
/*------------------------------------------------------------------------
 * Vulkan Conformance Tests
 * ------------------------
 *
 * Copyright (c) 2019 The Khronos Group Inc.
 *
 * Licensed under the Apache License, Version 2.0 (the "License");
 * you may not use this file except in compliance with the License.
 * You may obtain a copy of the License at
 *
 *      http://www.apache.org/licenses/LICENSE-2.0
 *
 * Unless required by applicable law or agreed to in writing, software
 * distributed under the License is distributed on an "AS IS" BASIS,
 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 * See the License for the specific language governing permissions and
 * limitations under the License.
 *
 *//*!
 * \file
 * \brief Signal ordering tests
 *//*--------------------------------------------------------------------*/

#include "vktSynchronizationSignalOrderTests.hpp"
#include "vktSynchronizationOperation.hpp"
#include "vktSynchronizationOperationTestData.hpp"
#include "vktSynchronizationOperationResources.hpp"
#include "vktTestCaseUtil.hpp"
#include "vktSynchronizationUtil.hpp"
#include "vktExternalMemoryUtil.hpp"
#include "vkBarrierUtil.hpp"

#include "vkDefs.hpp"
#include "vkPlatform.hpp"
#include "vkQueryUtil.hpp"
#include "vkCmdUtil.hpp"
#include "vkImageUtil.hpp"
#include "vkRef.hpp"
#include "vkTypeUtil.hpp"

#include "tcuTestLog.hpp"

#include "deRandom.hpp"
#include "deThread.hpp"
#include "deUniquePtr.hpp"

#include <limits>
#include <set>

namespace vkt
{
namespace synchronization
{
namespace
{

using namespace vk;
using namespace vkt::ExternalMemoryUtil;
using de::MovePtr;
using de::SharedPtr;
using de::UniquePtr;

template<typename T>
inline SharedPtr<Move<T> > makeVkSharedPtr (Move<T> move)
{
	return SharedPtr<Move<T> >(new Move<T>(move));
}

template<typename T>
inline SharedPtr<T> makeSharedPtr (de::MovePtr<T> move)
{
	return SharedPtr<T>(move.release());
}

template<typename T>
inline SharedPtr<T> makeSharedPtr (T* ptr)
{
	return SharedPtr<T>(ptr);
}

void hostSignal (const DeviceInterface& vk, const VkDevice& device, VkSemaphore semaphore, const deUint64 timelineValue)
{
	VkSemaphoreSignalInfoKHR	ssi	=
	{
		VK_STRUCTURE_TYPE_SEMAPHORE_SIGNAL_INFO,	// VkStructureType				sType;
		DE_NULL,									// const void*					pNext;
		semaphore,									// VkSemaphore					semaphore;
		timelineValue,								// deUint64						value;
	};

	VK_CHECK(vk.signalSemaphore(device, &ssi));
}

Move<VkDevice> createDevice (const Context& context)
{
	const float									priority				= 0.0f;
	const std::vector<VkQueueFamilyProperties>	queueFamilyProperties	= getPhysicalDeviceQueueFamilyProperties(context.getInstanceInterface(), context.getPhysicalDevice());
	std::vector<deUint32>						queueFamilyIndices		(queueFamilyProperties.size(), 0xFFFFFFFFu);
	std::vector<const char*>					extensions;

	if (context.isDeviceFunctionalitySupported("VK_KHR_timeline_semaphore"))
		extensions.push_back("VK_KHR_timeline_semaphore");

	if (!isCoreDeviceExtension(context.getUsedApiVersion(), "VK_KHR_external_semaphore"))
		extensions.push_back("VK_KHR_external_semaphore");
	if (!isCoreDeviceExtension(context.getUsedApiVersion(), "VK_KHR_external_memory"))
		extensions.push_back("VK_KHR_external_memory");

	if (context.isDeviceFunctionalitySupported("VK_KHR_external_semaphore_fd"))
		extensions.push_back("VK_KHR_external_semaphore_fd");

	if (context.isDeviceFunctionalitySupported("VK_KHR_external_semaphore_win32"))
		extensions.push_back("VK_KHR_external_semaphore_win32");

	try
	{
		std::vector<VkDeviceQueueCreateInfo>	queues;

		for (size_t ndx = 0; ndx < queueFamilyProperties.size(); ndx++)
		{
			const VkDeviceQueueCreateInfo	createInfo	=
			{
				VK_STRUCTURE_TYPE_DEVICE_QUEUE_CREATE_INFO,
				DE_NULL,
				0u,

				(deUint32)ndx,
				queueFamilyProperties[ndx].queueCount,
				&priority
			};

			queues.push_back(createInfo);
		}

		const VkPhysicalDeviceFeatures2			createPhysicalFeature	=
		{
			VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_FEATURES_2,
			DE_NULL,
			context.getDeviceFeatures(),
		};
		const VkDeviceCreateInfo				createInfo				=
		{
			VK_STRUCTURE_TYPE_DEVICE_CREATE_INFO,
			&createPhysicalFeature,
			0u,

			(deUint32)queues.size(),
			&queues[0],

			0u,
			DE_NULL,

			(deUint32)extensions.size(),
			extensions.empty() ? DE_NULL : &extensions[0],
			0u
		};

		return createDevice(context.getPlatformInterface(), context.getInstance(), context.getInstanceInterface(), context.getPhysicalDevice(), &createInfo);
	}
	catch (const vk::Error& error)
	{
		if (error.getError() == VK_ERROR_EXTENSION_NOT_PRESENT)
			TCU_THROW(NotSupportedError, "Required extensions not supported");
		else
			throw;
	}
}

// Class to wrap a singleton instance and device
class SingletonDevice
{
	SingletonDevice	(const Context& context)
		: m_logicalDevice	(createDevice(context))
	{
	}

public:

	static const Unique<vk::VkDevice>& getDevice(const Context& context)
	{
		if (!m_singletonDevice)
			m_singletonDevice = SharedPtr<SingletonDevice>(new SingletonDevice(context));

		DE_ASSERT(m_singletonDevice);
		return m_singletonDevice->m_logicalDevice;
	}

	static void destroy()
	{
		m_singletonDevice.clear();
	}

private:
	const Unique<vk::VkDevice>					m_logicalDevice;

	static SharedPtr<SingletonDevice>	m_singletonDevice;
};
SharedPtr<SingletonDevice>		SingletonDevice::m_singletonDevice;

static void cleanupGroup ()
{
	// Destroy singleton object
	SingletonDevice::destroy();
}

class SimpleAllocation : public Allocation
{
public:
	SimpleAllocation	(const DeviceInterface&	vkd,
						 VkDevice				device,
						 const VkDeviceMemory	memory);
	~SimpleAllocation	(void);

private:
	const DeviceInterface&	m_vkd;
	const VkDevice			m_device;
};

SimpleAllocation::SimpleAllocation (const DeviceInterface&	vkd,
									VkDevice				device,
									const VkDeviceMemory	memory)
	: Allocation	(memory, 0, DE_NULL)
	, m_vkd			(vkd)
	, m_device		(device)
{
}

SimpleAllocation::~SimpleAllocation (void)
{
	m_vkd.freeMemory(m_device, getMemory(), DE_NULL);
}

vk::VkMemoryRequirements getMemoryRequirements (const DeviceInterface&				vkd,
												 VkDevice							device,
												 VkBuffer							buffer)
{
	const VkBufferMemoryRequirementsInfo2	requirementInfo =
	{
		VK_STRUCTURE_TYPE_BUFFER_MEMORY_REQUIREMENTS_INFO_2,
		DE_NULL,
		buffer
	};
	VkMemoryRequirements2					requirements	=
	{
		VK_STRUCTURE_TYPE_MEMORY_REQUIREMENTS_2,
		DE_NULL,
		{ 0u, 0u, 0u, }
	};
	vkd.getBufferMemoryRequirements2(device, &requirementInfo, &requirements);
	return requirements.memoryRequirements;
}

vk::VkMemoryRequirements getMemoryRequirements(const DeviceInterface&				vkd,
												VkDevice							device,
												VkImage								image)
{
	const VkImageMemoryRequirementsInfo2	requirementInfo =
	{
		VK_STRUCTURE_TYPE_IMAGE_MEMORY_REQUIREMENTS_INFO_2,
		DE_NULL,
		image
	};
	VkMemoryRequirements2					requirements =
	{
		VK_STRUCTURE_TYPE_MEMORY_REQUIREMENTS_2,
		DE_NULL,
		{ 0u, 0u, 0u, }
	};
	vkd.getImageMemoryRequirements2(device, &requirementInfo, &requirements);

	return requirements.memoryRequirements;
}

MovePtr<Allocation> importAndBindMemory (const DeviceInterface&					vkd,
										 VkDevice								device,
										 VkBuffer								buffer,
										 NativeHandle&							nativeHandle,
										 VkExternalMemoryHandleTypeFlagBits		externalType,
										 const deUint32							exportedMemoryTypeIndex)
{
	const VkMemoryRequirements	requirements			= getBufferMemoryRequirements(vkd, device, buffer);
	Move<VkDeviceMemory>		memory;

	if (!!buffer)
		memory = importDedicatedMemory(vkd, device, buffer, requirements, externalType, exportedMemoryTypeIndex, nativeHandle);
	else
		memory = importMemory(vkd, device, requirements, externalType, exportedMemoryTypeIndex, nativeHandle);

	VK_CHECK(vkd.bindBufferMemory(device, buffer, *memory, 0u));

	return MovePtr<Allocation>(new SimpleAllocation(vkd, device, memory.disown()));
}

MovePtr<Allocation> importAndBindMemory (const DeviceInterface&					vkd,
										 VkDevice								device,
										 VkImage								image,
										 NativeHandle&							nativeHandle,
										 VkExternalMemoryHandleTypeFlagBits		externalType,
										 deUint32								exportedMemoryTypeIndex)
{
	const VkMemoryRequirements	requirements	= getImageMemoryRequirements(vkd, device, image);
	Move<VkDeviceMemory>		memory;

	if (!!image)
		memory = importDedicatedMemory(vkd, device, image, requirements, externalType, exportedMemoryTypeIndex, nativeHandle);
	else
		memory = importMemory(vkd, device, requirements, externalType, exportedMemoryTypeIndex, nativeHandle);

	VK_CHECK(vkd.bindImageMemory(device, image, *memory, 0u));

	return MovePtr<Allocation>(new SimpleAllocation(vkd, device, memory.disown()));
}

struct QueueTimelineIteration
{
	QueueTimelineIteration(const SharedPtr<OperationSupport>&	_opSupport,
						   deUint64								lastValue,
						   VkQueue								_queue,
						   deUint32								_queueFamilyIdx,
						   de::Random&							rng)
		: opSupport(_opSupport)
		, queue(_queue)
		, queueFamilyIdx(_queueFamilyIdx)
	{
		timelineValue	= lastValue + rng.getInt(1, 100);
	}
	~QueueTimelineIteration() {}

	SharedPtr<OperationSupport>	opSupport;
	VkQueue						queue;
	deUint32					queueFamilyIdx;
	deUint64					timelineValue;
	SharedPtr<Operation>		op;
};

de::MovePtr<Resource> importResource (const DeviceInterface&				vkd,
									  VkDevice								device,
									  const ResourceDescription&			resourceDesc,
									  const deUint32						queueFamilyIndex,
									  const OperationSupport&				readOp,
									  const OperationSupport&				writeOp,
									  NativeHandle&							nativeHandle,
									  VkExternalMemoryHandleTypeFlagBits	externalType,
									  deUint32								exportedMemoryTypeIndex)
{
	if (resourceDesc.type == RESOURCE_TYPE_IMAGE)
	{
		const VkExtent3D					extent					=
		{
			(deUint32)resourceDesc.size.x(),
			de::max(1u, (deUint32)resourceDesc.size.y()),
			de::max(1u, (deUint32)resourceDesc.size.z())
		};
		const VkImageSubresourceRange	subresourceRange		=
		{
			resourceDesc.imageAspect,
			0u,
			1u,
			0u,
			1u
		};
		const VkImageSubresourceLayers	subresourceLayers		=
		{
			resourceDesc.imageAspect,
			0u,
			0u,
			1u
		};
		const VkExternalMemoryImageCreateInfo externalInfo =
		{
			VK_STRUCTURE_TYPE_EXTERNAL_MEMORY_IMAGE_CREATE_INFO,
			DE_NULL,
			(VkExternalMemoryHandleTypeFlags)externalType
		};
		const VkImageCreateInfo			createInfo				=
		{
			VK_STRUCTURE_TYPE_IMAGE_CREATE_INFO,
			&externalInfo,
			0u,

			resourceDesc.imageType,
			resourceDesc.imageFormat,
			extent,
			1u,
			1u,
			VK_SAMPLE_COUNT_1_BIT,
			VK_IMAGE_TILING_OPTIMAL,
			readOp.getInResourceUsageFlags() | writeOp.getOutResourceUsageFlags(),
			VK_SHARING_MODE_EXCLUSIVE,

			1u,
			&queueFamilyIndex,
			VK_IMAGE_LAYOUT_UNDEFINED
		};

		Move<VkImage>			image		= createImage(vkd, device, &createInfo);
		MovePtr<Allocation>		allocation	= importAndBindMemory(vkd, device, *image, nativeHandle, externalType, exportedMemoryTypeIndex);

		return MovePtr<Resource>(new Resource(image, allocation, extent, resourceDesc.imageType, resourceDesc.imageFormat, subresourceRange, subresourceLayers));
	}
	else
	{
		const VkDeviceSize						offset			= 0u;
		const VkDeviceSize						size			= static_cast<VkDeviceSize>(resourceDesc.size.x());
		const VkBufferUsageFlags				usage			= readOp.getInResourceUsageFlags() | writeOp.getOutResourceUsageFlags();
		const VkExternalMemoryBufferCreateInfo	externalInfo	=
		{
			VK_STRUCTURE_TYPE_EXTERNAL_MEMORY_BUFFER_CREATE_INFO,
			DE_NULL,
			(VkExternalMemoryHandleTypeFlags)externalType
		};
		const VkBufferCreateInfo				createInfo		=
		{
			VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO,
			&externalInfo,
			0u,

			size,
			usage,
			VK_SHARING_MODE_EXCLUSIVE,
			1u,
			&queueFamilyIndex
		};
		Move<VkBuffer>							buffer		= createBuffer(vkd, device, &createInfo);
		MovePtr<Allocation>						allocation	= importAndBindMemory(vkd,
																				  device,
																				  *buffer,
																				  nativeHandle,
																				  externalType,
																				  exportedMemoryTypeIndex);

		return MovePtr<Resource>(new Resource(resourceDesc.type, buffer, allocation, offset, size));
	}
}

struct QueueSubmitOrderSharedIteration
{
	QueueSubmitOrderSharedIteration() {}
	~QueueSubmitOrderSharedIteration() {}

	SharedPtr<Resource>			resourceA;
	SharedPtr<Resource>			resourceB;

	SharedPtr<Operation>		writeOp;
	SharedPtr<Operation>		readOp;
};

// Verifies the signaling order of the semaphores in multiple
// VkSubmitInfo given to vkQueueSubmit() with queueA & queueB from a
// different VkDevice.
//
// vkQueueSubmit(queueA, [write0, write1, write2, ..., write6])
// vkQueueSubmit(queueB, [read0-6])
//
// With read0-6 waiting on write6, all the data should be available
// for reading given that signal operations are supposed to happen in
// order.
class QueueSubmitSignalOrderSharedTestInstance : public TestInstance
{
public:
	QueueSubmitSignalOrderSharedTestInstance (Context&									context,
											  const SharedPtr<OperationSupport>			writeOpSupport,
											  const SharedPtr<OperationSupport>			readOpSupport,
											  const ResourceDescription&				resourceDesc,
											  VkExternalMemoryHandleTypeFlagBits		memoryHandleType,
											  VkSemaphoreType							semaphoreType,
											  VkExternalSemaphoreHandleTypeFlagBits		semaphoreHandleType,
											  PipelineCacheData&						pipelineCacheData)
		: TestInstance			(context)
		, m_writeOpSupport		(writeOpSupport)
		, m_readOpSupport		(readOpSupport)
		, m_resourceDesc		(resourceDesc)
		, m_memoryHandleType	(memoryHandleType)
		, m_semaphoreType		(semaphoreType)
		, m_semaphoreHandleType	(semaphoreHandleType)
		, m_pipelineCacheData	(pipelineCacheData)
		, m_rng					(1234)

	{
		const InstanceInterface&					vki					= context.getInstanceInterface();
		const VkSemaphoreTypeCreateInfoKHR			semaphoreTypeInfo	=
		{
			VK_STRUCTURE_TYPE_SEMAPHORE_TYPE_CREATE_INFO_KHR,
			DE_NULL,
			VK_SEMAPHORE_TYPE_TIMELINE_KHR,
			0,
		};
		const VkPhysicalDeviceExternalSemaphoreInfo	info				=
		{
			VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_EXTERNAL_SEMAPHORE_INFO,
			&semaphoreTypeInfo,
			semaphoreHandleType
		};
		VkExternalSemaphoreProperties				properties			=
		{
			VK_STRUCTURE_TYPE_EXTERNAL_SEMAPHORE_PROPERTIES,
			DE_NULL,
			0u,
			0u,
			0u
		};

		vki.getPhysicalDeviceExternalSemaphoreProperties(context.getPhysicalDevice(), &info, &properties);

		if (m_semaphoreType == VK_SEMAPHORE_TYPE_TIMELINE_KHR &&
			!context.getTimelineSemaphoreFeatures().timelineSemaphore)
			TCU_THROW(NotSupportedError, "Timeline semaphore not supported");

		if ((properties.externalSemaphoreFeatures & vk::VK_EXTERNAL_SEMAPHORE_FEATURE_EXPORTABLE_BIT_KHR) == 0
			|| (properties.externalSemaphoreFeatures & vk::VK_EXTERNAL_SEMAPHORE_FEATURE_IMPORTABLE_BIT_KHR) == 0)
			TCU_THROW(NotSupportedError, "Exporting and importing semaphore type not supported");

		if (!isResourceExportable())
			TCU_THROW(NotSupportedError, "Resource not exportable");

	}

	Move<VkImage> createImage (const vk::DeviceInterface&	vkd,
							   vk::VkDevice					device,
							   const vk::VkExtent3D&		extent,
							   deUint32						queueFamilyIndex)
	{
		const VkExternalMemoryImageCreateInfo externalInfo =
		{
			VK_STRUCTURE_TYPE_EXTERNAL_MEMORY_IMAGE_CREATE_INFO,
			DE_NULL,
			(VkExternalMemoryHandleTypeFlags)m_memoryHandleType
		};
		const VkImageCreateInfo createInfo =
		{
			VK_STRUCTURE_TYPE_IMAGE_CREATE_INFO,
			&externalInfo,
			0u,

			m_resourceDesc.imageType,
			m_resourceDesc.imageFormat,
			extent,
			1u,
			1u,
			VK_SAMPLE_COUNT_1_BIT,
			VK_IMAGE_TILING_OPTIMAL,
			m_readOpSupport->getInResourceUsageFlags() | m_writeOpSupport->getOutResourceUsageFlags(),
			VK_SHARING_MODE_EXCLUSIVE,

			1u,
			&queueFamilyIndex,
			VK_IMAGE_LAYOUT_UNDEFINED
		};

		return vk::createImage(vkd, device, &createInfo);
	}

	Move<VkBuffer> createBuffer (const vk::DeviceInterface&		vkd,
								 vk::VkDevice					device,
								 const vk::VkDeviceSize&		size,
								 deUint32						queueFamilyIndex)
	{
		const VkExternalMemoryBufferCreateInfo	externalInfo =
		{
			VK_STRUCTURE_TYPE_EXTERNAL_MEMORY_BUFFER_CREATE_INFO,
			DE_NULL,
			(VkExternalMemoryHandleTypeFlags)m_memoryHandleType
		};
		const VkBufferCreateInfo				createInfo =
		{
			VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO,
			&externalInfo,
			0u,

			size,
			m_readOpSupport->getInResourceUsageFlags() | m_writeOpSupport->getOutResourceUsageFlags(),
			VK_SHARING_MODE_EXCLUSIVE,
			1u,
			&queueFamilyIndex
		};
		return vk::createBuffer(vkd, device, &createInfo);
	}

	tcu::TestStatus iterate (void)
	{
		// We're using 2 devices to make sure we have 2 queues even on
		// implementations that only have a single queue.
		const VkDevice&										deviceA						= m_context.getDevice();
		const Unique<VkDevice>&								deviceB						(SingletonDevice::getDevice(m_context));
		const DeviceInterface&								vkA							= m_context.getDeviceInterface();
		const DeviceDriver									vkB							(m_context.getPlatformInterface(), m_context.getInstance(), *deviceB);
		UniquePtr<SimpleAllocator>							allocatorA					(new SimpleAllocator(vkA, deviceA, vk::getPhysicalDeviceMemoryProperties(m_context.getInstanceInterface(),
																																								 m_context.getPhysicalDevice())));
		UniquePtr<SimpleAllocator>							allocatorB					(new SimpleAllocator(vkB, *deviceB, vk::getPhysicalDeviceMemoryProperties(m_context.getInstanceInterface(),
																																								  m_context.getPhysicalDevice())));
		UniquePtr<OperationContext>							operationContextA			(new OperationContext(m_context, m_pipelineCacheData, vkA, deviceA, *allocatorA));
		UniquePtr<OperationContext>							operationContextB			(new OperationContext(m_context, m_pipelineCacheData, vkB, *deviceB, *allocatorB));
		const deUint32										universalQueueFamilyIndex	= m_context.getUniversalQueueFamilyIndex();
		const VkQueue										queueA						= m_context.getUniversalQueue();
		const VkQueue										queueB						= getDeviceQueue(vkB, *deviceB, m_context.getUniversalQueueFamilyIndex(), 0);
		Unique<VkFence>										fenceA						(createFence(vkA, deviceA));
		Unique<VkFence>										fenceB						(createFence(vkB, *deviceB));
		const Unique<VkCommandPool>							cmdPoolA					(createCommandPool(vkA, deviceA, VK_COMMAND_POOL_CREATE_RESET_COMMAND_BUFFER_BIT, universalQueueFamilyIndex));
		const Unique<VkCommandPool>							cmdPoolB					(createCommandPool(vkB, *deviceB, VK_COMMAND_POOL_CREATE_RESET_COMMAND_BUFFER_BIT, universalQueueFamilyIndex));
		std::vector<SharedPtr<Move<VkCommandBuffer> > >		ptrCmdBuffersA;
		SharedPtr<Move<VkCommandBuffer> >					ptrCmdBufferB;
		std::vector<VkCommandBuffer>						cmdBuffersA;
		VkCommandBuffer										cmdBufferB;
		std::vector<Move<VkSemaphore> >						semaphoresA;
		std::vector<Move<VkSemaphore> >						semaphoresB;
		std::vector<VkSemaphore>							semaphoreHandlesA;
		std::vector<VkSemaphore>							semaphoreHandlesB;
		std::vector<deUint64>								timelineValuesA;
		std::vector<deUint64>								timelineValuesB;
		std::vector<QueueSubmitOrderSharedIteration>		iterations(12);
		std::vector<VkPipelineStageFlags>					stageBits;

		// Create a dozen of set of write/read operations.
		for (deUint32 iterIdx = 0; iterIdx < iterations.size(); iterIdx++)
		{
			QueueSubmitOrderSharedIteration&	iter				= iterations[iterIdx];
			deUint32							memoryTypeIndex;
			NativeHandle						nativeMemoryHandle;

			if (m_resourceDesc.type == RESOURCE_TYPE_IMAGE)
			{
				const VkExtent3D				extent =
				{
					(deUint32)m_resourceDesc.size.x(),
					de::max(1u, (deUint32)m_resourceDesc.size.y()),
					de::max(1u, (deUint32)m_resourceDesc.size.z())
				};
				const VkImageSubresourceRange	subresourceRange =
				{
					m_resourceDesc.imageAspect,
					0u,
					1u,
					0u,
					1u
				};
				const VkImageSubresourceLayers	subresourceLayers =
				{
					m_resourceDesc.imageAspect,
					0u,
					0u,
					1u
				};

				Move<VkImage>							image			= createImage(vkA, deviceA, extent, universalQueueFamilyIndex);
				const vk::VkMemoryRequirements			requirements	= getMemoryRequirements(vkA, deviceA, *image);
														memoryTypeIndex = chooseMemoryType(requirements.memoryTypeBits);
				vk::Move<vk::VkDeviceMemory>			memory			= allocateExportableMemory(vkA, deviceA, requirements.size, memoryTypeIndex, m_memoryHandleType, *image);

				VK_CHECK(vkA.bindImageMemory(deviceA, *image, *memory, 0u));

				MovePtr<Allocation> allocation(new SimpleAllocation(vkA, deviceA, memory.disown()));
				iter.resourceA = makeSharedPtr(new Resource(image, allocation, extent, m_resourceDesc.imageType, m_resourceDesc.imageFormat, subresourceRange, subresourceLayers));
			}
			else
			{
				const VkDeviceSize						offset			= 0u;
				const VkDeviceSize						size			= static_cast<VkDeviceSize>(m_resourceDesc.size.x());
				Move<VkBuffer>							buffer			= createBuffer(vkA, deviceA, size, universalQueueFamilyIndex);
				const vk::VkMemoryRequirements			requirements	= getMemoryRequirements(vkA, deviceA, *buffer);
														memoryTypeIndex	= chooseMemoryType(requirements.memoryTypeBits);
				vk::Move<vk::VkDeviceMemory>			memory			= allocateExportableMemory(vkA, deviceA, requirements.size, memoryTypeIndex, m_memoryHandleType, *buffer);

				VK_CHECK(vkA.bindBufferMemory(deviceA, *buffer, *memory, 0u));

				MovePtr<Allocation> allocation(new SimpleAllocation(vkA, deviceA, memory.disown()));
				iter.resourceA = makeSharedPtr(new Resource(m_resourceDesc.type, buffer, allocation, offset, size));
			}

			getMemoryNative(vkA, deviceA, iter.resourceA->getMemory(), m_memoryHandleType, nativeMemoryHandle);
			iter.resourceB	= makeSharedPtr(importResource(vkB, *deviceB,
														   m_resourceDesc,
														   universalQueueFamilyIndex,
														   *m_readOpSupport,
														   *m_writeOpSupport,
														   nativeMemoryHandle,
														   m_memoryHandleType,
														   memoryTypeIndex));

			iter.writeOp = makeSharedPtr(m_writeOpSupport->build(*operationContextA,
																 *iter.resourceA));
			iter.readOp = makeSharedPtr(m_readOpSupport->build(*operationContextB,
															   *iter.resourceB));
		}

		// Record each write operation into its own command buffer.
		for (deUint32 iterIdx = 0; iterIdx < iterations.size(); iterIdx++)
		{
			QueueSubmitOrderSharedIteration&	iter		= iterations[iterIdx];
			const Resource&						resource	= *iter.resourceA;
			const SyncInfo						writeSync	= iter.writeOp->getOutSyncInfo();
			const SyncInfo						readSync	= iter.readOp->getInSyncInfo();

			ptrCmdBuffersA.push_back(makeVkSharedPtr(makeCommandBuffer(vkA, deviceA, *cmdPoolA)));

			cmdBuffersA.push_back(**(ptrCmdBuffersA.back()));

			beginCommandBuffer(vkA, cmdBuffersA.back());

			iter.writeOp->recordCommands(cmdBuffersA.back());

			{

				if (resource.getType() == RESOURCE_TYPE_IMAGE)
				{
					DE_ASSERT(writeSync.imageLayout != VK_IMAGE_LAYOUT_UNDEFINED);
					DE_ASSERT(readSync.imageLayout != VK_IMAGE_LAYOUT_UNDEFINED);
					const VkImageMemoryBarrier barrier =  makeImageMemoryBarrier(writeSync.accessMask, readSync.accessMask,
																				 writeSync.imageLayout, readSync.imageLayout,
																				 resource.getImage().handle,
																				 resource.getImage().subresourceRange);
					vkA.cmdPipelineBarrier(cmdBuffersA.back(), writeSync.stageMask, readSync.stageMask, (VkDependencyFlags)0,
										   0u, (const VkMemoryBarrier*)DE_NULL, 0u, (const VkBufferMemoryBarrier*)DE_NULL, 1u, &barrier);
				}
				else
				{
					const VkBufferMemoryBarrier barrier = makeBufferMemoryBarrier(writeSync.accessMask, readSync.accessMask,
																				  resource.getBuffer().handle, 0, VK_WHOLE_SIZE);
					vkA.cmdPipelineBarrier(cmdBuffersA.back(), writeSync.stageMask, readSync.stageMask, (VkDependencyFlags)0,
										   0u, (const VkMemoryBarrier*)DE_NULL, 1u, &barrier, 0u, (const VkImageMemoryBarrier*)DE_NULL);
				}

				stageBits.push_back(writeSync.stageMask);
			}

			endCommandBuffer(vkA, cmdBuffersA.back());

			addSemaphore(vkA, deviceA, semaphoresA, semaphoreHandlesA, timelineValuesA, iterIdx == (iterations.size() - 1), 2u);
		}

		DE_ASSERT(stageBits.size() == iterations.size());
		DE_ASSERT(semaphoreHandlesA.size() == iterations.size());

		// Record all read operations into a single command buffer and record the union of their stage masks.
		VkPipelineStageFlags readStages = 0;
		ptrCmdBufferB = makeVkSharedPtr(makeCommandBuffer(vkB, *deviceB, *cmdPoolB));
		cmdBufferB = **(ptrCmdBufferB);
		beginCommandBuffer(vkB, cmdBufferB);
		for (deUint32 iterIdx = 0; iterIdx < iterations.size(); iterIdx++)
		{
			QueueSubmitOrderSharedIteration& iter = iterations[iterIdx];
			readStages |= iter.readOp->getInSyncInfo().stageMask;
			iter.readOp->recordCommands(cmdBufferB);
		}
		endCommandBuffer(vkB, cmdBufferB);

		// Export the last semaphore for use on deviceB and create another semaphore to signal on deviceB.
		{
			VkSemaphore		lastSemaphoreA			= semaphoreHandlesA.back();
			NativeHandle	nativeSemaphoreHandle;

			addSemaphore(vkB, *deviceB, semaphoresB, semaphoreHandlesB, timelineValuesB, true, timelineValuesA.back());

			getSemaphoreNative(vkA, deviceA, lastSemaphoreA, m_semaphoreHandleType, nativeSemaphoreHandle);
			importSemaphore(vkB, *deviceB, semaphoreHandlesB.back(), m_semaphoreHandleType, nativeSemaphoreHandle, 0u);

			addSemaphore(vkB, *deviceB, semaphoresB, semaphoreHandlesB, timelineValuesB, false, timelineValuesA.back());
		}

		// Submit writes, each in its own VkSubmitInfo. With binary
		// semaphores, submission don't wait on anything, with
		// timeline semaphores, submissions wait on a host signal
		// operation done below.
		{
			std::vector<VkTimelineSemaphoreSubmitInfoKHR>	timelineSubmitInfos;
			std::vector<VkSubmitInfo>						submitInfos;
			const deUint64									waitValue				= 1u;

			submitInfos.resize(iterations.size());
			timelineSubmitInfos.resize(iterations.size());

			for (deUint32 iterIdx = 0; iterIdx < iterations.size(); iterIdx++)
			{
				const VkTimelineSemaphoreSubmitInfoKHR	timelineSubmitInfo	=
				{
					VK_STRUCTURE_TYPE_TIMELINE_SEMAPHORE_SUBMIT_INFO_KHR,	// VkStructureType	sType;
					DE_NULL,												// const void*		pNext;
					m_semaphoreType == VK_SEMAPHORE_TYPE_TIMELINE_KHR ?
					1u: 0,													// deUint32			waitSemaphoreValueCount
					&waitValue,												// const deUint64*	pWaitSemaphoreValues
					1u,														// deUint32			signalSemaphoreValueCount
					&timelineValuesA[iterIdx],								// const deUint64*	pSignalSemaphoreValues
				};
				const VkSubmitInfo						submitInfo			=
				{
					VK_STRUCTURE_TYPE_SUBMIT_INFO,							// VkStructureType				sType;
					m_semaphoreType == VK_SEMAPHORE_TYPE_TIMELINE_KHR ?
					&timelineSubmitInfos[iterIdx] : DE_NULL,				// const void*					pNext;
					m_semaphoreType == VK_SEMAPHORE_TYPE_TIMELINE_KHR ?
					1u : 0u,												// deUint32						waitSemaphoreCount;
					&semaphoreHandlesA.front(),								// const VkSemaphore*			pWaitSemaphores;
					&stageBits[iterIdx],									// const VkPipelineStageFlags*	pWaitDstStageMask;
					1u,														// deUint32						commandBufferCount;
					&cmdBuffersA[iterIdx],									// const VkCommandBuffer*		pCommandBuffers;
					1u,														// deUint32						signalSemaphoreCount;
					&semaphoreHandlesA[iterIdx],							// const VkSemaphore*			pSignalSemaphores;
				};

				timelineSubmitInfos[iterIdx] = timelineSubmitInfo;
				submitInfos[iterIdx] = submitInfo;
			}

			VK_CHECK(vkA.queueSubmit(queueA, (deUint32)submitInfos.size(), &submitInfos[0], *fenceA));
		}

		// Submit reads, only waiting waiting on the last write
		// operations, ordering of signaling should guarantee that
		// when read operations kick in all writes have completed.
		{
			const VkTimelineSemaphoreSubmitInfoKHR	timelineSubmitInfo	=
			{
				VK_STRUCTURE_TYPE_TIMELINE_SEMAPHORE_SUBMIT_INFO_KHR,	// VkStructureType	sType;
				DE_NULL,												// const void*		pNext;
				1u,														// deUint32			waitSemaphoreValueCount
				&timelineValuesA.back(),								// const deUint64*	pWaitSemaphoreValues
				1u,														// deUint32			signalSemaphoreValueCount
				&timelineValuesB.back(),								// const deUint64*	pSignalSemaphoreValues
			};
			const VkSubmitInfo						submitInfo			=
			{
				VK_STRUCTURE_TYPE_SUBMIT_INFO,							// VkStructureType				sType;
				m_semaphoreType == VK_SEMAPHORE_TYPE_TIMELINE_KHR ?
				&timelineSubmitInfo : DE_NULL,							// const void*					pNext;
				1u,														// deUint32						waitSemaphoreCount;
				&semaphoreHandlesB.front(),								// const VkSemaphore*			pWaitSemaphores;
				&readStages,											// const VkPipelineStageFlags*	pWaitDstStageMask;
				1u,														// deUint32						commandBufferCount;
				&cmdBufferB,											// const VkCommandBuffer*		pCommandBuffers;
				1u,														// deUint32						signalSemaphoreCount;
				&semaphoreHandlesB.back(),								// const VkSemaphore*			pSignalSemaphores;
			};

			VK_CHECK(vkB.queueSubmit(queueB, 1, &submitInfo, *fenceB));

			if (m_semaphoreType == VK_SEMAPHORE_TYPE_TIMELINE_KHR)
			{
				const VkSemaphoreWaitInfo		waitInfo	=
				{
					VK_STRUCTURE_TYPE_SEMAPHORE_WAIT_INFO,	// VkStructureType			sType;
					DE_NULL,								// const void*				pNext;
					0u,										// VkSemaphoreWaitFlagsKHR	flags;
					1u,										// deUint32					semaphoreCount;
					&semaphoreHandlesB.back(),				// const VkSemaphore*		pSemaphores;
					&timelineValuesB.back(),				// const deUint64*			pValues;
				};

				// Unblock the whole lot.
				hostSignal(vkA, deviceA, semaphoreHandlesA.front(), 1);

				VK_CHECK(vkB.waitSemaphores(*deviceB, &waitInfo, ~0ull));
			}
			else
			{
				VK_CHECK(vkB.waitForFences(*deviceB, 1, &fenceB.get(), VK_TRUE, ~0ull));
			}
		}

		// Verify the result of the operations.
		for (deUint32 iterIdx = 0; iterIdx < iterations.size(); iterIdx++)
		{
			QueueSubmitOrderSharedIteration&	iter		= iterations[iterIdx];
			const Data							expected	= iter.writeOp->getData();
			const Data							actual		= iter.readOp->getData();

			if (isIndirectBuffer(iter.resourceA->getType()))
			{
				const deUint32 expectedValue = reinterpret_cast<const deUint32*>(expected.data)[0];
				const deUint32 actualValue   = reinterpret_cast<const deUint32*>(actual.data)[0];

				if (actualValue < expectedValue)
					return tcu::TestStatus::fail("Counter value is smaller than expected");
			}
			else
			{
				if (0 != deMemCmp(expected.data, actual.data, expected.size))
					return tcu::TestStatus::fail("Memory contents don't match");
			}
		}

		VK_CHECK(vkA.deviceWaitIdle(deviceA));
		VK_CHECK(vkB.deviceWaitIdle(*deviceB));

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

private:
	void addSemaphore (const DeviceInterface&			vk,
					   VkDevice							device,
					   std::vector<Move<VkSemaphore> >&	semaphores,
					   std::vector<VkSemaphore>&		semaphoreHandles,
					   std::vector<deUint64>&			timelineValues,
					   bool								exportable,
					   deUint64							firstTimelineValue)
	{
		Move<VkSemaphore>	semaphore;

		if (m_semaphoreType == VK_SEMAPHORE_TYPE_TIMELINE_KHR)
		{
			// Only allocate a single exportable semaphore.
			if (semaphores.empty())
			{
				semaphores.push_back(createExportableSemaphoreType(vk, device, m_semaphoreType, m_semaphoreHandleType));
			}
		}
		else
		{
			if (exportable)
				semaphores.push_back(createExportableSemaphoreType(vk, device, m_semaphoreType, m_semaphoreHandleType));
			else
				semaphores.push_back(createSemaphoreType(vk, device, m_semaphoreType));
		}

		semaphoreHandles.push_back(*semaphores.back());
		timelineValues.push_back((timelineValues.empty() ? firstTimelineValue : timelineValues.back()) + m_rng.getInt(1, 100));
	}

	bool isResourceExportable ()
	{
		const InstanceInterface&					vki				= m_context.getInstanceInterface();
		VkPhysicalDevice							physicalDevice	= m_context.getPhysicalDevice();

		if (m_resourceDesc.type == RESOURCE_TYPE_IMAGE)
		{
			const VkPhysicalDeviceExternalImageFormatInfo	externalInfo		=
			{
				VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_EXTERNAL_IMAGE_FORMAT_INFO,
				DE_NULL,
				m_memoryHandleType
			};
			const VkPhysicalDeviceImageFormatInfo2			imageFormatInfo		=
			{
				VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_IMAGE_FORMAT_INFO_2,
				&externalInfo,
				m_resourceDesc.imageFormat,
				m_resourceDesc.imageType,
				VK_IMAGE_TILING_OPTIMAL,
				m_readOpSupport->getInResourceUsageFlags() | m_writeOpSupport->getOutResourceUsageFlags(),
				0u
			};
			VkExternalImageFormatProperties					externalProperties	=
			{
				VK_STRUCTURE_TYPE_EXTERNAL_IMAGE_FORMAT_PROPERTIES,
				DE_NULL,
				{ 0u, 0u, 0u }
			};
			VkImageFormatProperties2						formatProperties	=
			{
				VK_STRUCTURE_TYPE_IMAGE_FORMAT_PROPERTIES_2,
				&externalProperties,
				{
					{ 0u, 0u, 0u },
					0u,
					0u,
					0u,
					0u,
				}
			};

			{
				const VkResult res = vki.getPhysicalDeviceImageFormatProperties2(physicalDevice, &imageFormatInfo, &formatProperties);

				if (res == VK_ERROR_FORMAT_NOT_SUPPORTED)
					return false;

				VK_CHECK(res); // Check other errors
			}

			if ((externalProperties.externalMemoryProperties.externalMemoryFeatures & VK_EXTERNAL_MEMORY_FEATURE_EXPORTABLE_BIT_KHR) == 0)
				return false;

			if ((externalProperties.externalMemoryProperties.externalMemoryFeatures & VK_EXTERNAL_MEMORY_FEATURE_IMPORTABLE_BIT_KHR) == 0)
				return false;

			return true;
		}
		else
		{
			const VkPhysicalDeviceExternalBufferInfo	info	=
			{
				VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_EXTERNAL_BUFFER_INFO,
				DE_NULL,

				0u,
				m_readOpSupport->getInResourceUsageFlags() | m_writeOpSupport->getOutResourceUsageFlags(),
				m_memoryHandleType
			};
			VkExternalBufferProperties					properties			=
			{
				VK_STRUCTURE_TYPE_EXTERNAL_BUFFER_PROPERTIES,
				DE_NULL,
				{ 0u, 0u, 0u}
			};
			vki.getPhysicalDeviceExternalBufferProperties(physicalDevice, &info, &properties);

			if ((properties.externalMemoryProperties.externalMemoryFeatures & VK_EXTERNAL_MEMORY_FEATURE_EXPORTABLE_BIT_KHR) == 0
				|| (properties.externalMemoryProperties.externalMemoryFeatures & VK_EXTERNAL_MEMORY_FEATURE_IMPORTABLE_BIT_KHR) == 0)
				return false;

			return true;
		}
	}

	SharedPtr<OperationSupport>					m_writeOpSupport;
	SharedPtr<OperationSupport>					m_readOpSupport;
	const ResourceDescription&					m_resourceDesc;
	VkExternalMemoryHandleTypeFlagBits			m_memoryHandleType;
	VkSemaphoreType								m_semaphoreType;
	VkExternalSemaphoreHandleTypeFlagBits		m_semaphoreHandleType;
	PipelineCacheData&							m_pipelineCacheData;
	de::Random									m_rng;
};

class QueueSubmitSignalOrderSharedTestCase : public TestCase
{
public:
	QueueSubmitSignalOrderSharedTestCase (tcu::TestContext&						testCtx,
										  const std::string&					name,
										  OperationName							writeOp,
										  OperationName							readOp,
										  const ResourceDescription&			resourceDesc,
										  VkExternalMemoryHandleTypeFlagBits	memoryHandleType,
										  VkSemaphoreType						semaphoreType,
										  VkExternalSemaphoreHandleTypeFlagBits	semaphoreHandleType,
										  PipelineCacheData&					pipelineCacheData)
		: TestCase				(testCtx, name.c_str(), "")
		, m_writeOpSupport		(makeOperationSupport(writeOp, resourceDesc).release())
		, m_readOpSupport		(makeOperationSupport(readOp, resourceDesc).release())
		, m_resourceDesc		(resourceDesc)
		, m_memoryHandleType	(memoryHandleType)
		, m_semaphoreType		(semaphoreType)
		, m_semaphoreHandleType	(semaphoreHandleType)
		, m_pipelineCacheData	(pipelineCacheData)
	{
	}

	virtual void checkSupport(Context& context) const
	{
		if (m_semaphoreType == VK_SEMAPHORE_TYPE_TIMELINE_KHR &&
			!context.getTimelineSemaphoreFeatures().timelineSemaphore)
			TCU_THROW(NotSupportedError, "Timeline semaphore not supported");

		if ((m_semaphoreHandleType == VK_EXTERNAL_SEMAPHORE_HANDLE_TYPE_SYNC_FD_BIT ||
			 m_semaphoreHandleType == VK_EXTERNAL_SEMAPHORE_HANDLE_TYPE_OPAQUE_FD_BIT) &&
			 !context.isDeviceFunctionalitySupported("VK_KHR_external_semaphore_fd"))
			TCU_THROW(NotSupportedError, "VK_KHR_external_semaphore_fd not supported");

		if ((m_semaphoreHandleType == VK_EXTERNAL_SEMAPHORE_HANDLE_TYPE_OPAQUE_WIN32_BIT ||
			 m_semaphoreHandleType == VK_EXTERNAL_SEMAPHORE_HANDLE_TYPE_OPAQUE_WIN32_KMT_BIT) &&
			!context.isDeviceFunctionalitySupported("VK_KHR_external_semaphore_win32"))
			TCU_THROW(NotSupportedError, "VK_KHR_external_semaphore_win32 not supported");
	}

	TestInstance* createInstance (Context& context) const
	{
		return new QueueSubmitSignalOrderSharedTestInstance(context,
															m_writeOpSupport,
															m_readOpSupport,
															m_resourceDesc,
															m_memoryHandleType,
															m_semaphoreType,
															m_semaphoreHandleType,
															m_pipelineCacheData);
	}

	void initPrograms (SourceCollections& programCollection) const
	{
		m_writeOpSupport->initPrograms(programCollection);
		m_readOpSupport->initPrograms(programCollection);
	}

private:
	SharedPtr<OperationSupport>				m_writeOpSupport;
	SharedPtr<OperationSupport>				m_readOpSupport;
	const ResourceDescription&				m_resourceDesc;
	VkExternalMemoryHandleTypeFlagBits		m_memoryHandleType;
	VkSemaphoreType							m_semaphoreType;
	VkExternalSemaphoreHandleTypeFlagBits	m_semaphoreHandleType;
	PipelineCacheData&						m_pipelineCacheData;
};

class QueueSubmitSignalOrderSharedTests : public tcu::TestCaseGroup
{
public:
	QueueSubmitSignalOrderSharedTests (tcu::TestContext& testCtx, VkSemaphoreType semaphoreType, const char *name)
		: tcu::TestCaseGroup	(testCtx, name, "Signal ordering of semaphores")
		, m_semaphoreType		(semaphoreType)
	{
	}

	void init (void)
	{
		static const OperationName	writeOps[]	=
		{
			OPERATION_NAME_WRITE_COPY_BUFFER,
			OPERATION_NAME_WRITE_COPY_BUFFER_TO_IMAGE,
			OPERATION_NAME_WRITE_COPY_IMAGE_TO_BUFFER,
			OPERATION_NAME_WRITE_COPY_IMAGE,
			OPERATION_NAME_WRITE_BLIT_IMAGE,
			OPERATION_NAME_WRITE_SSBO_VERTEX,
			OPERATION_NAME_WRITE_SSBO_TESSELLATION_CONTROL,
			OPERATION_NAME_WRITE_SSBO_TESSELLATION_EVALUATION,
			OPERATION_NAME_WRITE_SSBO_GEOMETRY,
			OPERATION_NAME_WRITE_SSBO_FRAGMENT,
			OPERATION_NAME_WRITE_SSBO_COMPUTE,
			OPERATION_NAME_WRITE_SSBO_COMPUTE_INDIRECT,
			OPERATION_NAME_WRITE_IMAGE_VERTEX,
			OPERATION_NAME_WRITE_IMAGE_TESSELLATION_CONTROL,
			OPERATION_NAME_WRITE_IMAGE_TESSELLATION_EVALUATION,
			OPERATION_NAME_WRITE_IMAGE_GEOMETRY,
			OPERATION_NAME_WRITE_IMAGE_FRAGMENT,
			OPERATION_NAME_WRITE_IMAGE_COMPUTE,
			OPERATION_NAME_WRITE_IMAGE_COMPUTE_INDIRECT,
		};
		static const OperationName	readOps[]	=
		{
			OPERATION_NAME_READ_COPY_BUFFER,
			OPERATION_NAME_READ_COPY_BUFFER_TO_IMAGE,
			OPERATION_NAME_READ_COPY_IMAGE_TO_BUFFER,
			OPERATION_NAME_READ_COPY_IMAGE,
			OPERATION_NAME_READ_BLIT_IMAGE,
			OPERATION_NAME_READ_UBO_VERTEX,
			OPERATION_NAME_READ_UBO_TESSELLATION_CONTROL,
			OPERATION_NAME_READ_UBO_TESSELLATION_EVALUATION,
			OPERATION_NAME_READ_UBO_GEOMETRY,
			OPERATION_NAME_READ_UBO_FRAGMENT,
			OPERATION_NAME_READ_UBO_COMPUTE,
			OPERATION_NAME_READ_UBO_COMPUTE_INDIRECT,
			OPERATION_NAME_READ_SSBO_VERTEX,
			OPERATION_NAME_READ_SSBO_TESSELLATION_CONTROL,
			OPERATION_NAME_READ_SSBO_TESSELLATION_EVALUATION,
			OPERATION_NAME_READ_SSBO_GEOMETRY,
			OPERATION_NAME_READ_SSBO_FRAGMENT,
			OPERATION_NAME_READ_SSBO_COMPUTE,
			OPERATION_NAME_READ_SSBO_COMPUTE_INDIRECT,
			OPERATION_NAME_READ_IMAGE_VERTEX,
			OPERATION_NAME_READ_IMAGE_TESSELLATION_CONTROL,
			OPERATION_NAME_READ_IMAGE_TESSELLATION_EVALUATION,
			OPERATION_NAME_READ_IMAGE_GEOMETRY,
			OPERATION_NAME_READ_IMAGE_FRAGMENT,
			OPERATION_NAME_READ_IMAGE_COMPUTE,
			OPERATION_NAME_READ_IMAGE_COMPUTE_INDIRECT,
			OPERATION_NAME_READ_INDIRECT_BUFFER_DRAW,
			OPERATION_NAME_READ_INDIRECT_BUFFER_DRAW_INDEXED,
			OPERATION_NAME_READ_INDIRECT_BUFFER_DISPATCH,
			OPERATION_NAME_READ_VERTEX_INPUT,
		};
		static const struct
		{
			VkExternalMemoryHandleTypeFlagBits		memoryType;
			VkExternalSemaphoreHandleTypeFlagBits	semaphoreType;
		}	exportCases[] =
		{
			// Only semaphore handle types having reference semantic
			// are valid for this test.
			{
				VK_EXTERNAL_MEMORY_HANDLE_TYPE_OPAQUE_FD_BIT,
				VK_EXTERNAL_SEMAPHORE_HANDLE_TYPE_OPAQUE_FD_BIT,
			},
			{
				VK_EXTERNAL_MEMORY_HANDLE_TYPE_OPAQUE_WIN32_KMT_BIT,
				VK_EXTERNAL_SEMAPHORE_HANDLE_TYPE_OPAQUE_WIN32_KMT_BIT,
			},
			{
				VK_EXTERNAL_MEMORY_HANDLE_TYPE_OPAQUE_WIN32_BIT,
				VK_EXTERNAL_SEMAPHORE_HANDLE_TYPE_OPAQUE_WIN32_BIT,
			},
		};

		for (deUint32 writeOpIdx = 0; writeOpIdx < DE_LENGTH_OF_ARRAY(writeOps); writeOpIdx++)
		for (deUint32 readOpIdx = 0; readOpIdx < DE_LENGTH_OF_ARRAY(readOps); readOpIdx++)
		{
			const OperationName	writeOp		= writeOps[writeOpIdx];
			const OperationName	readOp		= readOps[readOpIdx];
			const std::string	opGroupName = getOperationName(writeOp) + "_" + getOperationName(readOp);
			bool				empty		= true;

			de::MovePtr<tcu::TestCaseGroup> opGroup	(new tcu::TestCaseGroup(m_testCtx, opGroupName.c_str(), ""));

			for (int resourceNdx = 0; resourceNdx < DE_LENGTH_OF_ARRAY(s_resources); ++resourceNdx)
			{
				const ResourceDescription&	resource	= s_resources[resourceNdx];

				if (isResourceSupported(writeOp, resource) && isResourceSupported(readOp, resource))
				{
					for (deUint32 exportIdx = 0; exportIdx < DE_LENGTH_OF_ARRAY(exportCases); exportIdx++)
					{
						std::string					caseName	= getResourceName(resource) + "_" +
							externalSemaphoreTypeToName(exportCases[exportIdx].semaphoreType);

						opGroup->addChild(new QueueSubmitSignalOrderSharedTestCase(m_testCtx,
																				   caseName,
																				   writeOp,
																				   readOp,
																				   resource,
																				   exportCases[exportIdx].memoryType,
																				   m_semaphoreType,
																				   exportCases[exportIdx].semaphoreType,
																				   m_pipelineCacheData));
						empty = false;
					}
				}
			}
			if (!empty)
				addChild(opGroup.release());
		}
	}

	void deinit (void)
	{
		cleanupGroup();
	}

private:
	VkSemaphoreType		m_semaphoreType;
	// synchronization.op tests share pipeline cache data to speed up test
	// execution.
	PipelineCacheData	m_pipelineCacheData;
};

struct QueueSubmitOrderIteration
{
	QueueSubmitOrderIteration() {}
	~QueueSubmitOrderIteration() {}

	SharedPtr<Resource>			resource;

	SharedPtr<Operation>		writeOp;
	SharedPtr<Operation>		readOp;
};

// Verifies the signaling order of the semaphores in multiple
// VkSubmitInfo given to vkQueueSubmit() with queueA & queueB from the
// same VkDevice.
//
// vkQueueSubmit(queueA, [write0, write1, write2, ..., write6])
// vkQueueSubmit(queueB, [read0-6])
//
// With read0-6 waiting on write6, all the data should be available
// for reading given that signal operations are supposed to happen in
// order.
class QueueSubmitSignalOrderTestInstance : public TestInstance
{
public:
	QueueSubmitSignalOrderTestInstance (Context&									context,
										const SharedPtr<OperationSupport>			writeOpSupport,
										const SharedPtr<OperationSupport>			readOpSupport,
										const ResourceDescription&					resourceDesc,
										VkSemaphoreType								semaphoreType,
										PipelineCacheData&							pipelineCacheData)
		: TestInstance			(context)
		, m_writeOpSupport		(writeOpSupport)
		, m_readOpSupport		(readOpSupport)
		, m_resourceDesc		(resourceDesc)
		, m_semaphoreType		(semaphoreType)
		, m_device				(SingletonDevice::getDevice(context))
		, m_deviceInterface		(context.getPlatformInterface(), context.getInstance(), *m_device)
		, m_allocator			(new SimpleAllocator(m_deviceInterface,
													 *m_device,
													 getPhysicalDeviceMemoryProperties(context.getInstanceInterface(),
																					   context.getPhysicalDevice())))
		, m_operationContext	(new OperationContext(context, pipelineCacheData, m_deviceInterface, *m_device, *m_allocator))
		, m_queueA				(DE_NULL)
		, m_queueB				(DE_NULL)
		, m_rng					(1234)

	{
		const std::vector<VkQueueFamilyProperties> queueFamilyProperties	= getPhysicalDeviceQueueFamilyProperties(context.getInstanceInterface(),
																													 context.getPhysicalDevice());

		if (m_semaphoreType == VK_SEMAPHORE_TYPE_TIMELINE_KHR &&
			!context.getTimelineSemaphoreFeatures().timelineSemaphore)
			TCU_THROW(NotSupportedError, "Timeline semaphore not supported");

		VkQueueFlags writeOpQueueFlags = m_writeOpSupport->getQueueFlags(*m_operationContext);
		for (deUint32 familyIdx = 0; familyIdx < queueFamilyProperties.size(); familyIdx++) {
			if (((queueFamilyProperties[familyIdx].queueFlags & writeOpQueueFlags) == writeOpQueueFlags) ||
			((writeOpQueueFlags == VK_QUEUE_TRANSFER_BIT) &&
			(((queueFamilyProperties[familyIdx].queueFlags & VK_QUEUE_GRAPHICS_BIT) == VK_QUEUE_GRAPHICS_BIT) ||
			((queueFamilyProperties[familyIdx].queueFlags & VK_QUEUE_COMPUTE_BIT) == VK_QUEUE_COMPUTE_BIT)))) {
				m_queueA = getDeviceQueue(m_deviceInterface, *m_device, familyIdx, 0);
				m_queueFamilyIndexA = familyIdx;
				break;
			}
		}
		if (m_queueA == DE_NULL)
			TCU_THROW(NotSupportedError, "No queue supporting write operation");

		VkQueueFlags readOpQueueFlags = m_readOpSupport->getQueueFlags(*m_operationContext);
		for (deUint32 familyIdx = 0; familyIdx < queueFamilyProperties.size(); familyIdx++) {
			if (((queueFamilyProperties[familyIdx].queueFlags & readOpQueueFlags) == readOpQueueFlags) ||
			((readOpQueueFlags == VK_QUEUE_TRANSFER_BIT) &&
			(((queueFamilyProperties[familyIdx].queueFlags & VK_QUEUE_GRAPHICS_BIT) == VK_QUEUE_GRAPHICS_BIT) ||
			((queueFamilyProperties[familyIdx].queueFlags & VK_QUEUE_COMPUTE_BIT) == VK_QUEUE_COMPUTE_BIT)))) {
				for (deUint32 queueIdx = 0; queueIdx < queueFamilyProperties[familyIdx].queueCount; queueIdx++) {
					VkQueue queue = getDeviceQueue(m_deviceInterface, *m_device, familyIdx, queueIdx);

					if (queue == m_queueA)
						continue;

					m_queueB = queue;
					m_queueFamilyIndexB = familyIdx;
					break;
				}

				if (m_queueB != DE_NULL)
					break;
			}
		}
		if (m_queueB == DE_NULL)
			TCU_THROW(NotSupportedError, "No queue supporting read operation");
	}

	tcu::TestStatus iterate (void)
	{
		const VkDevice&										device						= *m_device;
		const DeviceInterface&								vk							= m_deviceInterface;
		Unique<VkFence>										fence						(createFence(vk, device));
		const Unique<VkCommandPool>							cmdPoolA					(createCommandPool(vk, device, VK_COMMAND_POOL_CREATE_RESET_COMMAND_BUFFER_BIT, m_queueFamilyIndexA));
		const Unique<VkCommandPool>							cmdPoolB					(createCommandPool(vk, device, VK_COMMAND_POOL_CREATE_RESET_COMMAND_BUFFER_BIT, m_queueFamilyIndexB));
		std::vector<SharedPtr<Move<VkCommandBuffer> > >		ptrCmdBuffersA;
		SharedPtr<Move<VkCommandBuffer> >					ptrCmdBufferB;
		std::vector<VkCommandBuffer>						cmdBuffersA;
		VkCommandBuffer										cmdBufferB;
		std::vector<Move<VkSemaphore> >						semaphoresA;
		std::vector<Move<VkSemaphore> >						semaphoresB;
		std::vector<VkSemaphore>							semaphoreHandlesA;
		std::vector<VkSemaphore>							semaphoreHandlesB;
		std::vector<deUint64>								timelineValuesA;
		std::vector<deUint64>								timelineValuesB;
		std::vector<QueueSubmitOrderIteration>				iterations;
		std::vector<VkPipelineStageFlags>					stageBits;
		std::vector<deUint32>								queueFamilies;

		queueFamilies.push_back(m_queueFamilyIndexA);
		queueFamilies.push_back(m_queueFamilyIndexB);

		// Create a dozen of set of write/read operations.
		iterations.resize(12);
		for (deUint32 iterIdx = 0; iterIdx < iterations.size(); iterIdx++)
		{
			QueueSubmitOrderIteration&		iter				= iterations[iterIdx];

			iter.resource	= makeSharedPtr(new Resource(*m_operationContext,
														 m_resourceDesc,
														 m_writeOpSupport->getOutResourceUsageFlags() |
														 m_readOpSupport->getInResourceUsageFlags(),
														 VK_SHARING_MODE_EXCLUSIVE,
														 queueFamilies));

			iter.writeOp = makeSharedPtr(m_writeOpSupport->build(*m_operationContext,
																 *iter.resource));
			iter.readOp = makeSharedPtr(m_readOpSupport->build(*m_operationContext,
															   *iter.resource));
		}

		// Record each write operation into its own command buffer.
		for (deUint32 iterIdx = 0; iterIdx < iterations.size(); iterIdx++)
		{
			QueueSubmitOrderIteration&	iter	= iterations[iterIdx];

			ptrCmdBuffersA.push_back(makeVkSharedPtr(makeCommandBuffer(vk, device, *cmdPoolA)));
			cmdBuffersA.push_back(**(ptrCmdBuffersA.back()));

			beginCommandBuffer(vk, cmdBuffersA.back());
			iter.writeOp->recordCommands(cmdBuffersA.back());

			{
				const SyncInfo	writeSync	= iter.writeOp->getOutSyncInfo();
				const SyncInfo	readSync	= iter.readOp->getInSyncInfo();
				const Resource&	resource	= *iter.resource;

				if (resource.getType() == RESOURCE_TYPE_IMAGE)
				{
					DE_ASSERT(writeSync.imageLayout != VK_IMAGE_LAYOUT_UNDEFINED);
					DE_ASSERT(readSync.imageLayout != VK_IMAGE_LAYOUT_UNDEFINED);
					const VkImageMemoryBarrier barrier =  makeImageMemoryBarrier(writeSync.accessMask, readSync.accessMask,
																				 writeSync.imageLayout, readSync.imageLayout,
																				 resource.getImage().handle,
																				 resource.getImage().subresourceRange);
					vk.cmdPipelineBarrier(cmdBuffersA.back(), writeSync.stageMask, readSync.stageMask, (VkDependencyFlags)0,
										  0u, (const VkMemoryBarrier*)DE_NULL, 0u, (const VkBufferMemoryBarrier*)DE_NULL, 1u, &barrier);
				}
				else
				{
					const VkBufferMemoryBarrier barrier = makeBufferMemoryBarrier(writeSync.accessMask, readSync.accessMask,
																				  resource.getBuffer().handle, 0, VK_WHOLE_SIZE);
					vk.cmdPipelineBarrier(cmdBuffersA.back(), writeSync.stageMask, readSync.stageMask, (VkDependencyFlags)0,
										  0u, (const VkMemoryBarrier*)DE_NULL, 1u, &barrier, 0u, (const VkImageMemoryBarrier*)DE_NULL);
				}

				stageBits.push_back(writeSync.stageMask);
			}

			endCommandBuffer(vk, cmdBuffersA.back());

			addSemaphore(vk, device, semaphoresA, semaphoreHandlesA, timelineValuesA, 2u);
		}

		DE_ASSERT(stageBits.size() == iterations.size());
		DE_ASSERT(semaphoreHandlesA.size() == iterations.size());

		// Record all read operations into a single command buffer and track the union of their execution stages.
		VkPipelineStageFlags readStages = 0;
		ptrCmdBufferB = makeVkSharedPtr(makeCommandBuffer(vk, device, *cmdPoolB));
		cmdBufferB = **(ptrCmdBufferB);
		beginCommandBuffer(vk, cmdBufferB);
		for (deUint32 iterIdx = 0; iterIdx < iterations.size(); iterIdx++)
		{
			QueueSubmitOrderIteration& iter = iterations[iterIdx];
			readStages |= iter.readOp->getInSyncInfo().stageMask;
			iter.readOp->recordCommands(cmdBufferB);
		}
		endCommandBuffer(vk, cmdBufferB);

		addSemaphore(vk, device, semaphoresB, semaphoreHandlesB, timelineValuesB, timelineValuesA.back());

		// Submit writes, each in its own VkSubmitInfo. With binary
		// semaphores, submission don't wait on anything, with
		// timeline semaphores, submissions wait on a host signal
		// operation done below.
		{
			std::vector<VkTimelineSemaphoreSubmitInfoKHR>	timelineSubmitInfos;
			std::vector<VkSubmitInfo>						submitInfos;
			const deUint64									waitValue				= 1u;

			submitInfos.resize(iterations.size());
			timelineSubmitInfos.resize(iterations.size());

			for (deUint32 iterIdx = 0; iterIdx < iterations.size(); iterIdx++)
			{
				const VkTimelineSemaphoreSubmitInfoKHR	timelineSubmitInfo	=
				{
					VK_STRUCTURE_TYPE_TIMELINE_SEMAPHORE_SUBMIT_INFO_KHR,	// VkStructureType	sType;
					DE_NULL,												// const void*		pNext;
					m_semaphoreType == VK_SEMAPHORE_TYPE_TIMELINE_KHR ?
					1u : 0u,												// deUint32			waitSemaphoreValueCount
					&waitValue,												// const deUint64*	pWaitSemaphoreValues
					1u,														// deUint32			signalSemaphoreValueCount
					&timelineValuesA[iterIdx],								// const deUint64*	pSignalSemaphoreValues
				};
				const VkSubmitInfo						submitInfo			=
				{
					VK_STRUCTURE_TYPE_SUBMIT_INFO,							// VkStructureType				sType;
					m_semaphoreType == VK_SEMAPHORE_TYPE_TIMELINE_KHR ?
					&timelineSubmitInfos[iterIdx] : DE_NULL,				// const void*					pNext;
					m_semaphoreType == VK_SEMAPHORE_TYPE_TIMELINE_KHR ?
					1u : 0u,												// deUint32						waitSemaphoreCount;
					&semaphoreHandlesA.front(),								// const VkSemaphore*			pWaitSemaphores;
					&stageBits[iterIdx],									// const VkPipelineStageFlags*	pWaitDstStageMask;
					1u,														// deUint32						commandBufferCount;
					&cmdBuffersA[iterIdx],									// const VkCommandBuffer*		pCommandBuffers;
					1u,														// deUint32						signalSemaphoreCount;
					&semaphoreHandlesA[iterIdx],							// const VkSemaphore*			pSignalSemaphores;
				};

				timelineSubmitInfos[iterIdx] = timelineSubmitInfo;
				submitInfos[iterIdx] = submitInfo;
			}

			VK_CHECK(vk.queueSubmit(m_queueA, (deUint32)submitInfos.size(), &submitInfos[0], DE_NULL));
		}

		// Submit reads, only waiting waiting on the last write
		// operations, ordering of signaling should guarantee that
		// when read operations kick in all writes have completed.
		{
			const VkTimelineSemaphoreSubmitInfoKHR	timelineSubmitInfo	=
			{
				VK_STRUCTURE_TYPE_TIMELINE_SEMAPHORE_SUBMIT_INFO_KHR,	// VkStructureType	sType;
				DE_NULL,												// const void*		pNext;
				1u,														// deUint32			waitSemaphoreValueCount
				&timelineValuesA.back(),								// const deUint64*	pWaitSemaphoreValues
				1u,														// deUint32			signalSemaphoreValueCount
				&timelineValuesB.back(),								// const deUint64*	pSignalSemaphoreValues
			};
			const VkSubmitInfo						submitInfo			=
			{
				VK_STRUCTURE_TYPE_SUBMIT_INFO,							// VkStructureType				sType;
				m_semaphoreType == VK_SEMAPHORE_TYPE_TIMELINE_KHR ?
				&timelineSubmitInfo : DE_NULL,							// const void*					pNext;
				1u,														// deUint32						waitSemaphoreCount;
				&semaphoreHandlesA.back(),								// const VkSemaphore*			pWaitSemaphores;
				&readStages,											// const VkPipelineStageFlags*	pWaitDstStageMask;
				1u,														// deUint32						commandBufferCount;
				&cmdBufferB,											// const VkCommandBuffer*		pCommandBuffers;
				1u,														// deUint32						signalSemaphoreCount;
				&semaphoreHandlesB.back(),								// const VkSemaphore*			pSignalSemaphores;
			};

			VK_CHECK(vk.queueSubmit(m_queueB, 1, &submitInfo, *fence));

			if (m_semaphoreType == VK_SEMAPHORE_TYPE_TIMELINE_KHR)
			{
				const VkSemaphoreWaitInfo		waitInfo	=
				{
					VK_STRUCTURE_TYPE_SEMAPHORE_WAIT_INFO,	// VkStructureType			sType;
					DE_NULL,								// const void*				pNext;
					0u,										// VkSemaphoreWaitFlagsKHR	flags;
					1u,										// deUint32					semaphoreCount;
					&semaphoreHandlesB.back(),				// const VkSemaphore*		pSemaphores;
					&timelineValuesB.back(),				// const deUint64*			pValues;
				};

				// Unblock the whole lot.
				hostSignal(vk, device, semaphoreHandlesA.front(), 1);

				VK_CHECK(vk.waitSemaphores(device, &waitInfo, ~0ull));
			}
			else
			{
				VK_CHECK(vk.waitForFences(device, 1, &fence.get(), VK_TRUE, ~0ull));
			}
		}

		// Verify the result of the operations.
		for (deUint32 iterIdx = 0; iterIdx < iterations.size(); iterIdx++)
		{
			QueueSubmitOrderIteration&		iter		= iterations[iterIdx];
			const Data						expected	= iter.writeOp->getData();
			const Data						actual		= iter.readOp->getData();

			if (isIndirectBuffer(iter.resource->getType()))
			{
				const deUint32 expectedValue = reinterpret_cast<const deUint32*>(expected.data)[0];
				const deUint32 actualValue   = reinterpret_cast<const deUint32*>(actual.data)[0];

				if (actualValue < expectedValue)
					return tcu::TestStatus::fail("Counter value is smaller than expected");
			}
			else
			{
				if (0 != deMemCmp(expected.data, actual.data, expected.size))
					return tcu::TestStatus::fail("Memory contents don't match");
			}
		}

		VK_CHECK(vk.deviceWaitIdle(device));

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

private:
	void addSemaphore (const DeviceInterface&			vk,
					   VkDevice							device,
					   std::vector<Move<VkSemaphore> >&	semaphores,
					   std::vector<VkSemaphore>&		semaphoreHandles,
					   std::vector<deUint64>&			timelineValues,
					   deUint64							firstTimelineValue)
	{
		Move<VkSemaphore>	semaphore;

		if (m_semaphoreType == VK_SEMAPHORE_TYPE_TIMELINE_KHR)
		{
			// Only allocate a single exportable semaphore.
			if (semaphores.empty())
			{
				semaphores.push_back(createSemaphoreType(vk, device, m_semaphoreType));
			}
		}
		else
		{
			semaphores.push_back(createSemaphoreType(vk, device, m_semaphoreType));
		}

		semaphoreHandles.push_back(*semaphores.back());
		timelineValues.push_back((timelineValues.empty() ? firstTimelineValue : timelineValues.back()) + m_rng.getInt(1, 100));
	}

	SharedPtr<OperationSupport>					m_writeOpSupport;
	SharedPtr<OperationSupport>					m_readOpSupport;
	const ResourceDescription&					m_resourceDesc;
	VkSemaphoreType								m_semaphoreType;
	const Unique<VkDevice>&						m_device;
	const DeviceDriver							m_deviceInterface;
	UniquePtr<SimpleAllocator>					m_allocator;
	UniquePtr<OperationContext>					m_operationContext;
	VkQueue										m_queueA;
	VkQueue										m_queueB;
	deUint32									m_queueFamilyIndexA;
	deUint32									m_queueFamilyIndexB;
	de::Random									m_rng;
};

class QueueSubmitSignalOrderTestCase : public TestCase
{
public:
	QueueSubmitSignalOrderTestCase (tcu::TestContext&			testCtx,
									const std::string&			name,
									OperationName				writeOp,
									OperationName				readOp,
									const ResourceDescription&	resourceDesc,
									VkSemaphoreType				semaphoreType,
									PipelineCacheData&			pipelineCacheData)
		: TestCase				(testCtx, name.c_str(), "")
		, m_writeOpSupport		(makeOperationSupport(writeOp, resourceDesc).release())
		, m_readOpSupport		(makeOperationSupport(readOp, resourceDesc).release())
		, m_resourceDesc		(resourceDesc)
		, m_semaphoreType		(semaphoreType)
		, m_pipelineCacheData	(pipelineCacheData)
	{
	}

	virtual void checkSupport(Context& context) const
	{
		if (m_semaphoreType == VK_SEMAPHORE_TYPE_TIMELINE_KHR &&
			!context.getTimelineSemaphoreFeatures().timelineSemaphore)
			TCU_THROW(NotSupportedError, "Timeline semaphore not supported");
	}

	TestInstance* createInstance (Context& context) const
	{
		return new QueueSubmitSignalOrderTestInstance(context,
													  m_writeOpSupport,
													  m_readOpSupport,
													  m_resourceDesc,
													  m_semaphoreType,
													  m_pipelineCacheData);
	}

	void initPrograms (SourceCollections& programCollection) const
	{
		m_writeOpSupport->initPrograms(programCollection);
		m_readOpSupport->initPrograms(programCollection);
	}

private:
	SharedPtr<OperationSupport>				m_writeOpSupport;
	SharedPtr<OperationSupport>				m_readOpSupport;
	const ResourceDescription&				m_resourceDesc;
	VkSemaphoreType							m_semaphoreType;
	PipelineCacheData&						m_pipelineCacheData;
};

class QueueSubmitSignalOrderTests : public tcu::TestCaseGroup
{
public:
	QueueSubmitSignalOrderTests (tcu::TestContext& testCtx, VkSemaphoreType semaphoreType, const char *name)
		: tcu::TestCaseGroup	(testCtx, name, "Signal ordering of semaphores")
		, m_semaphoreType		(semaphoreType)
	{
	}

	void init (void)
	{
		static const OperationName	writeOps[]	=
		{
			OPERATION_NAME_WRITE_COPY_BUFFER,
			OPERATION_NAME_WRITE_COPY_BUFFER_TO_IMAGE,
			OPERATION_NAME_WRITE_COPY_IMAGE_TO_BUFFER,
			OPERATION_NAME_WRITE_COPY_IMAGE,
			OPERATION_NAME_WRITE_BLIT_IMAGE,
			OPERATION_NAME_WRITE_SSBO_VERTEX,
			OPERATION_NAME_WRITE_SSBO_TESSELLATION_CONTROL,
			OPERATION_NAME_WRITE_SSBO_TESSELLATION_EVALUATION,
			OPERATION_NAME_WRITE_SSBO_GEOMETRY,
			OPERATION_NAME_WRITE_SSBO_FRAGMENT,
			OPERATION_NAME_WRITE_SSBO_COMPUTE,
			OPERATION_NAME_WRITE_SSBO_COMPUTE_INDIRECT,
			OPERATION_NAME_WRITE_IMAGE_VERTEX,
			OPERATION_NAME_WRITE_IMAGE_TESSELLATION_CONTROL,
			OPERATION_NAME_WRITE_IMAGE_TESSELLATION_EVALUATION,
			OPERATION_NAME_WRITE_IMAGE_GEOMETRY,
			OPERATION_NAME_WRITE_IMAGE_FRAGMENT,
			OPERATION_NAME_WRITE_IMAGE_COMPUTE,
			OPERATION_NAME_WRITE_IMAGE_COMPUTE_INDIRECT,
		};
		static const OperationName	readOps[]	=
		{
			OPERATION_NAME_READ_COPY_BUFFER,
			OPERATION_NAME_READ_COPY_BUFFER_TO_IMAGE,
			OPERATION_NAME_READ_COPY_IMAGE_TO_BUFFER,
			OPERATION_NAME_READ_COPY_IMAGE,
			OPERATION_NAME_READ_BLIT_IMAGE,
			OPERATION_NAME_READ_UBO_VERTEX,
			OPERATION_NAME_READ_UBO_TESSELLATION_CONTROL,
			OPERATION_NAME_READ_UBO_TESSELLATION_EVALUATION,
			OPERATION_NAME_READ_UBO_GEOMETRY,
			OPERATION_NAME_READ_UBO_FRAGMENT,
			OPERATION_NAME_READ_UBO_COMPUTE,
			OPERATION_NAME_READ_UBO_COMPUTE_INDIRECT,
			OPERATION_NAME_READ_SSBO_VERTEX,
			OPERATION_NAME_READ_SSBO_TESSELLATION_CONTROL,
			OPERATION_NAME_READ_SSBO_TESSELLATION_EVALUATION,
			OPERATION_NAME_READ_SSBO_GEOMETRY,
			OPERATION_NAME_READ_SSBO_FRAGMENT,
			OPERATION_NAME_READ_SSBO_COMPUTE,
			OPERATION_NAME_READ_SSBO_COMPUTE_INDIRECT,
			OPERATION_NAME_READ_IMAGE_VERTEX,
			OPERATION_NAME_READ_IMAGE_TESSELLATION_CONTROL,
			OPERATION_NAME_READ_IMAGE_TESSELLATION_EVALUATION,
			OPERATION_NAME_READ_IMAGE_GEOMETRY,
			OPERATION_NAME_READ_IMAGE_FRAGMENT,
			OPERATION_NAME_READ_IMAGE_COMPUTE,
			OPERATION_NAME_READ_IMAGE_COMPUTE_INDIRECT,
			OPERATION_NAME_READ_INDIRECT_BUFFER_DRAW,
			OPERATION_NAME_READ_INDIRECT_BUFFER_DRAW_INDEXED,
			OPERATION_NAME_READ_INDIRECT_BUFFER_DISPATCH,
			OPERATION_NAME_READ_VERTEX_INPUT,
		};

		for (deUint32 writeOpIdx = 0; writeOpIdx < DE_LENGTH_OF_ARRAY(writeOps); writeOpIdx++)
		for (deUint32 readOpIdx = 0; readOpIdx < DE_LENGTH_OF_ARRAY(readOps); readOpIdx++)
		{
			const OperationName	writeOp		= writeOps[writeOpIdx];
			const OperationName	readOp		= readOps[readOpIdx];
			const std::string	opGroupName = getOperationName(writeOp) + "_" + getOperationName(readOp);
			bool				empty		= true;

			de::MovePtr<tcu::TestCaseGroup> opGroup	(new tcu::TestCaseGroup(m_testCtx, opGroupName.c_str(), ""));

			for (int resourceNdx = 0; resourceNdx < DE_LENGTH_OF_ARRAY(s_resources); ++resourceNdx)
			{
				const ResourceDescription&	resource	= s_resources[resourceNdx];

				if (isResourceSupported(writeOp, resource) && isResourceSupported(readOp, resource))
				{
					opGroup->addChild(new QueueSubmitSignalOrderTestCase(m_testCtx,
																		 getResourceName(resource),
																		 writeOp,
																		 readOp,
																		 resource,
																		 m_semaphoreType,
																		 m_pipelineCacheData));
					empty = false;
				}
			}
			if (!empty)
				addChild(opGroup.release());
		}
	}

	void deinit (void)
	{
		cleanupGroup();
	}

private:
	VkSemaphoreType		m_semaphoreType;
	// synchronization.op tests share pipeline cache data to speed up test
	// execution.
	PipelineCacheData	m_pipelineCacheData;
};

} // anonymous

tcu::TestCaseGroup* createSignalOrderTests (tcu::TestContext& testCtx)
{
	de::MovePtr<tcu::TestCaseGroup> orderingTests(new tcu::TestCaseGroup(testCtx, "signal_order", "Signal ordering tests"));

	orderingTests->addChild(new QueueSubmitSignalOrderTests(testCtx, VK_SEMAPHORE_TYPE_BINARY_KHR, "binary_semaphore"));
	orderingTests->addChild(new QueueSubmitSignalOrderTests(testCtx, VK_SEMAPHORE_TYPE_TIMELINE_KHR, "timeline_semaphore"));
	orderingTests->addChild(new QueueSubmitSignalOrderSharedTests(testCtx, VK_SEMAPHORE_TYPE_BINARY_KHR, "shared_binary_semaphore"));
	orderingTests->addChild(new QueueSubmitSignalOrderSharedTests(testCtx, VK_SEMAPHORE_TYPE_TIMELINE_KHR, "shared_timeline_semaphore"));

	return orderingTests.release();
}

} // synchronization
} // vkt