// CHECK-NEXT:    return [[RES]], [[MATRES]] : tensor<1x1xf32>, tensor<1x1xf32>
//
module {
  func.func @serving_default(%arg0: tensor<1x3xf32>) -> (tensor<1x1xf32>, tensor<1x1xf32>) {
    %0 = "tf.VarHandleOp"() <{container = "", shared_name = "y"}> : () -> tensor<!tf_type.resource<tensor<3x1xf32>>>
    %2 = "tf.ReadVariableOp"(%0) : (tensor<!tf_type.resource<tensor<3x1xf32>>>) -> tensor<3x1xf32>

func.func @zeros_like(%arg0: tensor<8x16xf32>) -> tensor<8x16xf32> {
  %0 = "tf.ZerosLike"(%arg0) : (tensor<8x16xf32>) -> tensor<8x16xf32>
  func.return %0 : tensor<8x16xf32>
// CHECK-LABEL: zeros_like
// CHECK:  "tfl.zeros_like"(%arg0) : (tensor<8x16xf32>) -> tensor<8x16xf32>
}

func.func @div(%arg0: tensor<1xf32>, %arg1: tensor<1xf32>) -> tensor<1xf32> {
  %0 = "tf.Div"(%arg0, %arg1) : (tensor<1xf32>, tensor<1xf32>) -> tensor<1xf32>

  %2 = "tfl.add"(%arg0, %arg3) {fused_activation_function = "RELU6", per_device_costs = {CPU = 5.0 : f32, GPU = 1.0 : f32}, tac.device = "GPU"} : (tensor<1xf32>, tensor<1xf32>) -> tensor<1xf32>
  %3 = "tfl.pack"(%1, %2) {axis = 0 : i32, per_device_costs = {CPU = 2.0 : f32, GPU = -1.0 : f32}, values_count = 2 : i32, tac.device = "CPU"} : (tensor<1xf32>, tensor<1xf32>) -> tensor<2x1xf32>

  func.return %24 : tensor<1x4xf32>
}

    ```mlir
      %0 = "tf.Const"() {value = dense<[[42.0]]> : tensor<1x1xf32>} : () -> tensor<1x1xf32>
      %1 = "tf.Const"() {device = "", value = dense<[[42.0]]> : tensor<1x1xf32>} : () -> tensor<1x1xf32>
      %2 = "tf.Const"() {device = "baz", value = dense<[[42.0]]> : tensor<1x1xf32>} : () -> tensor<1x1xf32>
    ```

    then running this pass with 'default-device=foobar', we get:

    ```mlir

func.func private @gelu_decomp_2(%arg0: tensor<5x10xf32>) -> tensor<5x10xf32>
func.func @gelu_aten_approximate(%arg0: tensor<5x10xf32>) -> (tensor<*xf32>) {
  %0 = mhlo.composite "aten.gelu.default" %arg0 {composite_attributes = {approximate = "tanh"}, decomposition = @gelu_decomp_2} : (tensor<5x10xf32>) -> tensor<5x10xf32>

//                (tensor<4x2xf32>, tensor<4x2xf32>, tensor<4x2xf32>)
//
// will be converted into:
//
//   %0 = "mhlo.slice"(%input) {
//             limit_indices = dense<[4, 2]> : tensor<2xi64>,
//             start_indices = dense<0> : tensor<2xi64>,
//             strides = dense<1> : tensor<2xi64>} :
//        (tensor<4x6xf32>) -> tensor<4x2xf32>