[Announce] VA-API/VPP Support for vdr-plugin-softhddevice

rofafor · 20. Januar 2015

The 2015 is the year of the VA-API/VPP in VDR! The latest VA-API/VPP code is now available from Antti Seppälä's Github repositories instead of traditional patches and it comes at the moment in two flavors:

The main VA-API/VPP development:

https://github.com/pesintta/vdr-plugin-softhddevice (or https://github.com/pesintta/vd…hddevice/tree/vpp_support)
the successor of the former v10 patch
supports the va-api-glx output
supports ffmpeg's libavfilters before and after GPU post-processing
you can even mix yadif deinterlacer and VA-API color post-processing
pre/postfilters are executed 25Hz, unless VPP deinterlacing is enabled. In this case the PostAvFilter is 50Hz.
thou shall not modify frame rate nor resolution in avfilters!

Zitat

These GIT repos adds support for Intel VA-API VPP filters to softhddevice. It is expected that many
things that used to work are now broken (different frame rates, etc.) or does not work well enough for
production use.

It is strongly recommended to use git master version of libva-intel-driver together with this patch
(http://cgit.freedesktop.org/vaapi/intel-driver/?h=master). Minimum recommended released version is
libva-intel-driver git-5c239249 which contains necessary fixes for the plugin to function properly.

Ivybridge, Sandybridge, and Baytrail users will get better deinterlacing options by using the
libva-intel-driver version 4a1c4d21f3428b08ef765d7f7de75b97006514ac or greater.

The patch was/is tested and developed on:
* ffmpeg 2.2.13
* libva 1.5.0
* libva-intel-driver git-5c239249
* xf86-video-intel 2.99.917
+ Intel D54250WYK NUC kit (Haswell i5)

Alles anzeigen

Example of required command line parameters:

Code

~# vdr -P 'softhddevice -a hw:0,3 -d :0.0 -f -v va-api'
~# vdr -P 'softhddevice -a hw:0,3 -d :0.0 -f -v va-api-glx'

One can check the current GPU load with the intel_gpu_top application from x11-apps/intel-gpu-tools package:

Code

~# emerge x11-apps/intel-gpu-tools # using Gentoo
~# intel_gpu_top

All credits go to Antti Seppälä for his superb work!

fnu · 20. Januar 2015

Zitat von rofafor

All credits go to Antti Seppälä for his superb work!

Thanks for that passion

helau · 20. Januar 2015

Thanks for the plugin !
Are there any plans of merging the original softhddevice from Johns with this code ?

rofafor · 20. Januar 2015

Zitat von helau

Are there any plans of merging the original softhddevice from Johns with this code ?

That's up to Johns. One of the reason for providing a fork is to ease up a possible merge back to the Johns by providing the VA-API/VPP commit history.

johns · 21. Januar 2015

Zitat von rofafor

That's up to Johns. One of the reason for providing a fork is to ease up a possible merge back to the Johns by providing the VA-API/VPP commit history.

I have no idea how to handle the merge correctly.
New code is always good.

The problems are;

Coding style
I want understand the new code
New bugs in old part

Johns

johns · 21. Januar 2015

Zitat

The experimental libavfilter support for VA-API/VPP:

This was also on my todo list.

Have you added a new filter thread with its own ring buffer?

Johns

NetWraith · 21. Januar 2015

Great job! Thank you! Althought I have a lack of knowledge about coding !

I've started testing va-api support on an Intel NUC D54250WYKH yesterday with Gen2VDR V5.1. A good time to get in deeper .

helau:
I will post my experience with your new distri and VAAPI soon at your board.

mni tnx e vy 73 ...
NetWraith

rofafor · 21. Januar 2015

Zitat von johns

The problems are;

Coding style

I want understand the new code

New bugs in old part

... and now you have Github's tools to comment unwanted coding styles and ask any code specific questions easily. Your original code base is found and mirrored into the "master" branch.

rofafor · 21. Januar 2015

Zitat von johns

Have you added a new filter thread with its own ring buffer?

Nope. It's done in the display handler thread as you can easily see here: https://github.com/pesintta/vd…e/libavfilter_integration

johns · 21. Januar 2015

New thread for discussion?

I tested the today version:

gentoo-sources 3.18.1
xorg-server 1.16.2.901
xf86-video-intel 9999
mesa 10.3.5
libva 9999
libva-intel-driver 9999
ffmpeg 2.2.11

9999 is GIT from today.

Kernel cmdline: i915.powersave=1 i915.i915_enable_fbc=1 i915.i915_enable_rc6=1

Celeron(R) CPU G1610 @ 2.60GHz Monitor 50Hz 1920x1080 HDMI Audio Analog.
video/vaapi: motion adaptive deinterlace supported

softhddevice.xxxx.FirstField = 2
softhddevice.xxxx.SecondField = 1

with -v va-api

720p works fine.
576i N24 needs some seconds to run smooth
Every few minutes i get "vdr: video/vaapi: time/frame too long 36ms".
1080i permanent "video: speed up video, droping frame"
This is an old bug from Intel, this is why i added the GLX output.

There are some artifacts with scrolling text lines.

with -v va-api-glx

There is a scaling failure: yellow bar right and left. (I think my bug)

720p works fine.
576u N26 needs some seconds to run smooth
1080i seems to run fine now, with the first test i got many "vdr: video/vaapi: time/frame too long 36ms".

Johns

fnu · 21. Januar 2015

Zitat von johns

New thread for discussion?

Yes, sure, I can move it out starting with your last post?

Space · 21. Januar 2015

Hi Johns,

regarding the artifacts ... that might be solved with the branch by Gwenole (19.vpp.adi ... see announcement for git location) because that might fix these bugs and even provide MCDI on Ivy Bridge ...

Currently my new VDR with VA-API is connected to a monitor with 60Hz output only so I can not really test. Maybe this weekend

Thanks a lot to all people involved. This looks really really promising ...

Best regards,

Space

fnu · 22. Januar 2015

Zitat von Space

Gwenole (19.vpp.adi ... see announcement for git location) because that might fix these bugs and even provide MCDI on Ivy Bridge

I do run that branch on my BayTrail, very nice with "va-api-glx", but it doesn't contain Antti's latest fixes to the Intel drivers and therefor I cannot activate Bob with it.

The commit mentioned by rofa do contain the fixes, but there is no MCDI available for IVB/BayTrail. But difference between MADI and MCDI is subtle like at VDPAU's temporal to temporal_spatial. The scrolling text of N24 (SD) looks better with MADI, from my point of view.

Regards
fnu

Space · 22. Januar 2015

Zitat von fnu

I do run that branch on my BayTrail, very nice with "va-api-glx", but it doesn't contain Antti's latest fixes to the Intel drivers and therefor I cannot activate Bob with it.

The commit mentioned by rofa do contain the fixes, but there is no MCDI available for IVB/BayTrail. But difference between MADI and MCDI is subtle like at VDPAU's temporal to temporal_spatial. The scrolling text of N24 (SD) looks better with MADI, from my point of view.

Hi,

that's why I am currently running GIT master + cherry pick of the 3 commits from the 19.vpp.adi tree by Gwenole. And apart from the stuttering caused by 60Hz display it seems to look fine. Need to check on the big screen, hopefully this weekend.

Best regards,

Space

johns · 22. Januar 2015

Only the VA-API-GLX works good.

But my GLX code is not good. The complete decoder and display handler runs in the same thread.
As soon there is a little load on the CPU (Celeron G1610) i get the "vdr: video/vaapi: time/frame too long ...".

Johns

Space · 22. Januar 2015

If someone else wants to test the branch of Gwenole on top of the master vaapi-intel tree find instructions below how I added those commits to my local git master:

Code

git clone -b master http://anongit.freedesktop.org/git/vaapi/intel-driver.git
cd intel-driver
git remote add gbeauchesne https://github.com/gbeauchesne/libva-intel-driver.git
git fetch gbeauchesne
git cherry-pick 8294e43e809923eb6f8231a085c76595028cafad
git cherry-pick ec6e26656c580871787343c71b4c6f670911775f
git cherry-pick 19256196e01eb1184425f684eb19e74f71e2d438

But it seems that I have problems with it ... I zapped some time and it worked fine but now the system does not respond any more ... need to dig a little bit deeper into this on the weekend. Back to work now

Best regards,

Space

PS: Johns: no, I can not force the display to 50Hz mode ... it says invalid signal then ... it's a *very* old LCD monitor.

rofafor · 22. Januar 2015

FYI: Gwenole's repo contains also support for advanced (read: HQ) scaling if your hardware supports the feature.

johns · 22. Januar 2015

Zitat von Space

But it seems that I have problems with it ... I zapped some time and it worked fine but now the system does not respond any more ... need to dig a little bit deeper into this on the weekend. Back to work now

I had the same problem with all freedesktop GIT, i think this is a general intel problem.

Johns

rofafor · 22. Januar 2015

Zitat von johns

I had the same problem with all freedesktop GIT, i think this is a general intel problem.

Not that generic as my Haswell is working fine on both repos. I don't even get those "video: speed up video, droping frame" messages, so it must a hardware generation specific driver problem: a bug or glx execution path is just a bit better optimized.

fnu · 22. Januar 2015

Well, looking to the different hardware used, Haswell i5 with 40 EUs and IvyBridge with just 6 or even 4 on the sibling BayTrail might explain some of these issues. The 40 EUs are not stressed that much by the tiny little nicks and nacks here and there ...

I'm really happy that is does run in the BayTrail GPU as it is actually. Ok, there seems to be room for improvement, as always, and 1080i with "va-api-glx" drives the little GPU to it's boundaries. So having Bob available to lower pressure in case needed would be good, let's see how to find a good libva-intel-driver version ...

Regards
fnu

[Announce] VA-API/VPP Support for vdr-plugin-softhddevice

Jetzt mitmachen!

Teilen