c - 使用 MPI_Reduce 时检测到堆栈粉碎

Question

我学会了使用一些 MPI 函数。当我尝试使用 MPI_Reduce 时，我在运行代码时收到检测到堆栈粉碎:

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

void main(int argc, char **argv) {
    int i, rank, size;
    int sendBuf, recvBuf, count;
    MPI_Init(&argc, &argv);
    MPI_Comm_size(MPI_COMM_WORLD, &size);
    MPI_Comm_rank(MPI_COMM_WORLD, &rank);
    sendBuf = rank;
    count = size;
    MPI_Reduce(&sendBuf, &recvBuf, count, MPI_INT, 
        MPI_SUM, 0, MPI_COMM_WORLD);

    if (rank == 0) {
        printf("Sum is %d\n", recvBuf);
    }

    MPI_Finalize();
}

我的代码似乎没问题。它将打印 sum of all rank in recvBuf with process 0。在这种情况下，如果我运行我的代码，它将打印 Sum is 45 10 进程 mpirun -np 10 myexecutefile。但是我不知道为什么我的代码有错误:

Sum is 45
*** stack smashing detected ***: example6 terminated
[ubuntu:06538] *** Process received signal ***
[ubuntu:06538] Signal: Aborted (6)
[ubuntu:06538] Signal code:  (-6)
[ubuntu:06538] *** Process received signal ***
[ubuntu:06538] Signal: Segmentation fault (11)
[ubuntu:06538] Signal code:  (128)
[ubuntu:06538] Failing at address: (nil)
--------------------------------------------------------------------------
mpirun noticed that process rank 0 with PID 0 on node ubuntu exited on signal 11 (Segmentation fault).
--------------------------------------------------------------------------

问题是什么，我该如何解决？

Answer 1

在

MPI_Reduce(&sendBuf, &recvBuf, count, MPI_INT, MPI_SUM, 0, MPI_COMM_WORLD);

参数 count 必须是 number of elements in send buffer .由于 sendBuf 是单个整数，请尝试使用 count = 1; 而不是 count = size;。

Sum is 45 被正确打印的原因很难解释。越界访问值是未定义的行为:问题可能一直未被注意到，或者在打印 Sum is 45 之前可能已经引发段错误。未定义行为的魔力...