Function runBenchmark

test_Grad.cpp:29–62

template <typename Func>
BenchmarkStats runBenchmark(const Func &func, int iters, int repeats = 7, int warmup_iters = 200)
{
    for (int i = 0; i < warmup_iters; ++i)
    {
        func();
    }

    vector<double> samples_us;
    samples_us.reserve(repeats);
    for (int r = 0; r < repeats; ++r)
    {
        const auto t0 = high_resolution_clock::now();
        for (int i = 0; i < iters; ++i)
        {
            func();
        }
        const double avg_us =
            duration_cast<microseconds>(high_resolution_clock::now() - t0).count() / static_cast<double>(iters);
        samples_us.push_back(avg_us);
    }

    BenchmarkStats stats;
    stats.mean_us = accumulate(samples_us.begin(), samples_us.end(), 0.0) / static_cast<double>(samples_us.size());
    stats.min_us = *min_element(samples_us.begin(), samples_us.end());

    double var = 0.0;
    for (const double x : samples_us)
    {
        const double d = x - stats.mean_us;
        var += d * d;
    }
    stats.std_us = sqrt(var / static_cast<double>(samples_us.size()));
    return stats;
}

void printHeader(const string &title)
{