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
 107         struct radv_shader_module fs_module = {
 108                 .nir = radv_meta_build_nir_fs_noop(),
 109         };
 110
 111         if (!fs_module.nir) {
 112                 /* XXX: Need more accurate error */
 113                 result = VK_ERROR_OUT_OF_HOST_MEMORY;
 114                 goto cleanup;
 115         }
 116
 117         const VkGraphicsPipelineCreateInfo pipeline_create_info = {
 118                 .sType = VK_STRUCTURE_TYPE_GRAPHICS_PIPELINE_CREATE_INFO,
 119                 .stageCount = 2,
 120                 .pStages = (VkPipelineShaderStageCreateInfo[]) {
 121                        {
 122                                 .sType = VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_CREATE_INFO,
 123                                 .stage = VK_SHADER_STAGE_VERTEX_BIT,
 124                                 .module = vs_module_h,
 125                                 .pName = "main",
 126                         },
 127                         {
 128                                 .sType = VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_CREATE_INFO,
 129                                 .stage = VK_SHADER_STAGE_FRAGMENT_BIT,
 130                                 .module = radv_shader_module_to_handle(&fs_module),
 131                                 .pName = "main",
 132                         },
 133                 },
 134                 .pVertexInputState = &(VkPipelineVertexInputStateCreateInfo) {
 135                         .sType = VK_STRUCTURE_TYPE_PIPELINE_VERTEX_INPUT_STATE_CREATE_INFO,
 136                         .vertexBindingDescriptionCount = 0,
 137                         .vertexAttributeDescriptionCount = 0,
 138                 },
 139                 .pInputAssemblyState = &(VkPipelineInputAssemblyStateCreateInfo) {
 140                         .sType = VK_STRUCTURE_TYPE_PIPELINE_INPUT_ASSEMBLY_STATE_CREATE_INFO,
 141                         .topology = VK_PRIMITIVE_TOPOLOGY_TRIANGLE_STRIP,
 142                         .primitiveRestartEnable = false,
 143                 },
 144                 .pViewportState = &(VkPipelineViewportStateCreateInfo) {
 145                         .sType = VK_STRUCTURE_TYPE_PIPELINE_VIEWPORT_STATE_CREATE_INFO,
 146                         .viewportCount = 1,
 147                         .scissorCount = 1,
 148                 },
 149                 .pRasterizationState = &(VkPipelineRasterizationStateCreateInfo) {
 150                         .sType = VK_STRUCTURE_TYPE_PIPELINE_RASTERIZATION_STATE_CREATE_INFO,
 151                         .depthClampEnable = false,
 152                         .rasterizerDiscardEnable = false,
 153                         .polygonMode = VK_POLYGON_MODE_FILL,
 154                         .cullMode = VK_CULL_MODE_NONE,
 155                         .frontFace = VK_FRONT_FACE_COUNTER_CLOCKWISE,
 156                 },
 157                 .pMultisampleState = &(VkPipelineMultisampleStateCreateInfo) {
 158                         .sType = VK_STRUCTURE_TYPE_PIPELINE_MULTISAMPLE_STATE_CREATE_INFO,
 159                         .rasterizationSamples = samples,
 160                         .sampleShadingEnable = false,
 161                         .pSampleMask = NULL,
 162                         .alphaToCoverageEnable = false,
 163                         .alphaToOneEnable = false,
 164                 },
 165                 .pColorBlendState = &(VkPipelineColorBlendStateCreateInfo) {
 166                         .sType = VK_STRUCTURE_TYPE_PIPELINE_COLOR_BLEND_STATE_CREATE_INFO,
 167                         .logicOpEnable = false,
 168                         .attachmentCount = 0,
 169                         .pAttachments = NULL,
 170                 },
 171                 .pDepthStencilState = &(VkPipelineDepthStencilStateCreateInfo) {
 172                         .sType = VK_STRUCTURE_TYPE_PIPELINE_DEPTH_STENCIL_STATE_CREATE_INFO,
 173                         .depthTestEnable = false,
 174                         .depthWriteEnable = false,
 175                         .depthBoundsTestEnable = false,
 176                         .stencilTestEnable = false,
 177                 },
 178                 .pDynamicState = &(VkPipelineDynamicStateCreateInfo) {
 179                         .sType = VK_STRUCTURE_TYPE_PIPELINE_DYNAMIC_STATE_CREATE_INFO,
 180                         .dynamicStateCount = 2,
 181                         .pDynamicStates = (VkDynamicState[]) {
 182                                 VK_DYNAMIC_STATE_VIEWPORT,
 183                                 VK_DYNAMIC_STATE_SCISSOR,
 184                         },
 185                 },
 186                 .layout = layout,
 187                 .renderPass = pass,
 188                 .subpass = 0,
 189         };
 190
 191         result = radv_graphics_pipeline_create(device_h,
 192                                                radv_pipeline_cache_to_handle(&device->meta_state.cache),
 193                                                &pipeline_create_info,
 194                                                &(struct radv_graphics_pipeline_create_info) {
 195                                                         .use_rectlist = true,
 196                                                         .db_flush_depth_inplace = true,
 197                                                         .db_flush_stencil_inplace = true,
 198                                                },
 199                                                &device->meta_state.alloc,
 200                                                decompress_pipeline);
 201         if (result != VK_SUCCESS)
 202                 goto cleanup;
 203
 204         result = radv_graphics_pipeline_create(device_h,
 205                                                radv_pipeline_cache_to_handle(&device->meta_state.cache),
 206                                                &pipeline_create_info,
 207                                                &(struct radv_graphics_pipeline_create_info) {
 208                                                         .use_rectlist = true,
 209                                                         .db_flush_depth_inplace = true,
 210                                                         .db_flush_stencil_inplace = true,
 211                                                         .db_resummarize = true,
 212                                                },
 213                                                &device->meta_state.alloc,
 214                                                resummarize_pipeline);
 215         if (result != VK_SUCCESS)
 216                 goto cleanup;
 217
 218         goto cleanup;
 219
 220 cleanup:
 221         ralloc_free(fs_module.nir);
 222         return result;
 223 }
 224
 225 void
 226 radv_device_finish_meta_depth_decomp_state(struct radv_device *device)
 227 {
 228         struct radv_meta_state *state = &device->meta_state;
 229
 230         for (uint32_t i = 0; i < ARRAY_SIZE(state->depth_decomp); ++i) {
 231                 radv_DestroyRenderPass(radv_device_to_handle(device),
 232                                        state->depth_decomp[i].pass,
 233                                        &state->alloc);
 234                 radv_DestroyPipelineLayout(radv_device_to_handle(device),
 235                                            state->depth_decomp[i].p_layout,
 236                                            &state->alloc);
 237                 radv_DestroyPipeline(radv_device_to_handle(device),
 238                                      state->depth_decomp[i].decompress_pipeline,
 239                                      &state->alloc);
 240                 radv_DestroyPipeline(radv_device_to_handle(device),
 241                                      state->depth_decomp[i].resummarize_pipeline,
 242                                      &state->alloc);
 243         }
 244 }
 245
 246 VkResult
 247 radv_device_init_meta_depth_decomp_state(struct radv_device *device)
 248 {
 249         struct radv_meta_state *state = &device->meta_state;
 250         VkResult res = VK_SUCCESS;
 251
 252         struct radv_shader_module vs_module = { .nir = radv_meta_build_nir_vs_generate_vertices() };
 253         if (!vs_module.nir) {
 254                 /* XXX: Need more accurate error */
 255                 res = VK_ERROR_OUT_OF_HOST_MEMORY;
 256                 goto fail;
 257         }
 258
 259         VkShaderModule vs_module_h = radv_shader_module_to_handle(&vs_module);
 260
 261         for (uint32_t i = 0; i < ARRAY_SIZE(state->depth_decomp); ++i) {
 262                 uint32_t samples = 1 << i;
 263
 264                 res = create_pass(device, samples, &state->depth_decomp[i].pass);
 265                 if (res != VK_SUCCESS)
 266                         goto fail;
 267
 268                 res = create_pipeline_layout(device,
 269                                              &state->depth_decomp[i].p_layout);
 270                 if (res != VK_SUCCESS)
 271                         goto fail;
 272
 273                 res = create_pipeline(device, vs_module_h, samples,
 274                                       state->depth_decomp[i].pass,
 275                                       state->depth_decomp[i].p_layout,
 276                                       &state->depth_decomp[i].decompress_pipeline,
 277                                       &state->depth_decomp[i].resummarize_pipeline);
 278                 if (res != VK_SUCCESS)
 279                         goto fail;
 280         }
 281
 282         goto cleanup;
 283
 284 fail:
 285         radv_device_finish_meta_depth_decomp_state(device);
 286
 287 cleanup:
 288         ralloc_free(vs_module.nir);
 289
 290         return res;
 291 }
 292
 293 static void
 294 emit_depth_decomp(struct radv_cmd_buffer *cmd_buffer,
 295                   const VkExtent2D *depth_decomp_extent,
 296                   VkPipeline pipeline_h)
 297 {
 298         VkCommandBuffer cmd_buffer_h = radv_cmd_buffer_to_handle(cmd_buffer);
 299
 300         radv_CmdBindPipeline(cmd_buffer_h, VK_PIPELINE_BIND_POINT_GRAPHICS,
 301                              pipeline_h);
 302
 303         radv_CmdSetViewport(radv_cmd_buffer_to_handle(cmd_buffer), 0, 1, &(VkViewport) {
 304                 .x = 0,
 305                 .y = 0,
 306                 .width = depth_decomp_extent->width,
 307                 .height = depth_decomp_extent->height,
 308                 .minDepth = 0.0f,
 309                 .maxDepth = 1.0f
 310         });
 311
 312         radv_CmdSetScissor(radv_cmd_buffer_to_handle(cmd_buffer), 0, 1, &(VkRect2D) {
 313                 .offset = { 0, 0 },
 314                 .extent = *depth_decomp_extent,
 315         });
 316
 317         radv_CmdDraw(cmd_buffer_h, 3, 1, 0, 0);
 318 }
 319
 320
 321 enum radv_depth_op {
 322         DEPTH_DECOMPRESS,
 323         DEPTH_RESUMMARIZE,
 324 };
 325
 326 static void radv_process_depth_image_inplace(struct radv_cmd_buffer *cmd_buffer,
 327                                              struct radv_image *image,
 328                                              VkImageSubresourceRange *subresourceRange,
 329                                              enum radv_depth_op op)
 330 {
 331         struct radv_meta_saved_state saved_state;
 332         VkDevice device_h = radv_device_to_handle(cmd_buffer->device);
 333         VkCommandBuffer cmd_buffer_h = radv_cmd_buffer_to_handle(cmd_buffer);
 334         uint32_t width = radv_minify(image->info.width,
 335                                      subresourceRange->baseMipLevel);
 336         uint32_t height = radv_minify(image->info.height,
 337                                      subresourceRange->baseMipLevel);
 338         uint32_t samples = image->info.samples;
 339         uint32_t samples_log2 = ffs(samples) - 1;
 340         struct radv_meta_state *meta_state = &cmd_buffer->device->meta_state;
 341         VkPipeline pipeline_h;
 342
 343         if (!image->surface.htile_size)
 344                 return;
 345
 346         radv_meta_save(&saved_state, cmd_buffer,
 347                        RADV_META_SAVE_GRAPHICS_PIPELINE |
 348                        RADV_META_SAVE_PASS);
 349
 350         switch (op) {
 351         case DEPTH_DECOMPRESS:
 352                 pipeline_h = meta_state->depth_decomp[samples_log2].decompress_pipeline;
 353                 break;
 354         case DEPTH_RESUMMARIZE:
 355                 pipeline_h = meta_state->depth_decomp[samples_log2].resummarize_pipeline;
 356                 break;
 357         default:
 358                 unreachable("unknown operation");
 359         }
 360
 361         for (uint32_t layer = 0; layer < radv_get_layerCount(image, subresourceRange); layer++) {
 362                 struct radv_image_view iview;
 363
 364                 radv_image_view_init(&iview, cmd_buffer->device,
 365                                      &(VkImageViewCreateInfo) {
 366                                              .sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO,
 367                                              .image = radv_image_to_handle(image),
 368                                              .viewType = radv_meta_get_view_type(image),
 369                                              .format = image->vk_format,
 370                                              .subresourceRange = {
 371                                                      .aspectMask = VK_IMAGE_ASPECT_DEPTH_BIT,
 372                                                      .baseMipLevel = subresourceRange->baseMipLevel,
 373                                                      .levelCount = 1,
 374                                                      .baseArrayLayer = subresourceRange->baseArrayLayer + layer,
 375                                                      .layerCount = 1,
 376                                              },
 377                                      });
 378
 379
 380                 VkFramebuffer fb_h;
 381                 radv_CreateFramebuffer(device_h,
 382                                        &(VkFramebufferCreateInfo) {
 383                                                .sType = VK_STRUCTURE_TYPE_FRAMEBUFFER_CREATE_INFO,
 384                                                .attachmentCount = 1,
 385                                                        .pAttachments = (VkImageView[]) {
 386                                                        radv_image_view_to_handle(&iview)
 387                                                },
 388                                                .width = width,
 389                                                 .height = height,
 390                                                .layers = 1
 391                                        },
 392                                        &cmd_buffer->pool->alloc,
 393                                        &fb_h);
 394
 395                 radv_CmdBeginRenderPass(cmd_buffer_h,
 396                                               &(VkRenderPassBeginInfo) {
 397                                                       .sType = VK_STRUCTURE_TYPE_RENDER_PASS_BEGIN_INFO,
 398                                                               .renderPass = meta_state->depth_decomp[samples_log2].pass,
 399                                                               .framebuffer = fb_h,
 400                                                               .renderArea = {
 401                                                               .offset = {
 402                                                                       0,
 403                                                                       0,
 404                                                               },
 405                                                               .extent = {
 406                                                                       width,
 407                                                                       height,
 408                                                               }
 409                                                        },
 410                                                        .clearValueCount = 0,
 411                                                        .pClearValues = NULL,
 412                                            },
 413                                            VK_SUBPASS_CONTENTS_INLINE);
 414
 415                 emit_depth_decomp(cmd_buffer, &(VkExtent2D){width, height}, pipeline_h);
 416                 radv_CmdEndRenderPass(cmd_buffer_h);
 417
 418                 radv_DestroyFramebuffer(device_h, fb_h,
 419                                         &cmd_buffer->pool->alloc);
 420         }
 421         radv_meta_restore(&saved_state, cmd_buffer);
 422 }
 423
 424 void radv_decompress_depth_image_inplace(struct radv_cmd_buffer *cmd_buffer,
 425                                          struct radv_image *image,
 426                                          VkImageSubresourceRange *subresourceRange)
 427 {
 428         assert(cmd_buffer->queue_family_index == RADV_QUEUE_GENERAL);
 429         radv_process_depth_image_inplace(cmd_buffer, image, subresourceRange, DEPTH_DECOMPRESS);
 430 }
 431
 432 void radv_resummarize_depth_image_inplace(struct radv_cmd_buffer *cmd_buffer,
 433                                          struct radv_image *image,
 434                                          VkImageSubresourceRange *subresourceRange)
 435 {
 436         assert(cmd_buffer->queue_family_index == RADV_QUEUE_GENERAL);
 437         radv_process_depth_image_inplace(cmd_buffer, image, subresourceRange, DEPTH_RESUMMARIZE);
 438 }