NRI v1.87:

- API: exposed AccelerationStructure in wrapper interfaces
- VK: fixed internal type of VkFence
- VK: updated VK SDK headers
- D3D12: fixed incorrect cast (major bug)
- D3D12: fixed potential "double free"

.gitmodules

@@ -1,5 +1,5 @@
 [submodule "External/vulkan"]
 	path = External/vulkan
 	url = https://github.com/KhronosGroup/Vulkan-Headers.git
-	branch = sdk-1.3.216
+	branch = sdk-1.3.231.0
 	update = merge

Include/Extensions/NRIRayTracing.h

@@ -192,5 +192,7 @@ namespace nri
         void (NRI_CALL *CmdWriteAccelerationStructureSize)(CommandBuffer& commandBuffer, const AccelerationStructure* const* accelerationStructures, uint32_t accelerationStructureNum, QueryPool& queryPool, uint32_t queryPoolOffset);
 
         void (NRI_CALL *CmdDispatchRays)(CommandBuffer& commandBuffer, const DispatchRaysDesc& dispatchRaysDesc);
+
+        uint64_t (NRI_CALL* GetAccelerationStructureNativeObject)(const AccelerationStructure& accelerationStructure, uint32_t physicalDeviceIndex); // ID3D12Resource* or VkAccelerationStructureKHR
     };
 }

Include/Extensions/NRIWrapperD3D12.h

@@ -22,6 +22,8 @@ struct IDXGIAdapter;
 
 namespace nri
 {
+    struct AccelerationStructure;
+
     struct DeviceCreationD3D12Desc
     {
         ID3D12Device* d3d12Device;
@@ -57,6 +59,13 @@ namespace nri
         ID3D12Heap* d3d12Heap;
     };
 
+    struct AccelerationStructureD3D12Desc
+    {
+        ID3D12Resource* d3d12Resource;
+        uint64_t scratchDataSizeInBytes;
+        uint64_t updateScratchDataSizeInBytes;
+    };
+
     NRI_API Result NRI_CALL CreateDeviceFromD3D12Device(const DeviceCreationD3D12Desc& deviceDesc, Device*& device);
     NRI_API Format NRI_CALL ConvertDXGIFormatToNRI(uint32_t dxgiFormat);
     NRI_API uint32_t NRI_CALL ConvertNRIFormatToDXGI(Format format);
@@ -67,6 +76,7 @@ namespace nri
         Result (NRI_CALL *CreateBufferD3D12)(Device& device, const BufferD3D12Desc& bufferDesc, Buffer*& buffer);
         Result (NRI_CALL *CreateTextureD3D12)(Device& device, const TextureD3D12Desc& textureDesc, Texture*& texture);
         Result (NRI_CALL *CreateMemoryD3D12)(Device& device, const MemoryD3D12Desc& memoryDesc, Memory*& memory);
+        Result (NRI_CALL *CreateAccelerationStructureD3D12)(Device& device, const AccelerationStructureD3D12Desc& accelerationStructureDesc, AccelerationStructure*& accelerationStructure);
     };
 
 }

Include/Extensions/NRIWrapperVK.h

@@ -16,6 +16,8 @@ struct VkImageSubresourceRange;
 
 namespace nri
 {
+    struct AccelerationStructure;
+
     typedef uint64_t NRIVkCommandPool;
     typedef uint64_t NRIVkImage;
     typedef uint64_t NRIVkBuffer;
@@ -27,6 +29,7 @@ namespace nri
     typedef uint64_t NRIVkFence;
     typedef uint64_t NRIVkImageView;
     typedef uint64_t NRIVkBufferView;
+    typedef uint64_t NRIVkAccelerationStructureKHR;
 
     typedef void* NRIVkInstance;
     typedef void* NRIVkPhysicalDevice;
@@ -111,6 +114,14 @@ namespace nri
         uint32_t physicalDeviceMask;
     };
 
+    struct AccelerationStructureVulkanDesc
+    {
+        NRIVkAccelerationStructureKHR vkAccelerationStructure;
+        uint64_t buildScratchSize;
+        uint64_t updateScratchSize;
+        uint32_t physicalDeviceMask;
+    };
+
     NRI_API Result NRI_CALL CreateDeviceFromVkDevice(const DeviceCreationVulkanDesc& deviceDesc, Device*& device);
     NRI_API Format NRI_CALL ConvertVKFormatToNRI(uint32_t vkFormat);
     NRI_API uint32_t NRI_CALL ConvertNRIFormatToVK(Format format);
@@ -129,6 +140,7 @@ namespace nri
         Result (NRI_CALL *CreateQueryPoolVK)(Device& device, const QueryPoolVulkanDesc& queryPoolVulkanDesc, QueryPool*& queryPool);
         Result (NRI_CALL *CreateQueueSemaphoreVK)(Device& device, NRIVkSemaphore vkSemaphore, QueueSemaphore*& queueSemaphore);
         Result (NRI_CALL *CreateDeviceSemaphoreVK)(Device& device, NRIVkFence vkFence, DeviceSemaphore*& deviceSemaphore);
+        Result (NRI_CALL *CreateAccelerationStructureVK)(Device& device, const AccelerationStructureVulkanDesc& accelerationStructureDesc, AccelerationStructure*& accelerationStructure);
 
         NRIVkPhysicalDevice (NRI_CALL *GetVkPhysicalDevice)(const Device& device);
         NRIVkInstance (NRI_CALL *GetVkInstance)(const Device& device);

Include/NRI.h

@@ -16,8 +16,8 @@ license agreement from NVIDIA CORPORATION is strictly prohibited.
 #include <cstddef>
 
 #define NRI_VERSION_MAJOR 1
-#define NRI_VERSION_MINOR 86
-#define NRI_VERSION_DATE "19 September 2022"
+#define NRI_VERSION_MINOR 87
+#define NRI_VERSION_DATE "5 October 2022"
 #define NRI_INTERFACE( name ) #name, sizeof(name)
 
 #if _WIN32

README.md

@@ -1,4 +1,4 @@
-# NVIDIA Render Interface v1.86 (NRI)
+# NVIDIA Render Interface v1.87 (NRI)
 
 *NRI* is a low-level abstract render interface which currently supports three backends: D3D11, D3D12 and Vulkan. *NRI* has been designed to support all (at least major) low level features of D3D12 and Vulkan APIs, but at the same time to simplify usage and reduce the amount of code needed (especially compared with Vulkan).
 
@@ -10,6 +10,7 @@
  - default D3D11 behavior is changed to match D3D12/VK where applicable using *NVAPI* or *AMD AGS* libraries
 
  *NRI* is used in:
+ - [*NRI samples*](https://github.com/NVIDIAGameWorks/NRISamples)
  - [*NRD Sample*](https://github.com/NVIDIAGameWorks/NRDSample)
 
 ## Build instructions

Resources/Version.h

@@ -21,7 +21,7 @@ Versioning rules:
 */
 
 #define VERSION_MAJOR                   1
-#define VERSION_MINOR                   86
+#define VERSION_MINOR                   87
 #define VERSION_BUILD                   0
 #define VERSION_REVISION                0
 

Source/D3D12/AccelerationStructureD3D12.cpp

@@ -30,6 +30,15 @@ AccelerationStructureD3D12::~AccelerationStructureD3D12()
         Deallocate(m_Device.GetStdAllocator(), m_Buffer);
 }
 
+Result AccelerationStructureD3D12::Create(const AccelerationStructureD3D12Desc& accelerationStructureDesc)
+{
+    m_PrebuildInfo.ScratchDataSizeInBytes = accelerationStructureDesc.scratchDataSizeInBytes;
+    m_PrebuildInfo.UpdateScratchDataSizeInBytes = accelerationStructureDesc.updateScratchDataSizeInBytes;
+    BufferD3D12Desc bufferDesc = {};
+    bufferDesc.d3d12Resource = (ID3D12Resource*)accelerationStructureDesc.d3d12Resource;
+    return m_Device.CreateBuffer(bufferDesc, (Buffer*&)m_Buffer);
+}
+
 Result AccelerationStructureD3D12::Create(const AccelerationStructureDesc& accelerationStructureDesc)
 {
     D3D12_BUILD_RAYTRACING_ACCELERATION_STRUCTURE_INPUTS accelerationStructureInputs = {};

Source/D3D12/AccelerationStructureD3D12.h

@@ -22,8 +22,11 @@ namespace nri
         ~AccelerationStructureD3D12();
 
         DeviceD3D12& GetDevice() const;
+        operator ID3D12Resource*() const;
+        operator BufferD3D12* () const;
 
         Result Create(const AccelerationStructureDesc& accelerationStructureDesc);
+        Result Create(const AccelerationStructureD3D12Desc& accelerationStructureDesc);
         void GetMemoryInfo(MemoryDesc& memoryDesc) const;
         uint64_t GetUpdateScratchBufferSize() const;
         uint64_t GetBuildScratchBufferSize() const;
@@ -50,5 +53,15 @@ namespace nri
     {
         return m_Device;
     }
+
+    inline AccelerationStructureD3D12::operator ID3D12Resource*() const
+    {
+        return (ID3D12Resource*)m_Buffer;
+    }
+
+    inline AccelerationStructureD3D12::operator BufferD3D12* () const
+    {
+        return m_Buffer;
+    }
 }
 #endif

Source/D3D12/AccelerationStructureD3D12.hpp

@@ -44,6 +44,13 @@ static void NRI_CALL SetAccelerationStructureDebugName(AccelerationStructure& ac
     ((AccelerationStructureD3D12&)accelerationStructure).SetDebugName(name);
 }
 
+static uint64_t NRI_CALL GetAccelerationStructureNativeObject(const AccelerationStructure& accelerationStructure, uint32_t physicalDeviceIndex)
+{
+    MaybeUnused(physicalDeviceIndex); // TODO: use it
+
+    return uint64_t((ID3D12Resource*)((AccelerationStructureD3D12&)accelerationStructure));
+}
+
 void FillFunctionTableAccelerationStructureD3D12(RayTracingInterface& rayTracingInterface)
 {
     rayTracingInterface.CreateAccelerationStructureDescriptor = ::CreateAccelerationStructureDescriptor;
@@ -52,6 +59,7 @@ void FillFunctionTableAccelerationStructureD3D12(RayTracingInterface& rayTracing
     rayTracingInterface.GetAccelerationStructureBuildScratchBufferSize = ::GetAccelerationStructureBuildScratchBufferSize;
     rayTracingInterface.GetAccelerationStructureHandle = ::GetAccelerationStructureHandle;
     rayTracingInterface.SetAccelerationStructureDebugName = ::SetAccelerationStructureDebugName;
+    rayTracingInterface.GetAccelerationStructureNativeObject = ::GetAccelerationStructureNativeObject;
 }
 
 #pragma endregion

Source/D3D12/CommandBufferD3D12.cpp

@@ -517,8 +517,8 @@ inline void CommandBufferD3D12::PipelineBarrier(const TransitionBarrierDesc* tra
         {
             D3D12_RESOURCE_BARRIER& barrier = *ptr++;
             barrier.Type = D3D12_RESOURCE_BARRIER_TYPE_ALIASING;
-            barrier.Aliasing.pResourceBefore = *((BufferD3D12*)aliasingBarriers->textures[i].before);
-            barrier.Aliasing.pResourceAfter = *((BufferD3D12*)aliasingBarriers->textures[i].after);
+            barrier.Aliasing.pResourceBefore = *((TextureD3D12*)aliasingBarriers->textures[i].before);
+            barrier.Aliasing.pResourceAfter = *((TextureD3D12*)aliasingBarriers->textures[i].after);
         }
     }
 

Source/D3D12/DescriptorPoolD3D12.cpp

@@ -24,8 +24,9 @@ DescriptorPoolD3D12::DescriptorPoolD3D12(DeviceD3D12& device)
 
 DescriptorPoolD3D12::~DescriptorPoolD3D12()
 {
-    for (size_t i = 0; i < m_DescriptorSetNum; i++)
-        Deallocate(m_Device.GetStdAllocator(), m_DescriptorSets[i]);
+    Reset();
+
+    m_DescriptorSets.clear();
 }
 
 Result DescriptorPoolD3D12::Create(const DescriptorPoolDesc& descriptorPoolDesc)

Source/D3D12/DescriptorPoolD3D12.h

@@ -51,8 +51,8 @@ namespace nri
         std::array<DescriptorHeapDesc, DescriptorHeapType::MAX_NUM> m_DescriptorHeapDescs;
         std::array<uint32_t, DescriptorHeapType::MAX_NUM> m_DescriptorNum = {};
         std::array<ID3D12DescriptorHeap*, DescriptorHeapType::MAX_NUM> m_DescriptorHeaps = {};
-        uint32_t m_DescriptorHeapNum = 0;
         Vector<DescriptorSetD3D12*> m_DescriptorSets;
+        uint32_t m_DescriptorHeapNum = 0;
         uint32_t m_DescriptorSetNum = 0;
     };
 

Source/D3D12/DeviceD3D12.cpp

@@ -270,6 +270,11 @@ Result DeviceD3D12::CreateAccelerationStructure(const AccelerationStructureDesc&
     return CreateImplementation<AccelerationStructureD3D12>(accelerationStructure, accelerationStructureDesc);
 }
 
+Result DeviceD3D12::CreateAccelerationStructure(const AccelerationStructureD3D12Desc& accelerationStructureDesc, AccelerationStructure*& accelerationStructure)
+{
+    return CreateImplementation<AccelerationStructureD3D12>(accelerationStructure, accelerationStructureDesc);
+}
+
 inline void DeviceD3D12::DestroyAccelerationStructure(AccelerationStructure& accelerationStructure)
 {
     Deallocate(GetStdAllocator(), (AccelerationStructureD3D12*)&accelerationStructure);

Source/D3D12/DeviceD3D12.h

@@ -1,4 +1,4 @@
-/*
+/*
 Copyright (c) 2022, NVIDIA CORPORATION. All rights reserved.
 
 NVIDIA CORPORATION and its licensors retain all intellectual property
@@ -107,9 +107,10 @@ namespace nri
         Result CreateDeviceSemaphore(bool signaled, DeviceSemaphore*& deviceSemaphore);
 
         Result CreateCommandBuffer(const CommandBufferD3D12Desc& commandBufferDesc, CommandBuffer*& commandBuffer);
-        Result CreateBuffer(const BufferD3D12Desc& commandBufferDesc, Buffer*& commandBuffer);
-        Result CreateTexture(const TextureD3D12Desc& commandBufferDesc, Texture*& commandBuffer);
+        Result CreateBuffer(const BufferD3D12Desc& bufferDesc, Buffer*& buffer);
+        Result CreateTexture(const TextureD3D12Desc& textureDesc, Texture*& texture);
         Result CreateMemory(const MemoryD3D12Desc& memoryDesc, Memory*& memory);
+        Result CreateAccelerationStructure(const AccelerationStructureD3D12Desc& accelerationStructureDesc, AccelerationStructure*& accelerationStructure);
 
         void DestroyCommandAllocator(CommandAllocator& commandAllocator);
         void DestroyDescriptorPool(DescriptorPool& descriptorPool);

Source/D3D12/DeviceD3D12.hpp

@@ -389,13 +389,19 @@ static Result NRI_CALL CreateMemoryD3D12(Device& device, const MemoryD3D12Desc&
     return ((DeviceD3D12&)device).CreateMemory(memoryDesc, memory);
 }
 
+static Result NRI_CALL CreateAccelerationStructureD3D12(Device& device, const AccelerationStructureD3D12Desc& accelerationStructureDesc, AccelerationStructure*& memory)
+{
+    return ((DeviceD3D12&)device).CreateAccelerationStructure(accelerationStructureDesc, memory);
+}
+
 Result DeviceD3D12::FillFunctionTable(WrapperD3D12Interface& wrapperD3D12Interface) const
 {
     wrapperD3D12Interface = {};
     wrapperD3D12Interface.CreateCommandBufferD3D12 = ::CreateCommandBufferD3D12;
     wrapperD3D12Interface.CreateBufferD3D12 = ::CreateBufferD3D12;
     wrapperD3D12Interface.CreateTextureD3D12 = ::CreateTextureD3D12;
     wrapperD3D12Interface.CreateMemoryD3D12 = ::CreateMemoryD3D12;
+    wrapperD3D12Interface.CreateAccelerationStructureD3D12 = ::CreateAccelerationStructureD3D12;
 
     return ValidateFunctionTable(GetLog(), wrapperD3D12Interface);
 }

Source/VK/AccelerationStructureVK.cpp

@@ -22,6 +22,9 @@ AccelerationStructureVK::~AccelerationStructureVK()
 {
     const auto& vk = m_Device.GetDispatchTable();
 
+    if (!m_OwnsNativeObjects)
+        return;
+
     for (uint32_t i = 0; i < GetCountOf(m_Handles); i++)
     {
         if (m_Handles[i] != VK_NULL_HANDLE)
@@ -34,6 +37,7 @@ AccelerationStructureVK::~AccelerationStructureVK()
 
 Result AccelerationStructureVK::Create(const AccelerationStructureDesc& accelerationStructureDesc)
 {
+    m_OwnsNativeObjects = true;
     m_Type = GetAccelerationStructureType(accelerationStructureDesc.type);
     m_BuildFlags = GetAccelerationStructureBuildFlags(accelerationStructureDesc.flags);
 
@@ -58,6 +62,40 @@ Result AccelerationStructureVK::Create(const AccelerationStructureDesc& accelera
     return result;
 }
 
+Result AccelerationStructureVK::Create(const AccelerationStructureVulkanDesc& accelerationStructureDesc)
+{
+    m_OwnsNativeObjects = false;
+    m_Type = VK_ACCELERATION_STRUCTURE_TYPE_MAX_ENUM_KHR;
+    m_BuildFlags = 0;
+
+    uint32_t physicalDeviceMask = GetPhysicalDeviceGroupMask(accelerationStructureDesc.physicalDeviceMask);
+
+    VkAccelerationStructureDeviceAddressInfoKHR deviceAddressInfo = {};
+    deviceAddressInfo.sType = VK_STRUCTURE_TYPE_ACCELERATION_STRUCTURE_DEVICE_ADDRESS_INFO_KHR;
+    deviceAddressInfo.accelerationStructure = (VkAccelerationStructureKHR)accelerationStructureDesc.vkAccelerationStructure;
+
+    const auto& vk = m_Device.GetDispatchTable();
+
+    const uint64_t deviceAddress = vk.GetAccelerationStructureDeviceAddressKHR(m_Device, &deviceAddressInfo);
+
+    if (deviceAddress == 0)
+        return Result::FAILURE;
+
+    for (uint32_t i = 0; i < m_Device.GetPhyiscalDeviceGroupSize(); i++)
+    {
+        if ((1 << i) & physicalDeviceMask)
+        {
+            m_Handles[i] = (VkAccelerationStructureKHR)accelerationStructureDesc.vkAccelerationStructure;
+            m_DeviceAddresses[i] = deviceAddress;
+        }
+    }
+
+    m_BuildScratchSize = accelerationStructureDesc.buildScratchSize;
+    m_UpdateScratchSize = accelerationStructureDesc.updateScratchSize;
+
+    return Result::SUCCESS;
+}
+
 void AccelerationStructureVK::PrecreateBottomLevel(const AccelerationStructureDesc& accelerationStructureDesc)
 {
     const auto buildType = VK_ACCELERATION_STRUCTURE_BUILD_TYPE_DEVICE_KHR;

Source/VK/AccelerationStructureVK.h

@@ -21,6 +21,7 @@ namespace nri
 
         DeviceVK& GetDevice() const;
         Result Create(const AccelerationStructureDesc& accelerationStructureDesc);
+        Result Create(const AccelerationStructureVulkanDesc& accelerationStructureDesc);
         Result FinishCreation();
         VkAccelerationStructureKHR GetHandle(uint32_t physicalDeviceIndex) const;
         BufferVK* GetBuffer() const;
@@ -46,6 +47,7 @@ namespace nri
         uint32_t m_PhysicalDeviceMask = 0;
         VkAccelerationStructureTypeKHR m_Type = (VkAccelerationStructureTypeKHR)0;
         VkBuildAccelerationStructureFlagsKHR m_BuildFlags = (VkBuildAccelerationStructureFlagsKHR)0;
+        bool m_OwnsNativeObjects = false;
     };
 
     inline AccelerationStructureVK::AccelerationStructureVK(DeviceVK& device) :

Source/VK/AcceleratrionStructureVK.hpp

@@ -43,6 +43,11 @@ static Result NRI_CALL CreateAccelerationStructureDescriptor(const AccelerationS
     return ((AccelerationStructureVK&)accelerationStructure).CreateDescriptor(physicalDeviceIndex, descriptor);
 }
 
+static uint64_t NRI_CALL GetAccelerationStructureNativeObject(const AccelerationStructure& accelerationStructure, uint32_t physicalDeviceIndex)
+{
+    return uint64_t(((AccelerationStructureVK&)accelerationStructure).GetHandle(physicalDeviceIndex));
+}
+
 void FillFunctionTableAccelerationStructureVK(RayTracingInterface& rayTracingInterface)
 {
     rayTracingInterface.GetAccelerationStructureMemoryInfo = ::GetAccelerationStructureMemoryInfo;
@@ -51,6 +56,7 @@ void FillFunctionTableAccelerationStructureVK(RayTracingInterface& rayTracingInt
     rayTracingInterface.GetAccelerationStructureHandle = ::GetAccelerationStructureHandle;
     rayTracingInterface.SetAccelerationStructureDebugName = ::SetAccelerationStructureDebugName;
     rayTracingInterface.CreateAccelerationStructureDescriptor = ::CreateAccelerationStructureDescriptor;
+    rayTracingInterface.GetAccelerationStructureNativeObject = ::GetAccelerationStructureNativeObject;
 }
 
 #pragma endregion

Source/VK/DeviceSemaphoreVK.cpp

@@ -40,7 +40,7 @@ Result DeviceSemaphoreVK::Create(bool signaled)
     return Result::SUCCESS;
 }
 
-Result DeviceSemaphoreVK::Create(void* vkFence)
+Result DeviceSemaphoreVK::Create(NRIVkFence vkFence)
 {
     m_OwnsNativeObjects = false;
     m_Handle = (VkFence)vkFence;

Source/VK/DeviceSemaphoreVK.h

@@ -22,7 +22,7 @@ namespace nri
         operator VkFence() const;
         DeviceVK& GetDevice() const;
         Result Create(bool signaled);
-        Result Create(void* vkFence);
+        Result Create(NRIVkFence vkFence);
 
         void SetDebugName(const char* name);