Bad X-performance on 2.6.6 & 2.6.7-rc1 on x86-64

Bad X-performance on 2.6.6 & 2.6.7-rc1 on x86-64

[Malte Schröder]

[-- Attachment #1: Type: text/plain, Size: 426 bytes --]

Hi,
I build a 64-bit kernel (using gcc 3.3.3) on debian/sid for an Athlon 64 3200+. The System has a Radeon 9800pro as graphics card. 
When playing videos using xine in full PAL-Resolution these videos run choppy, top then shows a cpuload of roughly 50% system and 50% user.

Thanks.
-- 
---------------------------------------
Malte Schröder
MalteSch@gmx.de
ICQ# 68121508
---------------------------------------


[-- Attachment #2: Type: application/pgp-signature, Size: 189 bytes --]

Re: Bad X-performance on 2.6.6 & 2.6.7-rc1 on x86-64

[Andi Kleen]

Malte Schröder <MalteSch@gmx.de> writes:

> Hi,
> I build a 64-bit kernel (using gcc 3.3.3) on debian/sid for an Athlon 64 3200+. The System has a Radeon 9800pro as graphics card. 
> When playing videos using xine in full PAL-Resolution these videos run choppy, top then shows a cpuload of roughly 50% system and 50% user.

I would suggest using oprofile to find out where the CPU time is going.

2.6.7rc1 had a fix to help performance with hardware accelerated
DVD playback on x86-64, but apparently that's not what you're doing.

-Andi

Re: Bad X-performance on 2.6.6 & 2.6.7-rc1 on x86-64

[Malte Schröder]

[-- Attachment #1: Type: text/plain, Size: 551 bytes --]

New information :)
I didn't profile it yet but I think I found what caused the problem.
It turned out that I have to disable alsa mmap-support in xine (mplayer worked w/o problems, it does not offer alsa mmap), so X is not involved at all. Do you still need a profile or is this a known thing?
Ah, before I forget, my userspace is 32bit-only, I have a 64-bit chroot I wanted to do some testing with.

Greets
-- 
---------------------------------------
Malte Schröder
MalteSch@gmx.de
ICQ# 68121508
---------------------------------------


[-- Attachment #2: Type: application/pgp-signature, Size: 189 bytes --]

Re: Bad X-performance on 2.6.6 & 2.6.7-rc1 on x86-64

[Andi Kleen]

On Tue, May 25, 2004 at 12:26:59PM +0200, Malte Schr?der wrote:
> New information :)
> I didn't profile it yet but I think I found what caused the problem.
> It turned out that I have to disable alsa mmap-support in xine (mplayer worked w/o problems, it does not offer alsa mmap), so X is not involved at all. Do you still need a profile or is this a known thing?

Ask the xine guys if it's known, I don't know much about xine.
If a user space change fixes it then I don't need any profiles.

-Andi

Re: Bad X-performance on 2.6.6 & 2.6.7-rc1 on x86-64

[Stan Bubrouski]

On Tue, 2004-05-25 at 08:36, Andi Kleen wrote:
> On Tue, May 25, 2004 at 12:26:59PM +0200, Malte Schr?der wrote:
> > New information :)
> > I didn't profile it yet but I think I found what caused the problem.
> > It turned out that I have to disable alsa mmap-support in xine (mplayer worked w/o problems, it does not offer alsa mmap), so X is not involved at all. Do you still need a profile or is this a known thing?
> 
> Ask the xine guys if it's known, I don't know much about xine.
> If a user space change fixes it then I don't need any profiles.
> 

I can confirm that xine with alsa-mmap option set does cuase strange
behaviour, though I notice it mostly as audio and video getting out of
sync when playing videos.  I've noticed this behaviour since I started
doing weekly xine CVS builds.  I've never bothered reporting it however,
I just turned off the option... which leads me to my question, what is
the advantage of using alsa-mmap in an app if it is used correctly?

And Malte, are you using emu10k1 driver per chance?

-sb

> -Andi

Re: Bad X-performance on 2.6.6 & 2.6.7-rc1 on x86-64

[Tomasz Torcz]

On Tue, May 25, 2004 at 05:20:22PM -0400, Stan Bubrouski wrote:
> I can confirm that xine with alsa-mmap option set does cuase strange
> behaviour, though I notice it mostly as audio and video getting out of
> sync when playing videos.  I've noticed this behaviour since I started
> doing weekly xine CVS builds.  I've never bothered reporting it however,
> I just turned off the option... which leads me to my question, what is
> the advantage of using alsa-mmap in an app if it is used correctly?

 Could for completness test mplayer with alsa-mmap, too? 
Its '-ao alsa:mmap' option.

-- 
Tomasz Torcz                        To co nierealne - tutaj jest normalne.
zdzichu@irc.-nie.spam-.pl          Ziomale na życie mają tu patenty specjalne.

Re: Bad X-performance on 2.6.6 & 2.6.7-rc1 on x86-64

[Malte Schröder]

[-- Attachment #1: Type: text/plain, Size: 273 bytes --]

>  Could for completness test mplayer with alsa-mmap, too? 
> Its '-ao alsa:mmap' option.
Works w/o problems, testet with 1.0pre4.

-- 
---------------------------------------
Malte Schröder
MalteSch@gmx.de
ICQ# 68121508
---------------------------------------


[-- Attachment #2: Type: application/pgp-signature, Size: 189 bytes --]

Re: Bad X-performance on 2.6.6 & 2.6.7-rc1 on x86-64

[Malte Schröder]

[-- Attachment #1: Type: text/plain, Size: 227 bytes --]

> And Malte, are you using emu10k1 driver per chance?
Yes, emu10k1/alsa on sb audigy.

-- 
---------------------------------------
Malte Schröder
MalteSch@gmx.de
ICQ# 68121508
---------------------------------------


[-- Attachment #2: Type: application/pgp-signature, Size: 189 bytes --]

Re: Bad X-performance on 2.6.6 & 2.6.7-rc1 on x86-64

[Stan Bubrouski]

On Wed, 2004-05-26 at 06:26, Malte Schröder wrote:
> > And Malte, are you using emu10k1 driver per chance?
> Yes, emu10k1/alsa on sb audigy.

I asked because I'm using the emu10k1 driver and I've noticed frequent
bizarre behviour when watching videos in xine AND mplayer...

While this could be written off as a xine bug, my friends who have other
audio cards can watch the same files without issue in both media
players.  I hadn't thought much of it, until now.

So here's the low-down... I'm thinking there are some bad bugs in the
emu10k1 driver (not surprisingly, it's given me problems for *4* years,
using OSS and ALSA, though I must say the OSS driver seems to be ALOT
better).

For completeness could you test the OSS emu10k driver (which supports
PCM, while the ALSA driver does not) and see if you experience better
overall performance?  Like less CPU utilization etc... I'm very
interested in finding out where the bottlenecks are.  The emu10k1 driver
isn't perfect and neither are xine or mplayer so I'd like figure out
what  exactly  is going on here.  I'll of course do the same.  It's just
kind of hard to judge for me seeing as my system is now 4 years old.

-sb

Re: Bad X-performance on 2.6.6 & 2.6.7-rc1 on x86-64

[Malte Schröder]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

| For completeness could you test the OSS emu10k driver (which supports
| PCM, while the ALSA driver does not) and see if you experience better
| overall performance?  Like less CPU utilization etc... I'm very
| interested in finding out where the bottlenecks are.  The emu10k1
| driver
| isn't perfect and neither are xine or mplayer so I'd like figure out
| what  exactly  is going on here.  I'll of course do the same.  It's just
| kind of hard to judge for me seeing as my system is now 4 years old.

I dont think it would make a difference sice cpu-time used by system is
really low since switching off mmap. But my overall experience with the
emu10k1 alsa-driver since it hit version 1.0 is quite good (before there
where some issues or should I say "annoyances" with the mixer). But if
you really want this I can rebuild my kernel and swith all that
oss-stuff back on ;)

Greets
- --
- ---------------------------------------
Malte Schröder
MalteSch@gmx.de
ICQ# 68121508
- ---------------------------------------

-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.2.4 (GNU/Linux)
Comment: Using GnuPG with Thunderbird - http://enigmail.mozdev.org

iD8DBQFAta1w4q3E2oMjYtURAt+AAJ0XTwmxsRsAgASfpfeWqs+DlQqdKwCfQ2Hy
dYqPGy93HVcqjYJTFsKyayE=
=nxDc
-----END PGP SIGNATURE-----