GPU TPC: Avoid some code duplication

Author: davidrohr

GPU/GPUTracking/Merger/GPUTPCGMMergedTrack.h

@@ -49,6 +49,24 @@ class GPUTPCGMMergedTrack
   GPUd() bool MergedLooperConnected() const { return mFlags & 0x20; }
   GPUd() bool MergedLooper() const { return mFlags & 0x30; }
   GPUd() int32_t PrevSegment() const { return mPrevSegment; }
+  template <class T>
+  GPUd() static T* GetFirstSegment_helper(T* me, T* base)
+  {
+    if (me->mPrevSegment < 0) {
+      return me;
+    }
+    T* cur = &base[me->mPrevSegment];
+    while (cur->mPrevSegment >= 0) {
+      T* next = &base[cur->mPrevSegment];
+      if (next == me) {
+        return cur;
+      }
+      cur = next;
+    }
+    return cur;
+  }
+  GPUd() GPUTPCGMMergedTrack* GetFirstSegment(GPUTPCGMMergedTrack* base) { return GetFirstSegment_helper<GPUTPCGMMergedTrack>(this, base); }
+  GPUd() const GPUTPCGMMergedTrack* GetFirstSegment(const GPUTPCGMMergedTrack* base) const { return GetFirstSegment_helper<const GPUTPCGMMergedTrack>(this, base); }
   GPUd() uint8_t Leg() const { return mLeg; }
   GPUd() uint8_t Flags() const { return mFlags; }
 

GPU/GPUTracking/Merger/GPUTPCGMMerger.cxx

@@ -1770,14 +1770,7 @@ GPUd() void GPUTPCGMMerger::PrepareForFit1(int32_t nBlocks, int32_t nThreads, in
         CAMath::AtomicAdd(&mSharedCount[mClusters[trk.FirstClusterRef() + j].num], 1u);
       }
       if (!trk.CCE() && !trk.MergedLooper()) {
-        GPUTPCGMMergedTrack* updTrk = &trk;
-        while (updTrk->PrevSegment() >= 0) {
-          auto next = &mMergedTracks[updTrk->PrevSegment()];
-          if (next == &trk) {
-            break;
-          }
-          updTrk = next;
-        }
+        GPUTPCGMMergedTrack* updTrk = trk.GetFirstSegment(mMergedTracks);
         const auto &cl0 = mClusters[trk.FirstClusterRef()], &cln = mClusters[updTrk->FirstClusterRef() + updTrk->NClusters() - 1];
         const auto& GPUrestrict() cls = GetConstantMem()->ioPtrs.clustersNative->clustersLinear;
         float z0 = cls[cl0.num].getTime(), zn = cls[cln.num].getTime();
@@ -1946,14 +1939,7 @@ GPUd() void GPUTPCGMMerger::MergeLoopersMain(int32_t nBlocks, int32_t nThreads,
       const GPUTPCGMMergedTrack* trkI = &mMergedTracks[candidates[i].id];
       float refZI = candidates[i].refz;
       {
-        const auto* tmp = trkI;
-        while (tmp->PrevSegment() >= 0) {
-          const auto* next = &mMergedTracks[tmp->PrevSegment()];
-          if (next == trkI) {
-            break;
-          }
-          tmp = next;
-        }
+        const auto* tmp = trkI->GetFirstSegment(mMergedTracks);
         if (tmp != trkI && tmp->CSide() == trkI->CSide() && CAMath::Abs(tmp->GetParam().GetZ()) > CAMath::Abs(trkI->GetParam().GetZ())) {
           float tmpRefZ = refZI + tmp->GetParam().GetZ() - trkI->GetParam().GetZ();
           if (CAMath::Abs(tmpRefZ) < CAMath::Abs(candidates[j].refz) && CAMath::Abs(tmpRefZ) > CAMath::Abs(refZI)) {

GPU/GPUTracking/Merger/GPUTPCGMO2Output.cxx

@@ -212,14 +212,7 @@ GPUdii() void GPUTPCGMO2Output::Thread<GPUTPCGMO2Output::output>(int32_t nBlocks
     }
 
     if (track.PrevSegment() >= 0) {
-      const GPUTPCGMMergedTrack* chkTrk = &tracks[track.PrevSegment()];
-      while (chkTrk->PrevSegment() >= 0) {
-        auto next = &tracks[chkTrk->PrevSegment()];
-        if (next == &track) {
-          break;
-        }
-        chkTrk = next;
-      }
+      const GPUTPCGMMergedTrack* chkTrk = track.GetFirstSegment(tracks);
       const auto& firstPrevCluster = trackClusters[chkTrk->FirstClusterRef()];
       t1 = clusters->clustersLinear[firstPrevCluster.num].getTime();
       sector1 = firstPrevCluster.sector;