gFEXJetAlgo.cxx

Go to the documentation of this file.

/*
  Copyright (C) 2002-2026 CERN for the benefit of the ATLAS collaboration
*/
//***************************************************************************
//    gFEXJetAlgo - JetFinder algorithm for gFEX
//                              -------------------
//     begin                : 01 04 2021
//     email                : cecilia.tosciri@cern.ch
//***************************************************************************
 
 
 
#include "gFEXJetAlgo.h"
#include "L1CaloFEXSim/gFEXJetTOB.h"
#include "L1CaloFEXSim/gTowerContainer.h"
#include "L1CaloFEXSim/gTower.h"
 
#include <vector>
namespace LVL1 {
 
  // default constructor for persistency
gFEXJetAlgo::gFEXJetAlgo(const std::string& type, const std::string& name, const IInterface* parent):
    AthAlgTool(type, name, parent)
  {
    declareInterface<IgFEXJetAlgo>(this);
  }
 
 
StatusCode gFEXJetAlgo::initialize(){
 
  ATH_CHECK(m_gFEXJetAlgo_gTowerContainerKey.initialize());
 
  return StatusCode::SUCCESS;
 
}
 
 
 
std::vector<std::unique_ptr<gFEXJetTOB>> gFEXJetAlgo::largeRfinder(
                               const gTowersType& Atwr, 
                               const gTowersType& Btwr,
                               const gTowersType& Ctwr,
                               const gTowersType& Asat, 
                               const gTowersType& Bsat,
                               const gTowersType& Csat,
                               int pucA, int pucB, int pucC, int gLJ_seedThrA, int gLJ_seedThrB, int gLJ_seedThrC, 
                               int gJ_ptMinToTopoCounts1, int gJ_ptMinToTopoCounts2,
                               int jetThreshold, int gLJ_ptMinToTopoCounts1, int gLJ_ptMinToTopoCounts2,
                               std::array<uint32_t, 7> & ATOB1_dat, std::array<uint32_t, 7> & ATOB2_dat,
                               std::array<uint32_t, 7> & BTOB1_dat, std::array<uint32_t, 7> & BTOB2_dat,
                               std::array<uint32_t, 7> & CTOB1_dat, std::array<uint32_t, 7> & CTOB2_dat) const {
 
  // Arrays for gJets
  // s = status (1 if above threshold)
  // v = value 200 MeV LSB
  // eta and phi are bin numbers
  std::array<int, FEXAlgoSpaceDefs::gJetTOBfib> gTOBsat    = {{0,0,0,0,0,0}};
  std::array<int, FEXAlgoSpaceDefs::gJetTOBfib> gJetTOBs   = {{0,0,0,0,0,0}};
  std::array<int, FEXAlgoSpaceDefs::gJetTOBfib> gJetTOBv   = {{0,0,0,0,0,0}};
  std::array<int, FEXAlgoSpaceDefs::gJetTOBfib> gJetTOBeta = {{0,0,0,0,0,0}};
  std::array<int, FEXAlgoSpaceDefs::gJetTOBfib> gJetTOBphi = {{0,0,0,0,0,0}};
 
  // Arrays for gBlocks  (so far only 2 gBlocks per fiber)
  // first index is for the fiber number (A 0 and 1, B 0 and 1, C P and C N )
  // second index is for leading, sub-leading, and sub-sub-leading (not used)
  std::array<std::array<int, 3>, FEXAlgoSpaceDefs::BTOBFIB> gBlockTOBv   = {{ {{0,0,0}}, {{0,0,0}}, {{0,0,0}}, {{0,0,0}}, {{0,0,0}}, {{0,0,0}} }};
  std::array<std::array<int, 3>, FEXAlgoSpaceDefs::BTOBFIB> gBlockTOBeta = {{ {{0,0,0}}, {{0,0,0}}, {{0,0,0}}, {{0,0,0}}, {{0,0,0}}, {{0,0,0}} }};
  std::array<std::array<int, 3>, FEXAlgoSpaceDefs::BTOBFIB> gBlockTOBphi = {{ {{0,0,0}}, {{0,0,0}}, {{0,0,0}}, {{0,0,0}}, {{0,0,0}}, {{0,0,0}} }};
 
 
  //Alternate format for FPGC C (split in two)
  
  gTowersJetEngine CNtwr;
  gTowersJetEngine CPtwr;
 
  for(int irow=0; irow<FEXAlgoSpaceDefs::ABCrows; irow++){ 
    for(int jcolumn=0;jcolumn<FEXAlgoSpaceDefs::ABCcolumnsEng;jcolumn++){
      CNtwr[irow][jcolumn] =  Ctwr[irow][jcolumn]  ;     
      CPtwr[irow][jcolumn] =  Ctwr[irow][jcolumn+FEXAlgoSpaceDefs::ABCcolumnsEng] ;
    }
 
    for(unsigned int jcolumn=0;jcolumn<FEXAlgoSpaceDefs::ABcolumns;jcolumn++){
 
      if( jcolumn < 6 ) {
        // TOB1
        if( Asat[irow][jcolumn] == 1 )  gTOBsat[0] = 1;
        if( Bsat[irow][jcolumn] == 1 )  gTOBsat[2] = 1;
        if( Csat[irow][jcolumn] == 1 )  gTOBsat[4] = 1;
      } else {
        //TOB2 
        if( Asat[irow][jcolumn] == 1 ) gTOBsat[1] = 1;
        if( Bsat[irow][jcolumn] == 1 ) gTOBsat[3] = 1;
        if( Csat[irow][jcolumn] == 1 ) gTOBsat[5] = 1;
      }
    }
  }
 
 
 
  gTowersType AjetsAlt;
  singleHalf(Atwr,AjetsAlt);
 
  gTowersPartialSums AlpsAltOut;
  gTowersPartialSums ArpsAltOut;
 
  InternalPartialAB(Atwr, AlpsAltOut, ArpsAltOut);
 
  gTowersType BjetsAlt;
  singleHalf(Btwr,BjetsAlt);
 
  gTowersPartialSums BlpsAltOut;
  gTowersPartialSums BrpsAltOut;
 
  InternalPartialAB(Btwr, BlpsAltOut, BrpsAltOut);
 
  gTowersType CjetsAlt;
  singleHalf(Ctwr,CjetsAlt);
 
  // Add the left local partial sum (zero if right column)
  addInternalLin(AjetsAlt, AlpsAltOut);
  addInternalLin(BjetsAlt, BlpsAltOut);
 
  pileUpCorrectionAB(AjetsAlt,pucA);
  pileUpCorrectionAB(BjetsAlt,pucB);
  pileUpCorrectionAB(CjetsAlt,pucC);
 
  // If the local sum is less than zero set it to zero
  ZeroNegative(AjetsAlt);
  ZeroNegative(BjetsAlt);
  ZeroNegative(CjetsAlt);
 
  // Add the right local partial sum (zero if right column)
  addInternalRin(AjetsAlt, ArpsAltOut);
  addInternalRin(BjetsAlt, BrpsAltOut);
 
  for(unsigned int irow = 0; irow < FEXAlgoSpaceDefs::ABCrows; irow++ ){
    for(unsigned int icolumn =0; icolumn<FEXAlgoSpaceDefs::ABcolumns; icolumn++){
      // set 18 bits on
      if (Asat[irow][icolumn] != 0){
      }
    }
  }
 
  SaturateJets( AjetsAlt, Asat, 0 );
  SaturateJets( BjetsAlt, Bsat, 1 );
  SaturateJets( CjetsAlt, Csat, 2 );
 
  gTowersType AjetsRestricted;
  gTowersType BjetsRestricted;
  gTowersType CjetsRestricted;
 
 
  for(int irow=0; irow<FEXAlgoSpaceDefs::ABCrows; irow++){
    for(int icolumn=0; icolumn<FEXAlgoSpaceDefs::ABcolumns; icolumn++){
      AjetsRestricted[irow][icolumn]  = AjetsAlt[irow][icolumn]; 
      BjetsRestricted[irow][icolumn]  = BjetsAlt[irow][icolumn];
      CjetsRestricted[irow][icolumn]  = CjetsAlt[irow][icolumn]; 
    }
  }
 
  // find gBlocks
  gTowersType gBLKA;
  gTowersType hasSeedA;
  gBlockAB(Atwr, gBLKA, hasSeedA, gLJ_seedThrA);
 
  SaturateBlocks(gBLKA, Asat, 0);
 
  // TODO: Temporary fix to match FW saturation eta-alignment bug.
  // In firmware, seed_ovf forces et9 to max positive → have_seed=1 unconditionally.
  // The same −1 eta shift applies: overflow at col K forces have_seed at col K-1.
  // @see jet_eng.vhd lines 849-870: seed_ovf → et9=max → have_seed=1
  // Revert this block when the FW saturation alignment is fixed.
  for(unsigned int irow = 0; irow < FEXAlgoSpaceDefs::ABCrows; irow++){
    for(unsigned int icol = 0; icol < FEXAlgoSpaceDefs::ABcolumns; icol++){
      if(Asat[irow][icol] && (icol % 6 != 0)) hasSeedA[irow][icol - 1] = 1;
    }
  }
 
  gTowersType gBLKB;
  gTowersType hasSeedB;
  gBlockAB(Btwr, gBLKB, hasSeedB, gLJ_seedThrB);  
 SaturateBlocks(gBLKB, Bsat, 1);
 
  // TODO: Temporary fix to match FW saturation eta-alignment bug.
  // Revert this block when the FW saturation alignment is fixed.
  for(unsigned int irow = 0; irow < FEXAlgoSpaceDefs::ABCrows; irow++){
    for(unsigned int icol = 0; icol < FEXAlgoSpaceDefs::ABcolumns; icol++){
      if(Bsat[irow][icol] && (icol % 6 != 0) && icol != 11) hasSeedB[irow][icol - 1] = 1;
    }
  }
 
  gTowersType gBLKC;
  gTowersType hasSeedC;
  gBlockAB(Ctwr, gBLKC, hasSeedC, gLJ_seedThrC);  
 SaturateBlocks(gBLKC, Csat, 2);
 
  // TODO: Temporary fix to match FW saturation eta-alignment bug.
  // Revert this block when the FW saturation alignment is fixed.
  for(unsigned int irow = 0; irow < FEXAlgoSpaceDefs::ABCrows; irow++){
    for(unsigned int icol = 0; icol < FEXAlgoSpaceDefs::ABcolumns; icol++){
      if(Csat[irow][icol] && (icol % 6 != 0) && (icol % 6 != 5)) hasSeedC[irow][icol - 1] = 1;
    }
  }
 
  // sorting by jet engine -- not done in FPGA
  std::array<int, 32> AgBlockOutL{};
  std::array<int, 32> AgBlockEtaIndL{};
  std::array<int, 32> AgBlockOutR{};
  std::array<int, 32> AgBlockEtaIndR{};
  blkOutAB(gBLKA, AgBlockOutL, AgBlockEtaIndL, AgBlockOutR, AgBlockEtaIndR);
 
  std::array<int, 32> BgBlockOutL{};
  std::array<int, 32> BgBlockEtaIndL{};
  std::array<int, 32> BgBlockOutR{};
  std::array<int, 32> BgBlockEtaIndR{};
  blkOutAB(gBLKB, BgBlockOutL, BgBlockEtaIndL, BgBlockOutR, BgBlockEtaIndR);
  
  std::array<int, 32> CgBlockOutL{};
  std::array<int, 32> CgBlockEtaIndL{};
  std::array<int, 32> CgBlockOutR{};
  std::array<int, 32> CgBlockEtaIndR{};
  blkOutAB(gBLKC, CgBlockOutL, CgBlockEtaIndL, CgBlockOutR, CgBlockEtaIndR) ;
 
  // sorting by 192 blocks 0 = left, 1 = right
  // second index is column number, the eta value is 2 + 6*column number + local eta
 
  // find the leading and subleading gBlock  in the left column of FPGA A
  gBlockMax2(gBLKA, 1, 0, gBlockTOBv[0], gBlockTOBeta[0], gBlockTOBphi[0]);
   // find the leading and subleading gBlock  in the right column of FPGA A
  gBlockMax2(gBLKA, 2, 1, gBlockTOBv[1], gBlockTOBeta[1], gBlockTOBphi[1]);
 
  // find the leading and subleading gBlock  in the left column of FPGA B
  gBlockMax2(gBLKB, 3, 0, gBlockTOBv[2], gBlockTOBeta[2], gBlockTOBphi[2]);
  // find the leading and subleading gBlock  in the right column of FPGA B
  gBlockMax2(gBLKB, 4, 1, gBlockTOBv[3], gBlockTOBeta[3], gBlockTOBphi[3]);
 
  // find the leading and subleading gBlock  in the  CN column of FPGA C 
  gBlockMax2( gBLKC, 0, 0, gBlockTOBv[4], gBlockTOBeta[4], gBlockTOBphi[4]);
  // find the leading and subleading gBlock  in the CP column of FPGA C 
  gBlockMax2( gBLKC, 5, 1, gBlockTOBv[5], gBlockTOBeta[5], gBlockTOBphi[5]);
 
  // in hardware vetos happen before remote sums come in. 
  gBlockVetoAB(AjetsRestricted, hasSeedA);  
  gBlockVetoAB(BjetsRestricted, hasSeedB);  
  gBlockVetoAB(CjetsRestricted, hasSeedC);  
 
  // calculate A & B  remote partial sums first
  gTowersPartialSums RAlps_out, RArps_out;
  RemotePartialAB(Atwr, RAlps_out, RArps_out);
 
  gTowersPartialSums RBlps_out, RBrps_out;
  RemotePartialAB(Btwr, RBlps_out, RBrps_out);
 
  gTowersPartialSums RCNrps_out, RCPlps_out;
  RemotePartialCN(CNtwr,  RCNrps_out);
  RemotePartialCP(CPtwr,  RCPlps_out);
 
 
  if( FEXAlgoSpaceDefs::ENABLE_INTER_AB ) {
    
    // input partial sums from FPGA B to A (lps_out -> rps_in)
    addRemoteRin(AjetsRestricted, RBlps_out, FEXAlgoSpaceDefs::PS_UPPER_AB, FEXAlgoSpaceDefs::PS_LOWER_AB, FEXAlgoSpaceDefs::PS_SHIFT_AB);
 
    // input partial sums from FPGA B to A (lps_out -> rps_in)
    addRemoteLin(BjetsRestricted, RArps_out, FEXAlgoSpaceDefs::PS_UPPER_AB, FEXAlgoSpaceDefs::PS_LOWER_AB, FEXAlgoSpaceDefs::PS_SHIFT_AB);
 
  }
 
  if( FEXAlgoSpaceDefs::ENABLE_INTER_C ) {
 
    addRemoteLin(AjetsRestricted, RCNrps_out, FEXAlgoSpaceDefs::PS_UPPER_C, FEXAlgoSpaceDefs::PS_LOWER_C, FEXAlgoSpaceDefs::PS_SHIFT_C);
 
    addRemoteRin(BjetsRestricted, RCPlps_out, FEXAlgoSpaceDefs::PS_UPPER_C, FEXAlgoSpaceDefs::PS_LOWER_C, FEXAlgoSpaceDefs::PS_SHIFT_C);
 
  }
 
//  this are added into the jets you have to truncate to the number of bits on the interFPGA communication
  
  if( FEXAlgoSpaceDefs::ENABLE_INTER_ABC ) {
 
    addRemoteCNin(CjetsRestricted, RAlps_out, FEXAlgoSpaceDefs::PS_UPPER_C, FEXAlgoSpaceDefs::PS_LOWER_C, FEXAlgoSpaceDefs::PS_SHIFT_C );
    
    addRemoteCPin(CjetsRestricted, RBrps_out, FEXAlgoSpaceDefs::PS_UPPER_C, FEXAlgoSpaceDefs::PS_LOWER_C, FEXAlgoSpaceDefs::PS_SHIFT_C );
 
  }
 
  // Emulate firmware sum69o_del_ovf: after remote partial sums are added,
  // re-saturate jets whose input towers were saturated.
  // In firmware (jet_eng.vhd), sum69o_del_ovf = delayK(dat_ovf_in, SUM69O_DEL_DELAY)
  // overrides sum69o_del to max positive *instead of* adding remote_ps.
  // Without this, the simulation adds remote_ps on top of an already-max jet value,
  // biasing the max-finder (jetOutAB) toward columns that receive remote partial sums
  // and shifting the reported eta.
  // The same -1 eta shift as SaturateJets applies here.
  SaturateJets( AjetsRestricted, Asat, 0 );
  SaturateJets( BjetsRestricted, Bsat, 1 );
  SaturateJets( CjetsRestricted, Csat, 2 );
 
  //Emulate switch to unsigned values by zeroing everything below the jet threshold
  //https://gitlab.cern.ch/atlas-l1calo/gfex/firmware/-/blob/devel/common/jet_finder/HDL/jet_eng.vhd#L550
 
  gJetVetoAB(AjetsRestricted,  jetThreshold);
  gJetVetoAB(BjetsRestricted,  jetThreshold);
  gJetVetoAB(CjetsRestricted, jetThreshold);
 
  std::array<int, 32> AjetOutL;
  std::array<int, 32> AetaIndL;
  std::array<int, 32> AjetOutR;
  std::array<int, 32> AetaIndR;
 
 
  jetOutAB(AjetsRestricted, AjetOutL, AetaIndL, AjetOutR, AetaIndR);
 
  std::array<int, 32> BjetOutL;
  std::array<int, 32> BetaIndL;
  std::array<int, 32> BjetOutR;
  std::array<int, 32> BetaIndR;
 
  jetOutAB(BjetsRestricted, BjetOutL, BetaIndL, BjetOutR, BetaIndR);
 
  std::array<int, 32> CNjetOut;
  std::array<int, 32> CNetaInd;
  std::array<int, 32> CPjetOut;
  std::array<int, 32> CPetaInd;
 
  jetOutAB(CjetsRestricted, CNjetOut, CNetaInd, CPjetOut, CPetaInd);
 
  gJetTOBgen(AjetOutL, AetaIndL, 0, jetThreshold, gJetTOBs, gJetTOBv, gJetTOBeta, gJetTOBphi);
  gJetTOBgen(AjetOutR, AetaIndR, 1, jetThreshold, gJetTOBs, gJetTOBv, gJetTOBeta, gJetTOBphi);
 
  gJetTOBgen(BjetOutL, BetaIndL, 2, jetThreshold, gJetTOBs, gJetTOBv, gJetTOBeta, gJetTOBphi);
  gJetTOBgen(BjetOutR, BetaIndR, 3, jetThreshold, gJetTOBs, gJetTOBv, gJetTOBeta, gJetTOBphi);
 
  gJetTOBgen(CNjetOut, CNetaInd, 4, jetThreshold, gJetTOBs, gJetTOBv, gJetTOBeta, gJetTOBphi);
  gJetTOBgen(CPjetOut, CPetaInd, 5, jetThreshold, gJetTOBs, gJetTOBv, gJetTOBeta, gJetTOBphi);

  std::vector<std::unique_ptr<gFEXJetTOB>> tobs_v;
  tobs_v.resize(14);
 
  // fill in TOBs
  ATOB1_dat[0]  = 0;
  BTOB1_dat[0]  = 0;
  CTOB1_dat[0]  = 0;
  
  ATOB2_dat[0]  =  (( pucA & 0x0000FFFF ) << 8);
  BTOB2_dat[0]  =  (( pucB & 0x0000FFFF ) << 8);
  CTOB2_dat[0]  =  (( pucC & 0x0000FFFF ) << 8);
 
  // //First available TOBs are the gRho for each central FPGA
  tobs_v[0] = std::make_unique<gFEXJetTOB>();
  tobs_v[0]->setWord(ATOB2_dat[0]);
  tobs_v[0]->setET(pucA);
  tobs_v[0]->setEta(0);
  tobs_v[0]->setPhi(0);
  tobs_v[0]->setTobID(0);
  tobs_v[0]->setStatus(1);//is 1 only if rho calculation is valid, but rho is not calculated at the moment
 
  tobs_v[1] = std::make_unique<gFEXJetTOB>();
  tobs_v[1]->setWord(BTOB2_dat[0]);
  tobs_v[1]->setET(pucB);
  tobs_v[1]->setEta(0);
  tobs_v[1]->setPhi(0);
  tobs_v[1]->setTobID(0);
  tobs_v[1]->setStatus(1);//is 1 only if rho calculation is valid, but rho is not calculated at the moment
  
  // leading gBlocks  (available first and go in first TOB value)
  // TOBs 2-5 are leading gBlocks
  ATOB1_dat[1] =  0x00000001; //set the TOB ID in the corresponding slot (LSB)
  if(  gBlockTOBv[0][0] > gJ_ptMinToTopoCounts2 )  ATOB1_dat[1] = ATOB1_dat[1] | 0x00000080;//status
  ATOB1_dat[1] =  ATOB1_dat[1] | ( ( gBlockTOBv[0][0]   & 0x00000FFF ) << 8);
  ATOB1_dat[1] =  ATOB1_dat[1] | ( ( gBlockTOBeta[0][0] & 0x0000003F ) <<20);
  ATOB1_dat[1] =  ATOB1_dat[1] | ( ( gBlockTOBphi[0][0] & 0x0000001F ) <<26);
 
  tobs_v[2] = std::make_unique<gFEXJetTOB>();
  tobs_v[2]->setWord(ATOB1_dat[1]);
  tobs_v[2]->setET(gBlockTOBv[0][0]);
  tobs_v[2]->setEta(gBlockTOBeta[0][0]);
  tobs_v[2]->setPhi(gBlockTOBphi[0][0]);
  tobs_v[2]->setTobID(1);
  if(  gBlockTOBv[0][0] > gJ_ptMinToTopoCounts2 ) tobs_v[2]->setStatus(1);
  else tobs_v[2]->setStatus(0);
 
  ATOB2_dat[1] =  0x00000002;
  if(  gBlockTOBv[1][0] > gJ_ptMinToTopoCounts1 ) ATOB2_dat[1] =  ATOB2_dat[1] | 0x00000080;
  ATOB2_dat[1] =  ATOB2_dat[1] | ( ( gBlockTOBv[1][0]   & 0x00000FFF ) << 8);
  ATOB2_dat[1] =  ATOB2_dat[1] | ( ( gBlockTOBeta[1][0] & 0x0000003F ) <<20);
  ATOB2_dat[1] =  ATOB2_dat[1] | ( ( gBlockTOBphi[1][0] & 0x0000001F ) <<26);
 
  tobs_v[3] = std::make_unique<gFEXJetTOB>();
  tobs_v[3]->setWord(ATOB2_dat[1]);
  tobs_v[3]->setET(gBlockTOBv[1][0]);
  tobs_v[3]->setEta(gBlockTOBeta[1][0]);
  tobs_v[3]->setPhi(gBlockTOBphi[1][0]);
  tobs_v[3]->setTobID(2);
  if(  gBlockTOBv[1][0] > gJ_ptMinToTopoCounts1 ) tobs_v[3]->setStatus(1);
  else tobs_v[3]->setStatus(0);
 
 
  BTOB1_dat[1] =  0x00000001;
  if(  gBlockTOBv[2][0] > gJ_ptMinToTopoCounts1 ) BTOB1_dat[1] = BTOB1_dat[1] | 0x00000080;
  BTOB1_dat[1] =  BTOB1_dat[1] | ( ( gBlockTOBv[2][0]   & 0x00000FFF ) << 8);
  BTOB1_dat[1] =  BTOB1_dat[1] | ( ( gBlockTOBeta[2][0] & 0x0000003F ) <<20);
  BTOB1_dat[1] =  BTOB1_dat[1] | ( ( gBlockTOBphi[2][0] & 0x0000001F) <<26);
 
  tobs_v[4] = std::make_unique<gFEXJetTOB>();
  tobs_v[4]->setWord(BTOB1_dat[1]);
  tobs_v[4]->setET(gBlockTOBv[2][0]);
  tobs_v[4]->setEta(gBlockTOBeta[2][0]);
  tobs_v[4]->setPhi(gBlockTOBphi[2][0]);
  tobs_v[4]->setTobID(1);
  if(  gBlockTOBv[2][0] > gJ_ptMinToTopoCounts1 ) tobs_v[4]->setStatus(1);
  else tobs_v[4]->setStatus(0);
 
 
  BTOB2_dat[1] =  0x00000002;
  if(  gBlockTOBv[3][0] > gJ_ptMinToTopoCounts2 ) BTOB2_dat[1] = BTOB2_dat[1] | 0x00000080;
  BTOB2_dat[1] =  BTOB2_dat[1] | ( ( gBlockTOBv[3][0]   & 0x00000FFF ) << 8);
  BTOB2_dat[1] =  BTOB2_dat[1] | ( ( gBlockTOBeta[3][0] & 0x0000003F ) <<20);
  BTOB2_dat[1] =  BTOB2_dat[1] | ( ( gBlockTOBphi[3][0] & 0x0000001F ) <<26);
 
  tobs_v[5] = std::make_unique<gFEXJetTOB>();
  tobs_v[5]->setWord(BTOB2_dat[1]);
  tobs_v[5]->setET(gBlockTOBv[3][0]);
  tobs_v[5]->setEta(gBlockTOBeta[3][0]);
  tobs_v[5]->setPhi(gBlockTOBphi[3][0]);
  tobs_v[5]->setTobID(2);
  if(  gBlockTOBv[3][0] > gJ_ptMinToTopoCounts2 ) tobs_v[5]->setStatus(1);
  else tobs_v[5]->setStatus(0);
 
 
  CTOB1_dat[1] =  0x00000001;
  if(  gBlockTOBv[4][0] > gJ_ptMinToTopoCounts1 ) CTOB1_dat[1] = CTOB1_dat[1] | 0x00000080;
  CTOB1_dat[1] =  CTOB1_dat[1] | ( ( gBlockTOBv[4][0]   & 0x00000FFF ) << 8);
  CTOB1_dat[1] =  CTOB1_dat[1] | ( ( gBlockTOBeta[4][0] & 0x0000003F ) <<20);
  CTOB1_dat[1] =  CTOB1_dat[1] | ( ( gBlockTOBphi[4][0] & 0x0000001F) <<26);
 
 
  CTOB2_dat[1] =  0x00000002;
  if(  gBlockTOBv[5][0] > gJ_ptMinToTopoCounts2) CTOB2_dat[1] = CTOB2_dat[1] | 0x00000080;
  CTOB2_dat[1] =  CTOB2_dat[1] | ( ( gBlockTOBv[5][0]   & 0x00000FFF ) << 8);
  CTOB2_dat[1] =  CTOB2_dat[1] | ( ( gBlockTOBeta[5][0] & 0x0000003F ) <<20);
  CTOB2_dat[1] =  CTOB2_dat[1] | ( ( gBlockTOBphi[5][0] & 0x0000001F ) <<26);
 
 
  // subleading gBlocks
  // TOBs 6-9 are subleading gBlocks
  ATOB1_dat[2] =  0x00000003;
  if(  gBlockTOBv[0][1] > gJ_ptMinToTopoCounts2 ) ATOB1_dat[2] = ATOB1_dat[2] | 0x00000080;
  ATOB1_dat[2] =  ATOB1_dat[2] | ( ( gBlockTOBv[0][1]   & 0x00000FFF ) << 8);
  ATOB1_dat[2] =  ATOB1_dat[2] | ( ( gBlockTOBeta[0][1] & 0x0000003F ) <<20);
  ATOB1_dat[2] =  ATOB1_dat[2] | ( ( gBlockTOBphi[0][1] & 0x0000001F ) <<26);
 
  tobs_v[6] = std::make_unique<gFEXJetTOB>();
  tobs_v[6]->setWord(ATOB1_dat[2]);
  tobs_v[6]->setET(gBlockTOBv[0][1]);
  tobs_v[6]->setEta(gBlockTOBeta[0][1]);
  tobs_v[6]->setPhi(gBlockTOBphi[0][1]);
  tobs_v[6]->setTobID(3);
  if(  gBlockTOBv[0][1] > gJ_ptMinToTopoCounts2 ) tobs_v[6]->setStatus(1);
  else tobs_v[6]->setStatus(0);
 
 
  ATOB2_dat[2] =  0x00000004;
  if(  gBlockTOBv[1][1] > gJ_ptMinToTopoCounts1 ) ATOB2_dat[2] =  ATOB2_dat[2] | 0x00000080;
  ATOB2_dat[2] =  ATOB2_dat[2] | ( ( gBlockTOBv[1][1]   & 0x00000FFF ) << 8);
  ATOB2_dat[2] =  ATOB2_dat[2] | ( ( gBlockTOBeta[1][1] & 0x0000003F ) <<20);
  ATOB2_dat[2] =  ATOB2_dat[2] | ( ( gBlockTOBphi[1][1] & 0x0000001F ) <<26);
 
  tobs_v[7] = std::make_unique<gFEXJetTOB>();
  tobs_v[7]->setWord(ATOB2_dat[2]);
  tobs_v[7]->setET(gBlockTOBv[1][1]);
  tobs_v[7]->setEta(gBlockTOBeta[1][1]);
  tobs_v[7]->setPhi(gBlockTOBphi[1][1]);
  tobs_v[7]->setTobID(4);
  if(  gBlockTOBv[1][1] > gJ_ptMinToTopoCounts1 ) tobs_v[7]->setStatus(1);
  else tobs_v[7]->setStatus(0);
 
 
  BTOB1_dat[2] =  0x00000003;
  if(  gBlockTOBv[2][1] > gJ_ptMinToTopoCounts1 ) BTOB1_dat[2] = BTOB1_dat[2] | 0x00000080;
  BTOB1_dat[2] =  BTOB1_dat[2] | ( ( gBlockTOBv[2][1]   & 0x00000FFF ) << 8);
  BTOB1_dat[2] =  BTOB1_dat[2] | ( ( gBlockTOBeta[2][1] & 0x0000003F ) <<20);
  BTOB1_dat[2] =  BTOB1_dat[2] | ( ( gBlockTOBphi[2][1] & 0x0000001F ) <<26);
 
  tobs_v[8] = std::make_unique<gFEXJetTOB>();
  tobs_v[8]->setWord(BTOB1_dat[2]);
  tobs_v[8]->setET(gBlockTOBv[2][1]);
  tobs_v[8]->setEta(gBlockTOBeta[2][1]);
  tobs_v[8]->setPhi(gBlockTOBphi[2][1]);
  tobs_v[8]->setTobID(3);
  if(  gBlockTOBv[2][1] > gJ_ptMinToTopoCounts1 ) tobs_v[8]->setStatus(1);
  else tobs_v[8]->setStatus(0);
 
 
  BTOB2_dat[2] =  0x00000004;
  if(  gBlockTOBv[3][1] > gJ_ptMinToTopoCounts2 ) BTOB2_dat[2] = BTOB2_dat[2] | 0x00000080;
  BTOB2_dat[2] =  BTOB2_dat[2] | ( ( gBlockTOBv[3][1]   & 0x00000FFF ) << 8);
  BTOB2_dat[2] =  BTOB2_dat[2] | ( ( gBlockTOBeta[3][1] & 0x0000003F ) <<20);
  BTOB2_dat[2] =  BTOB2_dat[2] | ( ( gBlockTOBphi[3][1] & 0x0000001F ) <<26);
 
  tobs_v[9] = std::make_unique<gFEXJetTOB>();
  tobs_v[9]->setWord(BTOB2_dat[2]);
  tobs_v[9]->setET(gBlockTOBv[3][1]);
  tobs_v[9]->setEta(gBlockTOBeta[3][1]);
  tobs_v[9]->setPhi(gBlockTOBphi[3][1]);
  tobs_v[9]->setTobID(4);
  if(  gBlockTOBv[3][1] > gJ_ptMinToTopoCounts2 ) tobs_v[9]->setStatus(1);
  else tobs_v[9]->setStatus(0);
 
 
  CTOB1_dat[2] =  0x00000003;
  if(  gBlockTOBv[4][1] > gJ_ptMinToTopoCounts1 ) CTOB1_dat[2] = CTOB1_dat[2] | 0x00000080;
  CTOB1_dat[2] =  CTOB1_dat[2] | ( ( gBlockTOBv[4][1]   & 0x00000FFF ) << 8);
  CTOB1_dat[2] =  CTOB1_dat[2] | ( ( gBlockTOBeta[4][1] & 0x0000003F ) <<20);
  CTOB1_dat[2] =  CTOB1_dat[2] | ( ( gBlockTOBphi[4][1] & 0x0000001F ) <<26);
 
  CTOB2_dat[2] =  0x00000004;
  if(  gBlockTOBv[5][1] > gJ_ptMinToTopoCounts2 ) CTOB2_dat[2] = CTOB2_dat[2] | 0x00000080;
  CTOB2_dat[2] =  CTOB2_dat[2] | ( ( gBlockTOBv[5][1]   & 0x00000FFF ) << 8);
  CTOB2_dat[2] =  CTOB2_dat[2] | ( ( gBlockTOBeta[5][1] & 0x0000003F ) <<20);
  CTOB2_dat[2] =  CTOB2_dat[2] | ( ( gBlockTOBphi[5][1] & 0x0000001F ) <<26);
 
 
  // finally the main event -- lead gJET
  // according the specification https://docs.google.com/spreadsheets/d/15YVVtGofhXMtV7jXRFzWO0FVUtUAjS-X-aQjh3FKE_w/edit#gid=523371660
  // we should have an lsb of 3.2 GeV -- so ignore lower 4 bits.
 
  // TOBs 10-13 are leading gJets
  // shift is done before sorting as in firmware!
  int tobvShift = 8;
  int tobvMask   = 0x00000FFF;
 
 
  ATOB1_dat[3] =  0x00000005;
  if(  gJetTOBv[0] > gLJ_ptMinToTopoCounts2 ) ATOB1_dat[3] = ATOB1_dat[3] | 0x00000080;
  ATOB1_dat[3] =  ATOB1_dat[3] | ( ( gJetTOBv[0]   & tobvMask) << tobvShift);
  ATOB1_dat[3] =  ATOB1_dat[3] | ( ( gJetTOBeta[0] & 0x0000003F ) <<20);
  ATOB1_dat[3] =  ATOB1_dat[3] | ( ( gJetTOBphi[0] & 0x0000001F ) <<26);
 
  tobs_v[10] = std::make_unique<gFEXJetTOB>();
  tobs_v[10]->setWord(ATOB1_dat[3]);
  tobs_v[10]->setET(gJetTOBv[0]);
  tobs_v[10]->setEta(gJetTOBeta[0]);
  tobs_v[10]->setPhi(gJetTOBphi[0]);
  tobs_v[10]->setTobID(5);
  if(gJetTOBv[0] > gLJ_ptMinToTopoCounts2 ) tobs_v[10]->setStatus(1);
  else tobs_v[10]->setStatus(0);
 
 
  ATOB2_dat[3] =  0x00000006;
  if(  gJetTOBv[1] > gLJ_ptMinToTopoCounts1 )  ATOB2_dat[3] = ATOB2_dat[3] | 0x00000080;
  ATOB2_dat[3] =  ATOB2_dat[3] | ( ( gJetTOBv[1]   & tobvMask ) << tobvShift);
  ATOB2_dat[3] =  ATOB2_dat[3] | ( ( gJetTOBeta[1] & 0x0000003F ) <<20);
  ATOB2_dat[3] =  ATOB2_dat[3] | ( ( gJetTOBphi[1] & 0x0000001F ) <<26);
 
  tobs_v[11] = std::make_unique<gFEXJetTOB>();
  tobs_v[11]->setWord(ATOB2_dat[3]);
  tobs_v[11]->setET(gJetTOBv[1]);
  tobs_v[11]->setEta(gJetTOBeta[1]);
  tobs_v[11]->setPhi(gJetTOBphi[1]);
  tobs_v[11]->setTobID(6);
  if(  gJetTOBv[1] > gLJ_ptMinToTopoCounts1 ) tobs_v[11]->setStatus(1);
  else tobs_v[11]->setStatus(0);
 
 
  BTOB1_dat[3] =  0x00000005;
  if(  gJetTOBv[2] > gLJ_ptMinToTopoCounts1 ) BTOB1_dat[3] = BTOB1_dat[3] | 0x00000080;
  BTOB1_dat[3] =  BTOB1_dat[3] | ( ( gJetTOBv[2]   & tobvMask) << tobvShift);
  BTOB1_dat[3] =  BTOB1_dat[3] | ( ( gJetTOBeta[2] & 0x0000003F ) <<20);
  BTOB1_dat[3] =  BTOB1_dat[3] | ( ( gJetTOBphi[2] & 0x0000001F ) <<26);
 
  tobs_v[12] = std::make_unique<gFEXJetTOB>();
  tobs_v[12]->setWord(BTOB1_dat[3]);
  tobs_v[12]->setET(gJetTOBv[2]);
  tobs_v[12]->setEta(gJetTOBeta[2]);
  tobs_v[12]->setPhi(gJetTOBphi[2]);
  tobs_v[12]->setTobID(5);
  if(  gJetTOBv[2] > gLJ_ptMinToTopoCounts1 ) tobs_v[12]->setStatus(1);
  else tobs_v[12]->setStatus(0);
 
 
  BTOB2_dat[3] =  0x00000006;
  if(  gJetTOBv[3] > gLJ_ptMinToTopoCounts2 ) BTOB2_dat[3] = BTOB2_dat[3] | 0x00000080;
  BTOB2_dat[3] =  BTOB2_dat[3] | ( ( gJetTOBv[3]   & tobvMask   ) << tobvShift);
  BTOB2_dat[3] =  BTOB2_dat[3] | ( ( gJetTOBeta[3] & 0x0000003F ) <<20);
  BTOB2_dat[3] =  BTOB2_dat[3] | ( ( gJetTOBphi[3] & 0x0000001F ) <<26);
 
  tobs_v[13] = std::make_unique<gFEXJetTOB>();
  tobs_v[13]->setWord(BTOB2_dat[3]);
  tobs_v[13]->setET(gJetTOBv[3]);
  tobs_v[13]->setEta(gJetTOBeta[3]);
  tobs_v[13]->setPhi(gJetTOBphi[3]);
  tobs_v[13]->setTobID(6);
  if(  gJetTOBv[3] > gLJ_ptMinToTopoCounts2 ) tobs_v[13]->setStatus(1);
  else tobs_v[13]->setStatus(0);
 
  // // CECILIA - TO BE CHECKED
  CTOB1_dat[3] =  0x00000005;
  if(  gJetTOBv[4] > gLJ_ptMinToTopoCounts1 ) CTOB1_dat[3] = CTOB1_dat[3] | 0x00000080;
  CTOB1_dat[3] =  CTOB1_dat[3] | ( ( gJetTOBv[4]   & tobvMask) << tobvShift);
  CTOB1_dat[3] =  CTOB1_dat[3] | ( ( gJetTOBeta[4] & 0x0000003F ) <<20);
  CTOB1_dat[3] =  CTOB1_dat[3] | ( ( gJetTOBphi[4] & 0x0000001F ) <<26);
 
  CTOB2_dat[3] =  0x00000006;
  if(  gJetTOBv[5] > gLJ_ptMinToTopoCounts2 ) CTOB2_dat[3] = CTOB2_dat[3] | 0x00000080;
  CTOB2_dat[3] =  CTOB2_dat[3] | ( ( gJetTOBv[5]   & tobvMask   ) << tobvShift);
  CTOB2_dat[3] =  CTOB2_dat[3] | ( ( gJetTOBeta[5] & 0x0000003F ) <<20);
  CTOB2_dat[3] =  CTOB2_dat[3] | ( ( gJetTOBphi[5] & 0x0000001F ) <<26);
 
 
// saturation  currently not set in firmware
//
//
  bool setSat = 0; 
  if ( setSat) {
    if( gTOBsat[0] ) {
      ATOB1_dat[1] =  ( ATOB1_dat[1] | 0x80000000 );
      ATOB1_dat[3] =  ( ATOB1_dat[3] | 0x80000000 );
    }
 
    if( gTOBsat[1] ) {
      ATOB2_dat[1] =  ( ATOB2_dat[1] | 0x80000000 );
      ATOB2_dat[3] =  ( ATOB2_dat[3] | 0x80000000 );
    }
 
    if( gTOBsat[2] ) {
      BTOB1_dat[1] =  ( BTOB1_dat[1] | 0x80000000 );
      BTOB1_dat[3] =  ( BTOB1_dat[3] | 0x80000000 );
    }
 
    if( gTOBsat[3] ) {
      BTOB2_dat[1] =  ( BTOB2_dat[1] | 0x80000000 );
      BTOB2_dat[3] =  ( BTOB2_dat[3] | 0x80000000 );
    }
 
    if( gTOBsat[4] ) {
      CTOB1_dat[1] =  ( CTOB1_dat[1] | 0x80000000 );
      CTOB1_dat[3] =  ( CTOB1_dat[3] | 0x80000000 );
    }
 
    if( gTOBsat[5] ) {
      CTOB2_dat[1] =  ( CTOB2_dat[1] | 0x80000000 );
      CTOB2_dat[3] =  ( CTOB2_dat[3] | 0x80000000 );
    }
  }
 
  // zero tob 4 word
 
  ATOB1_dat[4] =  0 ;
  ATOB2_dat[4] =  0 ;
  BTOB1_dat[4] =  0 ;
  BTOB2_dat[4] =  0 ;
  CTOB1_dat[4] =  0 ;
  CTOB2_dat[4] =  0 ;
  
 
  // add seven bits as in the hardware of BCID in 6th TOB word ( index 5) 
  int BCID = 0;
 
  ATOB1_dat[5] =  ( (BCID&0x0000007F)<<8 ) ;
  ATOB2_dat[5] =  ( (BCID&0x0000007F)<<8 ) ;
  BTOB1_dat[5] =  ( (BCID&0x0000007F)<<8 ) ;
  BTOB2_dat[5] =  ( (BCID&0x0000007F)<<8 ) ;
  CTOB1_dat[5] =  ( (BCID&0x0000007F)<<8 ) ;
  CTOB2_dat[5] =  ( (BCID&0x0000007F)<<8 ) ;
  
  // 3 is gFEX FEX number, set CRC to zero for now 
  int CRC = 0;
 
  
  ATOB1_dat[6] = 0x000000BC | ( (BCID&0x0000000F)<<8 ) | (3<<12) | (CRC<<23);
  ATOB2_dat[6] = 0x000000BC | ( (BCID&0x0000000F)<<8 ) | (3<<12) | (CRC<<23);
  BTOB1_dat[6] = 0x000000BC | ( (BCID&0x0000000F)<<8 ) | (3<<12) | (CRC<<23);
  BTOB2_dat[6] = 0x000000BC | ( (BCID&0x0000000F)<<8 ) | (3<<12) | (CRC<<23);
  CTOB1_dat[6] = 0x000000BC | ( (BCID&0x0000000F)<<8 ) | (3<<12) | (CRC<<23);
  CTOB2_dat[6] = 0x000000BC | ( (BCID&0x0000000F)<<8 ) | (3<<12) | (CRC<<23);
 
 
 
  return tobs_v;
 
}
 
 
void gFEXJetAlgo::RemotePartialAB(const gTowersType& twrs, gTowersPartialSums & lps, gTowersPartialSums & rps) const {
 
  // Computes partial sums for FPGA A or B
  // twrs are the 32 x 12 = 284 gTowers in FPGA A or B
 
  // number of rows above/below for right partial sum
  // when sending to the right send values for largest partial sum first, i.e. for column 0
  typedef  std::array<std::array<int, 4>, 4> gTowersNeighbours;
  gTowersNeighbours NUpDwnR = {{ {{2,3,4,4}}, {{0,2,3,4}}, {{0,0,2,3}}, {{0,0,0,2}} }};
 
  // number of rows above or below for left partial sum
  // when sending to the left send values for smallest partial sum first (i.e. for column 8 )
  gTowersNeighbours NUpDwnL = {{ {{2,0,0,0}}, {{3,2,0,0}}, {{4,3,2,0}}, {{4,4,3,2}} }};
 
  // Do partial sum for output to right FPGA first
  for( int irow = 0; irow < FEXAlgoSpaceDefs::ABCrows; irow++ ){
    for(int rcolumn = 0; rcolumn<FEXAlgoSpaceDefs::n_partial; rcolumn++){
      // start calculating right partial sums for remote column rcolumn
      rps[irow][rcolumn] = 0;
      for(int lcolumn = 0; lcolumn<FEXAlgoSpaceDefs::n_partial; lcolumn++){
        // add in any energy from towers in this row
        // no need of modular arithmetic
        if ( NUpDwnR[rcolumn][lcolumn] > 0 ) {
          rps[irow][rcolumn] = rps[irow][rcolumn] + twrs[irow][lcolumn+8];
        }
        // now add rup1, rup2, rup3, rup4, ldn1, ldn2, ln3, ln4 -- use a loop instead of enumeratin in firmware
        for( int rowOff = 1 ; rowOff < NUpDwnR[rcolumn][lcolumn]+1; rowOff++){
          int rowModUp =  (irow + rowOff)%32;
          int rowModDn =  (irow - rowOff + 32 )%32;
          // printf("row %d remote column %d local column %d rowOff %d rowModUp %d rowModDn %d \n",
          rps[irow][rcolumn] =  rps[irow][rcolumn] + twrs[rowModUp][lcolumn+8] + twrs[rowModDn][lcolumn+8];
        }
      }
    }
  }
 
  // do  partial sum for output to left FPGA second
  for( int irow = 0; irow < FEXAlgoSpaceDefs::ABCrows; irow++ ){
    for(int rcolumn = 0; rcolumn<FEXAlgoSpaceDefs::n_partial; rcolumn++){
      // start calculating right partial sums for remote column rcolumn
      lps[irow][rcolumn] = 0;
      for(int lcolumn = 0; lcolumn<FEXAlgoSpaceDefs::n_partial; lcolumn++){
        // add in any energy from towers in this row
        // no need of modular arithmetic
        if ( NUpDwnL[rcolumn][lcolumn] > 0 ) {
          lps[irow][rcolumn] = lps[irow][rcolumn] + twrs[irow][lcolumn];
        }
        // now add rup1, rup2, rup3, rup4, ldn1, ldn2, ln3, ln4 -- use a loop instead of enumeratin in firmware
        for( int rowOff = 1 ; rowOff < NUpDwnL[rcolumn][lcolumn]+1; rowOff++){
          int rowModUp =  (irow + rowOff)%32;
          int rowModDn =  (irow - rowOff + 32 )%32;
          lps[irow][rcolumn] =  lps[irow][rcolumn] + twrs[rowModUp][lcolumn] + twrs[rowModDn][lcolumn];
        }
      }
    }
  }
 
}
 
 
void gFEXJetAlgo::RemotePartialCN(const gTowersJetEngine& twrs, gTowersPartialSums & rps ) const {
 
  typedef  std::array<std::array<int, 4>, 4> gTowersNeighbours;
  gTowersNeighbours NUpDwnR = {{ {{2,3,4,4}}, {{0,2,3,4}}, {{0,0,2,3}}, {{0,0,0,2}} }};
 
  // same as AB for right partial sum
  // starts from the righ most partial sum -- largest partial sum
  for( int irow = 0; irow < FEXAlgoSpaceDefs::ABCrows; irow++ ){
    for(int rcolumn = 0; rcolumn<FEXAlgoSpaceDefs::n_partial; rcolumn++){
      rps[irow][rcolumn] = 0;
      for(int lcolumn = 0; lcolumn<FEXAlgoSpaceDefs::n_partial; lcolumn++){
        if ( NUpDwnR[rcolumn][lcolumn] > 0 ) {
          rps[irow][rcolumn] = rps[irow][rcolumn] + twrs[irow][lcolumn+2];
        }
        for( int rowOff = 1 ; rowOff < NUpDwnR[rcolumn][lcolumn]+1; rowOff++){
          int rowModUp =  (irow + rowOff)%32;
          int rowModDn =  (irow - rowOff + 32 )%32;
          rps[irow][rcolumn] =  rps[irow][rcolumn] + twrs[rowModUp][lcolumn+2] + twrs[rowModDn][lcolumn+2];
        }
      }
    }
  }
}
 
 
void gFEXJetAlgo::RemotePartialCP(const gTowersJetEngine& twrs, gTowersPartialSums & lps ) const {
 
  typedef  std::array<std::array<int, 4>, 4> gTowersNeighbours;
  gTowersNeighbours NUpDwnL = {{ {{2,0,0,0}}, {{3,2,0,0}}, {{4,3,2,0}}, {{4,4,3,2}} }};
  // do  partial sum for output to left FPGA second
  for( int irow = 0; irow < FEXAlgoSpaceDefs::ABCrows; irow++ ){
    for(int rcolumn = 0; rcolumn<FEXAlgoSpaceDefs::n_partial; rcolumn++){
      // start calculating right partial sums for remote column rcolumn
      lps[irow][rcolumn] = 0;
      for(int lcolumn = 0; lcolumn<FEXAlgoSpaceDefs::n_partial; lcolumn++){
        // add in any energy from towers in this row
        // no need of modular arithmetic
        if ( NUpDwnL[rcolumn][lcolumn] > 0 ) {
          lps[irow][rcolumn] = lps[irow][rcolumn] + twrs[irow][lcolumn];
        }
        for( int rowOff = 1 ; rowOff < NUpDwnL[rcolumn][lcolumn]+1; rowOff++){
          int rowModUp =  (irow + rowOff)%32;
          int rowModDn =  (irow - rowOff + 32 )%32;
          lps[irow][rcolumn] =  lps[irow][rcolumn] + twrs[rowModUp][lcolumn] + twrs[rowModDn][lcolumn];
        }
      }
    }
  }
}
 
 
void gFEXJetAlgo::singleHalf(const gTowersType& twrs, gTowersType & FPGAsum) const {
  // Finds jets in a single FPGA
  // This version only gives sum in the right or left half of the FPGA
 
  // Number of up and down FPGAs to add
  std::array<int, 9> NUpDwn =  {{2,3,4,4,4,4,4,3,2}};
 
  //  Evaluate jet centered on each possible eta and phi gTower
  for( int irow = 0; irow < FEXAlgoSpaceDefs::ABCrows; irow++ ){
    for(int jcolumn = 0; jcolumn<FEXAlgoSpaceDefs::ABcolumns; jcolumn++){
      // zero jet sum here
      FPGAsum[irow][jcolumn] = 0;
 
      // local column goes from 0 to  8
      for(int localColumn = 0; localColumn<9; localColumn++){
        int sumColumn = jcolumn + localColumn - 4;
 
        // calculate min and max columns for FPGA half 
        int lmin = 0;
        int lmax = 5;
 
        if( jcolumn > 5 ) {
          lmin = 6;
          lmax = 11;
        }
 
        // check if tower to be summed is actually in FPGA 
        if( (sumColumn >= lmin)   && (sumColumn <= lmax) ) {
 
          // sum tower in the same row as jet center 
          FPGAsum[irow][jcolumn] = FPGAsum[irow][jcolumn] + twrs[irow][sumColumn]; 
 
          // add rows above and below according NUpDwn
          // localRow goes from 1 to 2, 1 to 3 or 1 to 4 
          for(int localRow = 1; localRow<=NUpDwn[localColumn]; localRow++){
            int krowUp = (irow + localRow);
            if( krowUp > 31 ) krowUp = krowUp - 32; 
            int krowDn = (irow - localRow);
            if( krowDn < 0 ) krowDn = krowDn + 32; 
            FPGAsum[irow][jcolumn] =  FPGAsum[irow][jcolumn] +
                                      twrs[krowUp][sumColumn] + 
                                      twrs[krowDn][sumColumn];
          }
        }   
      } 
    }
  }
}
 
 
void gFEXJetAlgo::InternalPartialAB(const gTowersType & twrs, gTowersPartialSums & lps, gTowersPartialSums & rps ) const{
  
  // Computes partial sums for FPGA A or B
  // twrs are the 32 x 12 = 284 gTowers in FPGA A or B
  //
  // lps:    partial sum for left hand side of FPGA
  // rps:    partial sum for  right half of the FPGA
 
  // NOTE rps is available first 
  
  // number of rows above/below for right partial sum
  // when sending to the right send values for largest partial sum first, i.e. for column 6
  unsigned int NUpDwnR[4][4] = { {2,3,4,4}, {0,2,3,4}, {0,0,2,3}, {0,0,0,2} };
  
  // number of rows above or below for left partial sum
  // when sending to the left send values for smallest partial sumn first (i.e. for column 2 ) 
  unsigned int NUpDwnL[4][4] = { {2,0,0,0}, {3,2,0,0}, {4,3,2,0}, {4,4,3,2} }; 
 
  
  // Do partial sum for output to right side FPGA first
  // for the right partial sum this starts from the right-most value first 
  for(unsigned int irow = 0; irow < FEXAlgoSpaceDefs::ABCrows; irow++ ){
    for(unsigned int rcolumn = 0; rcolumn<FEXAlgoSpaceDefs::n_partial; rcolumn++){
      // start calculating right partial sums for remote column rcolumn 
      rps[irow][rcolumn] = 0;
      for(unsigned int lcolumn = 0; lcolumn<FEXAlgoSpaceDefs::n_partial; lcolumn++){
        // add in any energy from towers in this row
        // no need of modular arithmetic
        if ( NUpDwnR[rcolumn][lcolumn] > 0 ) {
          // this is partial sum for the right half of the FPGA -- columns 2,3,4,5
          rps[irow][rcolumn] = rps[irow][rcolumn] + twrs[irow][lcolumn+2];
        }
        // now add rup1, rup2, rup3, rup4, ldn1, ldn2, ln3, ln4 -- use a loop instead of enumeratin in firmware  
        for(unsigned int rowOff = 1 ; rowOff < NUpDwnR[rcolumn][lcolumn]+1; rowOff++){
          const int row = static_cast<int>(irow);
          const int off = static_cast<int>(rowOff);
          int rowModUp =  (row + off)%32;
          int rowModDn =  (row - off + 32 )%32;
          // this is partial sum for the right half of the FPGA -- columns 2,3,4,5
          rps[irow][rcolumn] =  rps[irow][rcolumn] + twrs[rowModUp][lcolumn+2] + twrs[rowModDn][lcolumn+2];
        }
      }
    }
  }
  // do  partial sum for output to left half of the FPGA second
  // for the left partial sum this starts also with the right most column -- which is the smallest sum 
  for(unsigned int irow = 0; irow < FEXAlgoSpaceDefs::ABCrows; irow++ ){
    for(unsigned int rcolumn = 0; rcolumn<FEXAlgoSpaceDefs::n_partial; rcolumn++){
      // start calculating right partial sums for remote column rcolumn 
      lps[irow][rcolumn] = 0;
      for(unsigned int lcolumn = 0; lcolumn<FEXAlgoSpaceDefs::n_partial; lcolumn++){
        // add in any energy from towers in this row
        // no need of modular arithmetic
        if ( NUpDwnL[rcolumn][lcolumn] > 0 ) {
          lps[irow][rcolumn] = lps[irow][rcolumn] + twrs[irow][lcolumn+6];
        }
        // now add rup1, rup2, rup3, rup4, ldn1, ldn2, ln3, ln4 -- use a loop instead of enumeratin in firmware  
        for(unsigned int rowOff = 1 ; rowOff < NUpDwnL[rcolumn][lcolumn]+1; rowOff++){
          int rowModUp =  (irow + rowOff)%32;
          int rowModDn =  (irow - rowOff + 32 )%32;
          // this is partial sum for the left half of the FPGA -- columns 6,7,8,9 
          lps[irow][rcolumn] =  lps[irow][rcolumn] + twrs[rowModUp][lcolumn+6] + twrs[rowModDn][lcolumn+6];
        }
      }
    } 
  }
}
 
 
void gFEXJetAlgo::addInternalLin(gTowersType & jets, gTowersPartialSums & partial) const{
  // add parial sums for left side of FPGA
  for(unsigned int irow=0;irow<FEXAlgoSpaceDefs::ABCrows;irow++){
    for(unsigned int ipartial= 0; ipartial <FEXAlgoSpaceDefs::n_partial; ipartial++){
      jets[irow][2+ipartial] =  jets[irow][2+ipartial] + partial[irow][ipartial];
    }
  }
}
 
 
void gFEXJetAlgo::addInternalRin(gTowersType & jets, gTowersPartialSums & partial) const{
  // add parial sums for right side of FPGA 
  for(unsigned int irow=0;irow<FEXAlgoSpaceDefs::ABCrows;irow++){
    for(unsigned int ipartial= 0; ipartial <FEXAlgoSpaceDefs::n_partial; ipartial++){
      jets[irow][6+ipartial] =  jets[irow][6+ipartial] + partial[irow][ipartial];
    }
  }
}
 
 
void gFEXJetAlgo::pileUpCorrectionAB(gTowersType &jets, int puc) const {
  int rows = jets.size();
  int cols = jets[0].size();
  for(int irow=0;irow<rows;irow++){
    for(int icolumn=0;icolumn<cols;icolumn++){
      jets[irow][icolumn] =   jets[irow][icolumn] - puc;
    }
  }
}
 
 
void gFEXJetAlgo::ZeroNegative(gTowersType & jets) const{
  for(unsigned int irow = 0; irow < FEXAlgoSpaceDefs::ABCrows; irow++ ){
    for(unsigned int icolumn =0; icolumn<FEXAlgoSpaceDefs::ABcolumns; icolumn++){
      if(jets[irow][icolumn] < 0) jets[irow][icolumn] = 0; 
    }
  }
}
 
 
// https://gitlab.cern.ch/atlas-l1calo/gfex/firmware/-/blob/devel/common/jet_finder/HDL/jet_eng.vhd#L538
void gFEXJetAlgo::SaturateJets( gTowersType & jets, const gTowersType & sat, int fpga ) const {
  // TODO: Temporary fix to match FW saturation eta-alignment bug.
  // In firmware, sum69_ovf = delayK(dat_ovf_in, SUM69_OVF_DELAY=16).
  // dat_ovf_in for col K arrives at clock K; sum69_ovf fires at clock K+16.
  // sum69o for col J is computed at clock 17+J (LPS_A_DLY+J).
  // So overflow at col K fires during sum69o for col K+16-17 = K-1.
  // Engine-boundary columns:
  //  - First column (local col 0 = global 0,6): K-1 falls in the dead/gap
  //    phase of the serialiser → no valid target, skip (icol%6 != 0).
  //  - Last column (local col 5 = global 5,11): the override is gated by
  //    have_seed_d5 in jet_eng.vhd.  For extended columns the seed is
  //    computed from regular etower+htower only (the extended xetower
  //    contribution is added after the seed stage via lps_a), so the seed
  //    typically fails and the override is blocked.  This is FPGA-specific:
  //      FPGA A (fpga=0): cols 5,11 are regular → override goes through.
  //      FPGA B (fpga=1): col 11 is extended (xetower+xhtower) → skip.
  //                        col 5 is regular → override goes through.
  //      FPGA C (fpga=2): all columns extended → skip all last cols.
  // @see jet_eng.vhd: sum69_ovf forces sum69o to max positive
  // @see jet_eng.vhd: have_seed_d5 gates delm_din, blocking override
  // @see delayK.vhd: DLY FFs with combinatorial in/out
  // Revert this to use icol (not icol-1) when the FW saturation alignment is fixed.
  for(unsigned int irow = 0; irow < FEXAlgoSpaceDefs::ABCrows; irow++ ){
    for(unsigned int icol = 0; icol < FEXAlgoSpaceDefs::ABcolumns; icol++){
      if(sat[irow][icol] && (icol % 6 != 0)){
        if (fpga == 1 && icol == 11) continue;   // FPGA B col 11 extended
        if (fpga == 2 && icol % 6 == 5) continue; // FPGA C all last cols extended
        jets[irow][icol - 1] = 0x0003ffff;
      }
    }
  }
}
 
void gFEXJetAlgo::SaturateBlocks( gTowersType & gBlkSum, const gTowersType & sat, int fpga ) const {
  // TODO: Temporary fix to match FW saturation eta-alignment bug.
  // In firmware, seed_ovf = delayK(dat_ovf_in, SEED_OVF_DELAY=10).
  // dat_ovf_in for col K arrives at clock K; seed_ovf fires at clock K+10.
  // et9 for col J is computed at clock 11+J (SEED_DLY+J).
  // So overflow at col K fires during et9 for col K+10-11 = K-1.
  // First column (icol%6==0) is always skipped (pulse falls in dead/gap).
  // Last column skipping is FPGA-specific (see SaturateJets comments):
  //   FPGA A: no last-col skip (cols 5,11 regular).
  //   FPGA B: skip col 11 only (extended).
  //   FPGA C: skip all last cols (all extended).
  // @see jet_eng.vhd lines 849-863: seed_ovf forces et9/seed_out to max
  // @see delayK.vhd: DLY FFs with combinatorial in/out
  // Revert this to use icol (not icol-1) when the FW saturation alignment is fixed.
  for(unsigned int irow = 0; irow < FEXAlgoSpaceDefs::ABCrows; irow++ ){
    for(unsigned int icol = 0; icol < FEXAlgoSpaceDefs::ABcolumns; icol++){
      if(sat[irow][icol] && (icol % 6 != 0)){
        if (fpga == 1 && icol == 11) continue;   // FPGA B col 11 extended
        if (fpga == 2 && icol % 6 == 5) continue; // FPGA C all last cols extended
        gBlkSum[irow][icol - 1] = 0x00000fff;
      }
    }
  }
}
 
 
 
void gFEXJetAlgo::gBlockAB(const gTowersType & twrs, gTowersType & gBlkSum, gTowersType & hasSeed, int seedThreshold) const {
 
  const int rows = twrs.size();
  const int cols = twrs[0].size();
  for( int irow = 0; irow < rows; irow++ ){
    for(int jcolumn = 0; jcolumn<cols; jcolumn++){
      // zero jet sum here
      gBlkSum[irow][jcolumn] = 0;
      int krowUp = (irow + 1)%32;
      int krowDn = (irow - 1 +32)%32;
      if( (jcolumn == 0) || (jcolumn == 6) ) {
        //left edge case
        gBlkSum[irow][jcolumn] =
        twrs[irow][jcolumn]   + twrs[krowUp][jcolumn]   + twrs[krowDn][jcolumn] +
        twrs[irow][jcolumn+1] + twrs[krowUp][jcolumn+1] + twrs[krowDn][jcolumn+1];
      } else if( (jcolumn == 5) || (jcolumn == 11)) {
        //  right edge case
        gBlkSum[irow][jcolumn] =
        twrs[irow][jcolumn]   + twrs[krowUp][jcolumn]   + twrs[krowDn][jcolumn] +
        twrs[irow][jcolumn-1] + twrs[krowUp][jcolumn-1] + twrs[krowDn][jcolumn-1];
      } else{
        // normal case; jcolumn is not 11 so does not overrun
        //coverity[OVERRUN:FALSE]
        gBlkSum[irow][jcolumn] =
        twrs[irow][jcolumn]   + twrs[krowUp][jcolumn]   + twrs[krowDn][jcolumn]   +
        twrs[irow][jcolumn-1] + twrs[krowUp][jcolumn-1] + twrs[krowDn][jcolumn-1] +
        twrs[irow][jcolumn+1] + twrs[krowUp][jcolumn+1] + twrs[krowDn][jcolumn+1];
      }
 
      if( gBlkSum[irow][jcolumn]  > seedThreshold) {
        hasSeed[irow][jcolumn] = 1;
      } else {
        hasSeed[irow][jcolumn] = 0;
      }
    
      if ( gBlkSum[irow][jcolumn] < 0 )       
        gBlkSum[irow][jcolumn] = 0;
 
      // was bits 11+3 downto 3, now is 11 downto 0 
      if ( gBlkSum[irow][jcolumn] > FEXAlgoSpaceDefs::gBlockMax )   {
        gBlkSum[irow][jcolumn] =  FEXAlgoSpaceDefs::gBlockMax;
      }
    }
  }
}
 
 
 
void gFEXJetAlgo::blkOutAB(gTowersType & blocks, std::array<int, 32> & jetOutL, std::array<int, 32> & etaIndL, std::array<int, 32> & jetOutR, std::array<int, 32> & etaIndR) const{
  
  // find maximum in each jet engine for gBlocks  (not done in hardware) 
 
  //loop over left engines 
  for(unsigned int ieng=0; ieng<FEXAlgoSpaceDefs::ABCrows; ieng++){
    for(unsigned int localEta = 0; localEta<6; localEta++){
      if( blocks[ieng][localEta] > jetOutL[ieng]  ){
        jetOutL[ieng] = blocks[ieng][localEta];
        etaIndL[ieng] = localEta;
      }
    }
  }
  // loop over right engines 
  for(unsigned int ieng=0; ieng<FEXAlgoSpaceDefs::ABCrows; ieng++){
    for(unsigned int localEta = 0; localEta<6; localEta++){
      if(  blocks[ieng][localEta+6] > jetOutR[ieng] ) {
        jetOutR[ieng] = blocks[ieng][localEta+6];
        etaIndR[ieng] = localEta;
      }
    }
  }
}
 
 
void gFEXJetAlgo::gBlockMax2(const gTowersType & gBlkSum, int BjetColumn, int localColumn, std::array<int, 3> & gBlockV, std::array<int, 3> & gBlockEta, std::array<int, 3> & gBlockPhi) const {
  //  gBlkSum are the 9 or 6 gTower sums  
  //  BjetColumn is the Block Column -- 0 for CN, 1, 2 for A 3, 4 for B and 5 for CP 
  //  gBlockV is the array of values currently 2 
  //  gBlockEta is the eta in global 
 
  gTowersJetEngine gBlkSumC;
 
  // copy the correct column
  for( int irow = 0; irow<FEXAlgoSpaceDefs::ABCrows; irow++){
    for( int icolumn =0; icolumn<FEXAlgoSpaceDefs::ABCcolumnsEng; icolumn++){
      gBlkSumC[irow][icolumn] = gBlkSum[irow][icolumn + localColumn*FEXAlgoSpaceDefs::ABCcolumnsEng];
    }
  }
 
  gBlockMax192(gBlkSumC, gBlockV, gBlockEta, gBlockPhi, 0);
 
  for(int i = -1; i<2; i++){
    int iGlobal = i + gBlockPhi[0];
    // map row number in to 0-31
    if ( iGlobal < 0 ) iGlobal  = iGlobal + 32;
    if ( iGlobal > 31 ) iGlobal = iGlobal - 32;
    // don't do anything outside of the six columsn
    for( int j = -1; j<2; j++){
      int jGlobal = j + gBlockEta[0];
      if( (jGlobal > -1)  && (jGlobal < 6) ) {
        gBlkSumC[iGlobal][jGlobal] = 0;
      }
    }
  }
 
  gBlockMax192(gBlkSumC, gBlockV, gBlockEta, gBlockPhi, 1);
 
  // need to wait until the end for this!
  gBlockEta[0] =  gBlockEta[0] + 2 + 6*BjetColumn;
  gBlockEta[1] =  gBlockEta[1] + 2 + 6*BjetColumn;
}
 
 
void gFEXJetAlgo::gBlockMax192(const gTowersJetEngine& gBlkSum,
                               std::array<int, 3> & gBlockVp,
                               std::array<int, 3> & gBlockEtap,
                               std::array<int, 3> & gBlockPhip,
                               int index) const {
  // gBLKSum are the sums
  // gBlockVp is the returned value for the max block
  // gBlockEtap is the eta in the local  corrdinate system
  // gBlockPhip is the phi (global and local are the same) 
 
  int inpv[192]{};
  int maxv1[96]{};
  int maxv2[48]{};
  int maxv3[24]{};
  int maxv4[12]{};
  int maxv5[6]{};
  int maxv6[3]{};
 
  int inpi[192]{};
  int maxi1[96]{};
  int maxi2[48]{};
  int maxi3[24]{};
  int maxi4[12]{};
  int maxi5[6]{};
  int maxi6[3]{};
 
  int maxvall;
  int maxiall;
 
 
  // ABCcolumnsEng is 6 in hardware
  int maxv = 0;
 
  for(unsigned int icolumn = 0;  icolumn<FEXAlgoSpaceDefs::ABCcolumnsEng; icolumn++){
    for(unsigned int irow = 0; irow<FEXAlgoSpaceDefs::ABCrows; irow++){
      inpv[irow + icolumn*FEXAlgoSpaceDefs::ABCrows] = gBlkSum[irow][icolumn];
      inpi[irow + icolumn*FEXAlgoSpaceDefs::ABCrows] = irow  + icolumn*FEXAlgoSpaceDefs::ABCrows;
 
      if( gBlkSum[irow][icolumn] > maxv){
        maxv = gBlkSum[irow][icolumn];
      }
    }
  }
 
  for(int i = 0; i<96; i++){
    if( inpv[2*i+1] > inpv[2*i] ) {
      maxv1[i] = inpv[2*i+1];
      maxi1[i] = inpi[2*i+1];
    } else {
      maxv1[i] = inpv[2*i];
      maxi1[i] = inpi[2*i];
    }
  }
 
  for(int i = 0; i<48; i++){
    if( maxv1[2*i+1] > maxv1[2*i] ) {
      maxv2[i] = maxv1[2*i+1];
      maxi2[i] = maxi1[2*i+1];
    } else {
      maxv2[i] = maxv1[2*i];
      maxi2[i] = maxi1[2*i];
    }
  }
 
  for(int i = 0; i<24; i++){
    if( maxv2[2*i+1] > maxv2[2*i] ) {
      maxv3[i] = maxv2[2*i+1];
      maxi3[i] = maxi2[2*i+1];
    } else {
      maxv3[i] = maxv2[2*i];
      maxi3[i] = maxi2[2*i];
    }
  }
 
  for(int i = 0; i<12; i++){
    if( maxv3[2*i+1] > maxv3[2*i] ) {
      maxv4[i] = maxv3[2*i+1];
      maxi4[i] = maxi3[2*i+1];
    } else {
      maxv4[i] = maxv3[2*i];
      maxi4[i] = maxi3[2*i];
    }
  }
 
  for(int i = 0; i<6; i++){
    if( maxv4[2*i+1] > maxv4[2*i] ) {
      maxv5[i] = maxv4[2*i+1];
      maxi5[i] = maxi4[2*i+1];
    } else {
      maxv5[i] = maxv4[2*i];
      maxi5[i] = maxi4[2*i];
    }
  }
 
  for(int i = 0; i<3; i++){
    if( maxv5[2*i+1] > maxv5[2*i] ) {
      maxv6[i] = maxv5[2*i+1];
      maxi6[i] = maxi5[2*i+1];
    } else {
      maxv6[i] = maxv5[2*i];
      maxi6[i] = maxi5[2*i];
    }
  }
 
  if( ( maxv6[1] > maxv6[0] ) && ( maxv6[1] > maxv6[2] ) ) {
    maxvall = maxv6[1];
    maxiall = maxi6[1];
  } else {
    if( maxv6[0] > maxv6[2] ){
      maxvall = maxv6[0];
      maxiall = maxi6[0];
    } else {
      maxvall = maxv6[2];
      maxiall = maxi6[2];
    }
  }
 
 
  gBlockVp[index]   =  maxvall ;
  gBlockEtap[index] =  maxiall/32;
  gBlockPhip[index] =  maxiall%32;
 
}
 
 
void gFEXJetAlgo::gBlockVetoAB(gTowersType &jets, gTowersType& hasSeed) const {
  int rows = jets.size();
  int cols = jets[0].size();
  for( int irow = 0; irow < rows; irow++ ){
    for(int jcolumn = 0; jcolumn < cols; jcolumn++){
      if( hasSeed[irow][jcolumn] == 0 ) {
        // set to negative value as in hardware
        // https://gitlab.cern.ch/atlas-l1calo/gfex/firmware/-/blob/devel/common/jet_finder/HDL/jet_eng.vhd#L561
        jets[irow][jcolumn] = 0XFFFFF000; 
      }
    }
  }
}
 
 
void gFEXJetAlgo::addRemoteRin(gTowersType &jets, const gTowersPartialSums &partial,
                               int ps_upper, int ps_lower, int ps_shift) const {
 
  // See https://gitlab.cern.ch/atlas-l1calo/gfex/firmware/-/blob/kw-dev/jwj_verif/common/jet_finder/HDL/ab_encode.vhd
  
  int rows = partial.size();
  int cols = partial[0].size();
  // add partial sums
  for(int irow=0;irow<rows;irow++){
    for(int ipartial= 0; ipartial <cols; ipartial++){
      // current behavior
      int truncPart = -1;
      if( partial[irow][ipartial] > ps_upper ) {
        truncPart = ps_upper;
      } else if ( partial[irow][ipartial] < ps_lower ) {
        truncPart = ps_lower; 
      } else {
        truncPart = partial[irow][ipartial];
      }
      // change LSB from 200 MeV to 1600  MeV and then back to 200 MeV.   
      truncPart = (truncPart >> ps_shift );
      truncPart = (truncPart << ps_shift );
 
      jets[irow][8+ipartial] =  jets[irow][8+ipartial]  + truncPart;
    }
  }
}
 
void gFEXJetAlgo::addRemoteLin(gTowersType &jets, const gTowersPartialSums &partial,
                              int ps_upper, int ps_lower, int ps_shift) const {
 
  int rows = partial.size();
  int cols = partial[0].size();
  // add partial sums
  for(int irow=0;irow<rows;irow++){
    for(int ipartial= 0; ipartial <cols; ipartial++){
      // current behavior
      int truncPart = -1;
      if( partial[irow][ipartial] > ps_upper ) {
        truncPart = ps_upper;
      } else if ( partial[irow][ipartial] < ps_lower ) {
          truncPart = ps_lower; 
      } else {
          truncPart = partial[irow][ipartial];
      }
      // change LSB from 200 MeV to 1600  MeV and then back to 200 MeV. 
      truncPart = (truncPart >> ps_shift );
      truncPart = (truncPart << ps_shift );
 
      jets[irow][ipartial] =  jets[irow][ipartial]  + truncPart;
    }
  }
}
 
 
void gFEXJetAlgo::addRemoteCNin(gTowersType & jets, const gTowersPartialSums & partial,
                                int ps_upper, int ps_lower, int ps_shift ) const {
 
  int rows = partial.size();
  int cols = partial[0].size();
  // add partial sums 
  for(int irow=0; irow < rows; irow++){
    for(int ipartial= 0; ipartial <cols; ipartial++){
      // current behavior
      int truncPart = -1;
      if( partial[irow][ipartial] > ps_upper ) {
        truncPart = ps_upper;
      } else if ( partial[irow][ipartial] < ps_lower ) {
          truncPart = ps_lower; 
      } else {
          truncPart = partial[irow][ipartial];
      }
 
      truncPart = (truncPart >> ps_shift );
      truncPart = (truncPart << ps_shift );
 
      jets[irow][ipartial+2] =  jets[irow][ipartial+2]  + truncPart;
      
    }
  }
}
 
void gFEXJetAlgo::addRemoteCPin(gTowersType & jets, const gTowersPartialSums & partial,
                                int ps_upper, int ps_lower, int ps_shift ) const {
 
  int rows = partial.size();
  int cols = partial[0].size();
  // add partial sums 
  for(int irow=0;irow<rows;irow++){
    for(int ipartial= 0; ipartial <cols; ipartial++){
      int truncPart = -1;
      if( partial[irow][ipartial] > ps_upper ) {
        truncPart = ps_upper;
      } else if ( partial[irow][ipartial] < ps_lower ) {
          truncPart = ps_lower; 
      } else {
          truncPart = partial[irow][ipartial];
      }
 
      truncPart = (truncPart >> ps_shift );
      truncPart = (truncPart << ps_shift );
      
      jets[irow][ipartial+6] =  jets[irow][ipartial+6]  + truncPart;
      
    }
  }
}
 
 
void gFEXJetAlgo::gJetVetoAB( gTowersType &twrs ,int jet_threshold ) const {
  int rows = twrs.size();
  int cols = twrs[0].size();
  for( int irow = 0; irow < rows; irow++ ){
    for(int jcolumn = 0; jcolumn<cols; jcolumn++){
      if( twrs[irow][jcolumn] < jet_threshold+1 ) {
        twrs[irow][jcolumn] = 0; 
      }
    } 
  }
}
 
 
void gFEXJetAlgo::jetOutAB(const gTowersType & jets,
                           std::array<int, 32> & jetOutL, std::array<int, 32> & etaIndL,
                           std::array<int, 32> & jetOutR, std::array<int, 32> & etaIndR ) const {
  // find maximum in each jet engine or either gJets and requires corresponding gBlock be above threhsold
  //loop over left engines
  for(int ieng=0; ieng<FEXAlgoSpaceDefs::ABCrows; ieng++){
    jetOutL[ieng] = 0;
    etaIndL[ieng] = 0;
    for(int localEta = 0; localEta<6; localEta++){
      if( jets[ieng][localEta] > jetOutL[ieng] ){
        jetOutL[ieng] = jets[ieng][localEta];
        etaIndL[ieng] = localEta;
      }
    }
    // truncation and mapping from 12 to 15 bits 
    // https://gitlab.cern.ch/atlas-l1calo/gfex/firmware/-/blob/kw-dev/jet_finder_abc/common/jet_finder/HDL/jet_eng.vhd#L561
    // https://gitlab.cern.ch/atlas-l1calo/gfex/firmware/-/blob/kw-dev/jet_finder_abc/common/jet_finder/HDL/jet_eng.vhd#L561
    // Turncate to 15 bits as in VHDL  (Corresponds to 13 TeV)
    
    if( jetOutL[ieng] >  FEXAlgoSpaceDefs::gJetMax )  jetOutL[ieng] = FEXAlgoSpaceDefs::gJetMax;
    if( jetOutL[ieng] <  0  )       jetOutL[ieng] = 0;
  }
  // loop over right engines
  for(int ieng=0; ieng<FEXAlgoSpaceDefs::ABCrows; ieng++){
    jetOutR[ieng] = 0;
    etaIndR[ieng] = 0;
    for(int localEta = 0; localEta < 6; localEta++){
      if( jets[ieng][localEta+6] > jetOutR[ieng] ){
       jetOutR[ieng] = jets[ieng][localEta+6];
       etaIndR[ieng] = localEta;
      }
    }
    // Turncate to 15 bits as in VHDL (orresponds to 13 TeV) 
    if( jetOutR[ieng] >  FEXAlgoSpaceDefs::gJetMax )      jetOutR[ieng] = FEXAlgoSpaceDefs::gJetMax;
    if( jetOutR[ieng] <    0     )      jetOutR[ieng] = 0;
    // https://gitlab.cern.ch/atlas-l1calo/gfex/firmware/-/blob/kw-dev/jet_finder_abc/common/jet_finder/HDL/jet_finder_abc.vhd#L1103
    // https://gitlab.cern.ch/atlas-l1calo/gfex/firmware/-/blob/kw-dev/jet_finder_abc/common/jet_finder/HDL/jet_finder_abc.vhd#L1119
 
  }
 
}
 
 
void gFEXJetAlgo::pileUpCalculation(gTowersType &twrs, int rhoThreshold_Max, int inputScale,  int &PUCp, int &PUC_JWJ /*, int &PUChres*/) const {
  // input are 50 MeV "fine" scale towers (i.e. inputScale = 1)
  // to use 200 MeV towers use inputScale = 4  
  // PUCp output is the pileup correction for 69 towers at 200 MeV energy scale 
 
  int rows = twrs.size();
  int cols = twrs[0].size();
  int pucSum = 0;
  int pucSumJWJ = 0;
  int nSum   = 0; 
  for(int irow=0; irow<rows; irow++){
    for( int icolumn=0; icolumn<cols; icolumn++){
      int fineGT = twrs[irow][icolumn]*inputScale;
      //  set floor and ceiling here (currently 255 and -256 ) 
      if (fineGT > FEXAlgoSpaceDefs::fineCeiling ) fineGT = FEXAlgoSpaceDefs::fineCeiling;
      if (fineGT < FEXAlgoSpaceDefs::fineFloor ) fineGT = FEXAlgoSpaceDefs::fineFloor;
 
      if( fineGT < rhoThreshold_Max ) {
      pucSum = pucSum + fineGT;
      nSum = nSum + 1;
      }
    }
  }
  // in firmware this is done with a lookup table see https://gitlab.cern.ch/atlas-l1calo/gfex/firmware/-/blob/devel/common/jet_finder/HDL/inv_lut19.vhd
  // See also https://gitlab.cern.ch/atlas-l1calo/gfex/firmware/-/blob/devel/common/jet_finder/HDL/gt_build_all_AB.vhd#L1471
 
  // oneOverN is stored as a 32 bit number in inv_lut19
  unsigned int oneOverNTab = 69<<25;
  unsigned int JWJoneOverNTab = 69<<25;
  if( nSum > 0 && nSum < 385  ) {
    oneOverNTab = FEXAlgoSpaceDefs::inv19[nSum]; 
    JWJoneOverNTab = FEXAlgoSpaceDefs::jwjinv19[nSum];
  } else {
    oneOverNTab = 0;
    JWJoneOverNTab = 0;
  }
 
  int oneOverN = oneOverNTab;
  int JWJoneOverN = JWJoneOverNTab;
 
  // largest value should be 255*2^14 ~ 2^22  -- should easily fit in int -- try expliciting putting in int here.  
  pucSumJWJ = pucSum * JWJoneOverN;
  pucSum  = pucSum * oneOverN;
 
  // PUChres = pucSum;
 
  // current system 19 bits in table (69*4096 is max value) sign exten
  pucSum = ( pucSum >> 14); 
  pucSumJWJ = ( pucSumJWJ >> 16);
  PUCp   = pucSum;
  PUC_JWJ = pucSumJWJ;
 
}
 
void gFEXJetAlgo::gJetTOBgen(const std::array<int, FEXAlgoSpaceDefs::ABCrows>& jetOut,
                             const std::array<int, FEXAlgoSpaceDefs::ABCrows>&  etaInd,
                             int TOBnum, int jetThreshold, std::array<int, FEXAlgoSpaceDefs::gJetTOBfib> & gJetTOBs,
                             std::array<int, FEXAlgoSpaceDefs::gJetTOBfib> & gJetTOBv,
                             std::array<int, FEXAlgoSpaceDefs::gJetTOBfib> & gJetTOBeta,
                             std::array<int, FEXAlgoSpaceDefs::gJetTOBfib> & gJetTOBphi ) const {
 
  int jetOutZS[FEXAlgoSpaceDefs::ABCrows]{};
  // apply the tobthreshold to the values 
  //note that jetThreshold is not a configurable parameter in firmware, it is used to check that jet values are positive
  for( int irow =0; irow<FEXAlgoSpaceDefs::ABCrows; irow++){
    if(  jetOut[irow] > jetThreshold ) {
      jetOutZS[irow] = jetOut[irow];
    } else {
        jetOutZS[irow] = 0 ;
    }   
  }
 
 
  // offset of TOBs according to official format
  int etaOff[FEXAlgoSpaceDefs::gJetTOBfib] = {8,14,20,26,2,32};
 
  // see tob_gen.vhd
  int l1mv[16]{};
  int l1met[16]{};
  int l1mphi[16]{};
 
  int l2mv[8]{};
  int l2met[8]{};
  int l2mphi[8]{};
 
  int l3mv[4]{};
  int l3met[4]{};
  int l3mphi[4]{};
 
  int l4mv[2]{};
  int l4met[2]{};
  int l4mphi[2]{};
  
  for(int i=0; i<16; i++){
    if( jetOut[2*i + 1] > jetOutZS[2*i] ){
      l1mv[i]   = jetOutZS[2*i + 1];
      l1met[i]  = etaInd[2*i + 1];
      l1mphi[i] =        2*i + 1;
    } else {
      l1mv[i]   = jetOutZS[2*i ];
      l1met[i]  = etaInd[2*i ];
      l1mphi[i] =        2*i ;
    }
  }
 
  for(int i=0; i<8; i++){
    if( l1mv[2*i + 1] > l1mv[2*i] ){
      l2mv[i]   = l1mv[  2*i + 1 ];
      l2met[i]  = l1met[ 2*i + 1];
      l2mphi[i] = l1mphi[2*i + 1];
    } else {
      l2mv[i]   = l1mv[  2*i ];
      l2met[i]  = l1met[ 2*i ];
      l2mphi[i] = l1mphi[2*i ];
    }
  }
 
  for(int i=0; i<4; i++){
    if( l2mv[2*i + 1] > l2mv[2*i] ){
      l3mv[i]   = l2mv[  2*i + 1];
      l3met[i]  = l2met[ 2*i + 1];
      l3mphi[i] = l2mphi[2*i + 1];
    } else {
      l3mv[i]   = l2mv[2*i   ];
      l3met[i]  = l2met[2*i  ];
      l3mphi[i] = l2mphi[2*i ];
    }
  }
 
  for(int i=0; i<2; i++){
    if( l3mv[2*i + 1] > l3mv[2*i] ){
      l4mv[i]   = l3mv[  2*i + 1];
      l4met[i]  = l3met[ 2*i + 1];
      l4mphi[i] = l3mphi[2*i + 1];
    } else {
      l4mv[i]   = l3mv[2*i   ];
      l4met[i]  = l3met[2*i  ];
      l4mphi[i] = l3mphi[2*i ];
    }
  }
 
  if(  l4mv[1] > l4mv[0]  ){
    if(l4mv[1] > jetThreshold){
      gJetTOBv[TOBnum]   = l4mv[1];
      gJetTOBeta[TOBnum] = l4met[1] + etaOff[TOBnum];
      gJetTOBphi[TOBnum] = l4mphi[1];
      gJetTOBs[TOBnum]   = 1;
    } else {
      gJetTOBv[TOBnum]   = l4mv[1];
      gJetTOBeta[TOBnum] = l4met[1] + etaOff[TOBnum];
      gJetTOBphi[TOBnum] = l4mphi[1];
      gJetTOBs[TOBnum]   = 0;
    }
  } else {
    if(l4mv[0] > jetThreshold){
      gJetTOBv[TOBnum]   = l4mv[0];
      gJetTOBeta[TOBnum] = l4met[0] + etaOff[TOBnum];
      gJetTOBphi[TOBnum] = l4mphi[0];
      gJetTOBs[TOBnum]   = 1;
    } else {
      gJetTOBv[TOBnum]   = l4mv[0];
      gJetTOBeta[TOBnum] = l4met[0] + etaOff[TOBnum];
      gJetTOBphi[TOBnum] = l4mphi[0];
      gJetTOBs[TOBnum]   = 0;
    }
 }
 
}
 
 
} // namespace LVL1