Skip to content

Support group API with only SEND or RECV operations #193

@pkourdis

Description

@pkourdis

Summary

Group API of oneCCL expects the collectives grouped to finish before group_end() so the following pattern is not supported:

group_start();
send(send_buf_ptr, sendcount, ...);
group_end();
group_start();
recv(recv_buf_ptr, sendcount, ...);
group_end();

However, the above pattern is supported by NCCL.

Reproducer

See internal MLSL-3958.

Affected projects

XCCL backend for TorchComms.

Metadata

Metadata

Assignees

No one assigned

    Labels

    enhancementNew feature or request

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions