Adjusting launch bounds

Author: ChSonnabend

GPU/GPUTracking/Definitions/Parameters/GPUParameters.csv

@@ -76,7 +76,7 @@ GPUMemClean16,"[""GPUCA_THREAD_COUNT_DEFAULT"", 1]",,,,,,,,,,,,,,,
 GPUitoa,"[""GPUCA_THREAD_COUNT_DEFAULT"", 1]",,,,,,,,,,,,,,,
 GPUTPCCFNoiseSuppression_noiseSuppression,"""GPUCA_LB_GPUTPCCFNoiseSuppression""",,,,,,,,,,,,,,,448
 GPUTPCCFNoiseSuppression_updatePeaks,"""GPUCA_LB_GPUTPCCFNoiseSuppression""",,,,,,,,,,,,,,,448
-GPUTPCNNClusterizerKernels_runCfClusterizer,"""GPUCA_LB_GPUTPCNNClusterizerKernels""",,,,,,,,,,,,,,,
+GPUTPCNNClusterizerKernels_runCfClusterizer,"""GPUCA_LB_GPUTPCCFClusterizer""",,,,,,,,,,,,,,,
 GPUTPCNNClusterizerKernels_fillInputNNCPU,"""GPUCA_LB_GPUTPCNNClusterizerKernels""",,,,,,,,,,,,,,,
 GPUTPCNNClusterizerKernels_fillInputNNGPU,1024,,,,,,,,,,,,,,,
 GPUTPCNNClusterizerKernels_determineClass1Labels,"""GPUCA_LB_GPUTPCNNClusterizerKernels""",,,,,,,,,,,,,,,

GPU/GPUTracking/TPCClusterFinder/GPUTPCNNClusterizerKernels.cxx

@@ -36,6 +36,8 @@ using namespace o2::gpu::tpccf;
 #include "GPUTPCCFClusterizer.inc"
 #endif
 
+static_assert(GPUTPCNNClusterizerKernels::SCRATCH_PAD_WORK_GROUP_SIZE == GPUTPCCFClusterizer::SCRATCH_PAD_WORK_GROUP_SIZE, "Work group sizes do not match");
+
 // Defining individual thread functions for data filling, determining the class label and running the CF clusterizer
 template <>
 GPUdii() void GPUTPCNNClusterizerKernels::Thread<GPUTPCNNClusterizerKernels::runCfClusterizer>(int32_t nBlocks, int32_t nThreads, int32_t iBlock, int32_t iThread, GPUSharedMemory& smem, processorType& processors, uint8_t sector, int8_t dtype, int8_t withMC, uint32_t batchStart)

GPU/GPUTracking/TPCClusterFinder/GPUTPCNNClusterizerKernels.h

@@ -37,7 +37,7 @@ class GPUTPCNNClusterizerKernels : public GPUKernelTemplate
 {
  public:
   // Must all have same number of threads, since they use a common SCRATCH_PAD_WORK_GROUP_SIZE below
-  static constexpr size_t SCRATCH_PAD_WORK_GROUP_SIZE = GPUCA_GET_THREAD_COUNT(GPUCA_LB_GPUTPCNNClusterizerKernels_runCfClusterizer);
+  static constexpr size_t SCRATCH_PAD_WORK_GROUP_SIZE = GPUCA_GET_THREAD_COUNT(GPUCA_LB_GPUTPCCFClusterizer);
   using GPUSharedMemory = GPUTPCCFClusterizer::GPUSharedMemory;
 
   GPUhdi() constexpr static gpudatatypes::RecoStep GetRecoStep()