// CHECK: %[[XLA_CALL_MODULE:.+]] = "tf.XlaCallModule"(%[[CUSTOM_AGGREGATOR_0]], %[[CST]])
// CHECK-SAME: _entry_function = @composite_dot_general_fn_1, _original_entry_function = "composite_dot_general_fn_1"

  // CHECK-SAME: <{calibration_methods = [1 : i32, 1 : i32], ids = ["0", "1"], output_file_path = "main_0.pb"}> : (tensor<f32>, tensor<f32>, tensor<0xi64>, tensor<f32>, tensor<f32>, tensor<0xi64>) -> ()
  // CHECK: return

  %2 = mhlo.constant dense<3.40282347E+38> : tensor<f32>
  %3 = "mhlo.broadcast_in_dim"(%2) <{broadcast_dimensions = dense<> : tensor<0xi64>}> : (tensor<f32>) -> tensor<2xf32>
  %4 = mhlo.constant dense<3.000000e+00> : tensor<f32>
  %5 = "mhlo.broadcast_in_dim"(%4) <{broadcast_dimensions = dense<> : tensor<0xi64>}> : (tensor<f32>) -> tensor<2xf32>

    %0 = "tf.XlaCallModule"(%output, %cst) <{Sout = [#tf_type.shape<10x1x3>], dim_args_spec = [], disabled_checks = [], function_list = [], has_token_input_output = false, module = "", platforms = ["CPU"], version = 9 : i64}> {_entry_function...

// MIN-MAX-CHECK-NEXT:  [[lhs:%.*]], {{.*}}, {{.*}}, {{.*}} = "tf.CustomAggregator"(%arg0) <{calibration_method = 1 : i32, id = "composite_conv2d_with_relu6_fn_arg_0_calibration_method_1", max_percentile = 0.000000e+00 : f32, min_percentile = 0.000000e+00 : f32, num_bins = 0 : i32}> : (tensor<*xf32>) -> (tensor<*xf32>, tensor<f32>, tensor<f32>, tensor<0xi64>)

    func.func @main(%arg0: tensor<f32>) {
      %elem = "tf.Const"() <{value = dense<0.000000e+00> : tensor<f32>}> : () -> tensor<f32>
      %shape = "tf.Const"() <{value = dense<> : tensor<0xi64>}> : () -> tensor<0xi64>
      %zero = "tf.Fill"(%shape, %elem) : (tensor<0xi64>, tensor<f32>) -> tensor<f32>
      tf_device.replicate([%arg0, %zero] as %arg1: tensor<f32>) {n = 2 : i32} {
        %arg1_id = "tf_device.launch"() <{device = "TPU_REPLICATED_HOST_0"}> ({

  %32 = "tf.ConcatV2"(%29#0, %31, %14) {device = ""} : (tensor<2xi64>, tensor<1xi64>, tensor<i32>) -> tensor<3xi64>
  %33 = "tf.GatherV2"(%24#2, %29#1, %14) {batch_dims = 0 : i64, device = ""} : (tensor<?xi64>, tensor<?xi64>, tensor<i32>) -> tensor<?xi64>
  %34 = "tf.ConcatV2"(%33, %22, %14) {device = ""} : (tensor<?xi64>, tensor<1xi64>, tensor<i32>) -> tensor<?xi64>

  // CHECK: mhlo.reshape {{.*}} : (tensor<2x1xi64>) -> tensor<2xi64>
  // CHECK: "mhlo.slice"({{.*}}) <{limit_indices = dense<1> : tensor<1xi64>, start_indices = dense<0> : tensor<1xi64>, strides = dense<1> : tensor<1xi64>}> : (tensor<2xi64>) -> tensor<1xi64>

// CHECK-DAG:       %[[VAL_2:.*]] = "tf.Const"() <{value = dense<[1, 0]> : tensor<2xi64>}> : () -> tensor<2xi64>
// CHECK-DAG:       %[[VAL_3:.*]] = "tf.Const"() <{value = dense<[1, 0]> : tensor<2xi64>}> : () -> tensor<2xi64>
// CHECK:           %[[VAL_4:.*]] = "tf.Transpose"(%[[VAL_0]], %[[VAL_3]]) : (tensor<2x3xf32>, tensor<2xi64>) -> tensor<3x2xf32>

//        limit_indices = dense<[4, 1]> : tensor<2xi64>,
//        start_indices = dense<0> : tensor<2xi64>,
//        strides = dense<1> : tensor<2xi64>} :
//        (tensor<4x6xf32>) -> tensor<4x1xf32>
// %1 = "mhlo.slice"(%input) {
//        limit_indices = dense<[4, 3]> : tensor<2xi64>,
//        start_indices = dense<[0, 1]> : tensor<2xi64>,
//        strides = dense<1> : tensor<2xi64>} :
//        (tensor<4x6xf32>) -> tensor<4x2xf32>