Linux Pytorch C++

0 安装Linux环境

apt-get install cmake
apt-get install libnss3

1，下载Pytorch C++编译环境

https://download.pytorch.org/libtorch/nightly/cpu/libtorch-shared-with-deps-latest.zip

并解压

2 安装vscode

   下载地址：https://vscode.cdn.azure.cn/stable/7f3ce96ff4729c91352ae6def877e59c561f4850/code-stable-1539735949.tar.gz

   启动

./code

3 pytorch c++编写

创建一个工程，目录结构：

test/
  a.cpp
  CMakeLists.txt
  make # 执行脚本，删除build文件，然后编译

CMakeLists.txt，代码

cmake_minimum_required(VERSION 3.0 FATAL_ERROR)
project(custom_ops)

find_package(Torch REQUIRED)

add_executable(abc a.cpp)
target_link_libraries(abc "${TORCH_LIBRARIES}")
set_property(TARGET abc PROPERTY CXX_STANDARD 11)

a.cpp，代码：

#include <torch/script.h> // One-stop header.
#include <ATen/ATen.h>

#include <iostream>
#include <memory>

using namespace std;
using namespace at;
 
int main(int argc, const char* argv[]) {

  at::Tensor a = at::ones({2, 2}, at::kInt);
  std::cout << a << "\n";

  std::cout << "ok\n";
  getchar();

  return 1;
}

make文件，直接编译，并添加执行权限，chmod u+x make

#!/usr/bash

# delete build
path_project=`pwd`
path_project_build=$path_project/build

# delete build directory
rm -rf $path_project_build

# create build directory
mkdir $path_project_build

# join build directory
cd $path_project_build

# compile file
cmake -DCMAKE_PREFIX_PATH=path/to/libtorch ..

make

4，编译并执行。

4.1 编译

在工程下创建build,工程目录为:

test/
  a.cpp
  CMakeLists.txt
  make
  build --目录,存放编译生成后的文件

编译，sh make

4.2 执行，./build/abc