c++ - 这个线程池/多核模拟有什么问题

Question

大家好。我一直在尝试制作一种线程池，用于模拟多核处理器，其中我有许多线程一直在运行(核心)，我稍后会分派(dispatch)它们来处理一个(暂时固定)功能。让线程始终运行的想法是我没有线程创建/销毁开销。我现在做的事情存在三个问题。首先，结果都是错误的。二、测时函数报0ms第三，程序在退出时调用 abort。这是我正在使用的代码:

    auto fakeExpensiveOperation = [](int i) -> float
    {
        Sleep(10);
        return sqrt(i);
    };

    // Performance test
    int size = 4000;
    float* out = new float[size];

#define THREAD_RUNNING 0
#define THREAD_FINISHED (1 << 0)
#define THREAD_EXIT (1 << 1)
#define THREAD_STALL (1 << 2)

    const int coreCount = 8;
    thread cores[coreCount];
    atomic<unsigned int> msgArray[coreCount]; for (auto& msg : msgArray) msg.store(THREAD_STALL);

    auto kernel = [out, &fakeExpensiveOperation](int idx){ out[idx] = fakeExpensiveOperation(idx); };
    for (int i = 0; i < coreCount; i++)
    {
        cores[i] = thread([&msgArray, i, kernel]()
        {
            while (true)
            {
                unsigned int msg = msgArray[i].load();

                if((msg & THREAD_STALL) == THREAD_STALL)
                    continue;

                if ((msg & THREAD_EXIT) == THREAD_EXIT)
                    break;

                if ((msg & THREAD_RUNNING) == THREAD_RUNNING)
                {
                    int idx = (msg >> 3) + i;

                    // Do the function
                    kernel(idx);

                    msgArray[i].store(THREAD_FINISHED);
                }
            }
        });
    }

    auto t2 = time_call([&]()
    {
        for (int i = 0; i < size; i += coreCount)
        {
            for (int n = 0; n < coreCount; n++)
            {
                if((msgArray[n].load() & THREAD_RUNNING) == THREAD_RUNNING) continue; // The core is still working

                unsigned int msg = THREAD_RUNNING;
                msg |= (i << 3);
                msgArray[n].store(msg);
            }
        }
    });
    for (int n = 0; n < coreCount; n++) msgArray[n].store(THREAD_EXIT);

    cout << "sqrt 0 : " << out[0] << endl;
    cout << "sqrt 1 : " << out[1] << endl;
    cout << "sqrt 2 : " << out[2] << endl;
    cout << "sqrt 4 : " << out[4] << endl;
    cout << "sqrt 16 : " << out[16] << endl;

    cout << "Parallel : " << t2 << endl;
    system("pause");
    delete[] out;
    return 0;

我真的没脑子了。谁能指出这里出了什么问题？

编辑:我做了我提到的更改，但仍然得到错误的值。我更改了标志的值，并在创建线程后将其分离。

Answer 1

我可能是错的，(我对 C++11 线程的东西不是很熟悉)但看起来你正在运行一个线程 cores[i] = thread([&msgArray, i, kernel]( ) 然后等待该线程完成。然后创建下一个线程。本质上使其成为单线程。

我也喜欢用它来为 C++11 计时

std::chrono::time_point<std::chrono::system_clock> start, end;

start = std::chrono::system_clock::now();

// Do Stuff

end = std::chrono::system_clock::now();
std::chrono::duration<double> elapsed_seconds_d1 = end-start;
std::time_t end_time_d1 = std::chrono::system_clock::to_time_t(end);
std::cout << "elapsed time: " << elapsed_seconds_d1.count() << "s\n";