[MPICH] MPI_Comm_spawn and argument array_of_errcodes

Rajeev Thakur thakur at mcs.anl.gov
Wed Aug 29 13:48:47 CDT 2007


Wei-keng,
         You are right about both. I have fixed them. If you want to try the
fixed version, replace src/mpid/ch3/src/ch3u_comm_spawn_multiple.c and
src/mpi/spawn/comm_spawn_multiple.c with the attached files.

Rajeev 

> -----Original Message-----
> From: owner-mpich-discuss at mcs.anl.gov 
> [mailto:owner-mpich-discuss at mcs.anl.gov] On Behalf Of Wei-keng Liao
> Sent: Wednesday, August 29, 2007 1:42 AM
> To: mpich-discuss at mcs.anl.gov
> Subject: Re: [MPICH] MPI_Comm_spawn and argument array_of_errcodes
> 
> 
> MPI-2 also defines the argument array_of_info in 
> MPI_Comm_spawn_multiple() 
> is significant only at root. However, in the source files 
> src/mpi/spawn/comm_spawn_multiple.c, this argument is used by all MPI 
> processes (lines 112-114).
> 
> My test code failed when NULL pointer is used on processes other than 
> root.
> 
> Wei-keng
> 
> On Tue, 28 Aug 2007, Wei-keng Liao wrote:
> 
> > 
> > Is the argument array_of_errcodes significant only at root 
> in function 
> > MPI_Comm_spawn() and MPI_Comm_spawn_multiple()?
> > MPI-2 standard seems to indicate it is not.
> > 
> > My test of mpich2 shows the returned values in this 
> argument are only 
> > significant at root. If non-root processes check this 
> argument may see 
> > values other than MPI_SUCCESS, due to this uninitialized 
> integer array.
> > 
> > Wei-keng
> > 
> 
> 
-------------- next part --------------
A non-text attachment was scrubbed...
Name: ch3u_comm_spawn_multiple.c
Type: application/octet-stream
Size: 8578 bytes
Desc: not available
URL: <http://lists.mcs.anl.gov/pipermail/mpich-discuss/attachments/20070829/8a7f424b/attachment.obj>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: comm_spawn_multiple.c
Type: application/octet-stream
Size: 5310 bytes
Desc: not available
URL: <http://lists.mcs.anl.gov/pipermail/mpich-discuss/attachments/20070829/8a7f424b/attachment-0001.obj>


More information about the mpich-discuss mailing list