---
title: '58646'
pageLayout: default
slug: thoughts/inference/vllm-prs/58646
permalink: https://aarnphm.xyz/thoughts/inference/vllm-prs/58646.md
generator:
  quartz: v4.6.0
  hostedProvider: Cloudflare
  baseUrl: aarnphm.xyz
full: https://aarnphm.xyz/llms-full.txt
---
# [#58646](https://github.com/aarnphm/aarnphm.github.io/issues/58646) \[Skills\] Update kernel-microbenchmark to include ROCm

[vllm-project/vllm#58646](https://github.com/vllm-project/vllm/pull/58646) · gau-nernst · MERGED · created 2026-09-25T01:44:02Z · merged 2026-09-28T06:18:39Z

## Purpose

Update `kernel-microbenchmark` skill to include ROCm

- There is no flashinfer cupti, hence use HIP graph replay with rotating buffers exceeding LLC size.
- This is better than Triton `do_bench` because Triton `do_bench` still has significant launch overheads.

## Test Plan

## Test Result

---

<details>
<summary> Essential Elements of an Effective PR Description Checklist </summary>

- [ ] The purpose of the PR, such as “Fix some issue (link existing issues this PR will resolve)”.
- [ ] The test plan, such as providing test command.
- [ ] The test results, such as pasting the results comparison before and after, or e2e results
- [ ] (Optional) The necessary documentation update, such as updating `supported_models.md` and `examples` for a new model.

</details>

---

## comments

**gau-nernst** (2026-09-25T02:41:19Z):

/ci run

**github-actions** (2026-09-25T02:42:01Z):

✅ Triggered [Buildkite CI #91139](https://buildkite.com/vllm/ci/builds/91139) for commit `8f27fc6c891d`.

