Values at quadrature points for derivatives of coefficients are not computed

[gmarkall]

In UFL, we can have SpatialDerivative(Coefficient( ... ))

At the moment this is not treated correctly. It results in a totally messed up quadrature looop. E.g. from the diffusion RHS:

__shared__ double c_q1[24];
__shared__ double c_q0[6];
for(int i_g = 0; i_g < 6; i_g++)
{
  (c_q0[i_g] = 0.0);
  for(int i_r_0 = 0; i_r_0 < 3; i_r_0++)
  {
    (c_q0[i_g] += c0[i_r_0]);
  };
  for(int i_d_0 = 0; i_d_0 < 2; i_d_0++)
  {
    for(int i_d_1 = 0; i_d_1 < 2; i_d_1++)
    {
      (c_q1[((i_g + (6 * i_d_0)) + (2 * (6 * i_d_1)))] = 0.0);
      for(int i_r_0 = 0; i_r_0 < 3; i_r_0++)
      {
        (c_q1[((i_g + (6 * i_d_0)) + (2 * (6 * i_d_1)))] += c1[((i_d_1 + (2 * i_d_0)) + (2 * (2 * i_r_0)))]);
      };
    };
  };
};

We should be computing the gradient of c0. The loop nest should be completely nested.

[gmarkall]

Upon further inspection, there's nothing wrong with the loop nesting. However, we should still be computing the derivative at the quadrature points, which is currently not being done.

[gmarkall]

Fixed in commit 335eba5d0a342ab0485f.