Open MPI logo

Open MPI User's Mailing List Archives

  |   Home   |   Support   |   FAQ   |   all Open MPI User's mailing list

Subject: Re: [OMPI users] printing text fixes a problem?
From: Jeff Squyres (jsquyres) (jsquyres_at_[hidden])
Date: 2011-02-01 06:25:44


That's not quite right - a call to MPI-finalize does not terminate any processes.

If you're seeing this kind of instability, check the usual suspects such as ensuring you have a totally homogeneous environment (same OS, same version of OMPI, etc).

Sent from my PDA. No type good.

On Feb 1, 2011, at 4:03 AM, "David Zhang" <solarbikedz_at_[hidden]> wrote:

> According to the mpi_finalize documentation, a call to mpi_finalize terminate all processes. I have ran into this problem before where one process calls mpi_finalize before other processes reach the same line of code and cause errors/hang ups. Put a mpi_barrier(mpi_comm_world) before mpi_finalize would do the trick.
>
> On Mon, Jan 31, 2011 at 11:40 PM, abc def <cannonjunk_at_[hidden]> wrote:
> Hello,
>
> I'm having trouble with some MPI programming in Fortran, using openmpi.
> It seems that my program doesn't work unless I print some unrelated text to the screen. For example, if I have this situation:
>
> *** hundreds of lines cut ***
> IF (irank .eq. 0) THEN
> CALL print_results1(variable)
> CALL print_results2(more_variable)
> END IF
> print *, "done", irank
> CALL MPI_FINALIZE(ierr)
> END PROGRAM calculation
>
> The results are not printed unless I include this "print done irank" penultimate line.
> Also, despite seeing that all ranks reach the print statement, the program hangs, as if they have not all reached MPI_FINALIZE.
>
> Can anyone help me? Why does it do this?
>
> I also had many times where the program would crash if I didn't include a print statement in a loop. I've been doing Fortran programming for a while, and this is my nightmare debugging scenario since I've never been able to figure out why the simple printing of statements magically fixes the program, and I usually end up having to go back to a serial solution, which is really slow.
>
> If anyone might be able to help me, I would be really really grateful!!
>
> Thank you.
>
> Tom
>
>
> _______________________________________________
> users mailing list
> users_at_[hidden]
> http://www.open-mpi.org/mailman/listinfo.cgi/users
>
>
>
> --
> David Zhang
> University of California, San Diego
> _______________________________________________
> users mailing list
> users_at_[hidden]
> http://www.open-mpi.org/mailman/listinfo.cgi/users