OSDN Git Service

i965: Don't emit PIPELINE_SELECT from BLORP.
[android-x86/external-mesa.git] / src / mesa / drivers / dri / i965 / gen6_blorp.cpp
index 60c3ff1..bb87f6b 100644 (file)
                              * sizeof(float))
 /** \} */
 
-
-/**
- * Compute masks to determine how much of draw_x and draw_y should be
- * performed using the fine adjustment of "depth coordinate offset X/Y"
- * (dw5 of 3DSTATE_DEPTH_BUFFER).  See the emit_depthbuffer() function for
- * details.
- */
-void
-gen6_blorp_compute_tile_masks(const brw_blorp_params *params,
-                              uint32_t *tile_mask_x, uint32_t *tile_mask_y)
-{
-   uint32_t depth_mask_x, depth_mask_y, hiz_mask_x, hiz_mask_y;
-   intel_region_get_tile_masks(params->depth.mt->region,
-                               &depth_mask_x, &depth_mask_y, false);
-   intel_region_get_tile_masks(params->depth.mt->hiz_mt->region,
-                               &hiz_mask_x, &hiz_mask_y, false);
-
-   /* Each HiZ row represents 2 rows of pixels */
-   hiz_mask_y = hiz_mask_y << 1 | 1;
-
-   *tile_mask_x = depth_mask_x | hiz_mask_x;
-   *tile_mask_y = depth_mask_y | hiz_mask_y;
-}
-
-
 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();
-   }
 }
 
 
@@ -303,10 +259,30 @@ 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];
+
+   /* When blitting from an XRGB source to a ARGB destination, we need to
+    * interpret the missing channel as 1.0.  Blending can do that for us:
+    * we simply use the RGB values from the fragment shader ("source RGB"),
+    * but smash the alpha channel to 1.
+    */
+   if (params->src.mt &&
+       _mesa_get_format_bits(params->dst.mt->format, GL_ALPHA_BITS) > 0 &&
+       _mesa_get_format_bits(params->src.mt->format, GL_ALPHA_BITS) == 0) {
+      blend->blend0.blend_enable = 1;
+      blend->blend0.ia_blend_enable = 1;
+
+      blend->blend0.blend_func = BRW_BLENDFUNCTION_ADD;
+      blend->blend0.ia_blend_func = BRW_BLENDFUNCTION_ADD;
+
+      blend->blend0.source_blend_factor = BRW_BLENDFACTOR_SRC_COLOR;
+      blend->blend0.dest_blend_factor = BRW_BLENDFACTOR_ZERO;
+      blend->blend0.ia_source_blend_factor = BRW_BLENDFACTOR_ONE;
+      blend->blend0.ia_dest_blend_factor = BRW_BLENDFACTOR_ZERO;
+   }
 
    return cc_blend_state_offset;
 }
@@ -446,7 +422,7 @@ 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;
+   uint32_t pitch_bytes = region->pitch;
    if (surface->map_stencil_as_y_tiled)
       pitch_bytes *= 2;
    surf[3] = (tiling |
@@ -600,6 +576,15 @@ gen6_blorp_emit_vs_disable(struct brw_context *brw,
       intel_emit_post_sync_nonzero_flush(intel);
    }
 
+   /* 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);
@@ -621,6 +606,15 @@ gen6_blorp_emit_gs_disable(struct brw_context *brw,
 {
    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));
    OUT_BATCH(0);
@@ -737,7 +731,6 @@ gen6_blorp_emit_wm_config(struct brw_context *brw,
       assert(0);
       break;
    }
-   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;
@@ -803,6 +796,21 @@ 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)
+{
+   struct intel_context *intel = &brw->intel;
+
+   /* 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
@@ -830,11 +838,16 @@ 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 = &intel->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;
 
-   gen6_blorp_compute_tile_masks(params, &tile_mask_x, &tile_mask_y);
+   brw_get_depthstencil_tile_masks(params->depth.mt,
+                                   params->depth.level,
+                                   params->depth.layer,
+                                   NULL,
+                                   &tile_mask_x, &tile_mask_y);
 
    /* 3DSTATE_DEPTH_BUFFER */
    {
@@ -860,6 +873,9 @@ 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;
 
@@ -868,9 +884,7 @@ gen6_blorp_emit_depth_stencil_config(struct brw_context *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->region->pitch - 1) |
                 params->depth_format << 18 |
                 1 << 21 | /* separate stencil enable */
                 1 << 22 | /* hiz enable */
@@ -900,7 +914,7 @@ gen6_blorp_emit_depth_stencil_config(struct brw_context *brw,
 
       BEGIN_BATCH(3);
       OUT_BATCH((_3DSTATE_HIER_DEPTH_BUFFER << 16) | (3 - 2));
-      OUT_BATCH(hiz_region->pitch * hiz_region->cpp - 1);
+      OUT_BATCH(hiz_region->pitch - 1);
       OUT_RELOC(hiz_region->bo,
                 I915_GEM_DOMAIN_RENDER, I915_GEM_DOMAIN_RENDER,
                 hiz_offset);
@@ -1045,7 +1059,7 @@ gen6_blorp_exec(struct intel_context *intel,
    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);
+   gen6_emit_3dstate_sample_mask(brw, params->num_samples, 1.0, false, ~0u);
    gen6_blorp_emit_state_base_address(brw, params);
    gen6_blorp_emit_vertices(brw, params);
    gen6_blorp_emit_urb_config(brw, params);
@@ -1058,16 +1072,18 @@ 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);
       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,
@@ -1081,6 +1097,8 @@ 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);
@@ -1093,14 +1111,5 @@ 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;
 }