摘要
在龙芯2F平台上设计并实现性能调优工具Perf。分析Perf的实现原理,针对龙芯修改Perf内核层中体系相关部分代码,设计硬件性能计数器的操作函数集和溢出中断处理函数。使用Perf分析矩阵相乘测试用例,结果表明Perf得到的采样结果接近理论值,与Oprofile相比,采样数据更精确。
This paper designs and implements Perf on Loongson 2F.Perf is a tuning tool on linux.It analyses the implementation of Perf,and modifies the codes related to hardware architecture.It designs functions to control the performance counter in Loongson 2F and handle the overflow interruptions.The results of matrix multiplication sampled by Perf are close to the theoretical values,and more accurate than these sampled by Oprofile.
出处
《计算机工程》
CAS
CSCD
北大核心
2011年第19期236-238,共3页
Computer Engineering
基金
国家科技重大专项基金资助项目(2009ZX01036-001-002)
