使用memcpy函数的耗时测试（拷贝不同大小数据量耗时不同）

今天公司里的一个大神给我普及了一下知识，使用memcpy函数的耗时在拷贝不同大小数据的时候，速度是不一样的，于是我写了个程序测试了一下，具体如下：
目标：比较使用memcpy（）拷贝1k，4k，16k，512k，2M，4M，8M，16M，128M，500M数据的耗时

主要代码如下（编译时会自动区分当前是什么系统）：

#include "mainwindow.h"
#include "ui_mainwindow.h"
#include <QDebug>
#if defined(Q_OS_LINUX)
#include "time.h"
#else
#include <windows.h>
class chronograph
{
public:chronograph(){QueryPerformanceFrequency(&m_freq);QueryPerformanceCounter(&m_bgn);}void start(){QueryPerformanceCounter(&m_bgn);}double duration()
{QueryPerformanceCounter(&m_end);return (m_end.QuadPart - m_bgn.QuadPart) * 1000.0 / m_freq.QuadPart;
}
LARGE_INTEGER now()
{LARGE_INTEGER now;QueryPerformanceCounter(&now);return now;
}
double DoubleNow()
{LARGE_INTEGER now;QueryPerformanceCounter(&now);return now.QuadPart*1000.0 / m_freq.QuadPart;
}
private:LARGE_INTEGER m_freq;LARGE_INTEGER m_bgn;LARGE_INTEGER m_end;
};
#endif
MainWindow::MainWindow(QWidget *parent) :QMainWindow(parent),ui(new Ui::MainWindow)
{ui->setupUi(this);#if defined(Q_OS_LINUX)//在linux下测试memcpy的耗时double usetime1k,usetime4k,usetime16k,usetime512k,usetime2M,usetime4M,usetime8M,usetime16M,usetime128M,usetime500M;usetime1k = 0;usetime4k = 0;usetime16k = 0;usetime512k = 0;usetime2M = 0;usetime4M = 0;usetime8M = 0;usetime16M = 0;usetime128M = 0;usetime500M = 0;char *cData1k = new char[1024];//1k
char *cData4k = new char[1024*4];//4k
char *cData16k = new char[1024*16];//16k
char *cData512k = new char[1024*512];//512k
char *cData2M = new char[1024*1024*2];//2M  //(char*)malloc(1024*1024*1024*2); //
char *cData4M = new char[1024*1024*4];//4M
char *cData8M = new char[1024*1024*8];//8M
char *cData16M = new char[1024*1024*16];//64M
char *cData128M = new char[1024*1024*128];//128M
char *cData500M = new char[1024*1024*500];//128Mchar *cData1kCP = new char[1024];//1k
char *cData4kCP = new char[1024*4];//4k
char *cData16kCP = new char[1024*16];//16k
char *cData512kCP = new char[1024*512];//512k
char *cData2MCP = new char[1024*1024*2];//2M  //(char*)malloc(1024*1024*1024*2);//
char *cData4MCP = new char[1024*1024*4];//4M
char *cData8MCP = new char[1024*1024*8];//8M
char *cData16MCP = new char[1024*1024*16];//64M
char *cData128MCP = new char[1024*1024*128];//128M
char *cData500MCP = new char[1024*1024*500];//128Mmemset(cData1kCP,1,1024);
memset(cData4kCP,1,1024*4);
memset(cData16kCP,1,1024*16);
memset(cData512kCP,1,1024*512);
memset(cData2MCP,1,1024*1024*2);
memset(cData4MCP,1,1024*1024*4);
memset(cData8MCP,1,1024*1024*8);
memset(cData16MCP,1,1024*1024*16);
memset(cData128MCP,1,1024*1024*128);
memset(cData500MCP,1,1024*1024*500);struct timespec time1,time2;
for(int i = 0;i<100;i++)
{clock_gettime(CLOCK_MONOTONIC,&time1);//start timememcpy(cData1k,cData1kCP,1024*sizeof(char));clock_gettime(CLOCK_MONOTONIC,&time2);//end timeusetime1k += (time2.tv_sec-time1.tv_sec)*1000.0+(time2.tv_nsec-time1.tv_nsec)/1000000.0;//msclock_gettime(CLOCK_MONOTONIC,&time1);//start timememcpy(cData4k,cData4kCP,1024*4*sizeof(char));clock_gettime(CLOCK_MONOTONIC,&time2);//end timeusetime4k += (time2.tv_sec-time1.tv_sec)*1000.0+(time2.tv_nsec-time1.tv_nsec)/1000000.0;//msclock_gettime(CLOCK_MONOTONIC,&time1);//start timememcpy(cData16k,cData16kCP,1024*16*sizeof(char));clock_gettime(CLOCK_MONOTONIC,&time2);//end timeusetime16k += (time2.tv_sec-time1.tv_sec)*1000.0+(time2.tv_nsec-time1.tv_nsec)/1000000.0;//msclock_gettime(CLOCK_MONOTONIC,&time1);//start timememcpy(cData512k,cData512kCP,1024*512*sizeof(char));clock_gettime(CLOCK_MONOTONIC,&time2);//end timeusetime512k += (time2.tv_sec-time1.tv_sec)*1000.0+(time2.tv_nsec-time1.tv_nsec)/1000000.0;//msclock_gettime(CLOCK_MONOTONIC,&time1);//start timememcpy(cData2M,cData2MCP,1024*1024*2*sizeof(char));clock_gettime(CLOCK_MONOTONIC,&time2);//end timeusetime2M += (time2.tv_sec-time1.tv_sec)*1000.0+(time2.tv_nsec-time1.tv_nsec)/1000000.0;//msclock_gettime(CLOCK_MONOTONIC,&time1);//start timememcpy(cData4M,cData4MCP,1024*1024*4*sizeof(char));clock_gettime(CLOCK_MONOTONIC,&time2);//end timeusetime4M += (time2.tv_sec-time1.tv_sec)*1000.0+(time2.tv_nsec-time1.tv_nsec)/1000000.0;//msclock_gettime(CLOCK_MONOTONIC,&time1);//start timememcpy(cData8M,cData8MCP,1024*1024*8*sizeof(char));clock_gettime(CLOCK_MONOTONIC,&time2);//end timeusetime8M += (time2.tv_sec-time1.tv_sec)*1000.0+(time2.tv_nsec-time1.tv_nsec)/1000000.0;//msclock_gettime(CLOCK_MONOTONIC,&time1);//start timememcpy(cData16M,cData16MCP,1024*1024*16*sizeof(char));clock_gettime(CLOCK_MONOTONIC,&time2);//end timeusetime16M += (time2.tv_sec-time1.tv_sec)*1000.0+(time2.tv_nsec-time1.tv_nsec)/1000000.0;//msclock_gettime(CLOCK_MONOTONIC,&time1);//start timememcpy(cData128M,cData128MCP,1024*1024*128*sizeof(char));clock_gettime(CLOCK_MONOTONIC,&time2);//end timeusetime128M += (time2.tv_sec-time1.tv_sec)*1000.0+(time2.tv_nsec-time1.tv_nsec)/1000000.0;//msclock_gettime(CLOCK_MONOTONIC,&time1);//start timememcpy(cData500M,cData500MCP,1024*1024*500*sizeof(char));clock_gettime(CLOCK_MONOTONIC,&time2);//end timeusetime500M += (time2.tv_sec-time1.tv_sec)*1000.0+(time2.tv_nsec-time1.tv_nsec)/1000000.0;//ms
}
qDebug()<<"memcpy 1k data usetime:"<<usetime1k/100<<"ms,speed = "<<usetime1k*10<<"um/1kbit";
qDebug()<<"memcpy 4k data usetime:"<<usetime4k/100<<"ms,speed = "<<usetime4k/4*10<<"um/1kbit";
qDebug()<<"memcpy 16k data usetime:"<<usetime16k/100<<"ms,speed = "<<usetime16k/16*10<<"um/1kbit";
qDebug()<<"memcpy 512k data usetime:"<<usetime512k/100<<"ms,speed = "<<usetime512k/512*10<<"um/1kbit";
qDebug()<<"memcpy 2M data usetime:"<<usetime2M/100<<"ms,speed = "<<usetime2M/1024/4*10<<"um/1kbit";
qDebug()<<"memcpy 4M data usetime:"<<usetime4M/100<<"ms,speed = "<<usetime4M/1024/4*10<<"um/1kbit";
qDebug()<<"memcpy 8M data usetime:"<<usetime8M/100<<"ms,speed = "<<usetime8M/1024/8*10<<"um/1kbit";
qDebug()<<"memcpy 16M data usetime:"<<usetime16M/100<<"ms,speed = "<<usetime16M/1024/16*10<<"um/1kbit";
qDebug()<<"memcpy 128M data usetime:"<<usetime128M/100<<"ms,speed = "<<usetime128M/1024/128*10<<"um/1kbit";
qDebug()<<"memcpy 500M data usetime:"<<usetime500M/100<<"ms,speed = "<<usetime500M/1024/500*10<<"um/1kbit";
#else//在windows下测试memcpy()的耗时chronograph calTime;double usetime1k,usetime4k,usetime16k,usetime512k,usetime2M,usetime4M,usetime8M,usetime16M,usetime128M,usetime500M;usetime1k = 0;usetime4k = 0;usetime16k = 0;usetime512k = 0;usetime2M = 0;usetime4M = 0;usetime8M = 0;usetime16M = 0;usetime128M = 0;usetime500M = 0;
char *cData1k = new char[1024];//1k
char *cData4k = new char[1024*4];//4k
char *cData16k = new char[1024*16];//16k
char *cData512k = new char[1024*512];//512k
char *cData2M = new char[1024*1024*2];//2M  //(char*)malloc(1024*1024*1024*2); //
char *cData4M = new char[1024*1024*4];//4M
char *cData8M = new char[1024*1024*8];//8M
char *cData16M = new char[1024*1024*16];//64M
char *cData128M = new char[1024*1024*128];//128M
char *cData500M = new char[1024*1024*500];//128Mchar *cData1kCP = new char[1024];//1k
char *cData4kCP = new char[1024*4];//4k
char *cData16kCP = new char[1024*16];//16k
char *cData512kCP = new char[1024*512];//512k
char *cData2MCP = new char[1024*1024*2];//2M  //(char*)malloc(1024*1024*1024*2);//
char *cData4MCP = new char[1024*1024*4];//4M
char *cData8MCP = new char[1024*1024*8];//8M
char *cData16MCP = new char[1024*1024*16];//64M
char *cData128MCP = new char[1024*1024*128];//128M
char *cData500MCP = new char[1024*1024*500];//128Mmemset(cData1kCP,1,1024);
memset(cData4kCP,1,1024*4);
memset(cData16kCP,1,1024*16);
memset(cData512kCP,1,1024*512);
memset(cData2MCP,1,1024*1024*2);
memset(cData4MCP,1,1024*1024*4);
memset(cData8MCP,1,1024*1024*8);
memset(cData16MCP,1,1024*1024*16);
memset(cData128MCP,1,1024*1024*128);
memset(cData500MCP,1,1024*1024*500);for(int i = 0;i<100;i++)
{calTime.start();//start timememcpy(cData1k,cData1kCP,1024*sizeof(char));usetime1k += calTime.duration();calTime.start();//start timememcpy(cData4k,cData4kCP,1024*4*sizeof(char));usetime4k += calTime.duration();calTime.start();//start timememcpy(cData16k,cData16kCP,1024*16*sizeof(char));usetime16k += calTime.duration();calTime.start();//start timememcpy(cData512k,cData512kCP,1024*512*sizeof(char));usetime512k += calTime.duration();calTime.start();//start timememcpy(cData2M,cData2MCP,1024*1024*2*sizeof(char));usetime2M += calTime.duration();calTime.start();//start timememcpy(cData4M,cData4MCP,1024*1024*4*sizeof(char));usetime4M += calTime.duration();calTime.start();//start timememcpy(cData8M,cData8MCP,1024*1024*8*sizeof(char));usetime8M += calTime.duration();calTime.start();//start timememcpy(cData16M,cData16MCP,1024*1024*16*sizeof(char));usetime16M += calTime.duration();calTime.start();//start timememcpy(cData128M,cData128MCP,1024*1024*128*sizeof(char));usetime128M += calTime.duration();calTime.start();//start timememcpy(cData500M,cData500MCP,1024*1024*500*sizeof(char));usetime500M += calTime.duration();
}
qDebug()<<"memcpy 1k data usetime:"<<usetime1k/100<<"ms,speed = "<<usetime1k*10<<"um/1kbit";
qDebug()<<"memcpy 4k data usetime:"<<usetime4k/100<<"ms,speed = "<<usetime4k/4*10<<"um/1kbit";
qDebug()<<"memcpy 16k data usetime:"<<usetime16k/100<<"ms,speed = "<<usetime16k/16*10<<"um/1kbit";
qDebug()<<"memcpy 512k data usetime:"<<usetime512k/100<<"ms,speed = "<<usetime512k/512*10<<"um/1kbit";
qDebug()<<"memcpy 2M data usetime:"<<usetime2M/100<<"ms,speed = "<<usetime2M/1024/4*10<<"um/1kbit";
qDebug()<<"memcpy 4M data usetime:"<<usetime4M/100<<"ms,speed = "<<usetime4M/1024/4*10<<"um/1kbit";
qDebug()<<"memcpy 8M data usetime:"<<usetime8M/100<<"ms,speed = "<<usetime8M/1024/8*10<<"um/1kbit";
qDebug()<<"memcpy 16M data usetime:"<<usetime16M/100<<"ms,speed = "<<usetime16M/1024/16*10<<"um/1kbit";
qDebug()<<"memcpy 128M data usetime:"<<usetime128M/100<<"ms,speed = "<<usetime128M/1024/128*10<<"um/1kbit";
qDebug()<<"memcpy 500M data usetime:"<<usetime500M/100<<"ms,speed = "<<usetime500M/1024/500*10<<"um/1kbit";
#endif
}MainWindow::~MainWindow()
{delete ui;
}

在Tx2设备上运行结果如下：

可以看到拷贝数据在2M大小的时候，速度达到峰值。为什么拷贝2M数据速度最快呢，因为Tx2设备的处理器二级缓存就是2M，参数如下图：

在Thinkpad T570上运行的结果如下：

可以看到也是在拷贝2M数据的时候速度最快，CPU为i7-7500U，是3级缓存，大小为4M

使用memcpy函数的耗时测试（拷贝不同大小数据量耗时不同）相关推荐

使用memcpy函数时要注意拷贝数据的长度
memcpy函数简介 memcpy函数是C/C++语言中的一个用于内存复制的函数,声明在 string.h 中(C++是 cstring).其原型是: void *memcpy(void *desti ...
C语言之memcpy函数
昨天自己动手实现memcpy这个函数,用一个例程试了一下,结果正确,满心欢心,可是有些地方想不明白,于是百度了一下,结果自己写的函数简直无法直视. 觉得还是写个总结,以示教训. 先贴上我自己的函数: ...
linux memcpy需要头文件,memcpy函数详解
说明: 1.source和destin所指内存区域不能重叠,函数返回指向destin的指针. 2.如果目标数组destin本身已有数据,执行memcpy()后,将覆盖原有数据(最多覆盖n).如果要追加 ...
C语言memcpy函数和memmove函数
memcpy函数和memmove函数都是C语言的库函数,作用都是从一个空间copy一定长度的内容到另一个空间,不同的是memcpy只是单纯的拷贝,当两个空间发生局部重叠时无法保证拷贝的正确(正不正确取 ...
tensflower官方测试案例_大数据性能测试介绍
点击上方蓝色字体,选择"设为星标" 回复"资源"获取更多资源大数据技术与架构点击右侧关注,大数据开发领域最强公众号! 大数据真好玩点击右侧关注,大数据真好玩! ...
关于某些系统统计查询等业务进行大数据量的测试
大数据量测试分为两种:一种是针对某些系统存储.传输.统计查询等业务进行大数据量的测试:另一种是与并发测试相结合的极限状态下的综合数据测试.如专项的大数据量测试主要针对前者,后者尽量放在并发测试中.此外 ...
C语言 memcpy函数(增加dst 从第一字节拷贝判断)的内部简单实现方式
刚学习C语言编程的时候,感觉C内部函数深不可测,通过代码很难实现.后来经常慢慢的学习过程,理解了部分C函数的实现方法,下面简单说下memcpy函数的内部实现方式之一. 简单概括内存拷贝的含义:字节对齐 ...
c语言内存拷贝 memcpy()函数
需包含头文件:C 标准库 - <string.h> 文章目录描述声明参数返回值实例1 实例2 实例3:覆盖原有部分数据描述 C 库函数 void *memcpy(void *s ...
让你提前认识软件开发(8)：memset()与memcpy()函数
第1部分重新认识C语言 memset()与memcpy()函数软件项目中的代码,少则数千行,多则上万行,但并非无规律可循.阅读了那么多代码之后,我发现有两个函数出现的频率很高,那就是memset( ...
memcpy函数优化及DMA对比
一.背景与目的优化算法结构,提高芯片的使用效率,挖掘芯片的潜在能力,对提高产品质量,降低产品成本有着重要意义,在性能受限的嵌入式设备更加重要. 在使用C语言编程时,我们常用memcpy来复制内存数据 ...

使用memcpy函数的耗时测试（拷贝不同大小数据量耗时不同）

使用memcpy函数的耗时测试（拷贝不同大小数据量耗时不同）相关推荐

最新文章

热门文章