[llvm] [AArch64] Fix #94909: Optimize vector fmul(sitofp(x), 0.5) -> scvtf(x, 2) (PR #141480)

[David Green via llvm-commits]

https://github.com/davemgreen commented:

Thanks - I think the handling of MOVIshift might need to be tightened up a bit, otherwise it could ignore important certain bits of the constant. Detecting the constant correctly is the most awkward part of this with all the different combos, but luckily it looks like many of them use a similar node type.

https://github.com/llvm/llvm-project/pull/141480