I have a problem with sending a single column of a 2D array with MPI in C++. I found MPI_Type_create_subarray and it seems to work fine, except for the case where I want to receive the values in the last column. If I use column index 3 in a 4x4 array, for example, I always get a SIGSEGV.

Can anybody please help me?

I have created a small code example (and left the error checking etc. out here):

#include <iostream>
#include <mpi.h>

int main(int argc, char**argv){
    double array[4][4]{
            {1,1,1,1},
            {1,1,1,1},
            {1,1,1,1},
            {1,1,1,1}
    };

    MPI_Init(&argc, &argv);
    int numprocs, rank;
    MPI_Comm_size(MPI_COMM_WORLD, &numprocs);
    MPI_Comm_rank(MPI_COMM_WORLD, &rank);

    MPI_Request request[2];
    MPI_Status status[2];
    int request_count = 0;

    if( rank == 0 ){
        // Write column 3 ...
        for( int i = 0; i < 4; ++i)
            array[i][3] = i;

        // Send subarray, i.e. the 3rd column
        MPI_Datatype column_send;
        {
            int array_sizes[]{4,4};
            int array_subsizes[]{4,1};
            int starting_points[]{0,3};
            MPI_Type_create_subarray(2, array_sizes, array_subsizes, starting_points, MPI_ORDER_C, MPI_DOUBLE, &column_send);
            MPI_Type_commit(&column_send);
        }
        ierr = MPI_Isend(
                array[0],
                4,
                column_send,
                1,
                0,
                MPI_COMM_WORLD,
                request+request_count++
        );
    }
    else if(rank == 1){
        // Just overwrite the values here for testing ...
        for( int i = 0; i < 4; ++i){
            for( int j = 0; j < 4; ++j){
                array[i][j] = -1;
            }
        }

        // Receive the 3rd column
        MPI_Datatype column_send;
        {
            int array_sizes[]{4,4};
            int array_subsizes[]{4,1};
            // NOTE: IF I USE {0,0}, {0,1} or {0,2} HERE, IT WORKS!
            // But with {0,3}, I get a SIGSEGV.
            int starting_points[]{0,3};
            MPI_Type_create_subarray(2, array_sizes, array_subsizes, starting_points, MPI_ORDER_C, MPI_DOUBLE, &column_send);
            MPI_Type_commit(&column_send);
        }
        MPI_Irecv(
                array[0],
                4,
                column_send,
                0,
                0,
                MPI_COMM_WORLD,
                request+request_count++
        );
    }
    MPI_Waitall(request_count, request, status);

    // Print array ...
    if( rank == 1 ){
        for( int i = 0; i < 4; ++i ){
            for( int j = 0; j < 4; ++j ){
                std::cout << array[i][j] << " ";
            }
            std::cout << std::endl;
        }
    }

    MPI_Finalize();
}

The output for {0,3} is the following, while in the other cases only the array is printed:

[xxx@yyy ~]$ mpirun --mca orte_base_help_aggregate 0 -np 2 ./test
-1 -1 -1 0 
-1 -1 -1 1 
-1 -1 -1 2 
-1 -1 -1 3 
[yyy:2436997] *** Process received signal ***
[yyy:2436997] Signal: Segmentation fault (11)
[yyy:2436997] Signal code: Address not mapped (1)
[yyy:2436997] Failing at address: 0x7fc36e0fb493
...

Thank you very much in advance!

EDIT: The SIGSEGV actually happens at the MPI_Finalize() and not at sending or receiving.