ALPAKA_ACCELERATOR_NAMESPACE::TopoClusterContraction Class Reference

Public Member Functions
template<typename TAcc , typename = std::enable_if_t<alpaka::isAccelerator<TAcc>>>
ALPAKA_FN_ACC void	operator() (const TAcc &acc, const reco::PFRecHitHostCollection::ConstView pfRecHits, reco::PFClusteringVarsDeviceCollection::View pfClusteringVars, reco::PFClusterDeviceCollection::View clusterView, uint32_t *__restrict__ nSeeds, uint32_t *__restrict__ nRHF) const

Detailed Description

Definition at line 1194 of file PFClusterSoAProducerKernel.dev.cc.

Member Function Documentation

operator()()

template<typename TAcc , typename = std::enable_if_t<alpaka::isAccelerator<TAcc>>>

ALPAKA_FN_ACC void ALPAKA_ACCELERATOR_NAMESPACE::TopoClusterContraction::operator() ( const TAcc &  acc, const reco::PFRecHitHostCollection::ConstView  pfRecHits, reco::PFClusteringVarsDeviceCollection::View  pfClusteringVars, reco::PFClusterDeviceCollection::View  clusterView, uint32_t *__restrict__  nSeeds, uint32_t *__restrict__  nRHF  ) const

inline

Definition at line 1197 of file PFClusterSoAProducerKernel.dev.cc.

References cms::cudacompat::atomicAdd(), dqmdumpme::k, hltrates_dqm_sourceclient-live_cfg::offset, cms::alpakatools::once_per_block(), and HLT_2024v14_cff::pfRecHits.

                                                                     {
      const int nRH = pfRecHits.size();
      int& totalSeedOffset = alpaka::declareSharedVar<int, __COUNTER__>(acc);
      int& totalSeedFracOffset = alpaka::declareSharedVar<int, __COUNTER__>(acc);

      // rhCount, topoRHCount, topoSeedCount initialized earlier
      if (once_per_block(acc)) {
        pfClusteringVars.nTopos() = 0;
        pfClusteringVars.nRHFracs() = 0;
        totalSeedOffset = 0;
        totalSeedFracOffset = 0;
        pfClusteringVars.pcrhFracSize() = 0;
      }

      alpaka::syncBlockThreads(acc);  // all threads call sync

      // Now determine the number of seeds and rechits in each topo cluster [topoRHCount, topoSeedCount]
      // Also get the list of topoIds (smallest rhIdx of each topo cluser)
      for (int rhIdx = alpaka::getIdx<alpaka::Block, alpaka::Threads>(acc)[0u]; rhIdx < nRH;
           rhIdx += alpaka::getWorkDiv<alpaka::Block, alpaka::Threads>(acc)[0u]) {
        pfClusteringVars[rhIdx].rhIdxToSeedIdx() = -1;
        int topoId = pfClusteringVars[rhIdx].pfrh_topoId();
        if (topoId > -1) {
          // Valid topo cluster
          alpaka::atomicAdd(acc, &pfClusteringVars[topoId].topoRHCount(), 1);
          // Valid topoId not counted yet
          if (topoId == rhIdx) {  // For every topo cluster, there is one rechit that meets this condition.
            int topoIdx = alpaka::atomicAdd(acc, &pfClusteringVars.nTopos(), 1);
            pfClusteringVars[topoIdx].topoIds() =
                topoId;  // topoId: the smallest index of rechits that belong to a topo cluster.
          }
          // This is a cluster seed
          if (pfClusteringVars[rhIdx].pfrh_isSeed()) {  // # of seeds in this topo cluster
            alpaka::atomicAdd(acc, &pfClusteringVars[topoId].topoSeedCount(), 1);
          }
        }
      }

      alpaka::syncBlockThreads(acc);  // all threads call sync

      // Determine offsets for topo ID seed array [topoSeedOffsets]
      for (int topoId = alpaka::getIdx<alpaka::Block, alpaka::Threads>(acc)[0u]; topoId < nRH;
           topoId += alpaka::getWorkDiv<alpaka::Block, alpaka::Threads>(acc)[0u]) {
        if (pfClusteringVars[topoId].topoSeedCount() > 0) {
          // This is a valid topo ID
          int offset = alpaka::atomicAdd(acc, &totalSeedOffset, pfClusteringVars[topoId].topoSeedCount());
          pfClusteringVars[topoId].topoSeedOffsets() = offset;
        }
      }
      alpaka::syncBlockThreads(acc);  // all threads call sync

      // Fill arrays of rechit indicies for each seed [topoSeedList] and rhIdx->seedIdx conversion for each seed [rhIdxToSeedIdx]
      // Also fill seedRHIdx, topoId, depth
      for (int rhIdx = alpaka::getIdx<alpaka::Block, alpaka::Threads>(acc)[0u]; rhIdx < nRH;
           rhIdx += alpaka::getWorkDiv<alpaka::Block, alpaka::Threads>(acc)[0u]) {
        int topoId = pfClusteringVars[rhIdx].pfrh_topoId();
        if (pfClusteringVars[rhIdx].pfrh_isSeed()) {
          // Valid topo cluster and this rhIdx corresponds to a seed
          int k = alpaka::atomicAdd(acc, &pfClusteringVars[topoId].rhCount(), 1);
          int seedIdx = pfClusteringVars[topoId].topoSeedOffsets() + k;
          if ((unsigned int)seedIdx >= *nSeeds)
            printf("Warning(contraction) %8d > %8d should not happen, check topoId: %d has %d rh\n",
                   seedIdx,
                   *nSeeds,
                   topoId,
                   k);
          pfClusteringVars[seedIdx].topoSeedList() = rhIdx;
          pfClusteringVars[rhIdx].rhIdxToSeedIdx() = seedIdx;
          clusterView[seedIdx].topoId() = topoId;
          clusterView[seedIdx].seedRHIdx() = rhIdx;
          clusterView[seedIdx].depth() = pfRecHits[rhIdx].depth();
        }
      }

      alpaka::syncBlockThreads(acc);  // all threads call sync

      // Determine seed offsets for rechit fraction array
      for (int rhIdx = alpaka::getIdx<alpaka::Block, alpaka::Threads>(acc)[0u]; rhIdx < nRH;
           rhIdx += alpaka::getWorkDiv<alpaka::Block, alpaka::Threads>(acc)[0u]) {
        pfClusteringVars[rhIdx].rhCount() = 1;  // Reset this counter array

        int topoId = pfClusteringVars[rhIdx].pfrh_topoId();
        if (pfClusteringVars[rhIdx].pfrh_isSeed() && topoId > -1) {
          // Allot the total number of rechits for this topo cluster for rh fractions
          int offset = alpaka::atomicAdd(acc, &totalSeedFracOffset, pfClusteringVars[topoId].topoRHCount());

          // Add offset for this PF cluster seed
          pfClusteringVars[rhIdx].seedFracOffsets() = offset;

          // Store recHitFraction offset & size information for each seed
          clusterView[pfClusteringVars[rhIdx].rhIdxToSeedIdx()].rhfracOffset() =
              pfClusteringVars[rhIdx].seedFracOffsets();
          clusterView[pfClusteringVars[rhIdx].rhIdxToSeedIdx()].rhfracSize() =
              pfClusteringVars[topoId].topoRHCount() - pfClusteringVars[topoId].topoSeedCount() + 1;
        }
      }

      alpaka::syncBlockThreads(acc);  // all threads call sync

      if (once_per_block(acc)) {
        pfClusteringVars.pcrhFracSize() = totalSeedFracOffset;
        pfClusteringVars.nRHFracs() = totalSeedFracOffset;
        clusterView.nRHFracs() = totalSeedFracOffset;
        *nRHF = totalSeedFracOffset;
        clusterView.nSeeds() = *nSeeds;
        clusterView.nTopos() = pfClusteringVars.nTopos();

        if (pfClusteringVars.pcrhFracSize() > 200000)  // Warning in case the fraction is too large
          printf("At the end of topoClusterContraction, found large *pcrhFracSize = %d\n",
                 pfClusteringVars.pcrhFracSize());
      }
    }