Address known thread-safety issues in `cuda.core`

Open leofang opened this issue 2 months ago • 2 comments

In general, cuda.core has a few internal, module-level states that should be guarded by a lock. We haven't done that extensively.

Originally posted by @leofang in https://github.com/NVIDIA/cuda-python/pull/1364#discussion_r2624683212

Dec 16 '25 21:12 leofang

Example: https://github.com/NVIDIA/cuda-python/blob/027ba105fc33ac148c6d7343a9f71801e9f5ff72/cuda_core/cuda/core/experimental/_launcher.pyx#L20-L35

Dec 16 '25 21:12 leofang

Sebastian's comment here is relevant: https://github.com/NVIDIA/cuda-python/pull/1390#discussion_r2626380632

Dec 17 '25 18:12 leofang