[c++] Cuda GPU mode isn't available on my env

C++
1

env : vs2019, libtorch 1.9.0+cu11.1 , cuda 11.1, cudnn 7.6.5

include dir
: libtorch/include
: libtorch/include/torch/csrc/api/incldue
: opencv/build/incldue
: cudnn/include
: cuda/include

lib dir
: libtorch/lib
: cudnn/lib
: opencv/build/x64/vc15/lib

add dependency
c10.lib
torch.lib
torch_cpu.lib
torch_cuda.lib
opencv_world453.lib

also i tried to add -INCLUDE:?warp_size@cuda@at@@YAHXZ on dependency

but it occured an error related to added dependency with following message.

PyTorch is not linked with support for cuda devices
Exception raised from getDeviceGuardImpl at …\c10/core/impl/DeviceGuardImplInterface.h:294 (most recent call first):

i can build/run the project with torch::cpu but i could’t run with gpu mode.

what can i do to handle the problem?

please answer me. thank you.

2

i solved it by changing a command from [-INCLUDE:?warp_size@cuda@at@@YAHXZ] to [/INCLUDE:"?ignore_this_library_placeholder@@YAHXZ"] the reason why cuda 11.1 ver has to use second command.

2 Likes