基于硬件性能计数器的软件性能数据采集与分析研究

引入了基于CPU硬件性能计数器的性能数据采集和分析方法，从软件运行时刻的细粒度参数入手分析软件运行时刻的性能表现，从而更为准确地反映系统实际的动态运行状态。实验证明，这种方法对于需要详细掌握系统动态运行状态的应用能够提供非常有效的分析数据，同时也在一定程度上对编译器的性能优化给出了相关参考数据。

Software performance data sampling and analysis based on hardware performance counter

High performance analysis is a major topic in science computation and large-scale software system.Using a better data sampling method will gain better performance analysis.Traditional performance data are sampled by operation system based on rough granularity source code level,such as process time,memory usage percentage,etc.These data cannot show up hypostasis of software and provide a better effective proposition for optimization.This paper introduced a much more precise sampling method using CPU hardware performance counters(CHPC) to provide CPU data like instruction cycles,cache misses,branch prediction,and so on,and given detail scene of software status.With CHPC data,more accuracy software performance analysis and compiler optimization can be reached.