}
  return absl::OkStatus();
}

// Constructor of the decompose context.
TFRDecomposeContext::TFRDecomposeContext(mlir::ModuleOp tfr_module)
    : tfr_module_(tfr_module), pm_(tfr_module_.getContext()) {
  mlir::OpPassManager& func_pm = pm_.nest<mlir::func::FuncOp>();

  // Prepare the imported graph.
  func_pm.addPass(mlir::CreateExecutorDialectToFunctionalConversionPass());

  // Runs the decompose passes on the user_module.
  Status DecomposeGraph(mlir::ModuleOp user_module);

  // Erases the tfr_module created.
  void Destroy();

 private:
  mlir::ModuleOp tfr_module_;
  mlir::PassManager pm_;

  GraphExportConfig export_confs_;
};

// Decomposes the NodeDef to a set of primitive ops according to the decompose

                                      RewritePatternSet &patterns);

// Decompose ops.
std::unique_ptr<OperationPass<func::FuncOp>> CreateDecomposeTFOpsPass(
    std::optional<ModuleOp> tfr_module = std::nullopt);

// Rewrites quantized operands and results with their storage types.
// This pass should be run at module level after decomposition, if there are
// quantized operands or results.

  mlir::MLIRContext context(registry);
  context.loadAllAvailableDialects();

  auto tf_module = mlir::parseSourceString<ModuleOp>(module_op_str, &context);
  if (!tf_module) {
    return absl::UnknownError("Failed to parse MLIR module");
  }

  auto mlir_file_path = ExportModule(*tf_module);
  if (!mlir_file_path.ok()) {
    return absl::UnknownError(

        "  ).call,\n",
        "    with_gradient=False,\n",
        "    native_serialization=True,\n",
        "    native_serialization_platforms=('cpu',))\n",
        "\n",
        "tf_module = tf.Module()\n",
        "tf_module.f = tf.function(\n",
        "    jax_callable,\n",
        "    autograph=False,\n",
        "    input_signature=[\n",
        "        tf.TensorSpec(input_shape, jnp.float32, 'lhs_operand')\n",

    return 2
  os.makedirs(export_path)

  tf_module = module_ctor()
  if version == 2:
    options = save_options.SaveOptions(save_debug_info=True)
    saved_model.save(tf_module, export_path, options=options)
  else:
    builder = saved_model.builder.SavedModelBuilder(export_path)
    builder.add_meta_graph_and_variables(tf_module, ["serve"])
    builder.save()


if __name__ == "__main__":

                    arith::ArithDialect, func::FuncDialect>();
  }

  explicit RaiseToTFOpsPass(std::optional<ModuleOp> tfr_module,
                            bool materialize_derived_attrs)
      : external_tfr_module_(tfr_module),
        materialize_derived_attrs_(materialize_derived_attrs) {}

  StringRef getArgument() const final { return "tfr-raise-to-tf"; }

  os.flush();

  output->os() << result;
  output->keep();

  return absl::OkStatus();
}

tensorflow::Status ConvertTFToStableHLO(
    ModuleOp tf_module, const PassPipelineCLParser& pass_pipeline) {
  PassManager pm(tf_module.getContext());
  if (failed(applyPassManagerCLOptions(pm))) {
    return tensorflow::errors::Aborted(
        "Failed to apply MLIR pass manager CL options.");
  }

// module itself without inserting into the tf module.
FailureOr<StringAttr> RenameStablehloFunctions(
    MLIRContext *context, SymbolTableCollection &symbol_tables,
    ModuleOp tf_module, ModuleOp stablehlo_module) {
  SymbolTable &tf_symbol_table = symbol_tables.getSymbolTable(tf_module);
  // `stablehlo_module` is deleted right after the deserialization, so no need
  // to store its `SymbolTable` to `SymbolTableCollection`.

}

}  // namespace

// Creates an instance of the pass to decompose the TF ops.
std::unique_ptr<OperationPass<func::FuncOp>> CreateDecomposeTFOpsPass(
    std::optional<ModuleOp> tfr_module) {
  return std::make_unique<DecomposeTFOpsPass>(tfr_module);
}

static PassRegistration<DecomposeTFOpsPass> pass([] {
  return CreateDecomposeTFOpsPass();
});

}  // namespace TFR