import logging
import os
from common_func.constant import Constant
from common_func.db_name_constant import DBNameConstant
from common_func.file_manager import FileManager
from common_func.file_manager import FileOpen
from common_func.ms_multi_process import MsMultiProcess
from common_func.path_manager import PathManager
from framework.offset_calculator import OffsetCalculator
from msmodel.npu_mem.npu_ai_stack_mem_model import NpuAiStackMemModel
from msparser.data_struct_size_constant import StructFmt
from msparser.interface.data_parser import DataParser
from msparser.npu_mem.npu_op_mem_bean import NpuOpMemDataBean
from profiling_bean.prof_enum.data_tag import DataTag
class NpuOpMemParser(DataParser, MsMultiProcess):
"""
parsing npu op mem data
"""
def __init__(self: any, file_list: dict, sample_config: dict) -> None:
super().__init__(sample_config)
super(DataParser, self).__init__(sample_config)
self._file_list = file_list
self._npu_op_mem_data = []
self._npu_op_mem_model = NpuAiStackMemModel(self._project_path,
DBNameConstant.DB_MEMORY_OP,
[DBNameConstant.TABLE_NPU_OP_MEM_RAW])
def parse(self: any) -> None:
"""
start parsing the data
"""
num_op_mem_files = self._file_list.get(DataTag.MEMORY_OP, [])
if not num_op_mem_files:
return
for _file in num_op_mem_files:
_file_path = PathManager.get_data_file_path(self._project_path, _file)
_file_size = os.path.getsize(_file_path)
if not _file_size:
logging.warning(
"The size of file: %s is zero. Check whether the file size is correct.", _file)
continue
logging.info(
"start parsing npu op mem data file: %s", _file)
self._process_npu_op_mem_data(_file_path, _file_size, num_op_mem_files)
FileManager.add_complete_file(self._project_path, _file)
def save(self: any) -> bool:
"""
save npu op mem data
"""
if self._npu_op_mem_data:
with self._npu_op_mem_model as _model:
_model.flush(DBNameConstant.TABLE_NPU_OP_MEM_RAW, self._npu_op_mem_data)
def ms_run(self: any) -> None:
"""
run function
"""
if not self._file_list.get(DataTag.MEMORY_OP, []):
return
try:
self.parse()
except (OSError, SystemError, ValueError, TypeError, RuntimeError) as err:
logging.error(str(err), exc_info=Constant.TRACE_BACK_SWITCH)
return
self.save()
def _process_npu_op_mem_data(self: any, file_path: str, file_size: int, file_list: list) -> None:
offset_calculator = OffsetCalculator(file_list, StructFmt.MEMORY_OP_SIZE,
self._project_path)
with FileOpen(file_path, "rb") as _npu_op_mem_file:
_all_npu_op_mem_data = offset_calculator.pre_process(_npu_op_mem_file.file_reader, file_size)
for _index in range(file_size // StructFmt.MEMORY_OP_SIZE):
npu_op_mem_data_bean = NpuOpMemDataBean().npu_op_mem_decode(
_all_npu_op_mem_data[_index * StructFmt.MEMORY_OP_SIZE:(_index + 1) * StructFmt.MEMORY_OP_SIZE])
if npu_op_mem_data_bean:
_device_type = "NPU" + ':' + str(npu_op_mem_data_bean.device_id)
self._npu_op_mem_data.append([
str(npu_op_mem_data_bean.node_id),
str(npu_op_mem_data_bean.addr),
npu_op_mem_data_bean.size,
npu_op_mem_data_bean.timestamp,
npu_op_mem_data_bean.thread_id,
npu_op_mem_data_bean.total_allocate_memory,
npu_op_mem_data_bean.total_reserve_memory,
npu_op_mem_data_bean.level,
npu_op_mem_data_bean.type,
_device_type
])
