func p256Selectznz(out1 *[4]uint64, arg1 p256Uint1, arg2 *[4]uint64, arg3 *[4]uint64) {
	var x1 uint64
	p256CmovznzU64(&x1, arg1, arg2[0], arg3[0])
	var x2 uint64
	p256CmovznzU64(&x2, arg1, arg2[1], arg3[1])
	var x3 uint64
	p256CmovznzU64(&x3, arg1, arg2[2], arg3[2])
	var x4 uint64
	p256CmovznzU64(&x4, arg1, arg2[3], arg3[3])
	out1[0] = x1
	out1[1] = x2
	out1[2] = x3

  func.return %arg0: tensor<1x!tf_type.qint32>
}

// -----

// CHECK-LABEL: func @id_quint8(%arg0: tensor<1xui8>) -> tensor<1xui8> {
func.func @id_quint8(%arg0: tensor<1x!tf_type.quint8>) -> tensor<1x!tf_type.quint8> {
  // CHECK-NEXT: return %arg0 : tensor<1xui8>
  func.return %arg0: tensor<1x!tf_type.quint8>
}

// -----

// CHECK-LABEL: func @id_quint16(%arg0: tensor<1xui16>) -> tensor<1xui16> {

                                                            arg0="${f:h(q)}" arg1="${f:h(allRecordCount)}"
                                                            arg2="${f:h(currentStartRecordNumber)}"
                                                            arg3="${f:h(currentEndRecordNumber)}"/>

}

func.func @branch0(%arg0: tensor<f32>, %arg1: tensor<f32>) -> tensor<f32> {
  %0 = "tf.Add" (%arg0, %arg1) {device = "/device:CPU:0"}  : (tensor<f32>, tensor<f32>) -> tensor<f32>
  func.return %0 : tensor<f32>
}

func.func @branch1(%arg0: tensor<f32>, %arg1: tensor<f32>) -> tensor<f32> {
  %0 = "tf.Add" (%arg0, %arg1) {device = "/device:CPU:0"}  : (tensor<f32>, tensor<f32>) -> tensor<f32>

    func.return %0, %1 : tensor<*xf32>, tensor<*xf32>
  }
  func.func private @composite_conv2d_with_bias_and_relu6_fn_2(%arg0: tensor<1x2x2x3xf32>, %arg1: tensor<2x2x3x2xf32>, %arg2: tensor<2xf32>) -> tensor<*xf32> attributes {tf_quant.composite_function} {

func p224Selectznz(out1 *[4]uint64, arg1 p224Uint1, arg2 *[4]uint64, arg3 *[4]uint64) {
	var x1 uint64
	p224CmovznzU64(&x1, arg1, arg2[0], arg3[0])
	var x2 uint64
	p224CmovznzU64(&x2, arg1, arg2[1], arg3[1])
	var x3 uint64
	p224CmovznzU64(&x3, arg1, arg2[2], arg3[2])
	var x4 uint64
	p224CmovznzU64(&x4, arg1, arg2[3], arg3[3])
	out1[0] = x1
	out1[1] = x2
	out1[2] = x3

  func.func private @composite_dot_general_fn_1(%arg0: tensor<1x1024xf32>, %arg1: tensor<1024x3xf32>, %arg2: tensor<1x3xf32>) -> tensor<1x3xf32> attributes {_from_xla_call_module, tf_quant.composite_function} {
    %0 = stablehlo.dot_general %arg0, %arg1, contracting_dims = [1] x [0], precision = [DEFAULT, DEFAULT] : (tensor<1x1024xf32>, tensor<1024x3xf32>) -> tensor<1x3xf32>
    %1 = stablehlo.add %0, %arg2 : tensor<1x3xf32>
    return %0 : tensor<1x3xf32>
  }

// CHECK: func.func private @serving_default
// CHECK-SAME: (%arg0: tensor<?x!tf_type.string>, %arg1: tensor<!tf_type.resource>) -> tensor<*xi64>
// CHECK-SAME: tf.entry_function = {control_outputs = "", inputs = "input_vocabs:0,hash_table_1:0", outputs = "FakeQuantWithMinMaxArgs_2:0"}
// CHECK: "tf.LookupTableSizeV2"(%arg1)
// CHECK: "tf.LookupTableFindV2"(%arg1

        tf_device.return %1 : tensor<i8>
      }
      return
    }
    ```

    will be rewritten as:

    ```mlir
    func @tf_tpu_rewrite(%arg0: tensor<i8>, %arg1: tensor<i8>) {

// CHECK-LABEL: func @main
// CHECK-SAME: (%[[ARG0:.*]]: tensor<10x8x9xf32>, %[[ARG1:.*]]: tensor<5xi32>, %[[ARG2:.*]]: tensor<5x8x9xf32>) -> tensor<10x8x9xf32>
func.func @main(%arg0: tensor<10x8x9xf32>, %arg1: tensor<5xi32>, %arg2: tensor<5x8x9xf32>) -> tensor<10x8x9xf32> {
  %elem_shape = "tf.Const"() {value = dense<[8, 9]> : tensor<2xi32>} : () -> tensor<2xi32>
  // CHECK: %[[BUFFER:.*]] = "tf.Identity"(%[[ARG0]]) : (tensor<10x8x9xf32>) -> tensor<10x8x9xf32>