// CHECK-NEXT:   "tf.LayerNorm"(%arg4, %arg5, %arg6, %arg7) {_tfl_quant_trait = "fully_quantizable", device = ""} : (tensor<128x128xf32>, tensor<1xf32>, tensor<1xf32>, tensor<1xi32>) -> tensor<128x128xf32>
// CHECK-NEXT:   "tfl.yield"
// CHECK-NEXT: }) {_tfl_quant_trait = "fully_quantizable", device = ""} :

    %1 = "tf.BiasAdd"(%0, %cst_0) {data_format = "NHWC", device = ""} : (tensor<*xf32>, tensor<6xf32>) -> tensor<*xf32>
    func.return %1: tensor<*xf32>
  }
  func.func private @composite_depthwise_conv2d_fn(%arg0: tensor<1x3x4x3xf32>, %arg1: tensor<2x3x3x2xf32>) -> tensor<*xf32> attributes {tf_quant.composite_function} {

  %cst_0 = "tf.Const"() {value = dense<0.0235294122> : tensor<f32>} : () -> tensor<f32>
  %cst_1 = "tf.Const"() {value = dense<0.00708661414> : tensor<1xf32>} : () -> tensor<1xf32>
  %cst_2 = "tf.Const"() {value = dense<1.799000e+03> : tensor<1xf32>} : () -> tensor<1xf32>

func.func @transpose_simple_1d() -> tensor<2xf32> {
  %0 = stablehlo.constant dense<[0.000000e+0, 1.000000e+0]> : tensor<2xf32>
  %1 = stablehlo.transpose %0, dims = [0] : (tensor<2xf32>) -> tensor<2xf32>
  return %1 : tensor<2xf32>
}
// CHECK-DAG: %[[CONST_0:.+]] = stablehlo.constant dense<[0.000000e+00, 1.000000e+00]> : tensor<2xf32>
// CHECK-NOT: transpose
// CHECK: return %[[CONST_0]] : tensor<2xf32>

// -----

// CHECK-LABEL: fuseMulIntoPerTensorConv2dWithQDQs
func.func @fuseMulIntoPerTensorConv2dWithQDQs(%arg0: tensor<256x32x32x3xf32>) -> tensor<256x8x7x3xf32> {
  %cst = arith.constant dense<1.5> : tensor<3xf32>
  %cst_0 = arith.constant dense<[1.0, 2.0, 3.0]> : tensor<3xf32>
  %w = arith.constant dense<2.0> : tensor<3x3x3x3xf32>

  ^bb1(%0: tensor<!tf_type.resource<tensor<?xf32>>>, %1: tensor<!tf_type.resource<tensor<?xf32>>>, %2: tensor<!tf_type.resource<tensor<?xf32>>>):
    "tf.AssignVariableOp"(%0, %arg0) : (tensor<!tf_type.resource<tensor<?xf32>>>, tensor<?xf32>) -> ()
    cf.br ^bb1(%1, %2, %0 : tensor<!tf_type.resource<tensor<?xf32>>>, tensor<!tf_type.resource<tensor<?xf32>>>, tensor<!tf_type.resource<tensor<?xf32>>>)
  }

// -----

func.func @testAddReluPack(%arg0: tensor<1xf32>, %arg1: tensor<1xf32>) {
   // CHECK: tac.device = "GPU", tac.inference_type = "FLOAT"
  %0 = "tfl.add"(%arg0, %arg1) {fused_activation_function = "RELU6"} : (tensor<1xf32>, tensor<1xf32>) -> tensor<1xf32>
   // CHECK: tac.device = "GPU", tac.inference_type = "FLOAT"
  %1 = "tfl.add"(%arg0, %0) {fused_activation_function = "RELU"} : (tensor<1xf32>, tensor<1xf32>) -> tensor<1xf32>

    %1 = "tf.BiasAdd"(%0, %cst_0) {data_format = "NHWC", device = ""} : (tensor<*xf32>, tensor<6xf32>) -> tensor<*xf32>
    func.return %1: tensor<*xf32>
  }
  func.func private @composite_depthwise_conv2d_fn(%arg0: tensor<1x3x4x3xf32>, %arg1: tensor<2x3x3x2xf32>) -> tensor<*xf32> attributes {tf_quant.composite_function} {

  %1 = stablehlo.constant dense<[4.0, 6.0]> : tensor<2xf32>
  %2 = stablehlo.divide %0, %1 : tensor<2xf32>
  return %2 : tensor<2xf32>
}

// CHECK-LABEL: foldDivLHSSplat
// CHECK: stablehlo.constant dense<[5.000000e-01, 0.333333343]> : tensor<2xf32>

// -----

func.func @foldDivRHSSplat() -> tensor<2xf32> {
  %0 = stablehlo.constant dense<[4.0, 6.0]> : tensor<2xf32>
  %1 = stablehlo.constant dense<2.0> : tensor<2xf32>

  %tmp0, %ctl0 = tfl.control_node controls "tfl.neg"(%arg0): (tensor<1xf32>) -> tensor<1xf32>
  %tmp1, %ctl1 = tfl.control_node controls "tfl.neg"(%tmp0): (tensor<1xf32>) -> tensor<1xf32>
  %tmp2, %ctl2 = tfl.control_node controls "tfl.neg"(%tmp1): (tensor<1xf32>) -> tensor<1xf32>
  %tmp3, %ctl3 = tfl.control_node controls "tfl.neg"(%tmp2): (tensor<1xf32>) -> tensor<1xf32>