Re: r200 exa performance regression in xserver-1.6?
On Thu, Mar 26, 2009 at 3:15 PM, Daniel Kasak daniel.ka...@247realmedia.com wrote: Hi all. I've brought an old r200 card out of retirement to bring some bling to my work desktop. I was running xserver-1.5.something with an integrated nvidia card. I did an upgrade to xserver-1.6, mesa-7.4-rc1 and xf86-video-radeon-6.12.0. I don't have any figures for previous software on this hardware, but I can say that back when I was using this card at home on my old server, it used to *fly* - particularly after the exa-zero-copy-tfp landed. I certainly didn't get jumpiness or pauses like I'm getting now. So anyway, I'm running ecomorph ( compiz port for E17 ), and most of the effects are very slow and jumpy. Try Option AccelDFS true Dave. I did a quick sysprof test while switching desktops, and it looks like it's all in memcpy: miClearToBackground 0.00 77.45 miPaintWindow 0.00 77.45 ValidateGC 0.00 77.45 In file /usr/bin/Xorg 0.00 77.45 In file /usr/lib/xorg/modules/libexa.so 0.00 77.45 exaPrepareAccessGC 0.00 77.45 exaPrepareAccess 0.00 77.45 exaPrepareAccessReg 0.00 77.45 exaDoMigration 0.00 77.45 In file /usr/lib/xorg/modules/libexa.so 0.00 77.45 memcpy 77.45 77.45 Full sysprof output: http://entropy.homelinux.org/r200_exa.sysprof X log: http://entropy.homelinux.org/Xorg.0.log I can do more extensive testing later. I suppose next is to actually go back to an earlier xserver and test, right? Dan ___ xorg mailing list xorg@lists.freedesktop.org http://lists.freedesktop.org/mailman/listinfo/xorg ___ xorg mailing list xorg@lists.freedesktop.org http://lists.freedesktop.org/mailman/listinfo/xorg
Re: r200 exa performance regression in xserver-1.6?
On Thu, 2009-03-26 at 16:15 +1100, Daniel Kasak wrote: So anyway, I'm running ecomorph ( compiz port for E17 ), and most of the effects are very slow and jumpy. I did a quick sysprof test while switching desktops, and it looks like it's all in memcpy: miClearToBackground 0.00 77.45 miPaintWindow 0.00 77.45 ValidateGC 0.00 77.45 In file /usr/bin/Xorg 0.00 77.45 In file /usr/lib/xorg/modules/libexa.so 0.00 77.45 exaPrepareAccessGC0.00 77.45 exaPrepareAccess0.00 77.45 exaPrepareAccessReg 0.00 77.45 exaDoMigration 0.00 77.45 In file /usr/lib/xorg/modules/libexa.so 0.00 77.45 memcpy 77.45 77.45 Unfortunately the EXA module is missing debugging symbols, so it isn't entirely clear what's going on. Would be interesting to see a new profile with that fixed. It might also be worth trying EXA from the xserver Git master branch. exaValidateGC has been reworked quite a bit, and there's a chance your use case could have improved. -- Earthling Michel Dänzer |http://www.vmware.com Libre software enthusiast | Debian, X and DRI developer ___ xorg mailing list xorg@lists.freedesktop.org http://lists.freedesktop.org/mailman/listinfo/xorg
Re: r200 exa performance regression in xserver-1.6?
Hi, On Thu, 2009-03-26 at 16:05 +1000, Dave Airlie wrote: Try Option AccelDFS true Any reason this kind of option isn't the default ? It seems xorg.conf is mandatory for radeons just for the don't suck true options. Xav ___ xorg mailing list xorg@lists.freedesktop.org http://lists.freedesktop.org/mailman/listinfo/xorg
Re: r200 exa performance regression in xserver-1.6?
I did a quick sysprof test while switching desktops, and it looks like it's all in memcpy: miClearToBackground 0.00 77.45 miPaintWindow 0.00 77.45 ValidateGC 0.00 77.45 In file /usr/bin/Xorg 0.00 77.45 In file /usr/lib/xorg/modules/libexa.so 0.00 77.45 exaPrepareAccessGC0.00 77.45 exaPrepareAccess0.00 77.45 exaPrepareAccessReg 0.00 77.45 exaDoMigration 0.00 77.45 In file /usr/lib/xorg/modules/libexa.so 0.00 77.45 memcpy 77.45 77.45 This looks a lot like what I had when I did not properly compile it to use DRI. Make sure DRI is available. ___ xorg mailing list xorg@lists.freedesktop.org http://lists.freedesktop.org/mailman/listinfo/xorg
Re: r200 exa performance regression in xserver-1.6?
On Thu, Mar 26, 2009 at 7:11 PM, Xavier Bestel xavier.bes...@free.fr wrote: Hi, On Thu, 2009-03-26 at 16:05 +1000, Dave Airlie wrote: Try Option AccelDFS true Any reason this kind of option isn't the default ? It seems xorg.conf is mandatory for radeons just for the don't suck true options. Its the default on PCI and PCIE systems, however it can make AGP system unreliable due to AGP being badly designed heap of crap. So on AGP people get to do it themselves. Dave. ___ xorg mailing list xorg@lists.freedesktop.org http://lists.freedesktop.org/mailman/listinfo/xorg
Re: r200 exa performance regression in xserver-1.6?
On Thu, 2009-03-26 at 20:29 +1000, Dave Airlie wrote: On Thu, Mar 26, 2009 at 7:11 PM, Xavier Bestel xavier.bes...@free.fr wrote: Hi, On Thu, 2009-03-26 at 16:05 +1000, Dave Airlie wrote: Try Option AccelDFS true Any reason this kind of option isn't the default ? It seems xorg.conf is mandatory for radeons just for the don't suck true options. Its the default on PCI and PCIE systems, however it can make AGP system unreliable due to AGP being badly designed heap of crap. That's pretty clear. So on AGP people get to do it themselves. Thanks for the explanation. Xav ___ xorg mailing list xorg@lists.freedesktop.org http://lists.freedesktop.org/mailman/listinfo/xorg
Re: r200 exa performance regression in xserver-1.6?
On Thursday 26 March 2009 06:05:32 Dave Airlie wrote: Option AccelDFS true Just out of curiosity, why is this needed? I thought compositing was mostly just texturing onto the root window back buffer and could mostly be done in h/w? -- Bill, also wondering why compositing seems to be slow on his (AGP) r200 at home. ___ xorg mailing list xorg@lists.freedesktop.org http://lists.freedesktop.org/mailman/listinfo/xorg
Re: r200 exa performance regression in xserver-1.6?
On Thu, 2009-03-26 at 16:05 +1000, Dave Airlie wrote: So anyway, I'm running ecomorph ( compiz port for E17 ), and most of the effects are very slow and jumpy. Try Option AccelDFS true Dave. Woah. OK that made things a lot better. It's still not perfect ( as per my memory of my home server with this card ), but a lot better. When I get some time ( I'm at work ) I'll try out Michel's suggestions ( try xserver git master, if this doesn't improve things, then rebuild with debugging symbols and sysprof again ). Thanks guys. This is a lot better now anyway. Dan ___ xorg mailing list xorg@lists.freedesktop.org http://lists.freedesktop.org/mailman/listinfo/xorg
r200 exa performance regression in xserver-1.6?
Hi all. I've brought an old r200 card out of retirement to bring some bling to my work desktop. I was running xserver-1.5.something with an integrated nvidia card. I did an upgrade to xserver-1.6, mesa-7.4-rc1 and xf86-video-radeon-6.12.0. I don't have any figures for previous software on this hardware, but I can say that back when I was using this card at home on my old server, it used to *fly* - particularly after the exa-zero-copy-tfp landed. I certainly didn't get jumpiness or pauses like I'm getting now. So anyway, I'm running ecomorph ( compiz port for E17 ), and most of the effects are very slow and jumpy. I did a quick sysprof test while switching desktops, and it looks like it's all in memcpy: miClearToBackground 0.00 77.45 miPaintWindow 0.00 77.45 ValidateGC 0.00 77.45 In file /usr/bin/Xorg 0.00 77.45 In file /usr/lib/xorg/modules/libexa.so 0.00 77.45 exaPrepareAccessGC0.00 77.45 exaPrepareAccess0.00 77.45 exaPrepareAccessReg 0.00 77.45 exaDoMigration 0.00 77.45 In file /usr/lib/xorg/modules/libexa.so 0.00 77.45 memcpy 77.45 77.45 Full sysprof output: http://entropy.homelinux.org/r200_exa.sysprof X log: http://entropy.homelinux.org/Xorg.0.log I can do more extensive testing later. I suppose next is to actually go back to an earlier xserver and test, right? Dan ___ xorg mailing list xorg@lists.freedesktop.org http://lists.freedesktop.org/mailman/listinfo/xorg
Re: r200 exa performance regression in xserver-1.6?
On 3/26/09, Daniel Kasak daniel.ka...@247realmedia.com wrote: Hi all. I've brought an old r200 card out of retirement to bring some bling to my work desktop. I was running xserver-1.5.something with an integrated nvidia card. I did an upgrade to xserver-1.6, mesa-7.4-rc1 and xf86-video-radeon-6.12.0. I don't have any figures for previous software on this hardware, but I can say that back when I was using this card at home on my old server, it used to *fly* - particularly after the exa-zero-copy-tfp landed. I certainly didn't get jumpiness or pauses like I'm getting now. So anyway, I'm running ecomorph ( compiz port for E17 ), and most of the effects are very slow and jumpy. Does adding a virtual line to the screen section of your config to limit the front buffer size help? Virtual 1280 1024 Alex ___ xorg mailing list xorg@lists.freedesktop.org http://lists.freedesktop.org/mailman/listinfo/xorg
Re: r200 exa performance regression in xserver-1.6?
On Thu, 2009-03-26 at 01:36 -0400, Alex Deucher wrote: On 3/26/09, Daniel Kasak daniel.ka...@247realmedia.com wrote: Hi all. I've brought an old r200 card out of retirement to bring some bling to my work desktop. I was running xserver-1.5.something with an integrated nvidia card. I did an upgrade to xserver-1.6, mesa-7.4-rc1 and xf86-video-radeon-6.12.0. I don't have any figures for previous software on this hardware, but I can say that back when I was using this card at home on my old server, it used to *fly* - particularly after the exa-zero-copy-tfp landed. I certainly didn't get jumpiness or pauses like I'm getting now. So anyway, I'm running ecomorph ( compiz port for E17 ), and most of the effects are very slow and jumpy. Does adding a virtual line to the screen section of your config to limit the front buffer size help? Virtual 1280 1024 Alex Hi Alex. Thanks for the quick response. Unfortunately, no that doesn't seem to affect it at all. xorg.conf which I forgot to attach last time: http://entropy.homelinux.org/xorg.conf Dan ___ xorg mailing list xorg@lists.freedesktop.org http://lists.freedesktop.org/mailman/listinfo/xorg