refactor: communication module #1162

yyihuang · 2025-06-20T01:25:07Z

📌 Description

Separate communication kernels from different sources.

🔍 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 @yyihuang, 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 refactors the communication module within FlashInfer to logically separate communication kernels based on their origin or intended framework, specifically for vLLM and NVIDIA's TensorRT-LLM. This change enhances modularity and ensures that only the relevant communication kernels are compiled and loaded for each specific use case, improving code organization and potentially reducing build times or memory footprint for specific deployments.

Highlights

Module Separation: The generic comm module has been split into two distinct modules: vllm_comm and trtllm_comm, each responsible for loading its respective set of communication kernels.
Function Renaming: The gen_comm_module and get_comm_module functions were renamed to gen_vllm_comm_module and get_vllm_comm_module respectively, now exclusively handling vLLM-related kernels.
New TRT-LLM Module: New functions gen_trtllm_comm_module and get_trtllm_comm_module were introduced to manage the compilation and loading of communication kernels specific to TensorRT-LLM.
Call Site Updates: All existing calls to communication functions (init_custom_ar, all_reduce, trtllm_lamport_initialize, etc.) have been updated to correctly reference either the vllm_comm or trtllm_comm module based on their functionality.

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.

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 refactors the communication module to separate vLLM and TRT-LLM kernels into distinct modules, improving modularity and maintainability. The changes are well-structured and the use of caching is appropriate. Suggestions are provided to improve code readability and consistency.

gemini-code-assist · 2025-06-20T01:27:24Z

flashinfer/comm.py

+def gen_vllm_comm_module() -> JitSpec:
    return gen_jit_spec(
-        "comm",
+        "vllm_comm",
        [
            jit_env.FLASHINFER_CSRC_DIR / "comm_pybind.cu",
            jit_env.FLASHINFER_CSRC_DIR / "custom_all_reduce.cu",
+        ],


Consider renaming gen_vllm_comm_module to gen_comm_module_vllm for better consistency with gen_trtllm_comm_module.

Suggested change

def gen_vllm_comm_module() -> JitSpec:

return gen_jit_spec(

"comm",

"vllm_comm",

[

jit_env.FLASHINFER_CSRC_DIR / "comm_pybind.cu",

jit_env.FLASHINFER_CSRC_DIR / "custom_all_reduce.cu",

],

def gen_comm_module_vllm() -> JitSpec:

return gen_jit_spec(

"vllm_comm",

[

jit_env.FLASHINFER_CSRC_DIR / "comm_pybind.cu",

jit_env.FLASHINFER_CSRC_DIR / "custom_all_reduce.cu",

],

)

gemini-code-assist · 2025-06-20T01:27:25Z

flashinfer/comm.py

+    return SimpleNamespace(
+        init_custom_ar=init_custom_ar,
+        dispose=dispose,
+        get_graph_buffer_ipc_meta=get_graph_buffer_ipc_meta,
+        register_buffer=register_buffer,
+        register_graph_buffers=register_graph_buffers,
+        meta_size=meta_size,
+        all_reduce=all_reduce,
+    )


Consider extracting the common arguments passed to SimpleNamespace into a dictionary or a separate function to reduce redundancy and improve readability.

Suggested change

return SimpleNamespace(

init_custom_ar=init_custom_ar,

dispose=dispose,

get_graph_buffer_ipc_meta=get_graph_buffer_ipc_meta,

register_buffer=register_buffer,

register_graph_buffers=register_graph_buffers,

meta_size=meta_size,

all_reduce=all_reduce,

)

common_args = {

"init_custom_ar": init_custom_ar,

"dispose": dispose,

"get_graph_buffer_ipc_meta": get_graph_buffer_ipc_meta,

"register_buffer": register_buffer,

"register_graph_buffers": register_graph_buffers,

"meta_size": meta_size,

"all_reduce": all_reduce,

}

return SimpleNamespace(**common_args)

gemini-code-assist · 2025-06-20T01:27:25Z

flashinfer/comm.py

+    return get_vllm_comm_module().init_custom_ar(
+        ipc_tensors, rank_data, rank, full_nvlink
+    )


Consider using named arguments for improved readability, e.g., return get_vllm_comm_module().init_custom_ar(ipc_tensors=ipc_tensors, rank_data=rank_data, rank=rank, full_nvlink=full_nvlink).

return get_vllm_comm_module().init_custom_ar(ipc_tensors=ipc_tensors, rank_data=rank_data, rank=rank, full_nvlink=full_nvlink)

gemini-code-assist · 2025-06-20T01:27:25Z

flashinfer/comm.py

        num_ctas: The number of CTAs to use for the all reduce.
        CTA upper bounds: 36. Generally, we can saturate the bandwidth even with small amount the SMs.
    """
-    get_comm_module().all_reduce(
+    get_vllm_comm_module().all_reduce(
        fa, inp, out, reg_buffer, reg_buffer_sz_bytes, num_ctas
    )


Consider using descriptive variable names for fa, inp, out, reg_buffer, reg_buffer_sz_bytes, and num_ctas to improve code clarity. For example, registered_allocator, input_tensor, output_tensor, registered_buffer, registered_buffer_size_bytes, and number_of_ctas.

get_vllm_comm_module().all_reduce( registered_allocator, input_tensor, output_tensor, registered_buffer, registered_buffer_size_bytes, num_ctas )

yzh119

We should further split the file into several python files under comm/ module. Let's do that in the next PR.

## 📌 Description Separate communication kernels from different sources. ## 🔍 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 - [x] I have installed `pre-commit` by running `pip install pre-commit` (or used your preferred method). - [x] I have installed the hooks with `pre-commit install`. - [x] 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](https://pre-commit.com/). ## 🧪 Tests - [x] Tests have been added or updated as needed. - [x] All tests are passing (`unittest`, etc.). ## Reviewer Notes

init

1cf9bd4

gemini-code-assist bot reviewed Jun 20, 2025

View reviewed changes

yyihuang requested a review from yzh119 June 20, 2025 01:25

gemini-code-assist bot reviewed Jun 20, 2025

View reviewed changes

yyihuang added 2 commits June 19, 2025 21:32

upd aot

b90bf92

fmt

302a6c4

yzh119 approved these changes Jun 20, 2025

View reviewed changes

yzh119 merged commit 15b3e65 into flashinfer-ai:main Jun 20, 2025
2 checks passed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

refactor: communication module #1162

refactor: communication module #1162

Uh oh!

yyihuang commented Jun 20, 2025

Uh oh!

gemini-code-assist bot left a comment

Uh oh!

gemini-code-assist bot left a comment

Uh oh!

gemini-code-assist bot Jun 20, 2025

Uh oh!

gemini-code-assist bot Jun 20, 2025

Uh oh!

gemini-code-assist bot Jun 20, 2025

Uh oh!

gemini-code-assist bot Jun 20, 2025

Uh oh!

yzh119 left a comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

-    return SimpleNamespace(
-        init_custom_ar=init_custom_ar,
-        dispose=dispose,
-        get_graph_buffer_ipc_meta=get_graph_buffer_ipc_meta,
-        register_buffer=register_buffer,
-        register_graph_buffers=register_graph_buffers,
-        meta_size=meta_size,
-        all_reduce=all_reduce,
-    )
+common_args = {
+    "init_custom_ar": init_custom_ar,
+    "dispose": dispose,
+    "get_graph_buffer_ipc_meta": get_graph_buffer_ipc_meta,
+    "register_buffer": register_buffer,
+    "register_graph_buffers": register_graph_buffers,
+    "meta_size": meta_size,
+    "all_reduce": all_reduce,
+}
+return SimpleNamespace(**common_args)

refactor: communication module #1162

refactor: communication module #1162

Uh oh!

Conversation

yyihuang commented Jun 20, 2025

📌 Description

🔍 Related Issues

🚀 Pull Request Checklist

✅ Pre-commit Checks

🧪 Tests

Reviewer Notes

Uh oh!

gemini-code-assist bot left a comment

Choose a reason for hiding this comment

Summary of Changes

Highlights

Footnotes

Uh oh!

gemini-code-assist bot left a comment

Choose a reason for hiding this comment

Code Review

Uh oh!

gemini-code-assist bot Jun 20, 2025

Choose a reason for hiding this comment

Uh oh!

gemini-code-assist bot Jun 20, 2025

Choose a reason for hiding this comment

Uh oh!

gemini-code-assist bot Jun 20, 2025

Choose a reason for hiding this comment

Uh oh!

gemini-code-assist bot Jun 20, 2025

Choose a reason for hiding this comment

Uh oh!

yzh119 left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants