The Jacobian of a ReLU Network

A two-layer network with a ReLU in the middle computes

$f(x) = W_2\,\mathrm{ReLU}(W_1 x)$

with

$W_1 = \begin{bmatrix} 1 & -2 \\ 2 & 1 \\ -1 & 1 \end{bmatrix} \qquad W_2 = \begin{bmatrix} 1 & 2 & 2 \\ 3 & -1 & 1 \end{bmatrix}$

so the input is 2-dimensional, the hidden layer has 3 units, and the output is 2-dimensional.

$\mathrm{ReLU}$ acts entrywise, and its derivative is $1$ wherever its input is positive and $0$ wherever its input is negative.

Evaluate at $x = (3,\ 1)$. What is the Jacobian $\partial f/\partial x$ at this point — a $2 \times 2$ matrix?