Skip to content

fix(core): unit-test peer access planning without multi-GPU hardware#1773

Open
cpcloud wants to merge 1 commit intoNVIDIA:mainfrom
cpcloud:issue-802
Open

fix(core): unit-test peer access planning without multi-GPU hardware#1773
cpcloud wants to merge 1 commit intoNVIDIA:mainfrom
cpcloud:issue-802

Conversation

@cpcloud
Copy link
Contributor

@cpcloud cpcloud commented Mar 16, 2026

Summary

  • extract DeviceMemoryResource.peer_accessible_by transition planning into a helper so the normalization and delta logic can be validated without multiple GPUs
  • add single-GPU tests for peer-access normalization, inaccessible-peer validation, and transition planning while keeping the existing x2/x3 integration tests
  • preserve the hardware-backed peer-access and IPC integration checks so multi-GPU machines still exercise the end-to-end behavior

Test plan

  • pixi run -e default pytest tests/test_memory_peer_access_utils.py tests/test_memory_peer_access.py -q
  • pixi run -e default pytest tests/memory_ipc/test_peer_access.py -q

Made with Cursor

Extract the peer-access transition planning from DeviceMemoryResource so stale-state regressions can be covered on single-GPU systems. Keep the existing multi-GPU integration tests for end-to-end peer access behavior.

Made-with: Cursor
@github-actions
Copy link

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.

1 participant