(j3.2006) [MPI3 Fortran] Feedback from Fortran J3 meeting
Thu May 29 13:35:00 EDT 2008
I generally agree with Aleks here. We should not embed particular
libraries like MPI in the standard - it is possibly subject to
obsolescence (we might have picked PVM before), and inevitably subject
to incompleteness (...I want LaPack!). Instead, we have chosen to
interface with C under the assumption that the combination of C and
Fortran calling conventions should cover most libraries. The problem
with some C library routines (including many in MPI) is that our
interface does not handle (void *) dummy arguments well. Fixing that is
a big part of the current discussion.
Vendors, for performance reasons, of course, can recognize MPI routines
as additional intrinsics and optimize their calls. Ditto for BLAS and
LaPack routines. But I think this sort of thing should remain outside
the Fortran standard. We're in the business of defining a language, not
writing people's programs or picking their library packages for them.
(Yeah, I know there is a gray area with the Fortran intrinsics, but most
of those are not really library routines.)
Aleksandar Donev wrote:
> On Wednesday 28 May 2008 10:11, Keith Bierman wrote:
>> it might just be a
>> lot easier to have an optional part of the Standard which puts MPI in
>> as special modules
> While not opposed to this in principle, I believe that it is a mistake to not
> spend the time making a general facility. There are a lot more libraries out
> there than MPI. Sure, vendors won't spend the time to support them explicitly
> because few users use each of them. But if you add all the libraries in use,
> it will come out to more than MPI.
> That said, MPI is a good prototype of what all we need to be able to handle,
> mostly because it is soo big and has so many things in it.
> Also, I am not convinced providing the generic support is that difficult.
> Especially if all vendors already do it by crutches (directives, compiler
> switches, special modules, whatever).
> J3 mailing list
> J3 at j3-fortran.org
Bill Long longb at cray.com
Fortran Technical Support & voice: 651-605-9024
Bioinformatics Software Development fax: 651-605-9142
Cray Inc., 1340 Mendota Heights Rd., Mendota Heights, MN, 55120
More information about the J3