A class of low complexity high concurrence algorithms
J.L. Aravena, A.O. Barbir · IEEE Transactions on Parallel and Distributed Systems · 1991
A nonconventional approach to the analysis of dedicated computing structures in which the number of compute cycles is used as a design parameter to determine families of transformations implementable in the structure is presented. Using this approach, a single architecture can be used to implement a family of transformations with varying degrees of complexity. The transformations generated by a matrix multiplication array are considered in detail. It is shown that, for some real-time applications it becomes possible to incorporate the compute time as a constraint for designs based in optimality criteria. In particular, a least square approximation problem is discussed.>