VideoReferenceClock: refactor, fixes, video sync for RPi

[FernetMenta]

The old VideoReferenceClock has become a big mess and combined all platform specific code into a single file. So any fix or addition required a refactoring first. I stated with it because I wanted to get rid of the brittle GLX implementation for Intel systems. The invisible window with its GLX context often caused trouble.

In parallel @popcornmix worked on a clock for the Pi which is now included here.

credits to @Memphiz who did the work for OSX/iOS. In order not to break future bisects I squashed your changes into the refactor commit.

[FernetMenta]

jenkins build this please

[FernetMenta]

jenkins build this please

[wsnipex]

the linux build error is caused by missing libdrm in unified deps.
you'll have to add /usr/lib/x86_64-linux-gnu/pkgconfig/libdrm.pc to https://github.com/xbmc/xbmc/blob/master/tools/depends/target/Makefile#L128

We might also need to install libdrm-dev on the jenkins slave, didn't check that yet.

[FernetMenta]

jenkins build this please

[MartijnKaijser]

@topfs2

[fritsch]

While this is running perfectly fine on my hardware (intel), I think this could also work nicely on AMD OSS hardware - can anyone test this? (I need to reinstall my amd boxes before getting that running).

Edit: This line needs adjusting: FernetMenta@9012809#diff-cd6e9ebd37bcd0088ec3f06fc3054fdeR100 to test it on AMD

[wsnipex]

from a quick test on my kabini this seems to be working fine on AMD as well

[FernetMenta]

Thanks for testing but I won't enable it for AMD in this PR. It needs much more testing on various hw. @fritsch if you feel comfortable with this change for AMD had have more data from testing, feel free to submit this as a separate PR later.

[FernetMenta]

jenkins build this please

[wsnipex]

I've watched >1h of IPTV news streams and didn't notice any issues, even when using 3:2 pulldown(50Hz stream on 60Hz in windowed mode).

@FernetMenta how about enabling this in your master for some enduser testing?

[FernetMenta]

@wsnipex good idea. will do

[fritsch]

@wsnipex @FernetMenta This should enable it on Intel and AMD: fritsch@65f08d5

[FernetMenta]

@fritsch fixed a compiler warning, thx
jenkins build this please

[MartijnKaijser]

it was approved and jenkins builds ok

[Voyager1]

something's wrong with the refactor for Windows (D3D). Sync playback to display at 23.976fps is extremely stuttering (getting around 14-15fps).
I noticed that when pressing "o" the refresh is still listed as 50fps (although the monitor has switched to 24fps). That could cause the issue.
I reverted this whole PR and the problem is solved.
When pressing "o", the refresh rate is now correctly listed as 24fps.
I'm going to take a closer look at the potential cause, but I guess it's a refactoring mistake in one of the VideoSync classes.

Note: my desktop res refresh is at 50fps - and auto refresh change is enabled. I have the impression that the new GetFPS logic doesn't detect the change - as was previously signaled by the graphicsContext.

[fritsch]

Does this help: #5679 ?

[Voyager1]

yes thanks that solved the issue.

[Memphiz]

@FernetMenta mhhh using beta2 on my atv2 now (which is a release build!) and when watching freetv (60hz refreshrate on tv - 25fps in the stream) i get frame drops maybe one a sec or so - not each sec but really constantly. The strange thing is - as soon as i enable debug logging it works flawless again (seeing the onscreen debug text and it lasts some time and then gets to a rock solid 25fps).

You have any idea how this could happen? (no debugger on atv2 - and without debug log its kinda hard to tell anything i guess?)

[FernetMenta]

you resample audio right? (otherwise this setting don't make any sense) what resample ration does the codec screen show? is it constant?

[fritsch]

Having him on IRC. He has Sync Playback to Display disabled - so most likely not this PR.