Skip to content

dynamo: keep native NCCL collectives out of Myelin ForeignNode fusion (cast-boundary) #4388

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

Draft
pkisfaludi-nv wants to merge 3 commits into
base: main
Choose a base branch
from
Draft

dynamo: keep native NCCL collectives out of Myelin ForeignNode fusion (cast-boundary) #4388

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
1d34143
dynamo: target native NCCL collectives at the op's process (sub)group
pkisfaludi-nv Jul 4, 2026
8daa20a
dynamo: keep native NCCL collectives out of Myelin ForeignNode fusion…
pkisfaludi-nv Jul 7, 2026
ed8d41a
dynamo: test compute and collective in one TRT engine
pkisfaludi-nv Jul 7, 2026
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: 8 additions & 2 deletions py/torch_tensorrt/dynamo/conversion/custom_ops_converters.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -44,6 +44,7 @@ def fused_nccl_all_gather(
SourceIR.ATEN,
name,
[args[0]],
group_name=args[2] if len(args) > 2 else None,
)

@dynamo_tensorrt_converter(
Expand All @@ -66,6 +67,7 @@ def fused_nccl_reduce_scatter(
name,
[args[0]],
reduce_op=reduce_op,
group_name=args[3] if len(args) > 3 else None,
)

@dynamo_tensorrt_converter(
Expand All @@ -87,6 +89,7 @@ def fused_nccl_all_reduce(
name,
[args[0]],
reduce_op=reduce_op,
group_name=args[2] if len(args) > 2 else None,
)

@dynamo_tensorrt_converter(
Expand All @@ -106,6 +109,7 @@ def fused_nccl_all_to_all(
SourceIR.ATEN,
name,
[args[0]],
group_name=args[3] if len(args) > 3 else None,
)

@dynamo_tensorrt_converter(
Expand All @@ -121,7 +125,8 @@ def fused_nccl_scatter(
"""Scatter using native TensorRT DistCollective API."""
root = args[1] if len(args) > 1 else 0
return impl.nccl_ops.nccl_scatter_native(
ctx, target, SourceIR.ATEN, name, [args[0]], root=root
ctx, target, SourceIR.ATEN, name, [args[0]], root=root,
group_name=args[2] if len(args) > 2 else None,
)

@dynamo_tensorrt_converter(
Expand All @@ -137,7 +142,8 @@ def fused_nccl_gather(
"""Gather using native TensorRT DistCollective API."""
root = args[1] if len(args) > 1 else 0
return impl.nccl_ops.nccl_gather_native(
ctx, target, SourceIR.ATEN, name, [args[0]], root=root
ctx, target, SourceIR.ATEN, name, [args[0]], root=root,
group_name=args[2] if len(args) > 2 else None,
)


Expand Down
Loading
Loading