#include #include #include #include #include using namespace std; //assume we want to do i++; in multiple threads: i++ is three instructions: // -load value in register // -add one to register // -store register at memory adress //if we do the first two in thread 1, then do i++; in thread two and the do the store in thread 1, we will have incremented i by one and not by two //atomic types solve this issue //we could also solve this by using a mutex to aquire the counter, but atomic types are faster for elementary types and should be preferred //atomic objects are free from data races, that means that they can be modified from multiple threads at the same time or read and modified at the same time without introducing errors //every TriviallyCopyable(basically means that you can use memcioy to cooy it)type can be atomic atomic i=0; int main() { vector ts; //since i is atomic we can use it without locking it with a mutex //only certain operations on are atomic //++i; i++; i+=number; i|=number; //int j=i+number;i = j+number; //are atomic //but there is no atomic multiply //and x=x+number is atomic read + atomic write, so not atomic overall function l = [](){++i;}; for(int j=0; j < 5; j++){ ts.push_back(thread(l)); } for(auto& x : ts){ x.join(); } cout << "i should be 5: i="<

Modern C++: Multithreading 02-Atomics