This paper describes several loop splitting methods for exploitingparallelism from single loops, and also proposes a generalized andoptimal loop transformation technique for exploiting parallelism fromsingle loops with nonuniform dependencies. The proposed algorithm isbased on partitioning a serial loop by using the size of dependencedistance such that it varies between different instances of thedependence. It outperforms the two methods proposed by C.D.Polychronopoulos (see Compiler optimizations for enhancing parallelismand their impact on architecture design, IEEE Trans. Comput., vol.37,no.8, p. 991-1004, 1988)
展开▼