[llvm] [LV] isScalarWithPredication to check for interleaved accesses (PR #194579)

[David Sherwood via llvm-commits]

david-arm wrote:

I've not looked into this, but rather than fixing up the legacy cost model would it be better to simply stop precomputing the costs? This is something I've started doing for divides, e.g. https://github.com/llvm/llvm-project/pull/187056. I think one of the main reasons we'd want to fix up the legacy cost model is if we're making the wrong widening decision for calls or memory operations early on. However, from what you've described we choose CM_Interleave correctly anyway.

https://github.com/llvm/llvm-project/pull/194579