Embedded Python can't import torch in a C++ project

distributed distributed-rpc
1

Hi, PyTorch community!
This is my first post, so please bear with me. I have been working with Network Simulator (NS-3), a C++ code base and the project was built using CMake. I am calling “import torch” using embedding Python Interpreter as follows:

Py_Initialize();
PyGILState_STATE gilState = PyGILState_Ensure();        // Get the GIL
PyRun_SimpleString("import torch");
PyRun_SimpleString("torch.__version__");
PyGILState_Release(gilState);   // Release the GIL

The above code is part of a file called “tcp-madras.cc”, and a dynamic library called “libinternet.so” gets built using CMake.

When I run the main() executable that includes “libinternet.so” the program throws a Segmentation Fault. I did some debugging using gdb and found the following:

gdb debug error
gdb debug error1778×666 105 KB

Please note that everything works fine if I call “import torch” using a Python interpreter from the terminal.
I would be happy if someone could point me in the right direction to resolve this issue.

Environment:

# PyTorch Version: ‘2.5.0+cu124’
# Python version: 3.12.3
# OS: Windows 10
# WSL version: 2.3.24.0
# CUDA/cuDNN version: 12.6.77
# GPU: RTX 4060

2

Hi Community,
This issue has been resolved. The problem is that the torch module requires OpenSSL, whereas my project references BoringSSL. This I came to know by checking the OpenSSL version using PyRun_SimpleString(“import ssl; print(ssl.OPENSSL_VERSION)”)

Hence by including the following lines in project’s CMakeLists.txt the issue is solved:
find_package(OpenSSL REQUIRED)
target_link_libraries(myproject OpenSSL::Crypto)

1 Like