fix repeated unnecessary device check

[b8zhong]

Motivation

Technical Details

It seems to be spending a non trivial time in CUDA & it generally should be cached. We assume up to 8 devices.

Submission Checklist

• Look over the contributing guidelines at https://github.com/ROCm/ROCm/blob/develop/CONTRIBUTING.md#pull-requests.

[Copilot]

Pull Request Overview

Cache results of device/arch detection to avoid repeated and costly runtime queries.

• Add functools.lru_cache to get_arch and get_device
• Import lru_cache and configure cache size to 8

---

_{Tip: Customize your code reviews with copilot-instructions.md. Create the file or learn how to get started.}

[valarLip]

LGTM

[b8zhong]

Let me fix the lint and the copilot suggestion, I think it's acc correct

[b8zhong]

@valarLip Thanks for reviewing.

I fixed the lint

[b8zhong]

Hi, it doesn't look like the CI failure is related.