[llvm-branch-commits] [llvm] release/23.x: [AArch64] Fix cost of vector shifts with non-uniform constants(#214206) (PR #215625)

[Douglas Yung via llvm-branch-commits]

dyung wrote:

> This change improves the performance ~4% for imagick and ~3% for leela. These measures are for neoverse-v1/v2.

What are the performance regressions that are mentioned?

https://github.com/llvm/llvm-project/pull/215625