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

[Lukas Sommer via llvm-commits]

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

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

Do you mean to bail the entire DAG mutation or just for this specific load? The latter should be fine, at least from the perspective of not introducing any cycles in the DAG.

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