FOR j := 0 to 3
	i := j*64
	dst[i+63:i] := a[i+63:i] + b[i+63:i]
ENDFOR
dst[MAX:256] := 0

uint64_t lowbit(uint64_t x)
{
    return x & (-x);
}
uint64_t extTo2Power(uint64_t n, int i)//arraysize datasize
{
    while(lowbit(n) < i)
        n += lowbit(n);
    return n;
}

template <typename T>
T* getArray(uint64_t size)
{
    uint64_t ExSize = extTo2Power(size, 32/sizeof(T));
    T* arr = new T[ExSize];
    for (uint64_t i = 0; i < size; i++)
        arr[i] = i+1;
    for (uint64_t i = size; i < ExSize; i++)
        arr[i] = 0;
    return arr;
}

template <typename T>
T simpleSum(T* arr, uint64_t size)
{
    T sum = 0;
    for (uint64_t i = 0; i < size; i++)
        sum += arr[i];
    return sum;
}

float avx2Sum(float* arr, uint64_t size)
{
    float sum[8] = {0};
    __m256 sum256 = _mm256_setzero_ps();
    __m256 load256 = _mm256_setzero_ps();
    for (uint64_t i = 0; i < size; i += 8)
    {
        load256 = _mm256_loadu_ps(&arr[i]);
        sum256 = _mm256_add_ps(sum256, load256);
    }
    sum256 = _mm256_hadd_ps(sum256, sum256);
    sum256 = _mm256_hadd_ps(sum256, sum256);
    _mm256_storeu_ps(sum, sum256);
    sum[0] += sum[4];
    return sum[0];
}

double avx2Sum(double* arr, uint64_t size)
{
    double sum[4] = {0};
    __m256d sum256 = _mm256_setzero_pd();
    __m256d load256 = _mm256_setzero_pd();
    for (uint64_t i = 0; i < size; i += 4)
    {
        load256 = _mm256_loadu_pd(&arr[i]);
        sum256 = _mm256_add_pd(sum256, load256);
    }
    sum256 = _mm256_hadd_pd(sum256, sum256);
    _mm256_storeu_pd(sum, sum256);
    sum[0] += sum[2];
    return sum[0];
}

uint64_t getTime()
{
    uint64_t timems = std::chrono::duration_cast<std::chrono::milliseconds>(std::chrono::system_clock::now().time_since_epoch()).count();
    return timems;
}

	uint64_t N = 1e9;
    // compare the performance of normal add and avx2 add
    uint64_t start, end;
    // test float
    cout << "compare float sum: " << endl;
    float* arr3 = getArray<float>(N);
    start = getTime();
    float sum3 = simpleSum(arr3, N);
    end = getTime();
    cout << "float simpleSum time: " << end - start << endl;
    cout << "float simpleSum sum: " << sum3 << endl;
    start = getTime();
    sum3 = avx2Sum(arr3, N);
    end = getTime();
    cout << "float avx2Sum time: " << end - start << endl;
    cout << "float avx2Sum sum: " << sum3 << endl;
    delete[] arr3;
    cout << endl << endl;
    // test double
    cout << "compare double sum: " << endl;
    double* arr4 = getArray<double>(N);
    start = getTime();
    double sum4 = simpleSum(arr4, N);
    end = getTime();
    cout << "double simpleSum time: " << end - start << endl;
    cout << "double simpleSum sum: " << sum4 << endl;
    start = getTime();
    sum4 = avx2Sum(arr4, N);
    end = getTime();
    cout << "double avx2Sum time: " << end - start << endl;
    cout << "double avx2Sum sum: " << sum4 << endl;
    delete[] arr4;
    cout << endl << endl;

g++ -mavx2 avx_big_integer.cpp 
./a.out

g++ -O2 -mavx2 avx_big_integer.cpp 
./a.out