SC²S Colloquium - March 24, 2017
|Date:||March 24, 2017|
|Time:||3:00 pm, s.t.|
Michael Lettrich: Parallel Incomplete Cholesky Decomposition for Data Mining with Sparse Grids
Fabio Gratl: Implementation and Evaluation of Task-based Approaches for Molecular Dynamics Simulations using Quicksched and ls1-mardyn =
Manuel Fasching: JIT compilation to realize flexible data access in simulation software
Input data or scenarios of simulation applications are frequently defined by mathematical functions. These functions are often implemented in the application’s code and are, hence, statically compiled. The dependency between scenario implementations and the actual application causes an adaption of the application’s code for every new scenario. The primary goal of this thesis is the development of a methology, based on Just-in-time compiling, capable of dissolving this dependency and enhancing the flexibility of data initialization. The absence of an easy but flexible solution was identified as the primary issue causing this dependency. A library capable of Just-in-time compiling scenarios was developed in order to solve it. This library was integrated in the simulation applications Shallow Water Equations and SeisSol. Tests validated the proper behavior and the aptitude for simulation applications of the developed library. The results show that the flexibility of data initialization can be improved by JIT compiling without causing unacceptable drawbacks.