A Cache Visualization Tool

• Authors: Eric van der Deijl;Gerco Kanbier;Olivier Temam;Elena D. Granston
• Venue: Computer
• Year: 1997

Cache performance strongly influences the overall performance of software. As a result, researchers continue to use cache simulators to analyze cache performance and optimization. Most cache simulators, however, provide only raw, global information. To improve cache performance, developers must better understand, for example, the impact of software optimizations and the behavior of new hardware cache designs. Cache behavior analysis is a two-step process: First, code sections with poor cache performance must be identified. Second, the causes for poor performance in these code sections must be understood. Cache profilers handle the first task. The authors' Cache Visualization Tool addresses the second task. It thus complements cache profilers. The tool both dynamically visualizes cache content and provides related statistics. A graphical X Windows tool, CVT has a main window that displays a grid representing the cache content. A cache is composed of cache blocks or cache lines (a set of words with consecutive addresses), and each box in the grid corresponds to a cache line. CVT is dedicated to visualizing cache behavior of selected code sections rather than identifying critical code sections. The authors therefore intend to plug the CVT into a profiler similar to CProf that would address the first phase. Furthermore, by collecting information during the profiling run, such as loop boundaries and array subscripts' coefficients, they intend to reduce the number of references that need to be traced.