The standard function std::async:
The template function async runs the function f asynchronously (potentially in a separate thread which may be part of a thread pool) and returns a std::future that will eventually hold the result of that function call.
There is two launch polices std::launch::async and std::launch::deferred. In my compiler's (GCC 6.2) standard library impelmentation, the first one always creates a new thread and the second one does lazy evaluation on the calling thread. By default std::launch::deferred
is used.
Is there some implementation which uses thread pool with size equal to the hardware threads available when std::launch::async
is specified to avoid creating two many threads when std::async
is used in recursive algorithm?
Microsoft's compiler and C++ runtime that it ships with Visual Studio does.
i am using this approach
class ThreadPool
{
public:
ThreadPool(size_t n)
: work_(io_service_)
{
AddThreads(n);
}
/**
* \brief Adds \a n threads to this thread pool
* \param n - count of threads to add
*/
void AddThreads(size_t n)
{
for (size_t i = 0; i < n; i++)
threads_.create_thread(boost::bind(&boost::asio::io_service::run, &io_service_));
}
/**
* \brief Count of thread in pool
* \return number
*/
size_t Size() const
{
return threads_.size();
}
~ThreadPool()
{
io_service_.stop();
threads_.join_all();
}
/**
* \brief Perform task \a pt. see io_service::post
* \tparam T - type with operator() defined
* \param pt - functional object to execute
*/
template <class T>
void post(std::shared_ptr<T> &pt)
{
io_service_.post(boost::bind(&T::operator(), pt));
}
/**
* \brief Perform task \a pt. see io_service::dispatch
* \tparam T - type with operator() defined
* \param pt - functional object to execute
*/
template <class T>
void dispatch(std::shared_ptr<T> &pt)
{
io_service_.dispatch(boost::bind(&T::operator(), pt));
}
private:
boost::thread_group threads_;
boost::asio::io_service io_service_;
boost::asio::io_service::work work_;
};
dispatch
is asynk(..., async)
;
post
is asynk(..., deferred)
;