提前编译安装好对应版本的CUDA

源码下载

https://gitee.com/paddlepaddle/Paddle/tree/release%2F2.0/

解压

unzip paddlepaddle-Paddle-release-2.0.zip
cd Paddle
mkdir build && cd build

编译配置

PADDLE_ROOT=/usr/local/paddle
cmake -DFLUID_INFERENCE_INSTALL_DIR=$PADDLE_ROOT \
  -DCMAKE_BUILD_TYPE=Release \
  -DWITH_PYTHON=OFF \
  -DWITH_MKL=ON \
  -DWITH_GPU=ON  \
  -DON_INFER=ON \
  -DWITH_NCCL=OFF \
  ..

paddle预测库编译选项.png

遇到问题

CMake Error at CMakeLists.txt:48 (enable_language):
  No CMAKE_CUDA_COMPILER could be found.

  Tell CMake where to find the compiler by setting either the environment
  variable "CUDACXX" or the CMake cache entry CMAKE_CUDA_COMPILER to 
  the full  path to the compiler, or to the compiler name if it is in the PATH.

解决问题

#查看cuda的安装路径
find -name cuda
...
./usr/local/cuda
#执行以下指令后重新配置
export PATH=/usr/local/cuda/bin:$PATH

编译

make -j8

遇到的问题

...
fatal: unable to access 'https://github.com/dmlc/dlpack.git/': Protocol "https" not supported or disabled in libcurl
...

解决问题

https://blog.csdn.net/yujuan110/article/details/80939311
#下载新版openssll并上传到指定目录
#解压
tar -zxvf openssl-1.1.1k.tar.gz
 cd openssl-1.1.1k/
#配置
./config --prefix=/usr/local/curl
#编译
make -j8
#安装
make install

参考链接

 https://www.paddlepaddle.org.cn/documentation/docs/zh/guides/05_inference_deployment/inference/build_and_install_lib_cn.html