Splitting of MPI communicators

慧伟

于 2012-08-15 10:44:02 发布

阅读量1.4k

点赞数

分类专栏：并行计算

并行计算专栏收录该内容

6 篇文章 0 订阅

订阅专栏

有时候为了创建2D的拓扑，需要对MPI_COMM_WORLD进行划分。所使用的MPI函数是：

MPI_Comm_split(MPI_Comm comm, int color, int key, MPI_Comm *newcomm)

例子程序（出自MPI Groups, Communicators and Topologies）：

#include <stdio.h> 
#include <mpi.h>

main (int argc, char **argv)
{
    MPI_Comm row_comm, col_comm;
    int myrank, size, P=4, Q=3, p, q;

    MPI_Init(&argc, &argv);
    MPI_Comm_rank(MPI_COMM_WORLD, &myrank);
    MPI_Comm_size(MPI_COMM_WORLD, &size);

    p = myrank / Q;
    q = myrank % Q;

    /* Split comm intor row and column comms */
    MPI_Comm_split(MPI_COMM_WORLD, p, q, &row_comm);
    MPI_Comm_split(MPI_COMM_WORLD, q, p, &col_comm);

    int sum, rowsum, colsum;
    sum = myrank;
    rowsum = myrank;
    colsum = myrank;
    MPI_Allreduce(MPI_IN_PLACE, &sum, 1, MPI_INT, MPI_SUM, MPI_COMM_WORLD);
    MPI_Allreduce(MPI_IN_PLACE, &rowsum, 1, MPI_INT, MPI_SUM, row_comm);
    MPI_Allreduce(MPI_IN_PLACE, &colsum, 1, MPI_INT, MPI_SUM, col_comm);
    if (myrank == 0)
        printf("The sum of all rank is %d\n", sum);

    printf("[%d]: My coordinates are (%d, %d), row sum is %d, col sum is %d\n",myrank,p,q,
rowsum,colsum);
    MPI_Finalize();
}

参考资料：

1. http://www.cs.utexas.edu/~pingali/CS378/2011sp/lectures/MPIGroups.ppt