Fix batch graph capture issue caused by pinned memory allocator #110
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
Kh4ster
requested review from
akifcorduk,
aliceb-nv and
chris-maes
and removed request for
akifcorduk and
chris-maes
Kh4ster
added
bug
Contributor
Author
|
/merge |
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Outdated suggestions cannot be applied.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.
This PR aims at fixing the intermittent error we are seeing in CI on the Python Batch PDLP test.
Previous PR were trying to achieve the same while not solving the root cause: work being pushed to the default stream.
As described here: NVIDIA/cccl#5027 creating a thrust::host pinned memory allocator is incorrectly launching work on the GPU on the default stream which might happen during a graph capture, creating an error.
To fix it, we now use a std::unique_ptr with a custom allocator to allocate host pinned memory.
This allows us to go back to a non blocking stream when using batch mode.