* Fix: nvidia-container-toolkit
Changes:
- Dropped rootless Docker from the post script because Fedora's Docker (`moby-engine`) doesn't support this the same way
- Added postun scriptlet
- Cleaned up scriptlet using `try-restart` which should not fail if there is no service to restart
- This is how Fedora seems to do `systemctl` scriptlets
- Added `/etc/cdi/nvidia.yaml` to tracked (but ghost) config files
- Added `nvidia-driver and podman` supplements flag
Signed-off-by: Gil <rockgrub@disroot.org>
* Change: Post and postun deps
Signed-off-by: Gil <rockgrub@disroot.org>
---------
Signed-off-by: Gil <rockgrub@disroot.org>
(cherry picked from commit 6a2e4ca667)
Co-authored-by: Gil <rockgrub@disroot.org>
* add nvidia-kmod
* add settings, xsettings, mock label
* oops
* add nvidia setting assets
* add libva-nvidia-driver
* add: nvidia-kmod-common w/ properietary defaults
* download the x86 version of the driver package when running common
* add nvidia-driver package
* I don't think we actually need i386
* add a readme, fix build error and clean up script
* add back the accursed tarball script, we can port the thing later
* add nvidia-modprobe
* update metadata, add nvidia-persistenced
* add update scripts to nvidia drivers
* add update scripts for every cuda package
* make the component fetch a separate function for memoization
* cusparse
* cusparselt is not from cuda toolkit
* add nvidia subrepo label for multirepo
* set version for nvidia tarball
* nvidia-driver: download tarballs on build time
* clean up inline prep script
* NVProf is dropped in Jetpack 5, so there's no ARM version of it.
(cherry picked from commit c72c975c16)
Co-authored-by: Cappy Ishihara <cappy@cappuchino.xyz>
Raboneko