!119 feature of profiler analyser and framework parser

Merge pull request !119 from chenchao99/profiler_analyser

mindinsight/profiler/__init__.py

@@ -0,0 +1,14 @@
+# Copyright 2020 Huawei Technologies Co., Ltd
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+# ============================================================================

mindinsight/profiler/analyser/__init__.py

@@ -0,0 +1,14 @@
+# Copyright 2020 Huawei Technologies Co., Ltd
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+# ============================================================================

mindinsight/profiler/analyser/analyser.py

@@ -0,0 +1,298 @@
+# Copyright 2020 Huawei Technologies Co., Ltd
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+# ============================================================================
+"""The specific analyser class."""
+import csv
+import json
+import os
+
+from mindinsight.profiler.analyser.base_analyser import BaseAnalyser
+from mindinsight.profiler.common.log import logger
+
+
+class AicoreTypeAnalyser(BaseAnalyser):
+    """
+    The analyser for analyzing the AICORE operator types.
+
+    Args:
+        profiling_dir (str): The directory where the parsed profiling files are
+            located.
+        device_id (str): The device ID.
+    """
+    __col_names__ = ['op_type', 'execution_time', 'execution_frequency',
+                     'percent']
+    _file_name_aicore_type_time = 'aicore_intermediate_{}_type.csv'
+
+    def _load(self):
+        """Load data according to the parsed AICORE operator types file."""
+        op_type_file_path = os.path.join(
+            self._profiling_dir,
+            self._file_name_aicore_type_time.format(self._device_id)
+        )
+        if not os.path.isfile(op_type_file_path):
+            logger.warning('The file <%s> does not exist.', op_type_file_path)
+            return
+
+        with open(op_type_file_path, 'r') as file:
+            csv_reader = csv.reader(file)
+            _ = next(csv_reader)
+            for info in csv_reader:
+                self._data.append(self._convert_field_type(info))
+
+    def _filter(self, filter_condition):
+        """
+        Filter the profiling data according to the filter condition.
+
+        Args:
+            filter_condition (dict): The filter condition.
+        """
+        def _inner_filter(item: list):
+            return self._default_filter(item, filter_condition)
+        self._result = list(filter(_inner_filter, self._data))
+
+    def _convert_field_type(self, row):
+        """
+        Convert the field type to the specific type.
+
+        Args:
+            row (list[str]): One row data from parsed data.
+
+        Returns:
+            list[Union[str, float]], the converted data.
+        """
+        return [row[0], float(row[1]), int(row[2]), float(row[3])]
+
+
+class AicoreDetailAnalyser(BaseAnalyser):
+    """
+    The analyser for analyzing all the AICORE operators.
+
+    Args:
+        profiling_dir (str): The directory where the parsed profiling files are
+            located.
+        device_id (str): The device ID.
+    """
+    __col_names__ = ['op_name', 'op_type', 'execution_time', 'subgraph',
+                     'full_op_name', 'op_info']
+    _file_name_aicore_detail_time = 'aicore_intermediate_{}_detail.csv'
+    _file_name_framework_info = 'framework_raw_{}.csv'
+
+    def __init__(self, profiling_dir, device_id):
+        super().__init__(profiling_dir, device_id)
+        self._none_filter_condition_key = [
+            'is_display_detail', 'is_display_full_op_name'
+        ]
+
+    def query_and_sort_by_op_type(self, filter_condition, op_type_order: list):
+        """
+        Query the AICORE operator detail information by `filter_condition`,
+        and sort by `op_type_order` and execution time.
+
+        Args:
+            filter_condition (dict): The filter condition.
+            op_type_order (list[str]): The name of the operator type in order.
+
+        Returns:
+            dict, The results are filtered and sorted.
+        """
+        if filter_condition is None:
+            filter_condition = {}
+        self._filter(filter_condition)
+
+        type_detail_cache = {}
+        for detail_info in self._result:
+            op_type = detail_info[1]
+            if op_type not in op_type_order:
+                continue
+            infos = type_detail_cache.get(op_type)
+            if infos:
+                infos.append(detail_info)
+            else:
+                type_detail_cache[op_type] = [detail_info]
+
+        result = []
+        for op_type in op_type_order:
+            detail_infos = type_detail_cache.get(op_type)
+            detail_infos.sort(key=lambda item: item[2], reverse=True)
+            result.extend(detail_infos)
+
+        return {
+            'col_name': self._display_col_names,
+            'object': result
+        }
+
+    def _load(self):
+        """Load data according to the parsed AICORE operator file."""
+        op_detail_file_path = os.path.join(
+            self._profiling_dir,
+            self._file_name_aicore_detail_time.format(self._device_id)
+        )
+        framework_file_path = os.path.join(
+            self._profiling_dir,
+            self._file_name_framework_info.format(self._device_id)
+        )
+        if not os.path.isfile(op_detail_file_path):
+            logger.warning('The file <%s> does not exist.', op_detail_file_path)
+            return
+        if not os.path.isfile(framework_file_path):
+            logger.warning('The file <%s> does not exist.', framework_file_path)
+            return
+
+        framework_infos = dict()
+        with open(framework_file_path, 'r') as file:
+            csv_reader = csv.reader(file)
+            _ = next(csv_reader)
+            for info in csv_reader:
+                framework_infos[info[3]] = self._convert_framework_field_type(
+                    info
+                )
+
+        with open(op_detail_file_path, 'r') as file:
+            csv_reader = csv.reader(file)
+            _ = next(csv_reader)
+            for info in csv_reader:
+                detail_info = self._get_op_detail_info(info, framework_infos)
+                self._data.append(detail_info)
+
+        del framework_infos
+
+    def _filter(self, filter_condition):
+        """
+        Filter the profiling data according to the filter condition.
+
+        Args:
+            filter_condition (dict): The filter condition.
+        """
+        def _inner_filter(item: list):
+            return self._default_filter(item, filter_condition)
+
+        def _inner_map(item: list):
+            inner_item = item[0:4]
+            if is_display_full_op_name:
+                inner_item.append(item[4])
+            if is_display_detail:
+                inner_item.append(item[5])
+            return inner_item
+
+        is_display_detail = filter_condition.get('is_display_detail', True)
+        is_display_full_op_name = filter_condition.get(
+            'is_display_full_op_name', True
+        )
+        self._set_display_col_name(is_display_detail, is_display_full_op_name)
+        if is_display_detail and is_display_full_op_name:
+            self._result = list(filter(_inner_filter, self._data))
+        else:
+            self._result = list(
+                map(_inner_map, filter(_inner_filter, self._data))
+            )
+
+    def _set_display_col_name(self, is_display_detail, is_display_full_op_name):
+        """
+        Set the display column name according to the filter condition.
+
+        Args:
+            is_display_detail (bool): Whether to display the detailed operator
+                information.
+            is_display_full_op_name (bool): Whether to display the operator full
+                name.
+        """
+        self._display_col_names = self.__col_names__[0:4]
+        if is_display_full_op_name:
+            self._display_col_names.append(self.__col_names__[4])
+        if is_display_detail:
+            self._display_col_names.append(self.__col_names__[5])
+
+    def _convert_framework_field_type(self, row):
+        """
+        Convert the field type of framework file to the specific type.
+
+        Args:
+            row (list[str]): One row data from parsed data.
+
+        Returns:
+            list[Union[str, float]], the converted data.
+        """
+        return [row[3], row[4], row[5], row[6],
+                json.loads(row[7]) if row[7] else None]
+
+    def _get_op_detail_info(self, row, framework_infos):
+        """
+        Get operator detail information.
+
+        Args:
+            row (list[str]): One row data from parsed operator file.
+            framework_infos (dict): All framework information.
+
+        Returns:
+            list[Union[str, float]], the operator detail information in one row.
+        """
+        framework_info = framework_infos.get(row[0])
+        return [framework_info[1], framework_info[2], float(row[1]),
+                framework_info[3], framework_info[0], framework_info[4]]
+
+
+class AicpuAnalyser(BaseAnalyser):
+    """
+    The analyser for analyzing all the AICPU operators.
+
+    Args:
+        profiling_dir (str): The directory where the parsed profiling files are
+            located.
+        device_id (str): The device ID.
+    """
+    __col_names__ = ['serial_number', 'op_name', 'total_time', 'dispatch_time',
+                     'RunV2_start', 'compute_start', 'memcpy_start',
+                     'memcpy_end', 'RunV2_end']
+    _file_name_aicpu_time = 'aicpu_intermediate_{}.csv'
+
+    def _load(self):
+        """Load data according to the parsed AICPU operator file."""
+        aicpu_file_path = os.path.join(
+            self._profiling_dir,
+            self._file_name_aicpu_time.format(self._device_id)
+        )
+        if not os.path.isfile(aicpu_file_path):
+            logger.warning('The file <%s> does not exist.', aicpu_file_path)
+            return
+
+        with open(aicpu_file_path, 'r') as file:
+            csv_reader = csv.reader(file)
+            _ = next(csv_reader)
+            for info in csv_reader:
+                aicpu_info = self._convert_field_type(info)
+                self._data.append(aicpu_info)
+
+    def _filter(self, filter_condition):
+        """
+        Filter the profiling data according to the filter condition.
+
+        Args:
+            filter_condition (dict): The filter condition.
+        """
+        def _inner_filter(item: list):
+            return self._default_filter(item, filter_condition)
+        self._result = list(filter(_inner_filter, self._data))
+
+    def _convert_field_type(self, row):
+        """
+        Convert the field type to the specific type.
+
+        Args:
+            row (list[str]): One row data from parsed data.
+
+        Returns:
+            list[Union[str, float]], the converted data.
+        """
+        return [int(row[0]), row[1], float(row[2]), float(row[3]), int(row[4]),
+                int(row[5]), int(row[6]), int(row[7]), int(row[8])]

mindinsight/profiler/analyser/analyser_factory.py

@@ -0,0 +1,76 @@
+# Copyright 2020 Huawei Technologies Co., Ltd
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+# ============================================================================
+"""The analyser factory."""
+import threading
+
+import mindinsight.profiler.analyser.analyser as analyser_module
+from mindinsight.profiler.common.exceptions.exceptions import \
+    ProfilerAnalyserNotExistException
+
+
+class AnalyserFactory:
+    """
+    The analyser factory is used to create analyser special instance.
+
+    Currently the factory supports creating `AicoreTypeAnalyser`,
+    `AicoreDetailAnalyser` and `AicpuAnalyser`. The `AicoreTypeAnalyser` is used
+    to analyze execution time according to AICORE operator type.
+    The `AicoreDetailAnalyser` is used to analyze execution time according to
+    all specific AICORE operator. The `AicpuAnalyser` is used to analyze
+    execution time according to all specific AICPU operator.
+
+    Examples:
+        >>> analyser = AnalyserFactory.instance().get_analyser(
+        >>>     'aicore_type', '/path/to/profiling/dir', '0'
+        >>> )
+    """
+    _lock = threading.Lock()
+    _instance = None
+
+    def __new__(cls, *args, **kwargs):
+        if cls._instance is None:
+            with cls._lock:
+                if cls._instance is None:
+                    cls._instance = super().__new__(cls, *args, **kwargs)
+        return cls._instance
+
+    @classmethod
+    def instance(cls):
+        """The factory instance."""
+        if cls._instance is None:
+            cls._instance = cls()
+        return cls._instance
+
+    def get_analyser(self, analyser_type, *args):
+        """
+        Get the specified analyser according to the analyser type.
+
+        Args:
+            analyser_type (str): The analyser type.
+            args (list): The parameters required for the specific analyser class.
+
+        Returns:
+            BaseAnalyser, the specified analyser instance.
+
+        Raises:
+            ProfilerAnalyserNotExistException: If the analyser type does not exist.
+        """
+        subnames = analyser_type.split('_')
+        analyser_class_name = ''.join([name.capitalize() for name in subnames])
+        analyser_class_name += 'Analyser'
+
+        if not hasattr(analyser_module, analyser_class_name):
+            raise ProfilerAnalyserNotExistException(analyser_type)
+        return getattr(analyser_module, analyser_class_name)(*args)

mindinsight/profiler/analyser/base_analyser.py

@@ -0,0 +1,240 @@
+# Copyright 2020 Huawei Technologies Co., Ltd
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+# ============================================================================
+"""The base analyser."""
+import functools
+from abc import ABC, abstractmethod
+
+from marshmallow import ValidationError
+
+from mindinsight.profiler.common.exceptions.exceptions import \
+    ProfilerColumnNotExistException, ProfilerPathErrorException, \
+    ProfilerIOException
+from mindinsight.profiler.common.log import logger
+from mindinsight.profiler.common.validator.validate_path import \
+    validate_and_normalize_path
+
+
+class BaseAnalyser(ABC):
+    """
+    The base analyser.
+
+    A concrete analyser class can be constructed by inheriting the class. The
+    analyser provides the ability to filter, sort and group. The subclass only
+    need to implement `_load`, `_filter`, `_sort` and `_group`. The condition
+    defines the rules for filtering, sorting and grouping.
+
+    Args:
+        profiling_dir (str): The directory where the parsed profiling files
+            are located.
+        device_id (str): The device ID.
+    """
+    __col_names__ = []
+
+    def __init__(self, profiling_dir, device_id):
+        self._profiling_dir = self._normalize_profiling_dir(profiling_dir)
+        self._device_id = device_id
+        self._data = []
+        self._result = None
+        self._display_col_names = None
+        self._size = 0
+        self._none_filter_condition_key = []
+
+        try:
+            self._load()
+        except IOError as err:
+            logger.exception(err)
+            raise ProfilerIOException()
+
+    @property
+    def col_names(self):
+        """The column names in the parsed profiling file."""
+        return self.__col_names__
+
+    @property
+    def data(self):
+        """The data in the parsed profiling file."""
+        return self._data
+
+    def query(self, condition=None):
+        """
+        Query data according to the condition.
+
+        Args:
+            condition (dict): The search condition, including filter condition,
+                sort condition and group condition. If the condition is `None`,
+                all data will be returned. Default: None.
+
+        Returns:
+            dict, the result after filtered, sorted and grouped.
+        """
+        if condition is None:
+            condition = {}
+        filter_condition = condition.get('filter_condition', {})
+        sort_condition = condition.get('sort_condition')
+        group_condition = condition.get('group_condition')
+
+        self._result = []
+        self._display_col_names = self.__col_names__[:]
+        self._filter(filter_condition)
+        self._size = len(self._result)
+        if sort_condition:
+            self._sort(sort_condition)
+        if group_condition:
+            self._group(group_condition)
+        return {
+            'col_name': self._display_col_names,
+            'object': self._result,
+            'size': self._size
+        }
+
+    @abstractmethod
+    def _load(self):
+        """Load data according to the parsed profiling files."""
+
+    @abstractmethod
+    def _filter(self, filter_condition):
+        """
+        Filter the profiling data according to the filter condition.
+
+        Args:
+            filter_condition (dict): The filter condition.
+        """
+
+    def _sort(self, sort_condition: dict):
+        """
+        Sort the profiling data according to the filter condition.
+
+        Args:
+            sort_condition (dict): The sort condition.
+
+        Raises:
+            ProfilerColumnNotExistException: If the sort name does not exist.
+        """
+        def _cmp(item1, item2):
+            value1 = item1[index]
+            value2 = item2[index]
+
+            if value1 is None and value2 is None:
+                cmp_result = 0
+            elif value1 is None:
+                cmp_result = -1
+            elif value2 is None:
+                cmp_result = 1
+            else:
+                try:
+                    cmp_result = (value1 > value2) - (value1 < value2)
+                except TypeError:
+                    type1 = type(value1).__name__
+                    type2 = type(value2).__name__
+                    cmp_result = (type1 > type2) - (type1 < type2)
+            return cmp_result
+
+        sort_name = sort_condition.get('name')
+        sort_type = sort_condition.get('type', 'descending')
+        reverse = sort_type == 'descending'
+        if not sort_name:
+            return
+        try:
+            index = self.__col_names__.index(sort_name)
+        except ValueError:
+            raise ProfilerColumnNotExistException(sort_name)
+        self._result.sort(key=functools.cmp_to_key(_cmp), reverse=reverse)
+
+    def _group(self, group_condition: dict):
+        """
+        Group the profiling data according to the group condition.
+
+        Args:
+            group_condition (dict): The group condition.
+        """
+        limit = group_condition.get('limit')
+        offset = group_condition.get('offset')
+        if limit is None and offset is None:
+            return
+        if limit is None:
+            limit = 10
+        if offset is None:
+            offset = 0
+        self._result = self._result[limit * offset: limit * (offset + 1)]
+
+    def _default_filter(self, item, condition):
+        """
+        The default filter method.
+
+        Args:
+            item (list[Union[str, float, int]]): A piece of data to be filtered.
+            condition (dict): The filter condition.
+
+        Returns:
+            bool, `True` if the item is satisfied.
+        """
+        for condition_key, condition_value in condition.items():
+            if condition_key in self._none_filter_condition_key:
+                continue
+            if condition_key in self.__col_names__:
+                index = self.__col_names__.index(condition_key)
+                actual_value = item[index]
+                for exp_key, exp_value in condition_value.items():
+                    if not self._is_match_condition(
+                            exp_key, exp_value, actual_value):
+                        return False
+        return True
+
+    def _is_match_condition(self, exp_key, exp_value, actual_value):
+        """
+        Check whether the actual value meets the expect condition.
+
+        Args:
+            exp_key (str): Expect key of the condition.
+            exp_value (str): Expect value.
+            actual_value (str): Actual value.
+
+        Returns:
+            bool, `True` if the actual meets the expect condition, else `False`.
+        """
+        if exp_key == 'in':
+            if actual_value not in exp_value:
+                return False
+        elif exp_key == 'not_in':
+            if actual_value in exp_value:
+                return False
+        elif exp_key == 'partial_match_str_in':
+            for partial_match_str in exp_value:
+                if partial_match_str in actual_value:
+                    return True
+            return False
+        else:
+            return False
+
+        return True
+
+    def _normalize_profiling_dir(self, profiling_dir):
+        """
+        Normalize the profiling dir.
+
+        Args:
+            profiling_dir (str): The directory where the parsed profiling files
+                are located.
+
+        Returns:
+            str, the normalized profiling dir.
+        """
+        try:
+            normalized_profiling_dir = validate_and_normalize_path(
+                profiling_dir, 'profiler'
+            )
+        except ValidationError:
+            raise ProfilerPathErrorException('The profiling dir is invalid.')
+        return normalized_profiling_dir

mindinsight/profiler/analyser/integrator.py

@@ -0,0 +1,157 @@
+# Copyright 2020 Huawei Technologies Co., Ltd
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+# ============================================================================
+"""The integrator for integrating parsed profiling files."""
+import csv
+import os
+from decimal import Decimal
+
+
+class Integrator:
+    """
+    The integrator for integrating parsed profiling files.
+
+    Args:
+        profiling_dir (str): The directory where the parsed profiling files are
+            located.
+        device_id (str): The device ID.
+    """
+    _file_name_aicore_detail_time = 'output_op_compute_time_{}.txt'
+    _file_name_aicpu_time = 'output_data_preprocess_aicpu_{}.txt'
+    _file_name_framework = 'framework_raw_{}.csv'
+    _header_aicore_type = ['op_type', 'execution_time', 'execution_frequency',
+                           'percent']
+    _header_aicore_detail = ['full_op_name', 'execution_time']
+    _header_aicpu = ['serial_number', 'op_name', 'total_time', 'dispatch_time',
+                     'RunV2_start', 'compute_start', 'memcpy_start',
+                     'memcpy_end', 'RunV2_end']
+
+    def __init__(self, profiling_dir, device_id):
+        self._profiling_dir = profiling_dir
+        self._device_id = device_id
+        self._op_time_cache = {}
+        self._total_time = Decimal('0.0')
+
+    def integrate(self):
+        """Integrate the parsed profiling files."""
+        self._parse_aicore_detail_time()
+        self._parse_aicore_type_time()
+        self._parse_aicpu_time()
+
+    def _parse_aicore_type_time(self):
+        """Parse the parsed AICORE operator type file."""
+        framework_file = os.path.join(
+            self._profiling_dir,
+            self._file_name_framework.format(self._device_id)
+        )
+        if not os.path.isfile(framework_file):
+            return
+
+        op_name_type_cache = {}
+        with open(framework_file, 'r') as src_file:
+            csv_reader = csv.reader(src_file)
+            _ = next(csv_reader)
+
+            for row in csv_reader:
+                op_name_type_cache[row[3]] = row[5]
+
+        op_type_time_cache = {}
+        for full_op_name, op_time in self._op_time_cache.items():
+            op_type = op_name_type_cache.get(full_op_name)
+            if op_type_time_cache.get(op_type) is None:
+                op_type_time_cache[op_type] = [op_time, 1]
+            else:
+                op_type_time_cache[op_type][0] += op_time
+                op_type_time_cache[op_type][1] += 1
+
+        op_type_file_name = 'aicore_intermediate_' + self._device_id + '_type.csv'
+        op_type_file_path = os.path.join(self._profiling_dir, op_type_file_name)
+        with open(op_type_file_path, 'w') as type_file:
+            csv_writer = csv.writer(type_file)
+            csv_writer.writerow(self._header_aicore_type)
+
+            for op_type, op_type_time_info in op_type_time_cache.items():
+                type_info = [
+                    op_type, op_type_time_info[0], op_type_time_info[1],
+                    round((op_type_time_info[0] / self._total_time) * 100, 2)
+                ]
+                csv_writer.writerow(type_info)
+
+    def _parse_aicore_detail_time(self):
+        """Parse the parsed AICORE operator time file."""
+        aicore_detail_file = os.path.join(
+            self._profiling_dir,
+            self._file_name_aicore_detail_time.format(self._device_id)
+        )
+        if not os.path.isfile(aicore_detail_file):
+            return
+
+        op_detail_file_name = 'aicore_intermediate_' + self._device_id + '_detail.csv'
+        op_detail_file_path = os.path.join(
+            self._profiling_dir, op_detail_file_name
+        )
+        with open(aicore_detail_file, 'r') as src_file:
+            row = src_file.readline()
+            if row.startswith('op_name'):
+                _ = src_file.readline()
+            elif row.startswith('====='):
+                _ = src_file.readline()
+                _ = src_file.readline()
+            else:
+                return
+
+            with open(op_detail_file_path, 'w') as detail_file:
+                csv_writer = csv.writer(detail_file)
+                csv_writer.writerow(self._header_aicore_detail)
+
+                while True:
+                    row = src_file.readline()
+                    if not row:
+                        break
+
+                    op_infos = row.split()
+                    if op_infos[0] == 'total':
+                        self._total_time = Decimal(op_infos[2])
+                        continue
+                    self._op_time_cache[op_infos[0]] = Decimal(op_infos[1])
+                    csv_writer.writerow([op_infos[0], op_infos[1]])
+
+    def _parse_aicpu_time(self):
+        """Parse the parsed AICPU operator time file."""
+        aicpu_file = os.path.join(
+            self._profiling_dir,
+            self._file_name_aicpu_time.format(self._device_id)
+        )
+        if not os.path.isfile(aicpu_file):
+            return
+
+        save_file_name = 'aicpu_intermediate_' + self._device_id + '.csv'
+        save_file_path = os.path.join(self._profiling_dir, save_file_name)
+        with open(aicpu_file, 'r') as src_file:
+            row = src_file.readline()
+            if not row.startswith('serial_number'):
+                return
+            _ = src_file.readline()
+            with open(save_file_path, 'w') as save_file:
+                csv_writer = csv.writer(save_file)
+                csv_writer.writerow(self._header_aicpu)
+
+                while True:
+                    row = src_file.readline()
+                    if not row:
+                        break
+                    infos = row.split()
+                    if infos[0] == 'AI':
+                        continue
+                    csv_writer.writerow(infos)

mindinsight/profiler/common/__init__.py

@@ -0,0 +1,14 @@
+# Copyright 2020 Huawei Technologies Co., Ltd
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+# ============================================================================

mindinsight/profiler/common/exceptions/__init__.py

@@ -0,0 +1,14 @@
+# Copyright 2020 Huawei Technologies Co., Ltd
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+# ============================================================================

mindinsight/profiler/common/exceptions/error_code.py

@@ -0,0 +1,63 @@
+# Copyright 2020 Huawei Technologies Co., Ltd
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+# ============================================================================
+"""Profiler error code and messages."""
+from enum import unique, Enum
+
+from mindinsight.utils.constant import ProfilerMgrErrors
+
+
+_GENERAL_MASK = 0b00001 << 7
+_PARSER_MASK = 0b00010 << 7
+_ANALYSER_MASK = 0b00011 << 7
+
+
+@unique
+class ProfilerErrors(ProfilerMgrErrors):
+    """Profiler error codes."""
+    # general error code
+    PARAM_VALUE_ERROR = 0 | _GENERAL_MASK
+    PATH_ERROR = 1 | _GENERAL_MASK
+    PARAM_TYPE_ERROR = 2 | _GENERAL_MASK
+    DIR_NOT_FOUND_ERROR = 3 | _GENERAL_MASK
+    FILE_NOT_FOUND_ERROR = 4 | _GENERAL_MASK
+    IO_ERROR = 5 | _GENERAL_MASK
+
+    # parser error code
+    DEVICE_ID_MISMATCH_ERROR = 0 | _PARSER_MASK
+    RAW_FILE_ERROR = 1 | _PARSER_MASK
+
+    # analyser error code
+    COLUMN_NOT_EXIST_ERROR = 0 | _ANALYSER_MASK
+    ANALYSER_NOT_EXIST_ERROR = 1 | _ANALYSER_MASK
+
+
+@unique
+class ProfilerErrorMsg(Enum):
+    """Profiler error messages."""
+    # general error msg
+    PARAM_VALUE_ERROR = 'Param value error. {}'
+    PATH_ERROR = 'Path error. {}'
+    PARAM_TYPE_ERROR = 'Param type error. {}'
+    DIR_NOT_FOUND_ERROR = 'The dir <{}> not found.'
+    FILE_NOT_FOUND_ERROR = 'The file <{}> not found.'
+    IO_ERROR = 'Read or write file fail.'
+
+    # parser error msg
+    DEVICE_ID_MISMATCH_ERROR = 'The device ID mismatch.'
+    RAW_FILE_ERROR = 'Raw file error. {}'
+
+    # analyser error msg
+    COLUMN_NOT_EXIST_ERROR = 'The column {} does not exist.'
+    ANALYSER_NOT_EXIST_ERROR = 'The analyser {} does not exist.'

mindinsight/profiler/common/exceptions/exceptions.py

@@ -0,0 +1,128 @@
+# Copyright 2020 Huawei Technologies Co., Ltd
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+# ============================================================================
+"""Definition of error code and relative messages in profiler module."""
+from mindinsight.profiler.common.exceptions.error_code import ProfilerErrors, \
+    ProfilerErrorMsg
+from mindinsight.utils.exceptions import MindInsightException
+
+
+class ProfilerParamValueErrorException(MindInsightException):
+    """The parameter value error in profiler module."""
+
+    def __init__(self, msg):
+        super(ProfilerParamValueErrorException, self).__init__(
+            error=ProfilerErrors.PARAM_VALUE_ERROR,
+            message=ProfilerErrorMsg.PARAM_VALUE_ERROR.value.format(msg),
+            http_code=400
+        )
+
+
+class ProfilerPathErrorException(MindInsightException):
+    """The path error in profiler module."""
+
+    def __init__(self, msg):
+        super(ProfilerPathErrorException, self).__init__(
+            error=ProfilerErrors.PATH_ERROR,
+            message=ProfilerErrorMsg.PATH_ERROR.value.format(msg),
+            http_code=400
+        )
+
+
+class ProfilerParamTypeErrorException(MindInsightException):
+    """The parameter type error in profiler module."""
+
+    def __init__(self, msg):
+        super(ProfilerParamTypeErrorException, self).__init__(
+            error=ProfilerErrors.PARAM_TYPE_ERROR,
+            message=ProfilerErrorMsg.PARAM_TYPE_ERROR.value.format(msg),
+            http_code=400
+        )
+
+
+class ProfilerDirNotFoundException(MindInsightException):
+    """The dir not found exception in profiler module."""
+
+    def __init__(self, msg):
+        super(ProfilerDirNotFoundException, self).__init__(
+            error=ProfilerErrors.DIR_NOT_FOUND_ERROR,
+            message=ProfilerErrorMsg.DIR_NOT_FOUND_ERROR.value.format(msg),
+            http_code=400
+        )
+
+
+class ProfilerFileNotFoundException(MindInsightException):
+    """The file not found exception in profiler module."""
+
+    def __init__(self, msg):
+        super(ProfilerFileNotFoundException, self).__init__(
+            error=ProfilerErrors.FILE_NOT_FOUND_ERROR,
+            message=ProfilerErrorMsg.FILE_NOT_FOUND_ERROR.value.format(msg),
+            http_code=400
+        )
+
+
+class ProfilerIOException(MindInsightException):
+    """The IO exception in profiler module."""
+
+    def __init__(self):
+        super(ProfilerIOException, self).__init__(
+            error=ProfilerErrors.IO_ERROR,
+            message=ProfilerErrorMsg.IO_ERROR.value,
+            http_code=400
+        )
+
+
+class ProfilerDeviceIdMismatchException(MindInsightException):
+    """The device id mismatch exception in profiler module."""
+
+    def __init__(self):
+        super(ProfilerDeviceIdMismatchException, self).__init__(
+            error=ProfilerErrors.DEVICE_ID_MISMATCH_ERROR,
+            message=ProfilerErrorMsg.DEVICE_ID_MISMATCH_ERROR.value,
+            http_code=400
+        )
+
+
+class ProfilerRawFileException(MindInsightException):
+    """The raw file exception in profiler module."""
+
+    def __init__(self, msg):
+        super(ProfilerRawFileException, self).__init__(
+            error=ProfilerErrors.RAW_FILE_ERROR,
+            message=ProfilerErrorMsg.RAW_FILE_ERROR.value.format(msg),
+            http_code=400
+        )
+
+
+class ProfilerColumnNotExistException(MindInsightException):
+    """The column does not exist exception in profiler module."""
+
+    def __init__(self, msg):
+        super(ProfilerColumnNotExistException, self).__init__(
+            error=ProfilerErrors.COLUMN_NOT_EXIST_ERROR,
+            message=ProfilerErrorMsg.COLUMN_NOT_EXIST_ERROR.value.format(msg),
+            http_code=400
+        )
+
+
+class ProfilerAnalyserNotExistException(MindInsightException):
+    """The analyser in profiler module."""
+
+    def __init__(self, msg):
+        super(ProfilerAnalyserNotExistException, self).__init__(
+            error=ProfilerErrors.ANALYSER_NOT_EXIST_ERROR,
+            message=ProfilerErrorMsg.ANALYSER_NOT_EXIST_ERROR.value.format(msg),
+            http_code=400
+        )

mindinsight/profiler/common/log.py

@@ -0,0 +1,20 @@
+# Copyright 2020 Huawei Technologies Co., Ltd
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+# ============================================================================
+"""Profiler logger."""
+from mindinsight.utils.log import setup_logger
+
+LOG_NAME = "profiler"
+LOG_MODULE = "profiler"
+logger = setup_logger(sub_module=LOG_MODULE, log_name=LOG_NAME)

mindinsight/profiler/common/validator/__init__.py

@@ -0,0 +1,14 @@
+# Copyright 2020 Huawei Technologies Co., Ltd
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+# ============================================================================

mindinsight/profiler/common/validator/validate_path.py

@@ -0,0 +1,119 @@
+# Copyright 2019 Huawei Technologies Co., Ltd
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+# ============================================================================
+"""Validate the input path."""
+import os
+from typing import Union, List
+from marshmallow import ValidationError
+
+
+def safe_normalize_path(
+        path,
+        raise_key,
+        safe_prefixes: Union[None, List[str]],
+        check_absolute_path=False,
+        allow_parent_dir=False,
+):
+    """
+    Returns safe normalized path.
+
+    This func validates given path, and returns its normalized form. If
+    safe_prefixes is given, this func will check whether the path is safe.
+
+    Note:
+        This func is not compatible with windows.
+
+        Caller should check returned path to ensure safety according to
+        business logic.
+
+        File scheme (rfc8089) is currently not supported.
+
+    Args:
+        path (str): Path to be normalized.
+
+        raise_key (str): The exception raise key
+
+        safe_prefixes (list[str]): If not none, path must startswith one of the
+            safe_prefixes. Set this arg to [] will cause all paths considered
+            unsafe. Normally, prefix in this arg should end with "/".
+
+        check_absolute_path (bool): Whether check path is absolute.
+
+        allow_parent_dir (bool): Whether allow parent dir in path.
+
+    Returns:
+        str, normalized path.
+    """
+    normalized_path = validate_and_normalize_path(
+        path,
+        raise_key=raise_key,
+        check_absolute_path=check_absolute_path,
+        allow_parent_dir=allow_parent_dir,
+    )
+
+    if safe_prefixes is None:
+        return normalized_path
+
+    normalized_str = str(normalized_path)
+    for prefix in safe_prefixes:
+        if normalized_str.startswith(prefix):
+            return normalized_path
+
+    raise ValidationError({raise_key: {"The path is invalid!"}})
+
+
+def validate_and_normalize_path(
+        path,
+        raise_key,
+        check_absolute_path=False,
+        allow_parent_dir=False,
+):
+    """
+    Validates path and returns its normalized form.
+
+    If path has a valid scheme, treat path as url, otherwise consider path a
+    unix local path.
+
+    Note:
+        File scheme (rfc8089) is currently not supported.
+
+    Args:
+        path (str): Path to be normalized.
+        raise_key (str): The exception raise key.
+        check_absolute_path (bool): Whether check path scheme is supported.
+        allow_parent_dir (bool): Whether allow parent dir in path.
+
+    Returns:
+        str, normalized path.
+    """
+    if not path:
+        raise ValidationError({raise_key: {"The path is invalid!"}})
+
+    path_str = str(path)
+    if not allow_parent_dir:
+        path_components = path_str.split("/")
+        if ".." in path_components:
+            raise ValidationError({raise_key: {"The path is invalid!"}})
+
+    # path does not have valid schema, treat it as unix local path.
+    if check_absolute_path:
+        if not path_str.startswith("/"):
+            raise ValidationError({raise_key: {"The path is invalid!"}})
+    try:
+        # most unix systems allow
+        normalized_path = os.path.realpath(path)
+    except ValueError:
+        raise ValidationError({raise_key: {"The path is invalid!"}})
+
+    return normalized_path

mindinsight/profiler/parser/__init__.py

@@ -0,0 +1,14 @@
+# Copyright 2020 Huawei Technologies Co., Ltd
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+# ============================================================================

mindinsight/profiler/parser/framework_parser.py

@@ -0,0 +1,520 @@
+# Copyright 2020 Huawei Technologies Co., Ltd
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+# ============================================================================
+"""Thr parser for parsing framework files."""
+import csv
+import enum
+import json
+import os
+import re
+
+from marshmallow import ValidationError
+
+from mindinsight.profiler.common.exceptions.exceptions import \
+    ProfilerPathErrorException, ProfilerDirNotFoundException, \
+    ProfilerFileNotFoundException, ProfilerDeviceIdMismatchException, \
+    ProfilerRawFileException
+from mindinsight.profiler.common.validator.validate_path import \
+    validate_and_normalize_path
+
+
+class VmDataType(enum.IntEnum):
+    """Definition of vm data type."""
+    NUMBER_TYPE_BEGIN = 26
+    NUMBER_TYPE_BOOL = 27
+    NUMBER_TYPE_INT = 28
+    NUMBER_TYPE_INT8 = 29
+    NUMBER_TYPE_INT16 = 30
+    NUMBER_TYPE_INT32 = 31
+    NUMBER_TYPE_INT64 = 32
+    NUMBER_TYPE_UINT = 33
+    NUMBER_TYPE_UINT8 = 34
+    NUMBER_TYPE_UINT16 = 35
+    NUMBER_TYPE_UINT32 = 36
+    NUMBER_TYPE_UINT64 = 37
+    NUMBER_TYPE_FLOAT = 38
+    NUMBER_TYPE_FLOAT16 = 39
+    NUMBER_TYPE_FLOAT32 = 40
+    NUMBER_TYPE_FLOAT64 = 41
+    NUMBER_TYPE_END = 42
+
+    @classmethod
+    def get_data_type_name(cls, num):
+        """
+        Get the name of data type by enum number.
+
+        Args:
+            num (int): Enum number.
+
+        Returns:
+            str, the name of data type.
+        """
+        data_type = cls._value2member_map_.get(num)
+        return 'UNKNOWN' if data_type is None else data_type.name
+
+
+class GeDataType(enum.IntEnum):
+    """Definition of ge data type."""
+    DT_FLOAT = 0
+    DT_FLOAT16 = 1
+    DT_INT8 = 2
+    DT_INT16 = 6
+    DT_UINT16 = 7
+    DT_UINT8 = 4
+    DT_INT32 = 3
+    DT_INT64 = 9
+    DT_UINT32 = 8
+    DT_UINT64 = 10
+    DT_BOOL = 12
+    DT_DOUBLE = 11
+    DT_STRING = 13
+    DT_DUAL_SUB_INT8 = 14
+    DT_DUAL_SUB_UINT8 = 15
+    DT_COMPLEX64 = 16
+    DT_COMPLEX128 = 17
+    DT_QINT8 = 18
+    DT_QINT16 = 19
+    DT_QINT32 = 20
+    DT_QUINT8 = 21
+    DT_QUINT16 = 22
+    DT_RESOURCE = 23
+    DT_STRING_REF = 24
+    DT_DUAL = 25
+    DT_UNDEFINED = 26
+
+    @classmethod
+    def get_data_type_name(cls, num):
+        """
+        Get the name of data type by enum number.
+
+        Args:
+            num (int): Enum number.
+
+        Returns:
+            str, the name of data type.
+        """
+        data_type = cls._value2member_map_.get(num)
+        return 'UNKNOWN' if data_type is None else data_type.name
+
+
+class GeFormat(enum.IntEnum):
+    """Definition of ge format type."""
+    FORMAT_NCHW = 0
+    FORMAT_NHWC = 1
+    FORMAT_ND = 2
+    FORMAT_NC1HWC0 = 3
+    FORMAT_FRACTAL_Z = 4
+    FORMAT_NC1C0HWPAD = 5
+    FORMAT_NHWC1C0 = 6
+    FORMAT_FSR_NCHW = 7
+    FORMAT_FRACTAL_DECONV = 8
+    FORMAT_C1HWNC0 = 9
+    FORMAT_FRACTAL_DECONV_TRANSPOSE = 10
+    FORMAT_FRACTAL_DECONV_SP_STRIDE_TRANS = 11
+    FORMAT_NC1HWC0_C04 = 12
+    FORMAT_FRACTAL_Z_C04 = 13
+    FORMAT_CHWN = 14
+    FORMAT_FRACTAL_DECONV_SP_STRIDE8_TRANS = 15
+    FORMAT_HWCN = 16
+    FORMAT_NC1KHKWHWC0 = 17
+    FORMAT_BN_WEIGHT = 18
+    FORMAT_FILTER_HWCK = 19
+    FORMAT_HASHTABLE_LOOKUP_LOOKUPS = 20
+    FORMAT_HASHTABLE_LOOKUP_KEYS = 21
+    FORMAT_HASHTABLE_LOOKUP_VALUE = 22
+    FORMAT_HASHTABLE_LOOKUP_OUTPUT = 23
+    FORMAT_HASHTABLE_LOOKUP_HITS = 24
+    FORMAT_C1HWNCOC0 = 25
+    FORMAT_MD = 26
+    FORMAT_NDHWC = 27
+    FORMAT_FRACTAL_ZZ = 28
+    FORMAT_FRACTAL_NZ = 29
+    FORMAT_NCDHW = 30
+    FORMAT_DHWCN = 31
+    FORMAT_NDC1HWC0 = 32
+    FORMAT_FRACTAL_Z_3D = 33
+    FORMAT_CN = 34
+    FORMAT_NC = 35
+    FORMAT_DHWNC = 36
+    FORMAT_FRACTAL_Z_3D_TRANSPOSE = 37
+    FORMAT_RESERVED = 38
+    FORMAT_ALL = 39
+
+    @classmethod
+    def get_format_name(cls, num):
+        """
+        Get the name of format type by enum number.
+
+        Args:
+            num (int): Enum number.
+
+        Returns:
+            str, the name of format type.
+        """
+        format_type = cls._value2member_map_.get(num)
+        return 'UNKNOWN' if format_type is None else format_type.name
+
+
+class FrameworkParser:
+    """
+    Thr parser for parsing framework files.
+
+    Args:
+        profiling_id (str): The profiling ID.
+        device_id (str): The device ID.
+        output_path (str): The directory of the parsed file. Default: `./`.
+    """
+    _raw_data_dir = '/var/log/npu/profiling'
+    _regex_framework = r'Framework\.host\.(?P<data_type>.+)\.(?P<device_id>\d).+'
+    _regex_framework_in_data = r'Framework\.host\.(?P<data_type>.+)\.' \
+                               r'(?P<device_id>\d)\.(?P<profiling_id>[a-zA-Z0-9]+).+'
+    _col_names = [
+        'task_id', 'stream_id', 'block_dim', 'full_op_name', 'op_name',
+        'op_type', 'subgraph', 'op_info'
+    ]
+    _graph_attr_name = [
+        'input_format', 'input_data_type', 'input_shape', 'output_format',
+        'output_data_type', 'output_shape'
+    ]
+
+    def __init__(self, profiling_id, device_id, output_path='./'):
+        self._profiling_path = self._get_raw_profiling_path(profiling_id)
+        self._backend_type = None
+        self._framework_path = {'graph': [], 'task': []}
+        self._search_file(profiling_id, device_id)
+        self._device_id = device_id
+        self._save_path = self._get_save_path(device_id, output_path)
+        self._task_id_full_op_name_dict = {}
+        self._task_cache = {}
+        self._parse_task_files()
+
+    @property
+    def save_path(self):
+        """
+        The property of save path.
+
+        Returns:
+            str, the save path.
+        """
+        return self._save_path
+
+    def to_task_id_full_op_name_dict(self):
+        """
+        Get the task id and full operator name dict.
+
+        Returns:
+            dict, the task id and full operator name dict.
+        """
+        return self._task_id_full_op_name_dict
+
+    def parse(self):
+        """Parse the framework files."""
+        self._parse_graph_files_and_save(self._task_cache)
+        del self._task_cache
+
+    def _get_raw_profiling_path(self, profiling_id):
+        """
+        Get raw profiling path.
+
+        Args:
+            profiling_id (str): The profiling ID.
+
+        Returns:
+            str, the raw profiling path.
+
+        Raises:
+            ProfilerPathErrorException: If the profiling path is invalid.
+            ProfilerDirNotFoundException: If the profiling dir is not found.
+        """
+        profiling_path = os.path.join(self._raw_data_dir, profiling_id)
+        try:
+            profiling_path = validate_and_normalize_path(
+                profiling_path, 'profiler'
+            )
+        except ValidationError:
+            raise ProfilerPathErrorException('Profiling path is invalid.')
+        if not os.path.isdir(profiling_path):
+            raise ProfilerDirNotFoundException(profiling_path)
+        return profiling_path
+
+    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.
+        """
+        self._search_file_from_job_path(device_id)
+        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 _search_file_from_job_path(self, device_id):
+        """
+        Search framework files from job path.
+
+        Args:
+            device_id (str): The device ID.
+
+        Raises:
+            ProfilerRawFileException: If the framework file type is inconsistent.
+            ProfilerDeviceIdMismatchException: If the device id is mismatch
+                with framework in the raw dir.
+        """
+        files = os.listdir(self._profiling_path)
+        for file in files:
+            pattern = re.search(self._regex_framework, file)
+            if not pattern or file.endswith('.done'):
+                continue
+            attrs = pattern.groupdict()
+
+            device_id_in_path = attrs.get('device_id')
+            if device_id_in_path != device_id:
+                raise ProfilerDeviceIdMismatchException()
+
+            data_type = attrs.get('data_type')
+            if data_type.startswith('vm.'):
+                if self._backend_type and self._backend_type != 'vm':
+                    raise ProfilerRawFileException('Backend type is inconsistent.')
+                self._backend_type = 'vm'
+                data_type = data_type.split('.')[1]
+            else:
+                if self._backend_type and self._backend_type != 'ge':
+                    raise ProfilerRawFileException('Backend type is inconsistent.')
+                self._backend_type = 'ge'
+            if data_type.startswith('graph_desc_info'):
+                self._framework_path['graph'].append(
+                    os.path.join(self._profiling_path, file)
+                )
+            elif data_type.startswith('task_desc_info'):
+                self._framework_path['task'].append(
+                    os.path.join(self._profiling_path, file)
+                )
+
+    def _search_file_from_data_path(self, profiling_id, device_id):
+        """
+        Search framework files from data path.
+
+        Args:
+            profiling_id (str): The profiling ID.
+            device_id (str): The device ID.
+
+        Raises:
+            ProfilerRawFileException: If the framework file type is inconsistent.
+            ProfilerDeviceIdMismatchException: If the device id is mismatch
+                with framework in the raw dir.
+        """
+        profiling_data_path = os.path.join(
+            self._raw_data_dir, 'container', device_id, 'data'
+        )
+        if not os.path.isdir(profiling_data_path):
+            return
+
+        files = os.listdir(profiling_data_path)
+        for file in files:
+            pattern = re.search(self._regex_framework_in_data, file)
+            if not pattern or file.endswith('.done') or file.endswith('.zip'):
+                continue
+            attrs = pattern.groupdict()
+
+            profiling_id_in_path = attrs.get('profiling_id')
+            if profiling_id_in_path != profiling_id:
+                continue
+
+            device_id_in_path = attrs.get('device_id')
+            if device_id_in_path != device_id:
+                raise ProfilerDeviceIdMismatchException()
+
+            data_type = attrs.get('data_type')
+            if data_type.startswith('vm.'):
+                if self._backend_type and self._backend_type != 'vm':
+                    raise ProfilerRawFileException('Backend type is inconsistent.')
+                self._backend_type = 'vm'
+                data_type = data_type.split('.')[1]
+            else:
+                if self._backend_type and self._backend_type != 'ge':
+                    raise ProfilerRawFileException('Backend type is inconsistent.')
+                self._backend_type = 'ge'
+            if data_type.startswith('graph_desc_info'):
+                self._framework_path['graph'].append(
+                    os.path.join(profiling_data_path, file)
+                )
+            elif data_type.startswith('task_desc_info'):
+                self._framework_path['task'].append(
+                    os.path.join(profiling_data_path, file)
+                )
+
+    def _get_save_path(self, device_id, output_path):
+        """
+        Get the save path.
+
+        Args:
+            device_id (str): The device ID.
+            output_path (str): The output dir.
+
+        Returns:
+            str, the save path.
+
+        Raises:
+            ProfilerPathErrorException: If the output path is invalid.
+            ProfilerDirNotFoundException: If the output dir is not found.
+        """
+        try:
+            output_dir = validate_and_normalize_path(output_path, 'profiler')
+        except ValidationError:
+            raise ProfilerPathErrorException('Output path is invalid.')
+        if not os.path.isdir(output_dir):
+            raise ProfilerDirNotFoundException(output_dir)
+        return os.path.join(
+            output_dir, '_'.join(['framework', 'raw', device_id]) + '.csv'
+        )
+
+    def _parse_task_files(self):
+        """Parse the framework task files."""
+        for path in self._framework_path['task']:
+            with open(path, 'r') as file:
+                for task_info in file:
+                    infos = task_info.strip('\n').split(' ')
+                    # key is op name, values is task id, stream id, block_dim
+                    self._task_cache[infos[0]] = [infos[2], infos[3], infos[1]]
+                    self._task_id_full_op_name_dict[infos[2]] = infos[0]
+
+    def _parse_graph_files_and_save(self, task_cache):
+        """
+        Parse the framework graph files and save the framework information.
+
+        Args:
+            task_cache (dict): The task information cache.
+        """
+        with open(self._save_path, 'w') as save_file:
+            csv_writer = csv.writer(save_file)
+            csv_writer.writerow(self._col_names)
+            for path in self._framework_path['graph']:
+                with open(path, 'r') as graph_file:
+                    for graph_info in graph_file:
+                        result = self._parse_one_row_graph_info(graph_info)
+                        task_info = task_cache.get(result[0])
+                        task_info.extend(result)
+                        csv_writer.writerow(task_info)
+                        del task_cache[result[0]]
+
+            none_list = [None, None, None, None]
+            for key, value in task_cache.items():
+                value.append(key)
+                value.extend(none_list)
+                csv_writer.writerow(value)
+
+    def _parse_one_row_graph_info(self, row_info):
+        """
+        Parse the graph information in one row.
+
+        Args:
+            row_info (str): One row graph information.
+
+        Returns:
+            list[str], the parsed graph information.
+        """
+        full_op_name = None
+        op_name = None
+        subgraph_name = None
+        op_type = None
+        op_info = dict()
+        cur_op_info_key = None
+
+        infos = row_info.strip('\n').split(' ')
+        for info in infos:
+            attr_name, attr_value = info.split(':', 1)
+            if attr_name == 'op_name':
+                full_op_name = attr_value
+                subgraph_name = self._get_subgraph_name(full_op_name)
+                op_name = self._get_op_name(full_op_name, subgraph_name)
+            elif attr_name == 'op_type':
+                op_type = attr_value
+            elif attr_name in ['input_id', 'output_id']:
+                cur_op_info_key = '{}_{}'.format(
+                    attr_name.split('_')[0], attr_value
+                )
+                op_info[cur_op_info_key] = dict()
+            elif attr_name in self._graph_attr_name:
+                op_attr = attr_name.split('_', 1)[1]
+                if op_attr == 'shape':
+                    attr_value = attr_value.strip('"')
+                if self._backend_type == 'vm':
+                    if op_attr == 'data_type':
+                        attr_value = VmDataType.get_data_type_name(
+                            int(attr_value)
+                        )
+                else:
+                    if op_attr == 'data_type':
+                        attr_value = GeDataType.get_data_type_name(
+                            int(attr_value)
+                        )
+                    elif op_attr == 'format':
+                        attr_value = GeFormat.get_format_name(int(attr_value))
+
+                op_info[cur_op_info_key][op_attr] = attr_value
+
+        # the list info are full_op_name, op_name, op_type, subgraph, op_info
+        return [full_op_name, op_name, op_type, subgraph_name,
+                json.dumps(op_info)]
+
+    def _get_subgraph_name(self, full_op_name):
+        """
+        Get subgraph name.
+
+        Args:
+            full_op_name (str): The full operator name.
+
+        Returns:
+            str, the subgraph name.
+        """
+        subgraph_name = full_op_name.split('/', 1)[0]
+        if subgraph_name in ['Default', 'Gradients']:
+            return subgraph_name
+        return None
+
+    def _get_op_name(self, full_op_name, subgraph_name):
+        """
+        Get operator name.
+
+        Args:
+            full_op_name (str): The full operator name.
+            subgraph_name (str): The subgraph name.
+
+        Returns:
+            str, the operator name.
+        """
+        if subgraph_name is None:
+            return full_op_name
+
+        if self._backend_type == 'vm':
+            return full_op_name.split('/')[-1]
+
+        strs = full_op_name.split(subgraph_name + '/')
+        op_name = None
+        for name_str in strs:
+            if not name_str:
+                continue
+            if op_name is None:
+                op_name = name_str.split('/')[-1]
+            else:
+                op_name = '+'.join([op_name, name_str.split('/')[-1]])
+        return op_name

mindinsight/utils/constant.py

@@ -29,6 +29,7 @@ class MindInsightModules(Enum):
     GENERAL = 0
     LINEAGEMGR = 2
     DATAVISUAL = 5
+    PROFILERMGR = 6
 
 
 class GeneralErrors(Enum):
@@ -43,6 +44,10 @@ class GeneralErrors(Enum):
     URL_DECODE_ERROR = 10
 
 
+class ProfilerMgrErrors(Enum):
+    """Enum definition for profiler errors."""
+
+
 class LineageMgrErrors(Enum):
     """Enum definition for lineage errors."""
 

requirements.txt

@@ -12,4 +12,5 @@ numpy>=1.17.0
 protobuf>=3.8.0
 psutil>=5.6.1
 six>=1.12.0
-Werkzeug>=1.0.0
+Werkzeug>=1.0.0
+tabulate>=0.8.6