Aspect-oriented programming of sparse matrix code

The expressiveness conferred by high-level and object-oriented languages is often impaired by concerns that cross-cut a program's basic functionality. Execution time, data representation, and numerical stability are three such concerns that are of great interest to numerical analysts. Using aspect-oriented programming we have created AML, a system for sparse matrix computation that deals with these concerns separately and explicitly while preserving the expressiveness of the original functional language. The resulting code maintains the efficiency of highly tuned low-level code, yet is ten times shorter.