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

// CHECK-LABEL: func @addN_variant
func.func @addN_variant(%arg0: tensor<!tf_type.variant<tensor<2xf32>>>, %arg1: tensor<!tf_type.variant<tensor<2xf32>>>, %arg2: tensor<!tf_type.variant<tensor<2xf32>>>) -> tensor<!tf_type.variant<tensor<2xf32>>> {
  // CHECK: tf.AddN

        tensor<2xf32>, tensor<2xf32>, tensor<2xf32>,
        tensor<2xf32>, tensor<2xf32>, tensor<2xf32>, tensor<2xf32>,
        tensor<4x2xf32>, tensor<4xf32>,
        tensor<1x4xf32>, tensor<1x2xf32>,
        tensor<2xf32>, tensor<2xf32>, tensor<2xf32>, tensor<2xf32>) -> tensor<*xf32>
  %24 = "quantfork.stats"(%23) {layerStats = dense<[-1.0, 2.0]> : tensor<2xf32>} : (tensor<*xf32>) -> tensor<*xf32>

    %4 = "quantfork.stats"(%3) {layerStats = dense<[0.000000e+00, 0.18486841]> : tensor<2xf32>} : (tensor<*xf32>) -> tensor<*xf32>
    %5 = "tf.Identity"(%4) {device = ""} : (tensor<*xf32>) -> tensor<*xf32>
    %6 = "quantfork.stats"(%5) {layerStats = dense<[0.000000e+00, 0.18486841]> : tensor<2xf32>} : (tensor<*xf32>) -> tensor<*xf32>

  // CHECK:  %[[MUL1:.*]] = "tf.Mul"(%arg1, %[[RSQRT]]) : (tensor<2xf32>, tensor<2xf32>) -> tensor<2xf32>
  // CHECK:  %[[MUL2:.*]] = "tf.Mul"(%arg0, %[[MUL1]]) : (tensor<1x1x6x2xf32>, tensor<2xf32>) -> tensor<1x1x6x2xf32>
  // CHECK:  %[[MUL3:.*]] = "tf.Mul"(%[[MEAN]], %[[MUL1]]) : (tensor<2xf32>, tensor<2xf32>) -> tensor<2xf32>
  // CHECK:  %[[SUB:.*]] = "tf.Sub"(%arg2, %[[MUL3]]) : (tensor<2xf32>, tensor<2xf32>) -> tensor<2xf32>

    %cst = "tf.Const"() {value = dense<3.00000000e-1> : tensor<2x3xf32>} : () -> tensor<2x3xf32>
    %cst_0 = "tf.Const"() {value = dense<4.00000000e-1> : tensor<3xf32>} : () -> tensor<3xf32>
    %0 = "quantfork.stats"(%arg0) {layerStats = dense<[6.00000000e-6, 9.00000000e-1]> : tensor<2xf32>} : (tensor<?x2xf32>) -> tensor<?x2xf32>

  %4 = "tfl.minimum"(%0, %cst) : (tensor<32xf32>, tensor<32xf32>) -> tensor<32xf32>
  %5 = "tfl.minimum"(%1, %cst) : (tensor<32xf32>, tensor<32xf32>) -> tensor<32xf32>
  %6 = "tfl.minimum"(%2, %cst) : (tensor<32xf32>, tensor<32xf32>) -> tensor<32xf32>
  %7 = "tfl.minimum"(%3, %cst) : (tensor<32xf32>, tensor<32xf32>) -> tensor<32xf32>

//     : (tensor<1x2xf32>) -> tensor<2xf32>
//   %items1:4 = "tf.Unpack"(%[[INP1]]) {axis = 0 : i64}
//     : (tensor<4x2xf32>) -> (tensor<2xf32>, tensor<2xf32>, tensor<2xf32>,
//     tensor<2xf32>)
//   %axis = "tf.Const"() {value = dense<0> : tensor<i64>}
//   %0 = "tf.Pack"(items1#3, items1#2, items1#1, items1#0, %items0, %axis)
//     : (tensor<2xf32>, tensor<2xf32>, tensor<2xf32>, tensor<2xf32>,

    %0 = "tf.Div"(%arg1, %cst_0) : (tensor<?x?xf32>, tensor<f32>) -> tensor<?x?xf32>
    %1 = "tf.AddV2"(%0, %cst_1) : (tensor<?x?xf32>, tensor<f32>) -> tensor<?x?xf32>
    %2 = "tf.Floor"(%1) : (tensor<?x?xf32>) -> tensor<?x?xf32>
    %3 = "tf.ClipByValue"(%2, %cst_5, %cst_6) : (tensor<?x?xf32>, tensor<f32>, tensor<f32>) -> tensor<?x?xf32>
    %4 = "tf.Cast"(%3) {Truncate = false} : (tensor<?x?xf32>) -> tensor<?x?xi8>

  // CHECK-SAME: (tensor<5x3xf32>, tensor<2xi32>, tensor<2xi32>) -> tensor<1x3xf32>
  // CHECK: %[[UPDATE_SLICE0_START:.*]] = "tf.Const"() <{value = dense<0> : tensor<2xi32>}> : () -> tensor<2xi32>
  // CHECK: %[[UPDATE_SLICE0:.*]] = "tf.Slice"(%[[GATHER]], %[[UPDATE_SLICE0_START]], %[[SLICE_SIZE]]) : (tensor<2x3xf32>, tensor<2xi32>, tensor<2xi32>) -> tensor<1x3xf32>

// CHECK:           %[[VAL_4:.*]]:2 = call @func_0_GPU_FLOAT(%[[VAL_0]], %[[VAL_1]], %[[VAL_2]], %[[VAL_3]]) {tac.device = "GPU", tac.inference_type = "FLOAT", tac.interface_name = "func_0"} : (tensor<1xf32>, tensor<1xf32>, tensor<1xf32>, tensor<1xf32>) -> (tensor<1xf32>, tensor<1xf32>)