In real code (not just benchmarks), be sure you look at axpy in the context of any other operations you are doing on the arrays. If you are doing other O(n) operations too, then you’ll probably want to fuse these into a single loop with the axpy anyway. If you are also doing Ω(n²) or similarly expensive operations, then aggressively optimizing the O(n) (BLAS-1) stuff is probably not worth it.
2 Likes