[llvm] [AMDGPU] Improve MemorySSA walking in isClobberedInFunction (PR #211817)

[Krzysztof Drewniak via llvm-commits]

krzysz00 wrote:

> It seems like isClobberedInFunction is only complicated because we're trying to apply extra AA knowledge after running MemorySSA. It would surely be better to run MemorySSA using an AA that already includes that extra knowledge.

How feasible is it to fix this?

https://github.com/llvm/llvm-project/pull/211817