#include <iostream>
#include <vector>
#include <string>
#include <hpx/hpx_init.hpp>
#include <hpx/util/high_resolution_timer.hpp>
#include <hpx/runtime/get_worker_thread_num.hpp>
#include <hpx/runtime/get_os_thread_count.hpp>
#include <hpx/parallel/algorithms/for_each.hpp>
#include <hpx/parallel/algorithms/sort.hpp>
#include <hpx/parallel/executors/static_chunk_size.hpp>
#include <hpx/parallel/executors/dynamic_chunk_size.hpp>
#include <boost/range/irange.hpp>
#include <blaze/Math.h>

double get_time() {
    return hpx::util::high_resolution_clock::now() * 1e-9;
}

int hpx_main(int argc, char *argv[]) {

    int N = std::stoi(argv[1], NULL);

    blaze::DynamicMatrix<int> A(N, N), B(N, N), C(N, N);

   for(size_t i(0); i < A.rows(); i++) {
        for(size_t j(0); j < A.columns(); j++) {
            A(i, j) = blaze::rand<int>(0, 1000);
            B(i, j) = blaze::rand<int>(0, 1000);
        }
    }

    std::cout << " matrices divided into " << blaze::getNumThreads() << std::endl;

    double t1 = get_time();
    C = A * B;
    double t2 = get_time();
    double delta = t2 - t1;
    std::cout << "Time spent with " << hpx::get_os_thread_count() << " is " << delta << " (s) \n";
    return hpx::finalize();

}

int main(int argc, char *argv[]) {
    return hpx::init(argc, argv);
}