Skip to content

Support profiling of point-to-point collective operations.  #47482

@rohan-varma

Description

@rohan-varma

🚀 Feature

#46471 is adding support for profiling torch.distributed APIs such as allgather, allreduce, etc. for NCCL and Gloo backends. Currently, we only support collective communication APIs such as the aformentioned ones, but not point-to-point ones (such as send/recv). We should add and test support for send/recv for backends that support it.

Metadata

Metadata

Assignees

Labels

featureA request for a proper, new feature.module: c10dIssues/PRs related to collective communications and process groupstriagedThis issue has been looked at a team member, and triaged and prioritized into an appropriate module

Type

No type

Projects

No projects

Milestone

No milestone

Relationships

None yet

Development

No branches or pull requests

Issue actions