How to Benchmark Code Execution Times on Intel
®
IA-32
and IA-64 Instruction Set Architectures
7
2 Problem Description
This section explains the issues involved in reading the timestamp register and
discusses the correct methodology to return precise and reliable clock cycles
measurements. It is expected that readers have knowledge of basic GCC and ICC
compiling techniques, basic Intel assembly syntax, and AT&T* assembly syntax.
Those not interested in the problem description and method justification can skip
this section and go to Section 3.2.1
(if their platform supports the RDTSCP
instruction) or Section 3.2.3
(if not) to acquire the code.
2.1 Introduction
Intel CPUs have a timestamp counter to keep track of every cycle that occurs on
the CPU. Starting with the Intel Pentium
®
processor, the devices have included a
per-core timestamp register that stores the value of the timestamp counter and
that can be accessed by the RDTSC and RDTSCP assembly instructions.
When running a Linux OS, the developer can check if his CPU supports the RDTSCP
instruction by looking at the flags field of “/proc/cpuinfo”; if rdtscp is one of the
flags, then it is supported.
2.2 Problems with RDTSC Instruction in C Inline
Assembly
Assume that you are working in a Linux environment, and are compiling by using
GCC. You have C code and want to know how many clock cycles are spent to
execute the code itself or just a part of it. To make sure that our measurements
are not tainted by any kind of interrupt (including scheduling preemption), we are
going to write a kernel module where we guarantee the exclusive ownership of the
CPU when executing the code that we want to benchmark.
To understand the practical implementation, let’s consider the following dummy
kernel module; it simply calls a function that is taking a pointer as input and is
setting the pointed value to “1”. We want to measure how many clock cycles it
takes to call such a function:
#include <linux/module.h>
#include <linux/kernel.h>
#include <linux/init.h>
#include <linux/hardirq.h>
#include <linux/preempt.h>
#include <linux/sched.h>
void inline measured_function(volatile int *var)
{
(*var) = 1;
剩余36页未读，继续阅读
bsdnewbie
- 粉丝: 0
- 资源: 4
我的内容管理
展开
最新资源
- C语言快速排序算法的实现与应用
- KityFormula 编辑器压缩包功能解析
- 离线搭建Kubernetes 1.17.0集群教程与资源包分享
- Java毕业设计教学平台完整教程与源码
- 综合数据集汇总：浏览记录与市场研究分析
- STM32智能家居控制系统：创新设计与无线通讯
- 深入浅出C++20标准：四大新特性解析
- Real-ESRGAN: 开源项目提升图像超分辨率技术
- 植物大战僵尸杂交版v2.0.88：新元素新挑战
- 掌握数据分析核心模型，预测未来不是梦
- Android平台蓝牙HC-06/08模块数据交互技巧
- Python源码分享：计算100至200之间的所有素数
- 免费视频修复利器：Digital Video Repair
- Chrome浏览器新版本Adblock Plus插件发布
- GifSplitter：Linux下GIF转BMP的核心工具
- Vue.js开发教程：全面学习资源指南
资源上传下载、课程学习等过程中有任何疑问或建议，欢迎提出宝贵意见哦~我们会及时处理！
点击此处反馈
