I can't understand weird std::atomic_short.load() behavior

Question

I'm having trouble understanding a part of the C++11 std::atomic_short behavior.
I set either 0 or 255 as value for an atomic_short variable.
But .load() says that the value is neither 0 or 255.
I want one thread to write the atomic variable and I want another thread read it.

Environment:
Intel Core i5
OSX 10.11.6
clang (Xcode7.3.1)

#include <iostream>
#include <atomic>
#include <thread>

std::atomic_short value = ATOMIC_VAR_INIT(0);

void process1() {
    bool flag = false;
    for (int i = 0; i < 100000; ++i){
        std::this_thread::yield;
        if (flag){
            value.store(255);
        } else {
            value.store(0);
        }
        flag = !flag;
    }
}

void process2() {
    for (int i = 0; i < 100000; ++i){
        std::this_thread::yield;
        if (value.load() != 255 && value.load() != 0){
            printf("warningA! %d\n", i);
        }
    }
}

int main(int argc, char** argv) {
    value.store(0);
    std::thread t1(process1);
    std::thread t2(process2);
    t1.join();
    t2.join();

    return 0;
}

---

warningA! 3
warningA! 1084
warningA! 1093

Answer 1

The problem is that you have 2 individual loads which makes your comparison non-atomic. Instead, load the value once and then compare:

void process2() {
    for (int i = 0; i < 100000; ++i){
        std::this_thread::yield;
        auto currentValue = value.load();
        if (currentValue != 255 && currentValue != 0){
            printf("warningA! %d\n", i);
        }
    }
}

live example