src/amd/vulkan/radv_meta_resolve.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 "vk_format.h"
  30 #include "nir/nir_builder.h"
  31 #include "sid.h"
  32
  33 /* emit 0, 0, 0, 1 */
  34 static nir_shader *
  35 build_nir_fs(void)
  36 {
  37         const struct glsl_type *vec4 = glsl_vec4_type();
  38         nir_builder b;
  39         nir_variable *f_color; /* vec4, fragment output color */
  40
  41         nir_builder_init_simple_shader(&b, NULL, MESA_SHADER_FRAGMENT, NULL);
  42         b.shader->info.name = ralloc_asprintf(b.shader,
  43                                                "meta_resolve_fs");
  44
  45         f_color = nir_variable_create(b.shader, nir_var_shader_out, vec4,
  46                                       "f_color");
  47         f_color->data.location = FRAG_RESULT_DATA0;
  48         nir_store_var(&b, f_color, nir_imm_vec4(&b, 0.0, 0.0, 0.0, 1.0), 0xf);
  49
  50         return b.shader;
  51 }
  52
  53 static VkResult
  54 create_pass(struct radv_device *device, VkFormat vk_format, VkRenderPass *pass)
  55 {
  56         VkResult result;
  57         VkDevice device_h = radv_device_to_handle(device);
  58         const VkAllocationCallbacks *alloc = &device->meta_state.alloc;
  59         VkAttachmentDescription attachments[2];
  60         int i;
  61
  62         for (i = 0; i < 2; i++) {
  63                 attachments[i].format = vk_format;
  64                 attachments[i].samples = 1;
  65                 attachments[i].loadOp = VK_ATTACHMENT_LOAD_OP_LOAD;
  66                 attachments[i].storeOp = VK_ATTACHMENT_STORE_OP_STORE;
  67         }
  68         attachments[0].initialLayout = VK_IMAGE_LAYOUT_GENERAL;
  69         attachments[0].finalLayout = VK_IMAGE_LAYOUT_GENERAL;
  70         attachments[1].initialLayout = VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL;
  71         attachments[1].finalLayout = VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL;
  72
  73         result = radv_CreateRenderPass(device_h,
  74                                        &(VkRenderPassCreateInfo) {
  75                                                .sType = VK_STRUCTURE_TYPE_RENDER_PASS_CREATE_INFO,
  76                                                        .attachmentCount = 2,
  77                                                        .pAttachments = attachments,
  78                                                        .subpassCount = 1,
  79                                                                 .pSubpasses = &(VkSubpassDescription) {
  80                                                        .pipelineBindPoint = VK_PIPELINE_BIND_POINT_GRAPHICS,
  81                                                        .inputAttachmentCount = 0,
  82                                                        .colorAttachmentCount = 2,
  83                                                        .pColorAttachments = (VkAttachmentReference[]) {
  84                                                                {
  85                                                                        .attachment = 0,
  86                                                                        .layout = VK_IMAGE_LAYOUT_GENERAL,
  87                                                                },
  88                                                                {
  89                                                                        .attachment = 1,
  90                                                                        .layout = VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL,
  91                                                                },
  92                                                        },
  93                                                        .pResolveAttachments = NULL,
  94                                                        .pDepthStencilAttachment = &(VkAttachmentReference) {
  95                                                                .attachment = VK_ATTACHMENT_UNUSED,
  96                                                        },
  97                                                        .preserveAttachmentCount = 0,
  98                                                        .pPreserveAttachments = NULL,
  99                                                },
 100                                                                 .dependencyCount = 0,
 101                                                                          },
 102                                        alloc,
 103                                        pass);
 104
 105         return result;
 106 }
 107
 108 static VkResult
 109 create_pipeline(struct radv_device *device,
 110                  VkShaderModule vs_module_h,
 111                  VkPipeline *pipeline,
 112                  VkRenderPass pass)
 113 {
 114         VkResult result;
 115         VkDevice device_h = radv_device_to_handle(device);
 116
 117         struct radv_shader_module fs_module = {
 118                 .nir = build_nir_fs(),
 119         };
 120
 121         if (!fs_module.nir) {
 122                 /* XXX: Need more accurate error */
 123                 result = VK_ERROR_OUT_OF_HOST_MEMORY;
 124                 goto cleanup;
 125         }
 126
 127         VkPipelineLayoutCreateInfo pl_create_info = {
 128                 .sType = VK_STRUCTURE_TYPE_PIPELINE_LAYOUT_CREATE_INFO,
 129                 .setLayoutCount = 0,
 130                 .pSetLayouts = NULL,
 131                 .pushConstantRangeCount = 0,
 132                 .pPushConstantRanges = NULL,
 133         };
 134
 135         if (!device->meta_state.resolve.p_layout) {
 136                 result = radv_CreatePipelineLayout(radv_device_to_handle(device),
 137                                                    &pl_create_info,
 138                                                    &device->meta_state.alloc,
 139                                                    &device->meta_state.resolve.p_layout);
 140                 if (result != VK_SUCCESS)
 141                         goto cleanup;
 142         }
 143
 144         result = radv_graphics_pipeline_create(device_h,
 145                                                radv_pipeline_cache_to_handle(&device->meta_state.cache),
 146                                                &(VkGraphicsPipelineCreateInfo) {
 147                                                        .sType = VK_STRUCTURE_TYPE_GRAPHICS_PIPELINE_CREATE_INFO,
 148                                                        .stageCount = 2,
 149                                                        .pStages = (VkPipelineShaderStageCreateInfo[]) {
 150                                                        {
 151                                                                .sType = VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_CREATE_INFO,
 152                                                                .stage = VK_SHADER_STAGE_VERTEX_BIT,
 153                                                                .module = vs_module_h,
 154                                                                .pName = "main",
 155                                                        },
 156                                                        {
 157                                                                .sType = VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_CREATE_INFO,
 158                                                                .stage = VK_SHADER_STAGE_FRAGMENT_BIT,
 159                                                                .module = radv_shader_module_to_handle(&fs_module),
 160                                                                .pName = "main",
 161                                                        },
 162                                                },
 163                                                .pVertexInputState = &(VkPipelineVertexInputStateCreateInfo) {
 164                                                        .sType = VK_STRUCTURE_TYPE_PIPELINE_VERTEX_INPUT_STATE_CREATE_INFO,
 165                                                        .vertexBindingDescriptionCount = 0,
 166                                                        .vertexAttributeDescriptionCount = 0,
 167                                                },
 168                                                .pInputAssemblyState = &(VkPipelineInputAssemblyStateCreateInfo) {
 169                                                        .sType = VK_STRUCTURE_TYPE_PIPELINE_INPUT_ASSEMBLY_STATE_CREATE_INFO,
 170                                                        .topology = VK_PRIMITIVE_TOPOLOGY_TRIANGLE_STRIP,
 171                                                        .primitiveRestartEnable = false,
 172                                                },
 173                                                .pViewportState = &(VkPipelineViewportStateCreateInfo) {
 174                                                        .sType = VK_STRUCTURE_TYPE_PIPELINE_VIEWPORT_STATE_CREATE_INFO,
 175                                                        .viewportCount = 1,
 176                                                        .scissorCount = 1,
 177                                                },
 178                                                .pRasterizationState = &(VkPipelineRasterizationStateCreateInfo) {
 179                                                        .sType = VK_STRUCTURE_TYPE_PIPELINE_RASTERIZATION_STATE_CREATE_INFO,
 180                                                        .depthClampEnable = false,
 181                                                        .rasterizerDiscardEnable = false,
 182                                                        .polygonMode = VK_POLYGON_MODE_FILL,
 183                                                        .cullMode = VK_CULL_MODE_NONE,
 184                                                        .frontFace = VK_FRONT_FACE_COUNTER_CLOCKWISE,
 185                                                },
 186                                                .pMultisampleState = &(VkPipelineMultisampleStateCreateInfo) {
 187                                                        .sType = VK_STRUCTURE_TYPE_PIPELINE_MULTISAMPLE_STATE_CREATE_INFO,
 188                                                        .rasterizationSamples = 1,
 189                                                        .sampleShadingEnable = false,
 190                                                        .pSampleMask = NULL,
 191                                                        .alphaToCoverageEnable = false,
 192                                                        .alphaToOneEnable = false,
 193                                                },
 194                                                .pColorBlendState = &(VkPipelineColorBlendStateCreateInfo) {
 195                                                        .sType = VK_STRUCTURE_TYPE_PIPELINE_COLOR_BLEND_STATE_CREATE_INFO,
 196                                                        .logicOpEnable = false,
 197                                                        .attachmentCount = 2,
 198                                                        .pAttachments = (VkPipelineColorBlendAttachmentState []) {
 199                                                                {
 200                                                                .colorWriteMask = VK_COLOR_COMPONENT_R_BIT |
 201                                                                                VK_COLOR_COMPONENT_G_BIT |
 202                                                                                VK_COLOR_COMPONENT_B_BIT |
 203                                                                                VK_COLOR_COMPONENT_A_BIT,
 204                                                                },
 205                                                                {
 206                                                                .colorWriteMask = 0,
 207
 208                                                                }
 209                                                        },
 210                                                 },
 211                                                 .pDynamicState = &(VkPipelineDynamicStateCreateInfo) {
 212                                                         .sType = VK_STRUCTURE_TYPE_PIPELINE_DYNAMIC_STATE_CREATE_INFO,
 213                                                         .dynamicStateCount = 2,
 214                                                         .pDynamicStates = (VkDynamicState[]) {
 215                                                                 VK_DYNAMIC_STATE_VIEWPORT,
 216                                                                 VK_DYNAMIC_STATE_SCISSOR,
 217                                                         },
 218                                                 },
 219                                                 .layout = device->meta_state.resolve.p_layout,
 220                                                 .renderPass = pass,
 221                                                                                                                                        .subpass = 0,
 222                                                                                                                                        },
 223                                                &(struct radv_graphics_pipeline_create_info) {
 224                                                        .use_rectlist = true,
 225                                                        .custom_blend_mode = V_028808_CB_RESOLVE,
 226                                                                },
 227                                                &device->meta_state.alloc, pipeline);
 228         if (result != VK_SUCCESS)
 229                 goto cleanup;
 230
 231         goto cleanup;
 232
 233 cleanup:
 234         ralloc_free(fs_module.nir);
 235         return result;
 236 }
 237
 238 void
 239 radv_device_finish_meta_resolve_state(struct radv_device *device)
 240 {
 241         struct radv_meta_state *state = &device->meta_state;
 242
 243         for (uint32_t j = 0; j < NUM_META_FS_KEYS; j++) {
 244                 radv_DestroyRenderPass(radv_device_to_handle(device),
 245                                        state->resolve.pass[j], &state->alloc);
 246                 radv_DestroyPipeline(radv_device_to_handle(device),
 247                                      state->resolve.pipeline[j], &state->alloc);
 248         }
 249         radv_DestroyPipelineLayout(radv_device_to_handle(device),
 250                                    state->resolve.p_layout, &state->alloc);
 251
 252 }
 253
 254 VkResult
 255 radv_device_init_meta_resolve_state(struct radv_device *device, bool on_demand)
 256 {
 257         if (on_demand)
 258                 return VK_SUCCESS;
 259
 260         VkResult res = VK_SUCCESS;
 261         struct radv_meta_state *state = &device->meta_state;
 262         struct radv_shader_module vs_module = { .nir = radv_meta_build_nir_vs_generate_vertices() };
 263         if (!vs_module.nir) {
 264                 /* XXX: Need more accurate error */
 265                 res = VK_ERROR_OUT_OF_HOST_MEMORY;
 266                 goto fail;
 267         }
 268
 269         for (uint32_t i = 0; i < NUM_META_FS_KEYS; ++i) {
 270                 VkFormat format = radv_fs_key_format_exemplars[i];
 271                 unsigned fs_key = radv_format_meta_fs_key(format);
 272                 res = create_pass(device, format, &state->resolve.pass[fs_key]);
 273                 if (res != VK_SUCCESS)
 274                         goto fail;
 275
 276                 VkShaderModule vs_module_h = radv_shader_module_to_handle(&vs_module);
 277                 res = create_pipeline(device, vs_module_h,
 278                                       &state->resolve.pipeline[fs_key], state->resolve.pass[fs_key]);
 279                 if (res != VK_SUCCESS)
 280                         goto fail;
 281         }
 282
 283         goto cleanup;
 284
 285 fail:
 286         radv_device_finish_meta_resolve_state(device);
 287
 288 cleanup:
 289         ralloc_free(vs_module.nir);
 290
 291         return res;
 292 }
 293
 294 static void
 295 emit_resolve(struct radv_cmd_buffer *cmd_buffer,
 296              VkFormat vk_format,
 297              const VkOffset2D *dest_offset,
 298              const VkExtent2D *resolve_extent)
 299 {
 300         struct radv_device *device = cmd_buffer->device;
 301         VkCommandBuffer cmd_buffer_h = radv_cmd_buffer_to_handle(cmd_buffer);
 302         unsigned fs_key = radv_format_meta_fs_key(vk_format);
 303
 304         cmd_buffer->state.flush_bits |= RADV_CMD_FLAG_FLUSH_AND_INV_CB;
 305
 306         radv_CmdBindPipeline(cmd_buffer_h, VK_PIPELINE_BIND_POINT_GRAPHICS,
 307                              device->meta_state.resolve.pipeline[fs_key]);
 308
 309         radv_CmdSetViewport(radv_cmd_buffer_to_handle(cmd_buffer), 0, 1, &(VkViewport) {
 310                 .x = dest_offset->x,
 311                 .y = dest_offset->y,
 312                 .width = resolve_extent->width,
 313                 .height = resolve_extent->height,
 314                 .minDepth = 0.0f,
 315                 .maxDepth = 1.0f
 316         });
 317
 318         radv_CmdSetScissor(radv_cmd_buffer_to_handle(cmd_buffer), 0, 1, &(VkRect2D) {
 319                 .offset = *dest_offset,
 320                 .extent = *resolve_extent,
 321         });
 322
 323         radv_CmdDraw(cmd_buffer_h, 3, 1, 0, 0);
 324         cmd_buffer->state.flush_bits |= RADV_CMD_FLAG_FLUSH_AND_INV_CB;
 325 }
 326
 327 enum radv_resolve_method {
 328         RESOLVE_HW,
 329         RESOLVE_COMPUTE,
 330         RESOLVE_FRAGMENT,
 331 };
 332
 333 static void radv_pick_resolve_method_images(struct radv_image *src_image,
 334                                             VkFormat src_format,
 335                                             struct radv_image *dest_image,
 336                                             VkImageLayout dest_image_layout,
 337                                             bool dest_render_loop,
 338                                             struct radv_cmd_buffer *cmd_buffer,
 339                                             enum radv_resolve_method *method)
 340
 341 {
 342         uint32_t queue_mask = radv_image_queue_family_mask(dest_image,
 343                                                            cmd_buffer->queue_family_index,
 344                                                            cmd_buffer->queue_family_index);
 345
 346         if (vk_format_is_color(src_format)) {
 347                 if (src_format == VK_FORMAT_R16G16_UNORM ||
 348                     src_format == VK_FORMAT_R16G16_SNORM)
 349                         *method = RESOLVE_COMPUTE;
 350                 else if (vk_format_is_int(src_format))
 351                         *method = RESOLVE_COMPUTE;
 352                 else if (src_image->info.array_size > 1 ||
 353                          dest_image->info.array_size > 1)
 354                         *method = RESOLVE_COMPUTE;
 355
 356                 if (radv_layout_dcc_compressed(dest_image, dest_image_layout,
 357                                                dest_render_loop, queue_mask)) {
 358                         *method = RESOLVE_FRAGMENT;
 359                 } else if (dest_image->planes[0].surface.micro_tile_mode !=
 360                            src_image->planes[0].surface.micro_tile_mode) {
 361                         *method = RESOLVE_COMPUTE;
 362                 }
 363         } else {
 364                 if (src_image->info.array_size > 1 ||
 365                     dest_image->info.array_size > 1)
 366                         *method = RESOLVE_COMPUTE;
 367                 else
 368                         *method = RESOLVE_FRAGMENT;
 369         }
 370 }
 371
 372 static VkResult
 373 build_resolve_pipeline(struct radv_device *device,
 374                        unsigned fs_key)
 375 {
 376         VkResult result = VK_SUCCESS;
 377
 378         if (device->meta_state.resolve.pipeline[fs_key])
 379                 return result;
 380
 381         mtx_lock(&device->meta_state.mtx);
 382         if (device->meta_state.resolve.pipeline[fs_key]) {
 383                 mtx_unlock(&device->meta_state.mtx);
 384                 return result;
 385         }
 386
 387         struct radv_shader_module vs_module = { .nir = radv_meta_build_nir_vs_generate_vertices() };
 388
 389         result = create_pass(device, radv_fs_key_format_exemplars[fs_key], &device->meta_state.resolve.pass[fs_key]);
 390         if (result != VK_SUCCESS)
 391                 goto fail;
 392
 393         VkShaderModule vs_module_h = radv_shader_module_to_handle(&vs_module);
 394         result = create_pipeline(device, vs_module_h, &device->meta_state.resolve.pipeline[fs_key], device->meta_state.resolve.pass[fs_key]);
 395
 396 fail:
 397         ralloc_free(vs_module.nir);
 398         mtx_unlock(&device->meta_state.mtx);
 399         return result;
 400 }
 401
 402 void radv_CmdResolveImage(
 403         VkCommandBuffer                             cmd_buffer_h,
 404         VkImage                                     src_image_h,
 405         VkImageLayout                               src_image_layout,
 406         VkImage                                     dest_image_h,
 407         VkImageLayout                               dest_image_layout,
 408         uint32_t                                    region_count,
 409         const VkImageResolve*                       regions)
 410 {
 411         RADV_FROM_HANDLE(radv_cmd_buffer, cmd_buffer, cmd_buffer_h);
 412         RADV_FROM_HANDLE(radv_image, src_image, src_image_h);
 413         RADV_FROM_HANDLE(radv_image, dest_image, dest_image_h);
 414         struct radv_device *device = cmd_buffer->device;
 415         struct radv_meta_saved_state saved_state;
 416         VkDevice device_h = radv_device_to_handle(device);
 417         enum radv_resolve_method resolve_method = RESOLVE_HW;
 418         /* we can use the hw resolve only for single full resolves */
 419         if (region_count == 1) {
 420                 if (regions[0].srcOffset.x ||
 421                     regions[0].srcOffset.y ||
 422                     regions[0].srcOffset.z)
 423                         resolve_method = RESOLVE_COMPUTE;
 424                 if (regions[0].dstOffset.x ||
 425                     regions[0].dstOffset.y ||
 426                     regions[0].dstOffset.z)
 427                         resolve_method = RESOLVE_COMPUTE;
 428
 429                 if (regions[0].extent.width != src_image->info.width ||
 430                     regions[0].extent.height != src_image->info.height ||
 431                     regions[0].extent.depth != src_image->info.depth)
 432                         resolve_method = RESOLVE_COMPUTE;
 433         } else
 434                 resolve_method = RESOLVE_COMPUTE;
 435
 436         radv_pick_resolve_method_images(src_image, src_image->vk_format,
 437                                         dest_image, dest_image_layout,
 438                                         false, cmd_buffer, &resolve_method);
 439
 440         if (resolve_method == RESOLVE_FRAGMENT) {
 441                 radv_meta_resolve_fragment_image(cmd_buffer,
 442                                                  src_image,
 443                                                  src_image_layout,
 444                                                  dest_image,
 445                                                  dest_image_layout,
 446                                                  region_count, regions);
 447                 return;
 448         }
 449
 450         if (resolve_method == RESOLVE_COMPUTE) {
 451                 radv_meta_resolve_compute_image(cmd_buffer,
 452                                                 src_image,
 453                                                 src_image->vk_format,
 454                                                 src_image_layout,
 455                                                 dest_image,
 456                                                 dest_image->vk_format,
 457                                                 dest_image_layout,
 458                                                 region_count, regions);
 459                 return;
 460         }
 461
 462         radv_meta_save(&saved_state, cmd_buffer,
 463                        RADV_META_SAVE_GRAPHICS_PIPELINE);
 464
 465         assert(src_image->info.samples > 1);
 466         if (src_image->info.samples <= 1) {
 467                 /* this causes GPU hangs if we get past here */
 468                 fprintf(stderr, "radv: Illegal resolve operation (src not multisampled), will hang GPU.");
 469                 return;
 470         }
 471         assert(dest_image->info.samples == 1);
 472
 473         if (src_image->info.array_size > 1)
 474                 radv_finishme("vkCmdResolveImage: multisample array images");
 475
 476         unsigned fs_key = radv_format_meta_fs_key(dest_image->vk_format);
 477         for (uint32_t r = 0; r < region_count; ++r) {
 478                 const VkImageResolve *region = &regions[r];
 479
 480                 /* From the Vulkan 1.0 spec:
 481                  *
 482                  *    - The aspectMask member of srcSubresource and dstSubresource must
 483                  *      only contain VK_IMAGE_ASPECT_COLOR_BIT
 484                  *
 485                  *    - The layerCount member of srcSubresource and dstSubresource must
 486                  *      match
 487                  */
 488                 assert(region->srcSubresource.aspectMask == VK_IMAGE_ASPECT_COLOR_BIT);
 489                 assert(region->dstSubresource.aspectMask == VK_IMAGE_ASPECT_COLOR_BIT);
 490                 assert(region->srcSubresource.layerCount ==
 491                        region->dstSubresource.layerCount);
 492
 493                 const uint32_t src_base_layer =
 494                         radv_meta_get_iview_layer(src_image, &region->srcSubresource,
 495                                                   &region->srcOffset);
 496
 497                 const uint32_t dest_base_layer =
 498                         radv_meta_get_iview_layer(dest_image, &region->dstSubresource,
 499                                                   &region->dstOffset);
 500
 501                 /**
 502                  * From Vulkan 1.0.6 spec: 18.6 Resolving Multisample Images
 503                  *
 504                  *    extent is the size in texels of the source image to resolve in width,
 505                  *    height and depth. 1D images use only x and width. 2D images use x, y,
 506                  *    width and height. 3D images use x, y, z, width, height and depth.
 507                  *
 508                  *    srcOffset and dstOffset select the initial x, y, and z offsets in
 509                  *    texels of the sub-regions of the source and destination image data.
 510                  *    extent is the size in texels of the source image to resolve in width,
 511                  *    height and depth. 1D images use only x and width. 2D images use x, y,
 512                  *    width and height. 3D images use x, y, z, width, height and depth.
 513                  */
 514                 const struct VkExtent3D extent =
 515                         radv_sanitize_image_extent(src_image->type, region->extent);
 516                 const struct VkOffset3D dstOffset =
 517                         radv_sanitize_image_offset(dest_image->type, region->dstOffset);
 518
 519                 if (radv_dcc_enabled(dest_image, region->dstSubresource.mipLevel)) {
 520                         VkImageSubresourceRange range = {
 521                                 .aspectMask = VK_IMAGE_ASPECT_COLOR_BIT,
 522                                 .baseMipLevel = region->dstSubresource.mipLevel,
 523                                 .levelCount = 1,
 524                                 .baseArrayLayer = dest_base_layer,
 525                                 .layerCount = region->dstSubresource.layerCount,
 526                         };
 527
 528                         radv_initialize_dcc(cmd_buffer, dest_image, &range, 0xffffffff);
 529                 }
 530
 531                 for (uint32_t layer = 0; layer < region->srcSubresource.layerCount;
 532                      ++layer) {
 533
 534                         VkResult ret = build_resolve_pipeline(device, fs_key);
 535                         if (ret != VK_SUCCESS) {
 536                                 cmd_buffer->record_result = ret;
 537                                 break;
 538                         }
 539
 540                         struct radv_image_view src_iview;
 541                         radv_image_view_init(&src_iview, cmd_buffer->device,
 542                                              &(VkImageViewCreateInfo) {
 543                                                      .sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO,
 544                                                              .image = src_image_h,
 545                                                              .viewType = radv_meta_get_view_type(src_image),
 546                                                              .format = src_image->vk_format,
 547                                                              .subresourceRange = {
 548                                                              .aspectMask = VK_IMAGE_ASPECT_COLOR_BIT,
 549                                                              .baseMipLevel = region->srcSubresource.mipLevel,
 550                                                              .levelCount = 1,
 551                                                              .baseArrayLayer = src_base_layer + layer,
 552                                                              .layerCount = 1,
 553                                                      },
 554                                              });
 555
 556                         struct radv_image_view dest_iview;
 557                         radv_image_view_init(&dest_iview, cmd_buffer->device,
 558                                              &(VkImageViewCreateInfo) {
 559                                                      .sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO,
 560                                                              .image = dest_image_h,
 561                                                              .viewType = radv_meta_get_view_type(dest_image),
 562                                                              .format = dest_image->vk_format,
 563                                                              .subresourceRange = {
 564                                                              .aspectMask = VK_IMAGE_ASPECT_COLOR_BIT,
 565                                                              .baseMipLevel = region->dstSubresource.mipLevel,
 566                                                              .levelCount = 1,
 567                                                              .baseArrayLayer = dest_base_layer + layer,
 568                                                              .layerCount = 1,
 569                                                      },
 570                                               });
 571
 572                         VkFramebuffer fb_h;
 573                         radv_CreateFramebuffer(device_h,
 574                                                &(VkFramebufferCreateInfo) {
 575                                                        .sType = VK_STRUCTURE_TYPE_FRAMEBUFFER_CREATE_INFO,
 576                                                                .attachmentCount = 2,
 577                                                                .pAttachments = (VkImageView[]) {
 578                                                                radv_image_view_to_handle(&src_iview),
 579                                                                radv_image_view_to_handle(&dest_iview),
 580                                                        },
 581                                                        .width = radv_minify(dest_image->info.width,
 582                                                                             region->dstSubresource.mipLevel),
 583                                                        .height = radv_minify(dest_image->info.height,
 584                                                                               region->dstSubresource.mipLevel),
 585                                                        .layers = 1
 586                                                },
 587                                                &cmd_buffer->pool->alloc,
 588                                                &fb_h);
 589
 590                         radv_CmdBeginRenderPass(cmd_buffer_h,
 591                                                       &(VkRenderPassBeginInfo) {
 592                                                               .sType = VK_STRUCTURE_TYPE_RENDER_PASS_BEGIN_INFO,
 593                                                                       .renderPass = device->meta_state.resolve.pass[fs_key],
 594                                                                       .framebuffer = fb_h,
 595                                                                       .renderArea = {
 596                                                                       .offset = {
 597                                                                               dstOffset.x,
 598                                                                               dstOffset.y,
 599                                                                       },
 600                                                                       .extent = {
 601                                                                               extent.width,
 602                                                                               extent.height,
 603                                                                       }
 604                                                               },
 605                                                               .clearValueCount = 0,
 606                                                               .pClearValues = NULL,
 607                                                       },
 608                                                       VK_SUBPASS_CONTENTS_INLINE);
 609
 610                         emit_resolve(cmd_buffer,
 611                                      dest_iview.vk_format,
 612                                      &(VkOffset2D) {
 613                                              .x = dstOffset.x,
 614                                              .y = dstOffset.y,
 615                                      },
 616                                      &(VkExtent2D) {
 617                                              .width = extent.width,
 618                                              .height = extent.height,
 619                                      });
 620
 621                         radv_CmdEndRenderPass(cmd_buffer_h);
 622
 623                         radv_DestroyFramebuffer(device_h, fb_h,
 624                                                 &cmd_buffer->pool->alloc);
 625                 }
 626         }
 627
 628         radv_meta_restore(&saved_state, cmd_buffer);
 629 }
 630
 631 /**
 632  * Emit any needed resolves for the current subpass.
 633  */
 634 void
 635 radv_cmd_buffer_resolve_subpass(struct radv_cmd_buffer *cmd_buffer)
 636 {
 637         struct radv_framebuffer *fb = cmd_buffer->state.framebuffer;
 638         const struct radv_subpass *subpass = cmd_buffer->state.subpass;
 639         struct radv_meta_saved_state saved_state;
 640         enum radv_resolve_method resolve_method = RESOLVE_HW;
 641
 642         if (subpass->ds_resolve_attachment) {
 643                 struct radv_subpass_attachment src_att = *subpass->depth_stencil_attachment;
 644                 struct radv_subpass_attachment dst_att = *subpass->ds_resolve_attachment;
 645                 struct radv_image_view *src_iview =
 646                         cmd_buffer->state.attachments[src_att.attachment].iview;
 647                 struct radv_image_view *dst_iview =
 648                         cmd_buffer->state.attachments[dst_att.attachment].iview;
 649
 650                 radv_pick_resolve_method_images(src_iview->image,
 651                                                 src_iview->vk_format,
 652                                                 dst_iview->image,
 653                                                 dst_att.layout,
 654                                                 dst_att.in_render_loop,
 655                                                 cmd_buffer,
 656                                                 &resolve_method);
 657
 658                 if ((src_iview->aspect_mask & VK_IMAGE_ASPECT_DEPTH_BIT) &&
 659                     subpass->depth_resolve_mode != VK_RESOLVE_MODE_NONE_KHR) {
 660                         if (resolve_method == RESOLVE_FRAGMENT) {
 661                                 radv_depth_stencil_resolve_subpass_fs(cmd_buffer,
 662                                                                       VK_IMAGE_ASPECT_DEPTH_BIT,
 663                                                                       subpass->depth_resolve_mode);
 664                         } else {
 665                                 assert(resolve_method == RESOLVE_COMPUTE);
 666                                 radv_depth_stencil_resolve_subpass_cs(cmd_buffer,
 667                                                                       VK_IMAGE_ASPECT_DEPTH_BIT,
 668                                                                       subpass->depth_resolve_mode);
 669                         }
 670                 }
 671
 672                 if ((src_iview->aspect_mask & VK_IMAGE_ASPECT_STENCIL_BIT) &&
 673                     subpass->stencil_resolve_mode != VK_RESOLVE_MODE_NONE_KHR) {
 674                         if (resolve_method == RESOLVE_FRAGMENT) {
 675                                 radv_depth_stencil_resolve_subpass_fs(cmd_buffer,
 676                                                                       VK_IMAGE_ASPECT_STENCIL_BIT,
 677                                                                       subpass->stencil_resolve_mode);
 678                         } else {
 679                                 assert(resolve_method == RESOLVE_COMPUTE);
 680                                 radv_depth_stencil_resolve_subpass_cs(cmd_buffer,
 681                                                                       VK_IMAGE_ASPECT_STENCIL_BIT,
 682                                                                       subpass->stencil_resolve_mode);
 683                         }
 684                 }
 685         }
 686
 687         if (!subpass->has_color_resolve)
 688                 return;
 689
 690         for (uint32_t i = 0; i < subpass->color_count; ++i) {
 691                 struct radv_subpass_attachment src_att = subpass->color_attachments[i];
 692                 struct radv_subpass_attachment dest_att = subpass->resolve_attachments[i];
 693
 694                 if (dest_att.attachment == VK_ATTACHMENT_UNUSED)
 695                         continue;
 696
 697                 /* Make sure to not clear color attachments after resolves. */
 698                 cmd_buffer->state.attachments[dest_att.attachment].pending_clear_aspects = 0;
 699
 700                 struct radv_image *dst_img = cmd_buffer->state.attachments[dest_att.attachment].iview->image;
 701                 struct radv_image_view *src_iview= cmd_buffer->state.attachments[src_att.attachment].iview;
 702                 struct radv_image *src_img = src_iview->image;
 703
 704                 radv_pick_resolve_method_images(src_img, src_iview->vk_format,
 705                                                 dst_img, dest_att.layout,
 706                                                 dest_att.in_render_loop,
 707                                                 cmd_buffer, &resolve_method);
 708
 709                 if (resolve_method == RESOLVE_FRAGMENT) {
 710                         break;
 711                 }
 712         }
 713
 714         if (resolve_method == RESOLVE_COMPUTE) {
 715                 radv_cmd_buffer_resolve_subpass_cs(cmd_buffer);
 716                 return;
 717         } else if (resolve_method == RESOLVE_FRAGMENT) {
 718                 radv_cmd_buffer_resolve_subpass_fs(cmd_buffer);
 719                 return;
 720         }
 721
 722         radv_meta_save(&saved_state, cmd_buffer,
 723                        RADV_META_SAVE_GRAPHICS_PIPELINE);
 724
 725         for (uint32_t i = 0; i < subpass->color_count; ++i) {
 726                 struct radv_subpass_attachment src_att = subpass->color_attachments[i];
 727                 struct radv_subpass_attachment dest_att = subpass->resolve_attachments[i];
 728
 729                 if (dest_att.attachment == VK_ATTACHMENT_UNUSED)
 730                         continue;
 731
 732                 struct radv_image_view *dest_iview = cmd_buffer->state.attachments[dest_att.attachment].iview;
 733                 struct radv_image *dst_img = dest_iview->image;
 734
 735                 if (radv_dcc_enabled(dst_img, dest_iview->base_mip)) {
 736                         VkImageSubresourceRange range = {
 737                                 .aspectMask = dest_iview->aspect_mask,
 738                                 .baseMipLevel = dest_iview->base_mip,
 739                                 .levelCount = dest_iview->level_count,
 740                                 .baseArrayLayer = dest_iview->base_layer,
 741                                 .layerCount = dest_iview->layer_count,
 742                         };
 743
 744                         radv_initialize_dcc(cmd_buffer, dst_img, &range, 0xffffffff);
 745                         cmd_buffer->state.attachments[dest_att.attachment].current_layout = VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL;
 746                 }
 747
 748                 struct radv_subpass resolve_subpass = {
 749                         .color_count = 2,
 750                         .color_attachments = (struct radv_subpass_attachment[]) { src_att, dest_att },
 751                         .depth_stencil_attachment = NULL,
 752                 };
 753
 754                 radv_cmd_buffer_set_subpass(cmd_buffer, &resolve_subpass);
 755
 756                 VkResult ret = build_resolve_pipeline(cmd_buffer->device, radv_format_meta_fs_key(dest_iview->vk_format));
 757                 if (ret != VK_SUCCESS) {
 758                         cmd_buffer->record_result = ret;
 759                         continue;
 760                 }
 761
 762                 emit_resolve(cmd_buffer,
 763                              dest_iview->vk_format,
 764                              &(VkOffset2D) { 0, 0 },
 765                              &(VkExtent2D) { fb->width, fb->height });
 766         }
 767
 768         radv_cmd_buffer_set_subpass(cmd_buffer, subpass);
 769
 770         radv_meta_restore(&saved_state, cmd_buffer);
 771 }
 772
 773 /**
 774  * Decompress CMask/FMask before resolving a multisampled source image inside a
 775  * subpass.
 776  */
 777 void
 778 radv_decompress_resolve_subpass_src(struct radv_cmd_buffer *cmd_buffer)
 779 {
 780         const struct radv_subpass *subpass = cmd_buffer->state.subpass;
 781         struct radv_framebuffer *fb = cmd_buffer->state.framebuffer;
 782         uint32_t layer_count = fb->layers;
 783
 784         if (subpass->view_mask)
 785                 layer_count = util_last_bit(subpass->view_mask);
 786
 787         for (uint32_t i = 0; i < subpass->color_count; ++i) {
 788                 struct radv_subpass_attachment src_att = subpass->color_attachments[i];
 789                 struct radv_subpass_attachment dest_att = subpass->resolve_attachments[i];
 790
 791                 if (dest_att.attachment == VK_ATTACHMENT_UNUSED)
 792                         continue;
 793
 794                 struct radv_image_view *src_iview = cmd_buffer->state.attachments[src_att.attachment].iview;
 795                 struct radv_image *src_image = src_iview->image;
 796
 797                 VkImageResolve region = {};
 798                 region.srcSubresource.aspectMask = VK_IMAGE_ASPECT_COLOR_BIT;
 799                 region.srcSubresource.mipLevel = 0;
 800                 region.srcSubresource.baseArrayLayer = src_iview->base_layer;
 801                 region.srcSubresource.layerCount = layer_count;
 802
 803                 radv_decompress_resolve_src(cmd_buffer, src_image,
 804                                             src_att.layout, 1, &region);
 805         }
 806
 807         if (subpass->ds_resolve_attachment) {
 808                 struct radv_subpass_attachment src_att = *subpass->depth_stencil_attachment;
 809                 struct radv_image_view *src_iview = fb->attachments[src_att.attachment];
 810                 struct radv_image *src_image = src_iview->image;
 811
 812                 VkImageResolve region = {};
 813                 region.srcSubresource.aspectMask = VK_IMAGE_ASPECT_DEPTH_BIT;
 814                 region.srcSubresource.mipLevel = 0;
 815                 region.srcSubresource.baseArrayLayer = src_iview->base_layer;
 816                 region.srcSubresource.layerCount = layer_count;
 817
 818                 radv_decompress_resolve_src(cmd_buffer, src_image,
 819                                             src_att.layout, 1, &region);
 820         }
 821 }
 822
 823 static struct radv_sample_locations_state *
 824 radv_get_resolve_sample_locations(struct radv_cmd_buffer *cmd_buffer)
 825 {
 826         struct radv_cmd_state *state = &cmd_buffer->state;
 827         uint32_t subpass_id = radv_get_subpass_id(cmd_buffer);
 828
 829         for (uint32_t i = 0; i < state->num_subpass_sample_locs; i++) {
 830                 if (state->subpass_sample_locs[i].subpass_idx == subpass_id)
 831                         return &state->subpass_sample_locs[i].sample_location;
 832         }
 833
 834         return NULL;
 835 }
 836
 837 /**
 838  * Decompress CMask/FMask before resolving a multisampled source image.
 839  */
 840 void
 841 radv_decompress_resolve_src(struct radv_cmd_buffer *cmd_buffer,
 842                             struct radv_image *src_image,
 843                             VkImageLayout src_image_layout,
 844                             uint32_t region_count,
 845                             const VkImageResolve *regions)
 846 {
 847         for (uint32_t r = 0; r < region_count; ++r) {
 848                 const VkImageResolve *region = &regions[r];
 849                 const uint32_t src_base_layer =
 850                         radv_meta_get_iview_layer(src_image, &region->srcSubresource,
 851                                                   &region->srcOffset);
 852
 853                 VkImageMemoryBarrier barrier = {};
 854                 barrier.srcAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT;
 855                 barrier.dstAccessMask = VK_ACCESS_TRANSFER_READ_BIT;
 856                 barrier.oldLayout = src_image_layout;
 857                 barrier.newLayout = VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL;
 858                 barrier.image = radv_image_to_handle(src_image);
 859                 barrier.subresourceRange = (VkImageSubresourceRange) {
 860                         .aspectMask = region->srcSubresource.aspectMask,
 861                         .baseMipLevel = region->srcSubresource.mipLevel,
 862                         .levelCount = 1,
 863                         .baseArrayLayer = src_base_layer,
 864                         .layerCount = region->srcSubresource.layerCount,
 865                 };
 866
 867                 if (src_image->flags & VK_IMAGE_CREATE_SAMPLE_LOCATIONS_COMPATIBLE_DEPTH_BIT_EXT) {
 868                         /* If the depth/stencil image uses different sample
 869                          * locations, we need them during HTILE decompressions.
 870                          */
 871                         struct radv_sample_locations_state *sample_locs =
 872                                 radv_get_resolve_sample_locations(cmd_buffer);
 873
 874                         barrier.pNext = &(VkSampleLocationsInfoEXT) {
 875                                 .sType = VK_STRUCTURE_TYPE_SAMPLE_LOCATIONS_INFO_EXT,
 876                                 .sampleLocationsPerPixel = sample_locs->per_pixel,
 877                                 .sampleLocationGridSize = sample_locs->grid_size,
 878                                 .sampleLocationsCount = sample_locs->count,
 879                                 .pSampleLocations = sample_locs->locations,
 880                         };
 881                 }
 882
 883                 radv_CmdPipelineBarrier(radv_cmd_buffer_to_handle(cmd_buffer),
 884                                         VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT,
 885                                         VK_PIPELINE_STAGE_BOTTOM_OF_PIPE_BIT,
 886                                         false, 0, NULL, 0, NULL, 1, &barrier);
 887         }
 888 }