/* This example handles a 12 x 12 mesh, on 4 processors only. */
#define maxn 12

int main( argc, argv )
int argc;
char **argv;
{
    int        rank, value, size, errcnt, toterr, i, j, itcnt;
    int        i_first, i_last;
    MPI_Status status;
    double     diffnorm, gdiffnorm;
    double     xlocal[(12/4)+2][12];
    double     xnew[(12/3)+2][12];

    static char progName[] = { "hpmtest" };
    char label[10];
    int taskID, instID;

    MPI_Init( &argc, &argv );

    MPI_Comm_rank( MPI_COMM_WORLD, &rank );
    MPI_Comm_size( MPI_COMM_WORLD, &size );

    taskID = 1;
    hpmInit( taskID, progName );
     hpmStart(1, "MPI section");
 

    if (size != 4) MPI_Abort( MPI_COMM_WORLD, 1 );

    /* xlocal[][0] is lower ghostpoints, xlocal[][maxn+2] is upper */

    /* Note that top and bottom processes have one less row of interior
       points */
    i_first = 1;
    i_last  = maxn/size;
    if (rank == 0)        i_first++;
    if (rank == size - 1) i_last--;
    hpmStart(2, "Fill loop");

    /* Fill the data as specified */
    for (i=1; i<=maxn/size; i++)
        for (j=0; j<maxn; j++)
            xlocal[i][j] = rank;
    for (j=0; j<maxn; j++) {
        xlocal[i_first-1][j] = -1;
        xlocal[i_last+1][j] = -1;
    }
     hpmStop( 2 );

    itcnt = 0;
    do {
        /* Send up unless I'm at the top, then receive from below */
        /* Note the use of xlocal[i] for &xlocal[i][0] */
        if (rank < size - 1)
            MPI_Send( xlocal[maxn/size], maxn, MPI_DOUBLE, rank + 1, 0,
                      MPI_COMM_WORLD );
        if (rank > 0)
            MPI_Recv( xlocal[0], maxn, MPI_DOUBLE, rank - 1, 0,
                      MPI_COMM_WORLD, &status );
        /* Send down unless I'm at the bottom */
        if (rank > 0)
            MPI_Send( xlocal[1], maxn, MPI_DOUBLE, rank - 1, 1,
                      MPI_COMM_WORLD );
        if (rank < size - 1)
            MPI_Recv( xlocal[maxn/size+1], maxn, MPI_DOUBLE, rank + 1, 1,
                      MPI_COMM_WORLD, &status );
 
        /* Compute new values (but not on boundary) */
        itcnt ++;
        diffnorm = 0.0;
        for (i=i_first; i<=i_last; i++)
            for (j=1; j<maxn-1; j++) {
                xnew[i][j] = (xlocal[i][j+1] + xlocal[i][j-1] +
                              xlocal[i+1][j] + xlocal[i-1][j]) / 4.0;
                diffnorm += (xnew[i][j] - xlocal[i][j]) *
                            (xnew[i][j] - xlocal[i][j]);
            }
        /* Only transfer the interior points */
        for (i=i_first; i<=i_last; i++)
            for (j=1; j<maxn-1; j++)
                xlocal[i][j] = xnew[i][j];

        MPI_Allreduce( &diffnorm, &gdiffnorm, 1, MPI_DOUBLE, MPI_SUM,
                       MPI_COMM_WORLD );
        gdiffnorm = sqrt( gdiffnorm );
        if (rank == 0) printf( "At iteration %d, diff is %e\n", itcnt,
                               gdiffnorm );
    } while (gdiffnorm > 1.0e-2 && itcnt < 100);

    hpmStop( 1);
    hpmTerminate( taskID );

    MPI_Finalize( );
    return 0;
}