Stream
An execution stream: an ordered queue of device work. Operations submitted to it run in order, asynchronously of the host. A created stream pipelines submission and overlaps independent work; the per-device default stream dispatches on the calling thread instead.
device (property)
The device this stream runs on.
__init__
__init__(self, /, *args, **kwargs)
Initialize self. See help(type(self)) for accurate signature.
create
createcreate(device: clika_runtime._core.Device) -> clika_runtime._core.Stream
create(device: clika_runtime._core.Device) -> clika_runtime._core.Stream
Allocate a fresh non-default stream on device (for overlapping work).
default_stream
default_streamdefault_stream(device: clika_runtime._core.Device) -> clika_runtime._core.Stream
default_stream(device: clika_runtime._core.Device) -> clika_runtime._core.Stream
The per-device default stream; always valid.
from_native_handle
from_native_handlefrom_native_handle(device: clika_runtime._core.Device, native_handle: int) -> clika_runtime._core.Stream
from_native_handle(device: clika_runtime._core.Device, native_handle: int) -> clika_runtime._core.Stream
Adopt a native stream you own (the integer handle of a cudaStream_t, such as torch.cuda.Stream().cuda_stream) as a stream on device. Work placed on it runs on that native stream, and the runtime never destroys it. Adopting the same native stream on the same device again returns the same stream. An adopted stream is not released before the process ends. The runtime may synchronize it, query it and reclaim memory on it at any point until then, so keep the native stream alive until the process ends. A handle of 0 raises InvalidArgumentError. It is CUDA's legacy default stream (torch's current stream until one is set), which the runtime cannot order against its own streams, so pass a stream you created. A device whose backend cannot adopt a native stream raises UnsupportedError.
is_default
is_defaultis_default(self) -> bool
is_default(self) -> bool
True if this is the device's default stream.
query_idle
query_idlequery_idle(self) -> bool
query_idle(self) -> bool
True if the stream currently has no work in flight (a hint, not a barrier).
synchronize
synchronizesynchronize(self) -> None
synchronize(self) -> None
Block until all work submitted on this stream has completed.