// A molecular dynamics application that calculates the potential and relocation of particles within a large 3D space.

CLI::App app{"FFT"};

// num_threads should be roughly equal to NUMBER_PAR_PER_BOX for best performance

unsigned num_threads {1};

app.add_option("-t,--num_threads", num_threads, "number of threads (default=1)");

unsigned num_boxes {1};

app.add_option("-b,--boxes1d", num_boxes, "number of threads (default=1)");

int num_values {1024};

app.add_option("-v,--num_values", num_values, "number of values (default=1024)");

std::string model = "tf";

app.add_option("-m,--model", model, "model name tbb|omp|tf (default=tf)")

->check([] (const std::string& m) {

if(m != "tbb" && m != "omp" && m != "tf") {

return "model name should be \"tbb\", \"omp\", or \"tf\"";

}

return "";

});

CLI11_PARSE(app, argc, argv);

// Print configuration

std::cout << "model=" << model << ' '

<< "num_threads=" << num_threads << ' '

<< "num_values=" << num_values << ' '

<< std::endl;

int i;

COMPLEX *in, *out1=NULL, *out2=NULL;

in = (COMPLEX *)malloc(num_values * sizeof(COMPLEX));

out1 = (COMPLEX *)malloc(num_values * sizeof(COMPLEX));

for (i = 0; i < num_values; ++i) {

c_re(in[i]) = 1.0;

c_im(in[i]) = 1.0;

}

auto t1 = std::chrono::high_resolution_clock::now();

//fft(num_values, in, out1);

fft_tf(num_values, in, out1);

auto t2 = std::chrono::high_resolution_clock::now();

std::cout << std::chrono::duration_cast<std::chrono::microseconds>(t2 - t1).count()/1e3 << " ms\n";

{

out2 = (COMPLEX *)malloc(num_values * sizeof(COMPLEX));

for (i = 0; i < num_values; ++i) {

c_re(in[i]) = 1.0;

c_im(in[i]) = 1.0;

}

fft_seq(num_values, in, out2);

test_correctness(num_values, out1, out2);

}

return 0;