What's the official status of 3.5.x, anyway?

Duncan 1i5t5.duncan at cox.net
Tue Jul 28 20:27:48 BST 2009


Irving Leonard <irving.lp at feestudiantes.cujae.edu.cu> posted
200907281244.20644.irving.lp at feestudiantes.cujae.edu.cu, excerpted below,
on  Tue, 28 Jul 2009 12:44:20 -0400:

> I'm guessing here but maybe I can help you. It seems that you're using
> the openradeon driver, because the ati drivers works with 9600 and
> newers. I already post some issues in the kubuntu mailing list about the
> openradeon and kde4. It have huge problems, making a simple action like
> hovering over a mail in kmail consumes over 80% of CPU power. I get no
> answer, try some tweaks in xorg with no results and that was a show
> stopper for me, I had to return to kde3. I use a Mobility Radeon 9000IGP
> (misdetected by the kernel as 9100) and a Celeron at 2.8Ghz with 512MB
> RAM. I try the same kde4 in a similar but Intel powered graphics
> computer and worked just fine, even when the Intel chip (GPU) have fewer
> graphic features. So I guess there are major issues with kde4 and the
> radeon driver.

Yeah, couldn't legally run the closed source if I wanted to as I don't 
agree to waive liability rights on a black-box I can't even open to see 
what it is I'm actually agreeing to take on the liability of, so it's 
definitely the open drivers.

Meanwhile, as gentoo allows installing both kde3 and kde4 in parallel, 
and the setup is patched so the ksycoca files, etc, don't interfere with 
each other so it's actually possible to run apps from the one on the 
other without breaking either one, I decided to break down the problems I 
was having into chunks I could actually manage, by first converting to 
kde4 versions of individual apps like kmail and konsole, while still 
running on an otherwise working kde3 general desktop.

That actually worked out surprisingly well, and one at a time I got each 
app up and running, including killing kde3's kwin and starting kde4's 
kwin, to set it up.  All in all, it took quite some time configuring away 
from the defaults, but by the time I had finished stage one, the major 
individual kde4 apps I run were running fine on kde3.  Stage two was then 
to switch to actually starting kde4 instead of kde3, now that I had the 
individual apps including kwin generally working, and go to work on the 
kde4 desktop, plasma itself.  Again, this worked MUCH better this time, 
since konqueror, dolphin, kwin, kmail, all that, were already working 
reasonably well now, and the big problem was plasma, which I could now 
address on its own, without a whole bunch of other issues assaulting me 
all at once, stymieing almost all progress on anything at all!

Turns out that there were actually two, well, three, but I knew one 
already, issues affecting performance on kde4, which together slowed 
things down to the point it was almost unusable.

1) The r200 based chips max out OpenGL at 2048 px square.  I run dual 
1920x1200 monitors, stacked, for 1920x2400 total area.  All but the 
bottom 352 px of that can handle OpenGL, but apparently there's no 
provision for that in kwin4, it's an all or nothing thing, so it won't 
run OpenGL at all.  That leaves Composite based effects, which work quite 
well (as long as EXA is enabled, not XAA, in xorg) in kwin3, but in kwin4 
it limits to a small subset of those available, with no indication 
whatsoever of which ones work and which ones won't do a thing.

That's the one I already knew about.  But if it would do OpenGL on the 
part that actually worked, it would have been nice, and potentially 
faster.

2) kwin4's composite effects, namely transparency (I keep the other 
normal window composit effects, drop shadows, fading, disabled), SEEMED 
to be far slower than the same effects on kde3.

This one ended up being due to a particular default setting.  It's under 
System Settings > Look&Feel > Desktop > Desktop Effects > All Effects > 
Translucency.  I'd discovered that much and noted the various sliders, 
but OVERLOOKED the Fading Duration spinbox at the bottom of the left hand 
column.

I'm not sure what the actual "default" was, since that's exactly what it 
was labeled, "default", but the spinbox increments were in 100ms units, 
so it must have been several hundred milliseconds.  As I had the fade 
effects entirely turned off in kde3, setting that to zero was effectively 
the same thing, but I found I could actually use a fade, it just had to 
be far lower than the default 100ms spinbox increment -- I had to type it 
in.

Right now I'm running 50ms.  But after actual use for a few days, I think 
I'll lower that, to 20ms or so.

What seems to be happening is that while the duration labeling would seem 
to indicate that's the total length of the effect, and a several hundred 
ms (say 1/2 second) effect should be fine, at least on the Radeons, that 
setting seems to be the individual incremental step duration.  So if it's 
a 20-step effect and the duration is set for 500 ms, the actual effect 
runs for 10 full seconds!  That's fine just switching between two 
individual windows, but when there's multiple windows on the desktop, and 
you move the mouse across the desktop and thus across several windows, 
the effect at default duration will take several seconds for each, and 
you'll find yourself trying to interact with windows that haven't been 
activated yet as the system is still running the effect on the second 
window back!

Thus shortening the effect to tens of ms, as I said, 50 is usable, but I 
think 20 will be better, makes for a FAR more responsive desktop!  I may 
ultimately disable the fading by setting zero, and having it take 
immediate effect, as I think I had it set in kde3/kwin3.

3) Desktop plasmoids.

The card is apparently simply too slow, at least without OpenGL, to 
dynamically render transparency with dynamic (which might be all even 
apparently static) plasmoids rendering them thru the transparency of 
perhaps several layers of windows.  As kde3's kdesktop was static, this 
wasn't a big issue there.  But with the loss of hard separation between 
panel and desktop, I had initially configured a bunch of plasmoids on the 
desktop.  As I said, this simply didn't work well for this card, which 
apparently couldn't well handle rendering the plasmoids dynamically thru 
several layers of semi-transparent windows.

I was able to work around this issue by returning to a mostly panel 
widget oriented "dashboard", clearing the desktop of plasmoids, returning 
it to the static state it was on kde3.  Since the panels are either 
autohide or docked-always-on-top (I don't use the let windows cover 
option), panel plasmoids don't get rendered underneath semi-transparent 
windows, and thus don't tax the system as do desktop plasmoids.

As indicated, it seems to me that multiple window layers increase the 
slowdown, and I also suspect that plasma is dynamically redrawing even 
apparently static plasmoids, such that the system must recomposite them 
thru the window transparency as if they were actually dynamically 
changing displays.


Anyway, by experimenting with kwin4 on an otherwise general kde3 system 
-- in particular, without plasma running too -- I was able to find and 
set that kwin transparency effect fade duration appropriately, whereas 
I'm not sure /how/ long it would have taken with plasma compounding the 
problem at the same time, since with both of them, the system was really 
hardly practically workable due to lack of responsiveness.  And once I 
had that fixed, when I rebooted to kde4 and was finally working on just 
plasma, as the other apps were now working, working with the desktop 
plasmoids I noticed the effect they had too, and confirmed it relatively 
quickly by either moving off or shutting down all the desktop plasmoids.  
Now the system is actually usable, altho I'm still dealing with 
individual tweaks here and there to work around issues like no khotkeys4 
global multikey hotkeys (multi-modifier single-key works, but not multi-
key) that I HEAVILY depended on in kde3.

Of course killing transparency entirely should work as well, since X 
won't be rendering the multiple layers any more, but as I knew it worked 
fine in kde3, I was unwilling to settle for that.

So anyway, try enabling EXA in xorg.conf, instead of XAA.  If you're 
going to run transparency, turn the delay WAY down or off.  And again, if 
you're running transparency, kill desktop plasmoids.  Put them on panels 
or simply don't run them at all.  (Or if you do, make multiple activities 
and use the activity switcher, with no plasmoids on the desktop if you're 
actually using apps, switching to the plasmoid desktop only when you're 
not going to be running apps over top of it all.)

That should VASTLY improve the kde4 experience for any Radeon thru 
probably the r300 chip series at least.  I'm actually on kde4 by default, 
now, and it's actually usable. =:^)

-- 
Duncan - List replies preferred.   No HTML msgs.
"Every nonfree program has a lord, a master --
and if you use the program, he is your master."  Richard Stallman

___________________________________________________
This message is from the kde mailing list.
Account management:  https://mail.kde.org/mailman/listinfo/kde.
Archives: http://lists.kde.org/.
More info: http://www.kde.org/faq.html.




More information about the kde mailing list