Fix the legacy loop indexing traversal #3373
Open
+90
−0
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
|
!test --diff |
naoyam
changed the title
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Outdated suggestions cannot be applied.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 is a temporary WAR for #3374. It's temporary since the repro has no problem with the IdModel-based indexer. This is for unblocking @IvanYashchuk until we can make the new indexer enabled by default.
The root cause of the issue is when we attempt to find a correct indexing path from the loop domain to the allocation domain of the indexed tensor, the algorithm fails to find a path visiting a backward merge when the indexed tensor has only one of the inputs. That happens when the tensor is broadcast and gets inlined with broadcast forwarding. In the current code, in that case, it just picks the first traversal option, which I think happens to be working fine, but that's not necessarily the right chose, particularly because we are looking at all candidate next traversal targets that are permissively mapped.
The WAR is simply picking a candidate as long as it has at least one mapped ID. I think this would be good enough as a temporary WAR.
Fixes #3374