NAME

std::atomic::fetch_add - std::atomic::fetch_add

Synopsis

member only of atomic<Integral>(C++11) and atomic<Floating>(C++20)
template specializations
T fetch_add( T arg,
std::memory_order order = std::memory_order_seq_cst ) noexcept;
T fetch_add( T arg,
std::memory_order order = std::memory_order_seq_cst ) volatile noexcept;
member only of atomic<T*> template specialization (1)
T* fetch_add( std::ptrdiff_t arg,
std::memory_order order = std::memory_order_seq_cst ) noexcept; (2)
T* fetch_add( std::ptrdiff_t arg,
std::memory_order order = std::memory_order_seq_cst ) volatile noexcept;

Atomically replaces the current value with the result of arithmetic addition of the
value and arg. That is, it performs atomic post-increment. The operation is
read-modify-write operation. Memory is affected according to the value of order.

For signed Integral types, arithmetic is defined to use two’s complement
representation. There are no undefined results.

For T* types, the result may be an undefined address, but the operation otherwise
has no undefined behavior. The program is ill-formed if T is not an object type.

For floating-point types, the floating-point environment in effect may
be different from the calling thread's floating-point environment. The
operation need not conform to the corresponding std::numeric_limits
traits but is encouraged to do so. If the result is not a
representable value for its type, the result is unspecified but the (since C++20)
operation otherwise has no undefined behavior.

The volatile-qualified versions are deprecated if
std::atomic<T>::is_always_lock_free is false.

Parameters

arg - the other argument of arithmetic addition
order - memory order constraints to enforce

Return value

The value immediately preceding the effects of this function in the modification
order of *this.

Example

// Run this code

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

std::atomic<long long> data{10};
std::array<long long, 5> return_values{};

void do_work(int thread_num)
{
long long val = data.fetch_add(1, std::memory_order_relaxed);
return_values[thread_num] = val;
}

int main()
{
{
std::jthread th0{do_work, 0};
std::jthread th1{do_work, 1};
std::jthread th2{do_work, 2};
std::jthread th3{do_work, 3};
std::jthread th4{do_work, 4};
}

std::cout << "Result : " << data << '\n';

for (long long val : return_values) {
std::cout << "Seen return value : " << val << std::endl;
}
}

Possible output:

Result : 15
Seen return value : 11
Seen return value : 10
Seen return value : 14
Seen return value : 12
Seen return value : 13

Defect reports

The following behavior-changing defect reports were applied retroactively to
previously published C++ standards.