Another point is that the CPU can sequence the multiple calls to its internal SIMD unit internally without that having to be done by user code. This in extreme

Another point is that the CPU can sequence the multiple calls to its internal SIMD unit internally without that having to be done by user code. This in extreme case degrades to the Cray-1-like vector unit, which still has measurable preformance impact and can be implementated even in very resource constrained environments.