使用onnx+libtorch在c++环境下的GPU推理设置

记录一下

onnx

包含目录

C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.3\bin;
G:\d_disk\onnxruntime-win-x64-gpu-1.10.0\include;
%(AdditionalIncludeDirectories)

附加库目录

G:\d_disk\onnxruntime-win-x64-gpu-1.10.0\lib;%(AdditionalLibraryDirectories)

库输入

onnxruntime.lib
onnxruntime_providers_cuda.lib
onnxruntime_providers_shared.lib
onnxruntime_providers_tensorrt.lib

生成後事件(dll)

xcopy /y /d /s /e "G:\d_disk\onnxruntime-win-x64-gpu-1.14.1\lib\*.dll" "$(OutDir)"

libtorch

包含目录

G:\d_disk\libtorch-win-shared-with-deps-1.12.0+cu113\libtorch\include\torch\csrc\api\include;
G:\d_disk\libtorch-win-shared-with-deps-1.12.0+cu113\libtorch\include;
%(AdditionalIncludeDirectories)

附加库目录

G:\d_disk\libtorch-win-shared-with-deps-1.12.0+cu113\libtorch\lib;
%(AdditionalLibraryDirectories)

库输入

asmjit.lib
G:\d_disk\libtorch-win-shared-with-deps-1.12.0+cu113\libtorch\lib\*.lib
c10.lib
c10_cuda.lib
caffe2_nvrtc.lib
clog.lib
cpuinfo.lib
dnnl.lib
fbgemm.lib
kineto.lib
libprotobuf-lite.lib
libprotobuf.lib
libprotoc.lib
pthreadpool.lib
torch.lib
torch_cpu.lib
torch_cuda.lib
torch_cuda_cpp.lib
torch_cuda_cu.lib
XNNPACK.lib