Critical Path Heuristic for Automatic Parallelization

This paper presents an algorithm to heuristically partition the data dependence graph representing a basic block or a hyper block of code in order to assign instructions to available processing cores. The algorithm relies on the observation that the critical path in the data dependence graph defines the lower bound on the schedule length of a block and hence the instructions on the critical path should be assigned to the same core. The algorithm has been implemented in the Trimaran compiler framework. The given heuristic algorithm is much simpler and evaluation shows that it is as good as previous state-of-the-art heuristic algorithms.