[llvm] [AMDGPU] Split vgpr regalloc pipeline (PR #93526)

[Christudasan Devadasan via llvm-commits]

cdevadas wrote:

> > It depends on #92809. The downstream compiler would break some applications without this dependent patch.
> 
> It would really be better if it didn't. Did the other implementation depend on it as well?

The regalloc strategy changes and noted some tests (part of the extended testing) caused the assertion in SplitKit while trying to recompute the insertion point for COPY at the BB prologue. Yes, the same behavior was observed with the other implementation too.

https://github.com/llvm/llvm-project/pull/93526