Study And Implementation Of Parallel Program Performance Analysis

Performance analysis is critical for tuning, optimizing and algorithm modifying of parallel program. For these purpose, performance analysis tools are widely used to expose behaviour of running program, to find bottleneck of the whole program, and to guide users performing optimization. When implementing a performance analysis tool, data collection is remarkable part, for it has very important influence on the accuracy and completeness of the tool.In this paper, we focus on the main technologies on performance analysis, and present a event-based model for parallel program. Then, a performance analysis tool called AutoPAN (Automatic Performance ANalyzer) is designed and implemented. AutoPAN visualizes parallel program execution and qualifies performance features to users.We use hardware counters, which are special registers in processors, to perform performance data collection. And PAPI is what we use to access those counters.