Woptim/update local packages

[adrienbernede]

Summary

• This PR is an improvement.
• It does the following:
  • Update local Spack packages based on current spack@develop version. Keep the logic that was only present locally when necessary to pass tests (see commit messages)
  • Modifies the CI to add necessary variants.

Important notes

The raja package diff is hard to read because of the switch to CachedCMakePackages. Reviewers should compare the new version with https://github.com/spack/spack/blob/develop/var/spack/repos/builtin/packages/raja/package.py.

By default, "shared" is "True" in the upstream spack package. I set "shared" to "false" in the global variants settings in CI here:

By default, "tests" is "False" in the upstream spack package. I set "tests" to "true" in the global variants settings in CI here:

RAJA/.gitlab/custom-jobs-and-variables.yml

Line 21 in eb97f73

PROJECT_RUBY_VARIANTS: "~shared +openmp +tests"

RAJA/.gitlab/custom-jobs-and-variables.yml

Line 31 in eb97f73

PROJECT_CORONA_VARIANTS: "~shared ~openmp +tests"

RAJA/.gitlab/custom-jobs-and-variables.yml

Line 40 in eb97f73

PROJECT_LASSEN_VARIANTS: "~shared +openmp +tests"

Relates to:
LLNL/Umpire#793.
LLNL/CHAI#218

[adrienbernede]

@ reviewers:

I took the packages from spack@develop, ran the CI, and added the missing parts from the initial local packages to pass the tests.

It is possible that not all the diffs were ported to the new local packages. But since we are now passing the tests with packages much closer to the upstream packages, it seems to me that we should take this as a new base.

Also remember that the goal is to minimize the diffs with the upstream package. We should consider pushing the new local packages to Spack upstream.

This is part of the process of choosing a Spack reference for CI testing described here:
https://radiuss-shared-ci.readthedocs.io/en/latest/sphinx/user_guide/how_to.html#choose-a-spack-reference-commit-or-branch

[rhornung67]

Adding more reviewers since we will discuss Gitlab CI, including this at the project meeting tomorrow morning.

[rhornung67]

@adrienbernede there is an issue with the spack spec in one of the lassen xl+cuda builds that is causing the failure.

[rhornung67]

A couple of spack package version updates.

[adrienbernede]

@rhornung67 will do.

[rhornung67]

@adrienbernede this looks good to me. I would like @davidbeckingsale to review.

[adrienbernede]

@rhornung67 OK.

For the records, I am trying to sync things with Umpire and CHAI.

Illustration of the challenges:

• Since CHAI@develop needs 2022.10 releases, we need the updated packages locally in CHAI.
• the 2022.10 releases require BLT 5.2.0 which is not present in [email protected]. This means we need a local BLT package for BLT... again. Except if we push all our new packages to a Spack PR and get it merged.

Suggested path forward:

1. Make things work without updating Spack yet.
2. Push the new packages to Spack in a joint PR (RAJA, Umpire, CHAI, camp?).
3. Update Spack in our project and remove identical local packages.

[adrienbernede]

Note: This is not ready, I am finding things I need to improve first. Will report back.

[rhornung67]

Looks good to me. Thank you @adrienbernede

[rhornung67]

@davidbeckingsale should I merge this? I don't foresee any issues with it.

[davidbeckingsale]

@davidbeckingsale should I merge this? I don't foresee any issues with it.

Yup, sounds good!