[MPICH] Program crashes when multiple threads make MPI Calls

[Ashok Babu Amara]

Hi All,

I am making MPI calls from two threads and it doesn't seem to be
working. I have one thread constantly doing MPI_Send's and MPI_Recv's
and another thread which is constantly doing MPI_Allreduce.

When I run the program it either hangs or prints this message :
[0] Abort: header->id != c->next_packet_expected at line 730 in file
mpid/vapi/viacheck.c

The version of MPICH being used is 1.2.6 and the thread level support
is  MPI_THREAD_FUNNELED.

Can someone advise what is possibly wrong ?

I have attached the code that is being tested.

Thanks,
- Ashok
-------------- next part --------------
#include <stdio.h>
#include <assert.h>
#include <pthread.h>
#include "mpi.h"

int myRank;
int numProcs;
MPI_Comm com1;
MPI_Comm com2;

void* t1Main(void*);
void* t2Main(void*);

void* t1Main(void* data)
{
    int retCode;
    int val;
    MPI_Status status;

    while(1)
    {
        if(myRank == 0)
        {
            retCode = MPI_Send(&val, 1, MPI_INT, 1, 0, com1);
            assert(retCode == MPI_SUCCESS);
            retCode = MPI_Recv(&val, 1, MPI_INT, 1, 0, com1, &status);
            assert(retCode == MPI_SUCCESS);
        }
        else
        {
            retCode = MPI_Recv(&val, 1, MPI_INT, 0, 0, com1, &status);
            assert(retCode == MPI_SUCCESS);
            retCode = MPI_Send(&val, 1, MPI_INT, 0, 0, com1);
            assert(retCode == MPI_SUCCESS);
        }
        printf("%d: Send & Recv done\n", myRank);
    }
}

void* t2Main(void* data)
{
    int val;
    int retCode;

    val = 20;

    while(1)
    {
        retCode = MPI_Allreduce(&myRank, &val, 1, MPI_INT, MPI_MAX, com2);
        assert(retCode == MPI_SUCCESS);
        printf("%d: MPI_Allreduce done\n", myRank);
    }
}


int main(int argc, char** argv)
{
    int provided;
    int retCode;
    pthread_t thandle1;
    pthread_t thandle2;

    MPI_Init_thread(&argc, &argv, MPI_THREAD_MULTIPLE, &provided);
    MPI_Comm_rank(MPI_COMM_WORLD, &myRank);
    MPI_Comm_size(MPI_COMM_WORLD, &numProcs);

    retCode = MPI_Comm_dup(MPI_COMM_WORLD, &com1);
    assert(retCode == MPI_SUCCESS);

    retCode = MPI_Comm_dup(MPI_COMM_WORLD, &com2);
    assert(retCode == MPI_SUCCESS);

    retCode = pthread_create( &thandle1, NULL, t1Main, NULL);
    assert(retCode == 0);

    retCode = pthread_create( &thandle2, NULL, t2Main, NULL);
    assert(retCode == 0);

    pthread_join(thandle1, NULL);
    pthread_join(thandle2, NULL);
}