Merge remote branch 'origin/master' into gallium_draw_llvm

42 files changed, 1386 insertions, 1118 deletions

diff --git a/src/gallium/drivers/llvmpipe/Makefile b/src/gallium/drivers/llvmpipe/Makefile
index 3173251437..89c06ea3ad 100644
--- a/src/gallium/drivers/llvmpipe/Makefile
+++ b/src/gallium/drivers/llvmpipe/Makefile
@@ -37,6 +37,7 @@ C_SOURCES = \
 	lp_surface.c \
 	lp_tex_sample_llvm.c \
 	lp_texture.c \
+	lp_tile_image.c \
 	lp_tile_soa.c
 
 CPP_SOURCES = \
@@ -54,7 +55,7 @@ testprogs := lp_test_format	\
 
 LIBS += $(GL_LIB_DEPS) -L. -lllvmpipe -L../../auxiliary/ -lgallium
 
-$(testprogs): lp_test_% : lp_test_%.o lp_test_main.o libllvmpipe.a
-	$(LD) $(filter %.o,$^) -o $@ -Wl,--start-group  $(LIBS) -Wl,--end-group
+#$(testprogs): lp_test_% : lp_test_%.o lp_test_main.o libllvmpipe.a
+#	$(LD) $(filter %.o,$^) -o $@ -Wl,--start-group  $(LIBS) -Wl,--end-group
 
-default: $(testprogs)
+#default: $(testprogs)
diff --git a/src/gallium/drivers/llvmpipe/README b/src/gallium/drivers/llvmpipe/README
index 72d9f39658..3c3fd386b5 100644
--- a/src/gallium/drivers/llvmpipe/README
+++ b/src/gallium/drivers/llvmpipe/README
@@ -12,7 +12,11 @@ Done so far is:
    
    - depth testing
  
-   - texture sampling (not all state/formats are supported) 
+   - texture sampling
+     - 1D/2D/3D/cube maps supported
+     - all texture wrap modes supported
+     - all texture filtering modes supported
+     - perhaps not all texture formats yet supported
    
    - fragment shader TGSI translation
      - same level of support as the TGSI SSE2 exec machine, with the exception
@@ -37,8 +41,6 @@ To do (probably by this order):
 
  - code generate stipple and stencil testing
 
- - translate the remaining bits of texture sampling state
-
  - translate TGSI control flow instructions, and all other remaining opcodes
  
  - integrate with the draw module for VS code generation
@@ -49,8 +51,6 @@ To do (probably by this order):
 Requirements
 ============
 
- - Linux
- 
  - A x86 or amd64 processor.  64bit mode is preferred.
  
    Support for sse2 is strongly encouraged.  Support for ssse3, and sse4.1 will
@@ -59,7 +59,7 @@ Requirements
    
    See /proc/cpuinfo to know what your CPU supports.
  
- - LLVM 2.6.
+ - LLVM 2.6 (or later)
  
    For Linux, on a recent Debian based distribution do:
  
@@ -69,6 +69,9 @@ Requirements
    http://people.freedesktop.org/~jrfonseca/llvm/ and set the LLVM environment
    variable to the extracted path.
 
+   The version of LLVM from SVN ("2.7svn") from mid-March 2010 seems pretty
+   stable and has some features not in version 2.6.
+
  - scons (optional)
 
  - udis86, http://udis86.sourceforge.net/ (optional):
@@ -86,7 +89,7 @@ Building
 
 To build everything on Linux invoke scons as:
 
-  scons debug=yes statetrackers=mesa drivers=trace,llvmpipe winsys=xlib dri=false
+  scons debug=yes statetrackers=mesa drivers=llvmpipe winsys=xlib dri=false
 
 Alternatively, you can build it with GNU make, if you prefer, by invoking it as
 
@@ -96,7 +99,7 @@ but the rest of these instructions assume that scons is used.
 
 For windows is everything the except except the winsys:
 
-  scons debug=yes statetrackers=mesa drivers=trace,llvmpipe winsys=gdi dri=false
+  scons debug=yes statetrackers=mesa drivers=llvmpipe winsys=gdi dri=false
 
 Using
 =====
@@ -142,11 +145,13 @@ Development Notes
   then skim through the lp_bld_* functions called in there, and the comments
   at the top of the lp_bld_*.c functions.  
 
-- All lp_bld_*.[ch] are isolated from the rest of the driver, and could/may be 
-  put in a stand-alone Gallium state -> LLVM IR translation module.
+- The driver-independent parts of the LLVM / Gallium code are found in
+  src/gallium/auxiliary/gallivm/.  The filenames and function prefixes
+  need to be renamed from "lp_bld_" to something else though.
 
 - We use LLVM-C bindings for now. They are not documented, but follow the C++
   interfaces very closely, and appear to be complete enough for code
   generation. See 
   http://npcontemplation.blogspot.com/2008/06/secret-of-llvm-c-bindings.html
   for a stand-alone example.
+  See the llvm-c/Core.h file for reference.
diff --git a/src/gallium/drivers/llvmpipe/SConscript b/src/gallium/drivers/llvmpipe/SConscript
index a39283e5e8..13c1a13e87 100644
--- a/src/gallium/drivers/llvmpipe/SConscript
+++ b/src/gallium/drivers/llvmpipe/SConscript
@@ -18,6 +18,13 @@ env.CodeGenerate(
 	command = 'python $SCRIPT $SOURCE > $TARGET'
 )
 
+# XXX: Our dependency scanner only finds depended modules in relative dirs.
+env.Depends('lp_tile_soa.c', [
+    '#src/gallium/auxiliary/util/u_format_parse.py', 
+    '#src/gallium/auxiliary/util/u_format_pack.py', 
+    '#src/gallium/auxiliary/util/u_format_access.py',
+])
+
 llvmpipe = env.ConvenienceLibrary(
 	target = 'llvmpipe',
 	source = [
@@ -52,6 +59,7 @@ llvmpipe = env.ConvenienceLibrary(
 		'lp_surface.c',
 		'lp_tex_sample_llvm.c',
 		'lp_texture.c',
+		'lp_tile_image.c',
 		'lp_tile_soa.c',
 	])
 
diff --git a/src/gallium/drivers/llvmpipe/lp_buffer.c b/src/gallium/drivers/llvmpipe/lp_buffer.c
index 9eda972081..6e0f37393e 100644
--- a/src/gallium/drivers/llvmpipe/lp_buffer.c
+++ b/src/gallium/drivers/llvmpipe/lp_buffer.c
@@ -30,7 +30,6 @@
 #include "util/u_memory.h"
 #include "util/u_math.h"
 
-#include "lp_winsys.h"
 #include "lp_screen.h"
 #include "lp_buffer.h"
 
diff --git a/src/gallium/drivers/llvmpipe/lp_context.c b/src/gallium/drivers/llvmpipe/lp_context.c
index 3edc62d0c6..e0676c80a2 100644
--- a/src/gallium/drivers/llvmpipe/lp_context.c
+++ b/src/gallium/drivers/llvmpipe/lp_context.c
@@ -147,6 +147,10 @@ llvmpipe_create_context( struct pipe_screen *screen, void *priv )
    llvmpipe->pipe.bind_vs_state   = llvmpipe_bind_vs_state;
    llvmpipe->pipe.delete_vs_state = llvmpipe_delete_vs_state;
 
+   llvmpipe->pipe.create_vertex_elements_state = llvmpipe_create_vertex_elements_state;
+   llvmpipe->pipe.bind_vertex_elements_state = llvmpipe_bind_vertex_elements_state;
+   llvmpipe->pipe.delete_vertex_elements_state = llvmpipe_delete_vertex_elements_state;
+
    llvmpipe->pipe.set_blend_color = llvmpipe_set_blend_color;
    llvmpipe->pipe.set_stencil_ref = llvmpipe_set_stencil_ref;
    llvmpipe->pipe.set_clip_state = llvmpipe_set_clip_state;
@@ -159,7 +163,6 @@ llvmpipe_create_context( struct pipe_screen *screen, void *priv )
    llvmpipe->pipe.set_viewport_state = llvmpipe_set_viewport_state;
 
    llvmpipe->pipe.set_vertex_buffers = llvmpipe_set_vertex_buffers;
-   llvmpipe->pipe.set_vertex_elements = llvmpipe_set_vertex_elements;
 
    llvmpipe->pipe.draw_arrays = llvmpipe_draw_arrays;
    llvmpipe->pipe.draw_elements = llvmpipe_draw_elements;
@@ -172,6 +175,7 @@ llvmpipe_create_context( struct pipe_screen *screen, void *priv )
    llvmpipe->pipe.is_buffer_referenced = llvmpipe_is_buffer_referenced;
 
    llvmpipe_init_query_funcs( llvmpipe );
+   llvmpipe_init_context_texture_funcs( &llvmpipe->pipe );
 
    /*
     * Create drawing context and plug our rendering stage into it.
@@ -185,7 +189,7 @@ llvmpipe_create_context( struct pipe_screen *screen, void *priv )
    if (debug_get_bool_option( "LP_NO_RAST", FALSE ))
       llvmpipe->no_rast = TRUE;
 
-   llvmpipe->setup = lp_setup_create( screen,
+   llvmpipe->setup = lp_setup_create( &llvmpipe->pipe,
                                       llvmpipe->draw );
    if (!llvmpipe->setup)
       goto fail;
diff --git a/src/gallium/drivers/llvmpipe/lp_context.h b/src/gallium/drivers/llvmpipe/lp_context.h
index 955c7eb8e0..f391871b0e 100644
--- a/src/gallium/drivers/llvmpipe/lp_context.h
+++ b/src/gallium/drivers/llvmpipe/lp_context.h
@@ -45,7 +45,8 @@ struct draw_stage;
 struct lp_fragment_shader;
 struct lp_vertex_shader;
 struct lp_blend_state;
-struct setup_context;
+struct lp_setup_context;
+struct lp_velems_state;
 
 struct llvmpipe_context {
    struct pipe_context pipe;  /**< base class */
@@ -58,6 +59,7 @@ struct llvmpipe_context {
    const struct pipe_rasterizer_state *rasterizer;
    struct lp_fragment_shader *fs;
    const struct lp_vertex_shader *vs;
+   const struct lp_velems_state *velems;
 
    /** Other rendering state */
    struct pipe_blend_color blend_color;
@@ -71,13 +73,11 @@ struct llvmpipe_context {
    struct pipe_texture *vertex_textures[PIPE_MAX_VERTEX_SAMPLERS];
    struct pipe_viewport_state viewport;
    struct pipe_vertex_buffer vertex_buffer[PIPE_MAX_ATTRIBS];
-   struct pipe_vertex_element vertex_element[PIPE_MAX_ATTRIBS];
 
    unsigned num_samplers;
    unsigned num_textures;
    unsigned num_vertex_samplers;
    unsigned num_vertex_textures;
-   unsigned num_vertex_elements;
    unsigned num_vertex_buffers;
 
    unsigned dirty; /**< Mask of LP_NEW_x flags */
@@ -98,7 +98,7 @@ struct llvmpipe_context {
    int psize_slot;
 
    /** The tiling engine */
-   struct setup_context *setup;
+   struct lp_setup_context *setup;
 
    /** The primitive drawing context */
    struct draw_context *draw;
diff --git a/src/gallium/drivers/llvmpipe/lp_flush.c b/src/gallium/drivers/llvmpipe/lp_flush.c
index bf832433be..636d72a9bb 100644
--- a/src/gallium/drivers/llvmpipe/lp_flush.c
+++ b/src/gallium/drivers/llvmpipe/lp_flush.c
@@ -79,12 +79,12 @@ llvmpipe_flush( struct pipe_context *pipe,
 
       for (i = 0; i < llvmpipe->framebuffer.nr_cbufs; i++) {
 	 util_snprintf(filename, sizeof(filename), "cbuf%u_%u", i, frame_no);
-         debug_dump_surface(filename, llvmpipe->framebuffer.cbufs[i]);
+         debug_dump_surface_bmp(&llvmpipe->pipe, filename, llvmpipe->framebuffer.cbufs[0]);
       }
 
       if (0) {
          util_snprintf(filename, sizeof(filename), "zsbuf_%u", frame_no);
-         debug_dump_surface(filename, llvmpipe->framebuffer.zsbuf);
+         debug_dump_surface_bmp(&llvmpipe->pipe, filename, llvmpipe->framebuffer.zsbuf);
       }
 
       ++frame_no;
@@ -92,3 +92,68 @@ llvmpipe_flush( struct pipe_context *pipe,
 #endif
 }
 
+
+/**
+ * Flush context if necessary.
+ *
+ * TODO: move this logic to an auxiliary library?
+ *
+ * FIXME: We must implement DISCARD/DONTBLOCK/UNSYNCHRONIZED/etc for
+ * textures to avoid blocking.
+ */
+boolean
+llvmpipe_flush_texture(struct pipe_context *pipe,
+                       struct pipe_texture *texture,
+                       unsigned face,
+                       unsigned level,
+                       unsigned flush_flags,
+                       boolean read_only,
+                       boolean cpu_access,
+                       boolean do_not_flush)
+{
+   struct pipe_fence_handle *last_fence = NULL;
+   unsigned referenced;
+
+   referenced = pipe->is_texture_referenced(pipe, texture, face, level);
+
+   if ((referenced & PIPE_REFERENCED_FOR_WRITE) ||
+       ((referenced & PIPE_REFERENCED_FOR_READ) && !read_only)) {
+
+      if (do_not_flush)
+         return FALSE;
+
+      /*
+       * TODO: The semantics of these flush flags are too obtuse. They should
+       * disappear and the pipe driver should just ensure that all visible
+       * side-effects happen when they need to happen.
+       */
+      if (referenced & PIPE_REFERENCED_FOR_WRITE)
+         flush_flags |= PIPE_FLUSH_RENDER_CACHE;
+
+      if (referenced & PIPE_REFERENCED_FOR_READ)
+         flush_flags |= PIPE_FLUSH_TEXTURE_CACHE;
+
+      if (cpu_access) {
+         /*
+          * Flush and wait.
+          */
+
+         struct pipe_fence_handle *fence = NULL;
+
+         pipe->flush(pipe, flush_flags, &fence);
+
+         if (last_fence) {
+            pipe->screen->fence_finish(pipe->screen, fence, 0);
+            pipe->screen->fence_reference(pipe->screen, &fence, NULL);
+         }
+      } else {
+         /*
+          * Just flush.
+          */
+
+         pipe->flush(pipe, flush_flags, NULL);
+      }
+   }
+
+   return TRUE;
+}
diff --git a/src/gallium/drivers/llvmpipe/lp_flush.h b/src/gallium/drivers/llvmpipe/lp_flush.h
index 10b2b52583..e13f57ccec 100644
--- a/src/gallium/drivers/llvmpipe/lp_flush.h
+++ b/src/gallium/drivers/llvmpipe/lp_flush.h
@@ -28,10 +28,22 @@
 #ifndef LP_FLUSH_H
 #define LP_FLUSH_H
 
+#include "pipe/p_compiler.h"
+
 struct pipe_context;
 struct pipe_fence_handle;
 
 void llvmpipe_flush(struct pipe_context *pipe, unsigned flags,
                     struct pipe_fence_handle **fence);
 
+boolean
+llvmpipe_flush_texture(struct pipe_context *pipe,
+                       struct pipe_texture *texture,
+                       unsigned face,
+                       unsigned level,
+                       unsigned flush_flags,
+                       boolean read_only,
+                       boolean cpu_access,
+                       boolean do_not_flush);
+
 #endif
diff --git a/src/gallium/drivers/llvmpipe/lp_jit.c b/src/gallium/drivers/llvmpipe/lp_jit.c
index d3d7e26882..5887613120 100644
--- a/src/gallium/drivers/llvmpipe/lp_jit.c
+++ b/src/gallium/drivers/llvmpipe/lp_jit.c
@@ -37,6 +37,7 @@
 
 #include "util/u_memory.h"
 #include "util/u_cpu_detect.h"
+#include "gallivm/lp_bld_init.h"
 #include "lp_debug.h"
 #include "lp_screen.h"
 #include "gallivm/lp_bld_intr.h"
@@ -50,12 +51,17 @@ lp_jit_init_globals(struct llvmpipe_screen *screen)
 
    /* struct lp_jit_texture */
    {
-      LLVMTypeRef elem_types[4];
+      LLVMTypeRef elem_types[6];
 
       elem_types[LP_JIT_TEXTURE_WIDTH]  = LLVMInt32Type();
       elem_types[LP_JIT_TEXTURE_HEIGHT] = LLVMInt32Type();
-      elem_types[LP_JIT_TEXTURE_STRIDE] = LLVMInt32Type();
-      elem_types[LP_JIT_TEXTURE_DATA]   = LLVMPointerType(LLVMInt8Type(), 0);
+      elem_types[LP_JIT_TEXTURE_DEPTH] = LLVMInt32Type();
+      elem_types[LP_JIT_TEXTURE_LAST_LEVEL] = LLVMInt32Type();
+      elem_types[LP_JIT_TEXTURE_ROW_STRIDE] =
+         LLVMArrayType(LLVMInt32Type(), LP_MAX_TEXTURE_2D_LEVELS);
+      elem_types[LP_JIT_TEXTURE_DATA] =
+         LLVMArrayType(LLVMPointerType(LLVMInt8Type(), 0),
+                       LP_MAX_TEXTURE_2D_LEVELS);
 
       texture_type = LLVMStructType(elem_types, Elements(elem_types), 0);
 
@@ -65,9 +71,15 @@ lp_jit_init_globals(struct llvmpipe_screen *screen)
       LP_CHECK_MEMBER_OFFSET(struct lp_jit_texture, height,
                              screen->target, texture_type,
                              LP_JIT_TEXTURE_HEIGHT);
-      LP_CHECK_MEMBER_OFFSET(struct lp_jit_texture, stride,
+      LP_CHECK_MEMBER_OFFSET(struct lp_jit_texture, depth,
                              screen->target, texture_type,
-                             LP_JIT_TEXTURE_STRIDE);
+                             LP_JIT_TEXTURE_DEPTH);
+      LP_CHECK_MEMBER_OFFSET(struct lp_jit_texture, last_level,
+                             screen->target, texture_type,
+                             LP_JIT_TEXTURE_LAST_LEVEL);
+      LP_CHECK_MEMBER_OFFSET(struct lp_jit_texture, row_stride,
+                             screen->target, texture_type,
+                             LP_JIT_TEXTURE_ROW_STRIDE);
       LP_CHECK_MEMBER_OFFSET(struct lp_jit_texture, data,
                              screen->target, texture_type,
                              LP_JIT_TEXTURE_DATA);
@@ -148,8 +160,7 @@ lp_jit_screen_init(struct llvmpipe_screen *screen)
    util_cpu_caps.has_sse4_1 = 0;
 #endif
 
-   LLVMLinkInJIT();
-   LLVMInitializeNativeTarget();
+   lp_build_init();
 
    screen->module = LLVMModuleCreateWithName("llvmpipe");
 
diff --git a/src/gallium/drivers/llvmpipe/lp_jit.h b/src/gallium/drivers/llvmpipe/lp_jit.h
index 8df3015d4b..13167ae3bf 100644
--- a/src/gallium/drivers/llvmpipe/lp_jit.h
+++ b/src/gallium/drivers/llvmpipe/lp_jit.h
@@ -39,6 +39,7 @@
 #include "gallivm/lp_bld_struct.h"
 
 #include "pipe/p_state.h"
+#include "lp_texture.h"
 
 
 struct llvmpipe_screen;
@@ -48,15 +49,19 @@ struct lp_jit_texture
 {
    uint32_t width;
    uint32_t height;
-   uint32_t stride;
-   const void *data;
+   uint32_t depth;
+   uint32_t last_level;
+   uint32_t row_stride[LP_MAX_TEXTURE_2D_LEVELS];
+   const void *data[LP_MAX_TEXTURE_2D_LEVELS];
 };
 
 
 enum {
    LP_JIT_TEXTURE_WIDTH = 0,
    LP_JIT_TEXTURE_HEIGHT,
-   LP_JIT_TEXTURE_STRIDE,
+   LP_JIT_TEXTURE_DEPTH,
+   LP_JIT_TEXTURE_LAST_LEVEL,
+   LP_JIT_TEXTURE_ROW_STRIDE,
    LP_JIT_TEXTURE_DATA
 };
 
diff --git a/src/gallium/drivers/llvmpipe/lp_public.h b/src/gallium/drivers/llvmpipe/lp_public.h
new file mode 100644
index 0000000000..ec6b660b48
--- /dev/null
+++ b/src/gallium/drivers/llvmpipe/lp_public.h
@@ -0,0 +1,10 @@
+#ifndef LP_PUBLIC_H
+#define LP_PUBLIC_H
+
+struct pipe_screen;
+struct sw_winsys;
+
+struct pipe_screen *
+llvmpipe_create_screen(struct sw_winsys *winsys);
+
+#endif
diff --git a/src/gallium/drivers/llvmpipe/lp_rast.c b/src/gallium/drivers/llvmpipe/lp_rast.c
index 6dbcb3c9b3..81ea11a16b 100644
--- a/src/gallium/drivers/llvmpipe/lp_rast.c
+++ b/src/gallium/drivers/llvmpipe/lp_rast.c
@@ -42,128 +42,75 @@
 #include "lp_scene.h"
 
 
-/**
- * Begin the rasterization phase.
- * Map the framebuffer surfaces.  Initialize the 'rast' state.
+/* Begin rasterizing a scene:
  */
 static boolean
 lp_rast_begin( struct lp_rasterizer *rast,
-               const struct pipe_framebuffer_state *fb,
-               boolean write_color,
-               boolean write_zstencil )
+               struct lp_scene *scene )
 {
-   struct pipe_screen *screen = rast->screen;
-   struct pipe_surface *cbuf, *zsbuf;
+   const struct pipe_framebuffer_state *fb = &scene->fb;
+   boolean write_color = fb->nr_cbufs != 0;
+   boolean write_zstencil = fb->zsbuf != NULL;
    int i;
 
-   LP_DBG(DEBUG_RAST, "%s\n", __FUNCTION__);
+   rast->curr_scene = scene;
 
-   util_copy_framebuffer_state(&rast->state.fb, fb);
+   LP_DBG(DEBUG_RAST, "%s\n", __FUNCTION__);
 
+   rast->state.nr_cbufs = scene->fb.nr_cbufs;
    rast->state.write_zstencil = write_zstencil;
    rast->state.write_color = write_color;
-
-   rast->check_for_clipped_tiles = (fb->width % TILE_SIZE != 0 ||
-                                    fb->height % TILE_SIZE != 0);
-
    
-   for (i = 0; i < rast->state.fb.nr_cbufs; i++) {
-      cbuf = rast->state.fb.cbufs[i];
-      if (cbuf) {
-	 rast->cbuf_transfer[i] = screen->get_tex_transfer(rast->screen,
-							   cbuf->texture,
-							   cbuf->face,
-							   cbuf->level,
-							   cbuf->zslice,
-							   PIPE_TRANSFER_READ_WRITE,
-							   0, 0,
-							   cbuf->width, 
-							   cbuf->height);
-	 if (!rast->cbuf_transfer[i])
-	    goto fail;
-
-	 rast->cbuf_map[i] = screen->transfer_map(rast->screen, 
-						  rast->cbuf_transfer[i]);
-	 if (!rast->cbuf_map[i])
-	    goto fail;
-      }
+   for (i = 0; i < rast->state.nr_cbufs; i++) {
+      struct pipe_surface *cbuf = scene->fb.cbufs[i];
+      rast->cbuf[i].map = scene->cbuf_map[i];
+      rast->cbuf[i].format = cbuf->texture->format;
+      rast->cbuf[i].width = cbuf->width;
+      rast->cbuf[i].height = cbuf->height;
+      rast->cbuf[i].stride = llvmpipe_texture_stride(cbuf->texture, cbuf->level);
    }
 
-   zsbuf = rast->state.fb.zsbuf;
-   if (zsbuf) {
-      rast->zsbuf_transfer = screen->get_tex_transfer(rast->screen,
-                                                      zsbuf->texture,
-                                                      zsbuf->face,
-                                                      zsbuf->level,
-                                                      zsbuf->zslice,
-                                                      PIPE_TRANSFER_READ_WRITE,
-                                                      0, 0,
-                                                      zsbuf->width,
-						      zsbuf->height);
-      if (!rast->zsbuf_transfer)
-         goto fail;
-
-      rast->zsbuf_map = screen->transfer_map(rast->screen, 
-                                            rast->zsbuf_transfer);
-      if (!rast->zsbuf_map)
-	 goto fail;
+   if (write_zstencil) {
+      struct pipe_surface *zsbuf = scene->fb.zsbuf;
+      rast->zsbuf.map = scene->zsbuf_map;
+      rast->zsbuf.stride = llvmpipe_texture_stride(zsbuf->texture, zsbuf->level);
+      rast->zsbuf.blocksize = 
+         util_format_get_blocksize(zsbuf->texture->format);
    }
 
+   lp_scene_bin_iter_begin( scene );
+   
    return TRUE;
-
-fail:
-   /* Unmap and release transfers?
-    */
-   return FALSE;
 }
 
 
-/**
- * Finish the rasterization phase.
- * Unmap framebuffer surfaces.
- */
 static void
 lp_rast_end( struct lp_rasterizer *rast )
 {
-   struct pipe_screen *screen = rast->screen;
-   unsigned i;
-
-   for (i = 0; i < rast->state.fb.nr_cbufs; i++) {
-      if (rast->cbuf_map[i]) 
-	 screen->transfer_unmap(screen, rast->cbuf_transfer[i]);
-
-      if (rast->cbuf_transfer[i])
-	 screen->tex_transfer_destroy(rast->cbuf_transfer[i]);
-
-      rast->cbuf_transfer[i] = NULL;
-      rast->cbuf_map[i] = NULL;
-   }
+   int i;
 
-   if (rast->zsbuf_map) 
-      screen->transfer_unmap(screen, rast->zsbuf_transfer);
+   lp_scene_reset( rast->curr_scene );
 
-   if (rast->zsbuf_transfer)
-      screen->tex_transfer_destroy(rast->zsbuf_transfer);
+   for (i = 0; i < rast->state.nr_cbufs; i++)
+      rast->cbuf[i].map = NULL;
 
-   rast->zsbuf_transfer = NULL;
-   rast->zsbuf_map = NULL;
+   rast->zsbuf.map = NULL;
+   rast->curr_scene = NULL;
 }
 
-
 /**
  * Begining rasterization of a tile.
  * \param x  window X position of the tile, in pixels
  * \param y  window Y position of the tile, in pixels
  */
 static void
-lp_rast_start_tile( struct lp_rasterizer *rast,
-                    unsigned thread_index,
-                    unsigned x, unsigned y )
+lp_rast_start_tile(struct lp_rasterizer_task *task,
+                   unsigned x, unsigned y)
 {
    LP_DBG(DEBUG_RAST, "%s %d,%d\n", __FUNCTION__, x, y);
 
-   rast->tasks[thread_index].x = x;
-   rast->tasks[thread_index].y = y;
+   task->x = x;
+   task->y = y;
 }
 
 
@@ -171,12 +118,13 @@ lp_rast_start_tile( struct lp_rasterizer *rast,
  * Clear the rasterizer's current color tile.
  * This is a bin command called during bin processing.
  */
-void lp_rast_clear_color( struct lp_rasterizer *rast,
-                          unsigned thread_index,
-                          const union lp_rast_cmd_arg arg )
+void
+lp_rast_clear_color(struct lp_rasterizer_task *task,
+                    const union lp_rast_cmd_arg arg)
 {
+   struct lp_rasterizer *rast = task->rast;
    const uint8_t *clear_color = arg.clear_color;
-   uint8_t **color_tile = rast->tasks[thread_index].tile.color;
+   uint8_t **color_tile = task->tile.color;
    unsigned i;
 
    LP_DBG(DEBUG_RAST, "%s 0x%x,0x%x,0x%x,0x%x\n", __FUNCTION__, 
@@ -189,7 +137,7 @@ void lp_rast_clear_color( struct lp_rasterizer *rast,
        clear_color[1] == clear_color[2] &&
        clear_color[2] == clear_color[3]) {
       /* clear to grayscale value {x, x, x, x} */
-      for (i = 0; i < rast->state.fb.nr_cbufs; i++) {
+      for (i = 0; i < rast->state.nr_cbufs; i++) {
 	 memset(color_tile[i], clear_color[0], TILE_SIZE * TILE_SIZE * 4);
       }
    }
@@ -200,7 +148,7 @@ void lp_rast_clear_color( struct lp_rasterizer *rast,
        * works.
        */
       const unsigned chunk = TILE_SIZE / 4;
-      for (i = 0; i < rast->state.fb.nr_cbufs; i++) {
+      for (i = 0; i < rast->state.nr_cbufs; i++) {
          uint8_t *c = color_tile[i];
          unsigned j;
          for (j = 0; j < 4 * TILE_SIZE; j++) {
@@ -225,24 +173,24 @@ void lp_rast_clear_color( struct lp_rasterizer *rast,
  * Clear the rasterizer's current z/stencil tile.
  * This is a bin command called during bin processing.
  */
-void lp_rast_clear_zstencil( struct lp_rasterizer *rast,
-                             unsigned thread_index,
-                             const union lp_rast_cmd_arg arg)
+void
+lp_rast_clear_zstencil(struct lp_rasterizer_task *task,
+                       const union lp_rast_cmd_arg arg)
 {
-   struct lp_rasterizer_task *task = &rast->tasks[thread_index];
+   struct lp_rasterizer *rast = task->rast;
    const unsigned tile_x = task->x;
    const unsigned tile_y = task->y;
-   const unsigned height = TILE_SIZE/TILE_VECTOR_HEIGHT;
-   const unsigned width = TILE_SIZE*TILE_VECTOR_HEIGHT;
-   unsigned block_size = util_format_get_blocksize(rast->zsbuf_transfer->texture->format);
+   const unsigned height = TILE_SIZE / TILE_VECTOR_HEIGHT;
+   const unsigned width = TILE_SIZE * TILE_VECTOR_HEIGHT;
+   unsigned block_size = rast->zsbuf.blocksize;
    uint8_t *dst;
-   unsigned dst_stride = rast->zsbuf_transfer->stride*TILE_VECTOR_HEIGHT;
+   unsigned dst_stride = rast->zsbuf.stride * TILE_VECTOR_HEIGHT;
    unsigned i, j;
 
    LP_DBG(DEBUG_RAST, "%s 0x%x\n", __FUNCTION__, arg.clear_zstencil);
 
-   assert(rast->zsbuf_map);
-   if (!rast->zsbuf_map)
+   assert(rast->zsbuf.map);
+   if (!rast->zsbuf.map)
       return;
 
    LP_DBG(DEBUG_RAST, "%s\n", __FUNCTION__);
@@ -278,8 +226,8 @@ void lp_rast_clear_zstencil( struct lp_rasterizer *rast,
       }
       break;
    default:
-         assert(0);
-         break;
+      assert(0);
+      break;
    }
 }
 
@@ -288,55 +236,42 @@ void lp_rast_clear_zstencil( struct lp_rasterizer *rast,
  * Load tile color from the framebuffer surface.
  * This is a bin command called during bin processing.
  */
-void lp_rast_load_color( struct lp_rasterizer *rast,
-                         unsigned thread_index,
-                         const union lp_rast_cmd_arg arg)
+void
+lp_rast_load_color(struct lp_rasterizer_task *task,
+                   const union lp_rast_cmd_arg arg)
 {
-   struct lp_rasterizer_task *task = &rast->tasks[thread_index];
-   const unsigned x = task->x;
-   const unsigned y = task->y;
+   struct lp_rasterizer *rast = task->rast;
+   const unsigned x = task->x, y = task->y;
    unsigned i;
 
    LP_DBG(DEBUG_RAST, "%s at %u, %u\n", __FUNCTION__, x, y);
 
-   for (i = 0; i < rast->state.fb.nr_cbufs; i++) {
-      struct pipe_transfer *transfer = rast->cbuf_transfer[i];
-      int w = TILE_SIZE;
-      int h = TILE_SIZE;
-
-      if (x >= transfer->width)
-	 continue;
-
-      if (y >= transfer->height)
+   for (i = 0; i < rast->state.nr_cbufs; i++) {
+      if (x >= rast->cbuf[i].width || y >= rast->cbuf[i].height)
 	 continue;
 
-      assert(w >= 0);
-      assert(h >= 0);
-      assert(w <= TILE_SIZE);
-      assert(h <= TILE_SIZE);
-
-      lp_tile_read_4ub(transfer->texture->format,
+      lp_tile_read_4ub(rast->cbuf[i].format,
 		       task->tile.color[i],
-		       rast->cbuf_map[i], 
-		       transfer->stride,
+		       rast->cbuf[i].map, 
+		       rast->cbuf[i].stride,
 		       x, y,
-		       w, h);
+		       TILE_SIZE, TILE_SIZE);
 
       LP_COUNT(nr_color_tile_load);
    }
 }
 
 
-void lp_rast_set_state( struct lp_rasterizer *rast,
-                        unsigned thread_index,
-                        const union lp_rast_cmd_arg arg )
+void
+lp_rast_set_state(struct lp_rasterizer_task *task,
+                  const union lp_rast_cmd_arg arg)
 {
    const struct lp_rast_state *state = arg.set_state;
 
    LP_DBG(DEBUG_RAST, "%s %p\n", __FUNCTION__, (void *) state);
 
    /* just set the current state pointer for this rasterizer */
-   rast->tasks[thread_index].current_state = state;
+   task->current_state = state;
 }
 
 
@@ -346,16 +281,15 @@ void lp_rast_set_state( struct lp_rasterizer *rast,
  * completely contained inside a triangle.
  * This is a bin command called during bin processing.
  */
-void lp_rast_shade_tile( struct lp_rasterizer *rast,
-                         unsigned thread_index,
-                         const union lp_rast_cmd_arg arg )
+void
+lp_rast_shade_tile(struct lp_rasterizer_task *task,
+                   const union lp_rast_cmd_arg arg)
 {
-   struct lp_rasterizer_task *task = &rast->tasks[thread_index];
+   struct lp_rasterizer *rast = task->rast;
    const struct lp_rast_state *state = task->current_state;
    struct lp_rast_tile *tile = &task->tile;
    const struct lp_rast_shader_inputs *inputs = arg.shade_tile;
-   const unsigned tile_x = task->x;
-   const unsigned tile_y = task->y;
+   const unsigned tile_x = task->x, tile_y = task->y;
    unsigned x, y;
 
    LP_DBG(DEBUG_RAST, "%s\n", __FUNCTION__);
@@ -371,7 +305,7 @@ void lp_rast_shade_tile( struct lp_rasterizer *rast,
          block_offset = ((y / 4) * (16 * 16) + (x / 4) * 16);
 
          /* color buffer */
-         for (i = 0; i < rast->state.fb.nr_cbufs; i++)
+         for (i = 0; i < rast->state.nr_cbufs; i++)
             color[i] = tile->color[i] + 4 * block_offset;
 
          /* depth buffer */
@@ -396,14 +330,13 @@ void lp_rast_shade_tile( struct lp_rasterizer *rast,
  * Compute shading for a 4x4 block of pixels.
  * This is a bin command called during bin processing.
  */
-void lp_rast_shade_quads( struct lp_rasterizer *rast,
-                          unsigned thread_index,
+void lp_rast_shade_quads( struct lp_rasterizer_task *task,
                           const struct lp_rast_shader_inputs *inputs,
                           unsigned x, unsigned y,
                           int32_t c1, int32_t c2, int32_t c3)
 {
-   struct lp_rasterizer_task *task = &rast->tasks[thread_index];
    const struct lp_rast_state *state = task->current_state;
+   struct lp_rasterizer *rast = task->rast;
    struct lp_rast_tile *tile = &task->tile;
    uint8_t *color[PIPE_MAX_COLOR_BUFS];
    void *depth;
@@ -411,7 +344,6 @@ void lp_rast_shade_quads( struct lp_rasterizer *rast,
    unsigned ix, iy;
    int block_offset;
 
-#ifdef DEBUG
    assert(state);
 
    /* Sanity checks */
@@ -420,7 +352,6 @@ void lp_rast_shade_quads( struct lp_rasterizer *rast,
 
    assert((x % 4) == 0);
    assert((y % 4) == 0);
-#endif
 
    ix = x % TILE_SIZE;
    iy = y % TILE_SIZE;
@@ -429,22 +360,19 @@ void lp_rast_shade_quads( struct lp_rasterizer *rast,
    block_offset = ((iy / 4) * (16 * 16) + (ix / 4) * 16);
 
    /* color buffer */
-   for (i = 0; i < rast->state.fb.nr_cbufs; i++)
+   for (i = 0; i < rast->state.nr_cbufs; i++)
       color[i] = tile->color[i] + 4 * block_offset;
 
    /* depth buffer */
    depth = lp_rast_depth_pointer(rast, x, y);
 
 
-
-#ifdef DEBUG
    assert(lp_check_alignment(tile->color[0], 16));
    assert(lp_check_alignment(state->jit_context.blend_color, 16));
 
    assert(lp_check_alignment(inputs->step[0], 16));
    assert(lp_check_alignment(inputs->step[1], 16));
    assert(lp_check_alignment(inputs->step[2], 16));
-#endif
 
    /* run shader */
    state->jit_function[1]( &state->jit_context,
@@ -515,66 +443,48 @@ outline_subtiles(uint8_t *tile)
 /**
  * Write the rasterizer's color tile to the framebuffer.
  */
-static void lp_rast_store_color( struct lp_rasterizer *rast,
-                                 unsigned thread_index)
+static void
+lp_rast_store_color(struct lp_rasterizer_task *task)
 {
-   struct lp_rasterizer_task *task = &rast->tasks[thread_index];
-   const unsigned x = task->x;
-   const unsigned y = task->y;
+   struct lp_rasterizer *rast = task->rast;
+   const unsigned x = task->x, y = task->y;
    unsigned i;
 
-   for (i = 0; i < rast->state.fb.nr_cbufs; i++) {
-      struct pipe_transfer *transfer = rast->cbuf_transfer[i];
-      int w = TILE_SIZE;
-      int h = TILE_SIZE;
-
-      if (x >= transfer->width)
+   for (i = 0; i < rast->state.nr_cbufs; i++) {
+      if (x >= rast->cbuf[i].width)
 	 continue;
 
-      if (y >= transfer->height)
+      if (y >= rast->cbuf[i].height)
 	 continue;
 
-      LP_DBG(DEBUG_RAST, "%s [%u] %d,%d %dx%d\n", __FUNCTION__,
-	     thread_index, x, y, w, h);
+      LP_DBG(DEBUG_RAST, "%s [%u] %d,%d\n", __FUNCTION__,
+	     task->thread_index, x, y);
 
       if (LP_DEBUG & DEBUG_SHOW_SUBTILES)
          outline_subtiles(task->tile.color[i]);
       else if (LP_DEBUG & DEBUG_SHOW_TILES)
          outline_tile(task->tile.color[i]);
 
-      lp_tile_write_4ub(transfer->texture->format,
+      lp_tile_write_4ub(rast->cbuf[i].format,
 			task->tile.color[i],
-			rast->cbuf_map[i], 
-			transfer->stride,
+			rast->cbuf[i].map, 
+			rast->cbuf[i].stride,
 			x, y,
-			w, h);
+			TILE_SIZE, TILE_SIZE);
 
       LP_COUNT(nr_color_tile_store);
    }
 }
 
 
-/**
- * Write the rasterizer's tiles to the framebuffer.
- */
-static void
-lp_rast_end_tile( struct lp_rasterizer *rast,
-                  unsigned thread_index )
-{
-   LP_DBG(DEBUG_RAST, "%s\n", __FUNCTION__);
-
-   if (rast->state.write_color)
-      lp_rast_store_color(rast, thread_index);
-}
-
 
 /**
  * Signal on a fence.  This is called during bin execution/rasterization.
  * Called per thread.
  */
-void lp_rast_fence( struct lp_rasterizer *rast,
-                    unsigned thread_index,
-                    const union lp_rast_cmd_arg arg )
+void
+lp_rast_fence(struct lp_rasterizer_task *task,
+              const union lp_rast_cmd_arg arg)
 {
    struct lp_fence *fence = arg.fence;
 
@@ -592,20 +502,6 @@ void lp_rast_fence( struct lp_rasterizer *rast,
 }
 
 
-/**
- * When all the threads are done rasterizing a scene, one thread will
- * call this function to reset the scene and put it onto the empty queue.
- */
-static void
-release_scene( struct lp_rasterizer *rast,
-	       struct lp_scene *scene )
-{
-   util_unreference_framebuffer_state( &scene->fb );
-
-   lp_scene_reset( scene );
-   lp_scene_enqueue( rast->empty_scenes, scene );
-   rast->curr_scene = NULL;
-}
 
 
 /**
@@ -615,25 +511,31 @@ release_scene( struct lp_rasterizer *rast,
  * Called per thread.
  */
 static void
-rasterize_bin( struct lp_rasterizer *rast,
-               unsigned thread_index,
-               const struct cmd_bin *bin,
-               int x, int y)
+rasterize_bin(struct lp_rasterizer_task *task,
+              const struct cmd_bin *bin,
+              int x, int y)
 {
    const struct cmd_block_list *commands = &bin->commands;
    struct cmd_block *block;
    unsigned k;
 
-   lp_rast_start_tile( rast, thread_index, x, y );
+   lp_rast_start_tile( task, x * TILE_SIZE, y * TILE_SIZE );
 
    /* simply execute each of the commands in the block list */
    for (block = commands->head; block; block = block->next) {
       for (k = 0; k < block->count; k++) {
-         block->cmd[k]( rast, thread_index, block->arg[k] );
+         block->cmd[k]( task, block->arg[k] );
       }
    }
 
-   lp_rast_end_tile( rast, thread_index );
+   /* Write the rasterizer's tiles to the framebuffer.
+    */
+   if (task->rast->state.write_color)
+      lp_rast_store_color(task);
+
+   /* Free data for this bin.
+    */
+   lp_scene_bin_reset( task->rast->curr_scene, x, y);
 }
 
 
@@ -717,10 +619,8 @@ is_empty_bin( const struct cmd_bin *bin )
  * Called per thread.
  */
 static void
-rasterize_scene( struct lp_rasterizer *rast,
-                unsigned thread_index,
-                struct lp_scene *scene,
-                bool write_depth )
+rasterize_scene(struct lp_rasterizer_task *task,
+                struct lp_scene *scene)
 {
    /* loop over scene bins, rasterize each */
 #if 0
@@ -728,9 +628,8 @@ rasterize_scene( struct lp_rasterizer *rast,
       unsigned i, j;
       for (i = 0; i < scene->tiles_x; i++) {
          for (j = 0; j < scene->tiles_y; j++) {
-            struct cmd_bin *bin = lp_get_bin(scene, i, j);
-            rasterize_bin( rast, thread_index,
-                           bin, i * TILE_SIZE, j * TILE_SIZE );
+            struct cmd_bin *bin = lp_scene_get_bin(scene, i, j);
+            rasterize_bin(task, bin, i, j);
          }
       }
    }
@@ -742,7 +641,7 @@ rasterize_scene( struct lp_rasterizer *rast,
       assert(scene);
       while ((bin = lp_scene_bin_iter_next(scene, &x, &y))) {
          if (!is_empty_bin( bin ))
-            rasterize_bin( rast, thread_index, bin, x * TILE_SIZE, y * TILE_SIZE);
+            rasterize_bin(task, bin, x, y);
       }
    }
 #endif
@@ -753,44 +652,20 @@ rasterize_scene( struct lp_rasterizer *rast,
  * Called by setup module when it has something for us to render.
  */
 void
-lp_rasterize_scene( struct lp_rasterizer *rast,
-                   struct lp_scene *scene,
-                   const struct pipe_framebuffer_state *fb,
-                   bool write_depth )
+lp_rast_queue_scene( struct lp_rasterizer *rast,
+                     struct lp_scene *scene)
 {
-   boolean debug = false;
-
    LP_DBG(DEBUG_SETUP, "%s\n", __FUNCTION__);
 
-   if (debug) {
-      unsigned x, y;
-      debug_printf("rasterize scene:\n");
-      debug_printf("  data size: %u\n", lp_scene_data_size(scene));
-      for (y = 0; y < scene->tiles_y; y++) {
-         for (x = 0; x < scene->tiles_x; x++) {
-            debug_printf("  bin %u, %u size: %u\n", x, y,
-                         lp_scene_bin_size(scene, x, y));
-         }
-      }
-   }
-
-   /* save framebuffer state in the bin */
-   util_copy_framebuffer_state(&scene->fb, fb);
-   scene->write_depth = write_depth;
-
    if (rast->num_threads == 0) {
       /* no threading */
 
-      lp_rast_begin( rast, fb,
-                     fb->nr_cbufs != 0, /* always write color if cbufs present */
-                     fb->zsbuf != NULL && write_depth );
+      lp_rast_begin( rast, scene );
 
-      lp_scene_bin_iter_begin( scene );
-      rasterize_scene( rast, 0, scene, write_depth );
+      rasterize_scene( &rast->tasks[0], scene );
 
-      release_scene( rast, scene );
-
-      lp_rast_end( rast );
+      lp_scene_reset( scene );
+      rast->curr_scene = NULL;
    }
    else {
       /* threaded rendering! */
@@ -802,14 +677,26 @@ lp_rasterize_scene( struct lp_rasterizer *rast,
       for (i = 0; i < rast->num_threads; i++) {
          pipe_semaphore_signal(&rast->tasks[i].work_ready);
       }
+   }
+
+   LP_DBG(DEBUG_SETUP, "%s done \n", __FUNCTION__);
+}
+
+
+void
+lp_rast_finish( struct lp_rasterizer *rast )
+{
+   if (rast->num_threads == 0) {
+      /* nothing to do */
+   }
+   else {
+      int i;
 
       /* wait for work to complete */
       for (i = 0; i < rast->num_threads; i++) {
          pipe_semaphore_wait(&rast->tasks[i].work_done);
       }
    }
-
-   LP_DBG(DEBUG_SETUP, "%s done \n", __FUNCTION__);
 }
 
 
@@ -832,24 +719,16 @@ static PIPE_THREAD_ROUTINE( thread_func, init_data )
          debug_printf("thread %d waiting for work\n", task->thread_index);
       pipe_semaphore_wait(&task->work_ready);
 
+      if (rast->exit_flag)
+         break;
+
       if (task->thread_index == 0) {
          /* thread[0]:
           *  - get next scene to rasterize
           *  - map the framebuffer surfaces
           */
-         const struct pipe_framebuffer_state *fb;
-         boolean write_depth;
-
-         rast->curr_scene = lp_scene_dequeue( rast->full_scenes, TRUE );
-
-         lp_scene_bin_iter_begin( rast->curr_scene );
-
-         fb = &rast->curr_scene->fb;
-         write_depth = rast->curr_scene->write_depth;
-
-         lp_rast_begin( rast, fb,
-                        fb->nr_cbufs != 0,
-                        fb->zsbuf != NULL && write_depth );
+         lp_rast_begin( rast, 
+                        lp_scene_dequeue( rast->full_scenes, TRUE ) );
       }
 
       /* Wait for all threads to get here so that threads[1+] don't
@@ -860,26 +739,23 @@ static PIPE_THREAD_ROUTINE( thread_func, init_data )
       /* do work */
       if (debug)
          debug_printf("thread %d doing work\n", task->thread_index);
-      rasterize_scene(rast, 
-		     task->thread_index,
-                     rast->curr_scene, 
-		     rast->curr_scene->write_depth);
+
+      rasterize_scene(task,
+                      rast->curr_scene);
       
       /* wait for all threads to finish with this scene */
       pipe_barrier_wait( &rast->barrier );
 
+      /* XXX: shouldn't be necessary:
+       */
       if (task->thread_index == 0) {
-         /* thread[0]:
-          * - release the scene object
-          * - unmap the framebuffer surfaces
-          */
-         release_scene( rast, rast->curr_scene );
          lp_rast_end( rast );
       }
 
       /* signal done with work */
       if (debug)
          debug_printf("thread %d done working\n", task->thread_index);
+
       pipe_semaphore_signal(&task->work_done);
    }
 
@@ -922,7 +798,7 @@ create_rast_threads(struct lp_rasterizer *rast)
  *               processing them.
  */
 struct lp_rasterizer *
-lp_rast_create( struct pipe_screen *screen, struct lp_scene_queue *empty )
+lp_rast_create( void )
 {
    struct lp_rasterizer *rast;
    unsigned i, cbuf;
@@ -931,9 +807,6 @@ lp_rast_create( struct pipe_screen *screen, struct lp_scene_queue *empty )
    if(!rast)
       return NULL;
 
-   rast->screen = screen;
-
-   rast->empty_scenes = empty;
    rast->full_scenes = lp_scene_queue_create();
 
    for (i = 0; i < Elements(rast->tasks); i++) {
@@ -961,13 +834,31 @@ void lp_rast_destroy( struct lp_rasterizer *rast )
 {
    unsigned i, cbuf;
 
-   util_unreference_framebuffer_state(&rast->state.fb);
-
    for (i = 0; i < Elements(rast->tasks); i++) {
       for (cbuf = 0; cbuf < PIPE_MAX_COLOR_BUFS; cbuf++ )
 	 align_free(rast->tasks[i].tile.color[cbuf]);
    }
 
+   /* Set exit_flag and signal each thread's work_ready semaphore.
+    * Each thread will be woken up, notice that the exit_flag is set and
+    * break out of its main loop.  The thread will then exit.
+    */
+   rast->exit_flag = TRUE;
+   for (i = 0; i < rast->num_threads; i++) {
+      pipe_semaphore_signal(&rast->tasks[i].work_ready);
+   }
+
+   /* Wait for threads to terminate before cleaning up per-thread data */
+   for (i = 0; i < rast->num_threads; i++) {
+      pipe_thread_wait(rast->threads[i]);
+   }
+
+   /* Clean up per-thread data */
+   for (i = 0; i < rast->num_threads; i++) {
+      pipe_semaphore_destroy(&rast->tasks[i].work_ready);
+      pipe_semaphore_destroy(&rast->tasks[i].work_done);
+   }
+
    /* for synchronizing rasterization threads */
    pipe_barrier_destroy( &rast->barrier );
 
diff --git a/src/gallium/drivers/llvmpipe/lp_rast.h b/src/gallium/drivers/llvmpipe/lp_rast.h
index 875f18e0c0..303f6e3f7e 100644
--- a/src/gallium/drivers/llvmpipe/lp_rast.h
+++ b/src/gallium/drivers/llvmpipe/lp_rast.h
@@ -43,16 +43,17 @@
 
 struct lp_rasterizer;
 struct lp_scene;
-struct lp_scene_queue;
 struct lp_fence;
 struct cmd_bin;
-struct pipe_screen;
 
 /** For sub-pixel positioning */
 #define FIXED_ORDER 4
 #define FIXED_ONE (1<<FIXED_ORDER)
 
 
+struct lp_rasterizer_task;
+
+
 /**
  * Rasterization state.
  * Objects of this type are put into the shared data bin and pointed
@@ -94,9 +95,13 @@ struct lp_rast_shader_inputs {
  * Rasterization information for a triangle known to be in this bin,
  * plus inputs to run the shader:
  * These fields are tile- and bin-independent.
- * Objects of this type are put into the setup_context::data buffer.
+ * Objects of this type are put into the lp_setup_context::data buffer.
  */
 struct lp_rast_triangle {
+#ifdef DEBUG
+   float v[3][2];
+#endif
+
    /* one-pixel sized trivial accept offsets for each plane */
    int ei1;                   
    int ei2;
@@ -126,18 +131,21 @@ struct lp_rast_triangle {
 
 
 
-struct lp_rasterizer *lp_rast_create( struct pipe_screen *screen,
-                                      struct lp_scene_queue *empty );
+struct lp_rasterizer *
+lp_rast_create( void );
 
-void lp_rast_destroy( struct lp_rasterizer * );
+void
+lp_rast_destroy( struct lp_rasterizer * );
 
-unsigned lp_rast_get_num_threads( struct lp_rasterizer * );
+unsigned
+lp_rast_get_num_threads( struct lp_rasterizer * );
 
-void lp_rasterize_scene( struct lp_rasterizer *rast,
-			 struct lp_scene *scene,
-			 const struct pipe_framebuffer_state *fb,
-			 bool write_depth );
+void 
+lp_rast_queue_scene( struct lp_rasterizer *rast,
+                     struct lp_scene *scene );
 
+void
+lp_rast_finish( struct lp_rasterizer *rast );
 
 
 union lp_rast_cmd_arg {
@@ -201,32 +209,25 @@ lp_rast_arg_null( void )
  * the bins are executed.
  */
 
-void lp_rast_clear_color( struct lp_rasterizer *, 
-                          unsigned thread_index,
+void lp_rast_clear_color( struct lp_rasterizer_task *, 
                           const union lp_rast_cmd_arg );
 
-void lp_rast_clear_zstencil( struct lp_rasterizer *, 
-                             unsigned thread_index,
+void lp_rast_clear_zstencil( struct lp_rasterizer_task *, 
                              const union lp_rast_cmd_arg );
 
-void lp_rast_load_color( struct lp_rasterizer *, 
-                         unsigned thread_index,
+void lp_rast_load_color( struct lp_rasterizer_task *, 
                          const union lp_rast_cmd_arg );
 
-void lp_rast_set_state( struct lp_rasterizer *, 
-                        unsigned thread_index,
+void lp_rast_set_state( struct lp_rasterizer_task *, 
                         const union lp_rast_cmd_arg );
 
-void lp_rast_triangle( struct lp_rasterizer *, 
-                       unsigned thread_index,
+void lp_rast_triangle( struct lp_rasterizer_task *, 
                        const union lp_rast_cmd_arg );
 
-void lp_rast_shade_tile( struct lp_rasterizer *,
-                         unsigned thread_index,
+void lp_rast_shade_tile( struct lp_rasterizer_task *,
                          const union lp_rast_cmd_arg );
 
-void lp_rast_fence( struct lp_rasterizer *,
-                    unsigned thread_index,
+void lp_rast_fence( struct lp_rasterizer_task *,
                     const union lp_rast_cmd_arg );
 
 #endif
diff --git a/src/gallium/drivers/llvmpipe/lp_rast_priv.h b/src/gallium/drivers/llvmpipe/lp_rast_priv.h
index 5c5497e092..39bf2c2587 100644
--- a/src/gallium/drivers/llvmpipe/lp_rast_priv.h
+++ b/src/gallium/drivers/llvmpipe/lp_rast_priv.h
@@ -38,8 +38,6 @@
 #define MAX_THREADS 8  /* XXX probably temporary here */
 
 
-struct pipe_transfer;
-struct pipe_screen;
 struct lp_rasterizer;
 
 
@@ -82,19 +80,26 @@ struct lp_rasterizer_task
  */
 struct lp_rasterizer
 {
-   boolean clipped_tile;
-   boolean check_for_clipped_tiles;
+   boolean exit_flag;
 
    /* Framebuffer stuff
     */
-   struct pipe_screen *screen;
-   struct pipe_transfer *cbuf_transfer[PIPE_MAX_COLOR_BUFS];
-   struct pipe_transfer *zsbuf_transfer;
-   void *cbuf_map[PIPE_MAX_COLOR_BUFS];
-   uint8_t *zsbuf_map;
+   struct {
+      void *map;
+      unsigned stride;
+      unsigned width;
+      unsigned height;
+      enum pipe_format format;
+   } cbuf[PIPE_MAX_COLOR_BUFS];
+
+   struct {
+      uint8_t *map;
+      unsigned stride;
+      unsigned blocksize;
+   } zsbuf;
 
    struct {
-      struct pipe_framebuffer_state fb;
+      unsigned nr_cbufs;
       boolean write_color;
       boolean write_zstencil;
       unsigned clear_color;
@@ -104,7 +109,14 @@ struct lp_rasterizer
 
    /** The incoming queue of scenes ready to rasterize */
    struct lp_scene_queue *full_scenes;
-   /** The outgoing queue of processed scenes to return to setup modulee */
+
+   /**
+    * The outgoing queue of processed scenes to return to setup module
+    *
+    * XXX: while scenes are per-context but the rasterizer is
+    * (potentially) shared, these empty scenes should be returned to
+    * the context which created them rather than retained here.
+    */
    struct lp_scene_queue *empty_scenes;
 
    /** The scene currently being rasterized by the threads */
@@ -121,8 +133,7 @@ struct lp_rasterizer
 };
 
 
-void lp_rast_shade_quads( struct lp_rasterizer *rast,
-                          unsigned thread_index,
+void lp_rast_shade_quads( struct lp_rasterizer_task *task,
                           const struct lp_rast_shader_inputs *inputs,
                           unsigned x, unsigned y,
                           int32_t c1, int32_t c2, int32_t c3);
@@ -137,17 +148,18 @@ lp_rast_depth_pointer( struct lp_rasterizer *rast,
                        unsigned x, unsigned y )
 {
    void * depth;
+
    assert((x % TILE_VECTOR_WIDTH) == 0);
    assert((y % TILE_VECTOR_HEIGHT) == 0);
-   if(!rast->zsbuf_map)
+
+   if (!rast->zsbuf.map)
       return NULL;
-   assert(rast->zsbuf_transfer);
-   depth = rast->zsbuf_map +
-           y*rast->zsbuf_transfer->stride +
-           TILE_VECTOR_HEIGHT*x*util_format_get_blocksize(rast->zsbuf_transfer->texture->format);
-#ifdef DEBUG
+
+   depth = (rast->zsbuf.map +
+            rast->zsbuf.stride * y +
+            rast->zsbuf.blocksize * x * TILE_VECTOR_HEIGHT);
+
    assert(lp_check_alignment(depth, 16));
-#endif
    return depth;
 }
 
@@ -159,13 +171,13 @@ lp_rast_depth_pointer( struct lp_rasterizer *rast,
  * \param x, y location of 4x4 block in window coords
  */
 static INLINE void
-lp_rast_shade_quads_all( struct lp_rasterizer *rast,
-                         unsigned thread_index,
+lp_rast_shade_quads_all( struct lp_rasterizer_task *task,
                          const struct lp_rast_shader_inputs *inputs,
                          unsigned x, unsigned y )
 {
-   const struct lp_rast_state *state = rast->tasks[thread_index].current_state;
-   struct lp_rast_tile *tile = &rast->tasks[thread_index].tile;
+   struct lp_rasterizer *rast = task->rast;
+   const struct lp_rast_state *state = task->current_state;
+   struct lp_rast_tile *tile = &task->tile;
    const unsigned ix = x % TILE_SIZE, iy = y % TILE_SIZE;
    uint8_t *color[PIPE_MAX_COLOR_BUFS];
    void *depth;
@@ -175,7 +187,7 @@ lp_rast_shade_quads_all( struct lp_rasterizer *rast,
    block_offset = (iy / 4) * (16 * 16) + (ix / 4) * 16;
 
    /* color buffer */
-   for (i = 0; i < rast->state.fb.nr_cbufs; i++)
+   for (i = 0; i < rast->state.nr_cbufs; i++)
       color[i] = tile->color[i] + 4 * block_offset;
 
    depth = lp_rast_depth_pointer(rast, x, y);
diff --git a/src/gallium/drivers/llvmpipe/lp_rast_tri.c b/src/gallium/drivers/llvmpipe/lp_rast_tri.c
index 0334705ef7..a5f0d14c95 100644
--- a/src/gallium/drivers/llvmpipe/lp_rast_tri.c
+++ b/src/gallium/drivers/llvmpipe/lp_rast_tri.c
@@ -89,14 +89,11 @@ static const int pos_table16[16][2] = {
  * Shade all pixels in a 4x4 block.
  */
 static void
-block_full_4( struct lp_rasterizer_task *rast_task,
-              const struct lp_rast_triangle *tri,
-              int x, int y )
+block_full_4(struct lp_rasterizer_task *task,
+             const struct lp_rast_triangle *tri,
+             int x, int y)
 {
-   lp_rast_shade_quads_all(rast_task->rast,
-                           rast_task->thread_index,
-                           &tri->inputs, 
-                           x, y);
+   lp_rast_shade_quads_all(task, &tri->inputs, x, y);
 }
 
 
@@ -104,16 +101,16 @@ block_full_4( struct lp_rasterizer_task *rast_task,
  * Shade all pixels in a 16x16 block.
  */
 static void
-block_full_16( struct lp_rasterizer_task *rast_task,
-               const struct lp_rast_triangle *tri,
-               int x, int y )
+block_full_16(struct lp_rasterizer_task *task,
+              const struct lp_rast_triangle *tri,
+              int x, int y)
 {
    unsigned ix, iy;
    assert(x % 16 == 0);
    assert(y % 16 == 0);
    for (iy = 0; iy < 16; iy += 4)
       for (ix = 0; ix < 16; ix += 4)
-	 block_full_4(rast_task, tri, x + ix, y + iy);
+	 block_full_4(task, tri, x + ix, y + iy);
 }
 
 
@@ -123,18 +120,15 @@ block_full_16( struct lp_rasterizer_task *rast_task,
  * will be done as part of the fragment shader.
  */
 static void
-do_block_4( struct lp_rasterizer_task *rast_task,
-	    const struct lp_rast_triangle *tri,
-	    int x, int y,
-	    int c1,
-	    int c2,
-	    int c3 )
+do_block_4(struct lp_rasterizer_task *task,
+           const struct lp_rast_triangle *tri,
+           int x, int y,
+           int c1, int c2, int c3)
 {
-   lp_rast_shade_quads(rast_task->rast,
-                       rast_task->thread_index,
-                       &tri->inputs, 
-                       x, y,
-                       -c1, -c2, -c3);
+   assert(x >= 0);
+   assert(y >= 0);
+
+   lp_rast_shade_quads(task, &tri->inputs, x, y, -c1, -c2, -c3);
 }
 
 
@@ -143,18 +137,18 @@ do_block_4( struct lp_rasterizer_task *rast_task,
  * of the triangle's bounds.
  */
 static void
-do_block_16( struct lp_rasterizer_task *rast_task,
-             const struct lp_rast_triangle *tri,
-             int x, int y,
-             int c0,
-             int c1,
-             int c2 )
+do_block_16(struct lp_rasterizer_task *task,
+            const struct lp_rast_triangle *tri,
+            int x, int y,
+            int c0, int c1, int c2)
 {
    unsigned mask = 0;
    int eo[3];
    int c[3];
    int i, j;
 
+   assert(x >= 0);
+   assert(y >= 0);
    assert(x % 16 == 0);
    assert(y % 16 == 0);
 
@@ -193,7 +187,7 @@ do_block_16( struct lp_rasterizer_task *rast_task,
        * the triangle.  It's a little faster to do it in the jit code.
        */
       LP_COUNT(nr_non_empty_4);
-      do_block_4(rast_task, tri, px, py, cx1, cx2, cx3);
+      do_block_4(task, tri, px, py, cx1, cx2, cx3);
    }
 }
 
@@ -203,15 +197,11 @@ do_block_16( struct lp_rasterizer_task *rast_task,
  * for this triangle.
  */
 void
-lp_rast_triangle( struct lp_rasterizer *rast,
-                  unsigned thread_index,
-                  const union lp_rast_cmd_arg arg )
+lp_rast_triangle(struct lp_rasterizer_task *task,
+                 const union lp_rast_cmd_arg arg)
 {
-   struct lp_rasterizer_task *rast_task = &rast->tasks[thread_index];
    const struct lp_rast_triangle *tri = arg.triangle;
-
-   int x = rast_task->x;
-   int y = rast_task->y;
+   const int x = task->x, y = task->y;
    int ei[3], eo[3], c[3];
    unsigned outmask, inmask, partial_mask;
    unsigned i, j;
@@ -272,7 +262,7 @@ lp_rast_triangle( struct lp_rasterizer *rast,
       partial_mask &= ~(1 << i);
 
       LP_COUNT(nr_partially_covered_16);
-      do_block_16(rast_task, tri, px, py, cx1, cx2, cx3);
+      do_block_16(task, tri, px, py, cx1, cx2, cx3);
    }
 
    /* Iterate over fulls: 
@@ -285,6 +275,6 @@ lp_rast_triangle( struct lp_rasterizer *rast,
       inmask &= ~(1 << i);
 
       LP_COUNT(nr_fully_covered_16);
-      block_full_16(rast_task, tri, px, py);
+      block_full_16(task, tri, px, py);
    }
 }
diff --git a/src/gallium/drivers/llvmpipe/lp_scene.c b/src/gallium/drivers/llvmpipe/lp_scene.c
index b7116297ec..681ce674d4 100644
--- a/src/gallium/drivers/llvmpipe/lp_scene.c
+++ b/src/gallium/drivers/llvmpipe/lp_scene.c
@@ -29,44 +29,67 @@
 #include "util/u_memory.h"
 #include "util/u_inlines.h"
 #include "util/u_simple_list.h"
+#include "util/u_surface.h"
 #include "lp_scene.h"
+#include "lp_scene_queue.h"
+#include "lp_debug.h"
 
 
 struct lp_scene *
-lp_scene_create(void)
+lp_scene_create( struct pipe_context *pipe,
+                 struct lp_scene_queue *queue )
 {
+   unsigned i, j;
    struct lp_scene *scene = CALLOC_STRUCT(lp_scene);
-   if (scene)
-      lp_scene_init(scene);
+   if (!scene)
+      return NULL;
+
+   scene->pipe = pipe;
+   scene->empty_queue = queue;
+
+   for (i = 0; i < TILES_X; i++) {
+      for (j = 0; j < TILES_Y; j++) {
+         struct cmd_bin *bin = lp_scene_get_bin(scene, i, j);
+         bin->commands.head = bin->commands.tail = CALLOC_STRUCT(cmd_block);
+      }
+   }
+
+   scene->data.head =
+      scene->data.tail = CALLOC_STRUCT(data_block);
+
+   make_empty_list(&scene->textures);
+
+   pipe_mutex_init(scene->mutex);
+
    return scene;
 }
 
 
+/**
+ * Free all data associated with the given scene, and free(scene).
+ */
 void
 lp_scene_destroy(struct lp_scene *scene)
 {
-   lp_scene_reset(scene);
-   lp_scene_free_bin_data(scene);
-   FREE(scene);
-}
+   unsigned i, j;
 
+   lp_scene_reset(scene);
 
-void
-lp_scene_init(struct lp_scene *scene)
-{
-   unsigned i, j;
    for (i = 0; i < TILES_X; i++)
       for (j = 0; j < TILES_Y; j++) {
          struct cmd_bin *bin = lp_scene_get_bin(scene, i, j);
-         bin->commands.head = bin->commands.tail = CALLOC_STRUCT(cmd_block);
+         assert(bin->commands.head == bin->commands.tail);
+         FREE(bin->commands.head);
+         bin->commands.head = NULL;
+         bin->commands.tail = NULL;
       }
 
-   scene->data.head =
-      scene->data.tail = CALLOC_STRUCT(data_block);
+   FREE(scene->data.head);
+   scene->data.head = NULL;
 
-   make_empty_list(&scene->textures);
+   pipe_mutex_destroy(scene->mutex);
 
-   pipe_mutex_init(scene->mutex);
+   FREE(scene);
 }
 
 
@@ -92,6 +115,9 @@ lp_scene_is_empty(struct lp_scene *scene )
 }
 
 
+/* Free data for one particular bin.  May be called from the
+ * rasterizer thread(s).
+ */
 void
 lp_scene_bin_reset(struct lp_scene *scene, unsigned x, unsigned y)
 {
@@ -100,6 +126,9 @@ lp_scene_bin_reset(struct lp_scene *scene, unsigned x, unsigned y)
    struct cmd_block *block;
    struct cmd_block *tmp;
 
+   assert(x < TILES_X);
+   assert(y < TILES_Y);
+
    for (block = list->head; block != list->tail; block = tmp) {
       tmp = block->next;
       FREE(block);
@@ -112,7 +141,8 @@ lp_scene_bin_reset(struct lp_scene *scene, unsigned x, unsigned y)
 
 
 /**
- * Set scene to empty state.
+ * Free all the temporary data in a scene.  May be called from the
+ * rasterizer thread(s).
  */
 void
 lp_scene_reset(struct lp_scene *scene )
@@ -159,40 +189,8 @@ lp_scene_reset(struct lp_scene *scene )
 }
 
 
-/**
- * Free all data associated with the given bin, but don't free(scene).
- */
-void
-lp_scene_free_bin_data(struct lp_scene *scene)
-{
-   unsigned i, j;
-
-   for (i = 0; i < TILES_X; i++)
-      for (j = 0; j < TILES_Y; j++) {
-         struct cmd_bin *bin = lp_scene_get_bin(scene, i, j);
-         /* lp_reset_scene() should have been already called */
-         assert(bin->commands.head == bin->commands.tail);
-         FREE(bin->commands.head);
-         bin->commands.head = NULL;
-         bin->commands.tail = NULL;
-      }
-
-   FREE(scene->data.head);
-   scene->data.head = NULL;
-
-   pipe_mutex_destroy(scene->mutex);
-}
-
 
-void
-lp_scene_set_framebuffer_size( struct lp_scene *scene,
-                               unsigned width, unsigned height )
-{
-   assert(lp_scene_is_empty(scene));
 
-   scene->tiles_x = align(width, TILE_SIZE) / TILE_SIZE;
-   scene->tiles_y = align(height, TILE_SIZE) / TILE_SIZE;
-}
 
 
 void
@@ -390,3 +388,136 @@ end:
    pipe_mutex_unlock(scene->mutex);
    return bin;
 }
+
+
+/**
+ * Prepare this scene for the rasterizer.
+ * Map the framebuffer surfaces.  Initialize the 'rast' state.
+ */
+static boolean
+lp_scene_map_buffers( struct lp_scene *scene )
+{
+   struct pipe_surface *cbuf, *zsbuf;
+   int i;
+
+   LP_DBG(DEBUG_RAST, "%s\n", __FUNCTION__);
+
+
+   /* Map all color buffers 
+    */
+   for (i = 0; i < scene->fb.nr_cbufs; i++) {
+      cbuf = scene->fb.cbufs[i];
+      if (cbuf) {
+	 scene->cbuf_map[i] = llvmpipe_texture_map(cbuf->texture,
+	                                           cbuf->face,
+                                                   cbuf->level,
+                                                   cbuf->zslice);
+	 if (!scene->cbuf_map[i])
+	    goto fail;
+      }
+   }
+
+   /* Map the zsbuffer
+    */
+   zsbuf = scene->fb.zsbuf;
+   if (zsbuf) {
+      scene->zsbuf_map = llvmpipe_texture_map(zsbuf->texture,
+                                              zsbuf->face,
+                                              zsbuf->level,
+                                              zsbuf->zslice);
+      if (!scene->zsbuf_map)
+	 goto fail;
+   }
+
+   return TRUE;
+
+fail:
+   /* Unmap and release transfers?
+    */
+   return FALSE;
+}
+
+
+
+/**
+ * Called after rasterizer as finished rasterizing a scene. 
+ * 
+ * We want to call this from the pipe_context's current thread to
+ * avoid having to have mutexes on the transfer functions.
+ */
+static void
+lp_scene_unmap_buffers( struct lp_scene *scene )
+{
+   unsigned i;
+
+   for (i = 0; i < scene->fb.nr_cbufs; i++) {
+      if (scene->cbuf_map[i]) {
+         struct pipe_surface *cbuf = scene->fb.cbufs[i];
+         llvmpipe_texture_unmap(cbuf->texture,
+                                cbuf->face,
+                                cbuf->level,
+                                cbuf->zslice);
+         scene->cbuf_map[i] = NULL;
+      }
+   }
+
+   if (scene->zsbuf_map) {
+      struct pipe_surface *zsbuf = scene->fb.zsbuf;
+      llvmpipe_texture_unmap(zsbuf->texture,
+                             zsbuf->face,
+                             zsbuf->level,
+                             zsbuf->zslice);
+      scene->zsbuf_map = NULL;
+   }
+
+   util_unreference_framebuffer_state( &scene->fb );
+}
+
+
+void lp_scene_begin_binning( struct lp_scene *scene,
+                             struct pipe_framebuffer_state *fb )
+{
+   assert(lp_scene_is_empty(scene));
+
+   util_copy_framebuffer_state(&scene->fb, fb);
+
+   scene->tiles_x = align(fb->width, TILE_SIZE) / TILE_SIZE;
+   scene->tiles_y = align(fb->height, TILE_SIZE) / TILE_SIZE;
+}
+
+
+void lp_scene_rasterize( struct lp_scene *scene,
+                         struct lp_rasterizer *rast,
+                         boolean write_depth )
+{
+   if (0) {
+      unsigned x, y;
+      debug_printf("rasterize scene:\n");
+      debug_printf("  data size: %u\n", lp_scene_data_size(scene));
+      for (y = 0; y < scene->tiles_y; y++) {
+         for (x = 0; x < scene->tiles_x; x++) {
+            debug_printf("  bin %u, %u size: %u\n", x, y,
+                         lp_scene_bin_size(scene, x, y));
+         }
+      }
+   }
+
+
+   scene->write_depth = (scene->fb.zsbuf != NULL &&
+                         write_depth);
+
+   lp_scene_map_buffers( scene );
+
+   /* Enqueue the scene for rasterization, then immediately wait for
+    * it to finish.
+    */
+   lp_rast_queue_scene( rast, scene );
+
+   /* Currently just wait for the rasterizer to finish.  Some
+    * threading interactions need to be worked out, particularly once
+    * transfers become per-context:
+    */
+   lp_rast_finish( rast );
+   lp_scene_unmap_buffers( scene );
+   lp_scene_enqueue( scene->empty_queue, scene );
+}
diff --git a/src/gallium/drivers/llvmpipe/lp_scene.h b/src/gallium/drivers/llvmpipe/lp_scene.h
index fb478cc2eb..b602b1e8a0 100644
--- a/src/gallium/drivers/llvmpipe/lp_scene.h
+++ b/src/gallium/drivers/llvmpipe/lp_scene.h
@@ -39,6 +39,7 @@
 #include "lp_tile_soa.h"
 #include "lp_rast.h"
 
+struct lp_scene_queue;
 
 /* We're limited to 2K by 2K for 32bit fixed point rasterization.
  * Will need a 64-bit version for larger framebuffers.
@@ -56,8 +57,7 @@
 
 /* switch to a non-pointer value for this:
  */
-typedef void (*lp_rast_cmd)( struct lp_rasterizer *,
-                             unsigned thread_index,
+typedef void (*lp_rast_cmd)( struct lp_rasterizer_task *,
                              const union lp_rast_cmd_arg );
 
 struct cmd_block {
@@ -113,8 +113,12 @@ struct texture_ref {
  * scenes:
  */
 struct lp_scene {
-   struct cmd_bin tile[TILES_X][TILES_Y];
-   struct data_block_list data;
+   struct pipe_context *pipe;
+
+   /* Scene's buffers are mapped at the time the scene is enqueued:
+    */
+   void *cbuf_map[PIPE_MAX_COLOR_BUFS];
+   uint8_t *zsbuf_map;
 
    /** the framebuffer to render the scene into */
    struct pipe_framebuffer_state fb;
@@ -132,25 +136,28 @@ struct lp_scene {
 
    int curr_x, curr_y;  /**< for iterating over bins */
    pipe_mutex mutex;
+
+   /* Where to place this scene once it has been rasterized:
+    */
+   struct lp_scene_queue *empty_queue;
+
+   struct cmd_bin tile[TILES_X][TILES_Y];
+   struct data_block_list data;
 };
 
 
 
-struct lp_scene *lp_scene_create(void);
+struct lp_scene *lp_scene_create(struct pipe_context *pipe,
+                                 struct lp_scene_queue *empty_queue);
 
 void lp_scene_destroy(struct lp_scene *scene);
 
 
-void lp_scene_init(struct lp_scene *scene);
 
 boolean lp_scene_is_empty(struct lp_scene *scene );
 
 void lp_scene_reset(struct lp_scene *scene );
 
-void lp_scene_free_bin_data(struct lp_scene *scene);
-
-void lp_scene_set_framebuffer_size( struct lp_scene *scene,
-                                  unsigned width, unsigned height );
 
 void lp_bin_new_data_block( struct data_block_list *list );
 
@@ -297,5 +304,13 @@ lp_scene_bin_iter_begin( struct lp_scene *scene );
 struct cmd_bin *
 lp_scene_bin_iter_next( struct lp_scene *scene, int *bin_x, int *bin_y );
 
+void
+lp_scene_rasterize( struct lp_scene *scene,
+                    struct lp_rasterizer *rast,
+                    boolean write_depth );
+
+void
+lp_scene_begin_binning( struct lp_scene *scene,
+                        struct pipe_framebuffer_state *fb );
 
 #endif /* LP_BIN_H */
diff --git a/src/gallium/drivers/llvmpipe/lp_screen.c b/src/gallium/drivers/llvmpipe/lp_screen.c
index 1cd3ea9a84..5093f58bb1 100644
--- a/src/gallium/drivers/llvmpipe/lp_screen.c
+++ b/src/gallium/drivers/llvmpipe/lp_screen.c
@@ -34,11 +34,13 @@
 #include "lp_texture.h"
 #include "lp_buffer.h"
 #include "lp_fence.h"
-#include "lp_winsys.h"
 #include "lp_jit.h"
 #include "lp_screen.h"
 #include "lp_context.h"
 #include "lp_debug.h"
+#include "lp_public.h"
+
+#include "state_tracker/sw_winsys.h"
 
 #ifdef DEBUG
 int LP_DEBUG = 0;
@@ -83,7 +85,7 @@ llvmpipe_get_param(struct pipe_screen *screen, int param)
    case PIPE_CAP_MAX_TEXTURE_IMAGE_UNITS:
       return PIPE_MAX_SAMPLERS;
    case PIPE_CAP_MAX_VERTEX_TEXTURE_UNITS:
-      return PIPE_MAX_VERTEX_SAMPLERS;
+      return 0;
    case PIPE_CAP_MAX_COMBINED_SAMPLERS:
       return PIPE_MAX_SAMPLERS + PIPE_MAX_VERTEX_SAMPLERS;
    case PIPE_CAP_NPOT_TEXTURES:
@@ -107,11 +109,11 @@ llvmpipe_get_param(struct pipe_screen *screen, int param)
    case PIPE_CAP_TEXTURE_SHADOW_MAP:
       return 1;
    case PIPE_CAP_MAX_TEXTURE_2D_LEVELS:
-      return 13; /* max 4Kx4K */
+      return LP_MAX_TEXTURE_2D_LEVELS;
    case PIPE_CAP_MAX_TEXTURE_3D_LEVELS:
-      return 9;  /* max 256x256x256 */
+      return LP_MAX_TEXTURE_3D_LEVELS;
    case PIPE_CAP_MAX_TEXTURE_CUBE_LEVELS:
-      return 13; /* max 4Kx4K */
+      return LP_MAX_TEXTURE_2D_LEVELS;
    case PIPE_CAP_TGSI_CONT_SUPPORTED:
       return 1;
    case PIPE_CAP_BLEND_EQUATION_SEPARATE:
@@ -167,7 +169,7 @@ llvmpipe_is_format_supported( struct pipe_screen *_screen,
                               unsigned geom_flags )
 {
    struct llvmpipe_screen *screen = llvmpipe_screen(_screen);
-   struct llvmpipe_winsys *winsys = screen->winsys;
+   struct sw_winsys *winsys = screen->winsys;
    const struct util_format_description *format_desc;
 
    format_desc = util_format_description(format);
@@ -194,9 +196,7 @@ llvmpipe_is_format_supported( struct pipe_screen *_screen,
          format_desc->block.height != 1)
          return FALSE;
 
-      if(format_desc->layout != UTIL_FORMAT_LAYOUT_SCALAR &&
-         format_desc->layout != UTIL_FORMAT_LAYOUT_ARITH &&
-         format_desc->layout != UTIL_FORMAT_LAYOUT_ARRAY)
+      if(format_desc->layout != UTIL_FORMAT_LAYOUT_PLAIN)
          return FALSE;
 
       if(format_desc->colorspace != UTIL_FORMAT_COLORSPACE_RGB &&
@@ -224,33 +224,22 @@ llvmpipe_is_format_supported( struct pipe_screen *_screen,
          format_desc->block.height != 1)
          return FALSE;
 
-      if(format_desc->layout != UTIL_FORMAT_LAYOUT_SCALAR &&
-         format_desc->layout != UTIL_FORMAT_LAYOUT_ARITH &&
-         format_desc->layout != UTIL_FORMAT_LAYOUT_ARRAY)
+      if(format_desc->layout != UTIL_FORMAT_LAYOUT_PLAIN)
          return FALSE;
 
       if(format_desc->colorspace != UTIL_FORMAT_COLORSPACE_RGB &&
-         format_desc->colorspace != UTIL_FORMAT_COLORSPACE_SRGB &&
          format_desc->colorspace != UTIL_FORMAT_COLORSPACE_ZS)
          return FALSE;
+
+      /* not supported yet */
+      if (format == PIPE_FORMAT_Z16_UNORM)
+         return FALSE;
    }
 
    return TRUE;
 }
 
 
-static struct pipe_buffer *
-llvmpipe_surface_buffer_create(struct pipe_screen *screen,
-                               unsigned width, unsigned height,
-                               enum pipe_format format,
-                               unsigned tex_usage,
-                               unsigned usage,
-                               unsigned *stride)
-{
-   /* This function should never be used */
-   assert(0);
-   return NULL;
-}
 
 
 static void
@@ -259,7 +248,7 @@ llvmpipe_flush_frontbuffer(struct pipe_screen *_screen,
                            void *context_private)
 {
    struct llvmpipe_screen *screen = llvmpipe_screen(_screen);
-   struct llvmpipe_winsys *winsys = screen->winsys;
+   struct sw_winsys *winsys = screen->winsys;
    struct llvmpipe_texture *texture = llvmpipe_texture(surface->texture);
 
    assert(texture->dt);
@@ -272,7 +261,7 @@ static void
 llvmpipe_destroy_screen( struct pipe_screen *_screen )
 {
    struct llvmpipe_screen *screen = llvmpipe_screen(_screen);
-   struct llvmpipe_winsys *winsys = screen->winsys;
+   struct sw_winsys *winsys = screen->winsys;
 
    lp_jit_screen_cleanup(screen);
 
@@ -289,7 +278,7 @@ llvmpipe_destroy_screen( struct pipe_screen *_screen )
  * Note: we're not presently subclassing pipe_screen (no llvmpipe_screen).
  */
 struct pipe_screen *
-llvmpipe_create_screen(struct llvmpipe_winsys *winsys)
+llvmpipe_create_screen(struct sw_winsys *winsys)
 {
    struct llvmpipe_screen *screen = CALLOC_STRUCT(llvmpipe_screen);
 
@@ -310,7 +299,6 @@ llvmpipe_create_screen(struct llvmpipe_winsys *winsys)
    screen->base.get_paramf = llvmpipe_get_paramf;
    screen->base.is_format_supported = llvmpipe_is_format_supported;
 
-   screen->base.surface_buffer_create = llvmpipe_surface_buffer_create;
    screen->base.context_create = llvmpipe_create_context;
    screen->base.flush_frontbuffer = llvmpipe_flush_frontbuffer;
 
diff --git a/src/gallium/drivers/llvmpipe/lp_screen.h b/src/gallium/drivers/llvmpipe/lp_screen.h
index 4a1b4d6f3e..d977f98cfa 100644
--- a/src/gallium/drivers/llvmpipe/lp_screen.h
+++ b/src/gallium/drivers/llvmpipe/lp_screen.h
@@ -34,23 +34,21 @@
 #ifndef LP_SCREEN_H
 #define LP_SCREEN_H
 
-#include <llvm-c/Core.h>
-#include <llvm-c/Analysis.h>
-#include <llvm-c/Target.h>
+#include "os/os_llvm.h"
 #include <llvm-c/ExecutionEngine.h>
 
 #include "pipe/p_screen.h"
 #include "pipe/p_defines.h"
 
 
-struct llvmpipe_winsys;
+struct sw_winsys;
 
 
 struct llvmpipe_screen
 {
    struct pipe_screen base;
 
-   struct llvmpipe_winsys *winsys;
+   struct sw_winsys *winsys;
 
    LLVMModuleRef module;
    LLVMExecutionEngineRef engine;
@@ -76,4 +74,5 @@ llvmpipe_screen( struct pipe_screen *pipe )
 }
 
 
+
 #endif /* LP_SCREEN_H */
diff --git a/src/gallium/drivers/llvmpipe/lp_setup.c b/src/gallium/drivers/llvmpipe/lp_setup.c
index cb873667a2..16128c34c8 100644
--- a/src/gallium/drivers/llvmpipe/lp_setup.c
+++ b/src/gallium/drivers/llvmpipe/lp_setup.c
@@ -46,17 +46,17 @@
 #include "lp_rast.h"
 #include "lp_setup_context.h"
 #include "lp_screen.h"
-#include "lp_winsys.h"
+#include "state_tracker/sw_winsys.h"
 
 #include "draw/draw_context.h"
 #include "draw/draw_vbuf.h"
 
 
-static void set_scene_state( struct setup_context *, unsigned );
+static void set_scene_state( struct lp_setup_context *, unsigned );
 
 
 struct lp_scene *
-lp_setup_get_current_scene(struct setup_context *setup)
+lp_setup_get_current_scene(struct lp_setup_context *setup)
 {
    if (!setup->scene) {
 
@@ -64,18 +64,17 @@ lp_setup_get_current_scene(struct setup_context *setup)
        */
       setup->scene = lp_scene_dequeue(setup->empty_scenes, TRUE);
 
-      if(0)lp_scene_reset( setup->scene ); /* XXX temporary? */
+      assert(lp_scene_is_empty(setup->scene));
 
-      lp_scene_set_framebuffer_size(setup->scene,
-                                    setup->fb.width, 
-                                    setup->fb.height);
+      lp_scene_begin_binning(setup->scene,
+                             &setup->fb );
    }
    return setup->scene;
 }
 
 
 static void
-first_triangle( struct setup_context *setup,
+first_triangle( struct lp_setup_context *setup,
                 const float (*v0)[4],
                 const float (*v1)[4],
                 const float (*v2)[4])
@@ -86,7 +85,7 @@ first_triangle( struct setup_context *setup,
 }
 
 static void
-first_line( struct setup_context *setup,
+first_line( struct lp_setup_context *setup,
 	    const float (*v0)[4],
 	    const float (*v1)[4])
 {
@@ -96,7 +95,7 @@ first_line( struct setup_context *setup,
 }
 
 static void
-first_point( struct setup_context *setup,
+first_point( struct lp_setup_context *setup,
 	     const float (*v0)[4])
 {
    set_scene_state( setup, SETUP_ACTIVE );
@@ -104,7 +103,7 @@ first_point( struct setup_context *setup,
    setup->point( setup, v0 );
 }
 
-static void reset_context( struct setup_context *setup )
+static void reset_context( struct lp_setup_context *setup )
 {
    LP_DBG(DEBUG_SETUP, "%s\n", __FUNCTION__);
 
@@ -132,14 +131,13 @@ static void reset_context( struct setup_context *setup )
 
 /** Rasterize all scene's bins */
 static void
-lp_setup_rasterize_scene( struct setup_context *setup,
-			 boolean write_depth )
+lp_setup_rasterize_scene( struct lp_setup_context *setup,
+                          boolean write_depth )
 {
    struct lp_scene *scene = lp_setup_get_current_scene(setup);
 
-   lp_rasterize_scene(setup->rast,
-                      scene,
-                      &setup->fb,
+   lp_scene_rasterize(scene,
+                      setup->rast,
                       write_depth);
 
    reset_context( setup );
@@ -150,7 +148,7 @@ lp_setup_rasterize_scene( struct setup_context *setup,
 
 
 static void
-begin_binning( struct setup_context *setup )
+begin_binning( struct lp_setup_context *setup )
 {
    struct lp_scene *scene = lp_setup_get_current_scene(setup);
 
@@ -186,7 +184,7 @@ begin_binning( struct setup_context *setup )
  * TODO: fast path for fullscreen clears and no triangles.
  */
 static void
-execute_clears( struct setup_context *setup )
+execute_clears( struct lp_setup_context *setup )
 {
    LP_DBG(DEBUG_SETUP, "%s\n", __FUNCTION__);
 
@@ -196,7 +194,7 @@ execute_clears( struct setup_context *setup )
 
 
 static void
-set_scene_state( struct setup_context *setup,
+set_scene_state( struct lp_setup_context *setup,
            unsigned new_state )
 {
    unsigned old_state = setup->state;
@@ -231,7 +229,7 @@ set_scene_state( struct setup_context *setup,
 
 
 void
-lp_setup_flush( struct setup_context *setup,
+lp_setup_flush( struct lp_setup_context *setup,
                 unsigned flags )
 {
    LP_DBG(DEBUG_SETUP, "%s\n", __FUNCTION__);
@@ -241,27 +239,24 @@ lp_setup_flush( struct setup_context *setup,
 
 
 void
-lp_setup_bind_framebuffer( struct setup_context *setup,
+lp_setup_bind_framebuffer( struct lp_setup_context *setup,
                            const struct pipe_framebuffer_state *fb )
 {
-   struct lp_scene *scene = lp_setup_get_current_scene(setup);
-
    LP_DBG(DEBUG_SETUP, "%s\n", __FUNCTION__);
 
+   /* Flush any old scene.
+    */
    set_scene_state( setup, SETUP_FLUSHED );
 
-   /* re-get scene pointer, may have a new scene after flushing */
-   (void) scene;
-   scene = lp_setup_get_current_scene(setup);
-
+   /* Set new state.  This will be picked up later when we next need a
+    * scene.
+    */
    util_copy_framebuffer_state(&setup->fb, fb);
-
-   lp_scene_set_framebuffer_size(scene, setup->fb.width, setup->fb.height);
 }
 
 
 void
-lp_setup_clear( struct setup_context *setup,
+lp_setup_clear( struct lp_setup_context *setup,
                 const float *color,
                 double depth,
                 unsigned stencil,
@@ -319,7 +314,7 @@ lp_setup_clear( struct setup_context *setup,
  * Emit a fence.
  */
 struct pipe_fence_handle *
-lp_setup_fence( struct setup_context *setup )
+lp_setup_fence( struct lp_setup_context *setup )
 {
    struct lp_scene *scene = lp_setup_get_current_scene(setup);
    const unsigned rank = lp_scene_get_num_bins( scene ); /* xxx */
@@ -339,10 +334,11 @@ lp_setup_fence( struct setup_context *setup )
 
 
 void 
-lp_setup_set_triangle_state( struct setup_context *setup,
+lp_setup_set_triangle_state( struct lp_setup_context *setup,
                              unsigned cull_mode,
                              boolean ccw_is_frontface,
-                             boolean scissor )
+                             boolean scissor,
+                             boolean gl_rasterization_rules)
 {
    LP_DBG(DEBUG_SETUP, "%s\n", __FUNCTION__);
 
@@ -350,12 +346,13 @@ lp_setup_set_triangle_state( struct setup_context *setup,
    setup->cullmode = cull_mode;
    setup->triangle = first_triangle;
    setup->scissor_test = scissor;
+   setup->pixel_offset = gl_rasterization_rules ? 0.5f : 0.0f;
 }
 
 
 
 void
-lp_setup_set_fs_inputs( struct setup_context *setup,
+lp_setup_set_fs_inputs( struct lp_setup_context *setup,
                         const struct lp_shader_input *input,
                         unsigned nr )
 {
@@ -366,7 +363,7 @@ lp_setup_set_fs_inputs( struct setup_context *setup,
 }
 
 void
-lp_setup_set_fs_functions( struct setup_context *setup,
+lp_setup_set_fs_functions( struct lp_setup_context *setup,
                            lp_jit_frag_func jit_function0,
                            lp_jit_frag_func jit_function1,
                            boolean opaque )
@@ -381,7 +378,7 @@ lp_setup_set_fs_functions( struct setup_context *setup,
 }
 
 void
-lp_setup_set_fs_constants(struct setup_context *setup,
+lp_setup_set_fs_constants(struct lp_setup_context *setup,
                           struct pipe_buffer *buffer)
 {
    LP_DBG(DEBUG_SETUP, "%s %p\n", __FUNCTION__, (void *) buffer);
@@ -393,7 +390,7 @@ lp_setup_set_fs_constants(struct setup_context *setup,
 
 
 void
-lp_setup_set_alpha_ref_value( struct setup_context *setup,
+lp_setup_set_alpha_ref_value( struct lp_setup_context *setup,
                               float alpha_ref_value )
 {
    LP_DBG(DEBUG_SETUP, "%s %f\n", __FUNCTION__, alpha_ref_value);
@@ -405,7 +402,7 @@ lp_setup_set_alpha_ref_value( struct setup_context *setup,
 }
 
 void
-lp_setup_set_blend_color( struct setup_context *setup,
+lp_setup_set_blend_color( struct lp_setup_context *setup,
                           const struct pipe_blend_color *blend_color )
 {
    LP_DBG(DEBUG_SETUP, "%s\n", __FUNCTION__);
@@ -420,7 +417,7 @@ lp_setup_set_blend_color( struct setup_context *setup,
 
 
 void
-lp_setup_set_scissor( struct setup_context *setup,
+lp_setup_set_scissor( struct lp_setup_context *setup,
                       const struct pipe_scissor_state *scissor )
 {
    LP_DBG(DEBUG_SETUP, "%s\n", __FUNCTION__);
@@ -435,7 +432,7 @@ lp_setup_set_scissor( struct setup_context *setup,
 
 
 void 
-lp_setup_set_flatshade_first( struct setup_context *setup,
+lp_setup_set_flatshade_first( struct lp_setup_context *setup,
                               boolean flatshade_first )
 {
    setup->flatshade_first = flatshade_first;
@@ -443,7 +440,7 @@ lp_setup_set_flatshade_first( struct setup_context *setup,
 
 
 void 
-lp_setup_set_vertex_info( struct setup_context *setup,
+lp_setup_set_vertex_info( struct lp_setup_context *setup,
                           struct vertex_info *vertex_info )
 {
    /* XXX: just silently holding onto the pointer:
@@ -456,7 +453,7 @@ lp_setup_set_vertex_info( struct setup_context *setup,
  * Called during state validation when LP_NEW_TEXTURE is set.
  */
 void
-lp_setup_set_sampler_textures( struct setup_context *setup,
+lp_setup_set_sampler_textures( struct lp_setup_context *setup,
                                unsigned num, struct pipe_texture **texture)
 {
    unsigned i;
@@ -474,20 +471,29 @@ lp_setup_set_sampler_textures( struct setup_context *setup,
          jit_tex = &setup->fs.current.jit_context.textures[i];
          jit_tex->width = tex->width0;
          jit_tex->height = tex->height0;
-         jit_tex->stride = lp_tex->stride[0];
-         if(!lp_tex->dt) {
-            jit_tex->data = lp_tex->data;
+         jit_tex->depth = tex->depth0;
+         jit_tex->last_level = tex->last_level;
+         if (!lp_tex->dt) {
+            /* regular texture - setup array of mipmap level pointers */
+            int j;
+            for (j = 0; j <= tex->last_level; j++) {
+               jit_tex->data[j] =
+                  (ubyte *) lp_tex->data + lp_tex->level_offset[j];
+               jit_tex->row_stride[j] = lp_tex->stride[j];
+            }
          }
          else {
+            /* display target texture/surface */
             /*
              * XXX: Where should this be unmapped?
              */
 
             struct llvmpipe_screen *screen = llvmpipe_screen(tex->screen);
-            struct llvmpipe_winsys *winsys = screen->winsys;
-            jit_tex->data = winsys->displaytarget_map(winsys, lp_tex->dt,
+            struct sw_winsys *winsys = screen->winsys;
+            jit_tex->data[0] = winsys->displaytarget_map(winsys, lp_tex->dt,
                                                       PIPE_BUFFER_USAGE_CPU_READ);
-            assert(jit_tex->data);
+            jit_tex->row_stride[0] = lp_tex->stride[0];
+            assert(jit_tex->data[0]);
          }
 
          /* the scene references this texture */
@@ -508,7 +514,7 @@ lp_setup_set_sampler_textures( struct setup_context *setup,
  * being rendered and the current scene being built.
  */
 unsigned
-lp_setup_is_texture_referenced( const struct setup_context *setup,
+lp_setup_is_texture_referenced( const struct lp_setup_context *setup,
                                 const struct pipe_texture *texture )
 {
    unsigned i;
@@ -537,7 +543,7 @@ lp_setup_is_texture_referenced( const struct setup_context *setup,
  * Called by vbuf code when we're about to draw something.
  */
 void
-lp_setup_update_state( struct setup_context *setup )
+lp_setup_update_state( struct lp_setup_context *setup )
 {
    struct lp_scene *scene = lp_setup_get_current_scene(setup);
 
@@ -655,7 +661,7 @@ lp_setup_update_state( struct setup_context *setup )
 /* Only caller is lp_setup_vbuf_destroy()
  */
 void 
-lp_setup_destroy( struct setup_context *setup )
+lp_setup_destroy( struct lp_setup_context *setup )
 {
    reset_context( setup );
 
@@ -680,12 +686,12 @@ lp_setup_destroy( struct setup_context *setup )
  * the draw module.  Currently also creates a rasterizer to use with
  * it.
  */
-struct setup_context *
-lp_setup_create( struct pipe_screen *screen,
+struct lp_setup_context *
+lp_setup_create( struct pipe_context *pipe,
                  struct draw_context *draw )
 {
    unsigned i;
-   struct setup_context *setup = CALLOC_STRUCT(setup_context);
+   struct lp_setup_context *setup = CALLOC_STRUCT(lp_setup_context);
 
    if (!setup)
       return NULL;
@@ -696,7 +702,9 @@ lp_setup_create( struct pipe_screen *screen,
    if (!setup->empty_scenes)
       goto fail;
 
-   setup->rast = lp_rast_create( screen, setup->empty_scenes );
+   /* XXX: move this to the screen and share between contexts:
+    */
+   setup->rast = lp_rast_create();
    if (!setup->rast) 
       goto fail;
 
@@ -709,7 +717,8 @@ lp_setup_create( struct pipe_screen *screen,
 
    /* create some empty scenes */
    for (i = 0; i < MAX_SCENES; i++) {
-      setup->scenes[i] = lp_scene_create();
+      setup->scenes[i] = lp_scene_create( pipe, setup->empty_scenes );
+
       lp_scene_enqueue(setup->empty_scenes, setup->scenes[i]);
    }
 
diff --git a/src/gallium/drivers/llvmpipe/lp_setup.h b/src/gallium/drivers/llvmpipe/lp_setup.h
index 0e155a7dc3..be1bf96f12 100644
--- a/src/gallium/drivers/llvmpipe/lp_setup.h
+++ b/src/gallium/drivers/llvmpipe/lp_setup.h
@@ -61,78 +61,79 @@ struct pipe_framebuffer_state;
 struct lp_fragment_shader;
 struct lp_jit_context;
 
-struct setup_context *
-lp_setup_create( struct pipe_screen *screen,
+struct lp_setup_context *
+lp_setup_create( struct pipe_context *pipe,
                  struct draw_context *draw );
 
 void
-lp_setup_clear(struct setup_context *setup,
+lp_setup_clear(struct lp_setup_context *setup,
                const float *clear_color,
                double clear_depth,
                unsigned clear_stencil,
                unsigned flags);
 
 struct pipe_fence_handle *
-lp_setup_fence( struct setup_context *setup );
+lp_setup_fence( struct lp_setup_context *setup );
 
 
 void
-lp_setup_flush( struct setup_context *setup,
+lp_setup_flush( struct lp_setup_context *setup,
                 unsigned flags );
 
 
 void
-lp_setup_bind_framebuffer( struct setup_context *setup,
+lp_setup_bind_framebuffer( struct lp_setup_context *setup,
                            const struct pipe_framebuffer_state *fb );
 
 void 
-lp_setup_set_triangle_state( struct setup_context *setup,
+lp_setup_set_triangle_state( struct lp_setup_context *setup,
                              unsigned cullmode,
                              boolean front_is_ccw,
-                             boolean scissor );
+                             boolean scissor,
+                             boolean gl_rasterization_rules );
 
 void
-lp_setup_set_fs_inputs( struct setup_context *setup,
+lp_setup_set_fs_inputs( struct lp_setup_context *setup,
                         const struct lp_shader_input *interp,
                         unsigned nr );
 
 void
-lp_setup_set_fs_functions( struct setup_context *setup,
+lp_setup_set_fs_functions( struct lp_setup_context *setup,
                            lp_jit_frag_func jit_function0,
                            lp_jit_frag_func jit_function1,
                            boolean opaque );
 
 void
-lp_setup_set_fs_constants(struct setup_context *setup,
+lp_setup_set_fs_constants(struct lp_setup_context *setup,
                           struct pipe_buffer *buffer);
 
 
 void
-lp_setup_set_alpha_ref_value( struct setup_context *setup,
+lp_setup_set_alpha_ref_value( struct lp_setup_context *setup,
                               float alpha_ref_value );
 
 void
-lp_setup_set_blend_color( struct setup_context *setup,
+lp_setup_set_blend_color( struct lp_setup_context *setup,
                           const struct pipe_blend_color *blend_color );
 
 void
-lp_setup_set_scissor( struct setup_context *setup,
+lp_setup_set_scissor( struct lp_setup_context *setup,
                       const struct pipe_scissor_state *scissor );
 
 void
-lp_setup_set_sampler_textures( struct setup_context *setup,
+lp_setup_set_sampler_textures( struct lp_setup_context *setup,
                                unsigned num, struct pipe_texture **texture);
 
 unsigned
-lp_setup_is_texture_referenced( const struct setup_context *setup,
+lp_setup_is_texture_referenced( const struct lp_setup_context *setup,
                                 const struct pipe_texture *texture );
 
 void
-lp_setup_set_flatshade_first( struct setup_context *setup, 
+lp_setup_set_flatshade_first( struct lp_setup_context *setup, 
                               boolean flatshade_first );
 
 void
-lp_setup_set_vertex_info( struct setup_context *setup, 
+lp_setup_set_vertex_info( struct lp_setup_context *setup, 
                           struct vertex_info *info );
 
 
diff --git a/src/gallium/drivers/llvmpipe/lp_setup_context.h b/src/gallium/drivers/llvmpipe/lp_setup_context.h
index a5fc34e54a..464fb36984 100644
--- a/src/gallium/drivers/llvmpipe/lp_setup_context.h
+++ b/src/gallium/drivers/llvmpipe/lp_setup_context.h
@@ -65,7 +65,7 @@ struct lp_scene_queue;
  * Subclass of vbuf_render, plugged directly into the draw module as
  * the rendering backend.
  */
-struct setup_context
+struct lp_setup_context
 {
    struct vbuf_render base;
 
@@ -89,6 +89,7 @@ struct setup_context
    boolean ccw_is_frontface;
    boolean scissor_test;
    unsigned cullmode;
+   float pixel_offset;
 
    struct pipe_framebuffer_state fb;
 
@@ -131,29 +132,29 @@ struct setup_context
 
    unsigned dirty;   /**< bitmask of LP_SETUP_NEW_x bits */
 
-   void (*point)( struct setup_context *,
+   void (*point)( struct lp_setup_context *,
                   const float (*v0)[4]);
 
-   void (*line)( struct setup_context *,
+   void (*line)( struct lp_setup_context *,
                  const float (*v0)[4],
                  const float (*v1)[4]);
 
-   void (*triangle)( struct setup_context *,
+   void (*triangle)( struct lp_setup_context *,
                      const float (*v0)[4],
                      const float (*v1)[4],
                      const float (*v2)[4]);
 };
 
-void lp_setup_choose_triangle( struct setup_context *setup );
-void lp_setup_choose_line( struct setup_context *setup );
-void lp_setup_choose_point( struct setup_context *setup );
+void lp_setup_choose_triangle( struct lp_setup_context *setup );
+void lp_setup_choose_line( struct lp_setup_context *setup );
+void lp_setup_choose_point( struct lp_setup_context *setup );
 
-struct lp_scene *lp_setup_get_current_scene(struct setup_context *setup);
+struct lp_scene *lp_setup_get_current_scene(struct lp_setup_context *setup);
 
-void lp_setup_init_vbuf(struct setup_context *setup);
+void lp_setup_init_vbuf(struct lp_setup_context *setup);
 
-void lp_setup_update_state( struct setup_context *setup );
+void lp_setup_update_state( struct lp_setup_context *setup );
 
-void lp_setup_destroy( struct setup_context *setup );
+void lp_setup_destroy( struct lp_setup_context *setup );
 
 #endif
diff --git a/src/gallium/drivers/llvmpipe/lp_setup_line.c b/src/gallium/drivers/llvmpipe/lp_setup_line.c
index feea79d394..be41c44e6f 100644
--- a/src/gallium/drivers/llvmpipe/lp_setup_line.c
+++ b/src/gallium/drivers/llvmpipe/lp_setup_line.c
@@ -31,7 +31,7 @@
 
 #include "lp_setup_context.h"
 
-static void line_nop( struct setup_context *setup,
+static void line_nop( struct lp_setup_context *setup,
                       const float (*v0)[4],
                       const float (*v1)[4] )
 {
@@ -39,7 +39,7 @@ static void line_nop( struct setup_context *setup,
 
 
 void 
-lp_setup_choose_line( struct setup_context *setup )
+lp_setup_choose_line( struct lp_setup_context *setup )
 {
    setup->line = line_nop;
 }
diff --git a/src/gallium/drivers/llvmpipe/lp_setup_point.c b/src/gallium/drivers/llvmpipe/lp_setup_point.c
index f03ca729b2..9f69e6c5ce 100644
--- a/src/gallium/drivers/llvmpipe/lp_setup_point.c
+++ b/src/gallium/drivers/llvmpipe/lp_setup_point.c
@@ -31,14 +31,14 @@
 
 #include "lp_setup_context.h"
 
-static void point_nop( struct setup_context *setup,
+static void point_nop( struct lp_setup_context *setup,
                        const float (*v0)[4] )
 {
 }
 
 
 void 
-lp_setup_choose_point( struct setup_context *setup )
+lp_setup_choose_point( struct lp_setup_context *setup )
 {
    setup->point = point_nop;
 }
diff --git a/src/gallium/drivers/llvmpipe/lp_setup_tri.c b/src/gallium/drivers/llvmpipe/lp_setup_tri.c
index a8bf540803..ac6264dc73 100644
--- a/src/gallium/drivers/llvmpipe/lp_setup_tri.c
+++ b/src/gallium/drivers/llvmpipe/lp_setup_tri.c
@@ -41,7 +41,8 @@
 /**
  * Compute a0 for a constant-valued coefficient (GL_FLAT shading).
  */
-static void constant_coef( struct lp_rast_triangle *tri,
+static void constant_coef( struct lp_setup_context *setup,
+                           struct lp_rast_triangle *tri,
                            unsigned slot,
 			   const float value,
                            unsigned i )
@@ -56,7 +57,8 @@ static void constant_coef( struct lp_rast_triangle *tri,
  * Compute a0, dadx and dady for a linearly interpolated coefficient,
  * for a triangle.
  */
-static void linear_coef( struct lp_rast_triangle *tri,
+static void linear_coef( struct lp_setup_context *setup,
+                         struct lp_rast_triangle *tri,
                          float oneoverarea,
                          unsigned slot,
                          const float (*v1)[4],
@@ -90,8 +92,8 @@ static void linear_coef( struct lp_rast_triangle *tri,
     * instead - i'll switch to this later.
     */
    tri->inputs.a0[slot][i] = (a1 -
-                              (dadx * (v1[0][0] - 0.5f) +
-                               dady * (v1[0][1] - 0.5f)));
+                              (dadx * (v1[0][0] - setup->pixel_offset) +
+                               dady * (v1[0][1] - setup->pixel_offset)));
 }
 
 
@@ -103,7 +105,8 @@ static void linear_coef( struct lp_rast_triangle *tri,
  * Later, when we compute the value at a particular fragment position we'll
  * divide the interpolated value by the interpolated W at that fragment.
  */
-static void perspective_coef( struct lp_rast_triangle *tri,
+static void perspective_coef( struct lp_setup_context *setup,
+                              struct lp_rast_triangle *tri,
                               float oneoverarea,
                               unsigned slot,
 			      const float (*v1)[4],
@@ -125,8 +128,8 @@ static void perspective_coef( struct lp_rast_triangle *tri,
    tri->inputs.dadx[slot][i] = dadx;
    tri->inputs.dady[slot][i] = dady;
    tri->inputs.a0[slot][i] = (a1 -
-                              (dadx * (v1[0][0] - 0.5f) +
-                               dady * (v1[0][1] - 0.5f)));
+                              (dadx * (v1[0][0] - setup->pixel_offset) +
+                               dady * (v1[0][1] - setup->pixel_offset)));
 }
 
 
@@ -137,7 +140,8 @@ static void perspective_coef( struct lp_rast_triangle *tri,
  * We could do a bit less work if we'd examine gl_FragCoord's swizzle mask.
  */
 static void
-setup_fragcoord_coef(struct lp_rast_triangle *tri,
+setup_fragcoord_coef(struct lp_setup_context *setup,
+                     struct lp_rast_triangle *tri,
                      float oneoverarea,
                      unsigned slot,
                      const float (*v1)[4],
@@ -153,27 +157,28 @@ setup_fragcoord_coef(struct lp_rast_triangle *tri,
    tri->inputs.dadx[slot][1] = 0.0;
    tri->inputs.dady[slot][1] = 1.0;
    /*Z*/
-   linear_coef(tri, oneoverarea, slot, v1, v2, v3, 0, 2);
+   linear_coef(setup, tri, oneoverarea, slot, v1, v2, v3, 0, 2);
    /*W*/
-   linear_coef(tri, oneoverarea, slot, v1, v2, v3, 0, 3);
+   linear_coef(setup, tri, oneoverarea, slot, v1, v2, v3, 0, 3);
 }
 
 
-static void setup_facing_coef( struct lp_rast_triangle *tri,
+static void setup_facing_coef( struct lp_setup_context *setup,
+                               struct lp_rast_triangle *tri,
                                unsigned slot,
                                boolean frontface )
 {
-   constant_coef( tri, slot, 1.0f - frontface, 0 );
-   constant_coef( tri, slot, 0.0f, 1 ); /* wasted */
-   constant_coef( tri, slot, 0.0f, 2 ); /* wasted */
-   constant_coef( tri, slot, 0.0f, 3 ); /* wasted */
+   constant_coef( setup, tri, slot, 1.0f - frontface, 0 );
+   constant_coef( setup, tri, slot, 0.0f, 1 ); /* wasted */
+   constant_coef( setup, tri, slot, 0.0f, 2 ); /* wasted */
+   constant_coef( setup, tri, slot, 0.0f, 3 ); /* wasted */
 }
 
 
 /**
  * Compute the tri->coef[] array dadx, dady, a0 values.
  */
-static void setup_tri_coefficients( struct setup_context *setup,
+static void setup_tri_coefficients( struct lp_setup_context *setup,
 				    struct lp_rast_triangle *tri,
                                     float oneoverarea,
 				    const float (*v1)[4],
@@ -185,7 +190,7 @@ static void setup_tri_coefficients( struct setup_context *setup,
 
    /* The internal position input is in slot zero:
     */
-   setup_fragcoord_coef(tri, oneoverarea, 0, v1, v2, v3);
+   setup_fragcoord_coef(setup, tri, oneoverarea, 0, v1, v2, v3);
 
    /* setup interpolation for all the remaining attributes:
     */
@@ -196,27 +201,27 @@ static void setup_tri_coefficients( struct setup_context *setup,
       switch (setup->fs.input[slot].interp) {
       case LP_INTERP_CONSTANT:
          for (i = 0; i < NUM_CHANNELS; i++)
-            constant_coef(tri, slot+1, v3[vert_attr][i], i);
+            constant_coef(setup, tri, slot+1, v3[vert_attr][i], i);
          break;
 
       case LP_INTERP_LINEAR:
          for (i = 0; i < NUM_CHANNELS; i++)
-            linear_coef(tri, oneoverarea, slot+1, v1, v2, v3, vert_attr, i);
+            linear_coef(setup, tri, oneoverarea, slot+1, v1, v2, v3, vert_attr, i);
          break;
 
       case LP_INTERP_PERSPECTIVE:
          for (i = 0; i < NUM_CHANNELS; i++)
-            perspective_coef(tri, oneoverarea, slot+1, v1, v2, v3, vert_attr, i);
+            perspective_coef(setup, tri, oneoverarea, slot+1, v1, v2, v3, vert_attr, i);
          break;
 
       case LP_INTERP_POSITION:
          /* XXX: fix me - duplicates the values in slot zero.
           */
-         setup_fragcoord_coef(tri, oneoverarea, slot+1, v1, v2, v3);
+         setup_fragcoord_coef(setup, tri, oneoverarea, slot+1, v1, v2, v3);
          break;
 
       case LP_INTERP_FACING:
-         setup_facing_coef(tri, slot+1, frontface);
+         setup_facing_coef(setup, tri, slot+1, frontface);
          break;
 
       default:
@@ -274,19 +279,19 @@ alloc_triangle(struct lp_scene *scene, unsigned nr_inputs, unsigned *tri_size)
  * bins for the tiles which we overlap.
  */
 static void 
-do_triangle_ccw(struct setup_context *setup,
+do_triangle_ccw(struct lp_setup_context *setup,
 		const float (*v1)[4],
 		const float (*v2)[4],
 		const float (*v3)[4],
 		boolean frontfacing )
 {
    /* x/y positions in fixed point */
-   const int x1 = subpixel_snap(v1[0][0]);
-   const int x2 = subpixel_snap(v2[0][0]);
-   const int x3 = subpixel_snap(v3[0][0]);
-   const int y1 = subpixel_snap(v1[0][1]);
-   const int y2 = subpixel_snap(v2[0][1]);
-   const int y3 = subpixel_snap(v3[0][1]);
+   const int x1 = subpixel_snap(v1[0][0] + 0.5 - setup->pixel_offset);
+   const int x2 = subpixel_snap(v2[0][0] + 0.5 - setup->pixel_offset);
+   const int x3 = subpixel_snap(v3[0][0] + 0.5 - setup->pixel_offset);
+   const int y1 = subpixel_snap(v1[0][1] + 0.5 - setup->pixel_offset);
+   const int y2 = subpixel_snap(v2[0][1] + 0.5 - setup->pixel_offset);
+   const int y3 = subpixel_snap(v3[0][1] + 0.5 - setup->pixel_offset);
 
    struct lp_scene *scene = lp_setup_get_current_scene(setup);
    struct lp_rast_triangle *tri;
@@ -297,6 +302,15 @@ do_triangle_ccw(struct setup_context *setup,
 
    tri = alloc_triangle(scene, setup->fs.nr_inputs, &tri_bytes);
 
+#ifdef DEBUG
+   tri->v[0][0] = v1[0][0];
+   tri->v[1][0] = v2[0][0];
+   tri->v[2][0] = v3[0][0];
+   tri->v[0][1] = v1[0][1];
+   tri->v[1][1] = v2[0][1];
+   tri->v[2][1] = v3[0][1];
+#endif
+
    tri->dx12 = x1 - x2;
    tri->dx23 = x2 - x3;
    tri->dx31 = x3 - x1;
@@ -556,7 +570,7 @@ do_triangle_ccw(struct setup_context *setup,
 }
 
 
-static void triangle_cw( struct setup_context *setup,
+static void triangle_cw( struct lp_setup_context *setup,
 			 const float (*v0)[4],
 			 const float (*v1)[4],
 			 const float (*v2)[4] )
@@ -565,7 +579,7 @@ static void triangle_cw( struct setup_context *setup,
 }
 
 
-static void triangle_ccw( struct setup_context *setup,
+static void triangle_ccw( struct lp_setup_context *setup,
 			 const float (*v0)[4],
 			 const float (*v1)[4],
 			 const float (*v2)[4] )
@@ -574,7 +588,7 @@ static void triangle_ccw( struct setup_context *setup,
 }
 
 
-static void triangle_both( struct setup_context *setup,
+static void triangle_both( struct lp_setup_context *setup,
 			   const float (*v0)[4],
 			   const float (*v1)[4],
 			   const float (*v2)[4] )
@@ -593,7 +607,7 @@ static void triangle_both( struct setup_context *setup,
 }
 
 
-static void triangle_nop( struct setup_context *setup,
+static void triangle_nop( struct lp_setup_context *setup,
 			  const float (*v0)[4],
 			  const float (*v1)[4],
 			  const float (*v2)[4] )
@@ -602,7 +616,7 @@ static void triangle_nop( struct setup_context *setup,
 
 
 void 
-lp_setup_choose_triangle( struct setup_context *setup )
+lp_setup_choose_triangle( struct lp_setup_context *setup )
 {
    switch (setup->cullmode) {
    case PIPE_WINDING_NONE:
diff --git a/src/gallium/drivers/llvmpipe/lp_setup_vbuf.c b/src/gallium/drivers/llvmpipe/lp_setup_vbuf.c
index 24291da91e..d7336d82b2 100644
--- a/src/gallium/drivers/llvmpipe/lp_setup_vbuf.c
+++ b/src/gallium/drivers/llvmpipe/lp_setup_vbuf.c
@@ -48,10 +48,10 @@
   
 
 /** cast wrapper */
-static struct setup_context *
-setup_context(struct vbuf_render *vbr)
+static struct lp_setup_context *
+lp_setup_context(struct vbuf_render *vbr)
 {
-   return (struct setup_context *) vbr;
+   return (struct lp_setup_context *) vbr;
 }
 
 
@@ -59,7 +59,7 @@ setup_context(struct vbuf_render *vbr)
 static const struct vertex_info *
 lp_setup_get_vertex_info(struct vbuf_render *vbr)
 {
-   struct setup_context *setup = setup_context(vbr);
+   struct lp_setup_context *setup = lp_setup_context(vbr);
    return setup->vertex_info;
 }
 
@@ -68,7 +68,7 @@ static boolean
 lp_setup_allocate_vertices(struct vbuf_render *vbr,
                           ushort vertex_size, ushort nr_vertices)
 {
-   struct setup_context *setup = setup_context(vbr);
+   struct lp_setup_context *setup = lp_setup_context(vbr);
    unsigned size = vertex_size * nr_vertices;
 
    if (setup->vertex_buffer_size < size) {
@@ -92,7 +92,7 @@ lp_setup_release_vertices(struct vbuf_render *vbr)
 static void *
 lp_setup_map_vertices(struct vbuf_render *vbr)
 {
-   struct setup_context *setup = setup_context(vbr);
+   struct lp_setup_context *setup = lp_setup_context(vbr);
    return setup->vertex_buffer;
 }
 
@@ -101,7 +101,7 @@ lp_setup_unmap_vertices(struct vbuf_render *vbr,
                        ushort min_index,
                        ushort max_index )
 {
-   struct setup_context *setup = setup_context(vbr);
+   struct lp_setup_context *setup = lp_setup_context(vbr);
    assert( setup->vertex_buffer_size >= (max_index+1) * setup->vertex_size );
    /* do nothing */
 }
@@ -110,7 +110,7 @@ lp_setup_unmap_vertices(struct vbuf_render *vbr,
 static boolean
 lp_setup_set_primitive(struct vbuf_render *vbr, unsigned prim)
 {
-   setup_context(vbr)->prim = prim;
+   lp_setup_context(vbr)->prim = prim;
    return TRUE;
 }
 
@@ -129,7 +129,7 @@ static INLINE const_float4_ptr get_vert( const void *vertex_buffer,
 static void
 lp_setup_draw(struct vbuf_render *vbr, const ushort *indices, uint nr)
 {
-   struct setup_context *setup = setup_context(vbr);
+   struct lp_setup_context *setup = lp_setup_context(vbr);
    const unsigned stride = setup->vertex_info->size * sizeof(float);
    const void *vertex_buffer = setup->vertex_buffer;
    unsigned i;
@@ -231,57 +231,29 @@ lp_setup_draw(struct vbuf_render *vbr, const ushort *indices, uint nr)
       break;
 
    case PIPE_PRIM_QUADS:
-      if (setup->flatshade_first) {
-         for (i = 3; i < nr; i += 4) {
-            setup->triangle( setup,
-                             get_vert(vertex_buffer, indices[i-2], stride),
-                             get_vert(vertex_buffer, indices[i-1], stride),
-                             get_vert(vertex_buffer, indices[i-3], stride) );
-            setup->triangle( setup,
-                             get_vert(vertex_buffer, indices[i-1], stride),
-                             get_vert(vertex_buffer, indices[i-0], stride),
-                             get_vert(vertex_buffer, indices[i-3], stride) );
-         }
-      }
-      else {
-         for (i = 3; i < nr; i += 4) {
-            setup->triangle( setup,
-                             get_vert(vertex_buffer, indices[i-3], stride),
-                             get_vert(vertex_buffer, indices[i-2], stride),
-                             get_vert(vertex_buffer, indices[i-0], stride) );
+      for (i = 3; i < nr; i += 4) {
+         setup->triangle( setup,
+                          get_vert(vertex_buffer, indices[i-3], stride),
+                          get_vert(vertex_buffer, indices[i-2], stride),
+                          get_vert(vertex_buffer, indices[i-0], stride) );
 
-            setup->triangle( setup,
-                             get_vert(vertex_buffer, indices[i-2], stride),
-                             get_vert(vertex_buffer, indices[i-1], stride),
-                             get_vert(vertex_buffer, indices[i-0], stride) );
-         }
+         setup->triangle( setup,
+                          get_vert(vertex_buffer, indices[i-2], stride),
+                          get_vert(vertex_buffer, indices[i-1], stride),
+                          get_vert(vertex_buffer, indices[i-0], stride) );
       }
       break;
 
    case PIPE_PRIM_QUAD_STRIP:
-      if (setup->flatshade_first) {
-         for (i = 3; i < nr; i += 2) {
-            setup->triangle( setup,
-                             get_vert(vertex_buffer, indices[i-0], stride),
-                             get_vert(vertex_buffer, indices[i-1], stride),
-                             get_vert(vertex_buffer, indices[i-3], stride));
-            setup->triangle( setup,
-                             get_vert(vertex_buffer, indices[i-2], stride),
-                             get_vert(vertex_buffer, indices[i-0], stride),
-                             get_vert(vertex_buffer, indices[i-3], stride) );
-         }
-      }
-      else {
-         for (i = 3; i < nr; i += 2) {
-            setup->triangle( setup,
-                             get_vert(vertex_buffer, indices[i-3], stride),
-                             get_vert(vertex_buffer, indices[i-2], stride),
-                             get_vert(vertex_buffer, indices[i-0], stride) );
-            setup->triangle( setup,
-                             get_vert(vertex_buffer, indices[i-1], stride),
-                             get_vert(vertex_buffer, indices[i-3], stride),
-                             get_vert(vertex_buffer, indices[i-0], stride) );
-         }
+      for (i = 3; i < nr; i += 2) {
+         setup->triangle( setup,
+                          get_vert(vertex_buffer, indices[i-3], stride),
+                          get_vert(vertex_buffer, indices[i-2], stride),
+                          get_vert(vertex_buffer, indices[i-0], stride) );
+         setup->triangle( setup,
+                          get_vert(vertex_buffer, indices[i-1], stride),
+                          get_vert(vertex_buffer, indices[i-3], stride),
+                          get_vert(vertex_buffer, indices[i-0], stride) );
       }
       break;
 
@@ -312,7 +284,7 @@ lp_setup_draw(struct vbuf_render *vbr, const ushort *indices, uint nr)
 static void
 lp_setup_draw_arrays(struct vbuf_render *vbr, uint start, uint nr)
 {
-   struct setup_context *setup = setup_context(vbr);
+   struct lp_setup_context *setup = lp_setup_context(vbr);
    const unsigned stride = setup->vertex_info->size * sizeof(float);
    const void *vertex_buffer =
       (void *) get_vert(setup->vertex_buffer, start, stride);
@@ -415,57 +387,28 @@ lp_setup_draw_arrays(struct vbuf_render *vbr, uint start, uint nr)
       break;
 
    case PIPE_PRIM_QUADS:
-      if (setup->flatshade_first) {
-         for (i = 3; i < nr; i += 4) {
-            setup->triangle( setup,
-                             get_vert(vertex_buffer, i-2, stride),
-                             get_vert(vertex_buffer, i-1, stride),
-                             get_vert(vertex_buffer, i-3, stride) );
-            setup->triangle( setup,
-                             get_vert(vertex_buffer, i-1, stride),
-                             get_vert(vertex_buffer, i-0, stride),
-                             get_vert(vertex_buffer, i-3, stride) );
-         }
-      }
-      else {
-         for (i = 3; i < nr; i += 4) {
-            setup->triangle( setup,
-                             get_vert(vertex_buffer, i-3, stride),
-                             get_vert(vertex_buffer, i-2, stride),
-                             get_vert(vertex_buffer, i-0, stride) );
-            setup->triangle( setup,
-                             get_vert(vertex_buffer, i-2, stride),
-                             get_vert(vertex_buffer, i-1, stride),
-                             get_vert(vertex_buffer, i-0, stride) );
-         }
+      for (i = 3; i < nr; i += 4) {
+         setup->triangle( setup,
+                          get_vert(vertex_buffer, i-3, stride),
+                          get_vert(vertex_buffer, i-2, stride),
+                          get_vert(vertex_buffer, i-0, stride) );
+         setup->triangle( setup,
+                          get_vert(vertex_buffer, i-2, stride),
+                          get_vert(vertex_buffer, i-1, stride),
+                          get_vert(vertex_buffer, i-0, stride) );
       }
       break;
 
    case PIPE_PRIM_QUAD_STRIP:
-      if (setup->flatshade_first) {
-         for (i = 3; i < nr; i += 2) {
-            setup->triangle( setup,
-                             get_vert(vertex_buffer, i-0, stride),
-                             get_vert(vertex_buffer, i-1, stride),
-                             get_vert(vertex_buffer, i-3, stride) );
-            setup->triangle( setup,
-
-                             get_vert(vertex_buffer, i-2, stride),
-                             get_vert(vertex_buffer, i-0, stride),
-                             get_vert(vertex_buffer, i-3, stride) );
-         }
-      }
-      else {
-         for (i = 3; i < nr; i += 2) {
-            setup->triangle( setup,
-                             get_vert(vertex_buffer, i-3, stride),
-                             get_vert(vertex_buffer, i-2, stride),
-                             get_vert(vertex_buffer, i-0, stride) );
-            setup->triangle( setup,
-                             get_vert(vertex_buffer, i-1, stride),
-                             get_vert(vertex_buffer, i-3, stride),
-                             get_vert(vertex_buffer, i-0, stride) );
-         }
+      for (i = 3; i < nr; i += 2) {
+         setup->triangle( setup,
+                          get_vert(vertex_buffer, i-3, stride),
+                          get_vert(vertex_buffer, i-2, stride),
+                          get_vert(vertex_buffer, i-0, stride) );
+         setup->triangle( setup,
+                          get_vert(vertex_buffer, i-1, stride),
+                          get_vert(vertex_buffer, i-3, stride),
+                          get_vert(vertex_buffer, i-0, stride) );
       }
       break;
 
@@ -493,7 +436,7 @@ lp_setup_draw_arrays(struct vbuf_render *vbr, uint start, uint nr)
 static void
 lp_setup_vbuf_destroy(struct vbuf_render *vbr)
 {
-   lp_setup_destroy(setup_context(vbr));
+   lp_setup_destroy(lp_setup_context(vbr));
 }
 
 
@@ -501,7 +444,7 @@ lp_setup_vbuf_destroy(struct vbuf_render *vbr)
  * Create the post-transform vertex handler for the given context.
  */
 void
-lp_setup_init_vbuf(struct setup_context *setup)
+lp_setup_init_vbuf(struct lp_setup_context *setup)
 {
    setup->base.max_indices = LP_MAX_VBUF_INDEXES;
    setup->base.max_vertex_buffer_bytes = LP_MAX_VBUF_SIZE;
diff --git a/src/gallium/drivers/llvmpipe/lp_state.h b/src/gallium/drivers/llvmpipe/lp_state.h
index 9beba32271..a5a1a72074 100644
--- a/src/gallium/drivers/llvmpipe/lp_state.h
+++ b/src/gallium/drivers/llvmpipe/lp_state.h
@@ -31,7 +31,7 @@
 #ifndef LP_STATE_H
 #define LP_STATE_H
 
-#include <llvm-c/Core.h>
+#include "os/os_llvm.h"
 
 #include "pipe/p_state.h"
 #include "tgsi/tgsi_scan.h"
@@ -119,6 +119,10 @@ struct lp_vertex_shader {
    struct draw_vertex_shader *draw_data;
 };
 
+struct lp_velems_state {
+   unsigned count;
+   struct pipe_vertex_element velem[PIPE_MAX_ATTRIBS];
+};
 
 
 void *
@@ -176,8 +180,14 @@ void *llvmpipe_create_vs_state(struct pipe_context *,
 void llvmpipe_bind_vs_state(struct pipe_context *, void *);
 void llvmpipe_delete_vs_state(struct pipe_context *, void *);
 
+void *llvmpipe_create_vertex_elements_state(struct pipe_context *,
+                                            unsigned count,
+                                            const struct pipe_vertex_element *);
+void llvmpipe_bind_vertex_elements_state(struct pipe_context *, void *);
+void llvmpipe_delete_vertex_elements_state(struct pipe_context *, void *);
+
 void llvmpipe_set_polygon_stipple( struct pipe_context *,
-				  const struct pipe_poly_stipple * );
+                                   const struct pipe_poly_stipple * );
 
 void llvmpipe_set_scissor_state( struct pipe_context *,
                                  const struct pipe_scissor_state * );
@@ -194,10 +204,6 @@ llvmpipe_set_vertex_sampler_textures(struct pipe_context *,
 void llvmpipe_set_viewport_state( struct pipe_context *,
                                   const struct pipe_viewport_state * );
 
-void llvmpipe_set_vertex_elements(struct pipe_context *,
-                                  unsigned count,
-                                  const struct pipe_vertex_element *);
-
 void llvmpipe_set_vertex_buffers(struct pipe_context *,
                                  unsigned count,
                                  const struct pipe_vertex_buffer *);
diff --git a/src/gallium/drivers/llvmpipe/lp_state_fs.c b/src/gallium/drivers/llvmpipe/lp_state_fs.c
index 90dae3f910..9a8de0edfd 100644
--- a/src/gallium/drivers/llvmpipe/lp_state_fs.c
+++ b/src/gallium/drivers/llvmpipe/lp_state_fs.c
@@ -40,7 +40,7 @@
  * - depth/stencil test (stencil TBI)
  * - blending
  *
- * This file has only the glue to assembly the fragment pipeline.  The actual
+ * This file has only the glue to assemble the fragment pipeline.  The actual
  * plumbing of converting Gallium state into LLVM IR is done elsewhere, in the
  * lp_bld_*.[ch] files, and in a complete generic and reusable way. Here we
  * muster the LLVM JIT execution engine to create a function that follows an
@@ -95,6 +95,9 @@
 #include "lp_tex_sample.h"
 
 
+#include <llvm-c/Analysis.h>
+
+
 static const unsigned char quad_offset_x[4] = {0, 1, 0, 1};
 static const unsigned char quad_offset_y[4] = {0, 0, 1, 1};
 
@@ -974,6 +977,13 @@ llvmpipe_delete_fs_state(struct pipe_context *pipe, void *fs)
    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
+    */
+   draw_flush(llvmpipe->draw);
+   lp_setup_flush(llvmpipe->setup, 0);
+
    variant = shader->variants;
    while(variant) {
       struct lp_fragment_shader_variant *next = variant->next;
diff --git a/src/gallium/drivers/llvmpipe/lp_state_rasterizer.c b/src/gallium/drivers/llvmpipe/lp_state_rasterizer.c
index feb012816c..6df3ef25b0 100644
--- a/src/gallium/drivers/llvmpipe/lp_state_rasterizer.c
+++ b/src/gallium/drivers/llvmpipe/lp_state_rasterizer.c
@@ -62,7 +62,8 @@ void llvmpipe_bind_rasterizer_state(struct pipe_context *pipe,
       lp_setup_set_triangle_state( llvmpipe->setup,
                    llvmpipe->rasterizer->cull_mode,
                    llvmpipe->rasterizer->front_winding == PIPE_WINDING_CCW,
-                   llvmpipe->rasterizer->scissor);
+                   llvmpipe->rasterizer->scissor,
+                   llvmpipe->rasterizer->gl_rasterization_rules);
    }
 
    llvmpipe->dirty |= LP_NEW_RASTERIZER;
diff --git a/src/gallium/drivers/llvmpipe/lp_state_vertex.c b/src/gallium/drivers/llvmpipe/lp_state_vertex.c
index 57ac25ea0c..f6427aa908 100644
--- a/src/gallium/drivers/llvmpipe/lp_state_vertex.c
+++ b/src/gallium/drivers/llvmpipe/lp_state_vertex.c
@@ -35,24 +35,41 @@
 #include "draw/draw_context.h"
 
 
+void *
+llvmpipe_create_vertex_elements_state(struct pipe_context *pipe,
+                                      unsigned count,
+                                      const struct pipe_vertex_element *attribs)
+{
+   struct lp_velems_state *velems;
+   assert(count <= PIPE_MAX_ATTRIBS);
+   velems = (struct lp_velems_state *) MALLOC(sizeof(struct lp_velems_state));
+   if (velems) {
+      velems->count = count;
+      memcpy(velems->velem, attribs, sizeof(*attribs) * count);
+   }
+   return velems;
+}
+
 void
-llvmpipe_set_vertex_elements(struct pipe_context *pipe,
-                             unsigned count,
-                             const struct pipe_vertex_element *attribs)
+llvmpipe_bind_vertex_elements_state(struct pipe_context *pipe,
+                                    void *velems)
 {
    struct llvmpipe_context *llvmpipe = llvmpipe_context(pipe);
+   struct lp_velems_state *lp_velems = (struct lp_velems_state *) velems;
 
-   assert(count <= PIPE_MAX_ATTRIBS);
-
-   memcpy(llvmpipe->vertex_element, attribs,
-          count * sizeof(struct pipe_vertex_element));
-   llvmpipe->num_vertex_elements = count;
+   llvmpipe->velems = lp_velems;
 
    llvmpipe->dirty |= LP_NEW_VERTEX;
 
-   draw_set_vertex_elements(llvmpipe->draw, count, attribs);
+   if (velems)
+      draw_set_vertex_elements(llvmpipe->draw, lp_velems->count, lp_velems->velem);
 }
 
+void
+llvmpipe_delete_vertex_elements_state(struct pipe_context *pipe, void *velems)
+{
+   FREE( velems );
+}
 
 void
 llvmpipe_set_vertex_buffers(struct pipe_context *pipe,
diff --git a/src/gallium/drivers/llvmpipe/lp_surface.c b/src/gallium/drivers/llvmpipe/lp_surface.c
index 6110b0a193..ca3d62c361 100644
--- a/src/gallium/drivers/llvmpipe/lp_surface.c
+++ b/src/gallium/drivers/llvmpipe/lp_surface.c
@@ -27,6 +27,7 @@
 
 #include "util/u_rect.h"
 #include "lp_context.h"
+#include "lp_flush.h"
 #include "lp_surface.h"
 
 
@@ -36,6 +37,20 @@ lp_surface_copy(struct pipe_context *pipe,
                 struct pipe_surface *src, unsigned srcx, unsigned srcy,
                 unsigned width, unsigned height)
 {
+   llvmpipe_flush_texture(pipe,
+                          dest->texture, dest->face, dest->level,
+                          0, /* flush_flags */
+                          FALSE, /* read_only */
+                          FALSE, /* cpu_access */
+                          FALSE); /* do_not_flush */
+
+   llvmpipe_flush_texture(pipe,
+                          src->texture, src->face, src->level,
+                          0, /* flush_flags */
+                          TRUE, /* read_only */
+                          FALSE, /* cpu_access */
+                          FALSE); /* do_not_flush */
+
    util_surface_copy(pipe, FALSE,
                      dest, destx, desty,
                      src, srcx, srcy,
diff --git a/src/gallium/drivers/llvmpipe/lp_test.h b/src/gallium/drivers/llvmpipe/lp_test.h
index a9b99945f9..1df9897898 100644
--- a/src/gallium/drivers/llvmpipe/lp_test.h
+++ b/src/gallium/drivers/llvmpipe/lp_test.h
@@ -41,7 +41,7 @@
 #include <stdio.h>
 #include <float.h>
 
-#include <llvm-c/Core.h>
+#include "os/os_llvm.h"
 #include <llvm-c/Analysis.h>
 #include <llvm-c/ExecutionEngine.h>
 #include <llvm-c/Target.h>
diff --git a/src/gallium/drivers/llvmpipe/lp_test_format.c b/src/gallium/drivers/llvmpipe/lp_test_format.c
index 48828bd0a0..2c4d7fb6e1 100644
--- a/src/gallium/drivers/llvmpipe/lp_test_format.c
+++ b/src/gallium/drivers/llvmpipe/lp_test_format.c
@@ -29,7 +29,7 @@
 #include <stdlib.h>
 #include <stdio.h>
 
-#include <llvm-c/Core.h>
+#include "os/os_llvm.h"
 #include <llvm-c/Analysis.h>
 #include <llvm-c/ExecutionEngine.h>
 #include <llvm-c/Target.h>
@@ -52,25 +52,25 @@ struct pixel_test_case
 
 struct pixel_test_case test_cases[] =
 {
-   {PIPE_FORMAT_R5G6B5_UNORM,   0x0000, {0.0, 0.0, 0.0, 1.0}},
-   {PIPE_FORMAT_R5G6B5_UNORM,   0x001f, {0.0, 0.0, 1.0, 1.0}},
-   {PIPE_FORMAT_R5G6B5_UNORM,   0x07e0, {0.0, 1.0, 0.0, 1.0}},
-   {PIPE_FORMAT_R5G6B5_UNORM,   0xf800, {1.0, 0.0, 0.0, 1.0}},
-   {PIPE_FORMAT_R5G6B5_UNORM,   0xffff, {1.0, 1.0, 1.0, 1.0}},
-
-   {PIPE_FORMAT_A1R5G5B5_UNORM, 0x0000, {0.0, 0.0, 0.0, 0.0}},
-   {PIPE_FORMAT_A1R5G5B5_UNORM, 0x001f, {0.0, 0.0, 1.0, 0.0}},
-   {PIPE_FORMAT_A1R5G5B5_UNORM, 0x03e0, {0.0, 1.0, 0.0, 0.0}},
-   {PIPE_FORMAT_A1R5G5B5_UNORM, 0x7c00, {1.0, 0.0, 0.0, 0.0}},
-   {PIPE_FORMAT_A1R5G5B5_UNORM, 0x8000, {0.0, 0.0, 0.0, 1.0}},
-   {PIPE_FORMAT_A1R5G5B5_UNORM, 0xffff, {1.0, 1.0, 1.0, 1.0}},
+   {PIPE_FORMAT_B5G6R5_UNORM,   0x0000, {0.0, 0.0, 0.0, 1.0}},
+   {PIPE_FORMAT_B5G6R5_UNORM,   0x001f, {0.0, 0.0, 1.0, 1.0}},
+   {PIPE_FORMAT_B5G6R5_UNORM,   0x07e0, {0.0, 1.0, 0.0, 1.0}},
+   {PIPE_FORMAT_B5G6R5_UNORM,   0xf800, {1.0, 0.0, 0.0, 1.0}},
+   {PIPE_FORMAT_B5G6R5_UNORM,   0xffff, {1.0, 1.0, 1.0, 1.0}},
+
+   {PIPE_FORMAT_B5G5R5A1_UNORM, 0x0000, {0.0, 0.0, 0.0, 0.0}},
+   {PIPE_FORMAT_B5G5R5A1_UNORM, 0x001f, {0.0, 0.0, 1.0, 0.0}},
+   {PIPE_FORMAT_B5G5R5A1_UNORM, 0x03e0, {0.0, 1.0, 0.0, 0.0}},
+   {PIPE_FORMAT_B5G5R5A1_UNORM, 0x7c00, {1.0, 0.0, 0.0, 0.0}},
+   {PIPE_FORMAT_B5G5R5A1_UNORM, 0x8000, {0.0, 0.0, 0.0, 1.0}},
+   {PIPE_FORMAT_B5G5R5A1_UNORM, 0xffff, {1.0, 1.0, 1.0, 1.0}},
 
-   {PIPE_FORMAT_A8R8G8B8_UNORM, 0x00000000, {0.0, 0.0, 0.0, 0.0}},
-   {PIPE_FORMAT_A8R8G8B8_UNORM, 0x000000ff, {0.0, 0.0, 1.0, 0.0}},
-   {PIPE_FORMAT_A8R8G8B8_UNORM, 0x0000ff00, {0.0, 1.0, 0.0, 0.0}},
-   {PIPE_FORMAT_A8R8G8B8_UNORM, 0x00ff0000, {1.0, 0.0, 0.0, 0.0}},
-   {PIPE_FORMAT_A8R8G8B8_UNORM, 0xff000000, {0.0, 0.0, 0.0, 1.0}},
-   {PIPE_FORMAT_A8R8G8B8_UNORM, 0xffffffff, {1.0, 1.0, 1.0, 1.0}},
+   {PIPE_FORMAT_B8G8R8A8_UNORM, 0x00000000, {0.0, 0.0, 0.0, 0.0}},
+   {PIPE_FORMAT_B8G8R8A8_UNORM, 0x000000ff, {0.0, 0.0, 1.0, 0.0}},
+   {PIPE_FORMAT_B8G8R8A8_UNORM, 0x0000ff00, {0.0, 1.0, 0.0, 0.0}},
+   {PIPE_FORMAT_B8G8R8A8_UNORM, 0x00ff0000, {1.0, 0.0, 0.0, 0.0}},
+   {PIPE_FORMAT_B8G8R8A8_UNORM, 0xff000000, {0.0, 0.0, 0.0, 1.0}},
+   {PIPE_FORMAT_B8G8R8A8_UNORM, 0xffffffff, {1.0, 1.0, 1.0, 1.0}},
 
 #if 0
    {PIPE_FORMAT_R8G8B8A8_UNORM, 0x00000000, {0.0, 0.0, 0.0, 0.0}},
@@ -81,12 +81,12 @@ struct pixel_test_case test_cases[] =
    {PIPE_FORMAT_R8G8B8A8_UNORM, 0xffffffff, {1.0, 1.0, 1.0, 1.0}},
 #endif
 
-   {PIPE_FORMAT_B8G8R8A8_UNORM, 0x00000000, {0.0, 0.0, 0.0, 0.0}},
-   {PIPE_FORMAT_B8G8R8A8_UNORM, 0x000000ff, {0.0, 0.0, 0.0, 1.0}},
-   {PIPE_FORMAT_B8G8R8A8_UNORM, 0x0000ff00, {1.0, 0.0, 0.0, 0.0}},
-   {PIPE_FORMAT_B8G8R8A8_UNORM, 0x00ff0000, {0.0, 1.0, 0.0, 0.0}},
-   {PIPE_FORMAT_B8G8R8A8_UNORM, 0xff000000, {0.0, 0.0, 1.0, 0.0}},
-   {PIPE_FORMAT_B8G8R8A8_UNORM, 0xffffffff, {1.0, 1.0, 1.0, 1.0}},
+   {PIPE_FORMAT_A8R8G8B8_UNORM, 0x00000000, {0.0, 0.0, 0.0, 0.0}},
+   {PIPE_FORMAT_A8R8G8B8_UNORM, 0x000000ff, {0.0, 0.0, 0.0, 1.0}},
+   {PIPE_FORMAT_A8R8G8B8_UNORM, 0x0000ff00, {1.0, 0.0, 0.0, 0.0}},
+   {PIPE_FORMAT_A8R8G8B8_UNORM, 0x00ff0000, {0.0, 1.0, 0.0, 0.0}},
+   {PIPE_FORMAT_A8R8G8B8_UNORM, 0xff000000, {0.0, 0.0, 1.0, 0.0}},
+   {PIPE_FORMAT_A8R8G8B8_UNORM, 0xffffffff, {1.0, 1.0, 1.0, 1.0}},
 };
 
 
diff --git a/src/gallium/drivers/llvmpipe/lp_test_main.c b/src/gallium/drivers/llvmpipe/lp_test_main.c
index 14ff00469b..f9dce8b9c2 100644
--- a/src/gallium/drivers/llvmpipe/lp_test_main.c
+++ b/src/gallium/drivers/llvmpipe/lp_test_main.c
@@ -37,7 +37,7 @@
 #include "util/u_cpu_detect.h"
 
 #include "gallivm/lp_bld_const.h"
-#include "gallivm/lp_bld_misc.h"
+#include "gallivm/lp_bld_init.h"
 #include "lp_test.h"
 
 
@@ -380,8 +380,7 @@ int main(int argc, char **argv)
          n = atoi(argv[i]);
    }
 
-   LLVMLinkInJIT();
-   LLVMInitializeNativeTarget();
+   lp_build_init();
 
    util_cpu_detect();
 
diff --git a/src/gallium/drivers/llvmpipe/lp_tex_sample.h b/src/gallium/drivers/llvmpipe/lp_tex_sample.h
index cb59a94464..799df182b6 100644
--- a/src/gallium/drivers/llvmpipe/lp_tex_sample.h
+++ b/src/gallium/drivers/llvmpipe/lp_tex_sample.h
@@ -29,7 +29,7 @@
 #define LP_TEX_SAMPLE_H
 
 
-#include <llvm-c/Core.h>
+#include "os/os_llvm.h"
 
 
 struct lp_sampler_static_state;
diff --git a/src/gallium/drivers/llvmpipe/lp_tex_sample_llvm.c b/src/gallium/drivers/llvmpipe/lp_tex_sample_llvm.c
index 2533275dc1..662508af61 100644
--- a/src/gallium/drivers/llvmpipe/lp_tex_sample_llvm.c
+++ b/src/gallium/drivers/llvmpipe/lp_tex_sample_llvm.c
@@ -51,10 +51,11 @@
 
 
 /**
- * This provides the bridge between the sampler state store in lp_jit_context
- * and lp_jit_texture and the sampler code generator. It provides the
- * texture layout information required by the texture sampler code generator
- * in terms of the state stored in lp_jit_context and lp_jit_texture in runtime.
+ * This provides the bridge between the sampler state store in
+ * lp_jit_context and lp_jit_texture and the sampler code
+ * generator. It provides the texture layout information required by
+ * the texture sampler code generator in terms of the state stored in
+ * lp_jit_context and lp_jit_texture in runtime.
  */
 struct llvmpipe_sampler_dynamic_state
 {
@@ -79,6 +80,9 @@ struct lp_llvm_sampler_soa
 
 /**
  * Fetch the specified member of the lp_jit_texture structure.
+ * \param emit_load  if TRUE, emit the LLVM load instruction to actually
+ *                   fetch the field's value.  Otherwise, just emit the
+ *                   GEP code to address the field.
  *
  * @sa http://llvm.org/docs/GetElementPtr.html
  */
@@ -87,9 +91,11 @@ lp_llvm_texture_member(struct lp_sampler_dynamic_state *base,
                        LLVMBuilderRef builder,
                        unsigned unit,
                        unsigned member_index,
-                       const char *member_name)
+                       const char *member_name,
+                       boolean emit_load)
 {
-   struct llvmpipe_sampler_dynamic_state *state = (struct llvmpipe_sampler_dynamic_state *)base;
+   struct llvmpipe_sampler_dynamic_state *state =
+      (struct llvmpipe_sampler_dynamic_state *)base;
    LLVMValueRef indices[4];
    LLVMValueRef ptr;
    LLVMValueRef res;
@@ -107,7 +113,10 @@ lp_llvm_texture_member(struct lp_sampler_dynamic_state *base,
 
    ptr = LLVMBuildGEP(builder, state->context_ptr, indices, Elements(indices), "");
 
-   res = LLVMBuildLoad(builder, ptr, "");
+   if (emit_load)
+      res = LLVMBuildLoad(builder, ptr, "");
+   else
+      res = ptr;
 
    lp_build_name(res, "context.texture%u.%s", unit, member_name);
 
@@ -116,26 +125,30 @@ lp_llvm_texture_member(struct lp_sampler_dynamic_state *base,
 
 
 /**
- * Helper macro to instantiate the functions that generate the code to fetch
- * the members of lp_jit_texture to fulfill the sampler code generator requests.
+ * Helper macro to instantiate the functions that generate the code to
+ * fetch the members of lp_jit_texture to fulfill the sampler code
+ * generator requests.
  *
- * This complexity is the price we have to pay to keep the texture sampler code
- * generator a reusable module without dependencies to llvmpipe internals.
+ * This complexity is the price we have to pay to keep the texture
+ * sampler code generator a reusable module without dependencies to
+ * llvmpipe internals.
  */
-#define LP_LLVM_TEXTURE_MEMBER(_name, _index) \
+#define LP_LLVM_TEXTURE_MEMBER(_name, _index, _emit_load)  \
    static LLVMValueRef \
    lp_llvm_texture_##_name( struct lp_sampler_dynamic_state *base, \
                             LLVMBuilderRef builder, \
                             unsigned unit) \
    { \
-      return lp_llvm_texture_member(base, builder, unit, _index, #_name ); \
+      return lp_llvm_texture_member(base, builder, unit, _index, #_name, _emit_load ); \
    }
 
 
-LP_LLVM_TEXTURE_MEMBER(width,    LP_JIT_TEXTURE_WIDTH)
-LP_LLVM_TEXTURE_MEMBER(height,   LP_JIT_TEXTURE_HEIGHT)
-LP_LLVM_TEXTURE_MEMBER(stride,   LP_JIT_TEXTURE_STRIDE)
-LP_LLVM_TEXTURE_MEMBER(data_ptr, LP_JIT_TEXTURE_DATA)
+LP_LLVM_TEXTURE_MEMBER(width,      LP_JIT_TEXTURE_WIDTH, TRUE)
+LP_LLVM_TEXTURE_MEMBER(height,     LP_JIT_TEXTURE_HEIGHT, TRUE)
+LP_LLVM_TEXTURE_MEMBER(depth,      LP_JIT_TEXTURE_DEPTH, TRUE)
+LP_LLVM_TEXTURE_MEMBER(last_level, LP_JIT_TEXTURE_LAST_LEVEL, TRUE)
+LP_LLVM_TEXTURE_MEMBER(row_stride, LP_JIT_TEXTURE_ROW_STRIDE, FALSE)
+LP_LLVM_TEXTURE_MEMBER(data_ptr,   LP_JIT_TEXTURE_DATA, FALSE)
 
 
 static void
@@ -145,6 +158,10 @@ lp_llvm_sampler_soa_destroy(struct lp_build_sampler_soa *sampler)
 }
 
 
+/**
+ * Fetch filtered values from texture.
+ * The 'texel' parameter returns four vectors corresponding to R, G, B, A.
+ */
 static void
 lp_llvm_sampler_soa_emit_fetch_texel(struct lp_build_sampler_soa *base,
                                      LLVMBuilderRef builder,
@@ -185,7 +202,9 @@ lp_llvm_sampler_soa_create(const struct lp_sampler_static_state *static_state,
    sampler->base.emit_fetch_texel = lp_llvm_sampler_soa_emit_fetch_texel;
    sampler->dynamic_state.base.width = lp_llvm_texture_width;
    sampler->dynamic_state.base.height = lp_llvm_texture_height;
-   sampler->dynamic_state.base.stride = lp_llvm_texture_stride;
+   sampler->dynamic_state.base.depth = lp_llvm_texture_depth;
+   sampler->dynamic_state.base.last_level = lp_llvm_texture_last_level;
+   sampler->dynamic_state.base.row_stride = lp_llvm_texture_row_stride;
    sampler->dynamic_state.base.data_ptr = lp_llvm_texture_data_ptr;
    sampler->dynamic_state.static_state = static_state;
    sampler->dynamic_state.context_ptr = context_ptr;
diff --git a/src/gallium/drivers/llvmpipe/lp_texture.c b/src/gallium/drivers/llvmpipe/lp_texture.c
index 022bf92cb4..9a85a42897 100644
--- a/src/gallium/drivers/llvmpipe/lp_texture.c
+++ b/src/gallium/drivers/llvmpipe/lp_texture.c
@@ -40,9 +40,10 @@
 
 #include "lp_context.h"
 #include "lp_screen.h"
+#include "lp_flush.h"
 #include "lp_texture.h"
 #include "lp_tile_size.h"
-#include "lp_winsys.h"
+#include "state_tracker/sw_winsys.h"
 
 
 /**
@@ -93,7 +94,7 @@ static boolean
 llvmpipe_displaytarget_layout(struct llvmpipe_screen *screen,
                               struct llvmpipe_texture *lpt)
 {
-   struct llvmpipe_winsys *winsys = screen->winsys;
+   struct sw_winsys *winsys = screen->winsys;
 
    /* Round up the surface size to a multiple of the tile size to
     * avoid tile clipping.
@@ -124,14 +125,9 @@ llvmpipe_texture_create(struct pipe_screen *_screen,
    pipe_reference_init(&lpt->base.reference, 1);
    lpt->base.screen = &screen->base;
 
-   /* XXX: The xlib state tracker is brain-dead and will request
-    * PIPE_FORMAT_Z16_UNORM no matter how much we tell it we don't support it.
-    */
-   if (lpt->base.format == PIPE_FORMAT_Z16_UNORM)
-      lpt->base.format = PIPE_FORMAT_Z32_UNORM;
-
    if (lpt->base.tex_usage & (PIPE_TEXTURE_USAGE_DISPLAY_TARGET |
-                              PIPE_TEXTURE_USAGE_PRIMARY)) {
+                              PIPE_TEXTURE_USAGE_SCANOUT |
+                              PIPE_TEXTURE_USAGE_SHARED)) {
       if (!llvmpipe_displaytarget_layout(screen, lpt))
          goto fail;
    }
@@ -148,43 +144,6 @@ llvmpipe_texture_create(struct pipe_screen *_screen,
 }
 
 
-static struct pipe_texture *
-llvmpipe_texture_blanket(struct pipe_screen * screen,
-                         const struct pipe_texture *base,
-                         const unsigned *stride,
-                         struct pipe_buffer *buffer)
-{
-   /* FIXME */
-#if 0
-   struct llvmpipe_texture *lpt;
-   assert(screen);
-
-   /* Only supports one type */
-   if (base->target != PIPE_TEXTURE_2D ||
-       base->last_level != 0 ||
-       base->depth0 != 1) {
-      return NULL;
-   }
-
-   lpt = CALLOC_STRUCT(llvmpipe_texture);
-   if (!lpt)
-      return NULL;
-
-   lpt->base = *base;
-   pipe_reference_init(&lpt->base.reference, 1);
-   lpt->base.screen = screen;
-   lpt->stride[0] = stride[0];
-
-   pipe_buffer_reference(&lpt->buffer, buffer);
-
-   return &lpt->base;
-#else
-   debug_printf("llvmpipe_texture_blanket() not implemented!");
-   return NULL;
-#endif
-}
-
-
 static void
 llvmpipe_texture_destroy(struct pipe_texture *pt)
 {
@@ -193,7 +152,7 @@ llvmpipe_texture_destroy(struct pipe_texture *pt)
 
    if (lpt->dt) {
       /* display target */
-      struct llvmpipe_winsys *winsys = screen->winsys;
+      struct sw_winsys *winsys = screen->winsys;
       winsys->displaytarget_destroy(winsys, lpt->dt);
    }
    else {
@@ -205,6 +164,92 @@ llvmpipe_texture_destroy(struct pipe_texture *pt)
 }
 
 
+/**
+ * Map a texture. Without any synchronization.
+ */
+void *
+llvmpipe_texture_map(struct pipe_texture *texture,
+                     unsigned face,
+                     unsigned level,
+                     unsigned zslice)
+{
+   struct llvmpipe_texture *lpt = llvmpipe_texture(texture);
+   uint8_t *map;
+
+   if (lpt->dt) {
+      /* display target */
+      struct llvmpipe_screen *screen = llvmpipe_screen(texture->screen);
+      struct sw_winsys *winsys = screen->winsys;
+      const unsigned usage = PIPE_BUFFER_USAGE_CPU_READ_WRITE;
+
+      assert(face == 0);
+      assert(level == 0);
+      assert(zslice == 0);
+
+      /* FIXME: keep map count? */
+      map = winsys->displaytarget_map(winsys, lpt->dt, usage);
+   }
+   else {
+      /* regular texture */
+      unsigned offset;
+      unsigned stride;
+
+      map = lpt->data;
+
+      assert(level < LP_MAX_TEXTURE_2D_LEVELS);
+
+      offset = lpt->level_offset[level];
+      stride = lpt->stride[level];
+
+      /* XXX shouldn't that rather be
+         tex_height = align(u_minify(texture->height0, level), 2)
+         to account for alignment done in llvmpipe_texture_layout ?
+      */
+      if (texture->target == PIPE_TEXTURE_CUBE) {
+         unsigned tex_height = u_minify(texture->height0, level);
+         offset += face *  util_format_get_nblocksy(texture->format, tex_height) * stride;
+      }
+      else if (texture->target == PIPE_TEXTURE_3D) {
+         unsigned tex_height = u_minify(texture->height0, level);
+         offset += zslice * util_format_get_nblocksy(texture->format, tex_height) * stride;
+      }
+      else {
+         assert(face == 0);
+         assert(zslice == 0);
+      }
+
+      map += offset;
+   }
+
+   return map;
+}
+
+
+/**
+ * Unmap a texture. Without any synchronization.
+ */
+void
+llvmpipe_texture_unmap(struct pipe_texture *texture,
+                       unsigned face,
+                       unsigned level,
+                       unsigned zslice)
+{
+   struct llvmpipe_texture *lpt = llvmpipe_texture(texture);
+
+   if (lpt->dt) {
+      /* display target */
+      struct llvmpipe_screen *lp_screen = llvmpipe_screen(texture->screen);
+      struct sw_winsys *winsys = lp_screen->winsys;
+
+      assert(face == 0);
+      assert(level == 0);
+      assert(zslice == 0);
+
+      winsys->displaytarget_unmap(winsys, lpt->dt);
+   }
+}
+
+
 static struct pipe_surface *
 llvmpipe_get_tex_surface(struct pipe_screen *screen,
                          struct pipe_texture *pt,
@@ -223,7 +268,6 @@ llvmpipe_get_tex_surface(struct pipe_screen *screen,
       ps->format = pt->format;
       ps->width = u_minify(pt->width0, level);
       ps->height = u_minify(pt->height0, level);
-      ps->offset = lpt->level_offset[level];
       ps->usage = usage;
 
       /* Because we are llvmpipe, anything that the state tracker
@@ -249,23 +293,6 @@ llvmpipe_get_tex_surface(struct pipe_screen *screen,
       ps->face = face;
       ps->level = level;
       ps->zslice = zslice;
-
-      /* XXX shouldn't that rather be
-         tex_height = align(ps->height, 2);
-         to account for alignment done in llvmpipe_texture_layout ?
-      */
-      if (pt->target == PIPE_TEXTURE_CUBE) {
-         unsigned tex_height = ps->height;
-         ps->offset += face * util_format_get_nblocksy(pt->format, tex_height) * lpt->stride[level];
-      }
-      else if (pt->target == PIPE_TEXTURE_3D) {
-         unsigned tex_height = ps->height;
-         ps->offset += zslice * util_format_get_nblocksy(pt->format, tex_height) * lpt->stride[level];
-      }
-      else {
-         assert(face == 0);
-         assert(zslice == 0);
-      }
    }
    return ps;
 }
@@ -285,7 +312,7 @@ llvmpipe_tex_surface_destroy(struct pipe_surface *surf)
 
 
 static struct pipe_transfer *
-llvmpipe_get_tex_transfer(struct pipe_screen *screen,
+llvmpipe_get_tex_transfer(struct pipe_context *pipe,
                           struct pipe_texture *texture,
                           unsigned face, unsigned level, unsigned zslice,
                           enum pipe_transfer_usage usage,
@@ -311,24 +338,6 @@ llvmpipe_get_tex_transfer(struct pipe_screen *screen,
       pt->level = level;
       pt->zslice = zslice;
 
-      lpt->offset = lptex->level_offset[level];
-
-      /* XXX shouldn't that rather be
-         tex_height = align(u_minify(texture->height0, level), 2)
-         to account for alignment done in llvmpipe_texture_layout ?
-      */
-      if (texture->target == PIPE_TEXTURE_CUBE) {
-         unsigned tex_height = u_minify(texture->height0, level);
-         lpt->offset += face *  util_format_get_nblocksy(texture->format, tex_height) * pt->stride;
-      }
-      else if (texture->target == PIPE_TEXTURE_3D) {
-         unsigned tex_height = u_minify(texture->height0, level);
-         lpt->offset += zslice * util_format_get_nblocksy(texture->format, tex_height) * pt->stride;
-      }
-      else {
-         assert(face == 0);
-         assert(zslice == 0);
-      }
       return pt;
    }
    return NULL;
@@ -336,7 +345,8 @@ llvmpipe_get_tex_transfer(struct pipe_screen *screen,
 
 
 static void 
-llvmpipe_tex_transfer_destroy(struct pipe_transfer *transfer)
+llvmpipe_tex_transfer_destroy(struct pipe_context *pipe,
+                              struct pipe_transfer *transfer)
 {
    /* Effectively do the texture_update work here - if texture images
     * needed post-processing to put them into hardware layout, this is
@@ -349,11 +359,11 @@ llvmpipe_tex_transfer_destroy(struct pipe_transfer *transfer)
 
 
 static void *
-llvmpipe_transfer_map( struct pipe_screen *_screen,
+llvmpipe_transfer_map( struct pipe_context *pipe,
                        struct pipe_transfer *transfer )
 {
-   struct llvmpipe_screen *screen = llvmpipe_screen(_screen);
-   ubyte *map, *xfer_map;
+   struct llvmpipe_screen *screen = llvmpipe_screen(pipe->screen);
+   ubyte *map;
    struct llvmpipe_texture *lpt;
    enum pipe_format format;
 
@@ -361,52 +371,45 @@ llvmpipe_transfer_map( struct pipe_screen *_screen,
    lpt = llvmpipe_texture(transfer->texture);
    format = lpt->base.format;
 
-   if (lpt->dt) {
-      /* display target */
-      struct llvmpipe_winsys *winsys = screen->winsys;
+   /*
+    * Transfers, like other pipe operations, must happen in order, so flush the
+    * context if necessary.
+    */
+   llvmpipe_flush_texture(pipe,
+                          transfer->texture, transfer->face, transfer->level,
+                          0, /* flush_flags */
+                          !(transfer->usage & PIPE_TRANSFER_WRITE), /* read_only */
+                          TRUE, /* cpu_access */
+                          FALSE); /* do_not_flush */
 
-      map = winsys->displaytarget_map(winsys, lpt->dt,
-                                      pipe_transfer_buffer_flags(transfer));
-      if (map == NULL)
-         return NULL;
-   }
-   else {
-      /* regular texture */
-      map = lpt->data;
-   }
+   map = llvmpipe_texture_map(transfer->texture,
+                              transfer->face, transfer->level, transfer->zslice);
 
    /* May want to different things here depending on read/write nature
     * of the map:
     */
-   if (transfer->texture && (transfer->usage & PIPE_TRANSFER_WRITE)) {
+   if (transfer->usage & PIPE_TRANSFER_WRITE) {
       /* Do something to notify sharing contexts of a texture change.
        */
       screen->timestamp++;
    }
    
-   xfer_map = map + llvmpipe_transfer(transfer)->offset +
+   map +=
       transfer->y / util_format_get_blockheight(format) * transfer->stride +
       transfer->x / util_format_get_blockwidth(format) * util_format_get_blocksize(format);
-   /*printf("map = %p  xfer map = %p\n", map, xfer_map);*/
-   return xfer_map;
+
+   return map;
 }
 
 
 static void
-llvmpipe_transfer_unmap(struct pipe_screen *screen,
-                       struct pipe_transfer *transfer)
+llvmpipe_transfer_unmap(struct pipe_context *pipe,
+                        struct pipe_transfer *transfer)
 {
-   struct llvmpipe_screen *lp_screen = llvmpipe_screen(screen);
-   struct llvmpipe_texture *lpt;
-
    assert(transfer->texture);
-   lpt = llvmpipe_texture(transfer->texture);
 
-   if (lpt->dt) {
-      /* display target */
-      struct llvmpipe_winsys *winsys = lp_screen->winsys;
-      winsys->displaytarget_unmap(winsys, lpt->dt);
-   }
+   llvmpipe_texture_unmap(transfer->texture,
+                          transfer->face, transfer->level, transfer->zslice);
 }
 
 
@@ -414,14 +417,18 @@ void
 llvmpipe_init_screen_texture_funcs(struct pipe_screen *screen)
 {
    screen->texture_create = llvmpipe_texture_create;
-   screen->texture_blanket = llvmpipe_texture_blanket;
    screen->texture_destroy = llvmpipe_texture_destroy;
 
    screen->get_tex_surface = llvmpipe_get_tex_surface;
    screen->tex_surface_destroy = llvmpipe_tex_surface_destroy;
+}
 
-   screen->get_tex_transfer = llvmpipe_get_tex_transfer;
-   screen->tex_transfer_destroy = llvmpipe_tex_transfer_destroy;
-   screen->transfer_map = llvmpipe_transfer_map;
-   screen->transfer_unmap = llvmpipe_transfer_unmap;
+
+void
+llvmpipe_init_context_texture_funcs(struct pipe_context *pipe)
+{
+   pipe->get_tex_transfer = llvmpipe_get_tex_transfer;
+   pipe->tex_transfer_destroy = llvmpipe_tex_transfer_destroy;
+   pipe->transfer_map = llvmpipe_transfer_map;
+   pipe->transfer_unmap = llvmpipe_transfer_unmap;
 }
diff --git a/src/gallium/drivers/llvmpipe/lp_texture.h b/src/gallium/drivers/llvmpipe/lp_texture.h
index 87c905bc02..2350c26e4f 100644
--- a/src/gallium/drivers/llvmpipe/lp_texture.h
+++ b/src/gallium/drivers/llvmpipe/lp_texture.h
@@ -32,24 +32,29 @@
 #include "pipe/p_state.h"
 
 
+#define LP_MAX_TEXTURE_2D_LEVELS 13  /* 4K x 4K for now */
+#define LP_MAX_TEXTURE_3D_LEVELS 10  /* 512 x 512 x 512 for now */
+
+
 struct pipe_context;
 struct pipe_screen;
 struct llvmpipe_context;
-struct llvmpipe_displaytarget;
+
+struct sw_displaytarget;
 
 
 struct llvmpipe_texture
 {
    struct pipe_texture base;
 
-   unsigned long level_offset[PIPE_MAX_TEXTURE_LEVELS];
-   unsigned stride[PIPE_MAX_TEXTURE_LEVELS];
+   unsigned long level_offset[LP_MAX_TEXTURE_2D_LEVELS];
+   unsigned stride[LP_MAX_TEXTURE_2D_LEVELS];
 
    /**
     * Display target, for textures with the PIPE_TEXTURE_USAGE_DISPLAY_TARGET
     * usage.
     */
-   struct llvmpipe_displaytarget *dt;
+   struct sw_displaytarget *dt;
 
    /**
     * Malloc'ed data for regular textures, or a mapping to dt above.
@@ -90,8 +95,32 @@ llvmpipe_transfer(struct pipe_transfer *pt)
 }
 
 
+static INLINE unsigned
+llvmpipe_texture_stride(struct pipe_texture *texture,
+                        unsigned level)
+{
+   struct llvmpipe_texture *lpt = llvmpipe_texture(texture);
+   assert(level < LP_MAX_TEXTURE_2D_LEVELS);
+   return lpt->stride[level];
+}
+
+
+void *
+llvmpipe_texture_map(struct pipe_texture *texture,
+                     unsigned face,
+                     unsigned level,
+                     unsigned zslice);
+
+void
+llvmpipe_texture_unmap(struct pipe_texture *texture,
+                       unsigned face,
+                       unsigned level,
+                       unsigned zslice);
+
 extern void
 llvmpipe_init_screen_texture_funcs(struct pipe_screen *screen);
 
+extern void
+llvmpipe_init_context_texture_funcs(struct pipe_context *pipe);
 
 #endif /* LP_TEXTURE_H */
diff --git a/src/gallium/drivers/llvmpipe/lp_tile_image.c b/src/gallium/drivers/llvmpipe/lp_tile_image.c
new file mode 100644
index 0000000000..c1980b316d
--- /dev/null
+++ b/src/gallium/drivers/llvmpipe/lp_tile_image.c
@@ -0,0 +1,126 @@
+/**************************************************************************
+ * 
+ * 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.
+ * 
+ **************************************************************************/
+
+
+#include "lp_tile_soa.h"
+#include "lp_tile_image.h"
+
+
+#define BYTES_PER_TILE (TILE_SIZE * TILE_SIZE * 4)
+
+
+/**
+ * Convert a tiled image into a linear image.
+ * \param src_stride  source row stride in bytes (bytes per row of tiles)
+ * \param dst_stride  dest row stride in bytes
+ */
+void
+lp_tiled_to_linear(const uint8_t *src,
+                   uint8_t *dst,
+                   unsigned width, unsigned height,
+                   enum pipe_format format,
+                   unsigned src_stride,
+                   unsigned dst_stride)
+{
+   const unsigned tiles_per_row = src_stride / BYTES_PER_TILE;
+   unsigned i, j;
+
+   for (j = 0; j < height; j += TILE_SIZE) {
+      for (i = 0; i < width; i += TILE_SIZE) {
+         unsigned tile_offset =
+            ((j / TILE_SIZE) * tiles_per_row + i / TILE_SIZE);
+         unsigned byte_offset = tile_offset * BYTES_PER_TILE;
+         const uint8_t *src_tile = src + byte_offset;
+
+         lp_tile_write_4ub(format,
+                           src_tile,
+                           dst,
+                           dst_stride,
+                           i, j, TILE_SIZE, TILE_SIZE);
+      }
+   }
+}
+
+
+/**
+ * Convert a linear image into a tiled image.
+ * \param src_stride  source row stride in bytes
+ * \param dst_stride  dest row stride in bytes (bytes per row of tiles)
+ */
+void
+lp_linear_to_tiled(const uint8_t *src,
+                   uint8_t *dst,
+                   unsigned width, unsigned height,
+                   enum pipe_format format,
+                   unsigned src_stride,
+                   unsigned dst_stride)
+{
+   const unsigned tiles_per_row = dst_stride / BYTES_PER_TILE;
+   unsigned i, j;
+
+   for (j = 0; j < height; j += TILE_SIZE) {
+      for (i = 0; i < width; i += TILE_SIZE) {
+         unsigned tile_offset =
+            ((j / TILE_SIZE) * tiles_per_row + i / TILE_SIZE);
+         unsigned byte_offset = tile_offset * BYTES_PER_TILE;
+         uint8_t *dst_tile = dst + byte_offset;
+
+         lp_tile_read_4ub(format,
+                          dst_tile,
+                          src,
+                          src_stride,
+                          i, j, TILE_SIZE, TILE_SIZE);
+      }
+   }
+}
+
+
+/**
+ * For testing only.
+ */
+void
+test_tiled_linear_conversion(uint8_t *data,
+                             enum pipe_format format,
+                             unsigned width, unsigned height,
+                             unsigned stride)
+{
+   /* size in tiles */
+   unsigned wt = (width + TILE_SIZE - 1) / TILE_SIZE;
+   unsigned ht = (height + TILE_SIZE - 1) / TILE_SIZE;
+
+   uint8_t *tiled = malloc(wt * ht * TILE_SIZE * TILE_SIZE * 4);
+
+   unsigned tiled_stride = wt * TILE_SIZE * TILE_SIZE * 4;
+
+   lp_linear_to_tiled(data, tiled, width, height, format,
+                      stride, tiled_stride);
+
+   lp_tiled_to_linear(tiled, data, width, height, format,
+                      tiled_stride, stride);
+
+   free(tiled);
+}
+
diff --git a/src/gallium/drivers/llvmpipe/lp_tile_image.h b/src/gallium/drivers/llvmpipe/lp_tile_image.h
new file mode 100644
index 0000000000..60d472e8c5
--- /dev/null
+++ b/src/gallium/drivers/llvmpipe/lp_tile_image.h
@@ -0,0 +1,57 @@
+/**************************************************************************
+ * 
+ * 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 LP_TILE_IMAGE_H
+#define LP_TILE_IMAGE_H
+
+
+void
+lp_tiled_to_linear(const uint8_t *src,
+                   uint8_t *dst,
+                   unsigned width, unsigned height,
+                   enum pipe_format format,
+                   unsigned src_stride,
+                   unsigned dst_stride);
+
+
+void
+lp_linear_to_tiled(const uint8_t *src,
+                   uint8_t *dst,
+                   unsigned width, unsigned height,
+                   enum pipe_format format,
+                   unsigned src_stride,
+                   unsigned dst_stride);
+
+
+void
+test_tiled_linear_conversion(uint8_t *data,
+                             enum pipe_format format,
+                             unsigned width, unsigned height,
+                             unsigned stride);
+
+
+#endif /* LP_TILE_IMAGE_H */
diff --git a/src/gallium/drivers/llvmpipe/lp_tile_soa.py b/src/gallium/drivers/llvmpipe/lp_tile_soa.py
index 5d53689a3d..00b8d4fc38 100644
--- a/src/gallium/drivers/llvmpipe/lp_tile_soa.py
+++ b/src/gallium/drivers/llvmpipe/lp_tile_soa.py
@@ -45,10 +45,10 @@ sys.path.insert(0, os.path.join(os.path.dirname(sys.argv[0]), '../../auxiliary/u
 from u_format_access import *
 
 
-def generate_format_read(format, dst_type, dst_native_type, dst_suffix):
+def generate_format_read(format, dst_channel, dst_native_type, dst_suffix):
     '''Generate the function to read pixels from a particular format'''
 
-    name = short_name(format)
+    name = format.short_name()
 
     src_native_type = native_type(format)
 
@@ -64,11 +64,11 @@ def generate_format_read(format, dst_type, dst_native_type, dst_suffix):
     names = ['']*4
     if format.colorspace == 'rgb':
         for i in range(4):
-            swizzle = format.out_swizzle[i]
+            swizzle = format.swizzles[i]
             if swizzle < 4:
                 names[swizzle] += 'rgba'[i]
     elif format.colorspace == 'zs':
-        swizzle = format.out_swizzle[0]
+        swizzle = format.swizzles[0]
         if swizzle < 4:
             names[swizzle] = 'z'
         else:
@@ -76,48 +76,49 @@ def generate_format_read(format, dst_type, dst_native_type, dst_suffix):
     else:
         assert False
 
-    if format.layout == ARITH:
-        print '         %s pixel = *src_pixel++;' % src_native_type
-        shift = 0;
-        for i in range(4):
-            src_type = format.in_types[i]
-            width = src_type.size
-            if names[i]:
-                value = 'pixel'
-                mask = (1 << width) - 1
-                if shift:
-                    value = '(%s >> %u)' % (value, shift)
-                if shift + width < format.block_size():
-                    value = '(%s & 0x%x)' % (value, mask)
-                value = conversion_expr(src_type, dst_type, dst_native_type, value)
-                print '         %s %s = %s;' % (dst_native_type, names[i], value)
-            shift += width
-    elif format.layout == ARRAY:
-        for i in range(4):
-            src_type = format.in_types[i]
-            if names[i]:
-                value = '(*src_pixel++)'
-                value = conversion_expr(src_type, dst_type, dst_native_type, value)
-                print '         %s %s = %s;' % (dst_native_type, names[i], value)
+    if format.layout == PLAIN:
+        if not format.is_array():
+            print '         %s pixel = *src_pixel++;' % src_native_type
+            shift = 0;
+            for i in range(4):
+                src_channel = format.channels[i]
+                width = src_channel.size
+                if names[i]:
+                    value = 'pixel'
+                    mask = (1 << width) - 1
+                    if shift:
+                        value = '(%s >> %u)' % (value, shift)
+                    if shift + width < format.block_size():
+                        value = '(%s & 0x%x)' % (value, mask)
+                    value = conversion_expr(src_channel, dst_channel, dst_native_type, value, clamp=False)
+                    print '         %s %s = %s;' % (dst_native_type, names[i], value)
+                shift += width
+        else:
+            for i in range(4):
+                src_channel = format.channels[i]
+                if names[i]:
+                    value = '(*src_pixel++)'
+                    value = conversion_expr(src_channel, dst_channel, dst_native_type, value, clamp=False)
+                    print '         %s %s = %s;' % (dst_native_type, names[i], value)
     else:
         assert False
 
     for i in range(4):
         if format.colorspace == 'rgb':
-            swizzle = format.out_swizzle[i]
+            swizzle = format.swizzles[i]
             if swizzle < 4:
                 value = names[swizzle]
             elif swizzle == SWIZZLE_0:
                 value = '0'
             elif swizzle == SWIZZLE_1:
-                value = '1'
+                value = get_one(dst_channel)
             else:
                 assert False
         elif format.colorspace == 'zs':
             if i < 3:
                 value = 'z'
             else:
-                value = '1'
+                value = get_one(dst_channel)
         else:
             assert False
         print '         TILE_PIXEL(dst, x, y, %u) = %s; /* %s */' % (i, value, 'rgba'[i])
@@ -129,31 +130,16 @@ def generate_format_read(format, dst_type, dst_native_type, dst_suffix):
     print
     
 
-def compute_inverse_swizzle(format):
-    '''Return an array[4] of inverse swizzle terms'''
-    inv_swizzle = [None]*4
-    if format.colorspace == 'rgb':
-        for i in range(4):
-            swizzle = format.out_swizzle[i]
-            if swizzle < 4:
-                inv_swizzle[swizzle] = i
-    elif format.colorspace == 'zs':
-        swizzle = format.out_swizzle[0]
-        if swizzle < 4:
-            inv_swizzle[swizzle] = 0
-    return inv_swizzle
-
-
-def pack_rgba(format, src_type, r, g, b, a):
+def pack_rgba(format, src_channel, r, g, b, a):
     """Return an expression for packing r, g, b, a into a pixel of the
     given format.  Ex: '(b << 24) | (g << 16) | (r << 8) | (a << 0)'
     """
     assert format.colorspace == 'rgb'
-    inv_swizzle = compute_inverse_swizzle(format)
+    inv_swizzle = format.inv_swizzles()
     shift = 0
     expr = None
     for i in range(4):
-		# choose r, g, b, or a depending on the inverse swizzle term
+        # choose r, g, b, or a depending on the inverse swizzle term
         if inv_swizzle[i] == 0:
             value = r
         elif inv_swizzle[i] == 1:
@@ -166,25 +152,25 @@ def pack_rgba(format, src_type, r, g, b, a):
             value = None
 
         if value:
-            dst_type = format.in_types[i]
+            dst_channel = format.channels[i]
             dst_native_type = native_type(format)
-            value = conversion_expr(src_type, dst_type, dst_native_type, value)
+            value = conversion_expr(src_channel, dst_channel, dst_native_type, value, clamp=False)
             term = "((%s) << %d)" % (value, shift)
             if expr:
                 expr = expr + " | " + term
             else:
                 expr = term
 
-        width = format.in_types[i].size
+        width = format.channels[i].size
         shift = shift + width
     return expr
 
 
-def emit_unrolled_write_code(format, src_type):
+def emit_unrolled_write_code(format, src_channel):
     '''Emit code for writing a block based on unrolled loops.
     This is considerably faster than the TILE_PIXEL-based code below.
     '''
-    dst_native_type = native_type(format)
+    dst_native_type = 'uint%u_t' % format.block_size()
     print '   const unsigned dstpix_stride = dst_stride / %d;' % format.stride()
     print '   %s *dstpix = (%s *) dst;' % (dst_native_type, dst_native_type)
     print '   unsigned int qx, qy, i;'
@@ -199,8 +185,8 @@ def emit_unrolled_write_code(format, src_type):
     print '         const uint8_t *a = src + 3 * TILE_C_STRIDE;'
     print '         (void) r; (void) g; (void) b; (void) a; /* silence warnings */'
     print '         for (i = 0; i < TILE_C_STRIDE; i += 2) {'
-    print '            const uint32_t pixel0 = %s;' % pack_rgba(format, src_type, "r[i+0]", "g[i+0]", "b[i+0]", "a[i+0]")
-    print '            const uint32_t pixel1 = %s;' % pack_rgba(format, src_type, "r[i+1]", "g[i+1]", "b[i+1]", "a[i+1]")
+    print '            const uint32_t pixel0 = %s;' % pack_rgba(format, src_channel, "r[i+0]", "g[i+0]", "b[i+0]", "a[i+0]")
+    print '            const uint32_t pixel1 = %s;' % pack_rgba(format, src_channel, "r[i+1]", "g[i+1]", "b[i+1]", "a[i+1]")
     print '            const unsigned offset = (py + tile_y_offset[i]) * dstpix_stride + (px + tile_x_offset[i]);'
     print '            dstpix[offset + 0] = pixel0;'
     print '            dstpix[offset + 1] = pixel1;'
@@ -210,11 +196,11 @@ def emit_unrolled_write_code(format, src_type):
     print '   }'
 
 
-def emit_tile_pixel_write_code(format, src_type):
+def emit_tile_pixel_write_code(format, src_channel):
     '''Emit code for writing a block based on the TILE_PIXEL macro.'''
     dst_native_type = native_type(format)
 
-    inv_swizzle = compute_inverse_swizzle(format)
+    inv_swizzle = format.inv_swizzles()
 
     print '   unsigned x, y;'
     print '   uint8_t *dst_row = dst + y0*dst_stride;'
@@ -222,27 +208,28 @@ def emit_tile_pixel_write_code(format, src_type):
     print '      %s *dst_pixel = (%s *)(dst_row + x0*%u);' % (dst_native_type, dst_native_type, format.stride())
     print '      for (x = 0; x < w; ++x) {'
 
-    if format.layout == ARITH:
-        print '         %s pixel = 0;' % dst_native_type
-        shift = 0;
-        for i in range(4):
-            dst_type = format.in_types[i]
-            width = dst_type.size
-            if inv_swizzle[i] is not None:
-                value = 'TILE_PIXEL(src, x, y, %u)' % inv_swizzle[i]
-                value = conversion_expr(src_type, dst_type, dst_native_type, value)
-                if shift:
-                    value = '(%s << %u)' % (value, shift)
-                print '         pixel |= %s;' % value
-            shift += width
-        print '         *dst_pixel++ = pixel;'
-    elif format.layout == ARRAY:
-        for i in range(4):
-            dst_type = format.in_types[i]
-            if inv_swizzle[i] is not None:
-                value = 'TILE_PIXEL(src, x, y, %u)' % inv_swizzle[i]
-                value = conversion_expr(src_type, dst_type, dst_native_type, value)
-                print '         *dst_pixel++ = %s;' % value
+    if format.layout == PLAIN:
+        if not format.is_array():
+            print '         %s pixel = 0;' % dst_native_type
+            shift = 0;
+            for i in range(4):
+                dst_channel = format.channels[i]
+                width = dst_channel.size
+                if inv_swizzle[i] is not None:
+                    value = 'TILE_PIXEL(src, x, y, %u)' % inv_swizzle[i]
+                    value = conversion_expr(src_channel, dst_channel, dst_native_type, value, clamp=False)
+                    if shift:
+                        value = '(%s << %u)' % (value, shift)
+                    print '         pixel |= %s;' % value
+                shift += width
+            print '         *dst_pixel++ = pixel;'
+        else:
+            for i in range(4):
+                dst_channel = format.channels[i]
+                if inv_swizzle[i] is not None:
+                    value = 'TILE_PIXEL(src, x, y, %u)' % inv_swizzle[i]
+                    value = conversion_expr(src_channel, dst_channel, dst_native_type, value, clamp=False)
+                    print '         *dst_pixel++ = %s;' % value
     else:
         assert False
 
@@ -251,28 +238,33 @@ def emit_tile_pixel_write_code(format, src_type):
     print '   }'
 
 
-def generate_format_write(format, src_type, src_native_type, src_suffix):
+def generate_format_write(format, src_channel, src_native_type, src_suffix):
     '''Generate the function to write pixels to a particular format'''
 
-    name = short_name(format)
+    name = format.short_name()
 
     print 'static void'
     print 'lp_tile_%s_write_%s(const %s *src, uint8_t *dst, unsigned dst_stride, unsigned x0, unsigned y0, unsigned w, unsigned h)' % (name, src_suffix, src_native_type)
     print '{'
-    if format.layout == ARITH and format.colorspace == 'rgb':
-        emit_unrolled_write_code(format, src_type)
+    if format.layout == PLAIN \
+        and format.colorspace == 'rgb' \
+        and format.block_size() <= 32 \
+        and format.is_pot() \
+        and not format.is_mixed() \
+        and format.channels[0].type == UNSIGNED:
+        emit_unrolled_write_code(format, src_channel)
     else:
-        emit_tile_pixel_write_code(format, src_type)
+        emit_tile_pixel_write_code(format, src_channel)
     print '}'
     print
     
 
-def generate_read(formats, dst_type, dst_native_type, dst_suffix):
+def generate_read(formats, dst_channel, dst_native_type, dst_suffix):
     '''Generate the dispatch function to read pixels from any format'''
 
     for format in formats:
         if is_format_supported(format):
-            generate_format_read(format, dst_type, dst_native_type, dst_suffix)
+            generate_format_read(format, dst_channel, dst_native_type, dst_suffix)
 
     print 'void'
     print 'lp_tile_read_%s(enum pipe_format format, %s *dst, const void *src, unsigned src_stride, unsigned x, unsigned y, unsigned w, unsigned h)' % (dst_suffix, dst_native_type)
@@ -282,7 +274,7 @@ def generate_read(formats, dst_type, dst_native_type, dst_suffix):
     for format in formats:
         if is_format_supported(format):
             print '   case %s:' % format.name
-            print '      func = &lp_tile_%s_read_%s;' % (short_name(format), dst_suffix)
+            print '      func = &lp_tile_%s_read_%s;' % (format.short_name(), dst_suffix)
             print '      break;'
     print '   default:'
     print '      debug_printf("unsupported format\\n");'
@@ -293,12 +285,12 @@ def generate_read(formats, dst_type, dst_native_type, dst_suffix):
     print
 
 
-def generate_write(formats, src_type, src_native_type, src_suffix):
+def generate_write(formats, src_channel, src_native_type, src_suffix):
     '''Generate the dispatch function to write pixels to any format'''
 
     for format in formats:
         if is_format_supported(format):
-            generate_format_write(format, src_type, src_native_type, src_suffix)
+            generate_format_write(format, src_channel, src_native_type, src_suffix)
 
     print 'void'
     print 'lp_tile_write_%s(enum pipe_format format, const %s *src, void *dst, unsigned dst_stride, unsigned x, unsigned y, unsigned w, unsigned h)' % (src_suffix, src_native_type)
@@ -309,7 +301,7 @@ def generate_write(formats, src_type, src_native_type, src_suffix):
     for format in formats:
         if is_format_supported(format):
             print '   case %s:' % format.name
-            print '      func = &lp_tile_%s_write_%s;' % (short_name(format), src_suffix)
+            print '      func = &lp_tile_%s_write_%s;' % (format.short_name(), src_suffix)
             print '      break;'
     print '   default:'
     print '      debug_printf("unsupported format\\n");'
@@ -359,12 +351,12 @@ def main():
 
     generate_clamp()
 
-    type = Type(UNSIGNED, True, 8)
+    channel = Channel(UNSIGNED, True, 8)
     native_type = 'uint8_t'
     suffix = '4ub'
 
-    generate_read(formats, type, native_type, suffix)
-    generate_write(formats, type, native_type, suffix)
+    generate_read(formats, channel, native_type, suffix)
+    generate_write(formats, channel, native_type, suffix)
 
 
 if __name__ == '__main__':
diff --git a/src/gallium/drivers/llvmpipe/lp_winsys.h b/src/gallium/drivers/llvmpipe/lp_winsys.h
deleted file mode 100644
index ce11fa9304..0000000000
--- a/src/gallium/drivers/llvmpipe/lp_winsys.h
+++ /dev/null
@@ -1,125 +0,0 @@
-/**************************************************************************
- * 
- * Copyright 2007-2009 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 TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
- * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
- * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
- * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
- * 
- **************************************************************************/
-
-/**
- * @file
- * llvmpipe public interface.
- */
-
-
-#ifndef LP_WINSYS_H
-#define LP_WINSYS_H
-
-
-#include "pipe/p_compiler.h" /* for boolean */
-#include "pipe/p_format.h"
-
-
-#ifdef __cplusplus
-extern "C" {
-#endif
-
-
-struct pipe_screen;
-struct pipe_context;
-
-
-/**
- * Opaque pointer.
- */
-struct llvmpipe_displaytarget;
-
-
-/**
- * This is the interface that llvmpipe expects any window system
- * hosting it to implement.
- * 
- * llvmpipe is for the most part a self sufficient driver. The only thing it
- * does not know is how to display a surface.
- */
-struct llvmpipe_winsys
-{
-   void 
-   (*destroy)( struct llvmpipe_winsys *ws );
-
-   boolean
-   (*is_displaytarget_format_supported)( struct llvmpipe_winsys *ws,
-                                         enum pipe_format format );
-   
-   /**
-    * Allocate storage for a render target.
-    * 
-    * Often surfaces which are meant to be blitted to the front screen (i.e.,
-    * display targets) must be allocated with special characteristics, memory 
-    * pools, or obtained directly from the windowing system.
-    *  
-    * This callback is invoked by the pipe_screen when creating a texture marked
-    * with the PIPE_TEXTURE_USAGE_DISPLAY_TARGET flag to get the underlying 
-    * storage.
-    */
-   struct llvmpipe_displaytarget *
-   (*displaytarget_create)( struct llvmpipe_winsys *ws,
-                            enum pipe_format format,
-                            unsigned width, unsigned height,
-                            unsigned alignment,
-                            unsigned *stride );
-
-   void *
-   (*displaytarget_map)( struct llvmpipe_winsys *ws, 
-                         struct llvmpipe_displaytarget *dt,
-                         unsigned flags );
-
-   void
-   (*displaytarget_unmap)( struct llvmpipe_winsys *ws,
-                           struct llvmpipe_displaytarget *dt );
-
-   /**
-    * @sa pipe_screen:flush_frontbuffer.
-    *
-    * This call will likely become asynchronous eventually.
-    */
-   void
-   (*displaytarget_display)( struct llvmpipe_winsys *ws, 
-                             struct llvmpipe_displaytarget *dt,
-                             void *context_private );
-
-   void 
-   (*displaytarget_destroy)( struct llvmpipe_winsys *ws, 
-                             struct llvmpipe_displaytarget *dt );
-};
-
-
-
-struct pipe_screen *
-llvmpipe_create_screen( struct llvmpipe_winsys * );
-
-
-#ifdef __cplusplus
-}
-#endif
-
-#endif /* LP_WINSYS_H */