https://gcc.gnu.org/bugzilla/show_bug.cgi?id=125584
--- Comment #9 from Richard Biener <rguenth at gcc dot gnu.org> --- (In reply to [email protected] from comment #7) > "vehre at gcc dot gnu.org" <gcc-bugz> --- Comment #6 from Andre Vehreschild > <vehre at gcc dot gnu.org> --- > > (In reply to [email protected] from comment #5) > >> Solaris doesn't do Linux-style lazy allocation, but requires backing > >> store. This way, processes aren't subject to the OOM killer... I > >> suspect other OSes behave the same way. > > > > Well, then the testdriver caf.exp probably should set the env variable for > > all > > OS, but Linux and the ones that support lazy backing memory allocation. But > > how > > do we figure which these are? > > But do the tests actually *need* the 256 MB allocations anywhere? As an > experiment, I just ran the caf.exp tests with > GFORTRAN_SHARED_MEMORY_SIZE=2M and they still PASS just fine. > > If there is any point in testing with the 256 MB default at all, > restrict that to targets *known* to support lazy allocation, which would > be linux at least initially. I agree to that, this sounds most reasonable. Having a fixed shared memory segment size sounds ... "interesting" though. Is this how competing coarray implementations handle this as well?
