std::atomic_fetch_add, std::atomic_fetch_add_explicit

From cppreference.com
< cpp‎ | atomic
Defined in header <atomic>
(1) (since C++11)
template< class Integral >
Integral atomic_fetch_add( std::atomic<Integral>* obj, Integral arg );
template< class Integral >
Integral atomic_fetch_add( volatile std::atomic<Integral>* obj, Integral arg );
(2) (since C++11)
template< class Integral >

Integral atomic_fetch_add_explicit( std::atomic<Integral>* obj, Integral arg,

                                    std::memory_order order );
template< class Integral >

Integral atomic_fetch_add_explicit( volatile std::atomic<Integral>* obj, Integral arg,

                                    std::memory_order order );
(3) (since C++11)
template< class T >
T* atomic_fetch_add( std::atomic<T*>* obj, std::ptrdiff_t arg );
template< class T >
T* atomic_fetch_add( volatile std::atomic<T*>* obj, std::ptrdiff_t arg );
(4) (since C++11)
template< class T >

T* atomic_fetch_add_explicit( std::atomic<T*>* obj, std::ptrdiff_t arg,

                              std::memory_order order );
template< class T >

T* atomic_fetch_add_explicit( volatile std::atomic<T*>* obj, std::ptrdiff_t arg,

                              std::memory_order order );

Performs atomic addition.

1-2) Atomically adds arg to the value pointed to by obj and returns the value obj held previously. The operation is performed as if the following was executed:
1) obj->fetch_add(arg)
2) obj->fetch_add(arg, order)
3-4) Atomically increments the pointer value, pointed to by obj, by arg, and returns the value obj held previously. The operation is performed as if the following was executed:
3) obj->fetch_add(arg)
4) obj->fetch_add(arg, order)

Contents

[edit] Parameters

obj - pointer to the atomic object to modify
arg - the value to add to the value stored in the atomic object
order - the memory sycnhronization ordering for this operation: all values are permitted.

[edit] Return value

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

[edit] Exceptions

noexcept specification:  
noexcept
  

[edit] Possible implementation

First version
template< class T >
typename std::enable_if<std::is_integral<T>::value && !std::is_same<T, bool>::value, T>::type
atomic_fetch_add( std::atomic<T>* obj, T arg );
{
    return obj->fetch_add(arg);
}
Second version
template< class T >
T* atomic_fetch_add( std::atomic<T*>* obj, std::ptrdiff_t arg)
{
    return obj->fetch_add(arg);
}

[edit] Example

Single-writer/multiple-reader lock can be made with fetch_add. Note that this simplistic implementation is not lockout-free

#include <string>
#include <thread>
#include <vector>
#include <iostream>
#include <atomic>
#include <chrono>
 
// meaning of cnt:
// 10: there are no active readers or writers.
// 1...9: there are 9...1 readers active, The writer is blocked
// 0: temporary value between fetch_sub and fetch_add in reader lock
// -1: there is a writer active. The readers are blocked.
const int N = 10; // nine concurrent readers are allowed
std::atomic<int> cnt = ATOMIC_VAR_INIT(N);
 
std::vector<int> data;
 
void reader(int id)
{
    for(;;)
    {
        // lock
        while(std::atomic_fetch_sub(&cnt, 1) <= 0)
            std::atomic_fetch_add(&cnt, 1);
        // read
        if(!data.empty())
            std::cout << (  "reader " + std::to_string(id)
                          + " sees " + std::to_string(*data.rbegin()) + '\n');
        if(data.size() == 100)
            break;
        // unlock
        std::atomic_fetch_add(&cnt, 1);
        // pause
        std::this_thread::sleep_for(std::chrono::milliseconds(1));
    }
}
 
void writer()
{
    for(int n = 0; n < 100; ++n)
    {
        // lock
        while(std::atomic_fetch_sub(&cnt, N+1) != N)
            std::atomic_fetch_add(&cnt, N+1);
        // write
        data.push_back(n);
        std::cout << "writer pushed back " << n << '\n';
        // unlock
        std::atomic_fetch_add(&cnt, N+1);
        // pause
        std::this_thread::sleep_for(std::chrono::milliseconds(1));
    }
}
 
int main()
{
    std::vector<std::thread> v;
    for (int n = 0; n < N; ++n) {
        v.emplace_back(reader, n);
    }
    v.emplace_back(writer);
    for (auto& t : v) {
        t.join();
    }
}

Output:

writer pushed back 0
reader 8 sees 0
reader 3 sees 0
reader 1 sees 0
<...>
reader 2 sees 99
reader 6 sees 99
reader 1 sees 99

[edit] See also

atomically adds the argument to the value stored in the atomic object and obtains the value held previously
(public member function of std::atomic)
subtracts a non-atomic value from an atomic object and obtains the previous value of the atomic
(function template)
C documentation for atomic_fetch_add, atomic_fetch_add_explicit