[CUDA] Fix cuModuleUnload crash during interpreter shutdown

[guan404ming]

Related

#18614 ci error

Why

The CUDAModuleNode destructor was using CUDA_DRIVER_CALL and CUDA_CALL macros that call LOG(FATAL) (throw an exception) when CUDA operations fail.

During interpreter shutdown, the CUDA context can become invalid, causing CUDA_ERROR_ILLEGAL_ADDRESS when cuModuleUnload is called. Throwing exceptions in destructors is undefined behavior and causes crashes.

How

1. Removed the throwing macros from the destructor
2. Check cudaSetDevice return value and skip cleanup if it fails
3. Ignore errors from cuModuleUnload - during shutdown these are benign since the OS will reclaim resources anyway

[gemini-code-assist]

Summary of Changes

Hello @guan404ming, I'm Gemini Code Assist¹! I'm currently reviewing this pull request and will post my feedback shortly. In the meantime, here's a summary to help you and other reviewers quickly get up to speed!

This pull request addresses a critical stability issue where the CUDAModuleNode destructor could cause crashes during interpreter shutdown. The changes ensure graceful handling of CUDA context invalidation by preventing exceptions from being thrown in the destructor and intelligently skipping cleanup operations when the CUDA runtime is already in the process of unloading or encountering device errors.

Highlights

• Destructor Error Handling: Modified the CUDAModuleNode destructor to prevent crashes during interpreter shutdown by removing macros that throw exceptions on CUDA operation failures.
• Device Cleanup Logic: Introduced a check for cudaSetDevice return values, skipping module cleanup if device setting fails or indicates CUDA runtime unloading.
• Module Unload Error Suppression: Explicitly ignores errors from cuModuleUnload during shutdown, as these are considered benign since the operating system will reclaim resources.

🧠 New Feature in Public Preview: You can now enable Memory to help Gemini Code Assist learn from your team's feedback. This makes future code reviews more consistent and personalized to your project's style. Click here to enable Memory in your admin console.

Using Gemini Code Assist

The full guide for Gemini Code Assist can be found on our documentation page, here are some quick tips.

Invoking Gemini

You can request assistance from Gemini at any point by creating a comment using either /gemini <command> or @gemini-code-assist <command>. Below is a summary of the supported commands on the current page.

Feature	Command	Description
Code Review	/gemini review	Performs a code review for the current pull request in its current state.
Pull Request Summary	/gemini summary	Provides a summary of the current pull request in its current state.
Comment	@gemini-code-assist	Responds in comments when explicitly tagged, both in pull request comments and review comments.
Help	/gemini help	Displays a list of available commands.

Customization

To customize Gemini Code Assist for GitHub experience, repository maintainers can create a configuration file and/or provide a custom code review style guide (such as PEP-8 for Python) by creating and adding files to a .gemini/ folder in the base of the repository. Detailed instructions can be found here.

Limitations & Feedback

Gemini Code Assist may make mistakes. Please leave feedback on any instances where its feedback is incorrect or counter productive. You can react with 👍 and 👎 on @gemini-code-assist comments. If you're interested in giving your feedback about your experience with Gemini Code Assist for Github and other Google products, sign up here.

You can also get AI-powered code generation, chat, as well as code reviews directly in the IDE at no cost with the Gemini Code Assist IDE Extension.

Footnotes

1. Review the Privacy Notices, Generative AI Prohibited Use Policy, Terms of Service, and learn how to configure Gemini Code Assist in GitHub here. Gemini can make mistakes, so double check it and use code with caution. ↩

[gemini-code-assist]

Code Review

This pull request addresses a crash during interpreter shutdown caused by throwing exceptions from the CUDAModuleNode destructor. The fix correctly replaces the exception-throwing macros with manual error checking for CUDA API calls. By checking the return value of cudaSetDevice and ignoring errors from cuModuleUnload during shutdown, the change robustly prevents crashes while ensuring that resource cleanup is attempted. The implementation is clean, correct, and aligns with best practices for writing destructors that interact with external APIs that can fail during program termination.

[guan404ming]

cc @tlopex @mshr-h