Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Add pin_memory support #326

Merged
merged 1 commit into from
Jan 17, 2025
Merged

Add pin_memory support #326

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
27 changes: 18 additions & 9 deletions src/spdl/dataloader/_pytorch_dataloader.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -128,6 +128,7 @@ def __init__(
sampler: "torch.utils.data.sampler.Sampler[K]",
fetch_fn: Callable[[K], U],
collate_fn: Callable[[list[U]], V],
transfer_fn: Callable[[V], V],
mp_ctx: mp.context.BaseContext,
num_workers: int,
timeout: float | None,
Expand All @@ -139,6 +140,7 @@ def __init__(
self._sampler = sampler
self._fetch_fn = fetch_fn
self._collate_fn = collate_fn
self._transfer_fn = transfer_fn
self._mp_ctx = mp_ctx
self._num_workers = num_workers
self._buffer_size = buffer_size
Expand All @@ -153,7 +155,7 @@ def _get_pipeline(self) -> tuple[ProcessPoolExecutor, Pipeline]:
executor = _get_executor(
self._shmem.name, self._collate_fn, self._num_workers, self._mp_ctx
)
pipeline = (
builder = (
PipelineBuilder()
.add_source(self._sampler)
.pipe(
Expand All @@ -162,9 +164,14 @@ def _get_pipeline(self) -> tuple[ProcessPoolExecutor, Pipeline]:
output_order=self._output_order,
concurrency=self._num_workers,
)
.add_sink(self._buffer_size)
.build(num_threads=1)
)
if self._transfer_fn:
builder.pipe(
self._transfer_fn,
output_order=self._output_order,
)

pipeline = builder.add_sink(self._buffer_size).build(num_threads=1)
return executor, pipeline

def __iter__(self) -> Iterator[V]:
Expand Down Expand Up @@ -231,7 +238,7 @@ def _resolve_sampler(
_collate_fn = collate_fn or default_collate
elif batch_size is not None:
_sampler = BatchSampler(
sampler or _get_sampler(dataset, shuffle, generator), # pyre-ignore: [6]
sampler or _get_sampler(dataset, shuffle, generator),
batch_size,
drop_last,
)
Expand Down Expand Up @@ -281,11 +288,8 @@ def get_pytorch_dataloader(
if worker_init_fn is not None:
raise ValueError("`worker_init_fn` is not supported.")

if pin_memory:
raise ValueError("`pin_memory` is not supported (yet).")

if pin_memory_device is not None:
raise ValueError("`pin_memory_device` is not supported (yet).")
raise ValueError("`pin_memory_device` is not supported.")

if persistent_workers:
raise ValueError("`persistent_workers` is not supported.")
Expand All @@ -309,6 +313,10 @@ def get_pytorch_dataloader(
generator,
)

from torch.utils.data._utils.pin_memory import pin_memory as pin_memory_fn

transfer_fn = pin_memory_fn if pin_memory else None

mp_ctx = (
multiprocessing_context
if isinstance(multiprocessing_context, mp.context.BaseContext)
Expand All @@ -321,8 +329,9 @@ def get_pytorch_dataloader(
dataset=dataset,
shmem=shmem,
sampler=_sampler,
fetch_fn=_fetch_fn, # pyre-ignore
fetch_fn=_fetch_fn,
collate_fn=_collate_fn,
transfer_fn=transfer_fn,
mp_ctx=mp_ctx,
num_workers=num_workers,
timeout=timeout,
Expand Down
Loading