Abstract Interpretation of FIFO Replacement

Interpretation of FIFO Replacement Daniel Grund and Jan Reineke Saarland University, Saarbrücken, Germany Abstract. In hard real-time systems, the execution time of programs must be bounded by static timing analysis. For today’s embedded systems featuring caches, static analyses must predict cache hits and misses with high precision to obtain useful bounds. For caches with least-recentlyused (LRU) replacement policy, efficient and precise cache analyses exist. However, for other widely-used policies like first-in first-out (FIFO), current cache analyses are much less precise. This paper discusses challenges in FIFO cache analysis and advances the state of the art. We identify a generic framework for cache analysis that couples mayand must-analyses by means of domain cooperation. Our main contribution is a more precise may-analysis for FIFO. It not only increases the number of predicted misses, but also—due to the domain cooperation—the number of predicted hits. We instantiate the framework with a canonical must-analysis and three different may-analyses, including our new one, and compare the resulting three analyses to the collecting semantics. Our evaluation results characterize the progress achieved by our new may-analysis and reveal room for further improvement. In hard real-time systems, the execution time of programs must be bounded by static timing analysis. For today’s embedded systems featuring caches, static analyses must predict cache hits and misses with high precision to obtain useful bounds. For caches with least-recentlyused (LRU) replacement policy, efficient and precise cache analyses exist. However, for other widely-used policies like first-in first-out (FIFO), current cache analyses are much less precise. This paper discusses challenges in FIFO cache analysis and advances the state of the art. We identify a generic framework for cache analysis that couples mayand must-analyses by means of domain cooperation. Our main contribution is a more precise may-analysis for FIFO. It not only increases the number of predicted misses, but also—due to the domain cooperation—the number of predicted hits. We instantiate the framework with a canonical must-analysis and three different may-analyses, including our new one, and compare the resulting three analyses to the collecting semantics. Our evaluation results characterize the progress achieved by our new may-analysis and reveal room for further improvement.