external/vulkancts/modules/vulkan/api/vktApiCopiesAndBlittingTests.cpp

/*------------------------------------------------------------------------
 * Vulkan Conformance Tests
 * ------------------------
 *
 * Copyright (c) 2015-2016 The Khronos Group Inc.
 * Copyright (c) 2015-2016 Samsung Electronics Co., Ltd.
 *
 * 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
 * \brief Vulkan Copies And Blitting Tests
 *//*--------------------------------------------------------------------*/

#include "vktApiCopiesAndBlittingTests.hpp"

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

#include "tcuImageCompare.hpp"
#include "tcuTexture.hpp"
#include "tcuTextureUtil.hpp"
#include "tcuVectorType.hpp"
#include "tcuVectorUtil.hpp"
#include "tcuTestLog.hpp"
#include "tcuTexLookupVerifier.hpp"

#include "vkImageUtil.hpp"
#include "vkMemUtil.hpp"
#include "vkPrograms.hpp"
#include "vkQueryUtil.hpp"
#include "vkRefUtil.hpp"
#include "vktTestCase.hpp"
#include "vktTestCaseUtil.hpp"
#include "vktTestGroupUtil.hpp"
#include "vkTypeUtil.hpp"
#include "vkCmdUtil.hpp"
#include "vkObjUtil.hpp"

#include <set>

namespace vkt
{

namespace api
{

namespace
{
enum MirrorMode
{
	MIRROR_MODE_NONE = 0,
	MIRROR_MODE_X = (1<<0),
	MIRROR_MODE_Y = (1<<1),
	MIRROR_MODE_XY = MIRROR_MODE_X | MIRROR_MODE_Y,

	MIRROR_MODE_LAST
};

enum AllocationKind
{
	ALLOCATION_KIND_SUBALLOCATED,
	ALLOCATION_KIND_DEDICATED,
};

template <typename Type>
class BinaryCompare
{
public:
	bool operator() (const Type& a, const Type& b) const
	{
		return deMemCmp(&a, &b, sizeof(Type)) < 0;
	}
};

typedef std::set<vk::VkFormat, BinaryCompare<vk::VkFormat> >	FormatSet;

FormatSet dedicatedAllocationImageToImageFormatsToTestSet;
FormatSet dedicatedAllocationBlittingFormatsToTestSet;

using namespace vk;

VkImageAspectFlags getAspectFlags (tcu::TextureFormat format)
{
	VkImageAspectFlags	aspectFlag	= 0;
	aspectFlag |= (tcu::hasDepthComponent(format.order)? VK_IMAGE_ASPECT_DEPTH_BIT : 0);
	aspectFlag |= (tcu::hasStencilComponent(format.order)? VK_IMAGE_ASPECT_STENCIL_BIT : 0);

	if (!aspectFlag)
		aspectFlag = VK_IMAGE_ASPECT_COLOR_BIT;

	return aspectFlag;
}

VkImageAspectFlags getAspectFlags (VkFormat format)
{
	if (isCompressedFormat(format))
		return VK_IMAGE_ASPECT_COLOR_BIT;
	else
		return getAspectFlags(mapVkFormat(format));
}

tcu::TextureFormat getSizeCompatibleTcuTextureFormat (VkFormat format)
{
	if (isCompressedFormat(format))
		return (getBlockSizeInBytes(format) == 8) ? mapVkFormat(VK_FORMAT_R16G16B16A16_UINT) : mapVkFormat(VK_FORMAT_R32G32B32A32_UINT);
	else
		return mapVkFormat(format);
}

// This is effectively same as vk::isFloatFormat(mapTextureFormat(format))
// except that it supports some formats that are not mappable to VkFormat.
// When we are checking combined depth and stencil formats, each aspect is
// checked separately, and in some cases we construct PBA with a format that
// is not mappable to VkFormat.
bool isFloatFormat (tcu::TextureFormat format)
{
	return tcu::getTextureChannelClass(format.type) == tcu::TEXTURECHANNELCLASS_FLOATING_POINT;
}

union CopyRegion
{
	VkBufferCopy		bufferCopy;
	VkImageCopy			imageCopy;
	VkBufferImageCopy	bufferImageCopy;
	VkImageBlit			imageBlit;
	VkImageResolve		imageResolve;
};

struct ImageParms
{
	VkImageType		imageType;
	VkFormat		format;
	VkExtent3D		extent;
	VkImageTiling	tiling;
	VkImageLayout	operationLayout;
};

struct TestParams
{
	union Data
	{
		struct Buffer
		{
			VkDeviceSize	size;
		} buffer;

		ImageParms	image;
	} src, dst;

	std::vector<CopyRegion>	regions;

	union
	{
		VkFilter				filter;
		VkSampleCountFlagBits	samples;
	};

	AllocationKind	allocationKind;
	deUint32		mipLevels;
	deBool			singleCommand;
	deUint32		barrierCount;
	deBool			separateDepthStencilLayouts;

	TestParams (void)
	{
		mipLevels					= 1u;
		singleCommand				= DE_TRUE;
		barrierCount				= 1u;
		separateDepthStencilLayouts	= DE_FALSE;
	}
};

de::MovePtr<Allocation> allocateBuffer (const InstanceInterface&	vki,
										const DeviceInterface&		vkd,
										const VkPhysicalDevice&		physDevice,
										const VkDevice				device,
										const VkBuffer&				buffer,
										const MemoryRequirement		requirement,
										Allocator&					allocator,
										AllocationKind				allocationKind)
{
	switch (allocationKind)
	{
		case ALLOCATION_KIND_SUBALLOCATED:
		{
			const VkMemoryRequirements memoryRequirements = getBufferMemoryRequirements(vkd, device, buffer);

			return allocator.allocate(memoryRequirements, requirement);
		}

		case ALLOCATION_KIND_DEDICATED:
		{
			return allocateDedicated(vki, vkd, physDevice, device, buffer, requirement);
		}

		default:
		{
			TCU_THROW(InternalError, "Invalid allocation kind");
		}
	}
}

de::MovePtr<Allocation> allocateImage (const InstanceInterface&		vki,
									   const DeviceInterface&		vkd,
									   const VkPhysicalDevice&		physDevice,
									   const VkDevice				device,
									   const VkImage&				image,
									   const MemoryRequirement		requirement,
									   Allocator&					allocator,
									   AllocationKind				allocationKind)
{
	switch (allocationKind)
	{
		case ALLOCATION_KIND_SUBALLOCATED:
		{
			const VkMemoryRequirements memoryRequirements = getImageMemoryRequirements(vkd, device, image);

			return allocator.allocate(memoryRequirements, requirement);
		}

		case ALLOCATION_KIND_DEDICATED:
		{
			return allocateDedicated(vki, vkd, physDevice, device, image, requirement);
		}

		default:
		{
			TCU_THROW(InternalError, "Invalid allocation kind");
		}
	}
}


inline deUint32 getArraySize(const ImageParms& parms)
{
	return (parms.imageType != VK_IMAGE_TYPE_3D) ? parms.extent.depth : 1u;
}

inline VkImageCreateFlags getCreateFlags(const ImageParms& parms)
{
	return parms.imageType == VK_IMAGE_TYPE_2D && parms.extent.depth % 6 == 0 ?
		VK_IMAGE_CREATE_CUBE_COMPATIBLE_BIT : 0;
}

inline VkExtent3D getExtent3D(const ImageParms& parms, deUint32 mipLevel = 0u)
{
	const bool			isCompressed	= isCompressedFormat(parms.format);
	const deUint32		blockWidth		= (isCompressed) ? getBlockWidth(parms.format) : 1u;
	const deUint32		blockHeight		= (isCompressed) ? getBlockHeight(parms.format) : 1u;

	if (isCompressed && mipLevel != 0u)
		DE_FATAL("Not implemented");

	const VkExtent3D	extent			=
	{
		(parms.extent.width >> mipLevel) * blockWidth,
		(parms.imageType != VK_IMAGE_TYPE_1D) ? ((parms.extent.height >> mipLevel) * blockHeight) : 1u,
		(parms.imageType == VK_IMAGE_TYPE_3D) ? parms.extent.depth : 1u,
	};
	return extent;
}

const tcu::TextureFormat mapCombinedToDepthTransferFormat (const tcu::TextureFormat& combinedFormat)
{
	tcu::TextureFormat format;
	switch (combinedFormat.type)
	{
		case tcu::TextureFormat::UNORM_INT16:
		case tcu::TextureFormat::UNSIGNED_INT_16_8_8:
			format = tcu::TextureFormat(tcu::TextureFormat::D, tcu::TextureFormat::UNORM_INT16);
			break;
		case tcu::TextureFormat::UNSIGNED_INT_24_8_REV:
			format = tcu::TextureFormat(tcu::TextureFormat::D, tcu::TextureFormat::UNSIGNED_INT_24_8_REV);
			break;
		case tcu::TextureFormat::FLOAT_UNSIGNED_INT_24_8_REV:
		case tcu::TextureFormat::FLOAT:
			format = tcu::TextureFormat(tcu::TextureFormat::D, tcu::TextureFormat::FLOAT);
			break;
		default:
			DE_ASSERT(false);
			break;
	}
	return format;
}

class CopiesAndBlittingTestInstance : public vkt::TestInstance
{
public:
										CopiesAndBlittingTestInstance		(Context&	context,
																			 TestParams	testParams);
	virtual tcu::TestStatus				iterate								(void) = 0;

	enum FillMode
	{
		FILL_MODE_GRADIENT = 0,
		FILL_MODE_WHITE,
		FILL_MODE_RED,
		FILL_MODE_MULTISAMPLE,

		FILL_MODE_LAST
	};

protected:
	const TestParams					m_params;

	Move<VkCommandPool>					m_cmdPool;
	Move<VkCommandBuffer>				m_cmdBuffer;
	Move<VkFence>						m_fence;
	de::MovePtr<tcu::TextureLevel>		m_sourceTextureLevel;
	de::MovePtr<tcu::TextureLevel>		m_destinationTextureLevel;
	de::MovePtr<tcu::TextureLevel>		m_expectedTextureLevel[16];

	VkCommandBufferBeginInfo			m_cmdBufferBeginInfo;

	void								generateBuffer						(tcu::PixelBufferAccess buffer, int width, int height, int depth = 1, FillMode = FILL_MODE_GRADIENT);
	virtual void						generateExpectedResult				(void);
	void								uploadBuffer						(tcu::ConstPixelBufferAccess bufferAccess, const Allocation& bufferAlloc);
	void								uploadImage							(const tcu::ConstPixelBufferAccess& src, VkImage dst, const ImageParms& parms, const deUint32 mipLevels = 1u);
	virtual tcu::TestStatus				checkTestResult						(tcu::ConstPixelBufferAccess result);
	virtual void						copyRegionToTextureLevel			(tcu::ConstPixelBufferAccess src, tcu::PixelBufferAccess dst, CopyRegion region, deUint32 mipLevel = 0u) = 0;
	deUint32							calculateSize						(tcu::ConstPixelBufferAccess src) const
										{
											return src.getWidth() * src.getHeight() * src.getDepth() * tcu::getPixelSize(src.getFormat());
										}

	de::MovePtr<tcu::TextureLevel>		readImage							(vk::VkImage				image,
																			 const ImageParms&			imageParms,
																			 const deUint32				mipLevel = 0u);

private:
	void								uploadImageAspect					(const tcu::ConstPixelBufferAccess&	src,
																			 const VkImage&						dst,
																			 const ImageParms&					parms,
																			 const deUint32						mipLevels = 1u);
	void								readImageAspect						(vk::VkImage						src,
																			 const tcu::PixelBufferAccess&		dst,
																			 const ImageParms&					parms,
																			 const deUint32						mipLevel = 0u);
};

CopiesAndBlittingTestInstance::CopiesAndBlittingTestInstance (Context& context, TestParams testParams)
	: vkt::TestInstance	(context)
	, m_params			(testParams)
{
	const DeviceInterface&		vk					= context.getDeviceInterface();
	const VkDevice				vkDevice			= context.getDevice();
	const deUint32				queueFamilyIndex	= context.getUniversalQueueFamilyIndex();

	// Create command pool
	m_cmdPool = createCommandPool(vk, vkDevice, VK_COMMAND_POOL_CREATE_RESET_COMMAND_BUFFER_BIT, queueFamilyIndex);

	// Create command buffer
	m_cmdBuffer = allocateCommandBuffer(vk, vkDevice, *m_cmdPool, VK_COMMAND_BUFFER_LEVEL_PRIMARY);

	// Create fence
	m_fence = createFence(vk, vkDevice);
}

void CopiesAndBlittingTestInstance::generateBuffer (tcu::PixelBufferAccess buffer, int width, int height, int depth, FillMode mode)
{
	const tcu::TextureChannelClass	channelClass	= tcu::getTextureChannelClass(buffer.getFormat().type);
	tcu::Vec4						maxValue		(1.0f);

	if (buffer.getFormat().order == tcu::TextureFormat::S)
	{
		// Stencil-only is stored in the first component. Stencil is always 8 bits.
		maxValue.x() = 1 << 8;
	}
	else if (buffer.getFormat().order == tcu::TextureFormat::DS)
	{
		// In a combined format, fillWithComponentGradients expects stencil in the fourth component.
		maxValue.w() = 1 << 8;
	}
	else if (channelClass == tcu::TEXTURECHANNELCLASS_SIGNED_INTEGER || channelClass == tcu::TEXTURECHANNELCLASS_UNSIGNED_INTEGER)
	{
		// The tcu::Vectors we use as pixels are 32-bit, so clamp to that.
		const tcu::IVec4	bits	= tcu::min(tcu::getTextureFormatBitDepth(buffer.getFormat()), tcu::IVec4(32));
		const int			signBit	= (channelClass == tcu::TEXTURECHANNELCLASS_SIGNED_INTEGER ? 1 : 0);

		for (int i = 0; i < 4; ++i)
		{
			if (bits[i] != 0)
				maxValue[i] = static_cast<float>((deUint64(1) << (bits[i] - signBit)) - 1);
		}
	}

	if (mode == FILL_MODE_GRADIENT)
	{
		tcu::fillWithComponentGradients(buffer, tcu::Vec4(0.0f, 0.0f, 0.0f, 0.0f), maxValue);
		return;
	}

	const tcu::Vec4		redColor	(maxValue.x(),	0.0,			0.0,			maxValue.w());
	const tcu::Vec4		greenColor	(0.0,			maxValue.y(),	0.0,			maxValue.w());
	const tcu::Vec4		blueColor	(0.0,			0.0,			maxValue.z(),	maxValue.w());
	const tcu::Vec4		whiteColor	(maxValue.x(),	maxValue.y(),	maxValue.z(),	maxValue.w());

	for (int z = 0; z < depth;  ++z)
	for (int y = 0; y < height; ++y)
	for (int x = 0; x < width;  ++x)
	{
		switch (mode)
		{
			case FILL_MODE_WHITE:
				if (tcu::isCombinedDepthStencilType(buffer.getFormat().type))
				{
					buffer.setPixDepth(1.0f, x, y, z);
					if (tcu::hasStencilComponent(buffer.getFormat().order))
						buffer.setPixStencil(255, x, y, z);
				}
				else
					buffer.setPixel(whiteColor, x, y, z);
				break;

			case FILL_MODE_RED:
				if (tcu::isCombinedDepthStencilType(buffer.getFormat().type))
				{
					buffer.setPixDepth(redColor[0], x, y, z);
					if (tcu::hasStencilComponent(buffer.getFormat().order))
						buffer.setPixStencil((int)redColor[3], x, y, z);
				}
				else
					buffer.setPixel(redColor, x, y, z);
				break;

			case FILL_MODE_MULTISAMPLE:
			{
				float xScaled = static_cast<float>(x) / static_cast<float>(width);
				float yScaled = static_cast<float>(y) / static_cast<float>(height);
				buffer.setPixel((xScaled == yScaled) ? tcu::Vec4(0.0, 0.5, 0.5, 1.0) : ((xScaled > yScaled) ? greenColor : blueColor), x, y, z);
				break;
			}

			default:
				break;
		}
	}
}

void CopiesAndBlittingTestInstance::uploadBuffer (tcu::ConstPixelBufferAccess bufferAccess, const Allocation& bufferAlloc)
{
	const DeviceInterface&		vk			= m_context.getDeviceInterface();
	const VkDevice				vkDevice	= m_context.getDevice();
	const deUint32				bufferSize	= calculateSize(bufferAccess);

	// Write buffer data
	deMemcpy(bufferAlloc.getHostPtr(), bufferAccess.getDataPtr(), bufferSize);
	flushAlloc(vk, vkDevice, bufferAlloc);
}

void CopiesAndBlittingTestInstance::uploadImageAspect (const tcu::ConstPixelBufferAccess& imageAccess, const VkImage& image, const ImageParms& parms, const deUint32 mipLevels)
{
	const InstanceInterface&		vki					= m_context.getInstanceInterface();
	const DeviceInterface&			vk					= m_context.getDeviceInterface();
	const VkPhysicalDevice			vkPhysDevice		= m_context.getPhysicalDevice();
	const VkDevice					vkDevice			= m_context.getDevice();
	const VkQueue					queue				= m_context.getUniversalQueue();
	const deUint32					queueFamilyIndex	= m_context.getUniversalQueueFamilyIndex();
	Allocator&						memAlloc			= m_context.getDefaultAllocator();
	Move<VkBuffer>					buffer;
	const deUint32					bufferSize			= calculateSize(imageAccess);
	de::MovePtr<Allocation>			bufferAlloc;
	const deUint32					arraySize			= getArraySize(parms);
	const VkExtent3D				imageExtent			= getExtent3D(parms);
	std::vector <VkBufferImageCopy>	copyRegions;

	// Create source buffer
	{
		const VkBufferCreateInfo	bufferParams		=
		{
			VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO,		// VkStructureType		sType;
			DE_NULL,									// const void*			pNext;
			0u,											// VkBufferCreateFlags	flags;
			bufferSize,									// VkDeviceSize			size;
			VK_BUFFER_USAGE_TRANSFER_SRC_BIT,			// VkBufferUsageFlags	usage;
			VK_SHARING_MODE_EXCLUSIVE,					// VkSharingMode		sharingMode;
			1u,											// deUint32				queueFamilyIndexCount;
			&queueFamilyIndex,							// const deUint32*		pQueueFamilyIndices;
		};

		buffer		= createBuffer(vk, vkDevice, &bufferParams);
		bufferAlloc = allocateBuffer(vki, vk, vkPhysDevice, vkDevice, *buffer, MemoryRequirement::HostVisible, memAlloc, m_params.allocationKind);
		VK_CHECK(vk.bindBufferMemory(vkDevice, *buffer, bufferAlloc->getMemory(), bufferAlloc->getOffset()));
	}

	// Barriers for copying buffer to image
	const VkBufferMemoryBarrier		preBufferBarrier	=
	{
		VK_STRUCTURE_TYPE_BUFFER_MEMORY_BARRIER,		// VkStructureType	sType;
		DE_NULL,										// const void*		pNext;
		VK_ACCESS_HOST_WRITE_BIT,						// VkAccessFlags	srcAccessMask;
		VK_ACCESS_TRANSFER_READ_BIT,					// VkAccessFlags	dstAccessMask;
		VK_QUEUE_FAMILY_IGNORED,						// deUint32			srcQueueFamilyIndex;
		VK_QUEUE_FAMILY_IGNORED,						// deUint32			dstQueueFamilyIndex;
		*buffer,										// VkBuffer			buffer;
		0u,												// VkDeviceSize		offset;
		bufferSize										// VkDeviceSize		size;
	};

	const VkImageAspectFlags		formatAspect		= (m_params.separateDepthStencilLayouts) ? getAspectFlags(imageAccess.getFormat()) : getAspectFlags(parms.format);
	const bool						skipPreImageBarrier	= (m_params.separateDepthStencilLayouts) ? false : ((formatAspect == (VK_IMAGE_ASPECT_DEPTH_BIT | VK_IMAGE_ASPECT_STENCIL_BIT) &&
														  getAspectFlags(imageAccess.getFormat()) == VK_IMAGE_ASPECT_STENCIL_BIT));

	const VkImageMemoryBarrier		preImageBarrier		=
	{
		VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER,			// VkStructureType			sType;
		DE_NULL,										// const void*				pNext;
		0u,												// VkAccessFlags			srcAccessMask;
		VK_ACCESS_TRANSFER_WRITE_BIT,					// VkAccessFlags			dstAccessMask;
		VK_IMAGE_LAYOUT_UNDEFINED,						// VkImageLayout			oldLayout;
		VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL,			// VkImageLayout			newLayout;
		VK_QUEUE_FAMILY_IGNORED,						// deUint32					srcQueueFamilyIndex;
		VK_QUEUE_FAMILY_IGNORED,						// deUint32					dstQueueFamilyIndex;
		image,											// VkImage					image;
		{												// VkImageSubresourceRange	subresourceRange;
			formatAspect,	// VkImageAspectFlags	aspect;
			0u,				// deUint32				baseMipLevel;
			mipLevels,		// deUint32				mipLevels;
			0u,				// deUint32				baseArraySlice;
			arraySize,		// deUint32				arraySize;
		}
	};

	const VkImageMemoryBarrier		postImageBarrier	=
	{
		VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER,			// VkStructureType			sType;
		DE_NULL,										// const void*				pNext;
		VK_ACCESS_TRANSFER_WRITE_BIT,					// VkAccessFlags			srcAccessMask;
		VK_ACCESS_TRANSFER_WRITE_BIT,					// VkAccessFlags			dstAccessMask;
		VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL,			// VkImageLayout			oldLayout;
		VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL,			// VkImageLayout			newLayout;
		VK_QUEUE_FAMILY_IGNORED,						// deUint32					srcQueueFamilyIndex;
		VK_QUEUE_FAMILY_IGNORED,						// deUint32					dstQueueFamilyIndex;
		image,											// VkImage					image;
		{												// VkImageSubresourceRange	subresourceRange;
			formatAspect,				// VkImageAspectFlags	aspect;
			0u,							// deUint32				baseMipLevel;
			mipLevels,					// deUint32				mipLevels;
			0u,							// deUint32				baseArraySlice;
			arraySize,					// deUint32				arraySize;
		}
	};

	for (deUint32 mipLevelNdx = 0; mipLevelNdx < mipLevels; mipLevelNdx++)
	{
		const VkExtent3D		copyExtent	=
		{
			imageExtent.width	>> mipLevelNdx,
			imageExtent.height	>> mipLevelNdx,
			imageExtent.depth
		};

		const VkBufferImageCopy	copyRegion	=
		{
			0u,												// VkDeviceSize				bufferOffset;
			copyExtent.width,								// deUint32					bufferRowLength;
			copyExtent.height,								// deUint32					bufferImageHeight;
			{
				getAspectFlags(imageAccess.getFormat()),		// VkImageAspectFlags	aspect;
				mipLevelNdx,									// deUint32				mipLevel;
				0u,												// deUint32				baseArrayLayer;
				arraySize,										// deUint32				layerCount;
			},												// VkImageSubresourceLayers	imageSubresource;
			{ 0, 0, 0 },									// VkOffset3D				imageOffset;
			copyExtent										// VkExtent3D				imageExtent;
		};

		copyRegions.push_back(copyRegion);
	}

	// Write buffer data
	deMemcpy(bufferAlloc->getHostPtr(), imageAccess.getDataPtr(), bufferSize);
	flushAlloc(vk, vkDevice, *bufferAlloc);

	// Copy buffer to image
	beginCommandBuffer(vk, *m_cmdBuffer);
	vk.cmdPipelineBarrier(*m_cmdBuffer, VK_PIPELINE_STAGE_HOST_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT, (VkDependencyFlags)0, 0, (const VkMemoryBarrier*)DE_NULL,
						  1, &preBufferBarrier, (skipPreImageBarrier ? 0 : 1), (skipPreImageBarrier ? DE_NULL : &preImageBarrier));
	vk.cmdCopyBufferToImage(*m_cmdBuffer, *buffer, image, VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, (deUint32)copyRegions.size(), &copyRegions[0]);
	vk.cmdPipelineBarrier(*m_cmdBuffer, VK_PIPELINE_STAGE_TRANSFER_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT, (VkDependencyFlags)0, 0, (const VkMemoryBarrier*)DE_NULL, 0, (const VkBufferMemoryBarrier*)DE_NULL, 1, &postImageBarrier);
	endCommandBuffer(vk, *m_cmdBuffer);

	submitCommandsAndWait(vk, vkDevice, queue, *m_cmdBuffer);
}

void CopiesAndBlittingTestInstance::uploadImage (const tcu::ConstPixelBufferAccess& src, VkImage dst, const ImageParms& parms, const deUint32 mipLevels)
{
	if (tcu::isCombinedDepthStencilType(src.getFormat().type))
	{
		if (tcu::hasDepthComponent(src.getFormat().order))
		{
			tcu::TextureLevel	depthTexture	(mapCombinedToDepthTransferFormat(src.getFormat()), src.getWidth(), src.getHeight(), src.getDepth());
			tcu::copy(depthTexture.getAccess(), tcu::getEffectiveDepthStencilAccess(src, tcu::Sampler::MODE_DEPTH));
			uploadImageAspect(depthTexture.getAccess(), dst, parms);
		}

		if (tcu::hasStencilComponent(src.getFormat().order))
		{
			tcu::TextureLevel	stencilTexture	(tcu::getEffectiveDepthStencilTextureFormat(src.getFormat(), tcu::Sampler::MODE_STENCIL), src.getWidth(), src.getHeight(), src.getDepth());
			tcu::copy(stencilTexture.getAccess(), tcu::getEffectiveDepthStencilAccess(src, tcu::Sampler::MODE_STENCIL));
			uploadImageAspect(stencilTexture.getAccess(), dst, parms);
		}
	}
	else
		uploadImageAspect(src, dst, parms, mipLevels);
}

tcu::TestStatus CopiesAndBlittingTestInstance::checkTestResult (tcu::ConstPixelBufferAccess result)
{
	const tcu::ConstPixelBufferAccess	expected	= m_expectedTextureLevel[0]->getAccess();

	if (isFloatFormat(result.getFormat()))
	{
		const tcu::Vec4	threshold (0.0f);
		if (!tcu::floatThresholdCompare(m_context.getTestContext().getLog(), "Compare", "Result comparison", expected, result, threshold, tcu::COMPARE_LOG_RESULT))
			return tcu::TestStatus::fail("CopiesAndBlitting test");
	}
	else
	{
		const tcu::UVec4 threshold (0u);
		if (tcu::hasDepthComponent(result.getFormat().order) || tcu::hasStencilComponent(result.getFormat().order))
		{
			if (!tcu::dsThresholdCompare(m_context.getTestContext().getLog(), "Compare", "Result comparison", expected, result, 0.1f, tcu::COMPARE_LOG_RESULT))
				return tcu::TestStatus::fail("CopiesAndBlitting test");
		}
		else
		{
			if (!tcu::intThresholdCompare(m_context.getTestContext().getLog(), "Compare", "Result comparison", expected, result, threshold, tcu::COMPARE_LOG_RESULT))
				return tcu::TestStatus::fail("CopiesAndBlitting test");
		}
	}

	return tcu::TestStatus::pass("CopiesAndBlitting test");
}

void CopiesAndBlittingTestInstance::generateExpectedResult (void)
{
	const tcu::ConstPixelBufferAccess	src	= m_sourceTextureLevel->getAccess();
	const tcu::ConstPixelBufferAccess	dst	= m_destinationTextureLevel->getAccess();

	m_expectedTextureLevel[0]	= de::MovePtr<tcu::TextureLevel>(new tcu::TextureLevel(dst.getFormat(), dst.getWidth(), dst.getHeight(), dst.getDepth()));
	tcu::copy(m_expectedTextureLevel[0]->getAccess(), dst);

	for (deUint32 i = 0; i < m_params.regions.size(); i++)
		copyRegionToTextureLevel(src, m_expectedTextureLevel[0]->getAccess(), m_params.regions[i]);
}

class CopiesAndBlittingTestCase : public vkt::TestCase
{
public:
							CopiesAndBlittingTestCase	(tcu::TestContext&			testCtx,
														 const std::string&			name,
														 const std::string&			description)
								: vkt::TestCase	(testCtx, name, description)
							{}

	virtual TestInstance*	createInstance				(Context&					context) const = 0;
};

void CopiesAndBlittingTestInstance::readImageAspect (vk::VkImage					image,
													 const tcu::PixelBufferAccess&	dst,
													 const ImageParms&				imageParms,
													 const deUint32					mipLevel)
{
	const InstanceInterface&	vki					= m_context.getInstanceInterface();
	const DeviceInterface&		vk					= m_context.getDeviceInterface();
	const VkPhysicalDevice		physDevice			= m_context.getPhysicalDevice();
	const VkDevice				device				= m_context.getDevice();
	const VkQueue				queue				= m_context.getUniversalQueue();
	Allocator&					allocator			= m_context.getDefaultAllocator();

	Move<VkBuffer>				buffer;
	de::MovePtr<Allocation>		bufferAlloc;
	const deUint32				queueFamilyIndex	= m_context.getUniversalQueueFamilyIndex();
	const VkDeviceSize			pixelDataSize		= calculateSize(dst);
	const VkExtent3D			imageExtent			= getExtent3D(imageParms, mipLevel);

	// Create destination buffer
	{
		const VkBufferCreateInfo			bufferParams			=
		{
			VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO,		// VkStructureType		sType;
			DE_NULL,									// const void*			pNext;
			0u,											// VkBufferCreateFlags	flags;
			pixelDataSize,								// VkDeviceSize			size;
			VK_BUFFER_USAGE_TRANSFER_DST_BIT,			// VkBufferUsageFlags	usage;
			VK_SHARING_MODE_EXCLUSIVE,					// VkSharingMode		sharingMode;
			1u,											// deUint32				queueFamilyIndexCount;
			&queueFamilyIndex,							// const deUint32*		pQueueFamilyIndices;
		};

		buffer		= createBuffer(vk, device, &bufferParams);
		bufferAlloc = allocateBuffer(vki, vk, physDevice, device, *buffer, MemoryRequirement::HostVisible, allocator, m_params.allocationKind);
		VK_CHECK(vk.bindBufferMemory(device, *buffer, bufferAlloc->getMemory(), bufferAlloc->getOffset()));

		deMemset(bufferAlloc->getHostPtr(), 0, static_cast<size_t>(pixelDataSize));
		flushAlloc(vk, device, *bufferAlloc);
	}

	// Barriers for copying image to buffer
	const VkImageAspectFlags				formatAspect			= getAspectFlags(imageParms.format);
	const VkImageMemoryBarrier				imageBarrier			=
	{
		VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER,		// VkStructureType			sType;
		DE_NULL,									// const void*				pNext;
		VK_ACCESS_TRANSFER_WRITE_BIT,				// VkAccessFlags			srcAccessMask;
		VK_ACCESS_TRANSFER_READ_BIT,				// VkAccessFlags			dstAccessMask;
		imageParms.operationLayout,					// VkImageLayout			oldLayout;
		VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL,		// VkImageLayout			newLayout;
		VK_QUEUE_FAMILY_IGNORED,					// deUint32					srcQueueFamilyIndex;
		VK_QUEUE_FAMILY_IGNORED,					// deUint32					dstQueueFamilyIndex;
		image,										// VkImage					image;
		{											// VkImageSubresourceRange	subresourceRange;
			formatAspect,			// VkImageAspectFlags	aspectMask;
			mipLevel,				// deUint32				baseMipLevel;
			1u,						// deUint32				mipLevels;
			0u,						// deUint32				baseArraySlice;
			getArraySize(imageParms)// deUint32				arraySize;
		}
	};

	const VkBufferMemoryBarrier				bufferBarrier			=
	{
		VK_STRUCTURE_TYPE_BUFFER_MEMORY_BARRIER,	// VkStructureType	sType;
		DE_NULL,									// const void*		pNext;
		VK_ACCESS_TRANSFER_WRITE_BIT,				// VkAccessFlags	srcAccessMask;
		VK_ACCESS_HOST_READ_BIT,					// VkAccessFlags	dstAccessMask;
		VK_QUEUE_FAMILY_IGNORED,					// deUint32			srcQueueFamilyIndex;
		VK_QUEUE_FAMILY_IGNORED,					// deUint32			dstQueueFamilyIndex;
		*buffer,									// VkBuffer			buffer;
		0u,											// VkDeviceSize		offset;
		pixelDataSize								// VkDeviceSize		size;
	};

	const VkImageMemoryBarrier				postImageBarrier		=
	{
		VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER,		// VkStructureType			sType;
		DE_NULL,									// const void*				pNext;
		VK_ACCESS_TRANSFER_READ_BIT,				// VkAccessFlags			srcAccessMask;
		VK_ACCESS_TRANSFER_WRITE_BIT,				// VkAccessFlags			dstAccessMask;
		VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL,		// VkImageLayout			oldLayout;
		imageParms.operationLayout,					// VkImageLayout			newLayout;
		VK_QUEUE_FAMILY_IGNORED,					// deUint32					srcQueueFamilyIndex;
		VK_QUEUE_FAMILY_IGNORED,					// deUint32					dstQueueFamilyIndex;
		image,										// VkImage					image;
		{
			formatAspect,								// VkImageAspectFlags	aspectMask;
			mipLevel,									// deUint32				baseMipLevel;
			1u,											// deUint32				mipLevels;
			0u,											// deUint32				baseArraySlice;
			getArraySize(imageParms)					// deUint32				arraySize;
		}											// VkImageSubresourceRange	subresourceRange;
	};

	// Copy image to buffer
	const VkImageAspectFlags	aspect			= getAspectFlags(dst.getFormat());
	const VkBufferImageCopy		copyRegion		=
	{
		0u,								// VkDeviceSize				bufferOffset;
		imageExtent.width,				// deUint32					bufferRowLength;
		imageExtent.height,				// deUint32					bufferImageHeight;
		{
			aspect,							// VkImageAspectFlags		aspect;
			mipLevel,						// deUint32					mipLevel;
			0u,								// deUint32					baseArrayLayer;
			getArraySize(imageParms),		// deUint32					layerCount;
		},								// VkImageSubresourceLayers	imageSubresource;
		{ 0, 0, 0 },					// VkOffset3D				imageOffset;
		imageExtent						// VkExtent3D				imageExtent;
	};

	beginCommandBuffer(vk, *m_cmdBuffer);
	vk.cmdPipelineBarrier(*m_cmdBuffer, VK_PIPELINE_STAGE_TRANSFER_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT, (VkDependencyFlags)0, 0, (const VkMemoryBarrier*)DE_NULL, 0, (const VkBufferMemoryBarrier*)DE_NULL, 1, &imageBarrier);
	vk.cmdCopyImageToBuffer(*m_cmdBuffer, image, VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL, *buffer, 1u, &copyRegion);
	vk.cmdPipelineBarrier(*m_cmdBuffer, VK_PIPELINE_STAGE_TRANSFER_BIT, VK_PIPELINE_STAGE_HOST_BIT|VK_PIPELINE_STAGE_TRANSFER_BIT, (VkDependencyFlags)0, 0, (const VkMemoryBarrier*)DE_NULL, 1, &bufferBarrier, 1, &postImageBarrier);
	endCommandBuffer(vk, *m_cmdBuffer);

	submitCommandsAndWait(vk, device, queue, *m_cmdBuffer);

	// Read buffer data
	invalidateAlloc(vk, device, *bufferAlloc);
	tcu::copy(dst, tcu::ConstPixelBufferAccess(dst.getFormat(), dst.getSize(), bufferAlloc->getHostPtr()));
}

de::MovePtr<tcu::TextureLevel> CopiesAndBlittingTestInstance::readImage	(vk::VkImage		image,
																		 const ImageParms&	parms,
																		 const deUint32		mipLevel)
{
	const tcu::TextureFormat		imageFormat	= getSizeCompatibleTcuTextureFormat(parms.format);
	de::MovePtr<tcu::TextureLevel>	resultLevel	(new tcu::TextureLevel(imageFormat, parms.extent.width >> mipLevel, parms.extent.height >> mipLevel, parms.extent.depth));

	if (tcu::isCombinedDepthStencilType(imageFormat.type))
	{
		if (tcu::hasDepthComponent(imageFormat.order))
		{
			tcu::TextureLevel	depthTexture	(mapCombinedToDepthTransferFormat(imageFormat), parms.extent.width, parms.extent.height, parms.extent.depth);
			readImageAspect(image, depthTexture.getAccess(), parms);
			tcu::copy(tcu::getEffectiveDepthStencilAccess(resultLevel->getAccess(), tcu::Sampler::MODE_DEPTH), depthTexture.getAccess());
		}

		if (tcu::hasStencilComponent(imageFormat.order))
		{
			tcu::TextureLevel	stencilTexture	(tcu::getEffectiveDepthStencilTextureFormat(imageFormat, tcu::Sampler::MODE_STENCIL), parms.extent.width, parms.extent.height, parms.extent.depth);
			readImageAspect(image, stencilTexture.getAccess(), parms);
			tcu::copy(tcu::getEffectiveDepthStencilAccess(resultLevel->getAccess(), tcu::Sampler::MODE_STENCIL), stencilTexture.getAccess());
		}
	}
	else
		readImageAspect(image, resultLevel->getAccess(), parms, mipLevel);

	return resultLevel;
}

// Copy from image to image.

class CopyImageToImage : public CopiesAndBlittingTestInstance
{
public:
										CopyImageToImage			(Context&	context,
																	 TestParams params);
	virtual tcu::TestStatus				iterate						(void);

protected:
	virtual tcu::TestStatus				checkTestResult				(tcu::ConstPixelBufferAccess result = tcu::ConstPixelBufferAccess());

private:
	Move<VkImage>						m_source;
	de::MovePtr<Allocation>				m_sourceImageAlloc;
	Move<VkImage>						m_destination;
	de::MovePtr<Allocation>				m_destinationImageAlloc;

	virtual void						copyRegionToTextureLevel	(tcu::ConstPixelBufferAccess src, tcu::PixelBufferAccess dst, CopyRegion region, deUint32 mipLevel = 0u);
};

CopyImageToImage::CopyImageToImage (Context& context, TestParams params)
	: CopiesAndBlittingTestInstance(context, params)
{
	const InstanceInterface&	vki					= context.getInstanceInterface();
	const DeviceInterface&		vk					= context.getDeviceInterface();
	const VkPhysicalDevice		vkPhysDevice		= context.getPhysicalDevice();
	const VkDevice				vkDevice			= context.getDevice();
	const deUint32				queueFamilyIndex	= context.getUniversalQueueFamilyIndex();
	Allocator&					memAlloc			= context.getDefaultAllocator();

	// Create source image
	{
		const VkImageCreateInfo	sourceImageParams		=
		{
			VK_STRUCTURE_TYPE_IMAGE_CREATE_INFO,	// VkStructureType		sType;
			DE_NULL,								// const void*			pNext;
			getCreateFlags(m_params.src.image),		// VkImageCreateFlags	flags;
			m_params.src.image.imageType,			// VkImageType			imageType;
			m_params.src.image.format,				// VkFormat				format;
			getExtent3D(m_params.src.image),		// VkExtent3D			extent;
			1u,										// deUint32				mipLevels;
			getArraySize(m_params.src.image),		// deUint32				arraySize;
			VK_SAMPLE_COUNT_1_BIT,					// deUint32				samples;
			VK_IMAGE_TILING_OPTIMAL,				// VkImageTiling		tiling;
			VK_IMAGE_USAGE_TRANSFER_SRC_BIT |
				VK_IMAGE_USAGE_TRANSFER_DST_BIT,	// VkImageUsageFlags	usage;
			VK_SHARING_MODE_EXCLUSIVE,				// VkSharingMode		sharingMode;
			1u,										// deUint32				queueFamilyCount;
			&queueFamilyIndex,						// const deUint32*		pQueueFamilyIndices;
			VK_IMAGE_LAYOUT_UNDEFINED,				// VkImageLayout		initialLayout;
		};

		m_source				= createImage(vk, vkDevice, &sourceImageParams);
		m_sourceImageAlloc		= allocateImage(vki, vk, vkPhysDevice, vkDevice, *m_source, MemoryRequirement::Any, memAlloc, m_params.allocationKind);
		VK_CHECK(vk.bindImageMemory(vkDevice, *m_source, m_sourceImageAlloc->getMemory(), m_sourceImageAlloc->getOffset()));
	}

	// Create destination image
	{
		const VkImageCreateInfo	destinationImageParams	=
		{
			VK_STRUCTURE_TYPE_IMAGE_CREATE_INFO,	// VkStructureType		sType;
			DE_NULL,								// const void*			pNext;
			getCreateFlags(m_params.dst.image),		// VkImageCreateFlags	flags;
			m_params.dst.image.imageType,			// VkImageType			imageType;
			m_params.dst.image.format,				// VkFormat				format;
			getExtent3D(m_params.dst.image),		// VkExtent3D			extent;
			1u,										// deUint32				mipLevels;
			getArraySize(m_params.dst.image),		// deUint32				arraySize;
			VK_SAMPLE_COUNT_1_BIT,					// deUint32				samples;
			VK_IMAGE_TILING_OPTIMAL,				// VkImageTiling		tiling;
			VK_IMAGE_USAGE_TRANSFER_SRC_BIT |
				VK_IMAGE_USAGE_TRANSFER_DST_BIT,	// VkImageUsageFlags	usage;
			VK_SHARING_MODE_EXCLUSIVE,				// VkSharingMode		sharingMode;
			1u,										// deUint32				queueFamilyCount;
			&queueFamilyIndex,						// const deUint32*		pQueueFamilyIndices;
			VK_IMAGE_LAYOUT_UNDEFINED,				// VkImageLayout		initialLayout;
		};

		m_destination			= createImage(vk, vkDevice, &destinationImageParams);
		m_destinationImageAlloc	= allocateImage(vki, vk, vkPhysDevice, vkDevice, *m_destination, MemoryRequirement::Any, memAlloc, m_params.allocationKind);
		VK_CHECK(vk.bindImageMemory(vkDevice, *m_destination, m_destinationImageAlloc->getMemory(), m_destinationImageAlloc->getOffset()));
	}
}

tcu::TestStatus CopyImageToImage::iterate (void)
{
	const bool					srcCompressed		= isCompressedFormat(m_params.src.image.format);
	const bool					dstCompressed		= isCompressedFormat(m_params.dst.image.format);

	const tcu::TextureFormat	srcTcuFormat		= getSizeCompatibleTcuTextureFormat(m_params.src.image.format);
	const tcu::TextureFormat	dstTcuFormat		= getSizeCompatibleTcuTextureFormat(m_params.dst.image.format);

	m_sourceTextureLevel = de::MovePtr<tcu::TextureLevel>(new tcu::TextureLevel(srcTcuFormat,
																				(int)m_params.src.image.extent.width,
																				(int)m_params.src.image.extent.height,
																				(int)m_params.src.image.extent.depth));
	generateBuffer(m_sourceTextureLevel->getAccess(), m_params.src.image.extent.width, m_params.src.image.extent.height, m_params.src.image.extent.depth, FILL_MODE_GRADIENT);
	m_destinationTextureLevel = de::MovePtr<tcu::TextureLevel>(new tcu::TextureLevel(dstTcuFormat,
																				(int)m_params.dst.image.extent.width,
																				(int)m_params.dst.image.extent.height,
																				(int)m_params.dst.image.extent.depth));
	generateBuffer(m_destinationTextureLevel->getAccess(), m_params.dst.image.extent.width, m_params.dst.image.extent.height, m_params.dst.image.extent.depth, FILL_MODE_GRADIENT);
	generateExpectedResult();

	uploadImage(m_sourceTextureLevel->getAccess(), m_source.get(), m_params.src.image);
	uploadImage(m_destinationTextureLevel->getAccess(), m_destination.get(), m_params.dst.image);

	const DeviceInterface&		vk					= m_context.getDeviceInterface();
	const VkDevice				vkDevice			= m_context.getDevice();
	const VkQueue				queue				= m_context.getUniversalQueue();

	std::vector<VkImageCopy>	imageCopies;
	for (deUint32 i = 0; i < m_params.regions.size(); i++)
	{
		VkImageCopy imageCopy = m_params.regions[i].imageCopy;

		// When copying between compressed and uncompressed formats the extent
		// members represent the texel dimensions of the source image.
		if (srcCompressed)
		{
			const deUint32	blockWidth	= getBlockWidth(m_params.src.image.format);
			const deUint32	blockHeight	= getBlockHeight(m_params.src.image.format);

			imageCopy.srcOffset.x *= blockWidth;
			imageCopy.srcOffset.y *= blockHeight;
			imageCopy.extent.width *= blockWidth;
			imageCopy.extent.height *= blockHeight;
		}

		if (dstCompressed)
		{
			const deUint32	blockWidth	= getBlockWidth(m_params.dst.image.format);
			const deUint32	blockHeight	= getBlockHeight(m_params.dst.image.format);

			imageCopy.dstOffset.x *= blockWidth;
			imageCopy.dstOffset.y *= blockHeight;
		}

		imageCopies.push_back(imageCopy);
	}

	const VkImageMemoryBarrier	imageBarriers[]		=
	{
		// source image
		{
			VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER,		// VkStructureType			sType;
			DE_NULL,									// const void*				pNext;
			VK_ACCESS_TRANSFER_WRITE_BIT,				// VkAccessFlags			srcAccessMask;
			VK_ACCESS_TRANSFER_READ_BIT,				// VkAccessFlags			dstAccessMask;
			VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL,		// VkImageLayout			oldLayout;
			m_params.src.image.operationLayout,			// VkImageLayout			newLayout;
			VK_QUEUE_FAMILY_IGNORED,					// deUint32					srcQueueFamilyIndex;
			VK_QUEUE_FAMILY_IGNORED,					// deUint32					dstQueueFamilyIndex;
			m_source.get(),								// VkImage					image;
			{											// VkImageSubresourceRange	subresourceRange;
				getAspectFlags(srcTcuFormat),	// VkImageAspectFlags	aspectMask;
				0u,								// deUint32				baseMipLevel;
				1u,								// deUint32				mipLevels;
				0u,								// deUint32				baseArraySlice;
				getArraySize(m_params.src.image)// deUint32				arraySize;
			}
		},
		// destination image
		{
			VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER,		// VkStructureType			sType;
			DE_NULL,									// const void*				pNext;
			VK_ACCESS_TRANSFER_WRITE_BIT,				// VkAccessFlags			srcAccessMask;
			VK_ACCESS_TRANSFER_WRITE_BIT,				// VkAccessFlags			dstAccessMask;
			VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL,		// VkImageLayout			oldLayout;
			m_params.dst.image.operationLayout,			// VkImageLayout			newLayout;
			VK_QUEUE_FAMILY_IGNORED,					// deUint32					srcQueueFamilyIndex;
			VK_QUEUE_FAMILY_IGNORED,					// deUint32					dstQueueFamilyIndex;
			m_destination.get(),						// VkImage					image;
			{											// VkImageSubresourceRange	subresourceRange;
				getAspectFlags(dstTcuFormat),	// VkImageAspectFlags	aspectMask;
				0u,								// deUint32				baseMipLevel;
				1u,								// deUint32				mipLevels;
				0u,								// deUint32				baseArraySlice;
				getArraySize(m_params.dst.image)// deUint32				arraySize;
			}
		},
	};

	beginCommandBuffer(vk, *m_cmdBuffer);
	vk.cmdPipelineBarrier(*m_cmdBuffer, VK_PIPELINE_STAGE_TRANSFER_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT, (VkDependencyFlags)0, 0, (const VkMemoryBarrier*)DE_NULL, 0, (const VkBufferMemoryBarrier*)DE_NULL, DE_LENGTH_OF_ARRAY(imageBarriers), imageBarriers);
	vk.cmdCopyImage(*m_cmdBuffer, m_source.get(), m_params.src.image.operationLayout, m_destination.get(), m_params.dst.image.operationLayout, (deUint32)imageCopies.size(), imageCopies.data());
	endCommandBuffer(vk, *m_cmdBuffer);

	submitCommandsAndWait (vk, vkDevice, queue, *m_cmdBuffer);

	de::MovePtr<tcu::TextureLevel>	resultTextureLevel	= readImage(*m_destination, m_params.dst.image);

	return checkTestResult(resultTextureLevel->getAccess());
}

tcu::TestStatus CopyImageToImage::checkTestResult (tcu::ConstPixelBufferAccess result)
{
	const tcu::Vec4	fThreshold (0.0f);
	const tcu::UVec4 uThreshold (0u);

	if (tcu::isCombinedDepthStencilType(result.getFormat().type))
	{
		if (tcu::hasDepthComponent(result.getFormat().order))
		{
			const tcu::Sampler::DepthStencilMode	mode				= tcu::Sampler::MODE_DEPTH;
			const tcu::ConstPixelBufferAccess		depthResult			= tcu::getEffectiveDepthStencilAccess(result, mode);
			const tcu::ConstPixelBufferAccess		expectedResult		= tcu::getEffectiveDepthStencilAccess(m_expectedTextureLevel[0]->getAccess(), mode);

			if (isFloatFormat(result.getFormat()))
			{
				if (!tcu::floatThresholdCompare(m_context.getTestContext().getLog(), "Compare", "Result comparison", expectedResult, depthResult, fThreshold, tcu::COMPARE_LOG_RESULT))
					return tcu::TestStatus::fail("CopiesAndBlitting test");
			}
			else
			{
				if (!tcu::intThresholdCompare(m_context.getTestContext().getLog(), "Compare", "Result comparison", expectedResult, depthResult, uThreshold, tcu::COMPARE_LOG_RESULT))
					return tcu::TestStatus::fail("CopiesAndBlitting test");
			}
		}

		if (tcu::hasStencilComponent(result.getFormat().order))
		{
			const tcu::Sampler::DepthStencilMode	mode				= tcu::Sampler::MODE_STENCIL;
			const tcu::ConstPixelBufferAccess		stencilResult		= tcu::getEffectiveDepthStencilAccess(result, mode);
			const tcu::ConstPixelBufferAccess		expectedResult		= tcu::getEffectiveDepthStencilAccess(m_expectedTextureLevel[0]->getAccess(), mode);

			if (isFloatFormat(result.getFormat()))
			{
				if (!tcu::floatThresholdCompare(m_context.getTestContext().getLog(), "Compare", "Result comparison", expectedResult, stencilResult, fThreshold, tcu::COMPARE_LOG_RESULT))
					return tcu::TestStatus::fail("CopiesAndBlitting test");
			}
			else
			{
				if (!tcu::intThresholdCompare(m_context.getTestContext().getLog(), "Compare", "Result comparison", expectedResult, stencilResult, uThreshold, tcu::COMPARE_LOG_RESULT))
					return tcu::TestStatus::fail("CopiesAndBlitting test");
			}
		}
	}
	else
	{
		if (isFloatFormat(result.getFormat()))
		{
			if (!tcu::floatThresholdCompare(m_context.getTestContext().getLog(), "Compare", "Result comparison", m_expectedTextureLevel[0]->getAccess(), result, fThreshold, tcu::COMPARE_LOG_RESULT))
				return tcu::TestStatus::fail("CopiesAndBlitting test");
		}
		else if (isSnormFormat(mapTextureFormat(result.getFormat())))
		{
			// There may be an ambiguity between two possible binary representations of 1.0.
			// Get rid of that by expanding the data to floats and re-normalizing again.

			tcu::TextureLevel resultSnorm	(result.getFormat(), result.getWidth(), result.getHeight(), result.getDepth());
			{
				tcu::TextureLevel resultFloat	(tcu::TextureFormat(resultSnorm.getFormat().order, tcu::TextureFormat::FLOAT), resultSnorm.getWidth(), resultSnorm.getHeight(), resultSnorm.getDepth());

				tcu::copy(resultFloat.getAccess(), result);
				tcu::copy(resultSnorm, resultFloat.getAccess());
			}

			tcu::TextureLevel expectedSnorm	(m_expectedTextureLevel[0]->getFormat(), m_expectedTextureLevel[0]->getWidth(), m_expectedTextureLevel[0]->getHeight(), m_expectedTextureLevel[0]->getDepth());

			{
				tcu::TextureLevel expectedFloat	(tcu::TextureFormat(expectedSnorm.getFormat().order, tcu::TextureFormat::FLOAT), expectedSnorm.getWidth(), expectedSnorm.getHeight(), expectedSnorm.getDepth());

				tcu::copy(expectedFloat.getAccess(), m_expectedTextureLevel[0]->getAccess());
				tcu::copy(expectedSnorm, expectedFloat.getAccess());
			}

			if (!tcu::intThresholdCompare(m_context.getTestContext().getLog(), "Compare", "Result comparison", expectedSnorm.getAccess(), resultSnorm.getAccess(), uThreshold, tcu::COMPARE_LOG_RESULT))
				return tcu::TestStatus::fail("CopiesAndBlitting test");
		}
		else
		{
			if (!tcu::intThresholdCompare(m_context.getTestContext().getLog(), "Compare", "Result comparison", m_expectedTextureLevel[0]->getAccess(), result, uThreshold, tcu::COMPARE_LOG_RESULT))
				return tcu::TestStatus::fail("CopiesAndBlitting test");
		}
	}

	return tcu::TestStatus::pass("CopiesAndBlitting test");
}

void CopyImageToImage::copyRegionToTextureLevel (tcu::ConstPixelBufferAccess src, tcu::PixelBufferAccess dst, CopyRegion region, deUint32 mipLevel)
{
	DE_UNREF(mipLevel);

	VkOffset3D	srcOffset	= region.imageCopy.srcOffset;
	VkOffset3D	dstOffset	= region.imageCopy.dstOffset;
	VkExtent3D	extent		= region.imageCopy.extent;

	if (m_params.src.image.imageType == VK_IMAGE_TYPE_3D && m_params.dst.image.imageType == VK_IMAGE_TYPE_2D)
	{
		dstOffset.z = srcOffset.z;
		extent.depth = std::max(region.imageCopy.extent.depth, region.imageCopy.dstSubresource.layerCount);
	}
	if (m_params.src.image.imageType == VK_IMAGE_TYPE_2D && m_params.dst.image.imageType == VK_IMAGE_TYPE_3D)
	{
		srcOffset.z = dstOffset.z;
		extent.depth = std::max(region.imageCopy.extent.depth, region.imageCopy.srcSubresource.layerCount);
	}


	if (tcu::isCombinedDepthStencilType(src.getFormat().type))
	{
		DE_ASSERT(src.getFormat() == dst.getFormat());

		// Copy depth.
		if (tcu::hasDepthComponent(src.getFormat().order))
		{
			const tcu::ConstPixelBufferAccess	srcSubRegion	= getEffectiveDepthStencilAccess(tcu::getSubregion(src, srcOffset.x, srcOffset.y, srcOffset.z, extent.width, extent.height, extent.depth), tcu::Sampler::MODE_DEPTH);
			const tcu::PixelBufferAccess		dstSubRegion	= getEffectiveDepthStencilAccess(tcu::getSubregion(dst, dstOffset.x, dstOffset.y, dstOffset.z, extent.width, extent.height, extent.depth), tcu::Sampler::MODE_DEPTH);
			tcu::copy(dstSubRegion, srcSubRegion);
		}

		// Copy stencil.
		if (tcu::hasStencilComponent(src.getFormat().order))
		{
			const tcu::ConstPixelBufferAccess	srcSubRegion	= getEffectiveDepthStencilAccess(tcu::getSubregion(src, srcOffset.x, srcOffset.y, srcOffset.z, extent.width, extent.height, extent.depth), tcu::Sampler::MODE_STENCIL);
			const tcu::PixelBufferAccess		dstSubRegion	= getEffectiveDepthStencilAccess(tcu::getSubregion(dst, dstOffset.x, dstOffset.y, dstOffset.z, extent.width, extent.height, extent.depth), tcu::Sampler::MODE_STENCIL);
			tcu::copy(dstSubRegion, srcSubRegion);
		}
	}
	else
	{
		const tcu::ConstPixelBufferAccess	srcSubRegion		= tcu::getSubregion(src, srcOffset.x, srcOffset.y, srcOffset.z, extent.width, extent.height, extent.depth);
		// CopyImage acts like a memcpy. Replace the destination format with the srcformat to use a memcpy.
		const tcu::PixelBufferAccess		dstWithSrcFormat	(srcSubRegion.getFormat(), dst.getSize(), dst.getDataPtr());
		const tcu::PixelBufferAccess		dstSubRegion		= tcu::getSubregion(dstWithSrcFormat, dstOffset.x, dstOffset.y, dstOffset.z, extent.width, extent.height, extent.depth);

		tcu::copy(dstSubRegion, srcSubRegion);
	}
}

class CopyImageToImageTestCase : public vkt::TestCase
{
public:
							CopyImageToImageTestCase	(tcu::TestContext&				testCtx,
														 const std::string&				name,
														 const std::string&				description,
														 const TestParams				params)
								: vkt::TestCase	(testCtx, name, description)
								, m_params		(params)
	{}

	virtual TestInstance*	createInstance				(Context&						context) const
	{
		return new CopyImageToImage(context, m_params);
	}

	virtual void			checkSupport				(Context&						context) const
	{
		if (m_params.allocationKind == ALLOCATION_KIND_DEDICATED)
		{
			if (!context.isDeviceFunctionalitySupported("VK_KHR_dedicated_allocation"))
				TCU_THROW(NotSupportedError, "VK_KHR_dedicated_allocation is not supported");
		}

		if (m_params.separateDepthStencilLayouts)
			if (!context.isDeviceFunctionalitySupported("VK_KHR_separate_depth_stencil_layouts"))
				TCU_THROW(NotSupportedError, "VK_KHR_separate_depth_stencil_layouts is not supported");

		if ((m_params.dst.image.imageType == VK_IMAGE_TYPE_3D && m_params.src.image.imageType == VK_IMAGE_TYPE_2D) ||
			(m_params.dst.image.imageType == VK_IMAGE_TYPE_2D && m_params.src.image.imageType == VK_IMAGE_TYPE_3D))
		{
			if (!context.isDeviceFunctionalitySupported("VK_KHR_maintenance1"))
				TCU_THROW(NotSupportedError, "Extension VK_KHR_maintenance1 not supported");
		}

		const VkPhysicalDeviceLimits	limits		= context.getDeviceProperties().limits;
		VkImageFormatProperties			properties;

		if ((context.getInstanceInterface().getPhysicalDeviceImageFormatProperties (context.getPhysicalDevice(),
																					m_params.src.image.format,
																					m_params.src.image.imageType,
																					VK_IMAGE_TILING_OPTIMAL,
																					VK_IMAGE_USAGE_TRANSFER_SRC_BIT,
																					0,
																					&properties) == VK_ERROR_FORMAT_NOT_SUPPORTED) ||
			(context.getInstanceInterface().getPhysicalDeviceImageFormatProperties (context.getPhysicalDevice(),
																					m_params.dst.image.format,
																					m_params.dst.image.imageType,
																					VK_IMAGE_TILING_OPTIMAL,
																					VK_IMAGE_USAGE_TRANSFER_DST_BIT,
																					0,
																					&properties) == VK_ERROR_FORMAT_NOT_SUPPORTED))
		{
			TCU_THROW(NotSupportedError, "Format not supported");
		}

		// Check maxImageDimension2D
		{
			if (m_params.src.image.imageType == VK_IMAGE_TYPE_2D && (m_params.src.image.extent.width > limits.maxImageDimension2D
				|| m_params.src.image.extent.height > limits.maxImageDimension2D))
			{
				TCU_THROW(NotSupportedError, "Requested 2D src image dimensions not supported");
			}

			if (m_params.dst.image.imageType == VK_IMAGE_TYPE_2D && (m_params.dst.image.extent.width > limits.maxImageDimension2D
				|| m_params.dst.image.extent.height > limits.maxImageDimension2D))
			{
				TCU_THROW(NotSupportedError, "Requested 2D dst image dimensions not supported");
			}
		}

		// Check maxImageDimension3D
		{
			if (m_params.src.image.imageType == VK_IMAGE_TYPE_3D && (m_params.src.image.extent.width > limits.maxImageDimension3D
				|| m_params.src.image.extent.height > limits.maxImageDimension3D
				|| m_params.src.image.extent.depth > limits.maxImageDimension3D))
			{
				TCU_THROW(NotSupportedError, "Requested 3D src image dimensions not supported");
			}

			if (m_params.dst.image.imageType == VK_IMAGE_TYPE_3D && (m_params.dst.image.extent.width > limits.maxImageDimension3D
				|| m_params.dst.image.extent.height > limits.maxImageDimension3D
				|| m_params.src.image.extent.depth > limits.maxImageDimension3D))
			{
				TCU_THROW(NotSupportedError, "Requested 3D dst image dimensions not supported");
			}
		}
	}

private:
	TestParams				m_params;
};

// Copy from buffer to buffer.

class CopyBufferToBuffer : public CopiesAndBlittingTestInstance
{
public:
								CopyBufferToBuffer			(Context& context, TestParams params);
	virtual tcu::TestStatus		iterate						(void);
private:
	virtual void				copyRegionToTextureLevel	(tcu::ConstPixelBufferAccess, tcu::PixelBufferAccess, CopyRegion, deUint32 mipLevel = 0u);
	Move<VkBuffer>				m_source;
	de::MovePtr<Allocation>		m_sourceBufferAlloc;
	Move<VkBuffer>				m_destination;
	de::MovePtr<Allocation>		m_destinationBufferAlloc;
};

CopyBufferToBuffer::CopyBufferToBuffer (Context& context, TestParams params)
	: CopiesAndBlittingTestInstance	(context, params)
{
	const InstanceInterface&	vki					= context.getInstanceInterface();
	const DeviceInterface&		vk					= context.getDeviceInterface();
	const VkPhysicalDevice		vkPhysDevice		= context.getPhysicalDevice();
	const VkDevice				vkDevice			= context.getDevice();
	const deUint32				queueFamilyIndex	= context.getUniversalQueueFamilyIndex();
	Allocator&					memAlloc			= context.getDefaultAllocator();

	// Create source buffer
	{
		const VkBufferCreateInfo	sourceBufferParams		=
		{
			VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO,		// VkStructureType		sType;
			DE_NULL,									// const void*			pNext;
			0u,											// VkBufferCreateFlags	flags;
			m_params.src.buffer.size,					// VkDeviceSize			size;
			VK_BUFFER_USAGE_TRANSFER_SRC_BIT,			// VkBufferUsageFlags	usage;
			VK_SHARING_MODE_EXCLUSIVE,					// VkSharingMode		sharingMode;
			1u,											// deUint32				queueFamilyIndexCount;
			&queueFamilyIndex,							// const deUint32*		pQueueFamilyIndices;
		};

		m_source				= createBuffer(vk, vkDevice, &sourceBufferParams);
		m_sourceBufferAlloc		= allocateBuffer(vki, vk, vkPhysDevice, vkDevice, *m_source, MemoryRequirement::HostVisible, memAlloc, m_params.allocationKind);
		VK_CHECK(vk.bindBufferMemory(vkDevice, *m_source, m_sourceBufferAlloc->getMemory(), m_sourceBufferAlloc->getOffset()));
	}

	// Create destination buffer
	{
		const VkBufferCreateInfo	destinationBufferParams	=
		{
			VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO,		// VkStructureType		sType;
			DE_NULL,									// const void*			pNext;
			0u,											// VkBufferCreateFlags	flags;
			m_params.dst.buffer.size,					// VkDeviceSize			size;
			VK_BUFFER_USAGE_TRANSFER_DST_BIT,			// VkBufferUsageFlags	usage;
			VK_SHARING_MODE_EXCLUSIVE,					// VkSharingMode		sharingMode;
			1u,											// deUint32				queueFamilyIndexCount;
			&queueFamilyIndex,							// const deUint32*		pQueueFamilyIndices;
		};

		m_destination				= createBuffer(vk, vkDevice, &destinationBufferParams);
		m_destinationBufferAlloc	= allocateBuffer(vki, vk, vkPhysDevice, vkDevice, *m_destination, MemoryRequirement::HostVisible, memAlloc, m_params.allocationKind);
		VK_CHECK(vk.bindBufferMemory(vkDevice, *m_destination, m_destinationBufferAlloc->getMemory(), m_destinationBufferAlloc->getOffset()));
	}
}

tcu::TestStatus CopyBufferToBuffer::iterate (void)
{
	const int srcLevelWidth		= (int)(m_params.src.buffer.size/4); // Here the format is VK_FORMAT_R32_UINT, we need to divide the buffer size by 4
	m_sourceTextureLevel		= de::MovePtr<tcu::TextureLevel>(new tcu::TextureLevel(mapVkFormat(VK_FORMAT_R32_UINT), srcLevelWidth, 1));
	generateBuffer(m_sourceTextureLevel->getAccess(), srcLevelWidth, 1, 1, FILL_MODE_RED);

	const int dstLevelWidth		= (int)(m_params.dst.buffer.size/4);
	m_destinationTextureLevel	= de::MovePtr<tcu::TextureLevel>(new tcu::TextureLevel(mapVkFormat(VK_FORMAT_R32_UINT), dstLevelWidth, 1));
	generateBuffer(m_destinationTextureLevel->getAccess(), dstLevelWidth, 1, 1, FILL_MODE_WHITE);

	generateExpectedResult();

	uploadBuffer(m_sourceTextureLevel->getAccess(), *m_sourceBufferAlloc);
	uploadBuffer(m_destinationTextureLevel->getAccess(), *m_destinationBufferAlloc);

	const DeviceInterface&		vk			= m_context.getDeviceInterface();
	const VkDevice				vkDevice	= m_context.getDevice();
	const VkQueue				queue		= m_context.getUniversalQueue();

	const VkBufferMemoryBarrier		srcBufferBarrier	=
	{
		VK_STRUCTURE_TYPE_BUFFER_MEMORY_BARRIER,	// VkStructureType	sType;
		DE_NULL,									// const void*		pNext;
		VK_ACCESS_HOST_WRITE_BIT,					// VkAccessFlags	srcAccessMask;
		VK_ACCESS_TRANSFER_READ_BIT,				// VkAccessFlags	dstAccessMask;
		VK_QUEUE_FAMILY_IGNORED,					// deUint32			srcQueueFamilyIndex;
		VK_QUEUE_FAMILY_IGNORED,					// deUint32			dstQueueFamilyIndex;
		*m_source,									// VkBuffer			buffer;
		0u,											// VkDeviceSize		offset;
		m_params.src.buffer.size					// VkDeviceSize		size;
	};

	const VkBufferMemoryBarrier		dstBufferBarrier	=
	{
		VK_STRUCTURE_TYPE_BUFFER_MEMORY_BARRIER,	// VkStructureType	sType;
		DE_NULL,									// const void*		pNext;
		VK_ACCESS_TRANSFER_WRITE_BIT,				// VkAccessFlags	srcAccessMask;
		VK_ACCESS_HOST_READ_BIT,					// VkAccessFlags	dstAccessMask;
		VK_QUEUE_FAMILY_IGNORED,					// deUint32			srcQueueFamilyIndex;
		VK_QUEUE_FAMILY_IGNORED,					// deUint32			dstQueueFamilyIndex;
		*m_destination,								// VkBuffer			buffer;
		0u,											// VkDeviceSize		offset;
		m_params.dst.buffer.size					// VkDeviceSize		size;
	};

	std::vector<VkBufferCopy>		bufferCopies;
	for (deUint32 i = 0; i < m_params.regions.size(); i++)
		bufferCopies.push_back(m_params.regions[i].bufferCopy);

	beginCommandBuffer(vk, *m_cmdBuffer);
	vk.cmdPipelineBarrier(*m_cmdBuffer, VK_PIPELINE_STAGE_HOST_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT, (VkDependencyFlags)0, 0, (const VkMemoryBarrier*)DE_NULL, 1, &srcBufferBarrier, 0, (const VkImageMemoryBarrier*)DE_NULL);
	vk.cmdCopyBuffer(*m_cmdBuffer, m_source.get(), m_destination.get(), (deUint32)m_params.regions.size(), &bufferCopies[0]);
	vk.cmdPipelineBarrier(*m_cmdBuffer, VK_PIPELINE_STAGE_TRANSFER_BIT, VK_PIPELINE_STAGE_HOST_BIT, (VkDependencyFlags)0, 0, (const VkMemoryBarrier*)DE_NULL, 1, &dstBufferBarrier, 0, (const VkImageMemoryBarrier*)DE_NULL);
	endCommandBuffer(vk, *m_cmdBuffer);
	submitCommandsAndWait(vk, vkDevice, queue, *m_cmdBuffer);

	// Read buffer data
	de::MovePtr<tcu::TextureLevel>	resultLevel		(new tcu::TextureLevel(mapVkFormat(VK_FORMAT_R32_UINT), dstLevelWidth, 1));
	invalidateAlloc(vk, vkDevice, *m_destinationBufferAlloc);
	tcu::copy(*resultLevel, tcu::ConstPixelBufferAccess(resultLevel->getFormat(), resultLevel->getSize(), m_destinationBufferAlloc->getHostPtr()));

	return checkTestResult(resultLevel->getAccess());
}

void CopyBufferToBuffer::copyRegionToTextureLevel (tcu::ConstPixelBufferAccess src, tcu::PixelBufferAccess dst, CopyRegion region, deUint32 mipLevel)
{
	DE_UNREF(mipLevel);

	deMemcpy((deUint8*) dst.getDataPtr() + region.bufferCopy.dstOffset,
			 (deUint8*) src.getDataPtr() + region.bufferCopy.srcOffset,
			 (size_t)region.bufferCopy.size);
}

class BufferToBufferTestCase : public vkt::TestCase
{
public:
							BufferToBufferTestCase	(tcu::TestContext&	testCtx,
													 const std::string&	name,
													 const std::string&	description,
													 const TestParams	params)
								: vkt::TestCase	(testCtx, name, description)
								, m_params		(params)
							{}

	virtual TestInstance*	createInstance			(Context& context) const
							{
								return new CopyBufferToBuffer(context, m_params);
							}
private:
	TestParams				m_params;
};

// Copy from image to buffer.

class CopyImageToBuffer : public CopiesAndBlittingTestInstance
{
public:
								CopyImageToBuffer			(Context&	context,
															 TestParams	testParams);
	virtual tcu::TestStatus		iterate						(void);
private:
	virtual void				copyRegionToTextureLevel	(tcu::ConstPixelBufferAccess src, tcu::PixelBufferAccess dst, CopyRegion region, deUint32 mipLevel = 0u);

	tcu::TextureFormat			m_textureFormat;
	VkDeviceSize				m_bufferSize;

	Move<VkImage>				m_source;
	de::MovePtr<Allocation>		m_sourceImageAlloc;
	Move<VkBuffer>				m_destination;
	de::MovePtr<Allocation>		m_destinationBufferAlloc;
};

CopyImageToBuffer::CopyImageToBuffer (Context& context, TestParams testParams)
	: CopiesAndBlittingTestInstance(context, testParams)
	, m_textureFormat(mapVkFormat(testParams.src.image.format))
	, m_bufferSize(m_params.dst.buffer.size * tcu::getPixelSize(m_textureFormat))
{
	const InstanceInterface&	vki					= context.getInstanceInterface();
	const DeviceInterface&		vk					= context.getDeviceInterface();
	const VkPhysicalDevice		vkPhysDevice		= context.getPhysicalDevice();
	const VkDevice				vkDevice			= context.getDevice();
	const deUint32				queueFamilyIndex	= context.getUniversalQueueFamilyIndex();
	Allocator&					memAlloc			= context.getDefaultAllocator();

	// Create source image
	{
		const VkImageCreateInfo		sourceImageParams		=
		{
			VK_STRUCTURE_TYPE_IMAGE_CREATE_INFO,	// VkStructureType		sType;
			DE_NULL,								// const void*			pNext;
			getCreateFlags(m_params.src.image),		// VkImageCreateFlags	flags;
			m_params.src.image.imageType,			// VkImageType			imageType;
			m_params.src.image.format,				// VkFormat				format;
			getExtent3D(m_params.src.image),		// VkExtent3D			extent;
			1u,										// deUint32				mipLevels;
			getArraySize(m_params.src.image),		// deUint32				arraySize;
			VK_SAMPLE_COUNT_1_BIT,					// deUint32				samples;
			VK_IMAGE_TILING_OPTIMAL,				// VkImageTiling		tiling;
			VK_IMAGE_USAGE_TRANSFER_SRC_BIT |
				VK_IMAGE_USAGE_TRANSFER_DST_BIT,	// VkImageUsageFlags	usage;
			VK_SHARING_MODE_EXCLUSIVE,				// VkSharingMode		sharingMode;
			1u,										// deUint32				queueFamilyCount;
			&queueFamilyIndex,						// const deUint32*		pQueueFamilyIndices;
			VK_IMAGE_LAYOUT_UNDEFINED,				// VkImageLayout		initialLayout;
		};

		m_source			= createImage(vk, vkDevice, &sourceImageParams);
		m_sourceImageAlloc	= allocateImage(vki, vk, vkPhysDevice, vkDevice, *m_source, MemoryRequirement::Any, memAlloc, m_params.allocationKind);
		VK_CHECK(vk.bindImageMemory(vkDevice, *m_source, m_sourceImageAlloc->getMemory(), m_sourceImageAlloc->getOffset()));
	}

	// Create destination buffer
	{
		const VkBufferCreateInfo	destinationBufferParams	=
		{
			VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO,		// VkStructureType		sType;
			DE_NULL,									// const void*			pNext;
			0u,											// VkBufferCreateFlags	flags;
			m_bufferSize,								// VkDeviceSize			size;
			VK_BUFFER_USAGE_TRANSFER_DST_BIT,			// VkBufferUsageFlags	usage;
			VK_SHARING_MODE_EXCLUSIVE,					// VkSharingMode		sharingMode;
			1u,											// deUint32				queueFamilyIndexCount;
			&queueFamilyIndex,							// const deUint32*		pQueueFamilyIndices;
		};

		m_destination				= createBuffer(vk, vkDevice, &destinationBufferParams);
		m_destinationBufferAlloc	= allocateBuffer(vki, vk, vkPhysDevice, vkDevice, *m_destination, MemoryRequirement::HostVisible, memAlloc, m_params.allocationKind);
		VK_CHECK(vk.bindBufferMemory(vkDevice, *m_destination, m_destinationBufferAlloc->getMemory(), m_destinationBufferAlloc->getOffset()));
	}
}

tcu::TestStatus CopyImageToBuffer::iterate (void)
{
	m_sourceTextureLevel = de::MovePtr<tcu::TextureLevel>(new tcu::TextureLevel(m_textureFormat,
																				m_params.src.image.extent.width,
																				m_params.src.image.extent.height,
																				m_params.src.image.extent.depth));
	generateBuffer(m_sourceTextureLevel->getAccess(), m_params.src.image.extent.width, m_params.src.image.extent.height, m_params.src.image.extent.depth);
	m_destinationTextureLevel = de::MovePtr<tcu::TextureLevel>(new tcu::TextureLevel(m_textureFormat, (int)m_params.dst.buffer.size, 1));
	generateBuffer(m_destinationTextureLevel->getAccess(), (int)m_params.dst.buffer.size, 1, 1);

	generateExpectedResult();

	uploadImage(m_sourceTextureLevel->getAccess(), *m_source, m_params.src.image);
	uploadBuffer(m_destinationTextureLevel->getAccess(), *m_destinationBufferAlloc);

	const DeviceInterface&		vk			= m_context.getDeviceInterface();
	const VkDevice				vkDevice	= m_context.getDevice();
	const VkQueue				queue		= m_context.getUniversalQueue();

	// Barriers for copying image to buffer
	const VkImageMemoryBarrier		imageBarrier		=
	{
		VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER,		// VkStructureType			sType;
		DE_NULL,									// const void*				pNext;
		VK_ACCESS_TRANSFER_WRITE_BIT,				// VkAccessFlags			srcAccessMask;
		VK_ACCESS_TRANSFER_READ_BIT,				// VkAccessFlags			dstAccessMask;
		VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL,		// VkImageLayout			oldLayout;
		VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL,		// VkImageLayout			newLayout;
		VK_QUEUE_FAMILY_IGNORED,					// deUint32					srcQueueFamilyIndex;
		VK_QUEUE_FAMILY_IGNORED,					// deUint32					dstQueueFamilyIndex;
		*m_source,									// VkImage					image;
		{											// VkImageSubresourceRange	subresourceRange;
			getAspectFlags(m_textureFormat),	// VkImageAspectFlags	aspectMask;
			0u,								// deUint32				baseMipLevel;
			1u,								// deUint32				mipLevels;
			0u,								// deUint32				baseArraySlice;
			1u								// deUint32				arraySize;
		}
	};

	const VkBufferMemoryBarrier		bufferBarrier		=
	{
		VK_STRUCTURE_TYPE_BUFFER_MEMORY_BARRIER,	// VkStructureType	sType;
		DE_NULL,									// const void*		pNext;
		VK_ACCESS_TRANSFER_WRITE_BIT,				// VkAccessFlags	srcAccessMask;
		VK_ACCESS_HOST_READ_BIT,					// VkAccessFlags	dstAccessMask;
		VK_QUEUE_FAMILY_IGNORED,					// deUint32			srcQueueFamilyIndex;
		VK_QUEUE_FAMILY_IGNORED,					// deUint32			dstQueueFamilyIndex;
		*m_destination,								// VkBuffer			buffer;
		0u,											// VkDeviceSize		offset;
		m_bufferSize								// VkDeviceSize		size;
	};

	// Copy from image to buffer
	std::vector<VkBufferImageCopy>	bufferImageCopies;
	for (deUint32 i = 0; i < m_params.regions.size(); i++)
		bufferImageCopies.push_back(m_params.regions[i].bufferImageCopy);

	beginCommandBuffer(vk, *m_cmdBuffer);
	vk.cmdPipelineBarrier(*m_cmdBuffer, VK_PIPELINE_STAGE_TRANSFER_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT, (VkDependencyFlags)0, 0, (const VkMemoryBarrier*)DE_NULL, 0, (const VkBufferMemoryBarrier*)DE_NULL, 1, &imageBarrier);
	vk.cmdCopyImageToBuffer(*m_cmdBuffer, m_source.get(), VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL, m_destination.get(), (deUint32)m_params.regions.size(), &bufferImageCopies[0]);
	vk.cmdPipelineBarrier(*m_cmdBuffer, VK_PIPELINE_STAGE_TRANSFER_BIT, VK_PIPELINE_STAGE_HOST_BIT, (VkDependencyFlags)0, 0, (const VkMemoryBarrier*)DE_NULL, 1, &bufferBarrier, 0, (const VkImageMemoryBarrier*)DE_NULL);
	endCommandBuffer(vk, *m_cmdBuffer);

	submitCommandsAndWait (vk, vkDevice, queue, *m_cmdBuffer);

	// Read buffer data
	de::MovePtr<tcu::TextureLevel>	resultLevel		(new tcu::TextureLevel(m_textureFormat, (int)m_params.dst.buffer.size, 1));
	invalidateAlloc(vk, vkDevice, *m_destinationBufferAlloc);
	tcu::copy(*resultLevel, tcu::ConstPixelBufferAccess(resultLevel->getFormat(), resultLevel->getSize(), m_destinationBufferAlloc->getHostPtr()));

	return checkTestResult(resultLevel->getAccess());
}

class CopyImageToBufferTestCase : public vkt::TestCase
{
public:
							CopyImageToBufferTestCase	(tcu::TestContext&		testCtx,
														 const std::string&		name,
														 const std::string&		description,
														 const TestParams		params)
								: vkt::TestCase	(testCtx, name, description)
								, m_params		(params)
							{}

	virtual TestInstance*	createInstance				(Context&				context) const
							{
								return new CopyImageToBuffer(context, m_params);
							}
private:
	TestParams				m_params;
};

void CopyImageToBuffer::copyRegionToTextureLevel (tcu::ConstPixelBufferAccess src, tcu::PixelBufferAccess dst, CopyRegion region, deUint32 mipLevel)
{
	DE_UNREF(mipLevel);

	deUint32			rowLength	= region.bufferImageCopy.bufferRowLength;
	if (!rowLength)
		rowLength = region.bufferImageCopy.imageExtent.width;

	deUint32			imageHeight	= region.bufferImageCopy.bufferImageHeight;
	if (!imageHeight)
		imageHeight = region.bufferImageCopy.imageExtent.height;

	const int			texelSize	= src.getFormat().getPixelSize();
	const VkExtent3D	extent		= region.bufferImageCopy.imageExtent;
	const VkOffset3D	srcOffset	= region.bufferImageCopy.imageOffset;
	const int			texelOffset	= (int) region.bufferImageCopy.bufferOffset / texelSize;

	for (deUint32 z = 0; z < extent.depth; z++)
	{
		for (deUint32 y = 0; y < extent.height; y++)
		{
			int									texelIndex		= texelOffset + (z * imageHeight + y) *	rowLength;
			const tcu::ConstPixelBufferAccess	srcSubRegion	= tcu::getSubregion(src, srcOffset.x, srcOffset.y + y, srcOffset.z + z,
																					region.bufferImageCopy.imageExtent.width, 1, 1);
			const tcu::PixelBufferAccess		dstSubRegion	= tcu::getSubregion(dst, texelIndex, 0, region.bufferImageCopy.imageExtent.width, 1);
			tcu::copy(dstSubRegion, srcSubRegion);
		}
	}
}

// Copy from buffer to image.

class CopyBufferToImage : public CopiesAndBlittingTestInstance
{
public:
								CopyBufferToImage			(Context&	context,
															 TestParams	testParams);
	virtual tcu::TestStatus		iterate						(void);

private:
	virtual void				copyRegionToTextureLevel	(tcu::ConstPixelBufferAccess src, tcu::PixelBufferAccess dst, CopyRegion region, deUint32 mipLevel = 0u);

	tcu::TextureFormat			m_textureFormat;
	VkDeviceSize				m_bufferSize;

	Move<VkBuffer>				m_source;
	de::MovePtr<Allocation>		m_sourceBufferAlloc;
	Move<VkImage>				m_destination;
	de::MovePtr<Allocation>		m_destinationImageAlloc;
};

CopyBufferToImage::CopyBufferToImage (Context& context, TestParams testParams)
	: CopiesAndBlittingTestInstance(context, testParams)
	, m_textureFormat(mapVkFormat(testParams.dst.image.format))
	, m_bufferSize(m_params.src.buffer.size * tcu::getPixelSize(m_textureFormat))
{
	const InstanceInterface&	vki					= context.getInstanceInterface();
	const DeviceInterface&		vk					= context.getDeviceInterface();
	const VkPhysicalDevice		vkPhysDevice		= context.getPhysicalDevice();
	const VkDevice				vkDevice			= context.getDevice();
	const deUint32				queueFamilyIndex	= context.getUniversalQueueFamilyIndex();
	Allocator&					memAlloc			= context.getDefaultAllocator();

	// Create source buffer
	{
		const VkBufferCreateInfo	sourceBufferParams		=
		{
			VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO,		// VkStructureType		sType;
			DE_NULL,									// const void*			pNext;
			0u,											// VkBufferCreateFlags	flags;
			m_bufferSize,								// VkDeviceSize			size;
			VK_BUFFER_USAGE_TRANSFER_SRC_BIT,			// VkBufferUsageFlags	usage;
			VK_SHARING_MODE_EXCLUSIVE,					// VkSharingMode		sharingMode;
			1u,											// deUint32				queueFamilyIndexCount;
			&queueFamilyIndex,							// const deUint32*		pQueueFamilyIndices;
		};

		m_source				= createBuffer(vk, vkDevice, &sourceBufferParams);
		m_sourceBufferAlloc		= allocateBuffer(vki, vk, vkPhysDevice, vkDevice, *m_source, MemoryRequirement::HostVisible, memAlloc, m_params.allocationKind);
		VK_CHECK(vk.bindBufferMemory(vkDevice, *m_source, m_sourceBufferAlloc->getMemory(), m_sourceBufferAlloc->getOffset()));
	}

	// Create destination image
	{
		const VkImageCreateInfo		destinationImageParams	=
		{
			VK_STRUCTURE_TYPE_IMAGE_CREATE_INFO,	// VkStructureType		sType;
			DE_NULL,								// const void*			pNext;
			getCreateFlags(m_params.dst.image),		// VkImageCreateFlags	flags;
			m_params.dst.image.imageType,			// VkImageType			imageType;
			m_params.dst.image.format,				// VkFormat				format;
			getExtent3D(m_params.dst.image),		// VkExtent3D			extent;
			1u,										// deUint32				mipLevels;
			getArraySize(m_params.dst.image),		// deUint32				arraySize;
			VK_SAMPLE_COUNT_1_BIT,					// deUint32				samples;
			VK_IMAGE_TILING_OPTIMAL,				// VkImageTiling		tiling;
			VK_IMAGE_USAGE_TRANSFER_SRC_BIT |
				VK_IMAGE_USAGE_TRANSFER_DST_BIT,	// VkImageUsageFlags	usage;
			VK_SHARING_MODE_EXCLUSIVE,				// VkSharingMode		sharingMode;
			1u,										// deUint32				queueFamilyCount;
			&queueFamilyIndex,						// const deUint32*		pQueueFamilyIndices;
			VK_IMAGE_LAYOUT_UNDEFINED,				// VkImageLayout		initialLayout;
		};

		m_destination			= createImage(vk, vkDevice, &destinationImageParams);
		m_destinationImageAlloc	= allocateImage(vki, vk, vkPhysDevice, vkDevice, *m_destination, MemoryRequirement::Any, memAlloc, m_params.allocationKind);
		VK_CHECK(vk.bindImageMemory(vkDevice, *m_destination, m_destinationImageAlloc->getMemory(), m_destinationImageAlloc->getOffset()));
	}
}

tcu::TestStatus CopyBufferToImage::iterate (void)
{
	m_sourceTextureLevel = de::MovePtr<tcu::TextureLevel>(new tcu::TextureLevel(m_textureFormat, (int)m_params.src.buffer.size, 1));
	generateBuffer(m_sourceTextureLevel->getAccess(), (int)m_params.src.buffer.size, 1, 1);
	m_destinationTextureLevel = de::MovePtr<tcu::TextureLevel>(new tcu::TextureLevel(m_textureFormat,
																					m_params.dst.image.extent.width,
																					m_params.dst.image.extent.height,
																					m_params.dst.image.extent.depth));

	generateBuffer(m_destinationTextureLevel->getAccess(), m_params.dst.image.extent.width, m_params.dst.image.extent.height, m_params.dst.image.extent.depth);

	generateExpectedResult();

	uploadBuffer(m_sourceTextureLevel->getAccess(), *m_sourceBufferAlloc);
	uploadImage(m_destinationTextureLevel->getAccess(), *m_destination, m_params.dst.image);

	const DeviceInterface&		vk			= m_context.getDeviceInterface();
	const VkDevice				vkDevice	= m_context.getDevice();
	const VkQueue				queue		= m_context.getUniversalQueue();

	const VkImageMemoryBarrier	imageBarrier	=
	{
		VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER,		// VkStructureType			sType;
		DE_NULL,									// const void*				pNext;
		VK_ACCESS_TRANSFER_WRITE_BIT,				// VkAccessFlags			srcAccessMask;
		VK_ACCESS_TRANSFER_WRITE_BIT,				// VkAccessFlags			dstAccessMask;
		VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL,		// VkImageLayout			oldLayout;
		VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL,		// VkImageLayout			newLayout;
		VK_QUEUE_FAMILY_IGNORED,					// deUint32					srcQueueFamilyIndex;
		VK_QUEUE_FAMILY_IGNORED,					// deUint32					dstQueueFamilyIndex;
		*m_destination,								// VkImage					image;
		{											// VkImageSubresourceRange	subresourceRange;
			getAspectFlags(m_textureFormat),	// VkImageAspectFlags	aspectMask;
			0u,								// deUint32				baseMipLevel;
			1u,								// deUint32				mipLevels;
			0u,								// deUint32				baseArraySlice;
			1u								// deUint32				arraySize;
		}
	};

	// Copy from buffer to image
	std::vector<VkBufferImageCopy>		bufferImageCopies;
	for (deUint32 i = 0; i < m_params.regions.size(); i++)
		bufferImageCopies.push_back(m_params.regions[i].bufferImageCopy);

	beginCommandBuffer(vk, *m_cmdBuffer);
	vk.cmdPipelineBarrier(*m_cmdBuffer, VK_PIPELINE_STAGE_TRANSFER_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT, (VkDependencyFlags)0, 0, (const VkMemoryBarrier*)DE_NULL, 0, (const VkBufferMemoryBarrier*)DE_NULL, 1, &imageBarrier);
	vk.cmdCopyBufferToImage(*m_cmdBuffer, m_source.get(), m_destination.get(), VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, (deUint32)m_params.regions.size(), bufferImageCopies.data());
	endCommandBuffer(vk, *m_cmdBuffer);

	submitCommandsAndWait (vk, vkDevice, queue, *m_cmdBuffer);

	de::MovePtr<tcu::TextureLevel>	resultLevel	= readImage(*m_destination, m_params.dst.image);

	return checkTestResult(resultLevel->getAccess());
}

class CopyBufferToImageTestCase : public vkt::TestCase
{
public:
							CopyBufferToImageTestCase	(tcu::TestContext&		testCtx,
														 const std::string&		name,
														 const std::string&		description,
														 const TestParams		params)
								: vkt::TestCase	(testCtx, name, description)
								, m_params		(params)
							{}

	virtual					~CopyBufferToImageTestCase	(void) {}

	virtual TestInstance*	createInstance				(Context&				context) const
							{
								return new CopyBufferToImage(context, m_params);
							}
private:
	TestParams				m_params;
};

void CopyBufferToImage::copyRegionToTextureLevel (tcu::ConstPixelBufferAccess src, tcu::PixelBufferAccess dst, CopyRegion region, deUint32 mipLevel)
{
	DE_UNREF(mipLevel);

	deUint32			rowLength	= region.bufferImageCopy.bufferRowLength;
	if (!rowLength)
		rowLength = region.bufferImageCopy.imageExtent.width;

	deUint32			imageHeight	= region.bufferImageCopy.bufferImageHeight;
	if (!imageHeight)
		imageHeight = region.bufferImageCopy.imageExtent.height;

	const int			texelSize	= dst.getFormat().getPixelSize();
	const VkExtent3D	extent		= region.bufferImageCopy.imageExtent;
	const VkOffset3D	dstOffset	= region.bufferImageCopy.imageOffset;
	const int			texelOffset	= (int) region.bufferImageCopy.bufferOffset / texelSize;

	for (deUint32 z = 0; z < extent.depth; z++)
	{
		for (deUint32 y = 0; y < extent.height; y++)
		{
			int									texelIndex		= texelOffset + (z * imageHeight + y) *	rowLength;
			const tcu::ConstPixelBufferAccess	srcSubRegion	= tcu::getSubregion(src, texelIndex, 0, region.bufferImageCopy.imageExtent.width, 1);
			const tcu::PixelBufferAccess		dstSubRegion	= tcu::getSubregion(dst, dstOffset.x, dstOffset.y + y, dstOffset.z + z,
																					region.bufferImageCopy.imageExtent.width, 1, 1);
			tcu::copy(dstSubRegion, srcSubRegion);
		}
	}
}

class CopyBufferToDepthStencil : public CopiesAndBlittingTestInstance
{
public:
								CopyBufferToDepthStencil	(Context& context,
															 TestParams	testParams);
	virtual tcu::TestStatus		iterate						(void);
private:
	virtual void				copyRegionToTextureLevel	(tcu::ConstPixelBufferAccess src, tcu::PixelBufferAccess dst, CopyRegion region, deUint32 mipLevel = 0u);

	tcu::TextureFormat			m_textureFormat;
	VkDeviceSize				m_bufferSize;

	Move<VkBuffer>				m_source;
	de::MovePtr<Allocation>		m_sourceBufferAlloc;
	Move<VkImage>				m_destination;
	de::MovePtr<Allocation>		m_destinationImageAlloc;
};

void CopyBufferToDepthStencil::copyRegionToTextureLevel(tcu::ConstPixelBufferAccess src, tcu::PixelBufferAccess dst, CopyRegion region, deUint32 mipLevel)
{
	DE_UNREF(mipLevel);

	deUint32			rowLength	= region.bufferImageCopy.bufferRowLength;
	if (!rowLength)
		rowLength = region.bufferImageCopy.imageExtent.width;

	deUint32			imageHeight = region.bufferImageCopy.bufferImageHeight;
	if (!imageHeight)
		imageHeight = region.bufferImageCopy.imageExtent.height;

	const int			texelSize	= dst.getFormat().getPixelSize();
	const VkExtent3D	extent		= region.bufferImageCopy.imageExtent;
	const VkOffset3D	dstOffset	= region.bufferImageCopy.imageOffset;
	const int			texelOffset = (int)region.bufferImageCopy.bufferOffset / texelSize;

	for (deUint32 z = 0; z < extent.depth; z++)
	{
		for (deUint32 y = 0; y < extent.height; y++)
		{
			int									texelIndex = texelOffset + (z * imageHeight + y) * rowLength;
			const tcu::ConstPixelBufferAccess	srcSubRegion = tcu::getSubregion(src, texelIndex, 0, region.bufferImageCopy.imageExtent.width, 1);
			const tcu::PixelBufferAccess		dstSubRegion = tcu::getSubregion(dst, dstOffset.x, dstOffset.y + y, dstOffset.z + z,
				region.bufferImageCopy.imageExtent.width, 1, 1);

			if (region.bufferImageCopy.imageSubresource.aspectMask == VK_IMAGE_ASPECT_DEPTH_BIT)
			{
				tcu::copy(dstSubRegion, tcu::getEffectiveDepthStencilAccess(srcSubRegion, tcu::Sampler::MODE_DEPTH), DE_FALSE);
			}
			else
			{
				tcu::copy(dstSubRegion, tcu::getEffectiveDepthStencilAccess(srcSubRegion, tcu::Sampler::MODE_STENCIL), DE_FALSE);
			}
		}
	}
}

bool isSupportedDepthStencilFormat(const InstanceInterface& vki, const VkPhysicalDevice physDevice, const VkFormat format)
{
	VkFormatProperties formatProps;
	vki.getPhysicalDeviceFormatProperties(physDevice, format, &formatProps);
	return (formatProps.optimalTilingFeatures & VK_FORMAT_FEATURE_DEPTH_STENCIL_ATTACHMENT_BIT) != 0;
}

CopyBufferToDepthStencil::CopyBufferToDepthStencil(Context& context, TestParams testParams)
	: CopiesAndBlittingTestInstance(context, testParams)
	, m_textureFormat(mapVkFormat(testParams.dst.image.format))
	, m_bufferSize(0)
{
	const InstanceInterface&	vki					= context.getInstanceInterface();
	const DeviceInterface&		vk					= context.getDeviceInterface();
	const VkPhysicalDevice		vkPhysDevice		= context.getPhysicalDevice();
	const VkDevice				vkDevice			= context.getDevice();
	const deUint32				queueFamilyIndex	= context.getUniversalQueueFamilyIndex();
	Allocator&					memAlloc			= context.getDefaultAllocator();
	const bool					hasDepth			= tcu::hasDepthComponent(mapVkFormat(m_params.dst.image.format).order);
	const bool					hasStencil			= tcu::hasStencilComponent(mapVkFormat(m_params.dst.image.format).order);

	if (!isSupportedDepthStencilFormat(vki, vkPhysDevice, testParams.dst.image.format))
	{
		TCU_THROW(NotSupportedError, "Image format not supported.");
	}

	if (hasDepth)
	{
		glw::GLuint texelSize = m_textureFormat.getPixelSize();
		if (texelSize > sizeof(float))
		{
			// We must have D32F_S8 format, depth must be packed so we only need
			// to allocate space for the D32F part. Stencil will be separate
			texelSize = sizeof(float);
		}
		m_bufferSize += static_cast<VkDeviceSize>(m_params.dst.image.extent.width) * static_cast<VkDeviceSize>(m_params.dst.image.extent.height) * static_cast<VkDeviceSize>(texelSize);
	}
	if (hasStencil)
	{
		// Stencil is always 8bits and packed.
		m_bufferSize += static_cast<VkDeviceSize>(m_params.dst.image.extent.width) * static_cast<VkDeviceSize>(m_params.dst.image.extent.height);
	}

	// Create source buffer, this is where the depth & stencil data will go that's used by test's regions.
	{
		const VkBufferCreateInfo	sourceBufferParams		=
		{
			VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO,		// VkStructureType		sType;
			DE_NULL,									// const void*			pNext;
			0u,											// VkBufferCreateFlags	flags;
			m_bufferSize,								// VkDeviceSize			size;
			VK_BUFFER_USAGE_TRANSFER_SRC_BIT,			// VkBufferUsageFlags	usage;
			VK_SHARING_MODE_EXCLUSIVE,					// VkSharingMode		sharingMode;
			1u,											// deUint32				queueFamilyIndexCount;
			&queueFamilyIndex,							// const deUint32*		pQueueFamilyIndices;
		};

		m_source				= createBuffer(vk, vkDevice, &sourceBufferParams);
		m_sourceBufferAlloc		= allocateBuffer(vki, vk, vkPhysDevice, vkDevice, *m_source, MemoryRequirement::HostVisible, memAlloc, m_params.allocationKind);
		VK_CHECK(vk.bindBufferMemory(vkDevice, *m_source, m_sourceBufferAlloc->getMemory(), m_sourceBufferAlloc->getOffset()));
	}

	// Create destination image
	{
		const VkImageCreateInfo		destinationImageParams	=
		{
			VK_STRUCTURE_TYPE_IMAGE_CREATE_INFO,	// VkStructureType		sType;
			DE_NULL,								// const void*			pNext;
			getCreateFlags(m_params.dst.image),		// VkImageCreateFlags	flags;
			m_params.dst.image.imageType,			// VkImageType			imageType;
			m_params.dst.image.format,				// VkFormat				format;
			getExtent3D(m_params.dst.image),		// VkExtent3D			extent;
			1u,										// deUint32				mipLevels;
			getArraySize(m_params.dst.image),		// deUint32				arraySize;
			VK_SAMPLE_COUNT_1_BIT,					// deUint32				samples;
			VK_IMAGE_TILING_OPTIMAL,				// VkImageTiling		tiling;
			VK_IMAGE_USAGE_TRANSFER_SRC_BIT |
				VK_IMAGE_USAGE_TRANSFER_DST_BIT,	// VkImageUsageFlags	usage;
			VK_SHARING_MODE_EXCLUSIVE,				// VkSharingMode		sharingMode;
			1u,										// deUint32				queueFamilyCount;
			&queueFamilyIndex,						// const deUint32*		pQueueFamilyIndices;
			VK_IMAGE_LAYOUT_UNDEFINED,				// VkImageLayout		initialLayout;
		};

		m_destination				= createImage(vk, vkDevice, &destinationImageParams);
		m_destinationImageAlloc		= allocateImage(vki, vk, vkPhysDevice, vkDevice, *m_destination, MemoryRequirement::Any, memAlloc, m_params.allocationKind);
		VK_CHECK(vk.bindImageMemory(vkDevice, *m_destination, m_destinationImageAlloc->getMemory(), m_destinationImageAlloc->getOffset()));
	}
}

tcu::TestStatus CopyBufferToDepthStencil::iterate(void)
{
	// Create source depth/stencil content. Treat as 1D texture to get different pattern
	m_sourceTextureLevel = de::MovePtr<tcu::TextureLevel>(new tcu::TextureLevel(m_textureFormat, (int)m_params.src.buffer.size, 1));
	// Fill buffer with linear gradiant
	generateBuffer(m_sourceTextureLevel->getAccess(), (int)m_params.src.buffer.size, 1, 1);

	// Create image layer for depth/stencil
	m_destinationTextureLevel = de::MovePtr<tcu::TextureLevel>(new tcu::TextureLevel(m_textureFormat,
		m_params.dst.image.extent.width,
		m_params.dst.image.extent.height,
		m_params.dst.image.extent.depth));

	// Fill image layer with 2D gradiant
	generateBuffer(m_destinationTextureLevel->getAccess(), m_params.dst.image.extent.width, m_params.dst.image.extent.height, m_params.dst.image.extent.depth);

	// Fill m_extendedTextureLevel with copy of m_destinationTextureLevel
	// Then iterate over each of the regions given in m_params.regions and copy m_sourceTextureLevel content to m_extendedTextureLevel
	// This emulates what the HW will be doing.
	generateExpectedResult();

	// Upload our source depth/stencil content to the source buffer
	// This is the buffer that will be used by region commands
	std::vector<VkBufferImageCopy>	bufferImageCopies;
	VkDeviceSize					bufferOffset	= 0;
	const VkDevice					vkDevice		= m_context.getDevice();
	const DeviceInterface&			vk				= m_context.getDeviceInterface();
	const VkQueue					queue			= m_context.getUniversalQueue();
	char*							dstPtr			= reinterpret_cast<char*>(m_sourceBufferAlloc->getHostPtr());
	bool							depthLoaded		= DE_FALSE;
	bool							stencilLoaded	= DE_FALSE;
	VkDeviceSize					depthOffset		= 0;
	VkDeviceSize					stencilOffset	= 0;

	// To be able to test ordering depth & stencil differently
	// we take the given copy regions and use that as the desired order
	// and copy the appropriate data into place and compute the appropriate
	// data offsets to be used in the copy command.
	for (deUint32 i = 0; i < m_params.regions.size(); i++)
	{
		tcu::ConstPixelBufferAccess bufferAccess	= m_sourceTextureLevel->getAccess();
		deUint32					bufferSize		= bufferAccess.getWidth() * bufferAccess.getHeight() * bufferAccess.getDepth();
		VkBufferImageCopy			copyData		= m_params.regions[i].bufferImageCopy;
		char*						srcPtr;

		if (copyData.imageSubresource.aspectMask == VK_IMAGE_ASPECT_DEPTH_BIT && !depthLoaded)
		{
			// Create level that is same component as depth buffer (e.g. D16, D24, D32F)
			tcu::TextureLevel	depthTexture(mapCombinedToDepthTransferFormat(bufferAccess.getFormat()), bufferAccess.getWidth(), bufferAccess.getHeight(), bufferAccess.getDepth());
			bufferSize *= tcu::getPixelSize(depthTexture.getFormat());
			// Copy depth component only from source data. This gives us packed depth-only data.
			tcu::copy(depthTexture.getAccess(), tcu::getEffectiveDepthStencilAccess(bufferAccess, tcu::Sampler::MODE_DEPTH));
			srcPtr = (char*)depthTexture.getAccess().getDataPtr();
			// Copy packed depth-only data to output buffer
			deMemcpy(dstPtr, srcPtr, bufferSize);
			depthLoaded = DE_TRUE;
			depthOffset = bufferOffset;
			dstPtr += bufferSize;
			bufferOffset += bufferSize;
			copyData.bufferOffset += depthOffset;
		}
		else if (!stencilLoaded)
		{
			// Create level that is same component as stencil buffer (always 8-bits)
			tcu::TextureLevel	stencilTexture(tcu::getEffectiveDepthStencilTextureFormat(bufferAccess.getFormat(), tcu::Sampler::MODE_STENCIL), bufferAccess.getWidth(), bufferAccess.getHeight(), bufferAccess.getDepth());
			// Copy stencil component only from source data. This gives us packed stencil-only data.
			tcu::copy(stencilTexture.getAccess(), tcu::getEffectiveDepthStencilAccess(bufferAccess, tcu::Sampler::MODE_STENCIL));
			srcPtr = (char*)stencilTexture.getAccess().getDataPtr();
			// Copy packed stencil-only data to output buffer
			deMemcpy(dstPtr, srcPtr, bufferSize);
			stencilLoaded = DE_TRUE;
			stencilOffset = bufferOffset;
			dstPtr += bufferSize;
			bufferOffset += bufferSize;

			// Reference image generation uses pixel offsets based on buffer offset.
			// We need to adjust the offset now that the stencil data is not interleaved.
			copyData.bufferOffset /= tcu::getPixelSize(m_textureFormat);

			copyData.bufferOffset += stencilOffset;
		}

		bufferImageCopies.push_back(copyData);
	}

	flushAlloc(vk, vkDevice, *m_sourceBufferAlloc);

	// Upload the depth/stencil data from m_destinationTextureLevel to initialize
	// depth and stencil to known values.
	// Uses uploadImageAspect so makes its own buffers for depth and stencil
	// aspects (as needed) and copies them with independent vkCmdCopyBufferToImage commands.
	uploadImage(m_destinationTextureLevel->getAccess(), *m_destination, m_params.dst.image);

	const VkImageMemoryBarrier	imageBarrier	=
	{
		VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER,		// VkStructureType			sType;
		DE_NULL,									// const void*				pNext;
		VK_ACCESS_TRANSFER_WRITE_BIT,				// VkAccessFlags			srcAccessMask;
		VK_ACCESS_TRANSFER_WRITE_BIT,				// VkAccessFlags			dstAccessMask;
		VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL,		// VkImageLayout			oldLayout;
		VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL,		// VkImageLayout			newLayout;
		VK_QUEUE_FAMILY_IGNORED,					// deUint32					srcQueueFamilyIndex;
		VK_QUEUE_FAMILY_IGNORED,					// deUint32					dstQueueFamilyIndex;
		*m_destination,								// VkImage					image;
		{											// VkImageSubresourceRange	subresourceRange;
			getAspectFlags(m_textureFormat),	// VkImageAspectFlags	aspectMask;
			0u,								// deUint32				baseMipLevel;
			1u,								// deUint32				mipLevels;
			0u,								// deUint32				baseArraySlice;
			1u								// deUint32				arraySize;
		}
	};

	// Copy from buffer to depth/stencil image

	beginCommandBuffer(vk, *m_cmdBuffer);
	vk.cmdPipelineBarrier(*m_cmdBuffer, VK_PIPELINE_STAGE_TRANSFER_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT, (VkDependencyFlags)0, 0, (const VkMemoryBarrier*)DE_NULL, 0, (const VkBufferMemoryBarrier*)DE_NULL, 1, &imageBarrier);

	if (m_params.singleCommand)
	{
		// Issue a single copy command with regions defined by the test.
		vk.cmdCopyBufferToImage(*m_cmdBuffer, m_source.get(), m_destination.get(), VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, (deUint32)m_params.regions.size(), bufferImageCopies.data());
	}
	else
	{
		// Issue a a copy command per region defined by the test.
		for (deUint32 i = 0; i < bufferImageCopies.size(); i++)
		{
			vk.cmdCopyBufferToImage(*m_cmdBuffer, m_source.get(), m_destination.get(), VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, 1, &bufferImageCopies[i]);
		}
	}
	endCommandBuffer(vk, *m_cmdBuffer);

	submitCommandsAndWait(vk, vkDevice, queue, *m_cmdBuffer);

	de::MovePtr<tcu::TextureLevel>	resultLevel = readImage(*m_destination, m_params.dst.image);

	// For combined depth/stencil formats both aspects are checked even when the test only
	// copies one. Clear such aspects here for both the result and the reference.
	if (tcu::hasDepthComponent(m_textureFormat.order) && !depthLoaded)
	{
		tcu::clearDepth(m_expectedTextureLevel[0]->getAccess(), 0.0f);
		tcu::clearDepth(resultLevel->getAccess(), 0.0f);
	}
	if (tcu::hasStencilComponent(m_textureFormat.order) && !stencilLoaded)
	{
		tcu::clearStencil(m_expectedTextureLevel[0]->getAccess(), 0);
		tcu::clearStencil(resultLevel->getAccess(), 0);
	}

	return checkTestResult(resultLevel->getAccess());
}

class CopyBufferToDepthStencilTestCase : public vkt::TestCase
{
public:
							CopyBufferToDepthStencilTestCase	(tcu::TestContext&		testCtx,
																 const std::string&		name,
																 const std::string&		description,
																 const TestParams		params)
								: vkt::TestCase(testCtx, name, description)
								, m_params(params)
							{}

	virtual					~CopyBufferToDepthStencilTestCase	(void) {}

	virtual TestInstance*	createInstance						(Context&				context) const
							{
								return new CopyBufferToDepthStencil(context, m_params);
							}
private:
	TestParams				m_params;
};

// Copy from image to image with scaling.

class BlittingImages : public CopiesAndBlittingTestInstance
{
public:
										BlittingImages					(Context&	context,
																		 TestParams params);
	virtual tcu::TestStatus				iterate							(void);
protected:
	virtual tcu::TestStatus				checkTestResult					(tcu::ConstPixelBufferAccess result);
	virtual void						copyRegionToTextureLevel		(tcu::ConstPixelBufferAccess src, tcu::PixelBufferAccess dst, CopyRegion region, deUint32 mipLevel = 0u);
	virtual void						generateExpectedResult			(void);
private:
	bool								checkNonNearestFilteredResult	(const tcu::ConstPixelBufferAccess&	result,
																		 const tcu::ConstPixelBufferAccess&	clampedReference,
																		 const tcu::ConstPixelBufferAccess&	unclampedReference,
																		 const tcu::TextureFormat&			sourceFormat);
	bool								checkNearestFilteredResult		(const tcu::ConstPixelBufferAccess&	result,
																		 const tcu::ConstPixelBufferAccess& source);

	Move<VkImage>						m_source;
	de::MovePtr<Allocation>				m_sourceImageAlloc;
	Move<VkImage>						m_destination;
	de::MovePtr<Allocation>				m_destinationImageAlloc;

	de::MovePtr<tcu::TextureLevel>		m_unclampedExpectedTextureLevel;
};

BlittingImages::BlittingImages (Context& context, TestParams params)
	: CopiesAndBlittingTestInstance(context, params)
{
	const InstanceInterface&	vki					= context.getInstanceInterface();
	const DeviceInterface&		vk					= context.getDeviceInterface();
	const VkPhysicalDevice		vkPhysDevice		= context.getPhysicalDevice();
	const VkDevice				vkDevice			= context.getDevice();
	const deUint32				queueFamilyIndex	= context.getUniversalQueueFamilyIndex();
	Allocator&					memAlloc			= context.getDefaultAllocator();

	// Create source image
	{
		const VkImageCreateInfo		sourceImageParams		=
		{
			VK_STRUCTURE_TYPE_IMAGE_CREATE_INFO,	// VkStructureType		sType;
			DE_NULL,								// const void*			pNext;
			getCreateFlags(m_params.src.image),		// VkImageCreateFlags	flags;
			m_params.src.image.imageType,			// VkImageType			imageType;
			m_params.src.image.format,				// VkFormat				format;
			getExtent3D(m_params.src.image),		// VkExtent3D			extent;
			1u,										// deUint32				mipLevels;
			getArraySize(m_params.src.image),		// deUint32				arraySize;
			VK_SAMPLE_COUNT_1_BIT,					// deUint32				samples;
			m_params.src.image.tiling,				// VkImageTiling		tiling;
			VK_IMAGE_USAGE_TRANSFER_SRC_BIT |
				VK_IMAGE_USAGE_TRANSFER_DST_BIT,	// VkImageUsageFlags	usage;
			VK_SHARING_MODE_EXCLUSIVE,				// VkSharingMode		sharingMode;
			1u,										// deUint32				queueFamilyCount;
			&queueFamilyIndex,						// const deUint32*		pQueueFamilyIndices;
			VK_IMAGE_LAYOUT_UNDEFINED,				// VkImageLayout		initialLayout;
		};

		m_source = createImage(vk, vkDevice, &sourceImageParams);
		m_sourceImageAlloc = allocateImage(vki, vk, vkPhysDevice, vkDevice, *m_source, MemoryRequirement::Any, memAlloc, m_params.allocationKind);
		VK_CHECK(vk.bindImageMemory(vkDevice, *m_source, m_sourceImageAlloc->getMemory(), m_sourceImageAlloc->getOffset()));
	}

	// Create destination image
	{
		const VkImageCreateInfo		destinationImageParams	=
		{
			VK_STRUCTURE_TYPE_IMAGE_CREATE_INFO,	// VkStructureType		sType;
			DE_NULL,								// const void*			pNext;
			getCreateFlags(m_params.dst.image),		// VkImageCreateFlags	flags;
			m_params.dst.image.imageType,			// VkImageType			imageType;
			m_params.dst.image.format,				// VkFormat				format;
			getExtent3D(m_params.dst.image),		// VkExtent3D			extent;
			1u,										// deUint32				mipLevels;
			getArraySize(m_params.dst.image),		// deUint32				arraySize;
			VK_SAMPLE_COUNT_1_BIT,					// deUint32				samples;
			m_params.dst.image.tiling,				// VkImageTiling		tiling;
			VK_IMAGE_USAGE_TRANSFER_SRC_BIT |
				VK_IMAGE_USAGE_TRANSFER_DST_BIT,	// VkImageUsageFlags	usage;
			VK_SHARING_MODE_EXCLUSIVE,				// VkSharingMode		sharingMode;
			1u,										// deUint32				queueFamilyCount;
			&queueFamilyIndex,						// const deUint32*		pQueueFamilyIndices;
			VK_IMAGE_LAYOUT_UNDEFINED,				// VkImageLayout		initialLayout;
		};

		m_destination = createImage(vk, vkDevice, &destinationImageParams);
		m_destinationImageAlloc = allocateImage(vki, vk, vkPhysDevice, vkDevice, *m_destination, MemoryRequirement::Any, memAlloc, m_params.allocationKind);
		VK_CHECK(vk.bindImageMemory(vkDevice, *m_destination, m_destinationImageAlloc->getMemory(), m_destinationImageAlloc->getOffset()));
	}
}

tcu::TestStatus BlittingImages::iterate (void)
{
	const tcu::TextureFormat	srcTcuFormat		= mapVkFormat(m_params.src.image.format);
	const tcu::TextureFormat	dstTcuFormat		= mapVkFormat(m_params.dst.image.format);
	m_sourceTextureLevel = de::MovePtr<tcu::TextureLevel>(new tcu::TextureLevel(srcTcuFormat,
																				m_params.src.image.extent.width,
																				m_params.src.image.extent.height,
																				m_params.src.image.extent.depth));
	generateBuffer(m_sourceTextureLevel->getAccess(), m_params.src.image.extent.width, m_params.src.image.extent.height, m_params.src.image.extent.depth, FILL_MODE_GRADIENT);
	m_destinationTextureLevel = de::MovePtr<tcu::TextureLevel>(new tcu::TextureLevel(dstTcuFormat,
																					 (int)m_params.dst.image.extent.width,
																					 (int)m_params.dst.image.extent.height,
																					 (int)m_params.dst.image.extent.depth));
	generateBuffer(m_destinationTextureLevel->getAccess(), m_params.dst.image.extent.width, m_params.dst.image.extent.height, m_params.dst.image.extent.depth, FILL_MODE_WHITE);
	generateExpectedResult();

	uploadImage(m_sourceTextureLevel->getAccess(), m_source.get(), m_params.src.image);
	uploadImage(m_destinationTextureLevel->getAccess(), m_destination.get(), m_params.dst.image);

	const DeviceInterface&		vk					= m_context.getDeviceInterface();
	const VkDevice				vkDevice			= m_context.getDevice();
	const VkQueue				queue				= m_context.getUniversalQueue();

	std::vector<VkImageBlit>	regions;
	for (deUint32 i = 0; i < m_params.regions.size(); i++)
		regions.push_back(m_params.regions[i].imageBlit);

	// Barriers for copying image to buffer
	const VkImageMemoryBarrier		srcImageBarrier		=
	{
		VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER,		// VkStructureType			sType;
		DE_NULL,									// const void*				pNext;
		VK_ACCESS_TRANSFER_WRITE_BIT,				// VkAccessFlags			srcAccessMask;
		VK_ACCESS_TRANSFER_READ_BIT,				// VkAccessFlags			dstAccessMask;
		VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL,		// VkImageLayout			oldLayout;
		m_params.src.image.operationLayout,			// VkImageLayout			newLayout;
		VK_QUEUE_FAMILY_IGNORED,					// deUint32					srcQueueFamilyIndex;
		VK_QUEUE_FAMILY_IGNORED,					// deUint32					dstQueueFamilyIndex;
		m_source.get(),								// VkImage					image;
		{											// VkImageSubresourceRange	subresourceRange;
			getAspectFlags(srcTcuFormat),	// VkImageAspectFlags	aspectMask;
			0u,								// deUint32				baseMipLevel;
			1u,								// deUint32				mipLevels;
			0u,								// deUint32				baseArraySlice;
			1u								// deUint32				arraySize;
		}
	};

	const VkImageMemoryBarrier		dstImageBarrier		=
	{
		VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER,		// VkStructureType			sType;
		DE_NULL,									// const void*				pNext;
		VK_ACCESS_TRANSFER_WRITE_BIT,				// VkAccessFlags			srcAccessMask;
		VK_ACCESS_TRANSFER_WRITE_BIT,				// VkAccessFlags			dstAccessMask;
		VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL,		// VkImageLayout			oldLayout;
		m_params.dst.image.operationLayout,			// VkImageLayout			newLayout;
		VK_QUEUE_FAMILY_IGNORED,					// deUint32					srcQueueFamilyIndex;
		VK_QUEUE_FAMILY_IGNORED,					// deUint32					dstQueueFamilyIndex;
		m_destination.get(),						// VkImage					image;
		{											// VkImageSubresourceRange	subresourceRange;
			getAspectFlags(dstTcuFormat),	// VkImageAspectFlags	aspectMask;
			0u,								// deUint32				baseMipLevel;
			1u,								// deUint32				mipLevels;
			0u,								// deUint32				baseArraySlice;
			1u								// deUint32				arraySize;
		}
	};

	beginCommandBuffer(vk, *m_cmdBuffer);
	vk.cmdPipelineBarrier(*m_cmdBuffer, VK_PIPELINE_STAGE_TRANSFER_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT, (VkDependencyFlags)0, 0, (const VkMemoryBarrier*)DE_NULL, 0, (const VkBufferMemoryBarrier*)DE_NULL, 1, &srcImageBarrier);
	vk.cmdPipelineBarrier(*m_cmdBuffer, VK_PIPELINE_STAGE_TRANSFER_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT, (VkDependencyFlags)0, 0, (const VkMemoryBarrier*)DE_NULL, 0, (const VkBufferMemoryBarrier*)DE_NULL, 1, &dstImageBarrier);
	vk.cmdBlitImage(*m_cmdBuffer, m_source.get(), m_params.src.image.operationLayout, m_destination.get(), m_params.dst.image.operationLayout, (deUint32)m_params.regions.size(), &regions[0], m_params.filter);
	endCommandBuffer(vk, *m_cmdBuffer);
	submitCommandsAndWait(vk, vkDevice, queue, *m_cmdBuffer);

	de::MovePtr<tcu::TextureLevel> resultTextureLevel = readImage(*m_destination, m_params.dst.image);

	return checkTestResult(resultTextureLevel->getAccess());
}

static float calculateFloatConversionError (int srcBits)
{
	if (srcBits > 0)
	{
		const int	clampedBits	= de::clamp<int>(srcBits, 0, 32);
		const float	srcMaxValue	= de::max((float)(1ULL<<clampedBits) - 1.0f, 1.0f);
		const float	error		= 1.0f / srcMaxValue;

		return de::clamp<float>(error, 0.0f, 1.0f);
	}
	else
		return 1.0f;
}

tcu::Vec4 getFormatThreshold (const tcu::TextureFormat& format)
{
	tcu::Vec4 threshold(0.01f);

	switch (format.type)
	{
	case tcu::TextureFormat::HALF_FLOAT:
		threshold = tcu::Vec4(0.005f);
		break;

	case tcu::TextureFormat::FLOAT:
	case tcu::TextureFormat::FLOAT64:
		threshold = tcu::Vec4(0.001f);
		break;

	case tcu::TextureFormat::UNSIGNED_INT_11F_11F_10F_REV:
		threshold = tcu::Vec4(0.02f, 0.02f, 0.0625f, 1.0f);
		break;

	case tcu::TextureFormat::UNSIGNED_INT_999_E5_REV:
		threshold = tcu::Vec4(0.05f, 0.05f, 0.05f, 1.0f);
		break;

	case tcu::TextureFormat::UNORM_INT_1010102_REV:
		threshold = tcu::Vec4(0.002f, 0.002f, 0.002f, 0.3f);
		break;

	case tcu:: TextureFormat::UNORM_INT8:
		threshold = tcu::Vec4(0.008f, 0.008f, 0.008f, 0.008f);
		break;

	default:
		const tcu::IVec4 bits = tcu::getTextureFormatMantissaBitDepth(format);
		threshold = tcu::Vec4(calculateFloatConversionError(bits.x()),
				      calculateFloatConversionError(bits.y()),
				      calculateFloatConversionError(bits.z()),
				      calculateFloatConversionError(bits.w()));
	}

	// Return value matching the channel order specified by the format
	if (format.order == tcu::TextureFormat::BGR || format.order == tcu::TextureFormat::BGRA)
		return threshold.swizzle(2, 1, 0, 3);
	else
		return threshold;
}

bool BlittingImages::checkNonNearestFilteredResult (const tcu::ConstPixelBufferAccess&	result,
													const tcu::ConstPixelBufferAccess&	clampedExpected,
													const tcu::ConstPixelBufferAccess&	unclampedExpected,
													const tcu::TextureFormat&			srcFormat)
{
	tcu::TestLog&					log				(m_context.getTestContext().getLog());
	const tcu::TextureFormat		dstFormat		= result.getFormat();
	const tcu::TextureChannelClass	dstChannelClass = tcu::getTextureChannelClass(dstFormat.type);
	const tcu::TextureChannelClass	srcChannelClass = tcu::getTextureChannelClass(srcFormat.type);
	bool							isOk			= false;

	log << tcu::TestLog::Section("ClampedSourceImage", "Region with clamped edges on source image.");

	// if either of srcImage or dstImage stores values as a signed/unsigned integer,
	// the other must also store values a signed/unsigned integer
	// e.g. blit unorm to uscaled is not allowed as uscaled formats store data as integers
	// despite the fact that both formats are sampled as floats
	bool dstImageIsIntClass = dstChannelClass == tcu::TEXTURECHANNELCLASS_SIGNED_INTEGER ||
							  dstChannelClass == tcu::TEXTURECHANNELCLASS_UNSIGNED_INTEGER;
	bool srcImageIsIntClass = srcChannelClass == tcu::TEXTURECHANNELCLASS_SIGNED_INTEGER ||
							  srcChannelClass == tcu::TEXTURECHANNELCLASS_UNSIGNED_INTEGER;
	if (dstImageIsIntClass != srcImageIsIntClass)
	{
		log << tcu::TestLog::EndSection;
		return false;
	}

	if (isFloatFormat(dstFormat))
	{
		const bool		srcIsSRGB	= tcu::isSRGB(srcFormat);
		const tcu::Vec4	srcMaxDiff	= getFormatThreshold(srcFormat) * tcu::Vec4(srcIsSRGB ? 2.0f : 1.0f);
		const tcu::Vec4	dstMaxDiff	= getFormatThreshold(dstFormat);
		const tcu::Vec4	threshold	= ( srcMaxDiff + dstMaxDiff ) * ((m_params.filter == VK_FILTER_CUBIC_EXT) ? 1.5f : 1.0f);

		isOk = tcu::floatThresholdCompare(log, "Compare", "Result comparsion", clampedExpected, result, threshold, tcu::COMPARE_LOG_RESULT);
		log << tcu::TestLog::EndSection;

		if (!isOk)
		{
			log << tcu::TestLog::Section("NonClampedSourceImage", "Region with non-clamped edges on source image.");
			isOk = tcu::floatThresholdCompare(log, "Compare", "Result comparsion", unclampedExpected, result, threshold, tcu::COMPARE_LOG_RESULT);
			log << tcu::TestLog::EndSection;
		}
	}
	else
	{
		tcu::UVec4	threshold;
		// Calculate threshold depending on channel width of destination format.
		const tcu::IVec4	dstBitDepth	= tcu::getTextureFormatBitDepth(dstFormat);
		const tcu::IVec4	srcBitDepth = tcu::getTextureFormatBitDepth(srcFormat);
		for (deUint32 i = 0; i < 4; ++i)
			threshold[i] = 1 + de::max( ( ( 1 << dstBitDepth[i] ) - 1 ) / de::clamp((1 << srcBitDepth[i]) - 1, 1, 256), 1);

		isOk = tcu::intThresholdCompare(log, "Compare", "Result comparsion", clampedExpected, result, threshold, tcu::COMPARE_LOG_RESULT);
		log << tcu::TestLog::EndSection;

		if (!isOk)
		{
			log << tcu::TestLog::Section("NonClampedSourceImage", "Region with non-clamped edges on source image.");
			isOk = tcu::intThresholdCompare(log, "Compare", "Result comparsion", unclampedExpected, result, threshold, tcu::COMPARE_LOG_RESULT);
			log << tcu::TestLog::EndSection;
		}
	}

	return isOk;
}

//! Utility to encapsulate coordinate computation and loops.
struct CompareEachPixelInEachRegion
{
	virtual		 ~CompareEachPixelInEachRegion  (void) {}
	virtual bool compare								(const void* pUserData, const int x, const int y, const int z, const tcu::Vec3& srcNormCoord) const = 0;

	bool forEach (const void*							pUserData,
				  const std::vector<CopyRegion>&		regions,
				  const int								sourceWidth,
				  const int								sourceHeight,
				  const int								sourceDepth,
				  const tcu::PixelBufferAccess&			errorMask) const
	{
		bool compareOk = true;

		for (std::vector<CopyRegion>::const_iterator regionIter = regions.begin(); regionIter != regions.end(); ++regionIter)
		{
			const VkImageBlit& blit = regionIter->imageBlit;

			const int	xStart	= deMin32(blit.dstOffsets[0].x, blit.dstOffsets[1].x);
			const int	yStart	= deMin32(blit.dstOffsets[0].y, blit.dstOffsets[1].y);
			const int	zStart	= deMin32(blit.dstOffsets[0].z, blit.dstOffsets[1].z);
			const int	xEnd	= deMax32(blit.dstOffsets[0].x, blit.dstOffsets[1].x);
			const int	yEnd	= deMax32(blit.dstOffsets[0].y, blit.dstOffsets[1].y);
			const int	zEnd	= deMax32(blit.dstOffsets[0].z, blit.dstOffsets[1].z);
			const float	xScale	= static_cast<float>(blit.srcOffsets[1].x - blit.srcOffsets[0].x) / static_cast<float>(blit.dstOffsets[1].x - blit.dstOffsets[0].x);
			const float	yScale	= static_cast<float>(blit.srcOffsets[1].y - blit.srcOffsets[0].y) / static_cast<float>(blit.dstOffsets[1].y - blit.dstOffsets[0].y);
			const float	zScale	= static_cast<float>(blit.srcOffsets[1].z - blit.srcOffsets[0].z) / static_cast<float>(blit.dstOffsets[1].z - blit.dstOffsets[0].z);
			const float srcInvW	= 1.0f / static_cast<float>(sourceWidth);
			const float srcInvH	= 1.0f / static_cast<float>(sourceHeight);
			const float srcInvD	= 1.0f / static_cast<float>(sourceDepth);

			for (int z = zStart; z < zEnd; z++)
			for (int y = yStart; y < yEnd; y++)
			for (int x = xStart; x < xEnd; x++)
			{
				const tcu::Vec3 srcNormCoord
				(
					(xScale * (static_cast<float>(x - blit.dstOffsets[0].x) + 0.5f) + static_cast<float>(blit.srcOffsets[0].x)) * srcInvW,
					(yScale * (static_cast<float>(y - blit.dstOffsets[0].y) + 0.5f) + static_cast<float>(blit.srcOffsets[0].y)) * srcInvH,
					(zScale * (static_cast<float>(z - blit.dstOffsets[0].z) + 0.5f) + static_cast<float>(blit.srcOffsets[0].z)) * srcInvD
				);

				if (!compare(pUserData, x, y, z, srcNormCoord))
				{
					errorMask.setPixel(tcu::Vec4(1.0f, 0.0f, 0.0f, 1.0f), x, y, z);
					compareOk = false;
				}
			}
		}
		return compareOk;
	}
};

tcu::Vec4 getFloatOrFixedPointFormatThreshold (const tcu::TextureFormat& format)
{
	const tcu::TextureChannelClass	channelClass	= tcu::getTextureChannelClass(format.type);
	const tcu::IVec4				bitDepth		= tcu::getTextureFormatBitDepth(format);

	if (channelClass == tcu::TEXTURECHANNELCLASS_FLOATING_POINT)
	{
		return getFormatThreshold(format);
	}
	else if (channelClass == tcu::TEXTURECHANNELCLASS_UNSIGNED_FIXED_POINT ||
			 channelClass == tcu::TEXTURECHANNELCLASS_SIGNED_FIXED_POINT)
	{
		const bool	isSigned	= (channelClass == tcu::TEXTURECHANNELCLASS_SIGNED_FIXED_POINT);
		const float	range		= isSigned ? 1.0f - (-1.0f)
										   : 1.0f -   0.0f;

		tcu::Vec4 v;
		for (int i = 0; i < 4; ++i)
		{
			if (bitDepth[i] == 0)
				v[i] = 1.0f;
			else
				v[i] = range / static_cast<float>((1 << bitDepth[i]) - 1);
		}
		return v;
	}
	else
	{
		DE_ASSERT(0);
		return tcu::Vec4();
	}
}

bool floatNearestBlitCompare (const tcu::ConstPixelBufferAccess&	source,
							  const tcu::ConstPixelBufferAccess&	result,
							  const tcu::PixelBufferAccess&			errorMask,
							  const std::vector<CopyRegion>&		regions)
{
	const tcu::Sampler		sampler		(tcu::Sampler::CLAMP_TO_EDGE, tcu::Sampler::CLAMP_TO_EDGE, tcu::Sampler::CLAMP_TO_EDGE, tcu::Sampler::NEAREST, tcu::Sampler::NEAREST);
	tcu::LookupPrecision	precision;

	{
		const tcu::IVec4	dstBitDepth	= tcu::getTextureFormatBitDepth(result.getFormat());
		const tcu::Vec4		srcMaxDiff	= getFloatOrFixedPointFormatThreshold(source.getFormat());
		const tcu::Vec4		dstMaxDiff	= getFloatOrFixedPointFormatThreshold(result.getFormat());

		precision.colorMask		 = tcu::notEqual(dstBitDepth, tcu::IVec4(0));
		precision.colorThreshold = tcu::max(srcMaxDiff, dstMaxDiff);
	}

	const struct Capture
	{
		const tcu::ConstPixelBufferAccess&	source;
		const tcu::ConstPixelBufferAccess&	result;
		const tcu::Sampler&					sampler;
		const tcu::LookupPrecision&			precision;
		const bool							isSRGB;
	} capture =
	{
		source, result, sampler, precision, tcu::isSRGB(result.getFormat())
	};

	const struct Loop : CompareEachPixelInEachRegion
	{
		Loop (void) {}

		bool compare (const void* pUserData, const int x, const int y, const int z, const tcu::Vec3& srcNormCoord) const
		{
			const Capture&					c					= *static_cast<const Capture*>(pUserData);
			const tcu::TexLookupScaleMode	lookupScaleDontCare	= tcu::TEX_LOOKUP_SCALE_MINIFY;
			tcu::Vec4						dstColor			= c.result.getPixel(x, y, z);

			// TexLookupVerifier performs a conversion to linear space, so we have to as well
			if (c.isSRGB)
				dstColor = tcu::sRGBToLinear(dstColor);

			return tcu::isLevel3DLookupResultValid(c.source, c.sampler, lookupScaleDontCare, c.precision, srcNormCoord, dstColor);
		}
	} loop;

	return loop.forEach(&capture, regions, source.getWidth(), source.getHeight(), source.getDepth(), errorMask);
}

bool intNearestBlitCompare (const tcu::ConstPixelBufferAccess&	source,
							const tcu::ConstPixelBufferAccess&	result,
							const tcu::PixelBufferAccess&		errorMask,
							const std::vector<CopyRegion>&		regions)
{
	const tcu::Sampler		sampler		(tcu::Sampler::CLAMP_TO_EDGE, tcu::Sampler::CLAMP_TO_EDGE, tcu::Sampler::CLAMP_TO_EDGE, tcu::Sampler::NEAREST, tcu::Sampler::NEAREST);
	tcu::IntLookupPrecision	precision;

	{
		const tcu::IVec4	srcBitDepth	= tcu::getTextureFormatBitDepth(source.getFormat());
		const tcu::IVec4	dstBitDepth	= tcu::getTextureFormatBitDepth(result.