Establishing an honest scalar baseline before reaching for intrinsics is standard practice in real performance engineering: numerical libraries like Eigen always benchmark against a naive scalar reference to prove a SIMD kernel actually helps. The restrict keyword and -O3 -march=native flags you use here are the exact same levers GCC and Clang's auto-vectorizer relies on before you ever write a single intrinsic.