double calcPiCpp11(const int numThreads, const long numSteps)
{
    double step = 1.0 / (double)numSteps;
    double x, pi;

    double* sums = new double[numThreads];
    for (int i = 0; i < numThreads; ++i)
    {
        sums[i] = 0.0f;
    }

    const int threadNumSteps = numSteps / numThreads;
    std::vector<std::thread> threads;

    for (int i = 0; i < numThreads; ++i)
    {
        int iStart = threadNumSteps * i + 1;
        threads.push_back(std::thread(calcPI, sums, step, threadNumSteps, iStart, i));
    }

    for (auto& thread : threads)
    {
        thread.join();
    }

    pi = step * std::accumulate(sums, sums + numThreads, 0.0);

    delete[] sums;
    return pi;
}