Vectorization is the action of converting (either manually or automatically) a program processing some kind of data (here a unicorn), into a program processing a vector of these data.


Recent processors can do arithmetic operations on very large data (up to 512 bits for the Xeon Phi), but the largest datatype in C is 64 bits : therefore each non-vectorized operation loses at least 448 bits. The idea is to merge as many data as possible in this 512 vector, and then do a single operation on it instead of multiple operations on almost empty vectors.
As shown on the picture, it's fastest to add up the 8 unicorn parts with a vectorized program !