/* builds n communicators each with m processors from a collection of
 * m*n processors using MPI_Comm_split
 *
 * here m is implicitly the number of rows in a virtual
 * process grid and n is the number of columns in a virtual
 * process grid.
 *
 * Input: m (the global total number of rows)
 *        Note: with m given n is then derived via n = p/m
 *
 * Output: n -- program tests correct creation of new communicator
 *     by broadcasting the value 1 to its members -- all other 
 *     processes have the value 0 -- global sum computed across
 *     ALL the processes.
 *
 * See Chap. 7, pp. 120 & ff in PPMPI
 * by John L. Weatherwax
 */
#include <stdio.h>
#include <math.h>
#include "mpi.h"

main(int argc, char* argv[]) {
    int       p,m,n;
    int       my_rank;
    MPI_Comm  my_col_comm;
    int       my_row,my_col;
    int       q;
    int       test,sum;
    int       my_rank_in_col;

    MPI_Init(&argc, &argv);
    MPI_Comm_size(MPI_COMM_WORLD, &p);
    MPI_Comm_rank(MPI_COMM_WORLD, &my_rank);

    /* Input arguments */ 
    if( my_rank==0 ){
      printf("Enter m: "); scanf("%d",&m);
      printf("Virtual grid is of size [%d,%d]\n",m,(int) (p/m)); 
    }
    MPI_Bcast(&m,1,MPI_INT,0,MPI_COMM_WORLD); 

    n = (int) (p/m); 
    
    /* my_rank is rank in MPI_COMM_WORLD. */ 
    my_row = my_rank/n;
    my_col = my_rank - (my_row)*n;
    MPI_Comm_split(MPI_COMM_WORLD, my_col, my_rank, &my_col_comm);

    /* Test the new communicators */
    MPI_Comm_rank(my_col_comm, &my_rank_in_col);
    if (my_rank_in_col == 0)
      test = 1; /*my_col;*/
    else
      test = 0;

    MPI_Bcast(&test, 1, MPI_INT, 0, my_col_comm);

    MPI_Reduce(&test, &sum, 1, MPI_INT, MPI_SUM, 0, MPI_COMM_WORLD);

    if (my_rank == 0) {
      printf("sum = %d\n", sum);
    }

    MPI_Finalize();
}  /* main */