GPU TPC Compression: Minor optimization to get rid of one synchronization point

Author: davidrohr

GPU/GPUTracking/DataCompression/GPUTPCCompressionKernels.cxx

@@ -242,17 +242,17 @@ GPUdii() void GPUTPCCompressionKernels::Thread<GPUTPCCompressionKernels::step1un
         storeCluster = 1;
       } while (false);
 
+      const uint32_t currentCount = smem.nCount;
       GPUbarrier();
       int32_t myIndex = work_group_scan_inclusive_add(storeCluster);
       int32_t storeLater = -1;
       if (storeCluster) {
-        if (smem.nCount + myIndex <= GPUCA_TPC_COMP_CHUNK_SIZE) {
-          sortBuffer[smem.nCount + myIndex - 1] = i;
+        if (currentCount + myIndex <= GPUCA_TPC_COMP_CHUNK_SIZE) {
+          sortBuffer[currentCount + myIndex - 1] = i;
         } else {
-          storeLater = smem.nCount + myIndex - 1 - GPUCA_TPC_COMP_CHUNK_SIZE;
+          storeLater = currentCount + myIndex - 1 - GPUCA_TPC_COMP_CHUNK_SIZE;
         }
       }
-      GPUbarrier();
       if (iThread == nThreads - 1) {
         smem.nCount += myIndex;
       }