i965: Move sampler state to state streaming.

Overall, across this series since the last set of numbers, gen6 3DMMES
taiji performance has dropped 0.8% +/- 0.3% (n=15), probably due to
the increased reissuing of state from some of the state objects that
otherwise never changed, and increased occurrence of the per-batch
overhead as we've increased how much we put in the batch BO without
increasing the batch BO's size.

Reviewed-by: Kenneth Graunke <[email protected]>

9 files changed, 143 insertions, 238 deletions

diff --git a/src/mesa/drivers/dri/i965/brw_context.h b/src/mesa/drivers/dri/i965/brw_context.h
index b61a6ff9bc3..748a7503833 100644
--- a/src/mesa/drivers/dri/i965/brw_context.h
+++ b/src/mesa/drivers/dri/i965/brw_context.h
@@ -311,7 +311,6 @@ enum brw_cache_id {
    BRW_CC_VP,
    BRW_CC_UNIT,
    BRW_WM_PROG,
-   BRW_SAMPLER_DEFAULT_COLOR,
    BRW_SAMPLER,
    BRW_WM_UNIT,
    BRW_SF_PROG,
@@ -384,7 +383,6 @@ struct brw_tracked_state {
 #define CACHE_NEW_CC_VP                  (1<<BRW_CC_VP)
 #define CACHE_NEW_CC_UNIT                (1<<BRW_CC_UNIT)
 #define CACHE_NEW_WM_PROG                (1<<BRW_WM_PROG)
-#define CACHE_NEW_SAMPLER_DEFAULT_COLOR  (1<<BRW_SAMPLER_DEFAULT_COLOR)
 #define CACHE_NEW_SAMPLER                (1<<BRW_SAMPLER)
 #define CACHE_NEW_WM_UNIT                (1<<BRW_WM_UNIT)
 #define CACHE_NEW_SF_PROG                (1<<BRW_SF_PROG)
@@ -684,8 +682,9 @@ struct brw_context
        */
       GLbitfield input_size_masks[4];
 
-      /** Array of surface default colors (texture border color) */
-      drm_intel_bo *sdc_bo[BRW_MAX_TEX_UNIT];
+      /** offsets in the batch to sampler default colors (texture border color)
+       */
+      uint32_t sdc_offset[BRW_MAX_TEX_UNIT];
 
       GLuint render_surf;
       GLuint nr_surfaces;      
@@ -694,7 +693,7 @@ struct brw_context
       drm_intel_bo *scratch_bo;
 
       GLuint sampler_count;
-      drm_intel_bo *sampler_bo;
+      uint32_t sampler_offset;
 
       /** Binding table of pointers to surf_bo entries */
       uint32_t bind_bo_offset;
diff --git a/src/mesa/drivers/dri/i965/brw_draw.c b/src/mesa/drivers/dri/i965/brw_draw.c
index 0c93a03151b..9ab533179b8 100644
--- a/src/mesa/drivers/dri/i965/brw_draw.c
+++ b/src/mesa/drivers/dri/i965/brw_draw.c
@@ -367,6 +367,9 @@ static GLboolean brw_try_draw_prims( struct gl_context *ctx,
       int estimated_max_prim_size;
 
       estimated_max_prim_size = 512; /* batchbuffer commands */
+      estimated_max_prim_size += (BRW_MAX_TEX_UNIT *
+				  (sizeof(struct brw_sampler_state) +
+				   sizeof(struct gen5_sampler_default_color)));
       estimated_max_prim_size += 1024; /* gen6 VS push constants */
       estimated_max_prim_size += 1024; /* gen6 WM push constants */
       estimated_max_prim_size += 512; /* misc. pad */
diff --git a/src/mesa/drivers/dri/i965/brw_state_cache.c b/src/mesa/drivers/dri/i965/brw_state_cache.c
index 01eeb19a684..82a0403b2dd 100644
--- a/src/mesa/drivers/dri/i965/brw_state_cache.c
+++ b/src/mesa/drivers/dri/i965/brw_state_cache.c
@@ -367,7 +367,6 @@ brw_init_non_surface_cache(struct brw_context *brw)
    brw_init_cache_id(cache, "CC_VP", BRW_CC_VP);
    brw_init_cache_id(cache, "CC_UNIT", BRW_CC_UNIT);
    brw_init_cache_id(cache, "WM_PROG", BRW_WM_PROG);
-   brw_init_cache_id(cache, "SAMPLER_DEFAULT_COLOR", BRW_SAMPLER_DEFAULT_COLOR);
    brw_init_cache_id(cache, "SAMPLER", BRW_SAMPLER);
    brw_init_cache_id(cache, "WM_UNIT", BRW_WM_UNIT);
    brw_init_cache_id(cache, "SF_PROG", BRW_SF_PROG);
diff --git a/src/mesa/drivers/dri/i965/brw_state_dump.c b/src/mesa/drivers/dri/i965/brw_state_dump.c
index 06cf97519ad..930eb5155db 100644
--- a/src/mesa/drivers/dri/i965/brw_state_dump.c
+++ b/src/mesa/drivers/dri/i965/brw_state_dump.c
@@ -144,12 +144,7 @@ static void dump_wm_sampler_state(struct brw_context *brw)
    struct gl_context *ctx = &brw->intel.ctx;
    int i;
 
-   if (!brw->wm.sampler_bo) {
-      fprintf(stderr, "WM_SAMPLER: NULL\n");
-      return;
-   }
-
-   drm_intel_bo_map(brw->wm.sampler_bo, GL_FALSE);
+   drm_intel_bo_map(intel->batch.bo, GL_FALSE);
    for (i = 0; i < BRW_MAX_TEX_UNIT; i++) {
       unsigned int offset;
       uint32_t sdc_offset;
@@ -161,9 +156,11 @@ static void dump_wm_sampler_state(struct brw_context *brw)
 	 continue;
       }
 
-      offset = brw->wm.sampler_bo->offset +
-	 i * sizeof(struct brw_sampler_state);
-      samp = (struct brw_sampler_state *)(brw->wm.sampler_bo->virtual +
+      offset = (intel->batch.bo->offset +
+		brw->wm.sampler_offset +
+		i * sizeof(struct brw_sampler_state));
+      samp = (struct brw_sampler_state *)(intel->batch.bo->virtual +
+					  brw->wm.sampler_offset +
 					  i * sizeof(struct brw_sampler_state));
 
       sprintf(name, "WM SAMP%d", i);
@@ -174,10 +171,10 @@ static void dump_wm_sampler_state(struct brw_context *brw)
 
       sprintf(name, " WM SDC%d", i);
 
-      drm_intel_bo_map(brw->wm.sdc_bo[i], GL_FALSE);
-      sdc_offset = brw->wm.sdc_bo[i]->offset;
+      sdc_offset = intel->batch.bo->offset + brw->wm.sdc_offset[i];
       if (intel->gen >= 5) {
-	 struct gen5_sampler_default_color *sdc = brw->wm.sdc_bo[i]->virtual;
+	 struct gen5_sampler_default_color *sdc = (intel->batch.bo->virtual +
+						   brw->wm.sdc_offset[i]);
 	 state_out(name, sdc, sdc_offset, 0, "unorm rgba\n");
 	 state_out(name, sdc, sdc_offset, 1, "r %f\n", sdc->f[0]);
 	 state_out(name, sdc, sdc_offset, 2, "b %f\n", sdc->f[1]);
@@ -191,15 +188,15 @@ static void dump_wm_sampler_state(struct brw_context *brw)
 	 state_out(name, sdc, sdc_offset, 10, "s16 ba\n");
 	 state_out(name, sdc, sdc_offset, 11, "s8 rgba\n");
       } else {
-	 struct brw_sampler_default_color *sdc = brw->wm.sdc_bo[i]->virtual;
+	 struct brw_sampler_default_color *sdc = (intel->batch.bo->virtual +
+						  brw->wm.sdc_offset[i]);
 	 state_out(name, sdc, sdc_offset, 0, "r %f\n", sdc->color[0]);
 	 state_out(name, sdc, sdc_offset, 1, "g %f\n", sdc->color[1]);
 	 state_out(name, sdc, sdc_offset, 2, "b %f\n", sdc->color[2]);
 	 state_out(name, sdc, sdc_offset, 3, "a %f\n", sdc->color[3]);
       }
-      drm_intel_bo_unmap(brw->wm.sdc_bo[i]);
    }
-   drm_intel_bo_unmap(brw->wm.sampler_bo);
+   drm_intel_bo_unmap(intel->batch.bo);
 }
 
 static void dump_sf_viewport_state(struct brw_context *brw)
diff --git a/src/mesa/drivers/dri/i965/brw_state_upload.c b/src/mesa/drivers/dri/i965/brw_state_upload.c
index de67114e024..e1485772149 100644
--- a/src/mesa/drivers/dri/i965/brw_state_upload.c
+++ b/src/mesa/drivers/dri/i965/brw_state_upload.c
@@ -317,7 +317,6 @@ static struct dirty_bit_map cache_bits[] = {
    DEFINE_BIT(CACHE_NEW_CC_VP),
    DEFINE_BIT(CACHE_NEW_CC_UNIT),
    DEFINE_BIT(CACHE_NEW_WM_PROG),
-   DEFINE_BIT(CACHE_NEW_SAMPLER_DEFAULT_COLOR),
    DEFINE_BIT(CACHE_NEW_SAMPLER),
    DEFINE_BIT(CACHE_NEW_WM_UNIT),
    DEFINE_BIT(CACHE_NEW_SF_PROG),
diff --git a/src/mesa/drivers/dri/i965/brw_vtbl.c b/src/mesa/drivers/dri/i965/brw_vtbl.c
index 7add92aab43..36c1cc8bba8 100644
--- a/src/mesa/drivers/dri/i965/brw_vtbl.c
+++ b/src/mesa/drivers/dri/i965/brw_vtbl.c
@@ -85,9 +85,6 @@ static void brw_destroy_context( struct intel_context *intel )
    dri_bo_release(&brw->clip.state_bo);
    dri_bo_release(&brw->sf.prog_bo);
    dri_bo_release(&brw->sf.state_bo);
-   for (i = 0; i < BRW_MAX_TEX_UNIT; i++)
-      dri_bo_release(&brw->wm.sdc_bo[i]);
-   dri_bo_release(&brw->wm.sampler_bo);
    dri_bo_release(&brw->wm.prog_bo);
    dri_bo_release(&brw->wm.const_bo);
    dri_bo_release(&brw->cc.prog_bo);
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 9ceac5adf05..c24494b645f 100644
--- a/src/mesa/drivers/dri/i965/brw_wm_sampler_state.c
+++ b/src/mesa/drivers/dri/i965/brw_wm_sampler_state.c
@@ -66,91 +66,93 @@ static GLuint translate_wrap_mode( GLenum wrap )
    }
 }
 
-static drm_intel_bo *upload_default_color( struct brw_context *brw,
-				     const GLfloat *color )
+static void
+upload_default_color(struct brw_context *brw, struct gl_sampler_object *sampler,
+		     int unit)
 {
    struct intel_context *intel = &brw->intel;
+   struct gl_context *ctx = &intel->ctx;
+   struct gl_texture_unit *texUnit = &ctx->Texture.Unit[unit];
+   struct gl_texture_object *texObj = texUnit->_Current;
+   struct gl_texture_image *firstImage = texObj->Image[0][texObj->BaseLevel];
+   float color[4];
+
+   if (firstImage->_BaseFormat == GL_DEPTH_COMPONENT) {
+      /* GL specs that border color for depth textures is taken from the
+       * R channel, while the hardware uses A.  Spam R into all the
+       * channels for safety.
+       */
+      color[0] = sampler->BorderColor.f[0];
+      color[1] = sampler->BorderColor.f[0];
+      color[2] = sampler->BorderColor.f[0];
+      color[3] = sampler->BorderColor.f[0];
+   } else {
+      color[0] = sampler->BorderColor.f[0];
+      color[1] = sampler->BorderColor.f[1];
+      color[2] = sampler->BorderColor.f[2];
+      color[3] = sampler->BorderColor.f[3];
+   }
 
    if (intel->gen >= 5) {
-      struct gen5_sampler_default_color sdc;
-
-      memset(&sdc, 0, sizeof(sdc));
-
-      UNCLAMPED_FLOAT_TO_UBYTE(sdc.ub[0], color[0]);
-      UNCLAMPED_FLOAT_TO_UBYTE(sdc.ub[1], color[1]);
-      UNCLAMPED_FLOAT_TO_UBYTE(sdc.ub[2], color[2]);
-      UNCLAMPED_FLOAT_TO_UBYTE(sdc.ub[3], color[3]);
-
-      UNCLAMPED_FLOAT_TO_USHORT(sdc.us[0], color[0]);
-      UNCLAMPED_FLOAT_TO_USHORT(sdc.us[1], color[1]);
-      UNCLAMPED_FLOAT_TO_USHORT(sdc.us[2], color[2]);
-      UNCLAMPED_FLOAT_TO_USHORT(sdc.us[3], color[3]);
-
-      UNCLAMPED_FLOAT_TO_SHORT(sdc.s[0], color[0]);
-      UNCLAMPED_FLOAT_TO_SHORT(sdc.s[1], color[1]);
-      UNCLAMPED_FLOAT_TO_SHORT(sdc.s[2], color[2]);
-      UNCLAMPED_FLOAT_TO_SHORT(sdc.s[3], color[3]);
-
-      sdc.hf[0] = _mesa_float_to_half(color[0]);
-      sdc.hf[1] = _mesa_float_to_half(color[1]);
-      sdc.hf[2] = _mesa_float_to_half(color[2]);
-      sdc.hf[3] = _mesa_float_to_half(color[3]);
-
-      sdc.b[0] = sdc.s[0] >> 8;
-      sdc.b[1] = sdc.s[1] >> 8;
-      sdc.b[2] = sdc.s[2] >> 8;
-      sdc.b[3] = sdc.s[3] >> 8;
-
-      sdc.f[0] = color[0];
-      sdc.f[1] = color[1];
-      sdc.f[2] = color[2];
-      sdc.f[3] = color[3];
-
-      return brw_cache_data(&brw->cache, BRW_SAMPLER_DEFAULT_COLOR,
-			    &sdc, sizeof(sdc));
-   } else {
-      struct brw_sampler_default_color sdc;
+      struct gen5_sampler_default_color *sdc;
 
-      COPY_4V(sdc.color, color);
+      sdc = brw_state_batch(brw, sizeof(*sdc), 32, &brw->wm.sdc_offset[unit]);
 
-      return brw_cache_data(&brw->cache, BRW_SAMPLER_DEFAULT_COLOR,
-			    &sdc, sizeof(sdc));
-   }
-}
+      memset(sdc, 0, sizeof(*sdc));
 
+      UNCLAMPED_FLOAT_TO_UBYTE(sdc->ub[0], color[0]);
+      UNCLAMPED_FLOAT_TO_UBYTE(sdc->ub[1], color[1]);
+      UNCLAMPED_FLOAT_TO_UBYTE(sdc->ub[2], color[2]);
+      UNCLAMPED_FLOAT_TO_UBYTE(sdc->ub[3], color[3]);
 
-struct wm_sampler_key {
-   int sampler_count;
+      UNCLAMPED_FLOAT_TO_USHORT(sdc->us[0], color[0]);
+      UNCLAMPED_FLOAT_TO_USHORT(sdc->us[1], color[1]);
+      UNCLAMPED_FLOAT_TO_USHORT(sdc->us[2], color[2]);
+      UNCLAMPED_FLOAT_TO_USHORT(sdc->us[3], color[3]);
 
-   struct wm_sampler_entry {
-      GLenum tex_target;
-      GLenum wrap_r, wrap_s, wrap_t;
-      float maxlod, minlod;
-      float lod_bias;
-      float max_aniso;
-      GLenum minfilter, magfilter;
-      GLenum comparemode, comparefunc;
+      UNCLAMPED_FLOAT_TO_SHORT(sdc->s[0], color[0]);
+      UNCLAMPED_FLOAT_TO_SHORT(sdc->s[1], color[1]);
+      UNCLAMPED_FLOAT_TO_SHORT(sdc->s[2], color[2]);
+      UNCLAMPED_FLOAT_TO_SHORT(sdc->s[3], color[3]);
 
-      /** If target is cubemap, take context setting.
-       */
-      GLboolean seamless_cube_map;
-   } sampler[BRW_MAX_TEX_UNIT];
-};
+      sdc->hf[0] = _mesa_float_to_half(color[0]);
+      sdc->hf[1] = _mesa_float_to_half(color[1]);
+      sdc->hf[2] = _mesa_float_to_half(color[2]);
+      sdc->hf[3] = _mesa_float_to_half(color[3]);
+
+      sdc->b[0] = sdc->s[0] >> 8;
+      sdc->b[1] = sdc->s[1] >> 8;
+      sdc->b[2] = sdc->s[2] >> 8;
+      sdc->b[3] = sdc->s[3] >> 8;
+
+      sdc->f[0] = color[0];
+      sdc->f[1] = color[1];
+      sdc->f[2] = color[2];
+      sdc->f[3] = color[3];
+   } else {
+      struct brw_sampler_default_color *sdc;
+
+      sdc = brw_state_batch(brw, sizeof(*sdc), 32, &brw->wm.sdc_offset[unit]);
+
+      COPY_4V(sdc->color, color);
+   }
+}
 
 /**
  * Sets the sampler state for a single unit based off of the sampler key
  * entry.
  */
 static void brw_update_sampler_state(struct brw_context *brw,
-				     struct wm_sampler_entry *key,
-				     drm_intel_bo *sdc_bo,
+				     int unit,
 				     struct brw_sampler_state *sampler)
 {
    struct intel_context *intel = &brw->intel;
+   struct gl_context *ctx = &intel->ctx;
+   struct gl_texture_unit *texUnit = &ctx->Texture.Unit[unit];
+   struct gl_texture_object *texObj = texUnit->_Current;
+   struct gl_sampler_object *gl_sampler = _mesa_get_samplerobj(ctx, unit);
 
-   memset(sampler, 0, sizeof(*sampler));
-
-   switch (key->minfilter) {
+   switch (gl_sampler->MinFilter) {
    case GL_NEAREST:
       sampler->ss0.min_filter = BRW_MAPFILTER_NEAREST;
       sampler->ss0.mip_filter = BRW_MIPFILTER_NONE;
@@ -181,17 +183,17 @@ static void brw_update_sampler_state(struct brw_context *brw,
 
    /* Set Anisotropy: 
     */
-   if (key->max_aniso > 1.0) {
+   if (gl_sampler->MaxAnisotropy > 1.0) {
       sampler->ss0.min_filter = BRW_MAPFILTER_ANISOTROPIC; 
       sampler->ss0.mag_filter = BRW_MAPFILTER_ANISOTROPIC;
 
-      if (key->max_aniso > 2.0) {
-	 sampler->ss3.max_aniso = MIN2((key->max_aniso - 2) / 2,
+      if (gl_sampler->MaxAnisotropy > 2.0) {
+	 sampler->ss3.max_aniso = MIN2((gl_sampler->MaxAnisotropy - 2) / 2,
 				       BRW_ANISORATIO_16);
       }
    }
    else {
-      switch (key->magfilter) {
+      switch (gl_sampler->MagFilter) {
       case GL_NEAREST:
 	 sampler->ss0.mag_filter = BRW_MAPFILTER_NEAREST;
 	 break;
@@ -203,9 +205,9 @@ static void brw_update_sampler_state(struct brw_context *brw,
       }  
    }
 
-   sampler->ss1.r_wrap_mode = translate_wrap_mode(key->wrap_r);
-   sampler->ss1.s_wrap_mode = translate_wrap_mode(key->wrap_s);
-   sampler->ss1.t_wrap_mode = translate_wrap_mode(key->wrap_t);
+   sampler->ss1.r_wrap_mode = translate_wrap_mode(gl_sampler->WrapR);
+   sampler->ss1.s_wrap_mode = translate_wrap_mode(gl_sampler->WrapS);
+   sampler->ss1.t_wrap_mode = translate_wrap_mode(gl_sampler->WrapT);
 
    if (intel->gen >= 6 &&
        sampler->ss0.min_filter != sampler->ss0.mag_filter)
@@ -214,9 +216,10 @@ static void brw_update_sampler_state(struct brw_context *brw,
    /* Cube-maps on 965 and later must use the same wrap mode for all 3
     * coordinate dimensions.  Futher, only CUBE and CLAMP are valid.
     */
-   if (key->tex_target == GL_TEXTURE_CUBE_MAP) {
-      if (key->seamless_cube_map &&
-	  (key->minfilter != GL_NEAREST || key->magfilter != GL_NEAREST)) {
+   if (texObj->Target == GL_TEXTURE_CUBE_MAP) {
+      if (ctx->Texture.CubeMapSeamless &&
+	  (gl_sampler->MinFilter != GL_NEAREST ||
+	   gl_sampler->MagFilter != GL_NEAREST)) {
 	 sampler->ss1.r_wrap_mode = BRW_TEXCOORDMODE_CUBE;
 	 sampler->ss1.s_wrap_mode = BRW_TEXCOORDMODE_CUBE;
 	 sampler->ss1.t_wrap_mode = BRW_TEXCOORDMODE_CUBE;
@@ -225,7 +228,7 @@ static void brw_update_sampler_state(struct brw_context *brw,
 	 sampler->ss1.s_wrap_mode = BRW_TEXCOORDMODE_CLAMP;
 	 sampler->ss1.t_wrap_mode = BRW_TEXCOORDMODE_CLAMP;
       }
-   } else if (key->tex_target == GL_TEXTURE_1D) {
+   } else if (texObj->Target == GL_TEXTURE_1D) {
       /* There's a bug in 1D texture sampling - it actually pays
        * attention to the wrap_t value, though it should not.
        * Override the wrap_t value here to GL_REPEAT to keep
@@ -237,18 +240,19 @@ static void brw_update_sampler_state(struct brw_context *brw,
 
    /* Set shadow function: 
     */
-   if (key->comparemode == GL_COMPARE_R_TO_TEXTURE_ARB) {
+   if (gl_sampler->CompareMode == GL_COMPARE_R_TO_TEXTURE_ARB) {
       /* Shadowing is "enabled" by emitting a particular sampler
        * message (sample_c).  So need to recompile WM program when
        * shadow comparison is enabled on each/any texture unit.
        */
       sampler->ss0.shadow_function =
-	 intel_translate_shadow_compare_func(key->comparefunc);
+	 intel_translate_shadow_compare_func(gl_sampler->CompareFunc);
    }
 
    /* Set LOD bias: 
     */
-   sampler->ss0.lod_bias = S_FIXED(CLAMP(key->lod_bias, -16, 15), 6);
+   sampler->ss0.lod_bias = S_FIXED(CLAMP(texUnit->LodBias +
+					 gl_sampler->LodBias, -16, 15), 6);
 
    sampler->ss0.lod_preclamp = 1; /* OpenGL mode */
    sampler->ss0.default_color_mode = 0; /* OpenGL/DX10 mode */
@@ -262,151 +266,63 @@ static void brw_update_sampler_state(struct brw_context *brw,
     */
    sampler->ss0.base_level = U_FIXED(0, 1);
 
-   sampler->ss1.max_lod = U_FIXED(CLAMP(key->maxlod, 0, 13), 6);
-   sampler->ss1.min_lod = U_FIXED(CLAMP(key->minlod, 0, 13), 6);
-   
-   sampler->ss2.default_color_pointer = sdc_bo->offset >> 5; /* reloc */
-}
+   sampler->ss1.max_lod = U_FIXED(CLAMP(gl_sampler->MaxLod, 0, 13), 6);
+   sampler->ss1.min_lod = U_FIXED(CLAMP(gl_sampler->MinLod, 0, 13), 6);
 
+   upload_default_color(brw, gl_sampler, unit);
 
-/** Sets up the cache key for sampler state for all texture units */
-static void
-brw_wm_sampler_populate_key(struct brw_context *brw,
-			    struct wm_sampler_key *key)
-{
-   struct gl_context *ctx = &brw->intel.ctx;
-   int unit;
-   char *last_entry_end = ((char*)&key->sampler_count) + 
-      sizeof(key->sampler_count);
-
-   key->sampler_count = 0;
-
-   for (unit = 0; unit < BRW_MAX_TEX_UNIT; unit++) {
-      if (ctx->Texture.Unit[unit]._ReallyEnabled) {
-	 struct wm_sampler_entry *entry = &key->sampler[unit];
-	 struct gl_texture_unit *texUnit = &ctx->Texture.Unit[unit];
-	 struct gl_texture_object *texObj = texUnit->_Current;
-	 struct gl_texture_image *firstImage =
-	    texObj->Image[0][texObj->BaseLevel];
-	 struct gl_sampler_object *sampler = _mesa_get_samplerobj(ctx, unit);
-
-	 memset(last_entry_end, 0, 
-		(char*)entry - last_entry_end + sizeof(*entry));
-	 last_entry_end = ((char*)entry) + sizeof(*entry);
-
-         entry->tex_target = texObj->Target;
-
-	 entry->seamless_cube_map = (texObj->Target == GL_TEXTURE_CUBE_MAP)
-	    ? ctx->Texture.CubeMapSeamless : GL_FALSE;
-
-	 entry->wrap_r = sampler->WrapR;
-	 entry->wrap_s = sampler->WrapS;
-	 entry->wrap_t = sampler->WrapT;
-
-	 entry->maxlod = sampler->MaxLod;
-	 entry->minlod = sampler->MinLod;
-	 entry->lod_bias = texUnit->LodBias + sampler->LodBias;
-	 entry->max_aniso = sampler->MaxAnisotropy;
-	 entry->minfilter = sampler->MinFilter;
-	 entry->magfilter = sampler->MagFilter;
-	 entry->comparemode = sampler->CompareMode;
-         entry->comparefunc = sampler->CompareFunc;
-
-	 drm_intel_bo_unreference(brw->wm.sdc_bo[unit]);
-	 if (firstImage->_BaseFormat == GL_DEPTH_COMPONENT) {
-	    float bordercolor[4] = {
-	       sampler->BorderColor.f[0],
-	       sampler->BorderColor.f[0],
-	       sampler->BorderColor.f[0],
-	       sampler->BorderColor.f[0]
-	    };
-	    /* GL specs that border color for depth textures is taken from the
-	     * R channel, while the hardware uses A.  Spam R into all the
-	     * channels for safety.
-	     */
-	    brw->wm.sdc_bo[unit] = upload_default_color(brw, bordercolor);
-	 } else {
-	    brw->wm.sdc_bo[unit] = upload_default_color(brw,
-							sampler->BorderColor.f);
-	 }
-	 key->sampler_count = unit + 1;
-      }
-   }
-   struct wm_sampler_entry *entry = &key->sampler[key->sampler_count];
-   memset(last_entry_end, 0, (char*)entry - last_entry_end);
+   /* reloc */
+   sampler->ss2.default_color_pointer = (intel->batch.bo->offset +
+					 brw->wm.sdc_offset[unit]) >> 5;
+
+   drm_intel_bo_emit_reloc(intel->batch.bo,
+			   brw->wm.sampler_offset +
+			   unit * sizeof(struct brw_sampler_state) +
+			   offsetof(struct brw_sampler_state, ss2),
+			   intel->batch.bo, brw->wm.sdc_offset[unit],
+			   I915_GEM_DOMAIN_SAMPLER, 0);
 }
 
+
 /* All samplers must be uploaded in a single contiguous array, which
  * complicates various things.  However, this is still too confusing -
  * FIXME: simplify all the different new texture state flags.
  */
-static void upload_wm_samplers( struct brw_context *brw )
+static void
+prepare_wm_samplers(struct brw_context *brw)
 {
    struct gl_context *ctx = &brw->intel.ctx;
-   struct wm_sampler_key key;
-   int i, sampler_key_size;
+   struct brw_sampler_state *samplers;
+   int i;
 
-   brw_wm_sampler_populate_key(brw, &key);
-
-   if (brw->wm.sampler_count != key.sampler_count) {
-      brw->wm.sampler_count = key.sampler_count;
-      brw->state.dirty.cache |= CACHE_NEW_SAMPLER;
+   brw->wm.sampler_count = 0;
+   for (i = 0; i < BRW_MAX_TEX_UNIT; i++) {
+      if (ctx->Texture.Unit[i]._ReallyEnabled)
+	 brw->wm.sampler_count = i + 1;
    }
 
-   drm_intel_bo_unreference(brw->wm.sampler_bo);
-   brw->wm.sampler_bo = NULL;
    if (brw->wm.sampler_count == 0)
       return;
 
-   /* Only include the populated portion of the key in the search. */
-   sampler_key_size = offsetof(struct wm_sampler_key,
-			       sampler[key.sampler_count]);
-   brw->wm.sampler_bo = brw_search_cache(&brw->cache, BRW_SAMPLER,
-					 &key, sampler_key_size,
-					 brw->wm.sdc_bo, key.sampler_count,
-					 NULL);
-
-   /* If we didnt find it in the cache, compute the state and put it in the
-    * cache.
-    */
-   if (brw->wm.sampler_bo == NULL) {
-      struct brw_sampler_state sampler[BRW_MAX_TEX_UNIT];
-
-      memset(sampler, 0, sizeof(sampler));
-      for (i = 0; i < key.sampler_count; i++) {
-	 if (brw->wm.sdc_bo[i] == NULL)
-	    continue;
+   samplers = brw_state_batch(brw, brw->wm.sampler_count * sizeof(*samplers),
+			      32, &brw->wm.sampler_offset);
+   memset(samplers, 0, brw->wm.sampler_count * sizeof(*samplers));
 
-	 brw_update_sampler_state(brw, &key.sampler[i], brw->wm.sdc_bo[i],
-				  &sampler[i]);
-      }
-
-      brw->wm.sampler_bo = brw_upload_cache(&brw->cache, BRW_SAMPLER,
-					    &key, sampler_key_size,
-					    brw->wm.sdc_bo, key.sampler_count,
-					    &sampler, sizeof(sampler));
-
-      /* Emit SDC relocations */
-      for (i = 0; i < BRW_MAX_TEX_UNIT; i++) {
-	 if (!ctx->Texture.Unit[i]._ReallyEnabled)
-	    continue;
-
-	 drm_intel_bo_emit_reloc(brw->wm.sampler_bo,
-				 i * sizeof(struct brw_sampler_state) +
-				 offsetof(struct brw_sampler_state, ss2),
-				 brw->wm.sdc_bo[i], 0,
-				 I915_GEM_DOMAIN_SAMPLER, 0);
-      }
+   for (i = 0; i < brw->wm.sampler_count; i++) {
+      if (ctx->Texture.Unit[i]._ReallyEnabled)
+	 brw_update_sampler_state(brw, i, &samplers[i]);
    }
+
+   brw->state.dirty.cache |= CACHE_NEW_SAMPLER;
 }
 
 const struct brw_tracked_state brw_wm_samplers = {
    .dirty = {
       .mesa = _NEW_TEXTURE,
-      .brw = 0,
+      .brw = BRW_NEW_BATCH,
       .cache = 0
    },
-   .prepare = upload_wm_samplers,
+   .prepare = prepare_wm_samplers,
 };
 
 
diff --git a/src/mesa/drivers/dri/i965/brw_wm_state.c b/src/mesa/drivers/dri/i965/brw_wm_state.c
index d7faf490cfa..a91ae511b7f 100644
--- a/src/mesa/drivers/dri/i965/brw_wm_state.c
+++ b/src/mesa/drivers/dri/i965/brw_wm_state.c
@@ -131,9 +131,10 @@ brw_prepare_wm_unit(struct brw_context *brw)
       wm->wm4.sampler_count = (brw->wm.sampler_count + 1) / 4;
    }
 
-   if (brw->wm.sampler_bo != NULL) {
+   if (brw->wm.sampler_count) {
       /* reloc */
-      wm->wm4.sampler_state_pointer = brw->wm.sampler_bo->offset >> 5;
+      wm->wm4.sampler_state_pointer = (intel->batch.bo->offset +
+				       brw->wm.sampler_offset) >> 5;
    } else {
       wm->wm4.sampler_state_pointer = 0;
    }
@@ -244,8 +245,9 @@ brw_prepare_wm_unit(struct brw_context *brw)
       drm_intel_bo_emit_reloc(intel->batch.bo,
 			      brw->wm.state_offset +
 			      offsetof(struct brw_wm_unit_state, wm4),
-			      brw->wm.sampler_bo, (wm->wm4.stats_enable |
-						   (wm->wm4.sampler_count << 2)),
+			      intel->batch.bo, (brw->wm.sampler_offset |
+						wm->wm4.stats_enable |
+						(wm->wm4.sampler_count << 2)),
 			      I915_GEM_DOMAIN_INSTRUCTION, 0);
    }
 
diff --git a/src/mesa/drivers/dri/i965/gen6_sampler_state.c b/src/mesa/drivers/dri/i965/gen6_sampler_state.c
index f65c651bdff..969780ac1df 100644
--- a/src/mesa/drivers/dri/i965/gen6_sampler_state.c
+++ b/src/mesa/drivers/dri/i965/gen6_sampler_state.c
@@ -43,27 +43,20 @@ upload_sampler_state_pointers(struct brw_context *brw)
 	     (4 - 2));
    OUT_BATCH(0); /* VS */
    OUT_BATCH(0); /* GS */
-   if (brw->wm.sampler_bo)
-      OUT_RELOC(brw->wm.sampler_bo, I915_GEM_DOMAIN_INSTRUCTION, 0, 0);
+   if (brw->wm.sampler_count)
+      OUT_RELOC(intel->batch.bo, I915_GEM_DOMAIN_INSTRUCTION, 0,
+		brw->wm.sampler_offset);
    else
       OUT_BATCH(0);
 
    ADVANCE_BATCH();
 }
 
-
-static void
-prepare_sampler_state_pointers(struct brw_context *brw)
-{
-   brw_add_validated_bo(brw, brw->wm.sampler_bo);
-}
-
 const struct brw_tracked_state gen6_sampler_state = {
    .dirty = {
       .mesa = 0,
       .brw = BRW_NEW_BATCH,
       .cache = CACHE_NEW_SAMPLER
    },
-   .prepare = prepare_sampler_state_pointers,
    .emit = upload_sampler_state_pointers,
 };