A Path-Driven Loop Scheduling Mapped onto Generalized Hypercubes

One of the important issues in automatic code parallelization is the scheduling and mapping of nested loop iterations to different processors. The optimal scheduling problem is known to be NP-complete. Many heuristic static and dynamic loop scheduling techniques have been studied in the past. Here we propose a new static loop scheduling heuristic method called path-driven scheduling, under the assumption that the loop dependence graph has been generated. This method clusters tasks according to the directed paths on the dependence graph and assigns them to processors in the target architecture. We make comparisons with the free scheduling and the refined free scheduling algorithms [8]. We schedule three widely used nested loops on a generalized hypercube architecture. Our algorithm exhibits the lowest communication cost compared to the other two algorithms, while the execution cost is the same for all three algorithms.