tensorflow binding undefined symbol: _ZTIN10tensorflow8OpKernelE

[Slyne]

Hi,
I used a docker with pip installed tensorflow1.14, cuda10, python3.5 and nvidia-415 on my desktop and the script passed the test_cpu, test_gpu.
When I ran

python setup.py install

and there seemed no error.

However, when I ran

python setup.py test

the error shows that 'tensorflow.python.framework.errors_impl.NotFoundError: /usr/local/lib/python3.5/dist-packages/warprnnt_tensorflow-0.1-py3.5-linux-x86_64.egg/warprnnt_tensorflow/kernels.cpython-35m-x86_64-linux-gnu.so: undefined symbol: _ZTIN10tensorflow8OpKernelE'

Before running setup.py install, I find that it first has an error like 'fatal error: not found tensorflow/core/kernels/bounds_check.h', so I just change kernels to framework and past the setup install step.

(But on my desktop, which has cuda9 and tensorflow-1.12, there seems no problem at all since bounds_check.h is under kernels/ in tf1.12 source code. I successfully built this on my desktop.)

UPDATE:
Previously I've encountered /usr/bin/ld: cannot find -ltensorflow_framework. So followed a solution on warp-ctc, I made a soft-link

ln -s /usr/local/lib/python3.5/dist-packages/tensorflow/libtensorflow_framework.so.1 /usr/local/lib/ibtensorflow_framework.so

. Then I encountered the "xxxxx as list() " problem, so I changed the registered op according to here.Then I encountered _ZTIN10tensorflow8OpKernelE error.

I successfully solved this problem by just

cd  /usr/local/lib/python3.5/dist-packages/tensorflow/; ln -s libtensorflow_framework.so.1 libtensorflow_framework.so

After that, if I directly run

python setup.py test

it will encounter '"Registering two gradient with name 'WarpRNNT'!'.
So to test the test cases, just

python tests/test_basic.py
&
python tests/test_warprnnt_op.py

[yjiangling]

Hi,
I used a docker with pip installed tensorflow1.14, cuda10, python3.5 and nvidia-415 on my desktop and the script passed the test_cpu, test_gpu.
When I ran

python setup.py install

and there seemed no error.

However, when I ran

python setup.py test

the error shows that _'tensorflow.python.framework.errors_impl.NotFoundError: /usr/local/lib/python3.5/dist-packages/warprnnt_tensorflow-0.1-py3.5-linux-x86_64.egg/warprnnt_tensorflow/kernels.cpython-35m-x86_64-linux-gnu.so: undefined symbol: ZTIN10tensorflow8OpKernelE'

Before running setup.py install, I find that it first has an error like 'fatal error: not found tensorflow/core/kernels/bounds_check.h', so I just change kernels to framework and past the setup install step.

(But on my desktop, which has cuda9 and tensorflow-1.12, there seems no problem at all since bounds_check.h is under kernels/ in tf1.12 source code. I successfully built this on my desktop.)

UPDATE:
Previously I've encountered /usr/bin/ld: cannot find -ltensorflow_framework. So followed a solution on warp-ctc, I made a soft-link

ln -s /usr/local/lib/python3.5/dist-packages/tensorflow/libtensorflow_framework.so.1 /usr/local/lib/ibtensorflow_framework.so

. Then I encountered the "xxxxx as list() " problem, so I changed the registered op according to here.Then I encountered _ZTIN10tensorflow8OpKernelE error.

I successfully solved this problem by just

cd  /usr/local/lib/python3.5/dist-packages/tensorflow/; ln -s libtensorflow_framework.so.1 libtensorflow_framework.so

After that, if I directly run

python setup.py test

it will encounter '"Registering two gradient with name 'WarpRNNT'!'.
So to test the test cases, just

python tests/test_basic.py
&
python tests/test_warprnnt_op.py

So why the error "KeyError: "Registering two gradient with name 'WarpRNNT' !" will appear when directly run python setup.py test? I still can not find out the reason.