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

Jump to bottom

Support one-sided halo for DNDarrays #1509

Merged
merged 1 commit into from
Jun 5, 2024
Merged

Support one-sided halo for DNDarrays #1509

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
38 changes: 23 additions & 15 deletions heat/core/dndarray.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -384,14 +384,18 @@ def __prephalo(self, start, end) -> torch.Tensor:

return self.__array[ix].clone().contiguous()

def get_halo(self, halo_size: int) -> torch.Tensor:
def get_halo(self, halo_size: int, prev: bool = True, next: bool = True) -> torch.Tensor:
"""
Fetch halos of size ``halo_size`` from neighboring ranks and save them in ``self.halo_next/self.halo_prev``.

Parameters
----------
halo_size : int
Size of the halo.
prev : bool, optional
If True, fetch the halo from the previous rank. Default: True.
next : bool, optional
If True, fetch the halo from the next rank. Default: True.
"""
if not isinstance(halo_size, int):
raise TypeError(
Expand Down Expand Up @@ -433,25 +437,29 @@ def get_halo(self, halo_size: int) -> torch.Tensor:
req_list = []

# exchange data with next populated process
if rank != last_rank:
self.comm.Isend(a_next, next_rank)
res_prev = torch.zeros(
a_prev.size(), dtype=a_prev.dtype, device=self.device.torch_device
)
req_list.append(self.comm.Irecv(res_prev, source=next_rank))
if prev:
if rank != last_rank:
self.comm.Isend(a_next, next_rank)
if rank != first_rank:
res_prev = torch.zeros(
a_prev.size(), dtype=a_prev.dtype, device=self.device.torch_device
)
req_list.append(self.comm.Irecv(res_prev, source=prev_rank))

if rank != first_rank:
self.comm.Isend(a_prev, prev_rank)
res_next = torch.zeros(
a_next.size(), dtype=a_next.dtype, device=self.device.torch_device
)
req_list.append(self.comm.Irecv(res_next, source=prev_rank))
if next:
if rank != first_rank:
self.comm.Isend(a_prev, prev_rank)
if rank != last_rank:
res_next = torch.zeros(
a_next.size(), dtype=a_next.dtype, device=self.device.torch_device
)
req_list.append(self.comm.Irecv(res_next, source=next_rank))

for req in req_list:
req.Wait()

self.__halo_next = res_prev
self.__halo_prev = res_next
self.__halo_next = res_next
self.__halo_prev = res_prev
self.__ishalo = True

def __cat_halo(self) -> torch.Tensor:
Expand Down
Loading