state *gallivm,
                    LLVMValueRef ptr,
                    LLVMValueRef index);
 
@@ -91,7 +91,7 @@ lp_build_array_get(LLVMBuilderRef builder,
  * Set the value of an array element.
  */
 void
-lp_build_array_set(LLVMBuilderRef builder,
+lp_build_array_set(struct gallivm_state *gallivm,
                    LLVMValueRef ptr,
                    LLVMValueRef index,
                    LLVMValueRef value);
diff --git a/src/gallium/auxiliary/gallivm/lp_bld_swizzle.c b/src/gallium/auxiliary/gallivm/lp_bld_swizzle.c
index 4685a90e41..71693603c1 100644
--- a/src/gallium/auxiliary/gallivm/lp_bld_swizzle.c
+++ b/src/gallium/auxiliary/gallivm/lp_bld_swizzle.c
@@ -37,12 +37,13 @@
 
 #include "lp_bld_type.h"
 #include "lp_bld_const.h"
+#include "lp_bld_init.h"
 #include "lp_bld_logic.h"
 #include "lp_bld_swizzle.h"
 
 
 LLVMValueRef
-lp_build_broadcast(LLVMBuilderRef builder,
+lp_build_broadcast(struct gallivm_state *gallivm,
                    LLVMTypeRef vec_type,
                    LLVMValueRef scalar)
 {
@@ -52,8 +53,8 @@ lp_build_broadcast(LLVMBuilderRef builder,
 
    res = LLVMGetUndef(vec_type);
    for(i = 0; i < n; ++i) {
-      LLVMValueRef index = LLVMConstInt(LLVMInt32Type(), i, 0);
-      res = LLVMBuildInsertElement(builder, res, scalar, index, "");
+      LLVMValueRef index = lp_build_const_int32(gallivm, i);
+      res = LLVMBuildInsertElement(gallivm->builder, res, scalar, index, "");
    }
 
    return res;
@@ -67,6 +68,7 @@ LLVMValueRef
 lp_build_broadcast_scalar(struct lp_build_context *bld,
                           LLVMValueRef scalar)
 {
+   LLVMBuilderRef builder = bld->gallivm->builder;
    const struct lp_type type = bld->type;
 
    assert(lp_check_elem_type(type, LLVMTypeOf(scalar)));
@@ -82,17 +84,17 @@ lp_build_broadcast_scalar(struct lp_build_context *bld,
       struct lp_type i32_vec_type = lp_type_int_vec(32);
       i32_vec_type.length = type.length;
 
-      res = LLVMBuildInsertElement(bld->builder, bld->undef, scalar,
-                                   LLVMConstInt(LLVMInt32Type(), 0, 0), "");
-      res = LLVMBuildShuffleVector(bld->builder, res, bld->undef,
-                                   lp_build_const_int_vec(i32_vec_type, 0), "");
+      res = LLVMBuildInsertElement(builder, bld->undef, scalar,
+                                   lp_build_const_int32(bld->gallivm, 0), "");
+      res = LLVMBuildShuffleVector(builder, res, bld->undef,
+                                   lp_build_const_int_vec(bld->gallivm, i32_vec_type, 0), "");
 #else
       /* XXX: The above path provokes a bug in LLVM 2.6 */
       unsigned i;
       res = bld->undef;
       for(i = 0; i < type.length; ++i) {
-         LLVMValueRef index = LLVMConstInt(LLVMInt32Type(), i, 0);
-         res = LLVMBuildInsertElement(bld->builder, res, scalar, index, "");
+         LLVMValueRef index = lp_build_const_int32(bld->gallivm, i);
+         res = LLVMBuildInsertElement(builder, res, scalar, index, "");
       }
 #endif
       return res;
@@ -104,13 +106,13 @@ lp_build_broadcast_scalar(struct lp_build_context *bld,
  * Combined extract and broadcast (or a mere shuffle when the two types match)
  */
 LLVMValueRef
-lp_build_extract_broadcast(LLVMBuilderRef builder,
+lp_build_extract_broadcast(struct gallivm_state *gallivm,
                            struct lp_type src_type,
                            struct lp_type dst_type,
                            LLVMValueRef vector,
                            LLVMValueRef index)
 {
-   LLVMTypeRef i32t = LLVMInt32Type();
+   LLVMTypeRef i32t = LLVMInt32TypeInContext(gallivm->context);
    LLVMValueRef res;
 
    assert(src_type.floating == dst_type.floating);
@@ -132,8 +134,8 @@ lp_build_extract_broadcast(LLVMBuilderRef builder,
           * Broadcast scalar -> vector.
           */
 
-         res = lp_build_broadcast(builder,
-                                  lp_build_vec_type(dst_type),
+         res = lp_build_broadcast(gallivm,
+                                  lp_build_vec_type(gallivm, dst_type),
                                   vector);
       }
    }
@@ -144,16 +146,16 @@ lp_build_extract_broadcast(LLVMBuilderRef builder,
           */
 
          LLVMValueRef shuffle;
-         shuffle = lp_build_broadcast(builder,
+         shuffle = lp_build_broadcast(gallivm,
                                       LLVMVectorType(i32t, dst_type.length),
                                       index);
-         res = LLVMBuildShuffleVector(builder, vector,
-                                      LLVMGetUndef(lp_build_vec_type(dst_type)),
+         res = LLVMBuildShuffleVector(gallivm->builder, vector,
+                                      LLVMGetUndef(lp_build_vec_type(gallivm, dst_type)),
                                       shuffle, "");
       }
       else {
          LLVMValueRef scalar;
-         scalar = LLVMBuildExtractElement(builder, vector, index, "");
+         scalar = LLVMBuildExtractElement(gallivm->builder, vector, index, "");
          if (dst_type.length == 1) {
             /*
              * Trivial extract scalar from vector.
@@ -166,8 +168,8 @@ lp_build_extract_broadcast(LLVMBuilderRef builder,
              * General case of different sized vectors.
              */
 
-            res = lp_build_broadcast(builder,
-                                     lp_build_vec_type(dst_type),
+            res = lp_build_broadcast(gallivm,
+                                     lp_build_vec_type(gallivm, dst_type),
                                      vector);
          }
       }
@@ -185,6 +187,7 @@ lp_build_swizzle_scalar_aos(struct lp_build_context *bld,
                             LLVMValueRef a,
                             unsigned channel)
 {
+   LLVMBuilderRef builder = bld->gallivm->builder;
    const struct lp_type type = bld->type;
    const unsigned n = type.length;
    unsigned i, j;
@@ -199,14 +202,14 @@ lp_build_swizzle_scalar_aos(struct lp_build_context *bld,
       /*
        * Shuffle.
        */
-      LLVMTypeRef elem_type = LLVMInt32Type();
+      LLVMTypeRef elem_type = LLVMInt32TypeInContext(bld->gallivm->context);
       LLVMValueRef shuffles[LP_MAX_VECTOR_LENGTH];
 
       for(j = 0; j < n; j += 4)
          for(i = 0; i < 4; ++i)
             shuffles[j + i] = LLVMConstInt(elem_type, j + channel, 0);
 
-      return LLVMBuildShuffleVector(bld->builder, a, bld->undef, LLVMConstVector(shuffles, n), "");
+      return LLVMBuildShuffleVector(builder, a, bld->undef, LLVMConstVector(shuffles, n), "");
    }
    else {
       /*
@@ -226,8 +229,9 @@ lp_build_swizzle_scalar_aos(struct lp_build_context *bld,
       };
       unsigned i;
 
-      a = LLVMBuildAnd(bld->builder, a,
-                       lp_build_const_mask_aos(type, 1 << channel), "");
+      a = LLVMBuildAnd(builder, a,
+                       lp_build_const_mask_aos(bld->gallivm,
+                                               type, 1 << channel), "");
 
       /*
        * Build a type where each element is an integer that cover the four
@@ -239,7 +243,7 @@ lp_build_swizzle_scalar_aos(struct lp_build_context *bld,
       type4.width *= 4;
       type4.length /= 4;
 
-      a = LLVMBuildBitCast(bld->builder, a, lp_build_vec_type(type4), "");
+      a = LLVMBuildBitCast(builder, a, lp_build_vec_type(bld->gallivm, type4), "");
 
       for(i = 0; i < 2; ++i) {
          LLVMValueRef tmp = NULL;
@@ -250,16 +254,16 @@ lp_build_swizzle_scalar_aos(struct lp_build_context *bld,
 #endif
 
          if(shift > 0)
-            tmp = LLVMBuildLShr(bld->builder, a, lp_build_const_int_vec(type4, shift*type.width), "");
+            tmp = LLVMBuildLShr(builder, a, lp_build_const_int_vec(bld->gallivm, type4, shift*type.width), "");
          if(shift < 0)
-            tmp = LLVMBuildShl(bld->builder, a, lp_build_const_int_vec(type4, -shift*type.width), "");
+            tmp = LLVMBuildShl(builder, a, lp_build_const_int_vec(bld->gallivm, type4, -shift*type.width), "");
 
          assert(tmp);
          if(tmp)
-            a = LLVMBuildOr(bld->builder, a, tmp, "");
+            a = LLVMBuildOr(builder, a, tmp, "");
       }
 
-      return LLVMBuildBitCast(bld->builder, a, lp_build_vec_type(type), "");
+      return LLVMBuildBitCast(builder, a, lp_build_vec_type(bld->gallivm, type), "");
    }
 }
 
@@ -269,6 +273,7 @@ lp_build_swizzle_aos(struct lp_build_context *bld,
                      LLVMValueRef a,
                      const unsigned char swizzles[4])
 {
+   LLVMBuilderRef builder = bld->gallivm->builder;
    const struct lp_type type = bld->type;
    const unsigned n = type.length;
    unsigned i, j;
@@ -303,8 +308,8 @@ lp_build_swizzle_aos(struct lp_build_context *bld,
       /*
        * Shuffle.
        */
-      LLVMValueRef undef = LLVMGetUndef(lp_build_elem_type(type));
-      LLVMTypeRef i32t = LLVMInt32Type();
+      LLVMValueRef undef = LLVMGetUndef(lp_build_elem_type(bld->gallivm, type));
+      LLVMTypeRef i32t = LLVMInt32TypeInContext(bld->gallivm->context);
       LLVMValueRef shuffles[LP_MAX_VECTOR_LENGTH];
       LLVMValueRef aux[LP_MAX_VECTOR_LENGTH];
 
@@ -326,13 +331,13 @@ lp_build_swizzle_aos(struct lp_build_context *bld,
             case PIPE_SWIZZLE_ZERO:
                shuffle = type.length + 0;
                if (!aux[0]) {
-                  aux[0] = lp_build_const_elem(type, 0.0);
+                  aux[0] = lp_build_const_elem(bld->gallivm, type, 0.0);
                }
                break;
             case PIPE_SWIZZLE_ONE:
                shuffle = type.length + 1;
                if (!aux[1]) {
-                  aux[1] = lp_build_const_elem(type, 1.0);
+                  aux[1] = lp_build_const_elem(bld->gallivm, type, 1.0);
                }
                break;
             }
@@ -346,7 +351,7 @@ lp_build_swizzle_aos(struct lp_build_context *bld,
          }
       }
 
-      return LLVMBuildShuffleVector(bld->builder, a,
+      return LLVMBuildShuffleVector(builder, a,
                                     LLVMConstVector(aux, n),
                                     LLVMConstVector(shuffles, n), "");
    } else {
@@ -387,8 +392,8 @@ lp_build_swizzle_aos(struct lp_build_context *bld,
       type4.width *= 4;
       type4.length /= 4;
 
-      a = LLVMBuildBitCast(bld->builder, a, lp_build_vec_type(type4), "");
-      res = LLVMBuildBitCast(bld->builder, res, lp_build_vec_type(type4), "");
+      a = LLVMBuildBitCast(builder, a, lp_build_vec_type(bld->gallivm, type4), "");
+      res = LLVMBuildBitCast(builder, res, lp_build_vec_type(bld->gallivm, type4), "");
 
       /*
        * Mask and shift the channels, trying to group as many channels in the
@@ -414,23 +419,24 @@ lp_build_swizzle_aos(struct lp_build_context *bld,
             if (0)
                debug_printf("shift = %i, mask = 0x%08llx\n", shift, mask);
 
-            masked = LLVMBuildAnd(bld->builder, a,
-                                  lp_build_const_int_vec(type4, mask), "");
+            masked = LLVMBuildAnd(builder, a,
+                                  lp_build_const_int_vec(bld->gallivm, type4, mask), "");
             if (shift > 0) {
-               shifted = LLVMBuildShl(bld->builder, masked,
-                                      lp_build_const_int_vec(type4, shift*type.width), "");
+               shifted = LLVMBuildShl(builder, masked,
+                                      lp_build_const_int_vec(bld->gallivm, type4, shift*type.width), "");
             } else if (shift < 0) {
-               shifted = LLVMBuildLShr(bld->builder, masked,
-                                       lp_build_const_int_vec(type4, -shift*type.width), "");
+               shifted = LLVMBuildLShr(builder, masked,
+                                       lp_build_const_int_vec(bld->gallivm, type4, -shift*type.width), "");
             } else {
                shifted = masked;
             }
 
-            res = LLVMBuildOr(bld->builder, res, shifted, "");
+            res = LLVMBuildOr(builder, res, shifted, "");
          }
       }
 
-      return LLVMBuildBitCast(bld->builder, res, lp_build_vec_type(type), "");
+      return LLVMBuildBitCast(builder, res,
+                              lp_build_vec_type(bld->gallivm, type), "");
    }
 }
 
diff --git a/src/gallium/auxiliary/gallivm/lp_bld_swizzle.h b/src/gallium/auxiliary/gallivm/lp_bld_swizzle.h
index fdea8442ae..c366a65103 100644
--- a/src/gallium/auxiliary/gallivm/lp_bld_swizzle.h
+++ b/src/gallium/auxiliary/gallivm/lp_bld_swizzle.h
@@ -45,7 +45,7 @@ struct lp_build_context;
 
 
 LLVMValueRef
-lp_build_broadcast(LLVMBuilderRef builder,
+lp_build_broadcast(struct gallivm_state *gallivm,
                    LLVMTypeRef vec_type,
                    LLVMValueRef scalar);
 
@@ -56,7 +56,7 @@ lp_build_broadcast_scalar(struct lp_build_context *bld,
 
 
 LLVMValueRef
-lp_build_extract_broadcast(LLVMBuilderRef builder,
+lp_build_extract_broadcast(struct gallivm_state *gallivm,
                            struct lp_type src_type,
                            struct lp_type dst_type,
                            LLVMValueRef vector,
diff --git a/src/gallium/auxiliary/gallivm/lp_bld_tgsi.h b/src/gallium/auxiliary/gallivm/lp_bld_tgsi.h
index a4d3b750c3..40186befb9 100644
--- a/src/gallium/auxiliary/gallivm/lp_bld_tgsi.h
+++ b/src/gallium/auxiliary/gallivm/lp_bld_tgsi.h
@@ -46,6 +46,7 @@ struct tgsi_shader_info;
 struct lp_type;
 struct lp_build_context;
 struct lp_build_mask_context;
+struct gallivm_state;
 
 
 enum lp_build_tex_modifier {
@@ -141,7 +142,7 @@ struct lp_build_sampler_soa
 
    void
    (*emit_fetch_texel)( const struct lp_build_sampler_soa *sampler,
-                        LLVMBuilderRef builder,
+                        struct gallivm_state *gallivm,
                         struct lp_type type,
                         unsigned unit,
                         unsigned num_coords,
@@ -174,7 +175,7 @@ lp_build_tgsi_info(const struct tgsi_token *tokens,
 
 
 void
-lp_build_tgsi_soa(LLVMBuilderRef builder,
+lp_build_tgsi_soa(struct gallivm_state *gallivm,
                   const struct tgsi_token *tokens,
                   struct lp_type type,
                   struct lp_build_mask_context *mask,
@@ -187,7 +188,7 @@ lp_build_tgsi_soa(LLVMBuilderRef builder,
 
 
 void
-lp_build_tgsi_aos(LLVMBuilderRef builder,
+lp_build_tgsi_aos(struct gallivm_state *gallivm,
                   const struct tgsi_token *tokens,
                   struct lp_type type,
                   const unsigned char swizzles[4],
diff --git a/src/gallium/auxiliary/gallivm/lp_bld_tgsi_aos.c b/src/gallium/auxiliary/gallivm/lp_bld_tgsi_aos.c
index c3c082b2b9..a021efd69f 100644
--- a/src/gallium/auxiliary/gallivm/lp_bld_tgsi_aos.c
+++ b/src/gallium/auxiliary/gallivm/lp_bld_tgsi_aos.c
@@ -151,6 +151,7 @@ emit_fetch(
    const struct tgsi_full_instruction *inst,
    unsigned src_op)
 {
+   LLVMBuilderRef builder = bld->base.gallivm->builder;
    struct lp_type type = bld->base.type;
    const struct tgsi_full_src_register *reg = &inst->Src[src_op];
    LLVMValueRef res;
@@ -175,14 +176,12 @@ emit_fetch(
          LLVMValueRef scalar;
          LLVMValueRef swizzle;
 
-         index = LLVMConstInt(LLVMInt32Type(),
-                              reg->Register.Index*4 + chan,
-                              0);
+         index = lp_build_const_int32(bld->base.gallivm, reg->Register.Index * 4 + chan);
 
-         scalar_ptr = LLVMBuildGEP(bld->base.builder, bld->consts_ptr,
+         scalar_ptr = LLVMBuildGEP(builder, bld->consts_ptr,
                                    &index, 1, "");
 
-         scalar = LLVMBuildLoad(bld->base.builder, scalar_ptr, "");
+         scalar = LLVMBuildLoad(builder, scalar_ptr, "");
 
          lp_build_name(scalar, "const[%u].%c", reg->Register.Index, "xyzw"[chan]);
 
@@ -190,9 +189,9 @@ emit_fetch(
           * NOTE: constants array is always assumed to be RGBA
           */
 
-         swizzle = LLVMConstInt(LLVMInt32Type(), chan, 0);
+         swizzle = lp_build_const_int32(bld->base.gallivm, chan);
 
-         res = LLVMBuildInsertElement(bld->base.builder, res, scalar, swizzle, "");
+         res = LLVMBuildInsertElement(builder, res, scalar, swizzle, "");
       }
 
       /*
@@ -206,14 +205,14 @@ emit_fetch(
          unsigned i;
 
          for (chan = 0; chan < 4; ++chan) {
-            shuffles[chan] = LLVMConstInt(LLVMInt32Type(), chan, 0);
+            shuffles[chan] = lp_build_const_int32(bld->base.gallivm, chan);
          }
 
          for (i = 4; i < type.length; ++i) {
             shuffles[i] = shuffles[i % 4];
          }
 
-         res = LLVMBuildShuffleVector(bld->base.builder,
+         res = LLVMBuildShuffleVector(builder,
                                       res, bld->base.undef,
                                       LLVMConstVector(shuffles, type.length),
                                       "");
@@ -234,7 +233,7 @@ emit_fetch(
       {
          LLVMValueRef temp_ptr;
          temp_ptr = bld->temps[reg->Register.Index];
-         res = LLVMBuildLoad(bld->base.builder, temp_ptr, "");
+         res = LLVMBuildLoad(builder, temp_ptr, "");
          if (!res)
             return bld->base.undef;
       }
@@ -281,6 +280,7 @@ emit_store(
    unsigned index,
    LLVMValueRef value)
 {
+   LLVMBuilderRef builder = bld->base.gallivm->builder;
    const struct tgsi_full_dst_register *reg = &inst->Dst[index];
    LLVMValueRef mask = NULL;
    LLVMValueRef ptr;
@@ -299,7 +299,7 @@ emit_store(
       break;
 
    case TGSI_SAT_MINUS_PLUS_ONE:
-      value = lp_build_max(&bld->base, value, lp_build_const_vec(bld->base.type, -1.0));
+      value = lp_build_max(&bld->base, value, lp_build_const_vec(bld->base.gallivm, bld->base.type, -1.0));
       value = lp_build_min(&bld->base, value, bld->base.one);
       break;
 
@@ -344,20 +344,20 @@ emit_store(
 
       assert(inst->Predicate.Index < LP_MAX_TGSI_PREDS);
 
-      pred = LLVMBuildLoad(bld->base.builder,
+      pred = LLVMBuildLoad(builder,
                            bld->preds[inst->Predicate.Index], "");
 
       /*
        * Convert the value to an integer mask.
        */
-      pred = lp_build_compare(bld->base.builder,
+      pred = lp_build_compare(bld->base.gallivm,
                                bld->base.type,
                                PIPE_FUNC_NOTEQUAL,
                                pred,
                                bld->base.zero);
 
       if (inst->Predicate.Negate) {
-         pred = LLVMBuildNot(bld->base.builder, pred, "");
+         pred = LLVMBuildNot(builder, pred, "");
       }
 
       pred = swizzle_aos(bld, pred,
@@ -367,7 +367,7 @@ emit_store(
                          inst->Predicate.SwizzleW);
 
       if (mask) {
-         mask = LLVMBuildAnd(bld->base.builder, mask, pred, "");
+         mask = LLVMBuildAnd(builder, mask, pred, "");
       } else {
          mask = pred;
       }
@@ -380,10 +380,11 @@ emit_store(
    if (reg->Register.WriteMask != TGSI_WRITEMASK_XYZW) {
       LLVMValueRef writemask;
 
-      writemask = lp_build_const_mask_aos(bld->base.type, reg->Register.WriteMask);
+      writemask = lp_build_const_mask_aos(bld->base.gallivm, bld->base.type,
+                                          reg->Register.WriteMask);
 
       if (mask) {
-         mask = LLVMBuildAnd(bld->base.builder, mask, writemask, "");
+         mask = LLVMBuildAnd(builder, mask, writemask, "");
       } else {
          mask = writemask;
       }
@@ -392,12 +393,12 @@ emit_store(
    if (mask) {
       LLVMValueRef orig_value;
 
-      orig_value = LLVMBuildLoad(bld->base.builder, ptr, "");
+      orig_value = LLVMBuildLoad(builder, ptr, "");
       value = lp_build_select(&bld->base,
                               mask, value, orig_value);
    }
 
-   LLVMBuildStore(bld->base.builder, value, ptr);
+   LLVMBuildStore(builder, value, ptr);
 }
 
 
@@ -454,7 +455,8 @@ emit_declaration(
    struct lp_build_tgsi_aos_context *bld,
    const struct tgsi_full_declaration *decl)
 {
-   LLVMTypeRef vec_type = lp_build_vec_type(bld->base.type);
+   struct gallivm_state *gallivm = bld->base.gallivm;
+   LLVMTypeRef vec_type = lp_build_vec_type(bld->base.gallivm, bld->base.type);
 
    unsigned first = decl->Range.First;
    unsigned last = decl->Range.Last;
@@ -465,31 +467,26 @@ emit_declaration(
       case TGSI_FILE_TEMPORARY:
          assert(idx < LP_MAX_TGSI_TEMPS);
          if (bld->indirect_files & (1 << TGSI_FILE_TEMPORARY)) {
-            LLVMValueRef array_size = LLVMConstInt(LLVMInt32Type(),
-                                                   last + 1, 0);
-            bld->temps_array = lp_build_array_alloca(bld->base.builder,
+            LLVMValueRef array_size = lp_build_const_int32(gallivm, last + 1);
+            bld->temps_array = lp_build_array_alloca(bld->base.gallivm,
                                                      vec_type, array_size, "");
          } else {
-            bld->temps[idx] = lp_build_alloca(bld->base.builder,
-                                              vec_type, "");
+            bld->temps[idx] = lp_build_alloca(gallivm, vec_type, "");
          }
          break;
 
       case TGSI_FILE_OUTPUT:
-         bld->outputs[idx] = lp_build_alloca(bld->base.builder,
-                                             vec_type, "");
+         bld->outputs[idx] = lp_build_alloca(gallivm, vec_type, "");
          break;
 
       case TGSI_FILE_ADDRESS:
          assert(idx < LP_MAX_TGSI_ADDRS);
-         bld->addr[idx] = lp_build_alloca(bld->base.builder,
-                                          vec_type, "");
+         bld->addr[idx] = lp_build_alloca(gallivm, vec_type, "");
          break;
 
       case TGSI_FILE_PREDICATE:
          assert(idx < LP_MAX_TGSI_PREDS);
-         bld->preds[idx] = lp_build_alloca(bld->base.builder,
-                                           vec_type, "");
+         bld->preds[idx] = lp_build_alloca(gallivm, vec_type, "");
          break;
 
       default:
@@ -644,7 +641,7 @@ emit_instruction(
       src0 = emit_fetch(bld, inst, 0);
       src1 = emit_fetch(bld, inst, 1);
       src2 = emit_fetch(bld, inst, 2);
-      tmp1 = lp_build_const_vec(bld->base.type, 0.5);
+      tmp1 = lp_build_const_vec(bld->base.gallivm, bld->base.type, 0.5);
       tmp0 = lp_build_cmp(&bld->base, PIPE_FUNC_GREATER, src2, tmp1);
       dst0 = lp_build_select(&bld->base, tmp0, src0, src1);
       break;
@@ -1039,7 +1036,7 @@ emit_instruction(
 
 
 void
-lp_build_tgsi_aos(LLVMBuilderRef builder,
+lp_build_tgsi_aos(struct gallivm_state *gallivm,
                   const struct tgsi_token *tokens,
                   struct lp_type type,
                   const unsigned char swizzles[4],
@@ -1058,8 +1055,8 @@ lp_build_tgsi_aos(LLVMBuilderRef builder,
 
    /* Setup build context */
    memset(&bld, 0, sizeof bld);
-   lp_build_context_init(&bld.base, builder, type);
-   lp_build_context_init(&bld.int_bld, builder, lp_int_type(type));
+   lp_build_context_init(&bld.base, gallivm, type);
+   lp_build_context_init(&bld.int_bld, gallivm, lp_int_type(type));
 
    for (chan = 0; chan < 4; ++chan) {
       bld.swizzles[chan] = swizzles[chan];
@@ -1131,7 +1128,7 @@ lp_build_tgsi_aos(LLVMBuilderRef builder,
                imm[swizzle] = parse.FullToken.FullImmediate.u[chan].Float;
             }
             bld.immediates[num_immediates] =
-                     lp_build_const_aos(type,
+                     lp_build_const_aos(gallivm, type,
                                         imm[0], imm[1], imm[2], imm[3],
                                         NULL);
             num_immediates++;
@@ -1156,7 +1153,7 @@ lp_build_tgsi_aos(LLVMBuilderRef builder,
    }
 
    if (0) {
-      LLVMBasicBlockRef block = LLVMGetInsertBlock(builder);
+      LLVMBasicBlockRef block = LLVMGetInsertBlock(gallivm->builder);
       LLVMValueRef function = LLVMGetBasicBlockParent(block);
       debug_printf("11111111111111111111111111111 \n");
       tgsi_dump(tokens, 0);
@@ -1167,7 +1164,7 @@ lp_build_tgsi_aos(LLVMBuilderRef builder,
 
    if (0) {
       LLVMModuleRef module = LLVMGetGlobalParent(
-         LLVMGetBasicBlockParent(LLVMGetInsertBlock(bld.base.builder)));
+         LLVMGetBasicBlockParent(LLVMGetInsertBlock(gallivm->builder)));
       LLVMDumpModule(module);
    }
 
diff --git a/src/gallium/auxiliary/gallivm/lp_bld_tgsi_soa.c b/src/gallium/auxiliary/gallivm/lp_bld_tgsi_soa.c
index 7f0f058c22..1b5a8a5903 100644
--- a/src/gallium/auxiliary/gallivm/lp_bld_tgsi_soa.c
+++ b/src/gallium/auxiliary/gallivm/lp_bld_tgsi_soa.c
@@ -51,6 +51,7 @@
 #include "lp_bld_arit.h"
 #include "lp_bld_bitarit.h"
 #include "lp_bld_gather.h"
+#include "lp_bld_init.h"
 #include "lp_bld_logic.h"
 #include "lp_bld_swizzle.h"
 #include "lp_bld_flow.h"
@@ -175,22 +176,24 @@ static void lp_exec_mask_init(struct lp_exec_mask *mask, struct lp_build_context
    mask->loop_stack_size = 0;
    mask->call_stack_size = 0;
 
-   mask->int_vec_type = lp_build_int_vec_type(mask->bld->type);
+   mask->int_vec_type = lp_build_int_vec_type(bld->gallivm, mask->bld->type);
    mask->exec_mask = mask->ret_mask = mask->break_mask = mask->cont_mask = mask->cond_mask =
          LLVMConstAllOnes(mask->int_vec_type);
 }
 
 static void lp_exec_mask_update(struct lp_exec_mask *mask)
 {
+   LLVMBuilderRef builder = mask->bld->gallivm->builder;
+
    if (mask->loop_stack_size) {
       /*for loops we need to update the entire mask at runtime */
       LLVMValueRef tmp;
       assert(mask->break_mask);
-      tmp = LLVMBuildAnd(mask->bld->builder,
+      tmp = LLVMBuildAnd(builder,
                          mask->cont_mask,
                          mask->break_mask,
                          "maskcb");
-      mask->exec_mask = LLVMBuildAnd(mask->bld->builder,
+      mask->exec_mask = LLVMBuildAnd(builder,
                                      mask->cond_mask,
                                      tmp,
                                      "maskfull");
@@ -198,7 +201,7 @@ static void lp_exec_mask_update(struct lp_exec_mask *mask)
       mask->exec_mask = mask->cond_mask;
 
    if (mask->call_stack_size) {
-      mask->exec_mask = LLVMBuildAnd(mask->bld->builder,
+      mask->exec_mask = LLVMBuildAnd(builder,
                                      mask->exec_mask,
                                      mask->ret_mask,
                                      "callmask");
@@ -212,13 +215,15 @@ static void lp_exec_mask_update(struct lp_exec_mask *mask)
 static void lp_exec_mask_cond_push(struct lp_exec_mask *mask,
                                    LLVMValueRef val)
 {
+   LLVMBuilderRef builder = mask->bld->gallivm->builder;
+
    assert(mask->cond_stack_size < LP_MAX_TGSI_NESTING);
    if (mask->cond_stack_size == 0) {
       assert(mask->cond_mask == LLVMConstAllOnes(mask->int_vec_type));
    }
    mask->cond_stack[mask->cond_stack_size++] = mask->cond_mask;
    assert(LLVMTypeOf(val) == mask->int_vec_type);
-   mask->cond_mask = LLVMBuildAnd(mask->bld->builder,
+   mask->cond_mask = LLVMBuildAnd(builder,
                                   mask->cond_mask,
                                   val,
                                   "");
@@ -227,6 +232,7 @@ static void lp_exec_mask_cond_push(struct lp_exec_mask *mask,
 
 static void lp_exec_mask_cond_invert(struct lp_exec_mask *mask)
 {
+   LLVMBuilderRef builder = mask->bld->gallivm->builder;
    LLVMValueRef prev_mask;
    LLVMValueRef inv_mask;
 
@@ -236,9 +242,9 @@ static void lp_exec_mask_cond_invert(struct lp_exec_mask *mask)
       assert(prev_mask == LLVMConstAllOnes(mask->int_vec_type));
    }
 
-   inv_mask = LLVMBuildNot(mask->bld->builder, mask->cond_mask, "");
+   inv_mask = LLVMBuildNot(builder, mask->cond_mask, "");
 
-   mask->cond_mask = LLVMBuildAnd(mask->bld->builder,
+   mask->cond_mask = LLVMBuildAnd(builder,
                                   inv_mask,
                                   prev_mask, "");
    lp_exec_mask_update(mask);
@@ -253,6 +259,8 @@ static void lp_exec_mask_cond_pop(struct lp_exec_mask *mask)
 
 static void lp_exec_bgnloop(struct lp_exec_mask *mask)
 {
+   LLVMBuilderRef builder = mask->bld->gallivm->builder;
+
    if (mask->loop_stack_size == 0) {
       assert(mask->loop_block == NULL);
       assert(mask->cont_mask == LLVMConstAllOnes(mask->int_vec_type));
@@ -268,25 +276,26 @@ static void lp_exec_bgnloop(struct lp_exec_mask *mask)
    mask->loop_stack[mask->loop_stack_size].break_var = mask->break_var;
    ++mask->loop_stack_size;
 
-   mask->break_var = lp_build_alloca(mask->bld->builder, mask->int_vec_type, "");
-   LLVMBuildStore(mask->bld->builder, mask->break_mask, mask->break_var);
+   mask->break_var = lp_build_alloca(mask->bld->gallivm, mask->int_vec_type, "");
+   LLVMBuildStore(builder, mask->break_mask, mask->break_var);
 
-   mask->loop_block = lp_build_insert_new_block(mask->bld->builder, "bgnloop");
-   LLVMBuildBr(mask->bld->builder, mask->loop_block);
-   LLVMPositionBuilderAtEnd(mask->bld->builder, mask->loop_block);
+   mask->loop_block = lp_build_insert_new_block(mask->bld->gallivm, "bgnloop");
+   LLVMBuildBr(builder, mask->loop_block);
+   LLVMPositionBuilderAtEnd(builder, mask->loop_block);
 
-   mask->break_mask = LLVMBuildLoad(mask->bld->builder, mask->break_var, "");
+   mask->break_mask = LLVMBuildLoad(builder, mask->break_var, "");
 
    lp_exec_mask_update(mask);
 }
 
 static void lp_exec_break(struct lp_exec_mask *mask)
 {
-   LLVMValueRef exec_mask = LLVMBuildNot(mask->bld->builder,
+   LLVMBuilderRef builder = mask->bld->gallivm->builder;
+   LLVMValueRef exec_mask = LLVMBuildNot(builder,
                                          mask->exec_mask,
                                          "break");
 
-   mask->break_mask = LLVMBuildAnd(mask->bld->builder,
+   mask->break_mask = LLVMBuildAnd(builder,
                                    mask->break_mask,
                                    exec_mask, "break_full");
 
@@ -295,11 +304,12 @@ static void lp_exec_break(struct lp_exec_mask *mask)
 
 static void lp_exec_continue(struct lp_exec_mask *mask)
 {
-   LLVMValueRef exec_mask = LLVMBuildNot(mask->bld->builder,
+   LLVMBuilderRef builder = mask->bld->gallivm->builder;
+   LLVMValueRef exec_mask = LLVMBuildNot(builder,
                                          mask->exec_mask,
                                          "");
 
-   mask->cont_mask = LLVMBuildAnd(mask->bld->builder,
+   mask->cont_mask = LLVMBuildAnd(builder,
                                   mask->cont_mask,
                                   exec_mask, "");
 
@@ -307,11 +317,14 @@ static void lp_exec_continue(struct lp_exec_mask *mask)
 }
 
 
-static void lp_exec_endloop(struct lp_exec_mask *mask)
+static void lp_exec_endloop(struct gallivm_state *gallivm,
+                            struct lp_exec_mask *mask)
 {
+   LLVMBuilderRef builder = mask->bld->gallivm->builder;
    LLVMBasicBlockRef endloop;
-   LLVMTypeRef reg_type = LLVMIntType(mask->bld->type.width*
-                                      mask->bld->type.length);
+   LLVMTypeRef reg_type = LLVMIntTypeInContext(gallivm->context,
+                                               mask->bld->type.width *
+                                               mask->bld->type.length);
    LLVMValueRef i1cond;
 
    assert(mask->break_mask);
@@ -327,21 +340,21 @@ static void lp_exec_endloop(struct lp_exec_mask *mask)
     * Unlike the continue mask, the break_mask must be preserved across loop
     * iterations
     */
-   LLVMBuildStore(mask->bld->builder, mask->break_mask, mask->break_var);
+   LLVMBuildStore(builder, mask->break_mask, mask->break_var);
 
    /* i1cond = (mask == 0) */
    i1cond = LLVMBuildICmp(
-      mask->bld->builder,
+      builder,
       LLVMIntNE,
-      LLVMBuildBitCast(mask->bld->builder, mask->exec_mask, reg_type, ""),
+      LLVMBuildBitCast(builder, mask->exec_mask, reg_type, ""),
       LLVMConstNull(reg_type), "");
 
-   endloop = lp_build_insert_new_block(mask->bld->builder, "endloop");
+   endloop = lp_build_insert_new_block(mask->bld->gallivm, "endloop");
 
-   LLVMBuildCondBr(mask->bld->builder,
+   LLVMBuildCondBr(builder,
                    i1cond, mask->loop_block, endloop);
 
-   LLVMPositionBuilderAtEnd(mask->bld->builder, endloop);
+   LLVMPositionBuilderAtEnd(builder, endloop);
 
    assert(mask->loop_stack_size);
    --mask->loop_stack_size;
@@ -363,10 +376,12 @@ static void lp_exec_mask_store(struct lp_exec_mask *mask,
                                LLVMValueRef val,
                                LLVMValueRef dst)
 {
+   LLVMBuilderRef builder = mask->bld->gallivm->builder;
+
    /* Mix the predicate and execution mask */
    if (mask->has_mask) {
       if (pred) {
-         pred = LLVMBuildAnd(mask->bld->builder, pred, mask->exec_mask, "");
+         pred = LLVMBuildAnd(builder, pred, mask->exec_mask, "");
       } else {
          pred = mask->exec_mask;
       }
@@ -375,14 +390,14 @@ static void lp_exec_mask_store(struct lp_exec_mask *mask,
    if (pred) {
       LLVMValueRef real_val, dst_val;
 
-      dst_val = LLVMBuildLoad(mask->bld->builder, dst, "");
+      dst_val = LLVMBuildLoad(builder, dst, "");
       real_val = lp_build_select(mask->bld,
                                  pred,
                                  val, dst_val);
 
-      LLVMBuildStore(mask->bld->builder, real_val, dst);
+      LLVMBuildStore(builder, real_val, dst);
    } else
-      LLVMBuildStore(mask->bld->builder, val, dst);
+      LLVMBuildStore(builder, val, dst);
 }
 
 static void lp_exec_mask_call(struct lp_exec_mask *mask,
@@ -398,6 +413,7 @@ static void lp_exec_mask_call(struct lp_exec_mask *mask,
 
 static void lp_exec_mask_ret(struct lp_exec_mask *mask, int *pc)
 {
+   LLVMBuilderRef builder = mask->bld->gallivm->builder;
    LLVMValueRef exec_mask;
 
    if (mask->call_stack_size == 0) {
@@ -405,11 +421,11 @@ static void lp_exec_mask_ret(struct lp_exec_mask *mask, int *pc)
       *pc = -1;
       return;
    }
-   exec_mask = LLVMBuildNot(mask->bld->builder,
+   exec_mask = LLVMBuildNot(builder,
                             mask->exec_mask,
                             "ret");
 
-   mask->ret_mask = LLVMBuildAnd(mask->bld->builder,
+   mask->ret_mask = LLVMBuildAnd(builder,
                                  mask->ret_mask,
                                  exec_mask, "ret_full");
 
@@ -441,10 +457,11 @@ get_temp_ptr(struct lp_build_tgsi_soa_context *bld,
              unsigned index,
              unsigned chan)
 {
+   LLVMBuilderRef builder = bld->base.gallivm->builder;
    assert(chan < 4);
    if (bld->indirect_files & (1 << TGSI_FILE_TEMPORARY)) {
-      LLVMValueRef lindex = lp_build_const_int32(index * 4 + chan);
-      return LLVMBuildGEP(bld->base.builder, bld->temps_array, &lindex, 1, "");
+      LLVMValueRef lindex = lp_build_const_int32(bld->base.gallivm, index * 4 + chan);
+      return LLVMBuildGEP(builder, bld->temps_array, &lindex, 1, "");
    }
    else {
       return bld->temps[index][chan];
@@ -462,10 +479,12 @@ get_output_ptr(struct lp_build_tgsi_soa_context *bld,
                unsigned index,
                unsigned chan)
 {
+   LLVMBuilderRef builder = bld->base.gallivm->builder;
    assert(chan < 4);
    if (bld->indirect_files & (1 << TGSI_FILE_OUTPUT)) {
-      LLVMValueRef lindex = lp_build_const_int32(index * 4 + chan);
-      return LLVMBuildGEP(bld->base.builder, bld->outputs_array, &lindex, 1, "");
+      LLVMValueRef lindex = lp_build_const_int32(bld->base.gallivm,
+                                                 index * 4 + chan);
+      return LLVMBuildGEP(builder, bld->outputs_array, &lindex, 1, "");
    }
    else {
       return bld->outputs[index][chan];
@@ -482,6 +501,7 @@ build_gather(struct lp_build_tgsi_soa_context *bld,
              LLVMValueRef base_ptr,
              LLVMValueRef indexes)
 {
+   LLVMBuilderRef builder = bld->base.gallivm->builder;
    LLVMValueRef res = bld->base.undef;
    unsigned i;
 
@@ -489,14 +509,14 @@ build_gather(struct lp_build_tgsi_soa_context *bld,
     * Loop over elements of index_vec, load scalar value, insert it into 'res'.
     */
    for (i = 0; i < bld->base.type.length; i++) {
-      LLVMValueRef ii = LLVMConstInt(LLVMInt32Type(), i, 0);
-      LLVMValueRef index = LLVMBuildExtractElement(bld->base.builder,
+      LLVMValueRef ii = lp_build_const_int32(bld->base.gallivm, i);
+      LLVMValueRef index = LLVMBuildExtractElement(builder,
                                                    indexes, ii, "");
-      LLVMValueRef scalar_ptr = LLVMBuildGEP(bld->base.builder, base_ptr,
+      LLVMValueRef scalar_ptr = LLVMBuildGEP(builder, base_ptr,
                                              &index, 1, "gather_ptr");
-      LLVMValueRef scalar = LLVMBuildLoad(bld->base.builder, scalar_ptr, "");
+      LLVMValueRef scalar = LLVMBuildLoad(builder, scalar_ptr, "");
 
-      res = LLVMBuildInsertElement(bld->base.builder, res, scalar, ii, "");
+      res = LLVMBuildInsertElement(builder, res, scalar, ii, "");
    }
 
    return res;
@@ -514,13 +534,14 @@ emit_mask_scatter(struct lp_build_tgsi_soa_context *bld,
                   struct lp_exec_mask *mask,
                   LLVMValueRef pred)
 {
-   LLVMBuilderRef builder = bld->base.builder;
+   struct gallivm_state *gallivm = bld->base.gallivm;
+   LLVMBuilderRef builder = gallivm->builder;
    unsigned i;
 
    /* Mix the predicate and execution mask */
    if (mask->has_mask) {
       if (pred) {
-         pred = LLVMBuildAnd(mask->bld->builder, pred, mask->exec_mask, "");
+         pred = LLVMBuildAnd(builder, pred, mask->exec_mask, "");
       }
       else {
          pred = mask->exec_mask;
@@ -531,7 +552,7 @@ emit_mask_scatter(struct lp_build_tgsi_soa_context *bld,
     * Loop over elements of index_vec, store scalar value.
     */
    for (i = 0; i < bld->base.type.length; i++) {
-      LLVMValueRef ii = LLVMConstInt(LLVMInt32Type(), i, 0);
+      LLVMValueRef ii = lp_build_const_int32(gallivm, i);
       LLVMValueRef index = LLVMBuildExtractElement(builder, indexes, ii, "");
       LLVMValueRef scalar_ptr = LLVMBuildGEP(builder, base_ptr, &index, 1, "scatter_ptr");
       LLVMValueRef val = LLVMBuildExtractElement(builder, values, ii, "scatter_val");
@@ -539,7 +560,7 @@ emit_mask_scatter(struct lp_build_tgsi_soa_context *bld,
          LLVMBuildExtractElement(builder, pred, ii, "scatter_pred") : NULL;
 
       if (0)
-         lp_build_printf(builder, "scatter %d: val %f at %d %p\n",
+         lp_build_printf(gallivm, "scatter %d: val %f at %d %p\n",
                          ii, val, index, scalar_ptr);
 
       if (scalar_pred) {
@@ -566,6 +587,7 @@ get_indirect_index(struct lp_build_tgsi_soa_context *bld,
                    unsigned reg_file, unsigned reg_index,
                    const struct tgsi_src_register *indirect_reg)
 {
+   LLVMBuilderRef builder = bld->base.gallivm->builder;
    struct lp_build_context *uint_bld = &bld->uint_bld;
    /* always use X component of address register */
    unsigned swizzle = indirect_reg->SwizzleX;
@@ -576,21 +598,22 @@ get_indirect_index(struct lp_build_tgsi_soa_context *bld,
 
    assert(bld->indirect_files & (1 << reg_file));
 
-   base = lp_build_const_int_vec(uint_bld->type, reg_index);
+   base = lp_build_const_int_vec(bld->base.gallivm, uint_bld->type, reg_index);
 
    assert(swizzle < 4);
-   rel = LLVMBuildLoad(bld->base.builder,
+   rel = LLVMBuildLoad(builder,
                         bld->addr[indirect_reg->Index][swizzle],
                         "load addr reg");
 
    /* for indexing we want integers */
-   rel = LLVMBuildFPToSI(bld->base.builder,
+   rel = LLVMBuildFPToSI(builder,
                          rel,
                          uint_bld->vec_type, "");
 
    index = lp_build_add(uint_bld, base, rel);
 
-   max_index = lp_build_const_int_vec(uint_bld->type,
+   max_index = lp_build_const_int_vec(bld->base.gallivm,
+                                      uint_bld->type,
                                       bld->info->file_max[reg_file]);
 
    assert(!uint_bld->type.sign);
@@ -610,6 +633,8 @@ emit_fetch(
    unsigned src_op,
    const unsigned chan_index )
 {
+   struct gallivm_state *gallivm = bld->base.gallivm;
+   LLVMBuilderRef builder = gallivm->builder;
    struct lp_build_context *uint_bld = &bld->uint_bld;
    const struct tgsi_full_src_register *reg = &inst->Src[src_op];
    const unsigned swizzle =
@@ -635,7 +660,7 @@ emit_fetch(
    case TGSI_FILE_CONSTANT:
       if (reg->Register.Indirect) {
          LLVMValueRef swizzle_vec =
-            lp_build_const_int_vec(uint_bld->type, swizzle);
+            lp_build_const_int_vec(bld->base.gallivm, uint_bld->type, swizzle);
          LLVMValueRef index_vec;  /* index into the const buffer */
 
          /* index_vec = indirect_index * 4 + swizzle */
@@ -649,11 +674,11 @@ emit_fetch(
          LLVMValueRef index;  /* index into the const buffer */
          LLVMValueRef scalar, scalar_ptr;
 
-         index = lp_build_const_int32(reg->Register.Index*4 + swizzle);
+         index = lp_build_const_int32(gallivm, reg->Register.Index*4 + swizzle);
 
-         scalar_ptr = LLVMBuildGEP(bld->base.builder, bld->consts_ptr,
+         scalar_ptr = LLVMBuildGEP(builder, bld->consts_ptr,
                                    &index, 1, "");
-         scalar = LLVMBuildLoad(bld->base.builder, scalar_ptr, "");
+         scalar = LLVMBuildLoad(builder, scalar_ptr, "");
 
          res = lp_build_broadcast_scalar(&bld->base, scalar);
       }
@@ -667,9 +692,9 @@ emit_fetch(
    case TGSI_FILE_INPUT:
       if (reg->Register.Indirect) {
          LLVMValueRef swizzle_vec =
-            lp_build_const_int_vec(uint_bld->type, swizzle);
+            lp_build_const_int_vec(gallivm, uint_bld->type, swizzle);
          LLVMValueRef length_vec =
-            lp_build_const_int_vec(uint_bld->type, bld->base.type.length);
+            lp_build_const_int_vec(gallivm, uint_bld->type, bld->base.type.length);
          LLVMValueRef index_vec;  /* index into the const buffer */
          LLVMValueRef inputs_array;
          LLVMTypeRef float4_ptr_type;
@@ -680,18 +705,19 @@ emit_fetch(
          index_vec = lp_build_mul(uint_bld, index_vec, length_vec);
 
          /* cast inputs_array pointer to float* */
-         float4_ptr_type = LLVMPointerType(LLVMFloatType(), 0);
-         inputs_array = LLVMBuildBitCast(uint_bld->builder, bld->inputs_array,
-                                        float4_ptr_type, "");
+         float4_ptr_type = LLVMPointerType(LLVMFloatTypeInContext(gallivm->context), 0);
+         inputs_array = LLVMBuildBitCast(builder, bld->inputs_array,
+                                         float4_ptr_type, "");
 
          /* Gather values from the temporary register array */
          res = build_gather(bld, inputs_array, index_vec);
       } else {
          if (bld->indirect_files & (1 << TGSI_FILE_INPUT)) {
-            LLVMValueRef lindex = lp_build_const_int32(reg->Register.Index * 4 + swizzle);
-            LLVMValueRef input_ptr =  LLVMBuildGEP(bld->base.builder,
+            LLVMValueRef lindex = lp_build_const_int32(gallivm,
+                                           reg->Register.Index * 4 + swizzle);
+            LLVMValueRef input_ptr =  LLVMBuildGEP(builder,
                                                    bld->inputs_array, &lindex, 1, "");
-            res = LLVMBuildLoad(bld->base.builder, input_ptr, "");
+            res = LLVMBuildLoad(builder, input_ptr, "");
          }
          else {
             res = bld->inputs[reg->Register.Index][swizzle];
@@ -703,9 +729,10 @@ emit_fetch(
    case TGSI_FILE_TEMPORARY:
       if (reg->Register.Indirect) {
          LLVMValueRef swizzle_vec =
-            lp_build_const_int_vec(uint_bld->type, swizzle);
+            lp_build_const_int_vec(bld->base.gallivm, uint_bld->type, swizzle);
          LLVMValueRef length_vec =
-            lp_build_const_int_vec(uint_bld->type, bld->base.type.length);
+            lp_build_const_int_vec(bld->base.gallivm, uint_bld->type,
+                                   bld->base.type.length);
          LLVMValueRef index_vec;  /* index into the const buffer */
          LLVMValueRef temps_array;
          LLVMTypeRef float4_ptr_type;
@@ -716,8 +743,8 @@ emit_fetch(
          index_vec = lp_build_mul(uint_bld, index_vec, length_vec);
 
          /* cast temps_array pointer to float* */
-         float4_ptr_type = LLVMPointerType(LLVMFloatType(), 0);
-         temps_array = LLVMBuildBitCast(uint_bld->builder, bld->temps_array,
+         float4_ptr_type = LLVMPointerType(LLVMFloatTypeInContext(bld->base.gallivm->context), 0);
+         temps_array = LLVMBuildBitCast(builder, bld->temps_array,
                                         float4_ptr_type, "");
 
          /* Gather values from the temporary register array */
@@ -726,7 +753,7 @@ emit_fetch(
       else {
          LLVMValueRef temp_ptr;
          temp_ptr = get_temp_ptr(bld, reg->Register.Index, swizzle);
-         res = LLVMBuildLoad(bld->base.builder, temp_ptr, "");
+         res = LLVMBuildLoad(builder, temp_ptr, "");
          if (!res)
             return bld->base.undef;
       }
@@ -796,6 +823,7 @@ emit_fetch_predicate(
    const struct tgsi_full_instruction *inst,
    LLVMValueRef *pred)
 {
+   LLVMBuilderRef builder = bld->base.gallivm->builder;
    unsigned index;
    unsigned char swizzles[4];
    LLVMValueRef unswizzled[4] = {NULL, NULL, NULL, NULL};
@@ -825,7 +853,7 @@ emit_fetch_predicate(
        * in the swizzles
        */
       if (!unswizzled[swizzle]) {
-         value = LLVMBuildLoad(bld->base.builder,
+         value = LLVMBuildLoad(builder,
                                bld->preds[index][swizzle], "");
 
          /*
@@ -835,13 +863,13 @@ emit_fetch_predicate(
           * is needlessly causing two comparisons due to storing the intermediate
           * result as float vector instead of an integer mask vector.
           */
-         value = lp_build_compare(bld->base.builder,
+         value = lp_build_compare(bld->base.gallivm,
                                   bld->base.type,
                                   PIPE_FUNC_NOTEQUAL,
                                   value,
                                   bld->base.zero);
          if (inst->Predicate.Negate) {
-            value = LLVMBuildNot(bld->base.builder, value, "");
+            value = LLVMBuildNot(builder, value, "");
          }
 
          unswizzled[swizzle] = value;
@@ -866,6 +894,8 @@ emit_store(
    LLVMValueRef pred,
    LLVMValueRef value)
 {
+   struct gallivm_state *gallivm = bld->base.gallivm;
+   LLVMBuilderRef builder = gallivm->builder;
    const struct tgsi_full_dst_register *reg = &inst->Dst[index];
    struct lp_build_context *uint_bld = &bld->uint_bld;
    LLVMValueRef indirect_index = NULL;
@@ -880,7 +910,7 @@ emit_store(
       break;
 
    case TGSI_SAT_MINUS_PLUS_ONE:
-      value = lp_build_max(&bld->base, value, lp_build_const_vec(bld->base.type, -1.0));
+      value = lp_build_max(&bld->base, value, lp_build_const_vec(bld->base.gallivm, bld->base.type, -1.0));
       value = lp_build_min(&bld->base, value, bld->base.one);
       break;
 
@@ -900,11 +930,10 @@ emit_store(
    switch( reg->Register.File ) {
    case TGSI_FILE_OUTPUT:
       if (reg->Register.Indirect) {
-         LLVMBuilderRef builder = bld->base.builder;
          LLVMValueRef chan_vec =
-            lp_build_const_int_vec(uint_bld->type, chan_index);
+            lp_build_const_int_vec(gallivm, uint_bld->type, chan_index);
          LLVMValueRef length_vec =
-            lp_build_const_int_vec(uint_bld->type, bld->base.type.length);
+            lp_build_const_int_vec(gallivm, uint_bld->type, bld->base.type.length);
          LLVMValueRef index_vec;  /* indexes into the temp registers */
          LLVMValueRef outputs_array;
          LLVMValueRef pixel_offsets;
@@ -914,7 +943,7 @@ emit_store(
          /* build pixel offset vector: {0, 1, 2, 3, ...} */
          pixel_offsets = uint_bld->undef;
          for (i = 0; i < bld->base.type.length; i++) {
-            LLVMValueRef ii = lp_build_const_int32(i);
+            LLVMValueRef ii = lp_build_const_int32(gallivm, i);
             pixel_offsets = LLVMBuildInsertElement(builder, pixel_offsets,
                                                    ii, ii, "");
          }
@@ -925,7 +954,8 @@ emit_store(
          index_vec = lp_build_mul(uint_bld, index_vec, length_vec);
          index_vec = lp_build_add(uint_bld, index_vec, pixel_offsets);
 
-         float_ptr_type = LLVMPointerType(LLVMFloatType(), 0);
+         float_ptr_type =
+            LLVMPointerType(LLVMFloatTypeInContext(gallivm->context), 0);
          outputs_array = LLVMBuildBitCast(builder, bld->outputs_array,
                                           float_ptr_type, "");
 
@@ -942,11 +972,11 @@ emit_store(
 
    case TGSI_FILE_TEMPORARY:
       if (reg->Register.Indirect) {
-         LLVMBuilderRef builder = bld->base.builder;
          LLVMValueRef chan_vec =
-            lp_build_const_int_vec(uint_bld->type, chan_index);
+            lp_build_const_int_vec(gallivm, uint_bld->type, chan_index);
          LLVMValueRef length_vec =
-            lp_build_const_int_vec(uint_bld->type, bld->base.type.length);
+            lp_build_const_int_vec(gallivm, uint_bld->type,
+                                   bld->base.type.length);
          LLVMValueRef index_vec;  /* indexes into the temp registers */
          LLVMValueRef temps_array;
          LLVMValueRef pixel_offsets;
@@ -956,7 +986,7 @@ emit_store(
          /* build pixel offset vector: {0, 1, 2, 3, ...} */
          pixel_offsets = uint_bld->undef; 
          for (i = 0; i < bld->base.type.length; i++) {
-            LLVMValueRef ii = lp_build_const_int32(i);
+            LLVMValueRef ii = lp_build_const_int32(gallivm, i);
             pixel_offsets = LLVMBuildInsertElement(builder, pixel_offsets,
                                                    ii, ii, "");
          }
@@ -967,7 +997,8 @@ emit_store(
          index_vec = lp_build_mul(uint_bld, index_vec, length_vec);
          index_vec = lp_build_add(uint_bld, index_vec, pixel_offsets);
 
-         float_ptr_type = LLVMPointerType(LLVMFloatType(), 0);
+         float_ptr_type =
+            LLVMPointerType(LLVMFloatTypeInContext(gallivm->context), 0);
          temps_array = LLVMBuildBitCast(builder, bld->temps_array,
                                         float_ptr_type, "");
 
@@ -984,7 +1015,7 @@ emit_store(
 
    case TGSI_FILE_ADDRESS:
       lp_exec_mask_store(&bld->exec_mask, pred, value,
-                         bld->addr[reg->Indirect.Index][chan_index]);
+                         bld->addr[reg->Register.Index][chan_index]);
       break;
 
    case TGSI_FILE_PREDICATE:
@@ -1008,6 +1039,7 @@ emit_tex( struct lp_build_tgsi_soa_context *bld,
           enum lp_build_tex_modifier modifier,
           LLVMValueRef *texel)
 {
+   LLVMBuilderRef builder = bld->base.gallivm->builder;
    unsigned unit;
    LLVMValueRef lod_bias, explicit_lod;
    LLVMValueRef oow = NULL;
@@ -1073,13 +1105,12 @@ emit_tex( struct lp_build_tgsi_soa_context *bld,
    }
 
    if (modifier == LP_BLD_TEX_MODIFIER_EXPLICIT_DERIV) {
-      LLVMTypeRef i32t = LLVMInt32Type();
-      LLVMValueRef index0 = LLVMConstInt(i32t, 0, 0);
+      LLVMValueRef index0 = lp_build_const_int32(bld->base.gallivm, 0);
       for (i = 0; i < num_coords; i++) {
          LLVMValueRef src1 = emit_fetch( bld, inst, 1, i );
          LLVMValueRef src2 = emit_fetch( bld, inst, 2, i );
-         ddx[i] = LLVMBuildExtractElement(bld->base.builder, src1, index0, "");
-         ddy[i] = LLVMBuildExtractElement(bld->base.builder, src2, index0, "");
+         ddx[i] = LLVMBuildExtractElement(builder, src1, index0, "");
+         ddy[i] = LLVMBuildExtractElement(builder, src2, index0, "");
       }
       unit = inst->Src[3].Register.Index;
    }  else {
@@ -1095,7 +1126,7 @@ emit_tex( struct lp_build_tgsi_soa_context *bld,
    }
 
    bld->sampler->emit_fetch_texel(bld->sampler,
-                                  bld->base.builder,
+                                  bld->base.gallivm,
                                   bld->base.type,
                                   unit, num_coords, coords,
                                   ddx, ddy,
@@ -1150,6 +1181,7 @@ emit_kil(
    const struct tgsi_full_instruction *inst,
    int pc)
 {
+   LLVMBuilderRef builder = bld->base.gallivm->builder;
    const struct tgsi_full_src_register *reg = &inst->Src[0];
    LLVMValueRef terms[NUM_CHANNELS];
    LLVMValueRef mask;
@@ -1181,7 +1213,7 @@ emit_kil(
          chan_mask = lp_build_cmp(&bld->base, PIPE_FUNC_GEQUAL, terms[chan_index], bld->base.zero);
 
          if(mask)
-            mask = LLVMBuildAnd(bld->base.builder, mask, chan_mask, "");
+            mask = LLVMBuildAnd(builder, mask, chan_mask, "");
          else
             mask = chan_mask;
       }
@@ -1207,13 +1239,14 @@ emit_kilp(struct lp_build_tgsi_soa_context *bld,
           const struct tgsi_full_instruction *inst,
 	  int pc)
 {
+   LLVMBuilderRef builder = bld->base.gallivm->builder;
    LLVMValueRef mask;
 
    /* For those channels which are "alive", disable fragment shader
     * execution.
     */
    if (bld->exec_mask.has_mask) {
-      mask = LLVMBuildNot(bld->base.builder, bld->exec_mask.exec_mask, "kilp");
+      mask = LLVMBuildNot(builder, bld->exec_mask.exec_mask, "kilp");
    }
    else {
       LLVMValueRef zero = LLVMConstNull(bld->base.int_vec_type);
@@ -1234,38 +1267,39 @@ emit_kilp(struct lp_build_tgsi_soa_context *bld,
 static void
 emit_dump_temps(struct lp_build_tgsi_soa_context *bld)
 {
-   LLVMBuilderRef builder = bld->base.builder;
+   struct gallivm_state *gallivm = bld->base.gallivm;
+   LLVMBuilderRef builder = gallivm->builder;
    LLVMValueRef temp_ptr;
-   LLVMValueRef i0 = lp_build_const_int32(0);
-   LLVMValueRef i1 = lp_build_const_int32(1);
-   LLVMValueRef i2 = lp_build_const_int32(2);
-   LLVMValueRef i3 = lp_build_const_int32(3);
+   LLVMValueRef i0 = lp_build_const_int32(gallivm, 0);
+   LLVMValueRef i1 = lp_build_const_int32(gallivm, 1);
+   LLVMValueRef i2 = lp_build_const_int32(gallivm, 2);
+   LLVMValueRef i3 = lp_build_const_int32(gallivm, 3);
    int index;
    int n = bld->info->file_max[TGSI_FILE_TEMPORARY];
 
    for (index = 0; index < n; index++) {
-      LLVMValueRef idx = lp_build_const_int32(index);
+      LLVMValueRef idx = lp_build_const_int32(gallivm, index);
       LLVMValueRef v[4][4], res;
       int chan;
 
-      lp_build_printf(builder, "TEMP[%d]:\n", idx);
+      lp_build_printf(gallivm, "TEMP[%d]:\n", idx);
 
       for (chan = 0; chan < 4; chan++) {
          temp_ptr = get_temp_ptr(bld, index, chan);
-         res = LLVMBuildLoad(bld->base.builder, temp_ptr, "");
+         res = LLVMBuildLoad(builder, temp_ptr, "");
          v[chan][0] = LLVMBuildExtractElement(builder, res, i0, "");
          v[chan][1] = LLVMBuildExtractElement(builder, res, i1, "");
          v[chan][2] = LLVMBuildExtractElement(builder, res, i2, "");
          v[chan][3] = LLVMBuildExtractElement(builder, res, i3, "");
       }
 
-      lp_build_printf(builder, "  X: %f %f %f %f\n",
+      lp_build_printf(gallivm, "  X: %f %f %f %f\n",
                       v[0][0], v[0][1], v[0][2], v[0][3]);
-      lp_build_printf(builder, "  Y: %f %f %f %f\n",
+      lp_build_printf(gallivm, "  Y: %f %f %f %f\n",
                       v[1][0], v[1][1], v[1][2], v[1][3]);
-      lp_build_printf(builder, "  Z: %f %f %f %f\n",
+      lp_build_printf(gallivm, "  Z: %f %f %f %f\n",
                       v[2][0], v[2][1], v[2][2], v[2][3]);
-      lp_build_printf(builder, "  W: %f %f %f %f\n",
+      lp_build_printf(gallivm, "  W: %f %f %f %f\n",
                       v[3][0], v[3][1], v[3][2], v[3][3]);
    }
 }
@@ -1277,6 +1311,7 @@ emit_declaration(
    struct lp_build_tgsi_soa_context *bld,
    const struct tgsi_full_declaration *decl)
 {
+   struct gallivm_state *gallivm = bld->base.gallivm;
    LLVMTypeRef vec_type = bld->base.vec_type;
    const unsigned first = decl->Range.First;
    const unsigned last = decl->Range.Last;
@@ -1289,15 +1324,14 @@ emit_declaration(
          assert(idx < LP_MAX_TGSI_TEMPS);
          if (!(bld->indirect_files & (1 << TGSI_FILE_TEMPORARY))) {
             for (i = 0; i < NUM_CHANNELS; i++)
-               bld->temps[idx][i] = lp_build_alloca(bld->base.builder,
-                                                    vec_type, "temp");
+               bld->temps[idx][i] = lp_build_alloca(gallivm, vec_type, "temp");
          }
          break;
 
       case TGSI_FILE_OUTPUT:
          if (!(bld->indirect_files & (1 << TGSI_FILE_OUTPUT))) {
             for (i = 0; i < NUM_CHANNELS; i++)
-               bld->outputs[idx][i] = lp_build_alloca(bld->base.builder,
+               bld->outputs[idx][i] = lp_build_alloca(gallivm,
                                                       vec_type, "output");
          }
          break;
@@ -1305,15 +1339,14 @@ emit_declaration(
       case TGSI_FILE_ADDRESS:
          assert(idx < LP_MAX_TGSI_ADDRS);
          for (i = 0; i < NUM_CHANNELS; i++)
-            bld->addr[idx][i] = lp_build_alloca(bld->base.builder,
-                                                vec_type, "addr");
+            bld->addr[idx][i] = lp_build_alloca(gallivm, vec_type, "addr");
          break;
 
       case TGSI_FILE_PREDICATE:
          assert(idx < LP_MAX_TGSI_PREDS);
          for (i = 0; i < NUM_CHANNELS; i++)
-            bld->preds[idx][i] = lp_build_alloca(bld->base.builder,
-                                                 vec_type, "predicate");
+            bld->preds[idx][i] = lp_build_alloca(gallivm, vec_type,
+                                                 "predicate");
          break;
 
       default:
@@ -1639,7 +1672,7 @@ emit_instruction(
          src0 = emit_fetch( bld, inst, 0, chan_index );
          src1 = emit_fetch( bld, inst, 1, chan_index );
          src2 = emit_fetch( bld, inst, 2, chan_index );
-         tmp1 = lp_build_const_vec(bld->base.type, 0.5);
+         tmp1 = lp_build_const_vec(bld->base.gallivm, bld->base.type, 0.5);
          tmp0 = lp_build_cmp( &bld->base, PIPE_FUNC_GREATER, src2, tmp1);
          dst0[chan_index] = lp_build_select( &bld->base, tmp0, src0, src1 );
       }
@@ -2151,7 +2184,7 @@ emit_instruction(
       break;
 
    case TGSI_OPCODE_ENDLOOP:
-      lp_exec_endloop(&bld->exec_mask);
+      lp_exec_endloop(bld->base.gallivm, &bld->exec_mask);
       break;
 
    case TGSI_OPCODE_ENDSUB:
@@ -2284,7 +2317,7 @@ emit_instruction(
 
 
 void
-lp_build_tgsi_soa(LLVMBuilderRef builder,
+lp_build_tgsi_soa(struct gallivm_state *gallivm,
                   const struct tgsi_token *tokens,
                   struct lp_type type,
                   struct lp_build_mask_context *mask,
@@ -2312,9 +2345,9 @@ lp_build_tgsi_soa(LLVMBuilderRef builder,
 
    /* Setup build context */
    memset(&bld, 0, sizeof bld);
-   lp_build_context_init(&bld.base, builder, type);
-   lp_build_context_init(&bld.uint_bld, builder, lp_uint_type(type));
-   lp_build_context_init(&bld.elem_bld, builder, lp_elem_type(type));
+   lp_build_context_init(&bld.base, gallivm, type);
+   lp_build_context_init(&bld.uint_bld, gallivm, lp_uint_type(type));
+   lp_build_context_init(&bld.elem_bld, gallivm, lp_elem_type(type));
    bld.mask = mask;
    bld.pos = pos;
    bld.inputs = inputs;
@@ -2334,17 +2367,19 @@ lp_build_tgsi_soa(LLVMBuilderRef builder,
    lp_exec_mask_init(&bld.exec_mask, &bld.base);
 
    if (bld.indirect_files & (1 << TGSI_FILE_TEMPORARY)) {
-      LLVMValueRef array_size = LLVMConstInt(LLVMInt32Type(),
-                                             info->file_max[TGSI_FILE_TEMPORARY]*4 + 4, 0);
-      bld.temps_array = lp_build_array_alloca(bld.base.builder,
+      LLVMValueRef array_size =
+         lp_build_const_int32(gallivm,
+                              info->file_max[TGSI_FILE_TEMPORARY] * 4 + 4);
+      bld.temps_array = lp_build_array_alloca(gallivm,
                                               bld.base.vec_type, array_size,
                                               "temp_array");
    }
 
    if (bld.indirect_files & (1 << TGSI_FILE_OUTPUT)) {
-      LLVMValueRef array_size = LLVMConstInt(LLVMInt32Type(),
-                                             info->file_max[TGSI_FILE_OUTPUT]*4 + 4, 0);
-      bld.outputs_array = lp_build_array_alloca(bld.base.builder,
+      LLVMValueRef array_size =
+         lp_build_const_int32(gallivm,
+                              info->file_max[TGSI_FILE_OUTPUT] * 4 + 4);
+      bld.outputs_array = lp_build_array_alloca(gallivm,
                                                 bld.base.vec_type, array_size,
                                                 "output_array");
    }
@@ -2354,9 +2389,9 @@ lp_build_tgsi_soa(LLVMBuilderRef builder,
    if (bld.indirect_files & (1 << TGSI_FILE_INPUT)) {
       unsigned index, chan;
       LLVMTypeRef vec_type = bld.base.vec_type;
-      LLVMValueRef array_size = LLVMConstInt(LLVMInt32Type(),
-                                             info->file_max[TGSI_FILE_INPUT]*4 + 4, 0);
-      bld.inputs_array = lp_build_array_alloca(bld.base.builder,
+      LLVMValueRef array_size =
+         lp_build_const_int32(gallivm, info->file_max[TGSI_FILE_INPUT]*4 + 4);
+      bld.inputs_array = lp_build_array_alloca(gallivm,
                                                vec_type, array_size,
                                                "input_array");
 
@@ -2364,13 +2399,14 @@ lp_build_tgsi_soa(LLVMBuilderRef builder,
 
       for (index = 0; index < info->num_inputs; ++index) {
          for (chan = 0; chan < NUM_CHANNELS; ++chan) {
-            LLVMValueRef lindex = lp_build_const_int32(index * 4 + chan);
+            LLVMValueRef lindex =
+               lp_build_const_int32(gallivm, index * 4 + chan);
             LLVMValueRef input_ptr =
-               LLVMBuildGEP(bld.base.builder, bld.inputs_array,
+               LLVMBuildGEP(gallivm->builder, bld.inputs_array,
                             &lindex, 1, "");
             LLVMValueRef value = bld.inputs[index][chan];
             if (value)
-               LLVMBuildStore(bld.base.builder, value, input_ptr);
+               LLVMBuildStore(gallivm->builder, value, input_ptr);
          }
       }
    }
@@ -2420,7 +2456,7 @@ lp_build_tgsi_soa(LLVMBuilderRef builder,
             assert(num_immediates < LP_MAX_TGSI_IMMEDIATES);
             for( i = 0; i < size; ++i )
                bld.immediates[num_immediates][i] =
-                  lp_build_const_vec(type, parse.FullToken.FullImmediate.u[i].Float);
+                  lp_build_const_vec(gallivm, type, parse.FullToken.FullImmediate.u[i].Float);
             for( i = size; i < 4; ++i )
                bld.immediates[num_immediates][i] = bld.base.undef;
             num_immediates++;
@@ -2457,7 +2493,7 @@ lp_build_tgsi_soa(LLVMBuilderRef builder,
    }
 
    if (0) {
-      LLVMBasicBlockRef block = LLVMGetInsertBlock(builder);
+      LLVMBasicBlockRef block = LLVMGetInsertBlock(gallivm->builder);
       LLVMValueRef function = LLVMGetBasicBlockParent(block);
       debug_printf("11111111111111111111111111111 \n");
       tgsi_dump(tokens, 0);
@@ -2468,7 +2504,7 @@ lp_build_tgsi_soa(LLVMBuilderRef builder,
 
    if (0) {
       LLVMModuleRef module = LLVMGetGlobalParent(
-         LLVMGetBasicBlockParent(LLVMGetInsertBlock(bld.base.builder)));
+         LLVMGetBasicBlockParent(LLVMGetInsertBlock(gallivm->builder)));
       LLVMDumpModule(module);
 
    }
diff --git a/src/gallium/auxiliary/gallivm/lp_bld_type.c b/src/gallium/auxiliary/gallivm/lp_bld_type.c
index 5205c7ada9..c5cf6d4a6c 100644
--- a/src/gallium/auxiliary/gallivm/lp_bld_type.c
+++ b/src/gallium/auxiliary/gallivm/lp_bld_type.c
@@ -30,34 +30,35 @@
 
 #include "lp_bld_type.h"
 #include "lp_bld_const.h"
+#include "lp_bld_init.h"
 
 
 LLVMTypeRef
-lp_build_elem_type(struct lp_type type)
+lp_build_elem_type(struct gallivm_state *gallivm, struct lp_type type)
 {
    if (type.floating) {
       switch(type.width) {
       case 32:
-         return LLVMFloatType();
+         return LLVMFloatTypeInContext(gallivm->context);
          break;
       case 64:
-         return LLVMDoubleType();
+         return LLVMDoubleTypeInContext(gallivm->context);
          break;
       default:
          assert(0);
-         return LLVMFloatType();
+         return LLVMFloatTypeInContext(gallivm->context);
       }
    }
    else {
-      return LLVMIntType(type.width);
+      return LLVMIntTypeInContext(gallivm->context, type.width);
    }
 }
 
 
 LLVMTypeRef
-lp_build_vec_type(struct lp_type type)
+lp_build_vec_type(struct gallivm_state *gallivm,struct lp_type type)
 {
-   LLVMTypeRef elem_type = lp_build_elem_type(type);
+   LLVMTypeRef elem_type = lp_build_elem_type(gallivm, type);
    if (type.length == 1)
       return elem_type;
    else
@@ -149,16 +150,16 @@ lp_check_value(struct lp_type type, LLVMValueRef val)
 
 
 LLVMTypeRef
-lp_build_int_elem_type(struct lp_type type)
+lp_build_int_elem_type(struct gallivm_state *gallivm, struct lp_type type)
 {
-   return LLVMIntType(type.width);
+   return LLVMIntTypeInContext(gallivm->context, type.width);
 }
 
 
 LLVMTypeRef
-lp_build_int_vec_type(struct lp_type type)
+lp_build_int_vec_type(struct gallivm_state *gallivm, struct lp_type type)
 {
-   LLVMTypeRef elem_type = lp_build_int_elem_type(type);
+   LLVMTypeRef elem_type = lp_build_int_elem_type(gallivm, type);
    if (type.length == 1)
       return elem_type;
    else
@@ -170,7 +171,7 @@ lp_build_int_vec_type(struct lp_type type)
  * Build int32[4] vector type
  */
 LLVMTypeRef
-lp_build_int32_vec4_type(void)
+lp_build_int32_vec4_type(struct gallivm_state *gallivm)
 {
    struct lp_type t;
    LLVMTypeRef type;
@@ -182,7 +183,7 @@ lp_build_int32_vec4_type(void)
    t.width = 32;       /* 32-bit int */
    t.length = 4;       /* 4 elements per vector */
 
-   type = lp_build_int_elem_type(t);
+   type = lp_build_int_elem_type(gallivm, t);
    return LLVMVectorType(type, t.length);
 }
 
@@ -383,15 +384,15 @@ lp_dump_llvmtype(LLVMTypeRef t)
 
 void
 lp_build_context_init(struct lp_build_context *bld,
-                      LLVMBuilderRef builder,
+                      struct gallivm_state *gallivm,
                       struct lp_type type)
 {
-   bld->builder = builder;
+   bld->gallivm = gallivm;
    bld->type = type;
 
-   bld->int_elem_type = lp_build_int_elem_type(type);
+   bld->int_elem_type = lp_build_int_elem_type(gallivm, type);
    if (type.floating)
-      bld->elem_type = lp_build_elem_type(type);
+      bld->elem_type = lp_build_elem_type(gallivm, type);
    else
       bld->elem_type = bld->int_elem_type;
 
@@ -406,5 +407,5 @@ lp_build_context_init(struct lp_build_context *bld,
 
    bld->undef = LLVMGetUndef(bld->vec_type);
    bld->zero = LLVMConstNull(bld->vec_type);
-   bld->one = lp_build_one(type);
+   bld->one = lp_build_one(gallivm, type);
 }
diff --git a/src/gallium/auxiliary/gallivm/lp_bld_type.h b/src/gallium/auxiliary/gallivm/lp_bld_type.h
index a135d0df84..5007e83ac5 100644
--- a/src/gallium/auxiliary/gallivm/lp_bld_type.h
+++ b/src/gallium/auxiliary/gallivm/lp_bld_type.h
@@ -120,7 +120,7 @@ struct lp_type {
  */
 struct lp_build_context
 {
-   LLVMBuilderRef builder;
+   struct gallivm_state *gallivm;
 
    /**
     * This not only describes the input/output LLVM types, but also whether
@@ -285,11 +285,11 @@ lp_type_ufixed(unsigned width)
 
 
 LLVMTypeRef
-lp_build_elem_type(struct lp_type type);
+lp_build_elem_type(struct gallivm_state *gallivm, struct lp_type type);
 
 
 LLVMTypeRef
-lp_build_vec_type(struct lp_type type);
+lp_build_vec_type(struct gallivm_state *gallivm, struct lp_type type);
 
 
 boolean
@@ -305,15 +305,15 @@ lp_check_value(struct lp_type type, LLVMValueRef val);
 
 
 LLVMTypeRef
-lp_build_int_elem_type(struct lp_type type);
+lp_build_int_elem_type(struct gallivm_state *gallivm, struct lp_type type);
 
 
 LLVMTypeRef
-lp_build_int_vec_type(struct lp_type type);
+lp_build_int_vec_type(struct gallivm_state *gallivm, struct lp_type type);
 
 
 LLVMTypeRef
-lp_build_int32_vec4_type(void);
+lp_build_int32_vec4_type(struct gallivm_state *gallivm);
 
 
 static INLINE struct lp_type
@@ -394,7 +394,7 @@ lp_dump_llvmtype(LLVMTypeRef t);
 
 void
 lp_build_context_init(struct lp_build_context *bld,
-                      LLVMBuilderRef builder,
+                      struct gallivm_state *gallivm,
                       struct lp_type type);
 
 
diff --git a/src/gallium/auxiliary/target-helpers/wrap_screen.c b/src/gallium/auxiliary/target-helpers/wrap_screen.c
deleted file mode 100644
index df5d56a53c..0000000000
--- a/src/gallium/auxiliary/target-helpers/wrap_screen.c
+++ /dev/null
@@ -1,68 +0,0 @@
-/**************************************************************************
- * 
- * Copyright 2007 Tungsten Graphics, Inc., Bismarck, ND., USA
- * All Rights Reserved.
- * 
- * Permission is hereby granted, free of charge, to any person obtaining a
- * copy of this software and associated documentation files (the
- * "Software"), to deal in the Software without restriction, including
- * without limitation the rights to use, copy, modify, merge, publish,
- * distribute, sub license, and/or sell copies of the Software, and to
- * permit persons to whom the Software is furnished to do so, subject to
- * the following conditions:
- * 
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
- * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
- * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
- * THE COPYRIGHT HOLDERS, AUTHORS AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM,
- * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR 
- * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE 
- * USE OR OTHER DEALINGS IN THE SOFTWARE.
- *
- * The above copyright notice and this permission notice (including the
- * next paragraph) shall be included in all copies or substantial portions
- * of the Software.
- * 
- * 
- **************************************************************************/
-
-/*
- * Authors:
- *   Keith Whitwell
- */
-
-#include "target-helpers/wrap_screen.h"
-#include "trace/tr_public.h"
-#include "rbug/rbug_public.h"
-#include "identity/id_public.h"
-#include "util/u_debug.h"
-
-
-/* Centralized code to inject common wrapping layers:
- */
-struct pipe_screen *
-gallium_wrap_screen( struct pipe_screen *screen )
-{
-   /* Screen wrapping functions are required not to fail.  If it is
-    * impossible to wrap a screen, the unwrapped screen should be
-    * returned instead.  Any failure condition should be returned in
-    * an OUT argument.
-    *
-    * Otherwise it is really messy trying to clean up in this code.
-    */
-   if (debug_get_bool_option("GALLIUM_WRAP", FALSE)) {
-      screen = identity_screen_create(screen);
-   }
-
-   /* Trace does its own checking if it should run */
-   screen = trace_screen_create(screen);
-
-   /* Rbug does its own checking if it should run */
-   screen = rbug_screen_create(screen);
-
-   return screen;
-}
-
-
-
-
diff --git a/src/gallium/auxiliary/target-helpers/wrap_screen.h b/src/gallium/auxiliary/target-helpers/wrap_screen.h
deleted file mode 100644
index 7e76beb7c5..0000000000
--- a/src/gallium/auxiliary/target-helpers/wrap_screen.h
+++ /dev/null
@@ -1,16 +0,0 @@
-#ifndef WRAP_SCREEN_HELPER_H
-#define WRAP_SCREEN_HELPER_H
-
-#include "pipe/p_compiler.h"
-
-struct pipe_screen;
-
-/* Centralized code to inject common wrapping layers.  Other layers
- * can be introduced by specific targets, but these are the generally
- * helpful ones we probably want everywhere.
- */
-struct pipe_screen *
-gallium_wrap_screen( struct pipe_screen *screen );
-
-
-#endif
diff --git a/src/gallium/auxiliary/tgsi/tgsi_exec.h b/src/gallium/auxiliary/tgsi/tgsi_exec.h
index 7b07778601..b5ebbfbfaa 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_exec.h
+++ b/src/gallium/auxiliary/tgsi/tgsi_exec.h
@@ -388,6 +388,8 @@ tgsi_exec_get_shader_param(enum pipe_shader_cap param)
    case PIPE_SHADER_CAP_INDIRECT_TEMP_ADDR:
    case PIPE_SHADER_CAP_INDIRECT_CONST_ADDR:
       return 1;
+   case PIPE_SHADER_CAP_SUBROUTINES:
+      return 1;
    default:
       return 0;
    }
diff --git a/src/gallium/auxiliary/tgsi/tgsi_parse.h b/src/gallium/auxiliary/tgsi/tgsi_parse.h
index d4df585176..2aafa2a6e8 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_parse.h
+++ b/src/gallium/auxiliary/tgsi/tgsi_parse.h
@@ -136,7 +136,8 @@ tgsi_parse_token(
 static INLINE unsigned
 tgsi_num_tokens(const struct tgsi_token *tokens)
 {
-   struct tgsi_header header = *(const struct tgsi_header *) tokens;
+   struct tgsi_header header;
+   memcpy(&header, tokens, sizeof(header));
    return header.HeaderSize + header.BodySize;
 }
 
diff --git a/src/gallium/auxiliary/tgsi/tgsi_text.c b/src/gallium/auxiliary/tgsi/tgsi_text.c
index b01d2ff468..9a38c37979 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_text.c
+++ b/src/gallium/auxiliary/tgsi/tgsi_text.c
@@ -1007,7 +1007,8 @@ static const char *semantic_names[TGSI_SEMANTIC_COUNT] =
    "FACE",
    "EDGEFLAG",
    "PRIM_ID",
-   "INSTANCEID"
+   "INSTANCEID",
+   "STENCIL"
 };
 
 static const char *interpolate_names[TGSI_INTERPOLATE_COUNT] =
diff --git a/src/gallium/auxiliary/util/u_blit.c b/src/gallium/auxiliary/util/u_blit.c
index dfb142b9e1..c11f7d383d 100644
--- a/src/gallium/auxiliary/util/u_blit.c
+++ b/src/gallium/auxiliary/util/u_blit.c
@@ -291,7 +291,7 @@ regions_overlap(int srcX0, int srcY0,
 void
 util_blit_pixels_writemask(struct blit_state *ctx,
                            struct pipe_resource *src_tex,
-                           struct pipe_subresource srcsub,
+                           unsigned src_level,
                            int srcX0, int srcY0,
                            int srcX1, int srcY1,
                            int srcZ0,
@@ -316,13 +316,12 @@ util_blit_pixels_writemask(struct blit_state *ctx,
    assert(filter == PIPE_TEX_MIPFILTER_NEAREST ||
           filter == PIPE_TEX_MIPFILTER_LINEAR);
 
-   assert(srcsub.level <= src_tex->last_level);
+   assert(src_level <= src_tex->last_level);
 
    /* do the regions overlap? */
    overlap = src_tex == dst->texture &&
-             dst->face == srcsub.face &&
-             dst->level == srcsub.level &&
-             dst->zslice == srcZ0 &&
+             dst->u.tex.level == src_level &&
+             dst->u.tex.first_layer == srcZ0 &&
       regions_overlap(srcX0, srcY0, srcX1, srcY1,
                       dstX0, dstY0, dstX1, dstY1);
 
@@ -339,16 +338,19 @@ util_blit_pixels_writemask(struct blit_state *ctx,
        (dstX1 - dstX0) == (srcX1 - srcX0) &&
        (dstY1 - dstY0) == (srcY1 - srcY0) &&
        !overlap) {
-      struct pipe_subresource subdst;
-      subdst.face = dst->face;
-      subdst.level = dst->level;
+      struct pipe_box src_box;
+      src_box.x = srcX0;
+      src_box.y = srcY0;
+      src_box.z = srcZ0;
+      src_box.width = srcW;
+      src_box.height = srcH;
+      src_box.depth = 1;
       pipe->resource_copy_region(pipe,
-                                 dst->texture, subdst,
-                                 dstX0, dstY0, dst->zslice,/* dest */
-                                 src_tex, srcsub,
-                                 srcX0, srcY0, srcZ0,/* src */
-                                 srcW, srcH);       /* size */
-      return;
+                                 dst->texture, dst->u.tex.level,
+                                 dstX0, dstY0, dst->u.tex.first_layer,/* dest */
+                                 src_tex, src_level,
+                                 &src_box);
+       return;
    }
 
    /* Create a temporary texture when src and dest alias or when src
@@ -359,16 +361,16 @@ util_blit_pixels_writemask(struct blit_state *ctx,
     * This can still be improved upon.
     */
    if ((src_tex == dst->texture &&
-       dst->face == srcsub.face &&
-       dst->level == srcsub.level &&
-       dst->zslice == srcZ0) ||
+       dst->u.tex.level == src_level &&
+       dst->u.tex.first_layer == srcZ0) ||
        (src_tex->target != PIPE_TEXTURE_2D &&
+       src_tex->target != PIPE_TEXTURE_2D &&
        src_tex->target != PIPE_TEXTURE_RECT))
    {
       struct pipe_resource texTemp;
       struct pipe_resource *tex;
       struct pipe_sampler_view sv_templ;
-      struct pipe_subresource texsub;
+      struct pipe_box src_box;
       const int srcLeft = MIN2(srcX0, srcX1);
       const int srcTop = MIN2(srcY0, srcY1);
 
@@ -394,19 +396,23 @@ util_blit_pixels_writemask(struct blit_state *ctx,
       texTemp.width0 = srcW;
       texTemp.height0 = srcH;
       texTemp.depth0 = 1;
+      texTemp.array_size = 1;
       texTemp.bind = PIPE_BIND_SAMPLER_VIEW;
 
       tex = screen->resource_create(screen, &texTemp);
       if (!tex)
          return;
 
-      texsub.face = 0;
-      texsub.level = 0;
+      src_box.x = srcLeft;
+      src_box.y = srcTop;
+      src_box.z = srcZ0;
+      src_box.width = srcW;
+      src_box.height = srcH;
+      src_box.depth = 1;
       /* load temp texture */
       pipe->resource_copy_region(pipe,
-                                 tex, texsub, 0, 0, 0,  /* dest */
-                                 src_tex, srcsub, srcLeft, srcTop, srcZ0, /* src */
-                                 srcW, srcH);     /* size */
+                                 tex, 0, 0, 0, 0,  /* dest */
+                                 src_tex, src_level, &src_box);
 
       normalized = tex->target != PIPE_TEXTURE_RECT;
       if(normalized) {
@@ -433,7 +439,6 @@ util_blit_pixels_writemask(struct blit_state *ctx,
    }
    else {
       u_sampler_view_default_template(&sv_templ, src_tex, src_tex->format);
-      sv_templ.first_level = sv_templ.last_level = srcsub.level;
       sampler_view = pipe->create_sampler_view(pipe, src_tex, &sv_templ);
 
       if (!sampler_view) {
@@ -447,10 +452,10 @@ util_blit_pixels_writemask(struct blit_state *ctx,
       normalized = sampler_view->texture->target != PIPE_TEXTURE_RECT;
       if(normalized)
       {
-         s0 /= (float)(u_minify(sampler_view->texture->width0, srcsub.level));
-         s1 /= (float)(u_minify(sampler_view->texture->width0, srcsub.level));
-         t0 /= (float)(u_minify(sampler_view->texture->height0, srcsub.level));
-         t1 /= (float)(u_minify(sampler_view->texture->height0, srcsub.level));
+         s0 /= (float)(u_minify(sampler_view->texture->width0, src_level));
+         s1 /= (float)(u_minify(sampler_view->texture->width0, src_level));
+         t0 /= (float)(u_minify(sampler_view->texture->height0, src_level));
+         t1 /= (float)(u_minify(sampler_view->texture->height0, src_level));
       }
    }
 
@@ -489,9 +494,8 @@ util_blit_pixels_writemask(struct blit_state *ctx,
    ctx->sampler.normalized_coords = normalized;
    ctx->sampler.min_img_filter = filter;
    ctx->sampler.mag_img_filter = filter;
-   /* we've limited this already with the sampler view but you never know... */
-   ctx->sampler.min_lod = srcsub.level;
-   ctx->sampler.max_lod = srcsub.level;
+   ctx->sampler.min_lod = src_level;
+   ctx->sampler.max_lod = src_level;
    cso_single_sampler(ctx->cso, 0, &ctx->sampler);
    cso_single_sampler_done(ctx->cso);
 
@@ -575,7 +579,7 @@ util_blit_pixels_writemask(struct blit_state *ctx,
 void
 util_blit_pixels(struct blit_state *ctx,
                  struct pipe_resource *src_tex,
-                 struct pipe_subresource srcsub,
+                 unsigned src_level,
                  int srcX0, int srcY0,
                  int srcX1, int srcY1,
                  int srcZ,
@@ -585,7 +589,7 @@ util_blit_pixels(struct blit_state *ctx,
                  float z, uint filter )
 {
    util_blit_pixels_writemask( ctx, src_tex,
-                               srcsub,
+                               src_level,
                                srcX0, srcY0,
                                srcX1, srcY1,
                                srcZ,
@@ -662,6 +666,7 @@ util_blit_pixels_tex(struct blit_state *ctx,
    cso_save_rasterizer(ctx->cso);
    cso_save_samplers(ctx->cso);
    cso_save_fragment_sampler_views(ctx->cso);
+   cso_save_viewport(ctx->cso);
    cso_save_framebuffer(ctx->cso);
    cso_save_fragment_shader(ctx->cso);
    cso_save_vertex_shader(ctx->cso);
@@ -729,6 +734,7 @@ util_blit_pixels_tex(struct blit_state *ctx,
    cso_restore_rasterizer(ctx->cso);
    cso_restore_samplers(ctx->cso);
    cso_restore_fragment_sampler_views(ctx->cso);
+   cso_restore_viewport(ctx->cso);
    cso_restore_framebuffer(ctx->cso);
    cso_restore_fragment_shader(ctx->cso);
    cso_restore_vertex_shader(ctx->cso);
diff --git a/src/gallium/auxiliary/util/u_blit.h b/src/gallium/auxiliary/util/u_blit.h
index b8a0dfce13..3009e25eca 100644
--- a/src/gallium/auxiliary/util/u_blit.h
+++ b/src/gallium/auxiliary/util/u_blit.h
@@ -42,7 +42,6 @@ struct cso_context;
 struct pipe_context;
 struct pipe_resource;
 struct pipe_sampler_view;
-struct pipe_subresource;
 struct pipe_surface;
 
 
@@ -55,7 +54,7 @@ util_destroy_blit(struct blit_state *ctx);
 extern void
 util_blit_pixels(struct blit_state *ctx,
                  struct pipe_resource *src_tex,
-                 struct pipe_subresource srcsub,
+                 unsigned src_level,
                  int srcX0, int srcY0,
                  int srcX1, int srcY1,
                  int srcZ0,
@@ -67,7 +66,7 @@ util_blit_pixels(struct blit_state *ctx,
 void
 util_blit_pixels_writemask(struct blit_state *ctx,
                            struct pipe_resource *src_tex,
-                           struct pipe_subresource srcsub,
+                           unsigned src_level,
                            int srcX0, int srcY0,
                            int srcX1, int srcY1,
                            int srcZ0,
diff --git a/src/gallium/auxiliary/util/u_blitter.c b/src/gallium/auxiliary/util/u_blitter.c
index a163f93cb8..545021d264 100644
--- a/src/gallium/auxiliary/util/u_blitter.c
+++ b/src/gallium/auxiliary/util/u_blitter.c
@@ -63,8 +63,7 @@ struct blitter_context_priv
 
    /* Constant state objects. */
    /* Vertex shaders. */
-   void *vs_col; /**< Vertex shader which passes {pos, color} to the output */
-   void *vs_tex; /**< Vertex shader which passes {pos, texcoord} to the output.*/
+   void *vs; /**< Vertex shader which passes {pos, generic} to the output.*/
 
    /* Fragment shaders. */
    /* The shader at index i outputs color to color buffers 0,1,...,i-1. */
@@ -211,20 +210,12 @@ struct blitter_context *util_blitter_create(struct pipe_context *pipe)
 
    /* fragment shaders are created on-demand */
 
-   /* vertex shaders */
-   {
-      const uint semantic_names[] = { TGSI_SEMANTIC_POSITION,
-                                      TGSI_SEMANTIC_COLOR };
-      const uint semantic_indices[] = { 0, 0 };
-      ctx->vs_col =
-         util_make_vertex_passthrough_shader(pipe, 2, semantic_names,
-                                             semantic_indices);
-   }
+   /* vertex shader */
    {
       const uint semantic_names[] = { TGSI_SEMANTIC_POSITION,
                                       TGSI_SEMANTIC_GENERIC };
       const uint semantic_indices[] = { 0, 0 };
-      ctx->vs_tex =
+      ctx->vs =
          util_make_vertex_passthrough_shader(pipe, 2, semantic_names,
                                              semantic_indices);
    }
@@ -257,8 +248,7 @@ void util_blitter_destroy(struct blitter_context *blitter)
    pipe->delete_depth_stencil_alpha_state(pipe, ctx->dsa_flush_depth_stencil);
 
    pipe->delete_rasterizer_state(pipe, ctx->rs_state);
-   pipe->delete_vs_state(pipe, ctx->vs_col);
-   pipe->delete_vs_state(pipe, ctx->vs_tex);
+   pipe->delete_vs_state(pipe, ctx->vs);
    pipe->delete_vertex_elements_state(pipe, ctx->velem_state);
 
    for (i = 0; i < PIPE_MAX_TEXTURE_TYPES; i++) {
@@ -419,17 +409,17 @@ static void blitter_set_clear_color(struct blitter_context_priv *ctx,
 }
 
 static void get_texcoords(struct pipe_resource *src,
-                                     struct pipe_subresource subsrc,
-                                     unsigned x1, unsigned y1,
-                                     unsigned x2, unsigned y2,
-                                     boolean normalized, float out[4])
+                          unsigned level,
+                          unsigned x1, unsigned y1,
+                          unsigned x2, unsigned y2,
+                          boolean normalized, float out[4])
 {
    if(normalized)
    {
-      out[0] = x1 / (float)u_minify(src->width0,  subsrc.level);
-      out[1] = y1 / (float)u_minify(src->height0, subsrc.level);
-      out[2] = x2 / (float)u_minify(src->width0,  subsrc.level);
-      out[3] = y2 / (float)u_minify(src->height0, subsrc.level);
+      out[0] = x1 / (float)u_minify(src->width0,  level);
+      out[1] = y1 / (float)u_minify(src->height0, level);
+      out[2] = x2 / (float)u_minify(src->width0,  level);
+      out[3] = y2 / (float)u_minify(src->height0, level);
    }
    else
    {
@@ -458,14 +448,14 @@ static void set_texcoords_in_vertices(const float coord[4],
 
 static void blitter_set_texcoords_2d(struct blitter_context_priv *ctx,
                                      struct pipe_resource *src,
-                                     struct pipe_subresource subsrc,
+                                     unsigned level,
                                      unsigned x1, unsigned y1,
                                      unsigned x2, unsigned y2)
 {
    unsigned i;
    float coord[4];
 
-   get_texcoords(src, subsrc, x1, y1, x2, y2, TRUE, coord);
+   get_texcoords(src, level, x1, y1, x2, y2, TRUE, coord);
    set_texcoords_in_vertices(coord, &ctx->vertices[0][1][0], 8);
 
    for (i = 0; i < 4; i++) {
@@ -476,15 +466,15 @@ static void blitter_set_texcoords_2d(struct blitter_context_priv *ctx,
 
 static void blitter_set_texcoords_3d(struct blitter_context_priv *ctx,
                                      struct pipe_resource *src,
-                                     struct pipe_subresource subsrc,
+                                     unsigned level,
                                      unsigned zslice,
                                      unsigned x1, unsigned y1,
                                      unsigned x2, unsigned y2)
 {
    int i;
-   float r = zslice / (float)u_minify(src->depth0, subsrc.level);
+   float r = zslice / (float)u_minify(src->depth0, level);
 
-   blitter_set_texcoords_2d(ctx, src, subsrc, x1, y1, x2, y2);
+   blitter_set_texcoords_2d(ctx, src, level, x1, y1, x2, y2);
 
    for (i = 0; i < 4; i++)
       ctx->vertices[i][1][2] = r; /*r*/
@@ -492,7 +482,7 @@ static void blitter_set_texcoords_3d(struct blitter_context_priv *ctx,
 
 static void blitter_set_texcoords_cube(struct blitter_context_priv *ctx,
                                        struct pipe_resource *src,
-                                       struct pipe_subresource subsrc,
+                                       unsigned level, unsigned face,
                                        unsigned x1, unsigned y1,
                                        unsigned x2, unsigned y2)
 {
@@ -500,10 +490,10 @@ static void blitter_set_texcoords_cube(struct blitter_context_priv *ctx,
    float coord[4];
    float st[4][2];
 
-   get_texcoords(src, subsrc, x1, y1, x2, y2, TRUE, coord);
+   get_texcoords(src, level, x1, y1, x2, y2, TRUE, coord);
    set_texcoords_in_vertices(coord, &st[0][0], 2);
 
-   util_map_texcoords2d_onto_cubemap(subsrc.face,
+   util_map_texcoords2d_onto_cubemap(face,
                                      /* pointer, stride in floats */
                                      &st[0][0], 2,
                                      &ctx->vertices[0][1][0], 8);
@@ -522,10 +512,13 @@ static void blitter_set_dst_dimensions(struct blitter_context_priv *ctx,
 static void blitter_draw_quad(struct blitter_context_priv *ctx)
 {
    struct pipe_context *pipe = ctx->base.pipe;
+   struct pipe_box box;
 
    /* write vertices and draw them */
-   pipe_buffer_write(pipe, ctx->vbuf,
-                     0, sizeof(ctx->vertices), ctx->vertices);
+   u_box_1d(0, sizeof(ctx->vertices), &box);
+   pipe->transfer_inline_write(pipe, ctx->vbuf, 0,
+                               PIPE_TRANSFER_WRITE | PIPE_TRANSFER_DISCARD,
+                               &box, ctx->vertices, sizeof(ctx->vertices), 0);
 
    util_draw_vertex_buffer(pipe, ctx->vbuf, 0, PIPE_PRIM_TRIANGLE_FAN,
                            4,  /* verts */
@@ -566,7 +559,9 @@ void *blitter_get_fs_col(struct blitter_context_priv *ctx, unsigned num_cbufs)
 
    if (!ctx->fs_col[num_cbufs])
       ctx->fs_col[num_cbufs] =
-         util_make_fragment_clonecolor_shader(pipe, num_cbufs);
+         util_make_fragment_cloneinput_shader(pipe, num_cbufs,
+                                              TGSI_SEMANTIC_GENERIC,
+                                              TGSI_INTERPOLATE_LINEAR);
 
    return ctx->fs_col[num_cbufs];
 }
@@ -697,7 +692,7 @@ void util_blitter_clear(struct blitter_context *blitter,
    pipe->bind_rasterizer_state(pipe, ctx->rs_state);
    pipe->bind_vertex_elements_state(pipe, ctx->velem_state);
    pipe->bind_fs_state(pipe, blitter_get_fs_col(ctx, num_cbufs));
-   pipe->bind_vs_state(pipe, ctx->vs_col);
+   pipe->bind_vs_state(pipe, ctx->vs);
 
    blitter_set_dst_dimensions(ctx, width, height);
    blitter->draw_rectangle(blitter, 0, 0, width, height, depth,
@@ -714,21 +709,22 @@ boolean is_overlap(unsigned sx1, unsigned sx2, unsigned sy1, unsigned sy2,
 
 void util_blitter_copy_region(struct blitter_context *blitter,
                               struct pipe_resource *dst,
-                              struct pipe_subresource subdst,
+                              unsigned dstlevel,
                               unsigned dstx, unsigned dsty, unsigned dstz,
                               struct pipe_resource *src,
-                              struct pipe_subresource subsrc,
-                              unsigned srcx, unsigned srcy, unsigned srcz,
-                              unsigned width, unsigned height,
+                              unsigned srclevel,
+                              const struct pipe_box *srcbox,
                               boolean ignore_stencil)
 {
    struct blitter_context_priv *ctx = (struct blitter_context_priv*)blitter;
    struct pipe_context *pipe = ctx->base.pipe;
    struct pipe_screen *screen = pipe->screen;
-   struct pipe_surface *dstsurf;
+   struct pipe_surface *dstsurf, surf_templ;
    struct pipe_framebuffer_state fb_state;
    struct pipe_sampler_view viewTempl, *view;
    unsigned bind;
+   unsigned width = srcbox->width;
+   unsigned height = srcbox->height;
    boolean is_stencil, is_depth;
    boolean normalized;
 
@@ -739,12 +735,15 @@ void util_blitter_copy_region(struct blitter_context *blitter,
 
    /* Sanity checks. */
    if (dst == src) {
-      assert(!is_overlap(srcx, srcx + width, srcy, srcy + height,
+      assert(!is_overlap(srcbox->x, srcbox->x + width, srcbox->y, srcbox->y + height,
                          dstx, dstx + width, dsty, dsty + height));
    } else {
-      assert(dst->format == src->format);
+      assert(util_is_format_compatible(util_format_description(src->format),
+                                       util_format_description(dst->format)));
    }
    assert(src->target < PIPE_MAX_TEXTURE_TYPES);
+   /* XXX should handle 3d regions */
+   assert(srcbox->depth == 1);
 
    /* Is this a ZS format? */
    is_depth = util_format_get_component_bits(src->format, UTIL_FORMAT_COLORSPACE_ZS, 0) != 0;
@@ -762,15 +761,18 @@ void util_blitter_copy_region(struct blitter_context *blitter,
                                     dst->nr_samples, bind, 0) ||
        !screen->is_format_supported(screen, src->format, src->target,
                                     src->nr_samples, PIPE_BIND_SAMPLER_VIEW, 0)) {
-      util_resource_copy_region(pipe, dst, subdst, dstx, dsty, dstz,
-                                src, subsrc, srcx, srcy, srcz, width, height);
+      util_resource_copy_region(pipe, dst, dstlevel, dstx, dsty, dstz,
+                                src, srclevel, srcbox);
       return;
    }
 
-   /* Get surfaces. */
-   dstsurf = screen->get_tex_surface(screen, dst,
-                                     subdst.face, subdst.level, dstz,
-                                     bind);
+   /* Get surface. */
+   memset(&surf_templ, 0, sizeof(surf_templ));
+   u_surface_default_template(&surf_templ, dst, bind);
+   surf_templ.u.tex.level = dstlevel;
+   surf_templ.u.tex.first_layer = dstz;
+   surf_templ.u.tex.last_layer = dstz;
+   dstsurf = pipe->create_surface(pipe, dst, &surf_templ);
 
    /* Check whether the states are properly saved. */
    blitter_check_saved_CSOs(ctx);
@@ -810,9 +812,9 @@ void util_blitter_copy_region(struct blitter_context *blitter,
 
    /* Set rasterizer state, shaders, and textures. */
    pipe->bind_rasterizer_state(pipe, ctx->rs_state);
-   pipe->bind_vs_state(pipe, ctx->vs_tex);
+   pipe->bind_vs_state(pipe, ctx->vs);
    pipe->bind_fragment_sampler_states(pipe, 1,
-                                      blitter_get_sampler_state(ctx, subsrc.level, normalized));
+                                      blitter_get_sampler_state(ctx, srclevel, normalized));
    pipe->bind_vertex_elements_state(pipe, ctx->velem_state);
    pipe->set_fragment_sampler_views(pipe, 1, &view);
    pipe->set_framebuffer_state(pipe, &fb_state);
@@ -827,8 +829,8 @@ void util_blitter_copy_region(struct blitter_context *blitter,
          {
             /* Set texture coordinates. */
             float coord[4];
-            get_texcoords(src, subsrc, srcx, srcy,
-                                     srcx+width, srcy+height, normalized, coord);
+            get_texcoords(src, srclevel, srcbox->x, srcbox->y,
+                          srcbox->x+width, srcbox->y+height, normalized, coord);
 
             /* Draw. */
             blitter->draw_rectangle(blitter, dstx, dsty, dstx+width, dsty+height, 0,
@@ -841,11 +843,13 @@ void util_blitter_copy_region(struct blitter_context *blitter,
       case PIPE_TEXTURE_CUBE:
          /* Set texture coordinates. */
          if (src->target == PIPE_TEXTURE_3D)
-            blitter_set_texcoords_3d(ctx, src, subsrc, srcz,
-                                     srcx, srcy, srcx+width, srcy+height);
+            blitter_set_texcoords_3d(ctx, src, srclevel, srcbox->z,
+                                     srcbox->x, srcbox->y,
+                                     srcbox->x + width, srcbox->y + height);
          else
-            blitter_set_texcoords_cube(ctx, src, subsrc,
-                                       srcx, srcy, srcx+width, srcy+height);
+            blitter_set_texcoords_cube(ctx, src, srclevel, srcbox->z,
+                                       srcbox->x, srcbox->y,
+                                       srcbox->x + width, srcbox->y + height);
 
          /* Draw. */
          blitter_set_rectangle(ctx, dstx, dsty, dstx+width, dsty+height, 0);
@@ -887,7 +891,7 @@ void util_blitter_clear_render_target(struct blitter_context *blitter,
    pipe->bind_depth_stencil_alpha_state(pipe, ctx->dsa_keep_depth_stencil);
    pipe->bind_rasterizer_state(pipe, ctx->rs_state);
    pipe->bind_fs_state(pipe, blitter_get_fs_col(ctx, 1));
-   pipe->bind_vs_state(pipe, ctx->vs_col);
+   pipe->bind_vs_state(pipe, ctx->vs);
    pipe->bind_vertex_elements_state(pipe, ctx->velem_state);
 
    /* set a framebuffer state */
@@ -947,7 +951,7 @@ void util_blitter_clear_depth_stencil(struct blitter_context *blitter,
 
    pipe->bind_rasterizer_state(pipe, ctx->rs_state);
    pipe->bind_fs_state(pipe, blitter_get_fs_col(ctx, 0));
-   pipe->bind_vs_state(pipe, ctx->vs_col);
+   pipe->bind_vs_state(pipe, ctx->vs);
    pipe->bind_vertex_elements_state(pipe, ctx->velem_state);
 
    /* set a framebuffer state */
@@ -988,7 +992,7 @@ void util_blitter_custom_depth_stencil(struct blitter_context *blitter,
 
    pipe->bind_rasterizer_state(pipe, ctx->rs_state);
    pipe->bind_fs_state(pipe, blitter_get_fs_col(ctx, 0));
-   pipe->bind_vs_state(pipe, ctx->vs_col);
+   pipe->bind_vs_state(pipe, ctx->vs);
    pipe->bind_vertex_elements_state(pipe, ctx->velem_state);
 
    /* set a framebuffer state */
diff --git a/src/gallium/auxiliary/util/u_blitter.h b/src/gallium/auxiliary/util/u_blitter.h
index f9f96f25c7..c5660cf2d0 100644
--- a/src/gallium/auxiliary/util/u_blitter.h
+++ b/src/gallium/auxiliary/util/u_blitter.h
@@ -164,12 +164,11 @@ void util_blitter_clear(struct blitter_context *blitter,
  */
 void util_blitter_copy_region(struct blitter_context *blitter,
                               struct pipe_resource *dst,
-                              struct pipe_subresource subdst,
+                              unsigned dstlevel,
                               unsigned dstx, unsigned dsty, unsigned dstz,
                               struct pipe_resource *src,
-                              struct pipe_subresource subsrc,
-                              unsigned srcx, unsigned srcy, unsigned srcz,
-                              unsigned width, unsigned height,
+                              unsigned srclevel,
+                              const struct pipe_box *srcbox,
                               boolean ignore_stencil);
 
 /**
diff --git a/src/gallium/auxiliary/util/u_box.h b/src/gallium/auxiliary/util/u_box.h
index e9c71743fc..0b28d0f12c 100644
--- a/src/gallium/auxiliary/util/u_box.h
+++ b/src/gallium/auxiliary/util/u_box.h
@@ -60,7 +60,6 @@ void u_box_2d_zslice( unsigned x,
    box->depth = 1;
 }
 
-
 static INLINE
 void u_box_3d( unsigned x,
 	       unsigned y,
@@ -78,15 +77,4 @@ void u_box_3d( unsigned x,
    box->depth = d;
 }
 
-
-static INLINE
-struct pipe_subresource u_subresource( unsigned face,
-				       unsigned level )
-{
-   struct pipe_subresource subresource;
-   subresource.face = face;
-   subresource.level = level;
-   return subresource;
-}
-
 #endif
diff --git a/src/gallium/auxiliary/util/u_debug.c b/src/gallium/auxiliary/util/u_debug.c
index 504e6d2a18..2ad2f95b13 100644
--- a/src/gallium/auxiliary/util/u_debug.c
+++ b/src/gallium/auxiliary/util/u_debug.c
@@ -40,7 +40,8 @@
 #include "util/u_string.h" 
 #include "util/u_math.h" 
 #include "util/u_tile.h" 
-#include "util/u_prim.h" 
+#include "util/u_prim.h"
+#include "util/u_surface.h"
 
 #include <limits.h> /* CHAR_BIT */
 
@@ -453,9 +454,10 @@ void debug_dump_image(const char *prefix,
 #endif
 }
 
+/* FIXME: dump resources, not surfaces... */
 void debug_dump_surface(struct pipe_context *pipe,
-			const char *prefix,
-                        struct pipe_surface *surface)     
+                        const char *prefix,
+                        struct pipe_surface *surface)
 {
    struct pipe_resource *texture;
    struct pipe_transfer *transfer;
@@ -472,23 +474,23 @@ void debug_dump_surface(struct pipe_context *pipe,
     */
    texture = surface->texture;
 
-   transfer = pipe_get_transfer(pipe, texture, surface->face,
-				     surface->level, surface->zslice,
-				     PIPE_TRANSFER_READ, 0, 0, surface->width,
-				     surface->height);
-   
+   transfer = pipe_get_transfer(pipe, texture, surface->u.tex.level,
+                                surface->u.tex.first_layer,
+                                PIPE_TRANSFER_READ,
+                                0, 0, surface->width, surface->height);
+
    data = pipe->transfer_map(pipe, transfer);
    if(!data)
       goto error;
-   
-   debug_dump_image(prefix, 
+
+   debug_dump_image(prefix,
                     texture->format,
-                    util_format_get_blocksize(texture->format), 
+                    util_format_get_blocksize(texture->format),
                     util_format_get_nblocksx(texture->format, surface->width),
                     util_format_get_nblocksy(texture->format, surface->height),
                     transfer->stride,
                     data);
-   
+
    pipe->transfer_unmap(pipe, transfer);
 error:
    pipe->transfer_destroy(pipe, transfer);
@@ -499,20 +501,18 @@ void debug_dump_texture(struct pipe_context *pipe,
                         const char *prefix,
                         struct pipe_resource *texture)
 {
-   struct pipe_surface *surface;
-   struct pipe_screen *screen;
+   struct pipe_surface *surface, surf_tmpl;
 
    if (!texture)
       return;
 
-   screen = texture->screen;
-
-   /* XXX for now, just dump image for face=0, level=0 */
-   surface = screen->get_tex_surface(screen, texture, 0, 0, 0,
-                                     PIPE_BIND_SAMPLER_VIEW);
+   /* XXX for now, just dump image for layer=0, level=0 */
+   memset(&surf_tmpl, 0, sizeof(surf_tmpl));
+   u_surface_default_template(&surf_tmpl, texture, 0 /* no bind flag - not a surface */);
+   surface = pipe->create_surface(pipe, texture, &surf_tmpl);
    if (surface) {
       debug_dump_surface(pipe, prefix, surface);
-      screen->tex_surface_destroy(surface);
+      pipe->surface_destroy(pipe, surface);
    }
 }
 
@@ -550,17 +550,16 @@ struct bmp_rgb_quad {
 
 void
 debug_dump_surface_bmp(struct pipe_context *pipe,
-		       const char *filename,
+                       const char *filename,
                        struct pipe_surface *surface)
 {
 #ifndef PIPE_SUBSYSTEM_WINDOWS_MINIPORT
    struct pipe_transfer *transfer;
    struct pipe_resource *texture = surface->texture;
 
-   transfer = pipe_get_transfer(pipe, texture, surface->face,
-				surface->level, surface->zslice,
-				PIPE_TRANSFER_READ, 0, 0, surface->width,
-				surface->height);
+   transfer = pipe_get_transfer(pipe, texture, surface->u.tex.level,
+                                surface->u.tex.first_layer, PIPE_TRANSFER_READ,
+                                0, 0, surface->width, surface->height);
 
    debug_dump_transfer_bmp(pipe, filename, transfer);
 
diff --git a/src/gallium/auxiliary/util/u_debug_describe.c b/src/gallium/auxiliary/util/u_debug_describe.c
index 1c90ff3106..7ed8ee608a 100644
--- a/src/gallium/auxiliary/util/u_debug_describe.c
+++ b/src/gallium/auxiliary/util/u_debug_describe.c
@@ -69,7 +69,7 @@ debug_describe_surface(char* buf, const struct pipe_surface *ptr)
 {
    char res[128];
    debug_describe_resource(res, ptr->texture);
-   util_sprintf(buf, "pipe_surface<%s,%u,%u,%u>", res, ptr->face, ptr->level, ptr->zslice);
+   util_sprintf(buf, "pipe_surface<%s,%u,%u,%u>", res, ptr->u.tex.level, ptr->u.tex.first_layer, ptr->u.tex.last_layer);
 }
 
 void
diff --git a/src/gallium/auxiliary/util/u_debug_stack.c b/src/gallium/auxiliary/util/u_debug_stack.c
index 528a1c394b..24e039fd22 100644
--- a/src/gallium/auxiliary/util/u_debug_stack.c
+++ b/src/gallium/auxiliary/util/u_debug_stack.c
@@ -48,7 +48,10 @@ debug_backtrace_capture(struct debug_stack_frame *backtrace,
    if(!nr_frames)
       return;
 
-#if defined(PIPE_CC_GCC)
+#if defined(PIPE_CC_GCC) && defined(PIPE_ARCH_X86)
+   __asm__ __volatile__("mov (%%ebp),%0": "=r" (frame_pointer));
+   frame_pointer = (const void **)frame_pointer[0];
+#elif defined(PIPE_CC_GCC)
    frame_pointer = ((const void **)__builtin_frame_address(1));
 #elif defined(PIPE_CC_MSVC) && defined(PIPE_ARCH_X86)
    __asm {
diff --git a/src/gallium/auxiliary/util/u_debug_symbol.c b/src/gallium/auxiliary/util/u_debug_symbol.c
index 332952af88..44d437747a 100644
--- a/src/gallium/auxiliary/util/u_debug_symbol.c
+++ b/src/gallium/auxiliary/util/u_debug_symbol.c
@@ -40,7 +40,7 @@
 #include "u_debug_symbol.h"
 #include "u_hash_table.h"
 
-#if defined(PIPE_SUBSYSTEM_WINDOWS_USER) && defined(PIPE_ARCH_X86)
+#if defined(PIPE_CC_MSVC) && defined(PIPE_ARCH_X86)
    
 #include <windows.h>
 #include <stddef.h>
@@ -165,7 +165,7 @@ debug_symbol_name_glibc(const void *addr, char* buf, unsigned size)
 void
 debug_symbol_name(const void *addr, char* buf, unsigned size)
 {
-#if defined(PIPE_SUBSYSTEM_WINDOWS_USER) && defined(PIPE_ARCH_X86)
+#if defined(PIPE_CC_MSVC) && defined(PIPE_ARCH_X86)
    debug_symbol_name_imagehlp(addr, buf, size);
    if(buf[0])
       return;
diff --git a/src/gallium/auxiliary/util/u_dirty_flags.h b/src/gallium/auxiliary/util/u_dirty_flags.h
index 7e1be45ad5..40539f0b0e 100644
--- a/src/gallium/auxiliary/util/u_dirty_flags.h
+++ b/src/gallium/auxiliary/util/u_dirty_flags.h
@@ -24,5 +24,9 @@
 #define U_NEW_VERTEX_BUFFER         0x10000
 #define U_NEW_QUERY                 0x20000
 #define U_NEW_DEPTH_STENCIL         0x40000
+#define U_NEW_GS                    0x80000
+#define U_NEW_GS_CONSTANTS          0x100000
+#define U_NEW_GS_SAMPLER_VIEW       0x200000
+#define U_NEW_GS_SAMPLER_STATES     0x400000
 
 #endif
diff --git a/src/gallium/auxiliary/util/u_dirty_surfaces.h b/src/gallium/auxiliary/util/u_dirty_surfaces.h
index fd1bbe5ffd..f3618d9be7 100644
--- a/src/gallium/auxiliary/util/u_dirty_surfaces.h
+++ b/src/gallium/auxiliary/util/u_dirty_surfaces.h
@@ -77,7 +77,7 @@ util_dirty_surfaces_use_levels_for_sampling(struct pipe_context *pipe, struct ut
       struct util_dirty_surface *ds = LIST_ENTRY(struct util_dirty_surface, p, dirty_list);
       next = p->next;
 
-      if(ds->base.level >= first && ds->base.level <= last)
+      if(ds->base.u.tex.level >= first && ds->base.u.tex.level <= last)
 	 flush(pipe, &ds->base);
    }
 }
@@ -86,7 +86,8 @@ static INLINE void
 util_dirty_surfaces_use_for_sampling_with(struct pipe_context *pipe, struct util_dirty_surfaces *dss, struct pipe_sampler_view *psv, struct pipe_sampler_state *pss, util_dirty_surface_flush_t flush)
 {
    if(!LIST_IS_EMPTY(&dss->dirty_list))
-      util_dirty_surfaces_use_levels_for_sampling(pipe, dss, (unsigned)pss->min_lod + psv->first_level, MIN2((unsigned)ceilf(pss->max_lod) + psv->first_level, psv->last_level), flush);
+      util_dirty_surfaces_use_levels_for_sampling(pipe, dss, (unsigned)pss->min_lod + psv->u.tex.first_level,
+						  MIN2((unsigned)ceilf(pss->max_lod) + psv->u.tex.first_level, psv->u.tex.last_level), flush);
 }
 
 static INLINE void
diff --git a/src/gallium/auxiliary/util/u_dump_state.c b/src/gallium/auxiliary/util/u_dump_state.c
index cda5b8ba51..b471d59eeb 100644
--- a/src/gallium/auxiliary/util/u_dump_state.c
+++ b/src/gallium/auxiliary/util/u_dump_state.c
@@ -279,6 +279,10 @@ util_dump_template(struct os_stream *stream, const struct pipe_resource *templat
    util_dump_uint(stream, templat->depth0);
    util_dump_member_end(stream);
 
+   util_dump_member_begin(stream, "array_size");
+   util_dump_uint(stream, templat->array_size);
+   util_dump_member_end(stream);
+
    util_dump_member(stream, uint, templat, last_level);
    util_dump_member(stream, uint, templat, usage);
    util_dump_member(stream, uint, templat, bind);
@@ -633,14 +637,12 @@ util_dump_surface(struct os_stream *stream, const struct pipe_surface *state)
    util_dump_member(stream, uint, state, width);
    util_dump_member(stream, uint, state, height);
 
-   util_dump_member(stream, uint, state, layout);
-   util_dump_member(stream, uint, state, offset);
    util_dump_member(stream, uint, state, usage);
 
    util_dump_member(stream, ptr, state, texture);
-   util_dump_member(stream, uint, state, face);
-   util_dump_member(stream, uint, state, level);
-   util_dump_member(stream, uint, state, zslice);
+   util_dump_member(stream, uint, state, u.tex.level);
+   util_dump_member(stream, uint, state, u.tex.first_layer);
+   util_dump_member(stream, uint, state, u.tex.last_layer);
 
    util_dump_struct_end(stream);
 }
@@ -660,7 +662,7 @@ util_dump_transfer(struct os_stream *stream, const struct pipe_transfer *state)
    /*util_dump_member(stream, uint, state, box);*/
 
    util_dump_member(stream, uint, state, stride);
-   util_dump_member(stream, uint, state, slice_stride);
+   util_dump_member(stream, uint, state, layer_stride);
 
    /*util_dump_member(stream, ptr, state, data);*/
 
diff --git a/src/gallium/auxiliary/util/u_gen_mipmap.c b/src/gallium/auxiliary/util/u_gen_mipmap.c
index 6a931a9581..d22ae8b375 100644
--- a/src/gallium/auxiliary/util/u_gen_mipmap.c
+++ b/src/gallium/auxiliary/util/u_gen_mipmap.c
@@ -48,6 +48,8 @@
 #include "util/u_simple_shaders.h"
 #include "util/u_math.h"
 #include "util/u_texture.h"
+#include "util/u_half.h"
+#include "util/u_surface.h"
 
 #include "cso_cache/cso_context.h"
 
@@ -65,7 +67,7 @@ struct gen_mipmap_state
    struct pipe_vertex_element velem[2];
 
    void *vs;
-   void *fs2d, *fsCube;
+   void *fs1d, *fs2d, *fs3d, *fsCube;
 
    struct pipe_resource *vbuf;  /**< quad vertices */
    unsigned vbuf_slot;
@@ -89,24 +91,7 @@ enum dtype
 };
 
 
-typedef ushort half_float;
-
-
-static half_float
-float_to_half(float f)
-{
-   /* XXX fix this */
-   return 0;
-}
-
-static float
-half_to_float(half_float h)
-{
-   /* XXX fix this */
-   return 0.0f;
-}
-
-
+typedef uint16_t half_float;
 
 
 /**
@@ -145,7 +130,7 @@ half_to_float(half_float h)
                                 rowC[j][e], rowC[k][e], \
                                 rowD[j][e], rowD[k][e]); \
    } while(0)
-   
+
 #define FILTER_F_3D(e) \
    do { \
       dst[i][e] = (rowA[j][e] + rowA[k][e] \
@@ -156,15 +141,15 @@ half_to_float(half_float h)
 
 #define FILTER_HF_3D(e) \
    do { \
-      const float aj = half_to_float(rowA[j][e]); \
-      const float ak = half_to_float(rowA[k][e]); \
-      const float bj = half_to_float(rowB[j][e]); \
-      const float bk = half_to_float(rowB[k][e]); \
-      const float cj = half_to_float(rowC[j][e]); \
-      const float ck = half_to_float(rowC[k][e]); \
-      const float dj = half_to_float(rowD[j][e]); \
-      const float dk = half_to_float(rowD[k][e]); \
-      dst[i][e] = float_to_half((aj + ak + bj + bk + cj + ck + dj + dk) \
+      const float aj = util_half_to_float(rowA[j][e]); \
+      const float ak = util_half_to_float(rowA[k][e]); \
+      const float bj = util_half_to_float(rowB[j][e]); \
+      const float bk = util_half_to_float(rowB[k][e]); \
+      const float cj = util_half_to_float(rowC[j][e]); \
+      const float ck = util_half_to_float(rowC[k][e]); \
+      const float dj = util_half_to_float(rowD[j][e]); \
+      const float dk = util_half_to_float(rowD[k][e]); \
+      dst[i][e] = util_float_to_half((aj + ak + bj + bk + cj + ck + dj + dk) \
                                       * 0.125F); \
    } while(0)
 /*@}*/
@@ -343,8 +328,7 @@ do_row(enum dtype datatype, uint comps, int srcWidth,
       }
    }
 
-#if 0
-   else if (datatype == HALF_DTYPE_FLOAT && comps == 4) {
+   else if (datatype == DTYPE_HALF_FLOAT && comps == 4) {
       uint i, j, k, comp;
       const half_float(*rowA)[4] = (const half_float(*)[4]) srcRowA;
       const half_float(*rowB)[4] = (const half_float(*)[4]) srcRowB;
@@ -353,11 +337,11 @@ do_row(enum dtype datatype, uint comps, int srcWidth,
            i++, j += colStride, k += colStride) {
          for (comp = 0; comp < 4; comp++) {
             float aj, ak, bj, bk;
-            aj = half_to_float(rowA[j][comp]);
-            ak = half_to_float(rowA[k][comp]);
-            bj = half_to_float(rowB[j][comp]);
-            bk = half_to_float(rowB[k][comp]);
-            dst[i][comp] = float_to_half((aj + ak + bj + bk) * 0.25F);
+            aj = util_half_to_float(rowA[j][comp]);
+            ak = util_half_to_float(rowA[k][comp]);
+            bj = util_half_to_float(rowB[j][comp]);
+            bk = util_half_to_float(rowB[k][comp]);
+            dst[i][comp] = util_float_to_half((aj + ak + bj + bk) * 0.25F);
          }
       }
    }
@@ -370,11 +354,11 @@ do_row(enum dtype datatype, uint comps, int srcWidth,
            i++, j += colStride, k += colStride) {
          for (comp = 0; comp < 3; comp++) {
             float aj, ak, bj, bk;
-            aj = half_to_float(rowA[j][comp]);
-            ak = half_to_float(rowA[k][comp]);
-            bj = half_to_float(rowB[j][comp]);
-            bk = half_to_float(rowB[k][comp]);
-            dst[i][comp] = float_to_half((aj + ak + bj + bk) * 0.25F);
+            aj = util_half_to_float(rowA[j][comp]);
+            ak = util_half_to_float(rowA[k][comp]);
+            bj = util_half_to_float(rowB[j][comp]);
+            bk = util_half_to_float(rowB[k][comp]);
+            dst[i][comp] = util_float_to_half((aj + ak + bj + bk) * 0.25F);
          }
       }
    }
@@ -387,11 +371,11 @@ do_row(enum dtype datatype, uint comps, int srcWidth,
            i++, j += colStride, k += colStride) {
          for (comp = 0; comp < 2; comp++) {
             float aj, ak, bj, bk;
-            aj = half_to_float(rowA[j][comp]);
-            ak = half_to_float(rowA[k][comp]);
-            bj = half_to_float(rowB[j][comp]);
-            bk = half_to_float(rowB[k][comp]);
-            dst[i][comp] = float_to_half((aj + ak + bj + bk) * 0.25F);
+            aj = util_half_to_float(rowA[j][comp]);
+            ak = util_half_to_float(rowA[k][comp]);
+            bj = util_half_to_float(rowB[j][comp]);
+            bk = util_half_to_float(rowB[k][comp]);
+            dst[i][comp] = util_float_to_half((aj + ak + bj + bk) * 0.25F);
          }
       }
    }
@@ -403,14 +387,13 @@ do_row(enum dtype datatype, uint comps, int srcWidth,
       for (i = j = 0, k = k0; i < (uint) dstWidth;
            i++, j += colStride, k += colStride) {
          float aj, ak, bj, bk;
-         aj = half_to_float(rowA[j]);
-         ak = half_to_float(rowA[k]);
-         bj = half_to_float(rowB[j]);
-         bk = half_to_float(rowB[k]);
-         dst[i] = float_to_half((aj + ak + bj + bk) * 0.25F);
+         aj = util_half_to_float(rowA[j]);
+         ak = util_half_to_float(rowA[k]);
+         bj = util_half_to_float(rowB[j]);
+         bk = util_half_to_float(rowB[k]);
+         dst[i] = util_float_to_half((aj + ak + bj + bk) * 0.25F);
       }
    }
-#endif
 
    else if (datatype == DTYPE_UINT && comps == 1) {
       uint i, j, k;
@@ -1036,32 +1019,34 @@ reduce_2d(enum pipe_format pformat,
 static void
 reduce_3d(enum pipe_format pformat,
           int srcWidth, int srcHeight, int srcDepth,
-          int srcRowStride, const ubyte *srcPtr,
+          int srcRowStride, int srcImageStride, const ubyte *srcPtr,
           int dstWidth, int dstHeight, int dstDepth,
-          int dstRowStride, ubyte *dstPtr)
+          int dstRowStride, int dstImageStride, ubyte *dstPtr)
 {
    const int bpt = util_format_get_blocksize(pformat);
-   const int border = 0;
    int img, row;
-   int bytesPerSrcImage, bytesPerDstImage;
-   int bytesPerSrcRow, bytesPerDstRow;
    int srcImageOffset, srcRowOffset;
    enum dtype datatype;
    uint comps;
 
    format_to_type_comps(pformat, &datatype, &comps);
 
-   bytesPerSrcImage = srcWidth * srcHeight * bpt;
-   bytesPerDstImage = dstWidth * dstHeight * bpt;
+   /* XXX I think we should rather assert those strides */
+   if (!srcImageStride)
+      srcImageStride = srcWidth * srcHeight * bpt;
+   if (!dstImageStride)
+      dstImageStride = dstWidth * dstHeight * bpt;
 
-   bytesPerSrcRow = srcWidth * bpt;
-   bytesPerDstRow = dstWidth * bpt;
+   if (!srcRowStride)
+      srcRowStride = srcWidth * bpt;
+   if (!dstRowStride)
+      dstRowStride = dstWidth * bpt;
 
    /* Offset between adjacent src images to be averaged together */
-   srcImageOffset = (srcDepth == dstDepth) ? 0 : bytesPerSrcImage;
+   srcImageOffset = (srcDepth == dstDepth) ? 0 : srcImageStride;
 
    /* Offset between adjacent src rows to be averaged together */
-   srcRowOffset = (srcHeight == dstHeight) ? 0 : srcWidth * bpt;
+   srcRowOffset = (srcHeight == dstHeight) ? 0 : srcRowStride;
 
    /*
     * Need to average together up to 8 src pixels for each dest pixel.
@@ -1077,16 +1062,13 @@ reduce_3d(enum pipe_format pformat,
    */
 
    for (img = 0; img < dstDepth; img++) {
-      /* first source image pointer, skipping border */
+      /* first source image pointer */
       const ubyte *imgSrcA = srcPtr
-         + (bytesPerSrcImage + bytesPerSrcRow + border) * bpt * border
-         + img * (bytesPerSrcImage + srcImageOffset);
-      /* second source image pointer, skipping border */
+         + img * (srcImageStride + srcImageOffset);
+      /* second source image pointer */
       const ubyte *imgSrcB = imgSrcA + srcImageOffset;
-      /* address of the dest image, skipping border */
-      ubyte *imgDst = dstPtr
-         + (bytesPerDstImage + bytesPerDstRow + border) * bpt * border
-         + img * bytesPerDstImage;
+      /* address of the dest image */
+      ubyte *imgDst = dstPtr + img * dstImageStride;
 
       /* setup the four source row pointers and the dest row pointer */
       const ubyte *srcImgARowA = imgSrcA;
@@ -1102,11 +1084,11 @@ reduce_3d(enum pipe_format pformat,
                    dstWidth, dstImgRow);
 
          /* advance to next rows */
-         srcImgARowA += bytesPerSrcRow + srcRowOffset;
-         srcImgARowB += bytesPerSrcRow + srcRowOffset;
-         srcImgBRowA += bytesPerSrcRow + srcRowOffset;
-         srcImgBRowB += bytesPerSrcRow + srcRowOffset;
-         dstImgRow += bytesPerDstRow;
+         srcImgARowA += srcRowStride + srcRowOffset;
+         srcImgARowB += srcRowStride + srcRowOffset;
+         srcImgBRowA += srcRowStride + srcRowOffset;
+         srcImgBRowB += srcRowStride + srcRowOffset;
+         dstImgRow += dstImageStride;
       }
    }
 }
@@ -1117,25 +1099,24 @@ reduce_3d(enum pipe_format pformat,
 static void
 make_1d_mipmap(struct gen_mipmap_state *ctx,
                struct pipe_resource *pt,
-               uint face, uint baseLevel, uint lastLevel)
+               uint layer, uint baseLevel, uint lastLevel)
 {
    struct pipe_context *pipe = ctx->pipe;
-   const uint zslice = 0;
    uint dstLevel;
 
    for (dstLevel = baseLevel + 1; dstLevel <= lastLevel; dstLevel++) {
       const uint srcLevel = dstLevel - 1;
       struct pipe_transfer *srcTrans, *dstTrans;
       void *srcMap, *dstMap;
-      
-      srcTrans = pipe_get_transfer(pipe, pt, face, srcLevel, zslice,
-                                          PIPE_TRANSFER_READ, 0, 0,
-                                          u_minify(pt->width0, srcLevel),
-                                          u_minify(pt->height0, srcLevel));
-      dstTrans = pipe_get_transfer(pipe, pt, face, dstLevel, zslice,
-                                          PIPE_TRANSFER_WRITE, 0, 0,
-                                          u_minify(pt->width0, dstLevel),
-                                          u_minify(pt->height0, dstLevel));
+
+      srcTrans = pipe_get_transfer(pipe, pt, srcLevel, layer,
+                                   PIPE_TRANSFER_READ, 0, 0,
+                                   u_minify(pt->width0, srcLevel),
+                                   u_minify(pt->height0, srcLevel));
+      dstTrans = pipe_get_transfer(pipe, pt, dstLevel, layer,
+                                   PIPE_TRANSFER_WRITE, 0, 0,
+                                   u_minify(pt->width0, dstLevel),
+                                   u_minify(pt->height0, dstLevel));
 
       srcMap = (ubyte *) pipe->transfer_map(pipe, srcTrans);
       dstMap = (ubyte *) pipe->transfer_map(pipe, dstTrans);
@@ -1156,12 +1137,11 @@ make_1d_mipmap(struct gen_mipmap_state *ctx,
 static void
 make_2d_mipmap(struct gen_mipmap_state *ctx,
                struct pipe_resource *pt,
-               uint face, uint baseLevel, uint lastLevel)
+               uint layer, uint baseLevel, uint lastLevel)
 {
    struct pipe_context *pipe = ctx->pipe;
-   const uint zslice = 0;
    uint dstLevel;
-   
+
    assert(util_format_get_blockwidth(pt->format) == 1);
    assert(util_format_get_blockheight(pt->format) == 1);
 
@@ -1169,15 +1149,15 @@ make_2d_mipmap(struct gen_mipmap_state *ctx,
       const uint srcLevel = dstLevel - 1;
       struct pipe_transfer *srcTrans, *dstTrans;
       ubyte *srcMap, *dstMap;
-      
-      srcTrans = pipe_get_transfer(pipe, pt, face, srcLevel, zslice,
-				   PIPE_TRANSFER_READ, 0, 0,
-				   u_minify(pt->width0, srcLevel),
-				   u_minify(pt->height0, srcLevel));
-      dstTrans = pipe_get_transfer(pipe, pt, face, dstLevel, zslice,
-				   PIPE_TRANSFER_WRITE, 0, 0,
-				   u_minify(pt->width0, dstLevel),
-				   u_minify(pt->height0, dstLevel));
+
+      srcTrans = pipe_get_transfer(pipe, pt, srcLevel, layer,
+                                   PIPE_TRANSFER_READ, 0, 0,
+                                   u_minify(pt->width0, srcLevel),
+                                   u_minify(pt->height0, srcLevel));
+      dstTrans = pipe_get_transfer(pipe, pt, dstLevel, layer,
+                                   PIPE_TRANSFER_WRITE, 0, 0,
+                                   u_minify(pt->width0, dstLevel),
+                                   u_minify(pt->height0, dstLevel));
 
       srcMap = (ubyte *) pipe->transfer_map(pipe, srcTrans);
       dstMap = (ubyte *) pipe->transfer_map(pipe, dstTrans);
@@ -1197,41 +1177,49 @@ make_2d_mipmap(struct gen_mipmap_state *ctx,
 }
 
 
+/* XXX looks a bit more like it could work now but need to test */
 static void
 make_3d_mipmap(struct gen_mipmap_state *ctx,
                struct pipe_resource *pt,
                uint face, uint baseLevel, uint lastLevel)
 {
-#if 0
    struct pipe_context *pipe = ctx->pipe;
-   struct pipe_screen *screen = pipe->screen;
-   uint dstLevel, zslice = 0;
+   uint dstLevel;
+   struct pipe_box src_box, dst_box;
 
    assert(util_format_get_blockwidth(pt->format) == 1);
    assert(util_format_get_blockheight(pt->format) == 1);
 
+   src_box.x = src_box.y = src_box.z = 0;
+   dst_box.x = dst_box.y = dst_box.z = 0;
+
    for (dstLevel = baseLevel + 1; dstLevel <= lastLevel; dstLevel++) {
       const uint srcLevel = dstLevel - 1;
       struct pipe_transfer *srcTrans, *dstTrans;
       ubyte *srcMap, *dstMap;
-      
-      srcTrans = pipe->get_transfer(pipe, pt, face, srcLevel, zslice,
-                                          PIPE_TRANSFER_READ, 0, 0,
-                                          u_minify(pt->width0, srcLevel),
-                                          u_minify(pt->height0, srcLevel));
-      dstTrans = pipe->get_transfer(pipe, pt, face, dstLevel, zslice,
-                                          PIPE_TRANSFER_WRITE, 0, 0,
-                                          u_minify(pt->width0, dstLevel),
-                                          u_minify(pt->height0, dstLevel));
+      struct pipe_box src_box, dst_box;
+      src_box.width = u_minify(pt->width0, srcLevel);
+      src_box.height = u_minify(pt->height0, srcLevel);
+      src_box.depth = u_minify(pt->depth0, srcLevel);
+      dst_box.width = u_minify(pt->width0, dstLevel);
+      dst_box.height = u_minify(pt->height0, dstLevel);
+      dst_box.depth = u_minify(pt->depth0, dstLevel);
+
+      srcTrans = pipe->get_transfer(pipe, pt, srcLevel,
+                                    PIPE_TRANSFER_READ,
+                                    &src_box);
+      dstTrans = pipe->get_transfer(pipe, pt, dstLevel,
+                                    PIPE_TRANSFER_WRITE,
+                                    &dst_box);
 
       srcMap = (ubyte *) pipe->transfer_map(pipe, srcTrans);
       dstMap = (ubyte *) pipe->transfer_map(pipe, dstTrans);
 
       reduce_3d(pt->format,
-                srcTrans->width, srcTrans->height,
-                srcTrans->stride, srcMap,
-                dstTrans->width, dstTrans->height,
-                dstTrans->stride, dstMap);
+                srcTrans->box.width, srcTrans->box.height, srcTrans->box.depth,
+                srcTrans->stride, srcTrans->layer_stride, srcMap,
+                dstTrans->box.width, dstTrans->box.height, dstTrans->box.depth,
+                dstTrans->stride, dstTrans->layer_stride, dstMap);
 
       pipe->transfer_unmap(pipe, srcTrans);
       pipe->transfer_unmap(pipe, dstTrans);
@@ -1239,28 +1227,25 @@ make_3d_mipmap(struct gen_mipmap_state *ctx,
       pipe->transfer_destroy(pipe, srcTrans);
       pipe->transfer_destroy(pipe, dstTrans);
    }
-#else
-   (void) reduce_3d;
-#endif
 }
 
 
 static void
 fallback_gen_mipmap(struct gen_mipmap_state *ctx,
                     struct pipe_resource *pt,
-                    uint face, uint baseLevel, uint lastLevel)
+                    uint layer, uint baseLevel, uint lastLevel)
 {
    switch (pt->target) {
    case PIPE_TEXTURE_1D:
-      make_1d_mipmap(ctx, pt, face, baseLevel, lastLevel);
+      make_1d_mipmap(ctx, pt, layer, baseLevel, lastLevel);
       break;
    case PIPE_TEXTURE_2D:
    case PIPE_TEXTURE_RECT:
    case PIPE_TEXTURE_CUBE:
-      make_2d_mipmap(ctx, pt, face, baseLevel, lastLevel);
+      make_2d_mipmap(ctx, pt, layer, baseLevel, lastLevel);
       break;
    case PIPE_TEXTURE_3D:
-      make_3d_mipmap(ctx, pt, face, baseLevel, lastLevel);
+      make_3d_mipmap(ctx, pt, layer, baseLevel, lastLevel);
       break;
    default:
       assert(0);
@@ -1328,8 +1313,12 @@ util_create_gen_mipmap(struct pipe_context *pipe,
    }
 
    /* fragment shader */
+   ctx->fs1d = util_make_fragment_tex_shader(pipe, TGSI_TEXTURE_1D,
+                                             TGSI_INTERPOLATE_LINEAR);
    ctx->fs2d = util_make_fragment_tex_shader(pipe, TGSI_TEXTURE_2D,
                                              TGSI_INTERPOLATE_LINEAR);
+   ctx->fs3d = util_make_fragment_tex_shader(pipe, TGSI_TEXTURE_3D,
+                                               TGSI_INTERPOLATE_LINEAR);
    ctx->fsCube = util_make_fragment_tex_shader(pipe, TGSI_TEXTURE_CUBE,
                                                TGSI_INTERPOLATE_LINEAR);
 
@@ -1371,7 +1360,7 @@ get_next_slot(struct gen_mipmap_state *ctx)
 static unsigned
 set_vertex_data(struct gen_mipmap_state *ctx,
                 enum pipe_texture_target tex_target,
-                uint face)
+                uint layer, float r)
 {
    unsigned offset;
 
@@ -1397,26 +1386,26 @@ set_vertex_data(struct gen_mipmap_state *ctx,
          {0.0f, 0.0f}, {1.0f, 0.0f}, {1.0f, 1.0f}, {0.0f, 1.0f}
       };
 
-      util_map_texcoords2d_onto_cubemap(face, &st[0][0], 2,
+      util_map_texcoords2d_onto_cubemap(layer, &st[0][0], 2,
                                         &ctx->vertices[0][1][0], 8);
    }
    else {
-      /* 1D/2D */
+      /* 1D/2D/3D */
       ctx->vertices[0][1][0] = 0.0f; /*s*/
       ctx->vertices[0][1][1] = 0.0f; /*t*/
-      ctx->vertices[0][1][2] = 0.0f; /*r*/
+      ctx->vertices[0][1][2] = r;    /*r*/
 
       ctx->vertices[1][1][0] = 1.0f;
       ctx->vertices[1][1][1] = 0.0f;
-      ctx->vertices[1][1][2] = 0.0f;
+      ctx->vertices[1][1][2] = r;
 
       ctx->vertices[2][1][0] = 1.0f;
       ctx->vertices[2][1][1] = 1.0f;
-      ctx->vertices[2][1][2] = 0.0f;
+      ctx->vertices[2][1][2] = r;
 
       ctx->vertices[3][1][0] = 0.0f;
       ctx->vertices[3][1][1] = 1.0f;
-      ctx->vertices[3][1][2] = 0.0f;
+      ctx->vertices[3][1][2] = r;
    }
 
    offset = get_next_slot( ctx );
@@ -1437,9 +1426,11 @@ util_destroy_gen_mipmap(struct gen_mipmap_state *ctx)
 {
    struct pipe_context *pipe = ctx->pipe;
 
-   pipe->delete_vs_state(pipe, ctx->vs);
-   pipe->delete_fs_state(pipe, ctx->fs2d);
    pipe->delete_fs_state(pipe, ctx->fsCube);
+   pipe->delete_fs_state(pipe, ctx->fs3d);
+   pipe->delete_fs_state(pipe, ctx->fs2d);
+   pipe->delete_fs_state(pipe, ctx->fs1d);
+   pipe->delete_vs_state(pipe, ctx->vs);
 
    pipe_resource_reference(&ctx->vbuf, NULL);
 
@@ -1478,9 +1469,8 @@ util_gen_mipmap(struct gen_mipmap_state *ctx,
    struct pipe_screen *screen = pipe->screen;
    struct pipe_framebuffer_state fb;
    struct pipe_resource *pt = psv->texture;
-   void *fs = (pt->target == PIPE_TEXTURE_CUBE) ? ctx->fsCube : ctx->fs2d;
+   void *fs;
    uint dstLevel;
-   uint zslice = 0;
    uint offset;
 
    /* The texture object should have room for the levels which we're
@@ -1494,8 +1484,28 @@ util_gen_mipmap(struct gen_mipmap_state *ctx,
    assert(filter == PIPE_TEX_FILTER_LINEAR ||
           filter == PIPE_TEX_FILTER_NEAREST);
 
+   switch (pt->target) {
+   case PIPE_TEXTURE_1D:
+      fs = ctx->fs1d;
+      break;
+   case PIPE_TEXTURE_2D:
+      fs = ctx->fs2d;
+      break;
+   case PIPE_TEXTURE_3D:
+      fs = ctx->fs3d;
+      break;
+   case PIPE_TEXTURE_CUBE:
+      fs = ctx->fsCube;
+      break;
+   case PIPE_TEXTURE_1D_ARRAY:
+   case PIPE_TEXTURE_2D_ARRAY:
+   default:
+      assert(0);
+      fs = ctx->fs2d;
+   }
+
    /* check if we can render in the texture's format */
-   if (!screen->is_format_supported(screen, psv->format, PIPE_TEXTURE_2D,
+   if (!screen->is_format_supported(screen, psv->format, pt->target,
                                     pt->nr_samples, PIPE_BIND_RENDER_TARGET, 0)) {
       fallback_gen_mipmap(ctx, pt, face, baseLevel, lastLevel);
       return;
@@ -1539,60 +1549,85 @@ util_gen_mipmap(struct gen_mipmap_state *ctx,
    for (dstLevel = baseLevel + 1; dstLevel <= lastLevel; dstLevel++) {
       const uint srcLevel = dstLevel - 1;
       struct pipe_viewport_state vp;
-
-      struct pipe_surface *surf = 
-         screen->get_tex_surface(screen, pt, face, dstLevel, zslice,
-                                 PIPE_BIND_RENDER_TARGET);
-
-      /*
-       * Setup framebuffer / dest surface
-       */
-      fb.cbufs[0] = surf;
-      fb.width = u_minify(pt->width0, dstLevel);
-      fb.height = u_minify(pt->height0, dstLevel);
-      cso_set_framebuffer(ctx->cso, &fb);
-
-      /* viewport */
-      vp.scale[0] = 0.5f * fb.width;
-      vp.scale[1] = 0.5f * fb.height;
-      vp.scale[2] = 1.0f;
-      vp.scale[3] = 1.0f;
-      vp.translate[0] = 0.5f * fb.width;
-      vp.translate[1] = 0.5f * fb.height;
-      vp.translate[2] = 0.0f;
-      vp.translate[3] = 0.0f;
-      cso_set_viewport(ctx->cso, &vp);
-
-      /*
-       * Setup sampler state
-       * Note: we should only have to set the min/max LOD clamps to ensure
-       * we grab texels from the right mipmap level.  But some hardware
-       * has trouble with min clamping so we also set the lod_bias to
-       * try to work around that.
-       */
-      ctx->sampler.min_lod = ctx->sampler.max_lod = (float) srcLevel;
-      ctx->sampler.lod_bias = (float) srcLevel;
-      cso_single_sampler(ctx->cso, 0, &ctx->sampler);
-      cso_single_sampler_done(ctx->cso);
-
-      cso_set_fragment_sampler_views(ctx->cso, 1, &psv);
-
-      /* quad coords in clip coords */
-      offset = set_vertex_data(ctx,
-                               pt->target,
-                               face);
-
-      util_draw_vertex_buffer(ctx->pipe, 
-                              ctx->vbuf,
-                              offset,
-                              PIPE_PRIM_TRIANGLE_FAN,
-                              4,  /* verts */
-                              2); /* attribs/vert */
-
-      pipe->flush(pipe, PIPE_FLUSH_RENDER_CACHE, NULL);
-
-      /* need to signal that the texture has changed _after_ rendering to it */
-      pipe_surface_reference( &surf, NULL );
+      unsigned nr_layers, layer, i;
+      float rcoord = 0.0f;
+
+      if (pt->target == PIPE_TEXTURE_3D)
+         nr_layers = u_minify(pt->depth0, dstLevel);
+      else
+         nr_layers = 1;
+
+      for (i = 0; i < nr_layers; i++) {
+         struct pipe_surface *surf, surf_templ;
+         if (pt->target == PIPE_TEXTURE_3D) {
+            /* in theory with geom shaders and driver with full layer support
+               could do that in one go. */
+            layer = i;
+            offset = 1.0f / (float)(nr_layers * 2);
+            /* XXX hmm really? */
+            rcoord = (float)layer / (float)nr_layers + 1.0f / (float)(nr_layers * 2);
+         }
+         else
+            layer = face;
+
+         memset(&surf_templ, 0, sizeof(surf_templ));
+         u_surface_default_template(&surf_templ, pt, PIPE_BIND_RENDER_TARGET);
+         surf_templ.u.tex.level = dstLevel;
+         surf_templ.u.tex.first_layer = layer;
+         surf_templ.u.tex.last_layer = layer;
+         surf = pipe->create_surface(pipe, pt, &surf_templ);
+
+         /*
+          * Setup framebuffer / dest surface
+          */
+         fb.cbufs[0] = surf;
+         fb.width = u_minify(pt->width0, dstLevel);
+         fb.height = u_minify(pt->height0, dstLevel);
+         cso_set_framebuffer(ctx->cso, &fb);
+
+         /* viewport */
+         vp.scale[0] = 0.5f * fb.width;
+         vp.scale[1] = 0.5f * fb.height;
+         vp.scale[2] = 1.0f;
+         vp.scale[3] = 1.0f;
+         vp.translate[0] = 0.5f * fb.width;
+         vp.translate[1] = 0.5f * fb.height;
+         vp.translate[2] = 0.0f;
+         vp.translate[3] = 0.0f;
+         cso_set_viewport(ctx->cso, &vp);
+
+         /*
+          * Setup sampler state
+          * Note: we should only have to set the min/max LOD clamps to ensure
+          * we grab texels from the right mipmap level.  But some hardware
+          * has trouble with min clamping so we also set the lod_bias to
+          * try to work around that.
+          */
+         ctx->sampler.min_lod = ctx->sampler.max_lod = (float) srcLevel;
+         ctx->sampler.lod_bias = (float) srcLevel;
+         cso_single_sampler(ctx->cso, 0, &ctx->sampler);
+         cso_single_sampler_done(ctx->cso);
+
+         cso_set_fragment_sampler_views(ctx->cso, 1, &psv);
+
+         /* quad coords in clip coords */
+         offset = set_vertex_data(ctx,
+                                  pt->target,
+                                  face,
+                                  rcoord);
+
+         util_draw_vertex_buffer(ctx->pipe, 
+                                 ctx->vbuf,
+                                 offset,
+                                 PIPE_PRIM_TRIANGLE_FAN,
+                                 4,  /* verts */
+                                 2); /* attribs/vert */
+
+         pipe->flush(pipe, PIPE_FLUSH_RENDER_CACHE, NULL);
+
+         /* need to signal that the texture has changed _after_ rendering to it */
+         pipe_surface_reference( &surf, NULL );
+      }
    }
 
    /* restore state we changed */
diff --git a/src/gallium/auxiliary/util/u_gen_mipmap.h b/src/gallium/auxiliary/util/u_gen_mipmap.h
index a7502b9982..a10b6a4aba 100644
--- a/src/gallium/auxiliary/util/u_gen_mipmap.h
+++ b/src/gallium/auxiliary/util/u_gen_mipmap.h
@@ -60,7 +60,7 @@ util_gen_mipmap_flush( struct gen_mipmap_state *ctx );
 extern void
 util_gen_mipmap(struct gen_mipmap_state *ctx,
                 struct pipe_sampler_view *psv,
-                uint face, uint baseLevel, uint lastLevel, uint filter);
+                uint layer, uint baseLevel, uint lastLevel, uint filter);
 
 
 #ifdef __cplusplus
diff --git a/src/gallium/auxiliary/util/u_inlines.h b/src/gallium/auxiliary/util/u_inlines.h
index 6ed39561fb..e55aafe90f 100644
--- a/src/gallium/auxiliary/util/u_inlines.h
+++ b/src/gallium/auxiliary/util/u_inlines.h
@@ -109,7 +109,7 @@ pipe_surface_reference(struct pipe_surface **ptr, struct pipe_surface *surf)
 
    if (pipe_reference_described(&(*ptr)->reference, &surf->reference, 
                                 (debug_reference_descriptor)debug_describe_surface))
-      old_surf->texture->screen->tex_surface_destroy(old_surf);
+      old_surf->context->surface_destroy(old_surf->context, old_surf);
    *ptr = surf;
 }
 
@@ -136,26 +136,28 @@ pipe_sampler_view_reference(struct pipe_sampler_view **ptr, struct pipe_sampler_
 }
 
 static INLINE void
-pipe_surface_reset(struct pipe_surface* ps, struct pipe_resource *pt,
-		unsigned face, unsigned level, unsigned zslice, unsigned flags)
+pipe_surface_reset(struct pipe_context *ctx, struct pipe_surface* ps,
+                   struct pipe_resource *pt, unsigned level, unsigned layer,
+                   unsigned flags)
 {
    pipe_resource_reference(&ps->texture, pt);
    ps->format = pt->format;
    ps->width = u_minify(pt->width0, level);
    ps->height = u_minify(pt->height0, level);
    ps->usage = flags;
-   ps->face = face;
-   ps->level = level;
-   ps->zslice = zslice;
+   ps->u.tex.level = level;
+   ps->u.tex.first_layer = ps->u.tex.last_layer = layer;
+   ps->context = ctx;
 }
 
 static INLINE void
-pipe_surface_init(struct pipe_surface* ps, struct pipe_resource *pt,
-                unsigned face, unsigned level, unsigned zslice, unsigned flags)
+pipe_surface_init(struct pipe_context *ctx, struct pipe_surface* ps,
+                  struct pipe_resource *pt, unsigned level, unsigned layer,
+                  unsigned flags)
 {
    ps->texture = 0;
    pipe_reference_init(&ps->reference, 1);
-   pipe_surface_reset(ps, pt, face, level, zslice, flags);
+   pipe_surface_reset(ctx, ps, pt, level, layer, flags);
 }
 
 /*
@@ -177,6 +179,7 @@ pipe_buffer_create( struct pipe_screen *screen,
    buffer.width0 = size;
    buffer.height0 = 1;
    buffer.depth0 = 1;
+   buffer.array_size = 1;
    return screen->resource_create(screen, &buffer);
 }
 
@@ -202,15 +205,15 @@ pipe_buffer_map_range(struct pipe_context *pipe,
    assert(offset < buffer->width0);
    assert(offset + length <= buffer->width0);
    assert(length);
-   
+
    u_box_1d(offset, length, &box);
 
    *transfer = pipe->get_transfer( pipe,
-				   buffer,
-				   u_subresource(0, 0),
-				   usage,
-				   &box);
-   
+                                   buffer,
+                                   0,
+                                   usage,
+                                   &box);
+
    if (*transfer == NULL)
       return NULL;
 
@@ -231,7 +234,7 @@ static INLINE void *
 pipe_buffer_map(struct pipe_context *pipe,
                 struct pipe_resource *buffer,
                 unsigned usage,
-		struct pipe_transfer **transfer)
+                struct pipe_transfer **transfer)
 {
    return pipe_buffer_map_range(pipe, buffer, 0, buffer->width0, usage, transfer);
 }
@@ -240,7 +243,7 @@ pipe_buffer_map(struct pipe_context *pipe,
 static INLINE void
 pipe_buffer_unmap(struct pipe_context *pipe,
                   struct pipe_resource *buf,
-		  struct pipe_transfer *transfer)
+                  struct pipe_transfer *transfer)
 {
    if (transfer) {
       pipe->transfer_unmap(pipe, transfer);
@@ -250,7 +253,7 @@ pipe_buffer_unmap(struct pipe_context *pipe,
 
 static INLINE void
 pipe_buffer_flush_mapped_range(struct pipe_context *pipe,
-			       struct pipe_transfer *transfer,
+                               struct pipe_transfer *transfer,
                                unsigned offset,
                                unsigned length)
 {
@@ -266,7 +269,7 @@ pipe_buffer_flush_mapped_range(struct pipe_context *pipe,
     * mapped range.
     */
    transfer_offset = offset - transfer->box.x;
-   
+
    u_box_1d(transfer_offset, length, &box);
 
    pipe->transfer_flush_region(pipe, transfer, &box);
@@ -276,7 +279,7 @@ static INLINE void
 pipe_buffer_write(struct pipe_context *pipe,
                   struct pipe_resource *buf,
                   unsigned offset,
-		  unsigned size,
+                  unsigned size,
                   const void *data)
 {
    struct pipe_box box;
@@ -284,13 +287,13 @@ pipe_buffer_write(struct pipe_context *pipe,
    u_box_1d(offset, size, &box);
 
    pipe->transfer_inline_write( pipe,
-				buf,
-				u_subresource(0,0),
-				PIPE_TRANSFER_WRITE,
-				&box,
-				data,
-				size,
-				0);
+                                buf,
+                                0,
+                                PIPE_TRANSFER_WRITE,
+                                &box,
+                                data,
+                                size,
+                                0);
 }
 
 /**
@@ -309,21 +312,21 @@ pipe_buffer_write_nooverlap(struct pipe_context *pipe,
 
    u_box_1d(offset, size, &box);
 
-   pipe->transfer_inline_write(pipe, 
-			       buf,
-			       u_subresource(0,0),
-			       (PIPE_TRANSFER_WRITE |
-				PIPE_TRANSFER_NOOVERWRITE),
-			       &box,
-			       data,
-			       0, 0);
+   pipe->transfer_inline_write(pipe,
+                               buf,
+                               0,
+                               (PIPE_TRANSFER_WRITE |
+                                PIPE_TRANSFER_NOOVERWRITE),
+                               &box,
+                               data,
+                               0, 0);
 }
 
 static INLINE void
 pipe_buffer_read(struct pipe_context *pipe,
                  struct pipe_resource *buf,
                  unsigned offset,
-		 unsigned size,
+                 unsigned size,
                  void *data)
 {
    struct pipe_transfer *src_transfer;
@@ -343,20 +346,19 @@ pipe_buffer_read(struct pipe_context *pipe,
 
 static INLINE struct pipe_transfer *
 pipe_get_transfer( struct pipe_context *context,
-		       struct pipe_resource *resource,
-		       unsigned face, unsigned level,
-		       unsigned zslice,
-		       enum pipe_transfer_usage usage,
-		       unsigned x, unsigned y,
-		       unsigned w, unsigned h)
+                   struct pipe_resource *resource,
+                   unsigned level, unsigned layer,
+                   enum pipe_transfer_usage usage,
+                   unsigned x, unsigned y,
+                   unsigned w, unsigned h)
 {
    struct pipe_box box;
-   u_box_2d_zslice( x, y, zslice, w, h, &box );
+   u_box_2d_zslice( x, y, layer, w, h, &box );
    return context->get_transfer( context,
-				 resource,
-				 u_subresource(face, level),
-				 usage,
-				 &box );
+                                 resource,
+                                 level,
+                                 usage,
+                                 &box );
 }
 
 static INLINE void *
@@ -376,7 +378,7 @@ pipe_transfer_unmap( struct pipe_context *context,
 
 static INLINE void
 pipe_transfer_destroy( struct pipe_context *context, 
-		       struct pipe_transfer *transfer )
+                       struct pipe_transfer *transfer )
 {
    context->transfer_destroy(context, transfer);
 }
diff --git a/src/gallium/auxiliary/util/u_resource.c b/src/gallium/auxiliary/util/u_resource.c
index 9e6474b83d..443c9f8067 100644
--- a/src/gallium/auxiliary/util/u_resource.c
+++ b/src/gallium/auxiliary/util/u_resource.c
@@ -10,85 +10,85 @@ u_resource( struct pipe_resource *res )
 }
 
 boolean u_resource_get_handle_vtbl(struct pipe_screen *screen,
-			  struct pipe_resource *resource,
-			  struct winsys_handle *handle)
+                                   struct pipe_resource *resource,
+                                   struct winsys_handle *handle)
 {
    struct u_resource *ur = u_resource(resource);
    return ur->vtbl->resource_get_handle(screen, resource, handle);
 }
 
 void u_resource_destroy_vtbl(struct pipe_screen *screen,
-		    struct pipe_resource *resource)
+                             struct pipe_resource *resource)
 {
    struct u_resource *ur = u_resource(resource);
    ur->vtbl->resource_destroy(screen, resource);
 }
 
 unsigned u_is_resource_referenced_vtbl( struct pipe_context *pipe,
-					struct pipe_resource *resource,
-					unsigned face, unsigned level)
+                                        struct pipe_resource *resource,
+                                        unsigned level, int layer)
 {
    struct u_resource *ur = u_resource(resource);
-   return ur->vtbl->is_resource_referenced(pipe, resource, face, level);
+   return ur->vtbl->is_resource_referenced(pipe, resource, level, layer);
 }
 
 struct pipe_transfer *u_get_transfer_vtbl(struct pipe_context *context,
-					  struct pipe_resource *resource,
-					  struct pipe_subresource sr,
-					  enum pipe_transfer_usage usage,
-					  const struct pipe_box *box)
+                                          struct pipe_resource *resource,
+                                          unsigned level,
+                                          enum pipe_transfer_usage usage,
+                                          const struct pipe_box *box)
 {
    struct u_resource *ur = u_resource(resource);
-   return ur->vtbl->get_transfer(context, resource, sr, usage, box);
+   return ur->vtbl->get_transfer(context, resource, level, usage, box);
 }
 
 void u_transfer_destroy_vtbl(struct pipe_context *pipe,
-			     struct pipe_transfer *transfer)
+                             struct pipe_transfer *transfer)
 {
    struct u_resource *ur = u_resource(transfer->resource);
    ur->vtbl->transfer_destroy(pipe, transfer);
 }
 
 void *u_transfer_map_vtbl( struct pipe_context *pipe,
-			   struct pipe_transfer *transfer )
+                           struct pipe_transfer *transfer )
 {
    struct u_resource *ur = u_resource(transfer->resource);
    return ur->vtbl->transfer_map(pipe, transfer);
 }
 
 void u_transfer_flush_region_vtbl( struct pipe_context *pipe,
-				   struct pipe_transfer *transfer,
-				   const struct pipe_box *box)
+                                   struct pipe_transfer *transfer,
+                                   const struct pipe_box *box)
 {
    struct u_resource *ur = u_resource(transfer->resource);
    ur->vtbl->transfer_flush_region(pipe, transfer, box);
 }
 
 void u_transfer_unmap_vtbl( struct pipe_context *pipe,
-			    struct pipe_transfer *transfer )
+                            struct pipe_transfer *transfer )
 {
    struct u_resource *ur = u_resource(transfer->resource);
    ur->vtbl->transfer_unmap(pipe, transfer);
 }
 
 void u_transfer_inline_write_vtbl( struct pipe_context *pipe,
-				   struct pipe_resource *resource,
-				   struct pipe_subresource sr,
-				   unsigned usage,
-				   const struct pipe_box *box,
-				   const void *data,
-				   unsigned stride,
-				   unsigned slice_stride)
+                                   struct pipe_resource *resource,
+                                   unsigned level,
+                                   unsigned usage,
+                                   const struct pipe_box *box,
+                                   const void *data,
+                                   unsigned stride,
+                                   unsigned layer_stride)
 {
    struct u_resource *ur = u_resource(resource);
-   ur->vtbl->transfer_inline_write(pipe, 
-				   resource,
-				   sr,
-				   usage,
-				   box,
-				   data,
-				   stride,
-				   slice_stride);
+   ur->vtbl->transfer_inline_write(pipe,
+                                   resource,
+                                   level,
+                                   usage,
+                                   box,
+                                   data,
+                                   stride,
+                                   layer_stride);
 }
 
 
diff --git a/src/gallium/auxiliary/util/u_sampler.c b/src/gallium/auxiliary/util/u_sampler.c
index e77f562ea2..bb26099b7e 100644
--- a/src/gallium/auxiliary/util/u_sampler.c
+++ b/src/gallium/auxiliary/util/u_sampler.c
@@ -40,8 +40,11 @@ default_template(struct pipe_sampler_view *view,
     */
 
    view->format = format;
-   view->first_level = 0;
-   view->last_level = texture->last_level;
+   view->u.tex.first_level = 0;
+   view->u.tex.last_level = texture->last_level;
+   view->u.tex.first_layer = 0;
+   view->u.tex.last_layer = texture->target == PIPE_TEXTURE_3D ?
+                               texture->depth0 - 1 : texture->array_size - 1;
    view->swizzle_r = PIPE_SWIZZLE_RED;
    view->swizzle_g = PIPE_SWIZZLE_GREEN;
    view->swizzle_b = PIPE_SWIZZLE_BLUE;
diff --git a/src/gallium/auxiliary/util/u_simple_screen.h b/src/gallium/auxiliary/util/u_simple_screen.h
index b52232f025..7139aaabc5 100644
--- a/src/gallium/auxiliary/util/u_simple_screen.h
+++ b/src/gallium/auxiliary/util/u_simple_screen.h
@@ -57,7 +57,8 @@ struct pipe_winsys
     * displayed, eg copy fake frontbuffer.
     */
    void (*flush_frontbuffer)( struct pipe_winsys *ws,
-                              struct pipe_surface *surf,
+                              struct pipe_resource *resource,
+                              unsigned level, unsigned layer,
                               void *context_private );
 
 
diff --git a/src/gallium/auxiliary/util/u_simple_shaders.c b/src/gallium/auxiliary/util/u_simple_shaders.c
index 58ef68377f..b0f2dd8aa2 100644
--- a/src/gallium/auxiliary/util/u_simple_shaders.c
+++ b/src/gallium/auxiliary/util/u_simple_shaders.c
@@ -204,7 +204,8 @@ util_make_fragment_tex_shader_writedepth(struct pipe_context *pipe,
 void *
 util_make_fragment_passthrough_shader(struct pipe_context *pipe)
 {
-   return util_make_fragment_clonecolor_shader(pipe, 1);
+   return util_make_fragment_cloneinput_shader(pipe, 1, TGSI_SEMANTIC_COLOR,
+                                               TGSI_INTERPOLATE_PERSPECTIVE);
 }
 
 
@@ -212,7 +213,9 @@ util_make_fragment_passthrough_shader(struct pipe_context *pipe)
  * Make a fragment shader that copies the input color to N output colors.
  */
 void *
-util_make_fragment_clonecolor_shader(struct pipe_context *pipe, int num_cbufs)
+util_make_fragment_cloneinput_shader(struct pipe_context *pipe, int num_cbufs,
+                                     int input_semantic,
+                                     int input_interpolate)
 {
    struct ureg_program *ureg;
    struct ureg_src src;
@@ -225,8 +228,8 @@ util_make_fragment_clonecolor_shader(struct pipe_context *pipe, int num_cbufs)
    if (ureg == NULL)
       return NULL;
 
-   src = ureg_DECL_fs_input( ureg, TGSI_SEMANTIC_COLOR, 0, 
-                             TGSI_INTERPOLATE_PERSPECTIVE );
+   src = ureg_DECL_fs_input( ureg, input_semantic, 0,
+                             input_interpolate );
 
    for (i = 0; i < num_cbufs; i++)
       dst[i] = ureg_DECL_output( ureg, TGSI_SEMANTIC_COLOR, i );
diff --git a/src/gallium/auxiliary/util/u_simple_shaders.h b/src/gallium/auxiliary/util/u_simple_shaders.h
index 4aa34bc475..1bfec183e3 100644
--- a/src/gallium/auxiliary/util/u_simple_shaders.h
+++ b/src/gallium/auxiliary/util/u_simple_shaders.h
@@ -71,7 +71,9 @@ util_make_fragment_passthrough_shader(struct pipe_context *pipe);
 
 
 extern void *
-util_make_fragment_clonecolor_shader(struct pipe_context *pipe, int num_cbufs);
+util_make_fragment_cloneinput_shader(struct pipe_context *pipe, int num_cbufs,
+                                     int input_semantic,
+                                     int input_interpolate);
 
 #ifdef __cplusplus
 }
diff --git a/src/gallium/auxiliary/util/u_mempool.c b/src/gallium/auxiliary/util/u_slab.c
index 1f336b39a1..21bf2d735a 100644
--- a/src/gallium/auxiliary/util/u_mempool.c
+++ b/src/gallium/auxiliary/util/u_slab.c
@@ -20,7 +20,7 @@
  * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
  * USE OR OTHER DEALINGS IN THE SOFTWARE. */
 
-#include "util/u_mempool.h"
+#include "util/u_slab.h"
 
 #include "util/u_math.h"
 #include "util/u_memory.h"
@@ -28,13 +28,13 @@
 
 #include <stdio.h>
 
-#define UTIL_MEMPOOL_MAGIC 0xcafe4321
+#define UTIL_SLAB_MAGIC 0xcafe4321
 
 /* The block is either allocated memory or free space. */
-struct util_mempool_block {
+struct util_slab_block {
    /* The header. */
    /* The first next free block. */
-   struct util_mempool_block *next_free;
+   struct util_slab_block *next_free;
 
    intptr_t magic;
 
@@ -42,19 +42,19 @@ struct util_mempool_block {
     * The allocated size is always larger than this structure. */
 };
 
-static struct util_mempool_block *
-util_mempool_get_block(struct util_mempool *pool,
-                       struct util_mempool_page *page, unsigned index)
+static struct util_slab_block *
+util_slab_get_block(struct util_slab_mempool *pool,
+                    struct util_slab_page *page, unsigned index)
 {
-   return (struct util_mempool_block*)
-          ((uint8_t*)page + sizeof(struct util_mempool_page) +
+   return (struct util_slab_block*)
+          ((uint8_t*)page + sizeof(struct util_slab_page) +
            (pool->block_size * index));
 }
 
-static void util_mempool_add_new_page(struct util_mempool *pool)
+static void util_slab_add_new_page(struct util_slab_mempool *pool)
 {
-   struct util_mempool_page *page;
-   struct util_mempool_block *block;
+   struct util_slab_page *page;
+   struct util_slab_block *block;
    int i;
 
    page = MALLOC(pool->page_size);
@@ -62,15 +62,15 @@ static void util_mempool_add_new_page(struct util_mempool *pool)
 
    /* Mark all blocks as free. */
    for (i = 0; i < pool->num_blocks-1; i++) {
-      block = util_mempool_get_block(pool, page, i);
-      block->next_free = util_mempool_get_block(pool, page, i+1);
-      block->magic = UTIL_MEMPOOL_MAGIC;
+      block = util_slab_get_block(pool, page, i);
+      block->next_free = util_slab_get_block(pool, page, i+1);
+      block->magic = UTIL_SLAB_MAGIC;
    }
 
-   block = util_mempool_get_block(pool, page, pool->num_blocks-1);
+   block = util_slab_get_block(pool, page, pool->num_blocks-1);
    block->next_free = pool->first_free;
-   block->magic = UTIL_MEMPOOL_MAGIC;
-   pool->first_free = util_mempool_get_block(pool, page, 0);
+   block->magic = UTIL_SLAB_MAGIC;
+   pool->first_free = util_slab_get_block(pool, page, 0);
    pool->num_pages++;
 
 #if 0
@@ -78,74 +78,74 @@ static void util_mempool_add_new_page(struct util_mempool *pool)
 #endif
 }
 
-static void *util_mempool_malloc_st(struct util_mempool *pool)
+static void *util_slab_alloc_st(struct util_slab_mempool *pool)
 {
-   struct util_mempool_block *block;
+   struct util_slab_block *block;
 
    if (!pool->first_free)
-      util_mempool_add_new_page(pool);
+      util_slab_add_new_page(pool);
 
    block = pool->first_free;
-   assert(block->magic == UTIL_MEMPOOL_MAGIC);
+   assert(block->magic == UTIL_SLAB_MAGIC);
    pool->first_free = block->next_free;
 
-   return (uint8_t*)block + sizeof(struct util_mempool_block);
+   return (uint8_t*)block + sizeof(struct util_slab_block);
 }
 
-static void util_mempool_free_st(struct util_mempool *pool, void *ptr)
+static void util_slab_free_st(struct util_slab_mempool *pool, void *ptr)
 {
-   struct util_mempool_block *block =
-         (struct util_mempool_block*)
-         ((uint8_t*)ptr - sizeof(struct util_mempool_block));
+   struct util_slab_block *block =
+         (struct util_slab_block*)
+         ((uint8_t*)ptr - sizeof(struct util_slab_block));
 
-   assert(block->magic == UTIL_MEMPOOL_MAGIC);
+   assert(block->magic == UTIL_SLAB_MAGIC);
    block->next_free = pool->first_free;
    pool->first_free = block;
 }
 
-static void *util_mempool_malloc_mt(struct util_mempool *pool)
+static void *util_slab_alloc_mt(struct util_slab_mempool *pool)
 {
    void *mem;
 
    pipe_mutex_lock(pool->mutex);
-   mem = util_mempool_malloc_st(pool);
+   mem = util_slab_alloc_st(pool);
    pipe_mutex_unlock(pool->mutex);
    return mem;
 }
 
-static void util_mempool_free_mt(struct util_mempool *pool, void *ptr)
+static void util_slab_free_mt(struct util_slab_mempool *pool, void *ptr)
 {
    pipe_mutex_lock(pool->mutex);
-   util_mempool_free_st(pool, ptr);
+   util_slab_free_st(pool, ptr);
    pipe_mutex_unlock(pool->mutex);
 }
 
-void util_mempool_set_thread_safety(struct util_mempool *pool,
-                                    enum util_mempool_threading threading)
+void util_slab_set_thread_safety(struct util_slab_mempool *pool,
+                                    enum util_slab_threading threading)
 {
    pool->threading = threading;
 
    if (threading) {
-      pool->malloc = util_mempool_malloc_mt;
-      pool->free = util_mempool_free_mt;
+      pool->alloc = util_slab_alloc_mt;
+      pool->free = util_slab_free_mt;
    } else {
-      pool->malloc = util_mempool_malloc_st;
-      pool->free = util_mempool_free_st;
+      pool->alloc = util_slab_alloc_st;
+      pool->free = util_slab_free_st;
    }
 }
 
-void util_mempool_create(struct util_mempool *pool,
-                         unsigned item_size,
-                         unsigned num_blocks,
-                         enum util_mempool_threading threading)
+void util_slab_create(struct util_slab_mempool *pool,
+                      unsigned item_size,
+                      unsigned num_blocks,
+                      enum util_slab_threading threading)
 {
    item_size = align(item_size, sizeof(intptr_t));
 
    pool->num_pages = 0;
    pool->num_blocks = num_blocks;
-   pool->block_size = sizeof(struct util_mempool_block) + item_size;
+   pool->block_size = sizeof(struct util_slab_block) + item_size;
    pool->block_size = align(pool->block_size, sizeof(intptr_t));
-   pool->page_size = sizeof(struct util_mempool_page) +
+   pool->page_size = sizeof(struct util_slab_page) +
                      num_blocks * pool->block_size;
    pool->first_free = NULL;
 
@@ -153,12 +153,12 @@ void util_mempool_create(struct util_mempool *pool,
 
    pipe_mutex_init(pool->mutex);
 
-   util_mempool_set_thread_safety(pool, threading);
+   util_slab_set_thread_safety(pool, threading);
 }
 
-void util_mempool_destroy(struct util_mempool *pool)
+void util_slab_destroy(struct util_slab_mempool *pool)
 {
-   struct util_mempool_page *page, *temp;
+   struct util_slab_page *page, *temp;
 
    foreach_s(page, temp, &pool->list) {
       remove_from_list(page);
diff --git a/src/gallium/auxiliary/util/u_mempool.h b/src/gallium/auxiliary/util/u_slab.h
index a5b5d6a9b7..6b9718d08a 100644
--- a/src/gallium/auxiliary/util/u_mempool.h
+++ b/src/gallium/auxiliary/util/u_slab.h
@@ -22,66 +22,66 @@
 
 /**
  * @file
- * Simple memory pool for equally sized memory allocations.
- * util_mempool_malloc and util_mempool_free are in O(1).
+ * Simple slab allocator for equally sized memory allocations.
+ * util_slab_alloc and util_slab_free have time complexity in O(1).
  *
  * Good for allocations which have very low lifetime and are allocated
- * and freed very often. Use a profiler first!
+ * and freed very often. Use a profiler first to know if it's worth using it!
  *
  * Candidates: get_transfer, user_buffer_create
  *
  * @author Marek Olšák
  */
 
-#ifndef U_MEMPOOL_H
-#define U_MEMPOOL_H
+#ifndef U_SLAB_H
+#define U_SLAB_H
 
 #include "os/os_thread.h"
 
-enum util_mempool_threading {
-   UTIL_MEMPOOL_SINGLETHREADED = FALSE,
-   UTIL_MEMPOOL_MULTITHREADED = TRUE
+enum util_slab_threading {
+   UTIL_SLAB_SINGLETHREADED = FALSE,
+   UTIL_SLAB_MULTITHREADED = TRUE
 };
 
 /* The page is an array of blocks (allocations). */
-struct util_mempool_page {
+struct util_slab_page {
    /* The header (linked-list pointers). */
-   struct util_mempool_page *prev, *next;
+   struct util_slab_page *prev, *next;
 
    /* Memory after the last member is dedicated to the page itself.
     * The allocated size is always larger than this structure. */
 };
 
-struct util_mempool {
+struct util_slab_mempool {
    /* Public members. */
-   void *(*malloc)(struct util_mempool *pool);
-   void (*free)(struct util_mempool *pool, void *ptr);
+   void *(*alloc)(struct util_slab_mempool *pool);
+   void (*free)(struct util_slab_mempool *pool, void *ptr);
 
    /* Private members. */
-   struct util_mempool_block *first_free;
+   struct util_slab_block *first_free;
 
-   struct util_mempool_page list;
+   struct util_slab_page list;
 
    unsigned block_size;
    unsigned page_size;
    unsigned num_blocks;
    unsigned num_pages;
-   enum util_mempool_threading threading;
+   enum util_slab_threading threading;
 
    pipe_mutex mutex;
 };
 
-void util_mempool_create(struct util_mempool *pool,
-                         unsigned item_size,
-                         unsigned num_blocks,
-                         enum util_mempool_threading threading);
+void util_slab_create(struct util_slab_mempool *pool,
+                      unsigned item_size,
+                      unsigned num_blocks,
+                      enum util_slab_threading threading);
 
-void util_mempool_destroy(struct util_mempool *pool);
+void util_slab_destroy(struct util_slab_mempool *pool);
 
-void util_mempool_set_thread_safety(struct util_mempool *pool,
-                                    enum util_mempool_threading threading);
+void util_slab_set_thread_safety(struct util_slab_mempool *pool,
+                                 enum util_slab_threading threading);
 
-#define util_mempool_malloc(pool)    (pool)->malloc(pool)
-#define util_mempool_free(pool, ptr) (pool)->free(pool, ptr)
+#define util_slab_alloc(pool)     (pool)->alloc(pool)
+#define util_slab_free(pool, ptr) (pool)->free(pool, ptr)
 
 #endif
diff --git a/src/gallium/auxiliary/util/u_staging.c b/src/gallium/auxiliary/util/u_staging.c
index c5d68f8df8..b6bf241a22 100644
--- a/src/gallium/auxiliary/util/u_staging.c
+++ b/src/gallium/auxiliary/util/u_staging.c
@@ -41,6 +41,7 @@ util_staging_resource_template(struct pipe_resource *pt, unsigned width, unsigne
    template->width0 = width;
    template->height0 = height;
    template->depth0 = depth;
+   template->array_size = 1;
    template->last_level = 0;
    template->nr_samples = pt->nr_samples;
    template->bind = 0;
@@ -51,7 +52,7 @@ util_staging_resource_template(struct pipe_resource *pt, unsigned width, unsigne
 struct util_staging_transfer *
 util_staging_transfer_init(struct pipe_context *pipe,
            struct pipe_resource *pt,
-           struct pipe_subresource sr,
+           unsigned level,
            unsigned usage,
            const struct pipe_box *box,
            bool direct, struct util_staging_transfer *tx)
@@ -61,7 +62,7 @@ util_staging_transfer_init(struct pipe_context *pipe,
    struct pipe_resource staging_resource_template;
 
    pipe_resource_reference(&tx->base.resource, pt);
-   tx->base.sr = sr;
+   tx->base.level = level;
    tx->base.usage = usage;
    tx->base.box = *box;
 
@@ -82,12 +83,20 @@ util_staging_transfer_init(struct pipe_context *pipe,
 
    if (usage & PIPE_TRANSFER_READ)
    {
-      struct pipe_subresource dstsr;
+      /* XXX this looks wrong dst is always the same but looping over src z? */
       unsigned zi;
-      dstsr.face = 0;
-      dstsr.level = 0;
-      for(zi = 0; zi < box->depth; ++zi)
-         pipe->resource_copy_region(pipe, tx->staging_resource, dstsr, 0, 0, 0, tx->base.resource, sr, box->x, box->y, box->z + zi, box->width, box->height);
+      struct pipe_box sbox;
+      sbox.x = box->x;
+      sbox.y = box->y;
+      sbox.z = box->z;
+      sbox.width = box->width;
+      sbox.height = box->height;
+      sbox.depth = 1;
+      for(zi = 0; zi < box->depth; ++zi) {
+         sbox.z = sbox.z + zi;
+         pipe->resource_copy_region(pipe, tx->staging_resource, 0, 0, 0, 0,
+                                    tx->base.resource, level, &sbox);
+      }
    }
 
    return tx;
@@ -101,12 +110,18 @@ util_staging_transfer_destroy(struct pipe_context *pipe, struct pipe_transfer *p
    if (tx->staging_resource != tx->base.resource)
    {
       if(tx->base.usage & PIPE_TRANSFER_WRITE) {
-         struct pipe_subresource srcsr;
+         /* XXX this looks wrong src is always the same but looping over dst z? */
          unsigned zi;
-         srcsr.face = 0;
-         srcsr.level = 0;
+         struct pipe_box sbox;
+         sbox.x = 0;
+         sbox.y = 0;
+         sbox.z = 0;
+         sbox.width = tx->base.box.width;
+         sbox.height = tx->base.box.height;
+         sbox.depth = 1;
          for(zi = 0; zi < tx->base.box.depth; ++zi)
-            pipe->resource_copy_region(pipe, tx->base.resource, tx->base.sr, tx->base.box.x, tx->base.box.y, tx->base.box.z + zi, tx->staging_resource, srcsr, 0, 0, 0, tx->base.box.width, tx->base.box.height);
+            pipe->resource_copy_region(pipe, tx->base.resource, tx->base.level, tx->base.box.x, tx->base.box.y, tx->base.box.z + zi,
+                                       tx->staging_resource, 0, &sbox);
       }
 
       pipe_resource_reference(&tx->staging_resource, NULL);
diff --git a/src/gallium/auxiliary/util/u_staging.h b/src/gallium/auxiliary/util/u_staging.h
index 1aab78cc88..49839d2543 100644
--- a/src/gallium/auxiliary/util/u_staging.h
+++ b/src/gallium/auxiliary/util/u_staging.h
@@ -52,7 +52,7 @@ struct util_staging_transfer {
 struct util_staging_transfer *
 util_staging_transfer_init(struct pipe_context *pipe,
            struct pipe_resource *pt,
-           struct pipe_subresource sr,
+           unsigned level,
            unsigned usage,
            const struct pipe_box *box,
            bool direct, struct util_staging_transfer *tx);
diff --git a/src/gallium/auxiliary/util/u_surface.c b/src/gallium/auxiliary/util/u_surface.c
index f78b6838a7..4eddd3f519 100644
--- a/src/gallium/auxiliary/util/u_surface.c
+++ b/src/gallium/auxiliary/util/u_surface.c
@@ -42,6 +42,18 @@
 #include "util/u_surface.h"
 #include "util/u_pack_color.h"
 
+void
+u_surface_default_template(struct pipe_surface *view,
+                           const struct pipe_resource *texture,
+                           unsigned bind)
+{
+   view->format = texture->format;
+   view->u.tex.level = 0;
+   view->u.tex.first_layer = 0;
+   view->u.tex.last_layer = 0;
+   /* XXX should filter out all non-rt/ds bind flags ? */
+   view->usage = bind;
+}
 
 /**
  * Helper to quickly create an RGBA rendering surface of a certain size.
@@ -50,9 +62,9 @@
  * \return TRUE for success, FALSE if failure
  */
 boolean
-util_create_rgba_surface(struct pipe_screen *screen,
+util_create_rgba_surface(struct pipe_context *pipe,
                          uint width, uint height,
-			 uint bind,
+                         uint bind,
                          struct pipe_resource **textureOut,
                          struct pipe_surface **surfaceOut)
 {
@@ -65,6 +77,8 @@ util_create_rgba_surface(struct pipe_screen *screen,
    const uint target = PIPE_TEXTURE_2D;
    enum pipe_format format = PIPE_FORMAT_NONE;
    struct pipe_resource templ;
+   struct pipe_surface surf_templ;
+   struct pipe_screen *screen = pipe->screen;
    uint i;
 
    /* Choose surface format */
@@ -86,17 +100,20 @@ util_create_rgba_surface(struct pipe_screen *screen,
    templ.width0 = width;
    templ.height0 = height;
    templ.depth0 = 1;
+   templ.array_size = 1;
    templ.bind = bind;
 
    *textureOut = screen->resource_create(screen, &templ);
    if (!*textureOut)
       return FALSE;
 
+   /* create surface */
+   memset(&surf_templ, 0, sizeof(surf_templ));
+   u_surface_default_template(&surf_templ, *textureOut, bind);
    /* create surface / view into texture */
-   *surfaceOut = screen->get_tex_surface(screen, 
-					 *textureOut,
-					 0, 0, 0,
-					 bind);
+   *surfaceOut = pipe->create_surface(pipe,
+                                      *textureOut,
+                                      &surf_templ);
    if (!*surfaceOut) {
       pipe_resource_reference(textureOut, NULL);
       return FALSE;
@@ -126,17 +143,18 @@ util_destroy_rgba_surface(struct pipe_resource *texture,
 void
 util_resource_copy_region(struct pipe_context *pipe,
                           struct pipe_resource *dst,
-                          struct pipe_subresource subdst,
+                          unsigned dst_level,
                           unsigned dst_x, unsigned dst_y, unsigned dst_z,
                           struct pipe_resource *src,
-                          struct pipe_subresource subsrc,
-                          unsigned src_x, unsigned src_y, unsigned src_z,
-                          unsigned w, unsigned h)
+                          unsigned src_level,
+                          const struct pipe_box *src_box)
 {
    struct pipe_transfer *src_trans, *dst_trans;
    void *dst_map;
    const void *src_map;
    enum pipe_format src_format, dst_format;
+   unsigned w = src_box->width;
+   unsigned h = src_box->height;
 
    assert(src && dst);
    if (!src || !dst)
@@ -146,20 +164,18 @@ util_resource_copy_region(struct pipe_context *pipe,
    dst_format = dst->format;
 
    src_trans = pipe_get_transfer(pipe,
-				 src,
-				 subsrc.face,
-				 subsrc.level,
-				 src_z,
-				 PIPE_TRANSFER_READ,
-				 src_x, src_y, w, h);
+                                 src,
+                                 src_level,
+                                 src_box->z,
+                                 PIPE_TRANSFER_READ,
+                                 src_box->x, src_box->y, w, h);
 
    dst_trans = pipe_get_transfer(pipe,
-				 dst,
-				 subdst.face,
-				 subdst.level,
-				 src_z,
-				 PIPE_TRANSFER_WRITE,
-				 dst_x, dst_y, w, h);
+                                 dst,
+                                 dst_level,
+                                 dst_z,
+                                 PIPE_TRANSFER_WRITE,
+                                 dst_x, dst_y, w, h);
 
    assert(util_format_get_blocksize(dst_format) == util_format_get_blocksize(src_format));
    assert(util_format_get_blockwidth(dst_format) == util_format_get_blockwidth(src_format));
@@ -216,14 +232,13 @@ util_clear_render_target(struct pipe_context *pipe,
    assert(dst->texture);
    if (!dst->texture)
       return;
-
+   /* XXX: should handle multiple layers */
    dst_trans = pipe_get_transfer(pipe,
-				 dst->texture,
-				 dst->face,
-				 dst->level,
-				 dst->zslice,
-				 PIPE_TRANSFER_WRITE,
-				 dstx, dsty, width, height);
+                                 dst->texture,
+                                 dst->u.tex.level,
+                                 dst->u.tex.first_layer,
+                                 PIPE_TRANSFER_WRITE,
+                                 dstx, dsty, width, height);
 
    dst_map = pipe->transfer_map(pipe, dst_trans);
 
@@ -271,9 +286,8 @@ util_clear_depth_stencil(struct pipe_context *pipe,
       return;
    dst_trans = pipe_get_transfer(pipe,
                                  dst->texture,
-                                 dst->face,
-                                 dst->level,
-                                 dst->zslice,
+                                 dst->u.tex.level,
+                                 dst->u.tex.first_layer,
                                  (need_rmw ? PIPE_TRANSFER_READ_WRITE :
                                      PIPE_TRANSFER_WRITE),
                                  dstx, dsty, width, height);
diff --git a/src/gallium/auxiliary/util/u_surface.h b/src/gallium/auxiliary/util/u_surface.h
index 6cd12af3a8..6a7cc82b05 100644
--- a/src/gallium/auxiliary/util/u_surface.h
+++ b/src/gallium/auxiliary/util/u_surface.h
@@ -33,8 +33,18 @@
 #include "pipe/p_state.h"
 
 
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+
+extern void
+u_surface_default_template(struct pipe_surface *view,
+                           const struct pipe_resource *texture,
+                           unsigned bind);
+
 extern boolean
-util_create_rgba_surface(struct pipe_screen *screen,
+util_create_rgba_surface(struct pipe_context *ctx,
                          uint width, uint height, uint bind,
                          struct pipe_resource **textureOut,
                          struct pipe_surface **surfaceOut);
@@ -49,12 +59,11 @@ util_destroy_rgba_surface(struct pipe_resource *texture,
 extern void
 util_resource_copy_region(struct pipe_context *pipe,
                           struct pipe_resource *dst,
-                          struct pipe_subresource subdst,
+                          unsigned dst_level,
                           unsigned dst_x, unsigned dst_y, unsigned dst_z,
                           struct pipe_resource *src,
-                          struct pipe_subresource subsrc,
-                          unsigned src_x, unsigned src_y, unsigned src_z,
-                          unsigned w, unsigned h);
+                          unsigned src_level,
+                          const struct pipe_box *src_box);
 
 extern void
 util_clear_render_target(struct pipe_context *pipe,
@@ -73,4 +82,9 @@ util_clear_depth_stencil(struct pipe_context *pipe,
                          unsigned width, unsigned height);
 
 
+#ifdef __cplusplus
+}
+#endif
+
+
 #endif /* U_SURFACE_H */
diff --git a/src/gallium/auxiliary/util/u_surfaces.c b/src/gallium/auxiliary/util/u_surfaces.c
index 404e121995..b0cfec2a82 100644
--- a/src/gallium/auxiliary/util/u_surfaces.c
+++ b/src/gallium/auxiliary/util/u_surfaces.c
@@ -29,8 +29,11 @@
 #include "util/u_inlines.h"
 #include "util/u_memory.h"
 
-struct pipe_surface *
-util_surfaces_do_get(struct util_surfaces *us, unsigned surface_struct_size, struct pipe_screen *pscreen, struct pipe_resource *pt, unsigned face, unsigned level, unsigned zslice, unsigned flags)
+boolean
+util_surfaces_do_get(struct util_surfaces *us, unsigned surface_struct_size,
+                     struct pipe_context *ctx, struct pipe_resource *pt,
+                     unsigned level, unsigned layer, unsigned flags,
+                     struct pipe_surface **res)
 {
    struct pipe_surface *ps;
 
@@ -39,7 +42,7 @@ util_surfaces_do_get(struct util_surfaces *us, unsigned surface_struct_size, str
       if(!us->u.hash)
          us->u.hash = cso_hash_create();
 
-      ps = cso_hash_iter_data(cso_hash_find(us->u.hash, ((zslice + face) << 8) | level));
+      ps = cso_hash_iter_data(cso_hash_find(us->u.hash, (layer << 8) | level));
    }
    else
    {
@@ -48,25 +51,29 @@ util_surfaces_do_get(struct util_surfaces *us, unsigned surface_struct_size, str
       ps = us->u.array[level];
    }
 
-   if(ps)
+   if(ps && ps->context == ctx)
    {
       p_atomic_inc(&ps->reference.count);
-      return ps;
+      *res = ps;
+      return FALSE;
    }
 
    ps = (struct pipe_surface *)CALLOC(1, surface_struct_size);
    if(!ps)
-      return NULL;
+   {
+      *res = NULL;
+      return FALSE;
+   }
 
-   pipe_surface_init(ps, pt, face, level, zslice, flags);
-   ps->offset = ~0;
+   pipe_surface_init(ctx, ps, pt, level, layer, flags);
 
    if(pt->target == PIPE_TEXTURE_3D || pt->target == PIPE_TEXTURE_CUBE)
-      cso_hash_insert(us->u.hash, ((zslice + face) << 8) | level, ps);
+      cso_hash_insert(us->u.hash, (layer << 8) | level, ps);
    else
       us->u.array[level] = ps;
 
-   return ps;
+   *res = ps;
+   return TRUE;
 }
 
 void
@@ -75,10 +82,10 @@ util_surfaces_do_detach(struct util_surfaces *us, struct pipe_surface *ps)
    struct pipe_resource *pt = ps->texture;
    if(pt->target == PIPE_TEXTURE_3D || pt->target == PIPE_TEXTURE_CUBE)
    {    /* or 2D array */
-      cso_hash_erase(us->u.hash, cso_hash_find(us->u.hash, ((ps->zslice + ps->face) << 8) | ps->level));
+      cso_hash_erase(us->u.hash, cso_hash_find(us->u.hash, (ps->u.tex.first_layer << 8) | ps->u.tex.level));
    }
    else
-      us->u.array[ps->level] = 0;
+      us->u.array[ps->u.tex.level] = 0;
 }
 
 void
diff --git a/src/gallium/auxiliary/util/u_surfaces.h b/src/gallium/auxiliary/util/u_surfaces.h
index 17d8a5d3a5..9581feda7c 100644
--- a/src/gallium/auxiliary/util/u_surfaces.h
+++ b/src/gallium/auxiliary/util/u_surfaces.h
@@ -42,33 +42,42 @@ struct util_surfaces
    } u;
 };
 
-struct pipe_surface *util_surfaces_do_get(struct util_surfaces *us, unsigned surface_struct_size, struct pipe_screen *pscreen, struct pipe_resource *pt, unsigned face, unsigned level, unsigned zslice, unsigned flags);
+/* Return value indicates if the pipe surface result is new */
+boolean
+util_surfaces_do_get(struct util_surfaces *us, unsigned surface_struct_size,
+                     struct pipe_context *ctx, struct pipe_resource *pt,
+                     unsigned level, unsigned layer, unsigned flags,
+                     struct pipe_surface **res);
 
 /* fast inline path for the very common case */
-static INLINE struct pipe_surface *
-util_surfaces_get(struct util_surfaces *us, unsigned surface_struct_size, struct pipe_screen *pscreen, struct pipe_resource *pt, unsigned face, unsigned level, unsigned zslice, unsigned flags)
+static INLINE boolean
+util_surfaces_get(struct util_surfaces *us, unsigned surface_struct_size,
+                  struct pipe_context *ctx, struct pipe_resource *pt,
+                  unsigned level, unsigned layer, unsigned flags,
+                  struct pipe_surface **res)
 {
    if(likely((pt->target == PIPE_TEXTURE_2D || pt->target == PIPE_TEXTURE_RECT) && us->u.array))
    {
       struct pipe_surface *ps = us->u.array[level];
-      if(ps)
+      if(ps && ps->context == ctx)
       {
 	 p_atomic_inc(&ps->reference.count);
-	 return ps;
+	 *res = ps;
+	 return FALSE;
       }
    }
 
-   return util_surfaces_do_get(us, surface_struct_size, pscreen, pt, face, level, zslice, flags);
+   return util_surfaces_do_get(us, surface_struct_size, ctx, pt, level, layer, flags, res);
 }
 
 static INLINE struct pipe_surface *
-util_surfaces_peek(struct util_surfaces *us, struct pipe_resource *pt, unsigned face, unsigned level, unsigned zslice)
+util_surfaces_peek(struct util_surfaces *us, struct pipe_resource *pt, unsigned level, unsigned layer)
 {
    if(!us->u.pv)
       return 0;
 
    if(unlikely(pt->target == PIPE_TEXTURE_3D || pt->target == PIPE_TEXTURE_CUBE))
-      return cso_hash_iter_data(cso_hash_find(us->u.hash, ((zslice + face) << 8) | level));
+      return cso_hash_iter_data(cso_hash_find(us->u.hash, (layer << 8) | level));
    else
       return us->u.array[level];
 }
@@ -80,7 +89,7 @@ util_surfaces_detach(struct util_surfaces *us, struct pipe_surface *ps)
 {
    if(likely(ps->texture->target == PIPE_TEXTURE_2D || ps->texture->target == PIPE_TEXTURE_RECT))
    {
-      us->u.array[ps->level] = 0;
+      us->u.array[ps->u.tex.level] = 0;
       return;
    }
 
diff --git a/src/gallium/auxiliary/util/u_transfer.c b/src/gallium/auxiliary/util/u_transfer.c
index 69f6fab950..e2828cfd99 100644
--- a/src/gallium/auxiliary/util/u_transfer.c
+++ b/src/gallium/auxiliary/util/u_transfer.c
@@ -8,22 +8,24 @@
  * pointer.  XXX: strides??
  */
 void u_default_transfer_inline_write( struct pipe_context *pipe,
-				      struct pipe_resource *resource,
-				      struct pipe_subresource sr,
-				      unsigned usage,
-				      const struct pipe_box *box,
-				      const void *data,
-				      unsigned stride,
-				      unsigned slice_stride)
+                                      struct pipe_resource *resource,
+                                      unsigned level,
+                                      unsigned usage,
+                                      const struct pipe_box *box,
+                                      const void *data,
+                                      unsigned stride,
+                                      unsigned layer_stride)
 {
    struct pipe_transfer *transfer = NULL;
    uint8_t *map = NULL;
-
-   transfer = pipe->get_transfer(pipe, 
-				 resource,
-				 sr,
-				 usage,
-				 box );
+   const uint8_t *src_data = data;
+   unsigned i;
+
+   transfer = pipe->get_transfer(pipe,
+                                 resource,
+                                 level,
+                                 usage,
+                                 box );
    if (transfer == NULL)
       goto out;
 
@@ -31,17 +33,19 @@ void u_default_transfer_inline_write( struct pipe_context *pipe,
    if (map == NULL)
       goto out;
 
-   assert(box->depth == 1);	/* XXX: fix me */
-   
-   util_copy_rect(map,
-		  resource->format,
-		  transfer->stride, /* bytes */
-		  0, 0,
-		  box->width,
-		  box->height,
-		  data,
-		  stride,       /* bytes */
-		  0, 0);
+   for (i = 0; i < box->depth; i++) {
+      util_copy_rect(map,
+                     resource->format,
+                     transfer->stride, /* bytes */
+                     0, 0,
+                     box->width,
+                     box->height,
+                     src_data,
+                     stride,       /* bytes */
+                     0, 0);
+      map += transfer->layer_stride;
+      src_data += layer_stride;
+   }
 
 out:
    if (map)
@@ -53,8 +57,8 @@ out:
 
 
 boolean u_default_resource_get_handle(struct pipe_screen *screen,
-				      struct pipe_resource *resource,
-				      struct winsys_handle *handle)
+                                      struct pipe_resource *resource,
+                                      struct winsys_handle *handle)
 {
    return FALSE;
 }
@@ -62,32 +66,32 @@ boolean u_default_resource_get_handle(struct pipe_screen *screen,
 
 
 void u_default_transfer_flush_region( struct pipe_context *pipe,
-				      struct pipe_transfer *transfer,
-				      const struct pipe_box *box)
+                                      struct pipe_transfer *transfer,
+                                      const struct pipe_box *box)
 {
    /* This is a no-op implementation, nothing to do.
     */
 }
 
 unsigned u_default_is_resource_referenced( struct pipe_context *pipe,
-					   struct pipe_resource *resource,
-					unsigned face, unsigned level)
+                                           struct pipe_resource *resource,
+                                           unsigned level, int layer)
 {
    return 0;
 }
 
 struct pipe_transfer * u_default_get_transfer(struct pipe_context *context,
-					      struct pipe_resource *resource,
-					      struct pipe_subresource sr,
-					      unsigned usage,
-					      const struct pipe_box *box)
+                                              struct pipe_resource *resource,
+                                              unsigned level,
+                                              unsigned usage,
+                                              const struct pipe_box *box)
 {
    struct pipe_transfer *transfer = CALLOC_STRUCT(pipe_transfer);
    if (transfer == NULL)
       return NULL;
 
    transfer->resource = resource;
-   transfer->sr = sr;
+   transfer->level = level;
    transfer->usage = usage;
    transfer->box = *box;
 
@@ -98,12 +102,12 @@ struct pipe_transfer * u_default_get_transfer(struct pipe_context *context,
 }
 
 void u_default_transfer_unmap( struct pipe_context *pipe,
-			      struct pipe_transfer *transfer )
+                               struct pipe_transfer *transfer )
 {
 }
 
 void u_default_transfer_destroy(struct pipe_context *pipe,
-				struct pipe_transfer *transfer)
+                                struct pipe_transfer *transfer)
 {
    FREE(transfer);
 }
diff --git a/src/gallium/auxiliary/util/u_transfer.h b/src/gallium/auxiliary/util/u_transfer.h
index e3a38730f2..3412e13c3c 100644
--- a/src/gallium/auxiliary/util/u_transfer.h
+++ b/src/gallium/auxiliary/util/u_transfer.h
@@ -11,37 +11,37 @@ struct pipe_context;
 struct winsys_handle;
 
 boolean u_default_resource_get_handle(struct pipe_screen *screen,
-				      struct pipe_resource *resource,
-				      struct winsys_handle *handle);
+                                      struct pipe_resource *resource,
+                                      struct winsys_handle *handle);
 
 void u_default_transfer_inline_write( struct pipe_context *pipe,
-			      struct pipe_resource *resource,
-			      struct pipe_subresource sr,
-			      unsigned usage,
-			      const struct pipe_box *box,
-			      const void *data,
-			      unsigned stride,
-			      unsigned slice_stride);
+                                      struct pipe_resource *resource,
+                                      unsigned level,
+                                      unsigned usage,
+                                      const struct pipe_box *box,
+                                      const void *data,
+                                      unsigned stride,
+                                      unsigned layer_stride);
 
 void u_default_transfer_flush_region( struct pipe_context *pipe,
-				      struct pipe_transfer *transfer,
-				      const struct pipe_box *box);
+                                      struct pipe_transfer *transfer,
+                                      const struct pipe_box *box);
 
 unsigned u_default_is_resource_referenced( struct pipe_context *pipe,
-					   struct pipe_resource *resource,
-					   unsigned face, unsigned level);
+                                           struct pipe_resource *resource,
+                                           unsigned level, int layer);
 
 struct pipe_transfer * u_default_get_transfer(struct pipe_context *context,
-					      struct pipe_resource *resource,
-					      struct pipe_subresource sr,
-					      unsigned usage,
-					      const struct pipe_box *box);
+                                              struct pipe_resource *resource,
+                                              unsigned level,
+                                              unsigned usage,
+                                              const struct pipe_box *box);
 
 void u_default_transfer_unmap( struct pipe_context *pipe,
-			       struct pipe_transfer *transfer );
+                               struct pipe_transfer *transfer );
 
 void u_default_transfer_destroy(struct pipe_context *pipe,
-				struct pipe_transfer *transfer);
+                                struct pipe_transfer *transfer);
 
 
 
@@ -51,43 +51,43 @@ void u_default_transfer_destroy(struct pipe_context *pipe,
 struct u_resource_vtbl {
 
    boolean (*resource_get_handle)(struct pipe_screen *,
-				  struct pipe_resource *tex,
-				  struct winsys_handle *handle);
+                                  struct pipe_resource *tex,
+                                  struct winsys_handle *handle);
 
    void (*resource_destroy)(struct pipe_screen *,
-			    struct pipe_resource *pt);
+                            struct pipe_resource *pt);
 
    unsigned (*is_resource_referenced)(struct pipe_context *pipe,
-				      struct pipe_resource *texture,
-				      unsigned face, unsigned level);
+                                      struct pipe_resource *texture,
+                                      unsigned level, int layer);
 
    struct pipe_transfer *(*get_transfer)(struct pipe_context *,
-					 struct pipe_resource *resource,
-					 struct pipe_subresource,
-					 unsigned usage,
-					 const struct pipe_box *);
+                                         struct pipe_resource *resource,
+                                         unsigned level,
+                                         unsigned usage,
+                                         const struct pipe_box *);
 
    void (*transfer_destroy)(struct pipe_context *,
-			    struct pipe_transfer *);
+                            struct pipe_transfer *);
 
    void *(*transfer_map)( struct pipe_context *,
-			  struct pipe_transfer *transfer );
+                          struct pipe_transfer *transfer );
 
    void (*transfer_flush_region)( struct pipe_context *,
-				  struct pipe_transfer *transfer,
-				  const struct pipe_box *);
+                                  struct pipe_transfer *transfer,
+                                  const struct pipe_box *);
 
    void (*transfer_unmap)( struct pipe_context *,
-			   struct pipe_transfer *transfer );
+   struct pipe_transfer *transfer );
 
    void (*transfer_inline_write)( struct pipe_context *pipe,
-				  struct pipe_resource *resource,
-				  struct pipe_subresource sr,
-				  unsigned usage,
-				  const struct pipe_box *box,
-				  const void *data,
-				  unsigned stride,
-				  unsigned slice_stride);
+                                  struct pipe_resource *resource,
+                                  unsigned level,
+                                  unsigned usage,
+                                  const struct pipe_box *box,
+                                  const void *data,
+                                  unsigned stride,
+                                  unsigned layer_stride);
 };
 
 
@@ -98,43 +98,43 @@ struct u_resource {
 
 
 boolean u_resource_get_handle_vtbl(struct pipe_screen *screen,
-			      struct pipe_resource *resource,
-			      struct winsys_handle *handle);
+                                   struct pipe_resource *resource,
+                                   struct winsys_handle *handle);
 
 void u_resource_destroy_vtbl(struct pipe_screen *screen,
-			struct pipe_resource *resource);
+                             struct pipe_resource *resource);
 
 unsigned u_is_resource_referenced_vtbl( struct pipe_context *pipe,
-					struct pipe_resource *resource,
-					unsigned face, unsigned level);
+                                        struct pipe_resource *resource,
+                                        unsigned level, int layer);
 
 struct pipe_transfer *u_get_transfer_vtbl(struct pipe_context *context,
-				     struct pipe_resource *resource,
-				     struct pipe_subresource sr,
-				     unsigned usage,
-				     const struct pipe_box *box);
+                                          struct pipe_resource *resource,
+                                          unsigned level,
+                                          unsigned usage,
+                                          const struct pipe_box *box);
 
 void u_transfer_destroy_vtbl(struct pipe_context *pipe,
-			struct pipe_transfer *transfer);
+                             struct pipe_transfer *transfer);
 
 void *u_transfer_map_vtbl( struct pipe_context *pipe,
-		      struct pipe_transfer *transfer );
+                           struct pipe_transfer *transfer );
 
 void u_transfer_flush_region_vtbl( struct pipe_context *pipe,
-			      struct pipe_transfer *transfer,
-			      const struct pipe_box *box);
+                                   struct pipe_transfer *transfer,
+                                   const struct pipe_box *box);
 
 void u_transfer_unmap_vtbl( struct pipe_context *rm_ctx,
-		       struct pipe_transfer *transfer );
+                            struct pipe_transfer *transfer );
 
 void u_transfer_inline_write_vtbl( struct pipe_context *rm_ctx,
-				   struct pipe_resource *resource,
-				   struct pipe_subresource sr,
-				   unsigned usage,
-				   const struct pipe_box *box,
-				   const void *data,
-				   unsigned stride,
-				   unsigned slice_stride);
+                                   struct pipe_resource *resource,
+                                   unsigned level,
+                                   unsigned usage,
+                                   const struct pipe_box *box,
+                                   const void *data,
+                                   unsigned stride,
+                                   unsigned layer_stride);
 
 
 
diff --git a/src/gallium/docs/d3d11ddi.txt b/src/gallium/docs/d3d11ddi.txt
index f8155c828b..11e7719089 100644
--- a/src/gallium/docs/d3d11ddi.txt
+++ b/src/gallium/docs/d3d11ddi.txt
@@ -162,8 +162,8 @@ CreateDepthStencilState -> create_depth_stencil_alpha_state
 	+ Gallium has per-face writemask/valuemasks, D3D11 uses the same value for back and front
 	+ Gallium supports the alpha test, which D3D11 lacks
 
-CreateDepthStencilView -> get_tex_surface
-CreateRenderTargetView -> get_tex_surface
+CreateDepthStencilView -> create_surface
+CreateRenderTargetView -> create_surface
 	! Gallium merges depthstencil and rendertarget views into pipe_surface, which also doubles as a 2D surface abstraction
 	- lack of texture array support
 	- lack of render-to-buffer support
@@ -221,7 +221,6 @@ CreateResource -> texture_create or buffer_create
 	! D3D11 specifies mapping flags (i.e. read/write/discard);:it's unclear what they are used for here
 	- D3D11 supports odd things in the D3D10_DDI_RESOURCE_MISC_FLAG enum (D3D10_DDI_RESOURCE_MISC_DISCARD_ON_PRESENT, D3D11_DDI_RESOURCE_MISC_BUFFER_ALLOW_RAW_VIEWS, D3D11_DDI_RESOURCE_MISC_BUFFER_STRUCTURED)
 	- Gallium does not support indirect draw call parameter buffers
-	- Gallium lacks array textures
 	! D3D11 supports specifying hardware modes and other stuff here for scanout resources
 	+ Gallium allows specifying minimum buffer alignment
 	! D3D11 implements cube maps as 2D array textures
diff --git a/src/gallium/docs/source/context.rst b/src/gallium/docs/source/context.rst
index e09a1304c4..c33cf7c573 100644
--- a/src/gallium/docs/source/context.rst
+++ b/src/gallium/docs/source/context.rst
@@ -84,7 +84,14 @@ in the result register. For example, ``swizzle_r`` specifies what is going to be
 placed in first component of result register.
 
 The ``first_level`` and ``last_level`` fields of sampler view template specify
-the LOD range the texture is going to be constrained to.
+the LOD range the texture is going to be constrained to. Note that these
+values are in addition to the respective min_lod, max_lod values in the
+pipe_sampler_state (that is if min_lod is 2.0, and first_level 3, the first mip
+level used for sampling from the resource is effectively the fifth).
+
+The ``first_layer`` and ``last_layer`` fields specify the layer range the
+texture is going to be constrained to. Similar to the LOD range, this is added
+to the array index which is used for sampling.
 
 * ``set_fragment_sampler_views`` binds an array of sampler views to
   fragment shader stage. Every binding point acquires a reference
@@ -103,6 +110,22 @@ the LOD range the texture is going to be constrained to.
 * ``sampler_view_destroy`` destroys a sampler view and releases its reference
   to associated texture.
 
+Surfaces
+^^^^^^^^
+
+These are the means to use resources as color render targets or depthstencil
+attachments. To create one, specify the mip level, the range of layers, and
+the bind flags (either PIPE_BIND_DEPTH_STENCIL or PIPE_BIND_RENDER_TARGET).
+Note that layer values are in addition to what is indicated by the geometry
+shader output variable XXX_FIXME (that is if first_layer is 3 and geometry
+shader indicates index 2, the 5th layer of the resource will be used). These
+first_layer and last_layer parameters will only be used for 1d array, 2d array,
+cube, and 3d textures otherwise they are 0.
+
+* ``create_surface`` creates a new surface.
+
+* ``surface_destroy`` destroys a surface and releases its reference to the
+  associated resource.
 
 Clearing
 ^^^^^^^^
@@ -118,8 +141,7 @@ used by GL), and always clears the whole surfaces (no scissoring as used by
 GL clear or explicit rectangles like d3d9 uses). It can, however, also clear
 only depth or stencil in a combined depth/stencil surface, if the driver
 supports PIPE_CAP_DEPTHSTENCIL_CLEAR_SEPARATE.
-If a surface includes several layers/slices (XXX: not yet...) then all layers
-will be cleared.
+If a surface includes several layers then all layers will be cleared.
 
 ``clear_render_target`` clears a single color rendertarget with the specified
 color value. While it is only possible to clear one surface at a time (which can
@@ -271,12 +293,12 @@ These methods operate directly on ``pipe_resource`` objects, and stand
 apart from any 3D state in the context.  Blitting functionality may be
 moved to a separate abstraction at some point in the future.
 
-``resource_copy_region`` blits a region of a subresource of a resource to a
-region of another subresource of a resource, provided that both resources have
-the same format, or compatible formats, i.e., formats for which copying the
-bytes from the source resource unmodified to the destination resource will
-achieve the same effect of a textured quad blitter. The source and destination
-may be the same resource, but overlapping blits are not permitted.
+``resource_copy_region`` blits a region of a resource to a region of another
+resource, provided that both resources have the same format, or compatible
+formats, i.e., formats for which copying the bytes from the source resource
+unmodified to the destination resource will achieve the same effect of a
+textured quad blitter.. The source and destination may be the same resource,
+but overlapping blits are not permitted.
 
 ``resource_resolve`` resolves a multisampled resource into a non-multisampled
 one. Formats and dimensions must match. This function must be present if a driver
diff --git a/src/gallium/docs/source/glossary.rst b/src/gallium/docs/source/glossary.rst
index acde56eafc..c749d0c955 100644
--- a/src/gallium/docs/source/glossary.rst
+++ b/src/gallium/docs/source/glossary.rst
@@ -22,6 +22,14 @@ Glossary
       Level of Detail. Also spelled "LoD." The value that determines when the
       switches between mipmaps occur during texture sampling.
 
+   layer
+      This term is used as the name of the "3rd coordinate" of a resource.
+      3D textures have zslices, cube maps have faces, 1D and 2D array textures
+      have array members (other resources do not have multiple layers).
+      Since the functions only take one parameter no matter what type of
+      resource is used, use the term "layer" instead of a resource type
+      specific one.
+
    GLSL
       GL Shading Language. The official, common high-level shader language used
       in GL 2.0 and above.
diff --git a/src/gallium/docs/source/screen.rst b/src/gallium/docs/source/screen.rst
index e3ef49c862..09edbaa673 100644
--- a/src/gallium/docs/source/screen.rst
+++ b/src/gallium/docs/source/screen.rst
@@ -258,25 +258,33 @@ resource_create
 Create a new resource from a template.
 The following fields of the pipe_resource must be specified in the template:
 
-target
+**target** one of the pipe_texture_target enums.
+Note that PIPE_BUFFER and PIPE_TEXTURE_X are not really fundamentally different.
+Modern APIs allow using buffers as shader resources.
 
-format
+**format** one of the pipe_format enums.
 
-width0
+**width0** the width of the base mip level of the texture or size of the buffer.
 
-height0
+**height0** the height of the base mip level of the texture
+(1 for 1D or 1D array textures).
 
-depth0
+**depth0** the depth of the base mip level of the texture
+(1 for everything else).
 
-last_level
+**array_size the array size for 1D and 2D array textures.
+For cube maps this must be 6, for other textures 1.
 
-nr_samples
+**last_level** the last mip map level present.
 
-usage
+**nr_samples** the nr of msaa samples. 0 (or 1) specifies a resource
+which isn't multisampled.
 
-bind
+**usage** one of the PIPE_USAGE flags.
 
-flags
+**bind** bitmask of the PIPE_BIND flags.
+
+**flags** bitmask of PIPE_RESOURCE_FLAG flags.
 
 
 
diff --git a/src/gallium/docs/source/tgsi.rst b/src/gallium/docs/source/tgsi.rst
index d99ed7c6d6..7eb6bd05d7 100644
--- a/src/gallium/docs/source/tgsi.rst
+++ b/src/gallium/docs/source/tgsi.rst
@@ -577,17 +577,45 @@ This instruction replicates its result.
 
 .. opcode:: TEX - Texture Lookup
 
-  TBD
+.. math::
+
+  coord = src0
+
+  bias = 0.0
+
+  dst = texture_sample(unit, coord, bias)
 
 
 .. opcode:: TXD - Texture Lookup with Derivatives
 
-  TBD
+.. math::
+
+  coord = src0
+
+  ddx = src1
+
+  ddy = src2
+
+  bias = 0.0
+
+  dst = texture_sample_deriv(unit, coord, bias, ddx, ddy)
 
 
 .. opcode:: TXP - Projective Texture Lookup
 
-  TBD
+.. math::
+
+  coord.x = src0.x / src.w
+
+  coord.y = src0.y / src.w
+
+  coord.z = src0.z / src.w
+
+  coord.w = src0.w
+
+  bias = 0.0
+
+  dst = texture_sample(unit, coord, bias)
 
 
 .. opcode:: UP2H - Unpack Two 16-Bit Floats
@@ -729,7 +757,19 @@ This instruction replicates its result.
 
 .. opcode:: TXB - Texture Lookup With Bias
 
-  TBD
+.. math::
+
+  coord.x = src.x
+
+  coord.y = src.y
+
+  coord.z = src.z
+
+  coord.w = 1.0
+
+  bias = src.z
+
+  dst = texture_sample(unit, coord, bias)
 
 
 .. opcode:: NRM - 3-component Vector Normalise
@@ -767,9 +807,21 @@ This instruction replicates its result.
   dst = src0.x \times src1.x + src0.y \times src1.y
 
 
-.. opcode:: TXL - Texture Lookup With LOD
+.. opcode:: TXL - Texture Lookup With explicit LOD
 
-  TBD
+.. math::
+
+  coord.x = src0.x
+
+  coord.y = src0.y
+
+  coord.z = src0.z
+
+  coord.w = 1.0
+
+  lod = src0.w
+
+  dst = texture_sample(unit, coord, lod)
 
 
 .. opcode:: BRK - Break
diff --git a/src/gallium/drivers/cell/ppu/cell_context.c b/src/gallium/drivers/cell/ppu/cell_context.c
index 143eca848f..b6b3a700cd 100644
--- a/src/gallium/drivers/cell/ppu/cell_context.c
+++ b/src/gallium/drivers/cell/ppu/cell_context.c
@@ -100,8 +100,8 @@ static const struct debug_named_value cell_debug_flags[] = {
 
 static unsigned int
 cell_is_resource_referenced( struct pipe_context *pipe,
-			    struct pipe_resource *texture,
-			    unsigned face, unsigned level)
+                             struct pipe_resource *texture,
+                             unsigned level, int layer)
 {
    /**
     * FIXME: Optimize.
diff --git a/src/gallium/drivers/cell/ppu/cell_texture.c b/src/gallium/drivers/cell/ppu/cell_texture.c
index b3042df779..946a7050e5 100644
--- a/src/gallium/drivers/cell/ppu/cell_texture.c
+++ b/src/gallium/drivers/cell/ppu/cell_texture.c
@@ -304,47 +304,34 @@ untwiddle_image_uint(uint w, uint h, uint tile_size, uint *dst,
 
 
 static struct pipe_surface *
-cell_get_tex_surface(struct pipe_screen *screen,
-                     struct pipe_resource *pt,
-                     unsigned face, unsigned level, unsigned zslice,
-                     unsigned usage)
+cell_create_surface(struct pipe_context *ctx,
+                    struct pipe_resource *pt,
+                    const struct pipe_surface *surf_tmpl)
 {
    struct cell_resource *ct = cell_resource(pt);
    struct pipe_surface *ps;
 
+   assert(surf_tmpl->u.tex.first_layer == surf_tmpl->u.tex.last_layer);
    ps = CALLOC_STRUCT(pipe_surface);
    if (ps) {
       pipe_reference_init(&ps->reference, 1);
       pipe_resource_reference(&ps->texture, pt);
-      ps->format = pt->format;
-      ps->width = u_minify(pt->width0, level);
-      ps->height = u_minify(pt->height0, level);
-      ps->offset = ct->level_offset[level];
+      ps->format = surf_tmpl->format;
+      ps->context = ctx;
+      ps->width = u_minify(pt->width0, surf_tmpl->u.tex.level);
+      ps->height = u_minify(pt->height0, surf_tmpl->u.tex.level);
       /* XXX may need to override usage flags (see sp_texture.c) */
-      ps->usage = usage;
-      ps->face = face;
-      ps->level = level;
-      ps->zslice = zslice;
-
-      if (pt->target == PIPE_TEXTURE_CUBE) {
-         unsigned h_tile = align(ps->height, TILE_SIZE);
-         ps->offset += face * util_format_get_nblocksy(ps->format, h_tile) * ct->stride[level];
-      }
-      else if (pt->target == PIPE_TEXTURE_3D) {
-         unsigned h_tile = align(ps->height, TILE_SIZE);
-         ps->offset += zslice * util_format_get_nblocksy(ps->format, h_tile) * ct->stride[level];
-      }
-      else {
-         assert(face == 0);
-         assert(zslice == 0);
-      }
+      ps->usage = surf_tmpl->usage;
+      ps->u.tex.level = surf_tmpl->u.tex.level;
+      ps->u.tex.first_layer = surf_tmpl->u.tex.first_layer;
+      ps->u.tex.last_layer = surf_tmpl->u.tex.last_layer;
    }
    return ps;
 }
 
 
 static void 
-cell_tex_surface_destroy(struct pipe_surface *surf)
+cell_surface_destroy(struct pipe_context *ctx, struct pipe_surface *surf)
 {
    pipe_resource_reference(&surf->texture, NULL);
    FREE(surf);
@@ -358,44 +345,39 @@ cell_tex_surface_destroy(struct pipe_surface *surf)
  */
 static struct pipe_transfer *
 cell_get_transfer(struct pipe_context *ctx,
-		  struct pipe_resource *resource,
-		  struct pipe_subresource sr,
-		  unsigned usage,
-		  const struct pipe_box *box)
+                  struct pipe_resource *resource,
+                  unsigned level,
+                  unsigned usage,
+                  const struct pipe_box *box)
 {
    struct cell_resource *ct = cell_resource(resource);
    struct cell_transfer *ctrans;
    enum pipe_format format = resource->format;
 
    assert(resource);
-   assert(sr.level <= resource->last_level);
+   assert(level <= resource->last_level);
 
    /* make sure the requested region is in the image bounds */
-   assert(box->x + box->width <= u_minify(resource->width0, sr.level));
-   assert(box->y + box->height <= u_minify(resource->height0, sr.level));
-   assert(box->z + box->depth <= u_minify(resource->depth0, sr.level));
+   assert(box->x + box->width <= u_minify(resource->width0, level));
+   assert(box->y + box->height <= u_minify(resource->height0, level));
+   assert(box->z + box->depth <= (u_minify(resource->depth0, level) + resource->array_size - 1));
 
    ctrans = CALLOC_STRUCT(cell_transfer);
    if (ctrans) {
       struct pipe_transfer *pt = &ctrans->base;
       pipe_resource_reference(&pt->resource, resource);
-      pt->sr = sr;
+      pt->level = level;
       pt->usage = usage;
       pt->box = *box;
-      pt->stride = ct->stride[sr.level];
+      pt->stride = ct->stride[level];
 
-      ctrans->offset = ct->level_offset[sr.level];
+      ctrans->offset = ct->level_offset[level];
 
-      if (resource->target == PIPE_TEXTURE_CUBE) {
-         unsigned h_tile = align(u_minify(resource->height0, sr.level), TILE_SIZE);
-         ctrans->offset += sr.face * util_format_get_nblocksy(format, h_tile) * pt->stride;
-      }
-      else if (resource->target == PIPE_TEXTURE_3D) {
-         unsigned h_tile = align(u_minify(resource->height0, sr.level), TILE_SIZE);
+      if (resource->target == PIPE_TEXTURE_CUBE || resource->target == PIPE_TEXTURE_3D) {
+         unsigned h_tile = align(u_minify(resource->height0, level), TILE_SIZE);
          ctrans->offset += box->z * util_format_get_nblocksy(format, h_tile) * pt->stride;
       }
       else {
-         assert(sr.face == 0);
          assert(box->z == 0);
       }
 
@@ -439,7 +421,7 @@ cell_transfer_map(struct pipe_context *ctx, struct pipe_transfer *transfer)
    /* Better test would be resource->is_linear
     */
    if (transfer->resource->target != PIPE_BUFFER) {
-      const uint level = ctrans->base.sr.level;
+      const uint level = ctrans->base.level;
       const uint texWidth = u_minify(pt->width0, level);
       const uint texHeight = u_minify(pt->height0, level);
       unsigned size;
@@ -500,7 +482,7 @@ cell_transfer_unmap(struct pipe_context *ctx,
    struct cell_transfer *ctrans = cell_transfer(transfer);
    struct pipe_resource *pt = transfer->resource;
    struct cell_resource *ct = cell_resource(pt);
-   const uint level = ctrans->base.sr.level;
+   const uint level = ctrans->base.level;
    const uint texWidth = u_minify(pt->width0, level);
    const uint texHeight = u_minify(pt->height0, level);
    const uint stride = ct->stride[level];
@@ -548,12 +530,13 @@ cell_transfer_unmap(struct pipe_context *ctx,
  */
 static void
 cell_flush_frontbuffer(struct pipe_screen *_screen,
-                       struct pipe_surface *surface,
+                       struct pipe_resource *resource,
+                       unsigned level, unsigned layer,
                        void *context_private)
 {
    struct cell_screen *screen = cell_screen(_screen);
    struct sw_winsys *winsys = screen->winsys;
-   struct cell_resource *ct = cell_resource(surface->texture);
+   struct cell_resource *ct = cell_resource(resource);
 
    if (!ct->dt)
       return;
@@ -564,10 +547,10 @@ cell_flush_frontbuffer(struct pipe_screen *_screen,
       unsigned *map = winsys->displaytarget_map(winsys, ct->dt,
                                                 (PIPE_TRANSFER_READ |
                                                  PIPE_TRANSFER_WRITE));
-      unsigned *src = (unsigned *)(ct->data + ct->level_offset[surface->level]);
+      unsigned *src = (unsigned *)(ct->data + ct->level_offset[level]);
 
-      untwiddle_image_uint(surface->width,
-                           surface->height,
+      untwiddle_image_uint(u_minify(resource->width0, level),
+                           u_minify(resource->height0, level),
                            TILE_SIZE,
                            map,
                            ct->dt_stride,
@@ -605,6 +588,7 @@ cell_user_buffer_create(struct pipe_screen *screen,
    buffer->base.width0 = bytes;
    buffer->base.height0 = 1;
    buffer->base.depth0 = 1;
+   buffer->base.array_size = 1;
    buffer->userBuffer = TRUE;
    buffer->data = ptr;
 
@@ -641,9 +625,6 @@ cell_init_screen_texture_funcs(struct pipe_screen *screen)
    screen->resource_get_handle = cell_resource_get_handle;
    screen->user_buffer_create = cell_user_buffer_create;
 
-   screen->get_tex_surface = cell_get_tex_surface;
-   screen->tex_surface_destroy = cell_tex_surface_destroy;
-
    screen->flush_frontbuffer = cell_flush_frontbuffer;
 }
 
@@ -657,4 +638,7 @@ cell_init_texture_transfer_funcs(struct cell_context *cell)
 
    cell->pipe.transfer_flush_region = u_default_transfer_flush_region;
    cell->pipe.transfer_inline_write = u_default_transfer_inline_write;
+
+   cell->pipe.create_surface = cell_create_surface;
+   cell->pipe.surface_destroy = cell_surface_destroy;
 }
diff --git a/src/gallium/drivers/failover/fo_context.c b/src/gallium/drivers/failover/fo_context.c
index ec3609291e..e4d289c8a4 100644
--- a/src/gallium/drivers/failover/fo_context.c
+++ b/src/gallium/drivers/failover/fo_context.c
@@ -89,14 +89,14 @@ static void failover_draw_vbo( struct pipe_context *pipe,
 
 static unsigned int
 failover_is_resource_referenced( struct pipe_context *_pipe,
-				 struct pipe_resource *resource,
-				 unsigned face, unsigned level)
+                                 struct pipe_resource *resource,
+                                 unsigned level, int layer)
 {
    struct failover_context *failover = failover_context( _pipe );
    struct pipe_context *pipe = (failover->mode == FO_HW) ?
       failover->hw : failover->sw;
 
-   return pipe->is_resource_referenced(pipe, resource, face, level);
+   return pipe->is_resource_referenced(pipe, resource, level, layer);
 }
 
 struct pipe_context *failover_create( struct pipe_context *hw,
@@ -137,10 +137,10 @@ struct pipe_context *failover_create( struct pipe_context *hw,
    failover->pipe.resource_copy_region = hw->resource_copy_region;
 
 #if 0
-   failover->pipe.texture_create = hw->texture_create;
-   failover->pipe.texture_destroy = hw->texture_destroy;
-   failover->pipe.get_tex_surface = hw->get_tex_surface;
-   failover->pipe.texture_update = hw->texture_update;
+   failover->pipe.resource_create = hw->resource_create;
+   failover->pipe.resource_destroy = hw->resource_destroy;
+   failover->pipe.create_surface = hw->create_surface;
+   failover->pipe.surface_destroy = hw->surface_destroy;
 #endif
 
    failover->pipe.flush = hw->flush;
diff --git a/src/gallium/drivers/galahad/glhd_context.c b/src/gallium/drivers/galahad/glhd_context.c
index 50f66079c2..a572ad22bd 100644
--- a/src/gallium/drivers/galahad/glhd_context.c
+++ b/src/gallium/drivers/galahad/glhd_context.c
@@ -662,17 +662,13 @@ galahad_set_index_buffer(struct pipe_context *_pipe,
 static void
 galahad_resource_copy_region(struct pipe_context *_pipe,
                               struct pipe_resource *_dst,
-                              struct pipe_subresource subdst,
+                              unsigned dst_level,
                               unsigned dstx,
                               unsigned dsty,
                               unsigned dstz,
                               struct pipe_resource *_src,
-                              struct pipe_subresource subsrc,
-                              unsigned srcx,
-                              unsigned srcy,
-                              unsigned srcz,
-                              unsigned width,
-                              unsigned height)
+                              unsigned src_level,
+                              const struct pipe_box *src_box)
 {
    struct galahad_context *glhd_pipe = galahad_context(_pipe);
    struct galahad_resource *glhd_resource_dst = galahad_resource(_dst);
@@ -689,17 +685,13 @@ galahad_resource_copy_region(struct pipe_context *_pipe,
 
    pipe->resource_copy_region(pipe,
                               dst,
-                              subdst,
+                              dst_level,
                               dstx,
                               dsty,
                               dstz,
                               src,
-                              subsrc,
-                              srcx,
-                              srcy,
-                              srcz,
-                              width,
-                              height);
+                              src_level,
+                              src_box);
 }
 
 static void
@@ -781,8 +773,8 @@ galahad_flush(struct pipe_context *_pipe,
 static unsigned int
 galahad_is_resource_referenced(struct pipe_context *_pipe,
                                 struct pipe_resource *_resource,
-                                unsigned face,
-                                unsigned level)
+                                unsigned level,
+                                int layer)
 {
    struct galahad_context *glhd_pipe = galahad_context(_pipe);
    struct galahad_resource *glhd_resource = galahad_resource(_resource);
@@ -791,8 +783,8 @@ galahad_is_resource_referenced(struct pipe_context *_pipe,
 
    return pipe->is_resource_referenced(pipe,
                                        resource,
-                                       face,
-                                       level);
+                                       level,
+                                       layer);
 }
 
 static struct pipe_sampler_view *
@@ -823,10 +815,40 @@ galahad_context_sampler_view_destroy(struct pipe_context *_pipe,
                                  galahad_sampler_view(_view));
 }
 
+static struct pipe_surface *
+galahad_context_create_surface(struct pipe_context *_pipe,
+                                struct pipe_resource *_resource,
+                                const struct pipe_surface *templ)
+{
+   struct galahad_context *glhd_context = galahad_context(_pipe);
+   struct galahad_resource *glhd_resource = galahad_resource(_resource);
+   struct pipe_context *pipe = glhd_context->pipe;
+   struct pipe_resource *resource = glhd_resource->resource;
+   struct pipe_surface *result;
+
+   result = pipe->create_surface(pipe,
+                                 resource,
+                                 templ);
+
+   if (result)
+      return galahad_surface_create(glhd_context, glhd_resource, result);
+   return NULL;
+}
+
+static void
+galahad_context_surface_destroy(struct pipe_context *_pipe,
+                                struct pipe_surface *_surface)
+{
+   galahad_surface_destroy(galahad_context(_pipe),
+                           galahad_surface(_surface));
+}
+
+
+
 static struct pipe_transfer *
 galahad_context_get_transfer(struct pipe_context *_context,
                               struct pipe_resource *_resource,
-                              struct pipe_subresource sr,
+                              unsigned level,
                               unsigned usage,
                               const struct pipe_box *box)
 {
@@ -838,7 +860,7 @@ galahad_context_get_transfer(struct pipe_context *_context,
 
    result = context->get_transfer(context,
                                   resource,
-                                  sr,
+                                  level,
                                   usage,
                                   box);
 
@@ -915,7 +937,7 @@ galahad_context_transfer_unmap(struct pipe_context *_context,
 static void
 galahad_context_transfer_inline_write(struct pipe_context *_context,
                                        struct pipe_resource *_resource,
-                                       struct pipe_subresource sr,
+                                       unsigned level,
                                        unsigned usage,
                                        const struct pipe_box *box,
                                        const void *data,
@@ -929,7 +951,7 @@ galahad_context_transfer_inline_write(struct pipe_context *_context,
 
    context->transfer_inline_write(context,
                                   resource,
-                                  sr,
+                                  level,
                                   usage,
                                   box,
                                   data,
@@ -1004,6 +1026,8 @@ galahad_context_create(struct pipe_screen *_screen, struct pipe_context *pipe)
    glhd_pipe->base.is_resource_referenced = galahad_is_resource_referenced;
    glhd_pipe->base.create_sampler_view = galahad_context_create_sampler_view;
    glhd_pipe->base.sampler_view_destroy = galahad_context_sampler_view_destroy;
+   glhd_pipe->base.create_surface = galahad_context_create_surface;
+   glhd_pipe->base.surface_destroy = galahad_context_surface_destroy;
    glhd_pipe->base.get_transfer = galahad_context_get_transfer;
    glhd_pipe->base.transfer_destroy = galahad_context_transfer_destroy;
    glhd_pipe->base.transfer_map = galahad_context_transfer_map;
diff --git a/src/gallium/drivers/galahad/glhd_objects.c b/src/gallium/drivers/galahad/glhd_objects.c
index 6c5a21ae70..b50d85655e 100644
--- a/src/gallium/drivers/galahad/glhd_objects.c
+++ b/src/gallium/drivers/galahad/glhd_objects.c
@@ -71,7 +71,8 @@ galahad_resource_destroy(struct galahad_resource *glhd_resource)
 
 
 struct pipe_surface *
-galahad_surface_create(struct galahad_resource *glhd_resource,
+galahad_surface_create(struct galahad_context *glhd_context,
+                        struct galahad_resource *glhd_resource,
                         struct pipe_surface *surface)
 {
    struct galahad_surface *glhd_surface;
@@ -100,10 +101,11 @@ error:
 }
 
 void
-galahad_surface_destroy(struct galahad_surface *glhd_surface)
+galahad_surface_destroy(struct galahad_context *glhd_context,
+                         struct galahad_surface *glhd_surface)
 {
    pipe_resource_reference(&glhd_surface->base.texture, NULL);
-   pipe_surface_reference(&glhd_surface->surface, NULL);
+   glhd_context->pipe->surface_destroy(glhd_context->pipe, glhd_surface->surface);
    FREE(glhd_surface);
 }
 
diff --git a/src/gallium/drivers/galahad/glhd_objects.h b/src/gallium/drivers/galahad/glhd_objects.h
index dc74c5bebc..13dc748588 100644
--- a/src/gallium/drivers/galahad/glhd_objects.h
+++ b/src/gallium/drivers/galahad/glhd_objects.h
@@ -149,11 +149,13 @@ void
 galahad_resource_destroy(struct galahad_resource *glhd_resource);
 
 struct pipe_surface *
-galahad_surface_create(struct galahad_resource *glhd_resource,
+galahad_surface_create(struct galahad_context *glhd_context,
+                       struct galahad_resource *glhd_resource,
                         struct pipe_surface *surface);
 
 void
-galahad_surface_destroy(struct galahad_surface *glhd_surface);
+galahad_surface_destroy(struct galahad_context *glhd_context,
+                         struct galahad_surface *glhd_surface);
 
 struct pipe_sampler_view *
 galahad_sampler_view_create(struct galahad_context *glhd_context,
diff --git a/src/gallium/drivers/galahad/glhd_screen.c b/src/gallium/drivers/galahad/glhd_screen.c
index b6cc41d908..b4825bef66 100644
--- a/src/gallium/drivers/galahad/glhd_screen.c
+++ b/src/gallium/drivers/galahad/glhd_screen.c
@@ -223,39 +223,6 @@ galahad_screen_resource_destroy(struct pipe_screen *screen,
    galahad_resource_destroy(galahad_resource(_resource));
 }
 
-static struct pipe_surface *
-galahad_screen_get_tex_surface(struct pipe_screen *_screen,
-                                struct pipe_resource *_resource,
-                                unsigned face,
-                                unsigned level,
-                                unsigned zslice,
-                                unsigned usage)
-{
-   struct galahad_screen *glhd_screen = galahad_screen(_screen);
-   struct galahad_resource *glhd_resource = galahad_resource(_resource);
-   struct pipe_screen *screen = glhd_screen->screen;
-   struct pipe_resource *resource = glhd_resource->resource;
-   struct pipe_surface *result;
-
-   result = screen->get_tex_surface(screen,
-                                    resource,
-                                    face,
-                                    level,
-                                    zslice,
-                                    usage);
-
-   if (result)
-      return galahad_surface_create(glhd_resource, result);
-   return NULL;
-}
-
-static void
-galahad_screen_tex_surface_destroy(struct pipe_surface *_surface)
-{
-   galahad_surface_destroy(galahad_surface(_surface));
-}
-
-
 
 static struct pipe_resource *
 galahad_screen_user_buffer_create(struct pipe_screen *_screen,
@@ -281,16 +248,18 @@ galahad_screen_user_buffer_create(struct pipe_screen *_screen,
 
 static void
 galahad_screen_flush_frontbuffer(struct pipe_screen *_screen,
-                                  struct pipe_surface *_surface,
+                                  struct pipe_resource *_resource,
+                                  unsigned level, unsigned layer,
                                   void *context_private)
 {
    struct galahad_screen *glhd_screen = galahad_screen(_screen);
-   struct galahad_surface *glhd_surface = galahad_surface(_surface);
+   struct galahad_resource *glhd_resource = galahad_resource(_resource);
    struct pipe_screen *screen = glhd_screen->screen;
-   struct pipe_surface *surface = glhd_surface->surface;
+   struct pipe_resource *resource = glhd_resource->resource;
 
    screen->flush_frontbuffer(screen,
-                             surface,
+                             resource,
+                             level, layer,
                              context_private);
 }
 
@@ -360,8 +329,6 @@ galahad_screen_create(struct pipe_screen *screen)
    glhd_screen->base.resource_from_handle = galahad_screen_resource_from_handle;
    glhd_screen->base.resource_get_handle = galahad_screen_resource_get_handle;
    glhd_screen->base.resource_destroy = galahad_screen_resource_destroy;
-   glhd_screen->base.get_tex_surface = galahad_screen_get_tex_surface;
-   glhd_screen->base.tex_surface_destroy = galahad_screen_tex_surface_destroy;
    glhd_screen->base.user_buffer_create = galahad_screen_user_buffer_create;
    glhd_screen->base.flush_frontbuffer = galahad_screen_flush_frontbuffer;
    glhd_screen->base.fence_reference = galahad_screen_fence_reference;
diff --git a/src/gallium/drivers/i915/TODO b/src/gallium/drivers/i915/TODO
new file mode 100644
index 0000000000..94c428bebf
--- /dev/null
+++ b/src/gallium/drivers/i915/TODO
@@ -0,0 +1,25 @@
+Random list of problems with i915g:
+
+- Dies with BadDrawable on GLXFBconfig changes/destruction. Makes piglit totally
+  unusable :( Upgrading xserver helped here, it doesn't crash anymore. Still
+  broken, it doesn't update the viewport/get new buffers.
+
+- Tends to hang the chip after a few minutes of openarena. Looks tiling related,
+  at the last frame rendered has tiling corruption over the complete frame.
+
+- Kills the chip in 3D_PRIMITIVE LINELIST with mesa-demos/fbotexture in
+  wireframe mode.
+
+- Tiling is funny: If unlucky, it renders/samples all black. No clue yet what's
+  going on. Seems to depend on tiny details like whethever the sampler
+  relocation is fenced/unfenced (broken _with_ fenced reloc using tiling bits!).
+
+- Y-tiling is even more fun. i915c doesn't use it, maybe there's a reason?
+  Texture sampling from Y-tiled buffers seems to work, though (save above
+  problems).
+
+- Need to validate buffers before usage. Currently do_exec on the batchbuffer
+  can fail with -ENOSPC.
+
+Other bugs can be found here:
+https://bugs.freedesktop.org/buglist.cgi?bug_status=NEW&bug_status=ASSIGNED&bug_status=REOPENED&component=Drivers/Gallium/i915g
diff --git a/src/gallium/drivers/i915/i915_batch.h b/src/gallium/drivers/i915/i915_batch.h
index c411b84ccd..6e93da7620 100644
--- a/src/gallium/drivers/i915/i915_batch.h
+++ b/src/gallium/drivers/i915/i915_batch.h
@@ -38,7 +38,10 @@
    i915_winsys_batchbuffer_dword(i915->batch, dword)
 
 #define OUT_RELOC(buf, usage, offset) \
-   i915_winsys_batchbuffer_reloc(i915->batch, buf, usage, offset)
+   i915_winsys_batchbuffer_reloc(i915->batch, buf, usage, offset, false)
+
+#define OUT_RELOC_FENCED(buf, usage, offset) \
+   i915_winsys_batchbuffer_reloc(i915->batch, buf, usage, offset, true)
 
 #define FLUSH_BATCH(fence) \
    i915_flush(i915, fence)
diff --git a/src/gallium/drivers/i915/i915_batchbuffer.h b/src/gallium/drivers/i915/i915_batchbuffer.h
index c1cd314e7b..d92b2ccb31 100644
--- a/src/gallium/drivers/i915/i915_batchbuffer.h
+++ b/src/gallium/drivers/i915/i915_batchbuffer.h
@@ -29,42 +29,47 @@
 #define I915_BATCHBUFFER_H
 
 #include "i915_winsys.h"
+#include "util/u_debug.h"
 
 struct i915_context;
 
+static INLINE size_t
+i915_winsys_batchbuffer_space(struct i915_winsys_batchbuffer *batch)
+{
+   return batch->size - (batch->ptr - batch->map);
+}
+
 static INLINE boolean
 i915_winsys_batchbuffer_check(struct i915_winsys_batchbuffer *batch,
                               size_t dwords,
                               size_t relocs)
 {
-   return dwords * 4 <= batch->size - (batch->ptr - batch->map) &&
+   return dwords * 4 <= i915_winsys_batchbuffer_space(batch) &&
           relocs <= (batch->max_relocs - batch->relocs);
 }
 
-static INLINE size_t
-i915_winsys_batchbuffer_space(struct i915_winsys_batchbuffer *batch)
+static INLINE void
+i915_winsys_batchbuffer_dword_unchecked(struct i915_winsys_batchbuffer *batch,
+                              unsigned dword)
 {
-   return batch->size - (batch->ptr - batch->map);
+   *(unsigned *)batch->ptr = dword;
+   batch->ptr += 4;
 }
 
 static INLINE void
 i915_winsys_batchbuffer_dword(struct i915_winsys_batchbuffer *batch,
                               unsigned dword)
 {
-   if (i915_winsys_batchbuffer_space(batch) < 4)
-      return;
-
-   *(unsigned *)batch->ptr = dword;
-   batch->ptr += 4;
+   assert (i915_winsys_batchbuffer_space(batch) >= 4);
+   i915_winsys_batchbuffer_dword_unchecked(batch, dword);
 }
 
 static INLINE void
 i915_winsys_batchbuffer_write(struct i915_winsys_batchbuffer *batch,
-                       void *data,
-                       size_t size)
+			      void *data,
+			      size_t size)
 {
-   if (i915_winsys_batchbuffer_space(batch) < size)
-      return;
+   assert (i915_winsys_batchbuffer_space(batch) >= size);
 
    memcpy(data, batch->ptr, size);
    batch->ptr += size;
@@ -74,9 +79,9 @@ static INLINE int
 i915_winsys_batchbuffer_reloc(struct i915_winsys_batchbuffer *batch,
                               struct i915_winsys_buffer *buffer,
                               enum i915_winsys_buffer_usage usage,
-                              size_t offset)
+                              size_t offset, bool fenced)
 {
-   return batch->iws->batchbuffer_reloc(batch, buffer, usage, offset);
+   return batch->iws->batchbuffer_reloc(batch, buffer, usage, offset, fenced);
 }
 
 #endif
diff --git a/src/gallium/drivers/i915/i915_blit.c b/src/gallium/drivers/i915/i915_blit.c
index cdf20c0055..97c2566515 100644
--- a/src/gallium/drivers/i915/i915_blit.c
+++ b/src/gallium/drivers/i915/i915_blit.c
@@ -74,7 +74,7 @@ i915_fill_blit(struct i915_context *i915,
    OUT_BATCH(BR13);
    OUT_BATCH((y << 16) | x);
    OUT_BATCH(((y + h) << 16) | (x + w));
-   OUT_RELOC(dst_buffer, I915_USAGE_2D_TARGET, dst_offset);
+   OUT_RELOC_FENCED(dst_buffer, I915_USAGE_2D_TARGET, dst_offset);
    OUT_BATCH(color);
 }
 
@@ -138,8 +138,8 @@ i915_copy_blit(struct i915_context *i915,
    OUT_BATCH(BR13);
    OUT_BATCH((dst_y << 16) | dst_x);
    OUT_BATCH((dst_y2 << 16) | dst_x2);
-   OUT_RELOC(dst_buffer, I915_USAGE_2D_TARGET, dst_offset);
+   OUT_RELOC_FENCED(dst_buffer, I915_USAGE_2D_TARGET, dst_offset);
    OUT_BATCH((src_y << 16) | src_x);
    OUT_BATCH(((int) src_pitch & 0xffff));
-   OUT_RELOC(src_buffer, I915_USAGE_2D_SOURCE, src_offset);
+   OUT_RELOC_FENCED(src_buffer, I915_USAGE_2D_SOURCE, src_offset);
 }
diff --git a/src/gallium/drivers/i915/i915_context.h b/src/gallium/drivers/i915/i915_context.h
index 3ae61d0ea7..d15e1723d8 100644
--- a/src/gallium/drivers/i915/i915_context.h
+++ b/src/gallium/drivers/i915/i915_context.h
@@ -193,8 +193,7 @@ struct i915_velems_state {
 };
 
 
-struct i915_context
-{
+struct i915_context {
    struct pipe_context base;
 
    struct i915_winsys *iws;
@@ -273,7 +272,7 @@ struct i915_context
 #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))
+#define I915_HW_INVARIANT         (1<<(I915_MAX_CACHE+1))
 
 
 /***********************************************************************
diff --git a/src/gallium/drivers/i915/i915_debug.c b/src/gallium/drivers/i915/i915_debug.c
index 57d3390dea..87c435a2f3 100644
--- a/src/gallium/drivers/i915/i915_debug.c
+++ b/src/gallium/drivers/i915/i915_debug.c
@@ -46,10 +46,12 @@ static const struct debug_named_value debug_options[] = {
 };
 
 unsigned i915_debug = 0;
+boolean i915_tiling = TRUE;
 
 void i915_debug_init(struct i915_screen *screen)
 {
    i915_debug = debug_get_flags_option("I915_DEBUG", debug_options, 0);
+   i915_tiling = !debug_get_bool_option("I915_NO_TILING", FALSE);
 }
 
 
@@ -974,7 +976,7 @@ i915_dump_hardware_dirty(struct i915_context *i915, const char *func)
       {I915_HW_PROGRAM,   "program"},
       {I915_HW_CONSTANTS, "constants"},
       {I915_HW_IMMEDIATE, "immediate"},
-      {I915_HW_INVARIENT, "invarient"},
+      {I915_HW_INVARIANT, "invariant"},
       {0, NULL},
    };
    int i;
diff --git a/src/gallium/drivers/i915/i915_debug.h b/src/gallium/drivers/i915/i915_debug.h
index fa60799d0c..11af7662f0 100644
--- a/src/gallium/drivers/i915/i915_debug.h
+++ b/src/gallium/drivers/i915/i915_debug.h
@@ -46,6 +46,7 @@ struct i915_winsys_batchbuffer;
 #define DBG_CONSTANTS 0x20
 
 extern unsigned i915_debug;
+extern boolean i915_tiling;
 
 #ifdef DEBUG
 static INLINE boolean
diff --git a/src/gallium/drivers/i915/i915_prim_vbuf.c b/src/gallium/drivers/i915/i915_prim_vbuf.c
index bd046bd905..baebbc7bae 100644
--- a/src/gallium/drivers/i915/i915_prim_vbuf.c
+++ b/src/gallium/drivers/i915/i915_prim_vbuf.c
@@ -172,6 +172,7 @@ i915_vbuf_render_reserve(struct i915_vbuf_render *i915_render, size_t size)
  *
  * Side effects:
  *    Updates hw_offset, sw_offset, index and allocates a new buffer.
+ *    Will set i915->vbo to null on buffer allocation.
  */
 static void
 i915_vbuf_render_new_buf(struct i915_vbuf_render *i915_render, size_t size)
@@ -179,8 +180,16 @@ i915_vbuf_render_new_buf(struct i915_vbuf_render *i915_render, size_t size)
    struct i915_context *i915 = i915_render->i915;
    struct i915_winsys *iws = i915->iws;
 
-   if (i915_render->vbo)
+   if (i915_render->vbo) {
       iws->buffer_destroy(iws, i915_render->vbo);
+      /*
+       * XXX If buffers where referenced then this should be done in
+       * update_vbo_state but since they arn't and malloc likes to reuse
+       * memory we need to set it to null
+       */
+      i915->vbo = NULL;
+      i915_render->vbo = NULL;
+   }
 
    i915->vbo_flushed = 0;
 
@@ -198,7 +207,7 @@ i915_vbuf_render_new_buf(struct i915_vbuf_render *i915_render, size_t size)
 #endif
 
    i915_render->vbo = iws->buffer_create(iws, i915_render->vbo_size,
-                                         64, I915_NEW_VERTEX);
+                                         I915_NEW_VERTEX);
 }
 
 /**
@@ -726,7 +735,7 @@ i915_vbuf_render_create(struct i915_context *i915)
    i915_render->pool_fifo = u_fifo_create(6);
    for (i = 0; i < 6; i++)
       u_fifo_add(i915_render->pool_fifo,
-                 iws->buffer_create(iws, i915_render->pool_buffer_size, 64,
+                 iws->buffer_create(iws, i915_render->pool_buffer_size,
                                     I915_NEW_VERTEX));
 #else
    (void)i;
diff --git a/src/gallium/drivers/i915/i915_reg.h b/src/gallium/drivers/i915/i915_reg.h
index 04620fec68..5e4e80ddf6 100644
--- a/src/gallium/drivers/i915/i915_reg.h
+++ b/src/gallium/drivers/i915/i915_reg.h
@@ -753,7 +753,7 @@
 #define    MT_COMPRESS_DXT1_RGB		   (4<<3)
 #define MS3_USE_FENCE_REGS              (1<<2)
 #define MS3_TILED_SURFACE             (1<<1)
-#define MS3_TILE_WALK                 (1<<0)
+#define MS3_TILE_WALK_Y                (1<<0)
 
 #define MS4_PITCH_SHIFT                 21
 #define MS4_CUBE_FACE_ENA_NEGX          (1<<20)
@@ -851,6 +851,7 @@
 #define MI_FLUSH                   ((0<<29)|(4<<23))
 #define FLUSH_MAP_CACHE            (1<<0)
 #define INHIBIT_FLUSH_RENDER_CACHE (1<<2)
+#define MI_NOOP                    0
 
 
 #define CMD_3D (0x3<<29)
@@ -973,6 +974,8 @@
 #define PCI_CHIP_G33_G			0x29C2
 #define PCI_CHIP_Q35_G			0x29B2
 #define PCI_CHIP_Q33_G			0x29D2
+#define PCI_CHIP_PINEVIEW_G		0xA001
+#define PCI_CHIP_PINEVIEW_M		0xA011
 
 
 #endif
diff --git a/src/gallium/drivers/i915/i915_resource.h b/src/gallium/drivers/i915/i915_resource.h
index 1093e8f41f..86620e6a12 100644
--- a/src/gallium/drivers/i915/i915_resource.h
+++ b/src/gallium/drivers/i915/i915_resource.h
@@ -32,6 +32,7 @@ struct i915_screen;
 
 #include "util/u_transfer.h"
 #include "util/u_debug.h"
+#include "i915_winsys.h"
 
 
 struct i915_context;
@@ -48,29 +49,36 @@ struct i915_buffer {
 #define I915_MAX_TEXTURE_3D_LEVELS  8  /* max 128x128x128 */
 
 
+struct offset_pair {
+	unsigned short nblocksx;
+	unsigned short nblocksy;
+};
 
 struct i915_texture {
    struct u_resource b;
 
+   /* tiling flags */
+   enum i915_winsys_buffer_tile tiling;
    unsigned stride;
    unsigned depth_stride;          /* per-image on i945? */
    unsigned total_nblocksy;
 
-   unsigned sw_tiled; /**< tiled with software flags */
-   unsigned hw_tiled; /**< tiled with hardware fences */
-
    unsigned nr_images[I915_MAX_TEXTURE_2D_LEVELS];
 
    /* Explicitly store the offset of each image for each cube face or
     * depth value.
+    *
+    * Array [depth] off offsets.
     */
-   unsigned *image_offset[I915_MAX_TEXTURE_2D_LEVELS];   /**< array [depth] of offsets */
+   struct offset_pair *image_offset[I915_MAX_TEXTURE_2D_LEVELS];
 
    /* The data is held here:
     */
    struct i915_winsys_buffer *buffer;
 };
 
+unsigned i915_texture_offset(struct i915_texture *tex,
+                             unsigned level, unsigned layer);
 void i915_init_screen_resource_functions(struct i915_screen *is);
 void i915_init_resource_functions(struct i915_context *i915);
 
diff --git a/src/gallium/drivers/i915/i915_resource_buffer.c b/src/gallium/drivers/i915/i915_resource_buffer.c
index 0d379497df..450203d60a 100644
--- a/src/gallium/drivers/i915/i915_resource_buffer.c
+++ b/src/gallium/drivers/i915/i915_resource_buffer.c
@@ -62,7 +62,7 @@ i915_buffer_destroy(struct pipe_screen *screen,
 
 static void *
 i915_buffer_transfer_map( struct pipe_context *pipe,
-			  struct pipe_transfer *transfer )
+                          struct pipe_transfer *transfer )
 {
    struct i915_buffer *buffer = i915_buffer(transfer->resource);
    return buffer->data + transfer->box.x;
@@ -71,19 +71,19 @@ i915_buffer_transfer_map( struct pipe_context *pipe,
 
 static void
 i915_buffer_transfer_inline_write( struct pipe_context *rm_ctx,
-				   struct pipe_resource *resource,
-				   struct pipe_subresource sr,
-				   unsigned usage,
-				   const struct pipe_box *box,
-				   const void *data,
-				   unsigned stride,
-				   unsigned slice_stride)
+                                   struct pipe_resource *resource,
+                                   unsigned level,
+                                   unsigned usage,
+                                   const struct pipe_box *box,
+                                   const void *data,
+                                   unsigned stride,
+                                   unsigned layer_stride)
 {
    struct i915_buffer *buffer = i915_buffer(resource);
 
    memcpy(buffer->data + box->x,
-	  data,
-	  box->width);
+          data,
+          box->width);
 }
 
 
@@ -115,7 +115,7 @@ i915_buffer_create(struct pipe_screen *screen,
    buf->b.vtbl = &i915_buffer_vtbl;
    pipe_reference_init(&buf->b.b.reference, 1);
    buf->b.b.screen = screen;
-   
+
    buf->data = MALLOC(template->width0);
    buf->free_on_destroy = TRUE;
 
@@ -135,7 +135,7 @@ struct pipe_resource *
 i915_user_buffer_create(struct pipe_screen *screen,
                         void *ptr,
                         unsigned bytes,
-			unsigned bind)
+                        unsigned bind)
 {
    struct i915_buffer *buf = CALLOC_STRUCT(i915_buffer);
 
@@ -152,6 +152,7 @@ i915_user_buffer_create(struct pipe_screen *screen,
    buf->b.b.width0 = bytes;
    buf->b.b.height0 = 1;
    buf->b.b.depth0 = 1;
+   buf->b.b.array_size = 1;
 
    buf->data = ptr;
    buf->free_on_destroy = FALSE;
diff --git a/src/gallium/drivers/i915/i915_resource_texture.c b/src/gallium/drivers/i915/i915_resource_texture.c
index c5c6179b16..f19106f341 100644
--- a/src/gallium/drivers/i915/i915_resource_texture.c
+++ b/src/gallium/drivers/i915/i915_resource_texture.c
@@ -106,6 +106,23 @@ get_pot_stride(enum pipe_format format, unsigned width)
    return util_next_power_of_two(util_format_get_stride(format, width));
 }
 
+static INLINE const char*
+get_tiling_string(enum i915_winsys_buffer_tile tile)
+{
+   switch(tile) {
+   case I915_TILE_NONE:
+      return "none";
+   case I915_TILE_X:
+      return "x";
+   case I915_TILE_Y:
+      return "y";
+   default:
+      assert(FALSE);
+      return "?";
+   }
+}
+
+
 /*
  * More advanced helper funcs
  */
@@ -120,28 +137,56 @@ i915_texture_set_level_info(struct i915_texture *tex,
    assert(!tex->image_offset[level]);
 
    tex->nr_images[level] = nr_images;
-   tex->image_offset[level] = (unsigned *) MALLOC(nr_images * sizeof(unsigned));
-   tex->image_offset[level][0] = 0;
+   tex->image_offset[level] = MALLOC(nr_images * sizeof(struct offset_pair));
+   tex->image_offset[level][0].nblocksx = 0;
+   tex->image_offset[level][0].nblocksy = 0;
+}
+
+INLINE unsigned i915_texture_offset(struct i915_texture *tex,
+                                    unsigned level, unsigned layer)
+{
+   unsigned x, y;
+   x = tex->image_offset[level][layer].nblocksx
+      * util_format_get_blocksize(tex->b.b.format);
+   y = tex->image_offset[level][layer].nblocksy;
+
+   return y * tex->stride + x;
 }
 
 static void
 i915_texture_set_image_offset(struct i915_texture *tex,
                               unsigned level, unsigned img,
-                              unsigned x, unsigned y)
+                              unsigned nblocksx, unsigned nblocksy)
 {
    /* for the first image and level make sure offset is zero */
-   assert(!(img == 0 && level == 0) || (x == 0 && y == 0));
+   assert(!(img == 0 && level == 0) || (nblocksx == 0 && nblocksy == 0));
    assert(img < tex->nr_images[level]);
 
-   tex->image_offset[level][img] = y * tex->stride + x * util_format_get_blocksize(tex->b.b.format);
+   tex->image_offset[level][img].nblocksx = nblocksx;
+   tex->image_offset[level][img].nblocksy = nblocksy;
 
 #if DEBUG_TEXTURES
-   debug_printf("%s: %p level %u, img %u (%u, %u) %p\n", __FUNCTION__,
-                tex, level, img, x, y,
-                (void*)(uintptr_t)tex->image_offset[level][img]);
+   debug_printf("%s: %p level %u, img %u (%u, %u)\n", __FUNCTION__,
+                tex, level, img, x, y);
 #endif
 }
 
+static enum i915_winsys_buffer_tile
+i915_texture_tiling(struct pipe_resource *pt)
+{
+   if (!i915_tiling)
+      return I915_TILE_NONE;
+
+   if (pt->target == PIPE_TEXTURE_1D)
+      return I915_TILE_NONE;
+
+   if (util_format_is_s3tc(pt->format))
+      /* XXX X-tiling might make sense */
+      return I915_TILE_NONE;
+
+   return I915_TILE_X;
+}
+
 
 /*
  * Shared layout functions
@@ -163,9 +208,10 @@ i9x5_scanout_layout(struct i915_texture *tex)
    i915_texture_set_image_offset(tex, 0, 0, 0, 0);
 
    if (pt->width0 >= 240) {
-      tex->stride = get_pot_stride(pt->format, pt->width0);
+      tex->stride = align(util_format_get_stride(pt->format, pt->width0), 64);
       tex->total_nblocksy = align_nblocksy(pt->format, pt->height0, 8);
-      tex->hw_tiled = I915_TILE_X;
+      tex->tiling = I915_TILE_X;
+   /* special case for cursors */
    } else if (pt->width0 == 64 && pt->height0 == 64) {
       tex->stride = get_pot_stride(pt->format, pt->width0);
       tex->total_nblocksy = align_nblocksy(pt->format, pt->height0, 8);
@@ -200,9 +246,9 @@ i9x5_display_target_layout(struct i915_texture *tex)
    i915_texture_set_level_info(tex, 0, 1);
    i915_texture_set_image_offset(tex, 0, 0, 0, 0);
 
-   tex->stride = get_pot_stride(pt->format, pt->width0);
+   tex->stride = align(util_format_get_stride(pt->format, pt->width0), 64);
    tex->total_nblocksy = align_nblocksy(pt->format, pt->height0, 8);
-   tex->hw_tiled = I915_TILE_X;
+   tex->tiling = I915_TILE_X;
 
 #if DEBUG_TEXTURE
    debug_printf("%s size: %d,%d,%d offset %d,%d (0x%x)\n", __FUNCTION__,
@@ -357,6 +403,8 @@ i915_texture_layout(struct i915_texture * tex)
 {
    struct pipe_resource *pt = &tex->b.b;
 
+   tex->tiling = i915_texture_tiling(pt);
+
    switch (pt->target) {
    case PIPE_TEXTURE_1D:
    case PIPE_TEXTURE_2D:
@@ -603,6 +651,8 @@ i945_texture_layout(struct i915_texture * tex)
 {
    struct pipe_resource *pt = &tex->b.b;
 
+   tex->tiling = i915_texture_tiling(pt);
+
    switch (pt->target) {
    case PIPE_TEXTURE_1D:
    case PIPE_TEXTURE_2D:
@@ -650,7 +700,7 @@ i915_texture_get_handle(struct pipe_screen * screen,
 
 static void
 i915_texture_destroy(struct pipe_screen *screen,
-		     struct pipe_resource *pt)
+                     struct pipe_resource *pt)
 {
    struct i915_texture *tex = i915_texture(pt);
    struct i915_winsys *iws = i915_screen(screen)->iws;
@@ -667,10 +717,10 @@ i915_texture_destroy(struct pipe_screen *screen,
 
 static struct pipe_transfer * 
 i915_texture_get_transfer(struct pipe_context *context,
-			  struct pipe_resource *resource,
-			  struct pipe_subresource sr,
-			  unsigned usage,
-			  const struct pipe_box *box)
+                          struct pipe_resource *resource,
+                          unsigned level,
+                          unsigned usage,
+                          const struct pipe_box *box)
 {
    struct i915_texture *tex = i915_texture(resource);
    struct pipe_transfer *transfer = CALLOC_STRUCT(pipe_transfer);
@@ -678,37 +728,31 @@ i915_texture_get_transfer(struct pipe_context *context,
       return NULL;
 
    transfer->resource = resource;
-   transfer->sr = sr;
+   transfer->level = level;
    transfer->usage = usage;
    transfer->box = *box;
    transfer->stride = tex->stride;
+   /* FIXME: layer_stride */
 
    return transfer;
 }
 
-
 static void *
 i915_texture_transfer_map(struct pipe_context *pipe,
-			  struct pipe_transfer *transfer)
+                          struct pipe_transfer *transfer)
 {
    struct pipe_resource *resource = transfer->resource;
    struct i915_texture *tex = i915_texture(resource);
    struct i915_winsys *iws = i915_screen(pipe->screen)->iws;
-   struct pipe_subresource sr = transfer->sr;
    struct pipe_box *box = &transfer->box;
    enum pipe_format format = resource->format;
    unsigned offset;
    char *map;
 
-   if (resource->target == PIPE_TEXTURE_CUBE) {
-      offset = tex->image_offset[sr.level][sr.face];
-   } else if (resource->target == PIPE_TEXTURE_3D) {
-      offset = tex->image_offset[sr.level][box->z];
-   } else {
-      offset = tex->image_offset[sr.level][0];
-      assert(sr.face == 0);
+   if (resource->target != PIPE_TEXTURE_3D &&
+       resource->target != PIPE_TEXTURE_CUBE)
       assert(box->z == 0);
-   }
+   offset = i915_texture_offset(tex, transfer->level, box->z);
 
    map = iws->buffer_map(iws, tex->buffer,
                          (transfer->usage & PIPE_TRANSFER_WRITE) ? TRUE : FALSE);
@@ -754,7 +798,6 @@ i915_texture_create(struct pipe_screen *screen,
    struct i915_screen *is = i915_screen(screen);
    struct i915_winsys *iws = is->iws;
    struct i915_texture *tex = CALLOC_STRUCT(i915_texture);
-   size_t tex_size;
    unsigned buf_usage = 0;
 
    if (!tex)
@@ -773,8 +816,6 @@ i915_texture_create(struct pipe_screen *screen,
          goto fail;
    }
 
-   tex_size = tex->stride * tex->total_nblocksy;
-
    /* for scanouts and cursors, cursors arn't scanouts */
 
    /* XXX: use a custom flag for cursors, don't rely on magically
@@ -785,28 +826,15 @@ i915_texture_create(struct pipe_screen *screen,
    else
       buf_usage = I915_NEW_TEXTURE;
 
-   tex->buffer = iws->buffer_create(iws, tex_size, 64, buf_usage);
+   tex->buffer = iws->buffer_create_tiled(iws, &tex->stride, tex->total_nblocksy,
+		   			  &tex->tiling, buf_usage);
    if (!tex->buffer)
       goto fail;
 
-   /* setup any hw fences */
-   if (tex->hw_tiled) {
-      assert(tex->sw_tiled == I915_TILE_NONE);
-      iws->buffer_set_fence_reg(iws, tex->buffer, tex->stride, tex->hw_tiled);
-   }
-
-   
-#if 0
-   void *ptr = ws->buffer_map(ws, tex->buffer,
-      PIPE_BUFFER_USAGE_CPU_WRITE);
-   memset(ptr, 0x80, tex_size);
-   ws->buffer_unmap(ws, tex->buffer);
-#endif
-
-   I915_DBG(DBG_TEXTURE, "%s: %p size %u, stride %u, blocks (%u, %u)\n", __func__,
-            tex, (unsigned int)tex_size, tex->stride,
+   I915_DBG(DBG_TEXTURE, "%s: %p stride %u, blocks (%u, %u) tiling %s\n", __func__,
+            tex, tex->stride,
             tex->stride / util_format_get_blocksize(tex->b.b.format),
-            tex->total_nblocksy);
+            tex->total_nblocksy, get_tiling_string(tex->tiling));
 
    return &tex->b.b;
 
@@ -825,10 +853,11 @@ i915_texture_from_handle(struct pipe_screen * screen,
    struct i915_winsys *iws = is->iws;
    struct i915_winsys_buffer *buffer;
    unsigned stride;
+   enum i915_winsys_buffer_tile tiling;
 
    assert(screen);
 
-   buffer = iws->buffer_from_handle(iws, whandle, &stride);
+   buffer = iws->buffer_from_handle(iws, whandle, &tiling, &stride);
 
    /* Only supports one type */
    if ((template->target != PIPE_TEXTURE_2D &&
@@ -848,6 +877,7 @@ i915_texture_from_handle(struct pipe_screen * screen,
    tex->b.b.screen = screen;
 
    tex->stride = stride;
+   tex->tiling = tiling;
    tex->total_nblocksy = align_nblocksy(tex->b.b.format, tex->b.b.height0, 8);
 
    i915_texture_set_level_info(tex, 0, 1);
@@ -855,10 +885,10 @@ i915_texture_from_handle(struct pipe_screen * screen,
 
    tex->buffer = buffer;
 
-   I915_DBG(DBG_TEXTURE, "%s: %p stride %u, blocks (%ux%u)\n", __func__,
+   I915_DBG(DBG_TEXTURE, "%s: %p stride %u, blocks (%u, %u) tiling %s\n", __func__,
             tex, tex->stride,
             tex->stride / util_format_get_blocksize(tex->b.b.format),
-            tex->total_nblocksy);
+            tex->total_nblocksy, get_tiling_string(tex->tiling));
 
    return &tex->b.b;
 }
diff --git a/src/gallium/drivers/i915/i915_screen.c b/src/gallium/drivers/i915/i915_screen.c
index 312847fba9..f66478e729 100644
--- a/src/gallium/drivers/i915/i915_screen.c
+++ b/src/gallium/drivers/i915/i915_screen.c
@@ -83,6 +83,12 @@ i915_get_name(struct pipe_screen *screen)
    case PCI_CHIP_Q33_G:
       chipset = "Q33";
       break;
+   case PCI_CHIP_PINEVIEW_G:
+      chipset = "Pineview G";
+      break;
+   case PCI_CHIP_PINEVIEW_M:
+      chipset = "Pineview M";
+      break;
    default:
       chipset = "unknown";
       break;
@@ -183,6 +189,8 @@ i915_get_shader_param(struct pipe_screen *screen, unsigned shader, enum pipe_sha
       case PIPE_SHADER_CAP_INDIRECT_TEMP_ADDR:
       case PIPE_SHADER_CAP_INDIRECT_CONST_ADDR:
          return 1;
+      case PIPE_SHADER_CAP_SUBROUTINES:
+         return 0;
       default:
          assert(0);
          return 0;
@@ -225,10 +233,6 @@ i915_is_format_supported(struct pipe_screen *screen,
    static const enum pipe_format tex_supported[] = {
       PIPE_FORMAT_B8G8R8A8_UNORM,
       PIPE_FORMAT_B8G8R8X8_UNORM,
-      PIPE_FORMAT_R8G8B8A8_UNORM,
-#if 0
-      PIPE_FORMAT_R8G8B8X8_UNORM,
-#endif
       PIPE_FORMAT_B5G6R5_UNORM,
       PIPE_FORMAT_L8_UNORM,
       PIPE_FORMAT_A8_UNORM,
@@ -351,6 +355,8 @@ i915_screen_create(struct i915_winsys *iws)
    case PCI_CHIP_G33_G:
    case PCI_CHIP_Q33_G:
    case PCI_CHIP_Q35_G:
+   case PCI_CHIP_PINEVIEW_G:
+   case PCI_CHIP_PINEVIEW_M:
       is->is_i945 = TRUE;
       break;
 
@@ -381,7 +387,6 @@ i915_screen_create(struct i915_winsys *iws)
    is->base.fence_finish = i915_fence_finish;
 
    i915_init_screen_resource_functions(is);
-   i915_init_screen_surface_functions(is);
 
    i915_debug_init(is);
 
diff --git a/src/gallium/drivers/i915/i915_state_emit.c b/src/gallium/drivers/i915/i915_state_emit.c
index bd059d5716..86c0297649 100644
--- a/src/gallium/drivers/i915/i915_state_emit.c
+++ b/src/gallium/drivers/i915/i915_state_emit.c
@@ -86,6 +86,22 @@ framebuffer_size(const struct pipe_framebuffer_state *fb,
    }
 }
 
+static inline uint32_t
+buf_3d_tiling_bits(enum i915_winsys_buffer_tile tiling)
+{
+         uint32_t tiling_bits = 0;
+
+         switch (tiling) {
+         case I915_TILE_Y:
+            tiling_bits |= BUF_3D_TILE_WALK_Y;
+         case I915_TILE_X:
+            tiling_bits |= BUF_3D_TILED_SURFACE;
+         case I915_TILE_NONE:
+            break;
+         }
+
+         return tiling_bits;
+}
 
 /* Push the state into the sarea and/or texture memory.
  */
@@ -126,7 +142,7 @@ i915_emit_hardware_state(struct i915_context *i915 )
    save_relocs = i915->batch->relocs;
 
    /* 14 dwords, 0 relocs */
-   if (i915->hardware_dirty & I915_HW_INVARIENT)
+   if (i915->hardware_dirty & I915_HW_INVARIANT)
    {
       OUT_BATCH(_3DSTATE_AA_CMD |
                 AA_LINE_ECAAR_WIDTH_ENABLE |
@@ -220,46 +236,39 @@ i915_emit_hardware_state(struct i915_context *i915 )
       struct pipe_surface *depth_surface = i915->framebuffer.zsbuf;
 
       if (cbuf_surface) {
-         unsigned ctile = BUF_3D_USE_FENCE;
          struct i915_texture *tex = i915_texture(cbuf_surface->texture);
          assert(tex);
 
-         if (tex && tex->sw_tiled) {
-            ctile = BUF_3D_TILED_SURFACE;
-         }
-
          OUT_BATCH(_3DSTATE_BUF_INFO_CMD);
 
          OUT_BATCH(BUF_3D_ID_COLOR_BACK |
                    BUF_3D_PITCH(tex->stride) |  /* pitch in bytes */
-                   ctile);
+                   buf_3d_tiling_bits(tex->tiling));
 
          OUT_RELOC(tex->buffer,
                    I915_USAGE_RENDER,
-                   cbuf_surface->offset);
+                   0);
       }
 
       /* What happens if no zbuf??
        */
       if (depth_surface) {
-         unsigned ztile = BUF_3D_USE_FENCE;
          struct i915_texture *tex = i915_texture(depth_surface->texture);
+         unsigned offset = i915_texture_offset(tex, depth_surface->u.tex.level,
+                                               depth_surface->u.tex.first_layer);
          assert(tex);
-
-         if (tex && tex->sw_tiled) {
-            ztile = BUF_3D_TILED_SURFACE;
-         }
+         assert(offset == 0);
 
          OUT_BATCH(_3DSTATE_BUF_INFO_CMD);
 
          assert(tex);
          OUT_BATCH(BUF_3D_ID_DEPTH |
                    BUF_3D_PITCH(tex->stride) |  /* pitch in bytes */
-                   ztile);
+                   buf_3d_tiling_bits(tex->tiling));
 
          OUT_RELOC(tex->buffer,
                    I915_USAGE_RENDER,
-                   depth_surface->offset);
+                   0);
       }
 
       {
@@ -301,12 +310,11 @@ i915_emit_hardware_state(struct i915_context *i915 )
                if (enabled & (1 << unit)) {
                   struct i915_texture *texture = i915_texture(i915->fragment_sampler_views[unit]->texture);
                   struct i915_winsys_buffer *buf = texture->buffer;
-                  uint offset = 0;
                   assert(buf);
 
                   count++;
 
-                  OUT_RELOC(buf, I915_USAGE_SAMPLER, offset);
+                  OUT_RELOC(buf, I915_USAGE_SAMPLER, 0);
                   OUT_BATCH(i915->current.texbuffer[unit][0]); /* MS3 */
                   OUT_BATCH(i915->current.texbuffer[unit][1]); /* MS4 */
                }
@@ -399,18 +407,33 @@ i915_emit_hardware_state(struct i915_context *i915 )
 #if 01
    /* drawing surface size */
    /* 6 dwords, 0 relocs */
+   if (i915->hardware_dirty & I915_HW_STATIC)
    {
       uint w, h;
-      boolean k = framebuffer_size(&i915->framebuffer, &w, &h);
-      (void)k;
-      assert(k);
+      struct pipe_surface *cbuf_surface = i915->framebuffer.cbufs[0];
+      struct i915_texture *tex = i915_texture(cbuf_surface->texture);
+      unsigned x, y;
+      int layer;
+      uint32_t draw_offset;
+      boolean ret;
+
+      ret = framebuffer_size(&i915->framebuffer, &w, &h);
+      assert(ret);
+
+      layer = cbuf_surface->u.tex.first_layer;
+
+      x = tex->image_offset[cbuf_surface->u.tex.level][layer].nblocksx;
+      y = tex->image_offset[cbuf_surface->u.tex.level][layer].nblocksy;
+
+      draw_offset = x | (y << 16);
 
+      /* XXX flush only required when the draw_offset changes! */
+      OUT_BATCH(MI_FLUSH | INHIBIT_FLUSH_RENDER_CACHE);
       OUT_BATCH(_3DSTATE_DRAW_RECT_CMD);
-      OUT_BATCH(0);
-      OUT_BATCH(0);
-      OUT_BATCH(((w - 1) & 0xffff) | ((h - 1) << 16));
-      OUT_BATCH(0);
-      OUT_BATCH(0);
+      OUT_BATCH(DRAW_RECT_DIS_DEPTH_OFS);
+      OUT_BATCH(draw_offset);
+      OUT_BATCH((w - 1 + x) | ((h - 1 + y) << 16));
+      OUT_BATCH(draw_offset);
    }
 #endif
 
diff --git a/src/gallium/drivers/i915/i915_state_sampler.c b/src/gallium/drivers/i915/i915_state_sampler.c
index 4667e0b78d..916cb76753 100644
--- a/src/gallium/drivers/i915/i915_state_sampler.c
+++ b/src/gallium/drivers/i915/i915_state_sampler.c
@@ -243,6 +243,23 @@ static uint translate_texture_format(enum pipe_format pipeFormat)
    }
 }
 
+static inline uint32_t
+ms3_tiling_bits(enum i915_winsys_buffer_tile tiling)
+{
+         uint32_t tiling_bits = 0;
+
+         switch (tiling) {
+         case I915_TILE_Y:
+            tiling_bits |= MS3_TILE_WALK_Y;
+         case I915_TILE_X:
+            tiling_bits |= MS3_TILED_SURFACE;
+         case I915_TILE_NONE:
+            break;
+         }
+
+         return tiling_bits;
+}
+
 static void update_map(struct i915_context *i915,
                        uint unit,
                        const struct i915_texture *tex,
@@ -254,7 +271,6 @@ static void update_map(struct i915_context *i915,
    const uint width = pt->width0, height = pt->height0, depth = pt->depth0;
    const uint num_levels = pt->last_level;
    unsigned max_lod = num_levels * 4;
-   unsigned tiled = MS3_USE_FENCE_REGS;
 
    assert(tex);
    assert(width);
@@ -267,17 +283,12 @@ static void update_map(struct i915_context *i915,
    assert(format);
    assert(pitch);
 
-   if (tex->sw_tiled) {
-      assert(!((pitch - 1) & pitch));
-      tiled = MS3_TILED_SURFACE;
-   }
-
    /* MS3 state */
    state[0] =
       (((height - 1) << MS3_HEIGHT_SHIFT)
        | ((width - 1) << MS3_WIDTH_SHIFT)
        | format
-       | tiled);
+       | ms3_tiling_bits(tex->tiling));
 
    /*
     * XXX When min_filter != mag_filter and there's just one mipmap level,
diff --git a/src/gallium/drivers/i915/i915_surface.c b/src/gallium/drivers/i915/i915_surface.c
index f40876e708..becc6e93c2 100644
--- a/src/gallium/drivers/i915/i915_surface.c
+++ b/src/gallium/drivers/i915/i915_surface.c
@@ -43,11 +43,10 @@
  */
 static void
 i915_surface_copy(struct pipe_context *pipe,
-                  struct pipe_resource *dst, struct pipe_subresource subdst,
+                  struct pipe_resource *dst, unsigned dst_level,
                   unsigned dstx, unsigned dsty, unsigned dstz,
-                  struct pipe_resource *src, struct pipe_subresource subsrc,
-                  unsigned srcx, unsigned srcy, unsigned srcz,
-                  unsigned width, unsigned height)
+                  struct pipe_resource *src, unsigned src_level,
+                  const struct pipe_box *src_box)
 {
    struct i915_texture *dst_tex = i915_texture(dst);
    struct i915_texture *src_tex = i915_texture(src);
@@ -55,29 +54,17 @@ i915_surface_copy(struct pipe_context *pipe,
    struct pipe_resource *spt = &src_tex->b.b;
    unsigned dst_offset, src_offset;  /* in bytes */
 
-   if (dst->target == PIPE_TEXTURE_CUBE) {
-      dst_offset = dst_tex->image_offset[subdst.level][subdst.face];
-   }
-   else if (dst->target == PIPE_TEXTURE_3D) {
-      dst_offset = dst_tex->image_offset[subdst.level][dstz];
-   }
-   else {
-      dst_offset = dst_tex->image_offset[subdst.level][0];
-      assert(subdst.face == 0);
+   /* XXX cannot copy 3d regions at this time */
+   assert(src_box->depth == 1);
+   if (dst->target != PIPE_TEXTURE_CUBE &&
+       dst->target != PIPE_TEXTURE_3D)
       assert(dstz == 0);
-   }
-   if (src->target == PIPE_TEXTURE_CUBE) {
-      src_offset = src_tex->image_offset[subsrc.level][subsrc.face];
-   }
-   else if (src->target == PIPE_TEXTURE_3D) {
-      src_offset = src_tex->image_offset[subsrc.level][srcz];
-   }
-   else {
-      src_offset = src_tex->image_offset[subsrc.level][0];
-      assert(subsrc.face == 0);
-      assert(srcz == 0);
-   }
+   dst_offset = i915_texture_offset(dst_tex, dst_level, dstz);
 
+   if (src->target != PIPE_TEXTURE_CUBE &&
+       src->target != PIPE_TEXTURE_3D)
+      assert(src_box->z == 0);
+   src_offset = i915_texture_offset(src_tex, src_level, src_box->z);
 
    assert( dst != src );
    assert( util_format_get_blocksize(dpt->format) == util_format_get_blocksize(spt->format) );
@@ -90,7 +77,8 @@ i915_surface_copy(struct pipe_context *pipe,
                    util_format_get_blocksize(dpt->format),
                    (unsigned short) src_tex->stride, src_tex->buffer, src_offset,
                    (unsigned short) dst_tex->stride, dst_tex->buffer, dst_offset,
-                   (short) srcx, (short) srcy, (short) dstx, (short) dsty, (short) width, (short) height );
+                   (short) src_box->x, (short) src_box->y, (short) dstx, (short) dsty,
+                   (short) src_box->width, (short) src_box->height );
 }
 
 
@@ -104,6 +92,7 @@ i915_clear_render_target(struct pipe_context *pipe,
    struct i915_texture *tex = i915_texture(dst->texture);
    struct pipe_resource *pt = &tex->b.b;
    union util_color uc;
+   unsigned offset = i915_texture_offset(tex, dst->u.tex.level, dst->u.tex.first_layer);
 
    assert(util_format_get_blockwidth(pt->format) == 1);
    assert(util_format_get_blockheight(pt->format) == 1);
@@ -113,7 +102,7 @@ i915_clear_render_target(struct pipe_context *pipe,
                    util_format_get_blocksize(pt->format),
                    XY_COLOR_BLT_WRITE_ALPHA | XY_COLOR_BLT_WRITE_RGB,
                    (unsigned short) tex->stride,
-                   tex->buffer, dst->offset,
+                   tex->buffer, offset,
                    (short) dstx, (short) dsty,
                    (short) width, (short) height,
                    uc.ui );
@@ -132,6 +121,7 @@ i915_clear_depth_stencil(struct pipe_context *pipe,
    struct pipe_resource *pt = &tex->b.b;
    unsigned packedds;
    unsigned mask = 0;
+   unsigned offset = i915_texture_offset(tex, dst->u.tex.level, dst->u.tex.first_layer);
 
    assert(util_format_get_blockwidth(pt->format) == 1);
    assert(util_format_get_blockheight(pt->format) == 1);
@@ -151,7 +141,7 @@ i915_clear_depth_stencil(struct pipe_context *pipe,
                    util_format_get_blocksize(pt->format),
                    mask,
                    (unsigned short) tex->stride,
-                   tex->buffer, dst->offset,
+                   tex->buffer, offset,
                    (short) dstx, (short) dsty,
                    (short) width, (short) height,
                    packedds );
@@ -163,42 +153,37 @@ i915_clear_depth_stencil(struct pipe_context *pipe,
 
 
 static struct pipe_surface *
-i915_get_tex_surface(struct pipe_screen *screen,
-                     struct pipe_resource *pt,
-                     unsigned face, unsigned level, unsigned zslice,
-                     unsigned flags)
+i915_create_surface(struct pipe_context *ctx,
+                    struct pipe_resource *pt,
+                    const struct pipe_surface *surf_tmpl)
 {
-   struct i915_texture *tex = i915_texture(pt);
    struct pipe_surface *ps;
-   unsigned offset;  /* in bytes */
 
-   if (pt->target == PIPE_TEXTURE_CUBE) {
-      offset = tex->image_offset[level][face];
-   }
-   else if (pt->target == PIPE_TEXTURE_3D) {
-      offset = tex->image_offset[level][zslice];
-   }
-   else {
-      offset = tex->image_offset[level][0];
-      assert(face == 0);
-      assert(zslice == 0);
-   }
+   assert(surf_tmpl->u.tex.first_layer == surf_tmpl->u.tex.last_layer);
+   if (pt->target != PIPE_TEXTURE_CUBE &&
+       pt->target != PIPE_TEXTURE_3D)
+      assert(surf_tmpl->u.tex.first_layer == 0);
 
    ps = CALLOC_STRUCT(pipe_surface);
    if (ps) {
+      /* could subclass pipe_surface and store offset as it used to do */
       pipe_reference_init(&ps->reference, 1);
       pipe_resource_reference(&ps->texture, pt);
-      ps->format = pt->format;
-      ps->width = u_minify(pt->width0, level);
-      ps->height = u_minify(pt->height0, level);
-      ps->offset = offset;
-      ps->usage = flags;
+      ps->format = surf_tmpl->format;
+      ps->width = u_minify(pt->width0, surf_tmpl->u.tex.level);
+      ps->height = u_minify(pt->height0, surf_tmpl->u.tex.level);
+      ps->u.tex.level = surf_tmpl->u.tex.level;
+      ps->u.tex.first_layer = surf_tmpl->u.tex.first_layer;
+      ps->u.tex.last_layer = surf_tmpl->u.tex.last_layer;
+      ps->usage = surf_tmpl->usage;
+      ps->context = ctx;
    }
    return ps;
 }
 
 static void
-i915_tex_surface_destroy(struct pipe_surface *surf)
+i915_surface_destroy(struct pipe_context *ctx,
+                     struct pipe_surface *surf)
 {
    pipe_resource_reference(&surf->texture, NULL);
    FREE(surf);
@@ -211,13 +196,6 @@ i915_init_surface_functions(struct i915_context *i915)
    i915->base.resource_copy_region = i915_surface_copy;
    i915->base.clear_render_target = i915_clear_render_target;
    i915->base.clear_depth_stencil = i915_clear_depth_stencil;
-}
-
-/* No good reason for these to be in the screen.
- */
-void
-i915_init_screen_surface_functions(struct i915_screen *is)
-{
-   is->base.get_tex_surface = i915_get_tex_surface;
-   is->base.tex_surface_destroy = i915_tex_surface_destroy;
+   i915->base.create_surface = i915_create_surface;
+   i915->base.surface_destroy = i915_surface_destroy;
 }
diff --git a/src/gallium/drivers/i915/i915_surface.h b/src/gallium/drivers/i915/i915_surface.h
index 448106d566..70b61de80f 100644
--- a/src/gallium/drivers/i915/i915_surface.h
+++ b/src/gallium/drivers/i915/i915_surface.h
@@ -32,7 +32,6 @@ struct i915_context;
 struct i915_screen;
 
 void i915_init_surface_functions( struct i915_context *i915 );
-void i915_init_screen_surface_functions( struct i915_screen *is );
 
 
 #endif /* I915_SCREEN_H */
diff --git a/src/gallium/drivers/i915/i915_winsys.h b/src/gallium/drivers/i915/i915_winsys.h
index 5385e403d2..24ea416f01 100644
--- a/src/gallium/drivers/i915/i915_winsys.h
+++ b/src/gallium/drivers/i915/i915_winsys.h
@@ -53,6 +53,7 @@ enum i915_winsys_buffer_type
    I915_NEW_VERTEX
 };
 
+/* These need to be in sync with the definitions of libdrm-intel! */
 enum i915_winsys_buffer_tile
 {
    I915_TILE_NONE,
@@ -106,7 +107,7 @@ struct i915_winsys {
    int (*batchbuffer_reloc)(struct i915_winsys_batchbuffer *batch,
                             struct i915_winsys_buffer *reloc,
                             enum i915_winsys_buffer_usage usage,
-                            unsigned offset);
+                            unsigned offset, bool fenced);
 
    /**
     * Flush a bufferbatch.
@@ -130,10 +131,24 @@ struct i915_winsys {
     */
    struct i915_winsys_buffer *
       (*buffer_create)(struct i915_winsys *iws,
-                       unsigned size, unsigned alignment,
+                       unsigned size,
                        enum i915_winsys_buffer_type type);
 
    /**
+    * Create a tiled buffer.
+    *
+    * *stride, height are in bytes. The winsys tries to allocate the buffer with
+    * the tiling mode provide in *tiling. If tiling is no possible, *tiling will
+    * be set to I915_TILE_NONE. The calculated stride (incorporateing hw/kernel
+    * requirements) is always returned in *stride.
+    */
+   struct i915_winsys_buffer *
+      (*buffer_create_tiled)(struct i915_winsys *iws,
+                             unsigned *stride, unsigned height,
+                             enum i915_winsys_buffer_tile *tiling,
+                             enum i915_winsys_buffer_type type);
+
+   /**
     * Creates a buffer from a handle.
     * Used to implement pipe_screen::resource_from_handle.
     * Also provides the stride information needed for the
@@ -142,6 +157,7 @@ struct i915_winsys {
    struct i915_winsys_buffer *
       (*buffer_from_handle)(struct i915_winsys *iws,
                             struct winsys_handle *whandle,
+                            enum i915_winsys_buffer_tile *tiling,
                             unsigned *stride);
 
    /**
@@ -154,15 +170,6 @@ struct i915_winsys {
                                 unsigned stride);
 
    /**
-    * Fence a buffer with a fence reg.
-    * Not to be confused with pipe_fence_handle.
-    */
-   int (*buffer_set_fence_reg)(struct i915_winsys *iws,
-                               struct i915_winsys_buffer *buffer,
-                               unsigned stride,
-                               enum i915_winsys_buffer_tile tile);
-
-   /**
     * Map a buffer.
     */
    void *(*buffer_map)(struct i915_winsys *iws,
diff --git a/src/gallium/drivers/i965/Makefile b/src/gallium/drivers/i965/Makefile
index b0b0970338..a0331f8058 100644
--- a/src/gallium/drivers/i965/Makefile
+++ b/src/gallium/drivers/i965/Makefile
@@ -33,6 +33,7 @@ C_SOURCES = \
 	brw_pipe_flush.c \
 	brw_pipe_misc.c \
 	brw_pipe_sampler.c \
+        brw_pipe_surface.c \
 	brw_pipe_vertex.c \
 	brw_pipe_clear.c \
 	brw_pipe_rast.c \
@@ -66,7 +67,6 @@ C_SOURCES = \
 	brw_resource_buffer.c \
 	brw_resource_texture.c \
 	brw_resource_texture_layout.c \
-	brw_screen_surface.c \
 	brw_batchbuffer.c \
 	brw_winsys_debug.c \
 	intel_decode.c
diff --git a/src/gallium/drivers/i965/SConscript b/src/gallium/drivers/i965/SConscript
index 019af682f6..3ef6c88030 100644
--- a/src/gallium/drivers/i965/SConscript
+++ b/src/gallium/drivers/i965/SConscript
@@ -36,6 +36,8 @@ i965 = env.ConvenienceLibrary(
 		'brw_pipe_query.c',
 		'brw_pipe_rast.c',
 		'brw_pipe_sampler.c',
+                'brw_pipe_surface.c',
+                'brw_pipe_surface.c',
 		'brw_pipe_shader.c',
 		'brw_pipe_vertex.c',
 		'brw_resource.c',
@@ -43,7 +45,6 @@ i965 = env.ConvenienceLibrary(
 		'brw_resource_texture.c',
 		'brw_resource_texture_layout.c',
 		'brw_screen.c',
-		'brw_screen_surface.c',
 		'brw_structs_dump.c',
 		'brw_sf.c',
 		'brw_sf_emit.c',
diff --git a/src/gallium/drivers/i965/brw_context.c b/src/gallium/drivers/i965/brw_context.c
index 227bc790de..a2736f783d 100644
--- a/src/gallium/drivers/i965/brw_context.c
+++ b/src/gallium/drivers/i965/brw_context.c
@@ -131,6 +131,7 @@ struct pipe_context *brw_create_context(struct pipe_screen *screen,
    brw_pipe_shader_init( brw );
    brw_pipe_vertex_init( brw );
    brw_pipe_clear_init( brw );
+   brw_pipe_surface_init( brw );
 
    brw_hw_cc_init( brw );
 
diff --git a/src/gallium/drivers/i965/brw_context.h b/src/gallium/drivers/i965/brw_context.h
index 56d351f97d..d927f382d5 100644
--- a/src/gallium/drivers/i965/brw_context.h
+++ b/src/gallium/drivers/i965/brw_context.h
@@ -821,6 +821,7 @@ void brw_pipe_sampler_cleanup( struct brw_context *brw );
 void brw_pipe_shader_cleanup( struct brw_context *brw );
 void brw_pipe_vertex_cleanup( struct brw_context *brw );
 void brw_pipe_clear_cleanup( struct brw_context *brw );
+void brw_pipe_surface_init( struct brw_context *brw );
 
 void brw_hw_cc_init( struct brw_context *brw );
 void brw_hw_cc_cleanup( struct brw_context *brw );
diff --git a/src/gallium/drivers/i965/brw_misc_state.c b/src/gallium/drivers/i965/brw_misc_state.c
index b5029ceb69..c635d69661 100644
--- a/src/gallium/drivers/i965/brw_misc_state.c
+++ b/src/gallium/drivers/i965/brw_misc_state.c
@@ -287,11 +287,12 @@ static int emit_depthbuffer(struct brw_context *brw)
       OUT_BATCH(((pitch * cpp) - 1) |
 		(format << 18) |
 		(BRW_TILEWALK_YMAJOR << 26) |
-		((surface->layout != PIPE_SURFACE_LAYOUT_LINEAR) << 27) |
+                /* always linear ?
+		((surface->layout != PIPE_SURFACE_LAYOUT_LINEAR) << 27) |*/
 		(BRW_SURFACE_2D << 29));
       OUT_RELOC(bo,
 		BRW_USAGE_DEPTH_BUFFER,
-		surface->offset);
+		brw_surface(surface)->offset);
       OUT_BATCH((BRW_SURFACE_MIPMAPLAYOUT_BELOW << 1) |
 		((pitch - 1) << 6) |
 		((surface->height - 1) << 19));
@@ -362,10 +363,10 @@ const struct brw_tracked_state brw_line_stipple = {
 
 
 /***********************************************************************
- * Misc invarient state packets
+ * Misc invariant state packets
  */
 
-static int upload_invarient_state( struct brw_context *brw )
+static int upload_invariant_state( struct brw_context *brw )
 {
    {
       /* 0x61040000  Pipeline Select */
@@ -438,7 +439,7 @@ static int upload_invarient_state( struct brw_context *brw )
    {
       struct brw_polygon_stipple_offset bpso;
       
-      /* This is invarient state in gallium:
+      /* This is invariant state in gallium:
        */
       memset(&bpso, 0, sizeof(bpso));
       bpso.header.opcode = CMD_POLY_STIPPLE_OFFSET;
@@ -452,13 +453,13 @@ static int upload_invarient_state( struct brw_context *brw )
    return 0;
 }
 
-const struct brw_tracked_state brw_invarient_state = {
+const struct brw_tracked_state brw_invariant_state = {
    .dirty = {
       .mesa = 0,
       .brw = BRW_NEW_CONTEXT,
       .cache = 0
    },
-   .emit = upload_invarient_state
+   .emit = upload_invariant_state
 };
 
 
diff --git a/src/gallium/drivers/i965/brw_pipe_clear.c b/src/gallium/drivers/i965/brw_pipe_clear.c
index d5cff338a6..7bf3ea6994 100644
--- a/src/gallium/drivers/i965/brw_pipe_clear.c
+++ b/src/gallium/drivers/i965/brw_pipe_clear.c
@@ -64,7 +64,7 @@ try_clear( struct brw_context *brw,
    debug_printf("%s dst:buf(%p)/%d+%d %d,%d sz:%dx%d\n",
                 __FUNCTION__,
                 (void *)surface->bo, pitch * cpp,
-                surface->base.offset,
+                surface->offset,
                 x1, y1, x2 - x1, y2 - y1);
 
    BR13 = 0xf0 << 16;
@@ -99,7 +99,7 @@ try_clear( struct brw_context *brw,
    OUT_BATCH((y2 << 16) | x2);
    OUT_RELOC(surface->bo,
              BRW_USAGE_BLIT_DEST,
-             surface->base.offset);
+             surface->offset);
    OUT_BATCH(value);
    ADVANCE_BATCH();
 
diff --git a/src/gallium/drivers/i965/brw_screen_surface.c b/src/gallium/drivers/i965/brw_pipe_surface.c
index f288fdbcd3..4deead98b1 100644
--- a/src/gallium/drivers/i965/brw_screen_surface.c
+++ b/src/gallium/drivers/i965/brw_pipe_surface.c
@@ -35,6 +35,7 @@
 
 #include "pipe/p_screen.h"
 #include "brw_screen.h"
+#include "brw_context.h"
 #include "brw_defines.h"
 #include "brw_resource.h"
 #include "brw_winsys.h"
@@ -108,9 +109,10 @@ void brw_update_texture( struct brw_screen *brw_screen,
  * where it would be illegal (perhaps due to tiling constraints) to do
  * this in-place.
  * 
- * Currently not implmented, not sure if it's needed.
+ * Currently not implemented, not sure if it's needed.
  */
 static struct brw_surface *create_linear_view( struct brw_screen *brw_screen,
+                                               struct pipe_context *pipe,
 					       struct brw_texture *tex,
 					       union brw_surface_id id,
 					       unsigned usage )
@@ -123,9 +125,10 @@ static struct brw_surface *create_linear_view( struct brw_screen *brw_screen,
  * texture's storage.
  */
 static struct brw_surface *create_in_place_view( struct brw_screen *brw_screen,
-						  struct brw_texture *tex,
-						  union brw_surface_id id,
-						  unsigned usage )
+                                                 struct pipe_context *pipe,
+                                                 struct brw_texture *tex,
+                                                 union brw_surface_id id,
+                                                 unsigned usage )
 {
    struct brw_surface *surface;
 
@@ -137,17 +140,18 @@ static struct brw_surface *create_in_place_view( struct brw_screen *brw_screen,
 
    /* XXX: ignoring render-to-slice-of-3d-texture
     */
-   assert(id.bits.zslice == 0);
+   assert(tex->b.b.target != PIPE_TEXTURE_3D || id.bits.layer == 0);
 
+   surface->base.context = pipe;
    surface->base.format = tex->b.b.format;
    surface->base.width = u_minify(tex->b.b.width0, id.bits.level);
    surface->base.height = u_minify(tex->b.b.height0, id.bits.level);
-   surface->base.offset = tex->image_offset[id.bits.level][id.bits.face];
    surface->base.usage = usage;
-   surface->base.zslice = id.bits.zslice;
-   surface->base.face = id.bits.face;
-   surface->base.level = id.bits.level;
+   surface->base.u.tex.first_layer = id.bits.layer;
+   surface->base.u.tex.last_layer = surface->base.u.tex.first_layer;
+   surface->base.u.tex.level = id.bits.level;
    surface->id = id;
+   surface->offset = tex->image_offset[id.bits.level][id.bits.layer];
    surface->cpp = tex->cpp;
    surface->pitch = tex->pitch;
    surface->tiling = tex->tiling;
@@ -159,11 +163,11 @@ static struct brw_surface *create_in_place_view( struct brw_screen *brw_screen,
    surface->ss.ss0.surface_type = BRW_SURFACE_2D;
 
    if (tex->tiling == BRW_TILING_NONE) {
-      surface->ss.ss1.base_addr = surface->base.offset;
+      surface->ss.ss1.base_addr = surface->offset;
    } else {
-      uint32_t tile_offset = surface->base.offset % 4096;
+      uint32_t tile_offset = surface->offset % 4096;
 
-      surface->ss.ss1.base_addr = surface->base.offset - tile_offset;
+      surface->ss.ss1.base_addr = surface->offset - tile_offset;
 
       if (brw_screen->chipset.is_g4x) {
 	 if (tex->tiling == BRW_TILING_X) {
@@ -198,23 +202,21 @@ static struct brw_surface *create_in_place_view( struct brw_screen *brw_screen,
 
 /* Get a surface which is view into a texture 
  */
-static struct pipe_surface *brw_get_tex_surface(struct pipe_screen *screen,
-						struct pipe_resource *pt,
-						unsigned face, unsigned level,
-						unsigned zslice,
-						unsigned usage )
+static struct pipe_surface *brw_create_surface(struct pipe_context *pipe,
+                                               struct pipe_resource *pt,
+                                               const struct pipe_surface *surf_tmpl)
 {
    struct brw_texture *tex = brw_texture(pt);
-   struct brw_screen *bscreen = brw_screen(screen);
+   struct brw_screen *bscreen = brw_screen(pipe->screen);
    struct brw_surface *surface;
    union brw_surface_id id;
    int type;
 
-   id.bits.face = face;
-   id.bits.level = level;
-   id.bits.zslice = zslice;
+   assert(surf_tmpl->u.tex.first_layer == surf_tmpl->u.tex.last_layer);
+   id.bits.level = surf_tmpl->u.tex.level;
+   id.bits.layer = surf_tmpl->u.tex.first_layer;
 
-   if (need_linear_view(bscreen, tex, id, usage)) 
+   if (need_linear_view(bscreen, tex, id, surf_tmpl->usage))
       type = BRW_VIEW_LINEAR;
    else
       type = BRW_VIEW_IN_PLACE;
@@ -227,10 +229,10 @@ static struct pipe_surface *brw_get_tex_surface(struct pipe_screen *screen,
 
    switch (type) {
    case BRW_VIEW_LINEAR:
-      surface = create_linear_view( bscreen, tex, id, usage );
+      surface = create_linear_view( bscreen, pipe, tex, id, surf_tmpl->usage );
       break;
    case BRW_VIEW_IN_PLACE:
-      surface = create_in_place_view( bscreen, tex, id, usage );
+      surface = create_in_place_view( bscreen, pipe, tex, id, surf_tmpl->usage );
       break;
    }
 
@@ -239,7 +241,8 @@ static struct pipe_surface *brw_get_tex_surface(struct pipe_screen *screen,
 }
 
 
-static void brw_tex_surface_destroy( struct pipe_surface *surf )
+static void brw_surface_destroy( struct pipe_context *pipe,
+                                 struct pipe_surface *surf )
 {
    struct brw_surface *surface = brw_surface(surf);
 
@@ -249,13 +252,12 @@ static void brw_tex_surface_destroy( struct pipe_surface *surf )
    bo_reference(&surface->bo, NULL);
    pipe_resource_reference( &surface->base.texture, NULL );
 
-
    FREE(surface);
 }
 
 
-void brw_screen_tex_surface_init( struct brw_screen *brw_screen )
+void brw_pipe_surface_init( struct brw_context *brw )
 {
-   brw_screen->base.get_tex_surface = brw_get_tex_surface;
-   brw_screen->base.tex_surface_destroy = brw_tex_surface_destroy;
+   brw->base.create_surface = brw_create_surface;
+   brw->base.surface_destroy = brw_surface_destroy;
 }
diff --git a/src/gallium/drivers/i965/brw_resource_buffer.c b/src/gallium/drivers/i965/brw_resource_buffer.c
index 5f9e8a87c9..afb96ee3e7 100644
--- a/src/gallium/drivers/i965/brw_resource_buffer.c
+++ b/src/gallium/drivers/i965/brw_resource_buffer.c
@@ -92,9 +92,9 @@ brw_buffer_transfer_unmap( struct pipe_context *pipe,
 
 
 static unsigned brw_buffer_is_referenced( struct pipe_context *pipe,
-					 struct pipe_resource *resource,
-					 unsigned face,
-					 unsigned level)
+                                          struct pipe_resource *resource,
+                                          unsigned level,
+                                          int layer)
 {
    struct brw_context *brw = brw_context(pipe);
    struct brw_winsys_buffer *batch_bo = brw->batch->buf;
@@ -194,6 +194,7 @@ brw_user_buffer_create(struct pipe_screen *screen,
    buf->b.b.width0 = bytes;
    buf->b.b.height0 = 1;
    buf->b.b.depth0 = 1;
+   buf->b.b.array_size = 1;
 
    buf->user_buffer = ptr;
    
diff --git a/src/gallium/drivers/i965/brw_resource_texture.c b/src/gallium/drivers/i965/brw_resource_texture.c
index 3860d18a7a..fded2da382 100644
--- a/src/gallium/drivers/i965/brw_resource_texture.c
+++ b/src/gallium/drivers/i965/brw_resource_texture.c
@@ -229,8 +229,8 @@ static void brw_texture_destroy(struct pipe_screen *screen,
 
 static unsigned brw_texture_is_referenced( struct pipe_context *pipe,
 					   struct pipe_resource *texture,
-					   unsigned face, 
-					   unsigned level )
+					   unsigned level,
+					   int layer )
 {
    struct brw_context *brw = brw_context(pipe);
    struct brw_screen *bscreen = brw_screen(pipe->screen);
@@ -246,7 +246,7 @@ static unsigned brw_texture_is_referenced( struct pipe_context *pipe,
    if (bscreen->sws->bo_references( batch_bo, tex->bo ))
       return PIPE_REFERENCED_FOR_READ | PIPE_REFERENCED_FOR_WRITE;
 
-   /* Find any view on this texture for this face/level and see if it
+   /* Find any view on this texture for this level/layer and see if it
     * is referenced:
     */
    for (i = 0; i < 2; i++) {
@@ -254,7 +254,7 @@ static unsigned brw_texture_is_referenced( struct pipe_context *pipe,
          if (surf->bo == tex->bo)
             continue;
 
-         if (surf->id.bits.face != face ||
+         if (!(layer == -1 || surf->id.bits.layer == layer) ||
              surf->id.bits.level != level)
             continue;
          
@@ -274,10 +274,10 @@ static unsigned brw_texture_is_referenced( struct pipe_context *pipe,
 
 static struct pipe_transfer * 
 brw_texture_get_transfer(struct pipe_context *context,
-			  struct pipe_resource *resource,
-			  struct pipe_subresource sr,
-			  unsigned usage,
-			  const struct pipe_box *box)
+                         struct pipe_resource *resource,
+                         unsigned level,
+                         unsigned usage,
+                         const struct pipe_box *box)
 {
    struct brw_texture *tex = brw_texture(resource);
    struct pipe_transfer *transfer = CALLOC_STRUCT(pipe_transfer);
@@ -285,10 +285,11 @@ brw_texture_get_transfer(struct pipe_context *context,
       return NULL;
 
    transfer->resource = resource;
-   transfer->sr = sr;
+   transfer->level = level;
    transfer->usage = usage;
    transfer->box = *box;
    transfer->stride = tex->pitch * tex->cpp;
+   /* FIXME: layer_stride */
 
    return transfer;
 }
@@ -301,24 +302,16 @@ brw_texture_transfer_map(struct pipe_context *pipe,
    struct pipe_resource *resource = transfer->resource;
    struct brw_texture *tex = brw_texture(transfer->resource);
    struct brw_winsys_screen *sws = brw_screen(pipe->screen)->sws;
-   struct pipe_subresource sr = transfer->sr;
    struct pipe_box *box = &transfer->box;
    enum pipe_format format = resource->format;
    unsigned usage = transfer->usage;
    unsigned offset;
    char *map;
 
-   if (resource->target == PIPE_TEXTURE_CUBE) {
-      offset = tex->image_offset[sr.level][sr.face];
-   }
-   else if (resource->target == PIPE_TEXTURE_3D) {
-      offset = tex->image_offset[sr.level][box->z];
-   }
-   else {
-      offset = tex->image_offset[sr.level][0];
-      assert(sr.face == 0);
+   if (resource->target != PIPE_TEXTURE_3D &&
+       resource->target != PIPE_TEXTURE_CUBE)
       assert(box->z == 0);
-   }
+   offset = tex->image_offset[transfer->level][box->z];
 
    map = sws->bo_map(tex->bo, 
                      BRW_DATA_OTHER,
diff --git a/src/gallium/drivers/i965/brw_screen.c b/src/gallium/drivers/i965/brw_screen.c
index 57160ebb29..f5b75b17e3 100644
--- a/src/gallium/drivers/i965/brw_screen.c
+++ b/src/gallium/drivers/i965/brw_screen.c
@@ -240,6 +240,8 @@ brw_get_shader_param(struct pipe_screen *screen, unsigned shader, enum pipe_shad
       case PIPE_SHADER_CAP_INDIRECT_TEMP_ADDR:
       case PIPE_SHADER_CAP_INDIRECT_CONST_ADDR:
           return 1;
+      case PIPE_SHADER_CAP_SUBROUTINES:
+          return 1;
       default:
          assert(0);
          return 0;
@@ -468,7 +470,6 @@ brw_screen_create(struct brw_winsys_screen *sws)
    bscreen->base.fence_finish = brw_fence_finish;
 
    brw_init_screen_resource_functions(bscreen);
-   brw_screen_tex_surface_init(bscreen);
 
    bscreen->no_tiling = debug_get_option("BRW_NO_TILING", FALSE) != NULL;
    
diff --git a/src/gallium/drivers/i965/brw_screen.h b/src/gallium/drivers/i965/brw_screen.h
index 522a3bf899..58e293bc76 100644
--- a/src/gallium/drivers/i965/brw_screen.h
+++ b/src/gallium/drivers/i965/brw_screen.h
@@ -52,9 +52,8 @@ struct brw_screen
 
 union brw_surface_id {
    struct {
-      unsigned face:3;
-      unsigned zslice:13;
       unsigned level:16;
+      unsigned layer:16;
    } bits;
    unsigned value;
 };
@@ -63,8 +62,9 @@ union brw_surface_id {
 struct brw_surface
 {
    struct pipe_surface base;
-   
+
    union brw_surface_id id;
+   unsigned offset;
    unsigned cpp;
    unsigned pitch;
    unsigned draw_offset;
@@ -96,7 +96,5 @@ brw_surface(struct pipe_surface *surface)
 unsigned
 brw_surface_pitch( const struct pipe_surface *surface );
 
-void brw_screen_tex_surface_init( struct brw_screen *brw_screen );
-
 
 #endif /* BRW_SCREEN_H */
diff --git a/src/gallium/drivers/i965/brw_state.h b/src/gallium/drivers/i965/brw_state.h
index d2bbd0123d..380d511f9b 100644
--- a/src/gallium/drivers/i965/brw_state.h
+++ b/src/gallium/drivers/i965/brw_state.h
@@ -56,7 +56,7 @@ const struct brw_tracked_state brw_clip_prog;
 const struct brw_tracked_state brw_clip_unit;
 const struct brw_tracked_state brw_curbe_buffer;
 const struct brw_tracked_state brw_curbe_offsets;
-const struct brw_tracked_state brw_invarient_state;
+const struct brw_tracked_state brw_invariant_state;
 const struct brw_tracked_state brw_gs_prog;
 const struct brw_tracked_state brw_gs_unit;
 const struct brw_tracked_state brw_line_stipple;
diff --git a/src/gallium/drivers/i965/brw_state_upload.c b/src/gallium/drivers/i965/brw_state_upload.c
index f8b91eff81..cdbf270e06 100644
--- a/src/gallium/drivers/i965/brw_state_upload.c
+++ b/src/gallium/drivers/i965/brw_state_upload.c
@@ -69,7 +69,7 @@ const struct brw_tracked_state *atoms[] =
 
    /* Command packets:
     */
-   &brw_invarient_state,
+   &brw_invariant_state,
    &brw_state_base_address,
 
    &brw_binding_table_pointers,
diff --git a/src/gallium/drivers/i965/brw_wm_fp.c b/src/gallium/drivers/i965/brw_wm_fp.c
index f7ee55cc1c..a65e16edec 100644
--- a/src/gallium/drivers/i965/brw_wm_fp.c
+++ b/src/gallium/drivers/i965/brw_wm_fp.c
@@ -812,7 +812,7 @@ static void precalc_tex( struct brw_wm_compile *c,
    }
 
    /* XXX: add GL_EXT_texture_swizzle support to gallium -- by
-    * generating shader varients in mesa state tracker.
+    * generating shader variants in mesa state tracker.
     */
 
    /* Release this temp if we ended up allocating it:
diff --git a/src/gallium/drivers/identity/SConscript b/src/gallium/drivers/identity/SConscript
index b364e0acc8..d24d1ec7c6 100644
--- a/src/gallium/drivers/identity/SConscript
+++ b/src/gallium/drivers/identity/SConscript
@@ -10,4 +10,6 @@ identity = env.ConvenienceLibrary(
 		'id_screen.c',
 	])
 
+env.Alias('identity', identity)
+
 Export('identity')
diff --git a/src/gallium/drivers/identity/id_context.c b/src/gallium/drivers/identity/id_context.c
index de83c24905..3efbd6a246 100644
--- a/src/gallium/drivers/identity/id_context.c
+++ b/src/gallium/drivers/identity/id_context.c
@@ -577,17 +577,13 @@ identity_set_index_buffer(struct pipe_context *_pipe,
 static void
 identity_resource_copy_region(struct pipe_context *_pipe,
                               struct pipe_resource *_dst,
-                              struct pipe_subresource subdst,
+                              unsigned dst_level,
                               unsigned dstx,
                               unsigned dsty,
                               unsigned dstz,
                               struct pipe_resource *_src,
-                              struct pipe_subresource subsrc,
-                              unsigned srcx,
-                              unsigned srcy,
-                              unsigned srcz,
-                              unsigned width,
-                              unsigned height)
+                              unsigned src_level,
+                              const struct pipe_box *src_box)
 {
    struct identity_context *id_pipe = identity_context(_pipe);
    struct identity_resource *id_resource_dst = identity_resource(_dst);
@@ -598,17 +594,13 @@ identity_resource_copy_region(struct pipe_context *_pipe,
 
    pipe->resource_copy_region(pipe,
                               dst,
-                              subdst,
+                              dst_level,
                               dstx,
                               dsty,
                               dstz,
                               src,
-                              subsrc,
-                              srcx,
-                              srcy,
-                              srcz,
-                              width,
-                              height);
+                              src_level,
+                              src_box);
 }
 
 static void
@@ -690,8 +682,8 @@ identity_flush(struct pipe_context *_pipe,
 static unsigned int
 identity_is_resource_referenced(struct pipe_context *_pipe,
                                 struct pipe_resource *_resource,
-                                unsigned face,
-                                unsigned level)
+                                unsigned level,
+                                int layer)
 {
    struct identity_context *id_pipe = identity_context(_pipe);
    struct identity_resource *id_resource = identity_resource(_resource);
@@ -700,8 +692,8 @@ identity_is_resource_referenced(struct pipe_context *_pipe,
 
    return pipe->is_resource_referenced(pipe,
                                        resource,
-                                       face,
-                                       level);
+                                       level,
+                                       layer);
 }
 
 static struct pipe_sampler_view *
@@ -732,10 +724,38 @@ identity_context_sampler_view_destroy(struct pipe_context *_pipe,
                                  identity_sampler_view(_view));
 }
 
+static struct pipe_surface *
+identity_context_create_surface(struct pipe_context *_pipe,
+                                struct pipe_resource *_resource,
+                                const struct pipe_surface *templ)
+{
+   struct identity_context *id_context = identity_context(_pipe);
+   struct identity_resource *id_resource = identity_resource(_resource);
+   struct pipe_context *pipe = id_context->pipe;
+   struct pipe_resource *resource = id_resource->resource;
+   struct pipe_surface *result;
+
+   result = pipe->create_surface(pipe,
+                                 resource,
+                                 templ);
+
+   if (result)
+      return identity_surface_create(id_context, id_resource, result);
+   return NULL;
+}
+
+static void
+identity_context_surface_destroy(struct pipe_context *_pipe,
+                                 struct pipe_surface *_surf)
+{
+   identity_surface_destroy(identity_context(_pipe),
+                            identity_surface(_surf));
+}
+
 static struct pipe_transfer *
 identity_context_get_transfer(struct pipe_context *_context,
                               struct pipe_resource *_resource,
-                              struct pipe_subresource sr,
+                              unsigned level,
                               unsigned usage,
                               const struct pipe_box *box)
 {
@@ -747,7 +767,7 @@ identity_context_get_transfer(struct pipe_context *_context,
 
    result = context->get_transfer(context,
                                   resource,
-                                  sr,
+                                  level,
                                   usage,
                                   box);
 
@@ -812,12 +832,12 @@ identity_context_transfer_unmap(struct pipe_context *_context,
 static void 
 identity_context_transfer_inline_write(struct pipe_context *_context,
                                        struct pipe_resource *_resource,
-                                       struct pipe_subresource sr,
+                                       unsigned level,
                                        unsigned usage,
                                        const struct pipe_box *box,
                                        const void *data,
                                        unsigned stride,
-                                       unsigned slice_stride)
+                                       unsigned layer_stride)
 {
    struct identity_context *id_context = identity_context(_context);
    struct identity_resource *id_resource = identity_resource(_resource);
@@ -826,12 +846,12 @@ identity_context_transfer_inline_write(struct pipe_context *_context,
 
    context->transfer_inline_write(context,
                                   resource,
-                                  sr,
+                                  level,
                                   usage,
                                   box,
                                   data,
                                   stride,
-                                  slice_stride);
+                                  layer_stride);
 }
 
 
@@ -899,6 +919,8 @@ identity_context_create(struct pipe_screen *_screen, struct pipe_context *pipe)
    id_pipe->base.clear_depth_stencil = identity_clear_depth_stencil;
    id_pipe->base.flush = identity_flush;
    id_pipe->base.is_resource_referenced = identity_is_resource_referenced;
+   id_pipe->base.create_surface = identity_context_create_surface;
+   id_pipe->base.surface_destroy = identity_context_surface_destroy;
    id_pipe->base.create_sampler_view = identity_context_create_sampler_view;
    id_pipe->base.sampler_view_destroy = identity_context_sampler_view_destroy;
    id_pipe->base.get_transfer = identity_context_get_transfer;
diff --git a/src/gallium/drivers/identity/id_objects.c b/src/gallium/drivers/identity/id_objects.c
index 593928f399..6345441052 100644
--- a/src/gallium/drivers/identity/id_objects.c
+++ b/src/gallium/drivers/identity/id_objects.c
@@ -71,7 +71,8 @@ identity_resource_destroy(struct identity_resource *id_resource)
 
 
 struct pipe_surface *
-identity_surface_create(struct identity_resource *id_resource,
+identity_surface_create(struct identity_context *id_context,
+                        struct identity_resource *id_resource,
                         struct pipe_surface *surface)
 {
    struct identity_surface *id_surface;
@@ -100,10 +101,12 @@ error:
 }
 
 void
-identity_surface_destroy(struct identity_surface *id_surface)
+identity_surface_destroy(struct identity_context *id_context,
+                         struct identity_surface *id_surface)
 {
    pipe_resource_reference(&id_surface->base.texture, NULL);
-   pipe_surface_reference(&id_surface->surface, NULL);
+   id_context->pipe->surface_destroy(id_context->pipe,
+                                     id_surface->surface);
    FREE(id_surface);
 }
 
diff --git a/src/gallium/drivers/identity/id_objects.h b/src/gallium/drivers/identity/id_objects.h
index e8deabf4fc..181f2d6623 100644
--- a/src/gallium/drivers/identity/id_objects.h
+++ b/src/gallium/drivers/identity/id_objects.h
@@ -147,11 +147,13 @@ void
 identity_resource_destroy(struct identity_resource *id_resource);
 
 struct pipe_surface *
-identity_surface_create(struct identity_resource *id_resource,
+identity_surface_create(struct identity_context *id_context,
+                        struct identity_resource *id_resource,
                         struct pipe_surface *surface);
 
 void
-identity_surface_destroy(struct identity_surface *id_surface);
+identity_surface_destroy(struct identity_context *id_context,
+                         struct identity_surface *id_surface);
 
 struct pipe_sampler_view *
 identity_sampler_view_create(struct identity_context *id_context,
diff --git a/src/gallium/drivers/identity/id_screen.c b/src/gallium/drivers/identity/id_screen.c
index 5fb464b414..644481bb74 100644
--- a/src/gallium/drivers/identity/id_screen.c
+++ b/src/gallium/drivers/identity/id_screen.c
@@ -189,39 +189,6 @@ identity_screen_resource_destroy(struct pipe_screen *screen,
    identity_resource_destroy(identity_resource(_resource));
 }
 
-static struct pipe_surface *
-identity_screen_get_tex_surface(struct pipe_screen *_screen,
-                                struct pipe_resource *_resource,
-                                unsigned face,
-                                unsigned level,
-                                unsigned zslice,
-                                unsigned usage)
-{
-   struct identity_screen *id_screen = identity_screen(_screen);
-   struct identity_resource *id_resource = identity_resource(_resource);
-   struct pipe_screen *screen = id_screen->screen;
-   struct pipe_resource *resource = id_resource->resource;
-   struct pipe_surface *result;
-
-   result = screen->get_tex_surface(screen,
-                                    resource,
-                                    face,
-                                    level,
-                                    zslice,
-                                    usage);
-
-   if (result)
-      return identity_surface_create(id_resource, result);
-   return NULL;
-}
-
-static void
-identity_screen_tex_surface_destroy(struct pipe_surface *_surface)
-{
-   identity_surface_destroy(identity_surface(_surface));
-}
-
-
 
 static struct pipe_resource *
 identity_screen_user_buffer_create(struct pipe_screen *_screen,
@@ -247,16 +214,18 @@ identity_screen_user_buffer_create(struct pipe_screen *_screen,
 
 static void
 identity_screen_flush_frontbuffer(struct pipe_screen *_screen,
-                                  struct pipe_surface *_surface,
+                                  struct pipe_resource *_resource,
+                                  unsigned level, unsigned layer,
                                   void *context_private)
 {
    struct identity_screen *id_screen = identity_screen(_screen);
-   struct identity_surface *id_surface = identity_surface(_surface);
+   struct identity_resource *id_resource = identity_resource(_resource);
    struct pipe_screen *screen = id_screen->screen;
-   struct pipe_surface *surface = id_surface->surface;
+   struct pipe_resource *resource = id_resource->resource;
 
    screen->flush_frontbuffer(screen,
-                             surface,
+                             resource,
+                             level, layer,
                              context_private);
 }
 
@@ -323,8 +292,6 @@ identity_screen_create(struct pipe_screen *screen)
    id_screen->base.resource_from_handle = identity_screen_resource_from_handle;
    id_screen->base.resource_get_handle = identity_screen_resource_get_handle;
    id_screen->base.resource_destroy = identity_screen_resource_destroy;
-   id_screen->base.get_tex_surface = identity_screen_get_tex_surface;
-   id_screen->base.tex_surface_destroy = identity_screen_tex_surface_destroy;
    id_screen->base.user_buffer_create = identity_screen_user_buffer_create;
    id_screen->base.flush_frontbuffer = identity_screen_flush_frontbuffer;
    id_screen->base.fence_reference = identity_screen_fence_reference;
diff --git a/src/gallium/drivers/llvmpipe/Makefile b/src/gallium/drivers/llvmpipe/Makefile
index 669e42e300..4068bed393 100644
--- a/src/gallium/drivers/llvmpipe/Makefile
+++ b/src/gallium/drivers/llvmpipe/Makefile
@@ -3,8 +3,6 @@ include $(TOP)/configs/current
 
 LIBNAME = llvmpipe
 
-DEFINES += -D__STDC_CONSTANT_MACROS -D__STDC_LIMIT_MACROS
-
 C_SOURCES = \
 	lp_bld_alpha.c \
 	lp_bld_blend_aos.c \
diff --git a/src/gallium/drivers/llvmpipe/lp_bld_alpha.c b/src/gallium/drivers/llvmpipe/lp_bld_alpha.c
index e50643790c..518969c320 100644
--- a/src/gallium/drivers/llvmpipe/lp_bld_alpha.c
+++ b/src/gallium/drivers/llvmpipe/lp_bld_alpha.c
@@ -43,7 +43,7 @@
 
 
 void
-lp_build_alpha_test(LLVMBuilderRef builder,
+lp_build_alpha_test(struct gallivm_state *gallivm,
                     unsigned func,
                     struct lp_type type,
                     struct lp_build_mask_context *mask,
@@ -54,7 +54,7 @@ lp_build_alpha_test(LLVMBuilderRef builder,
    struct lp_build_context bld;
    LLVMValueRef test;
 
-   lp_build_context_init(&bld, builder, type);
+   lp_build_context_init(&bld, gallivm, type);
 
    test = lp_build_cmp(&bld, func, alpha, ref);
 
diff --git a/src/gallium/drivers/llvmpipe/lp_bld_alpha.h b/src/gallium/drivers/llvmpipe/lp_bld_alpha.h
index 27ca8aad4d..5c9392504f 100644
--- a/src/gallium/drivers/llvmpipe/lp_bld_alpha.h
+++ b/src/gallium/drivers/llvmpipe/lp_bld_alpha.h
@@ -43,7 +43,7 @@ struct lp_build_mask_context;
 
 
 void
-lp_build_alpha_test(LLVMBuilderRef builder,
+lp_build_alpha_test(struct gallivm_state *gallivm,
                     unsigned func,
                     struct lp_type type,
                     struct lp_build_mask_context *mask,
diff --git a/src/gallium/drivers/llvmpipe/lp_bld_blend.h b/src/gallium/drivers/llvmpipe/lp_bld_blend.h
index 5cecec3d7f..f82ae30bb7 100644
--- a/src/gallium/drivers/llvmpipe/lp_bld_blend.h
+++ b/src/gallium/drivers/llvmpipe/lp_bld_blend.h
@@ -30,6 +30,7 @@
 
 
 #include "gallivm/lp_bld.h"
+#include "gallivm/lp_bld_init.h"
  
 #include "pipe/p_format.h"
 
@@ -61,7 +62,7 @@ lp_build_blend_func(struct lp_build_context *bld,
 
 
 LLVMValueRef
-lp_build_blend_aos(LLVMBuilderRef builder,
+lp_build_blend_aos(struct gallivm_state *gallivm,
                    const struct pipe_blend_state *blend,
                    struct lp_type type,
                    unsigned rt,
@@ -72,7 +73,7 @@ lp_build_blend_aos(LLVMBuilderRef builder,
 
 
 void
-lp_build_blend_soa(LLVMBuilderRef builder,
+lp_build_blend_soa(struct gallivm_state *gallivm,
                    const struct pipe_blend_state *blend,
                    struct lp_type type,
                    unsigned rt,
diff --git a/src/gallium/drivers/llvmpipe/lp_bld_blend_aos.c b/src/gallium/drivers/llvmpipe/lp_bld_blend_aos.c
index d1c9b88f9b..c342346a36 100644
--- a/src/gallium/drivers/llvmpipe/lp_bld_blend_aos.c
+++ b/src/gallium/drivers/llvmpipe/lp_bld_blend_aos.c
@@ -301,7 +301,7 @@ lp_build_blend_func(struct lp_build_context *bld,
 
 
 LLVMValueRef
-lp_build_blend_aos(LLVMBuilderRef builder,
+lp_build_blend_aos(struct gallivm_state *gallivm,
                    const struct pipe_blend_state *blend,
                    struct lp_type type,
                    unsigned rt,
@@ -322,7 +322,7 @@ lp_build_blend_aos(LLVMBuilderRef builder,
 
    /* Setup build context */
    memset(&bld, 0, sizeof bld);
-   lp_build_context_init(&bld.base, builder, type);
+   lp_build_context_init(&bld.base, gallivm, type);
    bld.src = src;
    bld.dst = dst;
    bld.const_ = const_;
diff --git a/src/gallium/drivers/llvmpipe/lp_bld_blend_soa.c b/src/gallium/drivers/llvmpipe/lp_bld_blend_soa.c
index 30d261e979..4d5bc9642d 100644
--- a/src/gallium/drivers/llvmpipe/lp_bld_blend_soa.c
+++ b/src/gallium/drivers/llvmpipe/lp_bld_blend_soa.c
@@ -73,6 +73,7 @@
 
 #include "gallivm/lp_bld_type.h"
 #include "gallivm/lp_bld_arit.h"
+#include "gallivm/lp_bld_init.h"
 #include "lp_bld_blend.h"
 
 
@@ -211,7 +212,7 @@ lp_build_blend_factor_complementary(unsigned src_factor, unsigned dst_factor)
  * \param res  the result/output
  */
 void
-lp_build_blend_soa(LLVMBuilderRef builder,
+lp_build_blend_soa(struct gallivm_state *gallivm,
                    const struct pipe_blend_state *blend,
                    struct lp_type type,
                    unsigned rt,
@@ -220,6 +221,7 @@ lp_build_blend_soa(LLVMBuilderRef builder,
                    LLVMValueRef con[4],
                    LLVMValueRef res[4])
 {
+   LLVMBuilderRef builder = gallivm->builder;
    struct lp_build_blend_soa_context bld;
    unsigned i, j, k;
 
@@ -227,7 +229,7 @@ lp_build_blend_soa(LLVMBuilderRef builder,
 
    /* Setup build context */
    memset(&bld, 0, sizeof bld);
-   lp_build_context_init(&bld.base, builder, type);
+   lp_build_context_init(&bld.base, gallivm, type);
    for (i = 0; i < 4; ++i) {
       bld.src[i] = src[i];
       bld.dst[i] = dst[i];
diff --git a/src/gallium/drivers/llvmpipe/lp_bld_depth.c b/src/gallium/drivers/llvmpipe/lp_bld_depth.c
index 7eb76d4fb3..1bf741194c 100644
--- a/src/gallium/drivers/llvmpipe/lp_bld_depth.c
+++ b/src/gallium/drivers/llvmpipe/lp_bld_depth.c
@@ -97,6 +97,7 @@ lp_build_stencil_test_single(struct lp_build_context *bld,
                              LLVMValueRef stencilRef,
                              LLVMValueRef stencilVals)
 {
+   LLVMBuilderRef builder = bld->gallivm->builder;
    const unsigned stencilMax = 255; /* XXX fix */
    struct lp_type type = bld->type;
    LLVMValueRef res;
@@ -107,10 +108,10 @@ lp_build_stencil_test_single(struct lp_build_context *bld,
 
    if (stencil->valuemask != stencilMax) {
       /* compute stencilRef = stencilRef & valuemask */
-      LLVMValueRef valuemask = lp_build_const_int_vec(type, stencil->valuemask);
-      stencilRef = LLVMBuildAnd(bld->builder, stencilRef, valuemask, "");
+      LLVMValueRef valuemask = lp_build_const_int_vec(bld->gallivm, type, stencil->valuemask);
+      stencilRef = LLVMBuildAnd(builder, stencilRef, valuemask, "");
       /* compute stencilVals = stencilVals & valuemask */
-      stencilVals = LLVMBuildAnd(bld->builder, stencilVals, valuemask, "");
+      stencilVals = LLVMBuildAnd(builder, stencilVals, valuemask, "");
    }
 
    res = lp_build_cmp(bld, stencil->func, stencilRef, stencilVals);
@@ -167,9 +168,10 @@ lp_build_stencil_op_single(struct lp_build_context *bld,
                            LLVMValueRef stencilVals)
 
 {
+   LLVMBuilderRef builder = bld->gallivm->builder;
    struct lp_type type = bld->type;
    LLVMValueRef res;
-   LLVMValueRef max = lp_build_const_int_vec(type, 0xff);
+   LLVMValueRef max = lp_build_const_int_vec(bld->gallivm, type, 0xff);
    unsigned stencil_op;
 
    assert(type.sign);
@@ -210,15 +212,15 @@ lp_build_stencil_op_single(struct lp_build_context *bld,
       break;
    case PIPE_STENCIL_OP_INCR_WRAP:
       res = lp_build_add(bld, stencilVals, bld->one);
-      res = LLVMBuildAnd(bld->builder, res, max, "");
+      res = LLVMBuildAnd(builder, res, max, "");
       break;
    case PIPE_STENCIL_OP_DECR_WRAP:
       res = lp_build_sub(bld, stencilVals, bld->one);
-      res = LLVMBuildAnd(bld->builder, res, max, "");
+      res = LLVMBuildAnd(builder, res, max, "");
       break;
    case PIPE_STENCIL_OP_INVERT:
-      res = LLVMBuildNot(bld->builder, stencilVals, "");
-      res = LLVMBuildAnd(bld->builder, res, max, "");
+      res = LLVMBuildNot(builder, stencilVals, "");
+      res = LLVMBuildAnd(builder, res, max, "");
       break;
    default:
       assert(0 && "bad stencil op mode");
@@ -242,6 +244,7 @@ lp_build_stencil_op(struct lp_build_context *bld,
                     LLVMValueRef front_facing)
 
 {
+   LLVMBuilderRef builder = bld->gallivm->builder;
    LLVMValueRef res;
 
    assert(stencil[0].enabled);
@@ -262,10 +265,11 @@ lp_build_stencil_op(struct lp_build_context *bld,
 
    if (stencil->writemask != 0xff) {
       /* mask &= stencil->writemask */
-      LLVMValueRef writemask = lp_build_const_int_vec(bld->type, stencil->writemask);
-      mask = LLVMBuildAnd(bld->builder, mask, writemask, "");
+      LLVMValueRef writemask = lp_build_const_int_vec(bld->gallivm, bld->type,
+                                                      stencil->writemask);
+      mask = LLVMBuildAnd(builder, mask, writemask, "");
       /* res = (res & mask) | (stencilVals & ~mask) */
-      res = lp_build_select_bitwise(bld, writemask, res, stencilVals);
+      res = lp_build_select_bitwise(bld, mask, res, stencilVals);
    }
    else {
       /* res = mask ? res : stencilVals */
@@ -411,25 +415,27 @@ get_s_shift_and_mask(const struct util_format_description *format_desc,
  * \param counter is a pointer of the uint32 counter.
  */
 void
-lp_build_occlusion_count(LLVMBuilderRef builder,
+lp_build_occlusion_count(struct gallivm_state *gallivm,
                          struct lp_type type,
                          LLVMValueRef maskvalue,
                          LLVMValueRef counter)
 {
-   LLVMValueRef countmask = lp_build_const_int_vec(type, 1);
+   LLVMBuilderRef builder = gallivm->builder;
+   LLVMContextRef context = gallivm->context;
+   LLVMValueRef countmask = lp_build_const_int_vec(gallivm, type, 1);
    LLVMValueRef countv = LLVMBuildAnd(builder, maskvalue, countmask, "countv");
-   LLVMTypeRef i8v16 = LLVMVectorType(LLVMInt8Type(), 16);
+   LLVMTypeRef i8v16 = LLVMVectorType(LLVMInt8TypeInContext(context), 16);
    LLVMValueRef counti = LLVMBuildBitCast(builder, countv, i8v16, "counti");
    LLVMValueRef maskarray[4] = {
-      LLVMConstInt(LLVMInt32Type(), 0, 0),
-      LLVMConstInt(LLVMInt32Type(), 4, 0),
-      LLVMConstInt(LLVMInt32Type(), 8, 0),
-      LLVMConstInt(LLVMInt32Type(), 12, 0),
+      lp_build_const_int32(gallivm, 0),
+      lp_build_const_int32(gallivm, 4),
+      lp_build_const_int32(gallivm, 8),
+      lp_build_const_int32(gallivm, 12)
    };
    LLVMValueRef shufflemask = LLVMConstVector(maskarray, 4);
    LLVMValueRef shufflev =  LLVMBuildShuffleVector(builder, counti, LLVMGetUndef(i8v16), shufflemask, "shufflev");
-   LLVMValueRef shuffle = LLVMBuildBitCast(builder, shufflev, LLVMInt32Type(), "shuffle");
-   LLVMValueRef count = lp_build_intrinsic_unary(builder, "llvm.ctpop.i32", LLVMInt32Type(), shuffle);
+   LLVMValueRef shuffle = LLVMBuildBitCast(builder, shufflev, LLVMInt32TypeInContext(context), "shuffle");
+   LLVMValueRef count = lp_build_intrinsic_unary(builder, "llvm.ctpop.i32", LLVMInt32TypeInContext(context), shuffle);
    LLVMValueRef orig = LLVMBuildLoad(builder, counter, "orig");
    LLVMValueRef incr = LLVMBuildAdd(builder, orig, count, "incr");
    LLVMBuildStore(builder, incr, counter);
@@ -452,7 +458,7 @@ lp_build_occlusion_count(LLVMBuilderRef builder,
  * \param facing  contains boolean value indicating front/back facing polygon
  */
 void
-lp_build_depth_stencil_test(LLVMBuilderRef builder,
+lp_build_depth_stencil_test(struct gallivm_state *gallivm,
                             const struct pipe_depth_state *depth,
                             const struct pipe_stencil_state stencil[2],
                             struct lp_type z_src_type,
@@ -465,6 +471,7 @@ lp_build_depth_stencil_test(LLVMBuilderRef builder,
                             LLVMValueRef *zs_value,
                             boolean do_branch)
 {
+   LLVMBuilderRef builder = gallivm->builder;
    struct lp_type z_type;
    struct lp_build_context z_bld;
    struct lp_build_context s_bld;
@@ -537,11 +544,11 @@ lp_build_depth_stencil_test(LLVMBuilderRef builder,
 
 
    /* Setup build context for Z vals */
-   lp_build_context_init(&z_bld, builder, z_type);
+   lp_build_context_init(&z_bld, gallivm, z_type);
 
    /* Setup build context for stencil vals */
    s_type = lp_type_int_vec(z_type.width);
-   lp_build_context_init(&s_bld, builder, s_type);
+   lp_build_context_init(&s_bld, gallivm, s_type);
 
    /* Load current z/stencil value from z/stencil buffer */
    zs_dst_ptr = LLVMBuildBitCast(builder,
@@ -559,14 +566,14 @@ lp_build_depth_stencil_test(LLVMBuilderRef builder,
 
       if (get_z_shift_and_mask(format_desc, &z_shift, &z_width, &z_mask)) {
          if (z_mask != 0xffffffff) {
-            z_bitmask = lp_build_const_int_vec(z_type, z_mask);
+            z_bitmask = lp_build_const_int_vec(gallivm, z_type, z_mask);
          }
 
          /*
           * Align the framebuffer Z 's LSB to the right.
           */
          if (z_shift) {
-            LLVMValueRef shift = lp_build_const_int_vec(z_type, z_shift);
+            LLVMValueRef shift = lp_build_const_int_vec(gallivm, z_type, z_shift);
             z_dst = LLVMBuildLShr(builder, zs_dst, shift, "z_dst");
          } else if (z_bitmask) {
 	    /* TODO: Instead of loading a mask from memory and ANDing, it's
@@ -580,7 +587,7 @@ lp_build_depth_stencil_test(LLVMBuilderRef builder,
 
       if (get_s_shift_and_mask(format_desc, &s_shift, &s_mask)) {
          if (s_shift) {
-            LLVMValueRef shift = lp_build_const_int_vec(s_type, s_shift);
+            LLVMValueRef shift = lp_build_const_int_vec(gallivm, s_type, s_shift);
             stencil_vals = LLVMBuildLShr(builder, zs_dst, shift, "");
             stencil_shift = shift;  /* used below */
          }
@@ -589,7 +596,7 @@ lp_build_depth_stencil_test(LLVMBuilderRef builder,
          }
 
          if (s_mask != 0xffffffff) {
-            LLVMValueRef mask = lp_build_const_int_vec(s_type, s_mask);
+            LLVMValueRef mask = lp_build_const_int_vec(gallivm, s_type, s_mask);
             stencil_vals = LLVMBuildAnd(builder, stencil_vals, mask, "");
          }
 
@@ -600,12 +607,13 @@ lp_build_depth_stencil_test(LLVMBuilderRef builder,
    if (stencil[0].enabled) {
 
       if (face) {
-         LLVMValueRef zero = LLVMConstInt(LLVMInt32Type(), 0, 0);
+         LLVMValueRef zero = lp_build_const_int32(gallivm, 0);
 
          /* front_facing = face != 0 ? ~0 : 0 */
          front_facing = LLVMBuildICmp(builder, LLVMIntNE, face, zero, "");
          front_facing = LLVMBuildSExt(builder, front_facing,
-                                      LLVMIntType(s_bld.type.length*s_bld.type.width),
+                                      LLVMIntTypeInContext(gallivm->context,
+                                             s_bld.type.length*s_bld.type.width),
                                       "");
          front_facing = LLVMBuildBitCast(builder, front_facing,
                                          s_bld.int_vec_type, "");
@@ -642,7 +650,7 @@ lp_build_depth_stencil_test(LLVMBuilderRef builder,
           */
 
          if (!z_type.floating) {
-            z_src = lp_build_clamped_float_to_unsigned_norm(builder,
+            z_src = lp_build_clamped_float_to_unsigned_norm(gallivm,
                                                             z_src_type,
                                                             z_width,
                                                             z_src);
@@ -657,7 +665,7 @@ lp_build_depth_stencil_test(LLVMBuilderRef builder,
          assert(z_src_type.norm);
          assert(!z_type.floating);
          if (z_src_type.width > z_width) {
-            LLVMValueRef shift = lp_build_const_int_vec(z_src_type,
+            LLVMValueRef shift = lp_build_const_int_vec(gallivm, z_src_type,
                                                         z_src_type.width - z_width);
             z_src = LLVMBuildLShr(builder, z_src, shift, "");
          }
@@ -710,7 +718,7 @@ lp_build_depth_stencil_test(LLVMBuilderRef builder,
                                             z_fail_mask, front_facing);
 
          /* apply Z-pass operator */
-         z_pass_mask = LLVMBuildAnd(z_bld.builder, orig_mask, z_pass, "");
+         z_pass_mask = LLVMBuildAnd(builder, orig_mask, z_pass, "");
          stencil_vals = lp_build_stencil_op(&s_bld, stencil, Z_PASS_OP,
                                             stencil_refs, stencil_vals,
                                             z_pass_mask, front_facing);
@@ -720,7 +728,7 @@ lp_build_depth_stencil_test(LLVMBuilderRef builder,
       /* No depth test: apply Z-pass operator to stencil buffer values which
        * passed the stencil test.
        */
-      s_pass_mask = LLVMBuildAnd(s_bld.builder, orig_mask, s_pass_mask, "");
+      s_pass_mask = LLVMBuildAnd(builder, orig_mask, s_pass_mask, "");
       stencil_vals = lp_build_stencil_op(&s_bld, stencil, Z_PASS_OP,
                                          stencil_refs, stencil_vals,
                                          s_pass_mask, front_facing);
@@ -728,11 +736,11 @@ lp_build_depth_stencil_test(LLVMBuilderRef builder,
 
    /* Put Z and ztencil bits in the right place */
    if (z_dst && z_shift) {
-      LLVMValueRef shift = lp_build_const_int_vec(z_type, z_shift);
+      LLVMValueRef shift = lp_build_const_int_vec(gallivm, z_type, z_shift);
       z_dst = LLVMBuildShl(builder, z_dst, shift, "");
    }
    if (stencil_vals && stencil_shift)
-      stencil_vals = LLVMBuildShl(s_bld.builder, stencil_vals,
+      stencil_vals = LLVMBuildShl(builder, stencil_vals,
                                   stencil_shift, "");
 
    /* Finally, merge/store the z/stencil values */
@@ -740,7 +748,7 @@ lp_build_depth_stencil_test(LLVMBuilderRef builder,
        (stencil[0].enabled && stencil[0].writemask)) {
 
       if (z_dst && stencil_vals)
-         zs_dst = LLVMBuildOr(z_bld.builder, z_dst, stencil_vals, "");
+         zs_dst = LLVMBuildOr(builder, z_dst, stencil_vals, "");
       else if (z_dst)
          zs_dst = z_dst;
       else
@@ -775,7 +783,7 @@ lp_build_depth_write(LLVMBuilderRef builder,
 
 
 void
-lp_build_deferred_depth_write(LLVMBuilderRef builder,
+lp_build_deferred_depth_write(struct gallivm_state *gallivm,
                               struct lp_type z_src_type,
                               const struct util_format_description *format_desc,
                               struct lp_build_mask_context *mask,
@@ -785,11 +793,12 @@ lp_build_deferred_depth_write(LLVMBuilderRef builder,
    struct lp_type z_type;
    struct lp_build_context z_bld;
    LLVMValueRef z_dst;
+   LLVMBuilderRef builder = gallivm->builder;
 
    /* XXX: pointlessly redo type logic:
     */
    z_type = lp_depth_type(format_desc, z_src_type.width*z_src_type.length);
-   lp_build_context_init(&z_bld, builder, z_type);
+   lp_build_context_init(&z_bld, gallivm, z_type);
 
    zs_dst_ptr = LLVMBuildBitCast(builder, zs_dst_ptr,
                                  LLVMPointerType(z_bld.vec_type, 0), "");
diff --git a/src/gallium/drivers/llvmpipe/lp_bld_depth.h b/src/gallium/drivers/llvmpipe/lp_bld_depth.h
index a54ef3a711..038b136a28 100644
--- a/src/gallium/drivers/llvmpipe/lp_bld_depth.h
+++ b/src/gallium/drivers/llvmpipe/lp_bld_depth.h
@@ -51,7 +51,7 @@ lp_depth_type(const struct util_format_description *format_desc,
 
 
 void
-lp_build_depth_stencil_test(LLVMBuilderRef builder,
+lp_build_depth_stencil_test(struct gallivm_state *gallivm,
                             const struct pipe_depth_state *depth,
                             const struct pipe_stencil_state stencil[2],
                             struct lp_type type,
@@ -71,7 +71,7 @@ lp_build_depth_write(LLVMBuilderRef builder,
                      LLVMValueRef zs_value);
 
 void
-lp_build_deferred_depth_write(LLVMBuilderRef builder,
+lp_build_deferred_depth_write(struct gallivm_state *gallivm,
                               struct lp_type z_src_type,
                               const struct util_format_description *format_desc,
                               struct lp_build_mask_context *mask,
@@ -79,7 +79,7 @@ lp_build_deferred_depth_write(LLVMBuilderRef builder,
                               LLVMValueRef zs_value);
 
 void
-lp_build_occlusion_count(LLVMBuilderRef builder,
+lp_build_occlusion_count(struct gallivm_state *gallivm,
                          struct lp_type type,
                          LLVMValueRef maskvalue,
                          LLVMValueRef counter);
diff --git a/src/gallium/drivers/llvmpipe/lp_bld_interp.c b/src/gallium/drivers/llvmpipe/lp_bld_interp.c
index c9da8900d0..45ddf547bf 100644
--- a/src/gallium/drivers/llvmpipe/lp_bld_interp.c
+++ b/src/gallium/drivers/llvmpipe/lp_bld_interp.c
@@ -127,13 +127,14 @@ coeffs_init(struct lp_build_interp_soa_context *bld,
             LLVMValueRef dady_ptr)
 {
    struct lp_build_context *coeff_bld = &bld->coeff_bld;
-   LLVMBuilderRef builder = coeff_bld->builder;
+   struct gallivm_state *gallivm = coeff_bld->gallivm;
+   LLVMBuilderRef builder = gallivm->builder;
    LLVMValueRef zero = LLVMConstNull(coeff_bld->elem_type);
    LLVMValueRef one = LLVMConstReal(coeff_bld->elem_type, 1.0);
-   LLVMValueRef i0 = LLVMConstInt(LLVMInt32Type(), 0, 0);
-   LLVMValueRef i1 = LLVMConstInt(LLVMInt32Type(), 1, 0);
-   LLVMValueRef i2 = LLVMConstInt(LLVMInt32Type(), 2, 0);
-   LLVMValueRef i3 = LLVMConstInt(LLVMInt32Type(), 3, 0);
+   LLVMValueRef i0 = lp_build_const_int32(gallivm, 0);
+   LLVMValueRef i1 = lp_build_const_int32(gallivm, 1);
+   LLVMValueRef i2 = lp_build_const_int32(gallivm, 2);
+   LLVMValueRef i3 = lp_build_const_int32(gallivm, 3);
    unsigned attrib;
    unsigned chan;
 
@@ -144,7 +145,8 @@ coeffs_init(struct lp_build_interp_soa_context *bld,
       const unsigned interp = bld->interp[attrib];
       for (chan = 0; chan < NUM_CHANNELS; ++chan) {
          if (mask & (1 << chan)) {
-            LLVMValueRef index = LLVMConstInt(LLVMInt32Type(), attrib*NUM_CHANNELS + chan, 0);
+            LLVMValueRef index = lp_build_const_int32(gallivm,
+                                      attrib * NUM_CHANNELS + chan);
             LLVMValueRef a0 = zero;
             LLVMValueRef dadx = zero;
             LLVMValueRef dady = zero;
@@ -231,7 +233,7 @@ coeffs_init(struct lp_build_interp_soa_context *bld,
              * a = {a, a, a, a}
              */
 
-            a = lp_build_broadcast(builder, coeff_bld->vec_type, a);
+            a = lp_build_broadcast(gallivm, coeff_bld->vec_type, a);
 
             /*
              * Compute the attrib values on the upper-left corner of each quad.
@@ -273,12 +275,14 @@ coeffs_init(struct lp_build_interp_soa_context *bld,
  */
 static void
 attribs_update(struct lp_build_interp_soa_context *bld,
+               struct gallivm_state *gallivm,
                int quad_index,
                int start,
                int end)
 {
+   LLVMBuilderRef builder = gallivm->builder;
    struct lp_build_context *coeff_bld = &bld->coeff_bld;
-   LLVMValueRef shuffle = lp_build_const_int_vec(coeff_bld->type, quad_index);
+   LLVMValueRef shuffle = lp_build_const_int_vec(gallivm, coeff_bld->type, quad_index);
    LLVMValueRef oow = NULL;
    unsigned attrib;
    unsigned chan;
@@ -308,7 +312,7 @@ attribs_update(struct lp_build_interp_soa_context *bld,
                 * Broadcast the attribute value for this quad into all elements
                 */
 
-               a = LLVMBuildShuffleVector(coeff_bld->builder,
+               a = LLVMBuildShuffleVector(builder,
                                           a, coeff_bld->undef, shuffle, "");
 
                /*
@@ -380,10 +384,11 @@ pos_init(struct lp_build_interp_soa_context *bld,
          LLVMValueRef x0,
          LLVMValueRef y0)
 {
+   LLVMBuilderRef builder = bld->coeff_bld.gallivm->builder;
    struct lp_build_context *coeff_bld = &bld->coeff_bld;
 
-   bld->x = LLVMBuildSIToFP(coeff_bld->builder, x0, coeff_bld->elem_type, "");
-   bld->y = LLVMBuildSIToFP(coeff_bld->builder, y0, coeff_bld->elem_type, "");
+   bld->x = LLVMBuildSIToFP(builder, x0, coeff_bld->elem_type, "");
+   bld->y = LLVMBuildSIToFP(builder, y0, coeff_bld->elem_type, "");
 }
 
 
@@ -392,6 +397,7 @@ pos_init(struct lp_build_interp_soa_context *bld,
  */
 void
 lp_build_interp_soa_init(struct lp_build_interp_soa_context *bld,
+                         struct gallivm_state *gallivm,
                          unsigned num_inputs,
                          const struct lp_shader_input *inputs,
                          LLVMBuilderRef builder,
@@ -417,7 +423,7 @@ lp_build_interp_soa_init(struct lp_build_interp_soa_context *bld,
    /* XXX: we don't support interpolating into any other types */
    assert(memcmp(&coeff_type, &type, sizeof coeff_type) == 0);
 
-   lp_build_context_init(&bld->coeff_bld, builder, coeff_type);
+   lp_build_context_init(&bld->coeff_bld, gallivm, coeff_type);
 
    /* For convenience */
    bld->pos = bld->attribs[0];
@@ -453,19 +459,21 @@ lp_build_interp_soa_init(struct lp_build_interp_soa_context *bld,
  */
 void
 lp_build_interp_soa_update_inputs(struct lp_build_interp_soa_context *bld,
+                                  struct gallivm_state *gallivm,
                                   int quad_index)
 {
    assert(quad_index < 4);
 
-   attribs_update(bld, quad_index, 1, bld->num_attribs);
+   attribs_update(bld, gallivm, quad_index, 1, bld->num_attribs);
 }
 
 void
 lp_build_interp_soa_update_pos(struct lp_build_interp_soa_context *bld,
+                                  struct gallivm_state *gallivm,
                                   int quad_index)
 {
    assert(quad_index < 4);
 
-   attribs_update(bld, quad_index, 0, 1);
+   attribs_update(bld, gallivm, quad_index, 0, 1);
 }
 
diff --git a/src/gallium/drivers/llvmpipe/lp_bld_interp.h b/src/gallium/drivers/llvmpipe/lp_bld_interp.h
index a7ebdd1bfa..b58b2dc115 100644
--- a/src/gallium/drivers/llvmpipe/lp_bld_interp.h
+++ b/src/gallium/drivers/llvmpipe/lp_bld_interp.h
@@ -102,6 +102,7 @@ struct lp_build_interp_soa_context
 
 void
 lp_build_interp_soa_init(struct lp_build_interp_soa_context *bld,
+                         struct gallivm_state *gallivm,
                          unsigned num_inputs,
                          const struct lp_shader_input *inputs,
                          LLVMBuilderRef builder,
@@ -114,11 +115,13 @@ lp_build_interp_soa_init(struct lp_build_interp_soa_context *bld,
 
 void
 lp_build_interp_soa_update_inputs(struct lp_build_interp_soa_context *bld,
-                           int quad_index);
+                                  struct gallivm_state *gallivm,
+                                  int quad_index);
 
 void
 lp_build_interp_soa_update_pos(struct lp_build_interp_soa_context *bld,
-                           int quad_index);
+                               struct gallivm_state *gallivm,
+                               int quad_index);
 
 
 #endif /* LP_BLD_INTERP_H */
diff --git a/src/gallium/drivers/llvmpipe/lp_context.c b/src/gallium/drivers/llvmpipe/lp_context.c
index 763432ed71..2de20d6e9a 100644
--- a/src/gallium/drivers/llvmpipe/lp_context.c
+++ b/src/gallium/drivers/llvmpipe/lp_context.c
@@ -50,6 +50,46 @@
 DEBUG_GET_ONCE_BOOL_OPTION(lp_no_rast, "LP_NO_RAST", FALSE)
 
 
+/** shared by all contexts */
+unsigned llvmpipe_variant_count;
+
+
+/**
+ * This function is called by the gallivm "garbage collector" when
+ * the LLVM global data structures are freed.  We must free all LLVM-related
+ * data.  Specifically, all JIT'd shader variants.
+ */
+static void
+garbage_collect_callback(void *cb_data)
+{
+   struct llvmpipe_context *lp = (struct llvmpipe_context *) cb_data;
+   struct lp_fs_variant_list_item *li;
+
+   /* Free all the context's shader variants */
+   li = first_elem(&lp->fs_variants_list);
+   while (!at_end(&lp->fs_variants_list, li)) {
+      struct lp_fs_variant_list_item *next = next_elem(li);
+      llvmpipe_remove_shader_variant(lp, li->base);
+      li = next;
+   }
+
+   /* Free all the context's primitive setup variants */
+   lp_delete_setup_variants(lp);
+
+   /* release references to setup variants, shaders */
+   lp_setup_set_setup_variant(lp->setup, NULL);
+   lp_setup_set_fs_variant(lp->setup, NULL);
+   lp_setup_reset(lp->setup);
+
+   /* This type will be recreated upon demand */
+   lp->jit_context_ptr_type = NULL;
+
+   /* mark all state as dirty to ensure new shaders are jit'd, etc. */
+   lp->dirty = ~0;
+}
+
+
+
 static void llvmpipe_destroy( struct pipe_context *pipe )
 {
    struct llvmpipe_context *llvmpipe = llvmpipe_context( pipe );
@@ -57,6 +97,9 @@ static void llvmpipe_destroy( struct pipe_context *pipe )
 
    lp_print_counters();
 
+   gallivm_remove_garbage_collector_callback(garbage_collect_callback,
+                                             llvmpipe);
+
    /* This will also destroy llvmpipe->setup:
     */
    if (llvmpipe->draw)
@@ -82,7 +125,7 @@ static void llvmpipe_destroy( struct pipe_context *pipe )
       }
    }
 
-   lp_delete_setup_variants(llvmpipe);
+   gallivm_destroy(llvmpipe->gallivm);
 
    align_free( llvmpipe );
 }
@@ -110,8 +153,10 @@ llvmpipe_create_context( struct pipe_screen *screen, void *priv )
    memset(llvmpipe, 0, sizeof *llvmpipe);
 
    make_empty_list(&llvmpipe->fs_variants_list);
+
    make_empty_list(&llvmpipe->setup_variants_list);
 
+
    llvmpipe->pipe.winsys = screen->winsys;
    llvmpipe->pipe.screen = screen;
    llvmpipe->pipe.priv = priv;
@@ -136,10 +181,12 @@ llvmpipe_create_context( struct pipe_screen *screen, void *priv )
    llvmpipe_init_context_resource_funcs( &llvmpipe->pipe );
    llvmpipe_init_surface_functions(llvmpipe);
 
+   llvmpipe->gallivm = gallivm_create();
+
    /*
     * Create drawing context and plug our rendering stage into it.
     */
-   llvmpipe->draw = draw_create(&llvmpipe->pipe);
+   llvmpipe->draw = draw_create_gallivm(&llvmpipe->pipe, llvmpipe->gallivm);
    if (!llvmpipe->draw)
       goto fail;
 
@@ -173,6 +220,9 @@ llvmpipe_create_context( struct pipe_screen *screen, void *priv )
 
    lp_reset_counters();
 
+   gallivm_register_garbage_collector_callback(garbage_collect_callback,
+                                               llvmpipe);
+
    return &llvmpipe->pipe;
 
  fail:
diff --git a/src/gallium/drivers/llvmpipe/lp_context.h b/src/gallium/drivers/llvmpipe/lp_context.h
index db09c95b27..503f09d810 100644
--- a/src/gallium/drivers/llvmpipe/lp_context.h
+++ b/src/gallium/drivers/llvmpipe/lp_context.h
@@ -104,9 +104,18 @@ struct llvmpipe_context {
    /** Vertex format */
    struct vertex_info vertex_info;
    
+   /** Which vertex shader output slot contains color */
+   int color_slot[2];
+
+   /** Which vertex shader output slot contains bcolor */
+   int bcolor_slot[2];
+
    /** Which vertex shader output slot contains point size */
    int psize_slot;
 
+   /**< minimum resolvable depth value, for polygon offset */   
+   double mrd;
+   
    /** The tiling engine */
    struct lp_setup_context *setup;
    struct lp_setup_variant setup_variant;
@@ -117,14 +126,27 @@ struct llvmpipe_context {
    unsigned tex_timestamp;
    boolean no_rast;
 
+   /** List of all fragment shader variants */
    struct lp_fs_variant_list_item fs_variants_list;
    unsigned nr_fs_variants;
 
+   /** JIT code generation */
+   struct gallivm_state *gallivm;
+   LLVMTypeRef jit_context_ptr_type;
+
    struct lp_setup_variant_list_item setup_variants_list;
    unsigned nr_setup_variants;
 };
 
 
+/**
+ * Fragment and setup variant count, used to trigger garbage collection.
+ * This is global since all variants in all contexts will be free when
+ * we do garbage collection.
+ */
+extern unsigned llvmpipe_variant_count;
+
+
 struct pipe_context *
 llvmpipe_create_context( struct pipe_screen *screen, void *priv );
 
diff --git a/src/gallium/drivers/llvmpipe/lp_flush.c b/src/gallium/drivers/llvmpipe/lp_flush.c
index e2c723b7a8..85e3cdec82 100644
--- a/src/gallium/drivers/llvmpipe/lp_flush.c
+++ b/src/gallium/drivers/llvmpipe/lp_flush.c
@@ -56,6 +56,13 @@ llvmpipe_flush( struct pipe_context *pipe,
    /* ask the setup module to flush */
    lp_setup_flush(llvmpipe->setup, flags, fence, reason);
 
+
+   if (llvmpipe_variant_count > 1000) {
+      /* time to do a garbage collection */
+      gallivm_garbage_collect(llvmpipe->gallivm);
+      llvmpipe_variant_count = 0;
+   }
+
    /* Enable to dump BMPs of the color/depth buffers each frame */
    if (0) {
       if (flags & PIPE_FLUSH_FRAME) {
@@ -101,8 +108,8 @@ llvmpipe_finish( struct pipe_context *pipe,
 boolean
 llvmpipe_flush_resource(struct pipe_context *pipe,
                         struct pipe_resource *resource,
-                        unsigned face,
                         unsigned level,
+                        int layer,
                         unsigned flush_flags,
                         boolean read_only,
                         boolean cpu_access,
@@ -111,7 +118,7 @@ llvmpipe_flush_resource(struct pipe_context *pipe,
 {
    unsigned referenced;
 
-   referenced = pipe->is_resource_referenced(pipe, resource, face, level);
+   referenced = pipe->is_resource_referenced(pipe, resource, level, layer);
 
    if ((referenced & PIPE_REFERENCED_FOR_WRITE) ||
        ((referenced & PIPE_REFERENCED_FOR_READ) && !read_only)) {
diff --git a/src/gallium/drivers/llvmpipe/lp_flush.h b/src/gallium/drivers/llvmpipe/lp_flush.h
index 3626ce4a86..579d24c68a 100644
--- a/src/gallium/drivers/llvmpipe/lp_flush.h
+++ b/src/gallium/drivers/llvmpipe/lp_flush.h
@@ -47,8 +47,8 @@ llvmpipe_finish( struct pipe_context *pipe,
 boolean
 llvmpipe_flush_resource(struct pipe_context *pipe,
                         struct pipe_resource *resource,
-                        unsigned face,
                         unsigned level,
+                        int layer,
                         unsigned flush_flags,
                         boolean read_only,
                         boolean cpu_access,
diff --git a/src/gallium/drivers/llvmpipe/lp_jit.c b/src/gallium/drivers/llvmpipe/lp_jit.c
index c540f9b362..a775990f92 100644
--- a/src/gallium/drivers/llvmpipe/lp_jit.c
+++ b/src/gallium/drivers/llvmpipe/lp_jit.c
@@ -33,82 +33,85 @@
  */
 
 
-#include <llvm-c/Transforms/Scalar.h>
-
 #include "util/u_memory.h"
 #include "gallivm/lp_bld_init.h"
 #include "gallivm/lp_bld_debug.h"
-#include "lp_screen.h"
 #include "gallivm/lp_bld_intr.h"
+#include "lp_context.h"
+#include "lp_screen.h"
 #include "lp_jit.h"
 
 
 static void
-lp_jit_init_globals(struct llvmpipe_screen *screen)
+lp_jit_create_types(struct llvmpipe_context *lp)
 {
+   struct gallivm_state *gallivm = lp->gallivm;
+   LLVMContextRef lc = gallivm->context;
    LLVMTypeRef texture_type;
 
    /* struct lp_jit_texture */
    {
       LLVMTypeRef elem_types[LP_JIT_TEXTURE_NUM_FIELDS];
 
-      elem_types[LP_JIT_TEXTURE_WIDTH]  = LLVMInt32Type();
-      elem_types[LP_JIT_TEXTURE_HEIGHT] = LLVMInt32Type();
-      elem_types[LP_JIT_TEXTURE_DEPTH] = LLVMInt32Type();
-      elem_types[LP_JIT_TEXTURE_LAST_LEVEL] = LLVMInt32Type();
+      elem_types[LP_JIT_TEXTURE_WIDTH]  =
+      elem_types[LP_JIT_TEXTURE_HEIGHT] =
+      elem_types[LP_JIT_TEXTURE_DEPTH] =
+      elem_types[LP_JIT_TEXTURE_LAST_LEVEL] =  LLVMInt32TypeInContext(lc);
       elem_types[LP_JIT_TEXTURE_ROW_STRIDE] =
-         LLVMArrayType(LLVMInt32Type(), LP_MAX_TEXTURE_LEVELS);
       elem_types[LP_JIT_TEXTURE_IMG_STRIDE] =
-         LLVMArrayType(LLVMInt32Type(), LP_MAX_TEXTURE_LEVELS);
+         LLVMArrayType(LLVMInt32TypeInContext(lc), LP_MAX_TEXTURE_LEVELS);
       elem_types[LP_JIT_TEXTURE_DATA] =
-         LLVMArrayType(LLVMPointerType(LLVMInt8Type(), 0),
+         LLVMArrayType(LLVMPointerType(LLVMInt8TypeInContext(lc), 0),
                        LP_MAX_TEXTURE_LEVELS);
-      elem_types[LP_JIT_TEXTURE_MIN_LOD] = LLVMFloatType();
-      elem_types[LP_JIT_TEXTURE_MAX_LOD] = LLVMFloatType();
-      elem_types[LP_JIT_TEXTURE_LOD_BIAS] = LLVMFloatType();
+      elem_types[LP_JIT_TEXTURE_MIN_LOD] =
+      elem_types[LP_JIT_TEXTURE_MAX_LOD] =
+      elem_types[LP_JIT_TEXTURE_LOD_BIAS] = LLVMFloatTypeInContext(lc);
       elem_types[LP_JIT_TEXTURE_BORDER_COLOR] = 
-         LLVMArrayType(LLVMFloatType(), 4);
+         LLVMArrayType(LLVMFloatTypeInContext(lc), 4);
 
-      texture_type = LLVMStructType(elem_types, Elements(elem_types), 0);
+      texture_type = LLVMStructTypeInContext(lc, elem_types,
+                                             Elements(elem_types), 0);
+
+      LLVMInvalidateStructLayout(gallivm->target, texture_type);
 
       LP_CHECK_MEMBER_OFFSET(struct lp_jit_texture, width,
-                             screen->target, texture_type,
+                             gallivm->target, texture_type,
                              LP_JIT_TEXTURE_WIDTH);
       LP_CHECK_MEMBER_OFFSET(struct lp_jit_texture, height,
-                             screen->target, texture_type,
+                             gallivm->target, texture_type,
                              LP_JIT_TEXTURE_HEIGHT);
       LP_CHECK_MEMBER_OFFSET(struct lp_jit_texture, depth,
-                             screen->target, texture_type,
+                             gallivm->target, texture_type,
                              LP_JIT_TEXTURE_DEPTH);
       LP_CHECK_MEMBER_OFFSET(struct lp_jit_texture, last_level,
-                             screen->target, texture_type,
+                             gallivm->target, texture_type,
                              LP_JIT_TEXTURE_LAST_LEVEL);
       LP_CHECK_MEMBER_OFFSET(struct lp_jit_texture, row_stride,
-                             screen->target, texture_type,
+                             gallivm->target, texture_type,
                              LP_JIT_TEXTURE_ROW_STRIDE);
       LP_CHECK_MEMBER_OFFSET(struct lp_jit_texture, img_stride,
-                             screen->target, texture_type,
+                             gallivm->target, texture_type,
                              LP_JIT_TEXTURE_IMG_STRIDE);
       LP_CHECK_MEMBER_OFFSET(struct lp_jit_texture, data,
-                             screen->target, texture_type,
+                             gallivm->target, texture_type,
                              LP_JIT_TEXTURE_DATA);
       LP_CHECK_MEMBER_OFFSET(struct lp_jit_texture, min_lod,
-                             screen->target, texture_type,
+                             gallivm->target, texture_type,
                              LP_JIT_TEXTURE_MIN_LOD);
       LP_CHECK_MEMBER_OFFSET(struct lp_jit_texture, max_lod,
-                             screen->target, texture_type,
+                             gallivm->target, texture_type,
                              LP_JIT_TEXTURE_MAX_LOD);
       LP_CHECK_MEMBER_OFFSET(struct lp_jit_texture, lod_bias,
-                             screen->target, texture_type,
+                             gallivm->target, texture_type,
                              LP_JIT_TEXTURE_LOD_BIAS);
       LP_CHECK_MEMBER_OFFSET(struct lp_jit_texture, border_color,
-                             screen->target, texture_type,
+                             gallivm->target, texture_type,
                              LP_JIT_TEXTURE_BORDER_COLOR);
 
       LP_CHECK_STRUCT_SIZE(struct lp_jit_texture,
-                           screen->target, texture_type);
+                           gallivm->target, texture_type);
 
-      LLVMAddTypeName(screen->module, "texture", texture_type);
+      LLVMAddTypeName(gallivm->module, "texture", texture_type);
    }
 
    /* struct lp_jit_context */
@@ -116,44 +119,47 @@ lp_jit_init_globals(struct llvmpipe_screen *screen)
       LLVMTypeRef elem_types[LP_JIT_CTX_COUNT];
       LLVMTypeRef context_type;
 
-      elem_types[LP_JIT_CTX_CONSTANTS] = LLVMPointerType(LLVMFloatType(), 0);
-      elem_types[LP_JIT_CTX_ALPHA_REF] = LLVMFloatType();
-      elem_types[LP_JIT_CTX_STENCIL_REF_FRONT] = LLVMInt32Type();
-      elem_types[LP_JIT_CTX_STENCIL_REF_BACK] = LLVMInt32Type();
-      elem_types[LP_JIT_CTX_BLEND_COLOR] = LLVMPointerType(LLVMInt8Type(), 0);
+      elem_types[LP_JIT_CTX_CONSTANTS] = LLVMPointerType(LLVMFloatTypeInContext(lc), 0);
+      elem_types[LP_JIT_CTX_ALPHA_REF] = LLVMFloatTypeInContext(lc);
+      elem_types[LP_JIT_CTX_STENCIL_REF_FRONT] =
+      elem_types[LP_JIT_CTX_STENCIL_REF_BACK] = LLVMInt32TypeInContext(lc);
+      elem_types[LP_JIT_CTX_BLEND_COLOR] = LLVMPointerType(LLVMInt8TypeInContext(lc), 0);
       elem_types[LP_JIT_CTX_TEXTURES] = LLVMArrayType(texture_type,
                                                       PIPE_MAX_SAMPLERS);
 
-      context_type = LLVMStructType(elem_types, Elements(elem_types), 0);
+      context_type = LLVMStructTypeInContext(lc, elem_types,
+                                             Elements(elem_types), 0);
+
+      LLVMInvalidateStructLayout(gallivm->target, context_type);
 
       LP_CHECK_MEMBER_OFFSET(struct lp_jit_context, constants,
-                             screen->target, context_type,
+                             gallivm->target, context_type,
                              LP_JIT_CTX_CONSTANTS);
       LP_CHECK_MEMBER_OFFSET(struct lp_jit_context, alpha_ref_value,
-                             screen->target, context_type,
+                             gallivm->target, context_type,
                              LP_JIT_CTX_ALPHA_REF);
       LP_CHECK_MEMBER_OFFSET(struct lp_jit_context, stencil_ref_front,
-                             screen->target, context_type,
+                             gallivm->target, context_type,
                              LP_JIT_CTX_STENCIL_REF_FRONT);
       LP_CHECK_MEMBER_OFFSET(struct lp_jit_context, stencil_ref_back,
-                             screen->target, context_type,
+                             gallivm->target, context_type,
                              LP_JIT_CTX_STENCIL_REF_BACK);
       LP_CHECK_MEMBER_OFFSET(struct lp_jit_context, blend_color,
-                             screen->target, context_type,
+                             gallivm->target, context_type,
                              LP_JIT_CTX_BLEND_COLOR);
       LP_CHECK_MEMBER_OFFSET(struct lp_jit_context, textures,
-                             screen->target, context_type,
+                             gallivm->target, context_type,
                              LP_JIT_CTX_TEXTURES);
       LP_CHECK_STRUCT_SIZE(struct lp_jit_context,
-                           screen->target, context_type);
+                           gallivm->target, context_type);
 
-      LLVMAddTypeName(screen->module, "context", context_type);
+      LLVMAddTypeName(gallivm->module, "context", context_type);
 
-      screen->context_ptr_type = LLVMPointerType(context_type, 0);
+      lp->jit_context_ptr_type = LLVMPointerType(context_type, 0);
    }
 
    if (gallivm_debug & GALLIVM_DEBUG_IR) {
-      LLVMDumpModule(screen->module);
+      LLVMDumpModule(gallivm->module);
    }
 }
 
@@ -161,8 +167,7 @@ lp_jit_init_globals(struct llvmpipe_screen *screen)
 void
 lp_jit_screen_cleanup(struct llvmpipe_screen *screen)
 {
-   if(screen->pass)
-      LLVMDisposePassManager(screen->pass);
+   /* nothing */
 }
 
 
@@ -170,30 +175,14 @@ void
 lp_jit_screen_init(struct llvmpipe_screen *screen)
 {
    lp_build_init();
+}
 
-   screen->module = lp_build_module;
-   screen->provider = lp_build_provider;
-   screen->engine = lp_build_engine;
-   screen->target = lp_build_target;
-
-   screen->pass = LLVMCreateFunctionPassManager(screen->provider);
-   LLVMAddTargetData(screen->target, screen->pass);
-
-   if ((gallivm_debug & GALLIVM_DEBUG_NO_OPT) == 0) {
-      /* These are the passes currently listed in llvm-c/Transforms/Scalar.h,
-       * but there are more on SVN. */
-      /* TODO: Add more passes */
-      LLVMAddCFGSimplificationPass(screen->pass);
-      LLVMAddPromoteMemoryToRegisterPass(screen->pass);
-      LLVMAddConstantPropagationPass(screen->pass);
-      LLVMAddInstructionCombiningPass(screen->pass);
-      LLVMAddGVNPass(screen->pass);
-   } else {
-      /* We need at least this pass to prevent the backends to fail in
-       * unexpected ways.
-       */
-      LLVMAddPromoteMemoryToRegisterPass(screen->pass);
-   }
 
-   lp_jit_init_globals(screen);
+LLVMTypeRef
+lp_jit_get_context_type(struct llvmpipe_context *lp)
+{
+   if (!lp->jit_context_ptr_type)
+      lp_jit_create_types(lp);
+
+   return lp->jit_context_ptr_type;
 }
diff --git a/src/gallium/drivers/llvmpipe/lp_jit.h b/src/gallium/drivers/llvmpipe/lp_jit.h
index 114f21f2d1..a6763dce17 100644
--- a/src/gallium/drivers/llvmpipe/lp_jit.h
+++ b/src/gallium/drivers/llvmpipe/lp_jit.h
@@ -120,23 +120,23 @@ enum {
 };
 
 
-#define lp_jit_context_constants(_builder, _ptr) \
-   lp_build_struct_get(_builder, _ptr, LP_JIT_CTX_CONSTANTS, "constants")
+#define lp_jit_context_constants(_gallivm, _ptr) \
+   lp_build_struct_get(_gallivm, _ptr, LP_JIT_CTX_CONSTANTS, "constants")
 
-#define lp_jit_context_alpha_ref_value(_builder, _ptr) \
-   lp_build_struct_get(_builder, _ptr, LP_JIT_CTX_ALPHA_REF, "alpha_ref_value")
+#define lp_jit_context_alpha_ref_value(_gallivm, _ptr) \
+   lp_build_struct_get(_gallivm, _ptr, LP_JIT_CTX_ALPHA_REF, "alpha_ref_value")
 
-#define lp_jit_context_stencil_ref_front_value(_builder, _ptr) \
-   lp_build_struct_get(_builder, _ptr, LP_JIT_CTX_STENCIL_REF_FRONT, "stencil_ref_front")
+#define lp_jit_context_stencil_ref_front_value(_gallivm, _ptr) \
+   lp_build_struct_get(_gallivm, _ptr, LP_JIT_CTX_STENCIL_REF_FRONT, "stencil_ref_front")
 
-#define lp_jit_context_stencil_ref_back_value(_builder, _ptr) \
-   lp_build_struct_get(_builder, _ptr, LP_JIT_CTX_STENCIL_REF_BACK, "stencil_ref_back")
+#define lp_jit_context_stencil_ref_back_value(_gallivm, _ptr) \
+   lp_build_struct_get(_gallivm, _ptr, LP_JIT_CTX_STENCIL_REF_BACK, "stencil_ref_back")
 
-#define lp_jit_context_blend_color(_builder, _ptr) \
-   lp_build_struct_get(_builder, _ptr, LP_JIT_CTX_BLEND_COLOR, "blend_color")
+#define lp_jit_context_blend_color(_gallivm, _ptr) \
+   lp_build_struct_get(_gallivm, _ptr, LP_JIT_CTX_BLEND_COLOR, "blend_color")
 
-#define lp_jit_context_textures(_builder, _ptr) \
-   lp_build_struct_get_ptr(_builder, _ptr, LP_JIT_CTX_TEXTURES, "textures")
+#define lp_jit_context_textures(_gallivm, _ptr) \
+   lp_build_struct_get_ptr(_gallivm, _ptr, LP_JIT_CTX_TEXTURES, "textures")
 
 
 
@@ -162,4 +162,8 @@ void
 lp_jit_screen_init(struct llvmpipe_screen *screen);
 
 
+LLVMTypeRef
+lp_jit_get_context_type(struct llvmpipe_context *lp);
+
+
 #endif /* LP_JIT_H */
diff --git a/src/gallium/drivers/llvmpipe/lp_rast.c b/src/gallium/drivers/llvmpipe/lp_rast.c
index decf3bd449..dafadc1ea9 100644
--- a/src/gallium/drivers/llvmpipe/lp_rast.c
+++ b/src/gallium/drivers/llvmpipe/lp_rast.c
@@ -47,6 +47,7 @@
 #ifdef DEBUG
 int jit_line = 0;
 const struct lp_rast_state *jit_state = NULL;
+const struct lp_rasterizer_task *jit_task = NULL;
 #endif
 
 
@@ -119,8 +120,8 @@ lp_rast_tile_begin(struct lp_rasterizer_task *task,
           * and update the tile's layout info.
           */
          (void) llvmpipe_get_texture_tile(lpt,
-                                          zsbuf->face + zsbuf->zslice,
-                                          zsbuf->level,
+                                          zsbuf->u.tex.first_layer,
+                                          zsbuf->u.tex.level,
                                           usage,
                                           task->x,
                                           task->y);
@@ -288,7 +289,6 @@ lp_rast_clear_zstencil(struct lp_rasterizer_task *task,
 
 
 
-
 /**
  * Convert the color tile from tiled to linear layout.
  * This is generally only done when we're flushing the scene just prior to
@@ -306,15 +306,15 @@ lp_rast_store_linear_color( struct lp_rasterizer_task *task )
 
    for (buf = 0; buf < scene->fb.nr_cbufs; buf++) {
       struct pipe_surface *cbuf = scene->fb.cbufs[buf];
-      const unsigned face_slice = cbuf->face + cbuf->zslice;
-      const unsigned level = cbuf->level;
+      const unsigned layer = cbuf->u.tex.first_layer;
+      const unsigned level = cbuf->u.tex.level;
       struct llvmpipe_resource *lpt = llvmpipe_resource(cbuf->texture);
 
       if (!task->color_tiles[buf])
          continue;
 
       llvmpipe_unswizzle_cbuf_tile(lpt,
-                                   face_slice,
+                                   layer,
                                    level,
                                    task->x, task->y,
                                    task->color_tiles[buf]);
@@ -362,7 +362,7 @@ lp_rast_shade_tile(struct lp_rasterizer_task *task,
          depth = lp_rast_get_depth_block_pointer(task, tile_x + x, tile_y + y);
 
          /* run shader on 4x4 block */
-         BEGIN_JIT_CALL(state);
+         BEGIN_JIT_CALL(state, task);
          variant->jit_function[RAST_WHOLE]( &state->jit_context,
                                             tile_x + x, tile_y + y,
                                             inputs->frontfacing,
@@ -443,7 +443,7 @@ lp_rast_shade_quads_mask(struct lp_rasterizer_task *task,
    assert(lp_check_alignment(state->jit_context.blend_color, 16));
 
    /* run shader on 4x4 block */
-   BEGIN_JIT_CALL(state);
+   BEGIN_JIT_CALL(state, task);
    variant->jit_function[RAST_EDGE_TEST](&state->jit_context,
                                          x, y,
                                          inputs->frontfacing,
diff --git a/src/gallium/drivers/llvmpipe/lp_rast_priv.h b/src/gallium/drivers/llvmpipe/lp_rast_priv.h
index b30408f097..cd686bc82c 100644
--- a/src/gallium/drivers/llvmpipe/lp_rast_priv.h
+++ b/src/gallium/drivers/llvmpipe/lp_rast_priv.h
@@ -45,13 +45,16 @@
  */
 #ifdef DEBUG
 
+struct lp_rasterizer_task;
 extern int jit_line;
 extern const struct lp_rast_state *jit_state;
+extern const struct lp_rasterizer_task *jit_task;
 
-#define BEGIN_JIT_CALL(state) \
+#define BEGIN_JIT_CALL(state, task)                  \
    do { \
       jit_line = __LINE__; \
       jit_state = state; \
+      jit_task = task; \
    } while (0)
 
 #define END_JIT_CALL() \
@@ -62,7 +65,7 @@ extern const struct lp_rast_state *jit_state;
 
 #else
 
-#define BEGIN_JIT_CALL(X)
+#define BEGIN_JIT_CALL(X, Y)
 #define END_JIT_CALL()
 
 #endif
@@ -191,8 +194,8 @@ lp_rast_get_color_tile_pointer(struct lp_rasterizer_task *task,
 
       if (usage != LP_TEX_USAGE_WRITE_ALL) {
          llvmpipe_swizzle_cbuf_tile(lpt,
-                                    cbuf->face + cbuf->zslice,
-                                    cbuf->level,
+                                    cbuf->u.tex.first_layer,
+                                    cbuf->u.tex.level,
                                     task->x, task->y,
                                     task->color_tiles[buf]);
       }
@@ -258,7 +261,7 @@ lp_rast_shade_quads_all( struct lp_rasterizer_task *task,
    depth = lp_rast_get_depth_block_pointer(task, x, y);
 
    /* run shader on 4x4 block */
-   BEGIN_JIT_CALL(state);
+   BEGIN_JIT_CALL(state, task);
    variant->jit_function[RAST_WHOLE]( &state->jit_context,
                                       x, y,
                                       inputs->frontfacing,
diff --git a/src/gallium/drivers/llvmpipe/lp_scene.c b/src/gallium/drivers/llvmpipe/lp_scene.c
index a4fdf7cff3..5d0f5f8b7b 100644
--- a/src/gallium/drivers/llvmpipe/lp_scene.c
+++ b/src/gallium/drivers/llvmpipe/lp_scene.c
@@ -74,6 +74,7 @@ lp_scene_create( struct pipe_context *pipe )
 void
 lp_scene_destroy(struct lp_scene *scene)
 {
+   lp_fence_reference(&scene->fence, NULL);
    pipe_mutex_destroy(scene->mutex);
    assert(scene->data.head->next == NULL);
    FREE(scene->data.head);
@@ -136,30 +137,30 @@ lp_scene_begin_rasterization(struct lp_scene *scene)
    int i;
 
    //LP_DBG(DEBUG_RAST, "%s\n", __FUNCTION__);
-   
+
    for (i = 0; i < scene->fb.nr_cbufs; i++) {
       struct pipe_surface *cbuf = scene->fb.cbufs[i];
+      assert(cbuf->u.tex.first_layer == cbuf->u.tex.last_layer);
       scene->cbufs[i].stride = llvmpipe_resource_stride(cbuf->texture,
-                                                        cbuf->level);
+                                                        cbuf->u.tex.level);
 
       scene->cbufs[i].map = llvmpipe_resource_map(cbuf->texture,
-                                                  cbuf->face,
-                                                  cbuf->level,
-                                                  cbuf->zslice,
+                                                  cbuf->u.tex.level,
+                                                  cbuf->u.tex.first_layer,
                                                   LP_TEX_USAGE_READ_WRITE,
                                                   LP_TEX_LAYOUT_LINEAR);
    }
 
    if (fb->zsbuf) {
       struct pipe_surface *zsbuf = scene->fb.zsbuf;
-      scene->zsbuf.stride = llvmpipe_resource_stride(zsbuf->texture, zsbuf->level);
+      assert(zsbuf->u.tex.first_layer == zsbuf->u.tex.last_layer);
+      scene->zsbuf.stride = llvmpipe_resource_stride(zsbuf->texture, zsbuf->u.tex.level);
       scene->zsbuf.blocksize = 
          util_format_get_blocksize(zsbuf->texture->format);
 
       scene->zsbuf.map = llvmpipe_resource_map(zsbuf->texture,
-                                               zsbuf->face,
-                                               zsbuf->level,
-                                               zsbuf->zslice,
+                                               zsbuf->u.tex.level,
+                                               zsbuf->u.tex.first_layer,
                                                LP_TEX_USAGE_READ_WRITE,
                                                LP_TEX_LAYOUT_NONE);
    }
@@ -181,9 +182,8 @@ lp_scene_end_rasterization(struct lp_scene *scene )
       if (scene->cbufs[i].map) {
          struct pipe_surface *cbuf = scene->fb.cbufs[i];
          llvmpipe_resource_unmap(cbuf->texture,
-                                 cbuf->face,
-                                 cbuf->level,
-                                 cbuf->zslice);
+                                 cbuf->u.tex.level,
+                                 cbuf->u.tex.first_layer);
          scene->cbufs[i].map = NULL;
       }
    }
@@ -192,9 +192,8 @@ lp_scene_end_rasterization(struct lp_scene *scene )
    if (scene->zsbuf.map) {
       struct pipe_surface *zsbuf = scene->fb.zsbuf;
       llvmpipe_resource_unmap(zsbuf->texture,
-                             zsbuf->face,
-                             zsbuf->level,
-                             zsbuf->zslice);
+                              zsbuf->u.tex.level,
+                              zsbuf->u.tex.first_layer);
       scene->zsbuf.map = NULL;
    }
 
diff --git a/src/gallium/drivers/llvmpipe/lp_screen.c b/src/gallium/drivers/llvmpipe/lp_screen.c
index ad0ea75b3a..9459a3cd11 100644
--- a/src/gallium/drivers/llvmpipe/lp_screen.c
+++ b/src/gallium/drivers/llvmpipe/lp_screen.c
@@ -287,12 +287,13 @@ llvmpipe_is_format_supported( struct pipe_screen *_screen,
 
 static void
 llvmpipe_flush_frontbuffer(struct pipe_screen *_screen,
-                           struct pipe_surface *surface,
+                           struct pipe_resource *resource,
+                           unsigned level, unsigned layer,
                            void *context_private)
 {
    struct llvmpipe_screen *screen = llvmpipe_screen(_screen);
    struct sw_winsys *winsys = screen->winsys;
-   struct llvmpipe_resource *texture = llvmpipe_resource(surface->texture);
+   struct llvmpipe_resource *texture = llvmpipe_resource(resource);
 
    assert(texture->dt);
    if (texture->dt)
diff --git a/src/gallium/drivers/llvmpipe/lp_screen.h b/src/gallium/drivers/llvmpipe/lp_screen.h
index 731526dfab..7f69a11a6e 100644
--- a/src/gallium/drivers/llvmpipe/lp_screen.h
+++ b/src/gallium/drivers/llvmpipe/lp_screen.h
@@ -34,12 +34,10 @@
 #ifndef LP_SCREEN_H
 #define LP_SCREEN_H
 
-#include "gallivm/lp_bld.h"
-#include <llvm-c/ExecutionEngine.h>
-
-#include "os/os_thread.h"
 #include "pipe/p_screen.h"
 #include "pipe/p_defines.h"
+#include "os/os_thread.h"
+#include "gallivm/lp_bld.h"
 
 
 struct sw_winsys;
@@ -51,14 +49,6 @@ struct llvmpipe_screen
 
    struct sw_winsys *winsys;
 
-   LLVMModuleRef module;
-   LLVMExecutionEngineRef engine;
-   LLVMModuleProviderRef provider;
-   LLVMTargetDataRef target;
-   LLVMPassManagerRef pass;
-
-   LLVMTypeRef context_ptr_type;
-
    unsigned num_threads;
 
    /* Increments whenever textures are modified.  Contexts can track this.
diff --git a/src/gallium/drivers/llvmpipe/lp_setup.c b/src/gallium/drivers/llvmpipe/lp_setup.c
index 6118434d3d..5d83a1e357 100644
--- a/src/gallium/drivers/llvmpipe/lp_setup.c
+++ b/src/gallium/drivers/llvmpipe/lp_setup.c
@@ -114,7 +114,7 @@ first_point( struct lp_setup_context *setup,
    setup->point( setup, v0 );
 }
 
-static void lp_setup_reset( struct lp_setup_context *setup )
+void lp_setup_reset( struct lp_setup_context *setup )
 {
    LP_DBG(DEBUG_SETUP, "%s\n", __FUNCTION__);
 
@@ -263,7 +263,6 @@ execute_clears( struct lp_setup_context *setup )
 
 const char *states[] = {
    "FLUSHED",
-   "EMPTY  ",
    "CLEARED",
    "ACTIVE "
 };
@@ -913,6 +912,12 @@ lp_setup_update_state( struct lp_setup_context *setup,
          llvmpipe_update_derived(lp);
       }
 
+      if (lp->setup->dirty) {
+         llvmpipe_update_setup(lp);
+      }
+
+      assert(setup->setup.variant);
+
       /* Will probably need to move this somewhere else, just need  
        * to know about vertex shader point size attribute.
        */
@@ -928,7 +933,7 @@ lp_setup_update_state( struct lp_setup_context *setup,
 		    setup->setup.variant->key.size) == 0);
    }
 
-   if (update_scene) {
+   if (update_scene && setup->state != SETUP_ACTIVE) {
       if (!set_scene_state( setup, SETUP_ACTIVE, __FUNCTION__ ))
          return FALSE;
    }
@@ -991,6 +996,8 @@ lp_setup_destroy( struct lp_setup_context *setup )
       lp_scene_destroy(scene);
    }
 
+   lp_fence_reference(&setup->last_fence, NULL);
+
    FREE( setup );
 }
 
diff --git a/src/gallium/drivers/llvmpipe/lp_setup.h b/src/gallium/drivers/llvmpipe/lp_setup.h
index ebb18f8134..0d6e161a21 100644
--- a/src/gallium/drivers/llvmpipe/lp_setup.h
+++ b/src/gallium/drivers/llvmpipe/lp_setup.h
@@ -45,6 +45,9 @@ struct lp_jit_context;
 struct llvmpipe_query;
 struct pipe_fence_handle;
 struct lp_setup_variant;
+struct lp_setup_context;
+
+void lp_setup_reset( struct lp_setup_context *setup );
 
 struct lp_setup_context *
 lp_setup_create( struct pipe_context *pipe,
diff --git a/src/gallium/drivers/llvmpipe/lp_setup_vbuf.c b/src/gallium/drivers/llvmpipe/lp_setup_vbuf.c
index 9c1f0fe793..384242f81d 100644
--- a/src/gallium/drivers/llvmpipe/lp_setup_vbuf.c
+++ b/src/gallium/drivers/llvmpipe/lp_setup_vbuf.c
@@ -141,6 +141,8 @@ lp_setup_draw_elements(struct vbuf_render *vbr, const ushort *indices, uint nr)
    const boolean flatshade_first = setup->flatshade_first;
    unsigned i;
 
+   assert(setup->setup.variant);
+
    if (!lp_setup_update_state(setup, TRUE))
       return;
 
diff --git a/src/gallium/drivers/llvmpipe/lp_state_derived.c b/src/gallium/drivers/llvmpipe/lp_state_derived.c
index 0f5f7369e0..8725ea39fe 100644
--- a/src/gallium/drivers/llvmpipe/lp_state_derived.c
+++ b/src/gallium/drivers/llvmpipe/lp_state_derived.c
@@ -53,6 +53,11 @@ compute_vertex_info(struct llvmpipe_context *llvmpipe)
    unsigned vs_index;
    uint i;
 
+   llvmpipe->color_slot[0] = ~0;
+   llvmpipe->color_slot[1] = ~0;
+   llvmpipe->bcolor_slot[0] = ~0;
+   llvmpipe->bcolor_slot[1] = ~0;
+
    /*
     * Match FS inputs against VS outputs, emitting the necessary
     * attributes.  Could cache these structs and look them up with a
@@ -76,12 +81,31 @@ compute_vertex_info(struct llvmpipe_context *llvmpipe)
                                          lpfs->info.base.input_semantic_name[i],
                                          lpfs->info.base.input_semantic_index[i]);
 
+      if (lpfs->info.base.input_semantic_name[i] == TGSI_SEMANTIC_COLOR &&
+          lpfs->info.base.input_semantic_index[i] < 2) {
+         int idx = lpfs->info.base.input_semantic_index[i];
+         llvmpipe->color_slot[idx] = vinfo->num_attribs;
+      }
+
       /*
        * Emit the requested fs attribute for all but position.
        */
       draw_emit_vertex_attr(vinfo, EMIT_4F, INTERP_PERSPECTIVE, vs_index);
    }
 
+   /* Figure out if we need bcolor as well.
+    */
+   for (i = 0; i < 2; i++) {
+      vs_index = draw_find_shader_output(llvmpipe->draw,
+                                         TGSI_SEMANTIC_BCOLOR, i);
+
+      if (vs_index > 0) {
+         llvmpipe->bcolor_slot[i] = vinfo->num_attribs;
+         draw_emit_vertex_attr(vinfo, EMIT_4F, INTERP_PERSPECTIVE, vs_index);
+      }
+   }
+
+
    /* Figure out if we need pointsize as well.
     */
    vs_index = draw_find_shader_output(llvmpipe->draw,
diff --git a/src/gallium/drivers/llvmpipe/lp_state_fs.c b/src/gallium/drivers/llvmpipe/lp_state_fs.c
index 48971510f2..2c4943a69f 100644
--- a/src/gallium/drivers/llvmpipe/lp_state_fs.c
+++ b/src/gallium/drivers/llvmpipe/lp_state_fs.c
@@ -102,10 +102,10 @@
 #include <llvm-c/BitWriter.h>
 
 
+/** Fragment shader number (for debugging) */
 static unsigned fs_no = 0;
 
 
-
 /**
  * Expand the relevent bits of mask_input to a 4-dword mask for the 
  * four pixels in a 2x2 quad.  This will set the four elements of the
@@ -115,13 +115,14 @@ static unsigned fs_no = 0;
  * \param mask_input  bitwise mask for the whole 4x4 stamp
  */
 static LLVMValueRef
-generate_quad_mask(LLVMBuilderRef builder,
+generate_quad_mask(struct gallivm_state *gallivm,
                    struct lp_type fs_type,
                    unsigned quad,
                    LLVMValueRef mask_input) /* int32 */
 {
+   LLVMBuilderRef builder = gallivm->builder;
    struct lp_type mask_type;
-   LLVMTypeRef i32t = LLVMInt32Type();
+   LLVMTypeRef i32t = LLVMInt32TypeInContext(gallivm->context);
    LLVMValueRef bits[4];
    LLVMValueRef mask;
    int shift;
@@ -136,7 +137,6 @@ generate_quad_mask(LLVMBuilderRef builder,
    /*
     * mask_input >>= (quad * 4)
     */
-   
    switch (quad) {
    case 0:
       shift = 0;
@@ -163,8 +163,9 @@ generate_quad_mask(LLVMBuilderRef builder,
    /*
     * mask = { mask_input & (1 << i), for i in [0,3] }
     */
-
-   mask = lp_build_broadcast(builder, lp_build_vec_type(mask_type), mask_input);
+   mask = lp_build_broadcast(gallivm,
+                             lp_build_vec_type(gallivm, mask_type),
+                             mask_input);
 
    bits[0] = LLVMConstInt(i32t, 1 << 0, 0);
    bits[1] = LLVMConstInt(i32t, 1 << 1, 0);
@@ -176,11 +177,10 @@ generate_quad_mask(LLVMBuilderRef builder,
    /*
     * mask = mask != 0 ? ~0 : 0
     */
-
-   mask = lp_build_compare(builder,
+   mask = lp_build_compare(gallivm,
                            mask_type, PIPE_FUNC_NOTEQUAL,
                            mask,
-                           lp_build_const_int_vec(mask_type, 0));
+                           lp_build_const_int_vec(gallivm, mask_type, 0));
 
    return mask;
 }
@@ -213,7 +213,8 @@ find_output_by_semantic( const struct tgsi_shader_info *info,
  * \param partial_mask  if 1, do mask_input testing
  */
 static void
-generate_fs(struct lp_fragment_shader *shader,
+generate_fs(struct gallivm_state *gallivm,
+            struct lp_fragment_shader *shader,
             const struct lp_fragment_shader_variant_key *key,
             LLVMBuilderRef builder,
             struct lp_type type,
@@ -278,42 +279,42 @@ generate_fs(struct lp_fragment_shader *shader,
 
    assert(i < 4);
 
-   stencil_refs[0] = lp_jit_context_stencil_ref_front_value(builder, context_ptr);
-   stencil_refs[1] = lp_jit_context_stencil_ref_back_value(builder, context_ptr);
+   stencil_refs[0] = lp_jit_context_stencil_ref_front_value(gallivm, context_ptr);
+   stencil_refs[1] = lp_jit_context_stencil_ref_back_value(gallivm, context_ptr);
 
-   vec_type = lp_build_vec_type(type);
+   vec_type = lp_build_vec_type(gallivm, type);
 
-   consts_ptr = lp_jit_context_constants(builder, context_ptr);
+   consts_ptr = lp_jit_context_constants(gallivm, context_ptr);
 
    memset(outputs, 0, sizeof outputs);
 
    /* Declare the color and z variables */
    for(cbuf = 0; cbuf < key->nr_cbufs; cbuf++) {
       for(chan = 0; chan < NUM_CHANNELS; ++chan) {
-	 color[cbuf][chan] = lp_build_alloca(builder, vec_type, "color");
+	 color[cbuf][chan] = lp_build_alloca(gallivm, vec_type, "color");
       }
    }
 
    /* do triangle edge testing */
    if (partial_mask) {
-      *pmask = generate_quad_mask(builder, type,
+      *pmask = generate_quad_mask(gallivm, type,
                                   i, mask_input);
    }
    else {
-      *pmask = lp_build_const_int_vec(type, ~0);
+      *pmask = lp_build_const_int_vec(gallivm, type, ~0);
    }
 
    /* 'mask' will control execution based on quad's pixel alive/killed state */
-   lp_build_mask_begin(&mask, builder, type, *pmask);
+   lp_build_mask_begin(&mask, gallivm, type, *pmask);
 
    if (!(depth_mode & EARLY_DEPTH_TEST) && !simple_shader)
       lp_build_mask_check(&mask);
 
-   lp_build_interp_soa_update_pos(interp, i);
+   lp_build_interp_soa_update_pos(interp, gallivm, i);
    z = interp->pos[2];
 
    if (depth_mode & EARLY_DEPTH_TEST) {
-      lp_build_depth_stencil_test(builder,
+      lp_build_depth_stencil_test(gallivm,
                                   &key->depth,
                                   key->stencil,
                                   type,
@@ -330,14 +331,13 @@ generate_fs(struct lp_fragment_shader *shader,
       }
    }
 
-   lp_build_interp_soa_update_inputs(interp, i);
+   lp_build_interp_soa_update_inputs(interp, gallivm, i);
    
    /* Build the actual shader */
-   lp_build_tgsi_soa(builder, tokens, type, &mask,
+   lp_build_tgsi_soa(gallivm, tokens, type, &mask,
                      consts_ptr, interp->pos, interp->inputs,
                      outputs, sampler, &shader->info.base);
 
-
    /* Alpha test */
    if (key->alpha.enabled) {
       int color0 = find_output_by_semantic(&shader->info.base,
@@ -348,10 +348,10 @@ generate_fs(struct lp_fragment_shader *shader,
          LLVMValueRef alpha = LLVMBuildLoad(builder, outputs[color0][3], "alpha");
          LLVMValueRef alpha_ref_value;
 
-         alpha_ref_value = lp_jit_context_alpha_ref_value(builder, context_ptr);
-         alpha_ref_value = lp_build_broadcast(builder, vec_type, alpha_ref_value);
+         alpha_ref_value = lp_jit_context_alpha_ref_value(gallivm, context_ptr);
+         alpha_ref_value = lp_build_broadcast(gallivm, vec_type, alpha_ref_value);
 
-         lp_build_alpha_test(builder, key->alpha.func, type,
+         lp_build_alpha_test(gallivm, key->alpha.func, type,
                              &mask, alpha, alpha_ref_value,
                              (depth_mode & LATE_DEPTH_TEST) != 0);
       }
@@ -367,7 +367,7 @@ generate_fs(struct lp_fragment_shader *shader,
          z = LLVMBuildLoad(builder, outputs[pos0][2], "output.z");
       }
 
-      lp_build_depth_stencil_test(builder,
+      lp_build_depth_stencil_test(gallivm,
                                   &key->depth,
                                   key->stencil,
                                   type,
@@ -390,7 +390,7 @@ generate_fs(struct lp_fragment_shader *shader,
        * depth value, update from zs_value with the new mask value and
        * write that out.
        */
-      lp_build_deferred_depth_write(builder,
+      lp_build_deferred_depth_write(gallivm,
                                     type,
                                     zs_format_desc,
                                     &mask,
@@ -420,7 +420,7 @@ generate_fs(struct lp_fragment_shader *shader,
    }
 
    if (counter)
-      lp_build_occlusion_count(builder, type,
+      lp_build_occlusion_count(gallivm, type,
                                lp_build_mask_value(&mask), counter);
 
    *pmask = lp_build_mask_end(&mask);
@@ -437,7 +437,8 @@ generate_fs(struct lp_fragment_shader *shader,
  * \param dst_ptr  the destination color buffer pointer
  */
 static void
-generate_blend(const struct pipe_blend_state *blend,
+generate_blend(struct gallivm_state *gallivm,
+               const struct pipe_blend_state *blend,
                unsigned rt,
                LLVMBuilderRef builder,
                struct lp_type type,
@@ -456,21 +457,21 @@ generate_blend(const struct pipe_blend_state *blend,
    LLVMValueRef res[4];
    unsigned chan;
 
-   lp_build_context_init(&bld, builder, type);
+   lp_build_context_init(&bld, gallivm, type);
 
-   lp_build_mask_begin(&mask_ctx, builder, type, mask);
+   lp_build_mask_begin(&mask_ctx, gallivm, type, mask);
    if (do_branch)
       lp_build_mask_check(&mask_ctx);
 
-   vec_type = lp_build_vec_type(type);
+   vec_type = lp_build_vec_type(gallivm, type);
 
-   const_ptr = lp_jit_context_blend_color(builder, context_ptr);
+   const_ptr = lp_jit_context_blend_color(gallivm, context_ptr);
    const_ptr = LLVMBuildBitCast(builder, const_ptr,
                                 LLVMPointerType(vec_type, 0), "");
 
    /* load constant blend color and colors from the dest color buffer */
    for(chan = 0; chan < 4; ++chan) {
-      LLVMValueRef index = LLVMConstInt(LLVMInt32Type(), chan, 0);
+      LLVMValueRef index = lp_build_const_int32(gallivm, chan);
       con[chan] = LLVMBuildLoad(builder, LLVMBuildGEP(builder, const_ptr, &index, 1, ""), "");
 
       dst[chan] = LLVMBuildLoad(builder, LLVMBuildGEP(builder, dst_ptr, &index, 1, ""), "");
@@ -480,12 +481,12 @@ generate_blend(const struct pipe_blend_state *blend,
    }
 
    /* do blend */
-   lp_build_blend_soa(builder, blend, type, rt, src, dst, con, res);
+   lp_build_blend_soa(gallivm, blend, type, rt, src, dst, con, res);
 
    /* store results to color buffer */
    for(chan = 0; chan < 4; ++chan) {
       if(blend->rt[rt].colormask & (1 << chan)) {
-         LLVMValueRef index = LLVMConstInt(LLVMInt32Type(), chan, 0);
+         LLVMValueRef index = lp_build_const_int32(gallivm, chan);
          lp_build_name(res[chan], "res.%c", "rgba"[chan]);
          res[chan] = lp_build_select(&bld, mask, res[chan], dst[chan]);
          LLVMBuildStore(builder, res[chan], LLVMBuildGEP(builder, dst_ptr, &index, 1, ""));
@@ -503,11 +504,12 @@ generate_blend(const struct pipe_blend_state *blend,
  * 2x2 pixels.
  */
 static void
-generate_fragment(struct llvmpipe_screen *screen,
+generate_fragment(struct llvmpipe_context *lp,
                   struct lp_fragment_shader *shader,
                   struct lp_fragment_shader_variant *variant,
                   unsigned partial_mask)
 {
+   struct gallivm_state *gallivm = lp->gallivm;
    const struct lp_fragment_shader_variant_key *key = &variant->key;
    struct lp_shader_input inputs[PIPE_MAX_SHADER_INPUTS];
    char func_name[256];
@@ -518,6 +520,8 @@ generate_fragment(struct llvmpipe_screen *screen,
    LLVMTypeRef blend_vec_type;
    LLVMTypeRef arg_types[11];
    LLVMTypeRef func_type;
+   LLVMTypeRef int32_type = LLVMInt32TypeInContext(gallivm->context);
+   LLVMTypeRef int8_type = LLVMInt8TypeInContext(gallivm->context);
    LLVMValueRef context_ptr;
    LLVMValueRef x;
    LLVMValueRef y;
@@ -579,29 +583,30 @@ generate_fragment(struct llvmpipe_screen *screen,
     * lp_jit.h's lp_jit_frag_func function pointer type, and vice-versa.
     */
 
-   fs_elem_type = lp_build_elem_type(fs_type);
-   fs_int_vec_type = lp_build_int_vec_type(fs_type);
+   fs_elem_type = lp_build_elem_type(gallivm, fs_type);
+   fs_int_vec_type = lp_build_int_vec_type(gallivm, fs_type);
 
-   blend_vec_type = lp_build_vec_type(blend_type);
+   blend_vec_type = lp_build_vec_type(gallivm, blend_type);
 
    util_snprintf(func_name, sizeof(func_name), "fs%u_variant%u_%s", 
 		 shader->no, variant->no, partial_mask ? "partial" : "whole");
 
-   arg_types[0] = screen->context_ptr_type;            /* context */
-   arg_types[1] = LLVMInt32Type();                     /* x */
-   arg_types[2] = LLVMInt32Type();                     /* y */
-   arg_types[3] = LLVMInt32Type();                     /* facing */
+   arg_types[0] = lp_jit_get_context_type(lp);         /* context */
+   arg_types[1] = int32_type;                          /* x */
+   arg_types[2] = int32_type;                          /* y */
+   arg_types[3] = int32_type;                          /* facing */
    arg_types[4] = LLVMPointerType(fs_elem_type, 0);    /* a0 */
    arg_types[5] = LLVMPointerType(fs_elem_type, 0);    /* dadx */
    arg_types[6] = LLVMPointerType(fs_elem_type, 0);    /* dady */
    arg_types[7] = LLVMPointerType(LLVMPointerType(blend_vec_type, 0), 0);  /* color */
-   arg_types[8] = LLVMPointerType(LLVMInt8Type(), 0);  /* depth */
-   arg_types[9] = LLVMInt32Type();                     /* mask_input */
-   arg_types[10] = LLVMPointerType(LLVMInt32Type(), 0);/* counter */
+   arg_types[8] = LLVMPointerType(int8_type, 0);       /* depth */
+   arg_types[9] = int32_type;                          /* mask_input */
+   arg_types[10] = LLVMPointerType(int32_type, 0);     /* counter */
 
-   func_type = LLVMFunctionType(LLVMVoidType(), arg_types, Elements(arg_types), 0);
+   func_type = LLVMFunctionType(LLVMVoidTypeInContext(gallivm->context),
+                                arg_types, Elements(arg_types), 0);
 
-   function = LLVMAddFunction(screen->module, func_name, func_type);
+   function = LLVMAddFunction(gallivm->module, func_name, func_type);
    LLVMSetFunctionCallConv(function, LLVMCCallConv);
 
    variant->function[partial_mask] = function;
@@ -643,8 +648,9 @@ generate_fragment(struct llvmpipe_screen *screen,
     * Function body
     */
 
-   block = LLVMAppendBasicBlock(function, "entry");
-   builder = LLVMCreateBuilder();
+   block = LLVMAppendBasicBlockInContext(gallivm->context, function, "entry");
+   builder = gallivm->builder;
+   assert(builder);
    LLVMPositionBuilderAtEnd(builder, block);
 
    /*
@@ -653,6 +659,7 @@ generate_fragment(struct llvmpipe_screen *screen,
     * already included in the shader key.
     */
    lp_build_interp_soa_init(&interp, 
+                            gallivm,
                             shader->info.base.num_inputs,
                             inputs,
                             builder, fs_type,
@@ -666,7 +673,7 @@ generate_fragment(struct llvmpipe_screen *screen,
    zs_format_desc = util_format_description(key->zsbuf_format);
 
    for(i = 0; i < num_fs; ++i) {
-      LLVMValueRef depth_offset = LLVMConstInt(LLVMInt32Type(),
+      LLVMValueRef depth_offset = LLVMConstInt(int32_type,
                                                i*fs_type.length*zs_format_desc->block.bits/8,
                                                0);
       LLVMValueRef out_color[PIPE_MAX_COLOR_BUFS][NUM_CHANNELS];
@@ -674,7 +681,8 @@ generate_fragment(struct llvmpipe_screen *screen,
 
       depth_ptr_i = LLVMBuildGEP(builder, depth_ptr, &depth_offset, 1, "");
 
-      generate_fs(shader, key,
+      generate_fs(gallivm,
+                  shader, key,
                   builder,
                   fs_type,
                   context_ptr,
@@ -700,7 +708,7 @@ generate_fragment(struct llvmpipe_screen *screen,
     */
    for(cbuf = 0; cbuf < key->nr_cbufs; cbuf++) {
       LLVMValueRef color_ptr;
-      LLVMValueRef index = LLVMConstInt(LLVMInt32Type(), cbuf, 0);
+      LLVMValueRef index = lp_build_const_int32(gallivm, cbuf);
       LLVMValueRef blend_in_color[NUM_CHANNELS];
       unsigned rt;
 
@@ -715,7 +723,7 @@ generate_fragment(struct llvmpipe_screen *screen,
                LLVMBuildLoad(builder, fs_out_color[cbuf][chan][i], "fs_color_vals");
          }
 
-	 lp_build_conv(builder, fs_type, blend_type,
+	 lp_build_conv(gallivm, fs_type, blend_type,
                        fs_color_vals,
                        num_fs,
 		       &blend_in_color[chan], 1);
@@ -724,11 +732,11 @@ generate_fragment(struct llvmpipe_screen *screen,
       }
 
       if (partial_mask || !variant->opaque) {
-         lp_build_conv_mask(builder, fs_type, blend_type,
+         lp_build_conv_mask(lp->gallivm, fs_type, blend_type,
                             fs_mask, num_fs,
                             &blend_mask, 1);
       } else {
-         blend_mask = lp_build_const_int_vec(blend_type, ~0);
+         blend_mask = lp_build_const_int_vec(lp->gallivm, blend_type, ~0);
       }
 
       color_ptr = LLVMBuildLoad(builder, 
@@ -749,7 +757,8 @@ generate_fragment(struct llvmpipe_screen *screen,
                               !key->alpha.enabled &&
                               !shader->info.base.uses_kill);
 
-         generate_blend(&key->blend,
+         generate_blend(lp->gallivm,
+                        &key->blend,
                         rt,
                         builder,
                         blend_type,
@@ -763,9 +772,6 @@ generate_fragment(struct llvmpipe_screen *screen,
 
    LLVMBuildRetVoid(builder);
 
-   LLVMDisposeBuilder(builder);
-
-
    /* Verify the LLVM IR.  If invalid, dump and abort */
 #ifdef DEBUG
    if(LLVMVerifyFunction(function, LLVMPrintMessageAction)) {
@@ -776,7 +782,7 @@ generate_fragment(struct llvmpipe_screen *screen,
 #endif
 
    /* Apply optimizations to LLVM IR */
-   LLVMRunFunctionPassManager(screen->pass, function);
+   LLVMRunFunctionPassManager(gallivm->passmgr, function);
 
    if ((gallivm_debug & GALLIVM_DEBUG_IR) || (LP_DEBUG & DEBUG_FS)) {
       /* Print the LLVM IR to stderr */
@@ -786,14 +792,14 @@ generate_fragment(struct llvmpipe_screen *screen,
 
    /* Dump byte code to a file */
    if (0) {
-      LLVMWriteBitcodeToFile(lp_build_module, "llvmpipe.bc");
+      LLVMWriteBitcodeToFile(gallivm->module, "llvmpipe.bc");
    }
 
    /*
     * Translate the LLVM IR into machine code.
     */
    {
-      void *f = LLVMGetPointerToGlobal(screen->engine, function);
+      void *f = LLVMGetPointerToGlobal(gallivm->engine, function);
 
       variant->jit_function[partial_mask] = (lp_jit_frag_func)pointer_to_func(f);
 
@@ -897,8 +903,13 @@ lp_debug_fs_variant(const struct lp_fragment_shader_variant *variant)
    debug_printf("\n");
 }
 
+
+/**
+ * Generate a new fragment shader variant from the shader code and
+ * other state indicated by the key.
+ */
 static struct lp_fragment_shader_variant *
-generate_variant(struct llvmpipe_screen *screen,
+generate_variant(struct llvmpipe_context *lp,
                  struct lp_fragment_shader *shader,
                  const struct lp_fragment_shader_variant_key *key)
 {
@@ -944,11 +955,11 @@ generate_variant(struct llvmpipe_screen *screen,
       lp_debug_fs_variant(variant);
    }
 
-   generate_fragment(screen, shader, variant, RAST_EDGE_TEST);
+   generate_fragment(lp, shader, variant, RAST_EDGE_TEST);
 
    if (variant->opaque) {
       /* Specialized shader, which doesn't need to read the color buffer. */
-      generate_fragment(screen, shader, variant, RAST_WHOLE);
+      generate_fragment(lp, shader, variant, RAST_WHOLE);
    } else {
       variant->jit_function[RAST_WHOLE] = variant->jit_function[RAST_EDGE_TEST];
    }
@@ -1033,7 +1044,8 @@ llvmpipe_create_fs_state(struct pipe_context *pipe,
 
    if (LP_DEBUG & DEBUG_TGSI) {
       unsigned attrib;
-      debug_printf("llvmpipe: Create fragment shader #%u %p:\n", shader->no, (void *) shader);
+      debug_printf("llvmpipe: Create fragment shader #%u %p:\n",
+                   shader->no, (void *) shader);
       tgsi_dump(templ->tokens, 0);
       debug_printf("usage masks:\n");
       for (attrib = 0; attrib < shader->info.base.num_inputs; ++attrib) {
@@ -1070,33 +1082,49 @@ llvmpipe_bind_fs_state(struct pipe_context *pipe, void *fs)
    llvmpipe->dirty |= LP_NEW_FS;
 }
 
-static void
-remove_shader_variant(struct llvmpipe_context *lp,
-                      struct lp_fragment_shader_variant *variant)
+
+/**
+ * Remove shader variant from two lists: the shader's variant list
+ * and the context's variant list.
+ */
+void
+llvmpipe_remove_shader_variant(struct llvmpipe_context *lp,
+                               struct lp_fragment_shader_variant *variant)
 {
-   struct llvmpipe_screen *screen = llvmpipe_screen(lp->pipe.screen);
    unsigned i;
 
    if (gallivm_debug & GALLIVM_DEBUG_IR) {
-      debug_printf("llvmpipe: del fs #%u var #%u v created #%u v cached #%u v total cached #%u\n",
-                    variant->shader->no, variant->no, variant->shader->variants_created,
-                    variant->shader->variants_cached, lp->nr_fs_variants);
+      debug_printf("llvmpipe: del fs #%u var #%u v created #%u v cached"
+                   " #%u v total cached #%u\n",
+                   variant->shader->no,
+                   variant->no,
+                   variant->shader->variants_created,
+                   variant->shader->variants_cached,
+                   lp->nr_fs_variants);
    }
+
+   /* free all the variant's JIT'd functions */
    for (i = 0; i < Elements(variant->function); i++) {
       if (variant->function[i]) {
          if (variant->jit_function[i])
-            LLVMFreeMachineCodeForFunction(screen->engine,
+            LLVMFreeMachineCodeForFunction(lp->gallivm->engine,
                                            variant->function[i]);
          LLVMDeleteFunction(variant->function[i]);
       }
    }
+
+   /* remove from shader's list */
    remove_from_list(&variant->list_item_local);
    variant->shader->variants_cached--;
+
+   /* remove from context's list */
    remove_from_list(&variant->list_item_global);
    lp->nr_fs_variants--;
+
    FREE(variant);
 }
 
+
 static void
 llvmpipe_delete_fs_state(struct pipe_context *pipe, void *fs)
 {
@@ -1105,23 +1133,23 @@ llvmpipe_delete_fs_state(struct pipe_context *pipe, void *fs)
    struct lp_fs_variant_list_item *li;
 
    assert(fs != llvmpipe->fs);
-   (void) llvmpipe;
 
    /*
     * XXX: we need to flush the context until we have some sort of reference
     * counting in fragment shaders as they may still be binned
     * Flushing alone might not sufficient we need to wait on it too.
     */
-
    llvmpipe_finish(pipe, __FUNCTION__);
 
+   /* Delete all the variants */
    li = first_elem(&shader->variants);
    while(!at_end(&shader->variants, li)) {
       struct lp_fs_variant_list_item *next = next_elem(li);
-      remove_shader_variant(llvmpipe, li->base);
+      llvmpipe_remove_shader_variant(llvmpipe, li->base);
       li = next;
    }
 
+   /* Delete draw module's data */
    draw_delete_fragment_shader(llvmpipe->draw, shader->draw_data);
 
    assert(shader->variants_cached == 0);
@@ -1277,14 +1305,15 @@ make_variant_key(struct llvmpipe_context *lp,
    }
 }
 
+
+
 /**
- * Update fragment state.  This is called just prior to drawing
+ * Update fragment shader state.  This is called just prior to drawing
  * something when some fragment-related state has changed.
  */
 void 
 llvmpipe_update_fs(struct llvmpipe_context *lp)
 {
-   struct llvmpipe_screen *screen = llvmpipe_screen(lp->pipe.screen);
    struct lp_fragment_shader *shader = lp->fs;
    struct lp_fragment_shader_variant_key key;
    struct lp_fragment_shader_variant *variant = NULL;
@@ -1292,6 +1321,7 @@ llvmpipe_update_fs(struct llvmpipe_context *lp)
 
    make_variant_key(lp, shader, &key);
 
+   /* Search the variants for one which matches the key */
    li = first_elem(&shader->variants);
    while(!at_end(&shader->variants, li)) {
       if(memcmp(&li->base->key, &key, shader->variant_key_size) == 0) {
@@ -1302,36 +1332,49 @@ llvmpipe_update_fs(struct llvmpipe_context *lp)
    }
 
    if (variant) {
+      /* Move this variant to the head of the list to implement LRU
+       * deletion of shader's when we have too many.
+       */
       move_to_head(&lp->fs_variants_list, &variant->list_item_global);
    }
    else {
-      int64_t t0, t1;
-      int64_t dt;
+      /* variant not found, create it now */
+      int64_t t0, t1, dt;
       unsigned i;
+
+      /* First, check if we've exceeded the max number of shader variants.
+       * If so, free 25% of them (the least recently used ones).
+       */
       if (lp->nr_fs_variants >= LP_MAX_SHADER_VARIANTS) {
          struct pipe_context *pipe = &lp->pipe;
 
          /*
-          * XXX: we need to flush the context until we have some sort of reference
-          * counting in fragment shaders as they may still be binned
+          * XXX: we need to flush the context until we have some sort of
+          * reference counting in fragment shaders as they may still be binned
           * Flushing alone might not be sufficient we need to wait on it too.
           */
          llvmpipe_finish(pipe, __FUNCTION__);
 
          for (i = 0; i < LP_MAX_SHADER_VARIANTS / 4; i++) {
-            struct lp_fs_variant_list_item *item = last_elem(&lp->fs_variants_list);
-            remove_shader_variant(lp, item->base);
+            struct lp_fs_variant_list_item *item;
+            item = last_elem(&lp->fs_variants_list);
+            llvmpipe_remove_shader_variant(lp, item->base);
          }
       }
-      t0 = os_time_get();
-
-      variant = generate_variant(screen, shader, &key);
 
+      /*
+       * Generate the new variant.
+       */
+      t0 = os_time_get();
+      variant = generate_variant(lp, shader, &key);
       t1 = os_time_get();
       dt = t1 - t0;
       LP_COUNT_ADD(llvm_compile_time, dt);
       LP_COUNT_ADD(nr_llvm_compiles, 2);  /* emit vs. omit in/out test */
 
+      llvmpipe_variant_count++;
+
+      /* Put the new variant into the list */
       if (variant) {
          insert_at_head(&shader->variants, &variant->list_item_local);
          insert_at_head(&lp->fs_variants_list, &variant->list_item_global);
@@ -1340,6 +1383,7 @@ llvmpipe_update_fs(struct llvmpipe_context *lp)
       }
    }
 
+   /* Bind this variant */
    lp_setup_set_fs_variant(lp->setup, variant);
 }
 
diff --git a/src/gallium/drivers/llvmpipe/lp_state_fs.h b/src/gallium/drivers/llvmpipe/lp_state_fs.h
index 7d58c4936c..98410c6935 100644
--- a/src/gallium/drivers/llvmpipe/lp_state_fs.h
+++ b/src/gallium/drivers/llvmpipe/lp_state_fs.h
@@ -69,12 +69,15 @@ struct lp_fragment_shader_variant_key
    struct lp_sampler_static_state sampler[PIPE_MAX_SAMPLERS];
 };
 
+
+/** doubly-linked list item */
 struct lp_fs_variant_list_item
 {
    struct lp_fragment_shader_variant *base;
    struct lp_fs_variant_list_item *next, *prev;
 };
 
+
 struct lp_fragment_shader_variant
 {
    struct lp_fragment_shader_variant_key key;
@@ -118,5 +121,9 @@ struct lp_fragment_shader
 void
 lp_debug_fs_variant(const struct lp_fragment_shader_variant *variant);
 
+void
+llvmpipe_remove_shader_variant(struct llvmpipe_context *lp,
+                               struct lp_fragment_shader_variant *variant);
+
 
 #endif /* LP_STATE_FS_H_ */
diff --git a/src/gallium/drivers/llvmpipe/lp_state_rasterizer.c b/src/gallium/drivers/llvmpipe/lp_state_rasterizer.c
index dbd73812e4..574f9e940e 100644
--- a/src/gallium/drivers/llvmpipe/lp_state_rasterizer.c
+++ b/src/gallium/drivers/llvmpipe/lp_state_rasterizer.c
@@ -32,16 +32,64 @@
 #include "lp_setup.h"
 #include "draw/draw_context.h"
 
+struct lp_rast_state {
+   struct pipe_rasterizer_state lp_state;
+   struct pipe_rasterizer_state draw_state;
+};
+
+/* State which might be handled in either the draw module or locally.
+ * This function is used to turn that state off in one of the two
+ * places.
+ */
+static void
+clear_flags(struct pipe_rasterizer_state *rast)
+{
+   rast->light_twoside = 0;
+   rast->offset_tri = 0;
+}
+
 
 
 static void *
 llvmpipe_create_rasterizer_state(struct pipe_context *pipe,
                                  const struct pipe_rasterizer_state *rast)
 {
-   /* We do nothing special with rasterizer state.
-    * The CSO handle is just a pointer to a pipe_rasterizer_state object.
+   boolean need_pipeline;
+
+   /* Partition rasterizer state into what we want the draw module to
+    * handle, and what we'll look after ourselves.
+    */
+   struct lp_rast_state *state = MALLOC_STRUCT(lp_rast_state);
+   if (state == NULL)
+      return NULL;
+
+   memcpy(&state->draw_state, rast, sizeof *rast);
+   memcpy(&state->lp_state, rast, sizeof *rast);
+
+   /* We rely on draw module to do unfilled polyons, AA lines and
+    * points and stipple.
+    * 
+    * Over time, reduce this list of conditions, and expand the list
+    * of flags which get cleared in clear_flags().
     */
-   return mem_dup(rast, sizeof(*rast));
+   need_pipeline = (rast->fill_front != PIPE_POLYGON_MODE_FILL ||
+		    rast->fill_back != PIPE_POLYGON_MODE_FILL ||
+		    rast->point_smooth ||
+		    rast->line_smooth ||
+		    rast->line_stipple_enable ||
+		    rast->poly_stipple_enable);
+
+   /* If not using the pipeline, clear out the flags which we can
+    * handle ourselves.  If we *are* using the pipeline, do everything
+    * on the pipeline and clear those flags on our internal copy of
+    * the state.
+    */
+   if (need_pipeline)
+      clear_flags(&state->lp_state);
+   else
+      clear_flags(&state->draw_state);
+
+   return state;
 }
 
 
@@ -50,36 +98,33 @@ static void
 llvmpipe_bind_rasterizer_state(struct pipe_context *pipe, void *handle)
 {
    struct llvmpipe_context *llvmpipe = llvmpipe_context(pipe);
-   const struct pipe_rasterizer_state *rasterizer =
-      (const struct pipe_rasterizer_state *) handle;
-
-   if (llvmpipe->rasterizer == rasterizer)
-      return;
+   const struct lp_rast_state *state =
+      (const struct lp_rast_state *) handle;
 
-   /* pass-through to draw module */
-   draw_set_rasterizer_state(llvmpipe->draw, rasterizer, handle);
+   if (state) {
+      llvmpipe->rasterizer = &state->lp_state;
+      draw_set_rasterizer_state(llvmpipe->draw, &state->draw_state, handle);
 
-   llvmpipe->rasterizer = rasterizer;
-
-   /* Note: we can immediately set the triangle state here and
-    * not worry about binning because we handle culling during
-    * triangle setup, not when rasterizing the bins.
-    */
-   if (llvmpipe->rasterizer) {
+      /* XXX: just pass lp_state directly to setup.
+       */
       lp_setup_set_triangle_state( llvmpipe->setup,
-                   llvmpipe->rasterizer->cull_face,
-                   llvmpipe->rasterizer->front_ccw,
-                   llvmpipe->rasterizer->scissor,
-                   llvmpipe->rasterizer->gl_rasterization_rules);
+				   state->lp_state.cull_face,
+				   state->lp_state.front_ccw,
+				   state->lp_state.scissor,
+				   state->lp_state.gl_rasterization_rules);
       lp_setup_set_flatshade_first( llvmpipe->setup,
-                   llvmpipe->rasterizer->flatshade_first);
+				    state->lp_state.flatshade_first);
       lp_setup_set_line_state( llvmpipe->setup,
-                   llvmpipe->rasterizer->line_width);
+			       state->lp_state.line_width);
       lp_setup_set_point_state( llvmpipe->setup,
-                   llvmpipe->rasterizer->point_size,
-                   llvmpipe->rasterizer->point_size_per_vertex,
-                   llvmpipe->rasterizer->sprite_coord_enable,
-                   llvmpipe->rasterizer->sprite_coord_mode);
+				state->lp_state.point_size,
+				state->lp_state.point_size_per_vertex,
+				state->lp_state.sprite_coord_enable,
+				state->lp_state.sprite_coord_mode);
+   }
+   else {
+      llvmpipe->rasterizer = NULL;
+      draw_set_rasterizer_state(llvmpipe->draw, NULL, handle);      
    }
 
    llvmpipe->dirty |= LP_NEW_RASTERIZER;
diff --git a/src/gallium/drivers/llvmpipe/lp_state_setup.c b/src/gallium/drivers/llvmpipe/lp_state_setup.c
index 2c8b8b9a92..ad751b9ef4 100644
--- a/src/gallium/drivers/llvmpipe/lp_state_setup.c
+++ b/src/gallium/drivers/llvmpipe/lp_state_setup.c
@@ -30,9 +30,13 @@
 #include "util/u_memory.h"
 #include "util/u_simple_list.h"
 #include "os/os_time.h"
+#include "gallivm/lp_bld_arit.h"
+#include "gallivm/lp_bld_const.h"
 #include "gallivm/lp_bld_debug.h"
 #include "gallivm/lp_bld_init.h"
 #include "gallivm/lp_bld_intr.h"
+#include "gallivm/lp_bld_flow.h"
+#include "gallivm/lp_bld_type.h"
 #include <llvm-c/Analysis.h>	/* for LLVMVerifyFunction */
 
 #include "lp_perf.h"
@@ -40,8 +44,6 @@
 #include "lp_flush.h"
 #include "lp_screen.h"
 #include "lp_context.h"
-#include "lp_setup_context.h"
-#include "lp_rast.h"
 #include "lp_state.h"
 #include "lp_state_fs.h"
 #include "lp_state_setup.h"
@@ -74,26 +76,37 @@ struct lp_setup_args
    LLVMValueRef dy01_ooa;
    LLVMValueRef dx20_ooa;
    LLVMValueRef dx01_ooa;
+
+   /* Temporary, per-attribute:
+    */
+   LLVMValueRef v0a;
+   LLVMValueRef v1a;
+   LLVMValueRef v2a;
 };
 
-static LLVMTypeRef type4f(void)
+
+
+static LLVMTypeRef
+type4f(struct gallivm_state *gallivm)
 {
-   return LLVMVectorType(LLVMFloatType(), 4);
+   return LLVMVectorType(LLVMFloatTypeInContext(gallivm->context), 4);
 }
 
 
 /* Equivalent of _mm_setr_ps(a,b,c,d)
  */
-static LLVMValueRef vec4f(LLVMBuilderRef bld,
-			  LLVMValueRef a, LLVMValueRef b, LLVMValueRef c, LLVMValueRef d,
-			  const char *name)
+static LLVMValueRef
+vec4f(struct gallivm_state *gallivm,
+      LLVMValueRef a, LLVMValueRef b, LLVMValueRef c, LLVMValueRef d,
+      const char *name)
 {
-   LLVMValueRef i0 = LLVMConstInt(LLVMInt32Type(), 0, 0);
-   LLVMValueRef i1 = LLVMConstInt(LLVMInt32Type(), 1, 0);
-   LLVMValueRef i2 = LLVMConstInt(LLVMInt32Type(), 2, 0);
-   LLVMValueRef i3 = LLVMConstInt(LLVMInt32Type(), 3, 0);
+   LLVMBuilderRef bld = gallivm->builder;
+   LLVMValueRef i0 = lp_build_const_int32(gallivm, 0);
+   LLVMValueRef i1 = lp_build_const_int32(gallivm, 1);
+   LLVMValueRef i2 = lp_build_const_int32(gallivm, 2);
+   LLVMValueRef i3 = lp_build_const_int32(gallivm, 3);
 
-   LLVMValueRef res = LLVMGetUndef(type4f());
+   LLVMValueRef res = LLVMGetUndef(type4f(gallivm));
 
    res = LLVMBuildInsertElement(bld, res, a, i0, "");
    res = LLVMBuildInsertElement(bld, res, b, i1, "");
@@ -105,15 +118,17 @@ static LLVMValueRef vec4f(LLVMBuilderRef bld,
 
 /* Equivalent of _mm_set1_ps(a)
  */
-static LLVMValueRef vec4f_from_scalar(LLVMBuilderRef bld,
-				      LLVMValueRef a,
-				      const char *name)
+static LLVMValueRef
+vec4f_from_scalar(struct gallivm_state *gallivm,
+                  LLVMValueRef a,
+                  const char *name)
 {
-   LLVMValueRef res = LLVMGetUndef(type4f());
+   LLVMBuilderRef bld = gallivm->builder;
+   LLVMValueRef res = LLVMGetUndef(type4f(gallivm));
    int i;
 
    for(i = 0; i < 4; ++i) {
-      LLVMValueRef index = LLVMConstInt(LLVMInt32Type(), i, 0);
+      LLVMValueRef index = lp_build_const_int32(gallivm, i);
       res = LLVMBuildInsertElement(bld, res, a, index, i == 3 ? name : "");
    }
 
@@ -121,14 +136,15 @@ static LLVMValueRef vec4f_from_scalar(LLVMBuilderRef bld,
 }
 
 static void
-store_coef(LLVMBuilderRef builder,
+store_coef(struct gallivm_state *gallivm,
 	   struct lp_setup_args *args,
 	   unsigned slot,
 	   LLVMValueRef a0,
 	   LLVMValueRef dadx,
 	   LLVMValueRef dady)
 {
-   LLVMValueRef idx = LLVMConstInt(LLVMInt32Type(), slot, 0);
+   LLVMBuilderRef builder = gallivm->builder;
+   LLVMValueRef idx = lp_build_const_int32(gallivm, slot);
    
    LLVMBuildStore(builder,
 		  a0, 
@@ -146,19 +162,14 @@ store_coef(LLVMBuilderRef builder,
 
 
 static void 
-emit_constant_coef4( LLVMBuilderRef builder,
+emit_constant_coef4(struct gallivm_state *gallivm,
 		     struct lp_setup_args *args,
 		     unsigned slot,
-		     LLVMValueRef vert,
-		     unsigned attr)
+		     LLVMValueRef vert)
 {
-   LLVMValueRef zero      = LLVMConstReal(LLVMFloatType(), 0.0);
-   LLVMValueRef zerovec   = vec4f_from_scalar(builder, zero, "zero");
-   LLVMValueRef idx       = LLVMConstInt(LLVMInt32Type(), attr, 0);
-   LLVMValueRef attr_ptr  = LLVMBuildGEP(builder, vert, &idx, 1, "attr_ptr");
-   LLVMValueRef vert_attr = LLVMBuildLoad(builder, attr_ptr, "vert_attr");
-
-   store_coef(builder, args, slot, vert_attr, zerovec, zerovec);
+   LLVMValueRef zero      = lp_build_const_float(gallivm, 0.0);
+   LLVMValueRef zerovec   = vec4f_from_scalar(gallivm, zero, "zero");
+   store_coef(gallivm, args, slot, vert, zerovec, zerovec);
 }
 
 
@@ -168,43 +179,194 @@ emit_constant_coef4( LLVMBuilderRef builder,
  * \param frontface  is the triangle front facing?
  */
 static void 
-emit_facing_coef( LLVMBuilderRef builder,
+emit_facing_coef(struct gallivm_state *gallivm,
 		  struct lp_setup_args *args,
 		  unsigned slot )
 {
+   LLVMBuilderRef builder = gallivm->builder;
+   LLVMTypeRef float_type = LLVMFloatTypeInContext(gallivm->context);
    LLVMValueRef a0_0 = args->facing;
-   LLVMValueRef a0_0f = LLVMBuildSIToFP(builder, a0_0, LLVMFloatType(), "");
-   LLVMValueRef zero = LLVMConstReal(LLVMFloatType(), 0.0);
-   LLVMValueRef a0      = vec4f(builder, a0_0f, zero, zero, zero, "facing");
-   LLVMValueRef zerovec = vec4f_from_scalar(builder, zero, "zero");
+   LLVMValueRef a0_0f = LLVMBuildSIToFP(builder, a0_0, float_type, "");
+   LLVMValueRef zero = lp_build_const_float(gallivm, 0.0);
+   LLVMValueRef a0 = vec4f(gallivm, a0_0f, zero, zero, zero, "facing");
+   LLVMValueRef zerovec = vec4f_from_scalar(gallivm, zero, "zero");
 
-   store_coef(builder, args, slot, a0, zerovec, zerovec);
+   store_coef(gallivm, args, slot, a0, zerovec, zerovec);
 }
 
 
 static LLVMValueRef
-vert_attrib(LLVMBuilderRef b,
+vert_attrib(struct gallivm_state *gallivm,
 	    LLVMValueRef vert,
 	    int attr,
 	    int elem,
 	    const char *name)
 {
+   LLVMBuilderRef b = gallivm->builder;
    LLVMValueRef idx[2];
-   idx[0] = LLVMConstInt(LLVMInt32Type(), attr, 0);
-   idx[1] = LLVMConstInt(LLVMInt32Type(), elem, 0);
+   idx[0] = lp_build_const_int32(gallivm, attr);
+   idx[1] = lp_build_const_int32(gallivm, elem);
    return LLVMBuildLoad(b, LLVMBuildGEP(b, vert, idx, 2, ""), name);
 }
 
+static LLVMValueRef
+vert_clamp(LLVMBuilderRef b,
+           LLVMValueRef x,
+           LLVMValueRef min,
+           LLVMValueRef max)
+{
+   LLVMValueRef min_result = LLVMBuildFCmp(b, LLVMRealUGT, min, x, "");
+   LLVMValueRef max_result = LLVMBuildFCmp(b, LLVMRealUGT, x, max, "");
+   LLVMValueRef clamp_value;
+
+   clamp_value = LLVMBuildSelect(b, min_result, min, x, "");
+   clamp_value = LLVMBuildSelect(b, max_result, max, x, "");
 
+   return clamp_value;
+}
+
+static void
+lp_twoside(struct gallivm_state *gallivm,
+           struct lp_setup_args *args,
+           const struct lp_setup_variant_key *key,
+           int bcolor_slot)
+{
+   LLVMBuilderRef b = gallivm->builder;
+   LLVMValueRef a0_back, a1_back, a2_back;
+   LLVMValueRef idx2 = lp_build_const_int32(gallivm, bcolor_slot);
+
+   LLVMValueRef facing = args->facing;
+   LLVMValueRef front_facing = LLVMBuildICmp(b, LLVMIntEQ, facing, lp_build_const_int32(gallivm, 0), ""); /** need i1 for if condition */
+   
+   a0_back = LLVMBuildLoad(b, LLVMBuildGEP(b, args->v0, &idx2, 1, ""), "v0a_back");
+   a1_back = LLVMBuildLoad(b, LLVMBuildGEP(b, args->v1, &idx2, 1, ""), "v1a_back");
+   a2_back = LLVMBuildLoad(b, LLVMBuildGEP(b, args->v2, &idx2, 1, ""), "v2a_back");
+
+   /* Possibly swap the front and back attrib values,
+    *
+    * Prefer select to if so we don't have to worry about phis or
+    * allocas.
+    */
+   args->v0a = LLVMBuildSelect(b, front_facing, a0_back, args->v0a, "");
+   args->v1a = LLVMBuildSelect(b, front_facing, a1_back, args->v1a, "");
+   args->v2a = LLVMBuildSelect(b, front_facing, a2_back, args->v2a, "");
+
+}
+
+static void
+lp_do_offset_tri(struct gallivm_state *gallivm,
+                 struct lp_setup_args *args,
+                 const struct lp_setup_variant_key *key)
+{
+   LLVMBuilderRef b = gallivm->builder;
+   struct lp_build_context bld;
+   LLVMValueRef zoffset, mult;
+   LLVMValueRef z0_new, z1_new, z2_new;
+   LLVMValueRef dzdx0, dzdx, dzdy0, dzdy;
+   LLVMValueRef max, max_value;
+   
+   LLVMValueRef one  = lp_build_const_float(gallivm, 1.0);
+   LLVMValueRef zero = lp_build_const_float(gallivm, 0.0);
+   LLVMValueRef two  = lp_build_const_int32(gallivm, 2);
+
+   /* edge vectors: e = v0 - v2, f = v1 - v2 */
+   LLVMValueRef v0_x = vert_attrib(gallivm, args->v0, 0, 0, "v0_x");
+   LLVMValueRef v1_x = vert_attrib(gallivm, args->v1, 0, 0, "v1_x");
+   LLVMValueRef v2_x = vert_attrib(gallivm, args->v2, 0, 0, "v2_x");
+   LLVMValueRef v0_y = vert_attrib(gallivm, args->v0, 0, 1, "v0_y");
+   LLVMValueRef v1_y = vert_attrib(gallivm, args->v1, 0, 1, "v1_y");
+   LLVMValueRef v2_y = vert_attrib(gallivm, args->v2, 0, 1, "v2_y");
+   LLVMValueRef v0_z = vert_attrib(gallivm, args->v0, 0, 2, "v0_z");
+   LLVMValueRef v1_z = vert_attrib(gallivm, args->v1, 0, 2, "v1_z");
+   LLVMValueRef v2_z = vert_attrib(gallivm, args->v2, 0, 2, "v2_z");
+ 
+   /* edge vectors: e = v0 - v2, f = v1 - v2 */
+   LLVMValueRef dx02 = LLVMBuildFSub(b, v0_x, v2_x, "dx02");
+   LLVMValueRef dy02 = LLVMBuildFSub(b, v0_y, v2_y, "dy02");
+   LLVMValueRef dz02 = LLVMBuildFSub(b, v0_z, v2_z, "dz02");
+   LLVMValueRef dx12 = LLVMBuildFSub(b, v1_x, v2_x, "dx12"); 
+   LLVMValueRef dy12 = LLVMBuildFSub(b, v1_y, v2_y, "dy12");
+   LLVMValueRef dz12 = LLVMBuildFSub(b, v1_z, v2_z, "dz12");
+ 
+   /* det = cross(e,f).z */
+   LLVMValueRef dx02_dy12  = LLVMBuildFMul(b, dx02, dy12, "dx02_dy12");
+   LLVMValueRef dy02_dx12  = LLVMBuildFMul(b, dy02, dx12, "dy02_dx12");
+   LLVMValueRef det  = LLVMBuildFSub(b, dx02_dy12, dy02_dx12, "det");
+   LLVMValueRef inv_det = LLVMBuildFDiv(b, one, det, "inv_det"); 
+   
+   /* (res1,res2) = cross(e,f).xy */
+   LLVMValueRef dy02_dz12    = LLVMBuildFMul(b, dy02, dz12, "dy02_dz12");
+   LLVMValueRef dz02_dy12    = LLVMBuildFMul(b, dz02, dy12, "dz02_dy12");
+   LLVMValueRef dz02_dx12    = LLVMBuildFMul(b, dz02, dx12, "dz02_dx12");
+   LLVMValueRef dx02_dz12    = LLVMBuildFMul(b, dx02, dz12, "dx02_dz12");
+   LLVMValueRef res1  = LLVMBuildFSub(b, dy02_dz12, dz02_dy12, "res1");
+   LLVMValueRef res2  = LLVMBuildFSub(b, dz02_dx12, dx02_dz12, "res2");
+
+   /* dzdx = fabsf(res1 * inv_det), dydx = fabsf(res2 * inv_det)*/
+   lp_build_context_init(&bld, gallivm, lp_type_float(32));
+   dzdx0 = LLVMBuildFMul(b, res1, inv_det, "dzdx");
+   dzdx  = lp_build_abs(&bld, dzdx0);
+   dzdy0 = LLVMBuildFMul(b, res2, inv_det, "dzdy");
+   dzdy  = lp_build_abs(&bld, dzdy0);
+
+   /* zoffset = offset->units + MAX2(dzdx, dzdy) * offset->scale */
+   max = LLVMBuildFCmp(b, LLVMRealUGT, dzdx, dzdy, "");
+   max_value = LLVMBuildSelect(b, max, dzdx, dzdy, "max"); 
+
+   mult = LLVMBuildFMul(b, max_value, lp_build_const_float(gallivm, key->scale), "");
+   zoffset = LLVMBuildFAdd(b, lp_build_const_float(gallivm, key->units), mult, "zoffset");
+
+   /* clamp and do offset */
+   z0_new = vert_clamp(b, LLVMBuildFAdd(b, v0_z, zoffset, ""), zero, one);
+   z1_new = vert_clamp(b, LLVMBuildFAdd(b, v1_z, zoffset, ""), zero, one);
+   z2_new = vert_clamp(b, LLVMBuildFAdd(b, v2_z, zoffset, ""), zero, one);
+
+   /* insert into args->a0.z, a1.z, a2.z:
+    */   
+   args->v0a = LLVMBuildInsertElement(b, args->v0a, z0_new, two, "");
+   args->v1a = LLVMBuildInsertElement(b, args->v1a, z1_new, two, "");
+   args->v2a = LLVMBuildInsertElement(b, args->v2a, z2_new, two, "");
+}
+
+static void
+load_attribute(struct gallivm_state *gallivm,
+               struct lp_setup_args *args,
+               const struct lp_setup_variant_key *key,
+               unsigned vert_attr)
+{
+   LLVMBuilderRef b = gallivm->builder;
+   LLVMValueRef idx = lp_build_const_int32(gallivm, vert_attr);
+
+   /* Load the vertex data
+    */
+   args->v0a = LLVMBuildLoad(b, LLVMBuildGEP(b, args->v0, &idx, 1, ""), "v0a");
+   args->v1a = LLVMBuildLoad(b, LLVMBuildGEP(b, args->v1, &idx, 1, ""), "v1a");
+   args->v2a = LLVMBuildLoad(b, LLVMBuildGEP(b, args->v2, &idx, 1, ""), "v2a");
+
+
+   /* Potentially modify it according to twoside, offset, etc:
+    */
+   if (vert_attr == 0 && (key->scale != 0.0f || key->units != 0.0f)) {
+      lp_do_offset_tri(gallivm, args, key);
+   }
+
+   if (key->twoside) {
+      if (vert_attr == key->color_slot && key->bcolor_slot != ~0)
+         lp_twoside(gallivm, args, key, key->bcolor_slot);
+      else if (vert_attr == key->spec_slot && key->bspec_slot != ~0)
+         lp_twoside(gallivm, args, key, key->bspec_slot);
+   }
+}
 
 static void 
-emit_coef4( LLVMBuilderRef b,
+emit_coef4( struct gallivm_state *gallivm,
 	    struct lp_setup_args *args,
 	    unsigned slot,
 	    LLVMValueRef a0,
 	    LLVMValueRef a1,
 	    LLVMValueRef a2)
 {
+   LLVMBuilderRef b = gallivm->builder;
    LLVMValueRef dy20_ooa = args->dy20_ooa;
    LLVMValueRef dy01_ooa = args->dy01_ooa;
    LLVMValueRef dx20_ooa = args->dx20_ooa;
@@ -236,27 +398,24 @@ emit_coef4( LLVMBuilderRef b,
    LLVMValueRef attr_v0       = LLVMBuildFAdd(b, dadx_x0, dady_y0, "attr_v0"); 
    LLVMValueRef attr_0        = LLVMBuildFSub(b, a0, attr_v0, "attr_0"); 
 
-   store_coef(b, args, slot, attr_0, dadx, dady);
+   store_coef(gallivm, args, slot, attr_0, dadx, dady);
 }
 
 
 static void 
-emit_linear_coef( LLVMBuilderRef b,
+emit_linear_coef( struct gallivm_state *gallivm,
 		  struct lp_setup_args *args,
-		  unsigned slot,
-		  unsigned vert_attr)
+		  unsigned slot)
 {
-   LLVMValueRef idx = LLVMConstInt(LLVMInt32Type(), vert_attr, 0);
-
-   LLVMValueRef a0 = LLVMBuildLoad(b, LLVMBuildGEP(b, args->v0, &idx, 1, ""), "v0a");
-   LLVMValueRef a1 = LLVMBuildLoad(b, LLVMBuildGEP(b, args->v1, &idx, 1, ""), "v1a");
-   LLVMValueRef a2 = LLVMBuildLoad(b, LLVMBuildGEP(b, args->v2, &idx, 1, ""), "v2a");
-
-   emit_coef4(b, args, slot, a0, a1, a2);
+   /* nothing to do anymore */
+   emit_coef4(gallivm,
+              args, slot, 
+              args->v0a,
+              args->v1a,
+              args->v2a);
 }
 
 
-
 /**
  * Compute a0, dadx and dady for a perspective-corrected interpolant,
  * for a triangle.
@@ -266,37 +425,32 @@ emit_linear_coef( LLVMBuilderRef b,
  * divide the interpolated value by the interpolated W at that fragment.
  */
 static void 
-emit_perspective_coef( LLVMBuilderRef b,
+emit_perspective_coef( struct gallivm_state *gallivm,
 		       struct lp_setup_args *args,
-		       unsigned slot,
-		       unsigned vert_attr)
+		       unsigned slot)
 {
+   LLVMBuilderRef b = gallivm->builder;
+
    /* premultiply by 1/w  (v[0][3] is always 1/w):
     */
-   LLVMValueRef idx = LLVMConstInt(LLVMInt32Type(), vert_attr, 0);
-
-   LLVMValueRef v0a = LLVMBuildLoad(b, LLVMBuildGEP(b, args->v0, &idx, 1, ""), "v0a");
-   LLVMValueRef v1a = LLVMBuildLoad(b, LLVMBuildGEP(b, args->v1, &idx, 1, ""), "v1a");
-   LLVMValueRef v2a = LLVMBuildLoad(b, LLVMBuildGEP(b, args->v2, &idx, 1, ""), "v2a");
-
-   LLVMValueRef v0_oow = vec4f_from_scalar(b, vert_attrib(b, args->v0, 0, 3, ""), "v0_oow");
-   LLVMValueRef v1_oow = vec4f_from_scalar(b, vert_attrib(b, args->v1, 0, 3, ""), "v1_oow");
-   LLVMValueRef v2_oow = vec4f_from_scalar(b, vert_attrib(b, args->v2, 0, 3, ""), "v2_oow");
+   LLVMValueRef v0_oow = vec4f_from_scalar(gallivm, vert_attrib(gallivm, args->v0, 0, 3, ""), "v0_oow");
+   LLVMValueRef v1_oow = vec4f_from_scalar(gallivm, vert_attrib(gallivm, args->v1, 0, 3, ""), "v1_oow");
+   LLVMValueRef v2_oow = vec4f_from_scalar(gallivm, vert_attrib(gallivm, args->v2, 0, 3, ""), "v2_oow");
 
-   LLVMValueRef v0_oow_v0a = LLVMBuildFMul(b, v0a, v0_oow, "v0_oow_v0a");
-   LLVMValueRef v1_oow_v1a = LLVMBuildFMul(b, v1a, v1_oow, "v1_oow_v1a");
-   LLVMValueRef v2_oow_v2a = LLVMBuildFMul(b, v2a, v2_oow, "v2_oow_v2a");
+   LLVMValueRef v0_oow_v0a = LLVMBuildFMul(b, args->v0a, v0_oow, "v0_oow_v0a");
+   LLVMValueRef v1_oow_v1a = LLVMBuildFMul(b, args->v1a, v1_oow, "v1_oow_v1a");
+   LLVMValueRef v2_oow_v2a = LLVMBuildFMul(b, args->v2a, v2_oow, "v2_oow_v2a");
 
-   emit_coef4(b, args, slot, v0_oow_v0a, v1_oow_v1a, v2_oow_v2a);
+   emit_coef4(gallivm, args, slot, v0_oow_v0a, v1_oow_v1a, v2_oow_v2a);
 }
 
 
 static void
-emit_position_coef( LLVMBuilderRef builder,
+emit_position_coef( struct gallivm_state *gallivm,
 		    struct lp_setup_args *args,
-		    int slot, int attrib )
+		    int slot )
 {
-   emit_linear_coef(builder, args, slot, attrib);
+   emit_linear_coef(gallivm, args, slot);
 }
 
 
@@ -306,7 +460,7 @@ emit_position_coef( LLVMBuilderRef builder,
  * Compute the inputs-> dadx, dady, a0 values.
  */
 static void 
-emit_tri_coef( LLVMBuilderRef builder,
+emit_tri_coef( struct gallivm_state *gallivm,
 	       const struct lp_setup_variant_key *key,
 	       struct lp_setup_args *args )
 {
@@ -314,29 +468,34 @@ emit_tri_coef( LLVMBuilderRef builder,
 
    /* The internal position input is in slot zero:
     */
-   emit_position_coef(builder, args, 0, 0);
+   load_attribute(gallivm, args, key, 0);
+   emit_position_coef(gallivm, args, 0);
 
    /* setup interpolation for all the remaining attributes:
     */
    for (slot = 0; slot < key->num_inputs; slot++) {
-      unsigned vert_attr = key->inputs[slot].src_index;
+
+      if (key->inputs[slot].interp == LP_INTERP_CONSTANT ||
+          key->inputs[slot].interp == LP_INTERP_LINEAR ||
+          key->inputs[slot].interp == LP_INTERP_PERSPECTIVE)
+         load_attribute(gallivm, args, key, key->inputs[slot].src_index);
 
       switch (key->inputs[slot].interp) {
       case LP_INTERP_CONSTANT:
 	 if (key->flatshade_first) {
-	    emit_constant_coef4(builder, args, slot+1, args->v0, vert_attr);
+	    emit_constant_coef4(gallivm, args, slot+1, args->v0a);
 	 }
 	 else {
-	    emit_constant_coef4(builder, args, slot+1, args->v2, vert_attr);
+	    emit_constant_coef4(gallivm, args, slot+1, args->v2a);
 	 }
 	 break;
 
       case LP_INTERP_LINEAR:
-	 emit_linear_coef(builder, args, slot+1, vert_attr);
+	 emit_linear_coef(gallivm, args, slot+1);
          break;
 
       case LP_INTERP_PERSPECTIVE:
-	 emit_perspective_coef(builder, args, slot+1, vert_attr);
+	 emit_perspective_coef(gallivm, args, slot+1);
          break;
 
       case LP_INTERP_POSITION:
@@ -347,7 +506,7 @@ emit_tri_coef( LLVMBuilderRef builder,
          break;
 
       case LP_INTERP_FACING:
-         emit_facing_coef(builder, args, slot+1);
+         emit_facing_coef(gallivm, args, slot+1);
          break;
 
       default:
@@ -360,7 +519,7 @@ emit_tri_coef( LLVMBuilderRef builder,
 /* XXX: This is generic code, share with fs/vs codegen:
  */
 static lp_jit_setup_triangle
-finalize_function(struct llvmpipe_screen *screen,
+finalize_function(struct gallivm_state *gallivm,
 		  LLVMBuilderRef builder,
 		  LLVMValueRef function)
 {
@@ -376,7 +535,7 @@ finalize_function(struct llvmpipe_screen *screen,
 #endif
 
    /* Apply optimizations to LLVM IR */
-   LLVMRunFunctionPassManager(screen->pass, function);
+   LLVMRunFunctionPassManager(gallivm->passmgr, function);
 
    if (gallivm_debug & GALLIVM_DEBUG_IR)
    {
@@ -388,7 +547,7 @@ finalize_function(struct llvmpipe_screen *screen,
    /*
     * Translate the LLVM IR into machine code.
     */
-   f = LLVMGetPointerToGlobal(screen->engine, function);
+   f = LLVMGetPointerToGlobal(gallivm->engine, function);
 
    if (gallivm_debug & GALLIVM_DEBUG_ASM)
    {
@@ -403,11 +562,12 @@ finalize_function(struct llvmpipe_screen *screen,
 /* XXX: Generic code:
  */
 static void
-lp_emit_emms(LLVMBuilderRef builder)
+lp_emit_emms(struct gallivm_state *gallivm)
 {
 #ifdef PIPE_ARCH_X86
    /* Avoid corrupting the FPU stack on 32bit OSes. */
-   lp_build_intrinsic(builder, "llvm.x86.mmx.emms", LLVMVoidType(), NULL, 0);
+   lp_build_intrinsic(gallivm->builder, "llvm.x86.mmx.emms",
+         LLVMVoidTypeInContext(gallivm->context), NULL, 0);
 #endif
 }
 
@@ -428,21 +588,23 @@ set_noalias(LLVMBuilderRef builder,
 }
 
 static void
-init_args(LLVMBuilderRef b, 
+init_args(struct gallivm_state *gallivm,
 	  struct lp_setup_args *args,
 	  const struct lp_setup_variant *variant)
 {
-   LLVMValueRef v0_x = vert_attrib(b, args->v0, 0, 0, "v0_x");
-   LLVMValueRef v0_y = vert_attrib(b, args->v0, 0, 1, "v0_y");
+   LLVMBuilderRef b = gallivm->builder;
+
+   LLVMValueRef v0_x = vert_attrib(gallivm, args->v0, 0, 0, "v0_x");
+   LLVMValueRef v0_y = vert_attrib(gallivm, args->v0, 0, 1, "v0_y");
 
-   LLVMValueRef v1_x = vert_attrib(b, args->v1, 0, 0, "v1_x");
-   LLVMValueRef v1_y = vert_attrib(b, args->v1, 0, 1, "v1_y");
+   LLVMValueRef v1_x = vert_attrib(gallivm, args->v1, 0, 0, "v1_x");
+   LLVMValueRef v1_y = vert_attrib(gallivm, args->v1, 0, 1, "v1_y");
 
-   LLVMValueRef v2_x = vert_attrib(b, args->v2, 0, 0, "v2_x");
-   LLVMValueRef v2_y = vert_attrib(b, args->v2, 0, 1, "v2_y");
+   LLVMValueRef v2_x = vert_attrib(gallivm, args->v2, 0, 0, "v2_x");
+   LLVMValueRef v2_y = vert_attrib(gallivm, args->v2, 0, 1, "v2_y");
 
-   LLVMValueRef pixel_center = LLVMConstReal(LLVMFloatType(),
-					     variant->key.pixel_center_half ? 0.5 : 0);
+   LLVMValueRef pixel_center = lp_build_const_float(gallivm,
+                                   variant->key.pixel_center_half ? 0.5 : 0);
 
    LLVMValueRef x0_center = LLVMBuildFSub(b, v0_x, pixel_center, "x0_center" );
    LLVMValueRef y0_center = LLVMBuildFSub(b, v0_y, pixel_center, "y0_center" );
@@ -452,7 +614,7 @@ init_args(LLVMBuilderRef b,
    LLVMValueRef dx20 = LLVMBuildFSub(b, v2_x, v0_x, "dx20");
    LLVMValueRef dy20 = LLVMBuildFSub(b, v2_y, v0_y, "dy20");
 
-   LLVMValueRef one  = LLVMConstReal(LLVMFloatType(), 1.0);
+   LLVMValueRef one  = lp_build_const_float(gallivm, 1.0);
    LLVMValueRef e    = LLVMBuildFMul(b, dx01, dy20, "e");
    LLVMValueRef f    = LLVMBuildFMul(b, dx20, dy01, "f");
    LLVMValueRef ooa  = LLVMBuildFDiv(b, one, LLVMBuildFSub(b, e, f, ""), "ooa");
@@ -462,14 +624,14 @@ init_args(LLVMBuilderRef b,
    LLVMValueRef dx20_ooa = LLVMBuildFMul(b, dx20, ooa, "dx20_ooa");
    LLVMValueRef dx01_ooa = LLVMBuildFMul(b, dx01, ooa, "dx01_ooa");
 
-   args->dy20_ooa  = vec4f_from_scalar(b, dy20_ooa, "dy20_ooa_4f");
-   args->dy01_ooa  = vec4f_from_scalar(b, dy01_ooa, "dy01_ooa_4f");
+   args->dy20_ooa  = vec4f_from_scalar(gallivm, dy20_ooa, "dy20_ooa_4f");
+   args->dy01_ooa  = vec4f_from_scalar(gallivm, dy01_ooa, "dy01_ooa_4f");
 
-   args->dx20_ooa  = vec4f_from_scalar(b, dx20_ooa, "dx20_ooa_4f");
-   args->dx01_ooa  = vec4f_from_scalar(b, dx01_ooa, "dx01_ooa_4f");
+   args->dx20_ooa  = vec4f_from_scalar(gallivm, dx20_ooa, "dx20_ooa_4f");
+   args->dx01_ooa  = vec4f_from_scalar(gallivm, dx01_ooa, "dx01_ooa_4f");
 
-   args->x0_center = vec4f_from_scalar(b, x0_center, "x0_center_4f");
-   args->y0_center = vec4f_from_scalar(b, y0_center, "y0_center_4f");
+   args->x0_center = vec4f_from_scalar(gallivm, x0_center, "x0_center_4f");
+   args->y0_center = vec4f_from_scalar(gallivm, y0_center, "y0_center_4f");
 }
 
 /**
@@ -477,8 +639,9 @@ init_args(LLVMBuilderRef b,
  *
  */
 static struct lp_setup_variant *
-generate_setup_variant(struct llvmpipe_screen *screen,
-		       struct lp_setup_variant_key *key)
+generate_setup_variant(struct gallivm_state *gallivm,
+		       struct lp_setup_variant_key *key,
+                       struct llvmpipe_context *lp)
 {
    struct lp_setup_variant *variant = NULL;
    struct lp_setup_args args;
@@ -487,7 +650,7 @@ generate_setup_variant(struct llvmpipe_screen *screen,
    LLVMTypeRef func_type;
    LLVMTypeRef arg_types[7];
    LLVMBasicBlockRef block;
-   LLVMBuilderRef builder;
+   LLVMBuilderRef builder = gallivm->builder;
    int64_t t0, t1;
 
    if (0)
@@ -512,19 +675,20 @@ generate_setup_variant(struct llvmpipe_screen *screen,
     * the vertices.
     */
 
-   vec4f_type = LLVMVectorType(LLVMFloatType(), 4);
+   vec4f_type = LLVMVectorType(LLVMFloatTypeInContext(gallivm->context), 4);
 
    arg_types[0] = LLVMPointerType(vec4f_type, 0);        /* v0 */
    arg_types[1] = LLVMPointerType(vec4f_type, 0);        /* v1 */
    arg_types[2] = LLVMPointerType(vec4f_type, 0);        /* v2 */
-   arg_types[3] = LLVMInt32Type();			/* facing */
+   arg_types[3] = LLVMInt32TypeInContext(gallivm->context); /* facing */
    arg_types[4] = LLVMPointerType(vec4f_type, 0);	/* a0, aligned */
    arg_types[5] = LLVMPointerType(vec4f_type, 0);	/* dadx, aligned */
    arg_types[6] = LLVMPointerType(vec4f_type, 0);	/* dady, aligned */
 
-   func_type = LLVMFunctionType(LLVMVoidType(), arg_types, Elements(arg_types), 0);
+   func_type = LLVMFunctionType(LLVMVoidTypeInContext(gallivm->context),
+                                arg_types, Elements(arg_types), 0);
 
-   variant->function = LLVMAddFunction(screen->module, func_name, func_type);
+   variant->function = LLVMAddFunction(gallivm->module, func_name, func_type);
    if (!variant->function)
       goto fail;
 
@@ -549,19 +713,18 @@ generate_setup_variant(struct llvmpipe_screen *screen,
    /*
     * Function body
     */
-   block = LLVMAppendBasicBlock(variant->function, "entry");
-   builder = LLVMCreateBuilder();
+   block = LLVMAppendBasicBlockInContext(gallivm->context,
+                                         variant->function, "entry");
    LLVMPositionBuilderAtEnd(builder, block);
 
    set_noalias(builder, variant->function, arg_types, Elements(arg_types));
-   init_args(builder, &args, variant);
-   emit_tri_coef(builder, &variant->key, &args);
+   init_args(gallivm, &args, variant);
+   emit_tri_coef(gallivm, &variant->key, &args);
 
-   lp_emit_emms(builder);
+   lp_emit_emms(gallivm);
    LLVMBuildRetVoid(builder);
-   LLVMDisposeBuilder(builder);
 
-   variant->jit_function = finalize_function(screen, builder,
+   variant->jit_function = finalize_function(gallivm, builder,
 					     variant->function);
    if (!variant->jit_function)
       goto fail;
@@ -581,7 +744,7 @@ fail:
    if (variant) {
       if (variant->function) {
 	 if (variant->jit_function)
-	    LLVMFreeMachineCodeForFunction(screen->engine,
+	    LLVMFreeMachineCodeForFunction(gallivm->engine,
 					   variant->function);
 	 LLVMDeleteFunction(variant->function);
       }
@@ -605,14 +768,20 @@ lp_make_setup_variant_key(struct llvmpipe_context *lp,
    key->num_inputs = fs->info.base.num_inputs;
    key->flatshade_first = lp->rasterizer->flatshade_first;
    key->pixel_center_half = lp->rasterizer->gl_rasterization_rules;
+   key->twoside = lp->rasterizer->light_twoside;
    key->size = Offset(struct lp_setup_variant_key,
 		      inputs[key->num_inputs]);
+   key->color_slot = lp->color_slot[0];
+   key->bcolor_slot = lp->bcolor_slot[0];
+   key->spec_slot = lp->color_slot[1];
+   key->bspec_slot = lp->bcolor_slot[1];
+   key->units = (float) (lp->rasterizer->offset_units * lp->mrd);
+   key->scale = lp->rasterizer->offset_scale;
    key->pad = 0;
-
    memcpy(key->inputs, fs->inputs, key->num_inputs * sizeof key->inputs[0]);
    for (i = 0; i < key->num_inputs; i++) {
       if (key->inputs[i].interp == LP_INTERP_COLOR) {
-	 if (lp->rasterizer->flatshade)
+         if (lp->rasterizer->flatshade)
 	    key->inputs[i].interp = LP_INTERP_CONSTANT;
 	 else
 	    key->inputs[i].interp = LP_INTERP_LINEAR;
@@ -626,8 +795,6 @@ static void
 remove_setup_variant(struct llvmpipe_context *lp,
 		     struct lp_setup_variant *variant)
 {
-   struct llvmpipe_screen *screen = llvmpipe_screen(lp->pipe.screen);
-
    if (gallivm_debug & GALLIVM_DEBUG_IR) {
       debug_printf("llvmpipe: del setup_variant #%u total %u\n",
 		   variant->no, lp->nr_setup_variants);
@@ -635,7 +802,7 @@ remove_setup_variant(struct llvmpipe_context *lp,
 
    if (variant->function) {
       if (variant->jit_function)
-	 LLVMFreeMachineCodeForFunction(screen->engine,
+	 LLVMFreeMachineCodeForFunction(lp->gallivm->engine,
 					variant->function);
       LLVMDeleteFunction(variant->function);
    }
@@ -678,8 +845,6 @@ cull_setup_variants(struct llvmpipe_context *lp)
 void 
 llvmpipe_update_setup(struct llvmpipe_context *lp)
 {
-   struct llvmpipe_screen *screen = llvmpipe_screen(lp->pipe.screen);
-
    struct lp_setup_variant_key *key = &lp->setup_variant.key;
    struct lp_setup_variant *variant = NULL;
    struct lp_setup_variant_list_item *li;
@@ -702,9 +867,11 @@ llvmpipe_update_setup(struct llvmpipe_context *lp)
 	 cull_setup_variants(lp);
       }
 
-      variant = generate_setup_variant(screen, key);
+      variant = generate_setup_variant(lp->gallivm, key, lp);
       insert_at_head(&lp->setup_variants_list, &variant->list_item_global);
       lp->nr_setup_variants++;
+
+      llvmpipe_variant_count++;
    }
 
    lp_setup_set_setup_variant(lp->setup,
diff --git a/src/gallium/drivers/llvmpipe/lp_state_setup.h b/src/gallium/drivers/llvmpipe/lp_state_setup.h
index b0c81baa75..90c55ca4ce 100644
--- a/src/gallium/drivers/llvmpipe/lp_state_setup.h
+++ b/src/gallium/drivers/llvmpipe/lp_state_setup.h
@@ -15,11 +15,20 @@ struct lp_setup_variant_list_item
 
 
 struct lp_setup_variant_key {   
+   unsigned size:16;
    unsigned num_inputs:8;
+   unsigned color_slot:8;
+
+   unsigned bcolor_slot:8;
+   unsigned spec_slot:8;
+   unsigned bspec_slot:8;
    unsigned flatshade_first:1;
    unsigned pixel_center_half:1;
-   unsigned pad:7;
-   unsigned size:16;
+   unsigned twoside:1;
+   unsigned pad:5;
+
+   float units;
+   float scale;      
    struct lp_shader_input inputs[PIPE_MAX_SHADER_INPUTS];
 };
 
diff --git a/src/gallium/drivers/llvmpipe/lp_state_surface.c b/src/gallium/drivers/llvmpipe/lp_state_surface.c
index cd1a5b1980..375ceb2b94 100644
--- a/src/gallium/drivers/llvmpipe/lp_state_surface.c
+++ b/src/gallium/drivers/llvmpipe/lp_state_surface.c
@@ -77,6 +77,7 @@ llvmpipe_set_framebuffer_state(struct pipe_context *pipe,
          else {
             mrd = 0.00002;
          }
+         lp->mrd = mrd;
          draw_set_mrd(lp->draw, mrd);
       }
 
diff --git a/src/gallium/drivers/llvmpipe/lp_surface.c b/src/gallium/drivers/llvmpipe/lp_surface.c
index 164242eda6..e7e46a628a 100644
--- a/src/gallium/drivers/llvmpipe/lp_surface.c
+++ b/src/gallium/drivers/llvmpipe/lp_surface.c
@@ -52,19 +52,23 @@ adjust_to_tile_bounds(unsigned x, unsigned y, unsigned width, unsigned height,
 
 static void
 lp_resource_copy(struct pipe_context *pipe,
-                 struct pipe_resource *dst, struct pipe_subresource subdst,
+                 struct pipe_resource *dst, unsigned dst_level,
                  unsigned dstx, unsigned dsty, unsigned dstz,
-                 struct pipe_resource *src, struct pipe_subresource subsrc,
-                 unsigned srcx, unsigned srcy, unsigned srcz,
-                 unsigned width, unsigned height)
+                 struct pipe_resource *src, unsigned src_level,
+                 const struct pipe_box *src_box)
 {
-   /* XXX what about the dstz/srcz parameters - zslice wasn't used... */
+   /* XXX this used to ignore srcz/dstz
+    * assume it works the same for cube and 3d
+    */
    struct llvmpipe_resource *src_tex = llvmpipe_resource(src);
    struct llvmpipe_resource *dst_tex = llvmpipe_resource(dst);
    const enum pipe_format format = src_tex->base.format;
+   unsigned width = src_box->width;
+   unsigned height = src_box->height;
+   assert(src_box->depth == 1);
 
    llvmpipe_flush_resource(pipe,
-                           dst, subdst.face, subdst.level,
+                           dst, dst_level, dstz,
                            0, /* flush_flags */
                            FALSE, /* read_only */
                            TRUE, /* cpu_access */
@@ -72,7 +76,7 @@ lp_resource_copy(struct pipe_context *pipe,
                            "blit dest");
 
    llvmpipe_flush_resource(pipe,
-                           src, subsrc.face, subsrc.level,
+                           src, src_level, src_box->z,
                            0, /* flush_flags */
                            TRUE, /* read_only */
                            TRUE, /* cpu_access */
@@ -80,9 +84,10 @@ lp_resource_copy(struct pipe_context *pipe,
                            "blit src");
 
    /*
-   printf("surface copy from %u to %u: %u,%u to %u,%u %u x %u\n",
-          src_tex->id, dst_tex->id,
-          srcx, srcy, dstx, dsty, width, height);
+   printf("surface copy from %u lvl %u to %u lvl %u: %u,%u,%u to %u,%u,%u %u x %u x %u\n",
+          src_tex->id, src_level, dst_tex->id, dst_level, 
+          src_box->x, src_box->y, src_box->z, dstx, dsty, dstz,
+          src_box->width, src_box->height, src_box->depth);
    */
 
    /* set src tiles to linear layout */
@@ -90,12 +95,13 @@ lp_resource_copy(struct pipe_context *pipe,
       unsigned tx, ty, tw, th;
       unsigned x, y;
 
-      adjust_to_tile_bounds(srcx, srcy, width, height, &tx, &ty, &tw, &th);
+      adjust_to_tile_bounds(src_box->x, src_box->y, width, height,
+                            &tx, &ty, &tw, &th);
 
       for (y = 0; y < th; y += TILE_SIZE) {
          for (x = 0; x < tw; x += TILE_SIZE) {
             (void) llvmpipe_get_texture_tile_linear(src_tex,
-                                                    subsrc.face, subsrc.level,
+                                                    src_box->z, src_level,
                                                     LP_TEX_USAGE_READ,
                                                     tx + x, ty + y);
          }
@@ -130,7 +136,7 @@ lp_resource_copy(struct pipe_context *pipe,
                usage = LP_TEX_USAGE_READ_WRITE;
 
             (void) llvmpipe_get_texture_tile_linear(dst_tex,
-                                                    subdst.face, subdst.level,
+                                                    dstz, dst_level,
                                                     usage,
                                                     tx + x, ty + y);
          }
@@ -140,22 +146,22 @@ lp_resource_copy(struct pipe_context *pipe,
    /* copy */
    {
       const ubyte *src_linear_ptr
-         = llvmpipe_get_texture_image_address(src_tex, subsrc.face,
-                                              subsrc.level,
+         = llvmpipe_get_texture_image_address(src_tex, src_box->z,
+                                              src_level,
                                               LP_TEX_LAYOUT_LINEAR);
       ubyte *dst_linear_ptr
-         = llvmpipe_get_texture_image_address(dst_tex, subdst.face,
-                                              subdst.level,
+         = llvmpipe_get_texture_image_address(dst_tex, dstz,
+                                              dst_level,
                                               LP_TEX_LAYOUT_LINEAR);
 
       if (dst_linear_ptr && src_linear_ptr) {
          util_copy_rect(dst_linear_ptr, format,
-                        llvmpipe_resource_stride(&dst_tex->base, subdst.level),
+                        llvmpipe_resource_stride(&dst_tex->base, dst_level),
                         dstx, dsty,
                         width, height,
                         src_linear_ptr,
-                        llvmpipe_resource_stride(&src_tex->base, subsrc.level),
-                        srcx, srcy);
+                        llvmpipe_resource_stride(&src_tex->base, src_level),
+                        src_box->x, src_box->y);
       }
    }
 }
diff --git a/src/gallium/drivers/llvmpipe/lp_test.h b/src/gallium/drivers/llvmpipe/lp_test.h
index 90422e4258..c64f3e149f 100644
--- a/src/gallium/drivers/llvmpipe/lp_test.h
+++ b/src/gallium/drivers/llvmpipe/lp_test.h
@@ -64,13 +64,14 @@ write_tsv_header(FILE *fp);
 
 
 boolean
-test_some(unsigned verbose, FILE *fp, unsigned long n);
+test_some(struct gallivm_state *gallivm,unsigned verbose, FILE *fp,
+          unsigned long n);
 
 boolean
-test_single(unsigned verbose, FILE *fp);
+test_single(struct gallivm_state *gallivm, unsigned verbose, FILE *fp);
 
 boolean
-test_all(unsigned verbose, FILE *fp);
+test_all(struct gallivm_state *gallivm, unsigned verbose, FILE *fp);
 
 
 #if defined(PIPE_CC_MSVC)
diff --git a/src/gallium/drivers/llvmpipe/lp_test_blend.c b/src/gallium/drivers/llvmpipe/lp_test_blend.c
index 8b6b5e1298..b3ca134131 100644
--- a/src/gallium/drivers/llvmpipe/lp_test_blend.c
+++ b/src/gallium/drivers/llvmpipe/lp_test_blend.c
@@ -163,11 +163,13 @@ dump_blend_type(FILE *fp,
 
 
 static LLVMValueRef
-add_blend_test(LLVMModuleRef module,
+add_blend_test(struct gallivm_state *gallivm,
                const struct pipe_blend_state *blend,
                enum vector_mode mode,
                struct lp_type type)
 {
+   LLVMModuleRef module = gallivm->module;
+   LLVMContextRef context = gallivm->context;
    LLVMTypeRef vec_type;
    LLVMTypeRef args[4];
    LLVMValueRef func;
@@ -179,18 +181,18 @@ add_blend_test(LLVMModuleRef module,
    LLVMBuilderRef builder;
    const unsigned rt = 0;
 
-   vec_type = lp_build_vec_type(type);
+   vec_type = lp_build_vec_type(gallivm, type);
 
    args[3] = args[2] = args[1] = args[0] = LLVMPointerType(vec_type, 0);
-   func = LLVMAddFunction(module, "test", LLVMFunctionType(LLVMVoidType(), args, 4, 0));
+   func = LLVMAddFunction(module, "test", LLVMFunctionType(LLVMVoidTypeInContext(context), args, 4, 0));
    LLVMSetFunctionCallConv(func, LLVMCCallConv);
    src_ptr = LLVMGetParam(func, 0);
    dst_ptr = LLVMGetParam(func, 1);
    const_ptr = LLVMGetParam(func, 2);
    res_ptr = LLVMGetParam(func, 3);
 
-   block = LLVMAppendBasicBlock(func, "entry");
-   builder = LLVMCreateBuilder();
+   block = LLVMAppendBasicBlockInContext(context, func, "entry");
+   builder = gallivm->builder;
    LLVMPositionBuilderAtEnd(builder, block);
 
    if (mode == AoS) {
@@ -203,7 +205,7 @@ add_blend_test(LLVMModuleRef module,
       dst = LLVMBuildLoad(builder, dst_ptr, "dst");
       con = LLVMBuildLoad(builder, const_ptr, "const");
 
-      res = lp_build_blend_aos(builder, blend, type, rt, src, dst, con, 3);
+      res = lp_build_blend_aos(gallivm, blend, type, rt, src, dst, con, 3);
 
       lp_build_name(res, "res");
 
@@ -218,7 +220,7 @@ add_blend_test(LLVMModuleRef module,
       unsigned i;
 
       for(i = 0; i < 4; ++i) {
-         LLVMValueRef index = LLVMConstInt(LLVMInt32Type(), i, 0);
+         LLVMValueRef index = LLVMConstInt(LLVMInt32TypeInContext(context), i, 0);
          src[i] = LLVMBuildLoad(builder, LLVMBuildGEP(builder, src_ptr, &index, 1, ""), "");
          dst[i] = LLVMBuildLoad(builder, LLVMBuildGEP(builder, dst_ptr, &index, 1, ""), "");
          con[i] = LLVMBuildLoad(builder, LLVMBuildGEP(builder, const_ptr, &index, 1, ""), "");
@@ -227,10 +229,10 @@ add_blend_test(LLVMModuleRef module,
          lp_build_name(dst[i], "dst.%c", "rgba"[i]);
       }
 
-      lp_build_blend_soa(builder, blend, type, rt, src, dst, con, res);
+      lp_build_blend_soa(gallivm, blend, type, rt, src, dst, con, res);
 
       for(i = 0; i < 4; ++i) {
-         LLVMValueRef index = LLVMConstInt(LLVMInt32Type(), i, 0);
+         LLVMValueRef index = LLVMConstInt(LLVMInt32TypeInContext(context), i, 0);
          lp_build_name(res[i], "res.%c", "rgba"[i]);
          LLVMBuildStore(builder, res[i], LLVMBuildGEP(builder, res_ptr, &index, 1, ""));
       }
@@ -238,7 +240,6 @@ add_blend_test(LLVMModuleRef module,
 
    LLVMBuildRetVoid(builder);;
 
-   LLVMDisposeBuilder(builder);
    return func;
 }
 
@@ -465,16 +466,16 @@ compute_blend_ref(const struct pipe_blend_state *blend,
 
 PIPE_ALIGN_STACK
 static boolean
-test_one(unsigned verbose,
+test_one(struct gallivm_state *gallivm,
+         unsigned verbose,
          FILE *fp,
          const struct pipe_blend_state *blend,
          enum vector_mode mode,
          struct lp_type type)
 {
-   LLVMModuleRef module = NULL;
+   LLVMModuleRef module = gallivm->module;
    LLVMValueRef func = NULL;
-   LLVMExecutionEngineRef engine = lp_build_engine;
-   LLVMPassManagerRef pass = NULL;
+   LLVMExecutionEngineRef engine = gallivm->engine;
    char *error = NULL;
    blend_test_ptr_t blend_test_ptr;
    boolean success;
@@ -487,9 +488,7 @@ test_one(unsigned verbose,
    if(verbose >= 1)
       dump_blend_type(stdout, blend, mode, type);
 
-   module = LLVMModuleCreateWithName("test");
-
-   func = add_blend_test(module, blend, mode, type);
+   func = add_blend_test(gallivm, blend, mode, type);
 
    if(LLVMVerifyModule(module, LLVMPrintMessageAction, &error)) {
       LLVMDumpModule(module);
@@ -497,24 +496,6 @@ test_one(unsigned verbose,
    }
    LLVMDisposeMessage(error);
 
-#if 0
-   pass = LLVMCreatePassManager();
-   LLVMAddTargetData(LLVMGetExecutionEngineTargetData(engine), pass);
-   /* These are the passes currently listed in llvm-c/Transforms/Scalar.h,
-    * but there are more on SVN. */
-   LLVMAddConstantPropagationPass(pass);
-   LLVMAddInstructionCombiningPass(pass);
-   LLVMAddPromoteMemoryToRegisterPass(pass);
-   LLVMAddGVNPass(pass);
-   LLVMAddCFGSimplificationPass(pass);
-   LLVMRunPassManager(pass, module);
-#else
-   (void)pass;
-#endif
-
-   if(verbose >= 2)
-      LLVMDumpModule(module);
-
    code = LLVMGetPointerToGlobal(engine, func);
    blend_test_ptr = voidptr_to_blend_test_ptr_t(code);
 
@@ -715,9 +696,6 @@ test_one(unsigned verbose,
 
    LLVMFreeMachineCodeForFunction(engine, func);
 
-   if(pass)
-      LLVMDisposePassManager(pass);
-
    return success;
 }
 
@@ -773,7 +751,7 @@ const unsigned num_types = sizeof(blend_types)/sizeof(blend_types[0]);
 
 
 boolean
-test_all(unsigned verbose, FILE *fp)
+test_all(struct gallivm_state *gallivm, unsigned verbose, FILE *fp)
 {
    const unsigned *rgb_func;
    const unsigned *rgb_src_factor;
@@ -809,7 +787,7 @@ test_all(unsigned verbose, FILE *fp)
                            blend.rt[0].alpha_dst_factor  = *alpha_dst_factor;
                            blend.rt[0].colormask         = PIPE_MASK_RGBA;
 
-                           if(!test_one(verbose, fp, &blend, mode, *type))
+                           if(!test_one(gallivm, verbose, fp, &blend, mode, *type))
                              success = FALSE;
 
                         }
@@ -826,7 +804,8 @@ test_all(unsigned verbose, FILE *fp)
 
 
 boolean
-test_some(unsigned verbose, FILE *fp, unsigned long n)
+test_some(struct gallivm_state *gallivm, unsigned verbose, FILE *fp,
+          unsigned long n)
 {
    const unsigned *rgb_func;
    const unsigned *rgb_src_factor;
@@ -868,7 +847,7 @@ test_some(unsigned verbose, FILE *fp, unsigned long n)
       blend.rt[0].alpha_dst_factor  = *alpha_dst_factor;
       blend.rt[0].colormask         = PIPE_MASK_RGBA;
 
-      if(!test_one(verbose, fp, &blend, mode, *type))
+      if(!test_one(gallivm, verbose, fp, &blend, mode, *type))
         success = FALSE;
    }
 
@@ -877,7 +856,7 @@ test_some(unsigned verbose, FILE *fp, unsigned long n)
 
 
 boolean
-test_single(unsigned verbose, FILE *fp)
+test_single(struct gallivm_state *gallivm, unsigned verbose, FILE *fp)
 {
    printf("no test_single()");
    return TRUE;
diff --git a/src/gallium/drivers/llvmpipe/lp_test_conv.c b/src/gallium/drivers/llvmpipe/lp_test_conv.c
index 3ba42bf11a..f4a2f360c7 100644
--- a/src/gallium/drivers/llvmpipe/lp_test_conv.c
+++ b/src/gallium/drivers/llvmpipe/lp_test_conv.c
@@ -97,64 +97,65 @@ dump_conv_types(FILE *fp,
 
 
 static LLVMValueRef
-add_conv_test(LLVMModuleRef module,
+add_conv_test(struct gallivm_state *gallivm,
               struct lp_type src_type, unsigned num_srcs,
               struct lp_type dst_type, unsigned num_dsts)
 {
+   LLVMModuleRef module = gallivm->module;
+   LLVMContextRef context = gallivm->context;
+   LLVMBuilderRef builder = gallivm->builder;
    LLVMTypeRef args[2];
    LLVMValueRef func;
    LLVMValueRef src_ptr;
    LLVMValueRef dst_ptr;
    LLVMBasicBlockRef block;
-   LLVMBuilderRef builder;
    LLVMValueRef src[LP_MAX_VECTOR_LENGTH];
    LLVMValueRef dst[LP_MAX_VECTOR_LENGTH];
    unsigned i;
 
-   args[0] = LLVMPointerType(lp_build_vec_type(src_type), 0);
-   args[1] = LLVMPointerType(lp_build_vec_type(dst_type), 0);
+   args[0] = LLVMPointerType(lp_build_vec_type(gallivm, src_type), 0);
+   args[1] = LLVMPointerType(lp_build_vec_type(gallivm, dst_type), 0);
 
-   func = LLVMAddFunction(module, "test", LLVMFunctionType(LLVMVoidType(), args, 2, 0));
+   func = LLVMAddFunction(module, "test",
+                          LLVMFunctionType(LLVMVoidTypeInContext(context),
+                                           args, 2, 0));
    LLVMSetFunctionCallConv(func, LLVMCCallConv);
    src_ptr = LLVMGetParam(func, 0);
    dst_ptr = LLVMGetParam(func, 1);
 
-   block = LLVMAppendBasicBlock(func, "entry");
-   builder = LLVMCreateBuilder();
+   block = LLVMAppendBasicBlockInContext(context, func, "entry");
    LLVMPositionBuilderAtEnd(builder, block);
 
    for(i = 0; i < num_srcs; ++i) {
-      LLVMValueRef index = LLVMConstInt(LLVMInt32Type(), i, 0);
+      LLVMValueRef index = LLVMConstInt(LLVMInt32TypeInContext(context), i, 0);
       LLVMValueRef ptr = LLVMBuildGEP(builder, src_ptr, &index, 1, "");
       src[i] = LLVMBuildLoad(builder, ptr, "");
    }
 
-   lp_build_conv(builder, src_type, dst_type, src, num_srcs, dst, num_dsts);
+   lp_build_conv(gallivm, src_type, dst_type, src, num_srcs, dst, num_dsts);
 
    for(i = 0; i < num_dsts; ++i) {
-      LLVMValueRef index = LLVMConstInt(LLVMInt32Type(), i, 0);
+      LLVMValueRef index = LLVMConstInt(LLVMInt32TypeInContext(context), i, 0);
       LLVMValueRef ptr = LLVMBuildGEP(builder, dst_ptr, &index, 1, "");
       LLVMBuildStore(builder, dst[i], ptr);
    }
 
    LLVMBuildRetVoid(builder);;
 
-   LLVMDisposeBuilder(builder);
    return func;
 }
 
 
 PIPE_ALIGN_STACK
 static boolean
-test_one(unsigned verbose,
+test_one(struct gallivm_state *gallivm, unsigned verbose,
          FILE *fp,
          struct lp_type src_type,
          struct lp_type dst_type)
 {
-   LLVMModuleRef module = NULL;
+   LLVMModuleRef module = gallivm->module;
+   LLVMExecutionEngineRef engine = gallivm->engine;
    LLVMValueRef func = NULL;
-   LLVMExecutionEngineRef engine = lp_build_engine;
-   LLVMPassManagerRef pass = NULL;
    char *error = NULL;
    conv_test_ptr_t conv_test_ptr;
    boolean success;
@@ -193,9 +194,7 @@ test_one(unsigned verbose,
 
    eps = MAX2(lp_const_eps(src_type), lp_const_eps(dst_type));
 
-   module = LLVMModuleCreateWithName("test");
-
-   func = add_conv_test(module, src_type, num_srcs, dst_type, num_dsts);
+   func = add_conv_test(gallivm, src_type, num_srcs, dst_type, num_dsts);
 
    if(LLVMVerifyModule(module, LLVMPrintMessageAction, &error)) {
       LLVMDumpModule(module);
@@ -203,21 +202,6 @@ test_one(unsigned verbose,
    }
    LLVMDisposeMessage(error);
 
-#if 0
-   pass = LLVMCreatePassManager();
-   LLVMAddTargetData(LLVMGetExecutionEngineTargetData(engine), pass);
-   /* These are the passes currently listed in llvm-c/Transforms/Scalar.h,
-    * but there are more on SVN. */
-   LLVMAddConstantPropagationPass(pass);
-   LLVMAddInstructionCombiningPass(pass);
-   LLVMAddPromoteMemoryToRegisterPass(pass);
-   LLVMAddGVNPass(pass);
-   LLVMAddCFGSimplificationPass(pass);
-   LLVMRunPassManager(pass, module);
-#else
-   (void)pass;
-#endif
-
    if(verbose >= 2)
       LLVMDumpModule(module);
 
@@ -342,9 +326,6 @@ test_one(unsigned verbose,
 
    LLVMFreeMachineCodeForFunction(engine, func);
 
-   if(pass)
-      LLVMDisposePassManager(pass);
-
    return success;
 }
 
@@ -390,7 +371,7 @@ const unsigned num_types = sizeof(conv_types)/sizeof(conv_types[0]);
 
 
 boolean
-test_all(unsigned verbose, FILE *fp)
+test_all(struct gallivm_state *gallivm, unsigned verbose, FILE *fp)
 {
    const struct lp_type *src_type;
    const struct lp_type *dst_type;
@@ -405,7 +386,7 @@ test_all(unsigned verbose, FILE *fp)
          if(src_type->norm != dst_type->norm)
             continue;
 
-         if(!test_one(verbose, fp, *src_type, *dst_type))
+         if(!test_one(gallivm, verbose, fp, *src_type, *dst_type))
            success = FALSE;
 
       }
@@ -416,7 +397,8 @@ test_all(unsigned verbose, FILE *fp)
 
 
 boolean
-test_some(unsigned verbose, FILE *fp, unsigned long n)
+test_some(struct gallivm_state *gallivm, unsigned verbose, FILE *fp,
+          unsigned long n)
 {
    const struct lp_type *src_type;
    const struct lp_type *dst_type;
@@ -430,7 +412,7 @@ test_some(unsigned verbose, FILE *fp, unsigned long n)
          dst_type = &conv_types[rand() % num_types];
       } while (src_type == dst_type || src_type->norm != dst_type->norm);
 
-      if(!test_one(verbose, fp, *src_type, *dst_type))
+      if(!test_one(gallivm, verbose, fp, *src_type, *dst_type))
         success = FALSE;
    }
 
@@ -439,7 +421,7 @@ test_some(unsigned verbose, FILE *fp, unsigned long n)
 
 
 boolean
-test_single(unsigned verbose, FILE *fp)
+test_single(struct gallivm_state *gallivm, unsigned verbose, FILE *fp)
 {
    /*    float, fixed,  sign,  norm, width, len */
    struct lp_type f32x4_type =
@@ -449,7 +431,7 @@ test_single(unsigned verbose, FILE *fp)
 
    boolean success;
 
-   success = test_one(verbose, fp, f32x4_type, ub8x4_type);
+   success = test_one(gallivm, verbose, fp, f32x4_type, ub8x4_type);
 
    return success;
 }
diff --git a/src/gallium/drivers/llvmpipe/lp_test_format.c b/src/gallium/drivers/llvmpipe/lp_test_format.c
index 2855d7cea4..4152ca6cf6 100644
--- a/src/gallium/drivers/llvmpipe/lp_test_format.c
+++ b/src/gallium/drivers/llvmpipe/lp_test_format.c
@@ -30,13 +30,6 @@
 #include <stdio.h>
 #include <float.h>
 
-#include "gallivm/lp_bld.h"
-#include "gallivm/lp_bld_debug.h"
-#include "gallivm/lp_bld_init.h"
-#include <llvm-c/Analysis.h>
-#include <llvm-c/Target.h>
-#include <llvm-c/Transforms/Scalar.h>
-
 #include "util/u_memory.h"
 #include "util/u_pointer.h"
 #include "util/u_string.h"
@@ -44,7 +37,11 @@
 #include "util/u_format_tests.h"
 #include "util/u_format_s3tc.h"
 
+#include "gallivm/lp_bld.h"
+#include "gallivm/lp_bld_debug.h"
 #include "gallivm/lp_bld_format.h"
+#include "gallivm/lp_bld_init.h"
+
 #include "lp_test.h"
 
 
@@ -78,56 +75,57 @@ typedef void
 
 
 static LLVMValueRef
-add_fetch_rgba_test(unsigned verbose,
+add_fetch_rgba_test(struct gallivm_state *gallivm, unsigned verbose,
                     const struct util_format_description *desc,
                     struct lp_type type)
 {
    char name[256];
+   LLVMContextRef context = gallivm->context;
+   LLVMModuleRef module = gallivm->module;
+   LLVMBuilderRef builder = gallivm->builder;
+   LLVMPassManagerRef passmgr = gallivm->passmgr;
    LLVMTypeRef args[4];
    LLVMValueRef func;
    LLVMValueRef packed_ptr;
-   LLVMValueRef offset = LLVMConstNull(LLVMInt32Type());
+   LLVMValueRef offset = LLVMConstNull(LLVMInt32TypeInContext(context));
    LLVMValueRef rgba_ptr;
    LLVMValueRef i;
    LLVMValueRef j;
    LLVMBasicBlockRef block;
-   LLVMBuilderRef builder;
    LLVMValueRef rgba;
 
    util_snprintf(name, sizeof name, "fetch_%s_%s", desc->short_name,
                  type.floating ? "float" : "unorm8");
 
-   args[0] = LLVMPointerType(lp_build_vec_type(type), 0);
-   args[1] = LLVMPointerType(LLVMInt8Type(), 0);
-   args[3] = args[2] = LLVMInt32Type();
+   args[0] = LLVMPointerType(lp_build_vec_type(gallivm, type), 0);
+   args[1] = LLVMPointerType(LLVMInt8TypeInContext(context), 0);
+   args[3] = args[2] = LLVMInt32TypeInContext(context);
 
-   func = LLVMAddFunction(lp_build_module, name,
-                          LLVMFunctionType(LLVMVoidType(), args, Elements(args), 0));
+   func = LLVMAddFunction(module, name,
+                          LLVMFunctionType(LLVMVoidTypeInContext(context),
+                                           args, Elements(args), 0));
    LLVMSetFunctionCallConv(func, LLVMCCallConv);
    rgba_ptr = LLVMGetParam(func, 0);
    packed_ptr = LLVMGetParam(func, 1);
    i = LLVMGetParam(func, 2);
    j = LLVMGetParam(func, 3);
 
-   block = LLVMAppendBasicBlock(func, "entry");
-   builder = LLVMCreateBuilder();
+   block = LLVMAppendBasicBlockInContext(context, func, "entry");
    LLVMPositionBuilderAtEnd(builder, block);
 
-   rgba = lp_build_fetch_rgba_aos(builder, desc, type,
+   rgba = lp_build_fetch_rgba_aos(gallivm, desc, type,
                                   packed_ptr, offset, i, j);
 
    LLVMBuildStore(builder, rgba, rgba_ptr);
 
    LLVMBuildRetVoid(builder);
 
-   LLVMDisposeBuilder(builder);
-
    if (LLVMVerifyFunction(func, LLVMPrintMessageAction)) {
       LLVMDumpValue(func);
       abort();
    }
 
-   LLVMRunFunctionPassManager(lp_build_pass, func);
+   LLVMRunFunctionPassManager(passmgr, func);
 
    if (verbose >= 1) {
       LLVMDumpValue(func);
@@ -139,10 +137,11 @@ add_fetch_rgba_test(unsigned verbose,
 
 PIPE_ALIGN_STACK
 static boolean
-test_format_float(unsigned verbose, FILE *fp,
+test_format_float(struct gallivm_state *gallivm, unsigned verbose, FILE *fp,
                   const struct util_format_description *desc)
 {
    LLVMValueRef fetch = NULL;
+   LLVMExecutionEngineRef engine = gallivm->engine;
    fetch_ptr_t fetch_ptr;
    PIPE_ALIGN_VAR(16) float unpacked[4];
    boolean first = TRUE;
@@ -150,9 +149,9 @@ test_format_float(unsigned verbose, FILE *fp,
    unsigned i, j, k, l;
    void *f;
 
-   fetch = add_fetch_rgba_test(verbose, desc, lp_float32_vec4_type());
+   fetch = add_fetch_rgba_test(gallivm, verbose, desc, lp_float32_vec4_type());
 
-   f = LLVMGetPointerToGlobal(lp_build_engine, fetch);
+   f = LLVMGetPointerToGlobal(engine, fetch);
    fetch_ptr = (fetch_ptr_t) pointer_to_func(f);
 
    if (verbose >= 2) {
@@ -208,7 +207,7 @@ test_format_float(unsigned verbose, FILE *fp,
       }
    }
 
-   LLVMFreeMachineCodeForFunction(lp_build_engine, fetch);
+   LLVMFreeMachineCodeForFunction(engine, fetch);
    LLVMDeleteFunction(fetch);
 
    if(fp)
@@ -220,7 +219,8 @@ test_format_float(unsigned verbose, FILE *fp,
 
 PIPE_ALIGN_STACK
 static boolean
-test_format_unorm8(unsigned verbose, FILE *fp,
+test_format_unorm8(struct gallivm_state *gallivm,
+                   unsigned verbose, FILE *fp,
                    const struct util_format_description *desc)
 {
    LLVMValueRef fetch = NULL;
@@ -231,9 +231,9 @@ test_format_unorm8(unsigned verbose, FILE *fp,
    unsigned i, j, k, l;
    void *f;
 
-   fetch = add_fetch_rgba_test(verbose, desc, lp_unorm8_vec4_type());
+   fetch = add_fetch_rgba_test(gallivm, verbose, desc, lp_unorm8_vec4_type());
 
-   f = LLVMGetPointerToGlobal(lp_build_engine, fetch);
+   f = LLVMGetPointerToGlobal(gallivm->engine, fetch);
    fetch_ptr = (fetch_ptr_t) pointer_to_func(f);
 
    if (verbose >= 2) {
@@ -290,7 +290,7 @@ test_format_unorm8(unsigned verbose, FILE *fp,
    if (!success)
       LLVMDumpValue(fetch);
 
-   LLVMFreeMachineCodeForFunction(lp_build_engine, fetch);
+   LLVMFreeMachineCodeForFunction(gallivm->engine, fetch);
    LLVMDeleteFunction(fetch);
 
    if(fp)
@@ -303,16 +303,17 @@ test_format_unorm8(unsigned verbose, FILE *fp,
 
 
 static boolean
-test_one(unsigned verbose, FILE *fp,
+test_one(struct gallivm_state *gallivm,
+         unsigned verbose, FILE *fp,
          const struct util_format_description *format_desc)
 {
    boolean success = TRUE;
 
-   if (!test_format_float(verbose, fp, format_desc)) {
+   if (!test_format_float(gallivm, verbose, fp, format_desc)) {
      success = FALSE;
    }
 
-   if (!test_format_unorm8(verbose, fp, format_desc)) {
+   if (!test_format_unorm8(gallivm, verbose, fp, format_desc)) {
      success = FALSE;
    }
 
@@ -321,7 +322,7 @@ test_one(unsigned verbose, FILE *fp,
 
 
 boolean
-test_all(unsigned verbose, FILE *fp)
+test_all(struct gallivm_state *gallivm, unsigned verbose, FILE *fp)
 {
    enum pipe_format format;
    boolean success = TRUE;
@@ -349,7 +350,7 @@ test_all(unsigned verbose, FILE *fp)
          continue;
       }
 
-      if (!test_one(verbose, fp, format_desc)) {
+      if (!test_one(gallivm, verbose, fp, format_desc)) {
            success = FALSE;
       }
    }
@@ -359,14 +360,15 @@ test_all(unsigned verbose, FILE *fp)
 
 
 boolean
-test_some(unsigned verbose, FILE *fp, unsigned long n)
+test_some(struct gallivm_state *gallivm, unsigned verbose, FILE *fp,
+          unsigned long n)
 {
-   return test_all(verbose, fp);
+   return test_all(gallivm, verbose, fp);
 }
 
 
 boolean
-test_single(unsigned verbose, FILE *fp)
+test_single(struct gallivm_state *gallivm, unsigned verbose, FILE *fp)
 {
    printf("no test_single()");
    return TRUE;
diff --git a/src/gallium/drivers/llvmpipe/lp_test_main.c b/src/gallium/drivers/llvmpipe/lp_test_main.c
index 7a0d06ae2c..149ee6f125 100644
--- a/src/gallium/drivers/llvmpipe/lp_test_main.c
+++ b/src/gallium/drivers/llvmpipe/lp_test_main.c
@@ -380,6 +380,7 @@ int main(int argc, char **argv)
    unsigned i;
    boolean success;
    boolean single = FALSE;
+   struct gallivm_state *gallivm;
 
    for(i = 1; i < argc; ++i) {
       if(strcmp(argv[i], "-v") == 0)
@@ -394,21 +395,23 @@ int main(int argc, char **argv)
 
    lp_build_init();
 
+   gallivm = gallivm_create();
+
    util_cpu_detect();
 
    if(fp) {
       /* Warm up the caches */
-      test_some(0, NULL, 100);
+      test_some(gallivm, 0, NULL, 100);
 
       write_tsv_header(fp);
    }
       
    if (single)
-      success = test_single(verbose, fp);
+      success = test_single(gallivm, verbose, fp);
    else if (n)
-      success = test_some(verbose, fp, n);
+      success = test_some(gallivm, verbose, fp, n);
    else
-      success = test_all(verbose, fp);
+      success = test_all(gallivm, verbose, fp);
 
    if(fp)
       fclose(fp);
diff --git a/src/gallium/drivers/llvmpipe/lp_test_printf.c b/src/gallium/drivers/llvmpipe/lp_test_printf.c
index 4653f30e39..620cdb57c1 100644
--- a/src/gallium/drivers/llvmpipe/lp_test_printf.c
+++ b/src/gallium/drivers/llvmpipe/lp_test_printf.c
@@ -35,11 +35,6 @@
 #include "gallivm/lp_bld_assert.h"
 #include "gallivm/lp_bld_printf.h"
 
-#include <llvm-c/Analysis.h>
-#include <llvm-c/ExecutionEngine.h>
-#include <llvm-c/Target.h>
-#include <llvm-c/Transforms/Scalar.h>
-
 #include "lp_test.h"
 
 
@@ -63,48 +58,45 @@ typedef void (*test_printf_t)(int i);
 
 
 static LLVMValueRef
-add_printf_test(LLVMModuleRef module)
+add_printf_test(struct gallivm_state *gallivm)
 {
-   LLVMTypeRef args[1] = { LLVMIntType(32) };
-   LLVMValueRef func = LLVMAddFunction(module, "test_printf", LLVMFunctionType(LLVMVoidType(), args, 1, 0));
-   LLVMBuilderRef builder = LLVMCreateBuilder();
-   LLVMBasicBlockRef block = LLVMAppendBasicBlock(func, "entry");
+   LLVMModuleRef module = gallivm->module;
+   LLVMTypeRef args[1] = { LLVMIntTypeInContext(gallivm->context, 32) };
+   LLVMValueRef func = LLVMAddFunction(module, "test_printf", LLVMFunctionType(LLVMVoidTypeInContext(gallivm->context), args, 1, 0));
+   LLVMBuilderRef builder = gallivm->builder;
+   LLVMBasicBlockRef block = LLVMAppendBasicBlockInContext(gallivm->context, func, "entry");
 
    LLVMSetFunctionCallConv(func, LLVMCCallConv);
 
    LLVMPositionBuilderAtEnd(builder, block);
-   lp_build_printf(builder, "hello, world\n");
-   lp_build_printf(builder, "print 5 6: %d %d\n", LLVMConstInt(LLVMInt32Type(), 5, 0),
-				LLVMConstInt(LLVMInt32Type(), 6, 0));
+   lp_build_printf(gallivm, "hello, world\n");
+   lp_build_printf(gallivm, "print 5 6: %d %d\n", LLVMConstInt(LLVMInt32TypeInContext(gallivm->context), 5, 0),
+				LLVMConstInt(LLVMInt32TypeInContext(gallivm->context), 6, 0));
 
    /* Also test lp_build_assert().  This should not fail. */
-   lp_build_assert(builder, LLVMConstInt(LLVMInt32Type(), 1, 0), "assert(1)");
+   lp_build_assert(gallivm, LLVMConstInt(LLVMInt32TypeInContext(gallivm->context), 1, 0), "assert(1)");
 
    LLVMBuildRetVoid(builder);
-   LLVMDisposeBuilder(builder);
+
    return func;
 }
 
 
 PIPE_ALIGN_STACK
 static boolean
-test_printf(unsigned verbose, FILE *fp, const struct printf_test_case *testcase)
+test_printf(struct gallivm_state *gallivm,
+            unsigned verbose, FILE *fp,
+            const struct printf_test_case *testcase)
 {
-   LLVMModuleRef module = NULL;
-   LLVMValueRef test = NULL;
-   LLVMExecutionEngineRef engine = NULL;
-   LLVMModuleProviderRef provider = NULL;
-   LLVMPassManagerRef pass = NULL;
+   LLVMExecutionEngineRef engine = gallivm->engine;
+   LLVMModuleRef module = gallivm->module;
+   LLVMValueRef test;
    char *error = NULL;
-   test_printf_t test_printf;
-   float unpacked[4];
-   unsigned packed;
+   test_printf_t test_printf_func;
    boolean success = TRUE;
    void *code;
 
-   module = LLVMModuleCreateWithName("test");
-
-   test = add_printf_test(module);
+   test = add_printf_test(gallivm);
 
    if(LLVMVerifyModule(module, LLVMPrintMessageAction, &error)) {
       LLVMDumpModule(module);
@@ -112,74 +104,40 @@ test_printf(unsigned verbose, FILE *fp, const struct printf_test_case *testcase)
    }
    LLVMDisposeMessage(error);
 
-   provider = LLVMCreateModuleProviderForExistingModule(module);
-#if 0
-   if (LLVMCreateJITCompiler(&engine, provider, 1, &error)) {
-      fprintf(stderr, "%s\n", error);
-      LLVMDisposeMessage(error);
-      abort();
-   }
-#else
-   (void) provider;
-   engine = lp_build_engine;
-#endif
-
-#if 0
-   pass = LLVMCreatePassManager();
-   LLVMAddTargetData(LLVMGetExecutionEngineTargetData(engine), pass);
-   /* These are the passes currently listed in llvm-c/Transforms/Scalar.h,
-    * but there are more on SVN. */
-   LLVMAddConstantPropagationPass(pass);
-   LLVMAddInstructionCombiningPass(pass);
-   LLVMAddPromoteMemoryToRegisterPass(pass);
-   LLVMAddGVNPass(pass);
-   LLVMAddCFGSimplificationPass(pass);
-   LLVMRunPassManager(pass, module);
-#else
-   (void)pass;
-#endif
-
    code = LLVMGetPointerToGlobal(engine, test);
-   test_printf = (test_printf_t)pointer_to_func(code);
-
-   memset(unpacked, 0, sizeof unpacked);
-   packed = 0;
-
+   test_printf_func = (test_printf_t) pointer_to_func(code);
 
    // LLVMDumpModule(module);
 
-   test_printf(0);
+   test_printf_func(0);
 
    LLVMFreeMachineCodeForFunction(engine, test);
 
-   LLVMDisposeExecutionEngine(engine);
-   if(pass)
-      LLVMDisposePassManager(pass);
-
    return success;
 }
 
 
 boolean
-test_all(unsigned verbose, FILE *fp)
+test_all(struct gallivm_state *gallivm, unsigned verbose, FILE *fp)
 {
    boolean success = TRUE;
 
-   test_printf(verbose, fp, NULL);
+   test_printf(gallivm, verbose, fp, NULL);
 
    return success;
 }
 
 
 boolean
-test_some(unsigned verbose, FILE *fp, unsigned long n)
+test_some(struct gallivm_state *gallivm, unsigned verbose, FILE *fp,
+          unsigned long n)
 {
-   return test_all(verbose, fp);
+   return test_all(gallivm, verbose, fp);
 }
 
 
 boolean
-test_single(unsigned verbose, FILE *fp)
+test_single(struct gallivm_state *gallivm, unsigned verbose, FILE *fp)
 {
    printf("no test_single()");
    return TRUE;
diff --git a/src/gallium/drivers/llvmpipe/lp_test_round.c b/src/gallium/drivers/llvmpipe/lp_test_round.c
index 816518e508..4edee4af12 100644
--- a/src/gallium/drivers/llvmpipe/lp_test_round.c
+++ b/src/gallium/drivers/llvmpipe/lp_test_round.c
@@ -34,11 +34,6 @@
 #include "gallivm/lp_bld_init.h"
 #include "gallivm/lp_bld_arit.h"
 
-#include <llvm-c/Analysis.h>
-#include <llvm-c/ExecutionEngine.h>
-#include <llvm-c/Target.h>
-#include <llvm-c/Transforms/Scalar.h>
-
 #include "lp_test.h"
 
 
@@ -64,18 +59,21 @@ typedef LLVMValueRef (*lp_func_t)(struct lp_build_context *, LLVMValueRef);
 
 
 static LLVMValueRef
-add_test(LLVMModuleRef module, const char *name, lp_func_t lp_func)
+add_test(struct gallivm_state *gallivm, const char *name, lp_func_t lp_func)
 {
-   LLVMTypeRef v4sf = LLVMVectorType(LLVMFloatType(), 4);
+   LLVMModuleRef module = gallivm->module;
+   LLVMContextRef context = gallivm->context;
+   LLVMBuilderRef builder = gallivm->builder;
+
+   LLVMTypeRef v4sf = LLVMVectorType(LLVMFloatTypeInContext(context), 4);
    LLVMTypeRef args[1] = { v4sf };
    LLVMValueRef func = LLVMAddFunction(module, name, LLVMFunctionType(v4sf, args, 1, 0));
    LLVMValueRef arg1 = LLVMGetParam(func, 0);
-   LLVMBuilderRef builder = LLVMCreateBuilder();
-   LLVMBasicBlockRef block = LLVMAppendBasicBlock(func, "entry");
+   LLVMBasicBlockRef block = LLVMAppendBasicBlockInContext(context, func, "entry");
    LLVMValueRef ret;
    struct lp_build_context bld;
 
-   lp_build_context_init(&bld, builder, lp_float32_vec4_type());
+   lp_build_context_init(&bld, gallivm, lp_float32_vec4_type());
 
    LLVMSetFunctionCallConv(func, LLVMCCallConv);
 
@@ -84,7 +82,7 @@ add_test(LLVMModuleRef module, const char *name, lp_func_t lp_func)
    ret = lp_func(&bld, arg1);
 
    LLVMBuildRet(builder, ret);
-   LLVMDisposeBuilder(builder);
+
    return func;
 }
 
@@ -117,12 +115,11 @@ compare(v4sf x, v4sf y)
 
 PIPE_ALIGN_STACK
 static boolean
-test_round(unsigned verbose, FILE *fp)
+test_round(struct gallivm_state *gallivm, unsigned verbose, FILE *fp)
 {
-   LLVMModuleRef module = NULL;
+   LLVMModuleRef module = gallivm->module;
    LLVMValueRef test_round = NULL, test_trunc, test_floor, test_ceil;
-   LLVMExecutionEngineRef engine = lp_build_engine;
-   LLVMPassManagerRef pass = NULL;
+   LLVMExecutionEngineRef engine = gallivm->engine;
    char *error = NULL;
    test_round_t round_func, trunc_func, floor_func, ceil_func;
    float unpacked[4];
@@ -130,12 +127,10 @@ test_round(unsigned verbose, FILE *fp)
    boolean success = TRUE;
    int i;
 
-   module = LLVMModuleCreateWithName("test");
-
-   test_round = add_test(module, "round", lp_build_round);
-   test_trunc = add_test(module, "trunc", lp_build_trunc);
-   test_floor = add_test(module, "floor", lp_build_floor);
-   test_ceil = add_test(module, "ceil", lp_build_ceil);
+   test_round = add_test(gallivm, "round", lp_build_round);
+   test_trunc = add_test(gallivm, "trunc", lp_build_trunc);
+   test_floor = add_test(gallivm, "floor", lp_build_floor);
+   test_ceil = add_test(gallivm, "ceil", lp_build_ceil);
 
    if(LLVMVerifyModule(module, LLVMPrintMessageAction, &error)) {
       printf("LLVMVerifyModule: %s\n", error);
@@ -144,21 +139,6 @@ test_round(unsigned verbose, FILE *fp)
    }
    LLVMDisposeMessage(error);
 
-#if 0
-   pass = LLVMCreatePassManager();
-   LLVMAddTargetData(LLVMGetExecutionEngineTargetData(engine), pass);
-   /* These are the passes currently listed in llvm-c/Transforms/Scalar.h,
-    * but there are more on SVN. */
-   LLVMAddConstantPropagationPass(pass);
-   LLVMAddInstructionCombiningPass(pass);
-   LLVMAddPromoteMemoryToRegisterPass(pass);
-   LLVMAddGVNPass(pass);
-   LLVMAddCFGSimplificationPass(pass);
-   LLVMRunPassManager(pass, module);
-#else
-   (void)pass;
-#endif
-
    round_func = (test_round_t) pointer_to_func(LLVMGetPointerToGlobal(engine, test_round));
    trunc_func = (test_round_t) pointer_to_func(LLVMGetPointerToGlobal(engine, test_trunc));
    floor_func = (test_round_t) pointer_to_func(LLVMGetPointerToGlobal(engine, test_floor));
@@ -229,17 +209,13 @@ test_round(unsigned verbose, FILE *fp)
    LLVMFreeMachineCodeForFunction(engine, test_floor);
    LLVMFreeMachineCodeForFunction(engine, test_ceil);
 
-   LLVMDisposeExecutionEngine(engine);
-   if(pass)
-      LLVMDisposePassManager(pass);
-
    return success;
 }
 
 #else /* !PIPE_ARCH_SSE */
 
 static boolean
-test_round(unsigned verbose, FILE *fp)
+test_round(struct gallivm_state *gallivm, unsigned verbose, FILE *fp)
 {
    return TRUE;
 }
@@ -248,20 +224,21 @@ test_round(unsigned verbose, FILE *fp)
 
 
 boolean
-test_all(unsigned verbose, FILE *fp)
+test_all(struct gallivm_state *gallivm, unsigned verbose, FILE *fp)
 {
-   return test_round(verbose, fp);
+   return test_round(gallivm, verbose, fp);
 }
 
 
 boolean
-test_some(unsigned verbose, FILE *fp, unsigned long n)
+test_some(struct gallivm_state *gallivm, unsigned verbose, FILE *fp,
+          unsigned long n)
 {
-   return test_all(verbose, fp);
+   return test_all(gallivm, verbose, fp);
 }
 
 boolean
-test_single(unsigned verbose, FILE *fp)
+test_single(struct gallivm_state *gallivm, unsigned verbose, FILE *fp)
 {
    printf("no test_single()");
    return TRUE;
diff --git a/src/gallium/drivers/llvmpipe/lp_test_sincos.c b/src/gallium/drivers/llvmpipe/lp_test_sincos.c
index 79939b1a39..066d633d44 100644
--- a/src/gallium/drivers/llvmpipe/lp_test_sincos.c
+++ b/src/gallium/drivers/llvmpipe/lp_test_sincos.c
@@ -29,15 +29,11 @@
 #include <stdlib.h>
 #include <stdio.h>
 
+#include "util/u_pointer.h"
+
 #include "gallivm/lp_bld.h"
 #include "gallivm/lp_bld_init.h"
 #include "gallivm/lp_bld_arit.h"
-#include "util/u_pointer.h"
-
-#include <llvm-c/Analysis.h>
-#include <llvm-c/ExecutionEngine.h>
-#include <llvm-c/Target.h>
-#include <llvm-c/Transforms/Scalar.h>
 
 #include "lp_test.h"
 
@@ -61,25 +57,25 @@ write_tsv_header(FILE *fp)
 typedef __m128 (*test_sincos_t)(__m128);
 
 static LLVMValueRef
-add_sincos_test(LLVMModuleRef module, boolean sin)
+add_sincos_test(struct gallivm_state *gallivm, LLVMModuleRef module,
+                LLVMContextRef context, boolean sin)
 {
-   LLVMTypeRef v4sf = LLVMVectorType(LLVMFloatType(), 4);
+   LLVMTypeRef v4sf = LLVMVectorType(LLVMFloatTypeInContext(context), 4);
    LLVMTypeRef args[1] = { v4sf };
    LLVMValueRef func = LLVMAddFunction(module, "sincos", LLVMFunctionType(v4sf, args, 1, 0));
    LLVMValueRef arg1 = LLVMGetParam(func, 0);
-   LLVMBuilderRef builder = LLVMCreateBuilder();
-   LLVMBasicBlockRef block = LLVMAppendBasicBlock(func, "entry");
+   LLVMBuilderRef builder = gallivm->builder;
+   LLVMBasicBlockRef block = LLVMAppendBasicBlockInContext(context, func, "entry");
    LLVMValueRef ret;
    struct lp_build_context bld;
 
-   lp_build_context_init(&bld, builder, lp_float32_vec4_type());
+   lp_build_context_init(&bld, gallivm, lp_float32_vec4_type());
 
    LLVMSetFunctionCallConv(func, LLVMCCallConv);
 
    LLVMPositionBuilderAtEnd(builder, block);
    ret = sin ? lp_build_sin(&bld, arg1) : lp_build_cos(&bld, arg1);
    LLVMBuildRet(builder, ret);
-   LLVMDisposeBuilder(builder);
    return func;
 }
 
@@ -95,22 +91,20 @@ printv(char* string, v4sf value)
 
 PIPE_ALIGN_STACK
 static boolean
-test_sincos(unsigned verbose, FILE *fp)
+test_sincos(struct gallivm_state *gallivm, unsigned verbose, FILE *fp)
 {
-   LLVMModuleRef module = NULL;
+   LLVMModuleRef module = gallivm->module;
    LLVMValueRef test_sin = NULL, test_cos = NULL;
-   LLVMExecutionEngineRef engine = lp_build_engine;
-   LLVMPassManagerRef pass = NULL;
+   LLVMExecutionEngineRef engine = gallivm->engine;
+   LLVMContextRef context = gallivm->context;
    char *error = NULL;
    test_sincos_t sin_func;
    test_sincos_t cos_func;
    float unpacked[4];
    boolean success = TRUE;
 
-   module = LLVMModuleCreateWithName("test");
-
-   test_sin = add_sincos_test(module, TRUE);
-   test_cos = add_sincos_test(module, FALSE);
+   test_sin = add_sincos_test(gallivm, module, context, TRUE);
+   test_cos = add_sincos_test(gallivm, module, context,FALSE);
 
    if(LLVMVerifyModule(module, LLVMPrintMessageAction, &error)) {
       printf("LLVMVerifyModule: %s\n", error);
@@ -119,21 +113,6 @@ test_sincos(unsigned verbose, FILE *fp)
    }
    LLVMDisposeMessage(error);
 
-#if 0
-   pass = LLVMCreatePassManager();
-   LLVMAddTargetData(LLVMGetExecutionEngineTargetData(engine), pass);
-   /* These are the passes currently listed in llvm-c/Transforms/Scalar.h,
-    * but there are more on SVN. */
-   LLVMAddConstantPropagationPass(pass);
-   LLVMAddInstructionCombiningPass(pass);
-   LLVMAddPromoteMemoryToRegisterPass(pass);
-   LLVMAddGVNPass(pass);
-   LLVMAddCFGSimplificationPass(pass);
-   LLVMRunPassManager(pass, module);
-#else
-   (void)pass;
-#endif
-
    sin_func = (test_sincos_t) pointer_to_func(LLVMGetPointerToGlobal(engine, test_sin));
    cos_func = (test_sincos_t) pointer_to_func(LLVMGetPointerToGlobal(engine, test_cos));
 
@@ -152,16 +131,13 @@ test_sincos(unsigned verbose, FILE *fp)
    LLVMFreeMachineCodeForFunction(engine, test_sin);
    LLVMFreeMachineCodeForFunction(engine, test_cos);
 
-   if(pass)
-      LLVMDisposePassManager(pass);
-
    return success;
 }
 
 #else /* !PIPE_ARCH_SSE */
 
 static boolean
-test_sincos(unsigned verbose, FILE *fp)
+test_sincos(struct gallivm_state *gallivm, unsigned verbose, FILE *fp)
 {
    return TRUE;
 }
@@ -170,24 +146,25 @@ test_sincos(unsigned verbose, FILE *fp)
 
 
 boolean
-test_all(unsigned verbose, FILE *fp)
+test_all(struct gallivm_state *gallivm, unsigned verbose, FILE *fp)
 {
    boolean success = TRUE;
 
-   test_sincos(verbose, fp);
+   test_sincos(gallivm, verbose, fp);
 
    return success;
 }
 
 
 boolean
-test_some(unsigned verbose, FILE *fp, unsigned long n)
+test_some(struct gallivm_state *gallivm, unsigned verbose, FILE *fp,
+          unsigned long n)
 {
-   return test_all(verbose, fp);
+   return test_all(gallivm, verbose, fp);
 }
 
 boolean
-test_single(unsigned verbose, FILE *fp)
+test_single(struct gallivm_state *gallivm, unsigned verbose, FILE *fp)
 {
    printf("no test_single()");
    return TRUE;
diff --git a/src/gallium/drivers/llvmpipe/lp_tex_sample.c b/src/gallium/drivers/llvmpipe/lp_tex_sample.c
index f417fc8a9e..ed4282937f 100644
--- a/src/gallium/drivers/llvmpipe/lp_tex_sample.c
+++ b/src/gallium/drivers/llvmpipe/lp_tex_sample.c
@@ -43,6 +43,7 @@
 #include "pipe/p_defines.h"
 #include "pipe/p_shader_tokens.h"
 #include "gallivm/lp_bld_debug.h"
+#include "gallivm/lp_bld_const.h"
 #include "gallivm/lp_bld_type.h"
 #include "gallivm/lp_bld_sample.h"
 #include "gallivm/lp_bld_tgsi.h"
@@ -89,7 +90,7 @@ struct lp_llvm_sampler_soa
  */
 static LLVMValueRef
 lp_llvm_texture_member(const struct lp_sampler_dynamic_state *base,
-                       LLVMBuilderRef builder,
+                       struct gallivm_state *gallivm,
                        unsigned unit,
                        unsigned member_index,
                        const char *member_name,
@@ -97,6 +98,7 @@ lp_llvm_texture_member(const struct lp_sampler_dynamic_state *base,
 {
    struct llvmpipe_sampler_dynamic_state *state =
       (struct llvmpipe_sampler_dynamic_state *)base;
+   LLVMBuilderRef builder = gallivm->builder;
    LLVMValueRef indices[4];
    LLVMValueRef ptr;
    LLVMValueRef res;
@@ -104,13 +106,13 @@ lp_llvm_texture_member(const struct lp_sampler_dynamic_state *base,
    assert(unit < PIPE_MAX_SAMPLERS);
 
    /* context[0] */
-   indices[0] = LLVMConstInt(LLVMInt32Type(), 0, 0);
+   indices[0] = lp_build_const_int32(gallivm, 0);
    /* context[0].textures */
-   indices[1] = LLVMConstInt(LLVMInt32Type(), LP_JIT_CTX_TEXTURES, 0);
+   indices[1] = lp_build_const_int32(gallivm, LP_JIT_CTX_TEXTURES);
    /* context[0].textures[unit] */
-   indices[2] = LLVMConstInt(LLVMInt32Type(), unit, 0);
+   indices[2] = lp_build_const_int32(gallivm, unit);
    /* context[0].textures[unit].member */
-   indices[3] = LLVMConstInt(LLVMInt32Type(), member_index, 0);
+   indices[3] = lp_build_const_int32(gallivm, member_index);
 
    ptr = LLVMBuildGEP(builder, state->context_ptr, indices, Elements(indices), "");
 
@@ -137,10 +139,10 @@ lp_llvm_texture_member(const struct lp_sampler_dynamic_state *base,
 #define LP_LLVM_TEXTURE_MEMBER(_name, _index, _emit_load)  \
    static LLVMValueRef \
    lp_llvm_texture_##_name( const struct lp_sampler_dynamic_state *base, \
-                            LLVMBuilderRef builder, \
+                            struct gallivm_state *gallivm, \
                             unsigned unit) \
    { \
-      return lp_llvm_texture_member(base, builder, unit, _index, #_name, _emit_load ); \
+      return lp_llvm_texture_member(base, gallivm, unit, _index, #_name, _emit_load ); \
    }
 
 
@@ -170,7 +172,7 @@ lp_llvm_sampler_soa_destroy(struct lp_build_sampler_soa *sampler)
  */
 static void
 lp_llvm_sampler_soa_emit_fetch_texel(const struct lp_build_sampler_soa *base,
-                                     LLVMBuilderRef builder,
+                                     struct gallivm_state *gallivm,
                                      struct lp_type type,
                                      unsigned unit,
                                      unsigned num_coords,
@@ -186,11 +188,11 @@ lp_llvm_sampler_soa_emit_fetch_texel(const struct lp_build_sampler_soa *base,
    assert(unit < PIPE_MAX_SAMPLERS);
    
    if (LP_PERF & PERF_NO_TEX) {
-      lp_build_sample_nop(type, texel);
+      lp_build_sample_nop(gallivm, type, texel);
       return;
    }
 
-   lp_build_sample_soa(builder,
+   lp_build_sample_soa(gallivm,
                        &sampler->dynamic_state.static_state[unit],
                        &sampler->dynamic_state.base,
                        type,
diff --git a/src/gallium/drivers/llvmpipe/lp_texture.c b/src/gallium/drivers/llvmpipe/lp_texture.c
index a4b9f2590a..9753da5e57 100644
--- a/src/gallium/drivers/llvmpipe/lp_texture.c
+++ b/src/gallium/drivers/llvmpipe/lp_texture.c
@@ -49,6 +49,7 @@
 #include "lp_tile_image.h"
 #include "lp_texture.h"
 #include "lp_setup.h"
+#include "lp_state.h"
 
 #include "state_tracker/sw_winsys.h"
 
@@ -242,6 +243,7 @@ llvmpipe_resource_create(struct pipe_screen *_screen,
       /* other data (vertex buffer, const buffer, etc) */
       const enum pipe_format format = templat->format;
       const uint w = templat->width0 / util_format_get_blockheight(format);
+      /* XXX buffers should only have one dimension, those values should be 1 */
       const uint h = templat->height0 / util_format_get_blockwidth(format);
       const uint d = templat->depth0;
       const uint bpp = util_format_get_blocksize(format);
@@ -329,17 +331,16 @@ llvmpipe_resource_destroy(struct pipe_screen *pscreen,
  */
 void *
 llvmpipe_resource_map(struct pipe_resource *resource,
-		      unsigned face,
-		      unsigned level,
-		      unsigned zslice,
+                      unsigned level,
+                      unsigned layer,
                       enum lp_texture_usage tex_usage,
                       enum lp_texture_layout layout)
 {
    struct llvmpipe_resource *lpr = llvmpipe_resource(resource);
    uint8_t *map;
 
-   assert(face < 6);
    assert(level < LP_MAX_TEXTURE_LEVELS);
+   assert(layer < (u_minify(resource->depth0, level) + resource->array_size - 1));
 
    assert(tex_usage == LP_TEX_USAGE_READ ||
           tex_usage == LP_TEX_USAGE_READ_WRITE ||
@@ -363,9 +364,8 @@ llvmpipe_resource_map(struct pipe_resource *resource,
          dt_usage = PIPE_TRANSFER_READ_WRITE;
       }
 
-      assert(face == 0);
       assert(level == 0);
-      assert(zslice == 0);
+      assert(layer == 0);
 
       /* FIXME: keep map count? */
       map = winsys->displaytarget_map(winsys, lpr->dt, dt_usage);
@@ -381,15 +381,8 @@ llvmpipe_resource_map(struct pipe_resource *resource,
       return map2;
    }
    else if (resource_is_texture(resource)) {
-      /* regular texture */
-      if (resource->target != PIPE_TEXTURE_CUBE) {
-         assert(face == 0);
-      }
-      if (resource->target != PIPE_TEXTURE_3D) {
-         assert(zslice == 0);
-      }
 
-      map = llvmpipe_get_texture_image(lpr, face + zslice, level,
+      map = llvmpipe_get_texture_image(lpr, layer, level,
                                        tex_usage, layout);
       return map;
    }
@@ -404,9 +397,8 @@ llvmpipe_resource_map(struct pipe_resource *resource,
  */
 void
 llvmpipe_resource_unmap(struct pipe_resource *resource,
-                       unsigned face,
                        unsigned level,
-                       unsigned zslice)
+                       unsigned layer)
 {
    struct llvmpipe_resource *lpr = llvmpipe_resource(resource);
 
@@ -415,12 +407,11 @@ llvmpipe_resource_unmap(struct pipe_resource *resource,
       struct llvmpipe_screen *lp_screen = llvmpipe_screen(resource->screen);
       struct sw_winsys *winsys = lp_screen->winsys;
 
-      assert(face == 0);
       assert(level == 0);
-      assert(zslice == 0);
+      assert(layer == 0);
 
       /* make sure linear image is up to date */
-      (void) llvmpipe_get_texture_image(lpr, face + zslice, level,
+      (void) llvmpipe_get_texture_image(lpr, layer, level,
                                         LP_TEX_USAGE_READ,
                                         LP_TEX_LAYOUT_LINEAR);
 
@@ -520,34 +511,35 @@ llvmpipe_resource_get_handle(struct pipe_screen *screen,
 
 
 static struct pipe_surface *
-llvmpipe_get_tex_surface(struct pipe_screen *screen,
-                         struct pipe_resource *pt,
-                         unsigned face, unsigned level, unsigned zslice,
-                         unsigned usage)
+llvmpipe_create_surface(struct pipe_context *pipe,
+                        struct pipe_resource *pt,
+                        const struct pipe_surface *surf_tmpl)
 {
    struct pipe_surface *ps;
 
-   assert(level <= pt->last_level);
+   assert(surf_tmpl->u.tex.level <= pt->last_level);
 
    ps = CALLOC_STRUCT(pipe_surface);
    if (ps) {
       pipe_reference_init(&ps->reference, 1);
       pipe_resource_reference(&ps->texture, pt);
-      ps->format = pt->format;
-      ps->width = u_minify(pt->width0, level);
-      ps->height = u_minify(pt->height0, level);
-      ps->usage = usage;
-
-      ps->face = face;
-      ps->level = level;
-      ps->zslice = zslice;
+      ps->context = pipe;
+      ps->format = surf_tmpl->format;
+      ps->width = u_minify(pt->width0, surf_tmpl->u.tex.level);
+      ps->height = u_minify(pt->height0, surf_tmpl->u.tex.level);
+      ps->usage = surf_tmpl->usage;
+
+      ps->u.tex.level = surf_tmpl->u.tex.level;
+      ps->u.tex.first_layer = surf_tmpl->u.tex.first_layer;
+      ps->u.tex.last_layer = surf_tmpl->u.tex.last_layer;
    }
    return ps;
 }
 
 
 static void 
-llvmpipe_tex_surface_destroy(struct pipe_surface *surf)
+llvmpipe_surface_destroy(struct pipe_context *pipe,
+                         struct pipe_surface *surf)
 {
    /* Effectively do the texture_update work here - if texture images
     * needed post-processing to put them into hardware layout, this is
@@ -561,16 +553,17 @@ llvmpipe_tex_surface_destroy(struct pipe_surface *surf)
 
 static struct pipe_transfer *
 llvmpipe_get_transfer(struct pipe_context *pipe,
-		      struct pipe_resource *resource,
-		      struct pipe_subresource sr,
-		      unsigned usage,
-		      const struct pipe_box *box)
+                      struct pipe_resource *resource,
+                      unsigned level,
+                      unsigned usage,
+                      const struct pipe_box *box)
 {
+   struct llvmpipe_context *llvmpipe = llvmpipe_context(pipe);
    struct llvmpipe_resource *lprex = llvmpipe_resource(resource);
    struct llvmpipe_transfer *lpr;
 
    assert(resource);
-   assert(sr.level <= resource->last_level);
+   assert(level <= resource->last_level);
 
    /*
     * Transfers, like other pipe operations, must happen in order, so flush the
@@ -580,7 +573,8 @@ llvmpipe_get_transfer(struct pipe_context *pipe,
       boolean read_only = !(usage & PIPE_TRANSFER_WRITE);
       boolean do_not_block = !!(usage & PIPE_TRANSFER_DONTBLOCK);
       if (!llvmpipe_flush_resource(pipe, resource,
-                                   sr.face, sr.level,
+                                   level,
+                                   box->depth > 1 ? -1 : box->z,
                                    0, /* flush_flags */
                                    read_only,
                                    TRUE, /* cpu_access */
@@ -594,14 +588,17 @@ llvmpipe_get_transfer(struct pipe_context *pipe,
       }
    }
 
+   if (resource == llvmpipe->constants[PIPE_SHADER_FRAGMENT][0])
+      llvmpipe->dirty |= LP_NEW_CONSTANTS;
+
    lpr = CALLOC_STRUCT(llvmpipe_transfer);
    if (lpr) {
       struct pipe_transfer *pt = &lpr->base;
       pipe_resource_reference(&pt->resource, resource);
       pt->box = *box;
-      pt->sr = sr;
-      pt->stride = lprex->row_stride[sr.level];
-      pt->slice_stride = lprex->img_stride[sr.level];
+      pt->level = level;
+      pt->stride = lprex->row_stride[level];
+      pt->layer_stride = lprex->img_stride[level];
       pt->usage = usage;
 
       return pt;
@@ -635,8 +632,7 @@ llvmpipe_transfer_map( struct pipe_context *pipe,
    enum lp_texture_usage tex_usage;
    const char *mode;
 
-   assert(transfer->sr.face < 6);
-   assert(transfer->sr.level < LP_MAX_TEXTURE_LEVELS);
+   assert(transfer->level < LP_MAX_TEXTURE_LEVELS);
 
    /*
    printf("tex_transfer_map(%d, %d  %d x %d of %d x %d,  usage %d )\n",
@@ -666,9 +662,8 @@ llvmpipe_transfer_map( struct pipe_context *pipe,
    format = lpr->base.format;
 
    map = llvmpipe_resource_map(transfer->resource,
-			       transfer->sr.face,
-			       transfer->sr.level,
-			       transfer->box.z,
+                               transfer->level,
+                               transfer->box.z,
                                tex_usage, LP_TEX_LAYOUT_LINEAR);
 
 
@@ -680,7 +675,7 @@ llvmpipe_transfer_map( struct pipe_context *pipe,
        */
       screen->timestamp++;
    }
-   
+
    map +=
       transfer->box.y / util_format_get_blockheight(format) * transfer->stride +
       transfer->box.x / util_format_get_blockwidth(format) * util_format_get_blocksize(format);
@@ -696,21 +691,20 @@ llvmpipe_transfer_unmap(struct pipe_context *pipe,
    assert(transfer->resource);
 
    llvmpipe_resource_unmap(transfer->resource,
-			   transfer->sr.face,
-			   transfer->sr.level,
-			   transfer->box.z);
+                           transfer->level,
+                           transfer->box.z);
 }
 
 static unsigned int
 llvmpipe_is_resource_referenced( struct pipe_context *pipe,
-				struct pipe_resource *presource,
-				unsigned face, unsigned level)
+                                 struct pipe_resource *presource,
+                                 unsigned level, int layer)
 {
    struct llvmpipe_context *llvmpipe = llvmpipe_context( pipe );
 
    if (presource->target == PIPE_BUFFER)
       return PIPE_UNREFERENCED;
-   
+
    return lp_setup_is_resource_referenced(llvmpipe->setup, presource);
 }
 
@@ -740,6 +734,7 @@ llvmpipe_user_buffer_create(struct pipe_screen *screen,
    buffer->base.width0 = bytes;
    buffer->base.height0 = 1;
    buffer->base.depth0 = 1;
+   buffer->base.array_size = 1;
    buffer->userBuffer = TRUE;
    buffer->data = ptr;
 
@@ -1396,8 +1391,6 @@ llvmpipe_init_screen_resource_funcs(struct pipe_screen *screen)
    screen->resource_get_handle = llvmpipe_resource_get_handle;
    screen->user_buffer_create = llvmpipe_user_buffer_create;
 
-   screen->get_tex_surface = llvmpipe_get_tex_surface;
-   screen->tex_surface_destroy = llvmpipe_tex_surface_destroy;
 }
 
 
@@ -1412,4 +1405,7 @@ llvmpipe_init_context_resource_funcs(struct pipe_context *pipe)
  
    pipe->transfer_flush_region = u_default_transfer_flush_region;
    pipe->transfer_inline_write = u_default_transfer_inline_write;
+
+   pipe->create_surface = llvmpipe_create_surface;
+   pipe->surface_destroy = llvmpipe_surface_destroy;
 }
diff --git a/src/gallium/drivers/llvmpipe/lp_texture.h b/src/gallium/drivers/llvmpipe/lp_texture.h
index 4e4a65dcb4..b789c0f409 100644
--- a/src/gallium/drivers/llvmpipe/lp_texture.h
+++ b/src/gallium/drivers/llvmpipe/lp_texture.h
@@ -172,17 +172,15 @@ llvmpipe_resource_stride(struct pipe_resource *resource,
 
 void *
 llvmpipe_resource_map(struct pipe_resource *resource,
-		      unsigned face_slice,
-		      unsigned level,
-		      unsigned zslice,
+                      unsigned level,
+                      unsigned layer,
                       enum lp_texture_usage tex_usage,
                       enum lp_texture_layout layout);
 
 void
 llvmpipe_resource_unmap(struct pipe_resource *resource,
-                       unsigned face_slice,
                        unsigned level,
-                       unsigned zslice);
+                       unsigned layer);
 
 
 void *
diff --git a/src/gallium/drivers/sw/Makefile b/src/gallium/drivers/noop/Makefile
index 2713a62ee9..29b8d73de2 100644
--- a/src/gallium/drivers/sw/Makefile
+++ b/src/gallium/drivers/noop/Makefile
@@ -4,7 +4,10 @@
 TOP = ../../../..
 include $(TOP)/configs/current
 
+LIBNAME = noop
+
 C_SOURCES = \
-	sw.c
+	noop_pipe.c \
+	noop_state.c
 
 include ../../Makefile.template
diff --git a/src/gallium/drivers/noop/SConscript b/src/gallium/drivers/noop/SConscript
new file mode 100644
index 0000000000..a4d0dcaf27
--- /dev/null
+++ b/src/gallium/drivers/noop/SConscript
@@ -0,0 +1,15 @@
+#######################################################################
+# SConscript for noop convenience library
+
+Import('*')
+
+env = env.Clone()
+
+noop = env.ConvenienceLibrary(
+	target = 'noop',
+	source = [
+		'noop_pipe.c',
+		'noop_state.c'
+		]
+    ) + extra
+Export('noop')
diff --git a/src/gallium/drivers/noop/noop_pipe.c b/src/gallium/drivers/noop/noop_pipe.c
new file mode 100644
index 0000000000..c9c463f470
--- /dev/null
+++ b/src/gallium/drivers/noop/noop_pipe.c
@@ -0,0 +1,515 @@
+/*
+ * Copyright 2010 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to 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
+ * THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
+ * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
+ * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
+ * USE OR OTHER DEALINGS IN THE SOFTWARE.
+ */
+#include <stdio.h>
+#include <errno.h>
+#include <pipe/p_defines.h>
+#include <pipe/p_state.h>
+#include <pipe/p_context.h>
+#include <pipe/p_screen.h>
+#include <util/u_memory.h>
+#include <util/u_inlines.h>
+#include <util/u_format.h>
+#include "noop_public.h"
+#include "state_tracker/sw_winsys.h"
+
+void noop_init_state_functions(struct pipe_context *ctx);
+
+/*
+ * query
+ */
+struct noop_query {
+	unsigned	query;
+};
+static struct pipe_query *noop_create_query(struct pipe_context *ctx, unsigned query_type)
+{
+	struct noop_query *query = CALLOC_STRUCT(noop_query);
+
+	return (struct pipe_query *)query;
+}
+
+static void noop_destroy_query(struct pipe_context *ctx, struct pipe_query *query)
+{
+	FREE(query);
+}
+
+static void noop_begin_query(struct pipe_context *ctx, struct pipe_query *query)
+{
+}
+
+static void noop_end_query(struct pipe_context *ctx, struct pipe_query *query)
+{
+}
+
+static boolean noop_get_query_result(struct pipe_context *ctx,
+					struct pipe_query *query,
+					boolean wait, void *vresult)
+{
+	uint64_t *result = (uint64_t*)vresult;
+
+	*result = 0;
+	return TRUE;
+}
+
+
+/*
+ * resource
+ */
+struct noop_resource {
+	struct pipe_resource	base;
+	unsigned		size;
+	char			*data;
+	struct sw_displaytarget	*dt;
+};
+
+static unsigned noop_is_resource_referenced(struct pipe_context *pipe,
+						struct pipe_resource *resource,
+						unsigned level, int layer)
+{
+	return PIPE_UNREFERENCED;
+}
+
+static struct pipe_resource *noop_resource_create(struct pipe_screen *screen,
+						const struct pipe_resource *templ)
+{
+	struct noop_resource *nresource;
+	unsigned stride;
+
+	nresource = CALLOC_STRUCT(noop_resource);
+	if (nresource == NULL)
+		return NULL;
+
+	stride = util_format_get_stride(templ->format, templ->width0);
+	nresource->base = *templ;
+	nresource->base.screen = screen;
+	nresource->size = stride * templ->height0 * templ->depth0;
+	nresource->data = malloc(nresource->size);
+	pipe_reference_init(&nresource->base.reference, 1);
+	if (nresource->data == NULL) {
+		FREE(nresource);
+		return NULL;
+	}
+#if 0
+	if (nresource->base.bind & (PIPE_BIND_DISPLAY_TARGET |
+					PIPE_BIND_SCANOUT |
+					PIPE_BIND_SHARED)) {
+		struct sw_winsys *winsys = (struct sw_winsys *)screen->winsys;
+		unsigned stride;
+
+		nresource->dt = winsys->displaytarget_create(winsys, nresource->base.bind,
+								nresource->base.format,
+								nresource->base.width0, 
+								nresource->base.height0,
+								16, &stride);
+	}
+#endif
+	return &nresource->base;
+}
+
+static struct pipe_resource *noop_resource_from_handle(struct pipe_screen * screen,
+							const struct pipe_resource *templ,
+							struct winsys_handle *whandle)
+{
+	struct sw_winsys *winsys = (struct sw_winsys *)screen->winsys;
+	struct noop_resource *nresource;
+	struct sw_displaytarget *dt;
+	unsigned stride;
+
+	dt = winsys->displaytarget_from_handle(winsys, templ, whandle, &stride);
+	if (dt == NULL) {
+		return NULL;
+	}
+	nresource = (struct noop_resource *)noop_resource_create(screen, templ);
+	nresource->dt = dt;
+	return &nresource->base;
+}
+
+static boolean noop_resource_get_handle(struct pipe_screen *screen,
+					struct pipe_resource *resource,
+					struct winsys_handle *handle)
+{
+	struct sw_winsys *winsys = (struct sw_winsys *)screen->winsys;
+	struct noop_resource *nresource = (struct noop_resource *)resource;
+
+	if (nresource->dt == NULL)
+		return FALSE;
+
+	return winsys->displaytarget_get_handle(winsys, nresource->dt, handle);
+}
+
+static void noop_resource_destroy(struct pipe_screen *screen,
+					struct pipe_resource *resource)
+{
+	struct noop_resource *nresource = (struct noop_resource *)resource;
+
+	if (nresource->dt) {
+		/* display target */
+		struct sw_winsys *winsys = (struct sw_winsys *)screen->winsys;
+		winsys->displaytarget_destroy(winsys, nresource->dt);
+	}
+	free(nresource->data);
+	FREE(resource);
+}
+
+static struct pipe_resource *noop_user_buffer_create(struct pipe_screen *screen,
+							void *ptr, unsigned bytes,
+							unsigned bind)
+{
+	struct pipe_resource templ;
+
+	templ.target = PIPE_BUFFER;
+	templ.format = PIPE_FORMAT_R8_UNORM;
+	templ.usage = PIPE_USAGE_IMMUTABLE;
+	templ.bind = bind;
+	templ.width0 = bytes;
+	templ.height0 = 1;
+	templ.depth0 = 1;
+	templ.flags = 0;
+	return noop_resource_create(screen, &templ);
+}
+
+
+/*
+ * transfer
+ */
+static struct pipe_transfer *noop_get_transfer(struct pipe_context *context,
+						struct pipe_resource *resource,
+						unsigned level,
+						enum pipe_transfer_usage usage,
+						const struct pipe_box *box)
+{
+	struct pipe_transfer *transfer;
+
+	transfer = CALLOC_STRUCT(pipe_transfer);
+	if (transfer == NULL)
+		return NULL;
+	pipe_resource_reference(&transfer->resource, resource);
+	transfer->level = level;
+	transfer->usage = usage;
+	transfer->box = *box;
+	transfer->stride = 1;
+	transfer->layer_stride = 1;
+	return transfer;
+}
+
+static void *noop_transfer_map(struct pipe_context *pipe,
+				struct pipe_transfer *transfer)
+{
+	struct noop_resource *nresource = (struct noop_resource *)transfer->resource;
+
+	return nresource->data;
+}
+
+static void noop_transfer_flush_region(struct pipe_context *pipe,
+					struct pipe_transfer *transfer,
+					const struct pipe_box *box)
+{
+}
+
+static void noop_transfer_unmap(struct pipe_context *pipe,
+				struct pipe_transfer *transfer)
+{
+}
+
+static void noop_transfer_destroy(struct pipe_context *pipe,
+					struct pipe_transfer *transfer)
+{
+	pipe_resource_reference(&transfer->resource, NULL);
+	FREE(transfer);
+}
+
+static void noop_transfer_inline_write(struct pipe_context *pipe,
+					struct pipe_resource *resource,
+					unsigned level,
+					unsigned usage,
+					const struct pipe_box *box,
+					const void *data,
+					unsigned stride,
+					unsigned layer_stride)
+{
+}
+
+
+/*
+ * clear/copy
+ */
+static void noop_clear(struct pipe_context *ctx, unsigned buffers,
+			const float *rgba, double depth, unsigned stencil)
+{
+}
+
+static void noop_clear_render_target(struct pipe_context *ctx,
+				     struct pipe_surface *dst,
+				     const float *rgba,
+				     unsigned dstx, unsigned dsty,
+				     unsigned width, unsigned height)
+{
+}
+
+static void noop_clear_depth_stencil(struct pipe_context *ctx,
+				     struct pipe_surface *dst,
+				     unsigned clear_flags,
+				     double depth,
+				     unsigned stencil,
+				     unsigned dstx, unsigned dsty,
+				     unsigned width, unsigned height)
+{
+}
+
+static void noop_resource_copy_region(struct pipe_context *ctx,
+				      struct pipe_resource *dst,
+				      unsigned dst_level,
+				      unsigned dstx, unsigned dsty, unsigned dstz,
+				      struct pipe_resource *src,
+				      unsigned src_level,
+				      const struct pipe_box *src_box)
+{
+}
+
+
+/*
+ * context
+ */
+static void noop_flush(struct pipe_context *ctx, unsigned flags,
+			struct pipe_fence_handle **fence)
+{
+}
+
+static void noop_destroy_context(struct pipe_context *ctx)
+{
+	FREE(ctx);
+}
+
+static struct pipe_context *noop_create_context(struct pipe_screen *screen, void *priv)
+{
+	struct pipe_context *ctx = CALLOC_STRUCT(pipe_context);
+
+	if (ctx == NULL)
+		return NULL;
+	ctx->winsys = screen->winsys;
+	ctx->screen = screen;
+	ctx->priv = priv;
+	ctx->destroy = noop_destroy_context;
+	ctx->flush = noop_flush;
+	ctx->clear = noop_clear;
+	ctx->clear_render_target = noop_clear_render_target;
+	ctx->clear_depth_stencil = noop_clear_depth_stencil;
+	ctx->resource_copy_region = noop_resource_copy_region;
+	ctx->create_query = noop_create_query;
+	ctx->destroy_query = noop_destroy_query;
+	ctx->begin_query = noop_begin_query;
+	ctx->end_query = noop_end_query;
+	ctx->get_query_result = noop_get_query_result;
+	ctx->get_transfer = noop_get_transfer;
+	ctx->transfer_map = noop_transfer_map;
+	ctx->transfer_flush_region = noop_transfer_flush_region;
+	ctx->transfer_unmap = noop_transfer_unmap;
+	ctx->transfer_destroy = noop_transfer_destroy;
+	ctx->transfer_inline_write = noop_transfer_inline_write;
+	ctx->is_resource_referenced = noop_is_resource_referenced;
+	noop_init_state_functions(ctx);
+
+	return ctx;
+}
+
+
+/*
+ * pipe_screen
+ */
+static void noop_flush_frontbuffer(struct pipe_screen *_screen,
+				   struct pipe_resource *resource,
+				   unsigned level, unsigned layer,
+				   void *context_private)
+{
+}
+
+static const char *noop_get_vendor(struct pipe_screen* pscreen)
+{
+	return "X.Org";
+}
+
+static const char *noop_get_name(struct pipe_screen* pscreen)
+{
+	return "NOOP";
+}
+
+static int noop_get_param(struct pipe_screen* pscreen, enum pipe_cap param)
+{
+	switch (param) {
+	/* Supported features (boolean caps). */
+	case PIPE_CAP_NPOT_TEXTURES:
+	case PIPE_CAP_TWO_SIDED_STENCIL:
+	case PIPE_CAP_GLSL:
+	case PIPE_CAP_OCCLUSION_QUERY:
+	case PIPE_CAP_POINT_SPRITE:
+	case PIPE_CAP_ANISOTROPIC_FILTER:
+	case PIPE_CAP_TEXTURE_MIRROR_CLAMP:
+	case PIPE_CAP_TEXTURE_MIRROR_REPEAT:
+	case PIPE_CAP_TEXTURE_SHADOW_MAP:
+	case PIPE_CAP_TEXTURE_SWIZZLE:
+	case PIPE_CAP_BLEND_EQUATION_SEPARATE:
+
+	  return 1;
+	case PIPE_CAP_DUAL_SOURCE_BLEND:
+
+	case PIPE_CAP_SM3:
+	case PIPE_CAP_INDEP_BLEND_ENABLE:
+	case PIPE_CAP_DEPTHSTENCIL_CLEAR_SEPARATE:
+	case PIPE_CAP_DEPTH_CLAMP:
+	case PIPE_CAP_SHADER_STENCIL_EXPORT:
+	case PIPE_CAP_TIMER_QUERY:
+	case PIPE_CAP_STREAM_OUTPUT:
+	case PIPE_CAP_PRIMITIVE_RESTART:
+	case PIPE_CAP_INDEP_BLEND_FUNC:
+		return 0;
+
+	/* Texturing. */
+	case PIPE_CAP_MAX_TEXTURE_2D_LEVELS:
+	case PIPE_CAP_MAX_TEXTURE_3D_LEVELS:
+	case PIPE_CAP_MAX_TEXTURE_CUBE_LEVELS:
+		return 14;
+	case PIPE_CAP_MAX_VERTEX_TEXTURE_UNITS:
+		return 16;
+	case PIPE_CAP_MAX_TEXTURE_IMAGE_UNITS:
+	case PIPE_CAP_MAX_COMBINED_SAMPLERS:
+		return 16;
+
+	/* Render targets. */
+	case PIPE_CAP_MAX_RENDER_TARGETS:
+		return 8;
+
+	/* Fragment coordinate conventions. */
+	case PIPE_CAP_TGSI_FS_COORD_ORIGIN_UPPER_LEFT:
+	case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_HALF_INTEGER:
+		return 1;
+	case PIPE_CAP_TGSI_FS_COORD_ORIGIN_LOWER_LEFT:
+	case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_INTEGER:
+		return 0;
+
+	default:
+		return 0;
+	}
+}
+
+static float noop_get_paramf(struct pipe_screen* pscreen, enum pipe_cap param)
+{
+	switch (param) {
+	case PIPE_CAP_MAX_LINE_WIDTH:
+	case PIPE_CAP_MAX_LINE_WIDTH_AA:
+	case PIPE_CAP_MAX_POINT_WIDTH:
+	case PIPE_CAP_MAX_POINT_WIDTH_AA:
+		return 8192.0f;
+	case PIPE_CAP_MAX_TEXTURE_ANISOTROPY:
+		return 16.0f;
+	case PIPE_CAP_MAX_TEXTURE_LOD_BIAS:
+		return 16.0f;
+	default:
+		return 0.0f;
+	}
+}
+
+static int noop_get_shader_param(struct pipe_screen* pscreen, unsigned shader, enum pipe_shader_cap param)
+{
+	switch(shader)
+	{
+	case PIPE_SHADER_FRAGMENT:
+	case PIPE_SHADER_VERTEX:
+	case PIPE_SHADER_GEOMETRY:
+		break;
+	default:
+		return 0;
+	}
+
+	switch (param) {
+	case PIPE_SHADER_CAP_MAX_INSTRUCTIONS:
+	case PIPE_SHADER_CAP_MAX_ALU_INSTRUCTIONS:
+	case PIPE_SHADER_CAP_MAX_TEX_INSTRUCTIONS:
+	case PIPE_SHADER_CAP_MAX_TEX_INDIRECTIONS:
+		return 16384;
+	case PIPE_SHADER_CAP_MAX_CONTROL_FLOW_DEPTH:
+		return 8;
+	case PIPE_SHADER_CAP_MAX_INPUTS:
+		return 16;
+	case PIPE_SHADER_CAP_MAX_TEMPS:
+		return 256;
+	case PIPE_SHADER_CAP_MAX_ADDRS:
+		return 1;
+	case PIPE_SHADER_CAP_MAX_CONSTS:
+		return 256;
+	case PIPE_SHADER_CAP_MAX_CONST_BUFFERS:
+		return 1;
+	case PIPE_SHADER_CAP_MAX_PREDS:
+		return 0;
+	case PIPE_SHADER_CAP_TGSI_CONT_SUPPORTED:
+		return 1;
+	case PIPE_SHADER_CAP_INDIRECT_INPUT_ADDR:
+	case PIPE_SHADER_CAP_INDIRECT_OUTPUT_ADDR:
+	case PIPE_SHADER_CAP_INDIRECT_TEMP_ADDR:
+	case PIPE_SHADER_CAP_INDIRECT_CONST_ADDR:
+		return 1;
+	default:
+		return 0;
+	}
+}
+
+static boolean noop_is_format_supported(struct pipe_screen* screen,
+					enum pipe_format format,
+					enum pipe_texture_target target,
+					unsigned sample_count,
+					unsigned usage,
+					unsigned geom_flags)
+{
+	return true;
+}
+
+static void noop_destroy_screen(struct pipe_screen *screen)
+{
+	FREE(screen);
+}
+
+struct pipe_screen *noop_screen_create(struct sw_winsys *winsys)
+{
+	struct pipe_screen *screen;
+
+	screen = CALLOC_STRUCT(pipe_screen);
+	if (screen == NULL) {
+		return NULL;
+	}
+
+	screen->winsys = (struct pipe_winsys*)winsys;
+	screen->destroy = noop_destroy_screen;
+	screen->get_name = noop_get_name;
+	screen->get_vendor = noop_get_vendor;
+	screen->get_param = noop_get_param;
+	screen->get_shader_param = noop_get_shader_param;
+	screen->get_paramf = noop_get_paramf;
+	screen->is_format_supported = noop_is_format_supported;
+	screen->context_create = noop_create_context;
+	screen->resource_create = noop_resource_create;
+	screen->resource_from_handle = noop_resource_from_handle;
+	screen->resource_get_handle = noop_resource_get_handle;
+	screen->resource_destroy = noop_resource_destroy;
+	screen->user_buffer_create = noop_user_buffer_create;
+	screen->flush_frontbuffer = noop_flush_frontbuffer;
+
+	return screen;
+}
diff --git a/src/gallium/winsys/radeon/drm/radeon_r300.h b/src/gallium/drivers/noop/noop_public.h
index 2703464ad8..8ce82bec69 100644
--- a/src/gallium/winsys/radeon/drm/radeon_r300.h
+++ b/src/gallium/drivers/noop/noop_public.h
@@ -1,5 +1,5 @@
 /*
- * Copyright 2008 Corbin Simpson <MostAwesomeDude@gmail.com>
+ * Copyright 2010 Red Hat Inc.
  *
  * Permission is hereby granted, free of charge, to any person obtaining a
  * copy of this software and associated documentation files (the "Software"),
@@ -18,13 +18,13 @@
  * THE AUTHOR(S) 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. */
+ * USE OR OTHER DEALINGS IN THE SOFTWARE.
+ */
+#ifndef NOOP_PUBLIC_H
+#define NOOP_PUBLIC_H
 
-#ifndef RADEON_R300_H
-#define RADEON_R300_H
+struct sw_winsys;
 
-#include "radeon_winsys.h"
+struct pipe_screen *noop_screen_create(struct sw_winsys *winsys);
 
-boolean radeon_setup_winsys(int fd, struct radeon_libdrm_winsys* winsys);
-
-#endif /* RADEON_R300_H */
+#endif
diff --git a/src/gallium/drivers/noop/noop_state.c b/src/gallium/drivers/noop/noop_state.c
new file mode 100644
index 0000000000..ad324774c0
--- /dev/null
+++ b/src/gallium/drivers/noop/noop_state.c
@@ -0,0 +1,290 @@
+/*
+ * Copyright 2010 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to 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
+ * THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
+ * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
+ * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
+ * USE OR OTHER DEALINGS IN THE SOFTWARE.
+ */
+#include <stdio.h>
+#include <errno.h>
+#include <pipe/p_defines.h>
+#include <pipe/p_state.h>
+#include <pipe/p_context.h>
+#include <pipe/p_screen.h>
+#include <util/u_memory.h>
+#include <util/u_inlines.h>
+
+static void noop_draw_vbo(struct pipe_context *ctx, const struct pipe_draw_info *info)
+{
+}
+
+static void noop_set_blend_color(struct pipe_context *ctx,
+					const struct pipe_blend_color *state)
+{
+}
+
+static void *noop_create_blend_state(struct pipe_context *ctx,
+					const struct pipe_blend_state *state)
+{
+	struct pipe_blend_state *nstate = CALLOC_STRUCT(pipe_blend_state);
+
+	if (nstate == NULL) {
+		return NULL;
+	}
+	*nstate = *state;
+	return nstate;
+}
+
+static void *noop_create_dsa_state(struct pipe_context *ctx,
+				   const struct pipe_depth_stencil_alpha_state *state)
+{
+	struct pipe_depth_stencil_alpha_state *nstate = CALLOC_STRUCT(pipe_depth_stencil_alpha_state);
+
+	if (nstate == NULL) {
+		return NULL;
+	}
+	*nstate = *state;
+	return nstate;
+}
+
+static void *noop_create_rs_state(struct pipe_context *ctx,
+					const struct pipe_rasterizer_state *state)
+{
+	struct pipe_rasterizer_state *nstate = CALLOC_STRUCT(pipe_rasterizer_state);
+
+	if (nstate == NULL) {
+		return NULL;
+	}
+	*nstate = *state;
+	return nstate;
+}
+
+static void *noop_create_sampler_state(struct pipe_context *ctx,
+					const struct pipe_sampler_state *state)
+{
+	struct pipe_sampler_state *nstate = CALLOC_STRUCT(pipe_sampler_state);
+
+	if (nstate == NULL) {
+		return NULL;
+	}
+	*nstate = *state;
+	return nstate;
+}
+
+static struct pipe_sampler_view *noop_create_sampler_view(struct pipe_context *ctx,
+							struct pipe_resource *texture,
+							const struct pipe_sampler_view *state)
+{
+	struct pipe_sampler_view *sampler_view = CALLOC_STRUCT(pipe_sampler_view);
+
+	if (sampler_view == NULL)
+		return NULL;
+	/* initialize base object */
+	pipe_resource_reference(&sampler_view->texture, texture);
+	pipe_reference_init(&sampler_view->reference, 1);
+	sampler_view->context = ctx;
+	return sampler_view;
+}
+
+static struct pipe_surface *noop_create_surface(struct pipe_context *ctx,
+						struct pipe_resource *texture,
+						const struct pipe_surface *surf_tmpl)
+{
+	struct pipe_surface *surface = CALLOC_STRUCT(pipe_surface);
+
+	if (surface == NULL)
+		return NULL;
+	pipe_reference_init(&surface->reference, 1);
+	pipe_resource_reference(&surface->texture, texture);
+	surface->context = ctx;
+	surface->format = surf_tmpl->format;
+	surface->width = texture->width0;
+	surface->height = texture->height0;
+	surface->usage = surf_tmpl->usage;
+	surface->texture = texture;
+	surface->u.tex.first_layer = surf_tmpl->u.tex.first_layer;
+	surface->u.tex.last_layer = surf_tmpl->u.tex.last_layer;
+	surface->u.tex.level = surf_tmpl->u.tex.level;
+
+	return surface;
+}
+static void noop_set_vs_sampler_view(struct pipe_context *ctx, unsigned count,
+					struct pipe_sampler_view **views)
+{
+}
+
+static void noop_set_ps_sampler_view(struct pipe_context *ctx, unsigned count,
+					struct pipe_sampler_view **views)
+{
+}
+
+static void noop_bind_sampler(struct pipe_context *ctx, unsigned count, void **states)
+{
+}
+
+static void noop_set_clip_state(struct pipe_context *ctx,
+				const struct pipe_clip_state *state)
+{
+}
+
+static void noop_set_polygon_stipple(struct pipe_context *ctx,
+					 const struct pipe_poly_stipple *state)
+{
+}
+
+static void noop_set_sample_mask(struct pipe_context *pipe, unsigned sample_mask)
+{
+}
+
+static void noop_set_scissor_state(struct pipe_context *ctx,
+					const struct pipe_scissor_state *state)
+{
+}
+
+static void noop_set_stencil_ref(struct pipe_context *ctx,
+				const struct pipe_stencil_ref *state)
+{
+}
+
+static void noop_set_viewport_state(struct pipe_context *ctx,
+					const struct pipe_viewport_state *state)
+{
+}
+
+static void noop_set_framebuffer_state(struct pipe_context *ctx,
+					const struct pipe_framebuffer_state *state)
+{
+}
+
+static void noop_set_constant_buffer(struct pipe_context *ctx,
+					uint shader, uint index,
+					struct pipe_resource *buffer)
+{
+}
+
+
+static void noop_sampler_view_destroy(struct pipe_context *ctx,
+				struct pipe_sampler_view *state)
+{
+	pipe_resource_reference(&state->texture, NULL);
+	FREE(state);
+}
+
+
+static void noop_surface_destroy(struct pipe_context *ctx,
+				 struct pipe_surface *surface)
+{
+	pipe_resource_reference(&surface->texture, NULL);
+	FREE(surface);
+}
+
+static void noop_bind_state(struct pipe_context *ctx, void *state)
+{
+}
+
+static void noop_delete_state(struct pipe_context *ctx, void *state)
+{
+	FREE(state);
+}
+
+static void noop_delete_vertex_element(struct pipe_context *ctx, void *state)
+{
+	FREE(state);
+}
+
+
+static void noop_set_index_buffer(struct pipe_context *ctx,
+					const struct pipe_index_buffer *ib)
+{
+}
+
+static void noop_set_vertex_buffers(struct pipe_context *ctx, unsigned count,
+					const struct pipe_vertex_buffer *buffers)
+{
+}
+
+static void *noop_create_vertex_elements(struct pipe_context *ctx,
+					unsigned count,
+					const struct pipe_vertex_element *state)
+{
+	struct pipe_vertex_element *nstate = CALLOC_STRUCT(pipe_vertex_element);
+
+	if (nstate == NULL) {
+		return NULL;
+	}
+	*nstate = *state;
+	return nstate;
+}
+
+static void *noop_create_shader_state(struct pipe_context *ctx,
+					const struct pipe_shader_state *state)
+{
+	struct pipe_shader_state *nstate = CALLOC_STRUCT(pipe_shader_state);
+
+	if (nstate == NULL) {
+		return NULL;
+	}
+	*nstate = *state;
+	return nstate;
+}
+
+void noop_init_state_functions(struct pipe_context *ctx);
+
+void noop_init_state_functions(struct pipe_context *ctx)
+{
+	ctx->create_blend_state = noop_create_blend_state;
+	ctx->create_depth_stencil_alpha_state = noop_create_dsa_state;
+	ctx->create_fs_state = noop_create_shader_state;
+	ctx->create_rasterizer_state = noop_create_rs_state;
+	ctx->create_sampler_state = noop_create_sampler_state;
+	ctx->create_sampler_view = noop_create_sampler_view;
+	ctx->create_surface = noop_create_surface;
+	ctx->create_vertex_elements_state = noop_create_vertex_elements;
+	ctx->create_vs_state = noop_create_shader_state;
+	ctx->bind_blend_state = noop_bind_state;
+	ctx->bind_depth_stencil_alpha_state = noop_bind_state;
+	ctx->bind_fragment_sampler_states = noop_bind_sampler;
+	ctx->bind_fs_state = noop_bind_state;
+	ctx->bind_rasterizer_state = noop_bind_state;
+	ctx->bind_vertex_elements_state = noop_bind_state;
+	ctx->bind_vertex_sampler_states = noop_bind_sampler;
+	ctx->bind_vs_state = noop_bind_state;
+	ctx->delete_blend_state = noop_delete_state;
+	ctx->delete_depth_stencil_alpha_state = noop_delete_state;
+	ctx->delete_fs_state = noop_delete_state;
+	ctx->delete_rasterizer_state = noop_delete_state;
+	ctx->delete_sampler_state = noop_delete_state;
+	ctx->delete_vertex_elements_state = noop_delete_vertex_element;
+	ctx->delete_vs_state = noop_delete_state;
+	ctx->set_blend_color = noop_set_blend_color;
+	ctx->set_clip_state = noop_set_clip_state;
+	ctx->set_constant_buffer = noop_set_constant_buffer;
+	ctx->set_fragment_sampler_views = noop_set_ps_sampler_view;
+	ctx->set_framebuffer_state = noop_set_framebuffer_state;
+	ctx->set_polygon_stipple = noop_set_polygon_stipple;
+	ctx->set_sample_mask = noop_set_sample_mask;
+	ctx->set_scissor_state = noop_set_scissor_state;
+	ctx->set_stencil_ref = noop_set_stencil_ref;
+	ctx->set_vertex_buffers = noop_set_vertex_buffers;
+	ctx->set_index_buffer = noop_set_index_buffer;
+	ctx->set_vertex_sampler_views = noop_set_vs_sampler_view;
+	ctx->set_viewport_state = noop_set_viewport_state;
+	ctx->sampler_view_destroy = noop_sampler_view_destroy;
+	ctx->surface_destroy = noop_surface_destroy;
+	ctx->draw_vbo = noop_draw_vbo;
+}
diff --git a/src/gallium/drivers/nv50/nv50_buffer.c b/src/gallium/drivers/nv50/nv50_buffer.c
index dacfee9799..45356f9f63 100644
--- a/src/gallium/drivers/nv50/nv50_buffer.c
+++ b/src/gallium/drivers/nv50/nv50_buffer.c
@@ -136,6 +136,7 @@ nv50_user_buffer_create(struct pipe_screen *pscreen,
 	buffer->base.width0 = bytes;
 	buffer->base.height0 = 1;
 	buffer->base.depth0 = 1;
+	buffer->base.array_size = 1;
 
 	buffer->bo = nouveau_screen_bo_user(pscreen, ptr, bytes);
 	if (!buffer->bo)
diff --git a/src/gallium/drivers/nv50/nv50_context.h b/src/gallium/drivers/nv50/nv50_context.h
index bf6a577188..b2b0b72fe2 100644
--- a/src/gallium/drivers/nv50/nv50_context.h
+++ b/src/gallium/drivers/nv50/nv50_context.h
@@ -108,6 +108,7 @@ get_tile_depth(uint32_t tile_mode)
 
 struct nv50_surface {
 	struct pipe_surface base;
+	unsigned offset;
 };
 
 static INLINE struct nv50_surface *
diff --git a/src/gallium/drivers/nv50/nv50_miptree.c b/src/gallium/drivers/nv50/nv50_miptree.c
index dd0e8fd41b..309b6503ca 100644
--- a/src/gallium/drivers/nv50/nv50_miptree.c
+++ b/src/gallium/drivers/nv50/nv50_miptree.c
@@ -276,46 +276,53 @@ nv50_miptree_from_handle(struct pipe_screen *pscreen,
  */
 
 struct pipe_surface *
-nv50_miptree_surface_new(struct pipe_screen *pscreen, struct pipe_resource *pt,
-			 unsigned face, unsigned level, unsigned zslice,
-			 unsigned flags)
+nv50_miptree_surface_new(struct pipe_context *pipe, struct pipe_resource *pt,
+			 const struct pipe_surface *surf_tmpl)
 {
+	unsigned level = surf_tmpl->u.tex.level;
 	struct nv50_miptree *mt = nv50_miptree(pt);
 	struct nv50_miptree_level *lvl = &mt->level[level];
-	struct pipe_surface *ps;
-	unsigned img = 0;
+	struct nv50_surface *ns;
+	unsigned img = 0, zslice = 0;
 
+	assert(surf_tmpl->u.tex.first_layer == surf_tmpl->u.tex.last_layer);
+
+	/* XXX can't unify these here? */
 	if (pt->target == PIPE_TEXTURE_CUBE)
-		img = face;
+		img = surf_tmpl->u.tex.first_layer;
+	else if (pt->target == PIPE_TEXTURE_3D)
+		zslice = surf_tmpl->u.tex.first_layer;
 
-	ps = CALLOC_STRUCT(pipe_surface);
-	if (!ps)
+	ns = CALLOC_STRUCT(nv50_surface);
+	if (!ns)
 		return NULL;
-	pipe_resource_reference(&ps->texture, pt);
-	ps->format = pt->format;
-	ps->width = u_minify(pt->width0, level);
-	ps->height = u_minify(pt->height0, level);
-	ps->usage = flags;
-	pipe_reference_init(&ps->reference, 1);
-	ps->face = face;
-	ps->level = level;
-	ps->zslice = zslice;
-	ps->offset = lvl->image_offset[img];
+	pipe_resource_reference(&ns->base.texture, pt);
+	ns->base.context = pipe;
+	ns->base.format = pt->format;
+	ns->base.width = u_minify(pt->width0, level);
+	ns->base.height = u_minify(pt->height0, level);
+	ns->base.usage = surf_tmpl->usage;
+	pipe_reference_init(&ns->base.reference, 1);
+	ns->base.u.tex.level = level;
+	ns->base.u.tex.first_layer = surf_tmpl->u.tex.first_layer;
+	ns->base.u.tex.last_layer = surf_tmpl->u.tex.last_layer;
+	ns->offset = lvl->image_offset[img];
 
 	if (pt->target == PIPE_TEXTURE_3D) {
-		unsigned nb_h = util_format_get_nblocksy(pt->format, ps->height);
-		ps->offset += get_zslice_offset(lvl->tile_mode, zslice,
+		unsigned nb_h = util_format_get_nblocksy(pt->format, ns->base.height);
+		ns->offset += get_zslice_offset(lvl->tile_mode, zslice,
 						lvl->pitch, nb_h);
 	}
 
-	return ps;
+	return &ns->base;
 }
 
 void
-nv50_miptree_surface_del(struct pipe_surface *ps)
+nv50_miptree_surface_del(struct pipe_context *pipe,
+			 struct pipe_surface *ps)
 {
 	struct nv50_surface *s = nv50_surface(ps);
 
-	pipe_resource_reference(&ps->texture, NULL);
+	pipe_resource_reference(&s->base.texture, NULL);
 	FREE(s);
 }
diff --git a/src/gallium/drivers/nv50/nv50_resource.c b/src/gallium/drivers/nv50/nv50_resource.c
index cfdb60418b..6c0a969635 100644
--- a/src/gallium/drivers/nv50/nv50_resource.c
+++ b/src/gallium/drivers/nv50/nv50_resource.c
@@ -15,7 +15,7 @@
 static unsigned int
 nv50_resource_is_referenced(struct pipe_context *pipe,
 			    struct pipe_resource *resource,
-			    unsigned face, unsigned level)
+			    unsigned level, int layer)
 {
 	return nouveau_reference_flags(nv50_resource(resource)->bo);
 }
@@ -51,6 +51,9 @@ nv50_init_resource_functions(struct pipe_context *pcontext)
 	pcontext->transfer_destroy = u_transfer_destroy_vtbl;
 	pcontext->transfer_inline_write = u_transfer_inline_write_vtbl;
 	pcontext->is_resource_referenced = nv50_resource_is_referenced;
+
+	pcontext->create_surface = nv50_miptree_surface_new;
+	pcontext->surface_destroy = nv50_miptree_surface_del;
 }
 
 void
@@ -61,7 +64,4 @@ nv50_screen_init_resource_functions(struct pipe_screen *pscreen)
 	pscreen->resource_get_handle = u_resource_get_handle_vtbl;
 	pscreen->resource_destroy = u_resource_destroy_vtbl;
 	pscreen->user_buffer_create = nv50_user_buffer_create;
-   
-	pscreen->get_tex_surface = nv50_miptree_surface_new;
-	pscreen->tex_surface_destroy = nv50_miptree_surface_del;
 }
diff --git a/src/gallium/drivers/nv50/nv50_resource.h b/src/gallium/drivers/nv50/nv50_resource.h
index f435a5892e..4b2a75e11a 100644
--- a/src/gallium/drivers/nv50/nv50_resource.h
+++ b/src/gallium/drivers/nv50/nv50_resource.h
@@ -87,12 +87,11 @@ nv50_user_buffer_create(struct pipe_screen *screen,
 
 
 struct pipe_surface *
-nv50_miptree_surface_new(struct pipe_screen *pscreen, struct pipe_resource *pt,
-			 unsigned face, unsigned level, unsigned zslice,
-			 unsigned flags);
+nv50_miptree_surface_new(struct pipe_context *pipe, struct pipe_resource *pt,
+			 const struct pipe_surface *surf_tmpl);
 
 void
-nv50_miptree_surface_del(struct pipe_surface *ps);
+nv50_miptree_surface_del(struct pipe_context *pipe, struct pipe_surface *ps);
 
 
 #endif
diff --git a/src/gallium/drivers/nv50/nv50_screen.c b/src/gallium/drivers/nv50/nv50_screen.c
index 51eab3a0b0..edc3d54d01 100644
--- a/src/gallium/drivers/nv50/nv50_screen.c
+++ b/src/gallium/drivers/nv50/nv50_screen.c
@@ -127,6 +127,10 @@ nv50_screen_get_param(struct pipe_screen *pscreen, enum pipe_cap param)
 		return 0;
 	case PIPE_CAP_DEPTH_CLAMP:
 		return 1;
+	case PIPE_CAP_SHADER_STENCIL_EXPORT:
+		return 0;
+	case PIPE_CAP_PRIMITIVE_RESTART:
+		return 0;
 	default:
 		NOUVEAU_ERR("Unknown PIPE_CAP %d\n", param);
 		return 0;
@@ -176,6 +180,8 @@ nv50_screen_get_shader_param(struct pipe_screen *pscreen, unsigned shader,
 	case PIPE_SHADER_CAP_INDIRECT_TEMP_ADDR:
 	case PIPE_SHADER_CAP_INDIRECT_CONST_ADDR:
 		return 1;
+	case PIPE_SHADER_CAP_SUBROUTINES:
+		return 0;
 	default:
 		return 0;
 	}
diff --git a/src/gallium/drivers/nv50/nv50_state_validate.c b/src/gallium/drivers/nv50/nv50_state_validate.c
index 16c2dab9af..ae02143e35 100644
--- a/src/gallium/drivers/nv50/nv50_state_validate.c
+++ b/src/gallium/drivers/nv50/nv50_state_validate.c
@@ -63,13 +63,13 @@ validate_fb(struct nv50_context *nv50)
 		so_data  (so, fb->cbufs[i]->height);
 
 		so_method(so, tesla, NV50TCL_RT_ADDRESS_HIGH(i), 5);
-		so_reloc (so, bo, fb->cbufs[i]->offset, NOUVEAU_BO_VRAM |
+		so_reloc (so, bo, ((struct nv50_surface *)fb->cbufs[i])->offset, NOUVEAU_BO_VRAM |
 			      NOUVEAU_BO_HIGH | NOUVEAU_BO_RDWR, 0, 0);
-		so_reloc (so, bo, fb->cbufs[i]->offset, NOUVEAU_BO_VRAM |
+		so_reloc (so, bo, ((struct nv50_surface *)fb->cbufs[i])->offset, NOUVEAU_BO_VRAM |
 			      NOUVEAU_BO_LOW | NOUVEAU_BO_RDWR, 0, 0);
 		so_data  (so, nv50_format_table[fb->cbufs[i]->format].rt);
 		so_data  (so, nv50_miptree(pt)->
-			      level[fb->cbufs[i]->level].tile_mode << 4);
+			      level[fb->cbufs[i]->u.tex.level].tile_mode << 4);
 		so_data(so, 0x00000000);
 
 		so_method(so, tesla, NV50TCL_RT_ARRAY_MODE, 1);
@@ -92,13 +92,13 @@ validate_fb(struct nv50_context *nv50)
 		assert(nv50_format_table[fb->zsbuf->format].rt);
 
 		so_method(so, tesla, NV50TCL_ZETA_ADDRESS_HIGH, 5);
-		so_reloc (so, bo, fb->zsbuf->offset, NOUVEAU_BO_VRAM |
+		so_reloc (so, bo, ((struct nv50_surface *)(fb->zsbuf))->offset, NOUVEAU_BO_VRAM |
 			      NOUVEAU_BO_HIGH | NOUVEAU_BO_RDWR, 0, 0);
-		so_reloc (so, bo, fb->zsbuf->offset, NOUVEAU_BO_VRAM |
+		so_reloc (so, bo, ((struct nv50_surface *)(fb->zsbuf))->offset, NOUVEAU_BO_VRAM |
 			      NOUVEAU_BO_LOW | NOUVEAU_BO_RDWR, 0, 0);
 		so_data  (so, nv50_format_table[fb->zsbuf->format].rt);
 		so_data  (so, nv50_miptree(pt)->
-			      level[fb->zsbuf->level].tile_mode << 4);
+			      level[fb->zsbuf->u.tex.level].tile_mode << 4);
 		so_data  (so, 0x00000000);
 
 		so_method(so, tesla, NV50TCL_ZETA_ENABLE, 1);
diff --git a/src/gallium/drivers/nv50/nv50_surface.c b/src/gallium/drivers/nv50/nv50_surface.c
index f70c138fe1..ce48022db4 100644
--- a/src/gallium/drivers/nv50/nv50_surface.c
+++ b/src/gallium/drivers/nv50/nv50_surface.c
@@ -97,23 +97,23 @@ nv50_surface_set(struct nv50_screen *screen, struct pipe_surface *ps, int dst)
  		OUT_RING  (chan, format);
  		OUT_RING  (chan, 1);
  		BEGIN_RING(chan, eng2d, mthd + 0x14, 5);
-		OUT_RING  (chan, mt->level[ps->level].pitch);
+		OUT_RING  (chan, mt->level[ps->u.tex.level].pitch);
  		OUT_RING  (chan, ps->width);
  		OUT_RING  (chan, ps->height);
- 		OUT_RELOCh(chan, bo, ps->offset, flags);
- 		OUT_RELOCl(chan, bo, ps->offset, flags);
+ 		OUT_RELOCh(chan, bo, ((struct nv50_surface *)ps)->offset, flags);
+ 		OUT_RELOCl(chan, bo, ((struct nv50_surface *)ps)->offset, flags);
  	} else {
  		BEGIN_RING(chan, eng2d, mthd, 5);
  		OUT_RING  (chan, format);
  		OUT_RING  (chan, 0);
-		OUT_RING  (chan, mt->level[ps->level].tile_mode << 4);
+		OUT_RING  (chan, mt->level[ps->u.tex.level].tile_mode << 4);
  		OUT_RING  (chan, 1);
  		OUT_RING  (chan, 0);
  		BEGIN_RING(chan, eng2d, mthd + 0x18, 4);
  		OUT_RING  (chan, ps->width);
  		OUT_RING  (chan, ps->height);
- 		OUT_RELOCh(chan, bo, ps->offset, flags);
- 		OUT_RELOCl(chan, bo, ps->offset, flags);
+ 		OUT_RELOCh(chan, bo, ((struct nv50_surface *)ps)->offset, flags);
+ 		OUT_RELOCl(chan, bo, ((struct nv50_surface *)ps)->offset, flags);
  	}
  
 #if 0
@@ -173,30 +173,41 @@ nv50_surface_do_copy(struct nv50_screen *screen, struct pipe_surface *dst,
 
 static void
 nv50_surface_copy(struct pipe_context *pipe,
-		  struct pipe_resource *dest, struct pipe_subresource subdst,
+		  struct pipe_resource *dest, unsigned dst_level,
 		  unsigned destx, unsigned desty, unsigned destz,
-		  struct pipe_resource *src, struct pipe_subresource subsrc,
-		  unsigned srcx, unsigned srcy, unsigned srcz,
-		  unsigned width, unsigned height)
+		  struct pipe_resource *src, unsigned src_level,
+		  const struct pipe_box *src_box)
 {
 	struct nv50_context *nv50 = nv50_context(pipe);
 	struct nv50_screen *screen = nv50->screen;
-	struct pipe_surface *ps_dst, *ps_src;
+	struct pipe_surface *ps_dst, *ps_src, surf_tmpl;
+
 
 	assert((src->format == dest->format) ||
 	       (nv50_2d_format_faithful(src->format) &&
 		nv50_2d_format_faithful(dest->format)));
-
-	ps_src = nv50_miptree_surface_new(pipe->screen, src, subsrc.face,
-					  subsrc.level, srcz, 0 /* bind flags */);
-	ps_dst = nv50_miptree_surface_new(pipe->screen, dest, subdst.face,
-					  subdst.level, destz, 0 /* bindflags */);
-
-	nv50_surface_do_copy(screen, ps_dst, destx, desty, ps_src, srcx,
-			     srcy, width, height);
-
-	nv50_miptree_surface_del(ps_src);
-	nv50_miptree_surface_del(ps_dst);
+	assert(src_box->depth == 1);
+
+	memset(&surf_tmpl, 0, sizeof(surf_tmpl));
+	surf_tmpl.format = src->format;
+	surf_tmpl.usage = 0; /* no bind flag - not a surface */
+	surf_tmpl.u.tex.level = src_level;
+	surf_tmpl.u.tex.first_layer = src_box->z;
+	surf_tmpl.u.tex.last_layer = src_box->z;
+	/* XXX really need surfaces here? */
+	ps_src = nv50_miptree_surface_new(pipe, src, &surf_tmpl);
+	surf_tmpl.format = dest->format;
+	surf_tmpl.usage = 0; /* no bind flag - not a surface */
+	surf_tmpl.u.tex.level = dst_level;
+	surf_tmpl.u.tex.first_layer = destz;
+	surf_tmpl.u.tex.last_layer = destz;
+	ps_dst = nv50_miptree_surface_new(pipe, dest, &surf_tmpl);
+
+	nv50_surface_do_copy(screen, ps_dst, destx, desty, ps_src, src_box->x,
+			     src_box->y, src_box->width, src_box->height);
+
+	nv50_miptree_surface_del(pipe, ps_src);
+	nv50_miptree_surface_del(pipe, ps_dst);
 }
 
 static void
@@ -225,10 +236,10 @@ nv50_clear_render_target(struct pipe_context *pipe,
 	BEGIN_RING(chan, tesla, NV50TCL_RT_CONTROL, 1);
 	OUT_RING  (chan, 1);
 	BEGIN_RING(chan, tesla, NV50TCL_RT_ADDRESS_HIGH(0), 5);
-	OUT_RELOCh(chan, bo, dst->offset, NOUVEAU_BO_VRAM | NOUVEAU_BO_WR);
-	OUT_RELOCl(chan, bo, dst->offset, NOUVEAU_BO_VRAM | NOUVEAU_BO_WR);
+	OUT_RELOCh(chan, bo, ((struct nv50_surface *)dst)->offset, NOUVEAU_BO_VRAM | NOUVEAU_BO_WR);
+	OUT_RELOCl(chan, bo, ((struct nv50_surface *)dst)->offset, NOUVEAU_BO_VRAM | NOUVEAU_BO_WR);
 	OUT_RING  (chan, nv50_format_table[dst->format].rt);
-	OUT_RING  (chan, mt->level[dst->level].tile_mode << 4);
+	OUT_RING  (chan, mt->level[dst->u.tex.level].tile_mode << 4);
 	OUT_RING  (chan, 0);
 	BEGIN_RING(chan, tesla, NV50TCL_RT_HORIZ(0), 2);
 	OUT_RING  (chan, dst->width);
@@ -281,10 +292,10 @@ nv50_clear_depth_stencil(struct pipe_context *pipe,
 		return;
 
 	BEGIN_RING(chan, tesla, NV50TCL_ZETA_ADDRESS_HIGH, 5);
-	OUT_RELOCh(chan, bo, dst->offset, NOUVEAU_BO_VRAM | NOUVEAU_BO_WR);
-	OUT_RELOCl(chan, bo, dst->offset, NOUVEAU_BO_VRAM | NOUVEAU_BO_WR);
+	OUT_RELOCh(chan, bo, ((struct nv50_surface *)dst)->offset, NOUVEAU_BO_VRAM | NOUVEAU_BO_WR);
+	OUT_RELOCl(chan, bo, ((struct nv50_surface *)dst)->offset, NOUVEAU_BO_VRAM | NOUVEAU_BO_WR);
 	OUT_RING  (chan, nv50_format_table[dst->format].rt);
-	OUT_RING  (chan, mt->level[dst->level].tile_mode << 4);
+	OUT_RING  (chan, mt->level[dst->u.tex.level].tile_mode << 4);
 	OUT_RING  (chan, 0);
 	BEGIN_RING(chan, tesla, NV50TCL_ZETA_ENABLE, 1);
 	OUT_RING  (chan, 1);
diff --git a/src/gallium/drivers/nv50/nv50_tex.c b/src/gallium/drivers/nv50/nv50_tex.c
index 658324ec5b..9243f9edce 100644
--- a/src/gallium/drivers/nv50/nv50_tex.c
+++ b/src/gallium/drivers/nv50/nv50_tex.c
@@ -106,7 +106,7 @@ nv50_tex_construct(struct nv50_sampler_view *view)
 
 	tic[6] = 0x03000000;
 
-	tic[7] = (view->pipe.last_level << 4) | view->pipe.first_level;
+	tic[7] = (view->pipe.u.tex.last_level << 4) | view->pipe.u.tex.first_level;
 
 	return TRUE;
 }
diff --git a/src/gallium/drivers/nv50/nv50_transfer.c b/src/gallium/drivers/nv50/nv50_transfer.c
index 0cc2f4a837..bf5af4ddc6 100644
--- a/src/gallium/drivers/nv50/nv50_transfer.c
+++ b/src/gallium/drivers/nv50/nv50_transfer.c
@@ -126,20 +126,23 @@ nv50_transfer_rect_m2mf(struct pipe_screen *pscreen,
 struct pipe_transfer *
 nv50_miptree_transfer_new(struct pipe_context *pcontext,
 			  struct pipe_resource *pt,
-			  struct pipe_subresource sr,
+			  unsigned level,
 			  unsigned usage,
 			  const struct pipe_box *box)
 {
         struct pipe_screen *pscreen = pcontext->screen;
 	struct nouveau_device *dev = nouveau_screen(pscreen)->device;
 	struct nv50_miptree *mt = nv50_miptree(pt);
-	struct nv50_miptree_level *lvl = &mt->level[sr.level];
+	struct nv50_miptree_level *lvl = &mt->level[level];
 	struct nv50_transfer *tx;
-	unsigned nx, ny, image = 0;
+	unsigned nx, ny, image = 0, boxz = 0;
 	int ret;
 
+	/* XXX can't unify these here? */
 	if (pt->target == PIPE_TEXTURE_CUBE)
-		image = sr.face;
+		image = box->z;
+	else if (pt->target == PIPE_TEXTURE_3D)
+		boxz = box->z;
 
 	tx = CALLOC_STRUCT(nv50_transfer);
 	if (!tx)
@@ -151,21 +154,21 @@ nv50_miptree_transfer_new(struct pipe_context *pcontext,
 
 
 	pipe_resource_reference(&tx->base.resource, pt);
-	tx->base.sr = sr;
+	tx->base.level = level;
 	tx->base.usage = usage;
 	tx->base.box = *box;
-	tx->nblocksx = util_format_get_nblocksx(pt->format, u_minify(pt->width0, sr.level));
-	tx->nblocksy = util_format_get_nblocksy(pt->format, u_minify(pt->height0, sr.level));
+	tx->nblocksx = util_format_get_nblocksx(pt->format, u_minify(pt->width0, level));
+	tx->nblocksy = util_format_get_nblocksy(pt->format, u_minify(pt->height0, level));
 	tx->base.stride = tx->nblocksx * util_format_get_blocksize(pt->format);
 	tx->base.usage = usage;
 
 	tx->level_pitch = lvl->pitch;
-	tx->level_width = u_minify(mt->base.base.width0, sr.level);
-	tx->level_height = u_minify(mt->base.base.height0, sr.level);
-	tx->level_depth = u_minify(mt->base.base.depth0, sr.level);
+	tx->level_width = u_minify(mt->base.base.width0, level);
+	tx->level_height = u_minify(mt->base.base.height0, level);
+	tx->level_depth = u_minify(mt->base.base.depth0, level);
 	tx->level_offset = lvl->image_offset[image];
 	tx->level_tiling = lvl->tile_mode;
-	tx->level_z = box->z;
+	tx->level_z = boxz;
 	tx->level_x = util_format_get_nblocksx(pt->format, box->x);
 	tx->level_y = util_format_get_nblocksy(pt->format, box->y);
 	ret = nouveau_bo_new(dev, NOUVEAU_BO_GART | NOUVEAU_BO_MAP, 0,
@@ -181,7 +184,7 @@ nv50_miptree_transfer_new(struct pipe_context *pcontext,
 
 		nv50_transfer_rect_m2mf(pscreen, mt->base.bo, tx->level_offset,
 					tx->level_pitch, tx->level_tiling,
-					box->x, box->y, box->z,
+					box->x, box->y, boxz,
 					tx->nblocksx, tx->nblocksy,
 					tx->level_depth,
 					tx->bo, 0,
diff --git a/src/gallium/drivers/nv50/nv50_transfer.h b/src/gallium/drivers/nv50/nv50_transfer.h
index 663503547c..6699bf546e 100644
--- a/src/gallium/drivers/nv50/nv50_transfer.h
+++ b/src/gallium/drivers/nv50/nv50_transfer.h
@@ -8,7 +8,7 @@
 struct pipe_transfer *
 nv50_miptree_transfer_new(struct pipe_context *pcontext,
 			  struct pipe_resource *pt,
-			  struct pipe_subresource sr,
+			  unsigned level,
 			  unsigned usage,
 			  const struct pipe_box *box);
 void
diff --git a/src/gallium/drivers/nvfx/nv30_fragtex.c b/src/gallium/drivers/nvfx/nv30_fragtex.c
index d6ede5b40a..951fb202ed 100644
--- a/src/gallium/drivers/nvfx/nv30_fragtex.c
+++ b/src/gallium/drivers/nvfx/nv30_fragtex.c
@@ -37,12 +37,12 @@ nv30_sampler_view_init(struct pipe_context *pipe,
 	struct pipe_resource* pt = sv->base.texture;
 	struct nvfx_texture_format *tf = &nvfx_texture_formats[sv->base.format];
 	unsigned txf;
-	unsigned level = pt->target == PIPE_TEXTURE_CUBE ? 0 : sv->base.first_level;
+	unsigned level = pt->target == PIPE_TEXTURE_CUBE ? 0 : sv->base.u.tex.first_level;
 
 	assert(tf->fmt[0] >= 0);
 
 	txf = sv->u.init_fmt;
-	txf |= (level != sv->base.last_level ? NV30_3D_TEX_FORMAT_MIPMAP : 0);
+	txf |= (level != sv->base.u.tex.last_level ? NV30_3D_TEX_FORMAT_MIPMAP : 0);
 	txf |= util_logbase2(u_minify(pt->width0, level)) << NV30_3D_TEX_FORMAT_BASE_SIZE_U__SHIFT;
 	txf |= util_logbase2(u_minify(pt->height0, level)) << NV30_3D_TEX_FORMAT_BASE_SIZE_V__SHIFT;
 	txf |= util_logbase2(u_minify(pt->depth0, level)) << NV30_3D_TEX_FORMAT_BASE_SIZE_W__SHIFT;
@@ -60,8 +60,8 @@ nv30_sampler_view_init(struct pipe_context *pipe,
 	else
 		sv->u.nv30.rect = !!(pt->flags & NVFX_RESOURCE_FLAG_LINEAR);
 
-	sv->lod_offset = sv->base.first_level - level;
-	sv->max_lod_limit = sv->base.last_level - level;
+	sv->lod_offset = sv->base.u.tex.first_level - level;
+	sv->max_lod_limit = sv->base.u.tex.last_level - level;
 }
 
 void
diff --git a/src/gallium/drivers/nvfx/nv40_fragtex.c b/src/gallium/drivers/nvfx/nv40_fragtex.c
index d4fb73702d..e8ab403f72 100644
--- a/src/gallium/drivers/nvfx/nv40_fragtex.c
+++ b/src/gallium/drivers/nvfx/nv40_fragtex.c
@@ -46,7 +46,7 @@ nv40_sampler_view_init(struct pipe_context *pipe,
 	struct nvfx_miptree* mt = (struct nvfx_miptree*)pt;
 	struct nvfx_texture_format *tf = &nvfx_texture_formats[sv->base.format];
 	unsigned txf;
-	unsigned level = pt->target == PIPE_TEXTURE_CUBE ? 0 : sv->base.first_level;
+	unsigned level = pt->target == PIPE_TEXTURE_CUBE ? 0 : sv->base.u.tex.first_level;
 	assert(tf->fmt[4] >= 0);
 
 	txf = sv->u.init_fmt;
@@ -54,7 +54,7 @@ nv40_sampler_view_init(struct pipe_context *pipe,
 	if(pt->target == PIPE_TEXTURE_CUBE)
 		txf |= ((pt->last_level + 1) << NV40_3D_TEX_FORMAT_MIPMAP_COUNT__SHIFT);
 	else
-		txf |= (((sv->base.last_level - sv->base.first_level) + 1) << NV40_3D_TEX_FORMAT_MIPMAP_COUNT__SHIFT);
+		txf |= (((sv->base.u.tex.last_level - sv->base.u.tex.first_level) + 1) << NV40_3D_TEX_FORMAT_MIPMAP_COUNT__SHIFT);
 
 	if (!mt->linear_pitch)
 		sv->u.nv40.npot_size2 = 0;
@@ -68,8 +68,8 @@ nv40_sampler_view_init(struct pipe_context *pipe,
 
 	sv->u.nv40.npot_size2 |= (u_minify(pt->depth0, level) << NV40_3D_TEX_SIZE1_DEPTH__SHIFT);
 
-	sv->lod_offset = (sv->base.first_level - level) * 256;
-	sv->max_lod_limit = (sv->base.last_level - level) * 256;
+	sv->lod_offset = (sv->base.u.tex.first_level - level) * 256;
+	sv->max_lod_limit = (sv->base.u.tex.last_level - level) * 256;
 }
 
 void
diff --git a/src/gallium/drivers/nvfx/nvfx_buffer.c b/src/gallium/drivers/nvfx/nvfx_buffer.c
index 041099e0e5..b407429731 100644
--- a/src/gallium/drivers/nvfx/nvfx_buffer.c
+++ b/src/gallium/drivers/nvfx/nvfx_buffer.c
@@ -64,6 +64,7 @@ nvfx_user_buffer_create(struct pipe_screen *pscreen,
 	buffer->base.base.width0 = bytes;
 	buffer->base.base.height0 = 1;
 	buffer->base.base.depth0 = 1;
+	buffer->base.base.array_size = 1;
 	buffer->data = ptr;
 	buffer->size = bytes;
 	buffer->bytes_to_draw_until_static = bytes * screen->static_reuse_threshold;
diff --git a/src/gallium/drivers/nvfx/nvfx_fragprog.c b/src/gallium/drivers/nvfx/nvfx_fragprog.c
index 13e8beed47..1740d72a8a 100644
--- a/src/gallium/drivers/nvfx/nvfx_fragprog.c
+++ b/src/gallium/drivers/nvfx/nvfx_fragprog.c
@@ -1189,12 +1189,12 @@ out_err:
 static inline void
 nvfx_fp_memcpy(void* dst, const void* src, size_t len)
 {
-#ifndef WORDS_BIGENDIAN
+#ifndef PIPE_ARCH_BIG_ENDIAN
 	memcpy(dst, src, len);
 #else
 	size_t i;
 	for(i = 0; i < len; i += 4) {
-		uint32_t v = (uint32_t*)((char*)src + i);
+		uint32_t v = *(uint32_t*)((char*)src + i);
 		*(uint32_t*)((char*)dst + i) = (v >> 16) | (v << 16);
 	}
 #endif
diff --git a/src/gallium/drivers/nvfx/nvfx_fragtex.c b/src/gallium/drivers/nvfx/nvfx_fragtex.c
index 1d6b4e24cb..fd0aff6a1a 100644
--- a/src/gallium/drivers/nvfx/nvfx_fragtex.c
+++ b/src/gallium/drivers/nvfx/nvfx_fragtex.c
@@ -122,8 +122,8 @@ nvfx_create_sampler_view(struct pipe_context *pipe,
 	}
 	else
 	{
-		sv->offset = nvfx_subresource_offset(pt, 0, sv->base.first_level, 0);
-		sv->npot_size = (u_minify(pt->width0, sv->base.first_level) << NV30_3D_TEX_NPOT_SIZE_W__SHIFT) | u_minify(pt->height0, sv->base.first_level);
+		sv->offset = nvfx_subresource_offset(pt, 0, sv->base.u.tex.first_level, 0);
+		sv->npot_size = (u_minify(pt->width0, sv->base.u.tex.first_level) << NV30_3D_TEX_NPOT_SIZE_W__SHIFT) | u_minify(pt->height0, sv->base.u.tex.first_level);
 
 		/* apparently, we need to ignore the t coordinate for 1D textures to fix piglit tex1d-2dborder */
 		if(pt->target == PIPE_TEXTURE_1D)
diff --git a/src/gallium/drivers/nvfx/nvfx_miptree.c b/src/gallium/drivers/nvfx/nvfx_miptree.c
index 7677fde40c..8c043b867b 100644
--- a/src/gallium/drivers/nvfx/nvfx_miptree.c
+++ b/src/gallium/drivers/nvfx/nvfx_miptree.c
@@ -190,25 +190,27 @@ nvfx_miptree_from_handle(struct pipe_screen *pscreen, const struct pipe_resource
 }
 
 struct pipe_surface *
-nvfx_miptree_surface_new(struct pipe_screen *pscreen, struct pipe_resource *pt,
-			 unsigned face, unsigned level, unsigned zslice,
-			 unsigned flags)
+nvfx_miptree_surface_new(struct pipe_context *pipe, struct pipe_resource *pt,
+			 const struct pipe_surface *surf_tmpl)
 {
-	struct nvfx_miptree* mt = (struct nvfx_miptree*)pt;
-	struct nvfx_surface *ns;
-
-	ns = (struct nvfx_surface*)util_surfaces_get(&mt->surfaces, sizeof(struct nvfx_surface), pscreen, pt, face, level, zslice, flags);
-	if(ns->base.base.offset == ~0) {
-		util_dirty_surface_init(&ns->base);
-		ns->pitch = nvfx_subresource_pitch(pt, level);
-		ns->base.base.offset = nvfx_subresource_offset(pt, face, level, zslice);
+	struct nvfx_miptree *mt = (struct nvfx_miptree *)pt;
+	unsigned level = surf_tmpl->u.tex.level;
+	struct nvfx_surface *ns = NULL;
+
+	assert(surf_tmpl->u.tex.first_layer == surf_tmpl->u.tex.last_layer);
+	if(util_surfaces_get(&mt->surfaces, sizeof(struct nvfx_surface), pipe,
+                             pt, level, surf_tmpl->u.tex.first_layer,
+                             surf_tmpl->usage, (struct pipe_surface **)&ns)) {
+                util_dirty_surface_init(&ns->base);
+                ns->pitch = nvfx_subresource_pitch(pt, level);
+                ns->offset = nvfx_subresource_offset(pt, surf_tmpl->u.tex.first_layer, level, surf_tmpl->u.tex.first_layer);
 	}
 
 	return &ns->base.base;
 }
 
 void
-nvfx_miptree_surface_del(struct pipe_surface *ps)
+nvfx_miptree_surface_del(struct pipe_context *pipe, struct pipe_surface *ps)
 {
 	struct nvfx_surface* ns = (struct nvfx_surface*)ps;
 
diff --git a/src/gallium/drivers/nvfx/nvfx_resource.c b/src/gallium/drivers/nvfx/nvfx_resource.c
index 39ae893f1b..c60a7bb8b9 100644
--- a/src/gallium/drivers/nvfx/nvfx_resource.c
+++ b/src/gallium/drivers/nvfx/nvfx_resource.c
@@ -7,7 +7,7 @@
 static unsigned int
 nvfx_resource_is_referenced(struct pipe_context *pipe,
 			    struct pipe_resource *pr,
-			    unsigned face, unsigned level)
+			    unsigned level, int layer)
 {
 	return !!nouveau_reference_flags(nvfx_resource(pr)->bo);
 }
@@ -59,6 +59,9 @@ void
 nvfx_init_resource_functions(struct pipe_context *pipe)
 {
 	pipe->is_resource_referenced = nvfx_resource_is_referenced;
+
+	pipe->create_surface = nvfx_miptree_surface_new;
+	pipe->surface_destroy = nvfx_miptree_surface_del;
 }
 
 void
@@ -69,7 +72,4 @@ nvfx_screen_init_resource_functions(struct pipe_screen *pscreen)
 	pscreen->resource_get_handle = nvfx_resource_get_handle;
 	pscreen->resource_destroy = nvfx_resource_destroy;
 	pscreen->user_buffer_create = nvfx_user_buffer_create;
-
-	pscreen->get_tex_surface = nvfx_miptree_surface_new;
-	pscreen->tex_surface_destroy = nvfx_miptree_surface_del;
 }
diff --git a/src/gallium/drivers/nvfx/nvfx_resource.h b/src/gallium/drivers/nvfx/nvfx_resource.h
index 583be4de2a..070f897944 100644
--- a/src/gallium/drivers/nvfx/nvfx_resource.h
+++ b/src/gallium/drivers/nvfx/nvfx_resource.h
@@ -74,6 +74,7 @@ struct nvfx_miptree {
 struct nvfx_surface {
 	struct util_dirty_surface base;
 	unsigned pitch;
+	unsigned offset;
 
 	struct nvfx_miptree* temp;
 };
@@ -116,12 +117,11 @@ nvfx_miptree_from_handle(struct pipe_screen *pscreen,
 			 struct winsys_handle *whandle);
 
 void
-nvfx_miptree_surface_del(struct pipe_surface *ps);
+nvfx_miptree_surface_del(struct pipe_context *pipe, struct pipe_surface *ps);
 
 struct pipe_surface *
-nvfx_miptree_surface_new(struct pipe_screen *pscreen, struct pipe_resource *pt,
-			 unsigned face, unsigned level, unsigned zslice,
-			 unsigned flags);
+nvfx_miptree_surface_new(struct pipe_context *pipe, struct pipe_resource *pt,
+			 const struct pipe_surface *surf_tmpl);
 
 /* only for miptrees, don't use for buffers */
 
diff --git a/src/gallium/drivers/nvfx/nvfx_screen.c b/src/gallium/drivers/nvfx/nvfx_screen.c
index 8bf0907a08..92e1d33090 100644
--- a/src/gallium/drivers/nvfx/nvfx_screen.c
+++ b/src/gallium/drivers/nvfx/nvfx_screen.c
@@ -37,7 +37,7 @@ nvfx_screen_get_param(struct pipe_screen *pscreen, enum pipe_cap param)
 	case PIPE_CAP_POINT_SPRITE:
 		return 1;
 	case PIPE_CAP_MAX_RENDER_TARGETS:
-		return screen->use_nv4x ? 4 : 2;
+		return screen->use_nv4x ? 4 : 1;
 	case PIPE_CAP_OCCLUSION_QUERY:
 		return 1;
         case PIPE_CAP_TIMER_QUERY:
@@ -123,6 +123,8 @@ nvfx_screen_get_shader_param(struct pipe_screen *pscreen, unsigned shader, enum
 		case PIPE_SHADER_CAP_INDIRECT_TEMP_ADDR:
 		case PIPE_SHADER_CAP_INDIRECT_CONST_ADDR:
 			return 0;
+		case PIPE_SHADER_CAP_SUBROUTINES:
+			return screen->use_nv4x ? 1 : 0;
 		default:
 			break;
 		}
@@ -161,6 +163,8 @@ nvfx_screen_get_shader_param(struct pipe_screen *pscreen, unsigned shader, enum
 			return 0;
 		case PIPE_SHADER_CAP_INDIRECT_CONST_ADDR:
 			return 1;
+		case PIPE_SHADER_CAP_SUBROUTINES:
+			return 1;
 		default:
 			break;
 		}
diff --git a/src/gallium/drivers/nvfx/nvfx_state_fb.c b/src/gallium/drivers/nvfx/nvfx_state_fb.c
index 30e48c8073..816bb89f2c 100644
--- a/src/gallium/drivers/nvfx/nvfx_state_fb.c
+++ b/src/gallium/drivers/nvfx/nvfx_state_fb.c
@@ -7,7 +7,7 @@ nvfx_surface_linear_renderable(struct pipe_surface* surf)
 {
 	/* TODO: precompute this in nvfx_surface creation */
 	return (surf->texture->flags & NVFX_RESOURCE_FLAG_LINEAR)
-		&& !(surf->offset & 63)
+		&& !(((struct nvfx_surface*)surf)->offset & 63)
 		&& !(((struct nvfx_surface*)surf)->pitch & 63);
 }
 
@@ -16,8 +16,8 @@ nvfx_surface_swizzled_renderable(struct pipe_framebuffer_state* fb, struct pipe_
 {
 	/* TODO: precompute this in nvfx_surface creation */
 	return !((struct nvfx_miptree*)surf->texture)->linear_pitch
-		&& (surf->texture->target != PIPE_TEXTURE_3D || u_minify(surf->texture->depth0, surf->level) <= 1)
-		&& !(surf->offset & 127)
+		&& (surf->texture->target != PIPE_TEXTURE_3D || u_minify(surf->texture->depth0, surf->u.tex.level) <= 1)
+		&& !(((struct nvfx_surface*)surf)->offset & 127)
 		&& (surf->width == fb->width)
 		&& (surf->height == fb->height)
 		&& !((struct nvfx_surface*)surf)->temp
@@ -31,7 +31,7 @@ nvfx_surface_get_render_target(struct pipe_surface* surf, int all_swizzled, stru
 	if(!ns->temp)
 	{
 		target->bo = ((struct nvfx_miptree*)surf->texture)->base.bo;
-		target->offset = surf->offset;
+		target->offset = ns->offset;
 		target->pitch = align(ns->pitch, 64);
 		assert(target->pitch);
 		return FALSE;
@@ -54,7 +54,7 @@ nvfx_framebuffer_prepare(struct nvfx_context *nvfx)
 	int all_swizzled = 1;
 
 	if(!nvfx->is_nv4x)
-		assert(fb->nr_cbufs <= 2);
+		assert(fb->nr_cbufs <= 1);
 	else
 		assert(fb->nr_cbufs <= 4);
 
@@ -113,7 +113,9 @@ nvfx_framebuffer_validate(struct nvfx_context *nvfx, unsigned prepare_result)
 		nvfx->state.render_temps |= nvfx_surface_get_render_target(fb->cbufs[i], prepare_result, &nvfx->hw_rt[i]) << i;
 
 	for(; i < 4; ++i)
-		nvfx->hw_rt[i].bo = 0;
+		nvfx->hw_rt[i].bo = NULL;
+
+	nvfx->hw_zeta.bo = NULL;
 
 	if (fb->zsbuf) {
 		nvfx->state.render_temps |= nvfx_surface_get_render_target(fb->zsbuf, prepare_result, &nvfx->hw_zeta) << 7;
diff --git a/src/gallium/drivers/nvfx/nvfx_surface.c b/src/gallium/drivers/nvfx/nvfx_surface.c
index 70adebc1be..6fd6c47081 100644
--- a/src/gallium/drivers/nvfx/nvfx_surface.c
+++ b/src/gallium/drivers/nvfx/nvfx_surface.c
@@ -99,17 +99,17 @@ nvfx_region_init_for_surface(struct nv04_region* rgn, struct nvfx_surface* surf,
 			util_dirty_surface_set_dirty(nvfx_surface_get_dirty_surfaces(&surf->base.base), &surf->base);
 	} else {
 		rgn->bo = ((struct nvfx_resource*)surf->base.base.texture)->bo;
-		rgn->offset = surf->base.base.offset;
+		rgn->offset = surf->offset;
 
 		if(surf->base.base.texture->flags & NVFX_RESOURCE_FLAG_LINEAR)
 			rgn->pitch = surf->pitch;
 	        else
 	        {
 		        rgn->pitch = 0;
-		        rgn->z = surf->base.base.zslice;
+		        rgn->z = surf->base.base.u.tex.first_layer;
 		        rgn->w = surf->base.base.width;
 		        rgn->h = surf->base.base.height;
-		        rgn->d = u_minify(surf->base.base.texture->depth0, surf->base.base.level);
+		        rgn->d = u_minify(surf->base.base.texture->depth0, surf->base.base.u.tex.level);
 	        }
 	}
 
@@ -119,11 +119,11 @@ nvfx_region_init_for_surface(struct nv04_region* rgn, struct nvfx_surface* surf,
 }
 
 static INLINE void
-nvfx_region_init_for_subresource(struct nv04_region* rgn, struct pipe_resource* pt, struct pipe_subresource sub, unsigned x, unsigned y, unsigned z, bool for_write)
+nvfx_region_init_for_subresource(struct nv04_region* rgn, struct pipe_resource* pt, unsigned level, unsigned x, unsigned y, unsigned z, bool for_write)
 {
 	if(pt->target != PIPE_BUFFER)
 	{
-		struct nvfx_surface* ns = (struct nvfx_surface*)util_surfaces_peek(&((struct nvfx_miptree*)pt)->surfaces, pt, sub.face, sub.level, z);
+		struct nvfx_surface* ns = (struct nvfx_surface*)util_surfaces_peek(&((struct nvfx_miptree*)pt)->surfaces, pt, level, z);
 		if(ns && util_dirty_surface_is_dirty(&ns->base))
 		{
 			nvfx_region_init_for_surface(rgn, ns, x, y, for_write);
@@ -132,22 +132,22 @@ nvfx_region_init_for_subresource(struct nv04_region* rgn, struct pipe_resource*
 	}
 
 	rgn->bo = ((struct nvfx_resource*)pt)->bo;
-	rgn->offset = nvfx_subresource_offset(pt, sub.face, sub.level, z);
+	rgn->offset = nvfx_subresource_offset(pt, z, level, z);
 	rgn->x = x;
 	rgn->y = y;
 
 	if(pt->flags & NVFX_RESOURCE_FLAG_LINEAR)
 	{
-		rgn->pitch = nvfx_subresource_pitch(pt, sub.level);
+		rgn->pitch = nvfx_subresource_pitch(pt, level);
 		rgn->z = 0;
 	}
 	else
 	{
 		rgn->pitch = 0;
 		rgn->z = z;
-		rgn->w = u_minify(pt->width0, sub.level);
-		rgn->h = u_minify(pt->height0, sub.level);
-		rgn->d = u_minify(pt->depth0, sub.level);
+		rgn->w = u_minify(pt->width0, level);
+		rgn->h = u_minify(pt->height0, level);
+		rgn->d = u_minify(pt->depth0, level);
 	}
 
 	nvfx_region_set_format(rgn, pt->format);
@@ -234,11 +234,10 @@ nvfx_region_clone(struct nv04_2d_context* ctx, struct nv04_region* rgn, unsigned
 
 static void
 nvfx_resource_copy_region(struct pipe_context *pipe,
-		  struct pipe_resource *dstr, struct pipe_subresource subdst,
-		  unsigned dstx, unsigned dsty, unsigned dstz,
-		  struct pipe_resource *srcr, struct pipe_subresource subsrc,
-		  unsigned srcx, unsigned srcy, unsigned srcz,
-		  unsigned w, unsigned h)
+			  struct pipe_resource *dstr, unsigned dst_level,
+			  unsigned dstx, unsigned dsty, unsigned dstz,
+			  struct pipe_resource *srcr, unsigned src_level,
+			  const struct pipe_box *src_box)
 {
 	static int copy_threshold = -1;
 	struct nv04_2d_context *ctx = nvfx_screen(pipe->screen)->eng2d;
@@ -247,6 +246,8 @@ nvfx_resource_copy_region(struct pipe_context *pipe,
 	int src_on_gpu;
 	boolean small;
 	int ret;
+	unsigned w = src_box->width;
+	unsigned h = src_box->height;
 
 	if(!w || !h)
 		return;
@@ -257,8 +258,8 @@ nvfx_resource_copy_region(struct pipe_context *pipe,
 	dst_to_gpu = dstr->usage != PIPE_USAGE_DYNAMIC && dstr->usage != PIPE_USAGE_STAGING;
 	src_on_gpu = nvfx_resource_on_gpu(srcr);
 
-	nvfx_region_init_for_subresource(&dst, dstr, subdst, dstx, dsty, dstz, TRUE);
-	nvfx_region_init_for_subresource(&src, srcr, subsrc, srcx, srcy, srcz, FALSE);
+	nvfx_region_init_for_subresource(&dst, dstr, dst_level, dstx, dsty, dstz, TRUE);
+	nvfx_region_init_for_subresource(&src, srcr, src_level, src_box->x, src_box->y, src_box->z, FALSE);
 	w = util_format_get_stride(dstr->format, w) >> dst.bpps;
 	h = util_format_get_nblocksy(dstr->format, h);
 
@@ -279,7 +280,7 @@ nvfx_resource_copy_region(struct pipe_context *pipe,
 		 * TODO: perhaps support reinterpreting the formats
 		 */
 		struct blitter_context* blitter = nvfx_get_blitter(pipe, 1);
-		util_blitter_copy_region(blitter, dstr, subdst, dstx, dsty, dstz, srcr, subsrc, srcx, srcy, srcz, w, h, TRUE);
+		util_blitter_copy_region(blitter, dstr, dst_level, dstx, dsty, dstz, srcr, src_level, src_box, TRUE);
 		nvfx_put_blitter(pipe, blitter);
 	}
 	else
@@ -371,7 +372,7 @@ static void
 nvfx_surface_copy_temp(struct pipe_context* pipe, struct pipe_surface* surf, int to_temp)
 {
 	struct nvfx_surface* ns = (struct nvfx_surface*)surf;
-	struct pipe_subresource tempsr, surfsr;
+	struct pipe_box box;
 	struct nvfx_context* nvfx = nvfx_context(pipe);
 	struct nvfx_miptree* temp;
 	unsigned use_vertex_buffers;
@@ -387,15 +388,20 @@ nvfx_surface_copy_temp(struct pipe_context* pipe, struct pipe_surface* surf, int
 	use_index_buffer = nvfx->use_index_buffer;
 	base_vertex = nvfx->base_vertex;
 
-	tempsr.face = 0;
-	tempsr.level = 0;
-	surfsr.face = surf->face;
-	surfsr.level = surf->level;
+	box.x = box.y = 0;
+	assert(surf->u.tex.first_layer == surf->u.tex.last_layer);
+	box.width = surf->width;
+	box.height = surf->height;
+	box.depth = 1;
 
-	if(to_temp)
-		nvfx_resource_copy_region(pipe, &temp->base.base, tempsr, 0, 0, 0, surf->texture, surfsr, 0, 0, surf->zslice, surf->width, surf->height);
-	else
-		nvfx_resource_copy_region(pipe, surf->texture, surfsr, 0, 0, surf->zslice, &temp->base.base, tempsr, 0, 0, 0, surf->width, surf->height);
+	if(to_temp) {
+	        box.z = surf->u.tex.first_layer;
+		nvfx_resource_copy_region(pipe, &temp->base.base, 0, 0, 0, 0, surf->texture, surf->u.tex.level, &box);
+	}
+	else {
+		box.z = 0;
+		nvfx_resource_copy_region(pipe, surf->texture, surf->u.tex.level, 0, 0, surf->u.tex.first_layer, &temp->base.base, 0, &box);
+	}
 
 	/* If this triggers, it probably means we attempted to use the blitter
 	 * but failed due to non-renderability of the target.
diff --git a/src/gallium/drivers/nvfx/nvfx_transfer.c b/src/gallium/drivers/nvfx/nvfx_transfer.c
index 7cb47a20f6..2debcb6eb8 100644
--- a/src/gallium/drivers/nvfx/nvfx_transfer.c
+++ b/src/gallium/drivers/nvfx/nvfx_transfer.c
@@ -21,10 +21,10 @@ struct nvfx_staging_transfer
 
 struct pipe_transfer *
 nvfx_transfer_new(struct pipe_context *pipe,
-			  struct pipe_resource *pt,
-			  struct pipe_subresource sr,
-			  unsigned usage,
-			  const struct pipe_box *box)
+		  struct pipe_resource *pt,
+		  unsigned level,
+		  unsigned usage,
+		  const struct pipe_box *box)
 {
         if((usage & (PIPE_TRANSFER_UNSYNCHRONIZED | PIPE_TRANSFER_DONTBLOCK)) == PIPE_TRANSFER_DONTBLOCK)
         {
@@ -44,11 +44,11 @@ nvfx_transfer_new(struct pipe_context *pipe,
 			return NULL;
 
 		pipe_resource_reference(&tx->resource, pt);
-		tx->sr = sr;
+		tx->level = level;
 		tx->usage = usage;
 		tx->box = *box;
 
-		tx->slice_stride = tx->stride = util_format_get_stride(pt->format, box->width);
+		tx->layer_stride = tx->stride = util_format_get_stride(pt->format, box->width);
 		tx->data = buffer->data + util_format_get_stride(pt->format, box->x);
 
 		return tx;
@@ -62,20 +62,20 @@ nvfx_transfer_new(struct pipe_context *pipe,
 	        if(!tx)
 	        	return NULL;
 
-	        util_staging_transfer_init(pipe, pt, sr, usage, box, direct, &tx->base);
+	        util_staging_transfer_init(pipe, pt, level, usage, box, direct, &tx->base);
 
 		if(direct)
 		{
-			tx->base.base.stride = nvfx_subresource_pitch(pt, sr.level);
-			tx->base.base.slice_stride = tx->base.base.stride * u_minify(pt->height0, sr.level);
-			tx->offset = nvfx_subresource_offset(pt, sr.face, sr.level, box->z)
+			tx->base.base.stride = nvfx_subresource_pitch(pt, level);
+			tx->base.base.layer_stride = tx->base.base.stride * u_minify(pt->height0, level);
+			tx->offset = nvfx_subresource_offset(pt, box->z, level, box->z)
 				+ util_format_get_2d_size(pt->format, tx->base.base.stride, box->y)
 				+ util_format_get_stride(pt->format, box->x);
 		}
 		else
 		{
 			tx->base.base.stride = nvfx_subresource_pitch(tx->base.staging_resource, 0);
-			tx->base.base.slice_stride = tx->base.base.stride * tx->base.staging_resource->height0;
+			tx->base.base.layer_stride = tx->base.base.stride * tx->base.staging_resource->height0;
 			tx->offset = 0;
 		}
 
@@ -187,7 +187,7 @@ nvfx_transfer_unmap(struct pipe_context *pipe, struct pipe_transfer *ptx)
 
 static void nvfx_transfer_inline_write( struct pipe_context *pipe,
 				      struct pipe_resource *pr,
-				      struct pipe_subresource sr,
+				      unsigned level,
 				      unsigned usage,
 				      const struct pipe_box *box,
 				      const void *data,
@@ -196,7 +196,7 @@ static void nvfx_transfer_inline_write( struct pipe_context *pipe,
 {
 	if(pr->target != PIPE_BUFFER)
 	{
-		u_default_transfer_inline_write(pipe, pr, sr, usage, box, data, stride, slice_stride);
+		u_default_transfer_inline_write(pipe, pr, level, usage, box, data, stride, slice_stride);
 	}
 	else
 	{
diff --git a/src/gallium/drivers/nvfx/nvfx_transfer.h b/src/gallium/drivers/nvfx/nvfx_transfer.h
index 20f20d5b0b..682f428b79 100644
--- a/src/gallium/drivers/nvfx/nvfx_transfer.h
+++ b/src/gallium/drivers/nvfx/nvfx_transfer.h
@@ -9,7 +9,7 @@
 struct pipe_transfer *
 nvfx_transfer_new(struct pipe_context *pcontext,
 			  struct pipe_resource *pt,
-			  struct pipe_subresource sr,
+			  unsigned level,
 			  unsigned usage,
 			  const struct pipe_box *box);
 
diff --git a/src/gallium/drivers/r300/r300_blit.c b/src/gallium/drivers/r300/r300_blit.c
index 0ac4e4c6f1..6e886433bc 100644
--- a/src/gallium/drivers/r300/r300_blit.c
+++ b/src/gallium/drivers/r300/r300_blit.c
@@ -186,12 +186,12 @@ static void r300_clear(struct pipe_context* pipe,
             r300_depth_clear_value(fb->zsbuf->format, depth, stencil);
 
         r300_mark_fb_state_dirty(r300, R300_CHANGED_ZCLEAR_FLAG);
-        if (zstex->zmask_mem[fb->zsbuf->level]) {
-            r300->zmask_clear.dirty = TRUE;
+        if (zstex->zmask_mem[fb->zsbuf->u.tex.level]) {
+            r300_mark_atom_dirty(r300, &r300->zmask_clear);
             buffers &= ~PIPE_CLEAR_DEPTHSTENCIL;
         }
-        if (zstex->hiz_mem[fb->zsbuf->level])
-            r300->hiz_clear.dirty = TRUE;
+        if (zstex->hiz_mem[fb->zsbuf->u.tex.level])
+            r300_mark_atom_dirty(r300, &r300->hiz_clear);
     }
 
     /* Enable CBZB clear. */
@@ -230,7 +230,7 @@ static void r300_clear(struct pipe_context* pipe,
                  r300_get_num_cs_end_dwords(r300);
 
         /* Reserve CS space. */
-        if (dwords > (r300->cs->ndw - r300->cs->cdw)) {
+        if (dwords > (R300_MAX_CMDBUF_DWORDS - r300->cs->cdw)) {
             r300->context.flush(&r300->context, 0, NULL);
         }
 
@@ -259,9 +259,9 @@ static void r300_clear(struct pipe_context* pipe,
      * If we cleared zmask/hiz, it's in use now. The Hyper-Z state update
      * looks if zmask/hiz is in use and enables fastfill accordingly. */
     if (zstex &&
-        (zstex->zmask_in_use[fb->zsbuf->level] ||
-         zstex->hiz_in_use[fb->zsbuf->level])) {
-        r300->hyperz_state.dirty = TRUE;
+        (zstex->zmask_in_use[fb->zsbuf->u.tex.level] ||
+         zstex->hiz_in_use[fb->zsbuf->u.tex.level])) {
+        r300_mark_atom_dirty(r300, &r300->hyperz_state);
     }
 }
 
@@ -300,58 +300,61 @@ static void r300_clear_depth_stencil(struct pipe_context *pipe,
 /* Flush a depth stencil buffer. */
 void r300_flush_depth_stencil(struct pipe_context *pipe,
                               struct pipe_resource *dst,
-                              struct pipe_subresource subdst,
-                              unsigned zslice)
+                              unsigned level,
+                              unsigned layer)
 {
     struct r300_context *r300 = r300_context(pipe);
-    struct pipe_surface *dstsurf;
+    struct pipe_surface *dstsurf, surf_tmpl;
     struct r300_texture *tex = r300_texture(dst);
 
-    if (!tex->zmask_mem[subdst.level])
+    if (!tex->zmask_mem[level])
         return;
-    if (!tex->zmask_in_use[subdst.level])
+    if (!tex->zmask_in_use[level])
         return;
 
-    dstsurf = pipe->screen->get_tex_surface(pipe->screen, dst,
-                                            subdst.face, subdst.level, zslice,
-                                            PIPE_BIND_DEPTH_STENCIL);
+    surf_tmpl.format = dst->format;
+    surf_tmpl.usage = PIPE_BIND_DEPTH_STENCIL;
+    surf_tmpl.u.tex.level = level;
+    surf_tmpl.u.tex.first_layer = layer;
+    surf_tmpl.u.tex.last_layer = layer;
+    dstsurf = pipe->create_surface(pipe, dst, &surf_tmpl);
+
     r300->z_decomp_rd = TRUE;
     r300_blitter_begin(r300, R300_CLEAR_SURFACE);
     util_blitter_flush_depth_stencil(r300->blitter, dstsurf);
     r300_blitter_end(r300);
     r300->z_decomp_rd = FALSE;
 
-    tex->zmask_in_use[subdst.level] = FALSE;
+    tex->zmask_in_use[level] = FALSE;
 }
 
 /* Copy a block of pixels from one surface to another using HW. */
 static void r300_hw_copy_region(struct pipe_context* pipe,
                                 struct pipe_resource *dst,
-                                struct pipe_subresource subdst,
+                                unsigned dst_level,
                                 unsigned dstx, unsigned dsty, unsigned dstz,
                                 struct pipe_resource *src,
-                                struct pipe_subresource subsrc,
-                                unsigned srcx, unsigned srcy, unsigned srcz,
-                                unsigned width, unsigned height)
+                                unsigned src_level,
+                                const struct pipe_box *src_box)
 {
     struct r300_context* r300 = r300_context(pipe);
 
     r300_blitter_begin(r300, R300_COPY);
-    util_blitter_copy_region(r300->blitter, dst, subdst, dstx, dsty, dstz,
-                             src, subsrc, srcx, srcy, srcz, width, height,
-                             TRUE);
+
+    /* Do a copy */
+    util_blitter_copy_region(r300->blitter, dst, dst_level, dstx, dsty, dstz,
+                             src, src_level, src_box, TRUE);
     r300_blitter_end(r300);
 }
 
 /* Copy a block of pixels from one surface to another. */
 static void r300_resource_copy_region(struct pipe_context *pipe,
                                       struct pipe_resource *dst,
-                                      struct pipe_subresource subdst,
+                                      unsigned dst_level,
                                       unsigned dstx, unsigned dsty, unsigned dstz,
                                       struct pipe_resource *src,
-                                      struct pipe_subresource subsrc,
-                                      unsigned srcx, unsigned srcy, unsigned srcz,
-                                      unsigned width, unsigned height)
+                                      unsigned src_level,
+                                      const struct pipe_box *src_box)
 {
     enum pipe_format old_format = dst->format;
     enum pipe_format new_format = old_format;
@@ -384,7 +387,7 @@ static void r300_resource_copy_region(struct pipe_context *pipe,
 
     is_depth = util_format_get_component_bits(src->format, UTIL_FORMAT_COLORSPACE_ZS, 0) != 0;
     if (is_depth) {
-        r300_flush_depth_stencil(pipe, src, subsrc, srcz);
+        r300_flush_depth_stencil(pipe, src, src_level, src_box->z);
     }
     if (old_format != new_format) {
         r300_texture_reinterpret_format(pipe->screen,
@@ -393,8 +396,8 @@ static void r300_resource_copy_region(struct pipe_context *pipe,
                                         src, new_format);
     }
 
-    r300_hw_copy_region(pipe, dst, subdst, dstx, dsty, dstz,
-                        src, subsrc, srcx, srcy, srcz, width, height);
+    r300_hw_copy_region(pipe, dst, dst_level, dstx, dsty, dstz,
+                        src, src_level, src_box);
 
     if (old_format != new_format) {
         r300_texture_reinterpret_format(pipe->screen,
diff --git a/src/gallium/drivers/r300/r300_chipset.c b/src/gallium/drivers/r300/r300_chipset.c
index 48c2409211..583e981a4d 100644
--- a/src/gallium/drivers/r300/r300_chipset.c
+++ b/src/gallium/drivers/r300/r300_chipset.c
@@ -424,4 +424,5 @@ void r300_parse_chipset(struct r300_capabilities* caps)
     }
 
     caps->is_rv350 = caps->family >= CHIP_FAMILY_RV350;
+    caps->dxtc_swizzle = caps->is_r400 || caps->is_r500;
 }
diff --git a/src/gallium/drivers/r300/r300_chipset.h b/src/gallium/drivers/r300/r300_chipset.h
index e7ca642b4f..f2035d2009 100644
--- a/src/gallium/drivers/r300/r300_chipset.h
+++ b/src/gallium/drivers/r300/r300_chipset.h
@@ -79,6 +79,10 @@ struct r300_capabilities {
     boolean is_r500;
     /* Whether or not the second pixel pipe is accessed with the high bit */
     boolean high_second_pipe;
+    /* DXTC texture swizzling. */
+    boolean dxtc_swizzle;
+    /* Index bias (AKA index offset). */
+    boolean index_bias_supported;
 };
 
 /* Enumerations for legibility and telling which card we're running on. */
diff --git a/src/gallium/drivers/r300/r300_context.c b/src/gallium/drivers/r300/r300_context.c
index fb099e2a7d..67b011a145 100644
--- a/src/gallium/drivers/r300/r300_context.c
+++ b/src/gallium/drivers/r300/r300_context.c
@@ -44,14 +44,14 @@ static void r300_update_num_contexts(struct r300_screen *r300screen,
         p_atomic_inc(&r300screen->num_contexts);
 
         if (r300screen->num_contexts > 1)
-            util_mempool_set_thread_safety(&r300screen->pool_buffers,
-                                           UTIL_MEMPOOL_MULTITHREADED);
+            util_slab_set_thread_safety(&r300screen->pool_buffers,
+                                        UTIL_SLAB_MULTITHREADED);
     } else {
         p_atomic_dec(&r300screen->num_contexts);
 
         if (r300screen->num_contexts <= 1)
-            util_mempool_set_thread_safety(&r300screen->pool_buffers,
-                                           UTIL_MEMPOOL_SINGLETHREADED);
+            util_slab_set_thread_safety(&r300screen->pool_buffers,
+                                        UTIL_SLAB_SINGLETHREADED);
     }
 }
 
@@ -79,6 +79,9 @@ static void r300_release_referenced_objects(struct r300_context *r300)
                 NULL);
     }
 
+    /* The dummy VBO. */
+    pipe_resource_reference(&r300->dummy_vb, NULL);
+
     /* The SWTCL VBO. */
     pipe_resource_reference(&r300->vbo, NULL);
 
@@ -97,23 +100,12 @@ static void r300_release_referenced_objects(struct r300_context *r300)
 static void r300_destroy_context(struct pipe_context* context)
 {
     struct r300_context* r300 = r300_context(context);
-    struct r300_atom *atom;
 
     if (r300->blitter)
         util_blitter_destroy(r300->blitter);
     if (r300->draw)
         draw_destroy(r300->draw);
 
-    /* Print stats, if enabled. */
-    if (SCREEN_DBG_ON(r300->screen, DBG_STATS)) {
-        fprintf(stderr, "r300: Stats for context %p:\n", r300);
-        fprintf(stderr, "    : Flushes: %" PRIu64 "\n", r300->flush_counter);
-        foreach(atom, &r300->atom_list) {
-            fprintf(stderr, "    : %s: %" PRIu64 " emits\n",
-                atom->name, atom->counter);
-        }
-    }
-
     if (r300->upload_vb)
         u_upload_destroy(r300->upload_vb);
     if (r300->upload_ib)
@@ -132,7 +124,7 @@ static void r300_destroy_context(struct pipe_context* context)
         r300->rws->cs_destroy(r300->cs);
 
     /* XXX: No way to tell if this was initialized or not? */
-    util_mempool_destroy(&r300->pool_transfers);
+    util_slab_destroy(&r300->pool_transfers);
 
     r300_update_num_contexts(r300->screen, -1);
 
@@ -174,10 +166,16 @@ void r300_flush_cb(void *data)
     r300->atomname.size = atomsize; \
     r300->atomname.emit = r300_emit_##atomname; \
     r300->atomname.dirty = FALSE; \
-    insert_at_tail(&r300->atom_list, &r300->atomname); \
  } while (0)
 
-static void r300_setup_atoms(struct r300_context* r300)
+#define R300_ALLOC_ATOM(atomname, statetype) \
+do { \
+    r300->atomname.state = CALLOC_STRUCT(statetype); \
+    if (r300->atomname.state == NULL) \
+        return FALSE; \
+} while (0)
+
+static boolean r300_setup_atoms(struct r300_context* r300)
 {
     boolean is_rv350 = r300->screen->caps.is_rv350;
     boolean is_r500 = r300->screen->caps.is_r500;
@@ -189,9 +187,6 @@ static void r300_setup_atoms(struct r300_context* r300)
 
     /* Create the actual atom list.
      *
-     * Each atom is examined and emitted in the order it appears here, which
-     * can affect performance and conformance if not handled with care.
-     *
      * Some atoms never change size, others change every emit - those have
      * the size of 0 here.
      *
@@ -203,7 +198,6 @@ static void r300_setup_atoms(struct r300_context* r300)
      * - fb_state_pipelined (pipelined regs)
      * The motivation behind this is to be able to emit a strict
      * subset of the regs, and to have reasonable register ordering. */
-    make_empty_list(&r300->atom_list);
     /* SC, GB (unpipelined), RB3D (unpipelined), ZB (unpipelined). */
     R300_INIT_ATOM(gpu_flush, 9);
     R300_INIT_ATOM(aa_state, 4);
@@ -258,23 +252,23 @@ static void r300_setup_atoms(struct r300_context* r300)
     }
 
     /* Some non-CSO atoms need explicit space to store the state locally. */
-    r300->aa_state.state = CALLOC_STRUCT(r300_aa_state);
-    r300->blend_color_state.state = CALLOC_STRUCT(r300_blend_color_state);
-    r300->clip_state.state = CALLOC_STRUCT(r300_clip_state);
-    r300->fb_state.state = CALLOC_STRUCT(pipe_framebuffer_state);
-    r300->gpu_flush.state = CALLOC_STRUCT(pipe_framebuffer_state);
-    r300->hyperz_state.state = CALLOC_STRUCT(r300_hyperz_state);
-    r300->invariant_state.state = CALLOC_STRUCT(r300_invariant_state);
-    r300->rs_block_state.state = CALLOC_STRUCT(r300_rs_block);
-    r300->scissor_state.state = CALLOC_STRUCT(pipe_scissor_state);
-    r300->textures_state.state = CALLOC_STRUCT(r300_textures_state);
-    r300->vap_invariant_state.state = CALLOC_STRUCT(r300_vap_invariant_state);
-    r300->viewport_state.state = CALLOC_STRUCT(r300_viewport_state);
-    r300->ztop_state.state = CALLOC_STRUCT(r300_ztop_state);
-    r300->fs_constants.state = CALLOC_STRUCT(r300_constant_buffer);
-    r300->vs_constants.state = CALLOC_STRUCT(r300_constant_buffer);
+    R300_ALLOC_ATOM(aa_state, r300_aa_state);
+    R300_ALLOC_ATOM(blend_color_state, r300_blend_color_state);
+    R300_ALLOC_ATOM(clip_state, r300_clip_state);
+    R300_ALLOC_ATOM(hyperz_state, r300_hyperz_state);
+    R300_ALLOC_ATOM(invariant_state, r300_invariant_state);
+    R300_ALLOC_ATOM(textures_state, r300_textures_state);
+    R300_ALLOC_ATOM(vap_invariant_state, r300_vap_invariant_state);
+    R300_ALLOC_ATOM(viewport_state, r300_viewport_state);
+    R300_ALLOC_ATOM(ztop_state, r300_ztop_state);
+    R300_ALLOC_ATOM(fb_state, pipe_framebuffer_state);
+    R300_ALLOC_ATOM(gpu_flush, pipe_framebuffer_state);
+    R300_ALLOC_ATOM(scissor_state, pipe_scissor_state);
+    R300_ALLOC_ATOM(rs_block_state, r300_rs_block);
+    R300_ALLOC_ATOM(fs_constants, r300_constant_buffer);
+    R300_ALLOC_ATOM(vs_constants, r300_constant_buffer);
     if (!r300->screen->caps.has_tcl) {
-        r300->vertex_stream_state.state = CALLOC_STRUCT(r300_vertex_stream_state);
+        R300_ALLOC_ATOM(vertex_stream_state, r300_vertex_stream_state);
     }
 
     /* Some non-CSO atoms don't use the state pointer. */
@@ -286,11 +280,13 @@ static void r300_setup_atoms(struct r300_context* r300)
 
     /* Some states must be marked as dirty here to properly set up
      * hardware in the first command stream. */
-    r300->invariant_state.dirty = TRUE;
-    r300->pvs_flush.dirty = TRUE;
-    r300->vap_invariant_state.dirty = TRUE;
-    r300->texture_cache_inval.dirty = TRUE;
-    r300->textures_state.dirty = TRUE;
+    r300_mark_atom_dirty(r300, &r300->invariant_state);
+    r300_mark_atom_dirty(r300, &r300->pvs_flush);
+    r300_mark_atom_dirty(r300, &r300->vap_invariant_state);
+    r300_mark_atom_dirty(r300, &r300->texture_cache_inval);
+    r300_mark_atom_dirty(r300, &r300->textures_state);
+
+    return TRUE;
 }
 
 /* Not every state tracker calls every driver function before the first draw
@@ -316,7 +312,7 @@ static void r300_init_states(struct pipe_context *pipe)
     pipe->set_scissor_state(pipe, &ss);
 
     /* Initialize the clip state. */
-    if (r300_context(pipe)->screen->caps.has_tcl) {
+    if (r300->screen->caps.has_tcl) {
         pipe->set_clip_state(pipe, &cs);
     } else {
         BEGIN_CB(clip->cb, 2);
@@ -418,9 +414,9 @@ struct pipe_context* r300_create_context(struct pipe_screen* screen,
 
     make_empty_list(&r300->query_list);
 
-    util_mempool_create(&r300->pool_transfers,
-                        sizeof(struct pipe_transfer), 64,
-                        UTIL_MEMPOOL_SINGLETHREADED);
+    util_slab_create(&r300->pool_transfers,
+                     sizeof(struct pipe_transfer), 64,
+                     UTIL_SLAB_SINGLETHREADED);
 
     r300->cs = rws->cs_create(rws);
     if (r300->cs == NULL)
@@ -429,6 +425,8 @@ struct pipe_context* r300_create_context(struct pipe_screen* screen,
     if (!r300screen->caps.has_tcl) {
         /* Create a Draw. This is used for SW TCL. */
         r300->draw = draw_create(&r300->context);
+        if (r300->draw == NULL)
+            goto fail;
         /* Enable our renderer. */
         draw_set_rasterize_stage(r300->draw, r300_draw_stage(r300));
         /* Disable converting points/lines to triangles. */
@@ -436,7 +434,8 @@ struct pipe_context* r300_create_context(struct pipe_screen* screen,
         draw_wide_point_threshold(r300->draw, 10000000.f);
     }
 
-    r300_setup_atoms(r300);
+    if (!r300_setup_atoms(r300))
+        goto fail;
 
     r300_init_blit_functions(r300);
     r300_init_flush_functions(r300);
@@ -488,6 +487,7 @@ struct pipe_context* r300_create_context(struct pipe_screen* screen,
         rtempl.target = PIPE_TEXTURE_2D;
         rtempl.format = PIPE_FORMAT_I8_UNORM;
         rtempl.bind = PIPE_BIND_SAMPLER_VIEW;
+        rtempl.usage = PIPE_USAGE_IMMUTABLE;
         rtempl.width0 = 1;
         rtempl.height0 = 1;
         rtempl.depth0 = 1;
@@ -501,6 +501,19 @@ struct pipe_context* r300_create_context(struct pipe_screen* screen,
         pipe_resource_reference(&tex, NULL);
     }
 
+    {
+        struct pipe_resource vb = {};
+        vb.target = PIPE_BUFFER;
+        vb.format = PIPE_FORMAT_R8_UNORM;
+        vb.bind = PIPE_BIND_VERTEX_BUFFER;
+        vb.usage = PIPE_USAGE_IMMUTABLE;
+        vb.width0 = sizeof(float) * 16;
+        vb.height0 = 1;
+        vb.depth0 = 1;
+
+        r300->dummy_vb = screen->resource_create(screen, &vb);
+    }
+
     return &r300->context;
 
  fail:
diff --git a/src/gallium/drivers/r300/r300_context.h b/src/gallium/drivers/r300/r300_context.h
index b59bc00261..39dcde0610 100644
--- a/src/gallium/drivers/r300/r300_context.h
+++ b/src/gallium/drivers/r300/r300_context.h
@@ -43,12 +43,8 @@ struct r300_vertex_shader;
 struct r300_stencilref_context;
 
 struct r300_atom {
-    /* List pointers. */
-    struct r300_atom *prev, *next;
     /* Name, for debugging. */
     const char* name;
-    /* Stat counter. */
-    uint64_t counter;
     /* Opaque state. */
     void* state;
     /* Emit the state to the context. */
@@ -258,6 +254,8 @@ struct r300_constant_buffer {
     uint32_t *ptr;
     /* Remapping table. */
     unsigned *remap_table;
+    /* const buffer base */
+    uint32_t buffer_base;
 };
 
 /* Query object.
@@ -282,6 +280,7 @@ struct r300_query {
 
     /* The buffer where query results are stored. */
     struct r300_winsys_buffer *buffer;
+    struct r300_winsys_cs_buffer *cs_buffer;
     /* The size of the buffer. */
     unsigned buffer_size;
     /* The domain of the buffer. */
@@ -313,6 +312,7 @@ struct r300_surface {
 
     /* Winsys buffer backing the texture. */
     struct r300_winsys_buffer *buffer;
+    struct r300_winsys_cs_buffer *cs_buffer;
 
     enum r300_buffer_domain domain;
 
@@ -396,6 +396,7 @@ struct r300_texture {
 
     /* Pipe buffer backing this texture. */
     struct r300_winsys_buffer *buffer;
+    struct r300_winsys_cs_buffer *cs_buffer;
 
     /* Registers carrying texture format data. */
     /* Only format-independent bits should be filled in. */
@@ -480,6 +481,10 @@ struct r300_context {
      * dummy texture there. */
     struct r300_sampler_view *texkill_sampler;
 
+    /* When no vertex buffer is set, this one is used instead to prevent
+     * hardlocks. */
+    struct pipe_resource *dummy_vb;
+
     /* The currently active query. */
     struct r300_query *query_current;
     /* The saved query for blitter operations. */
@@ -488,65 +493,68 @@ struct r300_context {
     struct r300_query query_list;
 
     /* Various CSO state objects. */
-    /* Beginning of atom list. */
-    struct r300_atom atom_list;
+
+    /* Each atom is emitted in the order it appears here, which can affect
+     * performance and stability if not handled with care. */
+    /* GPU flush. */
+    struct r300_atom gpu_flush;
     /* Anti-aliasing (MSAA) state. */
     struct r300_atom aa_state;
+    /* Framebuffer state. */
+    struct r300_atom fb_state;
+    /* HyperZ state (various SC/ZB bits). */
+    struct r300_atom hyperz_state;
+    /* ZTOP state. */
+    struct r300_atom ztop_state;
+    /* Depth, stencil, and alpha state. */
+    struct r300_atom dsa_state;
     /* Blend state. */
     struct r300_atom blend_state;
     /* Blend color state. */
     struct r300_atom blend_color_state;
+    /* Scissor state. */
+    struct r300_atom scissor_state;
+    /* Invariant state. This must be emitted to get the engine started. */
+    struct r300_atom invariant_state;
+    /* Viewport state. */
+    struct r300_atom viewport_state;
+    /* PVS flush. */
+    struct r300_atom pvs_flush;
+    /* VAP invariant state. */
+    struct r300_atom vap_invariant_state;
+    /* Vertex stream formatting state. */
+    struct r300_atom vertex_stream_state;
+    /* Vertex shader. */
+    struct r300_atom vs_state;
     /* User clip planes. */
     struct r300_atom clip_state;
-    /* Depth, stencil, and alpha state. */
-    struct r300_atom dsa_state;
+    /* RS block state + VAP (vertex shader) output mapping state. */
+    struct r300_atom rs_block_state;
+    /* Rasterizer state. */
+    struct r300_atom rs_state;
+    /* Framebuffer state (pipelined regs). */
+    struct r300_atom fb_state_pipelined;
     /* Fragment shader. */
     struct r300_atom fs;
     /* Fragment shader RC_CONSTANT_STATE variables. */
     struct r300_atom fs_rc_constant_state;
     /* Fragment shader constant buffer. */
     struct r300_atom fs_constants;
-    /* Framebuffer state. */
-    struct r300_atom fb_state;
-    /* Framebuffer state (pipelined regs). */
-    struct r300_atom fb_state_pipelined;
-    /* HyperZ state (various SC/ZB bits). */
-    struct r300_atom hyperz_state;
-    /* Occlusion query. */
-    struct r300_atom query_start;
-    /* Rasterizer state. */
-    struct r300_atom rs_state;
-    /* RS block state + VAP (vertex shader) output mapping state. */
-    struct r300_atom rs_block_state;
-    /* Scissor state. */
-    struct r300_atom scissor_state;
-    /* Textures state. */
-    struct r300_atom textures_state;
-    /* Vertex stream formatting state. */
-    struct r300_atom vertex_stream_state;
-    /* Vertex shader. */
-    struct r300_atom vs_state;
     /* Vertex shader constant buffer. */
     struct r300_atom vs_constants;
-    /* Viewport state. */
-    struct r300_atom viewport_state;
-    /* ZTOP state. */
-    struct r300_atom ztop_state;
-    /* PVS flush. */
-    struct r300_atom pvs_flush;
-    /* VAP invariant state. */
-    struct r300_atom vap_invariant_state;
     /* Texture cache invalidate. */
     struct r300_atom texture_cache_inval;
-    /* GPU flush. */
-    struct r300_atom gpu_flush;
+    /* Textures state. */
+    struct r300_atom textures_state;
     /* HiZ clear */
     struct r300_atom hiz_clear;
     /* zmask clear */
     struct r300_atom zmask_clear;
+    /* Occlusion query. */
+    struct r300_atom query_start;
 
-    /* Invariant state. This must be emitted to get the engine started. */
-    struct r300_atom invariant_state;
+    /* The pointers to the first and the last atom. */
+    struct r300_atom *first_dirty, *last_dirty;
 
     /* Vertex buffers for Gallium. */
     struct pipe_vertex_buffer vertex_buffer[PIPE_MAX_ATTRIBS];
@@ -595,12 +603,29 @@ struct r300_context {
     struct u_upload_mgr *upload_vb;
     struct u_upload_mgr *upload_ib;
 
-    struct util_mempool pool_transfers;
+    struct util_slab_mempool pool_transfers;
 
     /* Stat counter. */
     uint64_t flush_counter;
+
+    /* const tracking for VS */
+    int vs_const_base;
+
+    /* AOS (PACKET3_3D_LOAD_VBPNTR) command buffer for the case offset=0. */
+    uint32_t aos_cb[(16 * 3 + 1) / 2];
+    boolean aos_dirty;
+
+    /* Whether any buffer (FB, textures, VBOs) has been set, but buffers
+     * haven't been validated yet. */
+    boolean validate_buffers;
 };
 
+#define foreach_atom(r300, atom) \
+    for (atom = &r300->gpu_flush; atom != (&r300->query_start)+1; atom++)
+
+#define foreach_dirty_atom(r300, atom) \
+    for (atom = r300->first_dirty; atom != r300->last_dirty; atom++)
+
 /* Convenience cast wrappers. */
 static INLINE struct r300_query* r300_query(struct pipe_query* q)
 {
@@ -645,8 +670,8 @@ void r300_init_resource_functions(struct r300_context* r300);
 /* r300_blit.c */
 void r300_flush_depth_stencil(struct pipe_context *pipe,
                               struct pipe_resource *dst,
-                              struct pipe_subresource subdst,
-                              unsigned zslice);
+                              unsigned level,
+                              unsigned layer);
 
 /* r300_query.c */
 void r300_resume_query(struct r300_context *r300,
@@ -666,7 +691,6 @@ void r300_plug_in_stencil_ref_fallback(struct r300_context *r300);
 
 /* r300_render.c */
 void r300_draw_flush_vbuf(struct r300_context *r300);
-boolean r500_index_bias_supported(struct r300_context *r300);
 void r500_emit_index_bias(struct r300_context *r300, int index_bias);
 
 /* r300_state.c */
@@ -680,6 +704,22 @@ void r300_mark_fb_state_dirty(struct r300_context *r300,
                               enum r300_fb_state_change change);
 void r300_mark_fs_code_dirty(struct r300_context *r300);
 
+static INLINE void r300_mark_atom_dirty(struct r300_context *r300,
+                                        struct r300_atom *atom)
+{
+    atom->dirty = TRUE;
+
+    if (!r300->first_dirty) {
+        r300->first_dirty = atom;
+        r300->last_dirty = atom+1;
+    } else {
+        if (atom < r300->first_dirty)
+            r300->first_dirty = atom;
+        if (atom+1 > r300->last_dirty)
+            r300->last_dirty = atom+1;
+    }
+}
+
 /* r300_debug.c */
 void r500_dump_rs_block(struct r300_rs_block *rs);
 
diff --git a/src/gallium/drivers/r300/r300_cs.h b/src/gallium/drivers/r300/r300_cs.h
index c194d6a1b0..67fb0096a8 100644
--- a/src/gallium/drivers/r300/r300_cs.h
+++ b/src/gallium/drivers/r300/r300_cs.h
@@ -51,7 +51,7 @@
     int cs_count = 0; (void) cs_count; (void) cs_winsys;
 
 #define BEGIN_CS(size) do { \
-    assert(size <= (cs_copy->ndw - cs_copy->cdw)); \
+    assert(size <= (R300_MAX_CMDBUF_DWORDS - cs_copy->cdw)); \
     CS_DEBUG(cs_count = size;) \
 } while (0)
 
@@ -72,7 +72,7 @@
  */
 
 #define OUT_CS(value) do { \
-    cs_copy->ptr[cs_copy->cdw++] = (value); \
+    cs_copy->buf[cs_copy->cdw++] = (value); \
     CS_DEBUG(cs_count--;) \
 } while (0)
 
@@ -96,7 +96,7 @@
     OUT_CS(CP_PACKET3(op, count))
 
 #define OUT_CS_TABLE(values, count) do { \
-    memcpy(cs_copy->ptr + cs_copy->cdw, values, count * 4); \
+    memcpy(cs_copy->buf + cs_copy->cdw, values, count * 4); \
     cs_copy->cdw += count; \
     CS_DEBUG(cs_count -= count;) \
 } while (0)
@@ -115,17 +115,17 @@
 
 #define OUT_CS_BUF_RELOC(bo, offset, rd, wd) do { \
     assert(bo); \
-    OUT_CS_RELOC(r300_buffer(bo)->buf, offset, rd, wd); \
+    OUT_CS_RELOC(r300_buffer(bo)->cs_buf, offset, rd, wd); \
 } while (0)
 
 #define OUT_CS_TEX_RELOC(tex, offset, rd, wd) do { \
     assert(tex); \
-    OUT_CS_RELOC(tex->buffer, offset, rd, wd); \
+    OUT_CS_RELOC(tex->cs_buffer, offset, rd, wd); \
 } while (0)
 
 #define OUT_CS_BUF_RELOC_NO_OFFSET(bo, rd, wd) do { \
     assert(bo); \
-    cs_winsys->cs_write_reloc(cs_copy, r300_buffer(bo)->buf, rd, wd); \
+    cs_winsys->cs_write_reloc(cs_copy, r300_buffer(bo)->cs_buf, rd, wd); \
     CS_DEBUG(cs_count -= 2;) \
 } while (0)
 
@@ -136,7 +136,7 @@
 
 #define WRITE_CS_TABLE(values, count) do { \
     CS_DEBUG(assert(cs_count == 0);) \
-    memcpy(cs_copy->ptr + cs_copy->cdw, (values), (count) * 4); \
+    memcpy(cs_copy->buf + cs_copy->cdw, (values), (count) * 4); \
     cs_copy->cdw += (count); \
 } while (0)
 
diff --git a/src/gallium/drivers/r300/r300_debug.c b/src/gallium/drivers/r300/r300_debug.c
index f78fe34790..52031dd97b 100644
--- a/src/gallium/drivers/r300/r300_debug.c
+++ b/src/gallium/drivers/r300/r300_debug.c
@@ -40,7 +40,6 @@ static const struct debug_named_value debug_options[] = {
     { "rs", DBG_RS, "Log rasterizer" },
     { "fb", DBG_FB, "Log framebuffer" },
     { "cbzb", DBG_CBZB, "Log fast color clear info" },
-    { "stats", DBG_STATS, "Log emission statistics" },
     { "hyperz", DBG_HYPERZ, "Log HyperZ info" },
     { "scissor", DBG_SCISSOR, "Log scissor info" },
     { "fakeocc", DBG_FAKE_OCC, "Use fake occlusion queries" },
diff --git a/src/gallium/drivers/r300/r300_defines.h b/src/gallium/drivers/r300/r300_defines.h
index 896aeef395..2d111f9158 100644
--- a/src/gallium/drivers/r300/r300_defines.h
+++ b/src/gallium/drivers/r300/r300_defines.h
@@ -43,8 +43,8 @@ enum r300_buffer_tiling {
 };
 
 enum r300_buffer_domain { /* bitfield */
-    R300_DOMAIN_GTT  = 1,
-    R300_DOMAIN_VRAM = 2
+    R300_DOMAIN_GTT  = 2,
+    R300_DOMAIN_VRAM = 4
 };
 
 #endif
diff --git a/src/gallium/drivers/r300/r300_emit.c b/src/gallium/drivers/r300/r300_emit.c
index c187f115da..04a5bd92d1 100644
--- a/src/gallium/drivers/r300/r300_emit.c
+++ b/src/gallium/drivers/r300/r300_emit.c
@@ -29,6 +29,7 @@
 #include "util/u_simple_list.h"
 
 #include "r300_context.h"
+#include "r300_cb.h"
 #include "r300_cs.h"
 #include "r300_emit.h"
 #include "r300_fs.h"
@@ -83,16 +84,20 @@ void r300_emit_dsa_state(struct r300_context* r300, unsigned size, void* state)
     }
 }
 
-static const float * get_rc_constant_state(
+static void get_rc_constant_state(
+    float vec[4],
     struct r300_context * r300,
     struct rc_constant * constant)
 {
     struct r300_textures_state* texstate = r300->textures_state.state;
-    static float vec[4] = { 0.0, 0.0, 0.0, 1.0 };
     struct r300_texture *tex;
 
     assert(constant->Type == RC_CONSTANT_STATE);
 
+    /* vec should either be (0, 0, 0, 1), which should be a relatively safe
+     * RGBA or STRQ value, or it could be one of the RC_CONSTANT_STATE
+     * state factors. */
+
     switch (constant->u.State[0]) {
         /* Factor for converting rectangle coords to
          * normalized coords. Should only show up on non-r500. */
@@ -100,6 +105,8 @@ static const float * get_rc_constant_state(
             tex = r300_texture(texstate->sampler_views[constant->u.State[1]]->base.texture);
             vec[0] = 1.0 / tex->desc.width0;
             vec[1] = 1.0 / tex->desc.height0;
+            vec[2] = 0;
+            vec[3] = 1;
             break;
 
         case RC_STATE_R300_TEXSCALE_FACTOR:
@@ -108,29 +115,31 @@ static const float * get_rc_constant_state(
             vec[0] = tex->desc.b.b.width0  / (tex->desc.width0  + 0.001f);
             vec[1] = tex->desc.b.b.height0 / (tex->desc.height0 + 0.001f);
             vec[2] = tex->desc.b.b.depth0  / (tex->desc.depth0  + 0.001f);
+            vec[3] = 1;
             break;
 
         case RC_STATE_R300_VIEWPORT_SCALE:
             vec[0] = r300->viewport.scale[0];
             vec[1] = r300->viewport.scale[1];
             vec[2] = r300->viewport.scale[2];
+            vec[3] = 1;
             break;
 
         case RC_STATE_R300_VIEWPORT_OFFSET:
             vec[0] = r300->viewport.translate[0];
             vec[1] = r300->viewport.translate[1];
             vec[2] = r300->viewport.translate[2];
+            vec[3] = 1;
             break;
 
         default:
             fprintf(stderr, "r300: Implementation error: "
                 "Unknown RC_CONSTANT type %d\n", constant->u.State[0]);
+            vec[0] = 0;
+            vec[1] = 0;
+            vec[2] = 0;
+            vec[3] = 1;
     }
-
-    /* This should either be (0, 0, 0, 1), which should be a relatively safe
-     * RGBA or STRQ value, or it could be one of the RC_CONSTANT_STATE
-     * state factors. */
-    return vec;
 }
 
 /* Convert a normal single-precision float into the 7.16 format
@@ -220,8 +229,9 @@ void r300_emit_fs_rc_constant_state(struct r300_context* r300, unsigned size, vo
     BEGIN_CS(size);
     for(i = first; i < end; ++i) {
         if (constants->Constants[i].Type == RC_CONSTANT_STATE) {
-            const float *data =
-                    get_rc_constant_state(r300, &constants->Constants[i]);
+            float data[4];
+
+            get_rc_constant_state(data, r300, &constants->Constants[i]);
 
             OUT_CS_REG_SEQ(R300_PFS_PARAM_0_X + i * 16, 4);
             for (j = 0; j < 4; j++)
@@ -279,8 +289,9 @@ void r500_emit_fs_rc_constant_state(struct r300_context* r300, unsigned size, vo
     BEGIN_CS(size);
     for(i = first; i < end; ++i) {
         if (constants->Constants[i].Type == RC_CONSTANT_STATE) {
-            const float *data =
-                    get_rc_constant_state(r300, &constants->Constants[i]);
+            float data[4];
+
+            get_rc_constant_state(data, r300, &constants->Constants[i]);
 
             OUT_CS_REG(R500_GA_US_VECTOR_INDEX,
                        R500_GA_US_VECTOR_INDEX_TYPE_CONST |
@@ -343,10 +354,10 @@ void r300_emit_aa_state(struct r300_context *r300, unsigned size, void *state)
 
     if (aa->dest) {
         OUT_CS_REG_SEQ(R300_RB3D_AARESOLVE_OFFSET, 1);
-        OUT_CS_RELOC(aa->dest->buffer, aa->dest->offset, 0, aa->dest->domain);
+        OUT_CS_RELOC(aa->dest->cs_buffer, aa->dest->offset, 0, aa->dest->domain);
 
         OUT_CS_REG_SEQ(R300_RB3D_AARESOLVE_PITCH, 1);
-        OUT_CS_RELOC(aa->dest->buffer, aa->dest->pitch, 0, aa->dest->domain);
+        OUT_CS_RELOC(aa->dest->cs_buffer, aa->dest->pitch, 0, aa->dest->domain);
     }
 
     OUT_CS_REG(R300_RB3D_AARESOLVE_CTL, aa->aaresolve_ctl);
@@ -377,10 +388,10 @@ void r300_emit_fb_state(struct r300_context* r300, unsigned size, void* state)
         surf = r300_surface(fb->cbufs[i]);
 
         OUT_CS_REG_SEQ(R300_RB3D_COLOROFFSET0 + (4 * i), 1);
-        OUT_CS_RELOC(surf->buffer, surf->offset, 0, surf->domain);
+        OUT_CS_RELOC(surf->cs_buffer, surf->offset, 0, surf->domain);
 
         OUT_CS_REG_SEQ(R300_RB3D_COLORPITCH0 + (4 * i), 1);
-        OUT_CS_RELOC(surf->buffer, surf->pitch, 0, surf->domain);
+        OUT_CS_RELOC(surf->cs_buffer, surf->pitch, 0, surf->domain);
     }
 
     /* Set up the ZB part of the CBZB clear. */
@@ -390,10 +401,10 @@ void r300_emit_fb_state(struct r300_context* r300, unsigned size, void* state)
         OUT_CS_REG(R300_ZB_FORMAT, surf->cbzb_format);
 
         OUT_CS_REG_SEQ(R300_ZB_DEPTHOFFSET, 1);
-        OUT_CS_RELOC(surf->buffer, surf->cbzb_midpoint_offset, 0, surf->domain);
+        OUT_CS_RELOC(surf->cs_buffer, surf->cbzb_midpoint_offset, 0, surf->domain);
 
         OUT_CS_REG_SEQ(R300_ZB_DEPTHPITCH, 1);
-        OUT_CS_RELOC(surf->buffer, surf->cbzb_pitch, 0, surf->domain);
+        OUT_CS_RELOC(surf->cs_buffer, surf->cbzb_pitch, 0, surf->domain);
 
         DBG(r300, DBG_CBZB,
             "CBZB clearing cbuf %08x %08x\n", surf->cbzb_format,
@@ -406,15 +417,15 @@ void r300_emit_fb_state(struct r300_context* r300, unsigned size, void* state)
         OUT_CS_REG(R300_ZB_FORMAT, surf->format);
 
         OUT_CS_REG_SEQ(R300_ZB_DEPTHOFFSET, 1);
-        OUT_CS_RELOC(surf->buffer, surf->offset, 0, surf->domain);
+        OUT_CS_RELOC(surf->cs_buffer, surf->offset, 0, surf->domain);
 
         OUT_CS_REG_SEQ(R300_ZB_DEPTHPITCH, 1);
-        OUT_CS_RELOC(surf->buffer, surf->pitch, 0, surf->domain);
+        OUT_CS_RELOC(surf->cs_buffer, surf->pitch, 0, surf->domain);
 
         if (can_hyperz) {
             uint32_t surf_pitch;
             struct r300_texture *tex;
-            int level = surf->base.level;
+            int level = surf->base.u.tex.level;
             tex = r300_texture(surf->base.texture);
 
             surf_pitch = surf->pitch & R300_DEPTHPITCH_MASK;
@@ -549,7 +560,7 @@ static void r300_emit_query_end_frag_pipes(struct r300_context *r300,
                                            struct r300_query *query)
 {
     struct r300_capabilities* caps = &r300->screen->caps;
-    struct r300_winsys_buffer *buf = r300->query_current->buffer;
+    struct r300_winsys_cs_buffer *buf = r300->query_current->cs_buffer;
     CS_LOCALS(r300);
 
     assert(caps->num_frag_pipes);
@@ -605,7 +616,7 @@ static void r300_emit_query_end_frag_pipes(struct r300_context *r300,
 static void rv530_emit_query_end_single_z(struct r300_context *r300,
                                           struct r300_query *query)
 {
-    struct r300_winsys_buffer *buf = r300->query_current->buffer;
+    struct r300_winsys_cs_buffer *buf = r300->query_current->cs_buffer;
     CS_LOCALS(r300);
 
     BEGIN_CS(8);
@@ -619,7 +630,7 @@ static void rv530_emit_query_end_single_z(struct r300_context *r300,
 static void rv530_emit_query_end_double_z(struct r300_context *r300,
                                           struct r300_query *query)
 {
-    struct r300_winsys_buffer *buf = r300->query_current->buffer;
+    struct r300_winsys_cs_buffer *buf = r300->query_current->cs_buffer;
     CS_LOCALS(r300);
 
     BEGIN_CS(14);
@@ -796,39 +807,83 @@ void r300_emit_textures_state(struct r300_context *r300,
     END_CS;
 }
 
-void r300_emit_aos(struct r300_context* r300, int offset, boolean indexed)
+static void r300_update_aos_cb(struct r300_context *r300, unsigned packet_size)
 {
     struct pipe_vertex_buffer *vb1, *vb2, *vbuf = r300->vertex_buffer;
     struct pipe_vertex_element *velem = r300->velems->velem;
-    struct r300_buffer *buf;
-    int i;
     unsigned *hw_format_size = r300->velems->hw_format_size;
     unsigned size1, size2, aos_count = r300->velems->count;
-    unsigned packet_size = (aos_count * 3 + 1) / 2;
-    CS_LOCALS(r300);
-
-    BEGIN_CS(2 + packet_size + aos_count * 2);
-    OUT_CS_PKT3(R300_PACKET3_3D_LOAD_VBPNTR, packet_size);
-    OUT_CS(aos_count | (!indexed ? R300_VC_FORCE_PREFETCH : 0));
+    int i;
+    CB_LOCALS;
 
+    BEGIN_CB(r300->aos_cb, packet_size);
     for (i = 0; i < aos_count - 1; i += 2) {
         vb1 = &vbuf[velem[i].vertex_buffer_index];
         vb2 = &vbuf[velem[i+1].vertex_buffer_index];
         size1 = hw_format_size[i];
         size2 = hw_format_size[i+1];
 
-        OUT_CS(R300_VBPNTR_SIZE0(size1) | R300_VBPNTR_STRIDE0(vb1->stride) |
+        OUT_CB(R300_VBPNTR_SIZE0(size1) | R300_VBPNTR_STRIDE0(vb1->stride) |
                R300_VBPNTR_SIZE1(size2) | R300_VBPNTR_STRIDE1(vb2->stride));
-        OUT_CS(vb1->buffer_offset + velem[i].src_offset   + offset * vb1->stride);
-        OUT_CS(vb2->buffer_offset + velem[i+1].src_offset + offset * vb2->stride);
+        OUT_CB(vb1->buffer_offset + velem[i].src_offset);
+        OUT_CB(vb2->buffer_offset + velem[i+1].src_offset);
     }
 
     if (aos_count & 1) {
         vb1 = &vbuf[velem[i].vertex_buffer_index];
         size1 = hw_format_size[i];
 
-        OUT_CS(R300_VBPNTR_SIZE0(size1) | R300_VBPNTR_STRIDE0(vb1->stride));
-        OUT_CS(vb1->buffer_offset + velem[i].src_offset + offset * vb1->stride);
+        OUT_CB(R300_VBPNTR_SIZE0(size1) | R300_VBPNTR_STRIDE0(vb1->stride));
+        OUT_CB(vb1->buffer_offset + velem[i].src_offset);
+    }
+    END_CB;
+
+    r300->aos_dirty = FALSE;
+}
+
+void r300_emit_aos(struct r300_context* r300, int offset, boolean indexed)
+{
+    struct pipe_vertex_buffer *vbuf = r300->vertex_buffer;
+    struct pipe_vertex_element *velem = r300->velems->velem;
+    struct r300_buffer *buf;
+    int i;
+    unsigned aos_count = r300->velems->count;
+    unsigned packet_size = (aos_count * 3 + 1) / 2;
+    CS_LOCALS(r300);
+
+    BEGIN_CS(2 + packet_size + aos_count * 2);
+    OUT_CS_PKT3(R300_PACKET3_3D_LOAD_VBPNTR, packet_size);
+    OUT_CS(aos_count | (!indexed ? R300_VC_FORCE_PREFETCH : 0));
+
+    if (!offset) {
+        if (r300->aos_dirty) {
+            r300_update_aos_cb(r300, packet_size);
+        }
+        OUT_CS_TABLE(r300->aos_cb, packet_size);
+    } else {
+        struct pipe_vertex_buffer *vb1, *vb2;
+        unsigned *hw_format_size = r300->velems->hw_format_size;
+        unsigned size1, size2;
+
+        for (i = 0; i < aos_count - 1; i += 2) {
+            vb1 = &vbuf[velem[i].vertex_buffer_index];
+            vb2 = &vbuf[velem[i+1].vertex_buffer_index];
+            size1 = hw_format_size[i];
+            size2 = hw_format_size[i+1];
+
+            OUT_CS(R300_VBPNTR_SIZE0(size1) | R300_VBPNTR_STRIDE0(vb1->stride) |
+                   R300_VBPNTR_SIZE1(size2) | R300_VBPNTR_STRIDE1(vb2->stride));
+            OUT_CS(vb1->buffer_offset + velem[i].src_offset   + offset * vb1->stride);
+            OUT_CS(vb2->buffer_offset + velem[i+1].src_offset + offset * vb2->stride);
+        }
+
+        if (aos_count & 1) {
+            vb1 = &vbuf[velem[i].vertex_buffer_index];
+            size1 = hw_format_size[i];
+
+            OUT_CS(R300_VBPNTR_SIZE0(size1) | R300_VBPNTR_STRIDE0(vb1->stride));
+            OUT_CS(vb1->buffer_offset + velem[i].src_offset + offset * vb1->stride);
+        }
     }
 
     for (i = 0; i < aos_count; i++) {
@@ -914,7 +969,6 @@ void r300_emit_vs_state(struct r300_context* r300, unsigned size, void* state)
     struct r300_vertex_program_code* code = &vs->code;
     struct r300_screen* r300screen = r300->screen;
     unsigned instruction_count = code->length / 4;
-    unsigned i;
 
     unsigned vtx_mem_size = r300screen->caps.is_r500 ? 128 : 72;
     unsigned input_count = MAX2(util_bitcount(code->InputsRead), 1);
@@ -925,10 +979,6 @@ void r300_emit_vs_state(struct r300_context* r300, unsigned size, void* state)
                                   vtx_mem_size / output_count, 10);
     unsigned pvs_num_controllers = MIN2(vtx_mem_size / temp_count, 5);
 
-    unsigned imm_first = vs->externals_count;
-    unsigned imm_end = vs->code.constants.Count;
-    unsigned imm_count = vs->immediates_count;
-
     CS_LOCALS(r300);
 
     BEGIN_CS(size);
@@ -937,12 +987,10 @@ void r300_emit_vs_state(struct r300_context* r300, unsigned size, void* state)
      * R300_VAP_PVS_CONST_CNTL
      * R300_VAP_PVS_CODE_CNTL_1
      * See the r5xx docs for instructions on how to use these. */
-    OUT_CS_REG_SEQ(R300_VAP_PVS_CODE_CNTL_0, 3);
-    OUT_CS(R300_PVS_FIRST_INST(0) |
-            R300_PVS_XYZW_VALID_INST(instruction_count - 1) |
-            R300_PVS_LAST_INST(instruction_count - 1));
-    OUT_CS(R300_PVS_MAX_CONST_ADDR(code->constants.Count - 1));
-    OUT_CS(instruction_count - 1);
+    OUT_CS_REG(R300_VAP_PVS_CODE_CNTL_0, R300_PVS_FIRST_INST(0) |
+	       R300_PVS_XYZW_VALID_INST(instruction_count - 1) |
+	       R300_PVS_LAST_INST(instruction_count - 1));
+    OUT_CS_REG(R300_VAP_PVS_CODE_CNTL_1, instruction_count - 1);
 
     OUT_CS_REG(R300_VAP_PVS_VECTOR_INDX_REG, 0);
     OUT_CS_ONE_REG(R300_VAP_PVS_UPLOAD_DATA, code->length);
@@ -954,19 +1002,6 @@ void r300_emit_vs_state(struct r300_context* r300, unsigned size, void* state)
             R300_PVS_VF_MAX_VTX_NUM(12) |
             (r300screen->caps.is_r500 ? R500_TCL_STATE_OPTIMIZATION : 0));
 
-    /* Emit immediates. */
-    if (imm_count) {
-        OUT_CS_REG(R300_VAP_PVS_VECTOR_INDX_REG,
-                   (r300->screen->caps.is_r500 ?
-                   R500_PVS_CONST_START : R300_PVS_CONST_START) +
-                   imm_first);
-        OUT_CS_ONE_REG(R300_VAP_PVS_UPLOAD_DATA, imm_count * 4);
-        for (i = imm_first; i < imm_end; i++) {
-            const float *data = vs->code.constants.Constants[i].u.Immediate;
-            OUT_CS_TABLE(data, 4);
-        }
-    }
-
     /* Emit flow control instructions. */
     if (code->num_fc_ops) {
 
@@ -991,24 +1026,43 @@ void r300_emit_vs_constants(struct r300_context* r300,
     unsigned count =
         ((struct r300_vertex_shader*)r300->vs_state.state)->externals_count;
     struct r300_constant_buffer *buf = (struct r300_constant_buffer*)state;
+    struct r300_vertex_shader *vs = (struct r300_vertex_shader*)r300->vs_state.state;
     unsigned i;
+    int imm_first = vs->externals_count;
+    int imm_end = vs->code.constants.Count;
+    int imm_count = vs->immediates_count;
     CS_LOCALS(r300);
 
-    if (!count)
-        return;
-
     BEGIN_CS(size);
-    OUT_CS_REG(R300_VAP_PVS_VECTOR_INDX_REG,
-               (r300->screen->caps.is_r500 ?
-               R500_PVS_CONST_START : R300_PVS_CONST_START));
-    OUT_CS_ONE_REG(R300_VAP_PVS_UPLOAD_DATA, count * 4);
-    if (buf->remap_table){
-        for (i = 0; i < count; i++) {
-            uint32_t *data = &buf->ptr[buf->remap_table[i]*4];
+    OUT_CS_REG(R300_VAP_PVS_CONST_CNTL,
+               R300_PVS_CONST_BASE_OFFSET(buf->buffer_base) |
+               R300_PVS_MAX_CONST_ADDR(MAX2(imm_end - 1, 0)));
+    if (vs->externals_count) {
+        OUT_CS_REG(R300_VAP_PVS_VECTOR_INDX_REG,
+                   (r300->screen->caps.is_r500 ?
+                   R500_PVS_CONST_START : R300_PVS_CONST_START) + buf->buffer_base);
+        OUT_CS_ONE_REG(R300_VAP_PVS_UPLOAD_DATA, count * 4);
+        if (buf->remap_table){
+            for (i = 0; i < count; i++) {
+                uint32_t *data = &buf->ptr[buf->remap_table[i]*4];
+                OUT_CS_TABLE(data, 4);
+            }
+        } else {
+            OUT_CS_TABLE(buf->ptr, count * 4);
+        }
+    }
+
+    /* Emit immediates. */
+    if (imm_count) {
+        OUT_CS_REG(R300_VAP_PVS_VECTOR_INDX_REG,
+                   (r300->screen->caps.is_r500 ?
+                   R500_PVS_CONST_START : R300_PVS_CONST_START) +
+                   buf->buffer_base + imm_first);
+        OUT_CS_ONE_REG(R300_VAP_PVS_UPLOAD_DATA, imm_count * 4);
+        for (i = imm_first; i < imm_end; i++) {
+            const float *data = vs->code.constants.Constants[i].u.Immediate;
             OUT_CS_TABLE(data, 4);
         }
-    } else {
-        OUT_CS_TABLE(buf->ptr, count * 4);
     }
     END_CS;
 }
@@ -1063,8 +1117,8 @@ void r300_emit_hiz_clear(struct r300_context *r300, unsigned size, void *state)
 
     tex = r300_texture(fb->zsbuf->texture);
 
-    offset = tex->hiz_mem[fb->zsbuf->level]->ofs;
-    stride = tex->desc.stride_in_pixels[fb->zsbuf->level];
+    offset = tex->hiz_mem[fb->zsbuf->u.tex.level]->ofs;
+    stride = tex->desc.stride_in_pixels[fb->zsbuf->u.tex.level];
 
     /* convert from pixels to 4x4 blocks */
     stride = ALIGN_DIVUP(stride, 4);
@@ -1086,7 +1140,7 @@ void r300_emit_hiz_clear(struct r300_context *r300, unsigned size, void *state)
     z->current_func = -1;
 
     /* Mark the current zbuffer's hiz ram as in use. */
-    tex->hiz_in_use[fb->zsbuf->level] = TRUE;
+    tex->hiz_in_use[fb->zsbuf->u.tex.level] = TRUE;
 }
 
 void r300_emit_zmask_clear(struct r300_context *r300, unsigned size, void *state)
@@ -1100,9 +1154,9 @@ void r300_emit_zmask_clear(struct r300_context *r300, unsigned size, void *state
     int mult, offset_shift;
 
     tex = r300_texture(fb->zsbuf->texture);
-    stride = tex->desc.stride_in_pixels[fb->zsbuf->level];
+    stride = tex->desc.stride_in_pixels[fb->zsbuf->u.tex.level];
 
-    offset = tex->zmask_mem[fb->zsbuf->level]->ofs;
+    offset = tex->zmask_mem[fb->zsbuf->u.tex.level]->ofs;
 
     if (r300->z_compression == RV350_Z_COMPRESS_88)
         mult = 8;
@@ -1128,7 +1182,7 @@ void r300_emit_zmask_clear(struct r300_context *r300, unsigned size, void *state
     }
 
     /* Mark the current zbuffer's zmask as in use. */
-    tex->zmask_in_use[fb->zsbuf->level] = TRUE;
+    tex->zmask_in_use[fb->zsbuf->u.tex.level] = TRUE;
 }
 
 void r300_emit_ztop_state(struct r300_context* r300,
@@ -1165,12 +1219,6 @@ boolean r300_emit_buffer_validate(struct r300_context *r300,
     struct pipe_resource *pbuf;
     unsigned i;
 
-    /* upload buffers first */
-    if (r300->screen->caps.has_tcl && r300->any_user_vbs) {
-        r300_upload_user_buffers(r300);
-        r300->any_user_vbs = false;
-    }
-
     /* Clean out BOs. */
     r300->rws->cs_reset_buffers(r300->cs);
 
@@ -1178,14 +1226,14 @@ boolean r300_emit_buffer_validate(struct r300_context *r300,
     for (i = 0; i < fb->nr_cbufs; i++) {
         tex = r300_texture(fb->cbufs[i]->texture);
         assert(tex && tex->buffer && "cbuf is marked, but NULL!");
-        r300->rws->cs_add_buffer(r300->cs, tex->buffer, 0,
+        r300->rws->cs_add_buffer(r300->cs, tex->cs_buffer, 0,
                                  r300_surface(fb->cbufs[i])->domain);
     }
     /* ...depth buffer... */
     if (fb->zsbuf) {
         tex = r300_texture(fb->zsbuf->texture);
         assert(tex && tex->buffer && "zsbuf is marked, but NULL!");
-        r300->rws->cs_add_buffer(r300->cs, tex->buffer, 0,
+        r300->rws->cs_add_buffer(r300->cs, tex->cs_buffer, 0,
                                  r300_surface(fb->zsbuf)->domain);
     }
     /* ...textures... */
@@ -1195,28 +1243,30 @@ boolean r300_emit_buffer_validate(struct r300_context *r300,
         }
 
         tex = r300_texture(texstate->sampler_views[i]->base.texture);
-        r300->rws->cs_add_buffer(r300->cs, tex->buffer, tex->domain, 0);
+        r300->rws->cs_add_buffer(r300->cs, tex->cs_buffer, tex->domain, 0);
     }
     /* ...occlusion query buffer... */
     if (r300->query_current)
-        r300->rws->cs_add_buffer(r300->cs, r300->query_current->buffer,
+        r300->rws->cs_add_buffer(r300->cs, r300->query_current->cs_buffer,
                                  0, r300->query_current->domain);
     /* ...vertex buffer for SWTCL path... */
     if (r300->vbo)
-        r300->rws->cs_add_buffer(r300->cs, r300_buffer(r300->vbo)->buf,
+        r300->rws->cs_add_buffer(r300->cs, r300_buffer(r300->vbo)->cs_buf,
                                  r300_buffer(r300->vbo)->domain, 0);
     /* ...vertex buffers for HWTCL path... */
     if (do_validate_vertex_buffers) {
         for (i = 0; i < r300->velems->count; i++) {
             pbuf = vbuf[velem[i].vertex_buffer_index].buffer;
+            if (!pbuf)
+                continue;
 
-            r300->rws->cs_add_buffer(r300->cs, r300_buffer(pbuf)->buf,
+            r300->rws->cs_add_buffer(r300->cs, r300_buffer(pbuf)->cs_buf,
                                      r300_buffer(pbuf)->domain, 0);
         }
     }
     /* ...and index buffer for HWTCL path. */
     if (index_buffer)
-        r300->rws->cs_add_buffer(r300->cs, r300_buffer(index_buffer)->buf,
+        r300->rws->cs_add_buffer(r300->cs, r300_buffer(index_buffer)->cs_buf,
                                  r300_buffer(index_buffer)->domain, 0);
 
     if (!r300->rws->cs_validate(r300->cs)) {
@@ -1231,7 +1281,7 @@ unsigned r300_get_num_dirty_dwords(struct r300_context *r300)
     struct r300_atom* atom;
     unsigned dwords = 0;
 
-    foreach(atom, &r300->atom_list) {
+    foreach_dirty_atom(r300, atom) {
         if (atom->dirty) {
             dwords += atom->size;
         }
@@ -1250,7 +1300,7 @@ unsigned r300_get_num_cs_end_dwords(struct r300_context *r300)
     /* Emitted in flush. */
     dwords += 26; /* emit_query_end */
     dwords += r300->hyperz_state.size + 2; /* emit_hyperz_end + zcache flush */
-    if (r500_index_bias_supported(r300))
+    if (r300->screen->caps.index_bias_supported)
         dwords += 2;
 
     return dwords;
@@ -1259,17 +1309,16 @@ unsigned r300_get_num_cs_end_dwords(struct r300_context *r300)
 /* Emit all dirty state. */
 void r300_emit_dirty_state(struct r300_context* r300)
 {
-    struct r300_atom* atom;
+    struct r300_atom *atom;
 
-    foreach(atom, &r300->atom_list) {
+    foreach_dirty_atom(r300, atom) {
         if (atom->dirty) {
             atom->emit(r300, atom->size, atom->state);
-            if (SCREEN_DBG_ON(r300->screen, DBG_STATS)) {
-                atom->counter++;
-            }
             atom->dirty = FALSE;
         }
     }
 
+    r300->first_dirty = NULL;
+    r300->last_dirty = NULL;
     r300->dirty_hw++;
 }
diff --git a/src/gallium/drivers/r300/r300_flush.c b/src/gallium/drivers/r300/r300_flush.c
index 1afd27f093..451fe525b4 100644
--- a/src/gallium/drivers/r300/r300_flush.c
+++ b/src/gallium/drivers/r300/r300_flush.c
@@ -49,7 +49,7 @@ static void r300_flush(struct pipe_context* pipe,
     if (r300->dirty_hw) {
         r300_emit_hyperz_end(r300);
         r300_emit_query_end(r300);
-        if (r500_index_bias_supported(r300))
+        if (r300->screen->caps.index_bias_supported)
             r500_emit_index_bias(r300, 0);
 
         r300->flush_counter++;
@@ -57,9 +57,9 @@ static void r300_flush(struct pipe_context* pipe,
         r300->dirty_hw = 0;
 
         /* New kitchen sink, baby. */
-        foreach(atom, &r300->atom_list) {
+        foreach_atom(r300, atom) {
             if (atom->state || atom->allow_null_state) {
-                atom->dirty = TRUE;
+                r300_mark_atom_dirty(r300, atom);
             }
         }
 
@@ -68,6 +68,8 @@ static void r300_flush(struct pipe_context* pipe,
             r300->vs_state.dirty = FALSE;
             r300->vs_constants.dirty = FALSE;
         }
+
+        r300->validate_buffers = TRUE;
     }
 
     /* reset flushed query */
diff --git a/src/gallium/drivers/r300/r300_fs.c b/src/gallium/drivers/r300/r300_fs.c
index c91532eb7b..2936c3486e 100644
--- a/src/gallium/drivers/r300/r300_fs.c
+++ b/src/gallium/drivers/r300/r300_fs.c
@@ -390,7 +390,6 @@ static void r300_translate_fragment_shader(
     compiler.Base.max_temp_regs = compiler.Base.is_r500 ? 128 : 32;
     compiler.Base.max_constants = compiler.Base.is_r500 ? 256 : 32;
     compiler.Base.max_alu_insts = compiler.Base.is_r500 ? 512 : 64;
-    compiler.Base.remove_unused_constants = TRUE;
     compiler.AllocateHwInputs = &allocate_hardware_inputs;
     compiler.UserData = &shader->inputs;
 
@@ -408,6 +407,11 @@ static void r300_translate_fragment_shader(
 
     r300_tgsi_to_rc(&ttr, tokens);
 
+    if (!r300->screen->caps.is_r500 ||
+        compiler.Base.Program.Constants.Count > 200) {
+        compiler.Base.remove_unused_constants = TRUE;
+    }
+
     /**
      * Transform the program to support WPOS.
      *
diff --git a/src/gallium/drivers/r300/r300_hyperz.c b/src/gallium/drivers/r300/r300_hyperz.c
index 79f7f8abe9..c22e307c67 100644
--- a/src/gallium/drivers/r300/r300_hyperz.c
+++ b/src/gallium/drivers/r300/r300_hyperz.c
@@ -158,8 +158,8 @@ static void r300_update_hyperz(struct r300_context* r300)
     if (!r300->rws->get_value(r300->rws, R300_CAN_HYPERZ))
         return;
 
-    zmask_in_use = zstex->zmask_in_use[fb->zsbuf->level];
-    hiz_in_use = zstex->hiz_in_use[fb->zsbuf->level];
+    zmask_in_use = zstex->zmask_in_use[fb->zsbuf->u.tex.level];
+    hiz_in_use = zstex->hiz_in_use[fb->zsbuf->u.tex.level];
 
     /* Z fastfill. */
     if (zmask_in_use) {
@@ -282,7 +282,7 @@ static void r300_update_ztop(struct r300_context* r300)
         ztop_state->z_buffer_top = R300_ZTOP_ENABLE;
     }
     if (ztop_state->z_buffer_top != old_ztop)
-        r300->ztop_state.dirty = TRUE;
+        r300_mark_atom_dirty(r300, &r300->ztop_state);
 }
 
 #define ALIGN_DIVUP(x, y) (((x) + (y) - 1) / (y))
@@ -333,7 +333,7 @@ void r300_hiz_alloc_block(struct r300_context *r300, struct r300_surface *surf)
 {
     struct r300_texture *tex;
     uint32_t zsize, ndw;
-    int level = surf->base.level;
+    int level = surf->base.u.tex.level;
 
     tex = r300_texture(surf->base.texture);
 
@@ -352,7 +352,7 @@ void r300_zmask_alloc_block(struct r300_context *r300, struct r300_surface *surf
 {
     int bsize = 256;
     uint32_t zsize, ndw;
-    int level = surf->base.level;
+    int level = surf->base.u.tex.level;
     struct r300_texture *tex;
 
     tex = r300_texture(surf->base.texture);
diff --git a/src/gallium/drivers/r300/r300_query.c b/src/gallium/drivers/r300/r300_query.c
index 5f34fcb274..6223e04321 100644
--- a/src/gallium/drivers/r300/r300_query.c
+++ b/src/gallium/drivers/r300/r300_query.c
@@ -60,6 +60,7 @@ static struct pipe_query *r300_create_query(struct pipe_context *pipe,
     q->buffer = r300->rws->buffer_create(r300->rws, q->buffer_size, 4096,
                                          PIPE_BIND_CUSTOM, PIPE_USAGE_STREAM,
                                          q->domain);
+    q->cs_buffer = r300->rws->buffer_get_cs_handle(r300->rws, q->buffer);
 
     return (struct pipe_query*)q;
 }
@@ -79,7 +80,7 @@ void r300_resume_query(struct r300_context *r300,
                        struct r300_query *query)
 {
     r300->query_current = query;
-    r300->query_start.dirty = TRUE;
+    r300_mark_atom_dirty(r300, &r300->query_start);
 }
 
 static void r300_begin_query(struct pipe_context* pipe,
diff --git a/src/gallium/drivers/r300/r300_reg.h b/src/gallium/drivers/r300/r300_reg.h
index 6bea783f69..613186e815 100644
--- a/src/gallium/drivers/r300/r300_reg.h
+++ b/src/gallium/drivers/r300/r300_reg.h
@@ -427,7 +427,7 @@ USE OR OTHER DEALINGS IN THE SOFTWARE.
 #       define R300_PVS_CONST_START          512
 #       define R500_PVS_CONST_START          1024
 #       define R300_MAX_PVS_CONST_VECS       256
-#       define R500_MAX_PVS_CONST_VECS       1024
+#       define R500_MAX_PVS_CONST_VECS       256
 #       define R300_PVS_UCP_START            1024
 #       define R500_PVS_UCP_START            1536
 #       define R300_POINT_VPORT_SCALE_OFFSET 1030
@@ -553,6 +553,7 @@ USE OR OTHER DEALINGS IN THE SOFTWARE.
 /* Addresses are relative to the vertex program parameters area. */
 #define R300_VAP_PVS_CONST_CNTL             0x22D4
 #       define R300_PVS_CONST_BASE_OFFSET_SHIFT  0
+#       define R300_PVS_CONST_BASE_OFFSET(x)     (x)
 #       define R300_PVS_MAX_CONST_ADDR_SHIFT     16
 #       define R300_PVS_MAX_CONST_ADDR(x)        ((x) << 16)
 #define R300_VAP_PVS_CODE_CNTL_1	    0x22D8
@@ -1520,11 +1521,11 @@ USE OR OTHER DEALINGS IN THE SOFTWARE.
 #	define R300_TX_TRI_PERF_3_8            (3<<15)
 #	define R300_ANISO_THRESHOLD_MASK       (7<<17)
 
+#       define R400_DXTC_SWIZZLE_ENABLE        (1<<21)
 #	define R500_MACRO_SWITCH               (1<<22)
 #       define R500_TX_MAX_ANISO(x)            ((x) << 23)
 #       define R500_TX_MAX_ANISO_MASK          (63 << 23)
 #       define R500_TX_ANISO_HIGH_QUALITY      (1 << 30)
-
 #	define R500_BORDER_FIX                 (1<<31)
 
 #define R300_TX_FORMAT0_0                   0x4480
diff --git a/src/gallium/drivers/r300/r300_render.c b/src/gallium/drivers/r300/r300_render.c
index 2f00c878f5..1d26eb9f91 100644
--- a/src/gallium/drivers/r300/r300_render.c
+++ b/src/gallium/drivers/r300/r300_render.c
@@ -118,12 +118,6 @@ static uint32_t r300_provoking_vertex_fixes(struct r300_context *r300,
     return color_control;
 }
 
-boolean r500_index_bias_supported(struct r300_context *r300)
-{
-    return r300->screen->caps.is_r500 &&
-           r300->rws->get_value(r300->rws, R300_VID_DRM_2_3_0);
-}
-
 void r500_emit_index_bias(struct r300_context *r300, int index_bias)
 {
     CS_LOCALS(r300);
@@ -193,13 +187,12 @@ static boolean r300_reserve_cs_dwords(struct r300_context *r300,
     boolean first_draw     = flags & PREP_FIRST_DRAW;
     boolean emit_aos       = flags & PREP_EMIT_AOS;
     boolean emit_aos_swtcl = flags & PREP_EMIT_AOS_SWTCL;
-    boolean hw_index_bias  = r500_index_bias_supported(r300);
 
     /* Add dirty state, index offset, and AOS. */
     if (first_draw) {
         cs_dwords += r300_get_num_dirty_dwords(r300);
 
-        if (hw_index_bias)
+        if (r300->screen->caps.index_bias_supported)
             cs_dwords += 2; /* emit_index_offset */
 
         if (emit_aos)
@@ -212,7 +205,7 @@ static boolean r300_reserve_cs_dwords(struct r300_context *r300,
     cs_dwords += r300_get_num_cs_end_dwords(r300);
 
     /* Reserve requested CS space. */
-    if (cs_dwords > (r300->cs->ndw - r300->cs->cdw)) {
+    if (cs_dwords > (R300_MAX_CMDBUF_DWORDS - r300->cs->cdw)) {
         r300->context.flush(&r300->context, 0, NULL);
         flushed = TRUE;
     }
@@ -239,19 +232,32 @@ static boolean r300_emit_states(struct r300_context *r300,
     boolean emit_aos       = flags & PREP_EMIT_AOS;
     boolean emit_aos_swtcl = flags & PREP_EMIT_AOS_SWTCL;
     boolean indexed        = flags & PREP_INDEXED;
-    boolean hw_index_bias  = r500_index_bias_supported(r300);
+    boolean validate_vbos  = flags & PREP_VALIDATE_VBOS;
 
     /* Validate buffers and emit dirty state if needed. */
     if (first_draw) {
-        if (!r300_emit_buffer_validate(r300, flags & PREP_VALIDATE_VBOS,
-                                       index_buffer)) {
-            fprintf(stderr, "r300: CS space validation failed. "
-                    "(not enough memory?) Skipping rendering.\n");
-            return FALSE;
+        /* upload buffers first */
+        if (r300->screen->caps.has_tcl && r300->any_user_vbs) {
+            r300_upload_user_buffers(r300);
+            r300->any_user_vbs = false;
+        }
+
+        if (r300->validate_buffers) {
+            if (!r300_emit_buffer_validate(r300, validate_vbos,
+                                           index_buffer)) {
+                fprintf(stderr, "r300: CS space validation failed. "
+                        "(not enough memory?) Skipping rendering.\n");
+                return FALSE;
+            }
+
+            /* Consider the validation done only if everything was validated. */
+            if (validate_vbos) {
+                r300->validate_buffers = FALSE;
+            }
         }
 
         r300_emit_dirty_state(r300);
-        if (hw_index_bias) {
+        if (r300->screen->caps.index_bias_supported) {
             if (r300->screen->caps.has_tcl)
                 r500_emit_index_bias(r300, index_bias);
             else
@@ -535,30 +541,9 @@ static void r300_draw_range_elements(struct pipe_context* pipe,
                             r300->rws->get_value(r300->rws, R300_VID_DRM_2_3_0);
     unsigned short_count;
     int buffer_offset = 0, index_offset = 0; /* for index bias emulation */
-    boolean translate = FALSE;
     unsigned new_offset;
 
-    if (r300->skip_rendering) {
-        return;
-    }
-
-    if (!u_trim_pipe_prim(mode, &count)) {
-        return;
-    }
-
-    /* Index buffer range checking. */
-    if ((start + count) * indexSize > indexBuffer->width0) {
-        fprintf(stderr, "r300: Invalid index buffer range. Skipping rendering.\n");
-        return;
-    }
-
-    /* Set up fallback for incompatible vertex layout if needed. */
-    if (r300->incompatible_vb_layout || r300->velems->incompatible_layout) {
-        r300_begin_vertex_translate(r300);
-        translate = TRUE;
-    }
-
-    if (indexBias && !r500_index_bias_supported(r300)) {
+    if (indexBias && !r300->screen->caps.index_bias_supported) {
         r300_split_index_bias(r300, indexBias, &buffer_offset, &index_offset);
     }
 
@@ -566,7 +551,27 @@ static void r300_draw_range_elements(struct pipe_context* pipe,
                                 &start, count);
 
     r300_update_derived_state(r300);
-    r300_upload_index_buffer(r300, &indexBuffer, indexSize, start, count, &new_offset);
+
+    /* Fallback for misaligned ushort indices. */
+    if (indexSize == 2 && start % 2 == 1) {
+        struct pipe_transfer *transfer;
+        struct pipe_resource *userbuf;
+        uint16_t *ptr = pipe_buffer_map(pipe, indexBuffer,
+                                        PIPE_TRANSFER_READ, &transfer);
+
+        /* Copy the mapped index buffer directly to the upload buffer.
+         * The start index will be aligned simply from the fact that
+         * every sub-buffer in u_upload_mgr is aligned. */
+        userbuf = pipe->screen->user_buffer_create(pipe->screen,
+                                                   ptr + start, count * 2,
+                                                   PIPE_BIND_INDEX_BUFFER);
+        indexBuffer = userbuf;
+        r300_upload_index_buffer(r300, &indexBuffer, indexSize, 0, count, &new_offset);
+        pipe_resource_reference(&userbuf, NULL);
+        pipe_buffer_unmap(pipe, indexBuffer, transfer);
+    } else {
+        r300_upload_index_buffer(r300, &indexBuffer, indexSize, start, count, &new_offset);
+    }
 
     start = new_offset;
 
@@ -603,10 +608,6 @@ done:
     if (indexBuffer != orgIndexBuffer) {
         pipe_resource_reference( &indexBuffer, NULL );
     }
-
-    if (translate) {
-        r300_end_vertex_translate(r300);
-    }
 }
 
 static void r300_draw_arrays(struct pipe_context* pipe, unsigned mode,
@@ -617,21 +618,6 @@ static void r300_draw_arrays(struct pipe_context* pipe, unsigned mode,
                             count > 65536 &&
                             r300->rws->get_value(r300->rws, R300_VID_DRM_2_3_0);
     unsigned short_count;
-    boolean translate = FALSE;
-
-    if (r300->skip_rendering) {
-        return;
-    }
-
-    if (!u_trim_pipe_prim(mode, &count)) {
-        return;
-    }
-
-    /* Set up fallback for incompatible vertex layout if needed. */
-    if (r300->incompatible_vb_layout || r300->velems->incompatible_layout) {
-        r300_begin_vertex_translate(r300);
-        translate = TRUE;
-    }
 
     r300_update_derived_state(r300);
 
@@ -642,7 +628,7 @@ static void r300_draw_arrays(struct pipe_context* pipe, unsigned mode,
         if (!r300_prepare_for_rendering(r300,
                 PREP_FIRST_DRAW | PREP_VALIDATE_VBOS | PREP_EMIT_AOS,
                 NULL, 9, start, 0))
-            goto done;
+            return;
 
         if (alt_num_verts || count <= 65535) {
             r300_emit_draw_arrays(r300, mode, count);
@@ -659,32 +645,53 @@ static void r300_draw_arrays(struct pipe_context* pipe, unsigned mode,
                     if (!r300_prepare_for_rendering(r300,
                             PREP_VALIDATE_VBOS | PREP_EMIT_AOS, NULL, 9,
                             start, 0))
-                        goto done;
+                        return;
                 }
             } while (count);
         }
     }
-
-done:
-    if (translate) {
-        r300_end_vertex_translate(r300);
-    }
 }
 
 static void r300_draw_vbo(struct pipe_context* pipe,
                           const struct pipe_draw_info *info)
 {
     struct r300_context* r300 = r300_context(pipe);
+    unsigned count = info->count;
+    boolean translate = FALSE;
+    boolean indexed = info->indexed && r300->index_buffer.buffer;
+    unsigned start_indexed = 0;
 
-    if (!r300->velems->count || !r300->vertex_buffer_count)
-	    return;
+    if (r300->skip_rendering) {
+        return;
+    }
 
-    if (info->indexed && r300->index_buffer.buffer) {
-        unsigned offset;
+    if (!u_trim_pipe_prim(info->mode, &count)) {
+        return;
+    }
 
+    /* Index buffer range checking. */
+    if (indexed) {
         assert(r300->index_buffer.offset % r300->index_buffer.index_size == 0);
-        offset = r300->index_buffer.offset / r300->index_buffer.index_size;
 
+        /* Compute start for draw_elements, taking the offset into account. */
+        start_indexed =
+            info->start +
+            (r300->index_buffer.offset / r300->index_buffer.index_size);
+
+        if ((start_indexed + count) * r300->index_buffer.index_size >
+            r300->index_buffer.buffer->width0) {
+            fprintf(stderr, "r300: Invalid index buffer range. Skipping rendering.\n");
+            return;
+        }
+    }
+
+    /* Set up fallback for incompatible vertex layout if needed. */
+    if (r300->incompatible_vb_layout || r300->velems->incompatible_layout) {
+        r300_begin_vertex_translate(r300);
+        translate = TRUE;
+    }
+
+    if (indexed) {
         r300_draw_range_elements(pipe,
                                  r300->index_buffer.buffer,
                                  r300->index_buffer.index_size,
@@ -692,14 +699,17 @@ static void r300_draw_vbo(struct pipe_context* pipe,
                                  info->min_index,
                                  info->max_index,
                                  info->mode,
-                                 info->start + offset,
-                                 info->count);
-    }
-    else {
+                                 start_indexed,
+                                 count);
+    } else {
         r300_draw_arrays(pipe,
                          info->mode,
                          info->start,
-                         info->count);
+                         count);
+    }
+
+    if (translate) {
+        r300_end_vertex_translate(r300);
     }
 }
 
@@ -826,6 +836,7 @@ static boolean r300_render_allocate_vertices(struct vbuf_render* render,
 				       R300_MAX_DRAW_VBO_SIZE);
         r300->draw_vbo_offset = 0;
         r300->draw_vbo_size = R300_MAX_DRAW_VBO_SIZE;
+        r300->validate_buffers = TRUE;
     }
 
     r300render->vertex_size = vertex_size;
@@ -983,7 +994,7 @@ static void r300_render_draw_elements(struct vbuf_render* render,
     end_cs_dwords = r300_get_num_cs_end_dwords(r300);
 
     while (count) {
-        free_dwords = r300->cs->ndw - r300->cs->cdw;
+        free_dwords = R300_MAX_CMDBUF_DWORDS - r300->cs->cdw;
 
         short_count = MIN2(count, (free_dwords - end_cs_dwords - 6) * 2);
 
@@ -1104,6 +1115,8 @@ static void r300_blitter_draw_rectangle(struct blitter_context *blitter,
     const float zeros[4] = {0, 0, 0, 0};
     CS_LOCALS(r300);
 
+    r300->context.set_vertex_buffers(&r300->context, 0, NULL);
+
     if (type == UTIL_BLITTER_ATTRIB_TEXCOORD)
         r300->sprite_coord_enable = 1;
 
@@ -1160,37 +1173,45 @@ static void r300_blitter_draw_rectangle(struct blitter_context *blitter,
 
 done:
     /* Restore the state. */
-    r300->clip_state.dirty = TRUE;
-    r300->rs_state.dirty = TRUE;
-    r300->viewport_state.dirty = TRUE;
+    r300_mark_atom_dirty(r300, &r300->clip_state);
+    r300_mark_atom_dirty(r300, &r300->rs_state);
+    r300_mark_atom_dirty(r300, &r300->viewport_state);
 
     r300->sprite_coord_enable = last_sprite_coord_enable;
 }
 
 static void r300_resource_resolve(struct pipe_context* pipe,
                                   struct pipe_resource* dest,
-                                  struct pipe_subresource subdest,
+                                  unsigned dst_layer,
                                   struct pipe_resource* src,
-                                  struct pipe_subresource subsrc)
+                                  unsigned src_layer)
 {
     struct r300_context* r300 = r300_context(pipe);
+    struct pipe_surface* srcsurf, surf_tmpl;
     struct r300_aa_state *aa = (struct r300_aa_state*)r300->aa_state.state;
-    struct pipe_surface* srcsurf = src->screen->get_tex_surface(src->screen,
-            src, subsrc.face, subsrc.level, 0, 0);
     float color[] = {0, 0, 0, 0};
 
+    memset(&surf_tmpl, 0, sizeof(surf_tmpl));
+    surf_tmpl.format = src->format;
+    surf_tmpl.usage = 0; /* not really a surface hence no bind flags */
+    surf_tmpl.u.tex.level = 0; /* msaa resources cannot have mipmaps */
+    surf_tmpl.u.tex.first_layer = src_layer;
+    surf_tmpl.u.tex.last_layer = src_layer;
+    srcsurf = pipe->create_surface(pipe, src, &surf_tmpl);
+    surf_tmpl.format = dest->format;
+    surf_tmpl.u.tex.first_layer = dst_layer;
+    surf_tmpl.u.tex.last_layer = dst_layer;
+
     DBG(r300, DBG_DRAW, "r300: Resolving resource...\n");
 
     /* Enable AA resolve. */
-    aa->dest = r300_surface(
-            dest->screen->get_tex_surface(dest->screen, dest, subdest.face,
-                                          subdest.level, 0, 0));
+    aa->dest = r300_surface(pipe->create_surface(pipe, dest, &surf_tmpl));
 
     aa->aaresolve_ctl =
         R300_RB3D_AARESOLVE_CTL_AARESOLVE_MODE_RESOLVE |
         R300_RB3D_AARESOLVE_CTL_AARESOLVE_ALPHA_AVERAGE;
     r300->aa_state.size = 12;
-    r300->aa_state.dirty = TRUE;
+    r300_mark_atom_dirty(r300, &r300->aa_state);
 
     /* Resolve the surface. */
     r300->context.clear_render_target(pipe,
@@ -1199,7 +1220,7 @@ static void r300_resource_resolve(struct pipe_context* pipe,
     /* Disable AA resolve. */
     aa->aaresolve_ctl = 0;
     r300->aa_state.size = 4;
-    r300->aa_state.dirty = TRUE;
+    r300_mark_atom_dirty(r300, &r300->aa_state);
 
     pipe_surface_reference((struct pipe_surface**)&srcsurf, NULL);
     pipe_surface_reference((struct pipe_surface**)&aa->dest, NULL);
diff --git a/src/gallium/drivers/r300/r300_render_stencilref.c b/src/gallium/drivers/r300/r300_render_stencilref.c
index 1f035d64a2..747594afaf 100644
--- a/src/gallium/drivers/r300/r300_render_stencilref.c
+++ b/src/gallium/drivers/r300/r300_render_stencilref.c
@@ -66,7 +66,7 @@ static void r300_stencilref_begin(struct r300_context *r300)
     /* We *cull* pixels, therefore no need to mask out the bits. */
     rs->cb_main[rs->cull_mode_index] |= R300_CULL_BACK;
 
-    r300->rs_state.dirty = TRUE;
+    r300_mark_atom_dirty(r300, &r300->rs_state);
 }
 
 /* Set drawing for back faces. */
@@ -80,8 +80,8 @@ static void r300_stencilref_switch_side(struct r300_context *r300)
     dsa->stencil_ref_mask = dsa->stencil_ref_bf;
     r300->stencil_ref.ref_value[0] = r300->stencil_ref.ref_value[1];
 
-    r300->rs_state.dirty = TRUE;
-    r300->dsa_state.dirty = TRUE;
+    r300_mark_atom_dirty(r300, &r300->rs_state);
+    r300_mark_atom_dirty(r300, &r300->dsa_state);
 }
 
 /* Restore the original state. */
@@ -96,8 +96,8 @@ static void r300_stencilref_end(struct r300_context *r300)
     dsa->stencil_ref_mask = sr->zb_stencilrefmask;
     r300->stencil_ref.ref_value[0] = sr->ref_value_front;
 
-    r300->rs_state.dirty = TRUE;
-    r300->dsa_state.dirty = TRUE;
+    r300_mark_atom_dirty(r300, &r300->rs_state);
+    r300_mark_atom_dirty(r300, &r300->dsa_state);
 }
 
 static void r300_stencilref_draw_vbo(struct pipe_context *pipe,
diff --git a/src/gallium/drivers/r300/r300_render_translate.c b/src/gallium/drivers/r300/r300_render_translate.c
index 9247064508..90b2f40be5 100644
--- a/src/gallium/drivers/r300/r300_render_translate.c
+++ b/src/gallium/drivers/r300/r300_render_translate.c
@@ -145,6 +145,7 @@ void r300_begin_vertex_translate(struct r300_context *r300)
             vb->max_index = num_verts - 1;
             vb->stride = key.output_stride;
             r300->tran.vb_slot = i;
+            r300->validate_buffers = TRUE;
             break;
         }
     }
@@ -199,12 +200,14 @@ void r300_translate_index_buffer(struct r300_context *r300,
             util_shorten_ubyte_elts(&r300->context, index_buffer, index_offset, *start, count);
             *index_size = 2;
             *start = 0;
+            r300->validate_buffers = TRUE;
             break;
 
         case 2:
-            if (*start % 2 != 0 || index_offset) {
+            if (index_offset) {
                 util_rebuild_ushort_elts(&r300->context, index_buffer, index_offset, *start, count);
                 *start = 0;
+                r300->validate_buffers = TRUE;
             }
             break;
 
@@ -212,6 +215,7 @@ void r300_translate_index_buffer(struct r300_context *r300,
             if (index_offset) {
                 util_rebuild_uint_elts(&r300->context, index_buffer, index_offset, *start, count);
                 *start = 0;
+                r300->validate_buffers = TRUE;
             }
             break;
     }
diff --git a/src/gallium/drivers/r300/r300_resource.c b/src/gallium/drivers/r300/r300_resource.c
index f6f33028dc..dd1df97059 100644
--- a/src/gallium/drivers/r300/r300_resource.c
+++ b/src/gallium/drivers/r300/r300_resource.c
@@ -58,6 +58,8 @@ void r300_init_resource_functions(struct r300_context *r300)
    r300->context.transfer_destroy = u_transfer_destroy_vtbl;
    r300->context.transfer_inline_write = u_transfer_inline_write_vtbl;
    r300->context.is_resource_referenced = u_is_resource_referenced_vtbl;
+   r300->context.create_surface = r300_create_surface;
+   r300->context.surface_destroy = r300_surface_destroy;
 }
 
 void r300_init_screen_resource_functions(struct r300_screen *r300screen)
@@ -67,7 +69,4 @@ void r300_init_screen_resource_functions(struct r300_screen *r300screen)
    r300screen->screen.resource_get_handle = u_resource_get_handle_vtbl;
    r300screen->screen.resource_destroy = u_resource_destroy_vtbl;
    r300screen->screen.user_buffer_create = r300_user_buffer_create;
-
-   r300screen->screen.get_tex_surface = r300_get_tex_surface;
-   r300screen->screen.tex_surface_destroy = r300_tex_surface_destroy;
 }
diff --git a/src/gallium/drivers/r300/r300_screen.c b/src/gallium/drivers/r300/r300_screen.c
index 37563b5a94..921d6f1e67 100644
--- a/src/gallium/drivers/r300/r300_screen.c
+++ b/src/gallium/drivers/r300/r300_screen.c
@@ -32,6 +32,8 @@
 #include "r300_winsys.h"
 #include "r300_public.h"
 
+#include "draw/draw_context.h"
+
 /* Return the identifier behind whom the brave coders responsible for this
  * amalgamation of code, sweat, and duct tape, routinely obscure their names.
  *
@@ -114,8 +116,9 @@ static int r300_get_param(struct pipe_screen* pscreen, enum pipe_cap param)
         case PIPE_CAP_TEXTURE_MIRROR_CLAMP:
         case PIPE_CAP_TEXTURE_MIRROR_REPEAT:
         case PIPE_CAP_BLEND_EQUATION_SEPARATE:
-        case PIPE_CAP_TEXTURE_SWIZZLE:
             return 1;
+        case PIPE_CAP_TEXTURE_SWIZZLE:
+            return util_format_s3tc_enabled ? r300screen->caps.dxtc_swizzle : 1;
 
         /* Unsupported features (boolean caps). */
         case PIPE_CAP_TIMER_QUERY:
@@ -212,9 +215,15 @@ static int r300_get_shader_param(struct pipe_screen *pscreen, unsigned shader, e
         case PIPE_SHADER_CAP_INDIRECT_TEMP_ADDR:
         case PIPE_SHADER_CAP_INDIRECT_CONST_ADDR:
             return 0;
+        case PIPE_SHADER_CAP_SUBROUTINES:
+            return 0;
         }
         break;
     case PIPE_SHADER_VERTEX:
+        if (!r300screen->caps.has_tcl) {
+            return draw_get_shader_param(shader, param);
+        }
+
         switch (param)
         {
         case PIPE_SHADER_CAP_MAX_INSTRUCTIONS:
@@ -245,6 +254,8 @@ static int r300_get_shader_param(struct pipe_screen *pscreen, unsigned shader, e
             return 0;
         case PIPE_SHADER_CAP_INDIRECT_CONST_ADDR:
             return 1;
+        case PIPE_SHADER_CAP_SUBROUTINES:
+            return 0;
         default:
             break;
         }
@@ -277,6 +288,13 @@ static float r300_get_paramf(struct pipe_screen* pscreen, enum pipe_cap param)
             return 16.0f;
         case PIPE_CAP_MAX_TEXTURE_LOD_BIAS:
             return 16.0f;
+        case PIPE_CAP_GUARD_BAND_LEFT:
+        case PIPE_CAP_GUARD_BAND_TOP:
+        case PIPE_CAP_GUARD_BAND_RIGHT:
+        case PIPE_CAP_GUARD_BAND_BOTTOM:
+            /* XXX I don't know what these should be but the least we can do is
+             * silence the potential error message */
+            return 0.0f;
         default:
             debug_printf("r300: Warning: Unknown CAP %d in get_paramf.\n",
                          param);
@@ -382,7 +400,7 @@ static void r300_destroy_screen(struct pipe_screen* pscreen)
     struct r300_screen* r300screen = r300_screen(pscreen);
     struct r300_winsys_screen *rws = r300_winsys_screen(pscreen);
 
-    util_mempool_destroy(&r300screen->pool_buffers);
+    util_slab_destroy(&r300screen->pool_buffers);
 
     if (rws)
       rws->destroy(rws);
@@ -439,9 +457,13 @@ struct pipe_screen* r300_screen_create(struct r300_winsys_screen *rws)
     r300_init_debug(r300screen);
     r300_parse_chipset(&r300screen->caps);
 
-    util_mempool_create(&r300screen->pool_buffers,
-                        sizeof(struct r300_buffer), 64,
-                        UTIL_MEMPOOL_SINGLETHREADED);
+    r300screen->caps.index_bias_supported =
+            r300screen->caps.is_r500 &&
+            rws->get_value(rws, R300_VID_DRM_2_3_0);
+
+    util_slab_create(&r300screen->pool_buffers,
+                     sizeof(struct r300_buffer), 64,
+                     UTIL_SLAB_SINGLETHREADED);
 
     r300screen->rws = rws;
     r300screen->screen.winsys = (struct pipe_winsys*)rws;
diff --git a/src/gallium/drivers/r300/r300_screen.h b/src/gallium/drivers/r300/r300_screen.h
index 8b7f1fab61..5847fe1ffc 100644
--- a/src/gallium/drivers/r300/r300_screen.h
+++ b/src/gallium/drivers/r300/r300_screen.h
@@ -28,7 +28,7 @@
 
 #include "r300_chipset.h"
 
-#include "util/u_mempool.h"
+#include "util/u_slab.h"
 
 #include <stdio.h>
 
@@ -44,7 +44,7 @@ struct r300_screen {
     struct r300_capabilities caps;
 
     /* Memory pools. */
-    struct util_mempool pool_buffers;
+    struct util_slab_mempool pool_buffers;
 
     /** Combination of DBG_xxx flags */
     unsigned debug;
@@ -101,7 +101,6 @@ r300_winsys_screen(struct pipe_screen *screen) {
 #define DBG_NO_OPT      (1 << 20)
 #define DBG_NO_CBZB     (1 << 21)
 /* Statistics. */
-#define DBG_STATS       (1 << 24)
 #define DBG_P_STAT      (1 << 25)
 /*@}*/
 
diff --git a/src/gallium/drivers/r300/r300_screen_buffer.c b/src/gallium/drivers/r300/r300_screen_buffer.c
index 37a080ba48..4436443522 100644
--- a/src/gallium/drivers/r300/r300_screen_buffer.c
+++ b/src/gallium/drivers/r300/r300_screen_buffer.c
@@ -43,7 +43,7 @@ unsigned r300_buffer_is_referenced(struct pipe_context *context,
     if (r300_buffer_is_user_buffer(buf))
  	return PIPE_UNREFERENCED;
 
-    if (r300->rws->cs_is_buffer_referenced(r300->cs, rbuf->buf, domain))
+    if (r300->rws->cs_is_buffer_referenced(r300->cs, rbuf->cs_buf, domain))
         return PIPE_REFERENCED_FOR_READ | PIPE_REFERENCED_FOR_WRITE;
 
     return PIPE_UNREFERENCED;
@@ -51,7 +51,7 @@ unsigned r300_buffer_is_referenced(struct pipe_context *context,
 
 static unsigned r300_buffer_is_referenced_by_cs(struct pipe_context *context,
                                                 struct pipe_resource *buf,
-                                                unsigned face, unsigned level)
+                                                unsigned level, int layer)
 {
     return r300_buffer_is_referenced(context, buf, R300_REF_CS);
 }
@@ -118,6 +118,7 @@ int r300_upload_user_buffers(struct r300_context *r300)
             pipe_resource_reference(&vb->buffer, NULL);
             vb->buffer = upload_buffer;
             vb->buffer_offset = upload_offset;
+            r300->validate_buffers = TRUE;
         }
     }
     return ret;
@@ -136,26 +137,26 @@ static void r300_buffer_destroy(struct pipe_screen *screen,
     if (rbuf->buf)
         rws->buffer_reference(rws, &rbuf->buf, NULL);
 
-    util_mempool_free(&r300screen->pool_buffers, rbuf);
+    util_slab_free(&r300screen->pool_buffers, rbuf);
 }
 
 static struct pipe_transfer*
-r300_default_get_transfer(struct pipe_context *context,
-                          struct pipe_resource *resource,
-                          struct pipe_subresource sr,
-                          unsigned usage,
-                          const struct pipe_box *box)
+r300_buffer_get_transfer(struct pipe_context *context,
+                         struct pipe_resource *resource,
+                         unsigned level,
+                         unsigned usage,
+                         const struct pipe_box *box)
 {
    struct r300_context *r300 = r300_context(context);
    struct pipe_transfer *transfer =
-         util_mempool_malloc(&r300->pool_transfers);
+         util_slab_alloc(&r300->pool_transfers);
 
    transfer->resource = resource;
-   transfer->sr = sr;
+   transfer->level = level;
    transfer->usage = usage;
    transfer->box = *box;
    transfer->stride = 0;
-   transfer->slice_stride = 0;
+   transfer->layer_stride = 0;
    transfer->data = NULL;
 
    /* Note strides are zero, this is ok for buffers, but not for
@@ -164,11 +165,11 @@ r300_default_get_transfer(struct pipe_context *context,
    return transfer;
 }
 
-static void r300_default_transfer_destroy(struct pipe_context *pipe,
-                                          struct pipe_transfer *transfer)
+static void r300_buffer_transfer_destroy(struct pipe_context *pipe,
+                                         struct pipe_transfer *transfer)
 {
    struct r300_context *r300 = r300_context(pipe);
-   util_mempool_free(&r300->pool_transfers, transfer);
+   util_slab_free(&r300->pool_transfers, transfer);
 }
 
 static void *
@@ -206,6 +207,9 @@ r300_buffer_transfer_map( struct pipe_context *pipe,
                                                    rbuf->b.b.bind,
                                                    rbuf->b.b.usage,
                                                    rbuf->domain);
+                rbuf->cs_buf =
+                    r300screen->rws->buffer_get_cs_handle(r300screen->rws,
+                                                          rbuf->buf);
 		break;
 	    }
 	}
@@ -265,17 +269,45 @@ static void r300_buffer_transfer_unmap( struct pipe_context *pipe,
     }
 }
 
+static void r300_buffer_transfer_inline_write(struct pipe_context *pipe,
+                                              struct pipe_resource *resource,
+                                              unsigned level,
+                                              unsigned usage,
+                                              const struct pipe_box *box,
+                                              const void *data,
+                                              unsigned stride,
+                                              unsigned layer_stride)
+{
+    struct r300_buffer *rbuf = r300_buffer(resource);
+    struct pipe_transfer *transfer = NULL;
+    uint8_t *map = NULL;
+
+    if (rbuf->constant_buffer) {
+        memcpy(rbuf->constant_buffer + box->x, data, box->width);
+        return;
+    }
+
+    transfer = r300_buffer_get_transfer(pipe, resource, 0,
+                        PIPE_TRANSFER_WRITE | PIPE_TRANSFER_DISCARD, box);
+    map = r300_buffer_transfer_map(pipe, transfer);
+
+    memcpy(map, data, box->width);
+
+    r300_buffer_transfer_unmap(pipe, transfer);
+    r300_buffer_transfer_destroy(pipe, transfer);
+}
+
 struct u_resource_vtbl r300_buffer_vtbl = 
 {
    u_default_resource_get_handle,      /* get_handle */
    r300_buffer_destroy,                /* resource_destroy */
    r300_buffer_is_referenced_by_cs,    /* is_buffer_referenced */
-   r300_default_get_transfer,          /* get_transfer */
-   r300_default_transfer_destroy,      /* transfer_destroy */
+   r300_buffer_get_transfer,           /* get_transfer */
+   r300_buffer_transfer_destroy,       /* transfer_destroy */
    r300_buffer_transfer_map,           /* transfer_map */
    r300_buffer_transfer_flush_region,  /* transfer_flush_region */
    r300_buffer_transfer_unmap,         /* transfer_unmap */
-   u_default_transfer_inline_write     /* transfer_inline_write */
+   r300_buffer_transfer_inline_write   /* transfer_inline_write */
 };
 
 struct pipe_resource *r300_buffer_create(struct pipe_screen *screen,
@@ -285,7 +317,7 @@ struct pipe_resource *r300_buffer_create(struct pipe_screen *screen,
     struct r300_buffer *rbuf;
     unsigned alignment = 16;
 
-    rbuf = util_mempool_malloc(&r300screen->pool_buffers);
+    rbuf = util_slab_alloc(&r300screen->pool_buffers);
 
     rbuf->magic = R300_BUFFER_MAGIC;
 
@@ -310,9 +342,11 @@ struct pipe_resource *r300_buffer_create(struct pipe_screen *screen,
                                        rbuf->b.b.width0, alignment,
                                        rbuf->b.b.bind, rbuf->b.b.usage,
                                        rbuf->domain);
+    rbuf->cs_buf =
+        r300screen->rws->buffer_get_cs_handle(r300screen->rws, rbuf->buf);
 
     if (!rbuf->buf) {
-        util_mempool_free(&r300screen->pool_buffers, rbuf);
+        util_slab_free(&r300screen->pool_buffers, rbuf);
         return NULL;
     }
 
@@ -327,7 +361,7 @@ struct pipe_resource *r300_user_buffer_create(struct pipe_screen *screen,
     struct r300_screen *r300screen = r300_screen(screen);
     struct r300_buffer *rbuf;
 
-    rbuf = util_mempool_malloc(&r300screen->pool_buffers);
+    rbuf = util_slab_alloc(&r300screen->pool_buffers);
 
     rbuf->magic = R300_BUFFER_MAGIC;
 
@@ -341,6 +375,7 @@ struct pipe_resource *r300_user_buffer_create(struct pipe_screen *screen,
     rbuf->b.b.width0 = bytes;
     rbuf->b.b.height0 = 1;
     rbuf->b.b.depth0 = 1;
+    rbuf->b.b.array_size = 1;
     rbuf->b.b.flags = 0;
     rbuf->domain = R300_DOMAIN_GTT;
     rbuf->num_ranges = 0;
diff --git a/src/gallium/drivers/r300/r300_screen_buffer.h b/src/gallium/drivers/r300/r300_screen_buffer.h
index cafa9f96f2..0b3555dd81 100644
--- a/src/gallium/drivers/r300/r300_screen_buffer.h
+++ b/src/gallium/drivers/r300/r300_screen_buffer.h
@@ -51,11 +51,12 @@ struct r300_buffer
     uint32_t magic;
 
     struct r300_winsys_buffer *buf;
+    struct r300_winsys_cs_buffer *cs_buf;
 
     enum r300_buffer_domain domain;
 
-    void *user_buffer;
-    void *constant_buffer;
+    uint8_t *user_buffer;
+    uint8_t *constant_buffer;
     struct r300_buffer_range ranges[R300_BUFFER_MAX_RANGES];
     unsigned num_ranges;
 };
@@ -86,11 +87,7 @@ unsigned r300_buffer_is_referenced(struct pipe_context *context,
 
 static INLINE struct r300_buffer *r300_buffer(struct pipe_resource *buffer)
 {
-    if (buffer) {
-	assert(((struct r300_buffer *)buffer)->magic == R300_BUFFER_MAGIC);
-	return (struct r300_buffer *)buffer;
-    }
-    return NULL;
+    return (struct r300_buffer *)buffer;
 }
 
 static INLINE boolean r300_buffer_is_user_buffer(struct pipe_resource *buffer)
diff --git a/src/gallium/drivers/r300/r300_state.c b/src/gallium/drivers/r300/r300_state.c
index bd08bf2d3f..f902db54cc 100644
--- a/src/gallium/drivers/r300/r300_state.c
+++ b/src/gallium/drivers/r300/r300_state.c
@@ -52,7 +52,7 @@
 #define UPDATE_STATE(cso, atom) \
     if (cso != atom.state) { \
         atom.state = cso;    \
-        atom.dirty = TRUE;   \
+        r300_mark_atom_dirty(r300, &(atom));   \
     }
 
 static boolean blend_discard_if_src_alpha_0(unsigned srcRGB, unsigned srcA,
@@ -417,7 +417,7 @@ static void r300_set_blend_color(struct pipe_context* pipe,
         END_CB;
     }
 
-    r300->blend_color_state.dirty = TRUE;
+    r300_mark_atom_dirty(r300, &r300->blend_color_state);
 }
 
 static void r300_set_clip_state(struct pipe_context* pipe,
@@ -446,7 +446,7 @@ static void r300_set_clip_state(struct pipe_context* pipe,
                 (state->depth_clamp ? R300_CLIP_DISABLE : 0));
         END_CB;
 
-        r300->clip_state.dirty = TRUE;
+        r300_mark_atom_dirty(r300, &r300->clip_state);
     } else {
         draw_set_clip_state(r300->draw, state);
     }
@@ -594,7 +594,7 @@ static void r300_bind_dsa_state(struct pipe_context* pipe,
 
     UPDATE_STATE(state, r300->dsa_state);
 
-    r300->hyperz_state.dirty = TRUE; /* Will be updated before the emission. */
+    r300_mark_atom_dirty(r300, &r300->hyperz_state); /* Will be updated before the emission. */
     r300_dsa_inject_stencilref(r300);
 }
 
@@ -613,7 +613,7 @@ static void r300_set_stencil_ref(struct pipe_context* pipe,
     r300->stencil_ref = *sr;
 
     r300_dsa_inject_stencilref(r300);
-    r300->dsa_state.dirty = TRUE;
+    r300_mark_atom_dirty(r300, &r300->dsa_state);
 }
 
 static void r300_tex_set_tiling_flags(struct r300_context *r300,
@@ -626,7 +626,7 @@ static void r300_tex_set_tiling_flags(struct r300_context *r300,
         /* Tiling determines how DRM treats the buffer data.
          * We must flush CS when changing it if the buffer is referenced. */
         if (r300->rws->cs_is_buffer_referenced(r300->cs,
-                                               tex->buffer, R300_REF_CS))
+                                               tex->cs_buffer, R300_REF_CS))
             r300->context.flush(&r300->context, 0, NULL);
 
         r300->rws->buffer_set_tiling(r300->rws, tex->buffer,
@@ -647,12 +647,12 @@ static void r300_fb_set_tiling_flags(struct r300_context *r300,
     for (i = 0; i < state->nr_cbufs; i++) {
         r300_tex_set_tiling_flags(r300,
                                   r300_texture(state->cbufs[i]->texture),
-                                  state->cbufs[i]->level);
+                                  state->cbufs[i]->u.tex.level);
     }
     if (state->zsbuf) {
         r300_tex_set_tiling_flags(r300,
                                   r300_texture(state->zsbuf->texture),
-                                  state->zsbuf->level);
+                                  state->zsbuf->u.tex.level);
     }
 }
 
@@ -663,14 +663,14 @@ static void r300_print_fb_surf_info(struct pipe_surface *surf, unsigned index,
     struct r300_texture *rtex = r300_texture(tex);
 
     fprintf(stderr,
-            "r300:   %s[%i] Dim: %ix%i, Offset: %i, ZSlice: %i, "
-            "Face: %i, Level: %i, Format: %s\n"
+            "r300:   %s[%i] Dim: %ix%i, Firstlayer: %i, "
+            "Lastlayer: %i, Level: %i, Format: %s\n"
 
             "r300:     TEX: Macro: %s, Micro: %s, Pitch: %i, "
             "Dim: %ix%ix%i, LastLevel: %i, Format: %s\n",
 
-            binding, index, surf->width, surf->height, surf->offset,
-            surf->zslice, surf->face, surf->level,
+            binding, index, surf->width, surf->height,
+            surf->u.tex.first_layer, surf->u.tex.last_layer, surf->u.tex.level,
             util_format_short_name(surf->format),
 
             rtex->desc.macrotile[0] ? "YES" : " NO",
@@ -687,13 +687,13 @@ void r300_mark_fb_state_dirty(struct r300_context *r300,
     boolean can_hyperz = r300->rws->get_value(r300->rws, R300_CAN_HYPERZ);
 
     /* What is marked as dirty depends on the enum r300_fb_state_change. */
-    r300->gpu_flush.dirty = TRUE;
-    r300->fb_state.dirty = TRUE;
-    r300->hyperz_state.dirty = TRUE;
+    r300_mark_atom_dirty(r300, &r300->gpu_flush);
+    r300_mark_atom_dirty(r300, &r300->fb_state);
+    r300_mark_atom_dirty(r300, &r300->hyperz_state);
 
     if (change == R300_CHANGED_FB_STATE) {
-        r300->aa_state.dirty = TRUE;
-        r300->fb_state_pipelined.dirty = TRUE;
+        r300_mark_atom_dirty(r300, &r300->aa_state);
+        r300_mark_atom_dirty(r300, &r300->fb_state_pipelined);
     }
 
     /* Now compute the fb_state atom size. */
@@ -738,11 +738,11 @@ static void
 
     /* If nr_cbufs is changed from zero to non-zero or vice versa... */
     if (!!old_state->nr_cbufs != !!state->nr_cbufs) {
-        r300->blend_state.dirty = TRUE;
+        r300_mark_atom_dirty(r300, &r300->blend_state);
     }
     /* If zsbuf is set from NULL to non-NULL or vice versa.. */
     if (!!old_state->zsbuf != !!state->zsbuf) {
-        r300->dsa_state.dirty = TRUE;
+        r300_mark_atom_dirty(r300, &r300->dsa_state);
     }
 
     /* The tiling flags are dependent on the surface miplevel, unfortunately. */
@@ -751,6 +751,7 @@ static void
     util_copy_framebuffer_state(r300->fb_state.state, state);
 
     r300_mark_fb_state_dirty(r300, R300_CHANGED_FB_STATE);
+    r300->validate_buffers = TRUE;
 
     r300->z_compression = false;
     
@@ -768,7 +769,7 @@ static void
             struct r300_surface *zs_surf = r300_surface(state->zsbuf);
             struct r300_texture *tex;
             int compress = r300->screen->caps.is_rv350 ? RV350_Z_COMPRESS_88 : R300_Z_COMPRESS_44;
-            int level = zs_surf->base.level;
+            int level = zs_surf->base.u.tex.level;
 
             tex = r300_texture(zs_surf->base.texture);
 
@@ -795,7 +796,7 @@ static void
             r300->zbuffer_bpp = zbuffer_bpp;
 
             if (r300->polygon_offset_enabled)
-                r300->rs_state.dirty = TRUE;
+                r300_mark_atom_dirty(r300, &r300->rs_state);
         }
     }
 
@@ -853,9 +854,9 @@ void r300_mark_fs_code_dirty(struct r300_context *r300)
 {
     struct r300_fragment_shader* fs = r300_fs(r300);
 
-    r300->fs.dirty = TRUE;
-    r300->fs_rc_constant_state.dirty = TRUE;
-    r300->fs_constants.dirty = TRUE;
+    r300_mark_atom_dirty(r300, &r300->fs);
+    r300_mark_atom_dirty(r300, &r300->fs_rc_constant_state);
+    r300_mark_atom_dirty(r300, &r300->fs_constants);
     r300->fs.size = fs->shader->cb_code_size;
 
     if (r300->screen->caps.is_r500) {
@@ -885,7 +886,7 @@ static void r300_bind_fs_state(struct pipe_context* pipe, void* shader)
     r300_pick_fragment_shader(r300);
     r300_mark_fs_code_dirty(r300);
 
-    r300->rs_block_state.dirty = TRUE; /* Will be updated before the emission. */
+    r300_mark_atom_dirty(r300, &r300->rs_block_state); /* Will be updated before the emission. */
 }
 
 /* Delete fragment shader state. */
@@ -1137,7 +1138,7 @@ static void r300_bind_rs_state(struct pipe_context* pipe, void* state)
 
     if (last_sprite_coord_enable != r300->sprite_coord_enable ||
         last_two_sided_color != r300->two_sided_color) {
-        r300->rs_block_state.dirty = TRUE;
+        r300_mark_atom_dirty(r300, &r300->rs_block_state);
     }
 }
 
@@ -1235,7 +1236,7 @@ static void r300_bind_sampler_states(struct pipe_context* pipe,
     memcpy(state->sampler_states, states, sizeof(void*) * count);
     state->sampler_state_count = count;
 
-    r300->textures_state.dirty = TRUE;
+    r300_mark_atom_dirty(r300, &r300->textures_state);
 }
 
 static void r300_lacks_vertex_textures(struct pipe_context* pipe,
@@ -1297,29 +1298,27 @@ static void r300_set_fragment_sampler_views(struct pipe_context* pipe,
     }
 
     for (i = 0; i < count; i++) {
-        if (&state->sampler_views[i]->base != views[i]) {
-            pipe_sampler_view_reference(
-                    (struct pipe_sampler_view**)&state->sampler_views[i],
-                    views[i]);
+        pipe_sampler_view_reference(
+                (struct pipe_sampler_view**)&state->sampler_views[i],
+                views[i]);
 
-            if (!views[i]) {
-                continue;
-            }
+        if (!views[i]) {
+            continue;
+        }
 
-            /* A new sampler view (= texture)... */
-            dirty_tex = TRUE;
+        /* A new sampler view (= texture)... */
+        dirty_tex = TRUE;
 
-            /* Set the texrect factor in the fragment shader.
+        /* Set the texrect factor in the fragment shader.
              * Needed for RECT and NPOT fallback. */
-            texture = r300_texture(views[i]->texture);
-            if (texture->desc.is_npot) {
-                r300->fs_rc_constant_state.dirty = TRUE;
-            }
+        texture = r300_texture(views[i]->texture);
+        if (texture->desc.is_npot) {
+            r300_mark_atom_dirty(r300, &r300->fs_rc_constant_state);
+        }
 
-            state->sampler_views[i]->texcache_region =
+        state->sampler_views[i]->texcache_region =
                 r300_assign_texture_cache_region(view_index, real_num_views);
-            view_index++;
-        }
+        view_index++;
     }
 
     for (i = count; i < tex_units; i++) {
@@ -1332,10 +1331,11 @@ static void r300_set_fragment_sampler_views(struct pipe_context* pipe,
 
     state->sampler_view_count = count;
 
-    r300->textures_state.dirty = TRUE;
+    r300_mark_atom_dirty(r300, &r300->textures_state);
+    r300->validate_buffers = TRUE;
 
     if (dirty_tex) {
-        r300->texture_cache_inval.dirty = TRUE;
+        r300_mark_atom_dirty(r300, &r300->texture_cache_inval);
     }
 }
 
@@ -1347,6 +1347,7 @@ r300_create_sampler_view(struct pipe_context *pipe,
     struct r300_sampler_view *view = CALLOC_STRUCT(r300_sampler_view);
     struct r300_texture *tex = r300_texture(texture);
     boolean is_r500 = r300_screen(pipe->screen)->caps.is_r500;
+    boolean dxtc_swizzle = r300_screen(pipe->screen)->caps.dxtc_swizzle;
 
     if (view) {
         view->base = *templ;
@@ -1363,7 +1364,8 @@ r300_create_sampler_view(struct pipe_context *pipe,
         view->format = tex->tx_format;
         view->format.format1 |= r300_translate_texformat(templ->format,
                                                          view->swizzle,
-                                                         is_r500);
+                                                         is_r500,
+                                                         dxtc_swizzle);
         if (is_r500) {
             view->format.format2 |= r500_tx_format_msb_bit(templ->format);
         }
@@ -1388,7 +1390,7 @@ static void r300_set_scissor_state(struct pipe_context* pipe,
     memcpy(r300->scissor_state.state, state,
         sizeof(struct pipe_scissor_state));
 
-    r300->scissor_state.dirty = TRUE;
+    r300_mark_atom_dirty(r300, &r300->scissor_state);
 }
 
 static void r300_set_viewport_state(struct pipe_context* pipe,
@@ -1434,9 +1436,9 @@ static void r300_set_viewport_state(struct pipe_context* pipe,
         viewport->vte_control |= R300_VPORT_Z_OFFSET_ENA;
     }
 
-    r300->viewport_state.dirty = TRUE;
+    r300_mark_atom_dirty(r300, &r300->viewport_state);
     if (r300->fs.state && r300_fs(r300)->shader->inputs.wpos != ATTR_UNUSED) {
-        r300->fs_rc_constant_state.dirty = TRUE;
+        r300_mark_atom_dirty(r300, &r300->fs_rc_constant_state);
     }
 }
 
@@ -1448,6 +1450,15 @@ static void r300_set_vertex_buffers(struct pipe_context* pipe,
     struct pipe_vertex_buffer *vbo;
     unsigned i, max_index = (1 << 24) - 1;
     boolean any_user_buffer = FALSE;
+    struct pipe_vertex_buffer dummy_vb = {0};
+
+    /* There must be at least one vertex buffer set, otherwise it locks up. */
+    if (!count) {
+        dummy_vb.buffer = r300->dummy_vb;
+        dummy_vb.max_index = r300->dummy_vb->width0 / 4;
+        buffers = &dummy_vb;
+        count = 1;
+    }
 
     if (count == r300->vertex_buffer_count &&
         memcmp(r300->vertex_buffer, buffers,
@@ -1483,14 +1494,14 @@ static void r300_set_vertex_buffers(struct pipe_context* pipe,
                 any_user_buffer = TRUE;
             }
 
+            /* The stride of zero means we will be fetching only the first
+             * vertex, so don't care about max_index. */
+            if (!vbo->stride)
+                continue;
+
             if (vbo->max_index == ~0) {
-                /* if no VBO stride then only one vertex value so max index is 1 */
-                /* should think about converting to VS constants like svga does */
-                if (!vbo->stride)
-                    vbo->max_index = 1;
-                else
-                    vbo->max_index =
-                             (vbo->buffer->width0 - vbo->buffer_offset) / vbo->stride;
+                vbo->max_index =
+                        (vbo->buffer->width0 - vbo->buffer_offset) / vbo->stride;
             }
 
             max_index = MIN2(vbo->max_index, max_index);
@@ -1498,7 +1509,8 @@ static void r300_set_vertex_buffers(struct pipe_context* pipe,
 
         r300->any_user_vbs = any_user_buffer;
         r300->vertex_buffer_max_index = max_index;
-
+        r300->aos_dirty = TRUE;
+        r300->validate_buffers = TRUE;
     } else {
         /* SW TCL. */
         draw_set_vertex_buffers(r300->draw, count, buffers);
@@ -1534,10 +1546,10 @@ static void r300_set_index_buffer(struct pipe_context* pipe,
     }
 
     if (r300->screen->caps.has_tcl) {
-       /* TODO make this more like a state */
+        r300->validate_buffers = TRUE;
     }
     else {
-       draw_set_index_buffer(r300->draw, ib);
+        draw_set_index_buffer(r300->draw, ib);
     }
 }
 
@@ -1601,6 +1613,14 @@ static void* r300_create_vertex_elements_state(struct pipe_context* pipe,
     struct r300_vertex_element_state *velems;
     unsigned i;
     enum pipe_format *format;
+    struct pipe_vertex_element dummy_attrib = {0};
+
+    /* R300 Programmable Stream Control (PSC) doesn't support 0 vertex elements. */
+    if (!count) {
+        dummy_attrib.src_format = PIPE_FORMAT_R8G8B8A8_UNORM;
+        attribs = &dummy_attrib;
+        count = 1;
+    }
 
     assert(count <= PIPE_MAX_ATTRIBS);
     velems = CALLOC_STRUCT(r300_vertex_element_state);
@@ -1667,7 +1687,8 @@ static void* r300_create_vertex_elements_state(struct pipe_context* pipe,
              * swizzles are already set up.
              * Also compute the vertex size. */
             for (i = 0; i < count; i++) {
-                /* This is OK because we check for aligned strides too. */
+                /* This is OK because we check for aligned strides too
+                 * elsewhere. */
                 velems->hw_format_size[i] =
                     align(util_format_get_blocksize(velems->hw_format[i]), 4);
                 velems->vertex_size_dwords += velems->hw_format_size[i] / 4;
@@ -1696,6 +1717,7 @@ static void r300_bind_vertex_elements_state(struct pipe_context *pipe,
 
     UPDATE_STATE(&velems->vertex_stream, r300->vertex_stream_state);
     r300->vertex_stream_state.size = (1 + velems->vertex_stream.count) * 2;
+    r300->aos_dirty = TRUE;
 }
 
 static void r300_delete_vertex_elements_state(struct pipe_context *pipe, void *state)
@@ -1738,27 +1760,25 @@ static void r300_bind_vs_state(struct pipe_context* pipe, void* shader)
     r300->vs_state.state = vs;
 
     /* The majority of the RS block bits is dependent on the vertex shader. */
-    r300->rs_block_state.dirty = TRUE; /* Will be updated before the emission. */
+    r300_mark_atom_dirty(r300, &r300->rs_block_state); /* Will be updated before the emission. */
 
     if (r300->screen->caps.has_tcl) {
         unsigned fc_op_dwords = r300->screen->caps.is_r500 ? 3 : 2;
-        r300->vs_state.dirty = TRUE;
+        r300_mark_atom_dirty(r300, &r300->vs_state);
         r300->vs_state.size =
                 vs->code.length + 9 +
-                (vs->immediates_count ? vs->immediates_count * 4 + 3 : 0) +
         (vs->code.num_fc_ops ? vs->code.num_fc_ops * fc_op_dwords + 4 : 0);
 
-        if (vs->externals_count) {
-            r300->vs_constants.dirty = TRUE;
-            r300->vs_constants.size = vs->externals_count * 4 + 3;
-        } else {
-            r300->vs_constants.size = 0;
-        }
+        r300_mark_atom_dirty(r300, &r300->vs_constants);
+        r300->vs_constants.size =
+                2 +
+                (vs->externals_count ? vs->externals_count * 4 + 3 : 0) +
+                (vs->immediates_count ? vs->immediates_count * 4 + 3 : 0);
 
         ((struct r300_constant_buffer*)r300->vs_constants.state)->remap_table =
                 vs->code.constants_remap_table;
 
-        r300->pvs_flush.dirty = TRUE;
+        r300_mark_atom_dirty(r300, &r300->pvs_flush);
     } else {
         draw_bind_vertex_shader(r300->draw,
                 (struct draw_vertex_shader*)vs->draw_vs);
@@ -1799,33 +1819,44 @@ static void r300_set_constant_buffer(struct pipe_context *pipe,
             cbuf = (struct r300_constant_buffer*)r300->fs_constants.state;
             break;
         default:
-            assert(0);
             return;
     }
 
     if (buf == NULL || buf->width0 == 0 ||
-        (mapped = r300_buffer(buf)->constant_buffer) == NULL) {
+        (mapped = (uint32_t*)r300_buffer(buf)->constant_buffer) == NULL) {
         return;
     }
 
     if (shader == PIPE_SHADER_FRAGMENT ||
         (shader == PIPE_SHADER_VERTEX && r300->screen->caps.has_tcl)) {
         assert((buf->width0 % (4 * sizeof(float))) == 0);
-        cbuf->ptr = mapped + index*4;
+        cbuf->ptr = mapped;
     }
 
     if (shader == PIPE_SHADER_VERTEX) {
         if (r300->screen->caps.has_tcl) {
-            if (r300->vs_constants.size) {
-                r300->vs_constants.dirty = TRUE;
+            struct r300_vertex_shader *vs =
+                    (struct r300_vertex_shader*)r300->vs_state.state;
+
+            if (!vs) {
+                cbuf->buffer_base = 0;
+                return;
+            }
+
+            cbuf->buffer_base = r300->vs_const_base;
+            r300->vs_const_base += vs->code.constants.Count;
+            if (r300->vs_const_base > R500_MAX_PVS_CONST_VECS) {
+                r300->vs_const_base = vs->code.constants.Count;
+                cbuf->buffer_base = 0;
+                r300_mark_atom_dirty(r300, &r300->pvs_flush);
             }
-            r300->pvs_flush.dirty = TRUE;
+            r300_mark_atom_dirty(r300, &r300->vs_constants);
         } else if (r300->draw) {
             draw_set_mapped_constant_buffer(r300->draw, PIPE_SHADER_VERTEX,
                 0, mapped, buf->width0);
         }
     } else if (shader == PIPE_SHADER_FRAGMENT) {
-        r300->fs_constants.dirty = TRUE;
+        r300_mark_atom_dirty(r300, &r300->fs_constants);
     }
 }
 
diff --git a/src/gallium/drivers/r300/r300_state_derived.c b/src/gallium/drivers/r300/r300_state_derived.c
index 50366e32c2..d5fc8ece25 100644
--- a/src/gallium/drivers/r300/r300_state_derived.c
+++ b/src/gallium/drivers/r300/r300_state_derived.c
@@ -25,6 +25,7 @@
 
 #include "util/u_math.h"
 #include "util/u_memory.h"
+#include "util/u_pack_color.h"
 
 #include "r300_context.h"
 #include "r300_fs.h"
@@ -192,7 +193,7 @@ static void r300_swtcl_vertex_psc(struct r300_context *r300)
         (R300_LAST_VEC << (i & 1 ? 16 : 0));
 
     vstream->count = (i >> 1) + 1;
-    r300->vertex_stream_state.dirty = TRUE;
+    r300_mark_atom_dirty(r300, &r300->vertex_stream_state);
     r300->vertex_stream_state.size = (1 + vstream->count) * 2;
 }
 
@@ -433,6 +434,8 @@ static void r300_update_rs_block(struct r300_context *r300)
         fp_offset++;
         col_count++;
         DBG(r300, DBG_RS, "r300: Rasterized FACE written to FS.\n");
+    } else if (fs_inputs->face != ATTR_UNUSED) {
+        fprintf(stderr, "r300: ERROR: FS input FACE unassigned.\n");
     }
 
     /* Rasterize texture coordinates. */
@@ -484,12 +487,10 @@ static void r300_update_rs_block(struct r300_context *r300)
         }
     }
 
-    if (DBG_ON(r300, DBG_RS)) {
-        for (; i < ATTR_GENERIC_COUNT; i++) {
-            if (fs_inputs->generic[i] != ATTR_UNUSED) {
-                DBG(r300, DBG_RS,
-                    "r300: FS input generic %i unassigned.\n", i);
-            }
+    for (; i < ATTR_GENERIC_COUNT; i++) {
+        if (fs_inputs->generic[i] != ATTR_UNUSED) {
+            fprintf(stderr, "r300: ERROR: FS input generic %i unassigned, "
+                    "not enough hardware slots.\n", i);
         }
     }
 
@@ -520,7 +521,12 @@ static void r300_update_rs_block(struct r300_context *r300)
         if (fs_inputs->fog != ATTR_UNUSED) {
             fp_offset++;
 
-            DBG(r300, DBG_RS, "r300: FS input fog unassigned.\n");
+            if (tex_count < 8) {
+                DBG(r300, DBG_RS, "r300: FS input fog unassigned.\n");
+            } else {
+                fprintf(stderr, "r300: ERROR: FS input fog unassigned, "
+                        "not enough hardware slots.\n");
+            }
         }
     }
 
@@ -543,6 +549,11 @@ static void r300_update_rs_block(struct r300_context *r300)
         fp_offset++;
         tex_count++;
         tex_ptr += 4;
+    } else {
+        if (fs_inputs->wpos != ATTR_UNUSED && tex_count >= 8) {
+            fprintf(stderr, "r300: ERROR: FS input WPOS unassigned, "
+                    "not enough hardware slots.\n");
+        }
     }
 
     /* Invalidate the rest of the no-TCL (GA) stream locations. */
@@ -581,62 +592,93 @@ static void r300_update_rs_block(struct r300_context *r300)
 }
 
 static uint32_t r300_get_border_color(enum pipe_format format,
-                                      const float border[4])
+                                      const float border[4],
+                                      boolean is_r500)
 {
     const struct util_format_description *desc;
-    float border_swizzled[4] = {
-        border[2],
-        border[1],
-        border[0],
-        border[3]
-    };
-    uint32_t r;
+    float border_swizzled[4] = {0};
+    unsigned i;
+    union util_color uc = {0};
 
     desc = util_format_description(format);
 
-    /* We don't use util_pack_format because it does not handle the formats
-     * we want, e.g. R4G4B4A4 is non-existent in Gallium. */
+    /* Do depth formats first. */
+    if (util_format_is_depth_or_stencil(format)) {
+        switch (format) {
+        case PIPE_FORMAT_Z16_UNORM:
+            return util_pack_z(PIPE_FORMAT_Z16_UNORM, border[0]);
+        case PIPE_FORMAT_X8Z24_UNORM:
+        case PIPE_FORMAT_S8_USCALED_Z24_UNORM:
+            if (is_r500) {
+                return util_pack_z(PIPE_FORMAT_X8Z24_UNORM, border[0]);
+            } else {
+                return util_pack_z(PIPE_FORMAT_Z16_UNORM, border[0]) << 16;
+            }
+        default:
+            assert(0);
+            return 0;
+        }
+    }
+
+    /* Apply inverse swizzle of the format. */
+    for (i = 0; i < 4; i++) {
+        switch (desc->swizzle[i]) {
+        case UTIL_FORMAT_SWIZZLE_X:
+            border_swizzled[2] = border[i];
+            break;
+        case UTIL_FORMAT_SWIZZLE_Y:
+            border_swizzled[1] = border[i];
+            break;
+        case UTIL_FORMAT_SWIZZLE_Z:
+            border_swizzled[0] = border[i];
+            break;
+        case UTIL_FORMAT_SWIZZLE_W:
+            border_swizzled[3] = border[i];
+            break;
+        }
+    }
+
+    /* Compressed formats. */
+    if (util_format_is_compressed(format)) {
+        util_pack_color(border_swizzled, PIPE_FORMAT_R8G8B8A8_UNORM, &uc);
+        return uc.ui;
+    }
+
     switch (desc->channel[0].size) {
         case 4:
-            r = ((float_to_ubyte(border_swizzled[0]) & 0xf0) >> 4) |
-                ((float_to_ubyte(border_swizzled[1]) & 0xf0) << 0) |
-                ((float_to_ubyte(border_swizzled[2]) & 0xf0) << 4) |
-                ((float_to_ubyte(border_swizzled[3]) & 0xf0) << 8);
+            util_pack_color(border_swizzled, PIPE_FORMAT_B4G4R4A4_UNORM, &uc);
             break;
 
         case 5:
             if (desc->channel[1].size == 5) {
-                r = ((float_to_ubyte(border_swizzled[0]) & 0xf8) >> 3) |
-                    ((float_to_ubyte(border_swizzled[1]) & 0xf8) << 2) |
-                    ((float_to_ubyte(border_swizzled[2]) & 0xf8) << 7) |
-                    ((float_to_ubyte(border_swizzled[3]) & 0x80) << 8);
+                util_pack_color(border_swizzled, PIPE_FORMAT_B5G5R5A1_UNORM, &uc);
             } else if (desc->channel[1].size == 6) {
-                r = ((float_to_ubyte(border_swizzled[0]) & 0xf8) >> 3) |
-                    ((float_to_ubyte(border_swizzled[1]) & 0xfc) << 3) |
-                    ((float_to_ubyte(border_swizzled[2]) & 0xf8) << 8);
+                util_pack_color(border_swizzled, PIPE_FORMAT_B5G6R5_UNORM, &uc);
             } else {
                 assert(0);
-                r = 0;
             }
             break;
 
-        case 16:
-            r = ((float_to_ubyte(border_swizzled[2]) & 0xff) << 0) |
-                ((float_to_ubyte(border_swizzled[1]) & 0xff) << 8) |
-                ((float_to_ubyte(border_swizzled[0]) & 0xff) << 16) |
-                ((float_to_ubyte(border_swizzled[3]) & 0xff) << 24);
+        default:
+        case 8:
+            util_pack_color(border_swizzled, PIPE_FORMAT_B8G8R8A8_UNORM, &uc);
             break;
 
-        case 8:
-        default:
-            r = ((float_to_ubyte(border_swizzled[0]) & 0xff) << 0) |
-                ((float_to_ubyte(border_swizzled[1]) & 0xff) << 8) |
-                ((float_to_ubyte(border_swizzled[2]) & 0xff) << 16) |
-                ((float_to_ubyte(border_swizzled[3]) & 0xff) << 24);
+        case 10:
+            util_pack_color(border_swizzled, PIPE_FORMAT_B10G10R10A2_UNORM, &uc);
+            break;
+
+        case 16:
+            if (desc->nr_channels <= 2) {
+                border_swizzled[0] = border_swizzled[2];
+                util_pack_color(border_swizzled, PIPE_FORMAT_R16G16_UNORM, &uc);
+            } else {
+                util_pack_color(border_swizzled, PIPE_FORMAT_B8G8R8A8_UNORM, &uc);
+            }
             break;
     }
 
-    return r;
+    return uc.ui;
 }
 
 static void r300_merge_textures_and_samplers(struct r300_context* r300)
@@ -675,12 +717,13 @@ static void r300_merge_textures_and_samplers(struct r300_context* r300)
             /* Set the border color. */
             texstate->border_color =
                 r300_get_border_color(view->base.format,
-                                      sampler->state.border_color);
+                                      sampler->state.border_color,
+                                      r300->screen->caps.is_r500);
 
             /* determine min/max levels */
-            max_level = MIN3(sampler->max_lod + view->base.first_level,
-                             tex->desc.b.b.last_level, view->base.last_level);
-            min_level = MIN2(sampler->min_lod + view->base.first_level,
+            max_level = MIN3(sampler->max_lod + view->base.u.tex.first_level,
+                             tex->desc.b.b.last_level, view->base.u.tex.last_level);
+            min_level = MIN2(sampler->min_lod + view->base.u.tex.first_level,
                              max_level);
 
             if (tex->desc.is_npot && min_level > 0) {
@@ -721,13 +764,18 @@ static void r300_merge_textures_and_samplers(struct r300_context* r300)
                 if (sampler->state.compare_mode == PIPE_TEX_COMPARE_NONE) {
                     texstate->format.format1 |=
                         r300_get_swizzle_combined(depth_swizzle,
-                                                  view->swizzle);
+                                                  view->swizzle, FALSE);
                 } else {
                     texstate->format.format1 |=
-                        r300_get_swizzle_combined(depth_swizzle, 0);
+                        r300_get_swizzle_combined(depth_swizzle, 0, FALSE);
                 }
             }
 
+            if (r300->screen->caps.dxtc_swizzle &&
+                util_format_is_compressed(tex->desc.b.b.format)) {
+                texstate->filter1 |= R400_DXTC_SWIZZLE_ENABLE;
+            }
+
             /* to emulate 1D textures through 2D ones correctly */
             if (tex->desc.b.b.target == PIPE_TEXTURE_1D) {
                 texstate->filter0 &= ~R300_TX_WRAP_T_MASK;
@@ -839,9 +887,8 @@ static void r300_flush_depth_textures(struct r300_context *r300)
             for (level = 0; level <= tex->last_level; level++)
                 if (r300_texture(tex)->zmask_in_use[level]) {
                     /* We don't handle 3D textures and cubemaps yet. */
-                    r300_flush_depth_stencil(&r300->context, tex,
-                                             u_subresource(0, level), 0);
-                }
+                    r300_flush_depth_stencil(&r300->context, tex, level, 0);
+                 }
         }
 }
 
diff --git a/src/gallium/drivers/r300/r300_texture.c b/src/gallium/drivers/r300/r300_texture.c
index cee56bccdc..6d86bc282f 100644
--- a/src/gallium/drivers/r300/r300_texture.c
+++ b/src/gallium/drivers/r300/r300_texture.c
@@ -40,7 +40,8 @@
 #include "pipe/p_screen.h"
 
 unsigned r300_get_swizzle_combined(const unsigned char *swizzle_format,
-                                   const unsigned char *swizzle_view)
+                                   const unsigned char *swizzle_view,
+                                   boolean dxtc_swizzle)
 {
     unsigned i;
     unsigned char swizzle[4];
@@ -51,10 +52,10 @@ unsigned r300_get_swizzle_combined(const unsigned char *swizzle_format,
         R300_TX_FORMAT_B_SHIFT,
         R300_TX_FORMAT_A_SHIFT
     };
-    const uint32_t swizzle_bit[4] = {
-        R300_TX_FORMAT_X,
+    uint32_t swizzle_bit[4] = {
+        dxtc_swizzle ? R300_TX_FORMAT_Z : R300_TX_FORMAT_X,
         R300_TX_FORMAT_Y,
-        R300_TX_FORMAT_Z,
+        dxtc_swizzle ? R300_TX_FORMAT_X : R300_TX_FORMAT_Z,
         R300_TX_FORMAT_W
     };
 
@@ -107,7 +108,8 @@ unsigned r300_get_swizzle_combined(const unsigned char *swizzle_format,
  * makes available X, Y, Z, W, ZERO, and ONE for swizzling. */
 uint32_t r300_translate_texformat(enum pipe_format format,
                                   const unsigned char *swizzle_view,
-                                  boolean is_r500)
+                                  boolean is_r500,
+                                  boolean dxtc_swizzle)
 {
     uint32_t result = 0;
     const struct util_format_description *desc;
@@ -169,7 +171,8 @@ uint32_t r300_translate_texformat(enum pipe_format format,
             }
     }
 
-    result |= r300_get_swizzle_combined(desc->swizzle, swizzle_view);
+    result |= r300_get_swizzle_combined(desc->swizzle, swizzle_view,
+                    util_format_is_compressed(format) && dxtc_swizzle);
 
     /* S3TC formats. */
     if (desc->layout == UTIL_FORMAT_LAYOUT_S3TC) {
@@ -571,7 +574,7 @@ boolean r300_is_zs_format_supported(enum pipe_format format)
 
 boolean r300_is_sampler_format_supported(enum pipe_format format)
 {
-    return r300_translate_texformat(format, 0, TRUE) != ~0;
+    return r300_translate_texformat(format, 0, TRUE, FALSE) != ~0;
 }
 
 void r300_texture_setup_format_state(struct r300_screen *screen,
@@ -665,21 +668,21 @@ void r300_texture_reinterpret_format(struct pipe_screen *screen,
 }
 
 static unsigned r300_texture_is_referenced(struct pipe_context *context,
-					 struct pipe_resource *texture,
-					 unsigned face, unsigned level)
+                                           struct pipe_resource *texture,
+                                           unsigned level, int layer)
 {
     struct r300_context *r300 = r300_context(context);
     struct r300_texture *rtex = (struct r300_texture *)texture;
 
     if (r300->rws->cs_is_buffer_referenced(r300->cs,
-                                           rtex->buffer, R300_REF_CS))
+                                           rtex->cs_buffer, R300_REF_CS))
         return PIPE_REFERENCED_FOR_READ | PIPE_REFERENCED_FOR_WRITE;
 
     return PIPE_UNREFERENCED;
 }
 
 static void r300_texture_destroy(struct pipe_screen *screen,
-				 struct pipe_resource* texture)
+                                 struct pipe_resource* texture)
 {
     struct r300_texture* tex = (struct r300_texture*)texture;
     struct r300_winsys_screen *rws = (struct r300_winsys_screen *)texture->screen->winsys;
@@ -774,6 +777,8 @@ r300_texture_create_object(struct r300_screen *rscreen,
         }
     }
 
+    tex->cs_buffer = rws->buffer_get_cs_handle(rws, tex->buffer);
+
     rws->buffer_set_tiling(rws, tex->buffer,
             tex->desc.microtile, tex->desc.macrotile[0],
             tex->desc.stride_in_bytes[0]);
@@ -848,38 +853,40 @@ struct pipe_resource *r300_texture_from_handle(struct pipe_screen *screen,
 
 /* Not required to implement u_resource_vtbl, consider moving to another file:
  */
-struct pipe_surface* r300_get_tex_surface(struct pipe_screen* screen,
-                                          struct pipe_resource* texture,
-					  unsigned face,
-					  unsigned level,
-					  unsigned zslice,
-					  unsigned flags)
+struct pipe_surface* r300_create_surface(struct pipe_context * ctx,
+                                         struct pipe_resource* texture,
+                                         const struct pipe_surface *surf_tmpl)
 {
     struct r300_texture* tex = r300_texture(texture);
     struct r300_surface* surface = CALLOC_STRUCT(r300_surface);
+    unsigned level = surf_tmpl->u.tex.level;
+
+    assert(surf_tmpl->u.tex.first_layer == surf_tmpl->u.tex.last_layer);
 
     if (surface) {
         uint32_t offset, tile_height;
 
         pipe_reference_init(&surface->base.reference, 1);
         pipe_resource_reference(&surface->base.texture, texture);
-        surface->base.format = texture->format;
+        surface->base.context = ctx;
+        surface->base.format = surf_tmpl->format;
         surface->base.width = u_minify(texture->width0, level);
         surface->base.height = u_minify(texture->height0, level);
-        surface->base.usage = flags;
-        surface->base.zslice = zslice;
-        surface->base.face = face;
-        surface->base.level = level;
+        surface->base.usage = surf_tmpl->usage;
+        surface->base.u.tex.level = level;
+        surface->base.u.tex.first_layer = surf_tmpl->u.tex.first_layer;
+        surface->base.u.tex.last_layer = surf_tmpl->u.tex.last_layer;
 
         surface->buffer = tex->buffer;
+        surface->cs_buffer = tex->cs_buffer;
 
         /* Prefer VRAM if there are multiple domains to choose from. */
         surface->domain = tex->domain;
         if (surface->domain & R300_DOMAIN_VRAM)
             surface->domain &= ~R300_DOMAIN_GTT;
 
-        surface->offset = r300_texture_get_offset(&tex->desc,
-                                                  level, zslice, face);
+        surface->offset = r300_texture_get_offset(&tex->desc, level,
+                                                  surf_tmpl->u.tex.first_layer);
         surface->pitch = tex->fb_state.pitch[level];
         surface->format = tex->fb_state.format;
 
@@ -892,7 +899,7 @@ struct pipe_surface* r300_get_tex_surface(struct pipe_screen* screen,
                                                tex->desc.b.b.nr_samples,
                                                tex->desc.microtile,
                                                tex->desc.macrotile[level],
-                                               DIM_HEIGHT);
+                                               DIM_HEIGHT, 0);
 
         surface->cbzb_height = align((surface->base.height + 1) / 2,
                                      tile_height);
@@ -910,13 +917,13 @@ struct pipe_surface* r300_get_tex_surface(struct pipe_screen* screen,
         else
             surface->cbzb_format = R300_DEPTHFORMAT_16BIT_INT_Z;
 
-        SCREEN_DBG(r300_screen(screen), DBG_CBZB,
-                   "CBZB Allowed: %s, Dim: %ix%i, Misalignment: %i, Micro: %s, Macro: %s\n",
-                   surface->cbzb_allowed ? "YES" : " NO",
-                   surface->cbzb_width, surface->cbzb_height,
-                   offset & 2047,
-                   tex->desc.microtile ? "YES" : " NO",
-                   tex->desc.macrotile[level] ? "YES" : " NO");
+        DBG(r300_context(ctx), DBG_CBZB,
+            "CBZB Allowed: %s, Dim: %ix%i, Misalignment: %i, Micro: %s, Macro: %s\n",
+            surface->cbzb_allowed ? "YES" : " NO",
+            surface->cbzb_width, surface->cbzb_height,
+            offset & 2047,
+            tex->desc.microtile ? "YES" : " NO",
+            tex->desc.macrotile[level] ? "YES" : " NO");
     }
 
     return &surface->base;
@@ -924,7 +931,7 @@ struct pipe_surface* r300_get_tex_surface(struct pipe_screen* screen,
 
 /* Not required to implement u_resource_vtbl, consider moving to another file:
  */
-void r300_tex_surface_destroy(struct pipe_surface* s)
+void r300_surface_destroy(struct pipe_context *ctx, struct pipe_surface* s)
 {
     pipe_resource_reference(&s->texture, NULL);
     FREE(s);
diff --git a/src/gallium/drivers/r300/r300_texture.h b/src/gallium/drivers/r300/r300_texture.h
index c4588a0c90..0ab22f747e 100644
--- a/src/gallium/drivers/r300/r300_texture.h
+++ b/src/gallium/drivers/r300/r300_texture.h
@@ -27,6 +27,7 @@
 #include "pipe/p_format.h"
 
 struct pipe_screen;
+struct pipe_context;
 struct pipe_resource;
 struct winsys_handle;
 struct r300_texture_format_state;
@@ -35,11 +36,13 @@ struct r300_texture;
 struct r300_screen;
 
 unsigned r300_get_swizzle_combined(const unsigned char *swizzle_format,
-                                   const unsigned char *swizzle_view);
+                                   const unsigned char *swizzle_view,
+                                   boolean dxtc_swizzle);
 
 uint32_t r300_translate_texformat(enum pipe_format format,
                                   const unsigned char *swizzle_view,
-                                  boolean is_r500);
+                                  boolean is_r500,
+                                  boolean dxtc_swizzle);
 
 uint32_t r500_tx_format_msb_bit(enum pipe_format format);
 
@@ -68,13 +71,10 @@ r300_texture_create(struct pipe_screen* screen,
 		    const struct pipe_resource* templ);
 
 
-struct pipe_surface* r300_get_tex_surface(struct pipe_screen* screen,
-					  struct pipe_resource* texture,
-					  unsigned face,
-					  unsigned level,
-					  unsigned zslice,
-					  unsigned flags);
+struct pipe_surface* r300_create_surface(struct pipe_context *ctx,
+                                         struct pipe_resource* texture,
+                                         const struct pipe_surface *surf_tmpl);
 
-void r300_tex_surface_destroy(struct pipe_surface* s);
+void r300_surface_destroy(struct pipe_context *ctx, struct pipe_surface* s);
 
 #endif /* R300_TEXTURE_H */
diff --git a/src/gallium/drivers/r300/r300_texture_desc.c b/src/gallium/drivers/r300/r300_texture_desc.c
index 543d0fdc15..7b1739142d 100644
--- a/src/gallium/drivers/r300/r300_texture_desc.c
+++ b/src/gallium/drivers/r300/r300_texture_desc.c
@@ -34,7 +34,7 @@ unsigned r300_get_pixel_alignment(enum pipe_format format,
                                   unsigned num_samples,
                                   enum r300_buffer_tiling microtile,
                                   enum r300_buffer_tiling macrotile,
-                                  enum r300_dim dim)
+                                  enum r300_dim dim, boolean is_rs690)
 {
     static const unsigned table[2][5][3][2] =
     {
@@ -57,6 +57,7 @@ unsigned r300_get_pixel_alignment(enum pipe_format format,
             {{ 16, 8}, { 0,  0}, { 0,  0}}  /* 128 bits per pixel */
         }
     };
+
     static const unsigned aa_block[2] = {4, 8};
     unsigned tile = 0;
     unsigned pixsize = util_format_get_blocksize(format);
@@ -74,6 +75,14 @@ unsigned r300_get_pixel_alignment(enum pipe_format format,
     } else {
         /* Standard alignment. */
         tile = table[macrotile][util_logbase2(pixsize)][microtile][dim];
+        if (macrotile == 0 && is_rs690 && dim == DIM_WIDTH) {
+            int align;
+            int h_tile;
+            h_tile = table[macrotile][util_logbase2(pixsize)][microtile][DIM_HEIGHT];
+            align = 64 / (pixsize * h_tile);
+            if (tile < align)
+                tile = align;
+        }
     }
 
     assert(tile);
@@ -89,7 +98,7 @@ static boolean r300_texture_macro_switch(struct r300_texture_desc *desc,
     unsigned tile, texdim;
 
     tile = r300_get_pixel_alignment(desc->b.b.format, desc->b.b.nr_samples,
-                                    desc->microtile, R300_BUFFER_TILED, dim);
+                                    desc->microtile, R300_BUFFER_TILED, dim, 0);
     if (dim == DIM_WIDTH) {
         texdim = u_minify(desc->width0, level);
     } else {
@@ -113,6 +122,9 @@ static unsigned r300_texture_get_stride(struct r300_screen *screen,
                                         unsigned level)
 {
     unsigned tile_width, width, stride;
+    boolean is_rs690 = (screen->caps.family == CHIP_FAMILY_RS600 ||
+                        screen->caps.family == CHIP_FAMILY_RS690 ||
+                        screen->caps.family == CHIP_FAMILY_RS740);
 
     if (desc->stride_in_bytes_override)
         return desc->stride_in_bytes_override;
@@ -131,38 +143,14 @@ static unsigned r300_texture_get_stride(struct r300_screen *screen,
                                               desc->b.b.nr_samples,
                                               desc->microtile,
                                               desc->macrotile[level],
-                                              DIM_WIDTH);
+                                              DIM_WIDTH, is_rs690);
         width = align(width, tile_width);
 
         stride = util_format_get_stride(desc->b.b.format, width);
-
-        /* Some IGPs need a minimum stride of 64 bytes, hmm... */
-        if (!desc->macrotile[level] &&
-            (screen->caps.family == CHIP_FAMILY_RS600 ||
-             screen->caps.family == CHIP_FAMILY_RS690 ||
-             screen->caps.family == CHIP_FAMILY_RS740)) {
-            unsigned min_stride;
-
-            if (desc->microtile) {
-                unsigned tile_height =
-                        r300_get_pixel_alignment(desc->b.b.format,
-                                                 desc->b.b.nr_samples,
-                                                 desc->microtile,
-                                                 desc->macrotile[level],
-                                                 DIM_HEIGHT);
-
-                min_stride = 64 / tile_height;
-            } else {
-                min_stride = 64;
-            }
-
-            return stride < min_stride ? min_stride : stride;
-        }
-
         /* The alignment to 32 bytes is sort of implied by the layout... */
         return stride;
     } else {
-        return align(util_format_get_stride(desc->b.b.format, width), 32);
+        return align(util_format_get_stride(desc->b.b.format, width), is_rs690 ? 64 : 32);
     }
 }
 
@@ -179,7 +167,7 @@ static unsigned r300_texture_get_nblocksy(struct r300_texture_desc *desc,
                                                desc->b.b.nr_samples,
                                                desc->microtile,
                                                desc->macrotile[level],
-                                               DIM_HEIGHT);
+                                               DIM_HEIGHT, 0);
         height = align(height, tile_height);
 
         /* This is needed for the kernel checker, unfortunately. */
@@ -474,22 +462,17 @@ boolean r300_texture_desc_init(struct r300_screen *rscreen,
 }
 
 unsigned r300_texture_get_offset(struct r300_texture_desc *desc,
-                                 unsigned level, unsigned zslice,
-                                 unsigned face)
+                                 unsigned level, unsigned layer)
 {
     unsigned offset = desc->offset_in_bytes[level];
 
     switch (desc->b.b.target) {
         case PIPE_TEXTURE_3D:
-            assert(face == 0);
-            return offset + zslice * desc->layer_size_in_bytes[level];
-
         case PIPE_TEXTURE_CUBE:
-            assert(zslice == 0);
-            return offset + face * desc->layer_size_in_bytes[level];
+            return offset + layer * desc->layer_size_in_bytes[level];
 
         default:
-            assert(zslice == 0 && face == 0);
+            assert(layer == 0);
             return offset;
     }
 }
diff --git a/src/gallium/drivers/r300/r300_texture_desc.h b/src/gallium/drivers/r300/r300_texture_desc.h
index 3d7fe1fb47..121d215b4c 100644
--- a/src/gallium/drivers/r300/r300_texture_desc.h
+++ b/src/gallium/drivers/r300/r300_texture_desc.h
@@ -41,7 +41,7 @@ unsigned r300_get_pixel_alignment(enum pipe_format format,
                                   unsigned num_samples,
                                   enum r300_buffer_tiling microtile,
                                   enum r300_buffer_tiling macrotile,
-                                  enum r300_dim dim);
+                                  enum r300_dim dim, boolean is_rs690);
 
 boolean r300_texture_desc_init(struct r300_screen *rscreen,
                                struct r300_texture_desc *desc,
@@ -52,7 +52,6 @@ boolean r300_texture_desc_init(struct r300_screen *rscreen,
                                unsigned max_buffer_size);
 
 unsigned r300_texture_get_offset(struct r300_texture_desc *desc,
-                                 unsigned level, unsigned zslice,
-                                 unsigned face);
+                                 unsigned level, unsigned layer);
 
 #endif
diff --git a/src/gallium/drivers/r300/r300_tgsi_to_rc.c b/src/gallium/drivers/r300/r300_tgsi_to_rc.c
index a4911b9a2a..15a323989b 100644
--- a/src/gallium/drivers/r300/r300_tgsi_to_rc.c
+++ b/src/gallium/drivers/r300/r300_tgsi_to_rc.c
@@ -57,7 +57,7 @@ static unsigned translate_opcode(unsigned opcode)
      /* case TGSI_OPCODE_DP2A: return RC_OPCODE_DP2A; */
                                         /* gap */
         case TGSI_OPCODE_FRC: return RC_OPCODE_FRC;
-     /* case TGSI_OPCODE_CLAMP: return RC_OPCODE_CLAMP; */
+        case TGSI_OPCODE_CLAMP: return RC_OPCODE_CLAMP;
         case TGSI_OPCODE_FLR: return RC_OPCODE_FLR;
      /* case TGSI_OPCODE_ROUND: return RC_OPCODE_ROUND; */
         case TGSI_OPCODE_EX2: return RC_OPCODE_EX2;
@@ -363,10 +363,7 @@ void r300_tgsi_to_rc(struct tgsi_to_rc * ttr,
                 break;
             case TGSI_TOKEN_TYPE_INSTRUCTION:
                 inst = &parser.FullToken.FullInstruction;
-                /* This hack with the RET opcode woudn't work with
-                 * conditionals. */
-                if (inst->Instruction.Opcode == TGSI_OPCODE_END ||
-                    inst->Instruction.Opcode == TGSI_OPCODE_RET) {
+                if (inst->Instruction.Opcode == TGSI_OPCODE_END) {
                     break;
                 }
 
diff --git a/src/gallium/drivers/r300/r300_transfer.c b/src/gallium/drivers/r300/r300_transfer.c
index e9333b35ef..3b95af79bc 100644
--- a/src/gallium/drivers/r300/r300_transfer.c
+++ b/src/gallium/drivers/r300/r300_transfer.c
@@ -27,6 +27,7 @@
 
 #include "util/u_memory.h"
 #include "util/u_format.h"
+#include "util/u_box.h"
 
 struct r300_transfer {
     /* Parent class */
@@ -52,16 +53,10 @@ static void r300_copy_from_tiled_texture(struct pipe_context *ctx,
 {
     struct pipe_transfer *transfer = (struct pipe_transfer*)r300transfer;
     struct pipe_resource *tex = transfer->resource;
-    struct pipe_subresource subdst;
 
-    subdst.face = 0;
-    subdst.level = 0;
-
-    ctx->resource_copy_region(ctx, &r300transfer->linear_texture->desc.b.b, subdst,
-			      0, 0, 0,
-			      tex, transfer->sr,
-			      transfer->box.x, transfer->box.y, transfer->box.z,
-			      transfer->box.width, transfer->box.height);
+    ctx->resource_copy_region(ctx, &r300transfer->linear_texture->desc.b.b, 0,
+                              0, 0, 0,
+                              tex, transfer->level, &transfer->box);
 }
 
 /* Copy a detiled texture to a tiled one. */
@@ -70,26 +65,22 @@ static void r300_copy_into_tiled_texture(struct pipe_context *ctx,
 {
     struct pipe_transfer *transfer = (struct pipe_transfer*)r300transfer;
     struct pipe_resource *tex = transfer->resource;
-    struct pipe_subresource subsrc;
-
-    subsrc.face = 0;
-    subsrc.level = 0;
+    struct pipe_box src_box;
+    u_box_origin_2d(transfer->box.width, transfer->box.height, &src_box);
 
-    ctx->resource_copy_region(ctx, tex, transfer->sr,
-			      transfer->box.x, transfer->box.y, transfer->box.z,
-                              &r300transfer->linear_texture->desc.b.b, subsrc,
-			      0, 0, 0,
-			      transfer->box.width, transfer->box.height);
+    ctx->resource_copy_region(ctx, tex, transfer->level,
+                              transfer->box.x, transfer->box.y, transfer->box.z,
+                              &r300transfer->linear_texture->desc.b.b, 0, &src_box);
 
     ctx->flush(ctx, 0, NULL);
 }
 
 struct pipe_transfer*
 r300_texture_get_transfer(struct pipe_context *ctx,
-			  struct pipe_resource *texture,
-			  struct pipe_subresource sr,
-			  unsigned usage,
-			  const struct pipe_box *box)
+                          struct pipe_resource *texture,
+                          unsigned level,
+                          unsigned usage,
+                          const struct pipe_box *box)
 {
     struct r300_context *r300 = r300_context(ctx);
     struct r300_texture *tex = r300_texture(texture);
@@ -99,13 +90,13 @@ r300_texture_get_transfer(struct pipe_context *ctx,
 
     referenced_cs =
         r300->rws->cs_is_buffer_referenced(r300->cs,
-                                           tex->buffer, R300_REF_CS);
+                                           tex->cs_buffer, R300_REF_CS);
     if (referenced_cs) {
         referenced_hw = TRUE;
     } else {
         referenced_hw =
             r300->rws->cs_is_buffer_referenced(r300->cs,
-                                               tex->buffer, R300_REF_HW);
+                                               tex->cs_buffer, R300_REF_HW);
     }
 
     blittable = ctx->screen->is_format_supported(
@@ -116,25 +107,27 @@ r300_texture_get_transfer(struct pipe_context *ctx,
     if (trans) {
         /* Initialize the transfer object. */
         pipe_resource_reference(&trans->transfer.resource, texture);
-        trans->transfer.sr = sr;
+        trans->transfer.level = level;
         trans->transfer.usage = usage;
         trans->transfer.box = *box;
 
         /* If the texture is tiled, we must create a temporary detiled texture
          * for this transfer.
          * Also make write transfers pipelined. */
-        if (tex->desc.microtile || tex->desc.macrotile[sr.level] ||
+        if (tex->desc.microtile || tex->desc.macrotile[level] ||
             ((referenced_hw & !(usage & PIPE_TRANSFER_READ)) && blittable)) {
             base.target = PIPE_TEXTURE_2D;
             base.format = texture->format;
             base.width0 = box->width;
             base.height0 = box->height;
-            base.depth0 = 0;
+            /* XXX: was depth0 = 0 */
+            base.depth0 = 1;
+            base.array_size = 1;
             base.last_level = 0;
             base.nr_samples = 0;
             base.usage = PIPE_USAGE_DYNAMIC;
             base.bind = 0;
-	    base.flags = R300_RESOURCE_FLAG_TRANSFER;
+            base.flags = R300_RESOURCE_FLAG_TRANSFER;
 
             /* For texture reading, the temporary (detiled) texture is used as
              * a render target when blitting from a tiled texture. */
@@ -164,7 +157,7 @@ r300_texture_get_transfer(struct pipe_context *ctx,
                 if (!trans->linear_texture) {
                     /* For linear textures, it's safe to fallback to
                      * an unpipelined transfer. */
-                    if (!tex->desc.microtile && !tex->desc.macrotile[sr.level]) {
+                    if (!tex->desc.microtile && !tex->desc.macrotile[level]) {
                         goto unpipelined;
                     }
 
@@ -182,7 +175,7 @@ r300_texture_get_transfer(struct pipe_context *ctx,
             /* Set the stride.
 	     *
 	     * Even though we are using an internal texture for this,
-	     * the transfer sr, box and usage parameters still reflect
+	     * the transfer level, box and usage parameters still reflect
 	     * the arguments received to get_transfer.  We just do the
 	     * right thing internally.
 	     */
@@ -202,9 +195,8 @@ r300_texture_get_transfer(struct pipe_context *ctx,
 
     unpipelined:
         /* Unpipelined transfer. */
-        trans->transfer.stride = tex->desc.stride_in_bytes[sr.level];
-        trans->offset = r300_texture_get_offset(&tex->desc,
-                                                sr.level, box->z, sr.face);
+        trans->transfer.stride = tex->desc.stride_in_bytes[level];
+        trans->offset = r300_texture_get_offset(&tex->desc, level, box->z);
 
         if (referenced_cs)
             ctx->flush(ctx, PIPE_FLUSH_RENDER_CACHE, NULL);
diff --git a/src/gallium/drivers/r300/r300_transfer.h b/src/gallium/drivers/r300/r300_transfer.h
index 0d32a68d1f..7977ef516f 100644
--- a/src/gallium/drivers/r300/r300_transfer.h
+++ b/src/gallium/drivers/r300/r300_transfer.h
@@ -30,22 +30,22 @@ struct r300_context;
 
 struct pipe_transfer*
 r300_texture_get_transfer(struct pipe_context *ctx,
-			  struct pipe_resource *texture,
-			  struct pipe_subresource sr,
-			  unsigned usage,
-			  const struct pipe_box *box);
+                          struct pipe_resource *texture,
+                          unsigned level,
+                          unsigned usage,
+                          const struct pipe_box *box);
 
 void
 r300_texture_transfer_destroy(struct pipe_context *ctx,
-			      struct pipe_transfer *trans);
+                              struct pipe_transfer *trans);
 
 void*
 r300_texture_transfer_map(struct pipe_context *ctx,
-			  struct pipe_transfer *transfer);
+                          struct pipe_transfer *transfer);
 
 void
 r300_texture_transfer_unmap(struct pipe_context *ctx,
-			    struct pipe_transfer *transfer);
+                            struct pipe_transfer *transfer);
 
 
 #endif
diff --git a/src/gallium/drivers/r300/r300_vs.c b/src/gallium/drivers/r300/r300_vs.c
index 65696555ac..78021e2c5d 100644
--- a/src/gallium/drivers/r300/r300_vs.c
+++ b/src/gallium/drivers/r300/r300_vs.c
@@ -213,7 +213,6 @@ void r300_translate_vertex_shader(struct r300_context *r300,
     compiler.Base.max_temp_regs = 32;
     compiler.Base.max_constants = 256;
     compiler.Base.max_alu_insts = r300->screen->caps.is_r500 ? 1024 : 256;
-    compiler.Base.remove_unused_constants = TRUE;
 
     if (compiler.Base.Debug & RC_DBG_LOG) {
         DBG(r300, DBG_VP, "r300: Initial vertex program\n");
@@ -227,6 +226,10 @@ void r300_translate_vertex_shader(struct r300_context *r300,
 
     r300_tgsi_to_rc(&ttr, vs->state.tokens);
 
+    if (compiler.Base.Program.Constants.Count > 200) {
+        compiler.Base.remove_unused_constants = TRUE;
+    }
+
     compiler.RequiredOutputs = ~(~0 << (vs->info.num_outputs + 1));
     compiler.SetHwInputOutput = &set_vertex_inputs_outputs;
 
diff --git a/src/gallium/drivers/r300/r300_winsys.h b/src/gallium/drivers/r300/r300_winsys.h
index 4597332399..0dd330d101 100644
--- a/src/gallium/drivers/r300/r300_winsys.h
+++ b/src/gallium/drivers/r300/r300_winsys.h
@@ -33,15 +33,17 @@
 
 #include "r300_defines.h"
 
+#define R300_MAX_CMDBUF_DWORDS (16 * 1024)
+
 struct winsys_handle;
 struct r300_winsys_screen;
 
-struct r300_winsys_buffer;
+struct r300_winsys_buffer;      /* for map/unmap etc. */
+struct r300_winsys_cs_buffer;   /* for write_reloc etc. */
 
 struct r300_winsys_cs {
-    uint32_t *ptr;      /* Pointer to the beginning of the CS. */
     unsigned cdw;       /* Number of used dwords. */
-    unsigned ndw;       /* Size of the CS in dwords. */
+    uint32_t *buf;      /* The command buffer. */
 };
 
 enum r300_value_id {
@@ -102,6 +104,10 @@ struct r300_winsys_screen {
                                                 unsigned usage,
                                                 enum r300_buffer_domain domain);
 
+    struct r300_winsys_cs_buffer *(*buffer_get_cs_handle)(
+            struct r300_winsys_screen *ws,
+            struct r300_winsys_buffer *buf);
+
     /**
      * Reference a buffer object (assign with reference counting).
      *
@@ -242,7 +248,7 @@ struct r300_winsys_screen {
      *                  of the R300_DOMAIN_* flags.
      */
     void (*cs_add_buffer)(struct r300_winsys_cs *cs,
-                          struct r300_winsys_buffer *buf,
+                          struct r300_winsys_cs_buffer *buf,
                           enum r300_buffer_domain rd,
                           enum r300_buffer_domain wd);
 
@@ -263,7 +269,7 @@ struct r300_winsys_screen {
      * \param wd        A write domain containing a bitmask of the R300_DOMAIN_* flags.
      */
     void (*cs_write_reloc)(struct r300_winsys_cs *cs,
-                           struct r300_winsys_buffer *buf,
+                           struct r300_winsys_cs_buffer *buf,
                            enum r300_buffer_domain rd,
                            enum r300_buffer_domain wd);
 
@@ -303,7 +309,7 @@ struct r300_winsys_screen {
      * \param domain    A bitmask of the R300_REF_* enums.
      */
     boolean (*cs_is_buffer_referenced)(struct r300_winsys_cs *cs,
-                                       struct r300_winsys_buffer *buf,
+                                       struct r300_winsys_cs_buffer *buf,
                                        enum r300_reference_domain domain);
 };
 
diff --git a/src/gallium/drivers/r600/Makefile b/src/gallium/drivers/r600/Makefile
index a484f38e9f..b476b9af3b 100644
--- a/src/gallium/drivers/r600/Makefile
+++ b/src/gallium/drivers/r600/Makefile
@@ -21,6 +21,7 @@ C_SOURCES = \
 	evergreen_state.c \
 	eg_asm.c \
 	r600_translate.c \
-	r600_state_common.c
+	r600_state_common.c \
+	r600_upload.c
 
 include ../../Makefile.template
diff --git a/src/gallium/drivers/r600/SConscript b/src/gallium/drivers/r600/SConscript
index 3fc1fa94c2..6498014096 100644
--- a/src/gallium/drivers/r600/SConscript
+++ b/src/gallium/drivers/r600/SConscript
@@ -28,6 +28,7 @@ r600 = env.ConvenienceLibrary(
         'r600_state_common.c',
         'r600_texture.c',
         'r600_translate.c',
+        'r600_upload.c',
         'r700_asm.c',
         'evergreen_state.c',
         'eg_asm.c',
diff --git a/src/gallium/drivers/r600/eg_asm.c b/src/gallium/drivers/r600/eg_asm.c
index c30f09c394..b79875c7c7 100644
--- a/src/gallium/drivers/r600/eg_asm.c
+++ b/src/gallium/drivers/r600/eg_asm.c
@@ -27,6 +27,7 @@
 #include "r600_asm.h"
 #include "eg_sq.h"
 #include "r600_opcodes.h"
+#include "evergreend.h"
 
 int eg_bc_cf_build(struct r600_bc *bc, struct r600_bc_cf *cf)
 {
@@ -74,6 +75,8 @@ int eg_bc_cf_build(struct r600_bc *bc, struct r600_bc_cf *cf)
 	case EG_V_SQ_CF_WORD1_SQ_CF_INST_LOOP_END:
 	case EG_V_SQ_CF_WORD1_SQ_CF_INST_LOOP_CONTINUE:
 	case EG_V_SQ_CF_WORD1_SQ_CF_INST_LOOP_BREAK:
+	case EG_V_SQ_CF_WORD1_SQ_CF_INST_CALL_FS:
+	case EG_V_SQ_CF_WORD1_SQ_CF_INST_RETURN:
 		bc->bytecode[id++] = S_SQ_CF_WORD0_ADDR(cf->cf_addr >> 1);
 		bc->bytecode[id++] = S_SQ_CF_WORD1_CF_INST(cf->inst) |
 					S_SQ_CF_WORD1_BARRIER(1) |
@@ -87,3 +90,37 @@ int eg_bc_cf_build(struct r600_bc *bc, struct r600_bc_cf *cf)
 	}
 	return 0;
 }
+
+void eg_cf_vtx(struct r600_vertex_element *ve, u32 *bytecode, unsigned count)
+{
+	struct r600_pipe_state *rstate;
+	unsigned i = 0;
+
+	if (count > 8) {
+		bytecode[i++] = S_SQ_CF_WORD0_ADDR(8 >> 1);
+		bytecode[i++] = S_SQ_CF_WORD1_CF_INST(EG_V_SQ_CF_WORD1_SQ_CF_INST_VTX) |
+				S_SQ_CF_WORD1_BARRIER(1) |
+				S_SQ_CF_WORD1_COUNT(8 - 1);
+		bytecode[i++] = S_SQ_CF_WORD0_ADDR(40 >> 1);
+		bytecode[i++] = S_SQ_CF_WORD1_CF_INST(EG_V_SQ_CF_WORD1_SQ_CF_INST_VTX) |
+				S_SQ_CF_WORD1_BARRIER(1) |
+				S_SQ_CF_WORD1_COUNT(count - 8 - 1);
+	} else {
+		bytecode[i++] = S_SQ_CF_WORD0_ADDR(8 >> 1);
+		bytecode[i++] = S_SQ_CF_WORD1_CF_INST(EG_V_SQ_CF_WORD1_SQ_CF_INST_VTX) |
+				S_SQ_CF_WORD1_BARRIER(1) |
+				S_SQ_CF_WORD1_COUNT(count - 1);
+	}
+	bytecode[i++] = S_SQ_CF_WORD0_ADDR(0);
+	bytecode[i++] = S_SQ_CF_WORD1_CF_INST(EG_V_SQ_CF_WORD1_SQ_CF_INST_RETURN) |
+			S_SQ_CF_WORD1_BARRIER(1);
+
+	rstate = &ve->rstate;
+	rstate->id = R600_PIPE_STATE_FETCH_SHADER;
+	rstate->nregs = 0;
+	r600_pipe_state_add_reg(rstate, R_0288A8_SQ_PGM_RESOURCES_FS,
+				0x00000000, 0xFFFFFFFF, NULL);
+	r600_pipe_state_add_reg(rstate, R_0288A4_SQ_PGM_START_FS,
+				(r600_bo_offset(ve->fetch_shader)) >> 8,
+				0xFFFFFFFF, ve->fetch_shader);
+}
diff --git a/src/gallium/drivers/r600/eg_state_inlines.h b/src/gallium/drivers/r600/eg_state_inlines.h
index 5964197640..ecea1db4f1 100644
--- a/src/gallium/drivers/r600/eg_state_inlines.h
+++ b/src/gallium/drivers/r600/eg_state_inlines.h
@@ -290,6 +290,7 @@ static inline uint32_t r600_translate_colorswap(enum pipe_format format)
 	switch (format) {
 		/* 8-bit buffers. */
 	case PIPE_FORMAT_A8_UNORM:
+		return V_028C70_SWAP_ALT_REV;
 	case PIPE_FORMAT_I8_UNORM:
 	case PIPE_FORMAT_L8_UNORM:
 	case PIPE_FORMAT_R8_UNORM:
@@ -311,6 +312,8 @@ static inline uint32_t r600_translate_colorswap(enum pipe_format format)
 	case PIPE_FORMAT_Z16_UNORM:
 		return V_028C70_SWAP_STD;
 
+	case PIPE_FORMAT_L8A8_UNORM:
+		return V_028C70_SWAP_ALT;
 	case PIPE_FORMAT_R8G8_UNORM:
 		return V_028C70_SWAP_STD;
 
@@ -400,6 +403,7 @@ static INLINE uint32_t r600_translate_colorformat(enum pipe_format format)
 	case PIPE_FORMAT_Z16_UNORM:
 		return V_028C70_COLOR_16;
 
+	case PIPE_FORMAT_L8A8_UNORM:
 	case PIPE_FORMAT_R8G8_UNORM:
 		return V_028C70_COLOR_8_8;
 
@@ -447,8 +451,10 @@ static INLINE uint32_t r600_translate_colorformat(enum pipe_format format)
 		return V_028C70_COLOR_16_16;
 
 		/* 64-bit buffers. */
-	case PIPE_FORMAT_R16G16B16A16_SSCALED:
+	case PIPE_FORMAT_R16G16B16_USCALED:
+	case PIPE_FORMAT_R16G16B16A16_USCALED:
 	case PIPE_FORMAT_R16G16B16_SSCALED:
+	case PIPE_FORMAT_R16G16B16A16_SSCALED:
 	case PIPE_FORMAT_R16G16B16A16_UNORM:
 	case PIPE_FORMAT_R16G16B16A16_SNORM:
 		return V_028C70_COLOR_16_16_16_16;
@@ -460,6 +466,7 @@ static INLINE uint32_t r600_translate_colorformat(enum pipe_format format)
 	case PIPE_FORMAT_R32G32_FLOAT:
 		return V_028C70_COLOR_32_32_FLOAT;
 
+	case PIPE_FORMAT_R32G32_USCALED:
 	case PIPE_FORMAT_R32G32_SSCALED:
 		return V_028C70_COLOR_32_32;
 
@@ -634,38 +641,4 @@ out_unknown:
 	return ~0;
 }
 
-static INLINE uint32_t r600_translate_vertex_data_swizzle(enum pipe_format format)
-{
-	const struct util_format_description *desc = util_format_description(format);
-	unsigned i;
-	uint32_t word3;
-
-	assert(format);
-
-	if (desc->layout != UTIL_FORMAT_LAYOUT_PLAIN) {
-		fprintf(stderr, "r600: Bad format %s in %s:%d\n",
-				util_format_short_name(format), __FUNCTION__, __LINE__);
-		return 0;
-	}
-
-	word3 = 0;
-	for (i = 0; i < desc->nr_channels; i++) {
-		switch (i) {
-		case 0:
-			word3 |= S_03000C_DST_SEL_X(desc->swizzle[0]);
-			break;
-		case 1:
-			word3 |= S_03000C_DST_SEL_Y(desc->swizzle[1]);
-			break;
-		case 2:
-			word3 |= S_03000C_DST_SEL_Z(desc->swizzle[2]);
-			break;
-		case 3:
-			word3 |= S_03000C_DST_SEL_W(desc->swizzle[3]);
-			break;
-		}
-	}
-	return word3;
-}
-
 #endif
diff --git a/src/gallium/drivers/r600/evergreen_state.c b/src/gallium/drivers/r600/evergreen_state.c
index 4725b5e75e..07496ebf51 100644
--- a/src/gallium/drivers/r600/evergreen_state.c
+++ b/src/gallium/drivers/r600/evergreen_state.c
@@ -323,11 +323,11 @@ static void *evergreen_create_sampler_state(struct pipe_context *ctx,
 			S_03C000_BORDER_COLOR_TYPE(uc.ui ? V_03C000_SQ_TEX_BORDER_COLOR_REGISTER : 0), 0xFFFFFFFF, NULL);
 	/* FIXME LOD it depends on texture base level ... */
 	r600_pipe_state_add_reg(rstate, R_03C004_SQ_TEX_SAMPLER_WORD1_0,
-			S_03C004_MIN_LOD(S_FIXED(CLAMP(state->min_lod, 0, 15), 6)) |
-			S_03C004_MAX_LOD(S_FIXED(CLAMP(state->max_lod, 0, 15), 6)),
+			S_03C004_MIN_LOD(S_FIXED(CLAMP(state->min_lod, 0, 15), 8)) |
+			S_03C004_MAX_LOD(S_FIXED(CLAMP(state->max_lod, 0, 15), 8)),
 			0xFFFFFFFF, NULL);
 	r600_pipe_state_add_reg(rstate, R_03C008_SQ_TEX_SAMPLER_WORD2_0,
-				S_03C008_LOD_BIAS(S_FIXED(CLAMP(state->lod_bias, -16, 16), 6)) |
+				S_03C008_LOD_BIAS(S_FIXED(CLAMP(state->lod_bias, -16, 16), 8)) |
 				S_03C008_TYPE(1),
 				0xFFFFFFFF, NULL);
 
@@ -410,9 +410,9 @@ static struct pipe_sampler_view *evergreen_create_sampler_view(struct pipe_conte
 	r600_pipe_state_add_reg(rstate, R_030010_RESOURCE0_WORD4,
 				word4 | S_030010_NUM_FORMAT_ALL(V_030010_SQ_NUM_FORMAT_NORM) |
 				S_030010_SRF_MODE_ALL(V_030010_SFR_MODE_NO_ZERO) |
-				S_030010_BASE_LEVEL(state->first_level), 0xFFFFFFFF, NULL);
+				S_030010_BASE_LEVEL(state->u.tex.first_level), 0xFFFFFFFF, NULL);
 	r600_pipe_state_add_reg(rstate, R_030014_RESOURCE0_WORD5,
-				S_030014_LAST_LEVEL(state->last_level) |
+				S_030014_LAST_LEVEL(state->u.tex.last_level) |
 				S_030014_BASE_ARRAY(0) |
 				S_030014_LAST_ARRAY(0), 0xffffffff, NULL);
 	r600_pipe_state_add_reg(rstate, R_030018_RESOURCE0_WORD6, 0x0, 0xFFFFFFFF, NULL);
@@ -431,7 +431,7 @@ static void evergreen_set_vs_sampler_view(struct pipe_context *ctx, unsigned cou
 
 	for (int i = 0; i < count; i++) {
 		if (resource[i]) {
-			evergreen_context_pipe_state_set_vs_resource(&rctx->ctx, &resource[i]->state, i + PIPE_MAX_ATTRIBS);
+			evergreen_context_pipe_state_set_vs_resource(&rctx->ctx, &resource[i]->state, i);
 		}
 	}
 }
@@ -501,16 +501,16 @@ static void evergreen_set_clip_state(struct pipe_context *ctx,
 	rstate->id = R600_PIPE_STATE_CLIP;
 	for (int i = 0; i < state->nr; i++) {
 		r600_pipe_state_add_reg(rstate,
-					R_0285BC_PA_CL_UCP0_X + i * 4,
+					R_0285BC_PA_CL_UCP0_X + i * 16,
 					fui(state->ucp[i][0]), 0xFFFFFFFF, NULL);
 		r600_pipe_state_add_reg(rstate,
-					R_0285C0_PA_CL_UCP0_Y + i * 4,
+					R_0285C0_PA_CL_UCP0_Y + i * 16,
 					fui(state->ucp[i][1]) , 0xFFFFFFFF, NULL);
 		r600_pipe_state_add_reg(rstate,
-					R_0285C4_PA_CL_UCP0_Z + i * 4,
+					R_0285C4_PA_CL_UCP0_Z + i * 16,
 					fui(state->ucp[i][2]), 0xFFFFFFFF, NULL);
 		r600_pipe_state_add_reg(rstate,
-					R_0285C8_PA_CL_UCP0_W + i * 4,
+					R_0285C8_PA_CL_UCP0_W + i * 16,
 					fui(state->ucp[i][3]), 0xFFFFFFFF, NULL);
 	}
 	r600_pipe_state_add_reg(rstate, R_028810_PA_CL_CLIP_CNTL,
@@ -633,10 +633,11 @@ static void evergreen_cb(struct r600_pipe_context *rctx, struct r600_pipe_state
 	struct r600_resource_texture *rtex;
 	struct r600_resource *rbuffer;
 	struct r600_surface *surf;
-	unsigned level = state->cbufs[cb]->level;
+	unsigned level = state->cbufs[cb]->u.tex.level;
 	unsigned pitch, slice;
 	unsigned color_info;
 	unsigned format, swap, ntype;
+	unsigned offset;
 	const struct util_format_description *desc;
 	struct r600_bo *bo[3];
 
@@ -647,6 +648,9 @@ static void evergreen_cb(struct r600_pipe_context *rctx, struct r600_pipe_state
 	bo[1] = rbuffer->bo;
 	bo[2] = rbuffer->bo;
 
+	/* XXX quite sure for dx10+ hw don't need any offset hacks */
+	offset = r600_texture_get_offset((struct r600_resource_texture *)state->cbufs[cb]->texture,
+					 level, state->cbufs[cb]->u.tex.first_layer);
 	pitch = rtex->pitch_in_pixels[level] / 8 - 1;
 	slice = rtex->pitch_in_pixels[level] * surf->aligned_height / 64 - 1;
 	ntype = 0;
@@ -660,13 +664,13 @@ static void evergreen_cb(struct r600_pipe_context *rctx, struct r600_pipe_state
 		S_028C70_COMP_SWAP(swap) |
 		S_028C70_BLEND_CLAMP(1) |
 		S_028C70_NUMBER_TYPE(ntype);
-	if (desc->colorspace != UTIL_FORMAT_COLORSPACE_ZS) 
+	if (desc->colorspace != UTIL_FORMAT_COLORSPACE_ZS)
 		color_info |= S_028C70_SOURCE_FORMAT(1);
 
 	/* FIXME handle enabling of CB beyond BASE8 which has different offset */
 	r600_pipe_state_add_reg(rstate,
 				R_028C60_CB_COLOR0_BASE + cb * 0x3C,
-				(state->cbufs[cb]->offset +  r600_bo_offset(bo[0])) >> 8, 0xFFFFFFFF, bo[0]);
+				(offset +  r600_bo_offset(bo[0])) >> 8, 0xFFFFFFFF, bo[0]);
 	r600_pipe_state_add_reg(rstate,
 				R_028C78_CB_COLOR0_DIM + cb * 0x3C,
 				0x0, 0xFFFFFFFF, NULL);
@@ -698,11 +702,12 @@ static void evergreen_db(struct r600_pipe_context *rctx, struct r600_pipe_state
 	struct r600_surface *surf;
 	unsigned level;
 	unsigned pitch, slice, format, stencil_format;
+	unsigned offset;
 
 	if (state->zsbuf == NULL)
 		return;
 
-	level = state->zsbuf->level;
+	level = state->zsbuf->u.tex.level;
 
 	surf = (struct r600_surface *)state->zsbuf;
 	rtex = (struct r600_resource_texture*)state->zsbuf->texture;
@@ -712,24 +717,27 @@ static void evergreen_db(struct r600_pipe_context *rctx, struct r600_pipe_state
 	rtex->depth = 1;
 	rbuffer = &rtex->resource;
 
+	/* XXX quite sure for dx10+ hw don't need any offset hacks */
+	offset = r600_texture_get_offset((struct r600_resource_texture *)state->zsbuf->texture,
+					 level, state->zsbuf->u.tex.first_layer);
 	pitch = rtex->pitch_in_pixels[level] / 8 - 1;
 	slice = rtex->pitch_in_pixels[level] * surf->aligned_height / 64 - 1;
 	format = r600_translate_dbformat(state->zsbuf->texture->format);
 	stencil_format = r600_translate_stencilformat(state->zsbuf->texture->format);
 
 	r600_pipe_state_add_reg(rstate, R_028048_DB_Z_READ_BASE,
-				(state->zsbuf->offset + r600_bo_offset(rbuffer->bo)) >> 8, 0xFFFFFFFF, rbuffer->bo);
+				(offset + r600_bo_offset(rbuffer->bo)) >> 8, 0xFFFFFFFF, rbuffer->bo);
 	r600_pipe_state_add_reg(rstate, R_028050_DB_Z_WRITE_BASE,
-				(state->zsbuf->offset  + r600_bo_offset(rbuffer->bo)) >> 8, 0xFFFFFFFF, rbuffer->bo);
+				(offset  + r600_bo_offset(rbuffer->bo)) >> 8, 0xFFFFFFFF, rbuffer->bo);
 
 	if (stencil_format) {
 		uint32_t stencil_offset;
 
 		stencil_offset = ((surf->aligned_height * rtex->pitch_in_bytes[level]) + 255) & ~255;
 		r600_pipe_state_add_reg(rstate, R_02804C_DB_STENCIL_READ_BASE,
-					(state->zsbuf->offset + stencil_offset + r600_bo_offset(rbuffer->bo)) >> 8, 0xFFFFFFFF, rbuffer->bo);
+					(offset + stencil_offset + r600_bo_offset(rbuffer->bo)) >> 8, 0xFFFFFFFF, rbuffer->bo);
 		r600_pipe_state_add_reg(rstate, R_028054_DB_STENCIL_WRITE_BASE,
-					(state->zsbuf->offset + stencil_offset + r600_bo_offset(rbuffer->bo)) >> 8, 0xFFFFFFFF, rbuffer->bo);
+					(offset + stencil_offset + r600_bo_offset(rbuffer->bo)) >> 8, 0xFFFFFFFF, rbuffer->bo);
 	}
 
 	r600_pipe_state_add_reg(rstate, R_028008_DB_DEPTH_VIEW, 0x00000000, 0xFFFFFFFF, NULL);
@@ -825,6 +833,10 @@ static void evergreen_set_framebuffer_state(struct pipe_context *ctx,
 	free(rctx->states[R600_PIPE_STATE_FRAMEBUFFER]);
 	rctx->states[R600_PIPE_STATE_FRAMEBUFFER] = rstate;
 	r600_context_pipe_state_set(&rctx->ctx, rstate);
+
+	if (state->zsbuf) {
+		evergreen_polygon_offset_update(rctx);
+	}
 }
 
 static void evergreen_set_constant_buffer(struct pipe_context *ctx, uint shader, uint index,
@@ -1036,11 +1048,33 @@ void evergreen_init_config(struct r600_pipe_context *rctx)
 		num_hs_stack_entries = 85;
 		num_ls_stack_entries = 85;
 		break;
+	case CHIP_PALM:
+		num_ps_gprs = 93;
+		num_vs_gprs = 46;
+		num_temp_gprs = 4;
+		num_gs_gprs = 31;
+		num_es_gprs = 31;
+		num_hs_gprs = 23;
+		num_ls_gprs = 23;
+		num_ps_threads = 96;
+		num_vs_threads = 16;
+		num_gs_threads = 16;
+		num_es_threads = 16;
+		num_hs_threads = 16;
+		num_ls_threads = 16;
+		num_ps_stack_entries = 42;
+		num_vs_stack_entries = 42;
+		num_gs_stack_entries = 42;
+		num_es_stack_entries = 42;
+		num_hs_stack_entries = 42;
+		num_ls_stack_entries = 42;
+		break;
 	}
 
 	tmp = 0x00000000;
 	switch (family) {
 	case CHIP_CEDAR:
+	case CHIP_PALM:
 		break;
 	default:
 		tmp |= S_008C00_VC_ENABLE(1);
@@ -1172,29 +1206,101 @@ void evergreen_init_config(struct r600_pipe_context *rctx)
 	r600_pipe_state_add_reg(rstate, R_0283F8_SQ_VTX_SEMANTIC_30, 0x0, 0xFFFFFFFF, NULL);
 	r600_pipe_state_add_reg(rstate, R_0283FC_SQ_VTX_SEMANTIC_31, 0x0, 0xFFFFFFFF, NULL);
 
-r600_pipe_state_add_reg(rstate, R_028810_PA_CL_CLIP_CNTL,
-			0x0, 0xFFFFFFFF, NULL);
+	r600_pipe_state_add_reg(rstate, R_028810_PA_CL_CLIP_CNTL, 0x0, 0xFFFFFFFF, NULL);
 
 	r600_context_pipe_state_set(&rctx->ctx, rstate);
 }
 
-int r600_conv_pipe_prim(unsigned pprim, unsigned *prim);
-void evergreen_draw(struct pipe_context *ctx, const struct pipe_draw_info *info)
+void evergreen_polygon_offset_update(struct r600_pipe_context *rctx)
+{
+	struct r600_pipe_state state;
+
+	state.id = R600_PIPE_STATE_POLYGON_OFFSET;
+	state.nregs = 0;
+	if (rctx->rasterizer && rctx->framebuffer.zsbuf) {
+		float offset_units = rctx->rasterizer->offset_units;
+		unsigned offset_db_fmt_cntl = 0, depth;
+
+		switch (rctx->framebuffer.zsbuf->texture->format) {
+		case PIPE_FORMAT_Z24X8_UNORM:
+		case PIPE_FORMAT_Z24_UNORM_S8_USCALED:
+			depth = -24;
+			offset_units *= 2.0f;
+			break;
+		case PIPE_FORMAT_Z32_FLOAT:
+			depth = -23;
+			offset_units *= 1.0f;
+			offset_db_fmt_cntl |= S_028B78_POLY_OFFSET_DB_IS_FLOAT_FMT(1);
+			break;
+		case PIPE_FORMAT_Z16_UNORM:
+			depth = -16;
+			offset_units *= 4.0f;
+			break;
+		default:
+			return;
+		}
+		/* FIXME some of those reg can be computed with cso */
+		offset_db_fmt_cntl |= S_028B78_POLY_OFFSET_NEG_NUM_DB_BITS(depth);
+		r600_pipe_state_add_reg(&state,
+				R_028B80_PA_SU_POLY_OFFSET_FRONT_SCALE,
+				fui(rctx->rasterizer->offset_scale), 0xFFFFFFFF, NULL);
+		r600_pipe_state_add_reg(&state,
+				R_028B84_PA_SU_POLY_OFFSET_FRONT_OFFSET,
+				fui(offset_units), 0xFFFFFFFF, NULL);
+		r600_pipe_state_add_reg(&state,
+				R_028B88_PA_SU_POLY_OFFSET_BACK_SCALE,
+				fui(rctx->rasterizer->offset_scale), 0xFFFFFFFF, NULL);
+		r600_pipe_state_add_reg(&state,
+				R_028B8C_PA_SU_POLY_OFFSET_BACK_OFFSET,
+				fui(offset_units), 0xFFFFFFFF, NULL);
+		r600_pipe_state_add_reg(&state,
+				R_028B78_PA_SU_POLY_OFFSET_DB_FMT_CNTL,
+				offset_db_fmt_cntl, 0xFFFFFFFF, NULL);
+		r600_context_pipe_state_set(&rctx->ctx, &state);
+	}
+}
+
+static void evergreen_spi_update(struct r600_pipe_context *rctx)
+{
+	struct r600_pipe_shader *shader = rctx->ps_shader;
+	struct r600_pipe_state rstate;
+	struct r600_shader *rshader = &shader->shader;
+	unsigned i, tmp;
+
+	rstate.nregs = 0;
+	for (i = 0; i < rshader->ninput; i++) {
+		tmp = S_028644_SEMANTIC(r600_find_vs_semantic_index(&rctx->vs_shader->shader, rshader, i));
+		if (rshader->input[i].name == TGSI_SEMANTIC_COLOR ||
+				rshader->input[i].name == TGSI_SEMANTIC_BCOLOR ||
+				rshader->input[i].name == TGSI_SEMANTIC_POSITION) {
+			tmp |= S_028644_FLAT_SHADE(rctx->flatshade);
+		}
+		if (rshader->input[i].name == TGSI_SEMANTIC_GENERIC &&
+			rctx->sprite_coord_enable & (1 << rshader->input[i].sid)) {
+			tmp |= S_028644_PT_SPRITE_TEX(1);
+		}
+		r600_pipe_state_add_reg(&rstate, R_028644_SPI_PS_INPUT_CNTL_0 + i * 4, tmp, 0xFFFFFFFF, NULL);
+	}
+	r600_context_pipe_state_set(&rctx->ctx, &rstate);
+}
+
+void evergreen_vertex_buffer_update(struct r600_pipe_context *rctx)
 {
-	struct r600_pipe_context *rctx = (struct r600_pipe_context *)ctx;
 	struct r600_pipe_state *rstate;
 	struct r600_resource *rbuffer;
-	unsigned i, j, offset, prim;
-	u32 vgt_dma_index_type, vgt_draw_initiator, mask;
 	struct pipe_vertex_buffer *vertex_buffer;
-	struct r600_draw rdraw;
-	struct r600_pipe_state vgt;
-	struct r600_drawl draw;
-	boolean translate = FALSE;
+	unsigned i, offset;
+
+	/* we don't update until we know vertex elements */
+	if (rctx->vertex_elements == NULL || !rctx->nvertex_buffer)
+		return;
 
 	if (rctx->vertex_elements->incompatible_layout) {
+		/* translate rebind new vertex elements so
+		 * return once translated
+		 */
 		r600_begin_vertex_translate(rctx);
-		translate = TRUE;
+		return;
 	}
 
 	if (rctx->any_user_vbs) {
@@ -1202,6 +1308,72 @@ void evergreen_draw(struct pipe_context *ctx, const struct pipe_draw_info *info)
 		rctx->any_user_vbs = FALSE;
 	}
 
+	if (rctx->vertex_elements->vbuffer_need_offset) {
+		/* one resource per vertex elements */
+		rctx->nvs_resource = rctx->vertex_elements->count;
+	} else {
+		/* bind vertex buffer once */
+		rctx->nvs_resource = rctx->nvertex_buffer;
+	}
+
+	for (i = 0 ; i < rctx->nvs_resource; i++) {
+		rstate = &rctx->vs_resource[i];
+		rstate->id = R600_PIPE_STATE_RESOURCE;
+		rstate->nregs = 0;
+
+		if (rctx->vertex_elements->vbuffer_need_offset) {
+			/* one resource per vertex elements */
+			unsigned vbuffer_index;
+			vbuffer_index = rctx->vertex_elements->elements[i].vertex_buffer_index;
+			vertex_buffer = &rctx->vertex_buffer[vbuffer_index];
+			rbuffer = (struct r600_resource*)vertex_buffer->buffer;
+			offset = rctx->vertex_elements->vbuffer_offset[i] +
+				vertex_buffer->buffer_offset +
+				r600_bo_offset(rbuffer->bo);
+		} else {
+			/* bind vertex buffer once */
+			vertex_buffer = &rctx->vertex_buffer[i];
+			rbuffer = (struct r600_resource*)vertex_buffer->buffer;
+			offset = vertex_buffer->buffer_offset +
+				r600_bo_offset(rbuffer->bo);
+		}
+
+		r600_pipe_state_add_reg(rstate, R_030000_RESOURCE0_WORD0,
+					offset, 0xFFFFFFFF, rbuffer->bo);
+		r600_pipe_state_add_reg(rstate, R_030004_RESOURCE0_WORD1,
+					rbuffer->bo_size - offset - 1, 0xFFFFFFFF, NULL);
+		r600_pipe_state_add_reg(rstate, R_030008_RESOURCE0_WORD2,
+					S_030008_STRIDE(vertex_buffer->stride),
+					0xFFFFFFFF, NULL);
+		r600_pipe_state_add_reg(rstate, R_03000C_RESOURCE0_WORD3,
+					S_03000C_DST_SEL_X(V_03000C_SQ_SEL_X) |
+					S_03000C_DST_SEL_Y(V_03000C_SQ_SEL_Y) |
+					S_03000C_DST_SEL_Z(V_03000C_SQ_SEL_Z) |
+					S_03000C_DST_SEL_W(V_03000C_SQ_SEL_W),
+					0xFFFFFFFF, NULL);
+		r600_pipe_state_add_reg(rstate, R_030010_RESOURCE0_WORD4,
+					0x00000000, 0xFFFFFFFF, NULL);
+		r600_pipe_state_add_reg(rstate, R_030014_RESOURCE0_WORD5,
+					0x00000000, 0xFFFFFFFF, NULL);
+		r600_pipe_state_add_reg(rstate, R_030018_RESOURCE0_WORD6,
+					0x00000000, 0xFFFFFFFF, NULL);
+		r600_pipe_state_add_reg(rstate, R_03001C_RESOURCE0_WORD7,
+					0xC0000000, 0xFFFFFFFF, NULL);
+		evergreen_fs_resource_set(&rctx->ctx, rstate, i);
+	}
+}
+
+int r600_conv_pipe_prim(unsigned pprim, unsigned *prim);
+void evergreen_draw(struct pipe_context *ctx, const struct pipe_draw_info *info)
+{
+	struct r600_pipe_context *rctx = (struct r600_pipe_context *)ctx;
+	struct r600_resource *rbuffer;
+	u32 vgt_dma_index_type, vgt_draw_initiator, mask;
+	struct r600_draw rdraw;
+	struct r600_pipe_state vgt;
+	struct r600_drawl draw;
+	unsigned prim;
+
 	memset(&draw, 0, sizeof(struct r600_drawl));
 	draw.ctx = ctx;
 	draw.mode = info->mode;
@@ -1250,45 +1422,23 @@ void evergreen_draw(struct pipe_context *ctx, const struct pipe_draw_info *info)
 	}
 	if (r600_conv_pipe_prim(draw.mode, &prim))
 		return;
-
-	/* rebuild vertex shader if input format changed */
-	if (r600_pipe_shader_update(&rctx->context, rctx->vs_shader))
+	if (unlikely(rctx->ps_shader == NULL)) {
+		R600_ERR("missing vertex shader\n");
 		return;
-	if (r600_pipe_shader_update(&rctx->context, rctx->ps_shader))
+	}
+	if (unlikely(rctx->vs_shader == NULL)) {
+		R600_ERR("missing vertex shader\n");
+		return;
+	}
+	/* there should be enough input */
+	if (rctx->vertex_elements->count < rctx->vs_shader->shader.bc.nresource) {
+		R600_ERR("%d resources provided, expecting %d\n",
+			rctx->vertex_elements->count, rctx->vs_shader->shader.bc.nresource);
 		return;
-
-	for (i = 0 ; i < rctx->vertex_elements->count; i++) {
-		uint32_t word3, word2;
-		uint32_t format;
-		rstate = &rctx->vs_resource[i];
-
-		rstate->id = R600_PIPE_STATE_RESOURCE;
-		rstate->nregs = 0;
-
-		j = rctx->vertex_elements->elements[i].vertex_buffer_index;
-		vertex_buffer = &rctx->vertex_buffer[j];
-		rbuffer = (struct r600_resource*)vertex_buffer->buffer;
-		offset = rctx->vertex_elements->elements[i].src_offset +
-			vertex_buffer->buffer_offset +
-			r600_bo_offset(rbuffer->bo);
-
-		format = r600_translate_vertex_data_type(rctx->vertex_elements->hw_format[i]);
-
-		word2 = format | S_030008_STRIDE(vertex_buffer->stride);
-
-		word3 = r600_translate_vertex_data_swizzle(rctx->vertex_elements->hw_format[i]);
-
-		r600_pipe_state_add_reg(rstate, R_030000_RESOURCE0_WORD0, offset, 0xFFFFFFFF, rbuffer->bo);
-		r600_pipe_state_add_reg(rstate, R_030004_RESOURCE0_WORD1, rbuffer->size - offset - 1, 0xFFFFFFFF, NULL);
-		r600_pipe_state_add_reg(rstate, R_030008_RESOURCE0_WORD2, word2, 0xFFFFFFFF, NULL);
-		r600_pipe_state_add_reg(rstate, R_03000C_RESOURCE0_WORD3, word3, 0xFFFFFFFF, NULL);
-		r600_pipe_state_add_reg(rstate, R_030010_RESOURCE0_WORD4, 0x00000000, 0xFFFFFFFF, NULL);
-		r600_pipe_state_add_reg(rstate, R_030014_RESOURCE0_WORD5, 0x00000000, 0xFFFFFFFF, NULL);
-		r600_pipe_state_add_reg(rstate, R_030018_RESOURCE0_WORD6, 0x00000000, 0xFFFFFFFF, NULL);
-		r600_pipe_state_add_reg(rstate, R_03001C_RESOURCE0_WORD7, 0xC0000000, 0xFFFFFFFF, NULL);
-		evergreen_vs_resource_set(&rctx->ctx, rstate, i);
 	}
 
+	evergreen_spi_update(rctx);
+
 	mask = 0;
 	for (int i = 0; i < rctx->framebuffer.nr_cbufs; i++) {
 		mask |= (0xF << (i * 4));
@@ -1303,46 +1453,6 @@ void evergreen_draw(struct pipe_context *ctx, const struct pipe_draw_info *info)
 	r600_pipe_state_add_reg(&vgt, R_028404_VGT_MIN_VTX_INDX, draw.min_index, 0xFFFFFFFF, NULL);
 	r600_pipe_state_add_reg(&vgt, R_03CFF0_SQ_VTX_BASE_VTX_LOC, 0, 0xFFFFFFFF, NULL);
 	r600_pipe_state_add_reg(&vgt, R_03CFF4_SQ_VTX_START_INST_LOC, 0, 0xFFFFFFFF, NULL);
-
-	if (rctx->rasterizer && rctx->framebuffer.zsbuf) {
-		float offset_units = rctx->rasterizer->offset_units;
-		unsigned offset_db_fmt_cntl = 0, depth;
-
-		switch (rctx->framebuffer.zsbuf->texture->format) {
-		case PIPE_FORMAT_Z24X8_UNORM:
-		case PIPE_FORMAT_Z24_UNORM_S8_USCALED:
-			depth = -24;
-			offset_units *= 2.0f;
-			break;
-		case PIPE_FORMAT_Z32_FLOAT:
-			depth = -23;
-			offset_units *= 1.0f;
-			offset_db_fmt_cntl |= S_028B78_POLY_OFFSET_DB_IS_FLOAT_FMT(1);
-			break;
-		case PIPE_FORMAT_Z16_UNORM:
-			depth = -16;
-			offset_units *= 4.0f;
-			break;
-		default:
-			return;
-		}
-		offset_db_fmt_cntl |= S_028B78_POLY_OFFSET_NEG_NUM_DB_BITS(depth);
-		r600_pipe_state_add_reg(&vgt,
-				R_028B80_PA_SU_POLY_OFFSET_FRONT_SCALE,
-				fui(rctx->rasterizer->offset_scale), 0xFFFFFFFF, NULL);
-		r600_pipe_state_add_reg(&vgt,
-				R_028B84_PA_SU_POLY_OFFSET_FRONT_OFFSET,
-				fui(offset_units), 0xFFFFFFFF, NULL);
-		r600_pipe_state_add_reg(&vgt,
-				R_028B88_PA_SU_POLY_OFFSET_BACK_SCALE,
-				fui(rctx->rasterizer->offset_scale), 0xFFFFFFFF, NULL);
-		r600_pipe_state_add_reg(&vgt,
-				R_028B8C_PA_SU_POLY_OFFSET_BACK_OFFSET,
-				fui(offset_units), 0xFFFFFFFF, NULL);
-		r600_pipe_state_add_reg(&vgt,
-				R_028B78_PA_SU_POLY_OFFSET_DB_FMT_CNTL,
-				offset_db_fmt_cntl, 0xFFFFFFFF, NULL);
-	}
 	r600_context_pipe_state_set(&rctx->ctx, &vgt);
 
 	rdraw.vgt_num_indices = draw.count;
@@ -1357,28 +1467,22 @@ void evergreen_draw(struct pipe_context *ctx, const struct pipe_draw_info *info)
 	}
 	evergreen_context_draw(&rctx->ctx, &rdraw);
 
-	if (translate)
-		r600_end_vertex_translate(rctx);
-
 	pipe_resource_reference(&draw.index_buffer, NULL);
 }
 
 void evergreen_pipe_shader_ps(struct pipe_context *ctx, struct r600_pipe_shader *shader)
 {
-	struct r600_pipe_context *rctx = (struct r600_pipe_context *)ctx;
 	struct r600_pipe_state *rstate = &shader->rstate;
 	struct r600_shader *rshader = &shader->shader;
-	unsigned i, tmp, exports_ps, num_cout, spi_ps_in_control_0, spi_input_z, spi_ps_in_control_1;
+	unsigned i, exports_ps, num_cout, spi_ps_in_control_0, spi_input_z, spi_ps_in_control_1;
 	int pos_index = -1, face_index = -1;
 	int ninterp = 0;
 	boolean have_linear = FALSE, have_centroid = FALSE, have_perspective = FALSE;
 	unsigned spi_baryc_cntl;
 
-	/* clear previous register */
 	rstate->nregs = 0;
 
 	for (i = 0; i < rshader->ninput; i++) {
-		tmp = S_028644_SEMANTIC(r600_find_vs_semantic_index(&rctx->vs_shader->shader, rshader, i));
 		/* evergreen NUM_INTERP only contains values interpolated into the LDS,
 		   POSITION goes via GPRs from the SC so isn't counted */
 		if (rshader->input[i].name == TGSI_SEMANTIC_POSITION)
@@ -1396,16 +1500,6 @@ void evergreen_pipe_shader_ps(struct pipe_context *ctx, struct r600_pipe_shader
 			if (rshader->input[i].centroid)
 				have_centroid = TRUE;
 		}
-		if (rshader->input[i].name == TGSI_SEMANTIC_COLOR ||
-		    rshader->input[i].name == TGSI_SEMANTIC_BCOLOR ||
-		    rshader->input[i].name == TGSI_SEMANTIC_POSITION) {
-			tmp |= S_028644_FLAT_SHADE(rshader->flat_shade);
-		}
-		if (rshader->input[i].name == TGSI_SEMANTIC_GENERIC &&
-			rctx->sprite_coord_enable & (1 << rshader->input[i].sid)) {
-			tmp |= S_028644_PT_SPRITE_TEX(1);
-		}
-		r600_pipe_state_add_reg(rstate, R_028644_SPI_PS_INPUT_CNTL_0 + i * 4, tmp, 0xFFFFFFFF, NULL);
 	}
 	for (i = 0; i < rshader->noutput; i++) {
 		if (rshader->output[i].name == TGSI_SEMANTIC_POSITION)
@@ -1464,8 +1558,8 @@ void evergreen_pipe_shader_ps(struct pipe_context *ctx, struct r600_pipe_shader
 				  S_0286E0_PERSP_CENTROID_ENA(have_centroid);
 	if (have_linear)
 		spi_baryc_cntl |= S_0286E0_LINEAR_CENTER_ENA(1) |
-			          S_0286E0_LINEAR_CENTROID_ENA(have_centroid);
-				
+				  S_0286E0_LINEAR_CENTROID_ENA(have_centroid);
+
 	r600_pipe_state_add_reg(rstate, R_0286CC_SPI_PS_IN_CONTROL_0,
 				spi_ps_in_control_0, 0xFFFFFFFF, NULL);
 	r600_pipe_state_add_reg(rstate, R_0286D0_SPI_PS_IN_CONTROL_1,
@@ -1544,14 +1638,8 @@ void evergreen_pipe_shader_vs(struct pipe_context *ctx, struct r600_pipe_shader
 				R_028864_SQ_PGM_RESOURCES_2_VS,
 				0x0, 0xFFFFFFFF, NULL);
 	r600_pipe_state_add_reg(rstate,
-			R_0288A8_SQ_PGM_RESOURCES_FS,
-			0x00000000, 0xFFFFFFFF, NULL);
-	r600_pipe_state_add_reg(rstate,
 			R_02885C_SQ_PGM_START_VS,
 			(r600_bo_offset(shader->bo)) >> 8, 0xFFFFFFFF, shader->bo);
-	r600_pipe_state_add_reg(rstate,
-			R_0288A4_SQ_PGM_START_FS,
-			(r600_bo_offset(shader->bo)) >> 8, 0xFFFFFFFF, shader->bo);
 
 	r600_pipe_state_add_reg(rstate,
 				R_03A200_SQ_LOOP_CONST_0 + (32 * 4), 0x01000FFF,
diff --git a/src/gallium/drivers/r600/evergreend.h b/src/gallium/drivers/r600/evergreend.h
index a337916c09..e67254b256 100644
--- a/src/gallium/drivers/r600/evergreend.h
+++ b/src/gallium/drivers/r600/evergreend.h
@@ -45,6 +45,15 @@
 
 #define EVENT_TYPE_ZPASS_DONE                  0x15
 #define EVENT_TYPE_CACHE_FLUSH_AND_INV_EVENT   0x16
+#define		EVENT_TYPE(x)                           ((x) << 0)
+#define		EVENT_INDEX(x)                          ((x) << 8)
+                /* 0 - any non-TS event
+		 * 1 - ZPASS_DONE
+		 * 2 - SAMPLE_PIPELINESTAT
+		 * 3 - SAMPLE_STREAMOUTSTAT*
+		 * 4 - *S_PARTIAL_FLUSH
+		 * 5 - TS events
+		 */
 
 #define R600_TEXEL_PITCH_ALIGNMENT_MASK        0x7
 
diff --git a/src/gallium/drivers/r600/r600.h b/src/gallium/drivers/r600/r600.h
index 17858b2d38..aa456d493f 100644
--- a/src/gallium/drivers/r600/r600.h
+++ b/src/gallium/drivers/r600/r600.h
@@ -91,6 +91,7 @@ enum radeon_family {
 	CHIP_JUNIPER,
 	CHIP_CYPRESS,
 	CHIP_HEMLOCK,
+	CHIP_PALM,
 	CHIP_LAST,
 };
 
@@ -264,6 +265,7 @@ void r600_context_fini(struct r600_context *ctx);
 void r600_context_pipe_state_set(struct r600_context *ctx, struct r600_pipe_state *state);
 void r600_context_pipe_state_set_ps_resource(struct r600_context *ctx, struct r600_pipe_state *state, unsigned rid);
 void r600_context_pipe_state_set_vs_resource(struct r600_context *ctx, struct r600_pipe_state *state, unsigned rid);
+void r600_context_pipe_state_set_fs_resource(struct r600_context *ctx, struct r600_pipe_state *state, unsigned rid);
 void r600_context_pipe_state_set_ps_sampler(struct r600_context *ctx, struct r600_pipe_state *state, unsigned id);
 void r600_context_pipe_state_set_vs_sampler(struct r600_context *ctx, struct r600_pipe_state *state, unsigned id);
 void r600_context_flush(struct r600_context *ctx);
@@ -284,10 +286,14 @@ int evergreen_context_init(struct r600_context *ctx, struct radeon *radeon);
 void evergreen_context_draw(struct r600_context *ctx, const struct r600_draw *draw);
 void evergreen_ps_resource_set(struct r600_context *ctx, struct r600_pipe_state *state, unsigned rid);
 void evergreen_vs_resource_set(struct r600_context *ctx, struct r600_pipe_state *state, unsigned rid);
+void evergreen_fs_resource_set(struct r600_context *ctx, struct r600_pipe_state *state, unsigned rid);
 
 void evergreen_context_pipe_state_set_ps_resource(struct r600_context *ctx, struct r600_pipe_state *state, unsigned rid);
 void evergreen_context_pipe_state_set_vs_resource(struct r600_context *ctx, struct r600_pipe_state *state, unsigned rid);
+void evergreen_context_pipe_state_set_fs_resource(struct r600_context *ctx, struct r600_pipe_state *state, unsigned rid);
 void evergreen_context_pipe_state_set_ps_sampler(struct r600_context *ctx, struct r600_pipe_state *state, unsigned id);
 void evergreen_context_pipe_state_set_vs_sampler(struct r600_context *ctx, struct r600_pipe_state *state, unsigned id);
 
+struct radeon *radeon_decref(struct radeon *radeon);
+
 #endif
diff --git a/src/gallium/drivers/r600/r600_asm.c b/src/gallium/drivers/r600/r600_asm.c
index 8a7f3ce575..1f41269534 100644
--- a/src/gallium/drivers/r600/r600_asm.c
+++ b/src/gallium/drivers/r600/r600_asm.c
@@ -22,11 +22,15 @@
  */
 #include <stdio.h>
 #include <errno.h>
+#include "util/u_format.h"
 #include "util/u_memory.h"
+#include "pipe/p_shader_tokens.h"
 #include "r600_pipe.h"
 #include "r600_sq.h"
 #include "r600_opcodes.h"
 #include "r600_asm.h"
+#include "r600_formats.h"
+#include "r600d.h"
 
 static inline unsigned int r600_bc_get_num_operands(struct r600_bc_alu *alu)
 {
@@ -55,8 +59,8 @@ static inline unsigned int r600_bc_get_num_operands(struct r600_bc_alu *alu)
 	case V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_DOT4:
 	case V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_DOT4_IEEE:
 	case V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_CUBE:
-		return 2;  
-		
+		return 2;
+
 	case V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MOV: 
 	case V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MOVA_FLOOR:
 	case V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_FRACT:
@@ -74,7 +78,7 @@ static inline unsigned int r600_bc_get_num_operands(struct r600_bc_alu *alu)
 	default: R600_ERR(
 		"Need instruction operand number for 0x%x.\n", alu->inst); 
 	};
-	
+
 	return 3;
 }
 
@@ -137,20 +141,21 @@ int r600_bc_init(struct r600_bc *bc, enum radeon_family family)
 	case CHIP_RV635:
 	case CHIP_RS780:
 	case CHIP_RS880:
-		bc->chiprev = 0;
+		bc->chiprev = CHIPREV_R600;
 		break;
 	case CHIP_RV770:
 	case CHIP_RV730:
 	case CHIP_RV710:
 	case CHIP_RV740:
-		bc->chiprev = 1;
+		bc->chiprev = CHIPREV_R700;
 		break;
 	case CHIP_CEDAR:
 	case CHIP_REDWOOD:
 	case CHIP_JUNIPER:
 	case CHIP_CYPRESS:
 	case CHIP_HEMLOCK:
-		bc->chiprev = 2;
+	case CHIP_PALM:
+		bc->chiprev = CHIPREV_EVERGREEN;
 		break;
 	default:
 		R600_ERR("unknown family %d\n", bc->family);
@@ -199,9 +204,9 @@ const unsigned bank_swizzle_vec[8] = {SQ_ALU_VEC_210,  //000
 				      SQ_ALU_VEC_012}; //111
 
 const unsigned bank_swizzle_scl[8] = {SQ_ALU_SCL_210,  //000
-				      SQ_ALU_SCL_122,  //001 
+				      SQ_ALU_SCL_122,  //001
 				      SQ_ALU_SCL_122,  //010
-				      
+
 				      SQ_ALU_SCL_221,  //011
 				      SQ_ALU_SCL_212,  //100
 				      SQ_ALU_SCL_122,  //101
@@ -592,10 +597,34 @@ int r600_bc_add_cfinst(struct r600_bc *bc, int inst)
 /* common to all 3 families */
 static int r600_bc_vtx_build(struct r600_bc *bc, struct r600_bc_vtx *vtx, unsigned id)
 {
-	bc->bytecode[id++] = S_SQ_VTX_WORD0_BUFFER_ID(vtx->buffer_id) |
-				S_SQ_VTX_WORD0_SRC_GPR(vtx->src_gpr) |
-				S_SQ_VTX_WORD0_SRC_SEL_X(vtx->src_sel_x) |
-				S_SQ_VTX_WORD0_MEGA_FETCH_COUNT(vtx->mega_fetch_count);
+	unsigned fetch_resource_start = 0;
+
+	/* check if we are fetch shader */
+			/* fetch shader can also access vertex resource,
+			 * first fetch shader resource is at 160
+			 */
+	if (bc->type == -1) {
+		switch (bc->chiprev) {
+		/* r600 */
+		case CHIPREV_R600:
+		/* r700 */
+		case CHIPREV_R700:
+			fetch_resource_start = 160;
+			break;
+		/* evergreen */
+		case CHIPREV_EVERGREEN:
+			fetch_resource_start = 0;
+			break;
+		default:
+			fprintf(stderr,  "%s:%s:%d unknown chiprev %d\n",
+				__FILE__, __func__, __LINE__, bc->chiprev);
+			break;
+		}
+	}
+	bc->bytecode[id++] = S_SQ_VTX_WORD0_BUFFER_ID(vtx->buffer_id + fetch_resource_start) |
+			S_SQ_VTX_WORD0_SRC_GPR(vtx->src_gpr) |
+			S_SQ_VTX_WORD0_SRC_SEL_X(vtx->src_sel_x) |
+			S_SQ_VTX_WORD0_MEGA_FETCH_COUNT(vtx->mega_fetch_count);
 	bc->bytecode[id++] = S_SQ_VTX_WORD1_DST_SEL_X(vtx->dst_sel_x) |
 				S_SQ_VTX_WORD1_DST_SEL_Y(vtx->dst_sel_y) |
 				S_SQ_VTX_WORD1_DST_SEL_Z(vtx->dst_sel_z) |
@@ -678,8 +707,8 @@ static int r600_bc_alu_build(struct r600_bc *bc, struct r600_bc_alu *alu, unsign
 					S_SQ_ALU_WORD1_OP2_WRITE_MASK(alu->dst.write) |
 					S_SQ_ALU_WORD1_OP2_ALU_INST(alu->inst) |
 					S_SQ_ALU_WORD1_BANK_SWIZZLE(alu->bank_swizzle) |
-			                S_SQ_ALU_WORD1_OP2_UPDATE_EXECUTE_MASK(alu->predicate) |
-		 	                S_SQ_ALU_WORD1_OP2_UPDATE_PRED(alu->predicate);
+					S_SQ_ALU_WORD1_OP2_UPDATE_EXECUTE_MASK(alu->predicate) |
+					S_SQ_ALU_WORD1_OP2_UPDATE_PRED(alu->predicate);
 	}
 	if (alu->last) {
 		if (alu->nliteral && !alu->literal_added) {
@@ -710,7 +739,7 @@ static int r600_bc_cf_build(struct r600_bc *bc, struct r600_bc_cf *cf)
 			S_SQ_CF_ALU_WORD1_KCACHE_ADDR0(cf->kcache0_addr) |
 			S_SQ_CF_ALU_WORD1_KCACHE_ADDR1(cf->kcache1_addr) |
 					S_SQ_CF_ALU_WORD1_BARRIER(1) |
-					S_SQ_CF_ALU_WORD1_USES_WATERFALL(bc->chiprev == 0 ? cf->r6xx_uses_waterfall : 0) |
+					S_SQ_CF_ALU_WORD1_USES_WATERFALL(bc->chiprev == CHIPREV_R600 ? cf->r6xx_uses_waterfall : 0) |
 					S_SQ_CF_ALU_WORD1_COUNT((cf->ndw / 2) - 1);
 		break;
 	case V_SQ_CF_WORD1_SQ_CF_INST_TEX:
@@ -742,6 +771,8 @@ static int r600_bc_cf_build(struct r600_bc *bc, struct r600_bc_cf *cf)
 	case V_SQ_CF_WORD1_SQ_CF_INST_LOOP_END:
 	case V_SQ_CF_WORD1_SQ_CF_INST_LOOP_CONTINUE:
 	case V_SQ_CF_WORD1_SQ_CF_INST_LOOP_BREAK:
+	case V_SQ_CF_WORD1_SQ_CF_INST_CALL_FS:
+	case V_SQ_CF_WORD1_SQ_CF_INST_RETURN:
 		bc->bytecode[id++] = S_SQ_CF_WORD0_ADDR(cf->cf_addr >> 1);
 		bc->bytecode[id++] = S_SQ_CF_WORD1_CF_INST(cf->inst) |
 					S_SQ_CF_WORD1_BARRIER(1) |
@@ -766,7 +797,10 @@ int r600_bc_build(struct r600_bc *bc)
 	int r;
 
 	if (bc->callstack[0].max > 0)
-	    bc->nstack = ((bc->callstack[0].max + 3) >> 2) + 2;
+		bc->nstack = ((bc->callstack[0].max + 3) >> 2) + 2;
+	if (bc->type == TGSI_PROCESSOR_VERTEX && !bc->nstack) {
+		bc->nstack = 1;
+	}
 
 	/* first path compute addr of each CF block */
 	/* addr start after all the CF instructions */
@@ -795,6 +829,8 @@ int r600_bc_build(struct r600_bc *bc)
 		case V_SQ_CF_WORD1_SQ_CF_INST_LOOP_END:
 		case V_SQ_CF_WORD1_SQ_CF_INST_LOOP_CONTINUE:
 		case V_SQ_CF_WORD1_SQ_CF_INST_LOOP_BREAK:
+		case V_SQ_CF_WORD1_SQ_CF_INST_CALL_FS:
+		case V_SQ_CF_WORD1_SQ_CF_INST_RETURN:
 			break;
 		default:
 			R600_ERR("unsupported CF instruction (0x%X)\n", cf->inst);
@@ -810,7 +846,7 @@ int r600_bc_build(struct r600_bc *bc)
 		return -ENOMEM;
 	LIST_FOR_EACH_ENTRY(cf, &bc->cf, list) {
 		addr = cf->addr;
-		if (bc->chiprev == 2)
+		if (bc->chiprev == CHIPREV_EVERGREEN)
 			r = eg_bc_cf_build(bc, cf);
 		else
 			r = r600_bc_cf_build(bc, cf);
@@ -821,11 +857,11 @@ int r600_bc_build(struct r600_bc *bc)
 		case (V_SQ_CF_ALU_WORD1_SQ_CF_INST_ALU_PUSH_BEFORE << 3):
 			LIST_FOR_EACH_ENTRY(alu, &cf->alu, list) {
 				switch(bc->chiprev) {
-				case 0:
+				case CHIPREV_R600:
 					r = r600_bc_alu_build(bc, alu, addr);
 					break;
-				case 1:
-				case 2: /* eg alu is same encoding as r700 */
+				case CHIPREV_R700:
+				case CHIPREV_EVERGREEN: /* eg alu is same encoding as r700 */
 					r = r700_bc_alu_build(bc, alu, addr);
 					break;
 				default:
@@ -868,6 +904,8 @@ int r600_bc_build(struct r600_bc *bc)
 		case V_SQ_CF_WORD1_SQ_CF_INST_JUMP:
 		case V_SQ_CF_WORD1_SQ_CF_INST_ELSE:
 		case V_SQ_CF_WORD1_SQ_CF_INST_POP:
+		case V_SQ_CF_WORD1_SQ_CF_INST_CALL_FS:
+		case V_SQ_CF_WORD1_SQ_CF_INST_RETURN:
 			break;
 		default:
 			R600_ERR("unsupported CF instruction (0x%X)\n", cf->inst);
@@ -912,3 +950,342 @@ void r600_bc_clear(struct r600_bc *bc)
 
 	LIST_INITHEAD(&cf->list);
 }
+
+void r600_bc_dump(struct r600_bc *bc)
+{
+	unsigned i;
+	char chip = '6';
+
+	switch (bc->chiprev) {
+	case 1:
+		chip = '7';
+		break;
+	case 2:
+		chip = 'E';
+		break;
+	case 0:
+	default:
+		chip = '6';
+		break;
+	}
+	fprintf(stderr, "bytecode %d dw -----------------------\n", bc->ndw);
+	fprintf(stderr, "     %c\n", chip);
+	for (i = 0; i < bc->ndw; i++) {
+		fprintf(stderr, "0x%08X\n", bc->bytecode[i]);
+	}
+	fprintf(stderr, "--------------------------------------\n");
+}
+
+void r600_cf_vtx(struct r600_vertex_element *ve, u32 *bytecode, unsigned count)
+{
+	struct r600_pipe_state *rstate;
+	unsigned i = 0;
+
+	if (count > 8) {
+		bytecode[i++] = S_SQ_CF_WORD0_ADDR(8 >> 1);
+		bytecode[i++] = S_SQ_CF_WORD1_CF_INST(V_SQ_CF_WORD1_SQ_CF_INST_VTX) |
+						S_SQ_CF_WORD1_BARRIER(1) |
+						S_SQ_CF_WORD1_COUNT(8 - 1);
+		bytecode[i++] = S_SQ_CF_WORD0_ADDR(40 >> 1);
+		bytecode[i++] = S_SQ_CF_WORD1_CF_INST(V_SQ_CF_WORD1_SQ_CF_INST_VTX) |
+						S_SQ_CF_WORD1_BARRIER(1) |
+						S_SQ_CF_WORD1_COUNT(count - 8 - 1);
+	} else {
+		bytecode[i++] = S_SQ_CF_WORD0_ADDR(8 >> 1);
+		bytecode[i++] = S_SQ_CF_WORD1_CF_INST(V_SQ_CF_WORD1_SQ_CF_INST_VTX) |
+						S_SQ_CF_WORD1_BARRIER(1) |
+						S_SQ_CF_WORD1_COUNT(count - 1);
+	}
+	bytecode[i++] = S_SQ_CF_WORD0_ADDR(0);
+	bytecode[i++] = S_SQ_CF_WORD1_CF_INST(V_SQ_CF_WORD1_SQ_CF_INST_RETURN) |
+			S_SQ_CF_WORD1_BARRIER(1);
+
+	rstate = &ve->rstate;
+	rstate->id = R600_PIPE_STATE_FETCH_SHADER;
+	rstate->nregs = 0;
+	r600_pipe_state_add_reg(rstate, R_0288A4_SQ_PGM_RESOURCES_FS,
+				0x00000000, 0xFFFFFFFF, NULL);
+	r600_pipe_state_add_reg(rstate, R_0288DC_SQ_PGM_CF_OFFSET_FS,
+				0x00000000, 0xFFFFFFFF, NULL);
+	r600_pipe_state_add_reg(rstate, R_028894_SQ_PGM_START_FS,
+				r600_bo_offset(ve->fetch_shader) >> 8,
+				0xFFFFFFFF, ve->fetch_shader);
+}
+
+void r600_cf_vtx_tc(struct r600_vertex_element *ve, u32 *bytecode, unsigned count)
+{
+	struct r600_pipe_state *rstate;
+	unsigned i = 0;
+
+	if (count > 8) {
+		bytecode[i++] = S_SQ_CF_WORD0_ADDR(8 >> 1);
+		bytecode[i++] = S_SQ_CF_WORD1_CF_INST(V_SQ_CF_WORD1_SQ_CF_INST_VTX_TC) |
+						S_SQ_CF_WORD1_BARRIER(1) |
+						S_SQ_CF_WORD1_COUNT(8 - 1);
+		bytecode[i++] = S_SQ_CF_WORD0_ADDR(40 >> 1);
+		bytecode[i++] = S_SQ_CF_WORD1_CF_INST(V_SQ_CF_WORD1_SQ_CF_INST_VTX_TC) |
+						S_SQ_CF_WORD1_BARRIER(1) |
+						S_SQ_CF_WORD1_COUNT((count - 8) - 1);
+	} else {
+		bytecode[i++] = S_SQ_CF_WORD0_ADDR(8 >> 1);
+		bytecode[i++] = S_SQ_CF_WORD1_CF_INST(V_SQ_CF_WORD1_SQ_CF_INST_VTX_TC) |
+						S_SQ_CF_WORD1_BARRIER(1) |
+						S_SQ_CF_WORD1_COUNT(count - 1);
+	}
+	bytecode[i++] = S_SQ_CF_WORD0_ADDR(0);
+	bytecode[i++] = S_SQ_CF_WORD1_CF_INST(V_SQ_CF_WORD1_SQ_CF_INST_RETURN) |
+			S_SQ_CF_WORD1_BARRIER(1);
+
+	rstate = &ve->rstate;
+	rstate->id = R600_PIPE_STATE_FETCH_SHADER;
+	rstate->nregs = 0;
+	r600_pipe_state_add_reg(rstate, R_0288A4_SQ_PGM_RESOURCES_FS,
+				0x00000000, 0xFFFFFFFF, NULL);
+	r600_pipe_state_add_reg(rstate, R_0288DC_SQ_PGM_CF_OFFSET_FS,
+				0x00000000, 0xFFFFFFFF, NULL);
+	r600_pipe_state_add_reg(rstate, R_028894_SQ_PGM_START_FS,
+				r600_bo_offset(ve->fetch_shader) >> 8,
+				0xFFFFFFFF, ve->fetch_shader);
+}
+
+static void r600_vertex_data_type(enum pipe_format pformat, unsigned *format,
+				unsigned *num_format, unsigned *format_comp)
+{
+	const struct util_format_description *desc;
+	unsigned i;
+
+	*format = 0;
+	*num_format = 0;
+	*format_comp = 0;
+
+	desc = util_format_description(pformat);
+	if (desc->layout != UTIL_FORMAT_LAYOUT_PLAIN) {
+		goto out_unknown;
+	}
+
+	/* Find the first non-VOID channel. */
+	for (i = 0; i < 4; i++) {
+		if (desc->channel[i].type != UTIL_FORMAT_TYPE_VOID) {
+			break;
+		}
+	}
+
+	switch (desc->channel[i].type) {
+		/* Half-floats, floats, doubles */
+	case UTIL_FORMAT_TYPE_FLOAT:
+		switch (desc->channel[i].size) {
+		case 16:
+			switch (desc->nr_channels) {
+			case 1:
+				*format = FMT_16_FLOAT;
+				break;
+			case 2:
+				*format = FMT_16_16_FLOAT;
+				break;
+			case 3:
+				*format = FMT_16_16_16_FLOAT;
+				break;
+			case 4:
+				*format = FMT_16_16_16_16_FLOAT;
+				break;
+			}
+			break;
+		case 32:
+			switch (desc->nr_channels) {
+			case 1:
+				*format = FMT_32_FLOAT;
+				break;
+			case 2:
+				*format = FMT_32_32_FLOAT;
+				break;
+			case 3:
+				*format = FMT_32_32_32_FLOAT;
+				break;
+			case 4:
+				*format = FMT_32_32_32_32_FLOAT;
+				break;
+			}
+			break;
+		default:
+			goto out_unknown;
+		}
+		break;
+		/* Unsigned ints */
+	case UTIL_FORMAT_TYPE_UNSIGNED:
+		/* Signed ints */
+	case UTIL_FORMAT_TYPE_SIGNED:
+		switch (desc->channel[i].size) {
+		case 8:
+			switch (desc->nr_channels) {
+			case 1:
+				*format = FMT_8;
+				break;
+			case 2:
+				*format = FMT_8_8;
+				break;
+			case 3:
+			//	*format = FMT_8_8_8; /* fails piglit draw-vertices test */
+			//	break;
+			case 4:
+				*format = FMT_8_8_8_8;
+				break;
+			}
+			break;
+		case 16:
+			switch (desc->nr_channels) {
+			case 1:
+				*format = FMT_16;
+				break;
+			case 2:
+				*format = FMT_16_16;
+				break;
+			case 3:
+			//	*format = FMT_16_16_16; /* fails piglit draw-vertices test */
+			//	break;
+			case 4:
+				*format = FMT_16_16_16_16;
+				break;
+			}
+			break;
+		case 32:
+			switch (desc->nr_channels) {
+			case 1:
+				*format = FMT_32;
+				break;
+			case 2:
+				*format = FMT_32_32;
+				break;
+			case 3:
+				*format = FMT_32_32_32;
+				break;
+			case 4:
+				*format = FMT_32_32_32_32;
+				break;
+			}
+			break;
+		default:
+			goto out_unknown;
+		}
+		break;
+	default:
+		goto out_unknown;
+	}
+
+	if (desc->channel[i].type == UTIL_FORMAT_TYPE_SIGNED) {
+		*format_comp = 1;
+	}
+	if (desc->channel[i].normalized) {
+		*num_format = 0;
+	} else {
+		*num_format = 2;
+	}
+	return;
+out_unknown:
+	R600_ERR("unsupported vertex format %s\n", util_format_name(pformat));
+}
+
+static void r600_bc(unsigned ndw, unsigned chiprev, u32 *bytecode)
+{
+	unsigned i;
+	char chip = '6';
+
+	switch (chiprev) {
+	case 1:
+		chip = '7';
+		break;
+	case 2:
+		chip = 'E';
+		break;
+	case 0:
+	default:
+		chip = '6';
+		break;
+	}
+	fprintf(stderr, "bytecode %d dw -----------------------\n", ndw);
+	fprintf(stderr, "    %c\n", chip);
+	for (i = 0; i < ndw; i++) {
+		fprintf(stderr, "0x%08X\n", bytecode[i]);
+	}
+	fprintf(stderr, "--------------------------------------\n");
+}
+
+int r600_vertex_elements_build_fetch_shader(struct r600_pipe_context *rctx, struct r600_vertex_element *ve)
+{
+	unsigned ndw, i;
+	u32 *bytecode;
+	unsigned fetch_resource_start = 0, format, num_format, format_comp;
+	struct pipe_vertex_element *elements = ve->elements;
+	const struct util_format_description *desc;
+
+	/* 2 dwords for cf aligned to 4 + 4 dwords per input */
+	ndw = 8 + ve->count * 4;
+	ve->fs_size = ndw * 4;
+
+	/* use PIPE_BIND_VERTEX_BUFFER so we use the cache buffer manager */
+	ve->fetch_shader = r600_bo(rctx->radeon, ndw*4, 256, PIPE_BIND_VERTEX_BUFFER, 0);
+	if (ve->fetch_shader == NULL) {
+		return -ENOMEM;
+	}
+
+	bytecode = r600_bo_map(rctx->radeon, ve->fetch_shader, 0, NULL);
+	if (bytecode == NULL) {
+		r600_bo_reference(rctx->radeon, &ve->fetch_shader, NULL);
+		return -ENOMEM;
+	}
+
+	if (rctx->family >= CHIP_CEDAR) {
+		eg_cf_vtx(ve, &bytecode[0], (ndw - 8) / 4);
+	} else {
+		r600_cf_vtx(ve, &bytecode[0], (ndw - 8) / 4);
+		fetch_resource_start = 160;
+	}
+
+	/* vertex elements offset need special handling, if offset is bigger
+	 * than what we can put in fetch instruction then we need to alterate
+	 * the vertex resource offset. In such case in order to simplify code
+	 * we will bound one resource per elements. It's a worst case scenario.
+	 */
+	for (i = 0; i < ve->count; i++) {
+		ve->vbuffer_offset[i] = C_SQ_VTX_WORD2_OFFSET & elements[i].src_offset;
+		if (ve->vbuffer_offset[i]) {
+			ve->vbuffer_need_offset = 1;
+		}
+	}
+
+	for (i = 0; i < ve->count; i++) {
+		unsigned vbuffer_index;
+		r600_vertex_data_type(ve->hw_format[i], &format, &num_format, &format_comp);
+		desc = util_format_description(ve->hw_format[i]);
+		if (desc == NULL) {
+			R600_ERR("unknown format %d\n", ve->hw_format[i]);
+			r600_bo_reference(rctx->radeon, &ve->fetch_shader, NULL);
+			return -EINVAL;
+		}
+
+		/* see above for vbuffer_need_offset explanation */
+		vbuffer_index = elements[i].vertex_buffer_index;
+		if (ve->vbuffer_need_offset) {
+			bytecode[8 + i * 4 + 0] = S_SQ_VTX_WORD0_BUFFER_ID(i + fetch_resource_start);
+		} else {
+			bytecode[8 + i * 4 + 0] = S_SQ_VTX_WORD0_BUFFER_ID(vbuffer_index + fetch_resource_start);
+		}
+		bytecode[8 + i * 4 + 0] |= S_SQ_VTX_WORD0_SRC_GPR(0) |
+					S_SQ_VTX_WORD0_SRC_SEL_X(0) |
+					S_SQ_VTX_WORD0_MEGA_FETCH_COUNT(0x1F);
+		bytecode[8 + i * 4 + 1] = S_SQ_VTX_WORD1_DST_SEL_X(desc->swizzle[0]) |
+					S_SQ_VTX_WORD1_DST_SEL_Y(desc->swizzle[1]) |
+					S_SQ_VTX_WORD1_DST_SEL_Z(desc->swizzle[2]) |
+					S_SQ_VTX_WORD1_DST_SEL_W(desc->swizzle[3]) |
+					S_SQ_VTX_WORD1_USE_CONST_FIELDS(0) |
+					S_SQ_VTX_WORD1_DATA_FORMAT(format) |
+					S_SQ_VTX_WORD1_NUM_FORMAT_ALL(num_format) |
+					S_SQ_VTX_WORD1_FORMAT_COMP_ALL(format_comp) |
+					S_SQ_VTX_WORD1_SRF_MODE_ALL(1) |
+					S_SQ_VTX_WORD1_GPR_DST_GPR(i + 1);
+		bytecode[8 + i * 4 + 2] = S_SQ_VTX_WORD2_OFFSET(elements[i].src_offset) |
+					S_SQ_VTX_WORD2_MEGA_FETCH(1);
+		bytecode[8 + i * 4 + 3] = 0;
+	}
+	r600_bo_unmap(rctx->radeon, ve->fetch_shader);
+	return 0;
+}
diff --git a/src/gallium/drivers/r600/r600_asm.h b/src/gallium/drivers/r600/r600_asm.h
index 25cda16837..b147f0f5c8 100644
--- a/src/gallium/drivers/r600/r600_asm.h
+++ b/src/gallium/drivers/r600/r600_asm.h
@@ -28,6 +28,9 @@
 #define NUM_OF_CYCLES 3
 #define NUM_OF_COMPONENTS 4
 
+struct r600_vertex_element;
+struct r600_pipe_context;
+
 struct r600_bc_alu_src {
 	unsigned			sel;
 	unsigned			chan;
@@ -145,12 +148,12 @@ struct r600_bc_cf {
 	struct r600_bc_alu		*curr_bs_head;
 };
 
-#define FC_NONE 0
-#define FC_IF 1
-#define FC_LOOP 2
-#define FC_REP 3
-#define FC_PUSH_VPM 4
-#define FC_PUSH_WQM 5
+#define FC_NONE				0
+#define FC_IF				1
+#define FC_LOOP				2
+#define FC_REP				3
+#define FC_PUSH_VPM			4
+#define FC_PUSH_WQM			5
 
 struct r600_cf_stack_entry {
 	int				type;
@@ -166,10 +169,11 @@ struct r600_cf_callstack {
 	int				current;
 	int				max;
 };
-	
+
 struct r600_bc {
 	enum radeon_family		family;
 	int				chiprev; /* 0 - r600, 1 - r700, 2 - evergreen */
+	int				type;
 	struct list_head		cf;
 	struct r600_bc_cf		*cf_last;
 	unsigned			ndw;
@@ -187,6 +191,7 @@ struct r600_bc {
 
 /* eg_asm.c */
 int eg_bc_cf_build(struct r600_bc *bc, struct r600_bc_cf *cf);
+void eg_cf_vtx(struct r600_vertex_element *ve, u32 *bytecode, unsigned count);
 
 /* r600_asm.c */
 int r600_bc_init(struct r600_bc *bc, enum radeon_family family);
@@ -199,6 +204,11 @@ int r600_bc_add_output(struct r600_bc *bc, const struct r600_bc_output *output);
 int r600_bc_build(struct r600_bc *bc);
 int r600_bc_add_cfinst(struct r600_bc *bc, int inst);
 int r600_bc_add_alu_type(struct r600_bc *bc, const struct r600_bc_alu *alu, int type);
+void r600_bc_dump(struct r600_bc *bc);
+void r600_cf_vtx(struct r600_vertex_element *ve, u32 *bytecode, unsigned count);
+void r600_cf_vtx_tc(struct r600_vertex_element *ve, u32 *bytecode, unsigned count);
+
+int r600_vertex_elements_build_fetch_shader(struct r600_pipe_context *rctx, struct r600_vertex_element *ve);
 
 /* r700_asm.c */
 int r700_bc_alu_build(struct r600_bc *bc, struct r600_bc_alu *alu, unsigned id);
diff --git a/src/gallium/drivers/r600/r600_blit.c b/src/gallium/drivers/r600/r600_blit.c
index 50d47060c1..0f04136fb2 100644
--- a/src/gallium/drivers/r600/r600_blit.c
+++ b/src/gallium/drivers/r600/r600_blit.c
@@ -27,9 +27,9 @@
 
 enum r600_blitter_op /* bitmask */
 {
-    R600_CLEAR         = 1,
-    R600_CLEAR_SURFACE = 2,
-    R600_COPY          = 4
+	R600_CLEAR         = 1,
+	R600_CLEAR_SURFACE = 2,
+	R600_COPY          = 4
 };
 
 static void r600_blitter_begin(struct pipe_context *ctx, enum r600_blitter_op op)
@@ -81,16 +81,21 @@ static void r600_blitter_end(struct pipe_context *ctx)
 int r600_blit_uncompress_depth(struct pipe_context *ctx, struct r600_resource_texture *texture)
 {
 	struct r600_pipe_context *rctx = (struct r600_pipe_context *)ctx;
-	struct pipe_surface *zsurf, *cbsurf;
+	struct pipe_surface *zsurf, *cbsurf, surf_tmpl;
 	int level = 0;
 	float depth = 1.0f;
+	surf_tmpl.format = texture->resource.base.b.format;
+	surf_tmpl.u.tex.level = level;
+	surf_tmpl.u.tex.first_layer = 0;
+	surf_tmpl.u.tex.last_layer = 0;
+	surf_tmpl.usage = PIPE_BIND_DEPTH_STENCIL;
 
-	zsurf = ctx->screen->get_tex_surface(ctx->screen, &texture->resource.base.b, 0, level, 0,
-					     PIPE_BIND_DEPTH_STENCIL);
+	zsurf = ctx->create_surface(ctx, &texture->resource.base.b, &surf_tmpl);
 
-	cbsurf = ctx->screen->get_tex_surface(ctx->screen,
-			(struct pipe_resource*)texture->flushed_depth_texture,
-			0, level, 0, PIPE_BIND_RENDER_TARGET);
+	surf_tmpl.format = ((struct pipe_resource*)texture->flushed_depth_texture)->format;
+	surf_tmpl.usage = PIPE_BIND_RENDER_TARGET;
+	cbsurf = ctx->create_surface(ctx,
+			(struct pipe_resource*)texture->flushed_depth_texture, &surf_tmpl);
 
 	if (rctx->family == CHIP_RV610 || rctx->family == CHIP_RV630 ||
 	    rctx->family == CHIP_RV620 || rctx->family == CHIP_RV635)
@@ -154,42 +159,38 @@ static void r600_clear_depth_stencil(struct pipe_context *ctx,
 
 /* Copy a block of pixels from one surface to another using HW. */
 static void r600_hw_copy_region(struct pipe_context *ctx,
-                                struct pipe_resource *dst,
-                                struct pipe_subresource subdst,
-                                unsigned dstx, unsigned dsty, unsigned dstz,
-                                struct pipe_resource *src,
-                                struct pipe_subresource subsrc,
-                                unsigned srcx, unsigned srcy, unsigned srcz,
-                                unsigned width, unsigned height)
+				struct pipe_resource *dst,
+				unsigned dst_level,
+				unsigned dstx, unsigned dsty, unsigned dstz,
+				struct pipe_resource *src,
+				unsigned src_level,
+				const struct pipe_box *src_box)
 {
 	struct r600_pipe_context *rctx = (struct r600_pipe_context *)ctx;
 
 	r600_blitter_begin(ctx, R600_COPY);
-	util_blitter_copy_region(rctx->blitter, dst, subdst, dstx, dsty, dstz,
-				 src, subsrc, srcx, srcy, srcz, width, height,
-				 TRUE);
+	util_blitter_copy_region(rctx->blitter, dst, dst_level, dstx, dsty, dstz,
+				 src, src_level, src_box, TRUE);
 	r600_blitter_end(ctx);
 }
 
 static void r600_resource_copy_region(struct pipe_context *ctx,
 				      struct pipe_resource *dst,
-				      struct pipe_subresource subdst,
+				      unsigned dst_level,
 				      unsigned dstx, unsigned dsty, unsigned dstz,
 				      struct pipe_resource *src,
-				      struct pipe_subresource subsrc,
-				      unsigned srcx, unsigned srcy, unsigned srcz,
-				      unsigned width, unsigned height)
+				      unsigned src_level,
+				      const struct pipe_box *src_box)
 {
 	boolean is_depth;
 	/* there is something wrong with depth resource copies at the moment so avoid them for now */
 	is_depth = util_format_get_component_bits(src->format, UTIL_FORMAT_COLORSPACE_ZS, 0) != 0;
 	if (is_depth)
-		util_resource_copy_region(ctx, dst, subdst, dstx, dsty, dstz,
-					  src, subsrc, srcx, srcy, srcz, width, height);
+		util_resource_copy_region(ctx, dst, dst_level, dstx, dsty, dstz,
+					  src, src_level, src_box);
 	else
-		r600_hw_copy_region(ctx, dst, subdst, dstx, dsty, dstz,
-				    src, subsrc, srcx, srcy, srcz, width, height);
-
+		r600_hw_copy_region(ctx, dst, dst_level, dstx, dsty, dstz,
+				    src, src_level, src_box);
 }
 
 void r600_init_blit_functions(struct r600_pipe_context *rctx)
diff --git a/src/gallium/drivers/r600/r600_buffer.c b/src/gallium/drivers/r600/r600_buffer.c
index ed97b6e69a..a17c54d6ee 100644
--- a/src/gallium/drivers/r600/r600_buffer.c
+++ b/src/gallium/drivers/r600/r600_buffer.c
@@ -29,7 +29,6 @@
 #include <util/u_math.h>
 #include <util/u_inlines.h>
 #include <util/u_memory.h>
-#include <util/u_upload_mgr.h>
 #include "state_tracker/drm_driver.h"
 #include <xf86drm.h>
 #include "radeon_drm.h"
@@ -53,12 +52,13 @@ struct pipe_resource *r600_buffer_create(struct pipe_screen *screen,
 
 	rbuffer->magic = R600_BUFFER_MAGIC;
 	rbuffer->user_buffer = NULL;
-	rbuffer->num_ranges = 0;
 	rbuffer->r.base.b = *templ;
 	pipe_reference_init(&rbuffer->r.base.b.reference, 1);
 	rbuffer->r.base.b.screen = screen;
 	rbuffer->r.base.vtbl = &r600_buffer_vtbl;
 	rbuffer->r.size = rbuffer->r.base.b.width0;
+	rbuffer->r.bo_size = rbuffer->r.size;
+	rbuffer->uploaded = FALSE;
 	bo = r600_bo((struct radeon*)screen->winsys, rbuffer->r.base.b.width0, alignment, rbuffer->r.base.b.bind, rbuffer->r.base.b.usage);
 	if (bo == NULL) {
 		FREE(rbuffer);
@@ -89,10 +89,12 @@ struct pipe_resource *r600_user_buffer_create(struct pipe_screen *screen,
 	rbuffer->r.base.b.width0 = bytes;
 	rbuffer->r.base.b.height0 = 1;
 	rbuffer->r.base.b.depth0 = 1;
+	rbuffer->r.base.b.array_size = 1;
 	rbuffer->r.base.b.flags = 0;
-	rbuffer->num_ranges = 0;
 	rbuffer->r.bo = NULL;
+	rbuffer->r.bo_size = 0;
 	rbuffer->user_buffer = ptr;
+	rbuffer->uploaded = FALSE;
 	return &rbuffer->r.base.b;
 }
 
@@ -104,6 +106,7 @@ static void r600_buffer_destroy(struct pipe_screen *screen,
 	if (rbuffer->r.bo) {
 		r600_bo_reference((struct radeon*)screen->winsys, &rbuffer->r.bo, NULL);
 	}
+	rbuffer->r.bo = NULL;
 	FREE(rbuffer);
 }
 
@@ -113,29 +116,10 @@ static void *r600_buffer_transfer_map(struct pipe_context *pipe,
 	struct r600_resource_buffer *rbuffer = r600_buffer(transfer->resource);
 	int write = 0;
 	uint8_t *data;
-	int i;
-	boolean flush = FALSE;
 
 	if (rbuffer->user_buffer)
 		return (uint8_t*)rbuffer->user_buffer + transfer->box.x;
 
-	if (transfer->usage & PIPE_TRANSFER_DISCARD) {
-		for (i = 0; i < rbuffer->num_ranges; i++) {
-			if ((transfer->box.x >= rbuffer->ranges[i].start) &&
-			    (transfer->box.x < rbuffer->ranges[i].end))
-				flush = TRUE;
-			
-			if (flush) {
-				r600_bo_reference((struct radeon*)pipe->winsys, &rbuffer->r.bo, NULL);
-				rbuffer->num_ranges = 0;
-				rbuffer->r.bo = r600_bo((struct radeon*)pipe->winsys,
-                                                        rbuffer->r.base.b.width0, 0,
-                                                        rbuffer->r.base.b.bind,
-                                                        rbuffer->r.base.b.usage);
-				break;
-			}
-		}
-	}
 	if (transfer->usage & PIPE_TRANSFER_DONTBLOCK) {
 		/* FIXME */
 	}
@@ -154,41 +138,22 @@ static void r600_buffer_transfer_unmap(struct pipe_context *pipe,
 {
 	struct r600_resource_buffer *rbuffer = r600_buffer(transfer->resource);
 
+	if (rbuffer->user_buffer)
+		return;
+
 	if (rbuffer->r.bo)
 		r600_bo_unmap((struct radeon*)pipe->winsys, rbuffer->r.bo);
 }
 
 static void r600_buffer_transfer_flush_region(struct pipe_context *pipe,
-					      struct pipe_transfer *transfer,
-					      const struct pipe_box *box)
+						struct pipe_transfer *transfer,
+						const struct pipe_box *box)
 {
-	struct r600_resource_buffer *rbuffer = r600_buffer(transfer->resource);
-	unsigned i;
-	unsigned offset = transfer->box.x + box->x;
-	unsigned length = box->width;
-
-	assert(box->x + box->width <= transfer->box.width);
-
-	if (rbuffer->user_buffer)
-		return;
-
-	/* mark the range as used */
-	for(i = 0; i < rbuffer->num_ranges; ++i) {
-		if(offset <= rbuffer->ranges[i].end && rbuffer->ranges[i].start <= (offset+box->width)) {
-			rbuffer->ranges[i].start = MIN2(rbuffer->ranges[i].start, offset);
-			rbuffer->ranges[i].end   = MAX2(rbuffer->ranges[i].end, (offset+length));
-			return;
-		}
-	}
-	
-	rbuffer->ranges[rbuffer->num_ranges].start = offset;
-	rbuffer->ranges[rbuffer->num_ranges].end = offset+length;
-	rbuffer->num_ranges++;
 }
 
 unsigned r600_buffer_is_referenced_by_cs(struct pipe_context *context,
 					 struct pipe_resource *buf,
-					 unsigned face, unsigned level)
+					 unsigned level, int layer)
 {
 	/* FIXME */
 	return PIPE_REFERENCED_FOR_READ | PIPE_REFERENCED_FOR_WRITE;
@@ -235,29 +200,25 @@ struct u_resource_vtbl r600_buffer_vtbl =
 
 int r600_upload_index_buffer(struct r600_pipe_context *rctx, struct r600_drawl *draw)
 {
-	struct pipe_resource *upload_buffer = NULL;
-	unsigned index_offset = draw->index_buffer_offset;
-	int ret = 0;
-
 	if (r600_buffer_is_user_buffer(draw->index_buffer)) {
-		ret = u_upload_buffer(rctx->upload_ib,
-				      index_offset,
-				      draw->count * draw->index_size,
-				      draw->index_buffer,
-				      &index_offset,
-				      &upload_buffer);
-		if (ret) {
-			goto done;
-		}
-		draw->index_buffer_offset = index_offset;
-
-		/* Transfer ownership. */
-		pipe_resource_reference(&draw->index_buffer, upload_buffer);
-		pipe_resource_reference(&upload_buffer, NULL);
+		struct r600_resource_buffer *rbuffer = r600_buffer(draw->index_buffer);
+		unsigned upload_offset;
+		int ret = 0;
+
+		ret = r600_upload_buffer(rctx->rupload_vb,
+					draw->index_buffer_offset,
+					draw->count * draw->index_size,
+					rbuffer,
+					&upload_offset,
+					&rbuffer->r.bo_size,
+					&rbuffer->r.bo);
+		if (ret)
+			return ret;
+		rbuffer->uploaded = TRUE;
+		draw->index_buffer_offset = upload_offset;
 	}
 
-done:
-	return ret;
+	return 0;
 }
 
 int r600_upload_user_buffers(struct r600_pipe_context *rctx)
@@ -266,25 +227,24 @@ int r600_upload_user_buffers(struct r600_pipe_context *rctx)
 	int i, nr;
 
 	nr = rctx->vertex_elements->count;
+	nr = rctx->nvertex_buffer;
 
 	for (i = 0; i < nr; i++) {
-		struct pipe_vertex_buffer *vb =
-			&rctx->vertex_buffer[rctx->vertex_elements->elements[i].vertex_buffer_index];
+		struct pipe_vertex_buffer *vb = &rctx->vertex_buffer[i];
 
 		if (r600_buffer_is_user_buffer(vb->buffer)) {
-			struct pipe_resource *upload_buffer = NULL;
-			unsigned offset = 0; /*vb->buffer_offset * 4;*/
-			unsigned size = vb->buffer->width0;
+			struct r600_resource_buffer *rbuffer = r600_buffer(vb->buffer);
 			unsigned upload_offset;
-			ret = u_upload_buffer(rctx->upload_vb,
-					      offset, size,
-					      vb->buffer,
-					      &upload_offset, &upload_buffer);
+
+			ret = r600_upload_buffer(rctx->rupload_vb,
+						0, vb->buffer->width0,
+						rbuffer,
+						&upload_offset,
+						&rbuffer->r.bo_size,
+						&rbuffer->r.bo);
 			if (ret)
 				return ret;
-
-			pipe_resource_reference(&vb->buffer, NULL);
-			vb->buffer = upload_buffer;
+			rbuffer->uploaded = TRUE;
 			vb->buffer_offset = upload_offset;
 		}
 	}
diff --git a/src/gallium/drivers/r600/r600_opcodes.h b/src/gallium/drivers/r600/r600_opcodes.h
index 4f9b39a7fd..2ee0c83e5d 100644
--- a/src/gallium/drivers/r600/r600_opcodes.h
+++ b/src/gallium/drivers/r600/r600_opcodes.h
@@ -385,8 +385,13 @@
 #define     EG_V_SQ_CF_ALLOC_EXPORT_WORD1_SQ_CF_INST_MEM_EXPORT_COMBINED 0x0000005B
 #define     EG_V_SQ_CF_ALLOC_EXPORT_WORD1_SQ_CF_INST_MEM_RAT_COMBINED_CACHELESS  0x0000005C
 
-#define BC_INST(bc, x) ((bc)->chiprev == 2 ? EG_##x : x)
 
-#define CTX_INST(x) (ctx->bc->chiprev == 2 ? EG_##x : x)
+#define CHIPREV_R600      0
+#define CHIPREV_R700      1
+#define CHIPREV_EVERGREEN 2
+
+#define BC_INST(bc, x) ((bc)->chiprev == CHIPREV_EVERGREEN ? EG_##x : x)
+
+#define CTX_INST(x) (ctx->bc->chiprev == CHIPREV_EVERGREEN ? EG_##x : x)
 
 #endif
diff --git a/src/gallium/drivers/r600/r600_pipe.c b/src/gallium/drivers/r600/r600_pipe.c
index 2a113f0ea8..72988b946e 100644
--- a/src/gallium/drivers/r600/r600_pipe.c
+++ b/src/gallium/drivers/r600/r600_pipe.c
@@ -35,7 +35,6 @@
 #include <util/u_pack_color.h>
 #include <util/u_memory.h>
 #include <util/u_inlines.h>
-#include <util/u_upload_mgr.h>
 #include <pipebuffer/pb_buffer.h>
 #include "r600.h"
 #include "r600d.h"
@@ -59,9 +58,6 @@ static void r600_flush(struct pipe_context *ctx, unsigned flags,
 	if (!rctx->ctx.pm4_cdwords)
 		return;
 
-	u_upload_flush(rctx->upload_vb);
-	u_upload_flush(rctx->upload_ib);
-
 #if 0
 	sprintf(dname, "gallium-%08d.bof", dc);
 	if (dc < 20) {
@@ -71,6 +67,8 @@ static void r600_flush(struct pipe_context *ctx, unsigned flags,
 	dc++;
 #endif
 	r600_context_flush(&rctx->ctx);
+
+	r600_upload_flush(rctx->rupload_vb);
 }
 
 static void r600_destroy_context(struct pipe_context *context)
@@ -79,6 +77,8 @@ static void r600_destroy_context(struct pipe_context *context)
 
 	rctx->context.delete_depth_stencil_alpha_state(&rctx->context, rctx->custom_dsa_flush);
 
+	r600_end_vertex_translate(rctx);
+
 	r600_context_fini(&rctx->ctx);
 
 	util_blitter_destroy(rctx->blitter);
@@ -87,8 +87,7 @@ static void r600_destroy_context(struct pipe_context *context)
 		free(rctx->states[i]);
 	}
 
-	u_upload_destroy(rctx->upload_vb);
-	u_upload_destroy(rctx->upload_ib);
+	r600_upload_destroy(rctx->rupload_vb);
 
 	if (rctx->tran.translate_cache)
 		translate_cache_destroy(rctx->tran.translate_cache);
@@ -120,6 +119,7 @@ static struct pipe_context *r600_create_context(struct pipe_screen *screen, void
 	r600_init_blit_functions(rctx);
 	r600_init_query_functions(rctx);
 	r600_init_context_resource_functions(rctx);
+	r600_init_surface_functions(rctx);
 
 	switch (r600_get_family(rctx->radeon)) {
 	case CHIP_R600:
@@ -147,6 +147,7 @@ static struct pipe_context *r600_create_context(struct pipe_screen *screen, void
 	case CHIP_JUNIPER:
 	case CHIP_CYPRESS:
 	case CHIP_HEMLOCK:
+	case CHIP_PALM:
 		rctx->context.draw_vbo = evergreen_draw;
 		evergreen_init_state_functions(rctx);
 		if (evergreen_context_init(&rctx->ctx, rctx->radeon)) {
@@ -161,16 +162,8 @@ static struct pipe_context *r600_create_context(struct pipe_screen *screen, void
 		return NULL;
 	}
 
-	rctx->upload_ib = u_upload_create(&rctx->context, 32 * 1024, 16,
-					  PIPE_BIND_INDEX_BUFFER);
-	if (rctx->upload_ib == NULL) {
-		r600_destroy_context(&rctx->context);
-		return NULL;
-	}
-
-	rctx->upload_vb = u_upload_create(&rctx->context, 128 * 1024, 16,
-					  PIPE_BIND_VERTEX_BUFFER);
-	if (rctx->upload_vb == NULL) {
+	rctx->rupload_vb = r600_upload_create(rctx, 128 * 1024, 16);
+	if (rctx->rupload_vb == NULL) {
 		r600_destroy_context(&rctx->context);
 		return NULL;
 	}
@@ -186,7 +179,7 @@ static struct pipe_context *r600_create_context(struct pipe_screen *screen, void
 		FREE(rctx);
 		return NULL;
 	}
-	
+
 	rctx->vs_resource = CALLOC(R600_RESOURCE_ARRAY_SIZE, sizeof(struct r600_pipe_state));
 	if (!rctx->vs_resource) {
 		FREE(rctx);
@@ -238,6 +231,7 @@ static const char *r600_get_family_name(enum radeon_family family)
 	case CHIP_JUNIPER: return "AMD JUNIPER";
 	case CHIP_CYPRESS: return "AMD CYPRESS";
 	case CHIP_HEMLOCK: return "AMD HEMLOCK";
+	case CHIP_PALM: return "AMD PALM";
 	default: return "AMD unknown";
 	}
 }
@@ -252,6 +246,9 @@ static const char* r600_get_name(struct pipe_screen* pscreen)
 
 static int r600_get_param(struct pipe_screen* pscreen, enum pipe_cap param)
 {
+	struct r600_screen *rscreen = (struct r600_screen *)pscreen;
+	enum radeon_family family = r600_get_family(rscreen->radeon);
+
 	switch (param) {
 	/* Supported features (boolean caps). */
 	case PIPE_CAP_NPOT_TEXTURES:
@@ -284,7 +281,10 @@ static int r600_get_param(struct pipe_screen* pscreen, enum pipe_cap param)
 	case PIPE_CAP_MAX_TEXTURE_2D_LEVELS:
 	case PIPE_CAP_MAX_TEXTURE_3D_LEVELS:
 	case PIPE_CAP_MAX_TEXTURE_CUBE_LEVELS:
-		return 14;
+		if (family >= CHIP_CEDAR)
+			return 15;
+		else
+			return 14;
 	case PIPE_CAP_MAX_VERTEX_TEXTURE_UNITS:
 		/* FIXME allow this once infrastructure is there */
 		return 16;
@@ -313,12 +313,18 @@ static int r600_get_param(struct pipe_screen* pscreen, enum pipe_cap param)
 
 static float r600_get_paramf(struct pipe_screen* pscreen, enum pipe_cap param)
 {
+	struct r600_screen *rscreen = (struct r600_screen *)pscreen;
+	enum radeon_family family = r600_get_family(rscreen->radeon);
+
 	switch (param) {
 	case PIPE_CAP_MAX_LINE_WIDTH:
 	case PIPE_CAP_MAX_LINE_WIDTH_AA:
 	case PIPE_CAP_MAX_POINT_WIDTH:
 	case PIPE_CAP_MAX_POINT_WIDTH_AA:
-		return 8192.0f;
+		if (family >= CHIP_CEDAR)
+			return 16384.0f;
+		else
+			return 8192.0f;
 	case PIPE_CAP_MAX_TEXTURE_ANISOTROPY:
 		return 16.0f;
 	case PIPE_CAP_MAX_TEXTURE_LOD_BIAS:
@@ -375,6 +381,8 @@ static int r600_get_shader_param(struct pipe_screen* pscreen, unsigned shader, e
 	case PIPE_SHADER_CAP_INDIRECT_TEMP_ADDR:
 	case PIPE_SHADER_CAP_INDIRECT_CONST_ADDR:
 		return 1;
+	case PIPE_SHADER_CAP_SUBROUTINES:
+		return 0;
 	default:
 		return 0;
 	}
@@ -403,10 +411,10 @@ static boolean r600_is_format_supported(struct pipe_screen* screen,
 	}
 
 	if ((usage & (PIPE_BIND_RENDER_TARGET |
-                  PIPE_BIND_DISPLAY_TARGET |
-                  PIPE_BIND_SCANOUT |
-                  PIPE_BIND_SHARED)) &&
-	    r600_is_colorbuffer_format_supported(format)) {
+			PIPE_BIND_DISPLAY_TARGET |
+			PIPE_BIND_SCANOUT |
+			PIPE_BIND_SHARED)) &&
+			r600_is_colorbuffer_format_supported(format)) {
 		retval |= usage &
 			(PIPE_BIND_RENDER_TARGET |
 			 PIPE_BIND_DISPLAY_TARGET |
@@ -463,7 +471,6 @@ struct pipe_screen *r600_screen_create(struct radeon *radeon)
 	rscreen->screen.get_paramf = r600_get_paramf;
 	rscreen->screen.is_format_supported = r600_is_format_supported;
 	rscreen->screen.context_create = r600_create_context;
-	r600_init_screen_texture_functions(&rscreen->screen);
 	r600_init_screen_resource_functions(&rscreen->screen);
 
 	rscreen->tiling_info = r600_get_tiling_info(radeon);
diff --git a/src/gallium/drivers/r600/r600_pipe.h b/src/gallium/drivers/r600/r600_pipe.h
index 1c691f6b76..43dbee99b0 100644
--- a/src/gallium/drivers/r600/r600_pipe.h
+++ b/src/gallium/drivers/r600/r600_pipe.h
@@ -53,6 +53,8 @@ enum r600_pipe_state_id {
 	R600_PIPE_STATE_CONSTANT,
 	R600_PIPE_STATE_SAMPLER,
 	R600_PIPE_STATE_RESOURCE,
+	R600_PIPE_STATE_POLYGON_OFFSET,
+	R600_PIPE_STATE_FETCH_SHADER,
 	R600_PIPE_NSTATES
 };
 
@@ -86,13 +88,22 @@ struct r600_vertex_element
 	struct pipe_vertex_element	elements[PIPE_MAX_ATTRIBS];
 	enum pipe_format		hw_format[PIPE_MAX_ATTRIBS];
 	unsigned			hw_format_size[PIPE_MAX_ATTRIBS];
-	boolean incompatible_layout;
+	boolean				incompatible_layout;
+	struct r600_bo			*fetch_shader;
+	unsigned			fs_size;
+	struct r600_pipe_state		rstate;
+	/* if offset is to big for fetch instructio we need to alterate
+	 * offset of vertex buffer, record here the offset need to add
+	 */
+	unsigned			vbuffer_need_offset;
+	unsigned			vbuffer_offset[PIPE_MAX_ATTRIBS];
 };
 
 struct r600_pipe_shader {
 	struct r600_shader		shader;
 	struct r600_pipe_state		rstate;
 	struct r600_bo			*bo;
+	struct r600_bo			*bo_fetch;
 	struct r600_vertex_element	vertex_elements;
 };
 
@@ -100,25 +111,28 @@ struct r600_pipe_shader {
 #define NUM_TEX_UNITS 16
 
 struct r600_textures_info {
-	struct r600_pipe_sampler_view   *views[NUM_TEX_UNITS];
-	unsigned                        n_views;
+	struct r600_pipe_sampler_view	*views[NUM_TEX_UNITS];
+	unsigned			n_views;
 	void				*samplers[NUM_TEX_UNITS];
-	unsigned                        n_samplers;
+	unsigned			n_samplers;
 };
 
+/* vertex buffer translation context, used to translate vertex input that
+ * hw doesn't natively support, so far only FLOAT64 is unsupported.
+ */
 struct r600_translate_context {
 	/* Translate cache for incompatible vertex offset/stride/format fallback. */
-	struct translate_cache *translate_cache;
-
+	struct translate_cache		*translate_cache;
 	/* The vertex buffer slot containing the translated buffer. */
-	unsigned vb_slot;
-	/* Saved and new vertex element state. */
-	void *saved_velems, *new_velems;
+	unsigned			vb_slot;
+	void				*new_velems;
 };
 
 #define R600_CONSTANT_ARRAY_SIZE 256
 #define R600_RESOURCE_ARRAY_SIZE 160
 
+struct r600_upload;
+
 struct r600_pipe_context {
 	struct pipe_context		context;
 	struct blitter_context		*blitter;
@@ -139,6 +153,7 @@ struct r600_pipe_context {
 	struct pipe_stencil_ref		stencil_ref;
 	struct pipe_viewport_state	viewport;
 	struct pipe_clip_state		clip;
+	unsigned			nvs_resource;
 	struct r600_pipe_state		*vs_resource;
 	struct r600_pipe_state		*ps_resource;
 	struct r600_pipe_state		config;
@@ -150,14 +165,11 @@ struct r600_pipe_context {
 	/* shader information */
 	unsigned			sprite_coord_enable;
 	bool				flatshade;
-	struct u_upload_mgr		*upload_vb;
-	struct u_upload_mgr		*upload_ib;
+	struct r600_upload		*rupload_vb;
 	unsigned			any_user_vbs;
-	struct r600_textures_info       ps_samplers;
-
-	unsigned vb_max_index;
-	struct r600_translate_context tran;
-
+	struct r600_textures_info	ps_samplers;
+	unsigned			vb_max_index;
+	struct r600_translate_context	tran;
 };
 
 struct r600_drawl {
@@ -180,6 +192,8 @@ void evergreen_draw(struct pipe_context *ctx, const struct pipe_draw_info *info)
 void evergreen_pipe_shader_ps(struct pipe_context *ctx, struct r600_pipe_shader *shader);
 void evergreen_pipe_shader_vs(struct pipe_context *ctx, struct r600_pipe_shader *shader);
 void *evergreen_create_db_flush_dsa(struct r600_pipe_context *rctx);
+void evergreen_polygon_offset_update(struct r600_pipe_context *rctx);
+void evergreen_vertex_buffer_update(struct r600_pipe_context *rctx);
 
 /* r600_blit.c */
 void r600_init_blit_functions(struct r600_pipe_context *rctx);
@@ -193,7 +207,7 @@ struct pipe_resource *r600_user_buffer_create(struct pipe_screen *screen,
 					      unsigned bind);
 unsigned r600_buffer_is_referenced_by_cs(struct pipe_context *context,
 					 struct pipe_resource *buf,
-					 unsigned face, unsigned level);
+					 unsigned level, int layer);
 struct pipe_resource *r600_buffer_from_handle(struct pipe_screen *screen,
 					      struct winsys_handle *whandle);
 int r600_upload_index_buffer(struct r600_pipe_context *rctx, struct r600_drawl *draw);
@@ -206,7 +220,7 @@ void r600_init_query_functions(struct r600_pipe_context *rctx);
 void r600_init_context_resource_functions(struct r600_pipe_context *r600);
 
 /* r600_shader.c */
-int r600_pipe_shader_update(struct pipe_context *ctx, struct r600_pipe_shader *shader);
+int r600_pipe_shader(struct pipe_context *ctx, struct r600_pipe_shader *shader);
 int r600_pipe_shader_create(struct pipe_context *ctx, struct r600_pipe_shader *shader, const struct tgsi_token *tokens);
 void r600_pipe_shader_destroy(struct pipe_context *ctx, struct r600_pipe_shader *shader);
 int r600_find_vs_semantic_index(struct r600_shader *vs,
@@ -217,14 +231,20 @@ void r600_init_state_functions(struct r600_pipe_context *rctx);
 void r600_draw_vbo(struct pipe_context *ctx, const struct pipe_draw_info *info);
 void r600_init_config(struct r600_pipe_context *rctx);
 void *r600_create_db_flush_dsa(struct r600_pipe_context *rctx);
+void r600_polygon_offset_update(struct r600_pipe_context *rctx);
+void r600_vertex_buffer_update(struct r600_pipe_context *rctx);
+
 /* r600_helper.h */
 int r600_conv_pipe_prim(unsigned pprim, unsigned *prim);
 
 /* r600_texture.c */
 void r600_init_screen_texture_functions(struct pipe_screen *screen);
+void r600_init_surface_functions(struct r600_pipe_context *r600);
 uint32_t r600_translate_texformat(enum pipe_format format,
 				  const unsigned char *swizzle_view, 
 				  uint32_t *word4_p, uint32_t *yuv_format_p);
+unsigned r600_texture_get_offset(struct r600_resource_texture *rtex,
+					unsigned level, unsigned layer);
 
 /* r600_translate.c */
 void r600_begin_vertex_translate(struct r600_pipe_context *rctx);
@@ -251,13 +271,13 @@ void r600_sampler_view_destroy(struct pipe_context *ctx,
 void r600_bind_state(struct pipe_context *ctx, void *state);
 void r600_delete_state(struct pipe_context *ctx, void *state);
 void r600_bind_vertex_elements(struct pipe_context *ctx, void *state);
-
 void *r600_create_shader_state(struct pipe_context *ctx,
 			       const struct pipe_shader_state *state);
 void r600_bind_ps_shader(struct pipe_context *ctx, void *state);
 void r600_bind_vs_shader(struct pipe_context *ctx, void *state);
 void r600_delete_ps_shader(struct pipe_context *ctx, void *state);
 void r600_delete_vs_shader(struct pipe_context *ctx, void *state);
+
 /*
  * common helpers
  */
diff --git a/src/gallium/drivers/r600/r600_resource.h b/src/gallium/drivers/r600/r600_resource.h
index 7a2d1f4412..8ca2769920 100644
--- a/src/gallium/drivers/r600/r600_resource.h
+++ b/src/gallium/drivers/r600/r600_resource.h
@@ -46,6 +46,7 @@ struct r600_resource {
 	struct u_resource		base;
 	struct r600_bo			*bo;
 	u32				size;
+	unsigned			bo_size;
 };
 
 struct r600_resource_texture {
@@ -61,7 +62,21 @@ struct r600_resource_texture {
 	unsigned			tile_type;
 	unsigned			depth;
 	unsigned			dirty;
-	struct r600_resource_texture    *flushed_depth_texture;
+	struct r600_resource_texture	*flushed_depth_texture;
+};
+
+#define R600_BUFFER_MAGIC 0xabcd1600
+
+struct r600_resource_buffer {
+	struct r600_resource		r;
+	uint32_t			magic;
+	void				*user_buffer;
+	bool				uploaded;
+};
+
+struct r600_surface {
+	struct pipe_surface		base;
+	unsigned			aligned_height;
 };
 
 void r600_init_screen_resource_functions(struct pipe_screen *screen);
@@ -73,46 +88,30 @@ struct pipe_resource *r600_texture_from_handle(struct pipe_screen *screen,
 						const struct pipe_resource *base,
 						struct winsys_handle *whandle);
 
-#define R600_BUFFER_MAGIC 0xabcd1600
-#define R600_BUFFER_MAX_RANGES 32
-
-struct r600_buffer_range {
-	uint32_t start;
-	uint32_t end;
-};
-
-struct r600_resource_buffer {
-	struct r600_resource r;
-	uint32_t magic;
-	void *user_buffer;
-	struct r600_buffer_range ranges[R600_BUFFER_MAX_RANGES];
-	unsigned num_ranges;
-};
-
 /* r600_buffer */
 static INLINE struct r600_resource_buffer *r600_buffer(struct pipe_resource *buffer)
 {
 	if (buffer) {
 		assert(((struct r600_resource_buffer *)buffer)->magic == R600_BUFFER_MAGIC);
 		return (struct r600_resource_buffer *)buffer;
-    }
-    return NULL;
+	}
+	return NULL;
 }
 
 static INLINE boolean r600_buffer_is_user_buffer(struct pipe_resource *buffer)
 {
-    return r600_buffer(buffer)->user_buffer ? TRUE : FALSE;
+	if (r600_buffer(buffer)->uploaded)
+		return FALSE;
+	return r600_buffer(buffer)->user_buffer ? TRUE : FALSE;
 }
 
-int r600_texture_depth_flush(struct pipe_context *ctx,
-			     struct pipe_resource *texture);
-
-extern int (*r600_blit_uncompress_depth_ptr)(struct pipe_context *ctx, struct r600_resource_texture *texture);
+int r600_texture_depth_flush(struct pipe_context *ctx, struct pipe_resource *texture);
+int (*r600_blit_uncompress_depth_ptr)(struct pipe_context *ctx, struct r600_resource_texture *texture);
 
 /* r600_texture.c texture transfer functions. */
 struct pipe_transfer* r600_texture_get_transfer(struct pipe_context *ctx,
 						struct pipe_resource *texture,
-						struct pipe_subresource sr,
+						unsigned level,
 						unsigned usage,
 						const struct pipe_box *box);
 void r600_texture_transfer_destroy(struct pipe_context *ctx,
@@ -122,9 +121,15 @@ void* r600_texture_transfer_map(struct pipe_context *ctx,
 void r600_texture_transfer_unmap(struct pipe_context *ctx,
 				 struct pipe_transfer* transfer);
 
-struct r600_surface {
-	struct pipe_surface base;
-	unsigned aligned_height;
-};
+struct r600_pipe_context;
+struct r600_upload *r600_upload_create(struct r600_pipe_context *rctx,
+					unsigned default_size,
+					unsigned alignment);
+void r600_upload_flush(struct r600_upload *upload);
+void r600_upload_destroy(struct r600_upload *upload);
+int r600_upload_buffer(struct r600_upload *upload, unsigned offset,
+			unsigned size, struct r600_resource_buffer *in_buffer,
+			unsigned *out_offset, unsigned *out_size,
+			struct r600_bo **out_buffer);
 
 #endif
diff --git a/src/gallium/drivers/r600/r600_shader.c b/src/gallium/drivers/r600/r600_shader.c
index f6153c0e80..9c7b7f0a57 100644
--- a/src/gallium/drivers/r600/r600_shader.c
+++ b/src/gallium/drivers/r600/r600_shader.c
@@ -44,6 +44,9 @@ static void r600_pipe_shader_vs(struct pipe_context *ctx, struct r600_pipe_shade
 	rstate->nregs = 0;
 
 	/* so far never got proper semantic id from tgsi */
+	/* FIXME better to move this in config things so they get emited
+	 * only one time per cs
+	 */
 	for (i = 0; i < 10; i++) {
 		spi_vs_out_id[i] = 0;
 	}
@@ -67,20 +70,11 @@ static void r600_pipe_shader_vs(struct pipe_context *ctx, struct r600_pipe_shade
 			S_028868_STACK_SIZE(rshader->bc.nstack),
 			0xFFFFFFFF, NULL);
 	r600_pipe_state_add_reg(rstate,
-			R_0288A4_SQ_PGM_RESOURCES_FS,
-			0x00000000, 0xFFFFFFFF, NULL);
-	r600_pipe_state_add_reg(rstate,
 			R_0288D0_SQ_PGM_CF_OFFSET_VS,
 			0x00000000, 0xFFFFFFFF, NULL);
 	r600_pipe_state_add_reg(rstate,
-			R_0288DC_SQ_PGM_CF_OFFSET_FS,
-			0x00000000, 0xFFFFFFFF, NULL);
-	r600_pipe_state_add_reg(rstate,
 			R_028858_SQ_PGM_START_VS,
 			r600_bo_offset(shader->bo) >> 8, 0xFFFFFFFF, shader->bo);
-	r600_pipe_state_add_reg(rstate,
-			R_028894_SQ_PGM_START_FS,
-			r600_bo_offset(shader->bo) >> 8, 0xFFFFFFFF, shader->bo);
 
 	r600_pipe_state_add_reg(rstate,
 				R_03E200_SQ_LOOP_CONST_0 + (32 * 4), 0x01000FFF,
@@ -104,37 +98,20 @@ int r600_find_vs_semantic_index(struct r600_shader *vs,
 
 static void r600_pipe_shader_ps(struct pipe_context *ctx, struct r600_pipe_shader *shader)
 {
-	struct r600_pipe_context *rctx = (struct r600_pipe_context *)ctx;
 	struct r600_pipe_state *rstate = &shader->rstate;
 	struct r600_shader *rshader = &shader->shader;
-	unsigned i, tmp, exports_ps, num_cout, spi_ps_in_control_0, spi_input_z, spi_ps_in_control_1;
+	unsigned i, exports_ps, num_cout, spi_ps_in_control_0, spi_input_z, spi_ps_in_control_1;
 	int pos_index = -1, face_index = -1;
 
-	/* clear previous register */
 	rstate->nregs = 0;
 
 	for (i = 0; i < rshader->ninput; i++) {
-		tmp = S_028644_SEMANTIC(r600_find_vs_semantic_index(&rctx->vs_shader->shader, rshader, i));
-		if (rshader->input[i].centroid)
-			tmp |= S_028644_SEL_CENTROID(1);
-		if (rshader->input[i].interpolate == TGSI_INTERPOLATE_LINEAR)
-			tmp |= S_028644_SEL_LINEAR(1);
-
 		if (rshader->input[i].name == TGSI_SEMANTIC_POSITION)
 			pos_index = i;
-		if (rshader->input[i].name == TGSI_SEMANTIC_COLOR ||
-		    rshader->input[i].name == TGSI_SEMANTIC_BCOLOR ||
-		    rshader->input[i].name == TGSI_SEMANTIC_POSITION) {
-			tmp |= S_028644_FLAT_SHADE(rshader->flat_shade);
-		}
 		if (rshader->input[i].name == TGSI_SEMANTIC_FACE)
 			face_index = i;
-		if (rshader->input[i].name == TGSI_SEMANTIC_GENERIC &&
-			rctx->sprite_coord_enable & (1 << rshader->input[i].sid)) {
-			tmp |= S_028644_PT_SPRITE_TEX(1);
-		}
-		r600_pipe_state_add_reg(rstate, R_028644_SPI_PS_INPUT_CNTL_0 + i * 4, tmp, 0xFFFFFFFF, NULL);
 	}
+
 	for (i = 0; i < rshader->noutput; i++) {
 		if (rshader->output[i].name == TGSI_SEMANTIC_POSITION)
 			r600_pipe_state_add_reg(rstate,
@@ -210,7 +187,7 @@ static void r600_pipe_shader_ps(struct pipe_context *ctx, struct r600_pipe_shade
 				0xFFFFFFFF, NULL);
 }
 
-static int r600_pipe_shader(struct pipe_context *ctx, struct r600_pipe_shader *shader)
+int r600_pipe_shader(struct pipe_context *ctx, struct r600_pipe_shader *shader)
 {
 	struct r600_pipe_context *rctx = (struct r600_pipe_context *)ctx;
 	struct r600_shader *rshader = &shader->shader;
@@ -227,7 +204,6 @@ static int r600_pipe_shader(struct pipe_context *ctx, struct r600_pipe_shader *s
 		r600_bo_unmap(rctx->radeon, shader->bo);
 	}
 	/* build state */
-	rshader->flat_shade = rctx->flatshade;
 	switch (rshader->processor_type) {
 	case TGSI_PROCESSOR_VERTEX:
 		if (rshader->family >= CHIP_CEDAR) {
@@ -246,75 +222,9 @@ static int r600_pipe_shader(struct pipe_context *ctx, struct r600_pipe_shader *s
 	default:
 		return -EINVAL;
 	}
-	r600_context_pipe_state_set(&rctx->ctx, &shader->rstate);
 	return 0;
 }
 
-static int r600_shader_update(struct pipe_context *ctx, struct r600_pipe_shader *rshader)
-{
-	struct r600_pipe_context *rctx = (struct r600_pipe_context *)ctx;
-	struct r600_shader *shader = &rshader->shader;
-	const struct util_format_description *desc;
-	enum pipe_format resource_format[160];
-	unsigned i, nresources = 0;
-	struct r600_bc *bc = &shader->bc;
-	struct r600_bc_cf *cf;
-	struct r600_bc_vtx *vtx;
-
-	if (shader->processor_type != TGSI_PROCESSOR_VERTEX)
-		return 0;
-	/* doing a full memcmp fell over the refcount */
-	if ((rshader->vertex_elements.count == rctx->vertex_elements->count) &&
-	    (!memcmp(&rshader->vertex_elements.elements, &rctx->vertex_elements->elements, 32 * sizeof(struct pipe_vertex_element)))) {
-		return 0;
-	}
-	rshader->vertex_elements = *rctx->vertex_elements;
-	for (i = 0; i < rctx->vertex_elements->count; i++) {
-		resource_format[nresources++] = rctx->vertex_elements->hw_format[i];
-	}
-	r600_bo_reference(rctx->radeon, &rshader->bo, NULL);
-	LIST_FOR_EACH_ENTRY(cf, &bc->cf, list) {
-		switch (cf->inst) {
-		case V_SQ_CF_WORD1_SQ_CF_INST_VTX:
-		case V_SQ_CF_WORD1_SQ_CF_INST_VTX_TC:
-			LIST_FOR_EACH_ENTRY(vtx, &cf->vtx, list) {
-				desc = util_format_description(resource_format[vtx->buffer_id]);
-				if (desc == NULL) {
-					R600_ERR("unknown format %d\n", resource_format[vtx->buffer_id]);
-					return -EINVAL;
-				}
-				vtx->dst_sel_x = desc->swizzle[0];
-				vtx->dst_sel_y = desc->swizzle[1];
-				vtx->dst_sel_z = desc->swizzle[2];
-				vtx->dst_sel_w = desc->swizzle[3];
-			}
-			break;
-		default:
-			break;
-		}
-	}
-	return r600_bc_build(&shader->bc);
-}
-
-int r600_pipe_shader_update(struct pipe_context *ctx, struct r600_pipe_shader *shader)
-{
-	struct r600_pipe_context *rctx = (struct r600_pipe_context *)ctx;
-	int r;
-
-	if (shader == NULL)
-		return -EINVAL;
-	/* there should be enough input */
-	if (rctx->vertex_elements->count < shader->shader.bc.nresource) {
-		R600_ERR("%d resources provided, expecting %d\n",
-			rctx->vertex_elements->count, shader->shader.bc.nresource);
-		return -EINVAL;
-	}
-	r = r600_shader_update(ctx, shader);
-	if (r)
-		return r;
-	return r600_pipe_shader(ctx, shader);
-}
-
 int r600_shader_from_tgsi(const struct tgsi_token *tokens, struct r600_shader *shader);
 int r600_pipe_shader_create(struct pipe_context *ctx, struct r600_pipe_shader *shader, const struct tgsi_token *tokens)
 {
@@ -334,20 +244,17 @@ int r600_pipe_shader_create(struct pipe_context *ctx, struct r600_pipe_shader *s
 		R600_ERR("building bytecode failed !\n");
 		return r;
 	}
+//r600_bc_dump(&shader->shader.bc);
 //fprintf(stderr, "______________________________________________________________\n");
-	return 0;
+	return r600_pipe_shader(ctx, shader);
 }
 
-void
-r600_pipe_shader_destroy(struct pipe_context *ctx, struct r600_pipe_shader *shader)
+void r600_pipe_shader_destroy(struct pipe_context *ctx, struct r600_pipe_shader *shader)
 {
 	struct r600_pipe_context *rctx = (struct r600_pipe_context *)ctx;
 
 	r600_bo_reference(rctx->radeon, &shader->bo, NULL);
-
 	r600_bc_clear(&shader->shader.bc);
-
-	/* FIXME: is there more stuff to free? */
 }
 
 /*
@@ -444,7 +351,7 @@ static int evergreen_interp_alu(struct r600_shader_ctx *ctx, int input)
 		if (ctx->shader->input[input].centroid)
 			ij_index++;
 	}
-		
+
 	/* work out gpr and base_chan from index */
 	gpr = ij_index / 2;
 	base_chan = (2 * (ij_index % 2)) + 1;
@@ -477,15 +384,13 @@ static int evergreen_interp_alu(struct r600_shader_ctx *ctx, int input)
 			return r;
 	}
 	return 0;
-}	
-		
-			
+}
+
+
 static int tgsi_declaration(struct r600_shader_ctx *ctx)
 {
 	struct tgsi_full_declaration *d = &ctx->parse.FullToken.FullDeclaration;
-	struct r600_bc_vtx vtx;
 	unsigned i;
-	int r;
 
 	switch (d->Declaration.File) {
 	case TGSI_FILE_INPUT:
@@ -495,27 +400,7 @@ static int tgsi_declaration(struct r600_shader_ctx *ctx)
 		ctx->shader->input[i].interpolate = d->Declaration.Interpolate;
 		ctx->shader->input[i].centroid = d->Declaration.Centroid;
 		ctx->shader->input[i].gpr = ctx->file_offset[TGSI_FILE_INPUT] + i;
-		if (ctx->type == TGSI_PROCESSOR_VERTEX) {
-			/* turn input into fetch */
-			memset(&vtx, 0, sizeof(struct r600_bc_vtx));
-			vtx.inst = 0;
-			vtx.fetch_type = 0;
-			vtx.buffer_id = i;
-			/* register containing the index into the buffer */
-			vtx.src_gpr = 0;
-			vtx.src_sel_x = 0;
-			vtx.mega_fetch_count = 0x1F;
-			vtx.dst_gpr = ctx->shader->input[i].gpr;
-			vtx.dst_sel_x = 0;
-			vtx.dst_sel_y = 1;
-			vtx.dst_sel_z = 2;
-			vtx.dst_sel_w = 3;
-			vtx.use_const_fields = 1;
-			r = r600_bc_add_vtx(ctx->bc, &vtx);
-			if (r)
-				return r;
-		}
-		if (ctx->type == TGSI_PROCESSOR_FRAGMENT && ctx->bc->chiprev == 2) {
+		if (ctx->type == TGSI_PROCESSOR_FRAGMENT && ctx->bc->chiprev == CHIPREV_EVERGREEN) {
 			/* turn input into interpolate on EG */
 			if (ctx->shader->input[i].name != TGSI_SEMANTIC_POSITION) {
 				if (ctx->shader->input[i].interpolate > 0) {
@@ -549,7 +434,7 @@ static int r600_get_temp(struct r600_shader_ctx *ctx)
 	return ctx->temp_reg + ctx->max_driver_temp_used++;
 }
 
-/* 
+/*
  * for evergreen we need to scan the shader to find the number of GPRs we need to
  * reserve for interpolation.
  *
@@ -615,6 +500,7 @@ int r600_shader_from_tgsi(const struct tgsi_token *tokens, struct r600_shader *s
 	tgsi_parse_init(&ctx.parse, tokens);
 	ctx.type = ctx.parse.FullHeader.Processor.Processor;
 	shader->processor_type = ctx.type;
+	ctx.bc->type = shader->processor_type;
 
 	/* register allocations */
 	/* Values [0,127] correspond to GPR[0..127].
@@ -640,8 +526,13 @@ int r600_shader_from_tgsi(const struct tgsi_token *tokens, struct r600_shader *s
 	}
 	if (ctx.type == TGSI_PROCESSOR_VERTEX) {
 		ctx.file_offset[TGSI_FILE_INPUT] = 1;
+		if (ctx.bc->chiprev == CHIPREV_EVERGREEN) {
+			r600_bc_add_cfinst(ctx.bc, EG_V_SQ_CF_WORD1_SQ_CF_INST_CALL_FS);
+		} else {
+			r600_bc_add_cfinst(ctx.bc, V_SQ_CF_WORD1_SQ_CF_INST_CALL_FS);
+		}
 	}
-	if (ctx.type == TGSI_PROCESSOR_FRAGMENT && ctx.bc->chiprev == 2) {
+	if (ctx.type == TGSI_PROCESSOR_FRAGMENT && ctx.bc->chiprev == CHIPREV_EVERGREEN) {
 		ctx.file_offset[TGSI_FILE_INPUT] = evergreen_gpr_count(&ctx);
 	}
 	ctx.file_offset[TGSI_FILE_OUTPUT] = ctx.file_offset[TGSI_FILE_INPUT] +
@@ -687,7 +578,7 @@ int r600_shader_from_tgsi(const struct tgsi_token *tokens, struct r600_shader *s
 			/* reserve first tmp for everyone */
 			r600_get_temp(&ctx);
 			opcode = ctx.parse.FullToken.FullInstruction.Instruction.Opcode;
-			if (ctx.bc->chiprev == 2)
+			if (ctx.bc->chiprev == CHIPREV_EVERGREEN)
 				ctx.inst_info = &eg_shader_tgsi_instruction[opcode];
 			else
 				ctx.inst_info = &r600_shader_tgsi_instruction[opcode];
@@ -1001,7 +892,7 @@ static int tgsi_op2_s(struct r600_shader_ctx *ctx, int swap)
 		r = tgsi_dst(ctx, &inst->Dst[0], i, &alu.dst);
 		if (r)
 			return r;
-		
+
 		alu.inst = ctx->inst_info->r600_opcode;
 		if (!swap) {
 			for (j = 0; j < inst->Instruction.NumSrcRegs; j++) {
@@ -1046,7 +937,7 @@ static int tgsi_op2_swap(struct r600_shader_ctx *ctx)
 	return tgsi_op2_s(ctx, 1);
 }
 
-/* 
+/*
  * r600 - trunc to -PI..PI range
  * r700 - normalize by dividing by 2PI
  * see fdo bug 27901
@@ -1058,7 +949,7 @@ static int tgsi_setup_trig(struct r600_shader_ctx *ctx,
 	int r;
 	uint32_t lit_vals[4];
 	struct r600_bc_alu alu;
-	
+
 	memset(lit_vals, 0, 4*4);
 	r = tgsi_split_constant(ctx, r600_src);
 	if (r)
@@ -1067,10 +958,6 @@ static int tgsi_setup_trig(struct r600_shader_ctx *ctx,
 	if (r)
 		return r;
 
-	r = tgsi_split_literal_constant(ctx, r600_src);
-	if (r)
-		return r;
-
 	lit_vals[0] = fui(1.0 /(3.1415926535 * 2));
 	lit_vals[1] = fui(0.5f);
 
@@ -1084,7 +971,7 @@ static int tgsi_setup_trig(struct r600_shader_ctx *ctx,
 
 	alu.src[0] = r600_src[0];
 	alu.src[0].chan = tgsi_chan(&inst->Src[0], 0);
-		
+
 	alu.src[1].sel = V_SQ_ALU_SRC_LITERAL;
 	alu.src[1].chan = 0;
 	alu.src[2].sel = V_SQ_ALU_SRC_LITERAL;
@@ -1099,7 +986,7 @@ static int tgsi_setup_trig(struct r600_shader_ctx *ctx,
 
 	memset(&alu, 0, sizeof(struct r600_bc_alu));
 	alu.inst = CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_FRACT);
-		
+
 	alu.dst.chan = 0;
 	alu.dst.sel = ctx->temp_reg;
 	alu.dst.write = 1;
@@ -1111,7 +998,7 @@ static int tgsi_setup_trig(struct r600_shader_ctx *ctx,
 	if (r)
 		return r;
 
-	if (ctx->bc->chiprev == 0) {
+	if (ctx->bc->chiprev == CHIPREV_R600) {
 		lit_vals[0] = fui(3.1415926535897f * 2.0f);
 		lit_vals[1] = fui(-3.1415926535897f);
 	} else {
@@ -1129,7 +1016,7 @@ static int tgsi_setup_trig(struct r600_shader_ctx *ctx,
 
 	alu.src[0].sel = ctx->temp_reg;
 	alu.src[0].chan = 0;
-		
+
 	alu.src[1].sel = V_SQ_ALU_SRC_LITERAL;
 	alu.src[1].chan = 0;
 	alu.src[2].sel = V_SQ_ALU_SRC_LITERAL;
@@ -1564,7 +1451,7 @@ static int tgsi_pow(struct r600_shader_ctx *ctx)
 		return r;
 	/* b * LOG2(a) */
 	memset(&alu, 0, sizeof(struct r600_bc_alu));
-	alu.inst = CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MUL_IEEE);
+	alu.inst = CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MUL);
 	r = tgsi_src(ctx, &inst->Src[1], &alu.src[0]);
 	if (r)
 		return r;
@@ -1908,10 +1795,10 @@ static int tgsi_tex(struct r600_shader_ctx *ctx)
 		r = r600_bc_add_alu(ctx->bc, &alu);
 		if (r)
 			return r;
-		
+
 		/* MULADD R0.x,  R0.x,  PS1,  (0x3FC00000, 1.5f).x
 		 * MULADD R0.y,  R0.y,  PS1,  (0x3FC00000, 1.5f).x
-		 * muladd has no writemask, have to use another temp 
+		 * muladd has no writemask, have to use another temp
 		 */
 		memset(&alu, 0, sizeof(struct r600_bc_alu));
 		alu.inst = CTX_INST(V_SQ_ALU_WORD1_OP3_SQ_OP3_INST_MULADD);
@@ -1921,7 +1808,7 @@ static int tgsi_tex(struct r600_shader_ctx *ctx)
 		alu.src[0].chan = 0;
 		alu.src[1].sel = ctx->temp_reg;
 		alu.src[1].chan = 2;
-		
+
 		alu.src[2].sel = V_SQ_ALU_SRC_LITERAL;
 		alu.src[2].chan = 0;
 
@@ -1941,7 +1828,7 @@ static int tgsi_tex(struct r600_shader_ctx *ctx)
 		alu.src[0].chan = 1;
 		alu.src[1].sel = ctx->temp_reg;
 		alu.src[1].chan = 2;
-		
+
 		alu.src[2].sel = V_SQ_ALU_SRC_LITERAL;
 		alu.src[2].chan = 0;
 
@@ -1980,7 +1867,7 @@ static int tgsi_tex(struct r600_shader_ctx *ctx)
 		}
 		src_gpr = ctx->temp_reg;
 	}
-	
+
 	opcode = ctx->inst_info->r600_opcode;
 	if (opcode == SQ_TEX_INST_SAMPLE &&
 	    (inst->Texture.Texture == TGSI_TEXTURE_SHADOW1D || inst->Texture.Texture == TGSI_TEXTURE_SHADOW2D))
@@ -1990,8 +1877,6 @@ static int tgsi_tex(struct r600_shader_ctx *ctx)
 	tex.inst = opcode;
 	tex.sampler_id = ctx->file_offset[inst->Src[1].Register.File] + inst->Src[1].Register.Index;
 	tex.resource_id = tex.sampler_id;
-	if (ctx->shader->processor_type == TGSI_PROCESSOR_VERTEX)
-		tex.resource_id += PIPE_MAX_ATTRIBS;
 	tex.src_gpr = src_gpr;
 	tex.dst_gpr = ctx->file_offset[inst->Dst[0].Register.File] + inst->Dst[0].Register.Index;
 	tex.dst_sel_x = (inst->Dst[0].Register.WriteMask & 1) ? 0 : 7;
@@ -2026,7 +1911,6 @@ static int tgsi_tex(struct r600_shader_ctx *ctx)
 
 	/* add shadow ambient support  - gallium doesn't do it yet */
 	return 0;
-	
 }
 
 static int tgsi_lrp(struct r600_shader_ctx *ctx)
@@ -2156,7 +2040,7 @@ static int tgsi_cmp(struct r600_shader_ctx *ctx)
 		r = r600_bc_add_alu(ctx->bc, &alu);
 		if (r)
 			return r;
-	}       
+	}
 	if (use_temp)
 		return tgsi_helper_copy(ctx, inst);
 	return 0;
@@ -2342,7 +2226,7 @@ static int tgsi_exp(struct r600_shader_ctx *ctx)
 		if (r)
 			return r;
 	}
-		
+
 	/* result.y = tmp - floor(tmp); */
 	if ((inst->Dst[0].Register.WriteMask >> 1) & 1) {
 		memset(&alu, 0, sizeof(struct r600_bc_alu));
@@ -2621,16 +2505,25 @@ static int tgsi_log(struct r600_shader_ctx *ctx)
 	return tgsi_helper_copy(ctx, inst);
 }
 
-/* r6/7 only for now */
 static int tgsi_eg_arl(struct r600_shader_ctx *ctx)
 {
 	struct tgsi_full_instruction *inst = &ctx->parse.FullToken.FullInstruction;
 	struct r600_bc_alu alu;
 	int r;
-	
 	memset(&alu, 0, sizeof(struct r600_bc_alu));
 
-	alu.inst = EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_FLT_TO_INT_FLOOR;
+	switch (inst->Instruction.Opcode) {
+	case TGSI_OPCODE_ARL:
+		alu.inst = EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_FLT_TO_INT_FLOOR;
+		break;
+	case TGSI_OPCODE_ARR:
+		alu.inst = EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_FLT_TO_INT;
+		break;
+	default:
+		assert(0);
+		return -1;
+	}
+
 	r = tgsi_src(ctx, &inst->Src[0], &alu.src[0]);
 	if (r)
 		return r;
@@ -2663,18 +2556,18 @@ static int tgsi_r600_arl(struct r600_shader_ctx *ctx)
 	int r;
 	memset(&alu, 0, sizeof(struct r600_bc_alu));
 
-        switch (inst->Instruction.Opcode) {
-        case TGSI_OPCODE_ARL:
-           alu.inst = V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MOVA_FLOOR;
-           break;
-        case TGSI_OPCODE_ARR:
-           alu.inst = V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MOVA;
-           break;
-        default:
-           assert(0);
-           return -1;
-        }
-        
+	switch (inst->Instruction.Opcode) {
+	case TGSI_OPCODE_ARL:
+		alu.inst = V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MOVA_FLOOR;
+		break;
+	case TGSI_OPCODE_ARR:
+		alu.inst = V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MOVA;
+		break;
+	default:
+		assert(0);
+		return -1;
+	}
+
 
 	r = tgsi_src(ctx, &inst->Src[0], &alu.src[0]);
 	if (r)
@@ -2703,8 +2596,8 @@ static int tgsi_opdst(struct r600_shader_ctx *ctx)
 		r = tgsi_dst(ctx, &inst->Dst[0], i, &alu.dst);
 		if (r)
 			return r;
-	
-	        if (i == 0 || i == 3) {
+
+		if (i == 0 || i == 3) {
 			alu.src[0].sel = V_SQ_ALU_SRC_1;
 		} else {
 			r = tgsi_src(ctx, &inst->Src[0], &alu.src[0]);
@@ -2750,7 +2643,7 @@ static int emit_logic_pred(struct r600_shader_ctx *ctx, int opcode)
 	alu.src[0].chan = tgsi_chan(&inst->Src[0], 0);
 	alu.src[1].sel = V_SQ_ALU_SRC_0;
 	alu.src[1].chan = 0;
-	
+
 	alu.last = 1;
 
 	r = r600_bc_add_alu_type(ctx->bc, &alu, CTX_INST(V_SQ_CF_ALU_WORD1_SQ_CF_INST_ALU_PUSH_BEFORE));
@@ -2763,6 +2656,7 @@ static int pops(struct r600_shader_ctx *ctx, int pops)
 {
 	r600_bc_add_cfinst(ctx->bc, CTX_INST(V_SQ_CF_WORD1_SQ_CF_INST_POP));
 	ctx->bc->cf_last->pop_count = pops;
+	ctx->bc->cf_last->cf_addr = ctx->bc->cf_last->id + 2;
 	return 0;
 }
 
@@ -2804,7 +2698,7 @@ static inline void callstack_check_depth(struct r600_shader_ctx *ctx, unsigned r
 				ctx->bc->callstack[ctx->bc->call_sp].current + diff;
 		}
 		return;
-	}					
+	}
 	switch (reason) {
 	case FC_PUSH_VPM:
 		ctx->bc->callstack[ctx->bc->call_sp].current++;
@@ -2878,7 +2772,7 @@ static int emit_setret_in_loop_flag(struct r600_shader_ctx *ctx, unsigned flag_v
 
 static void emit_testflag(struct r600_shader_ctx *ctx)
 {
-	
+
 }
 
 static void emit_return_on_flag(struct r600_shader_ctx *ctx, unsigned ifidx)
@@ -3239,7 +3133,7 @@ static struct r600_shader_tgsi_instruction eg_shader_tgsi_instruction[] = {
 	{TGSI_OPCODE_UP4UB,	0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP, tgsi_unsupported},
 	{TGSI_OPCODE_X2D,	0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP, tgsi_unsupported},
 	{TGSI_OPCODE_ARA,	0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP, tgsi_unsupported},
-	{TGSI_OPCODE_ARR,	0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP, tgsi_unsupported},
+	{TGSI_OPCODE_ARR,	0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP, tgsi_eg_arl},
 	{TGSI_OPCODE_BRA,	0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP, tgsi_unsupported},
 	{TGSI_OPCODE_CAL,	0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP, tgsi_unsupported},
 	{TGSI_OPCODE_RET,	0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP, tgsi_unsupported},
diff --git a/src/gallium/drivers/r600/r600_shader.h b/src/gallium/drivers/r600/r600_shader.h
index f8bc595139..35b0331525 100644
--- a/src/gallium/drivers/r600/r600_shader.h
+++ b/src/gallium/drivers/r600/r600_shader.h
@@ -38,7 +38,6 @@ struct r600_shader_io {
 struct r600_shader {
 	unsigned		processor_type;
 	struct r600_bc		bc;
-	boolean			flat_shade;
 	unsigned		ninput;
 	unsigned		noutput;
 	unsigned		nlds;
diff --git a/src/gallium/drivers/r600/r600_state.c b/src/gallium/drivers/r600/r600_state.c
index 54cc79b154..cd5f0792d5 100644
--- a/src/gallium/drivers/r600/r600_state.c
+++ b/src/gallium/drivers/r600/r600_state.c
@@ -36,7 +36,6 @@
 #include <util/u_pack_color.h>
 #include <util/u_memory.h>
 #include <util/u_inlines.h>
-#include <util/u_upload_mgr.h>
 #include <util/u_framebuffer.h>
 #include <pipebuffer/pb_buffer.h>
 #include "r600.h"
@@ -46,14 +45,164 @@
 #include "r600_pipe.h"
 #include "r600_state_inlines.h"
 
+void r600_polygon_offset_update(struct r600_pipe_context *rctx)
+{
+	struct r600_pipe_state state;
+
+	state.id = R600_PIPE_STATE_POLYGON_OFFSET;
+	state.nregs = 0;
+	if (rctx->rasterizer && rctx->framebuffer.zsbuf) {
+		float offset_units = rctx->rasterizer->offset_units;
+		unsigned offset_db_fmt_cntl = 0, depth;
+
+		switch (rctx->framebuffer.zsbuf->texture->format) {
+		case PIPE_FORMAT_Z24X8_UNORM:
+		case PIPE_FORMAT_Z24_UNORM_S8_USCALED:
+			depth = -24;
+			offset_units *= 2.0f;
+			break;
+		case PIPE_FORMAT_Z32_FLOAT:
+			depth = -23;
+			offset_units *= 1.0f;
+			offset_db_fmt_cntl |= S_028DF8_POLY_OFFSET_DB_IS_FLOAT_FMT(1);
+			break;
+		case PIPE_FORMAT_Z16_UNORM:
+			depth = -16;
+			offset_units *= 4.0f;
+			break;
+		default:
+			return;
+		}
+		/* FIXME some of those reg can be computed with cso */
+		offset_db_fmt_cntl |= S_028DF8_POLY_OFFSET_NEG_NUM_DB_BITS(depth);
+		r600_pipe_state_add_reg(&state,
+				R_028E00_PA_SU_POLY_OFFSET_FRONT_SCALE,
+				fui(rctx->rasterizer->offset_scale), 0xFFFFFFFF, NULL);
+		r600_pipe_state_add_reg(&state,
+				R_028E04_PA_SU_POLY_OFFSET_FRONT_OFFSET,
+				fui(offset_units), 0xFFFFFFFF, NULL);
+		r600_pipe_state_add_reg(&state,
+				R_028E08_PA_SU_POLY_OFFSET_BACK_SCALE,
+				fui(rctx->rasterizer->offset_scale), 0xFFFFFFFF, NULL);
+		r600_pipe_state_add_reg(&state,
+				R_028E0C_PA_SU_POLY_OFFSET_BACK_OFFSET,
+				fui(offset_units), 0xFFFFFFFF, NULL);
+		r600_pipe_state_add_reg(&state,
+				R_028DF8_PA_SU_POLY_OFFSET_DB_FMT_CNTL,
+				offset_db_fmt_cntl, 0xFFFFFFFF, NULL);
+		r600_context_pipe_state_set(&rctx->ctx, &state);
+	}
+}
+
+/* FIXME optimize away spi update when it's not needed */
+static void r600_spi_update(struct r600_pipe_context *rctx)
+{
+	struct r600_pipe_shader *shader = rctx->ps_shader;
+	struct r600_pipe_state rstate;
+	struct r600_shader *rshader = &shader->shader;
+	unsigned i, tmp;
+
+	rstate.nregs = 0;
+	for (i = 0; i < rshader->ninput; i++) {
+		tmp = S_028644_SEMANTIC(r600_find_vs_semantic_index(&rctx->vs_shader->shader, rshader, i));
+		if (rshader->input[i].centroid)
+			tmp |= S_028644_SEL_CENTROID(1);
+		if (rshader->input[i].interpolate == TGSI_INTERPOLATE_LINEAR)
+			tmp |= S_028644_SEL_LINEAR(1);
+
+		if (rshader->input[i].name == TGSI_SEMANTIC_COLOR ||
+		    rshader->input[i].name == TGSI_SEMANTIC_BCOLOR ||
+		    rshader->input[i].name == TGSI_SEMANTIC_POSITION) {
+			tmp |= S_028644_FLAT_SHADE(rctx->flatshade);
+		}
+		if (rshader->input[i].name == TGSI_SEMANTIC_GENERIC &&
+			rctx->sprite_coord_enable & (1 << rshader->input[i].sid)) {
+			tmp |= S_028644_PT_SPRITE_TEX(1);
+		}
+		r600_pipe_state_add_reg(&rstate, R_028644_SPI_PS_INPUT_CNTL_0 + i * 4, tmp, 0xFFFFFFFF, NULL);
+	}
+	r600_context_pipe_state_set(&rctx->ctx, &rstate);
+}
+
+void r600_vertex_buffer_update(struct r600_pipe_context *rctx)
+{
+	struct r600_pipe_state *rstate;
+	struct r600_resource *rbuffer;
+	struct pipe_vertex_buffer *vertex_buffer;
+	unsigned i, offset;
+
+	/* we don't update until we know vertex elements */
+	if (rctx->vertex_elements == NULL || !rctx->nvertex_buffer)
+		return;
+
+	if (rctx->vertex_elements->incompatible_layout) {
+		/* translate rebind new vertex elements so
+		 * return once translated
+		 */
+		r600_begin_vertex_translate(rctx);
+		return;
+	}
+
+	if (rctx->any_user_vbs) {
+		r600_upload_user_buffers(rctx);
+		rctx->any_user_vbs = FALSE;
+	}
+
+	if (rctx->vertex_elements->vbuffer_need_offset) {
+		/* one resource per vertex elements */
+		rctx->nvs_resource = rctx->vertex_elements->count;
+	} else {
+		/* bind vertex buffer once */
+		rctx->nvs_resource = rctx->nvertex_buffer;
+	}
+
+	for (i = 0 ; i < rctx->nvs_resource; i++) {
+		rstate = &rctx->vs_resource[i];
+		rstate->id = R600_PIPE_STATE_RESOURCE;
+		rstate->nregs = 0;
+
+		if (rctx->vertex_elements->vbuffer_need_offset) {
+			/* one resource per vertex elements */
+			unsigned vbuffer_index;
+			vbuffer_index = rctx->vertex_elements->elements[i].vertex_buffer_index;
+			vertex_buffer = &rctx->vertex_buffer[vbuffer_index];
+			rbuffer = (struct r600_resource*)vertex_buffer->buffer;
+			offset = rctx->vertex_elements->vbuffer_offset[i] +
+				vertex_buffer->buffer_offset +
+				r600_bo_offset(rbuffer->bo);
+		} else {
+			/* bind vertex buffer once */
+			vertex_buffer = &rctx->vertex_buffer[i];
+			rbuffer = (struct r600_resource*)vertex_buffer->buffer;
+			offset = vertex_buffer->buffer_offset +
+				r600_bo_offset(rbuffer->bo);
+		}
+
+		r600_pipe_state_add_reg(rstate, R_038000_RESOURCE0_WORD0,
+					offset, 0xFFFFFFFF, rbuffer->bo);
+		r600_pipe_state_add_reg(rstate, R_038004_RESOURCE0_WORD1,
+					rbuffer->bo_size - offset - 1, 0xFFFFFFFF, NULL);
+		r600_pipe_state_add_reg(rstate, R_038008_RESOURCE0_WORD2,
+					S_038008_STRIDE(vertex_buffer->stride),
+					0xFFFFFFFF, NULL);
+		r600_pipe_state_add_reg(rstate, R_03800C_RESOURCE0_WORD3,
+					0x00000000, 0xFFFFFFFF, NULL);
+		r600_pipe_state_add_reg(rstate, R_038010_RESOURCE0_WORD4,
+					0x00000000, 0xFFFFFFFF, NULL);
+		r600_pipe_state_add_reg(rstate, R_038014_RESOURCE0_WORD5,
+					0x00000000, 0xFFFFFFFF, NULL);
+		r600_pipe_state_add_reg(rstate, R_038018_RESOURCE0_WORD6,
+					0xC0000000, 0xFFFFFFFF, NULL);
+		r600_context_pipe_state_set_fs_resource(&rctx->ctx, rstate, i);
+	}
+}
+
 static void r600_draw_common(struct r600_drawl *draw)
 {
 	struct r600_pipe_context *rctx = (struct r600_pipe_context *)draw->ctx;
-	struct r600_pipe_state *rstate;
 	struct r600_resource *rbuffer;
-	unsigned i, j, offset, prim;
+	unsigned prim;
 	u32 vgt_dma_index_type, vgt_draw_initiator, mask;
-	struct pipe_vertex_buffer *vertex_buffer;
 	struct r600_draw rdraw;
 	struct r600_pipe_state vgt;
 
@@ -76,42 +225,23 @@ static void r600_draw_common(struct r600_drawl *draw)
 	}
 	if (r600_conv_pipe_prim(draw->mode, &prim))
 		return;
-
-
-	/* rebuild vertex shader if input format changed */
-	if (r600_pipe_shader_update(&rctx->context, rctx->vs_shader))
+	if (unlikely(rctx->ps_shader == NULL)) {
+		R600_ERR("missing vertex shader\n");
 		return;
-	if (r600_pipe_shader_update(&rctx->context, rctx->ps_shader))
+	}
+	if (unlikely(rctx->vs_shader == NULL)) {
+		R600_ERR("missing vertex shader\n");
+		return;
+	}
+	/* there should be enough input */
+	if (rctx->vertex_elements->count < rctx->vs_shader->shader.bc.nresource) {
+		R600_ERR("%d resources provided, expecting %d\n",
+			rctx->vertex_elements->count, rctx->vs_shader->shader.bc.nresource);
 		return;
-
-	for (i = 0 ; i < rctx->vertex_elements->count; i++) {
-		uint32_t word2, format;
-
-		rstate = &rctx->vs_resource[i];
-		rstate->id = R600_PIPE_STATE_RESOURCE;
-		rstate->nregs = 0;
-
-		j = rctx->vertex_elements->elements[i].vertex_buffer_index;
-		vertex_buffer = &rctx->vertex_buffer[j];
-		rbuffer = (struct r600_resource*)vertex_buffer->buffer;
-		offset = rctx->vertex_elements->elements[i].src_offset +
-			vertex_buffer->buffer_offset +
-			r600_bo_offset(rbuffer->bo);
-
-		format = r600_translate_vertex_data_type(rctx->vertex_elements->hw_format[i]);
-
-		word2 = format | S_038008_STRIDE(vertex_buffer->stride);
-
-		r600_pipe_state_add_reg(rstate, R_038000_RESOURCE0_WORD0, offset, 0xFFFFFFFF, rbuffer->bo);
-		r600_pipe_state_add_reg(rstate, R_038004_RESOURCE0_WORD1, rbuffer->size - offset - 1, 0xFFFFFFFF, NULL);
-		r600_pipe_state_add_reg(rstate, R_038008_RESOURCE0_WORD2, word2, 0xFFFFFFFF, NULL);
-		r600_pipe_state_add_reg(rstate, R_03800C_RESOURCE0_WORD3, 0x00000000, 0xFFFFFFFF, NULL);
-		r600_pipe_state_add_reg(rstate, R_038010_RESOURCE0_WORD4, 0x00000000, 0xFFFFFFFF, NULL);
-		r600_pipe_state_add_reg(rstate, R_038014_RESOURCE0_WORD5, 0x00000000, 0xFFFFFFFF, NULL);
-		r600_pipe_state_add_reg(rstate, R_038018_RESOURCE0_WORD6, 0xC0000000, 0xFFFFFFFF, NULL);
-		r600_context_pipe_state_set_vs_resource(&rctx->ctx, rstate, i);
 	}
 
+	r600_spi_update(rctx);
+
 	mask = 0;
 	for (int i = 0; i < rctx->framebuffer.nr_cbufs; i++) {
 		mask |= (0xF << (i * 4));
@@ -126,46 +256,6 @@ static void r600_draw_common(struct r600_drawl *draw)
 	r600_pipe_state_add_reg(&vgt, R_028238_CB_TARGET_MASK, rctx->cb_target_mask & mask, 0xFFFFFFFF, NULL);
 	r600_pipe_state_add_reg(&vgt, R_03CFF0_SQ_VTX_BASE_VTX_LOC, 0, 0xFFFFFFFF, NULL);
 	r600_pipe_state_add_reg(&vgt, R_03CFF4_SQ_VTX_START_INST_LOC, 0, 0xFFFFFFFF, NULL);
-	/* build late state */
-	if (rctx->rasterizer && rctx->framebuffer.zsbuf) {
-		float offset_units = rctx->rasterizer->offset_units;
-		unsigned offset_db_fmt_cntl = 0, depth;
-
-		switch (rctx->framebuffer.zsbuf->texture->format) {
-		case PIPE_FORMAT_Z24X8_UNORM:
-		case PIPE_FORMAT_Z24_UNORM_S8_USCALED:
-			depth = -24;
-			offset_units *= 2.0f;
-			break;
-		case PIPE_FORMAT_Z32_FLOAT:
-			depth = -23;
-			offset_units *= 1.0f;
-			offset_db_fmt_cntl |= S_028DF8_POLY_OFFSET_DB_IS_FLOAT_FMT(1);
-			break;
-		case PIPE_FORMAT_Z16_UNORM:
-			depth = -16;
-			offset_units *= 4.0f;
-			break;
-		default:
-			return;
-		}
-		offset_db_fmt_cntl |= S_028DF8_POLY_OFFSET_NEG_NUM_DB_BITS(depth);
-		r600_pipe_state_add_reg(&vgt,
-				R_028E00_PA_SU_POLY_OFFSET_FRONT_SCALE,
-				fui(rctx->rasterizer->offset_scale), 0xFFFFFFFF, NULL);
-		r600_pipe_state_add_reg(&vgt,
-				R_028E04_PA_SU_POLY_OFFSET_FRONT_OFFSET,
-				fui(offset_units), 0xFFFFFFFF, NULL);
-		r600_pipe_state_add_reg(&vgt,
-				R_028E08_PA_SU_POLY_OFFSET_BACK_SCALE,
-				fui(rctx->rasterizer->offset_scale), 0xFFFFFFFF, NULL);
-		r600_pipe_state_add_reg(&vgt,
-				R_028E0C_PA_SU_POLY_OFFSET_BACK_OFFSET,
-				fui(offset_units), 0xFFFFFFFF, NULL);
-		r600_pipe_state_add_reg(&vgt,
-				R_028DF8_PA_SU_POLY_OFFSET_DB_FMT_CNTL,
-				offset_db_fmt_cntl, 0xFFFFFFFF, NULL);
-	}
 	r600_context_pipe_state_set(&rctx->ctx, &vgt);
 
 	rdraw.vgt_num_indices = draw->count;
@@ -185,17 +275,7 @@ void r600_draw_vbo(struct pipe_context *ctx, const struct pipe_draw_info *info)
 {
 	struct r600_pipe_context *rctx = (struct r600_pipe_context *)ctx;
 	struct r600_drawl draw;
-	boolean translate = FALSE;
 
-	if (rctx->vertex_elements->incompatible_layout) {
-		r600_begin_vertex_translate(rctx);
-		translate = TRUE;
-	}
-
-	if (rctx->any_user_vbs) {
-		r600_upload_user_buffers(rctx);
-		rctx->any_user_vbs = FALSE;
-	}
 	memset(&draw, 0, sizeof(struct r600_drawl));
 	draw.ctx = ctx;
 	draw.mode = info->mode;
@@ -226,9 +306,6 @@ void r600_draw_vbo(struct pipe_context *ctx, const struct pipe_draw_info *info)
 	}
 	r600_draw_common(&draw);
 
-	if (translate)
-		r600_end_vertex_translate(rctx);
-
 	pipe_resource_reference(&draw.index_buffer, NULL);
 }
 
@@ -296,7 +373,7 @@ static void *r600_create_blend_state(struct pipe_context *ctx,
 		unsigned eqRGB = state->rt[i].rgb_func;
 		unsigned srcRGB = state->rt[i].rgb_src_factor;
 		unsigned dstRGB = state->rt[i].rgb_dst_factor;
-		
+
 		unsigned eqA = state->rt[i].alpha_func;
 		unsigned srcA = state->rt[i].alpha_src_factor;
 		unsigned dstA = state->rt[i].alpha_dst_factor;
@@ -475,7 +552,7 @@ static void *r600_create_rs_state(struct pipe_context *ctx,
 	r600_pipe_state_add_reg(rstate, R_028A0C_PA_SC_LINE_STIPPLE, 0x00000005, 0xFFFFFFFF, NULL);
 	r600_pipe_state_add_reg(rstate, R_028A48_PA_SC_MPASS_PS_CNTL, 0x00000000, 0xFFFFFFFF, NULL);
 	r600_pipe_state_add_reg(rstate, R_028C00_PA_SC_LINE_CNTL, 0x00000400, 0xFFFFFFFF, NULL);
-        
+
 	r600_pipe_state_add_reg(rstate, R_028C08_PA_SU_VTX_CNTL,
 				S_028C08_PIX_CENTER_HALF(state->gl_rasterization_rules),
 				0xFFFFFFFF, NULL);
@@ -603,9 +680,9 @@ static struct pipe_sampler_view *r600_create_sampler_view(struct pipe_context *c
 				word4 | S_038010_NUM_FORMAT_ALL(V_038010_SQ_NUM_FORMAT_NORM) |
 				S_038010_SRF_MODE_ALL(V_038010_SFR_MODE_NO_ZERO) |
 				S_038010_REQUEST_SIZE(1) |
-				S_038010_BASE_LEVEL(state->first_level), 0xFFFFFFFF, NULL);
+				S_038010_BASE_LEVEL(state->u.tex.first_level), 0xFFFFFFFF, NULL);
 	r600_pipe_state_add_reg(rstate, R_038014_RESOURCE0_WORD5,
-				S_038014_LAST_LEVEL(state->last_level) |
+				S_038014_LAST_LEVEL(state->u.tex.last_level) |
 				S_038014_BASE_ARRAY(0) |
 				S_038014_LAST_ARRAY(0), 0xFFFFFFFF, NULL);
 	r600_pipe_state_add_reg(rstate, R_038018_RESOURCE0_WORD6,
@@ -622,7 +699,7 @@ static void r600_set_vs_sampler_view(struct pipe_context *ctx, unsigned count,
 
 	for (int i = 0; i < count; i++) {
 		if (resource[i]) {
-			r600_context_pipe_state_set_vs_resource(&rctx->ctx, &resource[i]->state, i + PIPE_MAX_ATTRIBS);
+			r600_context_pipe_state_set_vs_resource(&rctx->ctx, &resource[i]->state, i);
 		}
 	}
 }
@@ -692,16 +769,16 @@ static void r600_set_clip_state(struct pipe_context *ctx,
 	rstate->id = R600_PIPE_STATE_CLIP;
 	for (int i = 0; i < state->nr; i++) {
 		r600_pipe_state_add_reg(rstate,
-					R_028E20_PA_CL_UCP0_X + i * 4,
+					R_028E20_PA_CL_UCP0_X + i * 16,
 					fui(state->ucp[i][0]), 0xFFFFFFFF, NULL);
 		r600_pipe_state_add_reg(rstate,
-					R_028E24_PA_CL_UCP0_Y + i * 4,
+					R_028E24_PA_CL_UCP0_Y + i * 16,
 					fui(state->ucp[i][1]) , 0xFFFFFFFF, NULL);
 		r600_pipe_state_add_reg(rstate,
-					R_028E28_PA_CL_UCP0_Z + i * 4,
+					R_028E28_PA_CL_UCP0_Z + i * 16,
 					fui(state->ucp[i][2]), 0xFFFFFFFF, NULL);
 		r600_pipe_state_add_reg(rstate,
-					R_028E2C_PA_CL_UCP0_W + i * 4,
+					R_028E2C_PA_CL_UCP0_W + i * 16,
 					fui(state->ucp[i][3]), 0xFFFFFFFF, NULL);
 	}
 	r600_pipe_state_add_reg(rstate, R_028810_PA_CL_CLIP_CNTL,
@@ -824,10 +901,11 @@ static void r600_cb(struct r600_pipe_context *rctx, struct r600_pipe_state *rsta
 	struct r600_resource_texture *rtex;
 	struct r600_resource *rbuffer;
 	struct r600_surface *surf;
-	unsigned level = state->cbufs[cb]->level;
+	unsigned level = state->cbufs[cb]->u.tex.level;
 	unsigned pitch, slice;
 	unsigned color_info;
 	unsigned format, swap, ntype;
+	unsigned offset;
 	const struct util_format_description *desc;
 	struct r600_bo *bo[3];
 
@@ -838,6 +916,9 @@ static void r600_cb(struct r600_pipe_context *rctx, struct r600_pipe_state *rsta
 	bo[1] = rbuffer->bo;
 	bo[2] = rbuffer->bo;
 
+	/* XXX quite sure for dx10+ hw don't need any offset hacks */
+	offset = r600_texture_get_offset((struct r600_resource_texture *)state->cbufs[cb]->texture,
+					 level, state->cbufs[cb]->u.tex.first_layer);
 	pitch = rtex->pitch_in_pixels[level] / 8 - 1;
 	slice = rtex->pitch_in_pixels[level] * surf->aligned_height / 64 - 1;
 	ntype = 0;
@@ -852,12 +933,12 @@ static void r600_cb(struct r600_pipe_context *rctx, struct r600_pipe_state *rsta
 		S_0280A0_ARRAY_MODE(rtex->array_mode[level]) |
 		S_0280A0_BLEND_CLAMP(1) |
 		S_0280A0_NUMBER_TYPE(ntype);
-	if (desc->colorspace != UTIL_FORMAT_COLORSPACE_ZS) 
+	if (desc->colorspace != UTIL_FORMAT_COLORSPACE_ZS)
 		color_info |= S_0280A0_SOURCE_FORMAT(1);
 
 	r600_pipe_state_add_reg(rstate,
 				R_028040_CB_COLOR0_BASE + cb * 4,
-				(state->cbufs[cb]->offset + r600_bo_offset(bo[0])) >> 8, 0xFFFFFFFF, bo[0]);
+				(offset + r600_bo_offset(bo[0])) >> 8, 0xFFFFFFFF, bo[0]);
 	r600_pipe_state_add_reg(rstate,
 				R_0280A0_CB_COLOR0_INFO + cb * 4,
 				color_info, 0xFFFFFFFF, bo[0]);
@@ -888,11 +969,12 @@ static void r600_db(struct r600_pipe_context *rctx, struct r600_pipe_state *rsta
 	struct r600_surface *surf;
 	unsigned level;
 	unsigned pitch, slice, format;
+	unsigned offset;
 
 	if (state->zsbuf == NULL)
 		return;
 
-	level = state->zsbuf->level;
+	level = state->zsbuf->u.tex.level;
 
 	surf = (struct r600_surface *)state->zsbuf;
 	rtex = (struct r600_resource_texture*)state->zsbuf->texture;
@@ -902,12 +984,15 @@ static void r600_db(struct r600_pipe_context *rctx, struct r600_pipe_state *rsta
 	rtex->depth = 1;
 	rbuffer = &rtex->resource;
 
+	/* XXX quite sure for dx10+ hw don't need any offset hacks */
+	offset = r600_texture_get_offset((struct r600_resource_texture *)state->zsbuf->texture,
+					 level, state->zsbuf->u.tex.first_layer);
 	pitch = rtex->pitch_in_pixels[level] / 8 - 1;
 	slice = rtex->pitch_in_pixels[level] * surf->aligned_height / 64 - 1;
 	format = r600_translate_dbformat(state->zsbuf->texture->format);
 
 	r600_pipe_state_add_reg(rstate, R_02800C_DB_DEPTH_BASE,
-				(state->zsbuf->offset + r600_bo_offset(rbuffer->bo)) >> 8, 0xFFFFFFFF, rbuffer->bo);
+				(offset + r600_bo_offset(rbuffer->bo)) >> 8, 0xFFFFFFFF, rbuffer->bo);
 	r600_pipe_state_add_reg(rstate, R_028000_DB_DEPTH_SIZE,
 				S_028000_PITCH_TILE_MAX(pitch) | S_028000_SLICE_TILE_MAX(slice),
 				0xFFFFFFFF, NULL);
@@ -933,7 +1018,7 @@ static void r600_set_framebuffer_state(struct pipe_context *ctx,
 	rstate->id = R600_PIPE_STATE_FRAMEBUFFER;
 
 	util_copy_framebuffer_state(&rctx->framebuffer, state);
-	
+
 	rctx->pframebuffer = &rctx->framebuffer;
 
 	/* build states */
@@ -1015,6 +1100,10 @@ static void r600_set_framebuffer_state(struct pipe_context *ctx,
 	free(rctx->states[R600_PIPE_STATE_FRAMEBUFFER]);
 	rctx->states[R600_PIPE_STATE_FRAMEBUFFER] = rstate;
 	r600_context_pipe_state_set(&rctx->ctx, rstate);
+
+	if (state->zsbuf) {
+		r600_polygon_offset_update(rctx);
+	}
 }
 
 static void r600_set_constant_buffer(struct pipe_context *ctx, uint shader, uint index,
diff --git a/src/gallium/drivers/r600/r600_state_common.c b/src/gallium/drivers/r600/r600_state_common.c
index 210420e823..99b372caac 100644
--- a/src/gallium/drivers/r600/r600_state_common.c
+++ b/src/gallium/drivers/r600/r600_state_common.c
@@ -58,6 +58,12 @@ void r600_bind_rs_state(struct pipe_context *ctx, void *state)
 
 	rctx->states[rs->rstate.id] = &rs->rstate;
 	r600_context_pipe_state_set(&rctx->ctx, &rs->rstate);
+
+	if (rctx->family >= CHIP_CEDAR) {
+		evergreen_polygon_offset_update(rctx);
+	} else {
+		r600_polygon_offset_update(rctx);
+	}
 }
 
 void r600_delete_rs_state(struct pipe_context *ctx, void *state)
@@ -113,8 +119,23 @@ void r600_bind_vertex_elements(struct pipe_context *ctx, void *state)
 	struct r600_pipe_context *rctx = (struct r600_pipe_context *)ctx;
 	struct r600_vertex_element *v = (struct r600_vertex_element*)state;
 
+	/* delete previous translated vertex elements */
+	if (rctx->tran.new_velems) {
+		r600_end_vertex_translate(rctx);
+	}
+
 	rctx->vertex_elements = v;
 	if (v) {
+		rctx->states[v->rstate.id] = &v->rstate;
+		r600_context_pipe_state_set(&rctx->ctx, &v->rstate);
+		if (rctx->family >= CHIP_CEDAR) {
+			evergreen_vertex_buffer_update(rctx);
+		} else {
+			r600_vertex_buffer_update(rctx);
+		}
+	}
+
+	if (v) {
 //		rctx->vs_rebuild = TRUE;
 	}
 }
@@ -122,11 +143,16 @@ void r600_bind_vertex_elements(struct pipe_context *ctx, void *state)
 void r600_delete_vertex_element(struct pipe_context *ctx, void *state)
 {
 	struct r600_pipe_context *rctx = (struct r600_pipe_context *)ctx;
+	struct r600_vertex_element *v = (struct r600_vertex_element*)state;
 
-	FREE(state);
-
+	if (rctx->states[v->rstate.id] == &v->rstate) {
+		rctx->states[v->rstate.id] = NULL;
+	}
 	if (rctx->vertex_elements == state)
 		rctx->vertex_elements = NULL;
+
+	r600_bo_reference(rctx->radeon, &v->fetch_shader, NULL);
+	FREE(state);
 }
 
 
@@ -166,16 +192,23 @@ void r600_set_vertex_buffers(struct pipe_context *ctx, unsigned count,
 			rctx->any_user_vbs = TRUE;
 		pipe_resource_reference(&rctx->vertex_buffer[i].buffer, buffers[i].buffer);
 
+		/* The stride of zero means we will be fetching only the first
+		 * vertex, so don't care about max_index. */
+		if (!vbo->stride)
+			continue;
+
 		if (vbo->max_index == ~0) {
-			if (!vbo->stride)
-				vbo->max_index = 1;
-			else
-				vbo->max_index = (vbo->buffer->width0 - vbo->buffer_offset) / vbo->stride;
+			vbo->max_index = (vbo->buffer->width0 - vbo->buffer_offset) / vbo->stride;
 		}
 		max_index = MIN2(vbo->max_index, max_index);
 	}
 	rctx->nvertex_buffer = count;
 	rctx->vb_max_index = max_index;
+	if (rctx->family >= CHIP_CEDAR) {
+		evergreen_vertex_buffer_update(rctx);
+	} else {
+		r600_vertex_buffer_update(rctx);
+	}
 }
 
 
@@ -186,9 +219,10 @@ void *r600_create_vertex_elements(struct pipe_context *ctx,
 				  unsigned count,
 				  const struct pipe_vertex_element *elements)
 {
+	struct r600_pipe_context *rctx = (struct r600_pipe_context *)ctx;
 	struct r600_vertex_element *v = CALLOC_STRUCT(r600_vertex_element);
-	int i;
 	enum pipe_format *format;
+	int i;
 
 	assert(count < 32);
 	if (!v)
@@ -202,19 +236,22 @@ void *r600_create_vertex_elements(struct pipe_context *ctx,
 		format = &v->hw_format[i];
 
 		switch (*format) {
-                    FORMAT_REPLACE(R64_FLOAT,           R32_FLOAT);
-                    FORMAT_REPLACE(R64G64_FLOAT,        R32G32_FLOAT);
-                    FORMAT_REPLACE(R64G64B64_FLOAT,     R32G32B32_FLOAT);
-                    FORMAT_REPLACE(R64G64B64A64_FLOAT,  R32G32B32A32_FLOAT);
+		FORMAT_REPLACE(R64_FLOAT,           R32_FLOAT);
+		FORMAT_REPLACE(R64G64_FLOAT,        R32G32_FLOAT);
+		FORMAT_REPLACE(R64G64B64_FLOAT,     R32G32B32_FLOAT);
+		FORMAT_REPLACE(R64G64B64A64_FLOAT,  R32G32B32A32_FLOAT);
 		default:;
 		}
 		v->incompatible_layout =
 			v->incompatible_layout ||
-			v->elements[i].src_format != v->hw_format[i] ||
-			v->elements[i].src_offset % 4 != 0;
+			v->elements[i].src_format != v->hw_format[i];
+
+		v->hw_format_size[i] = align(util_format_get_blocksize(v->hw_format[i]), 4);
+	}
 
-                v->hw_format_size[i] =
-			align(util_format_get_blocksize(v->hw_format[i]), 4);
+	if (r600_vertex_elements_build_fetch_shader(rctx, v)) {
+		FREE(v);
+		return NULL;
 	}
 
 	return v;
@@ -239,6 +276,9 @@ void r600_bind_ps_shader(struct pipe_context *ctx, void *state)
 
 	/* TODO delete old shader */
 	rctx->ps_shader = (struct r600_pipe_shader *)state;
+	if (state) {
+		r600_context_pipe_state_set(&rctx->ctx, &rctx->ps_shader->rstate);
+	}
 }
 
 void r600_bind_vs_shader(struct pipe_context *ctx, void *state)
@@ -247,6 +287,9 @@ void r600_bind_vs_shader(struct pipe_context *ctx, void *state)
 
 	/* TODO delete old shader */
 	rctx->vs_shader = (struct r600_pipe_shader *)state;
+	if (state) {
+		r600_context_pipe_state_set(&rctx->ctx, &rctx->vs_shader->rstate);
+	}
 }
 
 void r600_delete_ps_shader(struct pipe_context *ctx, void *state)
diff --git a/src/gallium/drivers/r600/r600_state_inlines.h b/src/gallium/drivers/r600/r600_state_inlines.h
index 1be5b156d3..d994196e19 100644
--- a/src/gallium/drivers/r600/r600_state_inlines.h
+++ b/src/gallium/drivers/r600/r600_state_inlines.h
@@ -282,6 +282,7 @@ static inline uint32_t r600_translate_colorswap(enum pipe_format format)
 	switch (format) {
 		/* 8-bit buffers. */
 	case PIPE_FORMAT_A8_UNORM:
+		return V_0280A0_SWAP_ALT_REV;
 	case PIPE_FORMAT_I8_UNORM:
 	case PIPE_FORMAT_L8_UNORM:
 	case PIPE_FORMAT_R8_UNORM:
@@ -304,6 +305,7 @@ static inline uint32_t r600_translate_colorswap(enum pipe_format format)
 		return V_0280A0_SWAP_STD;
 
 	case PIPE_FORMAT_L8A8_UNORM:
+		return V_0280A0_SWAP_ALT;
 	case PIPE_FORMAT_R8G8_UNORM:
 		return V_0280A0_SWAP_STD;
 
diff --git a/src/gallium/drivers/r600/r600_texture.c b/src/gallium/drivers/r600/r600_texture.c
index e719f7fb98..d4d9b07c0e 100644
--- a/src/gallium/drivers/r600/r600_texture.c
+++ b/src/gallium/drivers/r600/r600_texture.c
@@ -45,14 +45,10 @@ static void r600_copy_to_staging_texture(struct pipe_context *ctx, struct r600_t
 {
 	struct pipe_transfer *transfer = (struct pipe_transfer*)rtransfer;
 	struct pipe_resource *texture = transfer->resource;
-	struct pipe_subresource subdst;
 
-	subdst.face = 0;
-	subdst.level = 0;
 	ctx->resource_copy_region(ctx, rtransfer->staging_texture,
-				subdst, 0, 0, 0, texture, transfer->sr,
-				transfer->box.x, transfer->box.y, transfer->box.z,
-				transfer->box.width, transfer->box.height);
+				0, 0, 0, 0, texture, transfer->level,
+				&transfer->box);
 }
 
 
@@ -61,34 +57,32 @@ static void r600_copy_from_staging_texture(struct pipe_context *ctx, struct r600
 {
 	struct pipe_transfer *transfer = (struct pipe_transfer*)rtransfer;
 	struct pipe_resource *texture = transfer->resource;
-	struct pipe_subresource subsrc;
-
-	subsrc.face = 0;
-	subsrc.level = 0;
-	ctx->resource_copy_region(ctx, texture, transfer->sr,
+	struct pipe_box sbox;
+
+	sbox.x = sbox.y = sbox.z = 0;
+	sbox.width = transfer->box.width;
+	sbox.height = transfer->box.height;
+	/* XXX that might be wrong */
+	sbox.depth = 1;
+	ctx->resource_copy_region(ctx, texture, transfer->level,
 				  transfer->box.x, transfer->box.y, transfer->box.z,
-				  rtransfer->staging_texture, subsrc,
-				  0, 0, 0,
-				  transfer->box.width, transfer->box.height);
+				  rtransfer->staging_texture,
+				  0, &sbox);
 
 	ctx->flush(ctx, 0, NULL);
 }
 
-static unsigned r600_texture_get_offset(struct r600_resource_texture *rtex,
-					unsigned level, unsigned zslice,
-					unsigned face)
+unsigned r600_texture_get_offset(struct r600_resource_texture *rtex,
+					unsigned level, unsigned layer)
 {
 	unsigned offset = rtex->offset[level];
 
 	switch (rtex->resource.base.b.target) {
 	case PIPE_TEXTURE_3D:
-		assert(face == 0);
-		return offset + zslice * rtex->layer_size[level];
 	case PIPE_TEXTURE_CUBE:
-		assert(zslice == 0);
-		return offset + face * rtex->layer_size[level];
+		return offset + layer * rtex->layer_size[level];
 	default:
-		assert(zslice == 0 && face == 0);
+		assert(layer == 0);
 		return offset;
 	}
 }
@@ -109,11 +103,11 @@ static unsigned r600_get_pixel_alignment(struct pipe_screen *screen,
 	case V_038000_ARRAY_2D_TILED_THIN1:
 		p_align = MAX2(rscreen->tiling_info->num_banks,
 			       (((rscreen->tiling_info->group_bytes / 8 / pixsize)) *
-				rscreen->tiling_info->num_banks));
+				rscreen->tiling_info->num_banks)) * 8;
 		break;
-	case 0:
+	case V_038000_ARRAY_LINEAR_GENERAL:
 	default:
-		p_align = 64;
+		p_align = rscreen->tiling_info->group_bytes / pixsize;
 		break;
 	}
 	return p_align;
@@ -139,6 +133,29 @@ static unsigned r600_get_height_alignment(struct pipe_screen *screen,
 	return h_align;
 }
 
+static unsigned r600_get_base_alignment(struct pipe_screen *screen,
+					enum pipe_format format,
+					unsigned array_mode)
+{
+	struct r600_screen* rscreen = (struct r600_screen *)screen;
+	unsigned pixsize = util_format_get_blocksize(format);
+	int p_align = r600_get_pixel_alignment(screen, format, array_mode);
+	int h_align = r600_get_height_alignment(screen, array_mode);
+	int b_align;
+
+	switch (array_mode) {
+	case V_038000_ARRAY_2D_TILED_THIN1:
+		b_align = MAX2(rscreen->tiling_info->num_banks * rscreen->tiling_info->num_channels * 8 * 8 * pixsize,
+			       p_align * pixsize * h_align);
+		break;
+	case V_038000_ARRAY_1D_TILED_THIN1:
+	default:
+		b_align = rscreen->tiling_info->group_bytes;
+		break;
+	}
+	return b_align;
+}
+
 static unsigned mip_minify(unsigned size, unsigned level)
 {
 	unsigned val;
@@ -153,10 +170,8 @@ static unsigned r600_texture_get_stride(struct pipe_screen *screen,
 					unsigned level)
 {
 	struct pipe_resource *ptex = &rtex->resource.base.b;
-	struct radeon *radeon = (struct radeon *)screen->winsys;
-	enum chip_class chipc = r600_get_family_class(radeon);
 	unsigned width, stride, tile_width;
-	
+
 	if (rtex->pitch_override)
 		return rtex->pitch_override;
 
@@ -167,11 +182,6 @@ static unsigned r600_texture_get_stride(struct pipe_screen *screen,
 		width = align(width, tile_width);
 	}
 	stride = util_format_get_stride(ptex->format, width);
-	if (chipc == EVERGREEN)
-		stride = align(stride, 512);
-
-        if (ptex->bind & PIPE_BIND_RENDER_TARGET)
-		stride = align(stride, 512);
 
 	return stride;
 }
@@ -193,11 +203,10 @@ static unsigned r600_texture_get_nblocksy(struct pipe_screen *screen,
 }
 
 /* Get a width in pixels from a stride in bytes. */
-static unsigned pitch_to_width(enum pipe_format format,
-                                unsigned pitch_in_bytes)
+static unsigned pitch_to_width(enum pipe_format format, unsigned pitch_in_bytes)
 {
-    return (pitch_in_bytes / util_format_get_blocksize(format)) *
-            util_format_get_blockwidth(format);
+	return (pitch_in_bytes / util_format_get_blocksize(format)) *
+		util_format_get_blockwidth(format);
 }
 
 static void r600_texture_set_array_mode(struct pipe_screen *screen,
@@ -257,6 +266,9 @@ static void r600_setup_miptree(struct pipe_screen *screen,
 		}
 		else
 			size = layer_size * u_minify(ptex->depth0, i);
+		/* align base image and start of miptree */
+		if ((i == 0) || (i == 1))
+			offset = align(offset, r600_get_base_alignment(screen, ptex->format, array_mode));
 		rtex->offset[i] = offset;
 		rtex->layer_size[i] = layer_size;
 		rtex->pitch_in_bytes[i] = pitch;
@@ -297,7 +309,10 @@ r600_texture_create_object(struct pipe_screen *screen,
 	resource->size = rtex->size;
 
 	if (!resource->bo) {
-		resource->bo = r600_bo(radeon, rtex->size, 4096, base->bind, base->usage);
+		struct pipe_resource *ptex = &rtex->resource.base.b;
+		int base_align = r600_get_base_alignment(screen, ptex->format, array_mode);
+
+		resource->bo = r600_bo(radeon, rtex->size, base_align, base->bind, base->usage);
 		if (!resource->bo) {
 			FREE(rtex);
 			return NULL;
@@ -310,12 +325,12 @@ struct pipe_resource *r600_texture_create(struct pipe_screen *screen,
 						const struct pipe_resource *templ)
 {
 	unsigned array_mode = 0;
-        static int force_tiling = -1;
+	static int force_tiling = -1;
 
-        /* Would like some magic "get_bool_option_once" routine.
+	/* Would like some magic "get_bool_option_once" routine.
 	 */
 	if (force_tiling == -1)
-                force_tiling = debug_get_bool_option("R600_FORCE_TILING", FALSE);
+		force_tiling = debug_get_bool_option("R600_FORCE_TILING", FALSE);
 
 	if (force_tiling) {
 		if (!(templ->flags & R600_RESOURCE_FLAG_TRANSFER) &&
@@ -346,8 +361,8 @@ static void r600_texture_destroy(struct pipe_screen *screen,
 }
 
 static boolean r600_texture_get_handle(struct pipe_screen* screen,
-                                       struct pipe_resource *ptex,
-                                       struct winsys_handle *whandle)
+					struct pipe_resource *ptex,
+					struct winsys_handle *whandle)
 {
 	struct r600_resource_texture *rtex = (struct r600_resource_texture*)ptex;
 	struct r600_resource *resource = &rtex->resource;
@@ -357,36 +372,39 @@ static boolean r600_texture_get_handle(struct pipe_screen* screen,
 			rtex->pitch_in_bytes[0], whandle);
 }
 
-static struct pipe_surface *r600_get_tex_surface(struct pipe_screen *screen,
+static struct pipe_surface *r600_create_surface(struct pipe_context *pipe,
 						struct pipe_resource *texture,
-						unsigned face, unsigned level,
-						unsigned zslice, unsigned flags)
+						const struct pipe_surface *surf_tmpl)
 {
 	struct r600_resource_texture *rtex = (struct r600_resource_texture*)texture;
 	struct r600_surface *surface = CALLOC_STRUCT(r600_surface);
-	unsigned offset, tile_height;
+	unsigned tile_height;
+	unsigned level = surf_tmpl->u.tex.level;
 
+	assert(surf_tmpl->u.tex.first_layer == surf_tmpl->u.tex.last_layer);
 	if (surface == NULL)
 		return NULL;
-	offset = r600_texture_get_offset(rtex, level, zslice, face);
+	/* XXX no offset */
+/*	offset = r600_texture_get_offset(rtex, level, surf_tmpl->u.tex.first_layer);*/
 	pipe_reference_init(&surface->base.reference, 1);
 	pipe_resource_reference(&surface->base.texture, texture);
-	surface->base.format = texture->format;
+	surface->base.context = pipe;
+	surface->base.format = surf_tmpl->format;
 	surface->base.width = mip_minify(texture->width0, level);
 	surface->base.height = mip_minify(texture->height0, level);
-	surface->base.offset = offset;
-	surface->base.usage = flags;
-	surface->base.zslice = zslice;
+	surface->base.usage = surf_tmpl->usage;
 	surface->base.texture = texture;
-	surface->base.face = face;
-	surface->base.level = level;
+	surface->base.u.tex.first_layer = surf_tmpl->u.tex.first_layer;
+	surface->base.u.tex.last_layer = surf_tmpl->u.tex.last_layer;
+	surface->base.u.tex.level = level;
 
-	tile_height = r600_get_height_alignment(screen, rtex->array_mode[level]);
+	tile_height = r600_get_height_alignment(pipe->screen, rtex->array_mode[level]);
 	surface->aligned_height = align(surface->base.height, tile_height);
 	return &surface->base;
 }
 
-static void r600_tex_surface_destroy(struct pipe_surface *surface)
+static void r600_surface_destroy(struct pipe_context *pipe,
+				 struct pipe_surface *surface)
 {
 	pipe_resource_reference(&surface->texture, NULL);
 	FREE(surface);
@@ -419,7 +437,7 @@ struct pipe_resource *r600_texture_from_handle(struct pipe_screen *screen,
 
 static unsigned int r600_texture_is_referenced(struct pipe_context *context,
 						struct pipe_resource *texture,
-						unsigned face, unsigned level)
+						unsigned level, int layer)
 {
 	/* FIXME */
 	return PIPE_REFERENCED_FOR_READ | PIPE_REFERENCED_FOR_WRITE;
@@ -466,7 +484,7 @@ out:
  */
 static INLINE unsigned u_box_volume( const struct pipe_box *box )
 {
-        return box->width * box->depth * box->height;
+	return box->width * box->depth * box->height;
 };
 
 
@@ -474,44 +492,44 @@ static INLINE unsigned u_box_volume( const struct pipe_box *box )
  * If so, don't use a staging resource.
  */
 static boolean permit_hardware_blit(struct pipe_screen *screen,
-                                    struct pipe_resource *res)
+					struct pipe_resource *res)
 {
-        unsigned bind;
+	unsigned bind;
 
-        if (util_format_is_depth_or_stencil(res->format))
-                bind = PIPE_BIND_DEPTH_STENCIL;
-        else
-                bind = PIPE_BIND_RENDER_TARGET;
+	if (util_format_is_depth_or_stencil(res->format))
+		bind = PIPE_BIND_DEPTH_STENCIL;
+	else
+		bind = PIPE_BIND_RENDER_TARGET;
 
 	/* See r600_resource_copy_region: there is something wrong
-         * with depth resource copies at the moment so avoid them for
-         * now.
-         */
+	 * with depth resource copies at the moment so avoid them for
+	 * now.
+	 */
 	if (util_format_get_component_bits(res->format,
-                                           UTIL_FORMAT_COLORSPACE_ZS,
-                                           0) != 0)
-                return FALSE;
-
-        if (!screen->is_format_supported(screen,
-                                         res->format,
-                                         res->target,
-                                         res->nr_samples,
-                                         bind, 0))
-                return FALSE;
-
-        if (!screen->is_format_supported(screen,
-                                         res->format,
-                                         res->target,
-                                         res->nr_samples,
-                                         PIPE_BIND_SAMPLER_VIEW, 0))
-                return FALSE;
-
-        return TRUE;
+				UTIL_FORMAT_COLORSPACE_ZS,
+				0) != 0)
+		return FALSE;
+
+	if (!screen->is_format_supported(screen,
+				res->format,
+				res->target,
+				res->nr_samples,
+				bind, 0))
+		return FALSE;
+
+	if (!screen->is_format_supported(screen,
+				res->format,
+				res->target,
+				res->nr_samples,
+				PIPE_BIND_SAMPLER_VIEW, 0))
+		return FALSE;
+
+	return TRUE;
 }
 
 struct pipe_transfer* r600_texture_get_transfer(struct pipe_context *ctx,
 						struct pipe_resource *texture,
-						struct pipe_subresource sr,
+						unsigned level,
 						unsigned usage,
 						const struct pipe_box *box)
 {
@@ -531,37 +549,36 @@ struct pipe_transfer* r600_texture_get_transfer(struct pipe_context *ctx,
 	if (rtex->tiled)
 		use_staging_texture = TRUE;
 
-	if ((usage & PIPE_TRANSFER_READ) &&
-            u_box_volume(box) > 1024)
-                use_staging_texture = TRUE;
-
-        /* XXX: Use a staging texture for uploads if the underlying BO
-         * is busy.  No interface for checking that currently? so do
-         * it eagerly whenever the transfer doesn't require a readback
-         * and might block.
-         */
-        if ((usage & PIPE_TRANSFER_WRITE) &&
-            !(usage & (PIPE_TRANSFER_READ |
-                       PIPE_TRANSFER_DONTBLOCK |
-                       PIPE_TRANSFER_UNSYNCHRONIZED)))
-                use_staging_texture = TRUE;
-
-        if (!permit_hardware_blit(ctx->screen, texture) ||
-            (texture->flags & R600_RESOURCE_FLAG_TRANSFER))
-                use_staging_texture = FALSE;
+	if ((usage & PIPE_TRANSFER_READ) && u_box_volume(box) > 1024)
+		use_staging_texture = TRUE;
+
+	/* XXX: Use a staging texture for uploads if the underlying BO
+	 * is busy.  No interface for checking that currently? so do
+	 * it eagerly whenever the transfer doesn't require a readback
+	 * and might block.
+	 */
+	if ((usage & PIPE_TRANSFER_WRITE) &&
+			!(usage & (PIPE_TRANSFER_READ |
+					PIPE_TRANSFER_DONTBLOCK |
+					PIPE_TRANSFER_UNSYNCHRONIZED)))
+		use_staging_texture = TRUE;
+
+	if (!permit_hardware_blit(ctx->screen, texture) ||
+		(texture->flags & R600_RESOURCE_FLAG_TRANSFER))
+		use_staging_texture = FALSE;
 
 	trans = CALLOC_STRUCT(r600_transfer);
 	if (trans == NULL)
 		return NULL;
 	pipe_resource_reference(&trans->transfer.resource, texture);
-	trans->transfer.sr = sr;
+	trans->transfer.level = level;
 	trans->transfer.usage = usage;
 	trans->transfer.box = *box;
 	if (rtex->depth) {
-                /* XXX: only readback the rectangle which is being mapped?
-                 */
-                /* XXX: when discard is true, no need to read back from depth texture
-                 */
+		/* XXX: only readback the rectangle which is being mapped?
+		*/
+		/* XXX: when discard is true, no need to read back from depth texture
+		*/
 		r = r600_texture_depth_flush(ctx, texture);
 		if (r < 0) {
 			R600_ERR("failed to create temporary texture to hold untiled copy\n");
@@ -575,6 +592,7 @@ struct pipe_transfer* r600_texture_get_transfer(struct pipe_context *ctx,
 		resource.width0 = box->width;
 		resource.height0 = box->height;
 		resource.depth0 = 1;
+		resource.array_size = 1;
 		resource.last_level = 0;
 		resource.nr_samples = 0;
 		resource.usage = PIPE_USAGE_STAGING;
@@ -600,7 +618,7 @@ struct pipe_transfer* r600_texture_get_transfer(struct pipe_context *ctx,
 		}
 
 		trans->transfer.stride =
-                        ((struct r600_resource_texture *)trans->staging_texture)->pitch_in_bytes[0];
+			((struct r600_resource_texture *)trans->staging_texture)->pitch_in_bytes[0];
 		if (usage & PIPE_TRANSFER_READ) {
 			r600_copy_to_staging_texture(ctx, trans);
 			/* Always referenced in the blit. */
@@ -608,8 +626,8 @@ struct pipe_transfer* r600_texture_get_transfer(struct pipe_context *ctx,
 		}
 		return &trans->transfer;
 	}
-	trans->transfer.stride = rtex->pitch_in_bytes[sr.level];
-	trans->offset = r600_texture_get_offset(rtex, sr.level, box->z, sr.face);
+	trans->transfer.stride = rtex->pitch_in_bytes[level];
+	trans->offset = r600_texture_get_offset(rtex, level, box->z);
 	return &trans->transfer;
 }
 
@@ -722,10 +740,10 @@ struct u_resource_vtbl r600_texture_vtbl =
 	u_default_transfer_inline_write	/* transfer_inline_write */
 };
 
-void r600_init_screen_texture_functions(struct pipe_screen *screen)
+void r600_init_surface_functions(struct r600_pipe_context *r600)
 {
-	screen->get_tex_surface = r600_get_tex_surface;
-	screen->tex_surface_destroy = r600_tex_surface_destroy;
+	r600->context.create_surface = r600_create_surface;
+	r600->context.surface_destroy = r600_surface_destroy;
 }
 
 static unsigned r600_get_swizzle_combined(const unsigned char *swizzle_format,
@@ -826,8 +844,8 @@ uint32_t r600_translate_texformat(enum pipe_format format,
 	case UTIL_FORMAT_COLORSPACE_YUV:
 		yuv_format |= (1 << 30);
 		switch (format) {
-                case PIPE_FORMAT_UYVY:
-                case PIPE_FORMAT_YUYV:
+		case PIPE_FORMAT_UYVY:
+		case PIPE_FORMAT_YUYV:
 		default:
 			break;
 		}
@@ -845,29 +863,29 @@ uint32_t r600_translate_texformat(enum pipe_format format,
 
 	/* S3TC formats. TODO */
 	if (desc->layout == UTIL_FORMAT_LAYOUT_S3TC) {
-                static int r600_enable_s3tc = -1;
+		static int r600_enable_s3tc = -1;
 
-                if (r600_enable_s3tc == -1)
-                        r600_enable_s3tc = 
-                                debug_get_bool_option("R600_ENABLE_S3TC", FALSE);
+		if (r600_enable_s3tc == -1)
+			r600_enable_s3tc = 
+				debug_get_bool_option("R600_ENABLE_S3TC", FALSE);
 
-                if (!r600_enable_s3tc)
-                        goto out_unknown;
+		if (!r600_enable_s3tc)
+			goto out_unknown;
 
 		switch (format) {
 		case PIPE_FORMAT_DXT1_RGB:
 		case PIPE_FORMAT_DXT1_RGBA:
-                        result = FMT_BC1;
-                        goto out_word4;
+			result = FMT_BC1;
+			goto out_word4;
 		case PIPE_FORMAT_DXT3_RGBA:
-                        result = FMT_BC2;
-                        goto out_word4;
+			result = FMT_BC2;
+			goto out_word4;
 		case PIPE_FORMAT_DXT5_RGBA:
-                        result = FMT_BC3;
-                        goto out_word4;
-                default:
-                        goto out_unknown;
-                }
+			result = FMT_BC3;
+			goto out_word4;
+		default:
+			goto out_unknown;
+		}
 	}
 
 
diff --git a/src/gallium/drivers/r600/r600_translate.c b/src/gallium/drivers/r600/r600_translate.c
index 9a07cf2073..ba12eee2b5 100644
--- a/src/gallium/drivers/r600/r600_translate.c
+++ b/src/gallium/drivers/r600/r600_translate.c
@@ -41,6 +41,8 @@ void r600_begin_vertex_translate(struct r600_pipe_context *rctx)
 	struct pipe_transfer *vb_transfer[PIPE_MAX_ATTRIBS] = {0}, *out_transfer;
 	struct pipe_resource *out_buffer;
 	unsigned i, num_verts;
+	struct pipe_vertex_element new_velems[PIPE_MAX_ATTRIBS];
+	void *tmp;
 
 	/* Initialize the translate key, i.e. the recipe how vertices should be
 	 * translated. */
@@ -51,9 +53,7 @@ void r600_begin_vertex_translate(struct r600_pipe_context *rctx)
 		unsigned output_format_size = ve->hw_format_size[i];
 
 		/* Check for support. */
-		if (ve->elements[i].src_format == ve->hw_format[i] &&
-		    (vb->buffer_offset + ve->elements[i].src_offset) % 4 == 0 &&
-		    vb->stride % 4 == 0) {
+		if (ve->elements[i].src_format == ve->hw_format[i]) {
 			continue;
 		}
 
@@ -147,29 +147,23 @@ void r600_begin_vertex_translate(struct r600_pipe_context *rctx)
 	}
 
 	/* Save and replace vertex elements. */
-	{
-		struct pipe_vertex_element new_velems[PIPE_MAX_ATTRIBS];
-
-		rctx->tran.saved_velems = rctx->vertex_elements;
-
-		for (i = 0; i < ve->count; i++) {
-			if (vb_translated[ve->elements[i].vertex_buffer_index]) {
-				te = &key.element[tr_elem_index[i]];
-				new_velems[i].instance_divisor = ve->elements[i].instance_divisor;
-				new_velems[i].src_format = te->output_format;
-				new_velems[i].src_offset = te->output_offset;
-				new_velems[i].vertex_buffer_index = rctx->tran.vb_slot;
-			} else {
-				memcpy(&new_velems[i], &ve->elements[i],
-				       sizeof(struct pipe_vertex_element));
-			}
+	for (i = 0; i < ve->count; i++) {
+		if (vb_translated[ve->elements[i].vertex_buffer_index]) {
+			te = &key.element[tr_elem_index[i]];
+			new_velems[i].instance_divisor = ve->elements[i].instance_divisor;
+			new_velems[i].src_format = te->output_format;
+			new_velems[i].src_offset = te->output_offset;
+			new_velems[i].vertex_buffer_index = rctx->tran.vb_slot;
+		} else {
+			memcpy(&new_velems[i], &ve->elements[i],
+					sizeof(struct pipe_vertex_element));
 		}
-
-		rctx->tran.new_velems =
-			pipe->create_vertex_elements_state(pipe, ve->count, new_velems);
-		pipe->bind_vertex_elements_state(pipe, rctx->tran.new_velems);
 	}
 
+	tmp = pipe->create_vertex_elements_state(pipe, ve->count, new_velems);
+	pipe->bind_vertex_elements_state(pipe, tmp);
+	rctx->tran.new_velems = tmp;
+
 	pipe_resource_reference(&out_buffer, NULL);
 }
 
@@ -177,13 +171,15 @@ void r600_end_vertex_translate(struct r600_pipe_context *rctx)
 {
 	struct pipe_context *pipe = &rctx->context;
 
+	if (rctx->tran.new_velems == NULL) {
+		return;
+	}
 	/* Restore vertex elements. */
-	pipe->bind_vertex_elements_state(pipe, rctx->tran.saved_velems);
 	pipe->delete_vertex_elements_state(pipe, rctx->tran.new_velems);
+	rctx->tran.new_velems = NULL;
 
 	/* Delete the now-unused VBO. */
-	pipe_resource_reference(&rctx->vertex_buffer[rctx->tran.vb_slot].buffer,
-				NULL);
+	pipe_resource_reference(&rctx->vertex_buffer[rctx->tran.vb_slot].buffer, NULL);
 }
 
 void r600_translate_index_buffer(struct r600_pipe_context *r600,
@@ -197,14 +193,7 @@ void r600_translate_index_buffer(struct r600_pipe_context *r600,
 		*index_size = 2;
 		*start = 0;
 		break;
-
 	case 2:
-		if (*start % 2 != 0) {
-			util_rebuild_ushort_elts(&r600->context, index_buffer, 0, *start, count);
-			*start = 0;
-		}
-		break;
-
 	case 4:
 		break;
 	}
diff --git a/src/gallium/drivers/r600/r600_upload.c b/src/gallium/drivers/r600/r600_upload.c
new file mode 100644
index 0000000000..44102ff55b
--- /dev/null
+++ b/src/gallium/drivers/r600/r600_upload.c
@@ -0,0 +1,114 @@
+/*
+ * Copyright 2010 Jerome Glisse <glisse@freedesktop.org>
+ *
+ * Permission is hereby granted, free of charge, to 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
+ * THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
+ * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
+ * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
+ * USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors:
+ *      Jerome Glisse <jglisse@redhat.com>
+ */
+#include <errno.h>
+#include "util/u_inlines.h"
+#include "util/u_memory.h"
+#include "r600.h"
+#include "r600_pipe.h"
+#include "r600_resource.h"
+
+struct r600_upload {
+	struct r600_pipe_context	*rctx;
+	struct r600_bo			*buffer;
+	char				*ptr;
+	unsigned			size;
+	unsigned			default_size;
+	unsigned			total_alloc_size;
+	unsigned			offset;
+	unsigned			alignment;
+};
+
+struct r600_upload *r600_upload_create(struct r600_pipe_context *rctx,
+					unsigned default_size,
+					unsigned alignment)
+{
+	struct r600_upload *upload = CALLOC_STRUCT(r600_upload);
+
+	if (upload == NULL)
+		return NULL;
+
+	upload->rctx = rctx;
+	upload->size = 0;
+	upload->default_size = default_size;
+	upload->alignment = alignment;
+	upload->ptr = NULL;
+	upload->buffer = NULL;
+	upload->total_alloc_size = 0;
+
+	return upload;
+}
+
+void r600_upload_flush(struct r600_upload *upload)
+{
+	if (upload->buffer) {
+		r600_bo_reference(upload->rctx->radeon, &upload->buffer, NULL);
+	}
+	upload->default_size = MAX2(upload->total_alloc_size, upload->default_size);
+	upload->total_alloc_size = 0;
+	upload->size = 0;
+	upload->offset = 0;
+	upload->ptr = NULL;
+	upload->buffer = NULL;
+}
+
+void r600_upload_destroy(struct r600_upload *upload)
+{
+	r600_upload_flush(upload);
+	FREE(upload);
+}
+
+int r600_upload_buffer(struct r600_upload *upload, unsigned offset,
+			unsigned size, struct r600_resource_buffer *in_buffer,
+			unsigned *out_offset, unsigned *out_size,
+			struct r600_bo **out_buffer)
+{
+	unsigned alloc_size = align(size, upload->alignment);
+	const void *in_ptr = NULL;
+
+	if (upload->offset + alloc_size > upload->size) {
+		if (upload->size) {
+			r600_bo_reference(upload->rctx->radeon, &upload->buffer, NULL);
+		}
+		upload->size = align(MAX2(upload->default_size, alloc_size), 4096);
+		upload->total_alloc_size += upload->size;
+		upload->offset = 0;
+		upload->buffer = r600_bo(upload->rctx->radeon, upload->size, 4096, PIPE_BIND_VERTEX_BUFFER, 0);
+		if (upload->buffer == NULL) {
+			return -ENOMEM;
+		}
+		upload->ptr = r600_bo_map(upload->rctx->radeon, upload->buffer, 0, NULL);
+	}
+
+	in_ptr = in_buffer->user_buffer;
+	memcpy(upload->ptr + upload->offset, (uint8_t *) in_ptr + offset, size);
+	*out_offset = upload->offset;
+	*out_size = upload->size;
+	*out_buffer = NULL;
+	r600_bo_reference(upload->rctx->radeon, out_buffer, upload->buffer);
+	upload->offset += alloc_size;
+
+	return 0;
+}
diff --git a/src/gallium/drivers/rbug/rbug_context.c b/src/gallium/drivers/rbug/rbug_context.c
index 413da59e55..94e57e40f8 100644
--- a/src/gallium/drivers/rbug/rbug_context.c
+++ b/src/gallium/drivers/rbug/rbug_context.c
@@ -707,17 +707,13 @@ rbug_set_sample_mask(struct pipe_context *_pipe,
 static void
 rbug_resource_copy_region(struct pipe_context *_pipe,
                           struct pipe_resource *_dst,
-                          struct pipe_subresource subdst,
+                          unsigned dst_level,
                           unsigned dstx,
                           unsigned dsty,
                           unsigned dstz,
                           struct pipe_resource *_src,
-                          struct pipe_subresource subsrc,
-                          unsigned srcx,
-                          unsigned srcy,
-                          unsigned srcz,
-                          unsigned width,
-                          unsigned height)
+                          unsigned src_level,
+                          const struct pipe_box *src_box)
 {
    struct rbug_context *rb_pipe = rbug_context(_pipe);
    struct rbug_resource *rb_resource_dst = rbug_resource(_dst);
@@ -728,17 +724,13 @@ rbug_resource_copy_region(struct pipe_context *_pipe,
 
    pipe->resource_copy_region(pipe,
                               dst,
-                              subdst,
+                              dst_level,
                               dstx,
                               dsty,
                               dstz,
                               src,
-                              subsrc,
-                              srcx,
-                              srcy,
-                              srcz,
-                              width,
-                              height);
+                              src_level,
+                              src_box);
 }
 
 static void
@@ -820,8 +812,8 @@ rbug_flush(struct pipe_context *_pipe,
 static unsigned int
 rbug_is_resource_referenced(struct pipe_context *_pipe,
                             struct pipe_resource *_resource,
-                            unsigned face,
-                            unsigned level)
+                            unsigned level,
+                            int layer)
 {
    struct rbug_context *rb_pipe = rbug_context(_pipe);
    struct rbug_resource *rb_resource = rbug_resource(_resource);
@@ -830,8 +822,8 @@ rbug_is_resource_referenced(struct pipe_context *_pipe,
 
    return pipe->is_resource_referenced(pipe,
                                        resource,
-                                       face,
-                                       level);
+                                       level,
+                                       layer);
 }
 
 static struct pipe_sampler_view *
@@ -862,10 +854,40 @@ rbug_context_sampler_view_destroy(struct pipe_context *_pipe,
                              rbug_sampler_view(_view));
 }
 
+static struct pipe_surface *
+rbug_context_create_surface(struct pipe_context *_pipe,
+                            struct pipe_resource *_resource,
+                            const struct pipe_surface *surf_tmpl)
+{
+   struct rbug_context *rb_pipe = rbug_context(_pipe);
+   struct rbug_resource *rb_resource = rbug_resource(_resource);
+   struct pipe_context *pipe = rb_pipe->pipe;
+   struct pipe_resource *resource = rb_resource->resource;
+   struct pipe_surface *result;
+
+   result = pipe->create_surface(pipe,
+                                 resource,
+                                 surf_tmpl);
+
+   if (result)
+      return rbug_surface_create(rb_pipe, rb_resource, result);
+   return NULL;
+}
+
+static void
+rbug_context_surface_destroy(struct pipe_context *_pipe,
+                             struct pipe_surface *_surface)
+{
+   rbug_surface_destroy(rbug_context(_pipe),
+                        rbug_surface(_surface));
+}
+
+
+
 static struct pipe_transfer *
 rbug_context_get_transfer(struct pipe_context *_context,
                           struct pipe_resource *_resource,
-                          struct pipe_subresource sr,
+                          unsigned level,
                           unsigned usage,
                           const struct pipe_box *box)
 {
@@ -877,7 +899,7 @@ rbug_context_get_transfer(struct pipe_context *_context,
 
    result = context->get_transfer(context,
                                   resource,
-                                  sr,
+                                  level,
                                   usage,
                                   box);
 
@@ -942,12 +964,12 @@ rbug_context_transfer_unmap(struct pipe_context *_context,
 static void
 rbug_context_transfer_inline_write(struct pipe_context *_context,
                                    struct pipe_resource *_resource,
-                                   struct pipe_subresource sr,
+                                   unsigned level,
                                    unsigned usage,
                                    const struct pipe_box *box,
                                    const void *data,
                                    unsigned stride,
-                                   unsigned slice_stride)
+                                   unsigned layer_stride)
 {
    struct rbug_context *rb_pipe = rbug_context(_context);
    struct rbug_resource *rb_resource = rbug_resource(_resource);
@@ -956,12 +978,12 @@ rbug_context_transfer_inline_write(struct pipe_context *_context,
 
    context->transfer_inline_write(context,
                                   resource,
-                                  sr,
+                                  level,
                                   usage,
                                   box,
                                   data,
                                   stride,
-                                  slice_stride);
+                                  layer_stride);
 }
 
 
@@ -1042,6 +1064,8 @@ rbug_context_create(struct pipe_screen *_screen, struct pipe_context *pipe)
    rb_pipe->base.is_resource_referenced = rbug_is_resource_referenced;
    rb_pipe->base.create_sampler_view = rbug_context_create_sampler_view;
    rb_pipe->base.sampler_view_destroy = rbug_context_sampler_view_destroy;
+   rb_pipe->base.create_surface = rbug_context_create_surface;
+   rb_pipe->base.surface_destroy = rbug_context_surface_destroy;
    rb_pipe->base.get_transfer = rbug_context_get_transfer;
    rb_pipe->base.transfer_destroy = rbug_context_transfer_destroy;
    rb_pipe->base.transfer_map = rbug_context_transfer_map;
diff --git a/src/gallium/drivers/rbug/rbug_core.c b/src/gallium/drivers/rbug/rbug_core.c
index 9dc663b079..eb772d19d0 100644
--- a/src/gallium/drivers/rbug/rbug_core.c
+++ b/src/gallium/drivers/rbug/rbug_core.c
@@ -266,9 +266,9 @@ rbug_texture_read(struct rbug_rbug *tr_rbug, struct rbug_header *header, uint32_
 
    tex = tr_tex->resource;
    t = pipe_get_transfer(context, tex,
-				 gptr->face, gptr->level, gptr->zslice,
-				 PIPE_TRANSFER_READ,
-				 gptr->x, gptr->y, gptr->w, gptr->h);
+                         gptr->level, gptr->face + gptr->zslice,
+                         PIPE_TRANSFER_READ,
+                         gptr->x, gptr->y, gptr->w, gptr->h);
 
    map = context->transfer_map(context, t);
 
@@ -279,7 +279,7 @@ rbug_texture_read(struct rbug_rbug *tr_rbug, struct rbug_header *header, uint32_
                                 util_format_get_blocksize(t->resource->format),
                                 (uint8_t*)map,
                                 t->stride * util_format_get_nblocksy(t->resource->format,
-								     t->box.height),
+                                                                     t->box.height),
                                 t->stride,
                                 NULL);
 
diff --git a/src/gallium/drivers/rbug/rbug_objects.c b/src/gallium/drivers/rbug/rbug_objects.c
index 0979fcff95..7d7cc482ae 100644
--- a/src/gallium/drivers/rbug/rbug_objects.c
+++ b/src/gallium/drivers/rbug/rbug_objects.c
@@ -79,7 +79,8 @@ rbug_resource_destroy(struct rbug_resource *rb_resource)
 
 
 struct pipe_surface *
-rbug_surface_create(struct rbug_resource *rb_resource,
+rbug_surface_create(struct rbug_context *rb_context,
+                    struct rbug_resource *rb_resource,
                     struct pipe_surface *surface)
 {
    struct rbug_surface *rb_surface;
@@ -108,10 +109,12 @@ error:
 }
 
 void
-rbug_surface_destroy(struct rbug_surface *rb_surface)
+rbug_surface_destroy(struct rbug_context *rb_context,
+                     struct rbug_surface *rb_surface)
 {
    pipe_resource_reference(&rb_surface->base.texture, NULL);
-   pipe_surface_reference(&rb_surface->surface, NULL);
+   rb_context->pipe->surface_destroy(rb_context->pipe,
+                                     rb_surface->surface);
    FREE(rb_surface);
 }
 
diff --git a/src/gallium/drivers/rbug/rbug_objects.h b/src/gallium/drivers/rbug/rbug_objects.h
index 49c128d3d1..3fba333422 100644
--- a/src/gallium/drivers/rbug/rbug_objects.h
+++ b/src/gallium/drivers/rbug/rbug_objects.h
@@ -189,11 +189,13 @@ void
 rbug_resource_destroy(struct rbug_resource *rb_resource);
 
 struct pipe_surface *
-rbug_surface_create(struct rbug_resource *rb_resource,
+rbug_surface_create(struct rbug_context *rb_context,
+                    struct rbug_resource *rb_resource,
                     struct pipe_surface *surface);
 
 void
-rbug_surface_destroy(struct rbug_surface *rb_surface);
+rbug_surface_destroy(struct rbug_context *rb_context,
+                     struct rbug_surface *rb_surface);
 
 struct pipe_sampler_view *
 rbug_sampler_view_create(struct rbug_context *rb_context,
diff --git a/src/gallium/drivers/rbug/rbug_screen.c b/src/gallium/drivers/rbug/rbug_screen.c
index 961df482c2..d635ce575c 100644
--- a/src/gallium/drivers/rbug/rbug_screen.c
+++ b/src/gallium/drivers/rbug/rbug_screen.c
@@ -188,40 +188,6 @@ rbug_screen_resource_destroy(struct pipe_screen *screen,
    rbug_resource_destroy(rbug_resource(_resource));
 }
 
-static struct pipe_surface *
-rbug_screen_get_tex_surface(struct pipe_screen *_screen,
-                            struct pipe_resource *_resource,
-                            unsigned face,
-                            unsigned level,
-                            unsigned zslice,
-                            unsigned usage)
-{
-   struct rbug_screen *rb_screen = rbug_screen(_screen);
-   struct rbug_resource *rb_resource = rbug_resource(_resource);
-   struct pipe_screen *screen = rb_screen->screen;
-   struct pipe_resource *resource = rb_resource->resource;
-   struct pipe_surface *result;
-
-   result = screen->get_tex_surface(screen,
-                                    resource,
-                                    face,
-                                    level,
-                                    zslice,
-                                    usage);
-
-   if (result)
-      return rbug_surface_create(rb_resource, result);
-   return NULL;
-}
-
-static void
-rbug_screen_tex_surface_destroy(struct pipe_surface *_surface)
-{
-   rbug_surface_destroy(rbug_surface(_surface));
-}
-
-
-
 static struct pipe_resource *
 rbug_screen_user_buffer_create(struct pipe_screen *_screen,
                                void *ptr,
@@ -246,16 +212,18 @@ rbug_screen_user_buffer_create(struct pipe_screen *_screen,
 
 static void
 rbug_screen_flush_frontbuffer(struct pipe_screen *_screen,
-                              struct pipe_surface *_surface,
+                              struct pipe_resource *_resource,
+                              unsigned level, unsigned layer,
                               void *context_private)
 {
    struct rbug_screen *rb_screen = rbug_screen(_screen);
-   struct rbug_surface *rb_surface = rbug_surface(_surface);
+   struct rbug_resource *rb_resource = rbug_resource(_resource);
    struct pipe_screen *screen = rb_screen->screen;
-   struct pipe_surface *surface = rb_surface->surface;
+   struct pipe_resource *resource = rb_resource->resource;
 
    screen->flush_frontbuffer(screen,
-                             surface,
+                             resource,
+                             level, layer,
                              context_private);
 }
 
@@ -336,8 +304,6 @@ rbug_screen_create(struct pipe_screen *screen)
    rb_screen->base.resource_from_handle = rbug_screen_resource_from_handle;
    rb_screen->base.resource_get_handle = rbug_screen_resource_get_handle;
    rb_screen->base.resource_destroy = rbug_screen_resource_destroy;
-   rb_screen->base.get_tex_surface = rbug_screen_get_tex_surface;
-   rb_screen->base.tex_surface_destroy = rbug_screen_tex_surface_destroy;
    rb_screen->base.user_buffer_create = rbug_screen_user_buffer_create;
    rb_screen->base.flush_frontbuffer = rbug_screen_flush_frontbuffer;
    rb_screen->base.fence_reference = rbug_screen_fence_reference;
diff --git a/src/gallium/drivers/softpipe/sp_context.c b/src/gallium/drivers/softpipe/sp_context.c
index b5d30bc6fc..e935ce6d21 100644
--- a/src/gallium/drivers/softpipe/sp_context.c
+++ b/src/gallium/drivers/softpipe/sp_context.c
@@ -145,15 +145,15 @@ softpipe_destroy( struct pipe_context *pipe )
  */
 static unsigned int
 softpipe_is_resource_referenced( struct pipe_context *pipe,
-				struct pipe_resource *texture,
-				unsigned face, unsigned level)
+                                 struct pipe_resource *texture,
+                                 unsigned level, int layer)
 {
    struct softpipe_context *softpipe = softpipe_context( pipe );
    unsigned i;
 
    if (texture->target == PIPE_BUFFER)
       return PIPE_UNREFERENCED;
-   
+
    /* check if any of the bound drawing surfaces are this texture */
    if (softpipe->dirty_render_cache) {
       for (i = 0; i < softpipe->framebuffer.nr_cbufs; i++) {
diff --git a/src/gallium/drivers/softpipe/sp_context.h b/src/gallium/drivers/softpipe/sp_context.h
index 9361a3df09..903574b7e1 100644
--- a/src/gallium/drivers/softpipe/sp_context.h
+++ b/src/gallium/drivers/softpipe/sp_context.h
@@ -154,9 +154,9 @@ struct softpipe_context {
 
    /** TGSI exec things */
    struct {
-      struct sp_sampler_varient *geom_samplers_list[PIPE_MAX_GEOMETRY_SAMPLERS];
-      struct sp_sampler_varient *vert_samplers_list[PIPE_MAX_VERTEX_SAMPLERS];
-      struct sp_sampler_varient *frag_samplers_list[PIPE_MAX_SAMPLERS];
+      struct sp_sampler_variant *geom_samplers_list[PIPE_MAX_GEOMETRY_SAMPLERS];
+      struct sp_sampler_variant *vert_samplers_list[PIPE_MAX_VERTEX_SAMPLERS];
+      struct sp_sampler_variant *frag_samplers_list[PIPE_MAX_SAMPLERS];
    } tgsi;
 
    struct tgsi_exec_machine *fs_machine;
@@ -192,7 +192,7 @@ softpipe_context( struct pipe_context *pipe )
 }
 
 void
-softpipe_reset_sampler_varients(struct softpipe_context *softpipe);
+softpipe_reset_sampler_variants(struct softpipe_context *softpipe);
 
 struct pipe_context *
 softpipe_create_context( struct pipe_screen *, void *priv );
diff --git a/src/gallium/drivers/softpipe/sp_flush.c b/src/gallium/drivers/softpipe/sp_flush.c
index 1071011db0..4258395063 100644
--- a/src/gallium/drivers/softpipe/sp_flush.c
+++ b/src/gallium/drivers/softpipe/sp_flush.c
@@ -120,8 +120,8 @@ softpipe_flush( struct pipe_context *pipe,
 boolean
 softpipe_flush_resource(struct pipe_context *pipe,
                         struct pipe_resource *texture,
-                        unsigned face,
                         unsigned level,
+                        int layer,
                         unsigned flush_flags,
                         boolean read_only,
                         boolean cpu_access,
@@ -129,7 +129,7 @@ softpipe_flush_resource(struct pipe_context *pipe,
 {
    unsigned referenced;
 
-   referenced = pipe->is_resource_referenced(pipe, texture, face, level);
+   referenced = pipe->is_resource_referenced(pipe, texture, level, layer);
 
    if ((referenced & PIPE_REFERENCED_FOR_WRITE) ||
        ((referenced & PIPE_REFERENCED_FOR_READ) && !read_only)) {
diff --git a/src/gallium/drivers/softpipe/sp_flush.h b/src/gallium/drivers/softpipe/sp_flush.h
index cb97482a71..22a5ceeb9e 100644
--- a/src/gallium/drivers/softpipe/sp_flush.h
+++ b/src/gallium/drivers/softpipe/sp_flush.h
@@ -40,8 +40,8 @@ softpipe_flush(struct pipe_context *pipe, unsigned flags,
 boolean
 softpipe_flush_resource(struct pipe_context *pipe,
                         struct pipe_resource *texture,
-                        unsigned face,
                         unsigned level,
+                        int layer,
                         unsigned flush_flags,
                         boolean read_only,
                         boolean cpu_access,
diff --git a/src/gallium/drivers/softpipe/sp_limits.h b/src/gallium/drivers/softpipe/sp_limits.h
new file mode 100644
index 0000000000..a7a24c98d5
--- /dev/null
+++ b/src/gallium/drivers/softpipe/sp_limits.h
@@ -0,0 +1,42 @@
+/**************************************************************************
+ * 
+ * Copyright 2010 VMware, Inc.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL THE AUTHORS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+#ifndef SP_LIMITS_H
+#define SP_LIMITS_H
+
+
+
+#define SP_MAX_TEXTURE_2D_LEVELS 15  /* 16K x 16K */
+#define SP_MAX_TEXTURE_3D_LEVELS 9   /* 512 x 512 x 512 */
+
+
+/** Max surface size */
+#define MAX_WIDTH (1 << (SP_MAX_TEXTURE_2D_LEVELS - 1))
+#define MAX_HEIGHT (1 << (SP_MAX_TEXTURE_2D_LEVELS - 1))
+
+
+#endif /* SP_LIMITS_H */
diff --git a/src/gallium/drivers/softpipe/sp_screen.c b/src/gallium/drivers/softpipe/sp_screen.c
index 5f171d314a..cbdea19af4 100644
--- a/src/gallium/drivers/softpipe/sp_screen.c
+++ b/src/gallium/drivers/softpipe/sp_screen.c
@@ -266,12 +266,13 @@ softpipe_destroy_screen( struct pipe_screen *screen )
  */
 static void
 softpipe_flush_frontbuffer(struct pipe_screen *_screen,
-                           struct pipe_surface *surface,
+                           struct pipe_resource *resource,
+                           unsigned level, unsigned layer,
                            void *context_private)
 {
    struct softpipe_screen *screen = softpipe_screen(_screen);
    struct sw_winsys *winsys = screen->winsys;
-   struct softpipe_resource *texture = softpipe_resource(surface->texture);
+   struct softpipe_resource *texture = softpipe_resource(resource);
 
    assert(texture->dt);
    if (texture->dt)
diff --git a/src/gallium/drivers/softpipe/sp_state_derived.c b/src/gallium/drivers/softpipe/sp_state_derived.c
index 3ba4d934fd..bf4c12701a 100644
--- a/src/gallium/drivers/softpipe/sp_state_derived.c
+++ b/src/gallium/drivers/softpipe/sp_state_derived.c
@@ -197,7 +197,7 @@ update_tgsi_samplers( struct softpipe_context *softpipe )
 {
    unsigned i;
 
-   softpipe_reset_sampler_varients( softpipe );
+   softpipe_reset_sampler_variants( softpipe );
 
    for (i = 0; i < PIPE_MAX_SAMPLERS; i++) {
       struct softpipe_tex_tile_cache *tc = softpipe->tex_cache[i];
diff --git a/src/gallium/drivers/softpipe/sp_state_sampler.c b/src/gallium/drivers/softpipe/sp_state_sampler.c
index b59fbc33ed..cfa211b60a 100644
--- a/src/gallium/drivers/softpipe/sp_state_sampler.c
+++ b/src/gallium/drivers/softpipe/sp_state_sampler.c
@@ -43,8 +43,8 @@
 
 struct sp_sampler {
    struct pipe_sampler_state base;
-   struct sp_sampler_varient *varients;
-   struct sp_sampler_varient *current;
+   struct sp_sampler_variant *variants;
+   struct sp_sampler_variant *current;
 };
 
 static struct sp_sampler *sp_sampler( struct pipe_sampler_state *sampler )
@@ -60,7 +60,7 @@ softpipe_create_sampler_state(struct pipe_context *pipe,
    struct sp_sampler *sp_sampler = CALLOC_STRUCT(sp_sampler);
 
    sp_sampler->base = *sampler;
-   sp_sampler->varients = NULL;
+   sp_sampler->variants = NULL;
 
    return (void *)sp_sampler;
 }
@@ -277,23 +277,24 @@ softpipe_set_geometry_sampler_views(struct pipe_context *pipe,
 
 
 /**
- * Find/create an sp_sampler_varient object for sampling the given texture,
+ * Find/create an sp_sampler_variant object for sampling the given texture,
  * sampler and tex unit.
  *
  * Note that the tex unit is significant.  We can't re-use a sampler
- * varient for multiple texture units because the sampler varient contains
+ * variant for multiple texture units because the sampler variant contains
  * the texture object pointer.  If the texture object pointer were stored
- * somewhere outside the sampler varient, we could re-use samplers for
+ * somewhere outside the sampler variant, we could re-use samplers for
  * multiple texture units.
  */
-static struct sp_sampler_varient *
-get_sampler_varient( unsigned unit,
+static struct sp_sampler_variant *
+get_sampler_variant( unsigned unit,
                      struct sp_sampler *sampler,
+                     struct pipe_sampler_view *view,
                      struct pipe_resource *resource,
                      unsigned processor )
 {
    struct softpipe_resource *sp_texture = softpipe_resource(resource);
-   struct sp_sampler_varient *v = NULL;
+   struct sp_sampler_variant *v = NULL;
    union sp_sampler_key key;
 
    /* if this fails, widen the key.unit field and update this assertion */
@@ -303,6 +304,10 @@ get_sampler_varient( unsigned unit,
    key.bits.is_pot = sp_texture->pot;
    key.bits.processor = processor;
    key.bits.unit = unit;
+   key.bits.swizzle_r = view->swizzle_r;
+   key.bits.swizzle_g = view->swizzle_g;
+   key.bits.swizzle_b = view->swizzle_b;
+   key.bits.swizzle_a = view->swizzle_a;
    key.bits.pad = 0;
 
    if (sampler->current && 
@@ -311,14 +316,14 @@ get_sampler_varient( unsigned unit,
    }
 
    if (v == NULL) {
-      for (v = sampler->varients; v; v = v->next)
+      for (v = sampler->variants; v; v = v->next)
          if (v->key.value == key.value)
             break;
 
       if (v == NULL) {
-         v = sp_create_sampler_varient( &sampler->base, key );
-         v->next = sampler->varients;
-         sampler->varients = v;
+         v = sp_create_sampler_variant( &sampler->base, key );
+         v->next = sampler->variants;
+         sampler->variants = v;
       }
    }
    
@@ -328,7 +333,7 @@ get_sampler_varient( unsigned unit,
 
 
 void
-softpipe_reset_sampler_varients(struct softpipe_context *softpipe)
+softpipe_reset_sampler_variants(struct softpipe_context *softpipe)
 {
    int i;
 
@@ -345,12 +350,13 @@ softpipe_reset_sampler_varients(struct softpipe_context *softpipe)
          }
 
          softpipe->tgsi.vert_samplers_list[i] = 
-            get_sampler_varient( i,
+            get_sampler_variant( i,
                                  sp_sampler(softpipe->vertex_samplers[i]),
+                                 softpipe->vertex_sampler_views[i],
                                  texture,
                                  TGSI_PROCESSOR_VERTEX );
 
-         sp_sampler_varient_bind_texture( softpipe->tgsi.vert_samplers_list[i], 
+         sp_sampler_variant_bind_texture( softpipe->tgsi.vert_samplers_list[i], 
                                           softpipe->vertex_tex_cache[i],
                                           texture );
       }
@@ -366,13 +372,14 @@ softpipe_reset_sampler_varients(struct softpipe_context *softpipe)
             }
 
             softpipe->tgsi.geom_samplers_list[i] =
-               get_sampler_varient(
+               get_sampler_variant(
                   i,
                   sp_sampler(softpipe->geometry_samplers[i]),
+                  softpipe->geometry_sampler_views[i],
                   texture,
                   TGSI_PROCESSOR_GEOMETRY );
 
-            sp_sampler_varient_bind_texture(
+            sp_sampler_variant_bind_texture(
                softpipe->tgsi.geom_samplers_list[i],
                softpipe->geometry_tex_cache[i],
                texture );
@@ -389,12 +396,13 @@ softpipe_reset_sampler_varients(struct softpipe_context *softpipe)
          }
 
          softpipe->tgsi.frag_samplers_list[i] =
-            get_sampler_varient( i,
+            get_sampler_variant( i,
                                  sp_sampler(softpipe->sampler[i]),
+                                 softpipe->sampler_views[i],
                                  texture,
                                  TGSI_PROCESSOR_FRAGMENT );
 
-         sp_sampler_varient_bind_texture( softpipe->tgsi.frag_samplers_list[i], 
+         sp_sampler_variant_bind_texture( softpipe->tgsi.frag_samplers_list[i], 
                                           softpipe->tex_cache[i],
                                           texture );
       }
@@ -406,11 +414,11 @@ softpipe_delete_sampler_state(struct pipe_context *pipe,
                               void *sampler)
 {
    struct sp_sampler *sp_sampler = (struct sp_sampler *)sampler;
-   struct sp_sampler_varient *v, *tmp;
+   struct sp_sampler_variant *v, *tmp;
 
-   for (v = sp_sampler->varients; v; v = tmp) {
+   for (v = sp_sampler->variants; v; v = tmp) {
       tmp = v->next;
-      sp_sampler_varient_destroy(v);
+      sp_sampler_variant_destroy(v);
    }
 
    FREE( sampler );
diff --git a/src/gallium/drivers/softpipe/sp_tex_sample.c b/src/gallium/drivers/softpipe/sp_tex_sample.c
index 2eac4c7a82..cbc40d4b44 100644
--- a/src/gallium/drivers/softpipe/sp_tex_sample.c
+++ b/src/gallium/drivers/softpipe/sp_tex_sample.c
@@ -545,7 +545,7 @@ wrap_linear_unorm_clamp_to_edge(const float s[4], unsigned size,
  * derivatives w.r.t X and Y, then compute lambda (level of detail).
  */
 static float
-compute_lambda_1d(const struct sp_sampler_varient *samp,
+compute_lambda_1d(const struct sp_sampler_variant *samp,
                   const float s[QUAD_SIZE],
                   const float t[QUAD_SIZE],
                   const float p[QUAD_SIZE])
@@ -560,7 +560,7 @@ compute_lambda_1d(const struct sp_sampler_varient *samp,
 
 
 static float
-compute_lambda_2d(const struct sp_sampler_varient *samp,
+compute_lambda_2d(const struct sp_sampler_variant *samp,
                   const float s[QUAD_SIZE],
                   const float t[QUAD_SIZE],
                   const float p[QUAD_SIZE])
@@ -579,7 +579,7 @@ compute_lambda_2d(const struct sp_sampler_varient *samp,
 
 
 static float
-compute_lambda_3d(const struct sp_sampler_varient *samp,
+compute_lambda_3d(const struct sp_sampler_variant *samp,
                   const float s[QUAD_SIZE],
                   const float t[QUAD_SIZE],
                   const float p[QUAD_SIZE])
@@ -608,7 +608,7 @@ compute_lambda_3d(const struct sp_sampler_varient *samp,
  * Since there aren't derivatives to use, just return 0.
  */
 static float
-compute_lambda_vert(const struct sp_sampler_varient *samp,
+compute_lambda_vert(const struct sp_sampler_variant *samp,
                     const float s[QUAD_SIZE],
                     const float t[QUAD_SIZE],
                     const float p[QUAD_SIZE])
@@ -634,7 +634,7 @@ compute_lambda_vert(const struct sp_sampler_varient *samp,
 
 
 static INLINE const float *
-get_texel_2d_no_border(const struct sp_sampler_varient *samp,
+get_texel_2d_no_border(const struct sp_sampler_variant *samp,
 		       union tex_tile_address addr, int x, int y)
 {
    const struct softpipe_tex_cached_tile *tile;
@@ -651,7 +651,7 @@ get_texel_2d_no_border(const struct sp_sampler_varient *samp,
 
 
 static INLINE const float *
-get_texel_2d(const struct sp_sampler_varient *samp,
+get_texel_2d(const struct sp_sampler_variant *samp,
 	     union tex_tile_address addr, int x, int y)
 {
    const struct pipe_resource *texture = samp->texture;
@@ -659,8 +659,7 @@ get_texel_2d(const struct sp_sampler_varient *samp,
 
    if (x < 0 || x >= (int) u_minify(texture->width0, level) ||
        y < 0 || y >= (int) u_minify(texture->height0, level)) {
-      return sp_tex_tile_cache_border_color(samp->cache,
-                                            samp->sampler->border_color);
+      return samp->sampler->border_color;
    }
    else {
       return get_texel_2d_no_border( samp, addr, x, y );
@@ -671,7 +670,7 @@ get_texel_2d(const struct sp_sampler_varient *samp,
 /* Gather a quad of adjacent texels within a tile:
  */
 static INLINE void
-get_texel_quad_2d_no_border_single_tile(const struct sp_sampler_varient *samp,
+get_texel_quad_2d_no_border_single_tile(const struct sp_sampler_variant *samp,
 					union tex_tile_address addr, 
 					unsigned x, unsigned y, 
 					const float *out[4])
@@ -695,7 +694,7 @@ get_texel_quad_2d_no_border_single_tile(const struct sp_sampler_varient *samp,
 /* Gather a quad of potentially non-adjacent texels:
  */
 static INLINE void
-get_texel_quad_2d_no_border(const struct sp_sampler_varient *samp,
+get_texel_quad_2d_no_border(const struct sp_sampler_variant *samp,
 			    union tex_tile_address addr,
 			    int x0, int y0, 
 			    int x1, int y1,
@@ -710,7 +709,7 @@ get_texel_quad_2d_no_border(const struct sp_sampler_varient *samp,
 /* Can involve a lot of unnecessary checks for border color:
  */
 static INLINE void
-get_texel_quad_2d(const struct sp_sampler_varient *samp,
+get_texel_quad_2d(const struct sp_sampler_variant *samp,
 		  union tex_tile_address addr,
 		  int x0, int y0, 
 		  int x1, int y1,
@@ -724,10 +723,10 @@ get_texel_quad_2d(const struct sp_sampler_varient *samp,
 
 
 
-/* 3d varients:
+/* 3d variants:
  */
 static INLINE const float *
-get_texel_3d_no_border(const struct sp_sampler_varient *samp,
+get_texel_3d_no_border(const struct sp_sampler_variant *samp,
                        union tex_tile_address addr, int x, int y, int z)
 {
    const struct softpipe_tex_cached_tile *tile;
@@ -745,7 +744,7 @@ get_texel_3d_no_border(const struct sp_sampler_varient *samp,
 
 
 static INLINE const float *
-get_texel_3d(const struct sp_sampler_varient *samp,
+get_texel_3d(const struct sp_sampler_variant *samp,
 	     union tex_tile_address addr, int x, int y, int z)
 {
    const struct pipe_resource *texture = samp->texture;
@@ -754,8 +753,7 @@ get_texel_3d(const struct sp_sampler_varient *samp,
    if (x < 0 || x >= (int) u_minify(texture->width0, level) ||
        y < 0 || y >= (int) u_minify(texture->height0, level) ||
        z < 0 || z >= (int) u_minify(texture->depth0, level)) {
-      return sp_tex_tile_cache_border_color(samp->cache,
-                                            samp->sampler->border_color);
+      return samp->sampler->border_color;
    }
    else {
       return get_texel_3d_no_border( samp, addr, x, y, z );
@@ -800,7 +798,7 @@ img_filter_2d_linear_repeat_POT(struct tgsi_sampler *tgsi_sampler,
                                 enum tgsi_sampler_control control,
                                 float rgba[NUM_CHANNELS][QUAD_SIZE])
 {
-   const struct sp_sampler_varient *samp = sp_sampler_varient(tgsi_sampler);
+   const struct sp_sampler_variant *samp = sp_sampler_variant(tgsi_sampler);
    unsigned  j;
    unsigned level = samp->level;
    unsigned xpot = pot_level_size(samp->xpot, level);
@@ -863,7 +861,7 @@ img_filter_2d_nearest_repeat_POT(struct tgsi_sampler *tgsi_sampler,
                                  enum tgsi_sampler_control control,
                                  float rgba[NUM_CHANNELS][QUAD_SIZE])
 {
-   const struct sp_sampler_varient *samp = sp_sampler_varient(tgsi_sampler);
+   const struct sp_sampler_variant *samp = sp_sampler_variant(tgsi_sampler);
    unsigned  j;
    unsigned level = samp->level;
    unsigned xpot = pot_level_size(samp->xpot, level);
@@ -907,7 +905,7 @@ img_filter_2d_nearest_clamp_POT(struct tgsi_sampler *tgsi_sampler,
                                 enum tgsi_sampler_control control,
                                 float rgba[NUM_CHANNELS][QUAD_SIZE])
 {
-   const struct sp_sampler_varient *samp = sp_sampler_varient(tgsi_sampler);
+   const struct sp_sampler_variant *samp = sp_sampler_variant(tgsi_sampler);
    unsigned  j;
    unsigned level = samp->level;
    unsigned xpot = pot_level_size(samp->xpot, level);
@@ -960,7 +958,7 @@ img_filter_1d_nearest(struct tgsi_sampler *tgsi_sampler,
                         enum tgsi_sampler_control control,
                         float rgba[NUM_CHANNELS][QUAD_SIZE])
 {
-   const struct sp_sampler_varient *samp = sp_sampler_varient(tgsi_sampler);
+   const struct sp_sampler_variant *samp = sp_sampler_variant(tgsi_sampler);
    const struct pipe_resource *texture = samp->texture;
    unsigned level0, j;
    int width;
@@ -1000,7 +998,7 @@ img_filter_2d_nearest(struct tgsi_sampler *tgsi_sampler,
                       enum tgsi_sampler_control control,
                       float rgba[NUM_CHANNELS][QUAD_SIZE])
 {
-   const struct sp_sampler_varient *samp = sp_sampler_varient(tgsi_sampler);
+   const struct sp_sampler_variant *samp = sp_sampler_variant(tgsi_sampler);
    const struct pipe_resource *texture = samp->texture;
    unsigned level0, j;
    int width, height;
@@ -1052,7 +1050,7 @@ img_filter_cube_nearest(struct tgsi_sampler *tgsi_sampler,
                         enum tgsi_sampler_control control,
                         float rgba[NUM_CHANNELS][QUAD_SIZE])
 {
-   const struct sp_sampler_varient *samp = sp_sampler_varient(tgsi_sampler);
+   const struct sp_sampler_variant *samp = sp_sampler_variant(tgsi_sampler);
    const struct pipe_resource *texture = samp->texture;
    const unsigned *faces = samp->faces; /* zero when not cube-mapping */
    unsigned level0, j;
@@ -1096,7 +1094,7 @@ img_filter_3d_nearest(struct tgsi_sampler *tgsi_sampler,
                       enum tgsi_sampler_control control,
                       float rgba[NUM_CHANNELS][QUAD_SIZE])
 {
-   const struct sp_sampler_varient *samp = sp_sampler_varient(tgsi_sampler);
+   const struct sp_sampler_variant *samp = sp_sampler_variant(tgsi_sampler);
    const struct pipe_resource *texture = samp->texture;
    unsigned level0, j;
    int width, height, depth;
@@ -1138,7 +1136,7 @@ img_filter_1d_linear(struct tgsi_sampler *tgsi_sampler,
                      enum tgsi_sampler_control control,
                      float rgba[NUM_CHANNELS][QUAD_SIZE])
 {
-   const struct sp_sampler_varient *samp = sp_sampler_varient(tgsi_sampler);
+   const struct sp_sampler_variant *samp = sp_sampler_variant(tgsi_sampler);
    const struct pipe_resource *texture = samp->texture;
    unsigned level0, j;
    int width;
@@ -1178,7 +1176,7 @@ img_filter_2d_linear(struct tgsi_sampler *tgsi_sampler,
                      enum tgsi_sampler_control control,
                      float rgba[NUM_CHANNELS][QUAD_SIZE])
 {
-   const struct sp_sampler_varient *samp = sp_sampler_varient(tgsi_sampler);
+   const struct sp_sampler_variant *samp = sp_sampler_variant(tgsi_sampler);
    const struct pipe_resource *texture = samp->texture;
    unsigned level0, j;
    int width, height;
@@ -1225,7 +1223,7 @@ img_filter_cube_linear(struct tgsi_sampler *tgsi_sampler,
                        enum tgsi_sampler_control control,
                        float rgba[NUM_CHANNELS][QUAD_SIZE])
 {
-   const struct sp_sampler_varient *samp = sp_sampler_varient(tgsi_sampler);
+   const struct sp_sampler_variant *samp = sp_sampler_variant(tgsi_sampler);
    const struct pipe_resource *texture = samp->texture;
    const unsigned *faces = samp->faces; /* zero when not cube-mapping */
    unsigned level0, j;
@@ -1274,7 +1272,7 @@ img_filter_3d_linear(struct tgsi_sampler *tgsi_sampler,
                      enum tgsi_sampler_control control,
                      float rgba[NUM_CHANNELS][QUAD_SIZE])
 {
-   const struct sp_sampler_varient *samp = sp_sampler_varient(tgsi_sampler);
+   const struct sp_sampler_variant *samp = sp_sampler_variant(tgsi_sampler);
    const struct pipe_resource *texture = samp->texture;
    unsigned level0, j;
    int width, height, depth;
@@ -1350,7 +1348,7 @@ mip_filter_linear(struct tgsi_sampler *tgsi_sampler,
                   enum tgsi_sampler_control control,
                   float rgba[NUM_CHANNELS][QUAD_SIZE])
 {
-   struct sp_sampler_varient *samp = sp_sampler_varient(tgsi_sampler);
+   struct sp_sampler_variant *samp = sp_sampler_variant(tgsi_sampler);
    const struct pipe_resource *texture = samp->texture;
    int level0;
    float lambda;
@@ -1417,7 +1415,7 @@ mip_filter_nearest(struct tgsi_sampler *tgsi_sampler,
                    enum tgsi_sampler_control control,
                    float rgba[NUM_CHANNELS][QUAD_SIZE])
 {
-   struct sp_sampler_varient *samp = sp_sampler_varient(tgsi_sampler);
+   struct sp_sampler_variant *samp = sp_sampler_variant(tgsi_sampler);
    const struct pipe_resource *texture = samp->texture;
    float lambda;
    float lod[QUAD_SIZE];
@@ -1460,7 +1458,7 @@ mip_filter_none(struct tgsi_sampler *tgsi_sampler,
                 enum tgsi_sampler_control control,
                 float rgba[NUM_CHANNELS][QUAD_SIZE])
 {
-   struct sp_sampler_varient *samp = sp_sampler_varient(tgsi_sampler);
+   struct sp_sampler_variant *samp = sp_sampler_variant(tgsi_sampler);
    float lambda;
    float lod[QUAD_SIZE];
 
@@ -1501,7 +1499,7 @@ mip_filter_linear_2d_linear_repeat_POT(
    enum tgsi_sampler_control control,
    float rgba[NUM_CHANNELS][QUAD_SIZE])
 {
-   struct sp_sampler_varient *samp = sp_sampler_varient(tgsi_sampler);
+   struct sp_sampler_variant *samp = sp_sampler_variant(tgsi_sampler);
    const struct pipe_resource *texture = samp->texture;
    int level0;
    float lambda;
@@ -1569,7 +1567,7 @@ sample_compare(struct tgsi_sampler *tgsi_sampler,
                enum tgsi_sampler_control control,
                float rgba[NUM_CHANNELS][QUAD_SIZE])
 {
-   struct sp_sampler_varient *samp = sp_sampler_varient(tgsi_sampler);
+   struct sp_sampler_variant *samp = sp_sampler_variant(tgsi_sampler);
    const struct pipe_sampler_state *sampler = samp->sampler;
    int j, k0, k1, k2, k3;
    float val;
@@ -1656,7 +1654,7 @@ sample_cube(struct tgsi_sampler *tgsi_sampler,
             enum tgsi_sampler_control control,
             float rgba[NUM_CHANNELS][QUAD_SIZE])
 {
-   struct sp_sampler_varient *samp = sp_sampler_varient(tgsi_sampler);
+   struct sp_sampler_variant *samp = sp_sampler_variant(tgsi_sampler);
    unsigned j;
    float ssss[4], tttt[4];
 
@@ -1731,6 +1729,86 @@ sample_cube(struct tgsi_sampler *tgsi_sampler,
 }
 
 
+static void
+sample_swizzle(struct tgsi_sampler *tgsi_sampler,
+               const float s[QUAD_SIZE],
+               const float t[QUAD_SIZE],
+               const float p[QUAD_SIZE],
+               const float c0[QUAD_SIZE],
+               enum tgsi_sampler_control control,
+               float rgba[NUM_CHANNELS][QUAD_SIZE])
+{
+   struct sp_sampler_variant *samp = sp_sampler_variant(tgsi_sampler);
+   float rgba_temp[NUM_CHANNELS][QUAD_SIZE];
+   const unsigned swizzle_r = samp->key.bits.swizzle_r;
+   const unsigned swizzle_g = samp->key.bits.swizzle_g;
+   const unsigned swizzle_b = samp->key.bits.swizzle_b;
+   const unsigned swizzle_a = samp->key.bits.swizzle_a;
+   unsigned j;
+
+   samp->sample_target(tgsi_sampler, s, t, p, c0, control, rgba_temp);
+
+   switch (swizzle_r) {
+   case PIPE_SWIZZLE_ZERO:
+      for (j = 0; j < 4; j++)
+         rgba[0][j] = 0.0f;
+      break;
+   case PIPE_SWIZZLE_ONE:
+      for (j = 0; j < 4; j++)
+         rgba[0][j] = 1.0f;
+      break;
+   default:
+      assert(swizzle_r < 4);
+      for (j = 0; j < 4; j++)
+         rgba[0][j] = rgba_temp[swizzle_r][j];
+   }
+
+   switch (swizzle_g) {
+   case PIPE_SWIZZLE_ZERO:
+      for (j = 0; j < 4; j++)
+         rgba[1][j] = 0.0f;
+      break;
+   case PIPE_SWIZZLE_ONE:
+      for (j = 0; j < 4; j++)
+         rgba[1][j] = 1.0f;
+      break;
+   default:
+      assert(swizzle_g < 4);
+      for (j = 0; j < 4; j++)
+         rgba[1][j] = rgba_temp[swizzle_g][j];
+   }
+
+   switch (swizzle_b) {
+   case PIPE_SWIZZLE_ZERO:
+      for (j = 0; j < 4; j++)
+         rgba[2][j] = 0.0f;
+      break;
+   case PIPE_SWIZZLE_ONE:
+      for (j = 0; j < 4; j++)
+         rgba[2][j] = 1.0f;
+      break;
+   default:
+      assert(swizzle_b < 4);
+      for (j = 0; j < 4; j++)
+         rgba[2][j] = rgba_temp[swizzle_b][j];
+   }
+
+   switch (swizzle_a) {
+   case PIPE_SWIZZLE_ZERO:
+      for (j = 0; j < 4; j++)
+         rgba[3][j] = 0.0f;
+      break;
+   case PIPE_SWIZZLE_ONE:
+      for (j = 0; j < 4; j++)
+         rgba[3][j] = 1.0f;
+      break;
+   default:
+      assert(swizzle_a < 4);
+      for (j = 0; j < 4; j++)
+         rgba[3][j] = rgba_temp[swizzle_a][j];
+   }
+}
+
 
 static wrap_nearest_func
 get_nearest_unorm_wrap(unsigned mode)
@@ -1909,10 +1987,10 @@ get_img_filter(const union sp_sampler_key key,
 
 
 /**
- * Bind the given texture object and texture cache to the sampler varient.
+ * Bind the given texture object and texture cache to the sampler variant.
  */
 void
-sp_sampler_varient_bind_texture( struct sp_sampler_varient *samp,
+sp_sampler_variant_bind_texture( struct sp_sampler_variant *samp,
                                  struct softpipe_tex_tile_cache *tex_cache,
                                  const struct pipe_resource *texture )
 {
@@ -1927,20 +2005,20 @@ sp_sampler_varient_bind_texture( struct sp_sampler_varient *samp,
 
 
 void
-sp_sampler_varient_destroy( struct sp_sampler_varient *samp )
+sp_sampler_variant_destroy( struct sp_sampler_variant *samp )
 {
    FREE(samp);
 }
 
 
 /**
- * Create a sampler varient for a given set of non-orthogonal state.
+ * Create a sampler variant for a given set of non-orthogonal state.
  */
-struct sp_sampler_varient *
-sp_create_sampler_varient( const struct pipe_sampler_state *sampler,
+struct sp_sampler_variant *
+sp_create_sampler_variant( const struct pipe_sampler_state *sampler,
                            const union sp_sampler_key key )
 {
-   struct sp_sampler_varient *samp = CALLOC_STRUCT(sp_sampler_varient);
+   struct sp_sampler_variant *samp = CALLOC_STRUCT(sp_sampler_variant);
    if (!samp)
       return NULL;
 
@@ -2015,7 +2093,7 @@ sp_create_sampler_varient( const struct pipe_sampler_state *sampler,
    }
    
    if (key.bits.target == PIPE_TEXTURE_CUBE) {
-      samp->base.get_samples = sample_cube;
+      samp->sample_target = sample_cube;
    }
    else {
       samp->faces[0] = 0;
@@ -2026,7 +2104,17 @@ sp_create_sampler_varient( const struct pipe_sampler_state *sampler,
       /* Skip cube face determination by promoting the compare
        * function pointer:
        */
-      samp->base.get_samples = samp->compare;
+      samp->sample_target = samp->compare;
+   }
+
+   if (key.bits.swizzle_r != PIPE_SWIZZLE_RED ||
+       key.bits.swizzle_g != PIPE_SWIZZLE_GREEN ||
+       key.bits.swizzle_b != PIPE_SWIZZLE_BLUE ||
+       key.bits.swizzle_a != PIPE_SWIZZLE_ALPHA) {
+      samp->base.get_samples = sample_swizzle;
+   }
+   else {
+      samp->base.get_samples = samp->sample_target;
    }
 
    return samp;
diff --git a/src/gallium/drivers/softpipe/sp_tex_sample.h b/src/gallium/drivers/softpipe/sp_tex_sample.h
index 6114acf737..ed99006ab0 100644
--- a/src/gallium/drivers/softpipe/sp_tex_sample.h
+++ b/src/gallium/drivers/softpipe/sp_tex_sample.h
@@ -32,7 +32,7 @@
 
 #include "tgsi/tgsi_exec.h"
 
-struct sp_sampler_varient;
+struct sp_sampler_variant;
 
 typedef void (*wrap_nearest_func)(const float s[4],
                                   unsigned size,
@@ -44,7 +44,7 @@ typedef void (*wrap_linear_func)(const float s[4],
                                  int icoord1[4],
                                  float w[4]);
 
-typedef float (*compute_lambda_func)(const struct sp_sampler_varient *sampler,
+typedef float (*compute_lambda_func)(const struct sp_sampler_variant *sampler,
                                      const float s[QUAD_SIZE],
                                      const float t[QUAD_SIZE],
                                      const float p[QUAD_SIZE]);
@@ -64,7 +64,11 @@ union sp_sampler_key {
       unsigned is_pot:1;
       unsigned processor:2;
       unsigned unit:4;
-      unsigned pad:22;
+      unsigned swizzle_r:3;
+      unsigned swizzle_g:3;
+      unsigned swizzle_b:3;
+      unsigned swizzle_a:3;
+      unsigned pad:10;
    } bits;
    unsigned value;
 };
@@ -72,7 +76,7 @@ union sp_sampler_key {
 /**
  * Subclass of tgsi_sampler
  */
-struct sp_sampler_varient
+struct sp_sampler_variant
 {
    struct tgsi_sampler base;  /**< base class */
 
@@ -113,32 +117,33 @@ struct sp_sampler_varient
 
    filter_func mip_filter;
    filter_func compare;
+   filter_func sample_target;
    
    /* Linked list:
     */
-   struct sp_sampler_varient *next;
+   struct sp_sampler_variant *next;
 };
 
 struct sp_sampler;
 
-/* Create a sampler varient for a given set of non-orthogonal state.  Currently the 
+/* Create a sampler variant for a given set of non-orthogonal state.  Currently the 
  */
-struct sp_sampler_varient *
-sp_create_sampler_varient( const struct pipe_sampler_state *sampler,
+struct sp_sampler_variant *
+sp_create_sampler_variant( const struct pipe_sampler_state *sampler,
                            const union sp_sampler_key key );
 
-void sp_sampler_varient_bind_texture( struct sp_sampler_varient *varient,
+void sp_sampler_variant_bind_texture( struct sp_sampler_variant *variant,
                                       struct softpipe_tex_tile_cache *tex_cache,
                                       const struct pipe_resource *tex );
 
-void sp_sampler_varient_destroy( struct sp_sampler_varient * );
+void sp_sampler_variant_destroy( struct sp_sampler_variant * );
 
 
 
-static INLINE struct sp_sampler_varient *
-sp_sampler_varient(const struct tgsi_sampler *sampler)
+static INLINE struct sp_sampler_variant *
+sp_sampler_variant(const struct tgsi_sampler *sampler)
 {
-   return (struct sp_sampler_varient *) sampler;
+   return (struct sp_sampler_variant *) sampler;
 }
 
 extern void
diff --git a/src/gallium/drivers/softpipe/sp_tex_tile_cache.c b/src/gallium/drivers/softpipe/sp_tex_tile_cache.c
index e817c0c8cf..e42015ad49 100644
--- a/src/gallium/drivers/softpipe/sp_tex_tile_cache.c
+++ b/src/gallium/drivers/softpipe/sp_tex_tile_cache.c
@@ -48,6 +48,9 @@ sp_create_tex_tile_cache( struct pipe_context *pipe )
    struct softpipe_tex_tile_cache *tc;
    uint pos;
 
+   /* make sure max texture size works */
+   assert((TILE_SIZE << TEX_ADDR_BITS) >= (1 << (SP_MAX_TEXTURE_2D_LEVELS-1)));
+
    tc = CALLOC_STRUCT( softpipe_tex_tile_cache );
    if (tc) {
       tc->pipe = pipe;
@@ -260,15 +263,14 @@ sp_find_cached_tile_tex(struct softpipe_tex_tile_cache *tc,
          }
 
          tc->tex_trans = 
-            pipe_get_transfer(tc->pipe, tc->texture, 
-			      addr.bits.face, 
-			      addr.bits.level, 
-			      addr.bits.z, 
-			      PIPE_TRANSFER_READ | PIPE_TRANSFER_UNSYNCHRONIZED,
-			      0, 0,
-			      u_minify(tc->texture->width0, addr.bits.level),
-			      u_minify(tc->texture->height0, addr.bits.level));
-         
+            pipe_get_transfer(tc->pipe, tc->texture,
+                              addr.bits.level,
+                              addr.bits.face + addr.bits.z,
+                              PIPE_TRANSFER_READ | PIPE_TRANSFER_UNSYNCHRONIZED,
+                              0, 0,
+                              u_minify(tc->texture->width0, addr.bits.level),
+                              u_minify(tc->texture->height0, addr.bits.level));
+
          tc->tex_trans_map = tc->pipe->transfer_map(tc->pipe, tc->tex_trans);
 
          tc->tex_face = addr.bits.face;
@@ -276,45 +278,26 @@ sp_find_cached_tile_tex(struct softpipe_tex_tile_cache *tc,
          tc->tex_z = addr.bits.z;
       }
 
-      /* get tile from the transfer (view into texture) */
+      /* get tile from the transfer (view into texture)
+       * Note we're using the swizzle version of this fuction only because
+       * we need to pass the texture cache's format explicitly.
+       */
       pipe_get_tile_swizzle(tc->pipe,
 			    tc->tex_trans,
                             addr.bits.x * TILE_SIZE, 
                             addr.bits.y * TILE_SIZE,
                             TILE_SIZE,
                             TILE_SIZE,
-                            tc->swizzle_r,
-                            tc->swizzle_g,
-                            tc->swizzle_b,
-                            tc->swizzle_a,
+                            PIPE_SWIZZLE_RED,
+                            PIPE_SWIZZLE_GREEN,
+                            PIPE_SWIZZLE_BLUE,
+                            PIPE_SWIZZLE_ALPHA,
                             tc->format,
                             (float *) tile->data.color);
+
       tile->addr = addr;
    }
 
    tc->last_tile = tile;
    return tile;
 }
-
-
-
-/**
- * Return the swizzled border color.
- */
-const float *
-sp_tex_tile_cache_border_color(struct softpipe_tex_tile_cache *tc,
-                               const float border_color[4])
-{
-   float rgba01[6];
-
-   COPY_4V(rgba01, border_color);
-   rgba01[PIPE_SWIZZLE_ZERO] = 0.0f;
-   rgba01[PIPE_SWIZZLE_ONE] = 1.0f;
-
-   tc->swz_border_color[0] = rgba01[tc->swizzle_r];
-   tc->swz_border_color[1] = rgba01[tc->swizzle_g];
-   tc->swz_border_color[2] = rgba01[tc->swizzle_b];
-   tc->swz_border_color[3] = rgba01[tc->swizzle_a];
-
-   return tc->swz_border_color;
-}
diff --git a/src/gallium/drivers/softpipe/sp_tex_tile_cache.h b/src/gallium/drivers/softpipe/sp_tex_tile_cache.h
index 05f25133da..2220955b71 100644
--- a/src/gallium/drivers/softpipe/sp_tex_tile_cache.h
+++ b/src/gallium/drivers/softpipe/sp_tex_tile_cache.h
@@ -30,6 +30,7 @@
 
 
 #include "pipe/p_compiler.h"
+#include "sp_limits.h"
 
 
 struct softpipe_context;
@@ -39,22 +40,26 @@ struct softpipe_tex_tile_cache;
 /**
  * Cache tile size (width and height). This needs to be a power of two.
  */
-#define TILE_SIZE 64
+#define TILE_SIZE_LOG2 6
+#define TILE_SIZE (1 << TILE_SIZE_LOG2)
 
 
-/* If we need to support > 4096, just expand this to be a 64 bit
- * union, or consider tiling in Z as well.
+#define TEX_ADDR_BITS (SP_MAX_TEXTURE_2D_LEVELS - 1 - TILE_SIZE_LOG2)
+#define TEX_Z_BITS (SP_MAX_TEXTURE_2D_LEVELS - 1)
+
+/**
+ * Texture tile address as a union for fast compares.
  */
 union tex_tile_address {
    struct {
-      unsigned x:6;             /* 4096 / TILE_SIZE */
-      unsigned y:6;             /* 4096 / TILE_SIZE */
-      unsigned z:12;            /* 4096 -- z not tiled */
+      unsigned x:TEX_ADDR_BITS;  /* 16K / TILE_SIZE */
+      unsigned y:TEX_ADDR_BITS;  /* 16K / TILE_SIZE */
+      unsigned z:TEX_Z_BITS;     /* 16K -- z not tiled */
       unsigned face:3;
       unsigned level:4;
       unsigned invalid:1;
    } bits;
-   unsigned value;
+   uint64_t value;
 };
 
 
@@ -90,8 +95,6 @@ struct softpipe_tex_tile_cache
    unsigned format;
 
    struct softpipe_tex_cached_tile *last_tile;  /**< most recently retrieved tile */
-
-   float swz_border_color[4]; /**< swizzled border color */
 };
 
 
@@ -126,10 +129,10 @@ sp_find_cached_tile_tex(struct softpipe_tex_tile_cache *tc,
 
 static INLINE union tex_tile_address
 tex_tile_address( unsigned x,
-		  unsigned y,
-		  unsigned z,
-		  unsigned face,
-		  unsigned level )
+                  unsigned y,
+                  unsigned z,
+                  unsigned face,
+                  unsigned level )
 {
    union tex_tile_address addr;
 
@@ -139,7 +142,7 @@ tex_tile_address( unsigned x,
    addr.bits.z = z;
    addr.bits.face = face;
    addr.bits.level = level;
-      
+
    return addr;
 }
 
@@ -156,10 +159,5 @@ sp_get_cached_tile_tex(struct softpipe_tex_tile_cache *tc,
 }
 
 
-const float *
-sp_tex_tile_cache_border_color(struct softpipe_tex_tile_cache *tc,
-                               const float border_color[4]);
-
-
 #endif /* SP_TEX_TILE_CACHE_H */
 
diff --git a/src/gallium/drivers/softpipe/sp_texture.c b/src/gallium/drivers/softpipe/sp_texture.c
index 4e6123fbd0..509d9982b1 100644
--- a/src/gallium/drivers/softpipe/sp_texture.c
+++ b/src/gallium/drivers/softpipe/sp_texture.c
@@ -220,23 +220,18 @@ softpipe_resource_get_handle(struct pipe_screen *screen,
  */
 static unsigned
 sp_get_tex_image_offset(const struct softpipe_resource *spr,
-                        unsigned level, unsigned face, unsigned zslice)
+                        unsigned level, unsigned layer)
 {
    const unsigned hgt = u_minify(spr->base.height0, level);
    const unsigned nblocksy = util_format_get_nblocksy(spr->base.format, hgt);
    unsigned offset = spr->level_offset[level];
 
-   if (spr->base.target == PIPE_TEXTURE_CUBE) {
-      assert(zslice == 0);
-      offset += face * nblocksy * spr->stride[level];
-   }
-   else if (spr->base.target == PIPE_TEXTURE_3D) {
-      assert(face == 0);
-      offset += zslice * nblocksy * spr->stride[level];
+   if (spr->base.target == PIPE_TEXTURE_CUBE ||
+       spr->base.target == PIPE_TEXTURE_3D) {
+      offset += layer * nblocksy * spr->stride[level];
    }
    else {
-      assert(face == 0);
-      assert(zslice == 0);
+      assert(layer == 0);
    }
 
    return offset;
@@ -247,39 +242,40 @@ sp_get_tex_image_offset(const struct softpipe_resource *spr,
  * Get a pipe_surface "view" into a texture resource.
  */
 static struct pipe_surface *
-softpipe_get_tex_surface(struct pipe_screen *screen,
-                         struct pipe_resource *pt,
-                         unsigned face, unsigned level, unsigned zslice,
-                         unsigned usage)
+softpipe_create_surface(struct pipe_context *pipe,
+                        struct pipe_resource *pt,
+                        const struct pipe_surface *surf_tmpl)
 {
-   struct softpipe_resource *spr = softpipe_resource(pt);
    struct pipe_surface *ps;
+   unsigned level = surf_tmpl->u.tex.level;
 
    assert(level <= pt->last_level);
+   assert(surf_tmpl->u.tex.first_layer == surf_tmpl->u.tex.last_layer);
 
    ps = CALLOC_STRUCT(pipe_surface);
    if (ps) {
       pipe_reference_init(&ps->reference, 1);
       pipe_resource_reference(&ps->texture, pt);
-      ps->format = pt->format;
+      ps->context = pipe;
+      ps->format = surf_tmpl->format;
       ps->width = u_minify(pt->width0, level);
       ps->height = u_minify(pt->height0, level);
-      ps->offset = sp_get_tex_image_offset(spr, level, face, zslice);
-      ps->usage = usage;
+      ps->usage = surf_tmpl->usage;
 
-      ps->face = face;
-      ps->level = level;
-      ps->zslice = zslice;
+      ps->u.tex.level = level;
+      ps->u.tex.first_layer = surf_tmpl->u.tex.first_layer;
+      ps->u.tex.last_layer = surf_tmpl->u.tex.last_layer;
    }
    return ps;
 }
 
 
 /**
- * Free a pipe_surface which was created with softpipe_get_tex_surface().
+ * Free a pipe_surface which was created with softpipe_create_surface().
  */
 static void 
-softpipe_tex_surface_destroy(struct pipe_surface *surf)
+softpipe_surface_destroy(struct pipe_context *pipe,
+                         struct pipe_surface *surf)
 {
    /* Effectively do the texture_update work here - if texture images
     * needed post-processing to put them into hardware layout, this is
@@ -302,21 +298,21 @@ softpipe_tex_surface_destroy(struct pipe_surface *surf)
  */
 static struct pipe_transfer *
 softpipe_get_transfer(struct pipe_context *pipe,
-		      struct pipe_resource *resource,
-		      struct pipe_subresource sr,
-		      unsigned usage,
-		      const struct pipe_box *box)
+                      struct pipe_resource *resource,
+                      unsigned level,
+                      unsigned usage,
+                      const struct pipe_box *box)
 {
    struct softpipe_resource *spr = softpipe_resource(resource);
    struct softpipe_transfer *spt;
 
    assert(resource);
-   assert(sr.level <= resource->last_level);
+   assert(level <= resource->last_level);
 
    /* make sure the requested region is in the image bounds */
-   assert(box->x + box->width <= u_minify(resource->width0, sr.level));
-   assert(box->y + box->height <= u_minify(resource->height0, sr.level));
-   assert(box->z + box->depth <= u_minify(resource->depth0, sr.level));
+   assert(box->x + box->width <= u_minify(resource->width0, level));
+   assert(box->y + box->height <= u_minify(resource->height0, level));
+   assert(box->z + box->depth <= (u_minify(resource->depth0, level) + resource->array_size - 1));
 
    /*
     * Transfers, like other pipe operations, must happen in order, so flush the
@@ -326,7 +322,7 @@ softpipe_get_transfer(struct pipe_context *pipe,
       boolean read_only = !(usage & PIPE_TRANSFER_WRITE);
       boolean do_not_block = !!(usage & PIPE_TRANSFER_DONTBLOCK);
       if (!softpipe_flush_resource(pipe, resource,
-                                   sr.face, sr.level,
+                                   level, box->depth > 1 ? -1 : box->z,
                                    0, /* flush_flags */
                                    read_only,
                                    TRUE, /* cpu_access */
@@ -343,21 +339,21 @@ softpipe_get_transfer(struct pipe_context *pipe,
    if (spt) {
       struct pipe_transfer *pt = &spt->base;
       enum pipe_format format = resource->format;
-      const unsigned hgt = u_minify(spr->base.height0, sr.level);
+      const unsigned hgt = u_minify(spr->base.height0, level);
       const unsigned nblocksy = util_format_get_nblocksy(format, hgt);
 
       pipe_resource_reference(&pt->resource, resource);
-      pt->sr = sr;
+      pt->level = level;
       pt->usage = usage;
       pt->box = *box;
-      pt->stride = spr->stride[sr.level];
-      pt->slice_stride = pt->stride * nblocksy;
+      pt->stride = spr->stride[level];
+      pt->layer_stride = pt->stride * nblocksy;
 
-      spt->offset = sp_get_tex_image_offset(spr, sr.level, sr.face, box->z);
+      spt->offset = sp_get_tex_image_offset(spr, level, box->z);
  
       spt->offset += 
-	 box->y / util_format_get_blockheight(format) * spt->base.stride +
-	 box->x / util_format_get_blockwidth(format) * util_format_get_blocksize(format);
+         box->y / util_format_get_blockheight(format) * spt->base.stride +
+         box->x / util_format_get_blockwidth(format) * util_format_get_blocksize(format);
 
       return pt;
    }
@@ -454,6 +450,7 @@ softpipe_user_buffer_create(struct pipe_screen *screen,
    spr->base.width0 = bytes;
    spr->base.height0 = 1;
    spr->base.depth0 = 1;
+   spr->base.array_size = 1;
    spr->userBuffer = TRUE;
    spr->data = ptr;
 
@@ -471,6 +468,9 @@ softpipe_init_texture_funcs(struct pipe_context *pipe)
 
    pipe->transfer_flush_region = u_default_transfer_flush_region;
    pipe->transfer_inline_write = u_default_transfer_inline_write;
+
+   pipe->create_surface = softpipe_create_surface;
+   pipe->surface_destroy = softpipe_surface_destroy;
 }
 
 
@@ -483,6 +483,4 @@ softpipe_init_screen_texture_funcs(struct pipe_screen *screen)
    screen->resource_get_handle = softpipe_resource_get_handle;
    screen->user_buffer_create = softpipe_user_buffer_create;
 
-   screen->get_tex_surface = softpipe_get_tex_surface;
-   screen->tex_surface_destroy = softpipe_tex_surface_destroy;
 }
diff --git a/src/gallium/drivers/softpipe/sp_texture.h b/src/gallium/drivers/softpipe/sp_texture.h
index 6b205dc532..5603110eeb 100644
--- a/src/gallium/drivers/softpipe/sp_texture.h
+++ b/src/gallium/drivers/softpipe/sp_texture.h
@@ -30,10 +30,7 @@
 
 
 #include "pipe/p_state.h"
-
-
-#define SP_MAX_TEXTURE_2D_LEVELS 13  /* 4K x 4K */
-#define SP_MAX_TEXTURE_3D_LEVELS 9   /* 512 x 512 x 512 */
+#include "sp_limits.h"
 
 
 struct pipe_context;
diff --git a/src/gallium/drivers/softpipe/sp_tile_cache.c b/src/gallium/drivers/softpipe/sp_tile_cache.c
index aa76b8aa1e..480860af63 100644
--- a/src/gallium/drivers/softpipe/sp_tile_cache.c
+++ b/src/gallium/drivers/softpipe/sp_tile_cache.c
@@ -92,6 +92,10 @@ sp_create_tile_cache( struct pipe_context *pipe )
    maxTexSize = 1 << (maxLevels - 1);
    assert(MAX_WIDTH >= maxTexSize);
 
+   assert(sizeof(union tile_address) == 4);
+
+   assert((TILE_SIZE << TILE_ADDR_BITS) >= MAX_WIDTH);
+
    tc = CALLOC_STRUCT( softpipe_tile_cache );
    if (tc) {
       tc->pipe = pipe;
@@ -170,11 +174,11 @@ sp_tile_cache_set_surface(struct softpipe_tile_cache *tc,
    tc->surface = ps;
 
    if (ps) {
-      tc->transfer = pipe_get_transfer(pipe, ps->texture, ps->face,
-					   ps->level, ps->zslice,
-					   PIPE_TRANSFER_READ_WRITE |
-					   PIPE_TRANSFER_UNSYNCHRONIZED,
-					   0, 0, ps->width, ps->height);
+      tc->transfer = pipe_get_transfer(pipe, ps->texture,
+                                       ps->u.tex.level, ps->u.tex.first_layer,
+                                       PIPE_TRANSFER_READ_WRITE |
+                                       PIPE_TRANSFER_UNSYNCHRONIZED,
+                                       0, 0, ps->width, ps->height);
 
       tc->depth_stencil = (ps->format == PIPE_FORMAT_Z24_UNORM_S8_USCALED ||
                            ps->format == PIPE_FORMAT_Z24X8_UNORM ||
diff --git a/src/gallium/drivers/softpipe/sp_tile_cache.h b/src/gallium/drivers/softpipe/sp_tile_cache.h
index 4151a47c32..68140b1d2f 100644
--- a/src/gallium/drivers/softpipe/sp_tile_cache.h
+++ b/src/gallium/drivers/softpipe/sp_tile_cache.h
@@ -30,6 +30,7 @@
 
 
 #include "pipe/p_compiler.h"
+#include "sp_texture.h"
 
 
 struct softpipe_tile_cache;
@@ -38,18 +39,22 @@ struct softpipe_tile_cache;
 /**
  * Cache tile size (width and height). This needs to be a power of two.
  */
-#define TILE_SIZE 64
+#define TILE_SIZE_LOG2 6
+#define TILE_SIZE (1 << TILE_SIZE_LOG2)
 
 
-/* If we need to support > 4096, just expand this to be a 64 bit
- * union, or consider tiling in Z as well.
+#define TILE_ADDR_BITS (SP_MAX_TEXTURE_2D_LEVELS - 1 - TILE_SIZE_LOG2)
+
+
+/**
+ * Surface tile address as a union for fast compares.
  */
 union tile_address {
    struct {
-      unsigned x:6;             /* 4096 / TILE_SIZE */
-      unsigned y:6;             /* 4096 / TILE_SIZE */
+      unsigned x:TILE_ADDR_BITS;     /* 16K / TILE_SIZE */
+      unsigned y:TILE_ADDR_BITS;     /* 16K / TILE_SIZE */
       unsigned invalid:1;
-      unsigned pad:19;
+      unsigned pad:15;
    } bits;
    unsigned value;
 };
@@ -70,11 +75,6 @@ struct softpipe_cached_tile
 #define NUM_ENTRIES 50
 
 
-/** XXX move these */
-#define MAX_WIDTH 4096
-#define MAX_HEIGHT 4096
-
-
 struct softpipe_tile_cache
 {
    struct pipe_context *pipe;
diff --git a/src/gallium/drivers/svga/SConscript b/src/gallium/drivers/svga/SConscript
index 12ce4732d1..5455ed0757 100644
--- a/src/gallium/drivers/svga/SConscript
+++ b/src/gallium/drivers/svga/SConscript
@@ -73,4 +73,6 @@ svga = env.ConvenienceLibrary(
 	source = sources,
 )
 
+env.Alias('svga', svga)
+
 Export('svga')
diff --git a/src/gallium/drivers/svga/svga_cmd.c b/src/gallium/drivers/svga/svga_cmd.c
index e975f3b02f..05eab8a517 100644
--- a/src/gallium/drivers/svga/svga_cmd.c
+++ b/src/gallium/drivers/svga/svga_cmd.c
@@ -455,8 +455,8 @@ SVGA3D_SurfaceDMA(struct svga_winsys_context *swc,
    cmd->guest.pitch = st->base.stride;
 
    swc->surface_relocation(swc, &cmd->host.sid, texture->handle, surface_flags);
-   cmd->host.face = st->base.sr.face; /* PIPE_TEX_FACE_* and SVGA3D_CUBEFACE_* match */
-   cmd->host.mipmap = st->base.sr.level;
+   cmd->host.face = st->face; /* PIPE_TEX_FACE_* and SVGA3D_CUBEFACE_* match */
+   cmd->host.mipmap = st->base.level;
 
    cmd->transfer = transfer;
 
diff --git a/src/gallium/drivers/svga/svga_context.c b/src/gallium/drivers/svga/svga_context.c
index cd3f6b8982..1e513f1039 100644
--- a/src/gallium/drivers/svga/svga_context.c
+++ b/src/gallium/drivers/svga/svga_context.c
@@ -109,6 +109,7 @@ struct pipe_context *svga_context_create( struct pipe_screen *screen,
    svga_init_vertex_functions(svga);
    svga_init_constbuffer_functions(svga);
    svga_init_query_functions(svga);
+   svga_init_surface_functions(svga);
 
 
    /* debug */
diff --git a/src/gallium/drivers/svga/svga_context.h b/src/gallium/drivers/svga/svga_context.h
index 1fb5a04887..04e281a506 100644
--- a/src/gallium/drivers/svga/svga_context.h
+++ b/src/gallium/drivers/svga/svga_context.h
@@ -422,6 +422,7 @@ void svga_init_vertex_functions( struct svga_context *svga );
 void svga_init_constbuffer_functions( struct svga_context *svga );
 void svga_init_draw_functions( struct svga_context *svga );
 void svga_init_query_functions( struct svga_context *svga );
+void svga_init_surface_functions(struct svga_context *svga);
 
 void svga_cleanup_vertex_state( struct svga_context *svga );
 void svga_cleanup_tss_binding( struct svga_context *svga );
diff --git a/src/gallium/drivers/svga/svga_pipe_blit.c b/src/gallium/drivers/svga/svga_pipe_blit.c
index ca036a6463..426698806c 100644
--- a/src/gallium/drivers/svga/svga_pipe_blit.c
+++ b/src/gallium/drivers/svga/svga_pipe_blit.c
@@ -32,37 +32,40 @@
 #define FILE_DEBUG_FLAG DEBUG_BLIT
 
 
-/* XXX I got my doubts about this, should maybe use svga_texture_copy_handle directly? */
+/* XXX still have doubts about this... */
 static void svga_surface_copy(struct pipe_context *pipe,
                               struct pipe_resource* dst_tex,
-                              struct pipe_subresource subdst,
+                              unsigned dst_level,
                               unsigned dstx, unsigned dsty, unsigned dstz,
                               struct pipe_resource* src_tex,
-                              struct pipe_subresource subsrc,
-                              unsigned srcx, unsigned srcy, unsigned srcz,
-                              unsigned width, unsigned height)
-{
+                              unsigned src_level,
+                              const struct pipe_box *src_box)
+ {
    struct svga_context *svga = svga_context(pipe);
-   struct pipe_screen *screen = pipe->screen;
+   struct svga_texture *stex = svga_texture(src_tex);
+   struct svga_texture *dtex = svga_texture(dst_tex);
+/*   struct pipe_screen *screen = pipe->screen;
    SVGA3dCopyBox *box;
    enum pipe_error ret;
-   struct pipe_surface *srcsurf, *dstsurf;
+   struct pipe_surface *srcsurf, *dstsurf;*/
+   unsigned dst_face, dst_z, src_face, src_z;
 
    svga_hwtnl_flush_retry( svga );
 
+#if 0
    srcsurf = screen->get_tex_surface(screen, src_tex,
-                                     subsrc.face, subsrc.level, srcz,
+                                     src_level, src_box->z, src_box->z,
                                      PIPE_BIND_SAMPLER_VIEW);
 
    dstsurf = screen->get_tex_surface(screen, dst_tex,
-                                     subdst.face, subdst.level, dstz,
+                                     dst_level, dst_box->z, dst_box->z,
                                      PIPE_BIND_RENDER_TARGET);
 
    SVGA_DBG(DEBUG_DMA, "blit to sid %p (%d,%d), from sid %p (%d,%d) sz %dx%d\n",
             svga_surface(dstsurf)->handle,
             dstx, dsty,
             svga_surface(srcsurf)->handle,
-            srcx, srcy,
+            src_box->x, src_box->y,
             width, height);
 
    ret = SVGA3D_BeginSurfaceCopy(svga->swc,
@@ -88,8 +91,8 @@ static void svga_surface_copy(struct pipe_context *pipe,
    box->w = width;
    box->h = height;
    box->d = 1;
-   box->srcx = srcx;
-   box->srcy = srcy;
+   box->srcx = src_box->x;
+   box->srcy = src_box->y;
    box->srcz = 0;
 
    SVGA_FIFOCommitAll(svga->swc);
@@ -100,6 +103,37 @@ static void svga_surface_copy(struct pipe_context *pipe,
    pipe_surface_reference(&srcsurf, NULL);
    pipe_surface_reference(&dstsurf, NULL);
 
+#else
+   if (src_tex->target == PIPE_TEXTURE_CUBE) {
+      src_face = src_box->z;
+      src_z = 0;
+      assert(src_box->depth == 1);
+   }
+   else {
+      src_face = 0;
+      src_z = src_box->z;
+   }
+   /* different src/dst type???*/
+   if (dst_tex->target == PIPE_TEXTURE_CUBE) {
+      dst_face = dstz;
+      dst_z = 0;
+      assert(src_box->depth == 1);
+   }
+   else {
+      dst_face = 0;
+      dst_z = dstz;
+   }
+   svga_texture_copy_handle(svga,
+                            stex->handle,
+                            src_box->x, src_box->y, src_z,
+                            src_level, src_face,
+                            dtex->handle,
+                            dstx, dsty, dst_z,
+                            dst_level, dst_face,
+                            src_box->width, src_box->height, src_box->depth);
+
+#endif
+
 }
 
 
diff --git a/src/gallium/drivers/svga/svga_pipe_rasterizer.c b/src/gallium/drivers/svga/svga_pipe_rasterizer.c
index 660eb0757a..e97b4e5741 100644
--- a/src/gallium/drivers/svga/svga_pipe_rasterizer.c
+++ b/src/gallium/drivers/svga/svga_pipe_rasterizer.c
@@ -68,7 +68,7 @@ svga_create_rasterizer_state(struct pipe_context *pipe,
    /* need this for draw module. */
    rast->templ = *templ;
 
-   /* light_twoside          - XXX: need fragment shader varient */
+   /* light_twoside          - XXX: need fragment shader variant */
    /* poly_smooth            - XXX: no fallback available */
    /* poly_stipple_enable    - draw module */
    /* sprite_coord_enable    - ? */
diff --git a/src/gallium/drivers/svga/svga_resource_buffer.c b/src/gallium/drivers/svga/svga_resource_buffer.c
index 198d401332..f12e2b6862 100644
--- a/src/gallium/drivers/svga/svga_resource_buffer.c
+++ b/src/gallium/drivers/svga/svga_resource_buffer.c
@@ -53,8 +53,8 @@ svga_buffer_needs_hw_storage(unsigned usage)
 
 static unsigned int
 svga_buffer_is_referenced( struct pipe_context *pipe,
-			     struct pipe_resource *buf,
-			     unsigned face, unsigned level)
+                           struct pipe_resource *buf,
+                           unsigned level, int layer)
 {
    struct svga_screen *ss = svga_screen(pipe->screen);
    struct svga_buffer *sbuf = svga_buffer(buf);
@@ -337,6 +337,7 @@ svga_user_buffer_create(struct pipe_screen *screen,
    sbuf->b.b.width0 = bytes;
    sbuf->b.b.height0 = 1;
    sbuf->b.b.depth0 = 1;
+   sbuf->b.b.array_size = 1;
 
    sbuf->swbuf = ptr;
    sbuf->user = TRUE;
diff --git a/src/gallium/drivers/svga/svga_resource_texture.c b/src/gallium/drivers/svga/svga_resource_texture.c
index 26eb03a895..7c9e600b9f 100644
--- a/src/gallium/drivers/svga/svga_resource_texture.c
+++ b/src/gallium/drivers/svga/svga_resource_texture.c
@@ -50,8 +50,8 @@
 
 static unsigned int
 svga_texture_is_referenced( struct pipe_context *pipe,
-			    struct pipe_resource *texture,
-			    unsigned face, unsigned level)
+                            struct pipe_resource *texture,
+                            unsigned level, int layer)
 {
    struct svga_texture *tex = svga_texture(texture);
    struct svga_screen *ss = svga_screen(pipe->screen);
@@ -171,20 +171,7 @@ svga_transfer_dma_band(struct svga_context *svga,
    struct svga_texture *texture = svga_texture(st->base.resource); 
    SVGA3dCopyBox box;
    enum pipe_error ret;
-   
-   SVGA_DBG(DEBUG_DMA, "dma %s sid %p, face %u, (%u, %u, %u) - (%u, %u, %u), %ubpp\n",
-                transfer == SVGA3D_WRITE_HOST_VRAM ? "to" : "from", 
-                texture->handle,
-                st->base.sr.face,
-                st->base.box.x,
-                y,
-                st->base.box.z,
-                st->base.box.x + st->base.box.width,
-                y + h,
-                st->base.box.z + 1,
-                util_format_get_blocksize(texture->b.b.format) * 8 /
-                (util_format_get_blockwidth(texture->b.b.format)*util_format_get_blockheight(texture->b.b.format)));
-   
+ 
    box.x = st->base.box.x;
    box.y = y;
    box.z = st->base.box.z;
@@ -195,6 +182,26 @@ svga_transfer_dma_band(struct svga_context *svga,
    box.srcy = srcy;
    box.srcz = 0;
 
+   if (st->base.resource->target == PIPE_TEXTURE_CUBE) {
+      st->face = st->base.box.z;
+      box.z = 0;
+   }
+   else
+      st->face = 0;
+
+   SVGA_DBG(DEBUG_DMA, "dma %s sid %p, face %u, (%u, %u, %u) - (%u, %u, %u), %ubpp\n",
+                transfer == SVGA3D_WRITE_HOST_VRAM ? "to" : "from", 
+                texture->handle,
+                st->face,
+                st->base.box.x,
+                y,
+                box.z,
+                st->base.box.x + st->base.box.width,
+                y + h,
+                box.z + 1,
+                util_format_get_blocksize(texture->b.b.format) * 8 /
+                (util_format_get_blockwidth(texture->b.b.format)*util_format_get_blockheight(texture->b.b.format)));
+
    ret = SVGA3D_SurfaceDMA(svga->swc, st, transfer, &box, 1);
    if(ret != PIPE_OK) {
       svga->swc->flush(svga->swc, NULL);
@@ -213,7 +220,7 @@ svga_transfer_dma(struct svga_context *svga,
    struct svga_screen *screen = svga_screen(texture->b.b.screen);
    struct svga_winsys_screen *sws = screen->sws;
    struct pipe_fence_handle *fence = NULL;
-   
+
    if (transfer == SVGA3D_READ_HOST_VRAM) {
       SVGA_DBG(DEBUG_PERF, "%s: readback transfer\n", __FUNCTION__);
    }
@@ -221,7 +228,7 @@ svga_transfer_dma(struct svga_context *svga,
 
    if(!st->swbuf) {
       /* Do the DMA transfer in a single go */
-      
+
       svga_transfer_dma_band(svga, st, transfer, st->base.box.y, st->base.box.height, 0);
 
       if(transfer == SVGA3D_READ_HOST_VRAM) {
@@ -245,12 +252,12 @@ svga_transfer_dma(struct svga_context *svga,
          /* Transfer band must be aligned to pixel block boundaries */
          assert(y % blockheight == 0);
          assert(h % blockheight == 0);
-         
+
          offset = y * st->base.stride / blockheight;
          length = h * st->base.stride / blockheight;
 
          sw = (uint8_t *)st->swbuf + offset;
-         
+
          if(transfer == SVGA3D_WRITE_HOST_VRAM) {
             /* Wait for the previous DMAs to complete */
             /* TODO: keep one DMA (at half the size) in the background */
@@ -267,9 +274,9 @@ svga_transfer_dma(struct svga_context *svga,
                sws->buffer_unmap(sws, st->hwbuf);
             }
          }
-         
+
          svga_transfer_dma_band(svga, st, transfer, y, h, srcy);
-         
+
          if(transfer == SVGA3D_READ_HOST_VRAM) {
             svga_context_flush(svga, &fence);
             sws->fence_finish(sws, fence, 0);
@@ -336,10 +343,10 @@ svga_texture_destroy(struct pipe_screen *screen,
  */
 static struct pipe_transfer *
 svga_texture_get_transfer(struct pipe_context *pipe,
-			  struct pipe_resource *texture,
-			  struct pipe_subresource sr,
-			  unsigned usage,
-			  const struct pipe_box *box)
+                          struct pipe_resource *texture,
+                          unsigned level,
+                          unsigned usage,
+                          const struct pipe_box *box)
 {
    struct svga_context *svga = svga_context(pipe);
    struct svga_screen *ss = svga_screen(pipe->screen);
@@ -352,19 +359,20 @@ svga_texture_get_transfer(struct pipe_context *pipe,
    if (usage & PIPE_TRANSFER_MAP_DIRECTLY)
       return NULL;
 
+   assert(box->depth == 1);
    st = CALLOC_STRUCT(svga_transfer);
    if (!st)
       return NULL;
-   
+
    pipe_resource_reference(&st->base.resource, texture);
-   st->base.sr = sr;
+   st->base.level = level;
    st->base.usage = usage;
    st->base.box = *box;
    st->base.stride = nblocksx*util_format_get_blocksize(texture->format);
-   st->base.slice_stride = 0;
+   st->base.layer_stride = 0;
 
    st->hw_nblocksy = nblocksy;
-   
+
    st->hwbuf = svga_winsys_buffer_create(svga,
                                          1, 
                                          0,
@@ -391,7 +399,7 @@ svga_texture_get_transfer(struct pipe_context *pipe,
       if(!st->swbuf)
          goto no_swbuf;
    }
-   
+
    if (usage & PIPE_TRANSFER_READ)
       svga_transfer_dma(svga, st, SVGA3D_READ_HOST_VRAM);
 
@@ -454,8 +462,11 @@ svga_texture_transfer_destroy(struct pipe_context *pipe,
    if (st->base.usage & PIPE_TRANSFER_WRITE) {
       svga_transfer_dma(svga, st, SVGA3D_WRITE_HOST_VRAM);
       ss->texture_timestamp++;
-      tex->view_age[transfer->sr.level] = ++(tex->age);
-      tex->defined[transfer->sr.face][transfer->sr.level] = TRUE;
+      tex->view_age[transfer->level] = ++(tex->age);
+      if (transfer->resource->target == PIPE_TEXTURE_CUBE)
+         tex->defined[transfer->box.z][transfer->level] = TRUE;
+      else
+         tex->defined[0][transfer->level] = TRUE;
    }
 
    pipe_resource_reference(&st->base.resource, NULL);
@@ -490,7 +501,7 @@ svga_texture_create(struct pipe_screen *screen,
 {
    struct svga_screen *svgascreen = svga_screen(screen);
    struct svga_texture *tex = CALLOC_STRUCT(svga_texture);
-   
+
    if (!tex)
       goto error1;
 
@@ -507,7 +518,7 @@ svga_texture_create(struct pipe_screen *screen,
    tex->key.size.width = template->width0;
    tex->key.size.height = template->height0;
    tex->key.size.depth = template->depth0;
-   
+
    if(template->target == PIPE_TEXTURE_CUBE) {
       tex->key.flags |= SVGA3D_SURFACE_CUBEMAP;
       tex->key.numFaces = 6;
diff --git a/src/gallium/drivers/svga/svga_resource_texture.h b/src/gallium/drivers/svga/svga_resource_texture.h
index 631937f2eb..9a2911c2a9 100644
--- a/src/gallium/drivers/svga/svga_resource_texture.h
+++ b/src/gallium/drivers/svga/svga_resource_texture.h
@@ -85,6 +85,8 @@ struct svga_transfer
 {
    struct pipe_transfer base;
 
+   unsigned face;
+
    struct svga_winsys_buffer *hwbuf;
 
    /* Height of the hardware buffer in pixel blocks */
diff --git a/src/gallium/drivers/svga/svga_screen.c b/src/gallium/drivers/svga/svga_screen.c
index af99c41901..078190342a 100644
--- a/src/gallium/drivers/svga/svga_screen.c
+++ b/src/gallium/drivers/svga/svga_screen.c
@@ -237,6 +237,8 @@ static int svga_get_shader_param(struct pipe_screen *screen, unsigned shader, en
       case PIPE_SHADER_CAP_INDIRECT_TEMP_ADDR:
       case PIPE_SHADER_CAP_INDIRECT_CONST_ADDR:
          return 0;
+      case PIPE_SHADER_CAP_SUBROUTINES:
+         return 0;
       }
       break;
    case PIPE_SHADER_VERTEX:
@@ -276,6 +278,8 @@ static int svga_get_shader_param(struct pipe_screen *screen, unsigned shader, en
          return 0;
       case PIPE_SHADER_CAP_INDIRECT_CONST_ADDR:
          return 1;
+      case PIPE_SHADER_CAP_SUBROUTINES:
+         return 0;
       default:
          break;
       }
@@ -495,7 +499,6 @@ svga_screen_create(struct svga_winsys_screen *sws)
    screen->fence_finish = svga_fence_finish;
    svgascreen->sws = sws;
 
-   svga_screen_init_surface_functions(screen);
    svga_init_screen_resource_functions(svgascreen);
 
    svgascreen->use_ps30 =
diff --git a/src/gallium/drivers/svga/svga_state_fs.c b/src/gallium/drivers/svga/svga_state_fs.c
index ad6f294713..9c04adec8e 100644
--- a/src/gallium/drivers/svga/svga_state_fs.c
+++ b/src/gallium/drivers/svga/svga_state_fs.c
@@ -136,7 +136,7 @@ static int make_fs_key( const struct svga_context *svga,
 
    /* The blend workaround for simulating logicop xor behaviour
     * requires that the incoming fragment color be white.  This change
-    * achieves that by creating a varient of the current fragment
+    * achieves that by creating a variant of the current fragment
     * shader that overrides all output colors with 1,1,1,1
     *   
     * This will work for most shaders, including those containing
diff --git a/src/gallium/drivers/svga/svga_state_need_swtnl.c b/src/gallium/drivers/svga/svga_state_need_swtnl.c
index d34d68f535..66fea02a4b 100644
--- a/src/gallium/drivers/svga/svga_state_need_swtnl.c
+++ b/src/gallium/drivers/svga/svga_state_need_swtnl.c
@@ -114,7 +114,7 @@ static int update_need_pipeline( struct svga_context *svga,
    /* SVGA_NEW_RAST, SVGA_NEW_REDUCED_PRIMITIVE
     */
    if (svga->curr.rast->need_pipeline & (1 << svga->curr.reduced_prim)) {
-      SVGA_DBG(DEBUG_SWTNL, "%s: rast need_pipeline (%d) & prim (%x)\n", 
+      SVGA_DBG(DEBUG_SWTNL, "%s: rast need_pipeline (0x%x) & prim (0x%x)\n",
                  __FUNCTION__,
                  svga->curr.rast->need_pipeline,
                  (1 << svga->curr.reduced_prim) );
diff --git a/src/gallium/drivers/svga/svga_state_tss.c b/src/gallium/drivers/svga/svga_state_tss.c
index 4a50b19474..f8b269a101 100644
--- a/src/gallium/drivers/svga/svga_state_tss.c
+++ b/src/gallium/drivers/svga/svga_state_tss.c
@@ -238,7 +238,6 @@ update_tss(struct svga_context *svga,
          // TEXCOORDINDEX -- hopefully not needed
 
          if (svga->curr.tex_flags.flag_1d & (1 << i)) {
-            debug_printf("wrap 1d tex %d\n", i);
             EMIT_TS(svga, i, SVGA3D_TEX_ADDRESS_WRAP, ADDRESSV, fail);
          }
          else
diff --git a/src/gallium/drivers/svga/svga_surface.c b/src/gallium/drivers/svga/svga_surface.c
index b21dc5fd9a..3e4bed76c0 100644
--- a/src/gallium/drivers/svga/svga_surface.c
+++ b/src/gallium/drivers/svga/svga_surface.c
@@ -179,36 +179,50 @@ svga_texture_view_surface(struct pipe_context *pipe,
 
 
 static struct pipe_surface *
-svga_get_tex_surface(struct pipe_screen *screen,
-                     struct pipe_resource *pt,
-                     unsigned face, unsigned level, unsigned zslice,
-                     unsigned flags)
+svga_create_surface(struct pipe_context *pipe,
+                    struct pipe_resource *pt,
+                    const struct pipe_surface *surf_tmpl)
 {
    struct svga_texture *tex = svga_texture(pt);
+   struct pipe_screen *screen = pipe->screen;
    struct svga_surface *s;
-   boolean render = (flags & (PIPE_BIND_RENDER_TARGET |
-			      PIPE_BIND_DEPTH_STENCIL)) ? TRUE : FALSE;
+   unsigned face, zslice;
+   /* XXX surfaces should only be used for rendering purposes nowadays */
+   boolean render = (surf_tmpl->usage & (PIPE_BIND_RENDER_TARGET |
+                                         PIPE_BIND_DEPTH_STENCIL)) ? TRUE : FALSE;
    boolean view = FALSE;
    SVGA3dSurfaceFormat format;
 
+   assert(surf_tmpl->u.tex.first_layer == surf_tmpl->u.tex.last_layer);
+
    s = CALLOC_STRUCT(svga_surface);
    if (!s)
       return NULL;
 
+   if (pt->target == PIPE_TEXTURE_CUBE) {
+         face = surf_tmpl->u.tex.first_layer;
+         zslice = 0;
+   }
+   else {
+      face = 0;
+      zslice = surf_tmpl->u.tex.first_layer;
+   }
+
    pipe_reference_init(&s->base.reference, 1);
    pipe_resource_reference(&s->base.texture, pt);
-   s->base.format = pt->format;
-   s->base.width = u_minify(pt->width0, level);
-   s->base.height = u_minify(pt->height0, level);
-   s->base.usage = flags;
-   s->base.level = level;
-   s->base.face = face;
-   s->base.zslice = zslice;
+   s->base.context = pipe;
+   s->base.format = surf_tmpl->format;
+   s->base.width = u_minify(pt->width0, surf_tmpl->u.tex.level);
+   s->base.height = u_minify(pt->height0, surf_tmpl->u.tex.level);
+   s->base.usage = surf_tmpl->usage;
+   s->base.u.tex.level = surf_tmpl->u.tex.level;
+   s->base.u.tex.first_layer = surf_tmpl->u.tex.first_layer;
+   s->base.u.tex.last_layer = surf_tmpl->u.tex.last_layer;
 
    if (!render)
-      format = svga_translate_format(pt->format);
+      format = svga_translate_format(surf_tmpl->format);
    else
-      format = svga_translate_format_render(pt->format);
+      format = svga_translate_format_render(surf_tmpl->format);
 
    assert(format != SVGA3D_FORMAT_INVALID);
 
@@ -217,11 +231,11 @@ svga_get_tex_surface(struct pipe_screen *screen,
 
    /* Currently only used for compressed textures */
    if (render && 
-       format != svga_translate_format(pt->format)) {
+       format != svga_translate_format(surf_tmpl->format)) {
       view = TRUE;
    }
 
-   if (level != 0 && 
+   if (surf_tmpl->u.tex.level != 0 &&
        svga_screen(screen)->debug.force_level_surface_view)
       view = TRUE;
 
@@ -233,22 +247,22 @@ svga_get_tex_surface(struct pipe_screen *screen,
 
    if (view) {
       SVGA_DBG(DEBUG_VIEWS, "svga: Surface view: yes %p, level %u face %u z %u, %p\n",
-               pt, level, face, zslice, s);
+               pt, surf_tmpl->u.tex.level, face, zslice, s);
 
-      s->handle = svga_texture_view_surface(NULL, tex, format, level, 1, face, zslice,
-                                            &s->key);
+      s->handle = svga_texture_view_surface(NULL, tex, format, surf_tmpl->u.tex.level,
+	                                    1, face, zslice, &s->key);
       s->real_face = 0;
       s->real_level = 0;
       s->real_zslice = 0;
    } else {
       SVGA_DBG(DEBUG_VIEWS, "svga: Surface view: no %p, level %u, face %u, z %u, %p\n",
-               pt, level, face, zslice, s);
+               pt, surf_tmpl->u.tex.level, face, zslice, s);
 
       memset(&s->key, 0, sizeof s->key);
       s->handle = tex->handle;
       s->real_face = face;
-      s->real_level = level;
       s->real_zslice = zslice;
+      s->real_level = surf_tmpl->u.tex.level;
    }
 
    return &s->base;
@@ -256,7 +270,8 @@ svga_get_tex_surface(struct pipe_screen *screen,
 
 
 static void
-svga_tex_surface_destroy(struct pipe_surface *surf)
+svga_surface_destroy(struct pipe_context *pipe,
+                     struct pipe_surface *surf)
 {
    struct svga_surface *s = svga_surface(surf);
    struct svga_texture *t = svga_texture(surf->texture);
@@ -282,8 +297,13 @@ svga_mark_surface_dirty(struct pipe_surface *surf)
 
       s->dirty = TRUE;
 
-      if (s->handle == tex->handle)
-         tex->defined[surf->face][surf->level] = TRUE;
+      if (s->handle == tex->handle) {
+         /* hmm so 3d textures always have all their slices marked ? */
+         if (surf->texture->target == PIPE_TEXTURE_CUBE)
+            tex->defined[surf->u.tex.first_layer][surf->u.tex.level] = TRUE;
+         else
+            tex->defined[0][surf->u.tex.level] = TRUE;
+      }
       else {
          /* this will happen later in svga_propagate_surface */
       }
@@ -314,22 +334,32 @@ svga_propagate_surface(struct pipe_context *pipe, struct pipe_surface *surf)
    struct svga_surface *s = svga_surface(surf);
    struct svga_texture *tex = svga_texture(surf->texture);
    struct svga_screen *ss = svga_screen(surf->texture->screen);
+   unsigned zslice, face;
 
    if (!s->dirty)
       return;
 
+   if (surf->texture->target == PIPE_TEXTURE_CUBE) {
+      zslice = 0;
+      face = surf->u.tex.first_layer;
+   }
+   else {
+      zslice = surf->u.tex.first_layer;
+      face = 0;
+   }
+
    s->dirty = FALSE;
    ss->texture_timestamp++;
-   tex->view_age[surf->level] = ++(tex->age);
+   tex->view_age[surf->u.tex.level] = ++(tex->age);
 
    if (s->handle != tex->handle) {
-      SVGA_DBG(DEBUG_VIEWS, "svga: Surface propagate: tex %p, level %u, from %p\n", tex, surf->level, surf);
+      SVGA_DBG(DEBUG_VIEWS, "svga: Surface propagate: tex %p, level %u, from %p\n", tex, surf->u.tex.level, surf);
       svga_texture_copy_handle(svga_context(pipe),
                                s->handle, 0, 0, 0, s->real_level, s->real_face,
-                               tex->handle, 0, 0, surf->zslice, surf->level, surf->face,
-                               u_minify(tex->b.b.width0, surf->level),
-                               u_minify(tex->b.b.height0, surf->level), 1);
-      tex->defined[surf->face][surf->level] = TRUE;
+                               tex->handle, 0, 0, zslice, surf->u.tex.level, face,
+                               u_minify(tex->b.b.width0, surf->u.tex.level),
+                               u_minify(tex->b.b.height0, surf->u.tex.level), 1);
+      tex->defined[face][surf->u.tex.level] = TRUE;
    }
 }
 
@@ -351,9 +381,9 @@ svga_surface_needs_propagation(struct pipe_surface *surf)
 
 
 void
-svga_screen_init_surface_functions(struct pipe_screen *screen)
+svga_init_surface_functions(struct svga_context *svga)
 {
-   screen->get_tex_surface = svga_get_tex_surface;
-   screen->tex_surface_destroy = svga_tex_surface_destroy;
+   svga->pipe.create_surface = svga_create_surface;
+   svga->pipe.surface_destroy = svga_surface_destroy;
 }
 
diff --git a/src/gallium/drivers/svga/svga_surface.h b/src/gallium/drivers/svga/svga_surface.h
index 13bd5b19b6..afb8326e1f 100644
--- a/src/gallium/drivers/svga/svga_surface.h
+++ b/src/gallium/drivers/svga/svga_surface.h
@@ -90,7 +90,4 @@ svga_surface(struct pipe_surface *surface)
    return (struct svga_surface *)surface;
 }
 
-void
-svga_screen_init_surface_functions(struct pipe_screen *screen);
-
 #endif
diff --git a/src/gallium/drivers/sw/SConscript b/src/gallium/drivers/sw/SConscript
deleted file mode 100644
index 40d01db2f6..0000000000
--- a/src/gallium/drivers/sw/SConscript
+++ /dev/null
@@ -1,37 +0,0 @@
-#######################################################################
-# SConscript for swrast convenience library
-#
-# This is a meta-driver which consists of any and all of the software
-# rasterizers into a single driver.  A software rasterizer is defined
-# as any driver which takes an sw_winsys pointer as the only argument
-# to create_screen.
-
-Import('*')
-
-env = env.Clone()
-
-# To avoid targets having to check extensively or add drivers on a whim, append
-# all referenced extra drivers to the exported symbol.
-extra = []
-if True:
-    env.Append(CPPDEFINES = 'GALLIUM_SOFTPIPE')
-    env.Prepend(LIBS = [softpipe])
-    extra.append(softpipe)
-
-if env['llvm']:
-    env.Append(CPPDEFINES = 'GALLIUM_LLVMPIPE')
-    env.Prepend(LIBS = [llvmpipe])
-    extra.append(llvmpipe)
-
-if 'cell' in env['drivers']:
-    env.Append(CPPDEFINES = 'GALLIUM_CELL')
-    env.Prepend(LIBS = [cell])
-    extra.append(cell)
-
-sw = env.ConvenienceLibrary(
-	target = 'sw',
-	source = [
-		'sw.c',
-		]
-    ) + extra
-Export('sw')
diff --git a/src/gallium/drivers/sw/sw.c b/src/gallium/drivers/sw/sw.c
deleted file mode 100644
index 6b873ecc1b..0000000000
--- a/src/gallium/drivers/sw/sw.c
+++ /dev/null
@@ -1,58 +0,0 @@
-#include "pipe/p_compiler.h"
-#include "util/u_debug.h"
-#include "target-helpers/wrap_screen.h"
-#include "sw_public.h"
-
-
-/* Helper function to choose and instantiate one of the software rasterizers:
- * cell, llvmpipe, softpipe.
- */
-
-#ifdef GALLIUM_SOFTPIPE
-#include "softpipe/sp_public.h"
-#endif
-
-#ifdef GALLIUM_LLVMPIPE
-#include "llvmpipe/lp_public.h"
-#endif
-
-#ifdef GALLIUM_CELL
-#include "cell/ppu/cell_public.h"
-#endif
-
-struct pipe_screen *
-swrast_create_screen(struct sw_winsys *winsys)
-{
-   const char *default_driver;
-   const char *driver;
-   struct pipe_screen *screen = NULL;
-
-#if defined(GALLIUM_CELL)
-   default_driver = "cell";
-#elif defined(GALLIUM_LLVMPIPE)
-   default_driver = "llvmpipe";
-#elif defined(GALLIUM_SOFTPIPE)
-   default_driver = "softpipe";
-#else
-   default_driver = "";
-#endif
-
-   driver = debug_get_option("GALLIUM_DRIVER", default_driver);
-
-#if defined(GALLIUM_CELL)
-   if (screen == NULL && strcmp(driver, "cell") == 0)
-      screen = cell_create_screen( winsys );
-#endif
-
-#if defined(GALLIUM_LLVMPIPE)
-   if (screen == NULL && strcmp(driver, "llvmpipe") == 0)
-      screen = llvmpipe_create_screen( winsys );
-#endif
-
-#if defined(GALLIUM_SOFTPIPE)
-   if (screen == NULL)
-      screen = softpipe_create_screen( winsys );
-#endif
-
-   return screen;
-}
diff --git a/src/gallium/drivers/sw/sw_public.h b/src/gallium/drivers/sw/sw_public.h
deleted file mode 100644
index 7085c5c85a..0000000000
--- a/src/gallium/drivers/sw/sw_public.h
+++ /dev/null
@@ -1,13 +0,0 @@
-#ifndef SW_PUBLIC_H
-#define SW_PUBLIC_H
-
-/* A convenience library, primarily to isolate the logic required to
- * figure out which if any software rasterizers have been built and
- * select between them.
- */
-struct sw_winsys;
-
-struct pipe_screen *
-swrast_create_screen(struct sw_winsys *winsys);
-
-#endif
diff --git a/src/gallium/drivers/trace/tr_context.c b/src/gallium/drivers/trace/tr_context.c
index 04f30f82c3..eaabae8ce4 100644
--- a/src/gallium/drivers/trace/tr_context.c
+++ b/src/gallium/drivers/trace/tr_context.c
@@ -314,6 +314,9 @@ trace_context_bind_vertex_sampler_states(struct pipe_context *_pipe,
    struct trace_context *tr_ctx = trace_context(_pipe);
    struct pipe_context *pipe = tr_ctx->pipe;
 
+   if (!pipe->bind_vertex_sampler_states)
+      return;
+
    trace_dump_call_begin("pipe_context", "bind_vertex_sampler_states");
 
    trace_dump_arg(ptr, pipe);
@@ -885,6 +888,60 @@ trace_sampler_view_destroy(struct pipe_context *_pipe,
    FREE(_view);
 }
 
+/********************************************************************
+ * surface
+ */
+
+
+static struct pipe_surface *
+trace_create_surface(struct pipe_context *_pipe,
+                     struct pipe_resource *_texture,
+                     const struct pipe_surface *surf_tmpl)
+{
+   struct trace_context *tr_ctx = trace_context(_pipe);
+   struct trace_resource *tr_tex = trace_resource(_texture);
+   struct pipe_context *pipe = tr_ctx->pipe;
+   struct pipe_resource *texture = tr_tex->resource;
+   struct pipe_surface *result = NULL;
+
+   trace_dump_call_begin("pipe_context", "create_surface");
+
+   trace_dump_arg(ptr, pipe);
+   trace_dump_arg(ptr, texture);
+   /* hmm some values unitialized there */
+   trace_dump_arg(surface, surf_tmpl);
+
+   result = pipe->create_surface(pipe, texture, surf_tmpl);
+
+   trace_dump_ret(ptr, result);
+
+   trace_dump_call_end();
+
+   result = trace_surf_create(tr_tex, result);
+
+   return result;
+}
+
+
+static void
+trace_surface_destroy(struct pipe_context *_pipe,
+                      struct pipe_surface *_surface)
+{
+   struct trace_context *tr_ctx = trace_context(_pipe);
+   struct pipe_context *pipe = tr_ctx->pipe;
+   struct trace_surface *tr_surf = trace_surface(_surface);
+   struct pipe_surface *surface = tr_surf->surface;
+
+   trace_dump_call_begin("pipe_context", "surface_destroy");
+
+   trace_dump_arg(ptr, pipe);
+   trace_dump_arg(ptr, surface);
+
+   trace_dump_call_end();
+
+   trace_surf_destroy(tr_surf);
+}
+
 
 static INLINE void
 trace_context_set_fragment_sampler_views(struct pipe_context *_pipe,
@@ -926,6 +983,9 @@ trace_context_set_vertex_sampler_views(struct pipe_context *_pipe,
    struct pipe_sampler_view *unwrapped_views[PIPE_MAX_VERTEX_SAMPLERS];
    unsigned i;
 
+   if (!pipe->set_vertex_sampler_views)
+      return;
+
    for(i = 0; i < num; ++i) {
       tr_view = trace_sampler_view(views[i]);
       unwrapped_views[i] = tr_view ? tr_view->sampler_view : NULL;
@@ -1004,12 +1064,11 @@ trace_context_set_index_buffer(struct pipe_context *_pipe,
 static INLINE void
 trace_context_resource_copy_region(struct pipe_context *_pipe,
                                    struct pipe_resource *dst,
-                                   struct pipe_subresource subdst,
+                                   unsigned dst_level,
                                    unsigned dstx, unsigned dsty, unsigned dstz,
                                    struct pipe_resource *src,
-                                   struct pipe_subresource subsrc,
-                                   unsigned srcx, unsigned srcy, unsigned srcz,
-                                   unsigned width, unsigned height)
+                                   unsigned src_level,
+                                   const struct pipe_box *src_box)
 {
    struct trace_context *tr_ctx = trace_context(_pipe);
    struct pipe_context *pipe = tr_ctx->pipe;
@@ -1021,21 +1080,17 @@ trace_context_resource_copy_region(struct pipe_context *_pipe,
 
    trace_dump_arg(ptr, pipe);
    trace_dump_arg(ptr, dst);
-   trace_dump_arg_struct(subresource, subdst);
+   trace_dump_arg(uint, dst_level);
    trace_dump_arg(uint, dstx);
    trace_dump_arg(uint, dsty);
    trace_dump_arg(uint, dstz);
    trace_dump_arg(ptr, src);
-   trace_dump_arg_struct(subresource, subsrc);
-   trace_dump_arg(uint, srcx);
-   trace_dump_arg(uint, srcy);
-   trace_dump_arg(uint, srcz);
-   trace_dump_arg(uint, width);
-   trace_dump_arg(uint, height);
+   trace_dump_arg(uint, src_level);
+   trace_dump_arg(box, src_box);
 
    pipe->resource_copy_region(pipe,
-                              dst, subdst, dstx, dsty, dstz,
-                              src, subsrc, srcx, srcy, srcz, width, height);
+                              dst, dst_level, dstx, dsty, dstz,
+                              src, src_level, src_box);
 
    trace_dump_call_end();
 }
@@ -1166,8 +1221,8 @@ trace_context_destroy(struct pipe_context *_pipe)
 
 static unsigned int
 trace_is_resource_referenced( struct pipe_context *_pipe,
-			      struct pipe_resource *_resource,
-			      unsigned face, unsigned level)
+                              struct pipe_resource *_resource,
+                              unsigned level, int layer)
 {
    struct trace_context *tr_ctx = trace_context(_pipe);
    struct trace_resource *tr_tex = trace_resource(_resource);
@@ -1178,10 +1233,10 @@ trace_is_resource_referenced( struct pipe_context *_pipe,
    trace_dump_call_begin("pipe_context", "is_resource_referenced");
    trace_dump_arg(ptr, pipe);
    trace_dump_arg(ptr, texture);
-   trace_dump_arg(uint, face);
    trace_dump_arg(uint, level);
+   trace_dump_arg(int, layer);
 
-   referenced = pipe->is_resource_referenced(pipe, texture, face, level);
+   referenced = pipe->is_resource_referenced(pipe, texture, level, layer);
 
    trace_dump_ret(uint, referenced);
    trace_dump_call_end();
@@ -1197,10 +1252,10 @@ trace_is_resource_referenced( struct pipe_context *_pipe,
 
 static struct pipe_transfer *
 trace_context_get_transfer(struct pipe_context *_context,
-			   struct pipe_resource *_resource,
-			   struct pipe_subresource sr,
-			   unsigned usage,
-			   const struct pipe_box *box)
+                           struct pipe_resource *_resource,
+                           unsigned level,
+                           unsigned usage,
+                           const struct pipe_box *box)
 {
    struct trace_context *tr_context = trace_context(_context);
    struct trace_resource *tr_tex = trace_resource(_resource);
@@ -1215,7 +1270,7 @@ trace_context_get_transfer(struct pipe_context *_context,
     * to transfer_inline_write and ignore read transfers.
     */
 
-   result = context->get_transfer(context, texture, sr, usage, box);
+   result = context->get_transfer(context, texture, level, usage, box);
 
    if (result)
       result = trace_transfer_create(tr_context, tr_tex, result);
@@ -1226,7 +1281,7 @@ trace_context_get_transfer(struct pipe_context *_context,
 
 static void
 trace_context_transfer_destroy(struct pipe_context *_context,
-                                   struct pipe_transfer *_transfer)
+                               struct pipe_transfer *_transfer)
 {
    struct trace_context *tr_context = trace_context(_context);
    struct trace_transfer *tr_trans = trace_transfer(_transfer);
@@ -1274,7 +1329,7 @@ trace_context_transfer_flush_region( struct pipe_context *_context,
 
 static void
 trace_context_transfer_unmap(struct pipe_context *_context,
-			     struct pipe_transfer *_transfer)
+                             struct pipe_transfer *_transfer)
 {
    struct trace_context *tr_ctx = trace_context(_context);
    struct trace_transfer *tr_trans = trace_transfer(_transfer);
@@ -1287,17 +1342,17 @@ trace_context_transfer_unmap(struct pipe_context *_context,
        */
 
       struct pipe_resource *resource = transfer->resource;
-      struct pipe_subresource sr = transfer->sr;
+      unsigned level = transfer->level;
       unsigned usage = transfer->usage;
       const struct pipe_box *box = &transfer->box;
       unsigned stride = transfer->stride;
-      unsigned slice_stride = transfer->slice_stride;
+      unsigned layer_stride = transfer->layer_stride;
 
       trace_dump_call_begin("pipe_context", "transfer_inline_write");
 
       trace_dump_arg(ptr, context);
       trace_dump_arg(ptr, resource);
-      trace_dump_arg_struct(subresource, sr);
+      trace_dump_arg(uint, level);
       trace_dump_arg(uint, usage);
       trace_dump_arg(box, box);
 
@@ -1306,11 +1361,11 @@ trace_context_transfer_unmap(struct pipe_context *_context,
                            resource->format,
                            box,
                            stride,
-                           slice_stride);
+                           layer_stride);
       trace_dump_arg_end();
 
       trace_dump_arg(uint, stride);
-      trace_dump_arg(uint, slice_stride);
+      trace_dump_arg(uint, layer_stride);
 
       trace_dump_call_end();
 
@@ -1323,13 +1378,13 @@ trace_context_transfer_unmap(struct pipe_context *_context,
 
 static void
 trace_context_transfer_inline_write(struct pipe_context *_context,
-				    struct pipe_resource *_resource,
-				    struct pipe_subresource sr,
-				    unsigned usage,
-				    const struct pipe_box *box,
-				    const void *data,
-				    unsigned stride,
-				    unsigned slice_stride)
+                                    struct pipe_resource *_resource,
+                                    unsigned level,
+                                    unsigned usage,
+                                    const struct pipe_box *box,
+                                    const void *data,
+                                    unsigned stride,
+                                    unsigned layer_stride)
 {
    struct trace_context *tr_context = trace_context(_context);
    struct trace_resource *tr_tex = trace_resource(_resource);
@@ -1342,7 +1397,7 @@ trace_context_transfer_inline_write(struct pipe_context *_context,
 
    trace_dump_arg(ptr, context);
    trace_dump_arg(ptr, resource);
-   trace_dump_arg_struct(subresource, sr);
+   trace_dump_arg(uint, level);
    trace_dump_arg(uint, usage);
    trace_dump_arg(box, box);
 
@@ -1351,16 +1406,16 @@ trace_context_transfer_inline_write(struct pipe_context *_context,
                         resource->format,
                         box,
                         stride,
-                        slice_stride);
+                        layer_stride);
    trace_dump_arg_end();
 
    trace_dump_arg(uint, stride);
-   trace_dump_arg(uint, slice_stride);
+   trace_dump_arg(uint, layer_stride);
 
    trace_dump_call_end();
 
    context->transfer_inline_write(context, resource,
-				  sr, usage, box, data, stride, slice_stride);
+                                  level, usage, box, data, stride, layer_stride);
 }
 
 
@@ -1434,6 +1489,8 @@ trace_context_create(struct trace_screen *tr_scr,
    tr_ctx->base.set_vertex_sampler_views = trace_context_set_vertex_sampler_views;
    tr_ctx->base.create_sampler_view = trace_create_sampler_view;
    tr_ctx->base.sampler_view_destroy = trace_sampler_view_destroy;
+   tr_ctx->base.create_surface = trace_create_surface;
+   tr_ctx->base.surface_destroy = trace_surface_destroy;
    tr_ctx->base.set_vertex_buffers = trace_context_set_vertex_buffers;
    tr_ctx->base.set_index_buffer = trace_context_set_index_buffer;
    tr_ctx->base.resource_copy_region = trace_context_resource_copy_region;
diff --git a/src/gallium/drivers/trace/tr_dump_state.c b/src/gallium/drivers/trace/tr_dump_state.c
index 8f81606032..155c869fbd 100644
--- a/src/gallium/drivers/trace/tr_dump_state.c
+++ b/src/gallium/drivers/trace/tr_dump_state.c
@@ -71,6 +71,10 @@ void trace_dump_resource_template(const struct pipe_resource *templat)
    trace_dump_uint(templat->depth0);
    trace_dump_member_end();
 
+   trace_dump_member_begin("array_size");
+   trace_dump_uint(templat->array_size);
+   trace_dump_member_end();
+
    trace_dump_member(uint, templat, last_level);
    trace_dump_member(uint, templat, usage);
    trace_dump_member(uint, templat, bind);
@@ -80,25 +84,6 @@ void trace_dump_resource_template(const struct pipe_resource *templat)
 }
 
 
-void trace_dump_subresource(const struct pipe_subresource *subresource)
-{
-   if (!trace_dumping_enabled_locked())
-      return;
-
-   if(!subresource) {
-      trace_dump_null();
-      return;
-   }
-
-   trace_dump_struct_begin("pipe_subresource");
-
-   trace_dump_member(uint, subresource, face);
-   trace_dump_member(uint, subresource, level);
-
-   trace_dump_struct_end();
-}
-
-
 void trace_dump_box(const struct pipe_box *box)
 {
    if (!trace_dumping_enabled_locked())
@@ -445,8 +430,13 @@ void trace_dump_sampler_view_template(const struct pipe_sampler_view *state)
    trace_dump_struct_begin("pipe_sampler_view");
 
    trace_dump_member(format, state, format);
-   trace_dump_member(uint, state, first_level);
-   trace_dump_member(uint, state, last_level);
+   /* XXX */
+   trace_dump_member(uint, state, u.tex.first_level);
+   trace_dump_member(uint, state, u.tex.last_level);
+   trace_dump_member(uint, state, u.tex.first_layer);
+   trace_dump_member(uint, state, u.tex.last_layer);
+   trace_dump_member(uint, state, u.buf.first_element);
+   trace_dump_member(uint, state, u.buf.last_element);
    trace_dump_member(uint, state, swizzle_r);
    trace_dump_member(uint, state, swizzle_g);
    trace_dump_member(uint, state, swizzle_b);
@@ -472,14 +462,14 @@ void trace_dump_surface(const struct pipe_surface *state)
    trace_dump_member(uint, state, width);
    trace_dump_member(uint, state, height);
 
-   trace_dump_member(uint, state, layout);
-   trace_dump_member(uint, state, offset);
    trace_dump_member(uint, state, usage);
 
    trace_dump_member(ptr, state, texture);
-   trace_dump_member(uint, state, face);
-   trace_dump_member(uint, state, level);
-   trace_dump_member(uint, state, zslice);
+   trace_dump_member(uint, state, u.tex.level);
+   trace_dump_member(uint, state, u.tex.first_layer);
+   trace_dump_member(uint, state, u.tex.last_layer);
+   trace_dump_member(uint, state, u.buf.first_element);
+   trace_dump_member(uint, state, u.buf.last_element);
 
    trace_dump_struct_end();
 }
@@ -497,16 +487,18 @@ void trace_dump_transfer(const struct pipe_transfer *state)
 
    trace_dump_struct_begin("pipe_transfer");
 
+   trace_dump_member(uint, state, box.x);
+   trace_dump_member(uint, state, box.y);
+   trace_dump_member(uint, state, box.z);
    trace_dump_member(uint, state, box.width);
    trace_dump_member(uint, state, box.height);
+   trace_dump_member(uint, state, box.depth);
 
    trace_dump_member(uint, state, stride);
+   trace_dump_member(uint, state, layer_stride);
    trace_dump_member(uint, state, usage);
 
    trace_dump_member(ptr, state, resource);
-   trace_dump_member(uint, state, sr.face);
-   trace_dump_member(uint, state, sr.level);
-   trace_dump_member(uint, state, box.z);
 
    trace_dump_struct_end();
 }
diff --git a/src/gallium/drivers/trace/tr_dump_state.h b/src/gallium/drivers/trace/tr_dump_state.h
index 078d208610..fe8ece78d4 100644
--- a/src/gallium/drivers/trace/tr_dump_state.h
+++ b/src/gallium/drivers/trace/tr_dump_state.h
@@ -37,8 +37,6 @@ void trace_dump_format(enum pipe_format format);
 
 void trace_dump_resource_template(const struct pipe_resource *templat);
 
-void trace_dump_subresource(const struct pipe_subresource *subresource);
-
 void trace_dump_box(const struct pipe_box *box);
 
 void trace_dump_rasterizer_state(const struct pipe_rasterizer_state *state);
diff --git a/src/gallium/drivers/trace/tr_screen.c b/src/gallium/drivers/trace/tr_screen.c
index 935831071e..c2de2daa88 100644
--- a/src/gallium/drivers/trace/tr_screen.c
+++ b/src/gallium/drivers/trace/tr_screen.c
@@ -210,23 +210,26 @@ trace_screen_context_create(struct pipe_screen *_screen, void *priv)
 
 static void
 trace_screen_flush_frontbuffer(struct pipe_screen *_screen,
-                               struct pipe_surface *_surface,
+                               struct pipe_resource *_resource,
+                               unsigned level, unsigned layer,
                                void *context_private)
 {
    struct trace_screen *tr_scr = trace_screen(_screen);
-   struct trace_surface *tr_surf = trace_surface(_surface);
+   struct trace_resource *tr_res = trace_resource(_resource);
    struct pipe_screen *screen = tr_scr->screen;
-   struct pipe_surface *surface = tr_surf->surface;
+   struct pipe_resource *resource = tr_res->resource;
 
    trace_dump_call_begin("pipe_screen", "flush_frontbuffer");
 
    trace_dump_arg(ptr, screen);
-   trace_dump_arg(ptr, surface);
+   trace_dump_arg(ptr, resource);
+   trace_dump_arg(uint, level);
+   trace_dump_arg(uint, layer);
    /* XXX: hide, as there is nothing we can do with this
    trace_dump_arg(ptr, context_private);
    */
 
-   screen->flush_frontbuffer(screen, surface, context_private);
+   screen->flush_frontbuffer(screen, resource, level, layer, context_private);
 
    trace_dump_call_end();
 }
@@ -318,68 +321,6 @@ trace_screen_resource_destroy(struct pipe_screen *_screen,
 }
 
 
-/********************************************************************
- * surface
- */
-
-
-static struct pipe_surface *
-trace_screen_get_tex_surface(struct pipe_screen *_screen,
-                             struct pipe_resource *_texture,
-                             unsigned face, unsigned level,
-                             unsigned zslice,
-                             unsigned usage)
-{
-   struct trace_screen *tr_scr = trace_screen(_screen);
-   struct trace_resource *tr_tex = trace_resource(_texture);
-   struct pipe_screen *screen = tr_scr->screen;
-   struct pipe_resource *texture = tr_tex->resource;
-   struct pipe_surface *result = NULL;
-
-   assert(texture->screen == screen);
-
-   trace_dump_call_begin("pipe_screen", "get_tex_surface");
-
-   trace_dump_arg(ptr, screen);
-   trace_dump_arg(ptr, texture);
-   trace_dump_arg(uint, face);
-   trace_dump_arg(uint, level);
-   trace_dump_arg(uint, zslice);
-   trace_dump_arg(uint, usage);
-
-   result = screen->get_tex_surface(screen, texture, face, level, zslice, usage);
-
-   trace_dump_ret(ptr, result);
-
-   trace_dump_call_end();
-
-   result = trace_surface_create(tr_tex, result);
-
-   return result;
-}
-
-
-static void
-trace_screen_tex_surface_destroy(struct pipe_surface *_surface)
-{
-   struct trace_screen *tr_scr = trace_screen(_surface->texture->screen);
-   struct trace_surface *tr_surf = trace_surface(_surface);
-   struct pipe_screen *screen = tr_scr->screen;
-   struct pipe_surface *surface = tr_surf->surface;
-
-   trace_dump_call_begin("pipe_screen", "tex_surface_destroy");
-
-   trace_dump_arg(ptr, screen);
-   trace_dump_arg(ptr, surface);
-
-   trace_dump_call_end();
-
-   trace_surface_destroy(tr_surf);
-}
-
-
-
-
 
 /********************************************************************
  * buffer
@@ -580,8 +521,6 @@ trace_screen_create(struct pipe_screen *screen)
    tr_scr->base.resource_from_handle = trace_screen_resource_from_handle;
    tr_scr->base.resource_get_handle = trace_screen_resource_get_handle;
    tr_scr->base.resource_destroy = trace_screen_resource_destroy;
-   tr_scr->base.get_tex_surface = trace_screen_get_tex_surface;
-   tr_scr->base.tex_surface_destroy = trace_screen_tex_surface_destroy;
    tr_scr->base.user_buffer_create = trace_screen_user_buffer_create;
    tr_scr->base.fence_reference = trace_screen_fence_reference;
    tr_scr->base.fence_signalled = trace_screen_fence_signalled;
diff --git a/src/gallium/drivers/trace/tr_texture.c b/src/gallium/drivers/trace/tr_texture.c
index 9914b98b39..2799734647 100644
--- a/src/gallium/drivers/trace/tr_texture.c
+++ b/src/gallium/drivers/trace/tr_texture.c
@@ -74,8 +74,8 @@ trace_resource_destroy(struct trace_screen *tr_scr,
 
 
 struct pipe_surface *
-trace_surface_create(struct trace_resource *tr_tex,
-                     struct pipe_surface *surface)
+trace_surf_create(struct trace_resource *tr_tex,
+                  struct pipe_surface *surface)
 {
    struct trace_surface *tr_surf;
 
@@ -104,7 +104,7 @@ error:
 
 
 void
-trace_surface_destroy(struct trace_surface *tr_surf)
+trace_surf_destroy(struct trace_surface *tr_surf)
 {
    pipe_resource_reference(&tr_surf->base.texture, NULL);
    pipe_surface_reference(&tr_surf->surface, NULL);
diff --git a/src/gallium/drivers/trace/tr_texture.h b/src/gallium/drivers/trace/tr_texture.h
index 6513995d50..3352c96e59 100644
--- a/src/gallium/drivers/trace/tr_texture.h
+++ b/src/gallium/drivers/trace/tr_texture.h
@@ -125,11 +125,11 @@ trace_resource_destroy(struct trace_screen *tr_scr,
 		       struct trace_resource *tr_tex);
 
 struct pipe_surface *
-trace_surface_create(struct trace_resource *tr_tex,
+trace_surf_create(struct trace_resource *tr_tex,
                      struct pipe_surface *surface);
 
 void
-trace_surface_destroy(struct trace_surface *tr_surf);
+trace_surf_destroy(struct trace_surface *tr_surf);
 
 struct pipe_transfer *
 trace_transfer_create(struct trace_context *tr_ctx,
diff --git a/src/gallium/include/pipe/p_context.h b/src/gallium/include/pipe/p_context.h
index 0e53aef6d2..589cac2ddd 100644
--- a/src/gallium/include/pipe/p_context.h
+++ b/src/gallium/include/pipe/p_context.h
@@ -54,7 +54,6 @@ struct pipe_scissor_state;
 struct pipe_shader_state;
 struct pipe_stencil_ref;
 struct pipe_stream_output_state;
-struct pipe_subresource;
 struct pipe_surface;
 struct pipe_vertex_buffer;
 struct pipe_vertex_element;
@@ -256,12 +255,11 @@ struct pipe_context {
     */
    void (*resource_copy_region)(struct pipe_context *pipe,
                                 struct pipe_resource *dst,
-                                struct pipe_subresource subdst,
+                                unsigned dst_level,
                                 unsigned dstx, unsigned dsty, unsigned dstz,
                                 struct pipe_resource *src,
-                                struct pipe_subresource subsrc,
-                                unsigned srcx, unsigned srcy, unsigned srcz,
-                                unsigned width, unsigned height);
+                                unsigned src_level,
+                                const struct pipe_box *src_box);
 
    /**
     * Resolve a multisampled resource into a non-multisampled one.
@@ -269,9 +267,9 @@ struct pipe_context {
     */
    void (*resource_resolve)(struct pipe_context *pipe,
                             struct pipe_resource *dst,
-                            struct pipe_subresource subdst,
+                            unsigned dst_layer,
                             struct pipe_resource *src,
-                            struct pipe_subresource subsrc);
+                            unsigned src_layer);
 
    /*@}*/
 
@@ -328,13 +326,13 @@ struct pipe_context {
     * PIPE_REFERENCED_FOR_READ | PIPE_REFERENCED_FOR_WRITE.
     * \param pipe  context whose unflushed hw commands will be checked.
     * \param texture  texture to check.
-    * \param face  cubemap face. Use 0 for non-cubemap texture.
     * \param level  mipmap level.
+    * \param layer  cubemap face, 2d array or 3d slice, 0 otherwise. Use -1 for any layer.
     * \return mask of PIPE_REFERENCED_FOR_READ/WRITE or PIPE_UNREFERENCED
     */
    unsigned int (*is_resource_referenced)(struct pipe_context *pipe,
-					  struct pipe_resource *texture,
-					  unsigned face, unsigned level);
+                                          struct pipe_resource *texture,
+                                          unsigned level, int layer);
 
    /**
     * Create a view on a texture to be used by a shader stage.
@@ -348,20 +346,32 @@ struct pipe_context {
 
 
    /**
+    * Get a surface which is a "view" into a resource, used by
+    * render target / depth stencil stages.
+    * \param usage  bitmaks of PIPE_BIND_* flags
+    */
+   struct pipe_surface *(*create_surface)(struct pipe_context *ctx,
+                                          struct pipe_resource *resource,
+                                          const struct pipe_surface *templat);
+
+   void (*surface_destroy)(struct pipe_context *ctx,
+                           struct pipe_surface *);
+
+   /**
     * Get a transfer object for transferring data to/from a texture.
     *
     * Transfers are (by default) context-private and allow uploads to be
     * interleaved with
     */
    struct pipe_transfer *(*get_transfer)(struct pipe_context *,
-					 struct pipe_resource *resource,
-					 struct pipe_subresource,
-					 unsigned usage,  /* a combination of PIPE_TRANSFER_x */
-					 const struct pipe_box *);
+                                         struct pipe_resource *resource,
+                                         unsigned level,
+                                         unsigned usage,  /* a combination of PIPE_TRANSFER_x */
+                                         const struct pipe_box *);
 
    void (*transfer_destroy)(struct pipe_context *,
-                                struct pipe_transfer *);
-   
+                            struct pipe_transfer *);
+
    void *(*transfer_map)( struct pipe_context *,
                           struct pipe_transfer *transfer );
 
@@ -381,13 +391,13 @@ struct pipe_context {
     * pointer.  XXX: strides??
     */
    void (*transfer_inline_write)( struct pipe_context *,
-				  struct pipe_resource *,
-				  struct pipe_subresource,
-				  unsigned usage, /* a combination of PIPE_TRANSFER_x */
-				  const struct pipe_box *,
-				  const void *data,
-				  unsigned stride,
-				  unsigned slice_stride);
+                                  struct pipe_resource *,
+                                  unsigned level,
+                                  unsigned usage, /* a combination of PIPE_TRANSFER_x */
+                                  const struct pipe_box *,
+                                  const void *data,
+                                  unsigned stride,
+                                  unsigned layer_stride);
 
 };
 
diff --git a/src/gallium/include/pipe/p_defines.h b/src/gallium/include/pipe/p_defines.h
index 6cca301ccc..f5af15ff9f 100644
--- a/src/gallium/include/pipe/p_defines.h
+++ b/src/gallium/include/pipe/p_defines.h
@@ -138,12 +138,14 @@ enum pipe_error {
 /** Texture types.
  * See the documentation for info on PIPE_TEXTURE_RECT vs PIPE_TEXTURE_2D */
 enum pipe_texture_target {
-   PIPE_BUFFER       = 0,
-   PIPE_TEXTURE_1D   = 1,
-   PIPE_TEXTURE_2D   = 2,
-   PIPE_TEXTURE_3D   = 3,
-   PIPE_TEXTURE_CUBE = 4,
-   PIPE_TEXTURE_RECT = 5,
+   PIPE_BUFFER           = 0,
+   PIPE_TEXTURE_1D       = 1,
+   PIPE_TEXTURE_2D       = 2,
+   PIPE_TEXTURE_3D       = 3,
+   PIPE_TEXTURE_CUBE     = 4,
+   PIPE_TEXTURE_RECT     = 5,
+   PIPE_TEXTURE_1D_ARRAY = 6,
+   PIPE_TEXTURE_2D_ARRAY = 7,
    PIPE_MAX_TEXTURE_TYPES
 };
 
@@ -178,14 +180,6 @@ enum pipe_texture_target {
 #define PIPE_TEX_COMPARE_NONE          0
 #define PIPE_TEX_COMPARE_R_TO_TEXTURE  1
 
-
-/**
- * Surface layout -- a hint?  Or some driver-internal poking out into
- * the interface?
- */
-#define PIPE_SURFACE_LAYOUT_LINEAR  0
-
-
 /**
  * Clear buffer bits
  */
@@ -281,9 +275,9 @@ enum pipe_transfer_usage {
  * Resource binding flags -- state tracker must specify in advance all
  * the ways a resource might be used.
  */
-#define PIPE_BIND_DEPTH_STENCIL        (1 << 0) /* get_tex_surface */
-#define PIPE_BIND_RENDER_TARGET        (1 << 1) /* get_tex_surface */
-#define PIPE_BIND_SAMPLER_VIEW         (1 << 2) /* get_sampler_view */
+#define PIPE_BIND_DEPTH_STENCIL        (1 << 0) /* create_surface */
+#define PIPE_BIND_RENDER_TARGET        (1 << 1) /* create_surface */
+#define PIPE_BIND_SAMPLER_VIEW         (1 << 2) /* create_sampler_view */
 #define PIPE_BIND_VERTEX_BUFFER        (1 << 3) /* set_vertex_buffers */
 #define PIPE_BIND_INDEX_BUFFER         (1 << 4) /* draw_elements */
 #define PIPE_BIND_CONSTANT_BUFFER      (1 << 5) /* set_constant_buffer */
@@ -461,6 +455,7 @@ enum pipe_cap {
    /** different blend funcs per rendertarget */
    PIPE_CAP_INDEP_BLEND_FUNC,
    PIPE_CAP_DEPTHSTENCIL_CLEAR_SEPARATE,
+   PIPE_CAP_ARRAY_TEXTURES,
    PIPE_CAP_TGSI_FS_COORD_ORIGIN_UPPER_LEFT,
    PIPE_CAP_TGSI_FS_COORD_ORIGIN_LOWER_LEFT,
    PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_HALF_INTEGER,
@@ -489,6 +484,7 @@ enum pipe_shader_cap
    PIPE_SHADER_CAP_INDIRECT_OUTPUT_ADDR,
    PIPE_SHADER_CAP_INDIRECT_TEMP_ADDR,
    PIPE_SHADER_CAP_INDIRECT_CONST_ADDR,
+   PIPE_SHADER_CAP_SUBROUTINES, /* BGNSUB, ENDSUB, CAL, RET */
 };
 
 /**
diff --git a/src/gallium/include/pipe/p_screen.h b/src/gallium/include/pipe/p_screen.h
index 912631242f..850eb84a3c 100644
--- a/src/gallium/include/pipe/p_screen.h
+++ b/src/gallium/include/pipe/p_screen.h
@@ -117,7 +117,7 @@ struct pipe_screen {
    /**
     * Create a texture from a winsys_handle. The handle is often created in
     * another process by first creating a pipe texture and then calling
-    * texture_get_handle.
+    * resource_get_handle.
     */
    struct pipe_resource * (*resource_from_handle)(struct pipe_screen *,
 						  const struct pipe_resource *templat,
@@ -136,18 +136,6 @@ struct pipe_screen {
    void (*resource_destroy)(struct pipe_screen *,
 			    struct pipe_resource *pt);
 
-   /** Get a 2D surface which is a "view" into a texture
-    * \param usage  bitmaks of PIPE_BIND_* flags
-    */
-   struct pipe_surface *(*get_tex_surface)(struct pipe_screen *,
-                                           struct pipe_resource *resource,
-                                           unsigned face, unsigned level,
-                                           unsigned zslice,
-                                           unsigned usage );
-
-   void (*tex_surface_destroy)(struct pipe_surface *);
-   
-
 
    /**
     * Create a buffer that wraps user-space data.
@@ -182,7 +170,8 @@ struct pipe_screen {
     *                                gets out-of-band
     */
    void (*flush_frontbuffer)( struct pipe_screen *screen,
-                              struct pipe_surface *surf,
+                              struct pipe_resource *resource,
+                              unsigned level, unsigned layer,
                               void *winsys_drawable_handle );
 
 
diff --git a/src/gallium/include/pipe/p_state.h b/src/gallium/include/pipe/p_state.h
index fc6dba346d..226ae8667b 100644
--- a/src/gallium/include/pipe/p_state.h
+++ b/src/gallium/include/pipe/p_state.h
@@ -271,25 +271,33 @@ struct pipe_sampler_state
 
 
 /**
- * 2D surface.  This is basically a view into a memory buffer.
- * May be a renderbuffer, texture mipmap level, etc.
+ * A view into a texture that can be bound to a color render target /
+ * depth stencil attachment point.
  */
 struct pipe_surface
 {
    struct pipe_reference reference;
    struct pipe_resource *texture; /**< resource into which this is a view  */
+   struct pipe_context *context; /**< context this view belongs to */
    enum pipe_format format;
 
+   /* XXX width/height should be removed */
    unsigned width;               /**< logical width in pixels */
    unsigned height;              /**< logical height in pixels */
 
-   unsigned layout;              /**< PIPE_SURFACE_LAYOUT_x */
-   unsigned offset;              /**< offset from start of buffer, in bytes */
    unsigned usage;               /**< bitmask of PIPE_BIND_x */
 
-   unsigned zslice;
-   unsigned face;
-   unsigned level;
+   union {
+      struct {
+         unsigned level;
+         unsigned first_layer:16;
+         unsigned last_layer:16;
+      } tex;
+      struct {
+         unsigned first_element;
+         unsigned last_element;
+      } buf;
+   } u;
 };
 
 
@@ -302,8 +310,18 @@ struct pipe_sampler_view
    enum pipe_format format;      /**< typed PIPE_FORMAT_x */
    struct pipe_resource *texture; /**< texture into which this is a view  */
    struct pipe_context *context; /**< context this view belongs to */
-   unsigned first_level:8;       /**< first mipmap level */
-   unsigned last_level:8;        /**< last mipmap level */
+   union {
+      struct {
+         unsigned first_layer:16;     /**< first layer to use for array textures */
+         unsigned last_layer:16;      /**< last layer to use for array textures */
+         unsigned first_level:8;      /**< first mipmap level to use */
+         unsigned last_level:8;       /**< last mipmap level to use */
+      } tex;
+      struct {
+         unsigned first_element;
+         unsigned last_element;
+      } buf;
+   } u;
    unsigned swizzle_r:3;         /**< PIPE_SWIZZLE_x for red component */
    unsigned swizzle_g:3;         /**< PIPE_SWIZZLE_x for green component */
    unsigned swizzle_b:3;         /**< PIPE_SWIZZLE_x for blue component */
@@ -338,13 +356,14 @@ struct pipe_resource
    unsigned width0;
    unsigned height0;
    unsigned depth0;
+   unsigned array_size;
 
    unsigned last_level:8;    /**< Index of last mipmap level present/defined */
    unsigned nr_samples:8;    /**< for multisampled surfaces, nr of samples */
    unsigned usage:8;         /**< PIPE_USAGE_x (not a bitmask) */
 
-   unsigned bind;	     /**< bitmask of PIPE_BIND_x */
-   unsigned flags;	     /**< bitmask of PIPE_RESOURCE_FLAG_x */
+   unsigned bind;            /**< bitmask of PIPE_BIND_x */
+   unsigned flags;           /**< bitmask of PIPE_RESOURCE_FLAG_x */
 };
 
 struct pipe_stream_output_state
@@ -363,15 +382,6 @@ struct pipe_stream_output_state
    unsigned stride;
 };
 
-/**
- * Extra indexing info for (cube) texture resources.
- */
-struct pipe_subresource
-{
-   unsigned face:16;
-   unsigned level:16;
-};
-
 
 /**
  * Transfer object.  For data transfer to/from a resource.
@@ -379,11 +389,11 @@ struct pipe_subresource
 struct pipe_transfer
 {
    struct pipe_resource *resource; /**< resource to transfer to/from  */
-   struct pipe_subresource sr;
+   unsigned level;
    enum pipe_transfer_usage usage;
    struct pipe_box box;
    unsigned stride;
-   unsigned slice_stride;
+   unsigned layer_stride;
    void *data;
 };
 
diff --git a/src/gallium/include/state_tracker/st_api.h b/src/gallium/include/state_tracker/st_api.h
index 21e2165ed9..1c2148b78f 100644
--- a/src/gallium/include/state_tracker/st_api.h
+++ b/src/gallium/include/state_tracker/st_api.h
@@ -164,9 +164,8 @@ struct st_egl_image
    /* this is owned by the caller */
    struct pipe_resource *texture;
 
-   unsigned face;
    unsigned level;
-   unsigned zslice;
+   unsigned layer;
 };
 
 /**
@@ -349,6 +348,12 @@ struct st_context_iface
                 struct st_context_iface *stsrci, unsigned mask);
 
    /**
+    * Used to implement wglShareLists.
+    */
+   boolean (*share)(struct st_context_iface *stctxi,
+                    struct st_context_iface *stsrci);
+
+   /**
     * Look up and return the info of a resource for EGLImage.
     *
     * This function is optional.
@@ -402,6 +407,11 @@ struct st_manager
 struct st_api
 {
    /**
+    * The name of the rendering API.  This is informative.
+    */
+   const char *name;
+
+   /**
     * The supported rendering API.
     */
    enum st_api_type api;
diff --git a/src/gallium/include/state_tracker/swrast_screen_create.h b/src/gallium/include/state_tracker/swrast_screen_create.h
deleted file mode 100644
index a08f83a325..0000000000
--- a/src/gallium/include/state_tracker/swrast_screen_create.h
+++ /dev/null
@@ -1,67 +0,0 @@
-#include "pipe/p_compiler.h"
-#include "util/u_debug.h"
-#include "target-helpers/wrap_screen.h"
-
-struct sw_winsys;
-
-#ifdef GALLIUM_SOFTPIPE
-#include "softpipe/sp_public.h"
-#endif
-
-#ifdef GALLIUM_LLVMPIPE
-#include "llvmpipe/lp_public.h"
-#endif
-
-#ifdef GALLIUM_CELL
-#include "cell/ppu/cell_public.h"
-#endif
-
-/*
- * Helper function to choose and instantiate one of the software rasterizers:
- * cell, llvmpipe, softpipe.
- *
- * This function could be shared, but currently causes headaches for
- * the build systems, particularly scons if we try.  Long term, want
- * to avoid having global #defines for things like GALLIUM_LLVMPIPE,
- * GALLIUM_CELL, etc.  Scons already eliminates those #defines, so
- * things that are painful for it now are likely to be painful for
- * other build systems in the future.
- */
-
-static INLINE struct pipe_screen *
-swrast_screen_create(struct sw_winsys *winsys)
-{
-   const char *default_driver;
-   const char *driver;
-   struct pipe_screen *screen = NULL;
-
-#if defined(GALLIUM_CELL)
-   default_driver = "cell";
-#elif defined(GALLIUM_LLVMPIPE)
-   default_driver = "llvmpipe";
-#elif defined(GALLIUM_SOFTPIPE)
-   default_driver = "softpipe";
-#else
-   default_driver = "";
-#endif
-
-   driver = debug_get_option("GALLIUM_DRIVER", default_driver);
-
-#if defined(GALLIUM_CELL)
-   if (screen == NULL && strcmp(driver, "cell") == 0)
-      screen = cell_create_screen( winsys );
-#endif
-
-#if defined(GALLIUM_LLVMPIPE)
-   if (screen == NULL && strcmp(driver, "llvmpipe") == 0)
-      screen = llvmpipe_create_screen( winsys );
-#endif
-
-#if defined(GALLIUM_SOFTPIPE)
-   if (screen == NULL)
-      screen = softpipe_create_screen( winsys );
-#endif
-
-   return gallium_wrap_screen( screen );
-}
-
diff --git a/src/gallium/include/state_tracker/xlib_sw_winsys.h b/src/gallium/include/state_tracker/xlib_sw_winsys.h
index f22c22bb62..4e7ccc13ac 100644
--- a/src/gallium/include/state_tracker/xlib_sw_winsys.h
+++ b/src/gallium/include/state_tracker/xlib_sw_winsys.h
@@ -6,7 +6,7 @@
 
 
 struct pipe_screen;
-struct pipe_surface;
+struct pipe_resource;
 
 /* This is what the xlib software winsys expects to find in the
  * "private" field of flush_frontbuffers().
diff --git a/src/gallium/state_trackers/d3d1x/dxgi/src/dxgi_native.cpp b/src/gallium/state_trackers/d3d1x/dxgi/src/dxgi_native.cpp
index c246fc5ef7..a54324a04f 100644
--- a/src/gallium/state_trackers/d3d1x/dxgi/src/dxgi_native.cpp
+++ b/src/gallium/state_trackers/d3d1x/dxgi/src/dxgi_native.cpp
@@ -1159,8 +1159,15 @@ struct GalliumDXGISwapChain : public GalliumDXGIObject<IDXGISwapChain, GalliumDX
 			unsigned blit_x, blit_y, blit_w, blit_h;
 			float black[4] = {0, 0, 0, 0};
 
-			if(!formats_compatible || src->width0 != dst_w || src->height0 != dst_h)
-				dst_surface = pipe->screen->get_tex_surface(pipe->screen, dst, 0, 0, 0, PIPE_BIND_RENDER_TARGET);
+			if(!formats_compatible || src->width0 != dst_w || src->height0 != dst_h) {
+				struct pipe_surface templat;
+				templat.usage = PIPE_BIND_RENDER_TARGET;
+				templat.format = dst->format;
+				templat.u.tex.level = 0;
+				templat.u.tex.first_layer = 0;
+				templat.u.tex.last_layer = 0;
+				dst_surface = pipe->create_surface(pipe, dst, &templat);
+			}
 
 			if(preserve_aspect_ratio)
 			{
@@ -1199,10 +1206,12 @@ struct GalliumDXGISwapChain : public GalliumDXGIObject<IDXGISwapChain, GalliumDX
 
 			if(formats_compatible && blit_w == src->width0 && blit_h == src->height0)
 			{
-				pipe_subresource sr;
-				sr.face = 0;
-				sr.level = 0;
-				pipe->resource_copy_region(pipe, dst, sr, rect.left, rect.top, 0, src, sr, 0, 0, 0, blit_w, blit_h);
+				pipe_box box;
+				box.x = box.y = box.z;
+				box.width = blit_w;
+				box.height = blit_h;
+				box.z = 1;
+				pipe->resource_copy_region(pipe, dst, 0, rect.left, rect.top, 0, src, 0, &box);
 			}
 			else
 			{
@@ -1218,7 +1227,7 @@ struct GalliumDXGISwapChain : public GalliumDXGIObject<IDXGISwapChain, GalliumDX
 		}
 
 		if(dst_surface)
-			pipe->screen->tex_surface_destroy(dst_surface);
+			pipe->surface_destroy(pipe, dst_surface);
 
 		pipe->flush(pipe, PIPE_FLUSH_RENDER_CACHE | PIPE_FLUSH_FRAME, 0);
 
diff --git a/src/gallium/state_trackers/d3d1x/gd3d11/d3d11_context.h b/src/gallium/state_trackers/d3d1x/gd3d11/d3d11_context.h
index 36110595c2..e1ba6c184f 100644
--- a/src/gallium/state_trackers/d3d1x/gd3d11/d3d11_context.h
+++ b/src/gallium/state_trackers/d3d1x/gd3d11/d3d11_context.h
@@ -1416,23 +1416,33 @@ changed:
 			*out_predicate_value = render_predicate_value;
 	}
 
-	static pipe_subresource d3d11_to_pipe_subresource(struct pipe_resource* resource, unsigned subresource)
+	static unsigned d3d11_subresource_to_level(struct pipe_resource* resource, unsigned subresource)
 	{
-		pipe_subresource sr;
 		if(subresource <= resource->last_level)
 		{
-			sr.level = subresource;
-			sr.face = 0;
+			return subresource;
 		}
 		else
 		{
 			unsigned levels = resource->last_level + 1;
-			sr.level = subresource % levels;
-			sr.face = subresource / levels;
+			return subresource % levels;
 		}
-		return sr;
 	}
 
+	static unsigned d3d11_subresource_to_face(struct pipe_resource* resource, unsigned subresource)
+	{
+		if(subresource <= resource->last_level)
+		{
+			return 0;
+		}
+		else
+		{
+			unsigned levels = resource->last_level + 1;
+			return subresource / levels;
+		}
+	}
+		
+	
 	/* TODO: deferred contexts will need a different implementation of this,
 	 * because we can't put the transfer info into the resource itself.
 	 * Also, there are very different restrictions, for obvious reasons.
@@ -1448,8 +1458,10 @@ changed:
 		GalliumD3D11Resource<>* resource = (GalliumD3D11Resource<>*)iresource;
 		if(resource->transfers.count(subresource))
 			return E_FAIL;
-		pipe_subresource sr = d3d11_to_pipe_subresource(resource->resource, subresource);
-		pipe_box box = d3d11_to_pipe_box(resource->resource, sr.level, 0);
+		unsigned level = d3d11_subresource_to_level(resource->resource, subresource);
+		unsigned face = d3d11_subresource_to_face(resource->resource, subresource);
+		pipe_box box = d3d11_to_pipe_box(resource->resource, level, 0);
+		/* XXX the translation from subresource to level/face(zslice/array layer) isn't quite right */
 		unsigned usage = 0;
 		if(map_type == D3D11_MAP_READ)
 			usage = PIPE_TRANSFER_READ;
@@ -1465,7 +1477,7 @@ changed:
 			return E_INVALIDARG;
 		if(map_type & D3D10_MAP_FLAG_DO_NOT_WAIT)
 			usage |= PIPE_TRANSFER_DONTBLOCK;
-		struct pipe_transfer* transfer = pipe->get_transfer(pipe, resource->resource, sr, usage, &box);
+		struct pipe_transfer* transfer = pipe->get_transfer(pipe, resource->resource, level, usage, &box);
 		if(!transfer) {
 			if(map_type & D3D10_MAP_FLAG_DO_NOT_WAIT)
 				return DXGI_ERROR_WAS_STILL_DRAWING;
@@ -1475,7 +1487,7 @@ changed:
 		resource->transfers[subresource] = transfer;
 		mapped_resource->pData = pipe->transfer_map(pipe, transfer);
 		mapped_resource->RowPitch = transfer->stride;
-		mapped_resource->DepthPitch = transfer->slice_stride;
+		mapped_resource->DepthPitch = transfer->layer_stride;
 		return S_OK;
 	}
 
@@ -1507,15 +1519,16 @@ changed:
 		SYNCHRONIZED;
 		GalliumD3D11Resource<>* dst = (GalliumD3D11Resource<>*)dst_resource;
 		GalliumD3D11Resource<>* src = (GalliumD3D11Resource<>*)src_resource;
-		pipe_subresource subdst = d3d11_to_pipe_subresource(dst->resource, dst_subresource);
-		pipe_subresource subsrc = d3d11_to_pipe_subresource(src->resource, src_subresource);
-		pipe_box box = d3d11_to_pipe_box(src->resource, subsrc.level, src_box);
-		for(unsigned i = 0; i < box.depth; ++i)
+		unsigned dst_level = d3d11_subresource_to_level(dst->resource, dst_subresource);
+		unsigned dst_face = d3d11_subresource_to_face(dst->resource, dst_subresource);
+		unsigned src_level = d3d11_subresource_to_level(src->resource, src_subresource);
+		unsigned src_face = d3d11_subresource_to_face(src->resource, src_subresource);
+		/* XXX the translation from subresource to level/face(zslice/array layer) isn't quite right */
+		pipe_box box = d3d11_to_pipe_box(src->resource, src_level, src_box);
 		{
 			pipe->resource_copy_region(pipe,
-				dst->resource, subdst, dst_x, dst_y, dst_z + i,
-				src->resource, subsrc, box.x, box.y, box.z + i,
-				box.width, box.height);
+				dst->resource, dst_level, dst_x, dst_y, dst_z,
+				src->resource, src_level, &box);
 		}
 	}
 
@@ -1526,24 +1539,23 @@ changed:
 		SYNCHRONIZED;
 		GalliumD3D11Resource<>* dst = (GalliumD3D11Resource<>*)dst_resource;
 		GalliumD3D11Resource<>* src = (GalliumD3D11Resource<>*)src_resource;
-		pipe_subresource sr;
-		unsigned faces = dst->resource->target == PIPE_TEXTURE_CUBE ? 6 : 1;
-
-		for(sr.face = 0; sr.face < faces; ++sr.face)
+		unsigned level;
+		for(level = 0; level <= dst->resource->last_level; ++level)
 		{
-			for(sr.level = 0; sr.level <= dst->resource->last_level; ++sr.level)
-			{
-				unsigned w = u_minify(dst->resource->width0, sr.level);
-				unsigned h = u_minify(dst->resource->height0, sr.level);
-				unsigned d = u_minify(dst->resource->depth0, sr.level);
-				for(unsigned i = 0; i < d; ++i)
-				{
-					pipe->resource_copy_region(pipe,
-							dst->resource, sr, 0, 0, i,
-							src->resource, sr, 0, 0, i,
-							w, h);
-				}
-			}
+		        unsigned layers = 1;
+			pipe_box box;
+			if (dst->resource->target == PIPE_TEXTURE_CUBE)
+				layers = 6;
+			else if (dst->resource->target == PIPE_TEXTURE_3D)
+				layers = u_minify(dst->resource->depth0, level);
+			/* else layers = dst->resource->array_size; */
+			box.x = box.y = box.z = 0;
+			box.width = u_minify(dst->resource->width0, level);
+			box.height = u_minify(dst->resource->height0, level);
+			box.depth = layers;
+			pipe->resource_copy_region(pipe,
+						   dst->resource, level, 0, 0, 0,
+						   src->resource, level, &box);
 		}
 	}
 
@@ -1557,9 +1569,10 @@ changed:
 	{
 		SYNCHRONIZED;
 		GalliumD3D11Resource<>* dst = (GalliumD3D11Resource<>*)dst_resource;
-		pipe_subresource subdst = d3d11_to_pipe_subresource(dst->resource, dst_subresource);
-		pipe_box box = d3d11_to_pipe_box(dst->resource, subdst.level, pDstBox);
-		pipe->transfer_inline_write(pipe, dst->resource, subdst, PIPE_TRANSFER_WRITE, &box, pSrcData, src_row_pitch, src_depth_pitch);
+		unsigned dst_level = d3d11_subresource_to_level(dst->resource, dst_subresource);
+		/* XXX the translation from subresource to level/face(zslice/array layer) isn't quite right */
+		pipe_box box = d3d11_to_pipe_box(dst->resource, dst_level, pDstBox);
+		pipe->transfer_inline_write(pipe, dst->resource, dst_level, PIPE_TRANSFER_WRITE, &box, pSrcData, src_row_pitch, src_depth_pitch);
 	}
 
 #if API >= 11
@@ -1714,9 +1727,9 @@ changed:
 		SYNCHRONIZED;
 		GalliumD3D11Resource<>* dst = (GalliumD3D11Resource<>*)dst_resource;
 		GalliumD3D11Resource<>* src = (GalliumD3D11Resource<>*)src_resource;
-		pipe_subresource subdst = d3d11_to_pipe_subresource(dst->resource, dst_subresource);
-		pipe_subresource subsrc = d3d11_to_pipe_subresource(src->resource, src_subresource);
-		pipe->resource_resolve(pipe, dst->resource, subdst, src->resource, subsrc);
+		unsigned dst_layer = d3d11_subresource_to_face(dst->resource, dst_subresource);
+		unsigned src_layer = d3d11_subresource_to_face(src->resource, src_subresource);
+		pipe->resource_resolve(pipe, dst->resource, dst_layer, src->resource, src_layer);
 	}
 
 #if API >= 11
diff --git a/src/gallium/state_trackers/d3d1x/gd3d11/d3d11_screen.h b/src/gallium/state_trackers/d3d1x/gd3d11/d3d11_screen.h
index 95ea4e00fc..9cfdc837d8 100644
--- a/src/gallium/state_trackers/d3d1x/gd3d11/d3d11_screen.h
+++ b/src/gallium/state_trackers/d3d1x/gd3d11/d3d11_screen.h
@@ -718,15 +718,13 @@ struct GalliumD3D11ScreenImpl : public GalliumD3D11Screen
 			{
 				for(unsigned level = 0; level <= templat.last_level; ++level)
 				{
-					struct pipe_subresource sr;
-					sr.level = level;
-					sr.face = slice;
 					struct pipe_box box;
-					box.x = box.y = box.z = 0;
+					box.x = box.y = 0;
+					box.z = slice;
 					box.width = u_minify(width, level);
 					box.height = u_minify(height, level);
-					box.depth = u_minify(depth, level);
-					immediate_pipe->transfer_inline_write(immediate_pipe, resource, sr, PIPE_TRANSFER_WRITE | PIPE_TRANSFER_DISCARD | PIPE_TRANSFER_UNSYNCHRONIZED, &box, initial_data->pSysMem, initial_data->SysMemPitch, initial_data->SysMemSlicePitch);
+					box.depth = 1;
+					immediate_pipe->transfer_inline_write(immediate_pipe, resource, level, PIPE_TRANSFER_WRITE | PIPE_TRANSFER_DISCARD | PIPE_TRANSFER_UNSYNCHRONIZED, &box, initial_data->pSysMem, initial_data->SysMemPitch, initial_data->SysMemSlicePitch);
 					++initial_data;
 				}
 			}
@@ -978,8 +976,8 @@ struct GalliumD3D11ScreenImpl : public GalliumD3D11Screen
 		case D3D11_SRV_DIMENSION_TEXTURE1DARRAY:
 		case D3D11_SRV_DIMENSION_TEXTURE2DARRAY:
 			/* yes, this works for all of these types (but TODO: texture arrays) */
-			templat.first_level = desc->Texture1D.MostDetailedMip;
-			templat.last_level = templat.first_level + desc->Texture1D.MipLevels - 1;
+			templat.u.tex.first_level = desc->Texture1D.MostDetailedMip;
+			templat.u.tex.last_level = templat.u.tex.first_level + desc->Texture1D.MipLevels - 1;
 			break;
 		case D3D11_SRV_DIMENSION_BUFFER:
 		case D3D11_SRV_DIMENSION_TEXTURE2DMS:
@@ -1054,30 +1052,34 @@ struct GalliumD3D11ScreenImpl : public GalliumD3D11Screen
 			desc = &def_desc;
 		}
 
-		unsigned zslice = 0;
-		unsigned face = 0;
-		unsigned level;
-		enum pipe_format format;
+		struct pipe_surface templat;
+		memset(&templat, 0, sizeof(templat));
 		if(invalid(desc->format >= DXGI_FORMAT_COUNT))
 			return E_INVALIDARG;
-		format = dxgi_to_pipe_format[desc->Format];
-		if(!format)
+		templat.format = dxgi_to_pipe_format[desc->Format];
+		if(!templat.format)
 			return E_NOTIMPL;
+		templat.usage = PIPE_BIND_RENDER_TARGET;
+		templat.texture = ((GalliumD3D11Resource<>*)iresource)->resource;
 
 		switch(desc->ViewDimension)
 		{
 		case D3D11_RTV_DIMENSION_TEXTURE1D:
 		case D3D11_RTV_DIMENSION_TEXTURE2D:
-			level = desc->Texture1D.MipSlice;
+			templat.u.tex.level = desc->Texture1D.MipSlice;
 			break;
 		case D3D11_RTV_DIMENSION_TEXTURE3D:
-			level = desc->Texture3D.MipSlice;
-			zslice = desc->Texture3D.FirstWSlice;
+			templat.u.tex.level = desc->Texture3D.MipSlice;
+			templat.u.tex.first_layer = desc->Texture3D.FirstWSlice;
+			/* XXX FIXME */
+			templat.u.tex.last_layer = desc->Texture3D.FirstWSlice;
 			break;
 		case D3D11_RTV_DIMENSION_TEXTURE1DARRAY:
 		case D3D11_RTV_DIMENSION_TEXTURE2DARRAY:
-			level = desc->Texture1DArray.MipSlice;
-			face = desc->Texture1DArray.FirstArraySlice;
+			templat.u.tex.level = desc->Texture1DArray.MipSlice;
+			templat.u.tex.first_layer = desc->Texture1DArray.FirstArraySlice;
+			/* XXX FIXME */
+			templat.u.tex.last_layer = desc->Texture1DArray.FirstArraySlice;
 			break;
 		case D3D11_RTV_DIMENSION_BUFFER:
 		case D3D11_RTV_DIMENSION_TEXTURE2DMS:
@@ -1090,13 +1092,9 @@ struct GalliumD3D11ScreenImpl : public GalliumD3D11Screen
 		if(!out_rtv)
 			return S_FALSE;
 
-		struct pipe_surface* surface = screen->get_tex_surface(screen,
-				((GalliumD3D11Resource<>*)iresource)->resource,
-				face, level, zslice, PIPE_BIND_RENDER_TARGET);
+		struct pipe_surface* surface = immediate_pipe->create_surface(immediate_pipe, templat.texture, &templat);
 		if(!surface)
 			return E_FAIL;
-		/* muhahahahaha, let's hope this actually works */
-		surface->format = format;
 		*out_rtv = new GalliumD3D11RenderTargetView(this, (GalliumD3D11Resource<>*)iresource, surface, *desc);
 		return S_OK;
 	}
@@ -1134,26 +1132,28 @@ struct GalliumD3D11ScreenImpl : public GalliumD3D11Screen
 			desc = &def_desc;
 		}
 
-		unsigned zslice = 0;
-		unsigned face = 0;
-		unsigned level;
-		enum pipe_format format;
+		struct pipe_surface templat;
+		memset(&templat, 0, sizeof(templat));
 		if(invalid(desc->format >= DXGI_FORMAT_COUNT))
 			return E_INVALIDARG;
-		format = dxgi_to_pipe_format[desc->Format];
-		if(!format)
+		templat.format = dxgi_to_pipe_format[desc->Format];
+		if(!templat.format)
 			return E_NOTIMPL;
+		templat.usage = PIPE_BIND_DEPTH_STENCIL;
+		templat.texture = ((GalliumD3D11Resource<>*)iresource)->resource;
 
 		switch(desc->ViewDimension)
 		{
 		case D3D11_DSV_DIMENSION_TEXTURE1D:
 		case D3D11_DSV_DIMENSION_TEXTURE2D:
-			level = desc->Texture1D.MipSlice;
+			templat.u.tex.level = desc->Texture1D.MipSlice;
 			break;
 		case D3D11_DSV_DIMENSION_TEXTURE1DARRAY:
 		case D3D11_DSV_DIMENSION_TEXTURE2DARRAY:
-			level = desc->Texture1DArray.MipSlice;
-			face = desc->Texture1DArray.FirstArraySlice;
+			templat.u.tex.level = desc->Texture1DArray.MipSlice;
+			templat.u.tex.first_layer = desc->Texture1DArray.FirstArraySlice;
+			/* XXX FIXME */
+			templat.u.tex.last_layer = desc->Texture1DArray.FirstArraySlice;
 			break;
 		case D3D11_DSV_DIMENSION_TEXTURE2DMS:
 		case D3D11_DSV_DIMENSION_TEXTURE2DMSARRAY:
@@ -1165,13 +1165,9 @@ struct GalliumD3D11ScreenImpl : public GalliumD3D11Screen
 		if(!out_depth_stencil_view)
 			return S_FALSE;
 
-		struct pipe_surface* surface = screen->get_tex_surface(screen,
-				((GalliumD3D11Resource<>*)iresource)->resource,
-				face, level, zslice, PIPE_BIND_DEPTH_STENCIL);
+		struct pipe_surface* surface = immediate_pipe->create_surface(immediate_pipe, templat.texture, &templat);
 		if(!surface)
 			return E_FAIL;
-		/* muhahahahaha, let's hope this actually works */
-		surface->format = format;
 		*out_depth_stencil_view = new GalliumD3D11DepthStencilView(this, (GalliumD3D11Resource<>*)iresource, surface, *desc);
 		return S_OK;
 	}
diff --git a/src/gallium/state_trackers/dri/common/dri_screen.c b/src/gallium/state_trackers/dri/common/dri_screen.c
index 1302e9bc01..f6e22c74b4 100644
--- a/src/gallium/state_trackers/dri/common/dri_screen.c
+++ b/src/gallium/state_trackers/dri/common/dri_screen.c
@@ -304,9 +304,8 @@ dri_get_egl_image(struct st_manager *smapi,
 
    stimg->texture = NULL;
    pipe_resource_reference(&stimg->texture, img->texture);
-   stimg->face = img->face;
    stimg->level = img->level;
-   stimg->zslice = img->zslice;
+   stimg->layer = img->layer;
 
    return TRUE;
 }
diff --git a/src/gallium/state_trackers/dri/common/dri_screen.h b/src/gallium/state_trackers/dri/common/dri_screen.h
index 0da9b5510f..8cb0a102c8 100644
--- a/src/gallium/state_trackers/dri/common/dri_screen.h
+++ b/src/gallium/state_trackers/dri/common/dri_screen.h
@@ -83,9 +83,8 @@ dri_screen(__DRIscreen * sPriv)
 
 struct __DRIimageRec {
    struct pipe_resource *texture;
-   unsigned face;
    unsigned level;
-   unsigned zslice;
+   unsigned layer;
 
    void *loader_private;
 };
diff --git a/src/gallium/state_trackers/dri/drm/dri2.c b/src/gallium/state_trackers/dri/drm/dri2.c
index 3c5b075617..a9d05a80fb 100644
--- a/src/gallium/state_trackers/dri/drm/dri2.c
+++ b/src/gallium/state_trackers/dri/drm/dri2.c
@@ -200,6 +200,7 @@ dri2_drawable_process_buffers(struct dri_drawable *drawable,
    templ.width0 = dri_drawable->w;
    templ.height0 = dri_drawable->h;
    templ.depth0 = 1;
+   templ.array_size = 1;
 
    memset(&whandle, 0, sizeof(whandle));
 
@@ -348,6 +349,7 @@ dri2_create_image_from_name(__DRIscreen *_screen,
    templ.width0 = width;
    templ.height0 = height;
    templ.depth0 = 1;
+   templ.array_size = 1;
 
    memset(&whandle, 0, sizeof(whandle));
    whandle.handle = name;
@@ -360,9 +362,8 @@ dri2_create_image_from_name(__DRIscreen *_screen,
       return NULL;
    }
 
-   img->face = 0;
    img->level = 0;
-   img->zslice = 0;
+   img->layer = 0;
    img->loader_private = loaderPrivate;
 
    return img;
@@ -430,9 +431,8 @@ dri2_create_image(__DRIscreen *_screen,
       return NULL;
    }
 
-   img->face = 0;
    img->level = 0;
-   img->zslice = 0;
+   img->layer = 0;
 
    img->loader_private = loaderPrivate;
    return img;
diff --git a/src/gallium/state_trackers/dri/sw/drisw.c b/src/gallium/state_trackers/dri/sw/drisw.c
index c48cc44036..30088b0968 100644
--- a/src/gallium/state_trackers/dri/sw/drisw.c
+++ b/src/gallium/state_trackers/dri/sw/drisw.c
@@ -87,40 +87,17 @@ drisw_put_image(struct dri_drawable *drawable,
    put_image(dPriv, data, width, height);
 }
 
-static struct pipe_surface *
-drisw_get_pipe_surface(struct dri_drawable *drawable, struct pipe_resource *ptex)
-{
-   struct pipe_screen *pipe_screen = dri_screen(drawable->sPriv)->base.screen;
-   struct pipe_surface *psurf = drawable->drisw_surface;
-
-   if (!psurf || psurf->texture != ptex) {
-      pipe_surface_reference(&drawable->drisw_surface, NULL);
-
-      drawable->drisw_surface = pipe_screen->get_tex_surface(pipe_screen,
-            ptex, 0, 0, 0, 0/* no bind flag???*/);
-
-      psurf = drawable->drisw_surface;
-   }
-
-   return psurf;
-}
-
 static INLINE void
 drisw_present_texture(__DRIdrawable *dPriv,
                       struct pipe_resource *ptex)
 {
    struct dri_drawable *drawable = dri_drawable(dPriv);
    struct dri_screen *screen = dri_screen(drawable->sPriv);
-   struct pipe_surface *psurf;
 
    if (swrast_no_present)
       return;
 
-   psurf = drisw_get_pipe_surface(drawable, ptex);
-   if (!psurf)
-      return;
-
-   screen->base.screen->flush_frontbuffer(screen->base.screen, psurf, drawable);
+   screen->base.screen->flush_frontbuffer(screen->base.screen, ptex, 0, 0, drawable);
 }
 
 static INLINE void
@@ -220,6 +197,7 @@ drisw_allocate_textures(struct dri_drawable *drawable,
    templ.width0 = width;
    templ.height0 = height;
    templ.depth0 = 1;
+   templ.array_size = 1;
    templ.last_level = 0;
 
    for (i = 0; i < count; i++) {
diff --git a/src/gallium/state_trackers/egl/common/egl_g3d.h b/src/gallium/state_trackers/egl/common/egl_g3d.h
index 72c14f0ac4..9873fee6ec 100644
--- a/src/gallium/state_trackers/egl/common/egl_g3d.h
+++ b/src/gallium/state_trackers/egl/common/egl_g3d.h
@@ -92,9 +92,8 @@ struct egl_g3d_config {
 struct egl_g3d_image {
    _EGLImage base;
    struct pipe_resource *texture;
-   unsigned face;
    unsigned level;
-   unsigned zslice;
+   unsigned layer;
 };
 
 /* standard typecasts */
diff --git a/src/gallium/state_trackers/egl/common/egl_g3d_api.c b/src/gallium/state_trackers/egl/common/egl_g3d_api.c
index fd7dc8f814..8e53e1dccb 100644
--- a/src/gallium/state_trackers/egl/common/egl_g3d_api.c
+++ b/src/gallium/state_trackers/egl/common/egl_g3d_api.c
@@ -30,6 +30,7 @@
 #include "pipe/p_screen.h"
 #include "util/u_memory.h"
 #include "util/u_inlines.h"
+#include "util/u_box.h"
 
 #include "egl_g3d.h"
 #include "egl_g3d_api.h"
@@ -140,11 +141,22 @@ egl_g3d_choose_config(_EGLDriver *drv, _EGLDisplay *dpy, const EGLint *attribs,
    if (!_eglParseConfigAttribList(&criteria, dpy, attribs))
       return _eglError(EGL_BAD_ATTRIBUTE, "eglChooseConfig");
 
-   tmp_configs = (_EGLConfig **) _eglFilterArray(dpy->Configs, &tmp_size,
+   /* get the number of matched configs */
+   tmp_size = _eglFilterArray(dpy->Configs, NULL, 0,
          (_EGLArrayForEach) egl_g3d_match_config, (void *) &criteria);
+   if (!tmp_size) {
+      *num_configs = tmp_size;
+      return EGL_TRUE;
+   }
+
+   tmp_configs = MALLOC(sizeof(tmp_configs[0]) * tmp_size);
    if (!tmp_configs)
       return _eglError(EGL_BAD_ALLOC, "eglChooseConfig(out of memory)");
 
+   /* get the matched configs */
+   _eglFilterArray(dpy->Configs, (void **) tmp_configs, tmp_size,
+         (_EGLArrayForEach) egl_g3d_match_config, (void *) &criteria);
+
    /* perform sorting of configs */
    if (tmp_configs && tmp_size) {
       _eglSortConfigs((const _EGLConfig **) tmp_configs, tmp_size,
@@ -154,7 +166,7 @@ egl_g3d_choose_config(_EGLDriver *drv, _EGLDisplay *dpy, const EGLint *attribs,
          configs[i] = _eglGetConfigHandle(tmp_configs[i]);
    }
 
-   free(tmp_configs);
+   FREE(tmp_configs);
 
    *num_configs = size;
 
@@ -665,15 +677,11 @@ egl_g3d_copy_buffers(_EGLDriver *drv, _EGLDisplay *dpy, _EGLSurface *surf,
    ptex = get_pipe_resource(gdpy->native, nsurf, NATIVE_ATTACHMENT_FRONT_LEFT);
    if (ptex) {
       struct pipe_resource *psrc = gsurf->render_texture;
-      struct pipe_subresource subsrc, subdst;
-      subsrc.face = 0;
-      subsrc.level = 0;
-      subdst.face = 0;
-      subdst.level = 0;
-
+      struct pipe_box src_box;
+      u_box_origin_2d(ptex->width0, ptex->height0, &src_box);
       if (psrc) {
-         gdpy->pipe->resource_copy_region(gdpy->pipe, ptex, subdst, 0, 0, 0,
-               gsurf->render_texture, subsrc, 0, 0, 0, ptex->width0, ptex->height0);
+         gdpy->pipe->resource_copy_region(gdpy->pipe, ptex, 0, 0, 0, 0,
+               gsurf->render_texture, 0, &src_box);
          nsurf->present(nsurf, NATIVE_ATTACHMENT_FRONT_LEFT, FALSE, 0);
       }
 
diff --git a/src/gallium/state_trackers/egl/common/egl_g3d_image.c b/src/gallium/state_trackers/egl/common/egl_g3d_image.c
index 6a1f8cc697..b2d6b433c5 100644
--- a/src/gallium/state_trackers/egl/common/egl_g3d_image.c
+++ b/src/gallium/state_trackers/egl/common/egl_g3d_image.c
@@ -173,6 +173,7 @@ egl_g3d_reference_drm_buffer(_EGLDisplay *dpy, EGLint name,
    templ.width0 = attrs.Width;
    templ.height0 = attrs.Height;
    templ.depth0 = 1;
+   templ.array_size = 1;
 
    memset(&wsh, 0, sizeof(wsh));
    wsh.handle = (unsigned) name;
@@ -191,7 +192,7 @@ egl_g3d_create_image(_EGLDriver *drv, _EGLDisplay *dpy, _EGLContext *ctx,
 {
    struct pipe_resource *ptex;
    struct egl_g3d_image *gimg;
-   unsigned face = 0, level = 0, zslice = 0;
+   unsigned level = 0, layer = 0;
 
    gimg = CALLOC_STRUCT(egl_g3d_image);
    if (!gimg) {
@@ -231,7 +232,7 @@ egl_g3d_create_image(_EGLDriver *drv, _EGLDisplay *dpy, _EGLContext *ctx,
       FREE(gimg);
       return NULL;
    }
-   if (zslice > ptex->depth0) {
+   if (layer >= (u_minify(ptex->depth0, level) + ptex->array_size - 1)) {
       _eglError(EGL_BAD_PARAMETER, "eglCreateEGLImageKHR");
       pipe_resource_reference(&gimg->texture, NULL);
       FREE(gimg);
@@ -240,9 +241,8 @@ egl_g3d_create_image(_EGLDriver *drv, _EGLDisplay *dpy, _EGLContext *ctx,
 
    /* transfer the ownership to the image */
    gimg->texture = ptex;
-   gimg->face = face;
    gimg->level = level;
-   gimg->zslice = zslice;
+   gimg->layer = layer;
 
    return &gimg->base;
 }
@@ -288,9 +288,8 @@ egl_g3d_create_drm_image(_EGLDriver *drv, _EGLDisplay *dpy,
 
    /* transfer the ownership to the image */
    gimg->texture = ptex;
-   gimg->face = 0;
    gimg->level = 0;
-   gimg->zslice = 0;
+   gimg->layer = 0;
 
    return &gimg->base;
 }
diff --git a/src/gallium/state_trackers/egl/common/egl_g3d_st.c b/src/gallium/state_trackers/egl/common/egl_g3d_st.c
index 25e2999590..2b944b521a 100644
--- a/src/gallium/state_trackers/egl/common/egl_g3d_st.c
+++ b/src/gallium/state_trackers/egl/common/egl_g3d_st.c
@@ -72,9 +72,8 @@ egl_g3d_st_manager_get_egl_image(struct st_manager *smapi,
 
    out->texture = NULL;
    pipe_resource_reference(&out->texture, gimg->texture);
-   out->face = gimg->face;
    out->level = gimg->level;
-   out->zslice = gimg->zslice;
+   out->layer = gimg->layer;
 
    _eglUnlockMutex(&gsmapi->display->Mutex);
 
@@ -140,6 +139,7 @@ pbuffer_allocate_render_texture(struct egl_g3d_surface *gsurf)
    templ.width0 = gsurf->base.Width;
    templ.height0 = gsurf->base.Height;
    templ.depth0 = 1;
+   templ.array_size = 1;
    templ.format = gsurf->stvis.color_format;
    templ.bind = PIPE_BIND_RENDER_TARGET;
 
diff --git a/src/gallium/state_trackers/egl/common/native_helper.c b/src/gallium/state_trackers/egl/common/native_helper.c
index 7832b2b693..0f2d02032b 100644
--- a/src/gallium/state_trackers/egl/common/native_helper.c
+++ b/src/gallium/state_trackers/egl/common/native_helper.c
@@ -40,7 +40,6 @@ struct resource_surface {
    uint bind;
 
    struct pipe_resource *resources[NUM_NATIVE_ATTACHMENTS];
-   struct pipe_surface *present_surfaces[NUM_NATIVE_ATTACHMENTS];
    uint resource_mask;
    uint width, height;
 };
@@ -67,8 +66,6 @@ resource_surface_free_resources(struct resource_surface *rsurf)
       int i;
 
       for (i = 0; i < NUM_NATIVE_ATTACHMENTS; i++) {
-         if (rsurf->present_surfaces[i])
-            pipe_surface_reference(&rsurf->present_surfaces[i], NULL);
          if (rsurf->resources[i])
             pipe_resource_reference(&rsurf->resources[i], NULL);
       }
@@ -130,6 +127,7 @@ resource_surface_add_resources(struct resource_surface *rsurf,
    templ.width0 = rsurf->width;
    templ.height0 = rsurf->height;
    templ.depth0 = 1;
+   templ.array_size = 1;
 
    for (i = 0; i < NUM_NATIVE_ATTACHMENTS; i++) {
       if (resource_mask & (1 <<i)) {
@@ -193,8 +191,6 @@ resource_surface_swap_buffers(struct resource_surface *rsurf,
 
    pointer_swap((const void **) &rsurf->resources[buf1],
                 (const void **) &rsurf->resources[buf2]);
-   pointer_swap((const void **) &rsurf->present_surfaces[buf1],
-                (const void **) &rsurf->present_surfaces[buf2]);
 
    /* swap mask bits */
    mask = rsurf->resource_mask & ~(buf1_bit | buf2_bit);
@@ -212,24 +208,12 @@ resource_surface_present(struct resource_surface *rsurf,
                          void *winsys_drawable_handle)
 {
    struct pipe_resource *pres = rsurf->resources[which];
-   struct pipe_surface *psurf = rsurf->present_surfaces[which];
 
    if (!pres)
       return TRUE;
 
-   if (!psurf) {
-      psurf = rsurf->screen->get_tex_surface(rsurf->screen,
-            pres, 0, 0, 0, PIPE_BIND_DISPLAY_TARGET);
-      if (!psurf)
-         return FALSE;
-
-      rsurf->present_surfaces[which] = psurf;
-   }
-
-   assert(psurf->texture == pres);
-
    rsurf->screen->flush_frontbuffer(rsurf->screen,
-         psurf, winsys_drawable_handle);
+         pres, 0, 0, winsys_drawable_handle);
 
    return TRUE;
 }
diff --git a/src/gallium/state_trackers/egl/drm/native_drm.c b/src/gallium/state_trackers/egl/drm/native_drm.c
index 3759c2a26d..2441b43fd8 100644
--- a/src/gallium/state_trackers/egl/drm/native_drm.c
+++ b/src/gallium/state_trackers/egl/drm/native_drm.c
@@ -34,7 +34,7 @@
 
 /* see get_drm_screen_name */
 #include <radeon_drm.h>
-#include "radeon/drm/radeon_drm.h"
+#include "radeon/drm/radeon_drm_public.h"
 
 static boolean
 drm_display_is_format_supported(struct native_display *ndpy,
diff --git a/src/gallium/state_trackers/egl/x11/native_dri2.c b/src/gallium/state_trackers/egl/x11/native_dri2.c
index 331a7de432..8108ce4586 100644
--- a/src/gallium/state_trackers/egl/x11/native_dri2.c
+++ b/src/gallium/state_trackers/egl/x11/native_dri2.c
@@ -136,6 +136,7 @@ dri2_surface_process_drawable_buffers(struct native_surface *nsurf,
    templ.width0 = dri2surf->width;
    templ.height0 = dri2surf->height;
    templ.depth0 = 1;
+   templ.array_size = 1;
    templ.format = dri2surf->color_format;
    templ.bind = PIPE_BIND_RENDER_TARGET;
 
diff --git a/src/gallium/state_trackers/glx/xlib/glx_api.c b/src/gallium/state_trackers/glx/xlib/glx_api.c
index dcd50e19d7..205a7e3c19 100644
--- a/src/gallium/state_trackers/glx/xlib/glx_api.c
+++ b/src/gallium/state_trackers/glx/xlib/glx_api.c
@@ -46,9 +46,6 @@
 #define SERVER_MAJOR_VERSION 1
 #define SERVER_MINOR_VERSION 4
 
-/* This is appended onto the glXGetClient/ServerString version strings. */
-#define MESA_GLX_VERSION "Mesa " MESA_VERSION_STRING
-
 /* Who implemented this GLX? */
 #define VENDOR "Brian Paul"
 
@@ -1672,7 +1669,7 @@ glXQueryServerString( Display *dpy, int screen, int name )
 {
    static char version[1000];
    sprintf(version, "%d.%d %s",
-	   SERVER_MAJOR_VERSION, SERVER_MINOR_VERSION, MESA_GLX_VERSION);
+	   SERVER_MAJOR_VERSION, SERVER_MINOR_VERSION, xmesa_get_name());
 
    (void) dpy;
    (void) screen;
@@ -1697,7 +1694,7 @@ glXGetClientString( Display *dpy, int name )
 {
    static char version[1000];
    sprintf(version, "%d.%d %s", CLIENT_MAJOR_VERSION,
-	   CLIENT_MINOR_VERSION, MESA_GLX_VERSION);
+	   CLIENT_MINOR_VERSION, xmesa_get_name());
 
    (void) dpy;
 
diff --git a/src/gallium/state_trackers/glx/xlib/xm_api.c b/src/gallium/state_trackers/glx/xlib/xm_api.c
index 8332633f01..7c47a25ceb 100644
--- a/src/gallium/state_trackers/glx/xlib/xm_api.c
+++ b/src/gallium/state_trackers/glx/xlib/xm_api.c
@@ -828,6 +828,16 @@ void XMesaDestroyVisual( XMesaVisual v )
 
 
 /**
+ * Return the informative name.
+ */
+const char *
+xmesa_get_name(void)
+{
+   return stapi->name;
+}
+
+
+/**
  * Do per-display initializations.
  */
 void
diff --git a/src/gallium/state_trackers/glx/xlib/xm_api.h b/src/gallium/state_trackers/glx/xlib/xm_api.h
index b8ac979edc..4ea42dc375 100644
--- a/src/gallium/state_trackers/glx/xlib/xm_api.h
+++ b/src/gallium/state_trackers/glx/xlib/xm_api.h
@@ -57,7 +57,7 @@ and create a window, you must do the following to use the X/Mesa interface:
 #define XMESA_H
 
 
-#include "main/core.h" /* for GLvisual and MESA_VERSION_STRING */
+#include "main/core.h" /* for gl_config */
 #include "state_tracker/st_api.h"
 #include "os/os_thread.h"
 
@@ -351,6 +351,9 @@ struct xmesa_buffer {
 
 
 
+extern const char *
+xmesa_get_name(void);
+
 extern void
 xmesa_init(Display *dpy);
 
diff --git a/src/gallium/state_trackers/glx/xlib/xm_st.c b/src/gallium/state_trackers/glx/xlib/xm_st.c
index e7466bdbee..6bfe8b0788 100644
--- a/src/gallium/state_trackers/glx/xlib/xm_st.c
+++ b/src/gallium/state_trackers/glx/xlib/xm_st.c
@@ -42,7 +42,7 @@ struct xmesa_st_framebuffer {
    unsigned texture_width, texture_height, texture_mask;
    struct pipe_resource *textures[ST_ATTACHMENT_COUNT];
 
-   struct pipe_surface *display_surface;
+   struct pipe_resource *display_resource;
 };
 
 static INLINE struct xmesa_st_framebuffer *
@@ -60,25 +60,19 @@ xmesa_st_framebuffer_display(struct st_framebuffer_iface *stfbi,
 {
    struct xmesa_st_framebuffer *xstfb = xmesa_st_framebuffer(stfbi);
    struct pipe_resource *ptex = xstfb->textures[statt];
-   struct pipe_surface *psurf;
+   struct pipe_resource *pres;
 
    if (!ptex)
       return TRUE;
 
-   psurf = xstfb->display_surface;
+   pres = xstfb->display_resource;
    /* (re)allocate the surface for the texture to be displayed */
-   if (!psurf || psurf->texture != ptex) {
-      pipe_surface_reference(&xstfb->display_surface, NULL);
-
-      psurf = xstfb->screen->get_tex_surface(xstfb->screen,
-            ptex, 0, 0, 0, PIPE_BIND_DISPLAY_TARGET);
-      if (!psurf)
-         return FALSE;
-
-      xstfb->display_surface = psurf;
+   if (!pres || pres != ptex) {
+      pipe_resource_reference(&xstfb->display_resource, ptex);
+      pres = xstfb->display_resource;
    }
 
-   xstfb->screen->flush_frontbuffer(xstfb->screen, psurf, &xstfb->buffer->ws);
+   xstfb->screen->flush_frontbuffer(xstfb->screen, pres, 0, 0, &xstfb->buffer->ws);
 
    return TRUE;
 }
@@ -96,7 +90,7 @@ xmesa_st_framebuffer_copy_textures(struct st_framebuffer_iface *stfbi,
    struct xmesa_st_framebuffer *xstfb = xmesa_st_framebuffer(stfbi);
    struct pipe_resource *src_ptex = xstfb->textures[src_statt];
    struct pipe_resource *dst_ptex = xstfb->textures[dst_statt];
-   struct pipe_subresource subsrc, subdst;
+   struct pipe_box src_box;
    struct pipe_context *pipe;
 
    if (!src_ptex || !dst_ptex)
@@ -110,14 +104,11 @@ xmesa_st_framebuffer_copy_textures(struct st_framebuffer_iface *stfbi,
       xstfb->display->pipe = pipe;
    }
 
-   subsrc.face = 0;
-   subsrc.level = 0;
-   subdst.face = 0;
-   subdst.level = 0;
+   u_box_2d(x, y, width, height, &src_box);
 
    if (src_ptex && dst_ptex)
-      pipe->resource_copy_region(pipe, dst_ptex, subdst, x, y, 0,
-                                 src_ptex, subsrc, x, y, 0, width, height);
+      pipe->resource_copy_region(pipe, dst_ptex, 0, x, y, 0,
+                                 src_ptex, 0, &src_box);
 }
 
 /**
@@ -144,6 +135,7 @@ xmesa_st_framebuffer_validate_textures(struct st_framebuffer_iface *stfbi,
    templ.width0 = width;
    templ.height0 = height;
    templ.depth0 = 1;
+   templ.array_size = 1;
    templ.last_level = 0;
 
    for (i = 0; i < ST_ATTACHMENT_COUNT; i++) {
@@ -321,7 +313,7 @@ xmesa_destroy_st_framebuffer(struct st_framebuffer_iface *stfbi)
    struct xmesa_st_framebuffer *xstfb = xmesa_st_framebuffer(stfbi);
    int i;
 
-   pipe_surface_reference(&xstfb->display_surface, NULL);
+   pipe_resource_reference(&xstfb->display_resource, NULL);
 
    for (i = 0; i < ST_ATTACHMENT_COUNT; i++)
       pipe_resource_reference(&xstfb->textures[i], NULL);
diff --git a/src/gallium/state_trackers/python/p_context.i b/src/gallium/state_trackers/python/p_context.i
index 40c4603fb9..5bdeaa8e4e 100644
--- a/src/gallium/state_trackers/python/p_context.i
+++ b/src/gallium/state_trackers/python/p_context.i
@@ -132,6 +132,8 @@ struct st_context {
                        enum pipe_format format = PIPE_FORMAT_NONE,
                        unsigned first_level = 0,
                        unsigned last_level = ~0,
+                       unsigned first_layer = 0,
+                       unsigned last_layer = ~0,
                        unsigned swizzle_r = 0,
                        unsigned swizzle_g = 1,
                        unsigned swizzle_b = 2,
@@ -146,9 +148,10 @@ struct st_context {
       } else {
          templat.format = format;
       }
-      templat.last_level = MIN2(last_level, texture->last_level);
-      templat.first_level = first_level;
-      templat.last_level = last_level;
+      templat.u.tex.last_level = MIN2(last_level, texture->last_level);
+      templat.u.tex.first_level = first_level;
+      templat.u.tex.first_layer = first_layer;
+      templat.u.tex.last_layer = last_layer;
       templat.swizzle_r = swizzle_r;
       templat.swizzle_g = swizzle_g;
       templat.swizzle_b = swizzle_b;
@@ -412,17 +415,15 @@ error1:
     */
 
    void resource_copy_region(struct pipe_resource *dst,
-                             struct pipe_subresource subdst,
+                             unsigned dst_level,
                              unsigned dstx, unsigned dsty, unsigned dstz,
                              struct pipe_resource *src,
-                             struct pipe_subresource subsrc,
-                             unsigned srcx, unsigned srcy, unsigned srcz,
-                             unsigned width, unsigned height)
+                             unsigned src_level,
+                             const struct pipe_box *src_box)
    {
       $self->pipe->resource_copy_region($self->pipe,
-                                        dst, subdst, dstx, dsty, dstz,
-                                        src, subsrc, srcx, srcy, srcz,
-                                        width, height);
+                                        dst, dst_level, dstx, dsty, dstz,
+                                        src, src_level, src_box);
    }
 
 
@@ -433,7 +434,7 @@ error1:
    {
       struct pipe_surface *_dst = NULL;
 
-     _dst = st_pipe_surface(dst, PIPE_BIND_RENDER_TARGET);
+     _dst = st_pipe_surface($self->pipe, dst, PIPE_BIND_RENDER_TARGET);
       if(!_dst)
          SWIG_exception(SWIG_ValueError, "couldn't acquire destination surface for writing");
 
@@ -452,7 +453,7 @@ error1:
    {
       struct pipe_surface *_dst = NULL;
 
-     _dst = st_pipe_surface(dst, PIPE_BIND_DEPTH_STENCIL);
+     _dst = st_pipe_surface($self->pipe, dst, PIPE_BIND_DEPTH_STENCIL);
       if(!_dst)
          SWIG_exception(SWIG_ValueError, "couldn't acquire destination surface for writing");
 
@@ -482,9 +483,8 @@ error1:
 
       transfer = pipe_get_transfer(pipe,
                                    surface->texture,
-                                   surface->face,
                                    surface->level,
-                                   surface->zslice,
+                                   surface->layer,
                                    PIPE_TRANSFER_READ,
                                    x, y, w, h);
       if(transfer) {
@@ -511,9 +511,8 @@ error1:
 
       transfer = pipe_get_transfer(pipe,
                                    surface->texture,
-                                   surface->face,
                                    surface->level,
-                                   surface->zslice,
+                                   surface->layer,
                                    PIPE_TRANSFER_WRITE,
                                    x, y, w, h);
       if(!transfer)
@@ -535,9 +534,8 @@ error1:
       struct pipe_transfer *transfer;
       transfer = pipe_get_transfer(pipe,
                                    surface->texture,
-                                   surface->face,
                                    surface->level,
-                                   surface->zslice,
+                                   surface->layer,
                                    PIPE_TRANSFER_READ,
                                    x, y, w, h);
       if(transfer) {
@@ -555,9 +553,8 @@ error1:
       struct pipe_transfer *transfer;
       transfer = pipe_get_transfer(pipe,
                                    surface->texture,
-                                   surface->face,
                                    surface->level,
-                                   surface->zslice,
+                                   surface->layer,
                                    PIPE_TRANSFER_WRITE,
                                    x, y, w, h);
       if(transfer) {
@@ -597,9 +594,8 @@ error1:
 
       transfer = pipe_get_transfer(pipe,
                                    surface->texture,
-                                   surface->face,
                                    surface->level,
-                                   surface->zslice,
+                                   surface->layer,
                                    PIPE_TRANSFER_READ,
                                    x, y, w, h);
       if(transfer) {
@@ -624,9 +620,8 @@ error1:
       struct pipe_transfer *transfer;
       transfer = pipe_get_transfer(pipe,
                                    surface->texture,
-                                   surface->face,
                                    surface->level,
-                                   surface->zslice,
+                                   surface->layer,
                                    PIPE_TRANSFER_READ,
                                    x, y, w, h);
       if(transfer) {
@@ -644,9 +639,8 @@ error1:
       struct pipe_transfer *transfer;
       transfer = pipe_get_transfer(pipe,
                                    surface->texture,
-                                   surface->face,
                                    surface->level,
-                                   surface->zslice,
+                                   surface->layer,
                                    PIPE_TRANSFER_WRITE,
                                    x, y, w, h);
       if(transfer) {
@@ -681,9 +675,8 @@ error1:
 
       transfer = pipe_get_transfer(pipe,
                                    surface->texture,
-                                   surface->face,
                                    surface->level,
-                                   surface->zslice,
+                                   surface->layer,
                                    PIPE_TRANSFER_READ,
                                    x, y, w, h);
       if(!transfer) {
@@ -715,16 +708,16 @@ error1:
    %cstring_input_binary(const char *STRING, unsigned LENGTH);
    void
    transfer_inline_write(struct pipe_resource *resource,
-                         struct pipe_subresource *sr,
+                         unsigned level,
                          unsigned usage,
                          const struct pipe_box *box,
                          const char *STRING, unsigned LENGTH,
                          unsigned stride,
-                         unsigned slice_stride)
+                         unsigned layer_stride)
    {
       struct pipe_context *pipe = $self->pipe;
 
-      pipe->transfer_inline_write(pipe, resource, *sr, usage, box, STRING, stride, slice_stride);
+      pipe->transfer_inline_write(pipe, resource, level, usage, box, STRING, stride, layer_stride);
    }
 
    %cstring_output_allocate_size(char **STRING, int *LENGTH, free(*$1));
diff --git a/src/gallium/state_trackers/python/p_state.i b/src/gallium/state_trackers/python/p_state.i
index c1e6ea1b43..0537557661 100644
--- a/src/gallium/state_trackers/python/p_state.i
+++ b/src/gallium/state_trackers/python/p_state.i
@@ -114,7 +114,8 @@
          SWIG_exception(SWIG_ValueError, "index out of bounds");
       
       if(surface) {
-         _surface = st_pipe_surface(surface, PIPE_BIND_RENDER_TARGET);
+         /* XXX need a context here */
+         _surface = st_pipe_surface(NULL, surface, PIPE_BIND_RENDER_TARGET);
          if(!_surface)
             SWIG_exception(SWIG_ValueError, "couldn't acquire surface for writing");
       }
@@ -131,7 +132,8 @@
       struct pipe_surface *_surface = NULL;
 
       if(surface) {
-         _surface = st_pipe_surface(surface, PIPE_BIND_DEPTH_STENCIL);
+         /* XXX need a context here */
+         _surface = st_pipe_surface(NULL, surface, PIPE_BIND_DEPTH_STENCIL);
          if(!_surface)
             SWIG_exception(SWIG_ValueError, "couldn't acquire surface for writing");
       }
diff --git a/src/gallium/state_trackers/python/p_texture.i b/src/gallium/state_trackers/python/p_texture.i
index ae506944c4..1f4b4fd596 100644
--- a/src/gallium/state_trackers/python/p_texture.i
+++ b/src/gallium/state_trackers/python/p_texture.i
@@ -42,9 +42,8 @@
 %ignore pipe_resource::screen;
 
 %immutable st_surface::texture;
-%immutable st_surface::face;
 %immutable st_surface::level;
-%immutable st_surface::zslice;
+%immutable st_surface::layer;
 
 %newobject pipe_resource::get_surface;
 
@@ -73,25 +72,23 @@
 
    /** Get a surface which is a "view" into a texture */
    struct st_surface *
-   get_surface(unsigned face=0, unsigned level=0, unsigned zslice=0)
+   get_surface(unsigned level=0, unsigned layer=0)
    {
       struct st_surface *surface;
 
-      if(face >= ($self->target == PIPE_TEXTURE_CUBE ? 6U : 1U))
-         SWIG_exception(SWIG_ValueError, "face out of bounds");
       if(level > $self->last_level)
          SWIG_exception(SWIG_ValueError, "level out of bounds");
-      if(zslice >= u_minify($self->depth0, level))
-         SWIG_exception(SWIG_ValueError, "zslice out of bounds");
+      if(layer >= ($self->target == PIPE_TEXTURE_3D ?
+                         u_minify($self->depth0, level) : $self->depth0))
+         SWIG_exception(SWIG_ValueError, "layer out of bounds");
 
       surface = CALLOC_STRUCT(st_surface);
       if(!surface)
          return NULL;
 
       pipe_resource_reference(&surface->texture, $self);
-      surface->face = face;
       surface->level = level;
-      surface->zslice = zslice;
+      surface->layer = layer;
 
       return surface;
 
@@ -113,9 +110,8 @@ struct st_surface
    %immutable;
 
    struct pipe_resource *texture;
-   unsigned face;
    unsigned level;
-   unsigned zslice;
+   unsigned layer;
 
 };
 
diff --git a/src/gallium/state_trackers/python/st_device.c b/src/gallium/state_trackers/python/st_device.c
index 29813456b5..d0a4295c7a 100644
--- a/src/gallium/state_trackers/python/st_device.c
+++ b/src/gallium/state_trackers/python/st_device.c
@@ -240,6 +240,7 @@ st_context_create(struct st_device *st_dev)
       templat.width0 = 1;
       templat.height0 = 1;
       templat.depth0 = 1;
+      templat.array_size = 1;
       templat.last_level = 0;
       templat.bind = PIPE_BIND_SAMPLER_VIEW;
    
@@ -252,7 +253,7 @@ st_context_create(struct st_device *st_dev)
 
 	 pipe->transfer_inline_write(pipe,
 				     st_ctx->default_texture,
-				     u_subresource(0,0),
+				     0,
 				     PIPE_TRANSFER_WRITE,
 				     &box,
 				     &zero,
diff --git a/src/gallium/state_trackers/python/st_device.h b/src/gallium/state_trackers/python/st_device.h
index 2dca7a1974..53e2556cb0 100644
--- a/src/gallium/state_trackers/python/st_device.h
+++ b/src/gallium/state_trackers/python/st_device.h
@@ -41,9 +41,8 @@ struct st_winsys;
 struct st_surface
 {
    struct pipe_resource *texture;
-   unsigned face;
    unsigned level;
-   unsigned zslice;
+   unsigned layer;
 };
 
 
@@ -83,11 +82,17 @@ struct st_device
 
 
 static INLINE struct pipe_surface *
-st_pipe_surface(struct st_surface *surface, unsigned usage) 
+st_pipe_surface(struct pipe_context *pipe, struct st_surface *surface, unsigned usage) 
 {
    struct pipe_resource *texture = surface->texture;
-   struct pipe_screen *screen = texture->screen;
-   return screen->get_tex_surface(screen, texture, surface->face, surface->level, surface->zslice, usage);
+   struct pipe_surface surf_tmpl;
+   memset(&surf_tmpl, 0, sizeof(surf_tmpl));
+   surf_tmpl.format = texture->format;
+   surf_tmpl.usage = usage;
+   surf_tmpl.u.tex.level = surface->level;
+   surf_tmpl.u.tex.first_layer = surface->layer;
+   surf_tmpl.u.tex.last_layer = surface->layer;
+   return pipe->create_surface(pipe, texture, &surf_tmpl);
 }
 
 struct st_context *
diff --git a/src/gallium/state_trackers/python/st_sample.c b/src/gallium/state_trackers/python/st_sample.c
index 25bfbf1ab7..cac9a1a620 100644
--- a/src/gallium/state_trackers/python/st_sample.c
+++ b/src/gallium/state_trackers/python/st_sample.c
@@ -555,9 +555,8 @@ st_sample_surface(struct pipe_context *pipe,
 
    transfer = pipe_get_transfer(pipe,
                                 surface->texture,
-                                surface->face,
                                 surface->level,
-                                surface->zslice,
+                                surface->layer,
                                 PIPE_TRANSFER_WRITE,
                                 0, 0,
                                 width,
diff --git a/src/gallium/state_trackers/vega/.gitignore b/src/gallium/state_trackers/vega/.gitignore
new file mode 100644
index 0000000000..c452229f7f
--- /dev/null
+++ b/src/gallium/state_trackers/vega/.gitignore
@@ -0,0 +1 @@
+api_tmp.h
diff --git a/src/gallium/state_trackers/vega/Makefile b/src/gallium/state_trackers/vega/Makefile
index e0a87151c4..7342c124c2 100644
--- a/src/gallium/state_trackers/vega/Makefile
+++ b/src/gallium/state_trackers/vega/Makefile
@@ -9,6 +9,9 @@ LIBRARY_INCLUDES = \
 	-I$(TOP)/include \
 	-I$(TOP)/src/mapi
 
+LIBRARY_DEFINES = \
+	-DVEGA_VERSION_STRING=\"$(MESA_VERSION)\"
+
 C_SOURCES = \
            api.c           \
            api_context.c   \
@@ -35,7 +38,14 @@ C_SOURCES = \
            renderer.c \
            stroker.c \
            mask.c \
+           text.c \
            shader.c \
            shaders_cache.c
 
+GENERATED_SOURCES := api_tmp.h
+
 include ../../Makefile.template
+
+MAPI := $(TOP)/src/mapi
+api_tmp.h: $(MAPI)/mapi/mapi_abi.py $(MAPI)/vgapi/vgapi.csv
+	$(PYTHON2) $< --printer vgapi --mode app $(MAPI)/vgapi/vgapi.csv > $@
diff --git a/src/gallium/state_trackers/vega/SConscript b/src/gallium/state_trackers/vega/SConscript
index a25b8474e4..4900135a1c 100644
--- a/src/gallium/state_trackers/vega/SConscript
+++ b/src/gallium/state_trackers/vega/SConscript
@@ -3,11 +3,16 @@
 
 Import('*')
 
+from sys import executable as python_cmd
+
 env = env.Clone()
 
 env.Append(CPPPATH = [
     '#/src/mapi',
 ])
+env.Append(CPPDEFINES = [
+    'VEGA_VERSION_STRING=',
+])
 
 vega_sources = [
     'api.c',
@@ -37,10 +42,16 @@ vega_sources = [
     'mask.c',
     'shader.c',
     'shaders_cache.c',
+    'text.c',
 ]
 
-# vgapi_header must be generated first
-env.Depends(vega_sources, vgapi_header)
+api_tmp = env.CodeGenerate(
+        target = '#/src/gallium/state_trackers/vega/api_tmp.h',
+        script = '#src/mapi/mapi/mapi_abi.py',
+        source = '#src/mapi/vgapi/vgapi.csv',
+        command = python_cmd + ' $SCRIPT --printer vgapi --mode app $SOURCE > $TARGET'
+)
+env.Depends(vega_sources, api_tmp)
 
 st_vega = env.ConvenienceLibrary(
     target = 'st_vega',
diff --git a/src/gallium/state_trackers/vega/api.c b/src/gallium/state_trackers/vega/api.c
index bf1d37493a..4bf7c71d45 100644
--- a/src/gallium/state_trackers/vega/api.c
+++ b/src/gallium/state_trackers/vega/api.c
@@ -28,23 +28,10 @@
 
 #include "mapi/mapi.h"
 
+/* define vega_spec and vega_procs for use with mapi */
+#define API_TMP_DEFINE_SPEC
 #include "api.h"
 
-static const char vega_spec[] =
-   "1"
-#define MAPI_ABI_ENTRY(ret, name, params) \
-   "\0" #name "\0"
-#define MAPI_ALIAS_ENTRY(alias, ret, name, params) \
-   #name "\0"
-#include "vgapi/vgapi_tmp.h"
-   "\0";
-
-static const mapi_proc vega_procs[] = {
-#define MAPI_ABI_ENTRY(ret, name, params) \
-   (mapi_proc) vega ## name,
-#include "vgapi/vgapi_tmp.h"
-};
-
 static void api_init(void)
 {
    static boolean initialized = FALSE;
diff --git a/src/gallium/state_trackers/vega/api.h b/src/gallium/state_trackers/vega/api.h
index 955508dae9..459dafb4bb 100644
--- a/src/gallium/state_trackers/vega/api.h
+++ b/src/gallium/state_trackers/vega/api.h
@@ -33,10 +33,7 @@
 #include "VG/vgext.h"
 #include "vg_manager.h"
 
-/* declare the prototypes */
-#define MAPI_ABI_ENTRY(ret, name, params) \
-   ret VG_API_ENTRY vega ## name params;
-#include "vgapi/vgapi_tmp.h"
+#include "api_tmp.h"
 
 struct mapi_table;
 
diff --git a/src/gallium/state_trackers/vega/api_context.c b/src/gallium/state_trackers/vega/api_context.c
index 0d04d8e871..d6bbda5e07 100644
--- a/src/gallium/state_trackers/vega/api_context.c
+++ b/src/gallium/state_trackers/vega/api_context.c
@@ -73,7 +73,8 @@ void vegaFinish(void)
    pipe = ctx->pipe;
 
    pipe->flush(pipe, PIPE_FLUSH_RENDER_CACHE | PIPE_FLUSH_FRAME, &fence);
-
-   pipe->screen->fence_finish(pipe->screen, fence, 0);
-   pipe->screen->fence_reference(pipe->screen, &fence, NULL);
+   if (fence) {
+      pipe->screen->fence_finish(pipe->screen, fence, 0);
+      pipe->screen->fence_reference(pipe->screen, &fence, NULL);
+   }
 }
diff --git a/src/gallium/state_trackers/vega/api_filters.c b/src/gallium/state_trackers/vega/api_filters.c
index 4bd5d7e4bc..724e38241b 100644
--- a/src/gallium/state_trackers/vega/api_filters.c
+++ b/src/gallium/state_trackers/vega/api_filters.c
@@ -31,7 +31,6 @@
 #include "api.h"
 #include "renderer.h"
 #include "shaders_cache.h"
-#include "st_inlines.h"
 
 #include "pipe/p_context.h"
 #include "pipe/p_state.h"
@@ -42,7 +41,6 @@
 #include "util/u_sampler.h"
 #include "util/u_string.h"
 
-
 #include "asm_filters.h"
 
 
@@ -73,6 +71,7 @@ static INLINE struct pipe_resource *create_texture_1d(struct vg_context *ctx,
    templ.width0 = color_data_len;
    templ.height0 = 1;
    templ.depth0 = 1;
+   templ.array_size = 1;
    templ.bind = PIPE_BIND_SAMPLER_VIEW;
 
    tex = screen->resource_create(screen, &templ);
@@ -80,9 +79,9 @@ static INLINE struct pipe_resource *create_texture_1d(struct vg_context *ctx,
    { /* upload color_data */
       struct pipe_transfer *transfer =
          pipe_get_transfer(pipe, tex,
-				0, 0, 0,
-				PIPE_TRANSFER_READ_WRITE ,
-				0, 0, tex->width0, tex->height0);
+                           0, 0,
+                           PIPE_TRANSFER_READ_WRITE ,
+                           0, 0, tex->width0, tex->height0);
       void *map = pipe->transfer_map(pipe, transfer);
       memcpy(map, color_data, sizeof(VGint)*color_data_len);
       pipe->transfer_unmap(pipe, transfer);
@@ -114,153 +113,11 @@ static INLINE struct pipe_sampler_view *create_texture_1d_view(struct vg_context
    return view;
 }
 
-static INLINE struct pipe_surface * setup_framebuffer(struct vg_image *dst)
-{
-   struct vg_context *ctx = vg_current_context();
-   struct pipe_context *pipe = ctx->pipe;
-   struct pipe_framebuffer_state fb;
-   struct pipe_surface *dst_surf = pipe->screen->get_tex_surface(
-      pipe->screen, dst->sampler_view->texture, 0, 0, 0,
-      PIPE_BIND_RENDER_TARGET);
-
-   /* drawing dest */
-   memset(&fb, 0, sizeof(fb));
-   fb.width  = dst->x + dst_surf->width;
-   fb.height = dst->y + dst_surf->height;
-   fb.nr_cbufs = 1;
-   fb.cbufs[0] = dst_surf;
-   {
-      VGint i;
-      for (i = 1; i < PIPE_MAX_COLOR_BUFS; ++i)
-         fb.cbufs[i] = 0;
-   }
-   cso_set_framebuffer(ctx->cso_context, &fb);
-
-   return dst_surf;
-}
-
-static void setup_viewport(struct vg_image *dst)
-{
-   struct vg_context *ctx = vg_current_context();
-   vg_set_viewport(ctx, VEGA_Y0_TOP);
-}
-
-static void setup_blend()
-{
-   struct vg_context *ctx = vg_current_context();
-   struct pipe_blend_state blend;
-   memset(&blend, 0, sizeof(blend));
-   blend.rt[0].rgb_src_factor = PIPE_BLENDFACTOR_ONE;
-   blend.rt[0].alpha_src_factor = PIPE_BLENDFACTOR_ONE;
-   blend.rt[0].rgb_dst_factor = PIPE_BLENDFACTOR_ZERO;
-   blend.rt[0].alpha_dst_factor = PIPE_BLENDFACTOR_ZERO;
-   if (ctx->state.vg.filter_channel_mask & VG_RED)
-      blend.rt[0].colormask |= PIPE_MASK_R;
-   if (ctx->state.vg.filter_channel_mask & VG_GREEN)
-      blend.rt[0].colormask |= PIPE_MASK_G;
-   if (ctx->state.vg.filter_channel_mask & VG_BLUE)
-      blend.rt[0].colormask |= PIPE_MASK_B;
-   if (ctx->state.vg.filter_channel_mask & VG_ALPHA)
-      blend.rt[0].colormask |= PIPE_MASK_A;
-   blend.rt[0].blend_enable = 0;
-   cso_set_blend(ctx->cso_context, &blend);
-}
-
-static void setup_constant_buffer(struct vg_context *ctx, const void *buffer,
-                                  VGint param_bytes)
-{
-   struct pipe_context *pipe = ctx->pipe;
-   struct pipe_resource **cbuf = &ctx->filter.buffer;
-
-   /* We always need to get a new buffer, to keep the drivers simple and
-    * avoid gratuitous rendering synchronization. */
-   pipe_resource_reference(cbuf, NULL);
-
-   *cbuf = pipe_buffer_create(pipe->screen, 
-                              PIPE_BIND_CONSTANT_BUFFER,
-                              param_bytes);
-
-   if (*cbuf) {
-      st_no_flush_pipe_buffer_write(ctx, *cbuf,
-                                    0, param_bytes, buffer);
-   }
-
-   ctx->pipe->set_constant_buffer(ctx->pipe, PIPE_SHADER_FRAGMENT, 0, *cbuf);
-}
-
-static void setup_samplers(struct vg_context *ctx, struct filter_info *info)
-{
-   struct pipe_sampler_state *samplers[PIPE_MAX_SAMPLERS];
-   struct pipe_sampler_view *sampler_views[PIPE_MAX_SAMPLERS];
-   struct pipe_sampler_state sampler[3];
-   int num_samplers = 0;
-   int num_textures = 0;
-
-   samplers[0] = NULL;
-   samplers[1] = NULL;
-   samplers[2] = NULL;
-   samplers[3] = NULL;
-   sampler_views[0] = NULL;
-   sampler_views[1] = NULL;
-   sampler_views[2] = NULL;
-   sampler_views[3] = NULL;
-
-   memset(&sampler[0], 0, sizeof(struct pipe_sampler_state));
-   sampler[0].wrap_s = PIPE_TEX_WRAP_CLAMP_TO_EDGE;
-   sampler[0].wrap_t = PIPE_TEX_WRAP_CLAMP_TO_EDGE;
-   sampler[0].wrap_r = PIPE_TEX_WRAP_CLAMP_TO_EDGE;
-   sampler[0].min_img_filter = PIPE_TEX_MIPFILTER_LINEAR;
-   sampler[0].mag_img_filter = PIPE_TEX_MIPFILTER_LINEAR;
-   sampler[0].normalized_coords = 1;
-
-   switch(info->tiling_mode) {
-   case VG_TILE_FILL:
-      sampler[0].wrap_s = PIPE_TEX_WRAP_CLAMP_TO_BORDER;
-      sampler[0].wrap_t = PIPE_TEX_WRAP_CLAMP_TO_BORDER;
-      memcpy(sampler[0].border_color,
-             ctx->state.vg.tile_fill_color,
-             sizeof(VGfloat) * 4);
-      break;
-   case VG_TILE_PAD:
-      sampler[0].wrap_s = PIPE_TEX_WRAP_CLAMP_TO_EDGE;
-      sampler[0].wrap_t = PIPE_TEX_WRAP_CLAMP_TO_EDGE;
-      break;
-   case VG_TILE_REPEAT:
-      sampler[0].wrap_s = PIPE_TEX_WRAP_REPEAT;
-      sampler[0].wrap_t = PIPE_TEX_WRAP_REPEAT;
-      break;
-   case VG_TILE_REFLECT:
-      sampler[0].wrap_s = PIPE_TEX_WRAP_MIRROR_REPEAT;
-      sampler[0].wrap_t = PIPE_TEX_WRAP_MIRROR_REPEAT;
-      break;
-   default:
-      debug_assert(!"Unknown tiling mode");
-   }
-
-   samplers[0] = &sampler[0];
-   sampler_views[0] = info->src->sampler_view;
-   ++num_samplers;
-   ++num_textures;
-
-   if (info->extra_texture_view) {
-      memcpy(&sampler[1], &sampler[0], sizeof(struct pipe_sampler_state));
-      samplers[1] = &sampler[1];
-      sampler_views[1] = info->extra_texture_view;
-      ++num_samplers;
-      ++num_textures;
-   }
-
-
-   cso_set_samplers(ctx->cso_context, num_samplers, (const struct pipe_sampler_state **)samplers);
-   cso_set_fragment_sampler_views(ctx->cso_context, num_textures, sampler_views);
-}
-
 static struct vg_shader * setup_color_matrix(struct vg_context *ctx, void *user_data)
 {
    struct vg_shader *shader =
       shader_create_from_text(ctx->pipe, color_matrix_asm, 200,
          PIPE_SHADER_FRAGMENT);
-   cso_set_fragment_shader_handle(ctx->cso_context, shader->driver);
    return shader;
 }
 
@@ -275,7 +132,6 @@ static struct vg_shader * setup_convolution(struct vg_context *ctx, void *user_d
    shader = shader_create_from_text(ctx->pipe, buffer, 200,
                                     PIPE_SHADER_FRAGMENT);
 
-   cso_set_fragment_shader_handle(ctx->cso_context, shader->driver);
    return shader;
 }
 
@@ -285,7 +141,6 @@ static struct vg_shader * setup_lookup(struct vg_context *ctx, void *user_data)
       shader_create_from_text(ctx->pipe, lookup_asm,
                               200, PIPE_SHADER_FRAGMENT);
 
-   cso_set_fragment_shader_handle(ctx->cso_context, shader->driver);
    return shader;
 }
 
@@ -316,49 +171,68 @@ static struct vg_shader * setup_lookup_single(struct vg_context *ctx, void *user
    shader = shader_create_from_text(ctx->pipe, buffer, 200,
                                     PIPE_SHADER_FRAGMENT);
 
-   cso_set_fragment_shader_handle(ctx->cso_context, shader->driver);
    return shader;
 }
 
 static void execute_filter(struct vg_context *ctx,
                            struct filter_info *info)
 {
-   struct pipe_surface *dst_surf;
    struct vg_shader *shader;
+   const struct pipe_sampler_state *samplers[2];
+   struct pipe_sampler_view *views[2];
+   struct pipe_sampler_state sampler;
+   uint tex_wrap;
+
+   memset(&sampler, 0, sizeof(sampler));
+   sampler.min_img_filter = PIPE_TEX_FILTER_LINEAR;
+   sampler.mag_img_filter = PIPE_TEX_FILTER_LINEAR;
+   sampler.min_mip_filter = PIPE_TEX_MIPFILTER_NONE;
+   sampler.normalized_coords = 1;
+
+   switch (info->tiling_mode) {
+   case VG_TILE_FILL:
+      tex_wrap = PIPE_TEX_WRAP_CLAMP_TO_BORDER;
+      /* copy border color */
+      memcpy(sampler.border_color, ctx->state.vg.tile_fill_color,
+            sizeof(sampler.border_color));
+      break;
+   case VG_TILE_PAD:
+      tex_wrap = PIPE_TEX_WRAP_CLAMP_TO_EDGE;;
+      break;
+   case VG_TILE_REPEAT:
+      tex_wrap = PIPE_TEX_WRAP_REPEAT;;
+      break;
+   case VG_TILE_REFLECT:
+      tex_wrap = PIPE_TEX_WRAP_MIRROR_REPEAT;
+      break;
+   default:
+      debug_assert(!"Unknown tiling mode");
+      tex_wrap = 0;
+      break;
+   }
+
+   sampler.wrap_s = tex_wrap;
+   sampler.wrap_t = tex_wrap;
+   sampler.wrap_r = PIPE_TEX_WRAP_CLAMP_TO_EDGE;
+
+   samplers[0] = samplers[1] = &sampler;
+   views[0] = info->src->sampler_view;
+   views[1] = info->extra_texture_view;
 
-   cso_save_framebuffer(ctx->cso_context);
-   cso_save_fragment_shader(ctx->cso_context);
-   cso_save_viewport(ctx->cso_context);
-   cso_save_blend(ctx->cso_context);
-   cso_save_samplers(ctx->cso_context);
-   cso_save_fragment_sampler_views(ctx->cso_context);
-
-   dst_surf = setup_framebuffer(info->dst);
-   setup_viewport(info->dst);
-   setup_blend();
-   setup_constant_buffer(ctx, info->const_buffer, info->const_buffer_len);
    shader = info->setup_shader(ctx, info->user_data);
-   setup_samplers(ctx, info);
-
-   renderer_draw_texture(ctx->renderer,
-                         info->src->sampler_view->texture,
-                         info->dst->x, info->dst->y,
-                         info->dst->x + info->dst->width,
-                         info->dst->y + info->dst->height,
-                         info->dst->x, info->dst->y,
-                         info->dst->x + info->dst->width,
-                         info->dst->y + info->dst->height);
-
-   cso_restore_framebuffer(ctx->cso_context);
-   cso_restore_fragment_shader(ctx->cso_context);
-   cso_restore_viewport(ctx->cso_context);
-   cso_restore_blend(ctx->cso_context);
-   cso_restore_samplers(ctx->cso_context);
-   cso_restore_fragment_sampler_views(ctx->cso_context);
 
-   vg_shader_destroy(ctx, shader);
+   if (renderer_filter_begin(ctx->renderer,
+            info->dst->sampler_view->texture, VG_TRUE,
+            ctx->state.vg.filter_channel_mask,
+            samplers, views, (info->extra_texture_view) ? 2 : 1,
+            shader->driver, info->const_buffer, info->const_buffer_len)) {
+      renderer_filter(ctx->renderer,
+            info->dst->x, info->dst->y, info->dst->width, info->dst->height,
+            info->src->x, info->src->y, info->src->width, info->src->height);
+      renderer_filter_end(ctx->renderer);
+   }
 
-   pipe_surface_reference(&dst_surf, NULL);
+   vg_shader_destroy(ctx, shader);
 }
 
 void vegaColorMatrix(VGImage dst, VGImage src,
@@ -796,6 +670,8 @@ void vegaLookupSingle(VGImage dst, VGImage src,
       return;
    }
 
+   vg_validate_state(ctx);
+
    for (i = 0; i < 256; ++i) {
       VGuint rgba = lookupTable[i];
       VGubyte blue, green, red, alpha;
diff --git a/src/gallium/state_trackers/vega/api_images.c b/src/gallium/state_trackers/vega/api_images.c
index 7054d9bb7e..ad95409cd0 100644
--- a/src/gallium/state_trackers/vega/api_images.c
+++ b/src/gallium/state_trackers/vega/api_images.c
@@ -37,6 +37,7 @@
 #include "pipe/p_screen.h"
 #include "util/u_inlines.h"
 #include "util/u_tile.h"
+#include "util/u_math.h"
 
 static INLINE VGboolean supported_image_format(VGImageFormat format)
 {
@@ -302,7 +303,8 @@ void vegaDrawImage(VGImage image)
    }
 
    vg_validate_state(ctx);
-   image_draw((struct vg_image*)image);
+   image_draw((struct vg_image*)image,
+         &ctx->state.vg.image_user_to_surface_matrix);
 }
 
 void vegaSetPixels(VGint dx, VGint dy,
@@ -398,11 +400,9 @@ void vegaReadPixels(void * data, VGint dataStride,
 
    struct st_framebuffer *stfb = ctx->draw_buffer;
    struct st_renderbuffer *strb = stfb->strb;
-   struct pipe_framebuffer_state *fb = &ctx->state.g3d.fb;
 
    VGfloat temp[VEGA_MAX_IMAGE_WIDTH][4];
    VGfloat *df = (VGfloat*)temp;
-   VGint y = (fb->height - sy) - 1, yStep = -1;
    VGint i;
    VGubyte *dst = (VGubyte *)data;
    VGint xoffset = 0, yoffset = 0;
@@ -430,18 +430,26 @@ void vegaReadPixels(void * data, VGint dataStride,
    }
    if (sy < 0) {
       yoffset = -sy;
+      yoffset *= dataStride;
       height += sy;
       sy = 0;
-      y = (fb->height - sy) - 1;
-      yoffset *= dataStride;
+   }
+
+   if (sx + width > stfb->width || sy + height > stfb->height) {
+      width = stfb->width - sx;
+      height = stfb->height - sy;
+      /* nothing to read */
+      if (width <= 0 || height <= 0)
+         return;
    }
 
    {
+      VGint y = (stfb->height - sy) - 1, yStep = -1;
       struct pipe_transfer *transfer;
 
-      transfer = pipe_get_transfer(pipe, strb->texture,  0, 0, 0,
-				   PIPE_TRANSFER_READ,
-				   0, 0, width, height);
+      transfer = pipe_get_transfer(pipe, strb->texture,  0, 0,
+                                   PIPE_TRANSFER_READ,
+                                   0, 0, sx + width, stfb->height - sy);
 
       /* Do a row at a time to flip image data vertically */
       for (i = 0; i < height; i++) {
@@ -464,8 +472,8 @@ void vegaCopyPixels(VGint dx, VGint dy,
                     VGint width, VGint height)
 {
    struct vg_context *ctx = vg_current_context();
-   struct pipe_framebuffer_state *fb = &ctx->state.g3d.fb;
-   struct st_renderbuffer *strb = ctx->draw_buffer->strb;
+   struct st_framebuffer *stfb = ctx->draw_buffer;
+   struct st_renderbuffer *strb = stfb->strb;
 
    if (width <= 0 || height <= 0) {
       vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR);
@@ -473,8 +481,8 @@ void vegaCopyPixels(VGint dx, VGint dy,
    }
 
    /* do nothing if we copy from outside the fb */
-   if (dx >= (VGint)fb->width || dy >= (VGint)fb->height ||
-       sx >= (VGint)fb->width || sy >= (VGint)fb->height)
+   if (dx >= (VGint)stfb->width || dy >= (VGint)stfb->height ||
+       sx >= (VGint)stfb->width || sy >= (VGint)stfb->height)
       return;
 
    vg_validate_state(ctx);
diff --git a/src/gallium/state_trackers/vega/api_masks.c b/src/gallium/state_trackers/vega/api_masks.c
index 189390ec2d..beb15c33a5 100644
--- a/src/gallium/state_trackers/vega/api_masks.c
+++ b/src/gallium/state_trackers/vega/api_masks.c
@@ -28,6 +28,7 @@
 
 #include "mask.h"
 #include "api.h"
+#include "renderer.h"
 
 #include "vg_context.h"
 #include "pipe/p_context.h"
@@ -36,119 +37,6 @@
 #include "util/u_pack_color.h"
 #include "util/u_draw_quad.h"
 
-#define DISABLE_1_1_MASKING 1
-
-/**
- * Draw a screen-aligned quadrilateral.
- * Coords are window coords with y=0=bottom.  These coords will be transformed
- * by the vertex shader and viewport transform.
- */
-static void
-draw_clear_quad(struct vg_context *st,
-                float x0, float y0, float x1, float y1, float z,
-                const VGfloat color[4])
-{
-   struct pipe_context *pipe = st->pipe;
-   struct pipe_resource *buf;
-   VGuint i;
-
-   /* positions */
-   st->clear.vertices[0][0][0] = x0;
-   st->clear.vertices[0][0][1] = y0;
-
-   st->clear.vertices[1][0][0] = x1;
-   st->clear.vertices[1][0][1] = y0;
-
-   st->clear.vertices[2][0][0] = x1;
-   st->clear.vertices[2][0][1] = y1;
-
-   st->clear.vertices[3][0][0] = x0;
-   st->clear.vertices[3][0][1] = y1;
-
-   /* same for all verts: */
-   for (i = 0; i < 4; i++) {
-      st->clear.vertices[i][0][2] = z;
-      st->clear.vertices[i][0][3] = 1.0;
-      st->clear.vertices[i][1][0] = color[0];
-      st->clear.vertices[i][1][1] = color[1];
-      st->clear.vertices[i][1][2] = color[2];
-      st->clear.vertices[i][1][3] = color[3];
-   }
-
-
-   /* put vertex data into vbuf */
-   buf =  pipe_user_buffer_create(pipe->screen,
-                                  st->clear.vertices,
-                                  sizeof(st->clear.vertices),
-				  PIPE_BIND_VERTEX_BUFFER);
-
-
-   /* draw */
-   if (buf) {
-      cso_set_vertex_elements(st->cso_context, 2, st->velems);
-
-      util_draw_vertex_buffer(pipe, buf, 0,
-                              PIPE_PRIM_TRIANGLE_FAN,
-                              4,  /* verts */
-                              2); /* attribs/vert */
-
-      pipe_resource_reference(&buf, NULL);
-   }
-}
-
-/**
- * Do vgClear by drawing a quadrilateral.
- */
-static void
-clear_with_quad(struct vg_context *st, float x0, float y0,
-                float width, float height, const VGfloat clear_color[4])
-{
-   VGfloat x1, y1;
-
-   vg_validate_state(st);
-
-   x1 = x0 + width;
-   y1 = y0 + height;
-
-   /*
-     printf("%s %f,%f %f,%f\n", __FUNCTION__,
-     x0, y0,
-     x1, y1);
-   */
-
-   cso_save_blend(st->cso_context);
-   cso_save_rasterizer(st->cso_context);
-   cso_save_fragment_shader(st->cso_context);
-   cso_save_vertex_shader(st->cso_context);
-
-   /* blend state: RGBA masking */
-   {
-      struct pipe_blend_state blend;
-      memset(&blend, 0, sizeof(blend));
-      blend.rt[0].rgb_src_factor = PIPE_BLENDFACTOR_ONE;
-      blend.rt[0].alpha_src_factor = PIPE_BLENDFACTOR_ONE;
-      blend.rt[0].rgb_dst_factor = PIPE_BLENDFACTOR_ZERO;
-      blend.rt[0].alpha_dst_factor = PIPE_BLENDFACTOR_ZERO;
-      blend.rt[0].colormask = PIPE_MASK_RGBA;
-      cso_set_blend(st->cso_context, &blend);
-   }
-
-   cso_set_rasterizer(st->cso_context, &st->clear.raster);
-
-   cso_set_fragment_shader_handle(st->cso_context, st->clear.fs);
-   cso_set_vertex_shader_handle(st->cso_context, vg_clear_vs(st));
-
-   /* draw quad matching scissor rect (XXX verify coord round-off) */
-   draw_clear_quad(st, x0, y0, x1, y1, 0, clear_color);
-
-   /* Restore pipe state */
-   cso_restore_blend(st->cso_context);
-   cso_restore_rasterizer(st->cso_context);
-   cso_restore_fragment_shader(st->cso_context);
-   cso_restore_vertex_shader(st->cso_context);
-}
-
-
 void vegaMask(VGHandle mask, VGMaskOperation operation,
               VGint x, VGint y,
               VGint width, VGint height)
@@ -176,12 +64,8 @@ void vegaMask(VGHandle mask, VGMaskOperation operation,
       struct vg_image *image = (struct vg_image *)mask;
       mask_using_image(image, operation, x, y, width, height);
    } else if (vg_object_is_valid((void*)mask, VG_OBJECT_MASK)) {
-#if DISABLE_1_1_MASKING
-      return;
-#else
       struct vg_mask_layer *layer = (struct vg_mask_layer *)mask;
       mask_using_layer(layer, operation, x, y, width, height);
-#endif
    } else {
       vg_set_error(ctx, VG_BAD_HANDLE_ERROR);
    }
@@ -191,7 +75,7 @@ void vegaClear(VGint x, VGint y,
                VGint width, VGint height)
 {
    struct vg_context *ctx = vg_current_context();
-   struct pipe_framebuffer_state *fb;
+   struct st_framebuffer *stfb = ctx->draw_buffer;
 
    if (width <= 0 || height <= 0) {
       vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR);
@@ -208,14 +92,15 @@ void vegaClear(VGint x, VGint y,
                 ctx->state.vg.clear_color[3]);
 #endif
 
-   fb = &ctx->state.g3d.fb;
    /* check for a whole surface clear */
    if (!ctx->state.vg.scissoring &&
-       (x == 0 && y == 0 && width == fb->width && height == fb->height)) {
+       (x == 0 && y == 0 && width == stfb->width && height == stfb->height)) {
       ctx->pipe->clear(ctx->pipe, PIPE_CLEAR_COLOR | PIPE_CLEAR_DEPTHSTENCIL,
                        ctx->state.vg.clear_color, 1., 0);
-   } else {
-      clear_with_quad(ctx, x, y, width, height, ctx->state.vg.clear_color);
+   } else if (renderer_clear_begin(ctx->renderer)) {
+      /* XXX verify coord round-off */
+      renderer_clear(ctx->renderer, x, y, width, height, ctx->state.vg.clear_color);
+      renderer_clear_end(ctx->renderer);
    }
 }
 
@@ -247,10 +132,6 @@ void vegaRenderToMask(VGPath path,
       return;
    }
 
-#if DISABLE_1_1_MASKING
-   return;
-#endif
-
    vg_validate_state(ctx);
 
    mask_render_to((struct path *)path, paintModes, operation);
@@ -328,9 +209,8 @@ void vegaFillMaskLayer(VGMaskLayer maskLayer,
       return;
    }
 
-#if DISABLE_1_1_MASKING
-   return;
-#endif
+   vg_validate_state(ctx);
+
    mask_layer_fill(mask, x, y, width, height, value);
 }
 
@@ -355,9 +235,7 @@ void vegaCopyMask(VGMaskLayer maskLayer,
       return;
    }
 
-#if DISABLE_1_1_MASKING
-   return;
-#endif
+   vg_validate_state(ctx);
 
    mask = (struct vg_mask_layer*)maskLayer;
    mask_copy(mask, sx, sy, dx, dy, width, height);
diff --git a/src/gallium/state_trackers/vega/api_misc.c b/src/gallium/state_trackers/vega/api_misc.c
index e648549745..9e2ab03e01 100644
--- a/src/gallium/state_trackers/vega/api_misc.c
+++ b/src/gallium/state_trackers/vega/api_misc.c
@@ -63,8 +63,8 @@ const VGubyte *vegaGetString(VGStringID name)
 {
    struct vg_context *ctx = vg_current_context();
    static const VGubyte *vendor = (VGubyte *)"Tungsten Graphics, Inc";
-   static const VGubyte *renderer = (VGubyte *)"Vega OpenVG 1.0";
-   static const VGubyte *version = (VGubyte *)"1.0";
+   static const VGubyte *renderer = (VGubyte *)"Vega OpenVG 1.1";
+   static const VGubyte *version = (VGubyte *)"1.1";
 
    if (!ctx)
       return NULL;
diff --git a/src/gallium/state_trackers/vega/api_params.c b/src/gallium/state_trackers/vega/api_params.c
index a10b009e63..a73b6c3eff 100644
--- a/src/gallium/state_trackers/vega/api_params.c
+++ b/src/gallium/state_trackers/vega/api_params.c
@@ -30,6 +30,7 @@
 #include "paint.h"
 #include "path.h"
 #include "image.h"
+#include "text.h"
 #include "matrix.h"
 #include "api_consts.h"
 #include "api.h"
@@ -174,6 +175,7 @@ void vegaSeti (VGParamType type, VGint value)
          error = VG_ILLEGAL_ARGUMENT_ERROR;
       else
          state->image_mode = value;
+      break;
 #ifdef OPENVG_VERSION_1_1
    case VG_COLOR_TRANSFORM:
       state->color_transform = value;
@@ -1500,7 +1502,8 @@ VGint vegaGetParameteri(VGHandle object,
 
 #ifdef OPENVG_VERSION_1_1
    case VG_FONT_NUM_GLYPHS: {
-      return 1;
+      struct vg_font *font = (struct vg_font*)object;
+      return font_num_glyphs(font);
    }
       break;
 #endif
diff --git a/src/gallium/state_trackers/vega/api_path.c b/src/gallium/state_trackers/vega/api_path.c
index f76adddb58..fe57b7671d 100644
--- a/src/gallium/state_trackers/vega/api_path.c
+++ b/src/gallium/state_trackers/vega/api_path.c
@@ -479,6 +479,7 @@ void vegaDrawPath(VGPath path, VGbitfield paintModes)
 
    if (path_is_empty((struct path*)path))
       return;
-   path_render((struct path*)path, paintModes);
+   path_render((struct path*)path, paintModes,
+         &ctx->state.vg.path_user_to_surface_matrix);
 }
 
diff --git a/src/gallium/state_trackers/vega/api_text.c b/src/gallium/state_trackers/vega/api_text.c
index 2a62da0a1d..7c6b479409 100644
--- a/src/gallium/state_trackers/vega/api_text.c
+++ b/src/gallium/state_trackers/vega/api_text.c
@@ -27,21 +27,15 @@
 #include "VG/openvg.h"
 
 #include "vg_context.h"
+#include "text.h"
+#include "api.h"
 
 #include "util/u_memory.h"
 
 #ifdef OPENVG_VERSION_1_1
 
-struct vg_font {
-   struct vg_object base;
-
-   VGint glyph_indices[200];
-   VGint num_glyphs;
-};
-
 VGFont vegaCreateFont(VGint glyphCapacityHint)
 {
-   struct vg_font *font = 0;
    struct vg_context *ctx = vg_current_context();
 
    if (glyphCapacityHint < 0) {
@@ -49,10 +43,7 @@ VGFont vegaCreateFont(VGint glyphCapacityHint)
       return VG_INVALID_HANDLE;
    }
 
-   font = CALLOC_STRUCT(vg_font);
-   vg_init_object(&font->base, ctx, VG_OBJECT_FONT);
-   vg_context_add_object(ctx, VG_OBJECT_FONT, font);
-   return (VGFont)font;
+   return (VGFont) font_create(glyphCapacityHint);
 }
 
 void vegaDestroyFont(VGFont f)
@@ -64,20 +55,23 @@ void vegaDestroyFont(VGFont f)
       vg_set_error(ctx, VG_BAD_HANDLE_ERROR);
       return;
    }
+   if (!vg_object_is_valid((void *) font, VG_OBJECT_FONT)) {
+      vg_set_error(ctx, VG_BAD_HANDLE_ERROR);
+      return;
+   }
 
-   vg_context_remove_object(ctx, VG_OBJECT_FONT, font);
-   /*free(font);*/
+   font_destroy(font);
 }
 
 void vegaSetGlyphToPath(VGFont font,
                         VGuint glyphIndex,
                         VGPath path,
                         VGboolean isHinted,
-                        VGfloat glyphOrigin [2],
-                        VGfloat escapement[2])
+                        const VGfloat glyphOrigin[2],
+                        const VGfloat escapement[2])
 {
    struct vg_context *ctx = vg_current_context();
-   struct vg_object *pathObj;
+   struct path *pathObj;
    struct vg_font *f;
 
    if (font == VG_INVALID_HANDLE ||
@@ -95,25 +89,22 @@ void vegaSetGlyphToPath(VGFont font,
       vg_set_error(ctx, VG_BAD_HANDLE_ERROR);
       return;
    }
-   pathObj = (struct vg_object*)path;
-   if (pathObj && pathObj->type != VG_OBJECT_PATH) {
-      vg_set_error(ctx, VG_BAD_HANDLE_ERROR);
-      return;
-   }
 
-   f = (struct vg_font*)font;
-   f->glyph_indices[f->num_glyphs] = glyphIndex;
-   ++f->num_glyphs;
+   pathObj = (struct path*) path;
+   f = (struct vg_font*) font;
+
+   font_set_glyph_to_path(f, glyphIndex, pathObj,
+         isHinted, glyphOrigin, escapement);
 }
 
 void vegaSetGlyphToImage(VGFont font,
                          VGuint glyphIndex,
                          VGImage image,
-                         VGfloat glyphOrigin [2],
-                         VGfloat escapement[2])
+                         const VGfloat glyphOrigin[2],
+                         const VGfloat escapement[2])
 {
    struct vg_context *ctx = vg_current_context();
-   struct vg_object *img_obj;
+   struct vg_image *img_obj;
    struct vg_font *f;
 
    if (font == VG_INVALID_HANDLE ||
@@ -131,26 +122,11 @@ void vegaSetGlyphToImage(VGFont font,
       vg_set_error(ctx, VG_BAD_HANDLE_ERROR);
       return;
    }
-   img_obj = (struct vg_object*)image;
-   if (img_obj && img_obj->type != VG_OBJECT_IMAGE) {
-      vg_set_error(ctx, VG_BAD_HANDLE_ERROR);
-      return;
-   }
+
+   img_obj = (struct vg_image*)image;
    f = (struct vg_font*)font;
-   f->glyph_indices[f->num_glyphs] = glyphIndex;
-   ++f->num_glyphs;
-}
 
-static INLINE VGboolean font_contains_glyph(struct vg_font *font,
-                                            VGuint glyph_index)
-{
-   VGint i;
-   for (i = 0; i < font->num_glyphs; ++i) {
-      if (font->glyph_indices[i] == glyph_index) {
-         return VG_TRUE;
-      }
-   }
-   return VG_FALSE;
+   font_set_glyph_to_image(f, glyphIndex, img_obj, glyphOrigin, escapement);
 }
 
 void vegaClearGlyph(VGFont font,
@@ -158,30 +134,15 @@ void vegaClearGlyph(VGFont font,
 {
    struct vg_context *ctx = vg_current_context();
    struct vg_font *f;
-   VGint i;
 
    if (font == VG_INVALID_HANDLE) {
       vg_set_error(ctx, VG_BAD_HANDLE_ERROR);
       return;
    }
-   if (glyphIndex <= 0) {
-      vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR);
-      return;
-   }
-   f = (struct vg_font*)font;
-   if (!font_contains_glyph(f, glyphIndex)) {
-      vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR);
-      return;
-   }
 
-   for (i = 0; i < f->num_glyphs; ++i) {
-      if (f->glyph_indices[i] == glyphIndex) {
-         /*FIXME*/
-         f->glyph_indices[f->num_glyphs] = 0;
-         --f->num_glyphs;
-         return;
-      }
-   }
+   f = (struct vg_font*) font;
+
+   font_clear_glyph(f, glyphIndex);
 }
 
 void vegaDrawGlyph(VGFont font,
@@ -196,31 +157,24 @@ void vegaDrawGlyph(VGFont font,
       vg_set_error(ctx, VG_BAD_HANDLE_ERROR);
       return;
    }
-   if (glyphIndex <= 0) {
-      vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR);
-      return;
-   }
    if (paintModes & (~(VG_STROKE_PATH|VG_FILL_PATH))) {
       vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR);
       return;
    }
    f = (struct vg_font*)font;
-   if (!font_contains_glyph(f, glyphIndex)) {
-      vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR);
-      return;
-   }
+
+   font_draw_glyph(f, glyphIndex, paintModes, allowAutoHinting);
 }
 
 void vegaDrawGlyphs(VGFont font,
                     VGint glyphCount,
-                    VGuint *glyphIndices,
-                    VGfloat *adjustments_x,
-                    VGfloat *adjustments_y,
+                    const VGuint *glyphIndices,
+                    const VGfloat *adjustments_x,
+                    const VGfloat *adjustments_y,
                     VGbitfield paintModes,
                     VGboolean allowAutoHinting)
 {
    struct vg_context *ctx = vg_current_context();
-   VGint i;
    struct vg_font *f;
 
    if (font == VG_INVALID_HANDLE) {
@@ -235,8 +189,8 @@ void vegaDrawGlyphs(VGFont font,
       vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR);
       return;
    }
-   if (!adjustments_x || !is_aligned(adjustments_x) ||
-       !adjustments_y || !is_aligned(adjustments_y)) {
+   if ((adjustments_x && !is_aligned(adjustments_x)) ||
+       (adjustments_y && !is_aligned(adjustments_y))) {
       vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR);
       return;
    }
@@ -246,13 +200,9 @@ void vegaDrawGlyphs(VGFont font,
    }
 
    f = (struct vg_font*)font;
-   for (i = 0; i < glyphCount; ++i) {
-      VGuint glyph_index = glyphIndices[i];
-      if (!font_contains_glyph(f, glyph_index)) {
-         vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR);
-         return;
-      }
-   }
+
+   font_draw_glyphs(f, glyphCount, glyphIndices,
+         adjustments_x, adjustments_y, paintModes, allowAutoHinting);
 }
 
-#endif
+#endif /* OPENVG_VERSION_1_1 */
diff --git a/src/gallium/state_trackers/vega/asm_fill.h b/src/gallium/state_trackers/vega/asm_fill.h
index 27773467fa..77e6a14fe9 100644
--- a/src/gallium/state_trackers/vega/asm_fill.h
+++ b/src/gallium/state_trackers/vega/asm_fill.h
@@ -44,9 +44,31 @@ solid_fill( struct ureg_program *ureg,
             struct ureg_dst *temp,
             struct ureg_src *constant)
 {
-   ureg_MOV(ureg, *out, constant[0]);
+   ureg_MOV(ureg, *out, constant[2]);
 }
 
+/**
+ * Perform frag-coord-to-paint-coord transform.  The transformation is in
+ * CONST[4..6].
+ */
+#define PAINT_TRANSFORM                                                 \
+   ureg_MOV(ureg, ureg_writemask(temp[0], TGSI_WRITEMASK_XY), in[0]);   \
+   ureg_MOV(ureg,                                                       \
+            ureg_writemask(temp[0], TGSI_WRITEMASK_Z),                  \
+            ureg_scalar(constant[3], TGSI_SWIZZLE_Y));                  \
+   ureg_DP3(ureg, temp[1], constant[4], ureg_src(temp[0]));             \
+   ureg_DP3(ureg, temp[2], constant[5], ureg_src(temp[0]));             \
+   ureg_DP3(ureg, temp[3], constant[6], ureg_src(temp[0]));             \
+   ureg_RCP(ureg, temp[3], ureg_src(temp[3]));                          \
+   ureg_MUL(ureg, temp[1], ureg_src(temp[1]), ureg_src(temp[3]));       \
+   ureg_MUL(ureg, temp[2], ureg_src(temp[2]), ureg_src(temp[3]));       \
+   ureg_MOV(ureg,                                                       \
+            ureg_writemask(temp[4], TGSI_WRITEMASK_X),                  \
+            ureg_src(temp[1]));                                         \
+   ureg_MOV(ureg,                                                       \
+            ureg_writemask(temp[4], TGSI_WRITEMASK_Y),                  \
+            ureg_src(temp[2]));
+
 static INLINE void
 linear_grad( struct ureg_program *ureg,
              struct ureg_dst *out,
@@ -55,30 +77,19 @@ linear_grad( struct ureg_program *ureg,
              struct ureg_dst *temp,
              struct ureg_src *constant)
 {
+   PAINT_TRANSFORM
 
-   ureg_MOV(ureg,
-            ureg_writemask(temp[0], TGSI_WRITEMASK_XY),
-            in[0]);
-   ureg_MOV(ureg,
-            ureg_writemask(temp[0], TGSI_WRITEMASK_Z),
-            ureg_scalar(constant[1], TGSI_SWIZZLE_Y));
-   ureg_DP3(ureg, temp[1], constant[2], ureg_src(temp[0]));
-   ureg_DP3(ureg, temp[2], constant[3], ureg_src(temp[0]));
-   ureg_DP3(ureg, temp[3], constant[4], ureg_src(temp[0]));
-   ureg_RCP(ureg, temp[3], ureg_src(temp[3]));
-   ureg_MUL(ureg, temp[1], ureg_src(temp[1]), ureg_src(temp[3]));
-   ureg_MUL(ureg, temp[2], ureg_src(temp[2]), ureg_src(temp[3]));
-   ureg_MOV(ureg, ureg_writemask(temp[4], TGSI_WRITEMASK_X), ureg_src(temp[1]));
-   ureg_MOV(ureg, ureg_writemask(temp[4], TGSI_WRITEMASK_Y), ureg_src(temp[2]));
+   /* grad = DP2((x, y), CONST[2].xy) * CONST[2].z */
    ureg_MUL(ureg, temp[0],
-            ureg_scalar(constant[0], TGSI_SWIZZLE_Y),
+            ureg_scalar(constant[2], TGSI_SWIZZLE_Y),
             ureg_scalar(ureg_src(temp[4]), TGSI_SWIZZLE_Y));
    ureg_MAD(ureg, temp[1],
-            ureg_scalar(constant[0], TGSI_SWIZZLE_X),
+            ureg_scalar(constant[2], TGSI_SWIZZLE_X),
             ureg_scalar(ureg_src(temp[4]), TGSI_SWIZZLE_X),
             ureg_src(temp[0]));
    ureg_MUL(ureg, temp[2], ureg_src(temp[1]),
-            ureg_scalar(constant[0], TGSI_SWIZZLE_Z));
+            ureg_scalar(constant[2], TGSI_SWIZZLE_Z));
+
    ureg_TEX(ureg, *out, TGSI_TEXTURE_1D, ureg_src(temp[2]), sampler[0]);
 }
 
@@ -90,52 +101,32 @@ radial_grad( struct ureg_program *ureg,
              struct ureg_dst *temp,
              struct ureg_src *constant)
 {
-
-   ureg_MOV(ureg, ureg_writemask(temp[0], TGSI_WRITEMASK_XY), in[0]);
-   ureg_MOV(ureg,
-            ureg_writemask(temp[0], TGSI_WRITEMASK_Z),
-            ureg_scalar(constant[1], TGSI_SWIZZLE_Y));
-   ureg_DP3(ureg, temp[1], constant[2], ureg_src(temp[0]));
-   ureg_DP3(ureg, temp[2], constant[3], ureg_src(temp[0]));
-   ureg_DP3(ureg, temp[3], constant[4], ureg_src(temp[0]));
+   PAINT_TRANSFORM
+
+   /*
+    * Calculate (sqrt(B^2 + AC) - B) / A, where
+    *
+    *   A is CONST[2].z,
+    *   B is DP2((x, y), CONST[2].xy), and
+    *   C is DP2((x, y), (x, y)).
+    */
+
+   /* B and C */
+   ureg_DP2(ureg, temp[0], ureg_src(temp[4]), constant[2]);
+   ureg_DP2(ureg, temp[1], ureg_src(temp[4]), ureg_src(temp[4]));
+
+   /* the square root */
+   ureg_MUL(ureg, temp[2], ureg_src(temp[0]), ureg_src(temp[0]));
+   ureg_MAD(ureg, temp[3], ureg_src(temp[1]),
+         ureg_scalar(constant[2], TGSI_SWIZZLE_Z), ureg_src(temp[2]));
+   ureg_RSQ(ureg, temp[3], ureg_src(temp[3]));
    ureg_RCP(ureg, temp[3], ureg_src(temp[3]));
-   ureg_MUL(ureg, temp[1], ureg_src(temp[1]), ureg_src(temp[3]));
-   ureg_MUL(ureg, temp[2], ureg_src(temp[2]), ureg_src(temp[3]));
-   ureg_MOV(ureg, ureg_writemask(temp[5], TGSI_WRITEMASK_X), ureg_src(temp[1]));
-   ureg_MOV(ureg, ureg_writemask(temp[5], TGSI_WRITEMASK_Y), ureg_src(temp[2]));
-   ureg_MUL(ureg, temp[0], ureg_scalar(constant[0], TGSI_SWIZZLE_Y),
-            ureg_scalar(ureg_src(temp[5]), TGSI_SWIZZLE_Y));
-   ureg_MAD(ureg, temp[1],
-            ureg_scalar(constant[0], TGSI_SWIZZLE_X),
-            ureg_scalar(ureg_src(temp[5]), TGSI_SWIZZLE_X), ureg_src(temp[0]));
-   ureg_ADD(ureg, temp[1], ureg_src(temp[1]), ureg_src(temp[1]));
-   ureg_MUL(ureg, temp[3],
-            ureg_scalar(ureg_src(temp[5]), TGSI_SWIZZLE_Y),
-            ureg_scalar(ureg_src(temp[5]), TGSI_SWIZZLE_Y));
-   ureg_MAD(ureg, temp[4],
-            ureg_scalar(ureg_src(temp[5]), TGSI_SWIZZLE_X),
-            ureg_scalar(ureg_src(temp[5]), TGSI_SWIZZLE_X),
-            ureg_src(temp[3]));
-   ureg_MOV(ureg, temp[4], ureg_negate(ureg_src(temp[4])));
-   ureg_MUL(ureg, temp[2],
-            ureg_scalar(constant[0], TGSI_SWIZZLE_Z),
-            ureg_src(temp[4]));
-   ureg_MUL(ureg, temp[0],
-            ureg_scalar(constant[1], TGSI_SWIZZLE_W),
-            ureg_src(temp[2]));
-   ureg_MUL(ureg, temp[3], ureg_src(temp[1]), ureg_src(temp[1]));
-
-   ureg_SUB(ureg, temp[2], ureg_src(temp[3]), ureg_src(temp[0]));
-   ureg_RSQ(ureg, temp[2], ureg_abs(ureg_src(temp[2])));
-   ureg_RCP(ureg, temp[2], ureg_src(temp[2]));
-   ureg_SUB(ureg, temp[1], ureg_src(temp[2]), ureg_src(temp[1]));
-   ureg_ADD(ureg, temp[0],
-            ureg_scalar(constant[0], TGSI_SWIZZLE_Z),
-            ureg_scalar(constant[0], TGSI_SWIZZLE_Z));
-   ureg_RCP(ureg, temp[0], ureg_src(temp[0]));
-   ureg_MUL(ureg, temp[2], ureg_src(temp[1]), ureg_src(temp[0]));
-   ureg_TEX(ureg, *out, TGSI_TEXTURE_1D, ureg_src(temp[2]), sampler[0]);
 
+   ureg_SUB(ureg, temp[3], ureg_src(temp[3]), ureg_src(temp[0]));
+   ureg_RCP(ureg, temp[0], ureg_scalar(constant[2], TGSI_SWIZZLE_Z));
+   ureg_MUL(ureg, temp[0], ureg_src(temp[0]), ureg_src(temp[3]));
+
+   ureg_TEX(ureg, *out, TGSI_TEXTURE_1D, ureg_src(temp[0]), sampler[0]);
 }
 
 
@@ -147,22 +138,11 @@ pattern( struct ureg_program *ureg,
          struct ureg_dst     *temp,
          struct ureg_src     *constant)
 {
-   ureg_MOV(ureg,
-            ureg_writemask(temp[0], TGSI_WRITEMASK_XY),
-            in[0]);
-   ureg_MOV(ureg,
-            ureg_writemask(temp[0], TGSI_WRITEMASK_Z),
-            ureg_scalar(constant[1], TGSI_SWIZZLE_Y));
-   ureg_DP3(ureg, temp[1], constant[2], ureg_src(temp[0]));
-   ureg_DP3(ureg, temp[2], constant[3], ureg_src(temp[0]));
-   ureg_DP3(ureg, temp[3], constant[4], ureg_src(temp[0]));
-   ureg_RCP(ureg, temp[3], ureg_src(temp[3]));
-   ureg_MUL(ureg, temp[1], ureg_src(temp[1]), ureg_src(temp[3]));
-   ureg_MUL(ureg, temp[2], ureg_src(temp[2]), ureg_src(temp[3]));
-   ureg_MOV(ureg, ureg_writemask(temp[4], TGSI_WRITEMASK_X), ureg_src(temp[1]));
-   ureg_MOV(ureg, ureg_writemask(temp[4], TGSI_WRITEMASK_Y), ureg_src(temp[2]));
+   PAINT_TRANSFORM
+
+   /* (s, t) = (x / tex_width, y / tex_height) */
    ureg_RCP(ureg, temp[0],
-            ureg_swizzle(constant[1],
+            ureg_swizzle(constant[3],
                          TGSI_SWIZZLE_Z,
                          TGSI_SWIZZLE_W,
                          TGSI_SWIZZLE_Z,
@@ -176,22 +156,21 @@ pattern( struct ureg_program *ureg,
             ureg_writemask(temp[1], TGSI_WRITEMASK_Y),
             ureg_src(temp[1]),
             ureg_src(temp[0]));
+
    ureg_TEX(ureg, *out, TGSI_TEXTURE_2D, ureg_src(temp[1]), sampler[0]);
 }
 
 static INLINE void
-mask( struct ureg_program *ureg,
-      struct ureg_dst *out,
-      struct ureg_src *in,
-      struct ureg_src *sampler,
-      struct ureg_dst *temp,
-      struct ureg_src *constant)
+paint_degenerate( struct ureg_program *ureg,
+                  struct ureg_dst *out,
+                  struct ureg_src *in,
+                  struct ureg_src *sampler,
+                  struct ureg_dst *temp,
+                  struct ureg_src *constant)
 {
-   ureg_TEX(ureg, temp[1], TGSI_TEXTURE_2D, in[0], sampler[1]);
-   ureg_MUL(ureg, ureg_writemask(temp[0], TGSI_WRITEMASK_W),
-            ureg_scalar(ureg_src(temp[0]), TGSI_SWIZZLE_W),
-            ureg_scalar(ureg_src(temp[1]), TGSI_SWIZZLE_W));
-   ureg_MOV(ureg, *out, ureg_src(temp[0]));
+   /* CONST[3].y is 1.0f */
+   ureg_MOV(ureg, temp[1], ureg_scalar(constant[3], TGSI_SWIZZLE_Y));
+   ureg_TEX(ureg, *out, TGSI_TEXTURE_1D, ureg_src(temp[1]), sampler[0]);
 }
 
 static INLINE void
@@ -202,7 +181,9 @@ image_normal( struct ureg_program *ureg,
               struct ureg_dst *temp,
               struct ureg_src *constant)
 {
-   ureg_TEX(ureg, *out, TGSI_TEXTURE_2D, in[1], sampler[3]);
+   /* store and pass image color in TEMP[1] */
+   ureg_TEX(ureg, temp[1], TGSI_TEXTURE_2D, in[1], sampler[3]);
+   ureg_MOV(ureg, *out, ureg_src(temp[1]));
 }
 
 
@@ -214,6 +195,7 @@ image_multiply( struct ureg_program *ureg,
                 struct ureg_dst *temp,
                 struct ureg_src *constant)
 {
+   /* store and pass image color in TEMP[1] */
    ureg_TEX(ureg, temp[1], TGSI_TEXTURE_2D, in[1], sampler[3]);
    ureg_MUL(ureg, *out, ureg_src(temp[0]), ureg_src(temp[1]));
 }
@@ -227,43 +209,279 @@ image_stencil( struct ureg_program *ureg,
                struct ureg_dst *temp,
                struct ureg_src *constant)
 {
+   /* store and pass image color in TEMP[1] */
    ureg_TEX(ureg, temp[1], TGSI_TEXTURE_2D, in[1], sampler[3]);
-   ureg_MUL(ureg, *out, ureg_src(temp[0]), ureg_src(temp[1]));
+   ureg_MOV(ureg, *out, ureg_src(temp[0]));
+}
+
+static INLINE void
+color_transform( struct ureg_program *ureg,
+                 struct ureg_dst *out,
+                 struct ureg_src *in,
+                 struct ureg_src *sampler,
+                 struct ureg_dst *temp,
+                 struct ureg_src *constant)
+{
+   /* note that TEMP[1] may already be used for image color */
+
+   ureg_MAD(ureg, temp[2], ureg_src(temp[0]), constant[0], constant[1]);
+   /* clamp to [0.0f, 1.0f] */
+   ureg_CLAMP(ureg, temp[2],
+              ureg_src(temp[2]),
+              ureg_scalar(constant[3], TGSI_SWIZZLE_X),
+              ureg_scalar(constant[3], TGSI_SWIZZLE_Y));
+   ureg_MOV(ureg, *out, ureg_src(temp[2]));
+}
+
+static INLINE void
+alpha_normal( struct ureg_program *ureg,
+              struct ureg_dst *out,
+              struct ureg_src *in,
+              struct ureg_src *sampler,
+              struct ureg_dst *temp,
+              struct ureg_src *constant)
+{
+   /* save per-channel alpha in TEMP[1] */
+   ureg_MOV(ureg, temp[1], ureg_scalar(ureg_src(temp[0]), TGSI_SWIZZLE_W));
+
+   ureg_MOV(ureg, *out, ureg_src(temp[0]));
 }
 
-#define EXTENDED_BLENDER_OVER_FUNC                                      \
-   ureg_SUB(ureg, temp[3],                                              \
-            ureg_scalar(constant[1], TGSI_SWIZZLE_Y),                   \
-            ureg_scalar(ureg_src(temp[1]), TGSI_SWIZZLE_W));            \
-   ureg_SUB(ureg, temp[3],                                              \
-            ureg_scalar(constant[1], TGSI_SWIZZLE_Y),                   \
-            ureg_scalar(ureg_src(temp[0]), TGSI_SWIZZLE_W));            \
-   ureg_MUL(ureg, temp[3], ureg_src(temp[0]), ureg_src(temp[3]));       \
-   ureg_MUL(ureg, temp[4], ureg_src(temp[1]), ureg_src(temp[4]));       \
-   ureg_ADD(ureg, temp[3], ureg_src(temp[3]), ureg_src(temp[4]));
+static INLINE void
+alpha_per_channel( struct ureg_program *ureg,
+                   struct ureg_dst *out,
+                   struct ureg_src *in,
+                   struct ureg_src *sampler,
+                   struct ureg_dst *temp,
+                   struct ureg_src *constant)
+{
+   /* save per-channel alpha in TEMP[1] */
+   ureg_MUL(ureg,
+            ureg_writemask(temp[1], TGSI_WRITEMASK_W),
+            ureg_src(temp[0]),
+            ureg_src(temp[1]));
+   ureg_MUL(ureg,
+            ureg_writemask(temp[1], TGSI_WRITEMASK_XYZ),
+            ureg_src(temp[1]),
+            ureg_scalar(ureg_src(temp[1]), TGSI_SWIZZLE_W));
 
+   /* update alpha */
+   ureg_MOV(ureg,
+            ureg_writemask(temp[0], TGSI_WRITEMASK_W),
+            ureg_src(temp[1]));
+   ureg_MOV(ureg, *out, ureg_src(temp[0]));
+}
 
+/**
+ * Premultiply src and dst.
+ */
 static INLINE void
-blend_multiply( struct ureg_program *ureg,
+blend_premultiply( struct ureg_program *ureg,
+                   struct ureg_src src,
+                   struct ureg_src src_channel_alpha,
+                   struct ureg_src dst)
+{
+   /* premultiply src */
+   ureg_MUL(ureg,
+            ureg_writemask(ureg_dst(src), TGSI_WRITEMASK_XYZ),
+            src,
+            src_channel_alpha);
+   /* premultiply dst */
+   ureg_MUL(ureg,
+            ureg_writemask(ureg_dst(dst), TGSI_WRITEMASK_XYZ),
+            dst,
+            ureg_scalar(dst, TGSI_SWIZZLE_W));
+}
+
+/**
+ * Unpremultiply src.
+ */
+static INLINE void
+blend_unpremultiply( struct ureg_program *ureg,
+                     struct ureg_src src,
+                     struct ureg_src one,
+                     struct ureg_dst temp[1])
+{
+   /* replace 0.0f by 1.0f before calculating reciprocal */
+   ureg_CMP(ureg,
+            temp[0],
+            ureg_negate(ureg_scalar(src, TGSI_SWIZZLE_W)),
+            ureg_scalar(src, TGSI_SWIZZLE_W),
+            one);
+   ureg_RCP(ureg, temp[0], ureg_src(temp[0]));
+
+   ureg_MUL(ureg,
+            ureg_writemask(ureg_dst(src), TGSI_WRITEMASK_XYZ),
+            src,
+            ureg_src(temp[0]));
+}
+
+/**
+ * Emit instructions for the specified blend mode.  Colors will be
+ * unpremultiplied.  Two temporary registers are required.
+ *
+ * The output is written back to src.
+ */
+static INLINE void
+blend_generic(struct ureg_program *ureg,
+              VGBlendMode mode,
+              struct ureg_src src,
+              struct ureg_src src_channel_alpha,
+              struct ureg_src dst,
+              struct ureg_src one,
+              struct ureg_dst temp[2])
+{
+   struct ureg_dst out;
+
+   blend_premultiply(ureg, src, src_channel_alpha, dst);
+
+   /* blend in-place */
+   out = ureg_dst(src);
+
+   switch (mode) {
+   case VG_BLEND_SRC:
+      ureg_MOV(ureg, out, src);
+      break;
+   case VG_BLEND_SRC_OVER:
+      /* RGBA_out = RGBA_src + (1 - A_src) * RGBA_dst */
+      ureg_SUB(ureg, temp[0], one, src_channel_alpha);
+      ureg_MAD(ureg, out, ureg_src(temp[0]), dst, src);
+      break;
+   case VG_BLEND_DST_OVER:
+      /* RGBA_out = RGBA_dst + (1 - A_dst) * RGBA_src */
+      ureg_SUB(ureg, temp[0], one, ureg_scalar(dst, TGSI_SWIZZLE_W));
+      ureg_MAD(ureg, out, ureg_src(temp[0]), src, dst);
+      break;
+   case VG_BLEND_SRC_IN:
+      ureg_MUL(ureg, out, src, ureg_scalar(dst, TGSI_SWIZZLE_W));
+      break;
+   case VG_BLEND_DST_IN:
+      ureg_MUL(ureg, out, dst, src_channel_alpha);
+      break;
+   case VG_BLEND_MULTIPLY:
+      /*
+       * RGB_out = (1 - A_dst) * RGB_src + (1 - A_src) * RGB_dst +
+       *           RGB_src * RGB_dst
+       */
+      ureg_MAD(ureg, temp[0],
+            ureg_scalar(dst, TGSI_SWIZZLE_W), ureg_negate(src), src);
+      ureg_MAD(ureg, temp[1],
+            src_channel_alpha, ureg_negate(dst), dst);
+      ureg_MAD(ureg, temp[0], src, dst, ureg_src(temp[0]));
+      ureg_ADD(ureg, out, ureg_src(temp[0]), ureg_src(temp[1]));
+      /* alpha is src over */
+      ureg_ADD(ureg, ureg_writemask(out, TGSI_WRITEMASK_W),
+            src, ureg_src(temp[1]));
+      break;
+   case VG_BLEND_SCREEN:
+      /* RGBA_out = RGBA_src + (1 - RGBA_src) * RGBA_dst */
+      ureg_SUB(ureg, temp[0], one, src);
+      ureg_MAD(ureg, out, ureg_src(temp[0]), dst, src);
+      break;
+   case VG_BLEND_DARKEN:
+   case VG_BLEND_LIGHTEN:
+      /* src over */
+      ureg_SUB(ureg, temp[0], one, src_channel_alpha);
+      ureg_MAD(ureg, temp[0], ureg_src(temp[0]), dst, src);
+      /* dst over */
+      ureg_SUB(ureg, temp[1], one, ureg_scalar(dst, TGSI_SWIZZLE_W));
+      ureg_MAD(ureg, temp[1], ureg_src(temp[1]), src, dst);
+      /* take min/max for colors */
+      if (mode == VG_BLEND_DARKEN) {
+         ureg_MIN(ureg, ureg_writemask(out, TGSI_WRITEMASK_XYZ),
+               ureg_src(temp[0]), ureg_src(temp[1]));
+      }
+      else {
+         ureg_MAX(ureg, ureg_writemask(out, TGSI_WRITEMASK_XYZ),
+               ureg_src(temp[0]), ureg_src(temp[1]));
+      }
+      break;
+   case VG_BLEND_ADDITIVE:
+      /* RGBA_out = RGBA_src + RGBA_dst */
+      ureg_ADD(ureg, temp[0], src, dst);
+      ureg_MIN(ureg, out, ureg_src(temp[0]), one);
+      break;
+   default:
+      assert(0);
+      break;
+   }
+
+   blend_unpremultiply(ureg, src, one, temp);
+}
+
+#define BLEND_GENERIC(mode) \
+   do { \
+      ureg_TEX(ureg, temp[2], TGSI_TEXTURE_2D, in[0], sampler[2]);         \
+      blend_generic(ureg, (mode), ureg_src(temp[0]), ureg_src(temp[1]),    \
+                    ureg_src(temp[2]),                                     \
+                    ureg_scalar(constant[3], TGSI_SWIZZLE_Y), temp + 3);   \
+      ureg_MOV(ureg, *out, ureg_src(temp[0]));                             \
+   } while (0)
+
+static INLINE void
+blend_src( struct ureg_program *ureg,
+           struct ureg_dst *out,
+           struct ureg_src *in,
+           struct ureg_src *sampler,
+           struct ureg_dst *temp,
+           struct ureg_src *constant)
+{
+   BLEND_GENERIC(VG_BLEND_SRC);
+}
+
+static INLINE void
+blend_src_over( struct ureg_program *ureg,
                 struct ureg_dst *out,
                 struct ureg_src *in,
                 struct ureg_src *sampler,
                 struct ureg_dst *temp,
                 struct ureg_src *constant)
 {
-   ureg_TEX(ureg, temp[1], TGSI_TEXTURE_2D, in[0], sampler[2]);
-   EXTENDED_BLENDER_OVER_FUNC
-   ureg_MUL(ureg, temp[4], ureg_src(temp[0]), ureg_src(temp[1]));
-   ureg_ADD(ureg, temp[1], ureg_src(temp[4]), ureg_src(temp[3]));
+   BLEND_GENERIC(VG_BLEND_SRC_OVER);
+}
 
-   ureg_MUL(ureg, temp[2], ureg_scalar(ureg_src(temp[0]), TGSI_SWIZZLE_W),
-            ureg_scalar(ureg_src(temp[1]), TGSI_SWIZZLE_W));
-   ureg_ADD(ureg, temp[3], ureg_scalar(ureg_src(temp[0]), TGSI_SWIZZLE_W),
-            ureg_scalar(ureg_src(temp[1]), TGSI_SWIZZLE_W));
-   ureg_SUB(ureg, ureg_writemask(temp[1], TGSI_WRITEMASK_W),
-            ureg_src(temp[3]), ureg_src(temp[2]));
+static INLINE void
+blend_dst_over( struct ureg_program *ureg,
+                struct ureg_dst *out,
+                struct ureg_src *in,
+                struct ureg_src *sampler,
+                struct ureg_dst *temp,
+                struct ureg_src *constant)
+{
+   BLEND_GENERIC(VG_BLEND_DST_OVER);
+}
 
-   ureg_MOV(ureg, *out, ureg_src(temp[1]));
+static INLINE void
+blend_src_in( struct ureg_program *ureg,
+              struct ureg_dst *out,
+              struct ureg_src *in,
+              struct ureg_src *sampler,
+              struct ureg_dst *temp,
+              struct ureg_src *constant)
+{
+   BLEND_GENERIC(VG_BLEND_SRC_IN);
+}
+
+static INLINE void
+blend_dst_in( struct ureg_program *ureg,
+              struct ureg_dst *out,
+              struct ureg_src *in,
+              struct ureg_src *sampler,
+              struct ureg_dst *temp,
+              struct ureg_src *constant)
+{
+   BLEND_GENERIC(VG_BLEND_DST_IN);
+}
+
+static INLINE void
+blend_multiply( struct ureg_program *ureg,
+                struct ureg_dst *out,
+                struct ureg_src *in,
+                struct ureg_src *sampler,
+                struct ureg_dst *temp,
+                struct ureg_src *constant)
+{
+   BLEND_GENERIC(VG_BLEND_MULTIPLY);
 }
 
 static INLINE void
@@ -274,10 +492,7 @@ blend_screen( struct ureg_program *ureg,
               struct ureg_dst     *temp,
               struct ureg_src     *constant)
 {
-   ureg_TEX(ureg, temp[1], TGSI_TEXTURE_2D, in[0], sampler[2]);
-   ureg_ADD(ureg, temp[3], ureg_src(temp[0]), ureg_src(temp[1]));
-   ureg_MUL(ureg, temp[2], ureg_src(temp[0]), ureg_src(temp[1]));
-   ureg_SUB(ureg, *out, ureg_src(temp[3]), ureg_src(temp[2]));
+   BLEND_GENERIC(VG_BLEND_SCREEN);
 }
 
 static INLINE void
@@ -288,23 +503,7 @@ blend_darken( struct ureg_program *ureg,
               struct ureg_dst     *temp,
               struct ureg_src     *constant)
 {
-   ureg_TEX(ureg, temp[1], TGSI_TEXTURE_2D, in[0], sampler[2]);
-   EXTENDED_BLENDER_OVER_FUNC
-   ureg_MUL(ureg, temp[4], ureg_src(temp[0]),
-            ureg_scalar(ureg_src(temp[1]), TGSI_SWIZZLE_W));
-   ureg_MUL(ureg, temp[5], ureg_src(temp[1]),
-            ureg_scalar(ureg_src(temp[0]), TGSI_SWIZZLE_W));
-   ureg_MIN(ureg, temp[4], ureg_src(temp[4]), ureg_src(temp[5]));
-   ureg_ADD(ureg, temp[1], ureg_src(temp[3]), ureg_src(temp[4]));
-
-   ureg_MUL(ureg, temp[2], ureg_scalar(ureg_src(temp[0]), TGSI_SWIZZLE_W),
-            ureg_scalar(ureg_src(temp[1]), TGSI_SWIZZLE_W));
-   ureg_ADD(ureg, temp[3], ureg_scalar(ureg_src(temp[0]), TGSI_SWIZZLE_W),
-            ureg_scalar(ureg_src(temp[1]), TGSI_SWIZZLE_W));
-   ureg_SUB(ureg, ureg_writemask(temp[1], TGSI_WRITEMASK_W),
-            ureg_src(temp[3]), ureg_src(temp[2]));
-
-   ureg_MOV(ureg, *out, ureg_src(temp[1]));
+   BLEND_GENERIC(VG_BLEND_DARKEN);
 }
 
 static INLINE void
@@ -315,23 +514,33 @@ blend_lighten( struct ureg_program *ureg,
                struct ureg_dst *temp,
                struct ureg_src     *constant)
 {
-   ureg_TEX(ureg, temp[1], TGSI_TEXTURE_2D, in[0], sampler[2]);
-   EXTENDED_BLENDER_OVER_FUNC
-   ureg_MUL(ureg, temp[4], ureg_src(temp[0]),
-            ureg_scalar(ureg_src(temp[1]), TGSI_SWIZZLE_W));
-   ureg_MUL(ureg, temp[5], ureg_src(temp[1]),
-            ureg_scalar(ureg_src(temp[0]), TGSI_SWIZZLE_W));
-   ureg_MAX(ureg, temp[4], ureg_src(temp[4]), ureg_src(temp[5]));
-   ureg_ADD(ureg, temp[1], ureg_src(temp[3]), ureg_src(temp[4]));
+   BLEND_GENERIC(VG_BLEND_LIGHTEN);
+}
 
-   ureg_MUL(ureg, temp[2], ureg_scalar(ureg_src(temp[0]), TGSI_SWIZZLE_W),
-            ureg_scalar(ureg_src(temp[1]), TGSI_SWIZZLE_W));
-   ureg_ADD(ureg, temp[3], ureg_scalar(ureg_src(temp[0]), TGSI_SWIZZLE_W),
-            ureg_scalar(ureg_src(temp[1]), TGSI_SWIZZLE_W));
-   ureg_SUB(ureg, ureg_writemask(temp[1], TGSI_WRITEMASK_W),
-            ureg_src(temp[3]), ureg_src(temp[2]));
+static INLINE void
+blend_additive( struct ureg_program *ureg,
+                struct ureg_dst *out,
+                struct ureg_src *in,
+                struct ureg_src *sampler,
+                struct ureg_dst *temp,
+                struct ureg_src *constant)
+{
+   BLEND_GENERIC(VG_BLEND_ADDITIVE);
+}
 
-   ureg_MOV(ureg, *out, ureg_src(temp[1]));
+static INLINE void
+mask( struct ureg_program *ureg,
+      struct ureg_dst *out,
+      struct ureg_src *in,
+      struct ureg_src *sampler,
+      struct ureg_dst *temp,
+      struct ureg_src *constant)
+{
+   ureg_TEX(ureg, temp[1], TGSI_TEXTURE_2D, in[0], sampler[1]);
+   ureg_MUL(ureg, ureg_writemask(temp[0], TGSI_WRITEMASK_W),
+            ureg_scalar(ureg_src(temp[0]), TGSI_SWIZZLE_W),
+            ureg_scalar(ureg_src(temp[1]), TGSI_SWIZZLE_W));
+   ureg_MOV(ureg, *out, ureg_src(temp[0]));
 }
 
 static INLINE void
@@ -369,11 +578,11 @@ color_bw( struct ureg_program *ureg,
                 struct ureg_src *constant)
 {
    ureg_ADD(ureg, temp[1],
-            ureg_scalar(constant[1], TGSI_SWIZZLE_Y),
-            ureg_scalar(constant[1], TGSI_SWIZZLE_Y));
+            ureg_scalar(constant[3], TGSI_SWIZZLE_Y),
+            ureg_scalar(constant[3], TGSI_SWIZZLE_Y));
    ureg_RCP(ureg, temp[2], ureg_src(temp[1]));
    ureg_ADD(ureg, temp[1],
-            ureg_scalar(constant[1], TGSI_SWIZZLE_Y),
+            ureg_scalar(constant[3], TGSI_SWIZZLE_Y),
             ureg_src(temp[2]));
    ureg_ADD(ureg, ureg_writemask(temp[2], TGSI_WRITEMASK_X),
             ureg_scalar(ureg_src(temp[0]), TGSI_SWIZZLE_X),
@@ -410,46 +619,75 @@ struct shader_asm_info {
 };
 
 
-static const struct shader_asm_info shaders_asm[] = {
-   /* fills */
+/* paint types */
+static const struct shader_asm_info shaders_paint_asm[] = {
    {VEGA_SOLID_FILL_SHADER, solid_fill,
-    VG_FALSE, 0, 1, 0, 0, 0, 0},
+    VG_FALSE, 2, 1, 0, 0, 0, 0},
    {VEGA_LINEAR_GRADIENT_SHADER, linear_grad,
-    VG_TRUE,  0, 5, 0, 1, 0, 5},
+    VG_TRUE,  2, 5, 0, 1, 0, 5},
    {VEGA_RADIAL_GRADIENT_SHADER, radial_grad,
-    VG_TRUE,  0, 5, 0, 1, 0, 6},
+    VG_TRUE,  2, 5, 0, 1, 0, 5},
    {VEGA_PATTERN_SHADER, pattern,
-    VG_TRUE,  1, 4, 0, 1, 0, 5},
+    VG_TRUE,  3, 4, 0, 1, 0, 5},
+   {VEGA_PAINT_DEGENERATE_SHADER, paint_degenerate,
+    VG_FALSE,  3, 1, 0, 1, 0, 2}
+};
 
-   /* image draw modes */
+/* image draw modes */
+static const struct shader_asm_info shaders_image_asm[] = {
    {VEGA_IMAGE_NORMAL_SHADER, image_normal,
-    VG_TRUE,  0, 0, 3, 1, 0, 0},
+    VG_TRUE,  0, 0, 3, 1, 0, 2},
    {VEGA_IMAGE_MULTIPLY_SHADER, image_multiply,
     VG_TRUE,  0, 0, 3, 1, 0, 2},
    {VEGA_IMAGE_STENCIL_SHADER, image_stencil,
-    VG_TRUE,  0, 0, 3, 1, 0, 2},
+    VG_TRUE,  0, 0, 3, 1, 0, 2}
+};
 
+static const struct shader_asm_info shaders_color_transform_asm[] = {
+   {VEGA_COLOR_TRANSFORM_SHADER, color_transform,
+    VG_FALSE, 0, 4, 0, 0, 0, 3}
+};
+
+static const struct shader_asm_info shaders_alpha_asm[] = {
+   {VEGA_ALPHA_NORMAL_SHADER, alpha_normal,
+    VG_FALSE, 0, 0, 0, 0, 0, 2},
+   {VEGA_ALPHA_PER_CHANNEL_SHADER, alpha_per_channel,
+    VG_FALSE, 0, 0, 0, 0, 0, 2}
+};
+
+/* extra blend modes */
+static const struct shader_asm_info shaders_blend_asm[] = {
+#define BLEND_ASM_INFO(id, func) { (id), (func), VG_TRUE, 3, 1, 2, 1, 0, 5 }
+   BLEND_ASM_INFO(VEGA_BLEND_SRC_SHADER, blend_src),
+   BLEND_ASM_INFO(VEGA_BLEND_SRC_OVER_SHADER, blend_src_over),
+   BLEND_ASM_INFO(VEGA_BLEND_DST_OVER_SHADER, blend_dst_over),
+   BLEND_ASM_INFO(VEGA_BLEND_SRC_IN_SHADER, blend_src_in),
+   BLEND_ASM_INFO(VEGA_BLEND_DST_IN_SHADER, blend_dst_in),
+   BLEND_ASM_INFO(VEGA_BLEND_MULTIPLY_SHADER, blend_multiply),
+   BLEND_ASM_INFO(VEGA_BLEND_SCREEN_SHADER, blend_screen),
+   BLEND_ASM_INFO(VEGA_BLEND_DARKEN_SHADER, blend_darken),
+   BLEND_ASM_INFO(VEGA_BLEND_LIGHTEN_SHADER, blend_lighten),
+   BLEND_ASM_INFO(VEGA_BLEND_ADDITIVE_SHADER, blend_additive)
+#undef BLEND_ASM_INFO
+};
+
+static const struct shader_asm_info shaders_mask_asm[] = {
    {VEGA_MASK_SHADER, mask,
-    VG_TRUE,  0, 0, 1, 1, 0, 2},
-
-   /* extra blend modes */
-   {VEGA_BLEND_MULTIPLY_SHADER, blend_multiply,
-    VG_TRUE,  1, 1, 2, 1, 0, 5},
-   {VEGA_BLEND_SCREEN_SHADER, blend_screen,
-    VG_TRUE,  0, 0, 2, 1, 0, 4},
-   {VEGA_BLEND_DARKEN_SHADER, blend_darken,
-    VG_TRUE,  1, 1, 2, 1, 0, 6},
-   {VEGA_BLEND_LIGHTEN_SHADER, blend_lighten,
-    VG_TRUE,  1, 1, 2, 1, 0, 6},
-
-   /* premultiply */
+    VG_TRUE,  0, 0, 1, 1, 0, 2}
+};
+
+/* premultiply */
+static const struct shader_asm_info shaders_premultiply_asm[] = {
    {VEGA_PREMULTIPLY_SHADER, premultiply,
     VG_FALSE,  0, 0, 0, 0, 0, 1},
    {VEGA_UNPREMULTIPLY_SHADER, unpremultiply,
     VG_FALSE,  0, 0, 0, 0, 0, 1},
+};
 
-   /* color transform to black and white */
+/* color transform to black and white */
+static const struct shader_asm_info shaders_bw_asm[] = {
    {VEGA_BW_SHADER, color_bw,
-    VG_FALSE,  1, 1, 0, 0, 0, 3},
+    VG_FALSE,  3, 1, 0, 0, 0, 3},
 };
+
 #endif
diff --git a/src/gallium/state_trackers/vega/asm_util.h b/src/gallium/state_trackers/vega/asm_util.h
index 903bfc88a4..ae1842a62c 100644
--- a/src/gallium/state_trackers/vega/asm_util.h
+++ b/src/gallium/state_trackers/vega/asm_util.h
@@ -27,16 +27,6 @@
 #ifndef ASM_UTIL_H
 #define ASM_UTIL_H
 
-
-static const char pass_through_depth_asm[] =
-   "FRAG\n"
-   "DCL IN[0], POSITION, LINEAR\n"
-   "DCL OUT[0].z, POSITION, CONSTANT\n"
-   "0: MOV OUT[0].z, IN[0].zzzz\n"
-   "1: END\n";
-
-
-
 /* μnew = μmask */
 static const char set_mask_asm[] =
    "FRAG\n"
@@ -92,45 +82,4 @@ static const char subtract_mask_asm[] =
    "3: MUL OUT[0], TEMP[2].wwww, TEMP[0].wwww\n"
    "4: END\n";
 
-
-static const char vs_plain_asm[] =
-   "VERT\n"
-   "DCL IN[0]\n"
-   "DCL OUT[0], POSITION\n"
-   "DCL TEMP[0]\n"
-   "DCL CONST[0..1]\n"
-   "0: MUL TEMP[0], IN[0], CONST[0]\n"
-   "1: ADD TEMP[0], TEMP[0], CONST[1]\n"
-   "2: MOV OUT[0], TEMP[0]\n"
-   "3: END\n";
-
-static const char vs_clear_asm[] =
-   "VERT\n"
-   "DCL IN[0]\n"
-   "DCL IN[1]\n"
-   "DCL OUT[0], POSITION\n"
-   "DCL OUT[1], COLOR\n"
-   "DCL TEMP[0]\n"
-   "DCL CONST[0..1]\n"
-   "0: MUL TEMP[0], IN[0], CONST[0]\n"
-   "1: ADD TEMP[0], TEMP[0], CONST[1]\n"
-   "2: MOV OUT[0], TEMP[0]\n"
-   "3: MOV OUT[1], IN[1]\n"
-   "4: END\n";
-
-
-static const char vs_texture_asm[] =
-   "VERT\n"
-   "DCL IN[0]\n"
-   "DCL IN[1]\n"
-   "DCL OUT[0], POSITION\n"
-   "DCL OUT[1], GENERIC\n"
-   "DCL TEMP[0]\n"
-   "DCL CONST[0..1]\n"
-   "0: MUL TEMP[0], IN[0], CONST[0]\n"
-   "1: ADD TEMP[0], TEMP[0], CONST[1]\n"
-   "2: MOV OUT[0], TEMP[0]\n"
-   "3: MOV OUT[1], IN[1]\n"
-   "4: END\n";
-
 #endif
diff --git a/src/gallium/state_trackers/vega/image.c b/src/gallium/state_trackers/vega/image.c
index 28bbe420a2..318ea94bdf 100644
--- a/src/gallium/state_trackers/vega/image.c
+++ b/src/gallium/state_trackers/vega/image.c
@@ -42,6 +42,7 @@
 #include "util/u_memory.h"
 #include "util/u_math.h"
 #include "util/u_sampler.h"
+#include "util/u_surface.h"
 
 static enum pipe_format vg_format_to_pipe(VGImageFormat format)
 {
@@ -77,33 +78,23 @@ static INLINE void vg_sync_size(VGfloat *src_loc, VGfloat *dst_loc)
    dst_loc[3] = src_loc[3];
 }
 
-
-static void vg_copy_texture(struct vg_context *ctx,
-                            struct pipe_resource *dst, VGint dx, VGint dy,
-                            struct pipe_sampler_view *src, VGint sx, VGint sy,
-                            VGint width, VGint height)
+static void vg_get_copy_coords(VGfloat *src_loc,
+                               VGfloat src_width, VGfloat src_height,
+                               VGfloat *dst_loc,
+                               VGfloat dst_width, VGfloat dst_height)
 {
-   VGfloat dst_loc[4], src_loc[4];
    VGfloat dst_bounds[4], src_bounds[4];
    VGfloat src_shift[4], dst_shift[4], shift[4];
 
-   dst_loc[0] = dx;
-   dst_loc[1] = dy;
-   dst_loc[2] = width;
-   dst_loc[3] = height;
    dst_bounds[0] = 0.f;
    dst_bounds[1] = 0.f;
-   dst_bounds[2] = dst->width0;
-   dst_bounds[3] = dst->height0;
+   dst_bounds[2] = dst_width;
+   dst_bounds[3] = dst_height;
 
-   src_loc[0] = sx;
-   src_loc[1] = sy;
-   src_loc[2] = width;
-   src_loc[3] = height;
    src_bounds[0] = 0.f;
    src_bounds[1] = 0.f;
-   src_bounds[2] = src->texture->width0;
-   src_bounds[3] = src->texture->height0;
+   src_bounds[2] = src_width;
+   src_bounds[3] = src_height;
 
    vg_bound_rect(src_loc, src_bounds, src_shift);
    vg_bound_rect(dst_loc, dst_bounds, dst_shift);
@@ -121,22 +112,44 @@ static void vg_copy_texture(struct vg_context *ctx,
       vg_shift_recty(dst_loc, dst_bounds, shift[1]);
 
    vg_sync_size(src_loc, dst_loc);
+}
+
+static void vg_copy_texture(struct vg_context *ctx,
+                            struct pipe_resource *dst, VGint dx, VGint dy,
+                            struct pipe_sampler_view *src, VGint sx, VGint sy,
+                            VGint width, VGint height)
+{
+   VGfloat dst_loc[4], src_loc[4];
+
+   dst_loc[0] = dx;
+   dst_loc[1] = dy;
+   dst_loc[2] = width;
+   dst_loc[3] = height;
+
+   src_loc[0] = sx;
+   src_loc[1] = sy;
+   src_loc[2] = width;
+   src_loc[3] = height;
+
+   vg_get_copy_coords(src_loc, src->texture->width0, src->texture->height0,
+                      dst_loc, dst->width0, dst->height0);
 
    if (src_loc[2] >= 0 && src_loc[3] >= 0 &&
        dst_loc[2] >= 0 && dst_loc[3] >= 0) {
-      renderer_copy_texture(ctx->renderer,
-                            src,
-                            src_loc[0],
-                            src_loc[1] + src_loc[3],
-                            src_loc[0] + src_loc[2],
-                            src_loc[1],
-                            dst,
-                            dst_loc[0],
-                            dst_loc[1] + dst_loc[3],
-                            dst_loc[0] + dst_loc[2],
-                            dst_loc[1]);
-   }
+      struct pipe_surface *surf, surf_tmpl;
+
+      /* get the destination surface */
+      u_surface_default_template(&surf_tmpl, dst, PIPE_BIND_RENDER_TARGET);
+      surf = ctx->pipe->create_surface(ctx->pipe, dst, &surf_tmpl);
+      if (surf && renderer_copy_begin(ctx->renderer, surf, VG_TRUE, src)) {
+         renderer_copy(ctx->renderer,
+               dst_loc[0], dst_loc[1], dst_loc[2], dst_loc[3],
+               src_loc[0], src_loc[1], src_loc[2], src_loc[3]);
+         renderer_copy_end(ctx->renderer);
+      }
 
+      pipe_surface_reference(&surf, NULL);
+   }
 }
 
 void vg_copy_surface(struct vg_context *ctx,
@@ -145,43 +158,19 @@ void vg_copy_surface(struct vg_context *ctx,
                      VGint width, VGint height)
 {
    VGfloat dst_loc[4], src_loc[4];
-   VGfloat dst_bounds[4], src_bounds[4];
-   VGfloat src_shift[4], dst_shift[4], shift[4];
 
    dst_loc[0] = dx;
    dst_loc[1] = dy;
    dst_loc[2] = width;
    dst_loc[3] = height;
-   dst_bounds[0] = 0.f;
-   dst_bounds[1] = 0.f;
-   dst_bounds[2] = dst->width;
-   dst_bounds[3] = dst->height;
 
    src_loc[0] = sx;
    src_loc[1] = sy;
    src_loc[2] = width;
    src_loc[3] = height;
-   src_bounds[0] = 0.f;
-   src_bounds[1] = 0.f;
-   src_bounds[2] = src->width;
-   src_bounds[3] = src->height;
-
-   vg_bound_rect(src_loc, src_bounds, src_shift);
-   vg_bound_rect(dst_loc, dst_bounds, dst_shift);
-   shift[0] = src_shift[0] - dst_shift[0];
-   shift[1] = src_shift[1] - dst_shift[1];
-
-   if (shift[0] < 0)
-      vg_shift_rectx(src_loc, src_bounds, -shift[0]);
-   else
-      vg_shift_rectx(dst_loc, dst_bounds, shift[0]);
-
-   if (shift[1] < 0)
-      vg_shift_recty(src_loc, src_bounds, -shift[1]);
-   else
-      vg_shift_recty(dst_loc, dst_bounds, shift[1]);
 
-   vg_sync_size(src_loc, dst_loc);
+   vg_get_copy_coords(src_loc, src->width, src->height,
+                      dst_loc, dst->width, dst->height);
 
    if (src_loc[2] > 0 && src_loc[3] > 0 &&
        dst_loc[2] > 0 && dst_loc[3] > 0) {
@@ -277,6 +266,7 @@ struct vg_image * image_create(VGImageFormat format,
    pt.width0 = width;
    pt.height0 = height;
    pt.depth0 = 1;
+   pt.array_size = 1;
    pt.bind = PIPE_BIND_SAMPLER_VIEW;
 
    newtex = screen->resource_create(screen, &pt);
@@ -284,6 +274,13 @@ struct vg_image * image_create(VGImageFormat format,
    debug_assert(newtex);
 
    u_sampler_view_default_template(&view_templ, newtex, newtex->format);
+   /* R, G, and B are treated as 1.0 for alpha-only formats in OpenVG */
+   if (newtex->format == PIPE_FORMAT_A8_UNORM) {
+      view_templ.swizzle_r = PIPE_SWIZZLE_ONE;
+      view_templ.swizzle_g = PIPE_SWIZZLE_ONE;
+      view_templ.swizzle_b = PIPE_SWIZZLE_ONE;
+   }
+
    view = pipe->create_sampler_view(pipe, newtex, &view_templ);
    /* want the texture to go away if the view is freed */
    pipe_resource_reference(&newtex, NULL);
@@ -420,7 +417,7 @@ void image_sub_data(struct vg_image *image,
 
    { /* upload color_data */
       struct pipe_transfer *transfer = pipe_get_transfer(
-         pipe, texture, 0, 0, 0,
+         pipe, texture, 0, 0,
          PIPE_TRANSFER_WRITE, 0, 0, texture->width0, texture->height0);
       src += (dataStride * yoffset);
       for (i = 0; i < height; i++) {
@@ -451,11 +448,11 @@ void image_get_sub_data(struct vg_image * image,
    {
       struct pipe_transfer *transfer =
          pipe_get_transfer(pipe,
-                                  image->sampler_view->texture,  0, 0, 0,
-                                  PIPE_TRANSFER_READ,
-                                  0, 0,
-                                  image->x + image->width,
-                                  image->y + image->height);
+                           image->sampler_view->texture,  0, 0,
+                           PIPE_TRANSFER_READ,
+                           0, 0,
+                           image->x + image->width,
+                           image->y + image->height);
       /* Do a row at a time to flip image data vertically */
       for (i = 0; i < height; i++) {
 #if 0
@@ -525,14 +522,20 @@ void image_copy(struct vg_image *dst, VGint dx, VGint dy,
                    src->sampler_view, src->x + sx, src->y + sy, width, height);
 }
 
-void image_draw(struct vg_image *img)
+void image_draw(struct vg_image *img, struct matrix *matrix)
 {
    struct vg_context *ctx = vg_current_context();
+   struct matrix paint_matrix;
    VGfloat x1, y1;
    VGfloat x2, y2;
    VGfloat x3, y3;
    VGfloat x4, y4;
-   struct matrix *matrix;
+
+   if (!vg_get_paint_matrix(ctx,
+                            &ctx->state.vg.fill_paint_to_user_matrix,
+                            matrix,
+                            &paint_matrix))
+      return;
 
    x1 = 0;
    y1 = 0;
@@ -543,15 +546,10 @@ void image_draw(struct vg_image *img)
    x4 = 0;
    y4 = img->height;
 
-   matrix = &ctx->state.vg.image_user_to_surface_matrix;
-
-   matrix_map_point(matrix, x1, y1, &x1, &y1);
-   matrix_map_point(matrix, x2, y2, &x2, &y2);
-   matrix_map_point(matrix, x3, y3, &x3, &y3);
-   matrix_map_point(matrix, x4, y4, &x4, &y4);
-
+   shader_set_surface_matrix(ctx->shader, matrix);
    shader_set_drawing_image(ctx->shader, VG_TRUE);
    shader_set_paint(ctx->shader, ctx->state.vg.fill_paint);
+   shader_set_paint_matrix(ctx->shader, &paint_matrix);
    shader_set_image(ctx->shader, img);
    shader_bind(ctx->shader);
 
@@ -566,20 +564,21 @@ void image_set_pixels(VGint dx, VGint dy,
 {
    struct vg_context *ctx = vg_current_context();
    struct pipe_context *pipe = ctx->pipe;
-   struct pipe_screen *screen = pipe->screen;
-   struct pipe_surface *surf;
+   struct pipe_surface *surf, surf_tmpl;
    struct st_renderbuffer *strb = ctx->draw_buffer->strb;
 
    /* make sure rendering has completed */
    pipe->flush(pipe, PIPE_FLUSH_RENDER_CACHE, NULL);
 
-   surf = screen->get_tex_surface(screen, image_texture(src),  0, 0, 0,
-                                  0 /* no bind flags as surf isn't actually used??? */);
+   memset(&surf_tmpl, 0, sizeof(surf_tmpl));
+   u_surface_default_template(&surf_tmpl, image_texture(src),
+                              0 /* no bind flag - not a surface*/);
+   surf = pipe->create_surface(pipe, image_texture(src), &surf_tmpl);
 
    vg_copy_surface(ctx, strb->surface, dx, dy,
                    surf, sx+src->x, sy+src->y, width, height);
 
-   screen->tex_surface_destroy(surf);
+   pipe->surface_destroy(pipe, surf);
 }
 
 void image_get_pixels(struct vg_image *dst, VGint dx, VGint dy,
@@ -588,8 +587,7 @@ void image_get_pixels(struct vg_image *dst, VGint dx, VGint dy,
 {
    struct vg_context *ctx = vg_current_context();
    struct pipe_context *pipe = ctx->pipe;
-   struct pipe_screen *screen = pipe->screen;
-   struct pipe_surface *surf;
+   struct pipe_surface *surf, surf_tmpl;
    struct st_renderbuffer *strb = ctx->draw_buffer->strb;
 
    /* flip the y coordinates */
@@ -598,8 +596,10 @@ void image_get_pixels(struct vg_image *dst, VGint dx, VGint dy,
    /* make sure rendering has completed */
    pipe->flush(pipe, PIPE_FLUSH_RENDER_CACHE, NULL);
 
-   surf = screen->get_tex_surface(screen, image_texture(dst),  0, 0, 0,
-                                  0 /* no bind flags as surf isn't actually used??? */);
+   memset(&surf_tmpl, 0, sizeof(surf_tmpl));
+   u_surface_default_template(&surf_tmpl, image_texture(dst),
+                              PIPE_BIND_RENDER_TARGET);
+   surf = pipe->create_surface(pipe, image_texture(dst), &surf_tmpl);
 
    vg_copy_surface(ctx, surf, dst->x + dx, dst->y + dy,
                    strb->surface, sx, sy, width, height);
diff --git a/src/gallium/state_trackers/vega/image.h b/src/gallium/state_trackers/vega/image.h
index a990c9c587..391c048594 100644
--- a/src/gallium/state_trackers/vega/image.h
+++ b/src/gallium/state_trackers/vega/image.h
@@ -79,7 +79,7 @@ void image_copy(struct vg_image *dst, VGint dx, VGint dy,
                 VGint width, VGint height,
                 VGboolean dither);
 
-void image_draw(struct vg_image *img);
+void image_draw(struct vg_image *img, struct matrix *matrix);
 
 void image_set_pixels(VGint dx, VGint dy,
                       struct vg_image *src, VGint sx, VGint sy,
diff --git a/src/gallium/state_trackers/vega/mask.c b/src/gallium/state_trackers/vega/mask.c
index ef28ebd740..dfd0600e44 100644
--- a/src/gallium/state_trackers/vega/mask.c
+++ b/src/gallium/state_trackers/vega/mask.c
@@ -31,13 +31,14 @@
 #include "shaders_cache.h"
 #include "renderer.h"
 #include "asm_util.h"
-#include "st_inlines.h"
 
 #include "pipe/p_context.h"
 #include "pipe/p_screen.h"
 #include "util/u_inlines.h"
 #include "util/u_format.h"
 #include "util/u_memory.h"
+#include "util/u_surface.h"
+#include "util/u_sampler.h"
 
 struct vg_mask_layer {
    struct vg_object base;
@@ -48,17 +49,6 @@ struct vg_mask_layer {
    struct pipe_sampler_view *sampler_view;
 };
 
-static INLINE struct pipe_surface *
-alpha_mask_surface(struct vg_context *ctx, int usage)
-{
-   struct pipe_screen *screen = ctx->pipe->screen;
-   struct st_framebuffer *stfb = ctx->draw_buffer;
-   return screen->get_tex_surface(screen,
-                                  stfb->alpha_mask_view->texture,
-                                  0, 0, 0,
-                                  usage);
-}
-
 static INLINE VGboolean
 intersect_rectangles(VGint dwidth, VGint dheight,
                      VGint swidth, VGint sheight,
@@ -110,15 +100,13 @@ static void read_alpha_mask(void * data, VGint dataStride,
 {
    struct vg_context *ctx = vg_current_context();
    struct pipe_context *pipe = ctx->pipe;
-   struct pipe_screen *screen = pipe->screen;
 
    struct st_framebuffer *stfb = ctx->draw_buffer;
    struct st_renderbuffer *strb = stfb->alpha_mask;
-   struct pipe_framebuffer_state *fb = &ctx->state.g3d.fb;
 
    VGfloat temp[VEGA_MAX_IMAGE_WIDTH][4];
    VGfloat *df = (VGfloat*)temp;
-   VGint y = (fb->height - sy) - 1, yStep = -1;
+   VGint y = (stfb->height - sy) - 1, yStep = -1;
    VGint i;
    VGubyte *dst = (VGubyte *)data;
    VGint xoffset = 0, yoffset = 0;
@@ -135,15 +123,15 @@ static void read_alpha_mask(void * data, VGint dataStride,
       yoffset = -sy;
       height += sy;
       sy = 0;
-      y = (fb->height - sy) - 1;
+      y = (stfb->height - sy) - 1;
       yoffset *= dataStride;
    }
 
    {
       struct pipe_surface *surf;
 
-      surf = screen->get_tex_surface(screen, strb->texture,  0, 0, 0,
-                                     PIPE_BIND_TRANSFER_READ);
+      surf = pipe->create_surface(pipe, strb->texture,  0, 0, 0,
+                                  PIPE_BIND_TRANSFER_READ);
 
       /* Do a row at a time to flip image data vertically */
       for (i = 0; i < height; i++) {
@@ -164,23 +152,23 @@ static void read_alpha_mask(void * data, VGint dataStride,
 void save_alpha_to_file(const char *filename)
 {
    struct vg_context *ctx = vg_current_context();
-   struct pipe_framebuffer_state *fb = &ctx->state.g3d.fb;
+   struct st_framebuffer *stfb = ctx->draw_buffer;
    VGint *data;
    int i, j;
 
-   data = malloc(sizeof(int) * fb->width * fb->height);
-   read_alpha_mask(data, fb->width * sizeof(int),
+   data = malloc(sizeof(int) * stfb->width * stfb->height);
+   read_alpha_mask(data, stfb->width * sizeof(int),
                    VG_sRGBA_8888,
-                   0, 0, fb->width, fb->height);
+                   0, 0, stfb->width, stfb->height);
    fprintf(stderr, "/*---------- start */\n");
    fprintf(stderr, "const int image_width = %d;\n",
-           fb->width);
+           stfb->width);
    fprintf(stderr, "const int image_height = %d;\n",
-           fb->height);
+           stfb->height);
    fprintf(stderr, "const int image_data = {\n");
-   for (i = 0; i < fb->height; ++i) {
-      for (j = 0; j < fb->width; ++j) {
-         int rgba = data[i * fb->height + j];
+   for (i = 0; i < stfb->height; ++i) {
+      for (j = 0; j < stfb->width; ++j) {
+         int rgba = data[i * stfb->height + j];
          int argb = 0;
          argb = (rgba >> 8);
          argb |= ((rgba & 0xff) << 24);
@@ -193,49 +181,12 @@ void save_alpha_to_file(const char *filename)
 }
 #endif
 
-static void setup_mask_framebuffer(struct pipe_surface *surf,
-                                   VGint surf_width, VGint surf_height)
-{
-   struct vg_context *ctx = vg_current_context();
-   struct pipe_framebuffer_state fb;
-
-   memset(&fb, 0, sizeof(fb));
-   fb.width = surf_width;
-   fb.height = surf_height;
-   fb.nr_cbufs = 1;
-   fb.cbufs[0] = surf;
-   {
-      VGint i;
-      for (i = 1; i < PIPE_MAX_COLOR_BUFS; ++i)
-         fb.cbufs[i] = 0;
-   }
-   cso_set_framebuffer(ctx->cso_context, &fb);
-}
-
-
-/* setup shader constants */
-static void setup_mask_operation(VGMaskOperation operation)
+/* setup mask shader */
+static void *setup_mask_operation(VGMaskOperation operation)
 {
    struct vg_context *ctx = vg_current_context();
-   struct pipe_resource **cbuf = &ctx->mask.cbuf;
-   const VGint param_bytes = 4 * sizeof(VGfloat);
-   const VGfloat ones[4] = {1.f, 1.f, 1.f, 1.f};
    void *shader = 0;
 
-   /* We always need to get a new buffer, to keep the drivers simple and
-    * avoid gratuitous rendering synchronization.
-    */
-   pipe_resource_reference(cbuf, NULL);
-
-   *cbuf = pipe_buffer_create(ctx->pipe->screen, 
-                              PIPE_BIND_CONSTANT_BUFFER,
-                              param_bytes);
-   if (*cbuf) {
-      st_no_flush_pipe_buffer_write(ctx, *cbuf,
-                                    0, param_bytes, ones);
-   }
-
-   ctx->pipe->set_constant_buffer(ctx->pipe, PIPE_SHADER_FRAGMENT, 0, *cbuf);
    switch (operation) {
    case VG_UNION_MASK: {
       if (!ctx->mask.union_fs) {
@@ -281,94 +232,21 @@ static void setup_mask_operation(VGMaskOperation operation)
          assert(0);
       break;
    }
-   cso_set_fragment_shader_handle(ctx->cso_context, shader);
-}
-
-static void setup_mask_samplers(struct pipe_sampler_view *umask)
-{
-   struct vg_context *ctx = vg_current_context();
-   struct pipe_sampler_state *samplers[PIPE_MAX_SAMPLERS];
-   struct pipe_sampler_view *sampler_views[PIPE_MAX_SAMPLERS];
-   struct st_framebuffer *fb_buffers = ctx->draw_buffer;
-   struct pipe_sampler_view *uprev = NULL;
-   struct pipe_sampler_state sampler;
-
-   uprev = fb_buffers->blend_texture_view;
-   sampler = ctx->mask.sampler;
-   sampler.normalized_coords = 1;
-
-   samplers[0] = NULL;
-   samplers[1] = NULL;
-   sampler_views[0] = NULL;
-   sampler_views[1] = NULL;
-
-   samplers[0] = &sampler;
-   samplers[1] = &ctx->mask.sampler;
-
-   sampler_views[0] = umask;
-   sampler_views[1] = uprev;
-
-   cso_set_samplers(ctx->cso_context, 2,
-                    (const struct pipe_sampler_state **)samplers);
-   cso_set_fragment_sampler_views(ctx->cso_context, 2, sampler_views);
-}
-
-
-/* setup shader constants */
-static void setup_mask_fill(const VGfloat color[4])
-{
-   struct vg_context *ctx = vg_current_context();
-   struct pipe_resource **cbuf = &ctx->mask.cbuf;
-   const VGint param_bytes = 4 * sizeof(VGfloat);
-
-   /* We always need to get a new buffer, to keep the drivers simple and
-    * avoid gratuitous rendering synchronization.
-    */
-   pipe_resource_reference(cbuf, NULL);
-
-   *cbuf = pipe_buffer_create(ctx->pipe->screen,
-                              PIPE_BIND_CONSTANT_BUFFER,
-                              param_bytes);
-   if (*cbuf) {
-      st_no_flush_pipe_buffer_write(ctx, *cbuf, 0, param_bytes, color);
-   }
-
-   ctx->pipe->set_constant_buffer(ctx->pipe, PIPE_SHADER_FRAGMENT, 0, *cbuf);
-   cso_set_fragment_shader_handle(ctx->cso_context,
-                                  shaders_cache_fill(ctx->sc,
-                                                     VEGA_SOLID_FILL_SHADER));
-}
-
-static void setup_mask_viewport()
-{
-   struct vg_context *ctx = vg_current_context();
-   vg_set_viewport(ctx, VEGA_Y0_TOP);
-}
-
-static void setup_mask_blend()
-{
-   struct vg_context *ctx = vg_current_context();
-
-   struct pipe_blend_state blend;
 
-   memset(&blend, 0, sizeof(struct pipe_blend_state));
-   blend.rt[0].blend_enable = 0;
-   blend.rt[0].colormask = PIPE_MASK_RGBA;
-   blend.rt[0].rgb_src_factor = PIPE_BLENDFACTOR_ONE;
-   blend.rt[0].alpha_src_factor = PIPE_BLENDFACTOR_ONE;
-   blend.rt[0].rgb_dst_factor = PIPE_BLENDFACTOR_ZERO;
-   blend.rt[0].alpha_dst_factor = PIPE_BLENDFACTOR_ZERO;
-
-   cso_set_blend(ctx->cso_context, &blend);
+   return shader;
 }
 
-
-static void surface_fill(struct pipe_surface *surf,
-                         int surf_width, int surf_height,
-                         int x, int y, int width, int height,
-                         const VGfloat color[4])
+static void mask_resource_fill(struct pipe_resource *dst,
+                               int x, int y, int width, int height,
+                               VGfloat coverage)
 {
    struct vg_context *ctx = vg_current_context();
+   VGfloat fs_consts[12] = {
+      0.0f, 0.0f, 0.0f, 0.0f, /* not used */
+      0.0f, 0.0f, 0.0f, 0.0f, /* not used */
+      0.0f, 0.0f, 0.0f, coverage /* color */
+   };
+   void *fs;
 
    if (x < 0) {
       width += x;
@@ -379,50 +257,38 @@ static void surface_fill(struct pipe_surface *surf,
       y = 0;
    }
 
-   cso_save_framebuffer(ctx->cso_context);
-   cso_save_blend(ctx->cso_context);
-   cso_save_fragment_shader(ctx->cso_context);
-   cso_save_viewport(ctx->cso_context);
-
-   setup_mask_blend();
-   setup_mask_fill(color);
-   setup_mask_framebuffer(surf, surf_width, surf_height);
-   setup_mask_viewport();
+   fs = shaders_cache_fill(ctx->sc, VEGA_SOLID_FILL_SHADER);
 
-   renderer_draw_quad(ctx->renderer, x, y,
-                      x + width, y + height, 0.0f/*depth should be disabled*/);
-
-
-   /* make sure rendering has completed */
-   ctx->pipe->flush(ctx->pipe,
-                    PIPE_FLUSH_RENDER_CACHE | PIPE_FLUSH_FRAME,
-                    NULL);
+   if (renderer_filter_begin(ctx->renderer, dst, VG_FALSE, ~0,
+            NULL, NULL, 0, fs, (const void *) fs_consts, sizeof(fs_consts))) {
+      renderer_filter(ctx->renderer, x, y, width, height, 0, 0, 0, 0);
+      renderer_filter_end(ctx->renderer);
+   }
 
 #if DEBUG_MASKS
    save_alpha_to_file(0);
 #endif
-
-   cso_restore_blend(ctx->cso_context);
-   cso_restore_framebuffer(ctx->cso_context);
-   cso_restore_fragment_shader(ctx->cso_context);
-   cso_restore_viewport(ctx->cso_context);
 }
 
 
 static void mask_using_texture(struct pipe_sampler_view *sampler_view,
+                               VGboolean is_layer,
                                VGMaskOperation operation,
                                VGint x, VGint y,
                                VGint width, VGint height)
 {
    struct vg_context *ctx = vg_current_context();
+   struct pipe_sampler_view *dst_view = vg_get_surface_mask(ctx);
+   struct pipe_resource *dst = dst_view->texture;
    struct pipe_resource *texture = sampler_view->texture;
-   struct pipe_surface *surface =
-      alpha_mask_surface(ctx, PIPE_BIND_RENDER_TARGET);
+   const struct pipe_sampler_state *samplers[2];
+   struct pipe_sampler_view *views[2];
+   struct pipe_sampler_state sampler;
    VGint offsets[4], loc[4];
+   const VGfloat ones[4] = {1.f, 1.f, 1.f, 1.f};
+   void *fs;
 
-   if (!surface)
-      return;
-   if (!intersect_rectangles(surface->width, surface->height,
+   if (!intersect_rectangles(dst->width0, dst->height0,
                              texture->width0, texture->height0,
                              x, y, width, height,
                              offsets, loc))
@@ -434,38 +300,30 @@ static void mask_using_texture(struct pipe_sampler_view *sampler_view,
                 loc[1], loc[2], loc[3]);
 #endif
 
+
+   sampler = ctx->mask.sampler;
+   sampler.normalized_coords = 1;
+   samplers[0] = &sampler;
+   views[0] = sampler_view;
+
    /* prepare our blend surface */
-   vg_prepare_blend_surface_from_mask(ctx);
-
-   cso_save_samplers(ctx->cso_context);
-   cso_save_fragment_sampler_views(ctx->cso_context);
-   cso_save_framebuffer(ctx->cso_context);
-   cso_save_blend(ctx->cso_context);
-   cso_save_fragment_shader(ctx->cso_context);
-   cso_save_viewport(ctx->cso_context);
-
-   setup_mask_samplers(sampler_view);
-   setup_mask_blend();
-   setup_mask_operation(operation);
-   setup_mask_framebuffer(surface, surface->width, surface->height);
-   setup_mask_viewport();
-
-   /* render the quad to propagate the rendering from stencil */
-   renderer_draw_texture(ctx->renderer, texture,
-                         offsets[0], offsets[1],
-                         offsets[0] + offsets[2], offsets[1] + offsets[3],
-                         loc[0], loc[1], loc[0] + loc[2], loc[1] + loc[3]);
+   samplers[1] = &ctx->mask.sampler;
+   views[1] = vg_prepare_blend_surface_from_mask(ctx);
 
-   /* make sure rendering has completed */
-   ctx->pipe->flush(ctx->pipe, PIPE_FLUSH_RENDER_CACHE, NULL);
-   cso_restore_blend(ctx->cso_context);
-   cso_restore_framebuffer(ctx->cso_context);
-   cso_restore_fragment_shader(ctx->cso_context);
-   cso_restore_samplers(ctx->cso_context);
-   cso_restore_fragment_sampler_views(ctx->cso_context);
-   cso_restore_viewport(ctx->cso_context);
-
-   pipe_surface_reference(&surface, NULL);
+   fs = setup_mask_operation(operation);
+
+   if (renderer_filter_begin(ctx->renderer, dst, VG_FALSE,
+            ~0, samplers, views, 2, fs, (const void *) ones, sizeof(ones))) {
+      /* layer should be flipped when used as a texture */
+      if (is_layer) {
+         offsets[1] += offsets[3];
+         offsets[3] = -offsets[3];
+      }
+      renderer_filter(ctx->renderer,
+            loc[0], loc[1], loc[2], loc[3],
+            offsets[0], offsets[1], offsets[2], offsets[3]);
+      renderer_filter_end(ctx->renderer);
+   }
 }
 
 
@@ -496,8 +354,8 @@ struct vg_mask_layer * mask_layer_create(VGint width, VGint height)
       pt.width0 = width;
       pt.height0 = height;
       pt.depth0 = 1;
+      pt.array_size = 1;
       pt.bind = PIPE_BIND_SAMPLER_VIEW;
-      pt.compressed = 0;
 
       texture = screen->resource_create(screen, &pt);
 
@@ -519,7 +377,7 @@ void mask_layer_destroy(struct vg_mask_layer *layer)
    struct vg_context *ctx = vg_current_context();
 
    vg_context_remove_object(ctx, VG_OBJECT_MASK, layer);
-   pipe_resource_release(&layer->texture);
+   pipe_sampler_view_reference(&layer->sampler_view, NULL);
    FREE(layer);
 }
 
@@ -528,22 +386,12 @@ void mask_layer_fill(struct vg_mask_layer *layer,
                      VGint width, VGint height,
                      VGfloat value)
 {
-   struct vg_context *ctx = vg_current_context();
    VGfloat alpha_color[4] = {0, 0, 0, 0};
-   struct pipe_surface *surface;
 
    alpha_color[3] = value;
 
-   surface = ctx->pipe->screen->get_tex_surface(
-      ctx->pipe->screen, layer->sampler_view->texture,
-      0, 0, 0,
-      PIPE_BIND_RENDER_TARGET);
-
-   surface_fill(surface,
-                layer->width, layer->height,
-                x, y, width, height, alpha_color);
-
-   ctx->pipe->screen->tex_surface_release(ctx->pipe->screen, &surface);
+   mask_resource_fill(layer->sampler_view->texture,
+                      x, y, width, height, value);
 }
 
 void mask_copy(struct vg_mask_layer *layer,
@@ -551,16 +399,27 @@ void mask_copy(struct vg_mask_layer *layer,
                VGint dx, VGint dy,
                VGint width, VGint height)
 {
-    struct vg_context *ctx = vg_current_context();
-    struct st_framebuffer *fb_buffers = ctx->draw_buffer;
-
-    renderer_copy_texture(ctx->renderer,
-                          layer->sampler_view,
-                          sx, sy,
-                          sx + width, sy + height,
-                          fb_buffers->alpha_mask_view->texture,
-                          dx, dy,
-                          dx + width, dy + height);
+   struct vg_context *ctx = vg_current_context();
+   struct pipe_sampler_view *src = vg_get_surface_mask(ctx);
+   struct pipe_surface *surf, surf_tmpl;
+
+   /* get the destination surface */
+   u_surface_default_template(&surf_tmpl, layer->sampler_view->texture,
+                              PIPE_BIND_RENDER_TARGET);
+   surf = ctx->pipe->create_surface(ctx->pipe, layer->sampler_view->texture,
+                                    &surf_tmpl);
+   if (surf && renderer_copy_begin(ctx->renderer, surf, VG_FALSE, src)) {
+      /* layer should be flipped when used as a texture */
+      sy += height;
+      height = -height;
+
+      renderer_copy(ctx->renderer,
+            dx, dy, width, height,
+            sx, sy, width, height);
+      renderer_copy_end(ctx->renderer);
+   }
+
+   pipe_surface_reference(&surf, NULL);
 }
 
 static void mask_layer_render_to(struct vg_mask_layer *layer,
@@ -568,41 +427,25 @@ static void mask_layer_render_to(struct vg_mask_layer *layer,
                                  VGbitfield paint_modes)
 {
    struct vg_context *ctx = vg_current_context();
-   const VGfloat fill_color[4] = {1.f, 1.f, 1.f, 1.f};
-   struct pipe_screen *screen = ctx->pipe->screen;
-   struct pipe_surface *surface;
-
-   surface = screen->get_tex_surface(screen, layer->sampler_view->texture,  0, 0, 0,
-                                     PIPE_BIND_RENDER_TARGET);
-
-   cso_save_framebuffer(ctx->cso_context);
-   cso_save_fragment_shader(ctx->cso_context);
-   cso_save_viewport(ctx->cso_context);
+   struct pipe_context *pipe = ctx->pipe;
+   struct pipe_sampler_view *view = vg_get_surface_mask(ctx);
+   struct matrix *mat = &ctx->state.vg.path_user_to_surface_matrix;
+   struct pipe_surface *surf, surf_tmpl;
+   u_surface_default_template(&surf_tmpl, view->texture,
+                              PIPE_BIND_RENDER_TARGET);
+   surf = pipe->create_surface(pipe, view->texture, &surf_tmpl);
 
-   setup_mask_blend();
-   setup_mask_fill(fill_color);
-   setup_mask_framebuffer(surface, layer->width, layer->height);
-   setup_mask_viewport();
+   renderer_validate_for_mask_rendering(ctx->renderer, surf, mat);
 
    if (paint_modes & VG_FILL_PATH) {
-      struct matrix *mat = &ctx->state.vg.path_user_to_surface_matrix;
-      path_fill(path, mat);
+      path_fill(path);
    }
 
    if (paint_modes & VG_STROKE_PATH){
       path_stroke(path);
    }
 
-
-   /* make sure rendering has completed */
-   ctx->pipe->flush(ctx->pipe, PIPE_FLUSH_RENDER_CACHE, NULL);
-
-   cso_restore_framebuffer(ctx->cso_context);
-   cso_restore_fragment_shader(ctx->cso_context);
-   cso_restore_viewport(ctx->cso_context);
-   ctx->state.dirty |= BLEND_DIRTY;
-
-   screen->tex_surface_release(ctx->pipe->screen, &surface);
+   pipe_surface_reference(&surf, NULL);
 }
 
 void mask_render_to(struct path *path,
@@ -610,19 +453,19 @@ void mask_render_to(struct path *path,
                     VGMaskOperation operation)
 {
    struct vg_context *ctx = vg_current_context();
-   struct st_framebuffer *fb_buffers = ctx->draw_buffer;
+   struct st_framebuffer *stfb = ctx->draw_buffer;
    struct vg_mask_layer *temp_layer;
    VGint width, height;
 
-   width = fb_buffers->alpha_mask_view->texture->width0;
-   height = fb_buffers->alpha_mask_view->texture->width0;
+   width = stfb->width;
+   height = stfb->height;
 
    temp_layer = mask_layer_create(width, height);
+   mask_layer_fill(temp_layer, 0, 0, width, height, 0.0f);
 
    mask_layer_render_to(temp_layer, path, paint_modes);
 
-   mask_using_layer(temp_layer, 0, 0, width, height,
-                    operation);
+   mask_using_layer(temp_layer, operation, 0, 0, width, height);
 
    mask_layer_destroy(temp_layer);
 }
@@ -632,7 +475,7 @@ void mask_using_layer(struct vg_mask_layer *layer,
                       VGint x, VGint y,
                       VGint width, VGint height)
 {
-   mask_using_texture(layer->sampler_view, operation,
+   mask_using_texture(layer->sampler_view, VG_TRUE, operation,
                       x, y, width, height);
 }
 
@@ -654,7 +497,7 @@ void mask_using_image(struct vg_image *image,
                       VGint x, VGint y,
                       VGint width, VGint height)
 {
-   mask_using_texture(image->sampler_view, operation,
+   mask_using_texture(image->sampler_view, VG_FALSE, operation,
                       x, y, width, height);
 }
 
@@ -662,23 +505,15 @@ void mask_fill(VGint x, VGint y, VGint width, VGint height,
                VGfloat value)
 {
    struct vg_context *ctx = vg_current_context();
-   VGfloat alpha_color[4] = {.0f, .0f, .0f, value};
-   struct pipe_surface *surf = alpha_mask_surface(
-      ctx, PIPE_BIND_RENDER_TARGET);
+   struct pipe_sampler_view *view = vg_get_surface_mask(ctx);
 
 #if DEBUG_MASKS
    debug_printf("mask_fill(%d, %d, %d, %d) with  rgba(%f, %f, %f, %f)\n",
                 x, y, width, height,
-                alpha_color[0], alpha_color[1],
-                alpha_color[2], alpha_color[3]);
-   debug_printf("XXX %f  === %f \n",
-                alpha_color[3], value);
+                0.0f, 0.0f, 0.0f, value);
 #endif
 
-   surface_fill(surf, surf->width, surf->height,
-                x, y, width, height, alpha_color);
-
-   pipe_surface_reference(&surf, NULL);
+   mask_resource_fill(view->texture, x, y, width, height, value);
 }
 
 VGint mask_bind_samplers(struct pipe_sampler_state **samplers,
@@ -687,10 +522,8 @@ VGint mask_bind_samplers(struct pipe_sampler_state **samplers,
    struct vg_context *ctx = vg_current_context();
 
    if (ctx->state.vg.masking) {
-      struct st_framebuffer *fb_buffers = ctx->draw_buffer;
-
       samplers[1] = &ctx->mask.sampler;
-      sampler_views[1] = fb_buffers->alpha_mask_view;
+      sampler_views[1] = vg_get_surface_mask(ctx);
       return 1;
    } else
       return 0;
diff --git a/src/gallium/state_trackers/vega/matrix.h b/src/gallium/state_trackers/vega/matrix.h
index 4c207f912a..bed2b3193d 100644
--- a/src/gallium/state_trackers/vega/matrix.h
+++ b/src/gallium/state_trackers/vega/matrix.h
@@ -129,7 +129,7 @@ static INLINE void matrix_make_affine(struct matrix *matrix)
 }
 
 static INLINE void matrix_mult(struct matrix *dst,
-                               struct matrix *src)
+                               const struct matrix *src)
 {
    VGfloat m11 = dst->m[0]*src->m[0] + dst->m[3]*src->m[1] + dst->m[6]*src->m[2];
    VGfloat m12 = dst->m[0]*src->m[3] + dst->m[3]*src->m[4] + dst->m[6]*src->m[5];
diff --git a/src/gallium/state_trackers/vega/paint.c b/src/gallium/state_trackers/vega/paint.c
index c0c8cddabe..2db8cbcf7c 100644
--- a/src/gallium/state_trackers/vega/paint.c
+++ b/src/gallium/state_trackers/vega/paint.c
@@ -28,7 +28,6 @@
 
 #include "matrix.h"
 #include "image.h"
-#include "st_inlines.h"
 
 #include "pipe/p_compiler.h"
 #include "util/u_inlines.h"
@@ -155,14 +154,15 @@ static INLINE struct pipe_resource *create_gradient_texture(struct vg_paint *p)
    templ.width0 = 1024;
    templ.height0 = 1;
    templ.depth0 = 1;
+   templ.array_size = 1;
    templ.bind = PIPE_BIND_SAMPLER_VIEW;
 
    tex = screen->resource_create(screen, &templ);
 
    { /* upload color_data */
       struct pipe_transfer *transfer =
-         st_no_flush_get_transfer(p->base.ctx, tex, 0, 0, 0,
-                                      PIPE_TRANSFER_WRITE, 0, 0, 1024, 1);
+         pipe_get_transfer(p->base.ctx->pipe, tex, 0, 0,
+                           PIPE_TRANSFER_WRITE, 0, 0, 1024, 1);
       void *map = pipe->transfer_map(pipe, transfer);
       memcpy(map, p->gradient.color_data, sizeof(VGint)*1024);
       pipe->transfer_unmap(pipe, transfer);
@@ -261,14 +261,18 @@ static INLINE void paint_color_buffer(struct vg_paint *paint, void *buffer)
    map[7] = 4.f;
 }
 
-static INLINE void paint_linear_gradient_buffer(struct vg_paint *paint, void *buffer)
+static INLINE void paint_linear_gradient_buffer(struct vg_paint *paint,
+                                                const struct matrix *inv,
+                                                void *buffer)
 {
-   struct vg_context *ctx = paint->base.ctx;
    VGfloat *map = (VGfloat*)buffer;
+   VGfloat dd;
 
    map[0] = paint->gradient.linear.coords[2] - paint->gradient.linear.coords[0];
    map[1] = paint->gradient.linear.coords[3] - paint->gradient.linear.coords[1];
-   map[2] = 1.f / (map[0] * map[0] + map[1] * map[1]);
+   dd = (map[0] * map[0] + map[1] * map[1]);
+
+   map[2] = (dd > 0.0f) ? 1.f / dd : 0.f;
    map[3] = 1.f;
 
    map[4] = 0.f;
@@ -277,15 +281,10 @@ static INLINE void paint_linear_gradient_buffer(struct vg_paint *paint, void *bu
    map[7] = 4.f;
    {
       struct matrix mat;
-      struct matrix inv;
       matrix_load_identity(&mat);
+      /* VEGA_LINEAR_GRADIENT_SHADER expects the first point to be at (0, 0) */
       matrix_translate(&mat, -paint->gradient.linear.coords[0], -paint->gradient.linear.coords[1]);
-      memcpy(&inv, &ctx->state.vg.fill_paint_to_user_matrix,
-             sizeof(struct matrix));
-      matrix_invert(&inv);
-      matrix_mult(&inv, &mat);
-      memcpy(&mat, &inv,
-             sizeof(struct matrix));
+      matrix_mult(&mat, inv);
 
       map[8]  = mat.m[0]; map[9]  = mat.m[3]; map[10] = mat.m[6]; map[11] = 0.f;
       map[12] = mat.m[1]; map[13] = mat.m[4]; map[14] = mat.m[7]; map[15] = 0.f;
@@ -298,17 +297,37 @@ static INLINE void paint_linear_gradient_buffer(struct vg_paint *paint, void *bu
 }
 
 
-static INLINE void paint_radial_gradient_buffer(struct vg_paint *paint, void *buffer)
+static INLINE void paint_radial_gradient_buffer(struct vg_paint *paint,
+                                                const struct matrix *inv,
+                                                void *buffer)
 {
-   VGfloat *radialCoords = paint->gradient.radial.vals;
-   struct vg_context *ctx = paint->base.ctx;
-
+   const VGfloat *center = &paint->gradient.radial.vals[0];
+   const VGfloat *focal = &paint->gradient.radial.vals[2];
+   VGfloat rr = paint->gradient.radial.vals[4];
    VGfloat *map = (VGfloat*)buffer;
+   VGfloat dd, new_focal[2];
+
+   rr *= rr;
+
+   map[0] = center[0] - focal[0];
+   map[1] = center[1] - focal[1];
+   dd = map[0] * map[0] + map[1] * map[1];
+
+   /* focal point must lie inside the circle */
+   if (0.998f * rr < dd) {
+      VGfloat scale;
+
+      scale = (dd > 0.0f) ? sqrt(0.998f * rr / dd) : 0.0f;
+      map[0] *= scale;
+      map[1] *= scale;
 
-   map[0] = radialCoords[0] - radialCoords[2];
-   map[1] = radialCoords[1] - radialCoords[3];
-   map[2] = -map[0] * map[0] - map[1] * map[1] +
-            radialCoords[4] * radialCoords[4];
+      new_focal[0] = center[0] - map[0];
+      new_focal[1] = center[1] - map[1];
+      dd = map[0] * map[0] + map[1] * map[1];
+      focal = new_focal;
+   }
+
+   map[2] = (rr > dd) ? rr - dd : 1.0f;
    map[3] = 1.f;
 
    map[4] = 0.f;
@@ -318,15 +337,9 @@ static INLINE void paint_radial_gradient_buffer(struct vg_paint *paint, void *bu
 
    {
       struct matrix mat;
-      struct matrix inv;
       matrix_load_identity(&mat);
-      matrix_translate(&mat, -radialCoords[2], -radialCoords[3]);
-      memcpy(&inv, &ctx->state.vg.fill_paint_to_user_matrix,
-             sizeof(struct matrix));
-      matrix_invert(&inv);
-      matrix_mult(&inv, &mat);
-      memcpy(&mat, &inv,
-             sizeof(struct matrix));
+      matrix_translate(&mat, -focal[0], -focal[1]);
+      matrix_mult(&mat, inv);
 
       map[8]  = mat.m[0]; map[9]  = mat.m[3]; map[10] = mat.m[6]; map[11] = 0.f;
       map[12] = mat.m[1]; map[13] = mat.m[4]; map[14] = mat.m[7]; map[15] = 0.f;
@@ -340,10 +353,10 @@ static INLINE void paint_radial_gradient_buffer(struct vg_paint *paint, void *bu
 }
 
 
-static INLINE void  paint_pattern_buffer(struct vg_paint *paint, void *buffer)
+static INLINE void  paint_pattern_buffer(struct vg_paint *paint,
+                                         const struct matrix *inv,
+                                         void *buffer)
 {
-   struct vg_context *ctx = paint->base.ctx;
-
    VGfloat *map = (VGfloat *)buffer;
    memcpy(map, paint->solid.color, 4 * sizeof(VGfloat));
 
@@ -353,17 +366,8 @@ static INLINE void  paint_pattern_buffer(struct vg_paint *paint, void *buffer)
    map[7] = paint->pattern.sampler_view->texture->height0;
    {
       struct matrix mat;
-      memcpy(&mat, &ctx->state.vg.fill_paint_to_user_matrix,
-             sizeof(struct matrix));
-      matrix_invert(&mat);
-      {
-         struct matrix pm;
-         memcpy(&pm, &ctx->state.vg.path_user_to_surface_matrix,
-                sizeof(struct matrix));
-         matrix_invert(&pm);
-         matrix_mult(&pm, &mat);
-         memcpy(&mat, &pm, sizeof(struct matrix));
-      }
+
+      memcpy(&mat, inv, sizeof(*inv));
 
       map[8]  = mat.m[0]; map[9]  = mat.m[3]; map[10] = mat.m[6]; map[11] = 0.f;
       map[12] = mat.m[1]; map[13] = mat.m[4]; map[14] = mat.m[7]; map[15] = 0.f;
@@ -672,6 +676,34 @@ void paint_resolve_type(struct vg_paint *paint)
    }
 }
 
+VGboolean paint_is_degenerate(struct vg_paint *paint)
+{
+   VGboolean degen;
+   VGfloat *vals;
+
+
+   switch (paint->type) {
+   case VG_PAINT_TYPE_LINEAR_GRADIENT:
+      vals = paint->gradient.linear.coords;
+      /* two points are coincident */
+      degen = (floatsEqual(vals[0], vals[2]) &&
+               floatsEqual(vals[1], vals[3]));
+      break;
+   case VG_PAINT_TYPE_RADIAL_GRADIENT:
+      vals = paint->gradient.radial.vals;
+      /* radius <= 0 */
+      degen = (vals[4] <= 0.0f);
+      break;
+   case VG_PAINT_TYPE_COLOR:
+   case VG_PAINT_TYPE_PATTERN:
+   default:
+      degen = VG_FALSE;
+      break;
+   }
+
+   return degen;
+}
+
 VGint paint_constant_buffer_size(struct vg_paint *paint)
 {
    switch(paint->type) {
@@ -695,6 +727,7 @@ VGint paint_constant_buffer_size(struct vg_paint *paint)
 }
 
 void paint_fill_constant_buffer(struct vg_paint *paint,
+                                const struct matrix *mat,
                                 void *buffer)
 {
    switch(paint->type) {
@@ -702,13 +735,13 @@ void paint_fill_constant_buffer(struct vg_paint *paint,
       paint_color_buffer(paint, buffer);
       break;
    case VG_PAINT_TYPE_LINEAR_GRADIENT:
-      paint_linear_gradient_buffer(paint, buffer);
+      paint_linear_gradient_buffer(paint, mat, buffer);
       break;
    case VG_PAINT_TYPE_RADIAL_GRADIENT:
-      paint_radial_gradient_buffer(paint, buffer);
+      paint_radial_gradient_buffer(paint, mat, buffer);
       break;
    case VG_PAINT_TYPE_PATTERN:
-      paint_pattern_buffer(paint, buffer);
+      paint_pattern_buffer(paint, mat, buffer);
       break;
 
    default:
diff --git a/src/gallium/state_trackers/vega/paint.h b/src/gallium/state_trackers/vega/paint.h
index 012cd3e561..3de3bbe12e 100644
--- a/src/gallium/state_trackers/vega/paint.h
+++ b/src/gallium/state_trackers/vega/paint.h
@@ -110,8 +110,12 @@ VGboolean paint_color_ramp_premultiplied(struct vg_paint *paint);
 VGint paint_bind_samplers(struct vg_paint *paint, struct pipe_sampler_state **samplers,
                           struct pipe_sampler_view **sampler_views);
 
+VGboolean paint_is_degenerate(struct vg_paint *paint);
+
 VGint paint_constant_buffer_size(struct vg_paint *paint);
+
 void paint_fill_constant_buffer(struct vg_paint *paint,
+                                const struct matrix *mat,
                                 void *buffer);
 
 
diff --git a/src/gallium/state_trackers/vega/path.c b/src/gallium/state_trackers/vega/path.c
index 05f8b0d997..d7253befd0 100644
--- a/src/gallium/state_trackers/vega/path.c
+++ b/src/gallium/state_trackers/vega/path.c
@@ -207,13 +207,29 @@ struct path * path_create(VGPathDatatype dt, VGfloat scale, VGfloat bias,
    return path;
 }
 
+static void polygon_array_cleanup(struct polygon_array *polyarray)
+{
+   if (polyarray->array) {
+      VGint i;
+
+      for (i = 0; i < polyarray->array->num_elements; i++) {
+         struct polygon *p = ((struct polygon **) polyarray->array->data)[i];
+         polygon_destroy(p);
+      }
+
+      array_destroy(polyarray->array);
+      polyarray->array = NULL;
+   }
+}
+
 void path_destroy(struct path *p)
 {
    vg_context_remove_object(vg_current_context(), VG_OBJECT_PATH, p);
 
    array_destroy(p->segments);
    array_destroy(p->control_points);
-   array_destroy(p->fill_polys.polygon_array.array);
+
+   polygon_array_cleanup(&p->fill_polys.polygon_array);
 
    if (p->stroked.path)
       path_destroy(p->stroked.path);
@@ -302,7 +318,6 @@ static void convert_path(struct path *p,
    }
 }
 
-
 static void polygon_array_calculate_bounds( struct polygon_array *polyarray )
 {
    struct array *polys = polyarray->array;
@@ -312,7 +327,15 @@ static void polygon_array_calculate_bounds( struct polygon_array *polyarray )
    unsigned i;
 
    assert(polys);
-   assert(polys->num_elements);
+
+   if (!polys->num_elements) {
+      polyarray->min_x = 0.0f;
+      polyarray->min_y = 0.0f;
+      polyarray->max_x = 0.0f;
+      polyarray->max_y = 0.0f;
+      return;
+   }
+
    polygon_bounding_rect((((struct polygon**)polys->data)[0]), bounds);
    min_x = bounds[0];
    min_y = bounds[1];
@@ -353,16 +376,17 @@ static struct polygon_array * path_get_fill_polygons(struct path *p, struct matr
          return &p->fill_polys.polygon_array;
       }
       else {
-         array_destroy( p->fill_polys.polygon_array.array );
-         p->fill_polys.polygon_array.array = NULL;
+         polygon_array_cleanup(&p->fill_polys.polygon_array);
       }
    }
 
-   array = array_create(sizeof(struct array*));
+   /* an array of pointers to polygons */
+   array = array_create(sizeof(struct polygon *));
 
    sx = sy = px = py = ox = oy = 0.f;
 
-   current = polygon_create(32);
+   if (p->num_segments)
+      current = polygon_create(32);
 
    for (i = 0; i < p->num_segments; ++i) {
       VGubyte segment = ((VGubyte*)(p->segments->data))[i];
@@ -1519,10 +1543,11 @@ struct path * path_create_stroke(struct path *p,
    return stroker.base.path;
 }
 
-void path_render(struct path *p, VGbitfield paintModes)
+void path_render(struct path *p, VGbitfield paintModes,
+                 struct matrix *mat)
 {
    struct vg_context *ctx = vg_current_context();
-   struct matrix *mat = &ctx->state.vg.path_user_to_surface_matrix;
+   struct matrix paint_matrix;
 
    vg_validate_state(ctx);
 
@@ -1534,29 +1559,45 @@ void path_render(struct path *p, VGbitfield paintModes)
            mat->m[3], mat->m[4], mat->m[5],
            mat->m[6], mat->m[7], mat->m[8]);
 #endif
-   if (paintModes & VG_FILL_PATH) {
+   if ((paintModes & VG_FILL_PATH) &&
+       vg_get_paint_matrix(ctx,
+                           &ctx->state.vg.fill_paint_to_user_matrix,
+                           mat,
+                           &paint_matrix)) {
       /* First the fill */
+      shader_set_surface_matrix(ctx->shader, mat);
       shader_set_paint(ctx->shader, ctx->state.vg.fill_paint);
+      shader_set_paint_matrix(ctx->shader, &paint_matrix);
       shader_bind(ctx->shader);
-      path_fill(p, mat);
+      path_fill(p);
    }
 
-   if (paintModes & VG_STROKE_PATH){
+   if ((paintModes & VG_STROKE_PATH) &&
+       vg_get_paint_matrix(ctx,
+                           &ctx->state.vg.stroke_paint_to_user_matrix,
+                           mat,
+                           &paint_matrix)) {
       /* 8.7.5: "line width less than or equal to 0 prevents stroking from
        *  taking place."*/
       if (ctx->state.vg.stroke.line_width.f <= 0)
          return;
+      shader_set_surface_matrix(ctx->shader, mat);
       shader_set_paint(ctx->shader, ctx->state.vg.stroke_paint);
+      shader_set_paint_matrix(ctx->shader, &paint_matrix);
       shader_bind(ctx->shader);
       path_stroke(p);
    }
 }
 
-void path_fill(struct path *p, struct matrix *mat)
+void path_fill(struct path *p)
 {
    struct vg_context *ctx = vg_current_context();
+   struct matrix identity;
+
+   matrix_load_identity(&identity);
+
    {
-      struct polygon_array *polygon_array = path_get_fill_polygons(p, mat);
+      struct polygon_array *polygon_array = path_get_fill_polygons(p, &identity);
       struct array *polys = polygon_array->array;
 
       if (!polygon_array || !polys || !polys->num_elements) {
@@ -1569,7 +1610,6 @@ void path_fill(struct path *p, struct matrix *mat)
 void path_stroke(struct path *p)
 {
    struct vg_context *ctx = vg_current_context();
-   struct matrix *mat = &ctx->state.vg.path_user_to_surface_matrix;
    VGFillRule old_fill = ctx->state.vg.fill_rule;
    struct matrix identity;
    struct path *stroke;
@@ -1579,7 +1619,7 @@ void path_stroke(struct path *p)
    if (stroke && !path_is_empty(stroke)) {
       ctx->state.vg.fill_rule = VG_NON_ZERO;
 
-      path_fill(stroke, mat);
+      path_fill(stroke);
 
       ctx->state.vg.fill_rule = old_fill;
    }
diff --git a/src/gallium/state_trackers/vega/path.h b/src/gallium/state_trackers/vega/path.h
index e34538b736..d84b1f083c 100644
--- a/src/gallium/state_trackers/vega/path.h
+++ b/src/gallium/state_trackers/vega/path.h
@@ -104,8 +104,8 @@ VGboolean path_interpolate(struct path *dst,
                            VGfloat amount);
 
 void path_clear(struct path *p, VGbitfield capabilities);
-void path_render(struct path *p, VGbitfield paintModes);
-void path_fill(struct path *p, struct matrix *mat);
+void path_render(struct path *p, VGbitfield paintModes, struct matrix *mat);
+void path_fill(struct path *p);
 void path_stroke(struct path *p);
 
 void path_move_to(struct path *p, float x, float y);
diff --git a/src/gallium/state_trackers/vega/polygon.c b/src/gallium/state_trackers/vega/polygon.c
index ca8831064c..a491de27fa 100644
--- a/src/gallium/state_trackers/vega/polygon.c
+++ b/src/gallium/state_trackers/vega/polygon.c
@@ -244,24 +244,11 @@ VGboolean polygon_is_closed(struct polygon *p)
    return floatsEqual(start[0], end[0]) && floatsEqual(start[1], end[1]);
 }
 
-static void set_blend_for_fill(struct pipe_blend_state *blend)
-{
-   memset(blend, 0, sizeof(struct pipe_blend_state));
-   blend->rt[0].colormask = 0; /*disable colorwrites*/
-
-   blend->rt[0].rgb_src_factor = PIPE_BLENDFACTOR_ONE;
-   blend->rt[0].alpha_src_factor = PIPE_BLENDFACTOR_ONE;
-   blend->rt[0].rgb_dst_factor = PIPE_BLENDFACTOR_INV_SRC_ALPHA;
-   blend->rt[0].alpha_dst_factor = PIPE_BLENDFACTOR_INV_SRC_ALPHA;
-}
-
-static void draw_polygon(struct vg_context *ctx,
-                         struct polygon *poly)
+static void polygon_prepare_buffer(struct vg_context *ctx,
+                                   struct polygon *poly)
 {
    int vert_size;
    struct pipe_context *pipe;
-   struct pipe_vertex_buffer vbuffer;
-   struct pipe_vertex_element velement;
 
    vert_size = poly->num_verts * COMPONENTS * sizeof(float);
 
@@ -281,35 +268,15 @@ static void draw_polygon(struct vg_context *ctx,
 					  PIPE_BIND_VERTEX_BUFFER);
       poly->dirty = VG_FALSE;
    }
-
-
-   /* tell pipe about the vertex buffer */
-   memset(&vbuffer, 0, sizeof(vbuffer));
-   vbuffer.buffer = poly->vbuf;
-   vbuffer.stride = COMPONENTS * sizeof(float);  /* vertex size */
-   vbuffer.buffer_offset = 0;
-   vbuffer.max_index = poly->num_verts - 1;
-   pipe->set_vertex_buffers(pipe, 1, &vbuffer);
-
-   /* tell pipe about the vertex attributes */
-   memset(&velement, 0, sizeof(velement));
-   velement.src_offset = 0;
-   velement.instance_divisor = 0;
-   velement.vertex_buffer_index = 0;
-   velement.src_format = PIPE_FORMAT_R32G32_FLOAT;
-   cso_set_vertex_elements(ctx->cso_context, 1, &velement);
-
-   /* draw */
-   util_draw_arrays(pipe, PIPE_PRIM_TRIANGLE_FAN, 0, (uint) poly->num_verts);
 }
 
 void polygon_fill(struct polygon *poly, struct vg_context *ctx)
 {
-   struct pipe_depth_stencil_alpha_state dsa;
-   struct pipe_stencil_ref sr;
-   struct pipe_blend_state blend;
+   struct pipe_vertex_element velement;
+   struct pipe_vertex_buffer vbuffer;
    VGfloat bounds[4];
    VGfloat min_x, min_y, max_x, max_y;
+
    assert(poly);
    polygon_bounding_rect(poly, bounds);
    min_x = bounds[0];
@@ -322,113 +289,42 @@ void polygon_fill(struct polygon *poly, struct vg_context *ctx)
                 min_x, min_y, max_x, max_y);
 #endif
 
-   set_blend_for_fill(&blend);
-
-   memset(&dsa, 0, sizeof(struct pipe_depth_stencil_alpha_state));
-   memset(&sr, 0, sizeof(struct pipe_stencil_ref));
-   /* only need a fixed 0. Rely on default or move it out at least? */
-   cso_set_stencil_ref(ctx->cso_context, &sr);
-
-   cso_save_blend(ctx->cso_context);
-   cso_save_depth_stencil_alpha(ctx->cso_context);
-
-   dsa.stencil[0].enabled = 1;
-   if (ctx->state.vg.fill_rule == VG_EVEN_ODD) {
-      dsa.stencil[0].writemask = 1;
-      dsa.stencil[0].fail_op = PIPE_STENCIL_OP_KEEP;
-      dsa.stencil[0].zfail_op = PIPE_STENCIL_OP_KEEP;
-      dsa.stencil[0].zpass_op = PIPE_STENCIL_OP_INVERT;
-      dsa.stencil[0].func = PIPE_FUNC_ALWAYS;
-      dsa.stencil[0].valuemask = ~0;
-
-      cso_set_blend(ctx->cso_context, &blend);
-      cso_set_depth_stencil_alpha(ctx->cso_context, &dsa);
-      draw_polygon(ctx, poly);
-   } else if (ctx->state.vg.fill_rule == VG_NON_ZERO) {
-      struct pipe_screen *screen = ctx->pipe->screen;
-
-      if (screen->get_param(screen, PIPE_CAP_TWO_SIDED_STENCIL)) {
-         /* front */
-         dsa.stencil[0].writemask = ~0;
-         dsa.stencil[0].fail_op = PIPE_STENCIL_OP_KEEP;
-         dsa.stencil[0].zfail_op = PIPE_STENCIL_OP_KEEP;
-         dsa.stencil[0].zpass_op = PIPE_STENCIL_OP_INCR_WRAP;
-         dsa.stencil[0].func = PIPE_FUNC_ALWAYS;
-         dsa.stencil[0].valuemask = ~0;
-
-         /* back */
-         dsa.stencil[1].enabled = 1;
-         dsa.stencil[1].writemask = ~0;
-         dsa.stencil[1].fail_op = PIPE_STENCIL_OP_KEEP;
-         dsa.stencil[1].zfail_op = PIPE_STENCIL_OP_KEEP;
-         dsa.stencil[1].zpass_op = PIPE_STENCIL_OP_DECR_WRAP;
-         dsa.stencil[1].func = PIPE_FUNC_ALWAYS;
-         dsa.stencil[1].valuemask = ~0;
-
-         cso_set_blend(ctx->cso_context, &blend);
-         cso_set_depth_stencil_alpha(ctx->cso_context, &dsa);
-         draw_polygon(ctx, poly);
-      } else {
-         struct pipe_rasterizer_state raster;
-
-         memcpy(&raster, &ctx->state.g3d.rasterizer, sizeof(struct pipe_rasterizer_state));
-
-         cso_save_rasterizer(ctx->cso_context);
-         dsa.stencil[0].func = PIPE_FUNC_ALWAYS;
-         dsa.stencil[0].valuemask = ~0;
-
-         raster.cull_face = PIPE_FACE_BACK;
-         dsa.stencil[0].fail_op = PIPE_STENCIL_OP_KEEP;
-         dsa.stencil[0].zfail_op = PIPE_STENCIL_OP_KEEP;
-         dsa.stencil[0].zpass_op = PIPE_STENCIL_OP_INCR_WRAP;
-
-         cso_set_blend(ctx->cso_context, &blend);
-         cso_set_depth_stencil_alpha(ctx->cso_context, &dsa);
-         cso_set_rasterizer(ctx->cso_context, &raster);
-         draw_polygon(ctx, poly);
-
-         raster.cull_face = PIPE_FACE_FRONT;
-         dsa.stencil[0].fail_op = PIPE_STENCIL_OP_KEEP;
-         dsa.stencil[0].zfail_op = PIPE_STENCIL_OP_KEEP;
-         dsa.stencil[0].zpass_op = PIPE_STENCIL_OP_DECR_WRAP;
-         cso_set_depth_stencil_alpha(ctx->cso_context, &dsa);
-         cso_set_rasterizer(ctx->cso_context, &raster);
-         draw_polygon(ctx, poly);
-
-         cso_restore_rasterizer(ctx->cso_context);
-      }
-   }
+   polygon_prepare_buffer(ctx, poly);
+
+   /* tell renderer about the vertex attributes */
+   memset(&velement, 0, sizeof(velement));
+   velement.src_offset = 0;
+   velement.instance_divisor = 0;
+   velement.vertex_buffer_index = 0;
+   velement.src_format = PIPE_FORMAT_R32G32_FLOAT;
+
+   /* tell renderer about the vertex buffer */
+   memset(&vbuffer, 0, sizeof(vbuffer));
+   vbuffer.buffer = poly->vbuf;
+   vbuffer.stride = COMPONENTS * sizeof(float);  /* vertex size */
+   vbuffer.buffer_offset = 0;
+   vbuffer.max_index = poly->num_verts - 1;
+
+   renderer_polygon_stencil_begin(ctx->renderer,
+         &velement, ctx->state.vg.fill_rule, VG_FALSE);
+   renderer_polygon_stencil(ctx->renderer, &vbuffer,
+         PIPE_PRIM_TRIANGLE_FAN, 0, (VGuint) poly->num_verts);
+   renderer_polygon_stencil_end(ctx->renderer);
 
-   /* restore color writes */
-   cso_restore_blend(ctx->cso_context);
-   /* setup stencil ops */
-   dsa.stencil[0].func = PIPE_FUNC_NOTEQUAL;
-   dsa.stencil[0].fail_op = PIPE_STENCIL_OP_REPLACE;
-   dsa.stencil[0].zfail_op = PIPE_STENCIL_OP_REPLACE;
-   dsa.stencil[0].zpass_op = PIPE_STENCIL_OP_REPLACE;
-   dsa.stencil[0].valuemask = dsa.stencil[0].writemask;
-   dsa.stencil[1].enabled = 0;
-   memcpy(&dsa.depth, &ctx->state.g3d.dsa.depth,
-          sizeof(struct pipe_depth_state));
-   cso_set_depth_stencil_alpha(ctx->cso_context, &dsa);
-
-   /* render the quad to propagate the rendering from stencil */
-   renderer_draw_quad(ctx->renderer, min_x, min_y,
-                      max_x, max_y, 0.0f/*depth should be disabled*/);
-
-   cso_restore_depth_stencil_alpha(ctx->cso_context);
+   renderer_polygon_fill_begin(ctx->renderer, VG_FALSE);
+   renderer_polygon_fill(ctx->renderer, min_x, min_y, max_x, max_y);
+   renderer_polygon_fill_end(ctx->renderer);
 }
 
 void polygon_array_fill(struct polygon_array *polyarray, struct vg_context *ctx)
 {
    struct array *polys = polyarray->array;
-   struct pipe_depth_stencil_alpha_state dsa;
-   struct pipe_stencil_ref sr;
-   struct pipe_blend_state blend;
    VGfloat min_x = polyarray->min_x;
    VGfloat min_y = polyarray->min_y;
    VGfloat max_x = polyarray->max_x;
    VGfloat max_y = polyarray->max_y;
+   struct pipe_vertex_element velement;
+   struct pipe_vertex_buffer vbuffer;
    VGint i;
 
 
@@ -438,111 +334,35 @@ void polygon_array_fill(struct polygon_array *polyarray, struct vg_context *ctx)
                 min_x, min_y, max_x, max_y);
 #endif
 
-   set_blend_for_fill(&blend);
-
-   memset(&dsa, 0, sizeof(struct pipe_depth_stencil_alpha_state));
-   memset(&sr, 0, sizeof(struct pipe_stencil_ref));
-   /* only need a fixed 0. Rely on default or move it out at least? */
-   cso_set_stencil_ref(ctx->cso_context, &sr);
-
-   cso_save_blend(ctx->cso_context);
-   cso_save_depth_stencil_alpha(ctx->cso_context);
-
-   dsa.stencil[0].enabled = 1;
-   if (ctx->state.vg.fill_rule == VG_EVEN_ODD) {
-      dsa.stencil[0].writemask = 1;
-      dsa.stencil[0].fail_op = PIPE_STENCIL_OP_KEEP;
-      dsa.stencil[0].zfail_op = PIPE_STENCIL_OP_KEEP;
-      dsa.stencil[0].zpass_op = PIPE_STENCIL_OP_INVERT;
-      dsa.stencil[0].func = PIPE_FUNC_ALWAYS;
-      dsa.stencil[0].valuemask = ~0;
-
-      cso_set_blend(ctx->cso_context, &blend);
-      cso_set_depth_stencil_alpha(ctx->cso_context, &dsa);
-      for (i = 0; i < polys->num_elements; ++i) {
-         struct polygon *poly = (((struct polygon**)polys->data)[i]);
-         draw_polygon(ctx, poly);
-      }
-   } else if (ctx->state.vg.fill_rule == VG_NON_ZERO) {
-      struct pipe_screen *screen = ctx->pipe->screen;
-
-      if (screen->get_param(screen, PIPE_CAP_TWO_SIDED_STENCIL)) {
-         /* front */
-         dsa.stencil[0].writemask = ~0;
-         dsa.stencil[0].fail_op = PIPE_STENCIL_OP_KEEP;
-         dsa.stencil[0].zfail_op = PIPE_STENCIL_OP_KEEP;
-         dsa.stencil[0].zpass_op = PIPE_STENCIL_OP_INCR_WRAP;
-         dsa.stencil[0].func = PIPE_FUNC_ALWAYS;
-         dsa.stencil[0].valuemask = ~0;
-
-         /* back */
-         dsa.stencil[1].enabled = 1;
-         dsa.stencil[1].writemask = ~0;
-         dsa.stencil[1].fail_op = PIPE_STENCIL_OP_KEEP;
-         dsa.stencil[1].zfail_op = PIPE_STENCIL_OP_KEEP;
-         dsa.stencil[1].zpass_op = PIPE_STENCIL_OP_DECR_WRAP;
-         dsa.stencil[1].func = PIPE_FUNC_ALWAYS;
-         dsa.stencil[1].valuemask = ~0;
-
-         cso_set_blend(ctx->cso_context, &blend);
-         cso_set_depth_stencil_alpha(ctx->cso_context, &dsa);
-         for (i = 0; i < polys->num_elements; ++i) {
-            struct polygon *poly = (((struct polygon**)polys->data)[i]);
-            draw_polygon(ctx, poly);
-         }
-      } else {
-         struct pipe_rasterizer_state raster;
-
-         memcpy(&raster, &ctx->state.g3d.rasterizer, sizeof(struct pipe_rasterizer_state));
-
-         cso_save_rasterizer(ctx->cso_context);
-         dsa.stencil[0].func = PIPE_FUNC_ALWAYS;
-         dsa.stencil[0].valuemask = ~0;
-
-         raster.cull_face = PIPE_FACE_BACK;
-         dsa.stencil[0].fail_op = PIPE_STENCIL_OP_KEEP;
-         dsa.stencil[0].zfail_op = PIPE_STENCIL_OP_KEEP;
-         dsa.stencil[0].zpass_op = PIPE_STENCIL_OP_INCR_WRAP;
-
-         cso_set_blend(ctx->cso_context, &blend);
-         cso_set_depth_stencil_alpha(ctx->cso_context, &dsa);
-         cso_set_rasterizer(ctx->cso_context, &raster);
-         for (i = 0; i < polys->num_elements; ++i) {
-            struct polygon *poly = (((struct polygon**)polys->data)[i]);
-            draw_polygon(ctx, poly);
-         }
-
-         raster.cull_face = PIPE_FACE_FRONT;
-         dsa.stencil[0].fail_op = PIPE_STENCIL_OP_KEEP;
-         dsa.stencil[0].zfail_op = PIPE_STENCIL_OP_KEEP;
-         dsa.stencil[0].zpass_op = PIPE_STENCIL_OP_DECR_WRAP;
-         cso_set_depth_stencil_alpha(ctx->cso_context, &dsa);
-         cso_set_rasterizer(ctx->cso_context, &raster);
-         for (i = 0; i < polys->num_elements; ++i) {
-            struct polygon *poly = (((struct polygon**)polys->data)[i]);
-            draw_polygon(ctx, poly);
-         }
-
-         cso_restore_rasterizer(ctx->cso_context);
-      }
+   /* tell renderer about the vertex attributes */
+   memset(&velement, 0, sizeof(velement));
+   velement.src_offset = 0;
+   velement.instance_divisor = 0;
+   velement.vertex_buffer_index = 0;
+   velement.src_format = PIPE_FORMAT_R32G32_FLOAT;
+
+   /* tell renderer about the vertex buffer */
+   memset(&vbuffer, 0, sizeof(vbuffer));
+   vbuffer.stride = COMPONENTS * sizeof(float);  /* vertex size */
+   vbuffer.buffer_offset = 0;
+
+   /* prepare the stencil buffer */
+   renderer_polygon_stencil_begin(ctx->renderer,
+         &velement, ctx->state.vg.fill_rule, VG_FALSE);
+   for (i = 0; i < polys->num_elements; ++i) {
+      struct polygon *poly = (((struct polygon**)polys->data)[i]);
+
+      polygon_prepare_buffer(ctx, poly);
+      vbuffer.buffer = poly->vbuf;
+      vbuffer.max_index = poly->num_verts - 1;
+
+      renderer_polygon_stencil(ctx->renderer, &vbuffer,
+            PIPE_PRIM_TRIANGLE_FAN, 0, (VGuint) poly->num_verts);
    }
+   renderer_polygon_stencil_end(ctx->renderer);
 
-   /* restore color writes */
-   cso_restore_blend(ctx->cso_context);
-   /* setup stencil ops */
-   dsa.stencil[0].func = PIPE_FUNC_NOTEQUAL;
-   dsa.stencil[0].fail_op = PIPE_STENCIL_OP_REPLACE;
-   dsa.stencil[0].zfail_op = PIPE_STENCIL_OP_REPLACE;
-   dsa.stencil[0].zpass_op = PIPE_STENCIL_OP_REPLACE;
-   dsa.stencil[0].valuemask = dsa.stencil[0].writemask;
-   dsa.stencil[1].enabled = 0;
-   memcpy(&dsa.depth, &ctx->state.g3d.dsa.depth,
-          sizeof(struct pipe_depth_state));
-   cso_set_depth_stencil_alpha(ctx->cso_context, &dsa);
-
-   /* render the quad to propagate the rendering from stencil */
-   renderer_draw_quad(ctx->renderer, min_x, min_y,
-                      max_x, max_y, 0.0f/*depth should be disabled*/);
-
-   cso_restore_depth_stencil_alpha(ctx->cso_context);
+   /* fill it */
+   renderer_polygon_fill_begin(ctx->renderer, VG_FALSE);
+   renderer_polygon_fill(ctx->renderer, min_x, min_y, max_x, max_y);
+   renderer_polygon_fill_end(ctx->renderer);
 }
diff --git a/src/gallium/state_trackers/vega/renderer.c b/src/gallium/state_trackers/vega/renderer.c
index c6f5f7a05b..7871c516c4 100644
--- a/src/gallium/state_trackers/vega/renderer.c
+++ b/src/gallium/state_trackers/vega/renderer.c
@@ -1,6 +1,7 @@
 /**************************************************************************
  *
  * Copyright 2009 VMware, Inc.  All Rights Reserved.
+ * Copyright 2010 LunarG, Inc.  All Rights Reserved.
  *
  * Permission is hereby granted, free of charge, to any person obtaining a
  * copy of this software and associated documentation files (the
@@ -27,6 +28,7 @@
 #include "renderer.h"
 
 #include "vg_context.h"
+#include "image.h"
 
 #include "pipe/p_context.h"
 #include "pipe/p_state.h"
@@ -38,364 +40,1382 @@
 #include "util/u_simple_shaders.h"
 #include "util/u_memory.h"
 #include "util/u_sampler.h"
+#include "util/u_surface.h"
+#include "util/u_math.h"
+#include "util/u_format.h"
 
 #include "cso_cache/cso_context.h"
+#include "tgsi/tgsi_ureg.h"
+
+typedef enum {
+   RENDERER_STATE_INIT,
+   RENDERER_STATE_COPY,
+   RENDERER_STATE_DRAWTEX,
+   RENDERER_STATE_SCISSOR,
+   RENDERER_STATE_CLEAR,
+   RENDERER_STATE_FILTER,
+   RENDERER_STATE_POLYGON_STENCIL,
+   RENDERER_STATE_POLYGON_FILL,
+   NUM_RENDERER_STATES
+} RendererState;
+
+typedef enum {
+   RENDERER_VS_PLAIN,
+   RENDERER_VS_COLOR,
+   RENDERER_VS_TEXTURE,
+   NUM_RENDERER_VS
+} RendererVs;
+
+typedef enum {
+   RENDERER_FS_COLOR,
+   RENDERER_FS_TEXTURE,
+   RENDERER_FS_SCISSOR,
+   RENDERER_FS_WHITE,
+   NUM_RENDERER_FS
+} RendererFs;
 
 struct renderer {
    struct pipe_context *pipe;
-   struct vg_context *owner;
-
    struct cso_context *cso;
 
-   void *fs;
+   VGbitfield dirty;
+   struct {
+      struct pipe_rasterizer_state rasterizer;
+      struct pipe_depth_stencil_alpha_state dsa;
+      struct pipe_framebuffer_state fb;
+   } g3d;
+   struct matrix projection;
+
+   struct matrix mvp;
+   struct pipe_resource *vs_cbuf;
+
+   struct pipe_resource *fs_cbuf;
+   VGfloat fs_cbuf_data[32];
+   VGint fs_cbuf_len;
 
+   struct pipe_vertex_element velems[2];
    VGfloat vertices[4][2][4];
+
+   void *cached_vs[NUM_RENDERER_VS];
+   void *cached_fs[NUM_RENDERER_FS];
+
+   RendererState state;
+
+   /* state data */
+   union {
+      struct {
+         VGint tex_width;
+         VGint tex_height;
+      } copy;
+
+      struct {
+         VGint tex_width;
+         VGint tex_height;
+      } drawtex;
+
+      struct {
+         VGboolean restore_dsa;
+      } scissor;
+
+      struct {
+         VGboolean use_sampler;
+         VGint tex_width, tex_height;
+      } filter;
+
+      struct {
+         struct pipe_depth_stencil_alpha_state dsa;
+         VGboolean manual_two_sides;
+         VGboolean restore_dsa;
+      } polygon_stencil;
+   } u;
 };
 
-static void setup_shaders(struct renderer *ctx)
+/**
+ * Return VG_TRUE if the renderer can use the resource as the asked bindings.
+ */
+static VGboolean renderer_can_support(struct renderer *renderer,
+                                      struct pipe_resource *res,
+                                      unsigned bindings)
 {
-   struct pipe_context *pipe = ctx->pipe;
-   /* fragment shader */
-   ctx->fs = util_make_fragment_tex_shader(pipe, TGSI_TEXTURE_2D,
-                                           TGSI_INTERPOLATE_LINEAR);
-}
-
-static struct pipe_resource *
-setup_vertex_data(struct renderer *ctx,
-                  float x0, float y0, float x1, float y1, float z)
-{
-   ctx->vertices[0][0][0] = x0;
-   ctx->vertices[0][0][1] = y0;
-   ctx->vertices[0][0][2] = z;
-   ctx->vertices[0][1][0] = 0.0f; /*s*/
-   ctx->vertices[0][1][1] = 0.0f; /*t*/
-
-   ctx->vertices[1][0][0] = x1;
-   ctx->vertices[1][0][1] = y0;
-   ctx->vertices[1][0][2] = z;
-   ctx->vertices[1][1][0] = 1.0f; /*s*/
-   ctx->vertices[1][1][1] = 0.0f; /*t*/
-
-   ctx->vertices[2][0][0] = x1;
-   ctx->vertices[2][0][1] = y1;
-   ctx->vertices[2][0][2] = z;
-   ctx->vertices[2][1][0] = 1.0f;
-   ctx->vertices[2][1][1] = 1.0f;
-
-   ctx->vertices[3][0][0] = x0;
-   ctx->vertices[3][0][1] = y1;
-   ctx->vertices[3][0][2] = z;
-   ctx->vertices[3][1][0] = 0.0f;
-   ctx->vertices[3][1][1] = 1.0f;
-
-   return pipe_user_buffer_create( ctx->pipe->screen,
-                                   ctx->vertices,
-                                   sizeof(ctx->vertices),
-				   PIPE_BIND_VERTEX_BUFFER);
-}
-
-static struct pipe_resource *
-setup_vertex_data_tex(struct renderer *ctx,
-                      float x0, float y0, float x1, float y1,
-                      float s0, float t0, float s1, float t1,
-                      float z)
-{
-   ctx->vertices[0][0][0] = x0;
-   ctx->vertices[0][0][1] = y0;
-   ctx->vertices[0][0][2] = z;
-   ctx->vertices[0][1][0] = s0; /*s*/
-   ctx->vertices[0][1][1] = t0; /*t*/
-
-   ctx->vertices[1][0][0] = x1;
-   ctx->vertices[1][0][1] = y0;
-   ctx->vertices[1][0][2] = z;
-   ctx->vertices[1][1][0] = s1; /*s*/
-   ctx->vertices[1][1][1] = t0; /*t*/
-
-   ctx->vertices[2][0][0] = x1;
-   ctx->vertices[2][0][1] = y1;
-   ctx->vertices[2][0][2] = z;
-   ctx->vertices[2][1][0] = s1;
-   ctx->vertices[2][1][1] = t1;
-
-   ctx->vertices[3][0][0] = x0;
-   ctx->vertices[3][0][1] = y1;
-   ctx->vertices[3][0][2] = z;
-   ctx->vertices[3][1][0] = s0;
-   ctx->vertices[3][1][1] = t1;
-
-   return pipe_user_buffer_create( ctx->pipe->screen,
-                                   ctx->vertices,
-                                   sizeof(ctx->vertices),
-				   PIPE_BIND_VERTEX_BUFFER);
-}
-
-
-static struct pipe_resource *
-setup_vertex_data_qtex(struct renderer *ctx,
-                       float x0, float y0, float x1, float y1,
-                       float x2, float y2, float x3, float y3,
-                       float s0, float t0, float s1, float t1,
-                       float z)
-{
-   ctx->vertices[0][0][0] = x0;
-   ctx->vertices[0][0][1] = y0;
-   ctx->vertices[0][0][2] = z;
-   ctx->vertices[0][1][0] = s0; /*s*/
-   ctx->vertices[0][1][1] = t0; /*t*/
-
-   ctx->vertices[1][0][0] = x1;
-   ctx->vertices[1][0][1] = y1;
-   ctx->vertices[1][0][2] = z;
-   ctx->vertices[1][1][0] = s1; /*s*/
-   ctx->vertices[1][1][1] = t0; /*t*/
-
-   ctx->vertices[2][0][0] = x2;
-   ctx->vertices[2][0][1] = y2;
-   ctx->vertices[2][0][2] = z;
-   ctx->vertices[2][1][0] = s1;
-   ctx->vertices[2][1][1] = t1;
-
-   ctx->vertices[3][0][0] = x3;
-   ctx->vertices[3][0][1] = y3;
-   ctx->vertices[3][0][2] = z;
-   ctx->vertices[3][1][0] = s0;
-   ctx->vertices[3][1][1] = t1;
-
-   return pipe_user_buffer_create( ctx->pipe->screen,
-                                   ctx->vertices,
-                                   sizeof(ctx->vertices),
-				   PIPE_BIND_VERTEX_BUFFER);
+   struct pipe_screen *screen = renderer->pipe->screen;
+
+   return screen->is_format_supported(screen,
+         res->format, res->target, 0, bindings, 0);
 }
 
-struct renderer * renderer_create(struct vg_context *owner)
+/**
+ * Set the model-view-projection matrix used by vertex shaders.
+ */
+static void renderer_set_mvp(struct renderer *renderer,
+                             const struct matrix *mvp)
 {
+   struct matrix *cur = &renderer->mvp;
+   struct pipe_resource *cbuf;
+   VGfloat consts[3][4];
    VGint i;
-   struct renderer *renderer = CALLOC_STRUCT(renderer);
 
-   if (!renderer)
+   /* projection only */
+   if (!mvp)
+      mvp = &renderer->projection;
+
+   /* re-upload only if necessary */
+   if (memcmp(cur, mvp, sizeof(*mvp)) == 0)
+      return;
+
+   /* 3x3 matrix to 3 constant vectors (no Z) */
+   for (i = 0; i < 3; i++) {
+      consts[i][0] = mvp->m[i + 0];
+      consts[i][1] = mvp->m[i + 3];
+      consts[i][2] = 0.0f;
+      consts[i][3] = mvp->m[i + 6];
+   }
+
+   cbuf = renderer->vs_cbuf;
+   pipe_resource_reference(&cbuf, NULL);
+   cbuf = pipe_buffer_create(renderer->pipe->screen,
+                             PIPE_BIND_CONSTANT_BUFFER,
+                             sizeof(consts));
+   if (cbuf) {
+      pipe_buffer_write(renderer->pipe, cbuf,
+            0, sizeof(consts), consts);
+   }
+   renderer->pipe->set_constant_buffer(renderer->pipe,
+         PIPE_SHADER_VERTEX, 0, cbuf);
+
+   memcpy(cur, mvp, sizeof(*mvp));
+   renderer->vs_cbuf = cbuf;
+}
+
+/**
+ * Create a simple vertex shader that passes through position and the given
+ * attribute.
+ */
+static void *create_passthrough_vs(struct pipe_context *pipe, int semantic_name)
+{
+   struct ureg_program *ureg;
+   struct ureg_src src[2], constants[3];
+   struct ureg_dst dst[2], tmp;
+   int i;
+
+   ureg = ureg_create(TGSI_PROCESSOR_VERTEX);
+   if (!ureg)
       return NULL;
 
-   renderer->owner = owner;
-   renderer->pipe = owner->pipe;
-   renderer->cso = owner->cso_context;
+   /* position is in user coordinates */
+   src[0] = ureg_DECL_vs_input(ureg, 0);
+   dst[0] = ureg_DECL_output(ureg, TGSI_SEMANTIC_POSITION, 0);
+   tmp = ureg_DECL_temporary(ureg);
+   for (i = 0; i < Elements(constants); i++)
+      constants[i] = ureg_DECL_constant(ureg, i);
+
+   /* transform to clipped coordinates */
+   ureg_DP4(ureg, ureg_writemask(tmp, TGSI_WRITEMASK_X), src[0], constants[0]);
+   ureg_DP4(ureg, ureg_writemask(tmp, TGSI_WRITEMASK_Y), src[0], constants[1]);
+   ureg_MOV(ureg, ureg_writemask(tmp, TGSI_WRITEMASK_Z), src[0]);
+   ureg_DP4(ureg, ureg_writemask(tmp, TGSI_WRITEMASK_W), src[0], constants[2]);
+   ureg_MOV(ureg, dst[0], ureg_src(tmp));
+
+   if (semantic_name >= 0) {
+      src[1] = ureg_DECL_vs_input(ureg, 1);
+      dst[1] = ureg_DECL_output(ureg, semantic_name, 0);
+      ureg_MOV(ureg, dst[1], src[1]);
+   }
 
-   setup_shaders(renderer);
+   ureg_END(ureg);
 
-   /* init vertex data that doesn't change */
-   for (i = 0; i < 4; i++) {
-      renderer->vertices[i][0][3] = 1.0f; /* w */
-      renderer->vertices[i][1][2] = 0.0f; /* r */
-      renderer->vertices[i][1][3] = 1.0f; /* q */
+   return ureg_create_shader_and_destroy(ureg, pipe);
+}
+
+/**
+ * Set renderer vertex shader.
+ *
+ * This function modifies vertex_shader state.
+ */
+static void renderer_set_vs(struct renderer *r, RendererVs id)
+{
+   /* create as needed */
+   if (!r->cached_vs[id]) {
+      int semantic_name = -1;
+
+      switch (id) {
+      case RENDERER_VS_PLAIN:
+         break;
+      case RENDERER_VS_COLOR:
+         semantic_name = TGSI_SEMANTIC_COLOR;
+         break;
+      case RENDERER_VS_TEXTURE:
+         semantic_name = TGSI_SEMANTIC_GENERIC;
+         break;
+      default:
+         assert(!"Unknown renderer vs id");
+         break;
+      }
+
+      r->cached_vs[id] = create_passthrough_vs(r->pipe, semantic_name);
    }
 
-   return renderer;
+   cso_set_vertex_shader_handle(r->cso, r->cached_vs[id]);
 }
 
-void renderer_destroy(struct renderer *ctx)
+/**
+ * Create a simple fragment shader that sets the depth to 0.0f.
+ */
+static void *create_scissor_fs(struct pipe_context *pipe)
+{
+   struct ureg_program *ureg;
+   struct ureg_dst out;
+   struct ureg_src imm;
+
+   ureg = ureg_create(TGSI_PROCESSOR_FRAGMENT);
+   out = ureg_DECL_output(ureg, TGSI_SEMANTIC_POSITION, 0);
+   imm = ureg_imm4f(ureg, 0.0f, 0.0f, 0.0f, 0.0f);
+
+   ureg_MOV(ureg, ureg_writemask(out, TGSI_WRITEMASK_Z), imm);
+   ureg_END(ureg);
+
+   return ureg_create_shader_and_destroy(ureg, pipe);
+}
+
+/**
+ * Create a simple fragment shader that sets the color to white.
+ */
+static void *create_white_fs(struct pipe_context *pipe)
+{
+   struct ureg_program *ureg;
+   struct ureg_dst out;
+   struct ureg_src imm;
+
+   ureg = ureg_create(TGSI_PROCESSOR_FRAGMENT);
+   out = ureg_DECL_output(ureg, TGSI_SEMANTIC_COLOR, 0);
+   imm = ureg_imm4f(ureg, 1.0f, 1.0f, 1.0f, 1.0f);
+
+   ureg_MOV(ureg, out, imm);
+   ureg_END(ureg);
+
+   return ureg_create_shader_and_destroy(ureg, pipe);
+}
+
+/**
+ * Set renderer fragment shader.
+ *
+ * This function modifies fragment_shader state.
+ */
+static void renderer_set_fs(struct renderer *r, RendererFs id)
 {
-#if 0
-   if (ctx->fs) {
-      cso_delete_fragment_shader(ctx->cso, ctx->fs);
-      ctx->fs = NULL;
+   /* create as needed */
+   if (!r->cached_fs[id]) {
+      void *fs = NULL;
+
+      switch (id) {
+      case RENDERER_FS_COLOR:
+         fs = util_make_fragment_passthrough_shader(r->pipe);
+         break;
+      case RENDERER_FS_TEXTURE:
+         fs = util_make_fragment_tex_shader(r->pipe,
+               TGSI_TEXTURE_2D, TGSI_INTERPOLATE_LINEAR);
+         break;
+      case RENDERER_FS_SCISSOR:
+         fs = create_scissor_fs(r->pipe);
+         break;
+      case RENDERER_FS_WHITE:
+         fs = create_white_fs(r->pipe);
+         break;
+      default:
+         assert(!"Unknown renderer fs id");
+         break;
+      }
+
+      r->cached_fs[id] = fs;
    }
-#endif
-   FREE(ctx);
+
+   cso_set_fragment_shader_handle(r->cso, r->cached_fs[id]);
 }
 
-void renderer_draw_quad(struct renderer *r,
-                        VGfloat x1, VGfloat y1,
-                        VGfloat x2, VGfloat y2,
-                        VGfloat depth)
+typedef enum {
+   VEGA_Y0_TOP,
+   VEGA_Y0_BOTTOM
+} VegaOrientation;
+
+static void vg_set_viewport(struct renderer *r,
+                            VegaOrientation orientation)
 {
-   struct pipe_resource *buf;
+   const struct pipe_framebuffer_state *fb = &r->g3d.fb;
+   struct pipe_viewport_state viewport;
+   VGfloat y_scale = (orientation == VEGA_Y0_BOTTOM) ? -2.f : 2.f;
+
+   viewport.scale[0] =  fb->width / 2.f;
+   viewport.scale[1] =  fb->height / y_scale;
+   viewport.scale[2] =  1.0;
+   viewport.scale[3] =  1.0;
+   viewport.translate[0] = fb->width / 2.f;
+   viewport.translate[1] = fb->height / 2.f;
+   viewport.translate[2] = 0.0;
+   viewport.translate[3] = 0.0;
+
+   cso_set_viewport(r->cso, &viewport);
+}
+
+/**
+ * Set renderer target.
+ *
+ * This function modifies framebuffer and viewport states.
+ */
+static void renderer_set_target(struct renderer *r,
+                                struct pipe_surface *cbuf,
+                                struct pipe_surface *zsbuf,
+                                VGboolean y0_top)
+{
+   struct pipe_framebuffer_state fb;
+
+   memset(&fb, 0, sizeof(fb));
+   fb.width = cbuf->width;
+   fb.height = cbuf->height;
+   fb.cbufs[0] = cbuf;
+   fb.nr_cbufs = 1;
+   fb.zsbuf = zsbuf;
+   cso_set_framebuffer(r->cso, &fb);
+
+   vg_set_viewport(r, (y0_top) ? VEGA_Y0_TOP : VEGA_Y0_BOTTOM);
+}
+
+/**
+ * Set renderer blend state.  Blending is disabled.
+ *
+ * This function modifies blend state.
+ */
+static void renderer_set_blend(struct renderer *r,
+                               VGbitfield channel_mask)
+{
+   struct pipe_blend_state blend;
+
+   memset(&blend, 0, sizeof(blend));
+
+   blend.rt[0].rgb_src_factor = PIPE_BLENDFACTOR_ONE;
+   blend.rt[0].alpha_src_factor = PIPE_BLENDFACTOR_ONE;
+   blend.rt[0].rgb_dst_factor = PIPE_BLENDFACTOR_ZERO;
+   blend.rt[0].alpha_dst_factor = PIPE_BLENDFACTOR_ZERO;
+
+   if (channel_mask & VG_RED)
+      blend.rt[0].colormask |= PIPE_MASK_R;
+   if (channel_mask & VG_GREEN)
+      blend.rt[0].colormask |= PIPE_MASK_G;
+   if (channel_mask & VG_BLUE)
+      blend.rt[0].colormask |= PIPE_MASK_B;
+   if (channel_mask & VG_ALPHA)
+      blend.rt[0].colormask |= PIPE_MASK_A;
+
+   cso_set_blend(r->cso, &blend);
+}
+
+/**
+ * Set renderer sampler and view states.
+ *
+ * This function modifies samplers and fragment_sampler_views states.
+ */
+static void renderer_set_samplers(struct renderer *r,
+                                  uint num_views,
+                                  struct pipe_sampler_view **views)
+{
+   struct pipe_sampler_state sampler;
+   unsigned tex_filter = PIPE_TEX_FILTER_NEAREST;
+   unsigned tex_wrap = PIPE_TEX_WRAP_CLAMP_TO_EDGE;
+   uint i;
+
+   memset(&sampler, 0, sizeof(sampler));
+
+   sampler.min_img_filter = tex_filter;
+   sampler.mag_img_filter = tex_filter;
+   sampler.min_mip_filter = PIPE_TEX_MIPFILTER_NONE;
+
+   sampler.wrap_s = tex_wrap;
+   sampler.wrap_t = tex_wrap;
+   sampler.wrap_r = PIPE_TEX_WRAP_CLAMP_TO_EDGE;
+
+   sampler.normalized_coords = 1;
+
+   /* set samplers */
+   for (i = 0; i < num_views; i++)
+      cso_single_sampler(r->cso, i, &sampler);
+   cso_single_sampler_done(r->cso);
+
+   /* set views */
+   cso_set_fragment_sampler_views(r->cso, num_views, views);
+}
+
+/**
+ * Set custom renderer fragment shader, and optionally set samplers and views
+ * and upload the fragment constant buffer.
+ *
+ * This function modifies fragment_shader, samplers and fragment_sampler_views
+ * states.
+ */
+static void renderer_set_custom_fs(struct renderer *renderer,
+                                   void *fs,
+                                   const struct pipe_sampler_state **samplers,
+                                   struct pipe_sampler_view **views,
+                                   VGint num_samplers,
+                                   const void *const_buffer,
+                                   VGint const_buffer_len)
+{
+   cso_set_fragment_shader_handle(renderer->cso, fs);
+
+   /* set samplers and views */
+   if (num_samplers) {
+      cso_set_samplers(renderer->cso, num_samplers, samplers);
+      cso_set_fragment_sampler_views(renderer->cso, num_samplers, views);
+   }
+
+   /* upload fs constant buffer */
+   if (const_buffer_len) {
+      struct pipe_resource *cbuf = renderer->fs_cbuf;
+
+      if (!cbuf || renderer->fs_cbuf_len != const_buffer_len ||
+          memcmp(renderer->fs_cbuf_data, const_buffer, const_buffer_len)) {
+         pipe_resource_reference(&cbuf, NULL);
+
+         cbuf = pipe_buffer_create(renderer->pipe->screen,
+               PIPE_BIND_CONSTANT_BUFFER, const_buffer_len);
+         pipe_buffer_write(renderer->pipe, cbuf, 0,
+               const_buffer_len, const_buffer);
+         renderer->pipe->set_constant_buffer(renderer->pipe,
+               PIPE_SHADER_FRAGMENT, 0, cbuf);
+
+         renderer->fs_cbuf = cbuf;
+         if (const_buffer_len <= sizeof(renderer->fs_cbuf_data)) {
+            memcpy(renderer->fs_cbuf_data, const_buffer, const_buffer_len);
+            renderer->fs_cbuf_len = const_buffer_len;
+         }
+         else {
+            renderer->fs_cbuf_len = 0;
+         }
+      }
+   }
+}
+
+/**
+ * Setup renderer quad position.
+ */
+static void renderer_quad_pos(struct renderer *r,
+                              VGfloat x0, VGfloat y0,
+                              VGfloat x1, VGfloat y1,
+                              VGboolean scissor)
+{
+   VGfloat z;
+
+   /* the depth test is used for scissoring */
+   z = (scissor) ? 0.0f : 1.0f;
+
+   /* positions */
+   r->vertices[0][0][0] = x0;
+   r->vertices[0][0][1] = y0;
+   r->vertices[0][0][2] = z;
 
-   buf = setup_vertex_data(r, x1, y1, x2, y2, depth);
+   r->vertices[1][0][0] = x1;
+   r->vertices[1][0][1] = y0;
+   r->vertices[1][0][2] = z;
 
+   r->vertices[2][0][0] = x1;
+   r->vertices[2][0][1] = y1;
+   r->vertices[2][0][2] = z;
+
+   r->vertices[3][0][0] = x0;
+   r->vertices[3][0][1] = y1;
+   r->vertices[3][0][2] = z;
+}
+
+/**
+ * Setup renderer quad texture coordinates.
+ */
+static void renderer_quad_texcoord(struct renderer *r,
+                                   VGfloat x0, VGfloat y0,
+                                   VGfloat x1, VGfloat y1,
+                                   VGint tex_width, VGint tex_height)
+{
+   VGfloat s0, t0, s1, t1, r0, q0;
+   VGint i;
+
+   s0 = x0 / tex_width;
+   s1 = x1 / tex_width;
+   t0 = y0 / tex_height;
+   t1 = y1 / tex_height;
+   r0 = 0.0f;
+   q0 = 1.0f;
+
+   /* texcoords */
+   r->vertices[0][1][0] = s0;
+   r->vertices[0][1][1] = t0;
+
+   r->vertices[1][1][0] = s1;
+   r->vertices[1][1][1] = t0;
+
+   r->vertices[2][1][0] = s1;
+   r->vertices[2][1][1] = t1;
+
+   r->vertices[3][1][0] = s0;
+   r->vertices[3][1][1] = t1;
+
+   for (i = 0; i < 4; i++) {
+      r->vertices[i][1][2] = r0;
+      r->vertices[i][1][3] = q0;
+   }
+}
+
+/**
+ * Draw renderer quad.
+ */
+static void renderer_quad_draw(struct renderer *r)
+{
+   struct pipe_resource *buf;
+
+   buf = pipe_user_buffer_create(r->pipe->screen,
+                                 r->vertices,
+                                 sizeof(r->vertices),
+                                 PIPE_BIND_VERTEX_BUFFER);
    if (buf) {
-      cso_set_vertex_elements(r->cso, 2, r->owner->velems);
       util_draw_vertex_buffer(r->pipe, buf, 0,
                               PIPE_PRIM_TRIANGLE_FAN,
-                              4,  /* verts */
-                              2); /* attribs/vert */
+                              Elements(r->vertices),     /* verts */
+                              Elements(r->vertices[0])); /* attribs/vert */
 
-      pipe_resource_reference( &buf,
-                             NULL );
+      pipe_resource_reference(&buf, NULL);
    }
 }
 
-void renderer_draw_texture(struct renderer *r,
-                           struct pipe_resource *tex,
-                           VGfloat x1offset, VGfloat y1offset,
-                           VGfloat x2offset, VGfloat y2offset,
-                           VGfloat x1, VGfloat y1,
-                           VGfloat x2, VGfloat y2)
+/**
+ * Prepare the renderer for copying.
+ */
+VGboolean renderer_copy_begin(struct renderer *renderer,
+                              struct pipe_surface *dst,
+                              VGboolean y0_top,
+                              struct pipe_sampler_view *src)
 {
-   struct pipe_context *pipe = r->pipe;
-   struct pipe_resource *buf;
-   VGfloat s0, t0, s1, t1;
+   assert(renderer->state == RENDERER_STATE_INIT);
 
-   assert(tex->width0 != 0);
-   assert(tex->height0 != 0);
+   /* sanity check */
+   if (!renderer_can_support(renderer,
+            dst->texture, PIPE_BIND_RENDER_TARGET) ||
+       !renderer_can_support(renderer,
+          src->texture, PIPE_BIND_SAMPLER_VIEW))
+      return VG_FALSE;
 
-   s0 = x1offset / tex->width0;
-   s1 = x2offset / tex->width0;
-   t0 = y1offset / tex->height0;
-   t1 = y2offset / tex->height0;
+   cso_save_framebuffer(renderer->cso);
+   cso_save_viewport(renderer->cso);
+   cso_save_blend(renderer->cso);
+   cso_save_samplers(renderer->cso);
+   cso_save_fragment_sampler_views(renderer->cso);
+   cso_save_fragment_shader(renderer->cso);
+   cso_save_vertex_shader(renderer->cso);
 
-   cso_save_vertex_shader(r->cso);
-   /* shaders */
-   cso_set_vertex_shader_handle(r->cso, vg_texture_vs(r->owner));
+   renderer_set_target(renderer, dst, NULL, y0_top);
 
-   /* draw quad */
-   buf = setup_vertex_data_tex(r, x1, y1, x2, y2,
-                               s0, t0, s1, t1, 0.0f);
+   renderer_set_blend(renderer, ~0);
+   renderer_set_samplers(renderer, 1, &src);
 
-   if (buf) {
-      cso_set_vertex_elements(r->cso, 2, r->owner->velems);
-      util_draw_vertex_buffer(pipe, buf, 0,
-                           PIPE_PRIM_TRIANGLE_FAN,
-                           4,  /* verts */
-                           2); /* attribs/vert */
-
-      pipe_resource_reference( &buf,
-                             NULL );
+   renderer_set_fs(renderer, RENDERER_FS_TEXTURE);
+   renderer_set_vs(renderer, RENDERER_VS_TEXTURE);
+
+   renderer_set_mvp(renderer, NULL);
+
+   /* remember the texture size */
+   renderer->u.copy.tex_width = src->texture->width0;
+   renderer->u.copy.tex_height = src->texture->height0;
+   renderer->state = RENDERER_STATE_COPY;
+
+   return VG_TRUE;
+}
+
+/**
+ * Draw into the destination rectangle given by (x, y, w, h).  The texture is
+ * sampled from within the rectangle given by (sx, sy, sw, sh).
+ *
+ * The coordinates are in surface coordinates.
+ */
+void renderer_copy(struct renderer *renderer,
+                   VGint x, VGint y, VGint w, VGint h,
+                   VGint sx, VGint sy, VGint sw, VGint sh)
+{
+   assert(renderer->state == RENDERER_STATE_COPY);
+
+   /* there is no depth buffer for scissoring anyway */
+   renderer_quad_pos(renderer, x, y, x + w, y + h, VG_FALSE);
+   renderer_quad_texcoord(renderer, sx, sy, sx + sw, sy + sh,
+         renderer->u.copy.tex_width,
+         renderer->u.copy.tex_height);
+
+   renderer_quad_draw(renderer);
+}
+
+/**
+ * End copying and restore the states.
+ */
+void renderer_copy_end(struct renderer *renderer)
+{
+   assert(renderer->state == RENDERER_STATE_COPY);
+
+   cso_restore_framebuffer(renderer->cso);
+   cso_restore_viewport(renderer->cso);
+   cso_restore_blend(renderer->cso);
+   cso_restore_samplers(renderer->cso);
+   cso_restore_fragment_sampler_views(renderer->cso);
+   cso_restore_fragment_shader(renderer->cso);
+   cso_restore_vertex_shader(renderer->cso);
+
+   renderer->state = RENDERER_STATE_INIT;
+}
+
+/**
+ * Prepare the renderer for textured drawing.
+ */
+VGboolean renderer_drawtex_begin(struct renderer *renderer,
+                                 struct pipe_sampler_view *src)
+{
+   assert(renderer->state == RENDERER_STATE_INIT);
+
+   if (!renderer_can_support(renderer, src->texture, PIPE_BIND_SAMPLER_VIEW))
+      return VG_FALSE;
+
+   cso_save_blend(renderer->cso);
+   cso_save_samplers(renderer->cso);
+   cso_save_fragment_sampler_views(renderer->cso);
+   cso_save_fragment_shader(renderer->cso);
+   cso_save_vertex_shader(renderer->cso);
+
+   renderer_set_blend(renderer, ~0);
+
+   renderer_set_samplers(renderer, 1, &src);
+
+   renderer_set_fs(renderer, RENDERER_FS_TEXTURE);
+   renderer_set_vs(renderer, RENDERER_VS_TEXTURE);
+
+   renderer_set_mvp(renderer, NULL);
+
+   /* remember the texture size */
+   renderer->u.drawtex.tex_width = src->texture->width0;
+   renderer->u.drawtex.tex_height = src->texture->height0;
+   renderer->state = RENDERER_STATE_DRAWTEX;
+
+   return VG_TRUE;
+}
+
+/**
+ * Draw into the destination rectangle given by (x, y, w, h).  The texture is
+ * sampled from within the rectangle given by (sx, sy, sw, sh).
+ *
+ * The coordinates are in surface coordinates.
+ */
+void renderer_drawtex(struct renderer *renderer,
+                      VGint x, VGint y, VGint w, VGint h,
+                      VGint sx, VGint sy, VGint sw, VGint sh)
+{
+   assert(renderer->state == RENDERER_STATE_DRAWTEX);
+
+   /* with scissoring */
+   renderer_quad_pos(renderer, x, y, x + w, y + h, VG_TRUE);
+   renderer_quad_texcoord(renderer, sx, sy, sx + sw, sy + sh,
+         renderer->u.drawtex.tex_width,
+         renderer->u.drawtex.tex_height);
+
+   renderer_quad_draw(renderer);
+}
+
+/**
+ * End textured drawing and restore the states.
+ */
+void renderer_drawtex_end(struct renderer *renderer)
+{
+   assert(renderer->state == RENDERER_STATE_DRAWTEX);
+
+   cso_restore_blend(renderer->cso);
+   cso_restore_samplers(renderer->cso);
+   cso_restore_fragment_sampler_views(renderer->cso);
+   cso_restore_fragment_shader(renderer->cso);
+   cso_restore_vertex_shader(renderer->cso);
+
+   renderer->state = RENDERER_STATE_INIT;
+}
+
+/**
+ * Prepare the renderer for scissor update.  This will reset the depth buffer
+ * to 1.0f.
+ */
+VGboolean renderer_scissor_begin(struct renderer *renderer,
+                                 VGboolean restore_dsa)
+{
+   struct pipe_depth_stencil_alpha_state dsa;
+
+   assert(renderer->state == RENDERER_STATE_INIT);
+
+   if (restore_dsa)
+      cso_save_depth_stencil_alpha(renderer->cso);
+   cso_save_blend(renderer->cso);
+   cso_save_fragment_shader(renderer->cso);
+
+   /* enable depth writes */
+   memset(&dsa, 0, sizeof(dsa));
+   dsa.depth.enabled = 1;
+   dsa.depth.writemask = 1;
+   dsa.depth.func = PIPE_FUNC_ALWAYS;
+   cso_set_depth_stencil_alpha(renderer->cso, &dsa);
+
+   /* disable color writes */
+   renderer_set_blend(renderer, 0);
+   renderer_set_fs(renderer, RENDERER_FS_SCISSOR);
+
+   renderer_set_mvp(renderer, NULL);
+
+   renderer->u.scissor.restore_dsa = restore_dsa;
+   renderer->state = RENDERER_STATE_SCISSOR;
+
+   /* clear the depth buffer to 1.0f */
+   renderer->pipe->clear(renderer->pipe,
+         PIPE_CLEAR_DEPTHSTENCIL, NULL, 1.0f, 0);
+
+   return VG_TRUE;
+}
+
+/**
+ * Add a scissor rectangle.  Depth values inside the rectangle will be set to
+ * 0.0f.
+ */
+void renderer_scissor(struct renderer *renderer,
+                      VGint x, VGint y, VGint width, VGint height)
+{
+   assert(renderer->state == RENDERER_STATE_SCISSOR);
+
+   renderer_quad_pos(renderer, x, y, x + width, y + height, VG_FALSE);
+   renderer_quad_draw(renderer);
+}
+
+/**
+ * End scissor update and restore the states.
+ */
+void renderer_scissor_end(struct renderer *renderer)
+{
+   assert(renderer->state == RENDERER_STATE_SCISSOR);
+
+   if (renderer->u.scissor.restore_dsa)
+      cso_restore_depth_stencil_alpha(renderer->cso);
+   cso_restore_blend(renderer->cso);
+   cso_restore_fragment_shader(renderer->cso);
+
+   renderer->state = RENDERER_STATE_INIT;
+}
+
+/**
+ * Prepare the renderer for clearing.
+ */
+VGboolean renderer_clear_begin(struct renderer *renderer)
+{
+   assert(renderer->state == RENDERER_STATE_INIT);
+
+   cso_save_blend(renderer->cso);
+   cso_save_fragment_shader(renderer->cso);
+   cso_save_vertex_shader(renderer->cso);
+
+   renderer_set_blend(renderer, ~0);
+   renderer_set_fs(renderer, RENDERER_FS_COLOR);
+   renderer_set_vs(renderer, RENDERER_VS_COLOR);
+
+   renderer_set_mvp(renderer, NULL);
+
+   renderer->state = RENDERER_STATE_CLEAR;
+
+   return VG_TRUE;
+}
+
+/**
+ * Clear the framebuffer with the specified region and color.
+ *
+ * The coordinates are in surface coordinates.
+ */
+void renderer_clear(struct renderer *renderer,
+                    VGint x, VGint y, VGint width, VGint height,
+                    const VGfloat color[4])
+{
+   VGuint i;
+
+   assert(renderer->state == RENDERER_STATE_CLEAR);
+
+   renderer_quad_pos(renderer, x, y, x + width, y + height, VG_TRUE);
+   for (i = 0; i < 4; i++)
+      memcpy(renderer->vertices[i][1], color, sizeof(VGfloat) * 4);
+
+   renderer_quad_draw(renderer);
+}
+
+/**
+ * End clearing and retore the states.
+ */
+void renderer_clear_end(struct renderer *renderer)
+{
+   assert(renderer->state == RENDERER_STATE_CLEAR);
+
+   cso_restore_blend(renderer->cso);
+   cso_restore_fragment_shader(renderer->cso);
+   cso_restore_vertex_shader(renderer->cso);
+
+   renderer->state = RENDERER_STATE_INIT;
+}
+
+/**
+ * Prepare the renderer for image filtering.
+ */
+VGboolean renderer_filter_begin(struct renderer *renderer,
+                                struct pipe_resource *dst,
+                                VGboolean y0_top,
+                                VGbitfield channel_mask,
+                                const struct pipe_sampler_state **samplers,
+                                struct pipe_sampler_view **views,
+                                VGint num_samplers,
+                                void *fs,
+                                const void *const_buffer,
+                                VGint const_buffer_len)
+{
+   struct pipe_surface *surf, surf_tmpl;
+
+   assert(renderer->state == RENDERER_STATE_INIT);
+
+   if (!fs)
+      return VG_FALSE;
+   if (!renderer_can_support(renderer, dst, PIPE_BIND_RENDER_TARGET))
+      return VG_FALSE;
+
+   u_surface_default_template(&surf_tmpl, dst,
+                              PIPE_BIND_RENDER_TARGET);
+   surf = renderer->pipe->create_surface(renderer->pipe, dst, &surf_tmpl);
+   if (!surf)
+      return VG_FALSE;
+
+   cso_save_framebuffer(renderer->cso);
+   cso_save_viewport(renderer->cso);
+   cso_save_blend(renderer->cso);
+
+   /* set the image as the target */
+   renderer_set_target(renderer, surf, NULL, y0_top);
+   pipe_surface_reference(&surf, NULL);
+
+   renderer_set_blend(renderer, channel_mask);
+
+   if (num_samplers) {
+      struct pipe_resource *tex;
+
+      cso_save_samplers(renderer->cso);
+      cso_save_fragment_sampler_views(renderer->cso);
+      cso_save_fragment_shader(renderer->cso);
+      cso_save_vertex_shader(renderer->cso);
+
+      renderer_set_custom_fs(renderer, fs,
+                             samplers, views, num_samplers,
+                             const_buffer, const_buffer_len);
+      renderer_set_vs(renderer, RENDERER_VS_TEXTURE);
+
+      tex = views[0]->texture;
+      renderer->u.filter.tex_width = tex->width0;
+      renderer->u.filter.tex_height = tex->height0;
+      renderer->u.filter.use_sampler = VG_TRUE;
    }
+   else {
+      cso_save_fragment_shader(renderer->cso);
 
-   cso_restore_vertex_shader(r->cso);
+      renderer_set_custom_fs(renderer, fs, NULL, NULL, 0,
+                             const_buffer, const_buffer_len);
+
+      renderer->u.filter.use_sampler = VG_FALSE;
+   }
+
+   renderer_set_mvp(renderer, NULL);
+
+   renderer->state = RENDERER_STATE_FILTER;
+
+   return VG_TRUE;
 }
 
-void renderer_copy_texture(struct renderer *ctx,
-                           struct pipe_sampler_view *src,
-                           VGfloat sx1, VGfloat sy1,
-                           VGfloat sx2, VGfloat sy2,
-                           struct pipe_resource *dst,
-                           VGfloat dx1, VGfloat dy1,
-                           VGfloat dx2, VGfloat dy2)
+/**
+ * Draw into a rectangle of the destination with the specified region of the
+ * texture(s).
+ *
+ * The coordinates are in surface coordinates.
+ */
+void renderer_filter(struct renderer *renderer,
+                    VGint x, VGint y, VGint w, VGint h,
+                    VGint sx, VGint sy, VGint sw, VGint sh)
 {
-   struct pipe_context *pipe = ctx->pipe;
-   struct pipe_screen *screen = pipe->screen;
-   struct pipe_resource *tex = src->texture;
-   struct pipe_resource *buf;
-   struct pipe_surface *dst_surf = screen->get_tex_surface(
-      screen, dst, 0, 0, 0,
-      PIPE_BIND_RENDER_TARGET);
-   struct pipe_framebuffer_state fb;
-   float s0, t0, s1, t1;
+   assert(renderer->state == RENDERER_STATE_FILTER);
 
-   assert(tex->width0 != 0);
-   assert(tex->height0 != 0);
-   assert(dst->width0 != 0);
-   assert(dst->height0 != 0);
-
-#if 0
-   debug_printf("copy texture [%f, %f, %f, %f], [%f, %f, %f, %f]\n",
-                sx1, sy1, sx2, sy2, dx1, dy1, dx2, dy2);
-#endif
-
-#if 1
-   s0 = sx1 / tex->width0;
-   s1 = sx2 / tex->width0;
-   t0 = sy1 / tex->height0;
-   t1 = sy2 / tex->height0;
-#else
-   s0 = 0;
-   s1 = 1;
-   t0 = 0;
-   t1 = 1;
-#endif
-
-   assert(screen->is_format_supported(screen, dst_surf->format, PIPE_TEXTURE_2D,
-                                      0, PIPE_BIND_RENDER_TARGET, 0));
+   renderer_quad_pos(renderer, x, y, x + w, y + h, VG_FALSE);
+   if (renderer->u.filter.use_sampler) {
+      renderer_quad_texcoord(renderer, sx, sy, sx + sw, sy + sh,
+            renderer->u.filter.tex_width,
+            renderer->u.filter.tex_height);
+   }
 
-   /* save state (restored below) */
-   cso_save_blend(ctx->cso);
-   cso_save_samplers(ctx->cso);
-   cso_save_fragment_sampler_views(ctx->cso);
-   cso_save_framebuffer(ctx->cso);
-   cso_save_fragment_shader(ctx->cso);
-   cso_save_vertex_shader(ctx->cso);
+   renderer_quad_draw(renderer);
+}
+
+/**
+ * End image filtering and restore the states.
+ */
+void renderer_filter_end(struct renderer *renderer)
+{
+   assert(renderer->state == RENDERER_STATE_FILTER);
 
-   cso_save_viewport(ctx->cso);
+   if (renderer->u.filter.use_sampler) {
+      cso_restore_samplers(renderer->cso);
+      cso_restore_fragment_sampler_views(renderer->cso);
+      cso_restore_vertex_shader(renderer->cso);
+   }
 
+   cso_restore_framebuffer(renderer->cso);
+   cso_restore_viewport(renderer->cso);
+   cso_restore_blend(renderer->cso);
+   cso_restore_fragment_shader(renderer->cso);
 
-   /* set misc state we care about */
-   {
-      struct pipe_blend_state blend;
-      memset(&blend, 0, sizeof(blend));
-      blend.rt[0].rgb_src_factor = PIPE_BLENDFACTOR_ONE;
-      blend.rt[0].alpha_src_factor = PIPE_BLENDFACTOR_ONE;
-      blend.rt[0].rgb_dst_factor = PIPE_BLENDFACTOR_ZERO;
-      blend.rt[0].alpha_dst_factor = PIPE_BLENDFACTOR_ZERO;
-      blend.rt[0].colormask = PIPE_MASK_RGBA;
-      cso_set_blend(ctx->cso, &blend);
+   renderer->state = RENDERER_STATE_INIT;
+}
+
+/**
+ * Prepare the renderer for polygon silhouette rendering.
+ */
+VGboolean renderer_polygon_stencil_begin(struct renderer *renderer,
+                                         struct pipe_vertex_element *velem,
+                                         VGFillRule rule,
+                                         VGboolean restore_dsa)
+{
+   struct pipe_depth_stencil_alpha_state *dsa;
+   VGboolean manual_two_sides;
+
+   assert(renderer->state == RENDERER_STATE_INIT);
+
+   cso_save_vertex_elements(renderer->cso);
+   cso_save_blend(renderer->cso);
+   cso_save_depth_stencil_alpha(renderer->cso);
+
+   cso_set_vertex_elements(renderer->cso, 1, velem);
+
+   /* disable color writes */
+   renderer_set_blend(renderer, 0);
+
+   manual_two_sides = VG_FALSE;
+   dsa = &renderer->u.polygon_stencil.dsa;
+   memset(dsa, 0, sizeof(*dsa));
+   if (rule == VG_EVEN_ODD) {
+      dsa->stencil[0].enabled = 1;
+      dsa->stencil[0].writemask = 1;
+      dsa->stencil[0].fail_op = PIPE_STENCIL_OP_KEEP;
+      dsa->stencil[0].zfail_op = PIPE_STENCIL_OP_KEEP;
+      dsa->stencil[0].zpass_op = PIPE_STENCIL_OP_INVERT;
+      dsa->stencil[0].func = PIPE_FUNC_ALWAYS;
+      dsa->stencil[0].valuemask = ~0;
    }
+   else {
+      assert(rule == VG_NON_ZERO);
+
+      /* front face */
+      dsa->stencil[0].enabled = 1;
+      dsa->stencil[0].writemask = ~0;
+      dsa->stencil[0].fail_op = PIPE_STENCIL_OP_KEEP;
+      dsa->stencil[0].zfail_op = PIPE_STENCIL_OP_KEEP;
+      dsa->stencil[0].zpass_op = PIPE_STENCIL_OP_INCR_WRAP;
+      dsa->stencil[0].func = PIPE_FUNC_ALWAYS;
+      dsa->stencil[0].valuemask = ~0;
+
+      if (renderer->pipe->screen->get_param(renderer->pipe->screen,
+                                            PIPE_CAP_TWO_SIDED_STENCIL)) {
+         /* back face */
+         dsa->stencil[1] = dsa->stencil[0];
+         dsa->stencil[1].zpass_op = PIPE_STENCIL_OP_DECR_WRAP;
+      }
+      else {
+         manual_two_sides = VG_TRUE;
+      }
+   }
+   cso_set_depth_stencil_alpha(renderer->cso, dsa);
+
+   if (manual_two_sides)
+      cso_save_rasterizer(renderer->cso);
 
-   /* sampler */
-   {
-      struct pipe_sampler_state sampler;
-      memset(&sampler, 0, sizeof(sampler));
-      sampler.wrap_s = PIPE_TEX_WRAP_CLAMP_TO_EDGE;
-      sampler.wrap_t = PIPE_TEX_WRAP_CLAMP_TO_EDGE;
-      sampler.wrap_r = PIPE_TEX_WRAP_CLAMP_TO_EDGE;
-      sampler.min_mip_filter = PIPE_TEX_MIPFILTER_NONE;
-      sampler.min_img_filter = PIPE_TEX_FILTER_NEAREST;
-      sampler.mag_img_filter = PIPE_TEX_FILTER_NEAREST;
-      sampler.normalized_coords = 1;
-      cso_single_sampler(ctx->cso, 0, &sampler);
-      cso_single_sampler_done(ctx->cso);
+   renderer->u.polygon_stencil.manual_two_sides = manual_two_sides;
+   renderer->u.polygon_stencil.restore_dsa = restore_dsa;
+   renderer->state = RENDERER_STATE_POLYGON_STENCIL;
+
+   return VG_TRUE;
+}
+
+/**
+ * Render a polygon silhouette to stencil buffer.
+ */
+void renderer_polygon_stencil(struct renderer *renderer,
+                              struct pipe_vertex_buffer *vbuf,
+                              VGuint mode, VGuint start, VGuint count)
+{
+   assert(renderer->state == RENDERER_STATE_POLYGON_STENCIL);
+
+   renderer->pipe->set_vertex_buffers(renderer->pipe, 1, vbuf);
+
+   if (!renderer->u.polygon_stencil.manual_two_sides) {
+      util_draw_arrays(renderer->pipe, mode, start, count);
    }
+   else {
+      struct pipe_rasterizer_state raster;
+      struct pipe_depth_stencil_alpha_state dsa;
 
-   vg_set_viewport(ctx->owner, VEGA_Y0_TOP);
+      raster = renderer->g3d.rasterizer;
+      dsa = renderer->u.polygon_stencil.dsa;
 
-   /* texture */
-   cso_set_fragment_sampler_views(ctx->cso, 1, &src);
+      /* front */
+      raster.cull_face = PIPE_FACE_BACK;
+      dsa.stencil[0].zpass_op = PIPE_STENCIL_OP_INCR_WRAP;
 
-   /* shaders */
-   cso_set_vertex_shader_handle(ctx->cso, vg_texture_vs(ctx->owner));
-   cso_set_fragment_shader_handle(ctx->cso, ctx->fs);
+      cso_set_rasterizer(renderer->cso, &raster);
+      cso_set_depth_stencil_alpha(renderer->cso, &dsa);
+      util_draw_arrays(renderer->pipe, mode, start, count);
 
-   /* drawing dest */
-   memset(&fb, 0, sizeof(fb));
-   fb.width = dst_surf->width;
-   fb.height = dst_surf->height;
-   fb.nr_cbufs = 1;
-   fb.cbufs[0] = dst_surf;
-   {
-      VGint i;
-      for (i = 1; i < PIPE_MAX_COLOR_BUFS; ++i)
-         fb.cbufs[i] = 0;
+      /* back */
+      raster.cull_face = PIPE_FACE_FRONT;
+      dsa.stencil[0].zpass_op = PIPE_STENCIL_OP_DECR_WRAP;
+
+      cso_set_rasterizer(renderer->cso, &raster);
+      cso_set_depth_stencil_alpha(renderer->cso, &dsa);
+      util_draw_arrays(renderer->pipe, mode, start, count);
    }
-   cso_set_framebuffer(ctx->cso, &fb);
+}
+
+/**
+ * End polygon silhouette rendering.
+ */
+void renderer_polygon_stencil_end(struct renderer *renderer)
+{
+   assert(renderer->state == RENDERER_STATE_POLYGON_STENCIL);
 
-   /* draw quad */
-   buf = setup_vertex_data_tex(ctx,
-                         dx1, dy1,
-                         dx2, dy2,
-                         s0, t0, s1, t1,
-                         0.0f);
+   if (renderer->u.polygon_stencil.manual_two_sides)
+      cso_restore_rasterizer(renderer->cso);
 
-   if (buf) {
-      cso_set_vertex_elements(ctx->cso, 2, ctx->owner->velems);
-      util_draw_vertex_buffer(ctx->pipe, buf, 0,
-                              PIPE_PRIM_TRIANGLE_FAN,
-                              4,  /* verts */
-                              2); /* attribs/vert */
+   cso_restore_vertex_elements(renderer->cso);
+
+   /* restore color writes */
+   cso_restore_blend(renderer->cso);
+
+   if (renderer->u.polygon_stencil.restore_dsa)
+      cso_restore_depth_stencil_alpha(renderer->cso);
+
+   renderer->state = RENDERER_STATE_INIT;
+}
+
+/**
+ * Prepare the renderer for polygon filling.
+ */
+VGboolean renderer_polygon_fill_begin(struct renderer *renderer,
+                                      VGboolean save_dsa)
+{
+   struct pipe_depth_stencil_alpha_state dsa;
+
+   assert(renderer->state == RENDERER_STATE_INIT);
+
+   if (save_dsa)
+      cso_save_depth_stencil_alpha(renderer->cso);
+
+   /* setup stencil ops */
+   memset(&dsa, 0, sizeof(dsa));
+   dsa.stencil[0].enabled = 1;
+   dsa.stencil[0].func = PIPE_FUNC_NOTEQUAL;
+   dsa.stencil[0].fail_op = PIPE_STENCIL_OP_REPLACE;
+   dsa.stencil[0].zfail_op = PIPE_STENCIL_OP_REPLACE;
+   dsa.stencil[0].zpass_op = PIPE_STENCIL_OP_REPLACE;
+   dsa.stencil[0].valuemask = ~0;
+   dsa.stencil[0].writemask = ~0;
+   dsa.depth = renderer->g3d.dsa.depth;
+   cso_set_depth_stencil_alpha(renderer->cso, &dsa);
+
+   renderer->state = RENDERER_STATE_POLYGON_FILL;
+
+   return VG_TRUE;
+}
+
+/**
+ * Fill a polygon.
+ */
+void renderer_polygon_fill(struct renderer *renderer,
+                           VGfloat min_x, VGfloat min_y,
+                           VGfloat max_x, VGfloat max_y)
+{
+   assert(renderer->state == RENDERER_STATE_POLYGON_FILL);
+
+   renderer_quad_pos(renderer, min_x, min_y, max_x, max_y, VG_TRUE);
+   renderer_quad_draw(renderer);
+}
+
+/**
+ * End polygon filling.
+ */
+void renderer_polygon_fill_end(struct renderer *renderer)
+{
+   assert(renderer->state == RENDERER_STATE_POLYGON_FILL);
+
+   cso_restore_depth_stencil_alpha(renderer->cso);
 
-      pipe_resource_reference( &buf,
-                             NULL );
+   renderer->state = RENDERER_STATE_INIT;
+}
+
+struct renderer * renderer_create(struct vg_context *owner)
+{
+   struct renderer *renderer;
+   struct pipe_rasterizer_state *raster;
+   struct pipe_stencil_ref sr;
+   VGint i;
+
+   renderer = CALLOC_STRUCT(renderer);
+   if (!renderer)
+      return NULL;
+
+   renderer->pipe = owner->pipe;
+   renderer->cso = owner->cso_context;
+
+   /* init vertex data that doesn't change */
+   for (i = 0; i < 4; i++)
+      renderer->vertices[i][0][3] = 1.0f; /* w */
+
+   for (i = 0; i < 2; i++) {
+      renderer->velems[i].src_offset = i * 4 * sizeof(float);
+      renderer->velems[i].instance_divisor = 0;
+      renderer->velems[i].vertex_buffer_index = 0;
+      renderer->velems[i].src_format = PIPE_FORMAT_R32G32B32A32_FLOAT;
+   }
+   cso_set_vertex_elements(renderer->cso, 2, renderer->velems);
+
+   /* GL rasterization rules */
+   raster = &renderer->g3d.rasterizer;
+   memset(raster, 0, sizeof(*raster));
+   raster->gl_rasterization_rules = 1;
+   cso_set_rasterizer(renderer->cso, raster);
+
+   /* fixed at 0 */
+   memset(&sr, 0, sizeof(sr));
+   cso_set_stencil_ref(renderer->cso, &sr);
+
+   renderer_set_vs(renderer, RENDERER_VS_PLAIN);
+
+   renderer->state = RENDERER_STATE_INIT;
+
+   return renderer;
+}
+
+void renderer_destroy(struct renderer *ctx)
+{
+   int i;
+
+   for (i = 0; i < NUM_RENDERER_VS; i++) {
+      if (ctx->cached_vs[i])
+         cso_delete_vertex_shader(ctx->cso, ctx->cached_vs[i]);
+   }
+   for (i = 0; i < NUM_RENDERER_FS; i++) {
+      if (ctx->cached_fs[i])
+         cso_delete_fragment_shader(ctx->cso, ctx->cached_fs[i]);
+   }
+
+   pipe_resource_reference(&ctx->vs_cbuf, NULL);
+   pipe_resource_reference(&ctx->fs_cbuf, NULL);
+
+   FREE(ctx);
+}
+
+static void update_clip_state(struct renderer *renderer,
+                              const struct vg_state *state)
+{
+   struct pipe_depth_stencil_alpha_state *dsa = &renderer->g3d.dsa;
+
+   memset(dsa, 0, sizeof(struct pipe_depth_stencil_alpha_state));
+
+   if (state->scissoring) {
+      struct pipe_framebuffer_state *fb = &renderer->g3d.fb;
+      int i;
+
+      renderer_scissor_begin(renderer, VG_FALSE);
+
+      for (i = 0; i < state->scissor_rects_num; ++i) {
+         const float x      = state->scissor_rects[i * 4 + 0].f;
+         const float y      = state->scissor_rects[i * 4 + 1].f;
+         const float width  = state->scissor_rects[i * 4 + 2].f;
+         const float height = state->scissor_rects[i * 4 + 3].f;
+         VGint x0, y0, x1, y1, iw, ih;
+
+         x0 = (VGint) x;
+         y0 = (VGint) y;
+         if (x0 < 0)
+            x0 = 0;
+         if (y0 < 0)
+            y0 = 0;
+
+         /* note that x1 and y1 are exclusive */
+         x1 = (VGint) ceilf(x + width);
+         y1 = (VGint) ceilf(y + height);
+         if (x1 > fb->width)
+            x1 = fb->width;
+         if (y1 > fb->height)
+            y1 = fb->height;
+
+         iw = x1 - x0;
+         ih = y1 - y0;
+         if (iw > 0 && ih> 0 )
+            renderer_scissor(renderer, x0, y0, iw, ih);
+      }
+
+      renderer_scissor_end(renderer);
+
+      dsa->depth.enabled = 1; /* glEnable(GL_DEPTH_TEST); */
+      dsa->depth.writemask = 0;/*glDepthMask(FALSE);*/
+      dsa->depth.func = PIPE_FUNC_GEQUAL;
    }
+}
+
+static void renderer_validate_blend(struct renderer *renderer,
+                                     const struct vg_state *state,
+                                     enum pipe_format fb_format)
+{
+   struct pipe_blend_state blend;
+
+   memset(&blend, 0, sizeof(blend));
+   blend.rt[0].colormask = PIPE_MASK_RGBA;
+   blend.rt[0].rgb_src_factor   = PIPE_BLENDFACTOR_ONE;
+   blend.rt[0].alpha_src_factor = PIPE_BLENDFACTOR_ONE;
+   blend.rt[0].rgb_dst_factor   = PIPE_BLENDFACTOR_ZERO;
+   blend.rt[0].alpha_dst_factor = PIPE_BLENDFACTOR_ZERO;
 
-   /* restore state we changed */
-   cso_restore_blend(ctx->cso);
-   cso_restore_samplers(ctx->cso);
-   cso_restore_fragment_sampler_views(ctx->cso);
-   cso_restore_framebuffer(ctx->cso);
-   cso_restore_vertex_shader(ctx->cso);
-   cso_restore_fragment_shader(ctx->cso);
-   cso_restore_viewport(ctx->cso);
+   /* TODO alpha masking happens after blending? */
+
+   switch (state->blend_mode) {
+   case VG_BLEND_SRC:
+      blend.rt[0].rgb_src_factor   = PIPE_BLENDFACTOR_ONE;
+      blend.rt[0].alpha_src_factor = PIPE_BLENDFACTOR_ONE;
+      blend.rt[0].rgb_dst_factor   = PIPE_BLENDFACTOR_ZERO;
+      blend.rt[0].alpha_dst_factor = PIPE_BLENDFACTOR_ZERO;
+      break;
+   case VG_BLEND_SRC_OVER:
+      if (!util_format_has_alpha(fb_format)) {
+         blend.rt[0].rgb_src_factor   = PIPE_BLENDFACTOR_SRC_ALPHA;
+         blend.rt[0].alpha_src_factor = PIPE_BLENDFACTOR_ONE;
+         blend.rt[0].rgb_dst_factor   = PIPE_BLENDFACTOR_INV_SRC_ALPHA;
+         blend.rt[0].alpha_dst_factor = PIPE_BLENDFACTOR_INV_SRC_ALPHA;
+         blend.rt[0].blend_enable = 1;
+      }
+      break;
+   case VG_BLEND_SRC_IN:
+      blend.rt[0].rgb_src_factor   = PIPE_BLENDFACTOR_ONE;
+      blend.rt[0].alpha_src_factor = PIPE_BLENDFACTOR_DST_ALPHA;
+      blend.rt[0].rgb_dst_factor   = PIPE_BLENDFACTOR_ZERO;
+      blend.rt[0].alpha_dst_factor = PIPE_BLENDFACTOR_ZERO;
+      blend.rt[0].blend_enable = 1;
+      break;
+   case VG_BLEND_DST_IN:
+      blend.rt[0].rgb_src_factor   = PIPE_BLENDFACTOR_ZERO;
+      blend.rt[0].alpha_src_factor = PIPE_BLENDFACTOR_ZERO;
+      blend.rt[0].rgb_dst_factor   = PIPE_BLENDFACTOR_ONE;
+      blend.rt[0].alpha_dst_factor = PIPE_BLENDFACTOR_SRC_ALPHA;
+      blend.rt[0].blend_enable = 1;
+      break;
+   case VG_BLEND_DST_OVER:
+   case VG_BLEND_MULTIPLY:
+   case VG_BLEND_SCREEN:
+   case VG_BLEND_DARKEN:
+   case VG_BLEND_LIGHTEN:
+   case VG_BLEND_ADDITIVE:
+      /* need a shader */
+      break;
+   default:
+      assert(!"not implemented blend mode");
+      break;
+   }
 
-   pipe_surface_reference(&dst_surf, NULL);
+   cso_set_blend(renderer->cso, &blend);
+}
+
+/**
+ * Propogate OpenVG state changes to the renderer.  Only framebuffer, blending
+ * and scissoring states are relevant here.
+ */
+void renderer_validate(struct renderer *renderer,
+                       VGbitfield dirty,
+                       const struct st_framebuffer *stfb,
+                       const struct vg_state *state)
+{
+   assert(renderer->state == RENDERER_STATE_INIT);
+
+   dirty |= renderer->dirty;
+   renderer->dirty = 0;
+
+   if (dirty & FRAMEBUFFER_DIRTY) {
+      struct pipe_framebuffer_state *fb = &renderer->g3d.fb;
+      struct matrix *proj = &renderer->projection;
+
+      memset(fb, 0, sizeof(struct pipe_framebuffer_state));
+      fb->width  = stfb->width;
+      fb->height = stfb->height;
+      fb->nr_cbufs = 1;
+      fb->cbufs[0] = stfb->strb->surface;
+      fb->zsbuf = stfb->dsrb->surface;
+
+      cso_set_framebuffer(renderer->cso, fb);
+      vg_set_viewport(renderer, VEGA_Y0_BOTTOM);
+
+      matrix_load_identity(proj);
+      matrix_translate(proj, -1.0f, -1.0f);
+      matrix_scale(proj, 2.0f / fb->width, 2.0f / fb->height);
+
+      /* we also got a new depth buffer */
+      if (dirty & DEPTH_STENCIL_DIRTY) {
+         renderer->pipe->clear(renderer->pipe,
+               PIPE_CLEAR_DEPTHSTENCIL, NULL, 0.0, 0);
+      }
+   }
+
+   /* must be last because it renders to the depth buffer*/
+   if (dirty & DEPTH_STENCIL_DIRTY) {
+      update_clip_state(renderer, state);
+      cso_set_depth_stencil_alpha(renderer->cso, &renderer->g3d.dsa);
+   }
+
+   if (dirty & BLEND_DIRTY)
+      renderer_validate_blend(renderer, state, stfb->strb->format);
+}
+
+/**
+ * Prepare the renderer for OpenVG pipeline.
+ */
+void renderer_validate_for_shader(struct renderer *renderer,
+                                  const struct pipe_sampler_state **samplers,
+                                  struct pipe_sampler_view **views,
+                                  VGint num_samplers,
+                                  const struct matrix *modelview,
+                                  void *fs,
+                                  const void *const_buffer,
+                                  VGint const_buffer_len)
+{
+   struct matrix mvp = renderer->projection;
+
+   /* will be used in POLYGON_STENCIL and POLYGON_FILL */
+   matrix_mult(&mvp, modelview);
+   renderer_set_mvp(renderer, &mvp);
+
+   renderer_set_custom_fs(renderer, fs,
+                          samplers, views, num_samplers,
+                          const_buffer, const_buffer_len);
+}
+
+void renderer_validate_for_mask_rendering(struct renderer *renderer,
+                                          struct pipe_surface *dst,
+                                          const struct matrix *modelview)
+{
+   struct matrix mvp = renderer->projection;
+
+   /* will be used in POLYGON_STENCIL and POLYGON_FILL */
+   matrix_mult(&mvp, modelview);
+   renderer_set_mvp(renderer, &mvp);
+
+   renderer_set_target(renderer, dst, renderer->g3d.fb.zsbuf, VG_FALSE);
+   renderer_set_blend(renderer, ~0);
+   renderer_set_fs(renderer, RENDERER_FS_WHITE);
+
+   /* set internal dirty flags (hacky!) */
+   renderer->dirty = FRAMEBUFFER_DIRTY | BLEND_DIRTY;
 }
 
 void renderer_copy_surface(struct renderer *ctx,
@@ -409,13 +1429,11 @@ void renderer_copy_surface(struct renderer *ctx,
 {
    struct pipe_context *pipe = ctx->pipe;
    struct pipe_screen *screen = pipe->screen;
-   struct pipe_resource *buf;
    struct pipe_sampler_view view_templ;
    struct pipe_sampler_view *view;
+   struct pipe_box src_box;
    struct pipe_resource texTemp, *tex;
-   struct pipe_subresource subsrc, subdst;
-   struct pipe_framebuffer_state fb;
-   struct st_framebuffer *stfb = ctx->owner->draw_buffer;
+   const struct pipe_framebuffer_state *fb = &ctx->g3d.fb;
    const int srcW = abs(srcX1 - srcX0);
    const int srcH = abs(srcY1 - srcY0);
    const int srcLeft = MIN2(srcX0, srcX1);
@@ -458,6 +1476,7 @@ void renderer_copy_surface(struct renderer *ctx,
    texTemp.width0 = srcW;
    texTemp.height0 = srcH;
    texTemp.depth0 = 1;
+   texTemp.array_size = 1;
    texTemp.bind = PIPE_BIND_SAMPLER_VIEW;
 
    tex = screen->resource_create(screen, &texTemp);
@@ -470,100 +1489,35 @@ void renderer_copy_surface(struct renderer *ctx,
    if (!view)
       return;
 
-   subdst.face = 0;
-   subdst.level = 0;
-   subsrc.face = src->face;
-   subsrc.level = src->level;
+   u_box_2d_zslice(srcLeft, srcTop, src->u.tex.first_layer, srcW, srcH, &src_box);
 
    pipe->resource_copy_region(pipe,
-                              tex, subdst, 0, 0, 0,  /* dest */
-                              src->texture, subsrc, srcLeft, srcTop, src->zslice, /* src */
-                              srcW, srcH);     /* size */
-
-   /* save state (restored below) */
-   cso_save_blend(ctx->cso);
-   cso_save_samplers(ctx->cso);
-   cso_save_fragment_sampler_views(ctx->cso);
-   cso_save_framebuffer(ctx->cso);
-   cso_save_fragment_shader(ctx->cso);
-   cso_save_vertex_shader(ctx->cso);
-   cso_save_viewport(ctx->cso);
-
-   /* set misc state we care about */
-   {
-      struct pipe_blend_state blend;
-      memset(&blend, 0, sizeof(blend));
-      blend.rt[0].rgb_src_factor = PIPE_BLENDFACTOR_ONE;
-      blend.rt[0].alpha_src_factor = PIPE_BLENDFACTOR_ONE;
-      blend.rt[0].rgb_dst_factor = PIPE_BLENDFACTOR_ZERO;
-      blend.rt[0].alpha_dst_factor = PIPE_BLENDFACTOR_ZERO;
-      blend.rt[0].colormask = PIPE_MASK_RGBA;
-      cso_set_blend(ctx->cso, &blend);
+                              tex, 0, 0, 0, 0,  /* dest */
+                              src->texture, 0, &src_box);
+
+   assert(floatsEqual(z, 0.0f));
+
+   /* draw */
+   if (fb->cbufs[0] == dst) {
+      /* transform back to surface coordinates */
+      dstY0 = dst->height - dstY0;
+      dstY1 = dst->height - dstY1;
+
+      if (renderer_drawtex_begin(ctx, view)) {
+         renderer_drawtex(ctx,
+               dstX0, dstY0, dstX1 - dstX0, dstY1 - dstY0,
+               0, 0, view->texture->width0, view->texture->height0);
+         renderer_drawtex_end(ctx);
+      }
    }
-
-   vg_set_viewport(ctx->owner, VEGA_Y0_TOP);
-
-   /* sampler */
-   {
-      struct pipe_sampler_state sampler;
-      memset(&sampler, 0, sizeof(sampler));
-      sampler.wrap_s = PIPE_TEX_WRAP_CLAMP_TO_EDGE;
-      sampler.wrap_t = PIPE_TEX_WRAP_CLAMP_TO_EDGE;
-      sampler.wrap_r = PIPE_TEX_WRAP_CLAMP_TO_EDGE;
-      sampler.min_mip_filter = PIPE_TEX_MIPFILTER_NONE;
-      sampler.min_img_filter = PIPE_TEX_FILTER_NEAREST;
-      sampler.mag_img_filter = PIPE_TEX_FILTER_NEAREST;
-      sampler.normalized_coords = 1;
-      cso_single_sampler(ctx->cso, 0, &sampler);
-      cso_single_sampler_done(ctx->cso);
-   }
-
-   /* texture */
-   cso_set_fragment_sampler_views(ctx->cso, 1, &view);
-
-   /* shaders */
-   cso_set_fragment_shader_handle(ctx->cso, ctx->fs);
-   cso_set_vertex_shader_handle(ctx->cso, vg_texture_vs(ctx->owner));
-
-   /* drawing dest */
-   if (stfb->strb->surface != dst) {
-      memset(&fb, 0, sizeof(fb));
-      fb.width = dst->width;
-      fb.height = dst->height;
-      fb.nr_cbufs = 1;
-      fb.cbufs[0] = dst;
-      fb.zsbuf = stfb->dsrb->surface;
-      cso_set_framebuffer(ctx->cso, &fb);
+   else {
+      if (renderer_copy_begin(ctx, dst, VG_TRUE, view)) {
+         renderer_copy(ctx,
+               dstX0, dstY0, dstX1 - dstX0, dstY1 - dstY0,
+               0, 0, view->texture->width0, view->texture->height0);
+         renderer_copy_end(ctx);
+      }
    }
-
-   /* draw quad */
-   buf = setup_vertex_data(ctx,
-                           (float) dstX0, (float) dstY0,
-                           (float) dstX1, (float) dstY1, z);
-
-   if (buf) {
-      cso_set_vertex_elements(ctx->cso, 2, ctx->owner->velems);
-      util_draw_vertex_buffer(ctx->pipe, buf, 0,
-                              PIPE_PRIM_TRIANGLE_FAN,
-                              4,  /* verts */
-                              2); /* attribs/vert */
-
-      pipe_resource_reference( &buf,
-                             NULL );
-   }
-
-
-   /* restore state we changed */
-   cso_restore_blend(ctx->cso);
-   cso_restore_samplers(ctx->cso);
-   cso_restore_fragment_sampler_views(ctx->cso);
-   cso_restore_framebuffer(ctx->cso);
-   cso_restore_fragment_shader(ctx->cso);
-   cso_restore_vertex_shader(ctx->cso);
-   cso_restore_viewport(ctx->cso);
-
-   pipe_resource_reference(&tex, NULL);
-   pipe_sampler_view_reference(&view, NULL);
 }
 
 void renderer_texture_quad(struct renderer *r,
@@ -575,36 +1529,38 @@ void renderer_texture_quad(struct renderer *r,
                            VGfloat x3, VGfloat y3,
                            VGfloat x4, VGfloat y4)
 {
-   struct pipe_context *pipe = r->pipe;
-   struct pipe_resource *buf;
-   VGfloat s0, t0, s1, t1;
+   const VGfloat z = 0.0f;
 
+   assert(r->state == RENDERER_STATE_INIT);
    assert(tex->width0 != 0);
    assert(tex->height0 != 0);
 
-   s0 = x1offset / tex->width0;
-   s1 = x2offset / tex->width0;
-   t0 = y1offset / tex->height0;
-   t1 = y2offset / tex->height0;
-
    cso_save_vertex_shader(r->cso);
-   /* shaders */
-   cso_set_vertex_shader_handle(r->cso, vg_texture_vs(r->owner));
 
-   /* draw quad */
-   buf = setup_vertex_data_qtex(r, x1, y1, x2, y2, x3, y3, x4, y4,
-                          s0, t0, s1, t1, 0.0f);
+   renderer_set_vs(r, RENDERER_VS_TEXTURE);
 
-   if (buf) {
-      cso_set_vertex_elements(r->cso, 2, r->owner->velems);
-      util_draw_vertex_buffer(pipe, buf, 0,
-                              PIPE_PRIM_TRIANGLE_FAN,
-                              4,  /* verts */
-                              2); /* attribs/vert */
+   /* manually set up positions */
+   r->vertices[0][0][0] = x1;
+   r->vertices[0][0][1] = y1;
+   r->vertices[0][0][2] = z;
 
-      pipe_resource_reference(&buf,
-                            NULL);
-   }
+   r->vertices[1][0][0] = x2;
+   r->vertices[1][0][1] = y2;
+   r->vertices[1][0][2] = z;
+
+   r->vertices[2][0][0] = x3;
+   r->vertices[2][0][1] = y3;
+   r->vertices[2][0][2] = z;
+
+   r->vertices[3][0][0] = x4;
+   r->vertices[3][0][1] = y4;
+   r->vertices[3][0][2] = z;
+
+   /* texcoords */
+   renderer_quad_texcoord(r, x1offset, y1offset,
+         x2offset, y2offset, tex->width0, tex->height0);
+
+   renderer_quad_draw(r);
 
    cso_restore_vertex_shader(r->cso);
 }
diff --git a/src/gallium/state_trackers/vega/renderer.h b/src/gallium/state_trackers/vega/renderer.h
index b1a9fb58be..fe71993658 100644
--- a/src/gallium/state_trackers/vega/renderer.h
+++ b/src/gallium/state_trackers/vega/renderer.h
@@ -1,6 +1,7 @@
 /**************************************************************************
  *
  * Copyright 2009 VMware, Inc.  All Rights Reserved.
+ * Copyright 2010 LunarG, Inc.  All Rights Reserved.
  *
  * Permission is hereby granted, free of charge, to any person obtaining a
  * copy of this software and associated documentation files (the
@@ -32,23 +33,110 @@
 struct renderer;
 
 struct vg_context;
+struct vg_state;
+struct st_framebuffer;
 struct pipe_resource;
+struct pipe_sampler_state;
 struct pipe_sampler_view;
 struct pipe_surface;
+struct pipe_vertex_element;
+struct pipe_vertex_buffer;
+struct matrix;
 
 struct renderer *renderer_create(struct vg_context *owner);
 void renderer_destroy(struct renderer *);
 
-void renderer_draw_quad(struct renderer *,
-                        VGfloat x1, VGfloat y1,
-                        VGfloat x2, VGfloat y2,
-                        VGfloat depth);
-void renderer_draw_texture(struct renderer *,
-                           struct pipe_resource *texture,
-                           VGfloat x1offset, VGfloat y1offset,
-                           VGfloat x2offset, VGfloat y2offset,
-                           VGfloat x1, VGfloat y1,
-                           VGfloat x2, VGfloat y2);
+void renderer_validate(struct renderer *renderer,
+                       VGbitfield dirty,
+                       const struct st_framebuffer *stfb,
+                       const struct vg_state *state);
+
+void renderer_validate_for_shader(struct renderer *renderer,
+                                  const struct pipe_sampler_state **samplers,
+                                  struct pipe_sampler_view **views,
+                                  VGint num_samplers,
+                                  const struct matrix *modelview,
+                                  void *fs,
+                                  const void *const_buffer,
+                                  VGint const_buffer_len);
+
+void renderer_validate_for_mask_rendering(struct renderer *renderer,
+                                          struct pipe_surface *dst,
+                                          const struct matrix *modelview);
+
+VGboolean renderer_copy_begin(struct renderer *renderer,
+                              struct pipe_surface *dst,
+                              VGboolean y0_top,
+                              struct pipe_sampler_view *src);
+
+void renderer_copy(struct renderer *renderer,
+                   VGint x, VGint y, VGint w, VGint h,
+                   VGint sx, VGint sy, VGint sw, VGint sh);
+
+void renderer_copy_end(struct renderer *renderer);
+
+VGboolean renderer_drawtex_begin(struct renderer *renderer,
+                                 struct pipe_sampler_view *src);
+
+void renderer_drawtex(struct renderer *renderer,
+                      VGint x, VGint y, VGint w, VGint h,
+                      VGint sx, VGint sy, VGint sw, VGint sh);
+
+void renderer_drawtex_end(struct renderer *renderer);
+
+VGboolean renderer_scissor_begin(struct renderer *renderer,
+                                 VGboolean restore_dsa);
+
+void renderer_scissor(struct renderer *renderer,
+                      VGint x, VGint y, VGint width, VGint height);
+
+void renderer_scissor_end(struct renderer *renderer);
+
+VGboolean renderer_clear_begin(struct renderer *renderer);
+
+void renderer_clear(struct renderer *renderer,
+                    VGint x, VGint y, VGint width, VGint height,
+                    const VGfloat color[4]);
+
+void renderer_clear_end(struct renderer *renderer);
+
+VGboolean renderer_filter_begin(struct renderer *renderer,
+                                struct pipe_resource *dst,
+                                VGboolean y0_top,
+                                VGbitfield channel_mask,
+                                const struct pipe_sampler_state **samplers,
+                                struct pipe_sampler_view **views,
+                                VGint num_samplers,
+                                void *fs,
+                                const void *const_buffer,
+                                VGint const_buffer_len);
+
+void renderer_filter(struct renderer *renderer,
+                     VGint x, VGint y, VGint w, VGint h,
+                     VGint sx, VGint sy, VGint sw, VGint sh);
+
+void renderer_filter_end(struct renderer *renderer);
+
+VGboolean renderer_polygon_stencil_begin(struct renderer *renderer,
+                                         struct pipe_vertex_element *velem,
+                                         VGFillRule rule,
+                                         VGboolean restore_dsa);
+
+void renderer_polygon_stencil(struct renderer *renderer,
+                              struct pipe_vertex_buffer *vbuf,
+                              VGuint mode, VGuint start, VGuint count);
+
+void renderer_polygon_stencil_end(struct renderer *renderer);
+
+VGboolean renderer_polygon_fill_begin(struct renderer *renderer,
+                                      VGboolean save_dsa);
+
+void renderer_polygon_fill(struct renderer *renderer,
+                           VGfloat min_x, VGfloat min_y,
+                           VGfloat max_x, VGfloat max_y);
+
+void renderer_polygon_fill_end(struct renderer *renderer);
+
 void renderer_texture_quad(struct renderer *,
                            struct pipe_resource *texture,
                            VGfloat x1offset, VGfloat y1offset,
@@ -57,13 +145,7 @@ void renderer_texture_quad(struct renderer *,
                            VGfloat x2, VGfloat y2,
                            VGfloat x3, VGfloat y3,
                            VGfloat x4, VGfloat y4);
-void renderer_copy_texture(struct renderer *r,
-                           struct pipe_sampler_view *src,
-                           VGfloat sx1, VGfloat sy1,
-                           VGfloat sx2, VGfloat sy2,
-                           struct pipe_resource *dst,
-                           VGfloat dx1, VGfloat dy1,
-                           VGfloat dx2, VGfloat dy2);
+
 void renderer_copy_surface(struct renderer *r,
                            struct pipe_surface *src,
                            int sx1, int sy1,
diff --git a/src/gallium/state_trackers/vega/shader.c b/src/gallium/state_trackers/vega/shader.c
index eab1349639..0ed721376c 100644
--- a/src/gallium/state_trackers/vega/shader.c
+++ b/src/gallium/state_trackers/vega/shader.c
@@ -31,22 +31,29 @@
 #include "paint.h"
 #include "mask.h"
 #include "image.h"
+#include "renderer.h"
 
 #include "pipe/p_context.h"
 #include "pipe/p_screen.h"
 #include "pipe/p_state.h"
 #include "util/u_inlines.h"
 #include "util/u_memory.h"
+#include "util/u_math.h"
+#include "util/u_format.h"
 
-#define MAX_CONSTANTS 20
+#define MAX_CONSTANTS 28
 
 struct shader {
    struct vg_context *context;
 
+   VGboolean color_transform;
    VGboolean masking;
    struct vg_paint *paint;
    struct vg_image *image;
 
+   struct matrix modelview;
+   struct matrix paint_matrix;
+
    VGboolean drawing_image;
    VGImageMode image_mode;
 
@@ -71,6 +78,11 @@ void shader_destroy(struct shader *shader)
    FREE(shader);
 }
 
+void shader_set_color_transform(struct shader *shader, VGboolean set)
+{
+   shader->color_transform = set;
+}
+
 void shader_set_masking(struct shader *shader, VGboolean set)
 {
    shader->masking = set;
@@ -91,49 +103,64 @@ struct vg_paint * shader_paint(struct shader *shader)
    return shader->paint;
 }
 
-
-static void setup_constant_buffer(struct shader *shader)
+static VGint setup_constant_buffer(struct shader *shader)
 {
-   struct vg_context *ctx = shader->context;
-   struct pipe_context *pipe = shader->context->pipe;
-   struct pipe_resource **cbuf = &shader->cbuf;
+   const struct vg_state *state = &shader->context->state.vg;
    VGint param_bytes = paint_constant_buffer_size(shader->paint);
-   float temp_buf[MAX_CONSTANTS];
+   VGint i;
+
+   param_bytes += sizeof(VGfloat) * 8;
+   assert(param_bytes <= sizeof(shader->constants));
+
+   if (state->color_transform) {
+      for (i = 0; i < 8; i++) {
+         VGfloat val = (i < 4) ? 127.0f : 1.0f;
+         shader->constants[i] =
+            CLAMP(state->color_transform_values[i], -val, val);
+      }
+   }
+   else {
+      memset(shader->constants, 0, sizeof(VGfloat) * 8);
+   }
 
-   assert(param_bytes <= sizeof(temp_buf));
-   paint_fill_constant_buffer(shader->paint, temp_buf);
+   paint_fill_constant_buffer(shader->paint,
+         &shader->paint_matrix, shader->constants + 8);
 
-   if (*cbuf == NULL ||
-       memcmp(temp_buf, shader->constants, param_bytes) != 0)
-   {
-      pipe_resource_reference(cbuf, NULL);
+   return param_bytes;
+}
+
+static VGboolean blend_use_shader(struct vg_context *ctx)
+{
+   VGboolean advanced_blending;
 
-      memcpy(shader->constants, temp_buf, param_bytes);
-      *cbuf = pipe_user_buffer_create(pipe->screen,
-                                      &shader->constants,
-                                      sizeof(shader->constants),
-				      PIPE_BIND_VERTEX_BUFFER);
+   switch (ctx->state.vg.blend_mode) {
+   case VG_BLEND_SRC_OVER:
+      advanced_blending =
+         util_format_has_alpha(ctx->draw_buffer->strb->format);
+      break;
+   case VG_BLEND_DST_OVER:
+   case VG_BLEND_MULTIPLY:
+   case VG_BLEND_SCREEN:
+   case VG_BLEND_DARKEN:
+   case VG_BLEND_LIGHTEN:
+   case VG_BLEND_ADDITIVE:
+      advanced_blending = VG_TRUE;
+      break;
+   default:
+      advanced_blending = VG_FALSE;
+      break;
    }
 
-   ctx->pipe->set_constant_buffer(ctx->pipe, PIPE_SHADER_FRAGMENT, 0, *cbuf);
+   return advanced_blending;
 }
 
 static VGint blend_bind_samplers(struct vg_context *ctx,
                                  struct pipe_sampler_state **samplers,
                                  struct pipe_sampler_view **sampler_views)
 {
-   VGBlendMode bmode = ctx->state.vg.blend_mode;
-
-   if (bmode == VG_BLEND_MULTIPLY ||
-       bmode == VG_BLEND_SCREEN ||
-       bmode == VG_BLEND_DARKEN ||
-       bmode == VG_BLEND_LIGHTEN) {
-      struct st_framebuffer *stfb = ctx->draw_buffer;
-
-      vg_prepare_blend_surface(ctx);
-
+   if (blend_use_shader(ctx)) {
       samplers[2] = &ctx->blend_sampler;
-      sampler_views[2] = stfb->blend_texture_view;
+      sampler_views[2] = vg_prepare_blend_surface(ctx);
 
       if (!samplers[0] || !sampler_views[0]) {
          samplers[0] = samplers[2];
@@ -149,10 +176,10 @@ static VGint blend_bind_samplers(struct vg_context *ctx,
    return 0;
 }
 
-static void setup_samplers(struct shader *shader)
+static VGint setup_samplers(struct shader *shader,
+                            struct pipe_sampler_state **samplers,
+                            struct pipe_sampler_view **sampler_views)
 {
-   struct pipe_sampler_state *samplers[PIPE_MAX_SAMPLERS];
-   struct pipe_sampler_view *sampler_views[PIPE_MAX_SAMPLERS];
    struct vg_context *ctx = shader->context;
    /* a little wonky: we use the num as a boolean that just says
     * whether any sampler/textures have been set. the actual numbering
@@ -179,10 +206,7 @@ static void setup_samplers(struct shader *shader)
    if (shader->drawing_image && shader->image)
       num += image_bind_samplers(shader->image, samplers, sampler_views);
 
-   if (num) {
-      cso_set_samplers(ctx->cso_context, 4, (const struct pipe_sampler_state **)samplers);
-      cso_set_fragment_sampler_views(ctx->cso_context, 4, sampler_views);
-   }
+   return (num) ? 4 : 0;
 }
 
 static INLINE VGboolean is_format_bw(struct shader *shader)
@@ -227,6 +251,9 @@ static void setup_shader_program(struct shader *shader)
       default:
          abort();
       }
+
+      if (paint_is_degenerate(shader->paint))
+         shader_id = VEGA_PAINT_DEGENERATE_SHADER;
    }
 
    /* second stage image */
@@ -246,43 +273,86 @@ static void setup_shader_program(struct shader *shader)
       }
    }
 
-   if (shader->masking)
-      shader_id |= VEGA_MASK_SHADER;
+   if (shader->color_transform)
+      shader_id |= VEGA_COLOR_TRANSFORM_SHADER;
 
-   switch(blend_mode) {
-   case VG_BLEND_MULTIPLY:
-      shader_id |= VEGA_BLEND_MULTIPLY_SHADER;
-      break;
-   case VG_BLEND_SCREEN:
-      shader_id |= VEGA_BLEND_SCREEN_SHADER;
-      break;
-   case VG_BLEND_DARKEN:
-      shader_id |= VEGA_BLEND_DARKEN_SHADER;
-      break;
-   case VG_BLEND_LIGHTEN:
-      shader_id |= VEGA_BLEND_LIGHTEN_SHADER;
-      break;
-   default:
-      /* handled by pipe_blend_state */
-      break;
+   if (blend_use_shader(ctx)) {
+      if (shader->drawing_image && shader->image_mode == VG_DRAW_IMAGE_STENCIL)
+         shader_id |= VEGA_ALPHA_PER_CHANNEL_SHADER;
+      else
+         shader_id |= VEGA_ALPHA_NORMAL_SHADER;
+
+      switch(blend_mode) {
+      case VG_BLEND_SRC:
+         shader_id |= VEGA_BLEND_SRC_SHADER;
+         break;
+      case VG_BLEND_SRC_OVER:
+         shader_id |= VEGA_BLEND_SRC_OVER_SHADER;
+         break;
+      case VG_BLEND_DST_OVER:
+         shader_id |= VEGA_BLEND_DST_OVER_SHADER;
+         break;
+      case VG_BLEND_SRC_IN:
+         shader_id |= VEGA_BLEND_SRC_IN_SHADER;
+         break;
+      case VG_BLEND_DST_IN:
+         shader_id |= VEGA_BLEND_DST_IN_SHADER;
+         break;
+      case VG_BLEND_MULTIPLY:
+         shader_id |= VEGA_BLEND_MULTIPLY_SHADER;
+         break;
+      case VG_BLEND_SCREEN:
+         shader_id |= VEGA_BLEND_SCREEN_SHADER;
+         break;
+      case VG_BLEND_DARKEN:
+         shader_id |= VEGA_BLEND_DARKEN_SHADER;
+         break;
+      case VG_BLEND_LIGHTEN:
+         shader_id |= VEGA_BLEND_LIGHTEN_SHADER;
+         break;
+      case VG_BLEND_ADDITIVE:
+         shader_id |= VEGA_BLEND_ADDITIVE_SHADER;
+         break;
+      default:
+         assert(0);
+         break;
+      }
+   }
+   else {
+      /* update alpha of the source */
+      if (shader->drawing_image && shader->image_mode == VG_DRAW_IMAGE_STENCIL)
+         shader_id |= VEGA_ALPHA_PER_CHANNEL_SHADER;
    }
 
+   if (shader->masking)
+      shader_id |= VEGA_MASK_SHADER;
+
    if (black_white)
       shader_id |= VEGA_BW_SHADER;
 
    shader->fs = shaders_cache_fill(ctx->sc, shader_id);
-   cso_set_fragment_shader_handle(ctx->cso_context, shader->fs);
 }
 
 
 void shader_bind(struct shader *shader)
 {
+   struct vg_context *ctx = shader->context;
+   struct pipe_sampler_state *samplers[PIPE_MAX_SAMPLERS];
+   struct pipe_sampler_view *sampler_views[PIPE_MAX_SAMPLERS];
+   VGint num_samplers, param_bytes;
+
    /* first resolve the real paint type */
    paint_resolve_type(shader->paint);
 
-   setup_constant_buffer(shader);
-   setup_samplers(shader);
+   num_samplers = setup_samplers(shader, samplers, sampler_views);
+   param_bytes = setup_constant_buffer(shader);
    setup_shader_program(shader);
+
+   renderer_validate_for_shader(ctx->renderer,
+         (const struct pipe_sampler_state **) samplers,
+         sampler_views, num_samplers,
+         &shader->modelview,
+         shader->fs, (const void *) shader->constants, param_bytes);
 }
 
 void shader_set_image_mode(struct shader *shader, VGImageMode image_mode)
@@ -309,3 +379,28 @@ void shader_set_image(struct shader *shader, struct vg_image *img)
 {
    shader->image = img;
 }
+
+/**
+ * Set the transformation to map a vertex to the surface coordinates.
+ */
+void shader_set_surface_matrix(struct shader *shader,
+                               const struct matrix *mat)
+{
+   shader->modelview = *mat;
+}
+
+/**
+ * Set the transformation to map a pixel to the paint coordinates.
+ */
+void shader_set_paint_matrix(struct shader *shader, const struct matrix *mat)
+{
+   const struct st_framebuffer *stfb = shader->context->draw_buffer;
+   const VGfloat px_center_offset = 0.5f;
+
+   memcpy(&shader->paint_matrix, mat, sizeof(*mat));
+
+   /* make it window-to-paint for the shaders */
+   matrix_translate(&shader->paint_matrix, px_center_offset,
+         stfb->height - 1.0f + px_center_offset);
+   matrix_scale(&shader->paint_matrix, 1.0f, -1.0f);
+}
diff --git a/src/gallium/state_trackers/vega/shader.h b/src/gallium/state_trackers/vega/shader.h
index 847eee6a31..8b97e537ef 100644
--- a/src/gallium/state_trackers/vega/shader.h
+++ b/src/gallium/state_trackers/vega/shader.h
@@ -33,10 +33,13 @@ struct shader;
 struct vg_paint;
 struct vg_context;
 struct vg_image;
+struct matrix;
 
 struct shader *shader_create(struct vg_context *context);
 void shader_destroy(struct shader *shader);
 
+void shader_set_color_transform(struct shader *shader, VGboolean set);
+
 void shader_set_masking(struct shader *shader, VGboolean set);
 VGboolean shader_is_masking(struct shader *shader);
 
@@ -51,6 +54,10 @@ VGboolean shader_drawing_image(struct shader *shader);
 
 void shader_set_image(struct shader *shader, struct vg_image *img);
 
+void shader_set_surface_matrix(struct shader *shader,
+                               const struct matrix *mat);
+void shader_set_paint_matrix(struct shader *shader, const struct matrix *mat);
+
 void shader_bind(struct shader *shader);
 
 #endif
diff --git a/src/gallium/state_trackers/vega/shaders_cache.c b/src/gallium/state_trackers/vega/shaders_cache.c
index e002a7ed42..023996ce2d 100644
--- a/src/gallium/state_trackers/vega/shaders_cache.c
+++ b/src/gallium/state_trackers/vega/shaders_cache.c
@@ -50,14 +50,16 @@
 
 /* Essentially we construct an ubber-shader based on the state
  * of the pipeline. The stages are:
- * 1) Fill (mandatory, solid color/gradient/pattern/image draw)
- * 2) Image composition (image mode multiply and stencil)
- * 3) Mask
- * 4) Extended blend (multiply/screen/darken/lighten)
- * 5) Premultiply/Unpremultiply
- * 6) Color transform (to black and white)
+ * 1) Paint generation (color/gradient/pattern)
+ * 2) Image composition (normal/multiply/stencil)
+ * 3) Color transform
+ * 4) Per-channel alpha generation
+ * 5) Extended blend (multiply/screen/darken/lighten)
+ * 6) Mask
+ * 7) Premultiply/Unpremultiply
+ * 8) Color transform (to black and white)
  */
-#define SHADER_STAGES 6
+#define SHADER_STAGES 8
 
 struct cached_shader {
    void *driver_shader;
@@ -86,13 +88,6 @@ static INLINE struct tgsi_token *tokens_from_assembly(const char *txt, int num_t
    return tokens;
 }
 
-#define ALL_FILLS (VEGA_SOLID_FILL_SHADER | \
-   VEGA_LINEAR_GRADIENT_SHADER | \
-   VEGA_RADIAL_GRADIENT_SHADER | \
-   VEGA_PATTERN_SHADER         | \
-   VEGA_IMAGE_NORMAL_SHADER)
-
-
 /*
 static const char max_shader_preamble[] =
    "FRAG\n"
@@ -257,98 +252,126 @@ create_shader(struct pipe_context *pipe,
               int id,
               struct pipe_shader_state *shader)
 {
-   int idx = 0;
+   int idx = 0, sh;
    const struct shader_asm_info * shaders[SHADER_STAGES];
 
-   /* the shader has to have a fill */
-   debug_assert(id & ALL_FILLS);
-
    /* first stage */
-   if (id & VEGA_SOLID_FILL_SHADER) {
-      debug_assert(idx == 0);
-      shaders[idx] = &shaders_asm[0];
-      debug_assert(shaders_asm[0].id == VEGA_SOLID_FILL_SHADER);
-      ++idx;
-   }
-   if ((id & VEGA_LINEAR_GRADIENT_SHADER)) {
-      debug_assert(idx == 0);
-      shaders[idx] = &shaders_asm[1];
-      debug_assert(shaders_asm[1].id == VEGA_LINEAR_GRADIENT_SHADER);
-      ++idx;
-   }
-   if ((id & VEGA_RADIAL_GRADIENT_SHADER)) {
-      debug_assert(idx == 0);
-      shaders[idx] = &shaders_asm[2];
-      debug_assert(shaders_asm[2].id == VEGA_RADIAL_GRADIENT_SHADER);
-      ++idx;
-   }
-   if ((id & VEGA_PATTERN_SHADER)) {
-      debug_assert(idx == 0);
-      debug_assert(shaders_asm[3].id == VEGA_PATTERN_SHADER);
-      shaders[idx] = &shaders_asm[3];
-      ++idx;
-   }
-   if ((id & VEGA_IMAGE_NORMAL_SHADER)) {
-      debug_assert(idx == 0);
-      debug_assert(shaders_asm[4].id == VEGA_IMAGE_NORMAL_SHADER);
-      shaders[idx] = &shaders_asm[4];
-      ++idx;
+   sh = SHADERS_GET_PAINT_SHADER(id);
+   switch (sh << SHADERS_PAINT_SHIFT) {
+   case VEGA_SOLID_FILL_SHADER:
+   case VEGA_LINEAR_GRADIENT_SHADER:
+   case VEGA_RADIAL_GRADIENT_SHADER:
+   case VEGA_PATTERN_SHADER:
+   case VEGA_PAINT_DEGENERATE_SHADER:
+      shaders[idx] = &shaders_paint_asm[(sh >> SHADERS_PAINT_SHIFT) - 1];
+      assert(shaders[idx]->id == sh);
+      idx++;
+      break;
+   default:
+      break;
    }
 
    /* second stage */
-   if ((id & VEGA_IMAGE_MULTIPLY_SHADER)) {
-      debug_assert(shaders_asm[5].id == VEGA_IMAGE_MULTIPLY_SHADER);
-      shaders[idx] = &shaders_asm[5];
-      ++idx;
-   } else if ((id & VEGA_IMAGE_STENCIL_SHADER)) {
-      debug_assert(shaders_asm[6].id == VEGA_IMAGE_STENCIL_SHADER);
-      shaders[idx] = &shaders_asm[6];
-      ++idx;
+   sh = SHADERS_GET_IMAGE_SHADER(id);
+   switch (sh) {
+   case VEGA_IMAGE_NORMAL_SHADER:
+   case VEGA_IMAGE_MULTIPLY_SHADER:
+   case VEGA_IMAGE_STENCIL_SHADER:
+      shaders[idx] = &shaders_image_asm[(sh >> SHADERS_IMAGE_SHIFT) - 1];
+      assert(shaders[idx]->id == sh);
+      idx++;
+      break;
+   default:
+      break;
    }
 
+   /* sanity check */
+   assert(idx == ((!sh || sh == VEGA_IMAGE_NORMAL_SHADER) ? 1 : 2));
+
    /* third stage */
-   if ((id & VEGA_MASK_SHADER)) {
-      debug_assert(idx == 1);
-      debug_assert(shaders_asm[7].id == VEGA_MASK_SHADER);
-      shaders[idx] = &shaders_asm[7];
-      ++idx;
+   sh = SHADERS_GET_COLOR_TRANSFORM_SHADER(id);
+   switch (sh) {
+   case VEGA_COLOR_TRANSFORM_SHADER:
+      shaders[idx] = &shaders_color_transform_asm[
+         (sh >> SHADERS_COLOR_TRANSFORM_SHIFT) - 1];
+      assert(shaders[idx]->id == sh);
+      idx++;
+      break;
+   default:
+      break;
    }
 
    /* fourth stage */
-   if ((id & VEGA_BLEND_MULTIPLY_SHADER)) {
-      debug_assert(shaders_asm[8].id == VEGA_BLEND_MULTIPLY_SHADER);
-      shaders[idx] = &shaders_asm[8];
-      ++idx;
-   } else if ((id & VEGA_BLEND_SCREEN_SHADER)) {
-      debug_assert(shaders_asm[9].id == VEGA_BLEND_SCREEN_SHADER);
-      shaders[idx] = &shaders_asm[9];
-      ++idx;
-   } else if ((id & VEGA_BLEND_DARKEN_SHADER)) {
-      debug_assert(shaders_asm[10].id == VEGA_BLEND_DARKEN_SHADER);
-      shaders[idx] = &shaders_asm[10];
-      ++idx;
-   } else if ((id & VEGA_BLEND_LIGHTEN_SHADER)) {
-      debug_assert(shaders_asm[11].id == VEGA_BLEND_LIGHTEN_SHADER);
-      shaders[idx] = &shaders_asm[11];
-      ++idx;
+   sh = SHADERS_GET_ALPHA_SHADER(id);
+   switch (sh) {
+   case VEGA_ALPHA_NORMAL_SHADER:
+   case VEGA_ALPHA_PER_CHANNEL_SHADER:
+      shaders[idx] = &shaders_alpha_asm[
+         (sh >> SHADERS_ALPHA_SHIFT) - 1];
+      assert(shaders[idx]->id == sh);
+      idx++;
+      break;
+   default:
+      break;
    }
 
    /* fifth stage */
-   if ((id & VEGA_PREMULTIPLY_SHADER)) {
-      debug_assert(shaders_asm[12].id == VEGA_PREMULTIPLY_SHADER);
-      shaders[idx] = &shaders_asm[12];
-      ++idx;
-   } else if ((id & VEGA_UNPREMULTIPLY_SHADER)) {
-      debug_assert(shaders_asm[13].id == VEGA_UNPREMULTIPLY_SHADER);
-      shaders[idx] = &shaders_asm[13];
-      ++idx;
+   sh = SHADERS_GET_BLEND_SHADER(id);
+   switch (sh) {
+   case VEGA_BLEND_SRC_SHADER:
+   case VEGA_BLEND_SRC_OVER_SHADER:
+   case VEGA_BLEND_DST_OVER_SHADER:
+   case VEGA_BLEND_SRC_IN_SHADER:
+   case VEGA_BLEND_DST_IN_SHADER:
+   case VEGA_BLEND_MULTIPLY_SHADER:
+   case VEGA_BLEND_SCREEN_SHADER:
+   case VEGA_BLEND_DARKEN_SHADER:
+   case VEGA_BLEND_LIGHTEN_SHADER:
+   case VEGA_BLEND_ADDITIVE_SHADER:
+      shaders[idx] = &shaders_blend_asm[(sh >> SHADERS_BLEND_SHIFT) - 1];
+      assert(shaders[idx]->id == sh);
+      idx++;
+      break;
+   default:
+      break;
    }
 
    /* sixth stage */
-   if ((id & VEGA_BW_SHADER)) {
-      debug_assert(shaders_asm[14].id == VEGA_BW_SHADER);
-      shaders[idx] = &shaders_asm[14];
-      ++idx;
+   sh = SHADERS_GET_MASK_SHADER(id);
+   switch (sh) {
+   case VEGA_MASK_SHADER:
+      shaders[idx] = &shaders_mask_asm[(sh >> SHADERS_MASK_SHIFT) - 1];
+      assert(shaders[idx]->id == sh);
+      idx++;
+      break;
+   default:
+      break;
+   }
+
+   /* seventh stage */
+   sh = SHADERS_GET_PREMULTIPLY_SHADER(id);
+   switch (sh) {
+   case VEGA_PREMULTIPLY_SHADER:
+   case VEGA_UNPREMULTIPLY_SHADER:
+      shaders[idx] = &shaders_premultiply_asm[
+         (sh >> SHADERS_PREMULTIPLY_SHIFT) - 1];
+      assert(shaders[idx]->id == sh);
+      idx++;
+      break;
+   default:
+      break;
+   }
+
+   /* eighth stage */
+   sh = SHADERS_GET_BW_SHADER(id);
+   switch (sh) {
+   case VEGA_BW_SHADER:
+      shaders[idx] = &shaders_bw_asm[(sh >> SHADERS_BW_SHIFT) - 1];
+      assert(shaders[idx]->id == sh);
+      idx++;
+      break;
+   default:
+      break;
    }
 
    return combine_shaders(shaders, idx, pipe, shader);
diff --git a/src/gallium/state_trackers/vega/shaders_cache.h b/src/gallium/state_trackers/vega/shaders_cache.h
index feca58b61a..05014f25dc 100644
--- a/src/gallium/state_trackers/vega/shaders_cache.h
+++ b/src/gallium/state_trackers/vega/shaders_cache.h
@@ -33,26 +33,69 @@ struct pipe_context;
 struct tgsi_token;
 struct shaders_cache;
 
+#define _SHADERS_PAINT_BITS            3
+#define _SHADERS_IMAGE_BITS            2
+#define _SHADERS_COLOR_TRANSFORM_BITS  1
+#define _SHADERS_ALPHA_BITS            2
+#define _SHADERS_BLEND_BITS            4
+#define _SHADERS_MASK_BITS             1
+#define _SHADERS_PREMULTIPLY_BITS      2
+#define _SHADERS_BW_BITS               1
+
+#define SHADERS_PAINT_SHIFT           (0)
+#define SHADERS_IMAGE_SHIFT           (SHADERS_PAINT_SHIFT + _SHADERS_PAINT_BITS)
+#define SHADERS_COLOR_TRANSFORM_SHIFT (SHADERS_IMAGE_SHIFT + _SHADERS_IMAGE_BITS)
+#define SHADERS_ALPHA_SHIFT           (SHADERS_COLOR_TRANSFORM_SHIFT + _SHADERS_COLOR_TRANSFORM_BITS)
+#define SHADERS_BLEND_SHIFT           (SHADERS_ALPHA_SHIFT + _SHADERS_ALPHA_BITS)
+#define SHADERS_MASK_SHIFT            (SHADERS_BLEND_SHIFT + _SHADERS_BLEND_BITS)
+#define SHADERS_PREMULTIPLY_SHIFT     (SHADERS_MASK_SHIFT + _SHADERS_MASK_BITS)
+#define SHADERS_BW_SHIFT              (SHADERS_PREMULTIPLY_SHIFT + _SHADERS_PREMULTIPLY_BITS)
+
+#define _SHADERS_GET_STAGE(stage, id) \
+   ((id) & (((1 << _SHADERS_ ## stage ## _BITS) - 1) << SHADERS_ ## stage ## _SHIFT))
+
+#define SHADERS_GET_PAINT_SHADER(id)           _SHADERS_GET_STAGE(PAINT, id)
+#define SHADERS_GET_IMAGE_SHADER(id)           _SHADERS_GET_STAGE(IMAGE, id)
+#define SHADERS_GET_COLOR_TRANSFORM_SHADER(id) _SHADERS_GET_STAGE(COLOR_TRANSFORM, id)
+#define SHADERS_GET_ALPHA_SHADER(id)           _SHADERS_GET_STAGE(ALPHA, id)
+#define SHADERS_GET_BLEND_SHADER(id)           _SHADERS_GET_STAGE(BLEND, id)
+#define SHADERS_GET_MASK_SHADER(id)            _SHADERS_GET_STAGE(MASK, id)
+#define SHADERS_GET_PREMULTIPLY_SHADER(id)     _SHADERS_GET_STAGE(PREMULTIPLY, id)
+#define SHADERS_GET_BW_SHADER(id)              _SHADERS_GET_STAGE(BW, id)
+
 enum VegaShaderType {
-   VEGA_SOLID_FILL_SHADER         = 1 <<  0,
-   VEGA_LINEAR_GRADIENT_SHADER    = 1 <<  1,
-   VEGA_RADIAL_GRADIENT_SHADER    = 1 <<  2,
-   VEGA_PATTERN_SHADER            = 1 <<  3,
-   VEGA_IMAGE_NORMAL_SHADER       = 1 <<  4,
-   VEGA_IMAGE_MULTIPLY_SHADER     = 1 <<  5,
-   VEGA_IMAGE_STENCIL_SHADER      = 1 <<  6,
+   VEGA_SOLID_FILL_SHADER         = 1 << SHADERS_PAINT_SHIFT,
+   VEGA_LINEAR_GRADIENT_SHADER    = 2 << SHADERS_PAINT_SHIFT,
+   VEGA_RADIAL_GRADIENT_SHADER    = 3 << SHADERS_PAINT_SHIFT,
+   VEGA_PATTERN_SHADER            = 4 << SHADERS_PAINT_SHIFT,
+   VEGA_PAINT_DEGENERATE_SHADER   = 5 << SHADERS_PAINT_SHIFT,
+
+   VEGA_IMAGE_NORMAL_SHADER       = 1 << SHADERS_IMAGE_SHIFT,
+   VEGA_IMAGE_MULTIPLY_SHADER     = 2 << SHADERS_IMAGE_SHIFT,
+   VEGA_IMAGE_STENCIL_SHADER      = 3 << SHADERS_IMAGE_SHIFT,
+
+   VEGA_COLOR_TRANSFORM_SHADER    = 1 <<  SHADERS_COLOR_TRANSFORM_SHIFT,
+
+   VEGA_ALPHA_NORMAL_SHADER       = 1 <<  SHADERS_ALPHA_SHIFT,
+   VEGA_ALPHA_PER_CHANNEL_SHADER  = 2 <<  SHADERS_ALPHA_SHIFT,
 
-   VEGA_MASK_SHADER               = 1 <<  7,
+   VEGA_BLEND_SRC_SHADER          = 1 << SHADERS_BLEND_SHIFT,
+   VEGA_BLEND_SRC_OVER_SHADER     = 2 << SHADERS_BLEND_SHIFT,
+   VEGA_BLEND_DST_OVER_SHADER     = 3 << SHADERS_BLEND_SHIFT,
+   VEGA_BLEND_SRC_IN_SHADER       = 4 << SHADERS_BLEND_SHIFT,
+   VEGA_BLEND_DST_IN_SHADER       = 5 << SHADERS_BLEND_SHIFT,
+   VEGA_BLEND_MULTIPLY_SHADER     = 6 << SHADERS_BLEND_SHIFT,
+   VEGA_BLEND_SCREEN_SHADER       = 7 << SHADERS_BLEND_SHIFT,
+   VEGA_BLEND_DARKEN_SHADER       = 8 << SHADERS_BLEND_SHIFT,
+   VEGA_BLEND_LIGHTEN_SHADER      = 9 << SHADERS_BLEND_SHIFT,
+   VEGA_BLEND_ADDITIVE_SHADER     = 10<< SHADERS_BLEND_SHIFT,
 
-   VEGA_BLEND_MULTIPLY_SHADER     = 1 <<  8,
-   VEGA_BLEND_SCREEN_SHADER       = 1 <<  9,
-   VEGA_BLEND_DARKEN_SHADER       = 1 << 10,
-   VEGA_BLEND_LIGHTEN_SHADER      = 1 << 11,
+   VEGA_MASK_SHADER               = 1 << SHADERS_MASK_SHIFT,
 
-   VEGA_PREMULTIPLY_SHADER        = 1 << 12,
-   VEGA_UNPREMULTIPLY_SHADER      = 1 << 13,
+   VEGA_PREMULTIPLY_SHADER        = 1 << SHADERS_PREMULTIPLY_SHIFT,
+   VEGA_UNPREMULTIPLY_SHADER      = 2 << SHADERS_PREMULTIPLY_SHIFT,
 
-   VEGA_BW_SHADER                 = 1 << 14
+   VEGA_BW_SHADER                 = 1 << SHADERS_BW_SHIFT
 };
 
 struct vg_shader {
diff --git a/src/gallium/state_trackers/vega/st_inlines.h b/src/gallium/state_trackers/vega/st_inlines.h
deleted file mode 100644
index 7eaa67c76a..0000000000
--- a/src/gallium/state_trackers/vega/st_inlines.h
+++ /dev/null
@@ -1,122 +0,0 @@
-/**************************************************************************
- * 
- * Copyright 2009 VMware, Inc.
- * All Rights Reserved.
- * 
- * Permission is hereby granted, free of charge, to any person obtaining a
- * copy of this software and associated documentation files (the
- * "Software"), to deal in the Software without restriction, including
- * without limitation 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 VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR
- * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
- * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
- * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
- * 
- **************************************************************************/
-
-/**
- * Functions for checking if buffers/textures are referenced when we need
- * to read/write from/to them.  Flush when needed.
- */
-
-#ifndef ST_INLINES_H
-#define ST_INLINES_H
-
-#include "vg_context.h"
-
-#include "pipe/p_context.h"
-#include "pipe/p_screen.h"
-#include "pipe/p_defines.h"
-#include "util/u_inlines.h"
-#include "pipe/p_state.h"
-
-static INLINE struct pipe_transfer *
-st_cond_flush_get_transfer(struct vg_context *st,
-			       struct pipe_resource *pt,
-			       unsigned int face,
-			       unsigned int level,
-			       unsigned int zslice,
-			       enum pipe_transfer_usage usage,
-			       unsigned int x, unsigned int y,
-			       unsigned int w, unsigned int h)
-{
-   struct pipe_context *pipe = st->pipe;
-
-   return pipe_get_transfer(pipe, pt, face, level, zslice, usage,
-			    x, y, w, h);
-}
-
-static INLINE struct pipe_transfer *
-st_no_flush_get_transfer(struct vg_context *st,
-			     struct pipe_resource *pt,
-			     unsigned int face,
-			     unsigned int level,
-			     unsigned int zslice,
-			     enum pipe_transfer_usage usage,
-			     unsigned int x, unsigned int y,
-			     unsigned int w, unsigned int h)
-{
-   struct pipe_context *pipe = st->pipe;
-
-   return pipe_get_transfer(pipe, pt, face, level,
-			    zslice, usage, x, y, w, h);
-}
-
-
-static INLINE void
-st_cond_flush_pipe_buffer_write(struct vg_context *st,
-				struct pipe_resource *buf,
-				unsigned int offset,
-				unsigned int size,
-				const void * data)
-{
-   struct pipe_context *pipe = st->pipe;
-
-   pipe_buffer_write(pipe, buf, offset, size, data);
-}
-
-static INLINE void
-st_no_flush_pipe_buffer_write(struct vg_context *st,
-			      struct pipe_resource *buf,
-			      unsigned int offset,
-			      unsigned int size,
-			      const void * data)
-{
-   pipe_buffer_write(st->pipe, buf, offset, size, data);
-}
-
-static INLINE void
-st_cond_flush_pipe_buffer_read(struct vg_context *st,
-			       struct pipe_resource *buf,
-			       unsigned int offset,
-			       unsigned int size,
-			       void * data)
-{
-   struct pipe_context *pipe = st->pipe;
-
-   pipe_buffer_read(pipe, buf, offset, size, data);
-}
-
-static INLINE void
-st_no_flush_pipe_buffer_read(struct vg_context *st,
-			     struct pipe_resource *buf,
-			     unsigned int offset,
-			     unsigned int size,
-			     void * data)
-{
-   pipe_buffer_read(st->pipe, buf, offset, size, data);
-}
-
-#endif
-
diff --git a/src/gallium/state_trackers/vega/text.c b/src/gallium/state_trackers/vega/text.c
new file mode 100644
index 0000000000..01ff602f58
--- /dev/null
+++ b/src/gallium/state_trackers/vega/text.c
@@ -0,0 +1,250 @@
+/**************************************************************************
+ *
+ * Copyright 2010 LunarG, Inc.  All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation 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 VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION 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 "util/u_memory.h"
+#include "cso_cache/cso_hash.h"
+
+#include "text.h"
+#include "image.h"
+#include "path.h"
+#include "api.h"
+
+#ifdef OPENVG_VERSION_1_1
+
+struct vg_font {
+   struct vg_object base;
+   struct cso_hash *glyphs;
+};
+
+struct vg_glyph {
+   struct vg_object *object; /* it could be NULL */
+   VGboolean is_hinted;
+   VGfloat glyph_origin[2];
+   VGfloat escapement[2];
+};
+
+static VGboolean del_glyph(struct vg_font *font,
+                           VGuint glyphIndex)
+{
+   struct vg_glyph *glyph;
+
+   glyph = (struct vg_glyph *)
+      cso_hash_take(font->glyphs, (unsigned) glyphIndex);
+   if (glyph)
+      FREE(glyph);
+
+   return (glyph != NULL);
+}
+
+static void add_glyph(struct vg_font *font,
+                      VGuint glyphIndex,
+                      struct vg_object *obj,
+                      VGboolean isHinted,
+                      const VGfloat glyphOrigin[2],
+                      const VGfloat escapement[2])
+{
+   struct vg_glyph *glyph;
+
+   /* remove the existing one */
+   del_glyph(font, glyphIndex);
+
+   glyph = CALLOC_STRUCT(vg_glyph);
+   glyph->object = obj;
+   glyph->is_hinted = isHinted;
+   memcpy(glyph->glyph_origin, glyphOrigin, sizeof(glyphOrigin));
+   memcpy(glyph->escapement, escapement, sizeof(escapement));
+
+   cso_hash_insert(font->glyphs, (unsigned) glyphIndex, glyph);
+}
+
+static struct vg_glyph *get_glyph(struct vg_font *font,
+                                  VGuint glyphIndex)
+{
+   struct cso_hash_iter iter;
+
+   iter = cso_hash_find(font->glyphs, (unsigned) glyphIndex);
+   return (struct vg_glyph *) cso_hash_iter_data(iter);
+}
+
+static void vg_render_glyph(struct vg_context *ctx,
+                            struct vg_glyph *glyph,
+                            VGbitfield paintModes,
+                            VGboolean allowAutoHinting)
+{
+   if (glyph->object && paintModes) {
+      struct vg_state *state = &ctx->state.vg;
+      struct matrix m;
+
+      m = state->glyph_user_to_surface_matrix;
+      matrix_translate(&m,
+            state->glyph_origin[0].f - glyph->glyph_origin[0],
+            state->glyph_origin[1].f - glyph->glyph_origin[1]);
+
+      if (glyph->object->type == VG_OBJECT_PATH) {
+         path_render((struct path *) glyph->object, paintModes, &m);
+      }
+      else {
+         assert(glyph->object->type == VG_OBJECT_IMAGE);
+         image_draw((struct vg_image *) glyph->object, &m);
+      }
+   }
+}
+
+static void vg_advance_glyph(struct vg_context *ctx,
+                             struct vg_glyph *glyph,
+                             VGfloat adjustment_x,
+                             VGfloat adjustment_y,
+                             VGboolean last)
+{
+   struct vg_value *glyph_origin = ctx->state.vg.glyph_origin;
+
+   glyph_origin[0].f += glyph->escapement[0] + adjustment_x;
+   glyph_origin[1].f += glyph->escapement[1] + adjustment_y;
+
+   if (last) {
+      glyph_origin[0].i = float_to_int_floor(glyph_origin[0].f);
+      glyph_origin[1].i = float_to_int_floor(glyph_origin[1].f);
+   }
+}
+
+struct vg_font *font_create(VGint glyphCapacityHint)
+{
+   struct vg_context *ctx = vg_current_context();
+   struct vg_font *font;
+
+   font = CALLOC_STRUCT(vg_font);
+   vg_init_object(&font->base, ctx, VG_OBJECT_FONT);
+   font->glyphs = cso_hash_create();
+
+   vg_context_add_object(ctx, VG_OBJECT_FONT, font);
+
+   return font;
+}
+
+void font_destroy(struct vg_font *font)
+{
+   struct vg_context *ctx = vg_current_context();
+   struct cso_hash_iter iter;
+
+   vg_context_remove_object(ctx, VG_OBJECT_FONT, font);
+
+   iter = cso_hash_first_node(font->glyphs);
+   while (!cso_hash_iter_is_null(iter)) {
+      struct vg_glyph *glyph = (struct vg_glyph *) cso_hash_iter_data(iter);
+      FREE(glyph);
+      iter = cso_hash_iter_next(iter);
+   }
+   cso_hash_delete(font->glyphs);
+
+   FREE(font);
+}
+
+void font_set_glyph_to_path(struct vg_font *font,
+                            VGuint glyphIndex,
+                            struct path *path,
+                            VGboolean isHinted,
+                            const VGfloat glyphOrigin[2],
+                            const VGfloat escapement[2])
+{
+   add_glyph(font, glyphIndex, (struct vg_object *) path,
+         isHinted, glyphOrigin, escapement);
+}
+
+void font_set_glyph_to_image(struct vg_font *font,
+                             VGuint glyphIndex,
+                             struct vg_image *image,
+                             const VGfloat glyphOrigin[2],
+                             const VGfloat escapement[2])
+{
+   add_glyph(font, glyphIndex, (struct vg_object *) image,
+         VG_TRUE, glyphOrigin, escapement);
+}
+
+void font_clear_glyph(struct vg_font *font,
+                      VGuint glyphIndex)
+{
+   if (!del_glyph(font, glyphIndex)) {
+      struct vg_context *ctx = vg_current_context();
+      vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR);
+   }
+}
+
+void font_draw_glyph(struct vg_font *font,
+                     VGuint glyphIndex,
+                     VGbitfield paintModes,
+                     VGboolean allowAutoHinting)
+{
+   struct vg_context *ctx = vg_current_context();
+   struct vg_glyph *glyph;
+
+   glyph = get_glyph(font, glyphIndex);
+   if (!glyph) {
+      vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR);
+      return;
+   }
+
+   vg_render_glyph(ctx, glyph, paintModes, allowAutoHinting);
+   vg_advance_glyph(ctx, glyph, 0.0f, 0.0f, VG_TRUE);
+}
+
+void font_draw_glyphs(struct vg_font *font,
+                      VGint glyphCount,
+                      const VGuint *glyphIndices,
+                      const VGfloat *adjustments_x,
+                      const VGfloat *adjustments_y,
+                      VGbitfield paintModes,
+                      VGboolean allowAutoHinting)
+{
+   struct vg_context *ctx = vg_current_context();
+   VGint i;
+
+   for (i = 0; i < glyphCount; ++i) {
+      if (!get_glyph(font, glyphIndices[i])) {
+         vg_set_error(ctx, VG_ILLEGAL_ARGUMENT_ERROR);
+         return;
+      }
+   }
+
+   for (i = 0; i < glyphCount; ++i) {
+      struct vg_glyph *glyph;
+      VGfloat adj_x, adj_y;
+
+      glyph = get_glyph(font, glyphIndices[i]);
+
+      vg_render_glyph(ctx, glyph, paintModes, allowAutoHinting);
+
+      adj_x = (adjustments_x) ? adjustments_x[i] : 0.0f;
+      adj_y = (adjustments_y) ? adjustments_y[i] : 0.0f;
+      vg_advance_glyph(ctx, glyph, adj_x, adj_y, (i == glyphCount - 1));
+   }
+}
+
+VGint font_num_glyphs(struct vg_font *font)
+{
+   return cso_hash_size(font->glyphs);
+}
+
+#endif /* OPENVG_VERSION_1_1 */
diff --git a/src/gallium/state_trackers/vega/text.h b/src/gallium/state_trackers/vega/text.h
new file mode 100644
index 0000000000..6b3fa7323e
--- /dev/null
+++ b/src/gallium/state_trackers/vega/text.h
@@ -0,0 +1,71 @@
+/**************************************************************************
+ *
+ * Copyright 2010 LunarG, Inc.  All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation 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 VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION 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 _TEXT_H
+#define _TEXT_H
+
+#include "vg_context.h"
+#include "cso_cache/cso_hash.h"
+
+struct vg_font;
+struct vg_image;
+struct path;
+
+struct vg_font *font_create(VGint glyphCapacityHint);
+void font_destroy(struct vg_font *font);
+
+void font_set_glyph_to_path(struct vg_font *font,
+                            VGuint glyphIndex,
+                            struct path *path,
+                            VGboolean isHinted,
+                            const VGfloat glyphOrigin[2],
+                            const VGfloat escapement[2]);
+
+void font_set_glyph_to_image(struct vg_font *font,
+                             VGuint glyphIndex,
+                             struct vg_image *image,
+                             const VGfloat glyphOrigin[2],
+                             const VGfloat escapement[2]);
+
+void font_clear_glyph(struct vg_font *font,
+                      VGuint glyphIndex);
+
+void font_draw_glyph(struct vg_font *font,
+                     VGuint glyphIndex,
+                     VGbitfield paintModes,
+                     VGboolean allowAutoHinting);
+
+void font_draw_glyphs(struct vg_font *font,
+                      VGint glyphCount,
+                      const VGuint *glyphIndices,
+                      const VGfloat *adjustments_x,
+                      const VGfloat *adjustments_y,
+                      VGbitfield paintModes,
+                      VGboolean allowAutoHinting);
+
+VGint font_num_glyphs(struct vg_font *font);
+
+#endif /* _TEXT_H */
diff --git a/src/gallium/state_trackers/vega/vg_context.c b/src/gallium/state_trackers/vega/vg_context.c
index b1f98fa477..0844012cc3 100644
--- a/src/gallium/state_trackers/vega/vg_context.c
+++ b/src/gallium/state_trackers/vega/vg_context.c
@@ -31,9 +31,9 @@
 #include "shaders_cache.h"
 #include "shader.h"
 #include "asm_util.h"
-#include "st_inlines.h"
 #include "vg_manager.h"
 #include "api.h"
+#include "mask.h"
 
 #include "pipe/p_context.h"
 #include "util/u_inlines.h"
@@ -44,6 +44,8 @@
 #include "util/u_memory.h"
 #include "util/u_blit.h"
 #include "util/u_sampler.h"
+#include "util/u_surface.h"
+#include "util/u_format.h"
 
 struct vg_context *_vg_context = 0;
 
@@ -52,19 +54,6 @@ struct vg_context * vg_current_context(void)
    return _vg_context;
 }
 
-static void init_clear(struct vg_context *st)
-{
-   struct pipe_context *pipe = st->pipe;
-
-   /* rasterizer state: bypass clipping */
-   memset(&st->clear.raster, 0, sizeof(st->clear.raster));
-   st->clear.raster.gl_rasterization_rules = 1;
-
-   /* fragment shader state: color pass-through program */
-   st->clear.fs =
-      util_make_fragment_passthrough_shader(pipe);
-}
-
 /**
  * A depth/stencil rb will be needed regardless of what the visual says.
  */
@@ -101,7 +90,6 @@ struct vg_context * vg_create_context(struct pipe_context *pipe,
                                       struct vg_context *share)
 {
    struct vg_context *ctx;
-   unsigned i;
 
    ctx = CALLOC_STRUCT(vg_context);
 
@@ -118,8 +106,6 @@ struct vg_context * vg_create_context(struct pipe_context *pipe,
 
    ctx->cso_context = cso_create_context(pipe);
 
-   init_clear(ctx);
-
    ctx->default_paint = paint_create(ctx);
    ctx->state.vg.stroke_paint = ctx->default_paint;
    ctx->state.vg.fill_paint = ctx->default_paint;
@@ -127,6 +113,7 @@ struct vg_context * vg_create_context(struct pipe_context *pipe,
 
    ctx->mask.sampler.wrap_s = PIPE_TEX_WRAP_CLAMP_TO_EDGE;
    ctx->mask.sampler.wrap_t = PIPE_TEX_WRAP_CLAMP_TO_EDGE;
+   ctx->mask.sampler.wrap_r = PIPE_TEX_WRAP_CLAMP_TO_EDGE;
    ctx->mask.sampler.min_mip_filter = PIPE_TEX_MIPFILTER_NONE;
    ctx->mask.sampler.min_img_filter = PIPE_TEX_FILTER_NEAREST;
    ctx->mask.sampler.mag_img_filter = PIPE_TEX_FILTER_NEAREST;
@@ -134,18 +121,12 @@ struct vg_context * vg_create_context(struct pipe_context *pipe,
 
    ctx->blend_sampler.wrap_s = PIPE_TEX_WRAP_CLAMP_TO_EDGE;
    ctx->blend_sampler.wrap_t = PIPE_TEX_WRAP_CLAMP_TO_EDGE;
+   ctx->blend_sampler.wrap_r = PIPE_TEX_WRAP_CLAMP_TO_EDGE;
    ctx->blend_sampler.min_mip_filter = PIPE_TEX_MIPFILTER_NONE;
    ctx->blend_sampler.min_img_filter = PIPE_TEX_FILTER_NEAREST;
    ctx->blend_sampler.mag_img_filter = PIPE_TEX_FILTER_NEAREST;
    ctx->blend_sampler.normalized_coords = 0;
 
-   for (i = 0; i < 2; i++) {
-      ctx->velems[i].src_offset = i * 4 * sizeof(float);
-      ctx->velems[i].instance_divisor = 0;
-      ctx->velems[i].vertex_buffer_index = 0;
-      ctx->velems[i].src_format = PIPE_FORMAT_R32G32B32A32_FLOAT;
-   }
-
    vg_set_error(ctx, VG_NO_ERROR);
 
    ctx->owned_objects[VG_OBJECT_PAINT] = cso_hash_create();
@@ -166,7 +147,6 @@ struct vg_context * vg_create_context(struct pipe_context *pipe,
 void vg_destroy_context(struct vg_context *ctx)
 {
    struct pipe_resource **cbuf = &ctx->mask.cbuf;
-   struct pipe_resource **vsbuf = &ctx->vs_const_buffer;
 
    util_destroy_blit(ctx->blit);
    renderer_destroy(ctx->renderer);
@@ -177,29 +157,6 @@ void vg_destroy_context(struct vg_context *ctx)
    if (*cbuf)
       pipe_resource_reference(cbuf, NULL);
 
-   if (*vsbuf)
-      pipe_resource_reference(vsbuf, NULL);
-
-   if (ctx->clear.fs) {
-      cso_delete_fragment_shader(ctx->cso_context, ctx->clear.fs);
-      ctx->clear.fs = NULL;
-   }
-
-   if (ctx->plain_vs) {
-      vg_shader_destroy(ctx, ctx->plain_vs);
-      ctx->plain_vs = NULL;
-   }
-   if (ctx->clear_vs) {
-      vg_shader_destroy(ctx, ctx->clear_vs);
-      ctx->clear_vs = NULL;
-   }
-   if (ctx->texture_vs) {
-      vg_shader_destroy(ctx, ctx->texture_vs);
-      ctx->texture_vs = NULL;
-   }
-
-   if (ctx->pass_through_depth_fs)
-      vg_shader_destroy(ctx, ctx->pass_through_depth_fs);
    if (ctx->mask.union_fs)
       vg_shader_destroy(ctx, ctx->mask.union_fs);
    if (ctx->mask.intersect_fs)
@@ -266,186 +223,195 @@ void vg_context_remove_object(struct vg_context *ctx,
    }
 }
 
-static void update_clip_state(struct vg_context *ctx)
+static struct pipe_resource *
+create_texture(struct pipe_context *pipe, enum pipe_format format,
+                    VGint width, VGint height)
 {
-   struct pipe_depth_stencil_alpha_state *dsa = &ctx->state.g3d.dsa;
-   struct vg_state *state =  &ctx->state.vg;
-
-   memset(dsa, 0, sizeof(struct pipe_depth_stencil_alpha_state));
-
-   if (state->scissoring) {
-      struct pipe_blend_state *blend = &ctx->state.g3d.blend;
-      struct pipe_framebuffer_state *fb = &ctx->state.g3d.fb;
-      int i;
-
-      dsa->depth.writemask = 1;/*glDepthMask(TRUE);*/
-      dsa->depth.func = PIPE_FUNC_ALWAYS;
-      dsa->depth.enabled = 1;
-
-      cso_save_blend(ctx->cso_context);
-      cso_save_fragment_shader(ctx->cso_context);
-      /* set a passthrough shader */
-      if (!ctx->pass_through_depth_fs)
-         ctx->pass_through_depth_fs = shader_create_from_text(ctx->pipe,
-                                                              pass_through_depth_asm,
-                                                              40,
-                                                              PIPE_SHADER_FRAGMENT);
-      cso_set_fragment_shader_handle(ctx->cso_context,
-                                     ctx->pass_through_depth_fs->driver);
-      cso_set_depth_stencil_alpha(ctx->cso_context, dsa);
-
-      ctx->pipe->clear(ctx->pipe, PIPE_CLEAR_DEPTHSTENCIL, NULL, 1.0, 0);
-
-      /* disable color writes */
-      blend->rt[0].colormask = 0; /*disable colorwrites*/
-      cso_set_blend(ctx->cso_context, blend);
-
-      /* enable scissoring */
-      for (i = 0; i < state->scissor_rects_num; ++i) {
-         const float x      = state->scissor_rects[i * 4 + 0].f;
-         const float y      = state->scissor_rects[i * 4 + 1].f;
-         const float width  = state->scissor_rects[i * 4 + 2].f;
-         const float height = state->scissor_rects[i * 4 + 3].f;
-         VGfloat minx, miny, maxx, maxy;
-
-         minx = 0;
-         miny = 0;
-         maxx = fb->width;
-         maxy = fb->height;
-
-         if (x > minx)
-            minx = x;
-         if (y > miny)
-            miny = y;
-
-         if (x + width < maxx)
-            maxx = x + width;
-         if (y + height < maxy)
-            maxy = y + height;
-
-         /* check for null space */
-         if (minx >= maxx || miny >= maxy)
-            minx = miny = maxx = maxy = 0;
-
-         /*glClear(GL_DEPTH_BUFFER_BIT);*/
-         renderer_draw_quad(ctx->renderer, minx, miny, maxx, maxy, 0.0f);
-      }
-
-      cso_restore_blend(ctx->cso_context);
-      cso_restore_fragment_shader(ctx->cso_context);
-
-      dsa->depth.enabled = 1; /* glEnable(GL_DEPTH_TEST); */
-      dsa->depth.writemask = 0;/*glDepthMask(FALSE);*/
-      dsa->depth.func = PIPE_FUNC_GEQUAL;
+   struct pipe_resource templ;
+
+   memset(&templ, 0, sizeof(templ));
+
+   if (format != PIPE_FORMAT_NONE) {
+      templ.format = format;
+   }
+   else {
+      templ.format = PIPE_FORMAT_B8G8R8A8_UNORM;
+   }
+
+   templ.target = PIPE_TEXTURE_2D;
+   templ.width0 = width;
+   templ.height0 = height;
+   templ.depth0 = 1;
+   templ.array_size = 1;
+   templ.last_level = 0;
+
+   if (util_format_get_component_bits(format, UTIL_FORMAT_COLORSPACE_ZS, 1)) {
+      templ.bind = PIPE_BIND_DEPTH_STENCIL;
+   } else {
+      templ.bind = (PIPE_BIND_DISPLAY_TARGET |
+                    PIPE_BIND_RENDER_TARGET |
+                    PIPE_BIND_SAMPLER_VIEW);
    }
+
+   return pipe->screen->resource_create(pipe->screen, &templ);
 }
 
-void vg_validate_state(struct vg_context *ctx)
+static struct pipe_sampler_view *
+create_tex_and_view(struct pipe_context *pipe, enum pipe_format format,
+                    VGint width, VGint height)
 {
-   vg_manager_validate_framebuffer(ctx);
+   struct pipe_resource *texture;
+   struct pipe_sampler_view view_templ;
+   struct pipe_sampler_view *view;
 
-   if ((ctx->state.dirty & BLEND_DIRTY)) {
-      struct pipe_blend_state *blend = &ctx->state.g3d.blend;
-      memset(blend, 0, sizeof(struct pipe_blend_state));
-      blend->rt[0].blend_enable = 1;
-      blend->rt[0].colormask = PIPE_MASK_RGBA;
-
-      switch (ctx->state.vg.blend_mode) {
-      case VG_BLEND_SRC:
-         blend->rt[0].rgb_src_factor   = PIPE_BLENDFACTOR_ONE;
-         blend->rt[0].alpha_src_factor = PIPE_BLENDFACTOR_ONE;
-         blend->rt[0].rgb_dst_factor   = PIPE_BLENDFACTOR_ZERO;
-         blend->rt[0].alpha_dst_factor = PIPE_BLENDFACTOR_ZERO;
-         blend->rt[0].blend_enable = 0;
-         break;
-      case VG_BLEND_SRC_OVER:
-         blend->rt[0].rgb_src_factor   = PIPE_BLENDFACTOR_SRC_ALPHA;
-         blend->rt[0].alpha_src_factor = PIPE_BLENDFACTOR_ONE;
-         blend->rt[0].rgb_dst_factor   = PIPE_BLENDFACTOR_INV_SRC_ALPHA;
-         blend->rt[0].alpha_dst_factor = PIPE_BLENDFACTOR_INV_SRC_ALPHA;
-         break;
-      case VG_BLEND_DST_OVER:
-         blend->rt[0].rgb_src_factor   = PIPE_BLENDFACTOR_INV_DST_ALPHA;
-         blend->rt[0].alpha_src_factor = PIPE_BLENDFACTOR_INV_DST_ALPHA;
-         blend->rt[0].rgb_dst_factor   = PIPE_BLENDFACTOR_DST_ALPHA;
-         blend->rt[0].alpha_dst_factor = PIPE_BLENDFACTOR_DST_ALPHA;
-         break;
-      case VG_BLEND_SRC_IN:
-         blend->rt[0].rgb_src_factor   = PIPE_BLENDFACTOR_DST_ALPHA;
-         blend->rt[0].alpha_src_factor = PIPE_BLENDFACTOR_DST_ALPHA;
-         blend->rt[0].rgb_dst_factor   = PIPE_BLENDFACTOR_ZERO;
-         blend->rt[0].alpha_dst_factor = PIPE_BLENDFACTOR_ZERO;
-         break;
-      case VG_BLEND_DST_IN:
-         blend->rt[0].rgb_src_factor   = PIPE_BLENDFACTOR_ZERO;
-         blend->rt[0].alpha_src_factor = PIPE_BLENDFACTOR_ZERO;
-         blend->rt[0].rgb_dst_factor   = PIPE_BLENDFACTOR_SRC_ALPHA;
-         blend->rt[0].alpha_dst_factor = PIPE_BLENDFACTOR_SRC_ALPHA;
-         break;
-      case VG_BLEND_MULTIPLY:
-      case VG_BLEND_SCREEN:
-      case VG_BLEND_DARKEN:
-      case VG_BLEND_LIGHTEN:
-         blend->rt[0].rgb_src_factor   = PIPE_BLENDFACTOR_ONE;
-         blend->rt[0].alpha_src_factor = PIPE_BLENDFACTOR_ONE;
-         blend->rt[0].rgb_dst_factor   = PIPE_BLENDFACTOR_ZERO;
-         blend->rt[0].alpha_dst_factor = PIPE_BLENDFACTOR_ZERO;
-         blend->rt[0].blend_enable = 0;
-         break;
-      case VG_BLEND_ADDITIVE:
-         blend->rt[0].rgb_src_factor   = PIPE_BLENDFACTOR_ONE;
-         blend->rt[0].alpha_src_factor = PIPE_BLENDFACTOR_ONE;
-         blend->rt[0].rgb_dst_factor   = PIPE_BLENDFACTOR_ONE;
-         blend->rt[0].alpha_dst_factor = PIPE_BLENDFACTOR_ONE;
-         break;
-      default:
-         assert(!"not implemented blend mode");
-      }
-      cso_set_blend(ctx->cso_context, &ctx->state.g3d.blend);
-   }
-   if ((ctx->state.dirty & RASTERIZER_DIRTY)) {
-      struct pipe_rasterizer_state *raster = &ctx->state.g3d.rasterizer;
-      memset(raster, 0, sizeof(struct pipe_rasterizer_state));
-      raster->gl_rasterization_rules = 1;
-      cso_set_rasterizer(ctx->cso_context, &ctx->state.g3d.rasterizer);
-   }
-   if ((ctx->state.dirty & VIEWPORT_DIRTY)) {
-      struct pipe_framebuffer_state *fb = &ctx->state.g3d.fb;
-      const VGint param_bytes = 8 * sizeof(VGfloat);
-      VGfloat vs_consts[8] = {
-         2.f/fb->width, 2.f/fb->height, 1, 1,
-         -1, -1, 0, 0
-      };
-      struct pipe_resource **cbuf = &ctx->vs_const_buffer;
+   texture = create_texture(pipe, format, width, height);
 
-      vg_set_viewport(ctx, VEGA_Y0_BOTTOM);
+   if (!texture)
+      return NULL;
 
-      pipe_resource_reference(cbuf, NULL);
-      *cbuf = pipe_buffer_create(ctx->pipe->screen, 
-				 PIPE_BIND_CONSTANT_BUFFER,
-				 param_bytes);
-
-      if (*cbuf) {
-         st_no_flush_pipe_buffer_write(ctx, *cbuf,
-                                       0, param_bytes, vs_consts);
-      }
-      ctx->pipe->set_constant_buffer(ctx->pipe, PIPE_SHADER_VERTEX, 0, *cbuf);
+   u_sampler_view_default_template(&view_templ, texture, texture->format);
+   view = pipe->create_sampler_view(pipe, texture, &view_templ);
+   /* want the texture to go away if the view is freed */
+   pipe_resource_reference(&texture, NULL);
+
+   return view;
+}
+
+static void
+vg_context_update_surface_mask_view(struct vg_context *ctx,
+                                    uint width, uint height)
+{
+   struct st_framebuffer *stfb = ctx->draw_buffer;
+   struct pipe_sampler_view *old_sampler_view = stfb->surface_mask_view;
+   struct pipe_context *pipe = ctx->pipe;
+
+   if (old_sampler_view &&
+       old_sampler_view->texture->width0 == width &&
+       old_sampler_view->texture->height0 == height)
+      return;
+
+   /*
+     we use PIPE_FORMAT_B8G8R8A8_UNORM because we want to render to
+     this texture and use it as a sampler, so while this wastes some
+     space it makes both of those a lot simpler
+   */
+   stfb->surface_mask_view = create_tex_and_view(pipe,
+         PIPE_FORMAT_B8G8R8A8_UNORM, width, height);
+
+   if (!stfb->surface_mask_view) {
+      if (old_sampler_view)
+         pipe_sampler_view_reference(&old_sampler_view, NULL);
+      return;
    }
-   if ((ctx->state.dirty & VS_DIRTY)) {
-      cso_set_vertex_shader_handle(ctx->cso_context,
-                                   vg_plain_vs(ctx));
+
+   /* XXX could this call be avoided? */
+   vg_validate_state(ctx);
+
+   /* alpha mask starts with 1.f alpha */
+   mask_fill(0, 0, width, height, 1.f);
+
+   /* if we had an old surface copy it over */
+   if (old_sampler_view) {
+      struct pipe_box src_box;
+      u_box_origin_2d(MIN2(old_sampler_view->texture->width0,
+                           stfb->surface_mask_view->texture->width0),
+                      MIN2(old_sampler_view->texture->height0,
+                           stfb->surface_mask_view->texture->height0),
+                      &src_box);
+
+      pipe->resource_copy_region(pipe,
+                                 stfb->surface_mask_view->texture,
+                                 0, 0, 0, 0,
+                                 old_sampler_view->texture,
+                                 0, &src_box);
    }
 
-   /* must be last because it renders to the depth buffer*/
-   if ((ctx->state.dirty & DEPTH_STENCIL_DIRTY)) {
-      update_clip_state(ctx);
-      cso_set_depth_stencil_alpha(ctx->cso_context, &ctx->state.g3d.dsa);
+   /* Free the old texture
+    */
+   if (old_sampler_view)
+      pipe_sampler_view_reference(&old_sampler_view, NULL);
+}
+
+static void
+vg_context_update_blend_texture_view(struct vg_context *ctx,
+                                     uint width, uint height)
+{
+   struct pipe_context *pipe = ctx->pipe;
+   struct st_framebuffer *stfb = ctx->draw_buffer;
+   struct pipe_sampler_view *old = stfb->blend_texture_view;
+
+   if (old &&
+       old->texture->width0 == width &&
+       old->texture->height0 == height)
+      return;
+
+   stfb->blend_texture_view = create_tex_and_view(pipe,
+         PIPE_FORMAT_B8G8R8A8_UNORM, width, height);
+
+   pipe_sampler_view_reference(&old, NULL);
+}
+
+static boolean
+vg_context_update_depth_stencil_rb(struct vg_context * ctx,
+                                   uint width, uint height)
+{
+   struct st_renderbuffer *dsrb = ctx->draw_buffer->dsrb;
+   struct pipe_context *pipe = ctx->pipe;
+   struct pipe_surface surf_tmpl;
+
+   if ((dsrb->width == width && dsrb->height == height) && dsrb->texture)
+      return FALSE;
+
+   /* unreference existing ones */
+   pipe_surface_reference(&dsrb->surface, NULL);
+   pipe_resource_reference(&dsrb->texture, NULL);
+   dsrb->width = dsrb->height = 0;
+
+   dsrb->texture = create_texture(pipe, dsrb->format, width, height);
+   if (!dsrb->texture)
+      return TRUE;
+
+   memset(&surf_tmpl, 0, sizeof(surf_tmpl));
+   u_surface_default_template(&surf_tmpl, dsrb->texture,
+                              PIPE_BIND_DEPTH_STENCIL);
+   dsrb->surface = pipe->create_surface(pipe,
+                                        dsrb->texture,
+                                        &surf_tmpl);
+   if (!dsrb->surface) {
+      pipe_resource_reference(&dsrb->texture, NULL);
+      return TRUE;
    }
 
+   dsrb->width = width;
+   dsrb->height = height;
+
+   assert(dsrb->surface->width == width);
+   assert(dsrb->surface->height == height);
+
+   return TRUE;
+}
+
+void vg_validate_state(struct vg_context *ctx)
+{
+   struct st_framebuffer *stfb = ctx->draw_buffer;
+
+   vg_manager_validate_framebuffer(ctx);
+
+   if (vg_context_update_depth_stencil_rb(ctx, stfb->width, stfb->height))
+      ctx->state.dirty |= DEPTH_STENCIL_DIRTY;
+
+   /* blend state depends on fb format */
+   if (ctx->state.dirty & FRAMEBUFFER_DIRTY)
+      ctx->state.dirty |= BLEND_DIRTY;
+
+   renderer_validate(ctx->renderer, ctx->state.dirty,
+         ctx->draw_buffer, &ctx->state.vg);
+
+   ctx->state.dirty = 0;
+
    shader_set_masking(ctx->shader, ctx->state.vg.masking);
    shader_set_image_mode(ctx->shader, ctx->state.vg.image_mode);
-
-   ctx->state.dirty = NONE_DIRTY;
+   shader_set_color_transform(ctx->shader, ctx->state.vg.color_transform);
 }
 
 VGboolean vg_object_is_valid(void *ptr, enum vg_object_type type)
@@ -468,130 +434,94 @@ void vg_set_error(struct vg_context *ctx,
       ctx->_error = code;
 }
 
-void vg_prepare_blend_surface(struct vg_context *ctx)
+static void vg_prepare_blend_texture(struct vg_context *ctx,
+                                     struct pipe_sampler_view *src)
+{
+   struct st_framebuffer *stfb = ctx->draw_buffer;
+   struct pipe_surface *surf;
+   struct pipe_surface surf_tmpl;
+
+   vg_context_update_blend_texture_view(ctx, stfb->width, stfb->height);
+
+   memset(&surf_tmpl, 0, sizeof(surf_tmpl));
+   u_surface_default_template(&surf_tmpl, stfb->blend_texture_view->texture,
+                              PIPE_BIND_RENDER_TARGET);
+   surf = ctx->pipe->create_surface(ctx->pipe,
+                                    stfb->blend_texture_view->texture,
+                                    &surf_tmpl);
+   if (surf) {
+      util_blit_pixels_tex(ctx->blit,
+                           src, 0, 0, stfb->width, stfb->height,
+                           surf, 0, 0, stfb->width, stfb->height,
+                           0.0, PIPE_TEX_MIPFILTER_NEAREST);
+
+      pipe_surface_reference(&surf, NULL);
+   }
+}
+
+struct pipe_sampler_view *vg_prepare_blend_surface(struct vg_context *ctx)
 {
-   struct pipe_surface *dest_surface = NULL;
    struct pipe_context *pipe = ctx->pipe;
    struct pipe_sampler_view *view;
    struct pipe_sampler_view view_templ;
    struct st_framebuffer *stfb = ctx->draw_buffer;
    struct st_renderbuffer *strb = stfb->strb;
 
-   /* first finish all pending rendering */
-   vgFinish();
+   vg_validate_state(ctx);
 
    u_sampler_view_default_template(&view_templ, strb->texture, strb->texture->format);
    view = pipe->create_sampler_view(pipe, strb->texture, &view_templ);
 
-   dest_surface = pipe->screen->get_tex_surface(pipe->screen,
-                                                stfb->blend_texture_view->texture,
-                                                0, 0, 0,
-                                                PIPE_BIND_RENDER_TARGET);
-   /* flip it, because we want to use it as a sampler */
-   util_blit_pixels_tex(ctx->blit,
-                        view,
-                        0, strb->height,
-                        strb->width, 0,
-                        dest_surface,
-                        0, 0,
-                        strb->width, strb->height,
-                        0.0, PIPE_TEX_MIPFILTER_NEAREST);
-
-   if (dest_surface)
-      pipe_surface_reference(&dest_surface, NULL);
-
-   /* make sure it's complete */
-   vgFinish();
+   vg_prepare_blend_texture(ctx, view);
 
    pipe_sampler_view_reference(&view, NULL);
+
+   return stfb->blend_texture_view;
 }
 
 
-void vg_prepare_blend_surface_from_mask(struct vg_context *ctx)
+struct pipe_sampler_view *vg_prepare_blend_surface_from_mask(struct vg_context *ctx)
 {
-   struct pipe_surface *dest_surface = NULL;
-   struct pipe_context *pipe = ctx->pipe;
    struct st_framebuffer *stfb = ctx->draw_buffer;
-   struct st_renderbuffer *strb = stfb->strb;
 
    vg_validate_state(ctx);
 
-   /* first finish all pending rendering */
-   vgFinish();
-
-   dest_surface = pipe->screen->get_tex_surface(pipe->screen,
-                                                stfb->blend_texture_view->texture,
-                                                0, 0, 0,
-                                                PIPE_BIND_RENDER_TARGET);
-
-   /* flip it, because we want to use it as a sampler */
-   util_blit_pixels_tex(ctx->blit,
-                        stfb->alpha_mask_view,
-                        0, strb->height,
-                        strb->width, 0,
-                        dest_surface,
-                        0, 0,
-                        strb->width, strb->height,
-                        0.0, PIPE_TEX_MIPFILTER_NEAREST);
-
-   /* make sure it's complete */
-   vgFinish();
-
-   if (dest_surface)
-      pipe_surface_reference(&dest_surface, NULL);
+   vg_context_update_surface_mask_view(ctx, stfb->width, stfb->height);
+   vg_prepare_blend_texture(ctx, stfb->surface_mask_view);
+
+   return stfb->blend_texture_view;
 }
 
-void * vg_plain_vs(struct vg_context *ctx)
+struct pipe_sampler_view *vg_get_surface_mask(struct vg_context *ctx)
 {
-   if (!ctx->plain_vs) {
-      ctx->plain_vs = shader_create_from_text(ctx->pipe,
-                                              vs_plain_asm,
-                                              200,
-                                              PIPE_SHADER_VERTEX);
-   }
+   struct st_framebuffer *stfb = ctx->draw_buffer;
 
-   return ctx->plain_vs->driver;
-}
+   vg_context_update_surface_mask_view(ctx, stfb->width, stfb->height);
 
+   return stfb->surface_mask_view;
+}
 
-void * vg_clear_vs(struct vg_context *ctx)
+/**
+ * A transformation from window coordinates to paint coordinates.
+ */
+VGboolean vg_get_paint_matrix(struct vg_context *ctx,
+                              const struct matrix *paint_to_user,
+                              const struct matrix *user_to_surface,
+                              struct matrix *mat)
 {
-   if (!ctx->clear_vs) {
-      ctx->clear_vs = shader_create_from_text(ctx->pipe,
-                                              vs_clear_asm,
-                                              200,
-                                              PIPE_SHADER_VERTEX);
-   }
+   struct matrix tmp;
 
-   return ctx->clear_vs->driver;
-}
+   /* get user-to-paint matrix */
+   memcpy(mat, paint_to_user, sizeof(*paint_to_user));
+   if (!matrix_invert(mat))
+      return VG_FALSE;
 
-void * vg_texture_vs(struct vg_context *ctx)
-{
-   if (!ctx->texture_vs) {
-      ctx->texture_vs = shader_create_from_text(ctx->pipe,
-                                                vs_texture_asm,
-                                                200,
-                                                PIPE_SHADER_VERTEX);
-   }
+   /* get surface-to-user matrix */
+   memcpy(&tmp, user_to_surface, sizeof(*user_to_surface));
+   if (!matrix_invert(&tmp))
+      return VG_FALSE;
 
-   return ctx->texture_vs->driver;
-}
+   matrix_mult(mat, &tmp);
 
-void vg_set_viewport(struct vg_context *ctx, VegaOrientation orientation)
-{
-   struct pipe_viewport_state viewport;
-   struct pipe_framebuffer_state *fb = &ctx->state.g3d.fb;
-   VGfloat y_scale = (orientation == VEGA_Y0_BOTTOM) ? -2.f : 2.f;
-
-   viewport.scale[0] =  fb->width / 2.f;
-   viewport.scale[1] =  fb->height / y_scale;
-   viewport.scale[2] =  1.0;
-   viewport.scale[3] =  1.0;
-   viewport.translate[0] = fb->width / 2.f;
-   viewport.translate[1] = fb->height / 2.f;
-   viewport.translate[2] = 0.0;
-   viewport.translate[3] = 0.0;
-
-   cso_set_viewport(ctx->cso_context, &viewport);
+   return VG_TRUE;
 }
diff --git a/src/gallium/state_trackers/vega/vg_context.h b/src/gallium/state_trackers/vega/vg_context.h
index 80a6c07c69..d616a20a3d 100644
--- a/src/gallium/state_trackers/vega/vg_context.h
+++ b/src/gallium/state_trackers/vega/vg_context.h
@@ -56,7 +56,7 @@ struct st_framebuffer {
    struct st_renderbuffer *strb;
    struct st_renderbuffer *dsrb;
 
-   struct pipe_sampler_view *alpha_mask_view;
+   struct pipe_sampler_view *surface_mask_view;
 
    struct pipe_sampler_view *blend_texture_view;
 
@@ -78,14 +78,13 @@ enum vg_object_type {
    VG_OBJECT_LAST
 };
 enum dirty_state {
-   NONE_DIRTY          = 0<<0,
-   BLEND_DIRTY         = 1<<1,
-   RASTERIZER_DIRTY    = 1<<2,
-   VIEWPORT_DIRTY      = 1<<3,
-   VS_DIRTY            = 1<<4,
-   DEPTH_STENCIL_DIRTY = 1<<5,
-   ALL_DIRTY           = BLEND_DIRTY | RASTERIZER_DIRTY |
-   VIEWPORT_DIRTY | VS_DIRTY | DEPTH_STENCIL_DIRTY
+   BLEND_DIRTY         = 1 << 0,
+   FRAMEBUFFER_DIRTY   = 1 << 1,
+   DEPTH_STENCIL_DIRTY = 1 << 2,
+
+   ALL_DIRTY           = BLEND_DIRTY |
+                         FRAMEBUFFER_DIRTY |
+                         DEPTH_STENCIL_DIRTY
 };
 
 struct vg_context
@@ -98,13 +97,6 @@ struct vg_context
 
    struct {
       struct vg_state vg;
-      struct {
-         struct pipe_blend_state blend;
-         struct pipe_rasterizer_state rasterizer;
-         struct pipe_shader_state vs_state;
-         struct pipe_depth_stencil_alpha_state dsa;
-         struct pipe_framebuffer_state fb;
-      } g3d;
       VGbitfield dirty;
    } state;
 
@@ -116,14 +108,6 @@ struct vg_context
    struct cso_hash *owned_objects[VG_OBJECT_LAST];
 
    struct {
-      struct pipe_shader_state vert_shader;
-      struct pipe_shader_state frag_shader;
-      struct pipe_rasterizer_state raster;
-      void *fs;
-      float vertices[4][2][4];  /**< vertex pos + color */
-   } clear;
-
-   struct {
       struct pipe_resource *cbuf;
       struct pipe_sampler_state sampler;
 
@@ -133,31 +117,16 @@ struct vg_context
       struct vg_shader *set_fs;
    } mask;
 
-   struct vg_shader *pass_through_depth_fs;
-
    struct cso_context *cso_context;
 
-   struct pipe_resource *stencil_quad;
-   VGfloat stencil_vertices[4][2][4];
-
    struct renderer *renderer;
    struct shaders_cache *sc;
    struct shader *shader;
 
    struct pipe_sampler_state blend_sampler;
-   struct {
-      struct pipe_resource *buffer;
-      void *color_matrix_fs;
-   } filter;
    struct vg_paint *default_paint;
 
    struct blit_state *blit;
-
-   struct vg_shader *plain_vs;
-   struct vg_shader *clear_vs;
-   struct vg_shader *texture_vs;
-   struct pipe_resource *vs_const_buffer;
-   struct pipe_vertex_element velems[2];
 };
 
 struct vg_object {
@@ -189,9 +158,15 @@ void vg_validate_state(struct vg_context *ctx);
 void vg_set_error(struct vg_context *ctx,
                   VGErrorCode code);
 
-void vg_prepare_blend_surface(struct vg_context *ctx);
-void vg_prepare_blend_surface_from_mask(struct vg_context *ctx);
+struct pipe_sampler_view *vg_prepare_blend_surface(struct vg_context *ctx);
+struct pipe_sampler_view *vg_prepare_blend_surface_from_mask(struct vg_context *ctx);
 
+struct pipe_sampler_view *vg_get_surface_mask(struct vg_context *ctx);
+
+VGboolean vg_get_paint_matrix(struct vg_context *ctx,
+                              const struct matrix *paint_to_user,
+                              const struct matrix *user_to_surface,
+                              struct matrix *mat);
 
 static INLINE VGboolean is_aligned_to(const void *ptr, VGbyte alignment)
 {
@@ -292,13 +267,4 @@ static INLINE void vg_bound_rect(VGfloat coords[4],
    }
 }
 
-void *vg_plain_vs(struct vg_context *ctx);
-void *vg_clear_vs(struct vg_context *ctx);
-void *vg_texture_vs(struct vg_context *ctx);
-typedef enum {
-   VEGA_Y0_TOP,
-   VEGA_Y0_BOTTOM
-} VegaOrientation;
-void vg_set_viewport(struct vg_context *ctx, VegaOrientation orientation);
-
 #endif
diff --git a/src/gallium/state_trackers/vega/vg_manager.c b/src/gallium/state_trackers/vega/vg_manager.c
index 232deefa16..de935768b2 100644
--- a/src/gallium/state_trackers/vega/vg_manager.c
+++ b/src/gallium/state_trackers/vega/vg_manager.c
@@ -33,168 +33,22 @@
 #include "pipe/p_screen.h"
 #include "util/u_memory.h"
 #include "util/u_inlines.h"
-#include "util/u_format.h"
 #include "util/u_sampler.h"
+#include "util/u_box.h"
+#include "util/u_surface.h"
 
 #include "vg_api.h"
 #include "vg_manager.h"
 #include "vg_context.h"
 #include "image.h"
-#include "mask.h"
 #include "api.h"
 
-static struct pipe_resource *
-create_texture(struct pipe_context *pipe, enum pipe_format format,
-                    VGint width, VGint height)
-{
-   struct pipe_resource templ;
-
-   memset(&templ, 0, sizeof(templ));
-
-   if (format != PIPE_FORMAT_NONE) {
-      templ.format = format;
-   }
-   else {
-      templ.format = PIPE_FORMAT_B8G8R8A8_UNORM;
-   }
-
-   templ.target = PIPE_TEXTURE_2D;
-   templ.width0 = width;
-   templ.height0 = height;
-   templ.depth0 = 1;
-   templ.last_level = 0;
-
-   if (util_format_get_component_bits(format, UTIL_FORMAT_COLORSPACE_ZS, 1)) {
-      templ.bind = PIPE_BIND_DEPTH_STENCIL;
-   } else {
-      templ.bind = (PIPE_BIND_DISPLAY_TARGET |
-                    PIPE_BIND_RENDER_TARGET |
-                    PIPE_BIND_SAMPLER_VIEW);
-   }
-
-   return pipe->screen->resource_create(pipe->screen, &templ);
-}
-
-static struct pipe_sampler_view *
-create_tex_and_view(struct pipe_context *pipe, enum pipe_format format,
-                    VGint width, VGint height)
-{
-   struct pipe_resource *texture;
-   struct pipe_sampler_view view_templ;
-   struct pipe_sampler_view *view;
-
-   texture = create_texture(pipe, format, width, height);
-
-   if (!texture)
-      return NULL;
-
-   u_sampler_view_default_template(&view_templ, texture, texture->format);
-   view = pipe->create_sampler_view(pipe, texture, &view_templ);
-   /* want the texture to go away if the view is freed */
-   pipe_resource_reference(&texture, NULL);
-
-   return view;
-}
-
-static void
-setup_new_alpha_mask(struct vg_context *ctx, struct st_framebuffer *stfb)
-{
-   struct pipe_context *pipe = ctx->pipe;
-   struct pipe_sampler_view *old_sampler_view = stfb->alpha_mask_view;
-
-   /*
-     we use PIPE_FORMAT_B8G8R8A8_UNORM because we want to render to
-     this texture and use it as a sampler, so while this wastes some
-     space it makes both of those a lot simpler
-   */
-   stfb->alpha_mask_view = create_tex_and_view(pipe,
-         PIPE_FORMAT_B8G8R8A8_UNORM, stfb->width, stfb->height);
-
-   if (!stfb->alpha_mask_view) {
-      if (old_sampler_view)
-         pipe_sampler_view_reference(&old_sampler_view, NULL);
-      return;
-   }
-
-   /* XXX could this call be avoided? */
-   vg_validate_state(ctx);
-
-   /* alpha mask starts with 1.f alpha */
-   mask_fill(0, 0, stfb->width, stfb->height, 1.f);
-
-   /* if we had an old surface copy it over */
-   if (old_sampler_view) {
-      struct pipe_subresource subsurf, subold_surf;
-      subsurf.face = 0;
-      subsurf.level = 0;
-      subold_surf.face = 0;
-      subold_surf.level = 0;
-      pipe->resource_copy_region(pipe,
-                                 stfb->alpha_mask_view->texture,
-                                 subsurf,
-                                 0, 0, 0,
-                                 old_sampler_view->texture,
-                                 subold_surf,
-                                 0, 0, 0,
-                                 MIN2(old_sampler_view->texture->width0,
-                                      stfb->alpha_mask_view->texture->width0),
-                                 MIN2(old_sampler_view->texture->height0,
-                                      stfb->alpha_mask_view->texture->height0));
-   }
-
-   /* Free the old texture
-    */
-   if (old_sampler_view)
-      pipe_sampler_view_reference(&old_sampler_view, NULL);
-}
-
-static boolean
-vg_context_update_depth_stencil_rb(struct vg_context * ctx,
-                                   uint width, uint height)
-{
-   struct st_renderbuffer *dsrb = ctx->draw_buffer->dsrb;
-   struct pipe_context *pipe = ctx->pipe;
-   unsigned surface_usage;
-
-   if ((dsrb->width == width && dsrb->height == height) && dsrb->texture)
-      return FALSE;
-
-   /* unreference existing ones */
-   pipe_surface_reference(&dsrb->surface, NULL);
-   pipe_resource_reference(&dsrb->texture, NULL);
-   dsrb->width = dsrb->height = 0;
-
-   /* Probably need dedicated flags for surface usage too:
-    */
-   surface_usage = PIPE_BIND_DEPTH_STENCIL; /* XXX: was: RENDER_TARGET */
-
-   dsrb->texture = create_texture(pipe, dsrb->format, width, height);
-   if (!dsrb->texture)
-      return TRUE;
-
-   dsrb->surface = pipe->screen->get_tex_surface(pipe->screen,
-                                                 dsrb->texture,
-                                                 0, 0, 0,
-                                                 surface_usage);
-   if (!dsrb->surface) {
-      pipe_resource_reference(&dsrb->texture, NULL);
-      return TRUE;
-   }
-
-   dsrb->width = width;
-   dsrb->height = height;
-
-   assert(dsrb->surface->width == width);
-   assert(dsrb->surface->height == height);
-
-   return TRUE;
-}
-
 static boolean
 vg_context_update_color_rb(struct vg_context *ctx, struct pipe_resource *pt)
 {
    struct st_renderbuffer *strb = ctx->draw_buffer->strb;
-   struct pipe_screen *screen = ctx->pipe->screen;
+   struct pipe_context *pipe = ctx->pipe;
+   struct pipe_surface surf_tmpl;
 
    if (strb->texture == pt) {
       pipe_resource_reference(&pt, NULL);
@@ -207,8 +61,12 @@ vg_context_update_color_rb(struct vg_context *ctx, struct pipe_resource *pt)
    strb->width = strb->height = 0;
 
    strb->texture = pt;
-   strb->surface = screen->get_tex_surface(screen, strb->texture, 0, 0, 0,
-                                           PIPE_BIND_RENDER_TARGET);
+
+   memset(&surf_tmpl, 0, sizeof(surf_tmpl));
+   u_surface_default_template(&surf_tmpl, strb->texture,
+                              PIPE_BIND_RENDER_TARGET);
+   strb->surface = pipe->create_surface(pipe, strb->texture, &surf_tmpl);
+
    if (!strb->surface) {
       pipe_resource_reference(&strb->texture, NULL);
       return TRUE;
@@ -220,49 +78,6 @@ vg_context_update_color_rb(struct vg_context *ctx, struct pipe_resource *pt)
    return TRUE;
 }
 
-static void
-vg_context_update_draw_buffer(struct vg_context *ctx, struct pipe_resource *pt)
-{
-   struct st_framebuffer *stfb = ctx->draw_buffer;
-   boolean new_cbuf, new_zsbuf, new_size;
-
-   new_cbuf = vg_context_update_color_rb(ctx, pt);
-   new_zsbuf =
-      vg_context_update_depth_stencil_rb(ctx, pt->width0, pt->height0);
-
-   new_size = (stfb->width != pt->width0 || stfb->height != pt->height0);
-   stfb->width = pt->width0;
-   stfb->height = pt->height0;
-
-   if (new_cbuf || new_zsbuf || new_size) {
-      struct pipe_framebuffer_state *state = &ctx->state.g3d.fb;
-
-      memset(state, 0, sizeof(struct pipe_framebuffer_state));
-      state->width  = stfb->width;
-      state->height = stfb->height;
-      state->nr_cbufs = 1;
-      state->cbufs[0] = stfb->strb->surface;
-      state->zsbuf = stfb->dsrb->surface;
-
-      cso_set_framebuffer(ctx->cso_context, state);
-   }
-
-   if (new_zsbuf || new_size) {
-      ctx->state.dirty |= VIEWPORT_DIRTY;
-      ctx->state.dirty |= DEPTH_STENCIL_DIRTY;/*to reset the scissors*/
-
-      ctx->pipe->clear(ctx->pipe, PIPE_CLEAR_DEPTHSTENCIL, NULL, 0.0, 0);
-
-      /* we need all the other state already set */
-
-      setup_new_alpha_mask(ctx, stfb);
-
-      pipe_sampler_view_reference( &stfb->blend_texture_view, NULL);
-      stfb->blend_texture_view = create_tex_and_view(ctx->pipe,
-            PIPE_FORMAT_B8G8R8A8_UNORM, stfb->width, stfb->height);
-   }
-}
-
 /**
  * Flush the front buffer if the current context renders to the front buffer.
  */
@@ -304,15 +119,16 @@ vg_manager_validate_framebuffer(struct vg_context *ctx)
    if (!stfb->iface->validate(stfb->iface, &stfb->strb_att, 1, &pt) || !pt)
       return;
 
-   /*
-    * unset draw_buffer_invalid first because vg_context_update_draw_buffer
-    * will cause the framebuffer to be validated again because of a call to
-    * vg_validate_state
-    */
    p_atomic_set(&ctx->draw_buffer_invalid, FALSE);
-   vg_context_update_draw_buffer(ctx, pt);
-}
 
+   if (vg_context_update_color_rb(ctx, pt) ||
+       stfb->width != pt->width0 ||
+       stfb->height != pt->height0)
+      ctx->state.dirty |= FRAMEBUFFER_DIRTY;
+
+   stfb->width = pt->width0;
+   stfb->height = pt->height0;
+}
 
 static void
 vg_context_notify_invalid_framebuffer(struct st_context_iface *stctxi,
@@ -336,7 +152,10 @@ static void
 vg_context_destroy(struct st_context_iface *stctxi)
 {
    struct vg_context *ctx = (struct vg_context *) stctxi;
+   struct pipe_context *pipe = ctx->pipe;
+
    vg_destroy_context(ctx);
+   pipe->destroy(pipe);
 }
 
 static struct st_context_iface *
@@ -535,6 +354,7 @@ vg_api_destroy(struct st_api *stapi)
 }
 
 static const struct st_api vg_api = {
+   "Vega " VEGA_VERSION_STRING,
    ST_API_OPENVG,
    ST_PROFILE_DEFAULT_MASK,
    vg_api_destroy,
diff --git a/src/gallium/state_trackers/wgl/SConscript b/src/gallium/state_trackers/wgl/SConscript
index ec55f042f9..1b7597de44 100644
--- a/src/gallium/state_trackers/wgl/SConscript
+++ b/src/gallium/state_trackers/wgl/SConscript
@@ -13,7 +13,6 @@ env.Append(CPPPATH = [
 env.AppendUnique(CPPDEFINES = [
     '_GDI32_', # prevent wgl* being declared __declspec(dllimport)
     'BUILD_GL32', # declare gl* as __declspec(dllexport) in Mesa headers 
-    'WIN32_THREADS', # use Win32 thread API
     'WIN32_LEAN_AND_MEAN', # http://msdn2.microsoft.com/en-us/library/6dwk3a1z.aspx
 ])
 
@@ -22,6 +21,7 @@ sources = [
     'stw_device.c',
     'stw_ext_extensionsstring.c',
     'stw_ext_gallium.c',
+    'stw_ext_pbuffer.c',
     'stw_ext_pixelformat.c',
     'stw_ext_swapinterval.c',
     'stw_framebuffer.c',
diff --git a/src/gallium/state_trackers/wgl/stw_context.c b/src/gallium/state_trackers/wgl/stw_context.c
index 85878b4673..cd4f3c8b3e 100644
--- a/src/gallium/state_trackers/wgl/stw_context.c
+++ b/src/gallium/state_trackers/wgl/stw_context.c
@@ -29,12 +29,10 @@
 
 #include "pipe/p_compiler.h"
 #include "pipe/p_context.h"
+#include "pipe/p_state.h"
+#include "util/u_memory.h"
 #include "state_tracker/st_api.h"
 
-/* for _mesa_share_state */
-#include "state_tracker/st_context.h"
-#include "main/core.h"
-
 #include "stw_icd.h"
 #include "stw_device.h"
 #include "stw_winsys.h"
@@ -102,13 +100,8 @@ DrvShareLists(
    ctx1 = stw_lookup_context_locked( dhglrc1 );
    ctx2 = stw_lookup_context_locked( dhglrc2 );
 
-   if (ctx1 && ctx2) {
-      struct st_context *st1, *st2;
-
-      st1 = (struct st_context *) ctx1->st;
-      st2 = (struct st_context *) ctx2->st;
-      ret = _mesa_share_state(st2->ctx, st1->ctx);
-   }
+   if (ctx1 && ctx2 && ctx2->st->share)
+      ret = ctx2->st->share(ctx2->st, ctx1->st);
 
    pipe_mutex_unlock( stw_dev->ctx_mutex );
    
@@ -271,75 +264,82 @@ stw_make_current(
    struct stw_context *curctx = NULL;
    struct stw_context *ctx = NULL;
    struct stw_framebuffer *fb = NULL;
+   BOOL ret = FALSE;
 
    if (!stw_dev)
-      goto fail;
+      return FALSE;
 
    curctx = stw_current_context();
    if (curctx != NULL) {
-      if (curctx->dhglrc != dhglrc)
+      if (curctx->dhglrc == dhglrc) {
+         if (curctx->hdc == hdc) {
+            /* Return if already current. */
+            return TRUE;
+         }
+      } else {
          curctx->st->flush(curctx->st, PIPE_FLUSH_RENDER_CACHE, NULL);
-      
-      /* Return if already current. */
-      if (curctx->dhglrc == dhglrc && curctx->hdc == hdc) {
-         ctx = curctx;
-         fb = stw_framebuffer_from_hdc( hdc );
-         goto success;
       }
-
-      stw_framebuffer_reference(&curctx->current_framebuffer, NULL);
-   }
-
-   if (hdc == NULL || dhglrc == 0) {
-      return stw_dev->stapi->make_current(stw_dev->stapi, NULL, NULL, NULL);
    }
 
-   pipe_mutex_lock( stw_dev->ctx_mutex ); 
-   ctx = stw_lookup_context_locked( dhglrc );
-   pipe_mutex_unlock( stw_dev->ctx_mutex ); 
-   if(!ctx)
-      goto fail;
-
-   fb = stw_framebuffer_from_hdc( hdc );
-   if (fb) {
-      stw_framebuffer_update(fb);
-   }
-   else {
-      /* Applications should call SetPixelFormat before creating a context,
-       * but not all do, and the opengl32 runtime seems to use a default pixel
-       * format in some cases, so we must create a framebuffer for those here
-       */
-      int iPixelFormat = GetPixelFormat(hdc);
-      if(iPixelFormat)
-         fb = stw_framebuffer_create( hdc, iPixelFormat );
-      if(!fb) 
+   if (dhglrc) {
+      pipe_mutex_lock( stw_dev->ctx_mutex );
+      ctx = stw_lookup_context_locked( dhglrc );
+      pipe_mutex_unlock( stw_dev->ctx_mutex );
+      if (!ctx) {
          goto fail;
-   }
-   
-   if(fb->iPixelFormat != ctx->iPixelFormat)
-      goto fail;
+      }
 
-   /* Bind the new framebuffer */
-   ctx->hdc = hdc;
+      fb = stw_framebuffer_from_hdc( hdc );
+      if (fb) {
+         stw_framebuffer_update(fb);
+      }
+      else {
+         /* Applications should call SetPixelFormat before creating a context,
+          * but not all do, and the opengl32 runtime seems to use a default pixel
+          * format in some cases, so we must create a framebuffer for those here
+          */
+         int iPixelFormat = GetPixelFormat(hdc);
+         if (iPixelFormat)
+            fb = stw_framebuffer_create( hdc, iPixelFormat );
+         if (!fb)
+            goto fail;
+      }
    
-   if (!stw_dev->stapi->make_current(stw_dev->stapi, ctx->st, fb->stfb, fb->stfb))
-      goto fail;
+      if (fb->iPixelFormat != ctx->iPixelFormat) {
+         SetLastError(ERROR_INVALID_PIXEL_FORMAT);
+         goto fail;
+      }
 
-   stw_framebuffer_reference(&ctx->current_framebuffer, fb);
+      /* Bind the new framebuffer */
+      ctx->hdc = hdc;
 
-success:
-   assert(fb);
-   if(fb) {
-      stw_framebuffer_release(fb);
+      ret = stw_dev->stapi->make_current(stw_dev->stapi, ctx->st, fb->stfb, fb->stfb);
+      stw_framebuffer_reference(&ctx->current_framebuffer, fb);
+   } else {
+      ret = stw_dev->stapi->make_current(stw_dev->stapi, NULL, NULL, NULL);
    }
    
-   return TRUE;
-
 fail:
-   if(fb)
+
+   if (fb) {
       stw_framebuffer_release(fb);
-   stw_dev->stapi->make_current(stw_dev->stapi, NULL, NULL, NULL);
-   return FALSE;
+   }
+
+   /* On failure, make the thread's current rendering context not current
+    * before returning */
+   if (!ret) {
+      stw_dev->stapi->make_current(stw_dev->stapi, NULL, NULL, NULL);
+      ctx = NULL;
+   }
+
+   /* Unreference the previous framebuffer if any. It must be done after
+    * make_current, as it can be referenced inside.
+    */
+   if (curctx && curctx != ctx) {
+      stw_framebuffer_reference(&curctx->current_framebuffer, NULL);
+   }
+
+   return ret;
 }
 
 /**
diff --git a/src/gallium/state_trackers/wgl/stw_device.c b/src/gallium/state_trackers/wgl/stw_device.c
index 37809d084c..c4822d4d8a 100644
--- a/src/gallium/state_trackers/wgl/stw_device.c
+++ b/src/gallium/state_trackers/wgl/stw_device.c
@@ -41,9 +41,7 @@
 #include "stw_framebuffer.h"
 #include "stw_st.h"
 
-#ifdef WIN32_THREADS
 extern _glthread_Mutex OneTimeLock;
-#endif
 
 
 struct stw_device *stw_dev = NULL;
@@ -76,9 +74,7 @@ stw_init(const struct stw_winsys *stw_winsys)
    
    stw_dev->stw_winsys = stw_winsys;
 
-#ifdef WIN32_THREADS
    _glthread_INIT_MUTEX(OneTimeLock);
-#endif
 
    stw_dev->stapi = stw_st_create_api();
    stw_dev->smapi = CALLOC_STRUCT(st_manager);
@@ -96,6 +92,10 @@ stw_init(const struct stw_winsys *stw_winsys)
    stw_dev->smapi->get_param = stw_get_param;
    stw_dev->screen = screen;
 
+   stw_dev->max_2d_levels =
+         screen->get_param(screen, PIPE_CAP_MAX_TEXTURE_2D_LEVELS);
+   stw_dev->max_2d_length = 1 << (stw_dev->max_2d_levels - 1);
+
    pipe_mutex_init( stw_dev->ctx_mutex );
    pipe_mutex_init( stw_dev->fb_mutex );
 
@@ -168,11 +168,9 @@ stw_cleanup(void)
 
    stw_dev->screen->destroy(stw_dev->screen);
 
-#ifdef WIN32_THREADS
    _glthread_DESTROY_MUTEX(OneTimeLock);
 
    _glapi_destroy_multithread();
-#endif
 
 #ifdef DEBUG
    debug_memory_end(stw_dev->memdbg_no);
diff --git a/src/gallium/state_trackers/wgl/stw_device.h b/src/gallium/state_trackers/wgl/stw_device.h
index 1b836960d0..3c2b6d9c07 100644
--- a/src/gallium/state_trackers/wgl/stw_device.h
+++ b/src/gallium/state_trackers/wgl/stw_device.h
@@ -50,6 +50,10 @@ struct stw_device
    
    struct pipe_screen *screen;
    
+   /* Cache some PIPE_CAP_* */
+   unsigned max_2d_levels;
+   unsigned max_2d_length;
+
    struct st_api *stapi;
    struct st_manager *smapi;
 
diff --git a/src/gallium/state_trackers/wgl/stw_ext_extensionsstring.c b/src/gallium/state_trackers/wgl/stw_ext_extensionsstring.c
index 62c859e1f9..ecb326f1cf 100644
--- a/src/gallium/state_trackers/wgl/stw_ext_extensionsstring.c
+++ b/src/gallium/state_trackers/wgl/stw_ext_extensionsstring.c
@@ -37,6 +37,7 @@
 static const char *stw_extension_string = 
    "WGL_ARB_extensions_string "
    "WGL_ARB_multisample "
+   "WGL_ARB_pbuffer "
    "WGL_ARB_pixel_format "
 /*   "WGL_EXT_swap_interval " */
    "WGL_EXT_extensions_string";
diff --git a/src/gallium/state_trackers/wgl/stw_ext_pbuffer.c b/src/gallium/state_trackers/wgl/stw_ext_pbuffer.c
new file mode 100644
index 0000000000..32636c603d
--- /dev/null
+++ b/src/gallium/state_trackers/wgl/stw_ext_pbuffer.c
@@ -0,0 +1,212 @@
+/**************************************************************************
+ * 
+ * Copyright 2010 VMware, Inc.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation 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 VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION 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 <windows.h>
+
+#define WGL_WGLEXT_PROTOTYPES
+
+#include <GL/gl.h>
+#include <GL/wglext.h>
+
+#include "pipe/p_defines.h"
+#include "pipe/p_screen.h"
+
+#include "stw_device.h"
+#include "stw_pixelformat.h"
+#include "stw_framebuffer.h"
+
+
+HPBUFFERARB WINAPI
+wglCreatePbufferARB(HDC _hDC,
+                    int iPixelFormat,
+                    int iWidth,
+                    int iHeight,
+                    const int *piAttribList)
+{
+   static boolean first = TRUE;
+   const int *piAttrib;
+   int useLargest = 0;
+   const struct stw_pixelformat_info *info;
+   struct stw_framebuffer *fb;
+   HWND hWnd;
+   HDC hDC;
+
+   info = stw_pixelformat_get_info(iPixelFormat);
+   if (!info) {
+      SetLastError(ERROR_INVALID_PIXEL_FORMAT);
+      return 0;
+   }
+
+   if (iWidth <= 0 || iHeight <= 0) {
+      SetLastError(ERROR_INVALID_DATA);
+      return 0;
+   }
+
+   for (piAttrib = piAttribList; *piAttrib; piAttrib++) {
+      switch (*piAttrib) {
+      case WGL_PBUFFER_LARGEST_ARB:
+         piAttrib++;
+         useLargest = *piAttrib;
+         break;
+      default:
+         SetLastError(ERROR_INVALID_DATA);
+         return 0;
+      }
+   }
+
+   if (iWidth > stw_dev->max_2d_length) {
+      if (useLargest) {
+         iWidth = stw_dev->max_2d_length;
+      } else {
+         SetLastError(ERROR_NO_SYSTEM_RESOURCES);
+         return 0;
+      }
+   }
+
+   if (iHeight > stw_dev->max_2d_length) {
+      if (useLargest) {
+         iHeight = stw_dev->max_2d_length;
+      } else {
+         SetLastError(ERROR_NO_SYSTEM_RESOURCES);
+         return 0;
+      }
+   }
+
+   /*
+    * Implement pbuffers through invisible windows
+    */
+
+   if (first) {
+      WNDCLASS wc;
+      memset(&wc, 0, sizeof wc);
+      wc.hbrBackground = (HBRUSH) (COLOR_BTNFACE + 1);
+      wc.hCursor = LoadCursor(NULL, IDC_ARROW);
+      wc.hIcon = LoadIcon(NULL, IDI_APPLICATION);
+      wc.lpfnWndProc = DefWindowProc;
+      wc.lpszClassName = "wglpbuffer";
+      wc.style = CS_OWNDC | CS_HREDRAW | CS_VREDRAW;
+      RegisterClass(&wc);
+      first = FALSE;
+   }
+
+   hWnd = CreateWindowEx(0,
+                         "wglpbuffer", /* wc.lpszClassName */
+                         "wglpbuffer",
+#if 0 /* Useful for debugging what the application is drawing */
+                         WS_VISIBLE |
+#endif
+                         WS_CLIPSIBLINGS | WS_CLIPCHILDREN,
+                         CW_USEDEFAULT, CW_USEDEFAULT, /* x, y */
+                         iWidth, iHeight,
+                         NULL,
+                         NULL,
+                         NULL,
+                         NULL);
+   if (!hWnd) {
+      return 0;
+   }
+
+   hDC = GetDC(hWnd);
+   if (!hDC) {
+      return 0;
+   }
+
+   SetPixelFormat(hDC, iPixelFormat, &info->pfd);
+
+   fb = stw_framebuffer_create(hDC, iPixelFormat);
+   if (!fb) {
+      SetLastError(ERROR_NO_SYSTEM_RESOURCES);
+   }
+
+   return (HPBUFFERARB)fb;
+}
+
+
+HDC WINAPI
+wglGetPbufferDCARB(HPBUFFERARB hPbuffer)
+{
+   struct stw_framebuffer *fb;
+   HDC hDC;
+
+   fb = (struct stw_framebuffer *)hPbuffer;
+
+   hDC = GetDC(fb->hWnd);
+   SetPixelFormat(hDC, fb->iPixelFormat, &fb->pfi->pfd);
+
+   return hDC;
+}
+
+
+int WINAPI
+wglReleasePbufferDCARB(HPBUFFERARB hPbuffer,
+                       HDC hDC)
+{
+   struct stw_framebuffer *fb;
+
+   fb = (struct stw_framebuffer *)hPbuffer;
+
+   return ReleaseDC(fb->hWnd, hDC);
+}
+
+
+BOOL WINAPI
+wglDestroyPbufferARB(HPBUFFERARB hPbuffer)
+{
+   struct stw_framebuffer *fb;
+
+   fb = (struct stw_framebuffer *)hPbuffer;
+
+   /* This will destroy all our data */
+   return DestroyWindow(fb->hWnd);
+}
+
+
+BOOL WINAPI
+wglQueryPbufferARB(HPBUFFERARB hPbuffer,
+                   int iAttribute,
+                   int *piValue)
+{
+   struct stw_framebuffer *fb;
+
+   fb = (struct stw_framebuffer *)hPbuffer;
+
+   switch (iAttribute) {
+   case WGL_PBUFFER_WIDTH_ARB:
+      *piValue = fb->width;
+      return TRUE;
+   case WGL_PBUFFER_HEIGHT_ARB:
+      *piValue = fb->width;
+      return TRUE;
+   case WGL_PBUFFER_LOST_ARB:
+      /* We assume that no content is ever lost due to display mode change */
+      *piValue = FALSE;
+      return TRUE;
+   default:
+      SetLastError(ERROR_INVALID_DATA);
+      return FALSE;
+   }
+}
diff --git a/src/gallium/state_trackers/wgl/stw_ext_pixelformat.c b/src/gallium/state_trackers/wgl/stw_ext_pixelformat.c
index ab56800e28..d0a95863bb 100644
--- a/src/gallium/state_trackers/wgl/stw_ext_pixelformat.c
+++ b/src/gallium/state_trackers/wgl/stw_ext_pixelformat.c
@@ -43,6 +43,7 @@
 
 #include "pipe/p_compiler.h"
 #include "util/u_memory.h"
+#include "stw_device.h"
 #include "stw_pixelformat.h"
 
 
@@ -234,6 +235,23 @@ stw_query_attrib(
       *pvalue = pfi->stvis.samples;
       break;
 
+
+   /* WGL_ARB_pbuffer */
+
+   case WGL_MAX_PBUFFER_WIDTH_ARB:
+   case WGL_MAX_PBUFFER_HEIGHT_ARB:
+      *pvalue = stw_dev->max_2d_length;
+      break;
+
+   case WGL_MAX_PBUFFER_PIXELS_ARB:
+      *pvalue = stw_dev->max_2d_length * stw_dev->max_2d_length;
+      break;
+
+   case WGL_DRAW_TO_PBUFFER_ARB:
+      *pvalue = 1;
+      break;
+
+
    default:
       return FALSE;
    }
diff --git a/src/gallium/state_trackers/wgl/stw_framebuffer.c b/src/gallium/state_trackers/wgl/stw_framebuffer.c
index 259b22f22c..733222aa21 100644
--- a/src/gallium/state_trackers/wgl/stw_framebuffer.c
+++ b/src/gallium/state_trackers/wgl/stw_framebuffer.c
@@ -469,7 +469,7 @@ DrvPresentBuffers(HDC hdc, PGLPRESENTBUFFERSDATA data)
 {
    struct stw_framebuffer *fb;
    struct pipe_screen *screen;
-   struct pipe_surface *surface;
+   struct pipe_resource *res;
 
    if (!stw_dev)
       return FALSE;
@@ -480,7 +480,7 @@ DrvPresentBuffers(HDC hdc, PGLPRESENTBUFFERSDATA data)
 
    screen = stw_dev->screen;
 
-   surface = (struct pipe_surface *)data->pPrivateData;
+   res = (struct pipe_resource *)data->pPrivateData;
 
    if(data->hSharedSurface != fb->hSharedSurface) {
       if(fb->shared_surface) {
@@ -498,13 +498,13 @@ DrvPresentBuffers(HDC hdc, PGLPRESENTBUFFERSDATA data)
 
    if(fb->shared_surface) {
       stw_dev->stw_winsys->compose(screen,
-                                   surface,
+                                   res,
                                    fb->shared_surface,
                                    &fb->client_rect,
                                    data->PresentHistoryToken);
    }
    else {
-      stw_dev->stw_winsys->present( screen, surface, hdc );
+      stw_dev->stw_winsys->present( screen, res, hdc );
    }
 
    stw_framebuffer_update(fb);
@@ -524,7 +524,7 @@ DrvPresentBuffers(HDC hdc, PGLPRESENTBUFFERSDATA data)
 BOOL
 stw_framebuffer_present_locked(HDC hdc,
                                struct stw_framebuffer *fb,
-                               struct pipe_surface *surface)
+                               struct pipe_resource *res)
 {
    if(stw_dev->callbacks.wglCbPresentBuffers &&
       stw_dev->stw_winsys->compose) {
@@ -535,7 +535,7 @@ stw_framebuffer_present_locked(HDC hdc,
       data.magic2 = 0;
       data.AdapterLuid = stw_dev->AdapterLuid;
       data.rect = fb->client_rect;
-      data.pPrivateData = (void *)surface;
+      data.pPrivateData = (void *)res;
 
       stw_notify_current_locked(fb);
       stw_framebuffer_release(fb);
@@ -545,7 +545,7 @@ stw_framebuffer_present_locked(HDC hdc,
    else {
       struct pipe_screen *screen = stw_dev->screen;
 
-      stw_dev->stw_winsys->present( screen, surface, hdc );
+      stw_dev->stw_winsys->present( screen, res, hdc );
 
       stw_framebuffer_update(fb);
       stw_notify_current_locked(fb);
diff --git a/src/gallium/state_trackers/wgl/stw_framebuffer.h b/src/gallium/state_trackers/wgl/stw_framebuffer.h
index 89d12300e6..6604e545cb 100644
--- a/src/gallium/state_trackers/wgl/stw_framebuffer.h
+++ b/src/gallium/state_trackers/wgl/stw_framebuffer.h
@@ -32,7 +32,7 @@
 
 #include "os/os_thread.h"
 
-struct pipe_surface;
+struct pipe_resource;
 struct st_framebuffer_iface;
 struct stw_pixelformat_info;
 
@@ -143,7 +143,7 @@ stw_framebuffer_from_hdc(
 BOOL
 stw_framebuffer_present_locked(HDC hdc,
                                struct stw_framebuffer *fb,
-                               struct pipe_surface *surface);
+                               struct pipe_resource *res);
 
 void
 stw_framebuffer_update(
diff --git a/src/gallium/state_trackers/wgl/stw_getprocaddress.c b/src/gallium/state_trackers/wgl/stw_getprocaddress.c
index d43a55fa9e..b0aef943d3 100644
--- a/src/gallium/state_trackers/wgl/stw_getprocaddress.c
+++ b/src/gallium/state_trackers/wgl/stw_getprocaddress.c
@@ -50,6 +50,13 @@ static const struct stw_extension_entry stw_extension_entries[] = {
    /* WGL_ARB_extensions_string */
    STW_EXTENSION_ENTRY( wglGetExtensionsStringARB ),
 
+   /* WGL_ARB_pbuffer */
+   STW_EXTENSION_ENTRY( wglCreatePbufferARB ),
+   STW_EXTENSION_ENTRY( wglGetPbufferDCARB ),
+   STW_EXTENSION_ENTRY( wglReleasePbufferDCARB ),
+   STW_EXTENSION_ENTRY( wglDestroyPbufferARB ),
+   STW_EXTENSION_ENTRY( wglQueryPbufferARB ),
+
    /* WGL_ARB_pixel_format */
    STW_EXTENSION_ENTRY( wglChoosePixelFormatARB ),
    STW_EXTENSION_ENTRY( wglGetPixelFormatAttribfvARB ),
diff --git a/src/gallium/state_trackers/wgl/stw_pixelformat.c b/src/gallium/state_trackers/wgl/stw_pixelformat.c
index 18ac482369..5ede1d4322 100644
--- a/src/gallium/state_trackers/wgl/stw_pixelformat.c
+++ b/src/gallium/state_trackers/wgl/stw_pixelformat.c
@@ -185,7 +185,7 @@ stw_pixelformat_add(
     */
    pfi->stvis.buffer_mask = ST_ATTACHMENT_FRONT_LEFT_MASK;
    if (doublebuffer)
-      pfi->stvis.buffer_mask = ST_ATTACHMENT_BACK_LEFT_MASK;
+      pfi->stvis.buffer_mask |= ST_ATTACHMENT_BACK_LEFT_MASK;
 
    pfi->stvis.color_format = color->format;
    pfi->stvis.depth_stencil_format = depth->format;
diff --git a/src/gallium/state_trackers/wgl/stw_st.c b/src/gallium/state_trackers/wgl/stw_st.c
index bcdd82e4f6..b58d91673b 100644
--- a/src/gallium/state_trackers/wgl/stw_st.c
+++ b/src/gallium/state_trackers/wgl/stw_st.c
@@ -43,8 +43,6 @@ struct stw_st_framebuffer {
    struct pipe_resource *textures[ST_ATTACHMENT_COUNT];
    unsigned texture_width, texture_height;
    unsigned texture_mask;
-
-   struct pipe_surface *front_surface, *back_surface;
 };
 
 static INLINE struct stw_st_framebuffer *
@@ -65,10 +63,6 @@ stw_st_framebuffer_validate_locked(struct st_framebuffer_iface *stfb,
    struct pipe_resource templ;
    unsigned i;
 
-   /* remove outdated surface */
-   pipe_surface_reference(&stwfb->front_surface, NULL);
-   pipe_surface_reference(&stwfb->back_surface, NULL);
-
    /* remove outdated textures */
    if (stwfb->texture_width != width || stwfb->texture_height != height) {
       for (i = 0; i < ST_ATTACHMENT_COUNT; i++)
@@ -80,6 +74,7 @@ stw_st_framebuffer_validate_locked(struct st_framebuffer_iface *stfb,
    templ.width0 = width;
    templ.height0 = height;
    templ.depth0 = 1;
+   templ.array_size = 1;
    templ.last_level = 0;
 
    for (i = 0; i < ST_ATTACHMENT_COUNT; i++) {
@@ -155,45 +150,6 @@ stw_st_framebuffer_validate(struct st_framebuffer_iface *stfb,
    return TRUE;
 }
 
-static struct pipe_surface *
-get_present_surface_locked(struct st_framebuffer_iface *stfb,
-                           enum st_attachment_type statt)
-{
-   struct stw_st_framebuffer *stwfb = stw_st_framebuffer(stfb);
-   struct pipe_resource *ptex;
-   struct pipe_surface *psurf, **cache;
-   
-   ptex = stwfb->textures[statt];
-   if (!ptex)
-      return NULL;
-
-   psurf = NULL;
-
-   switch (statt) {
-   case ST_ATTACHMENT_FRONT_LEFT:
-      cache = &stwfb->front_surface;
-      break;
-   case ST_ATTACHMENT_BACK_LEFT:
-      cache = &stwfb->back_surface;
-      break;
-   default:
-      cache = &psurf;
-      break;
-   }
-
-   if (!*cache) {
-      *cache = stw_dev->screen->get_tex_surface(stw_dev->screen,
-            ptex, 0, 0, 0,
-            PIPE_BIND_DISPLAY_TARGET |
-            PIPE_BIND_RENDER_TARGET);
-   }
-
-   if (psurf != *cache)
-      pipe_surface_reference(&psurf, *cache);
-
-   return psurf;
-}
-
 /**
  * Present an attachment of the framebuffer.
  */
@@ -202,12 +158,11 @@ stw_st_framebuffer_present_locked(struct st_framebuffer_iface *stfb,
                                   enum st_attachment_type statt)
 {
    struct stw_st_framebuffer *stwfb = stw_st_framebuffer(stfb);
-   struct pipe_surface *psurf;
-   
-   psurf = get_present_surface_locked(&stwfb->base, statt);
-   if (psurf) {
-      stw_framebuffer_present_locked(stwfb->fb->hDC, stwfb->fb, psurf);
-      pipe_surface_reference(&psurf, NULL);
+   struct pipe_resource *resource;
+
+   resource = stwfb->textures[statt];
+   if (resource) {
+      stw_framebuffer_present_locked(stwfb->fb->hDC, stwfb->fb, resource);
    }
 
    return TRUE;
@@ -255,9 +210,6 @@ stw_st_destroy_framebuffer_locked(struct st_framebuffer_iface *stfb)
    struct stw_st_framebuffer *stwfb = stw_st_framebuffer(stfb);
    int i;
 
-   pipe_surface_reference(&stwfb->front_surface, NULL);
-   pipe_surface_reference(&stwfb->back_surface, NULL);
-
    for (i = 0; i < ST_ATTACHMENT_COUNT; i++)
       pipe_resource_reference(&stwfb->textures[i], NULL);
 
@@ -273,7 +225,6 @@ stw_st_swap_framebuffer_locked(struct st_framebuffer_iface *stfb)
    struct stw_st_framebuffer *stwfb = stw_st_framebuffer(stfb);
    unsigned front = ST_ATTACHMENT_FRONT_LEFT, back = ST_ATTACHMENT_BACK_LEFT;
    struct pipe_resource *ptex;
-   struct pipe_surface *psurf;
    unsigned mask;
 
    /* swap the textures */
@@ -281,11 +232,6 @@ stw_st_swap_framebuffer_locked(struct st_framebuffer_iface *stfb)
    stwfb->textures[front] = stwfb->textures[back];
    stwfb->textures[back] = ptex;
 
-   /* swap the surfaces */
-   psurf = stwfb->front_surface;
-   stwfb->front_surface = stwfb->back_surface;
-   stwfb->back_surface = psurf;
-
    /* convert to mask */
    front = 1 << front;
    back = 1 << back;
diff --git a/src/gallium/state_trackers/wgl/stw_winsys.h b/src/gallium/state_trackers/wgl/stw_winsys.h
index 270fad56a1..397065fc15 100644
--- a/src/gallium/state_trackers/wgl/stw_winsys.h
+++ b/src/gallium/state_trackers/wgl/stw_winsys.h
@@ -34,7 +34,7 @@
 
 struct pipe_screen;
 struct pipe_context;
-struct pipe_surface;
+struct pipe_resource;
 
 struct stw_shared_surface;
 
@@ -43,12 +43,13 @@ struct stw_winsys
    struct pipe_screen *
    (*create_screen)( void );
 
+   /* XXX is it actually possible to have non-zero level/layer ??? */
    /**
     * Present the color buffer to the window associated with the device context.
     */
    void
    (*present)( struct pipe_screen *screen,
-               struct pipe_surface *surf,
+               struct pipe_resource *res,
                HDC hDC );
 
    /**
@@ -85,7 +86,7 @@ struct stw_winsys
     */
    void
    (*compose)( struct pipe_screen *screen,
-               struct pipe_surface *src,
+               struct pipe_resource *res,
                struct stw_shared_surface *dest,
                LPCRECT pRect,
                ULONGLONG PresentHistoryToken );
diff --git a/src/gallium/state_trackers/xorg/xorg_composite.c b/src/gallium/state_trackers/xorg/xorg_composite.c
index 4ff48026e5..d4dc84a122 100644
--- a/src/gallium/state_trackers/xorg/xorg_composite.c
+++ b/src/gallium/state_trackers/xorg/xorg_composite.c
@@ -473,7 +473,7 @@ boolean xorg_composite_bind_state(struct exa_context *exa,
                                   struct exa_pixmap_priv *pMask,
                                   struct exa_pixmap_priv *pDst)
 {
-   struct pipe_surface *dst_surf = xorg_gpu_surface(exa->scrn, pDst);
+   struct pipe_surface *dst_surf = xorg_gpu_surface(exa->pipe, pDst);
 
    renderer_bind_destination(exa->renderer, dst_surf,
                              pDst->width,
@@ -529,7 +529,7 @@ boolean xorg_solid_bind_state(struct exa_context *exa,
                               struct exa_pixmap_priv *pixmap,
                               Pixel fg)
 {
-   struct pipe_surface *dst_surf = xorg_gpu_surface(exa->scrn, pixmap);
+   struct pipe_surface *dst_surf = xorg_gpu_surface(exa->pipe, pixmap);
    unsigned vs_traits, fs_traits;
    struct xorg_shader shader;
 
diff --git a/src/gallium/state_trackers/xorg/xorg_crtc.c b/src/gallium/state_trackers/xorg/xorg_crtc.c
index 80af82d97b..28e30e09ff 100644
--- a/src/gallium/state_trackers/xorg/xorg_crtc.c
+++ b/src/gallium/state_trackers/xorg/xorg_crtc.c
@@ -210,6 +210,7 @@ crtc_load_cursor_argb_ga3d(xf86CrtcPtr crtc, CARD32 * image)
 	templat.target = PIPE_TEXTURE_2D;
 	templat.last_level = 0;
 	templat.depth0 = 1;
+	templat.array_size = 1;
 	templat.format = PIPE_FORMAT_B8G8R8A8_UNORM;
 	templat.width0 = 64;
 	templat.height0 = 64;
@@ -225,9 +226,9 @@ crtc_load_cursor_argb_ga3d(xf86CrtcPtr crtc, CARD32 * image)
     }
 
     transfer = pipe_get_transfer(ms->ctx, crtcp->cursor_tex,
-                                         0, 0, 0,
-                                         PIPE_TRANSFER_WRITE,
-                                         0, 0, 64, 64);
+                                 0, 0,
+                                 PIPE_TRANSFER_WRITE,
+                                 0, 0, 64, 64);
     ptr = ms->ctx->transfer_map(ms->ctx, transfer);
     util_copy_rect(ptr, crtcp->cursor_tex->format,
 		   transfer->stride, 0, 0,
diff --git a/src/gallium/state_trackers/xorg/xorg_dri2.c b/src/gallium/state_trackers/xorg/xorg_dri2.c
index b723a8e9cb..17c34b7eac 100644
--- a/src/gallium/state_trackers/xorg/xorg_dri2.c
+++ b/src/gallium/state_trackers/xorg/xorg_dri2.c
@@ -129,6 +129,7 @@ dri2_do_create_buffer(DrawablePtr pDraw, DRI2BufferPtr buffer, unsigned int form
 	    template.width0 = pDraw->width;
 	    template.height0 = pDraw->height;
 	    template.depth0 = 1;
+	    template.array_size = 1;
 	    template.last_level = 0;
 	    template.bind = PIPE_BIND_DEPTH_STENCIL |
 		PIPE_BIND_SHARED;
diff --git a/src/gallium/state_trackers/xorg/xorg_driver.c b/src/gallium/state_trackers/xorg/xorg_driver.c
index 1ee79ae177..66685ecec6 100644
--- a/src/gallium/state_trackers/xorg/xorg_driver.c
+++ b/src/gallium/state_trackers/xorg/xorg_driver.c
@@ -538,44 +538,37 @@ drv_pre_init(ScrnInfoPtr pScrn, int flags)
     return TRUE;
 }
 
-static void drv_block_handler(int i, pointer blockData, pointer pTimeout,
-                              pointer pReadmask)
+void xorg_flush(ScreenPtr pScreen)
 {
-    ScreenPtr pScreen = screenInfo.screens[i];
     modesettingPtr ms = modesettingPTR(xf86Screens[pScreen->myNum]);
 
-    pScreen->BlockHandler = ms->blockHandler;
-    pScreen->BlockHandler(i, blockData, pTimeout, pReadmask);
-    pScreen->BlockHandler = drv_block_handler;
-
     if (ms->ctx) {
-       int j;
+	int j;
 
-       ms->ctx->flush(ms->ctx, PIPE_FLUSH_RENDER_CACHE,
-		      ms->dirtyThrottling ?
-		      &ms->fence[XORG_NR_FENCES-1] :
-		      NULL);
+	ms->ctx->flush(ms->ctx, PIPE_FLUSH_RENDER_CACHE,
+		       ms->dirtyThrottling ?
+		       &ms->fence[XORG_NR_FENCES-1] :
+		       NULL);
        
-       if (ms->dirtyThrottling) {
-	   if (ms->fence[0])
-	       ms->ctx->screen->fence_finish(ms->ctx->screen,
-					     ms->fence[0], 0);
+	if (ms->dirtyThrottling) {
+	    if (ms->fence[0])
+		ms->ctx->screen->fence_finish(ms->ctx->screen,
+					      ms->fence[0], 0);
   
-	   /* The amount of rendering generated by a block handler can be
-	    * quite small.  Let us get a fair way ahead of hardware before
-	    * throttling.
-	    */
-	   for (j = 0; j < XORG_NR_FENCES - 1; j++)
-	       ms->screen->fence_reference(ms->screen,
-					   &ms->fence[j],
-					   ms->fence[j+1]);
-
-	   ms->screen->fence_reference(ms->screen,
-				       &ms->fence[XORG_NR_FENCES-1],
-				       NULL);
-       }
+	    /* The amount of rendering generated by a block handler can be
+	     * quite small.  Let us get a fair way ahead of hardware before
+	     * throttling.
+	     */
+	    for (j = 0; j < XORG_NR_FENCES - 1; j++)
+		ms->screen->fence_reference(ms->screen,
+					    &ms->fence[j],
+					    ms->fence[j+1]);
+
+	    ms->screen->fence_reference(ms->screen,
+					&ms->fence[XORG_NR_FENCES-1],
+					NULL);
+	}
     }
-        
 
 #ifdef DRM_MODE_FEATURE_DIRTYFB
     {
@@ -608,6 +601,19 @@ static void drv_block_handler(int i, pointer blockData, pointer pTimeout,
 #endif
 }
 
+static void drv_block_handler(int i, pointer blockData, pointer pTimeout,
+                              pointer pReadmask)
+{
+    ScreenPtr pScreen = screenInfo.screens[i];
+    modesettingPtr ms = modesettingPTR(xf86Screens[pScreen->myNum]);
+
+    pScreen->BlockHandler = ms->blockHandler;
+    pScreen->BlockHandler(i, blockData, pTimeout, pReadmask);
+    pScreen->BlockHandler = drv_block_handler;
+
+    xorg_flush(pScreen);
+}
+
 static Bool
 drv_create_screen_resources(ScreenPtr pScreen)
 {
diff --git a/src/gallium/state_trackers/xorg/xorg_exa.c b/src/gallium/state_trackers/xorg/xorg_exa.c
index 4b1c02bad4..718a345393 100644
--- a/src/gallium/state_trackers/xorg/xorg_exa.c
+++ b/src/gallium/state_trackers/xorg/xorg_exa.c
@@ -46,6 +46,8 @@
 #include "util/u_math.h"
 #include "util/u_debug.h"
 #include "util/u_format.h"
+#include "util/u_box.h"
+#include "util/u_surface.h"
 
 #define DEBUG_PRINT 0
 #define ROUND_UP_TEXTURES 1
@@ -188,8 +190,8 @@ ExaDownloadFromScreen(PixmapPtr pPix, int x,  int y, int w,  int h, char *dst,
     if (!priv || !priv->tex)
 	return FALSE;
 
-    transfer = pipe_get_transfer(exa->pipe, priv->tex, 0, 0, 0,
-					   PIPE_TRANSFER_READ, x, y, w, h);
+    transfer = pipe_get_transfer(exa->pipe, priv->tex, 0, 0,
+                                 PIPE_TRANSFER_READ, x, y, w, h);
     if (!transfer)
 	return FALSE;
 
@@ -222,8 +224,8 @@ ExaUploadToScreen(PixmapPtr pPix, int x, int y, int w, int h, char *src,
     if (!priv || !priv->tex)
 	return FALSE;
 
-    transfer = pipe_get_transfer(exa->pipe, priv->tex, 0, 0, 0,
-					   PIPE_TRANSFER_WRITE, x, y, w, h);
+    transfer = pipe_get_transfer(exa->pipe, priv->tex, 0, 0,
+                                 PIPE_TRANSFER_WRITE, x, y, w, h);
     if (!transfer)
 	return FALSE;
 
@@ -265,7 +267,7 @@ ExaPrepareAccess(PixmapPtr pPix, int index)
         assert(pPix->drawable.height <= priv->tex->height0);
 
 	priv->map_transfer =
-	   pipe_get_transfer(exa->pipe, priv->tex, 0, 0, 0,
+	   pipe_get_transfer(exa->pipe, priv->tex, 0, 0,
 #ifdef EXA_MIXED_PIXMAPS
 					PIPE_TRANSFER_MAP_DIRECTLY |
 #endif
@@ -449,6 +451,7 @@ ExaPrepareCopy(PixmapPtr pSrcPixmap, PixmapPtr pDstPixmap, int xdir,
        exa->copy.use_surface_copy = TRUE;
     }
     else {
+       struct pipe_surface surf_tmpl;
        exa->copy.use_surface_copy = FALSE;
 
        if (exa->copy.dst == exa->copy.src)
@@ -458,11 +461,13 @@ ExaPrepareCopy(PixmapPtr pSrcPixmap, PixmapPtr pDstPixmap, int xdir,
           pipe_resource_reference(&exa->copy.src_texture,
                                  exa->copy.src->tex);
 
+       memset(&surf_tmpl, 0, sizeof(surf_tmpl));
+       u_surface_default_template(&surf_tmpl, exa->copy.dst->tex,
+                                  PIPE_BIND_RENDER_TARGET);
        exa->copy.dst_surface =
-          exa->scrn->get_tex_surface(exa->scrn,
-                                     exa->copy.dst->tex,
-                                     0, 0, 0,
-                                     PIPE_BIND_RENDER_TARGET);
+          exa->pipe->create_surface(exa->pipe,
+                                    exa->copy.dst->tex,
+                                    &surf_tmpl);
 
 
        renderer_copy_prepare(exa->renderer, 
@@ -492,19 +497,14 @@ ExaCopy(PixmapPtr pDstPixmap, int srcX, int srcY, int dstX, int dstY,
    (void) priv;
 
    if (exa->copy.use_surface_copy) {
-      struct pipe_subresource subdst, subsrc;
-      subdst.face = 0;
-      subdst.level = 0;
-      subsrc.face = 0;
-      subsrc.level = 0;
+      struct pipe_box src_box;
+      u_box_2d(srcX, srcY, width, height, &src_box);
       exa->pipe->resource_copy_region( exa->pipe,
                                        exa->copy.dst->tex,
-                                       subdst,
+                                       0,
                                        dstX, dstY, 0,
                                        exa->copy.src->tex,
-                                       subsrc,
-                                       srcX, srcY, 0,
-                                       width, height );
+                                       0, &src_box);
    }
    else {
       renderer_copy_pixmap(exa->renderer, 
@@ -874,24 +874,21 @@ ExaModifyPixmapHeader(PixmapPtr pPixmap, int width, int height,
         }
 
 	template.depth0 = 1;
+	template.array_size = 1;
 	template.last_level = 0;
 	template.bind = PIPE_BIND_RENDER_TARGET | priv->flags;
 	priv->tex_flags = priv->flags;
 	texture = exa->scrn->resource_create(exa->scrn, &template);
 
 	if (priv->tex) {
-	    struct pipe_subresource subdst, subsrc;
-
-	    subdst.face = 0;
-	    subdst.level = 0;
-	    subsrc.face = 0;
-	    subsrc.level = 0;
+            struct pipe_box src_box;
+            u_box_origin_2d(min(width, texture->width0),
+                            min(height, texture->height0),
+                            &src_box);
             exa->pipe->resource_copy_region(exa->pipe, texture,
-                                            subdst, 0, 0, 0,
+                                            0, 0, 0, 0,
                                             priv->tex,
-                                            subsrc, 0, 0, 0,
-                                            min(width, texture->width0),
-                                            min(height, texture->height0));
+                                            0, &src_box);
 	}
 
 	pipe_resource_reference(&priv->tex, texture);
@@ -947,6 +944,7 @@ xorg_exa_create_root_texture(ScrnInfoPtr pScrn,
     template.width0 = width;
     template.height0 = height;
     template.depth0 = 1;
+    template.array_size = 1;
     template.last_level = 0;
     template.bind |= PIPE_BIND_RENDER_TARGET;
     template.bind |= PIPE_BIND_SCANOUT;
@@ -1063,10 +1061,14 @@ out_err:
 }
 
 struct pipe_surface *
-xorg_gpu_surface(struct pipe_screen *scrn, struct exa_pixmap_priv *priv)
+xorg_gpu_surface(struct pipe_context *pipe, struct exa_pixmap_priv *priv)
 {
-   return scrn->get_tex_surface(scrn, priv->tex, 0, 0, 0,
-                                PIPE_BIND_RENDER_TARGET);
+   struct pipe_surface surf_tmpl;
+   memset(&surf_tmpl, 0, sizeof(surf_tmpl));
+   u_surface_default_template(&surf_tmpl, priv->tex,
+                              PIPE_BIND_RENDER_TARGET);
+
+   return pipe->create_surface(pipe, priv->tex, &surf_tmpl);
 
 }
 
diff --git a/src/gallium/state_trackers/xorg/xorg_exa.h b/src/gallium/state_trackers/xorg/xorg_exa.h
index 86a1afc06e..1f78f60be7 100644
--- a/src/gallium/state_trackers/xorg/xorg_exa.h
+++ b/src/gallium/state_trackers/xorg/xorg_exa.h
@@ -72,7 +72,7 @@ do {                                                          \
 } while(0)
 
 struct pipe_surface *
-xorg_gpu_surface(struct pipe_screen *scrn, struct exa_pixmap_priv *priv);
+xorg_gpu_surface(struct pipe_context *pipe, struct exa_pixmap_priv *priv);
 
 void xorg_exa_flush(struct exa_context *exa, uint pipeFlushFlags,
                     struct pipe_fence_handle **fence);
diff --git a/src/gallium/state_trackers/xorg/xorg_renderer.c b/src/gallium/state_trackers/xorg/xorg_renderer.c
index 92f1cc5065..a3d7c5a70e 100644
--- a/src/gallium/state_trackers/xorg/xorg_renderer.c
+++ b/src/gallium/state_trackers/xorg/xorg_renderer.c
@@ -10,6 +10,7 @@
 #include "util/u_sampler.h"
 
 #include "util/u_inlines.h"
+#include "util/u_box.h"
 
 #include <math.h>
 
@@ -535,6 +536,7 @@ renderer_clone_texture(struct xorg_renderer *r,
    templ.width0 = src->width0;
    templ.height0 = src->height0;
    templ.depth0 = 1;
+   templ.array_size = 1;
    templ.bind = PIPE_BIND_SAMPLER_VIEW;
 
    pt = screen->resource_create(screen, &templ);
@@ -546,19 +548,15 @@ renderer_clone_texture(struct xorg_renderer *r,
 
    {
       /* copy source framebuffer surface into texture */
-      struct pipe_subresource subsrc, subdst;
-      subsrc.face = 0;
-      subsrc.level = 0;
-      subdst.face = 0;
-      subdst.level = 0;
+      struct pipe_box src_box;
+      u_box_origin_2d(src->width0, src->height0, &src_box);
+
       pipe->resource_copy_region(pipe,
                                  pt, /* dest */
-                                 subdst,
+                                 0, /* dest_level */
                                  0, 0, 0, /* destx/y/z */
                                  src,
-                                 subsrc,
-                                 0, 0, 0,
-                                 src->width0, src->height0);
+                                 0, &src_box);
    }
 
    return pt;
diff --git a/src/gallium/state_trackers/xorg/xorg_tracker.h b/src/gallium/state_trackers/xorg/xorg_tracker.h
index a3fb5e5dad..56397b8fea 100644
--- a/src/gallium/state_trackers/xorg/xorg_tracker.h
+++ b/src/gallium/state_trackers/xorg/xorg_tracker.h
@@ -156,6 +156,7 @@ CustomizerPtr xorg_customizer(ScrnInfoPtr pScrn);
 
 Bool xorg_has_gallium(ScrnInfoPtr pScrn);
 
+void xorg_flush(ScreenPtr pScreen);
 /***********************************************************************
  * xorg_exa.c
  */
diff --git a/src/gallium/state_trackers/xorg/xorg_xv.c b/src/gallium/state_trackers/xorg/xorg_xv.c
index f64959f00e..c72ba9ef8d 100644
--- a/src/gallium/state_trackers/xorg/xorg_xv.c
+++ b/src/gallium/state_trackers/xorg/xorg_xv.c
@@ -171,6 +171,7 @@ create_component_texture(struct pipe_context *pipe,
    templ.width0 = width;
    templ.height0 = height;
    templ.depth0 = 1;
+   templ.array_size = 1;
    templ.bind = PIPE_BIND_SAMPLER_VIEW;
 
    tex = screen->resource_create(screen, &templ);
@@ -312,17 +313,17 @@ copy_packed_data(ScrnInfoPtr pScrn,
    int y_array_size = w * h;
 
    ytrans = pipe_get_transfer(pipe, dst[0],
-                                   0, 0, 0,
-                                   PIPE_TRANSFER_WRITE,
-                                   left, top, w, h);
+                              0, 0,
+                              PIPE_TRANSFER_WRITE,
+                              left, top, w, h);
    utrans = pipe_get_transfer(pipe, dst[1],
-                                   0, 0, 0,
-                                   PIPE_TRANSFER_WRITE,
-                                   left, top, w, h);
+                              0, 0,
+                              PIPE_TRANSFER_WRITE,
+                              left, top, w, h);
    vtrans = pipe_get_transfer(pipe, dst[2],
-                                   0, 0, 0,
-                                   PIPE_TRANSFER_WRITE,
-                                   left, top, w, h);
+                              0, 0,
+                              PIPE_TRANSFER_WRITE,
+                              left, top, w, h);
 
    ymap = (char*)pipe->transfer_map(pipe, ytrans);
    umap = (char*)pipe->transfer_map(pipe, utrans);
@@ -533,7 +534,7 @@ display_video(ScrnInfoPtr pScrn, struct xorg_xv_port_priv *pPriv, int id,
    if (!dst || !dst->tex)
       XORG_FALLBACK("Xv destination %s", !dst ? "!dst" : "!dst->tex");
 
-   dst_surf = xorg_gpu_surface(pPriv->r->pipe->screen, dst);
+   dst_surf = xorg_gpu_surface(pPriv->r->pipe, dst);
    hdtv = ((src_w >= RES_720P_X) && (src_h >= RES_720P_Y));
 
 #ifdef COMPOSITE
diff --git a/src/gallium/targets/dri-noop/Makefile b/src/gallium/targets/dri-noop/Makefile
new file mode 100644
index 0000000000..21c5f4f9f2
--- /dev/null
+++ b/src/gallium/targets/dri-noop/Makefile
@@ -0,0 +1,34 @@
+TOP = ../../../..
+include $(TOP)/configs/current
+
+LIBNAME = noop_dri.so
+
+DRIVER_DEFINES = \
+	-D__NOT_HAVE_DRM_H
+
+PIPE_DRIVERS = \
+	$(TOP)/src/gallium/state_trackers/dri/sw/libdrisw.a \
+	$(TOP)/src/gallium/winsys/sw/dri/libswdri.a \
+	$(TOP)/src/gallium/drivers/softpipe/libsoftpipe.a \
+	$(TOP)/src/gallium/drivers/trace/libtrace.a \
+	$(TOP)/src/gallium/drivers/rbug/librbug.a \
+	$(TOP)/src/gallium/drivers/noop/libnoop.a
+
+SWRAST_COMMON_GALLIUM_SOURCES = \
+	$(TOP)/src/mesa/drivers/dri/common/utils.c \
+	$(TOP)/src/mesa/drivers/dri/common/drisw_util.c \
+	$(TOP)/src/mesa/drivers/dri/common/xmlconfig.c
+
+C_SOURCES = \
+	swrast_drm_api.c \
+	$(SWRAST_COMMON_GALLIUM_SOURCES) \
+	$(DRIVER_SOURCES)
+
+ASM_SOURCES =
+
+include ../Makefile.dri
+
+INCLUDES += \
+	-I$(TOP)/src/gallium/winsys/sw/dri
+
+symlinks:
diff --git a/src/gallium/targets/dri-noop/SConscript b/src/gallium/targets/dri-noop/SConscript
new file mode 100644
index 0000000000..9c04ee6631
--- /dev/null
+++ b/src/gallium/targets/dri-noop/SConscript
@@ -0,0 +1,31 @@
+Import('*')
+
+env = drienv.Clone()
+
+env.Append(CPPPATH = [
+    '#/src/gallium/winsys/sw/dri',
+])
+
+env.Prepend(LIBS = [
+    st_drisw,
+    ws_dri,
+    noop,
+    mesa,
+    glsl,
+    gallium,
+    COMMON_DRI_SW_OBJECTS
+])
+
+env.Prepend(LIBS = [noop])
+
+swrastg_sources = [
+    'swrast_drm_api.c'
+]
+
+module = env.LoadableModule(
+    target ='noop_dri.so',
+    source = swrastg_sources,
+    SHLIBPREFIX = '',
+)
+
+env.Alias('dri-noop', module)
diff --git a/src/gallium/targets/dri-noop/swrast_drm_api.c b/src/gallium/targets/dri-noop/swrast_drm_api.c
new file mode 100644
index 0000000000..a99779f183
--- /dev/null
+++ b/src/gallium/targets/dri-noop/swrast_drm_api.c
@@ -0,0 +1,63 @@
+/**************************************************************************
+ *
+ * Copyright 2009, VMware, Inc.
+ * All Rights Reserved.
+ * Copyright 2010 George Sapountzis <gsapountzis@gmail.com>
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and 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 VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ **************************************************************************/
+
+#include "pipe/p_compiler.h"
+#include "util/u_memory.h"
+#include "dri_sw_winsys.h"
+#include "noop/noop_public.h"
+
+#include "target-helpers/inline_debug_helper.h"
+#include "target-helpers/inline_sw_helper.h"
+
+
+struct pipe_screen *
+drisw_create_screen(struct drisw_loader_funcs *lf)
+{
+   struct sw_winsys *winsys = NULL;
+   struct pipe_screen *screen = NULL;
+
+   winsys = dri_create_sw_winsys(lf);
+   if (winsys == NULL)
+      return NULL;
+
+   screen = noop_screen_create(winsys);
+   if (!screen)
+      goto fail;
+
+   screen = debug_screen_wrap(screen);
+
+   return screen;
+
+fail:
+   if (winsys)
+      winsys->destroy(winsys);
+
+   return NULL;
+}
+
+/* vim: set sw=3 ts=8 sts=3 expandtab: */
diff --git a/src/gallium/targets/egl/egl.c b/src/gallium/targets/egl/egl.c
index 786d5d1105..bb182ba984 100644
--- a/src/gallium/targets/egl/egl.c
+++ b/src/gallium/targets/egl/egl.c
@@ -100,9 +100,14 @@ load_st_module(struct st_module *stmod,
 {
    struct st_api *(*create_api)(void);
 
-   _eglLog(_EGL_DEBUG, "searching for st module %s", name);
+   if (name) {
+      _eglLog(_EGL_DEBUG, "searching for st module %s", name);
+      stmod->name = loader_strdup(name);
+   }
+   else {
+      stmod->name = NULL;
+   }
 
-   stmod->name = loader_strdup(name);
    if (stmod->name)
       _eglSearchPathForEach(dlopen_st_module_cb, (void *) stmod);
    else
@@ -232,6 +237,21 @@ get_st_api_full(enum st_api_type api, enum st_profile_type profile)
          break;
    }
 
+   /* try again with libGL.so loaded */
+   if (!stmod->stapi && api == ST_API_OPENGL) {
+      struct util_dl_library *glapi = util_dl_open("libGL" UTIL_DL_EXT);
+
+      if (glapi) {
+         _eglLog(_EGL_DEBUG, "retry with libGL" UTIL_DL_EXT " loaded");
+         /* skip the last name (which is NULL) */
+         for (i = 0; i < count - 1; i++) {
+            if (load_st_module(stmod, names[i], symbol))
+               break;
+         }
+         util_dl_close(glapi);
+      }
+   }
+
    if (!stmod->stapi) {
       EGLint level = (egl_g3d_loader.profile_masks[api]) ?
          _EGL_WARNING : _EGL_DEBUG;
diff --git a/src/gallium/targets/graw-gdi/SConscript b/src/gallium/targets/graw-gdi/SConscript
index 8ee8915ace..352efe95d0 100644
--- a/src/gallium/targets/graw-gdi/SConscript
+++ b/src/gallium/targets/graw-gdi/SConscript
@@ -12,9 +12,6 @@ env.Append(CPPPATH = [
 env.Prepend(LIBS = [
     gallium,
     'gdi32',
-    identity,
-    rbug,
-    trace,
     'user32',
     'ws2_32',
 ])
@@ -24,8 +21,13 @@ sources = [
     graw_util,
 ]
 
-env.Append(CPPDEFINES = 'GALLIUM_SOFTPIPE')
-env.Prepend(LIBS = [softpipe])
+if True:
+    env.Append(CPPDEFINES = ['GALLIUM_TRACE', 'GALLIUM_RBUG', 'GALLIUM_GALAHAD', 'GALLIUM_SOFTPIPE'])
+    env.Prepend(LIBS = [trace, rbug, galahad, softpipe])
+
+if env['llvm']:
+    env.Append(CPPDEFINES = 'GALLIUM_LLVMPIPE')
+    env.Prepend(LIBS = [llvmpipe])
 
 graw = env.SharedLibrary(
     target = 'graw',
diff --git a/src/gallium/targets/graw-gdi/graw_gdi.c b/src/gallium/targets/graw-gdi/graw_gdi.c
index bd6242b077..17ca2a761c 100644
--- a/src/gallium/targets/graw-gdi/graw_gdi.c
+++ b/src/gallium/targets/graw-gdi/graw_gdi.c
@@ -29,7 +29,7 @@
 #include "gdi/gdi_sw_winsys.h"
 #include "pipe/p_screen.h"
 #include "state_tracker/graw.h"
-#include "target-helpers/wrap_screen.h"
+#include "target-helpers/inline_debug_helper.h"
 #include "target-helpers/inline_sw_helper.h"
 #include <windows.h>
 
@@ -116,7 +116,7 @@ graw_create_window_and_screen(int x,
 
    *handle = (void *)hDC;
 
-   return gallium_wrap_screen(screen);
+   return debug_screen_wrap(screen);
 
 fail:
    if (hWnd)
diff --git a/src/gallium/targets/graw-xlib/SConscript b/src/gallium/targets/graw-xlib/SConscript
index e50eb8a03d..42cb349cc9 100644
--- a/src/gallium/targets/graw-xlib/SConscript
+++ b/src/gallium/targets/graw-xlib/SConscript
@@ -7,9 +7,6 @@ env = env.Clone()
 
 env.Prepend(LIBS = [
     ws_xlib,
-    trace,
-    rbug,
-    identity,
     gallium,
 ])
 
@@ -20,14 +17,16 @@ env.Append(CPPPATH = [
     '#src/gallium/include/state_tracker',
 ])
 
+env.Append(CPPDEFINES = ['GALLIUM_RBUG', 'GALLIUM_TRACE', 'GALLIUM_GALAHAD'])
+
 sources = [
     'graw_xlib.c',
     graw_util
 ]
 
 if True:
-    env.Append(CPPDEFINES = 'GALLIUM_SOFTPIPE')
-    env.Prepend(LIBS = [softpipe])
+    env.Append(CPPDEFINES = ['GALLIUM_TRACE', 'GALLIUM_RBUG', 'GALLIUM_GALAHAD', 'GALLIUM_SOFTPIPE'])
+    env.Prepend(LIBS = [trace, rbug, galahad, softpipe])
 
 if env['llvm']:
     env.Append(CPPDEFINES = 'GALLIUM_LLVMPIPE')
diff --git a/src/gallium/targets/graw-xlib/graw_xlib.c b/src/gallium/targets/graw-xlib/graw_xlib.c
index 8658e19e3a..578086f8f9 100644
--- a/src/gallium/targets/graw-xlib/graw_xlib.c
+++ b/src/gallium/targets/graw-xlib/graw_xlib.c
@@ -3,8 +3,8 @@
 #include "pipe/p_screen.h"
 #include "util/u_debug.h"
 #include "util/u_memory.h"
-#include "target-helpers/wrap_screen.h"
 #include "target-helpers/inline_sw_helper.h"
+#include "target-helpers/inline_debug_helper.h"
 #include "state_tracker/xlib_sw_winsys.h"
 #include "state_tracker/graw.h"
 
@@ -36,7 +36,7 @@ graw_create_screen( void )
 
    /* Inject any wrapping layers we want to here:
     */
-   return gallium_wrap_screen( screen );
+   return debug_screen_wrap( screen );
 }
 
 
diff --git a/src/gallium/targets/libgl-gdi/SConscript b/src/gallium/targets/libgl-gdi/SConscript
index 339238756a..6fa0851a1a 100644
--- a/src/gallium/targets/libgl-gdi/SConscript
+++ b/src/gallium/targets/libgl-gdi/SConscript
@@ -18,16 +18,15 @@ env.Append(LIBS = [
     talloc,
 ])
 
-sources = []
+sources = ['libgl_gdi.c']
 drivers = []
 
 if True:
-    sources = ['gdi_softpipe_winsys.c']
-    drivers = [softpipe]
+    drivers += [softpipe]
 
 if env['llvm']:
-    sources = ['gdi_llvmpipe_winsys.c']
-    drivers = [llvmpipe]
+    env.Append(CPPDEFINES = 'HAVE_LLVMPIPE')
+    drivers += [llvmpipe]
 
 if env['gcc']:
     sources += ['#src/gallium/state_trackers/wgl/opengl32.mingw.def']
diff --git a/src/gallium/targets/libgl-gdi/gdi_softpipe_winsys.c b/src/gallium/targets/libgl-gdi/gdi_softpipe_winsys.c
deleted file mode 100644
index 4ac507ff9c..0000000000
--- a/src/gallium/targets/libgl-gdi/gdi_softpipe_winsys.c
+++ /dev/null
@@ -1,124 +0,0 @@
-/**************************************************************************
- *
- * Copyright 2009 VMware, Inc.
- * All Rights Reserved.
- *
- * Permission is hereby granted, free of charge, to any person obtaining a
- * copy of this software and associated documentation files (the
- * "Software"), to deal in the Software without restriction, including
- * without limitation the rights to use, copy, modify, merge, publish,
- * distribute, sub license, and/or sell copies of the Software, and to
- * permit persons to whom the Software is furnished to do so, subject to
- * the following conditions:
- *
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
- * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
- * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
- * THE COPYRIGHT HOLDERS, AUTHORS AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM,
- * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
- * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
- * USE OR OTHER DEALINGS IN THE SOFTWARE.
- *
- * The above copyright notice and this permission notice (including the
- * next paragraph) shall be included in all copies or substantial portions
- * of the Software.
- *
- *
- **************************************************************************/
-
-/**
- * @file
- * LLVMpipe support.
- *
- * @author Jose Fonseca <jfonseca@vmware.com>
- */
-
-
-#include <windows.h>
-
-#include "stw_winsys.h"
-#include "gdi/gdi_sw_winsys.h"
-#include "softpipe/sp_texture.h"
-#include "softpipe/sp_screen.h"
-#include "softpipe/sp_public.h"
-
-
-static struct pipe_screen *
-gdi_softpipe_screen_create(void)
-{
-   static struct sw_winsys *winsys;
-   struct pipe_screen *screen;
-
-   winsys = gdi_create_sw_winsys();
-   if(!winsys)
-      goto no_winsys;
-
-   screen = softpipe_create_screen(winsys);
-   if(!screen)
-      goto no_screen;
-
-   return screen;
-   
-no_screen:
-   winsys->destroy(winsys);
-no_winsys:
-   return NULL;
-}
-
-
-
-
-static void
-gdi_softpipe_present(struct pipe_screen *screen,
-                     struct pipe_surface *surface,
-                     HDC hDC)
-{
-   /* This will fail if any interposing layer (trace, debug, etc) has
-    * been introduced between the state-trackers and softpipe.
-    *
-    * Ideally this would get replaced with a call to
-    * pipe_screen::flush_frontbuffer().
-    *
-    * Failing that, it may be necessary for intervening layers to wrap
-    * other structs such as this stw_winsys as well...
-    */
-   gdi_sw_display(softpipe_screen(screen)->winsys,
-                  softpipe_resource(surface->texture)->dt,
-                  hDC);
-}
-
-
-static const struct stw_winsys stw_winsys = {
-   &gdi_softpipe_screen_create,
-   &gdi_softpipe_present,
-   NULL, /* get_adapter_luid */
-   NULL, /* shared_surface_open */
-   NULL, /* shared_surface_close */
-   NULL  /* compose */
-};
-
-
-BOOL WINAPI
-DllMain(HINSTANCE hinstDLL, DWORD fdwReason, LPVOID lpReserved)
-{
-   switch (fdwReason) {
-   case DLL_PROCESS_ATTACH:
-      stw_init(&stw_winsys);
-      stw_init_thread();
-      break;
-
-   case DLL_THREAD_ATTACH:
-      stw_init_thread();
-      break;
-
-   case DLL_THREAD_DETACH:
-      stw_cleanup_thread();
-      break;
-
-   case DLL_PROCESS_DETACH:
-      stw_cleanup_thread();
-      stw_cleanup();
-      break;
-   }
-   return TRUE;
-}
diff --git a/src/gallium/targets/libgl-gdi/gdi_llvmpipe_winsys.c b/src/gallium/targets/libgl-gdi/libgl_gdi.c
index 58c941a03b..112904ab5f 100644
--- a/src/gallium/targets/libgl-gdi/gdi_llvmpipe_winsys.c
+++ b/src/gallium/targets/libgl-gdi/libgl_gdi.c
@@ -1,6 +1,6 @@
 /**************************************************************************
  *
- * Copyright 2009 VMware, Inc.
+ * Copyright 2009-2010 VMware, Inc.
  * All Rights Reserved.
  *
  * Permission is hereby granted, free of charge, to any person obtaining a
@@ -28,7 +28,7 @@
 
 /**
  * @file
- * LLVMpipe support.
+ * Softpipe/LLVMpipe support.
  *
  * @author Jose Fonseca <jfonseca@vmware.com>
  */
@@ -36,24 +36,56 @@
 
 #include <windows.h>
 
+#include "util/u_debug.h"
 #include "stw_winsys.h"
 #include "gdi/gdi_sw_winsys.h"
+
+#include "softpipe/sp_texture.h"
+#include "softpipe/sp_screen.h"
+#include "softpipe/sp_public.h"
+
+#ifdef HAVE_LLVMPIPE
 #include "llvmpipe/lp_texture.h"
 #include "llvmpipe/lp_screen.h"
 #include "llvmpipe/lp_public.h"
+#endif
+
+
+static boolean use_llvmpipe = FALSE;
 
 
 static struct pipe_screen *
-gdi_llvmpipe_screen_create(void)
+gdi_screen_create(void)
 {
-   static struct sw_winsys *winsys;
-   struct pipe_screen *screen;
+   const char *default_driver;
+   const char *driver;
+   struct pipe_screen *screen = NULL;
+   struct sw_winsys *winsys;
 
    winsys = gdi_create_sw_winsys();
    if(!winsys)
       goto no_winsys;
 
-   screen = llvmpipe_create_screen(winsys);
+#ifdef HAVE_LLVMPIPE
+   default_driver = "llvmpipe";
+#else
+   default_driver = "softpipe";
+#endif
+
+   driver = debug_get_option("GALLIUM_DRIVER", default_driver);
+
+#ifdef HAVE_LLVMPIPE
+   if (strcmp(driver, "llvmpipe") == 0) {
+      screen = llvmpipe_create_screen( winsys );
+   }
+#endif
+
+   if (screen == NULL) {
+      screen = softpipe_create_screen( winsys );
+   } else {
+      use_llvmpipe = TRUE;
+   }
+
    if(!screen)
       goto no_screen;
 
@@ -66,15 +98,13 @@ no_winsys:
 }
 
 
-
-
 static void
-gdi_llvmpipe_present(struct pipe_screen *screen,
-                     struct pipe_surface *surface,
-                     HDC hDC)
+gdi_present(struct pipe_screen *screen,
+            struct pipe_resource *res,
+            HDC hDC)
 {
    /* This will fail if any interposing layer (trace, debug, etc) has
-    * been introduced between the state-trackers and llvmpipe.
+    * been introduced between the state-trackers and the pipe driver.
     *
     * Ideally this would get replaced with a call to
     * pipe_screen::flush_frontbuffer().
@@ -82,15 +112,28 @@ gdi_llvmpipe_present(struct pipe_screen *screen,
     * Failing that, it may be necessary for intervening layers to wrap
     * other structs such as this stw_winsys as well...
     */
-   gdi_sw_display(llvmpipe_screen(screen)->winsys,
-                  llvmpipe_resource(surface->texture)->dt,
-                  hDC);
+
+   struct sw_winsys *winsys = NULL;
+   struct sw_displaytarget *dt = NULL;
+
+#ifdef HAVE_LLVMPIPE
+   if (use_llvmpipe) {
+      winsys = llvmpipe_screen(screen)->winsys;
+      dt = llvmpipe_resource(res)->dt;
+      gdi_sw_display(winsys, dt, hDC);
+      return;
+   }
+#endif
+
+   winsys = softpipe_screen(screen)->winsys,
+   dt = softpipe_resource(res)->dt,
+   gdi_sw_display(winsys, dt, hDC);
 }
 
 
 static const struct stw_winsys stw_winsys = {
-   &gdi_llvmpipe_screen_create,
-   &gdi_llvmpipe_present,
+   &gdi_screen_create,
+   &gdi_present,
    NULL, /* get_adapter_luid */
    NULL, /* shared_surface_open */
    NULL, /* shared_surface_close */
diff --git a/src/gallium/targets/libgl-xlib/Makefile b/src/gallium/targets/libgl-xlib/Makefile
index 076a040a5a..fb537c3155 100644
--- a/src/gallium/targets/libgl-xlib/Makefile
+++ b/src/gallium/targets/libgl-xlib/Makefile
@@ -26,6 +26,8 @@ INCLUDE_DIRS = \
 
 DEFINES += \
 	-DGALLIUM_SOFTPIPE \
+	-DGALLIUM_RBUG \
+	-DGALLIUM_TRACE \
 	-DGALLIUM_GALAHAD
 #-DGALLIUM_CELL will be defined by the config */
 
@@ -44,7 +46,7 @@ LIBS = \
 	$(TOP)/src/gallium/winsys/sw/xlib/libws_xlib.a \
 	$(TOP)/src/gallium/drivers/trace/libtrace.a \
 	$(TOP)/src/gallium/drivers/rbug/librbug.a \
-	$(TOP)/src/gallium/drivers/identity/libidentity.a \
+	$(TOP)/src/gallium/drivers/galahad/libgalahad.a \
 	$(TOP)/src/mapi/glapi/libglapi.a \
 	$(TOP)/src/mesa/libmesagallium.a \
 	$(GALLIUM_AUXILIARIES) \
diff --git a/src/gallium/targets/libgl-xlib/SConscript b/src/gallium/targets/libgl-xlib/SConscript
index 582760eac9..d932736be7 100644
--- a/src/gallium/targets/libgl-xlib/SConscript
+++ b/src/gallium/targets/libgl-xlib/SConscript
@@ -19,9 +19,6 @@ env.Prepend(LIBS = env['X11_LIBS'])
 env.Prepend(LIBS = [
     st_xlib,
     ws_xlib,
-    trace,
-    rbug,
-    identity,
     glapi,
     mesa,
     glsl,
@@ -34,15 +31,11 @@ sources = [
 ]
 
 if True:
-    env.Append(CPPDEFINES = 'GALLIUM_SOFTPIPE')
-    env.Prepend(LIBS = [softpipe])
-
-if True:
-    env.Append(CPPDEFINES = 'GALLIUM_GALAHAD')
-    env.Prepend(LIBS = [galahad])
+    env.Append(CPPDEFINES = ['GALLIUM_TRACE', 'GALLIUM_RBUG', 'GALLIUM_GALAHAD', 'GALLIUM_SOFTPIPE'])
+    env.Prepend(LIBS = [trace, rbug, galahad, softpipe])
 
 if env['llvm']:
-    env.Append(CPPDEFINES = 'GALLIUM_LLVMPIPE')
+    env.Append(CPPDEFINES = ['GALLIUM_LLVMPIPE'])
     env.Prepend(LIBS = [llvmpipe])
     
 if False:
diff --git a/src/gallium/targets/libgl-xlib/xlib.c b/src/gallium/targets/libgl-xlib/xlib.c
index b0d1e52910..9a3e0e07b0 100644
--- a/src/gallium/targets/libgl-xlib/xlib.c
+++ b/src/gallium/targets/libgl-xlib/xlib.c
@@ -32,17 +32,15 @@
  */
 #include "pipe/p_compiler.h"
 #include "util/u_debug.h"
-#include "target-helpers/wrap_screen.h"
-#include "target-helpers/inline_sw_helper.h"
 #include "state_tracker/xlib_sw_winsys.h"
 #include "xm_public.h"
 
 #include "state_tracker/st_api.h"
 #include "state_tracker/st_gl_api.h"
+#include "target-helpers/inline_sw_helper.h"
+#include "target-helpers/inline_debug_helper.h"
+
 
-#if defined(GALLIUM_GALAHAD)
-#include "galahad/glhd_public.h"
-#endif
 
 /* Helper function to build a subset of a driver stack consisting of
  * one of the software rasterizers (cell, llvmpipe, softpipe) and the
@@ -67,18 +65,9 @@ swrast_xlib_create_screen( Display *display )
    if (screen == NULL)
       goto fail;
 
-   /* XXX will fix soon */
-#if defined(GALLIUM_GALAHAD)
-   if (screen) {
-      struct pipe_screen *galahad_screen = galahad_screen_create( screen );
-      if (galahad_screen)
-         screen = galahad_screen;
-   }
-#endif
-
    /* Inject any wrapping layers we want to here:
     */
-   return gallium_wrap_screen( screen );
+   return debug_screen_wrap( screen );
 
 fail:
    if (winsys)
diff --git a/src/gallium/targets/xorg-vmwgfx/vmw_video.c b/src/gallium/targets/xorg-vmwgfx/vmw_video.c
index 94465e5204..7979209993 100644
--- a/src/gallium/targets/xorg-vmwgfx/vmw_video.c
+++ b/src/gallium/targets/xorg-vmwgfx/vmw_video.c
@@ -362,8 +362,10 @@ vmw_video_close(struct vmw_customizer *vmw)
 	/* make sure the port is stoped as well */
 	vmw_xv_stop_video(pScrn, &video->port[i], TRUE);
 	vmw_ioctl_unref_stream(vmw, video->port[i].streamId);
+	REGION_UNINIT(pScreen, &video->port[i].clipBoxes);
     }
 
+
     /* XXX: I'm sure this function is missing code for turning off Xv */
 
     free(vmw->video_priv);
@@ -448,7 +450,16 @@ vmw_video_init_adaptor(ScrnInfoPtr pScrn, struct vmw_customizer *vmw)
     vmw->video_priv = video;
 
     adaptor->type = XvInputMask | XvImageMask | XvWindowMask;
-    adaptor->flags = VIDEO_OVERLAID_IMAGES | VIDEO_CLIP_TO_VIEWPORT;
+
+    /**
+     * Note: CLIP_TO_VIEWPORT was removed from the flags, since with the
+     * crtc/output based modesetting, the viewport is not updated on
+     * RandR modeswitches. Hence the video may incorrectly be clipped away.
+     * The correct approach, (if needed) would be to clip against the
+     * scanout area union of all active crtcs. Revisit if needed.
+     */
+
+    adaptor->flags = VIDEO_OVERLAID_IMAGES;
     adaptor->name = "VMware Video Engine";
     adaptor->nEncodings = VMWARE_VID_NUM_ENCODINGS;
     adaptor->pEncodings = vmwareVideoEncodings;
@@ -463,6 +474,7 @@ vmw_video_init_adaptor(ScrnInfoPtr pScrn, struct vmw_customizer *vmw)
         video->port[i].flags = SVGA_VIDEO_FLAG_COLORKEY;
         video->port[i].colorKey = VMWARE_VIDEO_COLORKEY;
         video->port[i].isAutoPaintColorkey = TRUE;
+        REGION_NULL(pScrn->pScreen, &video->port[i].clipBoxes);
         adaptor->pPortPrivates[i].ptr = &video->port[i];
     }
 
@@ -554,7 +566,9 @@ vmw_video_port_init(ScrnInfoPtr pScrn, struct vmw_video_port *port,
     REGION_COPY(pScrn->pScreen, &port->clipBoxes, clipBoxes);
 
     if (port->isAutoPaintColorkey)
-        xf86XVFillKeyHelper(pScrn->pScreen, port->colorKey, clipBoxes);
+	xf86XVFillKeyHelper(pScrn->pScreen, port->colorKey, clipBoxes);
+
+    xorg_flush(pScrn->pScreen);
 
     return port->play(pScrn, port, src_x, src_y, drw_x, drw_y, src_w, src_h,
                       drw_w, drw_h, format, buf, width, height, clipBoxes);
@@ -641,11 +655,12 @@ vmw_video_port_play(ScrnInfoPtr pScrn, struct vmw_video_port *port,
      */
     if (!REGION_EQUAL(pScrn->pScreen, &port->clipBoxes, clipBoxes)) {
         REGION_COPY(pScrn->pScreen, &port->clipBoxes, clipBoxes);
-        if (port->isAutoPaintColorkey) {
+        if (port->isAutoPaintColorkey)
             xf86XVFillKeyHelper(pScrn->pScreen, port->colorKey, clipBoxes);
-        }
     }
 
+    xorg_flush(pScrn->pScreen);
+
     ret = drmCommandWrite(vmw->fd, DRM_VMW_CONTROL_STREAM, &arg, sizeof(arg));
     if (ret) {
 	vmw_video_port_cleanup(pScrn, port);
@@ -865,6 +880,8 @@ vmw_xv_stop_video(ScrnInfoPtr pScrn, pointer data, Bool cleanup)
     if (!vmw->video_priv)
         return;
 
+    REGION_EMPTY(pScrn->pScreen, &port->clipBoxes);
+
     if (!cleanup)
         return;
 
diff --git a/src/gallium/tests/graw/clear.c b/src/gallium/tests/graw/clear.c
index ee4581ef1e..1ff80cadee 100644
--- a/src/gallium/tests/graw/clear.c
+++ b/src/gallium/tests/graw/clear.c
@@ -20,6 +20,7 @@ static const int HEIGHT = 300;
 struct pipe_screen *screen;
 struct pipe_context *ctx;
 struct pipe_surface *surf;
+struct pipe_resource *tex;
 static void *window = NULL;
 
 static void draw( void )
@@ -31,13 +32,14 @@ static void draw( void )
 
    graw_save_surface_to_file(ctx, surf, NULL);
 
-   screen->flush_frontbuffer(screen, surf, window);
+   screen->flush_frontbuffer(screen, tex, 0, 0, window);
 }
 
 static void init( void )
 {
    struct pipe_framebuffer_state fb;
-   struct pipe_resource *tex, templat;
+   struct pipe_resource templat;
+   struct pipe_surface surf_tmpl;
    int i;
 
    /* It's hard to say whether window or screen should be created
@@ -66,6 +68,7 @@ static void init( void )
    templat.width0 = WIDTH;
    templat.height0 = HEIGHT;
    templat.depth0 = 1;
+   templat.array_size = 1;
    templat.last_level = 0;
    templat.nr_samples = 1;
    templat.bind = (PIPE_BIND_RENDER_TARGET |
@@ -76,9 +79,12 @@ static void init( void )
    if (tex == NULL)
       exit(4);
 
-   surf = screen->get_tex_surface(screen, tex, 0, 0, 0,
-                                  PIPE_BIND_RENDER_TARGET |
-                                  PIPE_BIND_DISPLAY_TARGET);
+   surf_tmpl.format = templat.format;
+   surf_tmpl.usage = PIPE_BIND_RENDER_TARGET;
+   surf_tmpl.u.tex.level = 0;
+   surf_tmpl.u.tex.first_layer = 0;
+   surf_tmpl.u.tex.last_layer = 0;
+   surf = ctx->create_surface(ctx, tex, &surf_tmpl);
    if (surf == NULL)
       exit(5);
 
diff --git a/src/gallium/tests/graw/fs-test.c b/src/gallium/tests/graw/fs-test.c
index 19af83fda8..37be2d0830 100644
--- a/src/gallium/tests/graw/fs-test.c
+++ b/src/gallium/tests/graw/fs-test.c
@@ -119,6 +119,7 @@ static void init_fs_constbuf( void )
    templat.width0 = sizeof(constants1);
    templat.height0 = 1;
    templat.depth0 = 1;
+   templat.array_size = 1;
    templat.last_level = 0;
    templat.nr_samples = 1;
    templat.bind = PIPE_BIND_CONSTANT_BUFFER;
@@ -139,7 +140,7 @@ static void init_fs_constbuf( void )
 
       ctx->transfer_inline_write(ctx,
                                  constbuf1,
-                                 u_subresource(0,0),
+                                 0,
                                  PIPE_TRANSFER_WRITE,
                                  &box,
                                  constants1,
@@ -156,7 +157,7 @@ static void init_fs_constbuf( void )
 
       ctx->transfer_inline_write(ctx,
                                  constbuf2,
-                                 u_subresource(0,0),
+                                 0,
                                  PIPE_TRANSFER_WRITE,
                                  &box,
                                  constants2,
@@ -280,7 +281,7 @@ static void draw( void )
 
    graw_save_surface_to_file(ctx, surf, NULL);
 
-   screen->flush_frontbuffer(screen, surf, window);
+   screen->flush_frontbuffer(screen, rttex, 0, 0, window);
 }
 
 #define SIZE 16
@@ -340,6 +341,7 @@ static void init_tex( void )
    templat.width0 = SIZE;
    templat.height0 = SIZE;
    templat.depth0 = 1;
+   templat.array_size = 1;
    templat.last_level = 0;
    templat.nr_samples = 1;
    templat.bind = PIPE_BIND_SAMPLER_VIEW;
@@ -354,7 +356,7 @@ static void init_tex( void )
 
    ctx->transfer_inline_write(ctx,
                               samptex,
-                              u_subresource(0,0),
+                              0,
                               PIPE_TRANSFER_WRITE,
                               &box,
                               tex2d,
@@ -368,7 +370,7 @@ static void init_tex( void )
       struct pipe_transfer *t;
       uint32_t *ptr;
       t = pipe_get_transfer(ctx, samptex,
-                            0, 0, 0, /* face, level, zslice */
+                            0, 0, /* level, layer */
                             PIPE_TRANSFER_READ,
                             0, 0, SIZE, SIZE); /* x, y, width, height */
 
@@ -387,8 +389,6 @@ static void init_tex( void )
    memset(&sv_template, 0, sizeof sv_template);
    sv_template.format = samptex->format;
    sv_template.texture = samptex;
-   sv_template.first_level = 0;
-   sv_template.last_level = 0;
    sv_template.swizzle_r = 0;
    sv_template.swizzle_g = 1;
    sv_template.swizzle_b = 2;
@@ -424,6 +424,7 @@ static void init( void )
 {
    struct pipe_framebuffer_state fb;
    struct pipe_resource templat;
+   struct pipe_surface surf_tmpl;
    int i;
 
    /* It's hard to say whether window or screen should be created
@@ -450,6 +451,7 @@ static void init( void )
    templat.width0 = WIDTH;
    templat.height0 = HEIGHT;
    templat.depth0 = 1;
+   templat.array_size = 1;
    templat.last_level = 0;
    templat.nr_samples = 1;
    templat.bind = (PIPE_BIND_RENDER_TARGET |
@@ -460,9 +462,12 @@ static void init( void )
    if (rttex == NULL)
       exit(4);
 
-   surf = screen->get_tex_surface(screen, rttex, 0, 0, 0,
-                                  PIPE_BIND_RENDER_TARGET |
-                                  PIPE_BIND_DISPLAY_TARGET);
+   surf_tmpl.format = templat.format;
+   surf_tmpl.usage = PIPE_BIND_RENDER_TARGET;
+   surf_tmpl.u.tex.level = 0;
+   surf_tmpl.u.tex.first_layer = 0;
+   surf_tmpl.u.tex.last_layer = 0;
+   surf = ctx->create_surface(ctx, rttex, &surf_tmpl);
    if (surf == NULL)
       exit(5);
 
diff --git a/src/gallium/tests/graw/gs-test.c b/src/gallium/tests/graw/gs-test.c
index ef29f13498..812666a8c8 100644
--- a/src/gallium/tests/graw/gs-test.c
+++ b/src/gallium/tests/graw/gs-test.c
@@ -156,6 +156,7 @@ static void init_fs_constbuf( void )
    templat.width0 = sizeof(constants1);
    templat.height0 = 1;
    templat.depth0 = 1;
+   templat.array_size = 1;
    templat.last_level = 0;
    templat.nr_samples = 1;
    templat.bind = PIPE_BIND_CONSTANT_BUFFER;
@@ -172,7 +173,7 @@ static void init_fs_constbuf( void )
 
       ctx->transfer_inline_write(ctx,
                                  constbuf1,
-                                 u_subresource(0,0),
+                                 0,
                                  PIPE_TRANSFER_WRITE,
                                  &box,
                                  constants1,
@@ -189,7 +190,7 @@ static void init_fs_constbuf( void )
 
       ctx->transfer_inline_write(ctx,
                                  constbuf2,
-                                 u_subresource(0,0),
+                                 0,
                                  PIPE_TRANSFER_WRITE,
                                  &box,
                                  constants2,
@@ -344,7 +345,7 @@ static void draw( void )
 
    graw_save_surface_to_file(ctx, surf, NULL);
 
-   screen->flush_frontbuffer(screen, surf, window);
+   screen->flush_frontbuffer(screen, rttex, 0, 0, window);
 }
 
 #define SIZE 16
@@ -404,6 +405,7 @@ static void init_tex( void )
    templat.width0 = SIZE;
    templat.height0 = SIZE;
    templat.depth0 = 1;
+   templat.array_size = 1;
    templat.last_level = 0;
    templat.nr_samples = 1;
    templat.bind = PIPE_BIND_SAMPLER_VIEW;
@@ -418,7 +420,7 @@ static void init_tex( void )
 
    ctx->transfer_inline_write(ctx,
                               samptex,
-                              u_subresource(0,0),
+                              0,
                               PIPE_TRANSFER_WRITE,
                               &box,
                               tex2d,
@@ -432,7 +434,7 @@ static void init_tex( void )
       struct pipe_transfer *t;
       uint32_t *ptr;
       t = pipe_get_transfer(ctx, samptex,
-                            0, 0, 0, /* face, level, zslice */
+                            0, 0, /* level, layer */
                             PIPE_TRANSFER_READ,
                             0, 0, SIZE, SIZE); /* x, y, width, height */
 
@@ -451,8 +453,6 @@ static void init_tex( void )
    memset(&sv_template, 0, sizeof sv_template);
    sv_template.format = samptex->format;
    sv_template.texture = samptex;
-   sv_template.first_level = 0;
-   sv_template.last_level = 0;
    sv_template.swizzle_r = 0;
    sv_template.swizzle_g = 1;
    sv_template.swizzle_b = 2;
@@ -488,6 +488,7 @@ static void init( void )
 {
    struct pipe_framebuffer_state fb;
    struct pipe_resource templat;
+   struct pipe_surface surf_tmpl;
    int i;
 
    /* It's hard to say whether window or screen should be created
@@ -514,6 +515,7 @@ static void init( void )
    templat.width0 = WIDTH;
    templat.height0 = HEIGHT;
    templat.depth0 = 1;
+   templat.array_size = 1;
    templat.last_level = 0;
    templat.nr_samples = 1;
    templat.bind = (PIPE_BIND_RENDER_TARGET |
@@ -524,9 +526,12 @@ static void init( void )
    if (rttex == NULL)
       exit(4);
 
-   surf = screen->get_tex_surface(screen, rttex, 0, 0, 0,
-                                  PIPE_BIND_RENDER_TARGET |
-                                  PIPE_BIND_DISPLAY_TARGET);
+   surf_tmpl.format = templat.format;
+   surf_tmpl.usage = PIPE_BIND_RENDER_TARGET;
+   surf_tmpl.u.tex.level = 0;
+   surf_tmpl.u.tex.first_layer = 0;
+   surf_tmpl.u.tex.last_layer = 0;
+   surf = ctx->create_surface(ctx, rttex, &surf_tmpl);
    if (surf == NULL)
       exit(5);
 
diff --git a/src/gallium/tests/graw/quad-tex.c b/src/gallium/tests/graw/quad-tex.c
index 35eade939e..952131d765 100644
--- a/src/gallium/tests/graw/quad-tex.c
+++ b/src/gallium/tests/graw/quad-tex.c
@@ -151,7 +151,7 @@ static void draw( void )
 
    graw_save_surface_to_file(ctx, surf, NULL);
 
-   screen->flush_frontbuffer(screen, surf, window);
+   screen->flush_frontbuffer(screen, rttex, 0, 0, window);
 }
 
 #define SIZE 16
@@ -211,6 +211,7 @@ static void init_tex( void )
    templat.width0 = SIZE;
    templat.height0 = SIZE;
    templat.depth0 = 1;
+   templat.array_size = 1;
    templat.last_level = 0;
    templat.nr_samples = 1;
    templat.bind = PIPE_BIND_SAMPLER_VIEW;
@@ -225,7 +226,7 @@ static void init_tex( void )
 
    ctx->transfer_inline_write(ctx,
                               samptex,
-                              u_subresource(0,0),
+                              0,
                               PIPE_TRANSFER_WRITE,
                               &box,
                               tex2d,
@@ -239,7 +240,7 @@ static void init_tex( void )
       struct pipe_transfer *t;
       uint32_t *ptr;
       t = pipe_get_transfer(ctx, samptex,
-                            0, 0, 0, /* face, level, zslice */
+                            0, 0, /* level, layer */
                             PIPE_TRANSFER_READ,
                             0, 0, SIZE, SIZE); /* x, y, width, height */
 
@@ -258,8 +259,6 @@ static void init_tex( void )
    memset(&sv_template, 0, sizeof sv_template);
    sv_template.format = samptex->format;
    sv_template.texture = samptex;
-   sv_template.first_level = 0;
-   sv_template.last_level = 0;
    sv_template.swizzle_r = 0;
    sv_template.swizzle_g = 1;
    sv_template.swizzle_b = 2;
@@ -295,6 +294,7 @@ static void init( void )
 {
    struct pipe_framebuffer_state fb;
    struct pipe_resource templat;
+   struct pipe_surface surf_tmpl;
    int i;
 
    /* It's hard to say whether window or screen should be created
@@ -321,6 +321,7 @@ static void init( void )
    templat.width0 = WIDTH;
    templat.height0 = HEIGHT;
    templat.depth0 = 1;
+   templat.array_size = 1;
    templat.last_level = 0;
    templat.nr_samples = 1;
    templat.bind = (PIPE_BIND_RENDER_TARGET |
@@ -331,9 +332,12 @@ static void init( void )
    if (rttex == NULL)
       exit(4);
 
-   surf = screen->get_tex_surface(screen, rttex, 0, 0, 0,
-                                  PIPE_BIND_RENDER_TARGET |
-                                  PIPE_BIND_DISPLAY_TARGET);
+   surf_tmpl.format = templat.format;
+   surf_tmpl.usage = PIPE_BIND_RENDER_TARGET;
+   surf_tmpl.u.tex.level = 0;
+   surf_tmpl.u.tex.first_layer = 0;
+   surf_tmpl.u.tex.last_layer = 0;
+   surf = ctx->create_surface(ctx, rttex, &surf_tmpl);
    if (surf == NULL)
       exit(5);
 
diff --git a/src/gallium/tests/graw/shader-leak.c b/src/gallium/tests/graw/shader-leak.c
index 0a6c362d17..b53f0a046c 100644
--- a/src/gallium/tests/graw/shader-leak.c
+++ b/src/gallium/tests/graw/shader-leak.c
@@ -28,6 +28,7 @@ static const int HEIGHT = 300;
 static struct pipe_screen *screen = NULL;
 static struct pipe_context *ctx = NULL;
 static struct pipe_surface *surf = NULL;
+static struct pipe_resource *tex = NULL;
 static void *window = NULL;
 
 struct vertex {
@@ -155,7 +156,7 @@ static void draw( void )
       ctx->delete_fs_state(ctx, fs);
    }
 
-   screen->flush_frontbuffer(screen, surf, window);
+   screen->flush_frontbuffer(screen, tex, 0, 0, window);
    ctx->destroy(ctx);
 
    exit(0);
@@ -165,7 +166,8 @@ static void draw( void )
 static void init( void )
 {
    struct pipe_framebuffer_state fb;
-   struct pipe_resource *tex, templat;
+   struct pipe_resource templat;
+   struct pipe_surface surf_tmpl;
    int i;
 
    /* It's hard to say whether window or screen should be created
@@ -203,11 +205,16 @@ static void init( void )
       exit(4);
    }
 
-   surf = screen->get_tex_surface(screen, tex, 0, 0, 0,
-                                  PIPE_BIND_RENDER_TARGET |
-                                  PIPE_BIND_DISPLAY_TARGET);
-   if (surf == NULL)
+   surf_tmpl.format = templat.format;
+   surf_tmpl.usage = PIPE_BIND_RENDER_TARGET;
+   surf_tmpl.u.tex.level = 0;
+   surf_tmpl.u.tex.first_layer = 0;
+   surf_tmpl.u.tex.last_layer = 0;
+   surf = ctx->create_surface(ctx, tex, &surf_tmpl);
+   if (surf == NULL) {
+      fprintf(stderr, "Unable to create tex surface!\n");
       exit(5);
+   }
 
    memset(&fb, 0, sizeof fb);
    fb.nr_cbufs = 1;
diff --git a/src/gallium/tests/graw/tri-gs.c b/src/gallium/tests/graw/tri-gs.c
index 731c4e10cf..84ff3e6773 100644
--- a/src/gallium/tests/graw/tri-gs.c
+++ b/src/gallium/tests/graw/tri-gs.c
@@ -23,6 +23,7 @@ static const int HEIGHT = 300;
 static struct pipe_screen *screen = NULL;
 static struct pipe_context *ctx = NULL;
 static struct pipe_surface *surf = NULL;
+static struct pipe_resource *tex = NULL;
 static void *window = NULL;
 
 struct vertex {
@@ -164,14 +165,15 @@ static void draw( void )
    util_draw_arrays(ctx, PIPE_PRIM_TRIANGLES, 0, 3);
    ctx->flush(ctx, PIPE_FLUSH_RENDER_CACHE, NULL);
 
-   screen->flush_frontbuffer(screen, surf, window);
+   screen->flush_frontbuffer(screen, tex, 0, 0, window);
 }
 
 
 static void init( void )
 {
    struct pipe_framebuffer_state fb;
-   struct pipe_resource *tex, templat;
+   struct pipe_resource templat;
+   struct pipe_surface surf_tmpl;
    int i;
 
    /* It's hard to say whether window or screen should be created
@@ -198,6 +200,7 @@ static void init( void )
    templat.width0 = WIDTH;
    templat.height0 = HEIGHT;
    templat.depth0 = 1;
+   templat.array_size = 1;
    templat.last_level = 0;
    templat.nr_samples = 1;
    templat.bind = (PIPE_BIND_RENDER_TARGET |
@@ -208,9 +211,12 @@ static void init( void )
    if (tex == NULL)
       exit(4);
 
-   surf = screen->get_tex_surface(screen, tex, 0, 0, 0,
-                                  PIPE_BIND_RENDER_TARGET |
-                                  PIPE_BIND_DISPLAY_TARGET);
+   surf_tmpl.format = templat.format;
+   surf_tmpl.usage = PIPE_BIND_RENDER_TARGET;
+   surf_tmpl.u.tex.level = 0;
+   surf_tmpl.u.tex.first_layer = 0;
+   surf_tmpl.u.tex.last_layer = 0;
+   surf = ctx->create_surface(ctx, tex, &surf_tmpl);
    if (surf == NULL)
       exit(5);
 
diff --git a/src/gallium/tests/graw/tri-instanced.c b/src/gallium/tests/graw/tri-instanced.c
index 7644381162..f33c061b22 100644
--- a/src/gallium/tests/graw/tri-instanced.c
+++ b/src/gallium/tests/graw/tri-instanced.c
@@ -27,6 +27,7 @@ static const int HEIGHT = 300;
 static struct pipe_screen *screen = NULL;
 static struct pipe_context *ctx = NULL;
 static struct pipe_surface *surf = NULL;
+static struct pipe_resource *tex = NULL;
 static void *window = NULL;
 
 struct vertex {
@@ -216,14 +217,15 @@ static void draw( void )
 
    graw_save_surface_to_file(ctx, surf, NULL);
 
-   screen->flush_frontbuffer(screen, surf, window);
+   screen->flush_frontbuffer(screen, tex, 0, 0, window);
 }
 
 
 static void init( void )
 {
    struct pipe_framebuffer_state fb;
-   struct pipe_resource *tex, templat;
+   struct pipe_resource templat;
+   struct pipe_surface surf_tmpl;
    int i;
 
    /* It's hard to say whether window or screen should be created
@@ -250,6 +252,7 @@ static void init( void )
    templat.width0 = WIDTH;
    templat.height0 = HEIGHT;
    templat.depth0 = 1;
+   templat.array_size = 1;
    templat.last_level = 0;
    templat.nr_samples = 1;
    templat.bind = (PIPE_BIND_RENDER_TARGET |
@@ -260,9 +263,12 @@ static void init( void )
    if (tex == NULL)
       exit(4);
 
-   surf = screen->get_tex_surface(screen, tex, 0, 0, 0,
-                                  PIPE_BIND_RENDER_TARGET |
-                                  PIPE_BIND_DISPLAY_TARGET);
+   surf_tmpl.format = templat.format;
+   surf_tmpl.usage = PIPE_BIND_RENDER_TARGET;
+   surf_tmpl.u.tex.level = 0;
+   surf_tmpl.u.tex.first_layer = 0;
+   surf_tmpl.u.tex.last_layer = 0;
+   surf = ctx->create_surface(ctx, tex, &surf_tmpl);
    if (surf == NULL)
       exit(5);
 
diff --git a/src/gallium/tests/graw/tri.c b/src/gallium/tests/graw/tri.c
index 025a1470dc..2742c7c99e 100644
--- a/src/gallium/tests/graw/tri.c
+++ b/src/gallium/tests/graw/tri.c
@@ -25,6 +25,7 @@ static const int HEIGHT = 300;
 static struct pipe_screen *screen = NULL;
 static struct pipe_context *ctx = NULL;
 static struct pipe_surface *surf = NULL;
+static struct pipe_resource *tex = NULL;
 static void *window = NULL;
 
 struct vertex {
@@ -144,14 +145,15 @@ static void draw( void )
 
    graw_save_surface_to_file(ctx, surf, NULL);
 
-   screen->flush_frontbuffer(screen, surf, window);
+   screen->flush_frontbuffer(screen, tex, 0, 0, window);
 }
 
 
 static void init( void )
 {
    struct pipe_framebuffer_state fb;
-   struct pipe_resource *tex, templat;
+   struct pipe_resource templat;
+   struct pipe_surface surf_tmpl;
    int i;
 
    /* It's hard to say whether window or screen should be created
@@ -180,6 +182,7 @@ static void init( void )
    templat.width0 = WIDTH;
    templat.height0 = HEIGHT;
    templat.depth0 = 1;
+   templat.array_size = 1;
    templat.last_level = 0;
    templat.nr_samples = 1;
    templat.bind = (PIPE_BIND_RENDER_TARGET |
@@ -192,11 +195,14 @@ static void init( void )
       exit(4);
    }
 
-   surf = screen->get_tex_surface(screen, tex, 0, 0, 0,
-                                  PIPE_BIND_RENDER_TARGET |
-                                  PIPE_BIND_DISPLAY_TARGET);
+   surf_tmpl.format = templat.format;
+   surf_tmpl.usage = PIPE_BIND_RENDER_TARGET;
+   surf_tmpl.u.tex.level = 0;
+   surf_tmpl.u.tex.first_layer = 0;
+   surf_tmpl.u.tex.last_layer = 0;
+   surf = ctx->create_surface(ctx, tex, &surf_tmpl);
    if (surf == NULL) {
-      fprintf(stderr, "Unable to get tex surface!\n");
+      fprintf(stderr, "Unable to create tex surface!\n");
       exit(5);
    }
 
diff --git a/src/gallium/tests/graw/vs-test.c b/src/gallium/tests/graw/vs-test.c
index 440c40bdcd..58908f38a2 100644
--- a/src/gallium/tests/graw/vs-test.c
+++ b/src/gallium/tests/graw/vs-test.c
@@ -87,6 +87,7 @@ static void init_fs_constbuf( void )
    templat.width0 = sizeof(constants);
    templat.height0 = 1;
    templat.depth0 = 1;
+   templat.array_size = 1;
    templat.last_level = 0;
    templat.nr_samples = 1;
    templat.bind = PIPE_BIND_CONSTANT_BUFFER;
@@ -101,7 +102,7 @@ static void init_fs_constbuf( void )
 
    ctx->transfer_inline_write(ctx,
                               constbuf,
-                              u_subresource(0,0),
+                              0,
                               PIPE_TRANSFER_WRITE,
                               &box,
                               constants,
@@ -231,7 +232,7 @@ static void draw( void )
 
    graw_save_surface_to_file(ctx, surf, NULL);
 
-   screen->flush_frontbuffer(screen, surf, window);
+   screen->flush_frontbuffer(screen, rttex, 0, 0, window);
 }
 
 #define SIZE 16
@@ -291,6 +292,7 @@ static void init_tex( void )
    templat.width0 = SIZE;
    templat.height0 = SIZE;
    templat.depth0 = 1;
+   templat.array_size = 1;
    templat.last_level = 0;
    templat.nr_samples = 1;
    templat.bind = PIPE_BIND_SAMPLER_VIEW;
@@ -305,7 +307,7 @@ static void init_tex( void )
 
    ctx->transfer_inline_write(ctx,
                               samptex,
-                              u_subresource(0,0),
+                              0,
                               PIPE_TRANSFER_WRITE,
                               &box,
                               tex2d,
@@ -319,7 +321,7 @@ static void init_tex( void )
       struct pipe_transfer *t;
       uint32_t *ptr;
       t = pipe_get_transfer(ctx, samptex,
-                            0, 0, 0, /* face, level, zslice */
+                            0, 0, /* level, layer */
                             PIPE_TRANSFER_READ,
                             0, 0, SIZE, SIZE); /* x, y, width, height */
 
@@ -338,8 +340,6 @@ static void init_tex( void )
    memset(&sv_template, 0, sizeof sv_template);
    sv_template.format = samptex->format;
    sv_template.texture = samptex;
-   sv_template.first_level = 0;
-   sv_template.last_level = 0;
    sv_template.swizzle_r = 0;
    sv_template.swizzle_g = 1;
    sv_template.swizzle_b = 2;
@@ -375,6 +375,7 @@ static void init( void )
 {
    struct pipe_framebuffer_state fb;
    struct pipe_resource templat;
+   struct pipe_surface surf_tmpl;
    int i;
 
    /* It's hard to say whether window or screen should be created
@@ -401,6 +402,7 @@ static void init( void )
    templat.width0 = WIDTH;
    templat.height0 = HEIGHT;
    templat.depth0 = 1;
+   templat.array_size = 1;
    templat.last_level = 0;
    templat.nr_samples = 1;
    templat.bind = (PIPE_BIND_RENDER_TARGET |
@@ -411,9 +413,12 @@ static void init( void )
    if (rttex == NULL)
       exit(4);
 
-   surf = screen->get_tex_surface(screen, rttex, 0, 0, 0,
-                                  PIPE_BIND_RENDER_TARGET |
-                                  PIPE_BIND_DISPLAY_TARGET);
+   surf_tmpl.format = templat.format;
+   surf_tmpl.usage = PIPE_BIND_RENDER_TARGET;
+   surf_tmpl.u.tex.level = 0;
+   surf_tmpl.u.tex.first_layer = 0;
+   surf_tmpl.u.tex.last_layer = 0;
+   surf = ctx->create_surface(ctx, rttex, &surf_tmpl);
    if (surf == NULL)
       exit(5);
 
diff --git a/src/gallium/tests/python/retrace/interpreter.py b/src/gallium/tests/python/retrace/interpreter.py
index 954a701a53..84371223f6 100755
--- a/src/gallium/tests/python/retrace/interpreter.py
+++ b/src/gallium/tests/python/retrace/interpreter.py
@@ -251,14 +251,6 @@ class Screen(Object):
     def texture_release(self, surface):
         pass
 
-    def get_tex_surface(self, texture, face, level, zslice, usage):
-        if texture is None:
-            return None
-        return texture.get_surface(face, level, zslice)
-    
-    def tex_surface_destroy(self, surface):
-        self.interpreter.unregister_object(surface)
-
     def tex_surface_release(self, surface):
         pass
 
@@ -282,7 +274,7 @@ class Screen(Object):
     def fence_reference(self, dst, src):
         pass
     
-    def flush_frontbuffer(self, surface):
+    def flush_frontbuffer(self, resource):
         pass
 
 
@@ -581,7 +573,7 @@ class Context(Object):
         if transfer and usage & gallium.PIPE_TRANSFER_READ:
             if self.interpreter.options.all:
                 surface = texture.get_surface(sr.face, sr.level, box.z)
-                self.interpreter.present(self.real, transfer.surface, 'transf_read', box.x, box.y, box.w, box.h)
+                self.interpreter.present(self.real, transfer.surface, 'transf_read', box.x, box.y, box.width, box.height)
         return transfer
     
     def tex_transfer_destroy(self, transfer):
@@ -589,6 +581,10 @@ class Context(Object):
 
     def transfer_inline_write(self, resource, sr, usage, box, stride, slice_stride, data):
         self.real.transfer_inline_write(resource, sr, usage, box, data, stride, slice_stride)
+        if self.interpreter.options.all:
+            for z in range(box.z, box.z + box.depth):
+                surface = resource.get_surface(sr.face, sr.level, box.z)
+                self.interpreter.present(self.real, surface, 'transf_inline_write%u' % z, box.x, box.y, box.width, box.height)
 
     def _set_dirty(self):
         if self.interpreter.options.step:
@@ -627,7 +623,13 @@ class Context(Object):
         if self.zsbuf:
             if self.interpreter.options.all:
                 self.interpreter.present(self.real, self.zsbuf, "zsbuf")
-    
+    def create_surface(self, texture, level, layer, usage):
+        if texture is None:
+            return None
+        return texture.get_surface(level, layer)
+
+    def surface_destroy(self, surface):
+        self.interpreter.unregister_object(surface)
 
 class Interpreter(parser.TraceDumper):
     
diff --git a/src/gallium/tests/trivial/quad-tex.c b/src/gallium/tests/trivial/quad-tex.c
index cf88edcdc5..92c5b4dbb1 100644
--- a/src/gallium/tests/trivial/quad-tex.c
+++ b/src/gallium/tests/trivial/quad-tex.c
@@ -92,6 +92,7 @@ struct program
 
 static void init_prog(struct program *p)
 {
+	struct pipe_surface surf_tmpl;
 	/* create the software rasterizer */
 	p->screen = sw_screen_create(null_sw_create());
 	/* wrap the screen with any debugger */
@@ -141,6 +142,7 @@ static void init_prog(struct program *p)
 		tmplt.width0 = WIDTH;
 		tmplt.height0 = HEIGHT;
 		tmplt.depth0 = 1;
+		tmplt.array_size = 1;
 		tmplt.last_level = 0;
 		tmplt.bind = PIPE_BIND_RENDER_TARGET;
 
@@ -153,7 +155,6 @@ static void init_prog(struct program *p)
 		struct pipe_transfer *t;
 		struct pipe_resource t_tmplt;
 		struct pipe_sampler_view v_tmplt;
-		struct pipe_subresource sub;
 		struct pipe_box box;
 
 		memset(&t_tmplt, 0, sizeof(t_tmplt));
@@ -162,17 +163,17 @@ static void init_prog(struct program *p)
 		t_tmplt.width0 = 2;
 		t_tmplt.height0 = 2;
 		t_tmplt.depth0 = 1;
+		t_tmplt.array_size = 1;
 		t_tmplt.last_level = 0;
 		t_tmplt.bind = PIPE_BIND_RENDER_TARGET;
 
 		p->tex = p->screen->resource_create(p->screen, &t_tmplt);
 
-		memset(&sub, 0, sizeof(sub));
 		memset(&box, 0, sizeof(box));
 		box.width = 2;
 		box.height = 2;
 
-		t = p->pipe->get_transfer(p->pipe, p->tex, sub, PIPE_TRANSFER_WRITE, &box);
+		t = p->pipe->get_transfer(p->pipe, p->tex, 0, PIPE_TRANSFER_WRITE, &box);
 
 		ptr = p->pipe->transfer_map(p->pipe, t);
 		ptr[0] = 0xffff0000;
@@ -210,12 +211,17 @@ static void init_prog(struct program *p)
 	p->sampler.mag_img_filter = PIPE_TEX_MIPFILTER_LINEAR;
 	p->sampler.normalized_coords = 1;
 
+	surf_tmpl.format = templat.format;
+	surf_tmpl.usage = PIPE_BIND_RENDER_TARGET;
+	surf_tmpl.u.tex.level = 0;
+	surf_tmpl.u.tex.first_layer = 0;
+	surf_tmpl.u.tex.last_layer = 0;
 	/* drawing destination */
 	memset(&p->framebuffer, 0, sizeof(p->framebuffer));
 	p->framebuffer.width = WIDTH;
 	p->framebuffer.height = HEIGHT;
 	p->framebuffer.nr_cbufs = 1;
-	p->framebuffer.cbufs[0] = p->screen->get_tex_surface(p->screen, p->target, 0, 0, 0, PIPE_BIND_RENDER_TARGET);
+	p->framebuffer.cbufs[0] = p->pipe->create_surface(p->pipe, p->target, &surf_tmpl);
 
 	/* viewport, depth isn't really needed */
 	{
diff --git a/src/gallium/tests/trivial/tri.c b/src/gallium/tests/trivial/tri.c
index 667a27b28a..37c1573051 100644
--- a/src/gallium/tests/trivial/tri.c
+++ b/src/gallium/tests/trivial/tri.c
@@ -87,6 +87,7 @@ struct program
 
 static void init_prog(struct program *p)
 {
+	struct pipe_surface surf_tmpl;
 	/* create the software rasterizer */
 	p->screen = sw_screen_create(null_sw_create());
 	/* wrap the screen with any debugger */
@@ -132,6 +133,7 @@ static void init_prog(struct program *p)
 		tmplt.width0 = WIDTH;
 		tmplt.height0 = HEIGHT;
 		tmplt.depth0 = 1;
+		tmplt.array_size = 1;
 		tmplt.last_level = 0;
 		tmplt.bind = PIPE_BIND_RENDER_TARGET;
 
@@ -150,12 +152,17 @@ static void init_prog(struct program *p)
 	p->rasterizer.cull_face = PIPE_FACE_NONE;
 	p->rasterizer.gl_rasterization_rules = 1;
 
+	surf_tmpl.format = templat.format;
+	surf_tmpl.usage = PIPE_BIND_RENDER_TARGET;
+	surf_tmpl.u.tex.level = 0;
+	surf_tmpl.u.tex.first_layer = 0;
+	surf_tmpl.u.tex.last_layer = 0;
 	/* drawing destination */
 	memset(&p->framebuffer, 0, sizeof(p->framebuffer));
 	p->framebuffer.width = WIDTH;
 	p->framebuffer.height = HEIGHT;
 	p->framebuffer.nr_cbufs = 1;
-	p->framebuffer.cbufs[0] = p->screen->get_tex_surface(p->screen, p->target, 0, 0, 0, PIPE_BIND_RENDER_TARGET);
+	p->framebuffer.cbufs[0] = p->pipe->create_surface(p->pipe, p->target, &surf_tmpl);
 
 	/* viewport, depth isn't really needed */
 	{
diff --git a/src/gallium/winsys/i915/drm/i915_drm_batchbuffer.c b/src/gallium/winsys/i915/drm/i915_drm_batchbuffer.c
index e50e7801c0..ebe86dcf19 100644
--- a/src/gallium/winsys/i915/drm/i915_drm_batchbuffer.c
+++ b/src/gallium/winsys/i915/drm/i915_drm_batchbuffer.c
@@ -14,9 +14,6 @@
 #define INTEL_BATCH_CLIPRECTS    0x2
 
 #undef INTEL_RUN_SYNC
-#undef INTEL_MAP_BATCHBUFFER
-#undef INTEL_MAP_GTT
-#define INTEL_ALWAYS_FLUSH
 
 struct i915_drm_batchbuffer
 {
@@ -41,7 +38,7 @@ i915_drm_batchbuffer_reset(struct i915_drm_batchbuffer *batch)
 
    if (batch->bo)
       drm_intel_bo_unreference(batch->bo);
-   batch->bo = drm_intel_bo_alloc(idws->pools.gem,
+   batch->bo = drm_intel_bo_alloc(idws->gem_manager,
                                   "gallium3d_batchbuffer",
                                   batch->actual_size,
                                   4096);
@@ -72,11 +69,7 @@ i915_drm_batchbuffer_create(struct i915_winsys *iws)
 
    batch->actual_size = idws->max_batch_size;
 
-#ifdef INTEL_MAP_BATCHBUFFER
-   batch->base.map = NULL;
-#else
    batch->base.map = MALLOC(batch->actual_size);
-#endif
    batch->base.ptr = NULL;
    batch->base.size = 0;
 
@@ -94,7 +87,7 @@ static int
 i915_drm_batchbuffer_reloc(struct i915_winsys_batchbuffer *ibatch,
                             struct i915_winsys_buffer *buffer,
                             enum i915_winsys_buffer_usage usage,
-                            unsigned pre_add)
+                            unsigned pre_add, bool fenced)
 {
    struct i915_drm_batchbuffer *batch = i915_drm_batchbuffer(ibatch);
    unsigned write_domain = 0;
@@ -104,37 +97,44 @@ i915_drm_batchbuffer_reloc(struct i915_winsys_batchbuffer *ibatch,
 
    assert(batch->base.relocs < batch->base.max_relocs);
 
-   if (usage == I915_USAGE_SAMPLER) {
+   switch (usage) {
+   case I915_USAGE_SAMPLER:
       write_domain = 0;
       read_domain = I915_GEM_DOMAIN_SAMPLER;
-
-   } else if (usage == I915_USAGE_RENDER) {
+      break;
+   case I915_USAGE_RENDER:
       write_domain = I915_GEM_DOMAIN_RENDER;
       read_domain = I915_GEM_DOMAIN_RENDER;
-
-   } else if (usage == I915_USAGE_2D_TARGET) {
+      break;
+   case I915_USAGE_2D_TARGET:
       write_domain = I915_GEM_DOMAIN_RENDER;
       read_domain = I915_GEM_DOMAIN_RENDER;
-
-   } else if (usage == I915_USAGE_2D_SOURCE) {
+      break;
+   case I915_USAGE_2D_SOURCE:
       write_domain = 0;
       read_domain = I915_GEM_DOMAIN_RENDER;
-
-   } else if (usage == I915_USAGE_VERTEX) {
+      break;
+   case I915_USAGE_VERTEX:
       write_domain = 0;
       read_domain = I915_GEM_DOMAIN_VERTEX;
-
-   } else {
+      break;
+   default:
       assert(0);
       return -1;
    }
 
    offset = (unsigned)(batch->base.ptr - batch->base.map);
 
-   ret = drm_intel_bo_emit_reloc(batch->bo, offset,
-                                 intel_bo(buffer), pre_add,
-                                 read_domain,
-                                 write_domain);
+   if (fenced)
+      ret = drm_intel_bo_emit_reloc_fence(batch->bo, offset,
+				    intel_bo(buffer), pre_add,
+				    read_domain,
+				    write_domain);
+   else
+      ret = drm_intel_bo_emit_reloc(batch->bo, offset,
+				    intel_bo(buffer), pre_add,
+				    read_domain,
+				    write_domain);
 
    ((uint32_t*)batch->base.ptr)[0] = intel_bo(buffer)->offset + pre_add;
    batch->base.ptr += 4;
@@ -150,70 +150,32 @@ i915_drm_batchbuffer_flush(struct i915_winsys_batchbuffer *ibatch,
                             struct pipe_fence_handle **fence)
 {
    struct i915_drm_batchbuffer *batch = i915_drm_batchbuffer(ibatch);
-   unsigned used = 0;
-   int ret = 0;
+   unsigned used;
+   int ret;
 
-   assert(i915_winsys_batchbuffer_space(ibatch) >= 0);
+   /* MI_BATCH_BUFFER_END */
+   i915_winsys_batchbuffer_dword_unchecked(ibatch, (0xA<<23));
 
    used = batch->base.ptr - batch->base.map;
-   assert((used & 3) == 0);
-
-
-#ifdef INTEL_ALWAYS_FLUSH
-   /* MI_FLUSH | FLUSH_MAP_CACHE */
-   i915_winsys_batchbuffer_dword(ibatch, (0x4<<23)|(1<<0));
-   used += 4;
-#endif
-
-   if ((used & 4) == 0) {
+   if (used & 4) {
       /* MI_NOOP */
-      i915_winsys_batchbuffer_dword(ibatch, 0);
+      i915_winsys_batchbuffer_dword_unchecked(ibatch, 0);
+      used += 4;
    }
-   /* MI_BATCH_BUFFER_END */
-   i915_winsys_batchbuffer_dword(ibatch, (0xA<<23));
-
-   used = batch->base.ptr - batch->base.map;
-   assert((used & 4) == 0);
-
-#ifdef INTEL_MAP_BATCHBUFFER
-#ifdef INTEL_MAP_GTT
-   drm_intel_gem_bo_unmap_gtt(batch->bo);
-#else
-   drm_intel_bo_unmap(batch->bo);
-#endif
-#else
-   drm_intel_bo_subdata(batch->bo, 0, used, batch->base.map);
-#endif
 
    /* Do the sending to HW */
-   if (i915_drm_winsys(ibatch->iws)->send_cmd)
+   ret = drm_intel_bo_subdata(batch->bo, 0, used, batch->base.map);
+   if (ret == 0 && i915_drm_winsys(ibatch->iws)->send_cmd)
       ret = drm_intel_bo_exec(batch->bo, used, NULL, 0, 0);
-   else
-      ret = 0;
 
    if (ret != 0 || i915_drm_winsys(ibatch->iws)->dump_cmd) {
-#ifdef INTEL_MAP_BATCHBUFFER
-#ifdef INTEL_MAP_GTT
-      drm_intel_gem_bo_map_gtt(batch->bo);
-#else
-      drm_intel_bo_map(batch->bo, 0);
-#endif
-#endif
       i915_dump_batchbuffer(ibatch);
       assert(ret == 0);
-#ifdef INTEL_MAP_BATCHBUFFER
-#ifdef INTEL_MAP_GTT
-   drm_intel_gem_bo_unmap_gtt(batch->bo);
-#else
-   drm_intel_bo_unmap(batch->bo);
-#endif
-#endif
-   } else {
+   }
+
 #ifdef INTEL_RUN_SYNC
-      drm_intel_bo_map(batch->bo, FALSE);
-      drm_intel_bo_unmap(batch->bo);
+   drm_intel_bo_wait_rendering(batch->bo);
 #endif
-   }
 
    if (fence) {
       ibatch->iws->fence_reference(ibatch->iws, fence, NULL);
@@ -237,9 +199,7 @@ i915_drm_batchbuffer_destroy(struct i915_winsys_batchbuffer *ibatch)
    if (batch->bo)
       drm_intel_bo_unreference(batch->bo);
 
-#ifndef INTEL_MAP_BATCHBUFFER
    FREE(batch->base.map);
-#endif
    FREE(batch);
 }
 
diff --git a/src/gallium/winsys/i915/drm/i915_drm_buffer.c b/src/gallium/winsys/i915/drm/i915_drm_buffer.c
index 6b06e7ae99..01dd4bf062 100644
--- a/src/gallium/winsys/i915/drm/i915_drm_buffer.c
+++ b/src/gallium/winsys/i915/drm/i915_drm_buffer.c
@@ -5,15 +5,31 @@
 
 #include "i915_drm.h"
 
+static char *i915_drm_type_to_name(enum i915_winsys_buffer_type type)
+{
+   char *name;
+
+   if (type == I915_NEW_TEXTURE) {
+      name = "gallium3d_texture";
+   } else if (type == I915_NEW_VERTEX) {
+      name = "gallium3d_vertex";
+   } else if (type == I915_NEW_SCANOUT) {
+      name = "gallium3d_scanout";
+   } else {
+      assert(0);
+      name = "gallium3d_unknown";
+   }
+
+   return name;
+}
+
 static struct i915_winsys_buffer *
 i915_drm_buffer_create(struct i915_winsys *iws,
-                        unsigned size, unsigned alignment,
+                        unsigned size,
                         enum i915_winsys_buffer_type type)
 {
    struct i915_drm_buffer *buf = CALLOC_STRUCT(i915_drm_buffer);
    struct i915_drm_winsys *idws = i915_drm_winsys(iws);
-   drm_intel_bufmgr *pool;
-   char *name;
 
    if (!buf)
       return NULL;
@@ -21,30 +37,49 @@ i915_drm_buffer_create(struct i915_winsys *iws,
    buf->magic = 0xDEAD1337;
    buf->flinked = FALSE;
    buf->flink = 0;
-   buf->map_gtt = FALSE;
 
-   if (type == I915_NEW_TEXTURE) {
-      name = "gallium3d_texture";
-      pool = idws->pools.gem;
-   } else if (type == I915_NEW_VERTEX) {
-      name = "gallium3d_vertex";
-      pool = idws->pools.gem;
-      buf->map_gtt = TRUE;
-   } else if (type == I915_NEW_SCANOUT) {
-      name = "gallium3d_scanout";
-      pool = idws->pools.gem;
-      buf->map_gtt = TRUE;
-   } else {
-      assert(0);
-      name = "gallium3d_unknown";
-      pool = idws->pools.gem;
-   }
+   buf->bo = drm_intel_bo_alloc(idws->gem_manager,
+                                i915_drm_type_to_name(type), size, 0);
 
-   buf->bo = drm_intel_bo_alloc(pool, name, size, alignment);
+   if (!buf->bo)
+      goto err;
+
+   return (struct i915_winsys_buffer *)buf;
+
+err:
+   assert(0);
+   FREE(buf);
+   return NULL;
+}
+
+static struct i915_winsys_buffer *
+i915_drm_buffer_create_tiled(struct i915_winsys *iws,
+                             unsigned *stride, unsigned height, 
+                             enum i915_winsys_buffer_tile *tiling,
+                             enum i915_winsys_buffer_type type)
+{
+   struct i915_drm_buffer *buf = CALLOC_STRUCT(i915_drm_buffer);
+   struct i915_drm_winsys *idws = i915_drm_winsys(iws);
+   unsigned long pitch = 0;
+   uint32_t tiling_mode = *tiling;
+
+   if (!buf)
+      return NULL;
+
+   buf->magic = 0xDEAD1337;
+   buf->flinked = FALSE;
+   buf->flink = 0;
+
+   buf->bo = drm_intel_bo_alloc_tiled(idws->gem_manager,
+                                      i915_drm_type_to_name(type),
+		   		      *stride, height, 1,
+                                      &tiling_mode, &pitch, 0);
 
    if (!buf->bo)
       goto err;
 
+   *stride = pitch;
+   *tiling = tiling_mode;
    return (struct i915_winsys_buffer *)buf;
 
 err:
@@ -55,8 +90,9 @@ err:
 
 static struct i915_winsys_buffer *
 i915_drm_buffer_from_handle(struct i915_winsys *iws,
-                             struct winsys_handle *whandle,
-                             unsigned *stride)
+                            struct winsys_handle *whandle,
+                            enum i915_winsys_buffer_tile *tiling,
+                            unsigned *stride)
 {
    struct i915_drm_winsys *idws = i915_drm_winsys(iws);
    struct i915_drm_buffer *buf = CALLOC_STRUCT(i915_drm_buffer);
@@ -66,7 +102,7 @@ i915_drm_buffer_from_handle(struct i915_winsys *iws,
       return NULL;
 
    buf->magic = 0xDEAD1337;
-   buf->bo = drm_intel_bo_gem_create_from_name(idws->pools.gem, "gallium3d_from_handle", whandle->handle);
+   buf->bo = drm_intel_bo_gem_create_from_name(idws->gem_manager, "gallium3d_from_handle", whandle->handle);
    buf->flinked = TRUE;
    buf->flink = whandle->handle;
 
@@ -74,10 +110,9 @@ i915_drm_buffer_from_handle(struct i915_winsys *iws,
       goto err;
 
    drm_intel_bo_get_tiling(buf->bo, &tile, &swizzle);
-   if (tile != I915_TILE_NONE)
-      buf->map_gtt = TRUE;
 
    *stride = whandle->stride;
+   *tiling = tile;
 
    return (struct i915_winsys_buffer *)buf;
 
@@ -113,25 +148,6 @@ i915_drm_buffer_get_handle(struct i915_winsys *iws,
    return TRUE;
 }
 
-static int
-i915_drm_buffer_set_fence_reg(struct i915_winsys *iws,
-                               struct i915_winsys_buffer *buffer,
-                               unsigned stride,
-                               enum i915_winsys_buffer_tile tile)
-{
-   struct i915_drm_buffer *buf = i915_drm_buffer(buffer);
-   assert(I915_TILING_NONE == I915_TILE_NONE);
-   assert(I915_TILING_X == I915_TILE_X);
-   assert(I915_TILING_Y == I915_TILE_Y);
-
-   if (tile != I915_TILE_NONE) {
-      assert(buf->map_count == 0);
-      buf->map_gtt = TRUE;
-   }
-
-   return drm_intel_bo_set_tiling(buf->bo, &tile, stride);
-}
-
 static void *
 i915_drm_buffer_map(struct i915_winsys *iws,
                      struct i915_winsys_buffer *buffer,
@@ -146,10 +162,7 @@ i915_drm_buffer_map(struct i915_winsys *iws,
    if (buf->map_count)
       goto out;
 
-   if (buf->map_gtt)
-      ret = drm_intel_gem_bo_map_gtt(bo);
-   else
-      ret = drm_intel_bo_map(bo, write);
+   ret = drm_intel_gem_bo_map_gtt(bo);
 
    buf->ptr = bo->virtual;
 
@@ -171,10 +184,7 @@ i915_drm_buffer_unmap(struct i915_winsys *iws,
    if (--buf->map_count)
       return;
 
-   if (buf->map_gtt)
-      drm_intel_gem_bo_unmap_gtt(intel_bo(buffer));
-   else
-      drm_intel_bo_unmap(intel_bo(buffer));
+   drm_intel_gem_bo_unmap_gtt(intel_bo(buffer));
 }
 
 static int
@@ -207,9 +217,9 @@ void
 i915_drm_winsys_init_buffer_functions(struct i915_drm_winsys *idws)
 {
    idws->base.buffer_create = i915_drm_buffer_create;
+   idws->base.buffer_create_tiled = i915_drm_buffer_create_tiled;
    idws->base.buffer_from_handle = i915_drm_buffer_from_handle;
    idws->base.buffer_get_handle = i915_drm_buffer_get_handle;
-   idws->base.buffer_set_fence_reg = i915_drm_buffer_set_fence_reg;
    idws->base.buffer_map = i915_drm_buffer_map;
    idws->base.buffer_unmap = i915_drm_buffer_unmap;
    idws->base.buffer_write = i915_drm_buffer_write;
diff --git a/src/gallium/winsys/i915/drm/i915_drm_winsys.c b/src/gallium/winsys/i915/drm/i915_drm_winsys.c
index 179a84a704..2288b48b2b 100644
--- a/src/gallium/winsys/i915/drm/i915_drm_winsys.c
+++ b/src/gallium/winsys/i915/drm/i915_drm_winsys.c
@@ -40,7 +40,7 @@ i915_drm_winsys_destroy(struct i915_winsys *iws)
 {
    struct i915_drm_winsys *idws = i915_drm_winsys(iws);
 
-   drm_intel_bufmgr_destroy(idws->pools.gem);
+   drm_intel_bufmgr_destroy(idws->gem_manager);
 
    FREE(idws);
 }
@@ -67,8 +67,9 @@ i915_drm_winsys_create(int drmFD)
 
    idws->base.destroy = i915_drm_winsys_destroy;
 
-   idws->pools.gem = drm_intel_bufmgr_gem_init(idws->fd, idws->max_batch_size);
-   drm_intel_bufmgr_gem_enable_reuse(idws->pools.gem);
+   idws->gem_manager = drm_intel_bufmgr_gem_init(idws->fd, idws->max_batch_size);
+   drm_intel_bufmgr_gem_enable_reuse(idws->gem_manager);
+   drm_intel_bufmgr_gem_enable_fenced_relocs(idws->gem_manager);
 
    idws->dump_cmd = debug_get_bool_option("I915_DUMP_CMD", FALSE);
    idws->send_cmd = !debug_get_bool_option("I915_NO_HW", FALSE);
diff --git a/src/gallium/winsys/i915/drm/i915_drm_winsys.h b/src/gallium/winsys/i915/drm/i915_drm_winsys.h
index 88a71f2424..0d74d0270c 100644
--- a/src/gallium/winsys/i915/drm/i915_drm_winsys.h
+++ b/src/gallium/winsys/i915/drm/i915_drm_winsys.h
@@ -24,9 +24,7 @@ struct i915_drm_winsys
 
    size_t max_batch_size;
 
-   struct {
-      drm_intel_bufmgr *gem;
-   } pools;
+   drm_intel_bufmgr *gem_manager;
 };
 
 static INLINE struct i915_drm_winsys *
@@ -54,7 +52,6 @@ struct i915_drm_buffer {
 
    void *ptr;
    unsigned map_count;
-   boolean map_gtt;
 
    boolean flinked;
    unsigned flink;
diff --git a/src/gallium/winsys/i915/sw/i915_sw_batchbuffer.c b/src/gallium/winsys/i915/sw/i915_sw_batchbuffer.c
index a480cfed57..44773ae30e 100644
--- a/src/gallium/winsys/i915/sw/i915_sw_batchbuffer.c
+++ b/src/gallium/winsys/i915/sw/i915_sw_batchbuffer.c
@@ -61,7 +61,7 @@ static int
 i915_sw_batchbuffer_reloc(struct i915_winsys_batchbuffer *ibatch,
                           struct i915_winsys_buffer *buffer,
                           enum i915_winsys_buffer_usage usage,
-                          unsigned pre_add)
+                          unsigned pre_add, bool fenced)
 {
    struct i915_sw_batchbuffer *batch = i915_sw_batchbuffer(ibatch);
    int ret = 0;
diff --git a/src/gallium/winsys/i915/sw/i915_sw_buffer.c b/src/gallium/winsys/i915/sw/i915_sw_buffer.c
index df17568886..834805e621 100644
--- a/src/gallium/winsys/i915/sw/i915_sw_buffer.c
+++ b/src/gallium/winsys/i915/sw/i915_sw_buffer.c
@@ -4,28 +4,15 @@
 
 static struct i915_winsys_buffer *
 i915_sw_buffer_create(struct i915_winsys *iws,
-                      unsigned size, unsigned alignment,
+                      unsigned size,
                       enum i915_winsys_buffer_type type)
 {
    struct i915_sw_buffer *buf = CALLOC_STRUCT(i915_sw_buffer);
-   char *name;
 
    if (!buf)
       return NULL;
 
-   if (type == I915_NEW_TEXTURE) {
-      name = "gallium3d_texture";
-   } else if (type == I915_NEW_VERTEX) {
-      name = "gallium3d_vertex";
-   } else if (type == I915_NEW_SCANOUT) {
-      name = "gallium3d_scanout";
-   } else {
-      assert(0);
-      name = "gallium3d_unknown";
-   }
-
    buf->magic = 0xDEAD1337;
-   buf->name = name;
    buf->type = type;
    buf->ptr = CALLOC(size, 1);
 
@@ -40,21 +27,32 @@ err:
    return NULL;
 }
 
-static int
-i915_sw_buffer_set_fence_reg(struct i915_winsys *iws,
-                               struct i915_winsys_buffer *buffer,
-                               unsigned stride,
-                               enum i915_winsys_buffer_tile tile)
+static struct i915_winsys_buffer *
+i915_sw_buffer_create_tiled(struct i915_winsys *iws,
+                      unsigned *stride, unsigned height, 
+                      enum i915_winsys_buffer_tile *tiling,
+                      enum i915_winsys_buffer_type type)
 {
-   struct i915_sw_buffer *buf = i915_sw_buffer(buffer);
+   struct i915_sw_buffer *buf = CALLOC_STRUCT(i915_sw_buffer);
+
+   if (!buf)
+      return NULL;
+
+   buf->magic = 0xDEAD1337;
+   buf->type = type;
+   buf->ptr = CALLOC(*stride * height, 1);
+   buf->tiling = *tiling;
+   buf->stride = *stride;
 
-   if (tile != I915_TILE_NONE) {
-      assert(buf->map_count == 0);
-   }
+   if (!buf->ptr)
+      goto err;
 
-   buf->tile = tile;
+   return (struct i915_winsys_buffer *)buf;
 
-   return 0;
+err:
+   assert(0);
+   FREE(buf);
+   return NULL;
 }
 
 static void *
@@ -108,7 +106,7 @@ void
 i915_sw_winsys_init_buffer_functions(struct i915_sw_winsys *isws)
 {
    isws->base.buffer_create = i915_sw_buffer_create;
-   isws->base.buffer_set_fence_reg = i915_sw_buffer_set_fence_reg;
+   isws->base.buffer_create_tiled = i915_sw_buffer_create_tiled;
    isws->base.buffer_map = i915_sw_buffer_map;
    isws->base.buffer_unmap = i915_sw_buffer_unmap;
    isws->base.buffer_write = i915_sw_buffer_write;
diff --git a/src/gallium/winsys/i915/sw/i915_sw_winsys.h b/src/gallium/winsys/i915/sw/i915_sw_winsys.h
index b7b43669f3..3af2548419 100644
--- a/src/gallium/winsys/i915/sw/i915_sw_winsys.h
+++ b/src/gallium/winsys/i915/sw/i915_sw_winsys.h
@@ -43,8 +43,8 @@ struct i915_sw_buffer {
    void *ptr;
    unsigned map_count;
    enum i915_winsys_buffer_type type;
-   enum i915_winsys_buffer_tile tile;
-   const char *name;
+   enum i915_winsys_buffer_tile tiling;
+   unsigned stride;
 };
 
 static INLINE struct i915_sw_buffer *
diff --git a/src/gallium/winsys/i965/xlib/xlib_i965.c b/src/gallium/winsys/i965/xlib/xlib_i965.c
index baadd6e89c..c22df6643a 100644
--- a/src/gallium/winsys/i965/xlib/xlib_i965.c
+++ b/src/gallium/winsys/i965/xlib/xlib_i965.c
@@ -42,6 +42,7 @@
 
 #include "i965/brw_winsys.h"
 #include "i965/brw_screen.h"
+#include "i965/brw_resource.h"
 #include "i965/brw_reg.h"
 #include "i965/brw_structs_dump.h"
 
@@ -421,25 +422,28 @@ xlib_create_brw_winsys_screen( void )
 
 static void
 xlib_i965_display_surface(struct xmesa_buffer *xm_buffer,
-                          struct pipe_surface *surf)
+                          struct pipe_resource *resource,
+                          unsigned level, unsigned layer)
 {
-   struct brw_surface *surface = brw_surface(surf);
-   struct xlib_brw_buffer *bo = xlib_brw_buffer(surface->bo);
-
+   struct brw_texture *tex = brw_texture(resource);
+   struct xlib_brw_buffer *bo = xlib_brw_buffer(tex->bo);
+   /* not sure if the resource is really useful here but
+      since it was never implemented anyway... */
    if (BRW_DEBUG & DEBUG_WINSYS)
-      debug_printf("%s offset %x+%x sz %dx%d\n", __FUNCTION__, 
+      debug_printf("%s level %u layer %u offset %x base sz %dx%d\n", __FUNCTION__,
+                   level, layer,
                    bo->offset,
-                   surface->draw_offset,
-                   surf->width,
-                   surf->height);
+                   resource->width0,
+                   resource->height0);
 }
 
 static void
 xlib_i965_flush_frontbuffer(struct pipe_screen *screen,
-			    struct pipe_surface *surf,
-			    void *context_private)
+                            struct pipe_resource *resource,
+                            unsigned level, unsigned layer,
+                            void *context_private)
 {
-   xlib_i965_display_surface(NULL, surf);
+   xlib_i965_display_surface(NULL, resource, level, layer);
 }
 
 
diff --git a/src/gallium/winsys/r600/drm/Makefile b/src/gallium/winsys/r600/drm/Makefile
index a396205f89..91c65012c8 100644
--- a/src/gallium/winsys/r600/drm/Makefile
+++ b/src/gallium/winsys/r600/drm/Makefile
@@ -8,12 +8,12 @@ C_SOURCES = \
 	bof.c \
 	evergreen_hw_context.c \
 	radeon_bo.c \
-	radeon_bo_pb.c \
 	radeon_pciid.c \
 	r600.c \
 	r600_bo.c \
 	r600_drm.c \
-	r600_hw_context.c
+	r600_hw_context.c \
+	r600_bomgr.c
 
 LIBRARY_INCLUDES = -I$(TOP)/src/gallium/drivers/r600 \
 		   $(shell pkg-config libdrm --cflags-only-I)
diff --git a/src/gallium/winsys/r600/drm/SConscript b/src/gallium/winsys/r600/drm/SConscript
index cc053c06dd..dac0097f14 100644
--- a/src/gallium/winsys/r600/drm/SConscript
+++ b/src/gallium/winsys/r600/drm/SConscript
@@ -6,12 +6,12 @@ r600_sources = [
     'bof.c',
     'evergreen_hw_context.c',
     'radeon_bo.c',
-    'radeon_bo_pb.c',
     'radeon_pciid.c',
     'r600.c',
     'r600_bo.c',
     'r600_drm.c',
     'r600_hw_context.c',
+    'r600_bomgr.c',
 ]
 
 env.ParseConfig('pkg-config --cflags libdrm_radeon')
diff --git a/src/gallium/winsys/r600/drm/evergreen_hw_context.c b/src/gallium/winsys/r600/drm/evergreen_hw_context.c
index 7f21b53ace..2175d578ec 100644
--- a/src/gallium/winsys/r600/drm/evergreen_hw_context.c
+++ b/src/gallium/winsys/r600/drm/evergreen_hw_context.c
@@ -36,7 +36,6 @@
 #include "pipe/p_compiler.h"
 #include "util/u_inlines.h"
 #include "util/u_memory.h"
-#include <pipebuffer/pb_bufmgr.h>
 #include "r600_priv.h"
 
 #define GROUP_FORCE_NEW_BLOCK	0
@@ -577,6 +576,12 @@ int evergreen_context_init(struct r600_context *ctx, struct radeon *radeon)
 		if (r)
 			goto out_err;
 	}
+	/* FS RESOURCE */
+	for (int j = 0, offset = 0x7C00; j < 16; j++, offset += 0x20) {
+		r = evergreen_state_resource_init(ctx, offset);
+		if (r)
+			goto out_err;
+	}
 
 	/* PS loop const */
 	evergreen_loop_const_init(ctx, 0);
@@ -686,6 +691,13 @@ void evergreen_context_pipe_state_set_vs_resource(struct r600_context *ctx, stru
 	evergreen_context_pipe_state_set_resource(ctx, state, offset);
 }
 
+void evergreen_context_pipe_state_set_fs_resource(struct r600_context *ctx, struct r600_pipe_state *state, unsigned rid)
+{
+	unsigned offset = R_030000_SQ_TEX_RESOURCE_WORD0_0 + 0x7C00 + 0x20 * rid;
+
+	evergreen_context_pipe_state_set_resource(ctx, state, offset);
+}
+
 static inline void evergreen_context_pipe_state_set_sampler(struct r600_context *ctx, struct r600_pipe_state *state, unsigned offset)
 {
 	struct r600_range *range;
@@ -842,7 +854,7 @@ void evergreen_context_draw(struct r600_context *ctx, const struct r600_draw *dr
 		ctx->pm4[ctx->pm4_cdwords++] = draw->vgt_draw_initiator;
 	}
 	ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_EVENT_WRITE, 0);
-	ctx->pm4[ctx->pm4_cdwords++] = EVENT_TYPE_CACHE_FLUSH_AND_INV_EVENT;
+	ctx->pm4[ctx->pm4_cdwords++] = EVENT_TYPE(EVENT_TYPE_CACHE_FLUSH_AND_INV_EVENT) | EVENT_INDEX(0);
 
 	/* flush color buffer */
 	for (int i = 0; i < 12; i++) {
@@ -917,3 +929,10 @@ void evergreen_vs_resource_set(struct r600_context *ctx, struct r600_pipe_state
 
 	evergreen_resource_set(ctx, state, offset);
 }
+
+void evergreen_fs_resource_set(struct r600_context *ctx, struct r600_pipe_state *state, unsigned rid)
+{
+	unsigned offset = R_030000_RESOURCE0_WORD0 + 0x7C00 + 0x20 * rid;
+
+	evergreen_resource_set(ctx, state, offset);
+}
diff --git a/src/gallium/winsys/r600/drm/r600.c b/src/gallium/winsys/r600/drm/r600.c
index 0a4d2e791d..b88733f80f 100644
--- a/src/gallium/winsys/r600/drm/r600.c
+++ b/src/gallium/winsys/r600/drm/r600.c
@@ -27,7 +27,6 @@
 #include "radeon_drm.h"
 #include "pipe/p_compiler.h"
 #include "util/u_inlines.h"
-#include <pipebuffer/pb_bufmgr.h>
 #include "r600_priv.h"
 
 enum radeon_family r600_get_family(struct radeon *r600)
@@ -93,6 +92,12 @@ struct radeon *r600_new(int fd, unsigned device)
 	case CHIP_RV730:
 	case CHIP_RV710:
 	case CHIP_RV740:
+	case CHIP_CEDAR:
+	case CHIP_REDWOOD:
+	case CHIP_JUNIPER:
+	case CHIP_CYPRESS:
+	case CHIP_HEMLOCK:
+	case CHIP_PALM:
 		break;
 	case CHIP_R100:
 	case CHIP_RV100:
@@ -121,11 +126,6 @@ struct radeon *r600_new(int fd, unsigned device)
 	case CHIP_RV560:
 	case CHIP_RV570:
 	case CHIP_R580:
-	case CHIP_CEDAR:
-	case CHIP_REDWOOD:
-	case CHIP_JUNIPER:
-	case CHIP_CYPRESS:
-	case CHIP_HEMLOCK:
 	default:
 		R600_ERR("unknown or unsupported chipset 0x%04X\n", r600->device);
 		break;
@@ -154,6 +154,7 @@ struct radeon *r600_new(int fd, unsigned device)
 	case CHIP_JUNIPER:
 	case CHIP_CYPRESS:
 	case CHIP_HEMLOCK:
+	case CHIP_PALM:
 		r600->chip_class = EVERGREEN;
 		break;
 	default:
diff --git a/src/gallium/winsys/r600/drm/r600_bo.c b/src/gallium/winsys/r600/drm/r600_bo.c
index 251f009a6b..6a3737f0a4 100644
--- a/src/gallium/winsys/r600/drm/r600_bo.c
+++ b/src/gallium/winsys/r600/drm/r600_bo.c
@@ -36,142 +36,153 @@ struct r600_bo *r600_bo(struct radeon *radeon,
 			unsigned size, unsigned alignment,
 			unsigned binding, unsigned usage)
 {
-	struct r600_bo *ws_bo = calloc(1, sizeof(struct r600_bo));
-	struct pb_desc desc;
-	struct pb_manager *man;
+	struct r600_bo *bo;
+	struct radeon_bo *rbo;
 
-	desc.alignment = alignment;
-	desc.usage = (PB_USAGE_CPU_READ_WRITE | PB_USAGE_GPU_READ_WRITE);
-	ws_bo->size = size;
+	if (binding & (PIPE_BIND_CONSTANT_BUFFER | PIPE_BIND_VERTEX_BUFFER | PIPE_BIND_INDEX_BUFFER)) {
+		bo = r600_bomgr_bo_create(radeon->bomgr, size, alignment, *radeon->cfence);
+		if (bo) {
+			return bo;
+		}
+	}
 
-	if (binding & (PIPE_BIND_CONSTANT_BUFFER | PIPE_BIND_VERTEX_BUFFER | PIPE_BIND_INDEX_BUFFER))
-		man = radeon->cman;
-	else
-		man = radeon->kman;
+	rbo = radeon_bo(radeon, 0, size, alignment);
+	if (rbo == NULL) {
+		return NULL;
+	}
+
+	bo = calloc(1, sizeof(struct r600_bo));
+	bo->size = size;
+	bo->alignment = alignment;
+	bo->bo = rbo;
+	if (binding & (PIPE_BIND_CONSTANT_BUFFER | PIPE_BIND_VERTEX_BUFFER | PIPE_BIND_INDEX_BUFFER)) {
+		r600_bomgr_bo_init(radeon->bomgr, bo);
+	}
 
 	/* Staging resources particpate in transfers and blits only
 	 * and are used for uploads and downloads from regular
 	 * resources.  We generate them internally for some transfers.
 	 */
 	if (usage == PIPE_USAGE_STAGING)
-                ws_bo->domains = RADEON_GEM_DOMAIN_CPU | RADEON_GEM_DOMAIN_GTT;
-        else
-                ws_bo->domains = (RADEON_GEM_DOMAIN_CPU |
-                                  RADEON_GEM_DOMAIN_GTT |
-                                  RADEON_GEM_DOMAIN_VRAM);
-
-
-	ws_bo->pb = man->create_buffer(man, size, &desc);
-	if (ws_bo->pb == NULL) {
-		free(ws_bo);
-		return NULL;
-	}
+		bo->domains = RADEON_GEM_DOMAIN_CPU | RADEON_GEM_DOMAIN_GTT;
+	else
+		bo->domains = (RADEON_GEM_DOMAIN_CPU |
+				RADEON_GEM_DOMAIN_GTT |
+				RADEON_GEM_DOMAIN_VRAM);
 
-	pipe_reference_init(&ws_bo->reference, 1);
-	return ws_bo;
+	pipe_reference_init(&bo->reference, 1);
+	return bo;
 }
 
 struct r600_bo *r600_bo_handle(struct radeon *radeon,
 			       unsigned handle, unsigned *array_mode)
 {
-	struct r600_bo *ws_bo = calloc(1, sizeof(struct r600_bo));
-	struct radeon_bo *bo;
+	struct r600_bo *bo = calloc(1, sizeof(struct r600_bo));
+	struct radeon_bo *rbo;
 
-	ws_bo->pb = radeon_bo_pb_create_buffer_from_handle(radeon->kman, handle);
-	if (!ws_bo->pb) {
-		free(ws_bo);
+	rbo = bo->bo = radeon_bo(radeon, handle, 0, 0);
+	if (rbo == NULL) {
+		free(bo);
 		return NULL;
 	}
-	bo = radeon_bo_pb_get_bo(ws_bo->pb);
-	ws_bo->size = bo->size;
-	ws_bo->domains = (RADEON_GEM_DOMAIN_CPU |
-			  RADEON_GEM_DOMAIN_GTT |
-			  RADEON_GEM_DOMAIN_VRAM);
+	bo->size = rbo->size;
+	bo->domains = (RADEON_GEM_DOMAIN_CPU |
+			RADEON_GEM_DOMAIN_GTT |
+			RADEON_GEM_DOMAIN_VRAM);
 
-	pipe_reference_init(&ws_bo->reference, 1);
+	pipe_reference_init(&bo->reference, 1);
 
-	radeon_bo_get_tiling_flags(radeon, bo, &ws_bo->tiling_flags,
-				   &ws_bo->kernel_pitch);
+	radeon_bo_get_tiling_flags(radeon, rbo, &bo->tiling_flags, &bo->kernel_pitch);
 	if (array_mode) {
-		if (ws_bo->tiling_flags) {
-			if (ws_bo->tiling_flags & RADEON_TILING_MICRO)
+		if (bo->tiling_flags) {
+			if (bo->tiling_flags & RADEON_TILING_MICRO)
 				*array_mode = V_0280A0_ARRAY_1D_TILED_THIN1;
-			if ((ws_bo->tiling_flags & (RADEON_TILING_MICRO | RADEON_TILING_MACRO)) ==
+			if ((bo->tiling_flags & (RADEON_TILING_MICRO | RADEON_TILING_MACRO)) ==
 			    (RADEON_TILING_MICRO | RADEON_TILING_MACRO))
 				*array_mode = V_0280A0_ARRAY_2D_TILED_THIN1;
 		} else {
 			*array_mode = 0;
 		}
 	}
-	return ws_bo;
+	return bo;
 }
 
 void *r600_bo_map(struct radeon *radeon, struct r600_bo *bo, unsigned usage, void *ctx)
 {
-	return pb_map(bo->pb, usage, ctx);
+	struct pipe_context *pctx = ctx;
+
+	if (usage & PB_USAGE_UNSYNCHRONIZED) {
+		radeon_bo_map(radeon, bo->bo);
+		return (uint8_t *) bo->bo->data + bo->offset;
+	}
+
+	if (p_atomic_read(&bo->bo->reference.count) > 1) {
+		if (usage & PB_USAGE_DONTBLOCK) {
+			return NULL;
+		}
+		if (ctx) {
+			pctx->flush(pctx, 0, NULL);
+		}
+	}
+
+	if (usage & PB_USAGE_DONTBLOCK) {
+		uint32_t domain;
+
+		if (radeon_bo_busy(radeon, bo->bo, &domain))
+			return NULL;
+		if (radeon_bo_map(radeon, bo->bo)) {
+			return NULL;
+		}
+		goto out;
+	}
+
+	radeon_bo_map(radeon, bo->bo);
+	if (radeon_bo_wait(radeon, bo->bo)) {
+		radeon_bo_unmap(radeon, bo->bo);
+		return NULL;
+	}
+
+out:
+	return (uint8_t *) bo->bo->data + bo->offset;
 }
 
 void r600_bo_unmap(struct radeon *radeon, struct r600_bo *bo)
 {
-	pb_unmap(bo->pb);
+	radeon_bo_unmap(radeon, bo->bo);
 }
 
-static void r600_bo_destroy(struct radeon *radeon, struct r600_bo *bo)
+void r600_bo_destroy(struct radeon *radeon, struct r600_bo *bo)
 {
-	if (bo->pb)
-		pb_reference(&bo->pb, NULL);
+	if (bo->manager_id) {
+		if (!r600_bomgr_bo_destroy(radeon->bomgr, bo)) {
+			/* destroy is delayed by buffer manager */
+			return;
+		}
+	}
+	radeon_bo_reference(radeon, &bo->bo, NULL);
 	free(bo);
 }
 
-void r600_bo_reference(struct radeon *radeon, struct r600_bo **dst,
-			    struct r600_bo *src)
+void r600_bo_reference(struct radeon *radeon, struct r600_bo **dst, struct r600_bo *src)
 {
 	struct r600_bo *old = *dst;
- 		
+
 	if (pipe_reference(&(*dst)->reference, &src->reference)) {
 		r600_bo_destroy(radeon, old);
 	}
 	*dst = src;
 }
 
-unsigned r600_bo_get_handle(struct r600_bo *pb_bo)
-{
-	struct radeon_bo *bo;
-
-	bo = radeon_bo_pb_get_bo(pb_bo->pb);
-	if (!bo)
-		return 0;
-
-	return bo->handle;
-}
-
-unsigned r600_bo_get_size(struct r600_bo *pb_bo)
-{
-	struct radeon_bo *bo;
-
-	bo = radeon_bo_pb_get_bo(pb_bo->pb);
-	if (!bo)
-		return 0;
-
-	return bo->size;
-}
-
-boolean r600_bo_get_winsys_handle(struct radeon *radeon, struct r600_bo *pb_bo,
+boolean r600_bo_get_winsys_handle(struct radeon *radeon, struct r600_bo *bo,
 				unsigned stride, struct winsys_handle *whandle)
 {
-	struct radeon_bo *bo;
-
-	bo = radeon_bo_pb_get_bo(pb_bo->pb);
-	if (!bo)
-		return FALSE;
-
 	whandle->stride = stride;
 	switch(whandle->type) {
 	case DRM_API_HANDLE_TYPE_KMS:
-		whandle->handle = r600_bo_get_handle(pb_bo);
+		whandle->handle = r600_bo_get_handle(bo);
 		break;
 	case DRM_API_HANDLE_TYPE_SHARED:
-		if (radeon_bo_get_name(radeon, bo, &whandle->handle))
+		if (radeon_bo_get_name(radeon, bo->bo, &whandle->handle))
 			return FALSE;
 		break;
 	default:
diff --git a/src/gallium/winsys/r600/drm/r600_bomgr.c b/src/gallium/winsys/r600/drm/r600_bomgr.c
new file mode 100644
index 0000000000..446ef0f9cf
--- /dev/null
+++ b/src/gallium/winsys/r600/drm/r600_bomgr.c
@@ -0,0 +1,161 @@
+/*
+ * Copyright 2010 VMWare.
+ * Copyright 2010 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to 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
+ * THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
+ * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
+ * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
+ * USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors:
+ *      Jose Fonseca <jrfonseca-at-vmware-dot-com>
+ *      Thomas Hellström <thomas-at-vmware-dot-com>
+ *      Jerome Glisse <jglisse@redhat.com>
+ */
+#include <util/u_memory.h>
+#include <util/u_double_list.h>
+#include <util/u_time.h>
+#include <pipebuffer/pb_bufmgr.h>
+#include "r600_priv.h"
+
+static void r600_bomgr_timeout_flush(struct r600_bomgr *mgr)
+{
+	struct r600_bo *bo, *tmp;
+	int64_t now;
+
+	now = os_time_get();
+	LIST_FOR_EACH_ENTRY_SAFE(bo, tmp, &mgr->delayed, list) {
+		if(!os_time_timeout(bo->start, bo->end, now))
+			break;
+
+		mgr->num_delayed--;
+		bo->manager_id = 0;
+		LIST_DEL(&bo->list);
+		r600_bo_destroy(mgr->radeon, bo);
+	}
+}
+
+static INLINE int r600_bo_is_compat(struct r600_bomgr *mgr,
+					struct r600_bo *bo,
+					unsigned size,
+					unsigned alignment,
+					unsigned cfence)
+{
+	if(bo->size < size) {
+		return 0;
+	}
+
+	/* be lenient with size */
+	if(bo->size >= 2*size) {
+		return 0;
+	}
+
+	if(!pb_check_alignment(alignment, bo->alignment)) {
+		return 0;
+	}
+
+	if (!fence_is_after(cfence, bo->fence)) {
+		return 0;
+	}
+
+	return 1;
+}
+
+struct r600_bo *r600_bomgr_bo_create(struct r600_bomgr *mgr,
+					unsigned size,
+					unsigned alignment,
+					unsigned cfence)
+{
+	struct r600_bo *bo, *tmp;
+	int64_t now;
+
+
+	pipe_mutex_lock(mgr->mutex);
+
+	now = os_time_get();
+	LIST_FOR_EACH_ENTRY_SAFE(bo, tmp, &mgr->delayed, list) {
+		if(r600_bo_is_compat(mgr, bo, size, alignment, cfence)) {
+			LIST_DEL(&bo->list);
+			--mgr->num_delayed;
+			r600_bomgr_timeout_flush(mgr);
+			pipe_mutex_unlock(mgr->mutex);
+			LIST_INITHEAD(&bo->list);
+			pipe_reference_init(&bo->reference, 1);
+			return bo;
+		}
+
+		if(os_time_timeout(bo->start, bo->end, now)) {
+			mgr->num_delayed--;
+			bo->manager_id = 0;
+			LIST_DEL(&bo->list);
+			r600_bo_destroy(mgr->radeon, bo);
+		}
+	}
+
+	pipe_mutex_unlock(mgr->mutex);
+	return NULL;
+}
+
+void r600_bomgr_bo_init(struct r600_bomgr *mgr, struct r600_bo *bo)
+{
+	LIST_INITHEAD(&bo->list);
+	bo->manager_id = 1;
+}
+
+bool r600_bomgr_bo_destroy(struct r600_bomgr *mgr, struct r600_bo *bo)
+{
+	bo->start = os_time_get();
+	bo->end = bo->start + mgr->usecs;
+	pipe_mutex_lock(mgr->mutex);
+	LIST_ADDTAIL(&bo->list, &mgr->delayed);
+	++mgr->num_delayed;
+	pipe_mutex_unlock(mgr->mutex);
+	return FALSE;
+}
+
+void r600_bomgr_destroy(struct r600_bomgr *mgr)
+{
+	struct r600_bo *bo, *tmp;
+
+	pipe_mutex_lock(mgr->mutex);
+	LIST_FOR_EACH_ENTRY_SAFE(bo, tmp, &mgr->delayed, list) {
+		mgr->num_delayed--;
+		bo->manager_id = 0;
+		LIST_DEL(&bo->list);
+		r600_bo_destroy(mgr->radeon, bo);
+	}
+	pipe_mutex_unlock(mgr->mutex);
+
+	FREE(mgr);
+}
+
+struct r600_bomgr *r600_bomgr_create(struct radeon *radeon, unsigned usecs)
+{
+	struct r600_bomgr *mgr;
+
+	mgr = CALLOC_STRUCT(r600_bomgr);
+	if (mgr == NULL)
+		return NULL;
+
+	mgr->radeon = radeon;
+	mgr->usecs = usecs;
+	LIST_INITHEAD(&mgr->delayed);
+	mgr->num_delayed = 0;
+	pipe_mutex_init(mgr->mutex);
+
+	return mgr;
+}
diff --git a/src/gallium/winsys/r600/drm/r600_drm.c b/src/gallium/winsys/r600/drm/r600_drm.c
index 60c2f51fac..3cbbf91878 100644
--- a/src/gallium/winsys/r600/drm/r600_drm.c
+++ b/src/gallium/winsys/r600/drm/r600_drm.c
@@ -30,7 +30,6 @@
 #include <sys/ioctl.h>
 #include "util/u_inlines.h"
 #include "util/u_debug.h"
-#include <pipebuffer/pb_bufmgr.h>
 #include "r600.h"
 #include "r600_priv.h"
 #include "r600_drm_public.h"
@@ -40,6 +39,9 @@
 #ifndef RADEON_INFO_TILING_CONFIG
 #define RADEON_INFO_TILING_CONFIG 0x6
 #endif
+
+static struct radeon *radeon_new(int fd, unsigned device);
+
 static int radeon_get_device(struct radeon *radeon)
 {
 	struct drm_radeon_info info;
@@ -108,7 +110,7 @@ static int radeon_drm_get_tiling(struct radeon *radeon)
 	return 0;
 }
 
-struct radeon *radeon_new(int fd, unsigned device)
+static struct radeon *radeon_new(int fd, unsigned device)
 {
 	struct radeon *radeon;
 	int r;
@@ -150,6 +152,7 @@ struct radeon *radeon_new(int fd, unsigned device)
 	case CHIP_JUNIPER:
 	case CHIP_CYPRESS:
 	case CHIP_HEMLOCK:
+	case CHIP_PALM:
 		break;
 	case CHIP_R100:
 	case CHIP_RV100:
@@ -195,19 +198,26 @@ struct radeon *radeon_new(int fd, unsigned device)
 	case CHIP_RS780:
 	case CHIP_RS880:
 		radeon->chip_class = R600;
+		/* set default group bytes, overridden by tiling info ioctl */
+		radeon->tiling_info.group_bytes = 256;
 		break;
 	case CHIP_RV770:
 	case CHIP_RV730:
 	case CHIP_RV710:
 	case CHIP_RV740:
 		radeon->chip_class = R700;
+		/* set default group bytes, overridden by tiling info ioctl */
+		radeon->tiling_info.group_bytes = 256;
 		break;
 	case CHIP_CEDAR:
 	case CHIP_REDWOOD:
 	case CHIP_JUNIPER:
 	case CHIP_CYPRESS:
 	case CHIP_HEMLOCK:
+	case CHIP_PALM:
 		radeon->chip_class = EVERGREEN;
+		/* set default group bytes, overridden by tiling info ioctl */
+		radeon->tiling_info.group_bytes = 512;
 		break;
 	default:
 		fprintf(stderr, "%s unknown or unsupported chipset 0x%04X\n",
@@ -219,12 +229,10 @@ struct radeon *radeon_new(int fd, unsigned device)
 		if (radeon_drm_get_tiling(radeon))
 			return NULL;
 	}
-	radeon->kman = radeon_bo_pbmgr_create(radeon);
-	if (!radeon->kman)
-		return NULL;
-	radeon->cman = pb_cache_manager_create(radeon->kman, 100000);
-	if (!radeon->cman)
+	radeon->bomgr = r600_bomgr_create(radeon, 1000000);
+	if (radeon->bomgr == NULL) {
 		return NULL;
+	}
 	return radeon;
 }
 
@@ -241,14 +249,11 @@ struct radeon *radeon_decref(struct radeon *radeon)
 		return NULL;
 	}
 
-        if (radeon->cman)
-           radeon->cman->destroy(radeon->cman);
-
-        if (radeon->kman)
-           radeon->kman->destroy(radeon->kman);
+	if (radeon->bomgr)
+		r600_bomgr_destroy(radeon->bomgr);
 
-        if (radeon->fd >= 0)
-           drmClose(radeon->fd);
+	if (radeon->fd >= 0)
+		drmClose(radeon->fd);
 
 	free(radeon);
 	return NULL;
diff --git a/src/gallium/winsys/r600/drm/r600_hw_context.c b/src/gallium/winsys/r600/drm/r600_hw_context.c
index 37e5baf8de..d01ec3ee9b 100644
--- a/src/gallium/winsys/r600/drm/r600_hw_context.c
+++ b/src/gallium/winsys/r600/drm/r600_hw_context.c
@@ -28,16 +28,15 @@
 #include <string.h>
 #include <stdlib.h>
 #include <assert.h>
+#include <pipe/p_compiler.h>
+#include <util/u_inlines.h>
+#include <util/u_memory.h>
+#include <pipebuffer/pb_bufmgr.h>
 #include "xf86drm.h"
-#include "r600.h"
-#include "r600d.h"
 #include "radeon_drm.h"
-#include "bof.h"
-#include "pipe/p_compiler.h"
-#include "util/u_inlines.h"
-#include "util/u_memory.h"
-#include <pipebuffer/pb_bufmgr.h>
 #include "r600_priv.h"
+#include "bof.h"
+#include "r600d.h"
 
 #define GROUP_FORCE_NEW_BLOCK	0
 
@@ -50,6 +49,7 @@ int r600_context_init_fence(struct r600_context *ctx)
 	}
 	ctx->cfence = r600_bo_map(ctx->radeon, ctx->fence_bo, PB_USAGE_UNSYNCHRONIZED, NULL);
 	*ctx->cfence = 0;
+	ctx->radeon->cfence = ctx->cfence;
 	LIST_INITHEAD(&ctx->fenced_bo);
 	return 0;
 }
@@ -593,6 +593,17 @@ static int r600_loop_const_init(struct r600_context *ctx, u32 offset)
 	return r600_context_add_block(ctx, r600_loop_consts, nreg);
 }
 
+static void r600_context_clear_fenced_bo(struct r600_context *ctx)
+{
+	struct radeon_bo *bo, *tmp;
+
+	LIST_FOR_EACH_ENTRY_SAFE(bo, tmp, &ctx->fenced_bo, fencedlist) {
+		LIST_DELINIT(&bo->fencedlist);
+		bo->fence = 0;
+		bo->ctx = NULL;
+	}
+}
+
 /* initialize */
 void r600_context_fini(struct r600_context *ctx)
 {
@@ -607,6 +618,9 @@ void r600_context_fini(struct r600_context *ctx)
 					range = &ctx->range[CTX_RANGE_ID(ctx, offset)];
 					range->blocks[CTX_BLOCK_ID(ctx, offset)] = NULL;
 				}
+				for (int k = 1; k <= block->nbo; k++) {
+					r600_bo_reference(ctx->radeon, &block->reloc[k].bo, NULL);
+				}
 				free(block);
 			}
 		}
@@ -616,6 +630,8 @@ void r600_context_fini(struct r600_context *ctx)
 	free(ctx->reloc);
 	free(ctx->bo);
 	free(ctx->pm4);
+
+	r600_context_clear_fenced_bo(ctx);
 	if (ctx->fence_bo) {
 		r600_bo_reference(ctx->radeon, &ctx->fence_bo, NULL);
 	}
@@ -693,6 +709,12 @@ int r600_context_init(struct r600_context *ctx, struct radeon *radeon)
 		if (r)
 			goto out_err;
 	}
+	/* FS RESOURCE */
+	for (int j = 0, offset = 0x2300; j < 16; j++, offset += 0x1C) {
+		r = r600_state_resource_init(ctx, offset);
+		if (r)
+			goto out_err;
+	}
 
 	/* PS loop const */
 	r600_loop_const_init(ctx, 0);
@@ -792,6 +814,7 @@ void r600_context_bo_reloc(struct r600_context *ctx, u32 *pm4, struct r600_bo *r
 	ctx->reloc[ctx->creloc].write_domain = rbo->domains & (RADEON_GEM_DOMAIN_GTT | RADEON_GEM_DOMAIN_VRAM);
 	ctx->reloc[ctx->creloc].flags = 0;
 	radeon_bo_reference(ctx->radeon, &ctx->bo[ctx->creloc], bo);
+	rbo->fence = ctx->fence;
 	ctx->creloc++;
 	/* set PKT3 to point to proper reloc */
 	*pm4 = bo->reloc_id;
@@ -814,6 +837,7 @@ void r600_context_pipe_state_set(struct r600_context *ctx, struct r600_pipe_stat
 			/* find relocation */
 			id = block->pm4_bo_index[id];
 			r600_bo_reference(ctx->radeon, &block->reloc[id].bo, state->regs[i].bo);
+			state->regs[i].bo->fence = ctx->fence;
 		}
 		if (!(block->status & R600_BLOCK_STATUS_DIRTY)) {
 			block->status |= R600_BLOCK_STATUS_ENABLED;
@@ -853,10 +877,13 @@ static inline void r600_context_pipe_state_set_resource(struct r600_context *ctx
 		 */
 		r600_bo_reference(ctx->radeon, &block->reloc[1].bo, state->regs[0].bo);
 		r600_bo_reference(ctx->radeon, &block->reloc[2].bo, state->regs[0].bo);
+		state->regs[0].bo->fence = ctx->fence;
 	} else {
 		/* TEXTURE RESOURCE */
 		r600_bo_reference(ctx->radeon, &block->reloc[1].bo, state->regs[2].bo);
 		r600_bo_reference(ctx->radeon, &block->reloc[2].bo, state->regs[3].bo);
+		state->regs[2].bo->fence = ctx->fence;
+		state->regs[3].bo->fence = ctx->fence;
 	}
 	if (!(block->status & R600_BLOCK_STATUS_DIRTY)) {
 		block->status |= R600_BLOCK_STATUS_ENABLED;
@@ -880,6 +907,13 @@ void r600_context_pipe_state_set_vs_resource(struct r600_context *ctx, struct r6
 	r600_context_pipe_state_set_resource(ctx, state, offset);
 }
 
+void r600_context_pipe_state_set_fs_resource(struct r600_context *ctx, struct r600_pipe_state *state, unsigned rid)
+{
+	unsigned offset = R_038000_SQ_TEX_RESOURCE_WORD0_0 + 0x2300 + 0x1C * rid;
+
+	r600_context_pipe_state_set_resource(ctx, state, offset);
+}
+
 static inline void r600_context_pipe_state_set_sampler(struct r600_context *ctx, struct r600_pipe_state *state, unsigned offset)
 {
 	struct r600_range *range;
@@ -1049,7 +1083,7 @@ void r600_context_draw(struct r600_context *ctx, const struct r600_draw *draw)
 		ctx->pm4[ctx->pm4_cdwords++] = draw->vgt_draw_initiator;
 	}
 	ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_EVENT_WRITE, 0);
-	ctx->pm4[ctx->pm4_cdwords++] = EVENT_TYPE_CACHE_FLUSH_AND_INV_EVENT;
+	ctx->pm4[ctx->pm4_cdwords++] = EVENT_TYPE(EVENT_TYPE_CACHE_FLUSH_AND_INV_EVENT) | EVENT_INDEX(0);
 
 	/* flush color buffer */
 	for (int i = 0; i < 8; i++) {
@@ -1082,11 +1116,11 @@ void r600_context_flush(struct r600_context *ctx)
 	/* suspend queries */
 	r600_context_queries_suspend(ctx);
 
-	radeon_bo_pbmgr_flush_maps(ctx->radeon->kman);
-
 	/* emit fence */
+	ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_EVENT_WRITE, 0);
+	ctx->pm4[ctx->pm4_cdwords++] = EVENT_TYPE(EVENT_TYPE_PS_PARTIAL_FLUSH) | EVENT_INDEX(4);
 	ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_EVENT_WRITE_EOP, 4);
-	ctx->pm4[ctx->pm4_cdwords++] = EVENT_TYPE_CACHE_FLUSH_AND_INV_TS_EVENT | (5 << 8);
+	ctx->pm4[ctx->pm4_cdwords++] = EVENT_TYPE(EVENT_TYPE_CACHE_FLUSH_AND_INV_TS_EVENT) | EVENT_INDEX(5);
 	ctx->pm4[ctx->pm4_cdwords++] = 0;
 	ctx->pm4[ctx->pm4_cdwords++] = (1 << 29) | (0 << 24);
 	ctx->pm4[ctx->pm4_cdwords++] = ctx->fence;
@@ -1266,7 +1300,7 @@ void r600_query_begin(struct r600_context *ctx, struct r600_query *query)
 
 	/* emit begin query */
 	ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_EVENT_WRITE, 2);
-	ctx->pm4[ctx->pm4_cdwords++] = EVENT_TYPE_ZPASS_DONE;
+	ctx->pm4[ctx->pm4_cdwords++] = EVENT_TYPE(EVENT_TYPE_ZPASS_DONE) | EVENT_INDEX(1);
 	ctx->pm4[ctx->pm4_cdwords++] = query->num_results + r600_bo_offset(query->buffer);
 	ctx->pm4[ctx->pm4_cdwords++] = 0;
 	ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_NOP, 0);
@@ -1282,7 +1316,7 @@ void r600_query_end(struct r600_context *ctx, struct r600_query *query)
 {
 	/* emit begin query */
 	ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_EVENT_WRITE, 2);
-	ctx->pm4[ctx->pm4_cdwords++] = EVENT_TYPE_ZPASS_DONE;
+	ctx->pm4[ctx->pm4_cdwords++] = EVENT_TYPE(EVENT_TYPE_ZPASS_DONE) | EVENT_INDEX(1);
 	ctx->pm4[ctx->pm4_cdwords++] = query->num_results + 8 + r600_bo_offset(query->buffer);
 	ctx->pm4[ctx->pm4_cdwords++] = 0;
 	ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_NOP, 0);
diff --git a/src/gallium/winsys/r600/drm/r600_priv.h b/src/gallium/winsys/r600/drm/r600_priv.h
index 9fd77b71c7..056d0255be 100644
--- a/src/gallium/winsys/r600/drm/r600_priv.h
+++ b/src/gallium/winsys/r600/drm/r600_priv.h
@@ -30,24 +30,24 @@
 #include <stdint.h>
 #include <stdlib.h>
 #include <assert.h>
-#include <pipebuffer/pb_bufmgr.h>
-#include "util/u_double_list.h"
+#include <util/u_double_list.h>
+#include <util/u_inlines.h>
+#include <os/os_thread.h>
 #include "r600.h"
 
+struct r600_bomgr;
+
 struct radeon {
 	int				fd;
 	int				refcount;
 	unsigned			device;
 	unsigned			family;
 	enum chip_class			chip_class;
-	struct pb_manager *kman; /* kernel bo manager */
-	struct pb_manager *cman; /* cached bo manager */
-	struct r600_tiling_info tiling_info;
+	struct r600_tiling_info		tiling_info;
+	struct r600_bomgr		*bomgr;
+	unsigned			*cfence;
 };
 
-struct radeon *r600_new(int fd, unsigned device);
-void r600_delete(struct radeon *r600);
-
 struct r600_reg {
 	unsigned			opcode;
 	unsigned			offset_base;
@@ -75,28 +75,49 @@ struct radeon_bo {
 
 struct r600_bo {
 	struct pipe_reference		reference;
-	struct pb_buffer		*pb;
 	unsigned			size;
 	unsigned			tiling_flags;
-	unsigned                        kernel_pitch;
+	unsigned			kernel_pitch;
 	unsigned			domains;
+	struct radeon_bo		*bo;
+	unsigned			fence;
+	/* manager data */
+	struct list_head		list;
+	unsigned			manager_id;
+	unsigned			alignment;
+	unsigned			offset;
+	int64_t				start;
+	int64_t				end;
 };
 
+struct r600_bomgr {
+	struct radeon			*radeon;
+	unsigned			usecs;
+	pipe_mutex			mutex;
+	struct list_head		delayed;
+	unsigned			num_delayed;
+};
 
-/* radeon_pciid.c */
-unsigned radeon_family_from_device(unsigned device);
+/*
+ * r600_drm.c
+ */
+struct radeon *r600_new(int fd, unsigned device);
+void r600_delete(struct radeon *r600);
 
-/* r600_drm.c */
-struct radeon *radeon_decref(struct radeon *radeon);
+/*
+ * radeon_pciid.c
+ */
+unsigned radeon_family_from_device(unsigned device);
 
-/* radeon_bo.c */
+/*
+ * radeon_bo.c
+ */
 struct radeon_bo *radeon_bo(struct radeon *radeon, unsigned handle,
 			    unsigned size, unsigned alignment);
 void radeon_bo_reference(struct radeon *radeon, struct radeon_bo **dst,
 			 struct radeon_bo *src);
 int radeon_bo_wait(struct radeon *radeon, struct radeon_bo *bo);
 int radeon_bo_busy(struct radeon *radeon, struct radeon_bo *bo, uint32_t *domain);
-void radeon_bo_pbmgr_flush_maps(struct pb_manager *_mgr);
 int radeon_bo_fencelist(struct radeon *radeon, struct radeon_bo **bolist, uint32_t num_bo);
 int radeon_bo_get_tiling_flags(struct radeon *radeon,
 			       struct radeon_bo *bo,
@@ -106,13 +127,9 @@ int radeon_bo_get_name(struct radeon *radeon,
 		       struct radeon_bo *bo,
 		       uint32_t *name);
 
-/* radeon_bo_pb.c */
-struct radeon_bo *radeon_bo_pb_get_bo(struct pb_buffer *_buf);
-struct pb_manager *radeon_bo_pbmgr_create(struct radeon *radeon);
-struct pb_buffer *radeon_bo_pb_create_buffer_from_handle(struct pb_manager *_mgr,
-							 uint32_t handle);
-
-/* r600_hw_context.c */
+/*
+ * r600_hw_context.c
+ */
 int r600_context_init_fence(struct r600_context *ctx);
 void r600_context_bo_reloc(struct r600_context *ctx, u32 *pm4, struct r600_bo *rbo);
 void r600_context_bo_flush(struct r600_context *ctx, unsigned flush_flags,
@@ -120,14 +137,27 @@ void r600_context_bo_flush(struct r600_context *ctx, unsigned flush_flags,
 struct r600_bo *r600_context_reg_bo(struct r600_context *ctx, unsigned offset);
 int r600_context_add_block(struct r600_context *ctx, const struct r600_reg *reg, unsigned nreg);
 
-/* r600_bo.c */
-unsigned r600_bo_get_handle(struct r600_bo *bo);
-unsigned r600_bo_get_size(struct r600_bo *bo);
-static INLINE struct radeon_bo *r600_bo_get_bo(struct r600_bo *bo)
-{
-	return radeon_bo_pb_get_bo(bo->pb);
-}
+/*
+ * r600_bo.c
+ */
+void r600_bo_destroy(struct radeon *radeon, struct r600_bo *bo);
 
+/*
+ * r600_bomgr.c
+ */
+struct r600_bomgr *r600_bomgr_create(struct radeon *radeon, unsigned usecs);
+void r600_bomgr_destroy(struct r600_bomgr *mgr);
+bool r600_bomgr_bo_destroy(struct r600_bomgr *mgr, struct r600_bo *bo);
+void r600_bomgr_bo_init(struct r600_bomgr *mgr, struct r600_bo *bo);
+struct r600_bo *r600_bomgr_bo_create(struct r600_bomgr *mgr,
+					unsigned size,
+					unsigned alignment,
+					unsigned cfence);
+
+
+/*
+ * helpers
+ */
 #define CTX_RANGE_ID(ctx, offset) (((offset) >> (ctx)->hash_shift) & 255)
 #define CTX_BLOCK_ID(ctx, offset) ((offset) & ((1 << (ctx)->hash_shift) - 1))
 
@@ -175,6 +205,9 @@ static inline void r600_context_block_emit_dirty(struct r600_context *ctx, struc
 	LIST_DELINIT(&block->list);
 }
 
+/*
+ * radeon_bo.c
+ */
 static inline int radeon_bo_map(struct radeon *radeon, struct radeon_bo *bo)
 {
 	bo->map_count++;
@@ -187,4 +220,35 @@ static inline void radeon_bo_unmap(struct radeon *radeon, struct radeon_bo *bo)
 	assert(bo->map_count >= 0);
 }
 
+/*
+ * r600_bo
+ */
+static inline struct radeon_bo *r600_bo_get_bo(struct r600_bo *bo)
+{
+	return bo->bo;
+}
+
+static unsigned inline r600_bo_get_handle(struct r600_bo *bo)
+{
+	return bo->bo->handle;
+}
+
+static unsigned inline r600_bo_get_size(struct r600_bo *bo)
+{
+	return bo->size;
+}
+
+/*
+ * fence
+ */
+static inline bool fence_is_after(unsigned fence, unsigned ofence)
+{
+	/* handle wrap around */
+	if (fence < 0x80000000 && ofence > 0x80000000)
+		return TRUE;
+	if (fence > ofence)
+		return TRUE;
+	return FALSE;
+}
+
 #endif
diff --git a/src/gallium/winsys/r600/drm/r600d.h b/src/gallium/winsys/r600/drm/r600d.h
index 5ca7456e90..1c1ac76fe6 100644
--- a/src/gallium/winsys/r600/drm/r600d.h
+++ b/src/gallium/winsys/r600/drm/r600d.h
@@ -91,9 +91,19 @@
 #define PKT3_SET_CTL_CONST                     0x6F
 #define PKT3_SURFACE_BASE_UPDATE               0x73
 
+#define EVENT_TYPE_PS_PARTIAL_FLUSH            0x10
 #define EVENT_TYPE_CACHE_FLUSH_AND_INV_TS_EVENT 0x14
 #define EVENT_TYPE_ZPASS_DONE                  0x15
 #define EVENT_TYPE_CACHE_FLUSH_AND_INV_EVENT   0x16
+#define		EVENT_TYPE(x)                           ((x) << 0)
+#define		EVENT_INDEX(x)                          ((x) << 8)
+                /* 0 - any non-TS event
+		 * 1 - ZPASS_DONE
+		 * 2 - SAMPLE_PIPELINESTAT
+		 * 3 - SAMPLE_STREAMOUTSTAT*
+		 * 4 - *S_PARTIAL_FLUSH
+		 * 5 - TS events
+		 */
 
 #define PKT_TYPE_S(x)                   (((x) & 0x3) << 30)
 #define PKT_TYPE_G(x)                   (((x) >> 30) & 0x3)
diff --git a/src/gallium/winsys/r600/drm/radeon_bo_pb.c b/src/gallium/winsys/r600/drm/radeon_bo_pb.c
deleted file mode 100644
index 312552f075..0000000000
--- a/src/gallium/winsys/r600/drm/radeon_bo_pb.c
+++ /dev/null
@@ -1,287 +0,0 @@
-/*
- * Copyright 2010 Dave Airlie
- *
- * Permission is hereby granted, free of charge, to 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
- * THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
- * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
- * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
- * USE OR OTHER DEALINGS IN THE SOFTWARE.
- *
- * Authors:
- *      Dave Airlie
- */
-#include <util/u_inlines.h>
-#include <util/u_memory.h>
-#include <util/u_double_list.h>
-#include <pipebuffer/pb_buffer.h>
-#include <pipebuffer/pb_bufmgr.h>
-#include "r600_priv.h"
-
-struct radeon_bo_pb {
-	struct pb_buffer b;
-	struct radeon_bo *bo;
-
-	struct radeon_bo_pbmgr *mgr;
-	struct list_head maplist;
-};
-
-extern const struct pb_vtbl radeon_bo_pb_vtbl;
-
-static INLINE struct radeon_bo_pb *radeon_bo_pb(struct pb_buffer *buf)
-{
-	assert(buf);
-	assert(buf->vtbl == &radeon_bo_pb_vtbl);
-	return (struct radeon_bo_pb *)buf;
-}
-
-struct radeon_bo_pbmgr {
-	struct pb_manager b;
-	struct radeon *radeon;
-	struct list_head buffer_map_list;
-};
-
-static INLINE struct radeon_bo_pbmgr *radeon_bo_pbmgr(struct pb_manager *mgr)
-{
-	assert(mgr);
-	return (struct radeon_bo_pbmgr *)mgr;
-}
-
-static void radeon_bo_pb_destroy(struct pb_buffer *_buf)
-{
-	struct radeon_bo_pb *buf = radeon_bo_pb(_buf);
-
-	/* If this buffer is on the list of buffers to unmap,
-	 * do the unmapping now.
-	 */
-	if (!LIST_IS_EMPTY(&buf->maplist))
-		radeon_bo_unmap(buf->mgr->radeon, buf->bo);
-
-	LIST_DEL(&buf->maplist);
-	radeon_bo_reference(buf->mgr->radeon, &buf->bo, NULL);
-	FREE(buf);
-}
-
-static void *
-radeon_bo_pb_map_internal(struct pb_buffer *_buf,
-			  unsigned flags, void *ctx)
-{
-	struct radeon_bo_pb *buf = radeon_bo_pb(_buf);
-	struct pipe_context *pctx = ctx;
-
-	if (flags & PB_USAGE_UNSYNCHRONIZED) {
-		if (radeon_bo_map(buf->mgr->radeon, buf->bo)) {
-			return NULL;
-		}
-		LIST_DELINIT(&buf->maplist);
-		return buf->bo->data;
-	}
-
-	if (p_atomic_read(&buf->bo->reference.count) > 1) {
-		if (flags & PB_USAGE_DONTBLOCK) {
-			return NULL;
-		}
-		if (ctx) {
-			pctx->flush(pctx, 0, NULL);
-		}
-	}
-
-	if (flags & PB_USAGE_DONTBLOCK) {
-		uint32_t domain;
-		if (radeon_bo_busy(buf->mgr->radeon, buf->bo, &domain))
-			return NULL;
-		if (radeon_bo_map(buf->mgr->radeon, buf->bo)) {
-			return NULL;
-		}
-		goto out;
-	}
-
-	if (radeon_bo_map(buf->mgr->radeon, buf->bo)) {
-		return NULL;
-	}
-	if (radeon_bo_wait(buf->mgr->radeon, buf->bo)) {
-		radeon_bo_unmap(buf->mgr->radeon, buf->bo);
-		return NULL;
-	}
-out:
-	LIST_DELINIT(&buf->maplist);
-	return buf->bo->data;
-}
-
-static void radeon_bo_pb_unmap_internal(struct pb_buffer *_buf)
-{
-	struct radeon_bo_pb *buf = radeon_bo_pb(_buf);
-	LIST_ADDTAIL(&buf->maplist, &buf->mgr->buffer_map_list);
-}
-
-static void
-radeon_bo_pb_get_base_buffer(struct pb_buffer *buf,
-			     struct pb_buffer **base_buf,
-			     unsigned *offset)
-{
-	*base_buf = buf;
-	*offset = 0;
-}
-
-static enum pipe_error
-radeon_bo_pb_validate(struct pb_buffer *_buf, 
-		      struct pb_validate *vl,
-		      unsigned flags)
-{
-	/* Always pinned */
-	return PIPE_OK;
-}
-
-static void
-radeon_bo_pb_fence(struct pb_buffer *buf,
-		   struct pipe_fence_handle *fence)
-{
-}
-
-const struct pb_vtbl radeon_bo_pb_vtbl = {
-    radeon_bo_pb_destroy,
-    radeon_bo_pb_map_internal,
-    radeon_bo_pb_unmap_internal,
-    radeon_bo_pb_validate,
-    radeon_bo_pb_fence,
-    radeon_bo_pb_get_base_buffer,
-};
-
-struct pb_buffer *
-radeon_bo_pb_create_buffer_from_handle(struct pb_manager *_mgr,
-				       uint32_t handle)
-{
-	struct radeon_bo_pbmgr *mgr = radeon_bo_pbmgr(_mgr);
-	struct radeon *radeon = mgr->radeon;
-	struct radeon_bo_pb *bo;
-	struct radeon_bo *hw_bo;
-
-	hw_bo = radeon_bo(radeon, handle, 0, 0);
-	if (hw_bo == NULL)
-		return NULL;
-
-	bo = CALLOC_STRUCT(radeon_bo_pb);
-	if (!bo) {
-		radeon_bo_reference(radeon, &hw_bo, NULL);
-		return NULL;
-	}
-
-	LIST_INITHEAD(&bo->maplist);
-	pipe_reference_init(&bo->b.base.reference, 1);
-	bo->b.base.alignment = 0;
-	bo->b.base.usage = PB_USAGE_GPU_WRITE | PB_USAGE_GPU_READ;
-	bo->b.base.size = hw_bo->size;
-	bo->b.vtbl = &radeon_bo_pb_vtbl;
-	bo->mgr = mgr;
-
-	bo->bo = hw_bo;
-
-	return &bo->b;
-}
-
-static struct pb_buffer *
-radeon_bo_pb_create_buffer(struct pb_manager *_mgr,
-			   pb_size size,
-			   const struct pb_desc *desc)
-{
-	struct radeon_bo_pbmgr *mgr = radeon_bo_pbmgr(_mgr);
-	struct radeon *radeon = mgr->radeon;
-	struct radeon_bo_pb *bo;
-
-	bo = CALLOC_STRUCT(radeon_bo_pb);
-	if (!bo)
-		goto error1;
-
-	pipe_reference_init(&bo->b.base.reference, 1);
-	bo->b.base.alignment = desc->alignment;
-	bo->b.base.usage = desc->usage;
-	bo->b.base.size = size;
-	bo->b.vtbl = &radeon_bo_pb_vtbl;
-	bo->mgr = mgr;
-
-	LIST_INITHEAD(&bo->maplist);
-
-	bo->bo = radeon_bo(radeon, 0, size, desc->alignment);
-	if (bo->bo == NULL)
-		goto error2;
-	return &bo->b;
-
-error2:
-	FREE(bo);
-error1:
-	return NULL;
-}
-
-static void
-radeon_bo_pbmgr_flush(struct pb_manager *mgr)
-{
-    /* NOP */
-}
-
-static void
-radeon_bo_pbmgr_destroy(struct pb_manager *_mgr)
-{
-	struct radeon_bo_pbmgr *mgr = radeon_bo_pbmgr(_mgr);
-	FREE(mgr);
-}
-
-struct pb_manager *radeon_bo_pbmgr_create(struct radeon *radeon)
-{
-	struct radeon_bo_pbmgr *mgr;
-
-	mgr = CALLOC_STRUCT(radeon_bo_pbmgr);
-	if (!mgr)
-		return NULL;
-
-	mgr->b.destroy = radeon_bo_pbmgr_destroy;
-	mgr->b.create_buffer = radeon_bo_pb_create_buffer;
-	mgr->b.flush = radeon_bo_pbmgr_flush;
-
-	mgr->radeon = radeon;
-	LIST_INITHEAD(&mgr->buffer_map_list);
-	return &mgr->b;
-}
-
-void radeon_bo_pbmgr_flush_maps(struct pb_manager *_mgr)
-{
-	struct radeon_bo_pbmgr *mgr = radeon_bo_pbmgr(_mgr);
-	struct radeon_bo_pb *rpb = NULL;
-	struct radeon_bo_pb *t_rpb;
-
-	LIST_FOR_EACH_ENTRY_SAFE(rpb, t_rpb, &mgr->buffer_map_list, maplist) {
-		radeon_bo_unmap(mgr->radeon, rpb->bo);
-		LIST_DELINIT(&rpb->maplist);
-	}
-
-	LIST_INITHEAD(&mgr->buffer_map_list);
-}
-
-struct radeon_bo *radeon_bo_pb_get_bo(struct pb_buffer *_buf)
-{
-	struct radeon_bo_pb *buf;
-	if (_buf->vtbl == &radeon_bo_pb_vtbl) {
-		buf = radeon_bo_pb(_buf);
-		return buf->bo;
-	} else {
-		struct pb_buffer *base_buf;
-		pb_size offset;
-		pb_get_base_buffer(_buf, &base_buf, &offset);
-		if (base_buf->vtbl == &radeon_bo_pb_vtbl) {
-			buf = radeon_bo_pb(base_buf);
-			return buf->bo;
-		}
-	}
-	return NULL;
-}
diff --git a/src/gallium/winsys/r600/drm/radeon_pciid.c b/src/gallium/winsys/r600/drm/radeon_pciid.c
index 08cc1c41e3..92560a488a 100644
--- a/src/gallium/winsys/r600/drm/radeon_pciid.c
+++ b/src/gallium/winsys/r600/drm/radeon_pciid.c
@@ -24,7 +24,7 @@
  *      Jerome Glisse
  */
 #include <stdlib.h>
-#include "r600.h"
+#include "r600_priv.h"
 
 struct pci_id {
 	unsigned	vendor;
@@ -441,6 +441,10 @@ struct pci_id radeon_pci_id[] = {
 	{0x1002, 0x9713, CHIP_RS880},
 	{0x1002, 0x9714, CHIP_RS880},
 	{0x1002, 0x9715, CHIP_RS880},
+	{0x1002, 0x9802, CHIP_PALM},
+	{0x1002, 0x9803, CHIP_PALM},
+	{0x1002, 0x9804, CHIP_PALM},
+	{0x1002, 0x9805, CHIP_PALM},
 	{0, 0},
 };
 
@@ -456,73 +460,3 @@ unsigned radeon_family_from_device(unsigned device)
 	}
 	return CHIP_UNKNOWN;
 }
-
-int radeon_is_family_compatible(unsigned family1, unsigned family2)
-{
-	switch (family1) {
-	case CHIP_R600:
-	case CHIP_RV610:
-	case CHIP_RV630:
-	case CHIP_RV670:
-	case CHIP_RV620:
-	case CHIP_RV635:
-	case CHIP_RS780:
-	case CHIP_RS880:
-	case CHIP_RV770:
-	case CHIP_RV730:
-	case CHIP_RV710:
-	case CHIP_RV740:
-		switch (family2) {
-		case CHIP_R600:
-		case CHIP_RV610:
-		case CHIP_RV630:
-		case CHIP_RV670:
-		case CHIP_RV620:
-		case CHIP_RV635:
-		case CHIP_RS780:
-		case CHIP_RS880:
-		case CHIP_RV770:
-		case CHIP_RV730:
-		case CHIP_RV710:
-		case CHIP_RV740:
-			return 1;
-		default:
-			return 0;
-		}
-		break;
-	case CHIP_R100:
-	case CHIP_RV100:
-	case CHIP_RS100:
-	case CHIP_RV200:
-	case CHIP_RS200:
-	case CHIP_R200:
-	case CHIP_RV250:
-	case CHIP_RS300:
-	case CHIP_RV280:
-	case CHIP_R300:
-	case CHIP_R350:
-	case CHIP_RV350:
-	case CHIP_RV380:
-	case CHIP_R420:
-	case CHIP_R423:
-	case CHIP_RV410:
-	case CHIP_RS400:
-	case CHIP_RS480:
-	case CHIP_RS600:
-	case CHIP_RS690:
-	case CHIP_RS740:
-	case CHIP_RV515:
-	case CHIP_R520:
-	case CHIP_RV530:
-	case CHIP_RV560:
-	case CHIP_RV570:
-	case CHIP_R580:
-	case CHIP_CEDAR:
-	case CHIP_REDWOOD:
-	case CHIP_JUNIPER:
-	case CHIP_CYPRESS:
-	case CHIP_HEMLOCK:
-	default:
-		return 0;
-	}
-}
diff --git a/src/gallium/winsys/radeon/drm/Makefile b/src/gallium/winsys/radeon/drm/Makefile
index 7f69e39273..aa73edde34 100644
--- a/src/gallium/winsys/radeon/drm/Makefile
+++ b/src/gallium/winsys/radeon/drm/Makefile
@@ -6,7 +6,7 @@ LIBNAME = radeonwinsys
 
 C_SOURCES = \
 	radeon_drm_buffer.c \
-	radeon_drm.c \
+	radeon_drm_common.c \
 	radeon_r300.c
 
 LIBRARY_INCLUDES = -I$(TOP)/src/gallium/drivers/r300 \
diff --git a/src/gallium/winsys/radeon/drm/SConscript b/src/gallium/winsys/radeon/drm/SConscript
index 60e409fe10..2dbf61a7ba 100644
--- a/src/gallium/winsys/radeon/drm/SConscript
+++ b/src/gallium/winsys/radeon/drm/SConscript
@@ -4,7 +4,7 @@ env = env.Clone()
 
 radeon_sources = [
     'radeon_drm_buffer.c',
-    'radeon_drm.c',
+    'radeon_drm_common.c',
     'radeon_r300.c',
 ]
 
diff --git a/src/gallium/winsys/radeon/drm/radeon_buffer.h b/src/gallium/winsys/radeon/drm/radeon_buffer.h
deleted file mode 100644
index a8137d85e8..0000000000
--- a/src/gallium/winsys/radeon/drm/radeon_buffer.h
+++ /dev/null
@@ -1,109 +0,0 @@
-/*
- * Copyright © 2008 Jérôme Glisse
- * All Rights Reserved.
- *
- * Permission is hereby granted, free of charge, to any person obtaining
- * a copy of this software and associated documentation files (the
- * "Software"), to deal in the Software without restriction, including
- * without limitation the rights to use, copy, modify, merge, publish,
- * distribute, sub license, and/or sell copies of the Software, and to
- * permit persons to whom the Software is furnished to do so, subject to
- * the following conditions:
- *
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
- * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES
- * OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
- * NON-INFRINGEMENT. IN NO EVENT SHALL THE COPYRIGHT HOLDERS, AUTHORS
- * AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
- * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
- * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
- * USE OR OTHER DEALINGS IN THE SOFTWARE.
- *
- * The above copyright notice and this permission notice (including the
- * next paragraph) shall be included in all copies or substantial portions
- * of the Software.
- */
-/*
- * Authors:
- *      Jérôme Glisse <glisse@freedesktop.org>
- */
-#ifndef RADEON_BUFFER_H
-#define RADEON_BUFFER_H
-
-#include <stdio.h>
-
-#include "pipe/p_defines.h"
-#include "util/u_inlines.h"
-
-#include "pipebuffer/pb_buffer.h"
-#include "pipebuffer/pb_bufmgr.h"
-
-#include "radeon_bo.h"
-#include "radeon_cs.h"
-
-#include "radeon_winsys.h"
-
-#define RADEON_PB_USAGE_VERTEX      (1 << 28)
-#define RADEON_PB_USAGE_DOMAIN_GTT  (1 << 29)
-#define RADEON_PB_USAGE_DOMAIN_VRAM (1 << 30)
-
-static INLINE struct pb_buffer *
-radeon_pb_buffer(struct r300_winsys_buffer *buffer)
-{
-    return (struct pb_buffer *)buffer;
-}
-
-static INLINE struct r300_winsys_buffer *
-radeon_libdrm_winsys_buffer(struct pb_buffer *buffer)
-{
-    return (struct r300_winsys_buffer *)buffer;
-}
-
-struct pb_manager *
-radeon_drm_bufmgr_create(struct radeon_libdrm_winsys *rws);
-
-void radeon_drm_bufmgr_add_buffer(struct r300_winsys_cs *cs,
-                                  struct r300_winsys_buffer *buf,
-                                  enum r300_buffer_domain rd,
-                                  enum r300_buffer_domain wd);
-
-void radeon_drm_bufmgr_write_reloc(struct r300_winsys_cs *cs,
-                                   struct r300_winsys_buffer *buf,
-				   enum r300_buffer_domain rd,
-                                   enum r300_buffer_domain wd);
-
-struct pb_buffer *radeon_drm_bufmgr_create_buffer_from_handle(struct pb_manager *_mgr,
-							      uint32_t handle);
-
-void radeon_drm_bufmgr_get_tiling(struct r300_winsys_screen *ws,
-                                  struct r300_winsys_buffer *buf,
-                                  enum r300_buffer_tiling *microtiled,
-                                  enum r300_buffer_tiling *macrotiled);
-
-void radeon_drm_bufmgr_set_tiling(struct r300_winsys_screen *ws,
-                                  struct r300_winsys_buffer *buf,
-                                  enum r300_buffer_tiling microtiled,
-                                  enum r300_buffer_tiling macrotiled,
-                                  uint32_t pitch);
-
-void radeon_drm_bufmgr_flush_maps(struct pb_manager *_mgr);
-
-boolean radeon_drm_bufmgr_get_handle(struct pb_buffer *_buf,
-				     struct winsys_handle *whandle);
-
-boolean radeon_drm_bufmgr_is_buffer_referenced(struct r300_winsys_cs *cs,
-                                               struct r300_winsys_buffer *buf,
-                                               enum r300_reference_domain domain);
-
-void radeon_drm_bufmgr_wait(struct r300_winsys_screen *ws,
-                            struct r300_winsys_buffer *buf);
-
-void *radeon_drm_buffer_map(struct r300_winsys_screen *ws,
-                            struct r300_winsys_buffer *buf,
-                            struct r300_winsys_cs *cs,
-                            enum pipe_transfer_usage usage);
-
-void radeon_drm_buffer_unmap(struct r300_winsys_screen *ws,
-                             struct r300_winsys_buffer *buf);
-
-#endif
diff --git a/src/gallium/winsys/radeon/drm/radeon_drm_buffer.c b/src/gallium/winsys/radeon/drm/radeon_drm_buffer.c
index 78723948d4..dc4b51747f 100644
--- a/src/gallium/winsys/radeon/drm/radeon_drm_buffer.c
+++ b/src/gallium/winsys/radeon/drm/radeon_drm_buffer.c
@@ -1,19 +1,17 @@
-
-#include <sys/ioctl.h>
-#include "radeon_drm.h"
-#include "radeon_bo_gem.h"
 #include "radeon_cs_gem.h"
-#include "radeon_buffer.h"
+#include "radeon_drm_buffer.h"
 
 #include "util/u_hash_table.h"
-#include "util/u_inlines.h"
 #include "util/u_memory.h"
 #include "util/u_simple_list.h"
-#include "pipebuffer/pb_buffer.h"
 #include "pipebuffer/pb_bufmgr.h"
 #include "os/os_thread.h"
 
-#include "radeon_winsys.h"
+#include "state_tracker/drm_driver.h"
+
+#include <radeon_drm.h>
+#include <radeon_bo_gem.h>
+#include <sys/ioctl.h>
 
 struct radeon_drm_bufmgr;
 
@@ -45,7 +43,7 @@ struct radeon_drm_bufmgr {
     struct pb_manager base;
 
     /* Winsys. */
-    struct radeon_libdrm_winsys *rws;
+    struct radeon_drm_winsys *rws;
 
     /* List of mapped buffers and its mutex. */
     struct radeon_drm_buffer buffer_map_list;
@@ -115,7 +113,7 @@ radeon_drm_buffer_map_internal(struct pb_buffer *_buf,
 			       unsigned flags, void *flush_ctx)
 {
     struct radeon_drm_buffer *buf = radeon_drm_buffer(_buf);
-    struct radeon_libdrm_cs *cs = flush_ctx;
+    struct radeon_drm_cs *cs = flush_ctx;
     int write = 0;
 
     /* Note how we use radeon_bo_is_referenced_by_cs here. There are
@@ -225,7 +223,7 @@ radeon_drm_bufmgr_create_buffer_from_handle_unsafe(struct pb_manager *_mgr,
                                                    uint32_t handle)
 {
     struct radeon_drm_bufmgr *mgr = radeon_drm_bufmgr(_mgr);
-    struct radeon_libdrm_winsys *rws = mgr->rws;
+    struct radeon_drm_winsys *rws = mgr->rws;
     struct radeon_drm_buffer *buf;
     struct radeon_bo *bo;
 
@@ -284,7 +282,7 @@ radeon_drm_bufmgr_create_buffer(struct pb_manager *_mgr,
 				const struct pb_desc *desc)
 {
     struct radeon_drm_bufmgr *mgr = radeon_drm_bufmgr(_mgr);
-    struct radeon_libdrm_winsys *rws = mgr->rws;
+    struct radeon_drm_winsys *rws = mgr->rws;
     struct radeon_drm_buffer *buf;
     uint32_t domain;
 
@@ -345,7 +343,7 @@ static int handle_compare(void *key1, void *key2)
 }
 
 struct pb_manager *
-radeon_drm_bufmgr_create(struct radeon_libdrm_winsys *rws)
+radeon_drm_bufmgr_create(struct radeon_drm_winsys *rws)
 {
     struct radeon_drm_bufmgr *mgr;
 
@@ -383,18 +381,18 @@ static struct radeon_drm_buffer *get_drm_buffer(struct pb_buffer *_buf)
     return buf;
 }
 
-void *radeon_drm_buffer_map(struct r300_winsys_screen *ws,
-                            struct r300_winsys_buffer *buf,
-                            struct r300_winsys_cs *cs,
-                            enum pipe_transfer_usage usage)
+static void *radeon_drm_buffer_map(struct r300_winsys_screen *ws,
+                                   struct r300_winsys_buffer *buf,
+                                   struct r300_winsys_cs *cs,
+                                   enum pipe_transfer_usage usage)
 {
     struct pb_buffer *_buf = radeon_pb_buffer(buf);
 
-    return pb_map(_buf, get_pb_usage_from_transfer_flags(usage), radeon_libdrm_cs(cs));
+    return pb_map(_buf, get_pb_usage_from_transfer_flags(usage), radeon_drm_cs(cs));
 }
 
-void radeon_drm_buffer_unmap(struct r300_winsys_screen *ws,
-                             struct r300_winsys_buffer *buf)
+static void radeon_drm_buffer_unmap(struct r300_winsys_screen *ws,
+                                    struct r300_winsys_buffer *buf)
 {
     struct pb_buffer *_buf = radeon_pb_buffer(buf);
 
@@ -425,10 +423,10 @@ boolean radeon_drm_bufmgr_get_handle(struct pb_buffer *_buf,
     return TRUE;
 }
 
-void radeon_drm_bufmgr_get_tiling(struct r300_winsys_screen *ws,
-                                  struct r300_winsys_buffer *_buf,
-                                  enum r300_buffer_tiling *microtiled,
-                                  enum r300_buffer_tiling *macrotiled)
+static void radeon_drm_buffer_get_tiling(struct r300_winsys_screen *ws,
+                                         struct r300_winsys_buffer *_buf,
+                                         enum r300_buffer_tiling *microtiled,
+                                         enum r300_buffer_tiling *macrotiled)
 {
     struct radeon_drm_buffer *buf = get_drm_buffer(radeon_pb_buffer(_buf));
     uint32_t flags = 0, pitch;
@@ -444,11 +442,11 @@ void radeon_drm_bufmgr_get_tiling(struct r300_winsys_screen *ws,
 	*macrotiled = R300_BUFFER_TILED;
 }
 
-void radeon_drm_bufmgr_set_tiling(struct r300_winsys_screen *ws,
-                                  struct r300_winsys_buffer *_buf,
-                                  enum r300_buffer_tiling microtiled,
-                                  enum r300_buffer_tiling macrotiled,
-                                  uint32_t pitch)
+static void radeon_drm_buffer_set_tiling(struct r300_winsys_screen *ws,
+                                         struct r300_winsys_buffer *_buf,
+                                         enum r300_buffer_tiling microtiled,
+                                         enum r300_buffer_tiling macrotiled,
+                                         uint32_t pitch)
 {
     struct radeon_drm_buffer *buf = get_drm_buffer(radeon_pb_buffer(_buf));
     uint32_t flags = 0;
@@ -465,66 +463,60 @@ void radeon_drm_bufmgr_set_tiling(struct r300_winsys_screen *ws,
     radeon_bo_set_tiling(buf->bo, flags, pitch);
 }
 
-static uint32_t get_gem_domain(enum r300_buffer_domain domain)
-{
-    uint32_t res = 0;
-
-    if (domain & R300_DOMAIN_GTT)
-        res |= RADEON_GEM_DOMAIN_GTT;
-    if (domain & R300_DOMAIN_VRAM)
-        res |= RADEON_GEM_DOMAIN_VRAM;
-    return res;
-}
-
-void radeon_drm_bufmgr_add_buffer(struct r300_winsys_cs *rcs,
-                                  struct r300_winsys_buffer *_buf,
-                                  enum r300_buffer_domain rd,
-                                  enum r300_buffer_domain wd)
+static void radeon_drm_bufmgr_add_buffer(struct r300_winsys_cs *rcs,
+                                         struct r300_winsys_cs_buffer *_buf,
+                                         enum r300_buffer_domain rd,
+                                         enum r300_buffer_domain wd)
 {
-    struct radeon_libdrm_cs *cs = radeon_libdrm_cs(rcs);
-    struct radeon_drm_buffer *buf = get_drm_buffer(radeon_pb_buffer(_buf));
-    uint32_t gem_rd = get_gem_domain(rd);
-    uint32_t gem_wd = get_gem_domain(wd);
+    struct radeon_drm_cs *cs = radeon_drm_cs(rcs);
+    struct radeon_bo *bo = (struct radeon_bo*)_buf;
 
-    radeon_cs_space_add_persistent_bo(cs->cs, buf->bo, gem_rd, gem_wd);
+    radeon_cs_space_add_persistent_bo(cs->cs, bo, rd, wd);
 }
 
-void radeon_drm_bufmgr_write_reloc(struct r300_winsys_cs *rcs,
-                                   struct r300_winsys_buffer *_buf,
-				   enum r300_buffer_domain rd,
-                                   enum r300_buffer_domain wd)
+static void radeon_drm_bufmgr_write_reloc(struct r300_winsys_cs *rcs,
+                                          struct r300_winsys_cs_buffer *_buf,
+                                          enum r300_buffer_domain rd,
+                                          enum r300_buffer_domain wd)
 {
-    struct radeon_libdrm_cs *cs = radeon_libdrm_cs(rcs);
-    struct radeon_drm_buffer *buf = get_drm_buffer(radeon_pb_buffer(_buf));
+    struct radeon_drm_cs *cs = radeon_drm_cs(rcs);
+    struct radeon_bo *bo = (struct radeon_bo*)_buf;
     int retval;
-    uint32_t gem_rd = get_gem_domain(rd);
-    uint32_t gem_wd = get_gem_domain(wd);
 
     cs->cs->cdw = cs->base.cdw;
-    retval = radeon_cs_write_reloc(cs->cs, buf->bo, gem_rd, gem_wd, 0);
+    retval = radeon_cs_write_reloc(cs->cs, bo, rd, wd, 0);
     cs->base.cdw = cs->cs->cdw;
     if (retval) {
         fprintf(stderr, "radeon: Relocation of %p (%d, %d, %d) failed!\n",
-                buf, gem_rd, gem_wd, 0);
+                bo, rd, wd, 0);
     }
 }
 
-boolean radeon_drm_bufmgr_is_buffer_referenced(struct r300_winsys_cs *rcs,
-                                               struct r300_winsys_buffer *_buf,
+static struct r300_winsys_cs_buffer *radeon_drm_get_cs_handle(
+        struct r300_winsys_screen *rws,
+        struct r300_winsys_buffer *_buf)
+{
+    /* return pure radeon_bo. */
+    return (struct r300_winsys_cs_buffer*)
+            get_drm_buffer(radeon_pb_buffer(_buf))->bo;
+}
+
+static boolean radeon_drm_is_buffer_referenced(struct r300_winsys_cs *rcs,
+                                               struct r300_winsys_cs_buffer *_buf,
                                                enum r300_reference_domain domain)
 {
-    struct radeon_libdrm_cs *cs = radeon_libdrm_cs(rcs);
-    struct radeon_drm_buffer *buf = get_drm_buffer(radeon_pb_buffer(_buf));
+    struct radeon_drm_cs *cs = radeon_drm_cs(rcs);
+    struct radeon_bo *bo = (struct radeon_bo*)_buf;
     uint32_t tmp;
 
     if (domain & R300_REF_CS) {
-        if (radeon_bo_is_referenced_by_cs(buf->bo, cs->cs)) {
+        if (radeon_bo_is_referenced_by_cs(bo, cs->cs)) {
             return TRUE;
         }
     }
 
     if (domain & R300_REF_HW) {
-        if (radeon_bo_is_busy(buf->bo, &tmp)) {
+        if (radeon_bo_is_busy(bo, &tmp)) {
             return TRUE;
         }
     }
@@ -550,10 +542,23 @@ void radeon_drm_bufmgr_flush_maps(struct pb_manager *_mgr)
     pipe_mutex_unlock(mgr->buffer_map_list_mutex);
 }
 
-void radeon_drm_bufmgr_wait(struct r300_winsys_screen *ws,
-                            struct r300_winsys_buffer *_buf)
+static void radeon_drm_buffer_wait(struct r300_winsys_screen *ws,
+                                   struct r300_winsys_buffer *_buf)
 {
     struct radeon_drm_buffer *buf = get_drm_buffer(radeon_pb_buffer(_buf));
 
     radeon_bo_wait(buf->bo);
 }
+
+void radeon_drm_bufmgr_init_functions(struct radeon_drm_winsys *ws)
+{
+    ws->base.buffer_get_cs_handle = radeon_drm_get_cs_handle;
+    ws->base.buffer_set_tiling = radeon_drm_buffer_set_tiling;
+    ws->base.buffer_get_tiling = radeon_drm_buffer_get_tiling;
+    ws->base.buffer_map = radeon_drm_buffer_map;
+    ws->base.buffer_unmap = radeon_drm_buffer_unmap;
+    ws->base.buffer_wait = radeon_drm_buffer_wait;
+    ws->base.cs_is_buffer_referenced = radeon_drm_is_buffer_referenced;
+    ws->base.cs_add_buffer = radeon_drm_bufmgr_add_buffer;
+    ws->base.cs_write_reloc = radeon_drm_bufmgr_write_reloc;
+}
diff --git a/src/gallium/winsys/radeon/drm/radeon_drm.h b/src/gallium/winsys/radeon/drm/radeon_drm_buffer.h
index df6dd91ad5..494abdc0b4 100644
--- a/src/gallium/winsys/radeon/drm/radeon_drm.h
+++ b/src/gallium/winsys/radeon/drm/radeon_drm_buffer.h
@@ -1,7 +1,7 @@
-/* 
- * Copyright © 2009 Corbin Simpson
+/*
+ * Copyright © 2008 Jérôme Glisse
  * All Rights Reserved.
- * 
+ *
  * Permission is hereby granted, free of charge, to any person obtaining
  * a copy of this software and associated documentation files (the
  * "Software"), to deal in the Software without restriction, including
@@ -9,14 +9,14 @@
  * distribute, sub license, and/or sell copies of the Software, and to
  * permit persons to whom the Software is furnished to do so, subject to
  * the following conditions:
- * 
+ *
  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES
  * OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
  * NON-INFRINGEMENT. IN NO EVENT SHALL THE COPYRIGHT HOLDERS, AUTHORS
  * AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
- * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE 
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
  * USE OR OTHER DEALINGS IN THE SOFTWARE.
  *
  * The above copyright notice and this permission notice (including the
@@ -25,19 +25,29 @@
  */
 /*
  * Authors:
- *      Corbin Simpson <MostAwesomeDude@gmail.com>
+ *      Jérôme Glisse <glisse@freedesktop.org>
  */
-#ifndef RADEON_DRM_H
-#define RADEON_DRM_H
+#ifndef RADEON_DRM_BUFFER_H
+#define RADEON_DRM_BUFFER_H
 
-#include "state_tracker/drm_driver.h"
+#include "radeon_winsys.h"
 
-/* Guess at whether this chipset should use r300g.
- *
- * I believe that this check is valid, but I haven't been exhaustive. */
-static INLINE boolean is_r3xx(int pciid)
+#define RADEON_PB_USAGE_VERTEX      (1 << 28)
+#define RADEON_PB_USAGE_DOMAIN_GTT  (1 << 29)
+#define RADEON_PB_USAGE_DOMAIN_VRAM (1 << 30)
+
+static INLINE struct pb_buffer *
+radeon_pb_buffer(struct r300_winsys_buffer *buffer)
 {
-    return (pciid > 0x3150) && (pciid < 0x796f);
+    return (struct pb_buffer *)buffer;
 }
 
+struct pb_manager *radeon_drm_bufmgr_create(struct radeon_drm_winsys *rws);
+struct pb_buffer *radeon_drm_bufmgr_create_buffer_from_handle(struct pb_manager *_mgr,
+							      uint32_t handle);
+void radeon_drm_bufmgr_flush_maps(struct pb_manager *_mgr);
+boolean radeon_drm_bufmgr_get_handle(struct pb_buffer *_buf,
+				     struct winsys_handle *whandle);
+void radeon_drm_bufmgr_init_functions(struct radeon_drm_winsys *ws);
+
 #endif
diff --git a/src/gallium/winsys/radeon/drm/radeon_drm.c b/src/gallium/winsys/radeon/drm/radeon_drm_common.c
index 86d4f94969..6bc6244115 100644
--- a/src/gallium/winsys/radeon/drm/radeon_drm.c
+++ b/src/gallium/winsys/radeon/drm/radeon_drm_common.c
@@ -29,30 +29,21 @@
  *      Joakim Sindholt <opensource@zhasha.com>
  */
 
-#include "radeon_drm.h"
-#include "radeon_r300.h"
-#include "radeon_buffer.h"
+#include "radeon_winsys.h"
+#include "radeon_drm_buffer.h"
 #include "radeon_drm_public.h"
 
-#include "r300_winsys.h"
-
+#include "pipebuffer/pb_bufmgr.h"
 #include "util/u_memory.h"
 
-#include "xf86drm.h"
-
-static struct radeon_libdrm_winsys *
-radeon_winsys_create(int fd)
-{
-    struct radeon_libdrm_winsys *rws;
+#include "state_tracker/drm_driver.h"
 
-    rws = CALLOC_STRUCT(radeon_libdrm_winsys);
-    if (rws == NULL) {
-        return NULL;
-    }
+#include <radeon_drm.h>
+#include <radeon_bo_gem.h>
+#include <radeon_cs_gem.h>
+#include <xf86drm.h>
+#include <stdio.h>
 
-    rws->fd = fd;
-    return rws;
-}
 
 /* Enable/disable Hyper-Z access. Return TRUE on success. */
 static boolean radeon_set_hyperz_access(int fd, boolean enable)
@@ -80,7 +71,7 @@ static boolean radeon_set_hyperz_access(int fd, boolean enable)
 }
 
 /* Helper function to do the ioctls needed for setup and init. */
-static void do_ioctls(int fd, struct radeon_libdrm_winsys* winsys)
+static void do_ioctls(struct radeon_drm_winsys *winsys)
 {
     struct drm_radeon_gem_info gem_info = {0};
     struct drm_radeon_info info = {0};
@@ -107,7 +98,7 @@ static void do_ioctls(int fd, struct radeon_libdrm_winsys* winsys)
      * (see radeon_gem_info_ioctl in radeon_gem.c) but that's alright because
      * we don't actually use the info for anything yet. */
 
-    version = drmGetVersion(fd);
+    version = drmGetVersion(winsys->fd);
     if (version->version_major != 2) {
         fprintf(stderr, "%s: DRM version is %d.%d.%d but this driver is "
                 "only compatible with 2.x.x\n", __FUNCTION__,
@@ -132,7 +123,7 @@ static void do_ioctls(int fd, struct radeon_libdrm_winsys* winsys)
                          version->version_minor >= 6);
 
     info.request = RADEON_INFO_DEVICE_ID;
-    retval = drmCommandWriteRead(fd, DRM_RADEON_INFO, &info, sizeof(info));
+    retval = drmCommandWriteRead(winsys->fd, DRM_RADEON_INFO, &info, sizeof(info));
     if (retval) {
         fprintf(stderr, "%s: Failed to get PCI ID, "
                 "error number %d\n", __FUNCTION__, retval);
@@ -141,7 +132,7 @@ static void do_ioctls(int fd, struct radeon_libdrm_winsys* winsys)
     winsys->pci_id = target;
 
     info.request = RADEON_INFO_NUM_GB_PIPES;
-    retval = drmCommandWriteRead(fd, DRM_RADEON_INFO, &info, sizeof(info));
+    retval = drmCommandWriteRead(winsys->fd, DRM_RADEON_INFO, &info, sizeof(info));
     if (retval) {
         fprintf(stderr, "%s: Failed to get GB pipe count, "
                 "error number %d\n", __FUNCTION__, retval);
@@ -150,7 +141,7 @@ static void do_ioctls(int fd, struct radeon_libdrm_winsys* winsys)
     winsys->gb_pipes = target;
 
     info.request = RADEON_INFO_NUM_Z_PIPES;
-    retval = drmCommandWriteRead(fd, DRM_RADEON_INFO, &info, sizeof(info));
+    retval = drmCommandWriteRead(winsys->fd, DRM_RADEON_INFO, &info, sizeof(info));
     if (retval) {
         fprintf(stderr, "%s: Failed to get Z pipe count, "
                 "error number %d\n", __FUNCTION__, retval);
@@ -158,9 +149,9 @@ static void do_ioctls(int fd, struct radeon_libdrm_winsys* winsys)
     }
     winsys->z_pipes = target;
 
-    winsys->hyperz = radeon_set_hyperz_access(fd, TRUE);
+    winsys->hyperz = radeon_set_hyperz_access(winsys->fd, TRUE);
 
-    retval = drmCommandWriteRead(fd, DRM_RADEON_GEM_INFO,
+    retval = drmCommandWriteRead(winsys->fd, DRM_RADEON_GEM_INFO,
             &gem_info, sizeof(gem_info));
     if (retval) {
         fprintf(stderr, "%s: Failed to get MM info, error number %d\n",
@@ -184,29 +175,65 @@ static void do_ioctls(int fd, struct radeon_libdrm_winsys* winsys)
     drmFreeVersion(version);
 }
 
-/* Create a pipe_screen. */
-struct r300_winsys_screen* r300_drm_winsys_screen_create(int drmFB)
+static void radeon_winsys_destroy(struct r300_winsys_screen *rws)
+{
+    struct radeon_drm_winsys *ws = (struct radeon_drm_winsys*)rws;
+
+    ws->cman->destroy(ws->cman);
+    ws->kman->destroy(ws->kman);
+
+    radeon_bo_manager_gem_dtor(ws->bom);
+    radeon_cs_manager_gem_dtor(ws->csm);
+    FREE(rws);
+}
+
+struct r300_winsys_screen *r300_drm_winsys_screen_create(int fd)
 {
-    struct radeon_libdrm_winsys* rws; 
-    boolean ret;
-
-    rws = radeon_winsys_create(drmFB);
-    if (!rws)
-	return NULL;
-
-    do_ioctls(drmFB, rws);
-
-    /* The state tracker can organize a softpipe fallback if no hw
-     * driver is found.
-     */
-    if (is_r3xx(rws->pci_id)) {
-        ret = radeon_setup_winsys(drmFB, rws);
-	if (ret == FALSE)
-	    goto fail;
-        return &rws->base;
+    struct radeon_drm_winsys *ws = CALLOC_STRUCT(radeon_drm_winsys);
+    if (!ws) {
+        return NULL;
+    }
+
+    ws->fd = fd;
+    do_ioctls(ws);
+
+    if (!is_r3xx(ws->pci_id)) {
+        goto fail;
     }
 
+    /* Create managers. */
+    ws->bom = radeon_bo_manager_gem_ctor(fd);
+    if (!ws->bom)
+	goto fail;
+    ws->csm = radeon_cs_manager_gem_ctor(fd);
+    if (!ws->csm)
+	goto fail;
+    ws->kman = radeon_drm_bufmgr_create(ws);
+    if (!ws->kman)
+	goto fail;
+    ws->cman = pb_cache_manager_create(ws->kman, 1000000);
+    if (!ws->cman)
+	goto fail;
+
+    /* Set functions. */
+    ws->base.destroy = radeon_winsys_destroy;
+
+    radeon_drm_bufmgr_init_functions(ws);
+    radeon_winsys_init_functions(ws);
+
+    return &ws->base;
+
 fail:
-    FREE(rws);
+    if (ws->bom)
+	radeon_bo_manager_gem_dtor(ws->bom);
+    if (ws->csm)
+	radeon_cs_manager_gem_dtor(ws->csm);
+
+    if (ws->cman)
+	ws->cman->destroy(ws->cman);
+    if (ws->kman)
+	ws->kman->destroy(ws->kman);
+
+    FREE(ws);
     return NULL;
 }
diff --git a/src/gallium/winsys/radeon/drm/radeon_drm_public.h b/src/gallium/winsys/radeon/drm/radeon_drm_public.h
index 0d96ae8c47..3a208cdd4c 100644
--- a/src/gallium/winsys/radeon/drm/radeon_drm_public.h
+++ b/src/gallium/winsys/radeon/drm/radeon_drm_public.h
@@ -1,9 +1,222 @@
-
 #ifndef RADEON_DRM_PUBLIC_H
 #define RADEON_DRM_PUBLIC_H
 
+#include "pipe/p_defines.h"
+
 struct r300_winsys_screen;
 
-struct r300_winsys_screen *r300_drm_winsys_screen_create(int drmFD);
+struct r300_winsys_screen *r300_drm_winsys_screen_create(int fd);
+
+static INLINE boolean is_r3xx(int pciid)
+{
+      switch (pciid) {
+      case 0x4144: /* PCI_CHIP_R300_AD */
+      case 0x4145: /* PCI_CHIP_R300_AE */
+      case 0x4146: /* PCI_CHIP_R300_AF */
+      case 0x4147: /* PCI_CHIP_R300_AG */
+      case 0x4E44: /* PCI_CHIP_R300_ND */
+      case 0x4E45: /* PCI_CHIP_R300_NE */
+      case 0x4E46: /* PCI_CHIP_R300_NF */
+      case 0x4E47: /* PCI_CHIP_R300_NG */
+      case 0x4E48: /* PCI_CHIP_R350_NH */
+      case 0x4E49: /* PCI_CHIP_R350_NI */
+      case 0x4E4B: /* PCI_CHIP_R350_NK */
+      case 0x4148: /* PCI_CHIP_R350_AH */
+      case 0x4149: /* PCI_CHIP_R350_AI */
+      case 0x414A: /* PCI_CHIP_R350_AJ */
+      case 0x414B: /* PCI_CHIP_R350_AK */
+      case 0x4E4A: /* PCI_CHIP_R360_NJ */
+      case 0x4150: /* PCI_CHIP_RV350_AP */
+      case 0x4151: /* PCI_CHIP_RV350_AQ */
+      case 0x4152: /* PCI_CHIP_RV350_AR */
+      case 0x4153: /* PCI_CHIP_RV350_AS */
+      case 0x4154: /* PCI_CHIP_RV350_AT */
+      case 0x4155: /* PCI_CHIP_RV350_AU */
+      case 0x4156: /* PCI_CHIP_RV350_AV */
+      case 0x4E50: /* PCI_CHIP_RV350_NP */
+      case 0x4E51: /* PCI_CHIP_RV350_NQ */
+      case 0x4E52: /* PCI_CHIP_RV350_NR */
+      case 0x4E53: /* PCI_CHIP_RV350_NS */
+      case 0x4E54: /* PCI_CHIP_RV350_NT */
+      case 0x4E56: /* PCI_CHIP_RV350_NV */
+      case 0x5460: /* PCI_CHIP_RV370_5460 */
+      case 0x5462: /* PCI_CHIP_RV370_5462 */
+      case 0x5464: /* PCI_CHIP_RV370_5464 */
+      case 0x5B60: /* PCI_CHIP_RV370_5B60 */
+      case 0x5B62: /* PCI_CHIP_RV370_5B62 */
+      case 0x5B63: /* PCI_CHIP_RV370_5B63 */
+      case 0x5B64: /* PCI_CHIP_RV370_5B64 */
+      case 0x5B65: /* PCI_CHIP_RV370_5B65 */
+      case 0x3150: /* PCI_CHIP_RV380_3150 */
+      case 0x3152: /* PCI_CHIP_RV380_3152 */
+      case 0x3154: /* PCI_CHIP_RV380_3154 */
+      case 0x3155: /* PCI_CHIP_RV380_3155 */
+      case 0x3E50: /* PCI_CHIP_RV380_3E50 */
+      case 0x3E54: /* PCI_CHIP_RV380_3E54 */
+      case 0x4A48: /* PCI_CHIP_R420_JH */
+      case 0x4A49: /* PCI_CHIP_R420_JI */
+      case 0x4A4A: /* PCI_CHIP_R420_JJ */
+      case 0x4A4B: /* PCI_CHIP_R420_JK */
+      case 0x4A4C: /* PCI_CHIP_R420_JL */
+      case 0x4A4D: /* PCI_CHIP_R420_JM */
+      case 0x4A4E: /* PCI_CHIP_R420_JN */
+      case 0x4A4F: /* PCI_CHIP_R420_JO */
+      case 0x4A50: /* PCI_CHIP_R420_JP */
+      case 0x4A54: /* PCI_CHIP_R420_JT */
+      case 0x5548: /* PCI_CHIP_R423_UH */
+      case 0x5549: /* PCI_CHIP_R423_UI */
+      case 0x554A: /* PCI_CHIP_R423_UJ */
+      case 0x554B: /* PCI_CHIP_R423_UK */
+      case 0x5550: /* PCI_CHIP_R423_5550 */
+      case 0x5551: /* PCI_CHIP_R423_UQ */
+      case 0x5552: /* PCI_CHIP_R423_UR */
+      case 0x5554: /* PCI_CHIP_R423_UT */
+      case 0x5D57: /* PCI_CHIP_R423_5D57 */
+      case 0x554C: /* PCI_CHIP_R430_554C */
+      case 0x554D: /* PCI_CHIP_R430_554D */
+      case 0x554E: /* PCI_CHIP_R430_554E */
+      case 0x554F: /* PCI_CHIP_R430_554F */
+      case 0x5D48: /* PCI_CHIP_R430_5D48 */
+      case 0x5D49: /* PCI_CHIP_R430_5D49 */
+      case 0x5D4A: /* PCI_CHIP_R430_5D4A */
+      case 0x5D4C: /* PCI_CHIP_R480_5D4C */
+      case 0x5D4D: /* PCI_CHIP_R480_5D4D */
+      case 0x5D4E: /* PCI_CHIP_R480_5D4E */
+      case 0x5D4F: /* PCI_CHIP_R480_5D4F */
+      case 0x5D50: /* PCI_CHIP_R480_5D50 */
+      case 0x5D52: /* PCI_CHIP_R480_5D52 */
+      case 0x4B49: /* PCI_CHIP_R481_4B49 */
+      case 0x4B4A: /* PCI_CHIP_R481_4B4A */
+      case 0x4B4B: /* PCI_CHIP_R481_4B4B */
+      case 0x4B4C: /* PCI_CHIP_R481_4B4C */
+      case 0x564A: /* PCI_CHIP_RV410_564A */
+      case 0x564B: /* PCI_CHIP_RV410_564B */
+      case 0x564F: /* PCI_CHIP_RV410_564F */
+      case 0x5652: /* PCI_CHIP_RV410_5652 */
+      case 0x5653: /* PCI_CHIP_RV410_5653 */
+      case 0x5657: /* PCI_CHIP_RV410_5657 */
+      case 0x5E48: /* PCI_CHIP_RV410_5E48 */
+      case 0x5E4A: /* PCI_CHIP_RV410_5E4A */
+      case 0x5E4B: /* PCI_CHIP_RV410_5E4B */
+      case 0x5E4C: /* PCI_CHIP_RV410_5E4C */
+      case 0x5E4D: /* PCI_CHIP_RV410_5E4D */
+      case 0x5E4F: /* PCI_CHIP_RV410_5E4F */
+      case 0x5A41: /* PCI_CHIP_RS400_5A41 */
+      case 0x5A42: /* PCI_CHIP_RS400_5A42 */
+      case 0x5A61: /* PCI_CHIP_RC410_5A61 */
+      case 0x5A62: /* PCI_CHIP_RC410_5A62 */
+      case 0x5954: /* PCI_CHIP_RS480_5954 */
+      case 0x5955: /* PCI_CHIP_RS480_5955 */
+      case 0x5974: /* PCI_CHIP_RS482_5974 */
+      case 0x5975: /* PCI_CHIP_RS482_5975 */
+      case 0x7100: /* PCI_CHIP_R520_7100 */
+      case 0x7101: /* PCI_CHIP_R520_7101 */
+      case 0x7102: /* PCI_CHIP_R520_7102 */
+      case 0x7103: /* PCI_CHIP_R520_7103 */
+      case 0x7104: /* PCI_CHIP_R520_7104 */
+      case 0x7105: /* PCI_CHIP_R520_7105 */
+      case 0x7106: /* PCI_CHIP_R520_7106 */
+      case 0x7108: /* PCI_CHIP_R520_7108 */
+      case 0x7109: /* PCI_CHIP_R520_7109 */
+      case 0x710A: /* PCI_CHIP_R520_710A */
+      case 0x710B: /* PCI_CHIP_R520_710B */
+      case 0x710C: /* PCI_CHIP_R520_710C */
+      case 0x710E: /* PCI_CHIP_R520_710E */
+      case 0x710F: /* PCI_CHIP_R520_710F */
+      case 0x7140: /* PCI_CHIP_RV515_7140 */
+      case 0x7141: /* PCI_CHIP_RV515_7141 */
+      case 0x7142: /* PCI_CHIP_RV515_7142 */
+      case 0x7143: /* PCI_CHIP_RV515_7143 */
+      case 0x7144: /* PCI_CHIP_RV515_7144 */
+      case 0x7145: /* PCI_CHIP_RV515_7145 */
+      case 0x7146: /* PCI_CHIP_RV515_7146 */
+      case 0x7147: /* PCI_CHIP_RV515_7147 */
+      case 0x7149: /* PCI_CHIP_RV515_7149 */
+      case 0x714A: /* PCI_CHIP_RV515_714A */
+      case 0x714B: /* PCI_CHIP_RV515_714B */
+      case 0x714C: /* PCI_CHIP_RV515_714C */
+      case 0x714D: /* PCI_CHIP_RV515_714D */
+      case 0x714E: /* PCI_CHIP_RV515_714E */
+      case 0x714F: /* PCI_CHIP_RV515_714F */
+      case 0x7151: /* PCI_CHIP_RV515_7151 */
+      case 0x7152: /* PCI_CHIP_RV515_7152 */
+      case 0x7153: /* PCI_CHIP_RV515_7153 */
+      case 0x715E: /* PCI_CHIP_RV515_715E */
+      case 0x715F: /* PCI_CHIP_RV515_715F */
+      case 0x7180: /* PCI_CHIP_RV515_7180 */
+      case 0x7181: /* PCI_CHIP_RV515_7181 */
+      case 0x7183: /* PCI_CHIP_RV515_7183 */
+      case 0x7186: /* PCI_CHIP_RV515_7186 */
+      case 0x7187: /* PCI_CHIP_RV515_7187 */
+      case 0x7188: /* PCI_CHIP_RV515_7188 */
+      case 0x718A: /* PCI_CHIP_RV515_718A */
+      case 0x718B: /* PCI_CHIP_RV515_718B */
+      case 0x718C: /* PCI_CHIP_RV515_718C */
+      case 0x718D: /* PCI_CHIP_RV515_718D */
+      case 0x718F: /* PCI_CHIP_RV515_718F */
+      case 0x7193: /* PCI_CHIP_RV515_7193 */
+      case 0x7196: /* PCI_CHIP_RV515_7196 */
+      case 0x719B: /* PCI_CHIP_RV515_719B */
+      case 0x719F: /* PCI_CHIP_RV515_719F */
+      case 0x7200: /* PCI_CHIP_RV515_7200 */
+      case 0x7210: /* PCI_CHIP_RV515_7210 */
+      case 0x7211: /* PCI_CHIP_RV515_7211 */
+      case 0x71C0: /* PCI_CHIP_RV530_71C0 */
+      case 0x71C1: /* PCI_CHIP_RV530_71C1 */
+      case 0x71C2: /* PCI_CHIP_RV530_71C2 */
+      case 0x71C3: /* PCI_CHIP_RV530_71C3 */
+      case 0x71C4: /* PCI_CHIP_RV530_71C4 */
+      case 0x71C5: /* PCI_CHIP_RV530_71C5 */
+      case 0x71C6: /* PCI_CHIP_RV530_71C6 */
+      case 0x71C7: /* PCI_CHIP_RV530_71C7 */
+      case 0x71CD: /* PCI_CHIP_RV530_71CD */
+      case 0x71CE: /* PCI_CHIP_RV530_71CE */
+      case 0x71D2: /* PCI_CHIP_RV530_71D2 */
+      case 0x71D4: /* PCI_CHIP_RV530_71D4 */
+      case 0x71D5: /* PCI_CHIP_RV530_71D5 */
+      case 0x71D6: /* PCI_CHIP_RV530_71D6 */
+      case 0x71DA: /* PCI_CHIP_RV530_71DA */
+      case 0x71DE: /* PCI_CHIP_RV530_71DE */
+      case 0x7281: /* PCI_CHIP_RV560_7281 */
+      case 0x7283: /* PCI_CHIP_RV560_7283 */
+      case 0x7287: /* PCI_CHIP_RV560_7287 */
+      case 0x7290: /* PCI_CHIP_RV560_7290 */
+      case 0x7291: /* PCI_CHIP_RV560_7291 */
+      case 0x7293: /* PCI_CHIP_RV560_7293 */
+      case 0x7297: /* PCI_CHIP_RV560_7297 */
+      case 0x7280: /* PCI_CHIP_RV570_7280 */
+      case 0x7288: /* PCI_CHIP_RV570_7288 */
+      case 0x7289: /* PCI_CHIP_RV570_7289 */
+      case 0x728B: /* PCI_CHIP_RV570_728B */
+      case 0x728C: /* PCI_CHIP_RV570_728C */
+      case 0x7240: /* PCI_CHIP_R580_7240 */
+      case 0x7243: /* PCI_CHIP_R580_7243 */
+      case 0x7244: /* PCI_CHIP_R580_7244 */
+      case 0x7245: /* PCI_CHIP_R580_7245 */
+      case 0x7246: /* PCI_CHIP_R580_7246 */
+      case 0x7247: /* PCI_CHIP_R580_7247 */
+      case 0x7248: /* PCI_CHIP_R580_7248 */
+      case 0x7249: /* PCI_CHIP_R580_7249 */
+      case 0x724A: /* PCI_CHIP_R580_724A */
+      case 0x724B: /* PCI_CHIP_R580_724B */
+      case 0x724C: /* PCI_CHIP_R580_724C */
+      case 0x724D: /* PCI_CHIP_R580_724D */
+      case 0x724E: /* PCI_CHIP_R580_724E */
+      case 0x724F: /* PCI_CHIP_R580_724F */
+      case 0x7284: /* PCI_CHIP_R580_7284 */
+      case 0x793F: /* PCI_CHIP_RS600_793F */
+      case 0x7941: /* PCI_CHIP_RS600_7941 */
+      case 0x7942: /* PCI_CHIP_RS600_7942 */
+      case 0x791E: /* PCI_CHIP_RS690_791E */
+      case 0x791F: /* PCI_CHIP_RS690_791F */
+      case 0x796C: /* PCI_CHIP_RS740_796C */
+      case 0x796D: /* PCI_CHIP_RS740_796D */
+      case 0x796E: /* PCI_CHIP_RS740_796E */
+      case 0x796F: /* PCI_CHIP_RS740_796F */
+	      return TRUE;
+      default:
+	      return FALSE;
+      }
+}
 
 #endif
diff --git a/src/gallium/winsys/radeon/drm/radeon_r300.c b/src/gallium/winsys/radeon/drm/radeon_r300.c
index 420522f5c1..9f59b3de46 100644
--- a/src/gallium/winsys/radeon/drm/radeon_r300.c
+++ b/src/gallium/winsys/radeon/drm/radeon_r300.c
@@ -20,15 +20,14 @@
  * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
  * USE OR OTHER DEALINGS IN THE SOFTWARE. */
 
-#include "radeon_r300.h"
-#include "radeon_buffer.h"
+#include "radeon_drm_buffer.h"
+
+#include "util/u_memory.h"
+#include "pipebuffer/pb_bufmgr.h"
 
-#include "radeon_bo_gem.h"
 #include "radeon_cs_gem.h"
 #include "state_tracker/drm_driver.h"
 
-#include "util/u_memory.h"
-
 static unsigned get_pb_usage_from_create_flags(unsigned bind, unsigned usage,
                                                enum r300_buffer_domain domain)
 {
@@ -73,7 +72,7 @@ radeon_r300_winsys_buffer_create(struct r300_winsys_screen *rws,
                                  unsigned usage,
                                  enum r300_buffer_domain domain)
 {
-    struct radeon_libdrm_winsys *ws = radeon_libdrm_winsys(rws);
+    struct radeon_drm_winsys *ws = radeon_drm_winsys(rws);
     struct pb_desc desc;
     struct pb_manager *provider;
     struct pb_buffer *buffer;
@@ -92,7 +91,7 @@ radeon_r300_winsys_buffer_create(struct r300_winsys_screen *rws,
     if (!buffer)
 	return NULL;
 
-    return radeon_libdrm_winsys_buffer(buffer);
+    return (struct r300_winsys_buffer*)buffer;
 }
 
 static void radeon_r300_winsys_buffer_reference(struct r300_winsys_screen *rws,
@@ -104,7 +103,7 @@ static void radeon_r300_winsys_buffer_reference(struct r300_winsys_screen *rws,
 
     pb_reference(&_dst, _src);
 
-    *pdst = radeon_libdrm_winsys_buffer(_dst);
+    *pdst = (struct r300_winsys_buffer*)_dst;
 }
 
 static struct r300_winsys_buffer *radeon_r300_winsys_buffer_from_handle(struct r300_winsys_screen *rws,
@@ -112,7 +111,7 @@ static struct r300_winsys_buffer *radeon_r300_winsys_buffer_from_handle(struct r
                                                                         unsigned *stride,
                                                                         unsigned *size)
 {
-    struct radeon_libdrm_winsys *ws = radeon_libdrm_winsys(rws);
+    struct radeon_drm_winsys *ws = radeon_drm_winsys(rws);
     struct pb_buffer *_buf;
 
     _buf = radeon_drm_bufmgr_create_buffer_from_handle(ws->kman, whandle->handle);
@@ -122,7 +121,7 @@ static struct r300_winsys_buffer *radeon_r300_winsys_buffer_from_handle(struct r
     if (size)
         *size = _buf->base.size;
 
-    return radeon_libdrm_winsys_buffer(_buf);
+    return (struct r300_winsys_buffer*)_buf;
 }
 
 static boolean radeon_r300_winsys_buffer_get_handle(struct r300_winsys_screen *rws,
@@ -139,7 +138,7 @@ static void radeon_r300_winsys_cs_set_flush(struct r300_winsys_cs *rcs,
                                             void (*flush)(void *),
                                             void *user)
 {
-    struct radeon_libdrm_cs *cs = radeon_libdrm_cs(rcs);
+    struct radeon_drm_cs *cs = radeon_drm_cs(rcs);
     cs->flush_cs = flush;
     cs->flush_data = user;
     radeon_cs_space_set_flush(cs->cs, flush, user);
@@ -147,20 +146,20 @@ static void radeon_r300_winsys_cs_set_flush(struct r300_winsys_cs *rcs,
 
 static boolean radeon_r300_winsys_cs_validate(struct r300_winsys_cs *rcs)
 {
-    struct radeon_libdrm_cs *cs = radeon_libdrm_cs(rcs);
+    struct radeon_drm_cs *cs = radeon_drm_cs(rcs);
 
     return radeon_cs_space_check(cs->cs) >= 0;
 }
 
 static void radeon_r300_winsys_cs_reset_buffers(struct r300_winsys_cs *rcs)
 {
-    struct radeon_libdrm_cs *cs = radeon_libdrm_cs(rcs);
+    struct radeon_drm_cs *cs = radeon_drm_cs(rcs);
     radeon_cs_space_reset_bos(cs->cs);
 }
 
 static void radeon_r300_winsys_cs_flush(struct r300_winsys_cs *rcs)
 {
-    struct radeon_libdrm_cs *cs = radeon_libdrm_cs(rcs);
+    struct radeon_drm_cs *cs = radeon_drm_cs(rcs);
     int retval;
 
     /* Don't flush a zero-sized CS. */
@@ -190,15 +189,14 @@ static void radeon_r300_winsys_cs_flush(struct r300_winsys_cs *rcs)
      * spinning through one CS while another one is being filled. */
     radeon_cs_erase(cs->cs);
 
-    cs->base.ptr = cs->cs->packets;
+    cs->base.buf = cs->cs->packets;
     cs->base.cdw = cs->cs->cdw;
-    cs->base.ndw = cs->cs->ndw;
 }
 
 static uint32_t radeon_get_value(struct r300_winsys_screen *rws,
                                  enum r300_value_id id)
 {
-    struct radeon_libdrm_winsys *ws = (struct radeon_libdrm_winsys *)rws;
+    struct radeon_drm_winsys *ws = (struct radeon_drm_winsys *)rws;
 
     switch(id) {
     case R300_VID_PCI_ID:
@@ -221,8 +219,8 @@ static uint32_t radeon_get_value(struct r300_winsys_screen *rws,
 
 static struct r300_winsys_cs *radeon_r300_winsys_cs_create(struct r300_winsys_screen *rws)
 {
-    struct radeon_libdrm_winsys *ws = radeon_libdrm_winsys(rws);
-    struct radeon_libdrm_cs *cs = CALLOC_STRUCT(radeon_libdrm_cs);
+    struct radeon_drm_winsys *ws = radeon_drm_winsys(rws);
+    struct radeon_drm_cs *cs = CALLOC_STRUCT(radeon_drm_cs);
 
     if (!cs)
         return NULL;
@@ -240,83 +238,29 @@ static struct r300_winsys_cs *radeon_r300_winsys_cs_create(struct r300_winsys_sc
             RADEON_GEM_DOMAIN_VRAM, ws->vram_size);
 
     cs->ws = ws;
-    cs->base.ptr = cs->cs->packets;
+    cs->base.buf = cs->cs->packets;
     cs->base.cdw = cs->cs->cdw;
-    cs->base.ndw = cs->cs->ndw;
     return &cs->base;
 }
 
 static void radeon_r300_winsys_cs_destroy(struct r300_winsys_cs *rcs)
 {
-    struct radeon_libdrm_cs *cs = radeon_libdrm_cs(rcs);
+    struct radeon_drm_cs *cs = radeon_drm_cs(rcs);
     radeon_cs_destroy(cs->cs);
     FREE(cs);
 }
 
-static void radeon_winsys_destroy(struct r300_winsys_screen *rws)
-{
-    struct radeon_libdrm_winsys *ws = (struct radeon_libdrm_winsys *)rws;
-
-    ws->cman->destroy(ws->cman);
-    ws->kman->destroy(ws->kman);
-
-    radeon_bo_manager_gem_dtor(ws->bom);
-    radeon_cs_manager_gem_dtor(ws->csm);
-
-    FREE(rws);
-}
-
-boolean radeon_setup_winsys(int fd, struct radeon_libdrm_winsys* ws)
+void radeon_winsys_init_functions(struct radeon_drm_winsys *ws)
 {
-    ws->csm = radeon_cs_manager_gem_ctor(fd);
-    if (!ws->csm)
-	goto fail;
-    ws->bom = radeon_bo_manager_gem_ctor(fd);
-    if (!ws->bom)
-	goto fail;
-    ws->kman = radeon_drm_bufmgr_create(ws);
-    if (!ws->kman)
-	goto fail;
-
-    ws->cman = pb_cache_manager_create(ws->kman, 100000);
-    if (!ws->cman)
-	goto fail;
-
-    ws->base.destroy = radeon_winsys_destroy;
     ws->base.get_value = radeon_get_value;
-
     ws->base.buffer_create = radeon_r300_winsys_buffer_create;
-    ws->base.buffer_set_tiling = radeon_drm_bufmgr_set_tiling;
-    ws->base.buffer_get_tiling = radeon_drm_bufmgr_get_tiling;
-    ws->base.buffer_map = radeon_drm_buffer_map;
-    ws->base.buffer_unmap = radeon_drm_buffer_unmap;
-    ws->base.buffer_wait = radeon_drm_bufmgr_wait;
     ws->base.buffer_reference = radeon_r300_winsys_buffer_reference;
     ws->base.buffer_from_handle = radeon_r300_winsys_buffer_from_handle;
     ws->base.buffer_get_handle = radeon_r300_winsys_buffer_get_handle;
-
     ws->base.cs_create = radeon_r300_winsys_cs_create;
     ws->base.cs_destroy = radeon_r300_winsys_cs_destroy;
-    ws->base.cs_add_buffer = radeon_drm_bufmgr_add_buffer;
     ws->base.cs_validate = radeon_r300_winsys_cs_validate;
-    ws->base.cs_write_reloc = radeon_drm_bufmgr_write_reloc;
     ws->base.cs_flush = radeon_r300_winsys_cs_flush;
     ws->base.cs_reset_buffers = radeon_r300_winsys_cs_reset_buffers;
     ws->base.cs_set_flush = radeon_r300_winsys_cs_set_flush;
-    ws->base.cs_is_buffer_referenced = radeon_drm_bufmgr_is_buffer_referenced;
-    return TRUE;
-
-fail:
-    if (ws->csm)
-	radeon_cs_manager_gem_dtor(ws->csm);
-
-    if (ws->bom)
-	radeon_bo_manager_gem_dtor(ws->bom);
-
-    if (ws->cman)
-	ws->cman->destroy(ws->cman);
-    if (ws->kman)
-	ws->kman->destroy(ws->kman);
-
-    return FALSE;
 }
diff --git a/src/gallium/winsys/radeon/drm/radeon_winsys.h b/src/gallium/winsys/radeon/drm/radeon_winsys.h
index 6f4aa4bce3..81da1a25e0 100644
--- a/src/gallium/winsys/radeon/drm/radeon_winsys.h
+++ b/src/gallium/winsys/radeon/drm/radeon_winsys.h
@@ -32,63 +32,37 @@
 
 #include "r300_winsys.h"
 
-struct radeon_libdrm_winsys {
-    /* Parent class. */
+struct radeon_drm_winsys {
     struct r300_winsys_screen base;
 
-    struct pb_manager *kman;
+    int fd; /* DRM file descriptor */
 
+    struct radeon_bo_manager *bom; /* Radeon BO manager. */
+    struct pb_manager *kman;
     struct pb_manager *cman;
 
-    /* PCI ID */
-    uint32_t pci_id;
-
-    /* GB pipe count */
-    uint32_t gb_pipes;
-
-    /* Z pipe count (rv530 only) */
-    uint32_t z_pipes;
-
-    /* GART size. */
-    uint32_t gart_size;
-
-    /* VRAM size. */
-    uint32_t vram_size;
-
-    /* Square tiling support. */
-    boolean squaretiling;
-
-    /* DRM 2.3.0
-     *   - R500 VAP regs
-     *   - MSPOS regs
-     *   - Fixed texture 3D size calculation
-     */
+    uint32_t pci_id;        /* PCI ID */
+    uint32_t gb_pipes;      /* GB pipe count */
+    uint32_t z_pipes;       /* Z pipe count (rv530 only) */
+    uint32_t gart_size;     /* GART size. */
+    uint32_t vram_size;     /* VRAM size. */
+    boolean squaretiling;   /* Square tiling support. */
+    /* DRM 2.3.0 (R500 VAP regs, MSPOS regs, fixed tex3D size checking) */
     boolean drm_2_3_0;
-
-    /* DRM 2.6.0
-     *   - Hyper-Z
-     *   - GB_Z_PEQ_CONFIG allowed on rv350->r4xx, we should initialize it
-     */
+    /* DRM 2.6.0 (Hyper-Z, GB_Z_PEQ_CONFIG allowed on rv350->r4xx) */
     boolean drm_2_6_0;
-
-    /* hyperz user */
+    /* Hyper-Z user */
     boolean hyperz;
 
-    /* DRM FD */
-    int fd;
-
-    /* Radeon BO manager. */
-    struct radeon_bo_manager *bom;
-
     /* Radeon CS manager. */
     struct radeon_cs_manager *csm;
 };
 
-struct radeon_libdrm_cs {
+struct radeon_drm_cs {
     struct r300_winsys_cs base;
 
     /* The winsys. */
-    struct radeon_libdrm_winsys *ws;
+    struct radeon_drm_winsys *ws;
 
     /* The libdrm command stream. */
     struct radeon_cs *cs;
@@ -98,16 +72,18 @@ struct radeon_libdrm_cs {
     void *flush_data;
 };
 
-static INLINE struct radeon_libdrm_cs *
-radeon_libdrm_cs(struct r300_winsys_cs *base)
+static INLINE struct radeon_drm_cs *
+radeon_drm_cs(struct r300_winsys_cs *base)
 {
-    return (struct radeon_libdrm_cs*)base;
+    return (struct radeon_drm_cs*)base;
 }
 
-static INLINE struct radeon_libdrm_winsys *
-radeon_libdrm_winsys(struct r300_winsys_screen *base)
+static INLINE struct radeon_drm_winsys *
+radeon_drm_winsys(struct r300_winsys_screen *base)
 {
-    return (struct radeon_libdrm_winsys*)base;
+    return (struct radeon_drm_winsys*)base;
 }
 
+void radeon_winsys_init_functions(struct radeon_drm_winsys *ws);
+
 #endif
diff --git a/src/gallium/winsys/sw/wrapper/wrapper_sw_winsys.c b/src/gallium/winsys/sw/wrapper/wrapper_sw_winsys.c
index bc2623e7b7..8f9a90858d 100644
--- a/src/gallium/winsys/sw/wrapper/wrapper_sw_winsys.c
+++ b/src/gallium/winsys/sw/wrapper/wrapper_sw_winsys.c
@@ -93,9 +93,9 @@ wsw_dt_get_stride(struct wrapper_sw_displaytarget *wdt, unsigned *stride)
    struct pipe_resource *tex = wdt->tex;
    struct pipe_transfer *tr;
 
-   tr = pipe_get_transfer(pipe, tex, 0, 0, 0,
-			  PIPE_TRANSFER_READ_WRITE,
-			  0, 0, wdt->width, wdt->height);
+   tr = pipe_get_transfer(pipe, tex, 0, 0,
+                          PIPE_TRANSFER_READ_WRITE,
+                          0, 0, wdt->width, wdt->height);
    if (!tr)
       return FALSE;
 
@@ -149,6 +149,8 @@ wsw_dt_create(struct sw_winsys *ws,
    templ.target = wsw->target;
    templ.width0 = width;
    templ.height0 = height;
+   templ.depth0 = 1;
+   templ.array_size = 1;
    templ.format = format;
    templ.bind = bind;
 
@@ -204,9 +206,9 @@ wsw_dt_map(struct sw_winsys *ws,
 
       assert(!wdt->transfer);
 
-      tr = pipe_get_transfer(pipe, tex, 0, 0, 0,
-			     PIPE_TRANSFER_READ_WRITE,
-			     0, 0, wdt->width, wdt->height);
+      tr = pipe_get_transfer(pipe, tex, 0, 0,
+                             PIPE_TRANSFER_READ_WRITE,
+                             0, 0, wdt->width, wdt->height);
       if (!tr)
          return NULL;