You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
[DOC]: Add missing CUDA 12 NVML API reference #3517
Reported by GPT-6.1 Sol High, at the user's request, following a review of generated release documentation.
Problem
The CUDA 12.9.11 documentation advertises NVML additions and breaking changes in its release notes, but does not include an NVML API reference. Users cannot navigate to or search for reference documentation for the affected NVML functions, classes, and members.
The 12.9.11 release notes describe additions such as device_get_gpu_uptime_v1, changes to NvlinkFirmwareInfo.firmware_version, and the NvmlTimeoutError/NvmlMemoryError compatibility aliases.
The CUDA 12 API toctree omits module/nvml, and that source tree has no docs/source/module/nvml.rst entry point.
The generated release tree has no module/nvml.html or NVML detail pages. Its versioned objects.inv contains zerocuda.bindings.nvml entries.
In comparison, the CUDA 13 API toctree includes module/nvml and has an NVML reference source.
Reproduce
Build component=cuda-bindings release documentation for v12.9.11, or inspect the deployment snapshot above.
Open docs/cuda-bindings/12.9.11/api.html and inspect the module list.
Look for module/nvml.html and generated NVML detail pages; none are present.
Search the built documentation for a newly documented NVML function. The API reference is absent even though the release notes discuss it.
Expected behavior and suggested scope
Add an NVML API reference to the CUDA 12 documentation, including navigation and generated detail pages for the NVML surface actually supported by that package. Use the CUDA 13 documentation as a counterpart to assess, without copying unsupported CUDA 13-only APIs into CUDA 12 docs.
Validation should confirm that representative functions and the renamed exception classes have reference pages, search results, and versioned inventory entries. Check for missing autosummary imports when rebuilding.
Related: #3383 tracks documentation build warnings, including stale CUDA 13 NVML autosummary entries. This report concerns the separate omission of the CUDA 12 NVML reference altogether.
No source changes are included in this report. No GPU or driver is required to inspect the generated output.
Activity