[llvm] [AMDGPU] fix a scc clobber bug (PR #193143)

[Jay Foad via llvm-commits]

jayfoad wrote:

> > > IMHO, not relying on `SplitBB` makes sense to me since `SplitBB` is only one of the known sources of this problem so extending it to a more general case might make check more robust.
> > 
> > Are there other known sources of this problem?
> 
> Before RegAlloc with SSA, that's the only place I know.

Then I think fixing it in (or near) `splitKillBlock` is better than adding a whole new pass.

I'm still not sure if there are other ways of fixing it, e.g. somehow persuading SelectionDAG not to leave SCC live across a KILL pseudo.

https://github.com/llvm/llvm-project/pull/193143