Move to ubuntu 22.04 in docker files and update compute-runtime version

[kurapov-peter]

p.s. why do we have Dockerfile.clang that is identical to Dockerfile.cuda?

[kurapov-peter]

The cause of the installation failure is that cuda-drivers-525 fails to build the module (similar to https://forums.developer.nvidia.com/t/510-68-02-driver-breaks-installation-for-newer-linux-kernel-5-17-8-on-ubuntu-20-04/214531):

gcc: error: unrecognized command line option '-mharden-sls=all'
make[2]: *** [scripts/Makefile.build:297: /var/lib/dkms/nvidia/525.125.06/build/nvidia/nv.o] Error 1

Autoconf reports a compiler version mismatch:

warning: the compiler differs from the one used to build the kernel
  The kernel was built by: gcc (Ubuntu 11.3.0-1ubuntu1~22.04.1) 11.3.0
  You are using:           gcc (Ubuntu 9.5.0-1ubuntu1~22.04) 9.5.0

This is caused by our docker file to update-alternative and set gcc default version to 9.5. Setting the default version back to 11 (even though the installed one is 11.4) solved the problem locally.

[kurapov-peter]

Now it is fatal error: 'cstdint' file not found. We already saw this some time ago when we tried to update gcc. The stupid solution was to install g++-12. I tried installing libstdc++ separately and clang finds it after installation, but cstdint is still missing.

[kurapov-peter]

The build issue is resolved. Fun fact: cuda docker build now needs ~25 mins less to complete (2/3 of the usual time).

[leshikus]

LGTM

[leshikus]

@kurapov-peter

p.s. why do we have Dockerfile.clang that is identical to Dockerfile.cuda?

This reflects that the configurations match