-
Cédric Bastoul
Loop Optimization: A Matter of Art and Science
18 juin 2026 - 09:00Salle de conférences IRMA
Abstract: Despite the massive adoption of parallel architectures in modern electronic devices, parallel programming is still an expert's job. (SemAutomatic parallelization techniques are amongst the main hopes to bridge the gap between low parallel programming expertise and high and ubiquitous parallel architecture complexity. The polyhedral model has been at the root of many advances in the automatic parallelization and optimization field during the last decade. It is an algebraic representation of some classes of compute-intensive loops which allows us to analyze them in depth and to manipulate them aggressively. In this talk, I will present an introduction to the polyhedral model and how this formal representation complements the art of parallel programming with a science of loop transformation. I will review some ongoing academic and industrial work to rely on this model to integrate parallel programmer expertise to compilers and to design new tools to ease efficient parallel programming.
Speaker: Cedric Bastoul is full professor at the university of Strasbourg, associated with the Parallel Computing team at ICube laboratory and Inria. His research interests are in compiler technologies for optimization and parallelization with a strong emphasis on code restructuring using the polyhedral model. He created or maintains several well known tools in the high-level compilation community as the code generator CLooG and the parametric integer programming solver PIP. Recently, he served as Chief Scientist of Distributed and Parallel Software and director of Central Software Institute Paris at Huawei, and as Director at Qualcomm's Compiler Labs, heading R&D programs and working on generative AI software optimization.