fix: Repair invalid schema arising from lowering pass by gs-olive · Pull Request #1786 · pytorch/TensorRT (original) (raw)

Description

When remove_unnecessary_casts replaces both tensors in an aten::div call with the corresponding scalars, and the rounding mode (third argument) is specified, the schema becomes invalid
To avoid this, we intercept the call and split the result based on what the third argument in the input was, either delegating the result to aten::Int(aten::div(...)), for "trunc" or aten::floordiv(...) for "floor"
Update lowering pass to incorporate the above change, with appropriate documentation
Add testing to verify catching and correctly handling these edge cases

Fixes #1780

Please delete options that are not relevant and/or add your own.

[ x ] My code follows the style guidelines of this project (You can use the linters)
[ x ] I have performed a self-review of my own code
[ x ] I have commented my code, particularly in hard-to-understand areas and hacks
[ x ] I have made corresponding changes to the documentation
[ x ] I have added tests to verify my fix or my feature
[ x ] New and existing unit tests pass locally with my changes
[ x ] I have added the relevant labels to my PR in so that relevant reviewers are notified

When remove_unnecessary_casts replaces both tensors in an aten::div call with the corresponding scalars, and the rounding mode (third argument) is specified, the schema becomes invalid
To avoid this, we intercept the call and split the result based on what the third argument in the input was, either delegating the result to aten::Int(aten::div(...)), for "trunc" or aten::floordiv(...) for "floor"
Update lowering pass to incorporate the above change, with appropriate documentation
Add testing to verify catching and correctly handling these edge cases

The reason will be displayed to describe this comment to others. Learn more.

LGTM

bowang007 pushed a commit that referenced this pull request

bowang007 pushed a commit that referenced this pull request