vulkan/sparse_resources/vktSparseResourcesImageSparseResidency.cpp

/*------------------------------------------------------------------------
 * Vulkan Conformance Tests
 * ------------------------
 *
 * Copyright (c) 2016 The Khronos Group Inc.
 *
 * Licensed under the Apache License, Version 2.0 (the "License");
 * you may not use this file except in compliance with the License.
 * You may obtain a copy of the License at
 *
 *      http://www.apache.org/licenses/LICENSE-2.0
 *
 * Unless required by applicable law or agreed to in writing, software
 * distributed under the License is distributed on an "AS IS" BASIS,
 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 * See the License for the specific language governing permissions and
 * limitations under the License.
 *
 *//*!
 * \file  vktSparseResourcesImageSparseResidency.cpp
 * \brief Sparse partially resident images tests
 *//*--------------------------------------------------------------------*/

#include "vktSparseResourcesBufferSparseBinding.hpp"
#include "vktSparseResourcesTestsUtil.hpp"
#include "vktSparseResourcesBase.hpp"
#include "vktTestCaseUtil.hpp"

#include "vkDefs.hpp"
#include "vkRef.hpp"
#include "vkRefUtil.hpp"
#include "vkPlatform.hpp"
#include "vkPrograms.hpp"
#include "vkMemUtil.hpp"
#include "vkBuilderUtil.hpp"
#include "vkImageUtil.hpp"
#include "vkQueryUtil.hpp"
#include "vkTypeUtil.hpp"

#include "deUniquePtr.hpp"
#include "deStringUtil.hpp"

#include <string>
#include <vector>

using namespace vk;

namespace vkt
{
namespace sparse
{
namespace
{

const std::string getCoordStr  (const ImageType		imageType,
								const std::string&	x,
								const std::string&	y,
								const std::string&	z)
{
	switch (imageType)
	{
		case IMAGE_TYPE_1D:
		case IMAGE_TYPE_BUFFER:
			return x;

		case IMAGE_TYPE_1D_ARRAY:
		case IMAGE_TYPE_2D:
			return "ivec2(" + x + "," + y + ")";

		case IMAGE_TYPE_2D_ARRAY:
		case IMAGE_TYPE_3D:
		case IMAGE_TYPE_CUBE:
		case IMAGE_TYPE_CUBE_ARRAY:
			return "ivec3(" + x + "," + y + "," + z + ")";

		default:
			DE_ASSERT(false);
			return "";
	}
}

tcu::UVec3 alignedDivide (const VkExtent3D& extent, const VkExtent3D& divisor)
{
	tcu::UVec3 result;

	result.x() = extent.width  / divisor.width  + ((extent.width  % divisor.width)  ? 1u : 0u);
	result.y() = extent.height / divisor.height + ((extent.height % divisor.height) ? 1u : 0u);
	result.z() = extent.depth  / divisor.depth  + ((extent.depth  % divisor.depth)  ? 1u : 0u);

	return result;
}

tcu::UVec3 computeWorkGroupSize (const tcu::UVec3& gridSize)
{
	const deUint32		maxComputeWorkGroupInvocations	= 128u;
	const tcu::UVec3	maxComputeWorkGroupSize			= tcu::UVec3(128u, 128u, 64u);

	const deUint32 xWorkGroupSize = std::min(std::min(gridSize.x(), maxComputeWorkGroupSize.x()), maxComputeWorkGroupInvocations);
	const deUint32 yWorkGroupSize = std::min(std::min(gridSize.y(), maxComputeWorkGroupSize.y()), maxComputeWorkGroupInvocations /  xWorkGroupSize);
	const deUint32 zWorkGroupSize = std::min(std::min(gridSize.z(), maxComputeWorkGroupSize.z()), maxComputeWorkGroupInvocations / (xWorkGroupSize*yWorkGroupSize));

	return tcu::UVec3(xWorkGroupSize, yWorkGroupSize, zWorkGroupSize);
}

class ImageSparseResidencyCase : public TestCase
{
public:
					ImageSparseResidencyCase	(tcu::TestContext&			testCtx,
												 const std::string&			name,
												 const std::string&			description,
												 const ImageType			imageType,
												 const tcu::UVec3&			imageSize,
												 const tcu::TextureFormat&	format,
												 const glu::GLSLVersion		glslVersion,
												 const bool					useDeviceGroups);

	void			initPrograms				(SourceCollections&			sourceCollections) const;
	TestInstance*	createInstance				(Context&					context) const;

private:
	const bool					m_useDeviceGroups;
	const ImageType				m_imageType;
	const tcu::UVec3			m_imageSize;
	const tcu::TextureFormat	m_format;
	const glu::GLSLVersion		m_glslVersion;
};

ImageSparseResidencyCase::ImageSparseResidencyCase (tcu::TestContext&			testCtx,
													const std::string&			name,
													const std::string&			description,
													const ImageType				imageType,
													const tcu::UVec3&			imageSize,
													const tcu::TextureFormat&	format,
													const glu::GLSLVersion		glslVersion,
													const bool					useDeviceGroups)
	: TestCase				(testCtx, name, description)
	, m_useDeviceGroups		(useDeviceGroups)
	, m_imageType			(imageType)
	, m_imageSize			(imageSize)
	, m_format				(format)
	, m_glslVersion			(glslVersion)
{
}

void ImageSparseResidencyCase::initPrograms (SourceCollections&	sourceCollections) const
{
	// Create compute program
	const char* const versionDecl			= glu::getGLSLVersionDeclaration(m_glslVersion);
	const std::string imageTypeStr			= getShaderImageType(m_format, m_imageType);
	const std::string formatQualifierStr	= getShaderImageFormatQualifier(m_format);
	const std::string formatDataStr			= getShaderImageDataType(m_format);
	const tcu::UVec3  gridSize				= getShaderGridSize(m_imageType, m_imageSize);
	const tcu::UVec3  workGroupSize			= computeWorkGroupSize(gridSize);

	std::ostringstream src;
	src << versionDecl << "\n"
		<< "layout (local_size_x = " << workGroupSize.x() << ", local_size_y = " << workGroupSize.y() << ", local_size_z = " << workGroupSize.z() << ") in; \n"
		<< "layout (binding = 0, " << formatQualifierStr << ") writeonly uniform highp " << imageTypeStr << " u_image;\n"
		<< "void main (void)\n"
		<< "{\n"
		<< "	if( gl_GlobalInvocationID.x < " << gridSize.x() << " ) \n"
		<< "	if( gl_GlobalInvocationID.y < " << gridSize.y() << " ) \n"
		<< "	if( gl_GlobalInvocationID.z < " << gridSize.z() << " ) \n"
		<< "	{\n"
		<< "		imageStore(u_image, " << getCoordStr(m_imageType, "gl_GlobalInvocationID.x", "gl_GlobalInvocationID.y", "gl_GlobalInvocationID.z") << ","
		<< formatDataStr << "( int(gl_GlobalInvocationID.x) % 127, int(gl_GlobalInvocationID.y) % 127, int(gl_GlobalInvocationID.z) % 127, 1));\n"
		<< "	}\n"
		<< "}\n";

	sourceCollections.glslSources.add("comp") << glu::ComputeSource(src.str());
}

class ImageSparseResidencyInstance : public SparseResourcesBaseInstance
{
public:
					ImageSparseResidencyInstance(Context&									 context,
												 const ImageType							 imageType,
												 const tcu::UVec3&							 imageSize,
												 const tcu::TextureFormat&					 format,
												 const bool									 useDeviceGroups);


	tcu::TestStatus	iterate						(void);

private:
	const bool					m_useDeviceGroups;
	const ImageType				m_imageType;
	const tcu::UVec3			m_imageSize;
	const tcu::TextureFormat	m_format;
};

ImageSparseResidencyInstance::ImageSparseResidencyInstance (Context&					context,
															const ImageType				imageType,
															const tcu::UVec3&			imageSize,
															const tcu::TextureFormat&	format,
															const bool					useDeviceGroups)
	: SparseResourcesBaseInstance	(context, useDeviceGroups)
	, m_useDeviceGroups				(useDeviceGroups)
	, m_imageType					(imageType)
	, m_imageSize					(imageSize)
	, m_format						(format)
{
}

tcu::TestStatus ImageSparseResidencyInstance::iterate (void)
{
	const InstanceInterface&			instance = m_context.getInstanceInterface();

	{
		// Create logical device supporting both sparse and compute queues
		QueueRequirementsVec queueRequirements;
		queueRequirements.push_back(QueueRequirements(VK_QUEUE_SPARSE_BINDING_BIT, 1u));
		queueRequirements.push_back(QueueRequirements(VK_QUEUE_COMPUTE_BIT, 1u));

		createDeviceSupportingQueues(queueRequirements);
	}

	VkImageCreateInfo					imageCreateInfo;
	VkSparseImageMemoryRequirements		aspectRequirements;
	VkExtent3D							imageGranularity;
	std::vector<DeviceMemorySp>			deviceMemUniquePtrVec;

	const DeviceInterface&	deviceInterface	= getDeviceInterface();
	const Queue&			sparseQueue		= getQueue(VK_QUEUE_SPARSE_BINDING_BIT, 0);
	const Queue&			computeQueue	= getQueue(VK_QUEUE_COMPUTE_BIT, 0);

	// Go through all physical devices
	for (deUint32 physDevID = 0; physDevID < m_numPhysicalDevices; physDevID++)
	{
		const deUint32						firstDeviceID				= physDevID;
		const deUint32						secondDeviceID				= (firstDeviceID + 1) % m_numPhysicalDevices;

		const VkPhysicalDevice				physicalDevice				= getPhysicalDevice(firstDeviceID);
		const VkPhysicalDeviceProperties	physicalDeviceProperties	= getPhysicalDeviceProperties(instance, physicalDevice);

		// Check if image size does not exceed device limits
		if (!isImageSizeSupported(instance, physicalDevice, m_imageType, m_imageSize))
			TCU_THROW(NotSupportedError, "Image size not supported for device");

		// Check if device supports sparse operations for image type
		if (!checkSparseSupportForImageType(instance, physicalDevice, m_imageType))
			TCU_THROW(NotSupportedError, "Sparse residency for image type is not supported");

		imageCreateInfo.sType					= VK_STRUCTURE_TYPE_IMAGE_CREATE_INFO;
		imageCreateInfo.pNext					= DE_NULL;
		imageCreateInfo.flags					= VK_IMAGE_CREATE_SPARSE_RESIDENCY_BIT | VK_IMAGE_CREATE_SPARSE_BINDING_BIT;
		imageCreateInfo.imageType				= mapImageType(m_imageType);
		imageCreateInfo.format					= mapTextureFormat(m_format);
		imageCreateInfo.extent					= makeExtent3D(getLayerSize(m_imageType, m_imageSize));
		imageCreateInfo.mipLevels				= 1u;
		imageCreateInfo.arrayLayers				= getNumLayers(m_imageType, m_imageSize);
		imageCreateInfo.samples					= VK_SAMPLE_COUNT_1_BIT;
		imageCreateInfo.tiling					= VK_IMAGE_TILING_OPTIMAL;
		imageCreateInfo.initialLayout			= VK_IMAGE_LAYOUT_UNDEFINED;
		imageCreateInfo.usage					= VK_IMAGE_USAGE_TRANSFER_SRC_BIT |
												  VK_IMAGE_USAGE_STORAGE_BIT;
		imageCreateInfo.sharingMode				= VK_SHARING_MODE_EXCLUSIVE;
		imageCreateInfo.queueFamilyIndexCount	= 0u;
		imageCreateInfo.pQueueFamilyIndices		= DE_NULL;

		if (m_imageType == IMAGE_TYPE_CUBE || m_imageType == IMAGE_TYPE_CUBE_ARRAY)
		{
			imageCreateInfo.flags |= VK_IMAGE_CREATE_CUBE_COMPATIBLE_BIT;
		}

		// Check if device supports sparse operations for image format
		if (!checkSparseSupportForImageFormat(instance, physicalDevice, imageCreateInfo))
			TCU_THROW(NotSupportedError, "The image format does not support sparse operations");

		// Create sparse image
		const Unique<VkImage> sparseImage(createImage(deviceInterface, getDevice(), &imageCreateInfo));

		// Create sparse image memory bind semaphore
		const Unique<VkSemaphore> imageMemoryBindSemaphore(createSemaphore(deviceInterface, getDevice()));

		{
			// Get image general memory requirements
			const VkMemoryRequirements imageMemoryRequirements = getImageMemoryRequirements(deviceInterface, getDevice(), *sparseImage);

			if (imageMemoryRequirements.size > physicalDeviceProperties.limits.sparseAddressSpaceSize)
				TCU_THROW(NotSupportedError, "Required memory size for sparse resource exceeds device limits");

			DE_ASSERT((imageMemoryRequirements.size % imageMemoryRequirements.alignment) == 0);

			// Get sparse image sparse memory requirements
			const std::vector<VkSparseImageMemoryRequirements> sparseMemoryRequirements = getImageSparseMemoryRequirements(deviceInterface, getDevice(), *sparseImage);

			DE_ASSERT(sparseMemoryRequirements.size() != 0);

			const deUint32 colorAspectIndex		= getSparseAspectRequirementsIndex(sparseMemoryRequirements, VK_IMAGE_ASPECT_COLOR_BIT);
			const deUint32 metadataAspectIndex	= getSparseAspectRequirementsIndex(sparseMemoryRequirements, VK_IMAGE_ASPECT_METADATA_BIT);

			if (colorAspectIndex == NO_MATCH_FOUND)
				TCU_THROW(NotSupportedError, "Not supported image aspect - the test supports currently only VK_IMAGE_ASPECT_COLOR_BIT");

			aspectRequirements	= sparseMemoryRequirements[colorAspectIndex];
			imageGranularity	= aspectRequirements.formatProperties.imageGranularity;

			const VkImageAspectFlags aspectMask = aspectRequirements.formatProperties.aspectMask;

			DE_ASSERT((aspectRequirements.imageMipTailSize % imageMemoryRequirements.alignment) == 0);

			std::vector<VkSparseImageMemoryBind> imageResidencyMemoryBinds;
			std::vector<VkSparseMemoryBind>		 imageMipTailMemoryBinds;

			const deUint32						 memoryType = findMatchingMemoryType(instance, physicalDevice, imageMemoryRequirements, MemoryRequirement::Any);

			if (memoryType == NO_MATCH_FOUND)
				return tcu::TestStatus::fail("No matching memory type found");

			// Bind device memory for each aspect
			for (deUint32 layerNdx = 0; layerNdx < imageCreateInfo.arrayLayers; ++layerNdx)
			{
				for (deUint32 mipLevelNdx = 0; mipLevelNdx < aspectRequirements.imageMipTailFirstLod; ++mipLevelNdx)
				{
					const VkImageSubresource subresource		= { aspectMask, mipLevelNdx, layerNdx };
					const VkExtent3D		 mipExtent			= mipLevelExtents(imageCreateInfo.extent, mipLevelNdx);
					const tcu::UVec3		 numSparseBinds		= alignedDivide(mipExtent, imageGranularity);
					const tcu::UVec3		 lastBlockExtent	= tcu::UVec3(mipExtent.width  % imageGranularity.width  ? mipExtent.width   % imageGranularity.width  : imageGranularity.width,
																			 mipExtent.height % imageGranularity.height ? mipExtent.height  % imageGranularity.height : imageGranularity.height,
																			 mipExtent.depth  % imageGranularity.depth  ? mipExtent.depth   % imageGranularity.depth  : imageGranularity.depth);
					for (deUint32 z = 0; z < numSparseBinds.z(); ++z)
					for (deUint32 y = 0; y < numSparseBinds.y(); ++y)
					for (deUint32 x = 0; x < numSparseBinds.x(); ++x)
					{
						const deUint32 linearIndex = x + y*numSparseBinds.x() + z*numSparseBinds.x()*numSparseBinds.y() + layerNdx*numSparseBinds.x()*numSparseBinds.y()*numSparseBinds.z();

						if (linearIndex % 2u == 1u)
						{
							continue;
						}

						VkOffset3D offset;
						offset.x = x*imageGranularity.width;
						offset.y = y*imageGranularity.height;
						offset.z = z*imageGranularity.depth;

						VkExtent3D extent;
						extent.width  = (x == numSparseBinds.x() - 1) ? lastBlockExtent.x() : imageGranularity.width;
						extent.height = (y == numSparseBinds.y() - 1) ? lastBlockExtent.y() : imageGranularity.height;
						extent.depth  = (z == numSparseBinds.z() - 1) ? lastBlockExtent.z() : imageGranularity.depth;

						const VkSparseImageMemoryBind imageMemoryBind = makeSparseImageMemoryBind(deviceInterface, getDevice(),
							imageMemoryRequirements.alignment, memoryType, subresource, offset, extent);

						deviceMemUniquePtrVec.push_back(makeVkSharedPtr(Move<VkDeviceMemory>(check<VkDeviceMemory>(imageMemoryBind.memory), Deleter<VkDeviceMemory>(deviceInterface, getDevice(), DE_NULL))));

						imageResidencyMemoryBinds.push_back(imageMemoryBind);
					}
				}

				if (!(aspectRequirements.formatProperties.flags & VK_SPARSE_IMAGE_FORMAT_SINGLE_MIPTAIL_BIT) && aspectRequirements.imageMipTailFirstLod < imageCreateInfo.mipLevels)
				{
					const VkSparseMemoryBind imageMipTailMemoryBind = makeSparseMemoryBind(deviceInterface, getDevice(),
						aspectRequirements.imageMipTailSize, memoryType, aspectRequirements.imageMipTailOffset + layerNdx * aspectRequirements.imageMipTailStride);

					deviceMemUniquePtrVec.push_back(makeVkSharedPtr(Move<VkDeviceMemory>(check<VkDeviceMemory>(imageMipTailMemoryBind.memory), Deleter<VkDeviceMemory>(deviceInterface, getDevice(), DE_NULL))));

					imageMipTailMemoryBinds.push_back(imageMipTailMemoryBind);
				}

				// Metadata
				if (metadataAspectIndex != NO_MATCH_FOUND)
				{
					const VkSparseImageMemoryRequirements metadataAspectRequirements = sparseMemoryRequirements[metadataAspectIndex];

					if (!(metadataAspectRequirements.formatProperties.flags & VK_SPARSE_IMAGE_FORMAT_SINGLE_MIPTAIL_BIT))
					{
						const VkSparseMemoryBind imageMipTailMemoryBind = makeSparseMemoryBind(deviceInterface, getDevice(),
							metadataAspectRequirements.imageMipTailSize, memoryType,
							metadataAspectRequirements.imageMipTailOffset + layerNdx * metadataAspectRequirements.imageMipTailStride,
							VK_SPARSE_MEMORY_BIND_METADATA_BIT);

						deviceMemUniquePtrVec.push_back(makeVkSharedPtr(Move<VkDeviceMemory>(check<VkDeviceMemory>(imageMipTailMemoryBind.memory), Deleter<VkDeviceMemory>(deviceInterface, getDevice(), DE_NULL))));

						imageMipTailMemoryBinds.push_back(imageMipTailMemoryBind);
					}
				}
			}

			if ((aspectRequirements.formatProperties.flags & VK_SPARSE_IMAGE_FORMAT_SINGLE_MIPTAIL_BIT) && aspectRequirements.imageMipTailFirstLod < imageCreateInfo.mipLevels)
			{
				const VkSparseMemoryBind imageMipTailMemoryBind = makeSparseMemoryBind(deviceInterface, getDevice(),
					aspectRequirements.imageMipTailSize, memoryType, aspectRequirements.imageMipTailOffset);

				deviceMemUniquePtrVec.push_back(makeVkSharedPtr(Move<VkDeviceMemory>(check<VkDeviceMemory>(imageMipTailMemoryBind.memory), Deleter<VkDeviceMemory>(deviceInterface, getDevice(), DE_NULL))));

				imageMipTailMemoryBinds.push_back(imageMipTailMemoryBind);
			}

			// Metadata
			if (metadataAspectIndex != NO_MATCH_FOUND)
			{
				const VkSparseImageMemoryRequirements metadataAspectRequirements = sparseMemoryRequirements[metadataAspectIndex];

				if ((metadataAspectRequirements.formatProperties.flags & VK_SPARSE_IMAGE_FORMAT_SINGLE_MIPTAIL_BIT))
				{
					const VkSparseMemoryBind imageMipTailMemoryBind = makeSparseMemoryBind(deviceInterface, getDevice(),
						metadataAspectRequirements.imageMipTailSize, memoryType, metadataAspectRequirements.imageMipTailOffset,
						VK_SPARSE_MEMORY_BIND_METADATA_BIT);

					deviceMemUniquePtrVec.push_back(makeVkSharedPtr(Move<VkDeviceMemory>(check<VkDeviceMemory>(imageMipTailMemoryBind.memory), Deleter<VkDeviceMemory>(deviceInterface, getDevice(), DE_NULL))));

					imageMipTailMemoryBinds.push_back(imageMipTailMemoryBind);
				}
			}

			const VkDeviceGroupBindSparseInfo devGroupBindSparseInfo =
			{
				VK_STRUCTURE_TYPE_DEVICE_GROUP_BIND_SPARSE_INFO_KHR,	//VkStructureType							sType;
				DE_NULL,												//const void*								pNext;
				firstDeviceID,											//deUint32									resourceDeviceIndex;
				secondDeviceID,											//deUint32									memoryDeviceIndex;
			};

			VkBindSparseInfo bindSparseInfo =
			{
				VK_STRUCTURE_TYPE_BIND_SPARSE_INFO,						//VkStructureType							sType;
				m_useDeviceGroups ? &devGroupBindSparseInfo : DE_NULL,	//const void*								pNext;
				0u,														//deUint32									waitSemaphoreCount;
				DE_NULL,												//const VkSemaphore*						pWaitSemaphores;
				0u,														//deUint32									bufferBindCount;
				DE_NULL,												//const VkSparseBufferMemoryBindInfo*		pBufferBinds;
				0u,														//deUint32									imageOpaqueBindCount;
				DE_NULL,												//const VkSparseImageOpaqueMemoryBindInfo*	pImageOpaqueBinds;
				0u,														//deUint32									imageBindCount;
				DE_NULL,												//const VkSparseImageMemoryBindInfo*		pImageBinds;
				1u,														//deUint32									signalSemaphoreCount;
				&imageMemoryBindSemaphore.get()							//const VkSemaphore*						pSignalSemaphores;
			};

			VkSparseImageMemoryBindInfo		  imageResidencyBindInfo;
			VkSparseImageOpaqueMemoryBindInfo imageMipTailBindInfo;

			if (imageResidencyMemoryBinds.size() > 0)
			{
				imageResidencyBindInfo.image		= *sparseImage;
				imageResidencyBindInfo.bindCount	= static_cast<deUint32>(imageResidencyMemoryBinds.size());
				imageResidencyBindInfo.pBinds		= &imageResidencyMemoryBinds[0];

				bindSparseInfo.imageBindCount		= 1u;
				bindSparseInfo.pImageBinds			= &imageResidencyBindInfo;
			}

			if (imageMipTailMemoryBinds.size() > 0)
			{
				imageMipTailBindInfo.image			= *sparseImage;
				imageMipTailBindInfo.bindCount		= static_cast<deUint32>(imageMipTailMemoryBinds.size());
				imageMipTailBindInfo.pBinds			= &imageMipTailMemoryBinds[0];

				bindSparseInfo.imageOpaqueBindCount = 1u;
				bindSparseInfo.pImageOpaqueBinds	= &imageMipTailBindInfo;
			}

			// Submit sparse bind commands for execution
			VK_CHECK(deviceInterface.queueBindSparse(sparseQueue.queueHandle, 1u, &bindSparseInfo, DE_NULL));
		}

		// Create command buffer for compute and transfer oparations
		const Unique<VkCommandPool>	  commandPool(makeCommandPool(deviceInterface, getDevice(), computeQueue.queueFamilyIndex));
		const Unique<VkCommandBuffer> commandBuffer(allocateCommandBuffer(deviceInterface, getDevice(), *commandPool, VK_COMMAND_BUFFER_LEVEL_PRIMARY));

		// Start recording commands
		beginCommandBuffer(deviceInterface, *commandBuffer);

		// Create descriptor set layout
		const Unique<VkDescriptorSetLayout> descriptorSetLayout(
			DescriptorSetLayoutBuilder()
			.addSingleBinding(VK_DESCRIPTOR_TYPE_STORAGE_IMAGE, VK_SHADER_STAGE_COMPUTE_BIT)
			.build(deviceInterface, getDevice()));

		// Create and bind compute pipeline
		const Unique<VkShaderModule>	shaderModule(createShaderModule(deviceInterface, getDevice(), m_context.getBinaryCollection().get("comp"), DE_NULL));
		const Unique<VkPipelineLayout>	pipelineLayout(makePipelineLayout(deviceInterface, getDevice(), *descriptorSetLayout));
		const Unique<VkPipeline>		computePipeline(makeComputePipeline(deviceInterface, getDevice(), *pipelineLayout, *shaderModule));

		deviceInterface.cmdBindPipeline(*commandBuffer, VK_PIPELINE_BIND_POINT_COMPUTE, *computePipeline);

		// Create and bind descriptor set
		const Unique<VkDescriptorPool> descriptorPool(
			DescriptorPoolBuilder()
			.addType(VK_DESCRIPTOR_TYPE_STORAGE_IMAGE, 1u)
			.build(deviceInterface, getDevice(), VK_DESCRIPTOR_POOL_CREATE_FREE_DESCRIPTOR_SET_BIT, 1u));

		const Unique<VkDescriptorSet>	descriptorSet(makeDescriptorSet(deviceInterface, getDevice(), *descriptorPool, *descriptorSetLayout));

		const VkImageSubresourceRange	subresourceRange = makeImageSubresourceRange(VK_IMAGE_ASPECT_COLOR_BIT, 0u, 1u, 0u, getNumLayers(m_imageType, m_imageSize));
		const Unique<VkImageView>		imageView(makeImageView(deviceInterface, getDevice(), *sparseImage, mapImageViewType(m_imageType), mapTextureFormat(m_format), subresourceRange));
		const VkDescriptorImageInfo		sparseImageInfo  = makeDescriptorImageInfo(DE_NULL, *imageView, VK_IMAGE_LAYOUT_GENERAL);

		DescriptorSetUpdateBuilder()
			.writeSingle(*descriptorSet, DescriptorSetUpdateBuilder::Location::binding(0u), VK_DESCRIPTOR_TYPE_STORAGE_IMAGE, &sparseImageInfo)
			.update(deviceInterface, getDevice());

		deviceInterface.cmdBindDescriptorSets(*commandBuffer, VK_PIPELINE_BIND_POINT_COMPUTE, *pipelineLayout, 0u, 1u, &descriptorSet.get(), 0u, DE_NULL);

		{
			const VkImageMemoryBarrier sparseImageLayoutChangeBarrier = makeImageMemoryBarrier
			(
				0u,
				VK_ACCESS_SHADER_WRITE_BIT,
				VK_IMAGE_LAYOUT_UNDEFINED,
				VK_IMAGE_LAYOUT_GENERAL,
				sparseQueue.queueFamilyIndex != computeQueue.queueFamilyIndex ? sparseQueue.queueFamilyIndex : VK_QUEUE_FAMILY_IGNORED,
				sparseQueue.queueFamilyIndex != computeQueue.queueFamilyIndex ? computeQueue.queueFamilyIndex : VK_QUEUE_FAMILY_IGNORED,
				*sparseImage,
				subresourceRange
			);

			deviceInterface.cmdPipelineBarrier(*commandBuffer, VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT, VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, 0u, 0u, DE_NULL, 0u, DE_NULL, 1u, &sparseImageLayoutChangeBarrier);
		}

		const tcu::UVec3  gridSize = getShaderGridSize(m_imageType, m_imageSize);

		{
			const tcu::UVec3  workGroupSize = computeWorkGroupSize(gridSize);

			const deUint32 xWorkGroupCount = gridSize.x() / workGroupSize.x() + (gridSize.x() % workGroupSize.x() ? 1u : 0u);
			const deUint32 yWorkGroupCount = gridSize.y() / workGroupSize.y() + (gridSize.y() % workGroupSize.y() ? 1u : 0u);
			const deUint32 zWorkGroupCount = gridSize.z() / workGroupSize.z() + (gridSize.z() % workGroupSize.z() ? 1u : 0u);

			const tcu::UVec3 maxComputeWorkGroupCount = tcu::UVec3(65535u, 65535u, 65535u);

			if (maxComputeWorkGroupCount.x() < xWorkGroupCount ||
				maxComputeWorkGroupCount.y() < yWorkGroupCount ||
				maxComputeWorkGroupCount.z() < zWorkGroupCount)
			{
				TCU_THROW(NotSupportedError, "Image size is not supported");
			}

			deviceInterface.cmdDispatch(*commandBuffer, xWorkGroupCount, yWorkGroupCount, zWorkGroupCount);
		}

		{
			const VkImageMemoryBarrier sparseImageTrasferBarrier = makeImageMemoryBarrier
			(
				VK_ACCESS_SHADER_WRITE_BIT,
				VK_ACCESS_TRANSFER_READ_BIT,
				VK_IMAGE_LAYOUT_GENERAL,
				VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL,
				*sparseImage,
				subresourceRange
			);

			deviceInterface.cmdPipelineBarrier(*commandBuffer, VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT, 0u, 0u, DE_NULL, 0u, DE_NULL, 1u, &sparseImageTrasferBarrier);
		}

		const deUint32					imageSizeInBytes		= getNumPixels(m_imageType, m_imageSize) * tcu::getPixelSize(m_format);
		const VkBufferCreateInfo		outputBufferCreateInfo	= makeBufferCreateInfo(imageSizeInBytes, VK_BUFFER_USAGE_TRANSFER_DST_BIT);
		const Unique<VkBuffer>			outputBuffer			(createBuffer(deviceInterface, getDevice(), &outputBufferCreateInfo));
		const de::UniquePtr<Allocation>	outputBufferAlloc		(bindBuffer(deviceInterface, getDevice(), getAllocator(), *outputBuffer, MemoryRequirement::HostVisible));

		{
			const VkBufferImageCopy bufferImageCopy = makeBufferImageCopy(imageCreateInfo.extent, imageCreateInfo.arrayLayers);

			deviceInterface.cmdCopyImageToBuffer(*commandBuffer, *sparseImage, VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL, *outputBuffer, 1u, &bufferImageCopy);
		}

		{
			const VkBufferMemoryBarrier outputBufferHostReadBarrier = makeBufferMemoryBarrier
			(
				VK_ACCESS_TRANSFER_WRITE_BIT,
				VK_ACCESS_HOST_READ_BIT,
				*outputBuffer,
				0u,
				imageSizeInBytes
			);

			deviceInterface.cmdPipelineBarrier(*commandBuffer, VK_PIPELINE_STAGE_TRANSFER_BIT, VK_PIPELINE_STAGE_HOST_BIT, 0u, 0u, DE_NULL, 1u, &outputBufferHostReadBarrier, 0u, DE_NULL);
		}

		// End recording commands
		endCommandBuffer(deviceInterface, *commandBuffer);

		// The stage at which execution is going to wait for finish of sparse binding operations
		const VkPipelineStageFlags stageBits[] = { VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT };

		// Submit commands for execution and wait for completion
		submitCommandsAndWait(deviceInterface, getDevice(), computeQueue.queueHandle, *commandBuffer, 1u, &imageMemoryBindSemaphore.get(), stageBits,
			0, DE_NULL, m_useDeviceGroups, firstDeviceID);

		// Retrieve data from buffer to host memory
		invalidateMappedMemoryRange(deviceInterface, getDevice(), outputBufferAlloc->getMemory(), outputBufferAlloc->getOffset(), imageSizeInBytes);

		const deUint8* outputData = static_cast<const deUint8*>(outputBufferAlloc->getHostPtr());
		const tcu::ConstPixelBufferAccess pixelBuffer = tcu::ConstPixelBufferAccess(m_format, gridSize.x(), gridSize.y(), gridSize.z(), outputData);

		// Wait for sparse queue to become idle
		//vsk fails:
		deviceInterface.queueWaitIdle(sparseQueue.queueHandle);

		// Validate results
		if( aspectRequirements.imageMipTailFirstLod > 0u )
		{
			const VkExtent3D		 mipExtent		 = mipLevelExtents(imageCreateInfo.extent, 0u);
			const tcu::UVec3		 numSparseBinds  = alignedDivide(mipExtent, imageGranularity);
			const tcu::UVec3		 lastBlockExtent = tcu::UVec3(	mipExtent.width  % imageGranularity.width  ? mipExtent.width  % imageGranularity.width  : imageGranularity.width,
																	mipExtent.height % imageGranularity.height ? mipExtent.height % imageGranularity.height : imageGranularity.height,
																	mipExtent.depth  % imageGranularity.depth  ? mipExtent.depth  % imageGranularity.depth  : imageGranularity.depth);

			for (deUint32 layerNdx = 0; layerNdx < imageCreateInfo.arrayLayers; ++layerNdx)
			{
				for (deUint32 z = 0; z < numSparseBinds.z(); ++z)
				for (deUint32 y = 0; y < numSparseBinds.y(); ++y)
				for (deUint32 x = 0; x < numSparseBinds.x(); ++x)
				{
					VkExtent3D offset;
					offset.width  = x*imageGranularity.width;
					offset.height = y*imageGranularity.height;
					offset.depth  = z*imageGranularity.depth + layerNdx*numSparseBinds.z()*imageGranularity.depth;

					VkExtent3D extent;
					extent.width  = (x == numSparseBinds.x() - 1) ? lastBlockExtent.x() : imageGranularity.width;
					extent.height = (y == numSparseBinds.y() - 1) ? lastBlockExtent.y() : imageGranularity.height;
					extent.depth  = (z == numSparseBinds.z() - 1) ? lastBlockExtent.z() : imageGranularity.depth;

					const deUint32 linearIndex = x + y*numSparseBinds.x() + z*numSparseBinds.x()*numSparseBinds.y() + layerNdx*numSparseBinds.x()*numSparseBinds.y()*numSparseBinds.z();

					if (linearIndex % 2u == 0u)
					{
						for (deUint32 offsetZ = offset.depth;  offsetZ < offset.depth  + extent.depth;  ++offsetZ)
						for (deUint32 offsetY = offset.height; offsetY < offset.height + extent.height; ++offsetY)
						for (deUint32 offsetX = offset.width;  offsetX < offset.width  + extent.width;  ++offsetX)
						{
							const tcu::UVec4 referenceValue = tcu::UVec4(offsetX % 127u, offsetY % 127u, offsetZ % 127u, 1u);
							const tcu::UVec4 outputValue	= pixelBuffer.getPixelUint(offsetX, offsetY, offsetZ);

							if (deMemCmp(&outputValue, &referenceValue, sizeof(deUint32) * getNumUsedChannels(m_format.order)) != 0)
								return tcu::TestStatus::fail("Failed");
						}
					}
					else if (physicalDeviceProperties.sparseProperties.residencyNonResidentStrict)
					{
						for (deUint32 offsetZ = offset.depth;  offsetZ < offset.depth  + extent.depth;  ++offsetZ)
						for (deUint32 offsetY = offset.height; offsetY < offset.height + extent.height; ++offsetY)
						for (deUint32 offsetX = offset.width;  offsetX < offset.width  + extent.width;  ++offsetX)
						{
							const tcu::UVec4 referenceValue = tcu::UVec4(0u, 0u, 0u, 0u);
							const tcu::UVec4 outputValue = pixelBuffer.getPixelUint(offsetX, offsetY, offsetZ);

							if (deMemCmp(&outputValue, &referenceValue, sizeof(deUint32) * getNumUsedChannels(m_format.order)) != 0)
								return tcu::TestStatus::fail("Failed");
						}
					}
				}
			}
		}
		else
		{
			const VkExtent3D mipExtent = mipLevelExtents(imageCreateInfo.extent, 0u);

			for (deUint32 offsetZ = 0u; offsetZ < mipExtent.depth * imageCreateInfo.arrayLayers; ++offsetZ)
			for (deUint32 offsetY = 0u; offsetY < mipExtent.height; ++offsetY)
			for (deUint32 offsetX = 0u; offsetX < mipExtent.width;  ++offsetX)
			{
				const tcu::UVec4 referenceValue = tcu::UVec4(offsetX % 127u, offsetY % 127u, offsetZ % 127u, 1u);
				const tcu::UVec4 outputValue	= pixelBuffer.getPixelUint(offsetX, offsetY, offsetZ);

				if (deMemCmp(&outputValue, &referenceValue, sizeof(deUint32) * getNumUsedChannels(m_format.order)) != 0)
					return tcu::TestStatus::fail("Failed");
			}
		}
	}

	return tcu::TestStatus::pass("Passed");
}

TestInstance* ImageSparseResidencyCase::createInstance (Context& context) const
{
	return new ImageSparseResidencyInstance(context, m_imageType, m_imageSize, m_format, m_useDeviceGroups);
}

} // anonymous ns

tcu::TestCaseGroup* createImageSparseResidencyTestsCommon (tcu::TestContext& testCtx, de::MovePtr<tcu::TestCaseGroup> testGroup, const bool useDeviceGroup = false)
{
	static const deUint32 sizeCountPerImageType = 3u;

	struct ImageParameters
	{
		ImageType	imageType;
		tcu::UVec3	imageSizes[sizeCountPerImageType];
	};

	static const ImageParameters imageParametersArray[] =
	{
		{ IMAGE_TYPE_2D,		 { tcu::UVec3(512u, 256u, 1u),  tcu::UVec3(1024u, 128u, 1u), tcu::UVec3(11u,  137u, 1u) } },
		{ IMAGE_TYPE_2D_ARRAY,	 { tcu::UVec3(512u, 256u, 6u),	tcu::UVec3(1024u, 128u, 8u), tcu::UVec3(11u,  137u, 3u) } },
		{ IMAGE_TYPE_CUBE,		 { tcu::UVec3(256u, 256u, 1u),	tcu::UVec3(128u,  128u, 1u), tcu::UVec3(137u, 137u, 1u) } },
		{ IMAGE_TYPE_CUBE_ARRAY, { tcu::UVec3(256u, 256u, 6u),	tcu::UVec3(128u,  128u, 8u), tcu::UVec3(137u, 137u, 3u) } },
		{ IMAGE_TYPE_3D,		 { tcu::UVec3(512u, 256u, 16u), tcu::UVec3(1024u, 128u, 8u), tcu::UVec3(11u,  137u, 3u) } }
	};

	static const tcu::TextureFormat formats[] =
	{
		tcu::TextureFormat(tcu::TextureFormat::R,	 tcu::TextureFormat::SIGNED_INT32),
		tcu::TextureFormat(tcu::TextureFormat::R,	 tcu::TextureFormat::SIGNED_INT16),
		tcu::TextureFormat(tcu::TextureFormat::R,	 tcu::TextureFormat::SIGNED_INT8),
		tcu::TextureFormat(tcu::TextureFormat::RG,	 tcu::TextureFormat::SIGNED_INT32),
		tcu::TextureFormat(tcu::TextureFormat::RG,   tcu::TextureFormat::SIGNED_INT16),
		tcu::TextureFormat(tcu::TextureFormat::RG,   tcu::TextureFormat::SIGNED_INT8),
		tcu::TextureFormat(tcu::TextureFormat::RGBA, tcu::TextureFormat::UNSIGNED_INT32),
		tcu::TextureFormat(tcu::TextureFormat::RGBA, tcu::TextureFormat::UNSIGNED_INT16),
		tcu::TextureFormat(tcu::TextureFormat::RGBA, tcu::TextureFormat::UNSIGNED_INT8)
	};

	for (deInt32 imageTypeNdx = 0; imageTypeNdx < DE_LENGTH_OF_ARRAY(imageParametersArray); ++imageTypeNdx)
	{
		const ImageType					imageType = imageParametersArray[imageTypeNdx].imageType;
		de::MovePtr<tcu::TestCaseGroup> imageTypeGroup(new tcu::TestCaseGroup(testCtx, getImageTypeName(imageType).c_str(), ""));

		for (deInt32 formatNdx = 0; formatNdx < DE_LENGTH_OF_ARRAY(formats); ++formatNdx)
		{
			const tcu::TextureFormat&		format = formats[formatNdx];
			de::MovePtr<tcu::TestCaseGroup> formatGroup(new tcu::TestCaseGroup(testCtx, getShaderImageFormatQualifier(format).c_str(), ""));

			for (deInt32 imageSizeNdx = 0; imageSizeNdx < DE_LENGTH_OF_ARRAY(imageParametersArray[imageTypeNdx].imageSizes); ++imageSizeNdx)
			{
				const tcu::UVec3 imageSize = imageParametersArray[imageTypeNdx].imageSizes[imageSizeNdx];

				std::ostringstream stream;
				stream << imageSize.x() << "_" << imageSize.y() << "_" << imageSize.z();

				formatGroup->addChild(new ImageSparseResidencyCase(testCtx, stream.str(), "", imageType, imageSize, format, glu::GLSL_VERSION_440, useDeviceGroup));
			}
			imageTypeGroup->addChild(formatGroup.release());
		}
		testGroup->addChild(imageTypeGroup.release());
	}

	return testGroup.release();
}

tcu::TestCaseGroup* createImageSparseResidencyTests (tcu::TestContext& testCtx)
{
	de::MovePtr<tcu::TestCaseGroup> testGroup(new tcu::TestCaseGroup(testCtx, "image_sparse_residency", "Buffer Sparse Residency"));
	return createImageSparseResidencyTestsCommon(testCtx, testGroup);
}

tcu::TestCaseGroup* createDeviceGroupImageSparseResidencyTests (tcu::TestContext& testCtx)
{
	de::MovePtr<tcu::TestCaseGroup> testGroup(new tcu::TestCaseGroup(testCtx, "device_group_image_sparse_residency", "Buffer Sparse Residency"));
	return createImageSparseResidencyTestsCommon(testCtx, testGroup, true);
}

} // sparse
} // vkt