[llvm] [AMDGPU][AMDGPULateCodeGenPrepare] Combine scalarized selects back into vector selects (PR #173990)

[Matt Arsenault via llvm-commits]

arsenm wrote:

> It also requires AMDGPU-specific cost knowledge - on AMDGPU, a uniform i1 condition doesn't need splatting (uses SCC), making vector select cheap. 

This doesn't depend on SALU or not




https://github.com/llvm/llvm-project/pull/173990