Cblacs_pcoord() routine

[alex]

Hi,

Julien Langou wrote a code at the following url: http://icl.cs.utk.edu/lapack-forum/viewtopic.php?t=136

I modified that code in order to test the Cblacs_pcoord() routine.
And it became:

###################################################################
#include <stdio.h>
#include <string.h>
#include <stdlib.h>
#include "mpi.h"
#include <sys/time.h>

static int max( int a, int b ){
if (a>b) return(a); else return(b);
}

extern void Cblacs_pinfo( int* mypnum, int* nprocs);
extern void Cblacs_get( int context, int request, int* value);
extern void Cblacs_gridmap(int *ictxt,int *pmap,int ldpmap,int nprow,int npcol);
extern void Cblacs_pcoord(int ictxt,int pnum,int *prow,int *pcol);
extern void Cblacs_gridexit( int context);
extern void Cblacs_exit( int error_code);

int main(int argc, char **argv) {
int iam, nprocs;
int myrank_mpi, nprocs_mpi;
int ictxt1, nprow1, npcol1, myrow1, mycol1;
int ictxt2, nprow2, npcol2, myrow2, mycol2;
int np1, nq1, n1, nb1, nrhs1;
int np2, nq2, n2, nb2, nrhs2;
int i, j, k, info, itemp, seed;
int descA[9], descB[9];
double *A, *Acpy, *B, *X, *R, eps, *work;
double AnormF, XnormF, RnormF, BnormF, residF;
int *ippiv;
int izero=0,ione=1;
double mone=(-1.0e0),pone=(1.0e0);
int *usermap1, *usermap2, ldumap1, ldumap2;
/**/
double MPIt1, MPIt2, MPIelapsed;
/**/
MPI_Init( &argc, &argv);
MPI_Comm_rank(MPI_COMM_WORLD, &myrank_mpi);
MPI_Comm_size(MPI_COMM_WORLD, &nprocs_mpi);
/**/
Cblacs_pinfo( &iam, &nprocs ) ;
Cblacs_get( -1, 0, &ictxt1 );
Cblacs_get( -1, 0, &ictxt2 );
/**/
n1 = 500; nrhs1 = 1; nprow1 = 1; npcol1 = 2; nb1 = 64; ldumap1 = nprow1;
n2 = 250; nrhs2 = 2; nprow2 = 2; npcol2 = 1; nb2 = 50; ldumap2 = nprow2;
/**/
usermap1 = (int *)calloc(ldumap1*npcol1,sizeof(int)) ;
usermap2 = (int *)calloc(ldumap2*npcol2,sizeof(int)) ;
/**/
usermap1[0] = 0; usermap1[1] = 2;
usermap2[0] = 1; usermap2[1] = 3;
/**/
if ((nprow1*npcol1)+(nprow2*npcol2)>nprocs_mpi){
if (myrank_mpi==0)
printf(" **** ERROR : I want to use %d processes and only have %d\n",(nprow1*npcol1)+(nprow2*npcol2),nprocs_mpi);
MPI_Finalize(); exit(1);
}
/**/
Cblacs_gridmap ( &ictxt1, usermap1, ldumap1, nprow1, npcol1 );
free(usermap1);
Cblacs_gridmap ( &ictxt2, usermap2, ldumap2, nprow2, npcol2 );
free(usermap2);
/**/

if(ictxt1>=0){
Cblacs_pcoord(ictxt1,iam,&myrow1,&mycol1);
printf("I am %d, my coordinates are (%d,%d) on grid_1\n\n",iam,myrow1,mycol1);
}

if(ictxt2>=0){
Cblacs_pcoord(ictxt2,iam,&myrow2,&mycol2);
printf("I am %d, my coordinates are (%d,%d) on grid_2\n\n",iam,myrow2,mycol2);
}


Cblacs_exit(0);
exit(0);
}
####################################################################

the compilation is successful using:
mpif77 -o pcoord pcoord.c blacsF77init_MPI-LINUX-0.a blacs_MPI-LINUX-0.a blacsF77init_MPI-LINUX-0.a


When I run the executable, I get the following:

##################################################################
[root@localhost /root]#mpirun -np 4 ./pcoord
I am 0, my coordinates are (0,0) on grid_1

I am 2, my coordinates are (-1,-1) on grid_1

I am 1, my coordinates are (1,0) on grid_2

I am 3, my coordinates are (-1,-1) on grid_2

[root@localhost /root]#
###################################################################

only the coordinates of process 0 seems to be correct.
for example, the coordinates of process 1 should be (0,0) on grid_2, no?

Why the Cblacs_pcoord() routine doesn't return the correct coordinates?

Have you any idea ?

Thanks in advance,
alex

[alex]

I am glad to tell you that I have finally understood the meaning of the result
returned by the Cblacs_pcoord routine.
In fact, the process rank given in entry to this routine represents the target process's rank in the local grid and not its rank in the global system.

alex