A Survey on Vectorization on Intel Xeon Phi Coprocessor: Technique for Performance Optimization

In computer science, vectorization is the process of converting an algorithm from a scalar implementation to a vector process. It does an operation on all the pairs of operands stored in SIMD registers at a time. This is totally different from task parallelism using MPI, OpenMP or alternative parallel libraries wherever extra cores or nodes are added to take care of information belonging to separate tasks placed on completely different cores or nodes. It adds a form of parallelism to software. It is a well-known technique for performance optimization. It makes the full use of the features provided by the hardware for parallelism. We discuss two vectorization techniques in this paper which are The Loop Unrolling technique and the Two Way Vectorization technique. The aim of this work is to study the work done until now on vectorization and its related techniques for performance optimization.