// expected-remark@+1 {{Cost: 10}}
    %44 = "tf.ConcatV2"(%36, %37, %38, %39, %40, %41, %42, %43, %10) {device = "/job:localhost/replica:0/task:0/device:CPU:0"} : (tensor<?x1xf32>, tensor<?x1xf32>, tensor<?x1xf32>, tensor<?x1xf32>, tensor<?x1xf32>, tensor<?x1xf32>, tensor<?x1xf32>, tensor<?x1xf32>, tensor<i32>) -> tensor<?x8xf32>

    %3 = "tf.XlaCallModule"(%2, %1, %0) <{Sout = [#tf_type.shape<?x2>], module = "", version = 9 : i64}> {_entry_function = @composite_dot_general_with_bias_and_relu6_dynamic_fn_1, _original_entry_function = "composite_dot_general_with_bias_and_relu6_dynamic_fn_1", _tfl_quant_trait = "fully_quantizable"} : (tensor<?x2xf32>, tensor<2x2xf32>, tensor<2xf32>) -> tensor<?x2xf32>

      %1 = "tf.Const"() {value = dense<1.0> : tensor<2xf32>} : () -> tensor<2xf32>
      %2 = "tf.Const"() {value = dense<2> : tensor<2xi32>} : () -> tensor<2xi32>
      %3 = "tf.Const"() {value = dense<1.0> : tensor<2xf32>} : () -> tensor<2xf32>
      %4 = "tf.Const"() {value = dense<0> : tensor<2xi32>} : () -> tensor<2xi32>

    module {
      func.func @test_fold_constant(%arg0: tensor<*xf32>) -> (tensor<?x?x?x3xf32>) {
        %cst = "tf.Const"() {value = dense<2.000000e+00> : tensor<2x3x3x1xf32>} : () -> tensor<2x3x3x1xf32>
        %cst_0 = "tf.Const"() {value = dense<0.400000e+00> : tensor<3xf32>} : () -> tensor<3xf32>
        %cst_1 = "tf.Const"() {value = dense<0.500000e+00> : tensor<3xf32>} : () -> tensor<3xf32>

func.func @func(%arg0: tensor<4xf32>) -> tensor<4xf32> {
  %cst = "tf.Const"() {value = dense<23.0> : tensor<4xf32>} : () -> tensor<4xf32>
  %0 = "tf.XlaSharding"(%arg0) { _XlaSharding = "\01\02\03"} : (tensor<4xf32>) -> tensor<4xf32>
  %1 = "tf.AddV2"(%0, %cst) : (tensor<4xf32>, tensor<4xf32>) -> (tensor<4xf32>)
  %2 = "tf.AddV2"(%cst, %1) : (tensor<4xf32>, tensor<4xf32>) -> (tensor<4xf32>)

  %biasadd = "tf.BiasAdd"(%conv, %dq_bias) {data_format = "NHWC", device = ""} : (tensor<*xf32>, tensor<2xf32>) -> tensor<*xf32>
  %res = "tf.Relu6"(%biasadd) : (tensor<*xf32>) -> tensor<*xf32>
  %q_res = "quantfork.qcast"(%res) : (tensor<*xf32>) -> tensor<*x!quant.uniform<i8:f32, 0.023529411764705882:-128>>

  func.return %2 : tensor<2xf32>
}
func.func private @callee(%arg0: tensor<!tf_type.resource<tensor<2xf32>>>) -> tensor<2xf32> {
  %0 = "tf.ReadVariableOp"(%arg0) : (tensor<!tf_type.resource<tensor<2xf32>>>) -> tensor<2xf32>
  func.return %0 : tensor<2xf32>
}


// -----

// Tests main function with multiple blocks.

}

TEST_F(IsOpFullyQuantizedTest, FalseIfOpNotQuantized) {
  constexpr absl::string_view kNotQuantizedAdd = R"mlir(
    func.func @not_quantized_add(%arg0: tensor<2xf32>) -> tensor<2xf32> {
      %0 = stablehlo.add %arg0, %arg0 : tensor<2xf32>
      return %0 : tensor<2xf32>
    }
  )mlir";

  OwningOpRef<ModuleOp> module_op = ParseModuleOpString(kNotQuantizedAdd);
  ASSERT_TRUE(module_op);

    quantization_max_val = 127 : i64
  } : (tensor<10xf32>, tensor<f32>, tensor<i32>) -> tensor<10x!tf_type.qint8>
  %1 = "tf.UniformDequantize"(%0, %scale, %zp) {
    quantization_axis = -1 : i64,
    quantization_min_val = -128 : i64,
    quantization_max_val = 127 : i64
  } : (tensor<10x!tf_type.qint8>, tensor<f32>, tensor<i32>) -> tensor<10xf32>
  return %1 : tensor<10xf32>
})mlir";

  %2 = func.call @fold_all_test(%arg0, %0, %1) : (tensor<256x32x32x3xf32>, tensor<16x3x3x3xf32>, tensor<16xf32>) -> tensor<256x30x30x16xf32>
  func.return %2 : tensor<256x30x30x16xf32>
}

// ALL-LABEL: @fold_all_test
func.func @fold_all_test(%arg0: tensor<256x32x32x3xf32>, %arg1: tensor<16x3x3x3xf32>, %arg2: tensor<16xf32>) -> tensor<256x30x30x16xf32> {