Skip to content

GPU: Some fixes #15931

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 3 commits into from
Oct 11, 2026
Merged

GPU: Some fixes #15931

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
9a2427e
GPU: Fix uninitialized memory
davidrohr Oct 11, 2026
60eccfc
GPU TPC Clusterizer: Fix endless loop when running with 1 thread
davidrohr Oct 11, 2026
e8c3b96
GPU TPC Clusterizer: Make CPU code compatible to processing scheme as…
davidrohr Oct 11, 2026
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/GPUProcessor.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -51,6 +51,7 @@ class GPUProcessor
GPUd() GPUconstantref() const GPUConstantMem* GetConstantMem() const; // Body in GPUConstantMem.h to avoid circular headers
GPUd() GPUconstantref() const GPUParam& Param() const; // ...
GPUd() void raiseError(uint32_t code, uint32_t param1 = 0, uint32_t param2 = 0, uint32_t param3 = 0) const;
GPUd() bool IsDeviceProcessor() const { return mGPUProcessorType == PROCESSOR_TYPE_DEVICE; }
const GPUReconstruction& GetRec() const { return *mRec; }

#ifndef __OPENCL__
Expand Down
1 change: 1 addition & 0 deletions GPU/GPUTracking/Merger/GPUTPCGMMerger.cxx
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -640,6 +640,7 @@ GPUd() void GPUTPCGMMerger::UnpackSectorGlobal(int32_t nBlocks, int32_t nThreads
SetTrackClusterT(track, iSector, sectorTr);
track.Set(this, sectorTr, alpha, iSector);
track.SetGlobalSectorTrackCov();
track.SetX2(0.f);
track.SetPrevNeighbour(-1);
track.SetNextNeighbour(-1);
track.SetNextSegmentNeighbour(-1);
Expand Down
22 changes: 14 additions & 8 deletions GPU/GPUTracking/TPCClusterFinder/GPUTPCCFDecodeZS.cxx
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -48,7 +48,8 @@ GPUdii() void GPUTPCCFDecodeZS::decode(GPUTPCClusterFinder& clusterer, GPUShared
#ifdef GPUCA_GPUCODE
const uint32_t endpoint = clusterer.mPzsOffsets[iBlock].endpoint;
#else
const uint32_t endpoint = iBlock;
const bool deviceLayout = clusterer.IsDeviceProcessor(); // Host-compiled kernel running with threading as on a device
const uint32_t endpoint = deviceLayout ? clusterer.mPzsOffsets[iBlock].endpoint : iBlock;
#endif
const GPUTrackingInOutZS::GPUTrackingInOutZSSector& zs = clusterer.GetConstantMem()->ioPtrs.tpcZS->sector[sector];
if (zs.count[endpoint] == 0) {
Expand All @@ -74,9 +75,11 @@ GPUdii() void GPUTPCCFDecodeZS::decode(GPUTPCClusterFinder& clusterer, GPUShared
{
{
#else
for (uint32_t i = clusterer.mMinMaxCN[endpoint].zsPtrFirst; i < clusterer.mMinMaxCN[endpoint].zsPtrLast; i++) {
const uint32_t minJ = (i == clusterer.mMinMaxCN[endpoint].zsPtrFirst) ? clusterer.mMinMaxCN[endpoint].zsPageFirst : 0;
const uint32_t maxJ = (i + 1 == clusterer.mMinMaxCN[endpoint].zsPtrLast) ? clusterer.mMinMaxCN[endpoint].zsPageLast : zs.nZSPtr[endpoint][i];
const uint32_t iFirst = deviceLayout ? 0 : clusterer.mMinMaxCN[endpoint].zsPtrFirst;
const uint32_t iLast = deviceLayout ? 1 : clusterer.mMinMaxCN[endpoint].zsPtrLast;
for (uint32_t i = iFirst; i < iLast; i++) {
const uint32_t minJ = deviceLayout ? clusterer.mPzsOffsets[iBlock].num : ((i == iFirst) ? clusterer.mMinMaxCN[endpoint].zsPageFirst : 0);
const uint32_t maxJ = deviceLayout ? (minJ + 1) : ((i + 1 == iLast) ? clusterer.mMinMaxCN[endpoint].zsPageLast : zs.nZSPtr[endpoint][i]);
for (uint32_t j = minJ; j < maxJ; j++) {
#endif
const uint32_t* pageSrc = (const uint32_t*)(((const uint8_t*)zs.zsPtr[endpoint][i]) + j * TPCZSHDR::TPC_ZS_PAGE_SIZE);
Expand Down Expand Up @@ -370,7 +373,8 @@ GPUd() void GPUTPCCFDecodeZSLinkBase::Decode(int32_t nBlocks, int32_t nThreads,
#ifdef GPUCA_GPUCODE
const uint32_t endpoint = clusterer.mPzsOffsets[iBlock].endpoint;
#else // CPU
const uint32_t endpoint = iBlock;
const bool deviceLayout = clusterer.IsDeviceProcessor(); // Host-compiled kernel running as device kernel (CPUDevice backend): GPU work distribution, one block per page
const uint32_t endpoint = deviceLayout ? clusterer.mPzsOffsets[iBlock].endpoint : iBlock;
#endif

const GPUTrackingInOutZS::GPUTrackingInOutZSSector& zs = clusterer.GetConstantMem()->ioPtrs.tpcZS->sector[sector];
Expand All @@ -386,9 +390,11 @@ GPUd() void GPUTPCCFDecodeZSLinkBase::Decode(int32_t nBlocks, int32_t nThreads,
{
{
#else // CPU
for (uint32_t i = clusterer.mMinMaxCN[endpoint].zsPtrFirst; i < clusterer.mMinMaxCN[endpoint].zsPtrLast; i++) {
const uint32_t minJ = (i == clusterer.mMinMaxCN[endpoint].zsPtrFirst) ? clusterer.mMinMaxCN[endpoint].zsPageFirst : 0;
const uint32_t maxJ = (i + 1 == clusterer.mMinMaxCN[endpoint].zsPtrLast) ? clusterer.mMinMaxCN[endpoint].zsPageLast : zs.nZSPtr[endpoint][i];
const uint32_t iFirst = deviceLayout ? 0 : clusterer.mMinMaxCN[endpoint].zsPtrFirst;
const uint32_t iLast = deviceLayout ? 1 : clusterer.mMinMaxCN[endpoint].zsPtrLast;
for (uint32_t i = iFirst; i < iLast; i++) {
const uint32_t minJ = deviceLayout ? clusterer.mPzsOffsets[iBlock].num : ((i == iFirst) ? clusterer.mMinMaxCN[endpoint].zsPageFirst : 0);
const uint32_t maxJ = deviceLayout ? (minJ + 1) : ((i + 1 == iLast) ? clusterer.mMinMaxCN[endpoint].zsPageLast : zs.nZSPtr[endpoint][i]);
for (uint32_t j = minJ; j < maxJ; j++) {
#endif
const uint32_t* pageSrc = (const uint32_t*)(((const uint8_t*)zs.zsPtr[endpoint][i]) + j * TPCZSHDR::TPC_ZS_PAGE_SIZE);
Expand Down
3 changes: 3 additions & 0 deletions GPU/GPUTracking/TPCClusterFinder/GPUTPCClusterFinder.cxx
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -161,6 +161,9 @@ uint32_t GPUTPCClusterFinder::getNSteps(size_t items) const
}
uint32_t c = 1;
const size_t scanWorkgroupSize = mRec->getGPUParameters(mRec->GetRecoStepsGPU() & gpudatatypes::RecoStep::TPCClusterFinding).par_CF_SCAN_WORKGROUP_SIZE;
if (scanWorkgroupSize <= 1) {
return 1;
}
size_t capacity = scanWorkgroupSize;
while (items > capacity) {
capacity *= scanWorkgroupSize;
Expand Down
Loading