std::async
Defined in header <future>
|
||
template< class F, class... Args > std::future</* see below */> async( F&& f, Args&&... args ); |
(1) | (since C++11) |
template< class F, class... Args > std::future</* see below */> async( std::launch policy, F&& f, Args&&... args ); |
(2) | (since C++11) |
The function template std::async
runs the function f asynchronously (potentially in a separate thread which might be a part of a thread pool) and returns a std::future that will eventually hold the result of that function call.
The return type of std::async
is std::future<V> where V is:
typename std::result_of<typename std::decay<F>::type( |
(until C++17) |
std::invoke_result_t<std::decay_t<F>, std::decay_t<Args>...>. |
(since C++17) |
If any of the following conditions is satisfied, the program is ill-formed:
|
(until C++20) |
If any of the following is false, the program is ill-formed:
|
(since C++20) |
The call to std::async
synchronizes with the call to f, and the completion of f is sequenced before making the shared state ready.
Contents |
[edit] Parameters
f | - | Callable object to call |
args | - | parameters to pass to f |
policy | - | bitmask value, where individual bits control the allowed methods of execution |
[edit] Return value
std::future referring to the shared state created by this call to std::async
.
[edit] Launch policies
[edit] Async invocation
If the async flag is set, i.e. (policy & std::launch::async) != 0, then std::async
calls
INVOKE(decay-copy(std::forward<F>(f)), |
(until C++23) |
std::invoke(auto(std::forward<F>(f)), |
(since C++23) |
as if in a new thread of execution represented by a std::thread object.
The calls of decay-copy are evaluated in the current thread. |
(until C++23) |
The values produced by auto are materialized in the current thread. |
(since C++23) |
If the function f returns a value or throws an exception, it is stored in the shared state accessible through the std::future that std::async
returns to the caller.
[edit] Deferred invocation
If the deferred flag is set (i.e. (policy & std::launch::deferred) != 0), then std::async
stores
decay-copy(std::forward<F>(f)) and decay-copy(std::forward<Args>(args))... in the shared state. |
(until C++23) |
auto(std::forward<F>(f)) and auto(std::forward<Args>(args))... in the shared state. |
(since C++23) |
Lazy evaluation is performed:
- The first call to a non-timed wait function on the std::future that
std::async
returned to the caller will evaluate INVOKE(std::move(g), std::move(xyz)) in the thread that called the waiting function (which does not have to be the thread that originally calledstd::async
), where
|
(until C++23) |
|
(since C++23) |
- The result or exception is placed in the shared state associated with the returned std::future and only then it is made ready. All further accesses to the same std::future will return the result immediately.
[edit] Other policies
If neither std::launch::async nor std::launch::deferred, nor any implementation-defined policy flag is set in policy, the behavior is undefined.
[edit] Policy selection
If more than one flag is set, it is implementation-defined which policy is selected. For the default (both the std::launch::async and std::launch::deferred flags are set in policy), standard recommends (but does not require) utilizing available concurrency, and deferring any additional tasks.
If the std::launch::async policy is chosen,
- a call to a waiting function on an asynchronous return object that shares the shared state created by this
std::async
call blocks until the associated thread has completed, as if joined, or else time out; and - the associated thread completion synchronizes-with the successful return from the first function that is waiting on the shared state, or with the return of the last function that releases the shared state, whichever comes first.
[edit] Exceptions
Throws
- std::bad_alloc, if the memory for the internal data structures cannot be allocated, or
- std::system_error with error condition std::errc::resource_unavailable_try_again, if policy == std::launch::async and the implementation is unable to start a new thread.
- If policy is std::launch::async | std::launch::deferred or has additional bits set, it will fall back to deferred invocation or the implementation-defined policies in this case.
[edit] Notes
The implementation may extend the behavior of the first overload of std::async
by enabling additional (implementation-defined) bits in the default launch policy.
Examples of implementation-defined launch policies are the sync policy (execute immediately, within the std::async
call) and the task policy (similar to std::async
, but thread-locals are not cleared)
If the std::future obtained from std::async
is not moved from or bound to a reference, the destructor of the std::future will block at the end of the full expression until the asynchronous operation completes, essentially making code such as the following synchronous:
std::async(std::launch::async, []{ f(); }); // temporary's dtor waits for f() std::async(std::launch::async, []{ g(); }); // does not start until f() completes
Note that the destructors of std::futures obtained by means other than a call to std::async
never block.
[edit] Example
#include <algorithm> #include <future> #include <iostream> #include <mutex> #include <numeric> #include <string> #include <vector> std::mutex m; struct X { void foo(int i, const std::string& str) { std::lock_guard<std::mutex> lk(m); std::cout << str << ' ' << i << '\n'; } void bar(const std::string& str) { std::lock_guard<std::mutex> lk(m); std::cout << str << '\n'; } int operator()(int i) { std::lock_guard<std::mutex> lk(m); std::cout << i << '\n'; return i + 10; } }; template<typename RandomIt> int parallel_sum(RandomIt beg, RandomIt end) { auto len = end - beg; if (len < 1000) return std::accumulate(beg, end, 0); RandomIt mid = beg + len / 2; auto handle = std::async(std::launch::async, parallel_sum<RandomIt>, mid, end); int sum = parallel_sum(beg, mid); return sum + handle.get(); } int main() { std::vector<int> v(10000, 1); std::cout << "The sum is " << parallel_sum(v.begin(), v.end()) << '\n'; X x; // Calls (&x)->foo(42, "Hello") with default policy: // may print "Hello 42" concurrently or defer execution auto a1 = std::async(&X::foo, &x, 42, "Hello"); // Calls x.bar("world!") with deferred policy // prints "world!" when a2.get() or a2.wait() is called auto a2 = std::async(std::launch::deferred, &X::bar, x, "world!"); // Calls X()(43); with async policy // prints "43" concurrently auto a3 = std::async(std::launch::async, X(), 43); a2.wait(); // prints "world!" std::cout << a3.get() << '\n'; // prints "53" } // if a1 is not done at this point, destructor of a1 prints "Hello 42" here
Possible output:
The sum is 10000 43 world! 53 Hello 42
[edit] Defect reports
The following behavior-changing defect reports were applied retroactively to previously published C++ standards.
DR | Applied to | Behavior as published | Correct behavior |
---|---|---|---|
LWG 2021 | C++11 | return type incorrect and value category of arguments unclear in the deferred case |
corrected return type and clarified that rvalues are used |
LWG 2078 | C++11 | it was unclear whether std::system_error may be thrown if policy specifies other launch policies besides std::launch::async |
can only be thrown if policy == std::launch::async |
LWG 2100 | C++11 | timed waiting functions could not timeout if std::launch::async policy is used |
allowed |
LWG 2120 | C++11 | the behavior was unclear if no standard or implementation-defined policy is set |
the behavior is undefined in this case |
LWG 2752 | C++11 | std::async might not throw std::bad_alloc if thememory for the internal data structures cannot be allocated |
throws |
LWG 3476 | C++20 | (the decayed types of) F and the argument typeswere directly required to be move constructible |
removed these requirements[1] |
- ↑ The move-constructibility is already indirectly required by std::is_constructible_v.
[edit] See also
(C++11) |
waits for a value that is set asynchronously (class template) |