[rwgk]

Summary

Exercise the CUDA 12 and CUDA 13 wheels built from main, with explicit component and dependency-source coverage in the PR checks view.

• Name matrix checks for the enabled suites and CUDA major, distinguishing PR bindings from published-floor bindings and local from wheel-installed toolkits. Metapackage suites display meta12 and meta13.
• Reject PR/release filters that remove required local-wheel coverage. Retain mandatory Python 3.15/3.15t coverage and current main's JUnit reports, log folding, retries, and Windows ARM64 behavior.
• Validate metapackage installability against the selected bindings and published dependencies, without requiring an unrelated Core build for a bindings release.
• Route Pixi source checks, coverage, and documentation to the intended package root. Preserve the current Lychee implementation and rendered-link checks.
• Keep the cuDSS test dependencies below version 1, matching Pathfinder's currently supported library ABI.

Shared GPU jobs keep runner usage bounded; their names and summaries expose which suites actually run.

CUDA 12 NVML test maintenance moves here from #3173, alongside CI activation. The four files retain their previous contents: device visibility, PCI-information and Orin/Thor naming accommodations (#1590, #2065, #2437, #2788); serialized initialization and the driver 13.4 refcount skip (#2229, #2512); removal of the unstable energy-timing assertion (#2644); and the field-enum spelling correction (#2560). Driver probing occurs during test execution. These changes affect tests and their helper, not the installed runtime.

Stack / Review Boundary

Layer 5 of 8. Previous: #3175. Next: #3177. All layers and integrated validation.

After #3173, #3174, #3175, and #3176 form one merge group, with separate review scopes. The cumulative #3176 head is the full-CI and landing target; #3174 and #3175 are incomplete intermediate states and should not be merged individually. Release routing (#3177), maintenance policy (#3178), and optional runtime fixes (#3182) remain subsequent layers.

The stack is based on main dfad7e3658a9. Validation of the updated merge-group head is recorded on #3176.

Release Boundary

Future CUDA 13 releases tagged from this state retain the existing release flow. The transitional backport_branch field keeps the legacy release-note and backport consumers working; toolkit selection uses the package-root registry. Releasing CUDA 12 from main is enabled separately by #3177.

Starting at this layer, release-documentation reruns and dry-runs against pre-registry tags require compatible historical tooling: the new docs workflow reads package configuration from the release tag.

Suggested Review Order

1. ci/tools/run-tests
2. .github/workflows/test-wheel-linux.yml
3. .github/workflows/test-wheel-windows.yml
4. .github/workflows/build-docs.yml

Validation

At this head, 419 CI-tool tests and 55 subtests pass, and full local pre-commit run --all-files --show-diff-on-failure passes. The existing release-note checker succeeds for both CUDA 13 packages without a backport-branch override; the unchanged backport lookups resolve 12.9.x. Tag-planning checks select only the matching bindings and metapackage major.

Full hosted CI passes at the cumulative #3176 head 19865b05d5b0, including all 28 wheel builds and the CUDA 12/CUDA 13 GPU matrix. One RTX 2080 runner setup timeout passed on targeted retry. The earlier attempt exposed the same dependency incompatibility in CUDA 12 and CUDA 13 jobs: cuDSS 1 supplies new library names unsupported by the current loader. Both test groups now require cuDSS <1. A strict local cuDSS loading check passes with version 0.8.0.10, which has compatible Linux x86_64, Linux ARM64, and Windows wheels.

Earlier integrated-stack validation passed full local CUDA 12.9 and CUDA 13.4 QA. Those earlier hosted attempts are superseded by the successful run above; their history is recorded on #3178.

Checklist

• Relevant existing and focused regression tests cover the integration behavior.
• This description identifies the layer's scope and validation boundary.

[copy-pr-bot]

Auto-sync is disabled for draft pull requests in this repository. Workflows must be run manually.

Contributors can view more details about this message here.

[rwgk]

/ok to test 9b8ac76

[rwgk]

/ok to test 19865b0

[github-actions]

Doc Preview CI
🚀 View preview at https://nvidia.github.io/cuda-python/pr-preview/pr-3176/
https://nvidia.github.io/cuda-python/pr-preview/pr-3176/cuda-core/
https://nvidia.github.io/cuda-python/pr-preview/pr-3176/cuda-bindings/
https://nvidia.github.io/cuda-python/pr-preview/pr-3176/cuda-pathfinder/
Preview will be ready when the GitHub Pages deployment is complete.