[llvm] [LoopVectorize][LAA] Hoist load in memory IV to allow vectorization (PR #168312)

[Florian Hahn via llvm-commits]

https://github.com/fhahn commented:

Thanks for the update. it looks like there are a number of  unrelated/uneeded changes which woud be good ot remove to reduce the diff.

I am wondering if it would make sense to implement this as separate transform, as it seems like this may be useful also in some cases we don't vectorize? We already have LoopVersioningLICM which does something similar, although I don't think it would hoist the load + sink the store yet.

https://github.com/llvm/llvm-project/pull/168312