Merge pull request #8 from MooreThreads/update_doc

[MTAI] doc(README): update doc about mt-container-toolkit

README.md

@@ -52,7 +52,7 @@ pip install torch_musa_xxxxxx-cp39-cp39-linux_x86_64.whl
 - [MUDNN](https://new-developer.mthreads.com/sdk/download/musa?equipment=&os=&driverVersion=&version=)
 - Other Libs (including muThrust, muSparse, muAlg, muRand)
 - [PyTorch Source Code](https://github.com/pytorch/pytorch/tree/v2.0.0)
-- Docker Container Toolkits
+- [Docker Container Toolkits](https://mcconline.mthreads.com/software)
 
 **NOTE:** Since some of the dependent libraries are in beta and have not yet been officially released, we recommend using the [development docker](#docker-image-for-developer) provided below to compile **torch_musa**. If you really want to compile **torch_musa** in your own environment, then please contact us for additional dependencies.
 
@@ -111,6 +111,9 @@ python setup.py install
 ### Docker Image
 - [Development Docker Image Download Link](https://mcconline.mthreads.com/repo/musa-pytorch-dev-public?repoName=musa-pytorch-dev-public&repoNamespace=mcconline&displayName=MUSA%20Pytorch%20Dev%20Public)
 - [Release Docker Image Download Link](https://mcconline.mthreads.com/repo/musa-pytorch-release-public?repoName=musa-pytorch-release-public&repoNamespace=mcconline&displayName=MUSA%20Pytorch%20Release%20Public)
+
+**NOTE:** If you want to use **torch_musa** in docker container, please install [mt-container-toolkit](https://mcconline.mthreads.com/software/1?id=1) first and use '--env MTHREADS_VISIBLE_DEVICES=all' when starting a container.
+
 #### Docker Image for Developer
 
 ```bash

docs/developer_guide/source/4_installation/installation.rst

@@ -14,7 +14,7 @@
   * muRAND_dev1.0.0.tar.gz
   * muSPARSE_dev0.1.0.tar.gz
   * muThrust_dev-0.1.1-Linux.deb
-  * Docker Container Toolkits
+  * Docker Container Toolkits(https://mcconline.mthreads.com/software)
 
 
 
@@ -39,6 +39,9 @@ torch_musa是在PyTorch v2.0.0基础上以插件的方式来支持摩尔线程
 
 开发docker镜像中已经安装了必需的依赖包，包括一些未正式发布的依赖包。如果用户不想在我们提供的docker镜像中开发，请通过developers@mthreads.com邮箱联系我们获取必需的依赖包。
 
+.. attention::
+   | 使用docker时，请务必提前安装mt-container-toolkit(https://mcconline.mthreads.com/software/1?id=1)，并且在启动docker container时添加选项“--env MTHREADS_VISIBLE_DEVICES=all”，否则在docker container内部无法使用torch_musa。
+
 编译步骤
 ---------
 
@@ -58,13 +61,12 @@ torch_musa是在PyTorch v2.0.0基础上以插件的方式来支持摩尔线程
 .. code-block:: bash
 
   cd torch_musa
-  bash scripts/update_daily_mudnn.sh # update daily mudnn lib if needed
-  bash build.sh   # build original PyTorch and Torch_MUSA from scratch
+  bash build.sh   # build original PyTorch and torch_musa from scratch
   
   # Some important parameters are as follows:
   bash build.sh --torch  # build original PyTorch only
-  bash build.sh --musa   # build Torch_MUSA only
-  bash build.sh --fp64   # compile fp64 in kernels using mcc in Torch_MUSA
+  bash build.sh --musa   # build torch_musa only
+  bash build.sh --fp64   # compile fp64 in kernels using mcc in torch_musa
   bash build.sh --debug  # build in debug mode
   bash build.sh --asan   # build in asan mode
   bash build.sh --clean  # clean everything built

docs/developer_guide/source/8_FAQ/index.rst

@@ -17,6 +17,15 @@ Q：如果在更新过torch_musa最新代码后，编译报错？
 如果还报错，可能是因为需要更新MUSA软件栈中某个底层软件包。
 
 
+Docker容器
+-----------
+
+Q：如果在docker container内部使用torch_musa时，报错 ``ImportError: libsrv_um_MUSA.so: cannot open shared object file: No such file or directory`` ？
+
+#. 请确保成功安装mt-container-toolkit(https://mcconline.mthreads.com/software/1?id=1);
+#. 在启动docker container时请添加 ``--env MTHREADS_VISIBLE_DEVICES=all`` 。
+
+
 适配算子
 ----------
 Q：如果在CUDA-Porting适配新算子时，编译可以通过，在 ``import torch;import torch_musa`` 时报错找不到符号?