}
   func @_func(%input: tensor<2x112x112x12xf32>,
              %filter: tensor<7x7x3x64xf32>) {
     %filter_transform = "tf.Pad/tf.Transpose/tf.Reshape"(%filter):
       tensor<7x7x3x64xf32>) -> tensor<4x4x12x64xf32>
     %conv = "tf.Conv2D"(%input, %filter_transform) {strides = [1, 1, 1, 1]}:
       (tensor<2x112x112x12xf32>, tensor<4x4x12x64xf32>) ->
       tensor<2x112x112x64xf32>
   }
}
```

    // CHECK-SAME: strides = [1, 1, 1, 1]
    // CHECK-SAME: (tensor<2x115x115x12xf32>, tensor<4x4x12x64xf32>) -> tensor<2x112x112x64xf32>
    %6 = "tf.Conv2D"(%5, %arg1) {data_format = "NHWC", dilations = [1, 1, 1, 1], explicit_paddings = [], padding = "VALID", strides = [1, 2, 2, 1], use_cudnn_on_gpu = true} : (tensor<2x230x230x3xf32>, tensor<7x7x3x64xf32>) -> tensor<2x112x112x64xf32>
    // CHECK: %[[BACKPROP:.*]] = "tf.Conv2DBackpropFilter"

        %space_to_depth = "tf.SpaceToDepth"(%input) {block_size = 2, ...}: (tensor<2x224x224x3xf32>) -> tensor<2x112x112x12xf32>
        %device_launch = "tf_device.cluster_func"(%space_to_depth,...) {func = @_func,...)
        return ...
      }
      func @_func(%input: tensor<2x112x112x12xf32>, %filter: tensor<7x7x3x64xf32>) {

    return ...
  }
  func @_func(%input: tensor<2x112x112x12xf32>, %filter: tensor<7x7x3x64xf32>) {
    %filter_transform = "tf.Pad/tf.Transpose/tf.Reshape"(%filter): tensor<7x7x3x64xf32>) -> tensor<4x4x12x64xf32>
    %conv = "tf.Conv2D"(%input, %filter_transfrom) {strides = [1, 1, 1, 1]}: (tensor<2x112x112x12xf32>, tensor<4x4x12x64xf32>) -> tensor<2x112x112x64xf32>
  }
}
```

func.func @QuantizeFullyConnected(%arg0: tensor<1x224x224x3xf32>) -> tensor<1x112x112x512xf32> {
  %w = arith.constant dense<127.0> : tensor<512x12xf32>
  %b = arith.constant dense<0.0> : tensor<512xf32>
  %fc = "tfl.fully_connected"(%arg0, %w, %b) {fused_activation_function = "NONE", keep_num_dims = false, weights_format = "DEFAULT"} : (tensor<1x224x224x3xf32>, tensor<512x12xf32>, tensor<512xf32>) -> tensor<1x112x112x512xf32>
  func.return %fc : tensor<1x112x112x512xf32>

  %fc_s = "quantfork.stats"(%fc) {layerStats = dense<[0.000000e+00, 1.000000e+01]> : tensor<2xf32>} : (tensor<1x112x112x512xf32>) -> tensor<1x112x112x512xf32>
  func.return %fc : tensor<1x112x112x512xf32>

// CHECK-DAG: %[[w:.*]] = arith.constant dense<1.270000e+02> : tensor<512x12xf32>

    %7 = "tfl.quantize"(%6) {qtype = tensor<1x112x112x32x!quant.uniform<u8:f32, 1.0>>} : (tensor<1x112x112x32xf32>) -> tensor<1x112x112x32x!quant.uniform<u8:f32, 1.0>>
    func.return %7 : tensor<1x112x112x32x!quant.uniform<u8:f32, 1.0>>

// CHECK: %[[conv:.*]] = "tfl.conv_2d"(%arg0, %arg1, %arg2)
// CHECK-SAME: -> tensor<1x112x112x32x!quant.uniform<u8:f32, 0.0078431372549019607:128>>

  %3 = "tfl.quantize"(%conv1) {qtype = tensor<1x112x112x32xf32>} : (tensor<1x112x112x32xf32>) -> tensor<1x112x112x32x!quant.uniform<u8:f32, 1.0>>

  %4 = "tfl.dequantize"(%3) : (tensor<1x112x112x32x!quant.uniform<u8:f32, 1.0>>) -> tensor<1x112x112x32xf32>

  func.return %0 : tensor<1x112x112x32xf32>
}

// -----

func.func @testAvgPool(%arg0: tensor<256x32x32x3xf32>) -> tensor<256x30x30x16xf32> {
  // CHECK: tac.device = "GPU", tac.inference_type = "FLOAT"

  %6 = "tfl.quantize"(%5) {qtype = tensor<1x112x112x32x!quant.uniform<u8:f32, 0.023528476789885875>>} : (tensor<1x112x112x32xf32>) -> tensor<1x112x112x32x!quant.uniform<u8:f32, 0.023528476789885875>>
  func.return %6 : tensor<1x112x112x32x!quant.uniform<u8:f32, 0.023528476789885875>>