A MATLAB-Based Code Generator for Sparse Matrix Computations

Abstract. We present a matrix language compiler CMC which translates annotated MATLAB scripts into Fortran 90 programs. Distinguishing features of CMC include its applicability to programs with sparse matrix computations and its capability of source-level optimization in MATLAB language. Different from other existing similar translators, CMC has an ability to generate codes based on information on the shape of matrices such as triangular and diagonal. Integrating these functionalities, CMC provides the user with a simple way to develop fast large-scale numerical computation codes beyond prototyping. Experimental results show that the programs of SOR and CG methods generated by CMC can run several times as fast as the original MATLAB scripts.

• Type: Conference
• Year: 2004
• Where: APLAS
• Authors: Hideyuki Kawabata, Mutsumi Suzuki, Toshiaki Kitamura
• Keywords: APLAS 2004, Language Compiler Cmc, MATLAB Scripts, Sparse Matrix Computations