c++提供了互斥量：mutex和条件变量：condition_variable，但是并没有信号量：semaphore。而linux和windows系统库会提供的。下面简单介绍一下信号量的特性，然后给出一个简单的demo，使用mutex + condition_variable 来实现信号量。
信号量的定义：
信号量是一个整数count，提供两个原子操作：P操作和V操作
P操作(wait操作)：count减1；如果count < 0，那么挂起执行线程
V操作(signal操作)：count加1；如果count <= 0，那么唤醒一个执行线程
mutex互斥量相当于一把锁，lock的状态为0 ，1，也就是lock状态与unlock状态。如果现在有多把锁，数量count最初被设定为n，取1把锁count–，如果发现锁的数量小于0，也就是没有锁了，此时就需要wait，也可以说是suspend or block，直到别人释放出一把锁。取完锁要还一把锁，也就是count++，如果发现此时count<=0，说明此时有人在等待锁，就唤醒一个等待的线程，把锁给他。
当count = 1时，就相当于mutex了。
如何实现信号量：
1、首先需要一个int or long变量作为count；
2、然后由于PV操作是原子操作，所以至少需要一个mutex来保证互斥
3、需要挂起线程，又需要唤醒线程，所以也需要用到condition_variable
代码：

#include <iostream>
#include <mutex>
#include <condition_variable>
#include <thread>
namespace Solution1 {class semaphore {private:int count;int wakeups;    // 辅助变量，要唤醒的线程数，初值为0std::mutex mutex;std::condition_variable cond;public:semaphore(int value = 1) : count(value), wakeups(0) {}void wait()  // P操作{std::unique_lock<std::mutex> lock(mutex);if (--count < 0) {cond.wait(lock, [&]()->bool{return wakeups > 0;});--wakeups;}}void signal() // V操作{std::lock_guard<std::mutex> lock(mutex);if (++count <= 0) {++wakeups;cond.notify_one();}}};std::mutex printMutex;Solution1::semaphore ba(0) , cb(0), dc(0);void a(){ba.wait(); // b->astd::lock_guard<std::mutex> lock(printMutex);std::cout << "thread a" <<std::endl;}void b(){cb.wait(); // c->bstd::lock_guard<std::mutex> lock(printMutex);std::cout << "thread b" <<std::endl;ba.signal(); // b->a}void c(){dc.wait(); // d->cstd::lock_guard<std::mutex> lock(printMutex);std::cout << "thread c" <<std::endl;cb.signal(); // c->b}void d(){std::lock_guard<std::mutex> lock(printMutex);std::cout << "thread d" <<std::endl;dc.signal(); // d->c}
};namespace Solution2 {class semaphore {private:int count;std::mutex mutex;std::condition_variable cond;public:semaphore(int value = 1) : count(value){}void wait()  // P操作{std::unique_lock<std::mutex> lock(mutex);if (--count < 0) {cond.wait(lock);}}void signal() // V操作{std::lock_guard<std::mutex> lock(mutex);if (++count <= 0) {cond.notify_one();}}};std::mutex printMutex;Solution1::semaphore ba(0) , cb(0), dc(0);void a(){ba.wait(); // b->astd::lock_guard<std::mutex> lock(printMutex);std::cout << "thread a" <<std::endl;}void b(){cb.wait(); // c->bstd::lock_guard<std::mutex> lock(printMutex);std::cout << "thread b" <<std::endl;ba.signal(); // b->a}void c(){dc.wait(); // d->cstd::lock_guard<std::mutex> lock(printMutex);std::cout << "thread c" <<std::endl;cb.signal(); // c->b}void d(){std::lock_guard<std::mutex> lock(printMutex);std::cout << "thread d" <<std::endl;dc.signal(); // d->c}
};// 测量一个函数的运行时间
template <class T>
void measure(T&& func) {using namespace std::chrono;auto start = system_clock::now();// funcfunc();duration<double> diff = system_clock::now() - start;std::cout << "执行了" << diff.count() << "秒" << std::endl;
}int main()
{measure([](){std::thread th1(Solution1::a), th2(Solution1::b), th3(Solution1::c), th4(Solution1::d);th1.join();th2.join();th3.join();th4.join();std::cout << "ending" << std::endl;});measure([](){std::thread th1(Solution2::a), th2(Solution2::b), th3(Solution2::c), th4(Solution2::d);th1.join();th2.join();th3.join();th4.join();std::cout << "ending" << std::endl;});return 0;
}

在linux系统下运行效果如下：

dyy@dyy-Lenovo-ThinkBook-14-IIL:~/Desktop/HeartBeat$ touch main.cpp
dyy@dyy-Lenovo-ThinkBook-14-IIL:~/Desktop/HeartBeat$ vim main.cpp 
dyy@dyy-Lenovo-ThinkBook-14-IIL:~/Desktop/HeartBeat$ g++ -pthread -o main main.cpp 
dyy@dyy-Lenovo-ThinkBook-14-IIL:~/Desktop/HeartBeat$ ./main 
thread d
thread c
thread b
thread a
ending
执行了0.000541065秒
thread d
thread c
thread b
thread a
ending
执行了0.000292463秒
dyy@dyy-Lenovo-ThinkBook-14-IIL:~/Desktop/HeartBeat$ ./main 
thread d
thread c
thread b
thread a
ending
执行了0.000422226秒
thread d
thread c
thread b
thread a
ending
执行了0.000189556秒
dyy@dyy-Lenovo-ThinkBook-14-IIL:~/Desktop/HeartBeat$ ./main 
thread d
thread c
thread b
thread a
ending
执行了0.000549342秒
thread d
thread c
thread b
thread a
ending
执行了0.000312412秒
dyy@dyy-Lenovo-ThinkBook-14-IIL:~/Desktop/HeartBeat$ ./main 
thread d
thread c
thread b
thread a
ending
执行了0.000521796秒
thread d
thread c
thread b
thread a
ending
执行了0.00043399秒
dyy@dyy-Lenovo-ThinkBook-14-IIL:~/Desktop/HeartBeat$ ./main 
thread d
thread c
thread b
thread a
ending
执行了0.000426875秒
thread d
thread c
thread b
thread a
ending
执行了0.000244544秒
dyy@dyy-Lenovo-ThinkBook-14-IIL:~/Desktop/HeartBeat$ ./main 
thread d
thread c
thread b
thread a
ending
执行了0.000445409秒
thread d
thread c
thread b
thread a
ending
执行了0.000345057秒
dyy@dyy-Lenovo-ThinkBook-14-IIL:~/Desktop/HeartBeat$ ./main 
thread d
thread c
thread b
thread a
ending
执行了0.000377516秒
thread d
thread c
thread b
thread a
ending
执行了0.000258996秒
dyy@dyy-Lenovo-ThinkBook-14-IIL:~/Desktop/HeartBeat$ ./main 
thread d
thread c
thread b
thread a
ending
执行了0.000523052秒
thread d
thread c
thread b
thread a
ending
执行了0.00027911秒
dyy@dyy-Lenovo-ThinkBook-14-IIL:~/Desktop/HeartBeat$ ./main 
thread d
thread c
thread b
thread a
ending
执行了0.000517352秒
thread d
thread c
thread b
thread a
ending
执行了0.000395749秒
dyy@dyy-Lenovo-ThinkBook-14-IIL:~/Desktop/HeartBeat$ ./main 
thread d
thread c
thread b
thread a
ending
执行了0.000488651秒
thread d
thread c
thread b
thread a
ending
执行了0.000392515秒

参考
c++11中信号量(semaphore)的实现 | 陆仁贾
深层次探讨mutex与semaphore之间的区别
下面的这个代码比较复杂：
线程同步之信号量，代码实现方法2（条件变量+mutex互斥量）