Abstract:
This paper explores the way that simple but inherently parallel programs may be adapted to a
parallel vector computer. Starting with an example algorithm that is applied independently to
many sets of data, we proceed to convert it systematically into a highly parallel form by using
multiprocessing and vector processing. The various versions of the program are written in
Fortran for the Cray X-MP. Performance data are collected to show the extent of the
performance improvement that can be expected.