On the Inference of Resource Usage Upper and Lower Bounds

摘要

Cost analysis aims at determining the amount of resources required to run a program in terms of its input data sizes. The most challenging step is to infer the cost of executing the loops in the program. This requires bounding the number of iterations of each loop and finding tight bounds for the cost of each of its iterations. This article presents a novel approach to infer upper and lower bounds from cost relations. These relations are an extended form of standard recurrence equations that can be nondeterministic, contain inexact size constraints and have multiple arguments that increase and/or decrease. We propose novel techniques to automatically transform cost relations into worst-case and best-case deterministic one-argument recurrence relations. The solution of each recursive relation provides a precise upper-bound and lower-bound for executing a corresponding loop in the program. Importantly, since the approach is developed at the level of the cost equations, our techniques are programming language independent.