*    dA = U * B.T
     *    dB = A.T * U
     *
     *    where A.T means `transpose(A)`
     */
    AbstractTensorHandle* upstream_grad = grad_outputs[0];

    // Get transpose attrs
    bool t_a;
    TF_RETURN_IF_ERROR(forward_attrs_.Get("transpose_a", &t_a));

    bool t_b;
    TF_RETURN_IF_ERROR(forward_attrs_.Get("transpose_b", &t_b));

    // Conj each input
    AbstractTensorHandle* conj_output;

                       TF_Operation* r, const char* name,
                       bool transpose_a = false, bool transpose_b = false) {
    TF_OperationDescription* desc = TF_NewOperation(graph, "MatMul", name);
    if (transpose_a) {
      TF_SetAttrBool(desc, "transpose_a", 1);
    }
    if (transpose_b) {
      TF_SetAttrBool(desc, "transpose_b", 1);
    }
    TF_AddInput(desc, {l, 0});
    TF_AddInput(desc, {r, 0});