GPU TPC clusterizer: wait for the gather of the previous sector before clearing the maps

A lane reuses its scratch memory for its next sector. At the first
fragment the charge and peak maps were cleared without waiting for the
gather of the previous sector, which runs on another stream, and the maps
can overlap that sector's clusterByRow: on pbpb30 the peak map of sector 3
ends at row 0 of sector 0. When the gather starts late, row 0 of the
previous sector is gathered as zeros: on a 20 kHz Pb-Pb time frame on
Metal, the first run in a process found 0.42 % fewer sector tracks.

The existing wait on the gather sat on the clusterInRow clear, after the
map clears. Move it to the first map clear at fragment 0.

Author: ktf

GPU/GPUTracking/Global/GPUChainTrackingClusterizer.cxx

@@ -1038,7 +1038,12 @@ int32_t GPUChainTracking::RunTPCClusterizer(bool synchronizeOutput)
 
         using ChargeMapType = decltype(*clustererShadow.mPchargeMap);
         using PeakMapType = decltype(*clustererShadow.mPpeakMap);
-        runKernel<GPUMemClean16>({GetGridAutoStep(lane, RecoStep::TPCClusterFinding)}, clustererShadow.mPchargeMap, TPCMapMemoryLayout<ChargeMapType>::items(GetProcessingSettings().overrideClusterizerFragmentLen) * sizeof(ChargeMapType));
+        deviceEvent* waitEvent = nullptr;
+        if (frag.index == 0 && transferRunning[lane] == 1) { // The gather of the previous sector of this lane may still read the scratch memory the maps are cleared in
+          waitEvent = &mEvents->stream[lane];
+          transferRunning[lane] = 2;
+        }
+        runKernel<GPUMemClean16>({GetGridAutoStep(lane, RecoStep::TPCClusterFinding), krnlRunRangeNone, {nullptr, waitEvent}}, clustererShadow.mPchargeMap, TPCMapMemoryLayout<ChargeMapType>::items(GetProcessingSettings().overrideClusterizerFragmentLen) * sizeof(ChargeMapType));
         runKernel<GPUMemClean16>({GetGridAutoStep(lane, RecoStep::TPCClusterFinding)}, clustererShadow.mPpeakMap, TPCMapMemoryLayout<PeakMapType>::items(GetProcessingSettings().overrideClusterizerFragmentLen) * sizeof(PeakMapType));
         if (frag.index == 0) {
           runKernel<GPUMemClean16>({GetGridAutoStep(lane, RecoStep::TPCClusterFinding)}, clustererShadow.mPpadIsNoisy, TPC_CLUSTERER_STRIDED_PAD_COUNT * sizeof(*clustererShadow.mPpadIsNoisy));
@@ -1211,12 +1216,7 @@ int32_t GPUChainTracking::RunTPCClusterizer(bool synchronizeOutput)
         }
 
         if (resetClusterCounters) {
-          deviceEvent* waitEvent = nullptr;
-          if (transferRunning[lane] == 1) {
-            waitEvent = &mEvents->stream[lane];
-            transferRunning[lane] = 2;
-          }
-          runKernel<GPUMemClean16>({GetGridAutoStep(lane, RecoStep::TPCClusterFinding), krnlRunRangeNone, {nullptr, waitEvent}}, clustererShadow.mPclusterInRow, GPUTPCGeometry::NROWS * sizeof(*clustererShadow.mPclusterInRow));
+          runKernel<GPUMemClean16>({GetGridAutoStep(lane, RecoStep::TPCClusterFinding)}, clustererShadow.mPclusterInRow, GPUTPCGeometry::NROWS * sizeof(*clustererShadow.mPclusterInRow));
         }
 
         if (clusterer.mPmemory->counters.nPositions == 0) {