I have a not-so-trivial, but compact test case for you. I will try it out
with the receipe you gave :).
Regards,
Arjen
Op ma 21 jul 2025 om 13:31 schreef Andre Vehreschild <ve...@gmx.de>:
> Hi all,
>
> we are looking for people having old and/or modern Fortran codes available
> that use Coarrays more or less intensively. Jerry has build a test branch
> on
> gcc's git, so testing is easier than usual:
>
> > For those who need some guidance to the test branch:
> >
> > $ git clone git://gcc.gnu.org/git/gcc.git
> >
> > $ cd gcc
> > $ git checkout remotes/origin/devel/gfortran-test
> > $ git switch -c gfortran-test
> >
> > Configure and build as usual in a separate directory not the source
> directory.
> >
> > cd .. ; mkdir build ; cd build ; ../gcc/configure --prefix=<PREFIX>
> > gmake install
> >
> > Replace <PREFIX> with a writeable full path on your system, e.g.
> > ${HOME}/gcc-16
> >
> > To use the new gcc use:
> >
> > export PATH=${HOME}/gcc-16/bin:$PATH
> >
> > and
> >
> > export LD_LIBRARY_PATH=${HOME}/gcc-16/lib64:$LD_LIBRARY_PATH
> >
> > or
> >
> > export LD_LIBRARY_PATH=${HOME}/gcc-16/lib:$LD_LIBRARY_PATH
> >
> > depending on how your OS names the library directory. Just have a look
> into
> > gcc-16 and use lib64 if it is present, else use lib.
>
> We like everyone to test the new caf_shmem library and report back any
> problems, like "does not compile", "does not run" or "hangs during
> execution".
> If you can narrow down the problem, that would be of great help. If you can
> also share (whether in private or in public) any code, that has issues,
> please
> do not hesitate to contact me or the gfortran mailing list.
>
> To compile your Fortran coarray code add -lcaf_shmem instead of -lcaf_mpi,
> if
> you previously used OpenCoarrays. When using the OpenCoarrays compile
> helper
> `caf` replace it with `gfortran -fcoarray=lib` for comiling and `gfortran
> -fcoarray=lib -lcaf_shmem` for linking.
>
> caf_shmem is multi process shared memory library for using coarrays with
> gfortran from version 16 on. It can provide great speed improvements in
> comparison to MPI-based implementations, but is limited to a single node
> where
> all CPUs can share memory.
>
> Any feedback is greatly appreciated.
>
> Thanks and regards,
> Andre
> --
> Andre Vehreschild * Email: vehre ad gmx dot de
>
