If you're only running with a few MPI processes, you might be able to get away with:
mpirun -np 4 valgrind ./my_mpi_application If you run any more than that, the output gets too jumbled and you should output each process' valgrind stdout to a different file with the --log-file option (IIRC). I personally like these valgrind options: valgrind --num-callers=50 --db-attach=yes --tool=memcheck --leak-check=yes --show-reachable=yes On May 18, 2011, at 8:49 AM, Paul van der Walt wrote: > Hi Jeff, > > Thanks for the response. > > On 18 May 2011 13:30, Jeff Squyres <jsquy...@cisco.com> wrote: >> *Usually* when we see segv's in calls to alloc, it means that there was >> previously some kind of memory bug, such as an array overflow or something >> like that (i.e., something that stomped on the memory allocation tables, >> causing the next alloc to fail). >> >> Have you tried running your code through a memory-checking debugger? > > I sort-of tried with valgrind, but I'm not really sure how to > interpret the output (I'm not such a C-wizard). I'll have another look > a little later then and report back. I suppose I should RTFM on how to > properly invoke valgrind so it makes sense with an MPI program? > > Paul > > -- > O< ascii ribbon campaign - stop html mail - www.asciiribbon.org > _______________________________________________ > users mailing list > us...@open-mpi.org > http://www.open-mpi.org/mailman/listinfo.cgi/users -- Jeff Squyres jsquy...@cisco.com For corporate legal information go to: http://www.cisco.com/web/about/doing_business/legal/cri/
