(j3.2006) [MPI3 Fortran] Feedback from Fortran J3 meeting
Thu May 29 11:42:34 EDT 2008
Robert Corbett wrote:
> I very much doubt that Coarray Fortran will ever overtake MPI in usage.
> Failure to support MPI in Fortran might drive those MPI users away from
> using Fortran.
Because of the huge build up of existing code, I suspect you are right.
However, based on my experience writing parallel codes with coarrays.
I'd be surprised if many new codes were written with MPI once f08 was
widely available. Same argument applies to UPC. It's just that the
number of new codes is always relatively small.
I don't understand the second sentence. Despite our apparent lack of
support currently, almost every Fortran program I see uses MPI. If we
improve things by helping the MPI folks make a better interface module,
things should be even better. I see no evidence of people departing
Fortran because of MPI.
> MPI works very well at low optimization levels.
Very few of the MPI codes I see are compiled with low optimization
levels. MPI works well at high optimization levels, too. It's just a
pain in the neck to write the code in the first place.
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