Add CHLO square operation. #2625
Merged
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
pearu
force-pushed
the
pearu/chlo-square
branch
from
44a1dc8
to
6e60fb7
Compare
Update acos/asin/... tests.
pearu
force-pushed
the
pearu/chlo-square
branch
from
6e60fb7
to
b106a65
Compare
GleasonK
approved these changes
Nov 18, 2024
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Outdated suggestions cannot be applied.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.
As in the title.
Providing
squareas a CHLO operation enables fixing inaccuracy problems in jax.numpy.square for complex inputs with large real and imaginary parts.This PR replaces #2623 and enables eliminating
squarelowering introduced in jax-ml/jax#24874The details of square inaccuracies is given in pearu/functional_algorithms#52 . Roughly speaking, the source of inaccuracies in square(z).real is the usage of z.real ** 2 - z.imag ** 2 that terms will overflow when abs(z.real/imag) > sqrt(largest). The overflow does not happen when using (z.real - z.imag) * (z.real + imag) formula instead.
As a side-effect to using the 0.11.1 version of function_algorithms, several mlir test scripts for asin/asinh/acos/... functions are updated as well.