libcuda.so failed in docker containers.

[hgandhi17]

Hello,

I am new to docker environment. I am trying to run the Nvencode sample code from my customer docker file and image and while running nvenc sample code dlopen failed to load libcuda.so

My host machine :Ubuntu 14.04 LTS
Nvidia driver version : 375.26
My docker version : 1.13

I have installed the nvidia-docker on to my machine. How i can link my host machine libcuda.so in to my custom docker image so Nvenc can find it run time.

Please guide.
Thanks,

[3XX0]

libcuda should be automatically mounted when using nvidia-docker with the official images. Your issue probably comes from your Dockerfile.

The following issue #103 (comment) might help.

[hgandhi17]

Thanks for your quick reply.

i have gone throgh the #103 (comment) , but the issue is that i can not use nvidia/cuda image directly in to my solution and so i need to make my own custom image which should support Nvidia nvenc.
I have tried to load nvidia-docker from my docker image but it fails to load
Error: Could not load NVML library

So i could not make it work.

Please guide from basic how i can load nvidia-docker from my docker file and images.

Thanks,

[3XX0]

Could you paste a sample of your Dockerfile? You're probably missing the volumes.needed label (see here)