A design for a fast computer for scientific calculations

Abstract

Recently developed techniques, such as the associative fast store and Tomasulo's algorithm, will enable typical large scale computers to achieve 15 to 20 million instructions per second. The hardware of such machines has a very much greater potential power, but it is inefficiently used, being limited to decoding a single instruction per logic cycle. This paper proposes a technique whereby the programmer is provided with complex instructions capable of controlling the operation of the whole machine during one logic cycle. The use of such instructions for the inner loops of programs yields substantial performance improvements without significantly increased costs.