SiPixelClusterizer/plugins/SiPixelRawToClusterCUDA.cc

0001 // C++ includes
0002 #include <memory>
0003 #include <string>
0004 #include <vector>
0005
0006 // CMSSW includes
0007 #include "CUDADataFormats/Common/interface/Product.h"
0008 #include "CUDADataFormats/SiPixelCluster/interface/SiPixelClustersCUDA.h"
0009 #include "CUDADataFormats/SiPixelDigi/interface/SiPixelDigiErrorsCUDA.h"
0010 #include "CUDADataFormats/SiPixelDigi/interface/SiPixelDigisCUDA.h"
0011 #include "CalibTracker/Records/interface/SiPixelGainCalibrationForHLTGPURcd.h"
0012 #include "CalibTracker/SiPixelESProducers/interface/SiPixelROCsStatusAndMappingWrapper.h"
0013 #include "CalibTracker/SiPixelESProducers/interface/SiPixelGainCalibrationForHLTGPU.h"
0014 #include "CondFormats/DataRecord/interface/SiPixelFedCablingMapRcd.h"
0015 #include "CondFormats/SiPixelObjects/interface/SiPixelFedCablingMap.h"
0016 #include "CondFormats/SiPixelObjects/interface/SiPixelFedCablingTree.h"
0017 #include "DataFormats/FEDRawData/interface/FEDNumbering.h"
0018 #include "DataFormats/FEDRawData/interface/FEDRawData.h"
0019 #include "DataFormats/FEDRawData/interface/FEDRawDataCollection.h"
0020 #include "EventFilter/SiPixelRawToDigi/interface/PixelDataFormatter.h"
0021 #include "EventFilter/SiPixelRawToDigi/interface/PixelUnpackingRegions.h"
0022 #include "FWCore/Framework/interface/ConsumesCollector.h"
0023 #include "FWCore/Framework/interface/ESHandle.h"
0024 #include "FWCore/Framework/interface/ESTransientHandle.h"
0025 #include "FWCore/Framework/interface/ESWatcher.h"
0026 #include "FWCore/Framework/interface/Event.h"
0027 #include "FWCore/Framework/interface/EventSetup.h"
0028 #include "FWCore/Framework/interface/MakerMacros.h"
0029 #include "FWCore/Framework/interface/stream/EDProducer.h"
0030 #include "FWCore/MessageLogger/interface/MessageLogger.h"
0031 #include "FWCore/ParameterSet/interface/ConfigurationDescriptions.h"
0032 #include "FWCore/ParameterSet/interface/ParameterSet.h"
0033 #include "FWCore/ParameterSet/interface/ParameterSetDescription.h"
0034 #include "FWCore/ServiceRegistry/interface/Service.h"
0035 #include "HeterogeneousCore/CUDACore/interface/ScopedContext.h"
0036 #include "HeterogeneousCore/CUDAServices/interface/CUDAService.h"
0037 #include "RecoTracker/Record/interface/CkfComponentsRecord.h"
0038
0039 // local includes
0040 #include "SiPixelClusterThresholds.h"
0041 #include "SiPixelRawToClusterGPUKernel.h"
0042
0043 class SiPixelRawToClusterCUDA : public edm::stream::EDProducer<edm::ExternalWork> {
0044 public:
0045   explicit SiPixelRawToClusterCUDA(const edm::ParameterSet& iConfig);
0046   ~SiPixelRawToClusterCUDA() override = default;
0047
0048   static void fillDescriptions(edm::ConfigurationDescriptions& descriptions);
0049
0050 private:
0051   void acquire(const edm::Event& iEvent,
0052                const edm::EventSetup& iSetup,
0053                edm::WaitingTaskWithArenaHolder waitingTaskHolder) override;
0054   void produce(edm::Event& iEvent, const edm::EventSetup& iSetup) override;
0055
0056   edm::EDGetTokenT<FEDRawDataCollection> rawGetToken_;
0057
0058   edm::EDPutTokenT<cms::cuda::Product<SiPixelDigisCUDA>> digiPutToken_;
0059   edm::EDPutTokenT<cms::cuda::Product<SiPixelDigiErrorsCUDA>> digiErrorPutToken_;
0060   edm::EDPutTokenT<cms::cuda::Product<SiPixelClustersCUDA>> clusterPutToken_;
0061
0062   cms::cuda::ContextState ctxState_;
0063
0064   edm::ESWatcher<SiPixelFedCablingMapRcd> recordWatcher_;
0065   edm::ESGetToken<SiPixelROCsStatusAndMappingWrapper, CkfComponentsRecord> gpuMapToken_;
0066   edm::ESGetToken<SiPixelGainCalibrationForHLTGPU, SiPixelGainCalibrationForHLTGPURcd> gainsToken_;
0067   edm::ESGetToken<SiPixelFedCablingMap, SiPixelFedCablingMapRcd> cablingMapToken_;
0068
0069   std::unique_ptr<SiPixelFedCablingTree> cabling_;
0070   std::vector<unsigned int> fedIds_;
0071   const SiPixelFedCablingMap* cablingMap_ = nullptr;
0072   std::unique_ptr<PixelUnpackingRegions> regions_;
0073
0074   pixelgpudetails::SiPixelRawToClusterGPUKernel gpuAlgo_;
0075   std::unique_ptr<pixelgpudetails::SiPixelRawToClusterGPUKernel::WordFedAppender> wordFedAppender_;
0076   PixelDataFormatter::Errors errors_;
0077
0078   const bool isRun2_;
0079   const bool includeErrors_;
0080   const bool useQuality_;
0081   const uint32_t maxFedWords_;
0082   uint32_t nDigis_;
0083   const SiPixelClusterThresholds clusterThresholds_;
0084 };
0085
0086 SiPixelRawToClusterCUDA::SiPixelRawToClusterCUDA(const edm::ParameterSet& iConfig)
0087     : rawGetToken_(consumes<FEDRawDataCollection>(iConfig.getParameter<edm::InputTag>("InputLabel"))),
0088       digiPutToken_(produces<cms::cuda::Product<SiPixelDigisCUDA>>()),
0089       clusterPutToken_(produces<cms::cuda::Product<SiPixelClustersCUDA>>()),
0090       gpuMapToken_(esConsumes<SiPixelROCsStatusAndMappingWrapper, CkfComponentsRecord>()),
0091       gainsToken_(esConsumes<SiPixelGainCalibrationForHLTGPU, SiPixelGainCalibrationForHLTGPURcd>()),
0092       cablingMapToken_(esConsumes<SiPixelFedCablingMap, SiPixelFedCablingMapRcd>(
0093           edm::ESInputTag("", iConfig.getParameter<std::string>("CablingMapLabel")))),
0094       isRun2_(iConfig.getParameter<bool>("isRun2")),
0095       includeErrors_(iConfig.getParameter<bool>("IncludeErrors")),
0096       useQuality_(iConfig.getParameter<bool>("UseQualityInfo")),
0097       maxFedWords_(iConfig.getParameter<uint32_t>("MaxFEDWords")),
0098       clusterThresholds_{iConfig.getParameter<int32_t>("clusterThreshold_layer1"),
0099                          iConfig.getParameter<int32_t>("clusterThreshold_otherLayers")} {
0100   if (includeErrors_) {
0101     digiErrorPutToken_ = produces<cms::cuda::Product<SiPixelDigiErrorsCUDA>>();
0102   }
0103
0104   // regions
0105   if (!iConfig.getParameter<edm::ParameterSet>("Regions").getParameterNames().empty()) {
0106     regions_ = std::make_unique<PixelUnpackingRegions>(iConfig, consumesCollector());
0107   }
0108
0109   edm::Service<CUDAService> cs;
0110   if (cs->enabled()) {
0111     wordFedAppender_ = std::make_unique<pixelgpudetails::SiPixelRawToClusterGPUKernel::WordFedAppender>(maxFedWords_);
0112   }
0113 }
0114
0115 void SiPixelRawToClusterCUDA::fillDescriptions(edm::ConfigurationDescriptions& descriptions) {
0116   edm::ParameterSetDescription desc;
0117   desc.add<bool>("isRun2", true);
0118   desc.add<bool>("IncludeErrors", true);
0119   desc.add<bool>("UseQualityInfo", false);
0120   desc.add<uint32_t>("MaxFEDWords", pixelgpudetails::MAX_FED * pixelgpudetails::MAX_WORD);
0121   desc.add<int32_t>("clusterThreshold_layer1", kSiPixelClusterThresholdsDefaultPhase1.layer1);
0122   desc.add<int32_t>("clusterThreshold_otherLayers", kSiPixelClusterThresholdsDefaultPhase1.otherLayers);
0123   desc.add<edm::InputTag>("InputLabel", edm::InputTag("rawDataCollector"));
0124   {
0125     edm::ParameterSetDescription psd0;
0126     psd0.addOptional<std::vector<edm::InputTag>>("inputs");
0127     psd0.addOptional<std::vector<double>>("deltaPhi");
0128     psd0.addOptional<std::vector<double>>("maxZ");
0129     psd0.addOptional<edm::InputTag>("beamSpot");
0130     desc.add<edm::ParameterSetDescription>("Regions", psd0)
0131         ->setComment("## Empty Regions PSet means complete unpacking");
0132   }
0133   desc.add<std::string>("CablingMapLabel", "")->setComment("CablingMap label");  //Tav
0134   descriptions.addWithDefaultLabel(desc);
0135 }
0136
0137 void SiPixelRawToClusterCUDA::acquire(const edm::Event& iEvent,
0138                                       const edm::EventSetup& iSetup,
0139                                       edm::WaitingTaskWithArenaHolder waitingTaskHolder) {
0140   cms::cuda::ScopedContextAcquire ctx{iEvent.streamID(), std::move(waitingTaskHolder), ctxState_};
0141
0142   auto hgpuMap = iSetup.getHandle(gpuMapToken_);
0143   if (hgpuMap->hasQuality() != useQuality_) {
0144     throw cms::Exception("LogicError")
0145         << "UseQuality of the module (" << useQuality_
0146         << ") differs the one from SiPixelROCsStatusAndMappingWrapper. Please fix your configuration.";
0147   }
0148   // get the GPU product already here so that the async transfer can begin
0149   const auto* gpuMap = hgpuMap->getGPUProductAsync(ctx.stream());
0150
0151   auto hgains = iSetup.getHandle(gainsToken_);
0152   // get the GPU product already here so that the async transfer can begin
0153   const auto* gpuGains = hgains->getGPUProductAsync(ctx.stream());
0154
0155   cms::cuda::device::unique_ptr<unsigned char[]> modulesToUnpackRegional;
0156   const unsigned char* gpuModulesToUnpack;
0157
0158   if (regions_) {
0159     regions_->run(iEvent, iSetup);
0160     LogDebug("SiPixelRawToCluster") << "region2unpack #feds: " << regions_->nFEDs();
0161     LogDebug("SiPixelRawToCluster") << "region2unpack #modules (BPIX,EPIX,total): " << regions_->nBarrelModules() << " "
0162                                     << regions_->nForwardModules() << " " << regions_->nModules();
0163     modulesToUnpackRegional = hgpuMap->getModToUnpRegionalAsync(*(regions_->modulesToUnpack()), ctx.stream());
0164     gpuModulesToUnpack = modulesToUnpackRegional.get();
0165   } else {
0166     gpuModulesToUnpack = hgpuMap->getModToUnpAllAsync(ctx.stream());
0167   }
0168
0169   // initialize cabling map or update if necessary
0170   if (recordWatcher_.check(iSetup)) {
0171     // cabling map, which maps online address (fed->link->ROC->local pixel) to offline (DetId->global pixel)
0172     auto cablingMap = iSetup.getTransientHandle(cablingMapToken_);
0173     cablingMap_ = cablingMap.product();
0174     fedIds_ = cablingMap->fedIds();
0175     cabling_ = cablingMap->cablingTree();
0176     LogDebug("map version:") << cabling_->version();
0177   }
0178
0179   const auto& buffers = iEvent.get(rawGetToken_);
0180
0181   errors_.clear();
0182
0183   // GPU specific: Data extraction for RawToDigi GPU
0184   unsigned int wordCounterGPU = 0;
0185   unsigned int fedCounter = 0;
0186   bool errorsInEvent = false;
0187
0188   // In CPU algorithm this loop is part of PixelDataFormatter::interpretRawData()
0189   ErrorChecker errorcheck;
0190   for (int fedId : fedIds_) {
0191     if (regions_ && !regions_->mayUnpackFED(fedId))
0192       continue;
0193
0194     // for GPU
0195     // first 150 index stores the fedId and next 150 will store the
0196     // start index of word in that fed
0197     assert(fedId >= FEDNumbering::MINSiPixeluTCAFEDID);
0198     fedCounter++;
0199
0200     // get event data for this fed
0201     const FEDRawData& rawData = buffers.FEDData(fedId);
0202
0203     // GPU specific
0204     int nWords = rawData.size() / sizeof(cms_uint64_t);
0205     if (nWords == 0) {
0206       continue;
0207     }
0208
0209     // check CRC bit
0210     const cms_uint64_t* trailer = reinterpret_cast<const cms_uint64_t*>(rawData.data()) + (nWords - 1);
0211     if (not errorcheck.checkCRC(errorsInEvent, fedId, trailer, errors_)) {
0212       continue;
0213     }
0214
0215     // check headers
0216     const cms_uint64_t* header = reinterpret_cast<const cms_uint64_t*>(rawData.data());
0217     header--;
0218     bool moreHeaders = true;
0219     while (moreHeaders) {
0220       header++;
0221       bool headerStatus = errorcheck.checkHeader(errorsInEvent, fedId, header, errors_);
0222       moreHeaders = headerStatus;
0223     }
0224
0225     // check trailers
0226     bool moreTrailers = true;
0227     trailer++;
0228     while (moreTrailers) {
0229       trailer--;
0230       bool trailerStatus = errorcheck.checkTrailer(errorsInEvent, fedId, nWords, trailer, errors_);
0231       moreTrailers = trailerStatus;
0232     }
0233
0234     const cms_uint32_t* bw = (const cms_uint32_t*)(header + 1);
0235     const cms_uint32_t* ew = (const cms_uint32_t*)(trailer);
0236
0237     assert(0 == (ew - bw) % 2);
0238     wordFedAppender_->initializeWordFed(fedId, wordCounterGPU, bw, (ew - bw));
0239     wordCounterGPU += (ew - bw);
0240
0241   }  // end of for loop
0242
0243   nDigis_ = wordCounterGPU;
0244
0245   if (nDigis_ == 0)
0246     return;
0247
0248   gpuAlgo_.makeClustersAsync(isRun2_,
0249                              clusterThresholds_,
0250                              gpuMap,
0251                              gpuModulesToUnpack,
0252                              gpuGains,
0253                              *wordFedAppender_,
0254                              std::move(errors_),
0255                              wordCounterGPU,
0256                              fedCounter,
0257                              maxFedWords_,
0258                              useQuality_,
0259                              includeErrors_,
0260                              edm::MessageDrop::instance()->debugEnabled,
0261                              ctx.stream());
0262 }
0263
0264 void SiPixelRawToClusterCUDA::produce(edm::Event& iEvent, const edm::EventSetup& iSetup) {
0265   cms::cuda::ScopedContextProduce ctx{ctxState_};
0266
0267   if (nDigis_ == 0) {
0268     // default construct collections and place them in event
0269     auto tmp = std::make_pair(SiPixelDigisCUDA{}, SiPixelClustersCUDA{});
0270     ctx.emplace(iEvent, digiPutToken_, std::move(tmp.first));
0271     ctx.emplace(iEvent, clusterPutToken_, std::move(tmp.second));
0272     if (includeErrors_) {
0273       ctx.emplace(iEvent, digiErrorPutToken_, SiPixelDigiErrorsCUDA{});
0274     }
0275     return;
0276   }
0277
0278   auto tmp = gpuAlgo_.getResults();
0279   ctx.emplace(iEvent, digiPutToken_, std::move(tmp.first));
0280   ctx.emplace(iEvent, clusterPutToken_, std::move(tmp.second));
0281   if (includeErrors_) {
0282     ctx.emplace(iEvent, digiErrorPutToken_, gpuAlgo_.getErrors());
0283   }
0284 }
0285
0286 // define as framework plugin
0287 DEFINE_FWK_MODULE(SiPixelRawToClusterCUDA);