func.func @multiply(%arg0: tensor<1xf32>, %arg1: tensor<1xf32>) -> tensor<1xf32> {
  %0 = "vhlo.multiply_v1"(%arg0, %arg1) : (tensor<1xf32>, tensor<1xf32>) -> tensor<1xf32>
  func.return %0 : tensor<1xf32>
}

// CHECK:func.func private @multiply(%arg0: tensor<1xf32>, %arg1: tensor<1xf32>) -> tensor<1xf32> {
// CHECK: %0 = "vhlo.multiply_v1"(%arg0, %arg1) : (tensor<1xf32>, tensor<1xf32>) -> tensor<1xf32>
// CHECK: return %0 : tensor<1xf32>
// CHECK:}

func.func @add(%arg0: tensor<1xf32>, %arg1: tensor<1xf32>) -> tensor<1xf32> {
  %0 = "tf.Add"(%arg0, %arg1) : (tensor<1xf32>, tensor<1xf32>) -> tensor<1xf32>
  func.return %0: tensor<1xf32>
// CHECK: %[[CUSTOM_0:.*]] = "tfl.custom"(%arg0, %arg1) <{custom_code = "FlexAdd", custom_option = #tfl<const_bytes : "0x03416464001412034164641A001A002A070A015412023001320000021B171414042801">}> : (tensor<1xf32>, tensor<1xf32>) -> tensor<1xf32>

// CHECK:           [[VAL_5:%.*]] = call @func_1_GPU_FLOAT([[VAL_0]], [[VAL_3]]) {tac.device = "GPU", tac.inference_type = "FLOAT", tac.interface_name = "func_1"} : (tensor<100xf32>, tensor<100xf32>) -> tensor<100xf32>

    %4 = "tf.LayerNorm"(%a1, %a2, %a3, %a4) {_tfl_quant_trait = "fully_quantizable", device = ""} : (tensor<128x128xf32>, tensor<1xf32>, tensor<1xf32>, tensor<1xi32>) -> tensor<128x128xf32>
   "tfl.yield"(%4) : (tensor<128x128xf32>) -> ()
  }) {_tfl_quant_trait = "fully_quantizable", device = ""} : (tensor<128x128xf32>, tensor<1xf32>, tensor<1xf32>, tensor<1xi32>) -> tensor<128x128xf32>

    "tfl.yield"(%1#0, %1#1) : (tensor<*xi32>, tensor<*xf32>) -> ()
  }) : (tensor<i32>, tensor<1xf32>) -> (tensor<i32>, tensor<1xf32>) loc("WhileOp")
  // CHECK: (tensor<i32>, tensor<1xf32>, tensor<i32>) ->
  // CHECK-SAME: (tensor<i32>, tensor<1xf32>, tensor<i32>)
  func.return %0#1 : tensor<1xf32>
}

func.func private @WhileOp_cond(%arg0: tensor<*xi32>, %arg1: tensor<*xf32>, %arg2: tensor<i32>) -> tensor<i1> {

func.func @main(
    %arg0: tensor<10x10xf32>, %scale: tensor<10xf32>, %zp: tensor<10xi32>
  ) -> tensor<10x10xi8> {
  %0 = "tf.UniformQuantize"(%arg0, %scale, %zp) {
    quantization_axis = 1 : i64,
    quantization_min_val = -128 : i64,
    quantization_max_val = 127 : i64
  } : (tensor<10x10xf32>, tensor<10xf32>, tensor<10xi32>) -> tensor<10x10x!tf_type.qint8>

  func.return %0 : tensor<*xf32>
}

// CHECK-NOT: "tfl.reshape"
// CHECK: "tfl.pack"

// -----

func.func @squaredDifference(%arg0: tensor<4xf32>, %arg1: tensor<4xf32>) -> tensor<4xf32> {
  %0 = "tfl.squared_difference"(%arg0, %arg1) : (tensor<4xf32>, tensor<4xf32>) -> tensor<4xf32>
  func.return %0 : tensor<4xf32>
}

    %4 = "tf.BiasAdd"(%3, %cst) {data_format = "NHWC", device = ""} : (tensor<*xf32>, tensor<2xf32>) -> tensor<*xf32>
    %5 = "tf.Relu6"(%4) {device = ""} : (tensor<*xf32>) -> tensor<*xf32>

    func.return %2, %5 : tensor<*xf32>, tensor<*xf32>
  }

  func.func @float_conv_1(%arg0: tensor<1x3x4x3xf32>, %arg1: tensor<2x3x3x2xf32>) -> (tensor<*xf32>, tensor<*xf32>) {

  %0 = "quantfork.stats"(%arg0) {layerStats = dense<[-1.0, 1.0]> : tensor<2xf32>} : (tensor<1x1xf32>) -> tensor<1x1xf32>
  %1 = "tfl.tanh"(%0) : (tensor<1x1xf32>) -> tensor<1x1xf32>
  %2 = "quantfork.stats"(%1) {layerStats = dense<[-1.0, 1.0]> : tensor<2xf32>} : (tensor<1x1xf32>) -> tensor<1x1xf32>
  func.return %2 : tensor<1x1xf32>

  func.return %0 : tensor<? x ? x ? x f32>
}