[llvm] [AArch64] Enable Spillage Copy Elimination by default (PR #186093)

[David Green via llvm-commits]

https://github.com/davemgreen commented:

This makes sense to me and is something we have seen multiple times before. It is pretty horrendous for performance when it does happen. I was wondering if the implementation could be better (why do findLastSeenDefInCopy scan through instructions again, could we remove the outer set of mov's?).

Can you add a test case for the issue you had ran into. Either a .ll files (which might change as other changes are made), or a .mir test for the pass directly.

https://github.com/llvm/llvm-project/pull/186093