%fq = "tf.FakeQuantWithMinMaxVars"(%in, %mini, %maxi) {num_bits = 3, narrow_range = false} : (tensor<3x3x3x16xf32>, tensor<f32>, tensor<f32>) -> tensor<3x3x3x16xf32>
  %rst = "tf.Conv2D"(%arg, %fq) {T = "tfdtype$DT_FLOAT", data_format = "NHWC", dilations = [1, 2, 3, 1], padding = "SAME", strides = [1, 4, 5, 1]} : (tensor<256x32x32x3xf32>, tensor<3x3x3x16xf32>) -> tensor<256x8x7x16xf32>
  func.return %rst : tensor<256x8x7x16xf32>

  %fq = "tf.FakeQuantWithMinMaxVars"(%in, %mini, %maxi) {num_bits = 5, narrow_range = false} : (tensor<3x3x3x16xf32>, tensor<f32>, tensor<f32>) -> tensor<3x3x3x16xf32>
  %rst = "tf.Conv2D"(%arg, %fq) {T = "tfdtype$DT_FLOAT", data_format = "NHWC", dilations = [1, 2, 3, 1], padding = "SAME", strides = [1, 4, 5, 1]} : (tensor<256x32x32x3xf32>, tensor<3x3x3x16xf32>) -> tensor<256x8x7x16xf32>
  func.return %rst : tensor<256x8x7x16xf32>

  %0 = "tf.BiasAdd"(%arg0, %arg1) {T = "tfdtype$DT_FLOAT", data_format = "NHWC"} : (tensor<1x10x10x32xf32>, tensor<32xf32>) -> tensor<1x10x10x32xf32>
  func.return %0 : tensor<1x10x10x32xf32>
// CHECK: %[[BIASADD_0:.*]] = "tf.BiasAdd"(%arg0, %arg1) <{data_format = "NHWC"}> {T = "tfdtype$DT_FLOAT"} : (tensor<1x10x10x32xf32>, tensor<32xf32>) -> tensor<1x10x10x32xf32>

  } : (tensor<3x2x!tf_type.qint32>, tensor<3x2x!tf_type.qint32>, tensor<3x2x!tf_type.qint32>, tensor<2xf32>, tensor<2xi32>) -> tensor<3x2x!tf_type.qint32>
  func.return %1 : tensor<3x2x!tf_type.qint32>
}

// -----

// CHECK-LABEL: func @uniform_quantized_clip_by_value_min_not_const

  %output_0 = "tf.TensorArrayGatherV3"(%handle_1, %indices, %res_0#3) {device = "/job:localhost/replica:0/task:0/device:CPU:0", element_shape = #tf_type.shape<?x512>} : (tensor<2x!tf_type.resource<tensor<?x512xf32>>>,...

    %1 = "tf.EmptyTensorList"(%elem_shape, %max_size) { _xla_outside_compilation="0" } : (tensor<0xi32>, tensor<i32>) -> tensor<!tf_type.variant<tensor<f32>>>
    %2 = "tf.Identity"(%1) : (tensor<!tf_type.variant<tensor<f32>>>) -> (tensor<!tf_type.variant<tensor<f32>>>)
    "tf.Identity"(%2) : (tensor<!tf_type.variant<tensor<f32>>>) -> (tensor<!tf_type.variant<tensor<f32>>>)
    tf_device.return
  }) {cluster_attr = "cluster_attr"} : () -> ()
  func.return
}

  %5 = "tf.Identity"(%4) {__op_key = 10: i32, device = "/device:CPU:0"} : (tensor<!tf_type.variant>) -> tensor<!tf_type.variant>
  func.return %5 : tensor<!tf_type.variant>
}
// CHECK-LABEL: __inference_Dataset_map_lambda_160

  "tfl.assign_variable"(%1, %5) : (tensor<!tf_type.resource>, tensor<1x2x1x3xf32>) -> ()
  func.return %arg0 : tensor<1x2x1x3x!quant.uniform<i8:f64, 1.0>>

// CHECK-NEXT:  %[[vh:.*]] = "tfl.var_handle"() <{container = "", shared_name = ""}> : () -> tensor<*x!tf_type.resource<tensor<1x2x1x3x!quant.uniform<i8:f32, 1.000000e+00>>>>

    %5 = "tf.ReadVariableOp"(%arg5) : (tensor<*x!tf_type.resource<tensor<1024x1xf32>>>) -> tensor<1024x1xf32>
    %6 = "tf.ReadVariableOp"(%arg20) : (tensor<*x!tf_type.resource<tensor<1024xf32>>>) -> tensor<1024xf32>
    %7 = "tf.ReadVariableOp"(%arg21) : (tensor<*x!tf_type.resource<tensor<1024x1xf32>>>) -> tensor<1024x1xf32>

}

func.func @string() -> tensor<2x2x!tf_type.string> {
  // CHECK-LABEL: @string
  // CHECK: value = dense<{{\[\["1", "12"\], \["123", "1234"\]\]}}> : tensor<2x2x!tf_type.string>
  %0 = "tfl.pseudo_const"() { value = dense<[["1", "12"], ["123", "1234"]]> : tensor<2x2x!tf_type.string> } : () -> tensor<2x2x!tf_type.string>
  func.return %0 : tensor<2x2x!tf_type.string>
}

func.func @string_norank() -> tensor<!tf_type.string> {