[idd] publish frames after copy submission

Submit the D3D12 copy before exposing the frame through LGMP, preventing
clients from waiting on work that has not yet been queued. Track the
last published buffer separately for safe subscriber resends and handle
submission failures.

idd/LGIdd/CIndirectDeviceContext.cpp

@@ -820,7 +820,7 @@ bool CIndirectDeviceContext::SetupLGMP(size_t alignSize)
 
 void CIndirectDeviceContext::DeInitLGMP()
 {
-  m_hasFrame = false;
+  InterlockedExchange(&m_publishedFrameIndex, -1);
 
   // The retry timer callback dereferences this context, so make sure it is
   // stopped and drained before we tear anything down. Wait for any in-flight
@@ -910,8 +910,10 @@ void CIndirectDeviceContext::LGMPTimer()
 
   if (lgmpHostQueueNewSubs(m_frameQueue) && m_monitor)
   {
-    if (m_hasFrame)
-      lgmpHostQueuePost(m_frameQueue, 0, m_frameMemory[m_frameIndex]);
+    const LONG frameIndex =
+      InterlockedCompareExchange(&m_publishedFrameIndex, 0, 0);
+    if (frameIndex >= 0)
+      lgmpHostQueuePost(m_frameQueue, 0, m_frameMemory[frameIndex]);
   }
 
   if (lgmpHostQueueNewSubs(m_pointerQueue))
@@ -1047,15 +1049,33 @@ CIndirectDeviceContext::PreparedFrameBuffer CIndirectDeviceContext::PrepareFrame
   FrameBuffer* fb = m_frameBuffer[m_frameIndex];
   fb->wp = 0;
 
-  lgmpHostQueuePost(m_frameQueue, 0, m_frameMemory[m_frameIndex]);
-
   result.frameIndex = m_frameIndex;
   result.mem        = fb->data;
 
-  m_hasFrame = true;
   return result;
 }
 
+bool CIndirectDeviceContext::PublishFrameBuffer(unsigned frameIndex)
+{
+  if (!m_frameQueue || frameIndex >= LGMP_Q_FRAME_LEN)
+    return false;
+
+  /* Make resends select this submitted frame before posting it. This prevents
+   * a new subscriber racing publication from receiving the previous frame
+   * after the new one. */
+  InterlockedExchange(&m_publishedFrameIndex, (LONG)frameIndex);
+
+  const LGMP_STATUS status =
+    lgmpHostQueuePost(m_frameQueue, 0, m_frameMemory[frameIndex]);
+  if (status != LGMP_OK)
+  {
+    DEBUG_ERROR("Failed to publish frame: %s", lgmpStatusString(status));
+    return false;
+  }
+
+  return true;
+}
+
 void CIndirectDeviceContext::WriteFrameBuffer(unsigned frameIndex, void* src, size_t offset, size_t len, bool setWritePos) const
 {
   FrameBuffer * fb = m_frameBuffer[frameIndex];

idd/LGIdd/CIndirectDeviceContext.h

@@ -91,11 +91,12 @@ private:
   bool m_cursorVisible = false;
   int m_cursorX = 0, m_cursorY = 0;
 
-  size_t         m_alignSize    = 0;
-  size_t         m_maxFrameSize = 0;
-  int            m_frameIndex   = 0;
-  uint32_t       m_formatVer    = 0;
-  uint32_t       m_frameSerial  = 0;
+  size_t         m_alignSize           = 0;
+  size_t         m_maxFrameSize        = 0;
+  int            m_frameIndex          = 0;
+  volatile LONG  m_publishedFrameIndex = -1;
+  uint32_t       m_formatVer           = 0;
+  uint32_t       m_frameSerial         = 0;
   PLGMPMemory    m_frameMemory[LGMP_Q_FRAME_LEN] = {};
   KVMFRFrame   * m_frame      [LGMP_Q_FRAME_LEN] = {};
   FrameBuffer  * m_frameBuffer[LGMP_Q_FRAME_LEN] = {};
@@ -105,8 +106,6 @@ private:
   unsigned    m_pitch    = 0;
   DXGI_FORMAT m_format   = DXGI_FORMAT_UNKNOWN;
   FrameType   m_frameType = FRAME_TYPE_INVALID;
-  bool        m_hasFrame = false;
-
   UINT m_iddCxVersion = 0;
   bool m_canProcessFP16 = false;
 
@@ -211,6 +210,7 @@ public:
   };
 
   PreparedFrameBuffer PrepareFrameBuffer(unsigned pitch, const D12FrameFormat& srcFormat, const D12FrameFormat& dstFormat, const RECT * dirtyRects, unsigned nbDirtyRects);
+  bool PublishFrameBuffer(unsigned frameIndex);
   void WriteFrameBuffer(unsigned frameIndex, void* src, size_t offset, size_t len, bool setWritePos) const;
   void FinalizeFrameBuffer(unsigned frameIndex) const;
 

idd/LGIdd/CSwapChainProcessor.cpp

@@ -224,7 +224,8 @@ bool CSwapChainProcessor::SwapChainThreadCore()
       if (frameNumber != lastFrameNumber)
       {
         lastFrameNumber = frameNumber;
-        SwapChainNewFrame(surface, dirtyRectCount, colorSpace, sdrWhiteLevel);
+        if (!SwapChainNewFrame(surface, dirtyRectCount, colorSpace, sdrWhiteLevel))
+          DEBUG_WARN("Failed to submit frame");
 
         // report that all GPU processing for this frame has been queued
         hr = IddCxSwapChainFinishedProcessingFrame(m_hSwapChain);
@@ -582,9 +583,10 @@ bool CSwapChainProcessor::SwapChainNewFrame(ComPtr<IDXGIResource> acquiredBuffer
   memcpy(m_dirtyRects, currentDirtyRects, nbDirtyRects * sizeof(*m_dirtyRects));
   m_nbDirtyRects = nbDirtyRects;
 
-  copyQueue->Execute();
+  if (!copyQueue->Execute())
+    return false;
 
-  return true;
+  return m_devContext->PublishFrameBuffer(buffer.frameIndex);
 }
 
 DWORD CALLBACK CSwapChainProcessor::_CursorThread(LPVOID arg)