%dimension = "tf.Const"() { value = dense<3> : tensor<1xi64> } : () -> tensor<1xi64>
    %6 = "tf.Sum"(%3, %dimension) { keep_dims = true }: (tensor<1x3x2x2xf32>, tensor<1xi64>) -> tensor<1x3x2x1xf32>
    return %5, %6 : tensor<1x3x2x2xf32>, tensor<1x3x2x1xf32>
  }

// CHECK-LABEL: func @conv_with_multiple_uses

  func.return %0 : tensor<4xi32>
}

func.func @i64() -> tensor<4xi64> {
  // CHECK-LABEL: @i64
  // CHECK: value = dense<[1, 2, 3, 72623859790382856]> : tensor<4xi64>
  %0 = "tfl.pseudo_const" () { value = dense<[1, 2, 3, 72623859790382856]> : tensor<4xi64> } : () -> tensor<4xi64>
  func.return %0 : tensor<4xi64>
}

func.func @string() -> tensor<2x2x!tf_type.string> {
  // CHECK-LABEL: @string

  %1 = mhlo.convert %0 : (tensor<f64>) -> tensor<i64>
  %2 = mhlo.reshape %1 : (tensor<i64>) -> tensor<1xi64>
  // expected-error @+1 {{Node `mhlo.dynamic_iota` must have compile-time constant}}
  %3 = "mhlo.dynamic_iota"(%2) <{iota_dimension = 0 : i64}> : (tensor<1xi64>) -> tensor<?xi32>
  %4 = mhlo.multiply %3, %3 : tensor<?xi32>
  return %4 : tensor<?xi32>

}

// CHECK-LABEL: func @testShapeOp
// CHECK-SAME: ({{%.+}}: tensor<*xi64> {tf.device = "/job:localhost/replica:0/task:0/device:TPU:0"})
// CHECK-SAME: -> (tensor<?xi64> {tf.device = "/job:localhost/replica:0/task:0/device:TPU:0"})
func.func @testShapeOp(%arg0: tensor<*xi64> {tf.device = "/job:localhost/replica:0/task:0/device:TPU:0"}) -> tensor<?xi64> {
  %0 = tf_executor.graph {
    // CHECK:      tf.Shape

    %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...

  EXPECT_THAT(num_bytes, Eq(16));
}

TEST_F(GetSizeInBytesTest, Float64ConstOpSizeInBytes) {
  constexpr absl::string_view kConstOpExpr =
      R"mlir(%cst = "tf.Const"() {value = dense<3.0> : tensor<2xf64>} : () -> tensor<2xf64>)mlir";

  Block block{};
  TF::ConstOp int_tensor_const_op = ParseConstOp(kConstOpExpr, block, ctx_);

  const int64_t num_bytes = GetSizeInBytes(int_tensor_const_op);
  EXPECT_THAT(num_bytes, Eq(16));

# CHECK: operandSegmentSizes = array<i32: 1, 1, 0, 1, 1>
# CHECK: resultSegmentSizes = array<i32: 0, 0, 0, 1>
# CHECK: tf_executor.fetch %[[output]] : tensor<*xi64>

node {
  name: "serilaized"
  op: "Const"
  attr {
    key: "dtype"
    value {
      type: DT_STRING
    }
  }
  attr {
    key: "value"
    value {
      tensor {

  %5 = "tf.Identity"(%4) {_replication_info = "cluster", _xla_compile_device_type = "TPU"} : (tensor<?xi64>) -> tensor<?xi64>

  func.return %2, %5 : tensor<?xi32>, tensor<?xi64>
}

// -----

// CHECK-LABEL: func @two_clusters
func.func @two_clusters(%arg0: tensor<*xi32>, %arg1: tensor<?xi64>) -> (tensor<?xi32>, tensor<?xi64>) {

// 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>)

    %dim = "tf.Const"() { value = dense<0> : tensor<1xi64> } : () -> tensor<1xi64>

    // Check and include zero in the range so that zero value can be correctly
    // represented.
    %input_1d = "tf.Reshape"(%input, %shape) : (tensor<*xf32>, tensor<1xi32>) -> tensor<?xf32>
    %r_max_without_zero = "tf.Max"(%input_1d, %dim) { keep_dims = true }: (tensor<?xf32>, tensor<1xi64>) -> tensor<1xf32>