[idd] rgb24: compare stable candidate timings

Ignore allocation alignment when matching retained candidate resources.
D3D12 may resolve automatic alignment requests in GetDesc.
Comparing that value recreated the candidate for every benchmark sample.

Measure native and packed over matching wall-clock intervals.
Use preparation and publication intervals, excluding the cadence hold.
This avoids mixing GPU timestamps with fallback timing. The CPU IVSHMEM
copy is included only on indirect systems.

idd/LGIdd/CSwapChainProcessor.cpp

@@ -598,6 +598,7 @@ void CSwapChainProcessor::CompletionFunction(
   uint64_t prepareReady;
   uint64_t prepareGPUStart;
   uint64_t prepareGPUEnd;
+  uint64_t timingStart;
   bool     prepareTimingValid;
   AcquireSRWLockShared(&sc->m_candidateLock);
   const FrameCandidate& candidate = sc->m_candidates[candidateIndex];
@@ -605,6 +606,7 @@ void CSwapChainProcessor::CompletionFunction(
   prepareReady       = candidate.prepareReady;
   prepareGPUStart    = candidate.prepareGPUStart;
   prepareGPUEnd      = candidate.prepareGPUEnd;
+  timingStart        = candidate.timingStart;
   prepareTimingValid = candidate.prepareTimingValid;
   ReleaseSRWLockShared(&sc->m_candidateLock);
 
@@ -614,8 +616,8 @@ void CSwapChainProcessor::CompletionFunction(
   uint64_t       indirectCopyTime = 0;
   if (sc->m_dx12Device->IsIndirectCopy())
   {
-    // GPU timestamps end at the readback copy. Include the following CPU copy
+    // GPU timestamps end at the readback copy. Track the following CPU copy
-    // to IVSHMEM in effect benchmark samples without charging it to Ready.
+    // separately for frame metrics; benchmark wall time includes it directly.
     const uint64_t indirectCopyStart = CFrameScheduler::Nanotime();
     sc->m_devContext->WriteFrameBuffer(
       fbRes->GetFrameIndex(), fbRes->GetMap(), 0, fbRes->GetFrameSize(), false);
@@ -652,9 +654,18 @@ void CSwapChainProcessor::CompletionFunction(
   const uint64_t measured = postProcessTime + copyTime;
   const uint64_t readyTime = elapsed > measured ? elapsed - measured : 0;
 
-  sc->m_postProcessors[candidateIndex].RecordTiming(
+  // Use matching wall-clock boundaries for both modes. The split excludes the
-    fbRes->GetTimingEffectIndex(), fbRes->GetTimingToken(),
+  // cadence hold while including the indirect CPU copy only when it occurs.
-    fbRes->IsFullCopy(), postProcessTime + copyTime);
+  const uint64_t timingToken = fbRes->GetTimingToken();
+  if (timingToken && timingStart && prepareReady >= timingStart &&
+      readyEnd >= publishStart)
+  {
+    const uint64_t totalTime =
+      (prepareReady - timingStart) + (readyEnd - publishStart);
+    sc->m_postProcessors[candidateIndex].RecordTiming(
+      fbRes->GetTimingEffectIndex(), timingToken,
+      fbRes->IsFullCopy(), totalTime);
+  }
   sc->m_devContext->RecordFrameTiming(readyEnd - publishStart);
   sc->m_devContext->SetFrameTiming(fbRes->GetFrameIndex(),
     fbRes->GetCaptureTime(), postProcessTime, copyTime, readyTime);
@@ -943,9 +954,10 @@ void CSwapChainProcessor::ReleaseCandidate(unsigned candidateIndex)
 static bool ResourceDescMatches(
   const D3D12_RESOURCE_DESC& left, const D3D12_RESOURCE_DESC& right)
 {
+  // Alignment is allocation metadata. GetDesc may report the resolved value
+  // when the creation descriptor requested automatic alignment.
   return
     left.Dimension          == right.Dimension          &&
-    left.Alignment          == right.Alignment          &&
     left.Width              == right.Width              &&
     left.Height             == right.Height             &&
     left.DepthOrArraySize   == right.DepthOrArraySize   &&
@@ -963,7 +975,7 @@ bool CSwapChainProcessor::EnsureCandidateResource(
   FrameCandidate& candidate = m_candidates[candidateIndex];
   D3D12_RESOURCE_DESC desc  = source->GetDesc();
   desc.Alignment = 0;
-  desc.Flags = static_cast<D3D12_RESOURCE_FLAGS>(
+  desc.Flags     = static_cast<D3D12_RESOURCE_FLAGS>(
     static_cast<UINT>(desc.Flags) &
       ~static_cast<UINT>(D3D12_RESOURCE_FLAG_ALLOW_CROSS_ADAPTER));
 
@@ -1599,6 +1611,9 @@ bool CSwapChainProcessor::SwapChainNewFrame(ComPtr<IDXGIResource> acquiredBuffer
     SetFullPendingDamage();
     return false;
   }
+  // Candidate and copy-slot acquisition are common to both benchmark modes.
+  const uint64_t timingStart = timingToken ?
+    CFrameScheduler::Nanotime() : 0;
 
   ComPtr<ID3D12Resource> copySrcResource = srcRes->GetRes();
   CD3D12CommandSlot    * computeSlot     = nullptr;
@@ -1677,23 +1692,24 @@ bool CSwapChainProcessor::SwapChainNewFrame(ComPtr<IDXGIResource> acquiredBuffer
     return false;
   }
 
-  candidate.srcFormat         = srcFormat;
+  candidate.srcFormat          = srcFormat;
-  candidate.dstFormat         = dstFormat;
+  candidate.dstFormat          = dstFormat;
-  candidate.nbDirtyRects      = nbDirtyRects;
+  candidate.nbDirtyRects       = nbDirtyRects;
-  candidate.pitch             = postProcessor.GetOutputPitch();
+  candidate.pitch              = postProcessor.GetOutputPitch();
-  candidate.frameSize         = postProcessor.GetOutputSize();
+  candidate.frameSize          = postProcessor.GetOutputSize();
-  candidate.captureTime       = captureTime;
+  candidate.captureTime        = captureTime;
-  candidate.postProcessStart  = postProcessStart;
+  candidate.postProcessStart   = postProcessStart;
-  candidate.prepareCopyStart  = CFrameScheduler::Nanotime();
+  candidate.prepareCopyStart   = CFrameScheduler::Nanotime();
-  candidate.prepareReady      = 0;
+  candidate.prepareReady       = 0;
-  candidate.prepareGPUStart   = 0;
+  candidate.prepareGPUStart    = 0;
-  candidate.prepareGPUEnd     = 0;
+  candidate.prepareGPUEnd      = 0;
+  candidate.timingStart        = timingStart;
   candidate.prepareTimingValid = false;
   if (nbDirtyRects)
     memcpy(candidate.dirtyRects, currentDirtyRects,
       nbDirtyRects * sizeof(*candidate.dirtyRects));
-  candidate.timingEffectIndex = timingEffectIndex;
+  candidate.timingEffectIndex  = timingEffectIndex;
-  candidate.timingToken       = timingToken;
+  candidate.timingToken        = timingToken;
 
   copySlot->SetCompletionCallback(
     &CandidateCompletionFunction, this, &candidate);

idd/LGIdd/CSwapChainProcessor.h

@@ -66,24 +66,25 @@ private:
 
   struct FrameCandidate
   {
-    CandidateState         state             = CANDIDATE_FREE;
+    CandidateState         state               = CANDIDATE_FREE;
     ComPtr<ID3D12Resource> resource;
-    D12FrameFormat         srcFormat          = {};
+    D12FrameFormat         srcFormat           = {};
-    D12FrameFormat         dstFormat          = {};
+    D12FrameFormat         dstFormat           = {};
     RECT                   dirtyRects[LG_MAX_DIRTY_RECTS] = {};
-    unsigned               nbDirtyRects      = 0;
+    unsigned               nbDirtyRects        = 0;
-    unsigned               pitch             = 0;
+    unsigned               pitch               = 0;
-    size_t                 frameSize         = 0;
+    size_t                 frameSize           = 0;
-    uint64_t               sequence          = 0;
+    uint64_t               sequence            = 0;
-    uint64_t               captureTime       = 0;
+    uint64_t               captureTime         = 0;
-    uint64_t               postProcessStart  = 0;
+    uint64_t               postProcessStart    = 0;
-    uint64_t               prepareCopyStart  = 0;
+    uint64_t               prepareCopyStart    = 0;
-    uint64_t               prepareReady      = 0;
+    uint64_t               prepareReady        = 0;
-    uint64_t               prepareGPUStart   = 0;
+    uint64_t               prepareGPUStart     = 0;
-    uint64_t               prepareGPUEnd     = 0;
+    uint64_t               prepareGPUEnd       = 0;
-    unsigned               timingEffectIndex = 0;
+    uint64_t               timingStart         = 0;
-    uint64_t               timingToken       = 0;
+    unsigned               timingEffectIndex   = 0;
-    bool                   prepareTimingValid = false;
+    uint64_t               timingToken         = 0;
+    bool                   prepareTimingValid  = false;
   };
 
   struct CandidateDamageTail