Add ROCm support to openmpi.py

[zerefwayne]

This pull request adds rocm specific dependencies to known_dependencies and sanity check commands to ensure OpenMPI is properly linked to rocm libraries.

[zerefwayne]

UCX-ROCm and UCC-ROCm are built as complete standalone installations rather
than as additional component modules layered on top of the base UCX/UCC builds.

Two problems emerged with the layered approach:

1. When a plain UCX build and a UCX-ROCm component module were both present in the module environment, ucx_info reported the plain UCX configuration (without --with-rocm) because the plain UCX binary took precedence on PATH. The ROCm transport components were built, but the wrong ucx_info was being invoked, making it appear that ROCm support was absent even when it was not.

2. UCC-ROCm depends on UCX. When UCC-ROCm was built as a component module on top of a ROCm-aware UCX, loading it would also pull in the plain UCX module as a listed dependency, which then shadowed the ROCm-aware UCX on PATH and in LD_LIBRARY_PATH. The result was a UCC-ROCm build backed at runtime by a UCX with no ROCm transport support.

Building UCX-ROCm and UCC-ROCm as fully independent installations avoids both conflicts. The trade-off is a larger on-disk footprint, but I think it is acceptable given that these modules are only loaded in ROCm-aware toolchains.

[Thyre]

LGTM

[Thyre]

@boegelbot please test @ jsc-zen3
EB_ARGS="--installpath /tmp/$USER/ebpr-4132 OpenMPI-4.1.4-GCC-11.3.0.eb OpenMPI-5.0.10-llvm-compilers-21.1.8.eb OpenMPI-5.0.3-intel-compilers-2024.2.0.eb OpenMPI-4.1.6-GCC-13.2.0.eb"

[boegelbot]

@Thyre: Request for testing this PR well received on jsczen3l1.int.jsc-zen3.fz-juelich.de

PR test command 'if [[ develop != 'develop' ]]; then EB_BRANCH=develop ./easybuild_develop.sh 2> /dev/null 1>&2; EB_PREFIX=/home/boegelbot/easybuild/develop source init_env_easybuild_develop.sh; fi; EB_PR=4132 EB_ARGS="--installpath /tmp/$USER/ebpr-4132 OpenMPI-4.1.4-GCC-11.3.0.eb OpenMPI-5.0.10-llvm-compilers-21.1.8.eb OpenMPI-5.0.3-intel-compilers-2024.2.0.eb OpenMPI-4.1.6-GCC-13.2.0.eb" EB_CONTAINER= EB_REPO=easybuild-easyblocks EB_BRANCH=develop /opt/software/slurm/bin/sbatch --job-name test_PR_4132 --ntasks=8 ~/boegelbot/eb_from_pr_upload_jsc-zen3.sh' executed!

• exit code: 0
• output:

Submitted batch job 10412

Test results coming soon (I hope)...

Details

- notification for comment with ID 4431346672 processed

Message to humans: this is just bookkeeping information for me,
it is of no use to you (unless you think I have a bug, which I don't).

[boegelbot]

Test report by @boegelbot

Overview of tested easyconfigs (in order)

• FAIL OpenMPI-4.1.4-GCC-11.3.0.eb (build issue)
  (partial log available at https://gist.github.com/boegelbot/70a2cb937c842a071c57213c45f4b4f3)

• SUCCESS OpenMPI-5.0.10-llvm-compilers-21.1.8.eb

• SUCCESS OpenMPI-5.0.3-intel-compilers-2024.2.0.eb

• SUCCESS OpenMPI-4.1.6-GCC-13.2.0.eb

Build succeeded for 3 out of 4 (total: 1 hour 32 mins 57 secs) (4 easyconfigs in total)
jsczen3c1.int.jsc-zen3.fz-juelich.de - Linux Rocky Linux 9.7, x86_64, AMD EPYC-Milan Processor (zen3), Python 3.9.25
See https://gist.github.com/boegelbot/a36bf09b09cfc6879502f83ce8b2679a for a full test report.

[Thyre]

FAIL: opal_path_nfs is not unsurprising, so that's certainly unrelated to the changes here.