摘要：This paper considers the thread scheduling problem. The thread scheduling problem abstracts the problem of minimizing memory latency, using a directed data dependency graph generated form a compiler, to improve run time effciency. Two thread scheduling problems are formulated and shown to be strongly NP-complete. New methods and algorithms for analyzing a data dependency graph in order to compute the theoretical best runtime (lower bound of the finishing time) and to estimate the required minimum number of PEs needed to achieve certain finishing time are presented. The new methods and algorithms improve upon some of the existing analysis and transformation techniques.
Loading...
