[llvm] [LLVM][AArch64] Correctly lower funnel shifts by zero. (PR #140058)

[Paul Walker via llvm-commits]

paulwalker-arm wrote:

@RKSimon - If getNode() performs this canonicalisation, can I be sure there's no way for the lowering code to see `fshl(x,y,0)`? I'm wondering if the lowering could end up generating the "zero" shift amount without there being a new `ISD::FHSL` construction to canonicalise it.

If that can happen then the custom lowering code must still handle it?

https://github.com/llvm/llvm-project/pull/140058