%cst = "tf.Const"() {value = dense<0.000000e+00> : tensor<2xf32>} : () -> tensor<2xf32>
  %0 = "tf.Conv2D"(%arg0, %arg1) {data_format = "NHWC", device = "", dilations = [1, 1, 2, 1], explicit_paddings = [], padding = "SAME", strides = [1, 1, 2, 1], use_cudnn_on_gpu = true} : (tensor<1x3x4x3xf32>, tensor<2x3x3x2xf32>) -> tensor<*xf32>

      } : (tensor<1x3x4x3xf32>, tensor<2x3x3x2xf32>) -> tensor<1x3x4x2xf32>
    return %0 : tensor<1x3x4x2xf32>
  }

  // CHECK: func.func private @qdq_for_conv_weight_per_channel(%[[ARG0:.+]]: tensor<1x3x4x3xf32>)
  // CHECK: %[[CST:.+]] = "tf.Const"() <{value = dense<3.000000e-01> : tensor<2x3x3x2xf32>}> : () -> tensor<2x3x3x2xf32>

  %cst_1 = "tf.Const"() {value = dense<[0.1, 0.2]> : tensor<2xf32>} : () -> tensor<2xf32>
  %cst_2 = "tf.Const"() {value = dense<[0.3, 0.4]> : tensor<2xf32>} : () -> tensor<2xf32>

    return %1 : tensor<1x3x4x2xf32>
  }

  func.func private @composite_conv_fn(%arg0: tensor<1x3x4x3xf32>, %arg1: tensor<2x3x3x2xf32>) -> tensor<1x3x4x2xf32> attributes {_from_xla_call_module} {

    %cst = stablehlo.constant dense<3.000000e-01> : tensor<2x3x3x2xf32>
    %0 = "quantfork.qcast"(%cst) : (tensor<2x3x3x2xf32>) -> tensor<2x3x3x2x!quant.uniform<i8:f32, 6.000000e-03:-128>>
    %1 = "quantfork.dcast"(%0) : (tensor<2x3x3x2x!quant.uniform<i8:f32, 6.000000e-03:-128>>) -> tensor<2x3x3x2xf32>

  %q_weight = "quantfork.qcast"(%weight) : (tensor<2x3x3x2xf32>) -> tensor<2x3x3x2x!quant.uniform<i8:f32, 0.074855112561992565:-1>>
  %dq_weight = "quantfork.dcast"(%q_weight) : (tensor<2x3x3x2x!quant.uniform<i8:f32, 0.074855112561992565:-1>>) -> tensor<2x3x3x2xf32>
  %q_bias = "quantfork.qcast"(%bias) : (tensor<2xf32>) -> tensor<2x!quant.uniform<i32:f32, 0.044022349891595126>>

  %q_weight = "quantfork.qcast"(%weight) : (tensor<2x3x3x2xf32>) -> tensor<2x3x3x2x!quant.uniform<i8:f32, 0.074855112561992565:-1>>
  %dq_weight = "quantfork.dcast"(%q_weight) : (tensor<2x3x3x2x!quant.uniform<i8:f32, 0.074855112561992565:-1>>) -> tensor<2x3x3x2xf32>
  %q_bias = "quantfork.qcast"(%bias) : (tensor<2xf32>) -> tensor<2x!quant.uniform<i32:f32, 0.044022349891595126>>

  func.func private @quantize_conv_fn(%arg0: tensor<1x3x4x3xf32>) -> tensor<1x3x4x2xf32> attributes {tf._original_func_name = "main_0"} {
    %cst = "tf.Const"() {value = dense<3.00000000e-1> : tensor<2x3x3x2xf32>} : () -> tensor<2x3x3x2xf32>
    %0 = "quantfork.stats"(%arg0) {layerStats = dense<[6.00000000e-6, 9.00000000e-1]> : tensor<2xf32>} : (tensor<1x3x4x3xf32>) -> tensor<1x3x4x3xf32>
    %1 = "tf.XlaCallModule"(%0, %cst) {

  %output, %min, %max, %histogram = "tf.CustomAggregator"(%arg0) <{calibration_method = 5 : i32, id = "skipping_id", num_bins = 32 : i32, max_percentile = 0.000000e+00 : f32, min_percentile = 0.000000e+00 : f32}> : (tensor<1x3x4x3xf32>)...

// -----

// CHECK-LABEL: doubleTposeInputPermNotEqualNoChange
func.func @doubleTposeInputPermNotEqualNoChange(%arg0: tensor<2x4x3x5xf32>, %arg1: tensor<2x3x4x5xf32>) -> tensor<5x2x3x4xf32> {
  %perm = arith.constant dense<[3, 0, 2, 1]> : tensor<4xi32>
  %0 = "tfl.transpose"(%arg0, %perm) : (tensor<2x4x3x5xf32>, tensor<4xi32>) -> tensor<5x2x3x4xf32>
  %perm1 = arith.constant dense<[3, 0, 1, 2]> : tensor<4xi32>