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

[Axel Sorenson via llvm-commits]

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

----------------
axelcool1234 wrote:

Sorry about that. I had these for my own sake when I was testing this against a bunch of kernels. Is there any sort of norm for `LLVM_DEBUG` outputs in AMDGPU so that I can follow it faithfully?

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