Add small MPI_Alltoall after autotuning to work around MPI memory registration delaying cudaFree.#47
Merged
romerojosh merged 2 commits intomainfrom Jan 31, 2025
Merged
Conversation
…ns delaying cudaFree.
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
1 participant
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
We have observed that some versions of UCX have a behavior where buffers used in MPI operations may not be immediately freed, even once
cudaFreeis called on them. This is due to their internal caching of memory registration handles on these buffers. These cached handles are not immediately destroyed upon a call tocudaFreeand the CUDA driver will not reclaim this memory until these registration handles are freed.This issue impacts memory usage after cuDecomp autotuning, where the memory used for the autotuning test buffers may not be immediately allocatable by users after autotuning completes. I have implemented a workaround that attempts to alleviate this issue in this PR. The workaround is to launch a
MPI_Alltoallon small GPU buffers after autotuning completes to force UCX to invalidate and destroy the stale registration handles.