src/amd/vulkan/radv_meta_copy.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 "radv_meta.h"
  25 #include "vk_format.h"
  26
  27 static VkExtent3D
  28 meta_image_block_size(const struct radv_image *image)
  29 {
  30         const struct vk_format_description *desc = vk_format_description(image->vk_format);
  31         return (VkExtent3D) { desc->block.width, desc->block.height, 1 };
  32 }
  33
  34 /* Returns the user-provided VkBufferImageCopy::imageExtent in units of
  35  * elements rather than texels. One element equals one texel or one block
  36  * if Image is uncompressed or compressed, respectively.
  37  */
  38 static struct VkExtent3D
  39 meta_region_extent_el(const struct radv_image *image,
  40                       const VkImageType imageType,
  41                       const struct VkExtent3D *extent)
  42 {
  43         const VkExtent3D block = meta_image_block_size(image);
  44         return radv_sanitize_image_extent(imageType, (VkExtent3D) {
  45                         .width  = DIV_ROUND_UP(extent->width , block.width),
  46                                 .height = DIV_ROUND_UP(extent->height, block.height),
  47                                 .depth  = DIV_ROUND_UP(extent->depth , block.depth),
  48                                 });
  49 }
  50
  51 /* Returns the user-provided VkBufferImageCopy::imageOffset in units of
  52  * elements rather than texels. One element equals one texel or one block
  53  * if Image is uncompressed or compressed, respectively.
  54  */
  55 static struct VkOffset3D
  56 meta_region_offset_el(const struct radv_image *image,
  57                       const struct VkOffset3D *offset)
  58 {
  59         const VkExtent3D block = meta_image_block_size(image);
  60         return radv_sanitize_image_offset(image->type, (VkOffset3D) {
  61                         .x = offset->x / block.width,
  62                                 .y = offset->y / block.height,
  63                                 .z = offset->z / block.depth,
  64                                 });
  65 }
  66
  67 static VkFormat
  68 vk_format_for_size(int bs)
  69 {
  70         switch (bs) {
  71         case 1: return VK_FORMAT_R8_UINT;
  72         case 2: return VK_FORMAT_R8G8_UINT;
  73         case 4: return VK_FORMAT_R8G8B8A8_UINT;
  74         case 8: return VK_FORMAT_R16G16B16A16_UINT;
  75         case 12: return VK_FORMAT_R32G32B32_UINT;
  76         case 16: return VK_FORMAT_R32G32B32A32_UINT;
  77         default:
  78                 unreachable("Invalid format block size");
  79         }
  80 }
  81
  82 static struct radv_meta_blit2d_surf
  83 blit_surf_for_image_level_layer(struct radv_image *image,
  84                                 VkImageLayout layout,
  85                                 const VkImageSubresourceLayers *subres)
  86 {
  87         VkFormat format = image->vk_format;
  88         if (subres->aspectMask & VK_IMAGE_ASPECT_DEPTH_BIT)
  89                 format = vk_format_depth_only(format);
  90         else if (subres->aspectMask & VK_IMAGE_ASPECT_STENCIL_BIT)
  91                 format = vk_format_stencil_only(format);
  92
  93         if (!radv_image_has_dcc(image) &&
  94             !(radv_image_is_tc_compat_htile(image)))
  95                 format = vk_format_for_size(vk_format_get_blocksize(format));
  96
  97         format = vk_format_no_srgb(format);
  98
  99         return (struct radv_meta_blit2d_surf) {
 100                 .format = format,
 101                 .bs = vk_format_get_blocksize(format),
 102                 .level = subres->mipLevel,
 103                 .layer = subres->baseArrayLayer,
 104                 .image = image,
 105                 .aspect_mask = subres->aspectMask,
 106                 .current_layout = layout,
 107         };
 108 }
 109
 110 static bool
 111 image_is_renderable(struct radv_device *device, struct radv_image *image)
 112 {
 113         if (image->vk_format == VK_FORMAT_R32G32B32_UINT ||
 114             image->vk_format == VK_FORMAT_R32G32B32_SINT ||
 115             image->vk_format == VK_FORMAT_R32G32B32_SFLOAT)
 116                 return false;
 117
 118         if (device->physical_device->rad_info.chip_class >= GFX9 &&
 119             image->type == VK_IMAGE_TYPE_3D &&
 120             vk_format_get_blocksizebits(image->vk_format) == 128 &&
 121             vk_format_is_compressed(image->vk_format))
 122                 return false;
 123         return true;
 124 }
 125
 126 static void
 127 meta_copy_buffer_to_image(struct radv_cmd_buffer *cmd_buffer,
 128                           struct radv_buffer* buffer,
 129                           struct radv_image* image,
 130                           VkImageLayout layout,
 131                           uint32_t regionCount,
 132                           const VkBufferImageCopy* pRegions)
 133 {
 134         bool cs = cmd_buffer->queue_family_index == RADV_QUEUE_COMPUTE;
 135         struct radv_meta_saved_state saved_state;
 136         bool old_predicating;
 137
 138         /* The Vulkan 1.0 spec says "dstImage must have a sample count equal to
 139          * VK_SAMPLE_COUNT_1_BIT."
 140          */
 141         assert(image->info.samples == 1);
 142
 143         radv_meta_save(&saved_state, cmd_buffer,
 144                        (cs ? RADV_META_SAVE_COMPUTE_PIPELINE :
 145                         RADV_META_SAVE_GRAPHICS_PIPELINE) |
 146                        RADV_META_SAVE_CONSTANTS |
 147                        RADV_META_SAVE_DESCRIPTORS);
 148
 149         /* VK_EXT_conditional_rendering says that copy commands should not be
 150          * affected by conditional rendering.
 151          */
 152         old_predicating = cmd_buffer->state.predicating;
 153         cmd_buffer->state.predicating = false;
 154
 155         for (unsigned r = 0; r < regionCount; r++) {
 156
 157                 /**
 158                  * From the Vulkan 1.0.6 spec: 18.3 Copying Data Between Images
 159                  *    extent is the size in texels of the source image to copy in width,
 160                  *    height and depth. 1D images use only x and width. 2D images use x, y,
 161                  *    width and height. 3D images use x, y, z, width, height and depth.
 162                  *
 163                  *
 164                  * Also, convert the offsets and extent from units of texels to units of
 165                  * blocks - which is the highest resolution accessible in this command.
 166                  */
 167                 const VkOffset3D img_offset_el =
 168                         meta_region_offset_el(image, &pRegions[r].imageOffset);
 169                 const VkExtent3D bufferExtent = {
 170                         .width  = pRegions[r].bufferRowLength ?
 171                         pRegions[r].bufferRowLength : pRegions[r].imageExtent.width,
 172                         .height = pRegions[r].bufferImageHeight ?
 173                         pRegions[r].bufferImageHeight : pRegions[r].imageExtent.height,
 174                 };
 175                 const VkExtent3D buf_extent_el =
 176                         meta_region_extent_el(image, image->type, &bufferExtent);
 177
 178                 /* Start creating blit rect */
 179                 const VkExtent3D img_extent_el =
 180                         meta_region_extent_el(image, image->type, &pRegions[r].imageExtent);
 181                 struct radv_meta_blit2d_rect rect = {
 182                         .width = img_extent_el.width,
 183                         .height =  img_extent_el.height,
 184                 };
 185
 186                 /* Create blit surfaces */
 187                 struct radv_meta_blit2d_surf img_bsurf =
 188                         blit_surf_for_image_level_layer(image,
 189                                                         layout,
 190                                                         &pRegions[r].imageSubresource);
 191
 192                 struct radv_meta_blit2d_buffer buf_bsurf = {
 193                         .bs = img_bsurf.bs,
 194                         .format = img_bsurf.format,
 195                         .buffer = buffer,
 196                         .offset = pRegions[r].bufferOffset,
 197                         .pitch = buf_extent_el.width,
 198                 };
 199
 200                 if (image->type == VK_IMAGE_TYPE_3D)
 201                         img_bsurf.layer = img_offset_el.z;
 202                 /* Loop through each 3D or array slice */
 203                 unsigned num_slices_3d = img_extent_el.depth;
 204                 unsigned num_slices_array = pRegions[r].imageSubresource.layerCount;
 205                 unsigned slice_3d = 0;
 206                 unsigned slice_array = 0;
 207                 while (slice_3d < num_slices_3d && slice_array < num_slices_array) {
 208
 209                         rect.dst_x = img_offset_el.x;
 210                         rect.dst_y = img_offset_el.y;
 211
 212
 213                         /* Perform Blit */
 214                         if (cs ||
 215                             !image_is_renderable(cmd_buffer->device, img_bsurf.image)) {
 216                                 radv_meta_buffer_to_image_cs(cmd_buffer, &buf_bsurf, &img_bsurf, 1, &rect);
 217                         } else {
 218                                 radv_meta_blit2d(cmd_buffer, NULL, &buf_bsurf, &img_bsurf, 1, &rect);
 219                         }
 220
 221                         /* Once we've done the blit, all of the actual information about
 222                          * the image is embedded in the command buffer so we can just
 223                          * increment the offset directly in the image effectively
 224                          * re-binding it to different backing memory.
 225                          */
 226                         buf_bsurf.offset += buf_extent_el.width *
 227                                             buf_extent_el.height * buf_bsurf.bs;
 228                         img_bsurf.layer++;
 229                         if (image->type == VK_IMAGE_TYPE_3D)
 230                                 slice_3d++;
 231                         else
 232                                 slice_array++;
 233                 }
 234         }
 235
 236         /* Restore conditional rendering. */
 237         cmd_buffer->state.predicating = old_predicating;
 238
 239         radv_meta_restore(&saved_state, cmd_buffer);
 240 }
 241
 242 void radv_CmdCopyBufferToImage(
 243         VkCommandBuffer                             commandBuffer,
 244         VkBuffer                                    srcBuffer,
 245         VkImage                                     destImage,
 246         VkImageLayout                               destImageLayout,
 247         uint32_t                                    regionCount,
 248         const VkBufferImageCopy*                    pRegions)
 249 {
 250         RADV_FROM_HANDLE(radv_cmd_buffer, cmd_buffer, commandBuffer);
 251         RADV_FROM_HANDLE(radv_image, dest_image, destImage);
 252         RADV_FROM_HANDLE(radv_buffer, src_buffer, srcBuffer);
 253
 254         meta_copy_buffer_to_image(cmd_buffer, src_buffer, dest_image, destImageLayout,
 255                                   regionCount, pRegions);
 256 }
 257
 258 static void
 259 meta_copy_image_to_buffer(struct radv_cmd_buffer *cmd_buffer,
 260                           struct radv_buffer* buffer,
 261                           struct radv_image* image,
 262                           VkImageLayout layout,
 263                           uint32_t regionCount,
 264                           const VkBufferImageCopy* pRegions)
 265 {
 266         struct radv_meta_saved_state saved_state;
 267         bool old_predicating;
 268
 269         radv_meta_save(&saved_state, cmd_buffer,
 270                        RADV_META_SAVE_COMPUTE_PIPELINE |
 271                        RADV_META_SAVE_CONSTANTS |
 272                        RADV_META_SAVE_DESCRIPTORS);
 273
 274         /* VK_EXT_conditional_rendering says that copy commands should not be
 275          * affected by conditional rendering.
 276          */
 277         old_predicating = cmd_buffer->state.predicating;
 278         cmd_buffer->state.predicating = false;
 279
 280         for (unsigned r = 0; r < regionCount; r++) {
 281
 282                 /**
 283                  * From the Vulkan 1.0.6 spec: 18.3 Copying Data Between Images
 284                  *    extent is the size in texels of the source image to copy in width,
 285                  *    height and depth. 1D images use only x and width. 2D images use x, y,
 286                  *    width and height. 3D images use x, y, z, width, height and depth.
 287                  *
 288                  *
 289                  * Also, convert the offsets and extent from units of texels to units of
 290                  * blocks - which is the highest resolution accessible in this command.
 291                  */
 292                 const VkOffset3D img_offset_el =
 293                         meta_region_offset_el(image, &pRegions[r].imageOffset);
 294                 const VkExtent3D bufferExtent = {
 295                         .width  = pRegions[r].bufferRowLength ?
 296                         pRegions[r].bufferRowLength : pRegions[r].imageExtent.width,
 297                         .height = pRegions[r].bufferImageHeight ?
 298                         pRegions[r].bufferImageHeight : pRegions[r].imageExtent.height,
 299                 };
 300                 const VkExtent3D buf_extent_el =
 301                         meta_region_extent_el(image, image->type, &bufferExtent);
 302
 303                 /* Start creating blit rect */
 304                 const VkExtent3D img_extent_el =
 305                         meta_region_extent_el(image, image->type, &pRegions[r].imageExtent);
 306                 struct radv_meta_blit2d_rect rect = {
 307                         .width = img_extent_el.width,
 308                         .height =  img_extent_el.height,
 309                 };
 310
 311                 /* Create blit surfaces */
 312                 struct radv_meta_blit2d_surf img_info =
 313                         blit_surf_for_image_level_layer(image,
 314                                                         layout,
 315                                                         &pRegions[r].imageSubresource);
 316
 317                 struct radv_meta_blit2d_buffer buf_info = {
 318                         .bs = img_info.bs,
 319                         .format = img_info.format,
 320                         .buffer = buffer,
 321                         .offset = pRegions[r].bufferOffset,
 322                         .pitch = buf_extent_el.width,
 323                 };
 324
 325                 if (image->type == VK_IMAGE_TYPE_3D)
 326                         img_info.layer = img_offset_el.z;
 327                 /* Loop through each 3D or array slice */
 328                 unsigned num_slices_3d = img_extent_el.depth;
 329                 unsigned num_slices_array = pRegions[r].imageSubresource.layerCount;
 330                 unsigned slice_3d = 0;
 331                 unsigned slice_array = 0;
 332                 while (slice_3d < num_slices_3d && slice_array < num_slices_array) {
 333
 334                         rect.src_x = img_offset_el.x;
 335                         rect.src_y = img_offset_el.y;
 336
 337
 338                         /* Perform Blit */
 339                         radv_meta_image_to_buffer(cmd_buffer, &img_info, &buf_info, 1, &rect);
 340
 341                         buf_info.offset += buf_extent_el.width *
 342                                             buf_extent_el.height * buf_info.bs;
 343                         img_info.layer++;
 344                         if (image->type == VK_IMAGE_TYPE_3D)
 345                                 slice_3d++;
 346                         else
 347                                 slice_array++;
 348                 }
 349         }
 350
 351         /* Restore conditional rendering. */
 352         cmd_buffer->state.predicating = old_predicating;
 353
 354         radv_meta_restore(&saved_state, cmd_buffer);
 355 }
 356
 357 void radv_CmdCopyImageToBuffer(
 358         VkCommandBuffer                             commandBuffer,
 359         VkImage                                     srcImage,
 360         VkImageLayout                               srcImageLayout,
 361         VkBuffer                                    destBuffer,
 362         uint32_t                                    regionCount,
 363         const VkBufferImageCopy*                    pRegions)
 364 {
 365         RADV_FROM_HANDLE(radv_cmd_buffer, cmd_buffer, commandBuffer);
 366         RADV_FROM_HANDLE(radv_image, src_image, srcImage);
 367         RADV_FROM_HANDLE(radv_buffer, dst_buffer, destBuffer);
 368
 369         meta_copy_image_to_buffer(cmd_buffer, dst_buffer, src_image,
 370                                   srcImageLayout,
 371                                   regionCount, pRegions);
 372 }
 373
 374 static void
 375 meta_copy_image(struct radv_cmd_buffer *cmd_buffer,
 376                 struct radv_image *src_image,
 377                 VkImageLayout src_image_layout,
 378                 struct radv_image *dest_image,
 379                 VkImageLayout dest_image_layout,
 380                 uint32_t regionCount,
 381                 const VkImageCopy *pRegions)
 382 {
 383         bool cs = cmd_buffer->queue_family_index == RADV_QUEUE_COMPUTE;
 384         struct radv_meta_saved_state saved_state;
 385         bool old_predicating;
 386
 387         /* From the Vulkan 1.0 spec:
 388          *
 389          *    vkCmdCopyImage can be used to copy image data between multisample
 390          *    images, but both images must have the same number of samples.
 391          */
 392         assert(src_image->info.samples == dest_image->info.samples);
 393
 394         radv_meta_save(&saved_state, cmd_buffer,
 395                        (cs ? RADV_META_SAVE_COMPUTE_PIPELINE :
 396                         RADV_META_SAVE_GRAPHICS_PIPELINE) |
 397                        RADV_META_SAVE_CONSTANTS |
 398                        RADV_META_SAVE_DESCRIPTORS);
 399
 400         /* VK_EXT_conditional_rendering says that copy commands should not be
 401          * affected by conditional rendering.
 402          */
 403         old_predicating = cmd_buffer->state.predicating;
 404         cmd_buffer->state.predicating = false;
 405
 406         for (unsigned r = 0; r < regionCount; r++) {
 407                 assert(pRegions[r].srcSubresource.aspectMask ==
 408                        pRegions[r].dstSubresource.aspectMask);
 409
 410                 /* Create blit surfaces */
 411                 struct radv_meta_blit2d_surf b_src =
 412                         blit_surf_for_image_level_layer(src_image,
 413                                                         src_image_layout,
 414                                                         &pRegions[r].srcSubresource);
 415
 416                 struct radv_meta_blit2d_surf b_dst =
 417                         blit_surf_for_image_level_layer(dest_image,
 418                                                         dest_image_layout,
 419                                                         &pRegions[r].dstSubresource);
 420
 421                 uint32_t dst_queue_mask = radv_image_queue_family_mask(dest_image,
 422                                                                        cmd_buffer->queue_family_index,
 423                                                                        cmd_buffer->queue_family_index);
 424                 bool dst_compressed = radv_layout_dcc_compressed(dest_image, dest_image_layout, dst_queue_mask);
 425                 uint32_t src_queue_mask = radv_image_queue_family_mask(src_image,
 426                                                                        cmd_buffer->queue_family_index,
 427                                                                        cmd_buffer->queue_family_index);
 428                 bool src_compressed = radv_layout_dcc_compressed(src_image, src_image_layout, src_queue_mask);
 429
 430                 if (!src_compressed || radv_dcc_formats_compatible(b_src.format, b_dst.format)) {
 431                         b_src.format = b_dst.format;
 432                 } else if (!dst_compressed) {
 433                         b_dst.format = b_src.format;
 434                 } else {
 435                         radv_decompress_dcc(cmd_buffer, dest_image, &(VkImageSubresourceRange) {
 436                                                 .aspectMask = pRegions[r].dstSubresource.aspectMask,
 437                                                 .baseMipLevel = pRegions[r].dstSubresource.mipLevel,
 438                                                 .levelCount = 1,
 439                                                 .baseArrayLayer = pRegions[r].dstSubresource.baseArrayLayer,
 440                                                 .layerCount = pRegions[r].dstSubresource.layerCount,
 441                                             });
 442                         b_dst.format = b_src.format;
 443                         b_dst.current_layout = VK_IMAGE_LAYOUT_GENERAL;
 444                 }
 445
 446
 447                 /**
 448                  * From the Vulkan 1.0.6 spec: 18.4 Copying Data Between Buffers and Images
 449                  *    imageExtent is the size in texels of the image to copy in width, height
 450                  *    and depth. 1D images use only x and width. 2D images use x, y, width
 451                  *    and height. 3D images use x, y, z, width, height and depth.
 452                  *
 453                  * Also, convert the offsets and extent from units of texels to units of
 454                  * blocks - which is the highest resolution accessible in this command.
 455                  */
 456                 const VkOffset3D dst_offset_el =
 457                         meta_region_offset_el(dest_image, &pRegions[r].dstOffset);
 458                 const VkOffset3D src_offset_el =
 459                         meta_region_offset_el(src_image, &pRegions[r].srcOffset);
 460
 461                 /*
 462                  * From Vulkan 1.0.68, "Copying Data Between Images":
 463                  *    "When copying between compressed and uncompressed formats
 464                  *     the extent members represent the texel dimensions of the
 465                  *     source image and not the destination."
 466                  * However, we must use the destination image type to avoid
 467                  * clamping depth when copying multiple layers of a 2D image to
 468                  * a 3D image.
 469                  */
 470                 const VkExtent3D img_extent_el =
 471                         meta_region_extent_el(src_image, dest_image->type, &pRegions[r].extent);
 472
 473                 /* Start creating blit rect */
 474                 struct radv_meta_blit2d_rect rect = {
 475                         .width = img_extent_el.width,
 476                         .height = img_extent_el.height,
 477                 };
 478
 479                 if (src_image->type == VK_IMAGE_TYPE_3D)
 480                         b_src.layer = src_offset_el.z;
 481
 482                 if (dest_image->type == VK_IMAGE_TYPE_3D)
 483                         b_dst.layer = dst_offset_el.z;
 484
 485                 /* Loop through each 3D or array slice */
 486                 unsigned num_slices_3d = img_extent_el.depth;
 487                 unsigned num_slices_array = pRegions[r].dstSubresource.layerCount;
 488                 unsigned slice_3d = 0;
 489                 unsigned slice_array = 0;
 490                 while (slice_3d < num_slices_3d && slice_array < num_slices_array) {
 491
 492                         /* Finish creating blit rect */
 493                         rect.dst_x = dst_offset_el.x;
 494                         rect.dst_y = dst_offset_el.y;
 495                         rect.src_x = src_offset_el.x;
 496                         rect.src_y = src_offset_el.y;
 497
 498                         /* Perform Blit */
 499                         if (cs ||
 500                             !image_is_renderable(cmd_buffer->device, b_dst.image)) {
 501                                 radv_meta_image_to_image_cs(cmd_buffer, &b_src, &b_dst, 1, &rect);
 502                         } else {
 503                                 radv_meta_blit2d(cmd_buffer, &b_src, NULL, &b_dst, 1, &rect);
 504                         }
 505
 506                         b_src.layer++;
 507                         b_dst.layer++;
 508                         if (dest_image->type == VK_IMAGE_TYPE_3D)
 509                                 slice_3d++;
 510                         else
 511                                 slice_array++;
 512                 }
 513         }
 514
 515         /* Restore conditional rendering. */
 516         cmd_buffer->state.predicating = old_predicating;
 517
 518         radv_meta_restore(&saved_state, cmd_buffer);
 519 }
 520
 521 void radv_CmdCopyImage(
 522         VkCommandBuffer                             commandBuffer,
 523         VkImage                                     srcImage,
 524         VkImageLayout                               srcImageLayout,
 525         VkImage                                     destImage,
 526         VkImageLayout                               destImageLayout,
 527         uint32_t                                    regionCount,
 528         const VkImageCopy*                          pRegions)
 529 {
 530         RADV_FROM_HANDLE(radv_cmd_buffer, cmd_buffer, commandBuffer);
 531         RADV_FROM_HANDLE(radv_image, src_image, srcImage);
 532         RADV_FROM_HANDLE(radv_image, dest_image, destImage);
 533
 534         meta_copy_image(cmd_buffer,
 535                         src_image, srcImageLayout,
 536                         dest_image, destImageLayout,
 537                         regionCount, pRegions);
 538 }