Repository navigation
fix(hip-tests): limit peer-import VMM copy overlap test to 2 streams - #14829
Merged
px1624 merged 1 commit intoNov 3, 2026
Merged
Conversation
Unit_hipMemImportFromShareableHandle_ImportFromPeersCopiesOverlapAcrossStreams
failed on MI350X in CPX + NPS2 ("held back copy 1", other_copies_done == false).
True P2P copies should run on SDMA, so their parallelism is bounded by the copy
device's SDMA engines. A CPX partition has only 2, with no preferred engine to
the other partitions of the same GPU. The test used up to 4 streams: one blit
copy (device 0 -> device 0) and 3 SDMA copies (devices 1-3 -> device 0), so two
peer copies shared an engine and the one held by hipStreamWaitValue32 blocked
the other. This is expected behavior, not a runtime bug.
Use 2 streams, the minimum to show overlap: the device 0 <-> device 0 blit copy
and the device 1 -> device 0 SDMA copy should overlap. Document that the
SelfImport and ImportOnPeers cases allocate physical memory on device 0, so all
of their copies are blit copies.
JIRA ID: ROCM-33753
✅ All Policy Checks Passed
📖 Need help? See the Policy FAQ for details on every check and how to fix failures. |
Contributor
There was a problem hiding this comment.
🟢 Approval recommended
No active review issues were identified for the peer-import overlap test adjustment.
No open issues 🎉
What changed in this PR
Adjusts the HIP virtual-memory peer-import overlap test to avoid expected SDMA engine contention on MI350X CPX partitions while preserving overlap coverage.
Changes:
- Limits the peer-import test to two streams: one local blit copy and one peer SDMA copy.
- Updates comments to explain the stream limit and clarify that SelfImport and ImportOnPeers use only blit copies.
| File | Description |
|---|---|
projects/hip-tests/catch/unit/virtualMemoryManagement/hipMemImportFromShareableHandle.cc |
Limits peer-import overlap testing to two streams and documents copy-engine behavior. |
Review effort: Balanced
chrispaquot
approved these changes
Nov 2, 2026
Contributor
TheRock Submodule Bump ActivityNewest first
|
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
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.
Motivation
Unit_hipMemImportFromShareableHandle_ImportFromPeersCopiesOverlapAcrossStreamsfailed on MI350X in CPX + NPS2Technical Details
True P2P copies should run on SDMA, so their parallelism is bounded by the copy device's SDMA engines. A CPX partition has only 2, with no preferred engine to the other partitions of the same GPU. The test used up to 4 streams: one blit copy (device 0 -> device 0) and 3 SDMA copies (devices 1-3 -> device 0), so two peer copies shared an engine and the one held by hipStreamWaitValue32 blocked the other. This is expected behavior, not a runtime bug.
Use 2 streams, the minimum to show overlap: the device 0 <-> device 0 blit copy and the device 1 -> device 0 SDMA copy should overlap. Document that the SelfImport and ImportOnPeers cases allocate physical memory on device 0, so all of their copies are blit copies.
Issue Tracking
JIRA ID: ROCM-33753
Test Plan
Validate failed tests is passing with this PR
Test Result
The 5 repeat runs of the fixed ImportFromPeersCopiesOverlapAcrossStreams test:
All tests passed (11 assertions in 1 test case)
All tests passed (11 assertions in 1 test case)
All tests passed (11 assertions in 1 test case)
All tests passed (11 assertions in 1 test case)
All tests passed (11 assertions in 1 test case)
Submission Checklist