[llvm] [LV] Peek through bitcasts when performing CSE (PR #146856)

[Florian Hahn via llvm-commits]

https://github.com/fhahn commented:

Does this have any real-world impact with the byte type?

I'm curious because the case here looks very spefic: we need an interleave group, which stores an invariant value with the vector loop unrolled.

This isn't a case we can easily catch with VPlan CSE, because it is part of VPInterleaveRecipe::execute, and can't easily be broken out.

This will make it harder to replace the legacy CSE with VPlan-based CSE, and I don't think we should keep both around in parallel.


https://github.com/llvm/llvm-project/pull/146856