diff --git a/include/vk_mem_alloc.h b/include/vk_mem_alloc.h index 0278f5c..60dcea6 100644 --- a/include/vk_mem_alloc.h +++ b/include/vk_mem_alloc.h @@ -6077,41 +6077,85 @@ public: VmaWin32Handle(HANDLE hHandle) noexcept : m_hHandle(hHandle) { } ~VmaWin32Handle() noexcept { if (m_hHandle != VMA_NULL) { ::CloseHandle(m_hHandle); } } VMA_CLASS_NO_COPY_NO_MOVE(VmaWin32Handle) + public: - VkResult Init(VkDevice device, VkDeviceMemory memory, decltype(&vkGetMemoryWin32HandleKHR) vkGetMemoryWin32HandleKHR) noexcept + // Strengthened + VkResult GetHandle(VkDevice device, VkDeviceMemory memory, decltype(&vkGetMemoryWin32HandleKHR) pvkGetMemoryWin32HandleKHR, HANDLE hTargetProcess, HANDLE* pHandle) noexcept + { + *pHandle = VMA_NULL; + // We only care about atomicity of handle retrieval, not about memory order. + HANDLE handle = m_hHandle.load(std::memory_order_relaxed); + + // Try to get handle first. + if (handle != VMA_NULL) + { + *pHandle = Duplicate(hTargetProcess); + return VK_SUCCESS; + } + + HANDLE hCreatedHandle = VMA_NULL; + + // If failed, try to create it. + VkResult res = Create(device, memory, pvkGetMemoryWin32HandleKHR, &hCreatedHandle); + if (res == VK_SUCCESS) + { + // Successfully created handle, try to set it. + if (!m_hHandle.compare_exchange_strong(handle, hCreatedHandle, std::memory_order_relaxed)) + { + // AMD workaround + if (hCreatedHandle != m_hHandle.load(std::memory_order_relaxed)) + { + ::CloseHandle(hCreatedHandle); + } + } + } + + // If somehow it was set in the meantime, return it. + if (m_hHandle.load(std::memory_order_relaxed)) + { + *pHandle = Duplicate(hTargetProcess); + return VK_SUCCESS; + } + return res; +} + + operator bool() const noexcept { return m_hHandle != VMA_NULL; } +private: + // Not atomic + static VkResult Create(VkDevice device, VkDeviceMemory memory, decltype(&vkGetMemoryWin32HandleKHR) pvkGetMemoryWin32HandleKHR, HANDLE* pHandle) noexcept { VkResult res = VK_ERROR_FEATURE_NOT_PRESENT; - if (vkGetMemoryWin32HandleKHR != VMA_NULL) + if (pvkGetMemoryWin32HandleKHR != VMA_NULL) { VkMemoryGetWin32HandleInfoKHR handleInfo{ }; handleInfo.sType = VK_STRUCTURE_TYPE_MEMORY_GET_WIN32_HANDLE_INFO_KHR; handleInfo.memory = memory; handleInfo.handleType = VK_EXTERNAL_MEMORY_HANDLE_TYPE_OPAQUE_WIN32_BIT_KHR; - res = vkGetMemoryWin32HandleKHR(device, &handleInfo, &m_hHandle); + res = pvkGetMemoryWin32HandleKHR(device, &handleInfo, pHandle); } return res; } HANDLE Duplicate(HANDLE hTargetProcess = VMA_NULL) const noexcept { - if (!m_hHandle) - return m_hHandle; + HANDLE handle = m_hHandle.load(std::memory_order_relaxed); + if (!handle) + return handle; HANDLE hCurrentProcess = ::GetCurrentProcess(); HANDLE hDupHandle = VMA_NULL; - if (!::DuplicateHandle(hCurrentProcess, m_hHandle, hTargetProcess ? hTargetProcess : hCurrentProcess, &hDupHandle, 0, FALSE, DUPLICATE_SAME_ACCESS)) + if (!::DuplicateHandle(hCurrentProcess, handle, hTargetProcess ? hTargetProcess : hCurrentProcess, &hDupHandle, 0, FALSE, DUPLICATE_SAME_ACCESS)) { VMA_ASSERT(0 && "Failed to duplicate handle."); } return hDupHandle; } - operator bool() const noexcept { return m_hHandle != VMA_NULL; } - private: - HANDLE m_hHandle; + std::atomic m_hHandle; }; #else class VmaWin32Handle { + // ABI compatibility void* placeholder = VMA_NULL; }; #endif // VK_USE_PLATFORM_WIN32_KHR @@ -6186,6 +6230,7 @@ public: #ifdef VK_USE_PLATFORM_WIN32_KHR VkResult CreateWin32Handle( const VmaAllocator hAllocator, + decltype(&vkGetMemoryWin32HandleKHR) pvkGetMemoryWin32HandleKHR, HANDLE hTargetProcess, HANDLE* pHandle)noexcept; #endif // VK_USE_PLATFORM_WIN32_KHR @@ -6310,7 +6355,7 @@ private: void* m_pMappedData; // Not null means memory is mapped. VmaAllocation_T* m_Prev; VmaAllocation_T* m_Next; - VmaWin32Handle m_Handle; // Win32 handle + mutable VmaWin32Handle m_Handle; // Win32 handle }; union { @@ -10740,34 +10785,10 @@ VkResult VmaDeviceMemoryBlock::BindImageMemory( } #ifdef VK_USE_PLATFORM_WIN32_KHR -VkResult VmaDeviceMemoryBlock::CreateWin32Handle(const VmaAllocator hAllocator, HANDLE hTargetProcess, HANDLE* pHandle) noexcept +VkResult VmaDeviceMemoryBlock::CreateWin32Handle(const VmaAllocator hAllocator, decltype(&vkGetMemoryWin32HandleKHR) pvkGetMemoryWin32HandleKHR, HANDLE hTargetProcess, HANDLE* pHandle) noexcept { VMA_ASSERT(pHandle); - *pHandle = VMA_NULL; - - // relieves the mutex - if (m_Handle) - { - *pHandle = m_Handle.Duplicate(hTargetProcess); - return VK_SUCCESS; - } - - VmaMutexLock lock(m_MapAndBindMutex, hAllocator->m_UseMutex); - if (m_Handle) - { - *pHandle = m_Handle.Duplicate(hTargetProcess); - return VK_SUCCESS; - } - - // Do we pass the function manually or use the one from the allocator? - VkResult res = m_Handle.Init(hAllocator->m_hDevice, m_hMemory, &vkGetMemoryWin32HandleKHR); - if (res != VK_SUCCESS) - { - return res; - } - - *pHandle = m_Handle.Duplicate(hTargetProcess); - return VK_SUCCESS; + return m_Handle.GetHandle(hAllocator->m_hDevice, m_hMemory, &vkGetMemoryWin32HandleKHR, hTargetProcess, pHandle); } #endif // VK_USE_PLATFORM_WIN32_KHR #endif // _VMA_DEVICE_MEMORY_BLOCK_FUNCTIONS @@ -11071,14 +11092,16 @@ void VmaAllocation_T::PrintParameters(class VmaJsonWriter& json) const } } #ifdef VK_USE_PLATFORM_WIN32_KHR -inline VkResult VmaAllocation_T::GetWin32Handle(VmaAllocator hAllocator, HANDLE hTargetProcess, HANDLE* pHandle) const noexcept +VkResult VmaAllocation_T::GetWin32Handle(VmaAllocator hAllocator, HANDLE hTargetProcess, HANDLE* pHandle) const noexcept { + // Where do we get this function from? + auto pvkGetMemoryWin32HandleKHR = &vkGetMemoryWin32HandleKHR; switch (m_Type) { case ALLOCATION_TYPE_BLOCK: - return m_BlockAllocation.m_Block->CreateWin32Handle(hAllocator, hTargetProcess, pHandle); + return m_BlockAllocation.m_Block->CreateWin32Handle(hAllocator, pvkGetMemoryWin32HandleKHR, hTargetProcess, pHandle); case ALLOCATION_TYPE_DEDICATED: - // return m_DedicatedAllocation.m_hMemory; + return m_DedicatedAllocation.m_Handle.GetHandle(hAllocator->m_hDevice, m_DedicatedAllocation.m_hMemory, pvkGetMemoryWin32HandleKHR, hTargetProcess, pHandle); default: VMA_ASSERT(0); return VK_ERROR_FEATURE_NOT_PRESENT;