“未定义标识符 "size_t"”, undefined symbol: __cxa_call_terminate

[justliulong]

似乎按照教程上方法来配置前面的基础文件，但是我在cpp中写python的拓展模块时：

PYBIND11_MODULE(TORCH_EXTENSION_NAME, m) {
    m.def("demo", &demo, "cuda demo");
}

这里提示我“未定义标识符 "size_t"”，我确保的torch已经正确安装，并且能够使用：

Python 3.8.18 (default, Sep 11 2023, 13:40:15) 
[GCC 11.2.0] :: Anaconda, Inc. on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> import torch
>>> print(torch.cuda.is_available())
True
>>> print(torch.version.cuda)
12.1
>>> print(torch.__version__)
2.2.1+cu121

但是我就是不能使用能够编译的拓展模块，他给我这样的报错：

Traceback (most recent call last):
  File "test.py", line 2, in <module>
    import cuda_demo
ImportError: /home/ll/.conda/envs/dl/lib/python3.8/site-packages/cuda_demo.cpython-38-x86_64-linux-gnu.so: undefined symbol: __cxa_call_terminate

我尝试修改了.bashrc文件使其能够链接到我的动态链接上export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/home/ll/.conda/envs/dl/lib/python3.8/site-packages/torch/lib 即使现在能拓展模块能链接到对应的动态库：

$ ldd /home/ll/.conda/envs/dl/lib/python3.8/site-packages/cuda_demo.cpython-38-x86_64-linux-gnu.so
        linux-vdso.so.1 (0x000073c384dc7000)
        libc10.so => /home/ll/.conda/envs/dl/lib/python3.8/site-packages/torch/lib/libc10.so (0x000073c384c95000)
       ...

但是程序依旧没法成功运行，cpp文件中还是提示我“未定义标识符 "size_t"”，运行程序依旧报相同的问题undefined symbol: __cxa_call_terminate

[stanleyedward]

same issue here

[justliulong]

same issue here

I tried some ways to run it. This has nothing to do with the version of Python itself, but it seems to be related to the version of torch. At least in my test, I used a lower version of torch (torch == 1.13) to run successfully and avoid this error. Maybe other higher versions can also avoid this problem, but I tested torch==2.0.1 and above, and it didn't work. I don't know how the upstream encapsulates the updated version of torch.

Since I recently tried to build the mamba_ssm model and got the same undefined error, I haven't tried to use other versions of torch (due to cuda, my virtual environment is using a newer torch recently), and I hope to get more effective suggestions from others.

[Xu-jinhua]

这是因为2.0或者更新的版本已经弃用了template <typename scalar_t>的用法，为了配合教程请使用1.13之前的版本，或者使用更新后的教程Custom C++ and CUDA Operators

It's because this template <typename scalar_t> has been deprecated, and if you want to use this tutorial you need to choose older pytorch like 1.13. Or you can use a new tutorial like this one Custom C++ and CUDA Operators

[justliulong]

这是因为2.0或者更新的版本已经弃用了template <typename scalar_t>的用法，为了配合教程请使用1.13之前的版本，或者使用更新后的教程Custom C++ and CUDA Operators

It's because this template <typename scalar_t> has been deprecated, and if you want to use this tutorial you need to choose older pytorch like 1.13. Or you can use a new tutorial like this one Custom C++ and CUDA Operators

感谢你的回复和解释！我以后会更加注意官方的公布，我将关闭这个issue，也希望你在编程工作上前程似锦！

Thank you for your reply and explanation! I will pay more attention to official announcements in the future. I will close this issue and wish you good luck in your future work with programming!