Add nvapi lib to D3D11 for efficient UAV writes overlaps on NVIDIA GPUs

Source/Engine/Graphics/GPUPass.h

@@ -46,6 +46,12 @@ struct FLAXENGINE_API GPUMemoryPass : GPUPass
     {
         Context->MemoryBarrier();
     }
+
+    // Inserts a global memory barrier on data copies between resources. Use to ensure all writes and before submitting another commands.
+    void MemoryBarrier()
+    {
+        Context->MemoryBarrier();
+    }
 };
 
 /// <summary>

Source/Engine/GraphicsDevice/DirectX/DX11/GPUContextDX11.cpp

@@ -14,6 +14,10 @@
 #include "Engine/Core/Math/Viewport.h"
 #include "Engine/Core/Math/Rectangle.h"
 #include "Engine/Profiler/RenderStats.h"
+#if COMPILE_WITH_NVAPI
+#include <ThirdParty/nvapi/nvapi.h>
+extern bool EnableNvapi;
+#endif
 
 #define DX11_CLEAR_SR_ON_STAGE_DISABLE 0
 
@@ -903,6 +907,23 @@ void GPUContextDX11::CopySubresource(GPUResource* dstResource, uint32 dstSubreso
     _context->CopySubresourceRegion(dstResourceDX11->GetResource(), dstSubresource, 0, 0, 0, srcResourceDX11->GetResource(), srcSubresource, nullptr);
 }
 
+void GPUContextDX11::OverlapUA(bool end)
+{
+    // DirectX 11 doesn't support UAV barriers control but custom GPU driver extensions allow to manually specify overlap sections.
+#if COMPILE_WITH_NVAPI
+    if (EnableNvapi)
+    {
+        if (end)
+            NvAPI_D3D11_EndUAVOverlap(_context);
+        else
+            NvAPI_D3D11_BeginUAVOverlap(_context);
+        return;
+    }
+#endif
+    // TODO: add support for AMD extensions to overlap UAV writes (agsDriverExtensionsDX11_BeginUAVOverlap/agsDriverExtensionsDX11_EndUAVOverlap)
+    // TODO: add support for Intel extensions to overlap UAV writes (INTC_D3D11_BeginUAVOverlap/INTC_D3D11_EndUAVOverlap)
+}
+
 void GPUContextDX11::flushSRVs()
 {
 #define FLUSH_STAGE(STAGE) if (Current##STAGE) _context->STAGE##SetShaderResources(0, ARRAY_COUNT(_srHandles), _srHandles)

Source/Engine/GraphicsDevice/DirectX/DX11/GPUContextDX11.h

@@ -167,6 +167,7 @@ public:
     void CopyCounter(GPUBuffer* dstBuffer, uint32 dstOffset, GPUBuffer* srcBuffer) override;
     void CopyResource(GPUResource* dstResource, GPUResource* srcResource) override;
     void CopySubresource(GPUResource* dstResource, uint32 dstSubresource, GPUResource* srcResource, uint32 srcSubresource) override;
+    void OverlapUA(bool end) override;
 };
 
 #endif

Source/Engine/GraphicsDevice/DirectX/DX11/GPUDeviceDX11.cpp

@@ -20,7 +20,10 @@
 #include "Engine/Graphics/PixelFormatExtensions.h"
 #include "Engine/Engine/CommandLine.h"
 #include "Engine/Profiler/ProfilerMemory.h"
-
+#if COMPILE_WITH_NVAPI
+#include <ThirdParty/nvapi/nvapi.h>
+bool EnableNvapi = false;
+#endif
 #if !USE_EDITOR && PLATFORM_WINDOWS
 #include "Engine/Core/Config/PlatformSettings.h"
 #endif
@@ -420,6 +423,31 @@ bool GPUDeviceDX11::Init()
 {
     HRESULT result;
 
+    // Driver extensions
+#if COMPILE_WITH_NVAPI
+    if (_adapter->IsNVIDIA())
+    {
+        NvAPI_Status status = NvAPI_Initialize();
+        if (status == NVAPI_OK)
+        {
+            EnableNvapi = true;
+
+            NvU32 driverVersion;
+            NvAPI_ShortString buildBranch("");
+            if (NvAPI_SYS_GetDriverAndBranchVersion(&driverVersion, buildBranch) == NVAPI_OK)
+            {
+                LOG(Info, "NvApi driver version: {}, {}", driverVersion, StringAsUTF16<>(buildBranch).Get());
+            }
+        }
+        else
+        {
+            NvAPI_ShortString desc;
+            NvAPI_GetErrorMessage(status, desc);
+            LOG(Warning, "NvAPI_Initialize failed with result {} ({})", (int32)status, String(desc));
+        }
+    }
+#endif
+
     // Get DXGI adapter
     ComPtr<IDXGIAdapter> adapter;
     if (_factoryDXGI->EnumAdapters(_adapter->Index, &adapter) == DXGI_ERROR_NOT_FOUND || adapter == nullptr)

Source/Engine/GraphicsDevice/DirectX/DX11/GraphicsDeviceDX11.Build.cs

@@ -14,5 +14,7 @@ public class GraphicsDeviceDX11 : GraphicsDeviceBaseModule
 
         options.PublicDefinitions.Add("GRAPHICS_API_DIRECTX11");
         options.OutputFiles.Add("d3d11.lib");
+        if (nvapi.Use(options))
+            options.PrivateDependencies.Add("nvapi");
     }
 }