Abstract
A principal limitation in accuracy for scientific computation performed with floating-point arithmetic may be traced to the computation of repeated sums, such as those which arise in inner products. We propose the design of a systolic super summer, a cellular piece of hardware for the high throughput performance of repeated sums of floating-point numbers. The apparatus receives pipelined inputs of streams of summands from one or many sources (say as a coprocessor unit in a supercomputer). The floating-point summands are converted into a fixed-point form by a sieve-like pipelined cellular packet-switching device with signal combining. The emerging fixed-point numbers are then summed in a corresponding network of extremely long acumulators (i.e., super accumulators). At the cell level, the design uses a synchronous model of VLSI. The amount of time the apparatus needs to compute an entire sum depends on the values of the summands; at this architectural level, the design is asynchronous. The throughput per unit area of hardware approaches that of a tree network, but without the long wire and signal propagation delay that are intrinsic to tree networks. © 1988 IEEE