Module filtering preprocessing for module assignment problems in pipelined computing

In this paper, we extend previous works on the problem of mapping pipelined (chain-like) computation modules onto a set of (chain-like) linear array processors. We show how this problem can be solved more effectively by pre-filtering out those bad partition points, which can be determined by the proposed forward-backward filtering scheme. By eliminating those infeasible (bad) partition points, some modules can be merged virtually and thus the total number of modules to be considered is reduced. Consequently, search speed is accelerated Meanwhile, the sub-chain cost matrix, formed by the new module set after above filtering and merging scheme, can be proved being well (lattice-like) ordered. This elegant property avoids the need of matrix rearrangement while applying the two dimension binary search method to find optimal solutions.