[llvm] [AArch64] Fold COPY(y:gpr, DUP(x:fpr, i)) -> UMOV(y:gpr, x:fpr, i) (PR #89017)

[Dhruv Chawla via llvm-commits]

dc03-work wrote:

> AArch64PostSelectOptimize sounds like a decent compromise. Could it be made to fix the regressions you see at the same time?

Hmm, the problem is `COPY(FPR->GPR) -> COPY(GPR->FPR)` being converted to `UMOV(FPR->GPR) -> COPY(GPR->FPR)`. Fixing this would mean having to effectively undo the optimization when one of the users is such a `COPY(GPR->FPR)`, but I'm not sure where that would go?

https://github.com/llvm/llvm-project/pull/89017