Vectorization Experiment on the IBM 3090 Vector Facility

A comparison between sequential code and vectorization code for matrix multiplication is made and the possible ways to vectorize matrix multiplication are examined. The result is that vectorization of the correct loop results in a speedup of almost 3 for vectors that are long enough, i.e., longer than or equal to half the vector length of the machine. The number and order of memory references is a very important factor.