[Bugfix] Build moe_data for both sm100 and sm90

[mgoin]

Purpose

The fix in #19648 was disruptive for Blackwell's cutlass_moe_fp4 usage. This is because get_cutlass_moe_mm_data is used for both run_cutlass_moe_fp8 (sm90) and cutlass_moe_fp4 (sm100)

While trying to run the kernel, we see the following error:

NotImplementedError: No compiled get_cutlass_moe_mm_data: no cutlass_scaled_mm kernel for CUDA device capability: 100. Required capability: 90

Thanks to @dsikka and @pavanimajety for finding this issue

Test Plan

Run test on B200

pytest tests/kernels/moe/test_nvfp4_moe.py

Test Result

Before:

...
FAILED tests/kernels/moe/test_nvfp4_moe.py::test_cutlass_fp4_moe_no_graph[dtype1-8-256-224-1024-1536] - NotImplementedError: No compiled get_cutlass_moe_mm_data: no cutlass_scaled_mm kernel for CUDA device capability: 100. Required capability: 90
==== 180 failed in 68.79s (0:01:08) ====

After:

==== 180 passed in 75.16s (0:01:15) ====

[gemini-code-assist]

Summary of Changes

Hello @mgoin, 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 bug where the moe_data component, essential for Mixture-of-Experts (MoE) kernels, was not being correctly built for Blackwell (sm100) architectures. The changes ensure that moe_data is properly compiled and enabled for both Hopper (sm90) and Blackwell (sm100) GPUs, resolving a NotImplementedError that prevented cutlass_moe_fp4 usage on Blackwell.

Highlights

• Build System Enhancement: I've updated CMakeLists.txt to explicitly define the ENABLE_CUTLASS_MOE_SM100 compiler flag when building NVFP4 kernels. This ensures that the necessary flag for Blackwell (sm100) MoE support is present during compilation.
• Multi-Architecture Compilation for MoE Data: I've added a new build block in CMakeLists.txt specifically for moe_data.cu. This ensures that moe_data.cu is compiled for both sm90 (Hopper) and sm100 (Blackwell) architectures, provided the CUDA compiler version is 12.3 or greater. This is crucial for supporting MoE kernels across different GPU generations.
• Conditional Compilation Logic Correction: I've corrected a preprocessor directive in csrc/quantization/cutlass_w8a8/scaled_mm_entry.cu. The condition for enabling sm100 MoE kernels was incorrectly checking for ENABLE_SCALED_MM_SM90. It now correctly checks for ENABLE_CUTLASS_MOE_SM100, resolving a NotImplementedError when running on Blackwell.

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 in your pull request via creating an issue comment (i.e. comment on the pull request page) using either /gemini <command> or @gemini-code-assist <command>. Below is a summary of the supported commands.

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 issue 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 is currently in preview and 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 to provide feedback.

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. ↩

[github-actions]

👋 Hi! Thank you for contributing to the vLLM project.

💬 Join our developer Slack at https://slack.vllm.ai to discuss your PR in #pr-reviews, coordinate on features in #feat- channels, or join special interest groups in #sig- channels.

Just a reminder: PRs would not trigger full CI run by default. Instead, it would only run fastcheck CI which starts running only a small and essential subset of CI tests to quickly catch errors. You can run other CI tests on top of those by going to your fastcheck build on Buildkite UI (linked in the PR checks section) and unblock them. If you do not have permission to unblock, ping simon-mo or khluu to add you in our Buildkite org.

Once the PR is approved and ready to go, your PR reviewer(s) can run CI to test the changes comprehensively before merging.

To run CI, PR reviewers can either: Add ready label to the PR or enable auto-merge.

🚀

[gemini-code-assist]

Code Review

This pull request addresses a bug where the get_cutlass_moe_mm_data function was not compiled for the sm100 architecture, causing issues with cutlass_moe_fp4 usage. The changes include adding a compiler flag for sm100, ensuring moe_data.cu is built for relevant architectures, and correcting a conditional compilation check. I've identified a potentially incorrect condition that should be reviewed.

[pavanimajety]

LGTM

[dsikka]

thanks!