Implementation of a parallel algorithm for searching the global extremum of a function on Intel Xeon Phi

A parallel algorithm for solving multiextremal optimization problems is proposed. An implementation of the algorithm on modern computing systems using Intel Xeon Phi coprocessors is examined. Two approaches to algorithm parallelization are discussed with consideration of the available information on the computational cost for computing a given objective function. A number of numerical results obtained on a Lobachevsky supercomputer are analyzed. It is shown that the implementation of the algorithm using Xeon Phi is more efficient than that using CPU only. Computational experiments confirm this conclusion.