Merge remote branch 'origin/master' into lp-binning

Conflicts:
	Makefile
	src/gallium/auxiliary/util/u_surface.c
	src/gallium/drivers/llvmpipe/lp_flush.c
	src/gallium/drivers/llvmpipe/lp_setup.c
	src/gallium/drivers/llvmpipe/lp_state_derived.c
	src/gallium/drivers/llvmpipe/lp_state_fs.c
	src/gallium/drivers/llvmpipe/lp_state_surface.c
	src/gallium/drivers/llvmpipe/lp_tex_cache.c
	src/gallium/drivers/llvmpipe/lp_texture.c
	src/gallium/drivers/llvmpipe/lp_tile_cache.c
	src/mesa/state_tracker/st_cb_condrender.c

338 files changed, 6069 insertions, 6069 deletions

diff --git a/src/gallium/SConscript b/src/gallium/SConscript
index eea32b1314..d56c5c8461 100644
--- a/src/gallium/SConscript
+++ b/src/gallium/SConscript
@@ -8,9 +8,10 @@ for driver in env['drivers']:
 	SConscript(os.path.join('drivers', driver, 'SConscript'))
 
 SConscript('state_trackers/python/SConscript')
-SConscript('state_trackers/glx/xlib/SConscript')
-SConscript('state_trackers/dri/SConscript')
-SConscript('state_trackers/xorg/SConscript')
+if platform != 'embedded':
+	SConscript('state_trackers/glx/xlib/SConscript')
+	SConscript('state_trackers/dri/SConscript')
+	SConscript('state_trackers/xorg/SConscript')
 
 if platform == 'windows':
 	SConscript('state_trackers/wgl/SConscript')
diff --git a/src/gallium/auxiliary/Makefile b/src/gallium/auxiliary/Makefile
index 8f937e3b4e..da1fb6b299 100644
--- a/src/gallium/auxiliary/Makefile
+++ b/src/gallium/auxiliary/Makefile
@@ -48,12 +48,10 @@ C_SOURCES = \
 	draw/draw_vs_sse.c \
 	indices/u_indices_gen.c \
 	indices/u_unfilled_gen.c \
-	pipebuffer/pb_buffer_fenced.c \
 	pipebuffer/pb_buffer_malloc.c \
 	pipebuffer/pb_bufmgr_alt.c \
 	pipebuffer/pb_bufmgr_cache.c \
 	pipebuffer/pb_bufmgr_debug.c \
-	pipebuffer/pb_bufmgr_fenced.c \
 	pipebuffer/pb_bufmgr_mm.c \
 	pipebuffer/pb_bufmgr_ondemand.c \
 	pipebuffer/pb_bufmgr_pool.c \
@@ -92,6 +90,7 @@ C_SOURCES = \
 	util/u_debug_dump.c \
 	util/u_debug_symbol.c \
 	util/u_debug_stack.c \
+	util/u_bitmask.c \
 	util/u_blit.c \
 	util/u_blitter.c \
 	util/u_cache.c \
diff --git a/src/gallium/auxiliary/SConscript b/src/gallium/auxiliary/SConscript
index f957090b5f..3aa782f81e 100644
--- a/src/gallium/auxiliary/SConscript
+++ b/src/gallium/auxiliary/SConscript
@@ -87,7 +87,6 @@ source = [
     'pipebuffer/pb_bufmgr_alt.c',
     'pipebuffer/pb_bufmgr_cache.c',
     'pipebuffer/pb_bufmgr_debug.c',
-    'pipebuffer/pb_bufmgr_fenced.c',
     'pipebuffer/pb_bufmgr_mm.c',
     'pipebuffer/pb_bufmgr_ondemand.c',
     'pipebuffer/pb_bufmgr_pool.c',
diff --git a/src/gallium/auxiliary/cso_cache/cso_cache.c b/src/gallium/auxiliary/cso_cache/cso_cache.c
index e6dce3f0e5..a6a07e72c2 100644
--- a/src/gallium/auxiliary/cso_cache/cso_cache.c
+++ b/src/gallium/auxiliary/cso_cache/cso_cache.c
@@ -113,26 +113,6 @@ static struct cso_hash *_cso_hash_for_type(struct cso_cache *sc, enum cso_cache_
    return hash;
 }
 
-static int _cso_size_for_type(enum cso_cache_type type)
-{
-   switch(type) {
-   case CSO_BLEND:
-      return sizeof(struct pipe_blend_state);
-   case CSO_SAMPLER:
-      return sizeof(struct pipe_sampler_state);
-   case CSO_DEPTH_STENCIL_ALPHA:
-      return sizeof(struct pipe_depth_stencil_alpha_state);
-   case CSO_RASTERIZER:
-      return sizeof(struct pipe_rasterizer_state);
-   case CSO_FRAGMENT_SHADER:
-      return sizeof(struct pipe_shader_state);
-   case CSO_VERTEX_SHADER:
-      return sizeof(struct pipe_shader_state);
-   }
-   return 0;
-}
-
-
 static void delete_blend_state(void *state, void *data)
 {
    struct cso_blend *cso = (struct cso_blend *)state;
@@ -282,10 +262,9 @@ void *cso_hash_find_data_from_template( struct cso_hash *hash,
 
 struct cso_hash_iter cso_find_state_template(struct cso_cache *sc,
                                              unsigned hash_key, enum cso_cache_type type,
-                                             void *templ)
+                                             void *templ, unsigned size)
 {
    struct cso_hash_iter iter = cso_find_state(sc, hash_key, type);
-   int size = _cso_size_for_type(type);
    while (!cso_hash_iter_is_null(iter)) {
       void *iter_data = cso_hash_iter_data(iter);
       if (!memcmp(iter_data, templ, size))
diff --git a/src/gallium/auxiliary/cso_cache/cso_cache.h b/src/gallium/auxiliary/cso_cache/cso_cache.h
index 6b5c230e8f..eea60b940b 100644
--- a/src/gallium/auxiliary/cso_cache/cso_cache.h
+++ b/src/gallium/auxiliary/cso_cache/cso_cache.h
@@ -160,7 +160,7 @@ struct cso_hash_iter cso_find_state(struct cso_cache *sc,
                                     unsigned hash_key, enum cso_cache_type type);
 struct cso_hash_iter cso_find_state_template(struct cso_cache *sc,
                                              unsigned hash_key, enum cso_cache_type type,
-                                             void *templ);
+                                             void *templ, unsigned size);
 void cso_for_each_state(struct cso_cache *sc, enum cso_cache_type type,
                         cso_state_callback func, void *user_data);
 void * cso_take_state(struct cso_cache *sc, unsigned hash_key,
diff --git a/src/gallium/auxiliary/cso_cache/cso_context.c b/src/gallium/auxiliary/cso_cache/cso_context.c
index 2b16332e14..dec830ba93 100644
--- a/src/gallium/auxiliary/cso_cache/cso_context.c
+++ b/src/gallium/auxiliary/cso_cache/cso_context.c
@@ -310,18 +310,21 @@ void cso_destroy_context( struct cso_context *ctx )
 enum pipe_error cso_set_blend(struct cso_context *ctx,
                               const struct pipe_blend_state *templ)
 {
-   unsigned hash_key = cso_construct_key((void*)templ, sizeof(struct pipe_blend_state));
-   struct cso_hash_iter iter = cso_find_state_template(ctx->cache,
-                                                       hash_key, CSO_BLEND,
-                                                       (void*)templ);
+   unsigned key_size, hash_key;
+   struct cso_hash_iter iter;
    void *handle;
 
+   key_size = templ->independent_blend_enable ? sizeof(struct pipe_blend_state) :
+              (char *)&(templ->rt[1]) - (char *)templ;
+   hash_key = cso_construct_key((void*)templ, key_size);
+   iter = cso_find_state_template(ctx->cache, hash_key, CSO_BLEND, (void*)templ, key_size);
+
    if (cso_hash_iter_is_null(iter)) {
       struct cso_blend *cso = MALLOC(sizeof(struct cso_blend));
       if (!cso)
          return PIPE_ERROR_OUT_OF_MEMORY;
 
-      memcpy(&cso->state, templ, sizeof(*templ));
+      memcpy(&cso->state, templ, key_size);
       cso->data = ctx->pipe->create_blend_state(ctx->pipe, &cso->state);
       cso->delete_state = (cso_state_callback)ctx->pipe->delete_blend_state;
       cso->context = ctx->pipe;
@@ -369,10 +372,11 @@ enum pipe_error cso_single_sampler(struct cso_context *ctx,
    void *handle = NULL;
 
    if (templ != NULL) {
-      unsigned hash_key = cso_construct_key((void*)templ, sizeof(struct pipe_sampler_state));
+      unsigned key_size = sizeof(struct pipe_sampler_state);
+      unsigned hash_key = cso_construct_key((void*)templ, key_size);
       struct cso_hash_iter iter = cso_find_state_template(ctx->cache,
                                                           hash_key, CSO_SAMPLER,
-                                                          (void*)templ);
+                                                          (void*)templ, key_size);
 
       if (cso_hash_iter_is_null(iter)) {
          struct cso_sampler *cso = MALLOC(sizeof(struct cso_sampler));
@@ -409,10 +413,11 @@ cso_single_vertex_sampler(struct cso_context *ctx,
    void *handle = NULL;
 
    if (templ != NULL) {
-      unsigned hash_key = cso_construct_key((void*)templ, sizeof(struct pipe_sampler_state));
+      unsigned key_size = sizeof(struct pipe_sampler_state);
+      unsigned hash_key = cso_construct_key((void*)templ, key_size);
       struct cso_hash_iter iter = cso_find_state_template(ctx->cache,
                                                           hash_key, CSO_SAMPLER,
-                                                          (void*)templ);
+                                                          (void*)templ, key_size);
 
       if (cso_hash_iter_is_null(iter)) {
          struct cso_sampler *cso = MALLOC(sizeof(struct cso_sampler));
@@ -539,6 +544,38 @@ void cso_restore_samplers(struct cso_context *ctx)
    cso_single_sampler_done( ctx );
 }
 
+/*
+ * If the function encouters any errors it will return the
+ * last one. Done to always try to set as many samplers
+ * as possible.
+ */
+enum pipe_error cso_set_vertex_samplers(struct cso_context *ctx,
+                                        unsigned nr,
+                                        const struct pipe_sampler_state **templates)
+{
+   unsigned i;
+   enum pipe_error temp, error = PIPE_OK;
+
+   /* TODO: fastpath
+    */
+
+   for (i = 0; i < nr; i++) {
+      temp = cso_single_vertex_sampler( ctx, i, templates[i] );
+      if (temp != PIPE_OK)
+         error = temp;
+   }
+
+   for ( ; i < ctx->nr_samplers; i++) {
+      temp = cso_single_vertex_sampler( ctx, i, NULL );
+      if (temp != PIPE_OK)
+         error = temp;
+   }
+
+   cso_single_vertex_sampler_done( ctx );
+
+   return error;
+}
+
 void
 cso_save_vertex_samplers(struct cso_context *ctx)
 {
@@ -666,12 +703,12 @@ cso_restore_vertex_sampler_textures(struct cso_context *ctx)
 enum pipe_error cso_set_depth_stencil_alpha(struct cso_context *ctx,
                                             const struct pipe_depth_stencil_alpha_state *templ)
 {
-   unsigned hash_key = cso_construct_key((void*)templ,
-                                         sizeof(struct pipe_depth_stencil_alpha_state));
+   unsigned key_size = sizeof(struct pipe_depth_stencil_alpha_state);
+   unsigned hash_key = cso_construct_key((void*)templ, key_size);
    struct cso_hash_iter iter = cso_find_state_template(ctx->cache,
                                                        hash_key, 
-						       CSO_DEPTH_STENCIL_ALPHA,
-                                                       (void*)templ);
+                                                       CSO_DEPTH_STENCIL_ALPHA,
+                                                       (void*)templ, key_size);
    void *handle;
 
    if (cso_hash_iter_is_null(iter)) {
@@ -723,11 +760,11 @@ void cso_restore_depth_stencil_alpha(struct cso_context *ctx)
 enum pipe_error cso_set_rasterizer(struct cso_context *ctx,
                                    const struct pipe_rasterizer_state *templ)
 {
-   unsigned hash_key = cso_construct_key((void*)templ,
-                                         sizeof(struct pipe_rasterizer_state));
+   unsigned key_size = sizeof(struct pipe_rasterizer_state);
+   unsigned hash_key = cso_construct_key((void*)templ, key_size);
    struct cso_hash_iter iter = cso_find_state_template(ctx->cache,
                                                        hash_key, CSO_RASTERIZER,
-                                                       (void*)templ);
+                                                       (void*)templ, key_size);
    void *handle = NULL;
 
    if (cso_hash_iter_is_null(iter)) {
@@ -809,7 +846,8 @@ enum pipe_error cso_set_fragment_shader(struct cso_context *ctx,
    struct cso_hash_iter iter = cso_find_state_template(ctx->cache,
                                                        hash_key, 
                                                        CSO_FRAGMENT_SHADER,
-                                                       (void*)tokens);
+                                                       (void*)tokens,
+                                                       sizeof(*templ)); /* XXX correct? tokens_size? */
    void *handle = NULL;
 
    if (cso_hash_iter_is_null(iter)) {
@@ -888,7 +926,8 @@ enum pipe_error cso_set_vertex_shader(struct cso_context *ctx,
                                          sizeof(struct pipe_shader_state));
    struct cso_hash_iter iter = cso_find_state_template(ctx->cache,
                                                        hash_key, CSO_VERTEX_SHADER,
-                                                       (void*)templ);
+                                                       (void*)templ,
+                                                       sizeof(*templ));
    void *handle = NULL;
 
    if (cso_hash_iter_is_null(iter)) {
diff --git a/src/gallium/auxiliary/cso_cache/cso_context.h b/src/gallium/auxiliary/cso_cache/cso_context.h
index b9e313e32d..d2089b1c88 100644
--- a/src/gallium/auxiliary/cso_cache/cso_context.h
+++ b/src/gallium/auxiliary/cso_cache/cso_context.h
@@ -84,6 +84,10 @@ enum pipe_error cso_single_sampler( struct cso_context *cso,
 
 void cso_single_sampler_done( struct cso_context *cso );
 
+enum pipe_error cso_set_vertex_samplers(struct cso_context *cso,
+                                        unsigned count,
+                                        const struct pipe_sampler_state **states);
+
 void
 cso_save_vertex_samplers(struct cso_context *cso);
 
diff --git a/src/gallium/auxiliary/draw/draw_context.c b/src/gallium/auxiliary/draw/draw_context.c
index e90dfc5aec..d3084fd428 100644
--- a/src/gallium/auxiliary/draw/draw_context.c
+++ b/src/gallium/auxiliary/draw/draw_context.c
@@ -34,11 +34,8 @@
 #include "util/u_memory.h"
 #include "util/u_math.h"
 #include "draw_context.h"
-#include "draw_vbuf.h"
 #include "draw_vs.h"
 #include "draw_gs.h"
-#include "draw_pt.h"
-#include "draw_pipe.h"
 
 
 struct draw_context *draw_create( void )
@@ -237,17 +234,20 @@ draw_set_mapped_vertex_buffer(struct draw_context *draw,
 void
 draw_set_mapped_constant_buffer(struct draw_context *draw,
                                 unsigned shader_type,
+                                unsigned slot,
                                 const void *buffer,
                                 unsigned size )
 {
    debug_assert(shader_type == PIPE_SHADER_VERTEX ||
                 shader_type == PIPE_SHADER_GEOMETRY);
+   debug_assert(slot < PIPE_MAX_CONSTANT_BUFFERS);
+
    if (shader_type == PIPE_SHADER_VERTEX) {
-      draw->pt.user.vs_constants = buffer;
-      draw_vs_set_constants( draw, (const float (*)[4])buffer, size );
+      draw->pt.user.vs_constants[slot] = buffer;
+      draw_vs_set_constants(draw, slot, buffer, size);
    } else if (shader_type == PIPE_SHADER_GEOMETRY) {
-      draw->pt.user.gs_constants = buffer;
-      draw_gs_set_constants( draw, (const float (*)[4])buffer, size );
+      draw->pt.user.gs_constants[slot] = buffer;
+      draw_gs_set_constants(draw, slot, buffer, size);
    }
 }
 
diff --git a/src/gallium/auxiliary/draw/draw_context.h b/src/gallium/auxiliary/draw/draw_context.h
index 8a64c06efc..acd81b9712 100644
--- a/src/gallium/auxiliary/draw/draw_context.h
+++ b/src/gallium/auxiliary/draw/draw_context.h
@@ -151,10 +151,12 @@ void draw_set_mapped_element_buffer( struct draw_context *draw,
 void draw_set_mapped_vertex_buffer(struct draw_context *draw,
                                    unsigned attr, const void *buffer);
 
-void draw_set_mapped_constant_buffer(struct draw_context *draw,
-                                     unsigned shader_type,
-                                     const void *buffer,
-                                     unsigned size );
+void
+draw_set_mapped_constant_buffer(struct draw_context *draw,
+                                unsigned shader_type,
+                                unsigned slot,
+                                const void *buffer,
+                                unsigned size);
 
 
 /***********************************************************************
diff --git a/src/gallium/auxiliary/draw/draw_gs.c b/src/gallium/auxiliary/draw/draw_gs.c
index daf8d071f1..7069aa6b18 100644
--- a/src/gallium/auxiliary/draw/draw_gs.c
+++ b/src/gallium/auxiliary/draw/draw_gs.c
@@ -69,9 +69,11 @@ void draw_gs_destroy( struct draw_context *draw )
    tgsi_exec_machine_destroy(draw->gs.machine);
 }
 
-void draw_gs_set_constants( struct draw_context *draw,
-                            const float (*constants)[4],
-                            unsigned size )
+void
+draw_gs_set_constants(struct draw_context *draw,
+                      unsigned slot,
+                      const void *constants,
+                      unsigned size)
 {
 }
 
@@ -291,7 +293,7 @@ draw_geometry_fetch_outputs(struct draw_geometry_shader *shader,
 void draw_geometry_shader_run(struct draw_geometry_shader *shader,
                               const float (*input)[4],
                               float (*output)[4],
-                              const float (*constants)[4],
+                              const void *constants[PIPE_MAX_CONSTANT_BUFFERS],
                               unsigned count,
                               unsigned input_stride,
                               unsigned vertex_size)
@@ -302,7 +304,9 @@ void draw_geometry_shader_run(struct draw_geometry_shader *shader,
    unsigned num_primitives = count/num_vertices;
    unsigned inputs_from_vs = 0;
 
-   machine->Consts = constants;
+   for (i = 0; i < PIPE_MAX_CONSTANT_BUFFERS; i++) {
+      machine->Consts[i] = constants[i];
+   }
 
    for (i = 0; i < shader->info.num_inputs; ++i) {
       if (shader->info.input_semantic_name[i] != TGSI_SEMANTIC_PRIMID)
diff --git a/src/gallium/auxiliary/draw/draw_gs.h b/src/gallium/auxiliary/draw/draw_gs.h
index d6a97d9c4e..d8eb210343 100644
--- a/src/gallium/auxiliary/draw/draw_gs.h
+++ b/src/gallium/auxiliary/draw/draw_gs.h
@@ -62,7 +62,7 @@ struct draw_geometry_shader {
 void draw_geometry_shader_run(struct draw_geometry_shader *shader,
                               const float (*input)[4],
                               float (*output)[4],
-                              const float (*constants)[4],
+                              const void *constants[PIPE_MAX_CONSTANT_BUFFERS],
                               unsigned count,
                               unsigned input_stride,
                               unsigned output_stride);
diff --git a/src/gallium/auxiliary/draw/draw_pipe.c b/src/gallium/auxiliary/draw/draw_pipe.c
index 1c6d657297..11d6485dcf 100644
--- a/src/gallium/auxiliary/draw/draw_pipe.c
+++ b/src/gallium/auxiliary/draw/draw_pipe.c
@@ -106,10 +106,9 @@ void draw_pipeline_destroy( struct draw_context *draw )
 
 
 
-
-
-
-
+/**
+ * Build primitive to render a point with vertex at v0.
+ */
 static void do_point( struct draw_context *draw,
 		      const char *v0 )
 {
@@ -123,6 +122,10 @@ static void do_point( struct draw_context *draw,
 }
 
 
+/**
+ * Build primitive to render a line with vertices at v0, v1.
+ * \param flags  bitmask of DRAW_PIPE_EDGE_x, DRAW_PIPE_RESET_STIPPLE
+ */
 static void do_line( struct draw_context *draw,
                      ushort flags,
 		     const char *v0,
@@ -139,6 +142,10 @@ static void do_line( struct draw_context *draw,
 }
 
 
+/**
+ * Build primitive to render a triangle with vertices at v0, v1, v2.
+ * \param flags  bitmask of DRAW_PIPE_EDGE_x, DRAW_PIPE_RESET_STIPPLE
+ */
 static void do_triangle( struct draw_context *draw,
                          ushort flags,
 			 char *v0,
@@ -157,7 +164,10 @@ static void do_triangle( struct draw_context *draw,
 }
 
 
-
+/*
+ * Set up macros for draw_pt_decompose.h template code.
+ * This code uses vertex indexes / elements.
+ */
 #define QUAD(i0,i1,i2,i3)                       \
    do_triangle( draw,                           \
                 ( DRAW_PIPE_RESET_STIPPLE |     \
@@ -175,16 +185,16 @@ static void do_triangle( struct draw_context *draw,
 
 #define TRIANGLE(flags,i0,i1,i2)                                        \
    do_triangle( draw,                                                   \
-                elts[i0],  /* flags */                          \
+                elts[i0],  /* flags */                                  \
                 verts + stride * (elts[i0] & ~DRAW_PIPE_FLAG_MASK),     \
-                verts + stride * elts[i1],                              \
-                verts + stride * elts[i2])
+                verts + stride * (elts[i1] & ~DRAW_PIPE_FLAG_MASK),     \
+                verts + stride * (elts[i2] & ~DRAW_PIPE_FLAG_MASK) );
 
 #define LINE(flags,i0,i1)                                       \
    do_line( draw,                                               \
-            elts[i0],                                   \
+            elts[i0],                                           \
             verts + stride * (elts[i0] & ~DRAW_PIPE_FLAG_MASK), \
-            verts + stride * elts[i1])
+            verts + stride * (elts[i1] & ~DRAW_PIPE_FLAG_MASK) );
 
 #define POINT(i0)                               \
    do_point( draw,                              \
@@ -213,7 +223,9 @@ static void do_triangle( struct draw_context *draw,
 
 
 
-/* Code to run the pipeline on a fairly arbitary collection of vertices.
+/**
+ * Code to run the pipeline on a fairly arbitary collection of vertices.
+ * For drawing indexed primitives.
  *
  * Vertex headers must be pre-initialized with the
  * UNDEFINED_VERTEX_ID, this code will cause that id to become
@@ -243,6 +255,12 @@ void draw_pipeline_run( struct draw_context *draw,
    draw->pipeline.vertex_count = 0;
 }
 
+
+
+/*
+ * Set up macros for draw_pt_decompose.h template code.
+ * This code is for non-indexed rendering (no elts).
+ */
 #define QUAD(i0,i1,i2,i3)                                        \
    do_triangle( draw,                                            \
                 ( DRAW_PIPE_RESET_STIPPLE |                      \
@@ -293,6 +311,10 @@ void draw_pipeline_run( struct draw_context *draw,
 
 #include "draw_pt_decompose.h"
 
+
+/*
+ * For drawing non-indexed primitives.
+ */
 void draw_pipeline_run_linear( struct draw_context *draw,
                                unsigned prim,
                                struct vertex_header *vertices,
diff --git a/src/gallium/auxiliary/draw/draw_pipe_cull.c b/src/gallium/auxiliary/draw/draw_pipe_cull.c
index 11b39db599..dc66c65a56 100644
--- a/src/gallium/auxiliary/draw/draw_pipe_cull.c
+++ b/src/gallium/auxiliary/draw/draw_pipe_cull.c
@@ -50,8 +50,6 @@ static INLINE struct cull_stage *cull_stage( struct draw_stage *stage )
 }
 
 
-
-
 static void cull_tri( struct draw_stage *stage,
 		      struct prim_header *header )
 {
@@ -62,7 +60,7 @@ static void cull_tri( struct draw_stage *stage,
    const float *v1 = header->v[1]->data[pos];
    const float *v2 = header->v[2]->data[pos];
 
-   /* edge vectors e = v0 - v2, f = v1 - v2 */
+   /* edge vectors: e = v0 - v2, f = v1 - v2 */
    const float ex = v0[0] - v2[0];
    const float ey = v0[1] - v2[1];
    const float fx = v1[0] - v2[0];
@@ -72,7 +70,7 @@ static void cull_tri( struct draw_stage *stage,
    header->det = ex * fy - ey * fx;
 
    if (header->det != 0) {
-      /* if (det < 0 then Z points toward camera and triangle is 
+      /* if det < 0 then Z points toward the camera and the triangle is 
        * counter-clockwise winding.
        */
       unsigned winding = (header->det < 0) ? PIPE_WINDING_CCW : PIPE_WINDING_CW;
@@ -84,6 +82,7 @@ static void cull_tri( struct draw_stage *stage,
    }
 }
 
+
 static void cull_first_tri( struct draw_stage *stage, 
 			    struct prim_header *header )
 {
@@ -96,13 +95,13 @@ static void cull_first_tri( struct draw_stage *stage,
 }
 
 
-
 static void cull_flush( struct draw_stage *stage, unsigned flags )
 {
    stage->tri = cull_first_tri;
    stage->next->flush( stage->next, flags );
 }
 
+
 static void cull_reset_stipple_counter( struct draw_stage *stage )
 {
    stage->next->reset_stipple_counter( stage->next );
@@ -140,7 +139,7 @@ struct draw_stage *draw_cull_stage( struct draw_context *draw )
 
    return &cull->stage;
 
- fail:
+fail:
    if (cull)
       cull->stage.destroy( &cull->stage );
 
diff --git a/src/gallium/auxiliary/draw/draw_pipe_validate.c b/src/gallium/auxiliary/draw/draw_pipe_validate.c
index bea90e50d3..a69e2633be 100644
--- a/src/gallium/auxiliary/draw/draw_pipe_validate.c
+++ b/src/gallium/auxiliary/draw/draw_pipe_validate.c
@@ -151,8 +151,8 @@ static struct draw_stage *validate_pipeline( struct draw_stage *stage )
 {
    struct draw_context *draw = stage->draw;
    struct draw_stage *next = draw->pipeline.rasterize;
-   int need_det = 0;
-   int precalc_flat = 0;
+   boolean need_det = FALSE;
+   boolean precalc_flat = FALSE;
    boolean wide_lines, wide_points;
 
    /* Set the validate's next stage to the rasterize stage, so that it
@@ -194,7 +194,7 @@ static struct draw_stage *validate_pipeline( struct draw_stage *stage )
    if (wide_lines) {
       draw->pipeline.wide_line->next = next;
       next = draw->pipeline.wide_line;
-      precalc_flat = 1;
+      precalc_flat = TRUE;
    }
 
    if (wide_points || draw->rasterizer->point_sprite) {
@@ -205,7 +205,7 @@ static struct draw_stage *validate_pipeline( struct draw_stage *stage )
    if (draw->rasterizer->line_stipple_enable && draw->pipeline.line_stipple) {
       draw->pipeline.stipple->next = next;
       next = draw->pipeline.stipple;
-      precalc_flat = 1;		/* only needed for lines really */
+      precalc_flat = TRUE;		/* only needed for lines really */
    }
 
    if (draw->rasterizer->poly_stipple_enable
@@ -218,8 +218,8 @@ static struct draw_stage *validate_pipeline( struct draw_stage *stage )
        draw->rasterizer->fill_ccw != PIPE_POLYGON_MODE_FILL) {
       draw->pipeline.unfilled->next = next;
       next = draw->pipeline.unfilled;
-      precalc_flat = 1;		/* only needed for triangles really */
-      need_det = 1;
+      precalc_flat = TRUE;		/* only needed for triangles really */
+      need_det = TRUE;
    }
 
    if (draw->rasterizer->flatshade && precalc_flat) {
@@ -231,13 +231,13 @@ static struct draw_stage *validate_pipeline( struct draw_stage *stage )
        draw->rasterizer->offset_ccw) {
       draw->pipeline.offset->next = next;
       next = draw->pipeline.offset;
-      need_det = 1;
+      need_det = TRUE;
    }
 
    if (draw->rasterizer->light_twoside) {
       draw->pipeline.twoside->next = next;
       next = draw->pipeline.twoside;
-      need_det = 1;
+      need_det = TRUE;
    }
 
    /* Always run the cull stage as we calculate determinant there
diff --git a/src/gallium/auxiliary/draw/draw_private.h b/src/gallium/auxiliary/draw/draw_private.h
index ef49e57536..6a7190e975 100644
--- a/src/gallium/auxiliary/draw/draw_private.h
+++ b/src/gallium/auxiliary/draw/draw_private.h
@@ -153,8 +153,8 @@ struct draw_context
          const void *vbuffer[PIPE_MAX_ATTRIBS];
          
          /** constant buffer (for vertex/geometry shader) */
-         const void *vs_constants;
-         const void *gs_constants;
+         const void *vs_constants[PIPE_MAX_CONSTANT_BUFFERS];
+         const void *gs_constants[PIPE_MAX_CONSTANT_BUFFERS];
       } user;
 
       boolean test_fse;         /* enable FSE even though its not correct (eg for softpipe) */
@@ -202,10 +202,10 @@ struct draw_context
       struct aos_machine *aos_machine; 
 
 
-      const float (*aligned_constants)[4];
+      const void *aligned_constants[PIPE_MAX_CONSTANT_BUFFERS];
 
-      const float (*aligned_constant_storage)[4];
-      unsigned const_storage_size;
+      const void *aligned_constant_storage[PIPE_MAX_CONSTANT_BUFFERS];
+      unsigned const_storage_size[PIPE_MAX_CONSTANT_BUFFERS];
 
 
       struct translate *fetch;
@@ -256,9 +256,11 @@ void draw_vs_destroy( struct draw_context *draw );
 void draw_vs_set_viewport( struct draw_context *, 
                            const struct pipe_viewport_state * );
 
-void draw_vs_set_constants( struct draw_context *,
-                            const float (*constants)[4],
-                            unsigned size );
+void
+draw_vs_set_constants(struct draw_context *,
+                      unsigned slot,
+                      const void *constants,
+                      unsigned size);
 
 
 
@@ -266,9 +268,13 @@ void draw_vs_set_constants( struct draw_context *,
  * Geometry shading code:
  */
 boolean draw_gs_init( struct draw_context *draw );
-void draw_gs_set_constants( struct draw_context *,
-                            const float (*constants)[4],
-                            unsigned size );
+
+void
+draw_gs_set_constants(struct draw_context *,
+                      unsigned slot,
+                      const void *constants,
+                      unsigned size);
+
 void draw_gs_destroy( struct draw_context *draw );
 
 /*******************************************************************************
diff --git a/src/gallium/auxiliary/draw/draw_pt.c b/src/gallium/auxiliary/draw/draw_pt.c
index a5ddec5286..f5ed32d0b0 100644
--- a/src/gallium/auxiliary/draw/draw_pt.c
+++ b/src/gallium/auxiliary/draw/draw_pt.c
@@ -33,7 +33,6 @@
 #include "draw/draw_context.h"
 #include "draw/draw_private.h"
 #include "draw/draw_pt.h"
-#include "draw/draw_vs.h"
 #include "tgsi/tgsi_dump.h"
 #include "util/u_math.h"
 #include "util/u_prim.h"
diff --git a/src/gallium/auxiliary/draw/draw_pt_fetch.c b/src/gallium/auxiliary/draw/draw_pt_fetch.c
index 55e7a7b81a..252be5053e 100644
--- a/src/gallium/auxiliary/draw/draw_pt_fetch.c
+++ b/src/gallium/auxiliary/draw/draw_pt_fetch.c
@@ -30,7 +30,6 @@
 #include "draw/draw_context.h"
 #include "draw/draw_private.h"
 #include "draw/draw_vbuf.h"
-#include "draw/draw_vertex.h"
 #include "draw/draw_pt.h"
 #include "translate/translate.h"
 #include "translate/translate_cache.h"
diff --git a/src/gallium/auxiliary/draw/draw_pt_fetch_shade_emit.c b/src/gallium/auxiliary/draw/draw_pt_fetch_shade_emit.c
index 734c05f068..c5dfbcfa3c 100644
--- a/src/gallium/auxiliary/draw/draw_pt_fetch_shade_emit.c
+++ b/src/gallium/auxiliary/draw/draw_pt_fetch_shade_emit.c
@@ -40,7 +40,6 @@
 #include "draw/draw_pt.h"
 #include "draw/draw_vs.h"
 
-#include "translate/translate.h"
 
 struct fetch_shade_emit;
 
diff --git a/src/gallium/auxiliary/draw/draw_pt_fetch_shade_pipeline.c b/src/gallium/auxiliary/draw/draw_pt_fetch_shade_pipeline.c
index 23da556f79..56b69354b2 100644
--- a/src/gallium/auxiliary/draw/draw_pt_fetch_shade_pipeline.c
+++ b/src/gallium/auxiliary/draw/draw_pt_fetch_shade_pipeline.c
@@ -33,7 +33,6 @@
 #include "draw/draw_pt.h"
 #include "draw/draw_vs.h"
 #include "draw/draw_gs.h"
-#include "translate/translate.h"
 
 
 struct fetch_pipeline_middle_end {
@@ -164,7 +163,7 @@ static void fetch_pipeline_run( struct draw_pt_middle_end *middle,
       vshader->run_linear(vshader,
                           (const float (*)[4])pipeline_verts->data,
                           (      float (*)[4])pipeline_verts->data,
-                          (const float (*)[4])draw->pt.user.vs_constants,
+                          draw->pt.user.vs_constants,
                           fetch_count,
                           fpme->vertex_size,
                           fpme->vertex_size);
@@ -172,7 +171,7 @@ static void fetch_pipeline_run( struct draw_pt_middle_end *middle,
          draw_geometry_shader_run(gshader,
                                   (const float (*)[4])pipeline_verts->data,
                                   (      float (*)[4])pipeline_verts->data,
-                                  (const float (*)[4])draw->pt.user.gs_constants,
+                                  draw->pt.user.gs_constants,
                                   fetch_count,
                                   fpme->vertex_size,
                                   fpme->vertex_size);
@@ -249,7 +248,7 @@ static void fetch_pipeline_linear_run( struct draw_pt_middle_end *middle,
       shader->run_linear(shader,
 			 (const float (*)[4])pipeline_verts->data,
 			 (      float (*)[4])pipeline_verts->data,
-			 (const float (*)[4])draw->pt.user.vs_constants,
+                         draw->pt.user.vs_constants,
 			 count,
 			 fpme->vertex_size,
 			 fpme->vertex_size);
@@ -258,7 +257,7 @@ static void fetch_pipeline_linear_run( struct draw_pt_middle_end *middle,
          draw_geometry_shader_run(geometry_shader,
                                   (const float (*)[4])pipeline_verts->data,
                                   (      float (*)[4])pipeline_verts->data,
-                                  (const float (*)[4])draw->pt.user.gs_constants,
+                                  draw->pt.user.gs_constants,
                                   count,
                                   fpme->vertex_size,
                                   fpme->vertex_size);
@@ -329,7 +328,7 @@ static boolean fetch_pipeline_linear_run_elts( struct draw_pt_middle_end *middle
       shader->run_linear(shader,
 			 (const float (*)[4])pipeline_verts->data,
 			 (      float (*)[4])pipeline_verts->data,
-			 (const float (*)[4])draw->pt.user.vs_constants,
+                         draw->pt.user.vs_constants,
 			 count,
 			 fpme->vertex_size,
 			 fpme->vertex_size);
@@ -338,7 +337,7 @@ static boolean fetch_pipeline_linear_run_elts( struct draw_pt_middle_end *middle
          draw_geometry_shader_run(geometry_shader,
                                   (const float (*)[4])pipeline_verts->data,
                                   (      float (*)[4])pipeline_verts->data,
-                                  (const float (*)[4])draw->pt.user.gs_constants,
+                                  draw->pt.user.gs_constants,
                                   count,
                                   fpme->vertex_size,
                                   fpme->vertex_size);
diff --git a/src/gallium/auxiliary/draw/draw_pt_post_vs.c b/src/gallium/auxiliary/draw/draw_pt_post_vs.c
index 55151823a1..9728d5c2bd 100644
--- a/src/gallium/auxiliary/draw/draw_pt_post_vs.c
+++ b/src/gallium/auxiliary/draw/draw_pt_post_vs.c
@@ -30,7 +30,6 @@
 #include "draw/draw_context.h"
 #include "draw/draw_private.h"
 #include "draw/draw_vbuf.h"
-#include "draw/draw_vertex.h"
 #include "draw/draw_pt.h"
 
 struct pt_post_vs {
diff --git a/src/gallium/auxiliary/draw/draw_vs.c b/src/gallium/auxiliary/draw/draw_vs.c
index e03ac8c229..6bdd612e6f 100644
--- a/src/gallium/auxiliary/draw/draw_vs.c
+++ b/src/gallium/auxiliary/draw/draw_vs.c
@@ -48,24 +48,27 @@
 
 
 
-void draw_vs_set_constants( struct draw_context *draw,
-                            const float (*constants)[4],
-                            unsigned size )
+void
+draw_vs_set_constants(struct draw_context *draw,
+                      unsigned slot,
+                      const void *constants,
+                      unsigned size)
 {
    if (((uintptr_t)constants) & 0xf) {
-      if (size > draw->vs.const_storage_size) {
-         if (draw->vs.aligned_constant_storage)
-            align_free((void *)draw->vs.aligned_constant_storage);
-         draw->vs.aligned_constant_storage = align_malloc( size, 16 );
+      if (size > draw->vs.const_storage_size[slot]) {
+         if (draw->vs.aligned_constant_storage[slot]) {
+            align_free((void *)draw->vs.aligned_constant_storage[slot]);
+         }
+         draw->vs.aligned_constant_storage[slot] = align_malloc(size, 16);
       }
-      memcpy( (void*)draw->vs.aligned_constant_storage,
-              constants, 
-              size );
-      constants = draw->vs.aligned_constant_storage;
+      memcpy((void *)draw->vs.aligned_constant_storage[slot],
+             constants,
+             size);
+      constants = draw->vs.aligned_constant_storage[slot];
    }
-      
-   draw->vs.aligned_constants = constants;
-   draw_vs_aos_machine_constants( draw->vs.aos_machine, constants );
+
+   draw->vs.aligned_constants[slot] = constants;
+   draw_vs_aos_machine_constants(draw->vs.aos_machine, slot, constants);
 }
 
 
@@ -182,6 +185,8 @@ draw_vs_init( struct draw_context *draw )
 void
 draw_vs_destroy( struct draw_context *draw )
 {
+   uint i;
+
    if (draw->vs.fetch_cache)
       translate_cache_destroy(draw->vs.fetch_cache);
 
@@ -191,8 +196,11 @@ draw_vs_destroy( struct draw_context *draw )
    if (draw->vs.aos_machine)
       draw_vs_aos_machine_destroy(draw->vs.aos_machine);
 
-   if (draw->vs.aligned_constant_storage)
-      align_free((void*)draw->vs.aligned_constant_storage);
+   for (i = 0; i < PIPE_MAX_CONSTANT_BUFFERS; i++) {
+      if (draw->vs.aligned_constant_storage[i]) {
+         align_free((void *)draw->vs.aligned_constant_storage[i]);
+      }
+   }
 
    tgsi_exec_machine_destroy(draw->vs.machine);
 }
diff --git a/src/gallium/auxiliary/draw/draw_vs.h b/src/gallium/auxiliary/draw/draw_vs.h
index 00036cfe68..d095c9bad1 100644
--- a/src/gallium/auxiliary/draw/draw_vs.h
+++ b/src/gallium/auxiliary/draw/draw_vs.h
@@ -132,7 +132,7 @@ struct draw_vertex_shader {
    void (*run_linear)( struct draw_vertex_shader *shader,
 		       const float (*input)[4],
 		       float (*output)[4],
-		       const float (*constants)[4],
+                      const void *constants[PIPE_MAX_CONSTANT_BUFFERS],
 		       unsigned count,
 		       unsigned input_stride,
 		       unsigned output_stride );
@@ -212,8 +212,10 @@ static INLINE int draw_vs_varient_key_compare( const struct draw_vs_varient_key
 struct aos_machine *draw_vs_aos_machine( void );
 void draw_vs_aos_machine_destroy( struct aos_machine *machine );
 
-void draw_vs_aos_machine_constants( struct aos_machine *machine,
-                                    const float (*constants)[4] );
+void
+draw_vs_aos_machine_constants(struct aos_machine *machine,
+                              unsigned slot,
+                              const void *constants);
 
 void draw_vs_aos_machine_viewport( struct aos_machine *machine,
                                    const struct pipe_viewport_state *viewport );
diff --git a/src/gallium/auxiliary/draw/draw_vs_aos.c b/src/gallium/auxiliary/draw/draw_vs_aos.c
index 1aaae4ab7a..e7121f3654 100644
--- a/src/gallium/auxiliary/draw/draw_vs_aos.c
+++ b/src/gallium/auxiliary/draw/draw_vs_aos.c
@@ -2114,11 +2114,14 @@ static void PIPE_CDECL vaos_run_elts( struct draw_vs_varient *varient,
 {
    struct draw_vs_varient_aos_sse *vaos = (struct draw_vs_varient_aos_sse *)varient;
    struct aos_machine *machine = vaos->draw->vs.aos_machine;
+   unsigned i;
 
    if (0) debug_printf("%s %d\n", __FUNCTION__, count);
 
    machine->internal[IMM_PSIZE][0] = vaos->draw->rasterizer->point_size;
-   machine->constants = vaos->draw->vs.aligned_constants;
+   for (i = 0; i < PIPE_MAX_CONSTANT_BUFFERS; i++) {
+      machine->constants[i] = vaos->draw->vs.aligned_constants[i];
+   }
    machine->immediates = vaos->base.vs->immediates;
    machine->buffer = vaos->buffer;
 
@@ -2135,12 +2138,15 @@ static void PIPE_CDECL vaos_run_linear( struct draw_vs_varient *varient,
 {
    struct draw_vs_varient_aos_sse *vaos = (struct draw_vs_varient_aos_sse *)varient;
    struct aos_machine *machine = vaos->draw->vs.aos_machine;
+   unsigned i;
 
    if (0) debug_printf("%s %d %d const: %x\n", __FUNCTION__, start, count, 
                        vaos->base.key.const_vbuffers);
 
    machine->internal[IMM_PSIZE][0] = vaos->draw->rasterizer->point_size;
-   machine->constants = vaos->draw->vs.aligned_constants;
+   for (i = 0; i < PIPE_MAX_CONSTANT_BUFFERS; i++) {
+      machine->constants[i] = vaos->draw->vs.aligned_constants[i];
+   }
    machine->immediates = vaos->base.vs->immediates;
    machine->buffer = vaos->buffer;
 
diff --git a/src/gallium/auxiliary/draw/draw_vs_aos.h b/src/gallium/auxiliary/draw/draw_vs_aos.h
index 2cf72ddf7b..1911242f82 100644
--- a/src/gallium/auxiliary/draw/draw_vs_aos.h
+++ b/src/gallium/auxiliary/draw/draw_vs_aos.h
@@ -122,7 +122,7 @@ struct aos_machine {
    ushort fpucntl;              /* one of FPU_* above */
 
    const float (*immediates)[4];     /* points to shader data */
-   const float (*constants)[4];      /* points to draw data */
+   const void *constants[PIPE_MAX_CONSTANT_BUFFERS]; /* points to draw data */
 
    const struct aos_buffer *buffer; /* points to ? */
 };
diff --git a/src/gallium/auxiliary/draw/draw_vs_aos_machine.c b/src/gallium/auxiliary/draw/draw_vs_aos_machine.c
index 3240e3745d..0eda414ee6 100644
--- a/src/gallium/auxiliary/draw/draw_vs_aos_machine.c
+++ b/src/gallium/auxiliary/draw/draw_vs_aos_machine.c
@@ -219,10 +219,12 @@ static void PIPE_CDECL populate_lut( struct aos_machine *machine,
 }
 
 
-void draw_vs_aos_machine_constants( struct aos_machine *machine,
-                                    const float (*constants)[4] )
+void
+draw_vs_aos_machine_constants(struct aos_machine *machine,
+                              unsigned slot,
+                              const void *constants)
 {
-   machine->constants = constants;
+   machine->constants[slot] = constants;
 
    {
       unsigned i;
@@ -307,8 +309,10 @@ void draw_vs_aos_machine_viewport( struct aos_machine *machine,
 {
 }
 
-void draw_vs_aos_machine_constants( struct aos_machine *machine,
-                                    const float (*constants)[4] )
+void
+draw_vs_aos_machine_constants(struct aos_machine *machine,
+                              unsigned slot,
+                              const void *constants)
 {
 }
 
diff --git a/src/gallium/auxiliary/draw/draw_vs_exec.c b/src/gallium/auxiliary/draw/draw_vs_exec.c
index 41cc802613..7deca2b69d 100644
--- a/src/gallium/auxiliary/draw/draw_vs_exec.c
+++ b/src/gallium/auxiliary/draw/draw_vs_exec.c
@@ -85,7 +85,7 @@ static void
 vs_exec_run_linear( struct draw_vertex_shader *shader,
 		    const float (*input)[4],
 		    float (*output)[4],
-		    const float (*constants)[4],
+                   const void *constants[PIPE_MAX_CONSTANT_BUFFERS],
 		    unsigned count,
 		    unsigned input_stride,
 		    unsigned output_stride )
@@ -95,7 +95,9 @@ vs_exec_run_linear( struct draw_vertex_shader *shader,
    unsigned int i, j;
    unsigned slot;
 
-   machine->Consts = constants;
+   for (i = 0; i < PIPE_MAX_CONSTANT_BUFFERS; i++) {
+      machine->Consts[i] = constants[i];
+   }
 
    for (i = 0; i < count; i += MAX_TGSI_VERTICES) {
       unsigned int max_vertices = MIN2(MAX_TGSI_VERTICES, count - i);
diff --git a/src/gallium/auxiliary/draw/draw_vs_llvm.c b/src/gallium/auxiliary/draw/draw_vs_llvm.c
index b3535c0e48..fd9166fda5 100644
--- a/src/gallium/auxiliary/draw/draw_vs_llvm.c
+++ b/src/gallium/auxiliary/draw/draw_vs_llvm.c
@@ -64,7 +64,7 @@ static void
 vs_llvm_run_linear( struct draw_vertex_shader *base,
 		   const float (*input)[4],
 		   float (*output)[4],
-		   const float (*constants)[4],
+                   const void *constants[PIPE_MAX_CONSTANT_BUFFERS],
 		   unsigned count,
 		   unsigned input_stride,
 		   unsigned output_stride )
@@ -74,7 +74,8 @@ vs_llvm_run_linear( struct draw_vertex_shader *base,
 
    gallivm_cpu_vs_exec(shader->llvm_prog, shader->machine,
                        input, base->info.num_inputs, output, base->info.num_outputs,
-                       constants, count, input_stride, output_stride);
+                       (const float (*)[4])constants[0],
+                       count, input_stride, output_stride);
 }
 
 
diff --git a/src/gallium/auxiliary/draw/draw_vs_ppc.c b/src/gallium/auxiliary/draw/draw_vs_ppc.c
index da9f3e3d35..d869eecec5 100644
--- a/src/gallium/auxiliary/draw/draw_vs_ppc.c
+++ b/src/gallium/auxiliary/draw/draw_vs_ppc.c
@@ -85,7 +85,7 @@ static void
 vs_ppc_run_linear( struct draw_vertex_shader *base,
 		   const float (*input)[4],
 		   float (*output)[4],
-		   const float (*constants)[4],
+                  const void *constants[PIPE_MAX_CONSTANT_BUFFERS],
 		   unsigned count,
 		   unsigned input_stride,
 		   unsigned output_stride )
@@ -125,7 +125,7 @@ vs_ppc_run_linear( struct draw_vertex_shader *base,
        */
       shader->func(inputs_soa, outputs_soa, temps_soa,
 		   (float (*)[4]) shader->base.immediates,
-		   (float (*)[4]) constants,
+                   (const float (*)[4])constants[0],
                    ppc_builtin_constants);
 
       /* convert (up to) four output verts from SoA back to AoS format */
diff --git a/src/gallium/auxiliary/draw/draw_vs_sse.c b/src/gallium/auxiliary/draw/draw_vs_sse.c
index 702051387a..54e6423388 100644
--- a/src/gallium/auxiliary/draw/draw_vs_sse.c
+++ b/src/gallium/auxiliary/draw/draw_vs_sse.c
@@ -83,7 +83,7 @@ static void
 vs_sse_run_linear( struct draw_vertex_shader *base,
 		   const float (*input)[4],
 		   float (*output)[4],
-		   const float (*constants)[4],
+                  const void *constants[PIPE_MAX_CONSTANT_BUFFERS],
 		   unsigned count,
 		   unsigned input_stride,
 		   unsigned output_stride )
@@ -112,7 +112,7 @@ vs_sse_run_linear( struct draw_vertex_shader *base,
       /* run compiled shader
        */
       shader->func(machine,
-		   constants,
+                   (const float (*)[4])constants[0],
 		   shader->base.immediates,
                    input,
                    base->info.num_inputs,
diff --git a/src/gallium/auxiliary/draw/draw_vs_varient.c b/src/gallium/auxiliary/draw/draw_vs_varient.c
index 9f40030f39..5ed706cb4f 100644
--- a/src/gallium/auxiliary/draw/draw_vs_varient.c
+++ b/src/gallium/auxiliary/draw/draw_vs_varient.c
@@ -38,7 +38,6 @@
 #include "draw/draw_vertex.h"
 #include "draw/draw_vs.h"
 #include "translate/translate.h"
-#include "translate/translate_cache.h"
 
 /* A first pass at incorporating vertex fetch/emit functionality into 
  */
@@ -148,7 +147,7 @@ static void PIPE_CDECL vsvg_run_elts( struct draw_vs_varient *varient,
    vsvg->base.vs->run_linear( vsvg->base.vs, 
                               temp_buffer,
                               temp_buffer,
-                              (const float (*)[4])vsvg->base.vs->draw->pt.user.vs_constants,
+                             vsvg->base.vs->draw->pt.user.vs_constants,
                               count,
                               temp_vertex_stride, 
                               temp_vertex_stride);
@@ -211,7 +210,7 @@ static void PIPE_CDECL vsvg_run_linear( struct draw_vs_varient *varient,
    vsvg->base.vs->run_linear( vsvg->base.vs, 
                               temp_buffer,
                               temp_buffer,
-                              (const float (*)[4])vsvg->base.vs->draw->pt.user.vs_constants,
+                             vsvg->base.vs->draw->pt.user.vs_constants,
                               count,
                               temp_vertex_stride, 
                               temp_vertex_stride);
diff --git a/src/gallium/auxiliary/pipebuffer/Makefile b/src/gallium/auxiliary/pipebuffer/Makefile
new file mode 100644
index 0000000000..21d25d2474
--- /dev/null
+++ b/src/gallium/auxiliary/pipebuffer/Makefile
@@ -0,0 +1,18 @@
+TOP = ../../../..
+include $(TOP)/configs/current
+
+LIBNAME = pipebuffer
+
+C_SOURCES = \
+	pb_buffer_fenced.c \
+	pb_buffer_malloc.c \
+	pb_bufmgr_alt.c \
+	pb_bufmgr_cache.c \
+	pb_bufmgr_debug.c \
+	pb_bufmgr_mm.c \
+	pb_bufmgr_ondemand.c \
+	pb_bufmgr_pool.c \
+	pb_bufmgr_slab.c \
+	pb_validate.c
+
+include ../../Makefile.template
diff --git a/src/gallium/auxiliary/pipebuffer/SConscript b/src/gallium/auxiliary/pipebuffer/SConscript
new file mode 100644
index 0000000000..a074a55471
--- /dev/null
+++ b/src/gallium/auxiliary/pipebuffer/SConscript
@@ -0,0 +1,18 @@
+Import('*')
+
+pipebuffer = env.ConvenienceLibrary(
+	target = 'pipebuffer',
+	source = [
+		'pb_buffer_fenced.c',
+		'pb_buffer_malloc.c',
+		'pb_bufmgr_alt.c',
+		'pb_bufmgr_cache.c',
+		'pb_bufmgr_debug.c',
+		'pb_bufmgr_mm.c',
+		'pb_bufmgr_ondemand.c',
+		'pb_bufmgr_pool.c',
+		'pb_bufmgr_slab.c',
+		'pb_validate.c',
+	])
+
+auxiliaries.insert(0, pipebuffer)
diff --git a/src/gallium/auxiliary/pipebuffer/pb_buffer_fenced.c b/src/gallium/auxiliary/pipebuffer/pb_buffer_fenced.c
index a4b78f1494..ba087ac0f3 100644
--- a/src/gallium/auxiliary/pipebuffer/pb_buffer_fenced.c
+++ b/src/gallium/auxiliary/pipebuffer/pb_buffer_fenced.c
@@ -1,6 +1,6 @@
 /**************************************************************************
  *
- * Copyright 2007-2009 VMware, Inc.
+ * Copyright 2007-2010 VMware, Inc.
  * All Rights Reserved.
  *
  * Permission is hereby granted, free of charge, to any person obtaining a
@@ -28,9 +28,9 @@
 /**
  * \file
  * Implementation of fenced buffers.
- * 
- * \author Jose Fonseca <jrfonseca-at-tungstengraphics-dot-com>
- * \author Thomas Hellström <thomas-at-tungstengraphics-dot-com>
+ *
+ * \author Jose Fonseca <jfonseca-at-vmware-dot-com>
+ * \author Thomas Hellström <thellstrom-at-vmware-dot-com>
  */
 
 
@@ -50,6 +50,7 @@
 
 #include "pb_buffer.h"
 #include "pb_buffer_fenced.h"
+#include "pb_bufmgr.h"
 
 
 
@@ -59,32 +60,79 @@
 #define SUPER(__derived) (&(__derived)->base)
 
 
-struct fenced_buffer_list
+struct fenced_manager
 {
-   pipe_mutex mutex;
-   
+   struct pb_manager base;
+   struct pb_manager *provider;
    struct pb_fence_ops *ops;
-   
-   pb_size numDelayed;
-   struct list_head delayed;
-   
-#ifdef DEBUG
-   pb_size numUnfenced;
+
+   /**
+    * Maximum buffer size that can be safely allocated.
+    */
+   pb_size max_buffer_size;
+
+   /**
+    * Maximum cpu memory we can allocate before we start waiting for the
+    * GPU to idle.
+    */
+   pb_size max_cpu_total_size;
+
+   /**
+    * Following members are mutable and protected by this mutex.
+    */
+   pipe_mutex mutex;
+
+   /**
+    * Fenced buffer list.
+    *
+    * All fenced buffers are placed in this listed, ordered from the oldest
+    * fence to the newest fence.
+    */
+   struct list_head fenced;
+   pb_size num_fenced;
+
    struct list_head unfenced;
-#endif
+   pb_size num_unfenced;
+
+   /**
+    * How much temporary CPU memory is being used to hold unvalidated buffers.
+    */
+   pb_size cpu_total_size;
 };
 
 
 /**
+ * Fenced buffer.
+ *
  * Wrapper around a pipe buffer which adds fencing and reference counting.
  */
 struct fenced_buffer
 {
+   /*
+    * Immutable members.
+    */
+
    struct pb_buffer base;
-   
+   struct fenced_manager *mgr;
+
+   /*
+    * Following members are mutable and protected by fenced_manager::mutex.
+    */
+
+   struct list_head head;
+
+   /**
+    * Buffer with storage.
+    */
    struct pb_buffer *buffer;
+   pb_size size;
+   struct pb_desc desc;
 
-   /* FIXME: protect access with mutex */
+   /**
+    * Temporary CPU storage data. Used when there isn't enough GPU memory to
+    * store the buffer.
+    */
+   void *data;
 
    /**
     * A bitmask of PIPE_BUFFER_USAGE_CPU/GPU_READ/WRITE describing the current
@@ -93,15 +141,22 @@ struct fenced_buffer
    unsigned flags;
 
    unsigned mapcount;
+
    struct pb_validate *vl;
    unsigned validation_flags;
-   struct pipe_fence_handle *fence;
 
-   struct list_head head;
-   struct fenced_buffer_list *list;
+   struct pipe_fence_handle *fence;
 };
 
 
+static INLINE struct fenced_manager *
+fenced_manager(struct pb_manager *mgr)
+{
+   assert(mgr);
+   return (struct fenced_manager *)mgr;
+}
+
+
 static INLINE struct fenced_buffer *
 fenced_buffer(struct pb_buffer *buf)
 {
@@ -110,221 +165,568 @@ fenced_buffer(struct pb_buffer *buf)
 }
 
 
-static INLINE void
-_fenced_buffer_add(struct fenced_buffer *fenced_buf)
-{
-   struct fenced_buffer_list *fenced_list = fenced_buf->list;
+static void
+fenced_buffer_destroy_cpu_storage_locked(struct fenced_buffer *fenced_buf);
 
-   assert(pipe_is_referenced(&fenced_buf->base.base.reference));
-   assert(fenced_buf->flags & PIPE_BUFFER_USAGE_GPU_READ_WRITE);
-   assert(fenced_buf->fence);
+static enum pipe_error
+fenced_buffer_create_cpu_storage_locked(struct fenced_manager *fenced_mgr,
+                                        struct fenced_buffer *fenced_buf);
 
+static void
+fenced_buffer_destroy_gpu_storage_locked(struct fenced_buffer *fenced_buf);
+
+static enum pipe_error
+fenced_buffer_create_gpu_storage_locked(struct fenced_manager *fenced_mgr,
+                                        struct fenced_buffer *fenced_buf,
+                                        boolean wait);
+
+static enum pipe_error
+fenced_buffer_copy_storage_to_gpu_locked(struct fenced_buffer *fenced_buf);
+
+static enum pipe_error
+fenced_buffer_copy_storage_to_cpu_locked(struct fenced_buffer *fenced_buf);
+
+
+/**
+ * Dump the fenced buffer list.
+ *
+ * Useful to understand failures to allocate buffers.
+ */
+static void
+fenced_manager_dump_locked(struct fenced_manager *fenced_mgr)
+{
 #ifdef DEBUG
-   LIST_DEL(&fenced_buf->head);
-   assert(fenced_list->numUnfenced);
-   --fenced_list->numUnfenced;
+   struct pb_fence_ops *ops = fenced_mgr->ops;
+   struct list_head *curr, *next;
+   struct fenced_buffer *fenced_buf;
+
+   debug_printf("%10s %7s %8s %7s %10s %s\n",
+                "buffer", "size", "refcount", "storage", "fence", "signalled");
+
+   curr = fenced_mgr->unfenced.next;
+   next = curr->next;
+   while(curr != &fenced_mgr->unfenced) {
+      fenced_buf = LIST_ENTRY(struct fenced_buffer, curr, head);
+      assert(!fenced_buf->fence);
+      debug_printf("%10p %7u %8u %7s\n",
+                   (void *) fenced_buf,
+                   fenced_buf->base.base.size,
+                   p_atomic_read(&fenced_buf->base.base.reference.count),
+                   fenced_buf->buffer ? "gpu" : (fenced_buf->data ? "cpu" : "none"));
+      curr = next;
+      next = curr->next;
+   }
+
+   curr = fenced_mgr->fenced.next;
+   next = curr->next;
+   while(curr != &fenced_mgr->fenced) {
+      int signaled;
+      fenced_buf = LIST_ENTRY(struct fenced_buffer, curr, head);
+      assert(fenced_buf->buffer);
+      signaled = ops->fence_signalled(ops, fenced_buf->fence, 0);
+      debug_printf("%10p %7u %8u %7s %10p %s\n",
+                   (void *) fenced_buf,
+                   fenced_buf->base.base.size,
+                   p_atomic_read(&fenced_buf->base.base.reference.count),
+                   "gpu",
+                   (void *) fenced_buf->fence,
+                   signaled == 0 ? "y" : "n");
+      curr = next;
+      next = curr->next;
+   }
+#else
+   (void)fenced_mgr;
 #endif
-   LIST_ADDTAIL(&fenced_buf->head, &fenced_list->delayed);
-   ++fenced_list->numDelayed;
 }
 
 
-/**
- * Actually destroy the buffer.
- */
 static INLINE void
-_fenced_buffer_destroy(struct fenced_buffer *fenced_buf)
+fenced_buffer_destroy_locked(struct fenced_manager *fenced_mgr,
+                             struct fenced_buffer *fenced_buf)
 {
-   struct fenced_buffer_list *fenced_list = fenced_buf->list;
-   
    assert(!pipe_is_referenced(&fenced_buf->base.base.reference));
+
    assert(!fenced_buf->fence);
-#ifdef DEBUG
    assert(fenced_buf->head.prev);
    assert(fenced_buf->head.next);
    LIST_DEL(&fenced_buf->head);
-   assert(fenced_list->numUnfenced);
-   --fenced_list->numUnfenced;
-#else
-   (void)fenced_list;
-#endif
-   pb_reference(&fenced_buf->buffer, NULL);
+   assert(fenced_mgr->num_unfenced);
+   --fenced_mgr->num_unfenced;
+
+   fenced_buffer_destroy_gpu_storage_locked(fenced_buf);
+   fenced_buffer_destroy_cpu_storage_locked(fenced_buf);
+
    FREE(fenced_buf);
 }
 
 
+/**
+ * Add the buffer to the fenced list.
+ *
+ * Reference count should be incremented before calling this function.
+ */
 static INLINE void
-_fenced_buffer_remove(struct fenced_buffer_list *fenced_list,
-                      struct fenced_buffer *fenced_buf)
+fenced_buffer_add_locked(struct fenced_manager *fenced_mgr,
+                         struct fenced_buffer *fenced_buf)
+{
+   assert(pipe_is_referenced(&fenced_buf->base.base.reference));
+   assert(fenced_buf->flags & PIPE_BUFFER_USAGE_GPU_READ_WRITE);
+   assert(fenced_buf->fence);
+
+   p_atomic_inc(&fenced_buf->base.base.reference.count);
+
+   LIST_DEL(&fenced_buf->head);
+   assert(fenced_mgr->num_unfenced);
+   --fenced_mgr->num_unfenced;
+   LIST_ADDTAIL(&fenced_buf->head, &fenced_mgr->fenced);
+   ++fenced_mgr->num_fenced;
+}
+
+
+/**
+ * Remove the buffer from the fenced list, and potentially destroy the buffer
+ * if the reference count reaches zero.
+ *
+ * Returns TRUE if the buffer was detroyed.
+ */
+static INLINE boolean
+fenced_buffer_remove_locked(struct fenced_manager *fenced_mgr,
+                            struct fenced_buffer *fenced_buf)
 {
-   struct pb_fence_ops *ops = fenced_list->ops;
+   struct pb_fence_ops *ops = fenced_mgr->ops;
 
    assert(fenced_buf->fence);
-   assert(fenced_buf->list == fenced_list);
-   
+   assert(fenced_buf->mgr == fenced_mgr);
+
    ops->fence_reference(ops, &fenced_buf->fence, NULL);
    fenced_buf->flags &= ~PIPE_BUFFER_USAGE_GPU_READ_WRITE;
-   
+
    assert(fenced_buf->head.prev);
    assert(fenced_buf->head.next);
-   
+
    LIST_DEL(&fenced_buf->head);
-   assert(fenced_list->numDelayed);
-   --fenced_list->numDelayed;
-   
-#ifdef DEBUG
-   LIST_ADDTAIL(&fenced_buf->head, &fenced_list->unfenced);
-   ++fenced_list->numUnfenced;
-#endif
-   
-   /**
-    * FIXME!!!
-    */
+   assert(fenced_mgr->num_fenced);
+   --fenced_mgr->num_fenced;
+
+   LIST_ADDTAIL(&fenced_buf->head, &fenced_mgr->unfenced);
+   ++fenced_mgr->num_unfenced;
+
+   if (p_atomic_dec_zero(&fenced_buf->base.base.reference.count)) {
+      fenced_buffer_destroy_locked(fenced_mgr, fenced_buf);
+      return TRUE;
+   }
 
-   if(!pipe_is_referenced(&fenced_buf->base.base.reference))
-      _fenced_buffer_destroy(fenced_buf);
+   return FALSE;
 }
 
 
+/**
+ * Wait for the fence to expire, and remove it from the fenced list.
+ *
+ * This function will release and re-aquire the mutex, so any copy of mutable
+ * state must be discarded after calling it.
+ */
 static INLINE enum pipe_error
-_fenced_buffer_finish(struct fenced_buffer *fenced_buf)
+fenced_buffer_finish_locked(struct fenced_manager *fenced_mgr,
+                            struct fenced_buffer *fenced_buf)
 {
-   struct fenced_buffer_list *fenced_list = fenced_buf->list;
-   struct pb_fence_ops *ops = fenced_list->ops;
+   struct pb_fence_ops *ops = fenced_mgr->ops;
+   enum pipe_error ret = PIPE_ERROR;
 
 #if 0
    debug_warning("waiting for GPU");
 #endif
 
+   assert(pipe_is_referenced(&fenced_buf->base.base.reference));
    assert(fenced_buf->fence);
+
    if(fenced_buf->fence) {
-      if(ops->fence_finish(ops, fenced_buf->fence, 0) != 0) {
-	 return PIPE_ERROR;
+      struct pipe_fence_handle *fence = NULL;
+      int finished;
+      boolean proceed;
+
+      ops->fence_reference(ops, &fence, fenced_buf->fence);
+
+      pipe_mutex_unlock(fenced_mgr->mutex);
+
+      finished = ops->fence_finish(ops, fenced_buf->fence, 0);
+
+      pipe_mutex_lock(fenced_mgr->mutex);
+
+      assert(pipe_is_referenced(&fenced_buf->base.base.reference));
+
+      /*
+       * Only proceed if the fence object didn't change in the meanwhile.
+       * Otherwise assume the work has been already carried out by another
+       * thread that re-aquired the lock before us.
+       */
+      proceed = fence == fenced_buf->fence ? TRUE : FALSE;
+
+      ops->fence_reference(ops, &fence, NULL);
+
+      if(proceed && finished == 0) {
+         /*
+          * Remove from the fenced list
+          */
+
+         boolean destroyed;
+
+         destroyed = fenced_buffer_remove_locked(fenced_mgr, fenced_buf);
+
+         /* TODO: remove consequents buffers with the same fence? */
+
+         assert(!destroyed);
+
+         fenced_buf->flags &= ~PIPE_BUFFER_USAGE_GPU_READ_WRITE;
+
+         ret = PIPE_OK;
       }
-      /* Remove from the fenced list */
-      /* TODO: remove consequents */
-      _fenced_buffer_remove(fenced_list, fenced_buf);
    }
 
-   fenced_buf->flags &= ~PIPE_BUFFER_USAGE_GPU_READ_WRITE;
-   return PIPE_OK;
+   return ret;
 }
 
 
 /**
- * Free as many fenced buffers from the list head as possible. 
+ * Remove as many fenced buffers from the fenced list as possible.
+ *
+ * Returns TRUE if at least one buffer was removed.
  */
-static void
-_fenced_buffer_list_check_free(struct fenced_buffer_list *fenced_list, 
-                               int wait)
+static boolean
+fenced_manager_check_signalled_locked(struct fenced_manager *fenced_mgr,
+                                      boolean wait)
 {
-   struct pb_fence_ops *ops = fenced_list->ops;
+   struct pb_fence_ops *ops = fenced_mgr->ops;
    struct list_head *curr, *next;
    struct fenced_buffer *fenced_buf;
-   struct pb_buffer *pb_buf;
    struct pipe_fence_handle *prev_fence = NULL;
+   boolean ret = FALSE;
 
-   curr = fenced_list->delayed.next;
+   curr = fenced_mgr->fenced.next;
    next = curr->next;
-   while(curr != &fenced_list->delayed) {
+   while(curr != &fenced_mgr->fenced) {
       fenced_buf = LIST_ENTRY(struct fenced_buffer, curr, head);
 
       if(fenced_buf->fence != prev_fence) {
 	 int signaled;
-	 if (wait)
+
+	 if (wait) {
 	    signaled = ops->fence_finish(ops, fenced_buf->fence, 0);
-	 else
+
+	    /*
+	     * Don't return just now. Instead preemptively check if the
+	     * following buffers' fences already expired, without further waits.
+	     */
+	    wait = FALSE;
+	 }
+	 else {
 	    signaled = ops->fence_signalled(ops, fenced_buf->fence, 0);
-	 if (signaled != 0)
-	    break;
+	 }
+
+	 if (signaled != 0) {
+	    return ret;
+         }
+
 	 prev_fence = fenced_buf->fence;
       }
       else {
+         /* This buffer's fence object is identical to the previous buffer's
+          * fence object, so no need to check the fence again.
+          */
 	 assert(ops->fence_signalled(ops, fenced_buf->fence, 0) == 0);
       }
 
-      _fenced_buffer_remove(fenced_list, fenced_buf);
+      fenced_buffer_remove_locked(fenced_mgr, fenced_buf);
+
+      ret = TRUE;
+
+      curr = next;
+      next = curr->next;
+   }
+
+   return ret;
+}
+
+
+/**
+ * Try to free some GPU memory by backing it up into CPU memory.
+ *
+ * Returns TRUE if at least one buffer was freed.
+ */
+static boolean
+fenced_manager_free_gpu_storage_locked(struct fenced_manager *fenced_mgr)
+{
+   struct list_head *curr, *next;
+   struct fenced_buffer *fenced_buf;
 
-      curr = next; 
+   curr = fenced_mgr->unfenced.next;
+   next = curr->next;
+   while(curr != &fenced_mgr->unfenced) {
+      fenced_buf = LIST_ENTRY(struct fenced_buffer, curr, head);
+
+      /*
+       * We can only move storage if the buffer is not mapped and not
+       * validated.
+       */
+      if(fenced_buf->buffer &&
+         !fenced_buf->mapcount &&
+         !fenced_buf->vl) {
+         enum pipe_error ret;
+
+         ret = fenced_buffer_create_cpu_storage_locked(fenced_mgr, fenced_buf);
+         if(ret == PIPE_OK) {
+            ret = fenced_buffer_copy_storage_to_cpu_locked(fenced_buf);
+            if(ret == PIPE_OK) {
+               fenced_buffer_destroy_gpu_storage_locked(fenced_buf);
+               return TRUE;
+            }
+            fenced_buffer_destroy_cpu_storage_locked(fenced_buf);
+         }
+      }
+
+      curr = next;
       next = curr->next;
    }
+
+   return FALSE;
 }
 
 
+/**
+ * Destroy CPU storage for this buffer.
+ */
 static void
-fenced_buffer_destroy(struct pb_buffer *buf)
+fenced_buffer_destroy_cpu_storage_locked(struct fenced_buffer *fenced_buf)
 {
-   struct fenced_buffer *fenced_buf = fenced_buffer(buf);   
-   struct fenced_buffer_list *fenced_list = fenced_buf->list;
+   if(fenced_buf->data) {
+      align_free(fenced_buf->data);
+      fenced_buf->data = NULL;
+      assert(fenced_buf->mgr->cpu_total_size >= fenced_buf->size);
+      fenced_buf->mgr->cpu_total_size -= fenced_buf->size;
+   }
+}
 
-   pipe_mutex_lock(fenced_list->mutex);
-   assert(!pipe_is_referenced(&fenced_buf->base.base.reference));
-   if (fenced_buf->fence) {
-      struct pb_fence_ops *ops = fenced_list->ops;
-      if(ops->fence_signalled(ops, fenced_buf->fence, 0) == 0) {
-	 struct list_head *curr, *prev;
-	 curr = &fenced_buf->head;
-	 prev = curr->prev;
-	 do {
-	    fenced_buf = LIST_ENTRY(struct fenced_buffer, curr, head);
-	    assert(ops->fence_signalled(ops, fenced_buf->fence, 0) == 0);
-	    _fenced_buffer_remove(fenced_list, fenced_buf);
-	    curr = prev;
-	    prev = curr->prev;
-	 } while (curr != &fenced_list->delayed);
-      }	  
-      else {
-	 /* delay destruction */
+
+/**
+ * Create CPU storage for this buffer.
+ */
+static enum pipe_error
+fenced_buffer_create_cpu_storage_locked(struct fenced_manager *fenced_mgr,
+                                        struct fenced_buffer *fenced_buf)
+{
+   assert(!fenced_buf->data);
+   if(fenced_buf->data)
+      return PIPE_OK;
+
+   if (fenced_mgr->cpu_total_size + fenced_buf->size > fenced_mgr->max_cpu_total_size)
+      return PIPE_ERROR_OUT_OF_MEMORY;
+
+   fenced_buf->data = align_malloc(fenced_buf->size, fenced_buf->desc.alignment);
+   if(!fenced_buf->data)
+      return PIPE_ERROR_OUT_OF_MEMORY;
+
+   fenced_mgr->cpu_total_size += fenced_buf->size;
+
+   return PIPE_OK;
+}
+
+
+/**
+ * Destroy the GPU storage.
+ */
+static void
+fenced_buffer_destroy_gpu_storage_locked(struct fenced_buffer *fenced_buf)
+{
+   if(fenced_buf->buffer) {
+      pb_reference(&fenced_buf->buffer, NULL);
+   }
+}
+
+
+/**
+ * Try to create GPU storage for this buffer.
+ *
+ * This function is a shorthand around pb_manager::create_buffer for
+ * fenced_buffer_create_gpu_storage_locked()'s benefit.
+ */
+static INLINE boolean
+fenced_buffer_try_create_gpu_storage_locked(struct fenced_manager *fenced_mgr,
+                                            struct fenced_buffer *fenced_buf)
+{
+   struct pb_manager *provider = fenced_mgr->provider;
+
+   assert(!fenced_buf->buffer);
+
+   fenced_buf->buffer = provider->create_buffer(fenced_mgr->provider,
+                                                fenced_buf->size,
+                                                &fenced_buf->desc);
+   return fenced_buf->buffer ? TRUE : FALSE;
+}
+
+
+/**
+ * Create GPU storage for this buffer.
+ */
+static enum pipe_error
+fenced_buffer_create_gpu_storage_locked(struct fenced_manager *fenced_mgr,
+                                        struct fenced_buffer *fenced_buf,
+                                        boolean wait)
+{
+   assert(!fenced_buf->buffer);
+
+   /*
+    * Check for signaled buffers before trying to allocate.
+    */
+   fenced_manager_check_signalled_locked(fenced_mgr, FALSE);
+
+   fenced_buffer_try_create_gpu_storage_locked(fenced_mgr, fenced_buf);
+
+   /*
+    * Keep trying while there is some sort of progress:
+    * - fences are expiring,
+    * - or buffers are being being swapped out from GPU memory into CPU memory.
+    */
+   while(!fenced_buf->buffer &&
+         (fenced_manager_check_signalled_locked(fenced_mgr, FALSE) ||
+          fenced_manager_free_gpu_storage_locked(fenced_mgr))) {
+      fenced_buffer_try_create_gpu_storage_locked(fenced_mgr, fenced_buf);
+   }
+
+   if(!fenced_buf->buffer && wait) {
+      /*
+       * Same as before, but this time around, wait to free buffers if
+       * necessary.
+       */
+      while(!fenced_buf->buffer &&
+            (fenced_manager_check_signalled_locked(fenced_mgr, TRUE) ||
+             fenced_manager_free_gpu_storage_locked(fenced_mgr))) {
+         fenced_buffer_try_create_gpu_storage_locked(fenced_mgr, fenced_buf);
       }
    }
-   else {
-      _fenced_buffer_destroy(fenced_buf);
+
+   if(!fenced_buf->buffer) {
+      if(0)
+         fenced_manager_dump_locked(fenced_mgr);
+
+      /* give up */
+      return PIPE_ERROR_OUT_OF_MEMORY;
    }
-   pipe_mutex_unlock(fenced_list->mutex);
+
+   return PIPE_OK;
+}
+
+
+static enum pipe_error
+fenced_buffer_copy_storage_to_gpu_locked(struct fenced_buffer *fenced_buf)
+{
+   uint8_t *map;
+
+   assert(fenced_buf->data);
+   assert(fenced_buf->buffer);
+
+   map = pb_map(fenced_buf->buffer, PIPE_BUFFER_USAGE_CPU_WRITE);
+   if(!map)
+      return PIPE_ERROR;
+
+   memcpy(map, fenced_buf->data, fenced_buf->size);
+
+   pb_unmap(fenced_buf->buffer);
+
+   return PIPE_OK;
+}
+
+
+static enum pipe_error
+fenced_buffer_copy_storage_to_cpu_locked(struct fenced_buffer *fenced_buf)
+{
+   const uint8_t *map;
+
+   assert(fenced_buf->data);
+   assert(fenced_buf->buffer);
+
+   map = pb_map(fenced_buf->buffer, PIPE_BUFFER_USAGE_CPU_READ);
+   if(!map)
+      return PIPE_ERROR;
+
+   memcpy(fenced_buf->data, map, fenced_buf->size);
+
+   pb_unmap(fenced_buf->buffer);
+
+   return PIPE_OK;
+}
+
+
+static void
+fenced_buffer_destroy(struct pb_buffer *buf)
+{
+   struct fenced_buffer *fenced_buf = fenced_buffer(buf);
+   struct fenced_manager *fenced_mgr = fenced_buf->mgr;
+
+   assert(!pipe_is_referenced(&fenced_buf->base.base.reference));
+
+   pipe_mutex_lock(fenced_mgr->mutex);
+
+   fenced_buffer_destroy_locked(fenced_mgr, fenced_buf);
+
+   pipe_mutex_unlock(fenced_mgr->mutex);
 }
 
 
 static void *
-fenced_buffer_map(struct pb_buffer *buf, 
+fenced_buffer_map(struct pb_buffer *buf,
                   unsigned flags)
 {
    struct fenced_buffer *fenced_buf = fenced_buffer(buf);
-   struct fenced_buffer_list *fenced_list = fenced_buf->list;
-   struct pb_fence_ops *ops = fenced_list->ops;
-   void *map;
+   struct fenced_manager *fenced_mgr = fenced_buf->mgr;
+   struct pb_fence_ops *ops = fenced_mgr->ops;
+   void *map = NULL;
+
+   pipe_mutex_lock(fenced_mgr->mutex);
 
    assert(!(flags & PIPE_BUFFER_USAGE_GPU_READ_WRITE));
-   
-   /* Serialize writes */
-   if((fenced_buf->flags & PIPE_BUFFER_USAGE_GPU_WRITE) ||
-      ((fenced_buf->flags & PIPE_BUFFER_USAGE_GPU_READ) && (flags & PIPE_BUFFER_USAGE_CPU_WRITE))) {
-      if(flags & PIPE_BUFFER_USAGE_DONTBLOCK) {
-         /* Don't wait for the GPU to finish writing */
-         if(ops->fence_signalled(ops, fenced_buf->fence, 0) == 0)
-            _fenced_buffer_remove(fenced_list, fenced_buf);
-         else
-            return NULL;
+
+   /*
+    * Serialize writes.
+    */
+   while((fenced_buf->flags & PIPE_BUFFER_USAGE_GPU_WRITE) ||
+         ((fenced_buf->flags & PIPE_BUFFER_USAGE_GPU_READ) &&
+          (flags & PIPE_BUFFER_USAGE_CPU_WRITE))) {
+
+      /* 
+       * Don't wait for the GPU to finish accessing it, if blocking is forbidden.
+       */
+      if((flags & PIPE_BUFFER_USAGE_DONTBLOCK) &&
+          ops->fence_signalled(ops, fenced_buf->fence, 0) == 0) {
+         goto done;
       }
-      else {
-         /* Wait for the GPU to finish writing */
-         _fenced_buffer_finish(fenced_buf);
+
+      if (flags & PIPE_BUFFER_USAGE_UNSYNCHRONIZED) {
+         break;
       }
+
+      /*
+       * Wait for the GPU to finish accessing. This will release and re-acquire
+       * the mutex, so all copies of mutable state must be discarded.
+       */
+      fenced_buffer_finish_locked(fenced_mgr, fenced_buf);
    }
 
-#if 0
-   /* Check for CPU write access (read is OK) */
-   if(fenced_buf->flags & PIPE_BUFFER_USAGE_CPU_READ_WRITE) {
-      /* this is legal -- just for debugging */
-      debug_warning("concurrent CPU writes");
+   if(fenced_buf->buffer) {
+      map = pb_map(fenced_buf->buffer, flags);
    }
-#endif
-   
-   map = pb_map(fenced_buf->buffer, flags);
+   else {
+      assert(fenced_buf->data);
+      map = fenced_buf->data;
+   }
+
    if(map) {
       ++fenced_buf->mapcount;
       fenced_buf->flags |= flags & PIPE_BUFFER_USAGE_CPU_READ_WRITE;
    }
 
+done:
+   pipe_mutex_unlock(fenced_mgr->mutex);
+
    return map;
 }
 
@@ -333,13 +735,20 @@ static void
 fenced_buffer_unmap(struct pb_buffer *buf)
 {
    struct fenced_buffer *fenced_buf = fenced_buffer(buf);
+   struct fenced_manager *fenced_mgr = fenced_buf->mgr;
+
+   pipe_mutex_lock(fenced_mgr->mutex);
+
    assert(fenced_buf->mapcount);
    if(fenced_buf->mapcount) {
-      pb_unmap(fenced_buf->buffer);
+      if (fenced_buf->buffer)
+         pb_unmap(fenced_buf->buffer);
       --fenced_buf->mapcount;
       if(!fenced_buf->mapcount)
 	 fenced_buf->flags &= ~PIPE_BUFFER_USAGE_CPU_READ_WRITE;
    }
+
+   pipe_mutex_unlock(fenced_mgr->mutex);
 }
 
 
@@ -349,48 +758,72 @@ fenced_buffer_validate(struct pb_buffer *buf,
                        unsigned flags)
 {
    struct fenced_buffer *fenced_buf = fenced_buffer(buf);
+   struct fenced_manager *fenced_mgr = fenced_buf->mgr;
    enum pipe_error ret;
-   
+
+   pipe_mutex_lock(fenced_mgr->mutex);
+
    if(!vl) {
       /* invalidate */
       fenced_buf->vl = NULL;
       fenced_buf->validation_flags = 0;
-      return PIPE_OK;
+      ret = PIPE_OK;
+      goto done;
    }
-   
+
    assert(flags & PIPE_BUFFER_USAGE_GPU_READ_WRITE);
    assert(!(flags & ~PIPE_BUFFER_USAGE_GPU_READ_WRITE));
    flags &= PIPE_BUFFER_USAGE_GPU_READ_WRITE;
 
-   /* Buffer cannot be validated in two different lists */ 
-   if(fenced_buf->vl && fenced_buf->vl != vl)
-      return PIPE_ERROR_RETRY;
-   
-#if 0
-   /* Do not validate if buffer is still mapped */
-   if(fenced_buf->flags & PIPE_BUFFER_USAGE_CPU_READ_WRITE) {
-      /* TODO: wait for the thread that mapped the buffer to unmap it */
-      return PIPE_ERROR_RETRY;
+   /* Buffer cannot be validated in two different lists */
+   if(fenced_buf->vl && fenced_buf->vl != vl) {
+      ret = PIPE_ERROR_RETRY;
+      goto done;
    }
-   /* Final sanity checking */
-   assert(!(fenced_buf->flags & PIPE_BUFFER_USAGE_CPU_READ_WRITE));
-   assert(!fenced_buf->mapcount);
-#endif
 
    if(fenced_buf->vl == vl &&
       (fenced_buf->validation_flags & flags) == flags) {
       /* Nothing to do -- buffer already validated */
-      return PIPE_OK;
+      ret = PIPE_OK;
+      goto done;
+   }
+
+   /*
+    * Create and update GPU storage.
+    */
+   if(!fenced_buf->buffer) {
+      assert(!fenced_buf->mapcount);
+
+      ret = fenced_buffer_create_gpu_storage_locked(fenced_mgr, fenced_buf, TRUE);
+      if(ret != PIPE_OK) {
+         goto done;
+      }
+
+      ret = fenced_buffer_copy_storage_to_gpu_locked(fenced_buf);
+      if(ret != PIPE_OK) {
+         fenced_buffer_destroy_gpu_storage_locked(fenced_buf);
+         goto done;
+      }
+
+      if(fenced_buf->mapcount) {
+         debug_printf("warning: validating a buffer while it is still mapped\n");
+      }
+      else {
+         fenced_buffer_destroy_cpu_storage_locked(fenced_buf);
+      }
    }
-   
+
    ret = pb_validate(fenced_buf->buffer, vl, flags);
    if (ret != PIPE_OK)
-      return ret;
-   
+      goto done;
+
    fenced_buf->vl = vl;
    fenced_buf->validation_flags |= flags;
-   
-   return PIPE_OK;
+
+done:
+   pipe_mutex_unlock(fenced_mgr->mutex);
+
+   return ret;
 }
 
 
@@ -398,36 +831,37 @@ static void
 fenced_buffer_fence(struct pb_buffer *buf,
                     struct pipe_fence_handle *fence)
 {
-   struct fenced_buffer *fenced_buf;
-   struct fenced_buffer_list *fenced_list;
-   struct pb_fence_ops *ops;
+   struct fenced_buffer *fenced_buf = fenced_buffer(buf);
+   struct fenced_manager *fenced_mgr = fenced_buf->mgr;
+   struct pb_fence_ops *ops = fenced_mgr->ops;
 
-   fenced_buf = fenced_buffer(buf);
-   fenced_list = fenced_buf->list;
-   ops = fenced_list->ops;
-   
-   if(fence == fenced_buf->fence) {
-      /* Nothing to do */
-      return;
-   }
+   pipe_mutex_lock(fenced_mgr->mutex);
 
-   assert(fenced_buf->vl);
-   assert(fenced_buf->validation_flags);
-   
-   pipe_mutex_lock(fenced_list->mutex);
-   if (fenced_buf->fence)
-      _fenced_buffer_remove(fenced_list, fenced_buf);
-   if (fence) {
-      ops->fence_reference(ops, &fenced_buf->fence, fence);
-      fenced_buf->flags |= fenced_buf->validation_flags;
-      _fenced_buffer_add(fenced_buf);
+   assert(pipe_is_referenced(&fenced_buf->base.base.reference));
+   assert(fenced_buf->buffer);
+
+   if(fence != fenced_buf->fence) {
+      assert(fenced_buf->vl);
+      assert(fenced_buf->validation_flags);
+
+      if (fenced_buf->fence) {
+         boolean destroyed;
+         destroyed = fenced_buffer_remove_locked(fenced_mgr, fenced_buf);
+         assert(!destroyed);
+      }
+      if (fence) {
+         ops->fence_reference(ops, &fenced_buf->fence, fence);
+         fenced_buf->flags |= fenced_buf->validation_flags;
+         fenced_buffer_add_locked(fenced_mgr, fenced_buf);
+      }
+
+      pb_fence(fenced_buf->buffer, fence);
+
+      fenced_buf->vl = NULL;
+      fenced_buf->validation_flags = 0;
    }
-   pipe_mutex_unlock(fenced_list->mutex);
-   
-   pb_fence(fenced_buf->buffer, fence);
 
-   fenced_buf->vl = NULL;
-   fenced_buf->validation_flags = 0;
+   pipe_mutex_unlock(fenced_mgr->mutex);
 }
 
 
@@ -437,11 +871,29 @@ fenced_buffer_get_base_buffer(struct pb_buffer *buf,
                               pb_size *offset)
 {
    struct fenced_buffer *fenced_buf = fenced_buffer(buf);
-   pb_get_base_buffer(fenced_buf->buffer, base_buf, offset);
+   struct fenced_manager *fenced_mgr = fenced_buf->mgr;
+
+   pipe_mutex_lock(fenced_mgr->mutex);
+
+   /*
+    * This should only be called when the buffer is validated. Typically
+    * when processing relocations.
+    */
+   assert(fenced_buf->vl);
+   assert(fenced_buf->buffer);
+
+   if(fenced_buf->buffer)
+      pb_get_base_buffer(fenced_buf->buffer, base_buf, offset);
+   else {
+      *base_buf = buf;
+      *offset = 0;
+   }
+
+   pipe_mutex_unlock(fenced_mgr->mutex);
 }
 
 
-static const struct pb_vtbl 
+static const struct pb_vtbl
 fenced_buffer_vtbl = {
       fenced_buffer_destroy,
       fenced_buffer_map,
@@ -452,147 +904,166 @@ fenced_buffer_vtbl = {
 };
 
 
-struct pb_buffer *
-fenced_buffer_create(struct fenced_buffer_list *fenced_list, 
-                     struct pb_buffer *buffer)
+/**
+ * Wrap a buffer in a fenced buffer.
+ */
+static struct pb_buffer *
+fenced_bufmgr_create_buffer(struct pb_manager *mgr,
+                            pb_size size,
+                            const struct pb_desc *desc)
 {
-   struct fenced_buffer *buf;
-   
-   if(!buffer)
-      return NULL;
-   
-   buf = CALLOC_STRUCT(fenced_buffer);
-   if(!buf) {
-      pb_reference(&buffer, NULL);
-      return NULL;
+   struct fenced_manager *fenced_mgr = fenced_manager(mgr);
+   struct fenced_buffer *fenced_buf;
+   enum pipe_error ret;
+
+   /*
+    * Don't stall the GPU, waste time evicting buffers, or waste memory
+    * trying to create a buffer that will most likely never fit into the
+    * graphics aperture.
+    */
+   if(size > fenced_mgr->max_buffer_size) {
+      goto no_buffer;
    }
-   
-   pipe_reference_init(&buf->base.base.reference, 1);
-   buf->base.base.alignment = buffer->base.alignment;
-   buf->base.base.usage = buffer->base.usage;
-   buf->base.base.size = buffer->base.size;
-   
-   buf->base.vtbl = &fenced_buffer_vtbl;
-   buf->buffer = buffer;
-   buf->list = fenced_list;
-   
-#ifdef DEBUG
-   pipe_mutex_lock(fenced_list->mutex);
-   LIST_ADDTAIL(&buf->head, &fenced_list->unfenced);
-   ++fenced_list->numUnfenced;
-   pipe_mutex_unlock(fenced_list->mutex);
-#endif
 
-   return &buf->base;
-}
+   fenced_buf = CALLOC_STRUCT(fenced_buffer);
+   if(!fenced_buf)
+      goto no_buffer;
 
+   pipe_reference_init(&fenced_buf->base.base.reference, 1);
+   fenced_buf->base.base.alignment = desc->alignment;
+   fenced_buf->base.base.usage = desc->usage;
+   fenced_buf->base.base.size = size;
+   fenced_buf->size = size;
+   fenced_buf->desc = *desc;
 
-struct fenced_buffer_list *
-fenced_buffer_list_create(struct pb_fence_ops *ops) 
-{
-   struct fenced_buffer_list *fenced_list;
+   fenced_buf->base.vtbl = &fenced_buffer_vtbl;
+   fenced_buf->mgr = fenced_mgr;
 
-   fenced_list = CALLOC_STRUCT(fenced_buffer_list);
-   if (!fenced_list)
-      return NULL;
+   pipe_mutex_lock(fenced_mgr->mutex);
+
+   /*
+    * Try to create GPU storage without stalling,
+    */
+   ret = fenced_buffer_create_gpu_storage_locked(fenced_mgr, fenced_buf, FALSE);
 
-   fenced_list->ops = ops;
+   /*
+    * Attempt to use CPU memory to avoid stalling the GPU.
+    */
+   if(ret != PIPE_OK) {
+      ret = fenced_buffer_create_cpu_storage_locked(fenced_mgr, fenced_buf);
+   }
 
-   LIST_INITHEAD(&fenced_list->delayed);
-   fenced_list->numDelayed = 0;
-   
-#ifdef DEBUG
-   LIST_INITHEAD(&fenced_list->unfenced);
-   fenced_list->numUnfenced = 0;
-#endif
+   /*
+    * Create GPU storage, waiting for some to be available.
+    */
+   if(ret != PIPE_OK) {
+      ret = fenced_buffer_create_gpu_storage_locked(fenced_mgr, fenced_buf, TRUE);
+   }
+
+   /*
+    * Give up.
+    */
+   if(ret != PIPE_OK) {
+      goto no_storage;
+   }
 
-   pipe_mutex_init(fenced_list->mutex);
+   assert(fenced_buf->buffer || fenced_buf->data);
 
-   return fenced_list;
-}
+   LIST_ADDTAIL(&fenced_buf->head, &fenced_mgr->unfenced);
+   ++fenced_mgr->num_unfenced;
+   pipe_mutex_unlock(fenced_mgr->mutex);
 
+   return &fenced_buf->base;
 
-void
-fenced_buffer_list_check_free(struct fenced_buffer_list *fenced_list, 
-                              int wait)
-{
-   pipe_mutex_lock(fenced_list->mutex);
-   _fenced_buffer_list_check_free(fenced_list, wait);
-   pipe_mutex_unlock(fenced_list->mutex);
+no_storage:
+   pipe_mutex_unlock(fenced_mgr->mutex);
+   FREE(fenced_buf);
+no_buffer:
+   return NULL;
 }
 
 
-#ifdef DEBUG
-void
-fenced_buffer_list_dump(struct fenced_buffer_list *fenced_list)
+static void
+fenced_bufmgr_flush(struct pb_manager *mgr)
 {
-   struct pb_fence_ops *ops = fenced_list->ops;
-   struct list_head *curr, *next;
-   struct fenced_buffer *fenced_buf;
+   struct fenced_manager *fenced_mgr = fenced_manager(mgr);
 
-   pipe_mutex_lock(fenced_list->mutex);
+   pipe_mutex_lock(fenced_mgr->mutex);
+   while(fenced_manager_check_signalled_locked(fenced_mgr, TRUE))
+      ;
+   pipe_mutex_unlock(fenced_mgr->mutex);
 
-   debug_printf("%10s %7s %7s %10s %s\n",
-                "buffer", "size", "refcount", "fence", "signalled");
-   
-   curr = fenced_list->unfenced.next;
-   next = curr->next;
-   while(curr != &fenced_list->unfenced) {
-      fenced_buf = LIST_ENTRY(struct fenced_buffer, curr, head);
-      assert(!fenced_buf->fence);
-      debug_printf("%10p %7u %7u\n",
-                   (void *) fenced_buf,
-                   fenced_buf->base.base.size,
-                   p_atomic_read(&fenced_buf->base.base.reference.count));
-      curr = next; 
-      next = curr->next;
-   }
-   
-   curr = fenced_list->delayed.next;
-   next = curr->next;
-   while(curr != &fenced_list->delayed) {
-      int signaled;
-      fenced_buf = LIST_ENTRY(struct fenced_buffer, curr, head);
-      signaled = ops->fence_signalled(ops, fenced_buf->fence, 0);
-      debug_printf("%10p %7u %7u %10p %s\n",
-                   (void *) fenced_buf,
-                   fenced_buf->base.base.size,
-                   p_atomic_read(&fenced_buf->base.base.reference.count),
-                   (void *) fenced_buf->fence,
-                   signaled == 0 ? "y" : "n");
-      curr = next; 
-      next = curr->next;
-   }
-   
-   pipe_mutex_unlock(fenced_list->mutex);
+   assert(fenced_mgr->provider->flush);
+   if(fenced_mgr->provider->flush)
+      fenced_mgr->provider->flush(fenced_mgr->provider);
 }
-#endif
 
 
-void
-fenced_buffer_list_destroy(struct fenced_buffer_list *fenced_list)
+static void
+fenced_bufmgr_destroy(struct pb_manager *mgr)
 {
-   pipe_mutex_lock(fenced_list->mutex);
+   struct fenced_manager *fenced_mgr = fenced_manager(mgr);
+
+   pipe_mutex_lock(fenced_mgr->mutex);
 
    /* Wait on outstanding fences */
-   while (fenced_list->numDelayed) {
-      pipe_mutex_unlock(fenced_list->mutex);
+   while (fenced_mgr->num_fenced) {
+      pipe_mutex_unlock(fenced_mgr->mutex);
 #if defined(PIPE_OS_LINUX) || defined(PIPE_OS_BSD) || defined(PIPE_OS_SOLARIS)
       sched_yield();
 #endif
-      _fenced_buffer_list_check_free(fenced_list, 1);
-      pipe_mutex_lock(fenced_list->mutex);
+      pipe_mutex_lock(fenced_mgr->mutex);
+      while(fenced_manager_check_signalled_locked(fenced_mgr, TRUE))
+         ;
    }
 
 #ifdef DEBUG
-   /*assert(!fenced_list->numUnfenced);*/
+   /*assert(!fenced_mgr->num_unfenced);*/
 #endif
-      
-   pipe_mutex_unlock(fenced_list->mutex);
-   
-   fenced_list->ops->destroy(fenced_list->ops);
-   
-   FREE(fenced_list);
+
+   pipe_mutex_unlock(fenced_mgr->mutex);
+   pipe_mutex_destroy(fenced_mgr->mutex);
+
+   if(fenced_mgr->provider)
+      fenced_mgr->provider->destroy(fenced_mgr->provider);
+
+   fenced_mgr->ops->destroy(fenced_mgr->ops);
+
+   FREE(fenced_mgr);
 }
 
 
+struct pb_manager *
+fenced_bufmgr_create(struct pb_manager *provider,
+                     struct pb_fence_ops *ops,
+                     pb_size max_buffer_size,
+                     pb_size max_cpu_total_size)
+{
+   struct fenced_manager *fenced_mgr;
+
+   if(!provider)
+      return NULL;
+
+   fenced_mgr = CALLOC_STRUCT(fenced_manager);
+   if (!fenced_mgr)
+      return NULL;
+
+   fenced_mgr->base.destroy = fenced_bufmgr_destroy;
+   fenced_mgr->base.create_buffer = fenced_bufmgr_create_buffer;
+   fenced_mgr->base.flush = fenced_bufmgr_flush;
+
+   fenced_mgr->provider = provider;
+   fenced_mgr->ops = ops;
+   fenced_mgr->max_buffer_size = max_buffer_size;
+   fenced_mgr->max_cpu_total_size = max_cpu_total_size;
+
+   LIST_INITHEAD(&fenced_mgr->fenced);
+   fenced_mgr->num_fenced = 0;
+
+   LIST_INITHEAD(&fenced_mgr->unfenced);
+   fenced_mgr->num_unfenced = 0;
+
+   pipe_mutex_init(fenced_mgr->mutex);
+
+   return &fenced_mgr->base;
+}
diff --git a/src/gallium/auxiliary/pipebuffer/pb_buffer_fenced.h b/src/gallium/auxiliary/pipebuffer/pb_buffer_fenced.h
index 034ca1e024..0372f81d0a 100644
--- a/src/gallium/auxiliary/pipebuffer/pb_buffer_fenced.h
+++ b/src/gallium/auxiliary/pipebuffer/pb_buffer_fenced.h
@@ -98,43 +98,6 @@ struct pb_fence_ops
 };
 
 
-/**
- * Create a fenced buffer list.
- * 
- * See also fenced_bufmgr_create for a more convenient way to use this.
- */
-struct fenced_buffer_list *
-fenced_buffer_list_create(struct pb_fence_ops *ops);
-
-
-/**
- * Walk the fenced buffer list to check and free signalled buffers.
- */ 
-void
-fenced_buffer_list_check_free(struct fenced_buffer_list *fenced_list, 
-                              int wait);
-
-
-#ifdef DEBUG
-void
-fenced_buffer_list_dump(struct fenced_buffer_list *fenced_list);
-#endif
-
-
-void
-fenced_buffer_list_destroy(struct fenced_buffer_list *fenced_list);
-
-
-/**
- * Wrap a buffer in a fenced buffer.
- * 
- * NOTE: this will not increase the buffer reference count.
- */
-struct pb_buffer *
-fenced_buffer_create(struct fenced_buffer_list *fenced, 
-                     struct pb_buffer *buffer);
-
-
 #ifdef __cplusplus
 }
 #endif
diff --git a/src/gallium/auxiliary/pipebuffer/pb_bufmgr.h b/src/gallium/auxiliary/pipebuffer/pb_bufmgr.h
index 8c8d713078..06669917ff 100644
--- a/src/gallium/auxiliary/pipebuffer/pb_bufmgr.h
+++ b/src/gallium/auxiliary/pipebuffer/pb_bufmgr.h
@@ -175,7 +175,9 @@ struct pb_fence_ops;
  */
 struct pb_manager *
 fenced_bufmgr_create(struct pb_manager *provider,
-                     struct pb_fence_ops *ops);
+                     struct pb_fence_ops *ops,
+                     pb_size max_buffer_size,
+                     pb_size max_cpu_total_size);
 
 
 struct pb_manager *
diff --git a/src/gallium/auxiliary/pipebuffer/pb_bufmgr_debug.c b/src/gallium/auxiliary/pipebuffer/pb_bufmgr_debug.c
index 6e3214ca9c..8f74180a11 100644
--- a/src/gallium/auxiliary/pipebuffer/pb_bufmgr_debug.c
+++ b/src/gallium/auxiliary/pipebuffer/pb_bufmgr_debug.c
@@ -371,6 +371,9 @@ pb_debug_manager_create_buffer(struct pb_manager *_mgr,
    struct pb_desc real_desc;
    pb_size real_size;
    
+   assert(size);
+   assert(desc->alignment);
+
    buf = CALLOC_STRUCT(pb_debug_buffer);
    if(!buf)
       return NULL;
diff --git a/src/gallium/auxiliary/pipebuffer/pb_bufmgr_fenced.c b/src/gallium/auxiliary/pipebuffer/pb_bufmgr_fenced.c
deleted file mode 100644
index 97dd1427fd..0000000000
--- a/src/gallium/auxiliary/pipebuffer/pb_bufmgr_fenced.c
+++ /dev/null
@@ -1,152 +0,0 @@
-/**************************************************************************
- * 
- * Copyright 2006 Tungsten Graphics, Inc., Bismarck, ND., USA
- * All Rights Reserved.
- * 
- * Permission is hereby granted, free of charge, to any person obtaining a
- * copy of this software and associated documentation files (the
- * "Software"), to deal in the Software without restriction, including
- * without limitation the rights to use, copy, modify, merge, publish,
- * distribute, sub license, and/or sell copies of the Software, and to
- * permit persons to whom the Software is furnished to do so, subject to
- * the following conditions:
- * 
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
- * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
- * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
- * THE COPYRIGHT HOLDERS, AUTHORS AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM,
- * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR 
- * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE 
- * USE OR OTHER DEALINGS IN THE SOFTWARE.
- *
- * The above copyright notice and this permission notice (including the
- * next paragraph) shall be included in all copies or substantial portions
- * of the Software.
- * 
- * 
- **************************************************************************/
-
-/**
- * \file
- * A buffer manager that wraps buffers in fenced buffers.
- * 
- * \author Jose Fonseca <jrfonseca@tungstengraphics.dot.com>
- */
-
-
-#include "util/u_debug.h"
-#include "util/u_memory.h"
-
-#include "pb_buffer.h"
-#include "pb_buffer_fenced.h"
-#include "pb_bufmgr.h"
-
-
-struct fenced_pb_manager
-{
-   struct pb_manager base;
-
-   struct pb_manager *provider;
-   
-   struct fenced_buffer_list *fenced_list;
-};
-
-
-static INLINE struct fenced_pb_manager *
-fenced_pb_manager(struct pb_manager *mgr)
-{
-   assert(mgr);
-   return (struct fenced_pb_manager *)mgr;
-}
-
-
-static struct pb_buffer *
-fenced_bufmgr_create_buffer(struct pb_manager *mgr, 
-                            pb_size size,
-                            const struct pb_desc *desc)
-{
-   struct fenced_pb_manager *fenced_mgr = fenced_pb_manager(mgr);
-   struct pb_buffer *buf;
-   struct pb_buffer *fenced_buf;
-
-   /* check for free buffers before allocating new ones */
-   fenced_buffer_list_check_free(fenced_mgr->fenced_list, 0);
-   
-   buf = fenced_mgr->provider->create_buffer(fenced_mgr->provider, size, desc);
-   if(!buf) {
-      /* try harder to get a buffer */
-      fenced_buffer_list_check_free(fenced_mgr->fenced_list, 1);
-      
-      buf = fenced_mgr->provider->create_buffer(fenced_mgr->provider, size, desc);
-      if(!buf) {
-#if 0
-         fenced_buffer_list_dump(fenced_mgr->fenced_list);
-#endif
-         
-         /* give up */
-         return NULL;
-      }
-   }
-   
-   fenced_buf = fenced_buffer_create(fenced_mgr->fenced_list, buf);
-   if(!fenced_buf) {
-      pb_reference(&buf, NULL);
-   }
-   
-   return fenced_buf;
-}
-
-
-static void
-fenced_bufmgr_flush(struct pb_manager *mgr)
-{
-   struct fenced_pb_manager *fenced_mgr = fenced_pb_manager(mgr);
-
-   fenced_buffer_list_check_free(fenced_mgr->fenced_list, TRUE);
-
-   assert(fenced_mgr->provider->flush);
-   if(fenced_mgr->provider->flush)
-      fenced_mgr->provider->flush(fenced_mgr->provider);
-}
-
-
-static void
-fenced_bufmgr_destroy(struct pb_manager *mgr)
-{
-   struct fenced_pb_manager *fenced_mgr = fenced_pb_manager(mgr);
-
-   fenced_buffer_list_destroy(fenced_mgr->fenced_list);
-
-   if(fenced_mgr->provider)
-      fenced_mgr->provider->destroy(fenced_mgr->provider);
-   
-   FREE(fenced_mgr);
-}
-
-
-struct pb_manager *
-fenced_bufmgr_create(struct pb_manager *provider, 
-                     struct pb_fence_ops *ops) 
-{
-   struct fenced_pb_manager *fenced_mgr;
-
-   if(!provider)
-      return NULL;
-   
-   fenced_mgr = CALLOC_STRUCT(fenced_pb_manager);
-   if (!fenced_mgr)
-      return NULL;
-
-   fenced_mgr->base.destroy = fenced_bufmgr_destroy;
-   fenced_mgr->base.create_buffer = fenced_bufmgr_create_buffer;
-   fenced_mgr->base.flush = fenced_bufmgr_flush;
-
-   fenced_mgr->provider = provider;
-   fenced_mgr->fenced_list = fenced_buffer_list_create(ops);
-   if(!fenced_mgr->fenced_list) {
-      FREE(fenced_mgr);
-      return NULL;
-   }
-      
-   return &fenced_mgr->base;
-}
diff --git a/src/gallium/auxiliary/pipebuffer/pb_validate.c b/src/gallium/auxiliary/pipebuffer/pb_validate.c
index ce40c0cf0e..903afc749d 100644
--- a/src/gallium/auxiliary/pipebuffer/pb_validate.c
+++ b/src/gallium/auxiliary/pipebuffer/pb_validate.c
@@ -39,7 +39,6 @@
 #include "util/u_debug.h"
 
 #include "pb_buffer.h"
-#include "pb_buffer_fenced.h"
 #include "pb_validate.h"
 
 
diff --git a/src/gallium/auxiliary/tgsi/tgsi_build.c b/src/gallium/auxiliary/tgsi/tgsi_build.c
index de9cbc8630..e38b0be7ab 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_build.c
+++ b/src/gallium/auxiliary/tgsi/tgsi_build.c
@@ -103,6 +103,7 @@ tgsi_default_declaration( void )
    declaration.File = TGSI_FILE_NULL;
    declaration.UsageMask = TGSI_WRITEMASK_XYZW;
    declaration.Interpolate = TGSI_INTERPOLATE_CONSTANT;
+   declaration.Dimension = 0;
    declaration.Semantic = 0;
    declaration.Centroid = 0;
    declaration.Invariant = 0;
@@ -116,6 +117,7 @@ tgsi_build_declaration(
    unsigned file,
    unsigned usage_mask,
    unsigned interpolate,
+   unsigned dimension,
    unsigned semantic,
    unsigned centroid,
    unsigned invariant,
@@ -130,6 +132,7 @@ tgsi_build_declaration(
    declaration.File = file;
    declaration.UsageMask = usage_mask;
    declaration.Interpolate = interpolate;
+   declaration.Dimension = dimension;
    declaration.Semantic = semantic;
    declaration.Centroid = centroid;
    declaration.Invariant = invariant;
@@ -183,6 +186,7 @@ tgsi_build_full_declaration(
       full_decl->Declaration.File,
       full_decl->Declaration.UsageMask,
       full_decl->Declaration.Interpolate,
+      full_decl->Declaration.Dimension,
       full_decl->Declaration.Semantic,
       full_decl->Declaration.Centroid,
       full_decl->Declaration.Invariant,
@@ -199,6 +203,20 @@ tgsi_build_full_declaration(
       declaration,
       header );
 
+   if (full_decl->Declaration.Dimension) {
+      struct tgsi_declaration_dimension *dd;
+
+      if (maxsize <= size) {
+         return 0;
+      }
+      dd = (struct tgsi_declaration_dimension *)&tokens[size];
+      size++;
+
+      *dd = tgsi_build_declaration_dimension(full_decl->Dim.Index2D,
+                                             declaration,
+                                             header);
+   }
+
    if( full_decl->Declaration.Semantic ) {
       struct tgsi_declaration_semantic *ds;
 
@@ -249,6 +267,34 @@ tgsi_build_declaration_range(
    return declaration_range;
 }
 
+struct tgsi_declaration_dimension
+tgsi_default_declaration_dimension(void)
+{
+   struct tgsi_declaration_dimension dd;
+
+   dd.Index2D = 0;
+   dd.Padding = 0;
+
+   return dd;
+}
+
+struct tgsi_declaration_dimension
+tgsi_build_declaration_dimension(unsigned index_2d,
+                                 struct tgsi_declaration *declaration,
+                                 struct tgsi_header *header)
+{
+   struct tgsi_declaration_dimension dd;
+
+   assert(index_2d <= 0xFFFF);
+
+   dd = tgsi_default_declaration_dimension();
+   dd.Index2D = index_2d;
+
+   declaration_grow(declaration, header);
+
+   return dd;
+}
+
 struct tgsi_declaration_semantic
 tgsi_default_declaration_semantic( void )
 {
diff --git a/src/gallium/auxiliary/tgsi/tgsi_build.h b/src/gallium/auxiliary/tgsi/tgsi_build.h
index 9de2757fe4..ebee4ce5f6 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_build.h
+++ b/src/gallium/auxiliary/tgsi/tgsi_build.h
@@ -64,6 +64,7 @@ tgsi_build_declaration(
    unsigned file,
    unsigned usage_mask,
    unsigned interpolate,
+   unsigned dimension,
    unsigned semantic,
    unsigned centroid,
    unsigned invariant,
@@ -89,6 +90,14 @@ tgsi_build_declaration_range(
    struct tgsi_declaration *declaration,
    struct tgsi_header *header );
 
+struct tgsi_declaration_dimension
+tgsi_default_declaration_dimension(void);
+
+struct tgsi_declaration_dimension
+tgsi_build_declaration_dimension(unsigned index_2d,
+                                 struct tgsi_declaration *declaration,
+                                 struct tgsi_header *header);
+
 struct tgsi_declaration_semantic
 tgsi_default_declaration_semantic( void );
 
diff --git a/src/gallium/auxiliary/tgsi/tgsi_dump.c b/src/gallium/auxiliary/tgsi/tgsi_dump.c
index d7ff262f30..dd36555875 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_dump.c
+++ b/src/gallium/auxiliary/tgsi/tgsi_dump.c
@@ -159,7 +159,9 @@ static const char *property_names[] =
 {
    "GS_INPUT_PRIMITIVE",
    "GS_OUTPUT_PRIMITIVE",
-   "GS_MAX_OUTPUT_VERTICES"
+   "GS_MAX_OUTPUT_VERTICES",
+   "FS_COORD_ORIGIN",
+   "FS_COORD_PIXEL_CENTER"
 };
 
 static const char *primitive_names[] =
@@ -176,29 +178,18 @@ static const char *primitive_names[] =
    "POLYGON"
 };
 
-
-static void
-_dump_register_decl(
-   struct dump_ctx *ctx,
-   uint file,
-   int first,
-   int last )
+static const char *fs_coord_origin_names[] =
 {
-   ENM( file, file_names );
+   "UPPER_LEFT",
+   "LOWER_LEFT"
+};
 
-   /* all geometry shader inputs are two dimensional */
-   if (file == TGSI_FILE_INPUT &&
-       ctx->iter.processor.Processor == TGSI_PROCESSOR_GEOMETRY)
-      TXT("[]");
+static const char *fs_coord_pixel_center_names[] =
+{
+   "HALF_INTEGER",
+   "INTEGER"
+};
 
-   CHR( '[' );
-   SID( first );
-   if (first != last) {
-      TXT( ".." );
-      SID( last );
-   }
-   CHR( ']' );
-}
 
 static void
 _dump_register_dst(
@@ -219,8 +210,13 @@ _dump_register_src(
    struct dump_ctx *ctx,
    const struct tgsi_full_src_register *src )
 {
+   ENM(src->Register.File, file_names);
+   if (src->Register.Dimension) {
+      CHR('[');
+      SID(src->Dimension.Index);
+      CHR(']');
+   }
    if (src->Register.Indirect) {
-      ENM( src->Register.File, file_names );
       CHR( '[' );
       ENM( src->Indirect.File, file_names );
       CHR( '[' );
@@ -234,16 +230,10 @@ _dump_register_src(
       }
       CHR( ']' );
    } else {
-      ENM( src->Register.File, file_names );
       CHR( '[' );
       SID( src->Register.Index );
       CHR( ']' );
    }
-   if (src->Register.Dimension) {
-      CHR( '[' );
-      SID( src->Dimension.Index );
-      CHR( ']' );
-   }
 }
 
 static void
@@ -300,11 +290,28 @@ iter_declaration(
 
    TXT( "DCL " );
 
-   _dump_register_decl(
-      ctx,
-      decl->Declaration.File,
-      decl->Range.First,
-      decl->Range.Last );
+   ENM(decl->Declaration.File, file_names);
+
+   /* all geometry shader inputs are two dimensional */
+   if (decl->Declaration.File == TGSI_FILE_INPUT &&
+       iter->processor.Processor == TGSI_PROCESSOR_GEOMETRY) {
+      TXT("[]");
+   }
+
+   if (decl->Declaration.Dimension) {
+      CHR('[');
+      SID(decl->Dim.Index2D);
+      CHR(']');
+   }
+
+   CHR('[');
+   SID(decl->Range.First);
+   if (decl->Range.First != decl->Range.Last) {
+      TXT("..");
+      SID(decl->Range.Last);
+   }
+   CHR(']');
+
    _dump_writemask(
       ctx,
       decl->Declaration.UsageMask );
@@ -373,6 +380,12 @@ iter_property(
       case TGSI_PROPERTY_GS_OUTPUT_PRIM:
          ENM(prop->u[i].Data, primitive_names);
          break;
+      case TGSI_PROPERTY_FS_COORD_ORIGIN:
+         ENM(prop->u[i].Data, fs_coord_origin_names);
+         break;
+      case TGSI_PROPERTY_FS_COORD_PIXEL_CENTER:
+         ENM(prop->u[i].Data, fs_coord_pixel_center_names);
+         break;
       default:
          SID( prop->u[i].Data );
          break;
diff --git a/src/gallium/auxiliary/tgsi/tgsi_exec.c b/src/gallium/auxiliary/tgsi/tgsi_exec.c
index 118a638ab4..fbb9aa0e63 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_exec.c
+++ b/src/gallium/auxiliary/tgsi/tgsi_exec.c
@@ -953,107 +953,90 @@ micro_sub(
 }
 
 static void
-fetch_src_file_channel(
-   const struct tgsi_exec_machine *mach,
-   const uint file,
-   const uint swizzle,
-   const union tgsi_exec_channel *index,
-   union tgsi_exec_channel *chan )
-{
-   switch( swizzle ) {
-   case TGSI_SWIZZLE_X:
-   case TGSI_SWIZZLE_Y:
-   case TGSI_SWIZZLE_Z:
-   case TGSI_SWIZZLE_W:
-      switch( file ) {
-      case TGSI_FILE_CONSTANT:
-         assert(mach->Consts);
-         if (index->i[0] < 0)
-            chan->f[0] = 0.0f;
-         else
-            chan->f[0] = mach->Consts[index->i[0]][swizzle];
-         if (index->i[1] < 0)
-            chan->f[1] = 0.0f;
-         else
-            chan->f[1] = mach->Consts[index->i[1]][swizzle];
-         if (index->i[2] < 0)
-            chan->f[2] = 0.0f;
-         else
-            chan->f[2] = mach->Consts[index->i[2]][swizzle];
-         if (index->i[3] < 0)
-            chan->f[3] = 0.0f;
-         else
-            chan->f[3] = mach->Consts[index->i[3]][swizzle];
-         break;
+fetch_src_file_channel(const struct tgsi_exec_machine *mach,
+                       const uint file,
+                       const uint swizzle,
+                       const union tgsi_exec_channel *index,
+                       const union tgsi_exec_channel *index2D,
+                       union tgsi_exec_channel *chan)
+{
+   uint i;
 
-      case TGSI_FILE_INPUT:
-      case TGSI_FILE_SYSTEM_VALUE:
-         chan->u[0] = mach->Inputs[index->i[0]].xyzw[swizzle].u[0];
-         chan->u[1] = mach->Inputs[index->i[1]].xyzw[swizzle].u[1];
-         chan->u[2] = mach->Inputs[index->i[2]].xyzw[swizzle].u[2];
-         chan->u[3] = mach->Inputs[index->i[3]].xyzw[swizzle].u[3];
-         break;
+   switch (file) {
+   case TGSI_FILE_CONSTANT:
+      for (i = 0; i < QUAD_SIZE; i++) {
+         assert(index2D->i[i] >= 0 && index2D->i[i] < PIPE_MAX_CONSTANT_BUFFERS);
+         assert(mach->Consts[index2D->i[i]]);
 
-      case TGSI_FILE_TEMPORARY:
-         assert(index->i[0] < TGSI_EXEC_NUM_TEMPS);
-         chan->u[0] = mach->Temps[index->i[0]].xyzw[swizzle].u[0];
-         chan->u[1] = mach->Temps[index->i[1]].xyzw[swizzle].u[1];
-         chan->u[2] = mach->Temps[index->i[2]].xyzw[swizzle].u[2];
-         chan->u[3] = mach->Temps[index->i[3]].xyzw[swizzle].u[3];
-         break;
+         if (index->i[i] < 0) {
+            chan->u[i] = 0;
+         } else {
+            const uint *p = (const uint *)mach->Consts[index2D->i[i]];
 
-      case TGSI_FILE_IMMEDIATE:
-         assert( index->i[0] < (int) mach->ImmLimit );
-         chan->f[0] = mach->Imms[index->i[0]][swizzle];
-         assert( index->i[1] < (int) mach->ImmLimit );
-         chan->f[1] = mach->Imms[index->i[1]][swizzle];
-         assert( index->i[2] < (int) mach->ImmLimit );
-         chan->f[2] = mach->Imms[index->i[2]][swizzle];
-         assert( index->i[3] < (int) mach->ImmLimit );
-         chan->f[3] = mach->Imms[index->i[3]][swizzle];
-         break;
+            chan->u[i] = p[index->i[i] * 4 + swizzle];
+         }
+      }
+      break;
 
-      case TGSI_FILE_ADDRESS:
-         chan->u[0] = mach->Addrs[index->i[0]].xyzw[swizzle].u[0];
-         chan->u[1] = mach->Addrs[index->i[1]].xyzw[swizzle].u[1];
-         chan->u[2] = mach->Addrs[index->i[2]].xyzw[swizzle].u[2];
-         chan->u[3] = mach->Addrs[index->i[3]].xyzw[swizzle].u[3];
-         break;
+   case TGSI_FILE_INPUT:
+   case TGSI_FILE_SYSTEM_VALUE:
+      for (i = 0; i < QUAD_SIZE; i++) {
+         /* XXX: 2D indexing */
+         chan->u[i] = mach->Inputs[index2D->i[i] * TGSI_EXEC_MAX_INPUT_ATTRIBS + index->i[i]].xyzw[swizzle].u[i];
+      }
+      break;
 
-      case TGSI_FILE_PREDICATE:
-         assert(index->i[0] < TGSI_EXEC_NUM_PREDS);
-         assert(index->i[1] < TGSI_EXEC_NUM_PREDS);
-         assert(index->i[2] < TGSI_EXEC_NUM_PREDS);
-         assert(index->i[3] < TGSI_EXEC_NUM_PREDS);
-         chan->u[0] = mach->Predicates[0].xyzw[swizzle].u[0];
-         chan->u[1] = mach->Predicates[0].xyzw[swizzle].u[1];
-         chan->u[2] = mach->Predicates[0].xyzw[swizzle].u[2];
-         chan->u[3] = mach->Predicates[0].xyzw[swizzle].u[3];
-         break;
+   case TGSI_FILE_TEMPORARY:
+      for (i = 0; i < QUAD_SIZE; i++) {
+         assert(index->i[i] < TGSI_EXEC_NUM_TEMPS);
+         assert(index2D->i[i] == 0);
 
-      case TGSI_FILE_OUTPUT:
-         /* vertex/fragment output vars can be read too */
-         chan->u[0] = mach->Outputs[index->i[0]].xyzw[swizzle].u[0];
-         chan->u[1] = mach->Outputs[index->i[1]].xyzw[swizzle].u[1];
-         chan->u[2] = mach->Outputs[index->i[2]].xyzw[swizzle].u[2];
-         chan->u[3] = mach->Outputs[index->i[3]].xyzw[swizzle].u[3];
-         break;
+         chan->u[i] = mach->Temps[index->i[i]].xyzw[swizzle].u[i];
+      }
+      break;
 
-      default:
-         assert( 0 );
-         chan->u[0] = 0;
-         chan->u[1] = 0;
-         chan->u[2] = 0;
-         chan->u[3] = 0;
+   case TGSI_FILE_IMMEDIATE:
+      for (i = 0; i < QUAD_SIZE; i++) {
+         assert(index->i[i] >= 0 && index->i[i] < (int)mach->ImmLimit);
+         assert(index2D->i[i] == 0);
+
+         chan->f[i] = mach->Imms[index->i[i]][swizzle];
+      }
+      break;
+
+   case TGSI_FILE_ADDRESS:
+      for (i = 0; i < QUAD_SIZE; i++) {
+         assert(index->i[i] >= 0);
+         assert(index2D->i[i] == 0);
+
+         chan->u[i] = mach->Addrs[index->i[i]].xyzw[swizzle].u[i];
+      }
+      break;
+
+   case TGSI_FILE_PREDICATE:
+      for (i = 0; i < QUAD_SIZE; i++) {
+         assert(index->i[i] >= 0 && index->i[i] < TGSI_EXEC_NUM_PREDS);
+         assert(index2D->i[i] == 0);
+
+         chan->u[i] = mach->Predicates[0].xyzw[swizzle].u[i];
+      }
+      break;
+
+   case TGSI_FILE_OUTPUT:
+      /* vertex/fragment output vars can be read too */
+      for (i = 0; i < QUAD_SIZE; i++) {
+         assert(index->i[i] >= 0);
+         assert(index2D->i[i] == 0);
+
+         chan->u[i] = mach->Outputs[index->i[i]].xyzw[swizzle].u[i];
       }
       break;
 
    default:
-      assert( 0 );
-      chan->u[0] = 0;
-      chan->u[1] = 0;
-      chan->u[2] = 0;
-      chan->u[3] = 0;
+      assert(0);
+      for (i = 0; i < QUAD_SIZE; i++) {
+         chan->u[i] = 0;
+      }
    }
 }
 
@@ -1065,6 +1048,7 @@ fetch_source(const struct tgsi_exec_machine *mach,
              enum tgsi_exec_datatype src_datatype)
 {
    union tgsi_exec_channel index;
+   union tgsi_exec_channel index2D;
    uint swizzle;
 
    /* We start with a direct index into a register file.
@@ -1103,12 +1087,12 @@ fetch_source(const struct tgsi_exec_machine *mach,
 
       /* get current value of address register[swizzle] */
       swizzle = tgsi_util_get_src_register_swizzle( &reg->Indirect, CHAN_X );
-      fetch_src_file_channel(
-         mach,
-         reg->Indirect.File,
-         swizzle,
-         &index2,
-         &indir_index );
+      fetch_src_file_channel(mach,
+                             reg->Indirect.File,
+                             swizzle,
+                             &index2,
+                             &ZeroVec,
+                             &indir_index);
 
       /* add value of address register to the offset */
       index.i[0] += indir_index.i[0];
@@ -1129,44 +1113,22 @@ fetch_source(const struct tgsi_exec_machine *mach,
     * subscript to a register file. Effectively it means that
     * the register file is actually a 2D array of registers.
     *
-    *    file[1][3] == file[1*sizeof(file[1])+3],
+    *    file[3][1],
     *    where:
     *       [3] = Dimension.Index
     */
    if (reg->Register.Dimension) {
-      /* The size of the first-order array depends on the register file type.
-       * We need to multiply the index to the first array to get an effective,
-       * "flat" index that points to the beginning of the second-order array.
-       */
-      switch (reg->Register.File) {
-      case TGSI_FILE_INPUT:
-      case TGSI_FILE_SYSTEM_VALUE:
-         index.i[0] *= TGSI_EXEC_MAX_INPUT_ATTRIBS;
-         index.i[1] *= TGSI_EXEC_MAX_INPUT_ATTRIBS;
-         index.i[2] *= TGSI_EXEC_MAX_INPUT_ATTRIBS;
-         index.i[3] *= TGSI_EXEC_MAX_INPUT_ATTRIBS;
-         break;
-      case TGSI_FILE_CONSTANT:
-         index.i[0] *= TGSI_EXEC_MAX_CONST_BUFFER;
-         index.i[1] *= TGSI_EXEC_MAX_CONST_BUFFER;
-         index.i[2] *= TGSI_EXEC_MAX_CONST_BUFFER;
-         index.i[3] *= TGSI_EXEC_MAX_CONST_BUFFER;
-         break;
-      default:
-         assert( 0 );
-      }
-
-      index.i[0] += reg->Dimension.Index;
-      index.i[1] += reg->Dimension.Index;
-      index.i[2] += reg->Dimension.Index;
-      index.i[3] += reg->Dimension.Index;
+      index2D.i[0] =
+      index2D.i[1] =
+      index2D.i[2] =
+      index2D.i[3] = reg->Dimension.Index;
 
       /* Again, the second subscript index can be addressed indirectly
        * identically to the first one.
        * Nothing stops us from indirectly addressing the indirect register,
        * but there is no need for that, so we won't exercise it.
        *
-       *    file[1][ind[4].y+3],
+       *    file[ind[4].y+3][1],
        *    where:
        *       ind = DimIndirect.File
        *       [4] = DimIndirect.Index
@@ -1184,24 +1146,25 @@ fetch_source(const struct tgsi_exec_machine *mach,
          index2.i[3] = reg->DimIndirect.Index;
 
          swizzle = tgsi_util_get_src_register_swizzle( &reg->DimIndirect, CHAN_X );
-         fetch_src_file_channel(
-            mach,
-            reg->DimIndirect.File,
-            swizzle,
-            &index2,
-            &indir_index );
-
-         index.i[0] += indir_index.i[0];
-         index.i[1] += indir_index.i[1];
-         index.i[2] += indir_index.i[2];
-         index.i[3] += indir_index.i[3];
+         fetch_src_file_channel(mach,
+                                reg->DimIndirect.File,
+                                swizzle,
+                                &index2,
+                                &ZeroVec,
+                                &indir_index);
+
+         index2D.i[0] += indir_index.i[0];
+         index2D.i[1] += indir_index.i[1];
+         index2D.i[2] += indir_index.i[2];
+         index2D.i[3] += indir_index.i[3];
 
          /* for disabled execution channels, zero-out the index to
           * avoid using a potential garbage value.
           */
          for (i = 0; i < QUAD_SIZE; i++) {
-            if ((execmask & (1 << i)) == 0)
-               index.i[i] = 0;
+            if ((execmask & (1 << i)) == 0) {
+               index2D.i[i] = 0;
+            }
          }
       }
 
@@ -1209,15 +1172,20 @@ fetch_source(const struct tgsi_exec_machine *mach,
        * files, we would have to check whether Dimension is followed
        * by a dimension register and continue the saga.
        */
+   } else {
+      index2D.i[0] =
+      index2D.i[1] =
+      index2D.i[2] =
+      index2D.i[3] = 0;
    }
 
    swizzle = tgsi_util_get_full_src_register_swizzle( reg, chan_index );
-   fetch_src_file_channel(
-      mach,
-      reg->Register.File,
-      swizzle,
-      &index,
-      chan );
+   fetch_src_file_channel(mach,
+                          reg->Register.File,
+                          swizzle,
+                          &index,
+                          &index2D,
+                          chan);
 
    if (reg->Register.Absolute) {
       if (src_datatype == TGSI_EXEC_DATA_FLOAT) {
@@ -1280,12 +1248,12 @@ store_dest(struct tgsi_exec_machine *mach,
       swizzle = tgsi_util_get_src_register_swizzle( &reg->Indirect, CHAN_X );
 
       /* fetch values from the address/indirection register */
-      fetch_src_file_channel(
-         mach,
-         reg->Indirect.File,
-         swizzle,
-         &index,
-         &indir_index );
+      fetch_src_file_channel(mach,
+                             reg->Indirect.File,
+                             swizzle,
+                             &index,
+                             &ZeroVec,
+                             &indir_index);
 
       /* save indirection offset */
       offset = indir_index.i[0];
diff --git a/src/gallium/auxiliary/tgsi/tgsi_exec.h b/src/gallium/auxiliary/tgsi/tgsi_exec.h
index 59e3b445cc..a22873e4c2 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_exec.h
+++ b/src/gallium/auxiliary/tgsi/tgsi_exec.h
@@ -260,7 +260,7 @@ struct tgsi_exec_machine
    struct tgsi_sampler           **Samplers;
 
    unsigned                      ImmLimit;
-   const float                   (*Consts)[4];
+   const void *Consts[PIPE_MAX_CONSTANT_BUFFERS];
    const struct tgsi_token       *Tokens;   /**< Declarations, instructions */
    unsigned                      Processor; /**< TGSI_PROCESSOR_x */
 
diff --git a/src/gallium/auxiliary/tgsi/tgsi_parse.c b/src/gallium/auxiliary/tgsi/tgsi_parse.c
index 8c7062d850..fd37fc3079 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_parse.c
+++ b/src/gallium/auxiliary/tgsi/tgsi_parse.c
@@ -109,6 +109,10 @@ tgsi_parse_token(
 
       next_token( ctx, &decl->Range );
 
+      if (decl->Declaration.Dimension) {
+         next_token(ctx, &decl->Dim);
+      }
+
       if( decl->Declaration.Semantic ) {
          next_token( ctx, &decl->Semantic );
       }
diff --git a/src/gallium/auxiliary/tgsi/tgsi_parse.h b/src/gallium/auxiliary/tgsi/tgsi_parse.h
index 439a57269b..8150e3cd29 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_parse.h
+++ b/src/gallium/auxiliary/tgsi/tgsi_parse.h
@@ -58,6 +58,7 @@ struct tgsi_full_declaration
 {
    struct tgsi_declaration Declaration;
    struct tgsi_declaration_range Range;
+   struct tgsi_declaration_dimension Dim;
    struct tgsi_declaration_semantic Semantic;
 };
 
diff --git a/src/gallium/auxiliary/tgsi/tgsi_sanity.c b/src/gallium/auxiliary/tgsi/tgsi_sanity.c
index e1e4f97967..91e1b27da1 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_sanity.c
+++ b/src/gallium/auxiliary/tgsi/tgsi_sanity.c
@@ -335,10 +335,6 @@ iter_instruction(
          fill_scan_register1d(ind_reg,
                               inst->Src[i].Indirect.File,
                               inst->Src[i].Indirect.Index);
-         if (!(ind_reg->file == TGSI_FILE_ADDRESS || ind_reg->file == TGSI_FILE_LOOP) ||
-             ind_reg->indices[0] != 0) {
-            report_warning(ctx, "Indirect register neither ADDR[0] nor LOOP[0]");
-         }
          check_register_usage(
             ctx,
             ind_reg,
@@ -412,12 +408,16 @@ iter_declaration(
          uint vert;
          for (vert = 0; vert < ctx->implied_array_size; ++vert) {
             scan_register *reg = MALLOC(sizeof(scan_register));
-            fill_scan_register2d(reg, file, vert, i);
+            fill_scan_register2d(reg, file, i, vert);
             check_and_declare(ctx, reg);
          }
       } else {
          scan_register *reg = MALLOC(sizeof(scan_register));
-         fill_scan_register1d(reg, file, i);
+         if (decl->Declaration.Dimension) {
+            fill_scan_register2d(reg, file, i, decl->Dim.Index2D);
+         } else {
+            fill_scan_register1d(reg, file, i);
+         }
          check_and_declare(ctx, reg);
       }
    }
diff --git a/src/gallium/auxiliary/tgsi/tgsi_scan.c b/src/gallium/auxiliary/tgsi/tgsi_scan.c
index a6cc773003..b9be8dc0a3 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_scan.c
+++ b/src/gallium/auxiliary/tgsi/tgsi_scan.c
@@ -101,12 +101,10 @@ tgsi_scan_shader(const struct tgsi_token *tokens,
                       src->Register.File == TGSI_FILE_SYSTEM_VALUE) {
                      const int ind = src->Register.Index;
                      if (info->input_semantic_name[ind] == TGSI_SEMANTIC_FOG) {
-                        if (src->Register.SwizzleX == TGSI_SWIZZLE_X) {
-                           info->uses_fogcoord = TRUE;
-                        }
-                        else if (src->Register.SwizzleX == TGSI_SWIZZLE_Y) {
-                           info->uses_frontfacing = TRUE;
-                        }
+                        info->uses_fogcoord = TRUE;
+                     }
+                     else if (info->input_semantic_name[ind] == TGSI_SEMANTIC_FACE) {
+                        info->uses_frontfacing = TRUE;
                      }
                   }
                }
diff --git a/src/gallium/auxiliary/tgsi/tgsi_text.c b/src/gallium/auxiliary/tgsi/tgsi_text.c
index 9fcffeda36..96be353e26 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_text.c
+++ b/src/gallium/auxiliary/tgsi/tgsi_text.c
@@ -553,7 +553,7 @@ parse_register_dcl_bracket(
       report_error( ctx, "Expected literal unsigned integer" );
       return FALSE;
    }
-   bracket->first = (int) uindex;
+   bracket->first = uindex;
 
    eat_opt_white( &ctx->cur );
 
@@ -617,10 +617,12 @@ parse_register_dcl(
        * input primitive. so we want to declare just
        * the index relevant to the semantics which is in
        * the second bracket */
-      if (ctx->processor == TGSI_PROCESSOR_GEOMETRY) {
+      if (ctx->processor == TGSI_PROCESSOR_GEOMETRY && *file == TGSI_FILE_INPUT) {
          brackets[0] = brackets[1];
+         *num_brackets = 1;
+      } else {
+         *num_brackets = 2;
       }
-      *num_brackets = 2;
    }
 
    return TRUE;
@@ -738,6 +740,13 @@ parse_src_operand(
       return FALSE;
 
    src->Register.File = file;
+   if (parsed_opt_brackets) {
+      src->Register.Dimension = 1;
+      src->Dimension.Indirect = 0;
+      src->Dimension.Dimension = 0;
+      src->Dimension.Index = bracket[0].index;
+      bracket[0] = bracket[1];
+   }
    src->Register.Index = bracket[0].index;
    if (bracket[0].ind_file != TGSI_FILE_NULL) {
       src->Register.Indirect = 1;
@@ -748,12 +757,6 @@ parse_src_operand(
       src->Indirect.SwizzleZ = bracket[0].ind_comp;
       src->Indirect.SwizzleW = bracket[0].ind_comp;
    }
-   if (parsed_opt_brackets) {
-      src->Register.Dimension = 1;
-      src->Dimension.Indirect = 0;
-      src->Dimension.Dimension = 0;
-      src->Dimension.Index = bracket[1].index;
-   }
 
    /* Parse optional swizzle.
     */
@@ -933,7 +936,8 @@ static const char *semantic_names[TGSI_SEMANTIC_COUNT] =
    "NORMAL",
    "FACE",
    "EDGEFLAG",
-   "PRIM_ID"
+   "PRIM_ID",
+   "INSTANCEID"
 };
 
 static const char *interpolate_names[TGSI_INTERPOLATE_COUNT] =
@@ -968,8 +972,17 @@ static boolean parse_declaration( struct translate_ctx *ctx )
    decl = tgsi_default_full_declaration();
    decl.Declaration.File = file;
    decl.Declaration.UsageMask = writemask;
-   decl.Range.First = brackets[0].first;
-   decl.Range.Last = brackets[0].last;
+
+   if (num_brackets == 1) {
+      decl.Range.First = brackets[0].first;
+      decl.Range.Last = brackets[0].last;
+   } else {
+      decl.Range.First = brackets[1].first;
+      decl.Range.Last = brackets[1].last;
+
+      decl.Declaration.Dimension = 1;
+      decl.Dim.Index2D = brackets[0].first;
+   }
 
    cur = ctx->cur;
    eat_opt_white( &cur );
@@ -1116,7 +1129,9 @@ static const char *property_names[] =
 {
    "GS_INPUT_PRIMITIVE",
    "GS_OUTPUT_PRIMITIVE",
-   "GS_MAX_OUTPUT_VERTICES"
+   "GS_MAX_OUTPUT_VERTICES",
+   "FS_COORD_ORIGIN",
+   "FS_COORD_PIXEL_CENTER"
 };
 
 static const char *primitive_names[] =
@@ -1133,6 +1148,19 @@ static const char *primitive_names[] =
    "POLYGON"
 };
 
+static const char *fs_coord_origin_names[] =
+{
+   "UPPER_LEFT",
+   "LOWER_LEFT"
+};
+
+static const char *fs_coord_pixel_center_names[] =
+{
+   "HALF_INTEGER",
+   "INTEGER"
+};
+
+
 static boolean
 parse_primitive( const char **pcur, uint *primitive )
 {
@@ -1150,6 +1178,40 @@ parse_primitive( const char **pcur, uint *primitive )
    return FALSE;
 }
 
+static boolean
+parse_fs_coord_origin( const char **pcur, uint *fs_coord_origin )
+{
+   uint i;
+
+   for (i = 0; i < sizeof(fs_coord_origin_names) / sizeof(fs_coord_origin_names[0]); i++) {
+      const char *cur = *pcur;
+
+      if (str_match_no_case( &cur, fs_coord_origin_names[i])) {
+         *fs_coord_origin = i;
+         *pcur = cur;
+         return TRUE;
+      }
+   }
+   return FALSE;
+}
+
+static boolean
+parse_fs_coord_pixel_center( const char **pcur, uint *fs_coord_pixel_center )
+{
+   uint i;
+
+   for (i = 0; i < sizeof(fs_coord_pixel_center_names) / sizeof(fs_coord_pixel_center_names[0]); i++) {
+      const char *cur = *pcur;
+
+      if (str_match_no_case( &cur, fs_coord_pixel_center_names[i])) {
+         *fs_coord_pixel_center = i;
+         *pcur = cur;
+         return TRUE;
+      }
+   }
+   return FALSE;
+}
+
 
 static boolean parse_property( struct translate_ctx *ctx )
 {
@@ -1191,6 +1253,18 @@ static boolean parse_property( struct translate_ctx *ctx )
          ctx->implied_array_size = u_vertices_per_prim(values[0]);
       }
       break;
+   case TGSI_PROPERTY_FS_COORD_ORIGIN:
+      if (!parse_fs_coord_origin(&ctx->cur, &values[0] )) {
+         report_error( ctx, "Unknown coord origin as property: must be UPPER_LEFT or LOWER_LEFT!" );
+         return FALSE;
+      }
+      break;
+   case TGSI_PROPERTY_FS_COORD_PIXEL_CENTER:
+      if (!parse_fs_coord_pixel_center(&ctx->cur, &values[0] )) {
+         report_error( ctx, "Unknown coord pixel center as property: must be HALF_INTEGER or INTEGER!" );
+         return FALSE;
+      }
+      break;
    default:
       if (!parse_uint(&ctx->cur, &values[0] )) {
          report_error( ctx, "Expected unsigned integer as property!" );
diff --git a/src/gallium/auxiliary/tgsi/tgsi_ureg.c b/src/gallium/auxiliary/tgsi/tgsi_ureg.c
index 8bd6f68dcc..0ae46785bb 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_ureg.c
+++ b/src/gallium/auxiliary/tgsi/tgsi_ureg.c
@@ -44,6 +44,7 @@ union tgsi_any_token {
    struct tgsi_property_data prop_data;
    struct tgsi_declaration decl;
    struct tgsi_declaration_range decl_range;
+   struct tgsi_declaration_dimension decl_dim;
    struct tgsi_declaration_semantic decl_semantic;
    struct tgsi_immediate imm;
    union  tgsi_immediate_data imm_data;
@@ -75,6 +76,14 @@ struct ureg_tokens {
 #define UREG_MAX_LOOP 1
 #define UREG_MAX_PRED 1
 
+struct const_decl {
+   struct {
+      unsigned first;
+      unsigned last;
+   } constant_range[UREG_MAX_CONSTANT_RANGE];
+   unsigned nr_constant_ranges;
+};
+
 #define DOMAIN_DECL 0
 #define DOMAIN_INSN 1
 
@@ -127,13 +136,14 @@ struct ureg_program
    unsigned temps_active[UREG_MAX_TEMP / 32];
    unsigned nr_temps;
 
-   struct {
-      unsigned first;
-      unsigned last;
-   } constant_range[UREG_MAX_CONSTANT_RANGE];
-   unsigned nr_constant_ranges;
+   struct const_decl const_decls;
+   struct const_decl const_decls2D[PIPE_MAX_CONSTANT_BUFFERS];
 
    unsigned property_gs_input_prim;
+   unsigned property_gs_output_prim;
+   unsigned property_gs_max_vertices;
+   unsigned char property_fs_coord_origin; /* = TGSI_FS_COORD_ORIGIN_* */
+   unsigned char property_fs_coord_pixel_center; /* = TGSI_FS_COORD_PIXEL_CENTER_* */
 
    unsigned nr_addrs;
    unsigned nr_preds;
@@ -235,36 +245,40 @@ ureg_dst_register( unsigned file,
    return dst;
 }
 
-static INLINE struct ureg_src 
-ureg_src_register( unsigned file,
-                   unsigned index )
+
+void
+ureg_property_gs_input_prim(struct ureg_program *ureg,
+                            unsigned input_prim)
 {
-   struct ureg_src src;
-
-   src.File     = file;
-   src.SwizzleX = TGSI_SWIZZLE_X;
-   src.SwizzleY = TGSI_SWIZZLE_Y;
-   src.SwizzleZ = TGSI_SWIZZLE_Z;
-   src.SwizzleW = TGSI_SWIZZLE_W;
-   src.Indirect = 0;
-   src.IndirectIndex = 0;
-   src.IndirectSwizzle = 0;
-   src.Absolute = 0;
-   src.Index    = index;
-   src.Negate   = 0;
-   src.Dimension = 0;
-   src.DimensionIndex = 0;
-
-   return src;
+   ureg->property_gs_input_prim = input_prim;
 }
 
+void
+ureg_property_gs_output_prim(struct ureg_program *ureg,
+                             unsigned output_prim)
+{
+   ureg->property_gs_output_prim = output_prim;
+}
 
+void
+ureg_property_gs_max_vertices(struct ureg_program *ureg,
+                              unsigned max_vertices)
+{
+   ureg->property_gs_max_vertices = max_vertices;
+}
 
 void
-ureg_property_gs_input_prim(struct ureg_program *ureg,
-                            unsigned gs_input_prim)
+ureg_property_fs_coord_origin(struct ureg_program *ureg,
+                            unsigned fs_coord_origin)
+{
+   ureg->property_fs_coord_origin = fs_coord_origin;
+}
+
+void
+ureg_property_fs_coord_pixel_center(struct ureg_program *ureg,
+                            unsigned fs_coord_pixel_center)
 {
-   ureg->property_gs_input_prim = gs_input_prim;
+   ureg->property_fs_coord_pixel_center = fs_coord_pixel_center;
 }
 
 
@@ -374,62 +388,92 @@ out:
 /* Returns a new constant register.  Keep track of which have been
  * referred to so that we can emit decls later.
  *
+ * Constant operands declared with this function must be addressed
+ * with a two-dimensional index.
+ *
  * There is nothing in this code to bind this constant to any tracked
  * value or manage any constant_buffer contents -- that's the
  * resposibility of the calling code.
  */
-struct ureg_src ureg_DECL_constant(struct ureg_program *ureg, 
-                                   unsigned index )
+void
+ureg_DECL_constant2D(struct ureg_program *ureg,
+                     unsigned first,
+                     unsigned last,
+                     unsigned index2D)
+{
+   struct const_decl *decl = &ureg->const_decls2D[index2D];
+
+   assert(index2D < PIPE_MAX_CONSTANT_BUFFERS);
+
+   if (decl->nr_constant_ranges < UREG_MAX_CONSTANT_RANGE) {
+      uint i = decl->nr_constant_ranges++;
+
+      decl->constant_range[i].first = first;
+      decl->constant_range[i].last = last;
+   }
+}
+
+
+/* A one-dimensional, depricated version of ureg_DECL_constant2D().
+ *
+ * Constant operands declared with this function must be addressed
+ * with a one-dimensional index.
+ */
+struct ureg_src
+ureg_DECL_constant(struct ureg_program *ureg,
+                   unsigned index)
 {
+   struct const_decl *decl = &ureg->const_decls;
    unsigned minconst = index, maxconst = index;
    unsigned i;
 
    /* Inside existing range?
     */
-   for (i = 0; i < ureg->nr_constant_ranges; i++) {
-      if (ureg->constant_range[i].first <= index &&
-          ureg->constant_range[i].last >= index)
+   for (i = 0; i < decl->nr_constant_ranges; i++) {
+      if (decl->constant_range[i].first <= index &&
+          decl->constant_range[i].last >= index) {
          goto out;
+      }
    }
 
    /* Extend existing range?
     */
-   for (i = 0; i < ureg->nr_constant_ranges; i++) {
-      if (ureg->constant_range[i].last == index - 1) {
-         ureg->constant_range[i].last = index;
+   for (i = 0; i < decl->nr_constant_ranges; i++) {
+      if (decl->constant_range[i].last == index - 1) {
+         decl->constant_range[i].last = index;
          goto out;
       }
 
-      if (ureg->constant_range[i].first == index + 1) {
-         ureg->constant_range[i].first = index;
+      if (decl->constant_range[i].first == index + 1) {
+         decl->constant_range[i].first = index;
          goto out;
       }
 
-      minconst = MIN2(minconst, ureg->constant_range[i].first);
-      maxconst = MAX2(maxconst, ureg->constant_range[i].last);
+      minconst = MIN2(minconst, decl->constant_range[i].first);
+      maxconst = MAX2(maxconst, decl->constant_range[i].last);
    }
 
    /* Create new range?
     */
-   if (ureg->nr_constant_ranges < UREG_MAX_CONSTANT_RANGE) {
-      i = ureg->nr_constant_ranges++;
-      ureg->constant_range[i].first = index;
-      ureg->constant_range[i].last = index;
+   if (decl->nr_constant_ranges < UREG_MAX_CONSTANT_RANGE) {
+      i = decl->nr_constant_ranges++;
+      decl->constant_range[i].first = index;
+      decl->constant_range[i].last = index;
       goto out;
    }
 
    /* Collapse all ranges down to one:
     */
    i = 0;
-   ureg->constant_range[0].first = minconst;
-   ureg->constant_range[0].last = maxconst;
-   ureg->nr_constant_ranges = 1;
+   decl->constant_range[0].first = minconst;
+   decl->constant_range[0].last = maxconst;
+   decl->nr_constant_ranges = 1;
 
 out:
-   assert(i < ureg->nr_constant_ranges);
-   assert(ureg->constant_range[i].first <= index);
-   assert(ureg->constant_range[i].last >= index);
-   return ureg_src_register( TGSI_FILE_CONSTANT, index );
+   assert(i < decl->nr_constant_ranges);
+   assert(decl->constant_range[i].first <= index);
+   assert(decl->constant_range[i].last >= index);
+   return ureg_src_register(TGSI_FILE_CONSTANT, index);
 }
 
 
@@ -578,7 +622,7 @@ decl_immediate( struct ureg_program *ureg,
                 unsigned type )
 {
    unsigned i, j;
-   unsigned swizzle;
+   unsigned swizzle = 0;
 
    /* Could do a first pass where we examine all existing immediates
     * without expanding.
@@ -656,6 +700,35 @@ ureg_DECL_immediate_uint( struct ureg_program *ureg,
 
 
 struct ureg_src
+ureg_DECL_immediate_block_uint( struct ureg_program *ureg,
+                                const unsigned *v,
+                                unsigned nr )
+{
+   uint index;
+   uint i;
+
+   if (ureg->nr_immediates + (nr + 3) / 4 > UREG_MAX_IMMEDIATE) {
+      set_bad(ureg);
+      return ureg_src_register(TGSI_FILE_IMMEDIATE, 0);
+   }
+
+   index = ureg->nr_immediates;
+   ureg->nr_immediates += (nr + 3) / 4;
+
+   for (i = index; i < ureg->nr_immediates; i++) {
+      ureg->immediate[i].type = TGSI_IMM_UINT32;
+      ureg->immediate[i].nr = nr > 4 ? 4 : nr;
+      memcpy(ureg->immediate[i].value.u,
+             &v[(i - index) * 4],
+             ureg->immediate[i].nr * sizeof(uint));
+      nr -= 4;
+   }
+
+   return ureg_src_register(TGSI_FILE_IMMEDIATE, index);
+}
+
+
+struct ureg_src
 ureg_DECL_immediate_int( struct ureg_program *ureg,
                          const int *v,
                          unsigned nr )
@@ -691,7 +764,7 @@ ureg_emit_src( struct ureg_program *ureg,
    if (src.Indirect) {
       out[0].src.Indirect = 1;
       out[n].value = 0;
-      out[n].src.File = TGSI_FILE_ADDRESS;
+      out[n].src.File = src.IndirectFile;
       out[n].src.SwizzleX = src.IndirectSwizzle;
       out[n].src.SwizzleY = src.IndirectSwizzle;
       out[n].src.SwizzleZ = src.IndirectSwizzle;
@@ -1058,6 +1131,31 @@ static void emit_decl_range( struct ureg_program *ureg,
 }
 
 static void
+emit_decl_range2D(struct ureg_program *ureg,
+                  unsigned file,
+                  unsigned first,
+                  unsigned last,
+                  unsigned index2D)
+{
+   union tgsi_any_token *out = get_tokens(ureg, DOMAIN_DECL, 3);
+
+   out[0].value = 0;
+   out[0].decl.Type = TGSI_TOKEN_TYPE_DECLARATION;
+   out[0].decl.NrTokens = 3;
+   out[0].decl.File = file;
+   out[0].decl.UsageMask = 0xf;
+   out[0].decl.Interpolate = TGSI_INTERPOLATE_CONSTANT;
+   out[0].decl.Dimension = 1;
+
+   out[1].value = 0;
+   out[1].decl_range.First = first;
+   out[1].decl_range.Last = last;
+
+   out[2].value = 0;
+   out[2].decl_dim.Index2D = index2D;
+}
+
+static void
 emit_immediate( struct ureg_program *ureg,
                 const unsigned *v,
                 unsigned type )
@@ -1104,6 +1202,38 @@ static void emit_decls( struct ureg_program *ureg )
                     ureg->property_gs_input_prim);
    }
 
+   if (ureg->property_gs_output_prim != ~0) {
+      assert(ureg->processor == TGSI_PROCESSOR_GEOMETRY);
+
+      emit_property(ureg,
+                    TGSI_PROPERTY_GS_OUTPUT_PRIM,
+                    ureg->property_gs_output_prim);
+   }
+
+   if (ureg->property_gs_max_vertices != ~0) {
+      assert(ureg->processor == TGSI_PROCESSOR_GEOMETRY);
+
+      emit_property(ureg,
+                    TGSI_PROPERTY_GS_MAX_VERTICES,
+                    ureg->property_gs_max_vertices);
+   }
+
+   if (ureg->property_fs_coord_origin) {
+      assert(ureg->processor == TGSI_PROCESSOR_FRAGMENT);
+
+      emit_property(ureg,
+                    TGSI_PROPERTY_FS_COORD_ORIGIN,
+                    ureg->property_fs_coord_origin);
+   }
+
+   if (ureg->property_fs_coord_pixel_center) {
+      assert(ureg->processor == TGSI_PROCESSOR_FRAGMENT);
+
+      emit_property(ureg,
+                    TGSI_PROPERTY_FS_COORD_PIXEL_CENTER,
+                    ureg->property_fs_coord_pixel_center);
+   }
+
    if (ureg->processor == TGSI_PROCESSOR_VERTEX) {
       for (i = 0; i < UREG_MAX_INPUT; i++) {
          if (ureg->vs_inputs[i/32] & (1 << (i%32))) {
@@ -1152,13 +1282,29 @@ static void emit_decls( struct ureg_program *ureg )
                        ureg->sampler[i].Index, 1 );
    }
 
-   if (ureg->nr_constant_ranges) {
-      for (i = 0; i < ureg->nr_constant_ranges; i++)
-         emit_decl_range( ureg,
-                          TGSI_FILE_CONSTANT,
-                          ureg->constant_range[i].first, 
-                          (ureg->constant_range[i].last + 1 -
-                           ureg->constant_range[i].first) );
+   if (ureg->const_decls.nr_constant_ranges) {
+      for (i = 0; i < ureg->const_decls.nr_constant_ranges; i++) {
+         emit_decl_range(ureg,
+                         TGSI_FILE_CONSTANT,
+                         ureg->const_decls.constant_range[i].first,
+                         ureg->const_decls.constant_range[i].last - ureg->const_decls.constant_range[i].first + 1);
+      }
+   }
+
+   for (i = 0; i < PIPE_MAX_CONSTANT_BUFFERS; i++) {
+      struct const_decl *decl = &ureg->const_decls2D[i];
+
+      if (decl->nr_constant_ranges) {
+         uint j;
+
+         for (j = 0; j < decl->nr_constant_ranges; j++) {
+            emit_decl_range2D(ureg,
+                              TGSI_FILE_CONSTANT,
+                              decl->constant_range[j].first,
+                              decl->constant_range[j].last,
+                              i);
+         }
+      }
    }
 
    if (ureg->nr_temps) {
@@ -1314,6 +1460,8 @@ struct ureg_program *ureg_create( unsigned processor )
 
    ureg->processor = processor;
    ureg->property_gs_input_prim = ~0;
+   ureg->property_gs_output_prim = ~0;
+   ureg->property_gs_max_vertices = ~0;
    return ureg;
 }
 
diff --git a/src/gallium/auxiliary/tgsi/tgsi_ureg.h b/src/gallium/auxiliary/tgsi/tgsi_ureg.h
index 03eaf24854..e25f35c6dc 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_ureg.h
+++ b/src/gallium/auxiliary/tgsi/tgsi_ureg.h
@@ -52,9 +52,10 @@ struct ureg_src
    unsigned Absolute    : 1;  /* BOOL */
    unsigned Negate      : 1;  /* BOOL */
    int      Index       : 16; /* SINT */
+   unsigned IndirectFile    : 4;  /* TGSI_FILE_ */
    int      IndirectIndex   : 16; /* SINT */
-   int      IndirectSwizzle : 2;  /* TGSI_SWIZZLE_ */
-   int      DimensionIndex : 16;  /* SINT */
+   unsigned IndirectSwizzle : 2;  /* TGSI_SWIZZLE_ */
+   int      DimensionIndex  : 16; /* SINT */
 };
 
 /* Very similar to a tgsi_dst_register, removing unsupported fields
@@ -125,8 +126,23 @@ ureg_create_shader_and_destroy( struct ureg_program *p,
 
 void
 ureg_property_gs_input_prim(struct ureg_program *ureg,
-                            unsigned gs_input_prim);
+                            unsigned input_prim);
+
+void
+ureg_property_gs_output_prim(struct ureg_program *ureg,
+                             unsigned output_prim);
 
+void
+ureg_property_gs_max_vertices(struct ureg_program *ureg,
+                              unsigned max_vertices);
+
+void
+ureg_property_fs_coord_origin(struct ureg_program *ureg,
+                            unsigned fs_coord_origin);
+
+void
+ureg_property_fs_coord_pixel_center(struct ureg_program *ureg,
+                            unsigned fs_coord_pixel_center);
 
 /***********************************************************************
  * Build shader declarations:
@@ -168,10 +184,21 @@ ureg_DECL_immediate_uint( struct ureg_program *,
                           unsigned nr );
 
 struct ureg_src
+ureg_DECL_immediate_block_uint( struct ureg_program *,
+                                const unsigned *v,
+                                unsigned nr );
+
+struct ureg_src
 ureg_DECL_immediate_int( struct ureg_program *,
                          const int *v,
                          unsigned nr );
 
+void
+ureg_DECL_constant2D(struct ureg_program *ureg,
+                     unsigned first,
+                     unsigned last,
+                     unsigned index2D);
+
 struct ureg_src
 ureg_DECL_constant( struct ureg_program *,
                     unsigned index );
@@ -768,8 +795,9 @@ static INLINE struct ureg_src
 ureg_src_indirect( struct ureg_src reg, struct ureg_src addr )
 {
    assert(reg.File != TGSI_FILE_NULL);
-   assert(addr.File == TGSI_FILE_ADDRESS);
+   assert(addr.File == TGSI_FILE_ADDRESS || addr.File == TGSI_FILE_TEMPORARY);
    reg.Indirect = 1;
+   reg.IndirectFile = addr.File;
    reg.IndirectIndex = addr.Index;
    reg.IndirectSwizzle = addr.SwizzleX;
    return reg;
@@ -789,6 +817,8 @@ ureg_dst( struct ureg_src src )
 {
    struct ureg_dst dst;
 
+   assert(!src.Indirect || src.IndirectFile == TGSI_FILE_ADDRESS);
+
    dst.File      = src.File;
    dst.WriteMask = TGSI_WRITEMASK_XYZW;
    dst.Indirect  = src.Indirect;
@@ -807,6 +837,30 @@ ureg_dst( struct ureg_src src )
 }
 
 static INLINE struct ureg_src
+ureg_src_register(unsigned file,
+                  unsigned index)
+{
+   struct ureg_src src;
+
+   src.File = file;
+   src.SwizzleX = TGSI_SWIZZLE_X;
+   src.SwizzleY = TGSI_SWIZZLE_Y;
+   src.SwizzleZ = TGSI_SWIZZLE_Z;
+   src.SwizzleW = TGSI_SWIZZLE_W;
+   src.Indirect = 0;
+   src.IndirectFile = TGSI_FILE_NULL;
+   src.IndirectIndex = 0;
+   src.IndirectSwizzle = 0;
+   src.Absolute = 0;
+   src.Index = index;
+   src.Negate = 0;
+   src.Dimension = 0;
+   src.DimensionIndex = 0;
+
+   return src;
+}
+
+static INLINE struct ureg_src
 ureg_src( struct ureg_dst dst )
 {
    struct ureg_src src;
@@ -817,6 +871,7 @@ ureg_src( struct ureg_dst dst )
    src.SwizzleZ  = TGSI_SWIZZLE_Z;
    src.SwizzleW  = TGSI_SWIZZLE_W;
    src.Indirect  = dst.Indirect;
+   src.IndirectFile = TGSI_FILE_ADDRESS;
    src.IndirectIndex = dst.IndirectIndex;
    src.IndirectSwizzle = dst.IndirectSwizzle;
    src.Absolute  = 0;
@@ -863,6 +918,7 @@ ureg_src_undef( void )
    src.SwizzleZ  = 0;
    src.SwizzleW  = 0;
    src.Indirect  = 0;
+   src.IndirectFile = TGSI_FILE_NULL;
    src.IndirectIndex = 0;
    src.IndirectSwizzle = 0;
    src.Absolute  = 0;
diff --git a/src/gallium/auxiliary/tgsi/tgsi_util.c b/src/gallium/auxiliary/tgsi/tgsi_util.c
index f4ca9e21ed..0a7e4105a8 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_util.c
+++ b/src/gallium/auxiliary/tgsi/tgsi_util.c
@@ -28,7 +28,6 @@
 #include "util/u_debug.h"
 #include "pipe/p_shader_tokens.h"
 #include "tgsi_parse.h"
-#include "tgsi_build.h"
 #include "tgsi_util.h"
 
 union pointer_hack
diff --git a/src/gallium/auxiliary/util/u_blit.c b/src/gallium/auxiliary/util/u_blit.c
index 3f74e2aa8b..eb63bec7b5 100644
--- a/src/gallium/auxiliary/util/u_blit.c
+++ b/src/gallium/auxiliary/util/u_blit.c
@@ -92,7 +92,7 @@ util_create_blit(struct pipe_context *pipe, struct cso_context *cso)
 
    /* disabled blending/masking */
    memset(&ctx->blend, 0, sizeof(ctx->blend));
-   ctx->blend.colormask = PIPE_MASK_RGBA;
+   ctx->blend.rt[0].colormask = PIPE_MASK_RGBA;
 
    /* no-op depth/stencil/alpha */
    memset(&ctx->depthstencil, 0, sizeof(ctx->depthstencil));
@@ -226,8 +226,8 @@ setup_vertex_data_tex(struct blit_state *ctx,
 
    offset = get_next_slot( ctx );
 
-   pipe_buffer_write(ctx->pipe->screen, ctx->vbuf,
-                     offset, sizeof(ctx->vertices), ctx->vertices);
+   pipe_buffer_write_nooverlap(ctx->pipe->screen, ctx->vbuf,
+                               offset, sizeof(ctx->vertices), ctx->vertices);
 
    return offset;
 }
@@ -262,6 +262,10 @@ regions_overlap(int srcX0, int srcY0,
  * Copy pixel block from src surface to dst surface.
  * Overlapping regions are acceptable.
  * Flipping and stretching are supported.
+ * \param filter  one of PIPE_TEX_MIPFILTER_NEAREST/LINEAR
+ * \param writemask  controls which channels in the dest surface are sourced
+ *                   from the src surface.  Disabled channels are sourced
+ *                   from (0,0,0,1).
  * XXX what about clipping???
  * XXX need some control over blitting Z and/or stencil.
  */
diff --git a/src/gallium/auxiliary/util/u_blitter.c b/src/gallium/auxiliary/util/u_blitter.c
index 249a0375fc..935e11c5d8 100644
--- a/src/gallium/auxiliary/util/u_blitter.c
+++ b/src/gallium/auxiliary/util/u_blitter.c
@@ -125,7 +125,7 @@ struct blitter_context *util_blitter_create(struct pipe_context *pipe)
    memset(&blend, 0, sizeof(blend));
    ctx->blend_keep_color = pipe->create_blend_state(pipe, &blend);
 
-   blend.colormask = PIPE_MASK_RGBA;
+   blend.rt[0].colormask = PIPE_MASK_RGBA;
    ctx->blend_write_color = pipe->create_blend_state(pipe, &blend);
 
    /* depth stencil alpha state objects */
diff --git a/src/gallium/auxiliary/util/u_gen_mipmap.c b/src/gallium/auxiliary/util/u_gen_mipmap.c
index 76023794dc..8611231ed7 100644
--- a/src/gallium/auxiliary/util/u_gen_mipmap.c
+++ b/src/gallium/auxiliary/util/u_gen_mipmap.c
@@ -1287,7 +1287,7 @@ util_create_gen_mipmap(struct pipe_context *pipe,
 
    /* disabled blending/masking */
    memset(&ctx->blend, 0, sizeof(ctx->blend));
-   ctx->blend.colormask = PIPE_MASK_RGBA;
+   ctx->blend.rt[0].colormask = PIPE_MASK_RGBA;
 
    /* no-op depth/stencil/alpha */
    memset(&ctx->depthstencil, 0, sizeof(ctx->depthstencil));
@@ -1411,8 +1411,8 @@ set_vertex_data(struct gen_mipmap_state *ctx,
 
    offset = get_next_slot( ctx );
 
-   pipe_buffer_write(ctx->pipe->screen, ctx->vbuf,
-                     offset, sizeof(ctx->vertices), ctx->vertices);
+   pipe_buffer_write_nooverlap(ctx->pipe->screen, ctx->vbuf,
+                               offset, sizeof(ctx->vertices), ctx->vertices);
 
    return offset;
 }
diff --git a/src/gallium/auxiliary/util/u_surface.c b/src/gallium/auxiliary/util/u_surface.c
index f66376ad75..70de140ec9 100644
--- a/src/gallium/auxiliary/util/u_surface.c
+++ b/src/gallium/auxiliary/util/u_surface.c
@@ -37,7 +37,6 @@
 #include "pipe/p_defines.h"
 
 #include "util/u_memory.h"
-#include "util/u_format.h"
 #include "util/u_surface.h"
 
 
diff --git a/src/gallium/auxiliary/util/u_tile.c b/src/gallium/auxiliary/util/u_tile.c
index f9936eb1cb..c25e1e52e9 100644
--- a/src/gallium/auxiliary/util/u_tile.c
+++ b/src/gallium/auxiliary/util/u_tile.c
@@ -390,7 +390,7 @@ a4r4g4b4_put_tile_rgba(ushort *dst,
          g >>= 4;
          b >>= 4;
          a >>= 4;
-         *dst++ = (a << 12) | (r << 16) | (g << 4) | b;
+         *dst++ = (a << 12) | (r << 8) | (g << 4) | b;
       }
       p += src_stride;
    }
diff --git a/src/gallium/auxiliary/util/u_upload_mgr.c b/src/gallium/auxiliary/util/u_upload_mgr.c
index 975ee89c45..55a65375c8 100644
--- a/src/gallium/auxiliary/util/u_upload_mgr.c
+++ b/src/gallium/auxiliary/util/u_upload_mgr.c
@@ -85,7 +85,9 @@ my_buffer_write(struct pipe_screen *screen,
 
    map = pipe_buffer_map_range(screen, buf, offset, size, 
                                PIPE_BUFFER_USAGE_CPU_WRITE |
-                               PIPE_BUFFER_USAGE_FLUSH_EXPLICIT);
+                               PIPE_BUFFER_USAGE_FLUSH_EXPLICIT |
+                               PIPE_BUFFER_USAGE_DISCARD |
+                               PIPE_BUFFER_USAGE_UNSYNCHRONIZED);
    if (map == NULL) 
       return PIPE_ERROR_OUT_OF_MEMORY;
 
diff --git a/src/gallium/docs/source/context.rst b/src/gallium/docs/source/context.rst
index d394f5b4f1..a7669575b9 100644
--- a/src/gallium/docs/source/context.rst
+++ b/src/gallium/docs/source/context.rst
@@ -54,7 +54,10 @@ objects. They all follow simple, one-method binding calls, e.g.
 * ``set_blend_color``
 * ``set_clip_state``
 * ``set_polygon_stipple``
-* ``set_scissor_state``
+* ``set_scissor_state`` sets the bounds for the scissor test, which culls
+  pixels before blending to render targets. If the :ref:`Rasterizer` does
+  not have the scissor test enabled, then the scissor bounds never need to
+  be set since they will not be used.
 * ``set_viewport_state``
 * ``set_vertex_elements``
 
@@ -145,9 +148,51 @@ draws.  Queries may be nested, though no state tracker currently
 exercises this.  
 
 Queries can be created with ``create_query`` and deleted with
-``destroy_query``. To enable a query, use ``begin_query``, and when finished,
-use ``end_query`` to stop the query. Finally, ``get_query_result`` is used
-to retrieve the results.
+``destroy_query``. To start a query, use ``begin_query``, and when finished,
+use ``end_query`` to end the query.
+
+``get_query_result`` is used to retrieve the results of a query.  If
+the ``wait`` parameter is TRUE, then the ``get_query_result`` call
+will block until the results of the query are ready (and TRUE will be
+returned).  Otherwise, if the ``wait`` parameter is FALSE, the call
+will not block and the return value will be TRUE if the query has
+completed or FALSE otherwise.
+
+A common type of query is the occlusion query which counts the number of
+fragments/pixels which are written to the framebuffer (and not culled by
+Z/stencil/alpha testing or shader KILL instructions).
+
+
+Conditional Rendering
+^^^^^^^^^^^^^^^^^^^^^
+
+A drawing command can be skipped depending on the outcome of a query
+(typically an occlusion query).  The ``render_condition`` function specifies
+the query which should be checked prior to rendering anything.
+
+If ``render_condition`` is called with ``query`` = NULL, conditional
+rendering is disabled and drawing takes place normally.
+
+If ``render_condition`` is called with a non-null ``query`` subsequent
+drawing commands will be predicated on the outcome of the query.  If
+the query result is zero subsequent drawing commands will be skipped.
+
+If ``mode`` is PIPE_RENDER_COND_WAIT the driver will wait for the
+query to complete before deciding whether to render.
+
+If ``mode`` is PIPE_RENDER_COND_NO_WAIT and the query has not yet
+completed, the drawing command will be executed normally.  If the query
+has completed, drawing will be predicated on the outcome of the query.
+
+If ``mode`` is PIPE_RENDER_COND_BY_REGION_WAIT or
+PIPE_RENDER_COND_BY_REGION_NO_WAIT rendering will be predicated as above
+for the non-REGION modes but in the case that an occulusion query returns
+a non-zero result, regions which were occluded may be ommitted by subsequent
+drawing commands.  This can result in better performance with some GPUs.
+Normally, if the occlusion query returned a non-zero result subsequent
+drawing happens normally so fragments may be generated, shaded and
+processed even where they're known to be obscured.
+
 
 Flushing
 ^^^^^^^^
diff --git a/src/gallium/docs/source/cso/blend.rst b/src/gallium/docs/source/cso/blend.rst
index fd9e4a1e2d..55c0f32885 100644
--- a/src/gallium/docs/source/cso/blend.rst
+++ b/src/gallium/docs/source/cso/blend.rst
@@ -6,9 +6,50 @@ Blend
 This state controls blending of the final fragments into the target rendering
 buffers.
 
-XXX it is unresolved what behavior should result if blend_enable is off.
+Blend Factors
+-------------
+
+The blend factors largely follow the same pattern as their counterparts
+in other modern and legacy drawing APIs.
+
+XXX blurb about dual-source blends
 
 Members
 -------
 
-XXX undocumented members
+independent_blend_enable
+   If enabled, blend state is different for each render target, and
+   for each render target set in the respective member of the rt array.
+   If disabled, blend state is the same for all render targets, and only
+   the first member of the rt array contains valid data.
+logicop_enable
+   Enables logic ops. Cannot be enabled at the same time as blending, and
+   is always the same for all render targets.
+logicop_func
+   The logic operation to use if logic ops are enabled. One of PIPE_LOGICOP.
+dither
+   Whether dithering is enabled.
+rt
+   Contains the per-rendertarget blend state.
+
+Per-rendertarget Members
+------------------------
+
+blend_enable
+   If blending is enabled, perform a blend calculation according to blend
+   functions and source/destination factors. Otherwise, the incoming fragment
+   color gets passed unmodified (but colormask still applies).
+rgb_func
+   The blend function to use for rgb channels. One of PIPE_BLEND.
+rgb_src_factor
+   The blend source factor to use for rgb channels. One of PIPE_BLENDFACTOR.
+rgb_dst_factor
+   The blend destination factor to use for rgb channels. One of PIPE_BLENDFACTOR.
+alpha_func
+   The blend function to use for the alpha channel. One of PIPE_BLEND.
+alpha_src_factor
+   The blend source factor to use for the alpha channel. One of PIPE_BLENDFACTOR.
+alpha_dst_factor
+   The blend destination factor to use for alpha channel. One of PIPE_BLENDFACTOR.
+colormask
+   Bitmask of which channels to write. Combination of PIPE_MASK bits.
diff --git a/src/gallium/docs/source/cso/rasterizer.rst b/src/gallium/docs/source/cso/rasterizer.rst
index 4d8e1708e7..bfa4a1170a 100644
--- a/src/gallium/docs/source/cso/rasterizer.rst
+++ b/src/gallium/docs/source/cso/rasterizer.rst
@@ -7,32 +7,69 @@ The rasterizer state controls the rendering of points, lines and triangles.
 Attributes include polygon culling state, line width, line stipple,
 multisample state, scissoring and flat/smooth shading.
 
-
 Members
 -------
 
+bypass_vs_clip_and_viewport
+^^^^^^^^^^^^^^^^^^^^^^^^^^^
+
+Whether the entire TCL pipeline should be bypassed. This implies that
+vertices are pre-transformed for the viewport, and will not be run
+through the vertex shader.
+
+.. note::
+   
+   Implementations may still clip away vertices that are not in the viewport
+   when this is set.
+
 flatshade
-    If set, the provoking vertex of each polygon is used to determine the
-    color of the entire polygon.  If not set, fragment colors will be
-    interpolated between the vertex colors.
-    Note that this is separate from the fragment shader input attributes
-    CONSTANT, LINEAR and PERSPECTIVE.  We need the flatshade state at
+^^^^^^^^^
+
+If set, the provoking vertex of each polygon is used to determine the color
+of the entire polygon.  If not set, fragment colors will be interpolated
+between the vertex colors.
+
+The actual interpolated shading algorithm is obviously
+implementation-dependent, but will usually be Gourard for most hardware.
+
+.. note::
+
+    This is separate from the fragment shader input attributes
+    CONSTANT, LINEAR and PERSPECTIVE. The flatshade state is needed at
     clipping time to determine how to set the color of new vertices.
-    Also note that the draw module can implement flat shading by copying
-    the provoking vertex color to all the other vertices in the primitive.
+
+    :ref:`Draw` can implement flat shading by copying the provoking vertex
+    color to all the other vertices in the primitive.
 
 flatshade_first
-    Whether the first vertex should be the provoking vertex, for most
-    primitives. If not set, the last vertex is the provoking vertex.
+^^^^^^^^^^^^^^^
+
+Whether the first vertex should be the provoking vertex, for most primitives.
+If not set, the last vertex is the provoking vertex.
+
+There are several important exceptions to the specification of this rule.
+
+* ``PIPE_PRIMITIVE_POLYGON``: The provoking vertex is always the first
+  vertex. If the caller wishes to change the provoking vertex, they merely
+  need to rotate the vertices themselves.
+* ``PIPE_PRIMITIVE_QUAD``, ``PIPE_PRIMITIVE_QUAD_STRIP``: This option has no
+  effect; the provoking vertex is always the last vertex.
+* ``PIPE_PRIMITIVE_TRIANGLE_FAN``: When set, the provoking vertex is the
+  second vertex, not the first. This permits each segment of the fan to have
+  a different color.
+
+Other Members
+^^^^^^^^^^^^^
 
 light_twoside
-    If set, there are per-vertex back-facing colors.  The draw module
+    If set, there are per-vertex back-facing colors. :ref:`Draw`
     uses this state along with the front/back information to set the
     final vertex colors prior to rasterization.
 
 front_winding
     Indicates the window order of front-facing polygons, either
     PIPE_WINDING_CW or PIPE_WINDING_CCW
+
 cull_mode
     Indicates which polygons to cull, either PIPE_WINDING_NONE (cull no
     polygons), PIPE_WINDING_CW (cull clockwise-winding polygons),
@@ -68,7 +105,7 @@ line_stipple_enable
 line_stipple_pattern
     16-bit bitfield of on/off flags, used to pattern the line stipple.
 line_stipple_factor
-    When drawinga stippled line, each bit in the stipple pattern is
+    When drawing a stippled line, each bit in the stipple pattern is
     repeated N times, where N = line_stipple_factor + 1.
 line_last_pixel
     Controls whether the last pixel in a line is drawn or not.  OpenGL
@@ -96,7 +133,7 @@ sprite_coord_mode
     lower left vertex will have coordinate (0,0,0,1).
     For PIPE_SPRITE_COORD_UPPER_LEFT, the upper-left vertex will have
     coordinate (0,0,0,1).
-    This state is needed by the 'draw' module because that's where each
+    This state is needed by :ref:`Draw` because that's where each
     point vertex is converted into four quad vertices.  There's no other
     place to emit the new vertex texture coordinates which are required for
     sprite rendering.
@@ -108,45 +145,9 @@ scissor
     Whether the scissor test is enabled.
 
 multisample
-    Whether :ref:`MSAA` is enabled.
-
-bypass_vs_clip_and_viewport
-    Whether the entire TCL pipeline should be bypassed. This implies that
-    vertices are pre-transformed for the viewport, and will not be run
-    through the vertex shader. Note that implementations may still clip away
-    vertices that are not in the viewport.
+    Whether :term:`MSAA` is enabled.
 
 gl_rasterization_rules
     Whether the rasterizer should use (0.5, 0.5) pixel centers. When not set,
     the rasterizer will use (0, 0) for pixel centers.
 
-
-Notes
------
-
-flatshade
-^^^^^^^^^
-
-The actual interpolated shading algorithm is obviously
-implementation-dependent, but will usually be Gourard for most hardware.
-
-bypass_vs_clip_and_viewport
-^^^^^^^^^^^^^^^^^^^^^^^^^^^
-
-When set, this implies that vertices are pre-transformed for the viewport, and
-will not be run through the vertex shader. Note that implementations may still
-clip away vertices that are not visible.
-
-flatshade_first
-^^^^^^^^^^^^^^^
-
-There are several important exceptions to the specification of this rule.
-
-* ``PIPE_PRIMITIVE_POLYGON``: The provoking vertex is always the first
-  vertex. If the caller wishes to change the provoking vertex, they merely
-  need to rotate the vertices themselves.
-* ``PIPE_PRIMITIVE_QUAD``, ``PIPE_PRIMITIVE_QUAD_STRIP``: This option has no
-  effect; the provoking vertex is always the last vertex.
-* ``PIPE_PRIMITIVE_TRIANGLE_FAN``: When set, the provoking vertex is the
-  second vertex, not the first. This permits each segment of the fan to have
-  a different color.
diff --git a/src/gallium/docs/source/cso/sampler.rst b/src/gallium/docs/source/cso/sampler.rst
index e3f1757f57..8b67ca57f1 100644
--- a/src/gallium/docs/source/cso/sampler.rst
+++ b/src/gallium/docs/source/cso/sampler.rst
@@ -12,8 +12,6 @@ with the traditional (S, T, R, Q) notation.
 Members
 -------
 
-XXX undocumented compare_mode, compare_func
-
 wrap_s
     How to wrap the S coordinate. One of PIPE_TEX_WRAP.
 wrap_t
@@ -27,12 +25,18 @@ min_mip_filter
     PIPE_TEX_FILTER.
 mag_img_filter
     The filter to use when magnifying texels. One of PIPE_TEX_FILTER.
+compare_mode
+    If set to PIPE_TEX_COMPARE_R_TO_TEXTURE, texture output is computed
+    according to compare_func, using r coord and the texture value as operands.
+    If set to PIPE_TEX_COMPARE_NONE, no comparison calculation is performed.
+compare_func
+    How the comparison is computed. One of PIPE_FUNC.
 normalized_coords
     Whether the texture coordinates are normalized. If normalized, they will
     always be in [0, 1]. If not, they will be in the range of each dimension
     of the loaded texture.
 prefilter
-    XXX From the Doxy, "weird sampling state exposed by some APIs." Refine.
+    Cylindrical texcoord wrap enable per coord. Not exposed by most APIs.
 lod_bias
     The bias to apply to the level of detail.
 min_lod
diff --git a/src/gallium/docs/source/distro.rst b/src/gallium/docs/source/distro.rst
index a448203c6f..0ef9fe2645 100644
--- a/src/gallium/docs/source/distro.rst
+++ b/src/gallium/docs/source/distro.rst
@@ -61,10 +61,7 @@ VMWare SVGA
 ATI r300
 ^^^^^^^^
 
-AMD/ATI r600
-^^^^^^^^^^^^
-
-Highly experimental.
+Testing-quality.
 
 Softpipe
 ^^^^^^^^
@@ -109,20 +106,31 @@ Auxiliary
 CSO Cache
 ^^^^^^^^^
 
+The CSO cache is used to accelerate preparation of state by saving
+driver-specific state structures for later use.
+
+.. _draw:
+
 Draw
 ^^^^
 
+Draw is a software :term:`TCL` pipeline for hardware that lacks vertex shaders
+or other essential parts of pre-rasterization vertex preparation.
+
 Gallivm
 ^^^^^^^
 
 Indices
 ^^^^^^^
 
-Tool for translating or generating element indices for element-based
-rendering.
+Indices provides tools for translating or generating element indices for
+use with element-based rendering.
+
+Pipe Buffer Managers
+^^^^^^^^^^^^^^^^^^^^
 
-Pipe Buffer Manager
-^^^^^^^^^^^^^^^^^^^
+Each of these managers provides various services to drivers that are not
+fully utilizing a memory manager.
 
 Remote Debugger
 ^^^^^^^^^^^^^^^
@@ -133,7 +141,8 @@ Runtime Assembly Emission
 TGSI
 ^^^^
 
-Basic utilities for manipulating TGSI streams.
+The TGSI auxiliary module provides basic utilities for manipulating TGSI
+streams.
 
 Translate
 ^^^^^^^^^
diff --git a/src/gallium/docs/source/glossary.rst b/src/gallium/docs/source/glossary.rst
index 6a9110ce78..0696cb5d27 100644
--- a/src/gallium/docs/source/glossary.rst
+++ b/src/gallium/docs/source/glossary.rst
@@ -8,3 +8,16 @@ Glossary
       Multi-Sampled Anti-Aliasing. A basic anti-aliasing technique that takes
       multiple samples of the depth buffer, and uses this information to
       smooth the edges of polygons.
+
+   TCL
+      Transform, Clipping, & Lighting. The three stages of preparation in a
+      rasterizing pipeline prior to the actual rasterization of vertices into
+      fragments.
+
+   NPOT
+      Non-power-of-two. Usually applied to textures which have at least one
+      dimension which is not a power of two.
+
+   LOD
+      Level of Detail. Also spelled "LoD." The value that determines when the
+      switches between mipmaps occur during texture sampling.
diff --git a/src/gallium/docs/source/screen.rst b/src/gallium/docs/source/screen.rst
index 9631e6967e..3e57a282fd 100644
--- a/src/gallium/docs/source/screen.rst
+++ b/src/gallium/docs/source/screen.rst
@@ -3,6 +3,160 @@ Screen
 
 A screen is an object representing the context-independent part of a device.
 
+Useful Flags
+------------
+
+.. _pipe_cap:
+
+PIPE_CAP
+^^^^^^^^
+
+Pipe capabilities help expose hardware functionality not explicitly required
+by Gallium. For floating-point values, use :ref:`get_paramf`, and for boolean
+or integer values, use :ref:`get_param`.
+
+The integer capabilities:
+
+* ``MAX_TEXTURE_IMAGE_UNITS``: The maximum number of samplers available.
+* ``NPOT_TEXTURES``: Whether :term:`NPOT` textures may have repeat modes,
+  normalized coordinates, and mipmaps.
+* ``TWO_SIDED_STENCIL``: Whether the stencil test can also affect back-facing
+  polygons.
+* ``GLSL``: Deprecated.
+* ``DUAL_SOURCE_BLEND``: Whether dual-source blend factors are supported. See
+  :ref:`Blend` for more information.
+* ``ANISOTROPIC_FILTER``: Whether textures can be filtered anisotropically.
+* ``POINT_SPRITE``: Whether point sprites are available.
+* ``MAX_RENDER_TARGETS``: The maximum number of render targets that may be
+  bound.
+* ``OCCLUSION_QUERY``: Whether occlusion queries are available.
+* ``TEXTURE_SHADOW_MAP``: XXX
+* ``MAX_TEXTURE_2D_LEVELS``: The maximum number of mipmap levels available
+  for a 2D texture.
+* ``MAX_TEXTURE_3D_LEVELS``: The maximum number of mipmap levels available
+  for a 3D texture.
+* ``MAX_TEXTURE_CUBE_LEVELS``: The maximum number of mipmap levels available
+  for a cubemap.
+* ``TEXTURE_MIRROR_CLAMP``: Whether mirrored texture coordinates with clamp
+  are supported.
+* ``TEXTURE_MIRROR_REPEAT``: Whether mirrored repeating texture coordinates
+  are supported.
+* ``MAX_VERTEX_TEXTURE_UNITS``: The maximum number of samplers addressable
+  inside the vertex shader. If this is 0, then the vertex shader cannot
+  sample textures.
+* ``TGSI_CONT_SUPPORTED``: Whether the TGSI CONT opcode is supported.
+* ``BLEND_EQUATION_SEPARATE``: Whether alpha blend equations may be different
+  from color blend equations, in :ref:`Blend` state.
+* ``SM3``: Whether the vertex shader and fragment shader support equivalent
+  opcodes to the Shader Model 3 specification. XXX oh god this is horrible
+* ``MAX_PREDICATE_REGISTERS``: XXX
+* ``MAX_COMBINED_SAMPLERS``: The total number of samplers accessible from
+  the vertex and fragment shader, inclusive.
+* ``MAX_CONST_BUFFERS``: Maximum number of constant buffers that can be bound
+  to any shader stage using ``set_constant_buffer``. If 0 or 1, the pipe will
+  only permit binding one constant buffer per shader, and the shaders will
+  not permit two-dimensional access to constants.
+* ``MAX_CONST_BUFFER_SIZE``: Maximum byte size of a single constant buffer.
+* ``INDEP_BLEND_ENABLE``: Whether per-rendertarget blend enabling and channel
+  masks are supported. If 0, then the first rendertarget's blend mask is
+  replicated across all MRTs.
+* ``INDEP_BLEND_FUNC``: Whether per-rendertarget blend functions are
+  available. If 0, then the first rendertarget's blend functions affect all
+  MRTs.
+* ``PIPE_CAP_TGSI_FS_COORD_ORIGIN_UPPER_LEFT``: Whether the TGSI property
+  FS_COORD_ORIGIN with value UPPER_LEFT is supported.
+* ``PIPE_CAP_TGSI_FS_COORD_ORIGIN_LOWER_LEFT``: Whether the TGSI property
+  FS_COORD_ORIGIN with value LOWER_LEFT is supported.
+* ``PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_HALF_INTEGER``: Whether the TGSI
+  property FS_COORD_PIXEL_CENTER with value HALF_INTEGER is supported.
+* ``PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_INTEGER``: Whether the TGSI
+  property FS_COORD_PIXEL_CENTER with value INTEGER is supported.
+
+The floating-point capabilities:
+
+* ``MAX_LINE_WIDTH``: The maximum width of a regular line.
+* ``MAX_LINE_WIDTH_AA``: The maximum width of a smoothed line.
+* ``MAX_POINT_WIDTH``: The maximum width and height of a point.
+* ``MAX_POINT_WIDTH_AA``: The maximum width and height of a smoothed point.
+* ``MAX_TEXTURE_ANISOTROPY``: The maximum level of anisotropy that can be
+  applied to anisotropically filtered textures.
+* ``MAX_TEXTURE_LOD_BIAS``: The maximum :term:`LOD` bias that may be applied
+  to filtered textures.
+* ``GUARD_BAND_LEFT``, ``GUARD_BAND_TOP``, ``GUARD_BAND_RIGHT``,
+  ``GUARD_BAND_BOTTOM``: XXX
+
+XXX Is there a better home for this? vvv
+
+If 0 is returned, the driver is not aware of multiple constant buffers,
+supports binding of only one constant buffer, and does not support
+two-dimensional CONST register file access in TGSI shaders.
+
+If a value greater than 0 is returned, the driver can have multiple
+constant buffers bound to shader stages. The CONST register file can
+be accessed with two-dimensional indices, like in the example below.
+
+DCL CONST[0][0..7]       # declare first 8 vectors of constbuf 0
+DCL CONST[3][0]          # declare first vector of constbuf 3
+MOV OUT[0], CONST[0][3]  # copy vector 3 of constbuf 0
+
+For backwards compatibility, one-dimensional access to CONST register
+file is still supported. In that case, the constbuf index is assumed
+to be 0.
+
+.. _pipe_buffer_usage:
+
+PIPE_BUFFER_USAGE
+^^^^^^^^^^^^^^^^^
+
+These flags control buffer creation. Buffers may only have one role, so
+care should be taken to not allocate a buffer with the wrong usage.
+
+* ``PIXEL``: This is the flag to use for all textures.
+* ``VERTEX``: A vertex buffer.
+* ``INDEX``: An element buffer.
+* ``CONSTANT``: A buffer of shader constants.
+
+Buffers are inevitably abstracting the pipe's underlying memory management,
+so many of their usage flags can be used to direct the way the buffer is
+handled.
+
+* ``CPU_READ``, ``CPU_WRITE``: Whether the user will map and, in the case of
+  the latter, write to, the buffer. The convenience flag ``CPU_READ_WRITE`` is
+  available to signify a read/write buffer.
+* ``GPU_READ``, ``GPU_WRITE``: Whether the driver will internally need to
+  read from or write to the buffer. The latter will only happen if the buffer
+  is made into a render target.
+* ``DISCARD``: When set on a map, the contents of the map will be discarded
+  beforehand. Cannot be used with ``CPU_READ``.
+* ``DONTBLOCK``: When set on a map, the map will fail if the buffer cannot be
+  mapped immediately.
+* ``UNSYNCHRONIZED``: When set on a map, any outstanding operations on the
+  buffer will be ignored. The interaction of any writes to the map and any
+  operations pending with the buffer are undefined. Cannot be used with
+  ``CPU_READ``.
+* ``FLUSH_EXPLICIT``: When set on a map, written ranges of the map require
+  explicit flushes using :ref:`buffer_flush_mapped_range`. Requires
+  ``CPU_WRITE``.
+
+.. _pipe_texture_usage:
+
+PIPE_TEXTURE_USAGE
+^^^^^^^^^^^^^^^^^^
+
+These flags determine the possible roles a texture may be used for during its
+lifetime. Texture usage flags are cumulative and may be combined to create a
+texture that can be used as multiple things.
+
+* ``RENDER_TARGET``: A colorbuffer or pixelbuffer.
+* ``DISPLAY_TARGET``: A sharable buffer that can be given to another process.
+* ``PRIMARY``: A frontbuffer or scanout buffer.
+* ``DEPTH_STENCIL``: A depthbuffer, stencilbuffer, or Z buffer. Gallium does
+  not explicitly provide for stencil-only buffers, so any stencilbuffer
+  validated here is implicitly also a depthbuffer.
+* ``SAMPLER``: A texture that may be sampled from in a fragment or vertex
+  shader.
+* ``DYNAMIC``: A texture that will be mapped frequently.
+
 Methods
 -------
 
@@ -18,22 +172,87 @@ get_vendor
 
 Returns the screen vendor.
 
+.. _get_param:
+
 get_param
 ^^^^^^^^^
 
 Get an integer/boolean screen parameter.
 
+**param** is one of the :ref:`PIPE_CAP` names.
+
+.. _get_paramf:
+
 get_paramf
 ^^^^^^^^^^
 
 Get a floating-point screen parameter.
 
+**param** is one of the :ref:`PIPE_CAP` names.
+
 is_format_supported
 ^^^^^^^^^^^^^^^^^^^
 
 See if a format can be used in a specific manner.
 
+**usage** is a bitmask of :ref:`PIPE_TEXTURE_USAGE` flags.
+
+Returns TRUE if all usages can be satisfied.
+
+.. note::
+
+   ``PIPE_TEXTURE_USAGE_DYNAMIC`` is not a valid usage.
+
+.. _texture_create:
+
 texture_create
 ^^^^^^^^^^^^^^
 
-Given a template of texture setup, create a BO-backed texture.
+Given a template of texture setup, create a buffer and texture.
+
+texture_blanket
+^^^^^^^^^^^^^^^
+
+Like :ref:`texture_create`, but use a supplied buffer instead of creating a
+new one.
+
+texture_destroy
+^^^^^^^^^^^^^^^
+
+Destroy a texture. The buffer backing the texture is destroyed if it has no
+more references.
+
+buffer_map
+^^^^^^^^^^
+
+Map a buffer into memory.
+
+**usage** is a bitmask of :ref:`PIPE_TEXTURE_USAGE` flags.
+
+Returns a pointer to the map, or NULL if the mapping failed.
+
+buffer_map_range
+^^^^^^^^^^^^^^^^
+
+Map a range of a buffer into memory.
+
+The returned map is always relative to the beginning of the buffer, not the
+beginning of the mapped range.
+
+.. _buffer_flush_mapped_range:
+
+buffer_flush_mapped_range
+^^^^^^^^^^^^^^^^^^^^^^^^^
+
+Flush a range of mapped memory into a buffer.
+
+The buffer must have been mapped with ``PIPE_BUFFER_USAGE_FLUSH_EXPLICIT``.
+
+**usage** is a bitmask of :ref:`PIPE_TEXTURE_USAGE` flags.
+
+buffer_unmap
+^^^^^^^^^^^^
+
+Unmap a buffer from memory.
+
+Any pointers into the map should be considered invalid and discarded.
diff --git a/src/gallium/docs/source/tgsi.rst b/src/gallium/docs/source/tgsi.rst
index ebee4902b0..3e702ceeda 100644
--- a/src/gallium/docs/source/tgsi.rst
+++ b/src/gallium/docs/source/tgsi.rst
@@ -516,8 +516,11 @@ SEQ - Set On Equal
 .. math::
 
   dst.x = (src0.x == src1.x) ? 1 : 0
+
   dst.y = (src0.y == src1.y) ? 1 : 0
+
   dst.z = (src0.z == src1.z) ? 1 : 0
+
   dst.w = (src0.w == src1.w) ? 1 : 0
 
 
@@ -526,8 +529,11 @@ SFL - Set On False
 .. math::
 
   dst.x = 0
+
   dst.y = 0
+
   dst.z = 0
+
   dst.w = 0
 
 Considered for removal.
@@ -537,8 +543,11 @@ SGT - Set On Greater Than
 .. math::
 
   dst.x = (src0.x > src1.x) ? 1 : 0
+
   dst.y = (src0.y > src1.y) ? 1 : 0
+
   dst.z = (src0.z > src1.z) ? 1 : 0
+
   dst.w = (src0.w > src1.w) ? 1 : 0
 
 
@@ -560,8 +569,11 @@ SLE - Set On Less Equal Than
 .. math::
 
   dst.x = (src0.x <= src1.x) ? 1 : 0
+
   dst.y = (src0.y <= src1.y) ? 1 : 0
+
   dst.z = (src0.z <= src1.z) ? 1 : 0
+
   dst.w = (src0.w <= src1.w) ? 1 : 0
 
 
@@ -570,8 +582,11 @@ SNE - Set On Not Equal
 .. math::
 
   dst.x = (src0.x != src1.x) ? 1 : 0
+
   dst.y = (src0.y != src1.y) ? 1 : 0
+
   dst.z = (src0.z != src1.z) ? 1 : 0
+
   dst.w = (src0.w != src1.w) ? 1 : 0
 
 
@@ -580,8 +595,11 @@ STR - Set On True
 .. math::
 
   dst.x = 1
+
   dst.y = 1
+
   dst.z = 1
+
   dst.w = 1
 
 
@@ -629,8 +647,11 @@ X2D - 2D Coordinate Transformation
 .. math::
 
   dst.x = src0.x + src1.x \times src2.x + src1.y \times src2.y
+
   dst.y = src0.y + src1.x \times src2.z + src1.y \times src2.w
+
   dst.z = src0.x + src1.x \times src2.x + src1.y \times src2.y
+
   dst.w = src0.y + src1.x \times src2.z + src1.y \times src2.w
 
 Considered for removal.
@@ -979,13 +1000,13 @@ XOR - Bitwise Xor
 
 .. math::
 
-  dst.x = src0.x ^ src1.x
+  dst.x = src0.x \oplus src1.x
 
-  dst.y = src0.y ^ src1.y
+  dst.y = src0.y \oplus src1.y
 
-  dst.z = src0.z ^ src1.z
+  dst.z = src0.z \oplus src1.z
 
-  dst.w = src0.w ^ src1.w
+  dst.w = src0.w \oplus src1.w
 
 
 SAD - Sum Of Absolute Differences
@@ -1090,6 +1111,117 @@ BREAKC - Break Conditional
 
   TBD
 
+Double Opcodes
+^^^^^^^^^^^^^^^
+
+DADD - Add Double
+
+.. math::
+
+  dst.xy = src0.xy + src1.xy
+
+  dst.zw = src0.zw + src1.zw
+
+
+DDIV - Divide Double
+
+.. math::
+
+  dst.xy = src0.xy / src1.xy
+
+  dst.zw = src0.zw / src1.zw
+
+DSEQ - Set Double on Equal
+
+.. math::
+
+  dst.xy = src0.xy == src1.xy ? 1.0F : 0.0F
+
+  dst.zw = src0.zw == src1.zw ? 1.0F : 0.0F
+
+DSLT - Set Double on Less than
+
+.. math::
+
+  dst.xy = src0.xy < src1.xy ? 1.0F : 0.0F
+
+  dst.zw = src0.zw < src1.zw ? 1.0F : 0.0F
+
+DFRAC - Double Fraction
+
+.. math::
+
+  dst.xy = src.xy - \lfloor src.xy\rfloor
+
+  dst.zw = src.zw - \lfloor src.zw\rfloor
+
+
+DFRACEXP - Convert Double Number to Fractional and Integral Components
+
+.. math::
+
+  dst0.xy = frexp(src.xy, dst1.xy)
+
+  dst0.zw = frexp(src.zw, dst1.zw)
+
+DLDEXP - Multiple Double Number by Integral Power of 2
+
+.. math::
+
+  dst.xy = ldexp(src0.xy, src1.xy)
+
+  dst.zw = ldexp(src0.zw, src1.zw)
+
+DMIN - Minimum Double
+
+.. math::
+
+  dst.xy = min(src0.xy, src1.xy)
+
+  dst.zw = min(src0.zw, src1.zw)
+
+DMAX - Maximum Double
+
+.. math::
+
+  dst.xy = max(src0.xy, src1.xy)
+
+  dst.zw = max(src0.zw, src1.zw)
+
+DMUL - Multiply Double
+
+.. math::
+
+  dst.xy = src0.xy \times src1.xy
+
+  dst.zw = src0.zw \times src1.zw
+
+
+DMAD - Multiply And Add Doubles
+
+.. math::
+
+  dst.xy = src0.xy \times src1.xy + src2.xy
+
+  dst.zw = src0.zw \times src1.zw + src2.zw
+
+
+DRCP - Reciprocal Double
+
+.. math::
+
+   dst.xy = \frac{1}{src.xy}
+
+   dst.zw = \frac{1}{src.zw}
+
+DSQRT - Square root double
+
+.. math::
+
+   dst.xy = \sqrt{src.xy}
+
+   dst.zw = \sqrt{src.zw}
+
 
 Explanation of symbols used
 ------------------------------
@@ -1187,9 +1319,8 @@ are the Cartesian coordinates, and ``w`` is the homogenous coordinate and used
 for the perspective divide, if enabled.
 
 As a vertex shader output, position should be scaled to the viewport. When
-used in fragment shaders, position will ---
-
-XXX --- wait a minute. Should position be in [0,1] for x and y?
+used in fragment shaders, position will be in window coordinates. The convention
+used depends on the FS_COORD_ORIGIN and FS_COORD_PIXEL_CENTER properties.
 
 XXX additionally, is there a way to configure the perspective divide? it's
 accelerated on most chipsets AFAIK...
@@ -1268,3 +1399,45 @@ TGSI_SEMANTIC_EDGEFLAG
 """"""""""""""""""""""
 
 XXX no clue
+
+
+Properties
+^^^^^^^^^^^^^^^^^^^^^^^^
+
+
+  Properties are general directives that apply to the whole TGSI program.
+
+FS_COORD_ORIGIN
+"""""""""""""""
+
+Specifies the fragment shader TGSI_SEMANTIC_POSITION coordinate origin.
+The default value is UPPER_LEFT.
+
+If UPPER_LEFT, the position will be (0,0) at the upper left corner and
+increase downward and rightward.
+If LOWER_LEFT, the position will be (0,0) at the lower left corner and
+increase upward and rightward.
+
+OpenGL defaults to LOWER_LEFT, and is configurable with the
+GL_ARB_fragment_coord_conventions extension.
+
+DirectX 9/10 use UPPER_LEFT.
+
+FS_COORD_PIXEL_CENTER
+"""""""""""""""""""""
+
+Specifies the fragment shader TGSI_SEMANTIC_POSITION pixel center convention.
+The default value is HALF_INTEGER.
+
+If HALF_INTEGER, the fractionary part of the position will be 0.5
+If INTEGER, the fractionary part of the position will be 0.0
+
+Note that this does not affect the set of fragments generated by
+rasterization, which is instead controlled by gl_rasterization_rules in the
+rasterizer.
+
+OpenGL defaults to HALF_INTEGER, and is configurable with the
+GL_ARB_fragment_coord_conventions extension.
+
+DirectX 9 uses INTEGER.
+DirectX 10 uses HALF_INTEGER.
diff --git a/src/gallium/drivers/cell/ppu/cell_draw_arrays.c b/src/gallium/drivers/cell/ppu/cell_draw_arrays.c
index 0a4da8ecc8..c674d0be63 100644
--- a/src/gallium/drivers/cell/ppu/cell_draw_arrays.c
+++ b/src/gallium/drivers/cell/ppu/cell_draw_arrays.c
@@ -59,7 +59,7 @@ cell_map_constant_buffers(struct cell_context *sp)
       }
    }
 
-   draw_set_mapped_constant_buffer(sp->draw, PIPE_SHADER_VERTEX,
+   draw_set_mapped_constant_buffer(sp->draw, PIPE_SHADER_VERTEX, 0,
                                    sp->mapped_constants[PIPE_SHADER_VERTEX],
                                    sp->constants[PIPE_SHADER_VERTEX]->size);
 }
diff --git a/src/gallium/drivers/cell/ppu/cell_gen_fragment.c b/src/gallium/drivers/cell/ppu/cell_gen_fragment.c
index 66d4b3b6a3..0dab34075d 100644
--- a/src/gallium/drivers/cell/ppu/cell_gen_fragment.c
+++ b/src/gallium/drivers/cell/ppu/cell_gen_fragment.c
@@ -408,7 +408,7 @@ gen_blend(const struct pipe_blend_state *blend,
    int one_reg = -1;
    int constR_reg = -1, constG_reg = -1, constB_reg = -1, constA_reg = -1;
 
-   ASSERT(blend->blend_enable);
+   ASSERT(blend->rt[0].blend_enable);
 
    /* packed RGBA -> float colors */
    unpack_colors(f, color_format, fbRGBA_reg,
@@ -420,7 +420,7 @@ gen_blend(const struct pipe_blend_state *blend,
     * because in some cases (like PIPE_BLENDFACTOR_ONE and 
     * PIPE_BLENDFACTOR_ZERO) we can avoid doing unnecessary math.
     */
-   switch (blend->rgb_src_factor) {
+   switch (blend->rt[0].rgb_src_factor) {
    case PIPE_BLENDFACTOR_ONE:
       /* factors = (1,1,1), so term = (R,G,B) */
       spe_move(f, term1R_reg, fragR_reg);
@@ -574,7 +574,7 @@ gen_blend(const struct pipe_blend_state *blend,
     * the full term A*factor, not just the factor itself, because
     * in many cases we can avoid doing unnecessary multiplies.
     */
-   switch (blend->alpha_src_factor) {
+   switch (blend->rt[0].alpha_src_factor) {
    case PIPE_BLENDFACTOR_ZERO:
       /* factor = 0, so term = 0 */
       spe_load_float(f, term1A_reg, 0.0f);
@@ -648,7 +648,7 @@ gen_blend(const struct pipe_blend_state *blend,
     * the full term (Rfb,Gfb,Bfb)*(factor), not just the factor itself, because
     * in many cases we can avoid doing unnecessary multiplies.
     */
-   switch (blend->rgb_dst_factor) {
+   switch (blend->rt[0].rgb_dst_factor) {
    case PIPE_BLENDFACTOR_ONE:
       /* factors = (1,1,1), so term = (Rfb,Gfb,Bfb) */
       spe_move(f, term2R_reg, fbR_reg);
@@ -786,7 +786,7 @@ gen_blend(const struct pipe_blend_state *blend,
     * the full term Afb*factor, not just the factor itself, because
     * in many cases we can avoid doing unnecessary multiplies.
     */
-   switch (blend->alpha_dst_factor) {
+   switch (blend->rt[0].alpha_dst_factor) {
    case PIPE_BLENDFACTOR_ONE:
       /* factor = 1, so term = Afb */
       spe_move(f, term2A_reg, fbA_reg);
@@ -858,7 +858,7 @@ gen_blend(const struct pipe_blend_state *blend,
    /*
     * Combine Src/Dest RGB terms as per the blend equation.
     */
-   switch (blend->rgb_func) {
+   switch (blend->rt[0].rgb_func) {
    case PIPE_BLEND_ADD:
       spe_fa(f, fragR_reg, term1R_reg, term2R_reg);
       spe_fa(f, fragG_reg, term1G_reg, term2G_reg);
@@ -891,7 +891,7 @@ gen_blend(const struct pipe_blend_state *blend,
    /*
     * Combine Src/Dest A term
     */
-   switch (blend->alpha_func) {
+   switch (blend->rt[0].alpha_func) {
    case PIPE_BLEND_ADD:
       spe_fa(f, fragA_reg, term1A_reg, term2A_reg);
       break;
@@ -2118,7 +2118,7 @@ cell_gen_fragment_function(struct cell_context *cell,
    spe_comment(f, 0, "Fetch quad colors from tile");
    spe_lqx(f, fbRGBA_reg, color_tile_reg, quad_offset_reg);
 
-   if (blend->blend_enable) {
+   if (blend->rt[0].blend_enable) {
       spe_comment(f, 0, "Perform blending");
       gen_blend(blend, blend_color, f, color_format,
                 fragR_reg, fragG_reg, fragB_reg, fragA_reg, fbRGBA_reg);
@@ -2143,9 +2143,9 @@ cell_gen_fragment_function(struct cell_context *cell,
          gen_logicop(blend, f, rgba_reg, fbRGBA_reg);
       }
 
-      if (blend->colormask != PIPE_MASK_RGBA) {
+      if (blend->rt[0].colormask != PIPE_MASK_RGBA) {
          spe_comment(f, 0, "Compute color mask");
-         gen_colormask(f, blend->colormask, color_format, rgba_reg, fbRGBA_reg);
+         gen_colormask(f, blend->rt[0].colormask, color_format, rgba_reg, fbRGBA_reg);
       }
 
       /* Mix fragment colors with framebuffer colors using the quad/pixel mask:
diff --git a/src/gallium/drivers/cell/ppu/cell_screen.c b/src/gallium/drivers/cell/ppu/cell_screen.c
index d185c6b849..37b04396b3 100644
--- a/src/gallium/drivers/cell/ppu/cell_screen.c
+++ b/src/gallium/drivers/cell/ppu/cell_screen.c
@@ -86,6 +86,12 @@ cell_get_param(struct pipe_screen *screen, int param)
       return 0; /* XXX to do */
    case PIPE_CAP_TGSI_CONT_SUPPORTED:
       return 1;
+   case PIPE_CAP_TGSI_FS_COORD_ORIGIN_UPPER_LEFT:
+   case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_HALF_INTEGER:
+      return 1;
+   case PIPE_CAP_TGSI_FS_COORD_ORIGIN_LOWER_LEFT:
+   case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_INTEGER:
+      return 0;
    default:
       return 0;
    }
diff --git a/src/gallium/drivers/cell/ppu/cell_state_per_fragment.c b/src/gallium/drivers/cell/ppu/cell_state_per_fragment.c
index d97c22b2ef..21af7ed1c3 100644
--- a/src/gallium/drivers/cell/ppu/cell_state_per_fragment.c
+++ b/src/gallium/drivers/cell/ppu/cell_state_per_fragment.c
@@ -999,23 +999,23 @@ cell_generate_alpha_blend(struct cell_blend_state *cb)
    /* Does the selected blend mode make use of the source / destination
     * color (RGB) blend factors?
     */
-   boolean need_color_factor = b->blend_enable
-       && (b->rgb_func != PIPE_BLEND_MIN)
-       && (b->rgb_func != PIPE_BLEND_MAX);
+   boolean need_color_factor = b->rt[0].blend_enable
+       && (b->rt[0].rgb_func != PIPE_BLEND_MIN)
+       && (b->rt[0].rgb_func != PIPE_BLEND_MAX);
 
    /* Does the selected blend mode make use of the source / destination
     * alpha blend factors?
     */
-   boolean need_alpha_factor = b->blend_enable
-       && (b->alpha_func != PIPE_BLEND_MIN)
-       && (b->alpha_func != PIPE_BLEND_MAX);
+   boolean need_alpha_factor = b->rt[0].blend_enable
+       && (b->rt[0].alpha_func != PIPE_BLEND_MIN)
+       && (b->rt[0].alpha_func != PIPE_BLEND_MAX);
 
 
-   if (b->blend_enable) {
-      sF[0] = b->rgb_src_factor;
+   if (b->rt[0].blend_enable) {
+      sF[0] = b->rt[0].rgb_src_factor;
       sF[1] = sF[0];
       sF[2] = sF[0];
-      switch (b->alpha_src_factor & 0x0f) {
+      switch (b->rt[0].alpha_src_factor & 0x0f) {
       case PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE:
          sF[3] = PIPE_BLENDFACTOR_ONE;
          break;
@@ -1023,30 +1023,30 @@ cell_generate_alpha_blend(struct cell_blend_state *cb)
       case PIPE_BLENDFACTOR_DST_COLOR:
       case PIPE_BLENDFACTOR_CONST_COLOR:
       case PIPE_BLENDFACTOR_SRC1_COLOR:
-         sF[3] = b->alpha_src_factor + 1;
+         sF[3] = b->rt[0].alpha_src_factor + 1;
          break;
       default:
-         sF[3] = b->alpha_src_factor;
+         sF[3] = b->rt[0].alpha_src_factor;
       }
 
-      dF[0] = b->rgb_dst_factor;
+      dF[0] = b->rt[0].rgb_dst_factor;
       dF[1] = dF[0];
       dF[2] = dF[0];
-      switch (b->alpha_dst_factor & 0x0f) {
+      switch (b->rt[0].alpha_dst_factor & 0x0f) {
       case PIPE_BLENDFACTOR_SRC_COLOR:
       case PIPE_BLENDFACTOR_DST_COLOR:
       case PIPE_BLENDFACTOR_CONST_COLOR:
       case PIPE_BLENDFACTOR_SRC1_COLOR:
-         dF[3] = b->alpha_dst_factor + 1;
+         dF[3] = b->rt[0].alpha_dst_factor + 1;
          break;
       default:
-         dF[3] = b->alpha_dst_factor;
+         dF[3] = b->rt[0].alpha_dst_factor;
       }
 
-      func[0] = b->rgb_func;
+      func[0] = b->rt[0].rgb_func;
       func[1] = func[0];
       func[2] = func[0];
-      func[3] = b->alpha_func;
+      func[3] = b->rt[0].alpha_func;
    } else {
       sF[0] = PIPE_BLENDFACTOR_ONE;
       sF[1] = PIPE_BLENDFACTOR_ONE;
@@ -1067,7 +1067,7 @@ cell_generate_alpha_blend(struct cell_blend_state *cb)
    /* If alpha writing is enabled and the alpha blend mode requires use of
     * the alpha factor, calculate the alpha factor.
     */
-   if (((b->colormask & 8) != 0) && need_alpha_factor) {
+   if (((b->rt[0].colormask & 8) != 0) && need_alpha_factor) {
       src_factor[3] = emit_alpha_factor_calculation(f, sF[3], const_color[3],
                                                     frag[3], pixel[3]);
 
@@ -1091,8 +1091,8 @@ cell_generate_alpha_blend(struct cell_blend_state *cb)
       src_factor[2] = dst_factor[3];
    } else if (need_color_factor) {
       emit_color_factor_calculation(f,
-                                    b->rgb_src_factor,
-                                    b->colormask,
+                                    b->rt[0].rgb_src_factor,
+                                    b->rt[0].colormask,
                                     frag, pixel, const_color, src_factor);
    }
 
@@ -1111,15 +1111,15 @@ cell_generate_alpha_blend(struct cell_blend_state *cb)
       dst_factor[2] = src_factor[2];
    } else if (need_color_factor) {
       emit_color_factor_calculation(f,
-                                    b->rgb_dst_factor,
-                                    b->colormask,
+                                    b->rt[0].rgb_dst_factor,
+                                    b->rt[0].colormask,
                                     frag, pixel, const_color, dst_factor);
    }
 
 
 
    for (i = 0; i < 4; ++i) {
-      if ((b->colormask & (1U << i)) != 0) {
+      if ((b->rt[0].colormask & (1U << i)) != 0) {
          emit_blend_calculation(f,
                                 func[i], sF[i], dF[i],
                                 frag[i], src_factor[i],
@@ -1216,7 +1216,7 @@ cell_generate_logic_op(struct spe_function *f,
 
    /* Short-circuit the noop and invert cases.
     */
-   if ((logic_op == PIPE_LOGICOP_NOOP) || (blend->colormask == 0)) {
+   if ((logic_op == PIPE_LOGICOP_NOOP) || (blend->rt[0].colormask == 0)) {
       spe_bi(f, 0, 0, 0);
       return;
    } else if (logic_op == PIPE_LOGICOP_INVERT) {
diff --git a/src/gallium/drivers/cell/spu/spu_main.h b/src/gallium/drivers/cell/spu/spu_main.h
index b18f4c22ef..a9d72f84d5 100644
--- a/src/gallium/drivers/cell/spu/spu_main.h
+++ b/src/gallium/drivers/cell/spu/spu_main.h
@@ -100,8 +100,10 @@ struct spu_framebuffer
    void *depth_start;              /**< addr of depth surface in main memory */
    enum pipe_format color_format;
    enum pipe_format depth_format;
-   uint width, height;             /**< size in pixels */
-   uint width_tiles, height_tiles; /**< width and height in tiles */
+   uint width;                     /**< width in pixels */
+   uint height;                    /**< height in pixels */
+   uint width_tiles;               /**< width in tiles */
+   uint height_tiles;              /**< width in tiles */
 
    uint color_clear_value;
    uint depth_clear_value;
@@ -116,15 +118,23 @@ PIPE_ALIGN_TYPE(16,
 struct spu_texture_level
 {
    void *start;
-   ushort width, height, depth;
+   ushort width;
+   ushort height;
+   ushort depth;
    ushort tiles_per_row;
    uint bytes_per_image;
    /** texcoord scale factors */
-   vector float scale_s, scale_t, scale_r;
+   vector float scale_s;
+   vector float scale_t;
+   vector float scale_r;
    /** texcoord masks (if REPEAT then size-1, else ~0) */
-   vector signed int mask_s, mask_t, mask_r;
+   vector signed int mask_s;
+   vector signed int mask_t;
+   vector signed int mask_r;
    /** texcoord clamp limits */
-   vector signed int max_s, max_t, max_r;
+   vector signed int max_s;
+   vector signed int max_t;
+   vector signed int max_r;
 });
 
 
@@ -166,7 +176,8 @@ struct spu_global
    boolean read_depth_stencil;
 
    /** Current tiles' status */
-   ubyte cur_ctile_status, cur_ztile_status;
+   ubyte cur_ctile_status;
+   ubyte cur_ztile_status;
 
    /** Status of all tiles in framebuffer */
    PIPE_ALIGN_VAR(16) ubyte ctile_status[CELL_MAX_HEIGHT/TILE_SIZE][CELL_MAX_WIDTH/TILE_SIZE];
diff --git a/src/gallium/drivers/i915/i915_clear.c b/src/gallium/drivers/i915/i915_clear.c
index 90530f2826..0d0859f8f3 100644
--- a/src/gallium/drivers/i915/i915_clear.c
+++ b/src/gallium/drivers/i915/i915_clear.c
@@ -32,7 +32,6 @@
 
 #include "util/u_clear.h"
 #include "i915_context.h"
-#include "i915_state.h"
 
 
 /**
diff --git a/src/gallium/drivers/i915/i915_context.c b/src/gallium/drivers/i915/i915_context.c
index 89feeade75..a0c80d0228 100644
--- a/src/gallium/drivers/i915/i915_context.c
+++ b/src/gallium/drivers/i915/i915_context.c
@@ -29,12 +29,9 @@
 #include "i915_state.h"
 #include "i915_screen.h"
 #include "i915_batch.h"
-#include "i915_texture.h"
-#include "i915_reg.h"
 
 #include "draw/draw_context.h"
 #include "pipe/p_defines.h"
-#include "pipe/internal/p_winsys_screen.h"
 #include "pipe/p_inlines.h"
 #include "util/u_memory.h"
 #include "pipe/p_screen.h"
@@ -84,7 +81,7 @@ i915_draw_range_elements(struct pipe_context *pipe,
    }
 
 
-   draw_set_mapped_constant_buffer(draw, PIPE_SHADER_VERTEX,
+   draw_set_mapped_constant_buffer(draw, PIPE_SHADER_VERTEX, 0,
                                    i915->current.constants[PIPE_SHADER_VERTEX],
                                    (i915->current.num_user_constants[PIPE_SHADER_VERTEX] * 
                                       4 * sizeof(float)));
diff --git a/src/gallium/drivers/i915/i915_debug.c b/src/gallium/drivers/i915/i915_debug.c
index c6e6d6fd31..237654d26b 100644
--- a/src/gallium/drivers/i915/i915_debug.c
+++ b/src/gallium/drivers/i915/i915_debug.c
@@ -29,7 +29,6 @@
 #include "i915_context.h"
 #include "i915_debug.h"
 #include "i915_batch.h"
-#include "pipe/internal/p_winsys_screen.h"
 #include "util/u_debug.h"
 
 
diff --git a/src/gallium/drivers/i915/i915_debug_fp.c b/src/gallium/drivers/i915/i915_debug_fp.c
index 9c5b117b6d..f9c40d8a11 100644
--- a/src/gallium/drivers/i915/i915_debug_fp.c
+++ b/src/gallium/drivers/i915/i915_debug_fp.c
@@ -29,7 +29,6 @@
 #include "i915_reg.h"
 #include "i915_debug.h"
 #include "pipe/internal/p_winsys_screen.h"
-#include "util/u_memory.h"
 
 
 static void
diff --git a/src/gallium/drivers/i915/i915_screen.c b/src/gallium/drivers/i915/i915_screen.c
index d4ee8f5339..7b8d66b8aa 100644
--- a/src/gallium/drivers/i915/i915_screen.c
+++ b/src/gallium/drivers/i915/i915_screen.c
@@ -117,6 +117,12 @@ i915_get_param(struct pipe_screen *screen, int param)
       return 8;  /* max 128x128x128 */
    case PIPE_CAP_MAX_TEXTURE_CUBE_LEVELS:
       return 11; /* max 1024x1024 */
+   case PIPE_CAP_TGSI_FS_COORD_ORIGIN_UPPER_LEFT:
+   case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_HALF_INTEGER:
+      return 1;
+   case PIPE_CAP_TGSI_FS_COORD_ORIGIN_LOWER_LEFT:
+   case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_INTEGER:
+      return 0;
    default:
       return 0;
    }
diff --git a/src/gallium/drivers/i915/i915_state.c b/src/gallium/drivers/i915/i915_state.c
index 0fab6e1bc3..23e4d6b993 100644
--- a/src/gallium/drivers/i915/i915_state.c
+++ b/src/gallium/drivers/i915/i915_state.c
@@ -30,7 +30,6 @@
 
 
 #include "draw/draw_context.h"
-#include "pipe/internal/p_winsys_screen.h"
 #include "pipe/p_inlines.h"
 #include "util/u_math.h"
 #include "util/u_memory.h"
@@ -38,7 +37,6 @@
 
 #include "i915_context.h"
 #include "i915_reg.h"
-#include "i915_state.h"
 #include "i915_state_inlines.h"
 #include "i915_fpc.h"
 
@@ -105,13 +103,13 @@ i915_create_blend_state(struct pipe_context *pipe,
    struct i915_blend_state *cso_data = CALLOC_STRUCT( i915_blend_state );
 
    {
-      unsigned eqRGB  = blend->rgb_func;
-      unsigned srcRGB = blend->rgb_src_factor;
-      unsigned dstRGB = blend->rgb_dst_factor;
+      unsigned eqRGB  = blend->rt[0].rgb_func;
+      unsigned srcRGB = blend->rt[0].rgb_src_factor;
+      unsigned dstRGB = blend->rt[0].rgb_dst_factor;
 
-      unsigned eqA    = blend->alpha_func;
-      unsigned srcA   = blend->alpha_src_factor;
-      unsigned dstA   = blend->alpha_dst_factor;
+      unsigned eqA    = blend->rt[0].alpha_func;
+      unsigned srcA   = blend->rt[0].alpha_src_factor;
+      unsigned dstA   = blend->rt[0].alpha_dst_factor;
 
       /* Special handling for MIN/MAX filter modes handled at
        * state_tracker level.
@@ -148,22 +146,22 @@ i915_create_blend_state(struct pipe_context *pipe,
    if (blend->dither)
       cso_data->LIS5 |= S5_COLOR_DITHER_ENABLE;
 
-   if ((blend->colormask & PIPE_MASK_R) == 0)
+   if ((blend->rt[0].colormask & PIPE_MASK_R) == 0)
       cso_data->LIS5 |= S5_WRITEDISABLE_RED;
 
-   if ((blend->colormask & PIPE_MASK_G) == 0)
+   if ((blend->rt[0].colormask & PIPE_MASK_G) == 0)
       cso_data->LIS5 |= S5_WRITEDISABLE_GREEN;
 
-   if ((blend->colormask & PIPE_MASK_B) == 0)
+   if ((blend->rt[0].colormask & PIPE_MASK_B) == 0)
       cso_data->LIS5 |= S5_WRITEDISABLE_BLUE;
 
-   if ((blend->colormask & PIPE_MASK_A) == 0)
+   if ((blend->rt[0].colormask & PIPE_MASK_A) == 0)
       cso_data->LIS5 |= S5_WRITEDISABLE_ALPHA;
 
-   if (blend->blend_enable) {
-      unsigned funcRGB = blend->rgb_func;
-      unsigned srcRGB  = blend->rgb_src_factor;
-      unsigned dstRGB  = blend->rgb_dst_factor;
+   if (blend->rt[0].blend_enable) {
+      unsigned funcRGB = blend->rt[0].rgb_func;
+      unsigned srcRGB  = blend->rt[0].rgb_src_factor;
+      unsigned dstRGB  = blend->rt[0].rgb_dst_factor;
 
       cso_data->LIS6 |= (S6_CBUF_BLEND_ENABLE |
                          SRC_BLND_FACT(i915_translate_blend_factor(srcRGB)) |
diff --git a/src/gallium/drivers/i915/i915_state_derived.c b/src/gallium/drivers/i915/i915_state_derived.c
index 03dd5091a6..f5b0e9f011 100644
--- a/src/gallium/drivers/i915/i915_state_derived.c
+++ b/src/gallium/drivers/i915/i915_state_derived.c
@@ -33,7 +33,6 @@
 #include "i915_context.h"
 #include "i915_state.h"
 #include "i915_reg.h"
-#include "i915_fpc.h"
 
 
 
diff --git a/src/gallium/drivers/i915/i915_state_sampler.c b/src/gallium/drivers/i915/i915_state_sampler.c
index cbac4175c8..e5c6d87215 100644
--- a/src/gallium/drivers/i915/i915_state_sampler.c
+++ b/src/gallium/drivers/i915/i915_state_sampler.c
@@ -27,7 +27,6 @@
 
 #include "pipe/p_context.h"
 #include "pipe/p_state.h"
-#include "util/u_memory.h"
 
 #include "i915_state_inlines.h"
 #include "i915_context.h"
diff --git a/src/gallium/drivers/i915/i915_surface.c b/src/gallium/drivers/i915/i915_surface.c
index c693eb30e8..1ff6b9f4c6 100644
--- a/src/gallium/drivers/i915/i915_surface.c
+++ b/src/gallium/drivers/i915/i915_surface.c
@@ -27,14 +27,8 @@
 
 #include "i915_context.h"
 #include "i915_blit.h"
-#include "i915_state.h"
 #include "pipe/p_defines.h"
-#include "pipe/p_inlines.h"
-#include "pipe/p_inlines.h"
-#include "pipe/internal/p_winsys_screen.h"
 #include "util/u_format.h"
-#include "util/u_tile.h"
-#include "util/u_rect.h"
 
 
 /* Assumes all values are within bounds -- no checking at this level -
diff --git a/src/gallium/drivers/i915/i915_texture.c b/src/gallium/drivers/i915/i915_texture.c
index 50a9e19094..612e5c1cdd 100644
--- a/src/gallium/drivers/i915/i915_texture.c
+++ b/src/gallium/drivers/i915/i915_texture.c
@@ -34,14 +34,12 @@
 #include "pipe/p_context.h"
 #include "pipe/p_defines.h"
 #include "pipe/p_inlines.h"
-#include "pipe/internal/p_winsys_screen.h"
 #include "util/u_format.h"
 #include "util/u_math.h"
 #include "util/u_memory.h"
 
 #include "i915_context.h"
 #include "i915_texture.h"
-#include "i915_debug.h"
 #include "i915_screen.h"
 #include "intel_winsys.h"
 
diff --git a/src/gallium/drivers/i965/brw_batchbuffer.c b/src/gallium/drivers/i965/brw_batchbuffer.c
index 22607dc608..003b1fd5bf 100644
--- a/src/gallium/drivers/i965/brw_batchbuffer.c
+++ b/src/gallium/drivers/i965/brw_batchbuffer.c
@@ -155,7 +155,7 @@ _brw_batchbuffer_flush(struct brw_batchbuffer *batch,
 enum pipe_error
 brw_batchbuffer_emit_reloc(struct brw_batchbuffer *batch,
 			   struct brw_winsys_buffer *buffer,
-			   uint32_t usage,
+			   enum brw_buffer_usage usage,
 			   uint32_t delta)
 {
    int ret;
diff --git a/src/gallium/drivers/i965/brw_batchbuffer.h b/src/gallium/drivers/i965/brw_batchbuffer.h
index 7473f5bea4..6ca9f617f5 100644
--- a/src/gallium/drivers/i965/brw_batchbuffer.h
+++ b/src/gallium/drivers/i965/brw_batchbuffer.h
@@ -64,12 +64,12 @@ brw_batchbuffer_reset(struct brw_batchbuffer *batch);
  * Consider it a convenience function wrapping multple
  * intel_buffer_dword() calls.
  */
-int brw_batchbuffer_data(struct brw_batchbuffer *batch,
+enum pipe_error brw_batchbuffer_data(struct brw_batchbuffer *batch,
                             const void *data, GLuint bytes,
 			    enum cliprect_mode cliprect_mode);
 
 
-int brw_batchbuffer_emit_reloc(struct brw_batchbuffer *batch,
+enum pipe_error brw_batchbuffer_emit_reloc(struct brw_batchbuffer *batch,
 			       struct brw_winsys_buffer *buffer,
 			       enum brw_buffer_usage usage,
 			       uint32_t offset);
diff --git a/src/gallium/drivers/i965/brw_cc.c b/src/gallium/drivers/i965/brw_cc.c
index 3e070f5591..4a543276f5 100644
--- a/src/gallium/drivers/i965/brw_cc.c
+++ b/src/gallium/drivers/i965/brw_cc.c
@@ -32,7 +32,6 @@
 
 #include "brw_context.h"
 #include "brw_state.h"
-#include "brw_defines.h"
 
 
 static enum pipe_error prepare_cc_vp( struct brw_context *brw )
diff --git a/src/gallium/drivers/i965/brw_clip.c b/src/gallium/drivers/i965/brw_clip.c
index d67a1a6263..ccba205e8c 100644
--- a/src/gallium/drivers/i965/brw_clip.c
+++ b/src/gallium/drivers/i965/brw_clip.c
@@ -38,7 +38,6 @@
 #include "brw_defines.h"
 #include "brw_context.h"
 #include "brw_eu.h"
-#include "brw_util.h"
 #include "brw_state.h"
 #include "brw_pipe_rast.h"
 #include "brw_clip.h"
diff --git a/src/gallium/drivers/i965/brw_clip_line.c b/src/gallium/drivers/i965/brw_clip_line.c
index 54282d975e..66caadc4d5 100644
--- a/src/gallium/drivers/i965/brw_clip_line.c
+++ b/src/gallium/drivers/i965/brw_clip_line.c
@@ -33,7 +33,6 @@
 
 #include "brw_defines.h"
 #include "brw_eu.h"
-#include "brw_util.h"
 #include "brw_clip.h"
 
 
diff --git a/src/gallium/drivers/i965/brw_clip_point.c b/src/gallium/drivers/i965/brw_clip_point.c
index e0a5330556..124156c1b5 100644
--- a/src/gallium/drivers/i965/brw_clip_point.c
+++ b/src/gallium/drivers/i965/brw_clip_point.c
@@ -31,7 +31,6 @@
 
 #include "brw_defines.h"
 #include "brw_eu.h"
-#include "brw_util.h"
 #include "brw_clip.h"
 
 
diff --git a/src/gallium/drivers/i965/brw_clip_tri.c b/src/gallium/drivers/i965/brw_clip_tri.c
index 4cde7294ea..069524bc14 100644
--- a/src/gallium/drivers/i965/brw_clip_tri.c
+++ b/src/gallium/drivers/i965/brw_clip_tri.c
@@ -31,7 +31,6 @@
 
 #include "brw_defines.h"
 #include "brw_eu.h"
-#include "brw_util.h"
 #include "brw_clip.h"
 
 static void release_tmps( struct brw_clip_compile *c )
diff --git a/src/gallium/drivers/i965/brw_clip_util.c b/src/gallium/drivers/i965/brw_clip_util.c
index 97a5710310..23e51ee9bc 100644
--- a/src/gallium/drivers/i965/brw_clip_util.c
+++ b/src/gallium/drivers/i965/brw_clip_util.c
@@ -32,7 +32,6 @@
 
 #include "brw_defines.h"
 #include "brw_eu.h"
-#include "brw_util.h"
 #include "brw_clip.h"
 
 
diff --git a/src/gallium/drivers/i965/brw_context.c b/src/gallium/drivers/i965/brw_context.c
index e67551882d..d60b7b99b6 100644
--- a/src/gallium/drivers/i965/brw_context.c
+++ b/src/gallium/drivers/i965/brw_context.c
@@ -34,7 +34,6 @@
 #include "util/u_simple_list.h"
 
 #include "brw_context.h"
-#include "brw_defines.h"
 #include "brw_draw.h"
 #include "brw_state.h"
 #include "brw_batchbuffer.h"
diff --git a/src/gallium/drivers/i965/brw_curbe.c b/src/gallium/drivers/i965/brw_curbe.c
index 3f031577d5..4b215a001c 100644
--- a/src/gallium/drivers/i965/brw_curbe.c
+++ b/src/gallium/drivers/i965/brw_curbe.c
@@ -36,9 +36,7 @@
 #include "brw_context.h"
 #include "brw_defines.h"
 #include "brw_state.h"
-#include "brw_util.h"
 #include "brw_debug.h"
-#include "brw_screen.h"
 
 
 /**
diff --git a/src/gallium/drivers/i965/brw_draw.c b/src/gallium/drivers/i965/brw_draw.c
index ea8d39adaf..1b5cd23995 100644
--- a/src/gallium/drivers/i965/brw_draw.c
+++ b/src/gallium/drivers/i965/brw_draw.c
@@ -34,7 +34,6 @@
 #include "brw_context.h"
 #include "brw_state.h"
 #include "brw_debug.h"
-#include "brw_screen.h"
 
 #include "brw_batchbuffer.h"
 
diff --git a/src/gallium/drivers/i965/brw_gs.c b/src/gallium/drivers/i965/brw_gs.c
index 921b201bae..06826635a8 100644
--- a/src/gallium/drivers/i965/brw_gs.c
+++ b/src/gallium/drivers/i965/brw_gs.c
@@ -34,7 +34,6 @@
 #include "brw_defines.h"
 #include "brw_context.h"
 #include "brw_eu.h"
-#include "brw_util.h"
 #include "brw_state.h"
 #include "brw_gs.h"
 
diff --git a/src/gallium/drivers/i965/brw_gs_emit.c b/src/gallium/drivers/i965/brw_gs_emit.c
index fd8e2acced..9b58773b3b 100644
--- a/src/gallium/drivers/i965/brw_gs_emit.c
+++ b/src/gallium/drivers/i965/brw_gs_emit.c
@@ -35,7 +35,6 @@
 #include "brw_defines.h"
 #include "brw_context.h"
 #include "brw_eu.h"
-#include "brw_util.h"
 #include "brw_gs.h"
 
 static void brw_gs_alloc_regs( struct brw_gs_compile *c,
diff --git a/src/gallium/drivers/i965/brw_pipe_blend.c b/src/gallium/drivers/i965/brw_pipe_blend.c
index b759a910b6..21f786f871 100644
--- a/src/gallium/drivers/i965/brw_pipe_blend.c
+++ b/src/gallium/drivers/i965/brw_pipe_blend.c
@@ -118,14 +118,14 @@ static void *brw_create_blend_state( struct pipe_context *pipe,
       blend->cc2.logicop_enable = 1;
       blend->cc5.logicop_func = translate_logicop(templ->logicop_func);
    } 
-   else if (templ->blend_enable) {
-      blend->cc6.dest_blend_factor = translate_blend_factor(templ->rgb_dst_factor);
-      blend->cc6.src_blend_factor = translate_blend_factor(templ->rgb_src_factor);
-      blend->cc6.blend_function = translate_blend_equation(templ->rgb_func);
+   else if (templ->rt[0].blend_enable) {
+      blend->cc6.dest_blend_factor = translate_blend_factor(templ->rt[0].rgb_dst_factor);
+      blend->cc6.src_blend_factor = translate_blend_factor(templ->rt[0].rgb_src_factor);
+      blend->cc6.blend_function = translate_blend_equation(templ->rt[0].rgb_func);
 
-      blend->cc5.ia_dest_blend_factor = translate_blend_factor(templ->alpha_dst_factor);
-      blend->cc5.ia_src_blend_factor = translate_blend_factor(templ->alpha_src_factor);
-      blend->cc5.ia_blend_function = translate_blend_equation(templ->alpha_func);
+      blend->cc5.ia_dest_blend_factor = translate_blend_factor(templ->rt[0].alpha_dst_factor);
+      blend->cc5.ia_src_blend_factor = translate_blend_factor(templ->rt[0].alpha_src_factor);
+      blend->cc5.ia_blend_function = translate_blend_equation(templ->rt[0].alpha_func);
 
       blend->cc3.blend_enable = 1;
       blend->cc3.ia_blend_enable = 
@@ -146,10 +146,10 @@ static void *brw_create_blend_state( struct pipe_context *pipe,
 
    /* Per-surface color mask -- just follow global state:
     */
-   blend->ss0.writedisable_red   = (templ->colormask & PIPE_MASK_R) ? 0 : 1;
-   blend->ss0.writedisable_green = (templ->colormask & PIPE_MASK_G) ? 0 : 1;
-   blend->ss0.writedisable_blue  = (templ->colormask & PIPE_MASK_B) ? 0 : 1;
-   blend->ss0.writedisable_alpha = (templ->colormask & PIPE_MASK_A) ? 0 : 1;
+   blend->ss0.writedisable_red   = (templ->rt[0].colormask & PIPE_MASK_R) ? 0 : 1;
+   blend->ss0.writedisable_green = (templ->rt[0].colormask & PIPE_MASK_G) ? 0 : 1;
+   blend->ss0.writedisable_blue  = (templ->rt[0].colormask & PIPE_MASK_B) ? 0 : 1;
+   blend->ss0.writedisable_alpha = (templ->rt[0].colormask & PIPE_MASK_A) ? 0 : 1;
 
    return (void *)blend;
 }
diff --git a/src/gallium/drivers/i965/brw_pipe_fb.c b/src/gallium/drivers/i965/brw_pipe_fb.c
index 5d4e5025f9..c1f049272a 100644
--- a/src/gallium/drivers/i965/brw_pipe_fb.c
+++ b/src/gallium/drivers/i965/brw_pipe_fb.c
@@ -3,7 +3,6 @@
 #include "pipe/p_state.h"
 
 #include "brw_context.h"
-#include "brw_debug.h"
 
 /**
  * called from intelDrawBuffer()
diff --git a/src/gallium/drivers/i965/brw_pipe_sampler.c b/src/gallium/drivers/i965/brw_pipe_sampler.c
index 81712798a5..ef6c1bb315 100644
--- a/src/gallium/drivers/i965/brw_pipe_sampler.c
+++ b/src/gallium/drivers/i965/brw_pipe_sampler.c
@@ -7,7 +7,6 @@
 
 #include "brw_context.h"
 #include "brw_defines.h"
-#include "brw_debug.h"
 
 
 
diff --git a/src/gallium/drivers/i965/brw_screen.c b/src/gallium/drivers/i965/brw_screen.c
index 0ecacac9a3..a8f9892d71 100644
--- a/src/gallium/drivers/i965/brw_screen.c
+++ b/src/gallium/drivers/i965/brw_screen.c
@@ -138,6 +138,9 @@ brw_get_name(struct pipe_screen *screen)
    case PCI_CHIP_ILM_G:
       chipset = "ILM_G";
       break;
+   default:
+      chipset = "unknown";
+      break;
    }
 
    util_snprintf(buffer, sizeof(buffer), "i965 (chipset: %s)", chipset);
@@ -172,6 +175,12 @@ brw_get_param(struct pipe_screen *screen, int param)
       return 8;  /* max 128x128x128 */
    case PIPE_CAP_MAX_TEXTURE_CUBE_LEVELS:
       return 11; /* max 1024x1024 */
+   case PIPE_CAP_TGSI_FS_COORD_ORIGIN_UPPER_LEFT:
+   case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_HALF_INTEGER:
+      return 1;
+   case PIPE_CAP_TGSI_FS_COORD_ORIGIN_LOWER_LEFT:
+   case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_INTEGER:
+      return 0;
    default:
       return 0;
    }
diff --git a/src/gallium/drivers/i965/brw_sf.c b/src/gallium/drivers/i965/brw_sf.c
index e1986a9dbb..6e99eaa09d 100644
--- a/src/gallium/drivers/i965/brw_sf.c
+++ b/src/gallium/drivers/i965/brw_sf.c
@@ -36,7 +36,6 @@
 #include "brw_context.h"
 #include "brw_pipe_rast.h"
 #include "brw_eu.h"
-#include "brw_util.h"
 #include "brw_sf.h"
 #include "brw_state.h"
 
diff --git a/src/gallium/drivers/i965/brw_sf_emit.c b/src/gallium/drivers/i965/brw_sf_emit.c
index 3b85725e36..497634ec9e 100644
--- a/src/gallium/drivers/i965/brw_sf_emit.c
+++ b/src/gallium/drivers/i965/brw_sf_emit.c
@@ -35,7 +35,6 @@
 #include "brw_defines.h"
 #include "brw_context.h"
 #include "brw_eu.h"
-#include "brw_util.h"
 #include "brw_sf.h"
 
 
diff --git a/src/gallium/drivers/i965/brw_state_cache.c b/src/gallium/drivers/i965/brw_state_cache.c
index 16b643ceb2..85c20076fb 100644
--- a/src/gallium/drivers/i965/brw_state_cache.c
+++ b/src/gallium/drivers/i965/brw_state_cache.c
@@ -59,7 +59,6 @@
 
 #include "brw_debug.h"
 #include "brw_state.h"
-#include "brw_batchbuffer.h"
 
 /* XXX: Fixme - have to include these to get the sizes of the prog_key
  * structs:
diff --git a/src/gallium/drivers/i965/brw_util.c b/src/gallium/drivers/i965/brw_util.c
index 458058d668..1fd2e29713 100644
--- a/src/gallium/drivers/i965/brw_util.c
+++ b/src/gallium/drivers/i965/brw_util.c
@@ -30,8 +30,6 @@
   */
          
 
-#include "brw_util.h"
-#include "brw_defines.h"
 
 
 
diff --git a/src/gallium/drivers/i965/brw_vs.c b/src/gallium/drivers/i965/brw_vs.c
index e3ea5a3a13..ca8ee79550 100644
--- a/src/gallium/drivers/i965/brw_vs.c
+++ b/src/gallium/drivers/i965/brw_vs.c
@@ -33,9 +33,7 @@
 
 #include "brw_context.h"
 #include "brw_vs.h"
-#include "brw_util.h"
 #include "brw_state.h"
-#include "brw_pipe_rast.h"
 
 
 
diff --git a/src/gallium/drivers/i965/brw_vs_surface_state.c b/src/gallium/drivers/i965/brw_vs_surface_state.c
index 177a5170d2..004e3cb4e6 100644
--- a/src/gallium/drivers/i965/brw_vs_surface_state.c
+++ b/src/gallium/drivers/i965/brw_vs_surface_state.c
@@ -31,7 +31,6 @@
 
 #include "brw_context.h"
 #include "brw_state.h"
-#include "brw_defines.h"
 #include "brw_winsys.h"
 
 /* XXX: disabled true constant buffer functionality
diff --git a/src/gallium/drivers/i965/brw_wm_fp.c b/src/gallium/drivers/i965/brw_wm_fp.c
index 9c5b527f89..9c67759ad0 100644
--- a/src/gallium/drivers/i965/brw_wm_fp.c
+++ b/src/gallium/drivers/i965/brw_wm_fp.c
@@ -41,7 +41,6 @@
 #include "tgsi/tgsi_util.h"
 
 #include "brw_wm.h"
-#include "brw_util.h"
 #include "brw_debug.h"
 
 
diff --git a/src/gallium/drivers/i965/brw_wm_surface_state.c b/src/gallium/drivers/i965/brw_wm_surface_state.c
index f92b8198ed..b01a7f194b 100644
--- a/src/gallium/drivers/i965/brw_wm_surface_state.c
+++ b/src/gallium/drivers/i965/brw_wm_surface_state.c
@@ -34,7 +34,6 @@
 #include "brw_batchbuffer.h"
 #include "brw_context.h"
 #include "brw_state.h"
-#include "brw_defines.h"
 #include "brw_screen.h"
 
 
diff --git a/src/gallium/drivers/llvmpipe/lp_bld_alpha.c b/src/gallium/drivers/llvmpipe/lp_bld_alpha.c
index 2b4bc5c819..7245730350 100644
--- a/src/gallium/drivers/llvmpipe/lp_bld_alpha.c
+++ b/src/gallium/drivers/llvmpipe/lp_bld_alpha.c
@@ -35,7 +35,6 @@
 
 #include "lp_bld_type.h"
 #include "lp_bld_const.h"
-#include "lp_bld_arit.h"
 #include "lp_bld_logic.h"
 #include "lp_bld_flow.h"
 #include "lp_bld_debug.h"
diff --git a/src/gallium/drivers/llvmpipe/lp_bld_blend_aos.c b/src/gallium/drivers/llvmpipe/lp_bld_blend_aos.c
index ced7b9c11d..ad1dbbc9b1 100644
--- a/src/gallium/drivers/llvmpipe/lp_bld_blend_aos.c
+++ b/src/gallium/drivers/llvmpipe/lp_bld_blend_aos.c
@@ -314,9 +314,10 @@ lp_build_blend_aos(LLVMBuilderRef builder,
    LLVMValueRef dst_term;
 
    /* FIXME */
-   assert(blend->colormask == 0xf);
+   assert(blend->independent_blend_enable == 0);
+   assert(blend->rt[0].colormask == 0xf);
 
-   if(!blend->blend_enable)
+   if(!blend->rt[0].blend_enable)
       return src;
 
    /* It makes no sense to blend unless values are normalized */
@@ -333,14 +334,16 @@ lp_build_blend_aos(LLVMBuilderRef builder,
     * combinations it is possible to reorder the operations and therefore saving
     * some instructions. */
 
-   src_term = lp_build_blend_factor(&bld, src, blend->rgb_src_factor, blend->alpha_src_factor, alpha_swizzle);
-   dst_term = lp_build_blend_factor(&bld, dst, blend->rgb_dst_factor, blend->alpha_dst_factor, alpha_swizzle);
+   src_term = lp_build_blend_factor(&bld, src, blend->rt[0].rgb_src_factor,
+                                    blend->rt[0].alpha_src_factor, alpha_swizzle);
+   dst_term = lp_build_blend_factor(&bld, dst, blend->rt[0].rgb_dst_factor,
+                                    blend->rt[0].alpha_dst_factor, alpha_swizzle);
 
    lp_build_name(src_term, "src_term");
    lp_build_name(dst_term, "dst_term");
 
-   if(blend->rgb_func == blend->alpha_func) {
-      return lp_build_blend_func(&bld.base, blend->rgb_func, src_term, dst_term);
+   if(blend->rt[0].rgb_func == blend->rt[0].alpha_func) {
+      return lp_build_blend_func(&bld.base, blend->rt[0].rgb_func, src_term, dst_term);
    }
    else {
       /* Seperate RGB / A functions */
@@ -348,8 +351,8 @@ lp_build_blend_aos(LLVMBuilderRef builder,
       LLVMValueRef rgb;
       LLVMValueRef alpha;
 
-      rgb   = lp_build_blend_func(&bld.base, blend->rgb_func,   src_term, dst_term);
-      alpha = lp_build_blend_func(&bld.base, blend->alpha_func, src_term, dst_term);
+      rgb   = lp_build_blend_func(&bld.base, blend->rt[0].rgb_func,   src_term, dst_term);
+      alpha = lp_build_blend_func(&bld.base, blend->rt[0].alpha_func, src_term, dst_term);
 
       return lp_build_blend_swizzle(&bld, rgb, alpha, LP_BUILD_BLEND_SWIZZLE_RGBA, alpha_swizzle);
    }
diff --git a/src/gallium/drivers/llvmpipe/lp_bld_blend_soa.c b/src/gallium/drivers/llvmpipe/lp_bld_blend_soa.c
index 9511299d55..a73d1158e7 100644
--- a/src/gallium/drivers/llvmpipe/lp_bld_blend_soa.c
+++ b/src/gallium/drivers/llvmpipe/lp_bld_blend_soa.c
@@ -71,7 +71,6 @@
 #include "pipe/p_state.h"
 
 #include "lp_bld_type.h"
-#include "lp_bld_const.h"
 #include "lp_bld_arit.h"
 #include "lp_bld_blend.h"
 
@@ -218,7 +217,7 @@ lp_build_blend_soa(LLVMBuilderRef builder,
    }
 
    for (i = 0; i < 4; ++i) {
-      if (blend->colormask & (1 << i)) {
+      if (blend->rt[0].colormask & (1 << i)) {
          if (blend->logicop_enable) {
             if(!type.floating) {
                res[i] = lp_build_logicop(builder, blend->logicop_func, src[i], dst[i]);
@@ -226,10 +225,10 @@ lp_build_blend_soa(LLVMBuilderRef builder,
             else
                res[i] = dst[i];
          }
-         else if (blend->blend_enable) {
-            unsigned src_factor = i < 3 ? blend->rgb_src_factor : blend->alpha_src_factor;
-            unsigned dst_factor = i < 3 ? blend->rgb_dst_factor : blend->alpha_dst_factor;
-            unsigned func = i < 3 ? blend->rgb_func : blend->alpha_func;
+         else if (blend->rt[0].blend_enable) {
+            unsigned src_factor = i < 3 ? blend->rt[0].rgb_src_factor : blend->rt[0].alpha_src_factor;
+            unsigned dst_factor = i < 3 ? blend->rt[0].rgb_dst_factor : blend->rt[0].alpha_dst_factor;
+            unsigned func = i < 3 ? blend->rt[0].rgb_func : blend->rt[0].alpha_func;
             boolean func_commutative = lp_build_blend_func_commutative(func);
 
             /* It makes no sense to blend unless values are normalized */
@@ -270,7 +269,7 @@ lp_build_blend_soa(LLVMBuilderRef builder,
 
             /* See if this function has been previously applied */
             for(j = 0; j < i; ++j) {
-               unsigned prev_func = j < 3 ? blend->rgb_func : blend->alpha_func;
+               unsigned prev_func = j < 3 ? blend->rt[0].rgb_func : blend->rt[0].alpha_func;
                unsigned func_reverse = lp_build_blend_func_reverse(func, prev_func);
 
                if((!func_reverse &&
diff --git a/src/gallium/drivers/llvmpipe/lp_bld_conv.c b/src/gallium/drivers/llvmpipe/lp_bld_conv.c
index 9fa8837202..f77cf78721 100644
--- a/src/gallium/drivers/llvmpipe/lp_bld_conv.c
+++ b/src/gallium/drivers/llvmpipe/lp_bld_conv.c
@@ -63,11 +63,9 @@
 
 #include "util/u_debug.h"
 #include "util/u_math.h"
-#include "util/u_cpu_detect.h"
 
 #include "lp_bld_type.h"
 #include "lp_bld_const.h"
-#include "lp_bld_intr.h"
 #include "lp_bld_arit.h"
 #include "lp_bld_pack.h"
 #include "lp_bld_conv.h"
diff --git a/src/gallium/drivers/llvmpipe/lp_bld_format_aos.c b/src/gallium/drivers/llvmpipe/lp_bld_format_aos.c
index 10e82f120b..dfa080b853 100644
--- a/src/gallium/drivers/llvmpipe/lp_bld_format_aos.c
+++ b/src/gallium/drivers/llvmpipe/lp_bld_format_aos.c
@@ -38,7 +38,6 @@
 
 #include "lp_bld_type.h"
 #include "lp_bld_const.h"
-#include "lp_bld_logic.h"
 #include "lp_bld_swizzle.h"
 #include "lp_bld_format.h"
 
diff --git a/src/gallium/drivers/llvmpipe/lp_bld_tgsi_soa.c b/src/gallium/drivers/llvmpipe/lp_bld_tgsi_soa.c
index fb1eda4423..85e3b1bdd4 100644
--- a/src/gallium/drivers/llvmpipe/lp_bld_tgsi_soa.c
+++ b/src/gallium/drivers/llvmpipe/lp_bld_tgsi_soa.c
@@ -47,13 +47,11 @@
 #include "tgsi/tgsi_exec.h"
 #include "lp_bld_type.h"
 #include "lp_bld_const.h"
-#include "lp_bld_intr.h"
 #include "lp_bld_arit.h"
 #include "lp_bld_logic.h"
 #include "lp_bld_swizzle.h"
 #include "lp_bld_flow.h"
 #include "lp_bld_tgsi.h"
-#include "lp_bld_debug.h"
 
 
 #define LP_MAX_TEMPS 256
diff --git a/src/gallium/drivers/llvmpipe/lp_draw_arrays.c b/src/gallium/drivers/llvmpipe/lp_draw_arrays.c
index 3989cce744..3dd68d5794 100644
--- a/src/gallium/drivers/llvmpipe/lp_draw_arrays.c
+++ b/src/gallium/drivers/llvmpipe/lp_draw_arrays.c
@@ -33,8 +33,6 @@
 
 #include "pipe/p_defines.h"
 #include "pipe/p_context.h"
-#include "pipe/internal/p_winsys_screen.h"
-#include "pipe/p_inlines.h"
 #include "util/u_prim.h"
 
 #include "lp_buffer.h"
diff --git a/src/gallium/drivers/llvmpipe/lp_flush.c b/src/gallium/drivers/llvmpipe/lp_flush.c
index 07f32848c7..edd480d7ed 100644
--- a/src/gallium/drivers/llvmpipe/lp_flush.c
+++ b/src/gallium/drivers/llvmpipe/lp_flush.c
@@ -35,7 +35,6 @@
 #include "lp_flush.h"
 #include "lp_context.h"
 #include "lp_surface.h"
-#include "lp_winsys.h"
 #include "lp_setup.h"
 
 
diff --git a/src/gallium/drivers/llvmpipe/lp_jit.c b/src/gallium/drivers/llvmpipe/lp_jit.c
index 9fad7033db..310fc2b847 100644
--- a/src/gallium/drivers/llvmpipe/lp_jit.c
+++ b/src/gallium/drivers/llvmpipe/lp_jit.c
@@ -40,7 +40,6 @@
 #include "lp_debug.h"
 #include "lp_screen.h"
 #include "lp_bld_intr.h"
-#include "lp_bld_misc.h"
 #include "lp_jit.h"
 
 
diff --git a/src/gallium/drivers/llvmpipe/lp_screen.c b/src/gallium/drivers/llvmpipe/lp_screen.c
index a3adc81e9f..ca64c41827 100644
--- a/src/gallium/drivers/llvmpipe/lp_screen.c
+++ b/src/gallium/drivers/llvmpipe/lp_screen.c
@@ -115,6 +115,16 @@ llvmpipe_get_param(struct pipe_screen *screen, int param)
       return 1;
    case PIPE_CAP_BLEND_EQUATION_SEPARATE:
       return 1;
+   case PIPE_CAP_INDEP_BLEND_ENABLE:
+      return 0;
+   case PIPE_CAP_INDEP_BLEND_FUNC:
+      return 0;
+   case PIPE_CAP_TGSI_FS_COORD_ORIGIN_UPPER_LEFT:
+   case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_INTEGER:
+      return 1;
+   case PIPE_CAP_TGSI_FS_COORD_ORIGIN_LOWER_LEFT:
+   case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_HALF_INTEGER:
+      return 0;
    default:
       return 0;
    }
diff --git a/src/gallium/drivers/llvmpipe/lp_state_fs.c b/src/gallium/drivers/llvmpipe/lp_state_fs.c
index a7514ee011..0602e940d9 100644
--- a/src/gallium/drivers/llvmpipe/lp_state_fs.c
+++ b/src/gallium/drivers/llvmpipe/lp_state_fs.c
@@ -66,7 +66,6 @@
 #include "util/u_format.h"
 #include "util/u_debug_dump.h"
 #include "util/u_time.h"
-#include "pipe/internal/p_winsys_screen.h"
 #include "pipe/p_shader_tokens.h"
 #include "draw/draw_context.h"
 #include "tgsi/tgsi_dump.h"
@@ -568,7 +567,7 @@ generate_blend(const struct pipe_blend_state *blend,
    lp_build_blend_soa(builder, blend, type, src, dst, con, res);
 
    for(chan = 0; chan < 4; ++chan) {
-      if(blend->colormask & (1 << chan)) {
+      if(blend->rt[0].colormask & (1 << chan)) {
          LLVMValueRef index = LLVMConstInt(LLVMInt32Type(), chan, 0);
          lp_build_name(res[chan], "res.%c", "rgba"[chan]);
          res[chan] = lp_build_select(&bld, mask, res[chan], dst[chan]);
@@ -871,15 +870,15 @@ generate_variant(struct llvmpipe_context *lp,
       if(key->blend.logicop_enable) {
          debug_printf("blend.logicop_func = %u\n", key->blend.logicop_func);
       }
-      else if(key->blend.blend_enable) {
-         debug_printf("blend.rgb_func = %s\n",   debug_dump_blend_func  (key->blend.rgb_func, TRUE));
-         debug_printf("rgb_src_factor = %s\n",   debug_dump_blend_factor(key->blend.rgb_src_factor, TRUE));
-         debug_printf("rgb_dst_factor = %s\n",   debug_dump_blend_factor(key->blend.rgb_dst_factor, TRUE));
-         debug_printf("alpha_func = %s\n",       debug_dump_blend_func  (key->blend.alpha_func, TRUE));
-         debug_printf("alpha_src_factor = %s\n", debug_dump_blend_factor(key->blend.alpha_src_factor, TRUE));
-         debug_printf("alpha_dst_factor = %s\n", debug_dump_blend_factor(key->blend.alpha_dst_factor, TRUE));
+      else if(key->blend.rt[0].blend_enable) {
+         debug_printf("blend.rgb_func = %s\n",   debug_dump_blend_func  (key->blend.rt[0].rgb_func, TRUE));
+         debug_printf("rgb_src_factor = %s\n",   debug_dump_blend_factor(key->blend.rt[0].rgb_src_factor, TRUE));
+         debug_printf("rgb_dst_factor = %s\n",   debug_dump_blend_factor(key->blend.rt[0].rgb_dst_factor, TRUE));
+         debug_printf("alpha_func = %s\n",       debug_dump_blend_func  (key->blend.rt[0].alpha_func, TRUE));
+         debug_printf("alpha_src_factor = %s\n", debug_dump_blend_factor(key->blend.rt[0].alpha_src_factor, TRUE));
+         debug_printf("alpha_dst_factor = %s\n", debug_dump_blend_factor(key->blend.rt[0].alpha_dst_factor, TRUE));
       }
-      debug_printf("blend.colormask = 0x%x\n", key->blend.colormask);
+      debug_printf("blend.colormask = 0x%x\n", key->blend.rt[0].colormask);
       for(i = 0; i < PIPE_MAX_SAMPLERS; ++i) {
          if(key->sampler[i].format) {
             debug_printf("sampler[%u] = \n", i);
@@ -1020,7 +1019,7 @@ llvmpipe_set_constant_buffer(struct pipe_context *pipe,
    pipe_buffer_reference(&llvmpipe->constants[shader], constants);
 
    if(shader == PIPE_SHADER_VERTEX) {
-      draw_set_mapped_constant_buffer(llvmpipe->draw, PIPE_SHADER_VERTEX,
+      draw_set_mapped_constant_buffer(llvmpipe->draw, PIPE_SHADER_VERTEX, 0,
                                       data, size);
    }
 
@@ -1078,7 +1077,7 @@ make_variant_key(struct llvmpipe_context *lp,
          enum util_format_swizzle swizzle = format_desc->swizzle[chan];
 
          if(swizzle <= UTIL_FORMAT_SWIZZLE_W)
-            key->cbuf_blend[i].colormask |= (1 << chan);
+            key->blend.rt[0].colormask |= (1 << chan);
       }
    }
 
@@ -1128,8 +1127,8 @@ llvmpipe_update_fs(struct llvmpipe_context *lp)
    /* TODO: put this in the variant */
    /* TODO: most of these can be relaxed, in particular the colormask */
    opaque = !key.blend.logicop_enable &&
-            !key.blend.blend_enable &&
-            key.blend.colormask == 0xf &&
+            !key.blend.rt[0].blend_enable &&
+            key.blend.rt[0].colormask == 0xf &&
             !key.alpha.enabled &&
             !key.depth.enabled &&
             !key.scissor &&
diff --git a/src/gallium/drivers/llvmpipe/lp_state_surface.c b/src/gallium/drivers/llvmpipe/lp_state_surface.c
index 0e9f03b90b..aa4241a80d 100644
--- a/src/gallium/drivers/llvmpipe/lp_state_surface.c
+++ b/src/gallium/drivers/llvmpipe/lp_state_surface.c
@@ -32,7 +32,6 @@
 #include "util/u_surface.h"
 #include "lp_context.h"
 #include "lp_state.h"
-#include "lp_surface.h"
 #include "lp_setup.h"
 
 #include "draw/draw_context.h"
diff --git a/src/gallium/drivers/llvmpipe/lp_state_vertex.c b/src/gallium/drivers/llvmpipe/lp_state_vertex.c
index 1a17631a4c..57ac25ea0c 100644
--- a/src/gallium/drivers/llvmpipe/lp_state_vertex.c
+++ b/src/gallium/drivers/llvmpipe/lp_state_vertex.c
@@ -31,7 +31,6 @@
 
 #include "lp_context.h"
 #include "lp_state.h"
-#include "lp_surface.h"
 
 #include "draw/draw_context.h"
 
diff --git a/src/gallium/drivers/llvmpipe/lp_test_blend.c b/src/gallium/drivers/llvmpipe/lp_test_blend.c
index 6c29e8d8ac..ee72f6ce4f 100644
--- a/src/gallium/drivers/llvmpipe/lp_test_blend.c
+++ b/src/gallium/drivers/llvmpipe/lp_test_blend.c
@@ -38,7 +38,6 @@
 
 
 #include "lp_bld_type.h"
-#include "lp_bld_arit.h"
 #include "lp_bld_blend.h"
 #include "lp_bld_debug.h"
 #include "lp_test.h"
@@ -104,18 +103,18 @@ write_tsv_row(FILE *fp,
 
    fprintf(fp,
            "%s\t%s\t%s\t",
-           blend->rgb_func != blend->alpha_func ? "true" : "false",
-           blend->rgb_src_factor != blend->alpha_src_factor ? "true" : "false",
-           blend->rgb_dst_factor != blend->alpha_dst_factor ? "true" : "false");
+           blend->rt[0].rgb_func != blend->rt[0].alpha_func ? "true" : "false",
+           blend->rt[0].rgb_src_factor != blend->rt[0].alpha_src_factor ? "true" : "false",
+           blend->rt[0].rgb_dst_factor != blend->rt[0].alpha_dst_factor ? "true" : "false");
 
    fprintf(fp,
            "%s\t%s\t%s\t%s\t%s\t%s\n",
-           debug_dump_blend_func(blend->rgb_func, TRUE),
-           debug_dump_blend_factor(blend->rgb_src_factor, TRUE),
-           debug_dump_blend_factor(blend->rgb_dst_factor, TRUE),
-           debug_dump_blend_func(blend->alpha_func, TRUE),
-           debug_dump_blend_factor(blend->alpha_src_factor, TRUE),
-           debug_dump_blend_factor(blend->alpha_dst_factor, TRUE));
+           debug_dump_blend_func(blend->rt[0].rgb_func, TRUE),
+           debug_dump_blend_factor(blend->rt[0].rgb_src_factor, TRUE),
+           debug_dump_blend_factor(blend->rt[0].rgb_dst_factor, TRUE),
+           debug_dump_blend_func(blend->rt[0].alpha_func, TRUE),
+           debug_dump_blend_factor(blend->rt[0].alpha_src_factor, TRUE),
+           debug_dump_blend_factor(blend->rt[0].alpha_dst_factor, TRUE));
 
    fflush(fp);
 }
@@ -137,12 +136,12 @@ dump_blend_type(FILE *fp,
 
    fprintf(fp,
            " %s=%s %s=%s %s=%s %s=%s %s=%s %s=%s",
-           "rgb_func",         debug_dump_blend_func(blend->rgb_func, TRUE),
-           "rgb_src_factor",   debug_dump_blend_factor(blend->rgb_src_factor, TRUE),
-           "rgb_dst_factor",   debug_dump_blend_factor(blend->rgb_dst_factor, TRUE),
-           "alpha_func",       debug_dump_blend_func(blend->alpha_func, TRUE),
-           "alpha_src_factor", debug_dump_blend_factor(blend->alpha_src_factor, TRUE),
-           "alpha_dst_factor", debug_dump_blend_factor(blend->alpha_dst_factor, TRUE));
+           "rgb_func",         debug_dump_blend_func(blend->rt[0].rgb_func, TRUE),
+           "rgb_src_factor",   debug_dump_blend_factor(blend->rt[0].rgb_src_factor, TRUE),
+           "rgb_dst_factor",   debug_dump_blend_factor(blend->rt[0].rgb_dst_factor, TRUE),
+           "alpha_func",       debug_dump_blend_func(blend->rt[0].alpha_func, TRUE),
+           "alpha_src_factor", debug_dump_blend_factor(blend->rt[0].alpha_src_factor, TRUE),
+           "alpha_dst_factor", debug_dump_blend_factor(blend->rt[0].alpha_dst_factor, TRUE));
 
    fprintf(fp, " ...\n");
    fflush(fp);
@@ -401,13 +400,15 @@ compute_blend_ref(const struct pipe_blend_state *blend,
    double src_term[4];
    double dst_term[4];
 
-   compute_blend_ref_term(blend->rgb_src_factor, blend->alpha_src_factor, src, src, dst, con, src_term);
-   compute_blend_ref_term(blend->rgb_dst_factor, blend->alpha_dst_factor, dst, src, dst, con, dst_term);
+   compute_blend_ref_term(blend->rt[0].rgb_src_factor, blend->rt[0].alpha_src_factor,
+                          src, src, dst, con, src_term);
+   compute_blend_ref_term(blend->rt[0].rgb_dst_factor, blend->rt[0].alpha_dst_factor,
+                          dst, src, dst, con, dst_term);
 
    /*
     * Combine RGB terms
     */
-   switch (blend->rgb_func) {
+   switch (blend->rt[0].rgb_func) {
    case PIPE_BLEND_ADD:
       ADD_SAT(res[0], src_term[0], dst_term[0]); /* R */
       ADD_SAT(res[1], src_term[1], dst_term[1]); /* G */
@@ -440,7 +441,7 @@ compute_blend_ref(const struct pipe_blend_state *blend,
    /*
     * Combine A terms
     */
-   switch (blend->alpha_func) {
+   switch (blend->rt[0].alpha_func) {
    case PIPE_BLEND_ADD:
       ADD_SAT(res[3], src_term[3], dst_term[3]); /* A */
       break;
@@ -806,14 +807,14 @@ test_all(unsigned verbose, FILE *fp)
                               continue;
 
                            memset(&blend, 0, sizeof blend);
-                           blend.blend_enable      = 1;
-                           blend.rgb_func          = *rgb_func;
-                           blend.rgb_src_factor    = *rgb_src_factor;
-                           blend.rgb_dst_factor    = *rgb_dst_factor;
-                           blend.alpha_func        = *alpha_func;
-                           blend.alpha_src_factor  = *alpha_src_factor;
-                           blend.alpha_dst_factor  = *alpha_dst_factor;
-                           blend.colormask         = PIPE_MASK_RGBA;
+                           blend.rt[0].blend_enable      = 1;
+                           blend.rt[0].rgb_func          = *rgb_func;
+                           blend.rt[0].rgb_src_factor    = *rgb_src_factor;
+                           blend.rt[0].rgb_dst_factor    = *rgb_dst_factor;
+                           blend.rt[0].alpha_func        = *alpha_func;
+                           blend.rt[0].alpha_src_factor  = *alpha_src_factor;
+                           blend.rt[0].alpha_dst_factor  = *alpha_dst_factor;
+                           blend.rt[0].colormask         = PIPE_MASK_RGBA;
 
                            if(!test_one(verbose, fp, &blend, mode, *type))
                              success = FALSE;
@@ -865,14 +866,14 @@ test_some(unsigned verbose, FILE *fp, unsigned long n)
       type = &blend_types[rand() % num_types];
 
       memset(&blend, 0, sizeof blend);
-      blend.blend_enable      = 1;
-      blend.rgb_func          = *rgb_func;
-      blend.rgb_src_factor    = *rgb_src_factor;
-      blend.rgb_dst_factor    = *rgb_dst_factor;
-      blend.alpha_func        = *alpha_func;
-      blend.alpha_src_factor  = *alpha_src_factor;
-      blend.alpha_dst_factor  = *alpha_dst_factor;
-      blend.colormask         = PIPE_MASK_RGBA;
+      blend.rt[0].blend_enable      = 1;
+      blend.rt[0].rgb_func          = *rgb_func;
+      blend.rt[0].rgb_src_factor    = *rgb_src_factor;
+      blend.rt[0].rgb_dst_factor    = *rgb_dst_factor;
+      blend.rt[0].alpha_func        = *alpha_func;
+      blend.rt[0].alpha_src_factor  = *alpha_src_factor;
+      blend.rt[0].alpha_dst_factor  = *alpha_dst_factor;
+      blend.rt[0].colormask         = PIPE_MASK_RGBA;
 
       if(!test_one(verbose, fp, &blend, mode, *type))
         success = FALSE;
diff --git a/src/gallium/drivers/llvmpipe/lp_tex_sample_llvm.c b/src/gallium/drivers/llvmpipe/lp_tex_sample_llvm.c
index d59d768139..7f55f1ae83 100644
--- a/src/gallium/drivers/llvmpipe/lp_tex_sample_llvm.c
+++ b/src/gallium/drivers/llvmpipe/lp_tex_sample_llvm.c
@@ -44,7 +44,6 @@
 #include "pipe/p_shader_tokens.h"
 #include "lp_bld_debug.h"
 #include "lp_bld_type.h"
-#include "lp_bld_intr.h"
 #include "lp_bld_sample.h"
 #include "lp_bld_tgsi.h"
 #include "lp_jit.h"
diff --git a/src/gallium/drivers/llvmpipe/lp_texture.c b/src/gallium/drivers/llvmpipe/lp_texture.c
index 36e2ebb41a..c9b6eb180f 100644
--- a/src/gallium/drivers/llvmpipe/lp_texture.c
+++ b/src/gallium/drivers/llvmpipe/lp_texture.c
@@ -33,7 +33,6 @@
 #include "pipe/p_context.h"
 #include "pipe/p_defines.h"
 #include "pipe/p_inlines.h"
-#include "pipe/internal/p_winsys_screen.h"
 
 #include "util/u_format.h"
 #include "util/u_math.h"
diff --git a/src/gallium/drivers/nouveau/nouveau_screen.c b/src/gallium/drivers/nouveau/nouveau_screen.c
index 7ebc94ed6c..1ad539d285 100644
--- a/src/gallium/drivers/nouveau/nouveau_screen.c
+++ b/src/gallium/drivers/nouveau/nouveau_screen.c
@@ -260,6 +260,8 @@ nouveau_screen_init(struct nouveau_screen *screen, struct nouveau_device *dev)
 void
 nouveau_screen_fini(struct nouveau_screen *screen)
 {
+	struct pipe_winsys *ws = screen->base.winsys;
 	nouveau_channel_free(&screen->channel);
+	ws->destroy(ws);
 }
 
diff --git a/src/gallium/drivers/nv04/nv04_screen.c b/src/gallium/drivers/nv04/nv04_screen.c
index 7c5b6e8229..97538964af 100644
--- a/src/gallium/drivers/nv04/nv04_screen.c
+++ b/src/gallium/drivers/nv04/nv04_screen.c
@@ -45,6 +45,16 @@ nv04_screen_get_param(struct pipe_screen *screen, int param)
 	case NOUVEAU_CAP_HW_VTXBUF:
 	case NOUVEAU_CAP_HW_IDXBUF:
 		return 0;
+	case PIPE_CAP_INDEP_BLEND_ENABLE:
+		return 0;
+	case PIPE_CAP_INDEP_BLEND_FUNC:
+		return 0;
+	case PIPE_CAP_TGSI_FS_COORD_ORIGIN_LOWER_LEFT:
+	case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_HALF_INTEGER:
+		return 1;
+	case PIPE_CAP_TGSI_FS_COORD_ORIGIN_UPPER_LEFT:
+	case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_INTEGER:
+		return 0;
 	default:
 		NOUVEAU_ERR("Unknown PIPE_CAP %d\n", param);
 		return 0;
diff --git a/src/gallium/drivers/nv04/nv04_state.c b/src/gallium/drivers/nv04/nv04_state.c
index b67f1e16b1..1fdb764b53 100644
--- a/src/gallium/drivers/nv04/nv04_state.c
+++ b/src/gallium/drivers/nv04/nv04_state.c
@@ -17,11 +17,11 @@ nv04_blend_state_create(struct pipe_context *pipe,
 
 	cb = MALLOC(sizeof(struct nv04_blend_state));
 
-	cb->b_enable = cso->blend_enable ? 1 : 0;
-	cb->b_src = ((nvgl_blend_func(cso->alpha_src_factor)<<16) |
-			 (nvgl_blend_func(cso->rgb_src_factor)));
-	cb->b_dst = ((nvgl_blend_func(cso->alpha_dst_factor)<<16) |
-			 (nvgl_blend_func(cso->rgb_dst_factor)));
+	cb->b_enable = cso->rt[0].blend_enable ? 1 : 0;
+	cb->b_src = ((nvgl_blend_func(cso->rt[0].alpha_src_factor)<<16) |
+			 (nvgl_blend_func(cso->rt[0].rgb_src_factor)));
+	cb->b_dst = ((nvgl_blend_func(cso->rt[0].alpha_dst_factor)<<16) |
+			 (nvgl_blend_func(cso->rt[0].rgb_dst_factor)));
 	
 
 	return (void *)cb;
diff --git a/src/gallium/drivers/nv04/nv04_vbo.c b/src/gallium/drivers/nv04/nv04_vbo.c
index 3484771814..2db61d8941 100644
--- a/src/gallium/drivers/nv04/nv04_vbo.c
+++ b/src/gallium/drivers/nv04/nv04_vbo.c
@@ -45,7 +45,7 @@ void nv04_draw_elements( struct pipe_context *pipe,
 		draw_set_mapped_element_buffer(draw, 0, NULL);
 	}
 
-	draw_set_mapped_constant_buffer(draw, PIPE_SHADER_VERTEX,
+	draw_set_mapped_constant_buffer(draw, PIPE_SHADER_VERTEX, 0,
 					nv04->constbuf[PIPE_SHADER_VERTEX],
 					nv04->constbuf_nr[PIPE_SHADER_VERTEX]);
 
diff --git a/src/gallium/drivers/nv10/nv10_screen.c b/src/gallium/drivers/nv10/nv10_screen.c
index 69a6dab866..4e6bc12994 100644
--- a/src/gallium/drivers/nv10/nv10_screen.c
+++ b/src/gallium/drivers/nv10/nv10_screen.c
@@ -40,6 +40,16 @@ nv10_screen_get_param(struct pipe_screen *screen, int param)
 	case NOUVEAU_CAP_HW_VTXBUF:
 	case NOUVEAU_CAP_HW_IDXBUF:
 		return 0;
+	case PIPE_CAP_INDEP_BLEND_ENABLE:
+		return 0;
+	case PIPE_CAP_INDEP_BLEND_FUNC:
+		return 0;
+	case PIPE_CAP_TGSI_FS_COORD_ORIGIN_LOWER_LEFT:
+	case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_HALF_INTEGER:
+		return 1;
+	case PIPE_CAP_TGSI_FS_COORD_ORIGIN_UPPER_LEFT:
+	case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_INTEGER:
+		return 0;
 	default:
 		NOUVEAU_ERR("Unknown PIPE_CAP %d\n", param);
 		return 0;
diff --git a/src/gallium/drivers/nv10/nv10_state.c b/src/gallium/drivers/nv10/nv10_state.c
index ad7def53b1..9aafc60411 100644
--- a/src/gallium/drivers/nv10/nv10_state.c
+++ b/src/gallium/drivers/nv10/nv10_state.c
@@ -17,16 +17,16 @@ nv10_blend_state_create(struct pipe_context *pipe,
 
 	cb = MALLOC(sizeof(struct nv10_blend_state));
 
-	cb->b_enable = cso->blend_enable ? 1 : 0;
-	cb->b_srcfunc = ((nvgl_blend_func(cso->alpha_src_factor)<<16) |
-			 (nvgl_blend_func(cso->rgb_src_factor)));
-	cb->b_dstfunc = ((nvgl_blend_func(cso->alpha_dst_factor)<<16) |
-			 (nvgl_blend_func(cso->rgb_dst_factor)));
-
-	cb->c_mask = (((cso->colormask & PIPE_MASK_A) ? (0x01<<24) : 0) |
-		      ((cso->colormask & PIPE_MASK_R) ? (0x01<<16) : 0) |
-		      ((cso->colormask & PIPE_MASK_G) ? (0x01<< 8) : 0) |
-		      ((cso->colormask & PIPE_MASK_B) ? (0x01<< 0) : 0));
+	cb->b_enable = cso->rt[0].blend_enable ? 1 : 0;
+	cb->b_srcfunc = ((nvgl_blend_func(cso->rt[0].alpha_src_factor)<<16) |
+			 (nvgl_blend_func(cso->rt[0].rgb_src_factor)));
+	cb->b_dstfunc = ((nvgl_blend_func(cso->rt[0].alpha_dst_factor)<<16) |
+			 (nvgl_blend_func(cso->rt[0].rgb_dst_factor)));
+
+	cb->c_mask = (((cso->rt[0].colormask & PIPE_MASK_A) ? (0x01<<24) : 0) |
+		      ((cso->rt[0].colormask & PIPE_MASK_R) ? (0x01<<16) : 0) |
+		      ((cso->rt[0].colormask & PIPE_MASK_G) ? (0x01<< 8) : 0) |
+		      ((cso->rt[0].colormask & PIPE_MASK_B) ? (0x01<< 0) : 0));
 
 	cb->d_enable = cso->dither ? 1 : 0;
 
diff --git a/src/gallium/drivers/nv10/nv10_vbo.c b/src/gallium/drivers/nv10/nv10_vbo.c
index 9180c72c9b..da5c93f081 100644
--- a/src/gallium/drivers/nv10/nv10_vbo.c
+++ b/src/gallium/drivers/nv10/nv10_vbo.c
@@ -46,6 +46,7 @@ void nv10_draw_elements( struct pipe_context *pipe,
 
 	draw_set_mapped_constant_buffer(draw,
                                         PIPE_SHADER_VERTEX,
+                                        0,
 					nv10->constbuf[PIPE_SHADER_VERTEX],
 					nv10->constbuf_nr[PIPE_SHADER_VERTEX]);
 
diff --git a/src/gallium/drivers/nv20/nv20_screen.c b/src/gallium/drivers/nv20/nv20_screen.c
index d091335063..df9401bcce 100644
--- a/src/gallium/drivers/nv20/nv20_screen.c
+++ b/src/gallium/drivers/nv20/nv20_screen.c
@@ -40,6 +40,16 @@ nv20_screen_get_param(struct pipe_screen *screen, int param)
 	case NOUVEAU_CAP_HW_VTXBUF:
 	case NOUVEAU_CAP_HW_IDXBUF:
 		return 0;
+	case PIPE_CAP_INDEP_BLEND_ENABLE:
+		return 0;
+	case PIPE_CAP_INDEP_BLEND_FUNC:
+		return 0;
+	case PIPE_CAP_TGSI_FS_COORD_ORIGIN_LOWER_LEFT:
+	case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_HALF_INTEGER:
+		return 1;
+	case PIPE_CAP_TGSI_FS_COORD_ORIGIN_UPPER_LEFT:
+	case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_INTEGER:
+		return 0;
 	default:
 		NOUVEAU_ERR("Unknown PIPE_CAP %d\n", param);
 		return 0;
diff --git a/src/gallium/drivers/nv20/nv20_state.c b/src/gallium/drivers/nv20/nv20_state.c
index 45697a60ef..83335c790a 100644
--- a/src/gallium/drivers/nv20/nv20_state.c
+++ b/src/gallium/drivers/nv20/nv20_state.c
@@ -17,16 +17,16 @@ nv20_blend_state_create(struct pipe_context *pipe,
 
 	cb = MALLOC(sizeof(struct nv20_blend_state));
 
-	cb->b_enable = cso->blend_enable ? 1 : 0;
-	cb->b_srcfunc = ((nvgl_blend_func(cso->alpha_src_factor)<<16) |
-			 (nvgl_blend_func(cso->rgb_src_factor)));
-	cb->b_dstfunc = ((nvgl_blend_func(cso->alpha_dst_factor)<<16) |
-			 (nvgl_blend_func(cso->rgb_dst_factor)));
-
-	cb->c_mask = (((cso->colormask & PIPE_MASK_A) ? (0x01<<24) : 0) |
-		      ((cso->colormask & PIPE_MASK_R) ? (0x01<<16) : 0) |
-		      ((cso->colormask & PIPE_MASK_G) ? (0x01<< 8) : 0) |
-		      ((cso->colormask & PIPE_MASK_B) ? (0x01<< 0) : 0));
+	cb->b_enable = cso->rt[0].blend_enable ? 1 : 0;
+	cb->b_srcfunc = ((nvgl_blend_func(cso->rt[0].alpha_src_factor)<<16) |
+			 (nvgl_blend_func(cso->rt[0].rgb_src_factor)));
+	cb->b_dstfunc = ((nvgl_blend_func(cso->rt[0].alpha_dst_factor)<<16) |
+			 (nvgl_blend_func(cso->rt[0].rgb_dst_factor)));
+
+	cb->c_mask = (((cso->rt[0].colormask & PIPE_MASK_A) ? (0x01<<24) : 0) |
+		      ((cso->rt[0].colormask & PIPE_MASK_R) ? (0x01<<16) : 0) |
+		      ((cso->rt[0].colormask & PIPE_MASK_G) ? (0x01<< 8) : 0) |
+		      ((cso->rt[0].colormask & PIPE_MASK_B) ? (0x01<< 0) : 0));
 
 	cb->d_enable = cso->dither ? 1 : 0;
 
diff --git a/src/gallium/drivers/nv20/nv20_vbo.c b/src/gallium/drivers/nv20/nv20_vbo.c
index 52991a0d85..6dc9538901 100644
--- a/src/gallium/drivers/nv20/nv20_vbo.c
+++ b/src/gallium/drivers/nv20/nv20_vbo.c
@@ -45,7 +45,7 @@ void nv20_draw_elements( struct pipe_context *pipe,
 		draw_set_mapped_element_buffer(draw, 0, NULL);
 	}
 
-	draw_set_mapped_constant_buffer(draw, PIPE_SHADER_VERTEX,
+	draw_set_mapped_constant_buffer(draw, PIPE_SHADER_VERTEX, 0,
 					nv20->constbuf[PIPE_SHADER_VERTEX],
 					nv20->constbuf_nr[PIPE_SHADER_VERTEX]);
 
diff --git a/src/gallium/drivers/nv30/nv30_screen.c b/src/gallium/drivers/nv30/nv30_screen.c
index 9ed48178dc..ee83ec5676 100644
--- a/src/gallium/drivers/nv30/nv30_screen.c
+++ b/src/gallium/drivers/nv30/nv30_screen.c
@@ -67,6 +67,16 @@ nv30_screen_get_param(struct pipe_screen *pscreen, int param)
 	case NOUVEAU_CAP_HW_VTXBUF:
 	case NOUVEAU_CAP_HW_IDXBUF:
 		return 1;
+	case PIPE_CAP_INDEP_BLEND_ENABLE:
+		return 0;
+	case PIPE_CAP_INDEP_BLEND_FUNC:
+		return 0;
+	case PIPE_CAP_TGSI_FS_COORD_ORIGIN_LOWER_LEFT:
+	case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_HALF_INTEGER:
+		return 1;
+	case PIPE_CAP_TGSI_FS_COORD_ORIGIN_UPPER_LEFT:
+	case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_INTEGER:
+		return 0;
 	default:
 		NOUVEAU_ERR("Unknown PIPE_CAP %d\n", param);
 		return 0;
diff --git a/src/gallium/drivers/nv30/nv30_state.c b/src/gallium/drivers/nv30/nv30_state.c
index 065c927a10..1f4c964270 100644
--- a/src/gallium/drivers/nv30/nv30_state.c
+++ b/src/gallium/drivers/nv30/nv30_state.c
@@ -16,27 +16,27 @@ nv30_blend_state_create(struct pipe_context *pipe,
 	struct nv30_blend_state *bso = CALLOC(1, sizeof(*bso));
 	struct nouveau_stateobj *so = so_new(5, 8, 0);
 
-	if (cso->blend_enable) {
+	if (cso->rt[0].blend_enable) {
 		so_method(so, rankine, NV34TCL_BLEND_FUNC_ENABLE, 3);
 		so_data  (so, 1);
-		so_data  (so, (nvgl_blend_func(cso->alpha_src_factor) << 16) |
-			       nvgl_blend_func(cso->rgb_src_factor));
-		so_data  (so, nvgl_blend_func(cso->alpha_dst_factor) << 16 |
-			      nvgl_blend_func(cso->rgb_dst_factor));
+		so_data  (so, (nvgl_blend_func(cso->rt[0].alpha_src_factor) << 16) |
+			       nvgl_blend_func(cso->rt[0].rgb_src_factor));
+		so_data  (so, nvgl_blend_func(cso->rt[0].alpha_dst_factor) << 16 |
+			      nvgl_blend_func(cso->rt[0].rgb_dst_factor));
 		/* FIXME: Gallium assumes GL_EXT_blend_func_separate.
 		   It is not the case for NV30 */
 		so_method(so, rankine, NV34TCL_BLEND_EQUATION, 1);
-		so_data  (so, nvgl_blend_eqn(cso->rgb_func));
+		so_data  (so, nvgl_blend_eqn(cso->rt[0].rgb_func));
 	} else {
 		so_method(so, rankine, NV34TCL_BLEND_FUNC_ENABLE, 1);
 		so_data  (so, 0);
 	}
 
 	so_method(so, rankine, NV34TCL_COLOR_MASK, 1);
-	so_data  (so, (((cso->colormask & PIPE_MASK_A) ? (0x01 << 24) : 0) |
-		       ((cso->colormask & PIPE_MASK_R) ? (0x01 << 16) : 0) |
-		       ((cso->colormask & PIPE_MASK_G) ? (0x01 <<  8) : 0) |
-		       ((cso->colormask & PIPE_MASK_B) ? (0x01 <<  0) : 0)));
+	so_data  (so, (((cso->rt[0].colormask & PIPE_MASK_A) ? (0x01 << 24) : 0) |
+		       ((cso->rt[0].colormask & PIPE_MASK_R) ? (0x01 << 16) : 0) |
+		       ((cso->rt[0].colormask & PIPE_MASK_G) ? (0x01 <<  8) : 0) |
+		       ((cso->rt[0].colormask & PIPE_MASK_B) ? (0x01 <<  0) : 0)));
 
 	if (cso->logicop_enable) {
 		so_method(so, rankine, NV34TCL_COLOR_LOGIC_OP_ENABLE, 2);
diff --git a/src/gallium/drivers/nv40/nv40_draw.c b/src/gallium/drivers/nv40/nv40_draw.c
index d826f8c2f5..a2fd5f6c02 100644
--- a/src/gallium/drivers/nv40/nv40_draw.c
+++ b/src/gallium/drivers/nv40/nv40_draw.c
@@ -271,7 +271,7 @@ nv40_draw_elements_swtnl(struct pipe_context *pipe,
 		map = pipe_buffer_map(pscreen,
 				      nv40->constbuf[PIPE_SHADER_VERTEX],
 				      PIPE_BUFFER_USAGE_CPU_READ);
-		draw_set_mapped_constant_buffer(nv40->draw, PIPE_SHADER_VERTEX,
+		draw_set_mapped_constant_buffer(nv40->draw, PIPE_SHADER_VERTEX, 0,
                                                 map, nr);
 	}
 
diff --git a/src/gallium/drivers/nv40/nv40_screen.c b/src/gallium/drivers/nv40/nv40_screen.c
index 9e55e5a089..b1a7343b40 100644
--- a/src/gallium/drivers/nv40/nv40_screen.c
+++ b/src/gallium/drivers/nv40/nv40_screen.c
@@ -52,6 +52,16 @@ nv40_screen_get_param(struct pipe_screen *pscreen, int param)
 		if (screen->curie->grclass == NV40TCL)
 			return 1;
 		return 0;
+	case PIPE_CAP_INDEP_BLEND_ENABLE:
+		return 0;
+	case PIPE_CAP_INDEP_BLEND_FUNC:
+		return 0;
+	case PIPE_CAP_TGSI_FS_COORD_ORIGIN_LOWER_LEFT:
+	case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_HALF_INTEGER:
+		return 1;
+	case PIPE_CAP_TGSI_FS_COORD_ORIGIN_UPPER_LEFT:
+	case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_INTEGER:
+		return 0;
 	default:
 		NOUVEAU_ERR("Unknown PIPE_CAP %d\n", param);
 		return 0;
diff --git a/src/gallium/drivers/nv40/nv40_state.c b/src/gallium/drivers/nv40/nv40_state.c
index 7d990f7d56..7a63fba18c 100644
--- a/src/gallium/drivers/nv40/nv40_state.c
+++ b/src/gallium/drivers/nv40/nv40_state.c
@@ -18,26 +18,26 @@ nv40_blend_state_create(struct pipe_context *pipe,
 	struct nv40_blend_state *bso = CALLOC(1, sizeof(*bso));
 	struct nouveau_stateobj *so = so_new(5, 8, 0);
 
-	if (cso->blend_enable) {
+	if (cso->rt[0].blend_enable) {
 		so_method(so, curie, NV40TCL_BLEND_ENABLE, 3);
 		so_data  (so, 1);
-		so_data  (so, (nvgl_blend_func(cso->alpha_src_factor) << 16) |
-			       nvgl_blend_func(cso->rgb_src_factor));
-		so_data  (so, nvgl_blend_func(cso->alpha_dst_factor) << 16 |
-			      nvgl_blend_func(cso->rgb_dst_factor));
+		so_data  (so, (nvgl_blend_func(cso->rt[0].alpha_src_factor) << 16) |
+			       nvgl_blend_func(cso->rt[0].rgb_src_factor));
+		so_data  (so, nvgl_blend_func(cso->rt[0].alpha_dst_factor) << 16 |
+			      nvgl_blend_func(cso->rt[0].rgb_dst_factor));
 		so_method(so, curie, NV40TCL_BLEND_EQUATION, 1);
-		so_data  (so, nvgl_blend_eqn(cso->alpha_func) << 16 |
-			      nvgl_blend_eqn(cso->rgb_func));
+		so_data  (so, nvgl_blend_eqn(cso->rt[0].alpha_func) << 16 |
+			      nvgl_blend_eqn(cso->rt[0].rgb_func));
 	} else {
 		so_method(so, curie, NV40TCL_BLEND_ENABLE, 1);
 		so_data  (so, 0);
 	}
 
 	so_method(so, curie, NV40TCL_COLOR_MASK, 1);
-	so_data  (so, (((cso->colormask & PIPE_MASK_A) ? (0x01 << 24) : 0) |
-		       ((cso->colormask & PIPE_MASK_R) ? (0x01 << 16) : 0) |
-		       ((cso->colormask & PIPE_MASK_G) ? (0x01 <<  8) : 0) |
-		       ((cso->colormask & PIPE_MASK_B) ? (0x01 <<  0) : 0)));
+	so_data  (so, (((cso->rt[0].colormask & PIPE_MASK_A) ? (0x01 << 24) : 0) |
+		       ((cso->rt[0].colormask & PIPE_MASK_R) ? (0x01 << 16) : 0) |
+		       ((cso->rt[0].colormask & PIPE_MASK_G) ? (0x01 <<  8) : 0) |
+		       ((cso->rt[0].colormask & PIPE_MASK_B) ? (0x01 <<  0) : 0)));
 
 	if (cso->logicop_enable) {
 		so_method(so, curie, NV40TCL_COLOR_LOGIC_OP_ENABLE, 2);
diff --git a/src/gallium/drivers/nv50/nv50_context.c b/src/gallium/drivers/nv50/nv50_context.c
index 22a8498c69..5c705ccc8f 100644
--- a/src/gallium/drivers/nv50/nv50_context.c
+++ b/src/gallium/drivers/nv50/nv50_context.c
@@ -34,6 +34,11 @@ nv50_flush(struct pipe_context *pipe, unsigned flags,
 	struct nv50_context *nv50 = nv50_context(pipe);
 	struct nouveau_channel *chan = nv50->screen->base.channel;
 
+	if (flags & PIPE_FLUSH_TEXTURE_CACHE) {
+		BEGIN_RING(chan, nv50->screen->tesla, 0x1338, 1);
+		OUT_RING  (chan, 0x20);
+	}
+
 	if (flags & PIPE_FLUSH_FRAME)
 		FIRE_RING(chan);
 }
@@ -81,6 +86,10 @@ nv50_destroy(struct pipe_context *pipe)
 		so_ref(NULL, &nv50->state.vtxattr);
 
 	draw_destroy(nv50->draw);
+
+	if (nv50->screen->cur_ctx == nv50)
+		nv50->screen->cur_ctx = NULL;
+
 	FREE(nv50);
 }
 
@@ -104,7 +113,9 @@ nv50_create(struct pipe_screen *pscreen, unsigned pctx_id)
 	nv50->pipe.destroy = nv50_destroy;
 
 	nv50->pipe.draw_arrays = nv50_draw_arrays;
+	nv50->pipe.draw_arrays_instanced = nv50_draw_arrays_instanced;
 	nv50->pipe.draw_elements = nv50_draw_elements;
+	nv50->pipe.draw_elements_instanced = nv50_draw_elements_instanced;
 	nv50->pipe.clear = nv50_clear;
 
 	nv50->pipe.flush = nv50_flush;
diff --git a/src/gallium/drivers/nv50/nv50_context.h b/src/gallium/drivers/nv50/nv50_context.h
index d024be6ea9..bebcd95054 100644
--- a/src/gallium/drivers/nv50/nv50_context.h
+++ b/src/gallium/drivers/nv50/nv50_context.h
@@ -29,9 +29,7 @@
 #define NV50_CB_PVP		1
 #define NV50_CB_PFP		2
 #define NV50_CB_PGP		3
-#define NV50_CB_TIC		4
-#define NV50_CB_TSC		5
-#define NV50_CB_PUPLOAD         6
+#define NV50_CB_AUX		4
 
 #define NV50_NEW_BLEND		(1 << 0)
 #define NV50_NEW_ZSA		(1 << 1)
@@ -137,6 +135,7 @@ struct nv50_state {
 	struct nouveau_stateobj *vtxfmt;
 	struct nouveau_stateobj *vtxbuf;
 	struct nouveau_stateobj *vtxattr;
+	struct nouveau_stateobj *instbuf;
 	unsigned vtxelt_nr;
 };
 
@@ -198,11 +197,22 @@ extern struct draw_stage *nv50_draw_render_stage(struct nv50_context *nv50);
 /* nv50_vbo.c */
 extern void nv50_draw_arrays(struct pipe_context *, unsigned mode,
 				unsigned start, unsigned count);
+extern void nv50_draw_arrays_instanced(struct pipe_context *, unsigned mode,
+					unsigned start, unsigned count,
+					unsigned startInstance,
+					unsigned instanceCount);
 extern void nv50_draw_elements(struct pipe_context *pipe,
 				  struct pipe_buffer *indexBuffer,
 				  unsigned indexSize,
 				  unsigned mode, unsigned start,
 				  unsigned count);
+extern void nv50_draw_elements_instanced(struct pipe_context *pipe,
+					 struct pipe_buffer *indexBuffer,
+					 unsigned indexSize,
+					 unsigned mode, unsigned start,
+					 unsigned count,
+					 unsigned startInstance,
+					 unsigned instanceCount);
 extern void nv50_vbo_validate(struct nv50_context *nv50);
 
 /* nv50_clear.c */
diff --git a/src/gallium/drivers/nv50/nv50_program.c b/src/gallium/drivers/nv50/nv50_program.c
index acd695b39b..e74a526c62 100644
--- a/src/gallium/drivers/nv50/nv50_program.c
+++ b/src/gallium/drivers/nv50/nv50_program.c
@@ -95,6 +95,8 @@ struct nv50_reg {
 
 	int vtx; /* vertex index, for GP inputs (TGSI Dimension.Index) */
 	int indirect[2]; /* index into pc->addr, or -1 */
+
+	ubyte buf_index; /* c{0 .. 15}[] or g{0 .. 15}[] */
 };
 
 #define NV50_MOD_NEG 1
@@ -188,6 +190,7 @@ ctor_reg(struct nv50_reg *reg, unsigned type, int index, int hw)
 	reg->vtx = -1;
 	reg->acc = 0;
 	reg->indirect[0] = reg->indirect[1] = -1;
+	reg->buf_index = (type == P_CONST) ? 1 : 0;
 }
 
 static INLINE unsigned
@@ -474,6 +477,12 @@ is_join(struct nv50_program_exec *e)
 	return FALSE;
 }
 
+static INLINE boolean
+is_control_flow(struct nv50_program_exec *e)
+{
+	return (e->inst[0] & 2);
+}
+
 static INLINE void
 set_pred(struct nv50_pc *pc, unsigned pred, unsigned idx,
 	 struct nv50_program_exec *e)
@@ -631,7 +640,7 @@ set_data(struct nv50_pc *pc, struct nv50_reg *src, unsigned m, unsigned s,
 		set_addr(e, pc->addr[src->indirect[0]]);
 	}
 
-	e->inst[1] |= (((src->type == P_IMMD) ? 0 : 1) << 22);
+	e->inst[1] |= (src->buf_index << 22);
 }
 
 /* Never apply nv50_reg::mod in emit_mov, or carefully check the code !!! */
@@ -3163,7 +3172,9 @@ nv50_program_tx_insn(struct nv50_pc *pc,
 		if (!is_long(pc->p->exec_tail))
 			convert_to_long(pc, pc->p->exec_tail);
 		else
-		if (is_immd(pc->p->exec_tail) || is_join(pc->p->exec_tail))
+		if (is_immd(pc->p->exec_tail) ||
+		    is_join(pc->p->exec_tail) ||
+		    is_control_flow(pc->p->exec_tail))
 			emit_nop(pc);
 
 		pc->p->exec_tail->inst[1] |= 1; /* set exit bit */
@@ -3272,7 +3283,7 @@ prep_inspect_insn(struct nv50_pc *pc, const struct tgsi_full_instruction *insn)
 static unsigned
 nv50_revdep_reorder(unsigned m[4], unsigned rdep[4])
 {
-	unsigned i, c, x, unsafe;
+	unsigned i, c, x, unsafe = 0;
 
 	for (c = 0; c < 4; c++)
 		m[c] = c;
@@ -3483,6 +3494,19 @@ load_frontfacing(struct nv50_pc *pc, struct nv50_reg *sv)
 }
 
 static void
+load_instance_id(struct nv50_pc *pc, unsigned index)
+{
+	struct nv50_reg reg, mem;
+
+	ctor_reg(&reg, P_TEMP, -1, -1);
+	ctor_reg(&mem, P_CONST, -1, 24); /* startInstance */
+	mem.buf_index = 2;
+
+	emit_add_b32(pc, &reg, &pc->sysval[index], &mem);
+	pc->sysval[index] = reg;
+}
+
+static void
 copy_semantic_info(struct nv50_program *p)
 {
 	unsigned i, id;
@@ -3668,8 +3692,10 @@ nv50_program_tx_prep(struct nv50_pc *pc)
 			}
 			if (p->cfg.regs[0] & (1 << 0))
 				pc->sysval[vertex_id].hw = rid++;
-			if (p->cfg.regs[0] & (1 << 4))
+			if (p->cfg.regs[0] & (1 << 4)) {
 				pc->sysval[instance_id].hw = rid++;
+				load_instance_id(pc, instance_id);
+			}
 		}
 
 		for (i = 0, rid = 0; i < pc->result_nr; ++i) {
@@ -3702,13 +3728,21 @@ nv50_program_tx_prep(struct nv50_pc *pc)
 		copy_semantic_info(p);
 	} else
 	if (p->type == PIPE_SHADER_FRAGMENT) {
-		int rid, aid, base;
+		int rid, aid;
 		unsigned n = 0, m = pc->attr_nr - flat_nr;
 
 		pc->allow32 = TRUE;
 
-		base = (TGSI_SEMANTIC_POSITION ==
-			p->info.input_semantic_name[0]) ? 0 : 1;
+		/* do we read FragCoord ? */
+		if (pc->attr_nr &&
+		    p->info.input_semantic_name[0] == TGSI_SEMANTIC_POSITION) {
+			/* select FCRD components we want accessible */
+			for (c = 0; c < 4; ++c)
+				if (pc->attr[c].acc)
+					p->cfg.regs[1] |= 1 << (24 + c);
+			aid = 0;
+		} else /* offset by 1 if FCRD.w is needed for pinterp */
+			aid = popcnt4(p->cfg.regs[1] >> 24);
 
 		/* non-flat interpolants have to be mapped to
 		 * the lower hardware IDs, so sort them:
@@ -3724,12 +3758,6 @@ nv50_program_tx_prep(struct nv50_pc *pc)
 		}
 		copy_semantic_info(p);
 
-		if (!base) /* set w-coordinate mask from perspective interp */
-			p->cfg.in[0].mask |= p->cfg.regs[1] >> 24;
-
-		aid = popcnt4( /* if fcrd isn't contained in cfg.io */
-			base ? (p->cfg.regs[1] >> 24) : p->cfg.in[0].mask);
-
 		for (n = 0; n < pc->attr_nr; ++n) {
 			p->cfg.in[n].hw = rid = aid;
 			i = p->cfg.in[n].id;
@@ -3751,9 +3779,6 @@ nv50_program_tx_prep(struct nv50_pc *pc)
 			aid += popcnt4(p->cfg.in[n].mask);
 		}
 
-		if (!base)
-			p->cfg.regs[1] |= p->cfg.in[0].mask << 24;
-
 		m = popcnt4(p->cfg.regs[1] >> 24);
 
 		/* set count of non-position inputs and of non-flat
@@ -4641,6 +4666,7 @@ nv50_program_destroy(struct nv50_context *nv50, struct nv50_program *p)
 
 	nouveau_bo_ref(NULL, &p->bo);
 
+	FREE(p->immd);
 	nouveau_resource_free(&p->data[0]);
 
 	p->translated = 0;
diff --git a/src/gallium/drivers/nv50/nv50_query.c b/src/gallium/drivers/nv50/nv50_query.c
index 5a4ab3508b..9eba4c9611 100644
--- a/src/gallium/drivers/nv50/nv50_query.c
+++ b/src/gallium/drivers/nv50/nv50_query.c
@@ -48,7 +48,7 @@ nv50_query_create(struct pipe_context *pipe, unsigned type)
 	assert (q->type == PIPE_QUERY_OCCLUSION_COUNTER);
 	q->type = type;
 
-	ret = nouveau_bo_new(dev, NOUVEAU_BO_VRAM | NOUVEAU_BO_MAP, 256,
+	ret = nouveau_bo_new(dev, NOUVEAU_BO_GART | NOUVEAU_BO_MAP, 256,
 			     16, &q->bo);
 	if (ret) {
 		FREE(q);
@@ -95,11 +95,13 @@ nv50_query_end(struct pipe_context *pipe, struct pipe_query *pq)
 
 	MARK_RING (chan, 5, 2); /* flush on lack of space or relocs */
 	BEGIN_RING(chan, tesla, NV50TCL_QUERY_ADDRESS_HIGH, 4);
-	OUT_RELOCh(chan, q->bo, 0, NOUVEAU_BO_VRAM | NOUVEAU_BO_WR);
-	OUT_RELOCl(chan, q->bo, 0, NOUVEAU_BO_VRAM | NOUVEAU_BO_WR);
+	OUT_RELOCh(chan, q->bo, 0, NOUVEAU_BO_GART | NOUVEAU_BO_WR);
+	OUT_RELOCl(chan, q->bo, 0, NOUVEAU_BO_GART | NOUVEAU_BO_WR);
 	OUT_RING  (chan, 0x00000000);
 	OUT_RING  (chan, 0x0100f002);
-	FIRE_RING (chan);
+
+	BEGIN_RING(chan, tesla, NV50TCL_SAMPLECNT_ENABLE, 1);
+	OUT_RING  (chan, 0);
 }
 
 static boolean
@@ -123,6 +125,35 @@ nv50_query_result(struct pipe_context *pipe, struct pipe_query *pq,
 	return q->ready;
 }
 
+static void
+nv50_render_condition(struct pipe_context *pipe,
+		      struct pipe_query *pq, uint mode)
+{
+	struct nv50_context *nv50 = nv50_context(pipe);
+	struct nouveau_channel *chan = nv50->screen->base.channel;
+	struct nouveau_grobj *tesla = nv50->screen->tesla;
+	struct nv50_query *q;
+
+	if (!pq) {
+		BEGIN_RING(chan, tesla, NV50TCL_COND_MODE, 1);
+		OUT_RING  (chan, NV50TCL_COND_MODE_ALWAYS);
+		return;
+	}
+	q = nv50_query(pq);
+
+	if (mode == PIPE_RENDER_COND_WAIT ||
+	    mode == PIPE_RENDER_COND_BY_REGION_WAIT) {
+		/* XXX: big fence, FIFO semaphore might be better */
+		BEGIN_RING(chan, tesla, 0x0110, 1);
+		OUT_RING  (chan, 0);
+	}
+
+	BEGIN_RING(chan, tesla, NV50TCL_COND_ADDRESS_HIGH, 3);
+	OUT_RELOCh(chan, q->bo, 0, NOUVEAU_BO_GART | NOUVEAU_BO_RD);
+	OUT_RELOCl(chan, q->bo, 0, NOUVEAU_BO_GART | NOUVEAU_BO_RD);
+	OUT_RING  (chan, NV50TCL_COND_MODE_RES);
+}
+
 void
 nv50_init_query_functions(struct nv50_context *nv50)
 {
@@ -131,4 +162,5 @@ nv50_init_query_functions(struct nv50_context *nv50)
 	nv50->pipe.begin_query = nv50_query_begin;
 	nv50->pipe.end_query = nv50_query_end;
 	nv50->pipe.get_query_result = nv50_query_result;
+	nv50->pipe.render_condition = nv50_render_condition;
 }
diff --git a/src/gallium/drivers/nv50/nv50_screen.c b/src/gallium/drivers/nv50/nv50_screen.c
index 73205652cc..c4465a228c 100644
--- a/src/gallium/drivers/nv50/nv50_screen.c
+++ b/src/gallium/drivers/nv50/nv50_screen.c
@@ -135,6 +135,16 @@ nv50_screen_get_param(struct pipe_screen *pscreen, int param)
 		return 1;
 	case NOUVEAU_CAP_HW_IDXBUF:
 		return 0;
+	case PIPE_CAP_INDEP_BLEND_ENABLE:
+		return 1;
+	case PIPE_CAP_INDEP_BLEND_FUNC:
+		return 0;
+	case PIPE_CAP_TGSI_FS_COORD_ORIGIN_UPPER_LEFT:
+	case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_HALF_INTEGER:
+		return 1;
+	case PIPE_CAP_TGSI_FS_COORD_ORIGIN_LOWER_LEFT:
+	case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_INTEGER:
+		return 0;
 	default:
 		NOUVEAU_ERR("Unknown PIPE_CAP %d\n", param);
 		return 0;
@@ -329,7 +339,7 @@ nv50_screen_create(struct pipe_winsys *ws, struct nouveau_device *dev)
 	so_ref(NULL, &so);
 
 	/* Static tesla init */
-	so = so_new(44, 90, 22);
+	so = so_new(47, 95, 24);
 
 	so_method(so, screen->tesla, NV50TCL_COND_MODE, 1);
 	so_data  (so, NV50TCL_COND_MODE_ALWAYS);
@@ -372,7 +382,7 @@ nv50_screen_create(struct pipe_winsys *ws, struct nouveau_device *dev)
 	}
 
 	for (i = 0; i < 3; i++) {
-		ret = nouveau_bo_new(dev, NOUVEAU_BO_VRAM, 0, (128 * 4) * 4,
+		ret = nouveau_bo_new(dev, NOUVEAU_BO_VRAM, 0, (256 * 4) * 4,
 				     &screen->constbuf_parm[i]);
 		if (ret) {
 			nv50_screen_destroy(pscreen);
@@ -411,6 +421,18 @@ nv50_screen_create(struct pipe_winsys *ws, struct nouveau_device *dev)
 	so_method(so, screen->tesla, NV50TCL_SET_PROGRAM_CB, 1);
 	so_data  (so, 0x00000031 | (NV50_CB_PMISC << 12));
 
+	/* bind auxiliary constbuf to immediate data bo */
+	so_method(so, screen->tesla, NV50TCL_CB_DEF_ADDRESS_HIGH, 3);
+	so_reloc (so, screen->constbuf_misc[0], (128 * 4) * 4,
+		  NOUVEAU_BO_VRAM | NOUVEAU_BO_RD | NOUVEAU_BO_HIGH, 0, 0);
+	so_reloc (so, screen->constbuf_misc[0], (128 * 4) * 4,
+		  NOUVEAU_BO_VRAM | NOUVEAU_BO_RD | NOUVEAU_BO_LOW, 0, 0);
+	so_data  (so, (NV50_CB_AUX << 16) | 0x00000200);
+	so_method(so, screen->tesla, NV50TCL_SET_PROGRAM_CB, 1);
+	so_data  (so, 0x00000201 | (NV50_CB_AUX << 12));
+	so_method(so, screen->tesla, NV50TCL_SET_PROGRAM_CB, 1);
+	so_data  (so, 0x00000221 | (NV50_CB_AUX << 12));
+
 	so_method(so, screen->tesla, NV50TCL_CB_DEF_ADDRESS_HIGH, 3);
 	so_reloc (so, screen->constbuf_parm[PIPE_SHADER_VERTEX], 0,
 		  NOUVEAU_BO_VRAM | NOUVEAU_BO_RD | NOUVEAU_BO_HIGH, 0, 0);
diff --git a/src/gallium/drivers/nv50/nv50_screen.h b/src/gallium/drivers/nv50/nv50_screen.h
index 7a155ca9c4..0d786b0f2e 100644
--- a/src/gallium/drivers/nv50/nv50_screen.h
+++ b/src/gallium/drivers/nv50/nv50_screen.h
@@ -23,6 +23,8 @@ struct nv50_screen {
 	struct nouveau_resource *immd_heap[1];
 	struct nouveau_resource *parm_heap[PIPE_SHADER_TYPES];
 
+	struct pipe_buffer *strm_vbuf[16];
+
 	struct nouveau_bo *tic;
 	struct nouveau_bo *tsc;
 
diff --git a/src/gallium/drivers/nv50/nv50_state.c b/src/gallium/drivers/nv50/nv50_state.c
index 6ab33be663..da8dc933b0 100644
--- a/src/gallium/drivers/nv50/nv50_state.c
+++ b/src/gallium/drivers/nv50/nv50_state.c
@@ -31,6 +31,23 @@
 
 #include "nouveau/nouveau_stateobj.h"
 
+static INLINE uint32_t
+nv50_colormask(unsigned mask)
+{
+	uint32_t cmask = 0;
+
+	if (mask & PIPE_MASK_R)
+		cmask |= 0x0001;
+	if (mask & PIPE_MASK_G)
+		cmask |= 0x0010;
+	if (mask & PIPE_MASK_B)
+		cmask |= 0x0100;
+	if (mask & PIPE_MASK_A)
+		cmask |= 0x1000;
+
+	return cmask;
+}
+
 static void *
 nv50_blend_state_create(struct pipe_context *pipe,
 			const struct pipe_blend_state *cso)
@@ -38,28 +55,37 @@ nv50_blend_state_create(struct pipe_context *pipe,
 	struct nouveau_stateobj *so = so_new(5, 24, 0);
 	struct nouveau_grobj *tesla = nv50_context(pipe)->screen->tesla;
 	struct nv50_blend_stateobj *bso = CALLOC_STRUCT(nv50_blend_stateobj);
-	unsigned cmask = 0, i;
+	unsigned i, blend_enabled = 0;
 
 	/*XXX ignored:
 	 * 	- dither
 	 */
 
-	if (cso->blend_enable == 0) {
-		so_method(so, tesla, NV50TCL_BLEND_ENABLE(0), 8);
+	so_method(so, tesla, NV50TCL_BLEND_ENABLE(0), 8);
+	if (cso->independent_blend_enable) {
+		for (i = 0; i < 8; ++i) {
+			so_data(so, cso->rt[i].blend_enable);
+			if (cso->rt[i].blend_enable)
+				blend_enabled = 1;
+		}
+	} else
+	if (cso->rt[0].blend_enable) {
+		blend_enabled = 1;
 		for (i = 0; i < 8; i++)
-			so_data(so, 0);
+			so_data(so, 1);
 	} else {
-		so_method(so, tesla, NV50TCL_BLEND_ENABLE(0), 8);
 		for (i = 0; i < 8; i++)
-			so_data(so, 1);
+			so_data(so, 0);
+	}
+	if (blend_enabled) {
 		so_method(so, tesla, NV50TCL_BLEND_EQUATION_RGB, 5);
-		so_data  (so, nvgl_blend_eqn(cso->rgb_func));
-		so_data  (so, 0x4000 | nvgl_blend_func(cso->rgb_src_factor));
-		so_data  (so, 0x4000 | nvgl_blend_func(cso->rgb_dst_factor));
-		so_data  (so, nvgl_blend_eqn(cso->alpha_func));
-		so_data  (so, 0x4000 | nvgl_blend_func(cso->alpha_src_factor));
+		so_data  (so, nvgl_blend_eqn(cso->rt[0].rgb_func));
+		so_data  (so, 0x4000 | nvgl_blend_func(cso->rt[0].rgb_src_factor));
+		so_data  (so, 0x4000 | nvgl_blend_func(cso->rt[0].rgb_dst_factor));
+		so_data  (so, nvgl_blend_eqn(cso->rt[0].alpha_func));
+		so_data  (so, 0x4000 | nvgl_blend_func(cso->rt[0].alpha_src_factor));
 		so_method(so, tesla, NV50TCL_BLEND_FUNC_DST_ALPHA, 1);
-		so_data  (so, 0x4000 | nvgl_blend_func(cso->alpha_dst_factor));
+		so_data  (so, 0x4000 | nvgl_blend_func(cso->rt[0].alpha_dst_factor));
 	}
 
 	if (cso->logicop_enable == 0 ) {
@@ -71,17 +97,15 @@ nv50_blend_state_create(struct pipe_context *pipe,
 		so_data  (so, nvgl_logicop_func(cso->logicop_func));
 	}
 
-	if (cso->colormask & PIPE_MASK_R)
-		cmask |= (1 << 0);
-	if (cso->colormask & PIPE_MASK_G)
-		cmask |= (1 << 4);
-	if (cso->colormask & PIPE_MASK_B)
-		cmask |= (1 << 8);
-	if (cso->colormask & PIPE_MASK_A)
-		cmask |= (1 << 12);
 	so_method(so, tesla, NV50TCL_COLOR_MASK(0), 8);
-	for (i = 0; i < 8; i++)
-		so_data(so, cmask);
+	if (cso->independent_blend_enable)
+		for (i = 0; i < 8; ++i)
+			so_data(so, nv50_colormask(cso->rt[i].colormask));
+	else {
+		uint32_t cmask = nv50_colormask(cso->rt[0].colormask);
+		for (i = 0; i < 8; i++)
+			so_data(so, cmask);
+	}
 
 	bso->pipe = *cso;
 	so_ref(so, &bso->so);
diff --git a/src/gallium/drivers/nv50/nv50_state_validate.c b/src/gallium/drivers/nv50/nv50_state_validate.c
index fcd07b59cd..956da9b304 100644
--- a/src/gallium/drivers/nv50/nv50_state_validate.c
+++ b/src/gallium/drivers/nv50/nv50_state_validate.c
@@ -274,6 +274,9 @@ nv50_state_flush_notify(struct nouveau_channel *chan)
 	so_emit_reloc_markers(chan, nv50->state.fragprog);
 	so_emit_reloc_markers(chan, nv50->state.vtxbuf);
 	so_emit_reloc_markers(chan, nv50->screen->static_init);
+
+	if (nv50->state.instbuf)
+		so_emit_reloc_markers(chan, nv50->state.instbuf);
 }
 
 boolean
diff --git a/src/gallium/drivers/nv50/nv50_tex.c b/src/gallium/drivers/nv50/nv50_tex.c
index 871536dca9..9f1a171303 100644
--- a/src/gallium/drivers/nv50/nv50_tex.c
+++ b/src/gallium/drivers/nv50/nv50_tex.c
@@ -220,11 +220,8 @@ nv50_tex_validate(struct nv50_context *nv50)
 		return;
 	}
 
-	/* not sure if the following really do what I think: */
 	so_method(so, tesla, 0x1330, 1); /* flush TIC */
 	so_data  (so, 0);
-	so_method(so, tesla, 0x1338, 1); /* flush texture caches */
-	so_data  (so, 0x20);
 
 	so_ref(so, &nv50->state.tic_upload);
 	so_ref(NULL, &so);
diff --git a/src/gallium/drivers/nv50/nv50_vbo.c b/src/gallium/drivers/nv50/nv50_vbo.c
index 89a94d2fe8..bfb1b34d27 100644
--- a/src/gallium/drivers/nv50/nv50_vbo.c
+++ b/src/gallium/drivers/nv50/nv50_vbo.c
@@ -40,6 +40,8 @@ nv50_push_elements_u32(struct nv50_context *, uint32_t *, unsigned);
 static boolean
 nv50_push_arrays(struct nv50_context *, unsigned, unsigned);
 
+#define NV50_USING_LOATHED_EDGEFLAG(ctx) ((ctx)->vertprog->cfg.edgeflag_in < 16)
+
 static INLINE unsigned
 nv50_prim(unsigned mode)
 {
@@ -160,6 +162,309 @@ nv50_vbo_vtxelt_to_hw(struct pipe_vertex_element *ve)
 	return (hw_type | hw_size);
 }
 
+/* For instanced drawing from user buffers, hitting the FIFO repeatedly
+ * with the same vertex data is probably worse than uploading all data.
+ */
+static boolean
+nv50_upload_vtxbuf(struct nv50_context *nv50, unsigned i)
+{
+	struct nv50_screen *nscreen = nv50->screen;
+	struct pipe_screen *pscreen = &nscreen->base.base;
+	struct pipe_buffer *buf = nscreen->strm_vbuf[i];
+	struct pipe_vertex_buffer *vb = &nv50->vtxbuf[i];
+	uint8_t *src;
+	unsigned size = align(vb->buffer->size, 4096);
+
+	if (buf && buf->size < size)
+		pipe_buffer_reference(&nscreen->strm_vbuf[i], NULL);
+
+	if (!nscreen->strm_vbuf[i]) {
+		nscreen->strm_vbuf[i] = pipe_buffer_create(
+			pscreen, 0, PIPE_BUFFER_USAGE_VERTEX, size);
+		buf = nscreen->strm_vbuf[i];
+	}
+
+	src = pipe_buffer_map(pscreen, vb->buffer, PIPE_BUFFER_USAGE_CPU_READ);
+	if (!src)
+		return FALSE;
+	src += vb->buffer_offset;
+
+	size = (vb->max_index + 1) * vb->stride + 16; /* + 16 is for stride 0 */
+	if (vb->buffer_offset + size > vb->buffer->size)
+		size = vb->buffer->size - vb->buffer_offset;
+
+	pipe_buffer_write(pscreen, buf, vb->buffer_offset, size, src);
+	pipe_buffer_unmap(pscreen, vb->buffer);
+
+	vb->buffer = buf; /* don't pipe_reference, this is a private copy */
+	return TRUE;
+}
+
+static void
+nv50_upload_user_vbufs(struct nv50_context *nv50)
+{
+	unsigned i;
+
+	if (nv50->vbo_fifo)
+		nv50->dirty |= NV50_NEW_ARRAYS;
+	if (!(nv50->dirty & NV50_NEW_ARRAYS))
+		return;
+
+	for (i = 0; i < nv50->vtxbuf_nr; ++i) {
+		if (nv50->vtxbuf[i].buffer->usage & PIPE_BUFFER_USAGE_VERTEX)
+			continue;
+		nv50_upload_vtxbuf(nv50, i);
+	}
+}
+
+static void
+nv50_set_static_vtxattr(struct nv50_context *nv50, unsigned i, void *data)
+{
+	struct nouveau_grobj *tesla = nv50->screen->tesla;
+	struct nouveau_channel *chan = tesla->channel;
+	float v[4];
+
+	util_format_read_4f(nv50->vtxelt[i].src_format,
+			    v, 0, data, 0, 0, 0, 1, 1);
+
+	switch (nv50->vtxelt[i].nr_components) {
+	case 4:
+		BEGIN_RING(chan, tesla, NV50TCL_VTX_ATTR_4F_X(i), 4);
+		OUT_RINGf (chan, v[0]);
+		OUT_RINGf (chan, v[1]);
+		OUT_RINGf (chan, v[2]);
+		OUT_RINGf (chan, v[3]);
+		break;
+	case 3:
+		BEGIN_RING(chan, tesla, NV50TCL_VTX_ATTR_3F_X(i), 3);
+		OUT_RINGf (chan, v[0]);
+		OUT_RINGf (chan, v[1]);
+		OUT_RINGf (chan, v[2]);
+		break;
+	case 2:
+		BEGIN_RING(chan, tesla, NV50TCL_VTX_ATTR_2F_X(i), 2);
+		OUT_RINGf (chan, v[0]);
+		OUT_RINGf (chan, v[1]);
+		break;
+	case 1:
+		BEGIN_RING(chan, tesla, NV50TCL_VTX_ATTR_1F(i), 1);
+		OUT_RINGf (chan, v[0]);
+		break;
+	default:
+		assert(0);
+		break;
+	}
+}
+
+static unsigned
+init_per_instance_arrays_immd(struct nv50_context *nv50,
+			      unsigned startInstance,
+			      unsigned pos[16], unsigned step[16])
+{
+	struct nouveau_bo *bo;
+	unsigned i, b, count = 0;
+
+	for (i = 0; i < nv50->vtxelt_nr; ++i) {
+		if (!nv50->vtxelt[i].instance_divisor)
+			continue;
+		++count;
+		b = nv50->vtxelt[i].vertex_buffer_index;
+
+		pos[i] = nv50->vtxelt[i].src_offset +
+			nv50->vtxbuf[b].buffer_offset +
+			startInstance * nv50->vtxbuf[b].stride;
+		step[i] = startInstance % nv50->vtxelt[i].instance_divisor;
+
+		bo = nouveau_bo(nv50->vtxbuf[b].buffer);
+		if (!bo->map)
+			nouveau_bo_map(bo, NOUVEAU_BO_RD);
+
+		nv50_set_static_vtxattr(nv50, i, (uint8_t *)bo->map + pos[i]);
+	}
+
+	return count;
+}
+
+static unsigned
+init_per_instance_arrays(struct nv50_context *nv50,
+			 unsigned startInstance,
+			 unsigned pos[16], unsigned step[16])
+{
+	struct nouveau_grobj *tesla = nv50->screen->tesla;
+	struct nouveau_channel *chan = tesla->channel;
+	struct nouveau_bo *bo;
+	struct nouveau_stateobj *so;
+	unsigned i, b, count = 0;
+	const uint32_t rl = NOUVEAU_BO_VRAM | NOUVEAU_BO_GART | NOUVEAU_BO_RD;
+
+	if (nv50->vbo_fifo)
+		return init_per_instance_arrays_immd(nv50, startInstance,
+						     pos, step);
+
+	so = so_new(nv50->vtxelt_nr, nv50->vtxelt_nr * 2, nv50->vtxelt_nr * 2);
+
+	for (i = 0; i < nv50->vtxelt_nr; ++i) {
+		if (!nv50->vtxelt[i].instance_divisor)
+			continue;
+		++count;
+		b = nv50->vtxelt[i].vertex_buffer_index;
+
+		pos[i] = nv50->vtxelt[i].src_offset +
+			nv50->vtxbuf[b].buffer_offset +
+			startInstance * nv50->vtxbuf[b].stride;
+
+		if (!startInstance) {
+			step[i] = 0;
+			continue;
+		}
+		step[i] = startInstance % nv50->vtxelt[i].instance_divisor;
+
+		bo = nouveau_bo(nv50->vtxbuf[b].buffer);
+
+		so_method(so, tesla, NV50TCL_VERTEX_ARRAY_START_HIGH(i), 2);
+		so_reloc (so, bo, pos[i], rl | NOUVEAU_BO_HIGH, 0, 0);
+		so_reloc (so, bo, pos[i], rl | NOUVEAU_BO_LOW, 0, 0);
+	}
+
+	if (count && startInstance) {
+		so_ref (so, &nv50->state.instbuf); /* for flush notify */
+		so_emit(chan, nv50->state.instbuf);
+	}
+	so_ref (NULL, &so);
+
+	return count;
+}
+
+static void
+step_per_instance_arrays_immd(struct nv50_context *nv50,
+			      unsigned pos[16], unsigned step[16])
+{
+	struct nouveau_bo *bo;
+	unsigned i, b;
+
+	for (i = 0; i < nv50->vtxelt_nr; ++i) {
+		if (!nv50->vtxelt[i].instance_divisor)
+			continue;
+		if (++step[i] != nv50->vtxelt[i].instance_divisor)
+			continue;
+		b = nv50->vtxelt[i].vertex_buffer_index;
+		bo = nouveau_bo(nv50->vtxbuf[b].buffer);
+
+		step[i] = 0;
+		pos[i] += nv50->vtxbuf[b].stride;
+
+		nv50_set_static_vtxattr(nv50, i, (uint8_t *)bo->map + pos[i]);
+	}
+}
+
+static void
+step_per_instance_arrays(struct nv50_context *nv50,
+			 unsigned pos[16], unsigned step[16])
+{
+	struct nouveau_grobj *tesla = nv50->screen->tesla;
+	struct nouveau_channel *chan = tesla->channel;
+	struct nouveau_bo *bo;
+	struct nouveau_stateobj *so;
+	unsigned i, b;
+	const uint32_t rl = NOUVEAU_BO_VRAM | NOUVEAU_BO_GART | NOUVEAU_BO_RD;
+
+	if (nv50->vbo_fifo) {
+		step_per_instance_arrays_immd(nv50, pos, step);
+		return;
+	}
+
+	so = so_new(nv50->vtxelt_nr, nv50->vtxelt_nr * 2, nv50->vtxelt_nr * 2);
+
+	for (i = 0; i < nv50->vtxelt_nr; ++i) {
+		if (!nv50->vtxelt[i].instance_divisor)
+			continue;
+		b = nv50->vtxelt[i].vertex_buffer_index;
+
+		if (++step[i] == nv50->vtxelt[i].instance_divisor) {
+			step[i] = 0;
+			pos[i] += nv50->vtxbuf[b].stride;
+		}
+
+		bo = nouveau_bo(nv50->vtxbuf[b].buffer);
+
+		so_method(so, tesla, NV50TCL_VERTEX_ARRAY_START_HIGH(i), 2);
+		so_reloc (so, bo, pos[i], rl | NOUVEAU_BO_HIGH, 0, 0);
+		so_reloc (so, bo, pos[i], rl | NOUVEAU_BO_LOW, 0, 0);
+	}
+
+	so_ref (so, &nv50->state.instbuf); /* for flush notify */
+	so_ref (NULL, &so);
+
+	so_emit(chan, nv50->state.instbuf);
+}
+
+static INLINE void
+nv50_unmap_vbufs(struct nv50_context *nv50)
+{
+        unsigned i;
+
+        for (i = 0; i < nv50->vtxbuf_nr; ++i)
+                if (nouveau_bo(nv50->vtxbuf[i].buffer)->map)
+                        nouveau_bo_unmap(nouveau_bo(nv50->vtxbuf[i].buffer));
+}
+
+void
+nv50_draw_arrays_instanced(struct pipe_context *pipe,
+			   unsigned mode, unsigned start, unsigned count,
+			   unsigned startInstance, unsigned instanceCount)
+{
+	struct nv50_context *nv50 = nv50_context(pipe);
+	struct nouveau_channel *chan = nv50->screen->tesla->channel;
+	struct nouveau_grobj *tesla = nv50->screen->tesla;
+	unsigned i, nz_divisors;
+	unsigned step[16], pos[16];
+
+	if (!NV50_USING_LOATHED_EDGEFLAG(nv50))
+		nv50_upload_user_vbufs(nv50);
+
+	nv50_state_validate(nv50);
+
+	nz_divisors = init_per_instance_arrays(nv50, startInstance, pos, step);
+
+	BEGIN_RING(chan, tesla, NV50TCL_CB_ADDR, 2);
+	OUT_RING  (chan, NV50_CB_AUX | (24 << 8));
+	OUT_RING  (chan, startInstance);
+
+	BEGIN_RING(chan, tesla, NV50TCL_VERTEX_BEGIN, 1);
+	OUT_RING  (chan, nv50_prim(mode));
+
+	if (nv50->vbo_fifo)
+		nv50_push_arrays(nv50, start, count);
+	else {
+		BEGIN_RING(chan, tesla, NV50TCL_VERTEX_BUFFER_FIRST, 2);
+		OUT_RING  (chan, start);
+		OUT_RING  (chan, count);
+	}
+	BEGIN_RING(chan, tesla, NV50TCL_VERTEX_END, 1);
+	OUT_RING  (chan, 0);
+
+	for (i = 1; i < instanceCount; i++) {
+		if (nz_divisors) /* any non-zero array divisors ? */
+			step_per_instance_arrays(nv50, pos, step);
+
+		BEGIN_RING(chan, tesla, NV50TCL_VERTEX_BEGIN, 1);
+		OUT_RING  (chan, nv50_prim(mode) | (1 << 28));
+
+		if (nv50->vbo_fifo)
+			nv50_push_arrays(nv50, start, count);
+		else {
+			BEGIN_RING(chan, tesla, NV50TCL_VERTEX_BUFFER_FIRST, 2);
+			OUT_RING  (chan, start);
+			OUT_RING  (chan, count);
+		}
+		BEGIN_RING(chan, tesla, NV50TCL_VERTEX_END, 1);
+		OUT_RING  (chan, 0);
+	}
+	nv50_unmap_vbufs(nv50);
+
+	so_ref(NULL, &nv50->state.instbuf);
+}
+
 void
 nv50_draw_arrays(struct pipe_context *pipe, unsigned mode, unsigned start,
 		 unsigned count)
@@ -190,6 +495,8 @@ nv50_draw_arrays(struct pipe_context *pipe, unsigned mode, unsigned start,
 	BEGIN_RING(chan, tesla, NV50TCL_VERTEX_END, 1);
 	OUT_RING  (chan, 0);
 
+	nv50_unmap_vbufs(nv50);
+
         /* XXX: not sure what to do if ret != TRUE: flush and retry?
          */
         assert(ret);
@@ -218,7 +525,7 @@ nv50_draw_elements_inline_u08(struct nv50_context *nv50, uint8_t *map,
 		unsigned nr = count > 2046 ? 2046 : count;
 		int i;
 
-		BEGIN_RING(chan, tesla, NV50TCL_VB_ELEMENT_U16 | 0x40000000, nr >> 1);
+		BEGIN_RING_NI(chan, tesla, NV50TCL_VB_ELEMENT_U16, nr >> 1);
 		for (i = 0; i < nr; i += 2)
 			OUT_RING  (chan, (map[i + 1] << 16) | map[i]);
 
@@ -251,7 +558,7 @@ nv50_draw_elements_inline_u16(struct nv50_context *nv50, uint16_t *map,
 		unsigned nr = count > 2046 ? 2046 : count;
 		int i;
 
-		BEGIN_RING(chan, tesla, NV50TCL_VB_ELEMENT_U16 | 0x40000000, nr >> 1);
+		BEGIN_RING_NI(chan, tesla, NV50TCL_VB_ELEMENT_U16, nr >> 1);
 		for (i = 0; i < nr; i += 2)
 			OUT_RING  (chan, (map[i + 1] << 16) | map[i]);
 
@@ -276,7 +583,7 @@ nv50_draw_elements_inline_u32(struct nv50_context *nv50, uint32_t *map,
 	while (count) {
 		unsigned nr = count > 2047 ? 2047 : count;
 
-		BEGIN_RING(chan, tesla, NV50TCL_VB_ELEMENT_U32 | 0x40000000, nr);
+		BEGIN_RING_NI(chan, tesla, NV50TCL_VB_ELEMENT_U32, nr);
 		OUT_RINGp (chan, map, nr);
 
 		count -= nr;
@@ -285,6 +592,77 @@ nv50_draw_elements_inline_u32(struct nv50_context *nv50, uint32_t *map,
 	return TRUE;
 }
 
+static INLINE void
+nv50_draw_elements_inline(struct nv50_context *nv50,
+			  void *map, unsigned indexSize,
+			  unsigned start, unsigned count)
+{
+	switch (indexSize) {
+	case 1:
+		nv50_draw_elements_inline_u08(nv50, map, start, count);
+		break;
+	case 2:
+		nv50_draw_elements_inline_u16(nv50, map, start, count);
+		break;
+	case 4:
+		nv50_draw_elements_inline_u32(nv50, map, start, count);
+		break;
+	}
+}
+
+void
+nv50_draw_elements_instanced(struct pipe_context *pipe,
+			     struct pipe_buffer *indexBuffer,
+			     unsigned indexSize,
+			     unsigned mode, unsigned start, unsigned count,
+			     unsigned startInstance, unsigned instanceCount)
+{
+	struct nv50_context *nv50 = nv50_context(pipe);
+	struct nouveau_grobj *tesla = nv50->screen->tesla;
+	struct nouveau_channel *chan = tesla->channel;
+	struct pipe_screen *pscreen = pipe->screen;
+	void *map;
+	unsigned i, nz_divisors;
+	unsigned step[16], pos[16];
+
+	map = pipe_buffer_map(pscreen, indexBuffer, PIPE_BUFFER_USAGE_CPU_READ);
+
+	if (!NV50_USING_LOATHED_EDGEFLAG(nv50))
+		nv50_upload_user_vbufs(nv50);
+
+	nv50_state_validate(nv50);
+
+	nz_divisors = init_per_instance_arrays(nv50, startInstance, pos, step);
+
+	BEGIN_RING(chan, tesla, NV50TCL_CB_ADDR, 2);
+	OUT_RING  (chan, NV50_CB_AUX | (24 << 8));
+	OUT_RING  (chan, startInstance);
+
+	BEGIN_RING(chan, tesla, NV50TCL_VERTEX_BEGIN, 1);
+	OUT_RING  (chan, nv50_prim(mode));
+
+	nv50_draw_elements_inline(nv50, map, indexSize, start, count);
+
+	BEGIN_RING(chan, tesla, NV50TCL_VERTEX_END, 1);
+	OUT_RING  (chan, 0);
+
+	for (i = 1; i < instanceCount; ++i) {
+		if (nz_divisors) /* any non-zero array divisors ? */
+			step_per_instance_arrays(nv50, pos, step);
+
+		BEGIN_RING(chan, tesla, NV50TCL_VERTEX_BEGIN, 1);
+		OUT_RING  (chan, nv50_prim(mode) | (1 << 28));
+
+		nv50_draw_elements_inline(nv50, map, indexSize, start, count);
+
+		BEGIN_RING(chan, tesla, NV50TCL_VERTEX_END, 1);
+		OUT_RING  (chan, 0);
+	}
+	nv50_unmap_vbufs(nv50);
+
+	so_ref(NULL, &nv50->state.instbuf);
+}
+
 void
 nv50_draw_elements(struct pipe_context *pipe,
 		   struct pipe_buffer *indexBuffer, unsigned indexSize,
@@ -295,7 +673,6 @@ nv50_draw_elements(struct pipe_context *pipe,
 	struct nouveau_grobj *tesla = nv50->screen->tesla;
 	struct pipe_screen *pscreen = pipe->screen;
 	void *map;
-	boolean ret;
 	
 	map = pipe_buffer_map(pscreen, indexBuffer, PIPE_BUFFER_USAGE_CPU_READ);
 
@@ -308,29 +685,15 @@ nv50_draw_elements(struct pipe_context *pipe,
 
 	BEGIN_RING(chan, tesla, NV50TCL_VERTEX_BEGIN, 1);
 	OUT_RING  (chan, nv50_prim(mode));
-	switch (indexSize) {
-	case 1:
-		ret = nv50_draw_elements_inline_u08(nv50, map, start, count);
-		break;
-	case 2:
-		ret = nv50_draw_elements_inline_u16(nv50, map, start, count);
-		break;
-	case 4:
-		ret = nv50_draw_elements_inline_u32(nv50, map, start, count);
-		break;
-	default:
-		assert(0);
-		ret = FALSE;
-		break;
-	}
+
+	nv50_draw_elements_inline(nv50, map, indexSize, start, count);
+
 	BEGIN_RING(chan, tesla, NV50TCL_VERTEX_END, 1);
 	OUT_RING  (chan, 0);
 
+	nv50_unmap_vbufs(nv50);
+
 	pipe_buffer_unmap(pscreen, indexBuffer);
-        
-        /* XXX: what to do if ret != TRUE?  Flush and retry?
-         */
-	assert(ret);
 }
 
 static INLINE boolean
@@ -343,23 +706,16 @@ nv50_vbo_static_attrib(struct nv50_context *nv50, unsigned attrib,
 	struct nouveau_stateobj *so;
 	struct nouveau_grobj *tesla = nv50->screen->tesla;
 	struct nouveau_bo *bo = nouveau_bo(vb->buffer);
-	float *v;
+	float v[4];
 	int ret;
-	enum pipe_format pf = ve->src_format;
-	const struct util_format_description *desc;
-
-	desc = util_format_description(pf);
-	assert(desc);
-
-	if ((desc->channel[0].type != UTIL_FORMAT_TYPE_FLOAT) ||
-	    util_format_get_component_bits(pf, UTIL_FORMAT_COLORSPACE_RGB, 0) != 32)
-		return FALSE;
 
 	ret = nouveau_bo_map(bo, NOUVEAU_BO_RD);
 	if (ret)
 		return FALSE;
-	v = (float *)(bo->map + (vb->buffer_offset + ve->src_offset));
 
+	util_format_read_4f(ve->src_format, v, 0, (uint8_t *)bo->map +
+			    (vb->buffer_offset + ve->src_offset), 0,
+			    0, 0, 1, 1);
 	so = *pso;
 	if (!so)
 		*pso = so = so_new(nv50->vtxelt_nr, nv50->vtxelt_nr * 4, 0);
@@ -417,7 +773,7 @@ nv50_vbo_validate(struct nv50_context *nv50)
 		    !(nv50->vtxbuf[i].buffer->usage & PIPE_BUFFER_USAGE_VERTEX))
 			nv50->vbo_fifo = 0xffff;
 
-	if (nv50->vertprog->cfg.edgeflag_in < 16)
+	if (NV50_USING_LOATHED_EDGEFLAG(nv50))
 		nv50->vbo_fifo = 0xffff; /* vertprog can't set edgeflag */
 
 	n_ve = MAX2(nv50->vtxelt_nr, nv50->state.vtxelt_nr);
@@ -445,17 +801,20 @@ nv50_vbo_validate(struct nv50_context *nv50)
 			nv50->vbo_fifo &= ~(1 << i);
 			continue;
 		}
-		so_data(vtxfmt, hw | i);
 
 		if (nv50->vbo_fifo) {
+			so_data  (vtxfmt, hw |
+				  (ve->instance_divisor ? (1 << 4) : i));
 			so_method(vtxbuf, tesla,
 				  NV50TCL_VERTEX_ARRAY_FORMAT(i), 1);
 			so_data  (vtxbuf, 0);
 			continue;
 		}
+		so_data(vtxfmt, hw | i);
 
 		so_method(vtxbuf, tesla, NV50TCL_VERTEX_ARRAY_FORMAT(i), 3);
-		so_data  (vtxbuf, 0x20000000 | vb->stride);
+		so_data  (vtxbuf, 0x20000000 |
+			  (ve->instance_divisor ? 0 : vb->stride));
 		so_reloc (vtxbuf, bo, vb->buffer_offset +
 			  ve->src_offset, NOUVEAU_BO_VRAM | NOUVEAU_BO_GART |
 			  NOUVEAU_BO_RD | NOUVEAU_BO_HIGH, 0, 0);
@@ -493,7 +852,7 @@ typedef void (*pfn_push)(struct nouveau_channel *, void *);
 struct nv50_vbo_emitctx
 {
 	pfn_push push[16];
-	void *map[16];
+	uint8_t *map[16];
 	unsigned stride[16];
 	unsigned nr_ve;
 	unsigned vtx_dwords;
@@ -531,19 +890,18 @@ nv50_map_vbufs(struct nv50_context *nv50)
 
 	for (i = 0; i < nv50->vtxbuf_nr; ++i) {
 		struct pipe_vertex_buffer *vb = &nv50->vtxbuf[i];
-		unsigned size, delta;
+		unsigned size = vb->stride * (vb->max_index + 1) + 16;
 
 		if (nouveau_bo(vb->buffer)->map)
 			continue;
 
-		size = vb->stride * (vb->max_index + 1);
-		delta = vb->buffer_offset;
-
+		size = vb->stride * (vb->max_index + 1) + 16;
+		size = MIN2(size, vb->buffer->size);
 		if (!size)
-			size = vb->buffer->size - vb->buffer_offset;
+			size = vb->buffer->size;
 
 		if (nouveau_bo_map_range(nouveau_bo(vb->buffer),
-					 delta, size, NOUVEAU_BO_RD))
+					 0, size, NOUVEAU_BO_RD))
 			break;
 	}
 
@@ -554,16 +912,6 @@ nv50_map_vbufs(struct nv50_context *nv50)
 	return FALSE;
 }
 
-static INLINE void
-nv50_unmap_vbufs(struct nv50_context *nv50)
-{
-        unsigned i;
-
-        for (i = 0; i < nv50->vtxbuf_nr; ++i)
-                if (nouveau_bo(nv50->vtxbuf[i].buffer)->map)
-                        nouveau_bo_unmap(nouveau_bo(nv50->vtxbuf[i].buffer));
-}
-
 static void
 emit_b32_1(struct nouveau_channel *chan, void *data)
 {
@@ -658,12 +1006,13 @@ emit_prepare(struct nv50_context *nv50, struct nv50_vbo_emitctx *emit,
 
 		ve = &nv50->vtxelt[i];
 		vb = &nv50->vtxbuf[ve->vertex_buffer_index];
-		if (!(nv50->vbo_fifo & (1 << i)))
+		if (!(nv50->vbo_fifo & (1 << i)) || ve->instance_divisor)
 			continue;
 		n = emit->nr_ve++;
 
 		emit->stride[n] = vb->stride;
-		emit->map[n] = nouveau_bo(vb->buffer)->map +
+		emit->map[n] = (uint8_t *)nouveau_bo(vb->buffer)->map +
+			vb->buffer_offset +
 			(start * vb->stride + ve->src_offset);
 
 		desc = util_format_description(ve->src_format);
@@ -753,13 +1102,12 @@ nv50_push_arrays(struct nv50_context *nv50, unsigned start, unsigned count)
 
 		set_edgeflag(chan, tesla, &emit, 0); /* nr will be 1 */
 
-		BEGIN_RING(chan, tesla, NV50TCL_VERTEX_DATA | 0x40000000, dw);
+		BEGIN_RING_NI(chan, tesla, NV50TCL_VERTEX_DATA, dw);
 		for (i = 0; i < nr; ++i)
 			emit_vtx_next(chan, &emit);
 
 		count -= nr;
 	}
-	nv50_unmap_vbufs(nv50);
 
 	return TRUE;
 }
@@ -780,13 +1128,12 @@ nv50_push_elements_u32(struct nv50_context *nv50, uint32_t *map, unsigned count)
 
 		set_edgeflag(chan, tesla, &emit, *map);
 
-		BEGIN_RING(chan, tesla, NV50TCL_VERTEX_DATA | 0x40000000, dw);
+		BEGIN_RING_NI(chan, tesla, NV50TCL_VERTEX_DATA, dw);
 		for (i = 0; i < nr; ++i)
 			emit_vtx(chan, &emit, *map++);
 
 		count -= nr;
 	}
-	nv50_unmap_vbufs(nv50);
 
 	return TRUE;
 }
@@ -807,13 +1154,12 @@ nv50_push_elements_u16(struct nv50_context *nv50, uint16_t *map, unsigned count)
 
 		set_edgeflag(chan, tesla, &emit, *map);
 
-		BEGIN_RING(chan, tesla, NV50TCL_VERTEX_DATA | 0x40000000, dw);
+		BEGIN_RING_NI(chan, tesla, NV50TCL_VERTEX_DATA, dw);
 		for (i = 0; i < nr; ++i)
 			emit_vtx(chan, &emit, *map++);
 
 		count -= nr;
 	}
-	nv50_unmap_vbufs(nv50);
 
 	return TRUE;
 }
@@ -834,13 +1180,12 @@ nv50_push_elements_u08(struct nv50_context *nv50, uint8_t *map, unsigned count)
 
 		set_edgeflag(chan, tesla, &emit, *map);
 
-		BEGIN_RING(chan, tesla, NV50TCL_VERTEX_DATA | 0x40000000, dw);
+		BEGIN_RING_NI(chan, tesla, NV50TCL_VERTEX_DATA, dw);
 		for (i = 0; i < nr; ++i)
 			emit_vtx(chan, &emit, *map++);
 
 		count -= nr;
 	}
-	nv50_unmap_vbufs(nv50);
 
 	return TRUE;
 }
diff --git a/src/gallium/drivers/r300/r300_blit.c b/src/gallium/drivers/r300/r300_blit.c
index c14414fff6..cdedb30220 100644
--- a/src/gallium/drivers/r300/r300_blit.c
+++ b/src/gallium/drivers/r300/r300_blit.c
@@ -23,8 +23,6 @@
 #include "r300_blit.h"
 #include "r300_context.h"
 
-#include "util/u_rect.h"
-
 static void r300_blitter_save_states(struct r300_context* r300)
 {
     util_blitter_save_blend(r300->blitter, r300->blend_state.state);
@@ -75,13 +73,15 @@ void r300_clear(struct pipe_context* pipe,
      */
 
     struct r300_context* r300 = r300_context(pipe);
+    struct pipe_framebuffer_state* fb =
+        (struct pipe_framebuffer_state*)r300->fb_state.state;
 
     r300_blitter_save_states(r300);
 
     util_blitter_clear(r300->blitter,
-                       r300->framebuffer_state.width,
-                       r300->framebuffer_state.height,
-                       r300->framebuffer_state.nr_cbufs,
+                       fb->width,
+                       fb->height,
+                       fb->nr_cbufs,
                        buffers, rgba, depth, stencil);
 }
 
@@ -99,7 +99,7 @@ void r300_surface_copy(struct pipe_context* pipe,
      * is really transparent. The states will be restored by the blitter once
      * copying is done. */
     r300_blitter_save_states(r300);
-    util_blitter_save_framebuffer(r300->blitter, &r300->framebuffer_state);
+    util_blitter_save_framebuffer(r300->blitter, r300->fb_state.state);
 
     util_blitter_save_fragment_sampler_states(
         r300->blitter, r300->sampler_count, (void**)r300->sampler_states);
@@ -123,7 +123,7 @@ void r300_surface_fill(struct pipe_context* pipe,
     struct r300_context* r300 = r300_context(pipe);
 
     r300_blitter_save_states(r300);
-    util_blitter_save_framebuffer(r300->blitter, &r300->framebuffer_state);
+    util_blitter_save_framebuffer(r300->blitter, r300->fb_state.state);
 
     util_blitter_fill(r300->blitter,
                       dst, dstx, dsty, width, height, value);
diff --git a/src/gallium/drivers/r300/r300_context.c b/src/gallium/drivers/r300/r300_context.c
index 5e4f6552c3..9837530a7a 100644
--- a/src/gallium/drivers/r300/r300_context.c
+++ b/src/gallium/drivers/r300/r300_context.c
@@ -22,9 +22,6 @@
 
 #include "draw/draw_context.h"
 
-#include "tgsi/tgsi_scan.h"
-
-#include "util/u_hash_table.h"
 #include "util/u_memory.h"
 #include "util/u_simple_list.h"
 
@@ -35,30 +32,16 @@
 #include "r300_query.h"
 #include "r300_render.h"
 #include "r300_screen.h"
-#include "r300_state_derived.h"
 #include "r300_state_invariant.h"
 #include "r300_texture.h"
 #include "r300_winsys.h"
 
-static enum pipe_error r300_clear_hash_table(void* key, void* value,
-                                             void* data)
-{
-    FREE(key);
-    FREE(value);
-    return PIPE_OK;
-}
-
 static void r300_destroy_context(struct pipe_context* context)
 {
     struct r300_context* r300 = r300_context(context);
     struct r300_query* query, * temp;
 
     util_blitter_destroy(r300->blitter);
-
-    util_hash_table_foreach(r300->shader_hash_table, r300_clear_hash_table,
-        NULL);
-    util_hash_table_destroy(r300->shader_hash_table);
-
     draw_destroy(r300->draw);
 
     /* Free the OQ BO. */
@@ -72,9 +55,9 @@ static void r300_destroy_context(struct pipe_context* context)
 
     FREE(r300->blend_color_state.state);
     FREE(r300->clip_state.state);
-    FREE(r300->rs_block);
+    FREE(r300->rs_block_state.state);
     FREE(r300->scissor_state.state);
-    FREE(r300->vertex_info);
+    FREE(r300->vertex_format_state.state);
     FREE(r300->viewport_state.state);
     FREE(r300->ztop_state.state);
     FREE(r300);
@@ -87,7 +70,7 @@ r300_is_texture_referenced(struct pipe_context *pipe,
 {
     struct pipe_buffer* buf = 0;
 
-    r300_get_texture_buffer(texture, &buf, NULL);
+    r300_get_texture_buffer(pipe->screen, texture, &buf, NULL);
 
     return pipe->is_buffer_referenced(pipe, buf);
 }
@@ -129,14 +112,18 @@ static void r300_setup_atoms(struct r300_context* r300)
      * an upper bound on each atom, to keep the emission machinery from
      * underallocating space. */
     make_empty_list(&r300->atom_list);
+    R300_INIT_ATOM(invariant, 71);
     R300_INIT_ATOM(ztop, 2);
     R300_INIT_ATOM(blend, 8);
     R300_INIT_ATOM(blend_color, 3);
     R300_INIT_ATOM(clip, 29);
     R300_INIT_ATOM(dsa, 8);
-    R300_INIT_ATOM(rs, 22);
+    R300_INIT_ATOM(fb, 56);
+    R300_INIT_ATOM(rs, 25);
     R300_INIT_ATOM(scissor, 3);
     R300_INIT_ATOM(viewport, 9);
+    R300_INIT_ATOM(rs_block, 21);
+    R300_INIT_ATOM(vertex_format, 26);
 }
 
 struct pipe_context* r300_create_context(struct pipe_screen* screen,
@@ -153,8 +140,6 @@ struct pipe_context* r300_create_context(struct pipe_screen* screen,
     r300->context.winsys = (struct pipe_winsys*)radeon_winsys;
     r300->context.screen = screen;
 
-    r300_init_debug(r300);
-
     r300->context.destroy = r300_destroy_context;
 
     r300->context.clear = r300_clear;
@@ -184,16 +169,13 @@ struct pipe_context* r300_create_context(struct pipe_screen* screen,
     r300->context.is_texture_referenced = r300_is_texture_referenced;
     r300->context.is_buffer_referenced = r300_is_buffer_referenced;
 
-    r300->shader_hash_table = util_hash_table_create(r300_shader_key_hash,
-        r300_shader_key_compare);
-
     r300_setup_atoms(r300);
 
     r300->blend_color_state.state = CALLOC_STRUCT(r300_blend_color_state);
     r300->clip_state.state = CALLOC_STRUCT(pipe_clip_state);
-    r300->rs_block = CALLOC_STRUCT(r300_rs_block);
+    r300->rs_block_state.state = CALLOC_STRUCT(r300_rs_block);
     r300->scissor_state.state = CALLOC_STRUCT(pipe_scissor_state);
-    r300->vertex_info = CALLOC_STRUCT(r300_vertex_info);
+    r300->vertex_format_state.state = CALLOC_STRUCT(r300_vertex_info);
     r300->viewport_state.state = CALLOC_STRUCT(r300_viewport_state);
     r300->ztop_state.state = CALLOC_STRUCT(r300_ztop_state);
 
@@ -210,7 +192,7 @@ struct pipe_context* r300_create_context(struct pipe_screen* screen,
 
     r300_init_state_functions(r300);
 
-    r300_emit_invariant_state(r300);
+    r300->invariant_state.dirty = TRUE;
 
     r300->winsys->set_flush_cb(r300->winsys, r300_flush_cb, r300);
     r300->dirty_state = R300_NEW_KITCHEN_SINK;
diff --git a/src/gallium/drivers/r300/r300_context.h b/src/gallium/drivers/r300/r300_context.h
index 682b9179c8..3f461640a0 100644
--- a/src/gallium/drivers/r300/r300_context.h
+++ b/src/gallium/drivers/r300/r300_context.h
@@ -30,6 +30,8 @@
 #include "pipe/p_context.h"
 #include "pipe/p_inlines.h"
 
+#include "r300_screen.h"
+
 struct r300_context;
 
 struct r300_fragment_shader;
@@ -82,13 +84,14 @@ struct r300_rs_state {
     struct pipe_rasterizer_state rs;
 
     uint32_t vap_control_status;    /* R300_VAP_CNTL_STATUS: 0x2140 */
+    uint32_t antialiasing_config;   /* R300_GB_AA_CONFIG: 0x4020 */
     uint32_t point_size;            /* R300_GA_POINT_SIZE: 0x421c */
     uint32_t point_minmax;          /* R300_GA_POINT_MINMAX: 0x4230 */
     uint32_t line_control;          /* R300_GA_LINE_CNTL: 0x4234 */
-    uint32_t depth_scale_front;  /* R300_SU_POLY_OFFSET_FRONT_SCALE: 0x42a4 */
-    uint32_t depth_offset_front;/* R300_SU_POLY_OFFSET_FRONT_OFFSET: 0x42a8 */
-    uint32_t depth_scale_back;    /* R300_SU_POLY_OFFSET_BACK_SCALE: 0x42ac */
-    uint32_t depth_offset_back;  /* R300_SU_POLY_OFFSET_BACK_OFFSET: 0x42b0 */
+    float depth_scale;            /* R300_SU_POLY_OFFSET_FRONT_SCALE: 0x42a4 */
+                                  /* R300_SU_POLY_OFFSET_BACK_SCALE: 0x42ac */
+    float depth_offset;           /* R300_SU_POLY_OFFSET_FRONT_OFFSET: 0x42a8 */
+                                  /* R300_SU_POLY_OFFSET_BACK_OFFSET: 0x42b0 */
     uint32_t polygon_offset_enable; /* R300_SU_POLY_OFFSET_ENABLE: 0x42b4 */
     uint32_t cull_mode;             /* R300_SU_CULL_MODE: 0x42b8 */
     uint32_t line_stipple_config;   /* R300_GA_LINE_STIPPLE_CONFIG: 0x4328 */
@@ -136,15 +139,12 @@ struct r300_ztop_state {
     uint32_t z_buffer_top;      /* R300_ZB_ZTOP: 0x4f14 */
 };
 
-#define R300_NEW_FRAMEBUFFERS    0x00000010
 #define R300_NEW_FRAGMENT_SHADER 0x00000020
 #define R300_NEW_FRAGMENT_SHADER_CONSTANTS    0x00000040
-#define R300_NEW_RS_BLOCK        0x00000100
 #define R300_NEW_SAMPLER         0x00000200
 #define R300_ANY_NEW_SAMPLERS    0x0001fe00
 #define R300_NEW_TEXTURE         0x00040000
 #define R300_ANY_NEW_TEXTURES    0x03fc0000
-#define R300_NEW_VERTEX_FORMAT   0x04000000
 #define R300_NEW_VERTEX_SHADER   0x08000000
 #define R300_NEW_VERTEX_SHADER_CONSTANTS    0x10000000
 #define R300_NEW_QUERY           0x40000000
@@ -269,11 +269,8 @@ struct r300_context {
     struct r300_query *query_current;
     struct r300_query query_list;
 
-    /* Shader hash table. Used to store vertex formatting information, which
-     * depends on the combination of both currently loaded shaders. */
-    struct util_hash_table* shader_hash_table;
     /* Vertex formatting information. */
-    struct r300_vertex_info* vertex_info;
+    struct r300_atom vertex_format_state;
 
     /* Various CSO state objects. */
     /* Beginning of atom list. */
@@ -290,12 +287,12 @@ struct r300_context {
     struct r300_atom dsa_state;
     /* Fragment shader. */
     struct r300_fragment_shader* fs;
-    /* Framebuffer state. We currently don't need our own version of this. */
-    struct pipe_framebuffer_state framebuffer_state;
+    /* Framebuffer state. */
+    struct r300_atom fb_state;
     /* Rasterizer state. */
     struct r300_atom rs_state;
     /* RS block state. */
-    struct r300_rs_block* rs_block;
+    struct r300_atom rs_block_state;
     /* Sampler states. */
     struct r300_sampler_state* sampler_states[8];
     int sampler_count;
@@ -311,6 +308,9 @@ struct r300_context {
     /* ZTOP state. */
     struct r300_atom ztop_state;
 
+    /* Invariant state. This must be emitted to get the engine started. */
+    struct r300_atom invariant_state;
+
     /* Vertex buffers for Gallium. */
     struct pipe_vertex_buffer vertex_buffer[PIPE_MAX_ATTRIBS];
     int vertex_buffer_count;
@@ -324,9 +324,10 @@ struct r300_context {
     uint32_t dirty_hw;
     /* Whether the TCL engine should be in bypass mode. */
     boolean tcl_bypass;
-
-    /** Combination of DBG_xxx flags */
-    unsigned debug;
+    /* Whether polygon offset is enabled. */
+    boolean polygon_offset_enabled;
+    /* Z buffer bit depth. */
+    uint32_t zbuffer_bpp;
 };
 
 /* Convenience cast wrapper. */
@@ -340,35 +341,15 @@ struct draw_stage* r300_draw_stage(struct r300_context* r300);
 void r300_init_state_functions(struct r300_context* r300);
 void r300_init_surface_functions(struct r300_context* r300);
 
-/* Debug functionality. */
-
-/**
- * Debug flags to disable/enable certain groups of debugging outputs.
- *
- * \note These may be rather coarse, and the grouping may be impractical.
- * If you find, while debugging the driver, that a different grouping
- * of these flags would be beneficial, just feel free to change them
- * but make sure to update the documentation in r300_debug.c to reflect
- * those changes.
- */
-/*@{*/
-#define DBG_HELP    0x0000001
-#define DBG_FP      0x0000002
-#define DBG_VP      0x0000004
-#define DBG_CS      0x0000008
-#define DBG_DRAW    0x0000010
-#define DBG_TEX     0x0000020
-#define DBG_FALL    0x0000040
-/*@}*/
-
-static INLINE boolean DBG_ON(struct r300_context * ctx, unsigned flags)
+static INLINE boolean CTX_DBG_ON(struct r300_context * ctx, unsigned flags)
 {
-    return (ctx->debug & flags) ? TRUE : FALSE;
+    return SCREEN_DBG_ON(r300_screen(ctx->context.screen), flags);
 }
 
-static INLINE void DBG(struct r300_context * ctx, unsigned flags, const char * fmt, ...)
+static INLINE void CTX_DBG(struct r300_context * ctx, unsigned flags,
+                       const char * fmt, ...)
 {
-    if (DBG_ON(ctx, flags)) {
+    if (CTX_DBG_ON(ctx, flags)) {
         va_list va;
         va_start(va, fmt);
         debug_vprintf(fmt, va);
@@ -376,6 +357,8 @@ static INLINE void DBG(struct r300_context * ctx, unsigned flags, const char * f
     }
 }
 
-void r300_init_debug(struct r300_context * ctx);
+#define DBG_ON  CTX_DBG_ON
+#define DBG     CTX_DBG
 
 #endif /* R300_CONTEXT_H */
+
diff --git a/src/gallium/drivers/r300/r300_debug.c b/src/gallium/drivers/r300/r300_debug.c
index 2a6ed54ac9..b881730848 100644
--- a/src/gallium/drivers/r300/r300_debug.c
+++ b/src/gallium/drivers/r300/r300_debug.c
@@ -22,8 +22,6 @@
 
 #include "r300_context.h"
 
-#include <ctype.h>
-
 
 struct debug_option {
     const char * name;
@@ -46,7 +44,7 @@ static struct debug_option debug_options[] = {
     { 0, 0, 0 }
 };
 
-void r300_init_debug(struct r300_context * ctx)
+void r300_init_debug(struct r300_screen * screen)
 {
     const char * options = debug_get_option("RADEON_DEBUG", 0);
     boolean printhint = FALSE;
@@ -64,7 +62,7 @@ void r300_init_debug(struct r300_context * ctx)
 
             for(opt = debug_options; opt->name; ++opt) {
                 if (!strncmp(options, opt->name, length)) {
-                    ctx->debug |= opt->flag;
+                    screen->debug |= opt->flag;
                     break;
                 }
             }
@@ -77,11 +75,11 @@ void r300_init_debug(struct r300_context * ctx)
             options += length;
         }
 
-        if (!ctx->debug)
+        if (!screen->debug)
             printhint = TRUE;
     }
 
-    if (printhint || ctx->debug & DBG_HELP) {
+    if (printhint || screen->debug & DBG_HELP) {
         debug_printf("You can enable debug output by setting the RADEON_DEBUG environment variable\n"
                      "to a comma-separated list of debug options. Available options are:\n");
         for(opt = debug_options; opt->name; ++opt) {
diff --git a/src/gallium/drivers/r300/r300_emit.c b/src/gallium/drivers/r300/r300_emit.c
index 9f93327e59..a429457625 100644
--- a/src/gallium/drivers/r300/r300_emit.c
+++ b/src/gallium/drivers/r300/r300_emit.c
@@ -32,20 +32,20 @@
 #include "r300_emit.h"
 #include "r300_fs.h"
 #include "r300_screen.h"
-#include "r300_state_derived.h"
 #include "r300_state_inlines.h"
-#include "r300_texture.h"
 #include "r300_vs.h"
 
 void r300_emit_blend_state(struct r300_context* r300, void* state)
 {
     struct r300_blend_state* blend = (struct r300_blend_state*)state;
+    struct pipe_framebuffer_state* fb =
+        (struct pipe_framebuffer_state*)r300->fb_state.state;
     CS_LOCALS(r300);
 
     BEGIN_CS(8);
     OUT_CS_REG(R300_RB3D_ROPCNTL, blend->rop);
     OUT_CS_REG_SEQ(R300_RB3D_CBLEND, 3);
-    if (r300->framebuffer_state.nr_cbufs) {
+    if (fb->nr_cbufs) {
         OUT_CS(blend->blend_control);
         OUT_CS(blend->alpha_blend_control);
         OUT_CS(blend->color_channel_mask);
@@ -112,6 +112,8 @@ void r300_emit_dsa_state(struct r300_context* r300, void* state)
 {
     struct r300_dsa_state* dsa = (struct r300_dsa_state*)state;
     struct r300_screen* r300screen = r300_screen(r300->context.screen);
+    struct pipe_framebuffer_state* fb =
+        (struct pipe_framebuffer_state*)r300->fb_state.state;
     CS_LOCALS(r300);
 
     BEGIN_CS(r300screen->caps->is_r500 ? 8 : 6);
@@ -124,7 +126,7 @@ void r300_emit_dsa_state(struct r300_context* r300, void* state)
 
     OUT_CS_REG_SEQ(R300_ZB_CNTL, 3);
 
-    if (r300->framebuffer_state.zsbuf) {
+    if (fb->zsbuf) {
         OUT_CS(dsa->z_buffer_control);
         OUT_CS(dsa->z_stencil_control);
     } else {
@@ -168,9 +170,8 @@ static const float * get_shader_constant(
                     vec[1] = 1.0 / tex->height0;
                     break;
 
-                /* Texture compare-fail value. */
-                /* XXX Since Gallium doesn't support GL_ARB_shadow_ambient,
-                 * this is always (0,0,0,0), right? */
+                /* Texture compare-fail value. Shouldn't ever show up, but if
+                 * it does, we'll be ready. */
                 case RC_STATE_SHADOW_AMBIENT:
                     vec[3] = 0;
                     break;
@@ -384,17 +385,14 @@ void r500_emit_fs_constant_buffer(struct r300_context* r300,
     END_CS;
 }
 
-void r300_emit_fb_state(struct r300_context* r300,
-                        struct pipe_framebuffer_state* fb)
+void r300_emit_fb_state(struct r300_context* r300, void* state)
 {
+    struct pipe_framebuffer_state* fb = (struct pipe_framebuffer_state*)state;
     struct r300_texture* tex;
     struct pipe_surface* surf;
     int i;
     CS_LOCALS(r300);
 
-    /* Shouldn't fail unless there is a bug in the state tracker. */
-    assert(fb->nr_cbufs <= 4);
-
     BEGIN_CS((10 * fb->nr_cbufs) + (2 * (4 - fb->nr_cbufs)) +
              (fb->zsbuf ? 10 : 0) + 6);
 
@@ -407,7 +405,14 @@ void r300_emit_fb_state(struct r300_context* r300,
         R300_ZB_ZCACHE_CTLSTAT_ZC_FREE_FREE);
 
     /* Set the number of colorbuffers. */
-    OUT_CS_REG(R300_RB3D_CCTL, R300_RB3D_CCTL_NUM_MULTIWRITES(fb->nr_cbufs));
+    if (fb->nr_cbufs > 1) {
+        OUT_CS_REG(R300_RB3D_CCTL,
+            R300_RB3D_CCTL_NUM_MULTIWRITES(fb->nr_cbufs) |
+            R300_RB3D_CCTL_INDEPENDENT_COLOR_CHANNEL_MASK_ENABLE |
+            R300_RB3D_CCTL_INDEPENDENT_COLORFORMAT_ENABLE_ENABLE);
+    } else {
+        OUT_CS_REG(R300_RB3D_CCTL, 0x0);
+    }
 
     /* Set up colorbuffers. */
     for (i = 0; i < fb->nr_cbufs; i++) {
@@ -584,19 +589,40 @@ void r300_emit_query_end(struct r300_context* r300)
 void r300_emit_rs_state(struct r300_context* r300, void* state)
 {
     struct r300_rs_state* rs = (struct r300_rs_state*)state;
+    float scale, offset;
     CS_LOCALS(r300);
 
-    BEGIN_CS(22);
+    BEGIN_CS(20 + (rs->polygon_offset_enable ? 5 : 0));
     OUT_CS_REG(R300_VAP_CNTL_STATUS, rs->vap_control_status);
+
+    OUT_CS_REG(R300_GB_AA_CONFIG, rs->antialiasing_config);
+
     OUT_CS_REG(R300_GA_POINT_SIZE, rs->point_size);
     OUT_CS_REG_SEQ(R300_GA_POINT_MINMAX, 2);
     OUT_CS(rs->point_minmax);
     OUT_CS(rs->line_control);
-    OUT_CS_REG_SEQ(R300_SU_POLY_OFFSET_FRONT_SCALE, 6);
-    OUT_CS(rs->depth_scale_front);
-    OUT_CS(rs->depth_offset_front);
-    OUT_CS(rs->depth_scale_back);
-    OUT_CS(rs->depth_offset_back);
+
+    if (rs->polygon_offset_enable) {
+        scale = rs->depth_scale * 12;
+        offset = rs->depth_offset;
+
+        switch (r300->zbuffer_bpp) {
+            case 16:
+                offset *= 4;
+                break;
+            case 24:
+                offset *= 2;
+                break;
+        }
+
+        OUT_CS_REG_SEQ(R300_SU_POLY_OFFSET_FRONT_SCALE, 4);
+        OUT_CS_32F(scale);
+        OUT_CS_32F(offset);
+        OUT_CS_32F(scale);
+        OUT_CS_32F(offset);
+    }
+
+    OUT_CS_REG_SEQ(R300_SU_POLY_OFFSET_ENABLE, 2);
     OUT_CS(rs->polygon_offset_enable);
     OUT_CS(rs->cull_mode);
     OUT_CS_REG(R300_GA_LINE_STIPPLE_CONFIG, rs->line_stipple_config);
@@ -606,10 +632,10 @@ void r300_emit_rs_state(struct r300_context* r300, void* state)
     END_CS;
 }
 
-void r300_emit_rs_block_state(struct r300_context* r300,
-                              struct r300_rs_block* rs)
+void r300_emit_rs_block_state(struct r300_context* r300, void* state)
 {
-    int i;
+    struct r300_rs_block* rs = (struct r300_rs_block*)state;
+    unsigned i;
     struct r300_screen* r300screen = r300_screen(r300->context.screen);
     CS_LOCALS(r300);
 
@@ -652,11 +678,13 @@ void r300_emit_scissor_state(struct r300_context* r300, void* state)
     uint32_t top_left, bottom_right;
     struct r300_screen* r300screen = r300_screen(r300->context.screen);
     struct pipe_scissor_state* scissor = (struct pipe_scissor_state*)state;
+    struct pipe_framebuffer_state* fb =
+        (struct pipe_framebuffer_state*)r300->fb_state.state;
     CS_LOCALS(r300);
 
     minx = miny = 0;
-    maxx = r300->framebuffer_state.width;
-    maxy = r300->framebuffer_state.height;
+    maxx = fb->width;
+    maxy = fb->height;
 
     if (((struct r300_rs_state*)r300->rs_state.state)->rs.scissor) {
         minx = MAX2(minx, scissor->minx);
@@ -665,6 +693,22 @@ void r300_emit_scissor_state(struct r300_context* r300, void* state)
         maxy = MIN2(maxy, scissor->maxy);
     }
 
+    /* Special case for zero-area scissor.
+     *
+     * We can't allow the variables maxx and maxy to be zero because they are
+     * subtracted from later in the code, which would cause emitting ~0 and
+     * making the kernel checker angry.
+     *
+     * Let's consider we change maxx and maxy to 1, which is effectively
+     * a one-pixel area. We must then change minx and miny to a number which is
+     * greater than 1 to get the zero area back. */
+    if (!maxx || !maxy) {
+        minx = 2;
+        miny = 2;
+        maxx = 1;
+        maxy = 1;
+    }
+
     if (r300screen->caps->is_r500) {
         top_left =
             (minx << R300_SCISSORS_X_SHIFT) |
@@ -705,12 +749,18 @@ void r300_emit_texture(struct r300_context* r300,
         filter0 |= R300_TX_WRAP_T(R300_TX_CLAMP_TO_EDGE);
     }
 
-    /* determine min/max levels */
-    /* the MAX_MIP level is the largest (finest) one */
-    max_level = MIN2(sampler->max_lod, tex->tex.last_level);
-    min_level = MIN2(sampler->min_lod, max_level);
-    format0 |= R300_TX_NUM_LEVELS(max_level);
-    filter0 |= R300_TX_MAX_MIP_LEVEL(min_level);
+    if (tex->is_npot) {
+        /* NPOT textures don't support mip filter, unfortunately.
+         * This prevents incorrect rendering. */
+        filter0 &= ~R300_TX_MIN_FILTER_MIP_MASK;
+    } else {
+        /* determine min/max levels */
+        /* the MAX_MIP level is the largest (finest) one */
+        max_level = MIN2(sampler->max_lod, tex->tex.last_level);
+        min_level = MIN2(sampler->min_lod, max_level);
+        format0 |= R300_TX_NUM_LEVELS(max_level);
+        filter0 |= R300_TX_MAX_MIP_LEVEL(min_level);
+    }
 
     BEGIN_CS(16);
     OUT_CS_REG(R300_TX_FILTER0_0 + (offset * 4), filter0 |
@@ -729,22 +779,6 @@ void r300_emit_texture(struct r300_context* r300,
     END_CS;
 }
 
-static boolean r300_validate_aos(struct r300_context *r300)
-{
-    struct pipe_vertex_buffer *vbuf = r300->vertex_buffer;
-    struct pipe_vertex_element *velem = r300->vertex_element;
-    int i;
-
-    /* Check if formats and strides are aligned to the size of DWORD. */
-    for (i = 0; i < r300->vertex_element_count; i++) {
-        if (vbuf[velem[i].vertex_buffer_index].stride % 4 != 0 ||
-            util_format_get_blocksize(velem[i].src_format) % 4 != 0) {
-            return FALSE;
-        }
-    }
-    return TRUE;
-}
-
 void r300_emit_aos(struct r300_context* r300, unsigned offset)
 {
     struct pipe_vertex_buffer *vb1, *vb2, *vbuf = r300->vertex_buffer;
@@ -754,12 +788,6 @@ void r300_emit_aos(struct r300_context* r300, unsigned offset)
     unsigned packet_size = (aos_count * 3 + 1) / 2;
     CS_LOCALS(r300);
 
-    /* XXX Move this checking to a more approriate place. */
-    if (!r300_validate_aos(r300)) {
-        /* XXX We should fallback using Draw. */
-        assert(0);
-    }
-
     BEGIN_CS(2 + packet_size + aos_count * 2);
     OUT_CS_PKT3(R300_PACKET3_3D_LOAD_VBPNTR, packet_size);
     OUT_CS(aos_count);
@@ -791,38 +819,39 @@ void r300_emit_aos(struct r300_context* r300, unsigned offset)
     END_CS;
 }
 
-void r300_emit_vertex_format_state(struct r300_context* r300)
+void r300_emit_vertex_format_state(struct r300_context* r300, void* state)
 {
-    int i;
+    struct r300_vertex_info* vertex_info = (struct r300_vertex_info*)state;
+    unsigned i;
     CS_LOCALS(r300);
 
     DBG(r300, DBG_DRAW, "r300: VAP/PSC emit:\n");
 
     BEGIN_CS(26);
-    OUT_CS_REG(R300_VAP_VTX_SIZE, r300->vertex_info->vinfo.size);
+    OUT_CS_REG(R300_VAP_VTX_SIZE, vertex_info->vinfo.size);
 
     OUT_CS_REG_SEQ(R300_VAP_VTX_STATE_CNTL, 2);
-    OUT_CS(r300->vertex_info->vinfo.hwfmt[0]);
-    OUT_CS(r300->vertex_info->vinfo.hwfmt[1]);
+    OUT_CS(vertex_info->vinfo.hwfmt[0]);
+    OUT_CS(vertex_info->vinfo.hwfmt[1]);
     OUT_CS_REG_SEQ(R300_VAP_OUTPUT_VTX_FMT_0, 2);
-    OUT_CS(r300->vertex_info->vinfo.hwfmt[2]);
-    OUT_CS(r300->vertex_info->vinfo.hwfmt[3]);
+    OUT_CS(vertex_info->vinfo.hwfmt[2]);
+    OUT_CS(vertex_info->vinfo.hwfmt[3]);
     for (i = 0; i < 4; i++) {
        DBG(r300, DBG_DRAW, "    : hwfmt%d: 0x%08x\n", i,
-               r300->vertex_info->vinfo.hwfmt[i]);
+               vertex_info->vinfo.hwfmt[i]);
     }
 
     OUT_CS_REG_SEQ(R300_VAP_PROG_STREAM_CNTL_0, 8);
     for (i = 0; i < 8; i++) {
-        OUT_CS(r300->vertex_info->vap_prog_stream_cntl[i]);
+        OUT_CS(vertex_info->vap_prog_stream_cntl[i]);
         DBG(r300, DBG_DRAW, "    : prog_stream_cntl%d: 0x%08x\n", i,
-               r300->vertex_info->vap_prog_stream_cntl[i]);
+               vertex_info->vap_prog_stream_cntl[i]);
     }
     OUT_CS_REG_SEQ(R300_VAP_PROG_STREAM_CNTL_EXT_0, 8);
     for (i = 0; i < 8; i++) {
-        OUT_CS(r300->vertex_info->vap_prog_stream_cntl_ext[i]);
+        OUT_CS(vertex_info->vap_prog_stream_cntl_ext[i]);
         DBG(r300, DBG_DRAW, "    : prog_stream_cntl_ext%d: 0x%08x\n", i,
-               r300->vertex_info->vap_prog_stream_cntl_ext[i]);
+               vertex_info->vap_prog_stream_cntl_ext[i]);
     }
     END_CS;
 }
@@ -987,39 +1016,21 @@ static void r300_flush_pvs(struct r300_context* r300)
     END_CS;
 }
 
-/* Emit all dirty state. */
-void r300_emit_dirty_state(struct r300_context* r300)
+void r300_emit_buffer_validate(struct r300_context *r300)
 {
-    struct r300_screen* r300screen = r300_screen(r300->context.screen);
+    struct pipe_framebuffer_state* fb =
+        (struct pipe_framebuffer_state*)r300->fb_state.state;
     struct r300_texture* tex;
-    struct r300_atom* atom;
-    unsigned i, dwords = 1024;
-    int dirty_tex = 0;
+    unsigned i;
     boolean invalid = FALSE;
 
-    /* Check the required number of dwords against the space remaining in the
-     * current CS object. If we need more, then flush. */
-
-    foreach(atom, &r300->atom_list) {
-        if (atom->dirty || atom->always_dirty) {
-            dwords += atom->size;
-        }
-    }
-
-    /* Make sure we have at least 2*1024 spare dwords. */
-    /* XXX It would be nice to know the number of dwords we really need to
-     * XXX emit. */
-    if (!r300->winsys->check_cs(r300->winsys, dwords)) {
-        r300->context.flush(&r300->context, 0, NULL);
-    }
-
     /* Clean out BOs. */
     r300->winsys->reset_bos(r300->winsys);
 
 validate:
     /* Color buffers... */
-    for (i = 0; i < r300->framebuffer_state.nr_cbufs; i++) {
-        tex = (struct r300_texture*)r300->framebuffer_state.cbufs[i]->texture;
+    for (i = 0; i < fb->nr_cbufs; i++) {
+        tex = (struct r300_texture*)fb->cbufs[i]->texture;
         assert(tex && tex->buffer && "cbuf is marked, but NULL!");
         if (!r300->winsys->add_buffer(r300->winsys, tex->buffer,
                     0, RADEON_GEM_DOMAIN_VRAM)) {
@@ -1028,8 +1039,8 @@ validate:
         }
     }
     /* ...depth buffer... */
-    if (r300->framebuffer_state.zsbuf) {
-        tex = (struct r300_texture*)r300->framebuffer_state.zsbuf->texture;
+    if (fb->zsbuf) {
+        tex = (struct r300_texture*)fb->zsbuf->texture;
         assert(tex && tex->buffer && "zsbuf is marked, but NULL!");
         if (!r300->winsys->add_buffer(r300->winsys, tex->buffer,
                     0, RADEON_GEM_DOMAIN_VRAM)) {
@@ -1076,6 +1087,31 @@ validate:
         invalid = TRUE;
         goto validate;
     }
+}
+
+/* Emit all dirty state. */
+void r300_emit_dirty_state(struct r300_context* r300)
+{
+    struct r300_screen* r300screen = r300_screen(r300->context.screen);
+    struct r300_atom* atom;
+    unsigned i, dwords = 1024;
+    int dirty_tex = 0;
+
+    /* Check the required number of dwords against the space remaining in the
+     * current CS object. If we need more, then flush. */
+
+    foreach(atom, &r300->atom_list) {
+        if (atom->dirty || atom->always_dirty) {
+            dwords += atom->size;
+        }
+    }
+
+    /* Make sure we have at least 2*1024 spare dwords. */
+    /* XXX It would be nice to know the number of dwords we really need to
+     * XXX emit. */
+    while (!r300->winsys->check_cs(r300->winsys, dwords)) {
+        r300->context.flush(&r300->context, 0, NULL);
+    }
 
     if (r300->dirty_state & R300_NEW_QUERY) {
         r300_emit_query_start(r300);
@@ -1110,16 +1146,6 @@ validate:
         r300->dirty_state &= ~R300_NEW_FRAGMENT_SHADER_CONSTANTS;
     }
 
-    if (r300->dirty_state & R300_NEW_FRAMEBUFFERS) {
-        r300_emit_fb_state(r300, &r300->framebuffer_state);
-        r300->dirty_state &= ~R300_NEW_FRAMEBUFFERS;
-    }
-
-    if (r300->dirty_state & R300_NEW_RS_BLOCK) {
-        r300_emit_rs_block_state(r300, r300->rs_block);
-        r300->dirty_state &= ~R300_NEW_RS_BLOCK;
-    }
-
     /* Samplers and textures are tracked separately but emitted together. */
     if (r300->dirty_state &
             (R300_ANY_NEW_SAMPLERS | R300_ANY_NEW_TEXTURES)) {
@@ -1145,11 +1171,6 @@ validate:
         r300_flush_textures(r300);
     }
 
-    if (r300->dirty_state & R300_NEW_VERTEX_FORMAT) {
-        r300_emit_vertex_format_state(r300);
-        r300->dirty_state &= ~R300_NEW_VERTEX_FORMAT;
-    }
-
     if (r300->dirty_state & (R300_NEW_VERTEX_SHADER | R300_NEW_VERTEX_SHADER_CONSTANTS)) {
         r300_flush_pvs(r300);
     }
diff --git a/src/gallium/drivers/r300/r300_emit.h b/src/gallium/drivers/r300/r300_emit.h
index 05a6bfeae8..6b96d9b57c 100644
--- a/src/gallium/drivers/r300/r300_emit.h
+++ b/src/gallium/drivers/r300/r300_emit.h
@@ -51,8 +51,7 @@ void r500_emit_fragment_program_code(struct r300_context* r300,
 void r500_emit_fs_constant_buffer(struct r300_context* r300,
                                   struct rc_constant_list* constants);
 
-void r300_emit_fb_state(struct r300_context* r300,
-                        struct pipe_framebuffer_state* fb);
+void r300_emit_fb_state(struct r300_context* r300, void* state);
 
 void r300_emit_query_begin(struct r300_context* r300,
                            struct r300_query* query);
@@ -61,8 +60,7 @@ void r300_emit_query_end(struct r300_context* r300);
 
 void r300_emit_rs_state(struct r300_context* r300, void* state);
 
-void r300_emit_rs_block_state(struct r300_context* r300,
-                              struct r300_rs_block* rs);
+void r300_emit_rs_block_state(struct r300_context* r300, void* state);
 
 void r300_emit_scissor_state(struct r300_context* r300, void* state);
 
@@ -73,7 +71,7 @@ void r300_emit_texture(struct r300_context* r300,
 
 void r300_emit_vertex_buffer(struct r300_context* r300);
 
-void r300_emit_vertex_format_state(struct r300_context* r300);
+void r300_emit_vertex_format_state(struct r300_context* r300, void* state);
 
 void r300_emit_vertex_program_code(struct r300_context* r300,
                                    struct r300_vertex_program_code* code);
@@ -95,4 +93,6 @@ void r300_flush_textures(struct r300_context* r300);
 /* Emit all dirty state. */
 void r300_emit_dirty_state(struct r300_context* r300);
 
+void r300_emit_buffer_validate(struct r300_context *r300);
+
 #endif /* R300_EMIT_H */
diff --git a/src/gallium/drivers/r300/r300_flush.c b/src/gallium/drivers/r300/r300_flush.c
index 59819cb106..e37d309270 100644
--- a/src/gallium/drivers/r300/r300_flush.c
+++ b/src/gallium/drivers/r300/r300_flush.c
@@ -29,7 +29,6 @@
 #include "r300_cs.h"
 #include "r300_emit.h"
 #include "r300_flush.h"
-#include "r300_state_invariant.h"
 
 static void r300_flush(struct pipe_context* pipe,
                        unsigned flags,
@@ -53,7 +52,6 @@ static void r300_flush(struct pipe_context* pipe,
 
     if (r300->dirty_hw) {
         FLUSH_CS;
-        r300_emit_invariant_state(r300);
         r300->dirty_state = R300_NEW_KITCHEN_SINK;
         r300->dirty_hw = 0;
 
diff --git a/src/gallium/drivers/r300/r300_render.c b/src/gallium/drivers/r300/r300_render.c
index 710d850163..95b7cb5fa4 100644
--- a/src/gallium/drivers/r300/r300_render.c
+++ b/src/gallium/drivers/r300/r300_render.c
@@ -26,10 +26,9 @@
 #include "draw/draw_context.h"
 #include "draw/draw_vbuf.h"
 
-#include "indices/u_indices.h"
-
 #include "pipe/p_inlines.h"
 
+#include "util/u_format.h"
 #include "util/u_memory.h"
 #include "util/u_prim.h"
 
@@ -116,20 +115,60 @@ static uint32_t r300_provoking_vertex_fixes(struct r300_context *r300,
     return color_control;
 }
 
-static void r300_emit_draw_immediate(struct r300_context *r300,
-                                     unsigned mode,
-                                     unsigned start,
-                                     unsigned count)
+static boolean immd_is_good_idea(struct r300_context *r300,
+                                      unsigned count)
 {
-    struct pipe_buffer* vbo = r300->vertex_buffer[0].buffer;
-    unsigned vertex_size = r300->vertex_buffer[0].stride / sizeof(float);
-    unsigned i;
-    uint32_t* map;
+    return count <= 4;
+}
+
+static void r300_emit_draw_arrays_immediate(struct r300_context *r300,
+                                            unsigned mode,
+                                            unsigned start,
+                                            unsigned count)
+{
+    struct pipe_vertex_element* velem;
+    struct pipe_vertex_buffer* vbuf;
+    unsigned vertex_element_count = r300->vertex_element_count;
+    unsigned i, v, vbi, dw, elem_offset;
+
+    /* Size of the vertex, in dwords. */
+    unsigned vertex_size = 0;
+
+    /* Offsets of the attribute, in dwords, from the start of the vertex. */
+    unsigned offset[PIPE_MAX_ATTRIBS];
+
+    /* Size of the vertex element, in dwords. */
+    unsigned size[PIPE_MAX_ATTRIBS];
+
+    /* Stride to the same attrib in the next vertex in the vertex buffer,
+     * in dwords. */
+    unsigned stride[PIPE_MAX_ATTRIBS];
+
+    /* Mapped vertex buffers. */
+    uint32_t* map[PIPE_MAX_ATTRIBS] = {0};
+
     CS_LOCALS(r300);
 
-    map = (uint32_t*)pipe_buffer_map_range(r300->context.screen, vbo,
-            start * vertex_size, count * vertex_size,
-            PIPE_BUFFER_USAGE_CPU_READ);
+    /* Calculate the vertex size, offsets, strides etc. and map the buffers. */
+    for (i = 0; i < vertex_element_count; i++) {
+        velem = &r300->vertex_element[i];
+        offset[i] = velem->src_offset / 4;
+        size[i] = util_format_get_blocksize(velem->src_format) / 4;
+        vertex_size += size[i];
+        vbi = velem->vertex_buffer_index;
+
+        /* Map the buffer. */
+        if (!map[vbi]) {
+            vbuf = &r300->vertex_buffer[vbi];
+            map[vbi] = (uint32_t*)pipe_buffer_map(r300->context.screen,
+                                                  vbuf->buffer,
+                                                  PIPE_BUFFER_USAGE_CPU_READ);
+            map[vbi] += vbuf->buffer_offset / 4;
+            stride[vbi] = vbuf->stride / 4;
+        }
+    }
+
+    r300_emit_dirty_state(r300);
 
     BEGIN_CS(10 + count * vertex_size);
     OUT_CS_REG(R300_GA_COLOR_CONTROL,
@@ -140,18 +179,31 @@ static void r300_emit_draw_immediate(struct r300_context *r300,
     OUT_CS_PKT3(R300_PACKET3_3D_DRAW_IMMD_2, count * vertex_size);
     OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_EMBEDDED | (count << 16) |
             r300_translate_primitive(mode));
-    //debug_printf("r300: Immd %d verts, %d attrs\n", count, vertex_size);
-    for (i = 0; i < count * vertex_size; i++) {
-        if (i % vertex_size == 0) {
-            //debug_printf("r300: -- vert --\n");
+
+    /* Emit vertices. */
+    for (v = 0; v < count; v++) {
+        for (i = 0; i < vertex_element_count; i++) {
+            velem = &r300->vertex_element[i];
+            vbi = velem->vertex_buffer_index;
+            elem_offset = offset[i] + stride[vbi] * (v + start);
+
+            for (dw = 0; dw < size[i]; dw++) {
+                OUT_CS(map[vbi][elem_offset + dw]);
+            }
         }
-        //debug_printf("r300: 0x%08x\n", *map);
-        OUT_CS(*map);
-        map++;
     }
     END_CS;
 
-    pipe_buffer_unmap(r300->context.screen, vbo);
+    /* Unmap buffers. */
+    for (i = 0; i < vertex_element_count; i++) {
+        vbi = r300->vertex_element[i].vertex_buffer_index;
+
+        if (map[vbi]) {
+            vbuf = &r300->vertex_buffer[vbi];
+            pipe_buffer_unmap(r300->context.screen, vbuf->buffer);
+            map[vbi] = NULL;
+        }
+    }
 }
 
 static void r300_emit_draw_arrays(struct r300_context *r300,
@@ -223,17 +275,18 @@ static void r300_emit_draw_elements(struct r300_context *r300,
     END_CS;
 }
 
-
 static boolean r300_setup_vertex_buffers(struct r300_context *r300)
 {
     struct pipe_vertex_buffer *vbuf = r300->vertex_buffer;
     struct pipe_vertex_element *velem = r300->vertex_element;
+    struct pipe_buffer *pbuf;
 
 validate:
     for (int i = 0; i < r300->vertex_element_count; i++) {
-        if (!r300->winsys->add_buffer(r300->winsys,
-                vbuf[velem[i].vertex_buffer_index].buffer,
-            RADEON_GEM_DOMAIN_GTT, 0)) {
+        pbuf = vbuf[velem[i].vertex_buffer_index].buffer;
+
+        if (!r300->winsys->add_buffer(r300->winsys, pbuf,
+                                      RADEON_GEM_DOMAIN_GTT, 0)) {
             r300->context.flush(&r300->context, 0, NULL);
             goto validate;
         }
@@ -247,40 +300,35 @@ validate:
     return TRUE;
 }
 
-static struct pipe_buffer* r300_translate_elts(struct r300_context* r300,
-                                               struct pipe_buffer* elts,
-                                               unsigned* size,
-                                               unsigned* mode,
-                                               unsigned* count)
+static void r300_shorten_ubyte_elts(struct r300_context* r300,
+                                    struct pipe_buffer** elts,
+                                    unsigned count)
 {
     struct pipe_screen* screen = r300->context.screen;
     struct pipe_buffer* new_elts;
-    void *in_map, *out_map;
-    unsigned out_prim, out_index_size, out_nr;
-    u_translate_func out_translate;
-
-    (void)u_index_translator(~0, *mode, *size, *count, PV_LAST, PV_LAST,
-        &out_prim, &out_index_size, &out_nr, &out_translate);
+    unsigned char *in_map;
+    unsigned short *out_map;
+    unsigned i;
 
     new_elts = screen->buffer_create(screen, 32,
                                      PIPE_BUFFER_USAGE_INDEX |
                                      PIPE_BUFFER_USAGE_CPU_WRITE |
                                      PIPE_BUFFER_USAGE_GPU_READ,
-                                     out_index_size * out_nr);
+                                     2 * count);
 
-    in_map = pipe_buffer_map(screen, elts, PIPE_BUFFER_USAGE_CPU_READ);
+    in_map = pipe_buffer_map(screen, *elts, PIPE_BUFFER_USAGE_CPU_READ);
     out_map = pipe_buffer_map(screen, new_elts, PIPE_BUFFER_USAGE_CPU_WRITE);
 
-    out_translate(in_map, *count, out_map);
+    for (i = 0; i < count; i++) {
+        *out_map = (unsigned short)*in_map;
+        in_map++;
+        out_map++;
+    }
 
-    pipe_buffer_unmap(screen, elts);
+    pipe_buffer_unmap(screen, *elts);
     pipe_buffer_unmap(screen, new_elts);
 
-    *size = out_index_size;
-    *mode = out_prim;
-    *count = out_nr;
-
-    return new_elts;
+    *elts = new_elts;
 }
 
 /* This is the fast-path drawing & emission for HW TCL. */
@@ -309,13 +357,15 @@ void r300_draw_range_elements(struct pipe_context* pipe,
 
     r300_update_derived_state(r300);
 
+    r300_emit_buffer_validate(r300);
+
     if (!r300_setup_vertex_buffers(r300)) {
         return;
     }
 
     if (indexSize == 1) {
-        indexBuffer = r300_translate_elts(r300, indexBuffer,
-            &indexSize, &mode, &count);
+        r300_shorten_ubyte_elts(r300, &indexBuffer, count);
+        indexSize = 2;
     }
 
     if (!r300->winsys->add_buffer(r300->winsys, indexBuffer,
@@ -368,15 +418,16 @@ void r300_draw_arrays(struct pipe_context* pipe, unsigned mode,
 
     r300_update_derived_state(r300);
 
-    if (!r300_setup_vertex_buffers(r300)) {
-        return;
-    }
-
-    r300_emit_dirty_state(r300);
+    r300_emit_buffer_validate(r300);
 
-    if (FALSE && count <= 4 && r300->vertex_buffer_count == 1) {
-        r300_emit_draw_immediate(r300, mode, start, count);
+    if (immd_is_good_idea(r300, count)) {
+        r300_emit_draw_arrays_immediate(r300, mode, start, count);
     } else {
+        if (!r300_setup_vertex_buffers(r300)) {
+            return;
+        }
+
+        r300_emit_dirty_state(r300);
         r300_emit_aos(r300, start);
         r300_emit_draw_arrays(r300, mode, count);
     }
@@ -411,6 +462,7 @@ void r300_swtcl_draw_arrays(struct pipe_context* pipe,
 
     draw_set_mapped_constant_buffer(r300->draw,
 				    PIPE_SHADER_VERTEX,
+                                    0,
 				    r300->shader_constants[PIPE_SHADER_VERTEX].constants,
 				    r300->shader_constants[PIPE_SHADER_VERTEX].count *
                 (sizeof(float) * 4));
@@ -455,6 +507,7 @@ void r300_swtcl_draw_range_elements(struct pipe_context* pipe,
 
     draw_set_mapped_constant_buffer(r300->draw,
 				    PIPE_SHADER_VERTEX,
+                                    0,
             r300->shader_constants[PIPE_SHADER_VERTEX].constants,
             r300->shader_constants[PIPE_SHADER_VERTEX].count *
                 (sizeof(float) * 4));
@@ -506,7 +559,7 @@ r300_render_get_vertex_info(struct vbuf_render* render)
 
     r300_update_derived_state(r300);
 
-    return &r300->vertex_info->vinfo;
+    return (struct vertex_info*)r300->vertex_format_state.state;
 }
 
 static boolean r300_render_allocate_vertices(struct vbuf_render* render,
diff --git a/src/gallium/drivers/r300/r300_screen.c b/src/gallium/drivers/r300/r300_screen.c
index 287664b1d2..fa8ed581da 100644
--- a/src/gallium/drivers/r300/r300_screen.c
+++ b/src/gallium/drivers/r300/r300_screen.c
@@ -113,6 +113,8 @@ static int r300_get_param(struct pipe_screen* pscreen, int param)
              * ~ C.
              */
             return 1;
+        case PIPE_CAP_DUAL_SOURCE_BLEND:
+            return 0;
         case PIPE_CAP_ANISOTROPIC_FILTER:
             return 1;
         case PIPE_CAP_POINT_SPRITE:
@@ -149,6 +151,20 @@ static int r300_get_param(struct pipe_screen* pscreen, int param)
             } else {
                 return 0;
             }
+        case PIPE_CAP_INDEP_BLEND_ENABLE:
+            if (r300screen->caps->is_r500) {
+                return 1;
+            } else {
+                return 0;
+            }
+        case PIPE_CAP_INDEP_BLEND_FUNC:
+            return 0;
+        case PIPE_CAP_TGSI_FS_COORD_ORIGIN_UPPER_LEFT:
+        case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_HALF_INTEGER:
+	    return 1;
+        case PIPE_CAP_TGSI_FS_COORD_ORIGIN_LOWER_LEFT:
+        case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_INTEGER:
+            return 0;
         default:
             debug_printf("r300: Implementation error: Bad param %d\n",
                 param);
@@ -183,10 +199,20 @@ static float r300_get_paramf(struct pipe_screen* pscreen, int param)
     }
 }
 
-static boolean check_tex_format(enum pipe_format format, uint32_t usage,
-                                boolean is_r500)
+static boolean r300_is_format_supported(struct pipe_screen* screen,
+                                        enum pipe_format format,
+                                        enum pipe_texture_target target,
+                                        unsigned usage,
+                                        unsigned geom_flags)
 {
     uint32_t retval = 0;
+    boolean is_r500 = r300_screen(screen)->caps->is_r500;
+
+    if (target >= PIPE_MAX_TEXTURE_TYPES) {
+        debug_printf("r300: Implementation error: Received bogus texture "
+            "target %d in %s\n", target, __FUNCTION__);
+        return FALSE;
+    }
 
     switch (format) {
         /* Supported formats. */
@@ -194,6 +220,8 @@ static boolean check_tex_format(enum pipe_format format, uint32_t usage,
         case PIPE_FORMAT_A4R4G4B4_UNORM:
         case PIPE_FORMAT_R5G6B5_UNORM:
         case PIPE_FORMAT_A1R5G5B5_UNORM:
+        case PIPE_FORMAT_A8_UNORM:
+        case PIPE_FORMAT_L8_UNORM:
             retval = usage &
                 (PIPE_TEXTURE_USAGE_RENDER_TARGET |
                  PIPE_TEXTURE_USAGE_DISPLAY_TARGET |
@@ -208,7 +236,6 @@ static boolean check_tex_format(enum pipe_format format, uint32_t usage,
         case PIPE_FORMAT_DXT3_RGBA:
         case PIPE_FORMAT_DXT5_RGBA:
         case PIPE_FORMAT_YCBCR:
-        case PIPE_FORMAT_L8_UNORM:
         case PIPE_FORMAT_A8L8_UNORM:
             retval = usage & PIPE_TEXTURE_USAGE_SAMPLER;
             break;
@@ -247,28 +274,13 @@ static boolean check_tex_format(enum pipe_format format, uint32_t usage,
         case PIPE_FORMAT_Z32_UNORM:
         case PIPE_FORMAT_S8Z24_UNORM:
         case PIPE_FORMAT_X8Z24_UNORM:
-            debug_printf("r300: Note: Got unsupported format: %s in %s\n",
-                pf_name(format), __FUNCTION__);
+            SCREEN_DBG(r300_screen(screen), DBG_TEX,
+                       "r300: Note: Got unsupported format: %s in %s\n",
+                       pf_name(format), __FUNCTION__);
             return FALSE;
 
-        /* XXX These don't even exist
-        case PIPE_FORMAT_A32R32G32B32:
-        case PIPE_FORMAT_A16R16G16B16: */
-        /* XXX What the deuce is UV88? (r3xx accel page 14)
-            debug_printf("r300: Warning: Got unimplemented format: %s in %s\n",
-                pf_name(format), __FUNCTION__);
-            return FALSE; */
-
-        /* XXX Supported yet unimplemented r5xx formats: */
-        /* XXX Again, what is UV1010 this time? (r5xx accel page 148) */
-        /* XXX Even more that don't exist
-        case PIPE_FORMAT_A10R10G10B10_UNORM:
-        case PIPE_FORMAT_A2R10G10B10_UNORM:
-        case PIPE_FORMAT_I10_UNORM:
-            debug_printf(
-                "r300: Warning: Got unimplemented r500 format: %s in %s\n",
-                pf_name(format), __FUNCTION__);
-            return FALSE; */
+        /* XXX Add all remaining gallium-supported formats,
+         * see util/u_format.csv. */
 
         default:
             /* Unknown format... */
@@ -286,30 +298,6 @@ static boolean check_tex_format(enum pipe_format format, uint32_t usage,
     return (retval >= usage);
 }
 
-static boolean r300_is_format_supported(struct pipe_screen* pscreen,
-                                        enum pipe_format format,
-                                        enum pipe_texture_target target,
-                                        unsigned tex_usage,
-                                        unsigned geom_flags)
-{
-    switch (target) {
-        case PIPE_TEXTURE_1D:   /* handle 1D textures as 2D ones */
-        case PIPE_TEXTURE_2D:
-        case PIPE_TEXTURE_3D:
-        case PIPE_TEXTURE_CUBE:
-            return check_tex_format(format, tex_usage,
-                r300_screen(pscreen)->caps->is_r500);
-
-        default:
-            debug_printf("r300: Fatal: This is not a format target: %d\n",
-                target);
-            assert(0);
-            break;
-    }
-
-    return FALSE;
-}
-
 static struct pipe_transfer*
 r300_get_tex_transfer(struct pipe_screen *screen,
                       struct pipe_texture *texture,
@@ -319,6 +307,7 @@ r300_get_tex_transfer(struct pipe_screen *screen,
 {
     struct r300_texture *tex = (struct r300_texture *)texture;
     struct r300_transfer *trans;
+    struct r300_screen *rscreen = r300_screen(screen);
     unsigned offset;
 
     offset = r300_texture_get_offset(tex, level, zslice, face);  /* in bytes */
@@ -330,11 +319,8 @@ r300_get_tex_transfer(struct pipe_screen *screen,
         trans->transfer.y = y;
         trans->transfer.width = w;
         trans->transfer.height = h;
-        trans->transfer.stride = r300_texture_get_stride(tex, level);
+        trans->transfer.stride = r300_texture_get_stride(rscreen, tex, level);
         trans->transfer.usage = usage;
-
-        /* XXX not sure whether it's required to set these two,
-               the driver doesn't use them */
         trans->transfer.zslice = zslice;
         trans->transfer.face = face;
 
@@ -396,6 +382,7 @@ struct pipe_screen* r300_create_screen(struct radeon_winsys* radeon_winsys)
     caps->num_frag_pipes = radeon_winsys->gb_pipes;
     caps->num_z_pipes = radeon_winsys->z_pipes;
 
+    r300_init_debug(r300screen);
     r300_parse_chipset(caps);
 
     r300screen->caps = caps;
diff --git a/src/gallium/drivers/r300/r300_screen.h b/src/gallium/drivers/r300/r300_screen.h
index 2217988add..580fda3984 100644
--- a/src/gallium/drivers/r300/r300_screen.h
+++ b/src/gallium/drivers/r300/r300_screen.h
@@ -35,6 +35,9 @@ struct r300_screen {
 
     /* Chipset capabilities */
     struct r300_capabilities* caps;
+
+    /** Combination of DBG_xxx flags */
+    unsigned debug;
 };
 
 struct r300_transfer {
@@ -60,4 +63,44 @@ r300_transfer(struct pipe_transfer* transfer)
 /* Creates a new r300 screen. */
 struct pipe_screen* r300_create_screen(struct radeon_winsys* radeon_winsys);
 
+/* Debug functionality. */
+
+/**
+ * Debug flags to disable/enable certain groups of debugging outputs.
+ *
+ * \note These may be rather coarse, and the grouping may be impractical.
+ * If you find, while debugging the driver, that a different grouping
+ * of these flags would be beneficial, just feel free to change them
+ * but make sure to update the documentation in r300_debug.c to reflect
+ * those changes.
+ */
+/*@{*/
+#define DBG_HELP    0x0000001
+#define DBG_FP      0x0000002
+#define DBG_VP      0x0000004
+#define DBG_CS      0x0000008
+#define DBG_DRAW    0x0000010
+#define DBG_TEX     0x0000020
+#define DBG_FALL    0x0000040
+/*@}*/
+
+static INLINE boolean SCREEN_DBG_ON(struct r300_screen * screen, unsigned flags)
+{
+    return (screen->debug & flags) ? TRUE : FALSE;
+}
+
+static INLINE void SCREEN_DBG(struct r300_screen * screen, unsigned flags,
+                              const char * fmt, ...)
+{
+    if (SCREEN_DBG_ON(screen, flags)) {
+        va_list va;
+        va_start(va, fmt);
+        debug_vprintf(fmt, va);
+        va_end(va);
+    }
+}
+
+void r300_init_debug(struct r300_screen* ctx);
+
 #endif /* R300_SCREEN_H */
+
diff --git a/src/gallium/drivers/r300/r300_state.c b/src/gallium/drivers/r300/r300_state.c
index 60ad763cf4..a5c0869066 100644
--- a/src/gallium/drivers/r300/r300_state.c
+++ b/src/gallium/drivers/r300/r300_state.c
@@ -30,7 +30,6 @@
 #include "tgsi/tgsi_parse.h"
 
 #include "pipe/p_config.h"
-#include "pipe/internal/p_winsys_screen.h"
 
 #include "r300_context.h"
 #include "r300_reg.h"
@@ -162,17 +161,18 @@ static boolean blend_discard_if_src_alpha_color_1(unsigned srcRGB, unsigned srcA
 static void* r300_create_blend_state(struct pipe_context* pipe,
                                      const struct pipe_blend_state* state)
 {
+    struct r300_screen* r300screen = r300_screen(pipe->screen);
     struct r300_blend_state* blend = CALLOC_STRUCT(r300_blend_state);
 
-    if (state->blend_enable)
+    if (state->rt[0].blend_enable)
     {
-        unsigned eqRGB = state->rgb_func;
-        unsigned srcRGB = state->rgb_src_factor;
-        unsigned dstRGB = state->rgb_dst_factor;
+        unsigned eqRGB = state->rt[0].rgb_func;
+        unsigned srcRGB = state->rt[0].rgb_src_factor;
+        unsigned dstRGB = state->rt[0].rgb_dst_factor;
 
-        unsigned eqA = state->alpha_func;
-        unsigned srcA = state->alpha_src_factor;
-        unsigned dstA = state->alpha_dst_factor;
+        unsigned eqA = state->rt[0].alpha_func;
+        unsigned srcA = state->rt[0].alpha_src_factor;
+        unsigned dstA = state->rt[0].alpha_dst_factor;
 
         /* despite the name, ALPHA_BLEND_ENABLE has nothing to do with alpha,
          * this is just the crappy D3D naming */
@@ -289,18 +289,18 @@ static void* r300_create_blend_state(struct pipe_context* pipe,
                 (state->logicop_func) << R300_RB3D_ROPCNTL_ROP_SHIFT;
     }
 
-    /* Color Channel Mask */
-    if (state->colormask & PIPE_MASK_R) {
-        blend->color_channel_mask |= RB3D_COLOR_CHANNEL_MASK_RED_MASK0;
-    }
-    if (state->colormask & PIPE_MASK_G) {
-        blend->color_channel_mask |= RB3D_COLOR_CHANNEL_MASK_GREEN_MASK0;
-    }
-    if (state->colormask & PIPE_MASK_B) {
-        blend->color_channel_mask |= RB3D_COLOR_CHANNEL_MASK_BLUE_MASK0;
-    }
-    if (state->colormask & PIPE_MASK_A) {
-        blend->color_channel_mask |= RB3D_COLOR_CHANNEL_MASK_ALPHA_MASK0;
+    /* Color channel masks for all MRTs. */
+    blend->color_channel_mask = state->rt[0].colormask;
+    if (r300screen->caps->is_r500 && state->independent_blend_enable) {
+        if (state->rt[1].blend_enable) {
+            blend->color_channel_mask |= (state->rt[1].colormask << 4);
+        }
+        if (state->rt[2].blend_enable) {
+            blend->color_channel_mask |= (state->rt[2].colormask << 8);
+        }
+        if (state->rt[3].blend_enable) {
+            blend->color_channel_mask |= (state->rt[3].colormask << 12);
+        }
     }
 
     if (state->dither) {
@@ -486,19 +486,47 @@ static void
                                const struct pipe_framebuffer_state* state)
 {
     struct r300_context* r300 = r300_context(pipe);
+    uint32_t zbuffer_bpp = 0;
+
+    r300->fb_state.size = (10 * state->nr_cbufs) +
+        (2 * (4 - state->nr_cbufs)) +
+        (state->zsbuf ? 10 : 0) + 6;
+
+    if (state->nr_cbufs > 4) {
+        debug_printf("r300: Implementation error: Too many MRTs in %s, "
+            "refusing to bind framebuffer state!\n", __FUNCTION__);
+        return;
+    }
 
     if (r300->draw) {
         draw_flush(r300->draw);
     }
 
-    r300->framebuffer_state = *state;
+    r300->fb_state.state = state;
 
     /* Don't rely on the order of states being set for the first time. */
-    r300->dirty_state |= R300_NEW_FRAMEBUFFERS;
-
+    /* XXX wait what */
     r300->blend_state.dirty = TRUE;
     r300->dsa_state.dirty = TRUE;
+    r300->fb_state.dirty = TRUE;
     r300->scissor_state.dirty = TRUE;
+
+    /* Polygon offset depends on the zbuffer bit depth. */
+    if (state->zsbuf && r300->polygon_offset_enabled) {
+        switch (util_format_get_blocksize(state->zsbuf->texture->format)) {
+            case 2:
+                zbuffer_bpp = 16;
+                break;
+            case 4:
+                zbuffer_bpp = 24;
+                break;
+        }
+
+        if (r300->zbuffer_bpp != zbuffer_bpp) {
+            r300->zbuffer_bpp = zbuffer_bpp;
+            r300->rs_state.dirty = TRUE;
+        }
+    }
 }
 
 /* Create fragment shader state. */
@@ -534,7 +562,7 @@ static void r300_bind_fs_state(struct pipe_context* pipe, void* shader)
     r300_pick_fragment_shader(r300);
 
     if (r300->vs && r300_vertex_shader_setup_wpos(r300)) {
-        r300->dirty_state |= R300_NEW_VERTEX_FORMAT;
+        r300->vertex_format_state.dirty = TRUE;
     }
 
     r300->dirty_state |= R300_NEW_FRAGMENT_SHADER | R300_NEW_FRAGMENT_SHADER_CONSTANTS;
@@ -602,9 +630,6 @@ static void* r300_create_rs_state(struct pipe_context* pipe,
     rs->line_control = pack_float_16_6x(state->line_width) |
         R300_GA_LINE_CNTL_END_TYPE_COMP;
 
-    /* XXX I think there is something wrong with the polygon mode,
-     * XXX re-test when r300g is in a better shape */
-
     /* Enable polygon mode */
     if (state->fill_cw != PIPE_POLYGON_MODE_FILL ||
         state->fill_ccw != PIPE_POLYGON_MODE_FILL) {
@@ -657,10 +682,8 @@ static void* r300_create_rs_state(struct pipe_context* pipe,
     }
 
     if (rs->polygon_offset_enable) {
-        rs->depth_offset_front = rs->depth_offset_back =
-            fui(state->offset_units);
-        rs->depth_scale_front = rs->depth_scale_back =
-            fui(state->offset_scale);
+        rs->depth_offset = state->offset_units;
+        rs->depth_scale = state->offset_scale;
     }
 
     if (state->line_stipple_enable) {
@@ -694,8 +717,10 @@ static void r300_bind_rs_state(struct pipe_context* pipe, void* state)
 
     if (rs) {
         r300->tcl_bypass = rs->rs.bypass_vs_clip_and_viewport;
+        r300->polygon_offset_enabled = rs->rs.offset_cw || rs->rs.offset_ccw;
     } else {
         r300->tcl_bypass = FALSE;
+        r300->polygon_offset_enabled = FALSE;
     }
 
     r300->rs_state.state = rs;
@@ -705,7 +730,6 @@ static void r300_bind_rs_state(struct pipe_context* pipe, void* state)
     r300->viewport_state.dirty = TRUE;
 
     /* XXX Clean these up when we move to atom emits */
-    r300->dirty_state |= R300_NEW_RS_BLOCK;
     if (r300->fs && r300->fs->inputs.wpos != ATTR_UNUSED) {
         r300->dirty_state |= R300_NEW_FRAGMENT_SHADER_CONSTANTS;
     }
@@ -906,7 +930,23 @@ static void r300_set_vertex_buffers(struct pipe_context* pipe,
         draw_set_vertex_buffers(r300->draw, count, buffers);
     }
 
-    r300->dirty_state |= R300_NEW_VERTEX_FORMAT;
+    r300->vertex_format_state.dirty = TRUE;
+}
+
+static boolean r300_validate_aos(struct r300_context *r300)
+{
+    struct pipe_vertex_buffer *vbuf = r300->vertex_buffer;
+    struct pipe_vertex_element *velem = r300->vertex_element;
+    int i;
+
+    /* Check if formats and strides are aligned to the size of DWORD. */
+    for (i = 0; i < r300->vertex_element_count; i++) {
+        if (vbuf[velem[i].vertex_buffer_index].stride % 4 != 0 ||
+            util_format_get_blocksize(velem[i].src_format) % 4 != 0) {
+            return FALSE;
+        }
+    }
+    return TRUE;
 }
 
 static void r300_set_vertex_elements(struct pipe_context* pipe,
@@ -924,6 +964,12 @@ static void r300_set_vertex_elements(struct pipe_context* pipe,
         draw_flush(r300->draw);
         draw_set_vertex_elements(r300->draw, count, elements);
     }
+
+    if (!r300_validate_aos(r300)) {
+        /* XXX We should fallback using draw. */
+        assert(0);
+        abort();
+    }
 }
 
 static void* r300_create_vs_state(struct pipe_context* pipe,
@@ -964,9 +1010,10 @@ static void r300_bind_vs_state(struct pipe_context* pipe, void* shader)
             r300_vertex_shader_setup_wpos(r300);
         }
 
+        r300->vertex_format_state.dirty = TRUE;
+
         r300->dirty_state |=
-            R300_NEW_VERTEX_SHADER | R300_NEW_VERTEX_SHADER_CONSTANTS |
-            R300_NEW_VERTEX_FORMAT;
+            R300_NEW_VERTEX_SHADER | R300_NEW_VERTEX_SHADER_CONSTANTS;
     } else {
         draw_flush(r300->draw);
         draw_bind_vertex_shader(r300->draw,
diff --git a/src/gallium/drivers/r300/r300_state_derived.c b/src/gallium/drivers/r300/r300_state_derived.c
index 192846411b..bad9e76067 100644
--- a/src/gallium/drivers/r300/r300_state_derived.c
+++ b/src/gallium/drivers/r300/r300_state_derived.c
@@ -37,32 +37,6 @@
 /* r300_state_derived: Various bits of state which are dependent upon
  * currently bound CSO data. */
 
-struct r300_shader_key {
-    struct r300_vertex_shader* vs;
-    struct r300_fragment_shader* fs;
-};
-
-struct r300_shader_derived_value {
-    struct r300_vertex_format* vformat;
-    struct r300_rs_block* rs_block;
-};
-
-unsigned r300_shader_key_hash(void* key) {
-    struct r300_shader_key* shader_key = (struct r300_shader_key*)key;
-    unsigned vs = (intptr_t)shader_key->vs;
-    unsigned fs = (intptr_t)shader_key->fs;
-
-    return (vs << 16) | (fs & 0xffff);
-}
-
-int r300_shader_key_compare(void* key1, void* key2) {
-    struct r300_shader_key* shader_key1 = (struct r300_shader_key*)key1;
-    struct r300_shader_key* shader_key2 = (struct r300_shader_key*)key2;
-
-    return (shader_key1->vs == shader_key2->vs) &&
-        (shader_key1->fs == shader_key2->fs);
-}
-
 static void r300_draw_emit_attrib(struct r300_context* r300,
                                   enum attrib_emit emit,
                                   enum interp_mode interp,
@@ -74,7 +48,9 @@ static void r300_draw_emit_attrib(struct r300_context* r300,
     output = draw_find_shader_output(r300->draw,
                                      info->output_semantic_name[index],
                                      info->output_semantic_index[index]);
-    draw_emit_vertex_attr(&r300->vertex_info->vinfo, emit, interp, output);
+    draw_emit_vertex_attr(
+        (struct vertex_info*)r300->vertex_format_state.state,
+        emit, interp, output);
 }
 
 static void r300_draw_emit_all_attribs(struct r300_context* r300)
@@ -130,7 +106,8 @@ static void r300_draw_emit_all_attribs(struct r300_context* r300)
 /* Update the PSC tables. */
 static void r300_vertex_psc(struct r300_context* r300)
 {
-    struct r300_vertex_info *vformat = r300->vertex_info;
+    struct r300_vertex_info *vformat =
+        (struct r300_vertex_info*)r300->vertex_format_state.state;
     uint16_t type, swizzle;
     enum pipe_format format;
     unsigned i;
@@ -182,7 +159,8 @@ static void r300_vertex_psc(struct r300_context* r300)
 /* Update the PSC tables for SW TCL, using Draw. */
 static void r300_swtcl_vertex_psc(struct r300_context* r300)
 {
-    struct r300_vertex_info *vformat = r300->vertex_info;
+    struct r300_vertex_info *vformat =
+        (struct r300_vertex_info*)r300->vertex_format_state.state;
     struct vertex_info* vinfo = &vformat->vinfo;
     uint16_t type, swizzle;
     enum pipe_format format;
@@ -327,7 +305,7 @@ static void r300_update_rs_block(struct r300_context* r300,
                                  struct r300_shader_semantics* vs_outputs,
                                  struct r300_shader_semantics* fs_inputs)
 {
-    struct r300_rs_block* rs = r300->rs_block;
+    struct r300_rs_block rs = { { 0 } };
     int i, col_count = 0, tex_count = 0, fp_offset = 0;
     void (*rX00_rs_col)(struct r300_rs_block*, int, int, boolean);
     void (*rX00_rs_col_write)(struct r300_rs_block*, int, int);
@@ -350,14 +328,15 @@ static void r300_update_rs_block(struct r300_context* r300,
 
     /* Rasterize colors. */
     for (i = 0; i < ATTR_COLOR_COUNT; i++) {
-        if (vs_outputs->color[i] != ATTR_UNUSED || any_bcolor_used) {
+        if (vs_outputs->color[i] != ATTR_UNUSED || any_bcolor_used ||
+            vs_outputs->color[1] != ATTR_UNUSED) {
             /* Always rasterize if it's written by the VS,
              * otherwise it locks up. */
-            rX00_rs_col(rs, col_count, i, FALSE);
+            rX00_rs_col(&rs, col_count, i, FALSE);
 
             /* Write it to the FS input register if it's used by the FS. */
             if (fs_inputs->color[i] != ATTR_UNUSED) {
-                rX00_rs_col_write(rs, col_count, fp_offset);
+                rX00_rs_col_write(&rs, col_count, fp_offset);
                 fp_offset++;
             }
             col_count++;
@@ -375,11 +354,11 @@ static void r300_update_rs_block(struct r300_context* r300,
         if (vs_outputs->generic[i] != ATTR_UNUSED) {
             /* Always rasterize if it's written by the VS,
              * otherwise it locks up. */
-            rX00_rs_tex(rs, tex_count, tex_count, FALSE);
+            rX00_rs_tex(&rs, tex_count, tex_count, FALSE);
 
             /* Write it to the FS input register if it's used by the FS. */
             if (fs_inputs->generic[i] != ATTR_UNUSED) {
-                rX00_rs_tex_write(rs, tex_count, fp_offset);
+                rX00_rs_tex_write(&rs, tex_count, fp_offset);
                 fp_offset++;
             }
             tex_count++;
@@ -396,11 +375,11 @@ static void r300_update_rs_block(struct r300_context* r300,
     if (vs_outputs->fog != ATTR_UNUSED) {
         /* Always rasterize if it's written by the VS,
          * otherwise it locks up. */
-        rX00_rs_tex(rs, tex_count, tex_count, TRUE);
+        rX00_rs_tex(&rs, tex_count, tex_count, TRUE);
 
         /* Write it to the FS input register if it's used by the FS. */
         if (fs_inputs->fog != ATTR_UNUSED) {
-            rX00_rs_tex_write(rs, tex_count, fp_offset);
+            rX00_rs_tex_write(&rs, tex_count, fp_offset);
             fp_offset++;
         }
         tex_count++;
@@ -415,8 +394,8 @@ static void r300_update_rs_block(struct r300_context* r300,
     /* Rasterize WPOS. */
     /* If the FS doesn't need it, it's not written by the VS. */
     if (fs_inputs->wpos != ATTR_UNUSED) {
-        rX00_rs_tex(rs, tex_count, tex_count, FALSE);
-        rX00_rs_tex_write(rs, tex_count, fp_offset);
+        rX00_rs_tex(&rs, tex_count, tex_count, FALSE);
+        rX00_rs_tex_write(&rs, tex_count, fp_offset);
 
         fp_offset++;
         tex_count++;
@@ -424,51 +403,33 @@ static void r300_update_rs_block(struct r300_context* r300,
 
     /* Rasterize at least one color, or bad things happen. */
     if (col_count == 0 && tex_count == 0) {
-        rX00_rs_col(rs, 0, 0, TRUE);
+        rX00_rs_col(&rs, 0, 0, TRUE);
         col_count++;
     }
 
-    rs->count = (tex_count*4) | (col_count << R300_IC_COUNT_SHIFT) |
+    rs.count = (tex_count*4) | (col_count << R300_IC_COUNT_SHIFT) |
         R300_HIRES_EN;
 
-    rs->inst_count = MAX3(col_count - 1, tex_count - 1, 0);
+    rs.inst_count = MAX3(col_count - 1, tex_count - 1, 0);
+
+    /* Now, after all that, see if we actually need to update the state. */
+    if (memcmp(r300->rs_block_state.state, &rs, sizeof(struct r300_rs_block))) {
+        memcpy(r300->rs_block_state.state, &rs, sizeof(struct r300_rs_block));
+        r300->rs_block_state.dirty = TRUE;
+    }
 }
 
-/* Update the vertex format. */
+/* Update the shader-dependant states. */
 static void r300_update_derived_shader_state(struct r300_context* r300)
 {
     struct r300_screen* r300screen = r300_screen(r300->context.screen);
+    struct r300_vertex_info *vformat =
+        (struct r300_vertex_info*)r300->vertex_format_state.state;
+    struct vertex_info* vinfo = &vformat->vinfo;
 
-    /*
-    struct r300_shader_key* key;
-    struct r300_shader_derived_value* value;
-    key = CALLOC_STRUCT(r300_shader_key);
-    key->vs = r300->vs;
-    key->fs = r300->fs;
-
-    value = (struct r300_shader_derived_value*)
-        util_hash_table_get(r300->shader_hash_table, (void*)key);
-    if (value) {
-        //vformat = value->vformat;
-        rs_block = value->rs_block;
-
-        FREE(key);
-    } else {
-        rs_block = CALLOC_STRUCT(r300_rs_block);
-        value = CALLOC_STRUCT(r300_shader_derived_value);
-
-        r300_update_rs_block(r300, rs_block);
-
-        //value->vformat = vformat;
-        value->rs_block = rs_block;
-        util_hash_table_set(r300->shader_hash_table,
-            (void*)key, (void*)value);
-    } */
-
-    /* Reset structures */
-    memset(r300->rs_block, 0, sizeof(struct r300_rs_block));
-    memset(r300->vertex_info, 0, sizeof(struct r300_vertex_info));
-    memcpy(r300->vertex_info->vinfo.hwfmt, r300->vs->hwfmt, sizeof(uint)*4);
+    /* Mmm, delicious hax */
+    memset(r300->vertex_format_state.state, 0, sizeof(struct r300_vertex_info));
+    memcpy(vinfo->hwfmt, r300->vs->hwfmt, sizeof(uint)*4);
 
     r300_update_rs_block(r300, &r300->vs->outputs, &r300->fs->inputs);
 
@@ -476,11 +437,10 @@ static void r300_update_derived_shader_state(struct r300_context* r300)
         r300_vertex_psc(r300);
     } else {
         r300_draw_emit_all_attribs(r300);
-        draw_compute_vertex_size(&r300->vertex_info->vinfo);
+        draw_compute_vertex_size(
+            (struct vertex_info*)r300->vertex_format_state.state);
         r300_swtcl_vertex_psc(r300);
     }
-
-    r300->dirty_state |= R300_NEW_RS_BLOCK;
 }
 
 static boolean r300_dsa_writes_depth_stencil(struct r300_dsa_state* dsa)
@@ -558,8 +518,8 @@ void r300_update_derived_state(struct r300_context* r300)
 {
     /* XXX */
     if (r300->dirty_state &
-        (R300_NEW_FRAGMENT_SHADER | R300_NEW_VERTEX_SHADER |
-         R300_NEW_VERTEX_FORMAT) || r300->rs_state.dirty) {
+        (R300_NEW_FRAGMENT_SHADER | R300_NEW_VERTEX_SHADER) ||
+        r300->vertex_format_state.dirty || r300->rs_state.dirty) {
         r300_update_derived_shader_state(r300);
     }
 
diff --git a/src/gallium/drivers/r300/r300_state_inlines.h b/src/gallium/drivers/r300/r300_state_inlines.h
index 35be00e1b0..5df6815221 100644
--- a/src/gallium/drivers/r300/r300_state_inlines.h
+++ b/src/gallium/drivers/r300/r300_state_inlines.h
@@ -81,9 +81,6 @@ static INLINE uint32_t r300_translate_blend_factor(int blend_fact)
             return R300_BLEND_GL_CONST_COLOR;
         case PIPE_BLENDFACTOR_CONST_ALPHA:
             return R300_BLEND_GL_CONST_ALPHA;
-        /* XXX WTF are these?
-        case PIPE_BLENDFACTOR_SRC1_COLOR:
-        case PIPE_BLENDFACTOR_SRC1_ALPHA: */
         case PIPE_BLENDFACTOR_ZERO:
             return R300_BLEND_GL_ZERO;
         case PIPE_BLENDFACTOR_INV_SRC_COLOR:
@@ -98,9 +95,16 @@ static INLINE uint32_t r300_translate_blend_factor(int blend_fact)
             return R300_BLEND_GL_ONE_MINUS_CONST_COLOR;
         case PIPE_BLENDFACTOR_INV_CONST_ALPHA:
             return R300_BLEND_GL_ONE_MINUS_CONST_ALPHA;
-        /* XXX see above
+
+        case PIPE_BLENDFACTOR_SRC1_COLOR:
+        case PIPE_BLENDFACTOR_SRC1_ALPHA:
         case PIPE_BLENDFACTOR_INV_SRC1_COLOR:
-        case PIPE_BLENDFACTOR_INV_SRC1_ALPHA: */
+        case PIPE_BLENDFACTOR_INV_SRC1_ALPHA:
+            debug_printf("r300: Implementation error: "
+                "Bad blend factor %d not supported!\n", blend_fact);
+            assert(0);
+            break;
+
         default:
             debug_printf("r300: Unknown blend factor %d\n", blend_fact);
             assert(0);
@@ -331,7 +335,10 @@ static INLINE uint32_t r300_translate_colorformat(enum pipe_format format)
 {
     switch (format) {
         /* 8-bit buffers */
+        case PIPE_FORMAT_A8_UNORM:
         case PIPE_FORMAT_I8_UNORM:
+        case PIPE_FORMAT_L8_UNORM:
+        /* case PIPE_FORMAT_S8_UNORM: ??? */
             return R300_COLOR_FORMAT_I8;
         /* 16-bit buffers */
         case PIPE_FORMAT_R5G6B5_UNORM:
@@ -408,6 +415,16 @@ static INLINE uint32_t r300_translate_out_fmt(enum pipe_format format)
             return R300_US_OUT_FMT_C4_8 |
                 R300_C0_SEL_A | R300_C1_SEL_B |
                 R300_C2_SEL_G | R300_C3_SEL_R;
+
+        /* 8-bit outputs */
+        case PIPE_FORMAT_A8_UNORM:
+            return R300_US_OUT_FMT_C4_8 |
+                R300_C0_SEL_A;
+        case PIPE_FORMAT_I8_UNORM:
+        case PIPE_FORMAT_L8_UNORM:
+            return R300_US_OUT_FMT_C4_8 |
+                R300_C0_SEL_R;
+ /* R300_OUT_SIGN(x) */
         default:
             debug_printf("r300: Implementation error: "
                 "Got unsupported output format %s in %s\n",
@@ -537,6 +554,7 @@ r300_translate_vertex_data_type(enum pipe_format format) {
 static INLINE uint16_t
 r300_translate_vertex_data_swizzle(enum pipe_format format) {
     const struct util_format_description *desc = util_format_description(format);
+    unsigned swizzle[4], i;
 
     assert(format);
 
@@ -547,11 +565,26 @@ r300_translate_vertex_data_swizzle(enum pipe_format format) {
         return 0;
     }
 
-    return ((desc->swizzle[0] << R300_SWIZZLE_SELECT_X_SHIFT) |
-        (desc->swizzle[1] << R300_SWIZZLE_SELECT_Y_SHIFT) |
-        (desc->swizzle[2] << R300_SWIZZLE_SELECT_Z_SHIFT) |
-        (desc->swizzle[3] << R300_SWIZZLE_SELECT_W_SHIFT) |
-        (0xf << R300_WRITE_ENA_SHIFT));
+    /* Swizzles for 8bits formats are in the reversed order, not sure why. */
+    if (desc->channel[0].size == 8) {
+        for (i = 0; i < 4; i++) {
+            if (desc->swizzle[i] <= 3) {
+                swizzle[i] = 3 - desc->swizzle[i];
+            } else {
+                swizzle[i] = desc->swizzle[i];
+            }
+        }
+    } else {
+        for (i = 0; i < 4; i++) {
+            swizzle[i] = desc->swizzle[i];
+        }
+    }
+
+    return ((swizzle[0] << R300_SWIZZLE_SELECT_X_SHIFT) |
+            (swizzle[1] << R300_SWIZZLE_SELECT_Y_SHIFT) |
+            (swizzle[2] << R300_SWIZZLE_SELECT_Z_SHIFT) |
+            (swizzle[3] << R300_SWIZZLE_SELECT_W_SHIFT) |
+            (0xf << R300_WRITE_ENA_SHIFT));
 }
 
 #endif /* R300_STATE_INLINES_H */
diff --git a/src/gallium/drivers/r300/r300_state_invariant.c b/src/gallium/drivers/r300/r300_state_invariant.c
index b0f309695c..97927acf1b 100644
--- a/src/gallium/drivers/r300/r300_state_invariant.c
+++ b/src/gallium/drivers/r300/r300_state_invariant.c
@@ -38,12 +38,12 @@ struct pipe_viewport_state r300_viewport_identity = {
  *
  * Note that eventually this should be empty, but it's useful for development
  * and general unduplication of code. */
-void r300_emit_invariant_state(struct r300_context* r300)
+void r300_emit_invariant_state(struct r300_context* r300, void* state)
 {
     struct r300_capabilities* caps = r300_screen(r300->context.screen)->caps;
     CS_LOCALS(r300);
 
-    BEGIN_CS(16 + (caps->has_tcl ? 2: 0));
+    BEGIN_CS(14 + (caps->has_tcl ? 2: 0));
 
     /*** Graphics Backend (GB) ***/
     /* Various GB enables */
@@ -58,8 +58,6 @@ void r300_emit_invariant_state(struct r300_context* r300)
      */
     /* Source of fog depth */
     OUT_CS_REG(R300_GB_SELECT, R300_GB_FOG_SELECT_1_1_W);
-    /* AA enable */
-    OUT_CS_REG(R300_GB_AA_CONFIG, 0x0);
 
     /*** Fog (FG) ***/
     OUT_CS_REG(R300_FG_FOG_BLEND, 0x0);
@@ -79,7 +77,8 @@ void r300_emit_invariant_state(struct r300_context* r300)
     END_CS;
 
     /* XXX unsorted stuff from surface_fill */
-    BEGIN_CS(44 + (caps->has_tcl ? 7 : 0) + (caps->is_r500 ? 4 : 0));
+    BEGIN_CS(44 + (caps->has_tcl ? 7 : 0) +
+             (caps->family >= CHIP_FAMILY_RV350 ? 4 : 0));
 
     if (caps->has_tcl) {
         /*Flushing PVS is required before the VAP_GB registers can be changed*/
diff --git a/src/gallium/drivers/r300/r300_state_invariant.h b/src/gallium/drivers/r300/r300_state_invariant.h
index 05cff0d6df..5d1a963654 100644
--- a/src/gallium/drivers/r300/r300_state_invariant.h
+++ b/src/gallium/drivers/r300/r300_state_invariant.h
@@ -25,6 +25,6 @@
 
 struct r300_context;
 
-void r300_emit_invariant_state(struct r300_context* r300);
+void r300_emit_invariant_state(struct r300_context* r300, void* state);
 
 #endif /* R300_STATE_INVARIANT_H */
diff --git a/src/gallium/drivers/r300/r300_texture.c b/src/gallium/drivers/r300/r300_texture.c
index a9bbdd56d8..67bf8ce13f 100644
--- a/src/gallium/drivers/r300/r300_texture.c
+++ b/src/gallium/drivers/r300/r300_texture.c
@@ -30,6 +30,8 @@
 #include "r300_texture.h"
 #include "r300_screen.h"
 
+#include "radeon_winsys.h"
+
 #define TILE_WIDTH 0
 #define TILE_HEIGHT 1
 
@@ -42,10 +44,11 @@ static const unsigned microblock_table[5][3][2] = {
     {{ 2, 1}, {0, 0}, {0, 0}}  /* 128 bits per pixel */
 };
 
-static void r300_setup_texture_state(struct r300_texture* tex, boolean is_r500)
+static void r300_setup_texture_state(struct r300_screen* screen, struct r300_texture* tex)
 {
     struct r300_texture_state* state = &tex->state;
     struct pipe_texture *pt = &tex->tex;
+    boolean is_r500 = screen->caps->is_r500;
 
     state->format0 = R300_TX_WIDTH((pt->width0 - 1) & 0x7ff) |
                      R300_TX_HEIGHT((pt->height0 - 1) & 0x7ff);
@@ -79,8 +82,8 @@ static void r300_setup_texture_state(struct r300_texture* tex, boolean is_r500)
     }
     assert(is_r500 || (pt->width0 <= 2048 && pt->height0 <= 2048));
 
-    debug_printf("r300: Set texture state (%dx%d, %d levels)\n",
-		 pt->width0, pt->height0, pt->last_level);
+    SCREEN_DBG(screen, DBG_TEX, "r300: Set texture state (%dx%d, %d levels)\n",
+               pt->width0, pt->height0, pt->last_level);
 }
 
 unsigned r300_texture_get_offset(struct r300_texture* tex, unsigned level,
@@ -123,7 +126,8 @@ static unsigned r300_texture_get_tile_size(struct r300_texture* tex, int dim)
  * Return the stride, in bytes, of the texture images of the given texture
  * at the given level.
  */
-unsigned r300_texture_get_stride(struct r300_texture* tex, unsigned level)
+unsigned r300_texture_get_stride(struct r300_screen* screen,
+                                 struct r300_texture* tex, unsigned level)
 {
     unsigned tile_width, width;
 
@@ -132,16 +136,20 @@ unsigned r300_texture_get_stride(struct r300_texture* tex, unsigned level)
 
     /* Check the level. */
     if (level > tex->tex.last_level) {
-        debug_printf("%s: level (%u) > last_level (%u)\n", __FUNCTION__,
-            level, tex->tex.last_level);
+        SCREEN_DBG(screen, DBG_TEX, "%s: level (%u) > last_level (%u)\n",
+                   __FUNCTION__, level, tex->tex.last_level);
         return 0;
     }
 
-    tile_width = r300_texture_get_tile_size(tex, TILE_WIDTH);
-    width = align(u_minify(tex->tex.width0, level), tile_width);
+    width = u_minify(tex->tex.width0, level);
 
-    /* Should already be aligned except for S3TC. */
-    return align(util_format_get_stride(tex->tex.format, width), 32);
+    if (!util_format_is_compressed(tex->tex.format)) {
+        tile_width = r300_texture_get_tile_size(tex, TILE_WIDTH);
+        width = align(width, tile_width);
+        return util_format_get_stride(tex->tex.format, width);
+    } else {
+        return align(util_format_get_stride(tex->tex.format, width), 32);
+    }
 }
 
 static unsigned r300_texture_get_nblocksy(struct r300_texture* tex,
@@ -149,21 +157,27 @@ static unsigned r300_texture_get_nblocksy(struct r300_texture* tex,
 {
     unsigned height, tile_height;
 
-    tile_height = r300_texture_get_tile_size(tex, TILE_HEIGHT);
-    height = align(u_minify(tex->tex.height0, level), tile_height);
+    height = u_minify(tex->tex.height0, level);
+
+    if (!util_format_is_compressed(tex->tex.format)) {
+        tile_height = r300_texture_get_tile_size(tex, TILE_HEIGHT);
+        height = align(height, tile_height);
+    }
 
     return util_format_get_nblocksy(tex->tex.format, height);
 }
 
-static void r300_setup_miptree(struct r300_texture* tex)
+static void r300_setup_miptree(struct r300_screen* screen,
+                               struct r300_texture* tex)
 {
     struct pipe_texture* base = &tex->tex;
     unsigned stride, size, layer_size, nblocksy, i;
 
-    debug_printf("r300: Making miptree for texture, format %s\n", pf_name(base->format));
+    SCREEN_DBG(screen, DBG_TEX, "r300: Making miptree for texture, format %s\n",
+               pf_name(base->format));
 
     for (i = 0; i <= base->last_level; i++) {
-        stride = r300_texture_get_stride(tex, i);
+        stride = r300_texture_get_stride(screen, tex, i);
         nblocksy = r300_texture_get_nblocksy(tex, i);
         layer_size = stride * nblocksy;
 
@@ -177,7 +191,7 @@ static void r300_setup_miptree(struct r300_texture* tex)
         tex->layer_size[i] = layer_size;
         tex->pitch[i] = stride / util_format_get_blocksize(base->format);
 
-        debug_printf("r300: Texture miptree: Level %d "
+        SCREEN_DBG(screen, DBG_TEX, "r300: Texture miptree: Level %d "
                 "(%dx%dx%d px, pitch %d bytes) %d bytes total\n",
                 i, u_minify(base->width0, i), u_minify(base->height0, i),
                 u_minify(base->depth0, i), stride, tex->size);
@@ -196,6 +210,8 @@ static struct pipe_texture*
                         const struct pipe_texture* template)
 {
     struct r300_texture* tex = CALLOC_STRUCT(r300_texture);
+    struct r300_screen* rscreen = r300_screen(screen);
+    struct radeon_winsys* winsys = (struct radeon_winsys*)screen->winsys;
 
     if (!tex) {
         return NULL;
@@ -206,12 +222,16 @@ static struct pipe_texture*
     tex->tex.screen = screen;
 
     r300_setup_flags(tex);
-    r300_setup_miptree(tex);
-    r300_setup_texture_state(tex, r300_screen(screen)->caps->is_r500);
+    r300_setup_miptree(rscreen, tex);
+    r300_setup_texture_state(rscreen, tex);
 
     tex->buffer = screen->buffer_create(screen, 2048,
                                         PIPE_BUFFER_USAGE_PIXEL,
                                         tex->size);
+    winsys->buffer_set_tiling(winsys, tex->buffer,
+                              tex->pitch[0],
+                              tex->microtile != R300_BUFFER_LINEAR,
+                              tex->macrotile != R300_BUFFER_LINEAR);
 
     if (!tex->buffer) {
         FREE(tex);
@@ -273,6 +293,7 @@ static struct pipe_texture*
                          struct pipe_buffer* buffer)
 {
     struct r300_texture* tex;
+    struct r300_screen* rscreen = r300_screen(screen);
 
     /* Support only 2D textures without mipmaps */
     if (base->target != PIPE_TEXTURE_2D ||
@@ -294,7 +315,7 @@ static struct pipe_texture*
     tex->pitch[0] = *stride / util_format_get_blocksize(base->format);
 
     r300_setup_flags(tex);
-    r300_setup_texture_state(tex, r300_screen(screen)->caps->is_r500);
+    r300_setup_texture_state(rscreen, tex);
 
     pipe_buffer_reference(&tex->buffer, buffer);
 
@@ -361,7 +382,8 @@ void r300_init_screen_texture_functions(struct pipe_screen* screen)
     screen->video_surface_destroy= r300_video_surface_destroy;
 }
 
-boolean r300_get_texture_buffer(struct pipe_texture* texture,
+boolean r300_get_texture_buffer(struct pipe_screen* screen,
+                                struct pipe_texture* texture,
                                 struct pipe_buffer** buffer,
                                 unsigned* stride)
 {
@@ -373,7 +395,7 @@ boolean r300_get_texture_buffer(struct pipe_texture* texture,
     pipe_buffer_reference(buffer, tex->buffer);
 
     if (stride) {
-        *stride = r300_texture_get_stride(tex, 0);
+        *stride = r300_texture_get_stride(r300_screen(screen), tex, 0);
     }
 
     return TRUE;
diff --git a/src/gallium/drivers/r300/r300_texture.h b/src/gallium/drivers/r300/r300_texture.h
index 55ceb1a513..453fb1accc 100644
--- a/src/gallium/drivers/r300/r300_texture.h
+++ b/src/gallium/drivers/r300/r300_texture.h
@@ -31,7 +31,8 @@ struct r300_texture;
 
 void r300_init_screen_texture_functions(struct pipe_screen* screen);
 
-unsigned r300_texture_get_stride(struct r300_texture* tex, unsigned level);
+unsigned r300_texture_get_stride(struct r300_screen* screen,
+                                 struct r300_texture* tex, unsigned level);
 
 unsigned r300_texture_get_offset(struct r300_texture* tex, unsigned level,
                                  unsigned zslice, unsigned face);
@@ -41,6 +42,7 @@ static INLINE uint32_t r300_translate_texformat(enum pipe_format format)
 {
     switch (format) {
         /* X8 */
+        case PIPE_FORMAT_A8_UNORM:
         case PIPE_FORMAT_I8_UNORM:
             return R300_EASY_TX_FORMAT(X, X, X, X, X8);
         case PIPE_FORMAT_L8_UNORM:
@@ -115,7 +117,8 @@ r300_video_surface(struct pipe_video_surface *pvs)
 
 #ifndef R300_WINSYS_H
 
-boolean r300_get_texture_buffer(struct pipe_texture* texture,
+boolean r300_get_texture_buffer(struct pipe_screen* screen,
+                                struct pipe_texture* texture,
                                 struct pipe_buffer** buffer,
                                 unsigned* stride);
 
diff --git a/src/gallium/drivers/r300/r300_vs.c b/src/gallium/drivers/r300/r300_vs.c
index 68aef70872..9fbb830047 100644
--- a/src/gallium/drivers/r300/r300_vs.c
+++ b/src/gallium/drivers/r300/r300_vs.c
@@ -124,7 +124,8 @@ static void r300_shader_vap_output_fmt(struct r300_vertex_shader* vs)
 
     /* Colors. */
     for (i = 0; i < ATTR_COLOR_COUNT; i++) {
-        if (vs_outputs->color[i] != ATTR_UNUSED || any_bcolor_used) {
+        if (vs_outputs->color[i] != ATTR_UNUSED || any_bcolor_used ||
+            vs_outputs->color[1] != ATTR_UNUSED) {
             hwfmt[1] |= R300_INPUT_CNTL_COLOR;
             hwfmt[2] |= R300_VAP_OUTPUT_VTX_FMT_0__COLOR_0_PRESENT << i;
         }
@@ -182,7 +183,8 @@ static void r300_stream_locations_notcl(
 
     /* Colors. */
     for (i = 0; i < ATTR_COLOR_COUNT; i++) {
-        if (vs_outputs->color[i] != ATTR_UNUSED || any_bcolor_used) {
+        if (vs_outputs->color[i] != ATTR_UNUSED || any_bcolor_used ||
+            vs_outputs->color[1] != ATTR_UNUSED) {
             stream_loc[tabi++] = 2 + i;
         }
     }
@@ -259,7 +261,8 @@ static void set_vertex_inputs_outputs(struct r300_vertex_program_compiler * c)
     for (i = 0; i < ATTR_COLOR_COUNT; i++) {
         if (outputs->color[i] != ATTR_UNUSED) {
             c->code->outputs[outputs->color[i]] = reg++;
-        } else if (any_bcolor_used) {
+        } else if (any_bcolor_used ||
+                   outputs->color[1] != ATTR_UNUSED) {
             reg++;
         }
     }
diff --git a/src/gallium/drivers/r300/r300_winsys.h b/src/gallium/drivers/r300/r300_winsys.h
index 1ae6de70fe..bdb8b54bab 100644
--- a/src/gallium/drivers/r300/r300_winsys.h
+++ b/src/gallium/drivers/r300/r300_winsys.h
@@ -40,7 +40,8 @@ extern "C" {
 struct pipe_context* r300_create_context(struct pipe_screen* screen,
                                          struct radeon_winsys* radeon_winsys);
 
-boolean r300_get_texture_buffer(struct pipe_texture* texture,
+boolean r300_get_texture_buffer(struct pipe_screen* screen,
+                                struct pipe_texture* texture,
                                 struct pipe_buffer** buffer,
                                 unsigned* stride);
 
diff --git a/src/gallium/drivers/softpipe/Makefile b/src/gallium/drivers/softpipe/Makefile
index bcb887a0b2..e4ac49fa85 100644
--- a/src/gallium/drivers/softpipe/Makefile
+++ b/src/gallium/drivers/softpipe/Makefile
@@ -32,6 +32,7 @@ C_SOURCES = \
 	sp_tex_tile_cache.c \
 	sp_tile_cache.c \
 	sp_surface.c \
-	sp_video_context.c
+	sp_video_context.c \
+	sp_winsys.c
 
 include ../../Makefile.template
diff --git a/src/gallium/drivers/softpipe/SConscript b/src/gallium/drivers/softpipe/SConscript
index aac9edf44e..3042e556c6 100644
--- a/src/gallium/drivers/softpipe/SConscript
+++ b/src/gallium/drivers/softpipe/SConscript
@@ -34,6 +34,7 @@ softpipe = env.ConvenienceLibrary(
 		'sp_texture.c',
 		'sp_tile_cache.c',
 		'sp_video_context.c',
+		'sp_winsys.c'
 	])
 
 Export('softpipe')
diff --git a/src/gallium/drivers/softpipe/sp_context.c b/src/gallium/drivers/softpipe/sp_context.c
index 8e01793940..53133d2888 100644
--- a/src/gallium/drivers/softpipe/sp_context.c
+++ b/src/gallium/drivers/softpipe/sp_context.c
@@ -43,7 +43,6 @@
 #include "sp_surface.h"
 #include "sp_tile_cache.h"
 #include "sp_tex_tile_cache.h"
-#include "sp_texture.h"
 #include "sp_winsys.h"
 #include "sp_query.h"
 
@@ -112,9 +111,13 @@ softpipe_destroy( struct pipe_context *pipe )
       pipe_texture_reference(&softpipe->vertex_textures[i], NULL);
    }
 
-   for (i = 0; i < Elements(softpipe->constants); i++) {
-      if (softpipe->constants[i]) {
-         pipe_buffer_reference(&softpipe->constants[i], NULL);
+   for (i = 0; i < PIPE_SHADER_TYPES; i++) {
+      uint j;
+
+      for (j = 0; j < PIPE_MAX_CONSTANT_BUFFERS; j++) {
+         if (softpipe->constants[i][j]) {
+            pipe_buffer_reference(&softpipe->constants[i][j], NULL);
+         }
       }
    }
 
diff --git a/src/gallium/drivers/softpipe/sp_context.h b/src/gallium/drivers/softpipe/sp_context.h
index da673c57ad..be4613b622 100644
--- a/src/gallium/drivers/softpipe/sp_context.h
+++ b/src/gallium/drivers/softpipe/sp_context.h
@@ -63,7 +63,7 @@ struct softpipe_context {
    /** Other rendering state */
    struct pipe_blend_color blend_color;
    struct pipe_clip_state clip;
-   struct pipe_buffer *constants[PIPE_SHADER_TYPES];
+   struct pipe_buffer *constants[PIPE_SHADER_TYPES][PIPE_MAX_CONSTANT_BUFFERS];
    struct pipe_framebuffer_state framebuffer;
    struct pipe_poly_stipple poly_stipple;
    struct pipe_scissor_state scissor;
@@ -92,7 +92,7 @@ struct softpipe_context {
    ubyte *mapped_vbuffer[PIPE_MAX_ATTRIBS];
 
    /** Mapped constant buffers */
-   void *mapped_constants[PIPE_SHADER_TYPES];
+   void *mapped_constants[PIPE_SHADER_TYPES][PIPE_MAX_CONSTANT_BUFFERS];
 
    /** Vertex format */
    struct vertex_info vertex_info;
diff --git a/src/gallium/drivers/softpipe/sp_draw_arrays.c b/src/gallium/drivers/softpipe/sp_draw_arrays.c
index 03b58d2fb7..2b82427d54 100644
--- a/src/gallium/drivers/softpipe/sp_draw_arrays.c
+++ b/src/gallium/drivers/softpipe/sp_draw_arrays.c
@@ -49,30 +49,36 @@ static void
 softpipe_map_constant_buffers(struct softpipe_context *sp)
 {
    struct pipe_winsys *ws = sp->pipe.winsys;
-   uint i, vssize, gssize;
+   uint i;
 
    for (i = 0; i < PIPE_SHADER_TYPES; i++) {
-      if (sp->constants[i] && sp->constants[i]->size)
-         sp->mapped_constants[i] = ws->buffer_map(ws, sp->constants[i],
-                                                  PIPE_BUFFER_USAGE_CPU_READ);
+      uint j;
+
+      for (j = 0; j < PIPE_MAX_CONSTANT_BUFFERS; j++) {
+         if (sp->constants[i][j] && sp->constants[i][j]->size) {
+            sp->mapped_constants[i][j] = ws->buffer_map(ws,
+                                                        sp->constants[i][j],
+                                                        PIPE_BUFFER_USAGE_CPU_READ);
+         }
+      }
    }
 
-   if (sp->constants[PIPE_SHADER_VERTEX])
-      vssize = sp->constants[PIPE_SHADER_VERTEX]->size;
-   else
-      vssize = 0;
-
-   if (sp->constants[PIPE_SHADER_GEOMETRY])
-      gssize = sp->constants[PIPE_SHADER_GEOMETRY]->size;
-   else
-      gssize = 0;
-
-   draw_set_mapped_constant_buffer(sp->draw, PIPE_SHADER_VERTEX,
-                                   sp->mapped_constants[PIPE_SHADER_VERTEX],
-                                   vssize);
-   draw_set_mapped_constant_buffer(sp->draw, PIPE_SHADER_GEOMETRY,
-                                   sp->mapped_constants[PIPE_SHADER_GEOMETRY],
-                                   gssize);
+   for (i = 0; i < PIPE_MAX_CONSTANT_BUFFERS; i++) {
+      if (sp->constants[PIPE_SHADER_VERTEX][i]) {
+         draw_set_mapped_constant_buffer(sp->draw,
+                                         PIPE_SHADER_VERTEX,
+                                         i,
+                                         sp->mapped_constants[PIPE_SHADER_VERTEX][i],
+                                         sp->constants[PIPE_SHADER_VERTEX][i]->size);
+      }
+      if (sp->constants[PIPE_SHADER_GEOMETRY][i]) {
+         draw_set_mapped_constant_buffer(sp->draw,
+                                         PIPE_SHADER_GEOMETRY,
+                                         i,
+                                         sp->mapped_constants[PIPE_SHADER_GEOMETRY][i],
+                                         sp->constants[PIPE_SHADER_GEOMETRY][i]->size);
+      }
+   }
 }
 
 
@@ -87,13 +93,28 @@ softpipe_unmap_constant_buffers(struct softpipe_context *sp)
     */
    draw_flush(sp->draw);
 
-   draw_set_mapped_constant_buffer(sp->draw, PIPE_SHADER_VERTEX, NULL, 0);
-   draw_set_mapped_constant_buffer(sp->draw, PIPE_SHADER_GEOMETRY, NULL, 0);
+   for (i = 0; i < PIPE_MAX_CONSTANT_BUFFERS; i++) {
+      draw_set_mapped_constant_buffer(sp->draw,
+                                      PIPE_SHADER_VERTEX,
+                                      i,
+                                      NULL,
+                                      0);
+      draw_set_mapped_constant_buffer(sp->draw,
+                                      PIPE_SHADER_GEOMETRY,
+                                      i,
+                                      NULL,
+                                      0);
+   }
 
    for (i = 0; i < PIPE_SHADER_TYPES; i++) {
-      if (sp->constants[i] && sp->constants[i]->size)
-         ws->buffer_unmap(ws, sp->constants[i]);
-      sp->mapped_constants[i] = NULL;
+      uint j;
+
+      for (j = 0; j < PIPE_MAX_CONSTANT_BUFFERS; j++) {
+         if (sp->constants[i][j] && sp->constants[i][j]->size) {
+            ws->buffer_unmap(ws, sp->constants[i][j]);
+         }
+         sp->mapped_constants[i][j] = NULL;
+      }
    }
 }
 
diff --git a/src/gallium/drivers/softpipe/sp_flush.c b/src/gallium/drivers/softpipe/sp_flush.c
index 75dac810a1..e8952bf4fb 100644
--- a/src/gallium/drivers/softpipe/sp_flush.c
+++ b/src/gallium/drivers/softpipe/sp_flush.c
@@ -34,11 +34,9 @@
 #include "draw/draw_context.h"
 #include "sp_flush.h"
 #include "sp_context.h"
-#include "sp_surface.h"
 #include "sp_state.h"
 #include "sp_tile_cache.h"
 #include "sp_tex_tile_cache.h"
-#include "sp_winsys.h"
 
 
 void
diff --git a/src/gallium/drivers/softpipe/sp_fs_sse.c b/src/gallium/drivers/softpipe/sp_fs_sse.c
index f912950658..acee213670 100644
--- a/src/gallium/drivers/softpipe/sp_fs_sse.c
+++ b/src/gallium/drivers/softpipe/sp_fs_sse.c
@@ -135,7 +135,7 @@ fs_sse_run( const struct sp_fragment_shader *base,
    tgsi_set_exec_mask(machine, 1, 1, 1, 1);
 
    shader->func( machine,
-		 machine->Consts,
+                 (const float (*)[4])machine->Consts[0],
                  (const float (*)[4])shader->immediates,
 		 machine->InterpCoefs
 		 /*, &machine->QuadPos*/
diff --git a/src/gallium/drivers/softpipe/sp_prim_vbuf.c b/src/gallium/drivers/softpipe/sp_prim_vbuf.c
index 7f573aef3c..98c08eaffa 100644
--- a/src/gallium/drivers/softpipe/sp_prim_vbuf.c
+++ b/src/gallium/drivers/softpipe/sp_prim_vbuf.c
@@ -526,6 +526,8 @@ static void
 sp_vbuf_destroy(struct vbuf_render *vbr)
 {
    struct softpipe_vbuf_render *cvbr = softpipe_vbuf_render(vbr);
+   if(cvbr->vertex_buffer)
+      align_free(cvbr->vertex_buffer);
    sp_setup_destroy_context(cvbr->setup);
    FREE(cvbr);
 }
@@ -541,7 +543,6 @@ sp_create_vbuf_backend(struct softpipe_context *sp)
 
    assert(sp->draw);
 
-
    cvbr->base.max_indices = SP_MAX_VBUF_INDEXES;
    cvbr->base.max_vertex_buffer_bytes = SP_MAX_VBUF_SIZE;
 
diff --git a/src/gallium/drivers/softpipe/sp_quad_blend.c b/src/gallium/drivers/softpipe/sp_quad_blend.c
index d9babe81da..d65307b7f6 100644
--- a/src/gallium/drivers/softpipe/sp_quad_blend.c
+++ b/src/gallium/drivers/softpipe/sp_quad_blend.c
@@ -35,7 +35,6 @@
 #include "util/u_memory.h"
 #include "sp_context.h"
 #include "sp_quad.h"
-#include "sp_surface.h"
 #include "sp_tile_cache.h"
 #include "sp_quad_pipe.h"
 
@@ -224,7 +223,8 @@ logicop_quad(struct quad_stage *qs,
 static void
 blend_quad(struct quad_stage *qs, 
            float (*quadColor)[4],
-           float (*dest)[4])
+           float (*dest)[4],
+           unsigned cbuf)
 {
    static const float zero[4] = { 0, 0, 0, 0 };
    static const float one[4] = { 1, 1, 1, 1 };
@@ -234,7 +234,7 @@ blend_quad(struct quad_stage *qs,
    /*
     * Compute src/first term RGB
     */
-   switch (softpipe->blend->rgb_src_factor) {
+   switch (softpipe->blend->rt[cbuf].rgb_src_factor) {
    case PIPE_BLENDFACTOR_ONE:
       VEC4_COPY(source[0], quadColor[0]); /* R */
       VEC4_COPY(source[1], quadColor[1]); /* G */
@@ -384,7 +384,7 @@ blend_quad(struct quad_stage *qs,
    /*
     * Compute src/first term A
     */
-   switch (softpipe->blend->alpha_src_factor) {
+   switch (softpipe->blend->rt[cbuf].alpha_src_factor) {
    case PIPE_BLENDFACTOR_ONE:
       VEC4_COPY(source[3], quadColor[3]); /* A */
       break;
@@ -453,7 +453,7 @@ blend_quad(struct quad_stage *qs,
    /*
     * Compute dest/second term RGB
     */
-   switch (softpipe->blend->rgb_dst_factor) {
+   switch (softpipe->blend->rt[cbuf].rgb_dst_factor) {
    case PIPE_BLENDFACTOR_ONE:
       /* dest = dest * 1   NO-OP, leave dest as-is */
       break;
@@ -593,7 +593,7 @@ blend_quad(struct quad_stage *qs,
    /*
     * Compute dest/second term A
     */
-   switch (softpipe->blend->alpha_dst_factor) {
+   switch (softpipe->blend->rt[cbuf].alpha_dst_factor) {
    case PIPE_BLENDFACTOR_ONE:
       /* dest = dest * 1   NO-OP, leave dest as-is */
       break;
@@ -656,7 +656,7 @@ blend_quad(struct quad_stage *qs,
    /*
     * Combine RGB terms
     */
-   switch (softpipe->blend->rgb_func) {
+   switch (softpipe->blend->rt[cbuf].rgb_func) {
    case PIPE_BLEND_ADD:
       VEC4_ADD_SAT(quadColor[0], source[0], dest[0]); /* R */
       VEC4_ADD_SAT(quadColor[1], source[1], dest[1]); /* G */
@@ -689,7 +689,7 @@ blend_quad(struct quad_stage *qs,
    /*
     * Combine A terms
     */
-   switch (softpipe->blend->alpha_func) {
+   switch (softpipe->blend->rt[cbuf].alpha_func) {
    case PIPE_BLEND_ADD:
       VEC4_ADD_SAT(quadColor[3], source[3], dest[3]); /* A */
       break;
@@ -711,26 +711,24 @@ blend_quad(struct quad_stage *qs,
 }
 
 static void
-colormask_quad(struct quad_stage *qs,
+colormask_quad(unsigned colormask,
                float (*quadColor)[4],
                float (*dest)[4])
 {
-   struct softpipe_context *softpipe = qs->softpipe;
-
    /* R */
-   if (!(softpipe->blend->colormask & PIPE_MASK_R))
+   if (!(colormask & PIPE_MASK_R))
       COPY_4V(quadColor[0], dest[0]);
 
    /* G */
-   if (!(softpipe->blend->colormask & PIPE_MASK_G))
+   if (!(colormask & PIPE_MASK_G))
       COPY_4V(quadColor[1], dest[1]);
 
    /* B */
-   if (!(softpipe->blend->colormask & PIPE_MASK_B))
+   if (!(colormask & PIPE_MASK_B))
       COPY_4V(quadColor[2], dest[2]);
 
    /* A */
-   if (!(softpipe->blend->colormask & PIPE_MASK_A))
+   if (!(colormask & PIPE_MASK_A))
       COPY_4V(quadColor[3], dest[3]);
 }
 
@@ -773,12 +771,12 @@ blend_fallback(struct quad_stage *qs,
          if (blend->logicop_enable) {
             logicop_quad( qs, quadColor, dest );
          }
-         else if (blend->blend_enable) {
-            blend_quad( qs, quadColor, dest );
+         else if (blend->rt[cbuf].blend_enable) {
+            blend_quad( qs, quadColor, dest, cbuf );
          }
 
-         if (blend->colormask != 0xf)
-            colormask_quad( qs, quadColor, dest );
+         if (blend->rt[cbuf].colormask != 0xf)
+            colormask_quad( blend->rt[cbuf].colormask, quadColor, dest);
    
          /* Output color values
           */
@@ -954,23 +952,23 @@ choose_blend_quad(struct quad_stage *qs,
       qs->run = blend_noop;
    }
    else if (!softpipe->blend->logicop_enable &&
-            softpipe->blend->colormask == 0xf &&
+            softpipe->blend->rt[0].colormask == 0xf &&
             softpipe->framebuffer.nr_cbufs == 1)
    {
-      if (!blend->blend_enable) {
+      if (!blend->rt[0].blend_enable) {
          qs->run = single_output_color;
       }
-      else if (blend->rgb_src_factor == blend->alpha_src_factor &&
-               blend->rgb_dst_factor == blend->alpha_dst_factor &&
-               blend->rgb_func == blend->alpha_func)
+      else if (blend->rt[0].rgb_src_factor == blend->rt[0].alpha_src_factor &&
+               blend->rt[0].rgb_dst_factor == blend->rt[0].alpha_dst_factor &&
+               blend->rt[0].rgb_func == blend->rt[0].alpha_func)
       {
-         if (blend->alpha_func == PIPE_BLEND_ADD) {
-            if (blend->rgb_src_factor == PIPE_BLENDFACTOR_ONE &&
-                blend->rgb_dst_factor == PIPE_BLENDFACTOR_ONE) {
+         if (blend->rt[0].alpha_func == PIPE_BLEND_ADD) {
+            if (blend->rt[0].rgb_src_factor == PIPE_BLENDFACTOR_ONE &&
+                blend->rt[0].rgb_dst_factor == PIPE_BLENDFACTOR_ONE) {
                qs->run = blend_single_add_one_one;
             }
-            else if (blend->rgb_src_factor == PIPE_BLENDFACTOR_SRC_ALPHA &&
-                blend->rgb_dst_factor == PIPE_BLENDFACTOR_INV_SRC_ALPHA)
+            else if (blend->rt[0].rgb_src_factor == PIPE_BLENDFACTOR_SRC_ALPHA &&
+                blend->rt[0].rgb_dst_factor == PIPE_BLENDFACTOR_INV_SRC_ALPHA)
                qs->run = blend_single_add_src_alpha_inv_src_alpha;
 
          }
diff --git a/src/gallium/drivers/softpipe/sp_quad_depth_test.c b/src/gallium/drivers/softpipe/sp_quad_depth_test.c
index 0ca86c4e1c..a981775cbd 100644
--- a/src/gallium/drivers/softpipe/sp_quad_depth_test.c
+++ b/src/gallium/drivers/softpipe/sp_quad_depth_test.c
@@ -30,11 +30,11 @@
  */
 
 #include "pipe/p_defines.h"
+#include "util/u_format.h"
 #include "util/u_memory.h"
 #include "tgsi/tgsi_scan.h"
 #include "sp_context.h"
 #include "sp_quad.h"
-#include "sp_surface.h"
 #include "sp_quad_pipe.h"
 #include "sp_tile_cache.h"
 #include "sp_state.h"           /* for sp_fragment_shader */
@@ -651,6 +651,20 @@ static unsigned mask_count[16] =
 
 
 
+/** helper to get number of Z buffer bits */
+static unsigned
+get_depth_bits(struct quad_stage *qs)
+{
+   struct pipe_surface *zsurf = qs->softpipe->framebuffer.zsbuf;
+   if (zsurf)
+      return util_format_get_component_bits(zsurf->format,
+                                            UTIL_FORMAT_COLORSPACE_ZS, 0);
+   else
+      return 0;
+}
+
+
+
 static void
 depth_test_quads_fallback(struct quad_stage *qs, 
                           struct quad_header *quads[],
@@ -666,7 +680,7 @@ depth_test_quads_fallback(struct quad_stage *qs,
       nr = alpha_test_quads(qs, quads, nr);
    }
 
-   if (qs->softpipe->framebuffer.zsbuf && 
+   if (get_depth_bits(qs) > 0 &&
        (qs->softpipe->depth_stencil->depth.enabled ||
         qs->softpipe->depth_stencil->stencil[0].enabled)) {
 
@@ -884,7 +898,7 @@ choose_depth_test(struct quad_stage *qs,
 
    boolean alpha = qs->softpipe->depth_stencil->alpha.enabled;
 
-   boolean depth = (qs->softpipe->framebuffer.zsbuf && 
+   boolean depth = (get_depth_bits(qs) > 0 &&
                     qs->softpipe->depth_stencil->depth.enabled);
 
    unsigned depthfunc = qs->softpipe->depth_stencil->depth.func;
@@ -895,7 +909,6 @@ choose_depth_test(struct quad_stage *qs,
 
    boolean occlusion = qs->softpipe->active_query_count;
 
-
    if (!alpha &&
        !depth &&
        !stencil) {
diff --git a/src/gallium/drivers/softpipe/sp_quad_fs.c b/src/gallium/drivers/softpipe/sp_quad_fs.c
index 1e7533d0f9..ad04dc2afc 100644
--- a/src/gallium/drivers/softpipe/sp_quad_fs.c
+++ b/src/gallium/drivers/softpipe/sp_quad_fs.c
@@ -45,8 +45,6 @@
 #include "sp_state.h"
 #include "sp_quad.h"
 #include "sp_quad_pipe.h"
-#include "sp_texture.h"
-#include "sp_tex_sample.h"
 
 
 struct quad_shade_stage
@@ -109,10 +107,11 @@ shade_quads(struct quad_stage *qs,
    struct quad_shade_stage *qss = quad_shade_stage( qs );
    struct softpipe_context *softpipe = qs->softpipe;
    struct tgsi_exec_machine *machine = qss->machine;
-
    unsigned i, pass = 0;
-   
-   machine->Consts = softpipe->mapped_constants[PIPE_SHADER_FRAGMENT];
+
+   for (i = 0; i < PIPE_MAX_CONSTANT_BUFFERS; i++) {
+      machine->Consts[i] = softpipe->mapped_constants[PIPE_SHADER_FRAGMENT][i];
+   }
    machine->InterpCoefs = quads[0]->coef;
 
    for (i = 0; i < nr; i++) {
diff --git a/src/gallium/drivers/softpipe/sp_screen.c b/src/gallium/drivers/softpipe/sp_screen.c
index bd3532de4f..714a1cf534 100644
--- a/src/gallium/drivers/softpipe/sp_screen.c
+++ b/src/gallium/drivers/softpipe/sp_screen.c
@@ -91,6 +91,19 @@ softpipe_get_param(struct pipe_screen *screen, int param)
       return 1;
    case PIPE_CAP_BLEND_EQUATION_SEPARATE:
       return 1;
+   case PIPE_CAP_MAX_CONST_BUFFERS:
+      return PIPE_MAX_CONSTANT_BUFFERS;
+   case PIPE_CAP_MAX_CONST_BUFFER_SIZE:
+      return 4096 * 4 * sizeof(float);
+   case PIPE_CAP_INDEP_BLEND_ENABLE:
+      return 1;
+   case PIPE_CAP_INDEP_BLEND_FUNC:
+      return 1;
+   case PIPE_CAP_TGSI_FS_COORD_ORIGIN_UPPER_LEFT:
+   case PIPE_CAP_TGSI_FS_COORD_ORIGIN_LOWER_LEFT:
+   case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_HALF_INTEGER:
+   case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_INTEGER:
+      return 1;
    default:
       return 0;
    }
diff --git a/src/gallium/drivers/softpipe/sp_setup.c b/src/gallium/drivers/softpipe/sp_setup.c
index 3da75364c5..bb1bff581c 100644
--- a/src/gallium/drivers/softpipe/sp_setup.c
+++ b/src/gallium/drivers/softpipe/sp_setup.c
@@ -38,10 +38,8 @@
 #include "sp_setup.h"
 #include "sp_state.h"
 #include "draw/draw_context.h"
-#include "draw/draw_private.h"
 #include "draw/draw_vertex.h"
 #include "pipe/p_shader_tokens.h"
-#include "pipe/p_thread.h"
 #include "util/u_math.h"
 #include "util/u_memory.h"
 
@@ -506,21 +504,24 @@ static void tri_persp_coeff( struct setup_context *setup,
 
 /**
  * Special coefficient setup for gl_FragCoord.
- * X and Y are trivial, though Y has to be inverted for OpenGL.
+ * X and Y are trivial, though Y may have to be inverted for OpenGL.
  * Z and W are copied from posCoef which should have already been computed.
  * We could do a bit less work if we'd examine gl_FragCoord's swizzle mask.
  */
 static void
 setup_fragcoord_coeff(struct setup_context *setup, uint slot)
 {
+   struct sp_fragment_shader* spfs = setup->softpipe->fs;
    /*X*/
-   setup->coef[slot].a0[0] = 0;
+   setup->coef[slot].a0[0] = spfs->pixel_center_integer ? 0.0 : 0.5;
    setup->coef[slot].dadx[0] = 1.0;
    setup->coef[slot].dady[0] = 0.0;
    /*Y*/
-   setup->coef[slot].a0[1] = 0.0;
+   setup->coef[slot].a0[1] =
+		   (spfs->origin_lower_left ? setup->softpipe->framebuffer.height : 0)
+		   + (spfs->pixel_center_integer ? 0.0 : 0.5);
    setup->coef[slot].dadx[1] = 0.0;
-   setup->coef[slot].dady[1] = 1.0;
+   setup->coef[slot].dady[1] = spfs->origin_lower_left ? -1.0 : 1.0;
    /*Z*/
    setup->coef[slot].a0[2] = setup->posCoef.a0[2];
    setup->coef[slot].dadx[2] = setup->posCoef.dadx[2];
diff --git a/src/gallium/drivers/softpipe/sp_state.h b/src/gallium/drivers/softpipe/sp_state.h
index 7f244c4fd4..a83cae7361 100644
--- a/src/gallium/drivers/softpipe/sp_state.h
+++ b/src/gallium/drivers/softpipe/sp_state.h
@@ -68,6 +68,9 @@ struct sp_fragment_shader {
 
    struct tgsi_shader_info info;
 
+   boolean origin_lower_left; /**< fragment shader uses lower left position origin? */
+   boolean pixel_center_integer; /**< fragment shader uses integer pixel center? */
+
    void (*prepare)( const struct sp_fragment_shader *shader,
 		    struct tgsi_exec_machine *machine,
 		    struct tgsi_sampler **samplers);
diff --git a/src/gallium/drivers/softpipe/sp_state_fs.c b/src/gallium/drivers/softpipe/sp_state_fs.c
index b7ed4441b4..de3edde976 100644
--- a/src/gallium/drivers/softpipe/sp_state_fs.c
+++ b/src/gallium/drivers/softpipe/sp_state_fs.c
@@ -44,6 +44,7 @@ softpipe_create_fs_state(struct pipe_context *pipe,
 {
    struct softpipe_context *softpipe = softpipe_context(pipe);
    struct sp_fragment_shader *state;
+   unsigned i;
 
    /* debug */
    if (softpipe->dump_fs) 
@@ -60,6 +61,13 @@ softpipe_create_fs_state(struct pipe_context *pipe,
    /* get/save the summary info for this shader */
    tgsi_scan_shader(templ->tokens, &state->info);
 
+   for (i = 0; i < state->info.num_properties; ++i) {
+      if (state->info.properties[i].name == TGSI_PROPERTY_FS_COORD_ORIGIN)
+         state->origin_lower_left = state->info.properties[i].data[0];
+      else if (state->info.properties[i].name == TGSI_PROPERTY_FS_COORD_PIXEL_CENTER)
+	 state->pixel_center_integer = state->info.properties[i].data[0];
+   }
+
    return state;
 }
 
@@ -164,12 +172,12 @@ softpipe_set_constant_buffer(struct pipe_context *pipe,
    struct softpipe_context *softpipe = softpipe_context(pipe);
 
    assert(shader < PIPE_SHADER_TYPES);
-   assert(index == 0);
+   assert(index < PIPE_MAX_CONSTANT_BUFFERS);
 
    draw_flush(softpipe->draw);
 
    /* note: reference counting */
-   pipe_buffer_reference(&softpipe->constants[shader], buf);
+   pipe_buffer_reference(&softpipe->constants[shader][index], buf);
 
    softpipe->dirty |= SP_NEW_CONSTANTS;
 }
diff --git a/src/gallium/drivers/softpipe/sp_state_surface.c b/src/gallium/drivers/softpipe/sp_state_surface.c
index f6154109ea..3946678219 100644
--- a/src/gallium/drivers/softpipe/sp_state_surface.c
+++ b/src/gallium/drivers/softpipe/sp_state_surface.c
@@ -30,7 +30,6 @@
 
 #include "sp_context.h"
 #include "sp_state.h"
-#include "sp_surface.h"
 #include "sp_tile_cache.h"
 
 #include "draw/draw_context.h"
diff --git a/src/gallium/drivers/softpipe/sp_state_vertex.c b/src/gallium/drivers/softpipe/sp_state_vertex.c
index 46b6991195..b491d92ed1 100644
--- a/src/gallium/drivers/softpipe/sp_state_vertex.c
+++ b/src/gallium/drivers/softpipe/sp_state_vertex.c
@@ -31,7 +31,6 @@
 
 #include "sp_context.h"
 #include "sp_state.h"
-#include "sp_surface.h"
 
 #include "draw/draw_context.h"
 
diff --git a/src/gallium/drivers/softpipe/sp_tex_sample.c b/src/gallium/drivers/softpipe/sp_tex_sample.c
index 1ae8fecacf..473ec3e150 100644
--- a/src/gallium/drivers/softpipe/sp_tex_sample.c
+++ b/src/gallium/drivers/softpipe/sp_tex_sample.c
@@ -517,7 +517,6 @@ compute_lambda_1d(const struct sp_sampler_varient *samp,
                   const float p[QUAD_SIZE])
 {
    const struct pipe_texture *texture = samp->texture;
-   const struct pipe_sampler_state *sampler = samp->sampler;
    float dsdx = fabsf(s[QUAD_BOTTOM_RIGHT] - s[QUAD_BOTTOM_LEFT]);
    float dsdy = fabsf(s[QUAD_TOP_LEFT]     - s[QUAD_BOTTOM_LEFT]);
    float rho = MAX2(dsdx, dsdy) * texture->width0;
@@ -533,7 +532,6 @@ compute_lambda_2d(const struct sp_sampler_varient *samp,
                   const float p[QUAD_SIZE])
 {
    const struct pipe_texture *texture = samp->texture;
-   const struct pipe_sampler_state *sampler = samp->sampler;
    float dsdx = fabsf(s[QUAD_BOTTOM_RIGHT] - s[QUAD_BOTTOM_LEFT]);
    float dsdy = fabsf(s[QUAD_TOP_LEFT]     - s[QUAD_BOTTOM_LEFT]);
    float dtdx = fabsf(t[QUAD_BOTTOM_RIGHT] - t[QUAD_BOTTOM_LEFT]);
@@ -553,7 +551,6 @@ compute_lambda_3d(const struct sp_sampler_varient *samp,
                   const float p[QUAD_SIZE])
 {
    const struct pipe_texture *texture = samp->texture;
-   const struct pipe_sampler_state *sampler = samp->sampler;
    float dsdx = fabsf(s[QUAD_BOTTOM_RIGHT] - s[QUAD_BOTTOM_LEFT]);
    float dsdy = fabsf(s[QUAD_TOP_LEFT]     - s[QUAD_BOTTOM_LEFT]);
    float dtdx = fabsf(t[QUAD_BOTTOM_RIGHT] - t[QUAD_BOTTOM_LEFT]);
diff --git a/src/gallium/drivers/softpipe/sp_tex_tile_cache.c b/src/gallium/drivers/softpipe/sp_tex_tile_cache.c
index e50a76a73b..50242d5bd6 100644
--- a/src/gallium/drivers/softpipe/sp_tex_tile_cache.c
+++ b/src/gallium/drivers/softpipe/sp_tex_tile_cache.c
@@ -37,7 +37,6 @@
 #include "util/u_tile.h"
 #include "util/u_math.h"
 #include "sp_context.h"
-#include "sp_surface.h"
 #include "sp_texture.h"
 #include "sp_tex_tile_cache.h"
 
diff --git a/src/gallium/drivers/softpipe/sp_texture.c b/src/gallium/drivers/softpipe/sp_texture.c
index a9436a3394..a5fff91507 100644
--- a/src/gallium/drivers/softpipe/sp_texture.c
+++ b/src/gallium/drivers/softpipe/sp_texture.c
@@ -38,7 +38,6 @@
 #include "util/u_memory.h"
 
 #include "sp_context.h"
-#include "sp_state.h"
 #include "sp_texture.h"
 #include "sp_screen.h"
 #include "sp_winsys.h"
@@ -57,13 +56,8 @@ softpipe_texture_layout(struct pipe_screen *screen,
    unsigned width = pt->width0;
    unsigned height = pt->height0;
    unsigned depth = pt->depth0;
-
    unsigned buffer_size = 0;
 
-   pt->width0 = width;
-   pt->height0 = height;
-   pt->depth0 = depth;
-
    for (level = 0; level <= pt->last_level; level++) {
       spt->stride[level] = util_format_get_stride(pt->format, width);
 
@@ -296,6 +290,10 @@ softpipe_get_tex_transfer(struct pipe_screen *screen,
    assert(texture);
    assert(level <= texture->last_level);
 
+   /* make sure the requested region is in the image bounds */
+   assert(x + w <= u_minify(texture->width0, level));
+   assert(y + h <= u_minify(texture->height0, level));
+
    spt = CALLOC_STRUCT(softpipe_transfer);
    if (spt) {
       struct pipe_transfer *pt = &spt->base;
diff --git a/src/gallium/drivers/softpipe/sp_video_context.c b/src/gallium/drivers/softpipe/sp_video_context.c
index cae2d3efc5..d0723ebe9e 100644
--- a/src/gallium/drivers/softpipe/sp_video_context.c
+++ b/src/gallium/drivers/softpipe/sp_video_context.c
@@ -185,17 +185,18 @@ init_pipe_state(struct sp_mpeg12_context *ctx)
    ctx->rast = ctx->pipe->create_rasterizer_state(ctx->pipe, &rast);
    ctx->pipe->bind_rasterizer_state(ctx->pipe, ctx->rast);
 
-   blend.blend_enable = 0;
-   blend.rgb_func = PIPE_BLEND_ADD;
-   blend.rgb_src_factor = PIPE_BLENDFACTOR_ONE;
-   blend.rgb_dst_factor = PIPE_BLENDFACTOR_ONE;
-   blend.alpha_func = PIPE_BLEND_ADD;
-   blend.alpha_src_factor = PIPE_BLENDFACTOR_ONE;
-   blend.alpha_dst_factor = PIPE_BLENDFACTOR_ONE;
+   blend.independent_blend_enable = 0;
+   blend.rt[0].blend_enable = 0;
+   blend.rt[0].rgb_func = PIPE_BLEND_ADD;
+   blend.rt[0].rgb_src_factor = PIPE_BLENDFACTOR_ONE;
+   blend.rt[0].rgb_dst_factor = PIPE_BLENDFACTOR_ONE;
+   blend.rt[0].alpha_func = PIPE_BLEND_ADD;
+   blend.rt[0].alpha_src_factor = PIPE_BLENDFACTOR_ONE;
+   blend.rt[0].alpha_dst_factor = PIPE_BLENDFACTOR_ONE;
    blend.logicop_enable = 0;
    blend.logicop_func = PIPE_LOGICOP_CLEAR;
    /* Needed to allow color writes to FB, even if blending disabled */
-   blend.colormask = PIPE_MASK_RGBA;
+   blend.rt[0].colormask = PIPE_MASK_RGBA;
    blend.dither = 0;
    ctx->blend = ctx->pipe->create_blend_state(ctx->pipe, &blend);
    ctx->pipe->bind_blend_state(ctx->pipe, ctx->blend);
diff --git a/src/gallium/drivers/softpipe/sp_winsys.c b/src/gallium/drivers/softpipe/sp_winsys.c
new file mode 100644
index 0000000000..38bcd64c6a
--- /dev/null
+++ b/src/gallium/drivers/softpipe/sp_winsys.c
@@ -0,0 +1,244 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Bismarck, ND., USA
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDERS, AUTHORS AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM,
+ * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR 
+ * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE 
+ * USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * 
+ **************************************************************************/
+
+/**
+ * @file
+ * Malloc softpipe winsys. Uses malloc for all memory allocations.
+ * 
+ * @author Keith Whitwell
+ * @author Brian Paul
+ * @author Jose Fonseca
+ */
+
+
+#include "pipe/internal/p_winsys_screen.h"/* port to just p_screen */
+#include "pipe/p_format.h"
+#include "pipe/p_context.h"
+#include "util/u_format.h"
+#include "util/u_math.h"
+#include "util/u_memory.h"
+#include "softpipe/sp_winsys.h"
+
+
+struct st_softpipe_buffer
+{
+   struct pipe_buffer base;
+   boolean userBuffer;  /** Is this a user-space buffer? */
+   void *data;
+   void *mapped;
+};
+
+
+/** Cast wrapper */
+static INLINE struct st_softpipe_buffer *
+st_softpipe_buffer( struct pipe_buffer *buf )
+{
+   return (struct st_softpipe_buffer *)buf;
+}
+
+
+static void *
+st_softpipe_buffer_map(struct pipe_winsys *winsys, 
+                       struct pipe_buffer *buf,
+                       unsigned flags)
+{
+   struct st_softpipe_buffer *st_softpipe_buf = st_softpipe_buffer(buf);
+   st_softpipe_buf->mapped = st_softpipe_buf->data;
+   return st_softpipe_buf->mapped;
+}
+
+
+static void
+st_softpipe_buffer_unmap(struct pipe_winsys *winsys, 
+                         struct pipe_buffer *buf)
+{
+   struct st_softpipe_buffer *st_softpipe_buf = st_softpipe_buffer(buf);
+   st_softpipe_buf->mapped = NULL;
+}
+
+
+static void
+st_softpipe_buffer_destroy(struct pipe_buffer *buf)
+{
+   struct st_softpipe_buffer *oldBuf = st_softpipe_buffer(buf);
+
+   if (oldBuf->data) {
+      if (!oldBuf->userBuffer)
+         align_free(oldBuf->data);
+
+      oldBuf->data = NULL;
+   }
+
+   FREE(oldBuf);
+}
+
+
+static void
+st_softpipe_flush_frontbuffer(struct pipe_winsys *winsys,
+                              struct pipe_surface *surf,
+                              void *context_private)
+{
+}
+
+
+
+static const char *
+st_softpipe_get_name(struct pipe_winsys *winsys)
+{
+   return "softpipe";
+}
+
+
+static struct pipe_buffer *
+st_softpipe_buffer_create(struct pipe_winsys *winsys, 
+                          unsigned alignment, 
+                          unsigned usage,
+                          unsigned size)
+{
+   struct st_softpipe_buffer *buffer = CALLOC_STRUCT(st_softpipe_buffer);
+
+   pipe_reference_init(&buffer->base.reference, 1);
+   buffer->base.alignment = alignment;
+   buffer->base.usage = usage;
+   buffer->base.size = size;
+
+   buffer->data = align_malloc(size, alignment);
+
+   return &buffer->base;
+}
+
+
+/**
+ * Create buffer which wraps user-space data.
+ */
+static struct pipe_buffer *
+st_softpipe_user_buffer_create(struct pipe_winsys *winsys, 
+                               void *ptr, 
+                               unsigned bytes)
+{
+   struct st_softpipe_buffer *buffer;
+   
+   buffer = CALLOC_STRUCT(st_softpipe_buffer);
+   if(!buffer)
+      return NULL;
+   
+   pipe_reference_init(&buffer->base.reference, 1);
+   buffer->base.size = bytes;
+   buffer->userBuffer = TRUE;
+   buffer->data = ptr;
+
+   return &buffer->base;
+}
+
+
+static struct pipe_buffer *
+st_softpipe_surface_buffer_create(struct pipe_winsys *winsys,
+                                  unsigned width, unsigned height,
+                                  enum pipe_format format,
+                                  unsigned usage,
+                                  unsigned tex_usage,
+                                  unsigned *stride)
+{
+   const unsigned alignment = 64;
+   unsigned nblocksy;
+
+   nblocksy = util_format_get_nblocksy(format, height);
+   *stride = align(util_format_get_stride(format, width), alignment);
+
+   return winsys->buffer_create(winsys, alignment,
+                                usage,
+                                *stride * nblocksy);
+}
+
+
+static void
+st_softpipe_fence_reference(struct pipe_winsys *winsys, 
+                            struct pipe_fence_handle **ptr,
+                            struct pipe_fence_handle *fence)
+{
+}
+
+
+static int
+st_softpipe_fence_signalled(struct pipe_winsys *winsys, 
+                            struct pipe_fence_handle *fence,
+                            unsigned flag)
+{
+   return 0;
+}
+
+
+static int
+st_softpipe_fence_finish(struct pipe_winsys *winsys, 
+                         struct pipe_fence_handle *fence,
+                         unsigned flag)
+{
+   return 0;
+}
+
+
+static void
+st_softpipe_destroy(struct pipe_winsys *winsys)
+{
+   FREE(winsys);
+}
+
+
+struct pipe_screen *
+softpipe_create_screen_malloc(void)
+{
+   static struct pipe_winsys *winsys;
+   struct pipe_screen *screen;
+
+   winsys = CALLOC_STRUCT(pipe_winsys);
+   if(!winsys)
+      return NULL;
+
+   winsys->destroy = st_softpipe_destroy;
+   
+   winsys->buffer_create = st_softpipe_buffer_create;
+   winsys->user_buffer_create = st_softpipe_user_buffer_create;
+   winsys->buffer_map = st_softpipe_buffer_map;
+   winsys->buffer_unmap = st_softpipe_buffer_unmap;
+   winsys->buffer_destroy = st_softpipe_buffer_destroy;
+
+   winsys->surface_buffer_create = st_softpipe_surface_buffer_create;
+
+   winsys->fence_reference = st_softpipe_fence_reference;
+   winsys->fence_signalled = st_softpipe_fence_signalled;
+   winsys->fence_finish = st_softpipe_fence_finish;
+
+   winsys->flush_frontbuffer = st_softpipe_flush_frontbuffer;
+   winsys->get_name = st_softpipe_get_name;
+
+   screen = softpipe_create_screen(winsys);
+   if(!screen)
+      st_softpipe_destroy(winsys);
+
+   return screen;
+}
diff --git a/src/gallium/drivers/softpipe/sp_winsys.h b/src/gallium/drivers/softpipe/sp_winsys.h
index f203ded29e..3042e01a05 100644
--- a/src/gallium/drivers/softpipe/sp_winsys.h
+++ b/src/gallium/drivers/softpipe/sp_winsys.h
@@ -49,10 +49,17 @@ struct pipe_buffer;
 
 struct pipe_context *softpipe_create( struct pipe_screen * );
 
+/**
+ * Create a softpipe screen that uses the
+ * given winsys for allocating buffers.
+ */
+struct pipe_screen *softpipe_create_screen( struct pipe_winsys * );
 
-struct pipe_screen *
-softpipe_create_screen(struct pipe_winsys *);
-
+/**
+ * Create a softpipe screen that uses
+ * regular malloc to create all its buffers.
+ */
+struct pipe_screen *softpipe_create_screen_malloc(void);
 
 boolean
 softpipe_get_texture_buffer( struct pipe_texture *texture,
diff --git a/src/gallium/drivers/svga/svga_draw_arrays.c b/src/gallium/drivers/svga/svga_draw_arrays.c
index 75492dffca..6b6ebc9b58 100644
--- a/src/gallium/drivers/svga/svga_draw_arrays.c
+++ b/src/gallium/drivers/svga/svga_draw_arrays.c
@@ -26,7 +26,6 @@
 #include "svga_cmd.h"
 
 #include "pipe/p_inlines.h"
-#include "util/u_prim.h"
 #include "indices/u_indices.h"
 
 #include "svga_hw_reg.h"
diff --git a/src/gallium/drivers/svga/svga_draw_elements.c b/src/gallium/drivers/svga/svga_draw_elements.c
index 167d817831..022b444eb9 100644
--- a/src/gallium/drivers/svga/svga_draw_elements.c
+++ b/src/gallium/drivers/svga/svga_draw_elements.c
@@ -24,7 +24,6 @@
  **********************************************************/
 
 #include "pipe/p_inlines.h"
-#include "util/u_prim.h"
 #include "util/u_upload_mgr.h"
 #include "indices/u_indices.h"
 
diff --git a/src/gallium/drivers/svga/svga_pipe_blend.c b/src/gallium/drivers/svga/svga_pipe_blend.c
index 855d228755..9df5485f46 100644
--- a/src/gallium/drivers/svga/svga_pipe_blend.c
+++ b/src/gallium/drivers/svga/svga_pipe_blend.c
@@ -29,7 +29,6 @@
 #include "util/u_memory.h"
 
 #include "svga_context.h"
-#include "svga_state.h"
 
 #include "svga_hw_reg.h"
 
@@ -182,15 +181,15 @@ svga_create_blend_state(struct pipe_context *pipe,
          }
       }
       else {
-         blend->rt[i].blend_enable   = templ->blend_enable;
+         blend->rt[i].blend_enable   = templ->rt[0].blend_enable;
 
-         if (templ->blend_enable) {
-            blend->rt[i].srcblend       = svga_translate_blend_factor(templ->rgb_src_factor);
-            blend->rt[i].dstblend       = svga_translate_blend_factor(templ->rgb_dst_factor);
-            blend->rt[i].blendeq        = svga_translate_blend_func(templ->rgb_func);
-            blend->rt[i].srcblend_alpha = svga_translate_blend_factor(templ->alpha_src_factor);
-            blend->rt[i].dstblend_alpha = svga_translate_blend_factor(templ->alpha_dst_factor);
-            blend->rt[i].blendeq_alpha  = svga_translate_blend_func(templ->alpha_func);
+         if (templ->rt[0].blend_enable) {
+            blend->rt[i].srcblend       = svga_translate_blend_factor(templ->rt[0].rgb_src_factor);
+            blend->rt[i].dstblend       = svga_translate_blend_factor(templ->rt[0].rgb_dst_factor);
+            blend->rt[i].blendeq        = svga_translate_blend_func(templ->rt[0].rgb_func);
+            blend->rt[i].srcblend_alpha = svga_translate_blend_factor(templ->rt[0].alpha_src_factor);
+            blend->rt[i].dstblend_alpha = svga_translate_blend_factor(templ->rt[0].alpha_dst_factor);
+            blend->rt[i].blendeq_alpha  = svga_translate_blend_func(templ->rt[0].alpha_func);
 
             if (blend->rt[i].srcblend_alpha != blend->rt[i].srcblend ||
                 blend->rt[i].dstblend_alpha != blend->rt[i].dstblend ||
@@ -201,7 +200,7 @@ svga_create_blend_state(struct pipe_context *pipe,
          }
       }
 
-      blend->rt[i].writemask = templ->colormask;
+      blend->rt[i].writemask = templ->rt[0].colormask;
    }
 
    return blend;
diff --git a/src/gallium/drivers/svga/svga_pipe_constants.c b/src/gallium/drivers/svga/svga_pipe_constants.c
index ca2c7c49d7..9022f7c439 100644
--- a/src/gallium/drivers/svga/svga_pipe_constants.c
+++ b/src/gallium/drivers/svga/svga_pipe_constants.c
@@ -26,13 +26,9 @@
 #include "pipe/p_inlines.h"
 #include "pipe/p_defines.h"
 #include "util/u_math.h"
-#include "util/u_memory.h"
 #include "tgsi/tgsi_parse.h"
 
 #include "svga_context.h"
-#include "svga_state.h"
-#include "svga_hw_reg.h"
-#include "svga_cmd.h"
 
 /***********************************************************************
  * Constant buffers 
diff --git a/src/gallium/drivers/svga/svga_pipe_depthstencil.c b/src/gallium/drivers/svga/svga_pipe_depthstencil.c
index df636c08a0..34e60cb341 100644
--- a/src/gallium/drivers/svga/svga_pipe_depthstencil.c
+++ b/src/gallium/drivers/svga/svga_pipe_depthstencil.c
@@ -29,7 +29,6 @@
 #include "util/u_memory.h"
 
 #include "svga_context.h"
-#include "svga_state.h"
 #include "svga_hw_reg.h"
 
 
diff --git a/src/gallium/drivers/svga/svga_pipe_draw.c b/src/gallium/drivers/svga/svga_pipe_draw.c
index 0f24ef4ee8..4e0c499dc3 100644
--- a/src/gallium/drivers/svga/svga_pipe_draw.c
+++ b/src/gallium/drivers/svga/svga_pipe_draw.c
@@ -33,7 +33,6 @@
 #include "svga_hw_reg.h"
 #include "svga_context.h"
 #include "svga_screen.h"
-#include "svga_winsys.h"
 #include "svga_draw.h"
 #include "svga_state.h"
 #include "svga_swtnl.h"
diff --git a/src/gallium/drivers/svga/svga_pipe_flush.c b/src/gallium/drivers/svga/svga_pipe_flush.c
index 0becb0765a..3eb10336c4 100644
--- a/src/gallium/drivers/svga/svga_pipe_flush.c
+++ b/src/gallium/drivers/svga/svga_pipe_flush.c
@@ -28,13 +28,8 @@
 #include "svga_screen_texture.h"
 #include "svga_context.h"
 #include "svga_winsys.h"
-#include "svga_draw.h"
 #include "svga_debug.h"
 
-#include "svga_hw_reg.h"
-
-
-
 
 static void svga_flush( struct pipe_context *pipe,
                         unsigned flags,
diff --git a/src/gallium/drivers/svga/svga_pipe_fs.c b/src/gallium/drivers/svga/svga_pipe_fs.c
index 5f1213e46a..32f07fb261 100644
--- a/src/gallium/drivers/svga/svga_pipe_fs.c
+++ b/src/gallium/drivers/svga/svga_pipe_fs.c
@@ -32,11 +32,9 @@
 
 #include "svga_screen.h"
 #include "svga_context.h"
-#include "svga_state.h"
 #include "svga_tgsi.h"
 #include "svga_hw_reg.h"
 #include "svga_cmd.h"
-#include "svga_draw.h"
 #include "svga_debug.h"
 
 
diff --git a/src/gallium/drivers/svga/svga_pipe_misc.c b/src/gallium/drivers/svga/svga_pipe_misc.c
index 58cb1e6e23..8cf1f2e083 100644
--- a/src/gallium/drivers/svga/svga_pipe_misc.c
+++ b/src/gallium/drivers/svga/svga_pipe_misc.c
@@ -27,12 +27,6 @@
 
 #include "svga_context.h"
 #include "svga_screen_texture.h"
-#include "svga_state.h"
-#include "svga_winsys.h"
-
-#include "svga_hw_reg.h"
-
-
 
 
 static void svga_set_scissor_state( struct pipe_context *pipe,
diff --git a/src/gallium/drivers/svga/svga_pipe_query.c b/src/gallium/drivers/svga/svga_pipe_query.c
index 01336b0a2c..08283e3731 100644
--- a/src/gallium/drivers/svga/svga_pipe_query.c
+++ b/src/gallium/drivers/svga/svga_pipe_query.c
@@ -32,7 +32,6 @@
 #include "svga_screen.h"
 #include "svga_screen_buffer.h"
 #include "svga_winsys.h"
-#include "svga_draw.h"
 #include "svga_debug.h"
 
 
diff --git a/src/gallium/drivers/svga/svga_pipe_rasterizer.c b/src/gallium/drivers/svga/svga_pipe_rasterizer.c
index b03f8eb9cf..9ea11aad9a 100644
--- a/src/gallium/drivers/svga/svga_pipe_rasterizer.c
+++ b/src/gallium/drivers/svga/svga_pipe_rasterizer.c
@@ -30,7 +30,6 @@
 #include "util/u_memory.h"
 
 #include "svga_context.h"
-#include "svga_state.h"
 
 #include "svga_hw_reg.h"
 
diff --git a/src/gallium/drivers/svga/svga_pipe_sampler.c b/src/gallium/drivers/svga/svga_pipe_sampler.c
index 460a101f8c..161c66dd4f 100644
--- a/src/gallium/drivers/svga/svga_pipe_sampler.c
+++ b/src/gallium/drivers/svga/svga_pipe_sampler.c
@@ -32,9 +32,6 @@
 
 #include "svga_context.h"
 #include "svga_screen_texture.h"
-#include "svga_state.h"
-
-#include "svga_hw_reg.h"
 
 #include "svga_debug.h"
 
diff --git a/src/gallium/drivers/svga/svga_pipe_vertex.c b/src/gallium/drivers/svga/svga_pipe_vertex.c
index 42f290d162..724166a335 100644
--- a/src/gallium/drivers/svga/svga_pipe_vertex.c
+++ b/src/gallium/drivers/svga/svga_pipe_vertex.c
@@ -26,16 +26,11 @@
 #include "pipe/p_inlines.h"
 #include "pipe/p_defines.h"
 #include "util/u_math.h"
-#include "util/u_memory.h"
 #include "tgsi/tgsi_parse.h"
 
 #include "svga_screen.h"
 #include "svga_screen_buffer.h"
 #include "svga_context.h"
-#include "svga_state.h"
-#include "svga_winsys.h"
-
-#include "svga_hw_reg.h"
 
 
 static void svga_set_vertex_buffers(struct pipe_context *pipe,
diff --git a/src/gallium/drivers/svga/svga_pipe_vs.c b/src/gallium/drivers/svga/svga_pipe_vs.c
index 7e6ab576ad..c4ac5304ac 100644
--- a/src/gallium/drivers/svga/svga_pipe_vs.c
+++ b/src/gallium/drivers/svga/svga_pipe_vs.c
@@ -33,7 +33,6 @@
 
 #include "svga_screen.h"
 #include "svga_context.h"
-#include "svga_state.h"
 #include "svga_tgsi.h"
 #include "svga_hw_reg.h"
 #include "svga_cmd.h"
diff --git a/src/gallium/drivers/svga/svga_screen.c b/src/gallium/drivers/svga/svga_screen.c
index fc1b3c980e..c9db9ac874 100644
--- a/src/gallium/drivers/svga/svga_screen.c
+++ b/src/gallium/drivers/svga/svga_screen.c
@@ -33,10 +33,8 @@
 #include "svga_screen.h"
 #include "svga_screen_texture.h"
 #include "svga_screen_buffer.h"
-#include "svga_cmd.h"
 #include "svga_debug.h"
 
-#include "svga_hw_reg.h"
 #include "svga3d_shaderdefs.h"
 
 
@@ -146,6 +144,13 @@ svga_get_paramf(struct pipe_screen *screen, int param)
    case PIPE_CAP_BLEND_EQUATION_SEPARATE: /* req. for GL 1.5 */
       return 1;
 
+   case PIPE_CAP_TGSI_FS_COORD_ORIGIN_UPPER_LEFT:
+   case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_HALF_INTEGER:
+      return 1;
+   case PIPE_CAP_TGSI_FS_COORD_ORIGIN_LOWER_LEFT:
+   case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_INTEGER:
+      return 0;
+
    default:
       return 0;
    }
@@ -393,8 +398,6 @@ svga_screen_create(struct svga_winsys_screen *sws)
    pipe_mutex_init(svgascreen->tex_mutex);
    pipe_mutex_init(svgascreen->swc_mutex);
 
-   LIST_INITHEAD(&svgascreen->cached_buffers);
-   
    svga_screen_cache_init(svgascreen);
 
    return screen;
diff --git a/src/gallium/drivers/svga/svga_screen.h b/src/gallium/drivers/svga/svga_screen.h
index b94ca7fc1c..a009b60720 100644
--- a/src/gallium/drivers/svga/svga_screen.h
+++ b/src/gallium/drivers/svga/svga_screen.h
@@ -68,12 +68,6 @@ struct svga_screen
    pipe_mutex tex_mutex; 
    pipe_mutex swc_mutex; /* Protects the use of swc and dirty_buffers */
    
-   /** 
-    * List of buffers with cached GMR. Ordered from the most recently used to
-    * the least recently used 
-    */
-   struct list_head cached_buffers;
-   
    struct svga_host_surface_cache cache;
 };
 
diff --git a/src/gallium/drivers/svga/svga_screen_buffer.c b/src/gallium/drivers/svga/svga_screen_buffer.c
index 58a1aba464..430a6970fd 100644
--- a/src/gallium/drivers/svga/svga_screen_buffer.c
+++ b/src/gallium/drivers/svga/svga_screen_buffer.c
@@ -113,68 +113,9 @@ svga_buffer_destroy_hw_storage(struct svga_screen *ss, struct svga_buffer *sbuf)
    if(sbuf->hw.buf) {
       sws->buffer_destroy(sws, sbuf->hw.buf);
       sbuf->hw.buf = NULL;
-      assert(sbuf->head.prev && sbuf->head.next);
-      LIST_DEL(&sbuf->head);
-#ifdef DEBUG
-      sbuf->head.next = sbuf->head.prev = NULL; 
-#endif
    }
 }
 
-static INLINE enum pipe_error
-svga_buffer_backup(struct svga_screen *ss, struct svga_buffer *sbuf)
-{
-   if (sbuf->hw.buf && sbuf->hw.num_ranges) {
-      void *src;
-
-      if (!sbuf->swbuf)
-	 sbuf->swbuf = align_malloc(sbuf->base.size, sbuf->base.alignment);
-      if (!sbuf->swbuf)
-	 return PIPE_ERROR_OUT_OF_MEMORY;
-
-      src = ss->sws->buffer_map(ss->sws, sbuf->hw.buf,
-				PIPE_BUFFER_USAGE_CPU_READ);
-      if (!src)
-	 return PIPE_ERROR;
-
-      memcpy(sbuf->swbuf, src, sbuf->base.size);
-      ss->sws->buffer_unmap(ss->sws, sbuf->hw.buf);
-   }
-
-   return PIPE_OK;
-}
-
-/**
- * Try to make GMR space available by freeing the hardware storage of 
- * unmapped
- */
-boolean
-svga_buffer_free_cached_hw_storage(struct svga_screen *ss)
-{
-   struct list_head *curr;
-   struct svga_buffer *sbuf;
-   enum pipe_error ret = PIPE_OK;
-
-   curr = ss->cached_buffers.prev;
-   
-   /* free the least recently used buffer's hw storage which is not mapped */
-   do {
-      if(curr == &ss->cached_buffers)
-         return FALSE;
-
-      sbuf = LIST_ENTRY(struct svga_buffer, curr, head);
-      
-      curr = curr->prev;
-      if (sbuf->map.count == 0)
-	 ret = svga_buffer_backup(ss, sbuf);
-
-   } while(sbuf->map.count != 0 || ret != PIPE_OK);
-   
-   svga_buffer_destroy_hw_storage(ss, sbuf);
-   
-   return TRUE;
-}
-
 struct svga_winsys_buffer *
 svga_winsys_buffer_create( struct svga_screen *ss,
                            unsigned alignment, 
@@ -195,12 +136,6 @@ svga_winsys_buffer_create( struct svga_screen *ss,
       svga_screen_flush(ss, NULL);
       buf = sws->buffer_create(sws, alignment, usage, size);
 
-      SVGA_DBG(DEBUG_DMA|DEBUG_PERF, "evicting buffers to find %d bytes GMR\n", 
-               size);
-
-      /* Try evicing all buffer storage */
-      while(!buf && svga_buffer_free_cached_hw_storage(ss))
-         buf = sws->buffer_create(sws, alignment, usage, size);
    }
    
    return buf;
@@ -226,8 +161,6 @@ svga_buffer_create_hw_storage(struct svga_screen *ss,
          return PIPE_ERROR_OUT_OF_MEMORY;
       
       assert(!sbuf->needs_flush);
-      assert(!sbuf->head.prev && !sbuf->head.next);
-      LIST_ADD(&sbuf->head, &ss->cached_buffers);
    }
    
    return PIPE_OK;
@@ -311,7 +244,6 @@ static void
 svga_buffer_upload_flush(struct svga_context *svga,
                          struct svga_buffer *sbuf)
 {
-   struct svga_screen *ss = svga_screen(svga->pipe.screen);
    SVGA3dCopyBox *boxes;
    unsigned i;
 
@@ -348,13 +280,16 @@ svga_buffer_upload_flush(struct svga_context *svga,
 
    assert(sbuf->head.prev && sbuf->head.next);
    LIST_DEL(&sbuf->head);
+#ifdef DEBUG
+   sbuf->head.next = sbuf->head.prev = NULL; 
+#endif
    sbuf->needs_flush = FALSE;
-   /* XXX: do we care about cached_buffers any more ?*/
-   LIST_ADD(&sbuf->head, &ss->cached_buffers);
 
    sbuf->hw.svga = NULL;
    sbuf->hw.boxes = NULL;
 
+   sbuf->host_written = TRUE;
+
    /* Decrement reference count */
    pipe_reference(&(sbuf->base.reference), NULL);
    sbuf = NULL;
@@ -437,17 +372,17 @@ svga_buffer_map_range( struct pipe_screen *screen,
    }
    else {
       if(!sbuf->hw.buf) {
-         struct svga_winsys_surface *handle = sbuf->handle;
-
          if(svga_buffer_create_hw_storage(ss, sbuf) != PIPE_OK)
             return NULL;
          
          /* Populate the hardware storage if the host surface pre-existed */
-         if((usage & PIPE_BUFFER_USAGE_CPU_READ) && handle) {
+         if(sbuf->host_written) {
             SVGA3dSurfaceDMAFlags flags;
             enum pipe_error ret;
             struct pipe_fence_handle *fence = NULL;
             
+            assert(sbuf->handle);
+
             SVGA_DBG(DEBUG_DMA|DEBUG_PERF, "dma from sid %p (buffer), bytes %u - %u\n", 
                      sbuf->handle, 0, sbuf->base.size);
 
@@ -478,17 +413,6 @@ svga_buffer_map_range( struct pipe_screen *screen,
             sws->fence_reference(sws, &fence, NULL);
          }
       }
-      else {
-         if((usage & PIPE_BUFFER_USAGE_CPU_READ) && !sbuf->needs_flush) {
-            /* We already had the hardware storage but we would have to issue
-             * a download if we hadn't, so move the buffer to the begginning
-             * of the LRU list.
-             */
-            assert(sbuf->head.prev && sbuf->head.next);
-            LIST_DEL(&sbuf->head);
-            LIST_ADD(&sbuf->head, &ss->cached_buffers);
-         }
-      }
          
       map = sws->buffer_map(sws, sbuf->hw.buf, usage);
    }
@@ -572,10 +496,8 @@ svga_buffer_destroy( struct pipe_buffer *buf )
    
    assert(!sbuf->needs_flush);
 
-   if(sbuf->handle) {
-      SVGA_DBG(DEBUG_DMA, "release sid %p sz %d\n", sbuf->handle, sbuf->base.size);
-      svga_screen_surface_destroy(ss, &sbuf->key, &sbuf->handle);
-   }
+   if(sbuf->handle)
+      svga_buffer_destroy_host_surface(ss, sbuf);
    
    if(sbuf->hw.buf)
       svga_buffer_destroy_hw_storage(ss, sbuf);
@@ -595,6 +517,9 @@ svga_buffer_create(struct pipe_screen *screen,
    struct svga_screen *ss = svga_screen(screen);
    struct svga_buffer *sbuf;
    
+   assert(size);
+   assert(alignment);
+
    sbuf = CALLOC_STRUCT(svga_buffer);
    if(!sbuf)
       goto error1;
@@ -755,8 +680,7 @@ svga_buffer_handle(struct svga_context *svga,
       assert(sbuf->hw.svga == svga);
 
       sbuf->needs_flush = TRUE;
-      assert(sbuf->head.prev && sbuf->head.next);
-      LIST_DEL(&sbuf->head);
+      assert(!sbuf->head.prev && !sbuf->head.next);
       LIST_ADDTAIL(&sbuf->head, &svga->dirty_buffers);
    }
 
diff --git a/src/gallium/drivers/svga/svga_screen_buffer.h b/src/gallium/drivers/svga/svga_screen_buffer.h
index 5d7af5a7c5..448ac107c7 100644
--- a/src/gallium/drivers/svga/svga_screen_buffer.h
+++ b/src/gallium/drivers/svga/svga_screen_buffer.h
@@ -135,6 +135,11 @@ struct svga_buffer
     */
    struct svga_winsys_surface *handle;
    
+   /**
+    * Whether the host has been ever written.
+    */
+   boolean host_written;
+
    struct {
       unsigned count;
       boolean writing;
@@ -178,9 +183,6 @@ svga_buffer_handle(struct svga_context *svga,
 void
 svga_context_flush_buffers(struct svga_context *svga);
 
-boolean
-svga_buffer_free_cached_hw_storage(struct svga_screen *ss);
-
 struct svga_winsys_buffer *
 svga_winsys_buffer_create(struct svga_screen *ss,
                           unsigned alignment, 
diff --git a/src/gallium/drivers/svga/svga_screen_texture.c b/src/gallium/drivers/svga/svga_screen_texture.c
index 2224c2d394..0d69007fd8 100644
--- a/src/gallium/drivers/svga/svga_screen_texture.c
+++ b/src/gallium/drivers/svga/svga_screen_texture.c
@@ -306,11 +306,19 @@ svga_texture_create(struct pipe_screen *screen,
       tex->key.numFaces = 1;
    }
 
+   tex->key.cachable = 1;
+
    if(templat->tex_usage & PIPE_TEXTURE_USAGE_SAMPLER)
       tex->key.flags |= SVGA3D_SURFACE_HINT_TEXTURE;
 
-   if(templat->tex_usage & PIPE_TEXTURE_USAGE_PRIMARY)
+   if(templat->tex_usage & PIPE_TEXTURE_USAGE_DISPLAY_TARGET) {
+      tex->key.cachable = 0;
+   }
+
+   if(templat->tex_usage & PIPE_TEXTURE_USAGE_PRIMARY) {
       tex->key.flags |= SVGA3D_SURFACE_HINT_SCANOUT;
+      tex->key.cachable = 0;
+   }
    
    /* 
     * XXX: Never pass the SVGA3D_SURFACE_HINT_RENDERTARGET hint. Mesa cannot
@@ -333,8 +341,6 @@ svga_texture_create(struct pipe_screen *screen,
    if(tex->key.format == SVGA3D_FORMAT_INVALID)
       goto error2;
 
-   tex->key.cachable = 1;
-   
    SVGA_DBG(DEBUG_DMA, "surface_create for texture\n", tex->handle);
    tex->handle = svga_screen_surface_create(svgascreen, &tex->key);
    if (tex->handle)
@@ -416,6 +422,62 @@ svga_texture_blanket(struct pipe_screen * screen,
 }
 
 
+struct pipe_texture *
+svga_screen_texture_wrap_surface(struct pipe_screen *screen,
+				 struct pipe_texture *base,
+				 enum SVGA3dSurfaceFormat format,
+				 struct svga_winsys_surface *srf)
+{
+   struct svga_texture *tex;
+   assert(screen);
+
+   /* Only supports one type */
+   if (base->target != PIPE_TEXTURE_2D ||
+       base->last_level != 0 ||
+       base->depth0 != 1) {
+      return NULL;
+   }
+
+   if (!srf)
+      return NULL;
+
+   if (svga_translate_format(base->format) != format) {
+      unsigned f1 = svga_translate_format(base->format);
+      unsigned f2 = format;
+
+      /* It's okay for XRGB and ARGB or depth with/out stencil to get mixed up */
+      if ( !( (f1 == SVGA3D_X8R8G8B8 && f2 == SVGA3D_A8R8G8B8) ||
+              (f1 == SVGA3D_A8R8G8B8 && f2 == SVGA3D_X8R8G8B8) ||
+              (f1 == SVGA3D_Z_D24X8 && f2 == SVGA3D_Z_D24S8) ) ) {
+         debug_printf("%s wrong format %u != %u\n", __FUNCTION__, f1, f2);
+         return NULL;
+      }
+   }
+
+   tex = CALLOC_STRUCT(svga_texture);
+   if (!tex)
+      return NULL;
+
+   tex->base = *base;
+   
+
+   if (format == 1)
+      tex->base.format = PIPE_FORMAT_X8R8G8B8_UNORM;
+   else if (format == 2)
+      tex->base.format = PIPE_FORMAT_A8R8G8B8_UNORM;
+
+   pipe_reference_init(&tex->base.reference, 1);
+   tex->base.screen = screen;
+
+   SVGA_DBG(DEBUG_DMA, "wrap surface sid %p\n", srf);
+
+   tex->key.cachable = 0;
+   tex->handle = srf;
+
+   return &tex->base;
+}
+
+
 static void
 svga_texture_destroy(struct pipe_texture *pt)
 {
diff --git a/src/gallium/drivers/svga/svga_state_framebuffer.c b/src/gallium/drivers/svga/svga_state_framebuffer.c
index cfdcae4ee4..eda1aefd67 100644
--- a/src/gallium/drivers/svga/svga_state_framebuffer.c
+++ b/src/gallium/drivers/svga/svga_state_framebuffer.c
@@ -32,8 +32,6 @@
 #include "svga_cmd.h"
 #include "svga_debug.h"
 
-#include "svga_hw_reg.h"
-
 
 /***********************************************************************
  * Hardware state update
diff --git a/src/gallium/drivers/svga/svga_state_rss.c b/src/gallium/drivers/svga/svga_state_rss.c
index 8b6803a285..2f9adaeb56 100644
--- a/src/gallium/drivers/svga/svga_state_rss.c
+++ b/src/gallium/drivers/svga/svga_state_rss.c
@@ -31,9 +31,6 @@
 #include "svga_state.h"
 #include "svga_cmd.h"
 
-#include "svga_hw_reg.h"
-
-
 
 struct rs_queue {
    unsigned rs_count;
diff --git a/src/gallium/drivers/svga/svga_state_tss.c b/src/gallium/drivers/svga/svga_state_tss.c
index b313794520..b3c9687b1a 100644
--- a/src/gallium/drivers/svga/svga_state_tss.c
+++ b/src/gallium/drivers/svga/svga_state_tss.c
@@ -33,8 +33,6 @@
 #include "svga_state.h"
 #include "svga_cmd.h"
 
-#include "svga_hw_reg.h"
-
 
 void svga_cleanup_tss_binding(struct svga_context *svga)
 {
diff --git a/src/gallium/drivers/svga/svga_swtnl_backend.c b/src/gallium/drivers/svga/svga_swtnl_backend.c
index b4f757a47a..aafb3e26b0 100644
--- a/src/gallium/drivers/svga/svga_swtnl_backend.c
+++ b/src/gallium/drivers/svga/svga_swtnl_backend.c
@@ -31,7 +31,6 @@
 #include "pipe/p_inlines.h"
 #include "util/u_math.h"
 #include "util/u_memory.h"
-#include "util/u_simple_shaders.h"
 
 #include "svga_context.h"
 #include "svga_state.h"
@@ -87,13 +86,13 @@ svga_vbuf_render_allocate_vertices( struct vbuf_render *render,
    if (!svga_render->vbuf) {
       svga_render->vbuf_size = MAX2(size, svga_render->vbuf_alloc_size);
       svga_render->vbuf = pipe_buffer_create(screen,
-                                             0,
+                                             16,
                                              PIPE_BUFFER_USAGE_VERTEX,
                                              svga_render->vbuf_size);
       if(!svga_render->vbuf) {
          svga_context_flush(svga, NULL);
          svga_render->vbuf = pipe_buffer_create(screen,
-                                                0,
+                                                16,
                                                 PIPE_BUFFER_USAGE_VERTEX,
                                                 svga_render->vbuf_size);
          assert(svga_render->vbuf);
@@ -123,7 +122,9 @@ svga_vbuf_render_map_vertices( struct vbuf_render *render )
    char *ptr = (char*)pipe_buffer_map(screen,
                                       svga_render->vbuf,
                                       PIPE_BUFFER_USAGE_CPU_WRITE | 
-                                      PIPE_BUFFER_USAGE_FLUSH_EXPLICIT);
+                                      PIPE_BUFFER_USAGE_FLUSH_EXPLICIT |
+                                      PIPE_BUFFER_USAGE_DISCARD |
+                                      PIPE_BUFFER_USAGE_UNSYNCHRONIZED);
    return ptr + svga_render->vbuf_offset;
 }
 
@@ -259,14 +260,14 @@ svga_vbuf_render_draw( struct vbuf_render *render,
    if (!svga_render->ibuf) {
       svga_render->ibuf_size = MAX2(size, svga_render->ibuf_alloc_size);
       svga_render->ibuf = pipe_buffer_create(screen,
-                                             0,
+                                             2,
                                              PIPE_BUFFER_USAGE_VERTEX,
                                              svga_render->ibuf_size);
       svga_render->ibuf_offset = 0;
    }
 
-   pipe_buffer_write(screen, svga_render->ibuf,
-                     svga_render->ibuf_offset, 2 * nr_indices, indices);
+   pipe_buffer_write_nooverlap(screen, svga_render->ibuf,
+                                 svga_render->ibuf_offset, 2 * nr_indices, indices);
 
 
    /* off to hardware */
diff --git a/src/gallium/drivers/svga/svga_swtnl_draw.c b/src/gallium/drivers/svga/svga_swtnl_draw.c
index 7655121bec..5e6e30c7df 100644
--- a/src/gallium/drivers/svga/svga_swtnl_draw.c
+++ b/src/gallium/drivers/svga/svga_swtnl_draw.c
@@ -27,7 +27,6 @@
 #include "draw/draw_vbuf.h"
 #include "pipe/p_inlines.h"
 #include "pipe/p_state.h"
-#include "util/u_memory.h"
 
 #include "svga_context.h"
 #include "svga_swtnl.h"
@@ -90,7 +89,7 @@ svga_swtnl_draw_range_elements(struct svga_context *svga,
                             PIPE_BUFFER_USAGE_CPU_READ);
       assert(map);
       draw_set_mapped_constant_buffer(
-         draw, PIPE_SHADER_VERTEX,
+         draw, PIPE_SHADER_VERTEX, 0,
          map,
          svga->curr.cb[PIPE_SHADER_VERTEX]->size);
    }
diff --git a/src/gallium/drivers/svga/svga_swtnl_state.c b/src/gallium/drivers/svga/svga_swtnl_state.c
index 94b6ccc62d..fe03e207ff 100644
--- a/src/gallium/drivers/svga/svga_swtnl_state.c
+++ b/src/gallium/drivers/svga/svga_swtnl_state.c
@@ -27,7 +27,6 @@
 #include "draw/draw_vbuf.h"
 #include "pipe/p_inlines.h"
 #include "pipe/p_state.h"
-#include "util/u_memory.h"
 
 #include "svga_context.h"
 #include "svga_swtnl.h"
diff --git a/src/gallium/drivers/svga/svga_tgsi_decl_sm20.c b/src/gallium/drivers/svga/svga_tgsi_decl_sm20.c
index 23b3ace7f3..1ae9906761 100644
--- a/src/gallium/drivers/svga/svga_tgsi_decl_sm20.c
+++ b/src/gallium/drivers/svga/svga_tgsi_decl_sm20.c
@@ -29,9 +29,6 @@
 #include "util/u_memory.h"
 
 #include "svga_tgsi_emit.h"
-#include "svga_context.h"
-
-
 
 
 static boolean ps20_input( struct svga_shader_emitter *emit,
diff --git a/src/gallium/drivers/svga/svga_tgsi_decl_sm30.c b/src/gallium/drivers/svga/svga_tgsi_decl_sm30.c
index d1c7336dec..43fc0d3235 100644
--- a/src/gallium/drivers/svga/svga_tgsi_decl_sm30.c
+++ b/src/gallium/drivers/svga/svga_tgsi_decl_sm30.c
@@ -29,7 +29,6 @@
 #include "util/u_memory.h"
 
 #include "svga_tgsi_emit.h"
-#include "svga_context.h"
 
 static boolean translate_vs_ps_semantic( struct tgsi_declaration_semantic semantic,
                                          unsigned *usage,
diff --git a/src/gallium/drivers/svga/svga_winsys.h b/src/gallium/drivers/svga/svga_winsys.h
index 59f299c185..27b99fe4c1 100644
--- a/src/gallium/drivers/svga/svga_winsys.h
+++ b/src/gallium/drivers/svga/svga_winsys.h
@@ -296,4 +296,10 @@ svga_screen_buffer_from_texture(struct pipe_texture *texture,
 				struct pipe_buffer **buffer,
 				unsigned *stride);
 
+struct pipe_texture *
+svga_screen_texture_wrap_surface(struct pipe_screen *screen,
+				 struct pipe_texture *base,
+				 enum SVGA3dSurfaceFormat format,
+				 struct svga_winsys_surface *srf);
+
 #endif /* SVGA_WINSYS_H_ */
diff --git a/src/gallium/drivers/trace/tr_drm.c b/src/gallium/drivers/trace/tr_drm.c
index 48d1c4051c..e7ca3a86ea 100644
--- a/src/gallium/drivers/trace/tr_drm.c
+++ b/src/gallium/drivers/trace/tr_drm.c
@@ -173,6 +173,7 @@ trace_drm_create(struct drm_api *api)
    if (!tr_api)
       goto error;
 
+   tr_api->base.driver_name = api->driver_name;
    tr_api->base.create_screen = trace_drm_create_screen;
    tr_api->base.create_context = trace_drm_create_context;
    tr_api->base.texture_from_shared_handle = trace_drm_texture_from_shared_handle;
diff --git a/src/gallium/drivers/trace/tr_dump_state.c b/src/gallium/drivers/trace/tr_dump_state.c
index 32f61f8c94..dbca5de0a5 100644
--- a/src/gallium/drivers/trace/tr_dump_state.c
+++ b/src/gallium/drivers/trace/tr_dump_state.c
@@ -323,9 +323,23 @@ void trace_dump_depth_stencil_alpha_state(const struct pipe_depth_stencil_alpha_
    trace_dump_struct_end();
 }
 
+static void trace_dump_rt_blend_state(const struct pipe_rt_blend_state *state)
+{
+   trace_dump_member(uint, state, rgb_func);
+   trace_dump_member(uint, state, rgb_src_factor);
+   trace_dump_member(uint, state, rgb_dst_factor);
+
+   trace_dump_member(uint, state, alpha_func);
+   trace_dump_member(uint, state, alpha_src_factor);
+   trace_dump_member(uint, state, alpha_dst_factor);
+
+   trace_dump_member(uint, state, colormask);
+
+}
 
 void trace_dump_blend_state(const struct pipe_blend_state *state)
 {
+   unsigned valid_entries = 1;
    if (!trace_dumping_enabled_locked())
       return;
 
@@ -336,21 +350,17 @@ void trace_dump_blend_state(const struct pipe_blend_state *state)
 
    trace_dump_struct_begin("pipe_blend_state");
 
-   trace_dump_member(bool, state, blend_enable);
-
-   trace_dump_member(uint, state, rgb_func);
-   trace_dump_member(uint, state, rgb_src_factor);
-   trace_dump_member(uint, state, rgb_dst_factor);
-
-   trace_dump_member(uint, state, alpha_func);
-   trace_dump_member(uint, state, alpha_src_factor);
-   trace_dump_member(uint, state, alpha_dst_factor);
+   trace_dump_member(bool, state, dither);
 
    trace_dump_member(bool, state, logicop_enable);
    trace_dump_member(uint, state, logicop_func);
 
-   trace_dump_member(uint, state, colormask);
-   trace_dump_member(bool, state, dither);
+   trace_dump_member(bool, state, independent_blend_enable);
+
+   if (state->independent_blend_enable)
+      valid_entries = PIPE_MAX_COLOR_BUFS;
+
+   trace_dump_struct_array(rt_blend_state, state->rt, valid_entries);
 
    trace_dump_struct_end();
 }
diff --git a/src/gallium/include/pipe/p_compiler.h b/src/gallium/include/pipe/p_compiler.h
index 18ebd0c948..6a9018aa3a 100644
--- a/src/gallium/include/pipe/p_compiler.h
+++ b/src/gallium/include/pipe/p_compiler.h
@@ -63,7 +63,7 @@
 #include <stdbool.h>
 
 
-#ifndef __HAIKU__
+#if !defined(__HAIKU__) && !defined(__USE_MISC)
 typedef unsigned int       uint;
 typedef unsigned short     ushort;
 #endif
@@ -104,7 +104,8 @@ typedef unsigned char boolean;
 
 /* Function visibility */
 #ifndef PUBLIC
-#  if defined(__GNUC__) && (__GNUC__ * 100 + __GNUC_MINOR__) >= 303
+#  if (defined(__GNUC__) && (__GNUC__ * 100 + __GNUC_MINOR__) >= 303) \
+	|| (defined(__SUNPRO_C) && (__SUNPRO_C >= 0x590))
 #    define PUBLIC __attribute__((visibility("default")))
 #  else
 #    define PUBLIC
@@ -140,7 +141,7 @@ typedef unsigned char boolean;
 
 
 /* Macros for data alignment. */
-#if defined(__GNUC__)
+#if defined(__GNUC__) || (defined(__SUNPRO_C) && (__SUNPRO_C >= 0x590))
 
 /* See http://gcc.gnu.org/onlinedocs/gcc-4.4.2/gcc/Type-Attributes.html */
 #define PIPE_ALIGN_TYPE(_alignment, _type) _type __attribute__((aligned(_alignment)))
@@ -162,6 +163,13 @@ typedef unsigned char boolean;
 
 #define PIPE_ALIGN_STACK
 
+#elif defined(SWIG)
+
+#define PIPE_ALIGN_TYPE(_alignment, _type) _type
+#define PIPE_ALIGN_VAR(_alignment)
+
+#define PIPE_ALIGN_STACK
+
 #else
 
 #error "Unsupported compiler"
diff --git a/src/gallium/include/pipe/p_context.h b/src/gallium/include/pipe/p_context.h
index 0b8f6da2f4..f1e6a60e04 100644
--- a/src/gallium/include/pipe/p_context.h
+++ b/src/gallium/include/pipe/p_context.h
@@ -103,7 +103,7 @@ struct pipe_context {
    /**
     * Predicate subsequent rendering on occlusion query result
     * \param query  the query predicate, or NULL if no predicate
-    * \param mode  one of PIPE_COND_RENDER_x
+    * \param mode  one of PIPE_RENDER_COND_x
     */
    void (*render_condition)( struct pipe_context *pipe,
                              struct pipe_query *query,
@@ -122,6 +122,11 @@ struct pipe_context {
    void (*begin_query)(struct pipe_context *pipe, struct pipe_query *q);
    void (*end_query)(struct pipe_context *pipe, struct pipe_query *q);
 
+   /**
+    * Get results of a query.
+    * \param wait  if true, this query will block until the result is ready
+    * \return TRUE if results are ready, FALSE otherwise
+    */
    boolean (*get_query_result)(struct pipe_context *pipe, 
                                struct pipe_query *q,
                                boolean wait,
@@ -271,30 +276,30 @@ struct pipe_context {
 
    /**
     * Check whether a texture is referenced by an unflushed hw command.
-    * The state-tracker uses this function to optimize away unnecessary
-    * flushes. It is safe (but wasteful) to always return.
+    * The state-tracker uses this function to avoid unnecessary flushes.
+    * It is safe (but wasteful) to always return
     * PIPE_REFERENCED_FOR_READ | PIPE_REFERENCED_FOR_WRITE.
-    * \param pipe  The pipe context whose unflushed hw commands will be
-    *              checked.
-    * \param level  mipmap level.
+    * \param pipe  context whose unflushed hw commands will be checked.
     * \param texture  texture to check.
     * \param face  cubemap face. Use 0 for non-cubemap texture.
+    * \param level  mipmap level.
+    * \return mask of PIPE_REFERENCED_FOR_READ/WRITE or PIPE_UNREFERENCED
     */
-   unsigned int (*is_texture_referenced) (struct pipe_context *pipe,
-					  struct pipe_texture *texture,
-					  unsigned face, unsigned level);
+   unsigned int (*is_texture_referenced)(struct pipe_context *pipe,
+					 struct pipe_texture *texture,
+					 unsigned face, unsigned level);
 
    /**
     * Check whether a buffer is referenced by an unflushed hw command.
-    * The state-tracker uses this function to optimize away unnecessary
-    * flushes. It is safe (but wasteful) to always return
+    * The state-tracker uses this function to avoid unnecessary flushes.
+    * It is safe (but wasteful) to always return
     * PIPE_REFERENCED_FOR_READ | PIPE_REFERENCED_FOR_WRITE.
-    * \param pipe  The pipe context whose unflushed hw commands will be
-    *              checked.
-    * \param buf  Buffer to check.
+    * \param pipe  context whose unflushed hw commands will be checked.
+    * \param buf  buffer to check.
+    * \return mask of PIPE_REFERENCED_FOR_READ/WRITE or PIPE_UNREFERENCED
     */
-   unsigned int (*is_buffer_referenced) (struct pipe_context *pipe,
-					 struct pipe_buffer *buf);
+   unsigned int (*is_buffer_referenced)(struct pipe_context *pipe,
+					struct pipe_buffer *buf);
 };
 
 
diff --git a/src/gallium/include/pipe/p_defines.h b/src/gallium/include/pipe/p_defines.h
index 35f3830ebc..63ba3119b8 100644
--- a/src/gallium/include/pipe/p_defines.h
+++ b/src/gallium/include/pipe/p_defines.h
@@ -377,7 +377,7 @@ enum pipe_transfer_usage {
 #define PIPE_CAP_NPOT_TEXTURES           2
 #define PIPE_CAP_TWO_SIDED_STENCIL       3
 #define PIPE_CAP_GLSL                    4  /* XXX need something better */
-#define PIPE_CAP_S3TC                    5  /* XXX: deprecated; cap determined via supported sampler formats */
+#define PIPE_CAP_DUAL_SOURCE_BLEND       5  
 #define PIPE_CAP_ANISOTROPIC_FILTER      6
 #define PIPE_CAP_POINT_SPRITE            7
 #define PIPE_CAP_MAX_RENDER_TARGETS      8
@@ -405,6 +405,14 @@ enum pipe_transfer_usage {
 #define PIPE_CAP_MAX_PREDICATE_REGISTERS 30
 #define PIPE_CAP_MAX_COMBINED_SAMPLERS   31  /*< Maximum texture image units accessible from vertex
                                                  and fragment shaders combined */
+#define PIPE_CAP_MAX_CONST_BUFFERS       32
+#define PIPE_CAP_MAX_CONST_BUFFER_SIZE   33  /*< In bytes */
+#define PIPE_CAP_INDEP_BLEND_ENABLE      34  /*< blend enables and write masks per rendertarget */
+#define PIPE_CAP_INDEP_BLEND_FUNC        35  /*< different blend funcs per rendertarget */
+#define PIPE_CAP_TGSI_FS_COORD_ORIGIN_UPPER_LEFT 36
+#define PIPE_CAP_TGSI_FS_COORD_ORIGIN_LOWER_LEFT 37
+#define PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_HALF_INTEGER 38
+#define PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_INTEGER 39
 
 
 /**
diff --git a/src/gallium/include/pipe/p_inlines.h b/src/gallium/include/pipe/p_inlines.h
index 5fbd62a03d..72f5c1dc2a 100644
--- a/src/gallium/include/pipe/p_inlines.h
+++ b/src/gallium/include/pipe/p_inlines.h
@@ -63,13 +63,6 @@ pipe_buffer_map(struct pipe_screen *screen,
    if(screen->buffer_map_range) {
       unsigned offset = 0;
       unsigned length = buf->size;
-
-      /* XXX: Actually we should be using/detecting DISCARD
-       * instead of assuming that WRITE implies discard */
-      if((usage & PIPE_BUFFER_USAGE_CPU_WRITE) &&
-         !(usage & PIPE_BUFFER_USAGE_DISCARD))
-         usage |= PIPE_BUFFER_USAGE_CPU_READ;
-
       return screen->buffer_map_range(screen, buf, offset, length, usage);
    }
    else
@@ -126,7 +119,39 @@ pipe_buffer_write(struct pipe_screen *screen,
 
    map = pipe_buffer_map_range(screen, buf, offset, size, 
                                PIPE_BUFFER_USAGE_CPU_WRITE | 
-                               PIPE_BUFFER_USAGE_FLUSH_EXPLICIT);
+                               PIPE_BUFFER_USAGE_FLUSH_EXPLICIT |
+                               PIPE_BUFFER_USAGE_DISCARD);
+   assert(map);
+   if(map) {
+      memcpy((uint8_t *)map + offset, data, size);
+      pipe_buffer_flush_mapped_range(screen, buf, offset, size);
+      pipe_buffer_unmap(screen, buf);
+   }
+}
+
+/**
+ * Special case for writing non-overlapping ranges.
+ *
+ * We can avoid GPU/CPU synchronization when writing range that has never
+ * been written before.
+ */
+static INLINE void
+pipe_buffer_write_nooverlap(struct pipe_screen *screen,
+                            struct pipe_buffer *buf,
+                            unsigned offset, unsigned size,
+                            const void *data)
+{
+   void *map;
+
+   assert(offset < buf->size);
+   assert(offset + size <= buf->size);
+   assert(size);
+
+   map = pipe_buffer_map_range(screen, buf, offset, size,
+                               PIPE_BUFFER_USAGE_CPU_WRITE |
+                               PIPE_BUFFER_USAGE_FLUSH_EXPLICIT |
+                               PIPE_BUFFER_USAGE_DISCARD |
+                               PIPE_BUFFER_USAGE_UNSYNCHRONIZED);
    assert(map);
    if(map) {
       memcpy((uint8_t *)map + offset, data, size);
diff --git a/src/gallium/include/pipe/p_shader_tokens.h b/src/gallium/include/pipe/p_shader_tokens.h
index b489b04466..21581a4674 100644
--- a/src/gallium/include/pipe/p_shader_tokens.h
+++ b/src/gallium/include/pipe/p_shader_tokens.h
@@ -109,10 +109,11 @@ struct tgsi_declaration
    unsigned File        : 4;  /**< one of TGSI_FILE_x */
    unsigned UsageMask   : 4;  /**< bitmask of TGSI_WRITEMASK_x flags */
    unsigned Interpolate : 4;  /**< one of TGSI_INTERPOLATE_x */
+   unsigned Dimension   : 1;  /**< any extra dimension info? */
    unsigned Semantic    : 1;  /**< BOOL, any semantic info? */
    unsigned Centroid    : 1;  /**< centroid sampling? */
    unsigned Invariant   : 1;  /**< invariant optimization? */
-   unsigned Padding     : 5;
+   unsigned Padding     : 4;
 };
 
 struct tgsi_declaration_range
@@ -121,6 +122,12 @@ struct tgsi_declaration_range
    unsigned Last    : 16; /**< UINT */
 };
 
+struct tgsi_declaration_dimension
+{
+   unsigned Index2D:16; /**< UINT */
+   unsigned Padding:16;
+};
+
 #define TGSI_SEMANTIC_POSITION   0
 #define TGSI_SEMANTIC_COLOR      1
 #define TGSI_SEMANTIC_BCOLOR     2  /**< back-face color */
@@ -163,7 +170,9 @@ union tgsi_immediate_data
 #define TGSI_PROPERTY_GS_INPUT_PRIM          0
 #define TGSI_PROPERTY_GS_OUTPUT_PRIM         1
 #define TGSI_PROPERTY_GS_MAX_VERTICES        2
-#define TGSI_PROPERTY_COUNT                  3
+#define TGSI_PROPERTY_FS_COORD_ORIGIN        3
+#define TGSI_PROPERTY_FS_COORD_PIXEL_CENTER  4
+#define TGSI_PROPERTY_COUNT                  5
 
 struct tgsi_property {
    unsigned Type         : 4;  /**< TGSI_TOKEN_TYPE_PROPERTY */
@@ -172,6 +181,12 @@ struct tgsi_property {
    unsigned Padding      : 12;
 };
 
+#define TGSI_FS_COORD_ORIGIN_UPPER_LEFT 0
+#define TGSI_FS_COORD_ORIGIN_LOWER_LEFT 1
+
+#define TGSI_FS_COORD_PIXEL_CENTER_HALF_INTEGER 0
+#define TGSI_FS_COORD_PIXEL_CENTER_INTEGER 1
+
 struct tgsi_property_data {
    unsigned Data;
 };
diff --git a/src/gallium/include/pipe/p_state.h b/src/gallium/include/pipe/p_state.h
index fdd29ed449..03cd74efed 100644
--- a/src/gallium/include/pipe/p_state.h
+++ b/src/gallium/include/pipe/p_state.h
@@ -58,7 +58,7 @@ extern "C" {
 #define PIPE_MAX_ATTRIBS          32
 #define PIPE_MAX_CLIP_PLANES       6
 #define PIPE_MAX_COLOR_BUFS        8
-#define PIPE_MAX_CONSTANT         32
+#define PIPE_MAX_CONSTANT_BUFFERS 32
 #define PIPE_MAX_SAMPLERS         16
 #define PIPE_MAX_VERTEX_SAMPLERS  16
 #define PIPE_MAX_SHADER_INPUTS    16
@@ -217,7 +217,7 @@ struct pipe_depth_stencil_alpha_state
 };
 
 
-struct pipe_blend_state
+struct pipe_rt_blend_state
 {
    unsigned blend_enable:1;
 
@@ -229,11 +229,16 @@ struct pipe_blend_state
    unsigned alpha_src_factor:5;  /**< PIPE_BLENDFACTOR_x */
    unsigned alpha_dst_factor:5;  /**< PIPE_BLENDFACTOR_x */
 
+   unsigned colormask:4;         /**< bitmask of PIPE_MASK_R/G/B/A */
+};
+
+struct pipe_blend_state
+{
+   unsigned independent_blend_enable:1;
    unsigned logicop_enable:1;
    unsigned logicop_func:4;      /**< PIPE_LOGICOP_x */
-
-   unsigned colormask:4;         /**< bitmask of PIPE_MASK_R/G/B/A */
    unsigned dither:1;
+   struct pipe_rt_blend_state rt[PIPE_MAX_COLOR_BUFS];
 };
 
 
@@ -269,7 +274,7 @@ struct pipe_sampler_state
    unsigned compare_mode:1;      /**< PIPE_TEX_COMPARE_x */
    unsigned compare_func:3;      /**< PIPE_FUNC_x */
    unsigned normalized_coords:1; /**< Are coords normalized to [0,1]? */
-   unsigned prefilter:4;         /**< Wierd sampling state exposed by some api's */
+   unsigned prefilter:4;         /**< Cylindrical texcoord wrap, per coord, exposed by some api's */
    float lod_bias;               /**< LOD/lambda bias */
    float min_lod, max_lod;       /**< LOD clamp range, after bias */
    float border_color[4];
diff --git a/src/gallium/include/state_tracker/drm_api.h b/src/gallium/include/state_tracker/drm_api.h
index bb928928c9..b248a81880 100644
--- a/src/gallium/include/state_tracker/drm_api.h
+++ b/src/gallium/include/state_tracker/drm_api.h
@@ -31,6 +31,11 @@ struct drm_api
         const char *name;
 
 	/**
+	 * Kernel driver name, as accepted by drmOpenByName.
+	 */
+	const char *driver_name;
+
+	/**
 	 * Special buffer functions
 	 */
 	/*@{*/
diff --git a/src/gallium/state_trackers/Makefile b/src/gallium/state_trackers/Makefile
index 265ca468c2..0900efc664 100644
--- a/src/gallium/state_trackers/Makefile
+++ b/src/gallium/state_trackers/Makefile
@@ -21,5 +21,9 @@ clean:
 	rm -f `find . -name depend`
 
 
-# Dummy install target
 install:
+	@for dir in $(SUBDIRS) ; do \
+		if [ -d $$dir ] ; then \
+			(cd $$dir && $(MAKE) $@) || exit 1 ; \
+		fi \
+	done
diff --git a/src/gallium/state_trackers/dri/dri_context.c b/src/gallium/state_trackers/dri/dri_context.c
index f2e5f3fb23..07f0554cc0 100644
--- a/src/gallium/state_trackers/dri/dri_context.c
+++ b/src/gallium/state_trackers/dri/dri_context.c
@@ -101,6 +101,12 @@ dri_destroy_context(__DRIcontext * cPriv)
 {
    struct dri_context *ctx = dri_context(cPriv);
 
+   /* note: we are freeing values and nothing more because
+    * driParseConfigFiles allocated values only - the rest
+    * is owned by screen optionCache.
+    */
+   FREE(ctx->optionCache.values);
+
    /* No particular reason to wait for command completion before
     * destroying a context, but it is probably worthwhile flushing it
     * to avoid having to add code elsewhere to cope with flushing a
diff --git a/src/gallium/state_trackers/dri/dri_drawable.c b/src/gallium/state_trackers/dri/dri_drawable.c
index 0fdfa96b35..1f456cb829 100644
--- a/src/gallium/state_trackers/dri/dri_drawable.c
+++ b/src/gallium/state_trackers/dri/dri_drawable.c
@@ -35,7 +35,6 @@
 
 #include "pipe/p_context.h"
 #include "pipe/p_screen.h"
-#include "pipe/p_inlines.h"
 #include "main/mtypes.h"
 #include "main/renderbuffer.h"
 #include "state_tracker/drm_api.h"
@@ -123,11 +122,12 @@ dri_get_buffers(__DRIdrawable * dPriv)
 
    struct dri_drawable *drawable = dri_drawable(dPriv);
    struct pipe_surface *surface = NULL;
-   struct pipe_screen *screen = dri_screen(drawable->sPriv)->pipe_screen;
+   struct dri_screen *st_screen = dri_screen(drawable->sPriv);
+   struct pipe_screen *screen = st_screen->pipe_screen;
    __DRIbuffer *buffers = NULL;
    __DRIscreen *dri_screen = drawable->sPriv;
    __DRIdrawable *dri_drawable = drawable->dPriv;
-   struct drm_api *api = ((struct dri_screen*)(dri_screen->private))->api;
+   struct drm_api *api = st_screen->api;
    boolean have_depth = FALSE;
    int i, count;
 
@@ -180,7 +180,9 @@ dri_get_buffers(__DRIdrawable * dPriv)
 
       switch (buffers[i].attachment) {
       case __DRI_BUFFER_FRONT_LEFT:
-	 continue;
+	 if (!st_screen->auto_fake_front)
+	    continue;
+	 /* fallthrough */
       case __DRI_BUFFER_FAKE_FRONT_LEFT:
 	 index = ST_SURFACE_FRONT_LEFT;
 	 format = drawable->color_format;
@@ -373,8 +375,8 @@ dri_create_buffer(__DRIscreen * sPriv,
    /* TODO incase of double buffer visual, delay fake creation */
    i = 0;
    drawable->attachments[i++] = __DRI_BUFFER_FRONT_LEFT;
-   drawable->attachments[i++] = __DRI_BUFFER_FAKE_FRONT_LEFT;
-
+   if (!screen->auto_fake_front)
+      drawable->attachments[i++] = __DRI_BUFFER_FAKE_FRONT_LEFT;
    if (visual->doubleBufferMode)
       drawable->attachments[i++] = __DRI_BUFFER_BACK_LEFT;
    if (visual->depthBits && visual->stencilBits)
diff --git a/src/gallium/state_trackers/dri/dri_extensions.c b/src/gallium/state_trackers/dri/dri_extensions.c
index 8b014a2a8b..1259813a41 100644
--- a/src/gallium/state_trackers/dri/dri_extensions.c
+++ b/src/gallium/state_trackers/dri/dri_extensions.c
@@ -50,6 +50,7 @@
 #define need_GL_EXT_blend_func_separate
 #define need_GL_EXT_blend_minmax
 #define need_GL_EXT_cull_vertex
+#define need_GL_EXT_draw_buffers2
 #define need_GL_EXT_fog_coord
 #define need_GL_EXT_framebuffer_object
 #define need_GL_EXT_multi_draw_arrays
@@ -98,6 +99,7 @@ static const struct dri_extension card_extensions[] = {
    {"GL_EXT_blend_minmax", GL_EXT_blend_minmax_functions},
    {"GL_EXT_blend_subtract", NULL},
    {"GL_EXT_cull_vertex", GL_EXT_cull_vertex_functions},
+   {"GL_EXT_draw_buffers2", GL_EXT_draw_buffers2_functions},
    {"GL_EXT_fog_coord", GL_EXT_fog_coord_functions},
    {"GL_EXT_framebuffer_object", GL_EXT_framebuffer_object_functions},
    {"GL_EXT_multi_draw_arrays", GL_EXT_multi_draw_arrays_functions},
@@ -130,6 +132,9 @@ dri_init_extensions(struct dri_context *ctx)
    /* The card_extensions list should be pruned according to the
     * capabilities of the pipe_screen. This is actually something
     * that can/should be done inside st_create_context().
+    * XXX Not pruning is very bogus. Always all these extensions above
+    * will be advertized, regardless what st_init_extensions
+    * (which depends on the pipe cap bits) does.
     */
    driInitExtensions(ctx->st->ctx, card_extensions, GL_TRUE);
 }
diff --git a/src/gallium/state_trackers/dri/dri_screen.c b/src/gallium/state_trackers/dri/dri_screen.c
index 793db087ee..a412b81256 100644
--- a/src/gallium/state_trackers/dri/dri_screen.c
+++ b/src/gallium/state_trackers/dri/dri_screen.c
@@ -37,14 +37,10 @@
 #include "dri_context.h"
 #include "dri_drawable.h"
 
-#include "pipe/p_context.h"
 #include "pipe/p_screen.h"
-#include "pipe/p_inlines.h"
 #include "pipe/p_format.h"
 #include "state_tracker/drm_api.h"
 #include "state_tracker/dri1_api.h"
-#include "state_tracker/st_public.h"
-#include "state_tracker/st_cb_fbo.h"
 
 PUBLIC const char __driConfigOptions[] =
    DRI_CONF_BEGIN DRI_CONF_SECTION_PERFORMANCE
@@ -83,7 +79,7 @@ dri_fill_in_modes(struct dri_screen *screen,
    unsigned num_modes;
    uint8_t depth_bits_array[5];
    uint8_t stencil_bits_array[5];
-   uint8_t msaa_samples_array[1];
+   uint8_t msaa_samples_array[2];
    unsigned depth_buffer_factor;
    unsigned back_buffer_factor;
    unsigned msaa_samples_factor;
@@ -147,8 +143,9 @@ dri_fill_in_modes(struct dri_screen *screen,
    }
 
    msaa_samples_array[0] = 0;
+   msaa_samples_array[1] = 4;
    back_buffer_factor = 3;
-   msaa_samples_factor = 1;
+   msaa_samples_factor = 2;
 
    num_modes =
       depth_buffer_factor * back_buffer_factor * msaa_samples_factor * 4;
@@ -158,7 +155,7 @@ dri_fill_in_modes(struct dri_screen *screen,
 				 depth_bits_array, stencil_bits_array,
 				 depth_buffer_factor, back_buffer_modes,
 				 back_buffer_factor,
-				 msaa_samples_array, 1);
+				 msaa_samples_array, msaa_samples_factor);
    } else {
       __DRIconfig **configs_a8r8g8b8 = NULL;
       __DRIconfig **configs_x8r8g8b8 = NULL;
@@ -170,7 +167,8 @@ dri_fill_in_modes(struct dri_screen *screen,
 					     depth_buffer_factor,
 					     back_buffer_modes,
 					     back_buffer_factor,
-					     msaa_samples_array, 1);
+					     msaa_samples_array,
+                                             msaa_samples_factor);
       if (pf_x8r8g8b8)
 	 configs_x8r8g8b8 = driCreateConfigs(GL_BGR, GL_UNSIGNED_INT_8_8_8_8_REV,
 					     depth_bits_array,
@@ -178,7 +176,8 @@ dri_fill_in_modes(struct dri_screen *screen,
 					     depth_buffer_factor,
 					     back_buffer_modes,
 					     back_buffer_factor,
-					     msaa_samples_array, 1);
+					     msaa_samples_array,
+                                             msaa_samples_factor);
 
       if (configs_a8r8g8b8 && configs_x8r8g8b8)
 	 configs = driConcatConfigs(configs_x8r8g8b8, configs_a8r8g8b8);
@@ -289,6 +288,8 @@ dri_init_screen2(__DRIscreen * sPriv)
 {
    struct dri_screen *screen;
    struct drm_create_screen_arg arg;
+   const __DRIdri2LoaderExtension *dri2_ext =
+     sPriv->dri2.loader;
 
    screen = CALLOC_STRUCT(dri_screen);
    if (!screen)
@@ -314,6 +315,9 @@ dri_init_screen2(__DRIscreen * sPriv)
    driParseOptionInfo(&screen->optionCache,
 		      __driConfigOptions, __driNConfigOptions);
 
+   screen->auto_fake_front = dri2_ext->base.version >= 3 &&
+      dri2_ext->getBuffersWithFormat != NULL;
+
    return dri_fill_in_modes(screen, 32);
  fail:
    return NULL;
@@ -323,8 +327,18 @@ static void
 dri_destroy_screen(__DRIscreen * sPriv)
 {
    struct dri_screen *screen = dri_screen(sPriv);
+   int i;
 
    screen->pipe_screen->destroy(screen->pipe_screen);
+   
+   for (i = 0; i < (1 << screen->optionCache.tableSize); ++i) {
+      FREE(screen->optionCache.info[i].name);
+      FREE(screen->optionCache.info[i].ranges);
+   }
+
+   FREE(screen->optionCache.info);
+   FREE(screen->optionCache.values);
+
    FREE(screen);
    sPriv->private = NULL;
 }
diff --git a/src/gallium/state_trackers/dri/dri_screen.h b/src/gallium/state_trackers/dri/dri_screen.h
index 03387a0e81..75a0ee4250 100644
--- a/src/gallium/state_trackers/dri/dri_screen.h
+++ b/src/gallium/state_trackers/dri/dri_screen.h
@@ -59,6 +59,7 @@ struct dri_screen
    struct pipe_screen *pipe_screen;
    boolean d_depth_bits_last;
    boolean sd_depth_bits_last;
+   boolean auto_fake_front;
 };
 
 /** cast wrapper */
diff --git a/src/gallium/state_trackers/egl/Makefile b/src/gallium/state_trackers/egl/Makefile
index e825aa718b..b696f2fae9 100644
--- a/src/gallium/state_trackers/egl/Makefile
+++ b/src/gallium/state_trackers/egl/Makefile
@@ -1,19 +1,75 @@
 TOP = ../../../..
 include $(TOP)/configs/current
 
-LIBNAME = egldrm
-
-LIBRARY_INCLUDES = \
+common_INCLUDES = \
+	-I. \
 	-I$(TOP)/src/gallium/include \
 	-I$(TOP)/src/gallium/auxiliary \
-	-I$(TOP)/src/mesa/drivers/dri/common \
-	-I$(TOP)/src/mesa \
-	-I$(TOP)/include \
 	-I$(TOP)/src/egl/main \
+	-I$(TOP)/include
+
+common_SOURCES = $(wildcard common/*.c)
+common_OBJECTS = $(common_SOURCES:.c=.o)
+
+
+x11_INCLUDES = \
+	-I$(TOP)/src/gallium/drivers \
+	-I$(TOP)/src/glx/x11 \
+	-I$(TOP)/src/egl/drivers/xdri \
+	-I$(TOP)/src/mesa \
 	$(shell pkg-config --cflags-only-I libdrm)
 
+x11_SOURCES = $(wildcard x11/*.c) \
+	      $(TOP)/src/glx/x11/dri2.c \
+	      $(TOP)/src/egl/drivers/xdri/glxinit.c
+x11_OBJECTS = $(x11_SOURCES:.c=.o)
+
+
+kms_INCLUDES = $(shell pkg-config --cflags-only-I libdrm)
+kms_SOURCES = $(wildcard kms/*.c)
+kms_OBJECTS = $(kms_SOURCES:.c=.o)
+
+
+ALL_INCLUDES = $(common_INCLUDES) $(x11_INCLUDES) $(kms_INCLUDES)
+ALL_SOURCES = $(common_SOURCES) $(x11_SOURCES) $(kms_SOURCES)
+ALL_OBJECTS = $(common_OBJECTS) $(x11_OBJECTS) $(kms_OBJECTS)
+
+##### TARGETS #####
+
+EGL_DISPLAYS_MODS = $(foreach dpy, $(EGL_DISPLAYS), libegl$(dpy).a)
+
+default: depend $(EGL_DISPLAYS_MODS)
+
+
+libeglx11.a: $(x11_OBJECTS) $(common_OBJECTS) Makefile
+	$(MKLIB) -o eglx11 -static $(x11_OBJECTS) $(common_OBJECTS)
+
+libeglkms.a: $(kms_OBJECTS) $(common_OBJECTS) Makefile
+	$(MKLIB) -o eglkms -static $(kms_OBJECTS) $(common_OBJECTS)
+
+depend: 
+	rm -f depend
+	touch depend
+	$(MKDEP) $(MKDEP_OPTIONS) $(ALL_INCLUDES) $(ALL_SOURCES) 2> /dev/null
+
+clean:
+	rm -f $(ALL_OBJECTS)
+	rm -f $(EGL_DISPLAYS_MODS)
+	rm -f depend depend.bak
+
+# Dummy target
+install:
+	@echo -n ""
+
+##### RULES #####
+
+$(common_OBJECTS): %.o: %.c
+	$(CC) -c $(common_INCLUDES) $(DEFINES) $(CFLAGS) $< -o $@
 
-C_SOURCES = $(wildcard ./*.c)
+$(x11_OBJECTS): %.o: %.c
+	$(CC) -c $(common_INCLUDES) $(x11_INCLUDES) $(DEFINES) $(CFLAGS) $< -o $@
 
+$(kms_OBJECTS): %.o: %.c
+	$(CC) -c $(common_INCLUDES) $(kms_INCLUDES) $(DEFINES) $(CFLAGS) $< -o $@
 
-include ../../Makefile.template
+sinclude depend
diff --git a/src/gallium/state_trackers/egl_g3d/common/egl_g3d.c b/src/gallium/state_trackers/egl/common/egl_g3d.c
index 51da8e19f5..6c8f3b9f79 100644
--- a/src/gallium/state_trackers/egl_g3d/common/egl_g3d.c
+++ b/src/gallium/state_trackers/egl/common/egl_g3d.c
@@ -26,6 +26,7 @@
 #include <string.h>
 #include "pipe/p_screen.h"
 #include "util/u_memory.h"
+#include "util/u_rect.h"
 #include "egldriver.h"
 #include "eglcurrent.h"
 #include "eglconfigutil.h"
@@ -44,8 +45,13 @@ egl_g3d_validate_context(_EGLDisplay *dpy, _EGLContext *ctx)
    struct egl_g3d_display *gdpy = egl_g3d_display(dpy);
    struct pipe_screen *screen = gdpy->native->screen;
    struct egl_g3d_context *gctx = egl_g3d_context(ctx);
-   EGLint num_surfaces;
-   EGLint s, i;
+   const uint st_att_map[NUM_NATIVE_ATTACHMENTS] = {
+      ST_SURFACE_FRONT_LEFT,
+      ST_SURFACE_BACK_LEFT,
+      ST_SURFACE_FRONT_RIGHT,
+      ST_SURFACE_BACK_RIGHT,
+   };
+   EGLint num_surfaces, s;
 
    /* validate draw and/or read buffers */
    num_surfaces = (gctx->base.ReadSurface == gctx->base.DrawSurface) ? 1 : 2;
@@ -53,6 +59,7 @@ egl_g3d_validate_context(_EGLDisplay *dpy, _EGLContext *ctx)
       struct pipe_texture *textures[NUM_NATIVE_ATTACHMENTS];
       struct egl_g3d_surface *gsurf;
       struct egl_g3d_buffer *gbuf;
+      EGLint att;
 
       if (s == 0) {
          gsurf = egl_g3d_surface(gctx->base.DrawSurface);
@@ -66,30 +73,31 @@ egl_g3d_validate_context(_EGLDisplay *dpy, _EGLContext *ctx)
       if (!gctx->force_validate) {
          unsigned int seq_num;
 
-         gsurf->native->validate(gsurf->native,
-               gbuf->native_atts, gbuf->num_atts,
+         gsurf->native->validate(gsurf->native, gbuf->attachment_mask,
                &seq_num, NULL, NULL, NULL);
          /* skip validation */
          if (gsurf->sequence_number == seq_num)
             continue;
       }
 
-      gsurf->native->validate(gsurf->native,
-            gbuf->native_atts, gbuf->num_atts,
+      pipe_surface_reference(&gsurf->render_surface, NULL);
+      memset(textures, 0, sizeof(textures));
+
+      gsurf->native->validate(gsurf->native, gbuf->attachment_mask,
             &gsurf->sequence_number, textures,
             &gsurf->base.Width, &gsurf->base.Height);
-      for (i = 0; i < gbuf->num_atts; i++) {
-         struct pipe_texture *pt = textures[i];
+      for (att = 0; att < NUM_NATIVE_ATTACHMENTS; att++) {
+         struct pipe_texture *pt = textures[att];
          struct pipe_surface *ps;
 
-         if (pt) {
+         if (native_attachment_mask_test(gbuf->attachment_mask, att) && pt) {
             ps = screen->get_tex_surface(screen, pt, 0, 0, 0,
                   PIPE_BUFFER_USAGE_GPU_READ |
                   PIPE_BUFFER_USAGE_GPU_WRITE);
             gctx->stapi->st_set_framebuffer_surface(gbuf->st_fb,
-                  gbuf->st_atts[i], ps);
+                  st_att_map[att], ps);
 
-            if (gbuf->native_atts[i] == gsurf->render_att)
+            if (gsurf->render_att == att)
                pipe_surface_reference(&gsurf->render_surface, ps);
 
             pipe_surface_reference(&ps, NULL);
@@ -128,13 +136,7 @@ static void
 egl_g3d_route_context(_EGLDisplay *dpy, _EGLContext *ctx)
 {
    struct egl_g3d_context *gctx = egl_g3d_context(ctx);
-   const uint st_att_map[NUM_NATIVE_ATTACHMENTS] = {
-      ST_SURFACE_FRONT_LEFT,
-      ST_SURFACE_BACK_LEFT,
-      ST_SURFACE_FRONT_RIGHT,
-      ST_SURFACE_BACK_RIGHT,
-   };
-   EGLint s, i;
+   EGLint s;
 
    /* route draw and read buffers' attachments */
    for (s = 0; s < 2; s++) {
@@ -150,11 +152,7 @@ egl_g3d_route_context(_EGLDisplay *dpy, _EGLContext *ctx)
          gbuf = &gctx->read;
       }
 
-      gbuf->native_atts[0] = gsurf->render_att;
-      gbuf->num_atts = 1;
-
-      for (i = 0; i < gbuf->num_atts; i++)
-         gbuf->st_atts[i] = st_att_map[gbuf->native_atts[i]];
+      gbuf->attachment_mask = (1 << gsurf->render_att);
 
       /* FIXME OpenGL defaults to draw the front or back buffer when the
        * context is single-buffered or double-buffered respectively.  In EGL,
@@ -196,19 +194,19 @@ egl_g3d_realloc_context(_EGLDisplay *dpy, _EGLContext *ctx)
       if (!gdraw || priv != (void *) &gdraw->base) {
          gctx->stapi->st_unreference_framebuffer(gctx->draw.st_fb);
          gctx->draw.st_fb = NULL;
-         gctx->draw.num_atts = 0;
+         gctx->draw.attachment_mask = 0x0;
       }
 
       if (is_equal) {
          gctx->read.st_fb = NULL;
-         gctx->draw.num_atts = 0;
+         gctx->draw.attachment_mask = 0x0;
       }
       else {
          priv = gctx->stapi->st_framebuffer_private(gctx->read.st_fb);
          if (!gread || priv != (void *) &gread->base) {
             gctx->stapi->st_unreference_framebuffer(gctx->read.st_fb);
             gctx->read.st_fb = NULL;
-            gctx->draw.num_atts = 0;
+            gctx->draw.attachment_mask = 0x0;
          }
       }
    }
@@ -246,17 +244,6 @@ egl_g3d_realloc_context(_EGLDisplay *dpy, _EGLContext *ctx)
 }
 
 /**
- * Return the current context of the given API.
- */
-static struct egl_g3d_context *
-egl_g3d_get_current_context(EGLint api)
-{
-   _EGLThreadInfo *t = _eglGetCurrentThread();
-   EGLint api_index = _eglConvertApiToIndex(api);
-   return egl_g3d_context(t->CurrentContexts[api_index]);
-}
-
-/**
  * Return the state tracker for the given context.
  */
 static const struct egl_g3d_st *
@@ -297,6 +284,71 @@ egl_g3d_choose_st(_EGLDriver *drv, _EGLContext *ctx)
 }
 
 /**
+ * Initialize the state trackers.
+ */
+static void
+egl_g3d_init_st(_EGLDriver *drv)
+{
+   struct egl_g3d_driver *gdrv = egl_g3d_driver(drv);
+   EGLint i;
+
+   /* already initialized */
+   if (gdrv->api_mask)
+      return;
+
+   for (i = 0; i < NUM_EGL_G3D_STS; i++) {
+      gdrv->stapis[i] = egl_g3d_get_st(i);
+      if (gdrv->stapis[i])
+         gdrv->api_mask |= gdrv->stapis[i]->api_bit;
+   }
+
+   if (gdrv->api_mask)
+      _eglLog(_EGL_DEBUG, "Driver API mask: 0x%x", gdrv->api_mask);
+   else
+      _eglLog(_EGL_WARNING, "No supported client API");
+}
+
+/**
+ * Get the probe object of the display.
+ *
+ * Note that this function may be called before the display is initialized.
+ */
+static struct native_probe *
+egl_g3d_get_probe(_EGLDriver *drv, _EGLDisplay *dpy)
+{
+   struct egl_g3d_driver *gdrv = egl_g3d_driver(drv);
+   struct native_probe *nprobe;
+
+   nprobe = (struct native_probe *) _eglGetProbeCache(gdrv->probe_key);
+   if (!nprobe || nprobe->display != dpy->NativeDisplay) {
+      if (nprobe)
+         nprobe->destroy(nprobe);
+      nprobe = native_create_probe(dpy->NativeDisplay);
+      _eglSetProbeCache(gdrv->probe_key, (void *) nprobe);
+   }
+
+   return nprobe;
+}
+
+/**
+ * Destroy the probe object of the display.  The display may be NULL.
+ *
+ * Note that this function may be called before the display is initialized.
+ */
+static void
+egl_g3d_destroy_probe(_EGLDriver *drv, _EGLDisplay *dpy)
+{
+   struct egl_g3d_driver *gdrv = egl_g3d_driver(drv);
+   struct native_probe *nprobe;
+
+   nprobe = (struct native_probe *) _eglGetProbeCache(gdrv->probe_key);
+   if (nprobe && (!dpy || nprobe->display == dpy->NativeDisplay)) {
+      nprobe->destroy(nprobe);
+      _eglSetProbeCache(gdrv->probe_key, NULL);
+   }
+}
+
+/**
  * Return an API mask that consists of the state trackers that supports the
  * given mode.
  *
@@ -412,18 +464,18 @@ egl_g3d_add_configs(_EGLDriver *drv, _EGLDisplay *dpy, EGLint id)
       struct egl_g3d_config *gconf;
       EGLBoolean valid;
 
+      gconf = CALLOC_STRUCT(egl_g3d_config);
+      if (!gconf)
+         continue;
+
+      _eglInitConfig(&gconf->base, id);
+
       api_mask = get_mode_api_mask(&native_configs[i]->mode, gdrv->api_mask);
       if (!api_mask) {
          _eglLog(_EGL_DEBUG, "no state tracker supports config 0x%x",
                native_configs[i]->mode.visualID);
-         continue;
       }
 
-      gconf = CALLOC_STRUCT(egl_g3d_config);
-      if (!gconf)
-         continue;
-
-      _eglInitConfig(&gconf->base, id);
       valid = _eglConfigFromContextModesRec(&gconf->base,
             &native_configs[i]->mode, api_mask, api_mask);
       if (valid) {
@@ -480,7 +532,7 @@ egl_g3d_update_buffer(struct pipe_screen *screen, void *context_private)
     * Set force_validate to skip an unnecessary check.
     */
    gctx->force_validate = EGL_TRUE;
-   egl_g3d_validate_context(gctx->base.Display, &gctx->base);
+   egl_g3d_validate_context(gctx->base.Resource.Display, &gctx->base);
 }
 
 static EGLBoolean
@@ -517,6 +569,9 @@ egl_g3d_initialize(_EGLDriver *drv, _EGLDisplay *dpy,
    struct egl_g3d_driver *gdrv = egl_g3d_driver(drv);
    struct egl_g3d_display *gdpy;
 
+   /* the probe object is unlikely to be needed again */
+   egl_g3d_destroy_probe(drv, dpy);
+
    gdpy = CALLOC_STRUCT(egl_g3d_display);
    if (!gdpy) {
       _eglError(EGL_BAD_ALLOC, "eglInitialize");
@@ -533,6 +588,7 @@ egl_g3d_initialize(_EGLDriver *drv, _EGLDisplay *dpy,
    gdpy->native->screen->flush_frontbuffer = egl_g3d_flush_frontbuffer;
    gdpy->native->screen->update_buffer = egl_g3d_update_buffer;
 
+   egl_g3d_init_st(&gdrv->base);
    dpy->ClientAPIsMask = gdrv->api_mask;
 
    if (egl_g3d_add_configs(drv, dpy, 1) == 1) {
@@ -605,21 +661,30 @@ egl_g3d_create_context(_EGLDriver *drv, _EGLDisplay *dpy, _EGLConfig *conf,
    return &gctx->base;
 }
 
-static EGLBoolean
-egl_g3d_destroy_context(_EGLDriver *drv, _EGLDisplay *dpy, _EGLContext *ctx)
+/**
+ * Destroy a context.
+ */
+static void
+destroy_context(_EGLDisplay *dpy, _EGLContext *ctx)
 {
    struct egl_g3d_context *gctx = egl_g3d_context(ctx);
 
-   if (_eglIsContextBound(&gctx->base))
-      return EGL_TRUE;
+   /* FIXME a context might live longer than its display */
+   if (!dpy->Initialized)
+      _eglLog(_EGL_FATAL, "destroy a context with an unitialized display");
 
    egl_g3d_realloc_context(dpy, &gctx->base);
-
-   /* it will destroy pipe context */
+   /* it will destroy the associated pipe context */
    gctx->stapi->st_destroy_context(gctx->st_ctx);
 
    free(gctx);
+}
 
+static EGLBoolean
+egl_g3d_destroy_context(_EGLDriver *drv, _EGLDisplay *dpy, _EGLContext *ctx)
+{
+   if (!_eglIsContextBound(ctx))
+      destroy_context(dpy, ctx);
    return EGL_TRUE;
 }
 
@@ -628,7 +693,7 @@ init_surface_geometry(_EGLSurface *surf)
 {
    struct egl_g3d_surface *gsurf = egl_g3d_surface(surf);
 
-   return gsurf->native->validate(gsurf->native, NULL, 0,
+   return gsurf->native->validate(gsurf->native, 0x0,
          &gsurf->sequence_number, NULL,
          &gsurf->base.Width, &gsurf->base.Height);
 }
@@ -750,17 +815,28 @@ egl_g3d_create_pbuffer_surface(_EGLDriver *drv, _EGLDisplay *dpy,
    return &gsurf->base;
 }
 
-static EGLBoolean
-egl_g3d_destroy_surface(_EGLDriver *drv, _EGLDisplay *dpy, _EGLSurface *surf)
+/**
+ * Destroy a surface.
+ */
+static void
+destroy_surface(_EGLDisplay *dpy, _EGLSurface *surf)
 {
    struct egl_g3d_surface *gsurf = egl_g3d_surface(surf);
 
-   if (_eglIsSurfaceBound(&gsurf->base))
-         return EGL_TRUE;
+   /* FIXME a surface might live longer than its display */
+   if (!dpy->Initialized)
+      _eglLog(_EGL_FATAL, "destroy a surface with an unitialized display");
 
    pipe_surface_reference(&gsurf->render_surface, NULL);
    gsurf->native->destroy(gsurf->native);
    free(gsurf);
+}
+
+static EGLBoolean
+egl_g3d_destroy_surface(_EGLDriver *drv, _EGLDisplay *dpy, _EGLSurface *surf)
+{
+   if (!_eglIsSurfaceBound(surf))
+      destroy_surface(dpy, surf);
    return EGL_TRUE;
 }
 
@@ -769,18 +845,14 @@ egl_g3d_make_current(_EGLDriver *drv, _EGLDisplay *dpy,
                      _EGLSurface *draw, _EGLSurface *read, _EGLContext *ctx)
 {
    struct egl_g3d_context *gctx = egl_g3d_context(ctx);
+   struct egl_g3d_surface *gdraw = egl_g3d_surface(draw);
    struct egl_g3d_context *old_gctx;
-   EGLint api;
    EGLBoolean ok = EGL_TRUE;
 
-   /* find the old context */
-   api = (gctx) ? gctx->base.ClientAPI : eglQueryAPI();
-   old_gctx = egl_g3d_get_current_context(api);
-   if (old_gctx && !_eglIsContextLinked(&old_gctx->base))
-         old_gctx = NULL;
-
-   if (!_eglMakeCurrent(drv, dpy, draw, read, ctx))
+   /* bind the new context and return the "orphaned" one */
+   if (!_eglBindContext(&ctx, &draw, &read))
       return EGL_FALSE;
+   old_gctx = egl_g3d_context(ctx);
 
    if (old_gctx) {
       /* flush old context */
@@ -796,8 +868,6 @@ egl_g3d_make_current(_EGLDriver *drv, _EGLDisplay *dpy,
    }
 
    if (gctx) {
-      struct egl_g3d_surface *gdraw = egl_g3d_surface(draw);
-
       ok = egl_g3d_realloc_context(dpy, &gctx->base);
       if (ok) {
          ok = gctx->stapi->st_make_current(gctx->st_ctx,
@@ -817,6 +887,13 @@ egl_g3d_make_current(_EGLDriver *drv, _EGLDisplay *dpy,
       old_gctx->base.WindowRenderBuffer = EGL_NONE;
    }
 
+   if (ctx && !_eglIsContextLinked(ctx))
+      destroy_context(dpy, ctx);
+   if (draw && !_eglIsSurfaceLinked(draw))
+      destroy_surface(dpy, draw);
+   if (read && read != draw && !_eglIsSurfaceLinked(read))
+      destroy_surface(dpy, read);
+
    return ok;
 }
 
@@ -864,6 +941,101 @@ egl_g3d_swap_buffers(_EGLDriver *drv, _EGLDisplay *dpy, _EGLSurface *surf)
    return EGL_TRUE;
 }
 
+/**
+ * Find a config that supports the pixmap.
+ */
+static _EGLConfig *
+find_pixmap_config(_EGLDisplay *dpy, EGLNativePixmapType pix)
+{
+   struct egl_g3d_display *gdpy = egl_g3d_display(dpy);
+   struct egl_g3d_config *gconf;
+   EGLint i;
+
+   for (i = 0; i < dpy->NumConfigs; i++) {
+      gconf = egl_g3d_config(dpy->Configs[i]);
+      if (gdpy->native->is_pixmap_supported(gdpy->native, pix, gconf->native))
+         break;
+   }
+
+   return (i < dpy->NumConfigs) ? &gconf->base : NULL;
+}
+
+/**
+ * Get the pipe surface of the given attachment of the native surface.
+ */
+static struct pipe_surface *
+get_pipe_surface(struct native_display *ndpy, struct native_surface *nsurf,
+                 enum native_attachment natt)
+{
+   struct pipe_texture *textures[NUM_NATIVE_ATTACHMENTS];
+   struct pipe_surface *psurf;
+
+   textures[natt] = NULL;
+   nsurf->validate(nsurf, 1 << natt, NULL, textures, NULL, NULL);
+   if (!textures[natt])
+      return NULL;
+
+   psurf = ndpy->screen->get_tex_surface(ndpy->screen, textures[natt],
+         0, 0, 0, PIPE_BUFFER_USAGE_CPU_WRITE);
+   pipe_texture_reference(&textures[natt], NULL);
+
+   return psurf;
+}
+
+static EGLBoolean
+egl_g3d_copy_buffers(_EGLDriver *drv, _EGLDisplay *dpy, _EGLSurface *surf,
+                     EGLNativePixmapType target)
+{
+   struct egl_g3d_display *gdpy = egl_g3d_display(dpy);
+   struct egl_g3d_surface *gsurf = egl_g3d_surface(surf);
+   _EGLContext *ctx = _eglGetCurrentContext();
+   struct egl_g3d_config *gconf;
+   struct native_surface *nsurf;
+   struct pipe_screen *screen = gdpy->native->screen;
+   struct pipe_surface *psurf;
+
+   if (!gsurf->render_surface)
+      return EGL_TRUE;
+
+   gconf = egl_g3d_config(find_pixmap_config(dpy, target));
+   if (!gconf)
+      return _eglError(EGL_BAD_NATIVE_PIXMAP, "eglCopyBuffers");
+
+   nsurf = gdpy->native->create_pixmap_surface(gdpy->native,
+         target, gconf->native);
+   if (!nsurf)
+      return _eglError(EGL_BAD_NATIVE_PIXMAP, "eglCopyBuffers");
+
+   /* flush if the surface is current */
+   if (ctx && ctx->DrawSurface == &gsurf->base) {
+      struct egl_g3d_context *gctx = egl_g3d_context(ctx);
+      gctx->stapi->st_flush(gctx->st_ctx,
+            PIPE_FLUSH_RENDER_CACHE | PIPE_FLUSH_FRAME, NULL);
+   }
+
+   psurf = get_pipe_surface(gdpy->native, nsurf, NATIVE_ATTACHMENT_FRONT_LEFT);
+   if (psurf) {
+      struct pipe_context pipe;
+
+      /**
+       * XXX This is hacky.  If we might allow the EGLDisplay to create a pipe
+       * context of its own and use the blitter context for this.
+       */
+      memset(&pipe, 0, sizeof(pipe));
+      pipe.screen = screen;
+
+      util_surface_copy(&pipe, FALSE, psurf, 0, 0,
+            gsurf->render_surface, 0, 0, psurf->width, psurf->height);
+
+      pipe_surface_reference(&psurf, NULL);
+      nsurf->flush_frontbuffer(nsurf);
+   }
+
+   nsurf->destroy(nsurf);
+
+   return EGL_TRUE;
+}
+
 static EGLBoolean
 egl_g3d_wait_client(_EGLDriver *drv, _EGLDisplay *dpy, _EGLContext *ctx)
 {
@@ -875,14 +1047,15 @@ egl_g3d_wait_client(_EGLDriver *drv, _EGLDisplay *dpy, _EGLContext *ctx)
 static EGLBoolean
 egl_g3d_wait_native(_EGLDriver *drv, _EGLDisplay *dpy, EGLint engine)
 {
-   _EGLSurface *surf = _eglGetCurrentSurface(EGL_DRAW);
-   struct egl_g3d_surface *gsurf = egl_g3d_surface(surf);
+   _EGLContext *ctx = _eglGetCurrentContext();
 
    if (engine != EGL_CORE_NATIVE_ENGINE)
       return _eglError(EGL_BAD_PARAMETER, "eglWaitNative");
 
-   if (gsurf)
+   if (ctx && ctx->DrawSurface) {
+      struct egl_g3d_surface *gsurf = egl_g3d_surface(ctx->DrawSurface);
       gsurf->native->wait(gsurf->native);
+   }
 
    return EGL_TRUE;
 }
@@ -894,6 +1067,9 @@ egl_g3d_get_proc_address(_EGLDriver *drv, const char *procname)
    _EGLProc proc;
    EGLint i;
 
+   /* in case this is called before a display is initialized */
+   egl_g3d_init_st(&gdrv->base);
+
    for (i = 0; i < NUM_EGL_G3D_STS; i++) {
       const struct egl_g3d_st *stapi = gdrv->stapis[i];
       if (stapi) {
@@ -911,7 +1087,8 @@ egl_g3d_bind_tex_image(_EGLDriver *drv, _EGLDisplay *dpy,
                        _EGLSurface *surf, EGLint buffer)
 {
    struct egl_g3d_surface *gsurf = egl_g3d_surface(surf);
-   struct egl_g3d_context *gctx;
+   _EGLContext *ctx = _eglGetAPIContext(EGL_OPENGL_ES_API);
+   struct egl_g3d_context *gctx = egl_g3d_context(ctx);
    enum pipe_format target_format;
    int target;
 
@@ -942,14 +1119,12 @@ egl_g3d_bind_tex_image(_EGLDriver *drv, _EGLDisplay *dpy,
    }
 
    /* flush properly if the surface is bound */
-   if (gsurf->base.Binding) {
-      gctx = egl_g3d_context(gsurf->base.Binding);
+   if (gsurf->base.CurrentContext) {
+      gctx = egl_g3d_context(gsurf->base.CurrentContext);
       gctx->stapi->st_flush(gctx->st_ctx,
             PIPE_FLUSH_RENDER_CACHE | PIPE_FLUSH_FRAME, NULL);
    }
 
-   /* XXX change to EGL_OPENGL_ES_API once OpenGL ES is merged */
-   gctx = egl_g3d_get_current_context(EGL_OPENGL_API);
    if (gctx) {
       if (!gsurf->render_surface)
          return EGL_FALSE;
@@ -975,10 +1150,8 @@ egl_g3d_release_tex_image(_EGLDriver *drv, _EGLDisplay *dpy,
       return _eglError(EGL_BAD_PARAMETER, "eglReleaseTexImage");
 
    if (gsurf->render_surface) {
-      _EGLThreadInfo *t = _eglGetCurrentThread();
-      /* XXX change to EGL_OPENGL_ES_API once OpenGL ES is merged */
-      struct egl_g3d_context *gctx = egl_g3d_context(
-            t->CurrentContexts[_eglConvertApiToIndex(EGL_OPENGL_API)]);
+      _EGLContext *ctx = _eglGetAPIContext(EGL_OPENGL_ES_API);
+      struct egl_g3d_context *gctx = egl_g3d_context(ctx);
 
       /* what if the context the surface binds to is no longer current? */
       if (gctx)
@@ -1084,10 +1257,41 @@ egl_g3d_show_screen_surface(_EGLDriver *drv, _EGLDisplay *dpy,
 
 #endif /* EGL_MESA_screen_surface */
 
+static EGLint
+egl_g3d_probe(_EGLDriver *drv, _EGLDisplay *dpy)
+{
+   struct native_probe *nprobe;
+   enum native_probe_result res;
+   EGLint score;
+
+   nprobe = egl_g3d_get_probe(drv, dpy);
+   res = native_get_probe_result(nprobe);
+
+   switch (res) {
+   case NATIVE_PROBE_UNKNOWN:
+   default:
+      score = 0;
+      break;
+   case NATIVE_PROBE_FALLBACK:
+      score = 40;
+      break;
+   case NATIVE_PROBE_SUPPORTED:
+      score = 50;
+      break;
+   case NATIVE_PROBE_EXACT:
+      score = 100;
+      break;
+   }
+
+   return score;
+}
+
 static void
 egl_g3d_unload(_EGLDriver *drv)
 {
    struct egl_g3d_driver *gdrv = egl_g3d_driver(drv);
+
+   egl_g3d_destroy_probe(drv, NULL);
    free(gdrv);
 }
 
@@ -1096,7 +1300,6 @@ _eglMain(const char *args)
 {
    static char driver_name[64];
    struct egl_g3d_driver *gdrv;
-   EGLint i;
 
    snprintf(driver_name, sizeof(driver_name),
          "Gallium/%s", native_get_name());
@@ -1117,6 +1320,7 @@ _eglMain(const char *args)
    gdrv->base.API.DestroySurface = egl_g3d_destroy_surface;
    gdrv->base.API.MakeCurrent = egl_g3d_make_current;
    gdrv->base.API.SwapBuffers = egl_g3d_swap_buffers;
+   gdrv->base.API.CopyBuffers = egl_g3d_copy_buffers;
    gdrv->base.API.WaitClient = egl_g3d_wait_client;
    gdrv->base.API.WaitNative = egl_g3d_wait_native;
    gdrv->base.API.GetProcAddress = egl_g3d_get_proc_address;
@@ -1130,18 +1334,11 @@ _eglMain(const char *args)
 #endif
 
    gdrv->base.Name = driver_name;
+   gdrv->base.Probe = egl_g3d_probe;
    gdrv->base.Unload = egl_g3d_unload;
 
-   for (i = 0; i < NUM_EGL_G3D_STS; i++) {
-      gdrv->stapis[i] = egl_g3d_get_st(i);
-      if (gdrv->stapis[i])
-         gdrv->api_mask |= gdrv->stapis[i]->api_bit;
-   }
-
-   if (gdrv->api_mask)
-      _eglLog(_EGL_DEBUG, "Driver API mask: 0x%x", gdrv->api_mask);
-   else
-      _eglLog(_EGL_WARNING, "No supported client API");
+   /* the key is " EGL G3D" */
+   gdrv->probe_key = 0x0E61063D;
 
    return &gdrv->base;
 }
diff --git a/src/gallium/state_trackers/egl_g3d/common/egl_g3d.h b/src/gallium/state_trackers/egl/common/egl_g3d.h
index 4c8b8dfe9e..3dae8c4052 100644
--- a/src/gallium/state_trackers/egl_g3d/common/egl_g3d.h
+++ b/src/gallium/state_trackers/egl/common/egl_g3d.h
@@ -44,6 +44,8 @@ struct egl_g3d_driver {
    _EGLDriver base;
    const struct egl_g3d_st *stapis[NUM_EGL_G3D_STS];
    EGLint api_mask;
+
+   EGLint probe_key;
 };
 
 struct egl_g3d_display {
@@ -52,9 +54,7 @@ struct egl_g3d_display {
 
 struct egl_g3d_buffer {
    struct st_framebuffer *st_fb;
-   EGLint num_atts;
-   enum native_attachment native_atts[NUM_NATIVE_ATTACHMENTS];
-   uint st_atts[NUM_NATIVE_ATTACHMENTS];
+   uint attachment_mask;
 };
 
 struct egl_g3d_context {
diff --git a/src/gallium/state_trackers/egl_g3d/common/egl_st.c b/src/gallium/state_trackers/egl/common/egl_st.c
index a88ff911cd..a88ff911cd 100644
--- a/src/gallium/state_trackers/egl_g3d/common/egl_st.c
+++ b/src/gallium/state_trackers/egl/common/egl_st.c
diff --git a/src/gallium/state_trackers/egl_g3d/common/egl_st.h b/src/gallium/state_trackers/egl/common/egl_st.h
index 8fb464bd3d..8fb464bd3d 100644
--- a/src/gallium/state_trackers/egl_g3d/common/egl_st.h
+++ b/src/gallium/state_trackers/egl/common/egl_st.h
diff --git a/src/gallium/state_trackers/egl_g3d/common/native.h b/src/gallium/state_trackers/egl/common/native.h
index 76f0e0c78a..72a9cec7ef 100644
--- a/src/gallium/state_trackers/egl_g3d/common/native.h
+++ b/src/gallium/state_trackers/egl/common/native.h
@@ -47,6 +47,27 @@ enum native_attachment {
    NUM_NATIVE_ATTACHMENTS
 };
 
+/**
+ * Enumerations for probe results.
+ */
+enum native_probe_result {
+   NATIVE_PROBE_UNKNOWN,
+   NATIVE_PROBE_FALLBACK,
+   NATIVE_PROBE_SUPPORTED,
+   NATIVE_PROBE_EXACT,
+};
+
+/**
+ * A probe object for display probe.
+ */
+struct native_probe {
+   int magic;
+   EGLNativeDisplayType display;
+   void *data;
+
+   void (*destroy)(struct native_probe *nprobe);
+};
+
 struct native_surface {
    void (*destroy)(struct native_surface *nsurf);
 
@@ -64,18 +85,18 @@ struct native_surface {
    boolean (*flush_frontbuffer)(struct native_surface *nsurf);
 
    /**
-    * Validate the buffers of the surface.  The returned textures are owned by
-    * the caller.  A sequence number is also returned.  The caller can use it
-    * to check if anything has changed since the last call. Any of the pointers
-    * may be NULL and it indicates the caller has no interest in those values.
+    * Validate the buffers of the surface.  textures, if not NULL, points to an
+    * array of size NUM_NATIVE_ATTACHMENTS and the returned textures are owned
+    * by the caller.  A sequence number is also returned.  The caller can use
+    * it to check if anything has changed since the last call. Any of the
+    * pointers may be NULL and it indicates the caller has no interest in those
+    * values.
     *
-    * If this function is called multiple times with different attachments,
-    * those not listed in the latest call might be destroyed.  This behavior
-    * might change in the future.
+    * If this function is called multiple times with different attachment
+    * masks, those not listed in the latest call might be destroyed.  This
+    * behavior might change in the future.
     */
-   boolean (*validate)(struct native_surface *nsurf,
-                       const enum native_attachment *natts,
-                       unsigned num_natts,
+   boolean (*validate)(struct native_surface *nsurf, uint attachment_mask,
                        unsigned int *seq_num, struct pipe_texture **textures,
                        int *width, int *height);
 
@@ -136,6 +157,17 @@ struct native_display {
                                                int *num_configs);
 
    /**
+    * Test if a pixmap is supported by the given config.  Required unless no
+    * config has GLX_PIXMAP_BIT set.
+    *
+    * This function is usually called to find a config that supports a given
+    * pixmap.  Thus, it is usually called with the same pixmap in a row.
+    */
+   boolean (*is_pixmap_supported)(struct native_display *ndpy,
+                                  EGLNativePixmapType pix,
+                                  const struct native_config *nconf);
+
+   /**
     * Create a pipe context.
     */
    struct pipe_context *(*create_context)(struct native_display *ndpy,
@@ -211,6 +243,31 @@ struct native_display_modeset {
                       const struct native_mode *nmode);
 };
 
+/**
+ * Test whether an attachment is set in the mask.
+ */
+static INLINE boolean
+native_attachment_mask_test(uint mask, enum native_attachment att)
+{
+   return !!(mask & (1 << att));
+}
+
+/**
+ * Return a probe object for the given display.
+ *
+ * Note that the returned object may be cached and used by different native
+ * display modules.  It allows fast probing when multiple modules probe the
+ * same display.
+ */
+struct native_probe *
+native_create_probe(EGLNativeDisplayType dpy);
+
+/**
+ * Probe the probe object.
+ */
+enum native_probe_result
+native_get_probe_result(struct native_probe *nprobe);
+
 const char *
 native_get_name(void);
 
diff --git a/src/gallium/state_trackers/egl_g3d/common/st_public_tmp.h b/src/gallium/state_trackers/egl/common/st_public_tmp.h
index 507a0ec402..507a0ec402 100644
--- a/src/gallium/state_trackers/egl_g3d/common/st_public_tmp.h
+++ b/src/gallium/state_trackers/egl/common/st_public_tmp.h
diff --git a/src/gallium/state_trackers/egl/egl_context.c b/src/gallium/state_trackers/egl/egl_context.c
deleted file mode 100644
index fee186c601..0000000000
--- a/src/gallium/state_trackers/egl/egl_context.c
+++ /dev/null
@@ -1,105 +0,0 @@
-
-#include "utils.h"
-#include <stdlib.h>
-#include <stdio.h>
-#include <string.h>
-#include "egl_tracker.h"
-
-#include "egllog.h"
-
-
-#include "pipe/p_context.h"
-#include "pipe/p_screen.h"
-
-#include "state_tracker/st_public.h"
-#include "state_tracker/drm_api.h"
-
-#include "GL/internal/glcore.h"
-
-_EGLContext *
-drm_create_context(_EGLDriver *drv, _EGLDisplay *dpy, _EGLConfig *conf, _EGLContext *share_list, const EGLint *attrib_list)
-{
-	struct drm_device *dev = lookup_drm_device(dpy);
-	struct drm_context *ctx;
-	struct drm_context *share = NULL;
-	struct st_context *st_share = NULL;
-	int i;
-	__GLcontextModes *visual;
-
-	for (i = 0; attrib_list && attrib_list[i] != EGL_NONE; i++) {
-		switch (attrib_list[i]) {
-			/* no attribs defined for now */
-			default:
-				_eglError(EGL_BAD_ATTRIBUTE, "eglCreateContext");
-				return EGL_NO_CONTEXT;
-		}
-	}
-
-	ctx = (struct drm_context *) calloc(1, sizeof(struct drm_context));
-	if (!ctx)
-		goto err_c;
-
-	_eglInitContext(drv, &ctx->base, conf, attrib_list);
-
-	ctx->pipe = dev->api->create_context(dev->api, dev->screen);
-	if (!ctx->pipe)
-		goto err_pipe;
-
-	if (share)
-		st_share = share->st;
-
-	visual = drm_visual_from_config(conf);
-	ctx->st = st_create_context(ctx->pipe, visual, st_share);
-	drm_visual_modes_destroy(visual);
-
-	if (!ctx->st)
-		goto err_gl;
-
-	return &ctx->base;
-
-err_gl:
-	ctx->pipe->destroy(ctx->pipe);
-err_pipe:
-	free(ctx);
-err_c:
-	return NULL;
-}
-
-EGLBoolean
-drm_destroy_context(_EGLDriver *drv, _EGLDisplay *dpy, _EGLContext *context)
-{
-	struct drm_context *c = lookup_drm_context(context);
-	if (!_eglIsContextBound(&c->base)) {
-		st_destroy_context(c->st);
-		free(c);
-	}
-	return EGL_TRUE;
-}
-
-EGLBoolean
-drm_make_current(_EGLDriver *drv, _EGLDisplay *dpy, _EGLSurface *draw, _EGLSurface *read, _EGLContext *context)
-{
-	struct drm_surface *readSurf = lookup_drm_surface(read);
-	struct drm_surface *drawSurf = lookup_drm_surface(draw);
-	struct drm_context *ctx = lookup_drm_context(context);
-	EGLBoolean b;
-
-	b = _eglMakeCurrent(drv, dpy, draw, read, context);
-	if (!b)
-		return EGL_FALSE;
-
-	if (ctx) {
-		if (!drawSurf || !readSurf)
-			return EGL_FALSE;
-
-		st_make_current(ctx->st, drawSurf->stfb, readSurf->stfb);
-
-		/* st_resize_framebuffer needs a bound context to work */
-		st_resize_framebuffer(drawSurf->stfb, drawSurf->w, drawSurf->h);
-		st_resize_framebuffer(readSurf->stfb, readSurf->w, readSurf->h);
-	} else {
-		st_make_current(NULL, NULL, NULL);
-	}
-
-	return EGL_TRUE;
-}
diff --git a/src/gallium/state_trackers/egl/egl_surface.c b/src/gallium/state_trackers/egl/egl_surface.c
deleted file mode 100644
index d55aa51b82..0000000000
--- a/src/gallium/state_trackers/egl/egl_surface.c
+++ /dev/null
@@ -1,443 +0,0 @@
-
-#include <stdlib.h>
-#include <stdio.h>
-#include <string.h>
-#include "egl_tracker.h"
-
-#include "egllog.h"
-
-#include "pipe/p_inlines.h"
-#include "pipe/p_screen.h"
-#include "pipe/p_context.h"
-
-#include "state_tracker/drm_api.h"
-
-#include "util/u_format.h"
-#include "util/u_rect.h"
-
-/*
- * Util functions
- */
-
-static drmModeModeInfoPtr
-drm_find_mode(drmModeConnectorPtr connector, _EGLMode *mode)
-{
-	int i;
-	drmModeModeInfoPtr m = NULL;
-
-	for (i = 0; i < connector->count_modes; i++) {
-		m = &connector->modes[i];
-		if (m->hdisplay == mode->Width && m->vdisplay == mode->Height && m->vrefresh == mode->RefreshRate)
-			break;
-		m = &connector->modes[0]; /* if we can't find one, return first */
-	}
-
-	return m;
-}
-
-static struct st_framebuffer *
-drm_create_framebuffer(struct pipe_screen *screen,
-                       const __GLcontextModes *visual,
-                       unsigned width,
-                       unsigned height,
-                       void *priv)
-{
-	enum pipe_format color_format, depth_stencil_format;
-	boolean d_depth_bits_last;
-	boolean ds_depth_bits_last;
-
-	d_depth_bits_last =
-		screen->is_format_supported(screen, PIPE_FORMAT_X8Z24_UNORM,
-		                            PIPE_TEXTURE_2D,
-		                            PIPE_TEXTURE_USAGE_DEPTH_STENCIL, 0);
-	ds_depth_bits_last =
-		screen->is_format_supported(screen, PIPE_FORMAT_S8Z24_UNORM,
-		                            PIPE_TEXTURE_2D,
-		                            PIPE_TEXTURE_USAGE_DEPTH_STENCIL, 0);
-
-	if (visual->redBits == 8) {
-		if (visual->alphaBits == 8)
-			color_format = PIPE_FORMAT_A8R8G8B8_UNORM;
-		else
-			color_format = PIPE_FORMAT_X8R8G8B8_UNORM;
-	} else {
-		color_format = PIPE_FORMAT_R5G6B5_UNORM;
-	}
-
-	switch(visual->depthBits) {
-		default:
-		case 0:
-			depth_stencil_format = PIPE_FORMAT_NONE;
-			break;
-		case 16:
-			depth_stencil_format = PIPE_FORMAT_Z16_UNORM;
-			break;
-		case 24:
-			if (visual->stencilBits == 0) {
-				depth_stencil_format = (d_depth_bits_last) ?
-					PIPE_FORMAT_X8Z24_UNORM:
-					PIPE_FORMAT_Z24X8_UNORM;
-			} else {
-				depth_stencil_format = (ds_depth_bits_last) ?
-					PIPE_FORMAT_S8Z24_UNORM:
-					PIPE_FORMAT_Z24S8_UNORM;
-			}
-			break;
-		case 32:
-			depth_stencil_format = PIPE_FORMAT_Z32_UNORM;
-			break;
-	}
-
-	return st_create_framebuffer(visual,
-	                             color_format,
-	                             depth_stencil_format,
-	                             depth_stencil_format,
-	                             width,
-	                             height,
-	                             priv);
-}
-
-static void
-drm_create_texture(_EGLDisplay *dpy,
-                   struct drm_screen *scrn,
-                   unsigned w, unsigned h)
-{
-	struct drm_device *dev = lookup_drm_device(dpy);
-	struct pipe_screen *screen = dev->screen;
-	struct pipe_surface *surface;
-	struct pipe_texture *texture;
-	struct pipe_texture templat;
-	struct pipe_buffer *buf = NULL;
-	unsigned pitch = 0;
-
-	memset(&templat, 0, sizeof(templat));
-	templat.tex_usage = PIPE_TEXTURE_USAGE_RENDER_TARGET;
-	templat.tex_usage |= PIPE_TEXTURE_USAGE_PRIMARY;
-	templat.target = PIPE_TEXTURE_2D;
-	templat.last_level = 0;
-	templat.depth0 = 1;
-	templat.format = PIPE_FORMAT_A8R8G8B8_UNORM;
-	templat.width0 = w;
-	templat.height0 = h;
-
-	texture = screen->texture_create(dev->screen,
-	                                 &templat);
-
-	if (!texture)
-		goto err_tex;
-
-	surface = screen->get_tex_surface(screen,
-	                                  texture,
-	                                  0,
-	                                  0,
-	                                  0,
-	                                  PIPE_BUFFER_USAGE_GPU_WRITE);
-
-	if (!surface)
-		goto err_surf;
-
-	scrn->tex = texture;
-	scrn->surface = surface;
-	scrn->front.width = w;
-	scrn->front.height = h;
-	scrn->front.pitch = pitch;
-	dev->api->local_handle_from_texture(dev->api, screen, texture,
-	                                    &scrn->front.pitch, &scrn->front.handle);
-	if (0)
-		goto err_handle;
-
-	return;
-
-err_handle:
-	pipe_surface_reference(&surface, NULL);
-err_surf:
-	pipe_texture_reference(&texture, NULL);
-err_tex:
-	pipe_buffer_reference(&buf, NULL);
-	return;
-}
-
-/*
- * Exported functions
- */
-
-void
-drm_takedown_shown_screen(_EGLDisplay *dpy, struct drm_screen *screen)
-{
-	struct drm_device *dev = lookup_drm_device(dpy);
-
-	screen->surf = NULL;
-
-	drmModeSetCrtc(
-		dev->drmFD,
-		screen->crtcID,
-		0, /* FD */
-		0, 0,
-		NULL, 0, /* List of output ids */
-		NULL);
-
-	drmModeRmFB(dev->drmFD, screen->fbID);
-	drmModeFreeFB(screen->fb);
-	screen->fb = NULL;
-
-	pipe_surface_reference(&screen->surface, NULL);
-	pipe_texture_reference(&screen->tex, NULL);
-
-	screen->shown = 0;
-}
-
-/**
- * Called by libEGL's eglCreateWindowSurface().
- */
-_EGLSurface *
-drm_create_window_surface(_EGLDriver *drv, _EGLDisplay *dpy, _EGLConfig *conf, NativeWindowType window, const EGLint *attrib_list)
-{
-	return NULL;
-}
-
-
-/**
- * Called by libEGL's eglCreatePixmapSurface().
- */
-_EGLSurface *
-drm_create_pixmap_surface(_EGLDriver *drv, _EGLDisplay *dpy, _EGLConfig *conf, NativePixmapType pixmap, const EGLint *attrib_list)
-{
-	return NULL;
-}
-
-
-/**
- * Called by libEGL's eglCreatePbufferSurface().
- */
-_EGLSurface *
-drm_create_pbuffer_surface(_EGLDriver *drv, _EGLDisplay *dpy, _EGLConfig *conf,
-                           const EGLint *attrib_list)
-{
-	struct drm_device *dev = lookup_drm_device(dpy);
-	int i;
-	int width = -1;
-	int height = -1;
-	struct drm_surface *surf = NULL;
-	__GLcontextModes *visual;
-
-	for (i = 0; attrib_list && attrib_list[i] != EGL_NONE; i++) {
-		switch (attrib_list[i]) {
-			case EGL_WIDTH:
-				width = attrib_list[++i];
-				break;
-			case EGL_HEIGHT:
-				height = attrib_list[++i];
-				break;
-			default:
-				_eglError(EGL_BAD_ATTRIBUTE, "eglCreatePbufferSurface");
-				return EGL_NO_SURFACE;
-		}
-	}
-
-	if (width < 1 || height < 1) {
-		_eglError(EGL_BAD_ATTRIBUTE, "eglCreatePbufferSurface");
-		return NULL;
-	}
-
-	surf = (struct drm_surface *) calloc(1, sizeof(struct drm_surface));
-	if (!surf)
-		goto err;
-
-	if (!_eglInitSurface(drv, &surf->base, EGL_PBUFFER_BIT, conf, attrib_list))
-		goto err_surf;
-
-	surf->w = width;
-	surf->h = height;
-
-	visual = drm_visual_from_config(conf);
-	surf->stfb = drm_create_framebuffer(dev->screen, visual,
-	                                    width, height,
-	                                    (void*)surf);
-	drm_visual_modes_destroy(visual);
-
-	return &surf->base;
-
-err_surf:
-	free(surf);
-err:
-	return NULL;
-}
-
-/**
- * Called by libEGL's eglCreateScreenSurfaceMESA().
- */
-_EGLSurface *
-drm_create_screen_surface_mesa(_EGLDriver *drv, _EGLDisplay *dpy, _EGLConfig *cfg,
-                               const EGLint *attrib_list)
-{
-	EGLSurface surf = drm_create_pbuffer_surface(drv, dpy, cfg, attrib_list);
-
-	return surf;
-}
-
-/**
- * Called by libEGL's eglShowScreenSurfaceMESA().
- */
-EGLBoolean
-drm_show_screen_surface_mesa(_EGLDriver *drv, _EGLDisplay *dpy,
-                             _EGLScreen *screen,
-                             _EGLSurface *surface, _EGLMode *mode)
-{
-	struct drm_device *dev = lookup_drm_device(dpy);
-	struct drm_surface *surf = lookup_drm_surface(surface);
-	struct drm_screen *scrn = lookup_drm_screen(screen);
-	int ret;
-	unsigned int i, k;
-
-	if (scrn->shown)
-		drm_takedown_shown_screen(dpy, scrn);
-
-
-	drm_create_texture(dpy, scrn, mode->Width, mode->Height);
-	if (!scrn->tex)
-		goto err_tex;
-
-	ret = drmModeAddFB(dev->drmFD,
-	                   scrn->front.width, scrn->front.height,
-	                   32, 32, scrn->front.pitch,
-	                   scrn->front.handle,
-	                   &scrn->fbID);
-
-	if (ret)
-		goto err_bo;
-
-	scrn->fb = drmModeGetFB(dev->drmFD, scrn->fbID);
-	if (!scrn->fb)
-		goto err_bo;
-
-	/* find a fitting crtc */
-	{
-		drmModeConnector *con = scrn->connector;
-
-		scrn->mode = drm_find_mode(con, mode);
-		if (!scrn->mode)
-			goto err_fb;
-
-		for (k = 0; k < con->count_encoders; k++) {
-			drmModeEncoder *enc = drmModeGetEncoder(dev->drmFD, con->encoders[k]);
-			for (i = 0; i < dev->res->count_crtcs; i++) {
-				if (enc->possible_crtcs & (1<<i)) {
-					/* save the ID */
-					scrn->crtcID = dev->res->crtcs[i];
-
-					/* skip the rest */
-					i = dev->res->count_crtcs;
-					k = dev->res->count_encoders;
-				}
-			}
-			drmModeFreeEncoder(enc);
-		}
-	}
-
-	ret = drmModeSetCrtc(dev->drmFD,
-	                     scrn->crtcID,
-	                     scrn->fbID,
-	                     0, 0,
-	                     &scrn->connectorID, 1,
-	                     scrn->mode);
-
-	if (ret)
-		goto err_crtc;
-
-
-	if (scrn->dpms)
-		drmModeConnectorSetProperty(dev->drmFD,
-		                            scrn->connectorID,
-		                            scrn->dpms->prop_id,
-		                            DRM_MODE_DPMS_ON);
-
-	surf->screen = scrn;
-
-	scrn->surf = surf;
-	scrn->shown = 1;
-
-	return EGL_TRUE;
-
-err_crtc:
-	scrn->crtcID = 0;
-
-err_fb:
-	drmModeRmFB(dev->drmFD, scrn->fbID);
-	drmModeFreeFB(scrn->fb);
-	scrn->fb = NULL;
-
-err_bo:
-	pipe_surface_reference(&scrn->surface, NULL);
-	pipe_texture_reference(&scrn->tex, NULL);
-
-err_tex:
-	return EGL_FALSE;
-}
-
-/**
- * Called by libEGL's eglDestroySurface().
- */
-EGLBoolean
-drm_destroy_surface(_EGLDriver *drv, _EGLDisplay *dpy, _EGLSurface *surface)
-{
-	struct drm_surface *surf = lookup_drm_surface(surface);
-	if (!_eglIsSurfaceBound(&surf->base)) {
-		if (surf->screen)
-			drm_takedown_shown_screen(dpy, surf->screen);
-		st_unreference_framebuffer(surf->stfb);
-		free(surf);
-	}
-	return EGL_TRUE;
-}
-
-/**
- * Called by libEGL's eglSwapBuffers().
- */
-EGLBoolean
-drm_swap_buffers(_EGLDriver *drv, _EGLDisplay *dpy, _EGLSurface *draw)
-{
-	struct drm_device *dev = lookup_drm_device(dpy);
-	struct drm_surface *surf = lookup_drm_surface(draw);
-	struct pipe_surface *back_surf;
-
-	if (!surf)
-		return EGL_FALSE;
-
-	st_get_framebuffer_surface(surf->stfb, ST_SURFACE_BACK_LEFT, &back_surf);
-
-	if (back_surf) {
-		struct drm_context *ctx = lookup_drm_context(draw->Binding);
-
-		st_notify_swapbuffers(surf->stfb);
-
-		if (ctx && surf->screen) {
-            if (ctx->pipe->surface_copy) {
-                ctx->pipe->surface_copy(ctx->pipe,
-                    surf->screen->surface,
-                    0, 0,
-                    back_surf,
-                    0, 0,
-                    surf->w, surf->h);
-            } else {
-                util_surface_copy(ctx->pipe, FALSE,
-                    surf->screen->surface,
-                    0, 0,
-                    back_surf,
-                    0, 0,
-                    surf->w, surf->h);
-            }
-			ctx->pipe->flush(ctx->pipe, PIPE_FLUSH_RENDER_CACHE | PIPE_FLUSH_TEXTURE_CACHE, NULL);
-
-#ifdef DRM_MODE_FEATURE_DIRTYFB
-			/* TODO query connector property to see if this is needed */
-			drmModeDirtyFB(dev->drmFD, surf->screen->fbID, NULL, 0);
-#else
-			(void)dev;
-#endif
-
-			/* TODO more stuff here */
-		}
-	}
-
-	return EGL_TRUE;
-}
diff --git a/src/gallium/state_trackers/egl/egl_tracker.c b/src/gallium/state_trackers/egl/egl_tracker.c
deleted file mode 100644
index 9345b0f490..0000000000
--- a/src/gallium/state_trackers/egl/egl_tracker.c
+++ /dev/null
@@ -1,274 +0,0 @@
-
-#include "utils.h"
-
-#include <stdlib.h>
-#include <stdio.h>
-#include <string.h>
-#include "egl_tracker.h"
-
-#include <fcntl.h>
-
-#include "egllog.h"
-#include "state_tracker/drm_api.h"
-
-#include "pipe/p_screen.h"
-#include "pipe/internal/p_winsys_screen.h"
-
-/** HACK */
-void* driDriverAPI;
-
-
-/*
- * Exported functions
- */
-
-/** Called by libEGL just prior to unloading/closing the driver.
- */
-static void
-drm_unload(_EGLDriver *drv)
-{
-	free(drv);
-}
-
-/**
- * The bootstrap function.  Return a new drm_driver object and
- * plug in API functions.
- * libEGL finds this function with dlopen()/dlsym() and calls it from
- * "load driver" function.
- */
-_EGLDriver *
-_eglMain(const char *args)
-{
-	_EGLDriver *drv;
-
-	drv = (_EGLDriver *) calloc(1, sizeof(_EGLDriver));
-	if (!drv) {
-		return NULL;
-	}
-
-	/* First fill in the dispatch table with defaults */
-	_eglInitDriverFallbacks(drv);
-	/* then plug in our Drm-specific functions */
-	drv->API.Initialize = drm_initialize;
-	drv->API.Terminate = drm_terminate;
-	drv->API.CreateContext = drm_create_context;
-	drv->API.MakeCurrent = drm_make_current;
-	drv->API.CreateWindowSurface = drm_create_window_surface;
-	drv->API.CreatePixmapSurface = drm_create_pixmap_surface;
-	drv->API.CreatePbufferSurface = drm_create_pbuffer_surface;
-	drv->API.DestroySurface = drm_destroy_surface;
-	drv->API.DestroyContext = drm_destroy_context;
-	drv->API.CreateScreenSurfaceMESA = drm_create_screen_surface_mesa;
-	drv->API.ShowScreenSurfaceMESA = drm_show_screen_surface_mesa;
-	drv->API.SwapBuffers = drm_swap_buffers;
-
-	drv->Name = "DRM/Gallium/Win";
-	drv->Unload = drm_unload;
-
-	return drv;
-}
-
-static void
-drm_get_device_id(struct drm_device *device)
-{
-	char path[512];
-	FILE *file;
-	char *ret;
-
-	/* TODO get the real minor */
-	int minor = 0;
-
-	device->deviceID = 0;
-
-	snprintf(path, sizeof(path), "/sys/class/drm/card%d/device/device", minor);
-	file = fopen(path, "r");
-	if (!file) {
-		_eglLog(_EGL_WARNING, "Could not retrive device ID\n");
-		return;
-	}
-
-	ret = fgets(path, sizeof( path ), file);
-	fclose(file);
-	if (!ret)
-		return;
-
-	sscanf(path, "%x", &device->deviceID);
-}
-
-static void
-drm_update_res(struct drm_device *dev)
-{
-	drmModeFreeResources(dev->res);
-	dev->res = drmModeGetResources(dev->drmFD);
-}
-
-static void
-drm_add_modes_from_connector(_EGLScreen *screen, drmModeConnectorPtr connector)
-{
-	drmModeModeInfoPtr m = NULL;
-	int i;
-
-	for (i = 0; i < connector->count_modes; i++) {
-		m = &connector->modes[i];
-		_eglAddNewMode(screen, m->hdisplay, m->vdisplay, m->vrefresh, m->name);
-	}
-}
-
-static void
-drm_find_dpms(struct drm_device *dev, struct drm_screen *screen)
-{
-	drmModeConnectorPtr c = screen->connector;
-	drmModePropertyPtr p;
-	int i;
-
-	for (i = 0; i < c->count_props; i++) {
-		p = drmModeGetProperty(dev->drmFD, c->props[i]);
-		if (!strcmp(p->name, "DPMS"))
-			break;
-
-		drmModeFreeProperty(p);
-		p = NULL;
-	}
-
-	screen->dpms = p;
-}
-
-static int drm_open_minor(int minor)
-{
-	char buf[64];
-
-	sprintf(buf, DRM_DEV_NAME, DRM_DIR_NAME, minor);
-	return open(buf, O_RDWR, 0);
-}
-
-EGLBoolean
-drm_initialize(_EGLDriver *drv, _EGLDisplay *disp, EGLint *major, EGLint *minor)
-{
-	struct drm_device *dev;
-	struct drm_screen *screen = NULL;
-	drmModeConnectorPtr connector = NULL;
-	drmModeResPtr res = NULL;
-	unsigned count_connectors = 0;
-	int num_screens = 0;
-	EGLint i;
-	int fd;
-	_EGLConfig *config;
-
-	dev = (struct drm_device *) calloc(1, sizeof(struct drm_device));
-	if (!dev)
-		return EGL_FALSE;
-	dev->api = drm_api_create();
-
-	/* try the first node */
-	fd = drm_open_minor(0);
-	if (fd < 0)
-		goto err_fd;
-
-	dev->drmFD = fd;
-	drm_get_device_id(dev);
-
-	dev->screen = dev->api->create_screen(dev->api, dev->drmFD, NULL);
-	if (!dev->screen)
-		goto err_screen;
-	dev->winsys = dev->screen->winsys;
-
-	driInitExtensions(NULL, NULL, GL_FALSE);
-
-	drm_update_res(dev);
-	res = dev->res;
-	if (res)
-		count_connectors = res->count_connectors;
-	else
-		_eglLog(_EGL_WARNING, "Could not retrive kms information\n");
-
-	for(i = 0; i < count_connectors && i < MAX_SCREENS; i++) {
-		connector = drmModeGetConnector(fd, res->connectors[i]);
-
-		if (!connector)
-			continue;
-
-		if (connector->connection != DRM_MODE_CONNECTED) {
-			drmModeFreeConnector(connector);
-			continue;
-		}
-
-		screen = malloc(sizeof(struct drm_screen));
-		memset(screen, 0, sizeof(*screen));
-		screen->connector = connector;
-		screen->connectorID = connector->connector_id;
-		_eglInitScreen(&screen->base);
-		_eglAddScreen(disp, &screen->base);
-		drm_add_modes_from_connector(&screen->base, connector);
-		drm_find_dpms(dev, screen);
-		dev->screens[num_screens++] = screen;
-	}
-	dev->count_screens = num_screens;
-
-	disp->DriverData = dev;
-
-	/* for now we only have one config */
-	config = calloc(1, sizeof(*config));
-	memset(config, 1, sizeof(*config));
-	_eglInitConfig(config, 1);
-	_eglSetConfigAttrib(config, EGL_RED_SIZE, 8);
-	_eglSetConfigAttrib(config, EGL_GREEN_SIZE, 8);
-	_eglSetConfigAttrib(config, EGL_BLUE_SIZE, 8);
-	_eglSetConfigAttrib(config, EGL_ALPHA_SIZE, 8);
-	_eglSetConfigAttrib(config, EGL_BUFFER_SIZE, 32);
-	_eglSetConfigAttrib(config, EGL_DEPTH_SIZE, 24);
-	_eglSetConfigAttrib(config, EGL_STENCIL_SIZE, 8);
-	_eglSetConfigAttrib(config, EGL_SURFACE_TYPE, EGL_PBUFFER_BIT);
-	_eglAddConfig(disp, config);
-
-	disp->ClientAPIsMask = EGL_OPENGL_BIT /*| EGL_OPENGL_ES_BIT*/;
-	/* enable supported extensions */
-	disp->Extensions.MESA_screen_surface = EGL_TRUE;
-	disp->Extensions.MESA_copy_context = EGL_TRUE;
-
-	*major = 1;
-	*minor = 4;
-
-	return EGL_TRUE;
-
-err_screen:
-	drmClose(fd);
-err_fd:
-	free(dev);
-	return EGL_FALSE;
-}
-
-EGLBoolean
-drm_terminate(_EGLDriver *drv, _EGLDisplay *dpy)
-{
-	struct drm_device *dev = lookup_drm_device(dpy);
-	struct drm_screen *screen;
-	int i = 0;
-
-	_eglReleaseDisplayResources(drv, dpy);
-	_eglCleanupDisplay(dpy);
-
-	drmFreeVersion(dev->version);
-
-	for (i = 0; i < dev->count_screens; i++) {
-		screen = dev->screens[i];
-
-		if (screen->shown)
-			drm_takedown_shown_screen(dpy, screen);
-
-		drmModeFreeProperty(screen->dpms);
-		drmModeFreeConnector(screen->connector);
-		_eglDestroyScreen(&screen->base);
-		dev->screens[i] = NULL;
-	}
-
-	dev->screen->destroy(dev->screen);
-	dev->winsys = NULL;
-
-	drmClose(dev->drmFD);
-
-	dev->api->destroy(dev->api);
-	free(dev);
-	dpy->DriverData = NULL;
-
-	return EGL_TRUE;
-}
diff --git a/src/gallium/state_trackers/egl/egl_tracker.h b/src/gallium/state_trackers/egl/egl_tracker.h
deleted file mode 100644
index 73eb1a1226..0000000000
--- a/src/gallium/state_trackers/egl/egl_tracker.h
+++ /dev/null
@@ -1,195 +0,0 @@
-
-#ifndef _EGL_TRACKER_H_
-#define _EGL_TRACKER_H_
-
-#include <stdint.h>
-
-#include "eglconfig.h"
-#include "eglcontext.h"
-#include "egldisplay.h"
-#include "egldriver.h"
-#include "eglglobals.h"
-#include "eglmode.h"
-#include "eglscreen.h"
-#include "eglsurface.h"
-
-#include "xf86drm.h"
-#include "xf86drmMode.h"
-
-#include "pipe/p_compiler.h"
-
-#include "state_tracker/st_public.h"
-
-#define MAX_SCREENS 16
-
-struct pipe_winsys;
-struct pipe_screen;
-struct pipe_context;
-struct state_tracker;
-
-struct drm_screen;
-struct drm_context;
-
-struct drm_device
-{
-	/*
-	 * pipe
-	 */
-
-	struct drm_api *api;
-	struct pipe_winsys *winsys;
-	struct pipe_screen *screen;
-
-	/*
-	 * drm
-	 */
-
-	int drmFD;
-	drmVersionPtr version;
-	int deviceID;
-
-	drmModeResPtr res;
-
-	struct drm_screen *screens[MAX_SCREENS];
-	size_t count_screens;
-};
-
-struct drm_surface
-{
-	_EGLSurface base; /* base class/object */
-
-	/*
-	 * pipe
-	 */
-
-
-	struct st_framebuffer *stfb;
-
-	/*
-	 * drm
-	 */
-
-	struct drm_screen *screen;
-
-	int w;
-	int h;
-};
-
-struct drm_context
-{
-	_EGLContext base; /* base class/object */
-
-	/* pipe */
-
-	struct pipe_context *pipe;
-	struct st_context *st;
-};
-
-struct drm_screen
-{
-	_EGLScreen base;
-
-	/*
-	 * pipe
-	 */
-
-	struct pipe_texture *tex;
-	struct pipe_surface *surface;
-
-	/*
-	 * drm
-	 */
-
-	struct {
-		unsigned height;
-		unsigned width;
-		unsigned pitch;
-		unsigned handle;
-	} front;
-
-	/* currently only support one connector */
-	drmModeConnectorPtr connector;
-	uint32_t connectorID;
-
-	/* dpms property */
-	drmModePropertyPtr dpms;
-
-	/* Has this screen been shown */
-	int shown;
-
-	/* Surface that is currently attached to this screen */
-	struct drm_surface *surf;
-
-	/* framebuffer */
-	drmModeFBPtr fb;
-	uint32_t fbID;
-
-	/* crtc and mode used */
-	/*drmModeCrtcPtr crtc;*/
-	uint32_t crtcID;
-
-	drmModeModeInfoPtr mode;
-};
-
-
-static INLINE struct drm_device *
-lookup_drm_device(_EGLDisplay *d)
-{
-	return (struct drm_device *) d->DriverData;
-}
-
-
-static INLINE struct drm_context *
-lookup_drm_context(_EGLContext *c)
-{
-	return (struct drm_context *) c;
-}
-
-
-static INLINE struct drm_surface *
-lookup_drm_surface(_EGLSurface *s)
-{
-	return (struct drm_surface *) s;
-}
-
-static INLINE struct drm_screen *
-lookup_drm_screen(_EGLScreen *s)
-{
-	return (struct drm_screen *) s;
-}
-
-/**
- * egl_visual.h
- */
-/*@{*/
-void drm_visual_modes_destroy(__GLcontextModes *modes);
-__GLcontextModes* drm_visual_modes_create(unsigned count, size_t minimum_size);
-__GLcontextModes* drm_visual_from_config(_EGLConfig *conf);
-/*@}*/
-
-/**
- * egl_surface.h
- */
-/*@{*/
-void drm_takedown_shown_screen(_EGLDisplay *dpy, struct drm_screen *screen);
-/*@}*/
-
-/**
- * All function exported to the egl side.
- */
-/*@{*/
-EGLBoolean drm_initialize(_EGLDriver *drv, _EGLDisplay *dpy, EGLint *major, EGLint *minor);
-EGLBoolean drm_terminate(_EGLDriver *drv, _EGLDisplay *dpy);
-_EGLContext *drm_create_context(_EGLDriver *drv, _EGLDisplay *dpy, _EGLConfig *conf, _EGLContext *share_list, const EGLint *attrib_list);
-EGLBoolean drm_destroy_context(_EGLDriver *drv, _EGLDisplay *dpy, _EGLContext *context);
-_EGLSurface *drm_create_window_surface(_EGLDriver *drv, _EGLDisplay *dpy, _EGLConfig *conf, NativeWindowType window, const EGLint *attrib_list);
-_EGLSurface *drm_create_pixmap_surface(_EGLDriver *drv, _EGLDisplay *dpy, _EGLConfig *conf, NativePixmapType pixmap, const EGLint *attrib_list);
-_EGLSurface *drm_create_pbuffer_surface(_EGLDriver *drv, _EGLDisplay *dpy, _EGLConfig *conf, const EGLint *attrib_list);
-_EGLSurface *drm_create_screen_surface_mesa(_EGLDriver *drv, _EGLDisplay *dpy, _EGLConfig *conf, const EGLint *attrib_list);
-EGLBoolean drm_show_screen_surface_mesa(_EGLDriver *drv, _EGLDisplay *dpy, _EGLScreen *screen, _EGLSurface *surface, _EGLMode *mode);
-EGLBoolean drm_destroy_surface(_EGLDriver *drv, _EGLDisplay *dpy, _EGLSurface *surface);
-EGLBoolean drm_make_current(_EGLDriver *drv, _EGLDisplay *dpy, _EGLSurface *draw, _EGLSurface *read, _EGLContext *context);
-EGLBoolean drm_swap_buffers(_EGLDriver *drv, _EGLDisplay *dpy, _EGLSurface *draw);
-/*@}*/
-
-#endif
diff --git a/src/gallium/state_trackers/egl/egl_visual.c b/src/gallium/state_trackers/egl/egl_visual.c
deleted file mode 100644
index e59f893851..0000000000
--- a/src/gallium/state_trackers/egl/egl_visual.c
+++ /dev/null
@@ -1,85 +0,0 @@
-
-#include "egl_tracker.h"
-
-#include "egllog.h"
-
-void
-drm_visual_modes_destroy(__GLcontextModes *modes)
-{
-   _eglLog(_EGL_DEBUG, "%s", __FUNCTION__);
-
-   while (modes) {
-      __GLcontextModes * const next = modes->next;
-      free(modes);
-      modes = next;
-   }
-}
-
-__GLcontextModes *
-drm_visual_modes_create(unsigned count, size_t minimum_size)
-{
-	/* This code copied from libGLX, and modified */
-	const size_t size = (minimum_size > sizeof(__GLcontextModes))
-		? minimum_size : sizeof(__GLcontextModes);
-	__GLcontextModes * head = NULL;
-	__GLcontextModes ** next;
-	unsigned   i;
-
-	_eglLog(_EGL_DEBUG, "%s %d %d", __FUNCTION__, count, minimum_size);
-
-	next = & head;
-	for (i = 0 ; i < count ; i++) {
-		*next = (__GLcontextModes *) calloc(1, size);
-		if (*next == NULL) {
-			drm_visual_modes_destroy(head);
-			head = NULL;
-			break;
-		}
-
-		(*next)->doubleBufferMode = 1;
-		(*next)->visualID = GLX_DONT_CARE;
-		(*next)->visualType = GLX_DONT_CARE;
-		(*next)->visualRating = GLX_NONE;
-		(*next)->transparentPixel = GLX_NONE;
-		(*next)->transparentRed = GLX_DONT_CARE;
-		(*next)->transparentGreen = GLX_DONT_CARE;
-		(*next)->transparentBlue = GLX_DONT_CARE;
-		(*next)->transparentAlpha = GLX_DONT_CARE;
-		(*next)->transparentIndex = GLX_DONT_CARE;
-		(*next)->xRenderable = GLX_DONT_CARE;
-		(*next)->fbconfigID = GLX_DONT_CARE;
-		(*next)->swapMethod = GLX_SWAP_UNDEFINED_OML;
-		(*next)->bindToTextureRgb = GLX_DONT_CARE;
-		(*next)->bindToTextureRgba = GLX_DONT_CARE;
-		(*next)->bindToMipmapTexture = GLX_DONT_CARE;
-		(*next)->bindToTextureTargets = 0;
-		(*next)->yInverted = GLX_DONT_CARE;
-
-		next = & ((*next)->next);
-	}
-
-	return head;
-}
-
-__GLcontextModes *
-drm_visual_from_config(_EGLConfig *conf)
-{
-	__GLcontextModes *visual;
-	(void)conf;
-
-	visual = drm_visual_modes_create(1, sizeof(*visual));
-	visual->redBits = 8;
-	visual->greenBits = 8;
-	visual->blueBits = 8;
-	visual->alphaBits = 8;
-
-	visual->rgbBits = 32;
-	visual->doubleBufferMode = 1;
-
-	visual->depthBits = 24;
-	visual->haveDepthBuffer = visual->depthBits > 0;
-	visual->stencilBits = 8;
-	visual->haveStencilBuffer = visual->stencilBits > 0;
-
-	return visual;
-}
diff --git a/src/gallium/state_trackers/egl_g3d/kms/native_kms.c b/src/gallium/state_trackers/egl/kms/native_kms.c
index 65829fc7b3..d5baf2c2f0 100644
--- a/src/gallium/state_trackers/egl_g3d/kms/native_kms.c
+++ b/src/gallium/state_trackers/egl/kms/native_kms.c
@@ -33,9 +33,7 @@
 #include "native_kms.h"
 
 static boolean
-kms_surface_validate(struct native_surface *nsurf,
-                     const enum native_attachment *natts,
-                     unsigned num_natts,
+kms_surface_validate(struct native_surface *nsurf, uint attachment_mask,
                      unsigned int *seq_num, struct pipe_texture **textures,
                      int *width, int *height)
 {
@@ -43,12 +41,9 @@ kms_surface_validate(struct native_surface *nsurf,
    struct kms_display *kdpy = ksurf->kdpy;
    struct pipe_screen *screen = kdpy->base.screen;
    struct pipe_texture templ, *ptex;
-   int i;
-
-   if (num_natts) {
-      if (textures)
-         memset(textures, 0, sizeof(*textures) * num_natts);
+   int att;
 
+   if (attachment_mask) {
       memset(&templ, 0, sizeof(templ));
       templ.target = PIPE_TEXTURE_2D;
       templ.last_level = 0;
@@ -62,17 +57,21 @@ kms_surface_validate(struct native_surface *nsurf,
    }
 
    /* create textures */
-   for (i = 0; i < num_natts; i++) {
-      enum native_attachment natt = natts[i];
+   for (att = 0; att < NUM_NATIVE_ATTACHMENTS; att++) {
+      /* delay the allocation */
+      if (!native_attachment_mask_test(attachment_mask, att))
+         continue;
 
-      ptex = ksurf->textures[natt];
+      ptex = ksurf->textures[att];
       if (!ptex) {
          ptex = screen->texture_create(screen, &templ);
-         ksurf->textures[natt] = ptex;
+         ksurf->textures[att] = ptex;
       }
 
-      if (textures)
-         pipe_texture_reference(&textures[i], ptex);
+      if (textures) {
+         textures[att] = NULL;
+         pipe_texture_reference(&textures[att], ptex);
+      }
    }
 
    if (seq_num)
@@ -113,7 +112,7 @@ kms_surface_init_framebuffers(struct native_surface *nsurf, boolean need_back)
 
       if (!fb->texture) {
          /* make sure the texture has been allocated */
-         kms_surface_validate(&ksurf->base, &natt, 1, NULL, NULL, NULL, NULL);
+         kms_surface_validate(&ksurf->base, 1 << natt, NULL, NULL, NULL, NULL);
          if (!ksurf->textures[natt])
             return FALSE;
 
@@ -821,6 +820,18 @@ kms_create_display(EGLNativeDisplayType dpy, struct drm_api *api)
    return &kdpy->base;
 }
 
+struct native_probe *
+native_create_probe(EGLNativeDisplayType dpy)
+{
+   return NULL;
+}
+
+enum native_probe_result
+native_get_probe_result(struct native_probe *nprobe)
+{
+   return NATIVE_PROBE_UNKNOWN;
+}
+
 /* the api is destroyed with the native display */
 static struct drm_api *drm_api;
 
diff --git a/src/gallium/state_trackers/egl_g3d/kms/native_kms.h b/src/gallium/state_trackers/egl/kms/native_kms.h
index 095186e3cf..095186e3cf 100644
--- a/src/gallium/state_trackers/egl_g3d/kms/native_kms.h
+++ b/src/gallium/state_trackers/egl/kms/native_kms.h
diff --git a/src/gallium/state_trackers/egl_g3d/x11/native_dri2.c b/src/gallium/state_trackers/egl/x11/native_dri2.c
index f497d8c1c7..07f82d878c 100644
--- a/src/gallium/state_trackers/egl_g3d/x11/native_dri2.c
+++ b/src/gallium/state_trackers/egl/x11/native_dri2.c
@@ -134,22 +134,18 @@ dri2_surface_swap_buffers(struct native_surface *nsurf)
 }
 
 static boolean
-dri2_surface_validate(struct native_surface *nsurf,
-                             const enum native_attachment *natts,
-                             unsigned num_natts,
-                             unsigned int *seq_num,
-                             struct pipe_texture **textures,
-                             int *width, int *height)
+dri2_surface_validate(struct native_surface *nsurf, uint attachment_mask,
+                      unsigned int *seq_num, struct pipe_texture **textures,
+                      int *width, int *height)
 {
    struct dri2_surface *dri2surf = dri2_surface(nsurf);
    struct dri2_display *dri2dpy = dri2surf->dri2dpy;
    unsigned int dri2atts[NUM_NATIVE_ATTACHMENTS];
-   EGLint texture_indices[NUM_NATIVE_ATTACHMENTS];
    struct pipe_texture templ;
    struct x11_drawable_buffer *xbufs;
-   int num_ins, num_outs, i;
+   int num_ins, num_outs, att, i;
 
-   if (num_natts) {
+   if (attachment_mask) {
       memset(&templ, 0, sizeof(templ));
       templ.target = PIPE_TEXTURE_2D;
       templ.last_level = 0;
@@ -160,24 +156,27 @@ dri2_surface_validate(struct native_surface *nsurf,
       templ.tex_usage = PIPE_TEXTURE_USAGE_RENDER_TARGET;
 
       if (textures)
-         memset(textures, 0, sizeof(*textures) * num_natts);
+         memset(textures, 0, sizeof(*textures) * NUM_NATIVE_ATTACHMENTS);
    }
 
    /* create textures for pbuffer */
    if (dri2surf->type == DRI2_SURFACE_TYPE_PBUFFER) {
       struct pipe_screen *screen = dri2dpy->base.screen;
 
-      for (i = 0; i < num_natts; i++) {
-         enum native_attachment natt = natts[i];
-         struct pipe_texture *ptex = dri2surf->pbuffer_textures[natt];
+      for (att = 0; att < NUM_NATIVE_ATTACHMENTS; att++) {
+         struct pipe_texture *ptex = dri2surf->pbuffer_textures[att];
+
+         /* delay the allocation */
+         if (!native_attachment_mask_test(attachment_mask, att))
+            continue;
 
          if (!ptex) {
             ptex = screen->texture_create(screen, &templ);
-            dri2surf->pbuffer_textures[natt] = ptex;
+            dri2surf->pbuffer_textures[att] = ptex;
          }
 
          if (textures)
-            pipe_texture_reference(&textures[i], ptex);
+            pipe_texture_reference(&textures[att], ptex);
       }
 
       if (seq_num)
@@ -190,34 +189,34 @@ dri2_surface_validate(struct native_surface *nsurf,
       return TRUE;
    }
 
-   for (i = 0; i < NUM_NATIVE_ATTACHMENTS; i++)
-      texture_indices[i] = -1;
-
    /* prepare the attachments */
-   num_ins = num_natts;
-   for (i = 0; i < num_natts; i++) {
-      unsigned int dri2att;
+   num_ins = 0;
+   for (att = 0; att < NUM_NATIVE_ATTACHMENTS; att++) {
+      if (native_attachment_mask_test(attachment_mask, att)) {
+         unsigned int dri2att;
+
+         switch (att) {
+         case NATIVE_ATTACHMENT_FRONT_LEFT:
+            dri2att = DRI2BufferFrontLeft;
+            break;
+         case NATIVE_ATTACHMENT_BACK_LEFT:
+            dri2att = DRI2BufferBackLeft;
+            break;
+         case NATIVE_ATTACHMENT_FRONT_RIGHT:
+            dri2att = DRI2BufferFrontRight;
+            break;
+         case NATIVE_ATTACHMENT_BACK_RIGHT:
+            dri2att = DRI2BufferBackRight;
+            break;
+         default:
+            assert(0);
+            dri2att = 0;
+            break;
+         }
 
-      switch (natts[i]) {
-      case NATIVE_ATTACHMENT_FRONT_LEFT:
-         dri2att = DRI2BufferFrontLeft;
-         break;
-      case NATIVE_ATTACHMENT_BACK_LEFT:
-         dri2att = DRI2BufferBackLeft;
-         break;
-      case NATIVE_ATTACHMENT_FRONT_RIGHT:
-         dri2att = DRI2BufferFrontRight;
-         break;
-      case NATIVE_ATTACHMENT_BACK_RIGHT:
-         dri2att = DRI2BufferBackRight;
-         break;
-      default:
-         assert(0);
-         dri2att = 0;
-         break;
+         dri2atts[num_ins] = dri2att;
+         num_ins++;
       }
-      dri2atts[i] = dri2att;
-      texture_indices[natts[i]] = i;
    }
 
    dri2surf->have_back = FALSE;
@@ -266,13 +265,13 @@ dri2_surface_validate(struct native_surface *nsurf,
          break;
       }
 
-      if (!desc || texture_indices[natt] < 0 ||
-          (textures && textures[texture_indices[natt]])) {
+      if (!desc || !native_attachment_mask_test(attachment_mask, natt) ||
+          (textures && textures[natt])) {
          if (!desc)
             _eglLog(_EGL_WARNING, "unknown buffer %d", xbuf->attachment);
-         else if (texture_indices[natt] < 0)
+         else if (!native_attachment_mask_test(attachment_mask, natt))
             _eglLog(_EGL_WARNING, "unexpected buffer %d", xbuf->attachment);
-         else if (textures && textures[texture_indices[natt]])
+         else
             _eglLog(_EGL_WARNING, "both real and fake front buffers are listed");
          continue;
       }
@@ -284,7 +283,7 @@ dri2_surface_validate(struct native_surface *nsurf,
                   desc, xbuf->pitch, xbuf->name);
          if (ptex) {
             /* the caller owns the textures */
-            textures[texture_indices[natt]] = ptex;
+            textures[natt] = ptex;
          }
       }
    }
@@ -589,6 +588,21 @@ dri2_display_get_configs(struct native_display *ndpy, int *num_configs)
    return configs;
 }
 
+static boolean
+dri2_display_is_pixmap_supported(struct native_display *ndpy,
+                                 EGLNativePixmapType pix,
+                                 const struct native_config *nconf)
+{
+   struct dri2_display *dri2dpy = dri2_display(ndpy);
+   uint depth, nconf_depth;
+
+   depth = x11_drawable_get_depth(dri2dpy->xscr, (Drawable) pix);
+   nconf_depth = util_format_get_blocksizebits(nconf->color_format);
+
+   /* simple depth match for now */
+   return (depth == nconf_depth || (depth == 24 && depth + 8 == nconf_depth));
+}
+
 static void
 dri2_display_destroy(struct native_display *ndpy)
 {
@@ -681,6 +695,7 @@ x11_create_dri2_display(EGLNativeDisplayType dpy, struct drm_api *api)
 
    dri2dpy->base.destroy = dri2_display_destroy;
    dri2dpy->base.get_configs = dri2_display_get_configs;
+   dri2dpy->base.is_pixmap_supported = dri2_display_is_pixmap_supported;
    dri2dpy->base.create_context = dri2_display_create_context;
    dri2dpy->base.create_window_surface = dri2_display_create_window_surface;
    dri2dpy->base.create_pixmap_surface = dri2_display_create_pixmap_surface;
diff --git a/src/gallium/state_trackers/egl_g3d/x11/native_x11.c b/src/gallium/state_trackers/egl/x11/native_x11.c
index 583ce3d329..dd3c9f8b6a 100644
--- a/src/gallium/state_trackers/egl_g3d/x11/native_x11.c
+++ b/src/gallium/state_trackers/egl/x11/native_x11.c
@@ -24,13 +24,90 @@
 
 #include <string.h>
 #include "util/u_debug.h"
+#include "util/u_memory.h"
 #include "state_tracker/drm_api.h"
 #include "egllog.h"
 
 #include "native_x11.h"
+#include "x11_screen.h"
+
+#define X11_PROBE_MAGIC 0x11980BE /* "X11PROBE" */
 
 static struct drm_api *api;
 
+static void
+x11_probe_destroy(struct native_probe *nprobe)
+{
+   if (nprobe->data)
+      free(nprobe->data);
+   free(nprobe);
+}
+
+struct native_probe *
+native_create_probe(EGLNativeDisplayType dpy)
+{
+   struct native_probe *nprobe;
+   struct x11_screen *xscr;
+   int scr;
+   const char *driver_name = NULL;
+   Display *xdpy;
+
+   nprobe = CALLOC_STRUCT(native_probe);
+   if (!nprobe)
+      return NULL;
+
+   xdpy = dpy;
+   if (!xdpy) {
+      xdpy = XOpenDisplay(NULL);
+      if (!xdpy) {
+         free(nprobe);
+         return NULL;
+      }
+   }
+
+   scr = DefaultScreen(xdpy);
+   xscr = x11_screen_create(xdpy, scr);
+   if (xscr) {
+      if (x11_screen_support(xscr, X11_SCREEN_EXTENSION_DRI2)) {
+         driver_name = x11_screen_probe_dri2(xscr);
+         if (driver_name)
+            nprobe->data = strdup(driver_name);
+      }
+
+      x11_screen_destroy(xscr);
+   }
+
+   if (xdpy != dpy)
+      XCloseDisplay(xdpy);
+
+   nprobe->magic = X11_PROBE_MAGIC;
+   nprobe->display = dpy;
+
+   nprobe->destroy = x11_probe_destroy;
+
+   return nprobe;
+}
+
+enum native_probe_result
+native_get_probe_result(struct native_probe *nprobe)
+{
+   if (!nprobe || nprobe->magic != X11_PROBE_MAGIC)
+      return NATIVE_PROBE_UNKNOWN;
+
+   if (!api)
+      api = drm_api_create();
+
+   /* this is a software driver */
+   if (!api)
+      return NATIVE_PROBE_SUPPORTED;
+
+   /* the display does not support DRI2 or the driver mismatches */
+   if (!nprobe->data || strcmp(api->name, (const char *) nprobe->data) != 0)
+      return NATIVE_PROBE_FALLBACK;
+
+   return NATIVE_PROBE_EXACT;
+}
+
 const char *
 native_get_name(void)
 {
diff --git a/src/gallium/state_trackers/egl_g3d/x11/native_x11.h b/src/gallium/state_trackers/egl/x11/native_x11.h
index 622ddac5df..622ddac5df 100644
--- a/src/gallium/state_trackers/egl_g3d/x11/native_x11.h
+++ b/src/gallium/state_trackers/egl/x11/native_x11.h
diff --git a/src/gallium/state_trackers/egl_g3d/x11/native_ximage.c b/src/gallium/state_trackers/egl/x11/native_ximage.c
index 24a50df7a0..dfa8df2223 100644
--- a/src/gallium/state_trackers/egl_g3d/x11/native_ximage.c
+++ b/src/gallium/state_trackers/egl/x11/native_ximage.c
@@ -66,7 +66,6 @@ struct ximage_buffer {
    XImage *ximage;
 
    struct pipe_texture *texture;
-   struct pipe_transfer *transfer;
    XShmSegmentInfo *shm_info;
    boolean xshm_attached;
 };
@@ -115,12 +114,7 @@ ximage_surface_free_buffer(struct native_surface *nsurf,
 {
    struct ximage_surface *xsurf = ximage_surface(nsurf);
    struct ximage_buffer *xbuf = &xsurf->buffers[which];
-   struct pipe_screen *screen = xsurf->xdpy->base.screen;
 
-   if (xbuf->transfer) {
-      screen->tex_transfer_destroy(xbuf->transfer);
-      xbuf->transfer = NULL;
-   }
    pipe_texture_reference(&xbuf->texture, NULL);
 
    if (xbuf->shm_info) {
@@ -193,13 +187,6 @@ ximage_surface_alloc_buffer(struct native_surface *nsurf,
       xbuf->texture = screen->texture_create(screen, &templ);
    }
 
-   if (xbuf->texture) {
-      xbuf->transfer = screen->get_tex_transfer(screen, xbuf->texture,
-            0, 0, 0, PIPE_TRANSFER_READ, 0, 0, xsurf->width, xsurf->height);
-      if (!xbuf->transfer)
-         pipe_texture_reference(&xbuf->texture, NULL);
-   }
-
    /* clean up the buffer if allocation failed */
    if (!xbuf->texture)
       ximage_surface_free_buffer(&xsurf->base, which);
@@ -214,13 +201,25 @@ ximage_surface_draw_buffer(struct native_surface *nsurf,
    struct ximage_surface *xsurf = ximage_surface(nsurf);
    struct ximage_buffer *xbuf = &xsurf->buffers[which];
    struct pipe_screen *screen = xsurf->xdpy->base.screen;
+   struct pipe_transfer *transfer;
 
    if (xsurf->type == XIMAGE_SURFACE_TYPE_PBUFFER)
       return TRUE;
 
    assert(xsurf->drawable && xbuf->ximage && xbuf->texture);
 
-   xbuf->ximage->data = screen->transfer_map(screen, xbuf->transfer);
+   transfer = screen->get_tex_transfer(screen, xbuf->texture,
+         0, 0, 0, PIPE_TRANSFER_READ, 0, 0, xsurf->width, xsurf->height);
+   if (!transfer)
+      return FALSE;
+
+   xbuf->ximage->bytes_per_line = transfer->stride;
+   xbuf->ximage->data = screen->transfer_map(screen, transfer);
+   if (!xbuf->ximage->data) {
+      screen->tex_transfer_destroy(transfer);
+      return FALSE;
+   }
+
 
    if (xbuf->shm_info)
       XShmPutImage(xsurf->xdpy->dpy, xsurf->drawable, xsurf->gc,
@@ -230,7 +229,13 @@ ximage_surface_draw_buffer(struct native_surface *nsurf,
             xbuf->ximage, 0, 0, 0, 0, xsurf->width, xsurf->height);
 
    xbuf->ximage->data = NULL;
-   screen->transfer_unmap(screen, xbuf->transfer);
+   screen->transfer_unmap(screen, transfer);
+
+   /*
+    * softpipe allows the pipe transfer to be re-used, but we don't want to
+    * rely on that behavior.
+    */
+   screen->tex_transfer_destroy(transfer);
 
    XSync(xsurf->xdpy->dpy, FALSE);
 
@@ -289,27 +294,21 @@ ximage_surface_update_geometry(struct native_surface *nsurf)
 }
 
 static boolean
-ximage_surface_validate(struct native_surface *nsurf,
-                               const enum native_attachment *natts,
-                               unsigned num_natts,
-                               unsigned int *seq_num,
-                               struct pipe_texture **textures,
-                               int *width, int *height)
+ximage_surface_validate(struct native_surface *nsurf, uint attachment_mask,
+                        unsigned int *seq_num, struct pipe_texture **textures,
+                        int *width, int *height)
 {
    struct ximage_surface *xsurf = ximage_surface(nsurf);
-   boolean new_buffers = FALSE, error = FALSE;
-   unsigned i;
+   boolean new_buffers = FALSE;
+   int att;
 
    ximage_surface_update_geometry(&xsurf->base);
 
-   if (textures)
-      memset(textures, 0, sizeof(*textures) * num_natts);
-
-   for (i = 0; i < num_natts; i++) {
-      enum native_attachment natt = natts[i];
-      struct ximage_buffer *xbuf = &xsurf->buffers[natt];
+   for (att = 0; att < NUM_NATIVE_ATTACHMENTS; att++) {
+      struct ximage_buffer *xbuf = &xsurf->buffers[att];
 
-      if (!xbuf)
+      /* delay the allocation */
+      if (!native_attachment_mask_test(attachment_mask, att))
          continue;
 
       /* reallocate the texture */
@@ -317,29 +316,19 @@ ximage_surface_validate(struct native_surface *nsurf,
           xsurf->width != xbuf->texture->width0 ||
           xsurf->height != xbuf->texture->height0) {
          new_buffers = TRUE;
-         if (ximage_surface_alloc_buffer(&xsurf->base, natt)) {
+         if (ximage_surface_alloc_buffer(&xsurf->base, att)) {
             /* update ximage */
             if (xbuf->ximage) {
-               xbuf->ximage->width = xbuf->transfer->width;
-               xbuf->ximage->height = xbuf->transfer->height;
-               xbuf->ximage->bytes_per_line = xbuf->transfer->stride;
+               xbuf->ximage->width = xsurf->width;
+               xbuf->ximage->height = xsurf->height;
             }
          }
       }
 
-      /* allocation failed */
-      if (!xbuf->texture) {
-         unsigned j;
-         for (j = 0; j < i; j++)
-            pipe_texture_reference(&textures[j], NULL);
-         for (j = i; j < num_natts; j++)
-            textures[j] = NULL;
-         error = TRUE;
-         break;
+      if (textures) {
+         textures[att] = NULL;
+         pipe_texture_reference(&textures[att], xbuf->texture);
       }
-
-      if (textures)
-         pipe_texture_reference(&textures[i], xbuf->texture);
    }
 
    /* increase the sequence number so that caller knows */
@@ -353,7 +342,7 @@ ximage_surface_validate(struct native_surface *nsurf,
    if (height)
       *height = xsurf->height;
 
-   return !error;
+   return TRUE;
 }
 
 static void
@@ -615,6 +604,34 @@ ximage_display_get_configs(struct native_display *ndpy, int *num_configs)
    return configs;
 }
 
+static boolean
+ximage_display_is_pixmap_supported(struct native_display *ndpy,
+                                   EGLNativePixmapType pix,
+                                   const struct native_config *nconf)
+{
+   struct ximage_display *xdpy = ximage_display(ndpy);
+   enum pipe_format fmt;
+   uint depth;
+
+   depth = x11_drawable_get_depth(xdpy->xscr, (Drawable) pix);
+   switch (depth) {
+   case 32:
+      fmt = PIPE_FORMAT_A8R8G8B8_UNORM;
+      break;
+   case 24:
+      fmt = PIPE_FORMAT_X8R8G8B8_UNORM;
+      break;
+   case 16:
+      fmt = PIPE_FORMAT_R5G6B5_UNORM;
+      break;
+   default:
+      fmt = PIPE_FORMAT_NONE;
+      break;
+   }
+
+   return (fmt == nconf->color_format);
+}
+
 static void
 ximage_display_destroy(struct native_display *ndpy)
 {
@@ -667,6 +684,7 @@ x11_create_ximage_display(EGLNativeDisplayType dpy, boolean use_xshm)
    xdpy->base.destroy = ximage_display_destroy;
 
    xdpy->base.get_configs = ximage_display_get_configs;
+   xdpy->base.is_pixmap_supported = ximage_display_is_pixmap_supported;
    xdpy->base.create_context = ximage_display_create_context;
    xdpy->base.create_window_surface = ximage_display_create_window_surface;
    xdpy->base.create_pixmap_surface = ximage_display_create_pixmap_surface;
diff --git a/src/gallium/state_trackers/egl_g3d/x11/sw_winsys.c b/src/gallium/state_trackers/egl/x11/sw_winsys.c
index 6ee3ede38c..6ee3ede38c 100644
--- a/src/gallium/state_trackers/egl_g3d/x11/sw_winsys.c
+++ b/src/gallium/state_trackers/egl/x11/sw_winsys.c
diff --git a/src/gallium/state_trackers/egl_g3d/x11/sw_winsys.h b/src/gallium/state_trackers/egl/x11/sw_winsys.h
index f96c5a14b0..f96c5a14b0 100644
--- a/src/gallium/state_trackers/egl_g3d/x11/sw_winsys.h
+++ b/src/gallium/state_trackers/egl/x11/sw_winsys.h
diff --git a/src/gallium/state_trackers/egl_g3d/x11/x11_screen.c b/src/gallium/state_trackers/egl/x11/x11_screen.c
index 1e98943242..76ce45ee57 100644
--- a/src/gallium/state_trackers/egl_g3d/x11/x11_screen.c
+++ b/src/gallium/state_trackers/egl/x11/x11_screen.c
@@ -30,6 +30,7 @@
 #include <X11/extensions/XShm.h>
 #include "util/u_memory.h"
 #include "util/u_math.h"
+#include "util/u_format.h"
 #include "xf86drm.h"
 #include "egllog.h"
 
@@ -41,6 +42,10 @@ struct x11_screen {
    Display *dpy;
    int number;
 
+   /*
+    * This is used to fetch GLX visuals/fbconfigs.  It uses code from egl_xdri.
+    * It might be better to rewrite the part in Xlib or XCB.
+    */
    __GLXdisplayPrivate *glx_dpy;
 
    int dri_major, dri_minor;
@@ -50,6 +55,10 @@ struct x11_screen {
 
    XVisualInfo *visuals;
    int num_visuals;
+
+   /* cached values for x11_drawable_get_depth */
+   Drawable last_drawable;
+   unsigned int last_depth;
 };
 
 
@@ -88,8 +97,8 @@ x11_screen_destroy(struct x11_screen *xscr)
    if (xscr->dri_device)
       Xfree(xscr->dri_device);
 
-   if (xscr->glx_dpy)
-      __glXRelease(xscr->glx_dpy);
+   /* xscr->glx_dpy will be destroyed with the X display */
+
    if (xscr->visuals)
       XFree(xscr->visuals);
    free(xscr);
@@ -245,6 +254,22 @@ x11_screen_is_driver_equal(struct x11_screen *xscr, const char *driver)
 }
 
 /**
+ * Probe the screen for the DRI2 driver name.
+ */
+const char *
+x11_screen_probe_dri2(struct x11_screen *xscr)
+{
+   /* get the driver name and the device name */
+   if (!xscr->dri_driver) {
+      if (!DRI2Connect(xscr->dpy, RootWindow(xscr->dpy, xscr->number),
+               &xscr->dri_driver, &xscr->dri_device))
+         xscr->dri_driver = xscr->dri_device = NULL;
+   }
+
+   return xscr->dri_driver;
+}
+
+/**
  * Enable DRI2 and returns the file descriptor of the DRM device.  The file
  * descriptor will be closed automatically when the screen is destoryed.
  */
@@ -256,13 +281,8 @@ x11_screen_enable_dri2(struct x11_screen *xscr, const char *driver)
       drm_magic_t magic;
 
       /* get the driver name and the device name first */
-      if (!xscr->dri_driver) {
-         if (!DRI2Connect(xscr->dpy, RootWindow(xscr->dpy, xscr->number),
-                  &xscr->dri_driver, &xscr->dri_device)) {
-            xscr->dri_driver = xscr->dri_device = NULL;
-            return -1;
-         }
-      }
+      if (!x11_screen_probe_dri2(xscr))
+         return -1;
 
       if (!x11_screen_is_driver_equal(xscr, driver)) {
          _eglLog(_EGL_WARNING, "Driver mismatch: %s != %s",
@@ -351,6 +371,37 @@ x11_drawable_get_buffers(struct x11_screen *xscr, Drawable drawable,
 }
 
 /**
+ * Return the depth of a drawable.
+ *
+ * Unlike other drawable functions, the drawable needs not be a DRI2 drawable.
+ */
+uint
+x11_drawable_get_depth(struct x11_screen *xscr, Drawable drawable)
+{
+   unsigned int depth;
+
+   if (drawable != xscr->last_drawable) {
+      Window root;
+      int x, y;
+      unsigned int w, h, border;
+      Status ok;
+
+      ok = XGetGeometry(xscr->dpy, drawable, &root,
+            &x, &y, &w, &h, &border, &depth);
+      if (!ok)
+         depth = 0;
+
+      xscr->last_drawable = drawable;
+      xscr->last_depth = depth;
+   }
+   else {
+      depth = xscr->last_depth;
+   }
+
+   return depth;
+}
+
+/**
  * Create a mode list of the given size.
  */
 __GLcontextModes *
diff --git a/src/gallium/state_trackers/egl_g3d/x11/x11_screen.h b/src/gallium/state_trackers/egl/x11/x11_screen.h
index 86e8e0501a..5432858ac3 100644
--- a/src/gallium/state_trackers/egl_g3d/x11/x11_screen.h
+++ b/src/gallium/state_trackers/egl/x11/x11_screen.h
@@ -70,6 +70,9 @@ x11_screen_get_glx_configs(struct x11_screen *xscr);
 const __GLcontextModes *
 x11_screen_get_glx_visuals(struct x11_screen *xscr);
 
+const char *
+x11_screen_probe_dri2(struct x11_screen *xscr);
+
 int
 x11_screen_enable_dri2(struct x11_screen *xscr, const char *driver);
 
@@ -96,4 +99,7 @@ x11_drawable_get_buffers(struct x11_screen *xscr, Drawable drawable,
                          int *width, int *height, unsigned int *attachments,
                          boolean with_format, int num_ins, int *num_outs);
 
+uint
+x11_drawable_get_depth(struct x11_screen *xscr, Drawable drawable);
+
 #endif /* _X11_SCREEN_H_ */
diff --git a/src/gallium/state_trackers/egl_g3d/Makefile b/src/gallium/state_trackers/egl_g3d/Makefile
deleted file mode 100644
index 213eb3e815..0000000000
--- a/src/gallium/state_trackers/egl_g3d/Makefile
+++ /dev/null
@@ -1,72 +0,0 @@
-TOP = ../../../..
-include $(TOP)/configs/current
-
-common_INCLUDES = \
-	-I. \
-	-I$(TOP)/src/gallium/include \
-	-I$(TOP)/src/gallium/auxiliary \
-	-I$(TOP)/src/egl/main \
-	-I$(TOP)/include
-
-common_SOURCES = $(wildcard common/*.c)
-common_OBJECTS = $(common_SOURCES:.c=.o)
-
-
-x11_INCLUDES = \
-	-I$(TOP)/src/gallium/drivers \
-	-I$(TOP)/src/glx/x11 \
-	-I$(TOP)/src/mesa \
-	$(shell pkg-config --cflags-only-I libdrm)
-
-x11_SOURCES = $(wildcard x11/*.c) $(TOP)/src/glx/x11/dri2.c
-x11_OBJECTS = $(x11_SOURCES:.c=.o)
-
-
-kms_INCLUDES = $(shell pkg-config --cflags-only-I libdrm)
-kms_SOURCES = $(wildcard kms/*.c)
-kms_OBJECTS = $(kms_SOURCES:.c=.o)
-
-
-ALL_INCLUDES = $(common_INCLUDES) $(x11_INCLUDES) $(kms_INCLUDES)
-ALL_SOURCES = $(common_SOURCES) $(x11_SOURCES) $(kms_SOURCES)
-ALL_OBJECTS = $(common_OBJECTS) $(x11_OBJECTS) $(kms_OBJECTS)
-
-##### TARGETS #####
-
-EGL_DISPLAYS_MODS = $(foreach dpy, $(EGL_DISPLAYS), libegl$(dpy).a)
-
-default: depend $(EGL_DISPLAYS_MODS)
-
-
-libeglx11.a: $(x11_OBJECTS) $(common_OBJECTS) Makefile
-	$(MKLIB) -o eglx11 -static $(x11_OBJECTS) $(common_OBJECTS)
-
-libeglkms.a: $(kms_OBJECTS) $(common_OBJECTS) Makefile
-	$(MKLIB) -o eglkms -static $(kms_OBJECTS) $(common_OBJECTS)
-
-depend: 
-	rm -f depend
-	touch depend
-	$(MKDEP) $(MKDEP_OPTIONS) $(ALL_INCLUDES) $(ALL_SOURCES) 2> /dev/null
-
-clean:
-	rm -f $(ALL_OBJECTS)
-	rm -f $(EGL_DISPLAYS_MODS)
-	rm -f depend depend.bak
-
-# Dummy target
-install:
-	@echo -n ""
-
-##### RULES #####
-
-$(common_OBJECTS): %.o: %.c
-	$(CC) -c $(common_INCLUDES) $(DEFINES) $(CFLAGS) $< -o $@
-
-$(x11_OBJECTS): %.o: %.c
-	$(CC) -c $(common_INCLUDES) $(x11_INCLUDES) $(DEFINES) $(CFLAGS) $< -o $@
-
-$(kms_OBJECTS): %.o: %.c
-	$(CC) -c $(common_INCLUDES) $(kms_INCLUDES) $(DEFINES) $(CFLAGS) $< -o $@
-
-sinclude depend
diff --git a/src/gallium/state_trackers/egl_g3d/x11/glxinit.c b/src/gallium/state_trackers/egl_g3d/x11/glxinit.c
deleted file mode 100644
index c955a908b9..0000000000
--- a/src/gallium/state_trackers/egl_g3d/x11/glxinit.c
+++ /dev/null
@@ -1,573 +0,0 @@
-/**
- * GLX initialization.  Code based on glxext.c, glx_query.c, and
- * glcontextmodes.c under src/glx/x11/.  The major difference is that no DRI
- * related code here.
- *
- */
-
-#include <assert.h>
-#include <X11/Xlib.h>
-#include <X11/Xproto.h>
-#include <X11/extensions/Xext.h>
-#include <X11/extensions/extutil.h>
-#include <sys/time.h>
-
-#include "x11_screen.h"
-#include "glxinit.h"
-
-typedef struct GLXGenericGetString
-{
-   CARD8 reqType;
-   CARD8 glxCode;
-   CARD16 length B16;
-   CARD32 for_whom B32;
-   CARD32 name B32;
-} xGLXGenericGetStringReq;
-
-#define sz_xGLXGenericGetStringReq 12
-#define X_GLXGenericGetString 0
-
-/* Extension required boiler plate */
-
-static char *__glXExtensionName = GLX_EXTENSION_NAME;
-static XExtensionInfo *__glXExtensionInfo = NULL;
-
-static /* const */ XExtensionHooks __glXExtensionHooks = { NULL };
-static
-XEXT_GENERATE_FIND_DISPLAY(__glXFindDisplay, __glXExtensionInfo,
-                           __glXExtensionName, &__glXExtensionHooks,
-                           __GLX_NUMBER_EVENTS, NULL)
-
-static GLint
-_gl_convert_from_x_visual_type(int visualType)
-{
-#define NUM_VISUAL_TYPES   6
-   static const int glx_visual_types[NUM_VISUAL_TYPES] = {
-      GLX_STATIC_GRAY, GLX_GRAY_SCALE,
-      GLX_STATIC_COLOR, GLX_PSEUDO_COLOR,
-      GLX_TRUE_COLOR, GLX_DIRECT_COLOR
-   };
-
-   return ((unsigned) visualType < NUM_VISUAL_TYPES)
-      ? glx_visual_types[visualType] : GLX_NONE;
-}
-
-_X_HIDDEN char *
-__glXQueryServerString(Display * dpy, int opcode, CARD32 screen, CARD32 name)
-{
-   xGLXGenericGetStringReq *req;
-   xGLXSingleReply reply;
-   int length;
-   int numbytes;
-   char *buf;
-   CARD32 for_whom = screen;
-   CARD32 glxCode = X_GLXQueryServerString;
-
-
-   LockDisplay(dpy);
-
-
-   /* All of the GLX protocol requests for getting a string from the server
-    * look the same.  The exact meaning of the for_whom field is usually
-    * either the screen number (for glXQueryServerString) or the context tag
-    * (for GLXSingle).
-    */
-
-   GetReq(GLXGenericGetString, req);
-   req->reqType = opcode;
-   req->glxCode = glxCode;
-   req->for_whom = for_whom;
-   req->name = name;
-
-   _XReply(dpy, (xReply *) & reply, 0, False);
-
-   length = reply.length * 4;
-   numbytes = reply.size;
-
-   buf = (char *) Xmalloc(numbytes);
-   if (buf != NULL) {
-      _XRead(dpy, buf, numbytes);
-      length -= numbytes;
-   }
-
-   _XEatData(dpy, length);
-
-   UnlockDisplay(dpy);
-   SyncHandle();
-
-   return buf;
-}
-
-/************************************************************************/
-/*
-** Free the per screen configs data as well as the array of
-** __glXScreenConfigs.
-*/
-static void
-FreeScreenConfigs(__GLXdisplayPrivate * priv)
-{
-   __GLXscreenConfigs *psc;
-   GLint i, screens;
-
-   /* Free screen configuration information */
-   psc = priv->screenConfigs;
-   screens = ScreenCount(priv->dpy);
-   for (i = 0; i < screens; i++, psc++) {
-      if (psc->configs) {
-         x11_context_modes_destroy(psc->configs);
-         psc->configs = NULL;   /* NOTE: just for paranoia */
-      }
-      if (psc->visuals) {
-         x11_context_modes_destroy(psc->visuals);
-         psc->visuals = NULL;   /* NOTE: just for paranoia */
-      }
-      Xfree((char *) psc->serverGLXexts);
-   }
-   XFree((char *) priv->screenConfigs);
-   priv->screenConfigs = NULL;
-}
-
-/************************************************************************/
-
-/*
-** Query the version of the GLX extension.  This procedure works even if
-** the client extension is not completely set up.
-*/
-static Bool
-QueryVersion(Display * dpy, int opcode, int *major, int *minor)
-{
-   xGLXQueryVersionReq *req;
-   xGLXQueryVersionReply reply;
-
-   /* Send the glXQueryVersion request */
-   LockDisplay(dpy);
-   GetReq(GLXQueryVersion, req);
-   req->reqType = opcode;
-   req->glxCode = X_GLXQueryVersion;
-   req->majorVersion = GLX_MAJOR_VERSION;
-   req->minorVersion = GLX_MINOR_VERSION;
-   _XReply(dpy, (xReply *) & reply, 0, False);
-   UnlockDisplay(dpy);
-   SyncHandle();
-
-   if (reply.majorVersion != GLX_MAJOR_VERSION) {
-      /*
-       ** The server does not support the same major release as this
-       ** client.
-       */
-      return GL_FALSE;
-   }
-   *major = reply.majorVersion;
-   *minor = min(reply.minorVersion, GLX_MINOR_VERSION);
-   return GL_TRUE;
-}
-
-_X_HIDDEN void
-__glXInitializeVisualConfigFromTags(__GLcontextModes * config, int count,
-                                    const INT32 * bp, Bool tagged_only,
-                                    Bool fbconfig_style_tags)
-{
-   int i;
-
-   if (!tagged_only) {
-      /* Copy in the first set of properties */
-      config->visualID = *bp++;
-
-      config->visualType = _gl_convert_from_x_visual_type(*bp++);
-
-      config->rgbMode = *bp++;
-
-      config->redBits = *bp++;
-      config->greenBits = *bp++;
-      config->blueBits = *bp++;
-      config->alphaBits = *bp++;
-      config->accumRedBits = *bp++;
-      config->accumGreenBits = *bp++;
-      config->accumBlueBits = *bp++;
-      config->accumAlphaBits = *bp++;
-
-      config->doubleBufferMode = *bp++;
-      config->stereoMode = *bp++;
-
-      config->rgbBits = *bp++;
-      config->depthBits = *bp++;
-      config->stencilBits = *bp++;
-      config->numAuxBuffers = *bp++;
-      config->level = *bp++;
-
-      count -= __GLX_MIN_CONFIG_PROPS;
-   }
-
-   /*
-    ** Additional properties may be in a list at the end
-    ** of the reply.  They are in pairs of property type
-    ** and property value.
-    */
-
-#define FETCH_OR_SET(tag) \
-    config-> tag = ( fbconfig_style_tags ) ? *bp++ : 1
-
-   for (i = 0; i < count; i += 2) {
-      switch (*bp++) {
-      case GLX_RGBA:
-         FETCH_OR_SET(rgbMode);
-         break;
-      case GLX_BUFFER_SIZE:
-         config->rgbBits = *bp++;
-         break;
-      case GLX_LEVEL:
-         config->level = *bp++;
-         break;
-      case GLX_DOUBLEBUFFER:
-         FETCH_OR_SET(doubleBufferMode);
-         break;
-      case GLX_STEREO:
-         FETCH_OR_SET(stereoMode);
-         break;
-      case GLX_AUX_BUFFERS:
-         config->numAuxBuffers = *bp++;
-         break;
-      case GLX_RED_SIZE:
-         config->redBits = *bp++;
-         break;
-      case GLX_GREEN_SIZE:
-         config->greenBits = *bp++;
-         break;
-      case GLX_BLUE_SIZE:
-         config->blueBits = *bp++;
-         break;
-      case GLX_ALPHA_SIZE:
-         config->alphaBits = *bp++;
-         break;
-      case GLX_DEPTH_SIZE:
-         config->depthBits = *bp++;
-         break;
-      case GLX_STENCIL_SIZE:
-         config->stencilBits = *bp++;
-         break;
-      case GLX_ACCUM_RED_SIZE:
-         config->accumRedBits = *bp++;
-         break;
-      case GLX_ACCUM_GREEN_SIZE:
-         config->accumGreenBits = *bp++;
-         break;
-      case GLX_ACCUM_BLUE_SIZE:
-         config->accumBlueBits = *bp++;
-         break;
-      case GLX_ACCUM_ALPHA_SIZE:
-         config->accumAlphaBits = *bp++;
-         break;
-      case GLX_VISUAL_CAVEAT_EXT:
-         config->visualRating = *bp++;
-         break;
-      case GLX_X_VISUAL_TYPE:
-         config->visualType = *bp++;
-         break;
-      case GLX_TRANSPARENT_TYPE:
-         config->transparentPixel = *bp++;
-         break;
-      case GLX_TRANSPARENT_INDEX_VALUE:
-         config->transparentIndex = *bp++;
-         break;
-      case GLX_TRANSPARENT_RED_VALUE:
-         config->transparentRed = *bp++;
-         break;
-      case GLX_TRANSPARENT_GREEN_VALUE:
-         config->transparentGreen = *bp++;
-         break;
-      case GLX_TRANSPARENT_BLUE_VALUE:
-         config->transparentBlue = *bp++;
-         break;
-      case GLX_TRANSPARENT_ALPHA_VALUE:
-         config->transparentAlpha = *bp++;
-         break;
-      case GLX_VISUAL_ID:
-         config->visualID = *bp++;
-         break;
-      case GLX_DRAWABLE_TYPE:
-         config->drawableType = *bp++;
-         break;
-      case GLX_RENDER_TYPE:
-         config->renderType = *bp++;
-         break;
-      case GLX_X_RENDERABLE:
-         config->xRenderable = *bp++;
-         break;
-      case GLX_FBCONFIG_ID:
-         config->fbconfigID = *bp++;
-         break;
-      case GLX_MAX_PBUFFER_WIDTH:
-         config->maxPbufferWidth = *bp++;
-         break;
-      case GLX_MAX_PBUFFER_HEIGHT:
-         config->maxPbufferHeight = *bp++;
-         break;
-      case GLX_MAX_PBUFFER_PIXELS:
-         config->maxPbufferPixels = *bp++;
-         break;
-      case GLX_OPTIMAL_PBUFFER_WIDTH_SGIX:
-         config->optimalPbufferWidth = *bp++;
-         break;
-      case GLX_OPTIMAL_PBUFFER_HEIGHT_SGIX:
-         config->optimalPbufferHeight = *bp++;
-         break;
-      case GLX_VISUAL_SELECT_GROUP_SGIX:
-         config->visualSelectGroup = *bp++;
-         break;
-      case GLX_SWAP_METHOD_OML:
-         config->swapMethod = *bp++;
-         break;
-      case GLX_SAMPLE_BUFFERS_SGIS:
-         config->sampleBuffers = *bp++;
-         break;
-      case GLX_SAMPLES_SGIS:
-         config->samples = *bp++;
-         break;
-      case GLX_BIND_TO_TEXTURE_RGB_EXT:
-         config->bindToTextureRgb = *bp++;
-         break;
-      case GLX_BIND_TO_TEXTURE_RGBA_EXT:
-         config->bindToTextureRgba = *bp++;
-         break;
-      case GLX_BIND_TO_MIPMAP_TEXTURE_EXT:
-         config->bindToMipmapTexture = *bp++;
-         break;
-      case GLX_BIND_TO_TEXTURE_TARGETS_EXT:
-         config->bindToTextureTargets = *bp++;
-         break;
-      case GLX_Y_INVERTED_EXT:
-         config->yInverted = *bp++;
-         break;
-      case None:
-         i = count;
-         break;
-      default:
-         break;
-      }
-   }
-
-   config->renderType =
-      (config->rgbMode) ? GLX_RGBA_BIT : GLX_COLOR_INDEX_BIT;
-
-   config->haveAccumBuffer = ((config->accumRedBits +
-                               config->accumGreenBits +
-                               config->accumBlueBits +
-                               config->accumAlphaBits) > 0);
-   config->haveDepthBuffer = (config->depthBits > 0);
-   config->haveStencilBuffer = (config->stencilBits > 0);
-}
-
-static __GLcontextModes *
-createConfigsFromProperties(Display * dpy, int nvisuals, int nprops,
-                            int screen, GLboolean tagged_only)
-{
-   INT32 buf[__GLX_TOTAL_CONFIG], *props;
-   unsigned prop_size;
-   __GLcontextModes *modes, *m;
-   int i;
-
-   if (nprops == 0)
-      return NULL;
-
-   /* FIXME: Is the __GLX_MIN_CONFIG_PROPS test correct for FBconfigs? */
-
-   /* Check number of properties */
-   if (nprops < __GLX_MIN_CONFIG_PROPS || nprops > __GLX_MAX_CONFIG_PROPS)
-      return NULL;
-
-   /* Allocate memory for our config structure */
-   modes = x11_context_modes_create(nvisuals);
-   if (!modes)
-      return NULL;
-
-   prop_size = nprops * __GLX_SIZE_INT32;
-   if (prop_size <= sizeof(buf))
-      props = buf;
-   else
-      props = Xmalloc(prop_size);
-
-   /* Read each config structure and convert it into our format */
-   m = modes;
-   for (i = 0; i < nvisuals; i++) {
-      _XRead(dpy, (char *) props, prop_size);
-      /* Older X servers don't send this so we default it here. */
-      m->drawableType = GLX_WINDOW_BIT;
-      __glXInitializeVisualConfigFromTags(m, nprops, props,
-                                     tagged_only, GL_TRUE);
-      m->screen = screen;
-      m = m->next;
-   }
-
-   if (props != buf)
-      Xfree(props);
-
-   return modes;
-}
-
-static GLboolean
-getVisualConfigs(Display * dpy, __GLXdisplayPrivate * priv, int screen)
-{
-   xGLXGetVisualConfigsReq *req;
-   __GLXscreenConfigs *psc;
-   xGLXGetVisualConfigsReply reply;
-
-   LockDisplay(dpy);
-
-   psc = priv->screenConfigs + screen;
-   psc->visuals = NULL;
-   GetReq(GLXGetVisualConfigs, req);
-   req->reqType = priv->majorOpcode;
-   req->glxCode = X_GLXGetVisualConfigs;
-   req->screen = screen;
-
-   if (!_XReply(dpy, (xReply *) & reply, 0, False))
-      goto out;
-
-   psc->visuals = createConfigsFromProperties(dpy,
-                                              reply.numVisuals,
-                                              reply.numProps,
-                                              screen, GL_FALSE);
-
- out:
-   UnlockDisplay(dpy);
-   return psc->visuals != NULL;
-}
-
-static GLboolean
-getFBConfigs(Display * dpy, __GLXdisplayPrivate * priv, int screen)
-{
-   xGLXGetFBConfigsReq *fb_req;
-   xGLXGetFBConfigsSGIXReq *sgi_req;
-   xGLXVendorPrivateWithReplyReq *vpreq;
-   xGLXGetFBConfigsReply reply;
-   __GLXscreenConfigs *psc;
-
-   psc = priv->screenConfigs + screen;
-   psc->serverGLXexts =
-      __glXQueryServerString(dpy, priv->majorOpcode, screen, GLX_EXTENSIONS);
-
-   LockDisplay(dpy);
-
-   psc->configs = NULL;
-   if (atof(priv->serverGLXversion) >= 1.3) {
-      GetReq(GLXGetFBConfigs, fb_req);
-      fb_req->reqType = priv->majorOpcode;
-      fb_req->glxCode = X_GLXGetFBConfigs;
-      fb_req->screen = screen;
-   }
-   else if (strstr(psc->serverGLXexts, "GLX_SGIX_fbconfig") != NULL) {
-      GetReqExtra(GLXVendorPrivateWithReply,
-                  sz_xGLXGetFBConfigsSGIXReq +
-                  sz_xGLXVendorPrivateWithReplyReq, vpreq);
-      sgi_req = (xGLXGetFBConfigsSGIXReq *) vpreq;
-      sgi_req->reqType = priv->majorOpcode;
-      sgi_req->glxCode = X_GLXVendorPrivateWithReply;
-      sgi_req->vendorCode = X_GLXvop_GetFBConfigsSGIX;
-      sgi_req->screen = screen;
-   }
-   else
-      goto out;
-
-   if (!_XReply(dpy, (xReply *) & reply, 0, False))
-      goto out;
-
-   psc->configs = createConfigsFromProperties(dpy,
-                                              reply.numFBConfigs,
-                                              reply.numAttribs * 2,
-                                              screen, GL_TRUE);
-
- out:
-   UnlockDisplay(dpy);
-   return psc->configs != NULL;
-}
-
-static GLboolean
-AllocAndFetchScreenConfigs(Display * dpy, __GLXdisplayPrivate * priv)
-{
-   __GLXscreenConfigs *psc;
-   GLint i, screens;
-
-   /*
-    ** First allocate memory for the array of per screen configs.
-    */
-   screens = ScreenCount(dpy);
-   psc = (__GLXscreenConfigs *) Xmalloc(screens * sizeof(__GLXscreenConfigs));
-   if (!psc) {
-      return GL_FALSE;
-   }
-   memset(psc, 0, screens * sizeof(__GLXscreenConfigs));
-   priv->screenConfigs = psc;
-
-   priv->serverGLXversion =
-      __glXQueryServerString(dpy, priv->majorOpcode, 0, GLX_VERSION);
-   if (priv->serverGLXversion == NULL) {
-      FreeScreenConfigs(priv);
-      return GL_FALSE;
-   }
-
-   for (i = 0; i < screens; i++, psc++) {
-      getFBConfigs(dpy, priv, i);
-      getVisualConfigs(dpy, priv, i);
-      psc->scr = i;
-      psc->dpy = dpy;
-   }
-
-   SyncHandle();
-
-   return GL_TRUE;
-}
-
-_X_HIDDEN void
-__glXRelease(__GLXdisplayPrivate *dpyPriv)
-{
-   FreeScreenConfigs(dpyPriv);
-
-   if (dpyPriv->serverGLXvendor) {
-      Xfree((char *) dpyPriv->serverGLXvendor);
-      dpyPriv->serverGLXvendor = NULL;
-   }
-   if (dpyPriv->serverGLXversion) {
-      Xfree((char *) dpyPriv->serverGLXversion);
-      dpyPriv->serverGLXversion = NULL;
-   }
-
-   Xfree(dpyPriv);
-}
-
-_X_HIDDEN __GLXdisplayPrivate *
-__glXInitialize(Display * dpy)
-{
-   XExtDisplayInfo *info = __glXFindDisplay(dpy);
-   __GLXdisplayPrivate *dpyPriv;
-   int major, minor;
-
-   if (!XextHasExtension(info))
-      return NULL;
-
-   /* See if the versions are compatible */
-   if (!QueryVersion(dpy, info->codes->major_opcode, &major, &minor))
-      return NULL;
-
-   dpyPriv = (__GLXdisplayPrivate *) Xcalloc(1, sizeof(__GLXdisplayPrivate));
-   if (!dpyPriv)
-      return NULL;
-
-   /*
-    ** Init the display private and then read in the screen config
-    ** structures from the server.
-    */
-   dpyPriv->majorOpcode = info->codes->major_opcode;
-   dpyPriv->majorVersion = major;
-   dpyPriv->minorVersion = minor;
-   dpyPriv->dpy = dpy;
-
-   dpyPriv->serverGLXvendor = NULL;
-   dpyPriv->serverGLXversion = NULL;
-
-   if (!AllocAndFetchScreenConfigs(dpy, dpyPriv)) {
-      Xfree(dpyPriv);
-      return NULL;
-   }
-
-   return dpyPriv;
-}
diff --git a/src/gallium/state_trackers/egl_g3d/x11/glxinit.h b/src/gallium/state_trackers/egl_g3d/x11/glxinit.h
deleted file mode 100644
index 515a825222..0000000000
--- a/src/gallium/state_trackers/egl_g3d/x11/glxinit.h
+++ /dev/null
@@ -1,14 +0,0 @@
-#ifndef GLXINIT_INCLUDED
-#define GLXINIT_INCLUDED
-
-#include <X11/Xlib.h>
-
-#ifndef GLX_DIRECT_RENDERING
-#define GLX_DIRECT_RENDERING
-#endif
-#include "glxclient.h"
-
-extern void
-__glXRelease(__GLXdisplayPrivate *dpyPriv);
-
-#endif /* GLXINIT_INCLUDED */
diff --git a/src/gallium/state_trackers/es/Makefile b/src/gallium/state_trackers/es/Makefile
new file mode 100644
index 0000000000..b036551271
--- /dev/null
+++ b/src/gallium/state_trackers/es/Makefile
@@ -0,0 +1,84 @@
+# src/gallium/state_trackers/es/Makefile
+
+# Build the ES 1/2 state tracker libraries
+# This consists of core Mesa ES, plus GL/gallium state tracker.
+
+TOP = ../../../..
+include $(TOP)/configs/current
+
+GLES_1_VERSION_MAJOR = 1
+GLES_1_VERSION_MINOR = 1
+GLES_1_VERSION_PATCH = 0
+
+GLES_2_VERSION_MAJOR = 2
+GLES_2_VERSION_MINOR = 0
+GLES_2_VERSION_PATCH = 0
+
+
+# Maybe move these into configs/default:
+GLES_1_LIB = GLESv1_CM
+GLES_1_LIB_NAME = lib$(GLES_1_LIB).so
+GLES_2_LIB = GLESv2
+GLES_2_LIB_NAME = lib$(GLES_2_LIB).so
+
+
+ES1_OBJECTS = st_es1.o
+ES2_OBJECTS = st_es2.o
+
+
+ES1_LIBS = \
+	$(TOP)/src/mesa/es/libes1gallium.a \
+	$(TOP)/src/mesa/es/libes1api.a
+
+ES2_LIBS = \
+	$(TOP)/src/mesa/es/libes2gallium.a \
+	$(TOP)/src/mesa/es/libes2api.a
+
+SYS_LIBS = -lm -pthread
+
+
+INCLUDE_DIRS = \
+	-I$(TOP)/src/gallium/include
+
+.c.o:
+	$(CC) -c $(INCLUDE_DIRS) $(DEFINES) $(CFLAGS) $< -o $@
+
+
+# Default: make both GL ES 1.1 and GL ES 2.0 libraries
+default: $(TOP)/$(LIB_DIR)/$(GLES_1_LIB_NAME) $(TOP)/$(LIB_DIR)/$(GLES_2_LIB_NAME)
+
+# Make the shared libs
+$(TOP)/$(LIB_DIR)/$(GLES_1_LIB_NAME): $(ES1_OBJECTS) $(ES1_LIBS) $(GALLIUM_AUXILIARIES)
+	$(MKLIB) -o $(GLES_1_LIB) -linker '$(CC)' -ldflags '$(LDFLAGS)' \
+		-major $(GLES_1_VERSION_MAJOR) \
+		-minor $(GLES_1_VERSION_MINOR) \
+		-patch $(GLES_1_VERSION_PATCH) \
+		-install $(TOP)/$(LIB_DIR) $(MKLIB_OPTIONS) \
+		$(ES1_OBJECTS) \
+		-Wl,--whole-archive $(ES1_LIBS) -Wl,--no-whole-archive \
+		$(GALLIUM_AUXILIARIES) $(SYS_LIBS)
+
+$(TOP)/$(LIB_DIR)/$(GLES_2_LIB_NAME): $(ES2_OBJECTS) $(ES1_LIBS) $(GALLIUM_AUXILIARIES)
+	$(MKLIB) -o $(GLES_2_LIB) -linker '$(CC)' -ldflags '$(LDFLAGS)' \
+		-major $(GLES_2_VERSION_MAJOR) \
+		-minor $(GLES_2_VERSION_MINOR) \
+		-patch $(GLES_2_VERSION_PATCH) \
+		-install $(TOP)/$(LIB_DIR) $(MKLIB_OPTIONS) \
+		$(ES2_OBJECTS) \
+		-Wl,--whole-archive $(ES2_LIBS) -Wl,--no-whole-archive \
+		$(GALLIUM_AUXILIARIES) $(SYS_LIBS)
+
+install: default
+	$(INSTALL) -d $(DESTDIR)$(INSTALL_DIR)/include/GLES
+	$(INSTALL) -m 644 $(TOP)/include/GLES/*.h $(DESTDIR)$(INSTALL_DIR)/include/GLES
+	$(INSTALL) -d $(DESTDIR)$(INSTALL_DIR)/include/GLES2
+	$(INSTALL) -m 644 $(TOP)/include/GLES2/*.h $(DESTDIR)$(INSTALL_DIR)/include/GLES2
+	$(INSTALL) -d $(DESTDIR)$(INSTALL_DIR)/$(LIB_DIR)
+	$(MINSTALL) $(TOP)/$(LIB_DIR)/libGLESv1* $(DESTDIR)$(INSTALL_DIR)/$(LIB_DIR)
+	$(MINSTALL) $(TOP)/$(LIB_DIR)/libGLESv2* $(DESTDIR)$(INSTALL_DIR)/$(LIB_DIR)
+
+clean:
+	-rm -f *.o *~
+	-rm -f $(TOP)/$(LIB_DIR)/$(GLES_1_LIB_NAME)* $(TOP)/$(LIB_DIR)/$(GLES_2_LIB_NAME)*
+
+depend:
diff --git a/src/gallium/state_trackers/es/st_es1.c b/src/gallium/state_trackers/es/st_es1.c
new file mode 100644
index 0000000000..25bc53b21e
--- /dev/null
+++ b/src/gallium/state_trackers/es/st_es1.c
@@ -0,0 +1,3 @@
+#include "pipe/p_compiler.h"
+
+PUBLIC const int st_api_OpenGL_ES1 = 1;
diff --git a/src/gallium/state_trackers/es/st_es2.c b/src/gallium/state_trackers/es/st_es2.c
new file mode 100644
index 0000000000..171ea62b97
--- /dev/null
+++ b/src/gallium/state_trackers/es/st_es2.c
@@ -0,0 +1,3 @@
+#include "pipe/p_compiler.h"
+
+PUBLIC const int st_api_OpenGL_ES2 = 1;
diff --git a/src/gallium/state_trackers/glx/xlib/glx_getproc.c b/src/gallium/state_trackers/glx/xlib/glx_getproc.c
index 84d47b12ed..bd4a85caa0 100644
--- a/src/gallium/state_trackers/glx/xlib/glx_getproc.c
+++ b/src/gallium/state_trackers/glx/xlib/glx_getproc.c
@@ -193,7 +193,7 @@ _glxapi_get_proc_address(const char *funcName)
 }
 
 
-__GLXextFuncPtr
+PUBLIC __GLXextFuncPtr
 glXGetProcAddressARB(const GLubyte *procName)
 {
    __GLXextFuncPtr f;
diff --git a/src/gallium/state_trackers/python/SConscript b/src/gallium/state_trackers/python/SConscript
index 8498a90812..527e065cd9 100644
--- a/src/gallium/state_trackers/python/SConscript
+++ b/src/gallium/state_trackers/python/SConscript
@@ -21,6 +21,7 @@ if 'python' in env['statetrackers']:
             'gdi32',
             'user32',
             'kernel32',
+            'ws2_32',
         ])
     else:
         env.Append(LIBS = [
diff --git a/src/gallium/state_trackers/python/samples/gs.py b/src/gallium/state_trackers/python/samples/gs.py
index a07cf557f2..cd68abac9a 100644
--- a/src/gallium/state_trackers/python/samples/gs.py
+++ b/src/gallium/state_trackers/python/samples/gs.py
@@ -72,11 +72,11 @@ def test(dev):
 
     # disabled blending/masking
     blend = Blend()
-    blend.rgb_src_factor = PIPE_BLENDFACTOR_ONE
-    blend.alpha_src_factor = PIPE_BLENDFACTOR_ONE
-    blend.rgb_dst_factor = PIPE_BLENDFACTOR_ZERO
-    blend.alpha_dst_factor = PIPE_BLENDFACTOR_ZERO
-    blend.colormask = PIPE_MASK_RGBA
+    blend.rt[0].rgb_src_factor = PIPE_BLENDFACTOR_ONE
+    blend.rt[0].alpha_src_factor = PIPE_BLENDFACTOR_ONE
+    blend.rt[0].rgb_dst_factor = PIPE_BLENDFACTOR_ZERO
+    blend.rt[0].alpha_dst_factor = PIPE_BLENDFACTOR_ZERO
+    blend.rt[0].colormask = PIPE_MASK_RGBA
     ctx.set_blend(blend)
 
     # depth/stencil/alpha
diff --git a/src/gallium/state_trackers/python/samples/tri.py b/src/gallium/state_trackers/python/samples/tri.py
index e5e168bdc8..f0b5e3dc98 100644
--- a/src/gallium/state_trackers/python/samples/tri.py
+++ b/src/gallium/state_trackers/python/samples/tri.py
@@ -72,11 +72,11 @@ def test(dev):
 
     # disabled blending/masking
     blend = Blend()
-    blend.rgb_src_factor = PIPE_BLENDFACTOR_ONE
-    blend.alpha_src_factor = PIPE_BLENDFACTOR_ONE
-    blend.rgb_dst_factor = PIPE_BLENDFACTOR_ZERO
-    blend.alpha_dst_factor = PIPE_BLENDFACTOR_ZERO
-    blend.colormask = PIPE_MASK_RGBA
+    blend.rt[0].rgb_src_factor = PIPE_BLENDFACTOR_ONE
+    blend.rt[0].alpha_src_factor = PIPE_BLENDFACTOR_ONE
+    blend.rt[0].rgb_dst_factor = PIPE_BLENDFACTOR_ZERO
+    blend.rt[0].alpha_dst_factor = PIPE_BLENDFACTOR_ZERO
+    blend.rt[0].colormask = PIPE_MASK_RGBA
     ctx.set_blend(blend)
 
     # depth/stencil/alpha
diff --git a/src/gallium/state_trackers/python/st_device.c b/src/gallium/state_trackers/python/st_device.c
index d144af2447..e41125f6f9 100644
--- a/src/gallium/state_trackers/python/st_device.c
+++ b/src/gallium/state_trackers/python/st_device.c
@@ -180,11 +180,11 @@ st_context_create(struct st_device *st_dev)
    {
       struct pipe_blend_state blend;
       memset(&blend, 0, sizeof(blend));
-      blend.rgb_src_factor = PIPE_BLENDFACTOR_ONE;
-      blend.alpha_src_factor = PIPE_BLENDFACTOR_ONE;
-      blend.rgb_dst_factor = PIPE_BLENDFACTOR_ZERO;
-      blend.alpha_dst_factor = PIPE_BLENDFACTOR_ZERO;
-      blend.colormask = PIPE_MASK_RGBA;
+      blend.rt[0].rgb_src_factor = PIPE_BLENDFACTOR_ONE;
+      blend.rt[0].alpha_src_factor = PIPE_BLENDFACTOR_ONE;
+      blend.rt[0].rgb_dst_factor = PIPE_BLENDFACTOR_ZERO;
+      blend.rt[0].alpha_dst_factor = PIPE_BLENDFACTOR_ZERO;
+      blend.rt[0].colormask = PIPE_MASK_RGBA;
       cso_set_blend(st_ctx->cso, &blend);
    }
 
diff --git a/src/gallium/state_trackers/python/st_softpipe_winsys.c b/src/gallium/state_trackers/python/st_softpipe_winsys.c
index a3294e877a..dfe3e465f7 100644
--- a/src/gallium/state_trackers/python/st_softpipe_winsys.c
+++ b/src/gallium/state_trackers/python/st_softpipe_winsys.c
@@ -35,225 +35,10 @@
  * @author Jose Fonseca
  */
 
-
-#include "pipe/internal/p_winsys_screen.h"/* port to just p_screen */
-#include "pipe/p_format.h"
-#include "pipe/p_context.h"
-#include "pipe/p_inlines.h"
-#include "util/u_format.h"
-#include "util/u_math.h"
-#include "util/u_memory.h"
 #include "softpipe/sp_winsys.h"
 #include "st_winsys.h"
 
-
-struct st_softpipe_buffer
-{
-   struct pipe_buffer base;
-   boolean userBuffer;  /** Is this a user-space buffer? */
-   void *data;
-   void *mapped;
-};
-
-
-/** Cast wrapper */
-static INLINE struct st_softpipe_buffer *
-st_softpipe_buffer( struct pipe_buffer *buf )
-{
-   return (struct st_softpipe_buffer *)buf;
-}
-
-
-static void *
-st_softpipe_buffer_map(struct pipe_winsys *winsys, 
-                       struct pipe_buffer *buf,
-                       unsigned flags)
-{
-   struct st_softpipe_buffer *st_softpipe_buf = st_softpipe_buffer(buf);
-   st_softpipe_buf->mapped = st_softpipe_buf->data;
-   return st_softpipe_buf->mapped;
-}
-
-
-static void
-st_softpipe_buffer_unmap(struct pipe_winsys *winsys, 
-                         struct pipe_buffer *buf)
-{
-   struct st_softpipe_buffer *st_softpipe_buf = st_softpipe_buffer(buf);
-   st_softpipe_buf->mapped = NULL;
-}
-
-
-static void
-st_softpipe_buffer_destroy(struct pipe_buffer *buf)
-{
-   struct st_softpipe_buffer *oldBuf = st_softpipe_buffer(buf);
-
-   if (oldBuf->data) {
-      if (!oldBuf->userBuffer)
-         align_free(oldBuf->data);
-
-      oldBuf->data = NULL;
-   }
-
-   FREE(oldBuf);
-}
-
-
-static void
-st_softpipe_flush_frontbuffer(struct pipe_winsys *winsys,
-                              struct pipe_surface *surf,
-                              void *context_private)
-{
-}
-
-
-
-static const char *
-st_softpipe_get_name(struct pipe_winsys *winsys)
-{
-   return "softpipe";
-}
-
-
-static struct pipe_buffer *
-st_softpipe_buffer_create(struct pipe_winsys *winsys, 
-                          unsigned alignment, 
-                          unsigned usage,
-                          unsigned size)
-{
-   struct st_softpipe_buffer *buffer = CALLOC_STRUCT(st_softpipe_buffer);
-
-   pipe_reference_init(&buffer->base.reference, 1);
-   buffer->base.alignment = alignment;
-   buffer->base.usage = usage;
-   buffer->base.size = size;
-
-   buffer->data = align_malloc(size, alignment);
-
-   return &buffer->base;
-}
-
-
-/**
- * Create buffer which wraps user-space data.
- */
-static struct pipe_buffer *
-st_softpipe_user_buffer_create(struct pipe_winsys *winsys, 
-                               void *ptr, 
-                               unsigned bytes)
-{
-   struct st_softpipe_buffer *buffer;
-   
-   buffer = CALLOC_STRUCT(st_softpipe_buffer);
-   if(!buffer)
-      return NULL;
-   
-   pipe_reference_init(&buffer->base.reference, 1);
-   buffer->base.size = bytes;
-   buffer->userBuffer = TRUE;
-   buffer->data = ptr;
-
-   return &buffer->base;
-}
-
-
-static struct pipe_buffer *
-st_softpipe_surface_buffer_create(struct pipe_winsys *winsys,
-                                  unsigned width, unsigned height,
-                                  enum pipe_format format,
-                                  unsigned usage,
-                                  unsigned tex_usage,
-                                  unsigned *stride)
-{
-   const unsigned alignment = 64;
-   unsigned nblocksy;
-
-   nblocksy = util_format_get_nblocksy(format, height);
-   *stride = align(util_format_get_stride(format, width), alignment);
-
-   return winsys->buffer_create(winsys, alignment,
-                                usage,
-                                *stride * nblocksy);
-}
-
-
-static void
-st_softpipe_fence_reference(struct pipe_winsys *winsys, 
-                            struct pipe_fence_handle **ptr,
-                            struct pipe_fence_handle *fence)
-{
-}
-
-
-static int
-st_softpipe_fence_signalled(struct pipe_winsys *winsys, 
-                            struct pipe_fence_handle *fence,
-                            unsigned flag)
-{
-   return 0;
-}
-
-
-static int
-st_softpipe_fence_finish(struct pipe_winsys *winsys, 
-                         struct pipe_fence_handle *fence,
-                         unsigned flag)
-{
-   return 0;
-}
-
-
-static void
-st_softpipe_destroy(struct pipe_winsys *winsys)
-{
-   FREE(winsys);
-}
-
-
-static struct pipe_screen *
-st_softpipe_screen_create(void)
-{
-   static struct pipe_winsys *winsys;
-   struct pipe_screen *screen;
-
-   winsys = CALLOC_STRUCT(pipe_winsys);
-   if(!winsys)
-      return NULL;
-
-   winsys->destroy = st_softpipe_destroy;
-   
-   winsys->buffer_create = st_softpipe_buffer_create;
-   winsys->user_buffer_create = st_softpipe_user_buffer_create;
-   winsys->buffer_map = st_softpipe_buffer_map;
-   winsys->buffer_unmap = st_softpipe_buffer_unmap;
-   winsys->buffer_destroy = st_softpipe_buffer_destroy;
-
-   winsys->surface_buffer_create = st_softpipe_surface_buffer_create;
-
-   winsys->fence_reference = st_softpipe_fence_reference;
-   winsys->fence_signalled = st_softpipe_fence_signalled;
-   winsys->fence_finish = st_softpipe_fence_finish;
-
-   winsys->flush_frontbuffer = st_softpipe_flush_frontbuffer;
-   winsys->get_name = st_softpipe_get_name;
-
-   screen = softpipe_create_screen(winsys);
-   if(!screen)
-      st_softpipe_destroy(winsys);
-
-   return screen;
-}
-
-
-static struct pipe_context *
-st_softpipe_context_create(struct pipe_screen *screen)
-{
-   return softpipe_create(screen);
-}
-
-
 const struct st_winsys st_softpipe_winsys = {
-   &st_softpipe_screen_create,
-   &st_softpipe_context_create,
+   &softpipe_create_screen_malloc,
+   &softpipe_create,
 };
diff --git a/src/gallium/state_trackers/python/tests/regress/fragment-shader/frag-cb-1d.sh b/src/gallium/state_trackers/python/tests/regress/fragment-shader/frag-cb-1d.sh
new file mode 100644
index 0000000000..85fb9ea4e7
--- /dev/null
+++ b/src/gallium/state_trackers/python/tests/regress/fragment-shader/frag-cb-1d.sh
@@ -0,0 +1,13 @@
+FRAG
+
+DCL IN[0], COLOR, LINEAR
+DCL OUT[0], COLOR
+DCL CONST[1]
+DCL CONST[3]
+DCL TEMP[0..1]
+
+ADD TEMP[0], IN[0], CONST[1]
+RCP TEMP[1], CONST[3].xxxx
+MUL OUT[0], TEMP[0], TEMP[1]
+
+END
diff --git a/src/gallium/state_trackers/python/tests/regress/fragment-shader/frag-cb-2d.sh b/src/gallium/state_trackers/python/tests/regress/fragment-shader/frag-cb-2d.sh
new file mode 100644
index 0000000000..f70a5146f4
--- /dev/null
+++ b/src/gallium/state_trackers/python/tests/regress/fragment-shader/frag-cb-2d.sh
@@ -0,0 +1,9 @@
+FRAG
+
+DCL IN[0], COLOR, LINEAR
+DCL OUT[0], COLOR
+DCL CONST[1][1..2]
+
+MAD OUT[0], IN[0], CONST[1][2], CONST[1][1]
+
+END
diff --git a/src/gallium/state_trackers/python/tests/regress/fragment-shader/frag-slt.sh b/src/gallium/state_trackers/python/tests/regress/fragment-shader/frag-slt.sh
index f2a1521cbf..d58b7886a1 100644
--- a/src/gallium/state_trackers/python/tests/regress/fragment-shader/frag-slt.sh
+++ b/src/gallium/state_trackers/python/tests/regress/fragment-shader/frag-slt.sh
@@ -1,4 +1,4 @@
-FRAG1.1
+FRAG
 
 DCL IN[0], COLOR, LINEAR
 DCL OUT[0], COLOR
diff --git a/src/gallium/state_trackers/python/tests/regress/fragment-shader/fragment-shader.py b/src/gallium/state_trackers/python/tests/regress/fragment-shader/fragment-shader.py
index 8d3bf9d4d7..41dd69d254 100644
--- a/src/gallium/state_trackers/python/tests/regress/fragment-shader/fragment-shader.py
+++ b/src/gallium/state_trackers/python/tests/regress/fragment-shader/fragment-shader.py
@@ -26,6 +26,7 @@
 # 
 ##########################################################################
 
+import struct
 
 from gallium import *
 
@@ -50,11 +51,11 @@ def test(dev, name):
 
     # disabled blending/masking
     blend = Blend()
-    blend.rgb_src_factor = PIPE_BLENDFACTOR_ONE
-    blend.alpha_src_factor = PIPE_BLENDFACTOR_ONE
-    blend.rgb_dst_factor = PIPE_BLENDFACTOR_ZERO
-    blend.alpha_dst_factor = PIPE_BLENDFACTOR_ZERO
-    blend.colormask = PIPE_MASK_RGBA
+    blend.rt[0].rgb_src_factor = PIPE_BLENDFACTOR_ONE
+    blend.rt[0].alpha_src_factor = PIPE_BLENDFACTOR_ONE
+    blend.rt[0].rgb_dst_factor = PIPE_BLENDFACTOR_ZERO
+    blend.rt[0].alpha_dst_factor = PIPE_BLENDFACTOR_ZERO
+    blend.rt[0].colormask = PIPE_MASK_RGBA
     ctx.set_blend(blend)
 
     # depth/stencil/alpha
@@ -146,6 +147,42 @@ def test(dev, name):
     fs = Shader(file('frag-' + name + '.sh', 'rt').read())
     ctx.set_fragment_shader(fs)
 
+    constbuf0 = dev.buffer_create(64,
+                                  (PIPE_BUFFER_USAGE_CONSTANT |
+                                   PIPE_BUFFER_USAGE_GPU_READ |
+                                   PIPE_BUFFER_USAGE_CPU_WRITE),
+                                  4 * 4 * 4)
+
+    cbdata = ''
+    cbdata += struct.pack('4f', 0.4, 0.0, 0.0, 1.0)
+    cbdata += struct.pack('4f', 1.0, 1.0, 1.0, 1.0)
+    cbdata += struct.pack('4f', 2.0, 2.0, 2.0, 2.0)
+    cbdata += struct.pack('4f', 4.0, 8.0, 16.0, 32.0)
+
+    constbuf0.write(cbdata, 0)
+
+    ctx.set_constant_buffer(PIPE_SHADER_FRAGMENT,
+                            0,
+                            constbuf0)
+
+    constbuf1 = dev.buffer_create(64,
+                                  (PIPE_BUFFER_USAGE_CONSTANT |
+                                   PIPE_BUFFER_USAGE_GPU_READ |
+                                   PIPE_BUFFER_USAGE_CPU_WRITE),
+                                  4 * 4 * 4)
+
+    cbdata = ''
+    cbdata += struct.pack('4f', 0.1, 0.1, 0.1, 0.1)
+    cbdata += struct.pack('4f', 0.25, 0.25, 0.25, 0.25)
+    cbdata += struct.pack('4f', 0.5, 0.5, 0.5, 0.5)
+    cbdata += struct.pack('4f', 0.75, 0.75, 0.75, 0.75)
+
+    constbuf1.write(cbdata, 0)
+
+    ctx.set_constant_buffer(PIPE_SHADER_FRAGMENT,
+                            1,
+                            constbuf1)
+
     xy = [
         -0.8, -0.8,
          0.8, -0.8,
@@ -184,6 +221,8 @@ def main():
     tests = [
         'abs',
         'add',
+        'cb-1d',
+        'cb-2d',
         'dp3',
         'dp4',
         'dst',
diff --git a/src/gallium/state_trackers/python/tests/regress/vertex-shader/vert-cb-1d.sh b/src/gallium/state_trackers/python/tests/regress/vertex-shader/vert-cb-1d.sh
new file mode 100644
index 0000000000..b41fe5dd38
--- /dev/null
+++ b/src/gallium/state_trackers/python/tests/regress/vertex-shader/vert-cb-1d.sh
@@ -0,0 +1,16 @@
+VERT
+
+DCL IN[0], POSITION
+DCL IN[1], COLOR
+DCL OUT[0], POSITION
+DCL OUT[1], COLOR
+DCL CONST[1]
+DCL CONST[3]
+DCL TEMP[0..1]
+
+MOV OUT[0], IN[0]
+ADD TEMP[0], IN[1], CONST[1]
+RCP TEMP[1], CONST[3].xxxx
+MUL OUT[1], TEMP[0], TEMP[1]
+
+END
diff --git a/src/gallium/state_trackers/python/tests/regress/vertex-shader/vert-cb-2d.sh b/src/gallium/state_trackers/python/tests/regress/vertex-shader/vert-cb-2d.sh
new file mode 100644
index 0000000000..45f5e6b729
--- /dev/null
+++ b/src/gallium/state_trackers/python/tests/regress/vertex-shader/vert-cb-2d.sh
@@ -0,0 +1,12 @@
+VERT
+
+DCL IN[0], POSITION
+DCL IN[1], COLOR
+DCL OUT[0], POSITION
+DCL OUT[1], COLOR
+DCL CONST[1][1..2]
+
+MOV OUT[0], IN[0]
+MAD OUT[1], IN[1], CONST[1][2], CONST[1][1]
+
+END
diff --git a/src/gallium/state_trackers/python/tests/regress/vertex-shader/vertex-shader.py b/src/gallium/state_trackers/python/tests/regress/vertex-shader/vertex-shader.py
index 01bf5a3210..2c44f872e1 100644
--- a/src/gallium/state_trackers/python/tests/regress/vertex-shader/vertex-shader.py
+++ b/src/gallium/state_trackers/python/tests/regress/vertex-shader/vertex-shader.py
@@ -27,6 +27,8 @@
 ##########################################################################
 
 
+import struct
+
 from gallium import *
 
 def make_image(surface):
@@ -50,11 +52,11 @@ def test(dev, name):
 
     # disabled blending/masking
     blend = Blend()
-    blend.rgb_src_factor = PIPE_BLENDFACTOR_ONE
-    blend.alpha_src_factor = PIPE_BLENDFACTOR_ONE
-    blend.rgb_dst_factor = PIPE_BLENDFACTOR_ZERO
-    blend.alpha_dst_factor = PIPE_BLENDFACTOR_ZERO
-    blend.colormask = PIPE_MASK_RGBA
+    blend.rt[0].rgb_src_factor = PIPE_BLENDFACTOR_ONE
+    blend.rt[0].alpha_src_factor = PIPE_BLENDFACTOR_ONE
+    blend.rt[0].rgb_dst_factor = PIPE_BLENDFACTOR_ZERO
+    blend.rt[0].alpha_dst_factor = PIPE_BLENDFACTOR_ZERO
+    blend.rt[0].colormask = PIPE_MASK_RGBA
     ctx.set_blend(blend)
 
     # depth/stencil/alpha
@@ -143,6 +145,42 @@ def test(dev, name):
     ''')
     ctx.set_fragment_shader(fs)
 
+    constbuf0 = dev.buffer_create(64,
+                                  (PIPE_BUFFER_USAGE_CONSTANT |
+                                   PIPE_BUFFER_USAGE_GPU_READ |
+                                   PIPE_BUFFER_USAGE_CPU_WRITE),
+                                  4 * 4 * 4)
+
+    cbdata = ''
+    cbdata += struct.pack('4f', 0.4, 0.0, 0.0, 1.0)
+    cbdata += struct.pack('4f', 1.0, 1.0, 1.0, 1.0)
+    cbdata += struct.pack('4f', 2.0, 2.0, 2.0, 2.0)
+    cbdata += struct.pack('4f', 4.0, 8.0, 16.0, 32.0)
+
+    constbuf0.write(cbdata, 0)
+
+    ctx.set_constant_buffer(PIPE_SHADER_VERTEX,
+                            0,
+                            constbuf0)
+
+    constbuf1 = dev.buffer_create(64,
+                                  (PIPE_BUFFER_USAGE_CONSTANT |
+                                   PIPE_BUFFER_USAGE_GPU_READ |
+                                   PIPE_BUFFER_USAGE_CPU_WRITE),
+                                  4 * 4 * 4)
+
+    cbdata = ''
+    cbdata += struct.pack('4f', 0.1, 0.1, 0.1, 0.1)
+    cbdata += struct.pack('4f', 0.25, 0.25, 0.25, 0.25)
+    cbdata += struct.pack('4f', 0.5, 0.5, 0.5, 0.5)
+    cbdata += struct.pack('4f', 0.75, 0.75, 0.75, 0.75)
+
+    constbuf1.write(cbdata, 0)
+
+    ctx.set_constant_buffer(PIPE_SHADER_VERTEX,
+                            1,
+                            constbuf1)
+
     xy = [
          0.0,  0.8,
         -0.2,  0.4,
@@ -213,6 +251,8 @@ def main():
         'add',
         'arl',
         'arr',
+        'cb-1d',
+        'cb-2d',
         'dp3',
         'dp4',
         'dst',
diff --git a/src/gallium/state_trackers/python/tests/texture_render.py b/src/gallium/state_trackers/python/tests/texture_render.py
index 79287f2cac..0fac1ea5ef 100755
--- a/src/gallium/state_trackers/python/tests/texture_render.py
+++ b/src/gallium/state_trackers/python/tests/texture_render.py
@@ -115,11 +115,11 @@ class TextureTest(TestCase):
     
         # disabled blending/masking
         blend = Blend()
-        blend.rgb_src_factor = PIPE_BLENDFACTOR_ONE
-        blend.alpha_src_factor = PIPE_BLENDFACTOR_ONE
-        blend.rgb_dst_factor = PIPE_BLENDFACTOR_ZERO
-        blend.alpha_dst_factor = PIPE_BLENDFACTOR_ZERO
-        blend.colormask = PIPE_MASK_RGBA
+        blend.rt[0].rgb_src_factor = PIPE_BLENDFACTOR_ONE
+        blend.rt[0].alpha_src_factor = PIPE_BLENDFACTOR_ONE
+        blend.rt[0].rgb_dst_factor = PIPE_BLENDFACTOR_ZERO
+        blend.rt[0].alpha_dst_factor = PIPE_BLENDFACTOR_ZERO
+        blend.rt[0].colormask = PIPE_MASK_RGBA
         ctx.set_blend(blend)
     
         # no-op depth/stencil/alpha
diff --git a/src/gallium/state_trackers/python/tests/texture_sample.py b/src/gallium/state_trackers/python/tests/texture_sample.py
index 520961c805..db32b537a1 100755
--- a/src/gallium/state_trackers/python/tests/texture_sample.py
+++ b/src/gallium/state_trackers/python/tests/texture_sample.py
@@ -140,11 +140,11 @@ class TextureColorSampleTest(TestCase):
     
         # disabled blending/masking
         blend = Blend()
-        blend.rgb_src_factor = PIPE_BLENDFACTOR_ONE
-        blend.alpha_src_factor = PIPE_BLENDFACTOR_ONE
-        blend.rgb_dst_factor = PIPE_BLENDFACTOR_ZERO
-        blend.alpha_dst_factor = PIPE_BLENDFACTOR_ZERO
-        blend.colormask = PIPE_MASK_RGBA
+        blend.rt[0].rgb_src_factor = PIPE_BLENDFACTOR_ONE
+        blend.rt[0].alpha_src_factor = PIPE_BLENDFACTOR_ONE
+        blend.rt[0].rgb_dst_factor = PIPE_BLENDFACTOR_ZERO
+        blend.rt[0].alpha_dst_factor = PIPE_BLENDFACTOR_ZERO
+        blend.rt[0].colormask = PIPE_MASK_RGBA
         ctx.set_blend(blend)
     
         # no-op depth/stencil/alpha
@@ -327,11 +327,11 @@ class TextureDepthSampleTest(TestCase):
     
         # disabled blending/masking
         blend = Blend()
-        blend.rgb_src_factor = PIPE_BLENDFACTOR_ONE
-        blend.alpha_src_factor = PIPE_BLENDFACTOR_ONE
-        blend.rgb_dst_factor = PIPE_BLENDFACTOR_ZERO
-        blend.alpha_dst_factor = PIPE_BLENDFACTOR_ZERO
-        blend.colormask = PIPE_MASK_RGBA
+        blend.rt[0].rgb_src_factor = PIPE_BLENDFACTOR_ONE
+        blend.rt[0].alpha_src_factor = PIPE_BLENDFACTOR_ONE
+        blend.rt[0].rgb_dst_factor = PIPE_BLENDFACTOR_ZERO
+        blend.rt[0].alpha_dst_factor = PIPE_BLENDFACTOR_ZERO
+        blend.rt[0].colormask = PIPE_MASK_RGBA
         ctx.set_blend(blend)
     
         # depth/stencil/alpha
diff --git a/src/gallium/state_trackers/vega/Makefile b/src/gallium/state_trackers/vega/Makefile
index fc97bf51f8..037d8dc911 100644
--- a/src/gallium/state_trackers/vega/Makefile
+++ b/src/gallium/state_trackers/vega/Makefile
@@ -1,8 +1,14 @@
-# src/mesa/Makefile
+# src/gallium/state_trackers/vega/Makefile
 
 TOP = ../../../..
 include $(TOP)/configs/current
-GALLIUM = $(TOP)
+
+VG_LIB = OpenVG
+VG_LIB_NAME = lib$(VG_LIB).so
+
+VG_MAJOR = 1
+VG_MINOR = 0
+VG_TINY = 0
 
 ### Lists of source files, included by Makefiles
 
@@ -34,88 +40,54 @@ VG_SOURCES = \
            shader.c \
            shaders_cache.c
 
+VG_OBJECTS = $(VG_SOURCES:.c=.o)
 
-### All the core C sources
-
-ALL_SOURCES = \
-        $(VG_SOURCES)
-
-
-### Object files
-VG_OBJECTS = \
-	$(VG_SOURCES:.c=.o)
+VG_LIBS = $(GALLIUM_AUXILIARIES) -lm
 
 ### Include directories
 
 INCLUDE_DIRS = \
 	-I$(TOP)/include \
-	-I$(GALLIUM)/include \
-	-I$(GALLIUM)/src/gallium/include \
-	-I$(GALLIUM)/src/gallium/auxiliary
+	-I$(TOP)/src/gallium/include \
+	-I$(TOP)/src/gallium/auxiliary
 
-VG_LIB = OpenVG
-VG_LIB_NAME = lib$(VG_LIB).so
-
-VG_MAJOR = 1
-VG_MINOR = 0
-VG_TINY = 0
-
-GALLIUM_LIBS = \
-	$(GALLIUM)/src/gallium/auxiliary/libgallium.a
-
-.SUFFIXES : .cpp
 
 .c.o:
-	$(CC) -c $(INCLUDE_DIRS) $(CFLAGS) $< -o $@
+	$(CC) -c $(INCLUDE_DIRS) $(DEFINES) $(CFLAGS) $< -o $@
 
-.cpp.o:
-	$(CXX) -c $(INCLUDE_DIRS) $(CXXFLAGS) $< -o $@
-
-.S.o:
-	$(CC) -c $(INCLUDE_DIRS) $(CFLAGS) $< -o $@
-
-
-default: depend subdirs $(TOP)/$(LIB_DIR)/$(VG_LIB_NAME)
+default: depend $(TOP)/$(LIB_DIR)/$(VG_LIB_NAME)
 
 # Make the OpenVG library
-$(TOP)/$(LIB_DIR)/$(VG_LIB_NAME): $(VG_OBJECTS) $(GALLIUM_LIBS)
-	$(TOP)/bin/mklib -o $(VG_LIB) \
+$(TOP)/$(LIB_DIR)/$(VG_LIB_NAME): $(VG_OBJECTS) $(VG_LIBS)
+	$(MKLIB) -o $(VG_LIB) -linker '$(CC)' -ldflags '$(LDFLAGS)' \
 		-major $(VG_MAJOR) \
 		-minor $(VG_MINOR) \
 		-patch $(VG_TINY) \
 		-install $(TOP)/$(LIB_DIR) \
-		$(VG_OBJECTS) $(GALLIUM_LIBS) \
-		-Wl,--whole-archive $(LIBS) -Wl,--no-whole-archive $(SYS_LIBS)
+		$(VG_OBJECTS) $(VG_LIBS)
 
 ######################################################################
 # Generic stuff
 
-depend: $(ALL_SOURCES)
+depend: $(VG_SOURCES)
 	@ echo "running $(MKDEP)"
 	@ rm -f depend  # workaround oops on gutsy?!?
 	@ touch depend
-	@ $(MKDEP) $(MKDEP_OPTIONS) $(DEFINES) $(INCLUDE_DIRS) $(ALL_SOURCES) \
+	@ $(MKDEP) $(MKDEP_OPTIONS) $(DEFINES) $(INCLUDE_DIRS) $(VG_SOURCES) \
 		> /dev/null 2>/dev/null
 
-
-subdirs:
-
 install: default
-	$(INSTALL) -d $(INSTALL_DIR)/include/VG
-	$(INSTALL) -d $(INSTALL_DIR)/$(LIB_DIR)
-	$(INSTALL) -m 644 $(TOP)/include/VG/*.h $(INSTALL_DIR)/include/VG
-	@if [ -e $(TOP)/$(LIB_DIR)/$(VG_LIB_NAME) ]; then \
-		$(INSTALL) $(TOP)/$(LIB_DIR)/libOpenVG* $(INSTALL_DIR)/$(LIB_DIR); \
-	fi
+	$(INSTALL) -d $(DESTDIR)$(INSTALL_DIR)/include/VG
+	$(INSTALL) -m 644 $(TOP)/include/VG/*.h $(DESTDIR)$(INSTALL_DIR)/include/VG
+	$(INSTALL) -d $(DESTDIR)$(INSTALL_DIR)/$(LIB_DIR)
+	$(MINSTALL) $(TOP)/$(LIB_DIR)/libOpenVG* $(DESTDIR)$(INSTALL_DIR)/$(LIB_DIR)
 
 # Emacs tags
 tags:
 	etags `find . -name \*.[ch]` $(TOP)/include/VG/*.h
 
 clean:
-	-rm -f *.o
-	-rm -f */*.o
-	-rm -f */*/*.o
-	-rm -f depend depend.bak
+	rm -f $(VG_OBJECTS)
+	rm -f depend depend.bak
 
-include depend
+sinclude depend
diff --git a/src/gallium/state_trackers/vega/api_filters.c b/src/gallium/state_trackers/vega/api_filters.c
index 8f69ee0109..cf4a11bb14 100644
--- a/src/gallium/state_trackers/vega/api_filters.c
+++ b/src/gallium/state_trackers/vega/api_filters.c
@@ -127,19 +127,19 @@ static void setup_blend()
    struct vg_context *ctx = vg_current_context();
    struct pipe_blend_state blend;
    memset(&blend, 0, sizeof(blend));
-   blend.rgb_src_factor = PIPE_BLENDFACTOR_ONE;
-   blend.alpha_src_factor = PIPE_BLENDFACTOR_ONE;
-   blend.rgb_dst_factor = PIPE_BLENDFACTOR_ZERO;
-   blend.alpha_dst_factor = PIPE_BLENDFACTOR_ZERO;
+   blend.rt[0].rgb_src_factor = PIPE_BLENDFACTOR_ONE;
+   blend.rt[0].alpha_src_factor = PIPE_BLENDFACTOR_ONE;
+   blend.rt[0].rgb_dst_factor = PIPE_BLENDFACTOR_ZERO;
+   blend.rt[0].alpha_dst_factor = PIPE_BLENDFACTOR_ZERO;
    if (ctx->state.vg.filter_channel_mask & VG_RED)
-      blend.colormask |= PIPE_MASK_R;
+      blend.rt[0].colormask |= PIPE_MASK_R;
    if (ctx->state.vg.filter_channel_mask & VG_GREEN)
-      blend.colormask |= PIPE_MASK_G;
+      blend.rt[0].colormask |= PIPE_MASK_G;
    if (ctx->state.vg.filter_channel_mask & VG_BLUE)
-      blend.colormask |= PIPE_MASK_B;
+      blend.rt[0].colormask |= PIPE_MASK_B;
    if (ctx->state.vg.filter_channel_mask & VG_ALPHA)
-      blend.colormask |= PIPE_MASK_A;
-   blend.blend_enable = 1;
+      blend.rt[0].colormask |= PIPE_MASK_A;
+   blend.rt[0].blend_enable = 0;
    cso_set_blend(ctx->cso_context, &blend);
 }
 
diff --git a/src/gallium/state_trackers/vega/api_masks.c b/src/gallium/state_trackers/vega/api_masks.c
index 4f9f3dae17..e6070c7353 100644
--- a/src/gallium/state_trackers/vega/api_masks.c
+++ b/src/gallium/state_trackers/vega/api_masks.c
@@ -32,7 +32,6 @@
 #include "vg_context.h"
 #include "pipe/p_context.h"
 #include "pipe/p_inlines.h"
-#include "pipe/internal/p_winsys_screen.h" /* for winsys->update_buffer */
 
 #include "util/u_pack_color.h"
 #include "util/u_draw_quad.h"
@@ -116,8 +115,8 @@ clear_with_quad(struct vg_context *st, float x0, float y0,
      x1, y1);
    */
 
-   if (st->pipe->winsys && st->pipe->winsys->update_buffer)
-      st->pipe->winsys->update_buffer( st->pipe->winsys,
+   if (st->pipe->screen && st->pipe->screen->update_buffer)
+      st->pipe->screen->update_buffer( st->pipe->screen,
                                        st->pipe->priv );
 
    cso_save_blend(st->cso_context);
@@ -129,14 +128,11 @@ clear_with_quad(struct vg_context *st, float x0, float y0,
    {
       struct pipe_blend_state blend;
       memset(&blend, 0, sizeof(blend));
-      blend.rgb_src_factor = PIPE_BLENDFACTOR_ONE;
-      blend.alpha_src_factor = PIPE_BLENDFACTOR_ONE;
-      blend.rgb_dst_factor = PIPE_BLENDFACTOR_ZERO;
-      blend.alpha_dst_factor = PIPE_BLENDFACTOR_ZERO;
-      blend.colormask |= PIPE_MASK_R;
-      blend.colormask |= PIPE_MASK_G;
-      blend.colormask |= PIPE_MASK_B;
-      blend.colormask |= PIPE_MASK_A;
+      blend.rt[0].rgb_src_factor = PIPE_BLENDFACTOR_ONE;
+      blend.rt[0].alpha_src_factor = PIPE_BLENDFACTOR_ONE;
+      blend.rt[0].rgb_dst_factor = PIPE_BLENDFACTOR_ZERO;
+      blend.rt[0].alpha_dst_factor = PIPE_BLENDFACTOR_ZERO;
+      blend.rt[0].colormask = PIPE_MASK_RGBA;
       cso_set_blend(st->cso_context, &blend);
    }
 
diff --git a/src/gallium/state_trackers/vega/mask.c b/src/gallium/state_trackers/vega/mask.c
index 3e260e7073..ba8ecef1d2 100644
--- a/src/gallium/state_trackers/vega/mask.c
+++ b/src/gallium/state_trackers/vega/mask.c
@@ -354,15 +354,12 @@ static void setup_mask_blend()
    struct pipe_blend_state blend;
 
    memset(&blend, 0, sizeof(struct pipe_blend_state));
-   blend.blend_enable = 1;
-   blend.colormask |= PIPE_MASK_R;
-   blend.colormask |= PIPE_MASK_G;
-   blend.colormask |= PIPE_MASK_B;
-   blend.colormask |= PIPE_MASK_A;
-   blend.rgb_src_factor = PIPE_BLENDFACTOR_ONE;
-   blend.alpha_src_factor = PIPE_BLENDFACTOR_ONE;
-   blend.rgb_dst_factor = PIPE_BLENDFACTOR_ZERO;
-   blend.alpha_dst_factor = PIPE_BLENDFACTOR_ZERO;
+   blend.rt[0].blend_enable = 0;
+   blend.rt[0].colormask = PIPE_MASK_RGBA;
+   blend.rt[0].rgb_src_factor = PIPE_BLENDFACTOR_ONE;
+   blend.rt[0].alpha_src_factor = PIPE_BLENDFACTOR_ONE;
+   blend.rt[0].rgb_dst_factor = PIPE_BLENDFACTOR_ZERO;
+   blend.rt[0].alpha_dst_factor = PIPE_BLENDFACTOR_ZERO;
 
    cso_set_blend(ctx->cso_context, &blend);
 }
diff --git a/src/gallium/state_trackers/vega/polygon.c b/src/gallium/state_trackers/vega/polygon.c
index d385ee567f..f5352ee84e 100644
--- a/src/gallium/state_trackers/vega/polygon.c
+++ b/src/gallium/state_trackers/vega/polygon.c
@@ -248,12 +248,12 @@ VGboolean polygon_is_closed(struct polygon *p)
 static void set_blend_for_fill(struct pipe_blend_state *blend)
 {
    memset(blend, 0, sizeof(struct pipe_blend_state));
-   blend->colormask = 0; /*disable colorwrites*/
+   blend->rt[0].colormask = 0; /*disable colorwrites*/
 
-   blend->rgb_src_factor = PIPE_BLENDFACTOR_ONE;
-   blend->alpha_src_factor = PIPE_BLENDFACTOR_ONE;
-   blend->rgb_dst_factor = PIPE_BLENDFACTOR_INV_SRC_ALPHA;
-   blend->alpha_dst_factor = PIPE_BLENDFACTOR_INV_SRC_ALPHA;
+   blend->rt[0].rgb_src_factor = PIPE_BLENDFACTOR_ONE;
+   blend->rt[0].alpha_src_factor = PIPE_BLENDFACTOR_ONE;
+   blend->rt[0].rgb_dst_factor = PIPE_BLENDFACTOR_INV_SRC_ALPHA;
+   blend->rt[0].alpha_dst_factor = PIPE_BLENDFACTOR_INV_SRC_ALPHA;
 }
 
 static void draw_polygon(struct vg_context *ctx,
diff --git a/src/gallium/state_trackers/vega/renderer.c b/src/gallium/state_trackers/vega/renderer.c
index 64e3a7c545..47e18dc1e2 100644
--- a/src/gallium/state_trackers/vega/renderer.c
+++ b/src/gallium/state_trackers/vega/renderer.c
@@ -317,11 +317,11 @@ void renderer_copy_texture(struct renderer *ctx,
    {
       struct pipe_blend_state blend;
       memset(&blend, 0, sizeof(blend));
-      blend.rgb_src_factor = PIPE_BLENDFACTOR_ONE;
-      blend.alpha_src_factor = PIPE_BLENDFACTOR_ONE;
-      blend.rgb_dst_factor = PIPE_BLENDFACTOR_ZERO;
-      blend.alpha_dst_factor = PIPE_BLENDFACTOR_ZERO;
-      blend.colormask = PIPE_MASK_RGBA;
+      blend.rt[0].rgb_src_factor = PIPE_BLENDFACTOR_ONE;
+      blend.rt[0].alpha_src_factor = PIPE_BLENDFACTOR_ONE;
+      blend.rt[0].rgb_dst_factor = PIPE_BLENDFACTOR_ZERO;
+      blend.rt[0].alpha_dst_factor = PIPE_BLENDFACTOR_ZERO;
+      blend.rt[0].colormask = PIPE_MASK_RGBA;
       cso_set_blend(ctx->cso, &blend);
    }
 
@@ -486,11 +486,11 @@ void renderer_copy_surface(struct renderer *ctx,
    {
       struct pipe_blend_state blend;
       memset(&blend, 0, sizeof(blend));
-      blend.rgb_src_factor = PIPE_BLENDFACTOR_ONE;
-      blend.alpha_src_factor = PIPE_BLENDFACTOR_ONE;
-      blend.rgb_dst_factor = PIPE_BLENDFACTOR_ZERO;
-      blend.alpha_dst_factor = PIPE_BLENDFACTOR_ZERO;
-      blend.colormask = PIPE_MASK_RGBA;
+      blend.rt[0].rgb_src_factor = PIPE_BLENDFACTOR_ONE;
+      blend.rt[0].alpha_src_factor = PIPE_BLENDFACTOR_ONE;
+      blend.rt[0].rgb_dst_factor = PIPE_BLENDFACTOR_ZERO;
+      blend.rt[0].alpha_dst_factor = PIPE_BLENDFACTOR_ZERO;
+      blend.rt[0].colormask = PIPE_MASK_RGBA;
       cso_set_blend(ctx->cso, &blend);
    }
 
diff --git a/src/gallium/state_trackers/vega/shader.c b/src/gallium/state_trackers/vega/shader.c
index bd5ae79e55..8e59d53dc7 100644
--- a/src/gallium/state_trackers/vega/shader.c
+++ b/src/gallium/state_trackers/vega/shader.c
@@ -135,8 +135,8 @@ static VGint blend_bind_samplers(struct vg_context *ctx,
       textures[2] = stfb->blend_texture;
 
       if (!samplers[0] || !textures[0]) {
-         samplers[1] = samplers[2];
-         textures[1] = textures[2];
+         samplers[0] = samplers[2];
+         textures[0] = textures[2];
       }
       if (!samplers[1] || !textures[1]) {
          samplers[1] = samplers[0];
diff --git a/src/gallium/state_trackers/vega/vg_context.c b/src/gallium/state_trackers/vega/vg_context.c
index c16ac036e3..64207d97ba 100644
--- a/src/gallium/state_trackers/vega/vg_context.c
+++ b/src/gallium/state_trackers/vega/vg_context.c
@@ -252,7 +252,7 @@ static void update_clip_state(struct vg_context *ctx)
       ctx->pipe->clear(ctx->pipe, PIPE_CLEAR_DEPTHSTENCIL, NULL, 1.0, 0);
 
       /* disable color writes */
-      blend->colormask = 0; /*disable colorwrites*/
+      blend->rt[0].colormask = 0; /*disable colorwrites*/
       cso_set_blend(ctx->cso_context, blend);
 
       /* enable scissoring */
@@ -286,7 +286,6 @@ static void update_clip_state(struct vg_context *ctx)
          renderer_draw_quad(ctx->renderer, minx, miny, maxx, maxy, 0.0f);
       }
 
-      blend->colormask = 1; /*enable colorwrites*/
       cso_restore_blend(ctx->cso_context);
       cso_restore_fragment_shader(ctx->cso_context);
 
@@ -301,57 +300,56 @@ void vg_validate_state(struct vg_context *ctx)
    if ((ctx->state.dirty & BLEND_DIRTY)) {
       struct pipe_blend_state *blend = &ctx->state.g3d.blend;
       memset(blend, 0, sizeof(struct pipe_blend_state));
-      blend->blend_enable = 1;
-      blend->colormask |= PIPE_MASK_R;
-      blend->colormask |= PIPE_MASK_G;
-      blend->colormask |= PIPE_MASK_B;
-      blend->colormask |= PIPE_MASK_A;
+      blend->rt[0].blend_enable = 1;
+      blend->rt[0].colormask = PIPE_MASK_RGBA;
 
       switch (ctx->state.vg.blend_mode) {
       case VG_BLEND_SRC:
-         blend->rgb_src_factor   = PIPE_BLENDFACTOR_ONE;
-         blend->alpha_src_factor = PIPE_BLENDFACTOR_ONE;
-         blend->rgb_dst_factor   = PIPE_BLENDFACTOR_ZERO;
-         blend->alpha_dst_factor = PIPE_BLENDFACTOR_ZERO;
+         blend->rt[0].rgb_src_factor   = PIPE_BLENDFACTOR_ONE;
+         blend->rt[0].alpha_src_factor = PIPE_BLENDFACTOR_ONE;
+         blend->rt[0].rgb_dst_factor   = PIPE_BLENDFACTOR_ZERO;
+         blend->rt[0].alpha_dst_factor = PIPE_BLENDFACTOR_ZERO;
+         blend->rt[0].blend_enable = 0;
          break;
       case VG_BLEND_SRC_OVER:
-         blend->rgb_src_factor   = PIPE_BLENDFACTOR_SRC_ALPHA;
-         blend->alpha_src_factor = PIPE_BLENDFACTOR_ONE;
-         blend->rgb_dst_factor   = PIPE_BLENDFACTOR_INV_SRC_ALPHA;
-         blend->alpha_dst_factor = PIPE_BLENDFACTOR_INV_SRC_ALPHA;
+         blend->rt[0].rgb_src_factor   = PIPE_BLENDFACTOR_SRC_ALPHA;
+         blend->rt[0].alpha_src_factor = PIPE_BLENDFACTOR_ONE;
+         blend->rt[0].rgb_dst_factor   = PIPE_BLENDFACTOR_INV_SRC_ALPHA;
+         blend->rt[0].alpha_dst_factor = PIPE_BLENDFACTOR_INV_SRC_ALPHA;
          break;
       case VG_BLEND_DST_OVER:
-         blend->rgb_src_factor   = PIPE_BLENDFACTOR_INV_DST_ALPHA;
-         blend->alpha_src_factor = PIPE_BLENDFACTOR_INV_DST_ALPHA;
-         blend->rgb_dst_factor   = PIPE_BLENDFACTOR_DST_ALPHA;
-         blend->alpha_dst_factor = PIPE_BLENDFACTOR_DST_ALPHA;
+         blend->rt[0].rgb_src_factor   = PIPE_BLENDFACTOR_INV_DST_ALPHA;
+         blend->rt[0].alpha_src_factor = PIPE_BLENDFACTOR_INV_DST_ALPHA;
+         blend->rt[0].rgb_dst_factor   = PIPE_BLENDFACTOR_DST_ALPHA;
+         blend->rt[0].alpha_dst_factor = PIPE_BLENDFACTOR_DST_ALPHA;
          break;
       case VG_BLEND_SRC_IN:
-         blend->rgb_src_factor   = PIPE_BLENDFACTOR_DST_ALPHA;
-         blend->alpha_src_factor = PIPE_BLENDFACTOR_DST_ALPHA;
-         blend->rgb_dst_factor   = PIPE_BLENDFACTOR_ZERO;
-         blend->alpha_dst_factor = PIPE_BLENDFACTOR_ZERO;
+         blend->rt[0].rgb_src_factor   = PIPE_BLENDFACTOR_DST_ALPHA;
+         blend->rt[0].alpha_src_factor = PIPE_BLENDFACTOR_DST_ALPHA;
+         blend->rt[0].rgb_dst_factor   = PIPE_BLENDFACTOR_ZERO;
+         blend->rt[0].alpha_dst_factor = PIPE_BLENDFACTOR_ZERO;
          break;
       case VG_BLEND_DST_IN:
-         blend->rgb_src_factor   = PIPE_BLENDFACTOR_ZERO;
-         blend->alpha_src_factor = PIPE_BLENDFACTOR_ZERO;
-         blend->rgb_dst_factor   = PIPE_BLENDFACTOR_SRC_ALPHA;
-         blend->alpha_dst_factor = PIPE_BLENDFACTOR_SRC_ALPHA;
+         blend->rt[0].rgb_src_factor   = PIPE_BLENDFACTOR_ZERO;
+         blend->rt[0].alpha_src_factor = PIPE_BLENDFACTOR_ZERO;
+         blend->rt[0].rgb_dst_factor   = PIPE_BLENDFACTOR_SRC_ALPHA;
+         blend->rt[0].alpha_dst_factor = PIPE_BLENDFACTOR_SRC_ALPHA;
          break;
       case VG_BLEND_MULTIPLY:
       case VG_BLEND_SCREEN:
       case VG_BLEND_DARKEN:
       case VG_BLEND_LIGHTEN:
-         blend->rgb_src_factor   = PIPE_BLENDFACTOR_ONE;
-         blend->alpha_src_factor = PIPE_BLENDFACTOR_ONE;
-         blend->rgb_dst_factor   = PIPE_BLENDFACTOR_ZERO;
-         blend->alpha_dst_factor = PIPE_BLENDFACTOR_ZERO;
+         blend->rt[0].rgb_src_factor   = PIPE_BLENDFACTOR_ONE;
+         blend->rt[0].alpha_src_factor = PIPE_BLENDFACTOR_ONE;
+         blend->rt[0].rgb_dst_factor   = PIPE_BLENDFACTOR_ZERO;
+         blend->rt[0].alpha_dst_factor = PIPE_BLENDFACTOR_ZERO;
+         blend->rt[0].blend_enable = 0;
          break;
       case VG_BLEND_ADDITIVE:
-         blend->rgb_src_factor   = PIPE_BLENDFACTOR_ONE;
-         blend->alpha_src_factor = PIPE_BLENDFACTOR_ONE;
-         blend->rgb_dst_factor   = PIPE_BLENDFACTOR_ONE;
-         blend->alpha_dst_factor = PIPE_BLENDFACTOR_ONE;
+         blend->rt[0].rgb_src_factor   = PIPE_BLENDFACTOR_ONE;
+         blend->rt[0].alpha_src_factor = PIPE_BLENDFACTOR_ONE;
+         blend->rt[0].rgb_dst_factor   = PIPE_BLENDFACTOR_ONE;
+         blend->rt[0].alpha_dst_factor = PIPE_BLENDFACTOR_ONE;
          break;
       default:
          assert(!"not implemented blend mode");
diff --git a/src/gallium/state_trackers/vega/vg_context.h b/src/gallium/state_trackers/vega/vg_context.h
index ff8c1a0421..bc88c8d139 100644
--- a/src/gallium/state_trackers/vega/vg_context.h
+++ b/src/gallium/state_trackers/vega/vg_context.h
@@ -50,7 +50,7 @@ struct st_renderbuffer {
 };
 
 struct st_framebuffer {
-   VGint init_width, init_height;
+   VGint width, height;
    struct st_renderbuffer *strb;
    struct st_renderbuffer *dsrb;
 
diff --git a/src/gallium/state_trackers/vega/vg_tracker.c b/src/gallium/state_trackers/vega/vg_tracker.c
index ff80aab03a..617c174eb6 100644
--- a/src/gallium/state_trackers/vega/vg_tracker.c
+++ b/src/gallium/state_trackers/vega/vg_tracker.c
@@ -193,8 +193,8 @@ struct st_framebuffer * st_create_framebuffer(const void *visual,
       */
       stfb->alpha_mask = 0;
 
-      stfb->init_width = width;
-      stfb->init_height = height;
+      stfb->width = width;
+      stfb->height = height;
       stfb->privateData = privateData;
    }
 
@@ -282,11 +282,14 @@ void st_resize_framebuffer(struct st_framebuffer *stfb,
 
    /* If this is a noop, exit early and don't do the clear, etc below.
     */
-   if (strb->width == width &&
-       strb->height == height &&
+   if (stfb->width == width &&
+       stfb->height == height &&
        state->zsbuf)
       return;
 
+   stfb->width = width;
+   stfb->height = height;
+
    if (strb->width != width || strb->height != height)
       st_renderbuffer_alloc_storage(ctx, strb,
                                  width, height);
diff --git a/src/gallium/state_trackers/wgl/stw_pixelformat.c b/src/gallium/state_trackers/wgl/stw_pixelformat.c
index 54cc361412..7d4c2430b0 100644
--- a/src/gallium/state_trackers/wgl/stw_pixelformat.c
+++ b/src/gallium/state_trackers/wgl/stw_pixelformat.c
@@ -95,8 +95,6 @@ stw_pf_depth_stencil[] = {
    { PIPE_FORMAT_Z24X8_UNORM, {24, 0} },
    { PIPE_FORMAT_X8Z24_UNORM, {24, 0} },
    { PIPE_FORMAT_Z16_UNORM,   {16, 0} },
-   /* pure stencil */
-   { PIPE_FORMAT_S8_UNORM,    { 0, 8} },
    /* combined depth-stencil */
    { PIPE_FORMAT_S8Z24_UNORM, {24, 8} },
    { PIPE_FORMAT_Z24S8_UNORM, {24, 8} }
@@ -220,7 +218,8 @@ stw_pixelformat_init( void )
          const struct stw_pf_color_info *color = &stw_pf_color[j];
          
          if(!screen->is_format_supported(screen, color->format, PIPE_TEXTURE_2D, 
-                                         PIPE_TEXTURE_USAGE_RENDER_TARGET, 0))
+                                         PIPE_TEXTURE_USAGE_RENDER_TARGET |
+                                         PIPE_TEXTURE_USAGE_DISPLAY_TARGET, 0))
             continue;
          
          for(k = 0; k < Elements(stw_pf_doublebuffer); ++k) {
diff --git a/src/gallium/state_trackers/xorg/xorg_composite.c b/src/gallium/state_trackers/xorg/xorg_composite.c
index 1c248a629e..c50873c150 100644
--- a/src/gallium/state_trackers/xorg/xorg_composite.c
+++ b/src/gallium/state_trackers/xorg/xorg_composite.c
@@ -4,10 +4,7 @@
 #include "xorg_exa_tgsi.h"
 
 #include "cso_cache/cso_context.h"
-#include "util/u_draw_quad.h"
-#include "util/u_math.h"
 
-#include "pipe/p_inlines.h"
 
 /*XXX also in Xrender.h but the including it here breaks compilition */
 #define XFixedToDouble(f)    (((double) (f)) / 65536.)
@@ -220,13 +217,13 @@ bind_blend_state(struct exa_context *exa, int op,
    blend_for_op(&blend_opt, op, pSrcPicture, pMaskPicture, pDstPicture);
 
    memset(&blend, 0, sizeof(struct pipe_blend_state));
-   blend.blend_enable = 1;
-   blend.colormask |= PIPE_MASK_RGBA;
+   blend.rt[0].blend_enable = 1;
+   blend.rt[0].colormask = PIPE_MASK_RGBA;
 
-   blend.rgb_src_factor   = blend_opt.rgb_src;
-   blend.alpha_src_factor = blend_opt.rgb_src;
-   blend.rgb_dst_factor   = blend_opt.rgb_dst;
-   blend.alpha_dst_factor = blend_opt.rgb_dst;
+   blend.rt[0].rgb_src_factor   = blend_opt.rgb_src;
+   blend.rt[0].alpha_src_factor = blend_opt.rgb_src;
+   blend.rt[0].rgb_dst_factor   = blend_opt.rgb_dst;
+   blend.rt[0].alpha_dst_factor = blend_opt.rgb_dst;
 
    cso_set_blend(exa->renderer->cso, &blend);
 }
diff --git a/src/gallium/state_trackers/xorg/xorg_crtc.c b/src/gallium/state_trackers/xorg/xorg_crtc.c
index 650d2c0d1d..2395d54975 100644
--- a/src/gallium/state_trackers/xorg/xorg_crtc.c
+++ b/src/gallium/state_trackers/xorg/xorg_crtc.c
@@ -50,7 +50,6 @@
 #endif
 
 #include "pipe/p_inlines.h"
-#include "util/u_format.h"
 #include "util/u_rect.h"
 
 #ifdef HAVE_LIBKMS
@@ -243,7 +242,11 @@ crtc_load_cursor_argb_kms(xf86CrtcPtr crtc, CARD32 * image)
 	unsigned attr[8];
 
 	attr[0] = KMS_BO_TYPE;
+#ifdef KMS_BO_TYPE_CURSOR_64X64_A8R8G8B8
+	attr[1] = KMS_BO_TYPE_CURSOR_64X64_A8R8G8B8;
+#else
 	attr[1] = KMS_BO_TYPE_CURSOR;
+#endif
 	attr[2] = KMS_WIDTH;
 	attr[3] = 64;
 	attr[4] = KMS_HEIGHT;
diff --git a/src/gallium/state_trackers/xorg/xorg_dri2.c b/src/gallium/state_trackers/xorg/xorg_dri2.c
index fd82f4fa1d..59588f0ff7 100644
--- a/src/gallium/state_trackers/xorg/xorg_dri2.c
+++ b/src/gallium/state_trackers/xorg/xorg_dri2.c
@@ -41,12 +41,14 @@
 #include "pipe/p_inlines.h"
 
 #include "util/u_format.h"
-#include "util/u_rect.h"
 
 /* Make all the #if cases in the code esier to read */
-/* XXX can it be set to 1? */
 #ifndef DRI2INFOREC_VERSION
-#define DRI2INFOREC_VERSION 0
+#define DRI2INFOREC_VERSION 1
+#endif
+
+#if DRI2INFOREC_VERSION == 2
+static Bool set_format_in_do_create_buffer;
 #endif
 
 typedef struct {
@@ -147,7 +149,9 @@ dri2_do_create_buffer(DrawablePtr pDraw, DRI2BufferPtr buffer, unsigned int form
     buffer->driverPrivate = private;
     buffer->flags = 0; /* not tiled */
 #if DRI2INFOREC_VERSION == 2
-    ((DRI2Buffer2Ptr)buffer)->format = 0;
+    /* ABI forwards/backwards compatibility */
+    if (set_format_in_do_create_buffer)
+	((DRI2Buffer2Ptr)buffer)->format = 0;
 #elif DRI2INFOREC_VERSION >= 3
     buffer->format = 0;
 #endif
@@ -211,7 +215,9 @@ dri2_destroy_buffer(DrawablePtr pDraw, DRI2Buffer2Ptr buffer)
     xfree(buffer);
 }
 
-#else /* DRI2INFOREC_VERSION < 2 */
+#endif /* DRI2INFOREC_VERSION >= 2 */
+
+#if DRI2INFOREC_VERSION <= 2
 
 static DRI2BufferPtr
 dri2_create_buffers(DrawablePtr pDraw, unsigned int *attachments, int count)
@@ -261,7 +267,7 @@ dri2_destroy_buffers(DrawablePtr pDraw, DRI2BufferPtr buffers, int count)
     }
 }
 
-#endif /* DRI2INFOREC_VERSION >= 2 */
+#endif /* DRI2INFOREC_VERSION <= 2 */
 
 static void
 dri2_copy_region(DrawablePtr pDraw, RegionPtr pRegion,
@@ -369,12 +375,19 @@ xorg_dri2_init(ScreenPtr pScreen)
     ScrnInfoPtr pScrn = xf86Screens[pScreen->myNum];
     modesettingPtr ms = modesettingPTR(pScrn);
     DRI2InfoRec dri2info;
-
 #if DRI2INFOREC_VERSION >= 2
-    dri2info.version = DRI2INFOREC_VERSION;
-#else
-    dri2info.version = 1;
+    int major, minor;
+
+    if (xf86LoaderCheckSymbol("DRI2Version")) {
+	DRI2Version(&major, &minor);
+    } else {
+	/* Assume version 1.0 */
+	major = 1;
+	minor = 0;
+    }
 #endif
+
+    dri2info.version = DRI2INFOREC_VERSION;
     dri2info.fd = ms->fd;
 
     dri2info.driverName = pScrn->driverName;
@@ -383,7 +396,22 @@ xorg_dri2_init(ScreenPtr pScreen)
 #if DRI2INFOREC_VERSION >= 2
     dri2info.CreateBuffer = dri2_create_buffer;
     dri2info.DestroyBuffer = dri2_destroy_buffer;
-#else
+#endif
+
+    /* For X servers in the 1.6.x series there where two DRI2 version.
+     * This allows us to build one binary that works on both servers.
+     */
+#if DRI2INFOREC_VERSION == 2
+    if (minor == 0) {
+	set_format_in_do_create_buffer = FALSE;
+	dri2info.CreateBuffers = dri2_create_buffers;
+	dri2info.DestroyBuffers = dri2_destroy_buffers;
+    } else
+	set_format_in_do_create_buffer = FALSE;
+#endif
+
+    /* For version 1 set these unconditionaly. */
+#if DRI2INFOREC_VERSION == 1
     dri2info.CreateBuffers = dri2_create_buffers;
     dri2info.DestroyBuffers = dri2_destroy_buffers;
 #endif
diff --git a/src/gallium/state_trackers/xorg/xorg_driver.c b/src/gallium/state_trackers/xorg/xorg_driver.c
index b02fe68f31..f53a879a14 100644
--- a/src/gallium/state_trackers/xorg/xorg_driver.c
+++ b/src/gallium/state_trackers/xorg/xorg_driver.c
@@ -45,7 +45,6 @@
 #include "miscstruct.h"
 #include "dixstruct.h"
 #include "xf86xv.h"
-#include <X11/extensions/Xv.h>
 #ifndef XSERVER_LIBPCIACCESS
 #error "libpciaccess needed"
 #endif
@@ -79,11 +78,13 @@ typedef enum
 {
     OPTION_SW_CURSOR,
     OPTION_2D_ACCEL,
+    OPTION_DEBUG_FALLBACK,
 } drv_option_enums;
 
 static const OptionInfoRec drv_options[] = {
     {OPTION_SW_CURSOR, "SWcursor", OPTV_BOOLEAN, {0}, FALSE},
     {OPTION_2D_ACCEL, "2DAccel", OPTV_BOOLEAN, {0}, FALSE},
+    {OPTION_DEBUG_FALLBACK, "DebugFallback", OPTV_BOOLEAN, {0}, FALSE},
     {-1, NULL, OPTV_NONE, {0}, FALSE}
 };
 
@@ -111,6 +112,28 @@ xorg_tracker_set_functions(ScrnInfoPtr scrn)
     scrn->ValidMode = drv_valid_mode;
 }
 
+Bool
+xorg_tracker_have_modesetting(ScrnInfoPtr pScrn, struct pci_device *device)
+{
+    char *BusID = xalloc(64);
+    sprintf(BusID, "pci:%04x:%02x:%02x.%d",
+	    device->domain, device->bus,
+	    device->dev, device->func);
+
+    if (drmCheckModesettingSupported(BusID)) {
+	xf86DrvMsgVerb(pScrn->scrnIndex, X_INFO, 0,
+		       "Drm modesetting not supported %s\n", BusID);
+	xfree(BusID);
+	return FALSE;
+    }
+
+    xf86DrvMsgVerb(pScrn->scrnIndex, X_INFO, 0,
+		   "Drm modesetting supported on %s\n", BusID);
+
+    xfree(BusID);
+    return TRUE;
+}
+
 
 /*
  * Internal function definitions
@@ -206,16 +229,41 @@ drv_init_drm(ScrnInfoPtr pScrn)
 		ms->PciInfo->dev, ms->PciInfo->func
 	    );
 
-	ms->fd = drmOpen(NULL, BusID);
 
-	if (ms->fd < 0)
-	    return FALSE;
+	ms->api = drm_api_create();
+	ms->fd = drmOpen(ms->api ? ms->api->driver_name : NULL, BusID);
+	xfree(BusID);
+
+	if (ms->fd >= 0)
+	    return TRUE;
+
+	if (ms->api && ms->api->destroy)
+	    ms->api->destroy(ms->api);
+
+	ms->api = NULL;
+
+	return FALSE;
     }
 
     return TRUE;
 }
 
 static Bool
+drv_close_drm(ScrnInfoPtr pScrn)
+{
+    modesettingPtr ms = modesettingPTR(pScrn);
+
+    if (ms->api && ms->api->destroy)
+	ms->api->destroy(ms->api);
+    ms->api = NULL;
+
+    drmClose(ms->fd);
+    ms->fd = -1;
+
+    return TRUE;
+}
+
+static Bool
 drv_init_resource_management(ScrnInfoPtr pScrn)
 {
     modesettingPtr ms = modesettingPTR(pScrn);
@@ -229,7 +277,6 @@ drv_init_resource_management(ScrnInfoPtr pScrn)
     if (ms->screen || ms->kms)
 	return TRUE;
 
-    ms->api = drm_api_create();
     if (ms->api) {
 	ms->screen = ms->api->create_screen(ms->api, ms->fd, NULL);
 
@@ -269,10 +316,6 @@ drv_close_resource_management(ScrnInfoPtr pScrn)
     }
     ms->screen = NULL;
 
-    if (ms->api && ms->api->destroy)
-	ms->api->destroy(ms->api);
-    ms->api = NULL;
-
 #ifdef HAVE_LIBKMS
     if (ms->kms)
 	kms_destroy(&ms->kms);
@@ -629,10 +672,11 @@ drv_screen_init(int scrnIndex, ScreenPtr pScreen, int argc, char **argv)
 
     xf86SetBlackWhitePixels(pScreen);
 
+    ms->accelerate_2d = xf86ReturnOptValBool(ms->Options, OPTION_2D_ACCEL, FALSE);
+    ms->debug_fallback = xf86ReturnOptValBool(ms->Options, OPTION_DEBUG_FALLBACK, TRUE);
+
     if (ms->screen) {
-	ms->exa = xorg_exa_init(pScrn, xf86ReturnOptValBool(ms->Options,
-							    OPTION_2D_ACCEL, TRUE));
-	ms->debug_fallback = debug_get_bool_option("XORG_DEBUG_FALLBACK", TRUE);
+	ms->exa = xorg_exa_init(pScrn, ms->accelerate_2d);
 
 	xorg_xv_init(pScreen);
 #ifdef DRI2
@@ -640,6 +684,17 @@ drv_screen_init(int scrnIndex, ScreenPtr pScreen, int argc, char **argv)
 #endif
     }
 
+    xf86DrvMsg(pScrn->scrnIndex, X_INFO, "2D Acceleration is %s\n",
+	       ms->screen && ms->accelerate_2d ? "enabled" : "disabled");
+    xf86DrvMsg(pScrn->scrnIndex, X_INFO, "Fallback debugging is %s\n",
+	       ms->debug_fallback ? "enabled" : "disabled");
+#ifdef DRI2
+    xf86DrvMsg(pScrn->scrnIndex, X_INFO, "3D Acceleration is %s\n",
+	       ms->screen ? "enabled" : "disabled");
+#else
+    xf86DrvMsg(pScrn->scrnIndex, X_INFO, "3D Acceleration is disabled\n");
+#endif
+
     miInitializeBackingStore(pScreen);
     xf86SetBackingStore(pScreen);
     xf86SetSilkenMouse(pScreen);
@@ -823,8 +878,7 @@ drv_close_screen(int scrnIndex, ScreenPtr pScreen)
 
     drv_close_resource_management(pScrn);
 
-    drmClose(ms->fd);
-    ms->fd = -1;
+    drv_close_drm(pScrn);
 
     pScrn->vtSema = FALSE;
     pScreen->CloseScreen = ms->CloseScreen;
@@ -954,7 +1008,11 @@ drv_create_front_buffer_kms(ScrnInfoPtr pScrn)
     int ret;
 
     attr[0] = KMS_BO_TYPE;
+#ifdef KMS_BO_TYPE_SCANOUT_X8R8G8B8
+    attr[1] = KMS_BO_TYPE_SCANOUT_X8R8G8B8;
+#else
     attr[1] = KMS_BO_TYPE_SCANOUT;
+#endif
     attr[2] = KMS_WIDTH;
     attr[3] = pScrn->virtualX;
     attr[4] = KMS_HEIGHT;
@@ -1012,12 +1070,22 @@ drv_bind_front_buffer_kms(ScrnInfoPtr pScrn)
 	goto err_destroy;
 
     pScreen->ModifyPixmapHeader(rootPixmap,
-				pScreen->width,
-				pScreen->height,
+				pScrn->virtualX,
+				pScrn->virtualY,
 				pScreen->rootDepth,
 				pScrn->bitsPerPixel,
 				stride,
 				ptr);
+
+    /* This a hack to work around EnableDisableFBAccess setting the pointer
+     * the real fix would be to replace pScrn->EnableDisableFBAccess hook
+     * and set the rootPixmap->devPrivate.ptr to something valid before that.
+     *
+     * But in its infinit visdome something uses either this some times before
+     * that, so our hook doesn't get called before the crash happens.
+     */
+    pScrn->pixmapPrivate.ptr = ptr;
+
     return TRUE;
 
 err_destroy:
diff --git a/src/gallium/state_trackers/xorg/xorg_exa.c b/src/gallium/state_trackers/xorg/xorg_exa.c
index d9432babf1..70af0c5fed 100644
--- a/src/gallium/state_trackers/xorg/xorg_exa.c
+++ b/src/gallium/state_trackers/xorg/xorg_exa.c
@@ -41,9 +41,7 @@
 #include "pipe/p_format.h"
 #include "pipe/p_context.h"
 #include "pipe/p_state.h"
-#include "pipe/p_inlines.h"
 
-#include "util/u_format.h"
 #include "util/u_rect.h"
 #include "util/u_math.h"
 #include "util/u_debug.h"
diff --git a/src/gallium/state_trackers/xorg/xorg_exa_tgsi.c b/src/gallium/state_trackers/xorg/xorg_exa_tgsi.c
index bed17caab7..3e5e6bd6a6 100644
--- a/src/gallium/state_trackers/xorg/xorg_exa_tgsi.c
+++ b/src/gallium/state_trackers/xorg/xorg_exa_tgsi.c
@@ -6,11 +6,9 @@
 #include "pipe/p_format.h"
 #include "pipe/p_context.h"
 #include "pipe/p_state.h"
-#include "pipe/p_inlines.h"
 #include "pipe/p_shader_tokens.h"
 
 #include "util/u_memory.h"
-#include "util/u_simple_shaders.h"
 
 #include "tgsi/tgsi_ureg.h"
 
diff --git a/src/gallium/state_trackers/xorg/xorg_output.c b/src/gallium/state_trackers/xorg/xorg_output.c
index 251f331ea7..13c3fb97e3 100644
--- a/src/gallium/state_trackers/xorg/xorg_output.c
+++ b/src/gallium/state_trackers/xorg/xorg_output.c
@@ -49,8 +49,6 @@
 #include <X11/extensions/dpms.h>
 #endif
 
-#include "X11/Xatom.h"
-
 #include "xorg_tracker.h"
 
 static char *output_enum_list[] = {
diff --git a/src/gallium/state_trackers/xorg/xorg_renderer.c b/src/gallium/state_trackers/xorg/xorg_renderer.c
index 8f729b565b..a3c3993ab8 100644
--- a/src/gallium/state_trackers/xorg/xorg_renderer.c
+++ b/src/gallium/state_trackers/xorg/xorg_renderer.c
@@ -5,7 +5,6 @@
 
 #include "cso_cache/cso_context.h"
 #include "util/u_draw_quad.h"
-#include "util/u_format.h"
 #include "util/u_math.h"
 #include "util/u_memory.h"
 #include "util/u_rect.h"
@@ -445,11 +444,11 @@ void renderer_copy_prepare(struct xorg_renderer *r,
    {
       struct pipe_blend_state blend;
       memset(&blend, 0, sizeof(blend));
-      blend.rgb_src_factor = PIPE_BLENDFACTOR_ONE;
-      blend.alpha_src_factor = PIPE_BLENDFACTOR_ONE;
-      blend.rgb_dst_factor = PIPE_BLENDFACTOR_ZERO;
-      blend.alpha_dst_factor = PIPE_BLENDFACTOR_ZERO;
-      blend.colormask = PIPE_MASK_RGBA;
+      blend.rt[0].rgb_src_factor = PIPE_BLENDFACTOR_ONE;
+      blend.rt[0].alpha_src_factor = PIPE_BLENDFACTOR_ONE;
+      blend.rt[0].rgb_dst_factor = PIPE_BLENDFACTOR_ZERO;
+      blend.rt[0].alpha_dst_factor = PIPE_BLENDFACTOR_ZERO;
+      blend.rt[0].colormask = PIPE_MASK_RGBA;
       cso_set_blend(r->cso, &blend);
    }
 
diff --git a/src/gallium/state_trackers/xorg/xorg_tracker.h b/src/gallium/state_trackers/xorg/xorg_tracker.h
index 4d5d4780dc..a88b1d520d 100644
--- a/src/gallium/state_trackers/xorg/xorg_tracker.h
+++ b/src/gallium/state_trackers/xorg/xorg_tracker.h
@@ -112,6 +112,7 @@ typedef struct _modesettingRec
     /* exa */
     struct exa_context *exa;
     Bool noEvict;
+    Bool accelerate_2d;
     Bool debug_fallback;
 
     /* winsys hocks */
diff --git a/src/gallium/state_trackers/xorg/xorg_winsys.h b/src/gallium/state_trackers/xorg/xorg_winsys.h
index 47ee4b9ffd..865733bca2 100644
--- a/src/gallium/state_trackers/xorg/xorg_winsys.h
+++ b/src/gallium/state_trackers/xorg/xorg_winsys.h
@@ -45,5 +45,6 @@
 
 void xorg_tracker_set_functions(ScrnInfoPtr scrn);
 const OptionInfoRec * xorg_tracker_available_options(int chipid, int busid);
+Bool xorg_tracker_have_modesetting(ScrnInfoPtr pScrn, struct pci_device *device);
 
 #endif
diff --git a/src/gallium/state_trackers/xorg/xorg_xv.c b/src/gallium/state_trackers/xorg/xorg_xv.c
index 5bf0e94b62..3dcef22c13 100644
--- a/src/gallium/state_trackers/xorg/xorg_xv.c
+++ b/src/gallium/state_trackers/xorg/xorg_xv.c
@@ -11,9 +11,6 @@
 #include "cso_cache/cso_context.h"
 
 #include "pipe/p_screen.h"
-#include "pipe/p_inlines.h"
-
-#include "util/u_format.h"
 
 /*XXX get these from pipe's texture limits */
 #define IMAGE_MAX_WIDTH		2048
@@ -403,14 +400,14 @@ bind_blend_state(struct xorg_xv_port_priv *port)
    struct pipe_blend_state blend;
 
    memset(&blend, 0, sizeof(struct pipe_blend_state));
-   blend.blend_enable = 1;
-   blend.colormask |= PIPE_MASK_RGBA;
+   blend.rt[0].blend_enable = 0;
+   blend.rt[0].colormask = PIPE_MASK_RGBA;
 
    /* porter&duff src */
-   blend.rgb_src_factor   = PIPE_BLENDFACTOR_ONE;
-   blend.alpha_src_factor = PIPE_BLENDFACTOR_ONE;
-   blend.rgb_dst_factor   = PIPE_BLENDFACTOR_ZERO;
-   blend.alpha_dst_factor = PIPE_BLENDFACTOR_ZERO;
+   blend.rt[0].rgb_src_factor   = PIPE_BLENDFACTOR_ONE;
+   blend.rt[0].alpha_src_factor = PIPE_BLENDFACTOR_ONE;
+   blend.rt[0].rgb_dst_factor   = PIPE_BLENDFACTOR_ZERO;
+   blend.rt[0].alpha_dst_factor = PIPE_BLENDFACTOR_ZERO;
 
    cso_set_blend(port->r->cso, &blend);
 }
diff --git a/src/gallium/winsys/drm/Makefile.egl_g3d b/src/gallium/winsys/drm/Makefile.egl
index 3ce2725852..b1f2038550 100644
--- a/src/gallium/winsys/drm/Makefile.egl_g3d
+++ b/src/gallium/winsys/drm/Makefile.egl
@@ -1,4 +1,4 @@
-# src/gallium/winsys/drm/Makefile.egl_g3d
+# src/gallium/winsys/drm/Makefile.egl
 
 # The driver Makefile should define
 #
@@ -13,10 +13,10 @@ EGL_DRIVER_OBJECTS = $(EGL_DRIVER_SOURCES:.c=.o)
 
 common_LIBS = -ldrm -lm -ldl
 
-x11_ST = $(TOP)/src/gallium/state_trackers/egl_g3d/libeglx11.a
+x11_ST = $(TOP)/src/gallium/state_trackers/egl/libeglx11.a
 x11_LIBS = $(common_LIBS) -lX11 -lXext -lXfixes
 
-kms_ST = $(TOP)/src/gallium/state_trackers/egl_g3d/libeglkms.a
+kms_ST = $(TOP)/src/gallium/state_trackers/egl/libeglkms.a
 kms_LIBS = $(common_LIBS)
 
 ##### RULES #####
@@ -29,17 +29,16 @@ kms_LIBS = $(common_LIBS)
 
 EGL_DISPLAY_DRIVERS = $(foreach dpy, $(EGL_DISPLAYS), egl_$(dpy)_$(EGL_DRIVER_NAME).so)
 
-LIB_GALLIUM_DIR = $(TOP)/$(LIB_DIR)/gallium
-EGL_DISPLAY_LIBS = $(foreach drv, $(EGL_DISPLAY_DRIVERS), $(LIB_GALLIUM_DIR)/$(drv))
+EGL_DISPLAY_LIBS = $(foreach drv, $(EGL_DISPLAY_DRIVERS), $(TOP)/$(LIB_DIR)/$(drv))
 
 default: $(EGL_DISPLAY_LIBS)
 
-$(EGL_DISPLAY_LIBS): $(LIB_GALLIUM_DIR)/%.so: %.so
-	@mkdir -p $(LIB_GALLIUM_DIR)
-	$(INSTALL) $^ $(LIB_GALLIUM_DIR)
+$(EGL_DISPLAY_LIBS): $(TOP)/$(LIB_DIR)/%.so: %.so
+	$(INSTALL) $< $(TOP)/$(LIB_DIR)
 
 define mklib-egl
-$(MKLIB) -noprefix -o $@ $(EGL_DRIVER_OBJECTS) \
+$(MKLIB) -o $@ -noprefix -linker '$(CC)' -ldflags '$(LDFLAGS)' \
+	$(MKLIB_OPTIONS) $(EGL_DRIVER_OBJECTS) \
 	-Wl,--whole-archive $($(1)_ST) -Wl,--no-whole-archive \
 	$(EGL_DRIVER_PIPES) $(GALLIUM_AUXILIARIES) $($(1)_LIBS) $(EGL_DRIVER_LIBS)
 endef
@@ -55,10 +54,9 @@ clean:
 	-rm -f $(EGL_DISPLAY_DRIVERS)
 
 install: $(EGL_DISPLAY_LIBS)
-	@$(INSTALL) -d $(DESTDIR)$(DRI_DRIVER_INSTALL_DIR)
-	@echo "Install $(EGL_DISPLAY_DRIVERS)"
-	@for lib in "$(EGL_DISPLAY_LIBS)"; do \
-		$(MINSTALL) -m 755 "$$lib" $(DESTDIR)$(DRI_DRIVER_INSTALL_DIR); \
+	$(INSTALL) -d $(DESTDIR)$(EGL_DRIVER_INSTALL_DIR)
+	for lib in $(EGL_DISPLAY_LIBS); do \
+		$(MINSTALL) -m 755 "$$lib" $(DESTDIR)$(EGL_DRIVER_INSTALL_DIR); \
 	done
 
 depend:
diff --git a/src/gallium/winsys/drm/Makefile.template b/src/gallium/winsys/drm/Makefile.template
index 9635c3c50e..960353a73d 100644
--- a/src/gallium/winsys/drm/Makefile.template
+++ b/src/gallium/winsys/drm/Makefile.template
@@ -82,18 +82,11 @@ SHARED_INCLUDES = \
 default: depend symlinks $(TOP)/$(LIB_DIR)/gallium/$(LIBNAME)
 
 $(LIBNAME): $(OBJECTS) $(MESA_MODULES) $(PIPE_DRIVERS) $(WINOBJ) Makefile $(TOP)/src/mesa/drivers/dri/Makefile.template
-	$(MKLIB) -noprefix -o $@ \
+	$(MKLIB) -o $@ -noprefix -linker '$(CC)' -ldflags '$(LDFLAGS)' \
 		$(OBJECTS) $(PIPE_DRIVERS) \
                 -Wl,--start-group $(MESA_MODULES) -Wl,--end-group \
                  $(WINOBJ) $(DRI_LIB_DEPS) $(DRIVER_EXTRAS)
 
-$(LIBNAME_EGL): $(WINSYS_OBJECTS) $(LIBS)
-	$(MKLIB) -o $(LIBNAME_EGL) \
-		-linker "$(CC)" \
-		-noprefix \
-		$(OBJECTS) $(MKLIB_OPTIONS) $(WINSYS_OBJECTS) $(PIPE_DRIVERS) $(WINOBJ) $(DRI_LIB_DEPS) \
-		--whole-archive $(LIBS) $(GALLIUM_AUXILIARIES) --no-whole-archive $(DRIVER_EXTRAS)
-
 $(TOP)/$(LIB_DIR)/gallium:
 	mkdir -p $@
 
diff --git a/src/gallium/winsys/drm/i965/egl/Makefile b/src/gallium/winsys/drm/i965/egl/Makefile
index a1b32eb2a7..1c13258200 100644
--- a/src/gallium/winsys/drm/i965/egl/Makefile
+++ b/src/gallium/winsys/drm/i965/egl/Makefile
@@ -1,29 +1,14 @@
 TOP = ../../../../../..
-GALLIUMDIR = ../../../..
 include $(TOP)/configs/current
 
-LIBNAME = EGL_i965.so
+EGL_DRIVER_NAME = i965
+EGL_DRIVER_SOURCES = dummy.c
+EGL_DRIVER_LIBS = -ldrm_intel
 
-PIPE_DRIVERS = \
-	$(TOP)/src/gallium/state_trackers/egl/libegldrm.a \
-	$(GALLIUMDIR)/winsys/drm/i965/gem/libi965drm.a \
+EGL_DRIVER_PIPES = \
+	$(TOP)/src/gallium/winsys/drm/i965/gem/libi965drm.a \
 	$(TOP)/src/gallium/drivers/softpipe/libsoftpipe.a \
 	$(TOP)/src/gallium/drivers/trace/libtrace.a \
 	$(TOP)/src/gallium/drivers/i965/libi965.a
 
-DRIVER_SOURCES =
-
-C_SOURCES = \
-	$(COMMON_GALLIUM_SOURCES) \
-	$(DRIVER_SOURCES)
-
-DRIVER_EXTRAS = -ldrm_intel
-
-ASM_SOURCES = 
-
-DRIVER_DEFINES = -I../gem $(shell pkg-config libdrm --atleast-version=2.3.1 \
-				&& echo "-DDRM_VBLANK_FLIP=DRM_VBLANK_FLIP")
-
-include ../../Makefile.template
-
-symlinks:
+include ../../Makefile.egl
diff --git a/src/gallium/winsys/drm/i965/egl_g3d/dummy.c b/src/gallium/winsys/drm/i965/egl/dummy.c
index 4a1bc28b0b..4a1bc28b0b 100644
--- a/src/gallium/winsys/drm/i965/egl_g3d/dummy.c
+++ b/src/gallium/winsys/drm/i965/egl/dummy.c
diff --git a/src/gallium/winsys/drm/i965/egl_g3d/Makefile b/src/gallium/winsys/drm/i965/egl_g3d/Makefile
deleted file mode 100644
index dd2efe2485..0000000000
--- a/src/gallium/winsys/drm/i965/egl_g3d/Makefile
+++ /dev/null
@@ -1,14 +0,0 @@
-TOP = ../../../../../..
-include $(TOP)/configs/current
-
-EGL_DRIVER_NAME = i965
-EGL_DRIVER_SOURCES = dummy.c
-EGL_DRIVER_LIBS = -ldrm_intel
-
-EGL_DRIVER_PIPES = \
-	$(TOP)/src/gallium/winsys/drm/i965/gem/libi965drm.a \
-	$(TOP)/src/gallium/drivers/softpipe/libsoftpipe.a \
-	$(TOP)/src/gallium/drivers/trace/libtrace.a \
-	$(TOP)/src/gallium/drivers/i965/libi965.a
-
-include ../../Makefile.egl_g3d
diff --git a/src/gallium/winsys/drm/i965/xorg/Makefile b/src/gallium/winsys/drm/i965/xorg/Makefile
index 65d521fc1c..c25726b0bb 100644
--- a/src/gallium/winsys/drm/i965/xorg/Makefile
+++ b/src/gallium/winsys/drm/i965/xorg/Makefile
@@ -36,7 +36,7 @@ all default: $(TARGET) $(TARGET_STAGING)
 
 $(TARGET): $(OBJECTS) Makefile $(GALLIUMDIR)/state_trackers/xorg/libxorgtracker.a $(LIBS)
 	$(TOP)/bin/mklib -noprefix -o $@ \
-	$(OBJECTS) $(LIBS) $(shell pkg-config --libs libdrm) -ldrm_radeon
+	$(OBJECTS) $(LIBS) $(shell pkg-config --libs libdrm) -ldrm_intel
 
 $(TOP)/$(LIB_DIR)/gallium:
 	mkdir -p $@
diff --git a/src/gallium/winsys/drm/intel/egl/Makefile b/src/gallium/winsys/drm/intel/egl/Makefile
index 1397e9f729..60d675ca73 100644
--- a/src/gallium/winsys/drm/intel/egl/Makefile
+++ b/src/gallium/winsys/drm/intel/egl/Makefile
@@ -1,29 +1,14 @@
 TOP = ../../../../../..
-GALLIUMDIR = ../../../..
 include $(TOP)/configs/current
 
-LIBNAME = EGL_i915.so
+EGL_DRIVER_NAME = i915
+EGL_DRIVER_SOURCES = dummy.c
+EGL_DRIVER_LIBS = -ldrm_intel
 
-PIPE_DRIVERS = \
-	$(TOP)/src/gallium/state_trackers/egl/libegldrm.a \
-	$(GALLIUMDIR)/winsys/drm/intel/gem/libinteldrm.a \
+EGL_DRIVER_PIPES = \
+	$(TOP)/src/gallium/winsys/drm/intel/gem/libinteldrm.a \
 	$(TOP)/src/gallium/drivers/softpipe/libsoftpipe.a \
 	$(TOP)/src/gallium/drivers/trace/libtrace.a \
 	$(TOP)/src/gallium/drivers/i915/libi915.a
 
-DRIVER_SOURCES =
-
-C_SOURCES = \
-	$(COMMON_GALLIUM_SOURCES) \
-	$(DRIVER_SOURCES)
-
-DRIVER_EXTRAS = -ldrm_intel
-
-ASM_SOURCES = 
-
-DRIVER_DEFINES = -I../gem $(shell pkg-config libdrm --atleast-version=2.3.1 \
-				&& echo "-DDRM_VBLANK_FLIP=DRM_VBLANK_FLIP")
-
-include ../../Makefile.template
-
-symlinks:
+include ../../Makefile.egl
diff --git a/src/gallium/winsys/drm/intel/egl_g3d/dummy.c b/src/gallium/winsys/drm/intel/egl/dummy.c
index 4a1bc28b0b..4a1bc28b0b 100644
--- a/src/gallium/winsys/drm/intel/egl_g3d/dummy.c
+++ b/src/gallium/winsys/drm/intel/egl/dummy.c
diff --git a/src/gallium/winsys/drm/intel/egl_g3d/Makefile b/src/gallium/winsys/drm/intel/egl_g3d/Makefile
deleted file mode 100644
index cdbb680773..0000000000
--- a/src/gallium/winsys/drm/intel/egl_g3d/Makefile
+++ /dev/null
@@ -1,14 +0,0 @@
-TOP = ../../../../../..
-include $(TOP)/configs/current
-
-EGL_DRIVER_NAME = i915
-EGL_DRIVER_SOURCES = dummy.c
-EGL_DRIVER_LIBS = -ldrm_intel
-
-EGL_DRIVER_PIPES = \
-	$(TOP)/src/gallium/winsys/drm/intel/gem/libinteldrm.a \
-	$(TOP)/src/gallium/drivers/softpipe/libsoftpipe.a \
-	$(TOP)/src/gallium/drivers/trace/libtrace.a \
-	$(TOP)/src/gallium/drivers/i915/libi915.a
-
-include ../../Makefile.egl_g3d
diff --git a/src/gallium/winsys/drm/intel/gem/intel_drm_api.c b/src/gallium/winsys/drm/intel/gem/intel_drm_api.c
index 450ae09b34..8c8176e44a 100644
--- a/src/gallium/winsys/drm/intel/gem/intel_drm_api.c
+++ b/src/gallium/winsys/drm/intel/gem/intel_drm_api.c
@@ -196,6 +196,7 @@ destroy(struct drm_api *api)
 struct drm_api intel_drm_api =
 {
    .name = "i915",
+   .driver_name = "i915",
    .create_context = intel_drm_create_context,
    .create_screen = intel_drm_create_screen,
    .texture_from_shared_handle = intel_drm_texture_from_shared_handle,
diff --git a/src/gallium/winsys/drm/nouveau/drm/nouveau_drm_api.c b/src/gallium/winsys/drm/nouveau/drm/nouveau_drm_api.c
index 4b2c6a1025..75cd1e2902 100644
--- a/src/gallium/winsys/drm/nouveau/drm/nouveau_drm_api.c
+++ b/src/gallium/winsys/drm/nouveau/drm/nouveau_drm_api.c
@@ -54,6 +54,16 @@ static struct dri1_api nouveau_dri1_api = {
 	nouveau_dri1_front_surface,
 };
 
+static void
+nouveau_drm_destroy_winsys(struct pipe_winsys *s)
+{
+	struct nouveau_winsys *nv_winsys = nouveau_winsys(s);
+	struct nouveau_screen *nv_screen= nouveau_screen(nv_winsys->pscreen);
+	nouveau_device_close(&nv_screen->device);
+	FREE(nv_winsys->pctx);
+	FREE(nv_winsys);
+}
+
 static struct pipe_screen *
 nouveau_drm_create_screen(struct drm_api *api, int fd,
 			  struct drm_create_screen_arg *arg)
@@ -105,6 +115,7 @@ nouveau_drm_create_screen(struct drm_api *api, int fd,
 		return NULL;
 	}
 	ws = &nvws->base;
+	ws->destroy = nouveau_drm_destroy_winsys;
 
 	nvws->pscreen = init(ws, dev);
 	if (!nvws->pscreen) {
@@ -255,6 +266,7 @@ nouveau_drm_handle_from_pt(struct drm_api *api, struct pipe_screen *pscreen,
 
 struct drm_api drm_api_hooks = {
 	.name = "nouveau",
+	.driver_name = "nouveau",
 	.create_screen = nouveau_drm_create_screen,
 	.create_context = nouveau_drm_create_context,
 	.texture_from_shared_handle = nouveau_drm_pt_from_name,
diff --git a/src/gallium/winsys/drm/nouveau/egl_g3d/Makefile b/src/gallium/winsys/drm/nouveau/egl/Makefile
index 865a5d56a9..8e812acc86 100644
--- a/src/gallium/winsys/drm/nouveau/egl_g3d/Makefile
+++ b/src/gallium/winsys/drm/nouveau/egl/Makefile
@@ -16,4 +16,4 @@ EGL_DRIVER_PIPES = \
 	$(TOP)/src/gallium/drivers/nouveau/libnouveau.a \
 	$(TOP)/src/gallium/drivers/softpipe/libsoftpipe.a
 
-include ../../Makefile.egl_g3d
+include ../../Makefile.egl
diff --git a/src/gallium/winsys/drm/nouveau/egl_g3d/dummy.c b/src/gallium/winsys/drm/nouveau/egl/dummy.c
index 4a1bc28b0b..4a1bc28b0b 100644
--- a/src/gallium/winsys/drm/nouveau/egl_g3d/dummy.c
+++ b/src/gallium/winsys/drm/nouveau/egl/dummy.c
diff --git a/src/gallium/winsys/drm/radeon/core/radeon_buffer.c b/src/gallium/winsys/drm/radeon/core/radeon_buffer.c
index 385fa857b5..f484503e0e 100644
--- a/src/gallium/winsys/drm/radeon/core/radeon_buffer.c
+++ b/src/gallium/winsys/drm/radeon/core/radeon_buffer.c
@@ -51,6 +51,26 @@ static const char *radeon_get_name(struct pipe_winsys *ws)
     return "Radeon/GEM+KMS";
 }
 
+static uint32_t radeon_domain_from_usage(unsigned usage)
+{
+    uint32_t domain = 0;
+
+    if (usage & PIPE_BUFFER_USAGE_GPU_WRITE) {
+        domain |= RADEON_GEM_DOMAIN_VRAM;
+    }
+    if (usage & PIPE_BUFFER_USAGE_PIXEL) {
+        domain |= RADEON_GEM_DOMAIN_VRAM;
+    }
+    if (usage & PIPE_BUFFER_USAGE_VERTEX) {
+        domain |= RADEON_GEM_DOMAIN_GTT;
+    }
+    if (usage & PIPE_BUFFER_USAGE_INDEX) {
+        domain |= RADEON_GEM_DOMAIN_GTT;
+    }
+
+    return domain;
+}
+
 static struct pipe_buffer *radeon_buffer_create(struct pipe_winsys *ws,
                                                 unsigned alignment,
                                                 unsigned usage,
@@ -58,6 +78,7 @@ static struct pipe_buffer *radeon_buffer_create(struct pipe_winsys *ws,
 {
     struct radeon_winsys *radeon_ws = (struct radeon_winsys *)ws;
     struct radeon_pipe_buffer *radeon_buffer;
+    struct pb_desc desc;
     uint32_t domain;
 
     radeon_buffer = CALLOC_STRUCT(radeon_pipe_buffer);
@@ -70,18 +91,16 @@ static struct pipe_buffer *radeon_buffer_create(struct pipe_winsys *ws,
     radeon_buffer->base.usage = usage;
     radeon_buffer->base.size = size;
 
-    domain = 0;
-
-    if (usage & PIPE_BUFFER_USAGE_PIXEL) {
-        domain |= RADEON_GEM_DOMAIN_VRAM;
-    }
-    if (usage & PIPE_BUFFER_USAGE_VERTEX) {
-        domain |= RADEON_GEM_DOMAIN_GTT;
-    }
-    if (usage & PIPE_BUFFER_USAGE_INDEX) {
-        domain |= RADEON_GEM_DOMAIN_GTT;
+    if (usage & PIPE_BUFFER_USAGE_CONSTANT && is_r3xx(radeon_ws->pci_id)) {
+        /* Don't bother allocating a BO, as it'll never get to the card. */
+        desc.alignment = alignment;
+        desc.usage = usage;
+        radeon_buffer->pb = pb_malloc_buffer_create(size, &desc);
+        return &radeon_buffer->base;
     }
 
+    domain = radeon_domain_from_usage(usage);
+
     radeon_buffer->bo = radeon_bo_open(radeon_ws->priv->bom, 0, size,
             alignment, domain, 0);
     if (radeon_buffer->bo == NULL) {
@@ -133,8 +152,16 @@ static void radeon_buffer_del(struct pipe_buffer *buffer)
     struct radeon_pipe_buffer *radeon_buffer =
         (struct radeon_pipe_buffer*)buffer;
 
-    radeon_bo_unref(radeon_buffer->bo);
-    free(radeon_buffer);
+    if (radeon_buffer->pb) {
+        pipe_reference_init(&radeon_buffer->pb->base.reference, 0);
+        pb_destroy(radeon_buffer->pb);
+    }
+
+    if (radeon_buffer->bo) {
+        radeon_bo_unref(radeon_buffer->bo);
+    }
+
+    FREE(radeon_buffer);
 }
 
 static void *radeon_buffer_map(struct pipe_winsys *ws,
@@ -146,6 +173,10 @@ static void *radeon_buffer_map(struct pipe_winsys *ws,
         (struct radeon_pipe_buffer*)buffer;
     int write = 0;
 
+    if (radeon_buffer->pb) {
+        return pb_map(radeon_buffer->pb, flags);
+    }
+
     if (flags & PIPE_BUFFER_USAGE_DONTBLOCK) {
         uint32_t domain;
 
@@ -174,7 +205,31 @@ static void radeon_buffer_unmap(struct pipe_winsys *ws,
     struct radeon_pipe_buffer *radeon_buffer =
         (struct radeon_pipe_buffer*)buffer;
 
-    radeon_bo_unmap(radeon_buffer->bo);
+    if (radeon_buffer->pb) {
+        pb_unmap(radeon_buffer->pb);
+    } else {
+        radeon_bo_unmap(radeon_buffer->bo);
+    }
+}
+
+static void radeon_buffer_set_tiling(struct radeon_winsys *ws,
+                                     struct pipe_buffer *buffer,
+                                     uint32_t pitch,
+                                     boolean microtiled,
+                                     boolean macrotiled)
+{
+    struct radeon_pipe_buffer *radeon_buffer =
+        (struct radeon_pipe_buffer*)buffer;
+    uint32_t flags = 0;
+
+    if (microtiled) {
+        flags |= RADEON_BO_FLAGS_MICRO_TILE;
+    }
+    if (macrotiled) {
+        flags |= RADEON_BO_FLAGS_MACRO_TILE;
+    }
+
+    radeon_bo_set_tiling(radeon_buffer->bo, flags, pitch);
 }
 
 static void radeon_fence_reference(struct pipe_winsys *ws,
@@ -279,5 +334,7 @@ struct radeon_winsys* radeon_pipe_winsys(int fd)
 
     radeon_ws->base.get_name = radeon_get_name;
 
+    radeon_ws->buffer_set_tiling = radeon_buffer_set_tiling;
+
     return radeon_ws;
 }
diff --git a/src/gallium/winsys/drm/radeon/core/radeon_buffer.h b/src/gallium/winsys/drm/radeon/core/radeon_buffer.h
index d7f17564a9..de71cb2f42 100644
--- a/src/gallium/winsys/drm/radeon/core/radeon_buffer.h
+++ b/src/gallium/winsys/drm/radeon/core/radeon_buffer.h
@@ -36,7 +36,7 @@
 #include "pipe/p_defines.h"
 #include "pipe/p_inlines.h"
 
-//#include "state_tracker/st_public.h"
+#include "pipebuffer/pb_buffer.h"
 
 #include "util/u_memory.h"
 
@@ -49,7 +49,10 @@
 
 struct radeon_pipe_buffer {
     struct pipe_buffer  base;
+    /* Pointer to GPU-backed BO. */
     struct radeon_bo    *bo;
+    /* Pointer to fallback PB buffer. */
+    struct pb_buffer    *pb;
     boolean flinked;
     uint32_t flink;
 };
diff --git a/src/gallium/winsys/drm/radeon/core/radeon_drm.c b/src/gallium/winsys/drm/radeon/core/radeon_drm.c
index 851c223697..bff6fdc1ad 100644
--- a/src/gallium/winsys/drm/radeon/core/radeon_drm.c
+++ b/src/gallium/winsys/drm/radeon/core/radeon_drm.c
@@ -40,12 +40,16 @@ static void do_ioctls(int fd, struct radeon_winsys* winsys)
     struct drm_radeon_info info = {0};
     int target = 0;
     int retval;
+    drmVersionPtr version;
 
     info.value = (unsigned long)&target;
 
     /* We do things in a specific order here.
      *
-     * First, the PCI ID. This is essential and should return usable numbers
+     * DRM version first. We need to be sure we're running on a KMS chipset.
+     * This is also for some features.
+     *
+     * Then, the PCI ID. This is essential and should return usable numbers
      * for all Radeons. If this fails, we probably got handed an FD for some
      * non-Radeon card.
      *
@@ -55,8 +59,18 @@ static void do_ioctls(int fd, struct radeon_winsys* winsys)
      *
      * The GEM info is actually bogus on the kernel side, as well as our side
      * (see radeon_gem_info_ioctl in radeon_gem.c) but that's alright because
-     * we don't actually use the info for anything yet.
-     * XXX update the above when we can safely use vram_size instead of vram_visible */
+     * we don't actually use the info for anything yet. */
+
+    version = drmGetVersion(fd);
+    if (version->version_major != 2) {
+        fprintf(stderr, "%s: DRM version is %d.%d.%d but this driver is "
+                "only compatible with 2.x.x\n", __FUNCTION__,
+                version->version_major, version->version_minor,
+                version->version_patchlevel);
+        drmFreeVersion(version);
+        exit(1);
+    }
+
     info.request = RADEON_INFO_DEVICE_ID;
     retval = drmCommandWriteRead(fd, DRM_RADEON_INFO, &info, sizeof(info));
     if (retval) {
@@ -92,16 +106,18 @@ static void do_ioctls(int fd, struct radeon_winsys* winsys)
         exit(1);
     }
     winsys->gart_size = gem_info.gart_size;
-    /* XXX */
-    winsys->vram_size = gem_info.vram_visible;
-}
-
-/* Guess at whether this chipset should use r300g.
- *
- * I believe that this check is valid, but I haven't been exhaustive. */
-static boolean is_r3xx(int pciid)
-{
-    return (pciid > 0x3150) && (pciid < 0x796f);
+    winsys->vram_size = gem_info.vram_size;
+
+    debug_printf("radeon: Successfully grabbed chipset info from kernel!\n"
+                 "radeon: DRM version: %d.%d.%d ID: 0x%04x GB: %d Z: %d\n"
+                 "radeon: GART size: %d MB VRAM size: %d MB\n",
+                 version->version_major, version->version_minor,
+                 version->version_patchlevel, winsys->pci_id,
+                 winsys->gb_pipes, winsys->z_pipes,
+                 winsys->gart_size / 1024 / 1024,
+                 winsys->vram_size / 1024 / 1024);
+
+    drmFreeVersion(version);
 }
 
 /* Create a pipe_screen. */
@@ -136,12 +152,13 @@ struct pipe_context* radeon_create_context(struct drm_api* api,
 }
 
 boolean radeon_buffer_from_texture(struct drm_api* api,
+                                   struct pipe_screen* screen,
                                    struct pipe_texture* texture,
                                    struct pipe_buffer** buffer,
                                    unsigned* stride)
 {
     /* XXX fix this */
-    return r300_get_texture_buffer(texture, buffer, stride);
+    return r300_get_texture_buffer(screen, texture, buffer, stride);
 }
 
 /* Create a buffer from a handle. */
@@ -208,7 +225,7 @@ static boolean radeon_shared_handle_from_texture(struct drm_api *api,
     struct radeon_pipe_buffer* radeon_buffer;
     struct pipe_buffer *buffer = NULL;
 
-    if (!radeon_buffer_from_texture(api, texture, &buffer, stride)) {
+    if (!radeon_buffer_from_texture(api, screen, texture, &buffer, stride)) {
         return FALSE;
     }
 
@@ -240,7 +257,7 @@ static boolean radeon_local_handle_from_texture(struct drm_api *api,
                                                 unsigned *handle)
 {
     struct pipe_buffer *buffer = NULL;
-    if (!radeon_buffer_from_texture(api, texture, &buffer, stride)) {
+    if (!radeon_buffer_from_texture(api, screen, texture, &buffer, stride)) {
         return FALSE;
     }
 
@@ -253,6 +270,7 @@ static boolean radeon_local_handle_from_texture(struct drm_api *api,
 
 struct drm_api drm_api_hooks = {
     .name = "radeon",
+    .driver_name = "radeon",
     .create_screen = radeon_create_screen,
     .create_context = radeon_create_context,
     .texture_from_shared_handle = radeon_texture_from_shared_handle,
diff --git a/src/gallium/winsys/drm/radeon/core/radeon_drm.h b/src/gallium/winsys/drm/radeon/core/radeon_drm.h
index bf0e78138d..077388ee02 100644
--- a/src/gallium/winsys/drm/radeon/core/radeon_drm.h
+++ b/src/gallium/winsys/drm/radeon/core/radeon_drm.h
@@ -56,6 +56,7 @@ struct pipe_context* radeon_create_context(struct drm_api* api,
                                            struct pipe_screen* screen);
 
 boolean radeon_buffer_from_texture(struct drm_api* api,
+                                   struct pipe_screen* screen,
                                    struct pipe_texture* texture,
                                    struct pipe_buffer** buffer,
                                    unsigned* stride);
@@ -76,4 +77,13 @@ boolean radeon_global_handle_from_buffer(struct drm_api* api,
                                          unsigned* handle);
 
 void radeon_destroy_drm_api(struct drm_api* api);
+
+/* Guess at whether this chipset should use r300g.
+ *
+ * I believe that this check is valid, but I haven't been exhaustive. */
+static INLINE boolean is_r3xx(int pciid)
+{
+    return (pciid > 0x3150) && (pciid < 0x796f);
+}
+
 #endif
diff --git a/src/gallium/winsys/drm/radeon/core/radeon_r300.c b/src/gallium/winsys/drm/radeon/core/radeon_r300.c
index 0875ee41cb..d759beaba1 100644
--- a/src/gallium/winsys/drm/radeon/core/radeon_r300.c
+++ b/src/gallium/winsys/drm/radeon/core/radeon_r300.c
@@ -81,9 +81,13 @@ static void radeon_write_cs_reloc(struct radeon_winsys* winsys,
                                   uint32_t flags)
 {
     int retval = 0;
+    struct radeon_pipe_buffer* radeon_buffer =
+        (struct radeon_pipe_buffer*)pbuffer;
 
-    retval = radeon_cs_write_reloc(winsys->priv->cs,
-            ((struct radeon_pipe_buffer*)pbuffer)->bo, rd, wd, flags);
+    assert(!radeon_buffer->pb);
+
+    retval = radeon_cs_write_reloc(winsys->priv->cs, radeon_buffer->bo,
+                                   rd, wd, flags);
 
     if (retval) {
         debug_printf("radeon: Relocation of %p (%d, %d, %d) failed!\n",
@@ -108,6 +112,11 @@ static void radeon_flush_cs(struct radeon_winsys* winsys)
 {
     int retval;
 
+    /* Don't flush a zero-sized CS. */
+    if (!winsys->priv->cs->cdw) {
+        return;
+    }
+
     /* Emit the CS. */
     retval = radeon_cs_emit(winsys->priv->cs);
     if (retval) {
diff --git a/src/gallium/winsys/drm/radeon/core/radeon_winsys.h b/src/gallium/winsys/drm/radeon/core/radeon_winsys.h
index 9edc9e038c..864082b99b 100644
--- a/src/gallium/winsys/drm/radeon/core/radeon_winsys.h
+++ b/src/gallium/winsys/drm/radeon/core/radeon_winsys.h
@@ -100,6 +100,12 @@ struct radeon_winsys {
 			 void (*flush_cb)(void *), void *data);
 
     void (*reset_bos)(struct radeon_winsys *winsys);
+
+    void (*buffer_set_tiling)(struct radeon_winsys* winsys,
+                              struct pipe_buffer* buffer,
+                              uint32_t pitch,
+                              boolean microtiled,
+                              boolean macrotiled);
 };
 
 #endif
diff --git a/src/gallium/winsys/drm/radeon/egl/Makefile b/src/gallium/winsys/drm/radeon/egl/Makefile
index 6a1448d1b9..cd4f9b20f0 100644
--- a/src/gallium/winsys/drm/radeon/egl/Makefile
+++ b/src/gallium/winsys/drm/radeon/egl/Makefile
@@ -1,26 +1,14 @@
 TOP = ../../../../../..
-GALLIUMDIR = ../../../..
 include $(TOP)/configs/current
 
-LIBNAME = EGL_r300.so
+EGL_DRIVER_NAME = radeon
+EGL_DRIVER_SOURCES = dummy.c
+EGL_DRIVER_LIBS = -ldrm_radeon
 
-PIPE_DRIVERS = \
-	$(TOP)/src/gallium/state_trackers/egl/libegldrm.a \
-	$(GALLIUMDIR)/winsys/drm/radeon/core/libradeonwinsys.a \
+EGL_DRIVER_PIPES = \
+	$(TOP)/src/gallium/winsys/drm/radeon/core/libradeonwinsys.a \
 	$(TOP)/src/gallium/drivers/softpipe/libsoftpipe.a \
 	$(TOP)/src/gallium/drivers/trace/libtrace.a \
 	$(TOP)/src/gallium/drivers/r300/libr300.a
 
-DRIVER_SOURCES =
-
-C_SOURCES = \
-	$(COMMON_GALLIUM_SOURCES) \
-	$(DRIVER_SOURCES)
-
-DRIVER_EXTRAS = -ldrm_radeon
-
-ASM_SOURCES = 
-
-include ../../Makefile.template
-
-symlinks:
+include ../../Makefile.egl
diff --git a/src/gallium/winsys/drm/radeon/egl_g3d/dummy.c b/src/gallium/winsys/drm/radeon/egl/dummy.c
index 4a1bc28b0b..4a1bc28b0b 100644
--- a/src/gallium/winsys/drm/radeon/egl_g3d/dummy.c
+++ b/src/gallium/winsys/drm/radeon/egl/dummy.c
diff --git a/src/gallium/winsys/drm/radeon/egl_g3d/Makefile b/src/gallium/winsys/drm/radeon/egl_g3d/Makefile
deleted file mode 100644
index 9027a5ff46..0000000000
--- a/src/gallium/winsys/drm/radeon/egl_g3d/Makefile
+++ /dev/null
@@ -1,14 +0,0 @@
-TOP = ../../../../../..
-include $(TOP)/configs/current
-
-EGL_DRIVER_NAME = r300
-EGL_DRIVER_SOURCES = dummy.c
-EGL_DRIVER_LIBS = -ldrm_radeon
-
-EGL_DRIVER_PIPES = \
-	$(TOP)/src/gallium/winsys/drm/radeon/core/libradeonwinsys.a \
-	$(TOP)/src/gallium/drivers/softpipe/libsoftpipe.a \
-	$(TOP)/src/gallium/drivers/trace/libtrace.a \
-	$(TOP)/src/gallium/drivers/r300/libr300.a
-
-include ../../Makefile.egl_g3d
diff --git a/src/gallium/winsys/drm/swrast/Makefile b/src/gallium/winsys/drm/swrast/Makefile
new file mode 100644
index 0000000000..363b89584f
--- /dev/null
+++ b/src/gallium/winsys/drm/swrast/Makefile
@@ -0,0 +1,12 @@
+# src/gallium/winsys/drm/swrast/Makefile
+TOP = ../../../../..
+include $(TOP)/configs/current
+
+SUBDIRS = core $(GALLIUM_STATE_TRACKERS_DIRS)
+
+default install clean:
+	@for dir in $(SUBDIRS) ; do \
+		if [ -d $$dir ] ; then \
+			(cd $$dir && $(MAKE) $@) || exit 1; \
+		fi \
+	done
diff --git a/src/gallium/winsys/drm/swrast/core/Makefile b/src/gallium/winsys/drm/swrast/core/Makefile
new file mode 100644
index 0000000000..93931ae22b
--- /dev/null
+++ b/src/gallium/winsys/drm/swrast/core/Makefile
@@ -0,0 +1,10 @@
+# src/gallium/winsys/drm/swrast/core/Makefile
+
+TOP = ../../../../../..
+include $(TOP)/configs/current
+
+LIBNAME = swrastdrm
+
+C_SOURCES = swrast_drm_api.c
+
+include ../../../../Makefile.template
diff --git a/src/gallium/winsys/drm/swrast/core/swrast_drm_api.c b/src/gallium/winsys/drm/swrast/core/swrast_drm_api.c
new file mode 100644
index 0000000000..8c9f80e2c1
--- /dev/null
+++ b/src/gallium/winsys/drm/swrast/core/swrast_drm_api.c
@@ -0,0 +1,13 @@
+#include "state_tracker/drm_api.h"
+
+static struct drm_api swrast_drm_api =
+{
+   .name = "swrast",
+};
+
+struct drm_api *
+drm_api_create()
+{
+   (void) swrast_drm_api;
+   return NULL;
+}
diff --git a/src/gallium/winsys/drm/swrast/egl/Makefile b/src/gallium/winsys/drm/swrast/egl/Makefile
new file mode 100644
index 0000000000..26fe2d2805
--- /dev/null
+++ b/src/gallium/winsys/drm/swrast/egl/Makefile
@@ -0,0 +1,12 @@
+TOP = ../../../../../..
+include $(TOP)/configs/current
+
+EGL_DRIVER_NAME = swrast
+EGL_DRIVER_SOURCES = dummy.c
+EGL_DRIVER_LIBS =
+
+EGL_DRIVER_PIPES = \
+	$(TOP)/src/gallium/winsys/drm/swrast/core/libswrastdrm.a \
+	$(TOP)/src/gallium/drivers/softpipe/libsoftpipe.a
+
+include ../../Makefile.egl
diff --git a/src/gallium/winsys/drm/vmware/egl_g3d/dummy.c b/src/gallium/winsys/drm/swrast/egl/dummy.c
index 4a1bc28b0b..4a1bc28b0b 100644
--- a/src/gallium/winsys/drm/vmware/egl_g3d/dummy.c
+++ b/src/gallium/winsys/drm/swrast/egl/dummy.c
diff --git a/src/gallium/winsys/drm/vmware/core/vmw_context.c b/src/gallium/winsys/drm/vmware/core/vmw_context.c
index b6997588de..b5fd4f5a6a 100644
--- a/src/gallium/winsys/drm/vmware/core/vmw_context.c
+++ b/src/gallium/winsys/drm/vmware/core/vmw_context.c
@@ -41,9 +41,18 @@
 
 #define VMW_COMMAND_SIZE (64*1024)
 #define VMW_SURFACE_RELOCS (1024)
+#define VMW_REGION_RELOCS (512)
 
 #define VMW_MUST_FLUSH_STACK 8
 
+struct vmw_region_relocation
+{
+   struct SVGAGuestPtr *where;
+   struct pb_buffer *buffer;
+   /* TODO: put offset info inside where */
+   uint32 offset;
+};
+
 struct vmw_svga_winsys_context
 {
    struct svga_winsys_context base;
@@ -69,10 +78,31 @@ struct vmw_svga_winsys_context
       uint32_t staged;
       uint32_t reserved;
    } surface;
+   
+   struct {
+      struct vmw_region_relocation relocs[VMW_REGION_RELOCS];
+      uint32_t size;
+      uint32_t used;
+      uint32_t staged;
+      uint32_t reserved;
+   } region;
 
    struct pb_validate *validate;
 
    uint32_t last_fence;
+
+   /**
+    * The amount of GMR that is referred by the commands currently batched
+    * in the context.
+    */
+   uint32_t seen_regions;
+
+   /**
+    * Whether this context should fail to reserve more commands, not because it
+    * ran out of command space, but because a substantial ammount of GMR was
+    * referred.
+    */
+   boolean preemptive_flush;
 };
 
 
@@ -96,6 +126,19 @@ vmw_swc_flush(struct svga_winsys_context *swc,
    ret = pb_validate_validate(vswc->validate);
    assert(ret == PIPE_OK);
    if(ret == PIPE_OK) {
+   
+      /* Apply relocations */
+      for(i = 0; i < vswc->region.used; ++i) {
+         struct vmw_region_relocation *reloc = &vswc->region.relocs[i];
+         struct SVGAGuestPtr ptr;
+
+         if(!vmw_gmr_bufmgr_region_ptr(reloc->buffer, &ptr))
+            assert(0);
+
+         ptr.offset += reloc->offset;
+
+         *reloc->where = ptr;
+      }
 
       if (vswc->command.used)
          vmw_ioctl_command(vswc->vws,
@@ -121,9 +164,18 @@ vmw_swc_flush(struct svga_winsys_context *swc,
    vswc->surface.used = 0;
    vswc->surface.reserved = 0;
 
+   for(i = 0; i < vswc->region.used + vswc->region.staged; ++i) {
+      pb_reference(&vswc->region.relocs[i].buffer, NULL);
+   }
+
+   vswc->region.used = 0;
+   vswc->region.reserved = 0;
+
 #ifdef DEBUG
    vswc->must_flush = FALSE;
 #endif
+   vswc->preemptive_flush = FALSE;
+   vswc->seen_regions = 0;
 
    if(pfence)
       *pfence = fence;
@@ -151,8 +203,10 @@ vmw_swc_reserve(struct svga_winsys_context *swc,
    if(nr_bytes > vswc->command.size)
       return NULL;
 
-   if(vswc->command.used + nr_bytes > vswc->command.size ||
-      vswc->surface.used + nr_relocs > vswc->surface.size) {
+   if(vswc->preemptive_flush ||
+      vswc->command.used + nr_bytes > vswc->command.size ||
+      vswc->surface.used + nr_relocs > vswc->surface.size ||
+      vswc->region.used + nr_relocs > vswc->region.size) {
 #ifdef DEBUG
       vswc->must_flush = TRUE;
       debug_backtrace_capture(vswc->must_flush_stack, 1,
@@ -163,11 +217,14 @@ vmw_swc_reserve(struct svga_winsys_context *swc,
 
    assert(vswc->command.used + nr_bytes <= vswc->command.size);
    assert(vswc->surface.used + nr_relocs <= vswc->surface.size);
-
+   assert(vswc->region.used + nr_relocs <= vswc->region.size);
+   
    vswc->command.reserved = nr_bytes;
    vswc->surface.reserved = nr_relocs;
    vswc->surface.staged = 0;
-
+   vswc->region.reserved = nr_relocs;
+   vswc->region.staged = 0;
+   
    return vswc->command.buffer + vswc->command.used;
 }
 
@@ -206,20 +263,41 @@ vmw_swc_region_relocation(struct svga_winsys_context *swc,
                           unsigned flags)
 {
    struct vmw_svga_winsys_context *vswc = vmw_svga_winsys_context(swc);
-   struct SVGAGuestPtr ptr;
-   struct pb_buffer *buf = vmw_pb_buffer(buffer);
+   struct vmw_region_relocation *reloc;
    enum pipe_error ret;
+   
+   assert(vswc->region.staged < vswc->region.reserved);
 
-   if(!vmw_gmr_bufmgr_region_ptr(buf, &ptr))
-      assert(0);
-
-   ptr.offset += offset;
+   reloc = &vswc->region.relocs[vswc->region.used + vswc->region.staged];
+   reloc->where = where;
+   pb_reference(&reloc->buffer, vmw_pb_buffer(buffer));
+   reloc->offset = offset;
 
-   *where = ptr;
+   ++vswc->region.staged;
 
-   ret = pb_validate_add_buffer(vswc->validate, buf, flags);
+   ret = pb_validate_add_buffer(vswc->validate, reloc->buffer, flags);
    /* TODO: Update pipebuffer to reserve buffers and not fail here */
    assert(ret == PIPE_OK);
+
+   /*
+    * Flush preemptively the FIFO commands to keep the GMR working set within
+    * the GMR pool size.
+    *
+    * This is necessary for applications like SPECviewperf that generate huge
+    * amounts of immediate vertex data, so that we don't pile up too much of
+    * that vertex data neither in the guest nor in the host.
+    *
+    * Note that in the current implementation if a region is referred twice in
+    * a command stream, it will be accounted twice. We could detect repeated
+    * regions and count only once, but there is no incentive to do that, since
+    * regions are typically short-lived; always referred in a single command;
+    * and at the worst we just flush the commands a bit sooner, which for the
+    * SVGA virtual device it's not a performance issue since flushing commands
+    * to the FIFO won't cause flushing in the host.
+    */
+   vswc->seen_regions += reloc->buffer->base.size;
+   if(vswc->seen_regions >= VMW_GMR_POOL_SIZE/2)
+      vswc->preemptive_flush = TRUE;
 }
 
 
@@ -238,6 +316,12 @@ vmw_swc_commit(struct svga_winsys_context *swc)
    vswc->surface.used += vswc->surface.staged;
    vswc->surface.staged = 0;
    vswc->surface.reserved = 0;
+
+   assert(vswc->region.staged <= vswc->region.reserved);
+   assert(vswc->region.used + vswc->region.staged <= vswc->region.size);
+   vswc->region.used += vswc->region.staged;
+   vswc->region.staged = 0;
+   vswc->region.reserved = 0;
 }
 
 
@@ -246,6 +330,11 @@ vmw_swc_destroy(struct svga_winsys_context *swc)
 {
    struct vmw_svga_winsys_context *vswc = vmw_svga_winsys_context(swc);
    unsigned i;
+
+   for(i = 0; i < vswc->region.used; ++i) {
+      pb_reference(&vswc->region.relocs[i].buffer, NULL);
+   }
+
    for(i = 0; i < vswc->surface.used; ++i) {
       p_atomic_dec(&vswc->surface.handles[i]->validated);
       vmw_svga_winsys_surface_reference(&vswc->surface.handles[i], NULL);
@@ -279,6 +368,7 @@ vmw_svga_winsys_context_create(struct svga_winsys_screen *sws)
 
    vswc->command.size = VMW_COMMAND_SIZE;
    vswc->surface.size = VMW_SURFACE_RELOCS;
+   vswc->region.size = VMW_REGION_RELOCS;
 
    vswc->validate = pb_validate_create();
    if(!vswc->validate) {
diff --git a/src/gallium/winsys/drm/vmware/core/vmw_screen.h b/src/gallium/winsys/drm/vmware/core/vmw_screen.h
index a875107370..f1d69865e7 100644
--- a/src/gallium/winsys/drm/vmware/core/vmw_screen.h
+++ b/src/gallium/winsys/drm/vmware/core/vmw_screen.h
@@ -40,6 +40,10 @@
 
 #include "svga_winsys.h"
 
+
+#define VMW_GMR_POOL_SIZE (16*1024*1024)
+
+
 struct pb_manager;
 struct vmw_region;
 
diff --git a/src/gallium/winsys/drm/vmware/core/vmw_screen_dri.c b/src/gallium/winsys/drm/vmware/core/vmw_screen_dri.c
index 4f5ccea467..d9abde3079 100644
--- a/src/gallium/winsys/drm/vmware/core/vmw_screen_dri.c
+++ b/src/gallium/winsys/drm/vmware/core/vmw_screen_dri.c
@@ -85,6 +85,23 @@ vmw_drm_create_screen(struct drm_api *drm_api,
    struct pipe_screen *screen;
    struct dri1_create_screen_arg *dri1;
 
+   if (!arg || arg->mode == DRM_CREATE_NORMAL) {
+      struct dri1_api_version drm_ver;
+      drmVersionPtr ver;
+
+      ver = drmGetVersion(fd);
+      if (ver == NULL)
+	 return NULL;
+
+      drm_ver.major = ver->version_major;
+      drm_ver.minor = ver->version_minor;
+
+      drmFreeVersion(ver);
+      if (!vmw_dri1_check_version(&drm_ver, &drm_required,
+				  &drm_compat, "vmwgfx drm driver"))
+	 return NULL;
+   }
+
    if (arg != NULL) {
       switch (arg->mode) {
       case DRM_CREATE_NORMAL:
@@ -220,22 +237,19 @@ vmw_dri1_present_locked(struct pipe_context *locked_pipe,
    vmw_svga_winsys_surface_reference(&vsrf, NULL);
 }
 
-/**
- * FIXME: We'd probably want to cache these buffers in the
- * screen, based on handle.
- */
-
-static struct pipe_buffer *
-vmw_drm_buffer_from_handle(struct drm_api *drm_api,
-                           struct pipe_screen *screen,
-			   const char *name,
-			   unsigned handle)
+static struct pipe_texture *
+vmw_drm_texture_from_handle(struct drm_api *drm_api,
+			    struct pipe_screen *screen,
+			    struct pipe_texture *templat,
+			    const char *name,
+			    unsigned stride,
+			    unsigned handle)
 {
     struct vmw_svga_winsys_surface *vsrf;
     struct svga_winsys_surface *ssrf;
     struct vmw_winsys_screen *vws =
 	vmw_winsys_screen(svga_winsys_screen(screen));
-    struct pipe_buffer *buf;
+    struct pipe_texture *tex;
     union drm_vmw_surface_reference_arg arg;
     struct drm_vmw_surface_arg *req = &arg.req;
     struct drm_vmw_surface_create_req *rep = &arg.rep;
@@ -282,43 +296,28 @@ vmw_drm_buffer_from_handle(struct drm_api *drm_api,
 
     pipe_reference_init(&vsrf->refcnt, 1);
     p_atomic_set(&vsrf->validated, 0);
+    vsrf->screen = vws;
     vsrf->sid = handle;
     ssrf = svga_winsys_surface(vsrf);
-    buf = svga_screen_buffer_wrap_surface(screen, rep->format, ssrf);
-    if (!buf)
+    tex = svga_screen_texture_wrap_surface(screen, templat, rep->format, ssrf);
+    if (!tex)
 	vmw_svga_winsys_surface_reference(&vsrf, NULL);
 
-    return buf;
+    return tex;
   out_mip:
     vmw_ioctl_surface_destroy(vws, handle);
     return NULL;
 }
 
-static struct pipe_texture *
-vmw_drm_texture_from_handle(struct drm_api *drm_api,
-			    struct pipe_screen *screen,
-			    struct pipe_texture *templat,
-			    const char *name,
-			    unsigned stride,
-			    unsigned handle)
-{
-    struct pipe_buffer *buffer;
-    buffer = vmw_drm_buffer_from_handle(drm_api, screen, name, handle);
-
-    if (!buffer)
-	return NULL;
-
-    return screen->texture_blanket(screen, templat, &stride, buffer);
-}
-
 static boolean
-vmw_drm_handle_from_buffer(struct drm_api *drm_api,
+vmw_drm_handle_from_texture(struct drm_api *drm_api,
                            struct pipe_screen *screen,
-			   struct pipe_buffer *buffer,
+			   struct pipe_texture *texture,
+			   unsigned *stride,
 			   unsigned *handle)
 {
     struct svga_winsys_surface *surface =
-	svga_screen_buffer_get_winsys_surface(buffer);
+	svga_screen_texture_get_winsys_surface(texture);
     struct vmw_svga_winsys_surface *vsrf;
 
     if (!surface)
@@ -326,25 +325,13 @@ vmw_drm_handle_from_buffer(struct drm_api *drm_api,
 
     vsrf = vmw_svga_winsys_surface(surface);
     *handle = vsrf->sid;
+    *stride = util_format_get_nblocksx(texture->format, texture->width0) *
+       util_format_get_blocksize(texture->format);
+
     vmw_svga_winsys_surface_reference(&vsrf, NULL);
     return TRUE;
 }
 
-static boolean
-vmw_drm_handle_from_texture(struct drm_api *drm_api,
-			    struct pipe_screen *screen,
-			    struct pipe_texture *texture,
-			    unsigned *stride,
-			    unsigned *handle)
-{
-    struct pipe_buffer *buffer;
-
-    if (!svga_screen_buffer_from_texture(texture, &buffer, stride))
-	return FALSE;
-
-    return vmw_drm_handle_from_buffer(drm_api, screen, buffer, handle);
-}
-
 static struct pipe_context*
 vmw_drm_create_context(struct drm_api *drm_api,
                        struct pipe_screen *screen)
@@ -359,6 +346,7 @@ static struct dri1_api dri1_api_hooks = {
 
 static struct drm_api vmw_drm_api_hooks = {
    .name = "vmwgfx",
+   .driver_name = "vmwgfx",
    .create_screen = vmw_drm_create_screen,
    .create_context = vmw_drm_create_context,
    .texture_from_shared_handle = vmw_drm_texture_from_handle,
diff --git a/src/gallium/winsys/drm/vmware/core/vmw_screen_pools.c b/src/gallium/winsys/drm/vmware/core/vmw_screen_pools.c
index b1c24b0cb6..b9823d7857 100644
--- a/src/gallium/winsys/drm/vmware/core/vmw_screen_pools.c
+++ b/src/gallium/winsys/drm/vmware/core/vmw_screen_pools.c
@@ -53,14 +53,32 @@ vmw_pools_init(struct vmw_winsys_screen *vws)
       goto error;
 
    vws->pools.gmr_mm = mm_bufmgr_create(vws->pools.gmr,
-                                        16*1024*1024,
+                                        VMW_GMR_POOL_SIZE,
                                         12 /* 4096 alignment */);
    if(!vws->pools.gmr_mm)
       goto error;
 
+   /*
+    * GMR buffers are typically shortlived, but it's possible that at a given
+    * instance a buffer is mapped. So to avoid stalling we tell pipebuffer to
+    * forbid creation of buffers beyond half the GMR pool size,
+    *
+    * XXX: It is unclear weather we want to limit the total amount of temporary
+    * malloc memory used to backup unvalidated GMR buffers. On one hand it is
+    * preferrable to fail an allocation than exhausting the guest memory with
+    * temporary data, but on the other hand it is possible that a stupid
+    * application creates large vertex buffers and does not use them for a long
+    * time -- since the svga pipe driver only emits the DMA uploads when a
+    * buffer is used for drawing this would effectively disabling swapping GMR
+    * buffers to memory. So far, the preemptively flush already seems to keep
+    * total allocated memory within relatively small numbers, so we don't
+    * limit.
+    */
    vws->pools.gmr_fenced = fenced_bufmgr_create(
       vws->pools.gmr_mm,
-      vmw_fence_ops_create(vws));
+      vmw_fence_ops_create(vws),
+      VMW_GMR_POOL_SIZE/2,
+      ~0);
 
 #ifdef DEBUG
    vws->pools.gmr_fenced = pb_debug_manager_create(vws->pools.gmr_fenced,
diff --git a/src/gallium/winsys/drm/vmware/egl/Makefile b/src/gallium/winsys/drm/vmware/egl/Makefile
index 8e2980c318..a3e73131c3 100644
--- a/src/gallium/winsys/drm/vmware/egl/Makefile
+++ b/src/gallium/winsys/drm/vmware/egl/Makefile
@@ -1,18 +1,14 @@
-
 TOP = ../../../../../..
 include $(TOP)/configs/current
 
-LIBNAME = EGL_svga.so
+EGL_DRIVER_NAME = vmwgfx
+EGL_DRIVER_SOURCES = dummy.c
+EGL_DRIVER_LIBS =
 
-PIPE_DRIVERS = \
-	$(TOP)/src/gallium/state_trackers/egl/libegldrm.a \
+EGL_DRIVER_PIPES = \
 	$(TOP)/src/gallium/winsys/drm/vmware/core/libsvgadrm.a \
+	$(TOP)/src/gallium/drivers/softpipe/libsoftpipe.a \
 	$(TOP)/src/gallium/drivers/trace/libtrace.a \
 	$(TOP)/src/gallium/drivers/svga/libsvga.a
 
-C_SOURCES = \
-	$(COMMON_GALLIUM_SOURCES)
-
-include ../../Makefile.template
-
-symlinks:
+include ../../Makefile.egl
diff --git a/src/gallium/winsys/drm/vmware/egl/dummy.c b/src/gallium/winsys/drm/vmware/egl/dummy.c
new file mode 100644
index 0000000000..4a1bc28b0b
--- /dev/null
+++ b/src/gallium/winsys/drm/vmware/egl/dummy.c
@@ -0,0 +1 @@
+/* mklib expects at least one object file */
diff --git a/src/gallium/winsys/drm/vmware/egl_g3d/Makefile b/src/gallium/winsys/drm/vmware/egl_g3d/Makefile
deleted file mode 100644
index 3cf79924e0..0000000000
--- a/src/gallium/winsys/drm/vmware/egl_g3d/Makefile
+++ /dev/null
@@ -1,14 +0,0 @@
-TOP = ../../../../../..
-include $(TOP)/configs/current
-
-EGL_DRIVER_NAME = vmwgfx
-EGL_DRIVER_SOURCES = dummy.c
-EGL_DRIVER_LIBS =
-
-EGL_DRIVER_PIPES = \
-	$(TOP)/src/gallium/winsys/drm/vmware/core/libsvgadrm.a \
-	$(TOP)/src/gallium/drivers/softpipe/libsoftpipe.a \
-	$(TOP)/src/gallium/drivers/trace/libtrace.a \
-	$(TOP)/src/gallium/drivers/svga/libsvga.a
-
-include ../../Makefile.egl_g3d
diff --git a/src/gallium/winsys/drm/vmware/xorg/vmw_video.c b/src/gallium/winsys/drm/vmware/xorg/vmw_video.c
index b065b96346..ff3b992d07 100644
--- a/src/gallium/winsys/drm/vmware/xorg/vmw_video.c
+++ b/src/gallium/winsys/drm/vmware/xorg/vmw_video.c
@@ -649,7 +649,8 @@ vmw_video_port_play(ScrnInfoPtr pScrn, struct vmw_video_port *port,
 	return XvBadAlloc;
     }
 
-    port->currBuf = ++port->currBuf & (VMWARE_VID_NUM_BUFFERS - 1);
+    if (++(port->currBuf) >= VMWARE_VID_NUM_BUFFERS)
+	port->currBuf = 0;
 
     return Success;
 }
diff --git a/src/gallium/winsys/drm/vmware/xorg/vmw_xorg.c b/src/gallium/winsys/drm/vmware/xorg/vmw_xorg.c
index 4b208719ca..cd273d091f 100644
--- a/src/gallium/winsys/drm/vmware/xorg/vmw_xorg.c
+++ b/src/gallium/winsys/drm/vmware/xorg/vmw_xorg.c
@@ -34,10 +34,10 @@
 #include "vmw_hook.h"
 
 static void vmw_xorg_identify(int flags);
-static Bool vmw_xorg_pci_probe(DriverPtr driver,
-			       int entity_num,
-			       struct pci_device *device,
-			       intptr_t match_data);
+_X_EXPORT Bool vmw_xorg_pci_probe(DriverPtr driver,
+				  int entity_num,
+				  struct pci_device *device,
+				  intptr_t match_data);
 
 static const struct pci_id_match vmw_xorg_device_match[] = {
     {0x15ad, PCI_MATCH_ANY, PCI_MATCH_ANY, PCI_MATCH_ANY, 0, 0, 0},
@@ -126,7 +126,7 @@ vmw_xorg_identify(int flags)
 		      vmw_xorg_chipsets);
 }
 
-static Bool
+_X_EXPORT Bool
 vmw_xorg_pci_probe(DriverPtr driver,
 	  int entity_num, struct pci_device *device, intptr_t match_data)
 {
diff --git a/src/gallium/winsys/egl_xlib/Makefile b/src/gallium/winsys/egl_xlib/Makefile
deleted file mode 100644
index 3efb7ed4af..0000000000
--- a/src/gallium/winsys/egl_xlib/Makefile
+++ /dev/null
@@ -1,89 +0,0 @@
-# src/gallium/winsys/egl_xlib/Makefile
-
-# Build softpipe/xlib/EGL driver library/object: "egl_softpipe.so"
-
-
-TOP = ../../../..
-include $(TOP)/configs/current
-
-
-DRIVER_NAME = egl_softpipe.so
-
-
-INCLUDE_DIRS = \
-	-I$(TOP)/include \
-	-I$(TOP)/src/egl/main \
-	-I$(TOP)/src/mesa \
-	-I$(TOP)/src/mesa/main \
-	-I$(TOP)/src/gallium/include \
-	-I$(TOP)/src/gallium/drivers \
-	-I$(TOP)/src/gallium/auxiliary
-
-WINSYS_SOURCES = \
-	egl_xlib.c \
-	sw_winsys.c
-
-WINSYS_OBJECTS = $(WINSYS_SOURCES:.c=.o)
-
-
-LIBS = \
-	$(GALLIUM_DRIVERS) \
-	$(GALLIUM_AUXILIARIES)
-
-# XXX temporary (should create a separate lib with the GL API funcs and
-# mesa code, as done for ES 1.x, 2.x, OpenVG, etc)
-UNUSED_LIBS = \
-	$(TOP)/src/mesa/libglapi.a \
-	$(TOP)/src/mesa/libmesagallium.a \
-
-
-LOCAL_CFLAGS =
-
-
-.c.o:
-	$(CC) -c $(INCLUDE_DIRS) $(CFLAGS) $(LOCAL_CFLAGS) $< -o $@
-
-
-.PHONY: library
-
-
-default: depend library Makefile
-
-
-library: $(TOP)/$(LIB_DIR)/$(DRIVER_NAME)
-
-
-# Make the egl_softpipe.so library
-$(TOP)/$(LIB_DIR)/$(DRIVER_NAME): $(WINSYS_OBJECTS) $(LIBS)
-	$(TOP)/bin/mklib -o $(DRIVER_NAME) \
-		-linker "$(CC)" \
-		-noprefix \
-		-install $(TOP)/$(LIB_DIR) \
-		$(MKLIB_OPTIONS) $(WINSYS_OBJECTS) \
-		-Wl,--whole-archive $(LIBS) -Wl,--no-whole-archive
-
-
-depend: $(ALL_SOURCES)
-	@ echo "running $(MKDEP)"
-	@ rm -f depend  # workaround oops on gutsy?!?
-	@ touch depend
-	@ $(MKDEP) $(MKDEP_OPTIONS) $(DEFINES) $(INCLUDE_DIRS) $(ALL_SOURCES) \
-		> /dev/null 2>/dev/null
-
-
-install: default
-	$(INSTALL) -d $(INSTALL_DIR)/$(LIB_DIR)
-	@if [ -e $(TOP)/$(LIB_DIR) ]; then \
-		$(MINSTALL) $(TOP)/$(LIB_DIR)/$(DRIVER_NAME) $(INSTALL_DIR)/$(LIB_DIR); \
-	fi
-
-
-# Emacs tags
-tags:
-	etags `find . -name \*.[ch]` $(TOP)/include/GL/*.h
-
-clean:
-	-rm -f *.o *~ *.bak
-
-
-include depend
diff --git a/src/gallium/winsys/egl_xlib/egl_xlib.c b/src/gallium/winsys/egl_xlib/egl_xlib.c
deleted file mode 100644
index 1d9bac3871..0000000000
--- a/src/gallium/winsys/egl_xlib/egl_xlib.c
+++ /dev/null
@@ -1,853 +0,0 @@
-/**************************************************************************
- * 
- * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, Texas.
- * All Rights Reserved.
- * 
- * Permission is hereby granted, free of charge, to any person obtaining a
- * copy of this software and associated documentation files (the
- * "Software"), to deal in the Software without restriction, including
- * without limitation the rights to use, copy, modify, merge, publish,
- * distribute, sub license, and/or sell copies of the Software, and to
- * permit persons to whom the Software is furnished to do so, subject to
- * the following conditions:
- * 
- * The above copyright notice and this permission notice (including the
- * next paragraph) shall be included in all copies or substantial portions
- * of the Software.
- * 
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
- * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
- * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
- * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
- * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
- * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
- * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
- * 
- **************************************************************************/
-
-/**
- * EGL / softpipe / xlib winsys module
- *
- * Authors: Brian Paul
- */
-
-
-#include <dlfcn.h>
-#include <X11/Xlib.h>
-#include <X11/Xutil.h>
-
-#include "pipe/p_compiler.h"
-#include "pipe/p_format.h"
-#include "pipe/p_state.h"
-#include "pipe/internal/p_winsys_screen.h"
-#include "util/u_memory.h"
-#include "util/u_math.h"
-#include "softpipe/sp_winsys.h"
-#include "softpipe/sp_texture.h"
-
-#include "eglconfig.h"
-#include "eglconfigutil.h"
-#include "eglcontext.h"
-#include "egldisplay.h"
-#include "egldriver.h"
-#include "eglglobals.h"
-#include "egllog.h"
-#include "eglsurface.h"
-
-#include "state_tracker/st_public.h"
-
-#include "sw_winsys.h"
-
-
-/** subclass of _EGLDriver */
-struct xlib_egl_driver
-{
-   _EGLDriver Base;   /**< base class */
-   EGLint apis;
-};
-
-
-/** driver data of _EGLDisplay */
-struct xlib_egl_display
-{
-   Display *Dpy;
-
-   struct pipe_winsys *winsys;
-   struct pipe_screen *screen;
-};
-
-
-/** subclass of _EGLContext */
-struct xlib_egl_context
-{
-   _EGLContext Base;   /**< base class */
-
-   struct pipe_context *pipe;   /**< Gallium driver context */
-   struct st_context *Context;  /**< Mesa/gallium state tracker context */
-};
-
-
-/** subclass of _EGLSurface */
-struct xlib_egl_surface
-{
-   _EGLSurface Base;   /**< base class */
-
-   /* These are set for window surface */
-   Display *Dpy;  /**< The X Display of the window */
-   Window Win;    /**< The user-created window ID */
-   GC Gc;
-   XVisualInfo VisInfo;
-
-   struct pipe_winsys *winsys;
-
-   struct st_framebuffer *Framebuffer;
-};
-
-
-static void
-flush_frontbuffer(struct pipe_winsys *pws,
-                  struct pipe_surface *psurf,
-                  void *context_private);
-
-
-/** cast wrapper */
-static INLINE struct xlib_egl_driver *
-xlib_egl_driver(_EGLDriver *drv)
-{
-   return (struct xlib_egl_driver *) drv;
-}
-
-
-static INLINE struct xlib_egl_display *
-xlib_egl_display(_EGLDisplay *dpy)
-{
-   return (struct xlib_egl_display *) dpy->DriverData;
-}
-
-
-static INLINE struct xlib_egl_surface *
-lookup_surface(_EGLSurface *surf)
-{
-   return (struct xlib_egl_surface *) surf;
-}
-
-
-static INLINE struct xlib_egl_context *
-lookup_context(_EGLContext *ctx)
-{
-   return (struct xlib_egl_context *) ctx;
-}
-
-
-/**
- * Create the EGLConfigs.  (one per X visual)
- */
-static void
-create_configs(struct xlib_egl_display *xdpy, _EGLDisplay *disp)
-{
-   static const EGLint all_apis = (EGL_OPENGL_ES_BIT |
-                                   EGL_OPENGL_ES2_BIT |
-                                   EGL_OPENVG_BIT |
-                                   EGL_OPENGL_BIT);
-   XVisualInfo *visInfo, visTemplate;
-   int num_visuals, i;
-
-   /* get list of all X visuals, create an EGL config for each */
-   visTemplate.screen = DefaultScreen(xdpy->Dpy);
-   visInfo = XGetVisualInfo(xdpy->Dpy, VisualScreenMask,
-                            &visTemplate, &num_visuals);
-   if (!visInfo) {
-      printf("egl_xlib.c: couldn't get any X visuals\n");
-      abort();
-   }
-
-   for (i = 0; i < num_visuals; i++) {
-      _EGLConfig *config = calloc(1, sizeof(_EGLConfig));
-      int id = i + 1;
-      int rbits = util_bitcount(visInfo[i].red_mask);
-      int gbits = util_bitcount(visInfo[i].green_mask);
-      int bbits = util_bitcount(visInfo[i].blue_mask);
-      int abits = bbits == 8 ? 8 : 0;
-      int zbits = 24;
-      int sbits = 8;
-      int visid = visInfo[i].visualid;
-#if defined(__cplusplus) || defined(c_plusplus)
-      int vistype = visInfo[i].c_class;
-#else
-      int vistype = visInfo[i].class;
-#endif
-
-      _eglInitConfig(config, id);
-      SET_CONFIG_ATTRIB(config, EGL_BUFFER_SIZE, rbits + gbits + bbits + abits);
-      SET_CONFIG_ATTRIB(config, EGL_RED_SIZE, rbits);
-      SET_CONFIG_ATTRIB(config, EGL_GREEN_SIZE, gbits);
-      SET_CONFIG_ATTRIB(config, EGL_BLUE_SIZE, bbits);
-      SET_CONFIG_ATTRIB(config, EGL_ALPHA_SIZE, abits);
-      SET_CONFIG_ATTRIB(config, EGL_DEPTH_SIZE, zbits);
-      SET_CONFIG_ATTRIB(config, EGL_STENCIL_SIZE, sbits);
-      SET_CONFIG_ATTRIB(config, EGL_NATIVE_VISUAL_ID, visid);
-      SET_CONFIG_ATTRIB(config, EGL_NATIVE_VISUAL_TYPE, vistype);
-      SET_CONFIG_ATTRIB(config, EGL_NATIVE_RENDERABLE, EGL_FALSE);
-      SET_CONFIG_ATTRIB(config, EGL_CONFORMANT, all_apis);
-      SET_CONFIG_ATTRIB(config, EGL_RENDERABLE_TYPE, all_apis);
-      SET_CONFIG_ATTRIB(config, EGL_SURFACE_TYPE, EGL_WINDOW_BIT | EGL_PBUFFER_BIT);
-      SET_CONFIG_ATTRIB(config, EGL_BIND_TO_TEXTURE_RGBA, EGL_TRUE);
-      SET_CONFIG_ATTRIB(config, EGL_BIND_TO_TEXTURE_RGB, EGL_TRUE);
-
-      _eglAddConfig(disp, config);
-   }
-
-   XFree(visInfo);
-}
-
-
-/**
- * Called via eglInitialize(), drv->API.Initialize().
- */
-static EGLBoolean
-xlib_eglInitialize(_EGLDriver *drv, _EGLDisplay *dpy,
-                   EGLint *major, EGLint *minor)
-{
-   struct xlib_egl_driver *xdrv = xlib_egl_driver(drv);
-   struct xlib_egl_display *xdpy;
-
-   xdpy = CALLOC_STRUCT(xlib_egl_display);
-   if (!xdpy)
-      return _eglError(EGL_BAD_ALLOC, "eglInitialize");
-
-   xdpy->Dpy = (Display *) dpy->NativeDisplay;
-   if (!xdpy->Dpy) {
-      xdpy->Dpy = XOpenDisplay(NULL);
-      if (!xdpy->Dpy) {
-         free(xdpy);
-         return EGL_FALSE;
-      }
-   }
-
-   /* create winsys and pipe screen */
-   xdpy->winsys = create_sw_winsys();
-   if (!xdpy->winsys) {
-      free(xdpy);
-      return _eglError(EGL_BAD_ALLOC, "eglInitialize");
-   }
-   xdpy->winsys->flush_frontbuffer = flush_frontbuffer;
-   xdpy->screen = softpipe_create_screen(xdpy->winsys);
-   if (!xdpy->screen) {
-      free(xdpy->winsys);
-      free(xdpy);
-      return _eglError(EGL_BAD_ALLOC, "eglInitialize");
-   }
-
-   dpy->DriverData = (void *) xdpy;
-   dpy->ClientAPIsMask = xdrv->apis;
-
-   create_configs(xdpy, dpy);
-
-   /* we're supporting EGL 1.4 */
-   *major = 1;
-   *minor = 4;
-
-   return EGL_TRUE;
-}
-
-
-/**
- * Called via eglTerminate(), drv->API.Terminate().
- */
-static EGLBoolean
-xlib_eglTerminate(_EGLDriver *drv, _EGLDisplay *dpy)
-{
-   struct xlib_egl_display *xdpy = xlib_egl_display(dpy);
-
-   _eglReleaseDisplayResources(drv, dpy);
-   _eglCleanupDisplay(dpy);
-
-   xdpy->screen->destroy(xdpy->screen);
-   free(xdpy->winsys);
-
-   if (!dpy->NativeDisplay)
-      XCloseDisplay(xdpy->Dpy);
-   free(xdpy);
-
-   return EGL_TRUE;
-}
-
-
-static _EGLProc
-xlib_eglGetProcAddress(_EGLDriver *drv, const char *procname)
-{
-   return (_EGLProc) st_get_proc_address(procname);
-}
-
-
-static void
-get_drawable_visual_info(Display *dpy, Drawable d, XVisualInfo *visInfo)
-{
-   XWindowAttributes attr;
-   XVisualInfo visTemp, *vis;
-   int num_visuals;
-
-   XGetWindowAttributes(dpy, d, &attr);
-
-   visTemp.screen = DefaultScreen(dpy);
-   visTemp.visualid = attr.visual->visualid;
-   vis = XGetVisualInfo(dpy,
-                        (VisualScreenMask | VisualIDMask),
-                        &visTemp, &num_visuals);
-   if (vis)
-      *visInfo = *vis;
-
-   XFree(vis);
-}
-
-
-
-/** Get size of given window */
-static Status
-get_drawable_size(Display *dpy, Drawable d, uint *width, uint *height)
-{
-   Window root;
-   Status stat;
-   int xpos, ypos;
-   unsigned int w, h, bw, depth;
-   stat = XGetGeometry(dpy, d, &root, &xpos, &ypos, &w, &h, &bw, &depth);
-   *width = w;
-   *height = h;
-   return stat;
-}
-
-
-static void
-check_and_update_buffer_size(struct xlib_egl_surface *surface)
-{
-   uint width, height;
-   if (surface->Base.Type == EGL_PBUFFER_BIT) {
-      width = surface->Base.Width;
-      height = surface->Base.Height;
-   }
-   else {
-      get_drawable_size(surface->Dpy, surface->Win, &width, &height);
-   }
-   st_resize_framebuffer(surface->Framebuffer, width, height);
-   surface->Base.Width = width;
-   surface->Base.Height = height;
-}
-
-
-
-static void
-display_surface(struct pipe_winsys *pws,
-                struct pipe_surface *psurf,
-                struct xlib_egl_surface *xsurf)
-{
-   struct softpipe_texture *spt = softpipe_texture(psurf->texture);
-   XImage *ximage;
-   void *data;
-
-   if (xsurf->Base.Type == EGL_PBUFFER_BIT)
-      return;
-
-   ximage = XCreateImage(xsurf->Dpy,
-                         xsurf->VisInfo.visual,
-                         xsurf->VisInfo.depth,
-                         ZPixmap, 0,   /* format, offset */
-                         NULL,         /* data */
-                         0, 0,         /* size */
-                         32,           /* bitmap_pad */
-                         0);           /* bytes_per_line */
-
-
-   assert(ximage->format);
-   assert(ximage->bitmap_unit);
-
-   data = pws->buffer_map(pws, spt->buffer, 0);
-
-   /* update XImage's fields */
-   ximage->data = data;
-   ximage->width = psurf->width;
-   ximage->height = psurf->height;
-   ximage->bytes_per_line = spt->stride[psurf->level];
-   
-   XPutImage(xsurf->Dpy, xsurf->Win, xsurf->Gc,
-             ximage, 0, 0, 0, 0, psurf->width, psurf->height);
-
-   XSync(xsurf->Dpy, 0);
-
-   ximage->data = NULL;
-   XDestroyImage(ximage);
-
-   pws->buffer_unmap(pws, spt->buffer);
-}
-
-
-
-/** Display gallium surface in X window */
-static void
-flush_frontbuffer(struct pipe_winsys *pws,
-                  struct pipe_surface *psurf,
-                  void *context_private)
-{
-   struct xlib_egl_surface *xsurf = (struct xlib_egl_surface *) context_private;
-   display_surface(pws, psurf, xsurf);
-}
-
-
-
-/**
- * Called via eglCreateContext(), drv->API.CreateContext().
- */
-static _EGLContext *
-xlib_eglCreateContext(_EGLDriver *drv, _EGLDisplay *dpy, _EGLConfig *conf,
-                      _EGLContext *share_list, const EGLint *attrib_list)
-{
-   struct xlib_egl_display *xdpy = xlib_egl_display(dpy);
-   struct xlib_egl_context *ctx;
-   struct st_context *share_ctx = NULL; /* XXX fix */
-   __GLcontextModes visual;
-
-   ctx = CALLOC_STRUCT(xlib_egl_context);
-   if (!ctx)
-      return NULL;
-
-   /* let EGL lib init the common stuff */
-   if (!_eglInitContext(drv, &ctx->Base, conf, attrib_list)) {
-      free(ctx);
-      return NULL;
-   }
-
-   /* API-dependent context creation */
-   switch (ctx->Base.ClientAPI) {
-   case EGL_OPENVG_API:
-   case EGL_OPENGL_ES_API:
-      _eglLog(_EGL_DEBUG, "Create Context for ES version %d\n",
-              ctx->Base.ClientVersion);
-      /* fall-through */
-   case EGL_OPENGL_API:
-      /* create a softpipe context */
-      ctx->pipe = softpipe_create(xdpy->screen);
-      /* Now do xlib / state tracker inits here */
-      _eglConfigToContextModesRec(conf, &visual);
-      ctx->Context = st_create_context(ctx->pipe, &visual, share_ctx);
-      break;
-   default:
-      _eglError(EGL_BAD_MATCH, "eglCreateContext(unsupported API)");
-      free(ctx);
-      return NULL;
-   }
-
-   return &ctx->Base;
-}
-
-
-static EGLBoolean
-xlib_eglDestroyContext(_EGLDriver *drv, _EGLDisplay *dpy, _EGLContext *ctx)
-{
-   struct xlib_egl_context *context = lookup_context(ctx);
-
-   if (!_eglIsContextBound(&context->Base)) {
-      /* API-dependent clean-up */
-      switch (context->Base.ClientAPI) {
-      case EGL_OPENGL_ES_API:
-      case EGL_OPENVG_API:
-         /* fall-through */
-      case EGL_OPENGL_API:
-         st_destroy_context(context->Context);
-         break;
-      default:
-         assert(0);
-      }
-      free(context);
-   }
-   return EGL_TRUE;
-}
-
-
-/**
- * Called via eglMakeCurrent(), drv->API.MakeCurrent().
- */
-static EGLBoolean
-xlib_eglMakeCurrent(_EGLDriver *drv, _EGLDisplay *dpy,
-                    _EGLSurface *draw, _EGLSurface *read, _EGLContext *ctx)
-{
-   struct xlib_egl_context *context = lookup_context(ctx);
-   struct xlib_egl_surface *draw_surf = lookup_surface(draw);
-   struct xlib_egl_surface *read_surf = lookup_surface(read);
-   struct st_context *oldcontext = NULL;
-   _EGLContext *oldctx;
-
-   oldctx = _eglGetCurrentContext();
-   if (oldctx && _eglIsContextLinked(oldctx))
-      oldcontext = st_get_current();
-
-   if (!_eglMakeCurrent(drv, dpy, draw, read, ctx))
-      return EGL_FALSE;
-
-   /* Flush before switching context.  Check client API? */
-   if (oldcontext)
-      st_flush(oldcontext, PIPE_FLUSH_RENDER_CACHE | PIPE_FLUSH_FRAME, NULL);
-   st_make_current((context ? context->Context : NULL),
-                   (draw_surf ? draw_surf->Framebuffer : NULL),
-                   (read_surf ? read_surf->Framebuffer : NULL));
-
-   if (draw_surf)
-      check_and_update_buffer_size(draw_surf);
-   if (read_surf && read_surf != draw_surf)
-      check_and_update_buffer_size(draw_surf);
-
-   return EGL_TRUE;
-}
-
-
-static enum pipe_format
-choose_color_format(const __GLcontextModes *visual)
-{
-   if (visual->redBits == 8 &&
-       visual->greenBits == 8 &&
-       visual->blueBits == 8 &&
-       visual->alphaBits == 8) {
-      /* XXX this really also depends on the ordering of R,G,B,A */
-      return PIPE_FORMAT_A8R8G8B8_UNORM;
-   }
-   else {
-      assert(0);
-      return PIPE_FORMAT_NONE;
-   }
-}
-
-
-static enum pipe_format
-choose_depth_format(const __GLcontextModes *visual)
-{
-   if (visual->depthBits > 0)
-      return PIPE_FORMAT_S8Z24_UNORM;
-   else
-      return PIPE_FORMAT_NONE;
-}
-
-
-static enum pipe_format
-choose_stencil_format(const __GLcontextModes *visual)
-{
-   if (visual->stencilBits > 0)
-      return PIPE_FORMAT_S8Z24_UNORM;
-   else
-      return PIPE_FORMAT_NONE;
-}
-
-
-/**
- * Called via eglCreateWindowSurface(), drv->API.CreateWindowSurface().
- */
-static _EGLSurface *
-xlib_eglCreateWindowSurface(_EGLDriver *drv, _EGLDisplay *disp, _EGLConfig *conf,
-                            NativeWindowType window, const EGLint *attrib_list)
-{
-   struct xlib_egl_display *xdpy = xlib_egl_display(disp);
-   struct xlib_egl_surface *surf;
-   __GLcontextModes visual;
-   uint width, height;
-
-   surf = CALLOC_STRUCT(xlib_egl_surface);
-   if (!surf)
-      return NULL;
-
-   /* Let EGL lib init the common stuff */
-   if (!_eglInitSurface(drv, &surf->Base, EGL_WINDOW_BIT,
-                        conf, attrib_list)) {
-      free(surf);
-      return NULL;
-   }
-
-   /*
-    * Now init the Xlib and gallium stuff
-    */
-   surf->Win = (Window) window;  /* The X window ID */
-   surf->Dpy = xdpy->Dpy;  /* The X display */
-   surf->Gc = XCreateGC(surf->Dpy, surf->Win, 0, NULL);
-
-   surf->winsys = xdpy->winsys;
-
-   _eglConfigToContextModesRec(conf, &visual);
-   get_drawable_size(surf->Dpy, surf->Win, &width, &height);
-   get_drawable_visual_info(surf->Dpy, surf->Win, &surf->VisInfo);
-
-   surf->Base.Width = width;
-   surf->Base.Height = height;
-
-   /* Create GL statetracker framebuffer */
-   surf->Framebuffer = st_create_framebuffer(&visual,
-                                             choose_color_format(&visual),
-                                             choose_depth_format(&visual),
-                                             choose_stencil_format(&visual),
-                                             width, height,
-                                             (void *) surf);
-
-   st_resize_framebuffer(surf->Framebuffer, width, height);
-
-   return &surf->Base;
-}
-
-
-static _EGLSurface *
-xlib_eglCreatePbufferSurface(_EGLDriver *drv, _EGLDisplay *disp, _EGLConfig *conf,
-                             const EGLint *attrib_list)
-{
-   struct xlib_egl_display *xdpy = xlib_egl_display(disp);
-   struct xlib_egl_surface *surf;
-   __GLcontextModes visual;
-   uint width, height;
-   EGLBoolean bind_texture;
-
-   surf = CALLOC_STRUCT(xlib_egl_surface);
-   if (!surf) {
-      _eglError(EGL_BAD_ALLOC, "eglCreatePbufferSurface");
-      return NULL;
-   }
-
-   if (!_eglInitSurface(drv, &surf->Base, EGL_PBUFFER_BIT,
-                        conf, attrib_list)) {
-      free(surf);
-      return NULL;
-   }
-   if (surf->Base.Width < 0 || surf->Base.Height < 0) {
-      _eglError(EGL_BAD_PARAMETER, "eglCreatePbufferSurface");
-      free(surf);
-      return NULL;
-   }
-
-   bind_texture = (surf->Base.TextureFormat != EGL_NO_TEXTURE);
-   width = (uint) surf->Base.Width;
-   height = (uint) surf->Base.Height;
-   if ((surf->Base.TextureTarget == EGL_NO_TEXTURE && bind_texture) ||
-       (surf->Base.TextureTarget != EGL_NO_TEXTURE && !bind_texture)) {
-      _eglError(EGL_BAD_MATCH, "eglCreatePbufferSurface");
-      free(surf);
-      return NULL;
-   }
-   /* a framebuffer of zero width or height confuses st */
-   if (width == 0 || height == 0) {
-      _eglError(EGL_BAD_MATCH, "eglCreatePbufferSurface");
-      free(surf);
-      return NULL;
-   }
-   /* no mipmap generation */
-   if (surf->Base.MipmapTexture) {
-      _eglError(EGL_BAD_MATCH, "eglCreatePbufferSurface");
-      free(surf);
-      return NULL;
-   }
-
-   surf->winsys = xdpy->winsys;
-
-   _eglConfigToContextModesRec(conf, &visual);
-
-   /* Create GL statetracker framebuffer */
-   surf->Framebuffer = st_create_framebuffer(&visual,
-                                             choose_color_format(&visual),
-                                             choose_depth_format(&visual),
-                                             choose_stencil_format(&visual),
-                                             width, height,
-                                             (void *) surf);
-   st_resize_framebuffer(surf->Framebuffer, width, height);
-
-   return &surf->Base;
-}
-
-
-static EGLBoolean
-xlib_eglDestroySurface(_EGLDriver *drv, _EGLDisplay *dpy, _EGLSurface *surface)
-{
-   struct xlib_egl_surface *surf = lookup_surface(surface);
-   if (!_eglIsSurfaceBound(&surf->Base)) {
-      if (surf->Base.Type != EGL_PBUFFER_BIT)
-         XFreeGC(surf->Dpy, surf->Gc);
-      st_unreference_framebuffer(surf->Framebuffer);
-      free(surf);
-   }
-   return EGL_TRUE;
-}
-
-
-static EGLBoolean
-xlib_eglBindTexImage(_EGLDriver *drv, _EGLDisplay *dpy,
-                     _EGLSurface *surface, EGLint buffer)
-{
-   struct xlib_egl_surface *xsurf = lookup_surface(surface);
-   struct xlib_egl_context *xctx;
-   struct pipe_surface *psurf;
-   enum pipe_format format;
-   int target;
-
-   if (!xsurf || xsurf->Base.Type != EGL_PBUFFER_BIT)
-      return _eglError(EGL_BAD_SURFACE, "eglBindTexImage");
-   if (buffer != EGL_BACK_BUFFER)
-      return _eglError(EGL_BAD_PARAMETER, "eglBindTexImage");
-   if (xsurf->Base.BoundToTexture)
-      return _eglError(EGL_BAD_ACCESS, "eglBindTexImage");
-
-   /* this should be updated when choose_color_format is */
-   switch (xsurf->Base.TextureFormat) {
-   case EGL_TEXTURE_RGB:
-      format = PIPE_FORMAT_R8G8B8_UNORM;
-      break;
-   case EGL_TEXTURE_RGBA:
-      format = PIPE_FORMAT_A8R8G8B8_UNORM;
-      break;
-   default:
-      return _eglError(EGL_BAD_MATCH, "eglBindTexImage");
-   }
-
-   switch (xsurf->Base.TextureTarget) {
-   case EGL_TEXTURE_2D:
-      target = ST_TEXTURE_2D;
-      break;
-   default:
-      return _eglError(EGL_BAD_MATCH, "eglBindTexImage");
-   }
-
-   /* flush properly */
-   if (eglGetCurrentSurface(EGL_DRAW) == surface) {
-      xctx = lookup_context(_eglGetCurrentContext());
-      st_flush(xctx->Context, PIPE_FLUSH_RENDER_CACHE | PIPE_FLUSH_FRAME,
-               NULL);
-   }
-   else if (_eglIsSurfaceBound(&xsurf->Base)) {
-      xctx = lookup_context(xsurf->Base.Binding);
-      if (xctx)
-         st_finish(xctx->Context);
-   }
-
-   st_get_framebuffer_surface(xsurf->Framebuffer, ST_SURFACE_BACK_LEFT,
-                              &psurf);
-   st_bind_texture_surface(psurf, target, xsurf->Base.MipmapLevel, format);
-   xsurf->Base.BoundToTexture = EGL_TRUE;
-
-   return EGL_TRUE;
-}
-
-
-static EGLBoolean
-xlib_eglReleaseTexImage(_EGLDriver *drv, _EGLDisplay *dpy, _EGLSurface *surface,
-                        EGLint buffer)
-{
-   struct xlib_egl_surface *xsurf = lookup_surface(surface);
-   struct pipe_surface *psurf;
-
-   if (!xsurf || xsurf->Base.Type != EGL_PBUFFER_BIT ||
-       !xsurf->Base.BoundToTexture)
-      return _eglError(EGL_BAD_SURFACE, "eglReleaseTexImage");
-   if (buffer != EGL_BACK_BUFFER)
-      return _eglError(EGL_BAD_PARAMETER, "eglReleaseTexImage");
-
-   st_get_framebuffer_surface(xsurf->Framebuffer, ST_SURFACE_BACK_LEFT,
-                              &psurf);
-   st_unbind_texture_surface(psurf, ST_TEXTURE_2D, xsurf->Base.MipmapLevel);
-   xsurf->Base.BoundToTexture = EGL_FALSE;
-
-   return EGL_TRUE;
-}
-
-
-static EGLBoolean
-xlib_eglSwapBuffers(_EGLDriver *drv, _EGLDisplay *dpy, _EGLSurface *draw)
-{
-   struct xlib_egl_surface *xsurf = lookup_surface(draw);
-   struct pipe_winsys *pws = xsurf->winsys;
-   struct pipe_surface *psurf;
-
-   st_get_framebuffer_surface(xsurf->Framebuffer, ST_SURFACE_BACK_LEFT,
-         &psurf);
-
-   st_notify_swapbuffers(xsurf->Framebuffer);
-
-   display_surface(pws, psurf, xsurf);
-
-   check_and_update_buffer_size(xsurf);
-
-   return EGL_TRUE;
-}
-
-
-/**
- * Determine which API(s) is(are) present by looking for some specific
- * global symbols.
- */
-static EGLint
-find_supported_apis(void)
-{
-   EGLint mask = 0;
-   void *handle;
-
-   handle = dlopen(NULL, RTLD_LAZY | RTLD_LOCAL);
-   if(!handle)
-      return mask;
-
-   if (dlsym(handle, "st_api_OpenGL_ES1"))
-      mask |= EGL_OPENGL_ES_BIT;
-
-   if (dlsym(handle, "st_api_OpenGL_ES2"))
-      mask |= EGL_OPENGL_ES2_BIT;
-
-   if (dlsym(handle, "st_api_OpenGL"))
-      mask |= EGL_OPENGL_BIT;
-
-   if (dlsym(handle, "st_api_OpenVG"))
-      mask |= EGL_OPENVG_BIT;
-
-   dlclose(handle);
-
-   return mask;
-}
-
-
-static void
-xlib_Unload(_EGLDriver *drv)
-{
-   struct xlib_egl_driver *xdrv = xlib_egl_driver(drv);
-   free(xdrv);
-}
-
-
-/**
- * This is the main entrypoint into the driver.
- * Called by libEGL to instantiate an _EGLDriver object.
- */
-_EGLDriver *
-_eglMain(const char *args)
-{
-   struct xlib_egl_driver *xdrv;
-
-   _eglLog(_EGL_INFO, "Entering EGL/Xlib _eglMain(%s)", args);
-
-   xdrv = CALLOC_STRUCT(xlib_egl_driver);
-   if (!xdrv)
-      return NULL;
-
-   _eglInitDriverFallbacks(&xdrv->Base);
-   xdrv->Base.API.Initialize = xlib_eglInitialize;
-   xdrv->Base.API.Terminate = xlib_eglTerminate;
-   xdrv->Base.API.GetProcAddress = xlib_eglGetProcAddress;
-   xdrv->Base.API.CreateContext = xlib_eglCreateContext;
-   xdrv->Base.API.DestroyContext = xlib_eglDestroyContext;
-   xdrv->Base.API.CreateWindowSurface = xlib_eglCreateWindowSurface;
-   xdrv->Base.API.CreatePbufferSurface = xlib_eglCreatePbufferSurface;
-   xdrv->Base.API.DestroySurface = xlib_eglDestroySurface;
-   xdrv->Base.API.BindTexImage = xlib_eglBindTexImage;
-   xdrv->Base.API.ReleaseTexImage = xlib_eglReleaseTexImage;
-   xdrv->Base.API.MakeCurrent = xlib_eglMakeCurrent;
-   xdrv->Base.API.SwapBuffers = xlib_eglSwapBuffers;
-
-   xdrv->apis = find_supported_apis();
-   if (xdrv->apis == 0x0) {
-      /* the app isn't directly linked with any EGL-supprted APIs
-       * (such as libGLESv2.so) so use an EGL utility to see what
-       * APIs might be loaded dynamically on this system.
-       */
-      xdrv->apis = _eglFindAPIs();
-   }
-
-   xdrv->Base.Name = "Xlib/softpipe";
-   xdrv->Base.Unload = xlib_Unload;
-
-   return &xdrv->Base;
-}
diff --git a/src/gallium/winsys/egl_xlib/sw_winsys.c b/src/gallium/winsys/egl_xlib/sw_winsys.c
deleted file mode 100644
index 6ee3ede38c..0000000000
--- a/src/gallium/winsys/egl_xlib/sw_winsys.c
+++ /dev/null
@@ -1,231 +0,0 @@
-/**************************************************************************
- * 
- * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, Texas.
- * All Rights Reserved.
- * 
- * Permission is hereby granted, free of charge, to any person obtaining a
- * copy of this software and associated documentation files (the
- * "Software"), to deal in the Software without restriction, including
- * without limitation the rights to use, copy, modify, merge, publish,
- * distribute, sub license, and/or sell copies of the Software, and to
- * permit persons to whom the Software is furnished to do so, subject to
- * the following conditions:
- * 
- * The above copyright notice and this permission notice (including the
- * next paragraph) shall be included in all copies or substantial portions
- * of the Software.
- * 
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
- * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
- * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
- * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
- * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
- * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
- * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
- * 
- **************************************************************************/
-
-/**
- * Totally software-based winsys layer.
- * Note that the one winsys function that we can't implement here
- * is flush_frontbuffer().
- * Whoever uses this code will have to provide that.
- *
- * Authors: Brian Paul
- */
-
-
-#include "pipe/internal/p_winsys_screen.h"
-#include "pipe/p_state.h"
-#include "pipe/p_inlines.h"
-#include "util/u_format.h"
-#include "util/u_math.h"
-#include "util/u_memory.h"
-
-#include "sw_winsys.h"
-
-
-
-/** Subclass of pipe_winsys */
-struct sw_pipe_winsys
-{
-   struct pipe_winsys Base;
-   /* no extra fields for now */
-};
-
-
-/** subclass of pipe_buffer */
-struct sw_pipe_buffer
-{
-   struct pipe_buffer Base;
-   boolean UserBuffer;  /** Is this a user-space buffer? */
-   void *Data;
-   void *Mapped;
-};
-
-
-/** cast wrapper */
-static INLINE struct sw_pipe_buffer *
-sw_pipe_buffer(struct pipe_buffer *b)
-{
-   return (struct sw_pipe_buffer *) b;
-}
-
-
-static const char *
-get_name(struct pipe_winsys *pws)
-{
-   return "software";
-}
-
-
-/** Create new pipe_buffer and allocate storage of given size */
-static struct pipe_buffer *
-buffer_create(struct pipe_winsys *pws, 
-              unsigned alignment, 
-              unsigned usage,
-              unsigned size)
-{
-   struct sw_pipe_buffer *buffer = CALLOC_STRUCT(sw_pipe_buffer);
-   if (!buffer)
-      return NULL;
-
-   pipe_reference_init(&buffer->Base.reference, 1);
-   buffer->Base.alignment = alignment;
-   buffer->Base.usage = usage;
-   buffer->Base.size = size;
-
-   /* align to 16-byte multiple for Cell */
-   buffer->Data = align_malloc(size, MAX2(alignment, 16));
-
-   return &buffer->Base;
-}
-
-
-/**
- * Create buffer which wraps user-space data.
- */
-static struct pipe_buffer *
-user_buffer_create(struct pipe_winsys *pws, void *ptr, unsigned bytes)
-{
-   struct sw_pipe_buffer *buffer = CALLOC_STRUCT(sw_pipe_buffer);
-   if (!buffer)
-      return NULL;
-
-   pipe_reference_init(&buffer->Base.reference, 1);
-   buffer->Base.size = bytes;
-   buffer->UserBuffer = TRUE;
-   buffer->Data = ptr;
-
-   return &buffer->Base;
-}
-
-
-static void *
-buffer_map(struct pipe_winsys *pws, struct pipe_buffer *buf, unsigned flags)
-{
-   struct sw_pipe_buffer *buffer = sw_pipe_buffer(buf);
-   buffer->Mapped = buffer->Data;
-   return buffer->Mapped;
-}
-
-
-static void
-buffer_unmap(struct pipe_winsys *pws, struct pipe_buffer *buf)
-{
-   struct sw_pipe_buffer *buffer = sw_pipe_buffer(buf);
-   buffer->Mapped = NULL;
-}
-
-
-static void
-buffer_destroy(struct pipe_buffer *buf)
-{
-   struct sw_pipe_buffer *buffer = sw_pipe_buffer(buf);
-
-   if (buffer->Data && !buffer->UserBuffer) {
-      align_free(buffer->Data);
-      buffer->Data = NULL;
-   }
-
-   free(buffer);
-}
-
-
-static struct pipe_buffer *
-surface_buffer_create(struct pipe_winsys *winsys,
-                      unsigned width, unsigned height,
-                      enum pipe_format format, 
-                      unsigned usage,
-                      unsigned tex_usage,
-                      unsigned *stride)
-{
-   const unsigned alignment = 64;
-   unsigned nblocksy;
-
-   nblocksy = util_format_get_nblocksy(format, height);
-   *stride = align(util_format_get_stride(format, width), alignment);
-
-   return winsys->buffer_create(winsys, alignment,
-                                usage,
-                                *stride * nblocksy);
-}
-
-
-static void
-fence_reference(struct pipe_winsys *sws, struct pipe_fence_handle **ptr,
-                struct pipe_fence_handle *fence)
-{
-   /* no-op */
-}
-
-
-static int
-fence_signalled(struct pipe_winsys *sws, struct pipe_fence_handle *fence,
-                unsigned flag)
-{
-   /* no-op */
-   return 0;
-}
-
-
-static int
-fence_finish(struct pipe_winsys *sws, struct pipe_fence_handle *fence,
-             unsigned flag)
-{
-   /* no-op */
-   return 0;
-}
-
-
-/**
- * Create/return a new pipe_winsys object.
- */
-struct pipe_winsys *
-create_sw_winsys(void)
-{
-   struct sw_pipe_winsys *ws = CALLOC_STRUCT(sw_pipe_winsys);
-   if (!ws)
-      return NULL;
-
-   /* Fill in this struct with callbacks that pipe will need to
-    * communicate with the window system, buffer manager, etc. 
-    */
-   ws->Base.buffer_create = buffer_create;
-   ws->Base.user_buffer_create = user_buffer_create;
-   ws->Base.buffer_map = buffer_map;
-   ws->Base.buffer_unmap = buffer_unmap;
-   ws->Base.buffer_destroy = buffer_destroy;
-
-   ws->Base.surface_buffer_create = surface_buffer_create;
-
-   ws->Base.fence_reference = fence_reference;
-   ws->Base.fence_signalled = fence_signalled;
-   ws->Base.fence_finish = fence_finish;
-
-   ws->Base.flush_frontbuffer = NULL; /* not implemented here! */
-
-   ws->Base.get_name = get_name;
-
-   return &ws->Base;
-}
diff --git a/src/gallium/winsys/egl_xlib/sw_winsys.h b/src/gallium/winsys/egl_xlib/sw_winsys.h
deleted file mode 100644
index f96c5a14b0..0000000000
--- a/src/gallium/winsys/egl_xlib/sw_winsys.h
+++ /dev/null
@@ -1,40 +0,0 @@
-/**************************************************************************
- * 
- * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, Texas.
- * All Rights Reserved.
- * 
- * Permission is hereby granted, free of charge, to any person obtaining a
- * copy of this software and associated documentation files (the
- * "Software"), to deal in the Software without restriction, including
- * without limitation the rights to use, copy, modify, merge, publish,
- * distribute, sub license, and/or sell copies of the Software, and to
- * permit persons to whom the Software is furnished to do so, subject to
- * the following conditions:
- * 
- * The above copyright notice and this permission notice (including the
- * next paragraph) shall be included in all copies or substantial portions
- * of the Software.
- * 
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
- * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
- * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
- * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
- * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
- * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
- * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
- * 
- **************************************************************************/
-
-
-#ifndef SW_WINSYS_H
-#define SW_WINSYS_H
-
-
-struct pipe_winsys;
-
-
-extern struct pipe_winsys *
-create_sw_winsys(void);
-
-
-#endif /* SW_WINSYS_H */
diff --git a/src/gallium/winsys/xlib/xlib.c b/src/gallium/winsys/xlib/xlib.c
index 13117712c2..67617a470d 100644
--- a/src/gallium/winsys/xlib/xlib.c
+++ b/src/gallium/winsys/xlib/xlib.c
@@ -105,3 +105,34 @@ extern void (*linker_foo(const unsigned char *procName))()
 {
    return glXGetProcAddress(procName);
 }
+
+
+/**
+ * When GLX_INDIRECT_RENDERING is defined, some symbols are missing in
+ * libglapi.a.  We need to define them here.
+ */
+#ifdef GLX_INDIRECT_RENDERING
+
+#define GL_GLEXT_PROTOTYPES
+#include "GL/gl.h"
+#include "glapi/glapi.h"
+#include "glapi/glapitable.h"
+#include "glapi/glapidispatch.h"
+
+#if defined(USE_MGL_NAMESPACE)
+#define NAME(func)  mgl##func
+#else
+#define NAME(func)  gl##func
+#endif
+
+#define DISPATCH(FUNC, ARGS, MESSAGE)		\
+   CALL_ ## FUNC(GET_DISPATCH(), ARGS);
+
+#define RETURN_DISPATCH(FUNC, ARGS, MESSAGE) 	\
+   return CALL_ ## FUNC(GET_DISPATCH(), ARGS);
+
+/* skip normal ones */
+#define _GLAPI_SKIP_NORMAL_ENTRY_POINTS
+#include "glapi/glapitemp.h"
+
+#endif /* GLX_INDIRECT_RENDERING */
diff --git a/src/gallium/winsys/xlib/xlib_softpipe.c b/src/gallium/winsys/xlib/xlib_softpipe.c
index f7c0099584..bf78aca686 100644
--- a/src/gallium/winsys/xlib/xlib_softpipe.c
+++ b/src/gallium/winsys/xlib/xlib_softpipe.c
@@ -63,7 +63,7 @@ struct xm_buffer
 
    XImage *tempImage;
 #ifdef USE_XSHM
-   int shm;
+   boolean shm;         /** Is this a shared memory buffer? */
    XShmSegmentInfo shminfo;
 #endif
 };
@@ -152,7 +152,7 @@ alloc_shm_ximage(struct xm_buffer *b, struct xmesa_buffer *xmb,
                                   &b->shminfo,
                                   width, height);
    if (b->tempImage == NULL) {
-      b->shm = 0;
+      b->shm = FALSE;
       return;
    }
 
@@ -169,12 +169,12 @@ alloc_shm_ximage(struct xm_buffer *b, struct xmesa_buffer *xmb,
       mesaXErrorFlag = 0;
       XDestroyImage(b->tempImage);
       b->tempImage = NULL;
-      b->shm = 0;
+      b->shm = FALSE;
       (void) XSetErrorHandler(old_handler);
       return;
    }
 
-   b->shm = 1;
+   b->shm = TRUE;
 }
 
 #endif /* USE_XSHM */
@@ -204,6 +204,14 @@ xm_buffer_destroy(struct pipe_buffer *buf)
 {
    struct xm_buffer *oldBuf = xm_buffer(buf);
 
+   /*
+    * Note oldBuf->data may point to one of three things:
+    * 1. XShm shared memory image data
+    * 2. User-provided (wrapped) memory, see xm_user_buffer_create()
+    * 3. Regular, malloc'd memory
+    * We need to be careful with freeing that data now.
+    */
+
    if (oldBuf->data) {
 #ifdef USE_XSHM
       if (oldBuf->shminfo.shmid >= 0) {
@@ -213,12 +221,20 @@ xm_buffer_destroy(struct pipe_buffer *buf)
          oldBuf->shminfo.shmid = -1;
          oldBuf->shminfo.shmaddr = (char *) -1;
       }
-      else
+
+      if (oldBuf->shm) {
+         oldBuf->data = NULL;
+      }
+
+      if (oldBuf->tempImage) {
+         XDestroyImage(oldBuf->tempImage);
+         oldBuf->tempImage = NULL;
+      }
 #endif
-      {
-         if (!oldBuf->userBuffer) {
-            align_free(oldBuf->data);
-         }
+
+      if (oldBuf->data && !oldBuf->userBuffer) {
+         /* this was regular malloc'd memory */
+         align_free(oldBuf->data);
       }
 
       oldBuf->data = NULL;
@@ -327,10 +343,8 @@ xm_buffer_create(struct pipe_winsys *pws,
    buffer->base.usage = usage;
    buffer->base.size = size;
 
-   if (buffer->data == NULL) {
-      /* align to 16-byte multiple for Cell */
-      buffer->data = align_malloc(size, max(alignment, 16));
-   }
+   /* align to 16-byte multiple for Cell */
+   buffer->data = align_malloc(size, max(alignment, 16));
 
    return &buffer->base;
 }