OpenMP我有虚假共享还是种族状况？

Question

我正在学习OpenMP，并且正在研究压缩的稀疏行乘法（Datatype std :: Complex＆lt; int＆gt;）。每次运行以下函数时，我都会收到不同的执行时间：

typedef std::vector < std::vector < std::complex < int >>> matrix;

struct CSR {
    std::vector<std::complex<int>> values; //non-zero values
    std::vector<int> row_ptr; //pointers of rows
    std::vector<int> cols_index; //indices of columns
    int rows; //number of rows
    int cols; //number of columns
    int NNZ; //number of non_zero elements
};

const matrix multiply_omp (const CSR& A,
    const CSR& B) {
    if (A.cols != B.rows)
        throw "Error";
    CSR B_t = sparse_transpose(B);
    matrix result(A.rows, std::vector < std::complex < int >>(B.rows, 0));
    #pragma omp parallel
    {
        #pragma omp for
        for (int i = 0; i < A.rows; i++) {
            for (int j = A.row_ptr[i]; j < A.row_ptr[i + 1]; j++) {
                int Ai = A.cols_index[j];
                std::complex<int> Avalue = A.values[j];
                for (int k = 0; k < B_t.rows; k++) {
                    std::complex < int > sum(0, 0);
                    for (int l = B_t.row_ptr[k]; l < B_t.row_ptr[k + 1]; l++)
                        if (Ai == B_t.cols_index[l]) {
                            sum += Avalue * B_t.values[l];
                            break;
                        }
                    if (sum != std::complex < int >(0, 0)) {
                        result[i][k] += sum;
                    }
                }
            }

        }
    }

    return result;
}

我设置了一个用于循环的函数10迭代，从而给它提供1000*1000矩阵，并使用过op_get_wtime（），这是结果：

iteration 1 : 0.751642 s
iteration 2 : 0.911264 s
iteration 3 : 1.553695 s
iteration 4 : 0.761839 s
iteration 5 : 0.603688 s
iteration 6 : 0.423919 s
iteration 7 : 0.423114 s
iteration 8 : 0.445878 s
iteration 9 : 0.892305 s
iteration 10 : 0.918682 s

正常吗？还是我有虚假的共享或比赛状况？