GPU TPC: Fix handling of ce-crossing looping tracks

Author: davidrohr

GPU/GPUTracking/Merger/GPUTPCGMMerger.cxx

@@ -1319,17 +1319,6 @@ GPUd() void GPUTPCGMMerger::MergeCE(int32_t nBlocks, int32_t nThreads, int32_t i
         continue;
       }
 
-      uint32_t newRef = CAMath::AtomicAdd(&mMemory->nMergedTrackClusters, trk[0]->NClusters() + trk[1]->NClusters());
-      if (newRef + trk[0]->NClusters() + trk[1]->NClusters() >= mNMaxMergedTrackClusters) {
-        raiseError(GPUErrors::ERROR_MERGER_CE_HIT_OVERFLOW, newRef + trk[0]->NClusters() + trk[1]->NClusters(), mNMaxMergedTrackClusters);
-        for (uint32_t k = newRef; k < mNMaxMergedTrackClusters; k++) {
-          mClusters[k].num = 0;
-          mClusters[k].state = 0;
-        }
-        CAMath::AtomicExch(&mMemory->nMergedTrackClusters, mNMaxMergedTrackClusters);
-        return;
-      }
-
       bool needswap = false;
       if (celooper) {
         const float z0max = -CAMath::Min(cls[mClusters[trk[0]->FirstClusterRef()].num].getTime(), cls[mClusters[trk[0]->FirstClusterRef() + trk[0]->NClusters() - 1].num].getTime());
@@ -1346,15 +1335,6 @@ GPUd() void GPUTPCGMMerger::MergeCE(int32_t nBlocks, int32_t nThreads, int32_t i
         GPUCommonAlgorithm::swap(trk[0], trk[1]);
       }
 
-      if (celooper) {
-        trk[0]->SetMergedLooperConnected(true);
-        trk[0]->SetCCE(true);
-        trk[0]->SetLooper(true);
-        trk[1]->SetCCE(true);
-        trk[1]->SetLooper(true);
-        continue;
-      }
-
       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(),
@@ -1364,7 +1344,32 @@ GPUd() void GPUTPCGMMerger::MergeCE(int32_t nBlocks, int32_t nThreads, int32_t i
         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) {
+        trk[0]->SetMergedLooperConnected(true);
+        trk[0]->SetCCE(true);
+        trk[0]->SetLooper(true);
+        trk[1]->SetCCE(true);
+        trk[1]->SetLooper(true);
+        continue;
+      }
+
+      uint32_t newRef = CAMath::AtomicAdd(&mMemory->nMergedTrackClusters, trk[0]->NClusters() + trk[1]->NClusters());
+      if (newRef + trk[0]->NClusters() + trk[1]->NClusters() >= mNMaxMergedTrackClusters) {
+        raiseError(GPUErrors::ERROR_MERGER_CE_HIT_OVERFLOW, newRef + trk[0]->NClusters() + trk[1]->NClusters(), mNMaxMergedTrackClusters);
+        for (uint32_t k = newRef; k < mNMaxMergedTrackClusters; k++) {
+          mClusters[k].num = 0;
+          mClusters[k].state = 0;
+        }
+        CAMath::AtomicExch(&mMemory->nMergedTrackClusters, mNMaxMergedTrackClusters);
+        return;
+      }
+
       int32_t pos = newRef;
 #pragma unroll
       for (int32_t k = 1; k >= 0; k--) {