Welcome to the Second Life Forums Archive

These forums are CLOSED. Please visit the new forums HERE

Testing the new graphics features

Aires Pierterson
New England Fur
Join date: 24 Dec 2005
Posts: 39
09-30-2006 08:44
Starting with hardware:

2.2 Ghz AMD64, 128MB AtiX200 graphics "card", 640MB RAM, running through a fully updated Suse 10.1. (I have 128 mb of shared system memory dedicated to the X200 because Suse crashes when I try running the card off of it's dedicated memory)

I was getting 4.4 - 4.5 fps when I logged in with a clean install. After enabling VBO, I went to 4.7 - 4.8 fps. not a very dramatic improvement, but an improvement nonetheless.

AGP lowered my framerate slightly, by maybe .1 or .2 fps, but was wiped from settings.xml when I logged out, which seems to be a common problem for everyone else. Shiny works ok. I haven't tried AVP (vertex program), cloth, or ripple water yet since i've yet to edit them in settings, but all in all it runs about the same as windows, maybe 1 fps slower, which is still bearable but nothing fancy.
Mathurin LeSuere
Registered User
Join date: 11 Jul 2006
Posts: 3
09-30-2006 09:11
From: ninjafoo Ng
Enabled everything, slight drop in fps, SHINEY !!!!!!!!!!!!11111111111

scary thing is ansotrpoic filtering onlys costs me a couple of fps - w o w


OpenGL + Linux with the highly improved drivers are going to allow for that sort of thing to happen. This is almost as good (in terms of visuals) as the Windows version. More crashing, but it PERFORMS BETTER!

OpenGL + Linux can do anything... And we have nVidia and ATI to thank for their vast improvement in driver performance over this year!
Llauren Mandelbrot
Twenty-Four Weeks Old.
Join date: 26 Apr 2006
Posts: 665
*** glibc detected *** double free or corruption
09-30-2006 09:13
CODE
2006-09-30T16:11:49Z WARNING: LLDir_Linux not fully implemented!
2006-09-30T16:11:49Z INFO: QUEUED THREAD STARTING
2006-09-30T16:11:49Z INFO: QUEUED THREAD STARTING
2006-09-30T16:11:49Z INFO: QUEUED THREAD STARTING
2006-09-30T16:11:49Z INFO: QUEUED THREAD STARTING
2006-09-30T16:11:50Z INFO: Checking marker file for lock...
2006-09-30T16:11:50Z INFO: Marker file isn't locked.
2006-09-30T16:11:50Z INFO: Checking marker file for lock...
2006-09-30T16:11:50Z INFO: Marker file created.
2006-09-30T16:11:50Z INFO: Removing message.log
2006-09-30T16:11:50Z INFO: Exiting init_marker_file().
2006-09-30T16:11:50Z INFO: Opening debug file /home/llauren/SecondLife_i686_1_12_1_13/SecondLife/logs/debug_info.log
2006-09-30T16:11:50Z INFO: Second Life version 1.12.1
2006-09-30T16:11:50Z INFO: Local time: 2006-09-30T12:11:50 EDT
2006-09-30T16:11:50Z INFO: CPU info:
2006-09-30T16:11:50Z processor : 0 vendor_id : AuthenticAMD cpu family : 6 model : 4 model name : AMD Athlon(tm) processor stepping: 4 cpu MHz : 1052.667 cache size : 256 KB fdiv_bug : no hlt_bug : no f00f_bug : no coma_bug : no fpu : yes fpu_exception : yes cpuid level : 1 wp : yes flags : fpu vme de pse tsc msr pae mce cx8 sep mtrr pge mca cmov pat pse36 mmx fxsr syscall mmxext 3dnowext 3dnow bogomips : 2107.09
2006-09-30T16:11:50Z INFO: Memory info:
2006-09-30T16:11:50Z MemTotal: 272484 kB MemFree: 3088 kB Buffers: 1208 kB Cached: 93328 kB SwapCached: 0 kB Active: 214012 kB Inactive: 30556 kB HighTotal: 0 kB HighFree: 0 kB LowTotal: 272484 kB LowFree: 3088 kB SwapTotal: 240932 kB SwapFree: 221972 kB Dirty: 24 kB Writeback: 0 kB Mapped: 226912 kB Slab: 14108 kB CommitLimit: 377172 kB Committed_AS: 327148 kB PageTables: 1680 kB VmallocTotal: 753656 kB VmallocUsed: 25084 kB VmallocChunk: 725492 kB
2006-09-30T16:11:50Z INFO: OS info: Linux 2.6.15-27-386 #1 PREEMPT Sat Sep 16 01:51:59 UTC 2006 i686
2006-09-30T16:11:50Z INFO: Loading feature tables.
2006-09-30T16:11:50Z INFO: Applying Feature Mask: CPUSlow
2006-09-30T16:11:50Z INFO: Loading configuration file /home/llauren/SecondLife_i686_1_12_1_13/SecondLife/user_settings/settings.xml
2006-09-30T16:11:50Z INFO: Loading art table from /home/llauren/SecondLife_i686_1_12_1_13/app_settings/viewerart.xml
2006-09-30T16:11:50Z INFO: Loading base colors from /home/llauren/SecondLife_i686_1_12_1_13/app_settings/colors_base.xml
2006-09-30T16:11:50Z INFO: Loading user colors from /home/llauren/SecondLife_i686_1_12_1_13/app_settings/colors.xml
2006-09-30T16:11:50Z INFO: Failed to load user colors from /home/llauren/SecondLife_i686_1_12_1_13/app_settings/colors.xml
2006-09-30T16:11:50Z INFO: Loading legacy colors from /home/llauren/SecondLife_i686_1_12_1_13/app_settings/colors.ini
2006-09-30T16:11:50Z INFO: Renaming /home/llauren/SecondLife_i686_1_12_1_13/SecondLife/cache/data.db2.x.493059183 to /home/llauren/SecondLife_i686_1_12_1_13/SecondLife/cache/data.db2.x.1300619291
2006-09-30T16:11:50Z INFO: Renaming /home/llauren/SecondLife_i686_1_12_1_13/SecondLife/cache/index.db2.x.493059183 to /home/llauren/SecondLife_i686_1_12_1_13/SecondLife/cache/index.db2.x.1300619291
2006-09-30T16:11:58Z WARNING: VFS corruption: 2675085e-00dc-7c19-0064-0000aafe1d45 (-20923) at index 212976 DS: 524288000
2006-09-30T16:11:58Z WARNING: Length: 99328 Location: 438105088 Size: 1662817898
2006-09-30T16:11:58Z WARNING: File has bad data - VFS removed
2006-09-30T16:11:59Z WARNING: VFS corrupt, deleted. Making new VFS.
*** glibc detected *** double free or corruption (!prev): 0x0a2f51b8 ***
Tofu Linden
Linden Lab Employee
Join date: 29 Aug 2006
Posts: 471
10-02-2006 06:28
From: SpacedOut Frye
I assume the mouse grabbing on alt and control was removed? It is working great here for alt and control etc in 1.12.1.13, thank you.
Thanks for the feedback. We're doing it differently and overall less aggressively (now going behind SDL's back) in 1.12.1.13 nowards.
Tofu Linden
Linden Lab Employee
Join date: 29 Aug 2006
Posts: 471
10-02-2006 06:39
Thanks everyone for the valuable feedback.

It seems that OpenGL extension driver stability on Linux is quite a mixed bag, as anticipated. Until we put a Linux-specific blacklist/whitelist in place such as the other platforms enjoy, I expect that we'll disable some or all GL extensions on Linux by default in the next release (don't worry - I'll leave some way to manually turn them on again).
Tofu Linden
Linden Lab Employee
Join date: 29 Aug 2006
Posts: 471
10-02-2006 06:45
When reporting any bug (or even success) relating to graphics or lockups, it's helpful if you could include the output from the following command:
CODE
glxinfo | grep string
This will help us to eventually build a whitelist/blacklist database for troublesome hardware.
If reporting a startup crash, it is helpful to include your whole SecondLife/logs/SecondLife.log file. An automated crash-reporter is high on my TODO list.
Thank you!
Llauren Mandelbrot
Twenty-Four Weeks Old.
Join date: 26 Apr 2006
Posts: 665
glxinfo | grep string
10-02-2006 07:02
CODE
server glx vendor string: NVIDIA Corporation
server glx version string: 1.4
client glx vendor string: NVIDIA Corporation
client glx version string: 1.4
OpenGL vendor string: NVIDIA Corporation
OpenGL renderer string: GeForce2 MX/AGP/3DNOW!
OpenGL version string: 1.5.6 NVIDIA 87.62
ninjafoo Ng
Just me :)
Join date: 11 Feb 2006
Posts: 713
10-02-2006 08:26
From: Tofu Linden
(don't worry - I'll leave some way to manually turn them on again).


Dont forget to document it somewhere too :)
_____________________
FooRoo : clothes,bdsm,cages,houses & scripts

QAvimator (Linux, MacOS X & Windows) : http://qavimator.org/
jrrdraco Oe
Insanity Fair
Join date: 28 Oct 2005
Posts: 372
Clicking BumpMapped is Instant crash
10-02-2006 09:26
I enter the game, VBO and AGP is on and runs very well, but when I try to turn BumpMapped Clot option on it crashes instantly

Lastrun last lines:

2006-10-02T16:23:29Z INFO: Link successful. The GLSL vertex shader will run in hardware. The GLSL fragment shader will run in hardware.
2006-10-02T16:23:29Z INFO: Attribute weight assigned to channel 3
2006-10-02T16:23:29Z INFO: Uniform matrixPalette[0] is at location 1 stored in index 5
2006-10-02T16:23:29Z INFO: Uniform diffuseMap is at location 46
2006-10-02T16:23:29Z INFO: Assigned to texture channel 0
2006-10-02T16:23:30Z INFO: remove_marker_file()

$glxinfo | grep string

server glx vendor string: SGI
server glx version string: 1.2
client glx vendor string: ATI
client glx version string: 1.3
OpenGL vendor string: ATI Technologies Inc.
OpenGL renderer string: RADEON 9600 XT Generic
OpenGL version string: 2.0.6065 (8.29.6)
_____________________
--
Linux Specs: http://www.immerdrauf.com/jrrhack/specs.txt
Kate Mommsen
Registered User
Join date: 22 Nov 2005
Posts: 10
10-05-2006 07:09
All options on, so far so good. Glxinfo data to follow once I get my ISDN back :-(

System specs: Gentoo 2006.1 / Linux 2.6.18 AMD64
2x Opteron 250
4GB RAM
2x GeForce 7800GT PCI-E

AGP made no real difference, VBO gave a decent improvement. No hit to speak of from enabling all the shineyness, but things look a heckuva lot better. If things get really hairy, I drop off, switch on SLi and come back - kills video lag dead. (Have to go to single-head for SLi which is why I usually don't use it.)
Seg Baphomet
Fedora Developer
Join date: 1 Oct 2005
Posts: 46
10-07-2006 19:39
Shiny, ripple water, avatar vertex program, and bumpmapped & cloth avatar rendering work fine here.

VBO does not work, crashes after login, right when it tries to render the world.

Not sure whats up with AGP, it says it can't allocate AGP ram, and it disappears from the prefs file, but its checkmarked in the GUI. If VBO is enabled, it will allocate AGP ram, but VBO don't work...

server glx vendor string: SGI
server glx version string: 1.2
client glx vendor string: ATI
client glx version string: 1.3
OpenGL vendor string: ATI Technologies Inc.
OpenGL renderer string: RADEON 9800 SE Generic
OpenGL version string: 2.0.6065 (8.29.6)
Seg Baphomet
Fedora Developer
Join date: 1 Oct 2005
Posts: 46
10-07-2006 20:09
Okay after noticing jrrdraco Oe's post, I turned off the avatar vertex program, and enabled VBO and it worked!

Turing on wavy cloth crashes it. So on ATI, it seems VBO and wavy cloth conflict. Bump mapping is fine though.

AGP just doesn't seem to stick in the prefs anymore. But it seems to work anyway. Weird.
Angel Sunset
Linutic
Join date: 7 Apr 2005
Posts: 636
10-08-2006 03:51
Confirmed!

I turned off Avatar Vertex Program with the GUI, and Bump Mapped and Cloth, and Bump Mapped both greyed out.

I added the VBO enabling lines, and SL started perfectly.

I cannot REALLY notice an improved frame rate, though...

However, at least with one skin that had problems, the rendering works now! With the others, no real change - black face, pink torso - the mess varies from skin to skin :(

I can also turn the avatar vertex program back on, and bump mapped, and everything STILL works fine! All I lose is fluttery clothes :p
_____________________
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Kubuntu Intrepid 8.10, KDE, linux 2.6.27-11, X.Org 11.0, server glx vendor: NVIDIA Corporation, server glx version: 1.5.2, OpenGL vendor: NVIDIA Corporation, OpenGL renderer: GeForce 9800 GTX+/PCI/SSE2, OpenGL version: 3.0.0 NVIDIA 180.29, glu version: 1.3, NVidia GEForce 9800 GTX+ 512 MB, Intel Core 2 Duo, Mem: 3371368k , Swap: 2570360k
Seg Baphomet
Fedora Developer
Join date: 1 Oct 2005
Posts: 46
10-08-2006 04:16
I'm still wondering how you convince the fglrx drivers to disable vsync. Its gotta be killing my framerates. I can't find a word about it, making use of strings + grep on the drivers provides no hints. And amusingly enough, the fireglcontrolpanel thingy doesn't seem to actually configure anything having to do with OpenGL.

Maybe they're expecting the app itself to set it.
Angel Sunset
Linutic
Join date: 7 Apr 2005
Posts: 636
10-08-2006 04:45
With NVIdia 87.74 drivers, the nvidia-settings program allows you to sync to VBlank. This DOES kill frame rates, but can be turned off in the OpenGL Settings and the XServer XVideo setting within the nvidia-settings GUI.

Maybe this will help?
_____________________
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Kubuntu Intrepid 8.10, KDE, linux 2.6.27-11, X.Org 11.0, server glx vendor: NVIDIA Corporation, server glx version: 1.5.2, OpenGL vendor: NVIDIA Corporation, OpenGL renderer: GeForce 9800 GTX+/PCI/SSE2, OpenGL version: 3.0.0 NVIDIA 180.29, glu version: 1.3, NVidia GEForce 9800 GTX+ 512 MB, Intel Core 2 Duo, Mem: 3371368k , Swap: 2570360k
Seg Baphomet
Fedora Developer
Join date: 1 Oct 2005
Posts: 46
10-09-2006 23:21
Okay apparently to disable vsync with fglrx you have to add/change this to your Device section in xorg.conf:

Option "Capabilities" "0x00000000"

SO OBVIOUS! WHY DIDN'T I THINK OF THAT?

With vsync off, best I can tell, VBO actually drops my framerate by about 3-5 fps. Hard to be sure though.
Major Senior
Registered User
Join date: 12 Apr 2006
Posts: 104
flgrx lockup related to window focus?
10-10-2006 02:15
Yah, this is a bit weird, but oh well. Seems I can start up SL just fine w/out using LL_GL_NOEXT=x if I take away window focus directly after doing it.

Similarly, I have periodically noticed that the SL viewer "hangs" when trying during "Connecting to Region", which also disapears if SL does not have window focus.

Once started, the entire world renders fine, shiny works fine, ect.. But the second I give SL focus, I lose the keyboard interface and the entire desktop locks up. Mouse cursor still functions, but X is effectively hung at that point and nothing is selectable. Even stuff like the system performance monitor stop updating. Logging in remotely and killing secondlife-bin free's up the system and I can use it again.

Toggling VBO has no effect on the system.

Toggling Shiny causes the lock-up regardless of mouse focus.

server glx vendor string: SGI
server glx version string: 1.2
client glx vendor string: ATI
client glx version string: 1.3
OpenGL vendor string: ATI Technologies Inc.
OpenGL renderer string: RADEON X300/X550 Series Generic
OpenGL version string: 2.0.5946 (8.27.10)
Major Senior
Registered User
Join date: 12 Apr 2006
Posts: 104
fglrx update
10-10-2006 04:16
Upgraded the ATI drivers to thier latest version, same problem, SL locks up the entire X desktop shortly after joining the region. Mouse pointer still moves, keyboard no longer functions. Caps lock can not be toggled, CTRL+ALT+F1 does not kick out to the VC.

System itself is still responsive and logging in remotely works. Killing the SecondLife system unlocks the X desktop.

Starting SL with LL_GL_NOEXT=x prevents the deadlock from occuring.

When the SL viewer does not have focus, the lock-up does not occure even when LL_GL_NOEXT is not set.

The observed behavior of the SL hanging when Connecting to Region only when SL has focus is not the case. I ran across it hanging a number of times with and w/out focus.

Have validated that the problem occures regardless of which version of kernel preemption is compiled into the kernel (including no-preemption).

Running out of ideas for what to look at next. That the problem only occures when SL has mouse focus is most frustrating in the end. The way in which it deadlocks X makes it feel like it is a very low level problem, but I have yet to find it. Pondering attaching to the process with gdb while it is locked to see what it was doing.

server glx vendor string: SGI
server glx version string: 1.2
client glx vendor string: ATI
client glx version string: 1.3
OpenGL vendor string: ATI Technologies Inc.
OpenGL renderer string: RADEON X300/X550 Series Generic
OpenGL version string: 2.0.6065 (8.29.6)
Major Senior
Registered User
Join date: 12 Apr 2006
Posts: 104
Debugging the fun stuff...
10-10-2006 04:47
From: Major Senior
Pondering attaching to the process with gdb while it is locked to see what it was doing.


Okay, as an added tidbit to all of this, so long as the MOUSE hasn't been moved in to hover over the SecondLife window, I can still use the keyboard to talk, type, fly around, ect.. It seems to be purely related to the mouse for some reason. Perhaps I should try changing the mouse type?

Anyway, the gdb backtrace looks pretty much confused, SL kicks a path of execution that dives down through the libraries and grabs an mlock() and the whole system sits there waiting on the mlock(). I have no idea what it is mlock()'ing at that point. Maybe the mouse? *shrug* The only thing realy weird is the thread addresses, it makes no sense that everything should be pointing to the same region of memory. gdb just confused?

Further more, the process keeps its Runnable flag the entire time, so this isn't a mutex sleeper problem. Time to go play with fglrx settings I suppose. Try a hardare cursor or something... Bleh.

(gdb) bt
#0 0xffffe410 in ?? ()
#1 0xbfd6b0d8 in ?? ()
#2 0x00000001 in ?? ()
#3 0x00002000 in ?? ()
#4 0xb76b2946 in mlock () from /lib/libc.so.6
#5 0xb4e36669 in s9341 () from /usr/lib/xorg/modules/dri/fglrx_dri.so
#6 0xb4e4e752 in s6654 () from /usr/lib/xorg/modules/dri/fglrx_dri.so
#7 0xb4b5b442 in s19110 () from /usr/lib/xorg/modules/dri/fglrx_dri.so
#8 0xb4b94bc0 in s14399 () from /usr/lib/xorg/modules/dri/fglrx_dri.so
#9 0xb4c0dc33 in s5200 () from /usr/lib/xorg/modules/dri/fglrx_dri.so
#10 0xb4bfce06 in s8194 () from /usr/lib/xorg/modules/dri/fglrx_dri.so
#11 0xb4bfcc57 in s8194 () from /usr/lib/xorg/modules/dri/fglrx_dri.so
#12 0xb4bfcd86 in s8194 () from /usr/lib/xorg/modules/dri/fglrx_dri.so
#13 0x092fd8ba in LLViewerWindow::performPick ()
#14 0x0908c5d9 in display ()
#15 0x0959e84f in main_loop ()
#16 0x095a5cda in main ()
(gdb) thread 2
[Switching to thread 2 (Thread -1453225056 (LWP 9309))]#0 0xffffe410 in ?? ()
(gdb) bt
#0 0xffffe410 in ?? ()
#1 0xa9618438 in ?? ()
#2 0x00056b5a in ?? ()
#3 0x00000000 in ?? ()
(gdb) thread 3
[Switching to thread 3 (Thread -1444832352 (LWP 9308))]#0 0xffffe410 in ?? ()
(gdb) bt
#0 0xffffe410 in ?? ()
#1 0xa9e193ec in ?? ()
#2 0x0a74b8e8 in ?? ()
#3 0x00000000 in ?? ()
(gdb) thread 4
[Switching to thread 4 (Thread -1246450784 (LWP 9305))]#0 0xffffe410 in ?? ()
(gdb) bt
#0 0xffffe410 in ?? ()
#1 0xb5b4a1b8 in ?? ()
#2 0x00000001 in ?? ()
#3 0x00000000 in ?? ()
(gdb) thread 5
[Switching to thread 5 (Thread -1238058080 (LWP 9304))]#0 0xffffe410 in ?? ()
(gdb) bt
#0 0xffffe410 in ?? ()
#1 0xb634b1b8 in ?? ()
#2 0x00000007 in ?? ()
#3 0x00000000 in ?? ()
(gdb) thread 6
[Switching to thread 6 (Thread -1229665376 (LWP 9303))]#0 0xffffe410 in ?? ()
(gdb) bt
#0 0xffffe410 in ?? ()
#1 0xb6b4c1b8 in ?? ()
#2 0x000003ef in ?? ()
#3 0x00000000 in ?? ()
(gdb) thread 7
[Switching to thread 7 (Thread -1221272672 (LWP 9302))]#0 0xffffe410 in ?? ()
(gdb) bt
#0 0xffffe410 in ?? ()
#1 0xb734d1b8 in ?? ()
#2 0x00000001 in ?? ()
#3 0x00000000 in ?? ()
Zi Ree
Mrrrew!
Join date: 25 Feb 2006
Posts: 723
10-10-2006 06:16
"LLViewerWindow::PerformPick ()"

This looks like a method that's supposed to return the object ID of what your mouse pointer is currently pointing at. Not sure if this helps in any way, though.
_____________________
Zi!

(SuSE Linux 10.2, Kernel 2.6.13-15, AMD64 3200+, 2GB RAM, NVidia GeForce 7800GS 512MB (AGP), KDE 3.5.5, Second Life 1.13.1 (6) alpha soon beta thingie)

Blog: http://ziree.wordpress.com/ - QAvimator: http://qavimator.org

Second Life Linux Users Group IRC Channel: irc.freenode.org #secondlifelug
Major Senior
Registered User
Join date: 12 Apr 2006
Posts: 104
ATI fglrx issue resolved...
10-10-2006 08:41
Okay, so apparently SecondLife doesn't get along with the fglrx kernel module's userpage locking. This took way to long to locate.

The solution is to kill X completely, unload the fglrx module, and fixup the X config file to add the following line to the flgrx Device section:

Option "KernelModuleParm" "locked-userpages=0"

While you are there, you might want to turn on fast local thread storage:

Option "UseFastTLS" "1"

And as mentioned before, for those wishing to shutdown vsync and try to push the FPS up a touch:

Option "Capabilities" "0x00000000"
Tofu Linden
Linden Lab Employee
Join date: 29 Aug 2006
Posts: 471
10-10-2006 09:14
Thanks, the locked-userpages thing is interesting data.
Major Senior
Registered User
Join date: 12 Apr 2006
Posts: 104
ATI fglrx locked-userpages
10-10-2006 09:44
It might be worth noting that the locked-userpages option is no longer used on ATI AGP cards as of the most recent drivers. While you can set the option to turn them on or off, it does nothing on AGP cards. The option currently only does anything on PCI/PCIE cards.
Seg Baphomet
Fedora Developer
Join date: 1 Oct 2005
Posts: 46
10-11-2006 15:53
Okay, after futher testing, VBO enabled seems to cause SL to hang in busy areas with LOTS of prims. This is with 1.12.1.x, haven't tried it with 1.12.2.x yet.

Things are fine here without messing with the userpage locking, but yeah I am using an AGP card.

Things have always been fine without LL_GL_BASICEXT=x.

server glx vendor string: SGI
server glx version string: 1.2
client glx vendor string: ATI
client glx version string: 1.3
OpenGL vendor string: ATI Technologies Inc.
OpenGL renderer string: RADEON 9800 SE Generic
OpenGL version string: 2.0.6065 (8.29.6)
1 2