Mirrored_Repos/LookingGlass
1
0
Fork 0
mirror of https://github.com/gnif/LookingGlass.git synced 2026-08-09 00:31:31 +00:00
Code Issues Actions 2 Packages Projects Releases Wiki Activity

[idd] rgb24: compare stable candidate timings

Browse Source 
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.
This commit is contained in:
Geoffrey McRae
2026-08-05 21:45:08 +10:00
parent f073099950
commit ab25601a61
2 changed files with 53 additions and 36 deletions
Download Patch File Download Diff File Expand all files Collapse all files

26
idd/LGIdd/CSwapChainProcessor.cpp
View File

@@ -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
// to IVSHMEM in effect benchmark samples without charging it to Ready.
// GPU timestamps end at the readback copy. Track the following CPU copy
// 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;
// Use matching wall-clock boundaries for both modes. The split excludes the
// cadence hold while including the indirect CPU copy only when it occurs.
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(), fbRes->GetTimingToken(),
fbRes->IsFullCopy(), postProcessTime + copyTime);
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 &&
@@ -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;
@@ -1688,6 +1703,7 @@ bool CSwapChainProcessor::SwapChainNewFrame(ComPtr<IDXGIResource> acquiredBuffer
candidate.prepareReady = 0;
candidate.prepareGPUStart = 0;
candidate.prepareGPUEnd = 0;
candidate.timingStart = timingStart;
candidate.prepareTimingValid = false;
if (nbDirtyRects)
memcpy(candidate.dirtyRects, currentDirtyRects,

1
idd/LGIdd/CSwapChainProcessor.h
View File

@@ -81,6 +81,7 @@ private:
uint64_t prepareReady = 0;
uint64_t prepareGPUStart = 0;
uint64_t prepareGPUEnd = 0;
uint64_t timingStart = 0;
unsigned timingEffectIndex = 0;
uint64_t timingToken = 0;
bool prepareTimingValid = false;