我实际上写了一个样本应用程序就是这么做的.作为一个参数,我给出了我想要检查的数字,以及要使用的线程数(每个线程都有一个相同大小的数字范围,试图将p除以 – 它们一起覆盖整个范围).

我的机器有8个核心.我开始使用大量的程序运行程序,我知道它是素数(2971215073),并且有1,2,3个线程等,直到达到8个线程 – 每次程序运行速度比前一个快,这是我的预期.但是,当我尝试大于8的数字时,计算时间实际上变得越来越小(即使是一点点)！

在我的线程中没有I / O或类似的东西,只是纯粹的cpu计算.当我通过8个线程时,我预计运行时间会变得更糟,因为会有更多的上下文切换,并行运行线程的数量保持在8个.很难说峰值在哪里因为差异非常小而且变化很大从一次运行到另一次运行,但很明显,即50个线程以某种方式运行速度超过8(约300毫秒)……

我的猜测是因为我有这么多线程,所以我得到更多的运行时间,因为我在系统的线程池中有更大的部分,所以我的线程被选中更多.但是,我创建的线程越多,程序运行得越快似乎没有意义(否则为什么不是每个人都创建1000个线程？).

任何人都可以提供一个解释,也许是一个最好的做法,关于创建相对于机器上的核心数量的线程数？

谢谢.

我感兴趣的人的代码(在Windows上编译,VS2012)：

#include <Windows.h>
#include <conio.h>
#include <iostream>
#include <thread>
#include <vector>

using namespace std;

typedef struct
{
    unsigned int primeCandidate;
    unsigned int rangeStart;
    unsigned int rangeEnd;
} param_t;


DWORD WINAPI isDivisible(LPVOID p)
{
    param_t* param = reinterpret_cast<param_t*>(p);

    for (unsigned int d = param->rangeStart; d < param->rangeEnd; ++d)
    {
        if (param->primeCandidate % d == 0)
        {
            cout << param->primeCandidate << " is divisible by " << d << endl;
            return 1;
        }
    }

    return 0;
}

bool isPrime(unsigned int primeCandidate,unsigned int numOfCores)
{
    vector<HANDLE> handles(numOfCores);
    vector<param_t> params(numOfCores);
    for (unsigned int i = 0; i < numOfCores; ++i)
    {
        params[i].primeCandidate = primeCandidate;
        params[i].rangeStart = (primeCandidate - 2) * (static_cast<double>(i) / numOfCores) + 2;
        params[i].rangeEnd = (primeCandidate - 2) * (static_cast<double>(i+1) / numOfCores) + 2;
        HANDLE h = CreateThread(nullptr,reinterpret_cast<LPTHREAD_START_ROUTINE>(isDivisible),&params[i],0);
        if (NULL == h)
        {
            cout << "ERROR creating thread: " << GetLastError() << endl;
            throw exception();
        }
        handles[i] = h;
    }

    DWORD ret = WaitForMultipleObjects(numOfCores,&handles[0],TRUE,INFINITE);
    if (ret >= WAIT_OBJECT_0 && ret <= WAIT_OBJECT_0 + numOfCores - 1)
    {
        for (unsigned int i = 0; i < numOfCores; ++i)
        {
            DWORD exitCode = -1;
            if (0 == GetExitCodeThread(handles[i],&exitCode))
            {
                cout << "Failed to get thread's exit code: " << GetLastError() << endl;
                throw exception();
            }

            if (1 == exitCode)
            {
                return false;
            }
        }

        return true;
    }
    else
    {
        cout << "ERROR waiting on threads: " << ret << endl;
        throw exception();
    }
}

int main()
{
    unsigned int primeCandidate = 1;
    unsigned int numOfCores = 1;

    cout << "Enter prime candidate: ";
    cin >> primeCandidate;
    cout << "Enter # of cores (0 means all): ";
    cin >> numOfCores;
    while (primeCandidate > 0)
    {
        if (0 == numOfCores) numOfCores = thread::hardware_concurrency();

        DWORD start = GetTickCount();
        bool res = isPrime(primeCandidate,numOfCores);
        DWORD end = GetTickCount();
        cout << "Time: " << end-start << endl;
        cout << primeCandidate << " is " << (res ? "" : "not ") << "prime!" << endl;

        cout << "Enter prime candidate: ";
        cin >> primeCandidate;
        cout << "Enter # of cores (0 means all): ";
        cin >> numOfCores;
    }

    return 0;
}