Linux - vaapi-sse4: Deinterlacing Testing

  Thread Rating:
  • 12 Votes - 4.75 Average
  • 1
  • 2
  • 3
  • 4
  • 5
Post Reply
FernetMenta Online
Team-Kodi Developer
Posts: 3,185
Joined: Jul 2010
Reputation: 63
Location: Munich
Post: #631
(2014-05-14 23:08)puithove Wrote:  So just out of curiosity - and to see if we really are playing with the same thing - does playing my sample through the vnsi parser work?

vnsi has trouble with PAT/PMT of your sample. I downloaded TS-Doctor, a little tool for analyzing mpegts. It did also indicate an issue with pat/pmt in this file. Not sure if this is an additional issue or related.
find quote
henfri Offline
Junior Member
Posts: 3
Joined: Mar 2009
Reputation: 0
Post: #632
Hello,

thanks for the work on this and the great tutorial (maybe you could add the information, that in the Video-Settings you need to activate the expert mode to see all the options).

What CPU-Load would you expect/do you get when (1) running xbmc (main-screen) and (2) playing a 1080i video on an Intel® Pentium® CPU G3220 @ 3.00GHz?
The Video I used is this (any better recommendations?)

I get (1) 20% and (2) 52%, which I find a bit high. Also, I am not sure, the deinterlacing works properly, as I see some horizontal lines in the Shuttle's exhaust..

When I open the video settings during playback, no deinterlacing is selected (!). When I change that to bob, the situation does not improve. Only the cpu-usage goes down to about 40%.

Is this expected?

Here the pastebin-links:
http://paste.ubuntu.com/7483192/
http://paste.ubuntu.com/7483193/
http://paste.ubuntu.com/7483194/
http://paste.ubuntu.com/7483195/
http://paste.ubuntu.com/7483198/

Regards,
Hendrik
(This post was last modified: 2014-05-18 14:00 by henfri.)
find quote
fritsch Online
Team-XBMC Developer
Posts: 5,937
Joined: Aug 2011
Reputation: 139
Location: Karlsruhe
Post: #633
Quote:When I open the video settings during playback, no deinterlacing is selected (!). When I change that to bob, the situation does not improve. Only the cpu-usage goes down to about 40%.

Is this expected?

Do the settings the howto wants. Deinterlacing: Auto -> Deinterlacing Method: Deinterlace (save for all files).

Concerning the idle usage, compare with the clock of your CPU. If it's 20% when the CPU is at 800 Mhz, then it's not an issue at all.

Yadif, e.g. Deinterlace, is running on the CPU so a 60% load is to be expected.

->>>>> Your user is not member of the audio group nor the video group, therefore you don't have any audio devices available. Did you miss that step in the howto?

You missed to post the xbmc.log
find quote
henfri Offline
Junior Member
Posts: 3
Joined: Mar 2009
Reputation: 0
Post: #634
Hello,

thanks for your reply.
On the audio:
Quote:groups xbmc
xbmc : xbmc dialout cdrom audio dip video plugdev users
Looks right, doesn't it?

Quote:Do the settings the howto wants. Deinterlacing: Auto -> Deinterlacing Method: Deinterlace (save for all files).
Where do you mean? Under Settings, I did do these. During the playing of the video?
Ah, I see, I missed this:
Quote:When playing some 720p file or lower (do you get why?) open the movie icon and set Deinterlace: Auto and Deinterlace Method to: Deinterlace (if it causes too much cpu load, choose Bob (Celeron Baytrail and Celeron 847 Nucs(!)) - also set Scaling algorithm to Lanczos 3 optimized (also not for Baytrail and Celeron Nucs, only for HD3000 and faster). Save for all files.
I will do that.

Regarding the xbmc.log: That was because I was logged in as myself, not as xbmc. Thus, line from your first post does not work, at least here.
cat ~/.xbmc/temp/xbmc.log | pastebinit leads to /home/henfri/.xbmc/temp/xbmc.log. The log is at /home/xbmc/.. though.
http://paste.ubuntu.com/7484914/

Regarding the cpu-load: Do you have a vision, if/when the GPU could do the deinterlacing?

Greetings,
Hendrik
find quote
fritsch Online
Team-XBMC Developer
Posts: 5,937
Joined: Aug 2011
Reputation: 139
Location: Karlsruhe
Post: #635
When VPP is fixed. I tested it today again. It works flawless on IVB, but Haswell only can do BOB with a special patch that was made cause of my constant nagging in vaapi-intel-driver ... I won't do anything to get VPP into xbmc direction if the driver is not fixed for hsw.

If you are on IVB, see my vpp branch, i have rebased btbn's work onto Gotham and parked it there.
find quote
stingray454 Offline
Junior Member
Posts: 8
Joined: Feb 2014
Reputation: 0
Post: #636
Hello, im using your ubuntu repo for my intel nuc, and it works very good.

i got a little issue thought i have tried two spotify plugins and both shows same sympthoms, that first song in playlist after a stopped playback plays as it should, sometimes it plays two full tracks before it cuts of, usually around a minute, after it has cutoff first time, its silent, but keeps playback timer and change track, next track plays 10-15seconds, sometimes up to 30-40seconds then it goes silent again.

i guess its the libspotifycore that is the issue, but it works for most people using regular gotham betas and i guess there might be something with intel audio hardware? but i guess not..

just thought i would ask here too, i have posted in the spotlight plugin thread too, just dont know whats wrong and why..

any thoughts?

ps sometimes it doesnt even play the first track fully and cutsoff randomly from middle to the end..
find quote
fritsch Online
Team-XBMC Developer
Posts: 5,937
Joined: Aug 2011
Reputation: 139
Location: Karlsruhe
Post: #637
No - no thoughts.

You are completely off topic and did not even provide logfiles.
find quote
puithove Offline
Junior Member
Posts: 42
Joined: Sep 2009
Reputation: 0
Post: #638
(2014-05-15 08:48)FernetMenta Wrote:  vnsi has trouble with PAT/PMT of your sample. I downloaded TS-Doctor, a little tool for analyzing mpegts. It did also indicate an issue with pat/pmt in this file. Not sure if this is an additional issue or related.

I had originally thought that forcing Deinterlace=ON would work. However, it does make some of the affected content play well, but other times it's still messed up. So at this point, I'm stuck leaving Deinterlace=Auto, Deinterlace Method=Auto Select. Which effectively puts me on Bob it seems.

I went as far as compiling a custom build with different yadif filter settings, and also tried the w3fdiff just for fun (yea, it couldn't keep up even with the "simple" coefficients - but it was a fun experiment anyway). Didn't fix my issue

FernetMenta & fritsch - I do want to say thanks for taking a look at my issue. Hopefully something can be worked out in the future. Please keep the samples I sent you to test with. If you need any more, PM me and I'll be glad to grab some.
find quote
fritsch Online
Team-XBMC Developer
Posts: 5,937
Joined: Aug 2011
Reputation: 139
Location: Karlsruhe
Post: #639
@puithove: What about the quality difference, could see any noticable difference? with w3fdiff?

I don't really:
https://dl.dropboxusercontent.com/u/5572...-yadif.mp4
https://dl.dropboxusercontent.com/u/5572...w3fdif.mp4

https://dl.dropboxusercontent.com/u/5572...-yadif.mp4
https://dl.dropboxusercontent.com/u/5572...w3fdif.mp4
(This post was last modified: 2014-05-20 09:28 by fritsch.)
find quote
puithove Offline
Junior Member
Posts: 42
Joined: Sep 2009
Reputation: 0
Post: #640
I did notice a small difference in aliasing - in fact in the sample you sent me, there was considerably less aliasing in the Seavus logo with w3 than there was with yadif. Not really noticeable with video in motion, but can definitely see it in the static logo. I didn't do a whole lot of other comparisons though. If the performance could somehow be optimized, it might actually bring some quality difference, but perhaps not enough until we have stronger CPUs.

I was able to play video OK on my i7 workhorse machine in XBMC with w3, but the i5 NUC just couldn't keep up.
find quote
akya Offline
Member
Posts: 96
Joined: Feb 2012
Reputation: 1
Location: /home/akya
Post: #641
Hi, I am not sure what happened recently but as of today XBMC started crashing every time I played a video. Checked the crash log and looks like the bug has been reported before -
Code:
https://www.mail-archive.com/debian-bugs-dist@lists.debian.org/msg1212597.html
My log -
Code:
http://xbmclogs.com/show.php?id=207623

XBMC 13.1 Beta 1 was successfully able to decode with Hardware acceleration. For now I switched to Software acceleration but thought I should post the log here. I hope I am posting in correct section.

~Web and graphic designer... Android Developer... A Gooner~
find quote
fritsch Online
Team-XBMC Developer
Posts: 5,937
Joined: Aug 2011
Reputation: 139
Location: Karlsruhe
Post: #642
Stop using VAAPI on an nvidia device. You have a broken wrapper installed - that crashes the driver.
(This post was last modified: 2014-05-22 17:14 by fritsch.)
find quote
akya Offline
Member
Posts: 96
Joined: Feb 2012
Reputation: 1
Location: /home/akya
Post: #643
(2014-05-22 17:14)fritsch Wrote:  Stop using VAAPI on an nvidia device. You have a broken wrapper installed - that crashes the driver.

That worked.. Thanks Smile

~Web and graphic designer... Android Developer... A Gooner~
find quote
noggin Offline
Posting Freak
Posts: 902
Joined: Oct 2008
Reputation: 25
Location: United Kingdom
Post: #644
(2014-05-20 09:27)fritsch Wrote:  @puithove: What about the quality difference, could see any noticable difference? with w3fdiff?

I don't really:
https://dl.dropboxusercontent.com/u/5572...-yadif.mp4
https://dl.dropboxusercontent.com/u/5572...w3fdif.mp4

https://dl.dropboxusercontent.com/u/5572...-yadif.mp4
https://dl.dropboxusercontent.com/u/5572...w3fdif.mp4

Interesting. The YADIF de-interlaced text looks a little bit worse to me than the W3FDIF stuff on the burosch1 sequence - particularly on the horizontal tops of the B and U capitals, and the www. Also the word "Frame" looks cleaner in the W3FDIF. Small differences - probably more visible when you freeze the image sequence.
find quote
-DDD- Offline
Fan
Posts: 391
Joined: Sep 2013
Reputation: 8
Post: #645
[PATCH][libva-intel-driver] VPP: MADI on SNB

http://permalink.gmane.org/gmane.comp.fr...libva/2097

| myHTPC |
(This post was last modified: 2014-05-27 12:38 by -DDD-.)
find quote
Post Reply