i965/gen6 depth surface: calculate depth (array size) for depth surface
[mesa.git] / src / mesa / drivers / dri / i965 / gen6_blorp.cpp
index 13dbd30d71f1ec6230f667bf0a92e23fa1953029..0b1d0977ccddab099e54a79a051fd31bb374c4a5 100644 (file)
                              * sizeof(float))
 /** \} */
 
-void
-gen6_blorp_emit_batch_head(struct brw_context *brw,
-                           const brw_blorp_params *params)
-{
-   struct gl_context *ctx = &brw->intel.ctx;
-   struct intel_context *intel = &brw->intel;
-
-   /* To ensure that the batch contains only the resolve, flush the batch
-    * before beginning and after finishing emitting the resolve packets.
-    *
-    * Ideally, we would not need to flush for the resolve op. But, I suspect
-    * that it's unsafe for CMD_PIPELINE_SELECT to occur multiple times in
-    * a single batch, and there is no safe way to ensure that other than by
-    * fencing the resolve with flushes. Ideally, we would just detect if
-    * a batch is in progress and do the right thing, but that would require
-    * the ability to *safely* access brw_context::state::dirty::brw
-    * outside of the brw_upload_state() codepath.
-    */
-   intel_flush(ctx);
-
-   /* CMD_PIPELINE_SELECT
-    *
-    * Select the 3D pipeline, as opposed to the media pipeline.
-    */
-   {
-      BEGIN_BATCH(1);
-      OUT_BATCH(brw->CMD_PIPELINE_SELECT << 16);
-      ADVANCE_BATCH();
-   }
-}
-
-
 /**
  * CMD_STATE_BASE_ADDRESS
  *
@@ -93,15 +61,18 @@ void
 gen6_blorp_emit_state_base_address(struct brw_context *brw,
                                    const brw_blorp_params *params)
 {
-   struct intel_context *intel = &brw->intel;
+   uint8_t mocs = brw->gen == 7 ? GEN7_MOCS_L3 : 0;
 
    BEGIN_BATCH(10);
    OUT_BATCH(CMD_STATE_BASE_ADDRESS << 16 | (10 - 2));
-   OUT_BATCH(1); /* GeneralStateBaseAddressModifyEnable */
+   OUT_BATCH(mocs << 8 | /* GeneralStateMemoryObjectControlState */
+             mocs << 4 | /* StatelessDataPortAccessMemoryObjectControlState */
+             1); /* GeneralStateBaseAddressModifyEnable */
+
    /* SurfaceStateBaseAddress */
-   OUT_RELOC(intel->batch.bo, I915_GEM_DOMAIN_SAMPLER, 0, 1);
+   OUT_RELOC(brw->batch.bo, I915_GEM_DOMAIN_SAMPLER, 0, 1);
    /* DynamicStateBaseAddress */
-   OUT_RELOC(intel->batch.bo, (I915_GEM_DOMAIN_RENDER |
+   OUT_RELOC(brw->batch.bo, (I915_GEM_DOMAIN_RENDER |
                                I915_GEM_DOMAIN_INSTRUCTION), 0, 1);
    OUT_BATCH(1); /* IndirectObjectBaseAddress */
    if (params->use_wm_prog) {
@@ -127,7 +98,6 @@ void
 gen6_blorp_emit_vertices(struct brw_context *brw,
                          const brw_blorp_params *params)
 {
-   struct intel_context *intel = &brw->intel;
    uint32_t vertex_offset;
 
    /* Setup VBO for the rectangle primitive..
@@ -182,17 +152,20 @@ gen6_blorp_emit_vertices(struct brw_context *brw,
       uint32_t dw0 = GEN6_VB0_ACCESS_VERTEXDATA |
                      (GEN6_BLORP_NUM_VUE_ELEMS * sizeof(float)) << BRW_VB0_PITCH_SHIFT;
 
-      if (intel->gen >= 7)
+      if (brw->gen >= 7)
          dw0 |= GEN7_VB0_ADDRESS_MODIFYENABLE;
 
+      if (brw->gen == 7)
+         dw0 |= GEN7_MOCS_L3 << 16;
+
       BEGIN_BATCH(batch_length);
       OUT_BATCH((_3DSTATE_VERTEX_BUFFERS << 16) | (batch_length - 2));
       OUT_BATCH(dw0);
       /* start address */
-      OUT_RELOC(intel->batch.bo, I915_GEM_DOMAIN_VERTEX, 0,
+      OUT_RELOC(brw->batch.bo, I915_GEM_DOMAIN_VERTEX, 0,
                vertex_offset);
       /* end address */
-      OUT_RELOC(intel->batch.bo, I915_GEM_DOMAIN_VERTEX, 0,
+      OUT_RELOC(brw->batch.bo, I915_GEM_DOMAIN_VERTEX, 0,
                vertex_offset + GEN6_BLORP_VBO_SIZE - 1);
       OUT_BATCH(0);
       ADVANCE_BATCH();
@@ -250,8 +223,6 @@ static void
 gen6_blorp_emit_urb_config(struct brw_context *brw,
                            const brw_blorp_params *params)
 {
-   struct intel_context *intel = &brw->intel;
-
    BEGIN_BATCH(3);
    OUT_BATCH(_3DSTATE_URB << 16 | (3 - 2));
    OUT_BATCH(brw->urb.max_vs_entries << GEN6_URB_VS_ENTRIES_SHIFT);
@@ -278,10 +249,10 @@ gen6_blorp_emit_blend_state(struct brw_context *brw,
    blend->blend1.post_blend_clamp_enable = 1;
    blend->blend1.clamp_range = BRW_RENDERTARGET_CLAMPRANGE_FORMAT;
 
-   blend->blend1.write_disable_r = false;
-   blend->blend1.write_disable_g = false;
-   blend->blend1.write_disable_b = false;
-   blend->blend1.write_disable_a = false;
+   blend->blend1.write_disable_r = params->color_write_disable[0];
+   blend->blend1.write_disable_g = params->color_write_disable[1];
+   blend->blend1.write_disable_b = params->color_write_disable[2];
+   blend->blend1.write_disable_a = params->color_write_disable[3];
 
    return cc_blend_state_offset;
 }
@@ -329,7 +300,7 @@ gen6_blorp_emit_depth_stencil_state(struct brw_context *brw,
    state->ds2.depth_write_enable = 1;
    if (params->hiz_op == GEN6_HIZ_OP_DEPTH_RESOLVE) {
       state->ds2.depth_test_enable = 1;
-      state->ds2.depth_test_func = COMPAREFUNC_NEVER;
+      state->ds2.depth_test_func = BRW_COMPAREFUNCTION_NEVER;
    }
 
    return depthstencil_offset;
@@ -350,8 +321,6 @@ gen6_blorp_emit_cc_state_pointers(struct brw_context *brw,
                                   uint32_t depthstencil_offset,
                                   uint32_t cc_state_offset)
 {
-   struct intel_context *intel = &brw->intel;
-
    BEGIN_BATCH(4);
    OUT_BATCH(_3DSTATE_CC_STATE_POINTERS << 16 | (4 - 2));
    OUT_BATCH(cc_blend_state_offset | 1); /* BLEND_STATE offset */
@@ -398,7 +367,7 @@ gen6_blorp_emit_surface_state(struct brw_context *brw,
       width /= 2;
       height /= 2;
    }
-   struct intel_region *region = surface->mt->region;
+   struct intel_mipmap_tree *mt = surface->mt;
    uint32_t tile_x, tile_y;
 
    uint32_t *surf = (uint32_t *)
@@ -412,7 +381,7 @@ gen6_blorp_emit_surface_state(struct brw_context *brw,
 
    /* reloc */
    surf[1] = (surface->compute_tile_offsets(&tile_x, &tile_y) +
-              region->bo->offset);
+              mt->bo->offset64);
 
    surf[2] = (0 << BRW_SURFACE_LOD_SHIFT |
               (width - 1) << BRW_SURFACE_WIDTH_SHIFT |
@@ -420,8 +389,8 @@ gen6_blorp_emit_surface_state(struct brw_context *brw,
 
    uint32_t tiling = surface->map_stencil_as_y_tiled
       ? BRW_SURFACE_TILED | BRW_SURFACE_TILED_Y
-      : brw_get_surface_tiling_bits(region->tiling);
-   uint32_t pitch_bytes = region->pitch * region->cpp;
+      : brw_get_surface_tiling_bits(mt->tiling);
+   uint32_t pitch_bytes = mt->pitch;
    if (surface->map_stencil_as_y_tiled)
       pitch_bytes *= 2;
    surf[3] = (tiling |
@@ -441,10 +410,10 @@ gen6_blorp_emit_surface_state(struct brw_context *brw,
                BRW_SURFACE_VERTICAL_ALIGN_ENABLE : 0));
 
    /* Emit relocation to surface contents */
-   drm_intel_bo_emit_reloc(brw->intel.batch.bo,
+   drm_intel_bo_emit_reloc(brw->batch.bo,
                            wm_surf_offset + 4,
-                           region->bo,
-                           surf[1] - region->bo->offset,
+                           mt->bo,
+                           surf[1] - mt->bo->offset64,
                            read_domains, write_domain);
 
    return wm_surf_offset;
@@ -476,55 +445,44 @@ gen6_blorp_emit_binding_table(struct brw_context *brw,
 /**
  * SAMPLER_STATE.  See brw_update_sampler_state().
  */
-static uint32_t
+uint32_t
 gen6_blorp_emit_sampler_state(struct brw_context *brw,
                               const brw_blorp_params *params)
 {
    uint32_t sampler_offset;
+   uint32_t *sampler_state = (uint32_t *)
+      brw_state_batch(brw, AUB_TRACE_SAMPLER_STATE, 16, 32, &sampler_offset);
 
-   struct brw_sampler_state *sampler = (struct brw_sampler_state *)
-      brw_state_batch(brw, AUB_TRACE_SAMPLER_STATE,
-                      sizeof(struct brw_sampler_state),
-                      32, &sampler_offset);
-   memset(sampler, 0, sizeof(*sampler));
-
-   sampler->ss0.min_filter = BRW_MAPFILTER_LINEAR;
-   sampler->ss0.mip_filter = BRW_MIPFILTER_NONE;
-   sampler->ss0.mag_filter = BRW_MAPFILTER_LINEAR;
+   unsigned address_rounding = BRW_ADDRESS_ROUNDING_ENABLE_U_MIN |
+                               BRW_ADDRESS_ROUNDING_ENABLE_V_MIN |
+                               BRW_ADDRESS_ROUNDING_ENABLE_R_MIN |
+                               BRW_ADDRESS_ROUNDING_ENABLE_U_MAG |
+                               BRW_ADDRESS_ROUNDING_ENABLE_V_MAG |
+                               BRW_ADDRESS_ROUNDING_ENABLE_R_MAG;
 
-   sampler->ss1.r_wrap_mode = BRW_TEXCOORDMODE_CLAMP;
-   sampler->ss1.s_wrap_mode = BRW_TEXCOORDMODE_CLAMP;
-   sampler->ss1.t_wrap_mode = BRW_TEXCOORDMODE_CLAMP;
-
-   sampler->ss0.min_mag_neq = 1;
-
-   /* Set LOD bias: 
-    */
-   sampler->ss0.lod_bias = 0;
-
-   sampler->ss0.lod_preclamp = 1; /* OpenGL mode */
-   sampler->ss0.default_color_mode = 0; /* OpenGL/DX10 mode */
-
-   /* Set BaseMipLevel, MaxLOD, MinLOD: 
-    *
-    * XXX: I don't think that using firstLevel, lastLevel works,
+   /* XXX: I don't think that using firstLevel, lastLevel works,
     * because we always setup the surface state as if firstLevel ==
     * level zero.  Probably have to subtract firstLevel from each of
     * these:
     */
-   sampler->ss0.base_level = U_FIXED(0, 1);
-
-   sampler->ss1.max_lod = U_FIXED(0, 6);
-   sampler->ss1.min_lod = U_FIXED(0, 6);
-
-   sampler->ss3.non_normalized_coord = 1;
-
-   sampler->ss3.address_round |= BRW_ADDRESS_ROUNDING_ENABLE_U_MIN |
-      BRW_ADDRESS_ROUNDING_ENABLE_V_MIN |
-      BRW_ADDRESS_ROUNDING_ENABLE_R_MIN;
-   sampler->ss3.address_round |= BRW_ADDRESS_ROUNDING_ENABLE_U_MAG |
-      BRW_ADDRESS_ROUNDING_ENABLE_V_MAG |
-      BRW_ADDRESS_ROUNDING_ENABLE_R_MAG;
+   brw_emit_sampler_state(brw,
+                          sampler_state,
+                          sampler_offset,
+                          BRW_MAPFILTER_LINEAR, /* min filter */
+                          BRW_MAPFILTER_LINEAR, /* mag filter */
+                          BRW_MIPFILTER_NONE,
+                          BRW_ANISORATIO_2,
+                          address_rounding,
+                          BRW_TEXCOORDMODE_CLAMP,
+                          BRW_TEXCOORDMODE_CLAMP,
+                          BRW_TEXCOORDMODE_CLAMP,
+                          0, /* min LOD */
+                          0, /* max LOD */
+                          0, /* LOD bias */
+                          0, /* base miplevel */
+                          0, /* shadow function */
+                          true, /* non-normalized coordinates */
+                          0); /* border color offset - unused */
 
    return sampler_offset;
 }
@@ -538,8 +496,6 @@ gen6_blorp_emit_sampler_state_pointers(struct brw_context *brw,
                                        const brw_blorp_params *params,
                                        uint32_t sampler_offset)
 {
-   struct intel_context *intel = &brw->intel;
-
    BEGIN_BATCH(4);
    OUT_BATCH(_3DSTATE_SAMPLER_STATE_POINTERS << 16 |
              VS_SAMPLER_STATE_CHANGE |
@@ -561,10 +517,8 @@ void
 gen6_blorp_emit_vs_disable(struct brw_context *brw,
                            const brw_blorp_params *params)
 {
-   struct intel_context *intel = &brw->intel;
-
-   if (intel->gen == 6) {
-      /* From the BSpec, Volume 2a, Part 3 "Vertex Shader", Section
+   if (brw->gen == 6) {
+      /* From the BSpec, 3D Pipeline > Geometry > Vertex Shader > State,
        * 3DSTATE_VS, Dword 5.0 "VS Function Enable":
        *
        *   [DevSNB] A pipeline flush must be programmed prior to a
@@ -572,9 +526,18 @@ gen6_blorp_emit_vs_disable(struct brw_context *brw,
        *   toggle. Pipeline flush can be executed by sending a PIPE_CONTROL
        *   command with CS stall bit set and a post sync operation.
        */
-      intel_emit_post_sync_nonzero_flush(intel);
+      intel_emit_post_sync_nonzero_flush(brw);
    }
 
+   /* Disable the push constant buffers. */
+   BEGIN_BATCH(5);
+   OUT_BATCH(_3DSTATE_CONSTANT_VS << 16 | (5 - 2));
+   OUT_BATCH(0);
+   OUT_BATCH(0);
+   OUT_BATCH(0);
+   OUT_BATCH(0);
+   ADVANCE_BATCH();
+
    BEGIN_BATCH(6);
    OUT_BATCH(_3DSTATE_VS << 16 | (6 - 2));
    OUT_BATCH(0);
@@ -594,7 +557,14 @@ void
 gen6_blorp_emit_gs_disable(struct brw_context *brw,
                            const brw_blorp_params *params)
 {
-   struct intel_context *intel = &brw->intel;
+   /* Disable all the constant buffers. */
+   BEGIN_BATCH(5);
+   OUT_BATCH(_3DSTATE_CONSTANT_GS << 16 | (5 - 2));
+   OUT_BATCH(0);
+   OUT_BATCH(0);
+   OUT_BATCH(0);
+   OUT_BATCH(0);
+   ADVANCE_BATCH();
 
    BEGIN_BATCH(7);
    OUT_BATCH(_3DSTATE_GS << 16 | (7 - 2));
@@ -626,8 +596,6 @@ void
 gen6_blorp_emit_clip_disable(struct brw_context *brw,
                              const brw_blorp_params *params)
 {
-   struct intel_context *intel = &brw->intel;
-
    BEGIN_BATCH(4);
    OUT_BATCH(_3DSTATE_CLIP << 16 | (4 - 2));
    OUT_BATCH(0);
@@ -659,15 +627,13 @@ static void
 gen6_blorp_emit_sf_config(struct brw_context *brw,
                           const brw_blorp_params *params)
 {
-   struct intel_context *intel = &brw->intel;
-
    BEGIN_BATCH(20);
    OUT_BATCH(_3DSTATE_SF << 16 | (20 - 2));
    OUT_BATCH((1 - 1) << GEN6_SF_NUM_OUTPUTS_SHIFT | /* only position */
              1 << GEN6_SF_URB_ENTRY_READ_LENGTH_SHIFT |
              0 << GEN6_SF_URB_ENTRY_READ_OFFSET_SHIFT);
    OUT_BATCH(0); /* dw2 */
-   OUT_BATCH(params->num_samples > 1 ? GEN6_SF_MSRAST_ON_PATTERN : 0);
+   OUT_BATCH(params->dst.num_samples > 1 ? GEN6_SF_MSRAST_ON_PATTERN : 0);
    for (int i = 0; i < 16; ++i)
       OUT_BATCH(0);
    ADVANCE_BATCH();
@@ -683,13 +649,12 @@ gen6_blorp_emit_wm_config(struct brw_context *brw,
                           uint32_t prog_offset,
                           brw_blorp_prog_data *prog_data)
 {
-   struct intel_context *intel = &brw->intel;
    uint32_t dw2, dw4, dw5, dw6;
 
    /* Even when thread dispatch is disabled, max threads (dw5.25:31) must be
-    * nonzero to prevent the GPU from hanging. See the valid ranges in the
-    * BSpec, Volume 2a.11 Windower, Section 3DSTATE_WM, Dword 5.25:31
-    * "Maximum Number Of Threads".
+    * nonzero to prevent the GPU from hanging.  While the documentation doesn't
+    * mention this explicitly, it notes that the valid range for the field is
+    * [1,39] = [2,40] threads, which excludes zero.
     *
     * To be safe (and to minimize extraneous code) we go ahead and fully
     * configure the WM state whether or not there is a WM program.
@@ -709,10 +674,8 @@ gen6_blorp_emit_wm_config(struct brw_context *brw,
    case GEN6_HIZ_OP_NONE:
       break;
    default:
-      assert(0);
-      break;
+      unreachable("not reached");
    }
-   dw4 |= GEN6_WM_STATISTICS_ENABLE;
    dw5 |= GEN6_WM_LINE_AA_WIDTH_1_0;
    dw5 |= GEN6_WM_LINE_END_CAP_AA_WIDTH_0_5;
    dw5 |= (brw->max_wm_threads - 1) << GEN6_WM_MAX_THREADS_SHIFT;
@@ -726,7 +689,7 @@ gen6_blorp_emit_wm_config(struct brw_context *brw,
       dw5 |= GEN6_WM_DISPATCH_ENABLE; /* We are rendering */
    }
 
-   if (params->num_samples > 1) {
+   if (params->dst.num_samples > 1) {
       dw6 |= GEN6_WM_MSRAST_ON_PATTERN;
       if (prog_data && prog_data->persample_msaa_dispatch)
          dw6 |= GEN6_WM_MSDISPMODE_PERSAMPLE;
@@ -756,8 +719,6 @@ gen6_blorp_emit_constant_ps(struct brw_context *brw,
                             const brw_blorp_params *params,
                             uint32_t wm_push_const_offset)
 {
-   struct intel_context *intel = &brw->intel;
-
    /* Make sure the push constants fill an exact integer number of
     * registers.
     */
@@ -778,6 +739,19 @@ gen6_blorp_emit_constant_ps(struct brw_context *brw,
    ADVANCE_BATCH();
 }
 
+static void
+gen6_blorp_emit_constant_ps_disable(struct brw_context *brw,
+                                    const brw_blorp_params *params)
+{
+   /* Disable the push constant buffers. */
+   BEGIN_BATCH(5);
+   OUT_BATCH(_3DSTATE_CONSTANT_PS << 16 | (5 - 2));
+   OUT_BATCH(0);
+   OUT_BATCH(0);
+   OUT_BATCH(0);
+   OUT_BATCH(0);
+   ADVANCE_BATCH();
+}
 
 /**
  * 3DSTATE_BINDING_TABLE_POINTERS
@@ -787,8 +761,6 @@ gen6_blorp_emit_binding_table_pointers(struct brw_context *brw,
                                        const brw_blorp_params *params,
                                        uint32_t wm_bind_bo_offset)
 {
-   struct intel_context *intel = &brw->intel;
-
    BEGIN_BATCH(4);
    OUT_BATCH(_3DSTATE_BINDING_TABLE_POINTERS << 16 |
              GEN6_BINDING_TABLE_MODIFY_PS |
@@ -804,12 +776,34 @@ static void
 gen6_blorp_emit_depth_stencil_config(struct brw_context *brw,
                                      const brw_blorp_params *params)
 {
-   struct intel_context *intel = &brw->intel;
+   struct gl_context *ctx = &brw->ctx;
    uint32_t draw_x = params->depth.x_offset;
    uint32_t draw_y = params->depth.y_offset;
    uint32_t tile_mask_x, tile_mask_y;
+   uint32_t surftype;
+   unsigned int depth = MAX2(params->depth.mt->logical_depth0, 1);
+   GLenum gl_target = params->depth.mt->target;
+
+   switch (gl_target) {
+   case GL_TEXTURE_CUBE_MAP_ARRAY:
+   case GL_TEXTURE_CUBE_MAP:
+      /* The PRM claims that we should use BRW_SURFACE_CUBE for this
+       * situation, but experiments show that gl_Layer doesn't work when we do
+       * this.  So we use BRW_SURFACE_2D, since for rendering purposes this is
+       * equivalent.
+       */
+      surftype = BRW_SURFACE_2D;
+      depth *= 6;
+      break;
+   default:
+      surftype = translate_tex_target(gl_target);
+      break;
+   }
 
-   brw_get_depthstencil_tile_masks(params->depth.mt, NULL,
+   brw_get_depthstencil_tile_masks(params->depth.mt,
+                                   params->depth.level,
+                                   params->depth.layer,
+                                   NULL,
                                    &tile_mask_x, &tile_mask_y);
 
    /* 3DSTATE_DEPTH_BUFFER */
@@ -817,9 +811,9 @@ gen6_blorp_emit_depth_stencil_config(struct brw_context *brw,
       uint32_t tile_x = draw_x & tile_mask_x;
       uint32_t tile_y = draw_y & tile_mask_y;
       uint32_t offset =
-         intel_region_get_aligned_offset(params->depth.mt->region,
-                                         draw_x & ~tile_mask_x,
-                                         draw_y & ~tile_mask_y, false);
+         intel_miptree_get_aligned_offset(params->depth.mt,
+                                          draw_x & ~tile_mask_x,
+                                          draw_y & ~tile_mask_y, false);
 
       /* According to the Sandy Bridge PRM, volume 2 part 1, pp326-327
        * (3DSTATE_DEPTH_BUFFER dw5), in the documentation for "Depth
@@ -836,24 +830,25 @@ gen6_blorp_emit_depth_stencil_config(struct brw_context *brw,
        * tile_x and tile_y to 0.  This is a temporary workaround until we come
        * up with a better solution.
        */
+      WARN_ONCE((tile_x & 7) || (tile_y & 7),
+                "Depth/stencil buffer needs alignment to 8-pixel boundaries.\n"
+                "Truncating offset, bad rendering may occur.\n");
       tile_x &= ~7;
       tile_y &= ~7;
 
-      intel_emit_post_sync_nonzero_flush(intel);
-      intel_emit_depth_stall_flushes(intel);
+      intel_emit_post_sync_nonzero_flush(brw);
+      intel_emit_depth_stall_flushes(brw);
 
       BEGIN_BATCH(7);
       OUT_BATCH(_3DSTATE_DEPTH_BUFFER << 16 | (7 - 2));
-      uint32_t pitch_bytes =
-         params->depth.mt->region->pitch * params->depth.mt->region->cpp;
-      OUT_BATCH((pitch_bytes - 1) |
+      OUT_BATCH((params->depth.mt->pitch - 1) |
                 params->depth_format << 18 |
                 1 << 21 | /* separate stencil enable */
                 1 << 22 | /* hiz enable */
                 BRW_TILEWALK_YMAJOR << 26 |
                 1 << 27 | /* y-tiled */
                 BRW_SURFACE_2D << 29);
-      OUT_RELOC(params->depth.mt->region->bo,
+      OUT_RELOC(params->depth.mt->bo,
                 I915_GEM_DOMAIN_RENDER, I915_GEM_DOMAIN_RENDER,
                 offset);
       OUT_BATCH(BRW_SURFACE_MIPMAPLAYOUT_BELOW << 1 |
@@ -868,16 +863,16 @@ gen6_blorp_emit_depth_stencil_config(struct brw_context *brw,
 
    /* 3DSTATE_HIER_DEPTH_BUFFER */
    {
-      struct intel_region *hiz_region = params->depth.mt->hiz_mt->region;
+      struct intel_mipmap_tree *hiz_mt = params->depth.mt->hiz_mt;
       uint32_t hiz_offset =
-         intel_region_get_aligned_offset(hiz_region,
-                                         draw_x & ~tile_mask_x,
-                                         (draw_y & ~tile_mask_y) / 2, false);
+         intel_miptree_get_aligned_offset(hiz_mt,
+                                          draw_x & ~tile_mask_x,
+                                          (draw_y & ~tile_mask_y) / 2, false);
 
       BEGIN_BATCH(3);
       OUT_BATCH((_3DSTATE_HIER_DEPTH_BUFFER << 16) | (3 - 2));
-      OUT_BATCH(hiz_region->pitch * hiz_region->cpp - 1);
-      OUT_RELOC(hiz_region->bo,
+      OUT_BATCH(hiz_mt->pitch - 1);
+      OUT_RELOC(hiz_mt->bo,
                 I915_GEM_DOMAIN_RENDER, I915_GEM_DOMAIN_RENDER,
                 hiz_offset);
       ADVANCE_BATCH();
@@ -898,7 +893,8 @@ static void
 gen6_blorp_emit_depth_disable(struct brw_context *brw,
                               const brw_blorp_params *params)
 {
-   struct intel_context *intel = &brw->intel;
+   intel_emit_post_sync_nonzero_flush(brw);
+   intel_emit_depth_stall_flushes(brw);
 
    BEGIN_BATCH(7);
    OUT_BATCH(_3DSTATE_DEPTH_BUFFER << 16 | (7 - 2));
@@ -910,6 +906,18 @@ gen6_blorp_emit_depth_disable(struct brw_context *brw,
    OUT_BATCH(0);
    OUT_BATCH(0);
    ADVANCE_BATCH();
+
+   BEGIN_BATCH(3);
+   OUT_BATCH(_3DSTATE_HIER_DEPTH_BUFFER << 16 | (3 - 2));
+   OUT_BATCH(0);
+   OUT_BATCH(0);
+   ADVANCE_BATCH();
+
+   BEGIN_BATCH(3);
+   OUT_BATCH(_3DSTATE_STENCIL_BUFFER << 16 | (3 - 2));
+   OUT_BATCH(0);
+   OUT_BATCH(0);
+   ADVANCE_BATCH();
 }
 
 
@@ -923,8 +931,6 @@ static void
 gen6_blorp_emit_clear_params(struct brw_context *brw,
                              const brw_blorp_params *params)
 {
-   struct intel_context *intel = &brw->intel;
-
    BEGIN_BATCH(2);
    OUT_BATCH(_3DSTATE_CLEAR_PARAMS << 16 |
             GEN5_DEPTH_CLEAR_VALID |
@@ -939,7 +945,8 @@ void
 gen6_blorp_emit_drawing_rectangle(struct brw_context *brw,
                                   const brw_blorp_params *params)
 {
-   struct intel_context *intel = &brw->intel;
+   if (brw->gen == 6)
+      intel_emit_post_sync_nonzero_flush(brw);
 
    BEGIN_BATCH(4);
    OUT_BATCH(_3DSTATE_DRAWING_RECTANGLE << 16 | (4 - 2));
@@ -955,7 +962,6 @@ void
 gen6_blorp_emit_viewport_state(struct brw_context *brw,
                               const brw_blorp_params *params)
 {
-   struct intel_context *intel = &brw->intel;
    struct brw_cc_viewport *ccv;
    uint32_t cc_vp_offset;
 
@@ -981,8 +987,6 @@ static void
 gen6_blorp_emit_primitive(struct brw_context *brw,
                           const brw_blorp_params *params)
 {
-   struct intel_context *intel = &brw->intel;
-
    BEGIN_BATCH(6);
    OUT_BATCH(CMD_3D_PRIM << 16 | (6 - 2) |
              _3DPRIM_RECTLIST << GEN4_3DPRIM_TOPOLOGY_TYPE_SHIFT |
@@ -993,8 +997,10 @@ gen6_blorp_emit_primitive(struct brw_context *brw,
    OUT_BATCH(0);
    OUT_BATCH(0);
    ADVANCE_BATCH();
-}
 
+   /* Only used on Sandybridge; harmless to set elsewhere. */
+   brw->batch.need_workaround_flush = true;
+}
 
 /**
  * \brief Execute a blit or render pass operation.
@@ -1006,11 +1012,9 @@ gen6_blorp_emit_primitive(struct brw_context *brw,
  * This function alters no GL state.
  */
 void
-gen6_blorp_exec(struct intel_context *intel,
+gen6_blorp_exec(struct brw_context *brw,
                 const brw_blorp_params *params)
 {
-   struct gl_context *ctx = &intel->ctx;
-   struct brw_context *brw = brw_context(ctx);
    brw_blorp_prog_data *prog_data = NULL;
    uint32_t cc_blend_state_offset = 0;
    uint32_t cc_state_offset = 0;
@@ -1019,9 +1023,14 @@ gen6_blorp_exec(struct intel_context *intel,
    uint32_t wm_bind_bo_offset = 0;
 
    uint32_t prog_offset = params->get_wm_prog(brw, &prog_data);
-   gen6_blorp_emit_batch_head(brw, params);
-   gen6_emit_3dstate_multisample(brw, params->num_samples);
-   gen6_emit_3dstate_sample_mask(brw, params->num_samples, 1.0, false);
+
+   /* Emit workaround flushes when we switch from drawing to blorping. */
+   brw->batch.need_workaround_flush = true;
+
+   gen6_emit_3dstate_multisample(brw, params->dst.num_samples);
+   gen6_emit_3dstate_sample_mask(brw,
+                                 params->dst.num_samples > 1 ?
+                                 (1 << params->dst.num_samples) - 1 : 1);
    gen6_blorp_emit_state_base_address(brw, params);
    gen6_blorp_emit_vertices(brw, params);
    gen6_blorp_emit_urb_config(brw, params);
@@ -1034,16 +1043,19 @@ gen6_blorp_exec(struct intel_context *intel,
                                      depthstencil_offset, cc_state_offset);
    if (params->use_wm_prog) {
       uint32_t wm_surf_offset_renderbuffer;
-      uint32_t wm_surf_offset_texture;
+      uint32_t wm_surf_offset_texture = 0;
       uint32_t sampler_offset;
       wm_push_const_offset = gen6_blorp_emit_wm_constants(brw, params);
+      intel_miptree_used_for_rendering(params->dst.mt);
       wm_surf_offset_renderbuffer =
          gen6_blorp_emit_surface_state(brw, params, &params->dst,
                                        I915_GEM_DOMAIN_RENDER,
                                        I915_GEM_DOMAIN_RENDER);
-      wm_surf_offset_texture =
-         gen6_blorp_emit_surface_state(brw, params, &params->src,
-                                       I915_GEM_DOMAIN_SAMPLER, 0);
+      if (params->src.mt) {
+         wm_surf_offset_texture =
+            gen6_blorp_emit_surface_state(brw, params, &params->src,
+                                          I915_GEM_DOMAIN_SAMPLER, 0);
+      }
       wm_bind_bo_offset =
          gen6_blorp_emit_binding_table(brw, params,
                                        wm_surf_offset_renderbuffer,
@@ -1057,6 +1069,8 @@ gen6_blorp_exec(struct intel_context *intel,
    gen6_blorp_emit_sf_config(brw, params);
    if (params->use_wm_prog)
       gen6_blorp_emit_constant_ps(brw, params, wm_push_const_offset);
+   else
+      gen6_blorp_emit_constant_ps_disable(brw, params);
    gen6_blorp_emit_wm_config(brw, params, prog_offset, prog_data);
    if (params->use_wm_prog)
       gen6_blorp_emit_binding_table_pointers(brw, params, wm_bind_bo_offset);
@@ -1069,14 +1083,5 @@ gen6_blorp_exec(struct intel_context *intel,
    gen6_blorp_emit_clear_params(brw, params);
    gen6_blorp_emit_drawing_rectangle(brw, params);
    gen6_blorp_emit_primitive(brw, params);
-
-   /* See comments above at first invocation of intel_flush() in
-    * gen6_blorp_emit_batch_head().
-    */
-   intel_flush(ctx);
-
-   /* Be safe. */
-   brw->state.dirty.brw = ~0;
-   brw->state.dirty.cache = ~0;
 }