[ROCm][perf] Shuffle KV cache to use paged_attention_common

[samutamm]

Purpose

For Qwen/Qwen3-235B-A22B-Instruct-2507-FP8 model, currently VLLM_ROCM_SHUFFLE_KV_CACHE_LAYOUT=1 performs worse on small concurrencies, compared to VLLM_ROCM_SHUFFLE_KV_CACHE_LAYOUT=0. This PR fixes the issue using paged_attention_common from aiter (see ROCm/aiter#1821).

Test Plan

For input and output lengths of 1k and 8k and concurrencies from 8, 18, 32, 64, 128, compare current main branch with and without VLLM_ROCM_SHUFFLE_KV_CACHE_LAYOUT (_vllm_main_shuffle1 and _vllm_main_shuffle0, respectively) to changes of this PR (_pr_shuffle1).

Also verified on MI355.
Also verified for Qwen/Qwen3-235B-A22B-Instruct-2507.

Test Result

For input length 8k and output length 1k (green lines), the changes of this PR (_pr_shuffle1, the solid line) outperform main branch, with or without shuffle kv cache.
For input length 1k and output length 8k (orange lines), the changes of this PR (_pr_shuffle1, the solid line) outperform main branch, with or without shuffle kv cache.
For input length 1k and output length 1k (blue lines), the changes of this PR (_pr_shuffle1, the solid line) are very close to main branch. This might require further adjustment in aiter paged_attention_common.

---

Essential Elements of an Effective PR Description Checklist

• 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.
• (Optional) Release notes update. If your change is user facing, please update the release notes draft in the Google Doc.

[github-actions]

👋 Hi! Thank you for contributing to the vLLM project.

💬 Join our developer Slack at https://slack.vllm.ai to discuss your PR in #pr-reviews, coordinate on features in #feat- channels, or join special interest groups in #sig- channels.

Just a reminder: PRs would not trigger full CI run by default. Instead, it would only run fastcheck CI which starts running only a small and essential subset of CI tests to quickly catch errors.

You ask your reviewers to trigger select CI tests on top of fastcheck CI.

Once the PR is approved and ready to go, your PR reviewer(s) can run CI to test the changes comprehensively before merging.

To run CI, PR reviewers can either: Add ready label to the PR or enable auto-merge.

If you have any questions, please reach out to us on Slack at https://slack.vllm.ai.

🚀

[gemini-code-assist]

Code Review

The pull request updates the AITER branch in the Dockerfile and integrates aiter.paged_attention_common for shuffle KV cache handling in rocm_aiter_fa.py. This change aims to fix performance issues with small concurrencies for specific Qwen models. The introduction of temporary tensors (tmp_out, exp_sums, max_logits) and new scaling parameters (K_QScale_hip, V_QScale_hip, K_QScale_asm, V_QScale_asm) to the paged_attention_common function is a significant update to the attention mechanism. I've identified a couple of issues related to variable redefinition and unreachable code that should be addressed.

[tjtanaa]

@gshtras what do you think about this aiter commit?

@samutamm before upgrading aiter commits, we also need to determine if all other popular models and settings does not have issue. So we will need time for reviewing this PR

E.g.
Llama4 tp, ep
Deepseek tp ep, MTP
Qwen3 moe models tp ep,
Gptoss

[tjtanaa]

So we would usually split this PR into

Upgrade Aiter version first, then only introduce new Kernel.

[tjtanaa]

We will keep this PR in check, once we have AITER commit version upgraded and if it contains the kernel, then we will continue with this PR.

[mergify]

Hi @samutamm, the pre-commit checks have failed. Please run:

uv pip install pre-commit
pre-commit install
pre-commit run --all-files

Then, commit the changes and push to your branch.

For future commits, pre-commit will run automatically on changed files before each commit.

Tip

Is mypy or markdownlint failing?

mypy and markdownlint are run differently in CI. If the failure is related to either of these checks, please use the following commands to run them locally:

# For mypy (substitute "3.10" with the failing version if needed)
pre-commit run --hook-stage manual mypy-3.10
# For markdownlint
pre-commit run --hook-stage manual markdownlint

[mergify]

Hi @samutamm, the pre-commit checks have failed. Please run:

uv pip install pre-commit
pre-commit install
pre-commit run --all-files

Then, commit the changes and push to your branch.

For future commits, pre-commit will run automatically on changed files before each commit.

Tip

Is mypy or markdownlint failing?

mypy and markdownlint are run differently in CI. If the failure is related to either of these checks, please use the following commands to run them locally:

# For mypy (substitute "3.10" with the failing version if needed)
pre-commit run --hook-stage manual mypy-3.10
# For markdownlint
pre-commit run --hook-stage manual markdownlint

[mergify]

Hi @samutamm, the pre-commit checks have failed. Please run:

uv pip install pre-commit
pre-commit install
pre-commit run --all-files

Then, commit the changes and push to your branch.

For future commits, pre-commit will run automatically on changed files before each commit.

Tip

Is mypy or markdownlint failing?

mypy and markdownlint are run differently in CI. If the failure is related to either of these checks, please use the following commands to run them locally:

# For mypy (substitute "3.10" with the failing version if needed)
pre-commit run --hook-stage manual mypy-3.10
# For markdownlint
pre-commit run --hook-stage manual markdownlint

[mergify]

Hi @samutamm, the pre-commit checks have failed. Please run:

uv pip install pre-commit
pre-commit install
pre-commit run --all-files

Then, commit the changes and push to your branch.

For future commits, pre-commit will run automatically on changed files before each commit.

Tip

Is mypy or markdownlint failing?

mypy and markdownlint are run differently in CI. If the failure is related to either of these checks, please use the following commands to run them locally:

# For mypy (substitute "3.10" with the failing version if needed)
pre-commit run --hook-stage manual mypy-3.10
# For markdownlint
pre-commit run --hook-stage manual markdownlint