[llvm] [SCEV] Add initial matchers for SCEV expressions. (NFC) (PR #119390)

[Florian Hahn via llvm-commits]

fhahn wrote:

> This looks reasonable, but I can't claim to understand the template magic going on here. I do wonder if it wouldn't be better to just have separate unary and binary implementations and not try to be overly generic.

Most of the complexity is coming from a missing utility to iterate over all members of a tuple. Happy to change it to a non-generic version. My original thought was that it might be convenient to match SCEV expressions with arbitrary number of arguments, but that would probably not be very common.

https://github.com/llvm/llvm-project/pull/119390