On the use of static branch prediction to reduce the worst-case execution time of real-time applications

Abstract

Nowadays, real-time applications need more and more hardware performance. This increasing performance demands the use of deterministic performance enhancement features such as static branch prediction. In this paper we propose a new technique which aims to use static branch prediction for worst-case execution time (WCET) reduction that can be applied on any processor that supports this type of prediction. The only requirement is the support of a WCET tool. This paper also describes how to estimate the maximum WCET reduction that can be obtained with static approaches. We show that our technique produces a slightly better result than a similar approach from the literature. We also compare WCET-centered techniques against standard compiler techniques not directly oriented to WCET reduction. We show that a very small or even no gain can be obtained with new techniques targeted to WCET reduction considering static branch prediction. That means the techniques considered in this paper are close to an optimal result. As a secondary contribution, we show that non WCET-aware techniques can also be used in real-time environments because they present good results and low complexity. We evaluate the prediction techniques using a set of examples from the Malardalen WCET benchmarks.