GPU: Improve debug dumps

Author: davidrohr

GPU/GPUTracking/Base/GPUReconstruction.cxx

@@ -269,6 +269,9 @@ int32_t GPUReconstruction::InitPhaseBeforeDevice()
 #ifndef GPUCA_DETERMINISTIC_MODE
     GPUError("WARNING, deterministicGPUReconstruction needs GPUCA_DETERMINISTIC_MODE for being fully deterministic, without only most indeterminism by concurrency is removed, but floating point effects remain!");
 #endif
+    if (mProcessingSettings->debugLevel >= 6 && ((mProcessingSettings->debugMask + 1) & mProcessingSettings->debugMask)) {
+      GPUError("WARNING: debugMask %d - debug output might not be deterministic with intermediate steps missing", mProcessingSettings->debugMask);
+    }
     mProcessingSettings->overrideClusterizerFragmentLen = TPC_MAX_FRAGMENT_LEN_GPU;
     if (GetProcessingSettings().createO2Output > 1) {
       mProcessingSettings->createO2Output = 1;

GPU/GPUTracking/Global/GPUChainTrackingDebug.h

@@ -28,8 +28,8 @@ enum GPUChainTrackingDebugFlags : uint32_t {
   TPCLinks = 1 << 2,
   TPCStartHits = 1 << 3,
   TPCTracklets = 1 << 4,
-  TPCSectorTracks = 1 << 5,
-  TPCHitWeights = 1 << 6,
+  TPCHitWeights = 1 << 5,
+  TPCSectorTracks = 1 << 6,
   TPCMergingRanges = 1 << 7,
   TPCMergingSectorTracks = 1 << 8,
   TPCMergingMatching = 1 << 9,

GPU/GPUTracking/Merger/GPUTPCGMMergerDump.cxx

@@ -51,7 +51,7 @@ void GPUTPCGMMerger::DumpSectorTracks(std::ostream& out) const
       out << "  Track type " << iGlobal << "\n";
       for (int32_t j = mSectorTrackInfoIndex[iSector + NSECTORS * iGlobal]; j < mSectorTrackInfoIndex[iSector + NSECTORS * iGlobal + 1]; j++) {
         const auto& trk = mSectorTrackInfos[j];
-        out << "    Track " << j << ": LocalId " << (iGlobal ? (trk.LocalTrackId() >> 24) : -1) << "/" << (iGlobal ? (trk.LocalTrackId() & 0xFFFFFF) : -1) << " X " << trk.X() << " offsetz " << trk.TOffset() << " A " << trk.Alpha() << " Y " << trk.Y() << " Z " << trk.Z() << " SinPhi " << trk.SinPhi() << " CosPhi " << trk.CosPhi() << " SecPhi " << trk.SecPhi() << " Tgl " << trk.DzDs() << " QPt " << trk.QPt() << "\n";
+        out << "    Track " << j << ": LocalId " << (iGlobal ? (trk.LocalTrackId() >> 24) : -1) << "/" << (iGlobal ? (trk.LocalTrackId() & 0xFFFFFF) : -1) << " NCl " << trk.NClusters() << " X " << trk.X() << " offsetz " << trk.TOffset() << " A " << trk.Alpha() << " Y " << trk.Y() << " Z " << trk.Z() << " SinPhi " << trk.SinPhi() << " CosPhi " << trk.CosPhi() << " SecPhi " << trk.SecPhi() << " Tgl " << trk.DzDs() << " QPt " << trk.QPt() << "\n";
       }
     }
   }
@@ -135,7 +135,7 @@ void GPUTPCGMMerger::DumpMergedBetweenSectors(std::ostream& out) const
 void GPUTPCGMMerger::DumpCollected(std::ostream& out) const
 {
   std::streamsize ss = out.precision();
-  out << std::setprecision(2);
+  out << std::setprecision(6);
   out << "\nTPC Merger Collected Tracks\n";
   for (uint32_t i = 0; i < mMemory->nMergedTracks; i++) {
     const auto& trk = mMergedTracks[i];