std:: barrier

The class template std::barrier provides a thread-coordination mechanism that blocks a group of threads of known size until all threads in that group have reached the barrier. Unlike std::latch , barriers are reusable: once a group of arriving threads are unblocked, the barrier can be reused. Unlike std::latch , barriers execute a possibly empty callable before unblocking threads.

A barrier object's lifetime consists of one or more phases. Each phase defines a phase synchronization point where waiting threads block. Threads can arrive at the barrier, but defer waiting on the phase synchronization point by calling arrive . Such threads can later block on the phase synchronization point by calling wait .

A barrier phase consists of the following steps:

1. The expected count is decremented by each call to arrive or arrive_and_drop .
2. When the expected count reaches zero, the phase completion step is run, meaning that the completion is invoked, and all threads blocked on the phase synchronization point are unblocked. The end of the completion step strongly happens-before all calls that were unblocked by the completion step return.
   Exactly once after the expected count reaches zero, a thread executes the completion step during its call to arrive , arrive_and_drop , or wait , except that it is implementation-defined whether the step executes if no thread calls wait .
3. When the completion step finishes, the expected count is reset to the value specified at construction less the number of calls to arrive_and_drop since, and the next barrier phase begins.

Concurrent invocations of the member functions of barrier , except for the destructor, do not introduce data races.

Template parameters

The default template argument of CompletionFunction is an unspecified function object type that additionally meets the requirements of DefaultConstructible . Calling an lvalue of it with no arguments has no effects.

Member types

Data members

Member functions

Notes

Example

#include <barrier>
#include <iostream>
#include <string>
#include <syncstream>
#include <thread>
#include <vector>
 
int main()
{
    const auto workers = {"Anil", "Busara", "Carl"};
 
    auto on_completion = []() noexcept
    {
        // locking not needed here
        static auto phase =
            "... done\n"
            "Cleaning up...\n";
        std::cout << phase;
        phase = "... done\n";
    };
 
    std::barrier sync_point(std::ssize(workers), on_completion);
 
    auto work = [&](std::string name)
    {
        std::string product = "  " + name + " worked\n";
        std::osyncstream(std::cout) << product;  // ok, op<< call is atomic
        sync_point.arrive_and_wait();
 
        product = "  " + name + " cleaned\n";
        std::osyncstream(std::cout) << product;
        sync_point.arrive_and_wait();
    };
 
    std::cout << "Starting...\n";
    std::vector<std::jthread> threads;
    threads.reserve(std::size(workers));
    for (auto const& worker : workers)
        threads.emplace_back(work, worker);
}

Starting...
  Anil worked
  Carl worked
  Busara worked
... done
Cleaning up...
  Busara cleaned
  Carl cleaned
  Anil cleaned
... done

Defect reports

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

See also