本文已参与「新人创作礼」活动，一起开启掘金创作之路。

Ascend910安装mindspore

本文档记录Ascend 910服务器Mindspore安装过程中遇到的各种环境、包兼容性问题，以期避免后续Mindspore在昇腾的安装遇到的问题。Mindspore的安装参考线上向导文档手动安装(www.mindspore.cn/install)。

gcc安装

按照mindspore安装向导，需要至少python3.7，为了解决包兼容性问题，gcc需要7以上，否则容易出现import异常，ImportError: /lib64/libstdc++.so.6: version `CXXABI_1.3.11' not found (required by /usr/local/lib/python3.7/site-packages/mindspore/_c_expression.cpython-37m-aarch64-linux-gnu.so)

Centos7.6默认安装libstdc++.so.6.0.19比较旧。ascend-toolkit后，默认带了libstdc++.so.6.0.24，依赖glibc、gcc环境包比较多。因此建议编译部署gcc7.5.0，下载地址ftp.tsukuba.wide.ad.jp/software/gc…。编译安装命令如下：

tar -Jxvf gcc-7.5.0.tar.xz; cd gcc-7.5.0

contrib/download_prerequisites

mkdir dist;cd dist

../configure --enable-checking=release --enable-bootstrap --enable-languages=c,c++ --enable-shared --enable-threads=posix  --disable-multilib

make –j64  ;make install  ###192 core

python安装

mindspore安装向导的python安装依靠conda环境管理。手动情况下可以自行快速安装。从registry.npmmirror.com/binary.html… 可以快速下载python3.7.13介质并安装。

python安装（确保openssl正确安装）

wget https://registry.npmmirror.com/-/binary/python/3.7.13/Python-3.7.13.tar.xz

 tar -Jxvf Python-3.7.13.tar.xz

yum install -y openssl-devel

./configure --prefix=/usr/local --with-openssl=/usr --enable-shared     #### 确保checking for openssl/ssl.h in /usr... yes

make –j64& make install  ###192 core，默认安装到/usr/local

设置pip快速index

python3 -m pip install --upgrade pip

pip3 config set global.index-url <https://pypi.tuna.tsinghua.edu.cn/simple/>

Firmware/driver选择

Firmware/driver版本的选择需要与cann、mindspore相匹配，否则mindspore运行中可能报出动态库异常错误。本文选自cann版本为如下图(www.hiascend.com/software/ca…

因此对应的firmware/driver版本应如下选择该版本。

下载包含firmware/driver的zip包到计算节点解开后(比如下载并解压到/usr/local/Ascend/A900-9000-npu_5.1.rc2_linux-aarch64)，执行安装脚本：

./install.sh install all

安装完成后可以通过npu-smi查看firmware、driver、npu-smi工具版本信息。

Cann安装

按照mindspore安装向导(安装昇腾AI处理器配套软件包)选定cann版本后，下载相应的nnae、nnrt、tookkit软件包后，默认安装到/usr/local/Ascend目录下(若安装程序报未找到python3，将python路径添加到PATH环境变量即可)。

cd /usr/local/Ascend

./Ascend-cann-nnae_5.1.RC2.alpha007_linux-aarch64.run --install

./Ascend-cann-nnrt_5.1.RC2.alpha007_linux-aarch64.run --install

./Ascend-cann-toolkit_5.1.RC2.alpha007_linux-aarch64.run --install

Mindspore安装和验证

前述gcc、python、firmware/driver、cann部署安装后，按照mindspore安装向导可以比较顺利完成安装。

pip3 uninstall te topi hccl -y

pip3 install sympy

pip3 install /usr/local/Ascend/ascend-toolkit/latest/lib64/topi-*-py3-none-any.whl

pip3 install /usr/local/Ascend/ascend-toolkit/latest/lib64/te-*-py3-none-any.whl

pip3 install /usr/local/Ascend/ascend-toolkit/latest/lib64/hccl-*-py3-none-any.whl

 

pip3 install https://ms-release.obs.cn-north-4.myhuaweicloud.com/1.8.0/MindSpore/ascend/aarch64/mindspore_ascend-1.8.0-cp37-cp37m-linux_aarch64.whl --trusted-host ms-release.obs.cn-north-4.myhuaweicloud.com -i https://pypi.tuna.tsinghua.edu.cn/simple

 ```

更新ldconfig库
```bash
ldconfig /usr/local/lib

更新环境变量

#/bin/bash

# control log level. 0-DEBUG, 1-INFO, 2-WARNING, 3-ERROR, 4-CRITICAL, default level is WARNING.

export GLOG_v=2

 

# Conda environmental options

LOCAL_ASCEND=/usr/local/Ascend # the root directory of run package

 

# lib libraries that the run package depends on

export LD_LIBRARY_PATH=${LOCAL_ASCEND}/ascend-toolkit/latest/lib64:${LOCAL_ASCEND}/driver/lib64:${LOCAL_ASCEND}/ascend-toolkit/latest/opp/op_impl/built-in/ai_core/tbe/op_tiling:/usr/local/lib:/usr/local/lib64:${LD_LIBRARY_PATH}:/lib64

 

# Environment variables that must be configured

## TBE operator implementation tool path

export TBE_IMPL_PATH=${LOCAL_ASCEND}/ascend-toolkit/latest/opp/op_impl/built-in/ai_core/tbe

## OPP path

export ASCEND_OPP_PATH=${LOCAL_ASCEND}/ascend-toolkit/latest/opp

## AICPU path

export ASCEND_AICPU_PATH=${ASCEND_OPP_PATH}/..

## TBE operator compilation tool path

export PATH=${LOCAL_ASCEND}/ascend-toolkit/latest/ccec_compiler/bin/:${PATH}

## Python library that TBE implementation depends on

export PYTHONPATH=${TBE_IMPL_PATH}:${PYTHONPATH}

 

 

 

source  /usr/local/Ascend/ascend-toolkit/set_env.sh

验证mindspore安装成功

昇腾Ascend910环境安装Mindspore