Undefined symbol issue

[xieshuqin]

Hello, I implement a custom cpp file, successfully compile it but when trying to import it via

import torch
import grid_sampler_cuda

, I encounter the following error:

ImportError: /home/shuqin/anaconda3/envs/pytorch_1.4_py3.7/lib/python3.7/site-pa
ckages/grid_sampler_cuda-0.0.0-py3.7-linux-x86_64.egg/grid_sampler_cuda.cpython-
37m-x86_64-linux-gnu.so: undefined symbol: _Z23my_grid_sampler_2d_cudaRKN2at6Ten
sorES2_b

My torch._C._GLIBCXX_USE_CXX11_ABI is False and my complier's output:

gcc -pthread -B /home/shuqin/anaconda3/envs/pytorch_1.4_py3.7/compiler_compat -W
l,--sysroot=/ -Wsign-compare -DNDEBUG -g -fwrapv -O3 -Wall -Wstrict-prototypes -
fPIC -I/home/shuqin/anaconda3/envs/pytorch_1.4_py3.7/lib/python3.7/site-packages
/torch/include -I/home/shuqin/anaconda3/envs/pytorch_1.4_py3.7/lib/python3.7/sit
e-packages/torch/include/torch/csrc/api/include -I/home/shuqin/anaconda3/envs/py
torch_1.4_py3.7/lib/python3.7/site-packages/torch/include/TH -I/home/shuqin/anac
onda3/envs/pytorch_1.4_py3.7/lib/python3.7/site-packages/torch/include/THC -I/us
r/local/cuda-10.0/include -I/home/shuqin/anaconda3/envs/pytorch_1.4_py3.7/includ
e/python3.7m -c grid_sampler_cuda.cpp -o build/temp.linux-x86_64-3.7/grid_sample
r_cuda.o -DTORCH_API_INCLUDE_EXTENSION_H -DTORCH_EXTENSION_NAME=grid_sampler_cud
a -D_GLIBCXX_USE_CXX11_ABI=0 -std=c++11 
cc1plus: warning: command line option '-Wstrict-prototypes'  is valid for C/ObjC
but not for C++
/usr/local/cuda-10.0/bin/nvcc -I/home/shuqin/anaconda3/envs/pytorch_1.4_py3.7/li
b/python3.7/site-packages/torch/include -I/home/shuqin/anaconda3/envs/pytorch_1.
4_py3.7/lib/python3.7/site-packages/torch/include/torch/csrc/api/include -I/home
/shuqin/anaconda3/envs/pytorch_1.4_py3.7/lib/python3.7/site-packages/torch/inclu
de/TH -I/home/shuqin/anaconda3/envs/pytorch_1.4_py3.7/lib/python3.7/site-package
s/torch/include/THC -I/usr/local/cuda-10.0/include -I/home/shuqin/anaconda3/envs
/pytorch_1.4_py3.7/include/python3.7m -c grid_sampler_cuda_kernel.cu -o build/te
mp.linux-x86_64-3.7/grid_sampler_cuda_kernel.o -D__CUDA_NO_HALF_OPERATORS__ -D__
CUDA_NO_HALF_CONVERSIONS__ -D__CUDA_NO_HALF2_OPERATORS__ --expt-relaxed-constexp
r --compiler-options '-fPIC' -DTORCH_API_INCLUDE_EXTENSION_H -DTORCH_EXTENSION_N
AME=grid_sampler_cuda -D_GLIBCXX_USE_CXX11_ABI=0 -gencode=arch=compute_75,code=s
m_75 -std=c++11

I browse some relative issues and many of them suggest adding -D_GLIBCXX_USE_CXX11_ABI=0 to compiler, however this is already satisfied in my case. My pytorch version is 1.4.0, installed via conda install pytorch torchvision cudatoolkit=10.0 -c pytorch. After installing pytorch, I might have upgrade my gcc version to 7.5.0, and install horovod which also install a gxx_linux-64 via conda install gxx_linux-64`.

Could somenbody please help me on this?