Skip to content

for asynchronous ring communication #958

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

Closed
dingirsu wants to merge 1 commit into from
Closed

for asynchronous ring communication #958

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
7 changes: 6 additions & 1 deletion lightx2v/common/ops/attn/utils/ring_comm.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -15,6 +15,8 @@ def __init__(self, process_group: dist.ProcessGroup = None):
self.send_rank = (self.rank + 1) % self.world_size
self.recv_rank = (self.rank - 1) % self.world_size

self.comm_stream = torch.cuda.Stream()

if process_group is not None:
self.send_rank = dist.get_global_rank(self._process_group, self.send_rank)
self.recv_rank = dist.get_global_rank(self._process_group, self.recv_rank)
Expand All @@ -35,7 +37,8 @@ def send_recv(self, to_send: torch.Tensor, recv_tensor: Optional[torch.Tensor] =
def commit(self):
if self._reqs is not None:
raise RuntimeError("commit called twice")
self._reqs = dist.batch_isend_irecv(self._ops)
with torch.cuda.stream(self.comm_stream):
self._reqs = dist.batch_isend_irecv(self._ops)

def wait(self):
if self._reqs is None:
Expand All @@ -44,3 +47,5 @@ def wait(self):
req.wait()
self._reqs = None
self._ops = []
# compute stream must wait for communication
torch.cuda.current_stream().wait_stream(self.comm_stream)
Loading