[llvm] [AMDGPU] DAG Mutation to solve load bunching in outer product matrix multiplications (PR #203095)

[Lukas Sommer via llvm-commits]

================

----------------
sommerlukas wrote:

I don't think we have fixed set of rules written down somewhere. Probably just look around in other backend passes or the CoExec scheduler to get a feel. 

In this example, I'd limit to the information that you'd need if you wanted to follow the flow through the code, i.e., the information like `Wmmas.size()` etc. printed below, but would skip the long text commentary above. 

https://github.com/llvm/llvm-project/pull/203095