%s1 = "tf.Const"() {value = dense<[1, 2]> : tensor<2xi32>} : () -> tensor<2xi32>
  %r0, %r1 = "tf.BroadcastGradientArgs"(%s0, %s1) {} : (tensor<2xi32>, tensor<2xi32>) -> (tensor<0xi32>, tensor<0xi32>)

  // CHECK-DAG: %[[R:.*]] = "tf.Const"() <{value = dense<> : tensor<0xi32>}> : () -> tensor<0xi32>
  // CHECK-NOT: tf.BroadcastGradientArgs
  // CHECK: return %[[R]], %[[R]]

  func.return %r0, %r1 : tensor<0xi32>, tensor<0xi32>
}

Clustered nodes: 2385
Unclustered nodes: 4221
Number of clusters: 30

unclustered size 4221
 Add 17
 AddN 1
 All 1
 ApplyAdam 38
 Assert 7
 Assign 47
 AssignAdd 2
 AssignSub 2
 BroadcastGradientArgs 44
 Cast 38
 ConcatV2 3
 Const 875
 ControlTrigger 5
 Enter 874
 Equal 4
 Exit 69
 ExpandDims 9
 Fill 5
 FloorMod 1
 GreaterEqual 7
 Identity 113
 IsVariableInitialized 1
 IteratorGetNext 1

module attributes {tf.versions = {producer = 179 : i32}} {
  func.func @main() -> (tensor<0xi32>, tensor<0xi32>) {
    %0 = "tf.Const"() {value = dense<[]> : tensor<0xi32>} : () -> tensor<0xi32>
    %r0, %r1 = "tf.BroadcastGradientArgs"(%0, %0) {T = i32} : (tensor<0xi32>, tensor<0xi32>) -> (tensor<0xi32>, tensor<0xi32>)
    func.return %r0, %r1 : tensor<0xi32>, tensor<0xi32>
  }
}

// CHECK-LABEL: HloModule main

  elif act == 'TANH':
    y = math_ops.conj(y)
    grad = gen_math_ops.tanh_grad(y, grad)

  broadcast_shape = tf.shape(y)
  input_value_shape = tf.shape(op.inputs[2])
  _, reduction_axes = tf.raw_ops.BroadcastGradientArgs(
      s0=broadcast_shape, s1=input_value_shape)
  updates_grad_reshaped = tf.reduce_sum(
      grad, axis=reduction_axes, keepdims=True)
  bias_grad = tf.reshape(updates_grad_reshaped, input_value_shape)

  Output s_max = Shape(scope, max);

  Output min_mask = Less(scope, x, min);
  Output max_mask = Greater(scope, x, max);

  auto r_min = internal::BroadcastGradientArgs(scope, s_x, s_min);
  auto r_max = internal::BroadcastGradientArgs(scope, s_x, s_max);

  Output grad = grad_inputs[0];
  Output zeros = ZerosLike(scope, grad);

  Output x_grad =

}

// -----

// CHECK-LABEL: func @testBroadcastGradientArgs
func.func @testBroadcastGradientArgs(%s0: tensor<4xi32>, %s1: tensor<4xi32>) -> (tensor<1xi32>, tensor<0xi32>) {
  // CHECK:     tf.BroadcastGradientArgs
  %r0, %r1 = "tf.BroadcastGradientArgs"(%s0, %s1) : (tensor<4xi32>, tensor<4xi32>) -> (tensor<1xi32>, tensor<0xi32>)
  func.return %r0, %r1 : tensor<1xi32>, tensor<0xi32>
}

// -----

// CHECK-LABEL: @acos

    %s1 = "tf.Const"() {value = dense<[  1, 1,  1, 1280]> : tensor<4xi32>} : () -> tensor<4xi32>

    // CHECK: "tf.BroadcastGradientArgs"
    // CHECK-NOT: _xla_outside_compilation
    %r0, %r1 = "tf.BroadcastGradientArgs"(%s0, %s1) {} : (tensor<4xi32>, tensor<4xi32>) -> (tensor<1xi32>, tensor<3xi32>)
    tf_device.return

REGISTER_NO_GRADIENT_OP("InvertPermutation");
REGISTER_NO_GRADIENT_OP("Shape");
REGISTER_NO_GRADIENT_OP("ShapeN");
REGISTER_NO_GRADIENT_OP("Rank");
REGISTER_NO_GRADIENT_OP("Size");
REGISTER_NO_GRADIENT_OP("BroadcastGradientArgs");
REGISTER_NO_GRADIENT_OP("OneHot");

Status PackGrad(const Scope& scope, const Operation& op,
                const std::vector<Output>& grad_inputs,
                std::vector<Output>* grad_outputs) {
  int N;

  }

  // CHECK-LABEL: func @testBroadcastGradientArgs
  func.func @testBroadcastGradientArgs(%s0: tensor<4xi32>, %s1: tensor<4xi32>) -> (tensor<1xi32>, tensor<0xi32>) {
    // CHECK:     tf.BroadcastGradientArgs
    %r0, %r1 = "tf.BroadcastGradientArgs"(%s0, %s1) : (tensor<4xi32>, tensor<4xi32>) -> (tensor<1xi32>, tensor<0xi32>)
    func.return %r0, %r1 : tensor<1xi32>, tensor<0xi32>
  }

  // CHECK-LABEL: @acos

  func.return %0 : tensor<3xf32>
}

// -----

// Test valid tf.BroadcastGradientArgs
// CHECK-LABEL: func @testBroadcastGradientArgs
func.func @testBroadcastGradientArgs(%s0: tensor<4xi32>, %s1: tensor<4xi32>) -> (tensor<1xi32>, tensor<0xi32>) {
  %r0, %r1 = "tf.BroadcastGradientArgs"(%s0, %s1) : (tensor<4xi32>, tensor<4xi32>) -> (tensor<1xi32>, tensor<0xi32>)