ABSTRACT
In §15 I mentioned several ways to reduce the wall-clock running time of a numerically-intensive program, including improving the algorithm, using compiler optimization, and tuning the FORTRAN source code. Another approach that can sometimes be used, along with those methods or by itself, is to divide the problem into pieces and do them all at the same time, using separate electronic circuits for each piece of the calculation. At first glance this might seem like a hare-brained scheme, but two different ways have in fact been devised to make concurrent processing practical for a wide variety of technical computing problems [94].
