CUDA 9.1 compile samples fails; looks in incorrect folder for lnvcuvid (nvidia-390.48 driver/ Ubuntu16.04)

Accelerated Computing CUDA CUDA Setup and Installation
1

Hello, I noticed that the findgllib.mk in the compile samples resolved UBUNTU_PKG_NAME to -L/usr/lib/nvidia-387 instead of /usr/lib/nvidia-390.
I have no /usr/lib/nvidia-387 but yes /usr/lib/nvidia-390 so compile fails

I installed the local deb after having installed driver 390.48.

  • which dpkg >dev/null 2>&1 && dpkg l ‘nvidia-*’ | grep ‘^ii’ | awk ‘{print $$2}’ | head -i shows both entries
    ii nvidia-387 …
    ii nvidia-387-dev
    ii nvidia-390
    ii nvidia-390-dev
    but the driver version for all of those is 390.48-0ubuntu0-gpu16.04.3 ; on the -387 entries the description reads, Transitional package.
  • nvidia-smi displays the correct driver 390.48

I cannot sudo apt-get uninstall nvidia-387 because it will carry 390 as well as cuda9.1 with it. It seems everything is correct except the presence of those transition packages, but the mk script fails because of this.

What is the simplest, cleanest way to fix this.?

thanks.

2

I just skipped it, every other one works fine, i just wonder whether those transitional packages arise from my installations and i should somehow remove them ( and then how) or they are there ok, and the distributed make scripts need adjustment.
thanks