A Generalized Directive-Based Approach for Accelerating PDE Solvers

Abstract

Cube-Flu is a Python software application that produces Fortran code for solving Partial Differential Equations (PDEs), according to the input provided by the user. The code produced by Cube-Flu is designed for exploiting distribuited memory architectures as well as Graphics Processing Units, as shown in the next section. The software solves equations of the form $\frac{\partial{\bf u}}{\partial{t}}=f(\bf u)$ on cartesian grids, using Runge-Kutta time integration, and finite difference schemes. The idea behind the application is to provide a simple framework for solving a wide class of systems of equations, using a natural and intuitive syntax.