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(struct radv_device *device,
  80                 VkShaderModule vs_module_h,
  81                 uint32_t samples,
  82                 VkRenderPass pass,
  83                 VkPipeline *decompress_pipeline,
  84                 VkPipeline *resummarize_pipeline)
  85 {
  86         VkResult result;
  87         VkDevice device_h = radv_device_to_handle(device);
  88
  89         struct radv_shader_module fs_module = {
  90                 .nir = radv_meta_build_nir_fs_noop(),
  91         };
  92
  93         if (!fs_module.nir) {
  94                 /* XXX: Need more accurate error */
  95                 result = VK_ERROR_OUT_OF_HOST_MEMORY;
  96                 goto cleanup;
  97         }
  98
  99         const VkGraphicsPipelineCreateInfo pipeline_create_info = {
 100                 .sType = VK_STRUCTURE_TYPE_GRAPHICS_PIPELINE_CREATE_INFO,
 101                 .stageCount = 2,
 102                 .pStages = (VkPipelineShaderStageCreateInfo[]) {
 103                        {
 104                                 .sType = VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_CREATE_INFO,
 105                                 .stage = VK_SHADER_STAGE_VERTEX_BIT,
 106                                 .module = vs_module_h,
 107                                 .pName = "main",
 108                         },
 109                         {
 110                                 .sType = VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_CREATE_INFO,
 111                                 .stage = VK_SHADER_STAGE_FRAGMENT_BIT,
 112                                 .module = radv_shader_module_to_handle(&fs_module),
 113                                 .pName = "main",
 114                         },
 115                 },
 116                 .pVertexInputState = &(VkPipelineVertexInputStateCreateInfo) {
 117                         .sType = VK_STRUCTURE_TYPE_PIPELINE_VERTEX_INPUT_STATE_CREATE_INFO,
 118                         .vertexBindingDescriptionCount = 0,
 119                         .vertexAttributeDescriptionCount = 0,
 120                 },
 121                 .pInputAssemblyState = &(VkPipelineInputAssemblyStateCreateInfo) {
 122                         .sType = VK_STRUCTURE_TYPE_PIPELINE_INPUT_ASSEMBLY_STATE_CREATE_INFO,
 123                         .topology = VK_PRIMITIVE_TOPOLOGY_TRIANGLE_STRIP,
 124                         .primitiveRestartEnable = false,
 125                 },
 126                 .pViewportState = &(VkPipelineViewportStateCreateInfo) {
 127                         .sType = VK_STRUCTURE_TYPE_PIPELINE_VIEWPORT_STATE_CREATE_INFO,
 128                         .viewportCount = 1,
 129                         .scissorCount = 1,
 130                 },
 131                 .pRasterizationState = &(VkPipelineRasterizationStateCreateInfo) {
 132                         .sType = VK_STRUCTURE_TYPE_PIPELINE_RASTERIZATION_STATE_CREATE_INFO,
 133                         .depthClampEnable = false,
 134                         .rasterizerDiscardEnable = false,
 135                         .polygonMode = VK_POLYGON_MODE_FILL,
 136                         .cullMode = VK_CULL_MODE_NONE,
 137                         .frontFace = VK_FRONT_FACE_COUNTER_CLOCKWISE,
 138                 },
 139                 .pMultisampleState = &(VkPipelineMultisampleStateCreateInfo) {
 140                         .sType = VK_STRUCTURE_TYPE_PIPELINE_MULTISAMPLE_STATE_CREATE_INFO,
 141                         .rasterizationSamples = samples,
 142                         .sampleShadingEnable = false,
 143                         .pSampleMask = NULL,
 144                         .alphaToCoverageEnable = false,
 145                         .alphaToOneEnable = false,
 146                 },
 147                 .pColorBlendState = &(VkPipelineColorBlendStateCreateInfo) {
 148                         .sType = VK_STRUCTURE_TYPE_PIPELINE_COLOR_BLEND_STATE_CREATE_INFO,
 149                         .logicOpEnable = false,
 150                         .attachmentCount = 0,
 151                         .pAttachments = NULL,
 152                 },
 153                 .pDepthStencilState = &(VkPipelineDepthStencilStateCreateInfo) {
 154                         .sType = VK_STRUCTURE_TYPE_PIPELINE_DEPTH_STENCIL_STATE_CREATE_INFO,
 155                         .depthTestEnable = false,
 156                         .depthWriteEnable = false,
 157                         .depthBoundsTestEnable = false,
 158                         .stencilTestEnable = false,
 159                 },
 160                 .pDynamicState = &(VkPipelineDynamicStateCreateInfo) {
 161                         .sType = VK_STRUCTURE_TYPE_PIPELINE_DYNAMIC_STATE_CREATE_INFO,
 162                         .dynamicStateCount = 2,
 163                         .pDynamicStates = (VkDynamicState[]) {
 164                                 VK_DYNAMIC_STATE_VIEWPORT,
 165                                 VK_DYNAMIC_STATE_SCISSOR,
 166                         },
 167                 },
 168                 .renderPass = pass,
 169                 .subpass = 0,
 170         };
 171
 172         result = radv_graphics_pipeline_create(device_h,
 173                                                radv_pipeline_cache_to_handle(&device->meta_state.cache),
 174                                                &pipeline_create_info,
 175                                                &(struct radv_graphics_pipeline_create_info) {
 176                                                         .use_rectlist = true,
 177                                                         .db_flush_depth_inplace = true,
 178                                                         .db_flush_stencil_inplace = true,
 179                                                },
 180                                                &device->meta_state.alloc,
 181                                                decompress_pipeline);
 182         if (result != VK_SUCCESS)
 183                 goto cleanup;
 184
 185         result = radv_graphics_pipeline_create(device_h,
 186                                                radv_pipeline_cache_to_handle(&device->meta_state.cache),
 187                                                &pipeline_create_info,
 188                                                &(struct radv_graphics_pipeline_create_info) {
 189                                                         .use_rectlist = true,
 190                                                         .db_flush_depth_inplace = true,
 191                                                         .db_flush_stencil_inplace = true,
 192                                                         .db_resummarize = true,
 193                                                },
 194                                                &device->meta_state.alloc,
 195                                                resummarize_pipeline);
 196         if (result != VK_SUCCESS)
 197                 goto cleanup;
 198
 199         goto cleanup;
 200
 201 cleanup:
 202         ralloc_free(fs_module.nir);
 203         return result;
 204 }
 205
 206 void
 207 radv_device_finish_meta_depth_decomp_state(struct radv_device *device)
 208 {
 209         struct radv_meta_state *state = &device->meta_state;
 210         VkDevice device_h = radv_device_to_handle(device);
 211         const VkAllocationCallbacks *alloc = &device->meta_state.alloc;
 212
 213         for (uint32_t i = 0; i < ARRAY_SIZE(state->depth_decomp); ++i) {
 214                 VkRenderPass pass_h = state->depth_decomp[i].pass;
 215                 if (pass_h) {
 216                         radv_DestroyRenderPass(device_h, pass_h, alloc);
 217                 }
 218                 VkPipeline pipeline_h = state->depth_decomp[i].decompress_pipeline;
 219                 if (pipeline_h) {
 220                         radv_DestroyPipeline(device_h, pipeline_h, alloc);
 221                 }
 222                 pipeline_h = state->depth_decomp[i].resummarize_pipeline;
 223                 if (pipeline_h) {
 224                         radv_DestroyPipeline(device_h, pipeline_h, alloc);
 225                 }
 226         }
 227 }
 228
 229 VkResult
 230 radv_device_init_meta_depth_decomp_state(struct radv_device *device)
 231 {
 232         struct radv_meta_state *state = &device->meta_state;
 233         VkResult res = VK_SUCCESS;
 234
 235         zero(state->depth_decomp);
 236
 237         struct radv_shader_module vs_module = { .nir = radv_meta_build_nir_vs_generate_vertices() };
 238         if (!vs_module.nir) {
 239                 /* XXX: Need more accurate error */
 240                 res = VK_ERROR_OUT_OF_HOST_MEMORY;
 241                 goto fail;
 242         }
 243
 244         VkShaderModule vs_module_h = radv_shader_module_to_handle(&vs_module);
 245
 246         for (uint32_t i = 0; i < ARRAY_SIZE(state->depth_decomp); ++i) {
 247                 uint32_t samples = 1 << i;
 248
 249                 res = create_pass(device, samples, &state->depth_decomp[i].pass);
 250                 if (res != VK_SUCCESS)
 251                         goto fail;
 252
 253                 res = create_pipeline(device, vs_module_h, samples,
 254                                       state->depth_decomp[i].pass,
 255                                       &state->depth_decomp[i].decompress_pipeline,
 256                                       &state->depth_decomp[i].resummarize_pipeline);
 257                 if (res != VK_SUCCESS)
 258                         goto fail;
 259         }
 260
 261         goto cleanup;
 262
 263 fail:
 264         radv_device_finish_meta_depth_decomp_state(device);
 265
 266 cleanup:
 267         ralloc_free(vs_module.nir);
 268
 269         return res;
 270 }
 271
 272 static void
 273 emit_depth_decomp(struct radv_cmd_buffer *cmd_buffer,
 274                   const VkOffset2D *dest_offset,
 275                   const VkExtent2D *depth_decomp_extent,
 276                   VkPipeline pipeline_h)
 277 {
 278         VkCommandBuffer cmd_buffer_h = radv_cmd_buffer_to_handle(cmd_buffer);
 279
 280         RADV_FROM_HANDLE(radv_pipeline, pipeline, pipeline_h);
 281
 282         if (cmd_buffer->state.pipeline != pipeline) {
 283                 radv_CmdBindPipeline(cmd_buffer_h, VK_PIPELINE_BIND_POINT_GRAPHICS,
 284                                      pipeline_h);
 285         }
 286
 287         radv_CmdSetViewport(radv_cmd_buffer_to_handle(cmd_buffer), 0, 1, &(VkViewport) {
 288                 .x = dest_offset->x,
 289                 .y = dest_offset->y,
 290                 .width = depth_decomp_extent->width,
 291                 .height = depth_decomp_extent->height,
 292                 .minDepth = 0.0f,
 293                 .maxDepth = 1.0f
 294         });
 295
 296         radv_CmdSetScissor(radv_cmd_buffer_to_handle(cmd_buffer), 0, 1, &(VkRect2D) {
 297                 .offset = *dest_offset,
 298                 .extent = *depth_decomp_extent,
 299         });
 300
 301         radv_CmdDraw(cmd_buffer_h, 3, 1, 0, 0);
 302 }
 303
 304
 305 enum radv_depth_op {
 306         DEPTH_DECOMPRESS,
 307         DEPTH_RESUMMARIZE,
 308 };
 309
 310 static void radv_process_depth_image_inplace(struct radv_cmd_buffer *cmd_buffer,
 311                                              struct radv_image *image,
 312                                              VkImageSubresourceRange *subresourceRange,
 313                                              enum radv_depth_op op)
 314 {
 315         struct radv_meta_saved_state saved_state;
 316         struct radv_meta_saved_pass_state saved_pass_state;
 317         VkDevice device_h = radv_device_to_handle(cmd_buffer->device);
 318         VkCommandBuffer cmd_buffer_h = radv_cmd_buffer_to_handle(cmd_buffer);
 319         uint32_t width = radv_minify(image->info.width,
 320                                      subresourceRange->baseMipLevel);
 321         uint32_t height = radv_minify(image->info.height,
 322                                      subresourceRange->baseMipLevel);
 323         uint32_t samples = image->info.samples;
 324         uint32_t samples_log2 = ffs(samples) - 1;
 325         struct radv_meta_state *meta_state = &cmd_buffer->device->meta_state;
 326
 327         if (!image->surface.htile_size)
 328                 return;
 329         radv_meta_save_pass(&saved_pass_state, cmd_buffer);
 330
 331         radv_meta_save_graphics_reset_vport_scissor_novertex(&saved_state, cmd_buffer);
 332
 333         for (uint32_t layer = 0; layer < radv_get_layerCount(image, subresourceRange); layer++) {
 334                 struct radv_image_view iview;
 335
 336                 radv_image_view_init(&iview, cmd_buffer->device,
 337                                      &(VkImageViewCreateInfo) {
 338                                              .sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO,
 339                                              .image = radv_image_to_handle(image),
 340                                              .format = image->vk_format,
 341                                              .subresourceRange = {
 342                                                      .aspectMask = VK_IMAGE_ASPECT_DEPTH_BIT,
 343                                                      .baseMipLevel = subresourceRange->baseMipLevel,
 344                                                      .levelCount = 1,
 345                                                      .baseArrayLayer = subresourceRange->baseArrayLayer + layer,
 346                                                      .layerCount = 1,
 347                                              },
 348                                      });
 349
 350
 351                 VkFramebuffer fb_h;
 352                 radv_CreateFramebuffer(device_h,
 353                                        &(VkFramebufferCreateInfo) {
 354                                                .sType = VK_STRUCTURE_TYPE_FRAMEBUFFER_CREATE_INFO,
 355                                                .attachmentCount = 1,
 356                                                        .pAttachments = (VkImageView[]) {
 357                                                        radv_image_view_to_handle(&iview)
 358                                                },
 359                                                .width = width,
 360                                                 .height = height,
 361                                                .layers = 1
 362                                        },
 363                                        &cmd_buffer->pool->alloc,
 364                                        &fb_h);
 365
 366                 radv_CmdBeginRenderPass(cmd_buffer_h,
 367                                               &(VkRenderPassBeginInfo) {
 368                                                       .sType = VK_STRUCTURE_TYPE_RENDER_PASS_BEGIN_INFO,
 369                                                               .renderPass = meta_state->depth_decomp[samples_log2].pass,
 370                                                               .framebuffer = fb_h,
 371                                                               .renderArea = {
 372                                                               .offset = {
 373                                                                       0,
 374                                                                       0,
 375                                                               },
 376                                                               .extent = {
 377                                                                       width,
 378                                                                       height,
 379                                                               }
 380                                                        },
 381                                                        .clearValueCount = 0,
 382                                                        .pClearValues = NULL,
 383                                            },
 384                                            VK_SUBPASS_CONTENTS_INLINE);
 385
 386                 VkPipeline pipeline_h;
 387                 switch (op) {
 388                 case DEPTH_DECOMPRESS:
 389                         pipeline_h = meta_state->depth_decomp[samples_log2].decompress_pipeline;
 390                         break;
 391                 case DEPTH_RESUMMARIZE:
 392                         pipeline_h = meta_state->depth_decomp[samples_log2].resummarize_pipeline;
 393                         break;
 394                 default:
 395                         unreachable("unknown operation");
 396                 }
 397
 398                 emit_depth_decomp(cmd_buffer, &(VkOffset2D){0, 0 }, &(VkExtent2D){width, height}, pipeline_h);
 399                 radv_CmdEndRenderPass(cmd_buffer_h);
 400
 401                 radv_DestroyFramebuffer(device_h, fb_h,
 402                                         &cmd_buffer->pool->alloc);
 403         }
 404         radv_meta_restore(&saved_state, cmd_buffer);
 405         radv_meta_restore_pass(&saved_pass_state, cmd_buffer);
 406 }
 407
 408 void radv_decompress_depth_image_inplace(struct radv_cmd_buffer *cmd_buffer,
 409                                          struct radv_image *image,
 410                                          VkImageSubresourceRange *subresourceRange)
 411 {
 412         assert(cmd_buffer->queue_family_index == RADV_QUEUE_GENERAL);
 413         radv_process_depth_image_inplace(cmd_buffer, image, subresourceRange, DEPTH_DECOMPRESS);
 414 }
 415
 416 void radv_resummarize_depth_image_inplace(struct radv_cmd_buffer *cmd_buffer,
 417                                          struct radv_image *image,
 418                                          VkImageSubresourceRange *subresourceRange)
 419 {
 420         assert(cmd_buffer->queue_family_index == RADV_QUEUE_GENERAL);
 421         radv_process_depth_image_inplace(cmd_buffer, image, subresourceRange, DEPTH_RESUMMARIZE);
 422 }