【发布时间】:2015-10-08 16:42:46
【问题描述】:
我需要帮助。我正在尝试做一些具体的事情,但我缺乏多线程技能正在杀死我。
基本上,我的主程序/线程需要管理许多必须运行多次的“通道”。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。
所以主线程必须等待所有通道(线程)完成它们的运行才能启动下一个。 并且所有通道都必须等待来自主线程的通知才能运行。
以下是我的做法 - 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;
}
但有时程序会阻塞,可能是线程相互等待,但我不明白为什么。 在任何情况下,加入线程(main 末尾的“stop”方法)会使我的程序无限期地运行,也不明白为什么。
编辑:感谢您的 cmets 和一些研究,我设法使用同步屏障获得了一个工作程序,因此主线程可以等待所有其他线程完成当前批处理,然后再告诉它们开始下一个。 我重用了此处引用 Anthony Wiiliams's book 的某人的屏障代码 - 这是 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();
}
}
};
这是使用屏障的 Channel 类的新 run 方法 - 请注意对“stop_th”标志的附加测试。当线程在最后一批之后并且在加入之前被解除阻塞时,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();
}
}
最后是main:
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 cmets / 回答!!!
【问题讨论】:
-
您应该能够将调试器连接到实时进程(例如 Linux 上的
gdb <exe name> <pid>)并列出线程的当前状态。我发现这通常可以很好地表明导致问题的原因。我怀疑您应该尝试坚持使用一个互斥锁或仅按严格顺序锁定/解锁。
标签: c++ multithreading concurrency conditional-statements