vk/formats: Document new meaning of anv_format::cpp

[mesa.git] / src / vulkan / pipeline.c

diff --git a/src/vulkan/pipeline.c b/src/vulkan/pipeline.c
index cf7562ae496073b05a8d596ec13526eba0369fe7..f574038f83f072bcde9aae1be7f78ee6aa058e46 100644 (file)
--- a/src/vulkan/pipeline.c
+++ b/src/vulkan/pipeline.c
@@ -96,13 +96,13 @@ emit_vertex_input(struct anv_pipeline *pipeline, VkPipelineVertexInputCreateInfo
       struct GEN8_VERTEX_ELEMENT_STATE element = {
          .VertexBufferIndex = desc->binding,
          .Valid = true,
-         .SourceElementFormat = format->format,
+         .SourceElementFormat = format->surface_format,
          .EdgeFlagEnable = false,
          .SourceElementOffset = desc->offsetInBytes,
          .Component0Control = VFCOMP_STORE_SRC,
-         .Component1Control = format->channels >= 2 ? VFCOMP_STORE_SRC : VFCOMP_STORE_0,
-         .Component2Control = format->channels >= 3 ? VFCOMP_STORE_SRC : VFCOMP_STORE_0,
-         .Component3Control = format->channels >= 4 ? VFCOMP_STORE_SRC : VFCOMP_STORE_1_FP
+         .Component1Control = format->num_channels >= 2 ? VFCOMP_STORE_SRC : VFCOMP_STORE_0,
+         .Component2Control = format->num_channels >= 3 ? VFCOMP_STORE_SRC : VFCOMP_STORE_0,
+         .Component3Control = format->num_channels >= 4 ? VFCOMP_STORE_SRC : VFCOMP_STORE_1_FP
       };
       GEN8_VERTEX_ELEMENT_STATE_pack(NULL, &p[1 + i * 2], &element);
 
@@ -573,8 +573,8 @@ anv_pipeline_create(
                      .BindingTableEntryCount = 0,
                      .ExpectedVertexCount = pipeline->gs_vertex_count,
                         
-                     .PerThreadScratchSpace = 0,
-                     .ScratchSpaceBasePointer = 0,
+                     .ScratchSpaceBasePointer = pipeline->scratch_start[VK_SHADER_STAGE_GEOMETRY],
+                     .PerThreadScratchSpace = ffs(gs_prog_data->base.base.total_scratch / 2048),
 
                      .OutputVertexSize = gs_prog_data->output_vertex_size_hwords * 2 - 1,
                      .OutputTopology = gs_prog_data->output_topology,
@@ -628,11 +628,8 @@ anv_pipeline_create(
                      .AccessesUAV = false,
                      .SoftwareExceptionEnable = false,
 
-                     /* FIXME: pointer needs to be assigned outside as it aliases
-                      * PerThreadScratchSpace.
-                      */
-                     .ScratchSpaceBasePointer = 0,
-                     .PerThreadScratchSpace = 0,
+                     .ScratchSpaceBasePointer = pipeline->scratch_start[VK_SHADER_STAGE_VERTEX],
+                     .PerThreadScratchSpace = ffs(vue_prog_data->base.total_scratch / 2048),
 
                      .DispatchGRFStartRegisterForURBData =
                      vue_prog_data->base.dispatch_grf_start_reg,
@@ -676,8 +673,8 @@ anv_pipeline_create(
                   .VectorMaskEnable = true,
                   .SamplerCount = 1,
 
-                  .ScratchSpaceBasePointer = 0,
-                  .PerThreadScratchSpace = 0,
+                  .ScratchSpaceBasePointer = pipeline->scratch_start[VK_SHADER_STAGE_FRAGMENT],
+                  .PerThreadScratchSpace = ffs(wm_prog_data->base.total_scratch / 2048),
                   
                   .MaximumNumberofThreadsPerPSD = 64 - 2,
                   .PositionXYOffsetSelect = wm_prog_data->uses_pos_offset ?
@@ -757,11 +754,13 @@ VkResult anv_CreateComputePipeline(
 
    anv_compiler_run(device->compiler, pipeline);
 
+   const struct brw_cs_prog_data *cs_prog_data = &pipeline->cs_prog_data;
+
    anv_batch_emit(&pipeline->batch, GEN8_MEDIA_VFE_STATE,
-                  .ScratchSpaceBasePointer = 0, /* FIXME: Scratch bo, this should be a reloc? */
-                  .StackSize = 0,
-                  .PerThreadScratchSpace = 0,
+                  .ScratchSpaceBasePointer = pipeline->scratch_start[VK_SHADER_STAGE_FRAGMENT],
+                  .PerThreadScratchSpace = ffs(cs_prog_data->base.total_scratch / 2048),
                   .ScratchSpaceBasePointerHigh = 0,
+                  .StackSize = 0,
 
                   .MaximumNumberofThreads = device->info.max_cs_threads - 1,
                   .NumberofURBEntries = 2,
@@ -770,6 +769,18 @@ VkResult anv_CreateComputePipeline(
                   .URBEntryAllocationSize = 2,
                   .CURBEAllocationSize = 0);
 
+   struct brw_cs_prog_data *prog_data = &pipeline->cs_prog_data;
+   uint32_t group_size = prog_data->local_size[0] *
+      prog_data->local_size[1] * prog_data->local_size[2];
+   pipeline->cs_thread_width_max = DIV_ROUND_UP(group_size, prog_data->simd_size);
+   uint32_t remainder = group_size & (prog_data->simd_size - 1);
+
+   if (remainder > 0)
+      pipeline->cs_right_mask = ~0u >> (32 - remainder);
+   else
+      pipeline->cs_right_mask = ~0u >> (32 - prog_data->simd_size);
+
+
    *pPipeline = (VkPipeline) pipeline;
 
    return VK_SUCCESS;