18 files changed, 748 insertions, 469 deletions

diff --git a/src/mesa/drivers/dri/i965/Makefile b/src/mesa/drivers/dri/i965/Makefile
index 2934414d99..9712c38725 100644
--- a/src/mesa/drivers/dri/i965/Makefile
+++ b/src/mesa/drivers/dri/i965/Makefile
@@ -14,6 +14,7 @@ DRIVER_SOURCES = \
 	intel_decode.c \
 	intel_extensions.c \
 	intel_fbo.c \
+	intel_generatemipmap.c \
 	intel_mipmap_tree.c \
 	intel_regions.c \
 	intel_screen.c \
@@ -69,6 +70,7 @@ DRIVER_SOURCES = \
 	brw_vs_constval.c \
 	brw_vs_emit.c \
 	brw_vs_state.c \
+	brw_vs_surface_state.c \
 	brw_vtbl.c \
 	brw_wm.c \
 	brw_wm_debug.c \
diff --git a/src/mesa/drivers/dri/i965/brw_context.c b/src/mesa/drivers/dri/i965/brw_context.c
index d96ff29310..4dbe551d83 100644
--- a/src/mesa/drivers/dri/i965/brw_context.c
+++ b/src/mesa/drivers/dri/i965/brw_context.c
@@ -118,6 +118,8 @@ GLboolean brwCreateContext( const __GLcontextModes *mesaVis,
    ctx->Const.MaxCubeTextureLevels = 12;
    ctx->Const.MaxTextureRectSize = (1<<12);
    
+   ctx->Const.MaxTextureMaxAnisotropy = 16.0;
+
    /* if conformance mode is set, swrast can handle any size AA point */
    ctx->Const.MaxPointSizeAA = 255.0;
 
diff --git a/src/mesa/drivers/dri/i965/brw_context.h b/src/mesa/drivers/dri/i965/brw_context.h
index a0b3b06309..873fc8ffff 100644
--- a/src/mesa/drivers/dri/i965/brw_context.h
+++ b/src/mesa/drivers/dri/i965/brw_context.h
@@ -131,6 +131,7 @@ struct brw_context;
 #define BRW_NEW_WM_INPUT_DIMENSIONS     0x100
 #define BRW_NEW_INPUT_VARYING           0x200
 #define BRW_NEW_PSP                     0x800
+#define BRW_NEW_WM_SURFACES		0x1000
 #define BRW_NEW_FENCE                   0x2000
 #define BRW_NEW_INDICES			0x4000
 #define BRW_NEW_VERTICES		0x8000
@@ -161,6 +162,7 @@ struct brw_vertex_program {
    struct gl_vertex_program program;
    GLuint id;
    dri_bo *const_buffer;    /** Program constant buffer/surface */
+   GLboolean use_const_buffer;
 };
 
 
@@ -171,6 +173,7 @@ struct brw_fragment_program {
    GLboolean isGLSL;  /**< really, any IF/LOOP/CONT/BREAK instructions */
 
    dri_bo *const_buffer;    /** Program constant buffer/surface */
+   GLboolean use_const_buffer;
 };
 
 
@@ -243,6 +246,9 @@ struct brw_vs_ouput_sizes {
 };
 
 
+/** Number of general purpose registers (VS, WM, etc) */
+#define BRW_MAX_GRF 128
+
 /** Number of texture sampler units */
 #define BRW_MAX_TEX_UNIT 16
 
@@ -448,8 +454,6 @@ struct brw_context
 
    struct {
       struct brw_state_flags dirty;
-      struct brw_tracked_state **atoms;
-      GLuint nr_atoms;
 
       GLuint nr_color_regions;
       struct intel_region *color_regions[MAX_DRAW_BUFFERS];
@@ -469,7 +473,8 @@ struct brw_context
       int validated_bo_count;
    } state;
 
-   struct brw_cache cache;
+   struct brw_cache cache;  /** non-surface items */
+   struct brw_cache surface_cache;  /* surface items */
    struct brw_cached_batch_item *cached_batch_items;
 
    struct {
@@ -553,11 +558,6 @@ struct brw_context
       GLuint vs_size;
       GLuint total_size;
 
-      /* Dynamic tracker which changes to reflect the state referenced
-       * by active fp and vp program parameters:
-       */
-      struct brw_tracked_state tracked_state;
-
       dri_bo *curbe_bo;
       /** Offset within curbe_bo of space for current curbe entry */
       GLuint curbe_offset;
diff --git a/src/mesa/drivers/dri/i965/brw_curbe.c b/src/mesa/drivers/dri/i965/brw_curbe.c
index 18b187ed1d..a1a6c53d0e 100644
--- a/src/mesa/drivers/dri/i965/brw_curbe.c
+++ b/src/mesa/drivers/dri/i965/brw_curbe.c
@@ -36,6 +36,7 @@
 #include "main/macros.h"
 #include "main/enums.h"
 #include "shader/prog_parameter.h"
+#include "shader/prog_print.h"
 #include "shader/prog_statevars.h"
 #include "intel_batchbuffer.h"
 #include "intel_regions.h"
@@ -188,13 +189,6 @@ static void prepare_constant_buffer(struct brw_context *brw)
    GLfloat *buf;
    GLuint i;
 
-   /* Update our own dependency flags.  This works because this
-    * function will also be called whenever fp or vp changes.
-    */
-   brw->curbe.tracked_state.dirty.mesa = (_NEW_TRANSFORM|_NEW_PROJECTION);
-   brw->curbe.tracked_state.dirty.mesa |= vp->program.Base.Parameters->StateFlags;
-   brw->curbe.tracked_state.dirty.mesa |= fp->program.Base.Parameters->StateFlags;
-
    if (sz == 0) {
       if (brw->curbe.last_buf) {
 	 free(brw->curbe.last_buf);
@@ -335,72 +329,11 @@ static void prepare_constant_buffer(struct brw_context *brw)
     */
 }
 
-
-/**
- * Copy Mesa program parameters into given constant buffer.
- */
-static void
-update_constant_buffer(struct brw_context *brw,
-                       const struct gl_program_parameter_list *params,
-                       dri_bo *const_buffer)
-{
-   const int size = params->NumParameters * 4 * sizeof(GLfloat);
-
-   /* copy Mesa program constants into the buffer */
-   if (const_buffer && size > 0) {
-      GLubyte *map;
-
-      assert(const_buffer);
-      assert(const_buffer->size >= size);
-
-      dri_bo_map(const_buffer, GL_TRUE);
-      map = const_buffer->virtual;
-      memcpy(map, params->ParameterValues, size);
-      dri_bo_unmap(const_buffer);
-
-      if (0) {
-         int i;
-         for (i = 0; i < params->NumParameters; i++) {
-            float *p = params->ParameterValues[i];
-            printf("%d: %f %f %f %f\n", i, p[0], p[1], p[2], p[3]);
-         }
-      }
-   }
-}
-
-
-/** Copy current vertex program's parameters into the constant buffer */
-static void
-update_vertex_constant_buffer(struct brw_context *brw)
-{
-   struct brw_vertex_program *vp =
-      (struct brw_vertex_program *) brw->vertex_program;
-   if (0) {
-      printf("update VS constants in buffer %p\n", vp->const_buffer);
-      printf("program %u\n", vp->program.Base.Id);
-   }
-   update_constant_buffer(brw, vp->program.Base.Parameters, vp->const_buffer);
-}
-
-
-/** Copy current fragment program's parameters into the constant buffer */
-static void
-update_fragment_constant_buffer(struct brw_context *brw)
-{
-   struct brw_fragment_program *fp =
-      (struct brw_fragment_program *) brw->fragment_program;
-   update_constant_buffer(brw, fp->program.Base.Parameters, fp->const_buffer);
-}
-
-
 static void emit_constant_buffer(struct brw_context *brw)
 {
    struct intel_context *intel = &brw->intel;
    GLuint sz = brw->curbe.total_size;
 
-   update_vertex_constant_buffer(brw);
-   update_fragment_constant_buffer(brw);
-
    BEGIN_BATCH(2, IGNORE_CLIPRECTS);
    if (sz == 0) {
       OUT_BATCH((CMD_CONST_BUFFER << 16) | (2 - 2));
@@ -422,7 +355,7 @@ static void emit_constant_buffer(struct brw_context *brw)
  */
 const struct brw_tracked_state brw_constant_buffer = {
    .dirty = {
-      .mesa = (_NEW_TRANSFORM|_NEW_PROJECTION),      /* plus fp and vp flags */
+      .mesa = _NEW_PROGRAM_CONSTANTS,
       .brw  = (BRW_NEW_FRAGMENT_PROGRAM |
 	       BRW_NEW_VERTEX_PROGRAM |
 	       BRW_NEW_URB_FENCE | /* Implicit - hardware requires this, not used above */
diff --git a/src/mesa/drivers/dri/i965/brw_misc_state.c b/src/mesa/drivers/dri/i965/brw_misc_state.c
index 9bc5c35139..4784254bc7 100644
--- a/src/mesa/drivers/dri/i965/brw_misc_state.c
+++ b/src/mesa/drivers/dri/i965/brw_misc_state.c
@@ -118,7 +118,10 @@ static void upload_binding_table_pointers(struct brw_context *brw)
 
    BEGIN_BATCH(6, IGNORE_CLIPRECTS);
    OUT_BATCH(CMD_BINDING_TABLE_PTRS << 16 | (6 - 2));
-   OUT_RELOC(brw->vs.bind_bo, I915_GEM_DOMAIN_SAMPLER, 0, 0); /* vs */
+   if (brw->vs.bind_bo != NULL)
+      OUT_RELOC(brw->vs.bind_bo, I915_GEM_DOMAIN_SAMPLER, 0, 0); /* vs */
+   else
+      OUT_BATCH(0);
    OUT_BATCH(0); /* gs */
    OUT_BATCH(0); /* clip */
    OUT_BATCH(0); /* sf */
diff --git a/src/mesa/drivers/dri/i965/brw_sf_state.c b/src/mesa/drivers/dri/i965/brw_sf_state.c
index fc4eddda0a..68fa9820b6 100644
--- a/src/mesa/drivers/dri/i965/brw_sf_state.c
+++ b/src/mesa/drivers/dri/i965/brw_sf_state.c
@@ -147,7 +147,7 @@ sf_unit_populate_key(struct brw_context *brw, struct brw_sf_unit_key *key)
    key->line_smooth = ctx->Line.SmoothFlag;
 
    key->point_sprite = ctx->Point.PointSprite;
-   key->point_size = ctx->Point.Size;
+   key->point_size = CLAMP(ctx->Point.Size, ctx->Point.MinSize, ctx->Point.MaxSize);
    key->point_attenuated = ctx->Point._Attenuated;
 
    key->render_to_fbo = brw->intel.ctx.DrawBuffer->Name != 0;
diff --git a/src/mesa/drivers/dri/i965/brw_state.h b/src/mesa/drivers/dri/i965/brw_state.h
index 81b0a45998..bf9f6cae55 100644
--- a/src/mesa/drivers/dri/i965/brw_state.h
+++ b/src/mesa/drivers/dri/i965/brw_state.h
@@ -72,11 +72,13 @@ const struct brw_tracked_state brw_sf_vp;
 const struct brw_tracked_state brw_state_base_address;
 const struct brw_tracked_state brw_urb_fence;
 const struct brw_tracked_state brw_vertex_state;
+const struct brw_tracked_state brw_vs_surfaces;
 const struct brw_tracked_state brw_vs_prog;
 const struct brw_tracked_state brw_vs_unit;
 const struct brw_tracked_state brw_wm_input_sizes;
 const struct brw_tracked_state brw_wm_prog;
 const struct brw_tracked_state brw_wm_samplers;
+const struct brw_tracked_state brw_wm_constant_surface;
 const struct brw_tracked_state brw_wm_surfaces;
 const struct brw_tracked_state brw_wm_unit;
 
@@ -91,6 +93,20 @@ const struct brw_tracked_state brw_drawing_rect;
 const struct brw_tracked_state brw_indices;
 const struct brw_tracked_state brw_vertices;
 
+/**
+ * Use same key for WM and VS surfaces.
+ */
+struct brw_surface_key {
+   GLenum target, depthmode;
+   dri_bo *bo;
+   GLint format, internal_format;
+   GLint first_level, last_level;
+   GLint width, height, depth;
+   GLint pitch, cpp;
+   uint32_t tiling;
+   GLuint offset;
+};
+
 /***********************************************************************
  * brw_state.c
  */
@@ -135,8 +151,8 @@ dri_bo *brw_search_cache( struct brw_cache *cache,
 			  void *aux_return);
 void brw_state_cache_check_size( struct brw_context *brw );
 
-void brw_init_cache( struct brw_context *brw );
-void brw_destroy_cache( struct brw_context *brw );
+void brw_init_caches( struct brw_context *brw );
+void brw_destroy_caches( struct brw_context *brw );
 
 /***********************************************************************
  * brw_state_batch.c
@@ -150,4 +166,9 @@ GLboolean brw_cached_batch_struct( struct brw_context *brw,
 void brw_destroy_batch_cache( struct brw_context *brw );
 void brw_clear_batch_cache_flush( struct brw_context *brw );
 
+/* brw_wm_surface_state.c */
+dri_bo *
+brw_create_constant_surface( struct brw_context *brw,
+                             struct brw_surface_key *key );
+
 #endif
diff --git a/src/mesa/drivers/dri/i965/brw_state_cache.c b/src/mesa/drivers/dri/i965/brw_state_cache.c
index d5b5166406..e40d7a0416 100644
--- a/src/mesa/drivers/dri/i965/brw_state_cache.c
+++ b/src/mesa/drivers/dri/i965/brw_state_cache.c
@@ -56,9 +56,9 @@
  * incorrect program is run for the other instance.
  */
 
+#include "main/imports.h"
 #include "brw_state.h"
 #include "intel_batchbuffer.h"
-#include "main/imports.h"
 
 /* XXX: Fixme - have to include these to get the sizes of the prog_key
  * structs:
@@ -69,8 +69,10 @@
 #include "brw_sf.h"
 #include "brw_gs.h"
 
-static GLuint hash_key( const void *key, GLuint key_size,
-			dri_bo **reloc_bufs, GLuint nr_reloc_bufs)
+
+static GLuint
+hash_key(const void *key, GLuint key_size,
+         dri_bo **reloc_bufs, GLuint nr_reloc_bufs)
 {
    GLuint *ikey = (GLuint *)key;
    GLuint hash = 0, i;
@@ -95,6 +97,7 @@ static GLuint hash_key( const void *key, GLuint key_size,
    return hash;
 }
 
+
 /**
  * Marks a new buffer as being chosen for the given cache id.
  */
@@ -111,6 +114,7 @@ update_cache_last(struct brw_cache *cache, enum brw_cache_id cache_id,
    cache->brw->state.dirty.cache |= 1 << cache_id;
 }
 
+
 static struct brw_cache_item *
 search_cache(struct brw_cache *cache, enum brw_cache_id cache_id,
 	     GLuint hash, const void *key, GLuint key_size,
@@ -143,7 +147,8 @@ search_cache(struct brw_cache *cache, enum brw_cache_id cache_id,
 }
 
 
-static void rehash( struct brw_cache *cache )
+static void
+rehash(struct brw_cache *cache)
 {
    struct brw_cache_item **items;
    struct brw_cache_item *c, *next;
@@ -164,15 +169,17 @@ static void rehash( struct brw_cache *cache )
    cache->size = size;
 }
 
+
 /**
  * Returns the buffer object matching cache_id and key, or NULL.
  */
-dri_bo *brw_search_cache( struct brw_cache *cache,
-			  enum brw_cache_id cache_id,
-			  const void *key,
-			  GLuint key_size,
-			  dri_bo **reloc_bufs, GLuint nr_reloc_bufs,
-			  void *aux_return )
+dri_bo *
+brw_search_cache(struct brw_cache *cache,
+                 enum brw_cache_id cache_id,
+                 const void *key,
+                 GLuint key_size,
+                 dri_bo **reloc_bufs, GLuint nr_reloc_bufs,
+                 void *aux_return)
 {
    struct brw_cache_item *item;
    GLuint hash = hash_key(key, key_size, reloc_bufs, nr_reloc_bufs);
@@ -192,6 +199,7 @@ dri_bo *brw_search_cache( struct brw_cache *cache,
    return item->bo;
 }
 
+
 dri_bo *
 brw_upload_cache( struct brw_cache *cache,
 		  enum brw_cache_id cache_id,
@@ -265,7 +273,9 @@ brw_upload_cache( struct brw_cache *cache,
    return bo;
 }
 
-/* This doesn't really work with aux data.  Use search/upload instead
+
+/**
+ * This doesn't really work with aux data.  Use search/upload instead
  */
 dri_bo *
 brw_cache_data_sz(struct brw_cache *cache,
@@ -296,6 +306,7 @@ brw_cache_data_sz(struct brw_cache *cache,
    return bo;
 }
 
+
 /**
  * Wrapper around brw_cache_data_sz using the cache_id's canonical key size.
  *
@@ -319,21 +330,22 @@ enum pool_type {
    DW_GENERAL_STATE
 };
 
+
 static void
-brw_init_cache_id( struct brw_context *brw,
-		const char *name,
-		enum brw_cache_id id,
-		GLuint key_size,
-		GLuint aux_size)
+brw_init_cache_id(struct brw_cache *cache,
+                  const char *name,
+                  enum brw_cache_id id,
+                  GLuint key_size,
+                  GLuint aux_size)
 {
-   struct brw_cache *cache = &brw->cache;
-
    cache->name[id] = strdup(name);
    cache->key_size[id] = key_size;
    cache->aux_size[id] = aux_size;
 }
 
-void brw_init_cache( struct brw_context *brw )
+
+static void
+brw_init_non_surface_cache(struct brw_context *brw)
 {
    struct brw_cache *cache = &brw->cache;
 
@@ -342,114 +354,136 @@ void brw_init_cache( struct brw_context *brw )
    cache->size = 7;
    cache->n_items = 0;
    cache->items = (struct brw_cache_item **)
-      _mesa_calloc(cache->size * 
-		   sizeof(struct brw_cache_item));
+      _mesa_calloc(cache->size * sizeof(struct brw_cache_item));
 
-   brw_init_cache_id(brw,
+   brw_init_cache_id(cache,
 		     "CC_VP",
 		     BRW_CC_VP,
 		     sizeof(struct brw_cc_viewport),
 		     0);
 
-   brw_init_cache_id(brw,
+   brw_init_cache_id(cache,
 		     "CC_UNIT",
 		     BRW_CC_UNIT,
 		     sizeof(struct brw_cc_unit_state),
 		     0);
 
-   brw_init_cache_id(brw,
+   brw_init_cache_id(cache,
 		     "WM_PROG",
 		     BRW_WM_PROG,
 		     sizeof(struct brw_wm_prog_key),
 		     sizeof(struct brw_wm_prog_data));
 
-   brw_init_cache_id(brw,
+   brw_init_cache_id(cache,
 		     "SAMPLER_DEFAULT_COLOR",
 		     BRW_SAMPLER_DEFAULT_COLOR,
 		     sizeof(struct brw_sampler_default_color),
 		     0);
 
-   brw_init_cache_id(brw,
+   brw_init_cache_id(cache,
 		     "SAMPLER",
 		     BRW_SAMPLER,
 		     0,		/* variable key/data size */
 		     0);
 
-   brw_init_cache_id(brw,
+   brw_init_cache_id(cache,
 		     "WM_UNIT",
 		     BRW_WM_UNIT,
 		     sizeof(struct brw_wm_unit_state),
 		     0);
 
-   brw_init_cache_id(brw,
+   brw_init_cache_id(cache,
 		     "SF_PROG",
 		     BRW_SF_PROG,
 		     sizeof(struct brw_sf_prog_key),
 		     sizeof(struct brw_sf_prog_data));
 
-   brw_init_cache_id(brw,
+   brw_init_cache_id(cache,
 		     "SF_VP",
 		     BRW_SF_VP,
 		     sizeof(struct brw_sf_viewport),
 		     0);
 
-   brw_init_cache_id(brw,
+   brw_init_cache_id(cache,
 		     "SF_UNIT",
 		     BRW_SF_UNIT,
 		     sizeof(struct brw_sf_unit_state),
 		     0);
 
-   brw_init_cache_id(brw,
+   brw_init_cache_id(cache,
 		     "VS_UNIT",
 		     BRW_VS_UNIT,
 		     sizeof(struct brw_vs_unit_state),
 		     0);
 
-   brw_init_cache_id(brw,
+   brw_init_cache_id(cache,
 		     "VS_PROG",
 		     BRW_VS_PROG,
 		     sizeof(struct brw_vs_prog_key),
 		     sizeof(struct brw_vs_prog_data));
 
-   brw_init_cache_id(brw,
+   brw_init_cache_id(cache,
 		     "CLIP_UNIT",
 		     BRW_CLIP_UNIT,
 		     sizeof(struct brw_clip_unit_state),
 		     0);
 
-   brw_init_cache_id(brw,
+   brw_init_cache_id(cache,
 		     "CLIP_PROG",
 		     BRW_CLIP_PROG,
 		     sizeof(struct brw_clip_prog_key),
 		     sizeof(struct brw_clip_prog_data));
 
-   brw_init_cache_id(brw,
+   brw_init_cache_id(cache,
 		     "GS_UNIT",
 		     BRW_GS_UNIT,
 		     sizeof(struct brw_gs_unit_state),
 		     0);
 
-   brw_init_cache_id(brw,
+   brw_init_cache_id(cache,
 		     "GS_PROG",
 		     BRW_GS_PROG,
 		     sizeof(struct brw_gs_prog_key),
 		     sizeof(struct brw_gs_prog_data));
+}
+
+
+static void
+brw_init_surface_cache(struct brw_context *brw)
+{
+   struct brw_cache *cache = &brw->surface_cache;
+
+   cache->brw = brw;
 
-   brw_init_cache_id(brw,
+   cache->size = 7;
+   cache->n_items = 0;
+   cache->items = (struct brw_cache_item **)
+      _mesa_calloc(cache->size * sizeof(struct brw_cache_item));
+
+   brw_init_cache_id(cache,
 		     "SS_SURFACE",
 		     BRW_SS_SURFACE,
 		     sizeof(struct brw_surface_state),
 		     0);
 
-   brw_init_cache_id(brw,
+   brw_init_cache_id(cache,
 		     "SS_SURF_BIND",
 		     BRW_SS_SURF_BIND,
 		     0,
 		     0);
 }
 
+
+void
+brw_init_caches(struct brw_context *brw)
+{
+   brw_init_non_surface_cache(brw);
+   brw_init_surface_cache(brw);
+}
+
+
 static void
-brw_clear_cache( struct brw_context *brw )
+brw_clear_cache(struct brw_context *brw, struct brw_cache *cache)
 {
    struct brw_cache_item *c, *next;
    GLuint i;
@@ -457,8 +491,8 @@ brw_clear_cache( struct brw_context *brw )
    if (INTEL_DEBUG & DEBUG_STATE)
       _mesa_printf("%s\n", __FUNCTION__);
 
-   for (i = 0; i < brw->cache.size; i++) {
-      for (c = brw->cache.items[i]; c; c = next) {
+   for (i = 0; i < cache->size; i++) {
+      for (c = cache->items[i]; c; c = next) {
 	 int j;
 
 	 next = c->next;
@@ -468,10 +502,10 @@ brw_clear_cache( struct brw_context *brw )
 	 free((void *)c->key);
 	 free(c);
       }
-      brw->cache.items[i] = NULL;
+      cache->items[i] = NULL;
    }
 
-   brw->cache.n_items = 0;
+   cache->n_items = 0;
 
    if (brw->curbe.last_buf) {
       _mesa_free(brw->curbe.last_buf);
@@ -483,25 +517,46 @@ brw_clear_cache( struct brw_context *brw )
    brw->state.dirty.cache |= ~0;
 }
 
-void brw_state_cache_check_size( struct brw_context *brw )
+
+void
+brw_state_cache_check_size(struct brw_context *brw)
 {
+   if (INTEL_DEBUG & DEBUG_STATE)
+      _mesa_printf("%s (n_items=%d)\n", __FUNCTION__, brw->cache.n_items);
+
    /* un-tuned guess.  We've got around 20 state objects for a total of around
     * 32k, so 1000 of them is around 1.5MB.
     */
    if (brw->cache.n_items > 1000)
-      brw_clear_cache(brw);
+      brw_clear_cache(brw, &brw->cache);
+
+   if (brw->surface_cache.n_items > 1000)
+      brw_clear_cache(brw, &brw->surface_cache);
 }
 
-void brw_destroy_cache( struct brw_context *brw )
+
+static void
+brw_destroy_cache(struct brw_context *brw, struct brw_cache *cache)
 {
    GLuint i;
 
-   brw_clear_cache(brw);
+   if (INTEL_DEBUG & DEBUG_STATE)
+      _mesa_printf("%s\n", __FUNCTION__);
+
+   brw_clear_cache(brw, cache);
    for (i = 0; i < BRW_MAX_CACHE; i++) {
-      dri_bo_unreference(brw->cache.last_bo[i]);
-      free(brw->cache.name[i]);
+      dri_bo_unreference(cache->last_bo[i]);
+      free(cache->name[i]);
    }
-   free(brw->cache.items);
-   brw->cache.items = NULL;
-   brw->cache.size = 0;
+   free(cache->items);
+   cache->items = NULL;
+   cache->size = 0;
+}
+
+
+void
+brw_destroy_caches(struct brw_context *brw)
+{
+   brw_destroy_cache(brw, &brw->cache);
+   brw_destroy_cache(brw, &brw->surface_cache);
 }
diff --git a/src/mesa/drivers/dri/i965/brw_state_upload.c b/src/mesa/drivers/dri/i965/brw_state_upload.c
index 5de1450e61..c6dfea4743 100644
--- a/src/mesa/drivers/dri/i965/brw_state_upload.c
+++ b/src/mesa/drivers/dri/i965/brw_state_upload.c
@@ -59,11 +59,12 @@ const struct brw_tracked_state *atoms[] =
    &brw_curbe_offsets,
    &brw_recalculate_urb_fence,
 
-
    &brw_cc_vp,
    &brw_cc_unit,
 
-   &brw_wm_surfaces,		/* must do before samplers */
+   &brw_vs_surfaces,		/* must do before unit */
+   &brw_wm_constant_surface,	/* must do before wm surfaces/bind bo */
+   &brw_wm_surfaces,		/* must do before samplers and unit */
    &brw_wm_samplers,
 
    &brw_wm_unit,
@@ -88,54 +89,26 @@ const struct brw_tracked_state *atoms[] =
 
    &brw_line_stipple,
    &brw_aa_line_parameters,
-   /* Ordering of the commands below is documented as fixed.  
-    */
-#if 0
-   &brw_pipelined_state_pointers,
-   &brw_urb_fence,
-   &brw_constant_buffer_state,
-#else
+
    &brw_psp_urb_cbs,
-#endif
 
    &brw_drawing_rect,
    &brw_indices,
    &brw_vertices,
 
-   NULL,			/* brw_constant_buffer */
+   &brw_constant_buffer
 };
 
 
 void brw_init_state( struct brw_context *brw )
 {
-   GLuint i;
-
-   brw_init_cache(brw);
-
-   brw->state.atoms = _mesa_malloc(sizeof(atoms));
-   brw->state.nr_atoms = sizeof(atoms)/sizeof(*atoms);
-   _mesa_memcpy(brw->state.atoms, atoms, sizeof(atoms));
-
-   /* Patch in a pointer to the dynamic state atom:
-    */
-   for (i = 0; i < brw->state.nr_atoms; i++)
-      if (brw->state.atoms[i] == NULL)
-	 brw->state.atoms[i] = &brw->curbe.tracked_state;
-
-   _mesa_memcpy(&brw->curbe.tracked_state, 
-		&brw_constant_buffer,
-		sizeof(brw_constant_buffer));
+   brw_init_caches(brw);
 }
 
 
 void brw_destroy_state( struct brw_context *brw )
 {
-   if (brw->state.atoms) {
-      _mesa_free(brw->state.atoms);
-      brw->state.atoms = NULL;
-   }
-
-   brw_destroy_cache(brw);
+   brw_destroy_caches(brw);
    brw_destroy_batch_cache(brw);
 }
 
@@ -218,6 +191,7 @@ static struct dirty_bit_map mesa_bits[] = {
    DEFINE_BIT(_NEW_MULTISAMPLE),
    DEFINE_BIT(_NEW_TRACK_MATRIX),
    DEFINE_BIT(_NEW_PROGRAM),
+   DEFINE_BIT(_NEW_PROGRAM_CONSTANTS),
    {0, 0, 0}
 };
 
@@ -336,7 +310,7 @@ void brw_validate_state( struct brw_context *brw )
 
    /* do prepare stage for all atoms */
    for (i = 0; i < Elements(atoms); i++) {
-      const struct brw_tracked_state *atom = brw->state.atoms[i];
+      const struct brw_tracked_state *atom = atoms[i];
 
       if (brw->intel.Fallback)
          break;
@@ -367,8 +341,8 @@ void brw_upload_state(struct brw_context *brw)
       _mesa_memset(&examined, 0, sizeof(examined));
       prev = *state;
 
-      for (i = 0; i < brw->state.nr_atoms; i++) {	 
-	 const struct brw_tracked_state *atom = brw->state.atoms[i];
+      for (i = 0; i < Elements(atoms); i++) {	 
+	 const struct brw_tracked_state *atom = atoms[i];
 	 struct brw_state_flags generated;
 
 	 assert(atom->dirty.mesa ||
@@ -397,7 +371,7 @@ void brw_upload_state(struct brw_context *brw)
    }
    else {
       for (i = 0; i < Elements(atoms); i++) {	 
-	 const struct brw_tracked_state *atom = brw->state.atoms[i];
+	 const struct brw_tracked_state *atom = atoms[i];
 
 	 if (brw->intel.Fallback)
 	    break;
diff --git a/src/mesa/drivers/dri/i965/brw_vs.h b/src/mesa/drivers/dri/i965/brw_vs.h
index d20cf78b8a..1e4f66091e 100644
--- a/src/mesa/drivers/dri/i965/brw_vs.h
+++ b/src/mesa/drivers/dri/i965/brw_vs.h
@@ -75,8 +75,6 @@ struct brw_vs_compile {
 
    struct brw_reg userplane[6];
 
-   /** using a real constant buffer? */
-   GLboolean use_const_buffer;
    /** we may need up to 3 constants per instruction (if use_const_buffer) */
    struct {
       GLint index;
diff --git a/src/mesa/drivers/dri/i965/brw_vs_emit.c b/src/mesa/drivers/dri/i965/brw_vs_emit.c
index 524f1211ce..d7f75e3685 100644
--- a/src/mesa/drivers/dri/i965/brw_vs_emit.c
+++ b/src/mesa/drivers/dri/i965/brw_vs_emit.c
@@ -69,13 +69,18 @@ static void brw_vs_alloc_regs( struct brw_vs_compile *c )
 {
    GLuint i, reg = 0, mrf;
 
-#if 0
-   if (c->vp->program.Base.Parameters->NumParameters >= 6)
-      c->use_const_buffer = 1;
+   /* Determine whether to use a real constant buffer or use a block
+    * of GRF registers for constants.  The later is faster but only
+    * works if everything fits in the GRF.
+    * XXX this heuristic/check may need some fine tuning...
+    */
+   if (c->vp->program.Base.Parameters->NumParameters +
+       c->vp->program.Base.NumTemporaries + 20 > BRW_MAX_GRF)
+      c->vp->use_const_buffer = GL_TRUE;
    else
-#endif
-      c->use_const_buffer = GL_FALSE;
-   /*printf("use_const_buffer = %d\n", c->use_const_buffer);*/
+      c->vp->use_const_buffer = GL_FALSE;
+
+   /*printf("use_const_buffer = %d\n", c->vp->use_const_buffer);*/
 
    /* r0 -- reserved as usual
     */
@@ -96,7 +101,7 @@ static void brw_vs_alloc_regs( struct brw_vs_compile *c )
 
    /* Vertex program parameters from curbe:
     */
-   if (c->use_const_buffer) {
+   if (c->vp->use_const_buffer) {
       /* get constants from a real constant buffer */
       c->prog_data.curb_read_length = 0;
       c->prog_data.nr_params = 4; /* XXX 0 causes a bug elsewhere... */
@@ -172,7 +177,7 @@ static void brw_vs_alloc_regs( struct brw_vs_compile *c )
       reg++;
    }
 
-   if (c->use_const_buffer) {
+   if (c->vp->use_const_buffer) {
       for (i = 0; i < 3; i++) {
          c->current_const[i].index = -1;
          c->current_const[i].reg = brw_vec8_grf(reg, 0);
@@ -709,10 +714,11 @@ get_constant(struct brw_vs_compile *c,
    struct brw_compile *p = &c->func;
    struct brw_reg const_reg;
    struct brw_reg const2_reg;
+   const GLboolean relAddr = src->RelAddr;
 
    assert(argIndex < 3);
 
-   if (c->current_const[argIndex].index != src->Index || src->RelAddr) {
+   if (c->current_const[argIndex].index != src->Index || relAddr) {
       struct brw_reg addrReg = c->regs[PROGRAM_ADDRESS][0];
 
       c->current_const[argIndex].index = src->Index;
@@ -725,13 +731,13 @@ get_constant(struct brw_vs_compile *c,
       brw_dp_READ_4_vs(p,
                        c->current_const[argIndex].reg,/* writeback dest */
                        0,                             /* oword */
-                       src->RelAddr,                  /* relative indexing? */
+                       relAddr,                       /* relative indexing? */
                        addrReg,                       /* address register */
                        16 * src->Index,               /* byte offset */
                        SURF_INDEX_VERT_CONST_BUFFER   /* binding table index */
                        );
 
-      if (src->RelAddr) {
+      if (relAddr) {
          /* second read */
          const2_reg = get_tmp(c);
 
@@ -742,7 +748,7 @@ get_constant(struct brw_vs_compile *c,
          brw_dp_READ_4_vs(p,
                           const2_reg,              /* writeback dest */
                           1,                       /* oword */
-                          src->RelAddr,            /* relative indexing? */
+                          relAddr,                 /* relative indexing? */
                           addrReg,                 /* address register */
                           16 * src->Index,         /* byte offset */
                           SURF_INDEX_VERT_CONST_BUFFER
@@ -752,7 +758,7 @@ get_constant(struct brw_vs_compile *c,
 
    const_reg = c->current_const[argIndex].reg;
 
-   if (src->RelAddr) {
+   if (relAddr) {
       /* merge the two Owords into the constant register */
       /* const_reg[7..4] = const2_reg[7..4] */
       brw_MOV(p,
@@ -869,7 +875,7 @@ get_src_reg( struct brw_vs_compile *c,
    case PROGRAM_STATE_VAR:
    case PROGRAM_CONSTANT:
    case PROGRAM_UNIFORM:
-      if (c->use_const_buffer) {
+      if (c->vp->use_const_buffer) {
          return get_constant(c, inst, argIndex);
       }
       else if (relAddr) {
@@ -1219,7 +1225,7 @@ void brw_vs_emit(struct brw_vs_compile *c )
 
    for (insn = 0; insn < nr_insns; insn++) {
 
-      struct prog_instruction *inst = &c->vp->program.Base.Instructions[insn];
+      const struct prog_instruction *inst = &c->vp->program.Base.Instructions[insn];
       struct brw_reg args[3], dst;
       GLuint i;
       
@@ -1232,7 +1238,7 @@ void brw_vs_emit(struct brw_vs_compile *c )
        */
       if (inst->Opcode != OPCODE_SWZ)
 	  for (i = 0; i < 3; i++) {
-	      struct prog_src_register *src = &inst->SrcReg[i];
+	      const struct prog_src_register *src = &inst->SrcReg[i];
 	      index = src->Index;
 	      file = src->File;	
 	      if (file == PROGRAM_OUTPUT && c->output_regs[index].used_in_src)
diff --git a/src/mesa/drivers/dri/i965/brw_vs_surface_state.c b/src/mesa/drivers/dri/i965/brw_vs_surface_state.c
new file mode 100644
index 0000000000..89f47522a1
--- /dev/null
+++ b/src/mesa/drivers/dri/i965/brw_vs_surface_state.c
@@ -0,0 +1,226 @@
+/*
+ Copyright (C) Intel Corp.  2006.  All Rights Reserved.
+ Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to
+ develop this 3D driver.
+
+ Permission is hereby granted, free of charge, to any person obtaining
+ a copy of this software and associated documentation files (the
+ "Software"), to deal in the Software without restriction, including
+ without limitation the rights to use, copy, modify, merge, publish,
+ distribute, sublicense, and/or sell copies of the Software, and to
+ permit persons to whom the Software is furnished to do so, subject to
+ the following conditions:
+
+ The above copyright notice and this permission notice (including the
+ next paragraph) shall be included in all copies or substantial
+ portions of the Software.
+
+ THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
+ IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
+ LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
+ OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+
+ **********************************************************************/
+ /*
+  * Authors:
+  *   Keith Whitwell <keith@tungstengraphics.com>
+  */
+
+#include "main/mtypes.h"
+#include "main/texformat.h"
+#include "main/texstore.h"
+#include "shader/prog_parameter.h"
+
+#include "brw_context.h"
+#include "brw_state.h"
+#include "brw_defines.h"
+
+/* Creates a new VS constant buffer reflecting the current VS program's
+ * constants, if needed by the VS program.
+ *
+ * Otherwise, constants go through the CURBEs using the brw_constant_buffer
+ * state atom.
+ */
+static drm_intel_bo *
+brw_vs_update_constant_buffer(struct brw_context *brw)
+{
+   struct intel_context *intel = &brw->intel;
+   struct brw_vertex_program *vp =
+      (struct brw_vertex_program *) brw->vertex_program;
+   const struct gl_program_parameter_list *params = vp->program.Base.Parameters;
+   const int size = params->NumParameters * 4 * sizeof(GLfloat);
+   drm_intel_bo *const_buffer;
+
+   /* BRW_NEW_VERTEX_PROGRAM */
+   if (!vp->use_const_buffer)
+      return NULL;
+
+   const_buffer = drm_intel_bo_alloc(intel->bufmgr, "vp_const_buffer",
+				     size, 64);
+
+   /* _NEW_PROGRAM_CONSTANTS */
+   dri_bo_subdata(const_buffer, 0, size, params->ParameterValues);
+
+   return const_buffer;
+}
+
+/**
+ * Update the surface state for a VS constant buffer.
+ *
+ * Sets brw->vs.surf_bo[surf] and brw->vp->const_buffer.
+ */
+static void
+brw_update_vs_constant_surface( GLcontext *ctx,
+                                GLuint surf)
+{
+   struct brw_context *brw = brw_context(ctx);
+   struct brw_surface_key key;
+   struct brw_vertex_program *vp =
+      (struct brw_vertex_program *) brw->vertex_program;
+   const struct gl_program_parameter_list *params = vp->program.Base.Parameters;
+
+   assert(surf == 0);
+
+   /* If we're in this state update atom, we need to update VS constants, so
+    * free the old buffer and create a new one for the new contents.
+    */
+   dri_bo_unreference(vp->const_buffer);
+   vp->const_buffer = brw_vs_update_constant_buffer(brw);
+
+   /* If there's no constant buffer, then no surface BO is needed to point at
+    * it.
+    */
+   if (vp->const_buffer == 0) {
+      drm_intel_bo_unreference(brw->vs.surf_bo[surf]);
+      brw->vs.surf_bo[surf] = NULL;
+      return;
+   }
+
+   memset(&key, 0, sizeof(key));
+
+   key.format = MESA_FORMAT_RGBA_FLOAT32;
+   key.internal_format = GL_RGBA;
+   key.bo = vp->const_buffer;
+   key.depthmode = GL_NONE;
+   key.pitch = params->NumParameters;
+   key.width = params->NumParameters;
+   key.height = 1;
+   key.depth = 1;
+   key.cpp = 16;
+
+   /*
+   printf("%s:\n", __FUNCTION__);
+   printf("  width %d  height %d  depth %d  cpp %d  pitch %d\n",
+          key.width, key.height, key.depth, key.cpp, key.pitch);
+   */
+
+   drm_intel_bo_unreference(brw->vs.surf_bo[surf]);
+   brw->vs.surf_bo[surf] = brw_search_cache(&brw->surface_cache,
+                                            BRW_SS_SURFACE,
+                                            &key, sizeof(key),
+                                            &key.bo, key.bo ? 1 : 0,
+                                            NULL);
+   if (brw->vs.surf_bo[surf] == NULL) {
+      brw->vs.surf_bo[surf] = brw_create_constant_surface(brw, &key);
+   }
+}
+
+
+/**
+ * Constructs the binding table for the VS surface state.
+ */
+static dri_bo *
+brw_vs_get_binding_table(struct brw_context *brw)
+{
+   dri_bo *bind_bo;
+
+   bind_bo = brw_search_cache(&brw->surface_cache, BRW_SS_SURF_BIND,
+			      NULL, 0,
+			      brw->vs.surf_bo, BRW_VS_MAX_SURF,
+			      NULL);
+
+   if (bind_bo == NULL) {
+      GLuint data_size = BRW_VS_MAX_SURF * sizeof(GLuint);
+      uint32_t *data = malloc(data_size);
+      int i;
+
+      for (i = 0; i < BRW_VS_MAX_SURF; i++)
+         if (brw->vs.surf_bo[i])
+            data[i] = brw->vs.surf_bo[i]->offset;
+         else
+            data[i] = 0;
+
+      bind_bo = brw_upload_cache( &brw->surface_cache, BRW_SS_SURF_BIND,
+				  NULL, 0,
+				  brw->vs.surf_bo, BRW_VS_MAX_SURF,
+				  data, data_size,
+				  NULL, NULL);
+
+      /* Emit binding table relocations to surface state */
+      for (i = 0; i < BRW_VS_MAX_SURF; i++) {
+	 if (brw->vs.surf_bo[i] != NULL) {
+	    /* The presumed offsets were set in the data values for
+	     * brw_upload_cache.
+	     */
+	    drm_intel_bo_emit_reloc(bind_bo, i * 4,
+				    brw->vs.surf_bo[i], 0,
+				    I915_GEM_DOMAIN_INSTRUCTION, 0);
+	 }
+      }
+
+      free(data);
+   }
+
+   return bind_bo;
+}
+
+/**
+ * Vertex shader surfaces (constant buffer).
+ *
+ * This consumes the state updates for the constant buffer needing
+ * to be updated, and produces BRW_NEW_NR_VS_SURFACES for the VS unit and
+ * CACHE_NEW_SURF_BIND for the binding table upload.
+ */
+static void prepare_vs_surfaces(struct brw_context *brw )
+{
+   GLcontext *ctx = &brw->intel.ctx;
+   int i;
+   int nr_surfaces = 0;
+
+   brw_update_vs_constant_surface(ctx, SURF_INDEX_VERT_CONST_BUFFER);
+
+   for (i = 0; i < BRW_VS_MAX_SURF; i++) {
+      if (brw->vs.surf_bo[i] != NULL) {
+	 nr_surfaces = i + 1;
+      }
+   }
+
+   if (brw->vs.nr_surfaces != nr_surfaces) {
+      brw->state.dirty.brw |= BRW_NEW_NR_VS_SURFACES;
+      brw->vs.nr_surfaces = nr_surfaces;
+   }
+
+   /* Note that we don't end up updating the bind_bo if we don't have a
+    * surface to be pointing at.  This should be relatively harmless, as it
+    * just slightly increases our working set size.
+    */
+   if (brw->vs.nr_surfaces != 0) {
+      dri_bo_unreference(brw->vs.bind_bo);
+      brw->vs.bind_bo = brw_vs_get_binding_table(brw);
+   }
+}
+
+const struct brw_tracked_state brw_vs_surfaces = {
+   .dirty = {
+      .mesa = (_NEW_PROGRAM_CONSTANTS),
+      .brw = (BRW_NEW_VERTEX_PROGRAM),
+      .cache = 0
+   },
+   .prepare = prepare_vs_surfaces,
+};
+
+
+
diff --git a/src/mesa/drivers/dri/i965/brw_wm.c b/src/mesa/drivers/dri/i965/brw_wm.c
index 90d74c2885..cd65f57bbc 100644
--- a/src/mesa/drivers/dri/i965/brw_wm.c
+++ b/src/mesa/drivers/dri/i965/brw_wm.c
@@ -146,6 +146,13 @@ static void do_wm_prog( struct brw_context *brw,
    if (c == NULL) {
       brw->wm.compile_data = calloc(1, sizeof(*brw->wm.compile_data));
       c = brw->wm.compile_data;
+      if (c == NULL) {
+         /* Ouch - big out of memory problem.  Can't continue
+          * without triggering a segfault, no way to signal,
+          * so just return.
+          */
+         return;
+      }
    } else {
       memset(c, 0, sizeof(*brw->wm.compile_data));
    }
diff --git a/src/mesa/drivers/dri/i965/brw_wm.h b/src/mesa/drivers/dri/i965/brw_wm.h
index d0ab3bdc65..59ead757b5 100644
--- a/src/mesa/drivers/dri/i965/brw_wm.h
+++ b/src/mesa/drivers/dri/i965/brw_wm.h
@@ -240,22 +240,25 @@ struct brw_wm_compile {
    GLuint max_wm_grf;
    GLuint last_scratch;
 
+   GLuint cur_inst;  /**< index of current instruction */
+
+   GLboolean out_of_regs;  /**< ran out of GRF registers? */
+
    /** Mapping from Mesa registers to hardware registers */
    struct {
       GLboolean inited;
       struct brw_reg reg;
    } wm_regs[PROGRAM_PAYLOAD+1][256][4];
 
+   GLboolean used_grf[BRW_WM_MAX_GRF];
+   GLuint first_free_grf;
    struct brw_reg stack;
    struct brw_reg emit_mask_reg;
-   GLuint reg_index;  /**< Index of next free GRF register */
    GLuint tmp_regs[BRW_WM_MAX_GRF];
    GLuint tmp_index;
    GLuint tmp_max;
    GLuint subroutines[BRW_WM_MAX_SUBROUTINE];
 
-   /** using a real constant buffer? */
-   GLboolean use_const_buffer;
    /** we may need up to 3 constants per instruction (if use_const_buffer) */
    struct {
       GLint index;
diff --git a/src/mesa/drivers/dri/i965/brw_wm_glsl.c b/src/mesa/drivers/dri/i965/brw_wm_glsl.c
index 22e17622c6..efe8b5126c 100644
--- a/src/mesa/drivers/dri/i965/brw_wm_glsl.c
+++ b/src/mesa/drivers/dri/i965/brw_wm_glsl.c
@@ -1,5 +1,7 @@
 #include "main/macros.h"
 #include "shader/prog_parameter.h"
+#include "shader/prog_print.h"
+#include "shader/prog_optimize.h"
 #include "brw_context.h"
 #include "brw_eu.h"
 #include "brw_wm.h"
@@ -42,6 +44,83 @@ GLboolean brw_wm_is_glsl(const struct gl_fragment_program *fp)
 }
 
 
+
+static void
+reclaim_temps(struct brw_wm_compile *c);
+
+
+/** Mark GRF register as used. */
+static void
+prealloc_grf(struct brw_wm_compile *c, int r)
+{
+   c->used_grf[r] = GL_TRUE;
+}
+
+
+/** Mark given GRF register as not in use. */
+static void
+release_grf(struct brw_wm_compile *c, int r)
+{
+   /*assert(c->used_grf[r]);*/
+   c->used_grf[r] = GL_FALSE;
+   c->first_free_grf = MIN2(c->first_free_grf, r);
+}
+
+
+/** Return index of a free GRF, mark it as used. */
+static int
+alloc_grf(struct brw_wm_compile *c)
+{
+   GLuint r;
+   for (r = c->first_free_grf; r < BRW_WM_MAX_GRF; r++) {
+      if (!c->used_grf[r]) {
+         c->used_grf[r] = GL_TRUE;
+         c->first_free_grf = r + 1;  /* a guess */
+         return r;
+      }
+   }
+
+   /* no free temps, try to reclaim some */
+   reclaim_temps(c);
+   c->first_free_grf = 0;
+
+   /* try alloc again */
+   for (r = c->first_free_grf; r < BRW_WM_MAX_GRF; r++) {
+      if (!c->used_grf[r]) {
+         c->used_grf[r] = GL_TRUE;
+         c->first_free_grf = r + 1;  /* a guess */
+         return r;
+      }
+   }
+
+   for (r = 0; r < BRW_WM_MAX_GRF; r++) {
+      assert(c->used_grf[r]);
+   }
+
+   /* really, no free GRF regs found */
+   if (!c->out_of_regs) {
+      /* print warning once per compilation */
+      _mesa_warning(NULL, "i965: ran out of registers for fragment program");
+      c->out_of_regs = GL_TRUE;
+   }
+
+   return -1;
+}
+
+
+/** Return number of GRF registers used */
+static int
+num_grf_used(const struct brw_wm_compile *c)
+{
+   int r;
+   for (r = BRW_WM_MAX_GRF - 1; r >= 0; r--)
+      if (c->used_grf[r])
+         return r + 1;
+   return 0;
+}
+
+
+
 /**
  * Record the mapping of a Mesa register to a hardware register.
  */
@@ -68,11 +147,23 @@ static int get_scalar_dst_index(const struct prog_instruction *inst)
 static struct brw_reg alloc_tmp(struct brw_wm_compile *c)
 {
     struct brw_reg reg;
-    if(c->tmp_index == c->tmp_max)
-	c->tmp_regs[ c->tmp_max++ ] = c->reg_index++;
-    
+
+    /* if we need to allocate another temp, grow the tmp_regs[] array */
+    if (c->tmp_index == c->tmp_max) {
+       int r = alloc_grf(c);
+       if (r < 0) {
+          /*printf("Out of temps in %s\n", __FUNCTION__);*/
+          r = 50; /* XXX random register! */
+       }
+       c->tmp_regs[ c->tmp_max++ ] = r;
+    }
+
+    /* form the GRF register */
     reg = brw_vec8_grf(c->tmp_regs[ c->tmp_index++ ], 0);
+    /*printf("alloc_temp %d\n", reg.nr);*/
+    assert(reg.nr < BRW_WM_MAX_GRF);
     return reg;
+
 }
 
 /**
@@ -130,35 +221,29 @@ get_reg(struct brw_wm_compile *c, int file, int index, int component,
 	    return brw_null_reg();
     }
 
+    assert(index < 256);
+    assert(component < 4);
+
     /* see if we've already allocated a HW register for this Mesa register */
     if (c->wm_regs[file][index][component].inited) {
-	/* yes, re-use */
-	reg = c->wm_regs[file][index][component].reg;
+       /* yes, re-use */
+       reg = c->wm_regs[file][index][component].reg;
     }
     else {
 	/* no, allocate new register */
-	reg = brw_vec8_grf(c->reg_index, 0);
-    }
+       int grf = alloc_grf(c);
+       /*printf("alloc grf %d for reg %d:%d.%d\n", grf, file, index, component);*/
+       if (grf < 0) {
+          /* totally out of temps */
+          grf = 51; /* XXX random register! */
+       }
 
-    /* if this is a new register allocation, record it in the table */
-    if (!c->wm_regs[file][index][component].inited) {
-	set_reg(c, file, index, component, reg);
-	c->reg_index++;
-    }
+       reg = brw_vec8_grf(grf, 0);
+       /*printf("Alloc new grf %d for %d.%d\n", reg.nr, index, component);*/
 
-    if (c->reg_index >= BRW_WM_MAX_GRF - 12) {
-	/* ran out of temporary registers! */
-#if 1
-        /* This is a big hack for now.
-         * Return bad register index, just don't hang the GPU.
-         */
-        _mesa_fprintf(stderr, "out of regs %d\n", c->reg_index);
-        c->reg_index = BRW_WM_MAX_GRF - 13;
-#else
-	return brw_null_reg();
-#endif
+       set_reg(c, file, index, component, reg);
     }
- 
+
     if (neg & (1 << component)) {
 	reg = negate(reg);
     }
@@ -168,6 +253,46 @@ get_reg(struct brw_wm_compile *c, int file, int index, int component,
 }
 
 
+
+/**
+ * This is called if we run out of GRF registers.  Examine the live intervals
+ * of temp regs in the program and free those which won't be used again.
+ */
+static void
+reclaim_temps(struct brw_wm_compile *c)
+{
+   GLint intBegin[MAX_PROGRAM_TEMPS];
+   GLint intEnd[MAX_PROGRAM_TEMPS];
+   int index;
+
+   /*printf("Reclaim temps:\n");*/
+
+   _mesa_find_temp_intervals(c->prog_instructions, c->nr_fp_insns,
+                             intBegin, intEnd);
+
+   for (index = 0; index < MAX_PROGRAM_TEMPS; index++) {
+      if (intEnd[index] != -1 && intEnd[index] < c->cur_inst) {
+         /* program temp[i] can be freed */
+         int component;
+         /*printf("  temp[%d] is dead\n", index);*/
+         for (component = 0; component < 4; component++) {
+            if (c->wm_regs[PROGRAM_TEMPORARY][index][component].inited) {
+               int r = c->wm_regs[PROGRAM_TEMPORARY][index][component].reg.nr;
+               release_grf(c, r);
+               /*
+               printf("  Reclaim temp %d, reg %d at inst %d\n",
+                      index, r, c->cur_inst);
+               */
+               c->wm_regs[PROGRAM_TEMPORARY][index][component].inited = GL_FALSE;
+            }
+         }
+      }
+   }
+}
+
+
+
+
 /**
  * Preallocate registers.  This sets up the Mesa to hardware register
  * mapping for certain registers, such as constants (uniforms/state vars)
@@ -179,6 +304,10 @@ static void prealloc_reg(struct brw_wm_compile *c)
     struct brw_reg reg;
     int nr_interp_regs = 0;
     GLuint inputs = FRAG_BIT_WPOS | c->fp_interp_emitted | c->fp_deriv_emitted;
+    GLuint reg_index = 0;
+
+    memset(c->used_grf, GL_FALSE, sizeof(c->used_grf));
+    c->first_free_grf = 0;
 
     for (i = 0; i < 4; i++) {
         if (i < c->key.nr_depth_regs) 
@@ -187,16 +316,22 @@ static void prealloc_reg(struct brw_wm_compile *c)
             reg = brw_vec8_grf(0, 0);
 	set_reg(c, PROGRAM_PAYLOAD, PAYLOAD_DEPTH, i, reg);
     }
-    c->reg_index += 2 * c->key.nr_depth_regs;
+    reg_index += 2 * c->key.nr_depth_regs;
 
     /* constants */
     {
-        const int nr_params = c->fp->program.Base.Parameters->NumParameters;
+        const GLuint nr_params = c->fp->program.Base.Parameters->NumParameters;
+        const GLuint nr_temps = c->fp->program.Base.NumTemporaries;
 
         /* use a real constant buffer, or just use a section of the GRF? */
-        c->use_const_buffer = GL_FALSE; /* (nr_params > 8);*/
+        /* XXX this heuristic may need adjustment... */
+        if ((nr_params + nr_temps) * 4 + reg_index > 80)
+           c->fp->use_const_buffer = GL_TRUE;
+        else
+           c->fp->use_const_buffer = GL_FALSE;
+        /*printf("WM use_const_buffer = %d\n", c->fp->use_const_buffer);*/
 
-        if (c->use_const_buffer) {
+        if (c->fp->use_const_buffer) {
            /* We'll use a real constant buffer and fetch constants from
             * it with a dataport read message.
             */
@@ -216,7 +351,7 @@ static void prealloc_reg(struct brw_wm_compile *c)
            for (i = 0; i < nr_params; i++) {
               /* loop over XYZW channels */
               for (j = 0; j < 4; j++, index++) {
-                 reg = brw_vec1_grf(c->reg_index + index / 8, index % 8);
+                 reg = brw_vec1_grf(reg_index + index / 8, index % 8);
                  /* Save pointer to parameter/constant value.
                   * Constants will be copied in prepare_constant_buffer()
                   */
@@ -226,7 +361,7 @@ static void prealloc_reg(struct brw_wm_compile *c)
            }
            /* number of constant regs used (each reg is float[8]) */
            c->nr_creg = 2 * ((4 * nr_params + 15) / 16);
-           c->reg_index += c->nr_creg;
+           reg_index += c->nr_creg;
         }
     }
 
@@ -234,34 +369,42 @@ static void prealloc_reg(struct brw_wm_compile *c)
     for (i = 0; i < FRAG_ATTRIB_MAX; i++) {
 	if (inputs & (1<<i)) {
 	    nr_interp_regs++;
-	    reg = brw_vec8_grf(c->reg_index, 0);
+	    reg = brw_vec8_grf(reg_index, 0);
 	    for (j = 0; j < 4; j++)
 		set_reg(c, PROGRAM_PAYLOAD, i, j, reg);
-	    c->reg_index += 2;
+	    reg_index += 2;
 	}
     }
 
     c->prog_data.first_curbe_grf = c->key.nr_depth_regs * 2;
     c->prog_data.urb_read_length = nr_interp_regs * 2;
     c->prog_data.curb_read_length = c->nr_creg;
-    c->emit_mask_reg = brw_uw1_reg(BRW_GENERAL_REGISTER_FILE, c->reg_index, 0);
-    c->reg_index++;
-    c->stack =  brw_uw16_reg(BRW_GENERAL_REGISTER_FILE, c->reg_index, 0);
-    c->reg_index += 2;
+    c->emit_mask_reg = brw_uw1_reg(BRW_GENERAL_REGISTER_FILE, reg_index, 0);
+    reg_index++;
+    c->stack =  brw_uw16_reg(BRW_GENERAL_REGISTER_FILE, reg_index, 0);
+    reg_index += 2;
+
+    /* mark GRF regs [0..reg_index-1] as in-use */
+    for (i = 0; i < reg_index; i++)
+       prealloc_grf(c, i);
+
+    /* Don't use GRF 126, 127.  Using them seems to lead to GPU lock-ups */
+    prealloc_grf(c, 126);
+    prealloc_grf(c, 127);
 
     /* An instruction may reference up to three constants.
      * They'll be found in these registers.
      * XXX alloc these on demand!
      */
-    if (c->use_const_buffer) {
+    if (c->fp->use_const_buffer) {
        for (i = 0; i < 3; i++) {
           c->current_const[i].index = -1;
-          c->current_const[i].reg = alloc_tmp(c);
+          c->current_const[i].reg = brw_vec8_grf(alloc_grf(c), 0);
        }
     }
 #if 0
-    printf("USE CONST BUFFER? %d\n", c->use_const_buffer);
-    printf("AFTER PRE_ALLOC, reg_index = %d\n", c->reg_index);
+    printf("USE CONST BUFFER? %d\n", c->fp->use_const_buffer);
+    printf("AFTER PRE_ALLOC, reg_index = %d\n", reg_index);
 #endif
 }
 
@@ -283,23 +426,21 @@ static void fetch_constants(struct brw_wm_compile *c,
       if (src->File == PROGRAM_STATE_VAR ||
           src->File == PROGRAM_CONSTANT ||
           src->File == PROGRAM_UNIFORM) {
-         if (c->current_const[i].index != src->Index) {
-            c->current_const[i].index = src->Index;
+	 c->current_const[i].index = src->Index;
 
 #if 0
-            printf("  fetch const[%d] for arg %d into reg %d\n",
-                   src->Index, i, c->current_const[i].reg.nr);
+	 printf("  fetch const[%d] for arg %d into reg %d\n",
+		src->Index, i, c->current_const[i].reg.nr);
 #endif
 
-            /* need to fetch the constant now */
-            brw_dp_READ_4(p,
-                          c->current_const[i].reg,  /* writeback dest */
-                          1,                        /* msg_reg */
-                          src->RelAddr,             /* relative indexing? */
-                          16 * src->Index,          /* byte offset */
-                          SURF_INDEX_FRAG_CONST_BUFFER/* binding table index */
-                          );
-         }
+	 /* need to fetch the constant now */
+	 brw_dp_READ_4(p,
+		       c->current_const[i].reg,  /* writeback dest */
+		       1,                        /* msg_reg */
+		       src->RelAddr,             /* relative indexing? */
+		       16 * src->Index,          /* byte offset */
+		       SURF_INDEX_FRAG_CONST_BUFFER/* binding table index */
+		       );
       }
    }
 }
@@ -368,7 +509,7 @@ static struct brw_reg get_src_reg(struct brw_wm_compile *c,
     const GLuint nr = 1;
     const GLuint component = GET_SWZ(src->Swizzle, channel);
 
-    if (c->use_const_buffer &&
+    if (c->fp->use_const_buffer &&
         (src->File == PROGRAM_STATE_VAR ||
          src->File == PROGRAM_CONSTANT ||
          src->File == PROGRAM_UNIFORM)) {
@@ -2595,7 +2736,8 @@ static void brw_wm_emit_glsl(struct brw_context *brw, struct brw_wm_compile *c)
     struct brw_compile *p = &c->func;
     struct brw_indirect stack_index = brw_indirect(0, 0);
 
-    c->reg_index = 0;
+    c->out_of_regs = GL_FALSE;
+
     prealloc_reg(c);
     brw_set_compression_control(p, BRW_COMPRESSION_NONE);
     brw_MOV(p, get_addr_reg(stack_index), brw_address(c->stack));
@@ -2603,13 +2745,15 @@ static void brw_wm_emit_glsl(struct brw_context *brw, struct brw_wm_compile *c)
     for (i = 0; i < c->nr_fp_insns; i++) {
         const struct prog_instruction *inst = &c->prog_instructions[i];
 
+        c->cur_inst = i;
+
 #if 0
         _mesa_printf("Inst %d: ", i);
         _mesa_print_instruction(inst);
 #endif
 
         /* fetch any constants that this instruction needs */
-        if (c->use_const_buffer)
+        if (c->fp->use_const_buffer)
            fetch_constants(c, inst);
 
 	if (inst->CondUpdate)
@@ -2833,17 +2977,13 @@ static void brw_wm_emit_glsl(struct brw_context *brw, struct brw_wm_compile *c)
 		_mesa_printf("unsupported IR in fragment shader %d\n",
 			inst->Opcode);
 	}
+
 	if (inst->CondUpdate)
 	    brw_set_predicate_control(p, BRW_PREDICATE_NORMAL);
 	else
 	    brw_set_predicate_control(p, BRW_PREDICATE_NONE);
     }
     post_wm_emit(c);
-
-    if (c->reg_index >= BRW_WM_MAX_GRF) {
-        _mesa_problem(NULL, "Ran out of registers in brw_wm_emit_glsl()");
-        /* XXX we need to do some proper error recovery here */
-    }
 }
 
 
@@ -2867,6 +3007,6 @@ void brw_wm_glsl_emit(struct brw_context *brw, struct brw_wm_compile *c)
         brw_wm_print_program(c, "brw_wm_glsl_emit done");
     }
 
-    c->prog_data.total_grf = c->reg_index;
+    c->prog_data.total_grf = num_grf_used(c);
     c->prog_data.total_scratch = 0;
 }
diff --git a/src/mesa/drivers/dri/i965/brw_wm_sampler_state.c b/src/mesa/drivers/dri/i965/brw_wm_sampler_state.c
index 1fc9f01372..c604ef0162 100644
--- a/src/mesa/drivers/dri/i965/brw_wm_sampler_state.c
+++ b/src/mesa/drivers/dri/i965/brw_wm_sampler_state.c
@@ -152,7 +152,7 @@ static void brw_update_sampler_state(struct wm_sampler_entry *key,
       sampler->ss0.mag_filter = BRW_MAPFILTER_ANISOTROPIC;
 
       if (key->max_aniso > 2.0) {
-	 sampler->ss3.max_aniso = MAX2((key->max_aniso - 2) / 2,
+	 sampler->ss3.max_aniso = MIN2((key->max_aniso - 2) / 2,
 				       BRW_ANISORATIO_16);
       }
    }
diff --git a/src/mesa/drivers/dri/i965/brw_wm_surface_state.c b/src/mesa/drivers/dri/i965/brw_wm_surface_state.c
index 71840d1e4e..c49a5f6b4e 100644
--- a/src/mesa/drivers/dri/i965/brw_wm_surface_state.c
+++ b/src/mesa/drivers/dri/i965/brw_wm_surface_state.c
@@ -176,22 +176,6 @@ static GLuint translate_tex_format( GLuint mesa_format, GLenum internal_format,
    }
 }
 
-
-/**
- * Use same key for WM and VS surfaces.
- */
-struct brw_surface_key {
-   GLenum target, depthmode;
-   dri_bo *bo;
-   GLint format, internal_format;
-   GLint first_level, last_level;
-   GLint width, height, depth;
-   GLint pitch, cpp;
-   uint32_t tiling;
-   GLuint offset;
-};
-
-
 static void
 brw_set_surface_tiling(struct brw_surface_state *surf, uint32_t tiling)
 {
@@ -268,7 +252,7 @@ brw_create_texture_surface( struct brw_context *brw,
       surf.ss0.cube_neg_z = 1;
    }
 
-   bo = brw_upload_cache(&brw->cache, BRW_SS_SURFACE,
+   bo = brw_upload_cache(&brw->surface_cache, BRW_SS_SURFACE,
 			 key, sizeof(*key),
 			 &key->bo, key->bo ? 1 : 0,
 			 &surf, sizeof(surf),
@@ -321,10 +305,11 @@ brw_update_texture_surface( GLcontext *ctx, GLuint unit )
    key.tiling = intelObj->mt->region->tiling;
 
    dri_bo_unreference(brw->wm.surf_bo[surf]);
-   brw->wm.surf_bo[surf] = brw_search_cache(&brw->cache, BRW_SS_SURFACE,
-                                         &key, sizeof(key),
-                                         &key.bo, key.bo ? 1 : 0,
-                                         NULL);
+   brw->wm.surf_bo[surf] = brw_search_cache(&brw->surface_cache,
+                                            BRW_SS_SURFACE,
+                                            &key, sizeof(key),
+                                            &key.bo, key.bo ? 1 : 0,
+                                            NULL);
    if (brw->wm.surf_bo[surf] == NULL) {
       brw->wm.surf_bo[surf] = brw_create_texture_surface(brw, &key);
    }
@@ -336,7 +321,7 @@ brw_update_texture_surface( GLcontext *ctx, GLuint unit )
  * Create the constant buffer surface.  Vertex/fragment shader constants will be
  * read from this buffer with Data Port Read instructions/messages.
  */
-static dri_bo *
+dri_bo *
 brw_create_constant_surface( struct brw_context *brw,
                              struct brw_surface_key *key )
 {
@@ -362,7 +347,7 @@ brw_create_constant_surface( struct brw_context *brw,
    surf.ss3.pitch = (key->pitch * key->cpp) - 1; /* ignored?? */
    brw_set_surface_tiling(&surf, key->tiling); /* tiling now allowed */
  
-   bo = brw_upload_cache(&brw->cache, BRW_SS_SURFACE,
+   bo = brw_upload_cache(&brw->surface_cache, BRW_SS_SURFACE,
 			 key, sizeof(*key),
 			 &key->bo, key->bo ? 1 : 0,
 			 &surf, sizeof(surf),
@@ -380,39 +365,70 @@ brw_create_constant_surface( struct brw_context *brw,
    return bo;
 }
 
+/* Creates a new WM constant buffer reflecting the current fragment program's
+ * constants, if needed by the fragment program.
+ *
+ * Otherwise, constants go through the CURBEs using the brw_constant_buffer
+ * state atom.
+ */
+static drm_intel_bo *
+brw_wm_update_constant_buffer(struct brw_context *brw)
+{
+   struct intel_context *intel = &brw->intel;
+   struct brw_fragment_program *fp =
+      (struct brw_fragment_program *) brw->fragment_program;
+   const struct gl_program_parameter_list *params = fp->program.Base.Parameters;
+   const int size = params->NumParameters * 4 * sizeof(GLfloat);
+   drm_intel_bo *const_buffer;
+
+   /* BRW_NEW_FRAGMENT_PROGRAM */
+   if (!fp->use_const_buffer)
+      return NULL;
+
+   const_buffer = drm_intel_bo_alloc(intel->bufmgr, "fp_const_buffer",
+				     size, 64);
+
+   /* _NEW_PROGRAM_CONSTANTS */
+   dri_bo_subdata(const_buffer, 0, size, params->ParameterValues);
+
+   return const_buffer;
+}
 
 /**
  * Update the surface state for a WM constant buffer.
  * The constant buffer will be (re)allocated here if needed.
  */
-static dri_bo *
+static void
 brw_update_wm_constant_surface( GLcontext *ctx,
-                                GLuint surf,
-                                dri_bo *const_buffer,
-                                const struct gl_program_parameter_list *params)
+                                GLuint surf)
 {
    struct brw_context *brw = brw_context(ctx);
    struct brw_surface_key key;
-   struct intel_context *intel = &brw->intel;
-   const int size = params->NumParameters * 4 * sizeof(GLfloat);
+   struct brw_fragment_program *fp =
+      (struct brw_fragment_program *) brw->fragment_program;
+   const struct gl_program_parameter_list *params =
+      fp->program.Base.Parameters;
 
-   /* free old const buffer if too small */
-   if (const_buffer && const_buffer->size < size) {
-      dri_bo_unreference(const_buffer);
-      const_buffer = NULL;
-   }
+   /* If we're in this state update atom, we need to update WM constants, so
+    * free the old buffer and create a new one for the new contents.
+    */
+   dri_bo_unreference(fp->const_buffer);
+   fp->const_buffer = brw_wm_update_constant_buffer(brw);
 
-   /* alloc new buffer if needed */
-   if (!const_buffer) {
-      const_buffer =
-         drm_intel_bo_alloc(intel->bufmgr, "fp_const_buffer", size, 64);
+   /* If there's no constant buffer, then no surface BO is needed to point at
+    * it.
+    */
+   if (fp->const_buffer == 0) {
+      drm_intel_bo_unreference(brw->wm.surf_bo[surf]);
+      brw->wm.surf_bo[surf] = NULL;
+      return;
    }
 
    memset(&key, 0, sizeof(key));
 
    key.format = MESA_FORMAT_RGBA_FLOAT32;
    key.internal_format = GL_RGBA;
-   key.bo = const_buffer;
+   key.bo = fp->const_buffer;
    key.depthmode = GL_NONE;
    key.pitch = params->NumParameters;
    key.width = params->NumParameters;
@@ -427,77 +443,59 @@ brw_update_wm_constant_surface( GLcontext *ctx,
    */
 
    dri_bo_unreference(brw->wm.surf_bo[surf]);
-   brw->wm.surf_bo[surf] = brw_search_cache(&brw->cache, BRW_SS_SURFACE,
+   brw->wm.surf_bo[surf] = brw_search_cache(&brw->surface_cache,
+                                            BRW_SS_SURFACE,
                                             &key, sizeof(key),
                                             &key.bo, key.bo ? 1 : 0,
                                             NULL);
    if (brw->wm.surf_bo[surf] == NULL) {
       brw->wm.surf_bo[surf] = brw_create_constant_surface(brw, &key);
    }
-
-   return const_buffer;
+   brw->state.dirty.brw |= BRW_NEW_WM_SURFACES;
 }
 
-
 /**
- * Update the surface state for a VS constant buffer.
- * The constant buffer will be (re)allocated here if needed.
+ * Updates surface / buffer for fragment shader constant buffer, if
+ * one is required.
+ *
+ * This consumes the state updates for the constant buffer, and produces
+ * BRW_NEW_WM_SURFACES to get picked up by brw_prepare_wm_surfaces for
+ * inclusion in the binding table.
  */
-static dri_bo *
-brw_update_vs_constant_surface( GLcontext *ctx,
-                                GLuint surf,
-                                dri_bo *const_buffer,
-                                const struct gl_program_parameter_list *params)
+static void prepare_wm_constant_surface(struct brw_context *brw )
 {
-   struct brw_context *brw = brw_context(ctx);
-   struct brw_surface_key key;
-   struct intel_context *intel = &brw->intel;
-   const int size = params->NumParameters * 4 * sizeof(GLfloat);
-
-   assert(surf == 0);
-
-   /* free old const buffer if too small */
-   if (const_buffer && const_buffer->size < size) {
-      dri_bo_unreference(const_buffer);
-      const_buffer = NULL;
-   }
-
-   /* alloc new buffer if needed */
-   if (!const_buffer) {
-      const_buffer =
-         drm_intel_bo_alloc(intel->bufmgr, "vp_const_buffer", size, 64);
-   }
-
-   memset(&key, 0, sizeof(key));
-
-   key.format = MESA_FORMAT_RGBA_FLOAT32;
-   key.internal_format = GL_RGBA;
-   key.bo = const_buffer;
-   key.depthmode = GL_NONE;
-   key.pitch = params->NumParameters;
-   key.width = params->NumParameters;
-   key.height = 1;
-   key.depth = 1;
-   key.cpp = 16;
+   GLcontext *ctx = &brw->intel.ctx;
+   struct brw_fragment_program *fp =
+      (struct brw_fragment_program *) brw->fragment_program;
+   GLuint surf = SURF_INDEX_FRAG_CONST_BUFFER;
 
-   /*
-   printf("%s:\n", __FUNCTION__);
-   printf("  width %d  height %d  depth %d  cpp %d  pitch %d\n",
-          key.width, key.height, key.depth, key.cpp, key.pitch);
-   */
+   drm_intel_bo_unreference(fp->const_buffer);
+   fp->const_buffer = brw_wm_update_constant_buffer(brw);
 
-   dri_bo_unreference(brw->vs.surf_bo[surf]);
-   brw->vs.surf_bo[surf] = brw_search_cache(&brw->cache, BRW_SS_SURFACE,
-                                            &key, sizeof(key),
-                                            &key.bo, key.bo ? 1 : 0,
-                                            NULL);
-   if (brw->vs.surf_bo[surf] == NULL) {
-      brw->vs.surf_bo[surf] = brw_create_constant_surface(brw, &key);
+   /* If there's no constant buffer, then no surface BO is needed to point at
+    * it.
+    */
+   if (fp->const_buffer == 0) {
+      if (brw->wm.surf_bo[surf] != NULL) {
+	 drm_intel_bo_unreference(brw->wm.surf_bo[surf]);
+	 brw->wm.surf_bo[surf] = NULL;
+	 brw->state.dirty.brw |= BRW_NEW_WM_SURFACES;
+      }
+      return;
    }
 
-   return const_buffer;
+   brw_update_wm_constant_surface(ctx, surf);
 }
 
+const struct brw_tracked_state brw_wm_constant_surface = {
+   .dirty = {
+      .mesa = (_NEW_PROGRAM_CONSTANTS),
+      .brw = (BRW_NEW_FRAGMENT_PROGRAM),
+      .cache = 0
+   },
+   .prepare = prepare_wm_constant_surface,
+};
+
 
 /**
  * Sets up a surface state structure to point at the given region.
@@ -507,7 +505,7 @@ brw_update_vs_constant_surface( GLcontext *ctx,
 static void
 brw_update_renderbuffer_surface(struct brw_context *brw,
 				struct gl_renderbuffer *rb,
-				unsigned int unit, GLboolean cached)
+				unsigned int unit)
 {
    GLcontext *ctx = &brw->intel.ctx;
    dri_bo *region_bo = NULL;
@@ -520,6 +518,7 @@ brw_update_renderbuffer_surface(struct brw_context *brw,
       GLubyte color_mask[4];
       GLboolean color_blend;
       uint32_t tiling;
+      uint32_t draw_offset;
    } key;
 
    memset(&key, 0, sizeof(key));
@@ -550,6 +549,7 @@ brw_update_renderbuffer_surface(struct brw_context *brw,
       key.height = region->height;
       key.pitch = region->pitch;
       key.cpp = region->cpp;
+      key.draw_offset = region->draw_offset; /* cur 3d or cube face offset */
    } else {
       key.surface_type = BRW_SURFACE_NULL;
       key.surface_format = BRW_SURFACEFORMAT_B8G8R8A8_UNORM;
@@ -557,6 +557,7 @@ brw_update_renderbuffer_surface(struct brw_context *brw,
       key.width = 1;
       key.height = 1;
       key.cpp = 4;
+      key.draw_offset = 0;
    }
    memcpy(key.color_mask, ctx->Color.ColorMask,
 	  sizeof(key.color_mask));
@@ -564,12 +565,11 @@ brw_update_renderbuffer_surface(struct brw_context *brw,
 		      ctx->Color.BlendEnabled);
 
    dri_bo_unreference(brw->wm.surf_bo[unit]);
-   brw->wm.surf_bo[unit] = NULL;
-   if (cached) 
-       brw->wm.surf_bo[unit] = brw_search_cache(&brw->cache, BRW_SS_SURFACE,
-	       &key, sizeof(key),
-	       &region_bo, 1,
-	       NULL);
+   brw->wm.surf_bo[unit] = brw_search_cache(&brw->surface_cache,
+					    BRW_SS_SURFACE,
+					    &key, sizeof(key),
+					    &region_bo, 1,
+					    NULL);
 
    if (brw->wm.surf_bo[unit] == NULL) {
       struct brw_surface_state surf;
@@ -578,8 +578,9 @@ brw_update_renderbuffer_surface(struct brw_context *brw,
 
       surf.ss0.surface_format = key.surface_format;
       surf.ss0.surface_type = key.surface_type;
+      surf.ss1.base_addr =  key.draw_offset;
       if (region_bo != NULL)
-	 surf.ss1.base_addr = region_bo->offset; /* reloc */
+	 surf.ss1.base_addr += region_bo->offset; /* reloc */
 
       surf.ss2.width = key.width - 1;
       surf.ss2.height = key.height - 1;
@@ -594,7 +595,8 @@ brw_update_renderbuffer_surface(struct brw_context *brw,
       surf.ss0.writedisable_alpha = !key.color_mask[3];
 
       /* Key size will never match key size for textures, so we're safe. */
-      brw->wm.surf_bo[unit] = brw_upload_cache(&brw->cache, BRW_SS_SURFACE,
+      brw->wm.surf_bo[unit] = brw_upload_cache(&brw->surface_cache,
+                                               BRW_SS_SURFACE,
                                                &key, sizeof(key),
 					       &region_bo, 1,
 					       &surf, sizeof(surf),
@@ -604,12 +606,12 @@ brw_update_renderbuffer_surface(struct brw_context *brw,
 	  * them both.  We might be able to figure out from other state
 	  * a more restrictive relocation to emit.
 	  */
-	 dri_bo_emit_reloc(brw->wm.surf_bo[unit],
-			   I915_GEM_DOMAIN_RENDER,
-			   I915_GEM_DOMAIN_RENDER,
-			   0,
-			   offsetof(struct brw_surface_state, ss1),
-			   region_bo);
+	 drm_intel_bo_emit_reloc(brw->wm.surf_bo[unit],
+				 offsetof(struct brw_surface_state, ss1),
+				 region_bo,
+				 key.draw_offset,
+				 I915_GEM_DOMAIN_RENDER,
+				 I915_GEM_DOMAIN_RENDER);
       }
    }
 }
@@ -626,7 +628,7 @@ brw_wm_get_binding_table(struct brw_context *brw)
 
    assert(brw->wm.nr_surfaces <= BRW_WM_MAX_SURF);
 
-   bind_bo = brw_search_cache(&brw->cache, BRW_SS_SURF_BIND,
+   bind_bo = brw_search_cache(&brw->surface_cache, BRW_SS_SURF_BIND,
 			      NULL, 0,
 			      brw->wm.surf_bo, brw->wm.nr_surfaces,
 			      NULL);
@@ -642,7 +644,7 @@ brw_wm_get_binding_table(struct brw_context *brw)
          else
             data[i] = 0;
 
-      bind_bo = brw_upload_cache( &brw->cache, BRW_SS_SURF_BIND,
+      bind_bo = brw_upload_cache( &brw->surface_cache, BRW_SS_SURF_BIND,
 				  NULL, 0,
 				  brw->wm.surf_bo, brw->wm.nr_surfaces,
 				  data, data_size,
@@ -678,27 +680,17 @@ static void prepare_wm_surfaces(struct brw_context *brw )
       for (i = 0; i < ctx->DrawBuffer->_NumColorDrawBuffers; i++) {
          brw_update_renderbuffer_surface(brw,
 					 ctx->DrawBuffer->_ColorDrawBuffers[i],
-					 i,
-					 GL_FALSE);
+					 i);
       }
    } else {
-      brw_update_renderbuffer_surface(brw, NULL, 0, GL_TRUE);
+      brw_update_renderbuffer_surface(brw, NULL, 0);
    }
 
    old_nr_surfaces = brw->wm.nr_surfaces;
    brw->wm.nr_surfaces = MAX_DRAW_BUFFERS;
 
-   /* Update surface / buffer for fragment shader constant buffer */
-   {
-      const GLuint surf = SURF_INDEX_FRAG_CONST_BUFFER;
-      struct brw_fragment_program *fp =
-         (struct brw_fragment_program *) brw->fragment_program;
-      fp->const_buffer =
-         brw_update_wm_constant_surface(ctx, surf, fp->const_buffer,
-                                     fp->program.Base.Parameters);
-
-      brw->wm.nr_surfaces = surf + 1;
-   }
+   if (brw->wm.surf_bo[SURF_INDEX_FRAG_CONST_BUFFER] != NULL)
+       brw->wm.nr_surfaces = SURF_INDEX_FRAG_CONST_BUFFER + 1;
 
    /* Update surfaces for textures */
    for (i = 0; i < BRW_MAX_TEX_UNIT; i++) {
@@ -731,100 +723,16 @@ static void prepare_wm_surfaces(struct brw_context *brw )
       brw->state.dirty.brw |= BRW_NEW_NR_WM_SURFACES;
 }
 
-
-/**
- * Constructs the binding table for the VS surface state.
- */
-static dri_bo *
-brw_vs_get_binding_table(struct brw_context *brw)
-{
-   dri_bo *bind_bo;
-
-   assert(brw->vs.nr_surfaces <= BRW_VS_MAX_SURF);
-
-   bind_bo = brw_search_cache(&brw->cache, BRW_SS_SURF_BIND,
-			      NULL, 0,
-			      brw->vs.surf_bo, brw->vs.nr_surfaces,
-			      NULL);
-
-   if (bind_bo == NULL) {
-      GLuint data_size = brw->vs.nr_surfaces * sizeof(GLuint);
-      uint32_t *data = malloc(data_size);
-      int i;
-
-      for (i = 0; i < brw->vs.nr_surfaces; i++)
-         if (brw->vs.surf_bo[i])
-            data[i] = brw->vs.surf_bo[i]->offset;
-         else
-            data[i] = 0;
-
-      bind_bo = brw_upload_cache( &brw->cache, BRW_SS_SURF_BIND,
-				  NULL, 0,
-				  brw->vs.surf_bo, brw->vs.nr_surfaces,
-				  data, data_size,
-				  NULL, NULL);
-
-      /* Emit binding table relocations to surface state */
-      for (i = 0; i < BRW_VS_MAX_SURF; i++) {
-	 if (brw->vs.surf_bo[i] != NULL) {
-	    dri_bo_emit_reloc(bind_bo,
-			      I915_GEM_DOMAIN_INSTRUCTION, 0,
-			      0,
-			      i * sizeof(GLuint),
-			      brw->vs.surf_bo[i]);
-	 }
-      }
-
-      free(data);
-   }
-
-   return bind_bo;
-}
-
-
-/**
- * Vertex shader surfaces.  Just constant buffer for now.  Could add vertex 
- * shader textures in the future.
- */
-static void prepare_vs_surfaces(struct brw_context *brw )
-{
-   GLcontext *ctx = &brw->intel.ctx;
-
-   /* Update surface / buffer for vertex shader constant buffer */
-   {
-      const GLuint surf = SURF_INDEX_VERT_CONST_BUFFER;
-      struct brw_vertex_program *vp =
-         (struct brw_vertex_program *) brw->vertex_program;
-      vp->const_buffer =
-         brw_update_vs_constant_surface(ctx, surf, vp->const_buffer,
-                                        vp->program.Base.Parameters);
-
-      brw->vs.nr_surfaces = 1;
-   }
-
-   dri_bo_unreference(brw->vs.bind_bo);
-   brw->vs.bind_bo = brw_vs_get_binding_table(brw);
-
-   if (1)
-      brw->state.dirty.brw |= BRW_NEW_NR_VS_SURFACES;
-}
-
-
-static void
-prepare_surfaces(struct brw_context *brw)
-{
-   prepare_wm_surfaces(brw);
-   prepare_vs_surfaces(brw);
-}
-
-
 const struct brw_tracked_state brw_wm_surfaces = {
    .dirty = {
-      .mesa = _NEW_COLOR | _NEW_TEXTURE | _NEW_BUFFERS | _NEW_PROGRAM,
-      .brw = BRW_NEW_CONTEXT,
+      .mesa = (_NEW_COLOR |
+               _NEW_TEXTURE |
+               _NEW_BUFFERS),
+      .brw = (BRW_NEW_CONTEXT |
+	      BRW_NEW_WM_SURFACES),
       .cache = 0
    },
-   .prepare = prepare_surfaces,
+   .prepare = prepare_wm_surfaces,
 };
 
 
diff --git a/src/mesa/drivers/dri/i965/intel_generatemipmap.c b/src/mesa/drivers/dri/i965/intel_generatemipmap.c
new file mode 120000
index 0000000000..4c6b37ada0
--- /dev/null
+++ b/src/mesa/drivers/dri/i965/intel_generatemipmap.c
@@ -0,0 +1 @@
+../intel/intel_generatemipmap.c
\ No newline at end of file