[trsqr]

Some gpus (as nvidia GTX 960 2nd gen) have some internal circuits to "alter the 10 bit to 8 bit" and then decode it as 8 bit. So we need to be careful in the future if someone really tells: 10 bit decoding :-)

Or would it actually decode the full 10 bits and then scale it down from 10 to 8 bits? What's the situation on the display side when it comes to 10-bit, is it already commonly supported? And if the display does not support it, then this kind of downscaling needs to happen also in the fully 10-bit capable GPU.

I wonder what's next after 10-bit 4k@60fps.. they'll improve something again so people keep buying new TVs.

Anyhow, there's no HW decoding of 10-bit content scaled down to 8-bit or not for Intel HW and to be honest, I don't see much need this year, probably not next year either. 2017 starts to look like a good time to buy 4k TV and a new HTPC.

[AndyFurniss]

Well 10 bit is yuv level and as AIUI this will still give a quality gain when displayed as 8bit RGB (going from full range 8 bit RGB -> studio 8 bit yuv then back eats millions of colours - I think you end up with 2.3m). So even without high bit displays you get a gain today if your display can really do 8bit RGB.

On higher display depths - it seems quite common for TVs to advertise deep colour - my 5 year old mid range Panasonic plasma does. It's even possible at hdmi level to drive it as such today with open source radeon driver and the right card. You would need to patch up xserver and use something capable to really drive it with full bits. I tested the patches when they went in (at hdmi level). The dev that did them did it for *scientist users" with oled displays that used s/w that he develops/supports (psychtoolbox-3).

As for the future - well I want HDR and 120fps already!

Edit: the app was called psychtoolbox-3

[john.cord]

Any Openelec Build with Kodi 15.1 out now?

6.0 final will be 15.1 afaik but no releasdate atm...

[fritsch]

And it will have nothing to do with this thread here ... as it ships the old GLX code.

[lmyllari]

Hi fritsch,

do you have any plans to update the egl-isengard branch? Or are you waiting for the OE release and then adding to that?

I'm thinking of making a more production ready 3dlut test build before moving the development on top of FernetMenta's master branch.

[fritsch]

I just rebased my egl-isengard to xbmc-upstream Isengard branch

For OE - I could also rebase that one - let's see.

Edit: both done

[trsqr]

I did a further testing with the OpenELEC build in post #2 and Braswell N3050 hardware with 2 gigs of RAM. I was able to play 20+ videos and when starting a new one, kodi crashed with following in dmesg:

[ 381.202509] Out of memory: Kill process 663 (kodi.bin) score 282 or sacrifice child
[ 381.202513] Killed process 663 (kodi.bin) total-vm:2558768kB, anon-rss:532540kB, file-rss:0kB

Logs that I collected are stored here: http://trsqr.net/kodi_crash/

Not sure if this helps, but at least now I have something more to indicate that it indeed is memory related. Though last time when I tried this, it kind of froze without kodi restarting, but today I've done this now 3 times and have reached the same result: kodi crash and restart.

It seems everytime I skip inside the video, the amount of free memory goes down. This can be seen easily if you enable debug logging, as the amount of free memory is all the time on the screen. Then skip 10 seconds forward repeatedly and watch the figure.

For comparison, if VAAPI is disabled, the memory consumption figure does not change significantly when skipping inside the video.

I'll do the valgrind testing in Ubuntu next.

[lmyllari]

I just rebased my egl-isengard to xbmc-upstream Isengard branch

For OE - I could also rebase that one - let's see.

Edit: both done

Thank you!

Did you push the OE changes?

[fritsch]

So basically you say: Skipping produces a memory leak?

[fritsch]

I just rebased my egl-isengard to xbmc-upstream Isengard branch

For OE - I could also rebase that one - let's see.

Edit: both done

Thank you!

Did you push the OE changes?

I rebased on OE-master, adjusted the patchset (dropped two of them) and force-pushed (EGL branch)

[fritsch]

I did a further testing with the OpenELEC build in post #2 and Braswell N3050 hardware with 2 gigs of RAM. I was able to play 20+ videos and when starting a new one, kodi crashed with following in dmesg:

[ 381.202509] Out of memory: Kill process 663 (kodi.bin) score 282 or sacrifice child
[ 381.202513] Killed process 663 (kodi.bin) total-vm:2558768kB, anon-rss:532540kB, file-rss:0kB

Logs that I collected are stored here: http://trsqr.net/kodi_crash/

Not sure if this helps, but at least now I have something more to indicate that it indeed is memory related. Though last time when I tried this, it kind of froze without kodi restarting, but today I've done this now 3 times and have reached the same result: kodi crash and restart.

It seems everytime I skip inside the video, the amount of free memory goes down. This can be seen easily if you enable debug logging, as the amount of free memory is all the time on the screen. Then skip 10 seconds forward repeatedly and watch the figure.

For comparison, if VAAPI is disabled, the memory consumption figure does not change significantly when skipping inside the video.

I'll do the valgrind testing in Ubuntu next.

I can verify this, but I don't know yet - if that's a leak or normal buffer management. Whenever I seek Memory usage increases, it always goes up and never down - top shows that easily.

[M4tt0]

I just rebased my egl-isengard to xbmc-upstream Isengard branch

For OE - I could also rebase that one - let's see.

Edit: both done

@fritsch
I think the dropbox OE links in post #2 still point at the previous July 25 packages. If so, and it is not just the filename that is outdated, could you please update those too?

Actually, have been comparing your EGL OE build to the new 15.1 DX11 build on my i5 HSW NUC, and with my setup and settings the EGL build looks better with SD Live-TV content. Not much difference w.r.t. HD content in my eyes. But will play around with this some more...

[fritsch]

I don't intend to build a new one currently, though you are free to do so :-)

[Roby77]

so it means that in the future will be in official oe ?

[fritsch]

That's the plan ... read the backlog please - this has been answered a lot of times.