[llvm] [X86] Fix VECREDUCE_XOR to PARITY lowering for 16-bit elements (PR #219216)

[Nikita Popov via llvm-commits]

nikic wrote:

> LGTM as a quick fix cheers - we can tweak the MOVMSK handling in a future patch to get this back.

What do you have in mind here? Something like masking the result with a 0b1010... mask, or is there some better way?

https://github.com/llvm/llvm-project/pull/219216