Re: PTP Communication problems when using a single process

--- William Gropp <[email protected]> wrote:

> >##########################
> >int id, ntasks, tag=100;
> >int msg[2];
> >
> >MPI::Init(argc, argv);
> >ntasks = MPI::COMM_WORLD.Get_size();
> >id = MPI::COMM_WORLD.Get_rank();
> >
> >msg[0] = id;
> >msg[1] = ntasks;
> >
> >MPI::COMM_WORLD.Send(msg, 2, MPI::INT, 0, tag);
> >MPI::COMM_WORLD.Recv(msg, 2, MPI::INT, 0, tag);
> >
> >MPI::Finalize();
> >###############################
> >
> >The problem is that the same process is both sender
> >and receiver (I know that it does not make sense
> too
> >much, but I need to test something with this code).
> >
> >If I use ethernet (ch_p4), it works fine.  But
> running
> >it on myri core-dumps during receiving.
> >
> >Does anybody have an idea what's happening?
> 
> It shouldn't dump core, but it would be valid for an
> MPI implementation to 
> hang here.
> 

Yes, I read about possible deadlock warning in the
documentation. Since it worked fine without any
problem with ch_p4 I thought it may work fine with
myri too.

> To check on the core dump, does it help to pass an
> MPI::Status to Recv?
> 

Nope. It looks like core dumps during Recv, it does
not return from there properly at all.

Eylul



__________________________________
Do you Yahoo!?
Yahoo! Mail - More reliable, more storage, less spam
http://mail.yahoo.com