def load_op_library(library_filename):
"""Loads a TensorFlow plugin, containing custom ops and kernels.
Pass "library_filename" to a platform-specific mechanism for dynamically
loading a library. The rules for determining the exact location of the
library are platform-specific and are not documented here. When the
library is loaded, ops and kernels registered in the library via the
`REGISTER_*` macros are made available in the TensorFlow process. Note
that ops with the same name as an existing op are rejected and not
registered with the process.
Args:
library_filename: Path to the plugin.
Relative or absolute filesystem path to a dynamic library file.
Returns:
A python module containing the Python wrappers for Ops defined in
the plugin.
Raises:
RuntimeError: when unable to load the library or get the python wrappers.
"""
status = py_tf.TF_NewStatus()
lib_handle = py_tf.TF_LoadLibrary(library_filename, status)
try:
error_code = py_tf.TF_GetCode(status)
if error_code != 0:
error_msg = compat.as_text(py_tf.TF_Message(status))
# pylint: disable=protected-access
raise errors_impl._make_specific_exception(None, None, error_msg,
error_code)
# pylint: enable=protected-access
finally:
py_tf.TF_DeleteStatus(status)
op_list_str = py_tf.TF_GetOpList(lib_handle)
op_list = op_def_pb2.OpList()
op_list.ParseFromString(compat.as_bytes(op_list_str))
wrappers = py_tf.GetPythonWrappers(op_list_str)
# Delete the library handle to release any memory held in C
# that are no longer needed.
py_tf.TF_DeleteLibraryHandle(lib_handle)
# Get a unique name for the module.
module_name = hashlib.md5(wrappers).hexdigest()
if module_name in sys.modules:
return sys.modules[module_name]
module = imp.new_module(module_name)
# pylint: disable=exec-used
exec(wrappers, module.__dict__)
# Stash away the library handle for making calls into the dynamic library.
module.LIB_HANDLE = lib_handle
# OpDefs of the list of ops defined in the library.
module.OP_LIST = op_list
sys.modules[module_name] = module
return module
def load_op_library(library_filename):
"""Loads a TensorFlow plugin, containing custom ops and kernels.
Pass "library_filename" to a platform-specific mechanism for dynamically
loading a library. The rules for determining the exact location of the
library are platform-specific and are not documented here. When the
library is loaded, ops and kernels registered in the library via the
`REGISTER_*` macros are made available in the TensorFlow process. Note
that ops with the same name as an existing op are rejected and not
registered with the process.
Args:
library_filename: Path to the plugin.
Relative or absolute filesystem path to a dynamic library file.
Returns:
A python module containing the Python wrappers for Ops defined in
the plugin.
Raises:
RuntimeError: when unable to load the library or get the python wrappers.
"""
lib_handle = py_tf.TF_LoadLibrary(library_filename)
op_list_str = py_tf.TF_GetOpList(lib_handle)
op_list = op_def_pb2.OpList()
op_list.ParseFromString(compat.as_bytes(op_list_str))
wrappers = py_tf.GetPythonWrappers(op_list_str)
# Delete the library handle to release any memory held in C
# that are no longer needed.
py_tf.TF_DeleteLibraryHandle(lib_handle)
# Get a unique name for the module.
module_name = hashlib.md5(wrappers).hexdigest()
if module_name in sys.modules:
return sys.modules[module_name]
module = imp.new_module(module_name)
# pylint: disable=exec-used
exec(wrappers, module.__dict__)
# Stash away the library handle for making calls into the dynamic library.
module.LIB_HANDLE = lib_handle
# OpDefs of the list of ops defined in the library.
module.OP_LIST = op_list
# Allow this to be recognized by AutoGraph.
setattr(module, '_IS_TENSORFLOW_PLUGIN', True)
sys.modules[module_name] = module
return module
