[llvm] [AMDGPU] Prefer lower total register usage in regions with spilling (PR #71882)

[Jeffrey Byrnes via llvm-commits]

jrbyrnes wrote:

Hi @arsenm , does the latest iteration of this PR resolve your change request? Also, do you have a comment on the excess RP calculation for the unified register file case (see Valery's latest comment). Thanks.

https://github.com/llvm/llvm-project/pull/71882