Vitis AI安装步骤-包括Ubuntu、Docker安装

环境安装流程

Ubuntu 20.04 安装

1. 下载镜像：ubuntu-20.04.6-desktop-amd64.iso
2. 下载 rufus: 官网地址
3. 制作USB启动盘
4. 开机按 Delete 键进入 BIOS
5. 启动项将首选项改为 USB 启动盘
6. 重新启动进入安装程序- 选择 English- 有线网账号密码：psd- 选择 Mininal installation- 选择 Erase disk and install Ubuntu- 设置用户名密码是：psd
7. 安装配置 ssh

sudoapt-get update

sudoaptinstall openssh-server

sudoapt-getinstallvimsudovim /etc/ssh/sshd_config

端口号修改为23321后保存，继续执行

sudo systemctl restart sshd

sudo ufw allow 23321

1. 关闭自动锁屏：- 点击右上角 Settings- 点击 Privacy- 点击 Screen Lock- 关闭自动锁屏

CUDA 11.3 & cuDNN 安装

1. 查看当前驱动

dpkg -l|grep nvidia

1. 卸载原本的驱动并清理链接

sudoapt-get purge nvidia*

sudoapt autoremove

1. 查询可用驱动

ubuntu-drivers devices

1. 自动安装推荐的驱动

sudo ubuntu-drivers autoinstall

1. 重启，然后验证驱动是否安装成功

sudo reboot

nvidia-smi

1. 下载并运行 CUDA 11.3.1 安装程序

cd ~

wget https://developer.download.nvidia.com/compute/cuda/11.3.1/local_installers/cuda_11.3.1_465.19.01_linux.run

sudoaptinstall gcc

sudosh cuda_11.3.1_465.19.01_linux.run

1. 只勾选 CUDA Toolkit 11.3，然后安装
2. 添加环境变量

sudovim /etc/profile

exportCUDA_HOME=/usr/local/cuda-11.3

exportPATH=$PATH:$CUDA_HOME/bin

exportLD_LIBRARY_PATH=${LD_LIBRARY_PATH}:/usr/local/cuda-11.3/lib64

1. 下载并安装 cuDNN 8.9.2.26

cd ~

wget https://developer.nvidia.com/downloads/compute/cudnn/secure/8.9.2/local_installers/11.x/cudnn-linux-x86_64-8.9.2.26_cuda11-archive.tar.xz

tar-xvf cudnn-linux-x86_64-8.9.2.26_cuda11-archive.tar.xz

cd cudnn-linux-x86_64-8.9.2.26_cuda11-archive

sudocp-r ./bin/* /usr/local/cuda-11.3/bin

sudocp-r ./lib/* /usr/local/cuda-11.3/lib64

1. 验证是否安装成功

source /etc/profile

nvcc -V

nvidia-smi

docker & nvidia docker 安装

1. 安装docker

sudoapt-getinstall-y docker.io

sudo systemctl start dockersudo systemctl enabledockerdocker version

1. 安装 nvidia container toolkit

sudoapt-getinstallcurlwget https://download.docker.com/linux/ubuntu/gpg

sudo apt-key add gpg

vim installNvidiaContainer.sh

distribution=$(. /etc/os-release;echo $ID$VERSION_ID)sudocurl-s-L https://nvidia.github.io/nvidia-docker/gpgkey |sudo apt-key add -

sudocurl-s-L https://nvidia.github.io/nvidia-docker/$distribution/nvidia-docker.list |sudotee /etc/apt/sources.list.d/nvidia-docker.list

保存后执行

./installNvidiaContainer.sh

sudoapt-get update &&sudoapt-getinstall-y nvidia-container-toolkit

sudo systemctl restart dockerrm installNvidiaContainer.sh

1. 创建多个新用户并添加docker用户组权限

cd ~

vim addUsers.sh

fornamein aifpga maomao yangyang lichangyv

doecho$nameuseradd-d /home/$name-m-s /bin/bash $nameecho$name:$name| chpasswd

    usermod-aGdocker${name}# passwd --expire $nameecho"$user add successfuly"done

保存后执行

./addUsers.sh

Vitis & Vivado & Vitis HLS 安装

1. 下载 Vitis 包并进入目录
2. 配置 dash（键盘选择 No）

sudo dpkg-reconfigure dash

1. 安装依赖包并执行安装程序

sudoapt-getinstall ocl-icd-libopencl1

sudoapt-getinstall opencl-headers

sudoapt-getinstall ocl-icd-opencl-dev

sudoaptinstall libstdc++6

sudoaptinstall libncurses5

sudoapt-getinstall libtinfo5

sudochmod +x xsetup

sudo ./xsetup

1. 选择安装内容（需在本机使用图形界面操作）1. 选择 Vitis2. 选择以下内容（共210.68GB）- Vitis Unified Software Platform- Vitis Model Composer- DocNav- Install devices for Alveo and edge acceleration platforms- Install Devices for Kria soMs and starter Kits- Devices for Custom Platforms- Engineering Sample Devices for Custom Platforms3. 其他配置默认，然后等待安装完成
2. 配置环境

sudovim /etc/profile

source /tools/Xilinx/Vivado/2023.1/settings64.sh

source /tools/Xilinx/Vitis/2023.1/settings64.sh

source /tools/Xilinx/Vitis_HLS/2023.1/settings64.sh

1. 安装 USB 驱动

cd /tools/Xilinx/Vivado/2023.1/data/xicom/cable_drivers/lin64/install_script/install_drivers

sudo ./install_drivers

1. 验证是否安装成功

source /etc/profile

vitis

vivado

vitis_hls

无论执行哪一个都有图形界面弹出

Vitis AI 安装

1. 克隆 Vitis AI 仓库

cd ~

git clone https://github.com/Xilinx/Vitis-AI

1. 构建基于 Pytorch-CUDA 的镜像

cd Vitis-AI/docker

./docker_build.sh -t gpu -f pytorch

1. 验证是否安装成功

cd../

./docker_run.sh xilinx/vitis-ai-pytorch-gpu:3.5.0.001-a350fc104