[llvm] [MVE] Expand64BitShift - handle all constant shift amounts less than 32 (RFC) (PR #81261)

[Simon Pilgrim via llvm-commits]

RKSimon wrote:

> Considering what this changes I think it looks OK to me. Were you running into other cases where it was getting worse too?

Thanks - I've seen no other cases, and what I'm seeing tends to be a lot of zext/sext nodes around illegal types that get promoted to i64 (or beyond). My guess would be that ComputeKnownBits/NumSignBits support for long shift target opcodes should help but I couldn't find a full description of how these instructions work to allow me to test it.

https://github.com/llvm/llvm-project/pull/81261