Bug 22759

Summary: [i845] xorg intel driver 2.7.99 36% slower than 2.7.0 on i845
Product: xorg Reporter: Bryce Harrington <bryce>
Component: Driver/intelAssignee: Carl Worth <cworth>
Status: RESOLVED FIXED QA Contact: Xorg Project Team <xorg-team>
Severity: major    
Priority: medium CC: jerrylamos
Version: 7.4 (2008.09)   
Hardware: x86 (IA32)   
OS: Linux (All)   
Whiteboard:
i915 platform: i915 features:
Attachments:
Description Flags
GtkPerf.2.7.99.ods
none
Xorg.0.log in KMS modeset=1
none
GtkPerf.karmic.i845.ods none

Description Bryce Harrington 2009-07-13 16:22:19 UTC 
Created attachment 27657 [details]
GtkPerf.2.7.99.ods

Forwarding this bug from Ubuntu reporter jerrylamos:
https://bugs.edge.launchpad.net/ubuntu/+source/xserver-xorg-video-intel/+bug/382017

[Problem]
Performance regression on 8xx as measured by GtkPerf between 2.7.0 to 2.7.99 on 2.6.30 kernel.

[Original Report]
GtkPerf from Synaptic measures several video performance characteristics. Results with Karmic Alpha 1 on i845 video on IBM ThinkCentre and i830 video on IBM Thinkpad R31 significant performance slowdown of driver 2.7.99 vs. 2.7.0, and for that matter VESA:

i845 2.7.99 42 seconds vs. 2.7.0 31 seconds, 2.799 being 36% slower
i830 2.7.99 65 seconds vs. 2.7.0 35 seconds, 2.7.99 being 90% slower

i845 2.7.99 42 seconds vs. VESA 27 seconds, 2.7.99 being 58% slower

Attached should be ubuntu-bug -p xorg data from the ThinkCentre i845

Also attached spreadsheet with details on which performance measures were worst

Linux version 2.6.30-5-generic (buildd@rothera) (gcc version 4.4.1 (Ubuntu 4.4.0-3ubuntu3) ) #6-Ubuntu SMP Mon May 11 19:56:30 UTC 2009

2.7.0 came with the Alternate CD. 2.7.99 was installed from "Edgers" as described in the Ubuntu Forums Intel Performance Guidelines.

Latest results from karmic 2.6.30.10.10:

I'm not sure what KMS is for on intel video graphics?

i830 Thinkpad R31 boots to blackscreen. nomodeset in kernel line turns off KMS then boots O.K. but of course KMS is turned off. That's another launchpad bug.

i845 ThinkCentre 2.0 gHz Celeron works O.K. but Noticeably slower than generic video driver vesa, for example GtkPerf from Synaptic:

06/24/09 karmic 2.6.30.10.10
karmic.i845_________ vesa__KMS=1__ KMS_slower
Entry_________________ 0.22 0.26 18%
ComboBox____________ 5.55 9.25 67%
ComboBoxEntry_______ 3.84 5.52 44%
SpinButton____________ 0.85 2.5 194%
ProgressBar___________ 0.82 3.97 384%
ToggleButton__________ 0.95 2.47 160%
CheckButton___________ 0.60 1.10 83%
RadioButton____________ 0.87 1.35 55%
TextView-Add-text______ 1.93 1.83 -5%
TextView-Scroll_________ 1.29 1.97 53%
DrawingArea-Lines______ 2.11 1.83 -13%
DrawingArea-Circles____ 4.15 9.81 136%
DrawingArea-Text_______ 4.77 5.53 16%
DrawingArea-Pixbufs____ 0.26 0.51 96%

Total_time_____________ 28.24 47.92 70%

Progress bar in particular visibly drags across the screen....

ProblemType: Bug
Architecture: i386
Date: Sat May 30 20:30:30 2009
DistroRelease: Ubuntu 9.10
Package: xorg 1:7.4~5ubuntu20
ProcEnviron:
 LANG=en_US.UTF-8
 SHELL=/bin/bash
ProcVersionSignature: Ubuntu 2.6.30-5.6-generic
RelatedPackageVersions:
 xserver-xorg 1:7.4~5ubuntu20
 libgl1-mesa-glx 7.5.0~git20090523+mesa-7-5-branch.1045481d-0ubuntu0sarvatt
 libdrm2 2.4.11+git20090519.f355ad89-0ubuntu0sarvatt~jaunty
 xserver-xorg-video-intel 2:2.7.99.1+git20090526.r1.8e942b70-0ubuntu0sarvatt~jaunty
 xserver-xorg-video-ati 1:6.12.99+git20090527.b34df233-0ubuntu0sarvatt~jaunty
SourcePackage: xorg
Uname: Linux 2.6.30-5-generic i686
system:
 distro: Ubuntu
 architecture: i686kernel: 2.6.30-5-generic
Comment 1 Gordon Jin 2009-07-13 20:36:59 UTC 
not considering performance issue on 8xx as high priority.
Comment 2 Jerry Amos 2009-07-14 05:23:30 UTC 
Created attachment 27679 [details]
Xorg.0.log in KMS modeset=1
Comment 3 Bryce Harrington 2009-07-14 10:38:40 UTC 
Created attachment 27684 [details]
GtkPerf.karmic.i845.ods

More data against more recent code...

Bryce,

Results with 2.6.31-2. Formatting columnar data is a bit of a bear on launchpad so I'll attach the spreadsheet. I also have results on IBM Thinkpad R31 i830 and IBM Thinkpad T40 Radeon Mobility 7400 which I hope to post tomorrow.

Jerry

1st results column: jauntyCD.i845.2.6.3.-0ubuntu9
2nd results column: karmic.2.26.31-2.i845.vesa
3rd results column: karmic.i845.2.27.99.901+git20090702.74227141-0ubuntu1

GtkPerf 0.40Startingtesting: Mon Jul 13 12:12:44 AM 2009
jauntyCD.i845.2.6.3.-0ubuntu9 jaunty Karmic vesa difference 2.79901 vs. jaunty vs. vesa
GtkEntrytime: 0.28 0.25 --------- 0.27 -4% 8%
GtkComboBoxtime: 6.66 6.22 --------- 7.21 8% 16%
GtkComboBoxEntrytim: 4.12 4.56 --------- 4.68 14% 3%
GtkSpinButtontime: 1.05 0.93 --------- 1.98 89% 113%
GtkProgressBartime: 1.98 0.85 --------- 2.08 5% 145%
GtkToggleButtontime: 1.1 1.04 --------- 1.55 41% 49%
GtkCheckButtontime: 0.85 0.64 --------- 1.22 44% 91%
GtkRadioButtontime: 1.6 0.96 --------- 1.12 -30% 17%
GtkTextViewAddtextti: 1.96 2.04 --------- 1.85 -6% -9%
GtkTextViewScrolltime: 1.57 1.5 --------- 1.61 3% 7%
GtkDrawingAreaLines: 2.85 3.07 --------- 3.05 7% -1%
GtkDrawingAreaCircles: 2.82 5.25 --------- 11.90 322% 127%
GtkDrawingAreaTextt: 6.09 5.65 --------- 6.44 6% 14%
GtkDrawingAreaPixbuf: 1.07 0.28 --------- 0.52 -51% 86%
 --- ---------
Totaltime: 34.02 33.26 --------- 45.49 34% 37%
Comment 4 Eric Anholt 2009-07-16 14:54:39 UTC 
The main thing 8xx 2D performance needs is support for A8 destinations so that glyph accumulation doesn't take so long.  I've got patches to do that, but they trigger 8xx render stability problems sooner, so I'm holding off.
Comment 5 Michel Dänzer 2009-07-17 01:35:48 UTC 
(In reply to comment #4)
> The main thing 8xx 2D performance needs is support for A8 destinations so that
> glyph accumulation doesn't take so long.

FWIW, EXA in xserver master automatically uses a 32 bpp picture if the driver can't render to A8.
Comment 6 Eric Anholt 2009-07-22 10:49:21 UTC 
commit 12c5aeca7a3db92d3522d00f5daf338d522e2176
Author: Eric Anholt <eric@anholt.net>
Date:   Thu Jul 16 13:34:09 2009 -0700

    8xx render: Add limited support for a8 dests.
    
    This improves aa10text performance from 74k to 569k on my 855 laptop.
    This also causes my 865 to hang on aa10text like it does on rgb10text,
    thanks to actually hitting render accel.

(the previous general performance fixes likely also impact gtkperf)

Use of freedesktop.org services, including Bugzilla, is subject to our Code of Conduct. How we collect and use information is described in our Privacy Policy.