func.func private @conv(%input: tensor<1x3x4x3xf32> {tf._user_specified_name = "input_tensor"}) -> tensor<*xf32> attributes {tf._construction_context = "kEagerRuntime", tf._input_shapes = [#tf_type.shape<1x3x4x3>]} {
  %weight = arith.constant dense_resource<__elided__> : tensor<2x3x3x2xf32>
  %bias = arith.constant dense<[7.11401462, 7.05456924]> : tensor<2xf32>

  %9 = "tfl.quantize"(%8) {qtype = tensor<1x3x1x1x!quant.uniform<i8:f32, 1.0:2>>, volatile} : (tensor<1x3x1x1xf32>) -> tensor<1x3x1x1x!quant.uniform<i8:f32, 1.0:2>>
  %10 = "tfl.dequantize"(%9) : (tensor<1x3x1x1x!quant.uniform<i8:f32, 1.0:2>>) -> tensor<1x3x1x1xf32>

// CHECK: %0 = "tfl.transpose"(%arg0, %cst_0) : (tensor<2x3x4x1xf32>, tensor<4xi32>) -> tensor<1x2x3x4xf32>
// CHECK: %1 = tfl.add(%0, %cst) <{fused_activation_function = "NONE"}> : (tensor<1x2x3x4xf32>, tensor<5x2x3x4xf32>) -> tensor<5x2x3x4xf32>

// -----

// CHECK-LABEL: doubleTposeOneBroadcastInput
func.func @doubleTposeOneBroadcastInput(%arg0: tensor<2x3x4x1xf32>, %arg1: tensor<2x3x4x5xf32>) -> tensor<5x2x3x4xf32> {

func.func @depth_to_space(%arg0: tensor<1x1x1x4xf32>) -> tensor<1x2x2x1xf32> {
  %0 = "tf.DepthToSpace"(%arg0) {block_size = 2: i64,  data_format = "NHWC"}: (tensor<1x1x1x4xf32>) -> tensor<1x2x2x1xf32>
  func.return %0 : tensor<1x2x2x1xf32>
// CHECK: %[[CUSTOM_0:.*]] = "tfl.custom"(%arg0) <{custom_code = "FlexDepthToSpace", custom_option = #tfl<const_bytes : "{{.*}}">}> : (tensor<1x1x1x4xf32>) -> tensor<1x2x2x1xf32>

  func.return %0, %1, %2 : tensor<1x4x4xf32>, tensor<1x4x4xf32>, tensor<14x4x4xf32>
}

// -----

func.func @testSplitVOpWithBadSizeSplitsConstantSum(%arg0: tensor<16x4x4xf32>) -> (tensor<0x4x4xf32>, tensor<16x4x4xf32>) {

  %2 = "tfl.select"(%cst_false, %arg0, %arg1) : (tensor<1x2x3x4xi1>, tensor<1x2x3x4xf32>, tensor<1x2x3x4xf32>) -> tensor<1x2x3x4xf32>
  %3 = "tfl.select_v2"(%cst_false, %arg0, %arg1) : (tensor<1x2x3x4xi1>, tensor<1x2x3x4xf32>, tensor<1x2x3x4xf32>) -> tensor<1x2x3x4xf32>
  func.return %0, %1, %2, %3 : tensor<1x2x3x4xf32>, tensor<1x2x3x4xf32>, tensor<1x2x3x4xf32>, tensor<1x2x3x4xf32>

// CHECK:  return
}

func.func @reverse_v2(%arg0: tensor<1x2x3x4xf32>, %arg1: tensor<1xi32>) -> tensor<1x2x3x4xf32> {
  %0 = "tf.ReverseV2"(%arg0, %arg1) : (tensor<1x2x3x4xf32>, tensor<1xi32>) -> tensor<1x2x3x4xf32>
  func.return %0 : tensor<1x2x3x4xf32>

// CHECK-LABEL:reverse_v2
// CHECK:  "tfl.reverse_v2"(%arg0, %arg1) : (tensor<1x2x3x4xf32>, tensor<1xi32>) -> tensor<1x2x3x4xf32>
// CHECK:  return
}

  %15 = "tf.Identity"(%14) {device = ""} : (tensor<1x2x4x1xf32>) -> tensor<1x2x4x1xf32>
  func.return %15 : tensor<1x2x4x1xf32>
}

// CHECK-LABEL: func @max_unpooling_2d(
// CHECK-SAME:                         %[[VAL_0:.*]]: tensor<1x1x2x1xf32>,
// CHECK-SAME:                         %[[VAL_1:.*]]: tensor<1x1x2x1xi32>) -> tensor<1x2x4x1xf32> attributes {tf._implements = "MaxUnpooling2D"} {

// CHECK:           %[[VAL_8:.*]] = "tf.Select"(%[[VAL_2]], %[[VAL_3]], %[[VAL_7]]) : (tensor<1x2x3x4xi1>, tensor<1x2x3x4xf32>, tensor<1x2x3x4xf32>) -> tensor<1x2x3x4xf32>
// CHECK:           return %[[VAL_8]] : tensor<1x2x3x4xf32>
// CHECK:         }
func.func @sign(%arg0: tensor<1x2x3x4xf32>, %arg1: tensor<1x2x3x4xf32>) -> tensor<1x2x3x4xf32> {

    // CHECK-NEXT: %[[CMP:.*]] = mhlo.compare GT, %[[INP]], %[[ZERO]], NOTYPE : (tensor<1x4x4x3xf32>, tensor<1x4x4x3xf32>) -> tensor<1x4x4x3xi1>
    // CHECK-NEXT: %[[RES:.*]] = mhlo.select %[[CMP]], %[[INP]], %[[LEAKY]] : tensor<1x4x4x3xi1>, tensor<1x4x4x3xf32>
    // CHECK-NEXT: return %[[RES]] : tensor<1x4x4x3xf32>
    %0 = "tf.LeakyRelu"(%arg0) {alpha = 2.000000e-01 : f32, device = ""} : (tensor<1x4x4x3xf32>) -> tensor<1x4x4x3xf32>