not found tf_sampling_so.so during the training.

[MrCrazyCrab]

@charlesq34 Charles R. Qi, i follow your advice to compile the tf, but when i run the program, the software warns me that the tf_sampling_so.so can not be found and the "undefined symbol: _ZN10tensorflow12OpDefBuilder4AttrESs". i'm sure that the progress of compiling the tf is right. i use tensorflow 1.13.1, cuda 10.0, anaconda 10.130 and spyder 3.3.6. Hope for you reply! Thanks!

[sheshap]

Use below commands and see if it works for you.

/usr/local/cuda/bin/nvcc --gpu-architecture=sm_35 --compiler-options -Wall -I/usr/local/cuda/include -L/usr/local/cuda/lib tf_sampling_g.cu -o tf_sampling_g.cu.o -c -O2 -DGOOGLE_CUDA=1 -x cu -Xcompiler -fPIC -lcusolver -lcurand -lcublas -lcusparse -g

TF_CFLAGS=$(python -c 'import tensorflow as tf; print(" ".join(tf.sysconfig.get_compile_flags()))') TF_LFLAGS=$(python -c 'import tensorflow as tf; print(" ".join(tf.sysconfig.get_link_flags()))')

g++ -std=c++11 -shared tf_sampling.cpp tf_sampling_g.cu.o -o tf_sampling_so.so -fPIC ${TF_CFLAGS[@]} ${TF_LFLAGS[@]} -I /usr/local/cuda/include -lcudart -L /usr/local/cuda/lib -O2 -D_GLIBCXX_USE_CXX11_ABI=0 -lcusolver -g

[jtx1999]

Please see if https://github.com/charlesq34/pointnet2/issues/48 solves your problem.

[wumingbai]

did you solve the problem

[jtx1999]

Yes, by removing -D_GLIBCXX_USE_CXX11_ABI=0

[MrCrazyCrab]

@jtx1999 @sheshappanavar it still won't work when running tensorflow1.13.1. do you know how to slove it?

[manishmaruthi]

@sheshap Thank you so much for posting the commands, in my case it solved the issue