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 "nir/nir_builder.h"
  30 #include "sid.h"
  31 /**
  32  * Vertex attributes used by all pipelines.
  33  */
  34 struct vertex_attrs {
  35         float position[2]; /**< 3DPRIM_RECTLIST */
  36 };
  37
  38 /* passthrough vertex shader */
  39 static nir_shader *
  40 build_nir_vs(void)
  41 {
  42         const struct glsl_type *vec4 = glsl_vec4_type();
  43
  44         nir_builder b;
  45         nir_variable *a_position;
  46         nir_variable *v_position;
  47
  48         nir_builder_init_simple_shader(&b, NULL, MESA_SHADER_VERTEX, NULL);
  49         b.shader->info->name = ralloc_strdup(b.shader, "meta_depth_decomp_vs");
  50
  51         a_position = nir_variable_create(b.shader, nir_var_shader_in, vec4,
  52                                          "a_position");
  53         a_position->data.location = VERT_ATTRIB_GENERIC0;
  54
  55         v_position = nir_variable_create(b.shader, nir_var_shader_out, vec4,
  56                                          "gl_Position");
  57         v_position->data.location = VARYING_SLOT_POS;
  58
  59         nir_copy_var(&b, v_position, a_position);
  60
  61         return b.shader;
  62 }
  63
  64 /* simple passthrough shader */
  65 static nir_shader *
  66 build_nir_fs(void)
  67 {
  68         nir_builder b;
  69
  70         nir_builder_init_simple_shader(&b, NULL, MESA_SHADER_FRAGMENT, NULL);
  71         b.shader->info->name = ralloc_asprintf(b.shader,
  72                                                "meta_depth_decomp_noop_fs");
  73
  74         return b.shader;
  75 }
  76
  77 static VkResult
  78 create_pass(struct radv_device *device)
  79 {
  80         VkResult result;
  81         VkDevice device_h = radv_device_to_handle(device);
  82         const VkAllocationCallbacks *alloc = &device->meta_state.alloc;
  83         VkAttachmentDescription attachment;
  84
  85         attachment.format = VK_FORMAT_UNDEFINED;
  86         attachment.samples = 1;
  87         attachment.loadOp = VK_ATTACHMENT_LOAD_OP_LOAD;
  88         attachment.storeOp = VK_ATTACHMENT_STORE_OP_STORE;
  89         attachment.initialLayout = VK_IMAGE_LAYOUT_DEPTH_STENCIL_ATTACHMENT_OPTIMAL;
  90         attachment.finalLayout = VK_IMAGE_LAYOUT_DEPTH_STENCIL_ATTACHMENT_OPTIMAL;
  91
  92         result = radv_CreateRenderPass(device_h,
  93                                        &(VkRenderPassCreateInfo) {
  94                                                .sType = VK_STRUCTURE_TYPE_RENDER_PASS_CREATE_INFO,
  95                                                        .attachmentCount = 1,
  96                                                        .pAttachments = &attachment,
  97                                                        .subpassCount = 1,
  98                                                         .pSubpasses = &(VkSubpassDescription) {
  99                                                        .pipelineBindPoint = VK_PIPELINE_BIND_POINT_GRAPHICS,
 100                                                        .inputAttachmentCount = 0,
 101                                                        .colorAttachmentCount = 0,
 102                                                        .pColorAttachments = NULL,
 103                                                        .pResolveAttachments = NULL,
 104                                                        .pDepthStencilAttachment = &(VkAttachmentReference) {
 105                                                                .attachment = 0,
 106                                                                .layout = VK_IMAGE_LAYOUT_DEPTH_STENCIL_ATTACHMENT_OPTIMAL,
 107                                                        },
 108                                                        .preserveAttachmentCount = 0,
 109                                                        .pPreserveAttachments = NULL,
 110                                                },
 111                                                                 .dependencyCount = 0,
 112                                                                    },
 113                                        alloc,
 114                                        &device->meta_state.depth_decomp.pass);
 115
 116         return result;
 117 }
 118
 119 static VkResult
 120 create_pipeline(struct radv_device *device,
 121                 VkShaderModule vs_module_h)
 122 {
 123         VkResult result;
 124         VkDevice device_h = radv_device_to_handle(device);
 125
 126         struct radv_shader_module fs_module = {
 127                 .nir = build_nir_fs(),
 128         };
 129
 130         if (!fs_module.nir) {
 131                 /* XXX: Need more accurate error */
 132                 result = VK_ERROR_OUT_OF_HOST_MEMORY;
 133                 goto cleanup;
 134         }
 135
 136         const VkGraphicsPipelineCreateInfo pipeline_create_info = {
 137                 .sType = VK_STRUCTURE_TYPE_GRAPHICS_PIPELINE_CREATE_INFO,
 138                 .stageCount = 2,
 139                 .pStages = (VkPipelineShaderStageCreateInfo[]) {
 140                        {
 141                                 .sType = VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_CREATE_INFO,
 142                                 .stage = VK_SHADER_STAGE_VERTEX_BIT,
 143                                 .module = vs_module_h,
 144                                 .pName = "main",
 145                         },
 146                         {
 147                                 .sType = VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_CREATE_INFO,
 148                                 .stage = VK_SHADER_STAGE_FRAGMENT_BIT,
 149                                 .module = radv_shader_module_to_handle(&fs_module),
 150                                 .pName = "main",
 151                         },
 152                 },
 153                 .pVertexInputState = &(VkPipelineVertexInputStateCreateInfo) {
 154                         .sType = VK_STRUCTURE_TYPE_PIPELINE_VERTEX_INPUT_STATE_CREATE_INFO,
 155                         .vertexBindingDescriptionCount = 1,
 156                         .pVertexBindingDescriptions = (VkVertexInputBindingDescription[]) {
 157                                 {
 158                                         .binding = 0,
 159                                         .stride = sizeof(struct vertex_attrs),
 160                                         .inputRate = VK_VERTEX_INPUT_RATE_VERTEX
 161                                 },
 162                         },
 163                         .vertexAttributeDescriptionCount = 1,
 164                         .pVertexAttributeDescriptions = (VkVertexInputAttributeDescription[]) {
 165                                 {
 166                                         /* Position */
 167                                         .location = 0,
 168                                         .binding = 0,
 169                                         .format = VK_FORMAT_R32G32_SFLOAT,
 170                                         .offset = offsetof(struct vertex_attrs, position),
 171                                 },
 172                         },
 173                 },
 174                 .pInputAssemblyState = &(VkPipelineInputAssemblyStateCreateInfo) {
 175                         .sType = VK_STRUCTURE_TYPE_PIPELINE_INPUT_ASSEMBLY_STATE_CREATE_INFO,
 176                         .topology = VK_PRIMITIVE_TOPOLOGY_TRIANGLE_STRIP,
 177                         .primitiveRestartEnable = false,
 178                 },
 179                 .pViewportState = &(VkPipelineViewportStateCreateInfo) {
 180                         .sType = VK_STRUCTURE_TYPE_PIPELINE_VIEWPORT_STATE_CREATE_INFO,
 181                         .viewportCount = 1,
 182                         .scissorCount = 1,
 183                 },
 184                 .pRasterizationState = &(VkPipelineRasterizationStateCreateInfo) {
 185                         .sType = VK_STRUCTURE_TYPE_PIPELINE_RASTERIZATION_STATE_CREATE_INFO,
 186                         .depthClampEnable = false,
 187                         .rasterizerDiscardEnable = false,
 188                         .polygonMode = VK_POLYGON_MODE_FILL,
 189                         .cullMode = VK_CULL_MODE_NONE,
 190                         .frontFace = VK_FRONT_FACE_COUNTER_CLOCKWISE,
 191                 },
 192                 .pMultisampleState = &(VkPipelineMultisampleStateCreateInfo) {
 193                         .sType = VK_STRUCTURE_TYPE_PIPELINE_MULTISAMPLE_STATE_CREATE_INFO,
 194                         .rasterizationSamples = 1,
 195                         .sampleShadingEnable = false,
 196                         .pSampleMask = NULL,
 197                         .alphaToCoverageEnable = false,
 198                         .alphaToOneEnable = false,
 199                 },
 200                 .pColorBlendState = &(VkPipelineColorBlendStateCreateInfo) {
 201                         .sType = VK_STRUCTURE_TYPE_PIPELINE_COLOR_BLEND_STATE_CREATE_INFO,
 202                         .logicOpEnable = false,
 203                         .attachmentCount = 0,
 204                         .pAttachments = NULL,
 205                 },
 206                 .pDepthStencilState = &(VkPipelineDepthStencilStateCreateInfo) {
 207                         .sType = VK_STRUCTURE_TYPE_PIPELINE_DEPTH_STENCIL_STATE_CREATE_INFO,
 208                         .depthTestEnable = false,
 209                         .depthWriteEnable = false,
 210                         .depthBoundsTestEnable = false,
 211                         .stencilTestEnable = false,
 212                 },
 213                 .pDynamicState = &(VkPipelineDynamicStateCreateInfo) {
 214                         .sType = VK_STRUCTURE_TYPE_PIPELINE_DYNAMIC_STATE_CREATE_INFO,
 215                         .dynamicStateCount = 2,
 216                         .pDynamicStates = (VkDynamicState[]) {
 217                                 VK_DYNAMIC_STATE_VIEWPORT,
 218                                 VK_DYNAMIC_STATE_SCISSOR,
 219                         },
 220                 },
 221                 .renderPass = device->meta_state.depth_decomp.pass,
 222                 .subpass = 0,
 223         };
 224
 225         result = radv_graphics_pipeline_create(device_h,
 226                                                radv_pipeline_cache_to_handle(&device->meta_state.cache),
 227                                                &pipeline_create_info,
 228                                                &(struct radv_graphics_pipeline_create_info) {
 229                                                         .use_rectlist = true,
 230                                                         .db_flush_depth_inplace = true,
 231                                                         .db_flush_stencil_inplace = true,
 232                                                },
 233                                                &device->meta_state.alloc,
 234                                                &device->meta_state.depth_decomp.decompress_pipeline);
 235         if (result != VK_SUCCESS)
 236                 goto cleanup;
 237
 238         result = radv_graphics_pipeline_create(device_h,
 239                                                radv_pipeline_cache_to_handle(&device->meta_state.cache),
 240                                                &pipeline_create_info,
 241                                                &(struct radv_graphics_pipeline_create_info) {
 242                                                         .use_rectlist = true,
 243                                                         .db_flush_depth_inplace = true,
 244                                                         .db_flush_stencil_inplace = true,
 245                                                         .db_resummarize = true,
 246                                                },
 247                                                &device->meta_state.alloc,
 248                                                &device->meta_state.depth_decomp.resummarize_pipeline);
 249         if (result != VK_SUCCESS)
 250                 goto cleanup;
 251
 252         goto cleanup;
 253
 254 cleanup:
 255         ralloc_free(fs_module.nir);
 256         return result;
 257 }
 258
 259 void
 260 radv_device_finish_meta_depth_decomp_state(struct radv_device *device)
 261 {
 262         struct radv_meta_state *state = &device->meta_state;
 263         VkDevice device_h = radv_device_to_handle(device);
 264         VkRenderPass pass_h = device->meta_state.depth_decomp.pass;
 265         const VkAllocationCallbacks *alloc = &device->meta_state.alloc;
 266
 267         if (pass_h)
 268                 radv_DestroyRenderPass(device_h, pass_h,
 269                                              &device->meta_state.alloc);
 270
 271         VkPipeline pipeline_h = state->depth_decomp.decompress_pipeline;
 272         if (pipeline_h) {
 273                 radv_DestroyPipeline(device_h, pipeline_h, alloc);
 274         }
 275         pipeline_h = state->depth_decomp.resummarize_pipeline;
 276         if (pipeline_h) {
 277                 radv_DestroyPipeline(device_h, pipeline_h, alloc);
 278         }
 279 }
 280
 281 VkResult
 282 radv_device_init_meta_depth_decomp_state(struct radv_device *device)
 283 {
 284         VkResult res = VK_SUCCESS;
 285
 286         zero(device->meta_state.depth_decomp);
 287
 288         struct radv_shader_module vs_module = { .nir = build_nir_vs() };
 289         if (!vs_module.nir) {
 290                 /* XXX: Need more accurate error */
 291                 res = VK_ERROR_OUT_OF_HOST_MEMORY;
 292                 goto fail;
 293         }
 294
 295         res = create_pass(device);
 296         if (res != VK_SUCCESS)
 297                 goto fail;
 298
 299         VkShaderModule vs_module_h = radv_shader_module_to_handle(&vs_module);
 300         res = create_pipeline(device, vs_module_h);
 301         if (res != VK_SUCCESS)
 302                 goto fail;
 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 void
 316 emit_depth_decomp(struct radv_cmd_buffer *cmd_buffer,
 317                   const VkOffset2D *dest_offset,
 318                   const VkExtent2D *depth_decomp_extent,
 319                   VkPipeline pipeline_h)
 320 {
 321         struct radv_device *device = cmd_buffer->device;
 322         VkCommandBuffer cmd_buffer_h = radv_cmd_buffer_to_handle(cmd_buffer);
 323         uint32_t offset;
 324         const struct vertex_attrs vertex_data[3] = {
 325                 {
 326                         .position = {
 327                                 -1.0,
 328                                 -1.0,
 329                         },
 330                 },
 331                 {
 332                         .position = {
 333                                 -1.0,
 334                                 1.0,
 335                         },
 336                 },
 337                 {
 338                         .position = {
 339                                 1.0,
 340                                 -1.0,
 341                         },
 342                 },
 343         };
 344
 345         radv_cmd_buffer_upload_data(cmd_buffer, sizeof(vertex_data), 16, vertex_data, &offset);
 346         struct radv_buffer vertex_buffer = {
 347                 .device = device,
 348                 .size = sizeof(vertex_data),
 349                 .bo = cmd_buffer->upload.upload_bo,
 350                 .offset = offset,
 351         };
 352
 353         VkBuffer vertex_buffer_h = radv_buffer_to_handle(&vertex_buffer);
 354
 355         radv_CmdBindVertexBuffers(cmd_buffer_h,
 356                                   /*firstBinding*/ 0,
 357                                   /*bindingCount*/ 1,
 358                                   (VkBuffer[]) { vertex_buffer_h },
 359                                   (VkDeviceSize[]) { 0 });
 360
 361         RADV_FROM_HANDLE(radv_pipeline, pipeline, pipeline_h);
 362
 363         if (cmd_buffer->state.pipeline != pipeline) {
 364                 radv_CmdBindPipeline(cmd_buffer_h, VK_PIPELINE_BIND_POINT_GRAPHICS,
 365                                      pipeline_h);
 366         }
 367
 368         radv_CmdSetViewport(radv_cmd_buffer_to_handle(cmd_buffer), 0, 1, &(VkViewport) {
 369                 .x = dest_offset->x,
 370                 .y = dest_offset->y,
 371                 .width = depth_decomp_extent->width,
 372                 .height = depth_decomp_extent->height,
 373                 .minDepth = 0.0f,
 374                 .maxDepth = 1.0f
 375         });
 376
 377         radv_CmdSetScissor(radv_cmd_buffer_to_handle(cmd_buffer), 0, 1, &(VkRect2D) {
 378                 .offset = *dest_offset,
 379                 .extent = *depth_decomp_extent,
 380         });
 381
 382         radv_CmdDraw(cmd_buffer_h, 3, 1, 0, 0);
 383 }
 384
 385
 386 static void radv_process_depth_image_inplace(struct radv_cmd_buffer *cmd_buffer,
 387                                              struct radv_image *image,
 388                                              VkImageSubresourceRange *subresourceRange,
 389                                              VkPipeline pipeline_h)
 390 {
 391         struct radv_meta_saved_state saved_state;
 392         struct radv_meta_saved_pass_state saved_pass_state;
 393         VkDevice device_h = radv_device_to_handle(cmd_buffer->device);
 394         VkCommandBuffer cmd_buffer_h = radv_cmd_buffer_to_handle(cmd_buffer);
 395         uint32_t width = radv_minify(image->extent.width,
 396                                      subresourceRange->baseMipLevel);
 397         uint32_t height = radv_minify(image->extent.height,
 398                                      subresourceRange->baseMipLevel);
 399
 400         if (!image->surface.htile_size)
 401                 return;
 402         radv_meta_save_pass(&saved_pass_state, cmd_buffer);
 403
 404         radv_meta_save_graphics_reset_vport_scissor(&saved_state, cmd_buffer);
 405
 406         for (uint32_t layer = 0; layer < radv_get_layerCount(image, subresourceRange); layer++) {
 407                 struct radv_image_view iview;
 408
 409                 radv_image_view_init(&iview, cmd_buffer->device,
 410                                      &(VkImageViewCreateInfo) {
 411                                              .sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO,
 412                                              .image = radv_image_to_handle(image),
 413                                              .format = image->vk_format,
 414                                              .subresourceRange = {
 415                                                      .aspectMask = VK_IMAGE_ASPECT_DEPTH_BIT,
 416                                                      .baseMipLevel = subresourceRange->baseMipLevel,
 417                                                      .levelCount = 1,
 418                                                      .baseArrayLayer = subresourceRange->baseArrayLayer + layer,
 419                                                      .layerCount = 1,
 420                                              },
 421                                      },
 422                                      cmd_buffer, VK_IMAGE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT);
 423
 424
 425                 VkFramebuffer fb_h;
 426                 radv_CreateFramebuffer(device_h,
 427                                        &(VkFramebufferCreateInfo) {
 428                                                .sType = VK_STRUCTURE_TYPE_FRAMEBUFFER_CREATE_INFO,
 429                                                .attachmentCount = 1,
 430                                                        .pAttachments = (VkImageView[]) {
 431                                                        radv_image_view_to_handle(&iview)
 432                                                },
 433                                                .width = width,
 434                                                 .height = height,
 435                                                .layers = 1
 436                                        },
 437                                        &cmd_buffer->pool->alloc,
 438                                        &fb_h);
 439
 440                 radv_CmdBeginRenderPass(cmd_buffer_h,
 441                                               &(VkRenderPassBeginInfo) {
 442                                                       .sType = VK_STRUCTURE_TYPE_RENDER_PASS_BEGIN_INFO,
 443                                                               .renderPass = cmd_buffer->device->meta_state.depth_decomp.pass,
 444                                                               .framebuffer = fb_h,
 445                                                               .renderArea = {
 446                                                               .offset = {
 447                                                                       0,
 448                                                                       0,
 449                                                               },
 450                                                               .extent = {
 451                                                                       width,
 452                                                                       height,
 453                                                               }
 454                                                        },
 455                                                        .clearValueCount = 0,
 456                                                        .pClearValues = NULL,
 457                                            },
 458                                            VK_SUBPASS_CONTENTS_INLINE);
 459
 460                 emit_depth_decomp(cmd_buffer, &(VkOffset2D){0, 0 }, &(VkExtent2D){width, height}, pipeline_h);
 461                 radv_CmdEndRenderPass(cmd_buffer_h);
 462
 463                 radv_DestroyFramebuffer(device_h, fb_h,
 464                                         &cmd_buffer->pool->alloc);
 465         }
 466         radv_meta_restore(&saved_state, cmd_buffer);
 467         radv_meta_restore_pass(&saved_pass_state, cmd_buffer);
 468 }
 469
 470 void radv_decompress_depth_image_inplace(struct radv_cmd_buffer *cmd_buffer,
 471                                          struct radv_image *image,
 472                                          VkImageSubresourceRange *subresourceRange)
 473 {
 474         assert(cmd_buffer->queue_family_index == RADV_QUEUE_GENERAL);
 475         radv_process_depth_image_inplace(cmd_buffer, image, subresourceRange,
 476                                          cmd_buffer->device->meta_state.depth_decomp.decompress_pipeline);
 477 }
 478
 479 void radv_resummarize_depth_image_inplace(struct radv_cmd_buffer *cmd_buffer,
 480                                          struct radv_image *image,
 481                                          VkImageSubresourceRange *subresourceRange)
 482 {
 483         assert(cmd_buffer->queue_family_index == RADV_QUEUE_GENERAL);
 484         radv_process_depth_image_inplace(cmd_buffer, image, subresourceRange,
 485                                          cmd_buffer->device->meta_state.depth_decomp.resummarize_pipeline);
 486 }