Autovectorizing is quite a long way away from being as or more performant from using intrinsics or asm. This makes sense when you think about it. The compiler

Autovectorizing is quite a long way away from being as or more performant from using intrinsics or asm. This makes sense when you think about it. The compiler would either need to be extremely capable of generalizing about some kinds of arithmetic, or it would need millions of special cases to recognize. By writing your own vectorization, you are basically covering your singular special case on your own.