vk/formats: Document new meaning of anv_format::cpp
[mesa.git] / src / vulkan / pipeline.c
index c0606dabba7075fd543895656e2784dff170c9fb..f574038f83f072bcde9aae1be7f78ee6aa058e46 100644 (file)
@@ -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);
 
@@ -129,17 +129,17 @@ emit_ia_state(struct anv_pipeline *pipeline,
               const struct anv_pipeline_create_info *extra)
 {
    static const uint32_t vk_to_gen_primitive_type[] = {
-      [VK_PRIMITIVE_TOPOLOGY_POINT_LIST] = _3DPRIM_POINTLIST,
-      [VK_PRIMITIVE_TOPOLOGY_LINE_LIST] = _3DPRIM_LINELIST,
-      [VK_PRIMITIVE_TOPOLOGY_LINE_STRIP] = _3DPRIM_LINESTRIP,
-      [VK_PRIMITIVE_TOPOLOGY_TRIANGLE_LIST] = _3DPRIM_TRILIST,
-      [VK_PRIMITIVE_TOPOLOGY_TRIANGLE_STRIP] = _3DPRIM_TRISTRIP,
-      [VK_PRIMITIVE_TOPOLOGY_TRIANGLE_FAN] = _3DPRIM_TRIFAN,
-      [VK_PRIMITIVE_TOPOLOGY_LINE_LIST_ADJ] = _3DPRIM_LINELIST_ADJ,
-      [VK_PRIMITIVE_TOPOLOGY_LINE_STRIP_ADJ] = _3DPRIM_LISTSTRIP_ADJ,
+      [VK_PRIMITIVE_TOPOLOGY_POINT_LIST]        = _3DPRIM_POINTLIST,
+      [VK_PRIMITIVE_TOPOLOGY_LINE_LIST]         = _3DPRIM_LINELIST,
+      [VK_PRIMITIVE_TOPOLOGY_LINE_STRIP]        = _3DPRIM_LINESTRIP,
+      [VK_PRIMITIVE_TOPOLOGY_TRIANGLE_LIST]     = _3DPRIM_TRILIST,
+      [VK_PRIMITIVE_TOPOLOGY_TRIANGLE_STRIP]    = _3DPRIM_TRISTRIP,
+      [VK_PRIMITIVE_TOPOLOGY_TRIANGLE_FAN]      = _3DPRIM_TRIFAN,
+      [VK_PRIMITIVE_TOPOLOGY_LINE_LIST_ADJ]     = _3DPRIM_LINELIST_ADJ,
+      [VK_PRIMITIVE_TOPOLOGY_LINE_STRIP_ADJ]    = _3DPRIM_LISTSTRIP_ADJ,
       [VK_PRIMITIVE_TOPOLOGY_TRIANGLE_LIST_ADJ] = _3DPRIM_TRILIST_ADJ,
       [VK_PRIMITIVE_TOPOLOGY_TRIANGLE_STRIP_ADJ] = _3DPRIM_TRISTRIP_ADJ,
-      [VK_PRIMITIVE_TOPOLOGY_PATCH] = _3DPRIM_PATCHLIST_1
+      [VK_PRIMITIVE_TOPOLOGY_PATCH]             = _3DPRIM_PATCHLIST_1
    };
    uint32_t topology = vk_to_gen_primitive_type[info->topology];
 
@@ -158,26 +158,26 @@ emit_rs_state(struct anv_pipeline *pipeline, VkPipelineRsStateCreateInfo *info,
               const struct anv_pipeline_create_info *extra)
 {
    static const uint32_t vk_to_gen_cullmode[] = {
-      [VK_CULL_MODE_NONE] = CULLMODE_NONE,
-      [VK_CULL_MODE_FRONT] = CULLMODE_FRONT,
-      [VK_CULL_MODE_BACK] = CULLMODE_BACK,
-      [VK_CULL_MODE_FRONT_AND_BACK] = CULLMODE_BOTH
+      [VK_CULL_MODE_NONE]                       = CULLMODE_NONE,
+      [VK_CULL_MODE_FRONT]                      = CULLMODE_FRONT,
+      [VK_CULL_MODE_BACK]                       = CULLMODE_BACK,
+      [VK_CULL_MODE_FRONT_AND_BACK]             = CULLMODE_BOTH
    };
 
    static const uint32_t vk_to_gen_fillmode[] = {
-      [VK_FILL_MODE_POINTS] = RASTER_POINT,
-      [VK_FILL_MODE_WIREFRAME] = RASTER_WIREFRAME,
-      [VK_FILL_MODE_SOLID] = RASTER_SOLID
+      [VK_FILL_MODE_POINTS]                     = RASTER_POINT,
+      [VK_FILL_MODE_WIREFRAME]                  = RASTER_WIREFRAME,
+      [VK_FILL_MODE_SOLID]                      = RASTER_SOLID
    };
 
    static const uint32_t vk_to_gen_front_face[] = {
-      [VK_FRONT_FACE_CCW] = CounterClockwise,
-      [VK_FRONT_FACE_CW] = Clockwise
+      [VK_FRONT_FACE_CCW]                       = CounterClockwise,
+      [VK_FRONT_FACE_CW]                        = Clockwise
    };
    
    static const uint32_t vk_to_gen_coordinate_origin[] = {
-      [VK_COORDINATE_ORIGIN_UPPER_LEFT] = UPPERLEFT,
-      [VK_COORDINATE_ORIGIN_LOWER_LEFT] = LOWERLEFT
+      [VK_COORDINATE_ORIGIN_UPPER_LEFT]         = UPPERLEFT,
+      [VK_COORDINATE_ORIGIN_LOWER_LEFT]         = LOWERLEFT
    };
 
    struct GEN8_3DSTATE_SF sf = {
@@ -314,30 +314,39 @@ emit_cb_state(struct anv_pipeline *pipeline, VkPipelineCbStateCreateInfo *info)
 }
 
 static const uint32_t vk_to_gen_compare_op[] = {
-   [VK_COMPARE_OP_NEVER] = COMPAREFUNCTION_NEVER,
-   [VK_COMPARE_OP_LESS] = COMPAREFUNCTION_LESS,
-   [VK_COMPARE_OP_EQUAL] = COMPAREFUNCTION_EQUAL,
-   [VK_COMPARE_OP_LESS_EQUAL] = COMPAREFUNCTION_LEQUAL,
-   [VK_COMPARE_OP_GREATER] = COMPAREFUNCTION_GREATER,
-   [VK_COMPARE_OP_NOT_EQUAL] = COMPAREFUNCTION_NOTEQUAL,
-   [VK_COMPARE_OP_GREATER_EQUAL] = COMPAREFUNCTION_GEQUAL,
-   [VK_COMPARE_OP_ALWAYS] = COMPAREFUNCTION_ALWAYS,
+   [VK_COMPARE_OP_NEVER]                        = COMPAREFUNCTION_NEVER,
+   [VK_COMPARE_OP_LESS]                         = COMPAREFUNCTION_LESS,
+   [VK_COMPARE_OP_EQUAL]                        = COMPAREFUNCTION_EQUAL,
+   [VK_COMPARE_OP_LESS_EQUAL]                   = COMPAREFUNCTION_LEQUAL,
+   [VK_COMPARE_OP_GREATER]                      = COMPAREFUNCTION_GREATER,
+   [VK_COMPARE_OP_NOT_EQUAL]                    = COMPAREFUNCTION_NOTEQUAL,
+   [VK_COMPARE_OP_GREATER_EQUAL]                = COMPAREFUNCTION_GEQUAL,
+   [VK_COMPARE_OP_ALWAYS]                       = COMPAREFUNCTION_ALWAYS,
 };
 
 static const uint32_t vk_to_gen_stencil_op[] = {
-   [VK_STENCIL_OP_KEEP] = 0,
-   [VK_STENCIL_OP_ZERO] = 0,
-   [VK_STENCIL_OP_REPLACE] = 0,
-   [VK_STENCIL_OP_INC_CLAMP] = 0,
-   [VK_STENCIL_OP_DEC_CLAMP] = 0,
-   [VK_STENCIL_OP_INVERT] = 0,
-   [VK_STENCIL_OP_INC_WRAP] = 0,
-   [VK_STENCIL_OP_DEC_WRAP] = 0
+   [VK_STENCIL_OP_KEEP]                         = 0,
+   [VK_STENCIL_OP_ZERO]                         = 0,
+   [VK_STENCIL_OP_REPLACE]                      = 0,
+   [VK_STENCIL_OP_INC_CLAMP]                    = 0,
+   [VK_STENCIL_OP_DEC_CLAMP]                    = 0,
+   [VK_STENCIL_OP_INVERT]                       = 0,
+   [VK_STENCIL_OP_INC_WRAP]                     = 0,
+   [VK_STENCIL_OP_DEC_WRAP]                     = 0
 };
 
 static void
 emit_ds_state(struct anv_pipeline *pipeline, VkPipelineDsStateCreateInfo *info)
 {
+   if (info == NULL) {
+      /* We're going to OR this together with the dynamic state.  We need
+       * to make sure it's initialized to something useful.
+       */
+      memset(pipeline->state_wm_depth_stencil, 0,
+             sizeof(pipeline->state_wm_depth_stencil));
+      return;
+   }
+
    /* bool32_t depthBoundsEnable;          // optional (depth_bounds_test) */
 
    struct GEN8_3DSTATE_WM_DEPTH_STENCIL wm_depth_stencil = {
@@ -378,7 +387,9 @@ anv_pipeline_destroy(struct anv_device *device,
    assert(obj_type == VK_OBJECT_TYPE_PIPELINE);
 
    anv_compiler_free(pipeline);
-   anv_batch_finish(&pipeline->batch);
+   anv_reloc_list_finish(&pipeline->batch.relocs, pipeline->device);
+   anv_state_stream_finish(&pipeline->program_stream);
+   anv_state_pool_free(&device->dynamic_state_pool, pipeline->blend_state);
    anv_device_free(pipeline->device, pipeline);
 }
 
@@ -397,7 +408,7 @@ anv_pipeline_create(
    VkPipelineRsStateCreateInfo *rs_info = NULL;
    VkPipelineDsStateCreateInfo *ds_info = NULL;
    VkPipelineCbStateCreateInfo *cb_info = NULL;
-   VkPipelineVertexInputCreateInfo *vi_info;
+   VkPipelineVertexInputCreateInfo *vi_info = NULL;
    VkResult result;
    uint32_t offset, length;
 
@@ -412,9 +423,14 @@ anv_pipeline_create(
    pipeline->device = device;
    pipeline->layout = (struct anv_pipeline_layout *) pCreateInfo->layout;
    memset(pipeline->shaders, 0, sizeof(pipeline->shaders));
-   result = anv_batch_init(&pipeline->batch, device);
-   if (result != VK_SUCCESS)
-      goto fail;
+
+   result = anv_reloc_list_init(&pipeline->batch.relocs, device);
+   if (result != VK_SUCCESS) {
+      anv_device_free(device, pipeline);
+      return result;
+   }
+   pipeline->batch.next = pipeline->batch.start = pipeline->batch_data;
+   pipeline->batch.end = pipeline->batch.start + sizeof(pipeline->batch_data);
 
    anv_state_stream_init(&pipeline->program_stream,
                          &device->instruction_block_pool);
@@ -472,16 +488,36 @@ anv_pipeline_create(
    emit_ia_state(pipeline, ia_info, extra);
    assert(rs_info);
    emit_rs_state(pipeline, rs_info, extra);
-   /* ds_info is optional if we're not using depth or stencil buffers, ps is
-    * optional for depth-only rendering. */
-   if (ds_info)
-      emit_ds_state(pipeline, ds_info);
-
+   emit_ds_state(pipeline, ds_info);
    emit_cb_state(pipeline, cb_info);
 
+   anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_VF_STATISTICS,
+                   .StatisticsEnable = true);
+   anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_HS, .Enable = false);
+   anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_TE, .TEEnable = false);
+   anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_DS, .FunctionEnable = false);
+   anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_STREAMOUT, .SOFunctionEnable = false);
+
+   anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_PUSH_CONSTANT_ALLOC_VS,
+                  .ConstantBufferOffset = 0,
+                  .ConstantBufferSize = 4);
+   anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_PUSH_CONSTANT_ALLOC_GS,
+                  .ConstantBufferOffset = 4,
+                  .ConstantBufferSize = 4);
+   anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_PUSH_CONSTANT_ALLOC_PS,
+                  .ConstantBufferOffset = 8,
+                  .ConstantBufferSize = 4);
+
+   anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_WM_CHROMAKEY,
+                  .ChromaKeyKillEnable = false);
+   anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_SBE_SWIZ);
+   anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_AA_LINE_PARAMETERS);
+
    anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_CLIP,
                   .ClipEnable = true,
-                  .ViewportXYClipTestEnable = !(extra && extra->disable_viewport));
+                  .ViewportXYClipTestEnable = !(extra && extra->disable_viewport),
+                  .MinimumPointWidth = 0.125,
+                  .MaximumPointWidth = 255.875);
 
    anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_WM,
                   .StatisticsEnable = true,
@@ -537,8 +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,
@@ -564,8 +600,6 @@ anv_pipeline_create(
                      .VertexURBEntryOutputReadOffset = offset,
                      .VertexURBEntryOutputLength = length);
 
-   //trp_generate_blend_hw_cmds(batch, pipeline);
-
    const struct brw_vue_prog_data *vue_prog_data = &pipeline->vs_prog_data.base;
    /* Skip the VUE header and position slots */
    offset = 1;
@@ -594,11 +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,
@@ -642,8 +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 ?
@@ -672,11 +703,6 @@ anv_pipeline_create(
    *pPipeline = (VkPipeline) pipeline;
 
    return VK_SUCCESS;
-
- fail:
-   anv_device_free(device, pipeline);
-   
-   return result;
 }
 
 VkResult anv_CreateGraphicsPipelineDerivative(
@@ -689,11 +715,75 @@ VkResult anv_CreateGraphicsPipelineDerivative(
 }
 
 VkResult anv_CreateComputePipeline(
-    VkDevice                                    device,
+    VkDevice                                    _device,
     const VkComputePipelineCreateInfo*          pCreateInfo,
     VkPipeline*                                 pPipeline)
 {
-   stub_return(VK_UNSUPPORTED);
+   struct anv_device *device = (struct anv_device *) _device;
+   struct anv_pipeline *pipeline;
+   VkResult result;
+
+   assert(pCreateInfo->sType == VK_STRUCTURE_TYPE_COMPUTE_PIPELINE_CREATE_INFO);
+
+   pipeline = anv_device_alloc(device, sizeof(*pipeline), 8,
+                               VK_SYSTEM_ALLOC_TYPE_API_OBJECT);
+   if (pipeline == NULL)
+      return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
+
+   pipeline->base.destructor = anv_pipeline_destroy;
+   pipeline->device = device;
+   pipeline->layout = (struct anv_pipeline_layout *) pCreateInfo->layout;
+
+   result = anv_reloc_list_init(&pipeline->batch.relocs, device);
+   if (result != VK_SUCCESS) {
+      anv_device_free(device, pipeline);
+      return result;
+   }
+   pipeline->batch.next = pipeline->batch.start = pipeline->batch_data;
+   pipeline->batch.end = pipeline->batch.start + sizeof(pipeline->batch_data);
+
+   anv_state_stream_init(&pipeline->program_stream,
+                         &device->instruction_block_pool);
+
+   memset(pipeline->shaders, 0, sizeof(pipeline->shaders));
+
+   pipeline->shaders[VK_SHADER_STAGE_COMPUTE] =
+      (struct anv_shader *) pCreateInfo->cs.shader;
+
+   pipeline->use_repclear = false;
+
+   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 = 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,
+                  .ResetGatewayTimer = true,
+                  .BypassGatewayControl = true,
+                  .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;
 }
 
 VkResult anv_StorePipeline(