Linking pytorch c++ (prebuilt GPU version) with Qt and GCC

1

I wrote a simple program in C++ to use my PyTorch model in libtorch 1.7.1 . I have CUDA 10.2 on my OS, thus I downloaded CUDA 10.2 prebuilt version of libtorch from this link. The program works fine with CMake and CUDA. CMake can load my model on GPU and returns “Yes” when I run this simple program:

#include <torch/torch.h>
#include <torch/nn.h>
#include <torch/script.h>
#include
int main(int argc, const char* argv) {
std::cout<<"Is CUDA avalaible ? "<<((torch::cuda::is_available() == 1 )?“Yes” : "No ")<<std::endl;
torch::Tensor a = torch::ones({ 2, 2 }, torch::requires_grad());
torch::Tensor b = torch::randn({ 2, 2 });
a = a.to(torch::kCUDA);
b = b.to(torch::kCUDA);
return 0;
}

Now, I would like to use the model in my Qt Application. It gives this result when I try it :

Screenshot from 2021-01-02 18-07-04
Screenshot from 2021-01-02 18-07-041107×319 88.3 KB

This is my project QMake file :

QT -= gui
CONFIG +=  c++14 console
CONFIG -= app_bundle
DEFINES += QT_DEPRECATED_WARNINGS
SOURCES += \
    main.cpp
qnx: target.path = /tmp/$${TARGET}/bin
else: unix:!android: target.path = /opt/$${TARGET}/bin
!isEmpty(target.path): INSTALLS += target

QMAKE_CXXFLAGS += -DGLIBCXX_USE_CXX11_ABI=0
CONFIG += no_keywords
QMAKE_CXXFLAGS += -Wno-unused-parameter
QMAKE_LFLAGS += -INCLUDE:?warp_size@cuda@at@@YAHXZ

INCLUDEPATH += /home/eaglesoft/libtorch/include/torch/csrc/api/include
INCLUDEPATH += /home/eaglesoft/libtorch/include
DEPENDPATH +=  /home/eaglesoft/libtorch/include/torch/csrc/api/include
DEPENDPATH +=  /home/eaglesoft/libtorch/include

LIBS += -L/home/eaglesoft/libtorch/lib -lc10 \
        -lc10_cuda \
        -lcaffe2_module_test_dynamic \
        -lcaffe2_observers \
        -ltorch_cpu \
        -ltorch \
        -lc10d_cuda_test \
        -lcaffe2_detectron_ops_gpu \
        -lcaffe2_nvrtc \
        -lfmt \
        -lnvrtc-builtins \
        -lshm \
        -ltensorpipe \
        -ltorch_cuda
CUDA_DIR = /usr/local/cuda-10.2
INCLUDEPATH  += $$CUDA_DIR/include
QMAKE_LIBDIR += $$CUDA_DIR/lib64

LIBS += -L/usr/local/cuda-10.2/lib64 -lcuda -lcudart -lcurand
LIBS += -L/usr/lib/x86_64-linux-gnu -lcublas
LIBS += -lstdc++fs