VA-API not available for AMD Strix Point GFX1150/GFX1151

[rcbevans]

This issue respects the following points:

• This is a bug, not a question or a configuration issue; Please visit our forum or chat rooms first to troubleshoot with volunteers, before creating a report. The links can be found here.
• This issue is not already reported on GitHub (I've searched it).
• I'm using an up to date version of Jellyfin Server stable, unstable or master; We generally do not support previous older versions. If possible, please update to the latest version before opening an issue.
• I agree to follow Jellyfin's Code of Conduct.
• This report addresses only a single issue; If you encounter multiple issues, kindly create separate reports for each one.

Description of the bug

I have a clean install of the latest jellyfin/jellyfin docker image.

I tried to enable VA-API hardware acceleration on my system with an AMD HX 370 (GFX1150).

All hardware accelerated streaming fails with an ffmpeg error

'gfx1150' is not a recognized processor for this target (ignoring processor)
'gfx1150' is not a recognized processor for this target (ignoring processor)
amd: LLVM doesn't support gfx1150, bailing out...
[AVHWDeviceContext @ 0x5fdca9b14580] libva: /usr/lib/jellyfin-ffmpeg/lib/dri/radeonsi_drv_video.so init failed
[AVHWDeviceContext @ 0x5fdca9b14580] Failed to initialise VAAPI connection: 2 (resource allocation failed).
[AVFilterGraph @ 0x5fdca9af0f00] Error initializing filters
[vost#0:0/h264_vaapi @ 0x5fdca9b5b500] Error initializing a simple filtergraph
Error opening output file /cache/transcodes/aad6d091b66a84ff6c205cb9f499f583.m3u8.
Error opening output files: Input/output error

Phoronix added GFX1150/1151 support last year and is available in LLVM 17.0.

I assume this issue here is the docker image being behind on the LLVM release?

Reproduction steps

Set up the docker image on a Strix Point AMD system.
Enable VA-API hardware acceleration.

What is the current bug behavior?

Hardware acceleration fails on HX 370 systems.

What is the expected correct behavior?

Hardware acceleration should be supported.

Jellyfin Server version

10.10.0+

Specify commit id

No response

Specify unstable release number

No response

Specify version number

No response

Specify the build version

10.10.3

Environment

- OS: Docker
- Linux Kernel:
- Virtualization:
- Clients:
- Browser:
- FFmpeg Version:
- Playback Method:
- Hardware Acceleration:
- GPU Model:
- Plugins:
- Reverse Proxy:
- Base URL:
- Networking:
- Storage:

Jellyfin logs

'gfx1150' is not a recognized processor for this target (ignoring processor)
'gfx1150' is not a recognized processor for this target (ignoring processor)
amd: LLVM doesn't support gfx1150, bailing out...
[AVHWDeviceContext @ 0x5fdca9b14580] libva: /usr/lib/jellyfin-ffmpeg/lib/dri/radeonsi_drv_video.so init failed
[AVHWDeviceContext @ 0x5fdca9b14580] Failed to initialise VAAPI connection: 2 (resource allocation failed).
[AVFilterGraph @ 0x5fdca9af0f00] Error initializing filters
[vost#0:0/h264_vaapi @ 0x5fdca9b5b500] Error initializing a simple filtergraph
Error opening output file /cache/transcodes/aad6d091b66a84ff6c205cb9f499f583.m3u8.
Error opening output files: Input/output error

FFmpeg logs

'gfx1150' is not a recognized processor for this target (ignoring processor)
'gfx1150' is not a recognized processor for this target (ignoring processor)
amd: LLVM doesn't support gfx1150, bailing out...
[AVHWDeviceContext @ 0x5fdca9b14580] libva: /usr/lib/jellyfin-ffmpeg/lib/dri/radeonsi_drv_video.so init failed
[AVHWDeviceContext @ 0x5fdca9b14580] Failed to initialise VAAPI connection: 2 (resource allocation failed).
[AVFilterGraph @ 0x5fdca9af0f00] Error initializing filters
[vost#0:0/h264_vaapi @ 0x5fdca9b5b500] Error initializing a simple filtergraph
Error opening output file /cache/transcodes/aad6d091b66a84ff6c205cb9f499f583.m3u8.
Error opening output files: Input/output error

Client / Browser logs

No response

Relevant screenshots or videos

No response

Additional information

No response

[rcbevans]

I took a look into jellyfin-ffmpeg in the image and sure enough I see radeonsi_drv_video.so is linked against libLLVM-16.so.1

[rcbevans]

In jellyfin-ffmpeg I see that if the cli-release is noble, it does use LLVM 17.

jellyfin-ffmpeg/build

Line 24 in 2f3b874

llvm_version="16"

    'noble')
        release="ubuntu:noble"
        gcc_version="13"
        llvm_version="17"
    ;;

Is there a docker image with noble as a base, or is it possible to do an in-place upgrade of the jellyfin-ffmpeg dependency to add the necessary hardware support?

[nyanmisaka]

I will prepare a Mesa/LLVM upgrade to support this new hardware. Our current linux package only supports AMD GPUs released before 2024, and we cannot upgrade it frequently to ensure stability.

[rcbevans]

Totally get that.

Is there any reason cloning jellyfin-ffmpeg, patching the bookworm gcc/llvm versions and producing a deb that I manually install into the container wouldn't work as a stopgap solution?

[nyanmisaka]

Nope, you can also do it now if you think you can handle the version upgrade and maintain the corresponding Mesa build options.

[rcbevans]

For anyone interested, I was able to get jellyfin-ffmpeg updated within the container and verify VA-API hardware acceleration is working.

• cloned jellyfin-ffmpeg,
• patched build
  • so that the bookworm debian packages use gcc 13 and llvm 17
  • patched the docker build to use host networking as the debian repo wasn't resolving to update packages
• Patched the Dockerfile to
  • Install lsb-release wget software-properties-common gnupg
  • Extend the initial "Prepare debian build environment" command to install llvm 17 per the following

&& wget https://apt.llvm.org/llvm.sh \
&& chmod +x llvm.sh \
&& ./llvm.sh 17 all \

• Ran configure
• Ran build bookworm amd64 to produce a custom deb
• Opened a shell into the running jellyfin/jellyfin container
  • Install llvm 17 using same wget, chmod, llvm.sh as above
  • Install custom jellyfin-ffmpeg7.deb
• Enable VA-API hw acceleration and verify that streams to client work as expected.

[nyanmisaka]

Debian/bookworm has backported LLVM19, so the steps of manually building LLVM can be skipped.

And Mesa needs to be updated to 24.3+.