nvapi support (#76)

* add nvapi support * rename SLANG_RHI_HAS_OPTIX -> SLANG_RHI_ENABLE_OPTIX
shader-slang · Oct 11, 2024 · b58c00d · b58c00d
1 parent 4f5ef3c
commit b58c00d
Show file tree

Hide file tree

Showing 18 changed files with 62 additions and 50 deletions.
diff --git a/CMakeLists.txt b/CMakeLists.txt
@@ -94,22 +94,6 @@ if(SLANG_RHI_MASTER_PROJECT)
     FetchPackage(slang URL ${SLANG_URL})
     set(SLANG_RHI_SLANG_INCLUDE_DIR ${slang_SOURCE_DIR}/include)
     set(SLANG_RHI_SLANG_BINARY_DIR ${slang_SOURCE_DIR})
-
-    if(CMAKE_SYSTEM_NAME STREQUAL "Windows")
-        # Fetch nvapi
-        # set(NVAPI_VERSION "4ba3384657149d63aa193f5a34e20efe1e42bf31")
-        # FetchPackage(nvapi URL "https://github.com/NVIDIA/nvapi/archive/${NVAPI_VERSION}.zip")
-        # set(NVAPI_ROOT_DIR ${nvapi_SOURCE_DIR})
-
-        # Fetch dxc
-        # set(DXC_VERSION "1.8.2407")
-        # set(DXC_TAG "2024_07_31")
-        # FetchPackage(dxc URL "https://github.com/microsoft/DirectXShaderCompiler/releases/download/v${DXC_VERSION}/dxc_${DXC_TAG}.zip")
-
-        # Fetch Agility SDK
-        # set(AGILITY_SDK_VERSION "1.611.2")
-        # FetchPackage(agility_sdk URL "https://www.nuget.org/api/v2/package/Microsoft.Direct3D.D3D12/${AGILITY_SDK_VERSION}")
-    endif()
 endif()
 
 set(SLANG_RHI_SLANG_INCLUDE_DIR ${SLANG_RHI_SLANG_INCLUDE_DIR} CACHE STRING "Slang include directory")
@@ -119,6 +103,31 @@ set(SLANG_RHI_SLANG_BINARY_DIR ${SLANG_RHI_SLANG_BINARY_DIR} CACHE STRING "Slang
 unset(SLANG_RHI_SLANG_INCLUDE_DIR)
 unset(SLANG_RHI_SLANG_BINARY_DIR)
 
+# Fetch NVAPI & Agility SDK
+if((SLANG_RHI_ENABLE_D3D11 OR SLANG_RHI_ENABLE_D3D12) AND (CMAKE_SYSTEM_NAME STREQUAL "Windows") AND (CMAKE_SYSTEM_PROCESSOR MATCHES "x86_64|AMD64"))
+    # NVAPI
+    set(NVAPI_VERSION "4ba3384657149d63aa193f5a34e20efe1e42bf31")
+    FetchPackage(nvapi URL "https://github.com/NVIDIA/nvapi/archive/${NVAPI_VERSION}.zip")
+    set(NVAPI_ROOT_DIR ${nvapi_SOURCE_DIR})
+
+    add_library(slang-rhi-nvapi INTERFACE)
+    target_include_directories(slang-rhi-nvapi INTERFACE ${NVAPI_ROOT_DIR})
+    target_link_libraries(slang-rhi-nvapi INTERFACE ${NVAPI_ROOT_DIR}/amd64/nvapi64.lib)
+    target_link_libraries(slang-rhi PUBLIC slang-rhi-nvapi)
+    set(SLANG_RHI_ENABLE_NVAPI ON)
+
+    # Fetch dxc
+    # set(DXC_VERSION "1.8.2407")
+    # set(DXC_TAG "2024_07_31")
+    # FetchPackage(dxc URL "https://github.com/microsoft/DirectXShaderCompiler/releases/download/v${DXC_VERSION}/dxc_${DXC_TAG}.zip")
+
+    # Agility SDK
+    set(AGILITY_SDK_VERSION "1.614.1")
+    FetchPackage(agility_sdk URL "https://www.nuget.org/api/v2/package/Microsoft.Direct3D.D3D12/${AGILITY_SDK_VERSION}")
+else()
+    set(SLANG_RHI_ENABLE_NVAPI OFF)
+endif()
+
 # Fetch and setup Google Dawn library (WebGPU implementation)
 if(SLANG_RHI_ENABLE_WGPU)
     set(DAWN_VERSION "131.0.6738.0")
@@ -186,13 +195,14 @@ if(SLANG_RHI_ENABLE_CUDA)
     # Find Optix
     set(Optix_ROOT_DIR ${CMAKE_CURRENT_SOURCE_DIR}/external/optix CACHE STRING "Optix root directory")
     find_package(OptiX)
-    set(SLANG_RHI_HAS_OPTIX ${OptiX_FOUND})
-    if(${SLANG_RHI_HAS_OPTIX})
+    set(SLANG_RHI_ENABLE_OPTIX ${OptiX_FOUND})
+    if(${SLANG_RHI_ENABLE_OPTIX})
         add_library(slang-rhi-optix INTERFACE)
         target_include_directories(slang-rhi-optix INTERFACE ${OptiX_INCLUDE_DIRS})
         target_link_libraries(slang-rhi PUBLIC slang-rhi-optix)
     endif()
-    target_compile_definitions(slang-rhi PRIVATE SLANG_RHI_HAS_OPTIX=$<BOOL:${SLANG_RHI_HAS_OPTIX}>)
+else()
+    set(SLANG_RHI_ENABLE_OPTIX OFF)
 endif()
 
 # Fetch glfw
@@ -466,6 +476,8 @@ target_compile_definitions(slang-rhi
     SLANG_RHI_ENABLE_METAL=$<BOOL:${SLANG_RHI_ENABLE_METAL}>
     SLANG_RHI_ENABLE_CUDA=$<BOOL:${SLANG_RHI_ENABLE_CUDA}>
     SLANG_RHI_ENABLE_WGPU=$<BOOL:${SLANG_RHI_ENABLE_WGPU}>
+    SLANG_RHI_ENABLE_NVAPI=$<BOOL:${SLANG_RHI_ENABLE_NVAPI}>
+    SLANG_RHI_ENABLE_OPTIX=$<BOOL:${SLANG_RHI_ENABLE_OPTIX}>
     $<$<PLATFORM_ID:Windows>:NOMINMAX> # do not define min/max macros
     $<$<PLATFORM_ID:Windows>:UNICODE> # force character map to unicode
 )

diff --git a/src/cuda/cuda-acceleration-structure.cpp b/src/cuda/cuda-acceleration-structure.cpp
@@ -1,7 +1,7 @@
 #include "cuda-acceleration-structure.h"
 #include "cuda-device.h"
 
-#if SLANG_RHI_HAS_OPTIX
+#if SLANG_RHI_ENABLE_OPTIX
 
 namespace rhi::cuda {
 
@@ -213,4 +213,4 @@ OptixVertexFormat AccelerationStructureBuildInputBuilder::translateVertexFormat(
 
 } // namespace rhi::cuda
 
-#endif // SLANG_RHI_HAS_OPTIX
+#endif // SLANG_RHI_ENABLE_OPTIX
diff --git a/src/cuda/cuda-acceleration-structure.h b/src/cuda/cuda-acceleration-structure.h
@@ -4,7 +4,7 @@
 
 #include "core/stable_vector.h"
 
-#if SLANG_RHI_HAS_OPTIX
+#if SLANG_RHI_ENABLE_OPTIX
 
 namespace rhi::cuda {
 
@@ -49,4 +49,4 @@ struct AccelerationStructureBuildInputBuilder
 
 } // namespace rhi::cuda
 
-#endif // SLANG_RHI_HAS_OPTIX
+#endif // SLANG_RHI_ENABLE_OPTIX
diff --git a/src/cuda/cuda-api.cpp b/src/cuda/cuda-api.cpp
@@ -249,6 +249,6 @@ void rhiCudaApiShutdown() {}
 
 #endif // SLANG_RHI_USE_DYNAMIC_CUDA
 
-#if SLANG_RHI_HAS_OPTIX
+#if SLANG_RHI_ENABLE_OPTIX
 #include <optix_function_table_definition.h>
 #endif
diff --git a/src/cuda/cuda-api.h b/src/cuda/cuda-api.h
@@ -863,7 +863,7 @@ enum CUmemAttach_flags
 
 #endif // SLANG_RHI_USE_DYNAMIC_CUDA
 
-#if SLANG_RHI_HAS_OPTIX
+#if SLANG_RHI_ENABLE_OPTIX
 #define OPTIX_DONT_INCLUDE_CUDA
 #include <optix.h>
 #include <optix_stubs.h>

diff --git a/src/cuda/cuda-command-buffer.cpp b/src/cuda/cuda-command-buffer.cpp
@@ -38,7 +38,7 @@ Result CommandBufferImpl::beginComputePass(IComputePassEncoder** outEncoder)
 
 Result CommandBufferImpl::beginRayTracingPass(IRayTracingPassEncoder** outEncoder)
 {
-#if SLANG_RHI_HAS_OPTIX
+#if SLANG_RHI_ENABLE_OPTIX
     m_rayTracingPassEncoder.init(this);
     *outEncoder = &m_rayTracingPassEncoder;
     return SLANG_OK;

diff --git a/src/cuda/cuda-command-buffer.h b/src/cuda/cuda-command-buffer.h
@@ -16,7 +16,7 @@ class CommandBufferImpl : public ICommandBuffer, public CommandWriter, public Co
     TransientResourceHeap* m_transientHeap;
     ResourcePassEncoderImpl m_resourcePassEncoder;
     ComputePassEncoderImpl m_computePassEncoder;
-#if SLANG_RHI_HAS_OPTIX
+#if SLANG_RHI_ENABLE_OPTIX
     RayTracingPassEncoderImpl m_rayTracingPassEncoder;
 #endif
 

diff --git a/src/cuda/cuda-command-encoder.cpp b/src/cuda/cuda-command-encoder.cpp
@@ -195,7 +195,7 @@ Result ComputePassEncoderImpl::dispatchComputeIndirect(IBuffer* argBuffer, Offse
 
 // RayTracingPassEncoderImpl
 
-#if SLANG_RHI_HAS_OPTIX
+#if SLANG_RHI_ENABLE_OPTIX
 
 void RayTracingPassEncoderImpl::init(CommandBufferImpl* cmdBuffer)
 {
@@ -329,6 +329,6 @@ Result RayTracingPassEncoderImpl::dispatchRays(
     return SLANG_E_NOT_IMPLEMENTED;
 }
 
-#endif // SLANG_RHI_HAS_OPTIX
+#endif // SLANG_RHI_ENABLE_OPTIX
 
 } // namespace rhi::cuda
diff --git a/src/cuda/cuda-command-encoder.h b/src/cuda/cuda-command-encoder.h
@@ -140,7 +140,7 @@ class ComputePassEncoderImpl : public IComputePassEncoder, public PassEncoderImp
     virtual SLANG_NO_THROW Result SLANG_MCALL dispatchComputeIndirect(IBuffer* argBuffer, Offset offset) override;
 };
 
-#if SLANG_RHI_HAS_OPTIX
+#if SLANG_RHI_ENABLE_OPTIX
 
 class RayTracingPassEncoderImpl : public IRayTracingPassEncoder, public PassEncoderImpl
 {
@@ -203,6 +203,6 @@ class RayTracingPassEncoderImpl : public IRayTracingPassEncoder, public PassEnco
         override;
 };
 
-#endif // SLANG_RHI_HAS_OPTIX
+#endif // SLANG_RHI_ENABLE_OPTIX
 
 } // namespace rhi::cuda
diff --git a/src/cuda/cuda-device.cpp b/src/cuda/cuda-device.cpp
@@ -132,7 +132,7 @@ DeviceImpl::~DeviceImpl()
 {
     m_queue.setNull();
 
-#if SLANG_RHI_HAS_OPTIX
+#if SLANG_RHI_ENABLE_OPTIX
     if (m_ctx.optixContext)
     {
         optixDeviceContextDestroy(m_ctx.optixContext);
@@ -149,7 +149,7 @@ Result DeviceImpl::getNativeDeviceHandles(DeviceNativeHandles* outHandles)
 {
     outHandles->handles[0].type = NativeHandleType::CUdevice;
     outHandles->handles[0].value = m_ctx.device;
-#if SLANG_RHI_HAS_OPTIX
+#if SLANG_RHI_ENABLE_OPTIX
     outHandles->handles[1].type = NativeHandleType::OptixDeviceContext;
     outHandles->handles[1].value = (uint64_t)m_ctx.optixContext;
 #else
@@ -208,7 +208,7 @@ Result DeviceImpl::initialize(const DeviceDesc& desc)
         m_features.push_back("has-ptr");
     }
 
-#if SLANG_RHI_HAS_OPTIX
+#if SLANG_RHI_ENABLE_OPTIX
     {
         SLANG_OPTIX_RETURN_ON_FAIL(optixInit());
 
@@ -1168,7 +1168,7 @@ Result DeviceImpl::getAccelerationStructureSizes(
     AccelerationStructureSizes* outSizes
 )
 {
-#if SLANG_RHI_HAS_OPTIX
+#if SLANG_RHI_ENABLE_OPTIX
     AccelerationStructureBuildInputBuilder builder;
     builder.build(desc, m_debugCallback);
     OptixAccelBufferSizes sizes;
@@ -1194,7 +1194,7 @@ Result DeviceImpl::createAccelerationStructure(
     IAccelerationStructure** outAccelerationStructure
 )
 {
-#if SLANG_RHI_HAS_OPTIX
+#if SLANG_RHI_ENABLE_OPTIX
     RefPtr<AccelerationStructureImpl> result = new AccelerationStructureImpl(this, desc);
     SLANG_CUDA_RETURN_ON_FAIL(cuMemAlloc(&result->m_buffer, desc.size));
     SLANG_CUDA_RETURN_ON_FAIL(cuMemAlloc(&result->m_propertyBuffer, 8));

diff --git a/src/cuda/cuda-device.h b/src/cuda/cuda-device.h
@@ -11,7 +11,7 @@ struct Context
 {
     CUdevice device = -1;
     CUcontext context = nullptr;
-#if SLANG_RHI_HAS_OPTIX
+#if SLANG_RHI_ENABLE_OPTIX
     OptixDeviceContext optixContext = nullptr;
 #endif
 };

diff --git a/src/cuda/cuda-helper-functions.cpp b/src/cuda/cuda-helper-functions.cpp
@@ -17,7 +17,7 @@ Result _handleCUDAError(CUresult cuResult, const char* file, int line)
     return info.handle();
 }
 
-#if SLANG_RHI_HAS_OPTIX
+#if SLANG_RHI_ENABLE_OPTIX
 
 bool _isError(OptixResult result)
 {
@@ -36,7 +36,7 @@ void _optixLogCallback(unsigned int level, const char* tag, const char* message,
     fprintf(stderr, "optix: %s (%s)\n", message, tag);
 }
 #endif
-#endif // SLANG_RHI_HAS_OPTIX
+#endif // SLANG_RHI_ENABLE_OPTIX
 
 AdapterLUID getAdapterLUID(int deviceIndex)
 {

diff --git a/src/cuda/cuda-helper-functions.h b/src/cuda/cuda-helper-functions.h
@@ -64,7 +64,7 @@ Result _handleCUDAError(CUresult cuResult, const char* file, int line);
         };                                                                                                             \
     }
 
-#if SLANG_RHI_HAS_OPTIX
+#if SLANG_RHI_ENABLE_OPTIX
 
 bool _isError(OptixResult result);
 
@@ -87,7 +87,7 @@ Result _handleOptixError(OptixResult result, char const* file, int line);
 
 void _optixLogCallback(unsigned int level, const char* tag, const char* message, void* userData);
 
-#endif // SLANG_RHI_HAS_OPTIX
+#endif // SLANG_RHI_ENABLE_OPTIX
 
 AdapterLUID getAdapterLUID(int deviceIndex);
 

diff --git a/src/d3d11/d3d11-device.cpp b/src/d3d11/d3d11-device.cpp
@@ -228,16 +228,16 @@ Result DeviceImpl::initialize(const DeviceDesc& desc)
 
         if (isSupportedNVAPIOp(m_device, NV_EXTN_OP_UINT64_ATOMIC))
         {
-            m_features.add("atomic-int64");
+            m_features.push_back("atomic-int64");
         }
         if (isSupportedNVAPIOp(m_device, NV_EXTN_OP_FP32_ATOMIC))
         {
-            m_features.add("atomic-float");
+            m_features.push_back("atomic-float");
         }
 
         // If we have NVAPI well assume we have realtime clock
         {
-            m_features.add("realtime-clock");
+            m_features.push_back("realtime-clock");
         }
 
         m_nvapi = true;

diff --git a/src/d3d12/d3d12-helper-functions.cpp b/src/d3d12/d3d12-helper-functions.cpp
@@ -3,7 +3,7 @@
 #include "d3d12-query.h"
 #include "d3d12-transient-heap.h"
 
-#ifdef SLANG_RHI_NVAPI
+#ifdef SLANG_RHI_ENABLE_NVAPI
 #include "../nvapi/nvapi-include.h"
 #endif
 
@@ -21,7 +21,7 @@ namespace rhi::d3d12 {
 
 bool isSupportedNVAPIOp(ID3D12Device* dev, uint32_t op)
 {
-#ifdef SLANG_RHI_NVAPI
+#ifdef SLANG_RHI_ENABLE_NVAPI
     {
         bool isSupported;
         NvAPI_Status status = NvAPI_D3D12_IsNvShaderExtnOpCodeSupported(dev, NvU32(op), &isSupported);

diff --git a/src/d3d12/d3d12-pipeline.cpp b/src/d3d12/d3d12-pipeline.cpp
@@ -4,7 +4,7 @@
 #include "d3d12-shader-program.h"
 #include "d3d12-input-layout.h"
 
-#ifdef SLANG_RHI_NVAPI
+#ifdef SLANG_RHI_ENABLE_NVAPI
 #include "../nvapi/nvapi-include.h"
 #endif
 
@@ -255,7 +255,7 @@ Result DeviceImpl::createComputePipeline2(const ComputePipelineDesc2& desc, ICom
                                      : program->m_rootObjectLayout->m_rootSignature;
     computeDesc.CS = {program->m_shaders[0].code.data(), SIZE_T(program->m_shaders[0].code.size())};
 
-#ifdef SLANG_RHI_NVAPI
+#ifdef SLANG_RHI_ENABLE_NVAPI
     if (m_nvapi)
     {
         // Also fill the extension structure.

diff --git a/src/nvapi/nvapi-include.h b/src/nvapi/nvapi-include.h
@@ -2,7 +2,7 @@
 
 // A helper that makes the NVAPI available across targets
 
-#ifdef SLANG_RHI_NVAPI
+#ifdef SLANG_RHI_ENABLE_NVAPI
 // On windows if we include NVAPI, we must include windows.h first
 
 #ifdef _WIN32

diff --git a/src/nvapi/nvapi-util.cpp b/src/nvapi/nvapi-util.cpp
@@ -7,7 +7,7 @@ static Result g_initStatus = SLANG_E_UNINITIALIZED;
 
 Result NVAPIUtil::initialize()
 {
-#ifdef SLANG_RHI_NVAPI
+#ifdef SLANG_RHI_ENABLE_NVAPI
     if (g_initStatus == SLANG_E_UNINITIALIZED)
     {
         NvAPI_Status ret = NVAPI_OK;