An Experimental Study of Global and Local Search Algorithms in Empirical Performance Tuning

Abstract

The increasing complexity, heterogeneity, and rapid evolution of modern computer architectures present obstacles for achieving high performance of scientific codes on different machines. Empirical performance tuning is a viable approach to obtain high-performing code variants based on their measured performance on the target machine. In previous work, we formulated the search for the best code variant as a numerical optimization problem. Two classes of algorithms are available to tackle this problem: global and local algorithms. We present an experimental study of some global and local search algorithms on a number of problems from the recently introduced SPAPT test suite. We show that local search algorithms are particularly attractive, where finding high-preforming code variants in a short computation time is crucial.Keywordsautomatic performance tuningsearchblack-box optimization