[Build error] libcudnn.so: error adding symbols: File in wrong format

[ll884856]

Issue description

When I build Caffe2 from source on my Jetson TX1, there are errors like this:

[ 93%] Linking CXX shared library ../lib/libcaffe2_gpu.so
/usr/local/cuda/lib64/libcudnn.so: error adding symbols: File in wrong format
collect2: error: ld returned 1 exit status
caffe2/CMakeFiles/caffe2_gpu.dir/build.make:2913: recipe for target 'lib/libcaffe2_gpu.so' failed
make[2]: *** [lib/libcaffe2_gpu.so] Error 1
CMakeFiles/Makefile2:1399: recipe for target 'caffe2/CMakeFiles/caffe2_gpu.dir/all' failed
make[1]: *** [caffe2/CMakeFiles/caffe2_gpu.dir/all] Error 2
Makefile:138: recipe for target 'all' failed
make: *** [all] Error 2

How can I solve the problem? Thank you!

System Info

• CUDA :8.0
• cuDNN: 6.0
• Ubuntu: 16.04

[DoctorYe]

same issue; CUDA: 9.1 cuDNN: 7.0; mac 10.13.4

[ll884856]

@DoctorYe I change to CUDA 8.0 and cuDNN 6.0 and the problem was solved.

[junweima]

same issue: CUDA 9.0, cudnn 7.2, linux 16.04

[perry0418]

I found the same issues on my TX2: CUDA: 9.0.252 cudnn: 7.3.1 linux 16.04

[ParmpalGill]

One Way to install it without cuda is use
cmake .. -DDLIB_USE_CUDA=0 -DUSE_AVX_INSTRUCTIONS=1

[Crystalxd]

I think this problem caused by the version of CUDA or CUDNN, the same issue when I compiled Pytorch on tx2. I found the version of CUDNN is 6.0.21 which is a low version for requirement, so I download higher version on Nvidia, I choose the CUDNN 7.1.4 for Linux. And now I know that this version is based on x86 instead of arm64(tx2 architecture), therefore, it shows this error. But another problem is that I don't know how to install CUDNN on tx2 what version we want to get.