OpenMP declares simd functions and / ref pointer arguments

Question

I need to vectorize a C ++ function that gives a lot (say 4) of results. The caller uses these results in local computing. Assuming that I use simdlen(4), I need a small constant space for numbers 16 doubleor 4 of the vector, each of which contains 4 elements. I do not want to store these numbers in a large global array. Let's say an example function looks like this:

#pragma omp declare simd simdlen(4) linear(out:4)
void multifun(double v1, double v2, double *out)
{
  out[0] = (v1+v2)/2;
  out[1] = (v1-v2)/2;
  out[2] = (v1+v2)/3;
  out[3] = (v1-v2)/3;
}

Question . Can this vectorized function be called with outsome small local / static variable? The following does not work, i.e. The call is multifunnot vectorized

#pragma omp declare simd simdlen(4)
double test1(double v1, double v2)
{
  double out[4];
  multifun(v1, v2, out);
  return out[0] + out[1] + out[2] + out[3];
}

- out , . ?

, , ,

#pragma omp declare simd simdlen(4) linear(out:4)
double test2(double v1, double v2, *out)
{
  multifun(v1, v2, out);
  return out[0] + out[1] + out[2] + out[3];
}

, : () .

test1 simd, :

void test_multifun(int n, const double *v1, const double *v2, double * result)
{

// should vectorize call to test1 - how??
#pragma omp for simd
  for(int i=0; i<n; i++){
    result[i] = test1(v1[i], v2[i]);
  }
}

// this works, but requires O(n) temporary storage
#pragma omp for simd
  for(int i=0; i<n; i++){
    result[i] = test2(v1[i], v2[i], out+i*4);
  }
}

OpenMP?