Apply fixes for CUDA 13

[Aidyn-A]

Purpose

Fixes #24464 "Building vLLM with CUDA 13.0".

Test Plan

No plans. Just build.

Test Result

Test results must be the same as on CUDA 12.x.

---

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.

[gemini-code-assist]

Code Review

This pull request introduces fixes for building with CUDA 13 by abstracting CUB API differences into a new cub_helpers.h header. The changes are well-implemented and consistently applied across all relevant CUDA kernel files. This is a good approach to handle version-specific API changes. I have one suggestion for CMakeLists.txt to improve the robustness of how compiler flags are added.

[atalman]

lgtm

[ProExpertProg]

Looks good apart from 1 nit, also Gemini suggestion is good. Did you try to build vLLM from source locally (pip install -e .) with cuda 13.0?

[Aidyn-A]

Did you try to build vLLM from source locally (pip install -e .) with cuda 13.0?

Yes, I did. I ran VLLM_TARGET_DEVICE=cuda pip install --verbose --no-deps --no-build-isolation -e . to confirm it:

$ pip show vllm
Name: vllm
Version: 0.10.2rc2.dev89+gf94619747.d20250911.cu130
...

[pytorch-bot]

No ciflow labels are configured for this repo.
For information on how to enable CIFlow bot see this wiki

[Aidyn-A]

@huydhn, @atalman are you able to merge it?