mindstudio-probe 1.0.4__py3-none-any.whl → 1.1.0__py3-none-any.whl

msprobe/core/common/const.py

@@ -14,11 +14,13 @@ class Const:
     REGEX_PREFIX_MAX_LENGTH = 20
     REGEX_PREFIX_PATTERN = r"^[a-zA-Z0-9_-]+$"
     FILE_PATTERN = r'^[a-zA-Z0-9_./-]+$'
+    STRING_BLACKLIST = r"^[＋－＝％＠\+\-=%@]|;[＋－＝％＠\+\-=%@]"
     COMMA = ","
     FLOAT_EPSILON = np.finfo(float).eps
     OFF = 'OFF'
     BACKWARD = 'backward'
     FORWARD = 'forward'
+    JIT = 'Jit'
     PRIMITIVE_PREFIX = 'Primitive'
     DEFAULT_LIST = []
     DEFAULT_PATH = './'
@@ -30,6 +32,7 @@ class Const:
     FOUR_SEGMENT = 4
     SIX_SEGMENT = 6
     SEVEN_SEGMENT = 7
+    MAX_DEPTH = 10
 
     # dump mode
     ALL = "all"
@@ -78,6 +81,7 @@ class Const:
     RUN_UT = "run_ut"
     GRAD_PROBE = "grad_probe"
     TASK_LIST = [TENSOR, STATISTICS, OVERFLOW_CHECK, FREE_BENCHMARK, RUN_UT, GRAD_PROBE]
+    DUMP_DATA_COLLECTION_LIST = [STATISTICS, TENSOR]
     LEVEL_L0 = "L0"
     LEVEL_L1 = "L1"
     LEVEL_L2 = "L2"
@@ -100,6 +104,30 @@ class Const:
     CUDA_LOWERCASE = 'cuda'
     DISTRIBUTED = 'Distributed'
 
+    # struct json param
+    ORIGIN_DATA = "origin_data"
+    SCOPE = "scope"
+    STACK = "stack"
+
+    ATEN = "Aten"
+    MODULE_WHITE_LIST = ["torch", "numpy"]
+
+    FUNC_SKIP_LIST = ["construct", "__call__"]
+
+    FILE_SKIP_LIST = ["site-packages/mindspore", "package/mindspore", "msprobe", "site-packages/torch", "package/torch"]
+
+    STACK_FILE_INDEX = 0
+
+    STACK_FUNC_INDEX = 2
+
+    STACK_FUNC_ELE_INDEX = 1
+
+    CONSTRUCT_NAME_INDEX = -3
+
+    NAME_FIRST_POSSIBLE_INDEX = -4
+
+    NAME_SECOND_POSSIBLE_INDEX = -5
+
     INPLACE_LIST = [
         "broadcast", "all_reduce", "reduce", "all_gather", "gather", "scatter", "reduce_scatter",
         "_reduce_scatter_base", "_all_gather_base", "send", "recv", "irecv", "isend", "all_to_all_single", "all_to_all",
@@ -114,6 +142,23 @@ class Const:
         "int32_to_int64": ["cross_entropy"]
     }
 
+    FILL_CHAR_NUMS = 50
+    TOOL_ENDS_SUCCESSFULLY = f"{TOOL_NAME} ends successfully."
+    WITHOUT_CALL_STACK = "The call stack retrieval failed."
+    
+    STEP = "step"
+    RANK = "rank"
+    HYPHEN = "-"
+    STEP_RANK_MAXIMUM_RANGE = [int(0), int(1e6)]
+
+    # data type const
+    FLOAT16 = "Float16"
+    FLOAT32 = "Float32"
+    BFLOAT16 = "BFloat16"
+    TORCH_FLOAT16 = "torch.float16"
+    TORCH_FLOAT32 = "torch.float32"
+    TORCH_BFLOAT16 = "torch.bfloat16"
+
 
 class CompareConst:
     """
@@ -159,6 +204,7 @@ class CompareConst:
     INPUT_STRUCT = "input_struct"
     OUTPUT_STRUCT = "output_struct"
     SUMMARY = "summary"
+    MAX_EXCEL_LENGTH = 1048576
 
     COMPARE_RESULT_HEADER = [
         NPU_NAME, BENCH_NAME, NPU_DTYPE, BENCH_DTYPE, NPU_SHAPE, BENCH_SHAPE, COSINE, MAX_ABS_ERR, MAX_RELATIVE_ERR,
@@ -197,6 +243,8 @@ class CompareConst:
     ERROR = 'error'
     SKIP = 'SKIP'
     N_A = 'N/A'
+    INF = 'inf'
+    NEG_INF = '-inf'
     BFLOAT16_MIN = -3.3895313892515355e+38
     BFLOAT16_MAX = 3.3895313892515355e+38
     BFLOAT16_EPS = 3.90625e-3  # 2 ** -8
@@ -274,7 +322,8 @@ class FileCheckConst:
     MAX_JSON_SIZE = 1073741824  # 1 * 1024 * 1024 * 1024
     MAX_PT_SIZE = 10737418240  # 10 * 1024 * 1024 * 1024
     MAX_CSV_SIZE = 1073741824  # 1 * 1024 * 1024 * 1024
-    MAX_YAML_SIZE = 1048576  # 10 * 1024 * 1024
+    MAX_YAML_SIZE = 1048576  # 1 * 1024 * 1024
+    COMMOM_FILE_SIZE = 1048576  # 1 * 1024 * 1024
     DIR = "dir"
     FILE = "file"
     DATA_DIR_AUTHORITY = 0o750
@@ -287,6 +336,7 @@ class FileCheckConst:
         CSV_SUFFIX: MAX_CSV_SIZE,
         YAML_SUFFIX: MAX_YAML_SIZE
     }
+    CSV_BLACK_LIST = r'^[＋－＝％＠\+\-=%@]|;[＋－＝％＠\+\-=%@]'
 
 
 class OverflowConst:
@@ -329,11 +379,22 @@ class MsgConst:
     """
     Class for log messages const
     """
-    CLEAR_SYMBOL = "\033[K"
     MSPROBE_LOG_LEVEL = "MSPROBE_LOG_LEVEL"
-    LEVEL = ["INFO", "WARNING", "ERROR", "DEBUG"]
+    LOG_LEVEL_ENUM = ["0", "1", "2", "3", "4"]
+    LOG_LEVEL = ["DEBUG", "INFO", "WARNING", "ERROR"]
+    class LogLevel:
+        class DEBUG:
+            value = 0
+        class INFO:
+            value = 1
+        class WARNING:
+            value = 2
+        class ERROR:
+            value = 3
     SPECIAL_CHAR = ["\n", "\r", "\u007F", "\b", "\f", "\t", "\u000B", "%08", "%0a", "%0b", "%0c", "%0d", "%7f"]
 
+    NOT_CREATED_INSTANCE = "PrecisionDebugger instance is not created."
+
 
 class GraphMode:
     NPY_MODE = "NPY_MODE"

msprobe/core/common/exceptions.py

@@ -13,8 +13,8 @@ class MsprobeException(CodedException):
     OVERFLOW_NUMS_ERROR = 1
 
     err_strs = {
-        INVALID_PARAM_ERROR: "[msprobe] 无效参数： ",
-        OVERFLOW_NUMS_ERROR: "[msprobe] 超过预设溢出次数 当前溢出次数:"
+        INVALID_PARAM_ERROR: "[msprobe] 无效参数：",
+        OVERFLOW_NUMS_ERROR: "[msprobe] 超过预设溢出次数 当前溢出次数："
     }
 
 

msprobe/core/common/file_utils.py

@@ -22,6 +22,7 @@ import re
 import shutil
 import yaml
 import numpy as np
+import pandas as pd
 
 from msprobe.core.common.log import logger
 from msprobe.core.common.exceptions import FileCheckException
@@ -187,7 +188,7 @@ def check_other_user_writable(path):
 
 def check_path_owner_consistent(path):
     file_owner = os.stat(path).st_uid
-    if file_owner != os.getuid():
+    if file_owner != os.getuid() and os.getuid() != 0:
         logger.error('The file path %s may be insecure because is does not belong to you.' % path)
         raise FileCheckException(FileCheckException.FILE_PERMISSION_ERROR)
 
@@ -214,7 +215,9 @@ def check_common_file_size(file_path):
         for suffix, max_size in FileCheckConst.FILE_SIZE_DICT.items():
             if file_path.endswith(suffix):
                 check_file_size(file_path, max_size)
-                break
+                return
+        check_file_size(file_path, FileCheckConst.COMMOM_FILE_SIZE)
+        
 
 
 def check_file_suffix(file_path, file_suffix):
@@ -322,7 +325,7 @@ def check_file_type(path):
     elif os.path.isfile(path):
         return FileCheckConst.FILE
     else:
-        logger.error('Neither a file nor a directory.')
+        logger.error(f'{path} does not exist, please check!')
         raise FileCheckException(FileCheckException.INVALID_FILE_ERROR)
 
 
@@ -338,10 +341,10 @@ def load_yaml(yaml_path):
     return yaml_data
 
 
-def load_npy(filepath, enable_pickle=False):
+def load_npy(filepath):
     check_file_or_directory_path(filepath)
     try:
-        npy = np.load(filepath, allow_pickle=enable_pickle)
+        npy = np.load(filepath)
     except Exception as e:
         logger.error(f"The numpy file failed to load. Please check the path: {filepath}.")
         raise RuntimeError(f"Load numpy file {filepath} failed.") from e
@@ -374,6 +377,20 @@ def save_json(json_path, data, indent=None):
     change_mode(json_path, FileCheckConst.DATA_FILE_AUTHORITY)
 
 
+def save_yaml(yaml_path, data):
+    yaml_path = os.path.realpath(yaml_path)
+    check_path_before_create(yaml_path)
+    try:
+        with FileOpen(yaml_path, 'w') as f:
+            fcntl.flock(f, fcntl.LOCK_EX)
+            yaml.dump(data, f, sort_keys=False)
+            fcntl.flock(f, fcntl.LOCK_UN)
+    except Exception as e:
+        logger.error(f'Save yaml file "{os.path.basename(yaml_path)}" failed.')
+        raise RuntimeError(f"Save yaml file {yaml_path} failed.") from e
+    change_mode(yaml_path, FileCheckConst.DATA_FILE_AUTHORITY)
+
+
 def move_file(src_path, dst_path):
     check_file_or_directory_path(src_path)
     check_path_before_create(dst_path)
@@ -396,9 +413,9 @@ def save_npy(data, filepath):
     change_mode(filepath, FileCheckConst.DATA_FILE_AUTHORITY)
 
 
-def save_npy_to_txt(self, data, dst_file='', align=0):
+def save_npy_to_txt(data, dst_file='', align=0):
     if os.path.exists(dst_file):
-        self.log.info("Dst file %s exists, will not save new one.", dst_file)
+        logger.info("Dst file %s exists, will not save new one." % dst_file)
         return
     shape = data.shape
     data = data.flatten()
@@ -411,7 +428,7 @@ def save_npy_to_txt(self, data, dst_file='', align=0):
     try:
         np.savetxt(dst_file, data.reshape((-1, align)), delimiter=' ', fmt='%g')
     except Exception as e:
-        self.log.error("An unexpected error occurred: %s when savetxt to %s" % (str(e)), dst_file)
+        logger.error("An unexpected error occurred: %s when savetxt to %s" % (str(e), dst_file))
     change_mode(dst_file, FileCheckConst.DATA_FILE_AUTHORITY)
 
 
@@ -431,7 +448,25 @@ def save_workbook(workbook, file_path):
     change_mode(file_path, FileCheckConst.DATA_FILE_AUTHORITY)
 
 
-def write_csv(data, filepath, mode="a+"):
+def write_csv(data, filepath, mode="a+", malicious_check=False):
+    def csv_value_is_valid(value: str) -> bool:
+        if not isinstance(value, str):
+            return True  
+        try:
+            # -1.00 or +1.00 should be consdiered 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
+    
+    if malicious_check:
+        for row in data:
+            for cell in row:
+                if not csv_value_is_valid(cell):
+                    raise RuntimeError(f"Malicious value [{cell}] is not allowed " \
+                                       f"to be written into the csv: {filepath}.")
+
     file_path = os.path.realpath(filepath)
     check_path_before_create(filepath)
     try:
@@ -444,6 +479,16 @@ def write_csv(data, filepath, mode="a+"):
     change_mode(filepath, FileCheckConst.DATA_FILE_AUTHORITY)
 
 
+def read_csv(filepath):
+    check_file_or_directory_path(filepath)
+    try:
+        csv_data = pd.read_csv(filepath)
+    except Exception as e:
+        logger.error(f"The csv file failed to load. Please check the path: {filepath}.")
+        raise RuntimeError(f"Read csv file {filepath} failed.") from e
+    return csv_data
+
+
 def remove_path(path):
     if not os.path.exists(path):
         return

msprobe/core/common/inplace_op_checker.py

@@ -0,0 +1,38 @@
+import os
+from msprobe.core.common.file_utils import load_yaml
+
+
+class InplaceOpChecker:
+    OP_FUNCTIONAL = 'functional'
+    OP_TENSOR = 'tensor'
+    OP_TORCH = 'torch'
+    OP_DISTRIBUTED = 'distributed'
+
+    INPLACE_OPS_DICT = None
+
+    @classmethod
+    def load_ops(cls):
+        if cls.INPLACE_OPS_DICT is None:
+            cls.INPLACE_OPS_DICT = dict()
+        cur_path = os.path.dirname(os.path.realpath(__file__))
+        yaml_path = os.path.join(cur_path, "inplace_ops.yaml")
+        all_ops = load_yaml(yaml_path)
+        cls.INPLACE_OPS_DICT[cls.OP_FUNCTIONAL] = all_ops.get('inplace_functional_op')
+        cls.INPLACE_OPS_DICT[cls.OP_TENSOR] = all_ops.get('inplace_tensor_op')
+        cls.INPLACE_OPS_DICT[cls.OP_TORCH] = all_ops.get('inplace_torch_op')
+        cls.INPLACE_OPS_DICT[cls.OP_DISTRIBUTED] = all_ops.get('inplace_distributed_op')
+
+    @classmethod
+    def check(cls, api, category='distributed'):
+        """
+            给定api和分类，检查其是否为inplace操作
+        """
+        if not cls.INPLACE_OPS_DICT:
+            cls.load_ops()
+
+        if category not in cls.INPLACE_OPS_DICT.keys():
+            return False
+        return api in cls.INPLACE_OPS_DICT[category]
+
+
+InplaceOpChecker.load_ops()

msprobe/core/common/inplace_ops.yaml

@@ -0,0 +1,251 @@
+inplace_functional_op:
+  - threshold_
+  - relu_
+  - hardtanh_
+  - elu_
+  - selu_
+  - celu_
+  - leaky_relu_
+  - rrelu_
+
+inplace_tensor_op:
+  - __iadd__
+  - __iand__
+  - __idiv__
+  - __ifloordiv__
+  - __ilshift__
+  - __imod__
+  - __imul__
+  - __ior__
+  - __irshift__
+  - __isub__
+  - __ixor__
+  - abs_
+  - absolute_
+  - acos_
+  - acosh_
+  - add_
+  - addbmm_
+  - addcdiv_
+  - addcmul_
+  - addmm_
+  - addmv_
+  - addr_
+  - arccos_
+  - arccosh_
+  - arcsin_
+  - arcsinh_
+  - arctan_
+  - arctanh_
+  - asin_
+  - asinh_
+  - atan2_
+  - atan_
+  - atanh_
+  - baddbmm_
+  - bernoulli_
+  - bitwise_and_
+  - bitwise_not_
+  - bitwise_or_
+  - bitwise_xor_
+  - cauchy_
+  - ceil_
+  - clamp_
+  - clamp_max_
+  - clamp_min_
+  - clip_
+  - copysign_
+  - cos_
+  - cosh_
+  - cumprod_
+  - cumsum_
+  - deg2rad_
+  - digamma_
+  - div_
+  - divide_
+  - eq_
+  - erf_
+  - erfc_
+  - erfinv_
+  - exp2_
+  - exp_
+  - expm1_
+  - exponential_
+  - fill_
+  - fill_diagonal_
+  - fix_
+  - float_power_
+  - floor_
+  - floor_divide_
+  - fmod_
+  - frac_
+  - gcd_
+  - ge_
+  - geometric_
+  - greater_
+  - gt_
+  - greater_equal_
+  - heaviside_
+  - hypot_
+  - igamma_
+  - igammac_
+  - index_add_
+  - index_copy_
+  - index_fill_
+  - index_put_
+  - lcm_
+  - ldexp_
+  - le_
+  - lerp_
+  - less_
+  - less_equal_
+  - lgamma_
+  - log10_
+  - log1p_
+  - log2_
+  - log_
+  - log_normal_
+  - logical_and_
+  - logical_not_
+  - logical_or_
+  - logical_xor_
+  - logit_
+  - lt_
+  - map2_
+  - map_
+  - masked_fill_
+  - masked_scatter_
+  - mul_
+  - multiply_
+  - mvlgamma_
+  - ne_
+  - neg_
+  - negative_
+  - normal_
+  - not_equal_
+  - pow_
+  - polygamma_
+  - put_
+  - rad2deg_
+  - reciprocal_
+  - relu_
+  - remainder_
+  - renorm_
+  - resize_
+  - resize_as_
+  - round_
+  - rsqrt_
+  - scatter_
+  - scatter_add_
+  - sgn_
+  - sigmoid_
+  - sign_
+  - sin_
+  - sinc_
+  - sinh_
+  - sqrt_
+  - square_
+  - squeeze_
+  - sub_
+  - t_
+  - tan_
+  - tanh_
+  - transpose_
+  - tril_
+  - triu_
+  - true_divide_
+  - trunc_
+  - unsqueeze_
+  - xlogy_
+
+inplace_torch_op:
+  - _add_relu_
+  - abs_
+  - acos_
+  - acosh_
+  - addmv_
+  - alpha_dropout_
+  - arccos_
+  - arccosh_
+  - arcsin_
+  - arcsinh_
+  - arctan_
+  - arctanh_
+  - asin_
+  - asinh_
+  - atan_
+  - atanh_
+  - ceil_
+  - celu_
+  - clamp_
+  - clamp_max_
+  - clamp_min_
+  - clip_
+  - cos_
+  - cosh_
+  - deg2rad_
+  - dropout_
+  - embedding_renorm_
+  - erf_
+  - erfc_
+  - exp2_
+  - exp_
+  - expm1_
+  - feature_alpha_dropout_
+  - feature_dropout_
+  - fill_
+  - fix_
+  - floor_
+  - frac_
+  - gcd_
+  - index_put_
+  - lcm_
+  - ldexp_
+  - log10_
+  - log1p_
+  - log2_
+  - log_
+  - logit_
+  - nan_to_num_
+  - neg_
+  - negative_
+  - rad2deg_
+  - reciprocal_
+  - relu_
+  - resize_as_
+  - round_
+  - rrelu_
+  - rsqrt_
+  - selu_
+  - sigmoid_
+  - sin_
+  - sinc_
+  - sinh_
+  - sqrt_
+  - square_
+  - tan_
+  - tanh_
+  - threshold_
+  - trunc_
+  - xlogy_
+
+inplace_distributed_op:
+  - broadcast
+  - all_reduce
+  - reduce
+  - all_gather
+  - gather
+  - scatter
+  - reduce_scatter
+  - _reduce_scatter_base
+  - _all_gather_base
+  - send
+  - recv
+  - irecv
+  - isend
+  - all_to_all_single
+  - all_to_all
+  - all_gather_into_tensor
+  - reduce_scatter_tensor
+
+

msprobe/core/common/log.py

@@ -4,13 +4,19 @@ import sys
 from functools import wraps
 from msprobe.core.common.const import MsgConst
 
-MSPROBE_LOG_LEVEL = os.environ.get(MsgConst.MSPROBE_LOG_LEVEL, "")
-
 
 class BaseLogger:
     def __init__(self):
         self.rank = None
+        self.level = self.get_level()
 
+    @staticmethod
+    def get_level():
+        input_level = os.environ.get(MsgConst.MSPROBE_LOG_LEVEL)
+        if input_level not in MsgConst.LOG_LEVEL_ENUM:
+            return MsgConst.LogLevel.INFO.value
+        else:
+            return int(input_level)
 
     def get_rank(self):
         return self.rank
@@ -22,23 +28,26 @@ class BaseLogger:
                 msg = msg.replace(char, '_')
             return func(self, msg, **kwargs)
         return func_level
-
-    @filter_special_chars
-    def info(self, msg, **kwargs):
-        self._print_log(MsgConst.LEVEL[0], msg, **kwargs)
-
+    
     @filter_special_chars
     def error(self, msg):
-        self._print_log(MsgConst.LEVEL[2], msg)
+        if self.level <= MsgConst.LogLevel.ERROR.value:
+            self._print_log(MsgConst.LOG_LEVEL[3], msg)
 
     @filter_special_chars
     def warning(self, msg):
-        self._print_log(MsgConst.LEVEL[1], msg)
+        if self.level <= MsgConst.LogLevel.WARNING.value:
+            self._print_log(MsgConst.LOG_LEVEL[2], msg)
+
+    @filter_special_chars
+    def info(self, msg):
+        if self.level <= MsgConst.LogLevel.INFO.value:
+            self._print_log(MsgConst.LOG_LEVEL[1], msg)
 
     @filter_special_chars
     def debug(self, msg):
-        if MSPROBE_LOG_LEVEL == MsgConst.LEVEL[3]:
-            self._print_log(MsgConst.LEVEL[3], msg)
+        if self.level <= MsgConst.LogLevel.DEBUG.value:
+            self._print_log(MsgConst.LOG_LEVEL[0], msg)
 
     def on_rank_0(self, func):
         def func_rank_0(*args, **kwargs):
@@ -73,4 +82,5 @@ class BaseLogger:
         print(full_msg, end=end)
         sys.stdout.flush()
 
+
 logger = BaseLogger()