[llvm] [LV][AArch64] Prefer Fixed over Scalable if cost-model is equal (Neoverse V2) (PR #95819)

[David Sherwood via llvm-commits]

david-arm wrote:

I do see issues with generating the addresses in the SVE version definitely and we should fix that, but if we assume those issues are fixed and we have a fair comparison with GCC do we still see the SVE loop being slower?

https://github.com/llvm/llvm-project/pull/95819