Various small fixes and improvements

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

@@ -95,6 +95,7 @@ void GPUContextDX11::FrameBegin()
     GPUContext::FrameBegin();
 
     // Setup
+    _flushOnDispatch = false;
     _omDirtyFlag = false;
     _uaDirtyFlag = false;
     _cbDirtyFlag = false;
@@ -497,50 +498,19 @@ void GPUContextDX11::UpdateCB(GPUConstantBuffer* cb, const void* data)
 
 void GPUContextDX11::Dispatch(GPUShaderProgramCS* shader, uint32 threadGroupCountX, uint32 threadGroupCountY, uint32 threadGroupCountZ)
 {
-    CurrentCS = (GPUShaderProgramCSDX11*)shader;
-
-    // Flush
-    flushCBs();
-    flushSRVs();
-    flushUAVs();
-    flushOM();
-
-    // Dispatch
-    auto compute = (ID3D11ComputeShader*)shader->GetBufferHandle();
-    if (_currentCompute != compute)
-    {
-        _currentCompute = compute;
-        _context->CSSetShader(compute, nullptr, 0);
-    }
+    onDispatch(shader);
     _context->Dispatch(threadGroupCountX, threadGroupCountY, threadGroupCountZ);
     RENDER_STAT_DISPATCH_CALL();
-
     CurrentCS = nullptr;
 }
 
 void GPUContextDX11::DispatchIndirect(GPUShaderProgramCS* shader, GPUBuffer* bufferForArgs, uint32 offsetForArgs)
 {
     ASSERT(bufferForArgs && EnumHasAnyFlags(bufferForArgs->GetFlags(), GPUBufferFlags::Argument));
-    CurrentCS = (GPUShaderProgramCSDX11*)shader;
-
     auto bufferForArgsDX11 = (GPUBufferDX11*)bufferForArgs;
-
-    // Flush
-    flushCBs();
-    flushSRVs();
-    flushUAVs();
-    flushOM();
-
-    // Dispatch
-    auto compute = (ID3D11ComputeShader*)shader->GetBufferHandle();
-    if (_currentCompute != compute)
-    {
-        _currentCompute = compute;
-        _context->CSSetShader(compute, nullptr, 0);
-    }
+    onDispatch(shader);
     _context->DispatchIndirect(bufferForArgsDX11->GetBuffer(), offsetForArgs);
     RENDER_STAT_DISPATCH_CALL();
-
     CurrentCS = nullptr;
 }
 
@@ -921,6 +891,7 @@ void GPUContextDX11::OverlapUA(bool end)
             NvAPI_D3D11_EndUAVOverlap(_context);
         else
             NvAPI_D3D11_BeginUAVOverlap(_context);
+        _flushOnDispatch |= end;
         return;
     }
 #endif
@@ -931,6 +902,7 @@ void GPUContextDX11::OverlapUA(bool end)
             agsDriverExtensionsDX11_EndUAVOverlap(AgsContext, _context);
         else
             agsDriverExtensionsDX11_BeginUAVOverlap(AgsContext, _context);
+        _flushOnDispatch |= end;
         return;
     }
 #endif
@@ -1046,6 +1018,7 @@ void GPUContextDX11::flushIA()
 
 void GPUContextDX11::onDrawCall()
 {
+    _flushOnDispatch = false;
     flushCBs();
     flushSRVs();
     flushUAVs();
@@ -1053,4 +1026,27 @@ void GPUContextDX11::onDrawCall()
     flushOM();
 }
 
+void GPUContextDX11::onDispatch(GPUShaderProgramCS* shader)
+{
+    CurrentCS = (GPUShaderProgramCSDX11*)shader;
+
+    flushCBs();
+    flushSRVs();
+    flushUAVs();
+    flushOM();
+
+    if (_flushOnDispatch)
+    {
+        _flushOnDispatch = false;
+        _context->Flush();
+    }
+
+    auto compute = (ID3D11ComputeShader*)shader->GetBufferHandle();
+    if (_currentCompute != compute)
+    {
+        _currentCompute = compute;
+        _context->CSSetShader(compute, nullptr, 0);
+    }
+}
+
 #endif

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

@@ -30,6 +30,7 @@ private:
     byte _tracyZone[TracyD3D11ZoneSize];
 #endif
     int32 _maxUASlots;
+    bool _flushOnDispatch;
 
     // Output Merger
     bool _omDirtyFlag;
@@ -111,6 +112,7 @@ private:
     void flushOM();
     void flushIA();
     void onDrawCall();
+    void onDispatch(GPUShaderProgramCS* shader);
 
 public:
 

Source/Engine/GraphicsDevice/DirectX/DX12/GPUBufferDX12.cpp

@@ -38,7 +38,7 @@ void* GPUBufferDX12::Map(GPUResourceMapMode mode)
 {
     D3D12_RANGE readRange;
     D3D12_RANGE* readRangePtr;
-    switch (mode)
+    switch (mode & GPUResourceMapMode::ReadWrite)
     {
     case GPUResourceMapMode::Read:
         readRangePtr = nullptr;

Source/Engine/Platform/Win32/Win32ConditionVariable.h

@@ -26,7 +26,7 @@ public:
     /// <summary>
     /// Initializes a new instance of the <see cref="Win32ConditionVariable"/> class.
     /// </summary>
-    Win32ConditionVariable()
+    __forceinline Win32ConditionVariable()
     {
         Windows::InitializeConditionVariable(&_cond);
     }
@@ -44,7 +44,7 @@ public:
     /// Blocks the current thread execution until the condition variable is woken up.
     /// </summary>
     /// <param name="lock">The critical section locked by the current thread.</param>
-    void Wait(const Win32CriticalSection& lock)
+    __forceinline void Wait(const Win32CriticalSection& lock)
     {
         Windows::SleepConditionVariableCS(&_cond, &lock._criticalSection, 0xFFFFFFFF);
     }
@@ -55,7 +55,7 @@ public:
     /// <param name="lock">The critical section locked by the current thread.</param>
     /// <param name="timeout">The time-out interval, in milliseconds. If the time-out interval elapses, the function re-acquires the critical section and returns zero. If timeout is zero, the function tests the states of the specified objects and returns immediately. If timeout is INFINITE, the function's time-out interval never elapses.</param>
     /// <returns>If the function succeeds, the return value is true, otherwise, if the function fails or the time-out interval elapses, the return value is false.</returns>
-    bool Wait(const Win32CriticalSection& lock, const int32 timeout)
+    __forceinline bool Wait(const Win32CriticalSection& lock, const int32 timeout)
     {
         return !!Windows::SleepConditionVariableCS(&_cond, &lock._criticalSection, timeout);
     }
@@ -63,7 +63,7 @@ public:
     /// <summary>
     /// Notifies one waiting thread.
     /// </summary>
-    void NotifyOne()
+    __forceinline void NotifyOne()
     {
         Windows::WakeConditionVariable(&_cond);
     }
@@ -71,7 +71,7 @@ public:
     /// <summary>
     /// Notifies all waiting threads.
     /// </summary>
-    void NotifyAll()
+    __forceinline void NotifyAll()
     {
         Windows::WakeAllConditionVariable(&_cond);
     }

Source/Engine/Platform/Win32/Win32CriticalSection.h

@@ -26,7 +26,7 @@ public:
     /// <summary>
     /// Initializes a new instance of the <see cref="Win32CriticalSection"/> class.
     /// </summary>
-    Win32CriticalSection()
+    __forceinline Win32CriticalSection()
     {
         Windows::InitializeCriticalSectionEx(&_criticalSection, 4000, 0x01000000);
     }
@@ -34,7 +34,7 @@ public:
     /// <summary>
     /// Finalizes an instance of the <see cref="Win32CriticalSection"/> class.
     /// </summary>
-    ~Win32CriticalSection()
+    __forceinline ~Win32CriticalSection()
     {
         Windows::DeleteCriticalSection(&_criticalSection);
     }
@@ -43,7 +43,7 @@ public:
     /// <summary>
     /// Locks the critical section.
     /// </summary>
-    void Lock() const
+    __forceinline void Lock() const
     {
         Windows::EnterCriticalSection(&_criticalSection);
     }
@@ -52,7 +52,7 @@ public:
     /// Attempts to enter a critical section without blocking. If the call is successful, the calling thread takes ownership of the critical section.
     /// </summary>
     /// <returns>True if calling thread took ownership of the critical section.</returns>
-    bool TryLock() const
+    __forceinline bool TryLock() const
     {
         return Windows::TryEnterCriticalSection(&_criticalSection) != 0;
     }
@@ -60,7 +60,7 @@ public:
     /// <summary>
     /// Releases the lock on the critical section.
     /// </summary>
-    void Unlock() const
+    __forceinline void Unlock() const
     {
         Windows::LeaveCriticalSection(&_criticalSection);
     }

Source/Engine/Renderer/GlobalSignDistanceFieldPass.cpp

@@ -924,12 +924,6 @@ bool GlobalSignDistanceFieldPass::Render(RenderContext& renderContext, GPUContex
                 }
             }
 
-#if PLATFORM_WINDOWS
-            // Hack to fix D3D11 bug that doesn't insert UAV barrier after overlap region ends (between two GPUComputePass)
-            if (context->GetDevice()->GetRendererType() == RendererType::DirectX11)
-                context->Dispatch(_csRasterizeModel0, chunkDispatchGroups, chunkDispatchGroups, chunkDispatchGroups);
-#endif
-
             // Rasterize non-empty chunks (additive layers so need combine with existing chunk data)
             for (uint32 layer = 0; layer <= maxLayer; layer++)
             {