if (grad_inputs_provided) {
      const float const3_val[] = {1.0, 1.0, 1.0, 1.0};
      const3 = FloatConst2x2(expected_graph_, s_, const3_val, "GradInputs");
    } else {
      const3 = OnesLike(expected_graph_, s_, matmul, "gradients/OnesLike");
    }

    TF_Operation* matmul1 = MatMul(expected_graph_, s_, const3, const1,
                                   "gradients/MatMul", false, true);

// `dx` are used as initial gradients (which represent the symbolic partial
// derivatives of some loss function `L` w.r.t. `y`).
// `dx` must be nullptr or have size `ny`.
// If `dx` is nullptr, the implementation will use dx of `OnesLike` for all
// shapes in `y`.
// The partial derivatives are returned in `dy`. `dy` should be allocated to
// size `nx`.
//
// Gradient nodes are automatically named under the "gradients/" prefix. To

*   Correctly handle CuDNN RNN weight loaded when nest in `TimeDistributed`.
*   Adding per-element weight support for `WALSComputePartialLhsAndRhsOp`.
*   ZerosLike and OnesLike ops treated as constants by Graph Transform Tool.
*   Gamma distribution and the derived distributions (Beta, Dirichlet, Student's
    t, inverse Gamma) now fully reparameterized.