Relaxed fetch_add operations between a release store and an acquire load

writer() runs once and performs a release store. Three separate threads then each call bumper() exactly once, performing a relaxed fetch_add on the same atomic counter. reader() spins on an acquire load until it observes counter's value has reached 4, the value produced by the last fetch_add to actually execute, not writer()'s original store of 1. Is reader() guaranteed to print 99?

main.cpp
std::atomic<int> counter{0};
int payload = 0;

void writer() {
    payload = 99;
    counter.store(1, std::memory_order_release);      // heads the release sequence
}

void bumper() {                                        // called once each, by 3 threads
    counter.fetch_add(1, std::memory_order_relaxed);   // relaxed RMW
}

void reader() {
    while (counter.load(std::memory_order_acquire) < 4) {}
    std::cout << payload << std::endl;   // guaranteed 99?
}

28 of 51 in Concurrency