GPU: Fix deterministic mode for cluster compression/decompression with looper merger afterburner + improved debug and cleanup

GPU/GPUTracking/Base/cuda/GPUReconstructionCUDAKernelsSpecialize.inc

@@ -14,88 +14,6 @@
 
 #if defined(GPUCA_SPECIALIZE_THRUST_SORTS) && !defined(GPUCA_GPUCODE_COMPILEKERNELS)
 
-namespace o2::gpu::internal
-{
-namespace // anonymous
-{
-struct MergeBorderTracks_compMax {
-  GPUd() bool operator()(const GPUTPCGMBorderRange& a, const GPUTPCGMBorderRange& b)
-  {
-    return GPUCA_DETERMINISTIC_CODE((a.fMax != b.fMax) ? (a.fMax < b.fMax) : (a.fId < b.fId), a.fMax < b.fMax);
-  }
-};
-struct MergeBorderTracks_compMin {
-  GPUd() bool operator()(const GPUTPCGMBorderRange& a, const GPUTPCGMBorderRange& b)
-  {
-    return GPUCA_DETERMINISTIC_CODE((a.fMin != b.fMin) ? (a.fMin < b.fMin) : (a.fId < b.fId), a.fMin < b.fMin);
-  }
-};
-
-struct GPUTPCGMMergerSortTracks_comp {
-  const GPUTPCGMMergedTrack* const mCmp;
-  GPUhd() GPUTPCGMMergerSortTracks_comp(GPUTPCGMMergedTrack* cmp) : mCmp(cmp) {}
-  GPUd() bool operator()(const int32_t aa, const int32_t bb)
-  {
-    const GPUTPCGMMergedTrack& GPUrestrict() a = mCmp[aa];
-    const GPUTPCGMMergedTrack& GPUrestrict() b = mCmp[bb];
-    if (a.CCE() != b.CCE()) {
-      return a.CCE() > b.CCE();
-    }
-    if (a.Legs() != b.Legs()) {
-      return a.Legs() > b.Legs();
-    }
-    GPUCA_DETERMINISTIC_CODE( // clang-format off
-      if (a.NClusters() != b.NClusters()) {
-        return a.NClusters() > b.NClusters();
-      } if (CAMath::Abs(a.GetParam().GetQPt()) != CAMath::Abs(b.GetParam().GetQPt())) {
-        return CAMath::Abs(a.GetParam().GetQPt()) > CAMath::Abs(b.GetParam().GetQPt());
-      } if (a.GetParam().GetY() != b.GetParam().GetY()) {
-        return a.GetParam().GetY() > b.GetParam().GetY();
-      }
-      return aa > bb;
-    , // !GPUCA_DETERMINISTIC_CODE
-      return a.NClusters() > b.NClusters();
-    ) // clang-format on
-  }
-};
-
-struct GPUTPCGMMergerSortTracksQPt_comp {
-  const GPUTPCGMMergedTrack* const mCmp;
-  GPUhd() GPUTPCGMMergerSortTracksQPt_comp(GPUTPCGMMergedTrack* cmp) : mCmp(cmp) {}
-  GPUd() bool operator()(const int32_t aa, const int32_t bb)
-  {
-    const GPUTPCGMMergedTrack& GPUrestrict() a = mCmp[aa];
-    const GPUTPCGMMergedTrack& GPUrestrict() b = mCmp[bb];
-    GPUCA_DETERMINISTIC_CODE( // clang-format off
-      if (CAMath::Abs(a.GetParam().GetQPt()) != CAMath::Abs(b.GetParam().GetQPt())) {
-        return CAMath::Abs(a.GetParam().GetQPt()) > CAMath::Abs(b.GetParam().GetQPt());
-      } if (a.GetParam().GetY() != b.GetParam().GetY()) {
-        return a.GetParam().GetY() > b.GetParam().GetY();
-      }
-      return a.GetParam().GetZ() > b.GetParam().GetZ();
-    , // !GPUCA_DETERMINISTIC_CODE
-      return CAMath::Abs(a.GetParam().GetQPt()) > CAMath::Abs(b.GetParam().GetQPt());
-    ) // clang-format on
-  }
-};
-
-struct GPUTPCGMMergerMergeLoopers_comp {
-  GPUd() bool operator()(const MergeLooperParam& a, const MergeLooperParam& b)
-  {
-    return CAMath::Abs(a.refz) < CAMath::Abs(b.refz);
-  }
-};
-
-struct GPUTPCGMO2OutputSort_comp {
-  GPUd() bool operator()(const GPUTPCGMMerger::tmpSort& a, const GPUTPCGMMerger::tmpSort& b)
-  {
-    return (a.y > b.y);
-  }
-};
-
-} // anonymous namespace
-} // namespace o2::gpu::internal
-
 template <>
 inline void GPUCA_M_CAT(GPUReconstruction, GPUCA_GPUTYPE)::runKernelBackendTimed<GPUTPCGMMergerMergeBorders, 3>(const krnlSetupTime& _xyz, GPUTPCGMBorderRange* const& range, int32_t const& N, int32_t const& cmpMax)
 {
@@ -109,13 +27,13 @@ inline void GPUCA_M_CAT(GPUReconstruction, GPUCA_GPUTYPE)::runKernelBackendTimed
 template <>
 inline void GPUCA_M_CAT(GPUReconstruction, GPUCA_GPUTYPE)::runKernelBackendTimed<GPUTPCGMMergerSortTracks, 0>(const krnlSetupTime& _xyz)
 {
-  GPUCommonAlgorithm::sortOnDevice(this, _xyz.x.stream, mProcessorsShadow->tpcMerger.TrackOrderProcess(), processors()->tpcMerger.NOutputTracks(), GPUTPCGMMergerSortTracks_comp(mProcessorsShadow->tpcMerger.OutputTracks()));
+  GPUCommonAlgorithm::sortOnDevice(this, _xyz.x.stream, mProcessorsShadow->tpcMerger.TrackOrderProcess(), processors()->tpcMerger.NMergedTracks(), GPUTPCGMMergerSortTracks_comp(mProcessorsShadow->tpcMerger.MergedTracks()));
 }
 
 template <>
 inline void GPUCA_M_CAT(GPUReconstruction, GPUCA_GPUTYPE)::runKernelBackendTimed<GPUTPCGMMergerSortTracksQPt, 0>(const krnlSetupTime& _xyz)
 {
-  GPUCommonAlgorithm::sortOnDevice(this, _xyz.x.stream, mProcessorsShadow->tpcMerger.TrackSort(), processors()->tpcMerger.NOutputTracks(), GPUTPCGMMergerSortTracksQPt_comp(mProcessorsShadow->tpcMerger.OutputTracks()));
+  GPUCommonAlgorithm::sortOnDevice(this, _xyz.x.stream, mProcessorsShadow->tpcMerger.TrackSort(), processors()->tpcMerger.NMergedTracks(), GPUTPCGMMergerSortTracksQPt_comp(mProcessorsShadow->tpcMerger.MergedTracks()));
 }
 
 template <>

GPU/GPUTracking/DataCompression/GPUTPCCompression.cxx

@@ -125,7 +125,7 @@ void GPUTPCCompression::SetMaxData(const GPUTrackingInOutPointers& io)
   mMaxClusterFactorBase1024 = mMaxClusters > 100000000 ? mRec->MemoryScalers()->NTPCUnattachedHitsBase1024(mRec->GetParam().rec.tpc.rejectionStrategy) : 1024;
   mMaxClustersInCache = mMaxClusters * mMaxClusterFactorBase1024 / 1024;
   mMaxTrackClusters = mRec->GetConstantMem().tpcMerger.NOutputTrackClusters(); // TODO: Why is this not using ioPtrs? Could remove GPUConstantMem.h include
-  mMaxTracks = mRec->GetConstantMem().tpcMerger.NOutputTracks();
+  mMaxTracks = mRec->GetConstantMem().tpcMerger.NMergedTracks();
   if (mMaxClusters % 16) {
     mMaxClusters += 16 - (mMaxClusters % 16);
   }

GPU/GPUTracking/DataCompression/GPUTPCCompressionKernels.cxx

@@ -201,7 +201,7 @@ GPUdii() void GPUTPCCompressionKernels::Thread<GPUTPCCompressionKernels::step1un
     const uint32_t iSector = iSectorRow / GPUCA_ROW_COUNT;
     const uint32_t iRow = iSectorRow % GPUCA_ROW_COUNT;
     const uint32_t idOffset = clusters->clusterOffset[iSector][iRow];
-    const uint32_t idOffsetOut = clusters->clusterOffset[iSector][iRow] * compressor.mMaxClusterFactorBase1024 / 1024;
+    const uint32_t idOffsetOut = clusters->clusterOffset[iSector][iRow] * compressor.mMaxClusterFactorBase1024 / 1024;                           // 32 bit enough for number of clusters per row * 1024
     const uint32_t idOffsetOutMax = ((const uint32_t*)clusters->clusterOffset[iSector])[iRow + 1] * compressor.mMaxClusterFactorBase1024 / 1024; // Array out of bounds access is ok, since it goes to the correct nClustersTotal
     if (iThread == nThreads - 1) {
       smem.nCount = 0;
@@ -214,7 +214,7 @@ GPUdii() void GPUTPCCompressionKernels::Thread<GPUTPCCompressionKernels::step1un
     const uint32_t nn = CAMath::nextMultipleOf<GPUCA_GET_THREAD_COUNT(GPUCA_LB_GPUTPCCompressionKernels_step1unattached)>(clusters->nClusters[iSector][iRow]);
     for (uint32_t i = iThread; i < nn + nThreads; i += nThreads) {
       const int32_t idx = idOffset + i;
-      int32_t cidx = 0;
+      int32_t storeCluster = 0;
       do {
         if (i >= clusters->nClusters[iSector][iRow]) {
           break;
@@ -239,13 +239,13 @@ GPUdii() void GPUTPCCompressionKernels::Thread<GPUTPCCompressionKernels::step1un
             break;
           }
         }
-        cidx = 1;
+        storeCluster = 1;
       } while (false);
 
       GPUbarrier();
-      int32_t myIndex = work_group_scan_inclusive_add(cidx);
+      int32_t myIndex = work_group_scan_inclusive_add(storeCluster);
       int32_t storeLater = -1;
-      if (cidx) {
+      if (storeCluster) {
         if (smem.nCount + myIndex <= GPUCA_TPC_COMP_CHUNK_SIZE) {
           sortBuffer[smem.nCount + myIndex - 1] = i;
         } else {

GPU/GPUTracking/Definitions/GPUSettingsList.h

@@ -295,7 +295,7 @@ AddOption(trdNCandidates, int32_t, 3, "", 0, "Number of branching track candidat
 AddOption(trdTrackModelO2, bool, false, "", 0, "Use O2 track model instead of GPU track model for TRD tracking")
 AddOption(debugLevel, int32_t, -1, "debug", 'd', "Set debug level (-2 = silent, -1 = autoselect (-2 for O2, 0 for standalone))")
 AddOption(allocDebugLevel, int32_t, 0, "allocDebug", 0, "Some debug output for memory allocations (without messing with normal debug level)")
-AddOption(debugMask, uint32_t, 262143, "", 0, "Mask for debug output dumps to file")
+AddOption(debugMask, uint32_t, (1 << 18) - 1, "debugMask", 0, "Mask for debug output dumps to file")
 AddOption(debugLogSuffix, std::string, "", "debugSuffix", 0, "Suffix for debug log files with --debug 6")
 AddOption(serializeGPU, int8_t, 0, "", 0, "Synchronize after each kernel call (bit 1) and DMA transfer (bit 2) and identify failures")
 AddOption(recoTaskTiming, bool, 0, "", 0, "Perform summary timing after whole reconstruction tasks")

GPU/GPUTracking/Global/GPUChainTrackingDebug.h

@@ -23,28 +23,29 @@ namespace o2::gpu
 {
 // NOTE: Values below 262144 are activated by default with --debug 6 in GPUSettingsList.h::debugMask
 enum GPUChainTrackingDebugFlags : uint32_t {
-  TPCSectorTrackingData = 1,
-  TPCPreLinks = 2,
-  TPCLinks = 4,
-  TPCStartHits = 8,
-  TPCTracklets = 16,
-  TPCSectorTracks = 32,
-  TPCHitWeights = 256,
-  TPCCompressedClusters = 512,
-  TPCDecompressedClusters = 1024,
-  TPCMergingRanges = 2048,
-  TPCMergingSectorTracks = 4096,
-  TPCMergingMergedTracks = 8192,
-  TPCMergingCollectedTracks = 16384,
-  TPCMergingCE = 32768,
-  TPCMergingRefit = 65536,
-  TPCClustererClusters = 131072,
-  TPCClusterer = 262144,
-  TPCClustererDigits = 262144 << 1,
-  TPCClustererPeaks = 262144 << 2,
-  TPCClustererSuppressedPeaks = 262144 << 3,
-  TPCClustererChargeMap = 262144 << 4,
-  TPCClustererZeroedCharges = 262144 << 5
+  TPCSectorTrackingData = 1 << 0,
+  TPCPreLinks = 1 << 1,
+  TPCLinks = 1 << 2,
+  TPCStartHits = 1 << 3,
+  TPCTracklets = 1 << 4,
+  TPCSectorTracks = 1 << 5,
+  TPCHitWeights = 1 << 6,
+  TPCMergingRanges = 1 << 7,
+  TPCMergingSectorTracks = 1 << 8,
+  TPCMergingMatching = 1 << 9,
+  TPCMergingCollectedTracks = 1 << 10,
+  TPCMergingCE = 1 << 11,
+  TPCMergingPrepareFit = 1 << 12,
+  TPCMergingRefit = 1 << 13,
+  TPCMergingLoopers = 1 << 14,
+  TPCCompressedClusters = 1 << 15,
+  TPCDecompressedClusters = 1 << 16,
+  TPCClustererClusters = 1 << 17,
+  TPCClustererDigits = 1 << 18,
+  TPCClustererPeaks = 1 << 19,
+  TPCClustererSuppressedPeaks = 1 << 20,
+  TPCClustererChargeMap = 1 << 21,
+  TPCClustererZeroedCharges = 1 << 22
 };
 
 template <class T, class S, typename... Args>

GPU/GPUTracking/Global/GPUChainTrackingDebugAndProfiling.cxx

@@ -153,7 +153,7 @@ void GPUChainTracking::PrintMemoryStatistics()
     addToMap("TPC Sector TrackHits", usageMap, *processors()->tpcTrackers[i].NTrackHits(), processors()->tpcTrackers[i].NMaxTrackHits());
   }
   addToMap("TPC Clusterer Clusters", usageMap, mRec->MemoryScalers()->nTPCHits, mRec->MemoryScalers()->NTPCClusters(mRec->MemoryScalers()->nTPCdigits));
-  addToMap("TPC Tracks", usageMap, processors()->tpcMerger.NOutputTracks(), processors()->tpcMerger.NMaxTracks());
+  addToMap("TPC Tracks", usageMap, processors()->tpcMerger.NMergedTracks(), processors()->tpcMerger.NMaxTracks());
   addToMap("TPC TrackHits", usageMap, processors()->tpcMerger.NOutputTrackClusters(), processors()->tpcMerger.NMaxOutputTrackClusters());
 
   if (mRec->GetProcessingSettings().createO2Output) {
@@ -181,7 +181,7 @@ void GPUChainTracking::PrintMemoryRelations()
     GPUInfo("MEMREL SectorTracks NCl %d NTrk %d", processors()->tpcTrackers[i].NHitsTotal(), *processors()->tpcTrackers[i].NTracks());
     GPUInfo("MEMREL SectorTrackHits NCl %d NTrkH %d", processors()->tpcTrackers[i].NHitsTotal(), *processors()->tpcTrackers[i].NTrackHits());
   }
-  GPUInfo("MEMREL Tracks NCl %d NTrk %d", processors()->tpcMerger.NMaxClusters(), processors()->tpcMerger.NOutputTracks());
+  GPUInfo("MEMREL Tracks NCl %d NTrk %d", processors()->tpcMerger.NMaxClusters(), processors()->tpcMerger.NMergedTracks());
   GPUInfo("MEMREL TrackHitss NCl %d NTrkH %d", processors()->tpcMerger.NMaxClusters(), processors()->tpcMerger.NOutputTrackClusters());
 }
 

GPU/GPUTracking/Global/GPUChainTrackingMerger.cxx

@@ -143,7 +143,7 @@ int32_t GPUChainTracking::RunTPCTrackingMerger(bool synchronizeOutput)
   runKernel<GPUTPCGMMergerMergeWithinPrepare>(GetGridAuto(0, deviceType));
   RunTPCTrackingMerger_MergeBorderTracks(1, 0, deviceType);
   RunTPCTrackingMerger_Resolve(0, 1, deviceType);
-  DoDebugAndDump(RecoStep::TPCMerging, GPUChainTrackingDebugFlags::TPCMergingMergedTracks, doGPU, Merger, &GPUTPCGMMerger::DumpMergedWithinSectors, *mDebugFile);
+  DoDebugAndDump(RecoStep::TPCMerging, GPUChainTrackingDebugFlags::TPCMergingMatching, doGPU, Merger, &GPUTPCGMMerger::DumpMergedWithinSectors, *mDebugFile);
 
   runKernel<GPUTPCGMMergerClearLinks>(GetGridAuto(0, deviceType), false);
   runKernel<GPUMemClean16>({{1, -WarpSize(), 0, deviceType, RecoStep::TPCMerging}}, MergerShadowAll.TmpCounter(), 2 * NSECTORS * sizeof(*MergerShadowAll.TmpCounter()));
@@ -158,7 +158,7 @@ int32_t GPUChainTracking::RunTPCTrackingMerger(bool synchronizeOutput)
   runKernel<GPUTPCGMMergerMergeSectorsPrepare>(GetGridBlk(std::max(2u, numBlocks), 0, deviceType), 0, 1, 1);
   RunTPCTrackingMerger_MergeBorderTracks(0, -1, deviceType);
   RunTPCTrackingMerger_Resolve(0, 1, deviceType);
-  DoDebugAndDump(RecoStep::TPCMerging, GPUChainTrackingDebugFlags::TPCMergingMergedTracks, doGPU, Merger, &GPUTPCGMMerger::DumpMergedBetweenSectors, *mDebugFile);
+  DoDebugAndDump(RecoStep::TPCMerging, GPUChainTrackingDebugFlags::TPCMergingMatching, doGPU, Merger, &GPUTPCGMMerger::DumpMergedBetweenSectors, *mDebugFile);
 
   runKernel<GPUMemClean16>({{1, -WarpSize(), 0, deviceType, RecoStep::TPCMerging}}, MergerShadowAll.TmpCounter(), 2 * NSECTORS * sizeof(*MergerShadowAll.TmpCounter()));
 
@@ -202,7 +202,7 @@ int32_t GPUChainTracking::RunTPCTrackingMerger(bool synchronizeOutput)
   runKernel<GPUTPCGMMergerPrepareClusters, 1>(GetGridAuto(0, deviceType));
   runKernel<GPUTPCGMMergerPrepareClusters, 2>(GetGridAuto(0, deviceType));
 
-  DoDebugAndDump(RecoStep::TPCMerging, GPUChainTrackingDebugFlags::TPCMergingRefit, doGPU, Merger, &GPUTPCGMMerger::DumpFitPrepare, *mDebugFile);
+  DoDebugAndDump(RecoStep::TPCMerging, GPUChainTrackingDebugFlags::TPCMergingPrepareFit, doGPU, Merger, &GPUTPCGMMerger::DumpFitPrepare, *mDebugFile);
 
   if (doGPU) {
     CondWaitEvent(waitForTransfer, &mEvents->single);
@@ -220,7 +220,7 @@ int32_t GPUChainTracking::RunTPCTrackingMerger(bool synchronizeOutput)
     mOutputQueue.clear();
   }
 
-  runKernel<GPUTPCGMMergerTrackFit>(doGPU ? GetGrid(Merger.NOutputTracks(), 0) : GetGridAuto(0), mergerSortTracks ? 1 : 0);
+  runKernel<GPUTPCGMMergerTrackFit>(doGPU ? GetGrid(Merger.NMergedTracks(), 0) : GetGridAuto(0), mergerSortTracks ? 1 : 0);
   if (param().rec.tpc.retryRefit == 1) {
     runKernel<GPUTPCGMMergerTrackFit>(GetGridAuto(0), -1);
   }
@@ -233,13 +233,14 @@ int32_t GPUChainTracking::RunTPCTrackingMerger(bool synchronizeOutput)
   runKernel<GPUTPCGMMergerFinalize, 1>(GetGridAuto(0, deviceType));
   runKernel<GPUTPCGMMergerFinalize, 2>(GetGridAuto(0, deviceType));
   if (param().rec.tpc.mergeLoopersAfterburner) {
-    runKernel<GPUTPCGMMergerMergeLoopers, 0>(doGPU ? GetGrid(Merger.NOutputTracks(), 0, deviceType) : GetGridAuto(0, deviceType));
+    runKernel<GPUTPCGMMergerMergeLoopers, 0>(doGPU ? GetGrid(Merger.NMergedTracks(), 0, deviceType) : GetGridAuto(0, deviceType));
     if (doGPU) {
       TransferMemoryResourceLinkToHost(RecoStep::TPCMerging, Merger.MemoryResMemory(), 0);
       SynchronizeStream(0); // TODO: could probably synchronize on an event after runKernel<GPUTPCGMMergerMergeLoopers, 1>
     }
     runKernel<GPUTPCGMMergerMergeLoopers, 1>(GetGridAuto(0, deviceType));
     runKernel<GPUTPCGMMergerMergeLoopers, 2>(doGPU ? GetGrid(Merger.Memory()->nLooperMatchCandidates, 0, deviceType) : GetGridAuto(0, deviceType));
+    DoDebugAndDump(RecoStep::TPCMerging, GPUChainTrackingDebugFlags::TPCMergingLoopers, Merger, &GPUTPCGMMerger::DumpLoopers, *mDebugFile);
   }
   DoDebugAndDump(RecoStep::TPCMerging, GPUChainTrackingDebugFlags::TPCMergingRefit, doGPU, Merger, &GPUTPCGMMerger::DumpFinal, *mDebugFile);
 
@@ -255,10 +256,10 @@ int32_t GPUChainTracking::RunTPCTrackingMerger(bool synchronizeOutput)
           throw std::runtime_error("QA Scratch buffer exceeded");
         }
       }
-      GPUMemCpy(RecoStep::TPCMerging, Merger.OutputTracks(), MergerShadowAll.OutputTracks(), Merger.NOutputTracks() * sizeof(*Merger.OutputTracks()), outputStream, 0, nullptr, waitEvent);
+      GPUMemCpy(RecoStep::TPCMerging, Merger.MergedTracks(), MergerShadowAll.MergedTracks(), Merger.NMergedTracks() * sizeof(*Merger.MergedTracks()), outputStream, 0, nullptr, waitEvent);
       waitEvent = nullptr;
       if (param().dodEdxEnabled) {
-        GPUMemCpy(RecoStep::TPCMerging, Merger.OutputTracksdEdx(), MergerShadowAll.OutputTracksdEdx(), Merger.NOutputTracks() * sizeof(*Merger.OutputTracksdEdx()), outputStream, 0);
+        GPUMemCpy(RecoStep::TPCMerging, Merger.MergedTracksdEdx(), MergerShadowAll.MergedTracksdEdx(), Merger.NMergedTracks() * sizeof(*Merger.MergedTracksdEdx()), outputStream, 0);
       }
       GPUMemCpy(RecoStep::TPCMerging, Merger.Clusters(), MergerShadowAll.Clusters(), Merger.NOutputTrackClusters() * sizeof(*Merger.Clusters()), outputStream, 0);
       if (param().par.earlyTpcTransform) {
@@ -325,8 +326,8 @@ int32_t GPUChainTracking::RunTPCTrackingMerger(bool synchronizeOutput)
     mRec->ReturnVolatileDeviceMemory();
   }
 
-  mIOPtrs.mergedTracks = Merger.OutputTracks();
-  mIOPtrs.nMergedTracks = Merger.NOutputTracks();
+  mIOPtrs.mergedTracks = Merger.MergedTracks();
+  mIOPtrs.nMergedTracks = Merger.NMergedTracks();
   mIOPtrs.mergedTrackHits = Merger.Clusters();
   mIOPtrs.mergedTrackHitsXYZ = Merger.ClustersXYZ();
   mIOPtrs.nMergedTrackHits = Merger.NOutputTrackClusters();
@@ -339,8 +340,8 @@ int32_t GPUChainTracking::RunTPCTrackingMerger(bool synchronizeOutput)
   mIOPtrs.outputTracksTPCO2MC = Merger.OutputTracksTPCO2MC();
 
   if (doGPU) {
-    processorsShadow()->ioPtrs.mergedTracks = MergerShadow.OutputTracks();
-    processorsShadow()->ioPtrs.nMergedTracks = Merger.NOutputTracks();
+    processorsShadow()->ioPtrs.mergedTracks = MergerShadow.MergedTracks();
+    processorsShadow()->ioPtrs.nMergedTracks = Merger.NMergedTracks();
     processorsShadow()->ioPtrs.mergedTrackHits = MergerShadow.Clusters();
     processorsShadow()->ioPtrs.mergedTrackHitsXYZ = MergerShadow.ClustersXYZ();
     processorsShadow()->ioPtrs.nMergedTrackHits = Merger.NOutputTrackClusters();

GPU/GPUTracking/Global/GPUChainTrackingRefit.cxx

@@ -33,7 +33,7 @@ int32_t GPUChainTracking::RunRefit()
   SetupGPUProcessor(&Refit, false);
   RefitShadow.SetPtrsFromGPUConstantMem(processorsShadow(), doGPU ? &processorsDevice()->param : nullptr);
   RefitShadow.SetPropagator(doGPU ? processorsShadow()->calibObjects.o2Propagator : GetO2Propagator());
-  RefitShadow.mPTracks = (doGPU ? processorsShadow() : processors())->tpcMerger.OutputTracks();
+  RefitShadow.mPTracks = (doGPU ? processorsShadow() : processors())->tpcMerger.MergedTracks();
   WriteToConstantMemory(RecoStep::Refit, (char*)&processors()->trackingRefit - (char*)processors(), &RefitShadow, sizeof(RefitShadow), 0);
   // TransferMemoryResourcesToGPU(RecoStep::Refit, &Refit, 0);
   if (param().rec.trackingRefitGPUModel) {

GPU/GPUTracking/Global/GPUChainTrackingSectorTracker.cxx

@@ -176,7 +176,9 @@ int32_t GPUChainTracking::RunTPCTrackingSectors_internal()
     }
 
     if (GetProcessingSettings().debugLevel >= 6) {
-      *mDebugFile << "\n\nReconstruction: Sector " << iSector << "/" << NSECTORS << std::endl;
+      if ((GetProcessingSettings().debugMask & 63)) {
+        *mDebugFile << "\n\nReconstruction: Sector " << iSector << "/" << NSECTORS << std::endl;
+      }
       if (GetProcessingSettings().debugMask & GPUChainTrackingDebugFlags::TPCSectorTrackingData) {
         if (doGPU) {
           TransferMemoryResourcesToHost(RecoStep::TPCSectorTracking, &trk, -1, true);