def _get_pipeline_path(self, source_dir):
"""
Get the minddata pipeline file path.
Args:
source_dir (str): The minddata pipeline source dir.
Returns:
str, the minddata pipeline file path.
"""
pipeline_path = os.path.join(
source_dir, self._raw_pipeline_file_name.format(self._device_id))
try:
pipeline_path = validate_and_normalize_path(pipeline_path)
except RuntimeError:
logger.warning('Minddata pipeline file is invalid.')
raise ProfilerPathErrorException(
'Minddata pipeline file is invalid.')
if not os.path.isfile(pipeline_path):
logger.warning('The minddata pipeline file <%s> not found.',
pipeline_path)
raise ProfilerFileNotFoundException(pipeline_path)
return pipeline_path
def _get_op_task_id_map(self):
"""
Read hwts data file, get the task time info.
Returns:
list: all hwts task time info.
"""
op_map_result = []
hwts_list = []
if not os.path.exists(self._hwts_output_file):
logger.error('The hwts output file does not exist.')
raise ProfilerFileNotFoundException('hwts output file')
with open(self._hwts_output_file, 'r') as data_file:
lines = data_file.readlines()
for line in lines:
if line.startswith("Start of task") or line.startswith("End of task"):
line_split = line.split()
container = HWTSContainer(line_split)
hwts_list.append(container)
# hwts op map by taskId
for hwts in hwts_list:
if hwts.task_id in self._op_task_info.keys():
hwts.op_name = self._op_task_info[hwts.task_id]
op_map_result.append(hwts)
return op_map_result
def _get_and_validate_path(self, file_name):
"""Generate op or activity file path from file name, and validate this path."""
file_path = os.path.join(self._profiling_dir,
file_name.format(self._device_id))
file_path = validate_and_normalize_path(file_path)
if not os.path.exists(file_path):
logger.error(f"Failed to find parsed timeline file {file_path}.")
raise ProfilerFileNotFoundException('parsed timeline file')
return file_path
def _get_file_path(self):
"""Get the proto file path."""
file_path = os.path.join(self._profiling_dir,
self._proto_file_path.format(self._device_id))
file_path = validate_and_normalize_path(file_path)
if not os.path.exists(file_path):
msg = 'The memory file does not exist!'
logger.error(msg)
raise ProfilerFileNotFoundException(msg=msg)
return file_path
def _get_file_path(self):
"""Get the proto file path."""
file_path = os.path.join(self._profiling_dir,
self._proto_file_path.format(self._device_id))
file_path = validate_and_normalize_path(file_path)
if not os.path.exists(file_path):
logger.warning(
'The memory file does not exist! Please ignore the warning '
'if you are running heterogeneous training.')
raise ProfilerFileNotFoundException(msg=file_path)
return file_path
def _aicore_trace_data_load(self):
"""Load data according to the parsed AICORE operator types file."""
file_path = query_latest_trace_time_file(self._profiling_dir, int(self._device_id))
if not file_path:
logger.error("Failed to find parsed trace time file.")
raise ProfilerFileNotFoundException('parsed step trace time file')
file_path = validate_and_normalize_path(file_path)
with open(file_path, 'r') as handle:
csv_reader = csv.reader(handle)
self.__column__ = next(csv_reader)
self._aicore_trace_data = list(csv_reader)
self._size = len(self._aicore_trace_data) - 1
self._load_point_info()
def _search_file(self, profiling_id, device_id):
"""
Search all framework files in raw profiling path.
Args:
profiling_id (str): The profiling ID.
device_id (str): The device ID.
Raises:
ProfilerFileNotFoundException: If the framework files are not found.
"""
# first search in the JOB dir, and if not, search in the sub directory
# in the JOB
self._search_file_from_job_path(device_id, search_in_sub_path=False)
if self._backend_type is None:
self._search_file_from_job_path(device_id, search_in_sub_path=True)
self._search_file_from_data_path(profiling_id, device_id)
if self._backend_type is None:
raise ProfilerFileNotFoundException('Framework')
self._framework_path['graph'].sort()
self._framework_path['task'].sort()
def _load_timeline_data(self):
"""Load timeline data from file."""
file_path = os.path.join(
self._profiling_dir,
self._output_timeline_data_file_path.format(self._device_id))
file_path = validate_and_normalize_path(file_path)
if not os.path.exists(file_path):
logger.error("Failed to find parsed timeline file.")
raise ProfilerFileNotFoundException('parsed timeline file')
timeline_list = []
try:
with open(file_path, 'r') as f_obj:
for line in f_obj:
if not line.startswith('op_name'):
line_list = line.strip('\n').split(',')
timeline_list.append(line_list)
except (IOError, OSError) as err:
logger.error(
'Error occurred when read timeline intermediate file: %s', err)
raise ProfilerIOException
return timeline_list
