将 pthread_cond_broadcast 移植到 std::atomic

Question

Trying to understand all the in and out of std::atomic (lock-free operations in C++ 11).

I was wondering if an operation like this:

bool SomeClass::waitMyVariableToBeSet() {
    pthread_mutex_lock(&mMutex);
    while (!MyVariableToBeSet) {
        r = pthread_cond_timedwait(&msCondVariableSet, &mMutex, &ts);
    }
    pthread_mutex_unlock(&mMutex);
}

AND

void SomeClass::setMyVariable(bool newVal) {
    pthread_mutex_lock(&mMutex);
    MyVariableToBeSet= newVal;
    pthread_cond_broadcast(&msCondVariableSet);
    pthread_mutex_unlock(&mMutex);
}

Could be replace like this with std::atomic:

std::atomic<bool> MyVariableToBeSet;

bool SomeClass::waitMyVariableToBeSet() {
    uint someTimeOutCnt = 100;
    while (!MyVariableToBeSet.compare_exchange_strong(false, true) && SomeTimeCnt) {
         someTimeCnt--;
         std::this_thread::yield(); // no sure of this here
   }
}

void SomeClass::setMyVariable(bool newVal) {
    MyVariableToBeSet= newVal;
}

Answer 1

C++ 20 implemented the solution

https://en.cppreference.com/w/cpp/atomic/atomic_flag

Otherwise, I believe a spinlock would be the compromise with all trade off that comes with it.

Answer 2

With C++20 this can be implemented using new atomic wait and notify operations:

std::atomic<bool> MyVariableToBeSet;

bool SomeClass::waitMyVariableToBeSet() {
    MyVariableToBeSet.wait(false);
}

void SomeClass::setMyVariable(bool newVal) {
    MyVariableToBeSet = newVal;
    MyVariableToBeSet.notify_all();
}

Internally those will be implemented using a hash map of queues or using kernel APIs such as futex , WaitOnAddress / WakeByAddressAll , _umtx_op when possible.

Before C++20 there is no direct equivalent and you have to use condition variables or platform-specific APIs directly.