[TIR] Fix InjectPTXLDG32 segfaults and skip non-CUDA targets

[LudovicoYIN]

Motivation

InjectPTXLDG32 rewrites BufferStore when encountering if_then_else, but it only
initializes temporary buffers when an Allocate node exists. For functions without
Allocate, this leads to uninitialized buffers and a hard segfault during compilation.
In addition, the PTX-only pass can run on CPU/LLVM targets when tir.ptx_ldg32=1,
injecting PTX intrinsics that are invalid for non-CUDA codegen.

This PR ensures temporary buffers are created even when no Allocate exists, and
skips InjectPTXLDG32 on non-CUDA targets, preventing segfaults and invalid PTX
intrinsics on CPU.

Changes

• Ensure temp buffers are created when the rewrite path is taken without Allocate
• Insert allocations at the function level when needed
• Guard InjectPTXLDG32 so it only runs on CUDA targets
• Add tests for CUDA (insertion) and CPU (skip) behavior

Testing

test_tir_transform_inject_ptx_ldg32.py

Fixes

• #18612
• #18617
• #18599

[gemini-code-assist]

Summary of Changes

Hello @LudovicoYIN, 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 critical stability and correctness issues within the InjectPTXLDG32 compiler pass. It resolves segfaults that occurred when functions lacked explicit allocation nodes by ensuring necessary temporary buffers are always created. Additionally, it prevents the generation of invalid code by restricting the pass's execution to only CUDA compilation targets, thereby improving the robustness of the TVM compiler for various backends.

Highlights

• Buffer Allocation Fix: Temporary buffers required by InjectPTXLDG32 are now correctly initialized and allocated at the function level, even when the original function body does not contain an Allocate node, preventing segfaults.
• Target-Specific Execution: The InjectPTXLDG32 pass now includes a guard to ensure it only executes for CUDA targets, preventing the injection of invalid PTX intrinsics into non-CUDA (e.g., CPU/LLVM) code.
• New Test Cases: New tests have been added to verify both the correct insertion of allocations for CUDA targets and the proper skipping of the pass for non-CUDA targets.

🧠 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 segmentation fault in InjectPTXLDG32 that occurred when processing functions without any Allocate nodes. It also correctly restricts the pass to run only on CUDA targets. The changes are well-implemented, ensuring temporary buffers are created when needed, either within an existing Allocate scope or at the function level. The addition of new tests effectively validates both the bug fix and the new target-specific behavior. I have one suggestion to refactor a small piece of duplicated code to enhance maintainability.

[LudovicoYIN]

@spectrometerHBH @vinx13
This PR fixes segfaults in InjectPTXLDG32 and adds a CUDA-only guard to avoid
invalid PTX intrinsics on CPU targets. All CI is green.
Could you please take a look and approve if it looks good to you? Thanks!