AVX2指令集优化整形数组求和算法

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;
}

int32_t avx2Sum(int32_t* arr, uint64_t size)
{
    int32_t sum[8] = {0};
    __m256i sum256 = _mm256_setzero_si256();
    __m256i load256 = _mm256_setzero_si256();
    for (uint64_t i = 0; i < size; i += 8)
    {
        load256 = _mm256_loadu_si256((__m256i*)&arr[i]);
        sum256 = _mm256_add_epi32(sum256, load256);
    }
    sum256 = _mm256_hadd_epi32(sum256, sum256);
    sum256 = _mm256_hadd_epi32(sum256, sum256);
    _mm256_storeu_si256((__m256i*)sum, sum256);
    sum[0] += sum[4];
    return sum[0];
}

int64_t avx2Sum(int64_t* arr, uint64_t size)
{
    int64_t sum[4] = {0};
    __m256i sum256 = _mm256_setzero_si256();
    __m256i load256 = _mm256_setzero_si256();
    for (uint64_t i = 0; i < size; i += 4)
    {
        load256 = _mm256_loadu_si256((__m256i*)&arr[i]);
        sum256 = _mm256_add_epi64(sum256, load256);
    }
    _mm256_storeu_si256((__m256i*)sum, sum256);
    sum[0] += sum[1] + sum[2] + sum[3];
    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 int32_t
    cout << "compare int32_t sum: " << endl;
    int32_t* arr = getArray<int32_t>(N);
    start = getTime();
    int32_t sum = simpleSum(arr, N);
    end = getTime();
    cout << "int32_t simpleSum time: " << end - start << endl;
    cout << "int32_t simpleSum sum: " << sum << endl;
    start = getTime();
    sum = avx2Sum(arr, N);
    end = getTime();
    cout << "int32_t avx2Sum time: " << end - start << endl;
    cout << "int32_t avx2Sum sum: " << sum << endl;
    delete[] arr;
    cout << endl << endl;
    // test int64_t
    cout << "compare int64_t sum: " << endl;
    int64_t* arr2 = getArray<int64_t>(N);
    start = getTime();
    int64_t sum2 = simpleSum(arr2, N);
    end = getTime();
    cout << "int64_t simpleSum time: " << end - start << endl;
    cout << "int64_t simpleSum sum: " << sum2 << endl;
    start = getTime();
    sum2 = avx2Sum(arr2, N);
    end = getTime();
    cout << "int64_t avx2Sum time: " << end - start << endl;
    cout << "int64_t avx2Sum sum: " << sum2 << endl;
    delete[] arr2;
    cout << endl << endl;

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

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