[fritsch]

Old ... won't help us on HSW and certainly not without a proper VPP implementation.

[Schasan]

I am trying the VPP branch.

Symptom: I just see BOB as available Deinterlacing method.

The first thing I discovered was that the deint methods from the XBMC logs showed confusing integers.
The mismatch is in /usr/include/va/va_vpp.h against ~/xbmc/xbmc/cores/dvdplayer/DVDCodecs/Video/VAAPI.h, with swapped locations of Weave and Bob. I am using stock Trusty in the moment at libva 0.35 (Intel i965 driver - 1.3.0).

Now I am getting familiar numbers matching my own capabilities queries done earlier this day.
Still I am not seeing the two methods in the GUI, could you point me to a location, how the Deint methods are mapped to the user frontend?

Code:

17:42:37 T:139979493078784   DEBUG: VAAPI_VPP - Successfully initialized deinterlacer 1. Need 1 forward and 0 backward refs.
17:42:37 T:139979493078784   DEBUG: VAAPI_VPP - Method 1 is supported!
...
17:42:37 T:139979493078784   DEBUG: VAAPI_VPP - Successfully initialized deinterlacer 3. Need 1 forward and 0 backward refs.
17:42:37 T:139979493078784   DEBUG: VAAPI_VPP - Method 3 is supported!

BTW: Why would a BOB deinterlacer - method 1 above - report to need 1 forward ref, it does every frame on its own.

[fritsch]

Post a complete log. MADI, MACI are disabled by default and are only to be enabled via advancedsettings.xml

Code:

<advancedsettings>
<video>
<vaapiadvanceddeint>true</vaapiadvanceddeint>
</video>
</advancedsettings>

VPP was (as in that code you use) just a test. A test that failed, cause performance sucked heavily.

[Schasan]

I thought VPP just has issues with Haswell.
But anyway, I am heavily in love with VPP so I would like to see it on my own screen.

So I collected the log, to me it still gives no clue.

(OOPS, I dont find the attach button)

[fritsch]

Link to a pastebin - you can easily try VPP from an OpenELEC image, that I posted in another thread.
One cannot be in love with something that never really worked (despite BOB).

[Schasan]

Here we are: http://pastebin.com/y8UaELs5

---

Link to a pastebin - you can easily try VPP from an OpenELEC image, that I posted in another thread.
One cannot be in love with something that never really worked (despite BOB).

I was motivated by your post http://forum.xbmc.org/showthread.php?tid...pid1713568
Quote: "flawless"

[fritsch]

Ah yes :-) IVB is working quite nice, but only if you have a HD4000. Stay tuned on that topic.

[Schasan]

Ah yes :-) IVB is working quite nice, but only if you have a HD4000. Stay tuned on that topic.

Thats exactly what I have, would not have started testing it otherwise.
Any hints for my system to get into debugging?
Why would, if capbilities are found according to the log, nothing show up in the GUI?
I cannot locate the position in XBMC code, where the capabilities are mapped into display text.

[fritsch]

the capabilities are enumerated in VAAPI_VPP.cpp and called via LinuxRenderGL - see here: https://github.com/BtbN/xbmc/commit/2123...5c53fR3542

[Schasan]

I do have it up and running now. I cannot fully reproduce my problems, but it may have been a mixture of:

• The enumeration mismatch mentioned earlier in this thread originating from different va_vpp.h and
  
• not understanding the mechanics of using advancedsettings.xml properly.
  

Thank you very much.

[fritsch]

What about the MADI quality in comparison to Yadif? Do you like it?

(I ask that everybody).

Here are some nice samples to compare those: http://samples.ffmpeg.org/MPEG2/interlaced/burosch1.mpg http://samples.ffmpeg.org/MPEG2/interlaced/burosch2.mpg

Don't forget to enable them properly.

[Schasan]

A short feedback:

TV looks ok, trying interlaced DVB-S transport streams in SD and HD. But difficult to judge by nature.
The recommended Burosch samples currently dump, I shall be trying something fresher than stock Trusty bundled VAAPI throughout the next weekend.

Code:

tv-i3:~$ xbmc
Running DIL (3.22.0) Version
DtsDeviceOpen: Opening HW in mode 0
DtsDeviceOpen: Create File Failed
libva info: VA-API version 0.35.0
libva info: va_getDriverName() returns 0
libva info: Trying to open /usr/lib/x86_64-linux-gnu/dri/i965_drv_video.so
libva info: Found init function __vaDriverInit_0_35
libva info: va_openDriver() returns 0
libva info: VA-API version 0.35.0
libva info: va_getDriverName() returns 0
libva info: Trying to open /usr/lib/x86_64-linux-gnu/dri/i965_drv_video.so
libva info: Found init function __vaDriverInit_0_35
libva info: va_openDriver() returns 0
xbmc.bin: i965_post_processing.c:4209: gen7_pp_nv12_dndi_initialize: Assertion `previous_in_obj_surface && previous_in_obj_surface->bo' failed.
Aborted (core dumped)

[fritsch]

Stock trusty works fine for me. Please provide more info about what you are doing. It seems you run the preliminary VPP builds. Please open a separate issue with those, as this thread is (not yet) about VPP.

Edit: It won't help to provide a second thread as this VPP code is not maintained, btbn gave up on it - so time to change onto the yadif solution proposed in this thread.

[FernetMenta]

I am working on a version which can do vpp and ffmpeg filters. And it will allow switching between those methods on-the-fly.

[wsnipex]

I am working on a version which can do vpp and ffmpeg filters. And it will allow switching between those methods on-the-fly.