A loop transformation for maximizing parallelism from single loops with nonuniform dependencies

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