[llvm] [AMDGPU] Eliminate unnecessary packing in wider f16 vectors for sdwa/opsel-able instruction (PR #137137)

[Vikash Gupta via llvm-commits]

vg0204 wrote:

I finally tried to incorporate my patch into existing APIs setup by SIPeepholeSDWA's infrastructure. It would be great if you guys can review it from a fresh perspective now. 

CC : @jayfoad , @krzysz00 , @frederik-h , @cdevadas , @arsenm 

https://github.com/llvm/llvm-project/pull/137137