// HOW TO COMPILE ON MACOS
// export CPPFLAGS="-I/opt/homebrew/opt/libomp/include"
// export LDFLAGS="-L/opt/homebrew/opt/libomp/lib"
// clang++ -std=c++17 -O3 -Xpreprocessor -fopenmp $CPPFLAGS main.cpp bench_funcs.cpp $LDFLAGS -lomp -o Ex5
// ./Ex5

#include <iostream>
#include <vector>
#include <chrono>
#include <omp.h>
#include "bench_funcs.h"

using namespace std;
using namespace std::chrono;

int main() {
    size_t n = 200000;       // FEM system size
    size_t maxit = 5000;
    double omega = 1.0;
    double tol = 1e-8;

    // show threads
    #pragma omp parallel
    {
        #pragma omp single
        cout << "Using " << omp_get_num_threads() << " OpenMP threads\n";
    }

    CSR K;
    vector<double> f, u;

    cout << "\nBuilding FEM system (parallel)\n";
    auto t0 = high_resolution_clock::now();
    build_fem_system(n, K, f);
    auto t1 = high_resolution_clock::now();
    cout << "Assembly time = " << duration<double>(t1 - t0).count() << " s\n";

    cout << "\nRunning Jacobi solver (parallel)\n";
    t0 = high_resolution_clock::now();
    jacobi_csr_parallel(K, f, u, maxit, omega, tol);
    t1 = high_resolution_clock::now();

    cout << "Jacobi time = " << duration<double>(t1 - t0).count() << " s\n";
    cout << "Done\n";
}