// CHECK-NEXT: [[r1:%.*]] = tfrt_fallback_async.executeop {{.*}} "tf.BiasAdd"([[r0]], [[result]])
  %3 = "tf.BiasAdd"(%2, %0) {T = f32, _output_shapes = ["tfshape$dim { size: 3 } dim { size: 3 }"], data_format = "NHWC", device = "/device:CPU:0"} : (tensor<3x3xf32>, tensor<3xf32>) -> tensor<3x3xf32>

  func.func private @composite_dot_general_fn_1(%arg0: tensor<3x3xf32>, %arg1: tensor<3x11xf32>) -> tensor<3x11xf32> attributes {_from_xla_call_module, tf_quant.composite_function} {
    %0 = stablehlo.dot_general %arg0, %arg1, contracting_dims = [1] x [0], precision = [DEFAULT, DEFAULT] : (tensor<3x3xf32>, tensor<3x11xf32>) -> tensor<3x11xf32>
    return %0 : tensor<3x11xf32>
  }
}

// -----

    %cst = arith.constant dense<0.000000e+00> : tensor<5xf32>
    %cst_0 = arith.constant dense<1.0> : tensor<5x3xf32>
    %0 = "tfl.fully_connected"(%arg0, %cst_0, %cst) {fused_activation_function = "NONE", keep_num_dims = false, weights_format = "DEFAULT"} : (tensor<?x3xf32>, tensor<5x3xf32>, tensor<5xf32>) -> tensor<?x5xf32>
    func.return %0, %0 : tensor<?x5xf32>, tensor<?x5xf32>
  }

  %2 = "tf.Const"() {value = dense<0.0> : tensor<2x2xf32>} : () -> tensor<2x2xf32>
  %3 = "tf.Const"() {value = dense<0.0> : tensor<4x4xf32>} : () -> tensor<4x4xf32>
  "tf.SendTPUEmbeddingGradients"(%2, %3) {_tpu_embedding_layer = "call1", config = "\0A\0B\0C\0D", operandSegmentSizes = array<i32: 2, 0>} : (tensor<2x2xf32>, tensor<4x4xf32>) -> ()

      func.func @main(%arg: tensor<1x2xf32> {tf_saved_model.index_path = ["input_tensor:0"]}) -> (tensor<1x2xf32> {tf_saved_model.index_path = ["output_tensor:0"]}) attributes {tf.entry_function = {inputs = "input_tensor:0", outputs = "output_tensor:0"}, tf_saved_model.exported_names = ["main"]} {
        %0 = tf_executor.graph {
          tf_executor.fetch %arg : tensor<1x2xf32>
        }
        return %0 : tensor<1x2xf32>
      }

func.func @RemoveRedundantPack(%arg0: tensor<2x5xf32>) -> (tensor<2x5xf32>, tensor<5xf32>) {
  %0:2 = "tfl.unpack"(%arg0) {axis = 0 : i32, num = 2 : i32} : (tensor<2x5xf32>) -> (tensor<5xf32>, tensor<5xf32>)
  %1 = "tfl.pack"(%0#0, %0#1) {axis = 0 : i32, values_count = 2 : i32} : (tensor<5xf32>, tensor<5xf32>) -> (tensor<2x5xf32>)
  func.return %1, %0#0: tensor<2x5xf32>, tensor<5xf32>
  // CHECK: %[[UNPACK:.*]]:2 = "tfl.unpack"

  %outputs =  "tf.Const"() {device = "/job:localhost/replica:0/task:0/device:CPU:0", value = dense<0> : tensor<i32>} : () -> tensor<i32>...

  %one = "glue.constant"() { value = 1: i32 } : () -> i32
  %done = "glue.compare" (%one, %one) { predicate = #glue<"compare LTE"> } : (i32, i32) -> i1
  %2 = mhlo.constant dense<[[1.1]]> : tensor<1x1xf32>
  %3 = mhlo.multiply %2, %2 : tensor<1x1xf32>
  %cst = "tf.Const"() {value = dense<0.0> : tensor<f32>} : () -> tensor<f32>
  %0 = "tf.AddV2"(%arg0, %cst) {device = "/device:CPU:0"} : (tensor<f32>, tensor<f32>) -> tensor<f32>

//     -> tensor<5x2xf32>
//
// is lowered to
//
//   %shape = "tf.Const"() {value = dense<[-1, 2]> : tensor<2xi64>}
//   %inp0 = "tf.Reshape"(%arg0, %shape)
//     : (tensor<2xf32>, tensor<2xi64>) -> tensor<1x2xf32>
//   %inp1 = "tf.Reshape"(%arg1, %shape)
//     : (tensor<2x2x2xf32>, tensor<2xi64>) -> tensor<4x2xf32>
//   %items0 = "tf.Unpack"(%[[INP0]]) {axis = 0 : i64}

  %4 = "tf.MatMul"(%arg0, %3) {device = "", transpose_a = false, transpose_b = false} : (tensor<2x3xf32>, tensor<3x4xf32>) -> tensor<2x4xf32>
  %5 = "tf.Identity"(%4) {device = ""} : (tensor<2x4xf32>) -> tensor<2x4xf32>
  %6 = "tf.Identity"(%5) {device = ""} : (tensor<2x4xf32>) -> tensor<2x4xf32>
  func.return %6 : tensor<2x4xf32>