%0 = "quantfork.stats"(%arg0) {layerStats = dense<[-1.0, 1.0]> : tensor<2xf32>} : (tensor<1x1xf32>) -> tensor<1x1xf32>
  %1 = "tfl.tanh"(%0) : (tensor<1x1xf32>) -> tensor<1x1xf32>
  %2 = "quantfork.stats"(%1) {layerStats = dense<[-1.0, 1.0]> : tensor<2xf32>} : (tensor<1x1xf32>) -> tensor<1x1xf32>
  func.return %2 : tensor<1x1xf32>

  func.return %0 : tensor<? x ? x ? x f32>
}

// CHECK-LABEL: func @init
func.func @init(%arg0: tensor<4xf32>) -> tensor<7xf32> {
    %0 = builtin.unrealized_conversion_cast to tensor<7xf32>
    return %0 : tensor<7xf32>
}

// CHECK-LABEL: func @next
func.func @next(%arg0: tensor<7xf32>, %arg1: tensor<3xf32>) -> tensor<6xf32> {
    %0 = builtin.unrealized_conversion_cast to tensor<6xf32>
    return %0 : tensor<6xf32>
}

// CHECK-LABEL: func @finalize

  %1 = "tf.Mul"(%arg1, %arg1) : (tensor<2xf32>, tensor<2xf32>) -> tensor<2xf32>
  %2 = "tf.Div"(%arg1, %arg1) : (tensor<2xf32>, tensor<2xf32>) -> tensor<2xf32>

  // CHECK: "tf.If"{{.+}}else_branch = @tf.IfRegion_else{{.+}}then_branch = @tf.IfRegion_then
  %3 = "tf.IfRegion"(%arg0) ({
     %4 = "tf.Select"(%arg0, %0, %1) : (tensor<i1>, tensor<2xf32>, tensor<2xf32>) -> tensor<2xf32>
     "tf.Yield"(%4) : (tensor<2xf32>) -> ()

// PerChannel-DAG: %[[scale2:.*]] = "tf.Const"() <{value = dense<{{[0-9\.Ee\+\-]+}}> : tensor<6xf32>}> : () -> tensor<6xf32>
// PerChannel-DAG: %[[zp1:.*]] = "tf.Const"() <{value = dense<{{[0-9]+}}> : tensor<3xi32>}> : () -> tensor<3xi32>
// PerChannel-DAG: %[[zp2:.*]] = "tf.Const"() <{value = dense<{{[0-9]+}}> : tensor<6xi32>}> : () -> tensor<6xi32>

  %2:2 = "tf.IteratorGetNext"(%arg0) {device = "/device:CPU:0"} : (tensor<*x!tf_type.resource>) -> (tensor<128xf32>, tensor<128xf32>)
  %3:2 = "tf.IteratorGetNext"(%arg1) {device = "/device:CPU:0"} : (tensor<*x!tf_type.resource>) -> (tensor<128xf32>, tensor<128xf32>)
  // CHECK: "tf.TPUCompileSucceededAssert"(%[[COMPILE]]#0)
  "tf_device.launch"() ({

    %0 = "tf.Sub"(%cst_0, %cst_1) : (tensor<8xf32>, tensor<8xf32>) -> tensor<8xf32>
    return %0 : tensor<8xf32>
  }
// CHECK: @serving_default
// CHECK-DAG: %[[VAR_HANDLE_2:.*]] = "tf.VarHandleOp"() {{.*shared_name = "const_0".*}} : () -> tensor<!tf_type.resource<tensor<8xf32>>>
// CHECK-DAG: %[[READ_VAR_0:.*]] = "tf.ReadVariableOp"(%[[VAR_HANDLE_2]]) : (tensor<!tf_type.resource<tensor<8xf32>>>) -> tensor<8xf32>

  %0 = "tfr.cast"(%arg0) : (tensor<10x!quant.uniform<i8:f32, 0.0039133410900831223:-128>>) -> !tfr.tensor
  %1 = tfr.quant_raw_data(%0) : (!tfr.tensor) -> !tfr.tensor
  %2 = "tfr.cast"(%1) : (!tfr.tensor) -> tensor<10xi8>
  %3 = "tf.Cast"(%2) {Truncate = false} : (tensor<10xi8>) -> tensor<10xi32>
  func.return %3 : tensor<10xi32>

  %0 = "tfr.constant_tensor"(%arg0) : (vector<1xi32>) -> tensor<1xf32>
  func.return %0 : tensor<1xf32>
}

// -----

func.func @constant_tensor_invalid_3(%arg0: vector<1xi32>) -> tensor<1x1xi32> {
    // expected-error@+1 {{input and output should have same shape and element type}}
  %0 = "tfr.constant_tensor"(%arg0) : (vector<1xi32>) -> tensor<1x1xi32>
  func.return %0 : tensor<1x1xi32>
}

// -----

  %cst_1 = arith.constant dense<0> : tensor<4xi32>
  %cst_2 = arith.constant dense<0.000000e+00> : tensor<1xf32>
  %0 = "tfl.quantize"(%cst_2) {qtype = tensor<1x!quant.uniform<i32:f32, 1.0>>, volatile} : (tensor<1xf32>) -> tensor<1x!quant.uniform<i32:f32, 1.0>>
  %1 = "tfl.dequantize"(%0) : (tensor<1x!quant.uniform<i32:f32, 1.0>>) -> tensor<1xf32>
  %cst_3 = arith.constant dense<[[[[1.0]], [[1.0]], [[1.0]]]]> : tensor<1x3x1x1xf32>