[llvm] AMDGPU: Shrink extracted dword vector loads (PR #203484)

[Matt Arsenault via llvm-commits]

https://github.com/arsenm requested changes to this pull request.

Please push https://github.com/llvm/llvm-project/pull/122671 to completion before adding some custom handling like this. I'd also prefer to get your VectorCombine patch done first as well. The DAG should be a last resort for patterns which only appear as a consequence of legalization 

https://github.com/llvm/llvm-project/pull/203484