[idd] capture: retain frames in linear buffers

Store retained frame candidates in the final linear transport layout.
Prepare texture output in local DEFAULT buffers before cadence holding.
Then publish full frames with a single buffer copy.

Keep partial damage on matching pitched layouts and leave effect-owned
buffer copies unchanged. This moves detiling out of the
deadline-critical IVSHMEM or readback transfer.

idd/LGIdd/CPostProcessor.cpp

@@ -351,15 +351,14 @@ bool CPostProcessor::ShouldCopyFully(
     m_copyEffect->ShouldCopyFully(dirtyRects, nbDirtyRects);
 }
 
-void CPostProcessor::CopyFrame(
+void CPostProcessor::CopyToCandidate(
   const ComPtr<ID3D12GraphicsCommandList>& commandList,
-  ID3D12Resource * dst, ID3D12Resource * src,
+  ID3D12Resource * dst, ID3D12Resource * src) const
-  const RECT dirtyRects[], unsigned nbDirtyRects, bool fullCopy) const
 {
   if (m_copyEffect)
   {
     m_copyEffect->CopyFrame(
-      commandList, dst, src, dirtyRects, nbDirtyRects, fullCopy);
+      commandList, dst, src, nullptr, 0, true);
     return;
   }
 
@@ -373,10 +372,36 @@ void CPostProcessor::CopyFrame(
   dstLoc.Type            = D3D12_TEXTURE_COPY_TYPE_PLACED_FOOTPRINT;
   dstLoc.PlacedFootprint = m_copyLayout;
 
+  commandList->CopyTextureRegion(
+    &dstLoc, 0, 0, 0, &srcLoc, nullptr);
+}
+
+void CPostProcessor::CopyFromCandidate(
+  const ComPtr<ID3D12GraphicsCommandList>& commandList,
+  ID3D12Resource * dst, ID3D12Resource * src,
+  const RECT dirtyRects[], unsigned nbDirtyRects, bool fullCopy) const
+{
+  if (m_copyEffect)
+  {
+    m_copyEffect->CopyFrame(
+      commandList, dst, src, dirtyRects, nbDirtyRects, fullCopy);
+    return;
+  }
+
+  D3D12_TEXTURE_COPY_LOCATION srcLoc = {};
+  srcLoc.pResource       = src;
+  srcLoc.Type            = D3D12_TEXTURE_COPY_TYPE_PLACED_FOOTPRINT;
+  srcLoc.PlacedFootprint = m_copyLayout;
+
+  D3D12_TEXTURE_COPY_LOCATION dstLoc = {};
+  dstLoc.pResource       = dst;
+  dstLoc.Type            = D3D12_TEXTURE_COPY_TYPE_PLACED_FOOTPRINT;
+  dstLoc.PlacedFootprint = m_copyLayout;
+
   if (fullCopy)
   {
-    commandList->CopyTextureRegion(
+    commandList->CopyBufferRegion(
-      &dstLoc, 0, 0, 0, &srcLoc, nullptr);
+      dst, 0, src, 0, m_frameSize);
     return;
   }
 

idd/LGIdd/CPostProcessor.h

@@ -183,7 +183,10 @@ public:
   size_t   GetOutputSize () const { return m_frameSize; }
   bool ShouldCopyFully(
     const RECT dirtyRects[], unsigned nbDirtyRects) const;
-  void CopyFrame(
+  void CopyToCandidate(
+    const ComPtr<ID3D12GraphicsCommandList>& commandList,
+    ID3D12Resource * dst, ID3D12Resource * src) const;
+  void CopyFromCandidate(
     const ComPtr<ID3D12GraphicsCommandList>& commandList,
     ID3D12Resource * dst, ID3D12Resource * src,
     const RECT dirtyRects[], unsigned nbDirtyRects, bool fullCopy) const;

idd/LGIdd/CSwapChainProcessor.cpp

@@ -970,14 +970,23 @@ static bool ResourceDescMatches(
 }
 
 bool CSwapChainProcessor::EnsureCandidateResource(
-  unsigned candidateIndex, ID3D12Resource * source)
+  unsigned candidateIndex, size_t frameSize)
 {
   FrameCandidate& candidate = m_candidates[candidateIndex];
-  D3D12_RESOURCE_DESC desc  = source->GetDesc();
+
-  desc.Alignment = 0;
+  // Keep the transport layout in local GPU memory so publication does not
-  desc.Flags     = static_cast<D3D12_RESOURCE_FLAGS>(
+  // combine texture detiling with the IVSHMEM or readback transfer.
-    static_cast<UINT>(desc.Flags) &
+  D3D12_RESOURCE_DESC desc = {};
-      ~static_cast<UINT>(D3D12_RESOURCE_FLAG_ALLOW_CROSS_ADAPTER));
+  desc.Dimension          = D3D12_RESOURCE_DIMENSION_BUFFER;
+  desc.Width              = frameSize;
+  desc.Height             = 1;
+  desc.DepthOrArraySize   = 1;
+  desc.MipLevels          = 1;
+  desc.Format             = DXGI_FORMAT_UNKNOWN;
+  desc.SampleDesc.Count   = 1;
+  desc.SampleDesc.Quality = 0;
+  desc.Layout             = D3D12_TEXTURE_LAYOUT_ROW_MAJOR;
+  desc.Flags              = D3D12_RESOURCE_FLAG_NONE;
 
   if (candidate.resource &&
       ResourceDescMatches(candidate.resource->GetDesc(), desc))
@@ -1178,7 +1187,7 @@ bool CSwapChainProcessor::PublishNewestCandidate(
   copySlot->SetCompletionCallback(&CompletionFunction, this, fbRes);
 
   copySlot->BeginTiming();
-  postProcessor.CopyFrame(
+  postProcessor.CopyFromCandidate(
     copySlot->GetGfxList(), fbRes->Get().Get(), candidate.resource.Get(),
     copyDirtyRects, nbCopyDirtyRects, fullCopy);
   copySlot->EndTiming();
@@ -1682,7 +1691,8 @@ bool CSwapChainProcessor::SwapChainNewFrame(ComPtr<IDXGIResource> acquiredBuffer
   ClipDirtyRects(currentDirtyRects, &nbDirtyRects,
     dstFormat.width, dstFormat.height);
 
-  if (!EnsureCandidateResource(candidateIndex, copySrcResource.Get()))
+  const size_t frameSize = postProcessor.GetOutputSize();
+  if (!EnsureCandidateResource(candidateIndex, frameSize))
   {
     copySlot->Cancel();
     if (computeSlot)
@@ -1696,7 +1706,7 @@ bool CSwapChainProcessor::SwapChainNewFrame(ComPtr<IDXGIResource> acquiredBuffer
   candidate.dstFormat          = dstFormat;
   candidate.nbDirtyRects       = nbDirtyRects;
   candidate.pitch              = postProcessor.GetOutputPitch();
-  candidate.frameSize          = postProcessor.GetOutputSize();
+  candidate.frameSize          = frameSize;
   candidate.captureTime        = captureTime;
   candidate.postProcessStart   = postProcessStart;
   candidate.prepareCopyStart   = CFrameScheduler::Nanotime();
@@ -1714,26 +1724,9 @@ bool CSwapChainProcessor::SwapChainNewFrame(ComPtr<IDXGIResource> acquiredBuffer
   copySlot->SetCompletionCallback(
     &CandidateCompletionFunction, this, &candidate);
   copySlot->BeginTiming();
-  const D3D12_RESOURCE_DESC copySrcDesc = copySrcResource->GetDesc();
+  postProcessor.CopyToCandidate(
-  if (copySrcDesc.Dimension == D3D12_RESOURCE_DIMENSION_BUFFER)
+    copySlot->GetGfxList(), candidate.resource.Get(),
-    copySlot->GetGfxList()->CopyBufferRegion(
+    copySrcResource.Get());
-      candidate.resource.Get(), 0, copySrcResource.Get(), 0,
-      copySrcDesc.Width);
-  else
-  {
-    D3D12_TEXTURE_COPY_LOCATION srcLocation = {};
-    srcLocation.pResource        = copySrcResource.Get();
-    srcLocation.Type             = D3D12_TEXTURE_COPY_TYPE_SUBRESOURCE_INDEX;
-    srcLocation.SubresourceIndex = 0;
-
-    D3D12_TEXTURE_COPY_LOCATION dstLocation = {};
-    dstLocation.pResource        = candidate.resource.Get();
-    dstLocation.Type             = D3D12_TEXTURE_COPY_TYPE_SUBRESOURCE_INDEX;
-    dstLocation.SubresourceIndex = 0;
-
-    copySlot->GetGfxList()->CopyTextureRegion(
-      &dstLocation, 0, 0, 0, &srcLocation, nullptr);
-  }
   copySlot->EndTiming();
 
   if (!copySlot->Execute())

idd/LGIdd/CSwapChainProcessor.h

@@ -149,8 +149,8 @@ private:
   bool HasReadyCandidate();
   int AcquireCandidate(bool exclusiveSample);
   void ReleaseCandidate(unsigned candidateIndex);
-  bool EnsureCandidateResource(unsigned candidateIndex,
+  bool EnsureCandidateResource(
-    ID3D12Resource * source);
+    unsigned candidateIndex, size_t frameSize);
   void ResetCandidates();
   void SignalCandidateState();