Context-sensitive Parametric WCET Analysis

Abstract

In this paper, we propose aWCET analysis that focuses on two aspects. First, it supports contextsensitive hardware and software timing effects, meaning that it is sensitive to the execution history of the program and thus can account for effects like cache persistence, triangular loop, etc. Second, it supports the introduction of parameters in both the software model (e.g. parametric loop bounds) and the hardware model (e.g. number of cache misses). WCET computation by static analysis is traditionally handled by the Implicit Path Enumeration Technique (IPET), using an Integer Linear Program (ILP) that is difficult to resolve parametrically. We suggest an alternative tree-based approach. We define a context-sensitive CFG format to express these effects, and we provide an efficient method to process it, giving a parametric WCET formula. Experimental results show that this new method is significantly faster and more accurate than existing parametric approaches.