为什么是多线程慢？ [英] Why is multithreaded slower?

问题描述

所以我想写一个找到素数的程序。该项目的真正目的是了解多线程。首先我写了一个单线程程序，它在1分钟内找到高达13,633,943。

So I am trying to write a program that finds prime numbers. The real purpose of the project is just to learn about multithreading. First I wrote a single thread program and it finds up to 13,633,943 in 1 minute. My multithreaded version only got to 10,025,627.

这是我的单线程程序代码

Here is my code for the single threaded program

#include <iostream>

using namespace std;

bool isprime(long num)
{
    long lim = num/2;
    if(num == 1)
    {
        return 0;
    }
    for(long i = 2; i <= lim; i++)
    {
        if (num % i == 0)
        {
            return 0;
        }
        else{ lim = num/i; }
    }
    return 1;
}

int main()
{
    long lim;
    cout << "How many numbers should I test: ";
    cin >> lim;
    for(long i = 1; i <= lim || lim == 0; i++)
    {
        if(isprime(i))
        {
            cout << i << endl;
        }
    }
}

多线程程序。

extern"C"
{
    #include <pthread.h>
    #include <unistd.h>
}
#include <iostream>

using namespace std;

bool isprime(long num);
void * iter1(void * arg);
void * iter2(void * arg);
void * iter3(void * arg);
void * iter4(void * arg);


int main()
{
    //long lim;
    //cout << "How many numbers should I test: ";
    //cin >> lim;
    pthread_t t1;
    char mem1[4096];//To avoid false sharing. Needed anywhere else?
    pthread_t t2;
    char mem2[4096];//These helped but did not solve problem.
    pthread_t t3;
    pthread_create(&t1, NULL, iter1, NULL);
    pthread_create(&t2, NULL, iter2, NULL);
    pthread_create(&t3, NULL, iter3, NULL);
    iter4(0);
}

bool isprime(long num)
{
    long lim = num/2;
    if(num == 1)
    {
        return 0;
    }
    for(long i = 2; i <= lim; i++)
    {
        if (num % i == 0)
        {
            return 0;
        }
        else{ lim = num/i; }
    }
    return 1;
}

void * iter1(void * arg)
{
    for(long i = 1;; i = i + 4)
    {
        if(isprime(i))
        {
            cout << i << endl;
        }
    }
return 0;
}

void * iter2(void * arg)
{
    for(long i = 2;; i = i + 4)
    {
        if(isprime(i))
        {
            cout << i << endl;
        }
    }
return 0;
}

void * iter3(void * arg)
{
    for(long i = 3;; i = i + 4)
    {
        if(isprime(i))
        {
            cout << i << endl;
        }
    }
return 0;
}

void * iter4(void * arg)
{
    for(long i = 4;; i = i + 4)
    {
        if(isprime(i))
        {
            cout << i << endl;
        }
    }
return 0;
}

令我特别困惑的是，系统监视器报告25％的CPU使用率单线程和多线程的100％使用。

Something that especially confuses me is that system monitor reports 25% CPU usage for the single thread and 100% usage for the multithread. Shouldn't that mean it is doing 4 times as many calculation?

推荐答案

我相当肯定 cout 充当共享资源 - 即使它实际上按正确的顺序正确打印每个数字，它也会减慢很多事情。

I'm fairly sure cout acts a shared resource - and even if it actually prints each number correctly and in the right order, it slows things down VERY much to do so.

我做了类似的事情（它更灵活，并使用原子操作来选择下一个数字），它在我的四核机器上几乎快了4倍。但是，只有当我不打印任何东西。如果它打印到控制台，这是一个很慢 - 因为很多时间是使用shuffling像素而不是实际计算。

I have done something similar (it is more flexible, and uses an atomic operation to "pick the next number"), and it's almost exactly 4x faster on my quad core machine. But that's only if I don't print anything. If it prints to the console, it's a lot slower - because a lot of the time is used shuffling pixels rather than actually calculating.

注释 cout< i<< endl; 行，它会运行得更快。

Comment out the cout << i << endl; line, and it will run much quicker.

编辑：使用我的测试程序，打印：

using my test program, with printing:

Single thread: 15.04s. 
Four threads: 11.25s

不打印：

Single threads: 12.63s.
Four threads: 3.69s.

3.69 * 4 = 14.76秒，但时间命令在我的Linux机器上显示12.792秒的总运行时间，所以显然有一点时间，当所有线程都不运行 - 或一些会计错误...

3.69 * 4 = 14.76s, but the time command on my Linux machine shows 12.792s total runtime, so there is obviously a little bit of time when all threads aren't running - or some accounting errors...

这篇关于为什么是多线程慢？的文章就介绍到这了，希望我们推荐的答案对大家有所帮助，也希望大家多多支持IT屋！