/**********************************************************

This software is part of J.-S. Caux's ABACUS library.

Copyright (c) J.-S. Caux.

-----------------------------------------------------------

File:  Heis_DSF_par.cc

Purpose:  Parallel version of ABACUS using MPICH.

***********************************************************/

#include "ABACUS.h"
#include "mpi.h"

using namespace ABACUS;

int main(int argc, char *argv[])
{
  char whichDSF;
  DP Delta;
  int N, M, iKneeded, iKmin, iKmax, Max_Secs;
  DP target_sumrule = 1.0e+6;  // effectively deactivated here
  bool refine;

  if (argc != 8) { // provide some info

    cout << endl << "Welcome to ABACUS\t(copyright J.-S. Caux)." << endl;
    cout << endl << "Usage of Heis_DSF_par executable: " << endl;
    cout << endl << "This function runs ABACUS in parallel mode, starting from a preexisting "
      "serial run (obtained using the Heis_DSF executable) using the same model parameters." << endl;
    cout << endl << "Provide the following arguments:" << endl << endl;
    cout << "char whichDSF \t\t Which structure factor should be calculated ?  Options are:  "
      "m for S- S+, z for Sz Sz, p for S+ S-." << endl;
    cout << "DP Delta \t\t Value of the anisotropy:  use positive real values only" << endl;
    cout << "int N \t\t\t Length (number of sites) of the system:  use positive even integer values only" << endl;
    cout << "int M \t\t\t Number of down spins:  use positive integer values between 1 and N/2" << endl;
    cout << "int iKmin" << endl << "int iKmax \t\t Min and max momentum integers to scan over:  "
      "recommended values:  0 and N" << endl;
    cout << "int Max_Secs \t\t Allowed computational time:  (in seconds)" << endl;
    cout << endl << "EXAMPLE: " << endl << endl;
    cout << "mpiexec -np 8 Heis_DSF_par z 1.0 100 40 0 100 600" << endl << endl;

    return(0);
  }

  else { // (argc == 8) correct nr of arguments
    whichDSF = *argv[1];
    Delta = atof(argv[2]);
    N = atoi(argv[3]);
    M = atoi(argv[4]);
    iKmin = atoi(argv[5]);
    iKmax = atoi(argv[6]);
    Max_Secs = atoi(argv[7]);
  }

  DP supercycle_time = 600.0; // allotted time per supercycle

  if (Max_Secs <= supercycle_time + 300) ABACUSerror("Please allow more time in Heis_DSF_par.");

  MPI::Init(argc, argv);

  DP tstart = MPI::Wtime();

  int rank = MPI::COMM_WORLD.Get_rank();
  int nr_processors = MPI::COMM_WORLD.Get_size();

  if (nr_processors < 2) ABACUSerror("Give at least 2 processors to ABACUS parallel !");

  refine = true;

  // ASSUMPTION:  preexisting files (raw, thr, ...) exist for the run.

  // IMPORTANT PRECONDITION:  no flags are being raised in General_Scan in parallel mode, so
  // the preinitializing serial run must be extensive enough to have flagged all base/type s necessary.

  DP tnow = MPI::Wtime();

  while (tnow - tstart < Max_Secs - supercycle_time - 300) { // space for one more supercycle, + 5 minutes safety

    if (rank == 0)
      // Split up thread list into chunks, one per processor
      Prepare_Parallel_Scan_Heis (whichDSF, Delta, N, M, iKmin, iKmax, nr_processors);

    // Barrier synchronization, to make sure other processes wait for process of rank 0
    // to have finished splitting up the thr file into pieces before starting:
    MPI_Barrier (MPI::COMM_WORLD);

    // then everybody gets going on their own chunk !
    Scan_Heis (whichDSF, Delta, N, M, iKmin, iKmax,
	       supercycle_time, target_sumrule, refine, rank, nr_processors);

    // Another barrier synchronization
    MPI_Barrier (MPI::COMM_WORLD);

    // Now that everybody is done, digest data into unique files
    if (rank == 0)
      Wrapup_Parallel_Scan_Heis (whichDSF, Delta, N, M, iKmin, iKmax, nr_processors);

    // Another barrier synchronization
    MPI_Barrier (MPI::COMM_WORLD);

    tnow = MPI::Wtime();

  } // while (tnow - tstart...


  MPI::Finalize();

  return(0);
}