Open MPI logo

Open MPI User's Mailing List Archives

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

Subject: Re: [OMPI users] MPI-3 Fortran feedback
From: Jeff Squyres (jsquyres_at_[hidden])
Date: 2009-10-27 16:50:07


On Oct 27, 2009, at 4:40 PM, Nifty Tom Mitchell wrote:

> If I recall correctly the OMPI build does not generate a full Fortran
> <--> C mapping of all types for all functions. Was the reason that
> the
> resulting library has too many permutations and symbols to test,
> build and
> link to.
>

The OMPI F90 module does not create prototypes for all possible MPI
F90 bindings because there are (IIRC) over 5M possibilities. There's
currently no compiler that can handle that many in a single F90 module
file.

> While not a co-array related comment is there a way to generate
> the necessary bindings on the fly and facilitate stronger type
> checking and perhaps open the door for richer optimizations
> from the compiler folk.
>
> http://www.open-mpi.org/papers/euro-pvmmpi-2005-fortran/euro-pvm-mpi-2005-fortran.pdf
>

Yes, Craig and I were the authors of that paper. :-) We used that as
the starting point for discussions in the MPI-3 working group to
improve the MPI F90 bindings problem. The final solution we ended up
on ended up being a little different -- see the MPI-3 Fortran WG wiki
page for a discussion about this issue:

     https://svn.mpi-forum.org/trac/mpi-forum-web/wiki/FtnWikiPage

> Will Fortran 2008 revisit some of these issues in a problematic way?
> Or is this such old news that folk just live with it?
>

We think we have a good solution to completely solve this particular
issue. See the wiki page for details.

-- 
Jeff Squyres
jsquyres_at_[hidden]