Fix fusion with inplace #1659
Closed
Fix fusion with inplace #1659
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
Codecov ReportAttention: Patch coverage is
Additional details and impacted files@@ Coverage Diff @@
## main #1659 +/- ##
==========================================
- Coverage 86.41% 86.36% -0.06%
==========================================
Files 694 694
Lines 80675 80690 +15
==========================================
- Hits 69719 69684 -35
- Misses 10956 11006 +50 ☔ View full report in Codecov by Sentry. |
Comment on lines
+196
to
+199
| if desc.shape == desc_input.shape | ||
| && input.item() == output.item() | ||
| && desc.id == desc_input.id | ||
| { |
There was a problem hiding this comment.
I don't think that's right, it's not a requirement! But I think we should verify before chosing an input to replace the output if the output is also an input, in which case we have to use it.
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Outdated suggestions cannot be applied.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.
The bug in #1656 seems to be caused by fusion when it works on several read-write tensors.
What happened was rather subtle: if the output of the fused kernel was a mutable tensor, then it was "Readwrite", even if there were only writes to it in this specific kernel. So while looking to match the outputs with potential inplace inputs it would not find itself and could wrongly assume that another tensor with the same shape could was it. So it did not write to the right tensor.
Batches of one allowed to detect it, but I think the bug happened with other batch sizes as well.
@nathanielsimard I'm not sure why this id comparison was not there before. Could adding it break something?
Fix #1656