GPU: Fix Debug Dump of deterministic mode without dEdx (was segfaulting in sync mode)

davidrohr · davidrohr · commit 272b78eaf71b · 2025-05-07T16:02:15.000+02:00
diff --git a/GPU/GPUTracking/Base/GPUParam.cxx b/GPU/GPUTracking/Base/GPUParam.cxx
@@ -135,9 +135,9 @@ void GPUParam::UpdateSettings(const GPUSettingsGRP* g, const GPUSettingsProcessi
     UpdateRun3ClusterErrors(p->param.tpcErrorParamY, p->param.tpcErrorParamZ);
   }
   if (w) {
-    par.dodEdx = dodEdxDownscaled = w->steps.isSet(GPUDataTypes::RecoStep::TPCdEdx);
-    if (par.dodEdx && p && p->tpcDownscaledEdx != 0) {
-      dodEdxDownscaled = (rand() % 100) < p->tpcDownscaledEdx;
+    par.dodEdx = dodEdxEnabled = w->steps.isSet(GPUDataTypes::RecoStep::TPCdEdx);
+    if (dodEdxEnabled && p && p->tpcDownscaledEdx != 0) {
+      dodEdxEnabled = (rand() % 100) < p->tpcDownscaledEdx;
     }
   }
   if (d) {
diff --git a/GPU/GPUTracking/Base/GPUParam.h b/GPU/GPUTracking/Base/GPUParam.h
@@ -54,7 +54,7 @@ struct GPUParam_t {
   float bzCLight;
   float qptB5Scaler;
 
-  int8_t dodEdxDownscaled;
+  int8_t dodEdxEnabled;
   int32_t continuousMaxTimeBin;
   int32_t tpcCutTimeBin;
 
diff --git a/GPU/GPUTracking/Global/GPUChainTrackingMerger.cxx b/GPU/GPUTracking/Global/GPUChainTrackingMerger.cxx
@@ -256,7 +256,7 @@ int32_t GPUChainTracking::RunTPCTrackingMerger(bool synchronizeOutput)
       }
       GPUMemCpy(RecoStep::TPCMerging, Merger.OutputTracks(), MergerShadowAll.OutputTracks(), Merger.NOutputTracks() * sizeof(*Merger.OutputTracks()), outputStream, 0, nullptr, waitEvent);
       waitEvent = nullptr;
-      if (param().dodEdxDownscaled) {
+      if (param().dodEdxEnabled) {
         GPUMemCpy(RecoStep::TPCMerging, Merger.OutputTracksdEdx(), MergerShadowAll.OutputTracksdEdx(), Merger.NOutputTracks() * sizeof(*Merger.OutputTracksdEdx()), outputStream, 0);
       }
       GPUMemCpy(RecoStep::TPCMerging, Merger.Clusters(), MergerShadowAll.Clusters(), Merger.NOutputTrackClusters() * sizeof(*Merger.Clusters()), outputStream, 0);
diff --git a/GPU/GPUTracking/Merger/GPUTPCGMMerger.cxx b/GPU/GPUTracking/Merger/GPUTPCGMMerger.cxx
@@ -299,7 +299,7 @@ void* GPUTPCGMMerger::SetPointersRefitScratch(void* mem)
 void* GPUTPCGMMerger::SetPointersOutput(void* mem)
 {
   computePointerWithAlignment(mem, mOutputTracks, mNMaxTracks);
-  if (mRec->GetParam().dodEdxDownscaled) {
+  if (mRec->GetParam().dodEdxEnabled) {
     computePointerWithAlignment(mem, mOutputTracksdEdx, mNMaxTracks);
     computePointerWithAlignment(mem, mOutputTracksdEdxAlt, mNMaxTracks);
   }
diff --git a/GPU/GPUTracking/Merger/GPUTPCGMMergerDump.cxx b/GPU/GPUTracking/Merger/GPUTPCGMMergerDump.cxx
@@ -203,7 +203,7 @@ void GPUTPCGMMerger::DumpRefit(std::ostream& out) const
     const auto& p = trk.GetParam();
     const auto& po = trk.OuterParam();
     out << "  Track " << i << ": OK " << trk.OK() << " Alpha " << trk.GetAlpha() << " X " << p.GetX() << " offset " << p.GetTZOffset() << " Y " << p.GetY() << " Z " << p.GetZ() << " SPhi " << p.GetSinPhi() << " Tgl " << p.GetDzDs() << " QPt " << p.GetQPt() << " NCl " << trk.NClusters() << " / " << trk.NClustersFitted() << " Cov " << p.GetErr2Y() << "/" << p.GetErr2Z()
-        << " dEdx " << (trk.OK() ? mOutputTracksdEdx[i].dEdxTotTPC : -1.f) << "/" << (trk.OK() ? mOutputTracksdEdx[i].dEdxMaxTPC : -1.f)
+        << " dEdx " << (trk.OK() && Param().dodEdxEnabled ? mOutputTracksdEdx[i].dEdxTotTPC : -1.f) << "/" << (trk.OK() && Param().dodEdxEnabled ? mOutputTracksdEdx[i].dEdxMaxTPC : -1.f)
         << " Outer " << po.P[0] << "/" << po.P[1] << "/" << po.P[2] << "/" << po.P[3] << "/" << po.P[4] << "\n";
   }
   out << std::setprecision(ss);
diff --git a/GPU/GPUTracking/Merger/GPUTPCGMO2Output.cxx b/GPU/GPUTracking/Merger/GPUTPCGMO2Output.cxx
@@ -44,7 +44,7 @@ GPUdii() void GPUTPCGMO2Output::Thread<GPUTPCGMO2Output::prepare>(int32_t nBlock
 
   constexpr uint8_t flagsReject = getFlagsReject();
   const uint32_t flagsRequired = getFlagsRequired(merger.Param().rec);
-  bool cutOnTrackdEdx = merger.Param().par.dodEdx && merger.Param().dodEdxDownscaled && merger.Param().rec.tpc.minTrackdEdxMax2Tot > 0.f;
+  bool cutOnTrackdEdx = merger.Param().par.dodEdx && merger.Param().dodEdxEnabled && merger.Param().rec.tpc.minTrackdEdxMax2Tot > 0.f;
 
   GPUTPCGMMerger::tmpSort* GPUrestrict() trackSort = merger.TrackSortO2();
   uint2* GPUrestrict() tmpData = merger.ClusRefTmp();
@@ -130,7 +130,7 @@ GPUdii() void GPUTPCGMO2Output::Thread<GPUTPCGMO2Output::output>(int32_t nBlocks
 
     oTrack.setChi2(tracks[i].GetParam().GetChi2());
     auto& outerPar = tracks[i].OuterParam();
-    if (merger.Param().par.dodEdx && merger.Param().dodEdxDownscaled) {
+    if (merger.Param().par.dodEdx && merger.Param().dodEdxEnabled) {
       oTrack.setdEdx(tracksdEdx[i]);
       oTrack.setdEdxAlt(tracksdEdxAlt[i]);
     }
@@ -148,7 +148,7 @@ GPUdii() void GPUTPCGMO2Output::Thread<GPUTPCGMO2Output::output>(int32_t nBlocks
        outerPar.C[6], outerPar.C[7], outerPar.C[8], outerPar.C[9], outerPar.C[10], outerPar.C[11],
        outerPar.C[12], outerPar.C[13], outerPar.C[14]}));
 
-    if (merger.Param().par.dodEdx && merger.Param().dodEdxDownscaled && merger.Param().rec.tpc.enablePID) {
+    if (merger.Param().par.dodEdx && merger.Param().dodEdxEnabled && merger.Param().rec.tpc.enablePID) {
       PIDResponse pidResponse{};
       auto pid = pidResponse.getMostProbablePID(oTrack, merger.Param().rec.tpc.PID_EKrangeMin, merger.Param().rec.tpc.PID_EKrangeMax, merger.Param().rec.tpc.PID_EPrangeMin, merger.Param().rec.tpc.PID_EPrangeMax, merger.Param().rec.tpc.PID_EDrangeMin, merger.Param().rec.tpc.PID_EDrangeMax, merger.Param().rec.tpc.PID_ETrangeMin, merger.Param().rec.tpc.PID_ETrangeMax, merger.Param().rec.tpc.PID_useNsigma, merger.Param().rec.tpc.PID_sigma);
       auto pidRemap = merger.Param().rec.tpc.PID_remap[pid];
diff --git a/GPU/GPUTracking/Merger/GPUTPCGMTrackParam.cxx b/GPU/GPUTracking/Merger/GPUTPCGMTrackParam.cxx
@@ -216,7 +216,7 @@ GPUd() bool GPUTPCGMTrackParam::Fit(GPUTPCGMMerger* GPUrestrict() merger, int32_
           continue;
         }
       } else if (allowModification && lastRow != 255 && CAMath::Abs(cluster.row - lastRow) > 1) {
-        bool dodEdx = param.par.dodEdx && param.dodEdxDownscaled && param.rec.tpc.adddEdxSubThresholdClusters && iWay == nWays - 1 && CAMath::Abs(cluster.row - lastRow) == 2 && cluster.leg == clusters[maxN - 1].leg;
+        bool dodEdx = param.par.dodEdx && param.dodEdxEnabled && param.rec.tpc.adddEdxSubThresholdClusters && iWay == nWays - 1 && CAMath::Abs(cluster.row - lastRow) == 2 && cluster.leg == clusters[maxN - 1].leg;
         dodEdx = AttachClustersPropagate(merger, cluster.sector, lastRow, cluster.row, iTrk, cluster.leg == clusters[maxN - 1].leg, prop, inFlyDirection, GPUCA_MAX_SIN_PHI, dodEdx);
         if (dodEdx) {
           dEdx.fillSubThreshold(lastRow - wayDirection);
@@ -367,7 +367,7 @@ GPUd() bool GPUTPCGMTrackParam::Fit(GPUTPCGMMerger* GPUrestrict() merger, int32_
           CADEBUG(printf("Reinit linearization\n"));
           prop.SetTrack(this, prop.GetAlpha());
         }
-        if (param.par.dodEdx && param.dodEdxDownscaled && iWay == nWays - 1 && cluster.leg == clusters[maxN - 1].leg) { // TODO: Costimize flag to remove, and option to remove double-clusters
+        if (param.par.dodEdx && param.dodEdxEnabled && iWay == nWays - 1 && cluster.leg == clusters[maxN - 1].leg) { // TODO: Costimize flag to remove, and option to remove double-clusters
           bool acc = (clusterState & param.rec.tpc.dEdxClusterRejectionFlagMask) == 0, accAlt = (clusterState & param.rec.tpc.dEdxClusterRejectionFlagMaskAlt) == 0;
           if (acc || accAlt) {
             float qtot = 0, qmax = 0, pad = 0, relTime = 0;
@@ -426,7 +426,7 @@ GPUd() bool GPUTPCGMTrackParam::Fit(GPUTPCGMMerger* GPUrestrict() merger, int32_
 
   // TODO: we have looping tracks here with 0 accepted clusters in the primary leg. In that case we should refit the track using only the primary leg.
 
-  if (param.par.dodEdx && param.dodEdxDownscaled) {
+  if (param.par.dodEdx && param.dodEdxEnabled) {
     dEdx.computedEdx(merger->OutputTracksdEdx()[iTrk], param);
     dEdxAlt.computedEdx(merger->OutputTracksdEdxAlt()[iTrk], param);
   }

Original file line number	Diff line number	Diff line change
`@@ -135,9 +135,9 @@ void GPUParam::UpdateSettings(const GPUSettingsGRP* g, const GPUSettingsProcessi`
`135`	`135`	`UpdateRun3ClusterErrors(p->param.tpcErrorParamY, p->param.tpcErrorParamZ);`
`136`	`136`	`}`
`137`	`137`	`if (w) {`
`138`		`- par.dodEdx = dodEdxDownscaled = w->steps.isSet(GPUDataTypes::RecoStep::TPCdEdx);`
`139`		`- if (par.dodEdx && p && p->tpcDownscaledEdx != 0) {`
`140`		`- dodEdxDownscaled = (rand() % 100) < p->tpcDownscaledEdx;`
	`138`	`+ par.dodEdx = dodEdxEnabled = w->steps.isSet(GPUDataTypes::RecoStep::TPCdEdx);`
	`139`	`+ if (dodEdxEnabled && p && p->tpcDownscaledEdx != 0) {`
	`140`	`+ dodEdxEnabled = (rand() % 100) < p->tpcDownscaledEdx;`
`141`	`141`	`}`
`142`	`142`	`}`
`143`	`143`	`if (d) {`
Original file line number	Diff line number	Diff line change
`@@ -256,7 +256,7 @@ int32_t GPUChainTracking::RunTPCTrackingMerger(bool synchronizeOutput)`
`256`	`256`	`}`
`257`	`257`	`GPUMemCpy(RecoStep::TPCMerging, Merger.OutputTracks(), MergerShadowAll.OutputTracks(), Merger.NOutputTracks() * sizeof(*Merger.OutputTracks()), outputStream, 0, nullptr, waitEvent);`
`258`	`258`	`waitEvent = nullptr;`
`259`		`- if (param().dodEdxDownscaled) {`
	`259`	`+ if (param().dodEdxEnabled) {`
`260`	`260`	`GPUMemCpy(RecoStep::TPCMerging, Merger.OutputTracksdEdx(), MergerShadowAll.OutputTracksdEdx(), Merger.NOutputTracks() * sizeof(*Merger.OutputTracksdEdx()), outputStream, 0);`
`261`	`261`	`}`
`262`	`262`	`GPUMemCpy(RecoStep::TPCMerging, Merger.Clusters(), MergerShadowAll.Clusters(), Merger.NOutputTrackClusters() * sizeof(*Merger.Clusters()), outputStream, 0);`
Original file line number	Diff line number	Diff line change
`@@ -299,7 +299,7 @@ void* GPUTPCGMMerger::SetPointersRefitScratch(void* mem)`
`299`	`299`	`void* GPUTPCGMMerger::SetPointersOutput(void* mem)`
`300`	`300`	`{`
`301`	`301`	`computePointerWithAlignment(mem, mOutputTracks, mNMaxTracks);`
`302`		`- if (mRec->GetParam().dodEdxDownscaled) {`
	`302`	`+ if (mRec->GetParam().dodEdxEnabled) {`
`303`	`303`	`computePointerWithAlignment(mem, mOutputTracksdEdx, mNMaxTracks);`
`304`	`304`	`computePointerWithAlignment(mem, mOutputTracksdEdxAlt, mNMaxTracks);`
`305`	`305`	`}`
Original file line number	Diff line number	Diff line change
`@@ -203,7 +203,7 @@ void GPUTPCGMMerger::DumpRefit(std::ostream& out) const`
`203`	`203`	`const auto& p = trk.GetParam();`
`204`	`204`	`const auto& po = trk.OuterParam();`
`205`	`205`	`out << " Track " << i << ": OK " << trk.OK() << " Alpha " << trk.GetAlpha() << " X " << p.GetX() << " offset " << p.GetTZOffset() << " Y " << p.GetY() << " Z " << p.GetZ() << " SPhi " << p.GetSinPhi() << " Tgl " << p.GetDzDs() << " QPt " << p.GetQPt() << " NCl " << trk.NClusters() << " / " << trk.NClustersFitted() << " Cov " << p.GetErr2Y() << "/" << p.GetErr2Z()`
`206`		`- << " dEdx " << (trk.OK() ? mOutputTracksdEdx[i].dEdxTotTPC : -1.f) << "/" << (trk.OK() ? mOutputTracksdEdx[i].dEdxMaxTPC : -1.f)`
	`206`	`+ << " dEdx " << (trk.OK() && Param().dodEdxEnabled ? mOutputTracksdEdx[i].dEdxTotTPC : -1.f) << "/" << (trk.OK() && Param().dodEdxEnabled ? mOutputTracksdEdx[i].dEdxMaxTPC : -1.f)`
`207`	`207`	`<< " Outer " << po.P[0] << "/" << po.P[1] << "/" << po.P[2] << "/" << po.P[3] << "/" << po.P[4] << "\n";`
`208`	`208`	`}`
`209`	`209`	`out << std::setprecision(ss);`