I can't install caffe2 on tx2

I try build caffe2 from source but failed
so I used tegra tx1 script to install on tx2,but have this problem

/usr/local/cuda-9.0/lib64/libcudnn.so: error adding symbols: file in wrong format
collect2: error: ld returned 1 exit status
/caffe2/CMakeFiles/caffe2_gpu.dir/build.make: 6255: recipe for target ‘lib/libcaffe2_gpu.so’ failed
Cmake[2]: ***[lib/libcaffe2_gpu.so] Error 1
CMakeFiles/makefile2:1638: recipe for target ‘caffe2/CMakeFiles/caffe2_gpu.dir/all’ failed
Cmake[1]: ***[caffe2/CMakeFiles/caffe2_gpu.dir/all] Error 2
CMakeFile:138: recipe for target ‘all’ failed
cmake : ***[all] Error 2

by the way, now my tx2 ubuntu system can not open terminal,and GUI has no menu and enter anything.
I just can use ctrl+alt +F1/F2/F3/F4/F5/F6 to control tx2.

how to fix this two problems? thanks…

Hi,

Which source do you use? Caffe2 or pyTorch?

You can reboot to fix the GUI issue.
Thanks.

I use pytorch .

Reboot is no useful.the case is just like install wrong gpu driver. But my device didn’t install gpu driver.

What’s more ,my mouse become x, I can’t use keyboard shortcuts to control system. I just use ctrl+alt+f1 to use command.

See if all are ok from:

sha1sum -c /etc/nv_tegra_release

This looks suspect. Have you installed CUDA 9 from JetPack or installed yourself ?
What gives:

file /usr/local/cuda-9.0/lib64/libcudnn.so

I have installed cuda all ready . I will try to use sha1sum command.

I used sha1sum, that showed all .so files are OK.

I used sha1sum, that showed all .so files are OK.

I used that, then they show as follow:

/usr/local/cuda-9.0/lib64/libcudnn.so: ELF 64-bit LSB shared object, x86-64, version 1(SYSV), dynamically linked, stripped

now I can enter gui normally.I’m trying to install caffe2 the N times.

Hi,

It looks like your CUDA package is for desktop:

/usr/local/cuda-9.0/lib64/libcudnn.so: ELF 64-bit LSB shared object, <b>x86-64</b>, version 1(SYSV), dynamically linked, stripped

Do you install it from JetPack?
Please noticed that you need to install ARM version for Jetson, which is only available in JetPack.

Thanks.

ok, I remembered I install cuda from JetPack,because I run the cuda simple after installed. I will check whether it install another version cuda. Maybe it soft links to the new version cuda.

Hi,
Do you have wheel of pytorch-1.0.0 -aarch64 based cuda8.0?
I have same error '/usr/local/cuda-8.0/lib64/libcudnn.so: error adding symbols: file in wrong format
collect2: error: ld returned 1 exit status ', when installed pytorch from source.

I will check this problem after exam on Friday.

Thanks .