[Intel-gfx] FireFox performance regressions XAA -> EXA -> UXA

Clemens Eisserer linuxhippy at gmail.com
Mon Jan 25 22:21:54 CET 2010


Hi,

Recently I worked on one of our university's internet terminals and
wondered how snappy scrolling my Gmail inbox felt,
compared to my personal Laptop which is way more powerful. (those
machines use Fedora-7 and intel+XAA).
I also remembered I "felt" some regressions during intel-2.4 ->
intel-2.5 and again when it switched to UXA, so I thought I
investigate a bit.

I did some comparisons using Carl's Firefox benchmark comparing
Fedora-12 (updated) with Fedora-8 (updated) using official FireFox-3.6
builds:

Results are         XAA /      EXA /    UXA:
GMail Inbox:     39ms / 100ms / 110ms
tomsguide:        56ms / 158ms / 316ms
Heise.de            12.6ms / 46ms / 53ms
phoronix.com   13.4ms / 57ms / 100ms

At least in this tests, XAA is at lot faster than both EXA/UXA. Keep
in mind Fedora-8 still has an old pixman version, without all the SSE2
optimizations added lately.
Ok, XAA is not fair, its comparing apples with oranges.
However even EXA from Xorg-server-1.3 which has been known for beeing
slow and unoptimized delivers better results than UXA, sometimes by a
factor of 2x!

Any idea why performance regressed so dramatically between 2.2.1+EXA
and 2.9.1+UXA?
I would call the workload above quite *real-world* and *common*, so
the fact that XA does not well in synthatic tests doesn't cound this
time.

Thank you in advance, Clemens

tomsguide: http://www.tomsguide.com/us/Nokia-N900-Maemo-iphone-android,review-1468.html
XAA: 2.6.26 / Xorg-1.3 / intel-2.1.1
EXA: 2.6.26 / Xorg-1.3 / intel-2.1.1
UXA: 2.6.31.12 + intel-2.9.1

My Laptop has an i945GM, C2D-2.0ghz/4MB-L2 cache and 3GB of DDR-533.
Xorg log is attached.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: Xorg.0.log
Type: application/octet-stream
Size: 33208 bytes
Desc: not available
URL: <http://lists.freedesktop.org/archives/intel-gfx/attachments/20100125/ed744dab/attachment.obj>


More information about the Intel-gfx mailing list