src/amd/vulkan/radv_meta_fast_clear.c

   1 /*
   2  * Copyright © 2016 Intel Corporation
   3  *
   4  * Permission is hereby granted, free of charge, to any person obtaining a
   5  * copy of this software and associated documentation files (the "Software"),
   6  * to deal in the Software without restriction, including without limitation
   7  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
   8  * and/or sell copies of the Software, and to permit persons to whom the
   9  * Software is furnished to do so, subject to the following conditions:
  10  *
  11  * The above copyright notice and this permission notice (including the next
  12  * paragraph) shall be included in all copies or substantial portions of the
  13  * Software.
  14  *
  15  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  16  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  17  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
  18  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  19  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
  20  * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
  21  * IN THE SOFTWARE.
  22  */
  23
  24 #include <assert.h>
  25 #include <stdbool.h>
  26
  27 #include "radv_meta.h"
  28 #include "radv_private.h"
  29 #include "sid.h"
  30
  31
  32 static nir_shader *
  33 build_dcc_decompress_compute_shader(struct radv_device *dev)
  34 {
  35         nir_builder b;
  36         const struct glsl_type *buf_type = glsl_sampler_type(GLSL_SAMPLER_DIM_2D,
  37                                                              false,
  38                                                              false,
  39                                                              GLSL_TYPE_FLOAT);
  40         const struct glsl_type *img_type = glsl_sampler_type(GLSL_SAMPLER_DIM_2D,
  41                                                              false,
  42                                                              false,
  43                                                              GLSL_TYPE_FLOAT);
  44         nir_builder_init_simple_shader(&b, NULL, MESA_SHADER_COMPUTE, NULL);
  45         b.shader->info.name = ralloc_strdup(b.shader, "dcc_decompress_compute");
  46
  47         /* We need at least 16/16/1 to cover an entire DCC block in a single workgroup. */
  48         b.shader->info.cs.local_size[0] = 16;
  49         b.shader->info.cs.local_size[1] = 16;
  50         b.shader->info.cs.local_size[2] = 1;
  51         nir_variable *input_img = nir_variable_create(b.shader, nir_var_uniform,
  52                                                       buf_type, "s_tex");
  53         input_img->data.descriptor_set = 0;
  54         input_img->data.binding = 0;
  55
  56         nir_variable *output_img = nir_variable_create(b.shader, nir_var_uniform,
  57                                                        img_type, "out_img");
  58         output_img->data.descriptor_set = 0;
  59         output_img->data.binding = 1;
  60
  61         nir_ssa_def *invoc_id = nir_load_system_value(&b, nir_intrinsic_load_local_invocation_id, 0);
  62         nir_ssa_def *wg_id = nir_load_system_value(&b, nir_intrinsic_load_work_group_id, 0);
  63         nir_ssa_def *block_size = nir_imm_ivec4(&b,
  64                                                 b.shader->info.cs.local_size[0],
  65                                                 b.shader->info.cs.local_size[1],
  66                                                 b.shader->info.cs.local_size[2], 0);
  67
  68         nir_ssa_def *global_id = nir_iadd(&b, nir_imul(&b, wg_id, block_size), invoc_id);
  69
  70         nir_tex_instr *tex = nir_tex_instr_create(b.shader, 2);
  71         tex->sampler_dim = GLSL_SAMPLER_DIM_2D;
  72         tex->op = nir_texop_txf;
  73         tex->src[0].src_type = nir_tex_src_coord;
  74         tex->src[0].src = nir_src_for_ssa(nir_channels(&b, global_id, 3));
  75         tex->src[1].src_type = nir_tex_src_lod;
  76         tex->src[1].src = nir_src_for_ssa(nir_imm_int(&b, 0));
  77         tex->dest_type = nir_type_float;
  78         tex->is_array = false;
  79         tex->coord_components = 2;
  80         tex->texture = nir_deref_var_create(tex, input_img);
  81         tex->sampler = NULL;
  82
  83         nir_ssa_dest_init(&tex->instr, &tex->dest, 4, 32, "tex");
  84         nir_builder_instr_insert(&b, &tex->instr);
  85
  86         nir_intrinsic_instr *membar = nir_intrinsic_instr_create(b.shader, nir_intrinsic_memory_barrier);
  87         nir_builder_instr_insert(&b, &membar->instr);
  88
  89         nir_intrinsic_instr *bar = nir_intrinsic_instr_create(b.shader, nir_intrinsic_barrier);
  90         nir_builder_instr_insert(&b, &bar->instr);
  91
  92         nir_ssa_def *outval = &tex->dest.ssa;
  93         nir_intrinsic_instr *store = nir_intrinsic_instr_create(b.shader, nir_intrinsic_image_store);
  94         store->src[0] = nir_src_for_ssa(global_id);
  95         store->src[1] = nir_src_for_ssa(nir_ssa_undef(&b, 1, 32));
  96         store->src[2] = nir_src_for_ssa(outval);
  97         store->variables[0] = nir_deref_var_create(store, output_img);
  98
  99         nir_builder_instr_insert(&b, &store->instr);
 100         return b.shader;
 101 }
 102
 103 static VkResult
 104 create_dcc_compress_compute(struct radv_device *device)
 105 {
 106         VkResult result = VK_SUCCESS;
 107         struct radv_shader_module cs = { .nir = NULL };
 108
 109         cs.nir = build_dcc_decompress_compute_shader(device);
 110
 111         VkDescriptorSetLayoutCreateInfo ds_create_info = {
 112                 .sType = VK_STRUCTURE_TYPE_DESCRIPTOR_SET_LAYOUT_CREATE_INFO,
 113                 .flags = VK_DESCRIPTOR_SET_LAYOUT_CREATE_PUSH_DESCRIPTOR_BIT_KHR,
 114                 .bindingCount = 2,
 115                 .pBindings = (VkDescriptorSetLayoutBinding[]) {
 116                         {
 117                                 .binding = 0,
 118                                 .descriptorType = VK_DESCRIPTOR_TYPE_SAMPLED_IMAGE,
 119                                 .descriptorCount = 1,
 120                                 .stageFlags = VK_SHADER_STAGE_COMPUTE_BIT,
 121                                 .pImmutableSamplers = NULL
 122                         },
 123                         {
 124                                 .binding = 1,
 125                                 .descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_IMAGE,
 126                                 .descriptorCount = 1,
 127                                 .stageFlags = VK_SHADER_STAGE_COMPUTE_BIT,
 128                                 .pImmutableSamplers = NULL
 129                         },
 130                 }
 131         };
 132
 133         result = radv_CreateDescriptorSetLayout(radv_device_to_handle(device),
 134                                                 &ds_create_info,
 135                                                 &device->meta_state.alloc,
 136                                                 &device->meta_state.fast_clear_flush.dcc_decompress_compute_ds_layout);
 137         if (result != VK_SUCCESS)
 138                 goto cleanup;
 139
 140
 141         VkPipelineLayoutCreateInfo pl_create_info = {
 142                 .sType = VK_STRUCTURE_TYPE_PIPELINE_LAYOUT_CREATE_INFO,
 143                 .setLayoutCount = 1,
 144                 .pSetLayouts = &device->meta_state.fast_clear_flush.dcc_decompress_compute_ds_layout,
 145                 .pushConstantRangeCount = 1,
 146                 .pPushConstantRanges = &(VkPushConstantRange){VK_SHADER_STAGE_COMPUTE_BIT, 0, 8},
 147         };
 148
 149         result = radv_CreatePipelineLayout(radv_device_to_handle(device),
 150                                           &pl_create_info,
 151                                           &device->meta_state.alloc,
 152                                           &device->meta_state.fast_clear_flush.dcc_decompress_compute_p_layout);
 153         if (result != VK_SUCCESS)
 154                 goto cleanup;
 155
 156         /* compute shader */
 157
 158         VkPipelineShaderStageCreateInfo pipeline_shader_stage = {
 159                 .sType = VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_CREATE_INFO,
 160                 .stage = VK_SHADER_STAGE_COMPUTE_BIT,
 161                 .module = radv_shader_module_to_handle(&cs),
 162                 .pName = "main",
 163                 .pSpecializationInfo = NULL,
 164         };
 165
 166         VkComputePipelineCreateInfo vk_pipeline_info = {
 167                 .sType = VK_STRUCTURE_TYPE_COMPUTE_PIPELINE_CREATE_INFO,
 168                 .stage = pipeline_shader_stage,
 169                 .flags = 0,
 170                 .layout = device->meta_state.fast_clear_flush.dcc_decompress_compute_p_layout,
 171         };
 172
 173         result = radv_CreateComputePipelines(radv_device_to_handle(device),
 174                                              radv_pipeline_cache_to_handle(&device->meta_state.cache),
 175                                              1, &vk_pipeline_info, NULL,
 176                                              &device->meta_state.fast_clear_flush.dcc_decompress_compute_pipeline);
 177         if (result != VK_SUCCESS)
 178                 goto cleanup;
 179
 180 cleanup:
 181         ralloc_free(cs.nir);
 182         return result;
 183 }
 184
 185 static VkResult
 186 create_pass(struct radv_device *device)
 187 {
 188         VkResult result;
 189         VkDevice device_h = radv_device_to_handle(device);
 190         const VkAllocationCallbacks *alloc = &device->meta_state.alloc;
 191         VkAttachmentDescription attachment;
 192
 193         attachment.format = VK_FORMAT_UNDEFINED;
 194         attachment.samples = 1;
 195         attachment.loadOp = VK_ATTACHMENT_LOAD_OP_LOAD;
 196         attachment.storeOp = VK_ATTACHMENT_STORE_OP_STORE;
 197         attachment.initialLayout = VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL;
 198         attachment.finalLayout = VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL;
 199
 200         result = radv_CreateRenderPass(device_h,
 201                                        &(VkRenderPassCreateInfo) {
 202                                                .sType = VK_STRUCTURE_TYPE_RENDER_PASS_CREATE_INFO,
 203                                                        .attachmentCount = 1,
 204                                                        .pAttachments = &attachment,
 205                                                        .subpassCount = 1,
 206                                                        .pSubpasses = &(VkSubpassDescription) {
 207                                                        .pipelineBindPoint = VK_PIPELINE_BIND_POINT_GRAPHICS,
 208                                                        .inputAttachmentCount = 0,
 209                                                        .colorAttachmentCount = 1,
 210                                                        .pColorAttachments = (VkAttachmentReference[]) {
 211                                                                {
 212                                                                        .attachment = 0,
 213                                                                        .layout = VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL,
 214                                                                },
 215                                                        },
 216                                                        .pResolveAttachments = NULL,
 217                                                        .pDepthStencilAttachment = &(VkAttachmentReference) {
 218                                                                .attachment = VK_ATTACHMENT_UNUSED,
 219                                                        },
 220                                                        .preserveAttachmentCount = 0,
 221                                                        .pPreserveAttachments = NULL,
 222                                                },
 223                                                                 .dependencyCount = 0,
 224                                        },
 225                                        alloc,
 226                                        &device->meta_state.fast_clear_flush.pass);
 227
 228         return result;
 229 }
 230
 231 static VkResult
 232 create_pipeline_layout(struct radv_device *device, VkPipelineLayout *layout)
 233 {
 234         VkPipelineLayoutCreateInfo pl_create_info = {
 235                 .sType = VK_STRUCTURE_TYPE_PIPELINE_LAYOUT_CREATE_INFO,
 236                 .setLayoutCount = 0,
 237                 .pSetLayouts = NULL,
 238                 .pushConstantRangeCount = 0,
 239                 .pPushConstantRanges = NULL,
 240         };
 241
 242         return radv_CreatePipelineLayout(radv_device_to_handle(device),
 243                                          &pl_create_info,
 244                                          &device->meta_state.alloc,
 245                                          layout);
 246 }
 247
 248 static VkResult
 249 create_pipeline(struct radv_device *device,
 250                 VkShaderModule vs_module_h,
 251                 VkPipelineLayout layout)
 252 {
 253         VkResult result;
 254         VkDevice device_h = radv_device_to_handle(device);
 255
 256         struct radv_shader_module fs_module = {
 257                 .nir = radv_meta_build_nir_fs_noop(),
 258         };
 259
 260         if (!fs_module.nir) {
 261                 /* XXX: Need more accurate error */
 262                 result = VK_ERROR_OUT_OF_HOST_MEMORY;
 263                 goto cleanup;
 264         }
 265
 266         const VkPipelineShaderStageCreateInfo stages[2] = {
 267                 {
 268                         .sType = VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_CREATE_INFO,
 269                         .stage = VK_SHADER_STAGE_VERTEX_BIT,
 270                         .module = vs_module_h,
 271                         .pName = "main",
 272                 },
 273                 {
 274                         .sType = VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_CREATE_INFO,
 275                         .stage = VK_SHADER_STAGE_FRAGMENT_BIT,
 276                         .module = radv_shader_module_to_handle(&fs_module),
 277                         .pName = "main",
 278                 },
 279         };
 280
 281         const VkPipelineVertexInputStateCreateInfo vi_state = {
 282                 .sType = VK_STRUCTURE_TYPE_PIPELINE_VERTEX_INPUT_STATE_CREATE_INFO,
 283                 .vertexBindingDescriptionCount = 0,
 284                 .vertexAttributeDescriptionCount = 0,
 285         };
 286
 287         const VkPipelineInputAssemblyStateCreateInfo ia_state = {
 288                 .sType = VK_STRUCTURE_TYPE_PIPELINE_INPUT_ASSEMBLY_STATE_CREATE_INFO,
 289                 .topology = VK_PRIMITIVE_TOPOLOGY_TRIANGLE_STRIP,
 290                 .primitiveRestartEnable = false,
 291         };
 292
 293         const VkPipelineColorBlendStateCreateInfo blend_state = {
 294                 .sType = VK_STRUCTURE_TYPE_PIPELINE_COLOR_BLEND_STATE_CREATE_INFO,
 295                 .logicOpEnable = false,
 296                 .attachmentCount = 1,
 297                 .pAttachments = (VkPipelineColorBlendAttachmentState []) {
 298                         {
 299                                 .colorWriteMask = VK_COLOR_COMPONENT_R_BIT |
 300                                 VK_COLOR_COMPONENT_G_BIT |
 301                                 VK_COLOR_COMPONENT_B_BIT |
 302                                 VK_COLOR_COMPONENT_A_BIT,
 303                         },
 304                 }
 305         };
 306         const VkPipelineRasterizationStateCreateInfo rs_state = {
 307                 .sType = VK_STRUCTURE_TYPE_PIPELINE_RASTERIZATION_STATE_CREATE_INFO,
 308                 .depthClampEnable = false,
 309                 .rasterizerDiscardEnable = false,
 310                 .polygonMode = VK_POLYGON_MODE_FILL,
 311                 .cullMode = VK_CULL_MODE_NONE,
 312                 .frontFace = VK_FRONT_FACE_COUNTER_CLOCKWISE,
 313         };
 314
 315         result = radv_graphics_pipeline_create(device_h,
 316                                                radv_pipeline_cache_to_handle(&device->meta_state.cache),
 317                                                &(VkGraphicsPipelineCreateInfo) {
 318                                                        .sType = VK_STRUCTURE_TYPE_GRAPHICS_PIPELINE_CREATE_INFO,
 319                                                        .stageCount = 2,
 320                                                        .pStages = stages,
 321
 322                                                        .pVertexInputState = &vi_state,
 323                                                        .pInputAssemblyState = &ia_state,
 324
 325                                                .pViewportState = &(VkPipelineViewportStateCreateInfo) {
 326                                                        .sType = VK_STRUCTURE_TYPE_PIPELINE_VIEWPORT_STATE_CREATE_INFO,
 327                                                        .viewportCount = 1,
 328                                                        .scissorCount = 1,
 329                                                },
 330                                                        .pRasterizationState = &rs_state,
 331                                                .pMultisampleState = &(VkPipelineMultisampleStateCreateInfo) {
 332                                                        .sType = VK_STRUCTURE_TYPE_PIPELINE_MULTISAMPLE_STATE_CREATE_INFO,
 333                                                        .rasterizationSamples = 1,
 334                                                        .sampleShadingEnable = false,
 335                                                        .pSampleMask = NULL,
 336                                                        .alphaToCoverageEnable = false,
 337                                                        .alphaToOneEnable = false,
 338                                                },
 339                                                 .pColorBlendState = &blend_state,
 340                                                 .pDynamicState = &(VkPipelineDynamicStateCreateInfo) {
 341                                                         .sType = VK_STRUCTURE_TYPE_PIPELINE_DYNAMIC_STATE_CREATE_INFO,
 342                                                         .dynamicStateCount = 2,
 343                                                         .pDynamicStates = (VkDynamicState[]) {
 344                                                                 VK_DYNAMIC_STATE_VIEWPORT,
 345                                                                 VK_DYNAMIC_STATE_SCISSOR,
 346                                                         },
 347                                                 },
 348                                                 .layout = layout,
 349                                                 .renderPass = device->meta_state.fast_clear_flush.pass,
 350                                                 .subpass = 0,
 351                                                },
 352                                                &(struct radv_graphics_pipeline_create_info) {
 353                                                        .use_rectlist = true,
 354                                                        .custom_blend_mode = V_028808_CB_ELIMINATE_FAST_CLEAR,
 355                                                },
 356                                                &device->meta_state.alloc,
 357                                                &device->meta_state.fast_clear_flush.cmask_eliminate_pipeline);
 358         if (result != VK_SUCCESS)
 359                 goto cleanup;
 360
 361         result = radv_graphics_pipeline_create(device_h,
 362                                                radv_pipeline_cache_to_handle(&device->meta_state.cache),
 363                                                &(VkGraphicsPipelineCreateInfo) {
 364                                                        .sType = VK_STRUCTURE_TYPE_GRAPHICS_PIPELINE_CREATE_INFO,
 365                                                        .stageCount = 2,
 366                                                        .pStages = stages,
 367
 368                                                        .pVertexInputState = &vi_state,
 369                                                        .pInputAssemblyState = &ia_state,
 370
 371                                                .pViewportState = &(VkPipelineViewportStateCreateInfo) {
 372                                                        .sType = VK_STRUCTURE_TYPE_PIPELINE_VIEWPORT_STATE_CREATE_INFO,
 373                                                        .viewportCount = 1,
 374                                                        .scissorCount = 1,
 375                                                },
 376                                                        .pRasterizationState = &rs_state,
 377                                                .pMultisampleState = &(VkPipelineMultisampleStateCreateInfo) {
 378                                                        .sType = VK_STRUCTURE_TYPE_PIPELINE_MULTISAMPLE_STATE_CREATE_INFO,
 379                                                        .rasterizationSamples = 1,
 380                                                        .sampleShadingEnable = false,
 381                                                        .pSampleMask = NULL,
 382                                                        .alphaToCoverageEnable = false,
 383                                                        .alphaToOneEnable = false,
 384                                                },
 385                                                 .pColorBlendState = &blend_state,
 386                                                 .pDynamicState = &(VkPipelineDynamicStateCreateInfo) {
 387                                                         .sType = VK_STRUCTURE_TYPE_PIPELINE_DYNAMIC_STATE_CREATE_INFO,
 388                                                         .dynamicStateCount = 2,
 389                                                         .pDynamicStates = (VkDynamicState[]) {
 390                                                                 VK_DYNAMIC_STATE_VIEWPORT,
 391                                                                 VK_DYNAMIC_STATE_SCISSOR,
 392                                                         },
 393                                                 },
 394                                                 .layout = layout,
 395                                                 .renderPass = device->meta_state.fast_clear_flush.pass,
 396                                                 .subpass = 0,
 397                                                },
 398                                                &(struct radv_graphics_pipeline_create_info) {
 399                                                        .use_rectlist = true,
 400                                                        .custom_blend_mode = V_028808_CB_FMASK_DECOMPRESS,
 401                                                },
 402                                                &device->meta_state.alloc,
 403                                                &device->meta_state.fast_clear_flush.fmask_decompress_pipeline);
 404         if (result != VK_SUCCESS)
 405                 goto cleanup;
 406
 407         result = radv_graphics_pipeline_create(device_h,
 408                                                radv_pipeline_cache_to_handle(&device->meta_state.cache),
 409                                                &(VkGraphicsPipelineCreateInfo) {
 410                                                        .sType = VK_STRUCTURE_TYPE_GRAPHICS_PIPELINE_CREATE_INFO,
 411                                                        .stageCount = 2,
 412                                                        .pStages = stages,
 413
 414                                                        .pVertexInputState = &vi_state,
 415                                                        .pInputAssemblyState = &ia_state,
 416
 417                                                .pViewportState = &(VkPipelineViewportStateCreateInfo) {
 418                                                        .sType = VK_STRUCTURE_TYPE_PIPELINE_VIEWPORT_STATE_CREATE_INFO,
 419                                                        .viewportCount = 1,
 420                                                        .scissorCount = 1,
 421                                                },
 422                                                        .pRasterizationState = &rs_state,
 423                                                .pMultisampleState = &(VkPipelineMultisampleStateCreateInfo) {
 424                                                        .sType = VK_STRUCTURE_TYPE_PIPELINE_MULTISAMPLE_STATE_CREATE_INFO,
 425                                                        .rasterizationSamples = 1,
 426                                                        .sampleShadingEnable = false,
 427                                                        .pSampleMask = NULL,
 428                                                        .alphaToCoverageEnable = false,
 429                                                        .alphaToOneEnable = false,
 430                                                },
 431                                                 .pColorBlendState = &blend_state,
 432                                                 .pDynamicState = &(VkPipelineDynamicStateCreateInfo) {
 433                                                         .sType = VK_STRUCTURE_TYPE_PIPELINE_DYNAMIC_STATE_CREATE_INFO,
 434                                                         .dynamicStateCount = 2,
 435                                                         .pDynamicStates = (VkDynamicState[]) {
 436                                                                 VK_DYNAMIC_STATE_VIEWPORT,
 437                                                                 VK_DYNAMIC_STATE_SCISSOR,
 438                                                         },
 439                                                 },
 440                                                 .layout = layout,
 441                                                 .renderPass = device->meta_state.fast_clear_flush.pass,
 442                                                 .subpass = 0,
 443                                                },
 444                                                &(struct radv_graphics_pipeline_create_info) {
 445                                                        .use_rectlist = true,
 446                                                        .custom_blend_mode = V_028808_CB_DCC_DECOMPRESS,
 447                                                },
 448                                                &device->meta_state.alloc,
 449                                                &device->meta_state.fast_clear_flush.dcc_decompress_pipeline);
 450         if (result != VK_SUCCESS)
 451                 goto cleanup;
 452
 453         goto cleanup;
 454
 455 cleanup:
 456         ralloc_free(fs_module.nir);
 457         return result;
 458 }
 459
 460 void
 461 radv_device_finish_meta_fast_clear_flush_state(struct radv_device *device)
 462 {
 463         struct radv_meta_state *state = &device->meta_state;
 464
 465         radv_DestroyPipeline(radv_device_to_handle(device),
 466                              state->fast_clear_flush.dcc_decompress_pipeline,
 467                              &state->alloc);
 468         radv_DestroyPipeline(radv_device_to_handle(device),
 469                              state->fast_clear_flush.fmask_decompress_pipeline,
 470                              &state->alloc);
 471         radv_DestroyPipeline(radv_device_to_handle(device),
 472                              state->fast_clear_flush.cmask_eliminate_pipeline,
 473                              &state->alloc);
 474         radv_DestroyRenderPass(radv_device_to_handle(device),
 475                                state->fast_clear_flush.pass, &state->alloc);
 476         radv_DestroyPipelineLayout(radv_device_to_handle(device),
 477                                    state->fast_clear_flush.p_layout,
 478                                    &state->alloc);
 479
 480         radv_DestroyPipeline(radv_device_to_handle(device),
 481                              state->fast_clear_flush.dcc_decompress_compute_pipeline,
 482                              &state->alloc);
 483         radv_DestroyPipelineLayout(radv_device_to_handle(device),
 484                                    state->fast_clear_flush.dcc_decompress_compute_p_layout,
 485                                    &state->alloc);
 486         radv_DestroyDescriptorSetLayout(radv_device_to_handle(device),
 487                                         state->fast_clear_flush.dcc_decompress_compute_ds_layout,
 488                                         &state->alloc);
 489 }
 490
 491 VkResult
 492 radv_device_init_meta_fast_clear_flush_state(struct radv_device *device)
 493 {
 494         VkResult res = VK_SUCCESS;
 495
 496         struct radv_shader_module vs_module = { .nir = radv_meta_build_nir_vs_generate_vertices() };
 497         if (!vs_module.nir) {
 498                 /* XXX: Need more accurate error */
 499                 res = VK_ERROR_OUT_OF_HOST_MEMORY;
 500                 goto fail;
 501         }
 502
 503         res = create_pass(device);
 504         if (res != VK_SUCCESS)
 505                 goto fail;
 506
 507         res = create_pipeline_layout(device,
 508                                      &device->meta_state.fast_clear_flush.p_layout);
 509         if (res != VK_SUCCESS)
 510                 goto fail;
 511
 512         VkShaderModule vs_module_h = radv_shader_module_to_handle(&vs_module);
 513         res = create_pipeline(device, vs_module_h,
 514                               device->meta_state.fast_clear_flush.p_layout);
 515         if (res != VK_SUCCESS)
 516                 goto fail;
 517
 518         res = create_dcc_compress_compute(device);
 519         if (res != VK_SUCCESS)
 520                 goto fail;
 521
 522         goto cleanup;
 523
 524 fail:
 525         radv_device_finish_meta_fast_clear_flush_state(device);
 526
 527 cleanup:
 528         ralloc_free(vs_module.nir);
 529
 530         return res;
 531 }
 532
 533 static void
 534 emit_fast_clear_flush(struct radv_cmd_buffer *cmd_buffer,
 535                       const VkExtent2D *resolve_extent,
 536                       VkPipeline pipeline)
 537 {
 538         VkCommandBuffer cmd_buffer_h = radv_cmd_buffer_to_handle(cmd_buffer);
 539
 540         radv_CmdBindPipeline(cmd_buffer_h, VK_PIPELINE_BIND_POINT_GRAPHICS,
 541                              pipeline);
 542
 543         radv_CmdSetViewport(radv_cmd_buffer_to_handle(cmd_buffer), 0, 1, &(VkViewport) {
 544                         .x = 0,
 545                         .y = 0,
 546                         .width = resolve_extent->width,
 547                         .height = resolve_extent->height,
 548                         .minDepth = 0.0f,
 549                         .maxDepth = 1.0f
 550                 });
 551
 552                 radv_CmdSetScissor(radv_cmd_buffer_to_handle(cmd_buffer), 0, 1, &(VkRect2D) {
 553                         .offset = (VkOffset2D) { 0, 0 },
 554                         .extent = (VkExtent2D) { resolve_extent->width, resolve_extent->height },
 555                 });
 556
 557         radv_CmdDraw(cmd_buffer_h, 3, 1, 0, 0);
 558         cmd_buffer->state.flush_bits |= (RADV_CMD_FLAG_FLUSH_AND_INV_CB |
 559                                          RADV_CMD_FLAG_FLUSH_AND_INV_CB_META);
 560 }
 561
 562 static void
 563 radv_emit_set_predication_state_from_image(struct radv_cmd_buffer *cmd_buffer,
 564                                       struct radv_image *image, bool value)
 565 {
 566         uint64_t va = 0;
 567
 568         if (value) {
 569                 va = radv_buffer_get_va(image->bo) + image->offset;
 570                 va += image->dcc_pred_offset;
 571         }
 572
 573         si_emit_set_predication_state(cmd_buffer, va);
 574 }
 575
 576 /**
 577  */
 578 static void
 579 radv_emit_color_decompress(struct radv_cmd_buffer *cmd_buffer,
 580                            struct radv_image *image,
 581                            const VkImageSubresourceRange *subresourceRange,
 582                            bool decompress_dcc)
 583 {
 584         struct radv_meta_saved_state saved_state;
 585         VkDevice device_h = radv_device_to_handle(cmd_buffer->device);
 586         VkCommandBuffer cmd_buffer_h = radv_cmd_buffer_to_handle(cmd_buffer);
 587         uint32_t layer_count = radv_get_layerCount(image, subresourceRange);
 588         VkPipeline pipeline;
 589
 590         assert(cmd_buffer->queue_family_index == RADV_QUEUE_GENERAL);
 591
 592         radv_meta_save(&saved_state, cmd_buffer,
 593                        RADV_META_SAVE_GRAPHICS_PIPELINE |
 594                        RADV_META_SAVE_PASS);
 595
 596         if (decompress_dcc && image->surface.dcc_size) {
 597                 pipeline = cmd_buffer->device->meta_state.fast_clear_flush.dcc_decompress_pipeline;
 598         } else if (image->fmask.size > 0) {
 599                pipeline = cmd_buffer->device->meta_state.fast_clear_flush.fmask_decompress_pipeline;
 600         } else {
 601                pipeline = cmd_buffer->device->meta_state.fast_clear_flush.cmask_eliminate_pipeline;
 602         }
 603
 604         if (!decompress_dcc && image->surface.dcc_size) {
 605                 radv_emit_set_predication_state_from_image(cmd_buffer, image, true);
 606                 cmd_buffer->state.predicating = true;
 607         }
 608         for (uint32_t layer = 0; layer < layer_count; ++layer) {
 609                 struct radv_image_view iview;
 610
 611                 radv_image_view_init(&iview, cmd_buffer->device,
 612                                      &(VkImageViewCreateInfo) {
 613                                              .sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO,
 614                                              .image = radv_image_to_handle(image),
 615                                              .viewType = radv_meta_get_view_type(image),
 616                                              .format = image->vk_format,
 617                                              .subresourceRange = {
 618                                                      .aspectMask = VK_IMAGE_ASPECT_COLOR_BIT,
 619                                                      .baseMipLevel = 0,
 620                                                      .levelCount = 1,
 621                                                      .baseArrayLayer = subresourceRange->baseArrayLayer + layer,
 622                                                      .layerCount = 1,
 623                                               },
 624                                      });
 625
 626                 VkFramebuffer fb_h;
 627                 radv_CreateFramebuffer(device_h,
 628                                 &(VkFramebufferCreateInfo) {
 629                                         .sType = VK_STRUCTURE_TYPE_FRAMEBUFFER_CREATE_INFO,
 630                                         .attachmentCount = 1,
 631                                         .pAttachments = (VkImageView[]) {
 632                                                 radv_image_view_to_handle(&iview)
 633                                         },
 634                                        .width = image->info.width,
 635                                        .height = image->info.height,
 636                                        .layers = 1
 637                                 },
 638                                 &cmd_buffer->pool->alloc,
 639                                 &fb_h);
 640
 641                 radv_CmdBeginRenderPass(cmd_buffer_h,
 642                                       &(VkRenderPassBeginInfo) {
 643                                               .sType = VK_STRUCTURE_TYPE_RENDER_PASS_BEGIN_INFO,
 644                                                       .renderPass = cmd_buffer->device->meta_state.fast_clear_flush.pass,
 645                                                       .framebuffer = fb_h,
 646                                                       .renderArea = {
 647                                                       .offset = {
 648                                                               0,
 649                                                               0,
 650                                                       },
 651                                                       .extent = {
 652                                                               image->info.width,
 653                                                               image->info.height,
 654                                                       }
 655                                               },
 656                                               .clearValueCount = 0,
 657                                               .pClearValues = NULL,
 658                                      },
 659                                      VK_SUBPASS_CONTENTS_INLINE);
 660
 661                 emit_fast_clear_flush(cmd_buffer,
 662                                       &(VkExtent2D) { image->info.width, image->info.height },
 663                                       pipeline);
 664                 radv_CmdEndRenderPass(cmd_buffer_h);
 665
 666                 radv_DestroyFramebuffer(device_h, fb_h,
 667                                         &cmd_buffer->pool->alloc);
 668
 669         }
 670         if (image->surface.dcc_size) {
 671                 cmd_buffer->state.predicating = false;
 672                 radv_emit_set_predication_state_from_image(cmd_buffer, image, false);
 673         }
 674         radv_meta_restore(&saved_state, cmd_buffer);
 675 }
 676
 677 void
 678 radv_fast_clear_flush_image_inplace(struct radv_cmd_buffer *cmd_buffer,
 679                                     struct radv_image *image,
 680                                     const VkImageSubresourceRange *subresourceRange)
 681 {
 682         radv_emit_color_decompress(cmd_buffer, image, subresourceRange, false);
 683 }
 684
 685 static void
 686 radv_decompress_dcc_gfx(struct radv_cmd_buffer *cmd_buffer,
 687                         struct radv_image *image,
 688                         const VkImageSubresourceRange *subresourceRange)
 689 {
 690         radv_emit_color_decompress(cmd_buffer, image, subresourceRange, true);
 691 }
 692
 693 static void
 694 radv_decompress_dcc_compute(struct radv_cmd_buffer *cmd_buffer,
 695                             struct radv_image *image,
 696                             const VkImageSubresourceRange *subresourceRange)
 697 {
 698         struct radv_meta_saved_state saved_state;
 699         struct radv_image_view iview = {0};
 700         struct radv_device *device = cmd_buffer->device;
 701
 702         /* This assumes the image is 2d with 1 layer and 1 mipmap level */
 703         struct radv_cmd_state *state = &cmd_buffer->state;
 704
 705         state->flush_bits |= RADV_CMD_FLAG_FLUSH_AND_INV_CB |
 706                              RADV_CMD_FLAG_FLUSH_AND_INV_CB_META;
 707
 708         radv_meta_save(&saved_state, cmd_buffer, RADV_META_SAVE_DESCRIPTORS |
 709                                                  RADV_META_SAVE_COMPUTE_PIPELINE);
 710
 711         radv_CmdBindPipeline(radv_cmd_buffer_to_handle(cmd_buffer),
 712                              VK_PIPELINE_BIND_POINT_COMPUTE,
 713                              device->meta_state.fast_clear_flush.dcc_decompress_compute_pipeline);
 714
 715         radv_image_view_init(&iview, cmd_buffer->device,
 716                              &(VkImageViewCreateInfo) {
 717                                      .sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO,
 718                                              .image = radv_image_to_handle(image),
 719                                              .viewType = VK_IMAGE_VIEW_TYPE_2D,
 720                                              .format = image->vk_format,
 721                                              .subresourceRange = {
 722                                                 .aspectMask = VK_IMAGE_ASPECT_COLOR_BIT,
 723                                                 .baseMipLevel = 0,
 724                                                 .levelCount = 1,
 725                                                 .baseArrayLayer = 0,
 726                                                 .layerCount = 1
 727                                              },
 728                              });
 729
 730         radv_meta_push_descriptor_set(cmd_buffer,
 731                                       VK_PIPELINE_BIND_POINT_COMPUTE,
 732                                       device->meta_state.fast_clear_flush.dcc_decompress_compute_p_layout,
 733                                       0, /* set */
 734                                       2, /* descriptorWriteCount */
 735                                       (VkWriteDescriptorSet[]) {
 736                                               {
 737                                                        .sType = VK_STRUCTURE_TYPE_WRITE_DESCRIPTOR_SET,
 738                                                        .dstBinding = 0,
 739                                                        .dstArrayElement = 0,
 740                                                        .descriptorCount = 1,
 741                                                        .descriptorType = VK_DESCRIPTOR_TYPE_SAMPLED_IMAGE,
 742                                                        .pImageInfo = (VkDescriptorImageInfo[]) {
 743                                                                {
 744                                                                        .sampler = VK_NULL_HANDLE,
 745                                                                        .imageView = radv_image_view_to_handle(&iview),
 746                                                                        .imageLayout = VK_IMAGE_LAYOUT_GENERAL,
 747                                                                },
 748                                                        }
 749                                               },
 750                                               {
 751                                                        .sType = VK_STRUCTURE_TYPE_WRITE_DESCRIPTOR_SET,
 752                                                        .dstBinding = 1,
 753                                                        .dstArrayElement = 0,
 754                                                        .descriptorCount = 1,
 755                                                        .descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_IMAGE,
 756                                                        .pImageInfo = (VkDescriptorImageInfo[]) {
 757                                                                {
 758                                                                        .sampler = VK_NULL_HANDLE,
 759                                                                        .imageView = radv_image_view_to_handle(&iview),
 760                                                                        .imageLayout = VK_IMAGE_LAYOUT_GENERAL,
 761                                                                },
 762                                                        }
 763                                               }
 764                                       });
 765
 766         radv_unaligned_dispatch(cmd_buffer, image->info.width, image->info.height, 1);
 767
 768         /* The fill buffer below does its own saving */
 769         radv_meta_restore(&saved_state, cmd_buffer);
 770
 771         state->flush_bits |= RADV_CMD_FLAG_CS_PARTIAL_FLUSH |
 772                              RADV_CMD_FLAG_INV_VMEM_L1;
 773
 774         state->flush_bits |= radv_fill_buffer(cmd_buffer, image->bo,
 775                                               image->offset + image->dcc_offset,
 776                                               image->surface.dcc_size, 0xffffffff);
 777
 778         state->flush_bits |= RADV_CMD_FLAG_FLUSH_AND_INV_CB |
 779                              RADV_CMD_FLAG_FLUSH_AND_INV_CB_META;
 780 }
 781
 782 void
 783 radv_decompress_dcc(struct radv_cmd_buffer *cmd_buffer,
 784                     struct radv_image *image,
 785                     const VkImageSubresourceRange *subresourceRange)
 786 {
 787         if (cmd_buffer->queue_family_index == RADV_QUEUE_GENERAL)
 788                 radv_decompress_dcc_gfx(cmd_buffer, image, subresourceRange);
 789         else
 790                 radv_decompress_dcc_compute(cmd_buffer, image, subresourceRange);
 791 }