[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:

> > @aemerson Any opinions on making reg-bank-select a little better? My understanding from a while ago was that the greedy mode did not work very well (maybe?), but I feel we will eventually need something like it to produce optimal code for all the cases that could be using gpr or fpr.
> 
> I don't even know if the greedy mode works at all for AArch64, we've always used the default mode. I think we can revisit it after we rework RBS once we have FP types.

@aemerson Until then, do you feel that it is viable to implement this as a fold in AArch64PostSelectOptimize? Unfortunately, this pattern has real-world implications for performance that we are running into, and I would like to have it fixed...

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