%1 = "arith.constant"() {value = dense<0> : tensor<16xi64>} : () -> tensor<16xi64>
  %2 = "tfl.conv_2d"(%arg0, %0, %1) {dilation_h_factor = 1 : i32, dilation_w_factor = 1 : i32, fused_activation_function = "NONE", padding = "SAME", stride_h = 1 : i32, stride_w = 1 : i32} : (tensor<1x32x32x8x!quant.uniform<i16:f32, 1.0>>, tensor<16x1x1x8x!quant.uniform<i8:f32, 1.0>>, tensor<16xi64>) -> tensor<1x32x32x16x!quant.uniform<i16:f32, 1.0>>

  %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>

  %0 = "vhlo.dot_general_v1"(%arg0, %arg1) <{
    lhs_batching_dimensions = #vhlo.tensor_v1<dense<0> : tensor<1xi64>>,
    lhs_contracting_dimensions = #vhlo.tensor_v1<dense<2> : tensor<1xi64>>,
    rhs_batching_dimensions = #vhlo.tensor_v1<dense<0> : tensor<1xi64>>,
    rhs_contracting_dimensions = #vhlo.tensor_v1<dense<0> : tensor<1xi64>>,

// CHECK-NEXT:      %18 = "tfl.pseudo_const"() <{value = dense<1> : tensor<1xi64>}> : () -> tensor<1xi64>
// CHECK-NEXT:      %19 = "tfl.gather"(%15, %18) <{axis = 0 : i32, batch_dims = 0 : i32}> : (tensor<3xi32>, tensor<1xi64>) -> tensor<1xi32>

func.func @testBroadcastGradientArgI64() -> (tensor<2xi64>, tensor<0xi64>) {
  %s0 = "tf.Const"() {value = dense<> : tensor<0xi64>} : () -> tensor<0xi64>
  %s1 = "tf.Const"() {value = dense<[2, 4]> : tensor<2xi64>} : () -> tensor<2xi64>
  %r0, %r1 = "tf.BroadcastGradientArgs"(%s0, %s1) {} : (tensor<0xi64>, tensor<2xi64>) -> (tensor<2xi64>, tensor<0xi64>)

    %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>)

  %0 = "quantfork.stats"(%arg0) {layerStats = dense<[0.000000e+00, 1.000000e+01]> : tensor<2xf32>} : (tensor<?x28x28x28x8xf32>) -> tensor<?x28x28x28x8xf32>
  %cst = arith.constant dense<16> : tensor<1xi64>
  %cst_0 = "tfl.no_value"() {value = unit} : () -> none
  %w = arith.constant dense<127.0> : tensor<3x3x3x8x16xf32>
  %b = arith.constant dense<0.0> : tensor<16xf32>

// CHECK-DAG:       [[VAL_36:%.*]] = arith.constant dense<1> : tensor<1xi64>
// CHECK:           [[VAL_37:%.*]] = "tf.Slice"([[VAL_3]], [[VAL_35]], [[VAL_36]]) : (tensor<2xf32>, tensor<1xi64>, tensor<1xi64>) -> tensor<1xf32>
// CHECK-DAG:       [[VAL_38:%.*]] = arith.constant dense<1> : tensor<1xi64>
// CHECK-DAG:       [[VAL_39:%.*]] = arith.constant dense<1> : tensor<1xi64>

    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"}> ({