main.cpp

#include <atomic>
#include <chrono>
#include <cmath>
#include <iomanip>
#include <iostream>
#include <mutex>
#include <thread>
#include <vector>
#include <omp.h>

constexpr double D_PI = 3.1415926535897932384626433832795;

template< typename FUNC, typename... Args >
inline std::chrono::nanoseconds time( FUNC &&func, Args &&... args )
{
    auto const start = std::chrono::high_resolution_clock::now();
    func( std::forward< Args >( args )... );
    auto const end = std::chrono::high_resolution_clock::now();
    auto const duration = end - start;
    return std::chrono::duration_cast< std::chrono::nanoseconds >( duration );
}

template< typename CHAR, typename TRAIT >
std::basic_ostream< CHAR, TRAIT > &operator<<( std::basic_ostream< CHAR, TRAIT > &os, unsigned __int128 value )
{
    char str[ 40 ] = {};
    int index = 38;
    if( value == 0 )
        str[ index-- ] = '0';
    else
    {
        while( value != 0 )
        {
            int v = value % 10;
            value /= 10;
            str[ index-- ] = v + '0';
        }
    }
    return os << &str[ index + 1 ];
}

void openmp()
{
    unsigned __int128 value = 0;
    #pragma omp parallel num_threads( 2 )
    {
        if( omp_get_thread_num() == 0 )
        {
            for( unsigned int i = 0; i < 200000; ++i )
            {
                if( i % 2 )
                    value = static_cast< unsigned __int128 >( 1 ) << 65;
                else
                    value = static_cast< unsigned __int128 >( 1 ) << 1;
                #pragma omp flush( value )
            }
        }
        else if( omp_get_thread_num() == 1 )
        {
            for( unsigned int i = 0; i < 2000; ++i )
            {
                #pragma omp flush( value )
                std::cout << value << std::endl;
            }
        }
    }
}

int main()
{
    constexpr unsigned long long int N = 1000ull * 1000 * 1000 * 10;
    constexpr unsigned int count = 1;
    std::cout << "omp_get_max_threads()               = " << omp_get_max_threads() << std::endl;
    std::cout << "std::thread::hardware_concurrency() = " << std::thread::hardware_concurrency() << std::endl;
    for( unsigned int i = 0; i < count; ++i )
        std::cout << "OpenMP Elapsed time [ns] = " << time( openmp ).count() << std::endl;
    return 0;
}