mindstudio-probe 1.2.2__py3-none-any.whl → 8.1.0__py3-none-any.whl

msprobe/core/compare/highlight.py

@@ -30,12 +30,7 @@ from msprobe.core.common.file_utils import save_workbook
 from msprobe.core.common.log import logger
 from msprobe.core.common.utils import get_header_index, safe_get_value
 from msprobe.core.compare.utils import table_value_is_valid, get_name_and_state, CompareException
-
-
-class HighlightCheck(abc.ABC):
-    @abc.abstractmethod
-    def apply(self, info, color_columns, dump_mode):
-        raise NotImplementedError
+from msprobe.core.compare.config import ModeConfig
 
 
 def add_highlight_row_info(color_list, num, highlight_err_msg):
@@ -46,6 +41,12 @@ def add_highlight_row_info(color_list, num, highlight_err_msg):
     color_list.append((num, [highlight_err_msg]))
 
 
+class HighlightCheck(abc.ABC):
+    @abc.abstractmethod
+    def apply(self, info, color_columns, dump_mode):
+        raise NotImplementedError
+
+
 class CheckOrderMagnitude(HighlightCheck):
     """检查Max diff的数量级差异"""
 
@@ -75,12 +76,12 @@ class CheckOneThousandErrorRatio(HighlightCheck):
         if (api_in[one_thousand_index] > CompareConst.ONE_THOUSAND_ERROR_IN_RED and
                 api_out[one_thousand_index] < CompareConst.ONE_THOUSAND_ERROR_OUT_RED):
             add_highlight_row_info(color_columns.red, num,
-                                   "The input/parameters's one thousandth err ratio exceeds 0.9, "
+                                   "The input/parameter's one thousandth err ratio exceeds 0.9, "
                                    "while the output's is below 0.6")
         elif api_in[one_thousand_index] - api_out[one_thousand_index] > CompareConst.ONE_THOUSAND_ERROR_DIFF_YELLOW:
             add_highlight_row_info(color_columns.yellow, num,
                                    "The output's one thousandth err ratio decreases by more than 0.1 "
-                                   "compared to the input/parameters's")
+                                   "compared to the input/parameter's")
 
 
 class CheckCosineSimilarity(HighlightCheck):
@@ -94,7 +95,7 @@ class CheckCosineSimilarity(HighlightCheck):
         if api_in[cosine_index] - api_out[cosine_index] > CompareConst.COSINE_DIFF_YELLOW:
             add_highlight_row_info(color_columns.yellow, num,
                                    "The output's cosine decreases by more than 0.1 "
-                                   "compared to the input/parameters's")
+                                   "compared to the input/parameter's")
 
 
 class CheckMaxRelativeDiff(HighlightCheck):
@@ -117,7 +118,7 @@ class CheckMaxRelativeDiff(HighlightCheck):
               input_max_relative_diff < CompareConst.MAX_RELATIVE_IN_YELLOW):
             add_highlight_row_info(color_columns.yellow, num,
                                    "The output's maximum relative error exceeds 0.1, "
-                                   "while the input/parameters's is below 0.01")
+                                   "while the input/parameter's is below 0.01")
 
 
 class CheckOverflow(HighlightCheck):
@@ -146,84 +147,19 @@ class HighlightRules:
     }
 
     # 用于比较输入和输出的规则
+    # 真实数据检查规则
     compare_rules = {
         "check_order_magnitude": CheckOrderMagnitude(),
         "check_one_thousand_error": CheckOneThousandErrorRatio(),
         "check_cosine_similarity": CheckCosineSimilarity()
     }
+    # 统计量数据检查规则
     summary_compare_rules = {
         "check_order_magnitude": CheckOrderMagnitude(),
         "check_max_relative_diff": CheckMaxRelativeDiff(),
     }
 
 
-def check_indices_numeric(api_items, indices: list):
-    """检查指定索引处的值是否都为数字类型（int 或 float）"""
-    return all(isinstance(api_items[i], (float, int)) for i in indices)
-
-
-def apply_comparison_rules(api_info, dump_mode, color_columns):
-    """output与input/params的比较"""
-    if dump_mode == Const.SUMMARY:
-        for rule in HighlightRules.summary_compare_rules.values():
-            rule.apply(api_info, color_columns, dump_mode)
-    else:
-        for rule in HighlightRules.compare_rules.values():
-            rule.apply(api_info, color_columns, dump_mode)
-
-
-def find_error_rows(result, api_batch, highlight_dict, dump_mode):
-    """找到单个API中需要高亮的行"""
-    if dump_mode == Const.MD5:
-        return
-    npu_max_index = get_header_index(CompareConst.NPU_MAX, dump_mode)
-    bench_max_index = get_header_index(CompareConst.BENCH_MAX, dump_mode)
-    max_diff_index = get_header_index(CompareConst.MAX_DIFF if dump_mode == Const.SUMMARY
-                                      else CompareConst.MAX_ABS_ERR, dump_mode)
-
-    red_lines, yellow_lines = [], []
-    LineInfo = namedtuple('LineInfo', ['line_data', 'num_pointer'])
-    ApiInfo = namedtuple('ApiInfo', ['api_input', 'api_output', 'num_pointer'])
-    ColorColumns = namedtuple('ColorColumns', ['red', 'yellow'])
-    color_columns = ColorColumns(red=red_lines, yellow=yellow_lines)
-
-    api_batch_start = api_batch.start  # result_df的input起始全局索引
-    api_batch_params_end_index = api_batch.params_end_index  # result_df的params结束全局索引 + 1
-    api_batch_output_end_index = api_batch.output_end_index  # result_df的output结束全局索引 + 1
-    api_batch_params_slice_index_local = api_batch_params_end_index - api_batch_start  # result的params结束局部切片索引
-    api_batch_output_slice_index_local = api_batch_output_end_index - api_batch_start  # result的output结束局部切片索引
-
-    # 对单行API的输入或输出进行误差判断
-    for i, line in enumerate(result):
-        index = api_batch_start + i
-        line_info = LineInfo(line_data=line, num_pointer=index)
-        for rule in HighlightRules.basic_rules.values():
-            rule.apply(line_info, color_columns, dump_mode)
-
-    # 对API的输出与输入比较，进行误差判断
-    for n, api_out in enumerate(result[api_batch_params_slice_index_local: api_batch_output_slice_index_local]):
-        index = api_batch_start + api_batch_params_slice_index_local + n
-        # 单行检查只有溢出检查（红色），如果已经溢出，不进一步检查
-        if index in red_lines:
-            continue
-        if not check_indices_numeric(api_out, [npu_max_index, bench_max_index, max_diff_index]):
-            continue
-
-        # input/parameters的比较检查, 这里api_in包括input、parameters
-        for _, api_in in enumerate(result[0: api_batch_params_slice_index_local]):
-            if not check_indices_numeric(api_in, [npu_max_index, bench_max_index, max_diff_index]):
-                continue
-            api_info = ApiInfo(api_input=api_in, api_output=api_out, num_pointer=index)
-            apply_comparison_rules(api_info, dump_mode, color_columns)
-
-    red_lines_num_set = {x[0] for x in red_lines}
-    yellow_lines_num_set = {x[0] for x in yellow_lines}
-    highlight_dict.get('red_rows', set()).update(red_lines_num_set)
-    highlight_dict.get('yellow_rows', set()).update(yellow_lines_num_set - red_lines_num_set)
-    highlight_dict.get('red_lines', []).extend(red_lines)
-    highlight_dict.get('yellow_lines', []).extend(yellow_lines)
-
-
 class ApiBatch:
     def __init__(self, api_name: str, start: int):
         self.api_name = api_name
@@ -257,159 +193,225 @@ class ApiBatch:
         self.params_grad_end_index += 1
 
 
-def api_batches_update(api_batches, api_name, state, index):
-    """
-    当一个api的所有item更新完后，input, output的索引范围：
-    input: [start: start+input_len]
-    output: [start+input_len: output_end_index]
-    params: [output_end_index: params_end_index]
-    """
-    if not api_batches:
-        api_batches.append(ApiBatch(api_name, index))
-    else:
-        api_batch = api_batches[-1]
-        if api_batch.api_name == api_name or (
-                not re.search(Const.REGEX_FORWARD_BACKWARD, api_name) and api_name in api_batch.api_name):
-            try:
-                api_batch.increment(state)
-            except ValueError as e:
-                logger.error(f"api_batch: {api_batch} with invalid state, please check! {e}")
-                raise CompareException(CompareException.INVALID_STATE_ERROR) from e
-        else:
-            api_batches.append(ApiBatch(api_name, index))
+class HighLight:
+    def __init__(self, mode_config: ModeConfig):
+        self.mode_config = mode_config
 
-
-def find_compare_result_error_rows(result_df, highlight_dict, dump_mode):
-    """将dataframe根据API分组，并找到有误差的算子用于高亮"""
-    result = result_df.values
-    api_batches = []
-    for i, res_i in enumerate(result):
-        api_full_name = safe_get_value(res_i, 0, "res_i")
-        api_name, state = get_name_and_state(api_full_name)
-        api_batches_update(api_batches, api_name, state, i)
-    with tqdm(total=len(api_batches), desc="API/Module Analyse Progress", unit="item", ncols=100) as progress_bar:
-        for api_batch in api_batches:
-            find_error_rows(result[api_batch.start: api_batch.params_grad_end_index], api_batch, highlight_dict, 
-                            dump_mode)
-            progress_bar.update(1)
-
-
-def value_check(value, api_name=None, i=None, result_df_columns=None):
-    if not table_value_is_valid(value):
-        if result_df_columns:
-            logger.error(f"Malicious value [{value}] at api_name [{api_name}], column [{result_df_columns[i]}], "
-                         f"is not allowed to be written into the compare result xlsx.")
+    @staticmethod
+    def api_batches_update(api_batches, api_name, state, index):
+        """
+        当一个api的所有item更新完后，input, output的索引范围：
+        input: [start: start+input_len]
+        output: [start+input_len: output_end_index]
+        params: [output_end_index: params_end_index]
+        """
+        if not api_batches:
+            api_batches.append(ApiBatch(api_name, index))
         else:
-            logger.error(f"Malicious value [{value}] is not allowed to be written into the compare result xlsx.")
-
-
-def df_malicious_value_check(df_chunk, result_df_columns):
-    for row in df_chunk.itertuples(index=False):
-        api_name = row[0]
-        for i, value in enumerate(row):
-            value_check(value, api_name, i, result_df_columns)
-
-
-def handle_multi_process_malicious_value_check(func, result_df):
-    result_total_nums = len(result_df)
-    process_num = int((multiprocessing.cpu_count() + 1) / 2)
-
-    if result_total_nums <= process_num:
-        process_num = 1
-        chunks = [result_df]
-    else:
-        chunk_size = result_total_nums // process_num
-        chunks = [result_df.iloc[i: i + chunk_size] for i in range(0, result_total_nums, chunk_size)]
-
-    pool = multiprocessing.Pool(process_num)
-
-    def err_call(args):
-        logger.error("Multiprocessing malicious value check failed! Reason: {}".format(args))
-        try:
-            pool.terminate()
-        except OSError:
-            logger.error("Pool terminate failed")
-
-    result_df_columns = result_df.columns.tolist()
-    for column in result_df_columns:
-        value_check(column)
-    for df_chunk in chunks:
-        pool.apply_async(func, args=(df_chunk, result_df_columns,), error_callback=err_call)
-
-    pool.close()
-    pool.join()
-
-
-def compare_result_df_convert(value):
-    if not isinstance(value, (float, int)) or isinstance(value, bool):  # bool类型或者非数字类型转str
-        value = f"{str(value)}\t" if str(value) in ("inf", "-inf", "nan") else str(value)
-    if isinstance(value, float):
-        value = f"{str(value)}\t" if str(value) in ("inf", "-inf", "nan") else value
-    return value
-
-
-def highlight_rows_xlsx(result_df, highlight_dict, file_path):
-    """Write and highlight results in Excel"""
+            api_batch = api_batches[-1]
+            if api_batch.api_name == api_name or (
+                    not re.search(Const.REGEX_FORWARD_BACKWARD, api_name) and api_name in api_batch.api_name):
+                try:
+                    api_batch.increment(state)
+                except ValueError as e:
+                    logger.error(f"api_batch: {api_batch} with invalid state, please check! {e}")
+                    raise CompareException(CompareException.INVALID_STATE_ERROR) from e
+            else:
+                api_batches.append(ApiBatch(api_name, index))
+
+    @staticmethod
+    def check_indices_numeric(api_items, indices: list):
+        """检查指定索引处的值是否都为数字类型（int 或 float）"""
+        return all(isinstance(api_items[i], (float, int)) for i in indices)
+
+    @staticmethod
+    def update_highlight_err_msg(result_df, highlight_dict):
+        if result_df.shape[1] <= 1:
+            return
 
-    update_highlight_err_msg(result_df, highlight_dict)  # add highlight err_msg
+        if CompareConst.NPU_MD5 in result_df.columns:
+            return
 
-    wb = openpyxl.Workbook()
-    ws = wb.active
+        err_msg = result_df.get(CompareConst.ERROR_MESSAGE)
+        red_lines_num_set = highlight_dict.get('red_rows')
+
+        for color in ['red', 'yellow']:
+            line_key = f'{color}_lines'
+            lines = highlight_dict.get(line_key, [])
+            for line_index, messages in lines:
+                if color == 'yellow' and line_index in red_lines_num_set:
+                    continue  # 如果是 yellow 行，且已被 red 行覆盖，跳过
+
+                for msg in messages:
+                    if err_msg[line_index] == '':
+                        err_msg[line_index] = msg
+                    else:
+                        err_msg[line_index] += '\n' + msg
+
+                if color == 'red':
+                    red_lines_num_set.add(line_index)
+
+        result_df[CompareConst.ERROR_MESSAGE] = err_msg
+
+    @staticmethod
+    def compare_result_df_convert(value):
+        if not isinstance(value, (float, int)) or isinstance(value, bool):  # bool类型或者非数字类型转str
+            value = f"{str(value)}\t" if str(value) in ("inf", "-inf", "nan") else str(value)
+        if isinstance(value, float):
+            value = f"{str(value)}\t" if str(value) in ("inf", "-inf", "nan") else value
+        return value
+
+    @staticmethod
+    def value_check(value, api_name=None, i=None, result_df_columns=None):
+        if not table_value_is_valid(value):
+            if result_df_columns:
+                logger.error(f"Malicious value [{value}] at api_name [{api_name}], column [{result_df_columns[i]}], "
+                             f"is not allowed to be written into the compare result xlsx.")
+            else:
+                logger.error(f"Malicious value [{value}] is not allowed to be written into the compare result xlsx.")
+
+    def find_compare_result_error_rows(self, result_df, highlight_dict):
+        """将dataframe根据API分组，并找到有误差的算子用于高亮"""
+        result = result_df.values
+        api_batches = []
+        for i, res_i in enumerate(result):
+            api_full_name = safe_get_value(res_i, 0, "res_i")
+            api_name, state = get_name_and_state(api_full_name)
+            self.api_batches_update(api_batches, api_name, state, i)
+        with tqdm(total=len(api_batches), desc="API/Module Analyse Progress", unit="item", ncols=100) as progress_bar:
+            for api_batch in api_batches:
+                self.find_error_rows(result[api_batch.start: api_batch.params_grad_end_index], api_batch,
+                                     highlight_dict)
+                progress_bar.update(1)
+
+    def find_error_rows(self, result, api_batch, highlight_dict):
+        """找到单个API中需要高亮的行"""
+        if self.mode_config.dump_mode == Const.MD5:
+            return
+        npu_max_index = get_header_index(CompareConst.NPU_MAX, self.mode_config.dump_mode)
+        bench_max_index = get_header_index(CompareConst.BENCH_MAX, self.mode_config.dump_mode)
+        max_diff_index = get_header_index(CompareConst.MAX_DIFF if self.mode_config.dump_mode == Const.SUMMARY
+                                          else CompareConst.MAX_ABS_ERR, self.mode_config.dump_mode)
+
+        red_lines, yellow_lines = [], []
+        LineInfo = namedtuple('LineInfo', ['line_data', 'num_pointer'])
+        ApiInfo = namedtuple('ApiInfo', ['api_input', 'api_output', 'num_pointer'])
+        ColorColumns = namedtuple('ColorColumns', ['red', 'yellow'])
+        color_columns = ColorColumns(red=red_lines, yellow=yellow_lines)
+
+        api_batch_start = api_batch.start  # result_df的input起始全局索引
+        api_batch_params_end_index = api_batch.params_end_index  # result_df的params结束全局索引 + 1
+        api_batch_output_end_index = api_batch.output_end_index  # result_df的output结束全局索引 + 1
+        api_batch_params_slice_index_local = api_batch_params_end_index - api_batch_start  # result的params结束局部切片索引
+        api_batch_output_slice_index_local = api_batch_output_end_index - api_batch_start  # result的output结束局部切片索引
+
+        # 对单行API的输入或输出进行误差判断
+        for i, line in enumerate(result):
+            index = api_batch_start + i
+            line_info = LineInfo(line_data=line, num_pointer=index)
+            for rule in HighlightRules.basic_rules.values():
+                rule.apply(line_info, color_columns, self.mode_config.dump_mode)
+
+        # 对API的输出与输入比较，进行误差判断
+        for n, api_out in enumerate(result[api_batch_params_slice_index_local: api_batch_output_slice_index_local]):
+            index = api_batch_start + api_batch_params_slice_index_local + n
+            # 单行检查只有溢出检查（红色），如果已经溢出，不进一步检查
+            if index in red_lines:
+                continue
+            if not self.check_indices_numeric(api_out, [npu_max_index, bench_max_index, max_diff_index]):
+                continue
 
-    # write header
-    logger.info('Initializing Excel file.')
+            # input/parameters的比较检查, 这里api_in包括input、parameters
+            for api_in in result[0: api_batch_params_slice_index_local]:
+                if not self.check_indices_numeric(api_in, [npu_max_index, bench_max_index, max_diff_index]):
+                    continue
+                api_info = ApiInfo(api_input=api_in, api_output=api_out, num_pointer=index)
+                self.apply_comparison_rules(api_info, color_columns)
+
+        red_lines_num_set = {x[0] for x in red_lines}
+        yellow_lines_num_set = {x[0] for x in yellow_lines}
+        highlight_dict.get('red_rows', set()).update(red_lines_num_set)
+        highlight_dict.get('yellow_rows', set()).update(yellow_lines_num_set - red_lines_num_set)
+        highlight_dict.get('red_lines', []).extend(red_lines)
+        highlight_dict.get('yellow_lines', []).extend(yellow_lines)
+
+    def apply_comparison_rules(self, api_info, color_columns):
+        """output与input/params的比较"""
+        if self.mode_config.dump_mode == Const.SUMMARY:
+            for rule in HighlightRules.summary_compare_rules.values():
+                rule.apply(api_info, color_columns, self.mode_config.dump_mode)
+        else:
+            for rule in HighlightRules.compare_rules.values():
+                rule.apply(api_info, color_columns, self.mode_config.dump_mode)
 
-    handle_multi_process_malicious_value_check(df_malicious_value_check, result_df)
+    def highlight_rows_xlsx(self, result_df, highlight_dict, file_path):
+        """Write and highlight results in Excel"""
 
-    result_df_convert = result_df.applymap(compare_result_df_convert)
+        self.update_highlight_err_msg(result_df, highlight_dict)  # add highlight err_msg
 
-    for row in dataframe_to_rows(result_df_convert, index=False, header=True):
-        ws.append(row)
+        wb = openpyxl.Workbook()
+        ws = wb.active
 
-    # 对可疑数据标色
-    logger.info('Coloring Excel in progress.')
-    col_len = len(result_df.columns)
-    red_fill = PatternFill(
-        start_color=CompareConst.RED, end_color=CompareConst.RED, fill_type="solid"
-    )
-    yellow_fill = PatternFill(
-        start_color=CompareConst.YELLOW, end_color=CompareConst.YELLOW, fill_type="solid",
-    )
-    for i in highlight_dict.get("red_rows", []):
-        for j in range(1, col_len + 1):
-            ws.cell(row=i + 2, column=j).fill = red_fill  # 2因为ws.cell中的row或column需要>=1,数据从第2行开始
-    for i in highlight_dict.get("yellow_rows", []):
-        for j in range(1, col_len + 1):
-            ws.cell(row=i + 2, column=j).fill = yellow_fill
+        # write header
+        logger.info('Initializing Excel file.')
 
-    logger.info('Saving Excel file to disk: %s' % file_path)
-    save_workbook(wb, file_path)
+        self.handle_multi_process_malicious_value_check(self.df_malicious_value_check, result_df)
 
+        result_df_convert = result_df.applymap(self.compare_result_df_convert)
 
-def update_highlight_err_msg(result_df, highlight_dict):
-    if result_df.shape[1] <= 1:
-        return
+        for row in dataframe_to_rows(result_df_convert, index=False, header=True):
+            ws.append(row)
 
-    if CompareConst.NPU_MD5 in result_df.columns:
-        return
+        # 对可疑数据标色
+        logger.info('Coloring Excel in progress.')
+        col_len = len(result_df.columns)
+        red_fill = PatternFill(
+            start_color=CompareConst.RED, end_color=CompareConst.RED, fill_type="solid"
+        )
+        yellow_fill = PatternFill(
+            start_color=CompareConst.YELLOW, end_color=CompareConst.YELLOW, fill_type="solid",
+        )
+        for i in highlight_dict.get("red_rows", []):
+            for j in range(1, col_len + 1):
+                ws.cell(row=i + 2, column=j).fill = red_fill  # 2因为ws.cell中的row或column需要>=1,数据从第2行开始
+        for i in highlight_dict.get("yellow_rows", []):
+            for j in range(1, col_len + 1):
+                ws.cell(row=i + 2, column=j).fill = yellow_fill
 
-    err_msg = result_df.get(CompareConst.ERROR_MESSAGE)
-    red_lines_num_set = highlight_dict.get('red_rows')
+        logger.info('Saving Excel file to disk: %s' % file_path)
+        save_workbook(wb, file_path)
 
-    for color in ['red', 'yellow']:
-        line_key = f'{color}_lines'
-        lines = highlight_dict.get(line_key, [])
-        for line_index, messages in lines:
-            if color == 'yellow' and line_index in red_lines_num_set:
-                continue  # 如果是 yellow 行，且已被 red 行覆盖，跳过
+    def handle_multi_process_malicious_value_check(self, func, result_df):
+        result_total_nums = len(result_df)
+        process_num = int((multiprocessing.cpu_count() + 1) / 2)
 
-            for msg in messages:
-                if err_msg[line_index] == '':
-                    err_msg[line_index] = msg
-                else:
-                    err_msg[line_index] += '\n' + msg
+        if result_total_nums <= process_num:
+            process_num = 1
+            chunks = [result_df]
+        else:
+            chunk_size = result_total_nums // process_num
+            chunks = [result_df.iloc[i: i + chunk_size] for i in range(0, result_total_nums, chunk_size)]
 
-            if color == 'red':
-                red_lines_num_set.add(line_index)
+        pool = multiprocessing.Pool(process_num)
 
-    result_df[CompareConst.ERROR_MESSAGE] = err_msg
+        def err_call(args):
+            logger.error("Multiprocessing malicious value check failed! Reason: {}".format(args))
+            try:
+                pool.close()
+            except OSError:
+                logger.error("Pool terminate failed")
+
+        result_df_columns = result_df.columns.tolist()
+        for column in result_df_columns:
+            self.value_check(column)
+        for df_chunk in chunks:
+            pool.apply_async(func, args=(df_chunk, result_df_columns,), error_callback=err_call)
+
+        pool.close()
+        pool.join()
+
+    def df_malicious_value_check(self, df_chunk, result_df_columns):
+        for row in df_chunk.itertuples(index=False):
+            api_name = row[0]
+            for i, value in enumerate(row):
+                self.value_check(value, api_name, i, result_df_columns)

msprobe/core/compare/layer_mapping/layer_mapping.py

@@ -23,7 +23,7 @@ from msprobe.core.common.utils import (add_time_with_yaml,
                                        get_stack_construct_by_dump_json_path)
 from msprobe.core.compare.layer_mapping.data_scope_parser import get_dump_data_items
 from msprobe.core.compare.utils import read_op, reorder_op_name_list
-
+from msprobe.core.common.decorator import recursion_depth_decorator
 
 
 class LayerTrie:
@@ -71,6 +71,7 @@ class LayerTrie:
         file_path = os.path.join(os.path.realpath(output_path), file_name)
         save_yaml(file_path, result)
 
+    @recursion_depth_decorator("LayerMapping: LayerTrie.convert_to_dict", max_depth=100)
     def convert_to_dict(self, node):
         result = {}
         result["data_item"] = {st: [dt.data_name for dt in dts] for st, dts in node.data_items.items()}
@@ -163,6 +164,8 @@ def preprocess_layer_mapping(mapping):
         for key, value in name_map.items():
             key_list = key.split('.')
             prefix = key_list[0]  # 取前缀
+            value_list = value.split('(')
+            value = value_list[0]  # 取前缀
             key_len = len(key_list)
             if prefix not in final_mapping[type_name]:
                 final_mapping[type_name][prefix] = []

msprobe/core/compare/merge_result/merge_result.py

@@ -1,4 +1,4 @@
-# Copyright (c) 2024-2024, Huawei Technologies Co., Ltd.
+# Copyright (c) 2024-2025, Huawei Technologies Co., Ltd.
 # All rights reserved.
 #
 # Licensed under the Apache License, Version 2.0  (the "License");
@@ -21,7 +21,8 @@ from functools import partial
 import pandas as pd
 from tqdm import tqdm
 
-from msprobe.core.common.file_utils import load_yaml, logger, FileChecker, save_excel, read_xlsx, create_directory
+from msprobe.core.common.file_utils import load_yaml, logger, FileChecker, save_excel, read_xlsx, create_directory, \
+    remove_path
 from msprobe.core.common.const import FileCheckConst, Const, CompareConst
 from msprobe.core.common.utils import CompareException, add_time_with_xlsx
 from msprobe.core.compare.utils import table_value_is_valid
@@ -32,8 +33,8 @@ def check_compare_result_name(file_name):
     """
     check whether the compare result name is as expected
     """
-    single_rank_pattern = r"^compare_result_rank-rank_\d{14}.xlsx$"
-    multi_ranks_pattern = r"^compare_result_rank(\d+)-rank\1_\d{14}.xlsx$"
+    single_rank_pattern = r"^compare_result_(rank|rank-rank)_\d{14}\.xlsx$"
+    multi_ranks_pattern = r"^compare_result_rank(\d+)(?:-rank\1)?_\d{14}\.xlsx$"
     if re.match(multi_ranks_pattern, file_name):
         return True
     if re.match(single_rank_pattern, file_name):
@@ -47,7 +48,7 @@ def reorder_path(compare_result_path_list):
     """
     reorder compare results by rank num
     """
-    rank_pattern = r"compare_result_rank(\d+)-rank"
+    rank_pattern = r"compare_result_rank(\d+)"
     reorder_path_list = sorted(
         compare_result_path_list,
         key=lambda path: int(re.search(rank_pattern, os.path.basename(path)).group(1))
@@ -63,6 +64,7 @@ def get_result_path(input_dir):
                                 for f in os.listdir(input_dir) if f.endswith(FileCheckConst.XLSX_SUFFIX)]
     filt_compare_result_path_list = []
     for file_path in compare_result_path_list:
+        FileChecker(file_path, FileCheckConst.FILE, FileCheckConst.READ_ABLE).common_check()
         file_name = os.path.basename(file_path)
         if check_compare_result_name(file_name):
             compare_result_path_checker = FileChecker(file_path, FileCheckConst.FILE, FileCheckConst.READ_ABLE)
@@ -236,7 +238,7 @@ def handle_multi_process(func, func_args, lock):
     def err_call(args):
         logger.error('Multiprocess merge result failed! Reason: {}'.format(args))
         try:
-            pool.terminate()
+            pool.close()
         except OSError:
             logger.error("Pool terminate failed")
 
@@ -329,6 +331,10 @@ def generate_merge_result(all_compare_index_dict_list, all_rank_num_list, all_co
     for i, df in enumerate(merge_df_list):
         # merge_df_list中df与compare_index_list中compare_index一一对应
         final_result_df_list.append((df, compare_index_list[i]))
+
+    if os.path.exists(output_path):
+        logger.warning(f"{output_path} will be deleted.")
+        remove_path(output_path)
     save_excel(output_path, final_result_df_list)
     logger.info(f"The compare results of the multi-ranks are merged and saved in: {output_path}.")