Skip to content

Fix SVE2/SME2 path of exp_f32 for some inputs

Mark Horvath requested to merge fix-exp-scalable into main

In case of SVE2/SME2 the order of some multiplications was incorrect and for inputs resulting almost positive infinte values the calculation was faulty.

In case of NEON the change is not functional, it just makes easier to the reader to understand the order of multiplications.

Merge request reports