1 /*------------------------------------------------------------------------
2 * Vulkan Conformance Tests
3 * ------------------------
5 * Copyright (c) 2016 The Khronos Group Inc.
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
11 * http://www.apache.org/licenses/LICENSE-2.0
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.
20 * \file vktSparseResourcesShaderIntrinsicsBase.cpp
21 * \brief Sparse Resources Shader Intrinsics Base Classes
22 *//*--------------------------------------------------------------------*/
24 #include "vktSparseResourcesShaderIntrinsicsBase.hpp"
33 tcu::UVec3 alignedDivide (const VkExtent3D& extent, const VkExtent3D& divisor)
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);
44 std::string getOpTypeImageComponent (const tcu::TextureFormat& format)
46 switch (tcu::getTextureChannelClass(format.type))
48 case tcu::TEXTURECHANNELCLASS_UNSIGNED_INTEGER:
49 return "OpTypeInt 32 0";
50 case tcu::TEXTURECHANNELCLASS_SIGNED_INTEGER:
51 return "OpTypeInt 32 1";
58 std::string getImageComponentTypeName (const tcu::TextureFormat& format)
60 switch (tcu::getTextureChannelClass(format.type))
62 case tcu::TEXTURECHANNELCLASS_UNSIGNED_INTEGER:
64 case tcu::TEXTURECHANNELCLASS_SIGNED_INTEGER:
72 std::string getImageComponentVec4TypeName (const tcu::TextureFormat& format)
74 switch (tcu::getTextureChannelClass(format.type))
76 case tcu::TEXTURECHANNELCLASS_UNSIGNED_INTEGER:
78 case tcu::TEXTURECHANNELCLASS_SIGNED_INTEGER:
86 std::string getOpTypeImageSparse (const ImageType imageType,
87 const tcu::TextureFormat& format,
88 const std::string& componentType,
89 const bool requiresSampler)
91 std::ostringstream src;
93 src << "OpTypeImage " << componentType << " ";
100 case IMAGE_TYPE_1D_ARRAY :
106 case IMAGE_TYPE_2D_ARRAY :
112 case IMAGE_TYPE_CUBE :
113 src << "Cube 0 0 0 ";
115 case IMAGE_TYPE_CUBE_ARRAY :
116 src << "Cube 0 1 0 ";
128 switch (format.order)
130 case tcu::TextureFormat::R:
133 case tcu::TextureFormat::RG:
136 case tcu::TextureFormat::RGB:
139 case tcu::TextureFormat::RGBA:
149 case tcu::TextureFormat::SIGNED_INT8:
152 case tcu::TextureFormat::SIGNED_INT16:
155 case tcu::TextureFormat::SIGNED_INT32:
158 case tcu::TextureFormat::UNSIGNED_INT8:
161 case tcu::TextureFormat::UNSIGNED_INT16:
164 case tcu::TextureFormat::UNSIGNED_INT32:
175 std::string getOpTypeImageResidency (const ImageType imageType)
177 std::ostringstream src;
179 src << "OpTypeImage %type_uint ";
184 src << "1D 0 0 0 2 R32ui";
186 case IMAGE_TYPE_1D_ARRAY :
187 src << "1D 0 1 0 2 R32ui";
190 src << "2D 0 0 0 2 R32ui";
192 case IMAGE_TYPE_2D_ARRAY :
193 src << "2D 0 1 0 2 R32ui";
196 src << "3D 0 0 0 2 R32ui";
198 case IMAGE_TYPE_CUBE :
199 src << "Cube 0 0 0 2 R32ui";
201 case IMAGE_TYPE_CUBE_ARRAY :
202 src << "Cube 0 1 0 2 R32ui";
212 tcu::TestStatus SparseShaderIntrinsicsInstanceBase::iterate (void)
214 const InstanceInterface& instance = m_context.getInstanceInterface();
215 const VkPhysicalDevice physicalDevice = m_context.getPhysicalDevice();
216 VkImageCreateInfo imageSparseInfo;
217 VkImageCreateInfo imageTexelsInfo;
218 VkImageCreateInfo imageResidencyInfo;
219 VkSparseImageMemoryRequirements aspectRequirements;
220 std::vector <deUint32> residencyReferenceData;
221 std::vector<DeviceMemorySp> deviceMemUniquePtrVec;
223 // Check if image size does not exceed device limits
224 if (!isImageSizeSupported(instance, physicalDevice, m_imageType, m_imageSize))
225 TCU_THROW(NotSupportedError, "Image size not supported for device");
227 // Check if device supports sparse operations for image type
228 if (!checkSparseSupportForImageType(instance, physicalDevice, m_imageType))
229 TCU_THROW(NotSupportedError, "Sparse residency for image type is not supported");
231 if (!getPhysicalDeviceFeatures(instance, physicalDevice).shaderResourceResidency)
232 TCU_THROW(NotSupportedError, "Sparse resource residency information not supported in shader code.");
234 imageSparseInfo.sType = VK_STRUCTURE_TYPE_IMAGE_CREATE_INFO;
235 imageSparseInfo.pNext = DE_NULL;
236 imageSparseInfo.flags = VK_IMAGE_CREATE_SPARSE_RESIDENCY_BIT | VK_IMAGE_CREATE_SPARSE_BINDING_BIT;
237 imageSparseInfo.imageType = mapImageType(m_imageType);
238 imageSparseInfo.format = mapTextureFormat(m_format);
239 imageSparseInfo.extent = makeExtent3D(getLayerSize(m_imageType, m_imageSize));
240 imageSparseInfo.arrayLayers = getNumLayers(m_imageType, m_imageSize);
241 imageSparseInfo.samples = VK_SAMPLE_COUNT_1_BIT;
242 imageSparseInfo.tiling = VK_IMAGE_TILING_OPTIMAL;
243 imageSparseInfo.initialLayout = VK_IMAGE_LAYOUT_UNDEFINED;
244 imageSparseInfo.usage = VK_IMAGE_USAGE_TRANSFER_DST_BIT | imageSparseUsageFlags();
245 imageSparseInfo.sharingMode = VK_SHARING_MODE_EXCLUSIVE;
246 imageSparseInfo.queueFamilyIndexCount = 0u;
247 imageSparseInfo.pQueueFamilyIndices = DE_NULL;
249 if (m_imageType == IMAGE_TYPE_CUBE || m_imageType == IMAGE_TYPE_CUBE_ARRAY)
251 imageSparseInfo.flags |= VK_IMAGE_CREATE_CUBE_COMPATIBLE_BIT;
255 // Assign maximum allowed mipmap levels to image
256 VkImageFormatProperties imageFormatProperties;
257 instance.getPhysicalDeviceImageFormatProperties(physicalDevice,
258 imageSparseInfo.format,
259 imageSparseInfo.imageType,
260 imageSparseInfo.tiling,
261 imageSparseInfo.usage,
262 imageSparseInfo.flags,
263 &imageFormatProperties);
265 imageSparseInfo.mipLevels = getImageMaxMipLevels(imageFormatProperties, imageSparseInfo.extent);
268 // Check if device supports sparse operations for image format
269 if (!checkSparseSupportForImageFormat(instance, physicalDevice, imageSparseInfo))
270 TCU_THROW(NotSupportedError, "The image format does not support sparse operations");
273 // Create logical device supporting both sparse and compute/graphics queues
274 QueueRequirementsVec queueRequirements;
275 queueRequirements.push_back(QueueRequirements(VK_QUEUE_SPARSE_BINDING_BIT, 1u));
276 queueRequirements.push_back(QueueRequirements(getQueueFlags(), 1u));
278 createDeviceSupportingQueues(queueRequirements);
281 const DeviceInterface& deviceInterface = getDeviceInterface();
283 // Create queues supporting sparse binding operations and compute/graphics operations
284 const Queue& sparseQueue = getQueue(VK_QUEUE_SPARSE_BINDING_BIT, 0);
285 const Queue& extractQueue = getQueue(getQueueFlags(), 0);
287 // Create sparse image
288 const Unique<VkImage> imageSparse(createImage(deviceInterface, getDevice(), &imageSparseInfo));
290 // Create sparse image memory bind semaphore
291 const Unique<VkSemaphore> memoryBindSemaphore(makeSemaphore(deviceInterface, getDevice()));
293 const deUint32 imageSparseSizeInBytes = getImageSizeInBytes(imageSparseInfo.extent, imageSparseInfo.arrayLayers, m_format, imageSparseInfo.mipLevels, BUFFER_IMAGE_COPY_OFFSET_GRANULARITY);
294 const deUint32 imageSizeInPixels = getImageSizeInBytes(imageSparseInfo.extent, imageSparseInfo.arrayLayers, m_format, imageSparseInfo.mipLevels) / tcu::getPixelSize(m_format);
296 residencyReferenceData.assign(imageSizeInPixels, MEMORY_BLOCK_NOT_BOUND_VALUE);
299 // Get sparse image general memory requirements
300 const VkMemoryRequirements imageMemoryRequirements = getImageMemoryRequirements(deviceInterface, getDevice(), *imageSparse);
302 // Check if required image memory size does not exceed device limits
303 if (imageMemoryRequirements.size > getPhysicalDeviceProperties(instance, physicalDevice).limits.sparseAddressSpaceSize)
304 TCU_THROW(NotSupportedError, "Required memory size for sparse resource exceeds device limits");
306 DE_ASSERT((imageMemoryRequirements.size % imageMemoryRequirements.alignment) == 0);
308 // Get sparse image sparse memory requirements
309 const std::vector<VkSparseImageMemoryRequirements> sparseMemoryRequirements = getImageSparseMemoryRequirements(deviceInterface, getDevice(), *imageSparse);
311 DE_ASSERT(sparseMemoryRequirements.size() != 0);
313 const deUint32 colorAspectIndex = getSparseAspectRequirementsIndex(sparseMemoryRequirements, VK_IMAGE_ASPECT_COLOR_BIT);
315 if (colorAspectIndex == NO_MATCH_FOUND)
316 TCU_THROW(NotSupportedError, "Not supported image aspect - the test supports currently only VK_IMAGE_ASPECT_COLOR_BIT");
318 aspectRequirements = sparseMemoryRequirements[colorAspectIndex];
320 DE_ASSERT((aspectRequirements.imageMipTailSize % imageMemoryRequirements.alignment) == 0);
322 const VkImageAspectFlags aspectMask = aspectRequirements.formatProperties.aspectMask;
323 const VkExtent3D imageGranularity = aspectRequirements.formatProperties.imageGranularity;
324 const deUint32 memoryType = findMatchingMemoryType(instance, physicalDevice, imageMemoryRequirements, MemoryRequirement::Any);
326 if (memoryType == NO_MATCH_FOUND)
327 return tcu::TestStatus::fail("No matching memory type found");
329 deUint32 pixelOffset = 0u;
331 std::vector<VkSparseImageMemoryBind> imageResidencyMemoryBinds;
332 std::vector<VkSparseMemoryBind> imageMipTailBinds;
334 // Bind memory for each mipmap level
335 for (deUint32 mipLevelNdx = 0; mipLevelNdx < aspectRequirements.imageMipTailFirstLod; ++mipLevelNdx)
337 const deUint32 mipLevelSizeInPixels = getImageMipLevelSizeInBytes(imageSparseInfo.extent, imageSparseInfo.arrayLayers, m_format, mipLevelNdx) / tcu::getPixelSize(m_format);
339 if (mipLevelNdx % MEMORY_BLOCK_TYPE_COUNT == MEMORY_BLOCK_NOT_BOUND)
341 pixelOffset += mipLevelSizeInPixels;
345 for (deUint32 pixelNdx = 0u; pixelNdx < mipLevelSizeInPixels; ++pixelNdx)
347 residencyReferenceData[pixelOffset + pixelNdx] = MEMORY_BLOCK_BOUND_VALUE;
350 pixelOffset += mipLevelSizeInPixels;
352 for (deUint32 layerNdx = 0; layerNdx < imageSparseInfo.arrayLayers; ++layerNdx)
354 const VkExtent3D mipExtent = mipLevelExtents(imageSparseInfo.extent, mipLevelNdx);
355 const tcu::UVec3 sparseBlocks = alignedDivide(mipExtent, imageGranularity);
356 const deUint32 numSparseBlocks = sparseBlocks.x() * sparseBlocks.y() * sparseBlocks.z();
357 const VkImageSubresource subresource = { aspectMask, mipLevelNdx, layerNdx };
359 const VkSparseImageMemoryBind imageMemoryBind = makeSparseImageMemoryBind(deviceInterface, getDevice(),
360 imageMemoryRequirements.alignment * numSparseBlocks, memoryType, subresource, makeOffset3D(0u, 0u, 0u), mipExtent);
362 deviceMemUniquePtrVec.push_back(makeVkSharedPtr(Move<VkDeviceMemory>(check<VkDeviceMemory>(imageMemoryBind.memory), Deleter<VkDeviceMemory>(deviceInterface, getDevice(), DE_NULL))));
364 imageResidencyMemoryBinds.push_back(imageMemoryBind);
368 if (aspectRequirements.imageMipTailFirstLod < imageSparseInfo.mipLevels)
370 if (aspectRequirements.formatProperties.flags & VK_SPARSE_IMAGE_FORMAT_SINGLE_MIPTAIL_BIT)
372 const VkSparseMemoryBind imageMipTailMemoryBind = makeSparseMemoryBind(deviceInterface, getDevice(),
373 aspectRequirements.imageMipTailSize, memoryType, aspectRequirements.imageMipTailOffset);
375 deviceMemUniquePtrVec.push_back(makeVkSharedPtr(Move<VkDeviceMemory>(check<VkDeviceMemory>(imageMipTailMemoryBind.memory), Deleter<VkDeviceMemory>(deviceInterface, getDevice(), DE_NULL))));
377 imageMipTailBinds.push_back(imageMipTailMemoryBind);
381 for (deUint32 layerNdx = 0; layerNdx < imageSparseInfo.arrayLayers; ++layerNdx)
383 const VkSparseMemoryBind imageMipTailMemoryBind = makeSparseMemoryBind(deviceInterface, getDevice(),
384 aspectRequirements.imageMipTailSize, memoryType, aspectRequirements.imageMipTailOffset + layerNdx * aspectRequirements.imageMipTailStride);
386 deviceMemUniquePtrVec.push_back(makeVkSharedPtr(Move<VkDeviceMemory>(check<VkDeviceMemory>(imageMipTailMemoryBind.memory), Deleter<VkDeviceMemory>(deviceInterface, getDevice(), DE_NULL))));
388 imageMipTailBinds.push_back(imageMipTailMemoryBind);
392 for (deUint32 pixelNdx = pixelOffset; pixelNdx < residencyReferenceData.size(); ++pixelNdx)
394 residencyReferenceData[pixelNdx] = MEMORY_BLOCK_BOUND_VALUE;
398 VkBindSparseInfo bindSparseInfo =
400 VK_STRUCTURE_TYPE_BIND_SPARSE_INFO, //VkStructureType sType;
401 DE_NULL, //const void* pNext;
402 0u, //deUint32 waitSemaphoreCount;
403 DE_NULL, //const VkSemaphore* pWaitSemaphores;
404 0u, //deUint32 bufferBindCount;
405 DE_NULL, //const VkSparseBufferMemoryBindInfo* pBufferBinds;
406 0u, //deUint32 imageOpaqueBindCount;
407 DE_NULL, //const VkSparseImageOpaqueMemoryBindInfo* pImageOpaqueBinds;
408 0u, //deUint32 imageBindCount;
409 DE_NULL, //const VkSparseImageMemoryBindInfo* pImageBinds;
410 1u, //deUint32 signalSemaphoreCount;
411 &memoryBindSemaphore.get() //const VkSemaphore* pSignalSemaphores;
414 VkSparseImageMemoryBindInfo imageResidencyBindInfo;
415 VkSparseImageOpaqueMemoryBindInfo imageMipTailBindInfo;
417 if (imageResidencyMemoryBinds.size() > 0)
419 imageResidencyBindInfo.image = *imageSparse;
420 imageResidencyBindInfo.bindCount = static_cast<deUint32>(imageResidencyMemoryBinds.size());
421 imageResidencyBindInfo.pBinds = &imageResidencyMemoryBinds[0];
423 bindSparseInfo.imageBindCount = 1u;
424 bindSparseInfo.pImageBinds = &imageResidencyBindInfo;
427 if (imageMipTailBinds.size() > 0)
429 imageMipTailBindInfo.image = *imageSparse;
430 imageMipTailBindInfo.bindCount = static_cast<deUint32>(imageMipTailBinds.size());
431 imageMipTailBindInfo.pBinds = &imageMipTailBinds[0];
433 bindSparseInfo.imageOpaqueBindCount = 1u;
434 bindSparseInfo.pImageOpaqueBinds = &imageMipTailBindInfo;
437 // Submit sparse bind commands for execution
438 VK_CHECK(deviceInterface.queueBindSparse(sparseQueue.queueHandle, 1u, &bindSparseInfo, DE_NULL));
441 // Create image to store texels copied from sparse image
442 imageTexelsInfo.sType = VK_STRUCTURE_TYPE_IMAGE_CREATE_INFO;
443 imageTexelsInfo.pNext = DE_NULL;
444 imageTexelsInfo.flags = 0u;
445 imageTexelsInfo.imageType = imageSparseInfo.imageType;
446 imageTexelsInfo.format = imageSparseInfo.format;
447 imageTexelsInfo.extent = imageSparseInfo.extent;
448 imageTexelsInfo.arrayLayers = imageSparseInfo.arrayLayers;
449 imageTexelsInfo.mipLevels = imageSparseInfo.mipLevels;
450 imageTexelsInfo.samples = imageSparseInfo.samples;
451 imageTexelsInfo.tiling = VK_IMAGE_TILING_OPTIMAL;
452 imageTexelsInfo.initialLayout = VK_IMAGE_LAYOUT_UNDEFINED;
453 imageTexelsInfo.usage = VK_IMAGE_USAGE_TRANSFER_SRC_BIT | imageOutputUsageFlags();
454 imageTexelsInfo.sharingMode = VK_SHARING_MODE_EXCLUSIVE;
455 imageTexelsInfo.queueFamilyIndexCount = 0u;
456 imageTexelsInfo.pQueueFamilyIndices = DE_NULL;
458 if (m_imageType == IMAGE_TYPE_CUBE || m_imageType == IMAGE_TYPE_CUBE_ARRAY)
460 imageTexelsInfo.flags |= VK_IMAGE_CREATE_CUBE_COMPATIBLE_BIT;
463 const Unique<VkImage> imageTexels (createImage(deviceInterface, getDevice(), &imageTexelsInfo));
464 const de::UniquePtr<Allocation> imageTexelsAlloc (bindImage(deviceInterface, getDevice(), getAllocator(), *imageTexels, MemoryRequirement::Any));
466 // Create image to store residency info copied from sparse image
467 imageResidencyInfo = imageTexelsInfo;
468 imageResidencyInfo.format = mapTextureFormat(m_residencyFormat);
470 const Unique<VkImage> imageResidency (createImage(deviceInterface, getDevice(), &imageResidencyInfo));
471 const de::UniquePtr<Allocation> imageResidencyAlloc (bindImage(deviceInterface, getDevice(), getAllocator(), *imageResidency, MemoryRequirement::Any));
473 // Create command buffer for compute and transfer oparations
474 const Unique<VkCommandPool> commandPool(makeCommandPool(deviceInterface, getDevice(), extractQueue.queueFamilyIndex));
475 const Unique<VkCommandBuffer> commandBuffer(makeCommandBuffer(deviceInterface, getDevice(), *commandPool));
477 std::vector <VkBufferImageCopy> bufferImageSparseCopy(imageSparseInfo.mipLevels);
480 deUint32 bufferOffset = 0u;
481 for (deUint32 mipLevelNdx = 0u; mipLevelNdx < imageSparseInfo.mipLevels; ++mipLevelNdx)
483 bufferImageSparseCopy[mipLevelNdx] = makeBufferImageCopy(mipLevelExtents(imageSparseInfo.extent, mipLevelNdx), imageSparseInfo.arrayLayers, mipLevelNdx, static_cast<VkDeviceSize>(bufferOffset));
484 bufferOffset += getImageMipLevelSizeInBytes(imageSparseInfo.extent, imageSparseInfo.arrayLayers, m_format, mipLevelNdx, BUFFER_IMAGE_COPY_OFFSET_GRANULARITY);
488 // Start recording commands
489 beginCommandBuffer(deviceInterface, *commandBuffer);
491 // Create input buffer
492 const VkBufferCreateInfo inputBufferCreateInfo = makeBufferCreateInfo(imageSparseSizeInBytes, VK_BUFFER_USAGE_TRANSFER_SRC_BIT);
493 const Unique<VkBuffer> inputBuffer (createBuffer(deviceInterface, getDevice(), &inputBufferCreateInfo));
494 const de::UniquePtr<Allocation> inputBufferAlloc (bindBuffer(deviceInterface, getDevice(), getAllocator(), *inputBuffer, MemoryRequirement::HostVisible));
496 // Fill input buffer with reference data
497 std::vector<deUint8> referenceData(imageSparseSizeInBytes);
499 for (deUint32 mipLevelNdx = 0u; mipLevelNdx < imageSparseInfo.mipLevels; ++mipLevelNdx)
501 const deUint32 mipLevelSizeinBytes = getImageMipLevelSizeInBytes(imageSparseInfo.extent, imageSparseInfo.arrayLayers, m_format, mipLevelNdx);
502 const deUint32 bufferOffset = static_cast<deUint32>(bufferImageSparseCopy[mipLevelNdx].bufferOffset);
504 for (deUint32 byteNdx = 0u; byteNdx < mipLevelSizeinBytes; ++byteNdx)
506 referenceData[bufferOffset + byteNdx] = (deUint8)(mipLevelNdx + byteNdx);
510 deMemcpy(inputBufferAlloc->getHostPtr(), &referenceData[0], imageSparseSizeInBytes);
511 flushMappedMemoryRange(deviceInterface, getDevice(), inputBufferAlloc->getMemory(), inputBufferAlloc->getOffset(), imageSparseSizeInBytes);
514 // Prepare input buffer for data transfer operation
515 const VkBufferMemoryBarrier inputBufferBarrier = makeBufferMemoryBarrier
517 VK_ACCESS_HOST_WRITE_BIT,
518 VK_ACCESS_TRANSFER_READ_BIT,
521 imageSparseSizeInBytes
524 deviceInterface.cmdPipelineBarrier(*commandBuffer, VK_PIPELINE_STAGE_HOST_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT, 0u, 0u, DE_NULL, 1u, &inputBufferBarrier, 0u, DE_NULL);
527 const VkImageSubresourceRange fullImageSubresourceRange = makeImageSubresourceRange(VK_IMAGE_ASPECT_COLOR_BIT, 0u, imageSparseInfo.mipLevels, 0u, imageSparseInfo.arrayLayers);
530 // Prepare sparse image for data transfer operation
531 const VkImageMemoryBarrier imageSparseTransferDstBarrier = makeImageMemoryBarrier
534 VK_ACCESS_TRANSFER_WRITE_BIT,
535 VK_IMAGE_LAYOUT_UNDEFINED,
536 VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL,
537 sparseQueue.queueFamilyIndex != extractQueue.queueFamilyIndex ? sparseQueue.queueFamilyIndex : VK_QUEUE_FAMILY_IGNORED,
538 sparseQueue.queueFamilyIndex != extractQueue.queueFamilyIndex ? extractQueue.queueFamilyIndex : VK_QUEUE_FAMILY_IGNORED,
540 fullImageSubresourceRange
543 deviceInterface.cmdPipelineBarrier(*commandBuffer, VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT, 0u, 0u, DE_NULL, 0u, DE_NULL, 1u, &imageSparseTransferDstBarrier);
546 // Copy reference data from input buffer to sparse image
547 deviceInterface.cmdCopyBufferToImage(*commandBuffer, *inputBuffer, *imageSparse, VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, static_cast<deUint32>(bufferImageSparseCopy.size()), &bufferImageSparseCopy[0]);
549 recordCommands(*commandBuffer, imageSparseInfo, *imageSparse, *imageTexels, *imageResidency);
551 const VkBufferCreateInfo bufferTexelsCreateInfo = makeBufferCreateInfo(imageSparseSizeInBytes, VK_BUFFER_USAGE_TRANSFER_DST_BIT);
552 const Unique<VkBuffer> bufferTexels (createBuffer(deviceInterface, getDevice(), &bufferTexelsCreateInfo));
553 const de::UniquePtr<Allocation> bufferTexelsAlloc (bindBuffer(deviceInterface, getDevice(), getAllocator(), *bufferTexels, MemoryRequirement::HostVisible));
555 // Copy data from texels image to buffer
556 deviceInterface.cmdCopyImageToBuffer(*commandBuffer, *imageTexels, VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL, *bufferTexels, static_cast<deUint32>(bufferImageSparseCopy.size()), &bufferImageSparseCopy[0]);
558 const deUint32 imageResidencySizeInBytes = getImageSizeInBytes(imageSparseInfo.extent, imageSparseInfo.arrayLayers, m_residencyFormat, imageSparseInfo.mipLevels, BUFFER_IMAGE_COPY_OFFSET_GRANULARITY);
560 const VkBufferCreateInfo bufferResidencyCreateInfo = makeBufferCreateInfo(imageResidencySizeInBytes, VK_BUFFER_USAGE_TRANSFER_DST_BIT);
561 const Unique<VkBuffer> bufferResidency (createBuffer(deviceInterface, getDevice(), &bufferResidencyCreateInfo));
562 const de::UniquePtr<Allocation> bufferResidencyAlloc (bindBuffer(deviceInterface, getDevice(), getAllocator(), *bufferResidency, MemoryRequirement::HostVisible));
564 // Copy data from residency image to buffer
565 std::vector <VkBufferImageCopy> bufferImageResidencyCopy(imageSparseInfo.mipLevels);
568 deUint32 bufferOffset = 0u;
569 for (deUint32 mipLevelNdx = 0u; mipLevelNdx < imageSparseInfo.mipLevels; ++mipLevelNdx)
571 bufferImageResidencyCopy[mipLevelNdx] = makeBufferImageCopy(mipLevelExtents(imageSparseInfo.extent, mipLevelNdx), imageSparseInfo.arrayLayers, mipLevelNdx, static_cast<VkDeviceSize>(bufferOffset));
572 bufferOffset += getImageMipLevelSizeInBytes(imageSparseInfo.extent, imageSparseInfo.arrayLayers, m_residencyFormat, mipLevelNdx, BUFFER_IMAGE_COPY_OFFSET_GRANULARITY);
576 deviceInterface.cmdCopyImageToBuffer(*commandBuffer, *imageResidency, VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL, *bufferResidency, static_cast<deUint32>(bufferImageResidencyCopy.size()), &bufferImageResidencyCopy[0]);
579 VkBufferMemoryBarrier bufferOutputHostReadBarriers[2];
581 bufferOutputHostReadBarriers[0] = makeBufferMemoryBarrier
583 VK_ACCESS_TRANSFER_WRITE_BIT,
584 VK_ACCESS_HOST_READ_BIT,
587 imageSparseSizeInBytes
590 bufferOutputHostReadBarriers[1] = makeBufferMemoryBarrier
592 VK_ACCESS_TRANSFER_WRITE_BIT,
593 VK_ACCESS_HOST_READ_BIT,
596 imageResidencySizeInBytes
599 deviceInterface.cmdPipelineBarrier(*commandBuffer, VK_PIPELINE_STAGE_TRANSFER_BIT, VK_PIPELINE_STAGE_HOST_BIT, 0u, 0u, DE_NULL, 2u, bufferOutputHostReadBarriers, 0u, DE_NULL);
602 // End recording commands
603 endCommandBuffer(deviceInterface, *commandBuffer);
605 const VkPipelineStageFlags stageBits[] = { VK_PIPELINE_STAGE_TRANSFER_BIT };
607 // Submit commands for execution and wait for completion
608 submitCommandsAndWait(deviceInterface, getDevice(), extractQueue.queueHandle, *commandBuffer, 1u, &memoryBindSemaphore.get(), stageBits);
610 // Wait for sparse queue to become idle
611 deviceInterface.queueWaitIdle(sparseQueue.queueHandle);
613 // Retrieve data from residency buffer to host memory
614 invalidateMappedMemoryRange(deviceInterface, getDevice(), bufferResidencyAlloc->getMemory(), bufferResidencyAlloc->getOffset(), imageResidencySizeInBytes);
616 const deUint32* bufferResidencyData = static_cast<const deUint32*>(bufferResidencyAlloc->getHostPtr());
618 deUint32 pixelOffsetNotAligned = 0u;
619 for (deUint32 mipmapNdx = 0; mipmapNdx < imageSparseInfo.mipLevels; ++mipmapNdx)
621 const deUint32 mipLevelSizeInBytes = getImageMipLevelSizeInBytes(imageSparseInfo.extent, imageSparseInfo.arrayLayers, m_residencyFormat, mipmapNdx);
622 const deUint32 pixelOffsetAligned = static_cast<deUint32>(bufferImageResidencyCopy[mipmapNdx].bufferOffset) / tcu::getPixelSize(m_residencyFormat);
624 if (deMemCmp(&bufferResidencyData[pixelOffsetAligned], &residencyReferenceData[pixelOffsetNotAligned], mipLevelSizeInBytes) != 0)
625 return tcu::TestStatus::fail("Failed");
627 pixelOffsetNotAligned += mipLevelSizeInBytes / tcu::getPixelSize(m_residencyFormat);
630 // Retrieve data from texels buffer to host memory
631 invalidateMappedMemoryRange(deviceInterface, getDevice(), bufferTexelsAlloc->getMemory(), bufferTexelsAlloc->getOffset(), imageSparseSizeInBytes);
633 const deUint8* bufferTexelsData = static_cast<const deUint8*>(bufferTexelsAlloc->getHostPtr());
635 for (deUint32 mipmapNdx = 0; mipmapNdx < imageSparseInfo.mipLevels; ++mipmapNdx)
637 const deUint32 mipLevelSizeInBytes = getImageMipLevelSizeInBytes(imageSparseInfo.extent, imageSparseInfo.arrayLayers, m_format, mipmapNdx);
638 const deUint32 bufferOffset = static_cast<deUint32>(bufferImageSparseCopy[mipmapNdx].bufferOffset);
640 if (mipmapNdx < aspectRequirements.imageMipTailFirstLod)
642 if (mipmapNdx % MEMORY_BLOCK_TYPE_COUNT == MEMORY_BLOCK_BOUND)
644 if (deMemCmp(&bufferTexelsData[bufferOffset], &referenceData[bufferOffset], mipLevelSizeInBytes) != 0)
645 return tcu::TestStatus::fail("Failed");
647 else if (getPhysicalDeviceProperties(instance, physicalDevice).sparseProperties.residencyNonResidentStrict)
649 std::vector<deUint8> zeroData;
650 zeroData.assign(mipLevelSizeInBytes, 0u);
652 if (deMemCmp(&bufferTexelsData[bufferOffset], &zeroData[0], mipLevelSizeInBytes) != 0)
653 return tcu::TestStatus::fail("Failed");
658 if (deMemCmp(&bufferTexelsData[bufferOffset], &referenceData[bufferOffset], mipLevelSizeInBytes) != 0)
659 return tcu::TestStatus::fail("Failed");
663 return tcu::TestStatus::pass("Passed");