Vulkan Working To Expose Video Encode/Decode, Machine Learning

[phoronix]

Phoronix: Vulkan Working To Expose Video Encode/Decode, Machine Learning

During this week's Game Developers Conference was the usual Khronos Dev Day where Vulkan, WebGL, glTF, and OpenXR took center stage. During the Vulkan State of the Union some details on their future endeavors were covered...

http://www.phoronix.com/scan.php?page=news_item&px=Vulkan-GDC-2019-Highlights

[GunpowaderGuy]

Are they keeping track of / developing their own zero cost abstractions ? It seems the new c++ standard will make the language fit to develop a vulkano alternative

[tajjada]

I, for one, approve of our new Vulkan overlords.

I am happy that the world is heading towards a common standard API for everything to do with the GPU.

[tildearrow]

Video encode/decode via Vulkan? I really hope it works out well... I wouldn't like to see it just being another competitor to VA-API/NV(ENC|DEC)/OpenMAX...

[cl333r]

exposing video encode/decode through Vulkan

Should take 5+ years till it's ready for use by end-users, e.g. handbrake fails (with no errors) to encode when choosing nvenc from the dropdown menu (libnvidia-encode-418 is installed).

[dwagner]

Sorry for the off-topic, but should we be concerned that at https://cgit.freedesktop.org/~agd5f/linux/ currently only "No repositories found" is displayed? Used to be the place where amd-staging-drm-next lived.

[M@yeulC]

Sorry for the off-topic, but should we be concerned that at https://cgit.freedesktop.org/~agd5f/linux/ currently only "No repositories found" is displayed? Used to be the place where amd-staging-drm-next lived.

Doesn't display that for me. Otherwise, I know that these repos have been slowly migrating to the freedesktop gitlab, but don't know how it is setup there.

[muncrief]

Gosh darn Vulkans.

They make me so mad.

I say expose everything!

[Imroy]

Exposing video encode/decode through Vulkan sounds like a good idea, but might be a problem on mobile devices. You see, those functions aren't actually done on the GPU and mobile SoC's have a mish-mash of functional units licensed from multiple other companies. So chips from AllWinner and Rockchip may both licence the CPU cores and Mali GPU from ARM, but they will have different encode/decode silicon. And a chip from Qualcomm will have their own Adreno GPU but may have the same encode/decode silicon as others (I don't know for sure, this is just a hypothetical).
The simple solution is to leave it up to the chip vendors - they will link together the main GPU Vulkan driver from the GPU company with a library for video encode/decode functions from another company. Anyone familiar with getting drivers out of SoC vendors (especially the Chinese ones) know that this is probably not a good idea. Good luck getting updates.
The more complicated solution is to allow video encode/decode functions to be loaded from a separate driver. This is probably better but kinda feels like the current situation with separate VDPAU/VA-API drivers.
I hope Khronos gives this some thought before making a release.

[zanny]

Obligatory XKCD link

Why did OpenMAX fail? It was Khronos last attempt at a universal standard of accelerated encode / decode support. Is there a postmortem on it to justify trying to add it to a graphics API?