[llvm] [X86] getScalarMaskingNode - FIXUPIMM scalar ops take upper elements from second operand (PR #179101)

[Simon Pilgrim via llvm-commits]

RKSimon wrote:

FIXUPIMM takes 3 vector operands so the dst/src ordering is a bit different to regular binops - I can't see any other intrinsics that use the getScalarMaskingNode path that requires this and most builtin intrinsics do the masking separately now so aren't affected.

 @phoebewang can you think of any others?

https://github.com/llvm/llvm-project/pull/179101