#sdy avoid sideways operand propagation for elementwise ops which can't propagate to the result. #382
+105
−18
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
![copybara-service[bot]](https://avatars.githubusercontent.com/in/44061?s=60&v=4){kind=small}
…'t propagate to the result.
We don't want to propagate sideways through operands if the element-wise factor is used in the result and is not sharded in the same way. We want to avoid this to avoid the following situation which can happen when a `sharding_constraint` is added onto the operand during Shardy import:
```mlir
%arg0: [{"a", ?}]
%arg1: [{?}]
%0 = add %arg0, %arg1 : [{}]
```
We don't want to do an all-gather on both %arg0 and %arg1 due to "a" propagating sideways. Instead with the code below, since "a" can't propagate to `%0`, we will only do an all-gather on %arg0.
Long term we should undo this and allow sideways propagation, but have our explicit reshard pass make sure the result is all-gathered instead of both operands.
PiperOrigin-RevId: 726920093
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Outdated suggestions cannot be applied.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.
#sdy avoid sideways operand propagation for elementwise ops which can't propagate to the result.
We don't want to propagate sideways through operands if the element-wise factor is used in the result and is not sharded in the same way. We want to avoid this to avoid the following situation which can happen when a
sharding_constraintis added onto the operand during Shardy import:We don't want to do an all-gather on both %arg0 and %arg1 due to "a" propagating sideways. Instead with the code below, since "a" can't propagate to
%0, we will only do an all-gather on %arg0.Long term we should undo this and allow sideways propagation, but have our explicit reshard pass make sure the result is all-gathered instead of both operands.