GPU TPC: Fix handling of ce-crossing looping tracks

Author: davidrohr

GPU/GPUTracking/Merger/GPUTPCGMMerger.cxx

@@ -1361,6 +1361,21 @@ GPUd() void GPUTPCGMMerger::MergeCE(int32_t nBlocks, int32_t nThreads, int32_t i
         GPUCommonAlgorithm::swap(trk[0], trk[1]);
       }
 
+      if (Param().par.continuousTracking) {
+        GPUTPCGMMergedTrackHit* clsmax;
+        const float tmax = CAMath::MaxWithRef(cls[mClusters[trk[0]->FirstClusterRef()].num].getTime(), cls[mClusters[trk[0]->FirstClusterRef() + trk[0]->NClusters() - 1].num].getTime(),
+                                              cls[mClusters[trk[1]->FirstClusterRef()].num].getTime(), cls[mClusters[trk[1]->FirstClusterRef() + trk[1]->NClusters() - 1].num].getTime(),
+                                              &mClusters[trk[0]->FirstClusterRef()], &mClusters[trk[0]->FirstClusterRef() + trk[0]->NClusters() - 1],
+                                              &mClusters[trk[1]->FirstClusterRef()], &mClusters[trk[1]->FirstClusterRef() + trk[1]->NClusters() - 1], clsmax);
+        const float offset = CAMath::Max(tmax - mConstantMem->calibObjects.fastTransformHelper->getCorrMap()->getMaxDriftTime(clsmax->sector, clsmax->row, cls[clsmax->num].getPad()), 0.f);
+        trk[1]->Param().Z() += mConstantMem->calibObjects.fastTransformHelper->getCorrMap()->convDeltaTimeToDeltaZinTimeFrame(trk[1]->CSide() * NSECTORS / 2, trk[1]->Param().TOffset() - offset);
+        trk[1]->Param().TOffset() = offset;
+        if (celooper) {
+          trk[0]->Param().Z() += mConstantMem->calibObjects.fastTransformHelper->getCorrMap()->convDeltaTimeToDeltaZinTimeFrame(trk[0]->CSide() * NSECTORS / 2, trk[0]->Param().TOffset() - offset);
+          trk[0]->Param().TOffset() = offset;
+        }
+      }
+
       if (celooper) { // TODO: Need propper handling, avoid falsely flagging the primary leg as looper
         trk[0]->SetMergedLooperConnected(true);
         trk[0]->SetCCE(true);
@@ -1382,16 +1397,6 @@ GPUd() void GPUTPCGMMerger::MergeCE(int32_t nBlocks, int32_t nThreads, int32_t i
         return;
       }
 
-      if (Param().par.continuousTracking) {
-        GPUTPCGMMergedTrackHit* clsmax;
-        const float tmax = CAMath::MaxWithRef(cls[mClusters[trk[0]->FirstClusterRef()].num].getTime(), cls[mClusters[trk[0]->FirstClusterRef() + trk[0]->NClusters() - 1].num].getTime(),
-                                              cls[mClusters[trk[1]->FirstClusterRef()].num].getTime(), cls[mClusters[trk[1]->FirstClusterRef() + trk[1]->NClusters() - 1].num].getTime(),
-                                              &mClusters[trk[0]->FirstClusterRef()], &mClusters[trk[0]->FirstClusterRef() + trk[0]->NClusters() - 1],
-                                              &mClusters[trk[1]->FirstClusterRef()], &mClusters[trk[1]->FirstClusterRef() + trk[1]->NClusters() - 1], clsmax);
-        const float offset = CAMath::Max(tmax - mConstantMem->calibObjects.fastTransformHelper->getCorrMap()->getMaxDriftTime(clsmax->sector, clsmax->row, cls[clsmax->num].getPad()), 0.f);
-        trk[1]->Param().Z() += mConstantMem->calibObjects.fastTransformHelper->getCorrMap()->convDeltaTimeToDeltaZinTimeFrame(trk[1]->CSide() * NSECTORS / 2, trk[1]->Param().TOffset() - offset);
-        trk[1]->Param().TOffset() = offset;
-      }
       int32_t pos = newRef;
 #pragma unroll
       for (int32_t k = 1; k >= 0; k--) {