[llvm] AMDGPU: Allow operand folding between loop body and its preheader (PR #137022)

[Jon Chesterfield via llvm-commits]

JonChesterfield wrote:

I thought this was in the category of stuff we can't really do because we've conflated exec mask mutation with the LLVM CFG. When is this optimisation sound?

https://github.com/llvm/llvm-project/pull/137022