mindstudio-probe 8.1.0__tar.gz → 8.1.2__tar.gz

{mindstudio_probe-8.1.0 → mindstudio_probe-8.1.2}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: mindstudio-probe
-Version: 8.1.0
+Version: 8.1.2
 Summary: Ascend Probe Utils
 Home-page: https://gitee.com/ascend/mstt/tree/master/debug/accuracy_tools/msprobe
 Author: Ascend Team
@@ -28,11 +28,12 @@ Requires-Dist: pyyaml
 Requires-Dist: rich
 Requires-Dist: tqdm
 Requires-Dist: openpyxl>=3.0.6
-Requires-Dist: pyopenssl
+Requires-Dist: pyopenssl==24.2.1
 Requires-Dist: twisted
 Requires-Dist: matplotlib
 Requires-Dist: tensorboard
 Requires-Dist: tabulate
 Requires-Dist: pwinput
+Requires-Dist: psutil
 
 MindStudio-Probe is a set of tools for diagnosing and improving model accuracy on Ascend NPU, including API acc checker, ptdbg, grad tool etc.

{mindstudio_probe-8.1.0 → mindstudio_probe-8.1.2}/mindstudio_probe.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: mindstudio-probe
-Version: 8.1.0
+Version: 8.1.2
 Summary: Ascend Probe Utils
 Home-page: https://gitee.com/ascend/mstt/tree/master/debug/accuracy_tools/msprobe
 Author: Ascend Team
@@ -28,11 +28,12 @@ Requires-Dist: pyyaml
 Requires-Dist: rich
 Requires-Dist: tqdm
 Requires-Dist: openpyxl>=3.0.6
-Requires-Dist: pyopenssl
+Requires-Dist: pyopenssl==24.2.1
 Requires-Dist: twisted
 Requires-Dist: matplotlib
 Requires-Dist: tensorboard
 Requires-Dist: tabulate
 Requires-Dist: pwinput
+Requires-Dist: psutil
 
 MindStudio-Probe is a set of tools for diagnosing and improving model accuracy on Ascend NPU, including API acc checker, ptdbg, grad tool etc.

{mindstudio_probe-8.1.0 → mindstudio_probe-8.1.2}/mindstudio_probe.egg-info/SOURCES.txt

@@ -83,6 +83,7 @@ msprobe/core/grad_probe/utils.py
 msprobe/core/kernel_dump/kernel_config.py
 msprobe/core/monitor/__init__.py
 msprobe/core/monitor/anomaly_processor.py
+msprobe/core/monitor/utils.py
 msprobe/core/overflow_check/abnormal_scene.py
 msprobe/core/overflow_check/api_info.py
 msprobe/core/overflow_check/checker.py
@@ -218,7 +219,6 @@ msprobe/mindspore/api_accuracy_checker/type_mapping.py
 msprobe/mindspore/api_accuracy_checker/utils.py
 msprobe/mindspore/api_accuracy_checker/bench_functions/flash_attention_score.py
 msprobe/mindspore/api_accuracy_checker/bench_functions/fusion_operator.py
-msprobe/mindspore/api_accuracy_checker/generate_op_script/config_op.json
 msprobe/mindspore/api_accuracy_checker/generate_op_script/op_generator.py
 msprobe/mindspore/api_accuracy_checker/generate_op_script/operator_replication.template
 msprobe/mindspore/code_mapping/__init__.py

{mindstudio_probe-8.1.0 → mindstudio_probe-8.1.2}/mindstudio_probe.egg-info/requires.txt

@@ -6,9 +6,10 @@ pyyaml
 rich
 tqdm
 openpyxl>=3.0.6
-pyopenssl
+pyopenssl==24.2.1
 twisted
 matplotlib
 tensorboard
 tabulate
 pwinput
+psutil

{mindstudio_probe-8.1.0 → mindstudio_probe-8.1.2}/msprobe/core/common/const.py

@@ -56,6 +56,9 @@ class Const:
     CPU_QUARTER = 4
     DUMP_MAX_DEPTH = 50
 
+    EXTERN_INPUT_LIST_MAX_LEN = 100
+    MAX_PROCESS_NUM = 128
+
     # dump mode
     ALL = "all"
     LIST = "list"
@@ -681,6 +684,7 @@ class FileCheckConst:
     MAX_IR_SIZE = 1073741824  # 1 * 1024 * 1024 * 1024
     MAX_ZIP_SIZE = 10737418240  # 10 * 1024 * 1024 * 1024
     MAX_FILE_IN_ZIP_SIZE = 1073741824  # 1 * 1024 * 1024 * 1024
+    MAX_FILE_SIZE = 1073741824  # 1 * 1024 * 1024 * 1024
     COMMOM_FILE_SIZE = 1048576  # 1 * 1024 * 1024
     DIR = "dir"
     FILE = "file"

{mindstudio_probe-8.1.0 → mindstudio_probe-8.1.2}/msprobe/core/common/file_utils.py

@@ -33,7 +33,7 @@ import pandas as pd
 from msprobe.core.common.decorator import recursion_depth_decorator
 from msprobe.core.common.log import logger
 from msprobe.core.common.exceptions import FileCheckException
-from msprobe.core.common.const import FileCheckConst
+from msprobe.core.common.const import FileCheckConst, CompareConst
 from msprobe.core.common.global_lock import global_lock, is_main_process
 
 proc_lock = multiprocessing.Lock()
@@ -384,7 +384,7 @@ def check_file_type(path):
     elif os.path.isfile(path):
         return FileCheckConst.FILE
     else:
-        logger.error(f'{path} does not exist, please check!')
+        logger.error(f'path does not exist, please check!')
         raise FileCheckException(FileCheckException.INVALID_FILE_ERROR)
 
 
@@ -460,6 +460,43 @@ def save_excel(path, data):
                 return "list"
         raise ValueError("Data must be a DataFrame or a list of (DataFrame, sheet_name) pairs.")
 
+    def check_value_is_valid(value: str) -> bool:
+        if not isinstance(value, str):
+            return True
+        try:
+            # -1.00 or +1.00 should be considered as digit numbers
+            float(value)
+        except ValueError:
+            # otherwise, they will be considered as formular injections
+            return not bool(re.compile(FileCheckConst.CSV_BLACK_LIST).search(value))
+        return True
+
+    def malicious_check(df):
+        for row_name in df.index:
+            if not check_value_is_valid(row_name):
+                raise RuntimeError(f"Malicious value [{row_name}] not allowed to be written into the excel: {path}.")
+
+        for col_name in df.columns:
+            if not check_value_is_valid(col_name):
+                raise RuntimeError(f"Malicious value [{col_name}] not allowed to be written into the excel: {path}.")
+
+        for _, row in df.iterrows():
+            for _, value in row.items():
+                if not check_value_is_valid(value):
+                    raise RuntimeError(f"Malicious value [{value}] not allowed to be written into the excel: {path}.")
+
+    def save_in_slice(df, base_name):
+        malicious_check(df)
+        df_length = len(df)
+        if df_length < CompareConst.MAX_EXCEL_LENGTH:
+            df.to_excel(writer, sheet_name=base_name if base_name else 'Sheet1', index=False)
+        else:
+            slice_num = (df_length + CompareConst.MAX_EXCEL_LENGTH - 1) // CompareConst.MAX_EXCEL_LENGTH
+            slice_size = (df_length + slice_num - 1) // slice_num
+            for i in range(slice_num):
+                df.iloc[i * slice_size: min((i + 1) * slice_size, df_length)] \
+                    .to_excel(writer, sheet_name=f'{base_name}_part_{i}' if base_name else f'part_{i}', index=False)
+
     check_path_before_create(path)
     path = os.path.realpath(path)
 
@@ -467,12 +504,12 @@ def save_excel(path, data):
     data_type = validate_data(data)
 
     try:
-        if data_type == "single":
-            data.to_excel(path, index=False)
-        elif data_type == "list":
-            with pd.ExcelWriter(path) as writer:
+        with pd.ExcelWriter(path) as writer:
+            if data_type == "single":
+                save_in_slice(data, None)
+            elif data_type == "list":
                 for data_df, sheet_name in data:
-                    data_df.to_excel(writer, sheet_name=sheet_name, index=False)
+                    save_in_slice(data_df, sheet_name)
     except Exception as e:
         logger.error(f'Save excel file "{os.path.basename(path)}" failed.')
         raise RuntimeError(f"Save excel file {path} failed.") from e
@@ -677,8 +714,23 @@ def os_walk_for_files(path, depth):
     return res
 
 
+def check_zip_file(zip_file_path):
+    with zipfile.ZipFile(zip_file_path, 'r') as zip_file:
+        total_size = 0
+        if len(zip_file.infolist()) > FileCheckConst.MAX_FILE_IN_ZIP_SIZE:
+            raise ValueError(f"Too many files in {os.path.basename(zip_file_path)}")
+        for file_info in zip_file.infolist():
+            if file_info.file_size > FileCheckConst.MAX_FILE_SIZE:
+                raise ValueError(f"File {file_info.filename} is too large to extract")
+
+            total_size += file_info.file_size
+            if total_size > FileCheckConst.MAX_ZIP_SIZE:
+                raise ValueError(f"Total extracted size exceeds the limit of {FileCheckConst.MAX_ZIP_SIZE} bytes")
+
+
 def read_xlsx(file_path, sheet_name=None):
     check_file_or_directory_path(file_path)
+    check_zip_file(file_path)
     try:
         if sheet_name:
             result_df = pd.read_excel(file_path, keep_default_na=False, sheet_name=sheet_name)
@@ -719,6 +771,13 @@ def create_file_with_content(data, filepath):
     change_mode(filepath, FileCheckConst.DATA_FILE_AUTHORITY)
 
 
+def check_file_whether_exist_or_not(filepath):
+    if os.path.exists(filepath):
+        check_file_or_directory_path(filepath)
+    else:
+        check_path_before_create(filepath)
+
+
 def add_file_to_zip(zip_file_path, file_path, arc_path=None):
     """
     Add a file to a ZIP archive, if zip does not exist, create one.
@@ -727,12 +786,13 @@ def add_file_to_zip(zip_file_path, file_path, arc_path=None):
     :param file_path: Path to the file to add
     :param arc_path: Optional path inside the ZIP archive where the file should be added
     """
+    check_file_or_directory_path(file_path)
     check_file_suffix(zip_file_path, FileCheckConst.ZIP_SUFFIX)
+    check_file_whether_exist_or_not(zip_file_path)
     check_file_size(file_path, FileCheckConst.MAX_FILE_IN_ZIP_SIZE)
     zip_size = os.path.getsize(zip_file_path) if os.path.exists(zip_file_path) else 0
     if zip_size + os.path.getsize(file_path) > FileCheckConst.MAX_ZIP_SIZE:
         raise RuntimeError(f"ZIP file size exceeds the limit of {FileCheckConst.MAX_ZIP_SIZE} bytes")
-    check_path_before_create(zip_file_path)
     try:
         proc_lock.acquire()
         with zipfile.ZipFile(zip_file_path, 'a') as zip_file:
@@ -754,7 +814,7 @@ def create_file_in_zip(zip_file_path, file_name, content):
     :param content: Content to write to the file
     """
     check_file_suffix(zip_file_path, FileCheckConst.ZIP_SUFFIX)
-    check_path_before_create(zip_file_path)
+    check_file_whether_exist_or_not(zip_file_path)
     zip_size = os.path.getsize(zip_file_path) if os.path.exists(zip_file_path) else 0
     if zip_size + sys.getsizeof(content) > FileCheckConst.MAX_ZIP_SIZE:
         raise RuntimeError(f"ZIP file size exceeds the limit of {FileCheckConst.MAX_ZIP_SIZE} bytes")
@@ -781,26 +841,22 @@ def extract_zip(zip_file_path, extract_dir):
     :param extract_dir: Directory to extract the contents to
     """
     check_file_suffix(zip_file_path, FileCheckConst.ZIP_SUFFIX)
+    check_file_or_directory_path(zip_file_path)
+    create_directory(extract_dir)
     try:
         proc_lock.acquire()
-        with zipfile.ZipFile(zip_file_path, 'r') as zip_file:
-            total_size = 0
-            if len(zip_file.infolist()) > FileCheckConst.MAX_FILE_IN_ZIP_SIZE:
-                raise ValueError(f"Too many files in {os.path.basename(zip_file_path)}")
-            for file_info in zip_file.infolist():
-                if file_info.file_size > FileCheckConst.MAX_FILE_IN_ZIP_SIZE:
-                    raise ValueError(f"File {file_info.filename} is too large to extract")
-
-                total_size += file_info.file_size
-                if total_size > FileCheckConst.MAX_ZIP_SIZE:
-                    raise ValueError(f"Total extracted size exceeds the limit of {FileCheckConst.MAX_ZIP_SIZE} bytes")
+        check_zip_file(zip_file_path)
     except Exception as e:
         logger.error(f'Save content to file "{os.path.basename(zip_file_path)}" failed.')
         raise RuntimeError(f"Save content to file {os.path.basename(zip_file_path)} failed.") from e
     finally:
         proc_lock.release()
-    with zipfile.ZipFile(zip_file_path, 'r') as zip_file:
-        zip_file.extractall(extract_dir)
+    try:
+        with zipfile.ZipFile(zip_file_path, 'r') as zip_file:
+            zip_file.extractall(extract_dir)
+    except Exception as e:
+        raise RuntimeError(f"extract zip file {os.path.basename(zip_file_path)} failed") from e
+    recursive_chmod(extract_dir)
 
 
 def split_zip_file_path(zip_file_path):
@@ -877,7 +933,8 @@ class SharedDict:
             self._shm = shared_memory.SharedMemory(create=False, name=name)
         except FileNotFoundError:
             try:
-                self._shm = shared_memory.SharedMemory(create=True, name=name, size=1024 * 1024)
+                # 共享内存空间增加至5M
+                self._shm = shared_memory.SharedMemory(create=True, name=name, size=1024 * 1024 * 5)
                 data = pickle.dumps({})
                 self._shm.buf[0:len(data)] = bytearray(data)
                 logger.debug(f'create shared memory, name: {name}')
@@ -892,6 +949,7 @@ class SharedDict:
             except Exception as e:
                 logger.debug(f'shared dict is unreadable, reason: {e}, create new dict.')
                 self._dict = {}
+                self._shm.buf[:] = bytearray(b'\x00' * len(self._shm.buf))  # 清空内存
                 self._changed = True
 
 

{mindstudio_probe-8.1.0 → mindstudio_probe-8.1.2}/msprobe/core/common/utils.py

@@ -14,21 +14,22 @@
 # limitations under the License.
 
 import collections
+import functools
+import inspect
 import os
 import re
-import subprocess
+import threading
 import time
-import inspect
+from collections import OrderedDict
 from datetime import datetime, timezone
 
 import numpy as np
 
-from msprobe.core.common.file_utils import (FileOpen, check_file_or_directory_path, load_json)
 from msprobe.core.common.const import Const, CompareConst
-from msprobe.core.common.log import logger
-from msprobe.core.common.exceptions import MsprobeException
 from msprobe.core.common.decorator import recursion_depth_decorator
-
+from msprobe.core.common.exceptions import MsprobeException
+from msprobe.core.common.file_utils import (FileOpen, check_file_or_directory_path, load_json)
+from msprobe.core.common.log import logger
 
 device = collections.namedtuple('device', ['type', 'index'])
 prefixes = ['api_stack', 'list', 'range', 'acl']
@@ -112,6 +113,82 @@ class DumpException(MsprobeBaseException):
         return f"Dump Error Code {self.code}: {self.error_info}"
 
 
+class ThreadSafe:
+    """
+    线程安全控制工具类，提供三种使用方式：
+    1.上下文管理器：with ThreadSafe()
+    2.主动加锁与释放锁：ThreadSafe.acquire()/ThreadSafe.release()
+    3.方法装饰器：@ThreadSafe.synchronized
+    """
+    _lock = threading.RLock()
+
+    def __enter__(self):
+        self.__class__._lock.acquire()
+
+    def __exit__(self, exc_type, exc_val, exc_tb):
+        self.__class__._lock.release()
+
+    @classmethod
+    def acquire(cls):
+        cls._lock.acquire()
+
+    @classmethod
+    def release(cls):
+        cls._lock.release()
+
+    @classmethod
+    def synchronized(cls, func):
+        @functools.wraps(func)
+        def wrapper(*args, **kwargs):
+            with cls._lock:
+                return func(*args, **kwargs)
+
+        return wrapper
+
+
+class ModuleQueue:
+    def __init__(self):
+        self.queue = OrderedDict()
+
+    def add_name(self, name):
+        self.queue[name] = True
+
+    def remove_name(self, name):
+        if name in self.queue:
+            del self.queue[name]
+
+    def find_last(self, name):
+        """
+        在队列中找到当前 Module/Cell 的父节点名称并返回，若找不到则返回None
+
+        Args:
+            name: 需要寻找父节点的 Module/Cell 的名称
+
+        Returns:
+            返回父节点名称，找不到则返回None
+
+        Examples:
+            父节点名称格式: Module.module1.module1.forward.0
+            子节点名称格式: Module.module1.module2.Module2.forward.0
+            匹配关系: Module/Cell 的名称总能被点(.)分割符分成5个部分及以上，子节点截断后4个点和父节点截断后3个点的前缀名称是匹配的
+        """
+        child_parts = name.split('.')
+        if len(child_parts) < 5:
+            return None
+        child_name_prefix = '.'.join(child_parts[:-4])
+        if child_name_prefix in Const.MODULE_PREFIX:
+            return None
+
+        for parent_name in reversed(self.queue):
+            parent_parts = parent_name.split('.')
+            if len(parent_parts) < 5:
+                return None
+            parent_name_prefix = '.'.join(parent_parts[:-3])
+            if parent_name_prefix == child_name_prefix:
+                return parent_name
+        return None
+
+
 def is_json_file(file_path):
     if isinstance(file_path, str) and file_path.lower().endswith('.json'):
         return True
@@ -156,9 +233,10 @@ def check_compare_param(input_param, output_path, dump_mode, stack_mode):
 
 def check_configuration_param(stack_mode=False, auto_analyze=True, fuzzy_match=False, is_print_compare_log=True):
     arg_list = [stack_mode, auto_analyze, fuzzy_match, is_print_compare_log]
-    for arg in arg_list:
+    arg_names = ['stack_mode', 'auto_analyze', 'fuzzy_match', 'is_print_compare_log']
+    for arg, name in zip(arg_list, arg_names):
         if not isinstance(arg, bool):
-            logger.error(f"Invalid input parameter, {arg} which should be only bool type.")
+            logger.error(f"Invalid input parameter, {name} which should be only bool type.")
             raise CompareException(CompareException.INVALID_PARAM_ERROR)
 
 
@@ -282,9 +360,9 @@ def set_dump_path(input_param):
     npu_path = input_param.get("npu_json_path", None)
     bench_path = input_param.get("bench_json_path", None)
     dump_json_path_valid = npu_path is not None and npu_path.endswith("dump.json") and \
-        bench_path is not None and bench_path.endswith("dump.json")
+                           bench_path is not None and bench_path.endswith("dump.json")
     debug_json_path_valid = npu_path is not None and npu_path.endswith("debug.json") and \
-        bench_path is not None and bench_path.endswith("debug.json")
+                            bench_path is not None and bench_path.endswith("debug.json")
     if not dump_json_path_valid and not debug_json_path_valid:
         logger.error(f"Please check the json path is valid and ensure that neither npu_path nor bench_path is None.")
         raise CompareException(CompareException.INVALID_PATH_ERROR)
@@ -303,6 +381,21 @@ def get_file_type(file_path):
     return file_type
 
 
+def check_dump_json_key(json_data, device_type):
+    task = json_data.get('task', None)
+    if not task:
+        logger.error(f"Task for {device_type} is empty, please check.")
+        raise CompareException(CompareException.INVALID_TASK_ERROR)
+    if 'data' not in json_data:
+        logger.error(f"Missing 'data' in dump.json, please check dump.json of {device_type}.")
+        raise CompareException(CompareException.INVALID_DATA_ERROR)
+    api_data = json_data.get('data')
+    if not isinstance(api_data, dict):
+        logger.error(f"Invalid type for 'data': expected a dict. Please check dump.json of {device_type}.")
+        raise CompareException(CompareException.INVALID_DATA_ERROR)
+    return task, api_data
+
+
 def get_dump_mode(input_param):
     npu_path = input_param.get("npu_json_path", None)
     bench_path = input_param.get("bench_json_path", None)
@@ -310,12 +403,8 @@ def get_dump_mode(input_param):
     bench_json_data = load_json(bench_path)
     json_type = get_file_type(file_path=npu_path)
 
-    npu_task = npu_json_data.get('task', None)
-    bench_task = bench_json_data.get('task', None)
-
-    if not npu_task or not bench_task:
-        logger.error(f"Please check the dump task is correct, npu's task is {npu_task}, bench's task is {bench_task}.")
-        raise CompareException(CompareException.INVALID_TASK_ERROR)
+    npu_task, npu_api_data = check_dump_json_key(npu_json_data, 'npu')
+    bench_task, bench_api_data = check_dump_json_key(bench_json_data, 'bench')
 
     if npu_task != bench_task:
         logger.error(f"Please check the dump task is consistent.")
@@ -328,8 +417,8 @@ def get_dump_mode(input_param):
         return Const.STRUCTURE
 
     if npu_task == Const.STATISTICS:
-        npu_md5_compare = md5_find(npu_json_data['data'], json_type)
-        bench_md5_compare = md5_find(bench_json_data['data'], json_type)
+        npu_md5_compare = md5_find(npu_api_data, json_type)
+        bench_md5_compare = md5_find(bench_api_data, json_type)
         if npu_md5_compare == bench_md5_compare:
             return Const.MD5 if npu_md5_compare else Const.SUMMARY
         else:
@@ -446,10 +535,10 @@ def get_real_step_or_rank(step_or_rank_input, obj):
 def check_init_step(step):
     if not is_int(step):
         raise MsprobeException(MsprobeException.INVALID_PARAM_ERROR,
-                        f"{step} must be an integer")
+                               f"{step} must be an integer")
     if not step >= 0:
         raise MsprobeException(MsprobeException.INVALID_PARAM_ERROR,
-                f"{step} must be greater than or equal to 0")
+                               f"{step} must be greater than or equal to 0")
 
 
 def check_token_range(token_range):
@@ -557,14 +646,25 @@ def replace_last_occurrence(text, old, new):
 
 def load_stack_json(stack_path):
     stack_dict = load_json(stack_path)
+
+    if not isinstance(stack_dict, dict):
+        raise MsprobeException(
+            MsprobeException.INVALID_PARAM_ERROR,
+            "The format of the stack.json is incorrect, the outermost layer of stack.json should be a dict type."
+        )
+
     if not stack_dict.get(Const.NEW_STACK_FLAG):
         return stack_dict
 
     new_stack_dict = {}
     for stack_info in stack_dict.values():
-        if len(stack_info) != 2:
+        if not isinstance(stack_info, list) or len(stack_info) != 2:
             continue
+
         api_list, stack_str = stack_info
+        if not isinstance(api_list, list):
+            continue
+
         for api_name in api_list:
             new_stack_dict.update({api_name: stack_str})
     return new_stack_dict
@@ -586,3 +686,18 @@ def analyze_api_call_stack(name):
     else:
         stack_str.append(Const.WITHOUT_CALL_STACK)
     return "".join(stack_str)
+
+
+def check_extern_input_list(input_list):
+    if not isinstance(input_list, list):
+        raise Exception("input is not a list")
+    if len(input_list) > Const.EXTERN_INPUT_LIST_MAX_LEN:
+        raise Exception(f"input list exceed max length {Const.EXTERN_INPUT_LIST_MAX_LEN}")
+
+
+def check_process_num(process_num):
+    if not is_int(process_num) or process_num <= 0:
+        raise ValueError(f"process_num({process_num}) is not a positive integer")
+    if process_num > Const.MAX_PROCESS_NUM:
+        raise ValueError(f"The maximum supported process_num is {Const.MAX_PROCESS_NUM}, current value: {process_num}.")
+

{mindstudio_probe-8.1.0 → mindstudio_probe-8.1.2}/msprobe/core/common_config.py

@@ -13,7 +13,9 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-from msprobe.core.common.const import Const, FileCheckConst
+import re
+
+from msprobe.core.common.const import Const
 from msprobe.core.common.log import logger
 from msprobe.core.common.exceptions import MsprobeException
 from msprobe.core.common.utils import get_real_step_or_rank
@@ -67,6 +69,7 @@ class BaseConfig:
         self.if_preheat = json_config.get("if_preheat")
         self.preheat_step = json_config.get("preheat_step")
         self.max_sample = json_config.get("max_sample")
+        self.is_regex_valid = True
 
     @staticmethod
     def _check_str_list_config(config_item, config_name):
@@ -83,6 +86,7 @@ class BaseConfig:
         self._check_str_list_config(self.scope, "scope")
         self._check_str_list_config(self.list, "list")
         self._check_data_mode()
+        self._check_regex_in_list()
 
     def _check_data_mode(self):
         if self.data_mode is not None:
@@ -118,3 +122,13 @@ class BaseConfig:
                         f"summary_mode is invalid, summary_mode is not in {Const.SUMMARY_MODE}.",
                         MsprobeException(MsprobeException.INVALID_PARAM_ERROR)
                     )
+
+    def _check_regex_in_list(self):
+        if self.list:
+            for name in self.list:
+                if name.startswith('name-regex(') and name.endswith(')'):
+                    try:
+                        re.compile(name[len('name-regex('):-1])
+                    except re.error:
+                        self.is_regex_valid = False
+                        break

{mindstudio_probe-8.1.0 → mindstudio_probe-8.1.2}/msprobe/core/compare/acc_compare.py

@@ -31,7 +31,7 @@ from msprobe.core.common.utils import CompareException, add_time_with_xlsx, chec
     set_dump_path, get_dump_mode, check_compare_param, check_configuration_param, load_stack_json, get_file_type
 from msprobe.core.compare.check import check_dump_json_str, check_stack_json_str, cross_dtype_mapping
 from msprobe.core.compare.utils import merge_tensor, print_compare_ends_info, read_op, \
-    reorder_op_x_list, set_stack_json_path
+    reorder_op_x_list, set_stack_json_path, check_api_info_len
 from msprobe.core.compare.config import ModeConfig, MappingConfig, MappingDict
 from msprobe.core.compare.multiprocessing_compute import CompareRealData
 from msprobe.core.compare.highlight import HighLight
@@ -105,7 +105,7 @@ class Comparator:
         stack_json = input_param.get("stack_json_path")
         result_df = self.compare_statistics([npu_json, bench_json, stack_json])
         if not result_df.values.tolist():
-            logger.warning("Can`t match any op.")
+            logger.warning("Can`t match any op. No compare result file generated.")
             return
 
         # compare real data
@@ -186,11 +186,16 @@ class ParseData:
         elif self.mode_config.dump_mode == Const.MD5:
             result[Const.MD5] = []
 
-        api_nums = len(data_json['data'])
+        apis_data = data_json.get('data', None)
+        if not apis_data:
+            logger.warning('No APIs found in dump.json.')
+            return pd.DataFrame(result)
+
+        api_nums = len(apis_data)
         progress_bar = tqdm(total=api_nums, desc="API/Module Read Progress", unit="api/module", ncols=100)
 
         # 从json中循环解析API数据，遍历所有API
-        for data_name in data_json['data']:
+        for data_name in apis_data:
             check_op_str_pattern_valid(data_name)
             merge_list = self.gen_merge_list(data_json, data_name, stack_json_data)
             if not merge_list:
@@ -206,25 +211,37 @@ class ParseData:
             for index, op_name in enumerate(op_name_reorder):
                 result[CompareConst.OP_NAME].append(op_name)
                 if (CompareConst.INPUT_PATTERN in op_name) or (CompareConst.KWARGS_PATTERN in op_name):
-                    struct = merge_list[CompareConst.INPUT_STRUCT].pop(0)
+                    info_list = merge_list[CompareConst.INPUT_STRUCT]
                 elif CompareConst.OUTPUT_PATTERN in op_name:
-                    struct = merge_list[CompareConst.OUTPUT_STRUCT].pop(0)
+                    info_list = merge_list[CompareConst.OUTPUT_STRUCT]
                 elif CompareConst.PARAMS_PATTERN in op_name:
-                    struct = merge_list[CompareConst.PARAMS_STRUCT].pop(0)
+                    info_list = merge_list[CompareConst.PARAMS_STRUCT]
                 elif CompareConst.PARAMS_GRAD_PATTERN in op_name:
-                    struct = merge_list[CompareConst.PARAMS_GRAD_STRUCT].pop(0)
+                    info_list = merge_list[CompareConst.PARAMS_GRAD_STRUCT]
                 else:
-                    struct = merge_list[CompareConst.DEBUG_STRUCT].pop(0)
+                    info_list = merge_list[CompareConst.DEBUG_STRUCT]
+                check_api_info_len(op_name, info_list, 1)
+                struct = info_list.pop(0)
+
+                check_api_info_len(op_name, struct, 2)
                 result[Const.DTYPE].append(struct[0])
                 result[Const.SHAPE].append(struct[1])
                 if self.mode_config.dump_mode == Const.MD5:
+                    check_api_info_len(op_name, struct, 3)
                     result[Const.MD5].append(struct[2])
+
+                check_api_info_len(op_name, summary_reorder, 1)
                 result[Const.SUMMARY].append(summary_reorder.pop(0))
-                result[Const.STACK_INFO].append(
-                    merge_list[Const.STACK_INFO][0] if index == 0 and self.mode_config.stack_mode else None)
+
+                if index == 0 and self.mode_config.stack_mode:
+                    check_api_info_len(op_name, merge_list[Const.STACK_INFO], 1)
+                    result[Const.STACK_INFO].append(merge_list[Const.STACK_INFO][0])
+                else:
+                    result[Const.STACK_INFO].append(None)
+
                 if self.mode_config.dump_mode == Const.ALL:
+                    check_api_info_len(op_name, data_name_reorder, 1)
                     result['data_name'].append(data_name_reorder.pop(0))
-
             progress_bar.update(1)
         progress_bar.close()
         return pd.DataFrame(result)
@@ -599,8 +616,12 @@ class CreateTable:
         npu_summary = [CompareConst.NPU_MAX, CompareConst.NPU_MIN, CompareConst.NPU_MEAN, CompareConst.NPU_NORM]
         bench_summary = [CompareConst.BENCH_MAX, CompareConst.BENCH_MIN, CompareConst.BENCH_MEAN,
                          CompareConst.BENCH_NORM]
-        result[npu_summary] = result['summary_x'].apply(self.set_summary).tolist()
-        result[bench_summary] = result['summary_y'].apply(self.set_summary).tolist()
+        if result.empty:
+            result[npu_summary] = pd.DataFrame(columns=npu_summary)
+            result[bench_summary] = pd.DataFrame(columns=bench_summary)
+        else:
+            result[npu_summary] = result['summary_x'].apply(self.set_summary).tolist()
+            result[bench_summary] = result['summary_y'].apply(self.set_summary).tolist()
 
         result_df = pd.DataFrame(columns=header)
         for h in header: