!496 Modify Asynchronous Data Dump doc

Merge pull request !496 from caifubi/r0.6

tutorials/source_en/advanced_use/customized_debugging_information.md

@@ -225,6 +225,8 @@ val:[[1 1]
 
 When the training result deviates from the expectation on Ascend, the input and output of the operator can be dumped for debugging through Asynchronous Data Dump.
 
+> `comm_ops` operators are not supported by Asynchronous Data Dump. `comm_ops` can be found in [Operator List](https://www.mindspore.cn/docs/en/master/operator_list.html).
+
 1. Turn on the switch to save graph IR: `context.set_context(save_graphs=True)`.
 2. Execute training script.
 3. Open `hwopt_d_end_graph_{graph id}.ir` in the directory you execute the script and find the name of the operators you want to Dump.
@@ -248,6 +250,9 @@ When the training result deviates from the expectation on Ascend, the input and
     }
     ```
 
+    > - Iteration should be set to 0 in non data sink mode and data of every iterationi will be dumped.
+    > - Iteration should increase by 1 in data sink mode. For example, data of GetNext will be dumped in iteration 0 and data of compute graph will be dumped in iteration 1.
+
 5. Set environment variables.
 
     ```bash
@@ -256,9 +261,8 @@ When the training result deviates from the expectation on Ascend, the input and
     export DATA_DUMP_CONFIG_PATH=data_dump.json
     ```
 
-    > Set the environment variables before executing the training script. Setting environment variables during training will not take effect.
-    
-    > Dump environment variables need to be configured before calling `mindspore.communication.management.init`.
+    > - Set the environment variables before executing the training script. Setting environment variables during training will not take effect.
+    > - Dump environment variables need to be configured before calling `mindspore.communication.management.init`.
 
 6. Execute the training script again.
 7. Parse the Dump file.

tutorials/source_zh_cn/advanced_use/customized_debugging_information.md

@@ -227,6 +227,8 @@ val:[[1 1]
 
 在Ascend环境上执行训练，当训练结果和预期有偏差时，可以通过异步数据Dump功能保存算子的输入输出进行调试。
 
+> 异步数据Dump不支持`comm_ops`类别的算子，算子类别详见[算子支持列表](https://www.mindspore.cn/docs/zh-CN/master/operator_list.html)。
+
 1. 开启IR保存开关： `context.set_context(save_graphs=True)`。
 2. 执行网络脚本。
 3. 查看执行目录下的`hwopt_d_end_graph_{graph id}.ir`，找到需要Dump的算子名称。
@@ -250,6 +252,9 @@ val:[[1 1]
     }
     ```
 
+    > - 非数据下沉模式下，iteration需要设置成0，并且会Dump出每个epoch的数据。
+    > - 数据下沉模式iteration需要增加1。例如iteration-0会Dump出GetNext算子的数据，而iteration-1才会去Dump真正的计算图的数据。
+
 5. 设置数据Dump的环境变量。
 
     ```bash
@@ -258,9 +263,8 @@ val:[[1 1]
     export DATA_DUMP_CONFIG_PATH=data_dump.json
     ```
 
-    > 在网络脚本执行前，设置好环境变量；网络脚本执行过程中设置将会不生效。
-
-    > 在分布式场景下，Dump环境变量需要调用`mindspore.communication.management.init`之前配置。
+    > - 在网络脚本执行前，设置好环境变量；网络脚本执行过程中设置将会不生效。
+    > - 在分布式场景下，Dump环境变量需要调用`mindspore.communication.management.init`之前配置。
 
 6. 再次执行用例进行异步数据Dump。
 7. 解析文件。