That's okay with the xserver. But still wrong driver.
apt-cache policy i965-va-driver | pastebinit please - or shortly step by in #xbmc-xvba
Here it is :
http://paste.ubuntu.com/8131313/
I think I see my problem. I hadn't upgraded i965-va-driver. I've just done an apt-get install and both new VAAPI modes are now working properly. Motion Compensated sitting at around 25% on both cores of the 2955U with a 25Mbs 1080/50i H264 file. Looks good and lower CPU than YADIF 2x in software?
After this I have :
http://paste.ubuntu.com/8131344/
So I obviously hadn't upgraded my i965-va-driver. My mistake. Thanks for the handholding.
I now get this with vainfo
http://paste.ubuntu.com/8131352/ The <unknown profile> is new though?
yes - it's done on the GPU :-)
If you try with enabled "Prefer VAAPI Output" - you can see what happens if you don't copy any memory back to system memory and measure that load.
Remember: It will crash every some minutes - cause of a bug in the prerelease driver I have mentioned earlier.
This is brilliant work fritsch and fernetmenta. We owe you guys a lot for working so hard on VAAPI and de-interlacing.
If Intel iron out the bugs (and because your constructive persistence they appear to be?) this makes things like the Chromebox even better! (The CPU load was a bit concerning with YADIF if you had other stuff in the background) and presumably this also means that the N2820 and J1800/1900 stuff now get decent de-interlacing (as they weren't able to use YADIF 2x ? Though they still have the low number of EUs which impacts scaling?)
The Baytrail is some oldish IVB based GPU with less Execution Units than the old Celeron 847. In theory it could work - but currently SNB/IVB drivers are broken and it's not really working. Additionally those IVB/SNB don't support MCDI only MADI.
(2014-08-24, 19:34)fritsch Wrote: [ -> ]The Baytrail is some oldish IVB based GPU with less Execution Units than the old Celeron 847. In theory it could work - but currently SNB/IVB drivers are broken and it's not really working. Additionally those IVB/SNB don't support MCDI only MADI.
Thanks for the clarification fritsch. This does make the Haswell Chromebox look like a very neat solution - particularly once this stuff gets rolled into OpenElec when (hopefully not if) Intel fix the remaining issues. It's a pity Intel don't make a 2955U based NUC.
The xbmc code already is in OpenELEC - but it won't help / work until the driver gets released.
I bumped the libva-driver-intel to 1.3.3~pre4-1~trusty and wsnipex is currently building it.
This version has the MCDI / MADI segfault fixed. Thx to Gwenole of intel.
He will port those fixes to IVB / Baytrail for the final release. So give this driver a good test.
Edit: Driver is build and in the ppa - thx again to wsnipex
:-)
Err. It's quite broken and needs fixing in the driver as mentioned above.
yea, the "He will port those fixes to IVB / Baytrail for the final release" should have given me a hint, lol
Just wanted to thank to all you guys involved. It is a great step forward !
Working great with Intel HSW on a Macbook Pro. Quality is great. With the new Intel driver it is really stable.
Cheers!
Chaps, I would like to test the new Intel driver too. Running OpenELEC, I'm not sure how to read fritsch's comment on the Intel driver release though (#877). Do I have to wait for a stable release to test within OE or can I test with the pre-4-1 already? If the latter, please advise how I can do that...
Thanks and best,
M4tt0