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 {
  34         VkResult result;
  35         VkDevice device_h = radv_device_to_handle(device);
  36         const VkAllocationCallbacks *alloc = &device->meta_state.alloc;
  37         VkAttachmentDescription attachment;
  38
  39         attachment.format = VK_FORMAT_UNDEFINED;
  40         attachment.samples = 1;
  41         attachment.loadOp = VK_ATTACHMENT_LOAD_OP_LOAD;
  42         attachment.storeOp = VK_ATTACHMENT_STORE_OP_STORE;
  43         attachment.initialLayout = VK_IMAGE_LAYOUT_DEPTH_STENCIL_ATTACHMENT_OPTIMAL;
  44         attachment.finalLayout = VK_IMAGE_LAYOUT_DEPTH_STENCIL_ATTACHMENT_OPTIMAL;
  45
  46         result = radv_CreateRenderPass(device_h,
  47                                        &(VkRenderPassCreateInfo) {
  48                                                .sType = VK_STRUCTURE_TYPE_RENDER_PASS_CREATE_INFO,
  49                                                        .attachmentCount = 1,
  50                                                        .pAttachments = &attachment,
  51                                                        .subpassCount = 1,
  52                                                         .pSubpasses = &(VkSubpassDescription) {
  53                                                        .pipelineBindPoint = VK_PIPELINE_BIND_POINT_GRAPHICS,
  54                                                        .inputAttachmentCount = 0,
  55                                                        .colorAttachmentCount = 0,
  56                                                        .pColorAttachments = NULL,
  57                                                        .pResolveAttachments = NULL,
  58                                                        .pDepthStencilAttachment = &(VkAttachmentReference) {
  59                                                                .attachment = 0,
  60                                                                .layout = VK_IMAGE_LAYOUT_DEPTH_STENCIL_ATTACHMENT_OPTIMAL,
  61                                                        },
  62                                                        .preserveAttachmentCount = 0,
  63                                                        .pPreserveAttachments = NULL,
  64                                                },
  65                                                                 .dependencyCount = 0,
  66                                                                    },
  67                                        alloc,
  68                                        &device->meta_state.depth_decomp.pass);
  69
  70         return result;
  71 }
  72
  73 static VkResult
  74 create_pipeline(struct radv_device *device,
  75                 VkShaderModule vs_module_h)
  76 {
  77         VkResult result;
  78         VkDevice device_h = radv_device_to_handle(device);
  79
  80         struct radv_shader_module fs_module = {
  81                 .nir = radv_meta_build_nir_fs_noop(),
  82         };
  83
  84         if (!fs_module.nir) {
  85                 /* XXX: Need more accurate error */
  86                 result = VK_ERROR_OUT_OF_HOST_MEMORY;
  87                 goto cleanup;
  88         }
  89
  90         const VkGraphicsPipelineCreateInfo pipeline_create_info = {
  91                 .sType = VK_STRUCTURE_TYPE_GRAPHICS_PIPELINE_CREATE_INFO,
  92                 .stageCount = 2,
  93                 .pStages = (VkPipelineShaderStageCreateInfo[]) {
  94                        {
  95                                 .sType = VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_CREATE_INFO,
  96                                 .stage = VK_SHADER_STAGE_VERTEX_BIT,
  97                                 .module = vs_module_h,
  98                                 .pName = "main",
  99                         },
 100                         {
 101                                 .sType = VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_CREATE_INFO,
 102                                 .stage = VK_SHADER_STAGE_FRAGMENT_BIT,
 103                                 .module = radv_shader_module_to_handle(&fs_module),
 104                                 .pName = "main",
 105                         },
 106                 },
 107                 .pVertexInputState = &(VkPipelineVertexInputStateCreateInfo) {
 108                         .sType = VK_STRUCTURE_TYPE_PIPELINE_VERTEX_INPUT_STATE_CREATE_INFO,
 109                         .vertexBindingDescriptionCount = 0,
 110                         .vertexAttributeDescriptionCount = 0,
 111                 },
 112                 .pInputAssemblyState = &(VkPipelineInputAssemblyStateCreateInfo) {
 113                         .sType = VK_STRUCTURE_TYPE_PIPELINE_INPUT_ASSEMBLY_STATE_CREATE_INFO,
 114                         .topology = VK_PRIMITIVE_TOPOLOGY_TRIANGLE_STRIP,
 115                         .primitiveRestartEnable = false,
 116                 },
 117                 .pViewportState = &(VkPipelineViewportStateCreateInfo) {
 118                         .sType = VK_STRUCTURE_TYPE_PIPELINE_VIEWPORT_STATE_CREATE_INFO,
 119                         .viewportCount = 1,
 120                         .scissorCount = 1,
 121                 },
 122                 .pRasterizationState = &(VkPipelineRasterizationStateCreateInfo) {
 123                         .sType = VK_STRUCTURE_TYPE_PIPELINE_RASTERIZATION_STATE_CREATE_INFO,
 124                         .depthClampEnable = false,
 125                         .rasterizerDiscardEnable = false,
 126                         .polygonMode = VK_POLYGON_MODE_FILL,
 127                         .cullMode = VK_CULL_MODE_NONE,
 128                         .frontFace = VK_FRONT_FACE_COUNTER_CLOCKWISE,
 129                 },
 130                 .pMultisampleState = &(VkPipelineMultisampleStateCreateInfo) {
 131                         .sType = VK_STRUCTURE_TYPE_PIPELINE_MULTISAMPLE_STATE_CREATE_INFO,
 132                         .rasterizationSamples = 1,
 133                         .sampleShadingEnable = false,
 134                         .pSampleMask = NULL,
 135                         .alphaToCoverageEnable = false,
 136                         .alphaToOneEnable = false,
 137                 },
 138                 .pColorBlendState = &(VkPipelineColorBlendStateCreateInfo) {
 139                         .sType = VK_STRUCTURE_TYPE_PIPELINE_COLOR_BLEND_STATE_CREATE_INFO,
 140                         .logicOpEnable = false,
 141                         .attachmentCount = 0,
 142                         .pAttachments = NULL,
 143                 },
 144                 .pDepthStencilState = &(VkPipelineDepthStencilStateCreateInfo) {
 145                         .sType = VK_STRUCTURE_TYPE_PIPELINE_DEPTH_STENCIL_STATE_CREATE_INFO,
 146                         .depthTestEnable = false,
 147                         .depthWriteEnable = false,
 148                         .depthBoundsTestEnable = false,
 149                         .stencilTestEnable = false,
 150                 },
 151                 .pDynamicState = &(VkPipelineDynamicStateCreateInfo) {
 152                         .sType = VK_STRUCTURE_TYPE_PIPELINE_DYNAMIC_STATE_CREATE_INFO,
 153                         .dynamicStateCount = 2,
 154                         .pDynamicStates = (VkDynamicState[]) {
 155                                 VK_DYNAMIC_STATE_VIEWPORT,
 156                                 VK_DYNAMIC_STATE_SCISSOR,
 157                         },
 158                 },
 159                 .renderPass = device->meta_state.depth_decomp.pass,
 160                 .subpass = 0,
 161         };
 162
 163         result = radv_graphics_pipeline_create(device_h,
 164                                                radv_pipeline_cache_to_handle(&device->meta_state.cache),
 165                                                &pipeline_create_info,
 166                                                &(struct radv_graphics_pipeline_create_info) {
 167                                                         .use_rectlist = true,
 168                                                         .db_flush_depth_inplace = true,
 169                                                         .db_flush_stencil_inplace = true,
 170                                                },
 171                                                &device->meta_state.alloc,
 172                                                &device->meta_state.depth_decomp.decompress_pipeline);
 173         if (result != VK_SUCCESS)
 174                 goto cleanup;
 175
 176         result = radv_graphics_pipeline_create(device_h,
 177                                                radv_pipeline_cache_to_handle(&device->meta_state.cache),
 178                                                &pipeline_create_info,
 179                                                &(struct radv_graphics_pipeline_create_info) {
 180                                                         .use_rectlist = true,
 181                                                         .db_flush_depth_inplace = true,
 182                                                         .db_flush_stencil_inplace = true,
 183                                                         .db_resummarize = true,
 184                                                },
 185                                                &device->meta_state.alloc,
 186                                                &device->meta_state.depth_decomp.resummarize_pipeline);
 187         if (result != VK_SUCCESS)
 188                 goto cleanup;
 189
 190         goto cleanup;
 191
 192 cleanup:
 193         ralloc_free(fs_module.nir);
 194         return result;
 195 }
 196
 197 void
 198 radv_device_finish_meta_depth_decomp_state(struct radv_device *device)
 199 {
 200         struct radv_meta_state *state = &device->meta_state;
 201         VkDevice device_h = radv_device_to_handle(device);
 202         VkRenderPass pass_h = device->meta_state.depth_decomp.pass;
 203         const VkAllocationCallbacks *alloc = &device->meta_state.alloc;
 204
 205         if (pass_h)
 206                 radv_DestroyRenderPass(device_h, pass_h,
 207                                              &device->meta_state.alloc);
 208
 209         VkPipeline pipeline_h = state->depth_decomp.decompress_pipeline;
 210         if (pipeline_h) {
 211                 radv_DestroyPipeline(device_h, pipeline_h, alloc);
 212         }
 213         pipeline_h = state->depth_decomp.resummarize_pipeline;
 214         if (pipeline_h) {
 215                 radv_DestroyPipeline(device_h, pipeline_h, alloc);
 216         }
 217 }
 218
 219 VkResult
 220 radv_device_init_meta_depth_decomp_state(struct radv_device *device)
 221 {
 222         VkResult res = VK_SUCCESS;
 223
 224         zero(device->meta_state.depth_decomp);
 225
 226         struct radv_shader_module vs_module = { .nir = radv_meta_build_nir_vs_generate_vertices() };
 227         if (!vs_module.nir) {
 228                 /* XXX: Need more accurate error */
 229                 res = VK_ERROR_OUT_OF_HOST_MEMORY;
 230                 goto fail;
 231         }
 232
 233         res = create_pass(device);
 234         if (res != VK_SUCCESS)
 235                 goto fail;
 236
 237         VkShaderModule vs_module_h = radv_shader_module_to_handle(&vs_module);
 238         res = create_pipeline(device, vs_module_h);
 239         if (res != VK_SUCCESS)
 240                 goto fail;
 241
 242         goto cleanup;
 243
 244 fail:
 245         radv_device_finish_meta_depth_decomp_state(device);
 246
 247 cleanup:
 248         ralloc_free(vs_module.nir);
 249
 250         return res;
 251 }
 252
 253 static void
 254 emit_depth_decomp(struct radv_cmd_buffer *cmd_buffer,
 255                   const VkOffset2D *dest_offset,
 256                   const VkExtent2D *depth_decomp_extent,
 257                   VkPipeline pipeline_h)
 258 {
 259         VkCommandBuffer cmd_buffer_h = radv_cmd_buffer_to_handle(cmd_buffer);
 260
 261         RADV_FROM_HANDLE(radv_pipeline, pipeline, pipeline_h);
 262
 263         if (cmd_buffer->state.pipeline != pipeline) {
 264                 radv_CmdBindPipeline(cmd_buffer_h, VK_PIPELINE_BIND_POINT_GRAPHICS,
 265                                      pipeline_h);
 266         }
 267
 268         radv_CmdSetViewport(radv_cmd_buffer_to_handle(cmd_buffer), 0, 1, &(VkViewport) {
 269                 .x = dest_offset->x,
 270                 .y = dest_offset->y,
 271                 .width = depth_decomp_extent->width,
 272                 .height = depth_decomp_extent->height,
 273                 .minDepth = 0.0f,
 274                 .maxDepth = 1.0f
 275         });
 276
 277         radv_CmdSetScissor(radv_cmd_buffer_to_handle(cmd_buffer), 0, 1, &(VkRect2D) {
 278                 .offset = *dest_offset,
 279                 .extent = *depth_decomp_extent,
 280         });
 281
 282         radv_CmdDraw(cmd_buffer_h, 3, 1, 0, 0);
 283 }
 284
 285
 286 static void radv_process_depth_image_inplace(struct radv_cmd_buffer *cmd_buffer,
 287                                              struct radv_image *image,
 288                                              VkImageSubresourceRange *subresourceRange,
 289                                              VkPipeline pipeline_h)
 290 {
 291         struct radv_meta_saved_state saved_state;
 292         struct radv_meta_saved_pass_state saved_pass_state;
 293         VkDevice device_h = radv_device_to_handle(cmd_buffer->device);
 294         VkCommandBuffer cmd_buffer_h = radv_cmd_buffer_to_handle(cmd_buffer);
 295         uint32_t width = radv_minify(image->extent.width,
 296                                      subresourceRange->baseMipLevel);
 297         uint32_t height = radv_minify(image->extent.height,
 298                                      subresourceRange->baseMipLevel);
 299
 300         if (!image->surface.htile_size)
 301                 return;
 302         radv_meta_save_pass(&saved_pass_state, cmd_buffer);
 303
 304         radv_meta_save_graphics_reset_vport_scissor_novertex(&saved_state, cmd_buffer);
 305
 306         for (uint32_t layer = 0; layer < radv_get_layerCount(image, subresourceRange); layer++) {
 307                 struct radv_image_view iview;
 308
 309                 radv_image_view_init(&iview, cmd_buffer->device,
 310                                      &(VkImageViewCreateInfo) {
 311                                              .sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO,
 312                                              .image = radv_image_to_handle(image),
 313                                              .format = image->vk_format,
 314                                              .subresourceRange = {
 315                                                      .aspectMask = VK_IMAGE_ASPECT_DEPTH_BIT,
 316                                                      .baseMipLevel = subresourceRange->baseMipLevel,
 317                                                      .levelCount = 1,
 318                                                      .baseArrayLayer = subresourceRange->baseArrayLayer + layer,
 319                                                      .layerCount = 1,
 320                                              },
 321                                      },
 322                                      cmd_buffer, VK_IMAGE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT);
 323
 324
 325                 VkFramebuffer fb_h;
 326                 radv_CreateFramebuffer(device_h,
 327                                        &(VkFramebufferCreateInfo) {
 328                                                .sType = VK_STRUCTURE_TYPE_FRAMEBUFFER_CREATE_INFO,
 329                                                .attachmentCount = 1,
 330                                                        .pAttachments = (VkImageView[]) {
 331                                                        radv_image_view_to_handle(&iview)
 332                                                },
 333                                                .width = width,
 334                                                 .height = height,
 335                                                .layers = 1
 336                                        },
 337                                        &cmd_buffer->pool->alloc,
 338                                        &fb_h);
 339
 340                 radv_CmdBeginRenderPass(cmd_buffer_h,
 341                                               &(VkRenderPassBeginInfo) {
 342                                                       .sType = VK_STRUCTURE_TYPE_RENDER_PASS_BEGIN_INFO,
 343                                                               .renderPass = cmd_buffer->device->meta_state.depth_decomp.pass,
 344                                                               .framebuffer = fb_h,
 345                                                               .renderArea = {
 346                                                               .offset = {
 347                                                                       0,
 348                                                                       0,
 349                                                               },
 350                                                               .extent = {
 351                                                                       width,
 352                                                                       height,
 353                                                               }
 354                                                        },
 355                                                        .clearValueCount = 0,
 356                                                        .pClearValues = NULL,
 357                                            },
 358                                            VK_SUBPASS_CONTENTS_INLINE);
 359
 360                 emit_depth_decomp(cmd_buffer, &(VkOffset2D){0, 0 }, &(VkExtent2D){width, height}, pipeline_h);
 361                 radv_CmdEndRenderPass(cmd_buffer_h);
 362
 363                 radv_DestroyFramebuffer(device_h, fb_h,
 364                                         &cmd_buffer->pool->alloc);
 365         }
 366         radv_meta_restore(&saved_state, cmd_buffer);
 367         radv_meta_restore_pass(&saved_pass_state, cmd_buffer);
 368 }
 369
 370 void radv_decompress_depth_image_inplace(struct radv_cmd_buffer *cmd_buffer,
 371                                          struct radv_image *image,
 372                                          VkImageSubresourceRange *subresourceRange)
 373 {
 374         assert(cmd_buffer->queue_family_index == RADV_QUEUE_GENERAL);
 375         radv_process_depth_image_inplace(cmd_buffer, image, subresourceRange,
 376                                          cmd_buffer->device->meta_state.depth_decomp.decompress_pipeline);
 377 }
 378
 379 void radv_resummarize_depth_image_inplace(struct radv_cmd_buffer *cmd_buffer,
 380                                          struct radv_image *image,
 381                                          VkImageSubresourceRange *subresourceRange)
 382 {
 383         assert(cmd_buffer->queue_family_index == RADV_QUEUE_GENERAL);
 384         radv_process_depth_image_inplace(cmd_buffer, image, subresourceRange,
 385                                          cmd_buffer->device->meta_state.depth_decomp.resummarize_pipeline);
 386 }