%0 = "tf.Div"(%arg1, %cst_0) : (tensor<2x2xf32>, tensor<f32>) -> tensor<2x2xf32>
    %1 = "tf.AddV2"(%0, %cst_1) : (tensor<2x2xf32>, tensor<f32>) -> tensor<2x2xf32>
    %2 = "tf.Floor"(%1) : (tensor<2x2xf32>) -> tensor<2x2xf32>
    %3 = "tf.ClipByValue"(%2, %cst_5, %cst_6) : (tensor<2x2xf32>, tensor<f32>, tensor<f32>) -> tensor<2x2xf32>
    %4 = "tf.Cast"(%3) {Truncate = false} : (tensor<2x2xf32>) -> tensor<2x2xi8>

    %3 = stablehlo.concatenate %2, %1, dim = 0 : (tensor<1x3xf32>, tensor<1x3xf32>) -> tensor<2x3xf32>
    return %3 : tensor<2x3xf32>
  }
  func.func private @composite_dot_general_fn_1(%arg0: tensor<1x2xf32>, %arg1: tensor<2x3xf32>) -> tensor<1x3xf32> attributes {_from_xla_call_module, tf_quant.composite_function} {

// CHECK-LABEL: hardcode_all
func.func @hardcode_all(%arg0: tensor<2x2xf32>, %arg1: tensor<2x1xf32>) -> tensor<2x2xf32> {
  %0 = "tfl.add"(%arg0, %arg1) {fused_activation_function="NONE"}: (tensor<2x2xf32>, tensor<2x1xf32>) -> tensor<2x2xf32>
  func.return %0 : tensor<2x2xf32>

// CHECK: %[[q0:.*]] = "tfl.quantize"(%arg1) <{qtype = tensor<2x1x!quant.uniform<u8:f32, 0.0078431372549019607:128>>}>

     %t2 = "tf.Acosh"(%arg1) : (tensor<2xf32>) -> tensor<2xf32>
    "tf.Yield"(%t0, %t1, %t2) : (tensor<2xf32>, tensor<2xf32>, tensor<2xf32>) -> ()
    }, {
     %e0 = "tf.Neg"(%arg1) : (tensor<2xf32>) -> tensor<2xf32>
     %e1 = "tf.Relu"(%arg1) : (tensor<2xf32>) -> tensor<2xf32>
     %e2 = "tf.Sin"(%arg1) : (tensor<2xf32>) -> tensor<2xf32>
     "tf.Yield"(%e0, %e1, %e2) : (tensor<2xf32>, tensor<2xf32>, tensor<2xf32>) -> ()

  %min = arith.constant dense<0.0> : tensor<16xf32>
  %max = arith.constant dense<255.0> : tensor<16xf32>
  %mini = "tf.Identity"(%min) : (tensor<16xf32>) -> tensor<16xf32>
  %maxi = "tf.Identity"(%max) : (tensor<16xf32>) -> tensor<16xf32>
  %fq = "tf.FakeQuantWithMinMaxVarsPerChannel"(%in, %mini, %maxi) {num_bits = 5, narrow_range = false} : (tensor<3x3x3x16xf32>, tensor<16xf32>, tensor<16xf32>) -> tensor<3x3x3x16xf32>

  %min = arith.constant dense<0.0> : tensor<16xf32>
  %max = arith.constant dense<15.0> : tensor<16xf32>
  %mini = "tf.Identity"(%min) : (tensor<16xf32>) -> tensor<16xf32>
  %maxi = "tf.Identity"(%max) : (tensor<16xf32>) -> tensor<16xf32>
  %fq = "tf.FakeQuantWithMinMaxVarsPerChannel"(%in, %mini, %maxi) {num_bits = 3, narrow_range = false} : (tensor<3x3x3x16xf32>, tensor<16xf32>, tensor<16xf32>) -> tensor<3x3x3x16xf32>

  // CHECK: tac.cost = 1.010000e+02
  %1 = "tfl.reshape"(%0, %cst) {tac.device = "CPU"} : (tensor<200xf32>, tensor<2xi64>) -> tensor<2x100xf32>
  func.return %1 : tensor<2x100xf32>
}

// -----

func.func @testConv2DCPU(tensor<256x32x32x3xf32>, tensor<16x3x3x3xf32>, tensor<16xf32>) -> tensor<256x32x32x16xf32> {

  %1:4 = "tfl.split"(%cst, %0) {num_splits = 4 : i32} : (tensor<i32>, tensor<4xf32>) -> (tensor<2xf32>, tensor<2xf32>,tensor<2xf32>, tensor<2xf32>)
  %2 = "tfl.quantize"(%1#0) {qtype = tensor<2x!quant.uniform<u8:f32, 1.0>>} : (tensor<2xf32>) -> tensor<2x!quant.uniform<u8:f32, 1.0>>
  %3 = "tfl.quantize"(%1#1) {qtype = tensor<2x!quant.uniform<u8:f32, 1.0>>} : (tensor<2xf32>) -> tensor<2x!quant.uniform<u8:f32, 1.0>>

  // CHECK-SAME: %[[ARG1:.*]]: tensor<2x4xf32>
  func.func private @composite_and_slice(%arg0: tensor<3x2xf32>, %arg1: tensor<2x4xf32>) -> tensor<2x2xf32> {
    // CHECK: %[[Q1:.*]] = "quantfork.qcast"(%[[ARG0]]) {volatile} : (tensor<3x2xf32>) -> tensor<3x2x!quant.uniform<i8:f32, 5.000000e-03>>

// CHECK: [[UNPACK:%.*]]:3 = "tf.Unpack"(%arg0) <{axis = 0 : i64}> : (tensor<3x2x2xf32>) -> (tensor<2x2xf32>, tensor<2x2xf32>, tensor<2x2xf32>)
// CHECK: [[SCALAR_ZERO:%.*]] = arith.constant dense<0> : tensor<i32>
// CHECK: [[CONCAT:%.*]] = "tf.Concat"([[SCALAR_ZERO]], [[UNPACK]]#0, [[UNPACK]]#1, [[UNPACK]]#2) : (tensor<i32>, tensor<2x2xf32>, tensor<2x2xf32>, tensor<2x2xf32>) -> tensor<?x2xf32>
// CHECK: [[LENGTHS:%.*]] = arith.constant dense<0> : tensor<0xi64>