[Rework] Add SwapAB Optimization for triton fused_moe_kernel on SM90.

[Insideyyy]

Motivation

Rework of reverted pr #15712, with AMD CI failures fixed.

Modifications

Accuracy Tests

Benchmarking and Profiling

Checklist

• Format your code according to the Format code with pre-commit.
• Add unit tests according to the Run and add unit tests.
• Update documentation according to Write documentations.
• Provide accuracy and speed benchmark results according to Test the accuracy and Benchmark the speed.
• Follow the SGLang code style guidance.

Review Process

1. Ping Merge Oncalls to start the PR flow. See the PR Merge Process.
2. Get approvals from CODEOWNERS and other reviewers.
3. Trigger CI tests with comments (/tag-run-ci-label, /rerun-failed-ci, /tag-and-rerun-ci) or contact authorized users to do so.
4. After green CI and required approvals, ask Merge Oncalls to merge.

[gemini-code-assist]

Summary of Changes

Hello @Insideyyy, 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 re-introduces a performance optimization, dubbed 'SwapAB', for the fused Mixture-of-Experts (MoE) kernel within the SGLang runtime. The core change involves conditionally transposing input matrices during the kernel's computation to improve efficiency on specific NVIDIA SM90 GPUs, particularly the H20. This re-implementation addresses and resolves the AMD CI failures that led to the previous reversion of this optimization, ensuring stability while delivering targeted performance enhancements.

Highlights

• Reintroduced SwapAB Optimization: This pull request re-implements the 'SwapAB' optimization for the Triton fused Mixture-of-Experts (MoE) kernel, which was previously reverted due to AMD CI failures. The current implementation includes fixes to prevent these failures.
• Conditional Optimization Activation: A new utility function, should_enable_swap_ab, has been added to conditionally activate the SwapAB optimization. It is currently enabled only for specific NVIDIA SM90 GPUs (specifically H20, excluding H200) and under certain BLOCK_SIZE_M (< 64) and BLOCK_SIZE_N (>= 64) conditions, primarily when using FP8 weight-activation quantization.
• Kernel Logic Modification: The fused_moe_kernel has been updated to accept a swap_ab flag. When enabled, it transposes the input matrices (A and B) before the dot product and transposes the final accumulator to maintain correct output dimensions, optimizing performance for specific hardware configurations.

🧠 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 introduces the SwapAB optimization for the Triton fused_moe_kernel on SM90 GPUs, specifically targeting H20 devices for now. The changes are well-contained and add the necessary logic to detect the target hardware and conditionally apply the optimization within the kernel. The implementation appears correct. I have a couple of suggestions to enhance code readability and reduce duplication, which will improve long-term maintainability.

[gemini-code-assist]

The logic to determine if the optimization should be enabled can be made more direct and readable by combining the checks into a single return statement and using is not None for explicit None checking. This removes the intermediate is_h20_device variable and makes the function's intent clearer at a glance.

Suggested change

	device_name = get_device_name()
	is_h20_device = device_name and "H20" in device_name and "H200" not in device_name
	return (
	is_h20_device
	and is_sm90_supported()
	and BLOCK_SIZE_M < 64
	and BLOCK_SIZE_N >= 64
	)
	device_name = get_device_name()
	return (
	device_name is not None
	and "H20" in device_name
	and "H200" not in device_name
	and is_sm90_supported()
	and BLOCK_SIZE_M < 64
	and BLOCK_SIZE_N >= 64
	)

[ClawSeven]

/tag-and-rerun-ci

[Fridge003]

/rerun-failed-ci