• 1
  • 17
  • 18
  • 19(current)
  • 20
  • 21
  • 22
No multithreaded Video decoding on XBMC INTREPID version (in-depth testing)
Just checked out svn and applied that multithread workaround patch. Will see if that fixes the multithreading for me on Intrepid ...

boba
Reply
The workaround is confirmed non-working in some cases. We're really back at square one on this issue.
Reply
headache Wrote:FYI, at this moment I'm running hardy with linux kernel 2.6.29. This kernel is build using make-kpkg and the default .config from the original 2.6.24-23. I've no problems with off-balanced cpu usage playing HD (1080p) material. Using xbmc svn r19517 and mediastream r1470.

Could you build another 2.6.29 with the default config for intrepid and report back? I wiped my hardy partition to install slackware so I could downgrade all of those stupid libraries. Sad
Reply
althekiller Wrote:The workaround is confirmed non-working in some cases. We're really back at square one on this issue.

Yea confirmed, the patch didn't really make a difference for me as well.

boba
Reply
althekiller Wrote:Could you build another 2.6.29 with the default config for intrepid and report back? I wiped my hardy partition to install slackware so I could downgrade all of those stupid libraries. Sad

Or someone could diff the (resulting) configs?
Reply
sandos Wrote:Or someone could diff the (resulting) configs?

It's 3700 lines, I haven't got around to it yet Smile
Reply
althekiller Wrote:Could you build another 2.6.29 with the default config for intrepid and report back? I wiped my hardy partition to install slackware so I could downgrade all of those stupid libraries. Sad

I've build a 2.6.29.1 with the config of 2.6.27-11-generic (intrepid). Also with this kernel there is no off-balanced cpu usage playing HD (1080p) material.

In the weekend i'll try to use the 2.6.27-11-generic intrepid kernel on hardy. For this to work I also need to build a correct nvidia kernel-module on an intrepid installation.
Reply
headache Wrote:I've build a 2.6.29.1 with the config of 2.6.27-11-generic (intrepid). Also with this kernel there is no off-balanced cpu usage playing HD (1080p) material.
Well, shit! Sad
Quote:In the weekend i'll try to use the 2.6.27-11-generic intrepid kernel on hardy. For this to work I also need to build a correct nvidia kernel-module on an intrepid installation.

Just grab the .run for w/e driver version you want from nvidia, it's dead simple.
Reply
boba23 Wrote:Yea confirmed, the patch didn't really make a difference for me as well.

boba
Use the full patch: http://xbmc.svn.sourceforge.net/viewvc/x...5&r2=19485
Reply
Anyone bothered to try with Jaunty as it was released today? I plan on giving it a go later on...

Side note: whats the best or at least a really good sample to tax the CPU usage (or really just a generally good 1080p test sample)

I've seen the Killa sample talked about a lot and I've seen a lot of people mention planet earth samples...
Reply
ERamseth Wrote:Anyone bothered to try with Jaunty as it was released today? I plan on giving it a go later on...

Side note: whats the best or at least a really good sample to tax the CPU usage (or really just a generally good 1080p test sample)

I've seen the Killa sample talked about a lot and I've seen a lot of people mention planet earth samples...

The alphas and betas still exhibit the problems so I have no hopes for the final release being any different.

"killa sampla" is a clip of the "bird scene" from "planet earth" with retarded encoding settings you'll likely never see in real life, but should be able to play it, it's unlikely you'll run into something you can't play.

I think I hit all keywords there that people use to describe the same clip. This clip is what exhibits the behavior best on my system, but it seems anything that your system can *just* play in hardy should do the trick.
Reply
First time I tested XBMC like 6 months ago in my intrepid system, multithreading was not working, XBMC cpu usage with software was like 110-120% (divided 100% 10-20%).

I just tested play killa sampla in software (advanced shaders) and
surprisingly cpu usage was max 160% (divided 90% 70%) with 50 frames dropped. Pretty good for 2.5GHz?
With plain software cpu usage was 140% (divided 70% 70%), 140 dropped frames.

So multithreading has got definitely got better. How much cpu usage should be when multithreading works properly?

My system

Software:

Ubuntu 8.10
Linux bogey 2.6.27-11-generic #1 SMP Wed Apr 1 20:53:41 UTC 2009 x86_64 GNU/Linux
libc6 2.8~20080505-0ubuntu9
XBMC r19063 (smoothvideo), no patches.

HW:
MB GA-E7AUM-DS2H, with IGP 9400 512MB shared
CPU E5200 @ 2.5GHz
Reply
please correct me if i'm wrong... i haven't been following this thread that closely.

with that (killa/planet earth) sample you're always going to max out your processor.

i believe the issue is that instead of evenly spreading the decoding process across the 2 cores, it is spiking one core while the other is relatively low and then it flips after a few seconds. if frames are dropped because one core is maxxed out while the other has cycles to spare, it's a problem.
Reply
bobb0 Wrote:please correct me if i'm wrong... i haven't been following this thread that closely.

with that (killa/planet earth) sample you're always going to max out your processor.

i believe the issue is that instead of evenly spreading the decoding process across the 2 cores, it is spiking one core while the other is relatively low and then it flips after a few seconds. if frames are dropped because one core is maxxed out while the other has cycles to spare, it's a problem.

that is correct, problem is that on ubuntu Hardy (8.04) same "killa sample" plays without the problem because load is distributed evenly between two cores
if you need to know more then read the thread Big Grin
Reply
First of all let me clear one thing up... multithreading is working fine. The threads are spawned as normal and function fine.

The PROBLEM is that both threads seem to get scheduled to the same core on intrepid. You can see this clearly in top with thread view enabled (hit 'H'). The two decode threads spawn and each is taking ~50% core load, but if you look at core usage one core is very high and the other very low.

This seems to be alleviated, but not resolved with jaunty. I'm seeing core usage splits of ~88/55 at worst on killa, where I would regularly see 98/7 on intrepid and 75/75 on hardy. There are still quite a few dropped frames ~200, but investigation hasn't gone too deep yet.

EDIT: Clarification: Multithreading was ALWAYS working, the topic of this thread is just misleading.
Reply
  • 1
  • 17
  • 18
  • 19(current)
  • 20
  • 21
  • 22

Logout Mark Read Team Forum Stats Members Help
No multithreaded Video decoding on XBMC INTREPID version (in-depth testing)1