radv: clean up tessellation state emission
[mesa.git] / src / amd / vulkan / radv_pipeline.c
index a05ea7f11fd4b0a598ebde5b1328d8305ee1db64..6d17fdbb2389d9500127aac17b7e6be11e45e5ee 100644 (file)
@@ -82,12 +82,6 @@ struct radv_dsa_order_invariance {
        bool pass_set;
 };
 
-struct radv_tessellation_state {
-       uint32_t ls_hs_config;
-       unsigned lds_size;
-       uint32_t tf_param;
-};
-
 static const VkPipelineMultisampleStateCreateInfo *
 radv_pipeline_get_multisample_state(const VkGraphicsPipelineCreateInfo *pCreateInfo)
 {
@@ -1985,99 +1979,6 @@ radv_get_shader(struct radv_pipeline *pipeline,
        return pipeline->shaders[stage];
 }
 
-static struct radv_tessellation_state
-calculate_tess_state(struct radv_pipeline *pipeline,
-                    const VkGraphicsPipelineCreateInfo *pCreateInfo)
-{
-       unsigned num_tcs_input_cp;
-       unsigned num_tcs_output_cp;
-       unsigned lds_size;
-       unsigned num_patches;
-       struct radv_tessellation_state tess = {0};
-
-       num_tcs_input_cp = pCreateInfo->pTessellationState->patchControlPoints;
-       num_tcs_output_cp = pipeline->shaders[MESA_SHADER_TESS_CTRL]->info.tcs.tcs_vertices_out; //TCS VERTICES OUT
-       num_patches = pipeline->shaders[MESA_SHADER_TESS_CTRL]->info.tcs.num_patches;
-
-       lds_size = pipeline->shaders[MESA_SHADER_TESS_CTRL]->info.tcs.lds_size;
-
-       if (pipeline->device->physical_device->rad_info.chip_class >= GFX7) {
-               assert(lds_size <= 65536);
-               lds_size = align(lds_size, 512) / 512;
-       } else {
-               assert(lds_size <= 32768);
-               lds_size = align(lds_size, 256) / 256;
-       }
-
-       tess.lds_size = lds_size;
-
-       tess.ls_hs_config = S_028B58_NUM_PATCHES(num_patches) |
-               S_028B58_HS_NUM_INPUT_CP(num_tcs_input_cp) |
-               S_028B58_HS_NUM_OUTPUT_CP(num_tcs_output_cp);
-
-       struct radv_shader_variant *tes = radv_get_shader(pipeline, MESA_SHADER_TESS_EVAL);
-       unsigned type = 0, partitioning = 0, topology = 0, distribution_mode = 0;
-
-       switch (tes->info.tes.primitive_mode) {
-       case GL_TRIANGLES:
-               type = V_028B6C_TESS_TRIANGLE;
-               break;
-       case GL_QUADS:
-               type = V_028B6C_TESS_QUAD;
-               break;
-       case GL_ISOLINES:
-               type = V_028B6C_TESS_ISOLINE;
-               break;
-       }
-
-       switch (tes->info.tes.spacing) {
-       case TESS_SPACING_EQUAL:
-               partitioning = V_028B6C_PART_INTEGER;
-               break;
-       case TESS_SPACING_FRACTIONAL_ODD:
-               partitioning = V_028B6C_PART_FRAC_ODD;
-               break;
-       case TESS_SPACING_FRACTIONAL_EVEN:
-               partitioning = V_028B6C_PART_FRAC_EVEN;
-               break;
-       default:
-               break;
-       }
-
-       bool ccw = tes->info.tes.ccw;
-       const VkPipelineTessellationDomainOriginStateCreateInfo *domain_origin_state =
-                     vk_find_struct_const(pCreateInfo->pTessellationState,
-                                          PIPELINE_TESSELLATION_DOMAIN_ORIGIN_STATE_CREATE_INFO);
-
-       if (domain_origin_state && domain_origin_state->domainOrigin != VK_TESSELLATION_DOMAIN_ORIGIN_UPPER_LEFT)
-               ccw = !ccw;
-
-       if (tes->info.tes.point_mode)
-               topology = V_028B6C_OUTPUT_POINT;
-       else if (tes->info.tes.primitive_mode == GL_ISOLINES)
-               topology = V_028B6C_OUTPUT_LINE;
-       else if (ccw)
-               topology = V_028B6C_OUTPUT_TRIANGLE_CCW;
-       else
-               topology = V_028B6C_OUTPUT_TRIANGLE_CW;
-
-       if (pipeline->device->physical_device->rad_info.has_distributed_tess) {
-               if (pipeline->device->physical_device->rad_info.family == CHIP_FIJI ||
-                   pipeline->device->physical_device->rad_info.family >= CHIP_POLARIS10)
-                       distribution_mode = V_028B6C_DISTRIBUTION_MODE_TRAPEZOIDS;
-               else
-                       distribution_mode = V_028B6C_DISTRIBUTION_MODE_DONUTS;
-       } else
-               distribution_mode = V_028B6C_DISTRIBUTION_MODE_NO_DIST;
-
-       tess.tf_param = S_028B6C_TYPE(type) |
-               S_028B6C_PARTITIONING(partitioning) |
-               S_028B6C_TOPOLOGY(topology) |
-               S_028B6C_DISTRIBUTION_MODE(distribution_mode);
-
-       return tess;
-}
-
 static const struct radv_vs_output_info *get_vs_output_info(const struct radv_pipeline *pipeline)
 {
        if (radv_pipeline_has_gs(pipeline))
@@ -2236,9 +2137,10 @@ radv_get_attrib_stride(const VkPipelineVertexInputStateCreateInfo *input_state,
 static struct radv_pipeline_key
 radv_generate_graphics_pipeline_key(struct radv_pipeline *pipeline,
                                     const VkGraphicsPipelineCreateInfo *pCreateInfo,
-                                    const struct radv_blend_state *blend,
-                                    bool has_view_index)
+                                    const struct radv_blend_state *blend)
 {
+       RADV_FROM_HANDLE(radv_render_pass, pass, pCreateInfo->renderPass);
+       struct radv_subpass *subpass = pass->subpasses + pCreateInfo->subpass;
        const VkPipelineVertexInputStateCreateInfo *input_state =
                                                 pCreateInfo->pVertexInputState;
        const VkPipelineVertexInputDivisorStateCreateInfoEXT *divisor_state =
@@ -2250,7 +2152,7 @@ radv_generate_graphics_pipeline_key(struct radv_pipeline *pipeline,
        if (pCreateInfo->flags & VK_PIPELINE_CREATE_DISABLE_OPTIMIZATION_BIT)
                key.optimisations_disabled = 1;
 
-       key.has_multiview_view_index = has_view_index;
+       key.has_multiview_view_index = !!subpass->view_mask;
 
        uint32_t binding_input_rate = 0;
        uint32_t instance_rate_divisors[MAX_VERTEX_ATTRIBS];
@@ -3907,9 +3809,9 @@ radv_pipeline_generate_hw_es(struct radeon_cmdbuf *cs,
 static void
 radv_pipeline_generate_hw_ls(struct radeon_cmdbuf *cs,
                             struct radv_pipeline *pipeline,
-                            struct radv_shader_variant *shader,
-                            const struct radv_tessellation_state *tess)
+                            struct radv_shader_variant *shader)
 {
+       unsigned num_lds_blocks = pipeline->shaders[MESA_SHADER_TESS_CTRL]->info.tcs.num_lds_blocks;
        uint64_t va = radv_buffer_get_va(shader->bo) + shader->bo_offset;
        uint32_t rsrc2 = shader->config.rsrc2;
 
@@ -3917,7 +3819,7 @@ radv_pipeline_generate_hw_ls(struct radeon_cmdbuf *cs,
        radeon_emit(cs, va >> 8);
        radeon_emit(cs, S_00B524_MEM_BASE(va >> 40));
 
-       rsrc2 |= S_00B52C_LDS_SIZE(tess->lds_size);
+       rsrc2 |= S_00B52C_LDS_SIZE(num_lds_blocks);
        if (pipeline->device->physical_device->rad_info.chip_class == GFX7 &&
            pipeline->device->physical_device->rad_info.family != CHIP_HAWAII)
                radeon_set_sh_reg(cs, R_00B52C_SPI_SHADER_PGM_RSRC2_LS, rsrc2);
@@ -4064,20 +3966,11 @@ radv_pipeline_generate_hw_ngg(struct radeon_cmdbuf *ctx_cs,
 static void
 radv_pipeline_generate_hw_hs(struct radeon_cmdbuf *cs,
                             struct radv_pipeline *pipeline,
-                            struct radv_shader_variant *shader,
-                            const struct radv_tessellation_state *tess)
+                            struct radv_shader_variant *shader)
 {
        uint64_t va = radv_buffer_get_va(shader->bo) + shader->bo_offset;
 
        if (pipeline->device->physical_device->rad_info.chip_class >= GFX9) {
-               unsigned hs_rsrc2 = shader->config.rsrc2;
-
-               if (pipeline->device->physical_device->rad_info.chip_class >= GFX10) {
-                       hs_rsrc2 |= S_00B42C_LDS_SIZE_GFX10(tess->lds_size);
-               } else {
-                       hs_rsrc2 |= S_00B42C_LDS_SIZE_GFX9(tess->lds_size);
-               }
-
                if (pipeline->device->physical_device->rad_info.chip_class >= GFX10) {
                        radeon_set_sh_reg_seq(cs, R_00B520_SPI_SHADER_PGM_LO_LS, 2);
                        radeon_emit(cs, va >> 8);
@@ -4090,7 +3983,7 @@ radv_pipeline_generate_hw_hs(struct radeon_cmdbuf *cs,
 
                radeon_set_sh_reg_seq(cs, R_00B428_SPI_SHADER_PGM_RSRC1_HS, 2);
                radeon_emit(cs, shader->config.rsrc1);
-               radeon_emit(cs, hs_rsrc2);
+               radeon_emit(cs, shader->config.rsrc2);
        } else {
                radeon_set_sh_reg_seq(cs, R_00B420_SPI_SHADER_PGM_LO_HS, 4);
                radeon_emit(cs, va >> 8);
@@ -4103,8 +3996,7 @@ radv_pipeline_generate_hw_hs(struct radeon_cmdbuf *cs,
 static void
 radv_pipeline_generate_vertex_shader(struct radeon_cmdbuf *ctx_cs,
                                     struct radeon_cmdbuf *cs,
-                                    struct radv_pipeline *pipeline,
-                                    const struct radv_tessellation_state *tess)
+                                    struct radv_pipeline *pipeline)
 {
        struct radv_shader_variant *vs;
 
@@ -4114,7 +4006,7 @@ radv_pipeline_generate_vertex_shader(struct radeon_cmdbuf *ctx_cs,
                return;
 
        if (vs->info.vs.as_ls)
-               radv_pipeline_generate_hw_ls(cs, pipeline, vs, tess);
+               radv_pipeline_generate_hw_ls(cs, pipeline, vs);
        else if (vs->info.vs.as_es)
                radv_pipeline_generate_hw_es(cs, pipeline, vs);
        else if (vs->info.is_ngg)
@@ -4126,12 +4018,8 @@ radv_pipeline_generate_vertex_shader(struct radeon_cmdbuf *ctx_cs,
 static void
 radv_pipeline_generate_tess_shaders(struct radeon_cmdbuf *ctx_cs,
                                    struct radeon_cmdbuf *cs,
-                                   struct radv_pipeline *pipeline,
-                                   const struct radv_tessellation_state *tess)
+                                   struct radv_pipeline *pipeline)
 {
-       if (!radv_pipeline_has_tess(pipeline))
-               return;
-
        struct radv_shader_variant *tes, *tcs;
 
        tcs = pipeline->shaders[MESA_SHADER_TESS_CTRL];
@@ -4146,17 +4034,7 @@ radv_pipeline_generate_tess_shaders(struct radeon_cmdbuf *ctx_cs,
                        radv_pipeline_generate_hw_vs(ctx_cs, cs, pipeline, tes);
        }
 
-       radv_pipeline_generate_hw_hs(cs, pipeline, tcs, tess);
-
-       radeon_set_context_reg(ctx_cs, R_028B6C_VGT_TF_PARAM,
-                              tess->tf_param);
-
-       if (pipeline->device->physical_device->rad_info.chip_class >= GFX7)
-               radeon_set_context_reg_idx(ctx_cs, R_028B58_VGT_LS_HS_CONFIG, 2,
-                                          tess->ls_hs_config);
-       else
-               radeon_set_context_reg(ctx_cs, R_028B58_VGT_LS_HS_CONFIG,
-                                      tess->ls_hs_config);
+       radv_pipeline_generate_hw_hs(cs, pipeline, tcs);
 
        if (pipeline->device->physical_device->rad_info.chip_class >= GFX10 &&
            !radv_pipeline_has_gs(pipeline) && !radv_pipeline_has_ngg(pipeline)) {
@@ -4167,6 +4045,91 @@ radv_pipeline_generate_tess_shaders(struct radeon_cmdbuf *ctx_cs,
        }
 }
 
+static void
+radv_pipeline_generate_tess_state(struct radeon_cmdbuf *ctx_cs,
+                                 struct radv_pipeline *pipeline,
+                                 const VkGraphicsPipelineCreateInfo *pCreateInfo)
+{
+       struct radv_shader_variant *tes = radv_get_shader(pipeline, MESA_SHADER_TESS_EVAL);
+       unsigned type = 0, partitioning = 0, topology = 0, distribution_mode = 0;
+       unsigned num_tcs_input_cp, num_tcs_output_cp, num_patches;
+       unsigned ls_hs_config;
+
+       num_tcs_input_cp = pCreateInfo->pTessellationState->patchControlPoints;
+       num_tcs_output_cp = pipeline->shaders[MESA_SHADER_TESS_CTRL]->info.tcs.tcs_vertices_out; //TCS VERTICES OUT
+       num_patches = pipeline->shaders[MESA_SHADER_TESS_CTRL]->info.tcs.num_patches;
+
+       ls_hs_config = S_028B58_NUM_PATCHES(num_patches) |
+                      S_028B58_HS_NUM_INPUT_CP(num_tcs_input_cp) |
+                      S_028B58_HS_NUM_OUTPUT_CP(num_tcs_output_cp);
+
+       if (pipeline->device->physical_device->rad_info.chip_class >= GFX7) {
+               radeon_set_context_reg_idx(ctx_cs, R_028B58_VGT_LS_HS_CONFIG,
+                                          2, ls_hs_config);
+       } else {
+               radeon_set_context_reg(ctx_cs, R_028B58_VGT_LS_HS_CONFIG,
+                                      ls_hs_config);
+       }
+
+       switch (tes->info.tes.primitive_mode) {
+       case GL_TRIANGLES:
+               type = V_028B6C_TESS_TRIANGLE;
+               break;
+       case GL_QUADS:
+               type = V_028B6C_TESS_QUAD;
+               break;
+       case GL_ISOLINES:
+               type = V_028B6C_TESS_ISOLINE;
+               break;
+       }
+
+       switch (tes->info.tes.spacing) {
+       case TESS_SPACING_EQUAL:
+               partitioning = V_028B6C_PART_INTEGER;
+               break;
+       case TESS_SPACING_FRACTIONAL_ODD:
+               partitioning = V_028B6C_PART_FRAC_ODD;
+               break;
+       case TESS_SPACING_FRACTIONAL_EVEN:
+               partitioning = V_028B6C_PART_FRAC_EVEN;
+               break;
+       default:
+               break;
+       }
+
+       bool ccw = tes->info.tes.ccw;
+       const VkPipelineTessellationDomainOriginStateCreateInfo *domain_origin_state =
+                     vk_find_struct_const(pCreateInfo->pTessellationState,
+                                          PIPELINE_TESSELLATION_DOMAIN_ORIGIN_STATE_CREATE_INFO);
+
+       if (domain_origin_state && domain_origin_state->domainOrigin != VK_TESSELLATION_DOMAIN_ORIGIN_UPPER_LEFT)
+               ccw = !ccw;
+
+       if (tes->info.tes.point_mode)
+               topology = V_028B6C_OUTPUT_POINT;
+       else if (tes->info.tes.primitive_mode == GL_ISOLINES)
+               topology = V_028B6C_OUTPUT_LINE;
+       else if (ccw)
+               topology = V_028B6C_OUTPUT_TRIANGLE_CCW;
+       else
+               topology = V_028B6C_OUTPUT_TRIANGLE_CW;
+
+       if (pipeline->device->physical_device->rad_info.has_distributed_tess) {
+               if (pipeline->device->physical_device->rad_info.family == CHIP_FIJI ||
+                   pipeline->device->physical_device->rad_info.family >= CHIP_POLARIS10)
+                       distribution_mode = V_028B6C_DISTRIBUTION_MODE_TRAPEZOIDS;
+               else
+                       distribution_mode = V_028B6C_DISTRIBUTION_MODE_DONUTS;
+       } else
+               distribution_mode = V_028B6C_DISTRIBUTION_MODE_NO_DIST;
+
+       radeon_set_context_reg(ctx_cs, R_028B6C_VGT_TF_PARAM,
+                              S_028B6C_TYPE(type) |
+                              S_028B6C_PARTITIONING(partitioning) |
+                              S_028B6C_TOPOLOGY(topology) |
+                              S_028B6C_DISTRIBUTION_MODE(distribution_mode));
+}
+
 static void
 radv_pipeline_generate_hw_gs(struct radeon_cmdbuf *ctx_cs,
                             struct radeon_cmdbuf *cs,
@@ -4619,7 +4582,6 @@ radv_pipeline_generate_pm4(struct radv_pipeline *pipeline,
                            const VkGraphicsPipelineCreateInfo *pCreateInfo,
                            const struct radv_graphics_pipeline_create_info *extra,
                            const struct radv_blend_state *blend,
-                           const struct radv_tessellation_state *tess,
                            unsigned gs_out)
 {
        struct radeon_cmdbuf *ctx_cs = &pipeline->ctx_cs;
@@ -4635,8 +4597,13 @@ radv_pipeline_generate_pm4(struct radv_pipeline *pipeline,
        radv_pipeline_generate_raster_state(ctx_cs, pipeline, pCreateInfo);
        radv_pipeline_generate_multisample_state(ctx_cs, pipeline);
        radv_pipeline_generate_vgt_gs_mode(ctx_cs, pipeline);
-       radv_pipeline_generate_vertex_shader(ctx_cs, cs, pipeline, tess);
-       radv_pipeline_generate_tess_shaders(ctx_cs, cs, pipeline, tess);
+       radv_pipeline_generate_vertex_shader(ctx_cs, cs, pipeline);
+
+       if (radv_pipeline_has_tess(pipeline)) {
+               radv_pipeline_generate_tess_shaders(ctx_cs, cs, pipeline);
+               radv_pipeline_generate_tess_state(ctx_cs, pipeline, pCreateInfo);
+       }
+
        radv_pipeline_generate_geometry_shader(ctx_cs, cs, pipeline);
        radv_pipeline_generate_fragment_shader(ctx_cs, cs, pipeline);
        radv_pipeline_generate_ps_inputs(ctx_cs, pipeline);
@@ -4779,12 +4746,6 @@ radv_pipeline_init(struct radv_pipeline *pipeline,
                   const struct radv_graphics_pipeline_create_info *extra)
 {
        VkResult result;
-       bool has_view_index = false;
-
-       RADV_FROM_HANDLE(radv_render_pass, pass, pCreateInfo->renderPass);
-       struct radv_subpass *subpass = pass->subpasses + pCreateInfo->subpass;
-       if (subpass->view_mask)
-               has_view_index = true;
 
        pipeline->device = device;
        pipeline->layout = radv_pipeline_layout_from_handle(pCreateInfo->layout);
@@ -4807,7 +4768,7 @@ radv_pipeline_init(struct radv_pipeline *pipeline,
                        stage_feedbacks[stage] = &creation_feedback->pPipelineStageCreationFeedbacks[i];
        }
 
-       struct radv_pipeline_key key = radv_generate_graphics_pipeline_key(pipeline, pCreateInfo, &blend, has_view_index);
+       struct radv_pipeline_key key = radv_generate_graphics_pipeline_key(pipeline, pCreateInfo, &blend);
 
        result = radv_create_shaders(pipeline, device, cache, &key, pStages,
                                     pCreateInfo->flags, pipeline_feedback,
@@ -4894,11 +4855,9 @@ radv_pipeline_init(struct radv_pipeline *pipeline,
                calculate_gs_ring_sizes(pipeline, &gs->info.gs_ring_info);
        }
 
-       struct radv_tessellation_state tess = {0};
        if (radv_pipeline_has_tess(pipeline)) {
                pipeline->graphics.tess_patch_control_points =
                        pCreateInfo->pTessellationState->patchControlPoints;
-               tess = calculate_tess_state(pipeline, pCreateInfo);
        }
 
        pipeline->graphics.ia_multi_vgt_param = radv_compute_ia_multi_vgt_param_helpers(pipeline);
@@ -4923,7 +4882,7 @@ radv_pipeline_init(struct radv_pipeline *pipeline,
        pipeline->streamout_shader = radv_pipeline_get_streamout_shader(pipeline);
 
        result = radv_pipeline_scratch_init(device, pipeline);
-       radv_pipeline_generate_pm4(pipeline, pCreateInfo, extra, &blend, &tess, gs_out);
+       radv_pipeline_generate_pm4(pipeline, pCreateInfo, extra, &blend, gs_out);
 
        return result;
 }