feat(amdsmi): [ROCM-3941] Add manylinux wheel job to the amdsmi CI matrix by amd-juwillia · Pull Request #7913 · ROCm/rocm-systems

amd-juwillia · 2026-06-28T01:30:30Z

Motivation

Wire the SONAME-isolated wheel build into amd-smi's own per-distro CI matrix so every change is checked against a manylinux wheel build.

Technical Details

CI (.github/workflows/amdsmi-build.yml):

Add a manylinux-build job that runs tools/build_wheel.py with -DBUILD_PYTHON_WHEEL=ON alongside the existing build-and-test matrix.

Split 2 of 3 from #5726. Depends on #7822 (the wheel build it exercises) and is stacked on it. Both target develop, so until #7822 merges this PR's diff also shows #7822's changes; it collapses to 2 files once #7822 lands. Merge order: #7822 → this.

JIRA ID

Resolves ROCM-3941

Test Plan

Trigger the manylinux-build job on this branch and confirm a .whl is produced and import-smoke-tested.

Test Result

The manylinux job builds and installs the wheel on the ubuntu-22.04 manylinux_2_28 target.

Submission Checklist

Look over the contributing guidelines at https://github.com/ROCm/ROCm/blob/develop/CONTRIBUTING.md#pull-requests.

therock-pr-bot · 2026-06-28T01:31:15Z

✅ All Policy Checks Passed

Check	Status	Details
🌿 Branch Name	✅ Pass	—
📝 PR Title/Description	✅ Pass	—
⛔ Forbidden Files	✅ Pass	—
🧪 Unit Test	✅ Pass	—
🚫 Draft PR	🔜 To Be Enabled	—
🚩 Feature Flag	🔜 To Be Enabled	—
📊 Code Coverage	🔜 To Be Enabled	—

🎉 All policy checks passed!

📖 Need help? See the Policy FAQ for details on every check and how to fix failures.

therock-pr-bot · 2026-06-28T01:31:17Z

🚫 Please fix the failed policies before requesting reviews.

The following policy checks failed:

❌ Unit Test

The Not ready to Review label has been added to this PR.
Once all policies pass, the label will be removed automatically.

geomin12

seems this is same as #7914 and #7912 , same comments apply here

but amdsmi code should be reviewed by amdsmi folks

dmitrii-galantsev · 2026-07-22T00:15:55Z

Generated by Claude (AI assistant), reviewed by @dmitrii-galantsev.

Part of the ROCM-3941 chain: #7914 -> #7913 (this) -> #7912. Rebased onto the updated #7914; history rewritten and force-pushed, so please re-pull.

Fixes applied to the manylinux job:

runs-on: ubuntu-latest -> ubuntu-24.04 (style guide requires pinned runners).
Removed the redundant manual pip wheel invocation. The CMake python_wheel target is part of ALL (gated on BUILD_PYTHON_WHEEL), so the build already produces the wheel via its proper path (which runs disable_system_fallback.py). The job now just builds and asserts the .whl exists, instead of hand-rolling a second wheel that bypassed that step.
Split the single run: block into discrete steps (select interpreter, install prereqs, configure, build, verify) so a failure points at the right stage, per review feedback.
Dropped -DBUILD_TESTS=ON (the gate never runs the tests); consistent quoting.
CHANGELOG wording: "workflow" -> "job".

Update: rebased on the follow-up #7914 (which adds the RHEL CLI-shebang / interpreter-detection fix). No changes to this commit itself. CI green: Manylinux Gate, RHEL8, Debian10 and the wheel-build jobs all pass.

…-3.8 - main() was annotated with list[str] (PEP 585), which is only valid at runtime on Python 3.9+. On the 3.6-3.8 interpreters AMD SMI still supports it raised TypeError: 'type' object is not subscriptable at import. - The wheel build runs this tool under the system python, so an older default interpreter broke the build. - Use typing.List[str], which works on every supported version. Signed-off-by: Maisam Arif <Maisam.Arif@amd.com>

- seven near-identical steps each cat one result log, which is noisy in the workflow and in the Actions log - collapse them into a single step that prints each log inside a collapsible ::group:: block keyed by a readable label Signed-off-by: Maisam Arif <Maisam.Arif@amd.com>

- the packaging rework made the CLI import site-packages first and only fall back to share/amd_smi, reversing the priority from #3082 - on a multi-version ROCm host the CLI must load the amdsmi library it shipped with, not another version's copy earlier on sys.path - resolve share/amd_smi (ROCM_PATH or CLI-relative) before importing, and fall back to site-packages only when no share copy exists Signed-off-by: Maisam Arif <Maisam.Arif@amd.com>

- sparse-checkout only projects/amdsmi and the workflows dir instead of the whole monorepo, so the checkout stays cheap - keep each test-result dump as its own step so a single failure is directly attributable instead of buried in one grouped step - resolve the project and runner script by their fixed sparse-checkout paths rather than a find plus existence guard - drop the Uninstall step; the job runs in a --rm container that is discarded after the run - note why the artifact glob matches the 99999 local version Signed-off-by: Maisam Arif <Maisam.Arif@amd.com>

- the upload feeds no downstream job; document that it exists for manual reviewer download so the step is not mistaken for dead weight Signed-off-by: Maisam Arif <Maisam.Arif@amd.com>

- the wheel-location check accepted any of site-packages, dist-packages or /opt/rocm, so it passed even when a system copy shadowed the pip install and never proved the pip module wins - assert import amdsmi resolves under pip's reported install location and not the /opt/rocm system copy, so a coexisting system module fails the check instead of masking it - document the CLI's share -> pip -> system resolution order, distinct from a user script where the pip install must win Signed-off-by: Maisam Arif <Maisam.Arif@amd.com>

- continue-on-error let every distro leg report success even when the package failed to build, install or import - drop it so any failure on the self-hosted runner CI is visible; these runners are expected to be stable Signed-off-by: Maisam Arif <Maisam.Arif@amd.com>

- the check asserted the pip install wins, but the CI container and ROCm installs set PYTHONPATH=/opt/rocm/share/amd_smi, which always precedes site-packages, so import amdsmi resolved to the system copy and the assertion could never pass on RHEL/SLES/Ubuntu22 - run the check with PYTHONPATH cleared so it tests the real scripting scenario (a bare interpreter where pip must win), not the environment's path config Signed-off-by: Maisam Arif <Maisam.Arif@amd.com>

- Add a manylinux-build job to amdsmi-build.yml that builds the PyPI-ready wheel via cmake -DBUILD_PYTHON_WHEEL=ON inside the manylinux_2_28 container, alongside the existing build-and-test matrix. Signed-off-by: Justin Williams <Justin.Williams@amd.com>

- sparse-checkout only projects/amdsmi and the workflows dir so the wheel gate does not clone the whole monorepo - pip install the built wheel and import amdsmi, so the gate proves the wheel is installable and loads its bundled library rather than only checking that the .whl file exists Signed-off-by: Maisam Arif <Maisam.Arif@amd.com>

The manylinux gate is internal CI infrastructure; the user-facing wheel build (-DBUILD_PYTHON_WHEEL=ON) is already covered by the packaging rework entry above it. Signed-off-by: Maisam Arif <Maisam.Arif@amd.com>

Reverts the manylinux wheel CI job (#7913) and the Python packaging rework (#7914): flat loader, dist-packages install, and wheel build. Restores the pre-rework loader in amdsmi_wrapper.py / tools/generator.py, the pip-based DEBIAN/RPM install scriptlets, and setup.py.in; removes pin_shebang.cmake, tools/disable_system_fallback.py, and tests/python/test_abi_compat.py. Accounts for changes that landed on develop after the rework merged: - Keeps the libamd_smi.so.27 SONAME and its CHANGELOG entry from the 27.0.0 version bump. The restored loader hardcoded the SONAME, so the literal is bumped from .so.26 to .so.27 in both generator.py and the regenerated wrapper to match AMDSMI_LIB_VERSION_MAJOR. - Keeps the ras --cper --json no-entries fix (#8819) and its CHANGELOG entry; only the rework's own CHANGELOG entry is removed. Signed-off-by: Maisam Arif <Maisam.Arif@amd.com>

Reintroduces the manylinux wheel CI job (#7913) and the Python packaging rework (#7914): flat loader, dist-packages install, and wheel build. Combined here with the amd-smi CLI shebang fixes so the rework and its follow-up corrections land as one reviewable change. Signed-off-by: Maisam Arif <Maisam.Arif@amd.com>

…trix (#7913) - Add a manylinux-build job to amdsmi-build.yml that builds the PyPI-ready wheel via cmake -DBUILD_PYTHON_WHEEL=ON inside the manylinux_2_28 container, alongside the build-and-test matrix. - sparse-checkout only projects/amdsmi and the workflows dir so the gate does not clone the whole monorepo. - pip install the built wheel and import amdsmi so the gate proves the wheel is installable and loads its bundled library, not just that the .whl file exists. --------- Signed-off-by: Maisam Arif <Maisam.Arif@amd.com> Co-authored-by: Maisam Arif <Maisam.Arif@amd.com> Co-authored-by: Galantsev, Dmitrii <dmitrii.galantsev@amd.com>

Reverts the manylinux wheel CI job (#7913) and the Python packaging rework (#7914): flat loader, dist-packages install, and wheel build. Restores the pre-rework loader in amdsmi_wrapper.py / tools/generator.py, the pip-based DEBIAN/RPM install scriptlets, and setup.py.in; removes pin_shebang.cmake, tools/disable_system_fallback.py, and tests/python/test_abi_compat.py. Accounts for changes that landed on develop after the rework merged: - Keeps the libamd_smi.so.27 SONAME and its CHANGELOG entry from the 27.0.0 version bump. The restored loader hardcoded the SONAME, so the literal is bumped from .so.26 to .so.27 in both generator.py and the regenerated wrapper to match AMDSMI_LIB_VERSION_MAJOR. - Keeps the ras --cper --json no-entries fix (#8819) and its CHANGELOG entry; only the rework's own CHANGELOG entry is removed. Signed-off-by: Maisam Arif <Maisam.Arif@amd.com>

Relands the manylinux wheel CI job (#7913) and the Python packaging rework (#7914) after they were reverted in #9339, together with the follow-up amd-smi CLI shebang fixes so the rework and its corrections land as one change. Packaging rework (reintroduced): - Install the amdsmi module into the system python's site-packages via install(DIRECTORY) instead of pip + a .pth file; drop the pip plumbing and python3-pip/setuptools/wheel deps from the DEBIAN/RPM scriptlets. - Replace the ROCm-path-walking resolver in amdsmi_wrapper.py (and tools/generator.py, regenerated in lockstep) with a flat loader: AMDSMI_LIB_OVERRIDE, a bundled libamd_smi_python.so, then the system libamd_smi.so; a _MissingLibrary sentinel keeps import tolerant. - Add BUILD_PYTHON_WHEEL (default OFF): build a SONAME-isolated libamd_smi_python.so and a wheel that disables the loader's system fallback. - Stage the module under share/amd_smi for ROCM_PATH consumers and TheRock. - Add test_abi_compat.py and the manylinux wheel CI job. CLI shebang fixes: - Use a portable "#!/usr/bin/env python3" shebang for the amd-smi CLI so it runs on relocatable installs with no /usr/bin/python3 (TheRock containers) and on wheel runners where /usr/bin/python3 lacks the amdsmi module. - Stop pinning the CLI shebang to a fixed interpreter: remove pin_shebang.cmake and the AMDSMI_CLI_SHEBANG_INTERP derivation. The pinned absolute path became a hard RPM "Requires:" (e.g. /usr/libexec/platform-python3.6) that RHEL10 and SLES16 could not satisfy; the RPM now depends only on python(abi). - Resolve the env-form shebang in the install-verify harness instead of running "/usr/bin/env -c". Signed-off-by: Maisam Arif <Maisam.Arif@amd.com>

amd-juwillia requested a review from a team June 28, 2026 01:30

amd-juwillia requested a review from a team as a code owner June 28, 2026 01:30

github-actions Bot added documentation Improvements or additions to documentation project: amdsmi github actions Pull requests that update GitHub Actions code labels Jun 28, 2026

therock-pr-bot Bot added the Not ready to Review label Jun 28, 2026

systems-assistant Bot added the organization: ROCm label Jun 28, 2026

amd-juwillia force-pushed the users/juwillia/py-wheel-build branch 2 times, most recently from 32a3914 to a6f2850 Compare June 29, 2026 17:37

amd-juwillia mentioned this pull request Jun 29, 2026

feat(amdsmi): [ROCM-3941] Add SONAME conflict test and build harness #7912

Merged

1 task

marifamd force-pushed the users/juwillia/py-wheel-build branch 2 times, most recently from 663a8b1 to eac9e6b Compare July 6, 2026 16:35

marifamd requested a review from geomin12 July 7, 2026 19:54

geomin12 reviewed Jul 11, 2026

View reviewed changes

dmitrii-galantsev changed the base branch from develop to users/juwillia/py-loader-dist-packages July 20, 2026 20:53

dmitrii-galantsev force-pushed the users/juwillia/py-wheel-build branch from eac9e6b to dfa7fc8 Compare July 22, 2026 00:11

dmitrii-galantsev mentioned this pull request Jul 22, 2026

ci(amdsmi): [ROCM-3941] Python packaging rework #7914

Merged

1 task

dmitrii-galantsev force-pushed the users/juwillia/py-wheel-build branch from dfa7fc8 to 4718bd1 Compare July 22, 2026 01:01

dmitrii-galantsev force-pushed the users/juwillia/py-wheel-build branch from 4718bd1 to 2122443 Compare July 22, 2026 01:08

dmitrii-galantsev force-pushed the users/juwillia/py-wheel-build branch from 2122443 to cccc84f Compare July 22, 2026 03:06

dmitrii-galantsev force-pushed the users/juwillia/py-wheel-build branch from cccc84f to 3eafae4 Compare July 22, 2026 04:49

dmitrii-galantsev approved these changes Jul 22, 2026

View reviewed changes

marifamd added 7 commits July 24, 2026 09:02

ci(amdsmi): note why the package artifact is uploaded

af9857b

- the upload feeds no downstream job; document that it exists for manual reviewer download so the step is not mistaken for dead weight Signed-off-by: Maisam Arif <Maisam.Arif@amd.com>

marifamd force-pushed the users/juwillia/py-wheel-build branch from 6491aa1 to 9222b97 Compare July 24, 2026 14:04

github-actions Bot removed the MINOR ABI BREAKAGE label Jul 24, 2026

marifamd and others added 3 commits July 24, 2026 09:26

marifamd force-pushed the users/juwillia/py-wheel-build branch from 9222b97 to fa32a56 Compare July 24, 2026 14:31

geomin12 approved these changes Jul 25, 2026

View reviewed changes

Base automatically changed from users/juwillia/py-loader-dist-packages to develop July 27, 2026 00:08

marifamd approved these changes Jul 27, 2026

View reviewed changes

docs(amdsmi): drop the manylinux CI changelog entry

76c2b99

The manylinux gate is internal CI infrastructure; the user-facing wheel build (-DBUILD_PYTHON_WHEEL=ON) is already covered by the packaging rework entry above it. Signed-off-by: Maisam Arif <Maisam.Arif@amd.com>

marifamd merged commit 6679d53 into develop Jul 27, 2026
32 of 33 checks passed

marifamd deleted the users/juwillia/py-wheel-build branch July 27, 2026 00:38

This was referenced Jul 27, 2026

revert(amdsmi): revert the Python packaging rework (#7913, #7914) #9339

Merged

feat(amdsmi): reland packaging rework with CLI shebang fixes #9340

Merged

Navigation Menu

Repository navigation

feat(amdsmi): [ROCM-3941] Add manylinux wheel job to the amdsmi CI matrix - #7913

feat(amdsmi): [ROCM-3941] Add manylinux wheel job to the amdsmi CI matrix#7913
marifamd merged 13 commits into
developfrom
users/juwillia/py-wheel-build

amd-juwillia commented Jun 28, 2026 •

edited

Loading

Uh oh!

therock-pr-bot Bot commented Jun 28, 2026 •

edited

Loading

Uh oh!

therock-pr-bot Bot commented Jun 28, 2026

Uh oh!

geomin12 left a comment

Uh oh!

dmitrii-galantsev commented Jul 22, 2026 •

edited

Loading

Uh oh!

Uh oh!

Reviewers

Uh oh!

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

Conversation

amd-juwillia commented Jun 28, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Motivation

Technical Details

JIRA ID

Test Plan

Test Result

Submission Checklist

Uh oh!

therock-pr-bot Bot commented Jun 28, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

✅ All Policy Checks Passed

Uh oh!

therock-pr-bot Bot commented Jun 28, 2026

Uh oh!

geomin12 left a comment

Choose a reason for hiding this comment

Uh oh!

dmitrii-galantsev commented Jul 22, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Uh oh!

Reviewers

Uh oh!

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

amd-juwillia commented Jun 28, 2026 •

edited

Loading

therock-pr-bot Bot commented Jun 28, 2026 •

edited

Loading

dmitrii-galantsev commented Jul 22, 2026 •

edited

Loading