%cst = "tf.Const"() <{value = dense<[0, 1]> : tensor<2xi32>}> {device = ""} : () -> tensor<2xi32>
    %cst_0 = "tf.Const"() <{value = dense<1.000000e+01> : tensor<f32>}> {device = ""} : () -> tensor<f32>
    %0 = "tf.Sum"(%arg0, %cst) <{keep_dims = false}> {device = ""} : (tensor<1x4xf32>, tensor<2xi32>) -> tensor<f32>
    %1 = "tf.Greater"(%0, %cst_0) {device = ""} : (tensor<f32>, tensor<f32>) -> tensor<i1>

// CustomOpNotWeightOnly-LABEL: QuantizeCustomOp
func.func @QuantizeCustomOp(%arg0: tensor<1x1x1x1xf32>) -> tensor<*xf32> attributes {tf.entry_function = {inputs = "input", outputs = "custom_op"}} {
  %0 = "quantfork.stats"(%arg0) {layerStats = dense<[0.000000e+00, 2.550000e+02]> : tensor<2xf32>} : (tensor<1x1x1x1xf32>) -> tensor<1x1x1x1xf32>
  %w = arith.constant dense<127.0> : tensor<1024x1x1x1xf32>

      while (auto cast_op =
                 llvm::dyn_cast_or_null<CastOp>(value.getDefiningOp())) {
        // Consider cast compatibility in case
        //    %cast = "tf.Cast"(%0) : (tensor<2xi64>) -> tensor<2xf32>
        // is skipped.
        if (cast_op.getSrcT() != cast_op.getDstT()) {
          break;
        }
        value = cast_op.getOperand();
      }
      return value;
    };

"TANH", input_to_cell_intermediate = tensor<0xf32>, input_to_forget_intermediate = tensor<0xf32>, input_to_input_intermediate = tensor<0xf32>, input_to_output_intermediate = tensor<0xf32>, proj_clip = 0.000000e+00 : f32, time_major = false} : (tensor<?x?x?xf32>, tensor<?x?xf32>, tensor<?x?xf32>, tensor<?x?xf32>, tensor<?x?xf32>, tensor<?x?xf32>, tensor<?x?xf32>, tensor<?x?xf32>, tensor<?x?xf32>, tensor<?xf32>, tensor<?xf32>, tensor<?xf32>, tensor<?xf32>, tensor<?xf32>, tensor<?xf32>, tensor<?xf32>,...

func.func @squeeze_folder(%arg0 : tensor<?x?xf32>) -> tensor<?x?xf32> {
  %0 = "tfl.squeeze"(%arg0) : (tensor<?x?xf32>) -> tensor<?x?xf32>
  // CHECK: return %arg0
  func.return %0 : tensor<?x?xf32>
}

// -----

// CHECK-LABEL: @squeeze_folder
func.func @squeeze_folder(%arg0 : tensor<?x?xf32>) -> tensor<*xf32> {
  %0 = "tfl.squeeze"(%arg0) : (tensor<?x?xf32>) -> tensor<*xf32>
  // CHECK: "tfl.squeeze"

module {
  func.func @wrap_composite_func(%arg0: tensor<*xf32>, %arg1: tensor<*xf32>) -> tensor<*xf32> {
    %0 = "tf.PartitionedCall"(%arg0, %arg1) <{f = @composite_conv2d_with_relu6_fn}> {_tfl_quant_trait = "fully_quantizable"}
          : (tensor<*xf32>, tensor<*xf32>) -> tensor<*xf32>
    func.return %0 : tensor<*xf32>
  }

  func.func @no_composite_func(%arg0: tensor<*xf32>, %arg1: tensor<*xf32>) -> tensor<*xf32> {

  %max = arith.constant dense<[14.0, 13.0, 8.0, 0.0]> : tensor<4xf32>
  %mini = "tf.Identity"(%min) : (tensor<4xf32>) -> tensor<4xf32>
  %maxi = "tf.Identity"(%max) : (tensor<4xf32>) -> tensor<4xf32>
  %fq = "tf.FakeQuantWithMinMaxVarsPerChannel"(%in, %mini, %maxi) {num_bits = 4, narrow_range = true} : (tensor<3x3x3x4xf32>, tensor<4xf32>, tensor<4xf32>) -> tensor<3x3x3x4xf32>

    "tfl.yield"(%2) : (tensor<8xf32>) -> ()
  }) {num_bits = 5, narrow_range = false} :  (tensor<8xf32>, tensor<f32>, tensor<f32>) -> tensor<8xf32>
  func.return %rst : tensor<8xf32>

// CHECK: %[[CONSTANT:.*]] = arith.constant dense<0.000000e+00> : tensor<8xf32>
// CHECK: %[[QUANTIZE:.*]] = "tfl.quantize"(%[[CONSTANT]]) <{qtype = tensor<8x!quant.uniform<u8:f32, 1.000000e+00>>}>

// CHECK-NEXT:      %28 = "tfl.concatenation"(%25, %27) <{axis = 0 : i32, fused_activation_function = "NONE"}> : (tensor<3xi32>, tensor<2xi32>) -> tensor<5xi32>
// CHECK-NEXT:      %29 = mhlo.dynamic_reshape %21, %28 : (tensor<2x?x?xf32>, tensor<5xi32>) -> tensor<2x2x?x4x?xf32>
// CHECK-NEXT:      return %29 : tensor<2x2x?x4x?xf32>

}

    %0 = "tfl.pseudo_const"() {value = dense<0.962260901> : tensor<1xf32>} : () -> tensor<1xf32>
    %1 = func.call @func_0_GPU_FLOAT(%arg0, %0) {tac.device = "GPU", tac.inference_type = "FLOAT", tac.interface_name = "func_0"} : (tensor<1x200x200x200xf32>, tensor<1xf32>) -> tensor<1x200x200x200xf32>
    %2 = "tfl.pseudo_const"() {value = dense<0.895973444> : tensor<1xf32>} : () -> tensor<1xf32>