Refactor hip kl

[amd-ruitang3]

Motivation

Technical Details

Test Plan

Test Result

Submission Checklist

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

[github-actions]

🏷️ CI Guide

Runs automatically on every PR:

• ✅ Pre-checks (submodule verification, code formatting)
• ✅ Aiter op tests (gfx942 + gfx950)
• ✅ Triton tests (only when aiter/ops/triton/** or related paths are changed)

Extended tests (opt-in via labels):

Label	Tests
ci:triton-355	Run Triton tests on MI355 in addition to MI325
ci:sglang	SGLang integration tests
ci:atom	ATOM benchmark (DeepSeek-R1 + GPT-OSS)
ci:vllm	vLLM benchmark
ci:all	All of the above

Add labels via the sidebar or gh pr edit 2624 --add-label <label>

[Copilot]

Pull request overview

Refactors parts of the HIP tensor utility layer by improving device correctness around memory ops and updating AiterTensor::empty_like() to preserve the source tensor’s strided layout when allocating storage. It also removes a Python-side HIP stream sync helper previously called during Torch→pybind tensor conversion.

Changes:

• Update AiterTensor::empty_like() to preserve shape/strides and allocate storage sized for the full positive-stride span.
• Add HipDeviceGuard around device-sensitive operations (hipMemset*, hipFree) and add a max-dims check in init_shape().
• Remove _sync_hip_stream() and its invocation from torch_to_aiter_pybind().

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 2 comments.

File	Description
csrc/include/aiter_tensor.h	Adjusts allocation semantics for empty_like(), adds device guarding for memset/free, and adds a dims bound check.
aiter/utility/dtypes.py	Removes implicit HIP stream synchronization from Torch→pybind tensor conversion.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.