fix: Remove CHECK_CONTIGUOUS for SFA/SFB

[yongwww]

📌 Description

Fix the regression in vLLM and SGLang with FI 0.4.0 in bmm_fp8

🔍 Related Issues

🚀 Pull Request Checklist

Thank you for contributing to FlashInfer! Before we review your pull request, please make sure the following items are complete.

✅ Pre-commit Checks

• I have installed pre-commit by running pip install pre-commit (or used your preferred method).
• I have installed the hooks with pre-commit install.
• I have run the hooks manually with pre-commit run --all-files and fixed any reported issues.

If you are unsure about how to set up pre-commit, see the pre-commit documentation.

🧪 Tests

• Tests have been added or updated as needed.
• All tests are passing (unittest, etc.).

Reviewer Notes

cc: @yzh119

Summary by CodeRabbit

• Backend Optimization
  • Streamlined tensor dispatch logic to support more flexible memory layout configurations, improving compatibility and processing efficiency for various tensor types while maintaining existing functionality.

[gemini-code-assist]

Summary of Changes

Hello @yongwww, 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 implements a fix for a regression affecting vLLM and SGLang when using FlashInfer 0.4.0 in bmm_fp8 operations. The core change involves removing redundant CHECK_CONTIGUOUS calls within the CutlassGemmGroupwiseScaledSM120 function, which were causing unexpected behavior. This ensures the correct and efficient execution of batch matrix multiplication with FP8 precision.

Highlights

• Redundant Contiguity Checks Removed: The CHECK_CONTIGUOUS calls for SFA and SFB have been removed from the CutlassGemmGroupwiseScaledSM120 function in csrc/gemm_groupwise_sm120.cu, as these checks were identified as redundant.
• Regression Fix: This change specifically addresses a regression encountered in vLLM and SGLang when utilizing FlashInfer 0.4.0 in bmm_fp8 operations, which was likely caused by the presence of these unnecessary contiguity checks.

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. ↩

[coderabbitai]

Walkthrough

The contiguity validation checks for SFA and SFB parameters are removed from the dispatch logic. The pipeline now proceeds without enforcing these constraints, deferring shape interpretation to downstream components or kernel layout specifications.

Changes

Cohort / File(s)	Summary
Contiguity Check Removal csrc/gemm_groupwise_sm120.cu	Removed SFA and SFB contiguity validation checks; dispatch pipeline no longer enforces contiguity constraints, delegating interpretation to downstream logic or kernel layouts. Other dispatch paths and parameter handling unchanged.

Estimated code review effort

🎯 2 (Simple) | ⏱️ ~12 minutes

Poem

🐰 No more checks to bind the way,
SFA and SFB now dance and play,
Contiguity fades to downstream care,
Flexibility flows through the air! ✨

Pre-merge checks and finishing touches

✅ Passed checks (3 passed)

Check name	Status	Explanation
Description Check	✅ Passed	The pull request description includes all required sections from the template: a Description section explaining the fix for the regression in vLLM and SGLang with FI 0.4.0, a Related Issues section (though with a placeholder rather than actual links), and the complete Pull Request Checklist with Pre-commit Checks, Tests, and Reviewer Notes sections. While the Related Issues section could be more detailed with specific issue links, the description is mostly complete with the key required information present and the overall structure matches the template.
Docstring Coverage	✅ Passed	Docstring coverage is 100.00% which is sufficient. The required threshold is 80.00%.
Title Check	✅ Passed	The pull request title "fix: Remove CHECK_CONTIGUOUS for SFA/SFB" directly and accurately summarizes the main change in the changeset. The raw summary confirms that the primary modification involves removing contiguity checks for SFA and SFB parameters in the csrc/gemm_groupwise_sm120.cu file. The title is specific and clear, identifying exactly what was removed (CHECK_CONTIGUOUS) and which parameters it affects (SFA/SFB), providing meaningful information without vagueness or unnecessary details.

✨ Finishing touches

• 📝 Generate docstrings

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Post copyable unit tests in a comment

---

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

[yzh119]

Do we have any assumptions on the layout of SFA or SFB?

[yongwww]

The layout of SFA and SFB: https://github.com/flashinfer-ai/flashinfer/blob/main/include/flashinfer/gemm/gemm_groupwise_sm120.cuh#L122-L123, looks like the contiguity is not required (https://github.com/NVIDIA/cutlass/blob/main/examples/87_blackwell_geforce_gemm_blockwise/87b_blackwell_geforce_fp8_bf16_gemm_groupwise.cu#L304-L305).

[gemini-code-assist]

Code Review

This pull request correctly removes redundant contiguity checks on scale factor tensors in the SM120 GEMM implementation. This fixes a regression where valid, non-contiguous tensors were being incorrectly rejected. The change is sound as the underlying CUTLASS kernel handles memory layout complexities for these tensors. For long-term stability, it would be beneficial to add a regression test case that specifically uses non-contiguous scale tensors to ensure this functionality remains correct in future changes.