Atomic CAS counting semphore

semaphore.cpp

#include <atomic>
#include <thread>
#include <emmintrin.h>

class Semaphore {
public:
    explicit Semaphore(uint32_t init_value);
    void acquire();
    void release();

private:
    std::atomic<uint32_t> counter;
};

Semaphore::Semaphore(uint32_t init_value) : counter(init_value) {}

void Semaphore::acquire() {
    while (true) {
        uint32_t curr;
        do {
            // Busy wait while checking for updates
            _mm_pause();
            curr = counter.load();
        } while (curr == 0);

        /* Acquire if no other thread has updated the counter since
           we last read the counter into our local register */
        if (counter.compare_exchange_strong(curr, curr - 1)) {
            return;
        }
        /* If we fail, another thread updated counter before us, so
           we need to retry */
    }
}

void Semaphore::release() {
    counter.fetch_add(1);
}