Open-Source Radeon Driver Enables Support For Vulkan Video H.264/H.265 Encode

[lu_tze]

https://www.phoronix.com/news/AMD-AMF-1.4.33

- AMF on Linux can now be used with AMD Pro Vulkan, and experimentally with RADV drivers

https://github.com/GPUOpen-LibrariesAndSDKs/AMF/releases/tag/v1.4.33

https://github.com/GPUOpen-LibrariesAndSDKs/AMF/issues/457

Exactly; you linked to AMF#457 which spells it out: you need the AMF runtime, which is shipping with the proprietary AMD drivers. You could try to install the proprietary packages and run them on the top of RADV, experimentally.

So VAAPI/Vulkan is still better API to use. Your users are more likely to have the requisite packages installed and your app would work out of the box.

I don't understand what AMD intends to achieve with this. Intel doesn't hide their media engine, it is open source (edit: not fully; some interesting bits are hidden in the GuC firmware).

[shmerl]

Why is RADV_PERFTEST still needed, the code doesn't pass tests?

[castlefox]

"....here's to hoping more multimedia/desktop software begins allow using this industry-standard API for video encode/decode."'

To someone smarter than myself: How likely / how long do you think it will take for that to realistically happen?

[Quackdoc]

"....here's to hoping more multimedia/desktop software begins allow using this industry-standard API for video encode/decode."'

To someone smarter than myself: How likely / how long do you think it will take for that to realistically happen?

5 days after ffmpeg merges support into mainline. People are already using vulkan decode.

[agd5f]

Exactly; you linked to AMF#457 which spells it out: you need the AMF runtime, which is shipping with the proprietary AMD drivers. You could try to install the proprietary packages and run them on the top of RADV, experimentally.

So VAAPI/Vulkan is still better API to use. Your users are more likely to have the requisite packages installed and your app would work out of the box.

I don't understand what AMD intends to achieve with this. Intel doesn't hide their media engine, it is open source (edit: not fully; some interesting bits are hidden in the GuC firmware).

AMF uses the same underlying hardware as VAAPI or Vulkan Video and uses the same kernel code. AMF is just an AMD specific API.

[M.Bahr]

These things are still based on the GPU general-purpose compute and shader units, whereas hardware video processing is purely ASIC. If Vulkan doesn't have standardized support for ASIC video processing offload, it's just going to cause Firefox and Chrome to shy away from it because they're more focused on power efficiency. But TSG clearly has no plans to do so in the near future.

Actually i would like to see a focus on flexible gpgpu and compute shaders for video processing apis. Hardwired video codecs are a pain when it comes to backwards compatibility. Developers hate fixed function gpu "features" by the way due to the aforementioned reason. And yes i know that ASIC fixed function blocks are more efficient. But FPGAs come close as a viable alternative.

[piotrj3]

They only disadvantage I see is that Vulkan is activating the 3D hardware. If you watch a video you don't that to be activated on a laptop. But maybe there is an other way.

that is pretty much always activated.

Also in case of Nvidia there is actually one more benefit - NVDEC actually does use CUDA to certain level and driver puts clocks high when CUDA load is detected. Vulkan Video doesn't have that issue.

[lu_tze]

AMF uses the same underlying hardware as VAAPI or Vulkan Video and uses the same kernel code. AMF is just an AMD specific API.

Nobody claimed otherwise; the claim was, that in order to use AMF, you need to have proprietary binaries installed, which basically nobody has. For distributions like Debian or Fedora, they are not even supported.

So as an API for linux apps, AMF is pretty useless. Compared to that, user's won't have any problems with VAAPI or Vulkan Video, no need to tinker with their systems.

[dec05eba]

This makes no sense, any serious program would continue to use AMF/VAAPI/QSV/NVENC and not switch to this toy.

There is a huge benefit to using this on nvidia at least. If you use nvenc then the driver will use cuda internally which reduces your gpus performance because the driver will lock the power level state to max - 1 and there is no way to disable that. That decreases performance in games while recording (on my gtx 1080 that reduces fps by 10%) and it also prevents your gpu from going to a lower power level state when the gpu can easily handle decoding the video.

[dec05eba]

They only disadvantage I see is that Vulkan is activating the 3D hardware. If you watch a video you don't that to be activated on a laptop. But maybe there is an other way.

Vulkan video doesn't touch the graphics rendering pipeline. It uses the dedicated video encoding/decoding unit on the gpu that is separate from the graphics. But when it comes to playing videos you want to use your graphics cards 3d processing anyways. That allows your gpu to decode the video and display the video directly without copying it to system ram. The video encoding/decoding unit uses the graphics cards ram so the 3d graphics unit can access it directly. It's the most power efficient and performant way to play a video.
Mobile devices (such as laptops) also support low power video encoding/decoding mode (and graphics) to use even less battery.