Cyclic architectures for concurrent processing applications

An optimal algorithm could be defined as one that exploits the properties of the available technology to the full. That is, it takes account of the inadequacies in an implementation and exploits the merits. For example, when multiplication was a very slow operation, algorithms aimed to reduce the number of multiplications needed. The radix-2 fast Fourier transform is an example of this type of algorithm. Similarly, number-theoretic transforms sought to replace multiplications with shifts and additions: operations that could be performed much faster within the constraints of the then existing technology. The advent of VLSI technology has created a fundamental change in the meaning of algorithm optimisation. In VLSI systems, global communication is very difficult but multiplications and additions are fast. The structure of the computation is the most important factor determining throughput. A simple computational structure is demanded. Algorithms must exhibit recursiveness and replication and they must require only local communications for a good VLSI implementation.<
>