SIMD is where the remaining single-core performance is: ffmpeg, x264, BLAS, and simdjson get their 4-16x wins from exactly the packed instructions you use here. Auto-vectorization often fails, so engineers at these projects still write intrinsics by hand — knowing SSE basics is the entry requirement.