Warn when ManagedMemoryResource is created without concurrent managed access #1618
+127
−2
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
… access Platforms where Device.properties.concurrent_managed_access is False silently allow managed memory allocation but segfault on unsynchronized host access. Emit a one-time UserWarning at construction time so users get actionable guidance before hitting a crash. Also skip test_build_hooks.py when setuptools is unavailable (Python 3.14+). Closes NVIDIA#1607 Co-authored-by: Cursor <cursoragent@cursor.com>
Andy-Jost
added
bug
Contributor
Contributor
Author
|
/ok to test f7eeec8 |
|
kkraus14
reviewed
Feb 13, 2026
| "allocation is forbidden while any GPU kernel is in flight, even " | ||
| "if the kernel does not touch that allocation. Failing to " | ||
| "synchronize before host access will cause a segfault. " | ||
| "See: https://docs.nvidia.com/cuda/cuda-c-programming-guide/index.html#um-coherency-hd", |
Collaborator
There was a problem hiding this comment.
I think we want https://docs.nvidia.com/cuda/cuda-c-programming-guide/index.html#gpu-exclusive-access-to-managed-memory as the link?
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
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Outdated suggestions cannot be applied.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.
Summary
UserWarningwhenManagedMemoryResourceis created on a platform whereDevice.properties.concurrent_managed_accessis False. On such platforms, unsynchronized host access to managed memory causes a segfault.test_build_hooks.pywhensetuptoolsis unavailable (Python 3.14+).Closes #1607
Changes
cuda/core/_memory/_managed_memory_resource.pyx: Add_check_concurrent_managed_accesshelper with double-checked locking (warn-once pattern), andreset_concurrent_access_warning()for testing.tests/test_managed_memory_warning.py: New test file verifying the warning fires and only fires once.tests/test_build_hooks.py: Addpytest.importorskip("setuptools").Test Coverage
test_warning_emitted: Verifies the warning content (category, message text).test_warning_emitted_only_once: Verifies warn-once behavior across multiple resource creations.concurrent_managed_access=False).Made with Cursor