Ubuntu20.04 CUDA11.8+Cudnn8+TensorRT8+pytorch-2.1.0(C++)安装

一、源码编译python3.8.17

1.下载pyhon3.8.17源码，放到apollo目录下

2.

cd python-3.8.17

./configure --enable-optimizations

make -j18

sudo make install

// make -j后面的数字为编译用到的核心数量

3.验证

python -V

如果不是3.8.17，将/usr/bin下原有的pyhton3删除，并链接到/usr/local/bin/python3.8

sudo rm -rf /usr/bin/python /usr/bin/python3

sudo ln -s /usr/local/bin/python3.8 /usr/bin/python

sudo ln -s /usr/local/bin/python3.8 /usr/bin/python3

二、安装cuda11.8

1.去Nvidia官网下载cuda_11.8.0_520.61.05_linux.run

2.安装cuda11.8

sudo bash cuda_11.8.0_520.61.05_linux.run

依次

a. 输入accept

b. 取消Driver的安装（在Driver选项按下回车），取消后Driver前应当没有[ X ]

c. 选中Install，回车

d. 提示A symlink already exists at /usr/local/cuda. Update to this installation?

    - 选择Yes，回车

3.验证

nvcc -V

输出为：

nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2022 NVIDIA Corporation
Built on Wed_Sep_21_10:33:58_PDT_2022
Cuda compilation tools, release 11.8, V11.8.89
Build cuda_11.8.r11.8/compiler.31833905_0

三、安装cudnn8

1.去Nvidia官网下载cudnn-local-repo-ubuntu2004-8.9.6.50_1.0-1_amd64.deb

2.卸载原有cudnn

sudo apt remove libcudnn*

3.安装cudnn8

sudo apt install ./cudnn-local-repo-ubuntu2004-8.9.6.50_1.0-1_amd64.deb
sudo cp /var/cudnn-local-repo-ubuntu2004-8.9.6.50/cudnn-local-5FA1A941-keyring.gpg /usr/share/keyrings/
sudo apt update
sudo apt install libcudnn8=8.9.6.50-1+cuda11.8
sudo apt install libcudnn8-dev=8.9.6.50-1+cuda11.8
sudo apt install libcudnn8-samples=8.9.6.50-1+cuda11.8

4.查看cudnn版本

cat /usr/include/cudnn_version.h

其中有三行

#define CUDNN_MAJOR 8
#define CUDNN_MINOR 9
#define CUDNN_PATCHLEVEL 6

四、安装TensorRT8

1.去Nvidia官网下载nv-tensorrt-local-repo-ubuntu2004-8.5.1-cuda-11.8_1.0-1_amd64.deb

2.安装tensorrt

sudo apt install ./nv-tensorrt-local-repo-ubuntu2004-8.5.1-cuda-11.8_1.0-1_amd64.deb
sudo cp /var/nv-tensorrt-local-repo-ubuntu2004-8.5.1-cuda-11.8/nv-tensorrt-local-3E18D84E-keyring.gpg /usr/share/keyrings/
sudo apt update
sudo aot install tensorrt

3.验证

ls /usr/lib/x86_64-linux-gnu/libnvinfer.so*

看到libnvinfer.so.8就算成功...

五、源码编译pytorch-2.1.0

1.下载pytorch-2.1.0

2.

cd pytorch
mkdir build && cd build
pip install pyyaml -i https://pypi.tuna.tsinghua.edu.cn/simple
cmake -D CMAKE_BUILD_TYPE=RELEASE -D CMAKE_INSTALL_PREFIX=/usr/local/libtorch_gpu -D CMAKE_CXX_STANDARD=17 -D CMAKE_CXX_STANDARD_REQUIRED=ON -D USE_CUDA=ON -D USE_CUDNN=ON -D USE_OPENCV=OFF -D BUILD_CAFFE2_MOBILE=OFF -D BUILD_PYTHON=OFF -D BUILD_CAFFE2_OPS=OFF -D BUILD_TEST=OFF -D USE_TBB=OFF ..
make -j18
sudo make install

// make -j后面的数字为编译用到的核心数量