GPU QA: Fix accounting of protected and unattached clusters with attachProtect flag

Author: davidrohr

GPU/GPUTracking/DataCompression/GPUTPCClusterRejection.h

@@ -16,12 +16,19 @@
 #define GPUTPCCLUSTERREJECTION_H
 
 #include "GPUTPCGMMergerTypes.h"
+#include "GPUCommonMath.h"
 
 namespace o2::gpu
 {
 struct GPUTPCClusterRejection {
+  template <class T, class S>
+  GPUdi() static bool IsTrackRejected(const T& trk, const S& param)
+  {
+    return CAMath::Abs(trk.GetParam().GetQPt() * param.qptB5Scaler) > param.rec.tpc.rejectQPtB5 || trk.MergedLooper();
+  }
+
   template <bool C, class T = void, class S = void>
-  static constexpr inline bool GetProtectionStatus(int32_t attach, bool& physics, bool& protect, T* counts = nullptr, S* mev200 = nullptr)
+  GPUdi() static constexpr bool GetRejectionStatus(int32_t attach, bool& physics, T* counts = nullptr, S* mev200 = nullptr)
   {
     (void)counts; // FIXME: Avoid incorrect -Wunused-but-set-parameter warning
     (void)mev200;
@@ -39,7 +46,6 @@ struct GPUTPCClusterRejection {
       }
       retVal = true;
     } else if (attach & gputpcgmmergertypes::attachTube) {
-      protect = true;
       if constexpr (C) {
         if (*mev200) {
           counts->nTube200++;
@@ -49,7 +55,6 @@ struct GPUTPCClusterRejection {
       }
       retVal = false;
     } else if ((attach & gputpcgmmergertypes::attachGood) == 0) {
-      protect = true;
       if constexpr (C) {
         counts->nRejected++;
       }
@@ -60,16 +65,15 @@ struct GPUTPCClusterRejection {
     }
 
     if (attach & gputpcgmmergertypes::attachProtect) {
-      protect = true;
       retVal = false;
     }
     return retVal;
   }
 
-  static constexpr inline bool GetIsRejected(int32_t attach)
+  GPUdi() static constexpr bool GetIsRejected(int32_t attach)
   {
-    bool physics = false, protect = false;
-    return GetProtectionStatus<false>(attach, physics, protect);
+    bool physics = false;
+    return GetRejectionStatus<false>(attach, physics);
   }
 };
 } // namespace o2::gpu

GPU/GPUTracking/DataCompression/GPUTPCCompression.cxx

@@ -111,7 +111,7 @@ void GPUTPCCompression::RegisterMemoryAllocation()
   if (gatherMode == 3) {
     mMemoryResOutputGPU = mRec->RegisterMemoryAllocation(this, &GPUTPCCompression::SetPointersOutputGPU, GPUMemoryResource::MEMORY_SCRATCH | GPUMemoryResource::MEMORY_GPU | GPUMemoryResource::MEMORY_CUSTOM | GPUMemoryResource::MEMORY_STACK, "TPCCompressionOutputGPU");
   }
-  uint32_t stackScratch = (gatherMode != 3) ? GPUMemoryResource::MEMORY_STACK : 0;
+  uint32_t stackScratch = (gatherMode != 3) ? GPUMemoryResource::MEMORY_STACK : 0; // TODO: Can we use stacked memory also with gather mode 3?
   if (gatherMode < 2) {
     mRec->RegisterMemoryAllocation(this, &GPUTPCCompression::SetPointersOutput, GPUMemoryResource::MEMORY_OUTPUT | stackScratch, "TPCCompressionOutput");
   }

GPU/GPUTracking/DataCompression/GPUTPCCompression.h

@@ -60,6 +60,7 @@ class GPUTPCCompression : public GPUProcessor
 #ifndef GPUCA_GPUCODE
   void DumpCompressedClusters(std::ostream& out);
 #endif
+  GPUd() bool rejectCluster(int32_t idx, const GPUParam& param, const GPUTrackingInOutPointers& ioPtrs) const;
 
  protected:
   struct memory {
@@ -89,7 +90,6 @@ class GPUTPCCompression : public GPUProcessor
   void SetPointersCompressedClusters(void*& mem, T& c, uint32_t nClA, uint32_t nTr, uint32_t nClU, bool reducedClA);
   template <class T>
   GPUd() static void truncateSignificantBits(T& val, uint32_t nBits, uint32_t max);
-  GPUd() bool rejectCluster(int32_t idx, GPUParam& param, const GPUTrackingInOutPointers& ioPtrs);
 
   int16_t mMemoryResOutputHost = -1;
   int16_t mMemoryResOutputGPU = -1;

GPU/GPUTracking/DataCompression/GPUTPCCompressionKernels.cxx

@@ -39,7 +39,7 @@ GPUdii() void GPUTPCCompressionKernels::Thread<GPUTPCCompressionKernels::step0at
     if (!trk.OK()) {
       continue;
     }
-    bool rejectTrk = CAMath::Abs(trk.GetParam().GetQPt() * processors.param.qptB5Scaler) > processors.param.rec.tpc.rejectQPtB5 || trk.MergedLooper();
+    bool rejectTrk = GPUTPCClusterRejection::IsTrackRejected(trk, param);
     uint32_t nClustersStored = 0;
     CompressedClustersPtrs& GPUrestrict() c = compressor.mPtrs;
     uint8_t lastRow = 0, lastSector = 0;
@@ -185,7 +185,7 @@ GPUd() bool GPUTPCCompressionKernels::GPUTPCCompressionKernels_Compare<4>::opera
   return mClsPtr[a].qTot < mClsPtr[b].qTot;
 }
 
-GPUd() bool GPUTPCCompression::rejectCluster(int32_t idx, GPUParam& GPUrestrict() param, const GPUTrackingInOutPointers& GPUrestrict() ioPtrs)
+GPUd() bool GPUTPCCompression::rejectCluster(int32_t idx, const GPUParam& GPUrestrict() param, const GPUTrackingInOutPointers& GPUrestrict() ioPtrs) const
 {
   if (mClusterStatus[idx]) {
     return true;
@@ -206,7 +206,7 @@ GPUd() bool GPUTPCCompression::rejectCluster(int32_t idx, GPUParam& GPUrestrict(
     }
     int32_t id = attach & gputpcgmmergertypes::attachTrackMask;
     auto& trk = ioPtrs.mergedTracks[id];
-    if (CAMath::Abs(trk.GetParam().GetQPt() * param.qptB5Scaler) > param.rec.tpc.rejectQPtB5 || trk.MergedLooper()) {
+    if (GPUTPCClusterRejection::IsTrackRejected(trk, param)) {
       return true;
     }
   }

GPU/GPUTracking/Global/GPUChainTrackingClusterizer.cxx

@@ -587,7 +587,7 @@ int32_t GPUChainTracking::RunTPCClusterizer(bool synchronizeOutput)
     return ForwardTPCDigits();
   }
 #ifdef GPUCA_TPC_GEOMETRY_O2
-  [[maybe_unused]] int32_t tpcTimeBinCut = mUpdateNewCalibObjects && mNewCalibValues->newTPCTimeBinCut ? mNewCalibValues->tpcTimeBinCut : param().tpcCutTimeBin;
+  [[maybe_unused]] int32_t tpcTimeBinCut = (mUpdateNewCalibObjects && mNewCalibValues->newTPCTimeBinCut) ? mNewCalibValues->tpcTimeBinCut : param().tpcCutTimeBin; // TODO: Implement time bin cut fultering
   mRec->PushNonPersistentMemory(qStr2Tag("TPCCLUST"));
   const auto& threadContext = GetThreadContext();
   const bool doGPU = GetRecoStepsGPU() & RecoStep::TPCClusterFinding;