In PjRtStreamExecutorBuffer::Delete, fix some bugs causing memory corruption. #67688
+104
−7
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Outdated suggestions cannot be applied.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.
In PjRtStreamExecutorBuffer::Delete, fix some bugs causing memory corruption.
The bug is that we don't wait for events using the buffer on the Compute stream, and also the definition events. This can cause a race condition where the buffer is deleted before it is read/written, which can lead to memory corruptions.
For events on the compute stream, the fix is to schedule the deallocation on the compute stream.
For the definition events, the fix is to wait for them on a borrowed stream if they are not defined on the compute stream.
This change probably wouldn't be needed if all writes are scheduled on the compute stream, but that is not the case today. We have a lot of usages of the h2d and d2d stream.