Skip to content

Commit 85d185e

Browse files
ChSonnabendChSonnabend
committed
Some weird formatting
1 parent 34419f3 commit 85d185e

File tree

1 file changed

+126
-126
lines changed

1 file changed

+126
-126
lines changed

‎GPU/GPUTracking/Global/GPUChainTrackingClusterizer.cxx‎

Lines changed: 126 additions & 126 deletions
Original file line numberDiff line numberDiff line change
@@ -870,132 +870,132 @@ int32_t GPUChainTracking::RunTPCClusterizer(bool synchronizeOutput)
870870
GPUInfo("Processing time bins [%d, %d) for sectors %d to %d", fragment.start, fragment.last(), iSectorBase, iSectorBase + GetProcessingSettings().nTPCClustererLanes - 1);
871871
}
872872
mRec->runParallelOuterLoop(doGPU, maxLane, [&](uint32_t lane)
873-
{
874-
if (doGPU && fragment.index != 0)
875-
{
876-
SynchronizeStream(lane); // Don't overwrite charge map from previous iteration until cluster computation is finished
877-
}
878-
879-
uint32_t iSector = iSectorBase + lane;
880-
GPUTPCClusterFinder &clusterer = processors()->tpcClusterer[iSector];
881-
GPUTPCClusterFinder &clustererShadow = doGPU ? processorsShadow()->tpcClusterer[iSector] : clusterer;
882-
clusterer.mPmemory->counters.nPeaks = clusterer.mPmemory->counters.nClusters = 0;
883-
clusterer.mPmemory->fragment = fragment;
884-
885-
if (mIOPtrs.tpcPackedDigits)
886-
{
887-
bool setDigitsOnGPU = doGPU && not mIOPtrs.tpcZS;
888-
bool setDigitsOnHost = (not doGPU && not mIOPtrs.tpcZS) || propagateMCLabels;
889-
auto *inDigits = mIOPtrs.tpcPackedDigits;
890-
size_t numDigits = inDigits->nTPCDigits[iSector];
891-
if (setDigitsOnGPU)
892-
{
893-
GPUMemCpy(RecoStep::TPCClusterFinding, clustererShadow.mPdigits, inDigits->tpcDigits[iSector], sizeof(clustererShadow.mPdigits[0]) * numDigits, lane, true);
894-
}
895-
if (setDigitsOnHost)
896-
{
897-
clusterer.mPdigits = const_cast<o2::tpc::Digit *>(inDigits->tpcDigits[iSector]); // TODO: Needs fixing (invalid const cast)
898-
}
899-
clusterer.mPmemory->counters.nDigits = numDigits;
900-
}
901-
902-
if (mIOPtrs.tpcZS)
903-
{
904-
if (mCFContext->nPagesSector[iSector] && mCFContext->zsVersion != -1)
905-
{
906-
clusterer.mPmemory->counters.nPositions = mCFContext->nextPos[iSector].first;
907-
clusterer.mPmemory->counters.nPagesSubsector = mCFContext->nextPos[iSector].second;
908-
}
909-
else
910-
{
911-
clusterer.mPmemory->counters.nPositions = clusterer.mPmemory->counters.nPagesSubsector = 0;
912-
}
913-
}
914-
TransferMemoryResourceLinkToGPU(RecoStep::TPCClusterFinding, clusterer.mMemoryId, lane);
915-
916-
using ChargeMapType = decltype(*clustererShadow.mPchargeMap);
917-
using PeakMapType = decltype(*clustererShadow.mPpeakMap);
918-
runKernel<GPUMemClean16>({GetGridAutoStep(lane, RecoStep::TPCClusterFinding)}, clustererShadow.mPchargeMap, TPCMapMemoryLayout<ChargeMapType>::items(GetProcessingSettings().overrideClusterizerFragmentLen) * sizeof(ChargeMapType)); // TODO: Not working in OpenCL2!!!
919-
runKernel<GPUMemClean16>({GetGridAutoStep(lane, RecoStep::TPCClusterFinding)}, clustererShadow.mPpeakMap, TPCMapMemoryLayout<PeakMapType>::items(GetProcessingSettings().overrideClusterizerFragmentLen) * sizeof(PeakMapType));
920-
if (fragment.index == 0)
921-
{
922-
runKernel<GPUMemClean16>({GetGridAutoStep(lane, RecoStep::TPCClusterFinding)}, clustererShadow.mPpadIsNoisy, TPC_PADS_IN_SECTOR * sizeof(*clustererShadow.mPpadIsNoisy));
923-
}
924-
DoDebugAndDump(RecoStep::TPCClusterFinding, 262144, clusterer, &GPUTPCClusterFinder::DumpChargeMap, *mDebugFile, "Zeroed Charges");
925-
926-
if (doGPU)
927-
{
928-
if (mIOPtrs.tpcZS && mCFContext->nPagesSector[iSector] && mCFContext->zsVersion != -1)
929-
{
930-
TransferMemoryResourceLinkToGPU(RecoStep::TPCClusterFinding, mInputsHost->mResourceZS, lane);
931-
SynchronizeStream(GetProcessingSettings().nTPCClustererLanes + lane);
932-
}
933-
SynchronizeStream(mRec->NStreams() - 1); // Wait for copying to constant memory
934-
}
935-
936-
if (mIOPtrs.tpcZS && (mCFContext->abandonTimeframe || !mCFContext->nPagesSector[iSector] || mCFContext->zsVersion == -1))
937-
{
938-
clusterer.mPmemory->counters.nPositions = 0;
939-
return;
940-
}
941-
if (!mIOPtrs.tpcZS && mIOPtrs.tpcPackedDigits->nTPCDigits[iSector] == 0)
942-
{
943-
clusterer.mPmemory->counters.nPositions = 0;
944-
return;
945-
}
946-
947-
if (propagateMCLabels && fragment.index == 0)
948-
{
949-
clusterer.PrepareMC();
950-
clusterer.mPinputLabels = digitsMC->v[iSector];
951-
if (clusterer.mPinputLabels == nullptr)
952-
{
953-
GPUFatal("MC label container missing, sector %d", iSector);
954-
}
955-
if (clusterer.mPinputLabels->getIndexedSize() != mIOPtrs.tpcPackedDigits->nTPCDigits[iSector])
956-
{
957-
GPUFatal("MC label container has incorrect number of entries: %d expected, has %d\n", (int32_t)mIOPtrs.tpcPackedDigits->nTPCDigits[iSector], (int32_t)clusterer.mPinputLabels->getIndexedSize());
958-
}
959-
}
960-
961-
if (GetProcessingSettings().tpcSingleSector == -1 || GetProcessingSettings().tpcSingleSector == (int32_t)iSector)
962-
{
963-
if (not mIOPtrs.tpcZS)
964-
{
965-
runKernel<GPUTPCCFChargeMapFiller, GPUTPCCFChargeMapFiller::findFragmentStart>({GetGrid(1, lane), {iSector}}, mIOPtrs.tpcZS == nullptr);
966-
TransferMemoryResourceLinkToHost(RecoStep::TPCClusterFinding, clusterer.mMemoryId, lane);
967-
}
968-
else if (propagateMCLabels)
969-
{
970-
runKernel<GPUTPCCFChargeMapFiller, GPUTPCCFChargeMapFiller::findFragmentStart>({GetGrid(1, lane, GPUReconstruction::krnlDeviceType::CPU), {iSector}}, mIOPtrs.tpcZS == nullptr);
971-
TransferMemoryResourceLinkToGPU(RecoStep::TPCClusterFinding, clusterer.mMemoryId, lane);
972-
}
973-
}
974-
975-
if (mIOPtrs.tpcZS)
976-
{
977-
int32_t firstHBF = (mIOPtrs.settingsTF && mIOPtrs.settingsTF->hasTfStartOrbit) ? mIOPtrs.settingsTF->tfStartOrbit : ((mIOPtrs.tpcZS->sector[iSector].count[0] && mIOPtrs.tpcZS->sector[iSector].nZSPtr[0][0]) ? o2::raw::RDHUtils::getHeartBeatOrbit(*(const o2::header::RAWDataHeader *)mIOPtrs.tpcZS->sector[iSector].zsPtr[0][0]) : 0);
978-
uint32_t nBlocks = doGPU ? clusterer.mPmemory->counters.nPagesSubsector : GPUTrackingInOutZS::NENDPOINTS;
979-
980-
(void)tpcTimeBinCut; // TODO: To be used in decoding kernels
981-
switch (mCFContext->zsVersion)
982-
{
983-
default:
984-
GPUFatal("Data with invalid TPC ZS mode (%d) received", mCFContext->zsVersion);
985-
break;
986-
case ZSVersionRowBased10BitADC:
987-
case ZSVersionRowBased12BitADC:
988-
runKernel<GPUTPCCFDecodeZS>({GetGridBlk(nBlocks, lane), {iSector}}, firstHBF);
989-
break;
990-
case ZSVersionLinkBasedWithMeta:
991-
runKernel<GPUTPCCFDecodeZSLink>({GetGridBlk(nBlocks, lane), {iSector}}, firstHBF);
992-
break;
993-
case ZSVersionDenseLinkBased:
994-
runKernel<GPUTPCCFDecodeZSDenseLink>({GetGridBlk(nBlocks, lane), {iSector}}, firstHBF);
995-
break;
996-
}
997-
TransferMemoryResourceLinkToHost(RecoStep::TPCClusterFinding, clusterer.mMemoryId, lane);
998-
} // clang-format off
873+
{
874+
if (doGPU && fragment.index != 0)
875+
{
876+
SynchronizeStream(lane); // Don't overwrite charge map from previous iteration until cluster computation is finished
877+
}
878+
879+
uint32_t iSector = iSectorBase + lane;
880+
GPUTPCClusterFinder &clusterer = processors()->tpcClusterer[iSector];
881+
GPUTPCClusterFinder &clustererShadow = doGPU ? processorsShadow()->tpcClusterer[iSector] : clusterer;
882+
clusterer.mPmemory->counters.nPeaks = clusterer.mPmemory->counters.nClusters = 0;
883+
clusterer.mPmemory->fragment = fragment;
884+
885+
if (mIOPtrs.tpcPackedDigits)
886+
{
887+
bool setDigitsOnGPU = doGPU && not mIOPtrs.tpcZS;
888+
bool setDigitsOnHost = (not doGPU && not mIOPtrs.tpcZS) || propagateMCLabels;
889+
auto *inDigits = mIOPtrs.tpcPackedDigits;
890+
size_t numDigits = inDigits->nTPCDigits[iSector];
891+
if (setDigitsOnGPU)
892+
{
893+
GPUMemCpy(RecoStep::TPCClusterFinding, clustererShadow.mPdigits, inDigits->tpcDigits[iSector], sizeof(clustererShadow.mPdigits[0]) * numDigits, lane, true);
894+
}
895+
if (setDigitsOnHost)
896+
{
897+
clusterer.mPdigits = const_cast<o2::tpc::Digit *>(inDigits->tpcDigits[iSector]); // TODO: Needs fixing (invalid const cast)
898+
}
899+
clusterer.mPmemory->counters.nDigits = numDigits;
900+
}
901+
902+
if (mIOPtrs.tpcZS)
903+
{
904+
if (mCFContext->nPagesSector[iSector] && mCFContext->zsVersion != -1)
905+
{
906+
clusterer.mPmemory->counters.nPositions = mCFContext->nextPos[iSector].first;
907+
clusterer.mPmemory->counters.nPagesSubsector = mCFContext->nextPos[iSector].second;
908+
}
909+
else
910+
{
911+
clusterer.mPmemory->counters.nPositions = clusterer.mPmemory->counters.nPagesSubsector = 0;
912+
}
913+
}
914+
TransferMemoryResourceLinkToGPU(RecoStep::TPCClusterFinding, clusterer.mMemoryId, lane);
915+
916+
using ChargeMapType = decltype(*clustererShadow.mPchargeMap);
917+
using PeakMapType = decltype(*clustererShadow.mPpeakMap);
918+
runKernel<GPUMemClean16>({GetGridAutoStep(lane, RecoStep::TPCClusterFinding)}, clustererShadow.mPchargeMap, TPCMapMemoryLayout<ChargeMapType>::items(GetProcessingSettings().overrideClusterizerFragmentLen) * sizeof(ChargeMapType)); // TODO: Not working in OpenCL2!!!
919+
runKernel<GPUMemClean16>({GetGridAutoStep(lane, RecoStep::TPCClusterFinding)}, clustererShadow.mPpeakMap, TPCMapMemoryLayout<PeakMapType>::items(GetProcessingSettings().overrideClusterizerFragmentLen) * sizeof(PeakMapType));
920+
if (fragment.index == 0)
921+
{
922+
runKernel<GPUMemClean16>({GetGridAutoStep(lane, RecoStep::TPCClusterFinding)}, clustererShadow.mPpadIsNoisy, TPC_PADS_IN_SECTOR * sizeof(*clustererShadow.mPpadIsNoisy));
923+
}
924+
DoDebugAndDump(RecoStep::TPCClusterFinding, 262144, clusterer, &GPUTPCClusterFinder::DumpChargeMap, *mDebugFile, "Zeroed Charges");
925+
926+
if (doGPU)
927+
{
928+
if (mIOPtrs.tpcZS && mCFContext->nPagesSector[iSector] && mCFContext->zsVersion != -1)
929+
{
930+
TransferMemoryResourceLinkToGPU(RecoStep::TPCClusterFinding, mInputsHost->mResourceZS, lane);
931+
SynchronizeStream(GetProcessingSettings().nTPCClustererLanes + lane);
932+
}
933+
SynchronizeStream(mRec->NStreams() - 1); // Wait for copying to constant memory
934+
}
935+
936+
if (mIOPtrs.tpcZS && (mCFContext->abandonTimeframe || !mCFContext->nPagesSector[iSector] || mCFContext->zsVersion == -1))
937+
{
938+
clusterer.mPmemory->counters.nPositions = 0;
939+
return;
940+
}
941+
if (!mIOPtrs.tpcZS && mIOPtrs.tpcPackedDigits->nTPCDigits[iSector] == 0)
942+
{
943+
clusterer.mPmemory->counters.nPositions = 0;
944+
return;
945+
}
946+
947+
if (propagateMCLabels && fragment.index == 0)
948+
{
949+
clusterer.PrepareMC();
950+
clusterer.mPinputLabels = digitsMC->v[iSector];
951+
if (clusterer.mPinputLabels == nullptr)
952+
{
953+
GPUFatal("MC label container missing, sector %d", iSector);
954+
}
955+
if (clusterer.mPinputLabels->getIndexedSize() != mIOPtrs.tpcPackedDigits->nTPCDigits[iSector])
956+
{
957+
GPUFatal("MC label container has incorrect number of entries: %d expected, has %d\n", (int32_t)mIOPtrs.tpcPackedDigits->nTPCDigits[iSector], (int32_t)clusterer.mPinputLabels->getIndexedSize());
958+
}
959+
}
960+
961+
if (GetProcessingSettings().tpcSingleSector == -1 || GetProcessingSettings().tpcSingleSector == (int32_t)iSector)
962+
{
963+
if (not mIOPtrs.tpcZS)
964+
{
965+
runKernel<GPUTPCCFChargeMapFiller, GPUTPCCFChargeMapFiller::findFragmentStart>({GetGrid(1, lane), {iSector}}, mIOPtrs.tpcZS == nullptr);
966+
TransferMemoryResourceLinkToHost(RecoStep::TPCClusterFinding, clusterer.mMemoryId, lane);
967+
}
968+
else if (propagateMCLabels)
969+
{
970+
runKernel<GPUTPCCFChargeMapFiller, GPUTPCCFChargeMapFiller::findFragmentStart>({GetGrid(1, lane, GPUReconstruction::krnlDeviceType::CPU), {iSector}}, mIOPtrs.tpcZS == nullptr);
971+
TransferMemoryResourceLinkToGPU(RecoStep::TPCClusterFinding, clusterer.mMemoryId, lane);
972+
}
973+
}
974+
975+
if (mIOPtrs.tpcZS)
976+
{
977+
int32_t firstHBF = (mIOPtrs.settingsTF && mIOPtrs.settingsTF->hasTfStartOrbit) ? mIOPtrs.settingsTF->tfStartOrbit : ((mIOPtrs.tpcZS->sector[iSector].count[0] && mIOPtrs.tpcZS->sector[iSector].nZSPtr[0][0]) ? o2::raw::RDHUtils::getHeartBeatOrbit(*(const o2::header::RAWDataHeader *)mIOPtrs.tpcZS->sector[iSector].zsPtr[0][0]) : 0);
978+
uint32_t nBlocks = doGPU ? clusterer.mPmemory->counters.nPagesSubsector : GPUTrackingInOutZS::NENDPOINTS;
979+
980+
(void)tpcTimeBinCut; // TODO: To be used in decoding kernels
981+
switch (mCFContext->zsVersion)
982+
{
983+
default:
984+
GPUFatal("Data with invalid TPC ZS mode (%d) received", mCFContext->zsVersion);
985+
break;
986+
case ZSVersionRowBased10BitADC:
987+
case ZSVersionRowBased12BitADC:
988+
runKernel<GPUTPCCFDecodeZS>({GetGridBlk(nBlocks, lane), {iSector}}, firstHBF);
989+
break;
990+
case ZSVersionLinkBasedWithMeta:
991+
runKernel<GPUTPCCFDecodeZSLink>({GetGridBlk(nBlocks, lane), {iSector}}, firstHBF);
992+
break;
993+
case ZSVersionDenseLinkBased:
994+
runKernel<GPUTPCCFDecodeZSDenseLink>({GetGridBlk(nBlocks, lane), {iSector}}, firstHBF);
995+
break;
996+
}
997+
TransferMemoryResourceLinkToHost(RecoStep::TPCClusterFinding, clusterer.mMemoryId, lane);
998+
} // clang-format off
999999
});
10001000
mRec->runParallelOuterLoop(doGPU, maxLane, [&](uint32_t lane) {
10011001
uint32_t iSector = iSectorBase + lane;

0 commit comments

Comments
 (0)