Skip to content

Fix distributed dataloader so it pins memory to current GPU not GPU 0. #4196

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

Merged
apaszke merged 1 commit into from
Dec 19, 2017
Merged

Fix distributed dataloader so it pins memory to current GPU not GPU 0. #4196

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
10 changes: 7 additions & 3 deletions torch/utils/data/dataloader.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -54,7 +54,10 @@ def _worker_loop(dataset, index_queue, data_queue, collate_fn):
data_queue.put((idx, samples))


def _worker_manager_loop(in_queue, out_queue, done_event, pin_memory):
def _worker_manager_loop(in_queue, out_queue, done_event, pin_memory, device_id):
if pin_memory:
torch.cuda.set_device(device_id)

while True:
try:
r = in_queue.get()
Expand Down Expand Up @@ -176,7 +179,7 @@ def __init__(self, loader):
self.collate_fn = loader.collate_fn
self.batch_sampler = loader.batch_sampler
self.num_workers = loader.num_workers
self.pin_memory = loader.pin_memory
self.pin_memory = loader.pin_memory and torch.cuda.is_available()
self.timeout = loader.timeout
self.done_event = threading.Event()

Expand Down Expand Up @@ -206,7 +209,8 @@ def __init__(self, loader):
self.data_queue = queue.Queue()
self.worker_manager_thread = threading.Thread(
target=_worker_manager_loop,
args=(self.worker_result_queue, self.data_queue, self.done_event, self.pin_memory))
args=(self.worker_result_queue, self.data_queue, self.done_event, self.pin_memory,
torch.cuda.current_device()))
self.worker_manager_thread.daemon = True
self.worker_manager_thread.start()
else:
Expand Down