MPI，通过其中一个过程产生一个孩子

Question

I am taking my first steps in the world of parallel programming with Open MPI. What I'm trying to achieve is to start an application with two processes (this can be achieved with mpirun -n 2 ...) and make the first process from these both spawn another instance of itself. Here is the sketch in c++:

// mpi_spawn_cpp2.cc
#include "mpi.h"
#include <iostream>

using namespace std;

int main( int argc, char *argv[] )
{
    int rank, size;
    int errcode;
    MPI::Intercomm intercomm;

    MPI::Init( argc, argv );

    rank = MPI::COMM_WORLD.Get_rank();
    size = MPI::COMM_WORLD.Get_size();
    cout << "world size: " << size << endl;

    if (rank == 0) {
        intercomm = MPI::COMM_WORLD.Spawn("./mpi_spawn_cpp2", MPI::ARGV_NULL, 1, MPI::INFO_NULL, 0, &errcode);
        cout << "intercomm size: " << intercomm.Get_size() << endl;
    }

    MPI::Finalize();
    return 0;
}

Although it compiles flawlessly the problem with the above code is that it doesn't seem to return from the Spawn function, which means the process is in fact not spawned. What is wrong with the above code and what should be corrected in order for it to work?

Your help is greatly appreciated.

Answer 1

MPI_Comm_spawn is collective over the intracommunicator but you are calling it only from rank 0 and that's why it hangs. Try this instead:

intercomm = MPI::COMM_WORLD.Spawn("./mpi_spawn_cpp2", MPI::ARGV_NULL, 1, MPI::INFO_NULL, 0, &errcode);
if (rank == 0) {
    cout << "intercomm size: " << intercomm.Get_size() << endl;
}

A word of advice. Since it is your first steps in MPI, I would recommend that you learn the C bindings instead. The C++ bindings are deprecated in the current MPI standard version 2.2 and will be completely dropped from the upcoming version 3.0.