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 enum radv_depth_op {
  32         DEPTH_DECOMPRESS,
  33         DEPTH_RESUMMARIZE,
  34 };
  35
  36 static VkResult
  37 create_pass(struct radv_device *device,
  38             uint32_t samples,
  39             VkRenderPass *pass)
  40 {
  41         VkResult result;
  42         VkDevice device_h = radv_device_to_handle(device);
  43         const VkAllocationCallbacks *alloc = &device->meta_state.alloc;
  44         VkAttachmentDescription attachment;
  45
  46         attachment.flags = 0;
  47         attachment.format = VK_FORMAT_D32_SFLOAT_S8_UINT;
  48         attachment.samples = samples;
  49         attachment.loadOp = VK_ATTACHMENT_LOAD_OP_LOAD;
  50         attachment.storeOp = VK_ATTACHMENT_STORE_OP_STORE;
  51         attachment.stencilLoadOp = VK_ATTACHMENT_LOAD_OP_LOAD;
  52         attachment.stencilStoreOp = VK_ATTACHMENT_STORE_OP_STORE;
  53         attachment.initialLayout = VK_IMAGE_LAYOUT_DEPTH_STENCIL_ATTACHMENT_OPTIMAL;
  54         attachment.finalLayout = VK_IMAGE_LAYOUT_DEPTH_STENCIL_ATTACHMENT_OPTIMAL;
  55
  56         result = radv_CreateRenderPass(device_h,
  57                                        &(VkRenderPassCreateInfo) {
  58                                                .sType = VK_STRUCTURE_TYPE_RENDER_PASS_CREATE_INFO,
  59                                                        .attachmentCount = 1,
  60                                                        .pAttachments = &attachment,
  61                                                        .subpassCount = 1,
  62                                                         .pSubpasses = &(VkSubpassDescription) {
  63                                                        .pipelineBindPoint = VK_PIPELINE_BIND_POINT_GRAPHICS,
  64                                                        .inputAttachmentCount = 0,
  65                                                        .colorAttachmentCount = 0,
  66                                                        .pColorAttachments = NULL,
  67                                                        .pResolveAttachments = NULL,
  68                                                        .pDepthStencilAttachment = &(VkAttachmentReference) {
  69                                                                .attachment = 0,
  70                                                                .layout = VK_IMAGE_LAYOUT_DEPTH_STENCIL_ATTACHMENT_OPTIMAL,
  71                                                        },
  72                                                        .preserveAttachmentCount = 0,
  73                                                        .pPreserveAttachments = NULL,
  74                                                },
  75                                                                 .dependencyCount = 0,
  76                                                                    },
  77                                        alloc,
  78                                        pass);
  79
  80         return result;
  81 }
  82
  83 static VkResult
  84 create_pipeline_layout(struct radv_device *device, VkPipelineLayout *layout)
  85 {
  86         VkPipelineLayoutCreateInfo pl_create_info = {
  87                 .sType = VK_STRUCTURE_TYPE_PIPELINE_LAYOUT_CREATE_INFO,
  88                 .setLayoutCount = 0,
  89                 .pSetLayouts = NULL,
  90                 .pushConstantRangeCount = 0,
  91                 .pPushConstantRanges = NULL,
  92         };
  93
  94         return radv_CreatePipelineLayout(radv_device_to_handle(device),
  95                                          &pl_create_info,
  96                                          &device->meta_state.alloc,
  97                                          layout);
  98 }
  99
 100 static VkResult
 101 create_pipeline(struct radv_device *device,
 102                 VkShaderModule vs_module_h,
 103                 uint32_t samples,
 104                 VkRenderPass pass,
 105                 VkPipelineLayout layout,
 106                 enum radv_depth_op op,
 107                 VkPipeline *pipeline)
 108 {
 109         VkResult result;
 110         VkDevice device_h = radv_device_to_handle(device);
 111         struct radv_shader_module vs_module = {0};
 112
 113         mtx_lock(&device->meta_state.mtx);
 114         if (*pipeline) {
 115                 mtx_unlock(&device->meta_state.mtx);
 116                 return VK_SUCCESS;
 117         }
 118
 119         if (!vs_module_h) {
 120                 vs_module.nir = radv_meta_build_nir_vs_generate_vertices();
 121                 vs_module_h = radv_shader_module_to_handle(&vs_module);
 122         }
 123
 124         struct radv_shader_module fs_module = {
 125                 .nir = radv_meta_build_nir_fs_noop(),
 126         };
 127
 128         if (!fs_module.nir) {
 129                 /* XXX: Need more accurate error */
 130                 result = VK_ERROR_OUT_OF_HOST_MEMORY;
 131                 goto cleanup;
 132         }
 133
 134         const VkPipelineSampleLocationsStateCreateInfoEXT sample_locs_create_info = {
 135                 .sType = VK_STRUCTURE_TYPE_PIPELINE_SAMPLE_LOCATIONS_STATE_CREATE_INFO_EXT,
 136                 .sampleLocationsEnable = false,
 137         };
 138
 139         const VkGraphicsPipelineCreateInfo pipeline_create_info = {
 140                 .sType = VK_STRUCTURE_TYPE_GRAPHICS_PIPELINE_CREATE_INFO,
 141                 .stageCount = 2,
 142                 .pStages = (VkPipelineShaderStageCreateInfo[]) {
 143                        {
 144                                 .sType = VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_CREATE_INFO,
 145                                 .stage = VK_SHADER_STAGE_VERTEX_BIT,
 146                                 .module = vs_module_h,
 147                                 .pName = "main",
 148                         },
 149                         {
 150                                 .sType = VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_CREATE_INFO,
 151                                 .stage = VK_SHADER_STAGE_FRAGMENT_BIT,
 152                                 .module = radv_shader_module_to_handle(&fs_module),
 153                                 .pName = "main",
 154                         },
 155                 },
 156                 .pVertexInputState = &(VkPipelineVertexInputStateCreateInfo) {
 157                         .sType = VK_STRUCTURE_TYPE_PIPELINE_VERTEX_INPUT_STATE_CREATE_INFO,
 158                         .vertexBindingDescriptionCount = 0,
 159                         .vertexAttributeDescriptionCount = 0,
 160                 },
 161                 .pInputAssemblyState = &(VkPipelineInputAssemblyStateCreateInfo) {
 162                         .sType = VK_STRUCTURE_TYPE_PIPELINE_INPUT_ASSEMBLY_STATE_CREATE_INFO,
 163                         .topology = VK_PRIMITIVE_TOPOLOGY_TRIANGLE_STRIP,
 164                         .primitiveRestartEnable = false,
 165                 },
 166                 .pViewportState = &(VkPipelineViewportStateCreateInfo) {
 167                         .sType = VK_STRUCTURE_TYPE_PIPELINE_VIEWPORT_STATE_CREATE_INFO,
 168                         .viewportCount = 1,
 169                         .scissorCount = 1,
 170                 },
 171                 .pRasterizationState = &(VkPipelineRasterizationStateCreateInfo) {
 172                         .sType = VK_STRUCTURE_TYPE_PIPELINE_RASTERIZATION_STATE_CREATE_INFO,
 173                         .depthClampEnable = false,
 174                         .rasterizerDiscardEnable = false,
 175                         .polygonMode = VK_POLYGON_MODE_FILL,
 176                         .cullMode = VK_CULL_MODE_NONE,
 177                         .frontFace = VK_FRONT_FACE_COUNTER_CLOCKWISE,
 178                 },
 179                 .pMultisampleState = &(VkPipelineMultisampleStateCreateInfo) {
 180                         .sType = VK_STRUCTURE_TYPE_PIPELINE_MULTISAMPLE_STATE_CREATE_INFO,
 181                         .pNext = &sample_locs_create_info,
 182                         .rasterizationSamples = samples,
 183                         .sampleShadingEnable = false,
 184                         .pSampleMask = NULL,
 185                         .alphaToCoverageEnable = false,
 186                         .alphaToOneEnable = false,
 187                 },
 188                 .pColorBlendState = &(VkPipelineColorBlendStateCreateInfo) {
 189                         .sType = VK_STRUCTURE_TYPE_PIPELINE_COLOR_BLEND_STATE_CREATE_INFO,
 190                         .logicOpEnable = false,
 191                         .attachmentCount = 0,
 192                         .pAttachments = NULL,
 193                 },
 194                 .pDepthStencilState = &(VkPipelineDepthStencilStateCreateInfo) {
 195                         .sType = VK_STRUCTURE_TYPE_PIPELINE_DEPTH_STENCIL_STATE_CREATE_INFO,
 196                         .depthTestEnable = false,
 197                         .depthWriteEnable = false,
 198                         .depthBoundsTestEnable = false,
 199                         .stencilTestEnable = false,
 200                 },
 201                 .pDynamicState = &(VkPipelineDynamicStateCreateInfo) {
 202                         .sType = VK_STRUCTURE_TYPE_PIPELINE_DYNAMIC_STATE_CREATE_INFO,
 203                         .dynamicStateCount = 3,
 204                         .pDynamicStates = (VkDynamicState[]) {
 205                                 VK_DYNAMIC_STATE_VIEWPORT,
 206                                 VK_DYNAMIC_STATE_SCISSOR,
 207                                 VK_DYNAMIC_STATE_SAMPLE_LOCATIONS_EXT,
 208                         },
 209                 },
 210                 .layout = layout,
 211                 .renderPass = pass,
 212                 .subpass = 0,
 213         };
 214
 215         struct radv_graphics_pipeline_create_info extra = {
 216                 .use_rectlist = true,
 217                 .db_flush_depth_inplace = true,
 218                 .db_flush_stencil_inplace = true,
 219                 .db_resummarize = op == DEPTH_RESUMMARIZE,
 220         };
 221
 222         result = radv_graphics_pipeline_create(device_h,
 223                                                radv_pipeline_cache_to_handle(&device->meta_state.cache),
 224                                                &pipeline_create_info, &extra,
 225                                                &device->meta_state.alloc,
 226                                                pipeline);
 227
 228 cleanup:
 229         ralloc_free(fs_module.nir);
 230         if (vs_module.nir)
 231                 ralloc_free(vs_module.nir);
 232         mtx_unlock(&device->meta_state.mtx);
 233         return result;
 234 }
 235
 236 void
 237 radv_device_finish_meta_depth_decomp_state(struct radv_device *device)
 238 {
 239         struct radv_meta_state *state = &device->meta_state;
 240
 241         for (uint32_t i = 0; i < ARRAY_SIZE(state->depth_decomp); ++i) {
 242                 radv_DestroyRenderPass(radv_device_to_handle(device),
 243                                        state->depth_decomp[i].pass,
 244                                        &state->alloc);
 245                 radv_DestroyPipelineLayout(radv_device_to_handle(device),
 246                                            state->depth_decomp[i].p_layout,
 247                                            &state->alloc);
 248                 radv_DestroyPipeline(radv_device_to_handle(device),
 249                                      state->depth_decomp[i].decompress_pipeline,
 250                                      &state->alloc);
 251                 radv_DestroyPipeline(radv_device_to_handle(device),
 252                                      state->depth_decomp[i].resummarize_pipeline,
 253                                      &state->alloc);
 254         }
 255 }
 256
 257 VkResult
 258 radv_device_init_meta_depth_decomp_state(struct radv_device *device, bool on_demand)
 259 {
 260         struct radv_meta_state *state = &device->meta_state;
 261         VkResult res = VK_SUCCESS;
 262
 263         struct radv_shader_module vs_module = { .nir = radv_meta_build_nir_vs_generate_vertices() };
 264         if (!vs_module.nir) {
 265                 /* XXX: Need more accurate error */
 266                 res = VK_ERROR_OUT_OF_HOST_MEMORY;
 267                 goto fail;
 268         }
 269
 270         VkShaderModule vs_module_h = radv_shader_module_to_handle(&vs_module);
 271
 272         for (uint32_t i = 0; i < ARRAY_SIZE(state->depth_decomp); ++i) {
 273                 uint32_t samples = 1 << i;
 274
 275                 res = create_pass(device, samples, &state->depth_decomp[i].pass);
 276                 if (res != VK_SUCCESS)
 277                         goto fail;
 278
 279                 res = create_pipeline_layout(device,
 280                                              &state->depth_decomp[i].p_layout);
 281                 if (res != VK_SUCCESS)
 282                         goto fail;
 283
 284                 if (on_demand)
 285                         continue;
 286
 287                 res = create_pipeline(device, vs_module_h, samples,
 288                                       state->depth_decomp[i].pass,
 289                                       state->depth_decomp[i].p_layout,
 290                                       DEPTH_DECOMPRESS,
 291                                       &state->depth_decomp[i].decompress_pipeline);
 292                 if (res != VK_SUCCESS)
 293                         goto fail;
 294
 295                 res = create_pipeline(device, vs_module_h, samples,
 296                                       state->depth_decomp[i].pass,
 297                                       state->depth_decomp[i].p_layout,
 298                                       DEPTH_RESUMMARIZE,
 299                                       &state->depth_decomp[i].resummarize_pipeline);
 300                 if (res != VK_SUCCESS)
 301                         goto fail;
 302         }
 303
 304         goto cleanup;
 305
 306 fail:
 307         radv_device_finish_meta_depth_decomp_state(device);
 308
 309 cleanup:
 310         ralloc_free(vs_module.nir);
 311
 312         return res;
 313 }
 314
 315 static VkPipeline *
 316 radv_get_depth_pipeline(struct radv_cmd_buffer *cmd_buffer,
 317                         struct radv_image *image, enum radv_depth_op op)
 318 {
 319         struct radv_meta_state *state = &cmd_buffer->device->meta_state;
 320         uint32_t samples = image->info.samples;
 321         uint32_t samples_log2 = ffs(samples) - 1;
 322         VkPipeline *pipeline;
 323
 324         if (!state->depth_decomp[samples_log2].decompress_pipeline) {
 325                 VkResult ret;
 326
 327                 ret = create_pipeline(cmd_buffer->device, VK_NULL_HANDLE, samples,
 328                                       state->depth_decomp[samples_log2].pass,
 329                                       state->depth_decomp[samples_log2].p_layout,
 330                                       DEPTH_DECOMPRESS,
 331                                       &state->depth_decomp[samples_log2].decompress_pipeline);
 332                 if (ret != VK_SUCCESS) {
 333                         cmd_buffer->record_result = ret;
 334                         return NULL;
 335                 }
 336
 337                 ret = create_pipeline(cmd_buffer->device, VK_NULL_HANDLE, samples,
 338                                       state->depth_decomp[samples_log2].pass,
 339                                       state->depth_decomp[samples_log2].p_layout,
 340                                       DEPTH_RESUMMARIZE,
 341                                       &state->depth_decomp[samples_log2].resummarize_pipeline);
 342                 if (ret != VK_SUCCESS) {
 343                         cmd_buffer->record_result = ret;
 344                         return NULL;
 345                 }
 346        }
 347
 348         switch (op) {
 349         case DEPTH_DECOMPRESS:
 350                 pipeline = &state->depth_decomp[samples_log2].decompress_pipeline;
 351                 break;
 352         case DEPTH_RESUMMARIZE:
 353                 pipeline = &state->depth_decomp[samples_log2].resummarize_pipeline;
 354                 break;
 355         default:
 356                 unreachable("unknown operation");
 357         }
 358
 359         return pipeline;
 360 }
 361
 362 static void
 363 radv_process_depth_image_layer(struct radv_cmd_buffer *cmd_buffer,
 364                                struct radv_image *image,
 365                                const VkImageSubresourceRange *range,
 366                                int level, int layer)
 367 {
 368         struct radv_device *device = cmd_buffer->device;
 369         struct radv_meta_state *state = &device->meta_state;
 370         uint32_t samples_log2 = ffs(image->info.samples) - 1;
 371         struct radv_image_view iview;
 372         uint32_t width, height;
 373
 374         width = radv_minify(image->info.width, range->baseMipLevel + level);
 375         height = radv_minify(image->info.height, range->baseMipLevel + level);
 376
 377         radv_image_view_init(&iview, device,
 378                              &(VkImageViewCreateInfo) {
 379                                         .sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO,
 380                                         .image = radv_image_to_handle(image),
 381                                         .viewType = radv_meta_get_view_type(image),
 382                                         .format = image->vk_format,
 383                                         .subresourceRange = {
 384                                                 .aspectMask = VK_IMAGE_ASPECT_DEPTH_BIT,
 385                                                 .baseMipLevel = range->baseMipLevel + level,
 386                                                 .levelCount = 1,
 387                                                 .baseArrayLayer = range->baseArrayLayer + layer,
 388                                                 .layerCount = 1,
 389                                         },
 390                              }, NULL);
 391
 392
 393         VkFramebuffer fb_h;
 394         radv_CreateFramebuffer(radv_device_to_handle(device),
 395                                &(VkFramebufferCreateInfo) {
 396                                         .sType = VK_STRUCTURE_TYPE_FRAMEBUFFER_CREATE_INFO,
 397                                         .attachmentCount = 1,
 398                                                 .pAttachments = (VkImageView[]) {
 399                                                         radv_image_view_to_handle(&iview)
 400                                         },
 401                                         .width = width,
 402                                         .height = height,
 403                                         .layers = 1
 404                                }, &cmd_buffer->pool->alloc, &fb_h);
 405
 406         radv_CmdBeginRenderPass(radv_cmd_buffer_to_handle(cmd_buffer),
 407                                 &(VkRenderPassBeginInfo) {
 408                                         .sType = VK_STRUCTURE_TYPE_RENDER_PASS_BEGIN_INFO,
 409                                         .renderPass = state->depth_decomp[samples_log2].pass,
 410                                         .framebuffer = fb_h,
 411                                         .renderArea = {
 412                                                 .offset = {
 413                                                         0,
 414                                                         0,
 415                                                 },
 416                                                 .extent = {
 417                                                         width,
 418                                                         height,
 419                                                 }
 420                                         },
 421                                         .clearValueCount = 0,
 422                                         .pClearValues = NULL,
 423                                 },
 424                                 VK_SUBPASS_CONTENTS_INLINE);
 425
 426         radv_CmdDraw(radv_cmd_buffer_to_handle(cmd_buffer), 3, 1, 0, 0);
 427         radv_CmdEndRenderPass(radv_cmd_buffer_to_handle(cmd_buffer));
 428
 429         radv_DestroyFramebuffer(radv_device_to_handle(device), fb_h,
 430                                 &cmd_buffer->pool->alloc);
 431 }
 432
 433 static void radv_process_depth_image_inplace(struct radv_cmd_buffer *cmd_buffer,
 434                                              struct radv_image *image,
 435                                              const VkImageSubresourceRange *subresourceRange,
 436                                              struct radv_sample_locations_state *sample_locs,
 437                                              enum radv_depth_op op)
 438 {
 439         struct radv_meta_saved_state saved_state;
 440         VkCommandBuffer cmd_buffer_h = radv_cmd_buffer_to_handle(cmd_buffer);
 441         VkPipeline *pipeline;
 442
 443         if (!radv_image_has_htile(image))
 444                 return;
 445
 446         radv_meta_save(&saved_state, cmd_buffer,
 447                        RADV_META_SAVE_GRAPHICS_PIPELINE |
 448                        RADV_META_SAVE_SAMPLE_LOCATIONS |
 449                        RADV_META_SAVE_PASS);
 450
 451         pipeline = radv_get_depth_pipeline(cmd_buffer, image, op);
 452
 453         radv_CmdBindPipeline(radv_cmd_buffer_to_handle(cmd_buffer),
 454                              VK_PIPELINE_BIND_POINT_GRAPHICS, *pipeline);
 455
 456         if (sample_locs) {
 457                 assert(image->flags & VK_IMAGE_CREATE_SAMPLE_LOCATIONS_COMPATIBLE_DEPTH_BIT_EXT);
 458
 459                 /* Set the sample locations specified during explicit or
 460                  * automatic layout transitions, otherwise the depth decompress
 461                  * pass uses the default HW locations.
 462                  */
 463                 radv_CmdSetSampleLocationsEXT(cmd_buffer_h, &(VkSampleLocationsInfoEXT) {
 464                         .sampleLocationsPerPixel = sample_locs->per_pixel,
 465                         .sampleLocationGridSize = sample_locs->grid_size,
 466                         .sampleLocationsCount = sample_locs->count,
 467                         .pSampleLocations = sample_locs->locations,
 468                 });
 469         }
 470
 471         for (uint32_t l = 0; l < radv_get_levelCount(image, subresourceRange); ++l) {
 472                 uint32_t width =
 473                         radv_minify(image->info.width,
 474                                     subresourceRange->baseMipLevel + l);
 475                 uint32_t height =
 476                         radv_minify(image->info.height,
 477                                     subresourceRange->baseMipLevel + l);
 478
 479                 radv_CmdSetViewport(cmd_buffer_h, 0, 1,
 480                                     &(VkViewport) {
 481                                         .x = 0,
 482                                         .y = 0,
 483                                         .width = width,
 484                                         .height = height,
 485                                         .minDepth = 0.0f,
 486                                         .maxDepth = 1.0f
 487                                     });
 488
 489                 radv_CmdSetScissor(cmd_buffer_h, 0, 1,
 490                                    &(VkRect2D) {
 491                                         .offset = { 0, 0 },
 492                                         .extent = { width, height },
 493                                    });
 494
 495                 for (uint32_t s = 0; s < radv_get_layerCount(image, subresourceRange); s++) {
 496                         radv_process_depth_image_layer(cmd_buffer, image,
 497                                                        subresourceRange, l, s);
 498                 }
 499         }
 500
 501         radv_meta_restore(&saved_state, cmd_buffer);
 502 }
 503
 504 void radv_decompress_depth_image_inplace(struct radv_cmd_buffer *cmd_buffer,
 505                                          struct radv_image *image,
 506                                          const VkImageSubresourceRange *subresourceRange,
 507                                          struct radv_sample_locations_state *sample_locs)
 508 {
 509         assert(cmd_buffer->queue_family_index == RADV_QUEUE_GENERAL);
 510         radv_process_depth_image_inplace(cmd_buffer, image, subresourceRange,
 511                                          sample_locs, DEPTH_DECOMPRESS);
 512 }
 513
 514 void radv_resummarize_depth_image_inplace(struct radv_cmd_buffer *cmd_buffer,
 515                                          struct radv_image *image,
 516                                          const VkImageSubresourceRange *subresourceRange,
 517                                          struct radv_sample_locations_state *sample_locs)
 518 {
 519         assert(cmd_buffer->queue_family_index == RADV_QUEUE_GENERAL);
 520         radv_process_depth_image_inplace(cmd_buffer, image, subresourceRange,
 521                                          sample_locs, DEPTH_RESUMMARIZE);
 522 }