C++ code design for multi-purpose explicit finite volume methods

Abstract

The Ernst-Mach-Institute (EMI) of the Fraunhofer-Society is dealing with a wide spectrum of subjects in the fields of applied physics, mechanical and civil engineering. The EMI department for numerical simulation supports the institute and external customers with high-performance software applications in the fields of compressible flows, structural dynamics, electro dynamics and multi-disciplinary couplings of these applications. The majority of the in-house codes are written in FORTRAN 95. Earlier performance studies of numerical simulations in terms of explicit finite element methods have shown that FORTRAN provides much better efficiency than C++. Nevertheless, potential benefits of object-oriented programming in C++ were recognized. Here we analyze the factors contributing to the code performance for the explicit finite volume scheme and show that C++ provides at least the same efficiency as FORTRAN by application of the new techniques.

Generic C++ Implementation of High-Performance BFS-RBF-based Mesh Motion Schemes

“…We applied the concept-bounded (static) polymorphism to achieve a better performance and maximal software reuse. The objects CELL, FACE and NODE were designed in a manner similar to [12]. All vector properties of the mentioned objects like position, velocity, list of the grid entities adjacent to the selected object like cell etc.…”

“…if As soon as the distances from the user-defined source vertex (line 6) to all mesh nodes are recorded (lines 7-8) and the total number of the BFS waves is known (lines 9-11), the sequences of the node indices are inserted to the corresponding wave (lines [12][13][14]. After spliting the node indices of each BFS wave into two groups (to constraints and approximation points, respectively), the iterative BFS-RBF interpolation begins.…”