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.format = VK_FORMAT_D32_SFLOAT_S8_UINT;
  42         attachment.samples = samples;
  43         attachment.loadOp = VK_ATTACHMENT_LOAD_OP_LOAD;
  44         attachment.storeOp = VK_ATTACHMENT_STORE_OP_STORE;
  45         attachment.initialLayout = VK_IMAGE_LAYOUT_DEPTH_STENCIL_ATTACHMENT_OPTIMAL;
  46         attachment.finalLayout = VK_IMAGE_LAYOUT_DEPTH_STENCIL_ATTACHMENT_OPTIMAL;
  47
  48         result = radv_CreateRenderPass(device_h,
  49                                        &(VkRenderPassCreateInfo) {
  50                                                .sType = VK_STRUCTURE_TYPE_RENDER_PASS_CREATE_INFO,
  51                                                        .attachmentCount = 1,
  52                                                        .pAttachments = &attachment,
  53                                                        .subpassCount = 1,
  54                                                         .pSubpasses = &(VkSubpassDescription) {
  55                                                        .pipelineBindPoint = VK_PIPELINE_BIND_POINT_GRAPHICS,
  56                                                        .inputAttachmentCount = 0,
  57                                                        .colorAttachmentCount = 0,
  58                                                        .pColorAttachments = NULL,
  59                                                        .pResolveAttachments = NULL,
  60                                                        .pDepthStencilAttachment = &(VkAttachmentReference) {
  61                                                                .attachment = 0,
  62                                                                .layout = VK_IMAGE_LAYOUT_DEPTH_STENCIL_ATTACHMENT_OPTIMAL,
  63                                                        },
  64                                                        .preserveAttachmentCount = 0,
  65                                                        .pPreserveAttachments = NULL,
  66                                                },
  67                                                                 .dependencyCount = 0,
  68                                                                    },
  69                                        alloc,
  70                                        pass);
  71
  72         return result;
  73 }
  74
  75 static VkResult
  76 create_pipeline(struct radv_device *device,
  77                 VkShaderModule vs_module_h,
  78                 uint32_t samples,
  79                 VkRenderPass pass,
  80                 VkPipeline *decompress_pipeline,
  81                 VkPipeline *resummarize_pipeline)
  82 {
  83         VkResult result;
  84         VkDevice device_h = radv_device_to_handle(device);
  85
  86         struct radv_shader_module fs_module = {
  87                 .nir = radv_meta_build_nir_fs_noop(),
  88         };
  89
  90         if (!fs_module.nir) {
  91                 /* XXX: Need more accurate error */
  92                 result = VK_ERROR_OUT_OF_HOST_MEMORY;
  93                 goto cleanup;
  94         }
  95
  96         const VkGraphicsPipelineCreateInfo pipeline_create_info = {
  97                 .sType = VK_STRUCTURE_TYPE_GRAPHICS_PIPELINE_CREATE_INFO,
  98                 .stageCount = 2,
  99                 .pStages = (VkPipelineShaderStageCreateInfo[]) {
 100                        {
 101                                 .sType = VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_CREATE_INFO,
 102                                 .stage = VK_SHADER_STAGE_VERTEX_BIT,
 103                                 .module = vs_module_h,
 104                                 .pName = "main",
 105                         },
 106                         {
 107                                 .sType = VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_CREATE_INFO,
 108                                 .stage = VK_SHADER_STAGE_FRAGMENT_BIT,
 109                                 .module = radv_shader_module_to_handle(&fs_module),
 110                                 .pName = "main",
 111                         },
 112                 },
 113                 .pVertexInputState = &(VkPipelineVertexInputStateCreateInfo) {
 114                         .sType = VK_STRUCTURE_TYPE_PIPELINE_VERTEX_INPUT_STATE_CREATE_INFO,
 115                         .vertexBindingDescriptionCount = 0,
 116                         .vertexAttributeDescriptionCount = 0,
 117                 },
 118                 .pInputAssemblyState = &(VkPipelineInputAssemblyStateCreateInfo) {
 119                         .sType = VK_STRUCTURE_TYPE_PIPELINE_INPUT_ASSEMBLY_STATE_CREATE_INFO,
 120                         .topology = VK_PRIMITIVE_TOPOLOGY_TRIANGLE_STRIP,
 121                         .primitiveRestartEnable = false,
 122                 },
 123                 .pViewportState = &(VkPipelineViewportStateCreateInfo) {
 124                         .sType = VK_STRUCTURE_TYPE_PIPELINE_VIEWPORT_STATE_CREATE_INFO,
 125                         .viewportCount = 1,
 126                         .scissorCount = 1,
 127                 },
 128                 .pRasterizationState = &(VkPipelineRasterizationStateCreateInfo) {
 129                         .sType = VK_STRUCTURE_TYPE_PIPELINE_RASTERIZATION_STATE_CREATE_INFO,
 130                         .depthClampEnable = false,
 131                         .rasterizerDiscardEnable = false,
 132                         .polygonMode = VK_POLYGON_MODE_FILL,
 133                         .cullMode = VK_CULL_MODE_NONE,
 134                         .frontFace = VK_FRONT_FACE_COUNTER_CLOCKWISE,
 135                 },
 136                 .pMultisampleState = &(VkPipelineMultisampleStateCreateInfo) {
 137                         .sType = VK_STRUCTURE_TYPE_PIPELINE_MULTISAMPLE_STATE_CREATE_INFO,
 138                         .rasterizationSamples = samples,
 139                         .sampleShadingEnable = false,
 140                         .pSampleMask = NULL,
 141                         .alphaToCoverageEnable = false,
 142                         .alphaToOneEnable = false,
 143                 },
 144                 .pColorBlendState = &(VkPipelineColorBlendStateCreateInfo) {
 145                         .sType = VK_STRUCTURE_TYPE_PIPELINE_COLOR_BLEND_STATE_CREATE_INFO,
 146                         .logicOpEnable = false,
 147                         .attachmentCount = 0,
 148                         .pAttachments = NULL,
 149                 },
 150                 .pDepthStencilState = &(VkPipelineDepthStencilStateCreateInfo) {
 151                         .sType = VK_STRUCTURE_TYPE_PIPELINE_DEPTH_STENCIL_STATE_CREATE_INFO,
 152                         .depthTestEnable = false,
 153                         .depthWriteEnable = false,
 154                         .depthBoundsTestEnable = false,
 155                         .stencilTestEnable = false,
 156                 },
 157                 .pDynamicState = &(VkPipelineDynamicStateCreateInfo) {
 158                         .sType = VK_STRUCTURE_TYPE_PIPELINE_DYNAMIC_STATE_CREATE_INFO,
 159                         .dynamicStateCount = 2,
 160                         .pDynamicStates = (VkDynamicState[]) {
 161                                 VK_DYNAMIC_STATE_VIEWPORT,
 162                                 VK_DYNAMIC_STATE_SCISSOR,
 163                         },
 164                 },
 165                 .renderPass = pass,
 166                 .subpass = 0,
 167         };
 168
 169         result = radv_graphics_pipeline_create(device_h,
 170                                                radv_pipeline_cache_to_handle(&device->meta_state.cache),
 171                                                &pipeline_create_info,
 172                                                &(struct radv_graphics_pipeline_create_info) {
 173                                                         .use_rectlist = true,
 174                                                         .db_flush_depth_inplace = true,
 175                                                         .db_flush_stencil_inplace = true,
 176                                                },
 177                                                &device->meta_state.alloc,
 178                                                decompress_pipeline);
 179         if (result != VK_SUCCESS)
 180                 goto cleanup;
 181
 182         result = radv_graphics_pipeline_create(device_h,
 183                                                radv_pipeline_cache_to_handle(&device->meta_state.cache),
 184                                                &pipeline_create_info,
 185                                                &(struct radv_graphics_pipeline_create_info) {
 186                                                         .use_rectlist = true,
 187                                                         .db_flush_depth_inplace = true,
 188                                                         .db_flush_stencil_inplace = true,
 189                                                         .db_resummarize = true,
 190                                                },
 191                                                &device->meta_state.alloc,
 192                                                resummarize_pipeline);
 193         if (result != VK_SUCCESS)
 194                 goto cleanup;
 195
 196         goto cleanup;
 197
 198 cleanup:
 199         ralloc_free(fs_module.nir);
 200         return result;
 201 }
 202
 203 void
 204 radv_device_finish_meta_depth_decomp_state(struct radv_device *device)
 205 {
 206         struct radv_meta_state *state = &device->meta_state;
 207         VkDevice device_h = radv_device_to_handle(device);
 208         const VkAllocationCallbacks *alloc = &device->meta_state.alloc;
 209
 210         for (uint32_t i = 0; i < ARRAY_SIZE(state->depth_decomp); ++i) {
 211                 VkRenderPass pass_h = state->depth_decomp[i].pass;
 212                 if (pass_h) {
 213                         radv_DestroyRenderPass(device_h, pass_h, alloc);
 214                 }
 215                 VkPipeline pipeline_h = state->depth_decomp[i].decompress_pipeline;
 216                 if (pipeline_h) {
 217                         radv_DestroyPipeline(device_h, pipeline_h, alloc);
 218                 }
 219                 pipeline_h = state->depth_decomp[i].resummarize_pipeline;
 220                 if (pipeline_h) {
 221                         radv_DestroyPipeline(device_h, pipeline_h, alloc);
 222                 }
 223         }
 224 }
 225
 226 VkResult
 227 radv_device_init_meta_depth_decomp_state(struct radv_device *device)
 228 {
 229         struct radv_meta_state *state = &device->meta_state;
 230         VkResult res = VK_SUCCESS;
 231
 232         zero(state->depth_decomp);
 233
 234         struct radv_shader_module vs_module = { .nir = radv_meta_build_nir_vs_generate_vertices() };
 235         if (!vs_module.nir) {
 236                 /* XXX: Need more accurate error */
 237                 res = VK_ERROR_OUT_OF_HOST_MEMORY;
 238                 goto fail;
 239         }
 240
 241         VkShaderModule vs_module_h = radv_shader_module_to_handle(&vs_module);
 242
 243         for (uint32_t i = 0; i < ARRAY_SIZE(state->depth_decomp); ++i) {
 244                 uint32_t samples = 1 << i;
 245
 246                 res = create_pass(device, samples, &state->depth_decomp[i].pass);
 247                 if (res != VK_SUCCESS)
 248                         goto fail;
 249
 250                 res = create_pipeline(device, vs_module_h, samples,
 251                                       state->depth_decomp[i].pass,
 252                                       &state->depth_decomp[i].decompress_pipeline,
 253                                       &state->depth_decomp[i].resummarize_pipeline);
 254                 if (res != VK_SUCCESS)
 255                         goto fail;
 256         }
 257
 258         goto cleanup;
 259
 260 fail:
 261         radv_device_finish_meta_depth_decomp_state(device);
 262
 263 cleanup:
 264         ralloc_free(vs_module.nir);
 265
 266         return res;
 267 }
 268
 269 static void
 270 emit_depth_decomp(struct radv_cmd_buffer *cmd_buffer,
 271                   const VkOffset2D *dest_offset,
 272                   const VkExtent2D *depth_decomp_extent,
 273                   VkPipeline pipeline_h)
 274 {
 275         VkCommandBuffer cmd_buffer_h = radv_cmd_buffer_to_handle(cmd_buffer);
 276
 277         RADV_FROM_HANDLE(radv_pipeline, pipeline, pipeline_h);
 278
 279         if (cmd_buffer->state.pipeline != pipeline) {
 280                 radv_CmdBindPipeline(cmd_buffer_h, VK_PIPELINE_BIND_POINT_GRAPHICS,
 281                                      pipeline_h);
 282         }
 283
 284         radv_CmdSetViewport(radv_cmd_buffer_to_handle(cmd_buffer), 0, 1, &(VkViewport) {
 285                 .x = dest_offset->x,
 286                 .y = dest_offset->y,
 287                 .width = depth_decomp_extent->width,
 288                 .height = depth_decomp_extent->height,
 289                 .minDepth = 0.0f,
 290                 .maxDepth = 1.0f
 291         });
 292
 293         radv_CmdSetScissor(radv_cmd_buffer_to_handle(cmd_buffer), 0, 1, &(VkRect2D) {
 294                 .offset = *dest_offset,
 295                 .extent = *depth_decomp_extent,
 296         });
 297
 298         radv_CmdDraw(cmd_buffer_h, 3, 1, 0, 0);
 299 }
 300
 301
 302 enum radv_depth_op {
 303         DEPTH_DECOMPRESS,
 304         DEPTH_RESUMMARIZE,
 305 };
 306
 307 static void radv_process_depth_image_inplace(struct radv_cmd_buffer *cmd_buffer,
 308                                              struct radv_image *image,
 309                                              VkImageSubresourceRange *subresourceRange,
 310                                              enum radv_depth_op op)
 311 {
 312         struct radv_meta_saved_state saved_state;
 313         struct radv_meta_saved_pass_state saved_pass_state;
 314         VkDevice device_h = radv_device_to_handle(cmd_buffer->device);
 315         VkCommandBuffer cmd_buffer_h = radv_cmd_buffer_to_handle(cmd_buffer);
 316         uint32_t width = radv_minify(image->info.width,
 317                                      subresourceRange->baseMipLevel);
 318         uint32_t height = radv_minify(image->info.height,
 319                                      subresourceRange->baseMipLevel);
 320         uint32_t samples = image->info.samples;
 321         uint32_t samples_log2 = ffs(samples) - 1;
 322         struct radv_meta_state *meta_state = &cmd_buffer->device->meta_state;
 323
 324         if (!image->surface.htile_size)
 325                 return;
 326         radv_meta_save_pass(&saved_pass_state, cmd_buffer);
 327
 328         radv_meta_save_graphics_reset_vport_scissor_novertex(&saved_state, cmd_buffer);
 329
 330         for (uint32_t layer = 0; layer < radv_get_layerCount(image, subresourceRange); layer++) {
 331                 struct radv_image_view iview;
 332
 333                 radv_image_view_init(&iview, cmd_buffer->device,
 334                                      &(VkImageViewCreateInfo) {
 335                                              .sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO,
 336                                              .image = radv_image_to_handle(image),
 337                                              .format = image->vk_format,
 338                                              .subresourceRange = {
 339                                                      .aspectMask = VK_IMAGE_ASPECT_DEPTH_BIT,
 340                                                      .baseMipLevel = subresourceRange->baseMipLevel,
 341                                                      .levelCount = 1,
 342                                                      .baseArrayLayer = subresourceRange->baseArrayLayer + layer,
 343                                                      .layerCount = 1,
 344                                              },
 345                                      });
 346
 347
 348                 VkFramebuffer fb_h;
 349                 radv_CreateFramebuffer(device_h,
 350                                        &(VkFramebufferCreateInfo) {
 351                                                .sType = VK_STRUCTURE_TYPE_FRAMEBUFFER_CREATE_INFO,
 352                                                .attachmentCount = 1,
 353                                                        .pAttachments = (VkImageView[]) {
 354                                                        radv_image_view_to_handle(&iview)
 355                                                },
 356                                                .width = width,
 357                                                 .height = height,
 358                                                .layers = 1
 359                                        },
 360                                        &cmd_buffer->pool->alloc,
 361                                        &fb_h);
 362
 363                 radv_CmdBeginRenderPass(cmd_buffer_h,
 364                                               &(VkRenderPassBeginInfo) {
 365                                                       .sType = VK_STRUCTURE_TYPE_RENDER_PASS_BEGIN_INFO,
 366                                                               .renderPass = meta_state->depth_decomp[samples_log2].pass,
 367                                                               .framebuffer = fb_h,
 368                                                               .renderArea = {
 369                                                               .offset = {
 370                                                                       0,
 371                                                                       0,
 372                                                               },
 373                                                               .extent = {
 374                                                                       width,
 375                                                                       height,
 376                                                               }
 377                                                        },
 378                                                        .clearValueCount = 0,
 379                                                        .pClearValues = NULL,
 380                                            },
 381                                            VK_SUBPASS_CONTENTS_INLINE);
 382
 383                 VkPipeline pipeline_h;
 384                 switch (op) {
 385                 case DEPTH_DECOMPRESS:
 386                         pipeline_h = meta_state->depth_decomp[samples_log2].decompress_pipeline;
 387                         break;
 388                 case DEPTH_RESUMMARIZE:
 389                         pipeline_h = meta_state->depth_decomp[samples_log2].resummarize_pipeline;
 390                         break;
 391                 default:
 392                         unreachable("unknown operation");
 393                 }
 394
 395                 emit_depth_decomp(cmd_buffer, &(VkOffset2D){0, 0 }, &(VkExtent2D){width, height}, pipeline_h);
 396                 radv_CmdEndRenderPass(cmd_buffer_h);
 397
 398                 radv_DestroyFramebuffer(device_h, fb_h,
 399                                         &cmd_buffer->pool->alloc);
 400         }
 401         radv_meta_restore(&saved_state, cmd_buffer);
 402         radv_meta_restore_pass(&saved_pass_state, cmd_buffer);
 403 }
 404
 405 void radv_decompress_depth_image_inplace(struct radv_cmd_buffer *cmd_buffer,
 406                                          struct radv_image *image,
 407                                          VkImageSubresourceRange *subresourceRange)
 408 {
 409         assert(cmd_buffer->queue_family_index == RADV_QUEUE_GENERAL);
 410         radv_process_depth_image_inplace(cmd_buffer, image, subresourceRange, DEPTH_DECOMPRESS);
 411 }
 412
 413 void radv_resummarize_depth_image_inplace(struct radv_cmd_buffer *cmd_buffer,
 414                                          struct radv_image *image,
 415                                          VkImageSubresourceRange *subresourceRange)
 416 {
 417         assert(cmd_buffer->queue_family_index == RADV_QUEUE_GENERAL);
 418         radv_process_depth_image_inplace(cmd_buffer, image, subresourceRange, DEPTH_RESUMMARIZE);
 419 }