logger().trace(message, arg1, arg2)
        logger().trace(message, arg1, arg2, arg3)
        logger().trace((Marker) null, message)
        logger().trace((Marker) null, message, new Exception())
        logger().trace((Marker) null, message, arg1)
        logger().trace((Marker) null, message, arg1, arg2)
        logger().trace((Marker) null, message, arg1, arg2, arg3)

        then:

  %arg1: tensor<1x28x28x64xf32>,
  %arg2: tensor<64xf32>
) -> tensor<1x28x28x64xf32> {

  // CHECK: "tf.FusedBatchNormGradV3"
  // CHECK-SAME: (%[[X_TRANSPOSE:[0-9]*]], %[[Y_TRANSPOSE:[0-9]*]],
  // CHECK-SAME: data_format = "NCHW"
  %x_backprop, %scale_backprop, %offset_backprop, %reserve_1, %reserve_2
    = "tf.FusedBatchNormGradV3"(%arg0, %arg1, %arg2, %arg2, %arg2, %arg2)
       {
         data_format = "NHWC",

  %0 = "stablehlo.reduce"(%arg0, %arg1) ({
  ^bb0(%arg2: tensor<5xf32>, %arg3: tensor<5xf32> ):
    // CHECK:      %1 = "vhlo.abs_v1"(%arg2) : (tensor<5xf32>) -> tensor<5xf32>
    // CHECK-NEXT: %2 = tfl.add %1, %arg2 {fused_activation_function = "NONE"} : tensor<5xf32>
    // CHECK-NEXT: %3 = "vhlo.abs_v1"(%2) : (tensor<5xf32>) -> tensor<5xf32>
    %1 = stablehlo.abs %arg2 : tensor<5xf32>
    %2 = tfl.add %1, %arg2 {fused_activation_function = "NONE"} : tensor<5xf32>

func.func @quant_act_range(%arg0: !tfr.attr, %arg1: f32, %arg2: i64) -> !tfr.tensor {
  %0:2 = "tfr.quant_act_range"(%arg0, %arg1, %arg2) : (!tfr.attr,f32,i64) -> (!tfr.tensor,!tfr.tensor)
  func.return %0#0 : !tfr.tensor
}

// -----

// CHECK-LABEL: quant_rescale
func.func @quant_rescale(%arg0: !tfr.tensor, %arg1: !tfr.tensor, %arg2: i64) -> !tfr.tensor {
  %0 = "tfr.quant_rescale"(%arg0, %arg1, %arg2) : (!tfr.tensor, !tfr.tensor, i64) -> (!tfr.tensor)

  func.return
}

// CHECK-LABEL: func @replace_single_output
// CHECK-SAME:  ([[ARG0:%.+]]: tensor<i1>, [[ARG1:%.+]]: tensor<i1>, [[ARG2:%.+]]: tensor<i1>, [[ARG3:%.+]]: tensor<!tf_type.resource>)
func.func @replace_single_output(%arg0: tensor<i1>, %arg1: tensor<i1>, %arg2: tensor<i1>, %arg3: tensor<!tf_type.resource>) {
  // CHECK:      {{%.+}}:2 = "tf_device.parallel_execute"
  %0:3 = "tf_device.parallel_execute"() ({

func.func @replicated(%arg0: tensor<!tf_type.resource<tensor<10x3xf32>>>, %arg1: tensor<!tf_type.resource<tensor<10x3xf32>>>, %arg2: tensor<!tf_type.resource<tensor<10x3xf32>>>, %arg3: tensor<!tf_type.resource<tensor<10x3xf32>>>) -> tensor<!tf_type.resourc...

  %0:2 = "tf.WhileRegion"(%arg0, %arg1) ({
  ^bb0(%arg2: tensor<i32>, %arg3: tensor<i32>):
    %1 = "tf.OpA"() {is_stateless = true} : () -> tensor<i1>
    "tf.Yield"(%1) : (tensor<i1>) -> ()
  }, {
  ^bb0(%arg2: tensor<i32>, %arg3: tensor<i32>):
    %1 = "tf.GlobalIterId"() : () -> tensor<i64>
    "tf.Yield"(%arg2, %arg3) : (tensor<i32>, tensor<i32>) -> ()

  // CHECK: %[[ARG1_TPOSE:[0-9]*]] = "tf.Transpose"(%arg1, %[[ARG_PERM]])

  // CHECK: "tf.FusedBatchNormGradV3"
  // CHECK-SAME: (%[[ARG0_TPOSE]], %[[ARG1_TPOSE]], %arg2, %arg2, %arg2, %arg2)
  // CHECK-SAME: data_format = "NCHW"
  // CHECK-SAME: (tensor<1x64x28x28xf32>, tensor<1x64x28x28xf32>,
  // CHECK-SAME: -> (tensor<1x64x28x28xf32>,

  // CHECK: %[[RES_PERM:.*]] = "tf.Const"()

func.func @testConv2D(tensor<256x32x32x3xf32>, tensor<16x3x3x3xf32>, tensor<16xf32>) -> tensor<256x32x32x16xf32> {
^bb0(%arg0: tensor<256x32x32x3xf32>, %arg1: tensor<16x3x3x3xf32>, %arg2: tensor<16xf32>):
  // CHECK: _arithmetic_count = 230686720 : i64

    def create(self, arg0: Location, arg1: Value, arg2: Value) -> Operation: ...

class Tf_GreaterEqualOp:
    def __init__(self, *args, **kwargs) -> None: ...
    def create(self, arg0: Location, arg1: Value, arg2: Value) -> Operation: ...

class Tf_GreaterOp:
    def __init__(self, *args, **kwargs) -> None: ...
    def create(self, arg0: Location, arg1: Value, arg2: Value) -> Operation: ...

class Tf_LegacyCallOp: