[llvm] [InterleavedLoadCombine] Index candidates to avoid quadratic matching (PR #213053)

[Madhur Amilkanthwar via llvm-commits]

madhur13490 wrote:

> This pass is wild, I had not looked into it before. It would be good if it just used SCEV instead of the polynomials it produces, but they would not work as well if the SCEVs are not canonicalized the same way.

Yes, turning them to SCEV can be another separate effort. 

https://github.com/llvm/llvm-project/pull/213053