patch v8 float compute lose precision