Language

unilab.ipc.replay_pipelines.transfer¶

Device transfer backends for replay pipelines.

class unilab.ipc.replay_pipelines.transfer.ReplayTransferBackend[source]¶

Bases: Protocol

Device-specific host-to-device transfer backend.

device: device¶

device_family: str¶

h2d_submitter: str¶

host_memory_kind: str¶

host_pinned: bool¶

direct_pinned_shared: bool¶

supports_async_submit: bool¶

supports_timing_events: bool¶

register_host_slots(slots)[source]¶

Parameters:: slots (list[Tensor])
Return type:: None

allocate_device_slots(*, count, shape, dtype)[source]¶

Parameters:

count (int)
shape (tuple[int, int])
dtype (dtype)

Return type:

list[Tensor]

submit_h2d(*, slot, dst, src, metadata, trace_recorder, trace_cuda_events, h2d_bytes, pack_layout, pack_executor)[source]¶

Parameters:

slot (int)
dst (Tensor)
src (Tensor)
metadata (ReplayTickMetadata | None)
trace_cuda_events (bool)
h2d_bytes (int)
pack_layout (str)
pack_executor (str)

Return type:

float

clear_ready(slot)[source]¶

Parameters:: slot (int)
Return type:: None

ready_query(slot)[source]¶

Parameters:: slot (int)
Return type:: bool

synchronize_ready(slot)[source]¶

Parameters:: slot (int)
Return type:: None

wait_current_stream_for_ready(slot)[source]¶

Parameters:: slot (int)
Return type:: None

close()[source]¶

Return type:: None

__init__(*args, **kwargs)¶

class unilab.ipc.replay_pipelines.transfer.XpuReplayTransferBackend[source]¶

Bases: object

XPU stream/event backend for packed replay batch transfer.

Parameters:

device (device)
ring_depth (int)

device_family = 'xpu'¶

h2d_submitter = 'torch_xpu_copy_stream'¶

host_memory_kind = 'pageable_shared'¶

host_pinned = False¶

direct_pinned_shared = False¶

supports_async_submit = True¶

supports_timing_events = False¶

__init__(*, device, ring_depth)[source]¶

Parameters:

device (device)
ring_depth (int)

register_host_slots(slots)[source]¶

Parameters:: slots (list[Tensor])
Return type:: None

allocate_device_slots(*, count, shape, dtype)[source]¶

Parameters:

count (int)
shape (tuple[int, int])
dtype (dtype)

Return type:

list[Tensor]

submit_h2d(*, slot, dst, src, metadata, trace_recorder, trace_cuda_events, h2d_bytes, pack_layout, pack_executor)[source]¶

Parameters:

slot (int)
dst (Tensor)
src (Tensor)
metadata (ReplayTickMetadata | None)
trace_cuda_events (bool)
h2d_bytes (int)
pack_layout (str)
pack_executor (str)

Return type:

float

clear_ready(slot)[source]¶

Parameters:: slot (int)
Return type:: None

ready_query(slot)[source]¶

Parameters:: slot (int)
Return type:: bool

synchronize_ready(slot)[source]¶

Parameters:: slot (int)
Return type:: None

wait_current_stream_for_ready(slot)[source]¶

Parameters:: slot (int)
Return type:: None

close()[source]¶

Return type:: None

unilab.ipc.replay_pipelines.transfer.build_replay_transfer_backend(*, device, ring_depth)[source]¶

Build the transfer backend for a learner device.

Parameters:

device (device)
ring_depth (int)

Return type:

ReplayTransferBackend

Modules

`base`	Replay transfer backend contract.
`cuda_like`	CUDA/ROCm replay transfer backend.
`factory`	Replay transfer backend factory.
`torch_copy`	Portable torch-copy replay transfer backend.
`xpu`	Intel XPU replay transfer backend.