libcuda.so: file not recognized: File truncated

[malekijoo]

I am using the image dustynv/ros:melodic-ros-base-l4t-r32.6.1. I set the /etc/docker/daemon.json as it is described in the Read.me, I also installed nvidia-docker2 as its instruction. My Nvidia driver is 470.63.01 and Cuda 11.4 on the host. The image has Cuda 10.2. When I compile an application called darknet-ros with C++ and make with catkin as its instruction the Errors come up.

Errors     << darknet_ros:make /root/yolo_ws/logs/darknet_ros/build.make.001.log                                                                                       
/usr/lib/gcc/aarch64-linux-gnu/7/../../../aarch64-linux-gnu/libcuda.so: file not recognized: File truncated
collect2: error: ld returned 1 exit status
make[2]: *** [/root/yolo_ws/devel/.private/darknet_ros/lib/libdarknet_ros_lib.so] Error 1
make[1]: *** [CMakeFiles/darknet_ros_lib.dir/all] Error 2
make: *** [all] Error 2

The LD_LIBRARY_PATH and the aarch64-linux-gnu in docker are as follows,

root@lab:/# ls -lah /usr/lib/x86_64-linux-gnu/libcuda*
lrwxrwxrwx 1 root root  12 Oct 26 08:21 /usr/lib/x86_64-linux-gnu/libcuda.so -> libcuda.so.1
lrwxrwxrwx 1 root root  20 Oct 26 08:21 /usr/lib/x86_64-linux-gnu/libcuda.so.1 -> libcuda.so.470.63.01
-rw-r--r-- 1 root root 23M Aug  3 20:51 /usr/lib/x86_64-linux-gnu/libcuda.so.470.63.01

root@lab:/# echo $LD_LIBRARY_PATH
/opt/ros/melodic/lib:/usr/local/cuda-10.2/targets/aarch64-linux/lib

In 856 a similar error came up in which building without GPU support and "default-runtime": "nvidia" was suggested to solve. The application I am using needs to access the GPU of the host to compile with C++ and run.

I was wondering if it is a bug originating from building dustynv/ros:melodic-ros-base-l4t-r32.6.1 or it is on my side? Can anyone tell me what is wrong?

[camelator]

same issue here following this step by step https://yunusmuhammad007.medium.com/build-and-install-opencl-on-jetson-nano-10bf4a7f0e65 as result, I am not able to find a way to compile opencv within this container

[wrickout]

same issue on r35.1 on noetic cuda 11.x

[dusty-nv]

My Nvidia driver is 470.63.01 and Cuda 11.4 on the host. The image has Cuda 10.2.

These container images are built for Jetson/ARM platform and not x86. On x86, you could either use OSRF's official ROS containers as your base container, or change the base container that I build with to something other than l4t-base. l4t-base is for Jetson/ARM has stubs for those cuda libraries in it which is probably what is causing you that error.