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

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

Copyright (c) J.-S. Caux.

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

File:  Heis_DSF_par_Run.cc

Purpose:  Parallel version of ABACUS using MPICH.

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

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

using namespace std;
using namespace ABACUS;

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

  if (argc < 10) { // provide some info

    cout << endl << "Welcome to ABACUS\t(copyright J.-S. Caux)." << endl;
    cout << endl << "Usage of Heis_DSF_par_Run 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 paralevel" << endl;
    cout << "rank[i], nr_processors[i] \t rank and nr_processors of each earlier paralevels." << endl;
    cout << "int Max_Secs \t\t Allowed computational time:  (in seconds)" << endl;
    cout << "int supercycle_time \t\t time for one supercycle (in seconds)" << endl;
    cout << endl << "EXAMPLE: " << endl << endl;
    cout << "mpiexec -np 8 Heis_DSF_par_Run z 1 128 64 0 128 [**UPDATE]" << endl << endl;

    return(0);
  }

  //else { // (argc == 9) correct nr of arguments
  int n = 1;
  whichDSF = *argv[n++];
  Delta = atof(argv[n++]);
  N = atoi(argv[n++]);
  M = atoi(argv[n++]);
  iKmin = atoi(argv[n++]);
  iKmax = atoi(argv[n++]);
  paralevel = atoi(argv[n++]); // paralevel == 1 means that we have one layer of parallelization, so no previous rank and nr_processors to specify
  if (argc != 10 + 2*(paralevel - 1)) ABACUSerror("Wrong nr of arguments in Heis_DSF_par_Run.");
  Vect<int> rank_lower_paralevels(paralevel - 1);
  Vect<int> nr_processors_lower_paralevels(paralevel - 1);
  for (int i = 0; i < paralevel - 1; ++i) {
    rank_lower_paralevels[i] = atoi(argv[n++]);
    nr_processors_lower_paralevels[i] = atoi(argv[n++]);
  }
  Max_Secs = atoi(argv[n++]);
  supercycle_time = atoi(argv[n++]);

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

  MPI::Init(argc, argv);

  DP tstart = MPI::Wtime();

  int rank_here = MPI::COMM_WORLD.Get_rank();
  int nr_processors_here = MPI::COMM_WORLD.Get_size();

  Vect<int> rank (paralevel);
  Vect<int> nr_processors (paralevel);
  for (int i = 0; i < paralevel - 1; ++i) {
    rank[i] = rank_lower_paralevels[i];
    nr_processors[i] = nr_processors_lower_paralevels[i];
  }
  rank[paralevel-1] = rank_here;
  nr_processors[paralevel-1] = nr_processors_here;

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

  refine = true;

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


  DP tnow = MPI::Wtime();

  string defaultScanStatename = "";

  while (tnow - tstart < Max_Secs - supercycle_time - 120) { // space for one more supercycle, + 2 minutes safety

    // 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, paralevel, rank, nr_processors);

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

    // Now that everybody is done, digest data into unique files

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

    tnow = MPI::Wtime();

  } // while (tnow - tstart...

  MPI::Finalize();

  return(0);
}