I try to send and receive a large array to self-rank using MPI. The code works well for small arrays (array size <10000) when I further increase the array size to 100000, it will deadlock. Here is my code:

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

int main(int argc, char *argv[])
{

MPI_Init(&argc, &argv);

int size;
MPI_Comm_size(MPI_COMM_WORLD, &size);

int rank;
MPI_Comm_rank(MPI_COMM_WORLD, &rank);

MPI_Status stat;


    int N = 100000;

    // Allocate memory for A on CPU
    double *A = (double*)malloc(N*sizeof(double));
    double *B = (double*)malloc(N*sizeof(double));

    // Initialize all elements of A to 0.0
    for(int i=0; i<N; i  ){
        A[i] = 0.0;
    }

    int tag1 = 10;
    int tag2 = 20;

//printf("rank=%d",rank);


    MPI_Send(A, N, MPI_DOUBLE, 0, tag1, MPI_COMM_WORLD);
    
    MPI_Recv(B, N, MPI_DOUBLE, 0, tag1, MPI_COMM_WORLD, &stat);

MPI_Finalize();

return 0;
}

I compile the code by: mpicc
command for the run: mpirun -np 1 ./

CodePudding user response：

That's the definition of send and receive: they are blocking. The statement after the send will not execute until the send has been successfully completed.

1. The safe way to solve this is using MPI_Isend and MPI_Irecv.
2. The case for a small messages which don't block is an optimization that is system dependent and you can not count on it.