Support H-max to 256 in JAX/SPDA for Blackwell.

[mingxu1067]

No description provided.

[gemini-code-assist]

Summary of Changes

Hello @mingxu1067, 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 modifies the fused_attention_stablehlo.py file to include support for the Blackwell architecture. Specifically, it updates the logic to allow for a higher H-max value (256) when running on Blackwell GPUs, similar to Hopper GPUs. This change enables optimized attention calculations on the newer architecture.

Highlights

• H-max Support: This PR extends support for H-max to 256 in JAX/SPDA for Blackwell architecture.
• Compute Capability Check: The code now checks for CUDA compute capability '10.0' to identify Blackwell architecture.
• Conditional H-max Assignment: H_max is set to 256 if the architecture is either Hopper or Blackwell, otherwise it defaults to 128.

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 adds support for NVIDIA's Blackwell architecture by increasing the maximum head dimension (H_max) to 256 for fused attention, consistent with the Hopper architecture. The change is correct, but I've suggested a small refactoring to make the check more robust for future GPU architectures by using a minimum compute capability check. This improves maintainability by avoiding the need to explicitly list each new supported architecture.

[gemini-code-assist]

To make this check more future-proof for upcoming GPU architectures, it would be better to check for a minimum compute capability rather than listing specific architectures. Assuming all architectures from Hopper (9.0) onwards will support H_max = 256, you can simplify this logic using check_compute_capability. This also aligns with how other capability checks are performed in this file (e.g., for is_packed and is_mla).

Suggested change

	is_on_hopper = is_cuda_compute_capability_equal("9.0")
	H_max = 256 if is_on_hopper else 128
	is_on_blackwell = is_cuda_compute_capability_equal("10.0")
	H_max = 256 if (is_on_hopper or is_on_blackwell) else 128
	is_hopper_or_later = check_compute_capability("9.0")
	H_max = 256 if is_hopper_or_later else 128