[abhilash1910]

Description

Solves Issue #2954

_prepare_nvvm_options_impl raises CUDAError for ProgramOptions fields that
libNVVM cannot use, but its rejection list stopped at minimal. Twelve fields
that the NVRTC path emits were neither emitted nor rejected on NVVM, so setting
one of them for code_type="nvvm" had no effect and produced no error or
warning:
no_cache, fdevice_time_trace, device_float128, frandom_seed,
ofast_compile, pch, create_pch, use_pch, pch_dir, pch_verbose,
pch_messages, instantiate_templates_in_pch
libNVVM 12.9 and 13.4 return NVVM_ERROR_INVALID_OPTION for these flags (see
the issue thread), so none of them was ever doing anything on NVVM.

@leofang @Andy-Jost @rwgk for a look.

[copy-pr-bot]

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.

[coderabbitai]

📝 Summary

Summary by CodeRabbit

• Bug Fixes
  • NVVM compilation now reports a clear error when unsupported options are supplied, including cache, device timing, random seed, fast compilation, float128, and precompiled-header settings.
  • This makes these option conflicts visible at compilation time instead of allowing them to pass through without an explicit unsupported-option error.

NVVM option preparation now rejects additional unsupported settings and reports them through the existing CUDAError path. The unsupported-option test includes cases for these settings and checks that each option name appears in the error.

Changes: NVVM option rejection

Layer / File(s)	Summary
Unsupported-option validation cuda_core/cuda/core/_program.pyx, cuda_core/tests/test_program.py	Option preparation adds enabled no_cache, device_float128, and pch settings, plus non-None time-trace, random-seed, fast-compilation, and PCH configuration settings, to the unsupported-options list. Tests check that each option is rejected and named in the error.

Possibly related issues

• Issue 2954: The change adds explicit rejection for the options identified in the issue, addressing the reported silent-ignore behavior.
• Issue 3206: The change rejects ofast_compile; it does not implement the issue’s future pass-through behavior.

Suggested labels: bug, cuda.core

Suggested reviewers: andy-jost, rluo8

Priority: Normal

Change: Bug fix

Merge Risk: Minimal · up to 746fa

The change is ready to merge; unsupported NVVM options are rejected consistently without affecting omitted defaults.

Finishing Touches: 💡 1

🛠️ Fix failing CI checks: 💡

• Commit to this branch
• Create a new PR

• Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

_{Comment @coderabbitai help to get the list of available commands.}

[leofang]

/ok to test 746fa2c

[github-actions]

Doc Preview CI
🚀 View preview at https://nvidia.github.io/cuda-python/pr-preview/pr-3289/
https://nvidia.github.io/cuda-python/pr-preview/pr-3289/cuda-core/
https://nvidia.github.io/cuda-python/pr-preview/pr-3289/cuda-bindings/
https://nvidia.github.io/cuda-python/pr-preview/pr-3289/cuda-pathfinder/
Preview will be ready when the GitHub Pages deployment is complete.

[leofang]

Why didn't the pre-commit checks run... 🤔