non deterministic results for ippsDiv_32f_I

[Josephnyu]

I'm using Intel Integrated Performance Primitives (windows, v2019 (2019.0.5.1052), 64bit) and I came across unexpected results.

When I run the following code 

const int N = 20;
std::vector<float> a(N, 1.0209482f);
const std::vector<float> b(N,0.032953884f);
    

ippsDiv_32f_I(b.data(), a.data(), N);
std::cout << "result : " << a[0] << std::endl;

sometimes I get 30.981117 and sometimes 30.98112

Even though initially a and b contain equal values, after calling ippsDiv_32f_I values in a varies.
At each run it gives me different output.

I understand that that the function has some degree of accuracy. I wouldn't expect non-deterministic behaviour.

Is it expected ? documented ?
@Ruqiu_C_Intel , @Gennady_F_Intel