Skip to content

GPU TPC Clusterizer: fix memory overlap between scratch memory and clusters leading to race condition #16103

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
davidrohr merged 1 commit into from
Nov 11, 2026
+11 −2
Merged

GPU TPC Clusterizer: fix memory overlap between scratch memory and clusters leading to race condition #16103

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
1 change: 1 addition & 0 deletions GPU/GPUTracking/Base/GPUMemoryResource.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -31,6 +31,7 @@ struct GPUMemoryReuse {
};
enum Group : uint16_t {
ClustererScratch,
ClustererScratchClusters,
NNClusterer,
ClustererZS,
TrackerScratch,
Expand Down
3 changes: 2 additions & 1 deletion GPU/GPUTracking/Global/GPUChainTrackingClusterizer.cxx
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1367,6 +1367,7 @@ int32_t GPUChainTracking::RunTPCClusterizer(bool synchronizeOutput)

if (doGPU && propagateMCLabels) {
TransferMemoryResourceLinkToHost(RecoStep::TPCClusterFinding, clusterer.mScratchId, lane);
TransferMemoryResourceLinkToHost(RecoStep::TPCClusterFinding, clusterer.mScratchClustersId, lane);
SynchronizeStream(lane);
runKernel<GPUTPCCFClusterizer>({GetGrid(clusterer.mPmemory->counters.nClusters, lane, GPUReconstruction::krnlDeviceType::CPU), {iSector}}, 1); // Computes MC labels
}
Expand All @@ -1382,7 +1383,7 @@ int32_t GPUChainTracking::RunTPCClusterizer(bool synchronizeOutput)
SynchronizeStream(lane);
}
if (doGPU && propagateMCLabels) {
TransferMemoryResourceLinkToHost(RecoStep::TPCClusterFinding, clusterer.mScratchId, lane);
TransferMemoryResourceLinkToHost(RecoStep::TPCClusterFinding, clusterer.mScratchClustersId, lane);
SynchronizeStream(lane);
runKernel<GPUTPCCFHIPClusterizer>({GetGrid(GPUTPCGeometry::NROWS, lane, GPUReconstruction::krnlDeviceType::CPU), {iSector}}, 1); // Computes MC labels
}
Expand Down
7 changes: 6 additions & 1 deletion GPU/GPUTracking/TPCClusterFinder/GPUTPCClusterFinder.cxx
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -96,10 +96,14 @@ void* GPUTPCClusterFinder::SetPointersScratch(void* mem)
computePointerWithAlignment(mem, mPisPeak, mNMaxDigitsFragment);
computePointerWithAlignment(mem, mPchargeMap, TPCMapMemoryLayout<decltype(*mPchargeMap)>::items(mRec->GetProcessingSettings().overrideClusterizerFragmentLen));
computePointerWithAlignment(mem, mPpeakMap, TPCMapMemoryLayout<decltype(*mPpeakMap)>::items(mRec->GetProcessingSettings().overrideClusterizerFragmentLen));
computePointerWithAlignment(mem, mPclusterByRow, GPUTPCGeometry::NROWS * mNMaxClusterPerRow);
if ((mRec->GetRecoStepsGPU() & gpudatatypes::RecoStep::TPCClusterFinding)) {
computePointerWithAlignment(mem, mPscanBuf, mBufSize * mNBufs);
}
return mem;
}
void* GPUTPCClusterFinder::SetPointersScratchClusters(void* mem)
{
computePointerWithAlignment(mem, mPclusterByRow, GPUTPCGeometry::NROWS * mNMaxClusterPerRow);
// TODO: Use memory scalers for MaxHIPTails.
// NOTE: Always allocate since Param() is not available during size computation.
computePointerWithAlignment(mem, mPhipTailsByRow, GPUTPCGeometry::NROWS * GPUTPCCFHIPClusterizer::MaxHIPTailsPerRow);
Expand All @@ -117,6 +121,7 @@ void GPUTPCClusterFinder::RegisterMemoryAllocation()
scratchType |= GPUMemoryResource::MEMORY_HOST | GPUMemoryResource::MEMORY_GPU;
}
mScratchId = mRec->RegisterMemoryAllocation(this, &GPUTPCClusterFinder::SetPointersScratch, scratchType, "TPCClustererScratch", GPUMemoryReuse{GPUMemoryReuse::REUSE_1TO1, GPUMemoryReuse::ClustererScratch, (uint16_t)(mISector % mRec->GetProcessingSettings().nTPCClustererLanes)});
mScratchClustersId = mRec->RegisterMemoryAllocation(this, &GPUTPCClusterFinder::SetPointersScratchClusters, scratchType, "TPCClustererScratchClusters", GPUMemoryReuse{GPUMemoryReuse::REUSE_1TO1, GPUMemoryReuse::ClustererScratchClusters, (uint16_t)(mISector % mRec->GetProcessingSettings().nTPCClustererLanes)});

mMemoryId = mRec->RegisterMemoryAllocation(this, &GPUTPCClusterFinder::SetPointersMemory, GPUMemoryResource::MEMORY_PERMANENT, "TPCClustererMemory");
mRec->RegisterMemoryAllocation(this, &GPUTPCClusterFinder::SetPointersOutput, GPUMemoryResource::MEMORY_OUTPUT | GPUMemoryResource::MEMORY_STACK, "TPCClustererOutput");
Expand Down
2 changes: 2 additions & 0 deletions GPU/GPUTracking/TPCClusterFinder/GPUTPCClusterFinder.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -85,6 +85,7 @@ class GPUTPCClusterFinder : public GPUProcessor
void* SetPointersInput(void* mem);
void* SetPointersOutput(void* mem);
void* SetPointersScratch(void* mem);
void* SetPointersScratchClusters(void* mem);
void* SetPointersMemory(void* mem);
void* SetPointersZS(void* mem);
void* SetPointersZSOffset(void* mem);
Expand Down Expand Up @@ -140,6 +141,7 @@ class GPUTPCClusterFinder : public GPUProcessor

int16_t mMemoryId = -1;
int16_t mScratchId = -1;
int16_t mScratchClustersId = -1;
int16_t mZSId = -1;
int16_t mZSOffsetId = -1;
int16_t mOutputId = -1;
Expand Down
Loading