作者: Erik Berg , Erik Hagersten
关键词: Operating system 、 Locality 、 Smart Cache 、 Cache 、 Profiling (computer programming) 、 Cache algorithms 、 Computer science 、 Embedded system 、 Cache coloring
摘要: Performance tools based on hardware counters can efficiently profile the cache behavior of an application and help software developers improve its utilization. Simulator-based potentially provide more insights flexibility model many different configurations, but have drawback large run-time overhead.We present StatCache, a performance tool statistical model. It has small overhead while providing much simulator-based tools. A monitor process running in background collects sparse memory access statistics about analyzed natively host computer. Generic locality information is derived presented code-centric and/or data-centric view.We evaluate accuracy using ten SPEC CPU2000 benchmarks. We also exemplify how be used to better understand characteristics cache-related problems.