Random FFTW test failures on aarch64/neoverse_v1

[casparvl]

We've experienced test failures with the FFTW make check command, in the MPI part of the unit tests. See e.g. this pr for FFTW 3.3.10 for foss-2022a and this pr with FFTW 3.3.10 for foss-2021b. The latter than build succesfully in another PR

Failures don't always happen in the same test. Furthermore, interactively, the tests seem to pass way more often, around 90% of the time, whereas building with the bot for foss-2022a seemed to fail consistently.

Created an upstream issue here

[lrbison]

@boegel

Should we rename this issue to "Open MPI test failures on aarch64/neoverse_v1" and coordinate here rather than bother the FFTW folks over in their repo?

[boegel]

@lrbison I'd rather keep this one specific to FFTW.MPI (because we use tracker issues like this in eessi-2023.06-known-issues.yml, see also #458

However, we can create a more general issue in the EESSI support portal @ https://gitlab.com/eessi/support

[lrbison]

Done. A more general issue against Open MPI on neoverse_v1: https://gitlab.com/eessi/support/-/issues/41