search

GEOS-ESM / jedi_bundle

Repo for building JEDI packages
https://geos-esm.github.io/jedi_bundle/
Apache License 2.0
2 stars 3 forks source link

Spack-Stack 1.7.0 and Discover SLES15 Modules #32

Closed Dooruk closed 6 months ago

Dooruk commented 6 months ago

Including SLES15 platform and updating to spack-stack 1.7.0 modules.

Also changes the gsibec version

gmao-jjin3 commented 6 months ago

@Dooruk I cannot checkout your branch "feature/discover_sles15". Strange.

gmao-jjin3 commented 6 months ago

@Dooruk I cannot checkout your branch "feature/discover_sles15". Strange.

My bad. I didn't see that it was in "jedi-bundle".

Dooruk commented 6 months ago

Looks reasonable. Apologies if I broke the platform lookup thing for Disco.

It was failing while discerning between the two Discover OS.

JEDI builds with the SLES12 Intel modules but pip install is giving Illegal instruction error. I will ask Dom, I didn't get this error with SLES15 previously.

@mathomp4 could this be because we are trying to implement a unified spack environment on two different OS's?

mathomp4 commented 6 months ago

Illegal instruction? Hmm. Well, you can get that if something compiled in a flag not supported on a chipset (say an Intel-only instruction that AMD chips have).

But often that's just a catch all for "things crashed".

Can you reproduce the crash on the commandline?

Dooruk commented 6 months ago

Illegal instruction? Hmm. Well, you can get that if something compiled in a flag not supported on a chipset (say an Intel-only instruction that AMD chips have).

Dom switched the intel compiler to 2021.6.0 and now all is good. I need to make the necessary updates now...

Potential explanation by Dom:

It is a combination of switching from zlib to zlib-ng AND using intel@2021.50 AND being on a particular system. We are only seeing this on Discover SLES12 and S4, and the problems go away when using an older 2021.3.0, 2021.4.0 or newer 2021.6.0+ version of the compiler. On other systems that use 2021.5.0 (e.g. the navy nautilus machine, which is penguin linux), we don’t see the issue.

Bonus: Screenshot 2024-04-04 at 10 16 49 AM