DeinoMPI

The Great and Terrible implementation of MPI-2

MPI_Comm_join

Create a communicator by joining two processes connected by a socket.

int MPI_Comm_join(
  int fd,
  MPI_Comm *intercomm
);

Parameters

fd: [in] socket file descriptor
intercomm: [out] new intercommunicator (handle)

Remarks

MPI_COMM_JOIN is intended for MPI implementations that exist in an environment supporting the Berkeley Socket interface [14,17]. Implementations that exist in an environment not supporting Berkeley Sockets should provide the entry point for MPI_COMM_JOIN and should return MPI_COMM_NULL.

This call creates an intercommunicator from the union of two MPI processes which are connected by a socket. MPI_COMM_JOIN should normally succeed if the local and remote processes have access to the same implementation-defined MPI communication universe.

The socket must be quiescent before MPI_COMM_JOIN is called and after MPI_COMM_JOIN returns. More specifically, on entry to MPI_COMM_JOIN, a read on the socket will not read any data that was written to the socket before the remote process called MPI_COMM_JOIN.

Advice to users. An MPI implementation may require a specific communication medium for MPI communication, such as a shared memory segment or a special switch. In this case, it may not be possible for two processes to successfully join even if there is a socket connecting them and they are using the same MPI implementation.

Thread and Interrupt Safety

This routine is thread-safe. This means that this routine may be safely used by multiple threads without the need for any user-provided thread locks. However, the routine is not interrupt safe. Typically, this is due to the use of memory allocation routines such as mallocor other non-MPICH runtime routines that are themselves not interrupt-safe.

Notes for Fortran

All MPI routines in Fortran (except for MPI_WTIME and MPI_WTICK) have an additional argument ierr at the end of the argument list. ierris an integer and has the same meaning as the return value of the routine in C. In Fortran, MPI routines are subroutines, and are invoked with the call statement.

All MPI objects (e.g., MPI_Datatype, MPI_Comm) are of type INTEGERin Fortran.

Errors

All MPI routines (except MPI_Wtime and MPI_Wtick) return an error value; C routines as the value of the function and Fortran routines in the last argument. Before the value is returned, the current MPI error handler is called. By default, this error handler aborts the MPI job. The error handler may be changed with MPI_Comm_set_errhandler (for communicators), MPI_File_set_errhandler (for files), and MPI_Win_set_errhandler (for RMA windows). The MPI-1 routine MPI_Errhandler_set may be used but its use is deprecated. The predefined error handler MPI_ERRORS_RETURN may be used to cause error values to be returned. Note that MPI does not guarentee that an MPI program can continue past an error; however, MPI implementations will attempt to continue whenever possible.

MPI_SUCCESS: No error; MPI routine completed successfully.

MPI_ERR_ARG: Invalid argument. Some argument is invalid and is not identified by a specific error class (e.g., MPI_ERR_RANK).

Example Code

The following sample code illustrates MPI_Comm_join.

#include "mpi.h"
#include <stdio.h>
#include <winsock2.h>
#include <ws2tcpip.h> /* socklen_t */
#include <string.h>

#define COUNT 1024

int main( int argc, char *argv[] )
{
    int sendbuf[COUNT], recvbuf[COUNT], i;
    int err=0, rank, nprocs, errs=0;
    MPI_Comm intercomm;
    int listenfd, connfd, port, namelen;
    struct sockaddr_in cliaddr, servaddr;
    struct hostent *h;
    char hostname[MPI_MAX_PROCESSOR_NAME];
    socklen_t len, clilen;

    MPI_Init( &argc, &argv );
    MPI_Comm_size(MPI_COMM_WORLD,&nprocs);
    MPI_Comm_rank(MPI_COMM_WORLD, &rank);
    if (nprocs != 2) {
        printf("Run this program with 2 processes\n");fflush(stdout);
        MPI_Abort(MPI_COMM_WORLD,1);
    }
    if (rank == 1) {
        /* server */
        listenfd = socket(AF_INET, SOCK_STREAM, 0);
        if (listenfd < 0) {
            printf("server cannot open socket\n");fflush(stdout);
            MPI_Abort(MPI_COMM_WORLD,1);
        }

        memset(&servaddr, 0, sizeof(servaddr));
        servaddr.sin_family = AF_INET;
        servaddr.sin_addr.s_addr = htonl(INADDR_ANY);
        servaddr.sin_port = 0;
        err = bind(listenfd, (struct sockaddr *) &servaddr, sizeof(servaddr));
        if (err < 0) {
            errs++;
            printf("bind failed\n");fflush(stdout);
            MPI_Abort(MPI_COMM_WORLD,1);
        }
        len = sizeof(servaddr);
        err = getsockname(listenfd, (struct sockaddr *) &servaddr, &len);
        if (err < 0) {
            errs++;
            printf("getsockname failed\n");fflush(stdout);
            MPI_Abort(MPI_COMM_WORLD,1);
        }
        port = ntohs(servaddr.sin_port);
        MPI_Get_processor_name(hostname, &namelen);
        MPI_Send(hostname, namelen+1, MPI_CHAR, 0, 0, MPI_COMM_WORLD);
        MPI_Send(&port, 1, MPI_INT, 0, 1, MPI_COMM_WORLD);
        err = listen(listenfd, 5);
        if (err < 0) {
            errs++;
            printf("listen failed\n");fflush(stdout);
            MPI_Abort(MPI_COMM_WORLD, 1);
        }
        clilen = sizeof(cliaddr);
        connfd = accept(listenfd, (struct sockaddr *) &cliaddr, &clilen);
        if (connfd < 0) {
            printf("accept failed\n");fflush(stdout);
            MPI_Abort(MPI_COMM_WORLD, 1);
        }
    }
    else {
        /* client */
        MPI_Recv(hostname, MPI_MAX_PROCESSOR_NAME, MPI_CHAR, 1, 0, MPI_COMM_WORLD, MPI_STATUS_IGNORE);
        MPI_Recv(&port, 1, MPI_INT, 1, 1, MPI_COMM_WORLD, MPI_STATUS_IGNORE);
        h = gethostbyname(hostname);
        if (h == NULL) {
            printf("gethostbyname failed\n");fflush(stdout);
            MPI_Abort(MPI_COMM_WORLD, 1);
        }
        servaddr.sin_family = h->h_addrtype;
        memcpy((char *) &servaddr.sin_addr.s_addr, h->h_addr_list[0], h->h_length);
        servaddr.sin_port = htons(port);
        /* create socket */
        connfd = socket(AF_INET, SOCK_STREAM, 0);
        if (connfd < 0) {
            printf("client cannot open socket\n");fflush(stdout);
            MPI_Abort(MPI_COMM_WORLD, 1);
        }
        /* connect to server */
        err = connect(connfd, (struct sockaddr *) &servaddr, sizeof(servaddr));
        if (err < 0) {
            errs++;
            printf("client cannot connect\n");fflush(stdout);
            MPI_Abort(MPI_COMM_WORLD, 1);
        }
    }

    MPI_Barrier(MPI_COMM_WORLD);
    /* To improve reporting of problems about operations, we change the error handler to errors return */
    MPI_Comm_set_errhandler( MPI_COMM_WORLD, MPI_ERRORS_RETURN );
    err = MPI_Comm_join(connfd, &intercomm);
    if (err)
    {
        errs++;
        printf("Error in MPI_Comm_join %d\n", err); fflush(stdout);
    }
    /* To improve reporting of problems about operations, we change the error handler to errors return */
    MPI_Comm_set_errhandler( intercomm, MPI_ERRORS_RETURN );
    for (i=0; i<COUNT; i++) {
        recvbuf[i] = -1;
        sendbuf[i] = i + COUNT*rank;
    }
    err = MPI_Sendrecv(sendbuf, COUNT, MPI_INT, 0, 0, recvbuf, COUNT, MPI_INT, 0, 0, intercomm, MPI_STATUS_IGNORE);
    if (err != MPI_SUCCESS) {
        errs++;
        printf( "Error in MPI_Sendrecv on new communicator\n" );fflush(stdout);
    }
    for (i=0; i<COUNT; i++) {
        if (recvbuf[i] != ((rank+1)%2) * COUNT + i)
            errs++;
    }
    MPI_Barrier(MPI_COMM_WORLD);
    err = MPI_Comm_disconnect(&intercomm);
    if (err != MPI_SUCCESS) {
        errs++;
        printf( "Error in MPI_Comm_disconnect\n" );fflush(stdout);
    }
    MPI_Finalize();
    return errs;
}

DOWNLOAD

Win32 DeinoMPI.2.0.1.msi
Win64 DeinoMPI.x64.2.0.1.msi