Building cuda files in a Jetson container using qemu

Robotics & Edge Computing Jetson & Embedded Systems Jetson AGX Xavier
1

Hi, as mentioned in this issue, I’ve been attempting to build cuda code in a Jetson container on an x86 workstation. This does not appear to be supported currently. Are there future plans to support this? It would be quite useful for cross compiling on my workstation.

Thanks,
Daniel

2

Hi,
It should be possible to cross-compile CUDA samples. Please check
https://devtalk.nvidia.com/default/topic/1049299/jetson-nano/cross-compile-cuda-engines/post/5325342/#5325342

3

Hi Dane,

I’m not able to cross compile CUDA samples in either docker build or docker run. If I run docker with my cuda volume mounted, as in

docker run -it -v /usr/local/cuda:/usr/local/cuda http://nvcr.io/nvidia/l4t-base:r32.3.1

then try to compile samples/5_Simulations/nbody I do get a different error message: /usr/local/cuda-10.0/bin/nvcc: Command not found

Either way, being able to build cuda code in both docker build and run would be ideal.

4

Hi,

The docker container only has L4T OS based.
You can try our deepstream container which pre-installs CUDA, cuDNN and TensorRT package.
https://ngc.nvidia.com/catalog/containers/nvidia:deepstream-l4t

Thanks.

5

Hi Aasta,

Looks like I have the same issue with the deepstream container.

Thanks,
Daniel

6

Hi,

The missing tool is deepstream container is g++ rather the nvcc.

Please try this:

$ sudo docker run -it --rm --net=host --runtime nvidia  -e DISPLAY=$DISPLAY -v /tmp/.X11-unix/:/tmp/.X11-unix nvcr.io/nvidia/deepstream-l4t:4.0.2-19.12-samples

$ apt-get install g++
$ /usr/local/cuda-10.0/bin/cuda-install-samples-10.0.sh .
$ cd NVIDIA_CUDA-10.0_Samples/5_Simulations/nbody
$ make

Thanks.

7

I was installing g++ before as specified in the container guide. Note that non cuda code compiles fine.
FYI, to run the above, I had to use gpus all instead of runtime nvidia. Also I didn’t have a cuda-install-samples-10.0.sh script anywhere, but I manually copied the samples over.

8

Hi,

Are you using deepstream-l4t:4.0.2-19.12-samples container?
If no, would you mind to give it a try?

Thanks.

9

Yup, I copied and pasted the command above.

10

Hi,

I realize the difference between us.
You are running these command on an x86 machine, right?

For the x86 environment, have you installed all these requirements?
https://github.com/NVIDIA/nvidia-docker/wiki/NVIDIA-Container-Runtime-on-Jetson#enabling-jetson-containers-on-an-x86-workstation-using-qemu

Thanks.

11

Yes I have installed them.

12

Hi,

We can reproduce this issue on our environment.

Based on the limitation mentioned here:
https://github.com/NVIDIA/nvidia-docker/wiki/NVIDIA-Container-Runtime-on-Jetson#building-jetson-containers-on-an-x86-workstation-using-qemu

Known limitation: Unfortunately you won’t be able to run any binary that calls into the NVIDIA driver on the x86 host.

The binary in the /usr/local/cuda/bin/ is not callable.
Sorry for the inconvenience.

Thanks.