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

[David Green via llvm-branch-commits]

davemgreen wrote:

The only regression I have seen has been in an oddly written matrix multiply routine, and the size of the regression was not very large. It was only a single case that I have seen so far, and I would not consider it important enough to hold this up given the improvements elsewhere. Thanks.

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