gtkperf Add Text bottleneck

[kinzeron]

Hi,

I keep getting a very long time with gtkperf for add Text: 46s!!! with IPP=2 and GC=1. I wonder if there is something wrong. from 16x to the new 177.68

I have a core 2 duo 2Ghz, 8400G, 2Gb RAM

Here is my xorg.conf:

Section "Monitor"
Identifier "Configured Monitor"
VendorName "Unknown"
ModelName "Unknown"
HorizSync 30.0 - 110.0
VertRefresh 50.0 - 150.0
Option "DPMS"
EndSection

Section "Device"
Identifier "Configured Video Device"
Driver "nvidia"
VendorName "NVIDIA Corporation"
Option "AddARGBGLXVisuals" "true"
Option "DamageEvents" "true"
Option "RenderAccel" "true"

Option "AllowGLXWithComposite" "true"

Option "AddARGBGLXVisuals" "true"

EndSection

Section "Screen"
Identifier "Default Screen"
Device "Configured Video Device"
Monitor "Configured Monitor"
DefaultDepth 24
SubSection "Display"
Depth 24
EndSubSection
Option "PixmapCacheSize" "2000000"
Option "AllowSHMPixmaps" "0"
EndSection

Section "Extensions"
Option "DAMAGE" "Enable"
Option "Composite" "Enable"
Option "RENDER" "Enable"
EndSection

Section "ServerFlags"
Option "AIGLX" "on"
EndSection

[Dragoran]

see http://www.nvnews.net/vbulletin/show...45#post1753045

[kinzeron]

Hmm, they said it is gona be fixed with 177.*, I have the 177.68 driver...
So, what's wrong?

[zander]

The problem in question is expected to be fixed in 177.xx driver releases more recent than 177.68 (the problem was identified and fixed after the latter was built).

[kinzeron]

don't want to be pushy, but I have been waiting for more than a year now, so is there a release date?

[zander]

Although I expect an updated BETA driver will be available very soon, I can't provide a release date.

[kinzeron]

ok, thx, hope it's gonna be soon

[AaronP]

I should point out that while the "add text" test does use the driver a little, it's mostly a benchmark of the libpango text layout engine:

CPU: Core 2, speed 800 MHz (estimated)
Counted CPU_CLK_UNHALTED events (Clock cycles when not halted) with a unit mask of 0x00 (Unhalted core cycles) count 100000
samples  %        image name               app name                 symbol name
118231   53.2714  libpangoft2-1.0.so.0.2002.3 libpangoft2-1.0.so.0.2002.3 (no symbols)
31761    14.3106  libpango-1.0.so.0.2002.3 libpango-1.0.so.0.2002.3 (no symbols)
10286     4.6346  libglib-2.0.so.0.1600.4  libglib-2.0.so.0.1600.4  (no symbols)
8817      3.9727  nvidia_drv.so            nvidia_drv.so            (no symbols)
7448      3.3558  libcairo.so.2.17.3       libcairo.so.2.17.3       (no symbols)
7150      3.2216  oprofiled                oprofiled                (no symbols)
6105      2.7507  libc-2.7.so              libc-2.7.so              (no symbols)
5612      2.5286  libpangocairo-1.0.so.0.2002.3 libpangocairo-1.0.so.0.2002.3 (no symbols)
4674      2.1060  Xorg                     Xorg                     (no symbols)
3309      1.4909  vmlinux-debug-2.6.24-19-generic vmlinux-debug-2.6.24-19-generic hpet_readl
2502      1.1273  pango-basic-fc.so        pango-basic-fc.so        (no symbols)
2354      1.0606  libpixman-1.so.0.10.0    libpixman-1.so.0.10.0    (no symbols)

As such, there's only so much that the driver can do for this test.

[kinzeron]

@AaronP: I see, but if there is so much difference (I m getting 46s while others are getting 5s), then the problem is from the driver, no? if not it defies logic, because, at least the experience shows it as such!

It is a driver problem, depends on the configuration, but it is still a driver problem.

I m willing to do other tests or beta tests to resolve this issue, but you have to acknowledge the problem.

Regards,

[AaronP]

Are you sure the people getting 5 sec were running with -c 1000? I wouldn't expect much better than about 30 sec even with an infinitely fast driver, at least for the version of libpango I have. Perhaps there's a version out there that's faster, but that's not a driver issue.

[kinzeron]

@AaronP: http://www.nvnews.net/vbulletin/showthread.php?t=115942

the 5-7secs seems to be from one post, however the rest got 20-30s, which is still two times better than mine. Maybe I m wrong, but I m assuming the numbers they are given are for gtkperf -c 1000. I will check to see if there is a faster pango, although that I doubt it.

If you need more tests, tell me.

[apaige]

The guy with the 5-7 seconds result has font anti-aliasing turned *off*, and uses the default theme, which has no bells and whistles and is very fast. Don't pay attention to it. I wouldn't be surprised if he had a fast CPU as well (or even overclocked).

I get 36s with a Phenom 9600 (2.3GHz), a 8600GT and the Mist theme. That's not too far off from your result. Try changing the theme, see how fast it goes.