[mpich-discuss] Issue regarding MPI_Comm_spawn_multiple

Pavan Balaji balaji at mcs.anl.gov
Mon May 6 19:46:11 CDT 2013


It's impossible to tell the problem from your description below.  Can
you please send us a small test program that reproduces your problem?

 -- Pavan

On 05/06/2013 06:45 AM US Central Time, Mahesh Doijade wrote:
> Hi,
>       I am using  MPI_Comm_spawn_multiple for dynamic process creation
> and passing different arguments to each process through this function, 
> but the code runs perfectly fine for some inputs and throws SEGFAULT for
> some other inputs the code sample is
> 
> 
> /*Arguments generation starts*/
> char ***procs_argv =     procs_argv = (char ***) malloc(sizeof(char **)
> * no_processes_to_launch);
> 
>     for(i=0; i < no_processes_to_launch; i++)
>     {
>         procs_argv[i] =    ( char **)malloc(sizeof(char *) * 2);
>         procs_argv[i][0] = ( char *) malloc(sizeof(char)*4);
>         procs_argv[i][1] = ( char *) malloc(sizeof(char)*10);
>         sprintf(procs_argv[i][0], "%d", ppstat_ptr[i].no_of_pivots);   
> //Total no. of pivot variables
>         sprintf(procs_argv[i][1], "%d",
> all_sizes_elems_fep[i+1]);//Total bytes of all sizes data.
>     }
> 
> /*Arguments generation finish*/
>  and process is launched as -- 
> MPI_Comm_spawn_multiple(no_processes_to_launch, cmds, procs_argv,
> nprocs, infos, 0, MPI_COMM_WORLD, &worker_comm, errcodes);
> 
> Error given when debugged through gdb :
> 
> in _IO_vfprintf_internal (s=<value optimized out>, format=<value
> optimized out>, ap=<value optimized out>) at vfprintf.c : 1603 ,
> process_string_arg(((struct printf_spec *)NULL));
> 
> 
> 
> 
> 
> -- 
> 
> Regards,
> -- Mahesh Doijade
> 
> 
>                                                                           
> 
> 
> _______________________________________________
> discuss mailing list     discuss at mpich.org
> To manage subscription options or unsubscribe:
> https://lists.mpich.org/mailman/listinfo/discuss
> 

-- 
Pavan Balaji
http://www.mcs.anl.gov/~balaji



More information about the discuss mailing list