return %result : tensor<1x1xf32>
  }
}

// -----
// Variable tensor for host can still be used.
//
// CHECK-LABEL:  func.func @serving_default(%arg0: tensor<1x3xf32>) -> (tensor<1x1xf32>, tensor<1x1xf32>) {
// CHECK:  "tf.VarHandleOp"
// CHECK-NOT:  [[VARIABLE:%.*]] = "tf.ReadVariableOp"

// CHECK-LABEL: func private @batched_function
func.func private @batched_function(%arg0: tensor<1x3xf32>) -> tensor<1x3xf32> {
  %2 = "tf.Identity"(%arg0) : (tensor<1x3xf32>) -> tensor<1x3xf32>
  func.return %2 : tensor<1x3xf32>
}

// CHECK-LABEL: func @main
func.func @main(%arg0: tensor<1x3xf32>) -> tensor<*xf32> {
  // CHECK:  "tf.BatchFunction"
  // CHECK-SAME: allowed_batch_sizes = [6]

// CHECK-NEXT:    "tf.IfrtCall"(%arg0, [[ARRAYKEY]]) <{program_id = 6515870160938153680 : i64, variable_arg_indices = [1 : i32]}> {__tpu_compile_metadata_text = "retvals { sharding { } }"} : (tensor<1x3xf32>, tensor<!tf_type.string>) -> tensor<1x1xf32>
// CHECK-NEXT:    return
//
 func.func @serving_default(%arg0: tensor<1x3xf32>) -> tensor<1x1xf32> {

func.func @main(%arg0: tensor<1x4xf32>) -> tensor<1x3xf32> {
  %0 = stablehlo.constant dense<1.0> : tensor<4x3xf32>
  %1 = stablehlo.dot_general %arg0, %0, contracting_dims = [1] x [0], precision = [DEFAULT, DEFAULT] : (tensor<1x4xf32>, tensor<4x3xf32>) -> tensor<1x3xf32>
  return %1 : tensor<1x3xf32>
}
// CHECK: @main(%[[ARG_0:.+]]: tensor<1x4xf32>) -> tensor<1x3xf32>

    return %2 : tensor<1x3xf32>
  }

  func.func private @composite_dot_general_fn(%arg0: tensor<1x2xf32>, %arg1: tensor<2x3xf32>) -> tensor<1x3xf32> attributes {_from_xla_call_module} {
    %0 = stablehlo.dot_general %arg0, %arg1, contracting_dims = [1] x [0] : (tensor<1x2xf32>, tensor<2x3xf32>) -> tensor<1x3xf32>
    return %0 : tensor<1x3xf32>
  }
}

    %1 = "tf.VarHandleOp"() <{container = "x", shared_name = "y"}> : () -> tensor<!tf_type.resource<tensor<3x1xf32>>>
    "tf.AssignVariableOp"(%1, %0#0) : (tensor<!tf_type.resource<tensor<3x1xf32>>>, tensor<3x1xf32>) -> ()
    %2 = "tf.VarHandleOp"() <{container = "x", shared_name = "z"}> : () -> tensor<!tf_type.resource<tensor<1x3xf32>>>

  // CHECK-NEXT:     %[[GROUP:.*]] = "tf.Const"()
  // CHECK-SAME:       [0, 1, 2, 3]
  // CHECK-NEXT:     %[[REDUCED:.*]] = "tf.XlaAllReduce"(%[[ID]], %[[GROUP]]) <{mode = "CrossReplica", reduce_op = "Add"}> : (tensor<f32>, tensor<1x4xi32>) -> tensor<f32>
  // CHECK-NEXT:     "tf.OpA"(%[[REDUCED]]) : (tensor<f32>) -> ()
  tf_device.replicate {n = 4 : i32} {
    "tf_device.cluster"() ({
      "tf.OpA"(%arg0) : (tensor<f32>) -> ()
      tf_device.return