[llvm] [AMDGPU] Allow hoisting of V_CMP feeding into SI_IF and SI_IF_BREAK (PR #71401)

[Carl Ritson via llvm-commits]

https://github.com/perlfu commented:

This looks reasonable.

I guess we could end up with "worse" code gen. for SI_IF_BREAK in some edge cases.
If the V_CMP is hoisted then an extra S_AND will be added.
Although the S_AND will probably have lower latency than a V_CMP?

https://github.com/llvm/llvm-project/pull/71401