[llvm] [AMDGPU] Use correct VGPR threshold for flagging ExcessRP regions in unified register file case (PR #85860)

[Jeffrey Byrnes via llvm-commits]

jrbyrnes wrote:

Latest iteration just uses getTotalNumVGPRs for MaxNumVGPRs and clamps to getAddressableNumArchVGPRs based on WholeRegisterFile as suggested.

Allows for unequal allocation (with a per register type max of 256) and PressureAfter.getVGPRNum(ST.hasGFX90AInsts()) > ST.getMaxNumVGPRs(MF, /*WholeRegisterFile*/ true) captures the unified limit

https://github.com/llvm/llvm-project/pull/85860