mindspore 2.2.14__cp39-cp39-win_amd64.whl → 2.3.0__cp39-cp39-win_amd64.whl

mindspore/parallel/parameter_broadcast.py

@@ -0,0 +1,152 @@
+# Copyright 2024 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.
+# ============================================================================
+"""Parameter broadcast"""
+from __future__ import absolute_import
+
+__all__ = ["parameter_broadcast"]
+
+import numpy as np
+
+
+def parameter_broadcast(net, layout, cur_rank=0, initial_rank=0):
+    """
+    Broadcast parameter to other rank in data parallel dimension.
+
+    .. warning::
+        This is an experimental API that is subject to change or deletion.
+
+    Args:
+        net (Cell): The network where the parameters will be broadcasted.
+        layout (Dict): Parameter layout dictionary. Come from
+            :func:`mindspore.nn.Cell.parameter_layout_dict`
+            or read from file(for example: "strategy.ckpt" saved by using the
+            `strategy_ckpt_config` parameter of :func:`mindspore.set_auto_parallel_context`).
+            The key is param name, the value is the layout of this parameter.
+        cur_rank (int, optional): current rank id. Default: ``0``.
+        initial_rank (int, optional): Start rank id for each pipeline. Default: ``0``.
+
+    Raises:
+        ValueError: `cur_rank` is not rank id of current rank.
+        ValueError: `initial_rank` is not the start rank id of current pipeline stage.
+        ValueError: Parameter name in `layout` can not be found in
+            :func:`mindspore.nn.Cell.parameters_dict`.
+
+    Examples:
+        >>> import os
+        >>> import mindspore as ms
+        >>> import mindspore.dataset as ds
+        >>> from mindspore import nn, ops
+        >>> from mindspore.communication import init
+        >>> from mindspore.common.initializer import initializer
+        >>> from mindspore.train import Model
+        >>> from mindspore.parallel.parameter_broadcast import parameter_broadcast
+        >>> from mindspore.train.serialization import load_checkpoint, load_param_into_net
+        >>> ms.set_context(mode=ms.GRAPH_MODE)
+        >>> ms.set_context(max_device_memory="28GB")
+        >>> ms.set_auto_parallel_context(parallel_mode=ms.ParallelMode.SEMI_AUTO_PARALLEL)
+        >>> init()
+        >>> ms.set_seed(1)
+        >>> class Network(nn.Cell):
+        ...     def __init__(self):
+        ...         super().__init__()
+        ...         self.flatten = ops.Flatten()
+        ...         self.fc1_weight = ms.Parameter(initializer("normal", [28*28, 512], ms.float32))
+        ...         self.fc2_weight = ms.Parameter(initializer("normal", [512, 512], ms.float32))
+        ...         self.fc3_weight = ms.Parameter(initializer("normal", [512, 10], ms.float32))
+        ...         self.matmul1 = ops.MatMul()
+        ...         self.relu1 = ops.ReLU()
+        ...         self.matmul2 = ops.MatMul()
+        ...         self.relu2 = ops.ReLU()
+        ...         self.matmul3 = ops.MatMul()
+        ...     def construct(self, x):
+        ...         x = self.flatten(x)
+        ...         x = self.matmul1(x, self.fc1_weight)
+        ...         x = self.relu1(x)
+        ...         x = self.matmul2(x, self.fc2_weight)
+        ...         x = self.relu2(x)
+        ...         logits = self.matmul3(x, self.fc3_weight)
+        ...         return logits
+        >>> net = Network()
+        >>> net.matmul1.shard(((2, 4), (4, 1)))
+        >>> net.relu1.shard(((4, 1),))
+        >>> net.matmul2.shard(((1, 8), (8, 1)))
+        >>> net.relu2.shard(((8, 1),))
+        >>> # Create the dataset taking MNIST as an example. Refer to
+        >>> # https://gitee.com/mindspore/docs/blob/master/docs/mindspore/code/mnist.py
+        >>> dataset = create_dataset()
+        >>> optim = nn.SGD(net.trainable_params(), 1e-2)
+        >>> loss = nn.CrossEntropyLoss()
+        >>> model = Model(net, loss_fn=loss, optimizer=optim)
+        >>> model.train(1, dataset)
+        >>> ms.save_checkpoint(net, "./simple.ckpt", False)
+        >>> layout = model.train_network.parameter_layout_dict
+        >>> param_dict = load_checkpoint("./simple.ckpt")
+        >>> load_param_into_net(net, param_dict)
+        >>> rank_id = os.environ["RANK_ID"]
+        >>> parameter_broadcast(model.train_network, layout, int(rank_id), 0)
+        >>> class LossCallBack(Callback):
+        ...     def step_end(self, run_context):
+        ...         cb_params = run_context.original_args()
+        ...         print("step end, cur step num: ", cb_params.cur_step_num, flush=True)
+        >>> model.train(1, dataset, callbacks=[LossCallBack()])
+    """
+    if not layout:
+        return
+    import mindspore as ms
+    from mindspore import Tensor
+    from mindspore.communication import get_rank, create_group, get_group_size
+    from mindspore.train._utils import get_parameter_redundancy, remove_param_redundancy
+    from mindspore.nn.wrap.cell_wrapper import AllreduceGraph
+    origin_parallel_mode = ms.get_auto_parallel_context("parallel_mode")
+    if origin_parallel_mode not in ("semi_auto_parallel", "auto_parallel"):
+        return
+    if cur_rank != get_rank():
+        raise ValueError(f"For parameter broadcast, the cur_rank: {cur_rank} is wrong.")
+    if initial_rank % (get_group_size() / ms.get_auto_parallel_context("pipeline_stages")) != 0:
+        raise ValueError(f"For parameter broadcast, the initial_rank: {initial_rank} is wrong.")
+    param_redundancy = get_parameter_redundancy(layout, initial_rank)
+    if not param_redundancy:
+        return
+    single_params = remove_param_redundancy(param_redundancy)
+    if not single_params:
+        return
+    param_redundancy_reversed = {}
+    for key, redundancy in param_redundancy.items():
+        for item in redundancy:
+            if len(item) == 1:
+                continue
+            if cur_rank in item:
+                param_redundancy_reversed.setdefault(item, []).append(key)
+    if not param_redundancy_reversed:
+        return
+    if cur_rank not in single_params:
+        return
+    net_param_dict = net.parameters_dict()
+    ms.set_auto_parallel_context(parallel_mode="hybrid_parallel")
+    for group, params in param_redundancy_reversed.items():
+        create_group(str(group), list(group))
+        allreduce_input = []
+        for param in params:
+            if param not in net_param_dict:
+                raise ValueError(f"For parameter broadcast, the param: {param} can not be found.")
+            real_param = net_param_dict[param]
+            if param not in single_params[cur_rank]:
+                real_param.set_data(Tensor(np.zeros(real_param.shape), dtype=real_param.dtype))
+            allreduce_input.append(real_param)
+        if not allreduce_input:
+            continue
+        allreduce_graph = AllreduceGraph(allreduce_input, str(group))
+        allreduce_graph()
+    ms.set_auto_parallel_context(parallel_mode=origin_parallel_mode)

mindspore/parallel/shard.py

@@ -1,4 +1,4 @@
-# Copyright 2022 Huawei Technologies Co., Ltd
+# Copyright 2023 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.
@@ -12,13 +12,125 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 # ============================================================================
-"""pynative shard"""
+"""shard"""
 
+import copy
 import mindspore as ms
 from mindspore import log as logger
 from mindspore._c_expression import Shard_
 
 
+class Layout():
+    """
+    Parallel layout describes the detailed sharding information.
+
+    Note:
+        - It is valid only in semi auto parallel or auto parallel mode.
+        - The multiplication result of the `device_matrix` must be equal to the device count in a pipeline stage.
+        - When the layout function is invoked to constructs a sharding strategy, each alias name is only allowed to be
+          used once to shard a tensor.
+
+    Args:
+        device_matrix (tuple): Describe the shape of devices arrangement, its element type is int.
+        alias_name (tuple): The alias name for each axis of device_matrix, its length shoits element type is string.
+                            When using "interleaved_parallel" as an alias name, the tensor would be split into multiple
+                            copies on the corresponding partition dimension on a single card. The corresponding value
+                            of "interleaved_parallel" in device_matrix must be 2.
+
+    Raises:
+        TypeError: `device_matrix` is not a tuple type.
+        TypeError: `alias_name` is not a tuple type.
+        ValueError: `device_matrix` length is not equal to `alias_name` length.
+        TypeError: The element of `device_matrix` is not int type.
+        TypeError: The element of `alias_name` is not a str type.
+        ValueError: The element of `alias_name` is an empty str.
+        ValueError: The element of `alias_name` is "None".
+        ValueError: `alias_name` contains repeated element.
+
+    Examples:
+        >>> from mindspore import Layout
+        >>> layout = Layout((2, 2, 2), ("dp", "sp", "mp"))
+        >>> layout0 = layout("dp", "mp")
+        >>> print(layout0.to_dict())
+        {"device_matrix": (2, 2, 2), "tensor_map": (2, 0)}
+        >>> # Total device num is 4, but split the tensor in local device into two copies.
+        >>> layout = Layout((2, 2, 2), ("dp", "sp", "interleaved_parallel"))
+        >>> layout1 = layout(("dp", "interleaved_parallel"), "sp")
+    """
+
+    def __init__(self, device_matrix, alias_name):
+        if not isinstance(device_matrix, tuple):
+            raise TypeError(f'device_matrix must be tuple type, but got:{type(device_matrix)}')
+        if not isinstance(alias_name, tuple):
+            raise TypeError(f'alias_name must be tuple type, but got:{type(alias_name)}')
+        if len(device_matrix) != len(alias_name):
+            raise ValueError(f'device_matrix length should be equal to alias_name length')
+        for in_ele in device_matrix:
+            if not isinstance(in_ele, int):
+                raise TypeError(f'The element of device_matrix must be int type, but got:{type(in_ele)}')
+        for in_ele in alias_name:
+            if not isinstance(in_ele, str):
+                raise TypeError(f'The element of alias_name must be str type, but got:{type(in_ele)}')
+            if not in_ele:
+                raise ValueError(f"The element of alias_name can not be empty.")
+            if in_ele == "None":
+                raise ValueError(f"The element of alias_name can not set 'None', because 'None' means no sharding.")
+        if len(set(alias_name)) != len(alias_name):
+            raise ValueError(f'Each element of alias_name {alias_name} should be different')
+        inter_key = "interleaved_parallel"
+        if inter_key in alias_name and alias_name.index(inter_key) != len(alias_name) - 1:
+            raise ValueError(f"When alias_name {alias_name} contains keyword 'interleaved_parallel',"
+                             f" it should be at the last dim of alias_name, which means the virtual sharding.")
+        if inter_key in alias_name and device_matrix[alias_name.index(inter_key)] != 2:
+            raise ValueError(f"When alias_name {alias_name} contains keyword 'interleaved_parallel',"
+                             f" the corresponding dim of device_matrix should be 2.")
+        self._device_shape = device_matrix
+        self._alias_name = alias_name
+        self._tensor_map = None
+
+    def __call__(self, *tensor_map):
+        self._tensor_map = ()
+        writed_map = ()
+        for ele in tensor_map:
+            if isinstance(ele, tuple):
+                ele_map = ()
+                for item in ele:
+                    if item == "None":
+                        ele_map += (-1,)
+                        continue
+                    if item not in self._alias_name:
+                        raise ValueError(f'The axis {item} is not found in {self._alias_name}')
+                    if item in writed_map:
+                        raise ValueError(f'The axis {item} has been set more than one in {self._alias_name}')
+                    ele_map += (len(self._alias_name) - 1 - self._alias_name.index(item),)
+                    writed_map += (item,)
+                self._tensor_map += (ele_map,)
+                continue
+            if ele == "None":
+                self._tensor_map += (-1,)
+                continue
+            if ele not in self._alias_name:
+                raise ValueError(f'The axis {ele} is not found in {self._alias_name}')
+            if ele in writed_map:
+                raise ValueError(f'The axis {ele} has been set more than one in {self._alias_name}')
+            self._tensor_map += (len(self._alias_name) - 1 - self._alias_name.index(ele),)
+            writed_map += (ele,)
+        return copy.deepcopy(self)
+
+    def to_dict(self):
+        """
+        Transform layout to a dictionary.
+        """
+        if self._device_shape is None:
+            raise ValueError("The device_shape of layout is None")
+        if self._tensor_map is None:
+            raise ValueError("The tensor_map of layout is None")
+        interleaved_parallel = "interleaved_parallel" in self._alias_name
+        return {"device_matrix": self._device_shape, "tensor_map": self._tensor_map,
+                "interleaved_parallel": interleaved_parallel}
+
+
+
 class Shard(Shard_):
     """Shard operation"""
 
@@ -34,16 +146,17 @@ class Shard(Shard_):
         self.level = None
 
     def __call__(self, fn, in_strategy, out_strategy=None, parameter_plan=None, device="Ascend", level=0):
-        if ms.context.get_context("mode") != ms.context.PYNATIVE_MODE or \
-                ms.context.get_auto_parallel_context("parallel_mode") not in ["auto_parallel"]:
+        parallel_mode = ms.context.get_auto_parallel_context("parallel_mode")
+        if parallel_mode not in ["auto_parallel", "semi_auto_parallel"]:
             raise AssertionError(
-                f"Cell shard only supports auto parallel under PyNative mode.")
+                f"Cell shard only supports auto parallel and semi auto parallel.")
         if ms.context.get_context("device_target") not in ["Ascend", "GPU"]:
             raise AssertionError(
                 f"'Shard' now only supports 'Ascend' and 'GPU'")
-        if ms.context.get_auto_parallel_context("search_mode") != "sharding_propagation":
-            raise AssertionError(
-                f"'search_mode' must be 'sharding_propagation' for 'Shard'")
+        if parallel_mode == "auto_parallel" and \
+            ms.context.get_auto_parallel_context("search_mode") != "sharding_propagation":
+            raise AssertionError(f"'search_mode' must be 'sharding_propagation' for 'Shard' when the "
+                                 f"'parallel_mode' is 'auto_parallel.'")
         if not isinstance(in_strategy, tuple):
             raise TypeError(
                 f"For 'Shard', the 'in_strategy' should be a tuple, but got {type(in_strategy).__name__}")
@@ -156,8 +269,9 @@ class Shard(Shard_):
 def shard(fn, in_strategy, out_strategy=None, parameter_plan=None, device="Ascend", level=0):
     """
     Defining the input and output layouts of this cell and the parallel strategies of remaining ops will be
-    generated by sharding propagation. In PyNative mode, use this method
-    to specify a Cell for distributed execution in graph mode.
+    generated by sharding propagation. In PyNative mode, use this method to specify a Cell for distributed
+    execution in graph mode. In Graph mode, use this method to specify distribution strategy for a Cell,
+    strategy for others will be set by sharding propagation.
     in_strategy and out_strategy define the input and output layout respectively.
     in_strategy/out_strategy should be a tuple, each element of which corresponds to the desired layout of
     this input/output, and None represents data_parallel,
@@ -165,9 +279,8 @@ def shard(fn, in_strategy, out_strategy=None, parameter_plan=None, device="Ascen
     The parallel strategies of remaining operators are derived from the strategy specified by the input and output.
 
     Note:
-        You need to set the execution mode to PyNative mode,
-        set the parallel mode in `set_auto_parallel_context` (parallel_mode) to "auto_parallel"
-        and the search mode (search_mode) to "sharding_propagation".
+        If ms.shard is called, the parallel mode in `set_auto_parallel_context` (parallel_mode) will be set to
+        "auto_parallel" and the search mode (search_mode) to "sharding_propagation".
         If the input contain Parameter, its strategy should be set in `in_strategy`.
 
     Args:
@@ -197,9 +310,7 @@ def shard(fn, in_strategy, out_strategy=None, parameter_plan=None, device="Ascen
         Function, return the function that will be executed under auto parallel process.
 
     Raises:
-        AssertionError: If execute mode is not PYNATIVE_MODE.
-        AssertionError: If parallel mode is not "auto_parallel".
-        AssertionError: If search_mode it not "sharding_propagation".
+        AssertionError: If parallel mode is not "auto_parallel" nor "semi_auto_parallel".
         AssertionError: If device_target it not "Ascend" or "GPU".
         TypeError: If `in_strategy` is not a tuple.
         TypeError: If `out_strategy` is not a tuple or None.
@@ -231,7 +342,7 @@ def shard(fn, in_strategy, out_strategy=None, parameter_plan=None, device="Ascen
 
     Tutorial Examples:
         - `Functional Operator Sharding
-          <https://www.mindspore.cn/tutorials/experts/en/r2.2/parallel/pynative_shard_function_parallel.html>`_
+          <https://www.mindspore.cn/tutorials/experts/en/master/parallel/shard_function_parallel.html>`_
     """
     if not isinstance(fn, (ms.nn.Cell)):
         logger.warning("'fn' is not a mindspore.nn.Cell, and its definition cannot involve Parameter; "

mindspore/profiler/__init__.py

@@ -20,7 +20,8 @@ Users can visualize the results using the MindInsight tool.
 Now, Profiler supports AICORE operator, AICPU operator, HostCPU operator, memory,
 correspondence, cluster, etc data analysis.
 """
+__all__ = ["Profiler", "EnvProfiler", "ProfilerLevel"]
+
 from mindspore.profiler.profiling import Profiler
 from mindspore.profiler.envprofiling import EnvProfiler
-
-__all__ = ["Profiler", "EnvProfiler"]
+from mindspore.profiler.profiling import ProfilerLevel

mindspore/profiler/common/process_pool.py

@@ -0,0 +1,41 @@
+# Copyright 2022-2024 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.
+# ============================================================================
+"""Analyze profiling data asynchronously by asynchronous process"""
+import atexit
+from typing import List
+from multiprocessing import Process
+
+from mindspore.profiler.common.singleton import Singleton
+
+
+@Singleton
+class MultiProcessPool:
+    """A ProcessPool to run task asynchronously"""
+
+    def __init__(self) -> None:
+        self.porcess_list: List[Process] = []
+        atexit.register(self.wait_all_job_finished)
+
+    def add_async_job(self, func):
+        """Add job and run in subprocess"""
+        process = Process(target=func)
+        process.start()
+        self.porcess_list.append(process)
+
+    def wait_all_job_finished(self):
+        """Wait all subprocess finished"""
+        for process in self.porcess_list:
+            process.join()
+        self.porcess_list = []

mindspore/profiler/common/singleton.py

@@ -0,0 +1,28 @@
+# Copyright 2022-2024 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.
+# ============================================================================
+"""A singleton implement in python"""
+
+
+class Singleton:
+    """A singleton implement"""
+
+    def __init__(self, cls):
+        self._cls = cls
+        self._instance = {}
+
+    def __call__(self):
+        if self._cls not in self._instance:
+            self._instance[self._cls] = self._cls()
+        return self._instance[self._cls]

mindspore/profiler/common/util.py

@@ -21,6 +21,7 @@ import os
 
 # one sys count takes 10 ns, 1 ms has 100000 system count
 import re
+import shutil
 import stat
 
 from mindspore import log as logger
@@ -317,3 +318,127 @@ def get_options(options):
 def combine_stream_task_id(stream_id, task_id):
     """Combine Stream ID and task ID into unique values."""
     return f'{stream_id}_{task_id}'
+
+
+def get_newest_file(file_list):
+    """
+    Find the newest files
+    :param file_list:
+    :return:
+    """
+    newest_file_list = []
+    newest_timestamp = '0'
+    for file_path in file_list:
+        timestamp = file_path.split('.')[0].split('/')[-1].split('_')[-1]
+        newest_timestamp = max(timestamp, newest_timestamp)
+
+    for file_path in file_list:
+        if file_path.split('.')[0].split('/')[-1].split('_')[-1] == newest_timestamp:
+            newest_file_list.append(file_path)
+
+    newest_file_list.sort()
+    return newest_file_list
+
+
+class ProfilerPathManager:
+    """A path manager to manage profiler path"""
+
+    FRAMEWORK_DIR = "FRAMEWORK"
+    INVALID_VALUE = -1
+
+    @classmethod
+    def get_fwk_path(cls, profiler_path: str) -> str:
+        """Get FRAMEWORK directory path"""
+        fwk_path = os.path.join(profiler_path, cls.FRAMEWORK_DIR)
+        if os.path.isdir(fwk_path):
+            return fwk_path
+        return ""
+
+    @classmethod
+    def get_cann_path(cls, profiler_path: str) -> str:
+        """Get CANN Prof directory path"""
+        sub_dirs = os.listdir(os.path.realpath(profiler_path))
+        for sub_dir in sub_dirs:
+            sub_path = os.path.join(profiler_path, sub_dir)
+            if os.path.isdir(sub_path) and re.match(r"^PROF_\d+_\d+_[0-9a-zA-Z]+", sub_dir):
+                return sub_path
+        return ""
+
+    @classmethod
+    def get_host_path(cls, cann_path: str) -> str:
+        """Get CANN Prof host directory path"""
+        host_path = os.path.join(cann_path, 'host')
+        if os.path.exists(host_path):
+            return host_path
+        return ""
+
+    @classmethod
+    def get_device_path(cls, cann_path: str) -> str:
+        """Get CANN Prof device directory path"""
+        sub_dirs = os.listdir(os.path.realpath(cann_path))
+        for sub_dir in sub_dirs:
+            sub_path = os.path.join(cann_path, sub_dir)
+            if os.path.isdir(sub_path) and re.match(r"^device_\d", sub_dir):
+                return sub_path
+        return ""
+
+    @classmethod
+    def remove_path_safety(cls, path: str):
+        """Remove directory"""
+        msg = f"Failed to remove path: {path}"
+        if os.path.islink(path):
+            raise RuntimeError(msg)
+        if not os.path.exists(path):
+            return
+        try:
+            shutil.rmtree(path)
+        except FileNotFoundError:
+            return
+        except Exception as err:
+            raise RuntimeError(msg) from err
+
+    @classmethod
+    def remove_file_safety(cls, file: str):
+        """Remove file"""
+        msg = f"Failed to remove file: {file}"
+        if os.path.islink(file):
+            raise RuntimeError(msg)
+        if not os.path.exists(file):
+            return
+        try:
+            os.remove(file)
+        except FileExistsError:
+            return
+        except Exception as err:
+            raise RuntimeError(msg) from err
+
+    @classmethod
+    def simplify_data(cls, profiler_path: str, simplify_flag: bool):
+        """Profiler simplify temporary data"""
+        cann_path = cls.get_cann_path(profiler_path)
+        device_path = cls.get_device_path(cann_path)
+        host_path = cls.get_host_path(cann_path)
+        rm_dirs = ['sqlite', 'summary', 'timeline'] if simplify_flag else ['sqlite']
+        for rm_dir in rm_dirs:
+            if device_path:
+                target_path = os.path.join(device_path, rm_dir)
+                cls.remove_path_safety(target_path)
+            if host_path:
+                target_path = os.path.join(host_path, rm_dir)
+                cls.remove_path_safety(target_path)
+        if simplify_flag:
+            fwk_path = cls.get_fwk_path(profiler_path)
+            cls.remove_path_safety(fwk_path)
+            if not cann_path:
+                return
+            cann_rm_dirs = ['analyze', 'mindstudio_profiler_log', 'mindstudio_profiler_output']
+            for cann_rm_dir in cann_rm_dirs:
+                target_path = os.path.join(cann_path, cann_rm_dir)
+                cls.remove_path_safety(target_path)
+            log_patten = r'msprof_anlysis_\d+\.log$'
+            for cann_file in os.listdir(cann_path):
+                file_path = os.path.join(cann_path, cann_file)
+                if not os.path.isfile(file_path):
+                    continue
+                if re.match(log_patten, cann_file):
+                    cls.remove_file_safety(file_path)

mindspore/profiler/envprofiling.py

@@ -68,7 +68,7 @@ def construct_profiling_options():
         logger.error(
             "The format of MS_PROFILER_OPTIONS is incorrect. "
             "The MS_PROFILER_OPTIONS parameter configuration may refer to "
-            "'https://www.mindspore.cn/mindinsight/docs/zh-CN/r2.2/performance_profiling_ascend.html'."
+            "'https://www.mindspore.cn/mindinsight/docs/zh-CN/master/performance_profiling_ascend.html'."
         )
         return error_config
     conbine_options = combine_profile_options(profiling_options)
@@ -160,7 +160,7 @@ def parse_ascend_args(options):
     if options.get("aicore_metrics") not in AICORE_METRICS_DICT:
         logger.warning(
             "The 'aicore_metrics' parameter of the environment variable MS_PROFILE_OPTIONS must be in "
-            f"[-1, 0, 1, 2, 3, 4, 5], but got {options.get('aicore_metrics')}, it will be set to 0.")
+            f"[-1, 0, 1, 2, 3, 4, 5, 6], but got {options.get('aicore_metrics')}, it will be set to 0.")
         options["aicore_metrics"] = 0
     if not isinstance(options.get("l2_cache"), bool):
         logger.warning(

mindspore/{_extends/parallel_compile/tbe_compiler → profiler/parser/ascend_analysis}/__init__.py

@@ -1,4 +1,4 @@
-# Copyright 2020 Huawei Technologies Co., Ltd
+# Copyright 2023 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.

mindspore/profiler/parser/ascend_analysis/constant.py

@@ -0,0 +1,53 @@
+# Copyright 2023 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.
+# ============================================================================
+"""Constant value for ascend profiling parser."""
+
+
+class Constant:
+    """Constant values"""
+
+    HOST_TO_DEVICE = "HostToDevice"
+    START_FLOW = "s"
+    END_FLOW = "f"
+    META_EVENT = 'M'
+    COMPLETE_EVENT = 'X'
+    FLOW_OP = "flow"
+    INVALID_FLOW_ID = 18446744073709551615
+
+    # file authority
+    MAX_FILE_SIZE = 1024 * 1024 * 1024 * 10
+    MAX_CSV_SIZE = 1024 * 1024 * 1024 * 5
+
+    # tlv constant struct
+    FIX_SIZE_BYTES = "fix_size_bytes"
+    NS_TO_US = 1e-3
+
+    # field name
+    SEQUENCE_UNMBER = "Sequence number"
+    FORWORD_THREAD_ID = "Fwd thread id"
+    OP_NAME = "op_name"
+    INPUT_SHAPES = "Input Dims"
+    INPUT_DTYPES = "Input type"
+    CALL_STACK = "Call stack"
+    MODULE_HIERARCHY = "Module Hierarchy"
+    FLOPS = "flops"
+    NAME = "name"
+
+    # trace constant
+    PROCESS_NAME = "process_name"
+    PROCESS_LABEL = "process_labels"
+    PROCESS_SORT = "process_sort_index"
+    THREAD_NAME = "thread_name"
+    THREAD_SORT = "thread_sort_index"