If I understand correctly, with gfotran I will be able to link to petsc cuda-c++ , but I wont be able to use the fortran cuda interface directly (so I wont be able to write my own gpu-accelerated fortran). Is that correct?
On Sat, Aug 2, 2025, 13:25 Satish Balay <balay....@fastmail.org> wrote: > On Sat, 2 Aug 2025, Renato E Poli wrote: > > > *>> And a recommendation to use gfortran instead of nvfortran* > > The other application is using nvfortran, and now we are trying to link > > PETSC to it. > > > > I guess I understood the main points. > > I need to work on it a bit now and get a better idea of what is going on. > > petsc-3.22 should work with nvfortran. > > Fortran interface in petsc is being over-hauled (in 3.23, and future 3.24 > releases) requiring F2008 (and F2023?) features - that some of the older > compilers don't support. > > [and older petsc releases might have other issues - say using latest cuda > version, cmake version etc...] > > Satish >
