MPI_Comm_split 不適用於 MPI_Bcast (MPI_Comm_split not working with MPI_Bcast)

問題描述

MPI_Comm_split 不適用於 MPI_Bcast (MPI_Comm_split not working with MPI_Bcast)

使用以下代碼，我將 4 個進程拆分為列組，然後從對角線 (0,3) 在同一列中廣播。進程 0 向 2 廣播。而 3 應該向 1 廣播。但它沒有按預期工作。有人能看出什麼問題嗎？

0 1

2 3

#include <stdio.h>
#include <stdlib.h>
#include <iostream>
#include <mpi.h>
#include <mpi.h>
using namespace std;

int main(int argc, char **argv){
    MPI_Comm col_comm,row_comm;
    int myrank, size, even, value=0;
    int localRank=0;
    MPI_Init (&argc, &argv);
    MPI_Comm_rank (MPI_COMM_WORLD, &myrank);
    MPI_Comm_size (MPI_COMM_WORLD, &size);      
    MPI_Comm_split(MPI_COMM_WORLD, myrank%2, myrank, &col_comm); 
    MPI_Comm_rank (col_comm, &localRank);

    if(myrank%3==0){
        value = myrank*5+1;
        MPI_Bcast(&value, 1, MPI_INT, localRank, col_comm);
    }

    printf("Rank=%d | LocalRank=%d | Got broadcast value of %d\n", myrank, localRank, value);
    MPI_Finalize();
    return 0;
}

</code></pre> 
輸出
 
ubuntu@root:~/matrixmult$ mpirun comtest ‑np 4
Rank=0 | LocalRank=0 | Got broadcast value of 1
Rank=1 | LocalRank=0 | Got broadcast value of 0
Rank=2 | LocalRank=1 | Got broadcast value of 0
Rank=3 | LocalRank=1 | Got broadcast value of 16






參考解法
方法 1:
 
MPI_Bcast


Broadcasts a message from the process with rank "root" to all other
processes of the communicator


</blockquote>

is a collective communication routine, hence it should be called by all the processes in a given communicator. Therefore, you need to remove the following condition if(myrank%3==0) and then you need to adapt the root process accordingly, instead of using localRank.

In your current code, only the processes with myrank 0 and 3 called the MPI_Bcast (both belonging to different communicators). So process 0 calls

MPI_Bcast(&value, 1, MPI_INT, localRank, col_comm);

which basically means that it broadcasted the value to itself. The same happens with process 3. Hence, the output:
Rank=0 | LocalRank=0 | Got broadcast value of 1
Rank=1 | LocalRank=0 | Got broadcast value of 0
Rank=2 | LocalRank=1 | Got broadcast value of 0
Rank=3 | LocalRank=1 | Got broadcast value of 16

Rank=0 and Rank=3 communicated with themselves, while the other processes where not part of the MPI_Bcast. Hence the value of 0 for both of them.
Try the following:
int main(int argc, char **argv){
    MPI_Comm col_comm,row_comm;
    int myrank, size, even, value=0;
    int localRank=0;
    MPI_Init (&argc, &argv);
    MPI_Comm_rank (MPI_COMM_WORLD, &myrank);
    MPI_Comm_size (MPI_COMM_WORLD, &size);      
    MPI_Comm_split(MPI_COMM_WORLD, myrank%2, myrank, &col_comm); 
    MPI_Comm_rank (col_comm, &localRank);

    if(myrank == 0 || myrank == 3)    
       value = myrank*5+1;

    MPI_Bcast(&value, 1, MPI_INT, myrank%2 != 0, col_comm);

    printf("Rank=%d | LocalRank=%d | Got broadcast value of %d\n", myrank, localRank, value);
    MPI_Finalize();
    return 0;
}


Process 0 broadcasts to 2. And 3 should broadcast to 1.

Output:
Rank=0 | LocalRank=0 | Got broadcast value of 1
Rank=1 | LocalRank=0 | Got broadcast value of 16
Rank=2 | LocalRank=1 | Got broadcast value of 1
Rank=3 | LocalRank=1 | Got broadcast value of 16
(by Pit Digger、dreamcrash)
參考文件


MPI_Comm_split not working with MPI_Bcast (CC BY‑SA 2.5/3.0/4.0)