from ._trilinear import lib as _lib, ffi as _ffi ImportError: /home/work/Project/Image-Adaptive-3DLUT-master/trilinear_c/_ext/trilinear/_trilinear.so: undefined symbol: __cudaRegisterFatBinaryEnd

[huizeng]

from ._trilinear import lib as _lib, ffi as _ffi ImportError: /home/work/Project/Image-Adaptive-3DLUT-master/trilinear_c/_ext/trilinear/_trilinear.so: undefined symbol: __cudaRegisterFatBinaryEnd

[HuiZeng]

Hi, I have no idea on this problem :(

[lld533]

Hello, your work is very good. How to solve this problem? Do you need to add this function trilinear_func.py? Can you update this function to your code in the github?

Hello, this may come from mismatch of your pytorch and CUDA version. To be short, please check your pytorch and CUDA version and compile the code again.
If you'd like to use the ffi for compilation, please make sure your PyTorch (pre-compiled, e.g. installed via pip or conda) version is lower than 1.0, e.g. 0.4.0 or 0.4.1. Meanwhile, you are using CUDA 9.0 or some other compatible version on your machine, instead of CUDA 10.1 or 10.2.

For NV 20/30 series card users, it would be appreciated if you could shift to PyTorch 1.0 or later and compile the cpp code, instead of c code.

Many thanks!

[henbucuoshanghai]

cuda 9.1 is ok? @lld533