Using MAQAO to Analyse and Optimise an Application

One of the major issues in the performance analysis of HPC codes is the difficulty to fully and accurately characterize the behaviour of an application. In particular, it is essential to precisely pinpoint bottlenecks and their true causes. Additionally, providing an estimation of the possible gain obtained after fixing a particular bottleneck would surely allow for a more thorough choice of the optimizations to apply or avoid. In this paper, we present MAQAO, a performance analysis framework aiming at providing a comprehensive human-friendly view of performance issues and also guide the user’s optimization efforts on the most promising performance bottlenecks.