Jellyfin Docker not using Vulkan with AMD for Tonemapping

[Jademalo]

I've been running Jellyfin as a regular install, and had both transcoding and hardware tonemapping enabled.
I decided to switch to docker to make things easier to orchestrate, but I'm having a bit of an issue with hardware tonemapping.

When enabling it, for some reason Jellyfin tries to use OpenCL, which doesn't work.

Code:

/usr/lib/jellyfin-ffmpeg/ffmpeg -analyzeduration 200M -probesize 1G -f mov,mp4,m4a,3gp,3g2,mj2 -init_hw_device opencl=ocl:0.0 -filter_hw_device ocl -hwaccel vaapi -hwaccel_output_format vaapi -noautorotate -i file:"/media/exodus.mp4" -noautoscale -map_metadata -1 -map_chapters -1 -threads 0 -map 0:0 -map 0:1 -map -0:s -codec:v:0 h264_vaapi -rc_mode VBR -b:v 42805084 -maxrate 42805084 -bufsize 85610168 -profile:v:0 high -sei -a53_cc -force_key_frames:0 "expr:gte(t,n_forced*3)" -vf "setparams=color_primaries=bt2020:color_trc=smpte2084:colorspace=bt2020nc,hwdownload,format=p010le,hwupload=derive_device=opencl,tonemap_opencl=format=nv12:p=bt709:t=bt709:m=bt709:tonemap=bt2390:peak=100:desat=0,hwdownload,format=nv12,hwupload_vaapi" -codec:a:0 libfdk_aac -ac 2 -ab 256000 -af "volume=2" -copyts -avoid_negative_ts disabled -max_muxing_queue_size 2048 -f hls -max_delay 5000000 -hls_time 3 -hls_segment_type fmp4 -hls_fmp4_init_filename "0d0ced89159a4038f8259906775d5e97-1.mp4" -start_number 0 -hls_segment_filename "/cache/transcodes/0d0ced89159a4038f8259906775d5e97%d.mp4" -hls_playlist_type vod -hls_list_size 0 -y "/cache/transcodes/0d0ced89159a4038f8259906775d5e97.m3u8"


ffmpeg version 7.0.2-Jellyfin Copyright (c) 2000-2024 the FFmpeg developers
  built with gcc 12 (Debian 12.2.0-14)
  configuration: --prefix=/usr/lib/jellyfin-ffmpeg --target-os=linux --extra-version=Jellyfin --disable-doc --disable-ffplay --disable-ptx-compression --disable-static --disable-libxcb --disable-sdl2 --disable-xlib --enable-lto=auto --enable-gpl --enable-version3 --enable-shared --enable-gmp --enable-gnutls --enable-chromaprint --enable-opencl --enable-libdrm --enable-libxml2 --enable-libass --enable-libfreetype --enable-libfribidi --enable-libfontconfig --enable-libharfbuzz --enable-libbluray --enable-libmp3lame --enable-libopus --enable-libtheora --enable-libvorbis --enable-libopenmpt --enable-libdav1d --enable-libsvtav1 --enable-libwebp --enable-libvpx --enable-libx264 --enable-libx265 --enable-libzvbi --enable-libzimg --enable-libfdk-aac --arch=amd64 --enable-libshaderc --enable-libplacebo --enable-vulkan --enable-vaapi --enable-amf --enable-libvpl --enable-ffnvcodec --enable-cuda --enable-cuda-llvm --enable-cuvid --enable-nvdec --enable-nvenc
  libavutil      59.  8.100 / 59.  8.100
  libavcodec    61.  3.100 / 61.  3.100
  libavformat    61.  1.100 / 61.  1.100
  libavdevice    61.  1.100 / 61.  1.100
  libavfilter    10.  1.100 / 10.  1.100
  libswscale      8.  1.100 /  8.  1.100
  libswresample  5.  1.100 /  5.  1.100
  libpostproc    58.  1.100 / 58.  1.100
[AVHWDeviceContext @ 0x5d84884f3080] Failed to get number of OpenCL platforms: -1001.
Device creation failed: -19.
Failed to set value 'opencl=ocl:0.0' for option 'init_hw_device': No such device
Error parsing global options: No such device

Running the command to check for vulkan compatibility shows it as being set up correctly, and that ffpmeg in the container is seeing vulkan;

Code:

docker exec -it jellyfin /usr/lib/jellyfin-ffmpeg/ffmpeg -v debug -init_hw_device vulkan
ffmpeg version 7.0.2-Jellyfin Copyright (c) 2000-2024 the FFmpeg developers
  built with gcc 12 (Debian 12.2.0-14)
  configuration: --prefix=/usr/lib/jellyfin-ffmpeg --target-os=linux --extra-version=Jellyfin --disable-doc --disable-ffplay --disable-ptx-compression --disable-static --disable-libxcb --disable-sdl2 --disable-xlib --enable-lto=auto --enable-gpl --enable-version3 --enable-shared --enable-gmp --enable-gnutls --enable-chromaprint --enable-opencl --enable-libdrm --enable-libxml2 --enable-libass --enable-libfreetype --enable-libfribidi --enable-libfontconfig --enable-libharfbuzz --enable-libbluray --enable-libmp3lame --enable-libopus --enable-libtheora --enable-libvorbis --enable-libopenmpt --enable-libdav1d --enable-libsvtav1 --enable-libwebp --enable-libvpx --enable-libx264 --enable-libx265 --enable-libzvbi --enable-libzimg --enable-libfdk-aac --arch=amd64 --enable-libshaderc --enable-libplacebo --enable-vulkan --enable-vaapi --enable-amf --enable-libvpl --enable-ffnvcodec --enable-cuda --enable-cuda-llvm --enable-cuvid --enable-nvdec --enable-nvenc
  libavutil      59.  8.100 / 59.  8.100
  libavcodec    61.  3.100 / 61.  3.100
  libavformat    61.  1.100 / 61.  1.100
  libavdevice    61.  1.100 / 61.  1.100
  libavfilter    10.  1.100 / 10.  1.100
  libswscale      8.  1.100 /  8.  1.100
  libswresample  5.  1.100 /  5.  1.100
  libpostproc    58.  1.100 / 58.  1.100
Splitting the commandline.
Reading option '-v' ... matched as option 'v' (set logging level) with argument 'debug'.
Reading option '-init_hw_device' ... matched as option 'init_hw_device' (initialise hardware device) with argument 'vulkan'.
Finished splitting the commandline.
Parsing a group of options: global .
Applying option v (set logging level) with argument debug.
Applying option init_hw_device (initialise hardware device) with argument vulkan.
[AVHWDeviceContext @ 0x580f822ab6c0] Supported layers:
[AVHWDeviceContext @ 0x580f822ab6c0]    VK_LAYER_MESA_device_select
[AVHWDeviceContext @ 0x580f822ab6c0]    VK_LAYER_MESA_overlay
[AVHWDeviceContext @ 0x580f822ab6c0] Using instance extension VK_KHR_portability_enumeration
[AVHWDeviceContext @ 0x580f822ab6c0] GPU listing:
[AVHWDeviceContext @ 0x580f822ab6c0]    0: AMD Radeon Graphics (RADV RAPHAEL_MENDOCINO) (integrated) (0x164e)
[AVHWDeviceContext @ 0x580f822ab6c0] Device 0 selected: AMD Radeon Graphics (RADV RAPHAEL_MENDOCINO) (integrated) (0x164e)
[AVHWDeviceContext @ 0x580f822ab6c0] Using device extension VK_KHR_push_descriptor
[AVHWDeviceContext @ 0x580f822ab6c0] Using device extension VK_EXT_descriptor_buffer
[AVHWDeviceContext @ 0x580f822ab6c0] Using device extension VK_EXT_physical_device_drm
[AVHWDeviceContext @ 0x580f822ab6c0] Using device extension VK_EXT_shader_atomic_float
[AVHWDeviceContext @ 0x580f822ab6c0] Using device extension VK_KHR_external_memory_fd
[AVHWDeviceContext @ 0x580f822ab6c0] Using device extension VK_EXT_external_memory_dma_buf
[AVHWDeviceContext @ 0x580f822ab6c0] Using device extension VK_EXT_image_drm_format_modifier
[AVHWDeviceContext @ 0x580f822ab6c0] Using device extension VK_KHR_external_semaphore_fd
[AVHWDeviceContext @ 0x580f822ab6c0] Using device extension VK_EXT_external_memory_host
[AVHWDeviceContext @ 0x580f822ab6c0] Queue families:
[AVHWDeviceContext @ 0x580f822ab6c0]    0: graphics compute transfer (queues: 1)
[AVHWDeviceContext @ 0x580f822ab6c0]    1: compute transfer (queues: 4)
[AVHWDeviceContext @ 0x580f822ab6c0]    2: sparse (queues: 1)
[AVHWDeviceContext @ 0x580f822ab6c0] Using device: AMD Radeon Graphics (RADV RAPHAEL_MENDOCINO)
[AVHWDeviceContext @ 0x580f822ab6c0] Alignments:
[AVHWDeviceContext @ 0x580f822ab6c0]    optimalBufferCopyRowPitchAlignment: 1
[AVHWDeviceContext @ 0x580f822ab6c0]    minMemoryMapAlignment:              4096
[AVHWDeviceContext @ 0x580f822ab6c0]    nonCoherentAtomSize:                64
[AVHWDeviceContext @ 0x580f822ab6c0]    minImportedHostPointerAlignment:    4096
[AVHWDeviceContext @ 0x580f822ab6c0] Using queue family 0 (queues: 1) for graphics
[AVHWDeviceContext @ 0x580f822ab6c0] Using queue family 1 (queues: 4) for compute transfers
Successfully parsed a group of options.
Universal media converter
usage: ffmpeg [options] [[infile options] -i infile]... {[outfile options] outfile}...

Use -h to get full help or, even better, run 'man ffmpeg'

Regular hardware transcode works fine, it's specifically tonemapping that fails.
Here's my Docker Compose file, group 104 is render and the render device is passed through correctly;

Code:

services:
  jellyfin:
    image: jellyfin/jellyfin
    container_name: jellyfin
    user: 103:112
    group_add:
      - "104"
      - "44"
    network_mode: host
    volumes:
      - /etc/jellyfin:/config
      - /var/cache/jellyfin:/cache
      - type: bind
        source: /media
        target: /media
      - type: bind
        source: /media/fonts
        target: /usr/local/share/fonts/custom
        read_only: true
    devices:
      - /dev/dri/renderD128:/dev/dri/renderD128
    restart: unless-stopped
    # Optional - alternative address used for autodiscovery
    environment:
      - JELLYFIN_PublishedServerUrl=http://example.com
    # Optional - may be necessary for docker healthcheck to pass if running in host network mode
    extra_hosts:
      - host.docker.internal:host-gateway

Any ideas? I'm totally stumped at this point.
Thanks

[Jademalo]

I've got it working, but I've got a feeling there's still a bug here.

Adding

Code:

- /dev/dri/card0:/dev/dri/card0

to the docker compose file fixed it, but from my understanding that shouldn't actually be necessary?
When running Jellyfin in an LXC directly I only need to pass /dev/dri/renderD128 through for it to work, but running a docker container within that LXC requires me to pass both card0 and renderD128.

If it worked directly with just renderD128, why does the containerised version need card0 as well?
I can only assume it's something strange with hardware detection in the container, since it was attempting to use OpenCL when Vulkan was there and working. I can only assume that when it sees card0 it realises it can use vulkan, but that doesn't explain why running it directly in the LXC was fine.