external/vulkancts/modules/vulkan/sparse_resources/vktSparseResourcesShaderIntrinsicsBase.cpp

   1 /*------------------------------------------------------------------------
   2  * Vulkan Conformance Tests
   3  * ------------------------
   4  *
   5  * Copyright (c) 2016 The Khronos Group Inc.
   6  *
   7  * Licensed under the Apache License, Version 2.0 (the "License");
   8  * you may not use this file except in compliance with the License.
   9  * You may obtain a copy of the License at
  10  *
  11  *      http://www.apache.org/licenses/LICENSE-2.0
  12  *
  13  * Unless required by applicable law or agreed to in writing, software
  14  * distributed under the License is distributed on an "AS IS" BASIS,
  15  * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
  16  * See the License for the specific language governing permissions and
  17  * limitations under the License.
  18  *
  19  *//*
  20  * \file  vktSparseResourcesShaderIntrinsicsBase.cpp
  21  * \brief Sparse Resources Shader Intrinsics Base Classes
  22  *//*--------------------------------------------------------------------*/
  23
  24 #include "vktSparseResourcesShaderIntrinsicsBase.hpp"
  25
  26 using namespace vk;
  27
  28 namespace vkt
  29 {
  30 namespace sparse
  31 {
  32
  33 tcu::UVec3 alignedDivide (const VkExtent3D& extent, const VkExtent3D& divisor)
  34 {
  35         tcu::UVec3 result;
  36
  37         result.x() = extent.width  / divisor.width  + ((extent.width  % divisor.width)  ? 1u : 0u);
  38         result.y() = extent.height / divisor.height + ((extent.height % divisor.height) ? 1u : 0u);
  39         result.z() = extent.depth  / divisor.depth  + ((extent.depth  % divisor.depth)  ? 1u : 0u);
  40
  41         return result;
  42 }
  43
  44 std::string getOpTypeImageComponent (const tcu::TextureFormat& format)
  45 {
  46         switch (tcu::getTextureChannelClass(format.type))
  47         {
  48                 case tcu::TEXTURECHANNELCLASS_UNSIGNED_INTEGER:
  49                         return "OpTypeInt 32 0";
  50                 case tcu::TEXTURECHANNELCLASS_SIGNED_INTEGER:
  51                         return "OpTypeInt 32 1";
  52                 default:
  53                         DE_ASSERT(0);
  54                         return "";
  55         }
  56 }
  57
  58 std::string getOpTypeImageSparse (const ImageType                       imageType,
  59                                                                   const tcu::TextureFormat&     format,
  60                                                                   const std::string&            componentType,
  61                                                                   const bool                            requiresSampler)
  62 {
  63         std::ostringstream      src;
  64
  65         src << "OpTypeImage " << componentType << " ";
  66
  67         switch (imageType)
  68         {
  69                 case IMAGE_TYPE_1D :
  70                         src << "1D 0 0 0 ";
  71                 break;
  72                 case IMAGE_TYPE_1D_ARRAY :
  73                         src << "1D 0 1 0 ";
  74                 break;
  75                 case IMAGE_TYPE_2D :
  76                         src << "2D 0 0 0 ";
  77                 break;
  78                 case IMAGE_TYPE_2D_ARRAY :
  79                         src << "2D 0 1 0 ";
  80                 break;
  81                 case IMAGE_TYPE_3D :
  82                         src << "3D 0 0 0 ";
  83                 break;
  84                 case IMAGE_TYPE_CUBE :
  85                         src << "Cube 0 0 0 ";
  86                 break;
  87                 case IMAGE_TYPE_CUBE_ARRAY :
  88                         src << "Cube 0 1 0 ";
  89                 break;
  90                 default :
  91                         DE_ASSERT(0);
  92                 break;
  93         };
  94
  95         if (requiresSampler)
  96                 src << "1 ";
  97         else
  98                 src << "2 ";
  99
 100         switch (format.order)
 101         {
 102                 case tcu::TextureFormat::R:
 103                         src << "R";
 104                 break;
 105                 case tcu::TextureFormat::RG:
 106                         src << "Rg";
 107                         break;
 108                 case tcu::TextureFormat::RGB:
 109                         src << "Rgb";
 110                         break;
 111                 case tcu::TextureFormat::RGBA:
 112                         src << "Rgba";
 113                 break;
 114                 default:
 115                         DE_ASSERT(0);
 116                 break;
 117         }
 118
 119         switch (format.type)
 120         {
 121                 case tcu::TextureFormat::SIGNED_INT8:
 122                         src << "8i";
 123                 break;
 124                 case tcu::TextureFormat::SIGNED_INT16:
 125                         src << "16i";
 126                 break;
 127                 case tcu::TextureFormat::SIGNED_INT32:
 128                         src << "32i";
 129                 break;
 130                 case tcu::TextureFormat::UNSIGNED_INT8:
 131                         src << "8ui";
 132                 break;
 133                 case tcu::TextureFormat::UNSIGNED_INT16:
 134                         src << "16ui";
 135                 break;
 136                 case tcu::TextureFormat::UNSIGNED_INT32:
 137                         src << "32ui";
 138                 break;
 139                 default:
 140                         DE_ASSERT(0);
 141                 break;
 142         };
 143
 144         return src.str();
 145 }
 146
 147 std::string getOpTypeImageResidency (const ImageType imageType)
 148 {
 149         std::ostringstream      src;
 150
 151         src << "OpTypeImage %type_uint ";
 152
 153         switch (imageType)
 154         {
 155                 case IMAGE_TYPE_1D :
 156                         src << "1D 0 0 0 2 R32ui";
 157                 break;
 158                 case IMAGE_TYPE_1D_ARRAY :
 159                         src << "1D 0 1 0 2 R32ui";
 160                 break;
 161                 case IMAGE_TYPE_2D :
 162                         src << "2D 0 0 0 2 R32ui";
 163                 break;
 164                 case IMAGE_TYPE_2D_ARRAY :
 165                         src << "2D 0 1 0 2 R32ui";
 166                 break;
 167                 case IMAGE_TYPE_3D :
 168                         src << "3D 0 0 0 2 R32ui";
 169                 break;
 170                 case IMAGE_TYPE_CUBE :
 171                         src << "Cube 0 0 0 2 R32ui";
 172                 break;
 173                 case IMAGE_TYPE_CUBE_ARRAY :
 174                         src << "Cube 0 1 0 2 R32ui";
 175                 break;
 176                 default :
 177                         DE_ASSERT(0);
 178                 break;
 179         };
 180
 181         return src.str();
 182 }
 183
 184 tcu::TestStatus SparseShaderIntrinsicsInstanceBase::iterate (void)
 185 {
 186         const InstanceInterface&                        instance                                = m_context.getInstanceInterface();
 187         const DeviceInterface&                          deviceInterface                 = m_context.getDeviceInterface();
 188         const VkPhysicalDevice                          physicalDevice                  = m_context.getPhysicalDevice();
 189         VkImageCreateInfo                                       imageSparseInfo;
 190         VkImageCreateInfo                                       imageTexelsInfo;
 191         VkImageCreateInfo                                       imageResidencyInfo;
 192         VkSparseImageMemoryRequirements         aspectRequirements;
 193         std::vector <deUint32>                          residencyReferenceData;
 194         std::vector<DeviceMemoryUniquePtr>      deviceMemUniquePtrVec;
 195
 196         // Check if image size does not exceed device limits
 197         if (!isImageSizeSupported(instance, physicalDevice, m_imageType, m_imageSize))
 198                 TCU_THROW(NotSupportedError, "Image size not supported for device");
 199
 200         // Check if device supports sparse operations for image type
 201         if (!checkSparseSupportForImageType(instance, physicalDevice, m_imageType))
 202                 TCU_THROW(NotSupportedError, "Sparse residency for image type is not supported");
 203
 204         if (!getPhysicalDeviceFeatures(instance, physicalDevice).shaderResourceResidency)
 205                 TCU_THROW(NotSupportedError, "Sparse resource residency information not supported in shader code.");
 206
 207         imageSparseInfo.sType                                   = VK_STRUCTURE_TYPE_IMAGE_CREATE_INFO;
 208         imageSparseInfo.pNext                                   = DE_NULL;
 209         imageSparseInfo.flags                                   = VK_IMAGE_CREATE_SPARSE_RESIDENCY_BIT | VK_IMAGE_CREATE_SPARSE_BINDING_BIT;
 210         imageSparseInfo.imageType                               = mapImageType(m_imageType);
 211         imageSparseInfo.format                                  = mapTextureFormat(m_format);
 212         imageSparseInfo.extent                                  = makeExtent3D(getLayerSize(m_imageType, m_imageSize));
 213         imageSparseInfo.arrayLayers                             = getNumLayers(m_imageType, m_imageSize);
 214         imageSparseInfo.samples                                 = VK_SAMPLE_COUNT_1_BIT;
 215         imageSparseInfo.tiling                                  = VK_IMAGE_TILING_OPTIMAL;
 216         imageSparseInfo.initialLayout                   = VK_IMAGE_LAYOUT_UNDEFINED;
 217         imageSparseInfo.usage                                   = VK_IMAGE_USAGE_TRANSFER_DST_BIT | imageSparseUsageFlags();
 218         imageSparseInfo.sharingMode                             = VK_SHARING_MODE_EXCLUSIVE;
 219         imageSparseInfo.queueFamilyIndexCount   = 0u;
 220         imageSparseInfo.pQueueFamilyIndices             = DE_NULL;
 221
 222         if (m_imageType == IMAGE_TYPE_CUBE || m_imageType == IMAGE_TYPE_CUBE_ARRAY)
 223         {
 224                 imageSparseInfo.flags |= VK_IMAGE_CREATE_CUBE_COMPATIBLE_BIT;
 225         }
 226
 227         {
 228                 // Assign maximum allowed mipmap levels to image
 229                 VkImageFormatProperties imageFormatProperties;
 230                 instance.getPhysicalDeviceImageFormatProperties(physicalDevice,
 231                         imageSparseInfo.format,
 232                         imageSparseInfo.imageType,
 233                         imageSparseInfo.tiling,
 234                         imageSparseInfo.usage,
 235                         imageSparseInfo.flags,
 236                         &imageFormatProperties);
 237
 238                 imageSparseInfo.mipLevels = getImageMaxMipLevels(imageFormatProperties, imageSparseInfo.extent);
 239         }
 240
 241         // Check if device supports sparse operations for image format
 242         if (!checkSparseSupportForImageFormat(instance, physicalDevice, imageSparseInfo))
 243                 TCU_THROW(NotSupportedError, "The image format does not support sparse operations");
 244
 245         {
 246                 // Create logical device supporting both sparse and compute/graphics queues
 247                 QueueRequirementsVec queueRequirements;
 248                 queueRequirements.push_back(QueueRequirements(VK_QUEUE_SPARSE_BINDING_BIT, 1u));
 249                 queueRequirements.push_back(QueueRequirements(getQueueFlags(), 1u));
 250
 251                 createDeviceSupportingQueues(queueRequirements);
 252         }
 253
 254         // Create queues supporting sparse binding operations and compute/graphics operations
 255         const Queue& sparseQueue        = getQueue(VK_QUEUE_SPARSE_BINDING_BIT, 0);
 256         const Queue& extractQueue       = getQueue(getQueueFlags(), 0);
 257
 258         // Create memory allocator for logical device
 259         const de::UniquePtr<Allocator> allocator(new SimpleAllocator(deviceInterface, *m_logicalDevice, getPhysicalDeviceMemoryProperties(instance, physicalDevice)));
 260
 261         // Create sparse image
 262         const Unique<VkImage> imageSparse(createImage(deviceInterface, *m_logicalDevice, &imageSparseInfo));
 263
 264         // Create sparse image memory bind semaphore
 265         const Unique<VkSemaphore> memoryBindSemaphore(makeSemaphore(deviceInterface, *m_logicalDevice));
 266
 267         const deUint32                    imageSparseSizeInBytes                = getImageSizeInBytes(imageSparseInfo.extent, imageSparseInfo.arrayLayers, m_format, imageSparseInfo.mipLevels, MEM_ALIGN_BUFFERIMAGECOPY_OFFSET);
 268         const deUint32                    imageSizeInPixels                             = getImageSizeInBytes(imageSparseInfo.extent, imageSparseInfo.arrayLayers, m_format, imageSparseInfo.mipLevels) / tcu::getPixelSize(m_format);
 269
 270         residencyReferenceData.assign(imageSizeInPixels, MEMORY_BLOCK_NOT_BOUND_VALUE);
 271
 272         {
 273                 // Get sparse image general memory requirements
 274                 const VkMemoryRequirements imageMemoryRequirements = getImageMemoryRequirements(deviceInterface, *m_logicalDevice, *imageSparse);
 275
 276                 // Check if required image memory size does not exceed device limits
 277                 if (imageMemoryRequirements.size > getPhysicalDeviceProperties(instance, physicalDevice).limits.sparseAddressSpaceSize)
 278                         TCU_THROW(NotSupportedError, "Required memory size for sparse resource exceeds device limits");
 279
 280                 DE_ASSERT((imageMemoryRequirements.size % imageMemoryRequirements.alignment) == 0);
 281
 282                 // Get sparse image sparse memory requirements
 283                 const std::vector<VkSparseImageMemoryRequirements> sparseMemoryRequirements = getImageSparseMemoryRequirements(deviceInterface, *m_logicalDevice, *imageSparse);
 284
 285                 DE_ASSERT(sparseMemoryRequirements.size() != 0);
 286
 287                 const deUint32 colorAspectIndex = getSparseAspectRequirementsIndex(sparseMemoryRequirements, VK_IMAGE_ASPECT_COLOR_BIT);
 288
 289                 if (colorAspectIndex == NO_MATCH_FOUND)
 290                         TCU_THROW(NotSupportedError, "Not supported image aspect - the test supports currently only VK_IMAGE_ASPECT_COLOR_BIT");
 291
 292                 aspectRequirements = sparseMemoryRequirements[colorAspectIndex];
 293
 294                 DE_ASSERT((aspectRequirements.imageMipTailSize % imageMemoryRequirements.alignment) == 0);
 295
 296                 const VkImageAspectFlags aspectMask                     = aspectRequirements.formatProperties.aspectMask;
 297                 const VkExtent3D                 imageGranularity       = aspectRequirements.formatProperties.imageGranularity;
 298                 const deUint32                   memoryType                     = findMatchingMemoryType(instance, physicalDevice, imageMemoryRequirements, MemoryRequirement::Any);
 299
 300                 if (memoryType == NO_MATCH_FOUND)
 301                         return tcu::TestStatus::fail("No matching memory type found");
 302
 303                 deUint32 pixelOffset = 0u;
 304
 305                 std::vector<VkSparseImageMemoryBind>  imageResidencyMemoryBinds;
 306                 std::vector<VkSparseMemoryBind>           imageMipTailBinds;
 307
 308                 // Bind memory for each mipmap level
 309                 for (deUint32 mipLevelNdx = 0; mipLevelNdx < aspectRequirements.imageMipTailFirstLod; ++mipLevelNdx)
 310                 {
 311                         const deUint32 mipLevelSizeInPixels = getImageMipLevelSizeInBytes(imageSparseInfo.extent, imageSparseInfo.arrayLayers, m_format, mipLevelNdx) / tcu::getPixelSize(m_format);
 312
 313                         if (mipLevelNdx % MEMORY_BLOCK_TYPE_COUNT == MEMORY_BLOCK_NOT_BOUND)
 314                         {
 315                                 pixelOffset += mipLevelSizeInPixels;
 316                                 continue;
 317                         }
 318
 319                         for (deUint32 pixelNdx = 0u; pixelNdx < mipLevelSizeInPixels; ++pixelNdx)
 320                         {
 321                                 residencyReferenceData[pixelOffset + pixelNdx] = MEMORY_BLOCK_BOUND_VALUE;
 322                         }
 323
 324                         pixelOffset += mipLevelSizeInPixels;
 325
 326                         for (deUint32 layerNdx = 0; layerNdx < imageSparseInfo.arrayLayers; ++layerNdx)
 327                         {
 328                                 const VkExtent3D                 mipExtent                      = mipLevelExtents(imageSparseInfo.extent, mipLevelNdx);
 329                                 const tcu::UVec3                 sparseBlocks           = alignedDivide(mipExtent, imageGranularity);
 330                                 const deUint32                   numSparseBlocks        = sparseBlocks.x() * sparseBlocks.y() * sparseBlocks.z();
 331                                 const VkImageSubresource subresource            = { aspectMask, mipLevelNdx, layerNdx };
 332
 333                                 const VkSparseImageMemoryBind imageMemoryBind = makeSparseImageMemoryBind(deviceInterface, *m_logicalDevice,
 334                                         imageMemoryRequirements.alignment * numSparseBlocks, memoryType, subresource, makeOffset3D(0u, 0u, 0u), mipExtent);
 335
 336                                 deviceMemUniquePtrVec.push_back(makeVkSharedPtr(Move<VkDeviceMemory>(check<VkDeviceMemory>(imageMemoryBind.memory), Deleter<VkDeviceMemory>(deviceInterface, *m_logicalDevice, DE_NULL))));
 337
 338                                 imageResidencyMemoryBinds.push_back(imageMemoryBind);
 339                         }
 340                 }
 341
 342                 if (aspectRequirements.imageMipTailFirstLod < imageSparseInfo.mipLevels)
 343                 {
 344                         if (aspectRequirements.formatProperties.flags & VK_SPARSE_IMAGE_FORMAT_SINGLE_MIPTAIL_BIT)
 345                         {
 346                                 const VkSparseMemoryBind imageMipTailMemoryBind = makeSparseMemoryBind(deviceInterface, *m_logicalDevice,
 347                                         aspectRequirements.imageMipTailSize, memoryType, aspectRequirements.imageMipTailOffset);
 348
 349                                 deviceMemUniquePtrVec.push_back(makeVkSharedPtr(Move<VkDeviceMemory>(check<VkDeviceMemory>(imageMipTailMemoryBind.memory), Deleter<VkDeviceMemory>(deviceInterface, *m_logicalDevice, DE_NULL))));
 350
 351                                 imageMipTailBinds.push_back(imageMipTailMemoryBind);
 352                         }
 353                         else
 354                         {
 355                                 for (deUint32 layerNdx = 0; layerNdx < imageSparseInfo.arrayLayers; ++layerNdx)
 356                                 {
 357                                         const VkSparseMemoryBind imageMipTailMemoryBind = makeSparseMemoryBind(deviceInterface, *m_logicalDevice,
 358                                                 aspectRequirements.imageMipTailSize, memoryType, aspectRequirements.imageMipTailOffset + layerNdx * aspectRequirements.imageMipTailStride);
 359
 360                                         deviceMemUniquePtrVec.push_back(makeVkSharedPtr(Move<VkDeviceMemory>(check<VkDeviceMemory>(imageMipTailMemoryBind.memory), Deleter<VkDeviceMemory>(deviceInterface, *m_logicalDevice, DE_NULL))));
 361
 362                                         imageMipTailBinds.push_back(imageMipTailMemoryBind);
 363                                 }
 364                         }
 365
 366                         for (deUint32 pixelNdx = pixelOffset; pixelNdx < residencyReferenceData.size(); ++pixelNdx)
 367                         {
 368                                 residencyReferenceData[pixelNdx] = MEMORY_BLOCK_BOUND_VALUE;
 369                         }
 370                 }
 371
 372                 VkBindSparseInfo bindSparseInfo =
 373                 {
 374                         VK_STRUCTURE_TYPE_BIND_SPARSE_INFO,     //VkStructureType                                                       sType;
 375                         DE_NULL,                                                        //const void*                                                           pNext;
 376                         0u,                                                                     //deUint32                                                                      waitSemaphoreCount;
 377                         DE_NULL,                                                        //const VkSemaphore*                                            pWaitSemaphores;
 378                         0u,                                                                     //deUint32                                                                      bufferBindCount;
 379                         DE_NULL,                                                        //const VkSparseBufferMemoryBindInfo*           pBufferBinds;
 380                         0u,                                                                     //deUint32                                                                      imageOpaqueBindCount;
 381                         DE_NULL,                                                        //const VkSparseImageOpaqueMemoryBindInfo*      pImageOpaqueBinds;
 382                         0u,                                                                     //deUint32                                                                      imageBindCount;
 383                         DE_NULL,                                                        //const VkSparseImageMemoryBindInfo*            pImageBinds;
 384                         1u,                                                                     //deUint32                                                                      signalSemaphoreCount;
 385                         &memoryBindSemaphore.get()                      //const VkSemaphore*                                            pSignalSemaphores;
 386                 };
 387
 388                 VkSparseImageMemoryBindInfo               imageResidencyBindInfo;
 389                 VkSparseImageOpaqueMemoryBindInfo imageMipTailBindInfo;
 390
 391                 if (imageResidencyMemoryBinds.size() > 0)
 392                 {
 393                         imageResidencyBindInfo.image            = *imageSparse;
 394                         imageResidencyBindInfo.bindCount        = static_cast<deUint32>(imageResidencyMemoryBinds.size());
 395                         imageResidencyBindInfo.pBinds           = &imageResidencyMemoryBinds[0];
 396
 397                         bindSparseInfo.imageBindCount           = 1u;
 398                         bindSparseInfo.pImageBinds                      = &imageResidencyBindInfo;
 399                 }
 400
 401                 if (imageMipTailBinds.size() > 0)
 402                 {
 403                         imageMipTailBindInfo.image                      = *imageSparse;
 404                         imageMipTailBindInfo.bindCount          = static_cast<deUint32>(imageMipTailBinds.size());
 405                         imageMipTailBindInfo.pBinds                     = &imageMipTailBinds[0];
 406
 407                         bindSparseInfo.imageOpaqueBindCount = 1u;
 408                         bindSparseInfo.pImageOpaqueBinds        = &imageMipTailBindInfo;
 409                 }
 410
 411                 // Submit sparse bind commands for execution
 412                 VK_CHECK(deviceInterface.queueBindSparse(sparseQueue.queueHandle, 1u, &bindSparseInfo, DE_NULL));
 413         }
 414
 415         // Create image to store texels copied from sparse image
 416         imageTexelsInfo.sType                                   = VK_STRUCTURE_TYPE_IMAGE_CREATE_INFO;
 417         imageTexelsInfo.pNext                                   = DE_NULL;
 418         imageTexelsInfo.flags                                   = 0u;
 419         imageTexelsInfo.imageType                               = imageSparseInfo.imageType;
 420         imageTexelsInfo.format                                  = imageSparseInfo.format;
 421         imageTexelsInfo.extent                                  = imageSparseInfo.extent;
 422         imageTexelsInfo.arrayLayers                             = imageSparseInfo.arrayLayers;
 423         imageTexelsInfo.mipLevels                               = imageSparseInfo.mipLevels;
 424         imageTexelsInfo.samples                                 = imageSparseInfo.samples;
 425         imageTexelsInfo.tiling                                  = VK_IMAGE_TILING_OPTIMAL;
 426         imageTexelsInfo.initialLayout                   = VK_IMAGE_LAYOUT_UNDEFINED;
 427         imageTexelsInfo.usage                                   = VK_IMAGE_USAGE_TRANSFER_SRC_BIT | imageOutputUsageFlags();
 428         imageTexelsInfo.sharingMode                             = VK_SHARING_MODE_EXCLUSIVE;
 429         imageTexelsInfo.queueFamilyIndexCount   = 0u;
 430         imageTexelsInfo.pQueueFamilyIndices             = DE_NULL;
 431
 432         if (m_imageType == IMAGE_TYPE_CUBE || m_imageType == IMAGE_TYPE_CUBE_ARRAY)
 433         {
 434                 imageTexelsInfo.flags |= VK_IMAGE_CREATE_CUBE_COMPATIBLE_BIT;
 435         }
 436
 437         const de::UniquePtr<Image> imageTexels(new Image(deviceInterface, *m_logicalDevice, *allocator, imageTexelsInfo, MemoryRequirement::Any));
 438
 439         // Create image to store residency info copied from sparse image
 440         imageResidencyInfo                      = imageTexelsInfo;
 441         imageResidencyInfo.format       = mapTextureFormat(m_residencyFormat);
 442
 443         const de::UniquePtr<Image> imageResidency(new Image(deviceInterface, *m_logicalDevice, *allocator, imageResidencyInfo, MemoryRequirement::Any));
 444
 445         // Create command buffer for compute and transfer oparations
 446         const Unique<VkCommandPool>       commandPool(makeCommandPool(deviceInterface, *m_logicalDevice, extractQueue.queueFamilyIndex));
 447         const Unique<VkCommandBuffer> commandBuffer(makeCommandBuffer(deviceInterface, *m_logicalDevice, *commandPool));
 448
 449         std::vector <VkBufferImageCopy> bufferImageSparseCopy(imageSparseInfo.mipLevels);
 450
 451         {
 452                 deUint32 bufferOffset = 0u;
 453                 for (deUint32 mipLevelNdx = 0u; mipLevelNdx < imageSparseInfo.mipLevels; ++mipLevelNdx)
 454                 {
 455                         bufferImageSparseCopy[mipLevelNdx] = makeBufferImageCopy(mipLevelExtents(imageSparseInfo.extent, mipLevelNdx), imageSparseInfo.arrayLayers, mipLevelNdx, static_cast<VkDeviceSize>(bufferOffset));
 456                         bufferOffset += getImageMipLevelSizeInBytes(imageSparseInfo.extent, imageSparseInfo.arrayLayers, m_format, mipLevelNdx, MEM_ALIGN_BUFFERIMAGECOPY_OFFSET);
 457                 }
 458         }
 459
 460         // Start recording commands
 461         beginCommandBuffer(deviceInterface, *commandBuffer);
 462
 463         // Create input buffer
 464         const VkBufferCreateInfo        inputBufferCreateInfo = makeBufferCreateInfo(imageSparseSizeInBytes, VK_BUFFER_USAGE_TRANSFER_SRC_BIT);
 465         const de::UniquePtr<Buffer>     inputBuffer(new Buffer(deviceInterface, *m_logicalDevice, *allocator, inputBufferCreateInfo, MemoryRequirement::HostVisible));
 466
 467         // Fill input buffer with reference data
 468         std::vector<deUint8> referenceData(imageSparseSizeInBytes);
 469
 470         for (deUint32 mipLevelNdx = 0u; mipLevelNdx < imageSparseInfo.mipLevels; ++mipLevelNdx)
 471         {
 472                 const deUint32 mipLevelSizeinBytes      = getImageMipLevelSizeInBytes(imageSparseInfo.extent, imageSparseInfo.arrayLayers, m_format, mipLevelNdx);
 473                 const deUint32 bufferOffset                     = static_cast<deUint32>(bufferImageSparseCopy[mipLevelNdx].bufferOffset);
 474
 475                 for (deUint32 byteNdx = 0u; byteNdx < mipLevelSizeinBytes; ++byteNdx)
 476                 {
 477                         referenceData[bufferOffset + byteNdx] = (deUint8)(mipLevelNdx + byteNdx);
 478                 }
 479         }
 480
 481         deMemcpy(inputBuffer->getAllocation().getHostPtr(), &referenceData[0], imageSparseSizeInBytes);
 482         flushMappedMemoryRange(deviceInterface, *m_logicalDevice, inputBuffer->getAllocation().getMemory(), inputBuffer->getAllocation().getOffset(), imageSparseSizeInBytes);
 483
 484         {
 485                 // Prepare input buffer for data transfer operation
 486                 const VkBufferMemoryBarrier inputBufferBarrier = makeBufferMemoryBarrier
 487                 (
 488                         VK_ACCESS_HOST_WRITE_BIT,
 489                         VK_ACCESS_TRANSFER_READ_BIT,
 490                         inputBuffer->get(),
 491                         0u,
 492                         imageSparseSizeInBytes
 493                 );
 494
 495                 deviceInterface.cmdPipelineBarrier(*commandBuffer, VK_PIPELINE_STAGE_HOST_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT, 0u, 0u, DE_NULL, 1u, &inputBufferBarrier, 0u, DE_NULL);
 496         }
 497
 498         const VkImageSubresourceRange fullImageSubresourceRange = makeImageSubresourceRange(VK_IMAGE_ASPECT_COLOR_BIT, 0u, imageSparseInfo.mipLevels, 0u, imageSparseInfo.arrayLayers);
 499
 500         {
 501                 // Prepare sparse image for data transfer operation
 502                 const VkImageMemoryBarrier imageSparseTransferDstBarrier = makeImageMemoryBarrier
 503                 (
 504                         0u,
 505                         VK_ACCESS_TRANSFER_WRITE_BIT,
 506                         VK_IMAGE_LAYOUT_UNDEFINED,
 507                         VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL,
 508                         sparseQueue.queueFamilyIndex != extractQueue.queueFamilyIndex ? sparseQueue.queueFamilyIndex  : VK_QUEUE_FAMILY_IGNORED,
 509                         sparseQueue.queueFamilyIndex != extractQueue.queueFamilyIndex ? extractQueue.queueFamilyIndex : VK_QUEUE_FAMILY_IGNORED,
 510                         *imageSparse,
 511                         fullImageSubresourceRange
 512                 );
 513
 514                 deviceInterface.cmdPipelineBarrier(*commandBuffer, VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT, 0u, 0u, DE_NULL, 0u, DE_NULL, 1u, &imageSparseTransferDstBarrier);
 515         }
 516
 517         // Copy reference data from input buffer to sparse image
 518         deviceInterface.cmdCopyBufferToImage(*commandBuffer, inputBuffer->get(), *imageSparse, VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, static_cast<deUint32>(bufferImageSparseCopy.size()), &bufferImageSparseCopy[0]);
 519
 520         recordCommands(*allocator, *commandBuffer, imageSparseInfo, *imageSparse, imageTexels->get(), imageResidency->get());
 521
 522         const VkBufferCreateInfo        bufferTexelsInfo = makeBufferCreateInfo(imageSparseSizeInBytes, VK_BUFFER_USAGE_TRANSFER_DST_BIT);
 523         const de::UniquePtr<Buffer>     bufferTexels(new Buffer(deviceInterface, *m_logicalDevice, *allocator, bufferTexelsInfo, MemoryRequirement::HostVisible));
 524
 525         // Copy data from texels image to buffer
 526         deviceInterface.cmdCopyImageToBuffer(*commandBuffer, imageTexels->get(), VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL, bufferTexels->get(), static_cast<deUint32>(bufferImageSparseCopy.size()), &bufferImageSparseCopy[0]);
 527
 528         const deUint32                          imageResidencySizeInBytes = getImageSizeInBytes(imageSparseInfo.extent, imageSparseInfo.arrayLayers, m_residencyFormat, imageSparseInfo.mipLevels, MEM_ALIGN_BUFFERIMAGECOPY_OFFSET);
 529
 530         const VkBufferCreateInfo        bufferResidencyInfo = makeBufferCreateInfo(imageResidencySizeInBytes, VK_BUFFER_USAGE_TRANSFER_DST_BIT);
 531         const de::UniquePtr<Buffer>     bufferResidency(new Buffer(deviceInterface, *m_logicalDevice, *allocator, bufferResidencyInfo, MemoryRequirement::HostVisible));
 532
 533         // Copy data from residency image to buffer
 534         std::vector <VkBufferImageCopy> bufferImageResidencyCopy(imageSparseInfo.mipLevels);
 535
 536         {
 537                 deUint32 bufferOffset = 0u;
 538                 for (deUint32 mipLevelNdx = 0u; mipLevelNdx < imageSparseInfo.mipLevels; ++mipLevelNdx)
 539                 {
 540                         bufferImageResidencyCopy[mipLevelNdx] = makeBufferImageCopy(mipLevelExtents(imageSparseInfo.extent, mipLevelNdx), imageSparseInfo.arrayLayers, mipLevelNdx, static_cast<VkDeviceSize>(bufferOffset));
 541                         bufferOffset += getImageMipLevelSizeInBytes(imageSparseInfo.extent, imageSparseInfo.arrayLayers, m_residencyFormat, mipLevelNdx, MEM_ALIGN_BUFFERIMAGECOPY_OFFSET);
 542                 }
 543         }
 544
 545         deviceInterface.cmdCopyImageToBuffer(*commandBuffer, imageResidency->get(), VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL, bufferResidency->get(), static_cast<deUint32>(bufferImageResidencyCopy.size()), &bufferImageResidencyCopy[0]);
 546
 547         {
 548                 VkBufferMemoryBarrier bufferOutputHostReadBarriers[2];
 549
 550                 bufferOutputHostReadBarriers[0] = makeBufferMemoryBarrier
 551                 (
 552                         VK_ACCESS_TRANSFER_WRITE_BIT,
 553                         VK_ACCESS_HOST_READ_BIT,
 554                         bufferTexels->get(),
 555                         0u,
 556                         imageSparseSizeInBytes
 557                 );
 558
 559                 bufferOutputHostReadBarriers[1] = makeBufferMemoryBarrier
 560                 (
 561                         VK_ACCESS_TRANSFER_WRITE_BIT,
 562                         VK_ACCESS_HOST_READ_BIT,
 563                         bufferResidency->get(),
 564                         0u,
 565                         imageResidencySizeInBytes
 566                 );
 567
 568                 deviceInterface.cmdPipelineBarrier(*commandBuffer, VK_PIPELINE_STAGE_TRANSFER_BIT, VK_PIPELINE_STAGE_HOST_BIT, 0u, 0u, DE_NULL, 2u, bufferOutputHostReadBarriers, 0u, DE_NULL);
 569         }
 570
 571         // End recording commands
 572         endCommandBuffer(deviceInterface, *commandBuffer);
 573
 574         const VkPipelineStageFlags stageBits[] = { VK_PIPELINE_STAGE_TRANSFER_BIT };
 575
 576         // Submit commands for execution and wait for completion
 577         submitCommandsAndWait(deviceInterface, *m_logicalDevice, extractQueue.queueHandle, *commandBuffer, 1u, &memoryBindSemaphore.get(), stageBits);
 578
 579         // Wait for sparse queue to become idle
 580         deviceInterface.queueWaitIdle(sparseQueue.queueHandle);
 581
 582         // Retrieve data from residency buffer to host memory
 583         const Allocation& bufferResidencyAllocation = bufferResidency->getAllocation();
 584         invalidateMappedMemoryRange(deviceInterface, *m_logicalDevice, bufferResidencyAllocation.getMemory(), bufferResidencyAllocation.getOffset(), imageResidencySizeInBytes);
 585
 586         const deUint32* bufferResidencyData = static_cast<const deUint32*>(bufferResidencyAllocation.getHostPtr());
 587
 588         deUint32 pixelOffsetNotAligned = 0u;
 589         for (deUint32 mipmapNdx = 0; mipmapNdx < imageSparseInfo.mipLevels; ++mipmapNdx)
 590         {
 591                 const deUint32 mipLevelSizeInBytes      = getImageMipLevelSizeInBytes(imageSparseInfo.extent, imageSparseInfo.arrayLayers, m_residencyFormat, mipmapNdx);
 592                 const deUint32 pixelOffsetAligned       = static_cast<deUint32>(bufferImageResidencyCopy[mipmapNdx].bufferOffset) / tcu::getPixelSize(m_residencyFormat);
 593
 594                 if (deMemCmp(&bufferResidencyData[pixelOffsetAligned], &residencyReferenceData[pixelOffsetNotAligned], mipLevelSizeInBytes) != 0)
 595                         return tcu::TestStatus::fail("Failed");
 596
 597                 pixelOffsetNotAligned += mipLevelSizeInBytes / tcu::getPixelSize(m_residencyFormat);
 598         }
 599
 600         // Retrieve data from texels buffer to host memory
 601         const Allocation& bufferTexelsAllocation = bufferTexels->getAllocation();
 602         invalidateMappedMemoryRange(deviceInterface, *m_logicalDevice, bufferTexelsAllocation.getMemory(), bufferTexelsAllocation.getOffset(), imageSparseSizeInBytes);
 603
 604         const deUint8* bufferTexelsData = static_cast<const deUint8*>(bufferTexelsAllocation.getHostPtr());
 605
 606         for (deUint32 mipmapNdx = 0; mipmapNdx < imageSparseInfo.mipLevels; ++mipmapNdx)
 607         {
 608                 const deUint32 mipLevelSizeInBytes      = getImageMipLevelSizeInBytes(imageSparseInfo.extent, imageSparseInfo.arrayLayers, m_format, mipmapNdx);
 609                 const deUint32 bufferOffset                     = static_cast<deUint32>(bufferImageSparseCopy[mipmapNdx].bufferOffset);
 610
 611                 if (mipmapNdx < aspectRequirements.imageMipTailFirstLod)
 612                 {
 613                         if (mipmapNdx % MEMORY_BLOCK_TYPE_COUNT == MEMORY_BLOCK_BOUND)
 614                         {
 615                                 if (deMemCmp(&bufferTexelsData[bufferOffset], &referenceData[bufferOffset], mipLevelSizeInBytes) != 0)
 616                                         return tcu::TestStatus::fail("Failed");
 617                         }
 618                         else if (getPhysicalDeviceProperties(instance, physicalDevice).sparseProperties.residencyNonResidentStrict)
 619                         {
 620                                 std::vector<deUint8> zeroData;
 621                                 zeroData.assign(mipLevelSizeInBytes, 0u);
 622
 623                                 if (deMemCmp(&bufferTexelsData[bufferOffset], &zeroData[0], mipLevelSizeInBytes) != 0)
 624                                         return tcu::TestStatus::fail("Failed");
 625                         }
 626                 }
 627                 else
 628                 {
 629                         if (deMemCmp(&bufferTexelsData[bufferOffset], &referenceData[bufferOffset], mipLevelSizeInBytes) != 0)
 630                                 return tcu::TestStatus::fail("Failed");
 631                 }
 632         }
 633
 634         return tcu::TestStatus::pass("Passed");
 635 }
 636
 637 } // sparse
 638 } // vkt