update allreduce to match trtllm

[nvjullin]

📌 Description

Updated allreduce launch config logic to match trtllm.
On llama3 concurrency=128 tp2 gen-only phase, the kernel time improved from ~26.8us to ~9.8us.

🔍 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

[gemini-code-assist]

Summary of Changes

Hello @nvjullin, 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 focuses on optimizing the allreduce kernel launch configuration within FlashInfer, specifically by adopting a strategy similar to TensorRT-LLM. The primary goal is to enhance performance, which has been demonstrated by a significant reduction in kernel execution time for Llama3 models under specific conditions.

Highlights

• Allreduce Kernel Optimization: The allreduce kernel launch configuration logic in trtllm_allreduce_fusion.cuh has been updated to align with TensorRT-LLM's approach.
• Performance Improvement: This change yielded substantial performance gains, reducing kernel time from ~26.8us to ~9.8us for Llama3 with concurrency=128 and tp2 in the generation-only phase.
• Dynamic Resource Allocation: A new dynamic adjustment mechanism for threads_per_block and cluster_size was introduced, which considers the GPU's Streaming Multiprocessor (SM) count to optimize resource utilization during kernel launches.

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 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 updates the kernel launch configuration for allreduce_fusion to better utilize hardware resources by considering the SM count. This is a good performance optimization that, according to the description, yields significant improvements. My review includes one suggestion to refactor the launch parameter calculation logic for improved conciseness and maintainability by merging two consecutive loops.