Speaking of ebuilds and fortran, I've been nosing around trying to get
things working with lam-mpi (I'm actually just administering a machine that
someone else is trying to use mpi on).
>From looking at the various ebuilds for lam-mpi, lapack, and blas, it looks
like newer ebuilds use fortran.eclass, while older ones hardcode the
fortran compiler they want to use into the ebuild.  There is no support
whatsoever for Portland Group (pgf).
To get everything working with pgf90, I had to setup some ebuilds in an
overlay which, a) avoided fortran.eclass and b) removed the FC/F77
hardcoding in the ebuilds for blas and lapack, so that my FC setting in
make.conf would be used.

fortran.eclass has to be avoided, because it simply punts if the compiler
you request is not in its hardcoded list of blessed compilers.
It looks like fortran.eclass is a work in progress.  Anybody have more
information on that?  I'm wondering if I can help so that I can pick a
fortran compiler to use on a system, without having to hack ebuilds in a
portage overlay.

-bryan

Donnie Berkholz writes:
> Bryan Green wrote:
> > Interesting question.  Is there a difference?
> > I am under the impression that lam-mpi needs to be built with pgf for
> > programs that use lam-mpi to be built with pgf.  Am I mistaken?
> > I any case, I found that I get get it to build by setting the FC environmen
> t
> > variable.  My initial error was apparently caused because of the lack of a
> > license server setting.
> 
> Good to hear it! If you ever come across an ebuild where FC/F77, CC, or
> CXX aren't respected, please file a bug.
> 
> Thanks,
> Donnie
-- 
[email protected] mailing list

Reply via email to