v4.3.5:

HIGHLIGHTS: - NRD INTEGRATION: various improvements - CMake: infrastructure improvements DETAILS: - NRD INTEGRATION: fixed a major bug: leaking descriptors with disabled descriptor caching - NRD INTEGRATION: optimized allocation of descriptor pools - NRD INTEGRATION: optimized suballocation in descriptor pools if samplers are in a dedicated register space - NRD INTEGRATION: more and better asserts - NRD INTEGRATION: "frameIndex" is hidden inside "NrdIntegration" instance - NRD: optimized "DescriptorPoolDesc" if samplers are in a dedicated register space - CMake: infrastructure improvements - CMake: "_Build" => "_Bin", "_Compiler" => "_Build" - updated dependencies
NVIDIAGameWorks · Oct 26, 2023 · fdb08f1 · fdb08f1
1 parent 2d94c1f
commit fdb08f1
Show file tree

Hide file tree

Showing 17 changed files with 165 additions and 103 deletions.
diff --git a/.gitignore b/.gitignore
@@ -45,6 +45,6 @@ compile_commands.json
 *.manifest
 
 # generated folders
+_Bin/
 _Build/
-_Compiler/
 _NRD_SDK/
diff --git a/1-Deploy.bat b/1-Deploy.bat
@@ -2,8 +2,8 @@
 
 git submodule update --init --recursive
 
-mkdir "_Compiler"
+mkdir "_Build"
 
-cd "_Compiler"
+cd "_Build"
 cmake .. -A x64
 cd ..
diff --git a/1-Deploy.sh b/1-Deploy.sh
@@ -6,8 +6,8 @@ chmod +x "4-Clean.sh"
 
 git submodule update --init --recursive
 
-mkdir -p "_Compiler"
+mkdir -p "_Build"
 
-cd "_Compiler"
+cd "_Build"
 cmake ..
 cd ..
diff --git a/2-Build.bat b/2-Build.bat
@@ -1,6 +1,6 @@
 @echo off
 
-cd "_Compiler"
+cd "_Build"
 cmake --build . --config Release -j 4
 cmake --build . --config Debug -j 4
 cd ..
diff --git a/2-Build.sh b/2-Build.sh
@@ -1,8 +1,8 @@
 #!/bin/sh
 
-mkdir -p "_Compiler"
+mkdir -p "_Build"
 
-cd "_Compiler"
+cd "_Build"
 cmake ..
 cmake --build . --config Release -j 4
 cmake --build . --config Debug -j 4

diff --git a/3-Prepare NRD SDK.bat b/3-Prepare NRD SDK.bat
@@ -34,12 +34,12 @@ mkdir "_NRD_SDK\Shaders\Include"
 cd "_NRD_SDK"
 
 copy "..\%NRD_DIR%\Include\*" "Include"
-copy "..\_Build\Debug\NRD.dll" "Lib\Debug"
-copy "..\_Build\Debug\NRD.lib" "Lib\Debug"
-copy "..\_Build\Debug\NRD.pdb" "Lib\Debug"
-copy "..\_Build\Release\NRD.dll" "Lib\Release"
-copy "..\_Build\Release\NRD.lib" "Lib\Release"
-copy "..\_Build\Release\NRD.pdb" "Lib\Release"
+copy "..\_Bin\Debug\NRD.dll" "Lib\Debug"
+copy "..\_Bin\Debug\NRD.lib" "Lib\Debug"
+copy "..\_Bin\Debug\NRD.pdb" "Lib\Debug"
+copy "..\_Bin\Release\NRD.dll" "Lib\Release"
+copy "..\_Bin\Release\NRD.lib" "Lib\Release"
+copy "..\_Bin\Release\NRD.pdb" "Lib\Release"
 copy "..\%NRD_DIR%\Shaders\Include\NRD.hlsli" "Shaders\Include"
 copy "..\%NRD_DIR%\Shaders\Include\NRDEncoding.hlsli" "Shaders\Include"
 copy "..\%NRD_DIR%\LICENSE.txt" "."

diff --git a/3-Prepare NRD SDK.sh b/3-Prepare NRD SDK.sh
@@ -13,8 +13,8 @@ mkdir -p "_NRD_SDK/Shaders/Include"
 cd "_NRD_SDK"
 
 cp -r ../$NRD_DIR/Include/ "Include"
-cp -H ../_Build/Debug/libNRD.so "Lib/Debug"
-cp -H ../_Build/Release/libNRD.so "Lib/Release"
+cp -H ../_Bin/Debug/libNRD.so "Lib/Debug"
+cp -H ../_Bin/Release/libNRD.so "Lib/Release"
 cp ../$NRD_DIR/Shaders/Include/NRD.hlsli "Shaders/Include"
 cp ../$NRD_DIR/Shaders/Include/NRDEncoding.hlsli "Shaders/Include"
 cp ../$NRD_DIR/LICENSE.txt "."

diff --git a/4-Clean.bat b/4-Clean.bat
@@ -2,7 +2,7 @@
 
 if exist "build" rd /q /s "build"
 
+if exist "_Bin" rd /q /s "_Bin"
 if exist "_Build" rd /q /s "_Build"
-if exist "_Compiler" rd /q /s "_Compiler"
 if exist "_Shaders" rd /q /s "_Shaders"
 if exist "_NRD_SDK" rd /q /s "_NRD_SDK"
diff --git a/4-Clean.sh b/4-Clean.sh
@@ -2,7 +2,7 @@
 
 rm -rf "build"
 
+rm -rf "_Bin"
 rm -rf "_Build"
-rm -rf "_Compiler"
 rm -rf "_Shaders"
 rm -rf "_NRD_SDK"
diff --git a/CMakeLists.txt b/CMakeLists.txt
@@ -23,17 +23,17 @@ endif ()
 # Cached
 if (NOT IS_SUBMODULE)
     set (CMAKE_CONFIGURATION_TYPES "Debug;Release" CACHE STRING "")
+    set (CMAKE_RUNTIME_OUTPUT_DIRECTORY "${CMAKE_SOURCE_DIR}/_Bin" CACHE STRING "")
 endif ()
 
-# Generate PDB for Release builds
-set (CMAKE_CXX_FLAGS_RELEASE "${CMAKE_CXX_FLAGS_RELEASE} /Zi")
-set (CMAKE_SHARED_LINKER_FLAGS_RELEASE "${CMAKE_SHARED_LINKER_FLAGS_RELEASE} /DEBUG /OPT:REF /OPT:ICF")
-
 set (NRD_DXC_CUSTOM_PATH "custom/path/to/dxc" CACHE STRING "Custom DXC to use if Vulkan SDK is not installed")
 set (NRD_SHADERS_PATH "" CACHE STRING "Shader output path override")
 set (NRD_NORMAL_ENCODING "2" CACHE STRING "Normal encoding variant (0-4, matches nrd::NormalEncoding)")
 set (NRD_ROUGHNESS_ENCODING "1" CACHE STRING "Roughness encoding variant (0-2, matches nrd::RoughnessEncoding)")
-set (GLOBAL_BIN_OUTPUT_PATH "${CMAKE_SOURCE_DIR}/_Build" CACHE STRING "")
+
+# Generate PDB for Release builds
+set (CMAKE_CXX_FLAGS_RELEASE "${CMAKE_CXX_FLAGS_RELEASE} /Zi")
+set (CMAKE_SHARED_LINKER_FLAGS_RELEASE "${CMAKE_SHARED_LINKER_FLAGS_RELEASE} /DEBUG /OPT:REF /OPT:ICF")
 
 # Create project
 file (READ "Include/NRD.h" ver_h)
@@ -149,8 +149,8 @@ target_compile_options (${PROJECT_NAME} PRIVATE ${COMPILE_OPTIONS})
 
 set_property (TARGET ${PROJECT_NAME} PROPERTY FOLDER "${PROJECT_NAME}")
 
-set_target_properties (${PROJECT_NAME} PROPERTIES RUNTIME_OUTPUT_DIRECTORY "${GLOBAL_BIN_OUTPUT_PATH}/$<CONFIG>")
-message ("NRD output path: '${GLOBAL_BIN_OUTPUT_PATH}'")
+set_target_properties (${PROJECT_NAME} PROPERTIES ARCHIVE_OUTPUT_DIRECTORY "${CMAKE_RUNTIME_OUTPUT_DIRECTORY}")
+message ("NRD output path: '${CMAKE_RUNTIME_OUTPUT_DIRECTORY}'")
 
 # Shaders
 if (NOT NRD_DISABLE_SHADER_COMPILATION)
@@ -161,7 +161,7 @@ if (NOT NRD_DISABLE_SHADER_COMPILATION)
 
     # External/ShaderMake
     if (NOT TARGET ShaderMake)
-        set (SHADERMAKE_BIN_OUTPUT_PATH ${GLOBAL_BIN_OUTPUT_PATH} CACHE STRING "")
+        set (SHADERMAKE_BIN_OUTPUT_PATH ${CMAKE_RUNTIME_OUTPUT_DIRECTORY} CACHE STRING "")
         add_subdirectory (External/ShaderMake)
     endif ()
 

diff --git a/External/ShaderMake b/External/ShaderMake
diff --git a/Include/NRD.h b/Include/NRD.h
@@ -29,8 +29,8 @@ license agreement from NVIDIA CORPORATION is strictly prohibited.
 
 #define NRD_VERSION_MAJOR 4
 #define NRD_VERSION_MINOR 3
-#define NRD_VERSION_BUILD 4
-#define NRD_VERSION_DATE "20 October 2023"
+#define NRD_VERSION_BUILD 5
+#define NRD_VERSION_DATE "25 October 2023"
 
 #if defined(_MSC_VER)
     #define NRD_CALL __fastcall

diff --git a/Integration/NRDIntegration.h b/Integration/NRDIntegration.h
@@ -23,15 +23,15 @@ license agreement from NVIDIA CORPORATION is strictly prohibited.
 #include <map>
 
 #define NRD_INTEGRATION_MAJOR 1
-#define NRD_INTEGRATION_MINOR 6
-#define NRD_INTEGRATION_DATE "30 April 2023"
+#define NRD_INTEGRATION_MINOR 7
+#define NRD_INTEGRATION_DATE "25 October 2023"
 #define NRD_INTEGRATION 1
 
 #define NRD_INTEGRATION_DEBUG_LOGGING 0
 
 #ifndef NRD_INTEGRATION_ASSERT
     #include <assert.h>
-    #define NRD_INTEGRATION_ASSERT(expr, msg) (assert(expr && msg))
+    #define NRD_INTEGRATION_ASSERT(expr, msg) assert(expr && msg)
 #endif
 
 // User inputs / outputs are not mipmapped, thus only 1 entry is needed.
@@ -58,12 +58,16 @@ inline void NrdIntegration_SetResource(NrdUserPool& pool, nrd::ResourceType slot
 class NrdIntegration
 {
 public:
-    // The application must provide number of buffered frames, it's needed to guarantee that
-    // constant data and descriptor sets are not overwritten while being executed on the GPU.
-    // Usually it's 2-3 frames.
-    NrdIntegration(uint32_t bufferedFramesNum, const char* persistentName = "") :
+    // bufferedFramesNum (usually 2-3 frames):
+    //      The application must provide number of buffered frames, it's needed to guarantee that
+    //      constant data and descriptor sets are not overwritten while being executed on the GPU.
+    // enableDescriptorCaching:
+    //      true - enables descriptor caching for the whole lifetime of an NrdIntegration instance
+    //      false - descriptors are cached only within a single "Denoise" call
+    NrdIntegration(uint32_t bufferedFramesNum, bool enableDescriptorCaching, const char* persistentName = "") :
         m_Name(persistentName)
         , m_BufferedFramesNum(bufferedFramesNum)
+        , m_IsDescriptorCachingEnabled(enableDescriptorCaching)
     {}
 
     ~NrdIntegration()
@@ -72,22 +76,16 @@ class NrdIntegration
     // There is no "Resize" functionality, because NRD full recreation costs nothing.
     // The main cost comes from render targets resizing, which needs to be done in any case
     // (call Destroy beforehand)
-    bool Initialize(const nrd::InstanceCreationDesc& instanceCreationDesc, nri::Device& nriDevice,
-        const nri::CoreInterface& nriCore, const nri::HelperInterface& nriHelper);
+    bool Initialize(const nrd::InstanceCreationDesc& instanceCreationDesc, nri::Device& nriDevice, const nri::CoreInterface& nriCore, const nri::HelperInterface& nriHelper);
 
     // Must be called once on a frame start
-    void NewFrame(uint32_t frameIndex);
+    void NewFrame();
 
-    // Explcitly calls eponymous NRD API functions
+    // Explicitly calls eponymous NRD API functions
     bool SetCommonSettings(const nrd::CommonSettings& commonSettings);
     bool SetDenoiserSettings(nrd::Identifier denoiser, const void* denoiserSettings);
 
-    // Better use "enableDescriptorCaching = true" if resources are not changing between frames
-    // (i.e. not suballocated from a heap)
-    void Denoise(const nrd::Identifier* denoisers, uint32_t denoisersNum,
-        nri::CommandBuffer& commandBuffer, const NrdUserPool& userPool,
-        bool enableDescriptorCaching
-    );
+    void Denoise(const nrd::Identifier* denoisers, uint32_t denoisersNum, nri::CommandBuffer& commandBuffer, const NrdUserPool& userPool);
 
     // This function assumes that the device is in the IDLE state, i.e. there is no work in flight
     void Destroy();
@@ -110,22 +108,19 @@ class NrdIntegration
 
     void CreateResources();
     void AllocateAndBindMemory();
-    void Dispatch
-    (
-        nri::CommandBuffer& commandBuffer, nri::DescriptorPool& descriptorPool,
-        const nrd::DispatchDesc& dispatchDesc, const NrdUserPool& userPool,
-        bool enableDescriptorCaching
-    );
+    void Dispatch(nri::CommandBuffer& commandBuffer, nri::DescriptorPool& descriptorPool, const nrd::DispatchDesc& dispatchDesc, const NrdUserPool& userPool);
 
 private:
     std::vector<NrdIntegrationTexture> m_TexturePool;
-    std::map<uint64_t, nri::Descriptor*> m_Descriptors;
+    std::map<uint64_t, nri::Descriptor*> m_CachedDescriptors;
+    std::vector<std::vector<nri::Descriptor*>> m_DescriptorsInFlight;
     std::vector<nri::TextureTransitionBarrierDesc> m_ResourceState;
     std::vector<nri::PipelineLayout*> m_PipelineLayouts;
     std::vector<nri::Pipeline*> m_Pipelines;
     std::vector<nri::Memory*> m_MemoryAllocations;
     std::vector<nri::Descriptor*> m_Samplers;
-    std::array<nri::DescriptorPool*, 16> m_DescriptorPools = {};
+    std::vector<nri::DescriptorPool*> m_DescriptorPools = {};
+    std::vector<nri::DescriptorSet*> m_DescriptorSetSamplers = {};
     const nri::CoreInterface* m_NRI = nullptr;
     const nri::HelperInterface* m_NRIHelper = nullptr;
     nri::Device* m_Device = nullptr;
@@ -140,7 +135,9 @@ class NrdIntegration
     uint32_t m_ConstantBufferOffset = 0;
     uint32_t m_BufferedFramesNum = 0;
     uint32_t m_DescriptorPoolIndex = 0;
+    uint32_t m_FrameIndex = 0;
     bool m_IsShadersReloadRequested = false;
+    bool m_IsDescriptorCachingEnabled = false;
 };
 
 #define NRD_INTEGRATION_ABORT_ON_FAILURE(result) if ((result) != nri::Result::SUCCESS) NRD_INTEGRATION_ASSERT(false, "Abort on failure!")