src/amd/vulkan/radv_meta_decompress.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 static VkResult
  32 create_pass(struct radv_device *device,
  33             uint32_t samples,
  34             VkRenderPass *pass)
  35 {
  36         VkResult result;
  37         VkDevice device_h = radv_device_to_handle(device);
  38         const VkAllocationCallbacks *alloc = &device->meta_state.alloc;
  39         VkAttachmentDescription attachment;
  40
  41         attachment.flags = 0;
  42         attachment.format = VK_FORMAT_D32_SFLOAT_S8_UINT;
  43         attachment.samples = samples;
  44         attachment.loadOp = VK_ATTACHMENT_LOAD_OP_LOAD;
  45         attachment.storeOp = VK_ATTACHMENT_STORE_OP_STORE;
  46         attachment.stencilLoadOp = VK_ATTACHMENT_LOAD_OP_LOAD;
  47         attachment.stencilStoreOp = VK_ATTACHMENT_STORE_OP_STORE;
  48         attachment.initialLayout = VK_IMAGE_LAYOUT_DEPTH_STENCIL_ATTACHMENT_OPTIMAL;
  49         attachment.finalLayout = VK_IMAGE_LAYOUT_DEPTH_STENCIL_ATTACHMENT_OPTIMAL;
  50
  51         result = radv_CreateRenderPass(device_h,
  52                                        &(VkRenderPassCreateInfo) {
  53                                                .sType = VK_STRUCTURE_TYPE_RENDER_PASS_CREATE_INFO,
  54                                                        .attachmentCount = 1,
  55                                                        .pAttachments = &attachment,
  56                                                        .subpassCount = 1,
  57                                                         .pSubpasses = &(VkSubpassDescription) {
  58                                                        .pipelineBindPoint = VK_PIPELINE_BIND_POINT_GRAPHICS,
  59                                                        .inputAttachmentCount = 0,
  60                                                        .colorAttachmentCount = 0,
  61                                                        .pColorAttachments = NULL,
  62                                                        .pResolveAttachments = NULL,
  63                                                        .pDepthStencilAttachment = &(VkAttachmentReference) {
  64                                                                .attachment = 0,
  65                                                                .layout = VK_IMAGE_LAYOUT_DEPTH_STENCIL_ATTACHMENT_OPTIMAL,
  66                                                        },
  67                                                        .preserveAttachmentCount = 0,
  68                                                        .pPreserveAttachments = NULL,
  69                                                },
  70                                                                 .dependencyCount = 0,
  71                                                                    },
  72                                        alloc,
  73                                        pass);
  74
  75         return result;
  76 }
  77
  78 static VkResult
  79 create_pipeline_layout(struct radv_device *device, VkPipelineLayout *layout)
  80 {
  81         VkPipelineLayoutCreateInfo pl_create_info = {
  82                 .sType = VK_STRUCTURE_TYPE_PIPELINE_LAYOUT_CREATE_INFO,
  83                 .setLayoutCount = 0,
  84                 .pSetLayouts = NULL,
  85                 .pushConstantRangeCount = 0,
  86                 .pPushConstantRanges = NULL,
  87         };
  88
  89         return radv_CreatePipelineLayout(radv_device_to_handle(device),
  90                                          &pl_create_info,
  91                                          &device->meta_state.alloc,
  92                                          layout);
  93 }
  94
  95 static VkResult
  96 create_pipeline(struct radv_device *device,
  97                 VkShaderModule vs_module_h,
  98                 uint32_t samples,
  99                 VkRenderPass pass,
 100                 VkPipelineLayout layout,
 101                 VkPipeline *decompress_pipeline,
 102                 VkPipeline *resummarize_pipeline)
 103 {
 104         VkResult result;
 105         VkDevice device_h = radv_device_to_handle(device);
 106         struct radv_shader_module vs_module = {0};
 107
 108         mtx_lock(&device->meta_state.mtx);
 109         if (*decompress_pipeline) {
 110                 mtx_unlock(&device->meta_state.mtx);
 111                 return VK_SUCCESS;
 112         }
 113
 114         if (!vs_module_h) {
 115                 vs_module.nir = radv_meta_build_nir_vs_generate_vertices();
 116                 vs_module_h = radv_shader_module_to_handle(&vs_module);
 117         }
 118
 119         struct radv_shader_module fs_module = {
 120                 .nir = radv_meta_build_nir_fs_noop(),
 121         };
 122
 123         if (!fs_module.nir) {
 124                 /* XXX: Need more accurate error */
 125                 result = VK_ERROR_OUT_OF_HOST_MEMORY;
 126                 goto cleanup;
 127         }
 128
 129         const VkPipelineSampleLocationsStateCreateInfoEXT sample_locs_create_info = {
 130                 .sType = VK_STRUCTURE_TYPE_PIPELINE_SAMPLE_LOCATIONS_STATE_CREATE_INFO_EXT,
 131                 .sampleLocationsEnable = false,
 132         };
 133
 134         const VkGraphicsPipelineCreateInfo pipeline_create_info = {
 135                 .sType = VK_STRUCTURE_TYPE_GRAPHICS_PIPELINE_CREATE_INFO,
 136                 .stageCount = 2,
 137                 .pStages = (VkPipelineShaderStageCreateInfo[]) {
 138                        {
 139                                 .sType = VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_CREATE_INFO,
 140                                 .stage = VK_SHADER_STAGE_VERTEX_BIT,
 141                                 .module = vs_module_h,
 142                                 .pName = "main",
 143                         },
 144                         {
 145                                 .sType = VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_CREATE_INFO,
 146                                 .stage = VK_SHADER_STAGE_FRAGMENT_BIT,
 147                                 .module = radv_shader_module_to_handle(&fs_module),
 148                                 .pName = "main",
 149                         },
 150                 },
 151                 .pVertexInputState = &(VkPipelineVertexInputStateCreateInfo) {
 152                         .sType = VK_STRUCTURE_TYPE_PIPELINE_VERTEX_INPUT_STATE_CREATE_INFO,
 153                         .vertexBindingDescriptionCount = 0,
 154                         .vertexAttributeDescriptionCount = 0,
 155                 },
 156                 .pInputAssemblyState = &(VkPipelineInputAssemblyStateCreateInfo) {
 157                         .sType = VK_STRUCTURE_TYPE_PIPELINE_INPUT_ASSEMBLY_STATE_CREATE_INFO,
 158                         .topology = VK_PRIMITIVE_TOPOLOGY_TRIANGLE_STRIP,
 159                         .primitiveRestartEnable = false,
 160                 },
 161                 .pViewportState = &(VkPipelineViewportStateCreateInfo) {
 162                         .sType = VK_STRUCTURE_TYPE_PIPELINE_VIEWPORT_STATE_CREATE_INFO,
 163                         .viewportCount = 1,
 164                         .scissorCount = 1,
 165                 },
 166                 .pRasterizationState = &(VkPipelineRasterizationStateCreateInfo) {
 167                         .sType = VK_STRUCTURE_TYPE_PIPELINE_RASTERIZATION_STATE_CREATE_INFO,
 168                         .depthClampEnable = false,
 169                         .rasterizerDiscardEnable = false,
 170                         .polygonMode = VK_POLYGON_MODE_FILL,
 171                         .cullMode = VK_CULL_MODE_NONE,
 172                         .frontFace = VK_FRONT_FACE_COUNTER_CLOCKWISE,
 173                 },
 174                 .pMultisampleState = &(VkPipelineMultisampleStateCreateInfo) {
 175                         .sType = VK_STRUCTURE_TYPE_PIPELINE_MULTISAMPLE_STATE_CREATE_INFO,
 176                         .pNext = &sample_locs_create_info,
 177                         .rasterizationSamples = samples,
 178                         .sampleShadingEnable = false,
 179                         .pSampleMask = NULL,
 180                         .alphaToCoverageEnable = false,
 181                         .alphaToOneEnable = false,
 182                 },
 183                 .pColorBlendState = &(VkPipelineColorBlendStateCreateInfo) {
 184                         .sType = VK_STRUCTURE_TYPE_PIPELINE_COLOR_BLEND_STATE_CREATE_INFO,
 185                         .logicOpEnable = false,
 186                         .attachmentCount = 0,
 187                         .pAttachments = NULL,
 188                 },
 189                 .pDepthStencilState = &(VkPipelineDepthStencilStateCreateInfo) {
 190                         .sType = VK_STRUCTURE_TYPE_PIPELINE_DEPTH_STENCIL_STATE_CREATE_INFO,
 191                         .depthTestEnable = false,
 192                         .depthWriteEnable = false,
 193                         .depthBoundsTestEnable = false,
 194                         .stencilTestEnable = false,
 195                 },
 196                 .pDynamicState = &(VkPipelineDynamicStateCreateInfo) {
 197                         .sType = VK_STRUCTURE_TYPE_PIPELINE_DYNAMIC_STATE_CREATE_INFO,
 198                         .dynamicStateCount = 3,
 199                         .pDynamicStates = (VkDynamicState[]) {
 200                                 VK_DYNAMIC_STATE_VIEWPORT,
 201                                 VK_DYNAMIC_STATE_SCISSOR,
 202                                 VK_DYNAMIC_STATE_SAMPLE_LOCATIONS_EXT,
 203                         },
 204                 },
 205                 .layout = layout,
 206                 .renderPass = pass,
 207                 .subpass = 0,
 208         };
 209
 210         result = radv_graphics_pipeline_create(device_h,
 211                                                radv_pipeline_cache_to_handle(&device->meta_state.cache),
 212                                                &pipeline_create_info,
 213                                                &(struct radv_graphics_pipeline_create_info) {
 214                                                         .use_rectlist = true,
 215                                                         .db_flush_depth_inplace = true,
 216                                                         .db_flush_stencil_inplace = true,
 217                                                },
 218                                                &device->meta_state.alloc,
 219                                                decompress_pipeline);
 220         if (result != VK_SUCCESS)
 221                 goto cleanup;
 222
 223         result = radv_graphics_pipeline_create(device_h,
 224                                                radv_pipeline_cache_to_handle(&device->meta_state.cache),
 225                                                &pipeline_create_info,
 226                                                &(struct radv_graphics_pipeline_create_info) {
 227                                                         .use_rectlist = true,
 228                                                         .db_flush_depth_inplace = true,
 229                                                         .db_flush_stencil_inplace = true,
 230                                                         .db_resummarize = true,
 231                                                },
 232                                                &device->meta_state.alloc,
 233                                                resummarize_pipeline);
 234         if (result != VK_SUCCESS)
 235                 goto cleanup;
 236
 237         goto cleanup;
 238
 239 cleanup:
 240         ralloc_free(fs_module.nir);
 241         if (vs_module.nir)
 242                 ralloc_free(vs_module.nir);
 243         mtx_unlock(&device->meta_state.mtx);
 244         return result;
 245 }
 246
 247 void
 248 radv_device_finish_meta_depth_decomp_state(struct radv_device *device)
 249 {
 250         struct radv_meta_state *state = &device->meta_state;
 251
 252         for (uint32_t i = 0; i < ARRAY_SIZE(state->depth_decomp); ++i) {
 253                 radv_DestroyRenderPass(radv_device_to_handle(device),
 254                                        state->depth_decomp[i].pass,
 255                                        &state->alloc);
 256                 radv_DestroyPipelineLayout(radv_device_to_handle(device),
 257                                            state->depth_decomp[i].p_layout,
 258                                            &state->alloc);
 259                 radv_DestroyPipeline(radv_device_to_handle(device),
 260                                      state->depth_decomp[i].decompress_pipeline,
 261                                      &state->alloc);
 262                 radv_DestroyPipeline(radv_device_to_handle(device),
 263                                      state->depth_decomp[i].resummarize_pipeline,
 264                                      &state->alloc);
 265         }
 266 }
 267
 268 VkResult
 269 radv_device_init_meta_depth_decomp_state(struct radv_device *device, bool on_demand)
 270 {
 271         struct radv_meta_state *state = &device->meta_state;
 272         VkResult res = VK_SUCCESS;
 273
 274         struct radv_shader_module vs_module = { .nir = radv_meta_build_nir_vs_generate_vertices() };
 275         if (!vs_module.nir) {
 276                 /* XXX: Need more accurate error */
 277                 res = VK_ERROR_OUT_OF_HOST_MEMORY;
 278                 goto fail;
 279         }
 280
 281         VkShaderModule vs_module_h = radv_shader_module_to_handle(&vs_module);
 282
 283         for (uint32_t i = 0; i < ARRAY_SIZE(state->depth_decomp); ++i) {
 284                 uint32_t samples = 1 << i;
 285
 286                 res = create_pass(device, samples, &state->depth_decomp[i].pass);
 287                 if (res != VK_SUCCESS)
 288                         goto fail;
 289
 290                 res = create_pipeline_layout(device,
 291                                              &state->depth_decomp[i].p_layout);
 292                 if (res != VK_SUCCESS)
 293                         goto fail;
 294
 295                 if (on_demand)
 296                         continue;
 297
 298                 res = create_pipeline(device, vs_module_h, samples,
 299                                       state->depth_decomp[i].pass,
 300                                       state->depth_decomp[i].p_layout,
 301                                       &state->depth_decomp[i].decompress_pipeline,
 302                                       &state->depth_decomp[i].resummarize_pipeline);
 303                 if (res != VK_SUCCESS)
 304                         goto fail;
 305         }
 306
 307         goto cleanup;
 308
 309 fail:
 310         radv_device_finish_meta_depth_decomp_state(device);
 311
 312 cleanup:
 313         ralloc_free(vs_module.nir);
 314
 315         return res;
 316 }
 317
 318 enum radv_depth_op {
 319         DEPTH_DECOMPRESS,
 320         DEPTH_RESUMMARIZE,
 321 };
 322
 323 static VkPipeline *
 324 radv_get_depth_pipeline(struct radv_cmd_buffer *cmd_buffer,
 325                         struct radv_image *image, enum radv_depth_op op)
 326 {
 327         struct radv_meta_state *state = &cmd_buffer->device->meta_state;
 328         uint32_t samples = image->info.samples;
 329         uint32_t samples_log2 = ffs(samples) - 1;
 330         VkPipeline *pipeline;
 331
 332         if (!state->depth_decomp[samples_log2].decompress_pipeline) {
 333                 VkResult ret;
 334
 335                 ret = create_pipeline(cmd_buffer->device, VK_NULL_HANDLE, samples,
 336                                       state->depth_decomp[samples_log2].pass,
 337                                       state->depth_decomp[samples_log2].p_layout,
 338                                       &state->depth_decomp[samples_log2].decompress_pipeline,
 339                                       &state->depth_decomp[samples_log2].resummarize_pipeline);
 340                 if (ret != VK_SUCCESS) {
 341                         cmd_buffer->record_result = ret;
 342                         return NULL;
 343                 }
 344        }
 345
 346         switch (op) {
 347         case DEPTH_DECOMPRESS:
 348                 pipeline = &state->depth_decomp[samples_log2].decompress_pipeline;
 349                 break;
 350         case DEPTH_RESUMMARIZE:
 351                 pipeline = &state->depth_decomp[samples_log2].resummarize_pipeline;
 352                 break;
 353         default:
 354                 unreachable("unknown operation");
 355         }
 356
 357         return pipeline;
 358 }
 359
 360 static void
 361 radv_process_depth_image_layer(struct radv_cmd_buffer *cmd_buffer,
 362                                struct radv_image *image,
 363                                const VkImageSubresourceRange *range,
 364                                int level, int layer)
 365 {
 366         struct radv_device *device = cmd_buffer->device;
 367         struct radv_meta_state *state = &device->meta_state;
 368         uint32_t samples_log2 = ffs(image->info.samples) - 1;
 369         struct radv_image_view iview;
 370         uint32_t width, height;
 371
 372         width = radv_minify(image->info.width, range->baseMipLevel + level);
 373         height = radv_minify(image->info.height, range->baseMipLevel + level);
 374
 375         radv_image_view_init(&iview, device,
 376                              &(VkImageViewCreateInfo) {
 377                                         .sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO,
 378                                         .image = radv_image_to_handle(image),
 379                                         .viewType = radv_meta_get_view_type(image),
 380                                         .format = image->vk_format,
 381                                         .subresourceRange = {
 382                                                 .aspectMask = VK_IMAGE_ASPECT_DEPTH_BIT,
 383                                                 .baseMipLevel = range->baseMipLevel + level,
 384                                                 .levelCount = 1,
 385                                                 .baseArrayLayer = range->baseArrayLayer + layer,
 386                                                 .layerCount = 1,
 387                                         },
 388                              }, NULL);
 389
 390
 391         VkFramebuffer fb_h;
 392         radv_CreateFramebuffer(radv_device_to_handle(device),
 393                                &(VkFramebufferCreateInfo) {
 394                                         .sType = VK_STRUCTURE_TYPE_FRAMEBUFFER_CREATE_INFO,
 395                                         .attachmentCount = 1,
 396                                                 .pAttachments = (VkImageView[]) {
 397                                                         radv_image_view_to_handle(&iview)
 398                                         },
 399                                         .width = width,
 400                                         .height = height,
 401                                         .layers = 1
 402                                }, &cmd_buffer->pool->alloc, &fb_h);
 403
 404         radv_CmdBeginRenderPass(radv_cmd_buffer_to_handle(cmd_buffer),
 405                                 &(VkRenderPassBeginInfo) {
 406                                         .sType = VK_STRUCTURE_TYPE_RENDER_PASS_BEGIN_INFO,
 407                                         .renderPass = state->depth_decomp[samples_log2].pass,
 408                                         .framebuffer = fb_h,
 409                                         .renderArea = {
 410                                                 .offset = {
 411                                                         0,
 412                                                         0,
 413                                                 },
 414                                                 .extent = {
 415                                                         width,
 416                                                         height,
 417                                                 }
 418                                         },
 419                                         .clearValueCount = 0,
 420                                         .pClearValues = NULL,
 421                                 },
 422                                 VK_SUBPASS_CONTENTS_INLINE);
 423
 424         radv_CmdDraw(radv_cmd_buffer_to_handle(cmd_buffer), 3, 1, 0, 0);
 425         radv_CmdEndRenderPass(radv_cmd_buffer_to_handle(cmd_buffer));
 426
 427         radv_DestroyFramebuffer(radv_device_to_handle(device), fb_h,
 428                                 &cmd_buffer->pool->alloc);
 429 }
 430
 431 static void radv_process_depth_image_inplace(struct radv_cmd_buffer *cmd_buffer,
 432                                              struct radv_image *image,
 433                                              VkImageSubresourceRange *subresourceRange,
 434                                              struct radv_sample_locations_state *sample_locs,
 435                                              enum radv_depth_op op)
 436 {
 437         struct radv_meta_saved_state saved_state;
 438         VkCommandBuffer cmd_buffer_h = radv_cmd_buffer_to_handle(cmd_buffer);
 439         VkPipeline *pipeline;
 440
 441         if (!radv_image_has_htile(image))
 442                 return;
 443
 444         radv_meta_save(&saved_state, cmd_buffer,
 445                        RADV_META_SAVE_GRAPHICS_PIPELINE |
 446                        RADV_META_SAVE_SAMPLE_LOCATIONS |
 447                        RADV_META_SAVE_PASS);
 448
 449         pipeline = radv_get_depth_pipeline(cmd_buffer, image, op);
 450
 451         radv_CmdBindPipeline(radv_cmd_buffer_to_handle(cmd_buffer),
 452                              VK_PIPELINE_BIND_POINT_GRAPHICS, *pipeline);
 453
 454         if (sample_locs) {
 455                 assert(image->flags & VK_IMAGE_CREATE_SAMPLE_LOCATIONS_COMPATIBLE_DEPTH_BIT_EXT);
 456
 457                 /* Set the sample locations specified during explicit or
 458                  * automatic layout transitions, otherwise the depth decompress
 459                  * pass uses the default HW locations.
 460                  */
 461                 radv_CmdSetSampleLocationsEXT(cmd_buffer_h, &(VkSampleLocationsInfoEXT) {
 462                         .sampleLocationsPerPixel = sample_locs->per_pixel,
 463                         .sampleLocationGridSize = sample_locs->grid_size,
 464                         .sampleLocationsCount = sample_locs->count,
 465                         .pSampleLocations = sample_locs->locations,
 466                 });
 467         }
 468
 469         for (uint32_t l = 0; l < radv_get_levelCount(image, subresourceRange); ++l) {
 470                 uint32_t width =
 471                         radv_minify(image->info.width,
 472                                     subresourceRange->baseMipLevel + l);
 473                 uint32_t height =
 474                         radv_minify(image->info.height,
 475                                     subresourceRange->baseMipLevel + l);
 476
 477                 radv_CmdSetViewport(cmd_buffer_h, 0, 1,
 478                                     &(VkViewport) {
 479                                         .x = 0,
 480                                         .y = 0,
 481                                         .width = width,
 482                                         .height = height,
 483                                         .minDepth = 0.0f,
 484                                         .maxDepth = 1.0f
 485                                     });
 486
 487                 radv_CmdSetScissor(cmd_buffer_h, 0, 1,
 488                                    &(VkRect2D) {
 489                                         .offset = { 0, 0 },
 490                                         .extent = { width, height },
 491                                    });
 492
 493                 for (uint32_t s = 0; s < radv_get_layerCount(image, subresourceRange); s++) {
 494                         radv_process_depth_image_layer(cmd_buffer, image,
 495                                                        subresourceRange, l, s);
 496                 }
 497         }
 498
 499         radv_meta_restore(&saved_state, cmd_buffer);
 500 }
 501
 502 void radv_decompress_depth_image_inplace(struct radv_cmd_buffer *cmd_buffer,
 503                                          struct radv_image *image,
 504                                          VkImageSubresourceRange *subresourceRange,
 505                                          struct radv_sample_locations_state *sample_locs)
 506 {
 507         assert(cmd_buffer->queue_family_index == RADV_QUEUE_GENERAL);
 508         radv_process_depth_image_inplace(cmd_buffer, image, subresourceRange,
 509                                          sample_locs, DEPTH_DECOMPRESS);
 510 }
 511
 512 void radv_resummarize_depth_image_inplace(struct radv_cmd_buffer *cmd_buffer,
 513                                          struct radv_image *image,
 514                                          VkImageSubresourceRange *subresourceRange,
 515                                          struct radv_sample_locations_state *sample_locs)
 516 {
 517         assert(cmd_buffer->queue_family_index == RADV_QUEUE_GENERAL);
 518         radv_process_depth_image_inplace(cmd_buffer, image, subresourceRange,
 519                                          sample_locs, DEPTH_RESUMMARIZE);
 520 }