An Experimental Evaluation of the OpenMP Thread Mapping for LU Factorisation on Xeon Phi Coprocessor and on Hybrid CPU-MIC Platform

Abstract

Efficient thread mapping relies upon matching the behaviour of the application with system characteristics. The main aim of this paper is to evaluate the influence of the OpenMP thread mapping on the computation performance of the matrix factorisations on Intel Xeon Phi coprocessor and hybrid CPU-MIC platforms. The authors consider parallel LU factorisations with and without pivoting, both from MKL (Math Kernel Library) library. The results show that the choice of thread affinity, the number of threads and the execution mode have a measurable impact on the performance and the scalability of the LU factorisations.