Function main

perf/perf_partial_sum.cpp:27–97

}

int main(int argc, char *argv[])
{
    using boost::compute::int_;

    perf_parse_args(argc, argv);

    std::cout << "size: " << PERF_N << std::endl;

    // setup context and queue for the default device
    boost::compute::device device = boost::compute::system::default_device();
    boost::compute::context context(device);
    boost::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_vector(PERF_N);
    std::generate(host_vector.begin(), host_vector.end(), rand_int);

    // create vector on the device and copy the data
    boost::compute::vector<int_> device_vector(PERF_N, context);
    boost::compute::vector<int_> device_res(PERF_N,context);
    boost::compute::copy(
        host_vector.begin(),
        host_vector.end(),
        device_vector.begin(),
        queue
    );

    // sum vector
    perf_timer t;
    for(size_t trial = 0; trial < PERF_TRIALS; trial++){
        boost::compute::copy(
            host_vector.begin(),
            host_vector.end(),
            device_vector.begin(),
            queue
        );

        t.start();
        boost::compute::partial_sum(
            device_vector.begin(),
            device_vector.end(),
            device_res.begin(),
            queue
        );
        queue.finish();
        t.stop();
    }
    std::cout << "time: " << t.min_time() / 1e6 << " ms" << std::endl;

    // verify sum is correct
    std::partial_sum(
        host_vector.begin(),
        host_vector.end(),
        host_vector.begin()
    );

    int device_sum = device_res.back();