Mapping techniques for parallel evaluation of chains of recurrences

This paper examines the parallelization of a technique for speeding up the evaluation of potentially-complex real-valued functions at a large number of points. The technique being parallelized generates a chain of recurrences (CR) which is then used to compute the function incrementally (i.e., by using the results of one iteration in calculating the value of the function in the next iteration). This paper examines the possibilities for mapping the CR technique onto parallel machines. The factors influencing the choice of mapping alternatives include efficiency, speedup, and most interestingly, the potential for improved error distributions.