The paper presents a nouvelle static heuristic-based algorithm for mapping precedence programs to a processor network. The algorithm deals with any precedence graph model of a program and maps it to any processor network. It aims at minimizing the total processing cost of the program by minimizing the inter-module communication cost. It extracts the parallelism of a graph confined with precedence constraints, produces maximum number of linear clusters and maps all clusters to processors based on the connectivity of processors and the connections between clusters. The algorithm has been applied to allocate various graph models to different processor network topologies |
