Skip to content

Introduce reduce-scatter algo - #866

Open
Binyang Li (Binyang2014) wants to merge 17 commits into
mainfrom
binyli/allgather
Open

Introduce reduce-scatter algo#866
Binyang Li (Binyang2014) wants to merge 17 commits into
mainfrom
binyli/allgather

Conversation

@Binyang2014

Copy link
Copy Markdown
Contributor

No description provided.

Caio Rocha and others added 11 commits July 30, 2026 22:05
Co-authored-by: Binyang2014 <9415966+Binyang2014@users.noreply.github.com>
Restore the previous FIFO protocol because the parity-based commit scheme stalls PortChannel proxy traffic on TP64.

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Copilot-Session: 60d476fd-fd9a-46ef-a7f8-c4d7fd05ef1d
@azure-pipelines

Copy link
Copy Markdown
Azure Pipelines:
There may be pipelines that require an authorized user to comment /azp run to run.

@Binyang2014
Binyang Li (Binyang2014) marked this pull request as ready for review August 11, 2026 17:24
@azure-pipelines

Copy link
Copy Markdown
Azure Pipelines:
There may be pipelines that require an authorized user to comment /azp run to run.

Copilot AI left a comment

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Pull request overview

Introduces a hierarchical, packet-based multi-node ReduceScatter algorithm and supporting executor/API updates.

Changes:

  • Adds and exports the ReduceScatter algorithm.
  • Adds thread-block-group support to port packet transfers.
  • Reuses cached execution contexts and limits IB registration to plans requiring IB.

Reviewed changes

Copilot reviewed 4 out of 4 changed files in this pull request and generated 3 comments.

File Description
src/core/executor/executor.cc Improves context reuse and IB registration.
python/mscclpp/language/channel.py Adds thread-block-group packet transfers.
python/mscclpp/default_algos/reducescatter_multi_nodes.py Implements hierarchical ReduceScatter.
python/mscclpp/default_algos/__init__.py Exports the new algorithm.

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

Comment thread python/mscclpp/default_algos/reducescatter_multi_nodes.py Outdated
Comment thread python/mscclpp/language/channel.py
Comment thread python/mscclpp/default_algos/reducescatter_multi_nodes.py Outdated
@Binyang2014
Binyang Li (Binyang2014) requested a review from a team August 11, 2026 23:04
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

4 participants