[llvm] [AMDGPU] Fix true16 high-half packing when the low operand is not a VGPR (PR #220843)

[Joe Nash via llvm-commits]

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

----------------
Sisyph wrote:

Still looks to be a difference between sdag and globalisel here, but It does not seem to be a problem and may not even be related to this patch. 

https://github.com/llvm/llvm-project/pull/220843