Ndefined symbol: _ZNK2at6Tensor6deviceEv when using c++ cuda extensions

[Kracozebr (Evgeniy)]

I’m building some custom ops with cuda and c++.
When I build my extensions in pytorch 1.8.1+cu102 it works properly.
Here is the code from my setup.py:

from setuptools import setup
from torch.utils.cpp_extension import BuildExtension, CUDAExtension

if __name__ == '__main__':
    setup(
    name='my_extension',
    ext_modules=[
        CUDAExtension(
        'my_extension',
        sources=['src/my_exteinsion.cpp','src/my_extension.cu'],
    )
    ],
    cmdclass={
        'build_ext': BuildExtension
    }
)

When I try to build with pytorch 1.10.0+cu102 it builds the extension but when i try to import my extension it gives me the error:

Traceback (most recent call last):
  File "/workplace/test_custom_op.py", line 1, in <module>
    import my_extension
ImportError: /home/appuser/.local/lib/python3.6/site-packages/my_extension-0.0.0-py3.6-linux-x86_64.egg/mod_dcn_op_v2.cpython-36m-x86_64-linux-gnu.so: undefined symbol: _ZNK2at6Tensor6deviceEv

I tryied to find the similar issues, but they were unresolved:

Colab not working

Briefly: you have installed : !pip install -q torch-scatter -f https://pytorch-geometric.com/whl/torch-1.7.0+cu101.html !pip install -q torch-sparse -f https://pytorch-geometric.com/whl/torch-1.7.0+cu101.html !pip install -q torch-geometric !pip install ogb when you do this import : from torch_geometric.datasets import TUDataset you get this error : --------------------------------------------------------------------------- /usr/local/lib/python3.7/dist-packages/torch_geometric/__init_…

Torch_geometric installation issue with Colab

I followed this colab notebook to install torch_geometric to Google Colab: (I turned on the GPU of the notebook and ensured the version of torch and Cuda is 1.9.0+cu102) !pip install -q torch-scatter -f https://pytorch-geometric.com/whl/torch-1.9.0+cu102.html !pip install -q torch-sparse -f https://pytorch-geometric.com/whl/torch-1.9.0+cu102.html !pip install -q torch-geometric And it appeared to be successful: Looking in links: https://pytorch-geometric.com/whl/torch-1.9.0+cu102.html Requir…

Undefined symbol: _ZNK2at6Tensor4dataIcEEPT_v when build pytorch cuda extension

HI, Dear guys, I’m build a cuda extention for my pytorch script. However, the following error occurs. File "<frozen importlib._bootstrap>", line 684, in _load File "<frozen importlib._bootstrap>", line 658, in _load_unlocked File "<frozen importlib._bootstrap>", line 571, in module_from_spec File "<frozen importlib._bootstrap_external>", line 922, in create_module File "<frozen importlib._bootstrap>", line 219, in _call_with_frames_removed ImportError: /tmp/torch_extensions/tools/tools.so: und…

I’ve read the tutorial that describes how to build the extensions:
https://pytorch.org/tutorials/advanced/cpp_extension.html
Here is the quote

A small note on compilers: Due to ABI versioning issues, the compiler you use to build your C++ extension must be ABI-compatible with the compiler PyTorch was built with. In practice, this means that you must use GCC version 4.9 and above on Linux.

So as I understand ther error could be occurred due to compiler version mismatch. I check the gcc version and it is v. 7.5.0
Am I write about my suggestion? If so what version I should use and if I’m wrong what can be the source of error?