TrigInDetTrackSeedingTool.cxx

Go to the documentation of this file.

/*
  Copyright (C) 2002-2025 CERN for the benefit of the ATLAS collaboration
*/
 
#include "InDetIdentifier/SCT_ID.h"
#include "InDetIdentifier/PixelID.h" 
 
#include "TrkSpacePoint/SpacePoint.h"
#include "TrkSpacePoint/SpacePointCollection.h"
#include "TrkSpacePoint/SpacePointContainer.h"
#include "InDetPrepRawData/PixelCluster.h"
#include "AtlasDetDescr/AtlasDetectorID.h"
 
#include "PathResolver/PathResolver.h"
 
#include "IRegionSelector/IRegSelTool.h"
 
#include "TrigInDetTrackSeedingTool.h"
 
#include "GNN_TrackingFilter.h"
 
//for GPU offloading
 
#include "TrigAccelEvent/TrigITkAccelEDM.h"
#include "TrigAccelEvent/TrigInDetAccelCodes.h"
 
TrigInDetTrackSeedingTool::TrigInDetTrackSeedingTool(const std::string& t, 
                         const std::string& n,
                         const IInterface*  p ) : SeedingToolBase(t,n,p)
{
 
}
 
StatusCode TrigInDetTrackSeedingTool::initialize() {
 
  ATH_CHECK(SeedingToolBase::initialize());
  
  ATH_CHECK(m_regsel_pix.retrieve());
  ATH_CHECK(m_regsel_sct.retrieve());
 
  ATH_CHECK(m_beamSpotKey.initialize());
 
  if(m_useGPU) {//for GPU offloading
    ATH_CHECK(m_accelSvc.retrieve());
    ATH_CHECK(m_accelSvc->isReady());
  }
 
  if (!m_usePixelSpacePoints && !m_useSctSpacePoints) {
    ATH_MSG_FATAL("Both usePixelSpacePoints and useSctSpacePoints set to False. At least one needs to be True");
    return StatusCode::FAILURE;
  }
 
  if (!m_useSctSpacePoints) ATH_MSG_INFO("Only using Pixel spacepoints => Pixel seeds only");
 
  if (!m_usePixelSpacePoints) ATH_MSG_INFO("Only using SCT spacepoints => Strip seeds only");
 
  if (m_usePixelSpacePoints && m_useSctSpacePoints) ATH_MSG_INFO("Using SCT and Pixel spacepoints");
 
  ATH_CHECK(m_pixelSpacePointsContainerKey.initialize(m_usePixelSpacePoints));
 
  ATH_CHECK(m_sctSpacePointsContainerKey.initialize(m_useSctSpacePoints));
  
  ATH_MSG_INFO("TrigInDetTrackSeedingTool initialized with " << m_useGPU);
  
  return StatusCode::SUCCESS;
}
 
StatusCode TrigInDetTrackSeedingTool::finalize() {
  return SeedingToolBase::finalize();
}
 
 
TrigInDetTrackSeedingResult TrigInDetTrackSeedingTool::findSeeds(const IRoiDescriptor& internalRoI, std::vector<TrigInDetTracklet>& output, const EventContext& ctx) const {
 
  TrigInDetTrackSeedingResult seedStats;
  
    output.clear();
 
  SG::ReadCondHandle<InDet::BeamSpotData> beamSpotHandle { m_beamSpotKey, ctx };  
  const Amg::Vector3D &vertex = beamSpotHandle->beamPos();
  float shift_x = vertex.x() - beamSpotHandle->beamTilt(0)*vertex.z();
  float shift_y = vertex.y() - beamSpotHandle->beamTilt(1)*vertex.z();
 
  std::unique_ptr<GNN_DataStorage> storage = std::make_unique<GNN_DataStorage>(*m_geo);
  
  int nPixels = 0;
  int nStrips = 0;
 
  const SpacePointContainer* sctSpacePointsContainer = nullptr;
  const SpacePointContainer* pixelSpacePointsContainer = nullptr;
 
  std::map<short, std::vector<IdentifierHash> > detIdMap;//relates global detector layer ID to constituent detector elements
 
  if (m_useSctSpacePoints) {
    
    SG::ReadHandle<SpacePointContainer> sctHandle(m_sctSpacePointsContainerKey, ctx);
 
    if(!sctHandle.isValid()) {
      ATH_MSG_WARNING("Invalid Strip Spacepoints handle: "<<sctHandle);
      return seedStats;
      
    }
    
    sctSpacePointsContainer = sctHandle.ptr();
 
    std::vector<IdentifierHash> listOfSctIds;
    m_regsel_sct->lookup(ctx)->HashIDList( internalRoI, listOfSctIds );
 
    const std::vector<short>* h2l = m_layerNumberTool->sctLayers();
    
    for(const auto& hashId : listOfSctIds) {
      
      short layerIndex = h2l->at(static_cast<int>(hashId));
 
      auto it = detIdMap.find(layerIndex);
      if(it != detIdMap.end()) (*it).second.push_back(hashId);
      else {
        std::vector<IdentifierHash> v = {hashId};
        detIdMap.insert(std::make_pair(layerIndex,v));
      }
    }
  }
 
  if (m_usePixelSpacePoints) {
    
    SG::ReadHandle<SpacePointContainer> pixHandle(m_pixelSpacePointsContainerKey, ctx);
 
    if(!pixHandle.isValid()) {
      ATH_MSG_WARNING("Invalid Pixel Spacepoints handle: "<<pixHandle);
      return seedStats;
    }
    
    pixelSpacePointsContainer = pixHandle.ptr();
    
    std::vector<IdentifierHash> listOfPixIds;
    
    m_regsel_pix->lookup(ctx)->HashIDList( internalRoI, listOfPixIds );
 
    const std::vector<short>* h2l = m_layerNumberTool->pixelLayers();
 
    for(const auto& hashId : listOfPixIds) {
      
      short layerIndex = h2l->at(static_cast<int>(hashId));
 
      auto it = detIdMap.find(layerIndex);
      if(it != detIdMap.end()) (*it).second.push_back(hashId);
      else {
        std::vector<IdentifierHash> v = {hashId};
        detIdMap.insert(std::make_pair(layerIndex,v));
      }
    }
  }
 
  if(!m_useGPU) {
 
    std::unique_ptr<GNN_DataStorage> storage = std::make_unique<GNN_DataStorage>(*m_geo);
 
    std::vector<const Trk::SpacePoint*> vSP;
 
    vSP.reserve(m_nMaxEdges);
    
    std::vector<std::vector<GNN_Node> > trigSpStorage[2];
    
    trigSpStorage[1].resize(m_layerNumberTool->sctLayers()->size());
 
    trigSpStorage[0].resize(m_layerNumberTool->pixelLayers()->size());
 
    for(const auto& lColl : detIdMap) {
 
      short layerIndex = lColl.first;
 
      int layerKey = m_geo->getTrigFTF_GNN_LayerByIndex(layerIndex)->m_layer.m_subdet;
      
      bool isPixel = layerKey > 20000;
 
      auto pCont = isPixel ? pixelSpacePointsContainer : sctSpacePointsContainer;
 
      int contIdx= isPixel ? 0 : 1;
 
      int nNewNodes = 0;
      
      for(const auto& idx : lColl.second) {
      
        std::vector<GNN_Node>& tmpColl = trigSpStorage[contIdx].at(static_cast<int>(idx));
 
        auto input_coll = pCont->indexFindPtr(idx);
 
        if(input_coll == nullptr) continue;
 
        createGraphNodes(input_coll, tmpColl, vSP, layerIndex, shift_x, shift_y);//TO-DO: if(m_useBeamTilt) SP full transform functor
 
        nNewNodes += (isPixel) ? storage->loadPixelGraphNodes(layerIndex, tmpColl, m_useML) : storage->loadStripGraphNodes(layerIndex, tmpColl);
      }
 
      if(isPixel) nPixels += nNewNodes;
      else nStrips += nNewNodes;      
    }
 
    seedStats.m_nPixelSPs = nPixels;
    seedStats.m_nStripSPs = nStrips;
  
    storage->sortByPhi();
    storage->initializeNodes(m_useML);
    storage->generatePhiIndexing(1.5*m_phiSliceWidth);
 
    std::vector<GNN_Edge> edgeStorage;
 
    std::pair<int, int> graphStats = buildTheGraph(internalRoI, storage, edgeStorage);
 
    ATH_MSG_DEBUG("Created graph with "<<graphStats.first<<" edges and "<<graphStats.second<< " edge links");
 
    seedStats.m_nGraphEdges = graphStats.first;
    seedStats.m_nEdgeLinks  = graphStats.second;
  
    if(graphStats.second == 0) return seedStats;
    
    int maxLevel = runCCA(graphStats.first, edgeStorage);
 
    ATH_MSG_DEBUG("Reached Level "<<maxLevel<<" after GNN iterations");
 
    int minLevel = 3;//a triplet + 1 confirmation
 
    if(m_LRTmode) {
      minLevel = 2;//a triplet
    }
  
    if(maxLevel < minLevel) return seedStats;
  
    std::vector<GNN_Edge*> vSeeds;
 
    vSeeds.reserve(graphStats.first/2);
 
    for(int edgeIndex=0;edgeIndex<graphStats.first;edgeIndex++) {
      GNN_Edge* pS = &(edgeStorage.at(edgeIndex));
 
      if(pS->m_level < minLevel) continue;
      
      vSeeds.push_back(pS);
    }
  
    if(vSeeds.empty()) return seedStats;
 
    std::sort(vSeeds.begin(), vSeeds.end(), GNN_Edge::CompareLevel());
 
    //backtracking
 
    TrigFTF_GNN_TrackingFilter tFilter(m_layerGeometry, edgeStorage);
 
    output.reserve(vSeeds.size());
  
    for(auto pS : vSeeds) {
      
      if(pS->m_level == -1) continue;
      
      TrigFTF_GNN_EdgeState rs(false);
      
      tFilter.followTrack(pS, rs);
      
      if(!rs.m_initialized) {
    continue;
      }
      
      if(static_cast<int>(rs.m_vs.size()) < minLevel) continue;
      
      std::vector<const GNN_Node*> vN;
      
      for(std::vector<GNN_Edge*>::reverse_iterator sIt=rs.m_vs.rbegin();sIt!=rs.m_vs.rend();++sIt) {
        
    (*sIt)->m_level = -1;//mark as collected
    
    if(sIt == rs.m_vs.rbegin()) {
      vN.push_back((*sIt)->m_n1);
    }
    vN.push_back((*sIt)->m_n2);
      }
      
      if(vN.size()<3) continue;
      
      unsigned int lastIdx = output.size();
      output.emplace_back(rs.m_J);
      
      for(const auto& n : vN) {
    output[lastIdx].addSpacePoint(vSP[n->m_idx]);
      }
    }
  
    ATH_MSG_DEBUG("Found "<<output.size()<<" tracklets");  
  }
  else {//GPU-accelerated graph building
        //1. data export
 
        std::vector<const Trk::SpacePoint*> vSP;
        std::vector<short> vL;
 
        vSP.reserve(TrigAccel::ITk::GBTS_MAX_NUMBER_SPACEPOINTS);
        vL.reserve(TrigAccel::ITk::GBTS_MAX_NUMBER_SPACEPOINTS);
 
        std::unique_ptr<TrigAccel::DATA_EXPORT_BUFFER> dataBuffer = std::unique_ptr<TrigAccel::DATA_EXPORT_BUFFER>(new TrigAccel::DATA_EXPORT_BUFFER(5000));
        
        size_t dataTypeSize = sizeof(TrigAccel::ITk::GRAPH_MAKING_INPUT_DATA);
        const size_t bufferOffset = 256;
        size_t totalSize = bufferOffset+dataTypeSize;//make room for the header
        if(!dataBuffer->fit(totalSize)) dataBuffer->reallocate(totalSize);
 
        TrigAccel::ITk::GRAPH_MAKING_INPUT_DATA* pJobData = reinterpret_cast<TrigAccel::ITk::GRAPH_MAKING_INPUT_DATA*>(dataBuffer->m_buffer + bufferOffset);
 
        unsigned int spIdx = 0;
        int sp_offset = 0;
        int nLayers = 0;
        int MaxEtaBin = 0;
        int nEtaBins  = 0;
        
        for(const auto& lColl : detIdMap) {
 
            short layerIndex = lColl.first;
            
            const TrigFTF_GNN_Layer* pL = m_geo->getTrigFTF_GNN_LayerByIndex(layerIndex);
 
            int layerKey = pL->m_layer.m_subdet;
            
            bool isPixel = layerKey > 20000;
            bool isBarrel = (pL->m_layer.m_type == 0);
            
            auto pCont = isPixel ? pixelSpacePointsContainer : sctSpacePointsContainer;
            
            pJobData->m_layerIdx[nLayers]        = layerIndex;
 
            pJobData->m_layerInfo[4*nLayers    ]   = spIdx;//layerInfo.x
            pJobData->m_layerInfo[4*nLayers + 2]   = pL->num_bins();//layerInfo.z
            pJobData->m_layerInfo[4*nLayers + 3]   = pL->m_bins[0];//layerInfo.w
            
            pJobData->m_layerGeo[2*nLayers    ]    = pL->m_minEta;//layerGeo.x
            pJobData->m_layerGeo[2*nLayers + 1]    = pL->m_etaBin;//layerGeo.y
            
            nEtaBins += pL->m_bins.size();
            
            for(auto b : pL->m_bins) {
                if(b > MaxEtaBin) MaxEtaBin = b;
            }
 
            for(const auto& idx : lColl.second) {
     
                auto input_coll = pCont->indexFindPtr(idx);
 
                if(input_coll == nullptr) continue;
 
                for(const auto sp : *input_coll) {
 
                    float cw = -1.0;
                    
                    const InDet::PixelCluster* pCL = dynamic_cast<const InDet::PixelCluster*>(sp->clusterList().first);
                    if(pCL != nullptr) {
                        cw = pCL->width().widthPhiRZ().y();
                        if(!isBarrel && m_useML) {
                            if(cw > 0.2) continue;
                            cw = -1.0;//set it to -1 so that it can be skipped later in the ML code on GPU
                        }
                    }
                    
                    vSP.emplace_back(sp);
                    vL.emplace_back(layerIndex);
        
                    const auto& p = sp->globalPosition();
 
                    float params[4] = {(float)(p.x() - shift_x), float(p.y() - shift_y), (float)(p.z()), cw};
 
                    memcpy(&pJobData->m_params[sp_offset], &params[0], sizeof(params));
                    
                    sp_offset += 4;
                    spIdx++;
                    if(spIdx >= TrigAccel::ITk::GBTS_MAX_NUMBER_SPACEPOINTS) break;
                }       
                if(spIdx >= TrigAccel::ITk::GBTS_MAX_NUMBER_SPACEPOINTS) break;
            }
 
            pJobData->m_layerInfo[4*nLayers + 1] = spIdx;//layerInfo.y
 
            if (isPixel) nPixels += spIdx - pJobData->m_layerInfo[4*nLayers];
            else nStrips += spIdx - pJobData->m_layerInfo[4*nLayers];
            
            nLayers++;
        }
 
        if(spIdx == 0) return seedStats;
        
        pJobData->m_nSpacepoints = spIdx;
        pJobData->m_nLayers      = nLayers;
        pJobData->m_nEtaBins     = nEtaBins;
        pJobData->m_maxEtaBin    = MaxEtaBin;
        pJobData->m_nMaxEdges    = static_cast<unsigned int>(7*spIdx); 
 
        //load bin pairs
 
        int pairIdx = 0;
        
        for(const auto& bg : m_geo->bin_groups()) {//loop over bin groups
        
            int bin1_idx = bg.first;
                
            for(const auto& bin2_idx : bg.second) {
                pJobData->m_bin_pairs[2*pairIdx  ] = bin1_idx;
                pJobData->m_bin_pairs[2*pairIdx+1] = bin2_idx;
                pairIdx++;
            }
        }
        
        pJobData->m_nBinPairs = pairIdx;
        
        //add algorithm parameters
        
        const float ptCoeff = 0.29997*1.9972/2.0;// ~0.3*B/2 - assuming nominal field of 2*T
 
        float tripletPtMin = 0.8*m_minPt;//correction due to limited pT resolution
 
        float maxCurv = ptCoeff/tripletPtMin;
 
        const float pt_scale = 900/m_minPt;
        
        const float min_deltaPhi_low_dr = 0.002*pt_scale;
        const float dphi_coeff_low_dr   = 4.33e-4*pt_scale;
        const float min_deltaPhi        = 0.015*pt_scale;
        const float dphi_coeff          = 2.2e-4*pt_scale;
        
        const float cut_dphi_max        = m_LRTmode ? 0.07 : 0.012;
        const float cut_dcurv_max       = m_LRTmode ? 0.015 : 0.001;
        const float cut_tau_ratio_max = m_LRTmode ? 0.015f : 0.01;
        const float min_z0              = m_LRTmode ? -600.0 : internalRoI.zedMinus();
        const float max_z0              = m_LRTmode ? 600.0 : internalRoI.zedPlus();
 
        const float maxOuterRadius      = m_LRTmode ? 1050.0 : 550.0;  
        const float minDeltaRadius      = 2.0;
                
        const float cut_zMinU = min_z0 + maxOuterRadius*internalRoI.dzdrMinus();
        const float cut_zMaxU = max_z0 + maxOuterRadius*internalRoI.dzdrPlus();
    
        const float maxKappa            = m_LRTmode ? 1.0*maxCurv : 0.9*maxCurv;
            
        pJobData->m_algo_params[0]  = min_deltaPhi;
        pJobData->m_algo_params[1]  = dphi_coeff;
        pJobData->m_algo_params[2]  = min_deltaPhi_low_dr;
        pJobData->m_algo_params[3]  = dphi_coeff_low_dr;
        pJobData->m_algo_params[4]  = minDeltaRadius;
        pJobData->m_algo_params[5]  = min_z0;
        pJobData->m_algo_params[6]  = max_z0;
        pJobData->m_algo_params[7]  = maxOuterRadius;
        pJobData->m_algo_params[8]  = cut_zMinU;
        pJobData->m_algo_params[9]  = cut_zMaxU;
        pJobData->m_algo_params[10] = maxKappa;
        pJobData->m_algo_params[11] = cut_dphi_max;
        pJobData->m_algo_params[12] = cut_dcurv_max;
        pJobData->m_algo_params[13] = cut_tau_ratio_max;    
        
        int minLevel = 3;//a triplet + 1 confirmation
 
        if(m_LRTmode) {
            minLevel = 2;//a triplet
        }
 
        pJobData->m_minLevel = minLevel;
 
        pJobData->m_useGPUseedExtraction = m_useGPUseedExtraction;          
 
        seedStats.m_nPixelSPs = nPixels;
        seedStats.m_nStripSPs = nStrips;
 
        ATH_MSG_DEBUG("Loaded "<<nPixels<< " Pixel Spacepoints and "<<nStrips<< " Strip SpacePoints");
        
        std::shared_ptr<TrigAccel::OffloadBuffer> pBuff = std::make_shared<TrigAccel::OffloadBuffer>(dataBuffer.get());
    
        std::unique_ptr<TrigAccel::Work> pWork = std::unique_ptr<TrigAccel::Work>(m_accelSvc->createWork(TrigAccel::InDetJobControlCode::RUN_GBTS, pBuff));
    
        if(!pWork) {
            ATH_MSG_WARNING("Failed to create a work item for task "<<TrigAccel::InDetJobControlCode::RUN_GBTS);
            return seedStats;
        }
 
        bool workSuccess = pWork->run();
        if(!workSuccess) {
            ATH_MSG_WARNING("Work item failed to complete");
            return seedStats;
        }    
        
        std::shared_ptr<TrigAccel::OffloadBuffer> pOutput = pWork->getOutput();
        
        TrigAccel::ITk::GRAPH_AND_SEEDS_OUTPUT* pGraphAndSeeds = reinterpret_cast<TrigAccel::ITk::GRAPH_AND_SEEDS_OUTPUT*>(pOutput->m_rawBuffer);
        TrigAccel::ITk::COMPRESSED_GRAPH* pGraph = &pGraphAndSeeds->m_CompressedGraph;
        TrigAccel::ITk::OUTPUT_SEEDS* pSeeds = &pGraphAndSeeds->m_OutputSeeds;
            
        if(m_useGPUseedExtraction) {
            if(pSeeds->m_nSeeds == 0) return seedStats;
            //converting tracklet into GBTS-CPU format
            for(unsigned int seed=0; seed<pSeeds->m_nSeeds; seed++) {
                
                unsigned int lastIdx = output.size();
                output.emplace_back(pSeeds->m_seedsArray[seed].m_Q);
                for(int node=0; node<pSeeds->m_seedsArray[seed].m_size; node++) {
                    output[lastIdx].addSpacePoint(vSP[pSeeds->m_seedsArray[seed].m_nodes[node]]);
                }
            }   
        }
        else {
 
            unsigned int nEdges = pGraph->m_nEdges;
            unsigned int nMaxNei = pGraph->m_nMaxNeighbours;
 
            //populating the edgeStorage
 
            std::vector<GNN_Node> nodes;
 
            nodes.reserve(vSP.size());
 
            for(unsigned int idx = 0;idx < vSP.size(); idx++) {
 
                nodes.emplace_back(vL[idx]);
                
                const auto& pos = vSP[idx]->globalPosition();
                float xs = pos.x() - shift_x;
                float ys = pos.y() - shift_y;
                float zs = pos.z();
                
                nodes[idx].m_x = xs;
                nodes[idx].m_y = ys;
                nodes[idx].m_z = zs;
                nodes[idx].m_r = std::sqrt(xs*xs + ys*ys);
 
                nodes[idx].m_idx = idx;
 
            }
            
            std::vector<GNN_Edge> edgeStorage;
 
            std::pair<int, int> graphStats(0,0);
 
            edgeStorage.resize(nEdges);
 
            unsigned int edgeSize = nMaxNei + 1 + 2;//neigbours, num_neighbours, 2 nodes
 
            for(unsigned int idx=0;idx<nEdges;idx++) {
                unsigned int pos = idx*edgeSize;
                
                int node1Idx = pGraph->m_graphArray[pos + TrigAccel::ITk::node1];
                int node2Idx = pGraph->m_graphArray[pos + TrigAccel::ITk::node2];
                int nNei     = pGraph->m_graphArray[pos + TrigAccel::ITk::nNei];
 
                if(nNei > N_SEG_CONNS) nNei = N_SEG_CONNS;
                
                edgeStorage[idx].m_n1 = &nodes[node1Idx];
                edgeStorage[idx].m_n2 = &nodes[node2Idx];
                edgeStorage[idx].m_level = 1;
                edgeStorage[idx].m_nNei = nNei;
                for(int k=0;k<nNei;k++) {
                    edgeStorage[idx].m_vNei[k] = pGraph->m_graphArray[pos + TrigAccel::ITk::nei_idx_start + k];
                }
            }
            
            graphStats.first = nEdges;
            graphStats.second = pGraph->m_nLinks;
            
            //run the rest of the GBTS workflow
 
            int maxLevel = runCCA(graphStats.first, edgeStorage);
 
            ATH_MSG_DEBUG("Reached Level "<<maxLevel<<" after GNN iterations");
     
            if(maxLevel < minLevel) return seedStats;
        
            std::vector<GNN_Edge*> vSeeds;
 
            vSeeds.reserve(graphStats.first/2);
 
            for(int edgeIndex=0;edgeIndex<graphStats.first;edgeIndex++) {
                GNN_Edge* pS = &(edgeStorage.at(edgeIndex));
 
                if(pS->m_level < minLevel) continue;
                
                vSeeds.push_back(pS);
            }
        
            if(vSeeds.empty()) return seedStats;
     
            std::sort(vSeeds.begin(), vSeeds.end(), GNN_Edge::CompareLevel());
 
            //backtracking
 
            TrigFTF_GNN_TrackingFilter tFilter(m_layerGeometry, edgeStorage);
 
            output.reserve(vSeeds.size());
        
            for(auto pS : vSeeds) {
                
                if(pS->m_level == -1) continue;
                
                TrigFTF_GNN_EdgeState rs(false);
                
                tFilter.followTrack(pS, rs);
      
                if(!rs.m_initialized) {
                    continue;
                }
      
                if(static_cast<int>(rs.m_vs.size()) < minLevel) continue;
      
                std::vector<const GNN_Node*> vN;
      
                for(std::vector<GNN_Edge*>::reverse_iterator sIt=rs.m_vs.rbegin();sIt!=rs.m_vs.rend();++sIt) {
        
                    (*sIt)->m_level = -1;//mark as collected
    
                    if(sIt == rs.m_vs.rbegin()) {
                        vN.push_back((*sIt)->m_n1);
                    }
                    vN.push_back((*sIt)->m_n2);
                }
      
                if(vN.size()<3) continue;
      
                unsigned int lastIdx = output.size();
                output.emplace_back(rs.m_J);
      
                for(const auto& n : vN) {
                    output[lastIdx].addSpacePoint(vSP[n->m_idx]);
                }
            }
        }
        ATH_MSG_DEBUG("Found "<<output.size()<<" tracklets");
  }
  return seedStats;  
}
 
void TrigInDetTrackSeedingTool::createGraphNodes(const SpacePointCollection* spColl, std::vector<GNN_Node>& tmpColl, std::vector<const Trk::SpacePoint*>& vSP, unsigned short layer, float shift_x, float shift_y) const {
  
  tmpColl.resize(spColl->size(), GNN_Node(layer));//all nodes belong to the same layer
  
  int idx = 0;
  int init_size = vSP.size();
  for(const auto sp : *spColl) {
    const auto& pos = sp->globalPosition();
    vSP.emplace_back(sp);
    float xs = pos.x() - shift_x;
    float ys = pos.y() - shift_y;
    float zs = pos.z();
    tmpColl[idx].m_x = xs;
    tmpColl[idx].m_y = ys;
    tmpColl[idx].m_z = zs;
    tmpColl[idx].m_r = std::sqrt(xs*xs + ys*ys);
    tmpColl[idx].m_phi = std::atan2(ys,xs);
    tmpColl[idx].m_idx = init_size + idx;
 
    const InDet::PixelCluster* pCL = dynamic_cast<const InDet::PixelCluster*>(sp->clusterList().first);
    if(pCL != nullptr){
      tmpColl[idx].m_pcw = pCL->width().z();
    }
 
    idx++;
  }
}