[llvm] AArch64: Optimize memmove for non-power-of-two sizes (PR #168633)

[David Green via llvm-commits]

https://github.com/davemgreen commented:

I think I would expect this to to work similarly to memcpy, where we already lower to unaligned loads. Providing we do the loads before the stores that sounds like it should be alright.

https://github.com/llvm/llvm-project/pull/168633