The correct answer is (c) looping over the faces in the computational domain
Easiest explanation: Computation of gradient for the overall domain is more efficient than calculating locally element by element. For this purpose, all the faces are looped over and the gradients are updated.