When observing program performances in practice, one could be faced to non-negligible variations, even if program input is fixed. When considering multiple code versions issued from many program transformation chains, it is not easy to select the best code version in presence of these performance variations.

In this seminar, we will tackle some rigorous statistical methods for selecting a practically most efficient code by choosing appropriate metrics of performances. We will recall some basis on statistical testing, the most popular ones, and explain why they could not be always helpful. Then we show a parametric approach for performance analysis and comparison and explain its difference with a non-parametric one. This seminar is devoted to people willing to use rigorous statistics to take cautious decisions to declare if a program is more efficient than others.

This research effort has been conducted with Julien Worms, an associate professor in mathematics at University of Versailles Saint Quentin en Yvelines.

Kurzbiographie von Professor Touati:
Sid Touati is a professor at University Nice Sophia Antipolis in France (faculty of science). He was a former associate professor at the University of Versailles Saint-Quentin en Yvelines. His research interests include formal and technical methods for code optimization and analysis for high performance and embedded processors, compilation and code generation, parallelism, statistics and code performance optimization. He has got a habilitation thesis on backend code optimization, a PhD thesis on register optimization and a master diploma in computer science. His research activities are conducted in "Institut National de Recherche en Informatique et Automatisme (INRIA)" as well as in "Centre National de Recherche Scientifique (CNRS)". He is the main author of the book Advanced Backend Code Optimization (ISTE, Wiley, 2014). ( <link www-sop.inria.fr/members/Sid.Touati _blank>http://www-sop.inria.fr/members/Sid.Touati</link> )