%0 = tf_executor.graph {
    %outputs, %control = tf_executor.island wraps "tf.GuaranteeConst"(%arg1) {T = f32, device = ""} : (tensor<f32>) -> tensor<f32>
    %outputs_0, %control_1 = tf_executor.island wraps "tf.GuaranteeConst"(%arg2) {T = f32, device = ""} : (tensor<f32>) -> tensor<f32>

    %3 = "tf.Const"() {value = dense<2.0> : tensor<f32>} : () -> tensor<f32>
    tf_device.return %2, %3 : tensor<i32>, tensor<f32>
  }) {cluster_attr = "cluster_attr"} : () -> (tensor<i32>, tensor<f32>)
  %4 = "tf.Const"() {value = dense<3.0> : tensor<f32>} : () -> tensor<f32>
  %5 = "tf.Add"(%4, %1) : (tensor<f32>, tensor<f32>) -> tensor<f32>
  %6 = "tf.Const"() {value = dense<4> : tensor<i32>} : () -> tensor<i32>

    %0 = "tf.FakeQuantWithMinMaxArgs"(%arg0) {device = "", max = 2.000000e-01 : f32, min = -1.000000e-01 : f32, narrow_range = false, num_bits = 8 : i64} : (tensor<1x3x4x3xf32>) -> tensor<1x3x4x3xf32>

    %2 = "quantfork.dcast"(%1) : (tensor<4x3x!quant.uniform<i8<-127:127>:f32:1, {5.000000e-03, 5.000000e-03, 5.000000e-03}>>) -> tensor<4x3xf32>
    %3 = "quantfork.qcast"(%arg0) {volatile} : (tensor<1x4xf32>) -> tensor<1x4x!quant.uniform<i8:f32, 6.000000e-03:-128>>
    %4 = "quantfork.dcast"(%3) : (tensor<1x4x!quant.uniform<i8:f32, 6.000000e-03:-128>>) -> tensor<1x4xf32>

      %min_range = "tf.Const"() { value = dense<1.0> : tensor<f32> } : () -> tensor<f32>
      %max_range = "tf.Const"() { value = dense<5.0> : tensor<f32> } : () -> tensor<f32>
      %0 = "tf.Dequantize"(%arg0, %min_range, %max_range) : (tensor<1x!tf_type.qint8>, tensor<f32>, tensor<f32>) -> tensor<1xf32>
      func.return %0 : tensor<1xf32>
    }
  })mlir";

  %2 = "tf.Const"() {value = dense<4.000000e+00> : tensor<f32>} : () -> tensor<f32>
  %3 = tf_executor.graph {
    %res, %ctl = tf_executor.island {
      %3 = "tf_device.cluster"() ({

        // In the device region, check that the 3 constants are materialized and
        // remapped to the uses.
        // CHECK: tf_device.cluster
        // CHECK-DAG: %[[CST2:.*]] = "tf.Const"{{.*}}2.0

// CHECK-SAME:    (%[[ARG0:.*]]: tensor<i32>, %[[ARG1:.*]]: tensor<f32>)
func.func @identity_n(%arg0: tensor<i32>, %arg1: tensor<f32>) -> (tensor<i32>, tensor<f32>) {
  // CHECK-NOT: "tf.IdentityN"
  %0:2 = "tf.IdentityN"(%arg0, %arg1) : (tensor<i32>, tensor<f32>) -> (tensor<i32>, tensor<f32>)
  // CHECK: return %[[ARG0]], %[[ARG1]]
  func.return %0#0, %0#1 : tensor<i32>, tensor<f32>

      %4 = "tf.opB"() : () -> tensor<f32>
      tf_executor.yield %4 : tensor<f32>
    }
    tf_executor.fetch %2#0 : tensor<f32>
  }
  func.return %0 : tensor<f32>
}

// CHECK-LABEL: @func2
func.func @func2(%arg0 : tensor<i1>) -> tensor<i1> {
  %0 = tf_executor.graph {
    %1:2 = tf_executor.island {
      %4 = "tf.opB"() : () -> tensor<f32>
      tf_executor.yield %4 : tensor<f32>
    }
// CHECK: island

# CHECK:       ENTRY %main.{{[0-9]+}} ([[ARG0:.*]]: f32[2], [[ARG1:.*]]: f32[2]) -> (f32[2]) {
# CHECK-NEXT:    %[[ARG1]] = f32[2]{0} parameter(1)
# CHECK-NEXT:    %[[ARG0]] = f32[2]{0} parameter(0)
# CHECK-NEXT:    ROOT %tuple.{{[0-9]+}} = (f32[2]{0}) tuple(f32[2]{0} %[[ARG0]])
# CHECK-NEXT:  }

# CHECK:       // InputMapping {0, 1}
# CHECK-NEXT:  // XlaInputShape f32[2]
# CHECK-NEXT:  // XlaInputShape f32[2]
# CHECK-NEXT:  // XlaOutputShape (f32[2])

  "tf.AssignVariableOp"(%arg2, %1#1) : (tensor<*x!tf_type.resource<tensor<i32>>>, tensor<i32>) -> ()
  func.return
}

// CHECK-LABEL: func @read_write_func
// CHECK-SAME: ({{%.*}}: tensor<i32>, {{%.*}}: tensor<f32>) -> (tensor<f32>, tensor<i32>)
func.func @read_write_func(%arg0: tensor<i32>, %arg1: tensor<f32>) -> (tensor<f32>, tensor<i32>) {