%outputs_9, %control_10 = tf_executor.island wraps "tf.Mul"(%outputs_7, %outputs) {T = "tfdtype$DT_FLOAT", _tpu_replicate = "cluster", device = "", name = "mul"} : (tensor<0xf32>, tensor<f32>) -> tensor<0xf32>
      %outputs_11, %control_12 = tf_executor.island wraps "tf.Placeholder"() {device = "", dtype = "tfdtype$DT_FLOAT", name = "y", shape = "tfshape$dim { }"} : () -> tensor<0xf32>

  }

  // CHECK-LABEL: @body(%arg0: tensor<0xf32>) -> tensor<0xf32>
  func.func private @body(%arg0:tensor<0xf32>, %arg1:tensor<*x!tf_type.resource>) -> (tensor<0xf32>, tensor<*x!tf_type.resource>) {
    %val = "tf.ReadVariableOp"(%arg1) : (tensor<*x!tf_type.resource>) -> (tensor<0xf32>)
    %0 = "tf.Mul"(%val, %arg0) : (tensor<0xf32>, tensor<0xf32>) -> tensor<0xf32>
    func.return %0, %arg1 : tensor<0xf32>, tensor<*x!tf_type.resource>
  }

  %2:2 = "tf.RecvTPUEmbeddingActivations"() {_tpu_embedding_layer = "call1", config = "\0A\0B\0C\0D"} : () -> (tensor<2x2xf32>, tensor<4x4xf32>)
  func.return
}

// -----

func.func @noinline0() -> (tensor<0xf32>) {
  %cst = "tf.Const"() {value = dense<1.0> : tensor<0xf32>} : () -> tensor<0xf32>
  return %cst : tensor<0xf32>
}

// -----

// Tests that the function not listed in the option `noinline-functions`
// is not marked tf._noinline = true.

// CHECK-LABEL: @inline
// CHECK-NOT: tf._noinline
func.func @inline() -> (tensor<0xf32>) {

    %outputs_2, %control_2 = tf_executor.island wraps "tf.Identity"(%outputs_1) {T = "tfdtype$DT_FLOAT", _tpu_input_identity = true, _tpu_replicate = "cluster", device = "", name = "replicated_input_1"} : (tensor<0xf32>) -> tensor<0xf32>

    %0 = tf_executor.graph {
      %outputs, %control = tf_executor.island wraps "tf.Const"() {device = "", dtype = f32, value = dense<[]> : tensor<0xf32>} : () -> tensor<0xf32>
      %outputs_0, %control_1 = tf_executor.island wraps "tf.Const"() {device = "", dtype = f32, value = dense<[]> : tensor<0xf32>} : () -> tensor<0xf32>

  %10 = "tfl.pseudo_const"() {value = dense<0.000000e+00> : tensor<3xf32>} : () -> tensor<3xf32>
  %11 = "tfl.pseudo_const"() {value = dense<1.000000e+00> : tensor<3xf32>} : () -> tensor<3xf32>
  %recurrent_input = "tfl.pseudo_const"() {value = dense<0.000000e+00> : tensor<1x3xf32>} : () -> tensor<1x3xf32>
  %recurrent_stats = "quantfork.stats"(%recurrent_input) {layerStats = dense<[0.0, 1.0]> : tensor<2xf32>} : (tensor<1x3xf32>) -> tensor<1x3xf32>

  %dense_default_0 = "tf.Const"() {device = "/device:CPU:0", dtype = f32, value = dense<[]> : tensor<0xf32>} : () -> tensor<0xf32>
  %dense_default_1 = "tf.Const"() {device = "/device:CPU:0", dtype = f32, value = dense<[]> : tensor<0xf32>} : () -> tensor<0xf32>
  %dense_keys = "tf.Const"() {device = "/device:CPU:0", dtype = !tf_type.string, value = dense<""> : tensor<2x!tf_type.string>} : () -> tensor<2x!tf_type.string>

  // CHECK-SAME: tf_saved_model.initializer_type = "restore_op"
  // CHECK: %[[VAR:.*]] = "tf.VarHandleOp"
  // CHECK-SAME: "var1"
  // CHECK: %[[CST:.*]] = arith.constant dense<> : tensor<0xf32>
  // CHECK: "tf.AssignVariableOp"(%[[VAR]], %[[CST]])
}

// -----

// Test with an existing session init function.
module attributes {tf_saved_model.semantics, tf_saved_model.under_construction} {

"TANH", input_to_cell_intermediate = tensor<0xf32>, input_to_forget_intermediate = tensor<0xf32>, input_to_input_intermediate = tensor<0xf32>, input_to_output_intermediate = tensor<0xf32>, proj_clip = 0.000000e+00 : f32, time_major = false} : (tensor<?x?x?xf32>, tensor<?x?xf32>, tensor<?x?xf32>, tensor<?x?xf32>, tensor<?x?xf32>, tensor<?x?xf32>, tensor<?x?xf32>, tensor<?x?xf32>, tensor<?x?xf32>, tensor<?xf32>, tensor<?xf32>, tensor<?xf32>, tensor<?xf32>, tensor<?xf32>, tensor<?xf32>, tensor<?xf32>,...