昇思MindSpore源码编译安装

企业开发 2023-06-18 21:51:11 阅读次数: 0

一、参考资料

MindSpore源码阅读系列(三)：统一调试MindSpore Python与C++
MindSpore源码阅读系列(二)：编译并调试MindSpore源码(对阅读源码很有帮助)

二、源码编译MindSpore（Ascend910版本）

源码编译方式安装MindSpore Ascend 910版本

1. 重要说明

  • 建议使用Anaconda虚拟环境，防止破坏系统中的Ascend环境。
  • MindSpore与CANN版本版本对齐，否则编译得到MindSpore软件包无法使用。
  • 当默认路径存在安装包的时候，LD_LIBRARY_PATH环境变量不起作用；默认路径优先级别为：/usr/local/Ascend/nnae 高于 /usr/loacl/Ascend/ascend-toolkit。原因是MindSpore采用DT_RPATH方式支持无环境变量启动，减少用户设置；DT_RPATH优先级比LD_LIBRARY_PATH环境变量高。

2. 准备工作

2.1 安装CANN和固件驱动

昇腾软件包提供商用版和社区版两种下载途径：

详细安装步骤，请参考：install/mindspore_ascend_install_source.md · MindSpore/docs - Gitee.com

2.2 安装昇腾AI处理器配套软件包

如果之前已经安装过昇腾AI处理器配套软件包，需要先卸载对应的whl包。

pip uninstall te topi hccl -y

重新安装昇腾AI处理器配套软件whl包

pip install sympy
pip install /usr/local/Ascend/ascend-toolkit/latest/lib64/topi-*-py3-none-any.whl
pip install /usr/local/Ascend/ascend-toolkit/latest/lib64/te-*-py3-none-any.whl
pip install /usr/local/Ascend/ascend-toolkit/latest/lib64/hccl-*-py3-none-any.whl

3. 下载源码

以MindSpore1.9版本为例。

git clone https://gitee.com/mindspore/mindspore.git -b r1.9

4. 编译

cd mindspore

# 编译Ascend910版本
bash build.sh -e ascend -S on

# 编译CPU版本
bash build.sh -e cpu -S on

time bash build.sh -d -t ut -S on

4.1 CPU版本

源码编译方式安装MindSpore CPU版本

源码编译方式安装MindSpore CPU版本（含第三方依赖）

# time bash build.sh -e cpu -S on
...
...
...
adding 'mindspore-1.9.1.dist-info/METADATA'
adding 'mindspore-1.9.1.dist-info/WHEEL'
adding 'mindspore-1.9.1.dist-info/entry_points.txt'
adding 'mindspore-1.9.1.dist-info/top_level.txt'
adding 'mindspore-1.9.1.dist-info/RECORD'
removing build/bdist.linux-aarch64/wheel
CPack: - package: /data/YOYOFile/mindspore/build/mindspore/mindspore generated.
success building mindspore project!
---------------- MindSpore: build end   ----------------

real    109m37.807s
user    822m49.398s
sys     54m14.307s

输出

output/mindspore-1.9.1-cp39-cp39-linux_aarch64.whl
output/mindspore-1.9.1-cp39-cp39-linux_aarch64.whl.sha256

4.2 Ascend910版本

# time bash build.sh -e ascend -S on
...
...
...
adding 'mindspore_ascend-1.9.1.dist-info/METADATA'
adding 'mindspore_ascend-1.9.1.dist-info/WHEEL'
adding 'mindspore_ascend-1.9.1.dist-info/entry_points.txt'
adding 'mindspore_ascend-1.9.1.dist-info/top_level.txt'
adding 'mindspore_ascend-1.9.1.dist-info/RECORD'
removing build/bdist.linux-aarch64/wheel
CPack: - package: /data/YOYOFile/mindspore/build/mindspore/mindspore generated.
success building mindspore project!
---------------- MindSpore: build end   ----------------

real    128m6.095s
user    925m10.682s
sys     56m32.674s

# time bash build.sh -e ascend -S on
...
...
...
[100%] Built target _mindspore_frontend_parallel_obj
[100%] Linking CXX static library libmindspore.a
[100%] Built target mindspore
Consolidate compiler generated dependencies of target _c_expression
[100%] Linking CXX shared library _c_dataengine.cpython-39-aarch64-linux-gnu.so
[100%] Building CXX object mindspore/ccsrc/CMakeFiles/_c_expression.dir/pipeline/jit/init.cc.o
[100%] Built target _c_dataengine
Consolidate compiler generated dependencies of target cache_admin
Consolidate compiler generated dependencies of target cache_server
[100%] Building CXX object mindspore/ccsrc/minddata/dataset/engine/cache/CMakeFiles/cache_server.dir/cache_main.cc.o
[100%] Building CXX object mindspore/ccsrc/minddata/dataset/engine/cache/CMakeFiles/cache_admin.dir/cache_admin.cc.o
[100%] Building CXX object mindspore/ccsrc/minddata/dataset/engine/cache/CMakeFiles/cache_admin.dir/cache_admin_arg.cc.o
[100%] Linking CXX executable cache_server
[100%] Built target cache_server
[100%] Linking CXX executable cache_admin
[100%] Built target cache_admin
[100%] Linking CXX shared module _c_expression.cpython-39-aarch64-linux-gnu.so
[100%] Built target _c_expression
Run CPack packaging tool...
CPack: Create package using External
CPack: Install projects
CPack: - Clean temporary : /data/YOYOFile/mindspore/build//package/mindspore
CPack: - Install project: MindSpore []
CPack: Create package
...
...
...
adding 'mindspore_ascend-1.9.1.dist-info/METADATA'
adding 'mindspore_ascend-1.9.1.dist-info/WHEEL'
adding 'mindspore_ascend-1.9.1.dist-info/entry_points.txt'
adding 'mindspore_ascend-1.9.1.dist-info/top_level.txt'
adding 'mindspore_ascend-1.9.1.dist-info/RECORD'
removing build/bdist.linux-aarch64/wheel
CPack: - package: /data/YOYOFile/mindspore/build/mindspore/mindspore generated.
success building mindspore project!
---------------- MindSpore: build end   ----------------

real    127m5.367s
user    911m16.674s
sys     48m24.587s

输出

output/mindspore_ascend-1.9.1-cp39-cp39-linux_aarch64.whl
output/mindspore_ascend-1.9.1-cp39-cp39-linux_aarch64.whl.sha256

5. 安装

在联网状态下，安装whl包时会自动下载MindSpore安装包的依赖项（依赖项详情参见setup.py中的required_package），其余情况需自行安装。运行模型时，需要根据ModelZoo中不同模型指定的requirements.txt安装额外依赖，常见依赖可以参考requirements.txt。

pip install output/mindspore-*.whl -i https://pypi.tuna.tsinghua.edu.cn/simple

6. 配置环境变量

LOCAL_ASCEND=/usr/local/Ascend 是昇腾AI处理器配套软件包默认安装路径。

# 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:${LOCAL_ASCEND}/ascend-toolkit/latest/opp/op_impl/built-in/ai_core/tbe/op_tiling:${LD_LIBRARY_PATH}

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

## Python library that TBE implementation depends on
export PYTHONPATH=${TBE_IMPL_PATH}:${PYTHONPATH}

# Set path to extracted MindSpore accordingly
export LD_LIBRARY_PATH={
    
    mindspore_path}:${LD_LIBRARY_PATH}

参数解释

  • {mindspore_path}表示MindSpore二进制包所在位置的绝对路径。

7. 验证是否成功安装

7.1 方法一

python -c "import mindspore;mindspore.run_check()"

MindSpore version: 版本号
The result of multiplication calculation is correct, MindSpore has been installed successfully!

(ms19) root@80f9a288c6ba:~# python -c "import mindspore;mindspore.run_check()"
MindSpore version:  1.9.0
The result of multiplication calculation is correct, MindSpore has been installed successfully!

7.2 方法二

import numpy as np
import mindspore as ms
import mindspore.ops as ops

ms.set_context(device_target="Ascend")
x = ms.Tensor(np.ones([1,3,3,4]).astype(np.float32))
y = ms.Tensor(np.ones([1,3,3,4]).astype(np.float32))
print(ops.add(x, y))

[[[[2. 2. 2. 2.]
   [2. 2. 2. 2.]
   [2. 2. 2. 2.]]

  [[2. 2. 2. 2.]
   [2. 2. 2. 2.]
   [2. 2. 2. 2.]]

  [[2. 2. 2. 2.]
   [2. 2. 2. 2.]
   [2. 2. 2. 2.]]]]

8. 本地源码编译升级

在源码根目录下执行编译脚本build.sh成功后，在output目录下找到编译生成的whl安装包，然后执行命令进行升级。

pip install --upgrade output/mindspore-*.whl

三、源码编译MindSpore（Ascend310版本）

源码编译方式安装mindspore-ascend-310版本

整体步骤与Ascend910一致，这里仅记录差异的地方，需要特别注意。

1. 编译

cd mindspore

# 编译Ascend310版本
bash build.sh -e ascend -V 310 -S on

# time bash build.sh -e ascend -V 310 -S on
...
...
...
Run CPack packaging tool...
CPack: Create package using TGZ
CPack: Install projects
CPack: - Run preinstall target for: MindSpore
CPack: - Install project: MindSpore []
CPack: Create package
CPack: - package: /data/YOYOFile/mindspore/output/mindspore_ascend-1.9.1-linux_aarch64.tar.gz generated.
CPack: - checksum file: /data/YOYOFile/mindspore/output/mindspore_ascend-1.9.1-linux_aarch64.tar.gz.sha256 generated.
success building mindspore project!
acl mode, skipping deploy phase
---------------- MindSpore: build end   ----------------

real    68m39.293s
user    511m36.950s
sys     25m28.390s

输出

output/mindspore_ascend-1.9.1-linux_aarch64.tar.gz
output/mindspore_ascend-1.9.1-linux_aarch64.tar.gz.sha256

2. 安装

由于Ascend310是推理卡，源码编译Ascend310版本的MindSpore软件包只能用于推理，用于真实生产环境。

tar -zxf output/mindspore_ascend-*.tar.gz

3. 配置环境变量

...
...
# 其他环境变量配置，与Ascend910一致
# Set path to extracted MindSpore accordingly
export LD_LIBRARY_PATH={
    
    mindspore_path}:${LD_LIBRARY_PATH}

解释说明

  • {mindspore_path}表示MindSpore二进制包所在位置的绝对路径。

4. 验证是否安装成功

创建目录放置样例代码工程，例如/home/HwHiAiUser/Ascend/ascend-toolkit/20.0.RC1/acllib_linux.arm64/sample/acl_execute_model/ascend310_single_op_sample，代码可以从官网示例下载获取，这是一个[1, 2, 3, 4]与[2, 3, 4, 5]相加的简单样例，代码工程目录结构如下:

└─ascend310_single_op_sample
    ├── CMakeLists.txt                    // 编译脚本
    ├── README.md                         // 使用说明
    ├── main.cc                           // 主函数
    └── tensor_add.mindir                 // MindIR模型文件

# 进入样例工程目录，按照实际情况修改路径路径
cd /home/HwHiAiUser/Ascend/ascend-toolkit/20.0.RC1/acllib_linux.arm64/sample/acl_execute_model/ascend310_single_op_sample

# 编译
# {mindspore_path}表示MindSpore二进制包所在位置的绝对路径
cmake . -DMINDSPORE_PATH={
    
    mindspore_path}
make

# 执行
./tensor_add_sample

root@44ff3bb0ad1d:/data/YOYOFile/Downloads/ascend310_single_op_sample# cmake . -DMINDSPORE_PATH=/data/YOYOFile/Downloads/mindspore_ascend-1.6.2-linux_aarch64
-- The C compiler identification is GNU 7.5.0
-- The CXX compiler identification is GNU 7.5.0
-- Detecting C compiler ABI info
-- Detecting C compiler ABI info - done
-- Check for working C compiler: /usr/bin/cc - skipped
-- Detecting C compile features
-- Detecting C compile features - done
-- Detecting CXX compiler ABI info
-- Detecting CXX compiler ABI info - done
-- Check for working CXX compiler: /usr/bin/c++ - skipped
-- Detecting CXX compile features
-- Detecting CXX compile features - done
-- Configuring done
-- Generating done
-- Build files have been written to: /data/YOYOFile/Downloads/ascend310_single_op_sample
root@44ff3bb0ad1d:/data/YOYOFile/Downloads/ascend310_single_op_sample# make -j4
Consolidate compiler generated dependencies of target tensor_add_sample
[ 50%] Building CXX object CMakeFiles/tensor_add_sample.dir/main.cc.o
[100%] Linking CXX executable tensor_add_sample
[100%] Built target tensor_add_sample
root@44ff3bb0ad1d:/data/YOYOFile/Downloads/ascend310_single_op_sample# ./tensor_add_sample
3
5
7
9

猜你喜欢

转载自blog.csdn.net/m0_37605642/article/details/128992269
今日推荐
openKylin 社区生态委员会第六次会议圆满召开
阿里云正式发布通义千问 2.5
Python 3.13 发布首个 Beta：实验性自由线程模式和 JIT、改进交互式解释器
Stack Overflow 拿我的代码去训练 AI 大模型，还封了我的账号
Pop!_OS 的 COSMIC 桌面完成 App Store 上架工作
报告：Django 仍然是 74% 开发者的首选
《2024 年一季度互联网投融资运行情况》研究报告
15 年前上了“FFmpeg 耻辱柱”，今天他还得谢谢咱——腾讯QQPlayer一雪前耻？
TIOBE 5 月榜单：Fortran “复活”进入 Top 10
GCC 14.1 发布
面壁智能发布 Eurux-8x22B 开源大模型 —— 堪称「理科状元」
开源日报 | 谷歌扶持鸿蒙上位；开源Rabbit R1；Docker加持的安卓手机；微软的焦虑和野心；海尔电器把开放平台关了
周排行
计算机组成与设计（七）—— 除法器
Integer Approximation(分治+枚举)
大话数据库索引
windows10系统JDK的配置及下载地址
mysql实现秒值转换中原六仔平台搭建
Codeforces Round #556 (Div. 1)
百练1064 网线主管
Codeforces 995F Cowmpany Cowmpensation
子集生成之增量构造法，位向量法，二进制法
ERROR: cmd.exe failed with args /c "/APK\gradle\rungradle.bat...
每日归档
更多
2024-05-10(38)
2024-05-09(35)
2024-05-08(42)
2024-05-07(14)
2024-05-06(40)
2024-05-05(0)
2024-05-04(7)
2024-05-03(19)
2024-05-02(0)
2024-05-01(4)

代码天地 © 2018 codetd.com

境外服务器   最新电视剧2022