Function main

compute/perf/perf_accumulate.cpp:89–140

}

int main(int argc, char *argv[])
{
    // setup command line arguments
    po::options_description options("options");
    options.add_options()
        ("help", "show usage instructions")
        ("size", po::value<size_t>()->default_value(8192), "input size")
        ("trials", po::value<size_t>()->default_value(3), "number of trials to run")
        ("tune", "run tuning procedure")
    ;
    po::positional_options_description positional_options;
    positional_options.add("size", 1);

    // parse command line
    po::variables_map vm;
    po::store(
        po::command_line_parser(argc, argv)
            .options(options).positional(positional_options).run(),
        vm
    );
    po::notify(vm);

    const size_t size = vm["size"].as<size_t>();
    const size_t trials = vm["trials"].as<size_t>();
    std::cout << "size: " << size << std::endl;

    // setup context and queue for the default device
    compute::device device = compute::system::default_device();
    compute::context context(device);
    compute::command_queue queue(context, device);
    std::cout << "device: " << device.name() << std::endl;

    // create vector of random numbers on the host
    std::vector<int> host_data(size);
    std::generate(host_data.begin(), host_data.end(), rand_int);

    // create vector on the device and copy the data
    compute::vector<int> device_data(
        host_data.begin(), host_data.end(), queue
    );

    // run tuning proceure (if requested)
    if(vm.count("tune")){
        tune_accumulate(device_data, trials, queue);
    }

    // run benchmark
    double t = perf_accumulate(device_data, trials, queue);
    std::cout << "time: " << t / 1e6 << " ms" << std::endl;

    return 0;
}