plugins/alpaka/PixelRecHitKernels.dev.cc

0001 // C++ headers
0002 #include <cassert>
0003 #include <cstdint>
0004 #include <type_traits>
0005
0006 // Alpaka headers
0007 #include <alpaka/alpaka.hpp>
0008
0009 // CMSSW headers
0010 #include "DataFormats/BeamSpot/interface/BeamSpotPOD.h"
0011 #include "DataFormats/SiPixelClusterSoA/interface/alpaka/SiPixelClustersSoACollection.h"
0012 #include "DataFormats/SiPixelDigiSoA/interface/alpaka/SiPixelDigisSoACollection.h"
0013 #include "DataFormats/TrackingRecHitSoA/interface/TrackingRecHitsSoA.h"
0014 #include "DataFormats/TrackingRecHitSoA/interface/alpaka/TrackingRecHitsSoACollection.h"
0015 #include "Geometry/CommonTopologies/interface/SimplePixelTopology.h"
0016 #include "HeterogeneousCore/AlpakaInterface/interface/HistoContainer.h"
0017 #include "HeterogeneousCore/AlpakaInterface/interface/config.h"
0018 #include "HeterogeneousCore/AlpakaInterface/interface/workdivision.h"
0019 #include "RecoLocalTracker/SiPixelRecHits/interface/pixelCPEforDevice.h"
0020
0021 // local headers
0022 #include "PixelRecHitKernel.h"
0023 #include "PixelRecHits.h"
0024
0025 //#define GPU_DEBUG
0026
0027 namespace ALPAKA_ACCELERATOR_NAMESPACE {
0028   using namespace cms::alpakatools;
0029   template <typename TrackerTraits>
0030   class setHitsLayerStart {
0031   public:
0032     template <typename TAcc, typename = std::enable_if_t<alpaka::isAccelerator<TAcc>>>
0033     ALPAKA_FN_ACC void operator()(TAcc const& acc,
0034                                   uint32_t const* __restrict__ hitsModuleStart,
0035                                   pixelCPEforDevice::ParamsOnDeviceT<TrackerTraits> const* __restrict__ cpeParams,
0036                                   uint32_t* __restrict__ hitsLayerStart) const {
0037       ALPAKA_ASSERT_ACC(0 == hitsModuleStart[0]);
0038
0039       for (int32_t i : cms::alpakatools::uniform_elements(acc, TrackerTraits::numberOfLayers + 1)) {
0040         hitsLayerStart[i] = hitsModuleStart[cpeParams->layerGeometry().layerStart[i]];
0041 #ifdef GPU_DEBUG
0042         int old = i == 0 ? 0 : hitsModuleStart[cpeParams->layerGeometry().layerStart[i - 1]];
0043         printf("LayerStart %d/%d at module %d: %d - %d\n",
0044                i,
0045                TrackerTraits::numberOfLayers,
0046                cpeParams->layerGeometry().layerStart[i],
0047                hitsLayerStart[i],
0048                hitsLayerStart[i] - old);
0049 #endif
0050       }
0051     }
0052   };
0053
0054   namespace pixelgpudetails {
0055
0056     template <typename TrackerTraits>
0057     TrackingRecHitsSoACollection<TrackerTraits> PixelRecHitKernel<TrackerTraits>::makeHitsAsync(
0058         SiPixelDigisSoACollection const& digis_d,
0059         SiPixelClustersSoACollection const& clusters_d,
0060         BeamSpotPOD const* bs_d,
0061         pixelCPEforDevice::ParamsOnDeviceT<TrackerTraits> const* cpeParams,
0062         Queue queue) const {
0063       using namespace pixelRecHits;
0064       auto nHits = clusters_d.nClusters();
0065       auto offsetBPIX2 = clusters_d.offsetBPIX2();
0066
0067       TrackingRecHitsSoACollection<TrackerTraits> hits_d(queue, nHits, offsetBPIX2, clusters_d->clusModuleStart());
0068
0069       int activeModulesWithDigis = digis_d.nModules();
0070
0071       // protect from empty events
0072       if (activeModulesWithDigis) {
0073         int threadsPerBlock = 128;
0074         // note: the kernel should work with an arbitrary number of blocks
0075         int blocks = activeModulesWithDigis;
0076         const auto workDiv1D = cms::alpakatools::make_workdiv<Acc1D>(blocks, threadsPerBlock);
0077
0078 #ifdef GPU_DEBUG
0079         std::cout << "launching GetHits kernel on " << alpaka::core::demangled<Acc1D> << " with " << blocks << " blocks"
0080                   << std::endl;
0081 #endif
0082         alpaka::exec<Acc1D>(queue,
0083                             workDiv1D,
0084                             GetHits<TrackerTraits>{},
0085                             cpeParams,
0086                             bs_d,
0087                             digis_d.view(),
0088                             digis_d.nDigis(),
0089                             digis_d.nModules(),
0090                             clusters_d.view(),
0091                             hits_d.view());
0092 #ifdef GPU_DEBUG
0093         alpaka::wait(queue);
0094 #endif
0095
0096         // assuming full warp of threads is better than a smaller number...
0097         if (nHits) {
0098           const auto workDiv1D = cms::alpakatools::make_workdiv<Acc1D>(1, 32);
0099           alpaka::exec<Acc1D>(queue,
0100                               workDiv1D,
0101                               setHitsLayerStart<TrackerTraits>{},
0102                               clusters_d->clusModuleStart(),
0103                               cpeParams,
0104                               hits_d.view().hitsLayerStart().data());
0105           constexpr auto nLayers = TrackerTraits::numberOfLayers;
0106
0107           // Use a view since it's runtime sized and can't use the implicit definition
0108           // see HeterogeneousCore/AlpakaInterface/interface/OneToManyAssoc.h:100
0109           typename TrackingRecHitSoA<TrackerTraits>::PhiBinnerView hrv_d;
0110           hrv_d.assoc = &(hits_d.view().phiBinner());
0111           hrv_d.offSize = -1;
0112           hrv_d.offStorage = nullptr;
0113           hrv_d.contentSize = nHits;
0114           hrv_d.contentStorage = hits_d.view().phiBinnerStorage();
0115
0116           cms::alpakatools::fillManyFromVector<Acc1D>(&(hits_d.view().phiBinner()),
0117                                                       hrv_d,
0118                                                       nLayers,
0119                                                       hits_d.view().iphi(),
0120                                                       hits_d.view().hitsLayerStart().data(),
0121                                                       nHits,
0122                                                       (uint32_t)256,
0123                                                       queue);
0124
0125 #ifdef GPU_DEBUG
0126           alpaka::wait(queue);
0127 #endif
0128         }
0129       }
0130
0131 #ifdef GPU_DEBUG
0132       alpaka::wait(queue);
0133       std::cout << "PixelRecHitKernel -> DONE!" << std::endl;
0134 #endif
0135
0136       return hits_d;
0137     }
0138
0139     template class PixelRecHitKernel<pixelTopology::Phase1>;
0140     template class PixelRecHitKernel<pixelTopology::Phase2>;
0141     template class PixelRecHitKernel<pixelTopology::HIonPhase1>;
0142
0143   }  // namespace pixelgpudetails
0144 }  // namespace ALPAKA_ACCELERATOR_NAMESPACE