libtorch1.8.1+cuda11.1+win10 vs2019环境搭建

环境说明

• libtorch: 1.8.1

• cuda: 11.1

• cudnn: 11.2

• visual studio: 2019

• NVIDIA: GeForce GTX 1050

• OS: win10

依赖安装

cuda

下载cuda11.1。

安装步骤参考CUDA Installation Guide for Microsoft Windows，一般默认下一步即可。

cudnn

下载cudnn11.2(cuDNN v8.1.1 for CUDA 11.0,11.1 and 11.2)。

安装步骤参考Installing cuDNN On Windows。

libtorch

下载libtorch1.8.1-windows-c++-CUDA11.1并解压到安装目录。

VS工程配置

1. 新建vs2019 c++空项目，右键项目->属性

   

2. 配置VC++目录下的包含目录

   

   F:\Workspace\ThirdPartLib\pytorch\cpp\libtorch-win-shared-with-deps-debug-1.8.1+cu111\libtorch\include
   F:\Workspace\ThirdPartLib\pytorch\cpp\libtorch-win-shared-with-deps-debug-1.8.1+cu111\libtorch\include\torch\csrc\api\include
   C:\Program Files\NVIDIA Corporation\NvToolsExt\include
   C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.1\include

3. 配置VC++目录下的库目录

   

   F:\Workspace\ThirdPartLib\pytorch\cpp\libtorch-win-shared-with-deps-debug-1.8.1+cu111\libtorch\lib
   C:\Program Files\NVIDIA Corporation\NvToolsExt\lib\x64
   C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.1\lib\x64

4. 配置链接器-输入下的附加依赖项

   

   

   c10.lib
   nvToolsExt64_1.lib
   cudart_static.lib
   caffe2_nvrtc.lib
   c10_cuda.lib
   torch.lib
   torch_cuda.lib
   torch_cuda_cu.lib
   -INCLUDE:?searchsorted_cuda@native@at@@YA?AVTensor@2@AEBV32@0_N1@Z
   torch_cuda_cpp.lib
   torch_cpu.lib
   cufft.lib
   curand.lib
   cublas.lib
   cudnn.lib
   -INCLUDE:?warp_size@cuda@at@@YAHXZ

5. 添加环境变量

   

   

   PATH=F:\Workspace\ThirdPartLib\pytorch\cpp\libtorch-win-shared-with-deps-debug-1.8.1+cu111\libtorch\lib;%CUDA_PATH%;%PATH%

6. 添加测试文件并编译

   #include 
   #include 
   ​
   int main(int argc, char* argv) {
     std::cout << "CUDA:   " << torch::cuda::is_available() << std::endl;
     std::cout << "CUDNN:  " << torch::cuda::cudnn_is_available() << std::endl;
     std::cout << "GPU(s): " << torch::cuda::device_count() << std::endl;
     return 0;
   }

环境验证

运行软件，得到以下结果