GPU nvidia c++ 开发流程

简介：

深度学习算法使用nvidia GPU可以加速推理速度。但开发具有一定的困难。记录一下自己的安装流程

1. native 开发流程

   1. 安装nvidia driver,请参考一下工程。

   https://gist.github.com/wangruohui/df039f0dc434d6486f5d4d098aa52d07
   

   安装完成后，重启电脑， 使用 nvidia-smi 查看显卡是否安装成功

   2. 安装 cuda

   https://docs.nvidia.com/cuda/cuda-installation-guide-linux/index.html#package-manager-installation
   

   3. cudnn 安装

   https://docs.nvidia.com/deeplearning/sdk/cudnn-install/index.html#install-linux
   

   建议tar包安装

   Navigate to your  directory containing the cuDNN Tar file.
   Unzip the cuDNN package.
   $ tar -xzvf cudnn-10.2-linux-x64-v7.6.5.32.tgz
   
   Copy the following files into the CUDA Toolkit directory, and change the file permissions.
   $ sudo cp cuda/include/cudnn.h /usr/local/cuda/include
   $ sudo cp cuda/lib64/libcudnn* /usr/local/cuda/lib64
   $ sudo chmod a+r /usr/local/cuda/include/cudnn.h /usr/local/cuda/lib64/libcudnn*
   

   4. tensorrt安装（optional，如果工程有tensorrt依赖的话）

   https://docs.nvidia.com/deeplearning/sdk/tensorrt-install-guide/index.html#installing
   

   建议tar包安装

   Download the TensorRT tar file that matches the Linux distribution you are using.
   Choose where you want to install TensorRT. This tar file will install everything into a subdirectory called TensorRT-7.x.x.x.
   Unpack the tar file.
   version=”7.x.x.x”
   os=””
   arch=$(uname -m)
   cuda=”cuda-x.x”
   cudnn=”cudnn7.x”
   tar xzvf TensorRT-${version}.${os}.${arch}-gnu.${cuda}.${cudnn}.tar.gz
   Where:
   7.x.x.x is your TensorRT version
    is:
   Ubuntu-16.04
   Ubuntu-18.04
   CentOS-7.6
   cuda-x.x is CUDA version 9.0, 10.0, or 10.2.
   cudnn7.x is cuDNN version 7.6.
   This directory will have sub-directories like lib, include, data, etc…
   ls TensorRT-${version}
   bin  data  doc  graphsurgeon  include  lib  python  samples  targets  TensorRT-Release-Notes.pdf  uff
   Add the absolute path to the TensorRTlib directory to the environment variable LD_LIBRARY_PATH:
   export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:
   

2. docker 开发流程

   １．安装nvidia driver,请参考上面流程

   ２．安装nvidia docker,请参考

   https://github.com/NVIDIA/nvidia-docker
   

   3 . 下载　nvidia/docker相关 镜像并编译

3 . PKG-CONFIG 方式管理cuda 和tensorrt

1 . 请先理解pkg-config

https://zh.wikipedia.org/wiki/Pkg-config

2 . cuda.pc模板如下

cudaroot=/usr/local/cuda
libdir=${cudaroot}/lib64
includedir=${cudaroot}/include

Name: cuda
Description: CUDA Driver Library
Version: 10.0
Libs: -L${libdir} -lcublas -lcudart -lcurand -lcudnn
Cflags: -I${includedir}

3 . tensorrt.pc 模板如下

root=/home/gh/TensorRT-7.0.0.11
libdir=${root}/lib
includedir=${root}/include
includecommondir=${root}/include

Name: tensorrt
Description: tensorrt for c++
Version: 7.0.0.11
Libs: -L${libdir} -lnvcaffe_parser -lnvinfer -lnvparsers -lnvinfer_plugin  -lnvonnxparser -lmyelin
Cflags: -I${includedir} -I${includecommondir}

4 . 　模板放到PKG_CONFIG_PATH使用的地址。比方说ubuntu 地址为 /usr/local/lib/pkgconfig , centos 路径为　/usr/local/lib64/pkgconfig

5 . CMakeLists.txt 使用

find_package(PkgConfig)

pkg_search_module(CUDA REQUIRED cuda)
include_directories(${CUDA_INCLUDE_DIRS})
link_directories(${CUDA_LIBRARY_DIRS})

pkg_search_module(TENSORRT REQUIRED tensorrt)
include_directories(${TENSORRT_INCLUDE_DIRS})
link_directories(${TENSORRT_LIBRARY_DIRS} )