I just wanted to let interested parties know that (spurred by the needs of scipy), we’re discussing in conda-forge which fortran compiler to use in the future. Flang is obviously on our radar, but ultimately it will be a pragmatic choice - if (say) LFortran works better/faster/earlier, we’ll most likely use that.
I thought I’d open this thread here in case this information is interesting to anyone.
PS. I haven’t forgotten my promise to try the status of compiling scipy with flang 15, but lack of time and infra changes (+ other breakage in the builds) in LLVM 15 have stalled this effort so far.
FYI the Fujitsu run was done with an old version of flang so the numbers are likely to be very different.
I know, and since SciPy additionally uses no newer Fortran features, I believe this would be realistic to solve even if there are still are bugs for compiling SciPy currently.
On a different note, has anyone thought about meson integration of flang yet? Feel free to join the discussion on the conda-forge, there’s some meson devs there as well, so might be more effective for this particular subject (just wanted to let people know without gratuitously tagging flang contributors).
Just for the record, Usman Nadeem has been looking into compiling SciPy with flang too: [LLVM-762] - Jira