Improve robustness of gfx lifetime management. (#1788)

* Improve robustness of gfx lifetime management.

* fix clang error

* fix clang error

* Fix clang warning

1 files changed, 63 insertions, 52 deletions

diff --git a/tools/gfx/cuda/render-cuda.cpp b/tools/gfx/cuda/render-cuda.cpp
index dbee0c5f2..f871e6246 100644
--- a/tools/gfx/cuda/render-cuda.cpp
+++ b/tools/gfx/cuda/render-cuda.cpp
@@ -181,6 +181,13 @@ void _optixLogCallback(unsigned int level, const char* tag, const char* message,
 
 #    endif
 
+class CUDAContext : public RefObject
+{
+public:
+    CUcontext m_context = nullptr;
+    ~CUDAContext() { cuCtxDestroy(m_context); }
+};
+
 class MemoryCUDAResource : public BufferResource
 {
 public:
@@ -199,6 +206,8 @@ public:
     uint64_t getBindlessHandle() { return (uint64_t)m_cudaMemory; }
 
     void* m_cudaMemory = nullptr;
+
+    RefPtr<CUDAContext> m_cudaContext;
 };
 
 class TextureCUDAResource : public TextureResource
@@ -238,12 +247,14 @@ public:
 
     CUarray m_cudaArray = CUarray();
     CUmipmappedArray m_cudaMipMappedArray = CUmipmappedArray();
+
+    RefPtr<CUDAContext> m_cudaContext;
 };
 
-class CUDAResourceView : public IResourceView, public RefObject
+class CUDAResourceView : public IResourceView, public ComObject
 {
 public:
-    SLANG_REF_OBJECT_IUNKNOWN_ALL
+    SLANG_COM_OBJECT_IUNKNOWN_ALL
     IResourceView* getInterface(const Guid& guid)
     {
         if (guid == GfxGUID::IID_ISlangUnknown || guid == GfxGUID::IID_IResourceView)
@@ -451,7 +462,7 @@ public:
         ComPtr<IBufferResource> constantBuffer;
         SLANG_RETURN_ON_FAIL(device->createBufferResource(
             IResource::Usage::ConstantBuffer, bufferDesc, nullptr, constantBuffer.writeRef()));
-        bufferResource = dynamic_cast<MemoryCUDAResource*>(constantBuffer.get());
+        bufferResource = static_cast<MemoryCUDAResource*>(constantBuffer.get());
         return SLANG_OK;
     }
 
@@ -514,8 +525,7 @@ public:
             *object = nullptr;
             return SLANG_OK;
         }
-        objects[subObjectIndex]->addRef();
-        *object = objects[subObjectIndex].Ptr();
+        returnComPtr(object, objects[subObjectIndex]);
         return SLANG_OK;
     }
     virtual SLANG_NO_THROW Result SLANG_MCALL
@@ -648,7 +658,7 @@ public:
         auto& bindingRange = layout->m_bindingRanges[bindingRangeIndex];
 
         auto viewIndex = bindingRange.baseIndex + offset.bindingArrayIndex;
-        auto cudaView = dynamic_cast<CUDAResourceView*>(resourceView);
+        auto cudaView = static_cast<CUDAResourceView*>(resourceView);
 
         resources[viewIndex] = cudaView;
 
@@ -720,7 +730,6 @@ public:
                 // type, we need to make sure to use that type as the specialization argument.
 
                 // TODO: need to implement the case where the field is an array of existential values.
-                SLANG_ASSERT(bindingRange.count == 1);
                 ExtendedShaderObjectType specializedSubObjType;
                 SLANG_RETURN_ON_FAIL(objects[subObjIndex]->getSpecializedShaderObjectType(&specializedSubObjType));
                 args.add(specializedSubObjType);
@@ -801,6 +810,12 @@ public:
 class CUDARootShaderObject : public CUDAShaderObject
 {
 public:
+    // Override default reference counting behavior to disable lifetime management.
+    // Root objects are managed by command buffer and does not need to be freed by the user.
+    SLANG_NO_THROW uint32_t SLANG_MCALL addRef() override { return 1; }
+    SLANG_NO_THROW uint32_t SLANG_MCALL release() override { return 1; }
+
+public:
     List<RefPtr<CUDAEntryPointShaderObject>> entryPointObjects;
     virtual SLANG_NO_THROW Result SLANG_MCALL
         init(IDevice* device, CUDAShaderObjectLayout* typeLayout) override;
@@ -808,8 +823,7 @@ public:
     virtual SLANG_NO_THROW Result SLANG_MCALL
         getEntryPoint(UInt index, IShaderObject** outEntryPoint) override
     {
-        *outEntryPoint = entryPointObjects[index].Ptr();
-        entryPointObjects[index]->addRef();
+        returnComPtr(outEntryPoint, entryPointObjects[index]);
         return SLANG_OK;
     }
     virtual Result collectSpecializationArgs(ExtendedShaderObjectTypeList& args) override
@@ -830,6 +844,7 @@ public:
     CUfunction cudaKernel;
     String kernelName;
     RefPtr<CUDAProgramLayout> layout;
+    RefPtr<CUDAContext> cudaContext;
     ~CUDAShaderProgram()
     {
         if (cudaModule)
@@ -969,7 +984,7 @@ private:
 private:
     int m_deviceIndex = -1;
     CUdevice m_device = 0;
-    CUcontext m_context = nullptr;
+    RefPtr<CUDAContext> m_context;
     DeviceInfo m_info;
     String m_adapterName;
 
@@ -979,10 +994,10 @@ public:
     class CommandBufferImpl
         : public ICommandBuffer
         , public CommandWriter
-        , public RefObject
+        , public ComObject
     {
     public:
-        SLANG_REF_OBJECT_IUNKNOWN_ALL
+        SLANG_COM_OBJECT_IUNKNOWN_ALL
         ICommandBuffer* getInterface(const Guid& guid)
         {
             if (guid == GfxGUID::IID_ISlangUnknown || guid == GfxGUID::IID_ICommandBuffer)
@@ -1010,7 +1025,7 @@ public:
             virtual SLANG_NO_THROW SlangResult SLANG_MCALL
                 queryInterface(SlangUUID const& uuid, void** outObject) override
             {
-                if (uuid == GfxGUID::IID_ISlangUnknown ||
+                if (uuid == GfxGUID::IID_ISlangUnknown || uuid == GfxGUID::IID_ICommandEncoder ||
                     uuid == GfxGUID::IID_IComputeCommandEncoder)
                 {
                     *outObject = static_cast<IComputeCommandEncoder*>(this);
@@ -1025,7 +1040,7 @@ public:
         public:
             CommandWriter* m_writer;
             CommandBufferImpl* m_commandBuffer;
-            ComPtr<IShaderObject> m_rootObject;
+            RefPtr<ShaderObjectBase> m_rootObject;
             virtual SLANG_NO_THROW void SLANG_MCALL endEncoding() override {}
             void init(CommandBufferImpl* cmdBuffer)
             {
@@ -1039,8 +1054,8 @@ public:
                 m_writer->setPipelineState(state);
                 PipelineStateBase* pipelineImpl = static_cast<PipelineStateBase*>(state);
                 SLANG_RETURN_ON_FAIL(m_commandBuffer->m_device->createRootShaderObject(
-                    pipelineImpl->m_program, outRootObject));
-                m_rootObject = *outRootObject;
+                    pipelineImpl->m_program, m_rootObject.writeRef()));
+                returnComPtr(outRootObject, m_rootObject);
                 return SLANG_OK;
             }
 
@@ -1066,7 +1081,7 @@ public:
             virtual SLANG_NO_THROW SlangResult SLANG_MCALL
                 queryInterface(SlangUUID const& uuid, void** outObject) override
             {
-                if (uuid == GfxGUID::IID_ISlangUnknown ||
+                if (uuid == GfxGUID::IID_ISlangUnknown || uuid == GfxGUID::IID_ICommandEncoder ||
                     uuid == GfxGUID::IID_IResourceCommandEncoder)
                 {
                     *outObject = static_cast<IResourceCommandEncoder*>(this);
@@ -1118,10 +1133,10 @@ public:
 
     class CommandQueueImpl
         : public ICommandQueue
-        , public RefObject
+        , public ComObject
     {
     public:
-        SLANG_REF_OBJECT_IUNKNOWN_ALL
+        SLANG_COM_OBJECT_IUNKNOWN_ALL
         ICommandQueue* getInterface(const Guid& guid)
         {
             if (guid == GfxGUID::IID_ISlangUnknown || guid == GfxGUID::IID_ICommandQueue)
@@ -1176,7 +1191,7 @@ public:
             currentPipeline = dynamic_cast<CUDAPipelineState*>(state);
         }
 
-        Result bindRootShaderObject(PipelineType pipelineType, IShaderObject* object)
+        Result bindRootShaderObject(IShaderObject* object)
         {
             currentRootObject = dynamic_cast<CUDARootShaderObject*>(object);
             if (currentRootObject)
@@ -1294,8 +1309,7 @@ public:
                     break;
                 case CommandName::BindRootShaderObject:
                     bindRootShaderObject(
-                        (PipelineType)cmd.operands[0],
-                        commandBuffer->getObject<IShaderObject>(cmd.operands[1]));
+                        commandBuffer->getObject<IShaderObject>(cmd.operands[0]));
                     break;
                 case CommandName::DispatchCompute:
                     dispatchCompute(
@@ -1324,13 +1338,6 @@ public:
     using TransientResourceHeapImpl = SimpleTransientResourceHeap<CUDADevice, CommandBufferImpl>;
 
 public:
-    ~CUDADevice()
-    {
-        if (m_context)
-        {
-            cuCtxDestroy(m_context);
-        }
-    }
     virtual SLANG_NO_THROW SlangResult SLANG_MCALL initialize(const Desc& desc) override
     {
         SLANG_RETURN_ON_FAIL(slangContext.initialize(desc.slang, SLANG_PTX, "sm_5_1"));
@@ -1342,15 +1349,12 @@ public:
         SLANG_RETURN_ON_FAIL(_findMaxFlopsDeviceIndex(&m_deviceIndex));
         SLANG_CUDA_RETURN_WITH_REPORT_ON_FAIL(cudaSetDevice(m_deviceIndex), reportType);
 
-        if (m_context)
-        {
-            cuCtxDestroy(m_context);
-            m_context = nullptr;
-        }
+        m_context = new CUDAContext();
 
         SLANG_CUDA_RETURN_ON_FAIL(cuDeviceGet(&m_device, m_deviceIndex));
 
-        SLANG_CUDA_RETURN_WITH_REPORT_ON_FAIL(cuCtxCreate(&m_context, 0, m_device), reportType);
+        SLANG_CUDA_RETURN_WITH_REPORT_ON_FAIL(
+            cuCtxCreate(&m_context->m_context, 0, m_device), reportType);
 
         // Initialize DeviceInfo
         {
@@ -1375,7 +1379,12 @@ public:
         const ITextureResource::SubresourceData* initData,
         ITextureResource** outResource) override
     {
-        RefPtr<TextureCUDAResource> tex = new TextureCUDAResource(desc);
+        TextureResource::Desc srcDesc(desc);
+        srcDesc.setDefaults(initialUsage);
+
+        RefPtr<TextureCUDAResource> tex = new TextureCUDAResource(srcDesc);
+        tex->m_cudaContext = m_context;
+
         CUresourcetype resourceType;
         size_t elementSize = 0;
 
@@ -1777,7 +1786,7 @@ public:
                 cuTexObjectCreate(&tex->m_cudaTexObj, &resDesc, &texDesc, nullptr));
         }
 
-        *outResource = tex.detach();
+        returnComPtr(outResource, tex);
         return SLANG_OK;
     }
 
@@ -1788,12 +1797,13 @@ public:
         IBufferResource** outResource) override
     {
         RefPtr<MemoryCUDAResource> resource = new MemoryCUDAResource(desc);
+        resource->m_cudaContext = m_context;
         SLANG_CUDA_RETURN_ON_FAIL(cudaMallocManaged(&resource->m_cudaMemory, desc.sizeInBytes));
         if (initData)
         {
             SLANG_CUDA_RETURN_ON_FAIL(cudaMemcpy(resource->m_cudaMemory, initData, desc.sizeInBytes, cudaMemcpyHostToDevice));
         }
-        *outResource = resource.detach();
+        returnComPtr(outResource, resource);
         return SLANG_OK;
     }
 
@@ -1803,7 +1813,7 @@ public:
         RefPtr<CUDAResourceView> view = new CUDAResourceView();
         view->desc = desc;
         view->textureResource = dynamic_cast<TextureCUDAResource*>(texture);
-        *outView = view.detach();
+        returnComPtr(outView, view);
         return SLANG_OK;
     }
 
@@ -1813,7 +1823,7 @@ public:
         RefPtr<CUDAResourceView> view = new CUDAResourceView();
         view->desc = desc;
         view->memoryResource = dynamic_cast<MemoryCUDAResource*>(buffer);
-        *outView = view.detach();
+        returnComPtr(outView, view);
         return SLANG_OK;
     }
 
@@ -1823,7 +1833,7 @@ public:
     {
         RefPtr<CUDAShaderObjectLayout> cudaLayout;
         cudaLayout = new CUDAShaderObjectLayout(this, typeLayout);
-        *outLayout = cudaLayout.detach();
+        returnRefPtrMove(outLayout, cudaLayout);
         return SLANG_OK;
     }
 
@@ -1833,18 +1843,18 @@ public:
     {
         RefPtr<CUDAShaderObject> result = new CUDAShaderObject();
         SLANG_RETURN_ON_FAIL(result->init(this, dynamic_cast<CUDAShaderObjectLayout*>(layout)));
-        *outObject = result.detach();
+        returnComPtr(outObject, result);
         return SLANG_OK;
     }
 
-    Result createRootShaderObject(IShaderProgram* program, IShaderObject** outObject)
+    Result createRootShaderObject(IShaderProgram* program, ShaderObjectBase** outObject)
     {
         auto cudaProgram = dynamic_cast<CUDAShaderProgram*>(program);
         auto cudaLayout = cudaProgram->layout;
 
         RefPtr<CUDARootShaderObject> result = new CUDARootShaderObject();
         SLANG_RETURN_ON_FAIL(result->init(this, cudaLayout));
-        *outObject = result.detach();
+        returnRefPtrMove(outObject, result);
         return SLANG_OK;
     }
 
@@ -1856,10 +1866,11 @@ public:
         // the shader object bindings.
         RefPtr<CUDAShaderProgram> cudaProgram = new CUDAShaderProgram();
         cudaProgram->slangProgram = desc.slangProgram;
+        cudaProgram->cudaContext = m_context;
         if (desc.slangProgram->getSpecializationParamCount() != 0)
         {
             cudaProgram->layout = new CUDAProgramLayout(this, desc.slangProgram->getLayout());
-            *outProgram = cudaProgram.detach();
+            returnComPtr(outProgram, cudaProgram);
             return SLANG_OK;
         }
 
@@ -1893,7 +1904,7 @@ public:
             cudaProgram->layout = cudaLayout;
         }
 
-        *outProgram = cudaProgram.detach();
+        returnComPtr(outProgram, cudaProgram);
         return SLANG_OK;
     }
 
@@ -1901,9 +1912,9 @@ public:
         const ComputePipelineStateDesc& desc, IPipelineState** outState) override
     {
         RefPtr<CUDAPipelineState> state = new CUDAPipelineState();
-        state->shaderProgram = dynamic_cast<CUDAShaderProgram*>(desc.program);
+        state->shaderProgram = static_cast<CUDAShaderProgram*>(desc.program);
         state->init(desc);
-        *outState = state.detach();
+        returnComPtr(outState, state);
         return Result();
     }
 
@@ -1929,7 +1940,7 @@ public:
     {
         RefPtr<TransientResourceHeapImpl> result = new TransientResourceHeapImpl();
         SLANG_RETURN_ON_FAIL(result->init(this, desc));
-        *outHeap = result.detach();
+        returnComPtr(outHeap, result);
         return SLANG_OK;
     }
 
@@ -1938,7 +1949,7 @@ public:
     {
         RefPtr<CommandQueueImpl> queue = new CommandQueueImpl();
         queue->init(this);
-        *outQueue = queue.detach();
+        returnComPtr(outQueue, queue);
         return SLANG_OK;
     }
     virtual SLANG_NO_THROW Result SLANG_MCALL createSwapchain(
@@ -2027,7 +2038,7 @@ public:
             (uint8_t*)bufferImpl->m_cudaMemory + offset,
             size,
             cudaMemcpyDefault);
-        *outBlob = blob.detach();
+        returnComPtr(outBlob, blob);
         return SLANG_OK;
     }
 };
@@ -2115,7 +2126,7 @@ SlangResult SLANG_MCALL createCUDADevice(const IDevice::Desc* desc, IDevice** ou
 {
     RefPtr<CUDADevice> result = new CUDADevice();
     SLANG_RETURN_ON_FAIL(result->initialize(*desc));
-    *outDevice = result.detach();
+    returnComPtr(outDevice, result);
     return SLANG_OK;
 }
 #else