[llvm] [SelectionDAG] Improve v2f16 maximumnum expansion (PR #160723)

[Artem Belevich via llvm-commits]

https://github.com/Artem-B commented:

LGTM for the effect on NVPTX.
AVX changes look like an improvement, too, but I'm not that familiar with the details. We still need someone from the X86 backend to take a look.

https://github.com/llvm/llvm-project/pull/160723