Simplify HIP cluster creation

Author: fweig

GPU/GPUTracking/TPCClusterFinder/GPUTPCCFCheckPadBaseline.cxx

@@ -519,28 +519,21 @@ GPUd() void GPUTPCCFHIPClusterizer::Thread<0>(int32_t nBlocks, int32_t nThreads,
 
   for (uint32_t iTail = iThread + 1; iTail <= nTails; iTail += nThreads) {
 
-    auto* tail = &tails[iTail];
-
+    const auto* tail = &tails[iTail];
     if (tail->iPrev != 0) {
       continue;
     }
 
-    float qTot = tail->qTot;
-    float qMax = tail->qMax;
-    const float firstWeight = tail->qTot;
-    const float firstPad = tail->pad;
-    const float firstTime = HIPTailTimeMean(*tail);
-    float padSum = firstWeight * firstPad;
-    float padSqSum = firstWeight * firstPad * firstPad;
-    float timeSum = firstWeight * firstTime;
-
-    uint32_t tailStart = tail->tailStart;
-    uint32_t tailEnd = tail->tailEnd;
-
-    while (tail->iNext != 0) {
-
-      tail = &tails[tail->iNext];
+    float qTot = 0;
+    float qMax = 0;
+    float padSum = 0;
+    float padSqSum = 0;
+    float timeSum = 0;
+    uint32_t tailStart = (uint32_t)-1;
+    uint32_t tailEnd = 0;
 
+    // Zero-th element is empty tail
+    for (; tail != tails; tail = &tails[tail->iNext]) {
       const float tailWeight = tail->qTot;
       const float tailPad = tail->pad;
       const float tailTime = HIPTailTimeMean(*tail);
@@ -554,9 +547,9 @@ GPUd() void GPUTPCCFHIPClusterizer::Thread<0>(int32_t nBlocks, int32_t nThreads,
     }
 
     const float weightSum = CAMath::Max(qTot, 1.f);
-    float padMean = padSum / weightSum;
-    float timeMean = timeSum / weightSum; // TODO: Use timebin of saturated signal instead! Time mean is biased for long tails.
-    float padSigma = CAMath::Sqrt(CAMath::Max(0.f, padSqSum / weightSum - padMean * padMean));
+    const float padMean = padSum / weightSum;
+    const float timeMean = timeSum / weightSum; // TODO: Use timebin of saturated signal instead! Time mean is biased for long tails.
+    const float padSigma = CAMath::Sqrt(CAMath::Max(0.f, padSqSum / weightSum - padMean * padMean));
 
     tpc::ClusterNative cn;
     cn.qMax = qMax;