Such loops can be written in terms of BLAS primitives-Intel
will then provide you with hand coded routines that will work now-
and in the future when the next process comes out.

If you don’t want intel code go through ATLAS, or some similar
library which abstracts out such simple constructs…

or otherwise buy an autovectorizing compiler- again several
on the market with gcc catching up.
http://gcc.gnu.org/projects/tree-ssa/vectorization.html