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];//128M
char *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];//128M
memset(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 time
    memcpy(cData1k,cData1kCP,1024*sizeof(char));
    clock_gettime(CLOCK_MONOTONIC,&time2);//end time
    usetime1k += (time2.tv_sec-time1.tv_sec)*1000.0+(time2.tv_nsec-time1.tv_nsec)/1000000.0;//ms
    clock_gettime(CLOCK_MONOTONIC,&time1);//start time
    memcpy(cData4k,cData4kCP,1024*4*sizeof(char));
    clock_gettime(CLOCK_MONOTONIC,&time2);//end time
    usetime4k += (time2.tv_sec-time1.tv_sec)*1000.0+(time2.tv_nsec-time1.tv_nsec)/1000000.0;//ms
    clock_gettime(CLOCK_MONOTONIC,&time1);//start time
    memcpy(cData16k,cData16kCP,1024*16*sizeof(char));
    clock_gettime(CLOCK_MONOTONIC,&time2);//end time
    usetime16k += (time2.tv_sec-time1.tv_sec)*1000.0+(time2.tv_nsec-time1.tv_nsec)/1000000.0;//ms
    clock_gettime(CLOCK_MONOTONIC,&time1);//start time
    memcpy(cData512k,cData512kCP,1024*512*sizeof(char));
    clock_gettime(CLOCK_MONOTONIC,&time2);//end time
    usetime512k += (time2.tv_sec-time1.tv_sec)*1000.0+(time2.tv_nsec-time1.tv_nsec)/1000000.0;//ms
    clock_gettime(CLOCK_MONOTONIC,&time1);//start time
    memcpy(cData2M,cData2MCP,1024*1024*2*sizeof(char));
    clock_gettime(CLOCK_MONOTONIC,&time2);//end time
    usetime2M += (time2.tv_sec-time1.tv_sec)*1000.0+(time2.tv_nsec-time1.tv_nsec)/1000000.0;//ms
    clock_gettime(CLOCK_MONOTONIC,&time1);//start time
    memcpy(cData4M,cData4MCP,1024*1024*4*sizeof(char));
    clock_gettime(CLOCK_MONOTONIC,&time2);//end time
    usetime4M += (time2.tv_sec-time1.tv_sec)*1000.0+(time2.tv_nsec-time1.tv_nsec)/1000000.0;//ms
    clock_gettime(CLOCK_MONOTONIC,&time1);//start time
    memcpy(cData8M,cData8MCP,1024*1024*8*sizeof(char));
    clock_gettime(CLOCK_MONOTONIC,&time2);//end time
    usetime8M += (time2.tv_sec-time1.tv_sec)*1000.0+(time2.tv_nsec-time1.tv_nsec)/1000000.0;//ms
    clock_gettime(CLOCK_MONOTONIC,&time1);//start time
    memcpy(cData16M,cData16MCP,1024*1024*16*sizeof(char));
    clock_gettime(CLOCK_MONOTONIC,&time2);//end time
    usetime16M += (time2.tv_sec-time1.tv_sec)*1000.0+(time2.tv_nsec-time1.tv_nsec)/1000000.0;//ms
    clock_gettime(CLOCK_MONOTONIC,&time1);//start time
    memcpy(cData128M,cData128MCP,1024*1024*128*sizeof(char));
    clock_gettime(CLOCK_MONOTONIC,&time2);//end time
    usetime128M += (time2.tv_sec-time1.tv_sec)*1000.0+(time2.tv_nsec-time1.tv_nsec)/1000000.0;//ms
    clock_gettime(CLOCK_MONOTONIC,&time1);//start time
    memcpy(cData500M,cData500MCP,1024*1024*500*sizeof(char));
    clock_gettime(CLOCK_MONOTONIC,&time2);//end time
    usetime500M += (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];//128M
char *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];//128M
memset(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 time
    memcpy(cData1k,cData1kCP,1024*sizeof(char));
    usetime1k += calTime.duration();
    calTime.start();//start time
    memcpy(cData4k,cData4kCP,1024*4*sizeof(char));
    usetime4k += calTime.duration();
    calTime.start();//start time
    memcpy(cData16k,cData16kCP,1024*16*sizeof(char));
    usetime16k += calTime.duration();
    calTime.start();//start time
    memcpy(cData512k,cData512kCP,1024*512*sizeof(char));
    usetime512k += calTime.duration();
    calTime.start();//start time
    memcpy(cData2M,cData2MCP,1024*1024*2*sizeof(char));
    usetime2M += calTime.duration();
    calTime.start();//start time
    memcpy(cData4M,cData4MCP,1024*1024*4*sizeof(char));
    usetime4M += calTime.duration();
    calTime.start();//start time
    memcpy(cData8M,cData8MCP,1024*1024*8*sizeof(char));
    usetime8M += calTime.duration();
    calTime.start();//start time
    memcpy(cData16M,cData16MCP,1024*1024*16*sizeof(char));
    usetime16M += calTime.duration();
    calTime.start();//start time
    memcpy(cData128M,cData128MCP,1024*1024*128*sizeof(char));
    usetime128M += calTime.duration();
    calTime.start();//start time
    memcpy(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