https://gcc.gnu.org/bugzilla/show_bug.cgi?id=120958
--- Comment #25 from Thomas Koenig <tkoenig at gcc dot gnu.org> --- (In reply to rguent...@suse.de from comment #24) > On Thu, 17 Jul 2025, anlauf at gcc dot gnu.org wrote: > > So if FOO is a wrapper subroutine that calls MPI_I*, one needs to find all > > of them in the code? > > Technically yes. You already have to be unlucky to see SRA do sth > given with Fortran the chances the address of something is taken > is quite high (everything is passed by reference). The chances of addresses taken by the compiler is high, but by the language definition something has to have the TARGET attribute to be pointed at by something. What we could maybe do is to make the relevant MPI functions intrinsics (sort of) and warn if an argument which should have ASYNCHRONOUS does not have it, to be included with -Wall.
