For-loop equivalents and massive parallelization in CasADi

Joris Gillis

KU Leuven, Belgien

Tuesday, July 19, 2016, 11:45

Room 01-012, Georges-Köhler-Allee 102, Freiburg 79110, Germany



In system identification, long horizons may degrade performance of CasADi when solving an optimal control method using a direct approach.

This talk describes how the 'map' and 'mapaccum' nodes can be used to avoid costly for-loops and memory overhead for this application.

Further, the map abstraction is shown to lead to easy parallelization capabilities.

A second application, optimization in image processing, is used to show massive speed-ups with the GPU within the CasADi framework.