[mpich-discuss] mpifort wrapper compiler?
Jeff Squyres (jsquyres)
jsquyres at cisco.com
Thu Dec 5 10:57:48 CST 2013
Amusingly enough, Bill Gropp was one of the ones who told me that consolidating down to a single "mpifort" would be a Good Idea.
FWIW, Rolf R. and Craig R. also told me this was a good idea -- we did quite a bit of legwork before making this decision.
My only reason for trying to convince you to do this is because it's helpful to users if they can just always rely on "mpifort".
Perhaps there are some special/obscure platforms where it *does* make sense to have a separate "77" compiler, but on those platforms, a) maybe you can/should have a different build of your MPI for that compiler, and/or b) perhaps you can create a separate "mpif77" *just for that platform* (i.e., usually, it's just a sym link to mpifort).
Just my $0.02. :-)
On Dec 5, 2013, at 11:54 AM, Pavan Balaji <balaji at mcs.anl.gov> wrote:
> I’m a Fortran nobody, so I can’t really say one way or another. But from my discussion with BillG and some vendors (e.g., Fujitsu), there still seem to be compilers that are special-optimized for F77. Now, I don’t know why that can’t be done absed on the suffix. If you can convince Bill that this is indeed no problem, I’m happy to change it, or even deprecate Fortran. :-)
> — Pavan
> On Dec 5, 2013, at 10:46 AM, Jeff Squyres (jsquyres) <jsquyres at cisco.com> wrote:
>> On Dec 5, 2013, at 11:09 AM, Pavan Balaji <balaji at mcs.anl.gov> wrote:
>>> It’s very unlikely we’ll drop F77 separation.
>> Just curious -- why? Fortran compiler vendors themselves are consolidating down to single executable names: ifort, pgifort, gfortran, ... etc.
>> This was one of the main motivations for OMPI: we're reflecting what the underlying compiler is, not adding our own distinctions on top of them. If the underlying compiler distinguishes between free/fixed form via the .suffix, then we (MPI implementations) shouldn't change that.
>> I guess my question is: why would you use "mpif77" instead of "mpifort"? Or, put differently, what do you need to compile with mpif77 that would not compile properly with mpifort?
>>> F90 and F08 separation might not be a big issue, but we’ll likely just maintain them as separate symbolic links to mpifort. For now, we could add a symbolic link from mpifort to mpif90.
>>> FWIW, we already use FC and FCFLAGS for F90+. But we still use F77 for F77 compilers.
>> Keep in mind that there have not been any "Fortran 77" compilers for over 30 years.
>> The distinction is artificial and arbitrary...
>> Jeff Squyres
>> jsquyres at cisco.com
>> For corporate legal information go to: http://www.cisco.com/web/about/doing_business/legal/cri/
>> discuss mailing list discuss at mpich.org
>> To manage subscription options or unsubscribe:
> Pavan Balaji
> discuss mailing list discuss at mpich.org
> To manage subscription options or unsubscribe:
jsquyres at cisco.com
For corporate legal information go to: http://www.cisco.com/web/about/doing_business/legal/cri/
More information about the discuss