Skip to content

More efficient cumsum for offset_per_key calculation for torch.export #1788

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
wants to merge 1 commit into from
Closed

More efficient cumsum for offset_per_key calculation for torch.export #1788

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
49 changes: 47 additions & 2 deletions torchrec/sparse/jagged_tensor.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -818,9 +818,54 @@ def _maybe_compute_offset_per_key(
offsets=offsets,
values=values,
)
return _length_per_key, _cumsum(_length_per_key)

if (
not torch.jit.is_scripting()
and not is_torchdynamo_compiling()
and torch.compiler.is_compiling()
):
return (
_length_per_key,
(
torch.ops.fbgemm.asynchronous_complete_cumsum(
torch._refs.tensor(
_length_per_key,
dtype=torch.int32,
device=torch.device("cpu"),
pin_memory=False,
requires_grad=False,
)
).tolist()
if len(_length_per_key) > 0
else []
),
)
else:
return _length_per_key, _cumsum(_length_per_key)
elif offset_per_key is None:
return length_per_key, _cumsum(length_per_key)
if (
not torch.jit.is_scripting()
and not is_torchdynamo_compiling()
and torch.compiler.is_compiling()
):
return (
length_per_key,
(
torch.ops.fbgemm.asynchronous_complete_cumsum(
torch._refs.tensor(
length_per_key,
dtype=torch.int32,
device=torch.device("cpu"),
pin_memory=False,
requires_grad=False,
)
).tolist()
if len(length_per_key) > 0
else []
),
)
else:
return length_per_key, _cumsum(length_per_key)
else:
return length_per_key, offset_per_key

Expand Down