Designing efficient algorithms for APL: A case study of a hydraulic-network solver

The interpretation time of APL instructions significantly influences the total execution time and therefore must be considered during the development of APL programs. As an example of the trade off between interpretation and execution time, the redesign process of a hydraulic network solver originally intended for implementation in FORTRAN is described. Efficiency is gained both by designing algorithms which have a more parallel nature and by extensive use of precomputations. It is shown that evaluating an algorithm using the classical method of counting the number of "basic steps" is insufficient in an APL context.