%3 = "tfl.dequantize"(%2) : (tensor<1x3x3x!quant.uniform<i8:f32, 0.004:-128>>) -> tensor<1x3x3xf32>
  %4 = "tfl.div"(%arg0, %3) {fused_activation_function = "NONE"} : (tensor<1x3x3xf32>, tensor<1x3x3xf32>) -> tensor<1x3x3xf32>
  func.return %4 : tensor<1x3x3xf32>
//  CHECK: %[[logistic:.*]] = "tfl.logistic"
//  CHECK: %[[dq:.*]] = "tfl.dequantize"(%[[logistic]])

  // CHECK-NEXT: [[out_ch:%.*]] = tfrt.merge.chains [[ch]], [[ch1]] : !tfrt.chain, !tfrt.chain
  // CHECK-NEXT: tfrt.return [[out_ch]], [[r2]] : !tfrt.chain, !tfrt_fallback.tf_tensor
  func.return %6#0 : tensor<3x3xf32>
}

    %cst_1 = "tf.Const"() <{value = dense<[[-0.795477629, 0.581315517, 0.921566545], [0.138622552, 0.463866323, 0.95474267], [-0.143770888, -0.796835303, 0.899996876], [0.0989735424, -0.483384758, -7.277030e-01]]> : tensor<4x3xf32>}> {device = ""} : () -> tensor<4x3xf32>

  func.return %1 : tensor<3x3xf32>
}

// CHECK-LABEL: func @gpu_device
func.func @gpu_device(%arg0: tensor<3x1xf32>, %arg1: tensor<!tf_type.resource<tensor<1x3xf32>>>) -> tensor<3x3xf32> {
  // CHECK: {{%.*}} = corert.get_op_handler %arg0 "/device:GPU:0"

      -> (tensor<3x3xf32> {tf_saved_model.index_path = []}) {
  // CHECK: {{%.*}} = corert.get_op_handler %arg0 "/device:GPU:0"
  %2 = "tf.MatMul"(%arg0, %arg1) {T = f32, _output_shapes = ["tfshape$dim { size: 3 } dim { size: 3 }"], device = "/device:GPU:0", transpose_a = false, transpose_b = false} : (tensor<3x1xf32>, tensor<1x3xf32>) -> tensor<3x3xf32>
  func.return %2 : tensor<3x3xf32>

  // attribute `_quantization_method` set to `"no_quantization {}"`.
  constexpr absl::string_view kXlaCallModuleOpWithQuantizationMethodAttr =
      R"mlir(
    func.func @main(%arg0: tensor<1x1x3xf32>, %arg1: tensor<3x4xf32>) -> tensor<1x1x4xf32> {

    %0 = stablehlo.constant dense<1.000000e+00> : tensor<4x3xf32>
    %1 = "quantfork.qcast"(%0) {volatile} : (tensor<4x3xf32>) -> tensor<4x3x!quant.uniform<i8<-127:127>:f32:1, {5.000000e-03, 5.000000e-03, 5.000000e-03}>>
    %2 = "quantfork.dcast"(%1) : (tensor<4x3x!quant.uniform<i8<-127:127>:f32:1, {5.000000e-03, 5.000000e-03, 5.000000e-03}>>) -> tensor<4x3xf32>

  %0 = stablehlo.constant dense<1.0> : tensor<4x3xf32>
  %1 = stablehlo.dot_general %arg0, %0, contracting_dims = [1] x [0], precision = [DEFAULT, DEFAULT] : (tensor<1x4xf32>, tensor<4x3xf32>) -> tensor<1x3xf32>
  return %1 : tensor<1x3xf32>
}
// CHECK: @main(%[[ARG_0:.+]]: tensor<1x4xf32>) -> tensor<1x3xf32>
// CHECK-DAG: %[[CST:.+]] = stablehlo.constant dense<1.000000e+00> : tensor<4x3xf32>

// CHECK-NEXT: return %[[dq]] : tensor<2x2xf32>
}

// CHECK-LABEL: prepareStatistics
func.func @prepareStatistics(%arg0: tensor<8x4x3xf32>) -> tensor<8x4x3xf32> {
  %0 = "quantfork.stats"(%arg0) {
    layerStats = dense<[-1.0, 1.0]> : tensor<2xf32>
  } : (tensor<8x4x3xf32>) -> tensor<8x4x3xf32>
  %1 = "quantfork.stats"(%0) {
    layerStats = dense<[-1.0, 1.0]> : tensor<2xf32>,
    axisStats = dense<[
      [-1.0, 1.0],

  func.return %0: tensor<2x3x7xf32>
}

// CHECK-LABEL: testDynamicBatchMatMulToV2
func.func @testDynamicBatchMatMulToV2(%arg0: tensor<2x3x5xf32>, %arg1: tensor<?x5x7xf32>) -> tensor<2x3x7xf32> {