// CHECK-SAME: %arg0,
  // CHECK-SAME: %[[DQ_1]], %[[DQ_2]], %[[DQ_3]], %[[DQ_4]],
  // CHECK-SAME: %[[DQ_5]], %[[DQ_6]], %[[DQ_7]], %[[DQ_8]],
  // CHECK-SAME: %[[NONE]], %[[NONE]], %[[NONE]],
  // CHECK-SAME: %[[DQ_9]], %[[DQ_10]], %[[DQ_9]], %[[DQ_9]],
  // CHECK-SAME: %[[NONE]], %[[NONE]],
  // CHECK-SAME: %[[DQ_11]], %[[DQ_12]],
  // CHECK-SAME: %[[NONE]], %[[NONE]], %[[NONE]], %[[NONE]]) <{

// PerChannel-SAME:                                             %arg0: tensor<1x3x4x3xf32>,
// PerChannel-SAME:                                             %arg1: tensor<2x3x3x2xf32>)

// PerChannel-LABEL: func private @composite_depthwise_conv2d_fn_0(
// PerChannel-SAME:                                             %arg0: tensor<1x3x4x3xf32>,

      // CHECK-SAME: f = @_func
      // CHECK-SAME: device = "/device:test_device:0"
      %3 = "tf_device.launch_func"(%2) {device = "/device:test_device:0", func = @_func} : (tensor<?xf32>) -> tensor<?xf32>

      // CHECK: %[[CALL_OUTPUT_1:[0-9]*]] = "tf.PartitionedCall"(%[[CALL_OUTPUT_0]])
      // CHECK-SAME: f = @_func
      // CHECK-SAME: device = "/device:test_device:1"

  // CHECK-SAME: Tin = [f32]
  // CHECK-SAME: Tout = [f32]
  // CHECK-SAME: allowed_batch_sizes = [6]
  // CHECK-SAME: batch_timeout_micros = 100000 : i64
  // CHECK-SAME: batching_queue = ""
  // CHECK-SAME: container = ""
  // CHECK-SAME: enable_large_batch_splitting = false
  // CHECK-SAME: max_batch_size = 6 : i64
  // CHECK-SAME: max_enqueued_batches = 10 : i64

# it is being invoked.
# CHECK: module
# CHECK-SAME: tf.versions
# CHECK-SAME: bad_consumers
# CHECK-SAME: min_consumer
# CHECK-SAME: producer

# CHECK: "tf_saved_model.global_tensor"()
# CHECK: "tf_saved_model.session_initializer"() <{initializers = [@[[init:.*]]]}> : () -> ()

# CHECK:      func @[[init]]
# CHECK-SAME: tf_saved_model.initializer_type = "init_op"
# CHECK-NEXT: [[R6:%.*]] = "tf.Const"()

    // CHECK: %[[CALL:.+]] = "tf.XlaCallModule"(%[[ARG0]], %[[ARG1]])
    // CHECK-SAME: _entry_function = @composite_dot_general_fn_1
    // CHECK-SAME: _original_entry_function = "composite_dot_general_fn_1"
    // CHECK: return %[[CALL]]
  }

  // CHECK: @composite_dot_general_fn_1
  // CHECK-SAME: %[[ARG2:[^:[:space:]]+]]
  // CHECK-SAME: %[[ARG3:[^:[:space:]]+]]

  // CHECK-SAME: quant.uniform<i8:f32, 0.023529411764705882:-128>
  // CHECK: %[[dq2:.*]] = "quantfork.dcast"(%[[q2]])
  // CHECK-SAME: quant.uniform<i8:f32, 0.023529411764705882:-128>
  // CHECK: %[[q3:.*]] = "quantfork.qcast"(%[[cst3]])
  // CHECK-SAME: quant.uniform<i8:f32, 3.9215686274509805E-9>
  // CHECK: %[[dq3:.*]] = "quantfork.dcast"(%[[q3]])
  // CHECK-SAME: quant.uniform<i8:f32, 3.9215686274509805E-9>

  // CHECK: %[[CONV2D:[0-9]*]] = "tf.Conv2D"(%[[ARG_TRANSPOSE]], %arg1)
  // CHECK-SAME: data_format = "NHWC"
  // CHECK-SAME: dilations = [1, 3, 4, 2]
  // CHECK-SAME: explicit_paddings = [1, 2, 5, 6, 7, 8, 3, 4]
  // CHECK-SAME: padding = "EXPLICIT"
  // CHECK-SAME: strides = [5, 7, 8, 6]
  // CHECK-SAME: (tensor<1x32x32x3xf32>, tensor<1x1x3x8xf32>) -> tensor<1x7x6x8xf32>

// CHECK-LABEL: func private @composite_depthwise_conv2d_fn(
// CHECK-SAME:                                             %arg0: tensor<1x3x4x3xf32>,
// CHECK-SAME:                                             %arg1: tensor<2x3x3x2xf32>)

// CHECK-LABEL: func private @composite_depthwise_conv2d_fn_0(
// CHECK-SAME:                                             %arg0: tensor<1x3x4x3xf32>,
// CHECK-SAME:                                             %arg1: tensor<2x3x1x6xf32>)

// CHECK-SAME: %[[ARG_0:[a-z0-9]*]]: tensor<?x2xi32>
// CHECK-SAME: %[[ARG_1:[a-z0-9]*]]: tensor<?x2xi32>
// CHECK-SAME: %[[ARG_2:[a-z0-9]*]]: tensor<?x2xi32>
// CHECK-SAME: %[[ARG_3:[a-z0-9]*]]: tensor<?xi32>
// CHECK-SAME: %[[ARG_4:[a-z0-9]*]]: tensor<?xi32>
// CHECK-SAME: %[[ARG_5:[a-z0-9]*]]: tensor<?xi32>
// CHECK-SAME: %[[ARG_6:[a-z0-9]*]]: tensor<!tf_type.string>
// CHECK-SAME: %[[ARG_7:[a-z0-9]*]]: tensor<!tf_type.string>