详解Windows下源码编译PaddlePaddle

Windows 下从源码编译 PaddlePaddle

1.环境准备

• Windows 10 专业版 (64bit)

• Python 版本 Python 3.9.16(64 bit)

• Visual Studio2022 专业版

1.1 cmake

cmake --version

cmake version 3.25.1-msvc1
CMake suite maintained and supported by Kitware (kitware.com/cmake).

1.2 python

(p2) PS D:\Paddle> python --version
Python 3.9.16

1.3 cuda

(p2) PS D:\Paddle> nvidia-smi
Wed Apr  5 17:06:19 2023
+-----------------------------------------------------------------------------+
| NVIDIA-SMI 516.94       Driver Version: 516.94       CUDA Version: 11.7     |

1.4 python包

(p2) PS D:\Paddle> pip install numpy protobuf  wheel ninja
Looking in indexes: https://pypi.tuna.tsinghua.edu.cn/simple
Requirement already satisfied: numpy in i:\miniconda3\envs\p2\lib\site-packages (1.23.2)
Requirement already satisfied: protobuf in i:\miniconda3\envs\p2\lib\site-packages (3.20.3)
Requirement already satisfied: wheel in i:\miniconda3\envs\p2\lib\site-packages (0.38.4)
Requirement already satisfied: ninja in i:\miniconda3\envs\p2\lib\site-packages (1.11.1)

1.5 创建编译文件夹

mkdir D:\workspace && cd /d D:\workspace
git clone https://github.com/PaddlePaddle/Paddle.git
cd Paddle

1.6 启用visual studio终端

**********************************************************************
** Visual Studio 2022 Developer Command Prompt v17.5.3
** Copyright (c) 2022 Microsoft Corporation
**********************************************************************
C:\Program Files\Microsoft Visual Studio\2022\Professional>d:
D:\>cd Paddle
D:\Paddle>

1.7 开始编译

先编译CPU版本吧，因为GPU的cuda和cudnn环境变量不会写（偷懒用conda安装的）

cmake .. -GNinja -DWITH_GPU=OFF -DWITH_UNITY_BUILD=ON
ninja

编译过程中会下载东西，所以建议不要断网。 此过程中：

• 特别费CPU,我的到了100％，风扇呼呼的；

• 内存使用不大，10GB左右没变过；

• 不断使用git下载，所以网速要好。

1.8 docker编译

docker run --name paddle-test -v 123456:/paddle --network=host -it registry.baidubce.com/paddlepaddle/paddle:latest-dev /bin/bash

3 times.
CMake Error at /paddle/build/third_party/dlpack/tmp/extern_dlpack-gitclone.cmake:31 (message):
 Failed to clone repository: 'https://github.com/dmlc/dlpack.git'
CMakeFiles/extern_dlpack.dir/build.make:91: recipe for target 'third_party/dlpack/src/extern_dlpack-stamp/extern_dlpack-download' failed
make[2]: *** [third_party/dlpack/src/extern_dlpack-stamp/extern_dlpack-download] Error 1
CMakeFiles/Makefile2:4791: recipe for target 'CMakeFiles/extern_dlpack.dir/all' failed
make[1]: *** [CMakeFiles/extern_dlpack.dir/all] Error 2
fatal: unable to access 'https://github.com/JuliaStrings/utf8proc.git/': Failed to connect to github.com port 443: Connection refused
-- Had to git clone more than once:
         3 times.
CMake Error at /paddle/build/third_party/utf8proc/tmp/extern_utf8proc-gitclone.cmake:31 (message):
 Failed to clone repository: 'https://github.com/JuliaStrings/utf8proc.git'
CMakeFiles/extern_utf8proc.dir/build.make:90: recipe for target 'third_party/utf8proc/src/extern_utf8proc-stamp/extern_utf8proc-download' failed
make[2]: *** [third_party/utf8proc/src/extern_utf8proc-stamp/extern_utf8proc-download] Error 1
CMakeFiles/Makefile2:4818: recipe for target 'CMakeFiles/extern_utf8proc.dir/all' failed

来源：https://juejin.cn/post/7218482973980082237