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

[Lukas Sommer via llvm-commits]

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

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

Due to SmallVector's semantics, you can also just do: 
```
unsigned WMMAPos = It - Wmmas.begin();
```

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