Optimize moe align block size kernel#7794
Conversation
There was a problem hiding this comment.
Summary of Changes
Hello @ispobock, 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 introduces a substantial optimization to the moe_align_block_size_kernel, a critical component for Mixture-of-Experts (MoE) models. The core improvement involves replacing the existing prefix sum computation with a highly efficient Blelloch scan algorithm, which is designed to minimize global memory access and enhance parallel processing on the GPU. These changes result in significant performance improvements across a range of configurations, making the MoE alignment process faster and more efficient.
Highlights
- Kernel Optimization: The
moe_align_block_size_kernelhas been significantly optimized by replacing its prefix sum calculation with a more efficient Blelloch scan algorithm. This change aims to reduce global memory access and improve performance. - Shared Memory Management: The kernel's shared memory usage has been re-architected to support the new Blelloch scan, including dedicated buffers for counts, prefixes, and scan operations, leading to more efficient resource utilization.
- Data Type Simplification: The custom
AlignedArraystruct for vector types has been removed in favor of directly usingint4, simplifying the code and potentially leveraging native vector operations for better performance. - Performance Improvements: The changes yield notable performance gains, particularly for the 'SGL Fusion' and 'SGL' metrics across various
num_tokens,num_experts, andtopkconfigurations, as evidenced by the provided benchmarks.
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
-
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. ↩
BBuf
left a comment
There was a problem hiding this comment.
Great job. Please add end-to-end accuracy testing for DeepSeek V3.
|
And maybe block_scan_warp_scan implementation will be better for specify num_experts, we can try it next time |
|
Tested on |
|
Motivation
main:
This PR: