[llvm-branch-commits] [llvm] [AArch64] Fold MIN/MAX(Vec[0], Vec[1]) to VECREDUCE_MIN/MAX(Vec) (PR #181162)

[Benjamin Maxwell via llvm-branch-commits]

MacDue wrote:

> Do the ll tests represent the expected input IR? If yes then I think this would be better implemented as an IR canonicalisation.

I have already implemented that, but it's pretty unlikely it'd be accepted. In the past I've been told that folding to the `vector.reduce.*` intrinsics is less canonical, as more combines/optimizations understand the scalar form. 

https://github.com/llvm/llvm-project/pull/181162