• 1
  • 13
  • 14
  • 15(current)
  • 16
  • 17
  • 22
No multithreaded Video decoding on XBMC INTREPID version (in-depth testing)
I certainly don't get better performance with Speedstep, but that might be because i only have a E7300, (clocks to default clock speed when i enable SpeedStep).

And i can't seem to get rid of the dropping of 15 frames i tried clocking it to 3.1GHz(from 3.06) and that didn't change anything so i guess it's a network issue right?
Reply
Hi,

I wanted to report my findings in this problem. I'm using ffplay compiled within the xbmc svn (no patches) so I guess I'm using the *same* ffmpeg that xbmc uses. I'm using ffplay as reference as I can not run xbmc in all my tests (when chrooted for instance, read below). My hardware is a mobo asus p5n7a-vm and e8400 at 3GHz (stock speed).

Hardy partition:
- Both xbmc and ffplay plays smooth killasample

Intrepid partition:
- Xbmc and stand alone ffplay drops frames
- running ffplay in the compile chroot (running it from intrepid) seems to *drop* frames

So, first of all we must take into account that I have no way (that I know of) to know the objective number of dropped frames in ffplay so it is more a subjective impression. To be exact, I can not use all hardy libs in the intrepid test, ffplay still links with intrepid libraries: libbz2.so.1.0 libpthread.so.0 libm.so.6 libdl.so.2 so it may well be one of them.

I'm going to try to reproduce the tests without gnome or any gnome deamons but I guess I need a way to know the dropped frames in standalone ffplay, anybody knows? How is xbmc doing it?

later and sorry for the long and not really so helping post :-(

EDIT: I have manage to run ffplay in intrepid using all hardy libraries but libc (obviously) and it *seems* to be better but as said I do need a way to report dropped frames in ffplay to be sure though!!!
Reply
Dropped frames is a piss poor metric, you need to watch the core utilization and ensure that it remains balanced through the entire sample.
Reply
althekiller Wrote:Dropped frames is a piss poor metric, you need to watch the core utilization and ensure that it remains balanced through the entire sample.

I know but let's face it, unfortunately dropped frames *is* the most objective measure we've got right know. I play killa sample on hardy -> no dropped frames, however, it drops like mad in intrepid. It is something we can easily ask other people to check rather than "check if core utilization remains balanced" or to see if the killa sample plays smooth...

Regarding the issue, I've changed the way I'm replacing libraries. Right now I'm copying hardy binary libraries to intrepid and link xbmc in intrepid with them. So the xbmc binary compiled normally in intrepid, now links with the hardy libraries. I've managed to change all libraries ffmpeg uses and it still the same lack of performance. The only strange message when I replace the libraries is something about not being able to load libasound_module_pcm_pulse.so. That makes me thing that it can not properly replace libasound and it may be a problem with alsa :confused2:

Has anyone install the latest alsa in their intrepid box? (I'm using intrepid official) Does it performs as bad as the normal intrepid version?
Reply
i can confirm that there still is a performance drop with upto alsa-1.0.19 and daily snapshots - and the most recent alsa-libs. Also i notice that lib is to do with pulse audio. i do not have pulse installed - no gdm/gnome either, xbmc compiled with --disable-pulse option, performance is still bad.
Reply
althekiller Wrote:Dropped frames is a piss poor metric, you need to watch the core utilization and ensure that it remains balanced through the entire sample.

My oppinion is that it the most important is to have the sum of the two cores utilization somewhere the 170-190 % zone when having the killa played with a 3ghz processor, as i relized for example with Jaunty that the core utilization looks pretty even, but only 50 -60 % maxed / core. The reason for this is that the two thread of video decoding is working sequential.This way we have the same bad performance as with Intrepid's 90-20 %. (At least with 1080p content)

I am really stuck as i changed the libraries, binaries and still the performance remains poor. Do we have to stuck with Hardy forever...:confused2:
Always read the XBMC online-manual, FAQ and search the forum before posting.
Do not e-mail XBMC-Team members directly asking for support. Read/follow the forum rules.
For troubleshooting and bug reporting please make sure you read this first.
Reply
alanwww1 Wrote:I am really stuck as i changed the libraries, binaries and still the performance remains poor. Do we have to stuck with Hardy forever...:confused2:

Same here alan, I've tried all hardy libraries, no gnome but I still see the performance drop in intrepid. I haven't been able to make xbmc play killa sample smooth. Sad

I seem to remember that aron was able at some point to play it smooth maybe he can re do the test and we can also try it... :confused2:

later,
Reply
pilluli Wrote:I seem to remember that aron was able at some point to play it smooth maybe he can re do the test and we can also try it... :confused2:
later,

I tried to ask Aron about this question, i even PMailed him, but no answer. Maybe he's gone skiing Smile or working on some other things.

Anyway he did it with a different way he made a debootstrab with Hardy (or Intrepid). I am afraid that in real life (full Intrepid system) it does not work for him eighter.

I still ask anyone who has more linux knowledge: what is librt.so because this is not loading in Hardy but is loading in Intrepid when i launch ffplay. I suspect some timing or syncing problem between libavc's two decoding threads.

I think only Devs can help who know the working nature of the two threads of the CABAC patch XBMC has.
Always read the XBMC online-manual, FAQ and search the forum before posting.
Do not e-mail XBMC-Team members directly asking for support. Read/follow the forum rules.
For troubleshooting and bug reporting please make sure you read this first.
Reply
Hi alan,

alanwww1 Wrote:Anyway he did it with a different way he made a debootstrab with Hardy (or Intrepid). I am afraid that in real life (full Intrepid system) it does not work for him eighter.

I also tried with a debootstrab hardy with no luck. I didn't manage to run xbmc or ffplay within the chroot (to try *all* hardy libraries but the intrepid kernel) but running ffplay compiled in the chroot from intrepid didn't help.

Quote:I still ask anyone who has more linux knowledge: what is librt.so because this is not loading in Hardy but is loading in Intrepid when i launch ffplay. I suspect some timing or syncing problem between libavc's two decoding threads.

As far as I know librt.so is part of the libc6 package, so part of the standard C library. I guess they split libc6 in newer versions. :confused2:

Quote:I think only Devs can help who know the working nature of the two threads of the CABAC patch XBMC has.

Don't really know, but the *same* code should be able to run fine in the *same* hardware so something else (rather than the CABAC patch) should be blamed for (say kernel, libc6 or whatever)...
Reply
pilluli Wrote:Don't really know, but the *same* code should be able to run fine in the *same* hardware so something else (rather than the CABAC patch) should be blamed for (say kernel, libc6 or whatever)...

You are right, i think the changes in libc makes the patch unworkable somehow. My knowledge is nothing to debuging this further. I think it could be a way to analize the patch and the changes in libc and how they work together, but not for me.

I tried to use Hardy's libc on Intrepid running ffplay but i got a nice segmentation fault. Is it possoble to make it without a segmentation fault ?
Always read the XBMC online-manual, FAQ and search the forum before posting.
Do not e-mail XBMC-Team members directly asking for support. Read/follow the forum rules.
For troubleshooting and bug reporting please make sure you read this first.
Reply
Is the CPU frequency scaling working for you guys?

I've found that I can only get acceptable XBMC performance in Hardy by setting my frequency governor to 'performance' using cpufreq-utils (3GHZ Dual-core AM2), however I wasn't able to do this on my fully patched up Intrepid installation (same hardware) because the kernel isn't properly detecting my AMD AM2's Powernow features.

It's a long shot, but I wondered if it could be related to this thread (I'm suffering with the same problem. A patched SVN xbmc plays perfectly on Intrepid, but as others have noted is unstable, so not exactly high on the WAF.)
Reply
I really gave up for now. I tested every possible scenarios with differnt libs, compiles.

We need to have someone who knows the working of the cabac patch for xbmc to check why it is not working.

I think you should try Ubuntu Jaunty as it could have better support for your hardware, but for now it is not stable yet.
Always read the XBMC online-manual, FAQ and search the forum before posting.
Do not e-mail XBMC-Team members directly asking for support. Read/follow the forum rules.
For troubleshooting and bug reporting please make sure you read this first.
Reply
I followed the guide here:

http://wiki.xbmc.org/?title=HOW-TO:_Inst...ep-by-step

And it fixed the problem I was having with playing back 1080p h.264 videos where the distribution of the load was uneven. Now I'm able to play it back properly.

It even fixed the problem where I couldn't play back music if I had navigation sounds, but now I can play back music and the navigation sounds work too. Perhaps it has to do with updating alsa from 1.0.17 to 1.0.19.

But I am still encountering the issue where when I set ac3 and dts to output compressed over spdif, nothing is sent over. Only if I untick those to send it over uncompressed does it work.
Reply
pyrates Wrote:I followed the guide here:

http://wiki.xbmc.org/?title=HOW-TO:_Inst...ep-by-step

And it fixed the problem I was having with playing back 1080p h.264 videos where the distribution of the load was uneven. Now I'm able to play it back properly.

Yes because in the guide there is Aron's patch for using ffmpeg-mt branch instead of xbmc's cabac entrophy decoding patch, but that version is unstable with ac3 encoded SD material when skippikng of fast forwarding. (At least for me)

We are loooking for a solution here for the original cabac patched xbmc to work with new kernels.
Always read the XBMC online-manual, FAQ and search the forum before posting.
Do not e-mail XBMC-Team members directly asking for support. Read/follow the forum rules.
For troubleshooting and bug reporting please make sure you read this first.
Reply
pyrates Wrote:I followed the guide here:

http://wiki.xbmc.org/?title=HOW-TO:_Inst...ep-by-step

And it fixed the problem I was having with playing back 1080p h.264 videos where the distribution of the load was uneven. Now I'm able to play it back properly.

It even fixed the problem where I couldn't play back music if I had navigation sounds, but now I can play back music and the navigation sounds work too. Perhaps it has to do with updating alsa from 1.0.17 to 1.0.19.

But I am still encountering the issue where when I set ac3 and dts to output compressed over spdif, nothing is sent over. Only if I untick those to send it over uncompressed does it work.

Fixed the sound problem by changing IEC958 to default.
Reply
  • 1
  • 13
  • 14
  • 15(current)
  • 16
  • 17
  • 22

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