[llvm] [X86] Avoid trunc/zext in narrow shift by working over i32 (PR #185539)

[Gergo Stomfai via llvm-commits]

================

----------------
stomfaig wrote:

This will have to be dealt with, but I think this is just a question of gating srl over i8 not being desirable for i686 

https://github.com/llvm/llvm-project/pull/185539