Use GpuIpcMem for NVLS connections
#719
Merged
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
- [x] Move hash specialization and equality operator from std/global namespace to custom namespace - [x] Update unordered_map to use custom hash and equality as template parameters - [x] Add noexcept to equality operator - [x] Verify the changes build correctly - [x] Run code review and security checks <!-- START COPILOT CODING AGENT TIPS --> --- ✨ Let Copilot coding agent [set things up for you](https://github.com/microsoft/mscclpp/issues/new?title=✨+Set+up+Copilot+instructions&body=Configure%20instructions%20for%20this%20repository%20as%20documented%20in%20%5BBest%20practices%20for%20Copilot%20coding%20agent%20in%20your%20repository%5D%28https://gh.io/copilot-coding-agent-tips%29%2E%0A%0A%3COnboard%20this%20repo%3E&assignees=copilot) — coding agent works faster and does higher quality work when set up for your repo. --------- Co-authored-by: copilot-swe-agent[bot] <198982749+Copilot@users.noreply.github.com> Co-authored-by: Binyang2014 <9415966+Binyang2014@users.noreply.github.com> Co-authored-by: Binyang Li <binyli@microsoft.com>
Contributor
There was a problem hiding this comment.
Pull request overview
This PR refactors NVLS (NVLink Sharp) connection handling to use the unified GpuIpcMem abstraction instead of directly managing CUDA multicast APIs. This simplifies the code by delegating multicast memory management to the existing GpuIpcMem infrastructure.
Key Changes
- Replaces manual multicast handle management with
GpuIpcMemandGpuIpcMemHandleabstractions - Removes manual buffer allocation tracking (
allocatedRanges_,freeRanges_) as this is now handled internally byGpuIpcMem - Removes explicit synchronization barriers in
connectNvlsCollective, relying instead on the blocking behavior ofcuMulticastBindAddrwithinGpuIpcMem::mapMulticast()
Reviewed changes
Copilot reviewed 3 out of 3 changed files in this pull request and generated 2 comments.
| File | Description |
|---|---|
| src/switch_channel.cc | Complete refactoring of NvlsConnection::Impl to use GpuIpcMem; updated license header; removed buffer allocation logic and synchronization barriers; simplified bindMemory implementation |
| python/csrc/switch_channel_py.cpp | Removed Python binding for getMultiCastMinGranularity method (breaking API change) |
| include/mscclpp/switch_channel.hpp | Removed public API methods addDevice() and getMultiCastMinGranularity() (breaking API changes) |
Contributor
Author
|
/azp run mscclpp-ut |
|
Azure Pipelines could not run because the pipeline triggers exclude this branch/path. |
Contributor
Author
|
/azp run mscclpp-ut |
|
Azure Pipelines successfully started running 1 pipeline(s). |
Contributor
Author
|
/azp run mscclpp-ut |
|
Azure Pipelines successfully started running 1 pipeline(s). |
Contributor
There was a problem hiding this comment.
Pull request overview
Copilot reviewed 7 out of 7 changed files in this pull request and generated 3 comments.
Comments suppressed due to low confidence (2)
python/csrc/switch_channel_py.cpp:2
- The license header should use "MIT License" instead of "MIT license" (capital L) to match the project's standard format seen in other C++ files.
// Licensed under the MIT license.
include/mscclpp/switch_channel.hpp:2
- The license header should use "MIT License" instead of "MIT license" (capital L) to match the project's standard format used in the implementation files.
// Licensed under the MIT license.
Closed
Binyang2014
reviewed
Jan 15, 2026
Binyang2014
approved these changes
Jan 15, 2026
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
NvlsConnectioninternally reusesGpuIpcMemfor multicast memory handling.connectNvlsCollective()(CUDA API handles this automatically).GpuIpcMem::map()andGpuIpcMem::mapMulticast()to return a shared pointer with custom deleter for unmapping, which prevents misuse of raw pointers and reduces states to be stored in theGpuIpcMeminstance.RuntimeIpctype handles, for consistency with other types,cudaIpcOpenMemHandlewill be called inGpuIpcMem::map()instead of the ctor ofGpuIpcMem.