cuda-python/cuda_core/docs/source/release/0.6.0-notes.rst at df811b9d1598f92331ad10f640bafedc8e38c8d1 · NVIDIA/cuda-python

.. currentmodule:: cuda.core

`cuda.core` 0.6.0 Release Notes

Added public access to default CUDA streams via module-level constants LEGACY_DEFAULT_STREAM and PER_THREAD_DEFAULT_STREAM

Users can now access default streams directly from the cuda.core namespace:
```
from cuda.core import LEGACY_DEFAULT_STREAM, PER_THREAD_DEFAULT_STREAM

# Use legacy default stream (synchronizes with all blocking streams)
LEGACY_DEFAULT_STREAM.sync()

# Use per-thread default stream (non-blocking, thread-local)
PER_THREAD_DEFAULT_STREAM.sync()
```
The legacy default stream synchronizes with all blocking streams in the same CUDA context, ensuring strict ordering but potentially limiting concurrency. The per-thread default stream is local to the calling thread and does not synchronize with other streams, enabling concurrent execution in multi-threaded applications.

This replaces the previous undocumented workaround of using Stream.from_handle(0) to access the legacy default stream.
Added :func:`~cuda.core.utils.make_aligned_dtype` utility for creating structured NumPy dtypes with GPU-compatible alignment. Field offsets and the total itemsize are recomputed so that each field is naturally aligned and the structure size is a multiple of the largest member alignment. An explicit alignment can be requested and is stored in the dtype's metadata under the key "__cuda_alignment__". (Resolves :issue:`734`.)

None.