[Waltman]

Jingai, i tested the second build, first one is was not available

Mac Mini 2011 intel HD3000
MACOS Sierra 01.12
kodi-20160930-27c4bc6-jingai-osx_vtb_timing_memphiz_422YpCbCr8_iosurface-x86_64
VTB = ON
Log at : http://pastebin.com/s8XVYxCD

conclusion
result: smooth playback
cpu at 20 - 30%
This test build solves the VTB problems.

[FernetMenta]

Can anybody make a short video of the screen when the issue shows? I would like to see what you see. Please do this with current nightly and turn component logging for video on, ffmpeg off. Please post debug log with video.

[jingai]

Can anybody make a short video of the screen when the issue shows? I would like to see what you see. Please do this with current nightly and turn component logging for video on, ffmpeg off. Please post debug log with video.

Video showing the problem.

Debug log with component logging for video on, everything else off.

[FernetMenta]

Thanks. dropbox seems incredible slow today. Still trying to download the video, 3rd attempt now...
In the meantime, could you please try this change? https://github.com/FernetMenta/xbmc/tree/vtb

[jingai]

In the meantime, could you please try this change? https://github.com/FernetMenta/xbmc/tree/vtb

Built it, but will have to test when I get home. I'll get back with you this evening.

[FernetMenta]

Got the video now. Looks pretty much like buffer reuse/overwrite before the GPU actually processed the buffer. I think the 422 format only worked by chance.

[FernetMenta]

I updated the my branch. Please make sure to get latest changes.

[Waltman]

kodi-20161002-25601d3-master-x86_64
video showing the problem
kodi log with component loggin for video on

[FernetMenta]

kodi-20161002-25601d3-master-x86_64
video showing the problem
kodi log with component loggin for video on

This is no performance issue, measured fps does not drop. You get this effect as shown in the video if a buffer is reused for decoding before the last content of the buffer has been processed by the GPU and rendered to screen. The new frame is in the future. This makes the video look jumping back and forth.
In my linked branch I implemented the same concept (fence) as I did for vdpau and vaapi.

[Waltman]

kodi-20161002-25601d3-master-x86_64
video showing the problem
kodi log with component loggin for video on

This is no performance issue, measured fps does not drop. You get this effect as shown in the video if a buffer is reused for decoding before the last content of the buffer has been processed by the GPU and rendered to screen. The new frame is in the future. This makes the video look jumping back and forth.
In my linked branch I implemented the same concept (fence) as I did for vdpau and vaapi.

Is there a build i should test with the changes you made ?

[jingai]

I will post a build with FernetMenta's patches tonight.

[jingai]

(removed test builds -- see here)

[jingai]

@FernetMenta,

Unfortunately your patches (commits 6d5af7d and 00682bc) did not solve the problem. The symptoms are the same as before (buffer reuse, videos appears to be jumping back and forth).

My debug log is here.

[jingai]

Ah, there was a typo on line 236 of RendererVTBGL.cpp. If I dump that return, it does indeed work fine

Debug log in case you wanted to see it.

For those that want to test it, here is a build with FernetMenta's patches applied.

[FernetMenta]

Ah, there was a typo on line 236 of RendererVTBGL.cpp. If I dump that return, it does indeed work fine

Debug log in case you wanted to see it.

For those that want to test it, here is a build with FernetMenta's patches applied.

Thank you very much. Great that you noticed the error in the code. Was a left over from my testing.