SCCS Colloquium - Dec 19, 2019
|Date:||December 19, 2019|
|Time:||15:00 - 16:00|
Bachelor's thesis submission talk. Tobias is advised by Fabio Gratl.
Molecular dynamics simulations are very compute-intensive tasks. In order to speed them up, massive parallelization for the force calculation is necessary. Verlet Lists are a data structure capable of reducing the naively seen quadratic computational complexity to a linear one. A library that solves the force calculation of these simulations is the C++ template library AutoPas [GST+19]. It includes many different approaches for this problem and automatically chooses the fastest one of them for the current state of the system at run time. This thesis extends AutoPas by adding several new approaches. All of them are built on the idea of Verlet Lists and are parallelized using OpenMP 4.5. Small parts of the library are refactored to prepare it for these new algorithms. The new approaches are tested on their Time to Solution and Strong Scaling behavior for different simulation settings on two different platforms. They are compared to an existing solution, and it is shown that they outperform it on many occasions.
Keywords: Verlet Lists, Shared-Memory Parallelizations