An evolutionary approach for improving the cache usage of FORTRAN codes

Recent years have seen the development of complex multidisciplinary computer applications, where software libraries can play an instrumental role in achieving high performance rates on state-of-the-art computers. Static performance analyses based on source code observation and linker generated maps may not be practical in such a context. We evaluate the potential of some evolutionary search techniques to identify optimal layouts of data in memory. Experiments are done using simulated traces generated using an ecosystem modeling application.