Use SSE intrinsics instead of asm() for vertical SSE, finally offloading the register...

Use SSE intrinsics instead of asm() for vertical SSE, finally offloading the register allocation to the compiler. The speed gain is immeasurable, but it sounds like a better idea in general.