%cluster = "tf_device.cluster_func"(%arg0) {func = @_func} : (tensor<i32>) -> tensor<i32>
      return %cluster : tensor<i32>
    }

    func @_func(%arg0: tensor<i32>) -> tensor<i32> {
      %identity = "tf.Identity"(%arg0) : (tensor<i32>) -> tensor<i32>
      return %identity : tensor<i32>
    }
    ```
  }];

  let constructor = "TFDevice::CreateClusterOutliningPass()";
}

      func.func @main() -> tensor<!tf_type.variant<tensor<64x1xbf16>>> {
      // unknown rank
      %elem_shape = "tf.Const"() <{value = dense<-1> : tensor<i32>}> {device = "/job:localhost/replica:0/task:0/device:CPU:0"} : () -> tensor<i32>
      // zero reserved elements

// CHECK-LABEL: func @move_broadcast
func.func @move_broadcast(%arg0: tensor<f32>) -> () {
  // CHECK:      %[[ELEM_0:.*]] = "tf.Const"()
  // CHECK:      {ici_weight_distribution_mlir_bridge_marker = true}

  %cst = "tf.Const"() <{value = dense<1> : tensor<i32>}> {device = ""} : () -> tensor<i32>
  %0 = "tf.MultiDeviceIteratorFromStringHandle"(%arg0) <{output_shapes = [#tf_type.shape<>], output_types = [!tf_type.string]}> {device = ""} : (tensor<!tf_type.string>) -> tensor<!tf_type.resource>
  %1 = "tf.MultiDeviceIteratorGetNextFromShard"(%0, %cst, %arg1) {device = ""} : (tensor<!tf_type.resource>, tensor<i32>, tensor<i64>) -> tensor<!tf_type.string>

  %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 %3 : tensor<2x1xf32>
})";
  const std::string kExpectedFB = CreateRuntimeMetadata();

// input tensor's full range. Strides for all resultant slices are all one.
//
// For example, the following source IR:
//
//   %dim = "tf.Const"() {value = dense<1> : tensor<i32>} : () -> tensor<i32>
//   %0:3 = "tf.Split"(%dim, %input) : (tensor<i32>, tensor<4x6xf32>) ->
//                (tensor<4x2xf32>, tensor<4x2xf32>, tensor<4x2xf32>)
//
// will be converted into:
//
//   %0 = "mhlo.slice"(%input) {