RunActionMaster Class Reference

#include <RunActionMaster.hh>

Inheritance diagram for RunActionMaster:

Collaboration diagram for RunActionMaster:

Public Member Functions
	RunActionMaster ()
	~RunActionMaster ()
virtual G4Run *	GenerateRun ()
virtual void	BeginOfRunAction (const G4Run *arun)
virtual void	EndOfRunAction (const G4Run *arun)
Public Member Functions inherited from G4UserRunAction
	G4UserRunAction ()
virtual	~G4UserRunAction ()
virtual void	SetMaster (G4bool val=true)
G4bool	IsMaster () const

Additional Inherited Members
Protected Attributes inherited from G4UserRunAction
G4bool	isMaster

Detailed Description

Definition at line 36 of file RunActionMaster.hh.

Constructor & Destructor Documentation

RunActionMaster::RunActionMaster

(

)

Definition at line 43 of file RunActionMaster.cc.

{
}

RunActionMaster::~RunActionMaster

(

)

Definition at line 54 of file RunActionMaster.cc.

{
}

Member Function Documentation

void RunActionMaster::BeginOfRunAction ( const G4Run *  arun )

virtual

Reimplemented from G4UserRunAction.

Definition at line 60 of file RunActionMaster.cc.

{
  Analysis* myana = Analysis::GetAnalysis();
  myana-> Clear();
  myana->Book();
}

Here is the call graph for this function:

void RunActionMaster::EndOfRunAction ( const G4Run *  arun )

virtual

Reimplemented from G4UserRunAction.

Definition at line 69 of file RunActionMaster.cc.

{
  //This is executed by master thread only. Worker threads have already
  //merged their results into this master threads.

  //We are going to merge the results via MPI for:
  // 1. User-defined "Run" object
  // 2. Command line scorers, if exists
  // 3. G4Analysis objects

  // For debugging purposes in the following we write out results twice:
  // BEFORE and AFTER the merging, so that rank 0 actually
  // writes two files, the one called "*rank000*" will contain the partial
  // results only from rank #0,
  // the file *merged* contains the reduction from all ranks.
  // It should be very easy to adapt this code

  const G4int rank = G4MPImanager::GetManager()-> GetRank();

  G4cout << "=====================================================" << G4endl;
  G4cout << "Start EndOfRunAction for master thread in rank: " << rank<<G4endl;
  G4cout << "=====================================================" << G4endl;

  //Merging of G4Run object:
  //All ranks > 0 merge to rank #0
  RunMerger rm(static_cast<const Run*>(arun));
  G4int ver = 0 ; //Use 4 for lots of info
  if ( rank == 0 && ver == 0) ver = 1;

  if ( ver > 1 ) {
  G4cout<<"Before merge the Run object has test counter value: "
      <<static_cast<const Run*>(arun)->GetCounter()<<G4endl;
  }
  rm.SetVerbosity( ver );
  rm.Merge();
  if ( ver > 1 ) {
  G4cout<<"After merge the Run object has test counter value: "
      <<static_cast<const Run*>(arun)->GetCounter()
      <<" (with 1 thread== number of ranks)"<<G4endl;
  }

  //Merge of scorers
  ver = 0;
  if (G4ScoringManager::GetScoringManagerIfExist()) {
      const auto scor = G4ScoringManager::GetScoringManager();
      G4MPIscorerMerger sm(scor);
      sm.SetVerbosity(ver);
      //Debug!
      auto debugme = [&scor](){
        for ( size_t idx = 0 ; idx < scor->GetNumberOfMesh() ; ++idx) {
            const auto m = scor->GetMesh(idx);
            const MeshScoreMap& map = m->GetScoreMap();
            std::for_each(map.begin(),map.end(),
              [](const MeshScoreMap::value_type& e) {
                G4cout<<e.first<<"("<<e.second<<"):"<<G4endl;
                const auto data = e.second->GetMap();
                for( auto it = data->begin() ; it != data->end() ; ++it ) {
                    G4cout<<it->first<<" => G4StatDouble(n,sum_w,sum_w2,sum_wx,sum_wx2): "
                    <<it->second->n()<<" "<<it->second->sum_w()<<" "
                    <<it->second->sum_w2()<<" "<<it->second->sum_wx()<<" "
                    <<it->second->sum_wx2()<<G4endl;
                }
            });
        }
      };
      //Debug!
      if ( ver > 4 ) {
          G4cout<<"Before merging: Meshes dump"<<G4endl;
          debugme();
      }
      //Write partial scorers from single ranks *before* merging
      //Do not rely on UI command to write out scorers, because rank-specific
      //files will have same file name: need to add rank # to file name
      if ( true ) {
       for ( size_t idx = 0 ; idx < scor->GetNumberOfMesh() ; ++idx) {
              const auto m = scor->GetMesh(idx);
              const auto& mn = m->GetWorldName();
              std::ostringstream fname;
              fname<<"scorer-"<<mn<<"-rank"<<rank<<".csv";
              scor->DumpAllQuantitiesToFile(mn,fname.str());
       }
      }

      //Now reduce all scorers to rank #0
      sm.Merge();

      //Debug!
      if ( ver > 4 ) {
          G4cout<<"After merging: Meshes dump"<<G4endl;
          debugme();
      }
      //For rank #0 write out the merged files
      if ( rank == 0 ) {
        for ( size_t idx = 0 ; idx < scor->GetNumberOfMesh() ; ++idx) {
            const auto m = scor->GetMesh(idx);
            const auto& mn = m->GetWorldName();
            std::ostringstream fname;
            fname<<"scorer-"<<mn<<"-merged.csv";
            scor->DumpAllQuantitiesToFile(mn,fname.str());
        }
    }
  }

  //Save histograms *before* MPI merging for rank #0
  if (rank == 0)
    {
      G4String fname("dose-rank0");
      Analysis* myana = Analysis::GetAnalysis();
      myana-> Save(fname);
    }
  //Merge of g4analysis objects
  ver=0;
  G4MPIhistoMerger hm(G4AnalysisManager::Instance());
  hm.SetVerbosity(ver);
  hm.Merge();

  //Save g4analysis objects to a file
  //NB: It is important that the save is done *after* MPI-merging of histograms

  //One can save all ranks or just rank0, chane the if
  if (true /*rank == 0*/)
    {
      std::ostringstream fname;
      fname<<"dose-rank"<<rank;
      if (rank == 0) {
        fname.str("dose-merged");
      }
      Analysis* myana = Analysis::GetAnalysis();
      myana-> Save(fname.str());
    }
    Analysis* myana = Analysis::GetAnalysis();
    myana-> Close();

    G4cout << "===================================================" << G4endl;
    G4cout << "End EndOfRunAction for master thread in rank: " << rank << G4endl;
    G4cout << "===================================================" << G4endl;

}

Here is the call graph for this function:

G4Run * RunActionMaster::GenerateRun ( void  )

virtual

Reimplemented from G4UserRunAction.

Definition at line 48 of file RunActionMaster.cc.

{
  return new Run;
}

---

The documentation for this class was generated from the following files:

• source/geant4.10.03.p02/examples/extended/parallel/MPI/examples/exMPI03/include/RunActionMaster.hh
• source/geant4.10.03.p02/examples/extended/parallel/MPI/examples/exMPI03/src/RunActionMaster.cc