NVIDIA / NVIDIA/raft

[BUG] mpi_comms comm_split does not split NCCL communicator

Open
#75 2 comments 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

bug inactive-30d inactive-90d
Dominant language
Cuda
Stars
1k
Forks
251
Avg merge
1d 8h
Merged PRs (30d)
31

Description

https://github.com/rapidsai/raft/blob/branch-0.16/cpp/include/raft/comms/mpi_comms.hpp#L136

  std::unique_ptr<comms_iface> comm_split(int color, int key) const {
    MPI_Comm new_comm;
    MPI_TRY(MPI_Comm_split(mpi_comm_, color, key, &new_comm));
    return std::unique_ptr<comms_iface>(new mpi_comms(new_comm, true));
  }

mpi_comms uses MPI for P2P and NCCL for collectives, but it split only MPI communicator in comm_split.

Contributor guide

No contributing guide indexed for this repository

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

Research direction

Start in cpp/include/raft/comms/mpi_comms.hpp at the comm_split implementation shown in the issue, then trace how mpi_comms stores and initializes its MPI and NCCL communicators. Update the split behavior so both communicators represent the requested partition, and verify that the resulting communicator is usable for MPI point-to-point and NCCL collective operations.

Written by the indexing model from the issue text.

Assessment

Tech stack
cpp
Domain
distributed-systems
Issue type
Bug
Difficulty
4/5
Estimated time
3-5 days
Activity status
Stale
Clarity
Mostly clear
Newbie friendliness
35/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.