Automatic generation of parallel C code for stencil applications written in MATLAB
Johannes Spazier, Steffen Christgau, Bettina Schnor · 2016
High-level programming languages such as MATLAB are widely used in scientific domains to implement prototypes based on mathematical models. These prototypes are finally often re-implemented in low-level languages to reach execution times required for the operational use. In order to exploit latest hardware architectures additional effort is necessary to add parallelism to the applications. This paper presents performance results of an automatic translation from a MATLAB subset into efficient parallelized C code for different architectures: multicores, compute clusters, and GPGPUs. We present the first compiler that generates native MPI code from MATLAB source and thereby showing significant performance improvements. The evaluation is done for two stencil applications which use different communication patterns, a Game-of-Life application and a Tsunami simulation. For the Game-of-Life application, the generated parallel code shows nearly optimal speedup. The generated parallel code of the Tsunami simulation reaches the performance of the available parallel reference implementations.