Repository navigation
Optional CUDA builds for UCX and OpenMPI; add UCX 1.22.0 and OpenMPI 5.0.11 - #191
Merged
Merged
Conversation
UCX has a single install dir, so rather than duplicating the build block (as PR #168 did, where both blocks installed to the same place), the one configure task drops --with-cuda when ucx_cuda is in ansible_skip_tags. Entries without a cuda_root, such as 1.9.0, no longer fail on an undefined variable and simply build without CUDA. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01FLJRERiNvARR3e8ftXi5TG
Split the 4.1.5+ build into a gpu block tagged openmpi_gpu and a plain block, so --skip-tags openmpi_gpu builds only the plain variant. The gpu module task carries the same tag so no module is left pointing at a missing install. Verified with a throwaway playbook that a skip tag on an inner block inside a tagged include drops exactly those tasks. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01FLJRERiNvARR3e8ftXi5TG
Built against CUDA 12.9.1 like 1.19.0. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01FLJRERiNvARR3e8ftXi5TG
Checksum computed from the release tarball. Companion versions mirror the 5.0.8 entry (Singularity 4.2.2, CUDA 12.9.1, PMIx 5.0.9) with UCX moved to 1.22.0. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01FLJRERiNvARR3e8ftXi5TG
Contributor
There was a problem hiding this comment.
Copilot review overview
🔵 Needs a closer look
Shared OpenMPI build state can cause later GPU configuration to be skipped after a plain build.
Review effort: Lite
Findings: None
What changed in this PR
Adds optional CUDA build paths for UCX/OpenMPI and registers UCX 1.22.0 and OpenMPI 5.0.11.
Changes:
- Adds UCX CUDA skip-tag handling.
- Separates OpenMPI GPU and plain builds.
- Adds new versions, dependencies, and checksums.
| File | Description |
|---|---|
ansible/roles/compiled/tasks/common/ucx.yaml |
Supports CUDA and non-CUDA UCX builds. |
ansible/roles/compiled/tasks/common/openmpi.yaml |
Adds tagged GPU/plain OpenMPI build flows. |
ansible/roles/compiled/tasks/common/main.yml |
Registers UCX 1.22.0 and OpenMPI 5.0.11. |
💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.
… one Derive the Singularity directory from the group_vars `singularity` path instead of a per-entry singularity_version, so every build follows the cluster default (4.4.1 today). OpenMPI 5's configure no longer recognises --with-singularity, so the flag was a silent no-op for the 5.x entries; it is now passed only to 4.x builds. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01FLJRERiNvARR3e8ftXi5TG
kennedydane
force-pushed
the
update/optional-cuda-builds
branch
from
September 23, 2026 09:29
20b0b81 to
8654939
Compare
MikeCTZA
approved these changes
Sep 23, 2026
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Ports the remaining build work from #168 onto master, and adds the current releases.
--skip-tags ucx_cudabuilds without CUDA. UCX has one install dir, so instead of the duplicated build blocks in Dane/update/cromwell and other important stuff #168 (which both installed to the same place), the single configure task drops--with-cudawhen the tag is skipped. Entries with nocuda_root(1.9.0) no longer trip on an undefined variable.--skip-tags openmpi_gpubuilds only the plain variant. The 4.1.5+ build is split into a gpu block carrying the tag and a plain block. The gpu module task carries the same tag so no_gpu.luais left pointing at a missing install. Verified with a throwaway playbook that a skip tag on an inner block inside a tagged include removes exactly those tasks.singularity_versionis gone; the directory derives from the group_varssingularitypath. OpenMPI 5's configure does not recognise--with-singularity, so the flag was a silent no-op for 5.x entries and is now passed only to 4.x builds.Test plan
uv run yamllint .anduv run ansible-lint(production profile) passansible-playbook site.yaml --syntax-checkpassesansible-playbook site.yaml -t ucx,ucx1.22.0then-t openmpi,openmpi5.0.11--skip-tags ucx_cuda,openmpi_gpu🤖 Generated with Claude Code
https://claude.ai/code/session_01FLJRERiNvARR3e8ftXi5TG