Cleanup.

Author: fweig

GPU/Common/GPUCommonAlgorithm.h

@@ -422,7 +422,7 @@ GPUdi() T warp_broadcast_FUNC(T v, int32_t i)
 #endif
 }
 
-[[nodiscard]] GPUdi() bool work_group_any(bool pred, int32_t &smemBcast)
+[[nodiscard]] GPUdi() bool work_group_any(bool pred)
 {
   return __syncthreads_or(pred);
 }

GPU/GPUTracking/TPCClusterFinder/GPUTPCCFCheckPadBaseline.cxx

@@ -97,11 +97,6 @@ GPUd() void GPUTPCCFCheckPadBaseline::CheckBaselineGPU(int32_t nBlocks, int32_t
 
   HipTailRange activeHIPTail{-1, -1};
 
-  if (iThread == 0) {
-    smem.bcast = 0;
-  }
-  GPUbarrier();
-
   const int16_t iPadOffset = iThread % MaxNPadsPerRow;
   const int16_t iTimeOffset = iThread / MaxNPadsPerRow;
   const int16_t iPadHandle = iThread;
@@ -119,13 +114,7 @@ GPUd() void GPUTPCCFCheckPadBaseline::CheckBaselineGPU(int32_t nBlocks, int32_t
     const Charge ql = iTimeLoad < lastTB && iPadOffset < rowinfo.nPads ? chargeMap[pos].unpack() : 0;
     smem.charges[iTimeOffset][iPadOffset] = ql;
 
-    const bool hasHIPTrigger = work_group_any(ql == MaxADC, smem.bcast);
-    // bool hasHIPTrigger = false;
-    // if (t % (NumOfCachedTBs * 100) == 0) {
-    //   hasHIPTrigger = work_group_any(ql == MaxADC);
-    // } else {
-    //   GPUbarrier();
-    // }
+    const bool hasHIPTrigger = work_group_any(ql == MaxADC);
 
     acc.HIPtb = -1;
 
@@ -152,9 +141,6 @@ GPUd() void GPUTPCCFCheckPadBaseline::CheckBaselineGPU(int32_t nBlocks, int32_t
 
     }
 
-    if (iThread == 0) {
-      smem.bcast = 0;
-    }
     GPUbarrier();
 
     if (hasHIPTrigger) [[unlikely]] {
@@ -237,7 +223,7 @@ GPUd() void GPUTPCCFCheckPadBaseline::CheckBaselineGPU(int32_t nBlocks, int32_t
   // Call `work_group_any` here, instead of always counting.
   // This is important as `work_group_count` is a lot slower
   // and has a lot of overhead if no HIPs were found.
-  const bool anyCloseTail = work_group_any(shouldCloseTail, smem.bcast);
+  const bool anyCloseTail = work_group_any(shouldCloseTail);
 
   if (anyCloseTail) {
 

GPU/GPUTracking/TPCClusterFinder/GPUTPCCFCheckPadBaseline.h

@@ -77,8 +77,6 @@ class GPUTPCCFCheckPadBaseline : public GPUKernelTemplate
     HipTailRange tails[MaxNPadsPerRow];
     uint8_t tailsClosedPad[MaxNPadsPerRow];
     HipTailRange tailsClosed[MaxNPadsPerRow];
-
-    int32_t bcast;
   };
 
   // Accumulated values from scanning cached charges in a pad