[AD] Fix missing masks on symbolic calls inside loops #338
Conversation
This file contains 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
When the `CallOp` node is traversed in forward/reverse-mode it is possible that the mask stack no longer matches what it was during the initial tracing. The `CallOp` therfore stores an extra reference to the mask at the top of the mask and applies it when it is traversed
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Outdated suggestions cannot be applied.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.
When running AD through evaluated-mode loops, any dynamic dispatch will not have the loop's mask applied.
This PR fixes this issue by saving the loop's mask (top of the mask stack) at the time of tracing the dynamic dispatch into the
CallOpcustom operation. This allows the custom operation to then apply that saved masked whenever it gets traversed.Without this patch, the issue would typically manifest itself as a CUDA error or segfault during kernel execution. Typically, lanes that should have been masked were accessing out-of-bounds memory.
A regression test was added in
test_switch.py::test18_apply_loop_mask.