Question

我需要帮助。我试图做一些具体的事情，而我缺乏多线程技能正在扼杀我。

基本上我的主程序/线程需要管理一些＆＃34;频道＆＃34;必须运行多次。由于这些运行是独立的，因此每个通道都包含一个执行它们的线程。

因此主线程必须等待所有通道（线程）完成运行才能启动下一个通道。并且所有渠道都必须等待主线程发出的可以运行的通知。

以下是我的表现 - 抱歉有点长！

#include <thread>
#include <mutex>
#include <condition_variable>
#include <iostream>
#include <atomic>

std::mutex              g_lockprint;
std::mutex              g_lockbatch;
std::condition_variable g_nextbatch;
std::mutex              g_lockready;
std::condition_variable g_ready;

int global_id = 0;
int nbChannels = 5;
std::atomic<int> nbChannelsLeftToEnd;

class Channel {

private:

    int _id;
    std::thread _th;
    std::atomic<bool> next_batch;
    std::atomic<bool> stop_th;

public:

    Channel() : _id(global_id++), _th(), next_batch(false), stop_th(false) {}

    void go_for_next_batch() { next_batch = true; }

    void start(int& start, int &end){
        _th = std::thread(&Channel::run, this, std::ref(start), std::ref(end));
    }

    void stop(){
        stop_th = true;
        _th.join();
    }

    void run(int& start, int& end){
        while (!stop_th){
            {
                std::unique_lock<std::mutex> locker(g_lockbatch);
                g_nextbatch.wait(locker, [&](){return (next_batch==true); });
            }

            // print a starting message
            {
                std::unique_lock<std::mutex> locker(g_lockprint);
                std::cout << "[channel " << _id << "]\trunning in [" << start << "," << end << "]" << std::endl;
            }

            // simulate work
            std::this_thread::sleep_for(std::chrono::seconds(1));

            // update the number of channels left to run
            nbChannelsLeftToEnd--;
            g_ready.notify_one();
            next_batch = false;
        }
    }
};

int main()
{
    int end = 100;
    int batch = 10;
    int startBatch = 0;
    int endBatch = startBatch + batch;

    // declare some channels (threads)
    std::vector<Channel> channels(nbChannels);

    // start the threads
    for (auto& ch : channels) ch.start(startBatch, endBatch);

    while (endBatch<=end){
        {
            std::unique_lock<std::mutex> locker(g_lockprint);
            std::cout << "[main]\trunning in [" << startBatch << "," << endBatch << "]" << std::endl;
        }
        nbChannelsLeftToEnd = nbChannels;
        for (auto& ch : channels) ch.go_for_next_batch();
        g_nextbatch.notify_all();

        std::unique_lock<std::mutex> locker(g_lockready);
        g_ready.wait(locker, [&](){return (nbChannelsLeftToEnd == 0); });

        startBatch += batch;
        endBatch += batch;
    }

    for (auto& ch : channels) ch.stop();

    return 0;
}

但有时程序阻塞，可能是线程在彼此等待，但我不明白为什么。在任何情况下，加入线程（＆＃34;停止＆＃34;主要结束时的方法）使我的程序无限期运行，不知道为什么。

编辑：感谢您的评论和一些研究，我设法使用同步屏障获取工作程序，因此主线程可以等待所有其他线程完成当前批处理，然后告诉它们启动下一个。我重复使用引用Anthony Wiiliams's book的人的屏障代码 - 这里是屏障：

class barrier
{
    unsigned const count;
    std::atomic<unsigned> spaces;
    std::atomic<unsigned> generation;

public:
    explicit barrier(unsigned count_) :
        count(count_), spaces(count), generation(0) {}

    void wait()
    {
        unsigned const my_generation = generation;
        if (!--spaces)
        {
            spaces = count;
            ++generation;
        }
        else
        {
            while (generation == my_generation)
                std::this_thread::yield();
        }
    }
};

这里是使用屏障的Channel类的新运行方法 - 请注意＆＃34; stop_th＆＃34;旗。当线程在最后一批之后和加入之前被解锁时，它不应该运行另一批，因此该测试。

void run(int& start, int& end, barrier& b)
{
    while (!stop_th){
        // wait for next batch notification - use the next_batch flag to avoid
        // spurious wake-ups
        {
            std::unique_lock<std::mutex> locker(g_lockbatch);
            g_nextbatch.wait(locker, [&](){return (next_batch==true); });
        }

        if (stop_th) return;

        // simulate work
        std::this_thread::sleep_for(std::chrono::seconds(1));

        // wait for everyone to meet
        next_batch = false;
        b.wait();
    }
}

最后，主要：

int main()
{
    int end = 100;
    int batch = 10;
    int startBatch = 0;
    int endBatch = startBatch + batch;

    // declare a barrier where all threads will meet
    barrier b(nbChannels+1);

    // declare some channels (threads)
    std::vector<Channel> channels(nbChannels);

    // start the threads
    for (auto& ch : channels) ch.start(startBatch, endBatch, b);

    while (endBatch<=end){

        // notify the channels they can process one batch
        for (auto& ch : channels) ch.go_for_next_batch();
        g_nextbatch.notify_all();

        // wait until all threads have finished their batch
        b.wait();

        // prepare the next one
        startBatch += batch;
        endBatch += batch;
    }

    // all channels are blocked by the next_batch condition
    // so notify a next batch and join them
    for (auto& ch : channels) ch.stop();
    for (auto& ch : channels) ch.go_for_next_batch();
    g_nextbatch.notify_all();
    for (auto& ch : channels) ch.wait_until_stopped();

    return 0;
}

再次感谢您的所有评论/答案!!!

Answer 1

我将注释更改为答案，因为我修复了cpp.sh中的代码，现在似乎已完成。

关于调用停止时不存在它们。请注意，它们可能仍在等待下一批锁定。考虑添加一个调用以从锁中释放它们，让它们检查它们是否在锁定步骤后停止。

将stop函数分成两个函数，一个是更改布尔值，另一个是等待的位置。让我们调用两个函数stop和wait_until_stopped

然后将以下代码添加到main函数中。

而不是

for (auto& ch : channels) ch.stop();

使用：

for (auto& ch : channels) ch.stop();

for (auto& ch : channels) ch.go_for_next_batch();

g_nextbatch.notify_all();

for (auto& ch : channels) ch.wait_until_stopped();

C ++主线程通知线程通知主线程

1 个答案: