 |
||
|
||
Home > TERATEC FORUM > Workshop 6
Algorithms and parallel software
Performance modeling and optimization of a Lagrange-Remap algorithm on multicore processors
Thibault GASC, Maison de la Simulation, CEA/DAM/DIF and CMLA
Performance modeling aims at building simple analytical models which allow to predict the performance of an algorithm on a target computing architecture from the knowledge of the main features of both the algorithm and the processor.
Such models provide a deep understanding of the performance of an algorithm and help to identify suitable optimization strategies. Performance models are robust and stable tools, since they are derived from algorithm study and are not based on one of its specific implementation. They can also be seen as decision-making tools.
We will present recent results of the performance study of Lagrange-Remap algorithm for compressible gas dynamics obtained using Roofline and ECM models.
 |
Thibault Gasc is a PhD student in Applied Mathematics at Maison de la Simulation and CEA DAM DIF. He graduated as an engineer from École Centrale de Nantes in 2013. His research interests focus on developing efficient Hydrodynamics solver for High Performance Computing systems. Performance modeling is used to characterise HPC properties of numerical methods in order to optimise them, highlight their computing limits and build more efficient numerical methods. |
