%9 = mhlo.floor %8 : tensor<64xf32>
  %10 = mhlo.convert %9 : (tensor<64xf32>) -> tensor<64xi32>
  %11 = mhlo.compare  LT, %10, %1,  SIGNED : (tensor<64xi32>, tensor<64xi32>) -> tensor<64xi1>
  %12 = mhlo.add %10, %0 : tensor<64xi32>
  %13 = mhlo.select %11, %12, %10 : tensor<64xi1>, tensor<64xi32>
  %14 = mhlo.reshape %13 : (tensor<64xi32>) -> tensor<64x1xi32>

  "tf_saved_model.global_tensor"() {is_mutable, sym_name = "y", type = tensor<3x1xi32>, value = dense<[[1], [2], [3]]> : tensor<3x1xi32>} : () -> ()
  "tf_saved_model.asset"() {sym_name = "z", filename = "file"} : () -> ()
  func.func @serving_default(
      %arg0: tensor<1x3xi32> {tf_saved_model.index_path = ["x"]},
      %arg1: tensor<!tf_type.resource<tensor<3x1xi32>>> {tf_saved_model.bound_input = @y},

// CHECK:           %[[VAL_4:.*]] = "tf.GatherNd"(%[[VAL_3]], %[[VAL_1]]) <{bad_indices_policy = ""}> : (tensor<20x1xi32>, tensor<1x1xi32>) -> tensor<1x1xi32>
// CHECK:           %[[VAL_5:.*]] = "tf.Const"() <{value = dense<[1, 0]> : tensor<2xi64>}> : () -> tensor<2xi64>
// CHECK:           %[[VAL_6:.*]] = "tf.Transpose"(%[[VAL_4]], %[[VAL_5]]) : (tensor<1x1xi32>, tensor<2xi64>) -> tensor<1x1xi32>

  func.return %resize: tensor<1x3x3x1xi32>
}

// CHECK-LABEL: func @imag_resize_nearest_full_dyn
func.func @imag_resize_nearest_full_dyn(%arg0: tensor<1x?x?x1xi32>, %arg1: tensor<2xi32>) -> tensor<1x?x?x1xi32> {

  // CHECK-DAG: [[VAL0:%.+]] = "tf.Const"() <{value = dense<1> : tensor<i32>}>

    %1 = "tf.Const"() {value = dense<0> : tensor<1x1xi32>} : () -> tensor<1x1xi32>
    %2 = "tf.Const"() {value = dense<[7, 7, 3, 64]> : tensor<4xi32>} : () -> tensor<4xi32>
    %3 = "tf.Const"() {value = dense<[[0, 0], [3, 3], [3, 3], [0, 0]]> : tensor<4x2xi32>} : () -> tensor<4x2xi32>
    %4 = "tf.Const"() {value = dense<0> : tensor<i32>} : () -> tensor<i32>

  %0 = "tf.ConcatV2"(%arg, %arg, %axis) : (tensor<8x16xf32>, tensor<8x16xf32>, tensor<1x1xi32>) -> tensor<*xf32>
  func.return %0 : tensor<*xf32>
}

// -----

func.func @testConcatV2(%arg: tensor<8x16xf32>, %axis: tensor<1x1xi32>) -> tensor<*xf32> {

  // CHECK: "mhlo.scatter"
  %0 = "tf.TensorScatterUpdate"(%tensor, %indices, %updates) : (tensor<4x3xi32>, tensor<2x1xi32>, tensor<i32>) -> tensor<4x3xi32>
  func.return %0 : tensor<4x3xi32>
}

// -----

// CHECK-LABEL: @tensor_scatter_add
func.func @tensor_scatter_add(%tensor: tensor<?x?x?xf32>, %indices: tensor<?x2xi32>, %updates: tensor<?x?xf32>) -> tensor<?x?x?xf32> {
  // CHECK: "mhlo.scatter"(%arg0, %arg1, %arg2)

  func.return %0#0 : tensor<*xi32>
}

// compute_3 has different argument types from compute_1, thus error is reported.
// CHECK: error: func_ops for BatchFunctionOp with the same shared name are different
func.func private @compute_3(%arg0: tensor<?x1xi32> {tf._user_specified_name = "0"}, %arg1: tensor<?x1xi32>) -> (tensor<?x1xi32>, tensor<?x1xi32>) {

  %group_assignment = "tf.Const"() { value = dense<[[0],[1]]> : tensor<2x1xi32> } : () -> tensor<2x1xi32>
  // CHECK: "mhlo.all_reduce"
  // CHECK{LITERAL}: replica_groups = dense<[[0], [1]]> : tensor<2x1xi64>
  // CHECK-NOT: channel_handle
  // CHECK: mhlo.add
  %0 = "tf.XlaAllReduce"(%input, %group_assignment) {reduce_op = "Add", mode = "CrossReplica"} : (tensor<f32>, tensor<2x1xi32>) -> tensor<f32>
  func.return %0 : tensor<f32>
}

// -----

  // converted to NHWC data format.

  // Padding in spatial dimension (NCHW)
  %0 = "tf.Const"() {value = dense<[[0, 0], [0, 0], [3, 3], [3, 3]]> : tensor<4x2xi32>} : () -> tensor<4x2xi32>

  // Reduce over spatial dimensions (NCHW)
  %1 = "tf.Const"() {value = dense<[2, 3]> : tensor<2xi32>} : () -> tensor<2xi32>

  // Transpose input: NHWC -> NCHW