Linux下操纵CPU曲线绘制心形

不久之前看了「编程之美」，里面有在windows下操纵CPU绘制正弦曲线的示例程序。思路很简单，但是需要知道几个windows的API函数。刚开始我想尝试在windows下绘制心形，不过没能做到，原因是CPU曲线是CPU利用率和时间构成的函数的关系，心形是一个封闭图形，不可能是一个函数，所以至少需要两条CPU曲线才能绘制（这需要你的机器为双核或多核或者多个CPU），然而windows的任务管理器，我没能找到将两条CPU曲线绘制在同一图表下的方法，所以没能成功。不过ubuntu下的任务管理器可以，而且还可以设置CPU曲线的颜色。以下是我在ubuntu下任务管理器的截图，我把两条CPU曲线都设置成了红色，设置CPU曲线的刷新速率为2s。

绘制正弦曲线的程序思路很简单，在看过「编程之美」后，完全可以仿照写出Linux版本，只是引用的库和调用的API函数不通而已。书中没有双核的参考程序，要让特定的线程在特定的cpu上运行需要设置cpu亲和度，windows下可参考这篇博客 http://blog.csdn.net/wesweeky/article/details/6402564

我本来想用经典的心形曲线X^2+(y-(x^2)^(1/3))^2=1来绘制，但是效果不是很好，所以干脆使用几个分段的二次函数。详细代码如下：

/*Programmed by Wang Zaijing 2011-08-01*/
#include <stdio.h>
#include <pthread.h>
#include <unistd.h>

#include <stdlib.h>
#include <math.h>
#include <time.h>
#include <sys/time.h>

#define SAMPLE_COUNT 800  //设置800个抽样点
#define INCREAMENT 0.0025 //增量为0.0025=2/800
#define SAMPLE_TIME 25    //设置每个抽样点的时间片


double GetSysTime()       //Windows中直接使用GetTickSAMPLE_COUNT()函数./ms
{
    struct timeval tv;
    gettimeofday(&tv, NULL);
    return (tv.tv_sec*1000+tv.tv_usec*1.0/1000);
}

int GetCpuNums()
{
    return (int)sysconf(_SC_NPROCESSORS_ONLN);
}

void *thread_cpu1(void*arg)          //线程函数1
{
    double busy_cache[SAMPLE_COUNT]; //CPU占用时间，缓存一个周期的计算结果
    double idle_cache[SAMPLE_COUNT]; //CPU空闲时间

    double radian = 0.0;             //增量从0开始，最大值到2=SAMPLE_COUNT*INCREAMENT

    double start_time;               //开始时间

    double temp;
    //一个周期内为分段函数分为四段
    for(int i=0; i<SAMPLE_COUNT/4; i++)
    {
        temp = (double)(-8.0*(radian-0.25)*(radian-0.25)+1);
        busy_cache[i] = (double)(SAMPLE_TIME*temp);
        idle_cache[i] = (double)(SAMPLE_TIME - busy_cache[i]);
        radian += INCREAMENT;
    }

    for(int i=SAMPLE_COUNT/4; i<SAMPLE_COUNT/2; i++)
    {
        temp = (double)(-8.0*(radian-0.75)*(radian-0.75)+1);
        busy_cache[i] = (double)(SAMPLE_TIME*temp);
        idle_cache[i] = (double)(SAMPLE_TIME - busy_cache[i]);
        radian += INCREAMENT;
    }

    for(int i= SAMPLE_COUNT/2; i < 3*SAMPLE_COUNT/4; i++)
    {
        temp = (double)(0.667*(radian-2)*(radian-2)-0.167);
        busy_cache[i] = (double)(SAMPLE_TIME*temp);
        idle_cache[i] = (double)(SAMPLE_TIME - busy_cache[i]);
        radian += INCREAMENT;
    }

    for(int i= 3*SAMPLE_COUNT/4; i < SAMPLE_COUNT; i++)
    {
        temp = (double)(0.667*(radian-1)*(radian-1)-0.167);
        busy_cache[i] = (double)(SAMPLE_TIME*temp);
        idle_cache[i] = (double)(SAMPLE_TIME - busy_cache[i]);
        radian += INCREAMENT;
    }

    for(int i = 0; ; i = (i + 1) % SAMPLE_COUNT)
    {
        start_time = GetSysTime();
        while((GetSysTime()-start_time) <= busy_cache[i]);
        usleep(idle_cache[i]*1000);
    }
    return NULL;
}
void *thread_cpu2(void*arg)//线程函数2，原理同线程函数1
{
    double busy_cache[SAMPLE_COUNT];
    double idle_cache[SAMPLE_COUNT];

    double radian = 0.0;

    double start_time;


    double temp;


    for(int i=0; i<SAMPLE_COUNT/4; i++)
    {

        temp = (double)(0.667*(radian-1)*(radian-1)-0.167);
        busy_cache[i] = (double)(SAMPLE_TIME*temp);
        idle_cache[i] = (double)(SAMPLE_TIME - busy_cache[i]);
        radian += INCREAMENT;
    }

    for(int i= SAMPLE_COUNT/4; i < SAMPLE_COUNT/2; i++)
    {
        temp = (double)(0.667*(radian)*(radian)-0.167);
        busy_cache[i] = (double)(SAMPLE_TIME*temp);
        idle_cache[i] = (double)(SAMPLE_TIME - busy_cache[i]);
        radian += INCREAMENT;
    }

    for(int i=SAMPLE_COUNT/2; i<3*SAMPLE_COUNT/4; i++)
    {
        temp = (double)(-8.0*(radian-1.25)*(radian-1.25)+1);
        busy_cache[i] = (double)(SAMPLE_TIME*temp);
        idle_cache[i] = (double)(SAMPLE_TIME - busy_cache[i]);
        radian += INCREAMENT;
    }

    for(int i=3*SAMPLE_COUNT/4; i<SAMPLE_COUNT; i++)
    {
        temp = (double)(-8.0*(radian-1.75)*(radian-1.75)+1);
        busy_cache[i] = (double)(SAMPLE_TIME*temp);
        idle_cache[i] = (double)(SAMPLE_TIME - busy_cache[i]);
        radian += INCREAMENT;
    }

    for(int i = 0; ; i = (i + 1) % SAMPLE_COUNT)
    {
        start_time = GetSysTime();
        while((GetSysTime()-start_time) <= busy_cache[i]);
        usleep(idle_cache[i]*1000);
    }
    return NULL;
}

int main()//主函数
{
    //int cpu_num = 0;
    //cpu_num  = GetCpuNums();
    //printf("The number of cpu is %d\n", cpu_num);

    pthread_t t1;
    pthread_t t2;
    pthread_attr_t attr1;
    pthread_attr_t attr2;

    pthread_attr_init(&attr1);
    pthread_attr_init(&attr2);

    if (0!=pthread_create(&t1, &attr1, thread_cpu1, NULL))//创建线程1
    {
        printf("error:Create thread 1\n");
        return 0;
    }

    if (0!=pthread_create(&t2, &attr2, thread_cpu2, NULL))//创建线程2
    {
        printf("error:Create thread 2\n");
        return 0;
    }

    cpu_set_t cpu_info;
    CPU_ZERO(&cpu_info);
    CPU_SET(0, &cpu_info);
    if (0!=pthread_setaffinity_np(t1, sizeof(cpu_set_t), &cpu_info))
    {
        printf("error:Thread 1 set affinity failed.\n");
    }

    CPU_ZERO(&cpu_info);
    CPU_SET(1, &cpu_info);
    if (0!=pthread_setaffinity_np(t2, sizeof(cpu_set_t), &cpu_info))
    {
        printf("error:Thread 2 set affinity failed.\n");
    }

    pthread_join(t1, NULL);//释放线程t1
    pthread_join(t2, NULL);//释放线程t2
    return 0;
}

总结几点：

Linux下的cpu亲和度 CPU 亲和性（affinity）就是进程要在某个给定的 CPU 上尽量长时间地运行而不被迁移到其他处理器的倾向性。Linux 内核进程调度器天生就具有被称为软 CPU 亲和性（affinity）的特性，这意味着进程通常不会在处理器之间频繁迁移，windows则不然，如果你的电脑是双核，你会看到两条cpu曲线近似相同，进程几乎被两个cpu均分。 Linux下的cpu亲和度可以参考以下博文，讲的非常清楚 http://www.ibm.com/developerworks/cn/linux/l-affinity.html
终端编译出错：undefined reference to’pthread_reate’ 加上-lpthread 如：gcc test.c –lpthread 。若使用IDE Code::Blocks 编译出错则要进行以下设置： Project -> Build options -> Linker settings -> Link libraries 里加上pthread
Happy Birthday to「某*」. This is just for you.

【参考文献】《编程之美》

【参考博客】管理处理器的亲和性 http://www.ibm.com/developerworks/cn/linux/l-affinity.html

【参考博客】让CPU占用率曲线听你指挥 http://blog.csdn.net/wesweeky/article/details/6402564

【参考博客】Linux下pthread的线程亲和性研究 http://blog.sina.com.cn/s/blog_6a1837e90100nrlj.html

王小兵 / 2011-08-03
Published under (CC) BY-NC-SA in categories C&C++ tagged with C语言 Linux CPU曲线 CPU亲和性编程之美