# CHECK:      func {{@[a-zA-Z_0-9]+}}(
# CHECK-SAME:   {{.*}})
# CHECK-SAME: attributes {{.*}} tf_saved_model.exported_names = ["key"]

# CHECK:      func {{@[a-zA-Z_0-9]+}}(
# CHECK-SAME:   {{.*}})
# CHECK-SAME: attributes {{.*}} tf_saved_model.exported_names = ["key2"]


def Test():

  x = tf.constant(1.0, shape=(3, 3))
  y = tf.constant(1.0, shape=(3, 3))

  s = tf.transpose(x)

  attributes {tf.entry_function = {control_outputs = "", inputs = "", outputs = "dense_2/Add:0"}, tf_saved_model.exported_names = ["serving_default"]} {
    %0 = "tf.VarHandleOp"() {_class = ["loc:@dense/kernel"], allowed_devices = [], container = "", device = "", shared_name = "dense/kernel"} : () -> tensor<!tf_type.resource<tensor<100x50xf32>>>

  # CHECK: attributes {{.*}} tf_saved_model.exported_names = ["my_predict"]
  @tf.function(input_signature=[
      tf.TensorSpec([1, 28, 28, 1], tf.float32),
  ])
  def my_predict(self, x):
    return self.model(x)


if __name__ == '__main__':

      absl::StrSplit(saved_model_exported_names, ',', absl::SkipEmpty());
  absl::Span<std::string> exported_names(exported_names_vector);

  if (import_saved_model_object_graph) {
    mlir::MLIRContext context;
    auto module_or = tensorflow::SavedModelObjectGraphToMlirImport(
        input_filename, tags, exported_names, &context);
    if (!module_or.status().ok()) return 1;

    module_or.value()->print(output->os());

# CHECK:      func @[[restore]](
# CHECK-SAME:   [[variable_path:%.*]]: tensor<!tf_type.string> {tf_saved_model.bound_input = @[[sym_name]]}
# CHECK-SAME: tf_saved_model.exported_names = ["{{__tf_saved_model_session_initializer.*}}"]
# CHECK-SAME: tf_saved_model.initializer_type = "restore_op"
# CHECK: [[v0:%.*]] = "tf.RestoreV2"([[variable_path]]
# CHECK: [[v1:%.*]] = "tf.Identity"([[v0]])

  attributes {tf.entry_function = {control_outputs = "", inputs = "", outputs = "dense_2/Add:0"}, tf_saved_model.exported_names = ["serving_default"]} {
    %0 = "tf.VarHandleOp"() {_class = ["loc:@dense/kernel"], allowed_devices = [], container = "", device = "", shared_name = "dense/kernel"} : () -> tensor<!tf_type.resource<tensor<100x50xf32>>>

# CHECK-SAME:             -> (tensor<3x3xf32> {tf_saved_model.index_path = ["r"]})
# CHECK-SAME: attributes {{.*}} tf_saved_model.exported_names = ["key"]

# CHECK:      func {{@[a-zA-Z_0-9]+}}(
# CHECK-SAME:   [[ARG2:%.*]]: tensor<3x1xf32> {tf_saved_model.index_path = ["x"]},

      module.getLoc(), name,
      FunctionType::get(module.getContext(), /*inputs=*/{}, /*results=*/{}));

  // Make savedmodel verification happy.
  init->setAttr("tf_saved_model.exported_names",
                builder.getStrArrayAttr({name}));

  builder.setInsertionPointToStart(init.addEntryBlock());

  for (func::FuncOp func : tf_saved_model::GetInitializerFunctions(module)) {

absl::StatusOr<mlir::OwningOpRef<mlir::ModuleOp>>
SavedModelObjectGraphToMlirImport(
    absl::string_view saved_model_dir,
    const std::unordered_set<std::string>& tags,
    absl::Span<std::string> exported_names, mlir::MLIRContext* context,
    bool unconditionally_use_set_output_shapes = false);

// Converts a TensorFlow V1 SavedModel stored in the directory with the given

                      ->cast<mlir::MLIRContext*>();

  // Load the saved model into MLIR TF dialect.
  absl::Span<std::string> exported_names(nullptr, 0);
  auto module_or =
      tensorflow::ConvertSavedModelToMlir(&bundle, context, exported_names);
  if (!module_or.status().ok()) {
    return None();
  }

  // Make a module to wrap MLIR module and allow getting strings and running
  // transforms.