AMD Catalyst 10.2 For Linux Gets Direct2D

[krionius]

Okay, so summarizing:

with Direct2DAccel:

- not too much faster but slightly faster 2D with compiz
- rendering glitches
- suspend/resume is borked, full freeze ! (without it it works!)

Well, i'm beginning to fear that this year's gonna be more like disaster than great improvement in 2d and catalyst... amd/ati hope you know what you're doing...

[Melcar]

Direct2DAccel=STRUE
interesting. only tried V1 and TRUE. let me see..

it works! Thanks Melcar.

So what kind of tests do you recommend?


Here's the distort i got using d2daccel on hd5650:

http://cdn.solidfiles.net/i/CH8t.jpg

window resize is indeed faster. compiz + gnome works okayish, but garbled parts are present..testing further.


any 2d tests recommended?

Wish I knew what is causing the black screens and segfaults with composite managers in some cases though. It works for some, but for others it doesn't. As of now I'm stuck with plain ol' 2D Kwin.
The corruption in rendering is expected. Direct2d on this driver revision is very early so expect issues. Corruption with certain rendering operations and slower than xaa (seems to be qorse with qt though). The people that are able to run composite managers with this driver report that direct2d seems a bit snappier. For testing I would just run the guitoolkits test on PTS and maybe some of the Java 2d tests; compare with xaa and direct2d without composition, and again with composition. Would be interesting to see the differences.

[energyman]

for me it is broken...

KDE 4.4 will not start with my 5770.

turn off effects and you can log in. Install previous drivers, turn off effects, update drivers, you can log in.

It SUCKS that this well known bug was not fixed.

[phtpht]

Upsetting many users is the fact that Catalyst 10.2 still lacks support for X Server 1.7, which has been released as stable since last October and has already had five point releases. With the X.Org Server 1.7-using Ubuntu 10.04 LTS coming in April, Catalyst 10.3 or 10.4 will support this updated X.Org Server finally. However, at the end of March this cycle will be restarted all over again once X Server 1.8 has been released.

What now, does ubuntu dictate the speed of development in AMD? What the heck? I want the driver to support today's X server, not the X server that was launched 4 months ago. Does this mean I'll have to wait another 4 months when 1.8 is out?

[Melcar]

I think AMD tries to coordinate the driver with the few distros they officially support (Ubuntu and OpenSuse if I'm not mistaken). Seems reasonable enough.

[phtpht]

I think AMD tries to coordinate the driver with the few distros they officially support (Ubuntu and OpenSuse if I'm not mistaken). Seems reasonable enough.

You mean: linger with stuff until the distros really need it.

[liangsuilong]

You mean: linger with stuff until the distros really need it.

In fact, Fedora has been the second most popular Linux distro.

I guess AMD will not support RHEL 6 as soon as it is released.

[energyman]

second most? where did you get that from?
and why would anybody use fedora at all?

[krionius]

My direct2dAccel vs Xaa result
on mobility hd5650 with GtkPerf Circles and Text (yeah, very bad test, but better than by eye)

Test run time in sec (lower better ;P)

	Xaa	Xaa + Compiz	D2d + Compiz	D2d
Circles	20	3.5	        4 	        9
Text	5.1	2.5	        4.6	        9
----------------------------------------------------
        ~13     ~3              ~4.3            ~9

So as of now Xaa + Compiz is faster by these numbers.

Still resizing/maximizing is much-much faster with D2d. Pity it corrupts the graphics and it hangs too (VT switch or suspend = hang)

[krionius]

oh, well, just found out that there's a patched Xorg for arch linux too that resolves the Maximizing/Resizing slowness for catalyst (xorg-server-catalyst-maximize-fix - which is an alternative package for xorg-server 1.6.3, patched)