Hybrid data dependence analysis for loop transformations

Abstract

Loop optimizations span from vectorization, scalar promotion, loop invariant code motion, software pipelining to loop fusion, skewing, tiling [6, Ch.9] and loop parallelization. These transformations are essential in the quest for automated high-performance code generation.