Help understanding vectorization (or lack thereof)

General Usage Performance
16

Try --math-mode=fast, then. It automatically SIMD-ed in both cases for me.