Radeon Driver Gets Textured Video (Xv)

[phoronix]

Phoronix: Radeon Driver Gets Textured Video (Xv)

We reported earlier today from FOSDEM that Alex Deucher had been successful in getting Textured Video (X-Video) with the Radeon driver on R300 class hardware and that the R500/600 series could even support it in the near future. Well, support for Textured Video on the R100 to R400 has today landed into the xf86-video-ati git tree...

http://www.phoronix.com/vr.php?view=NjM1MQ

[ioannis]

I have an R200 and it has Xv already. What's new to this? Does it mean that now Xv maps properly in compiz? (i.e. redirected direct rendering)

[agd5f]

I have an R200 and it has Xv already. What's new to this? Does it mean that now Xv maps properly in compiz? (i.e. redirected direct rendering)

This adds support for Xv adapters using the texture engine for display and colorspace conversion. Since the textures draw directly to the framebuffer, they work with composited desktops and rotation. So yes, using the textured adapter will work properly with compiz. We still expose the overlay adapter as well however, so make sure you use the right one when running compiz.

[agd5f]

Dave just committed textured video support for r5xx.

[maggot_brain]

Dave just committed textured video support for r5xx.

OMG! This is what I've been waiting for. I've been stuck using Windows to watch videos cos of the awful tearing problem in fglrx! What will I need to build the git code on Debian testing? Some instructions if somebody could knock some up would be absolutely fantastic.

[dungeon]

Well, i tried git version with new Xv support on ATI 9250 128 MB with or w/o compiz and i got just black screen and no X:(. Maybe it needs some option in xorg.conf, but i don't know which:confused:? Rolling back to 6.8.0 for now;).

[lucky_]

Does it induces any significant performance changes, for non composited desktop ?
I guess that before this commit it was the role of the cpu to do this job.....

Anyway thumbs up and have a good time at FOSDEM, don't drink too much beer :p, it's bad for the driver quality assurance...

[bitnick]

Does it induces any significant performance changes, for non composited desktop ?
I guess that before this commit it was the role of the cpu to do this job.....

Anyway thumbs up and have a good time at FOSDEM, don't drink too much beer :p, it's bad for the driver quality assurance...


There seems to be no difference in non-composited video playback between the recently checked in code and xf86-video-ati-6.6.3. (Tested with mplayer, video out xv, ~28% CPU usage for a 720p movie with both driver versions).

------
Edit:

I can find no differences with compiz either, neither in performace nor function. Tested with mplayer -vo xv, x11, gl, gl2:

xv: Works well, but the picture doesn't follow when the window is moved around, nor when using cube rotation.

x11: Works, but won't scale at all. Cube rotation and wrap-around-corner works.

gl: very slow.

gl2: picture totally garbled, very slow.

This is exactly the same results as with xf86-video-ati-6.6.3. My install procedure:
git-clone git://anongit.freedesktop.org/git/xorg/driver/xf86-video-ati
cd xf86-video-ati
./autogen.sh --prefix=/usr
make install
/etc/init.d/xdm restart

Did I miss something? How, exactly, should I be able to notice the improvement? Xv Texture support sounds cool; I'd like to get it to work. :)

--
Arvid

[ioannis]

This adds support for Xv adapters using the texture engine for display and colorspace conversion. Since the textures draw directly to the framebuffer, they work with composited desktops and rotation. So yes, using the textured adapter will work properly with compiz. We still expose the overlay adapter as well however, so make sure you use the right one when running compiz.

Lovely! :-D

I'll give it a try

[agd5f]

There seems to be no difference in non-composited video playback between the recently checked in code and xf86-video-ati-6.6.3. (Tested with mplayer, video out xv, ~28% CPU usage for a 720p movie with both driver versions).

xv: Works well, but the picture doesn't follow when the window is moved around, nor when using cube rotation.


Did I miss something? How, exactly, should I be able to notice the improvement? Xv Texture support sounds cool; I'd like to get it to work. :)

--
Arvid

The overlay adapter is still exposed as the first adapter so you will have to manually specify the textured adapter. Use xvinfo to lookup the port numbers and then I think you can use the port= option in mplayer to specify the Xv port to use.

[whatever]

*sorry for double posting, see below*

[whatever]

this is how texturedvideo looks on my radeon 9700:
http://www.250kb.de/u/080224/p/t/57e6c942.png (http://www.250kb.de/u/080224/p/57e6c942.png)

how to get rid of the checkboard pattern?

[bitnick]

Ok, here's the results from the Swedish jury ;):

xvinfo gives:
Adaptor #1: "Radeon Textured Video"
...
port base: 74

for both Metacity and Compiz. So I ran mplayer with -vo xv:port=74

Metacity:

Works, as long as the window is not partially covered. Whatever it is that renders the video seems to believe the real window size is equal to the visible window size, rounded down to the nearest 8? 16? pixels, and wraps around each scan line too early which garbles the picture completely.

Roughly twice the cpu usage compared to "ordinary" overlay. There's also an annoying 1px high blueish, broken line along the top of the window, and a greenish line along the bottom to the right.

Compiz:

No picture, just a bunch of "X11 error: BadAlloc (insufficient resources for operation)" messages.

[ioannis]

Lovely! :-D

I'll give it a try

I've gave a try, but I get an error during autoconfig
./configure: line 20296: syntax error near unexpected token `XINERAMA,'

tried in ubuntu 7.10 and Fedora 8. I have libXinerama-dev package installed. Where is the XINERAMA macro defined ?

[agd5f]

Ok, here's the results from the Swedish jury ;):

xvinfo gives:
Adaptor #1: "Radeon Textured Video"
...
port base: 74

for both Metacity and Compiz. So I ran mplayer with -vo xv:port=74

Metacity:

Works, as long as the window is not partially covered. Whatever it is that renders the video seems to believe the real window size is equal to the visible window size, rounded down to the nearest 8? 16? pixels, and wraps around each scan line too early which garbles the picture completely.

Roughly twice the cpu usage compared to "ordinary" overlay. There's also an annoying 1px high blueish, broken line along the top of the window, and a greenish line along the bottom to the right.


As noted in the commit message, there are currently some clipping issues with r300 that I haven't sorted out yet, hence the corruption.


Compiz:

No picture, just a bunch of "X11 error: BadAlloc (insufficient resources for operation)" messages.

Switch to EXA. XAA does not provide a mechanism to migrate pixmaps from system ram to vram.

[agd5f]

I've gave a try, but I get an error during autoconfig
./configure: line 20296: syntax error near unexpected token `XINERAMA,'

tried in ubuntu 7.10 and Fedora 8. I have libXinerama-dev package installed. Where is the XINERAMA macro defined ?

I think you need the xserver macros. I'm not sure what the package is called on fedora though.

[Shinigami2057]

I think you need the xserver macros. I'm not sure what the package is called on fedora though.

I believe that it's xorg-x11-server-sdk.

[ioannis]

I think you need the xserver macros. I'm not sure what the package is called on fedora though.

Thanks. You were right.

The package is called 'xorg-x11-server-sdk' on Fedora.

Using EXA acceleration works with compiz as advertised, but there is a pink hue on the video. Some colour conversion error? The video plays fine with video overlay (though with the usual problem of not being mapped on the window)

There isn't any measurable performance drop as others have mentioned as well. No difference between compiz and metacity. Both exhibit the same pink hue.

[bitnick]

As noted in the commit message, there are currently some clipping issues with r300 that I haven't sorted out yet, hence the corruption.

Ok, I'm not used to git and haven't figured out how to read commit messages yet. Sorry for the noise.


Switch to EXA. XAA does not provide a mechanism to migrate pixmaps from system ram to vram.

I'd love to try it out, but unfortunately, for some reason, with AccelMethod=EXA my desktop freezes over (no keyboard or screen updates) when I try to start gnome-terminal, which I usually use to start compiz. (Everything else seems to work fine. Weird.) At the moment I do not have another computer nearby to log in with to check what's going on. Ctrl-Alt-Fx doesn't work, not even SysRq.

I've never really gotten EXA to work; 2D updates like resizing a window is unbearably slow, like 2-3 draws per second. (This is on a R420.) Perhaps we should let others test this, and I'll get back with results if I get EXA going.

[koolmanoncampus]

Nice work... Dramatically less CPU usage when playing videos.

In related features anyone know if/when xvMC (or vaapi if that is the new thing) will make its way into the radeon driver?

[DocentDAMP]

I checked out xf86-video-ati from git yesterday to try xv on my RS690 equipped HTPC, but xvinfo only returned "no adaptors present".

Is xv supposed to work on RS960 chips? Do I need to add something to the configuration file?

[agd5f]

I checked out xf86-video-ati from git yesterday to try xv on my RS690 equipped HTPC, but xvinfo only returned "no adaptors present".

Is xv supposed to work on RS960 chips? Do I need to add something to the configuration file?

It should work. Are you sure you installed it properly? The drivers install to /usr/local unless you override when you run configure:

./autogen.sh --prefix=/usr

Make sure you back the current driver in case something goes wrong.

[xpgdk]

On my RS690 (Asus M2A-VM) I get nothing but strangely coloured triangles.
Could this be due to lack of proper DRI drivers?

/Paul

[DocentDAMP]

It should work. Are you sure you installed it properly? The drivers install to /usr/local unless you override when you run configure:

./autogen.sh --prefix=/usr

Make sure you back the current driver in case something goes wrong.

Thanks! I probably did something wrong with the updating of the git code because I checked out a clean copy, compiled it and now I get some decent output from xvinfo.

However, when I run "mplayer -vo xv any_film.avi" I get nothing but flickering red and green fields/triangles. This is probably the same problem xpgdk has.

[givemesugarr]

texturedvideo seems to be ignored on my x200m. the xv overlay worked also before, but rendering still results not present. since rs690 should be about the same i was thinking that rs480 would also support texturedvideo.
or do i have to compile something else beside xf86-video-ati?!

[agd5f]

texturedvideo seems to be ignored on my x200m. the xv overlay worked also before, but rendering still results not present. since rs690 should be about the same i was thinking that rs480 would also support texturedvideo.
or do i have to compile something else beside xf86-video-ati?!

It's not on by default on rs480 cards, but it should work, it's just that no one has tested it. You could go in and edit the code to enable it for rs480. Let me know and I can provide a patch.

[pedepy]

This adds support for Xv adapters using the texture engine for display and colorspace conversion. Since the textures draw directly to the framebuffer, they work with composited desktops and rotation. So yes, using the textured adapter will work properly with compiz. We still expose the overlay adapter as well however, so make sure you use the right one when running compiz.


I'm not sure I follow you on this: you say it draws directly to framebuffer, but then say it should get redirected properly by compiz.. I don't think that makes sense :confused:

[agd5f]

I'm not sure I follow you on this: you say it draws directly to framebuffer, but then say it should get redirected properly by compiz.. I don't think that makes sense :confused:

when you use the overlay, the video data is written to a separate overlay buffer and a colorkey is drawn to the part of the screen where the video would be. When the crtc scans out the frame, if the overlay is enabled and sourced to that crtc, the area with the colorkey is replaced by the video data when the pixel stream is sent to the output. When you use textured video, the texture engine converts the YUV data to RGB and writes it to a buffer: either the scanout buffer itself or a pixmap for composite. Since the data is actually there in a buffer it can be blended, etc. This doesn't work with the overlay because there is no data in the buffer, just colorkey; the streams are mixed during scanout.

[givemesugarr]

It's not on by default on rs480 cards, but it should work, it's just that no one has tested it. You could go in and edit the code to enable it for rs480. Let me know and I can provide a patch.

if that could help you devs to get it working i'd be glad to try it out and see if it works.

[whatever]

latest commit made the checkboard pattern on radeon 9700 disappear, texturedvideo is useable now :)
R300/R500: set the number graphics pipes properly

This should fix the checkerboard issues on r300/r350 cards.

btw: this also fixed exa-accelerated rotation.

[agd5f]

if that could help you devs to get it working i'd be glad to try it out and see if it works.

This patch enables exa render accel (for rotation) and textured video for XPRESS cards:

http://www.botchco.com/alex/xorg/xpress_enable_tvideo_and_render.diff

Let me know if it works.

[xpgdk]

On my RS690 (Asus M2A-VM) I get nothing but strangely coloured triangles.
Could this be due to lack of proper DRI drivers?


I have now upgraded to latest git of mesa, drm and xf86-video-ati and xvideo finally works. There is some tearing and the picture is wobbling -- not much, but enough for a headache.

Really nice to see the progress. Looking forward to be able to use my X1250 IGP :-)

/Paul

[givemesugarr]

This patch enables exa render accel (for rotation) and textured video for XPRESS cards:

http://www.botchco.com/alex/xorg/xpress_enable_tvideo_and_render.diff

Let me know if it works.

hmmm.... i was able to enable rendering and exa compositing but texturedvideo is still ignored on my board. i'll try to recompile the driver, drm and mesa and see if something changes. for now the first impression is that it works quite well. i'll try out also compiz and see what happens.
randr is still disabled.

[agd5f]

hmmm.... i was able to enable rendering and exa compositing but texturedvideo is still ignored on my board. i'll try to recompile the driver, drm and mesa and see if something changes. for now the first impression is that it works quite well. i'll try out also compiz and see what happens.
randr is still disabled.

You don't need to rebuild mesa or the drm, just the radeon ddx. make sure you apply the patch and install the driver in the right place.

[givemesugarr]

You don't need to rebuild mesa or the drm, just the radeon ddx. make sure you apply the patch and install the driver in the right place.

sorry, i had to remove the previous radeon module and then reload drm and radeon to have textured video enabled.

here's my log:

http://pastebin.com/m17dd4ae5

it seems that everything is working well. i haven't got any problem with the exception of aiglx:

#
(EE) AIGLX error: dlsym for __driCreateNewScreen_20050727 failed (/usr/lib64/dri/r300_dri.so: undefined symbol: __driCreateNewScreen_20050727)
#
(EE) AIGLX: reverting to software rendering

i'll try to see what happens with compiz as soon as i compile it (i'm on gentoo). it seems that i have xvmc extension loaded but as xvmc hasn't been implemented it's not used.
i'll try out some video and see what happens. i seemed to had divx5 video accel previous to this release. in the next days i'll let you know how i feel with rendering and textured video.
exa is as usual faulty (has the same artifacts that fglrx had without xaanoffscreenpixmaps set to true), but somehow faster that last week's version.
opengl apps don't work (hung on or don't even start).
if that's the case i don't think that compiz could run, but i'll give it a try, nevertheless.
i'll give you more info on request or some user impressions in the next days with different configurations.

[bridgman]

That divx5 acceleration of yours is still a mystery ;)

[sid350]

is any there .deb packages for xpress200m and x1600xt? does tvtime work with this drivers?

[sid350]

ok, i've compiled it from the git. It works fine on xpress200m, even XV support whoohoo, but without DRI option enabled. Once I enabled it - screen freezes. But I can't see any errors in Xorg.0.log - it looks like system working stable. Log writes than DRI enabled, acceleration enabled etc. But I see only corrupted wallpaper, no keyboard shortcuts work. Does it mean that on x200m direct rendering shouldn't work now? Or it is my problem?

[givemesugarr]

That divx5 acceleration of yours is still a mystery ;)

it's still a mistery to me too since it shouldn't be there.... but as i've said if on fglrx the same video in the same player with the same settings bumps the processor to his highest speed with big cpu usage while on radeon it's at less than 20% at the lowest speed value then the only explanation is that something's going at the hw level.... maybe my board is an alien made one that has something strange in it.... :D:D

ok, i've compiled it from the git. It works fine on xpress200m, even XV support whoohoo, but without DRI option enabled. Once I enabled it - screen freezes. But I can't see any errors in Xorg.0.log - it looks like system working stable. Log writes than DRI enabled, acceleration enabled etc. But I see only corrupted wallpaper, no keyboard shortcuts work. Does it mean that on x200m direct rendering shouldn't work now? Or it is my problem?

dri works, but on xv overlay you cannot use any opengl app. also you'd have to turn the mtrr to on, while on fglrx you have to have it on off. i've also seen that i have depth moves and page flipping enabled with radeon.

[whatever]

in the current driver version the image corruption when moving a texturedvideo window partially off-screen (without composite) is fixed when moving the window out at the right or bottom border.
but when moving the window out at the left border the right part of the video is not redrawn and you cannot move the video over the top border (you can move the window, but the video itself does not move higher).

somehow these changes also broke overscan as used by tvtime. i remember that this was also broken in fglrx-overlay for a long time, but they fixed it eventually...

[sid350]

dri works, but on xv overlay you cannot use any opengl app. also you'd have to turn the mtrr to on, while on fglrx you have to have it on off. i've also seen that i have depth moves and page flipping enabled with radeon.

unfortunately it didn't work. Only still picture with corrupted wallpaper after Option "DRI" "true" :( I can type login and password - Gnome loads, i hear sounds, but no video output.

Can it be because of my Samsung R40 laptop, which has no separate videomemory? Videocard can only use shared memory

//sorry for my english

[greg]

This patch enables exa render accel (for rotation) and textured video for XPRESS cards:

http://www.botchco.com/alex/xorg/xpress_enable_tvideo_and_render.diff

Let me know if it works.

This works for me. I get a notice about Render and textured video being enabled in Xorg.0.log and I can use it by forcing Xv port 74. However, the textured video ports are not listed in the xvinfo output and the scaling of textured video doesn't use any filtering (only nearest neighbour). Otherwise it works fine, also in conjunction with compositing.

I tested accelerated rotation and it worked fine, too. I wonder why the acceleration is still disabled by default.

EDIT: chipset is XPRESS 200M 5955.

[agd5f]

This works for me. I get a notice about Render and textured video being enabled in Xorg.0.log and I can use it by forcing Xv port 74. However, the textured video ports are not listed in the xvinfo output and the scaling of textured video doesn't use any filtering (only nearest neighbour). Otherwise it works fine, also in conjunction with compositing.

I tested accelerated rotation and it worked fine, too. I wonder why the acceleration is still disabled by default.

EDIT: chipset is XPRESS 200M 5955.

Thanks for confirming. I've gone ahead and committed the patch.

[chrisr]

I have just tested the new Radeon driver from git, and the Textured Video adapter for R100 seems to be working OK. Using mplayer -nosound -benchmark, I get the following information for Textured and Overlay adapters:

Textured:
BENCHMARKs: VC: 71.091s VO: 20.266s A: 0.000s Sys: 4.363s = 95.720s
BENCHMARK%: VC: 74.2695% VO: 21.1721% A: 0.0000% Sys: 4.5585% = 100.0000%

Overlay:
BENCHMARKs: VC: 70.628s VO: 12.017s A: 0.000s Sys: 3.725s = 86.369s
BENCHMARK%: VC: 81.7738% VO: 13.9137% A: 0.0000% Sys: 4.3125% = 100.0000%

So assuming that I am interpreting this correctly, the Overlay adapter uses 86.4 CPU seconds, whereas the Textured adapter uses 95.7 CPU seconds to play the same video file.

[NaterGator]

Here is the same for me:
Textured:
BENCHMARKs: VC: 99.624s VO: 0.901s A: 0.000s Sys: 6.640s = 107.166s
BENCHMARK%: VC: 92.9629% VO: 0.8409% A: 0.0000% Sys: 6.1962% = 100.0000%


Overlay:
BENCHMARKs: VC: 97.977s VO: 0.993s A: 0.000s Sys: 7.731s = 106.702s
BENCHMARK%: VC: 91.8236% VO: 0.9305% A: 0.0000% Sys: 7.2458% = 100.0000%


My gripe is that EXA is unusably slow for some reason. It just pegs my CPU usage for no good reason (scrolling in xterm, for example) where as XAA is fine. I get ~5900fps in glxgears with EXA and ~6200fps with XAA.


It's good to see Xv texture support coming, but at this point EXA is so much less mature compared to XAA that it makes everything else not worth it :(

[agd5f]

It's good to see Xv texture support coming, but at this point EXA is so much less mature compared to XAA that it makes everything else not worth it :(

Textured video works with XAA or EXA, however due to the limitations of XAA, there are limitations when using compositors with XAA and textured video.

[NaterGator]

Switch to EXA. XAA does not provide a mechanism to migrate pixmaps from system ram to vram.LoL sorry, I bet you get sick of repeating yourself. :p

I'd love to try it out, but unfortunately, for some reason, with AccelMethod=EXA my desktop freezes over (no keyboard or screen updates) when I try to start gnome-terminal, which I usually use to start compiz. (Everything else seems to work fine. Weird.) At the moment I do not have another computer nearby to log in with to check what's going on. Ctrl-Alt-Fx doesn't work, not even SysRq.

I've never really gotten EXA to work; 2D updates like resizing a window is unbearably slow, like 2-3 draws per second. (This is on a R420.) Perhaps we should let others test this, and I'll get back with results if I get EXA going.I have the exact same experience using the R420 (Radeon x800 xt pe). Random desktop lockups (not even reisub can bring it back) and horribly slow performance doing some 2D stuff.

Maybe this is just a R420 issue?

[NaterGator]

I can finally report success! After going through some of the source and bugzilla I came across the manpage for exa (duh!) and realized I was forgetting to set AccelDFS to on! After adding that to my device section Xv acceleration is properly compositing, and outputing planet earth in 720p resulted in 4% CPU utilization even when dragging around the wobbly window.

Things seem to be more stable than EXA with AccelDFS off as well. The only problem I'm yet to see (in about 30 minutes of use) is that emerald is messing up the title bar of my windows.

I'm going to see if I can do any patching of mplayer to autodetected the TexturedVideo pseudo adapter and automatically output to it.

Thanks again devs, this is a good step forward!

[karthikrg]

I tried downloading the latest git code via git-clone, but as I am behind a firewall, its not working and is timing out. Any way to get it as a http link <maybe some kind soul can host it as a tarball somewhere?> or can point to me to some alternate way to download the latest git snapshot via http?

I am fed up of video tearing in fglrx and wanna try radeon on my x1800 (R520).

Thanks :)

[NaterGator]

This includes git clones from about 10 minutes ago of the mesa/drm and video-ati projects.

http://www.nate-online.com/drm_and_radeon.tar.gz

-----Nate

[karthikrg]

hey thanx. i tried.. link doesnt seem 2 be working. can u please check it?

[NaterGator]

I just downloaded it... hmm.

Did you try:
wget http://www.nate-online.com/drm_and_radeon.tar.gz

[bitnick]

*snip*
I have the exact same experience using the R420 (Radeon x800 xt pe). Random desktop lockups (not even reisub can bring it back) and horribly slow performance doing some 2D stuff.

Maybe this is just a R420 issue?

I seem to have got rid of the lockups by setting AGPMode to 4 (instead of 8). (I've always had AGPFastWrite disabled.)

*snip*
I can finally report success! After going through some of the source and bugzilla I came across the manpage for exa (duh!) and realized I was forgetting to set AccelDFS to on! After adding that to my device section Xv acceleration is properly compositing, and outputing planet earth in 720p resulted in 4% CPU utilization even when dragging around the wobbly window.

I'll try the AccelDFS option (*sigh* Why have a separate document for EXA when XAA options are described in the radeon man page? Not exactly intuitive ;)). It would be cool to get this to work.

[karthikrg]

yaa got it.. thanx a LOTT!! :)

[siggma]

A bold comment?

Um, is there an ATI representative listening.

I own an ATI X1650 256 DDR2 PCIe card that performs flawlessly under Vista, but I despise VISTA, and Microsoft, so it's Linux.
BUT! I still cannot play video using the hardware and run compiz at the same time in any linux distribution using this card. It's been on the market for over a year now. What's UP?

Let's sum up the issue, from a user point of view. How you see it may be different but users are the one's spending the money.
Radeon
RadeonHD
ATI
FGLRX
AVIVOWhy do we need 5 different drivers for the same cards?

Why not just spend an afternoon and make ONE work like it should? We know the code is written and that it works because I can change a single word in a configuration file and REBOOT, and video plays.

As a loyal ATI customer, it's unrealistic to keep telling me why it won't work. You people make the cards and you CAN write a driver that WILL drive it under any operating system. You've proven you are capable. Is "intellectual property" so important you'll sacrifice your potential customer base?

Are we human beings living together on a planet as a society of individuals, or are we intellectual enemies in a no-holds-barred fight to the death?

IMHO, Vista was a massive error on the part of microsoft. From this day forward Microsoft goes downhill. It may take a while but it's going to happen. Historically all dishonest corporations self destruct, eventually. and microsoft is no exception.

When money becomes an end, the end of money is near... and revolutions borne of time and disgust leap forth in adoration of future freedoms.

-Tom

[jchapman]

First of all I want to say that I do appreciate all the effort of the people doing the open source drivers, and I also do appreciate that AMD is trying to support linux. But I have to agree with the general thrust of the previous message. I just put together a home theater system based on an x2 5000+ and an ATI 690 based motherboard.

I have tried running both the open source and ATI drivers under Gutsy and neither presents anything like satisfactory results. Really bad video tearing, horrible pixelation when increasing window size for video and the whole screen flashing pink/green for a variable amount of time whenever I start a video. And even though my monitor identifies the signal as 1080P it only fills about 90% of the actual screen so text etc. looks horrible as the 1920x1080 image is scaled down into a display grid of about 1800x1000 physical pixels.

Not to mention the difficulty switching back and forth between trying out the open and proprietary drivers (no novice is going to get more than 5 seconds into that process before giving up.

So I switched to an old copy of Win2KSP4 I had. Video tearing gone. Pink/green flashing gone. Smooth interpolation when maxing the video window. Yayyyy!

But things are not perfect on the Windows side either. The problem of a 1080P image only filling 90% of the monitor still happens, and has been reported by more than a few others on the net. Plus there don't seem to be Catalyst drivers for Win2K for the 690/X1250 so I have no way of adjusting anything, in particular the antialiasing or making the display size fit the physical screen (I know, PowerStrip... that's my next try).

This is exactly the sort of experience that is going to drive users away from Linux and to windows. And given that the 90% screen reduction on windows I'm not all that happy with AMD/ATI on windows either - I mean the 690 has been out for a long time now....

I wanted to switch all our machines to the same OS and for that to be Linux but that isn't possible so after a few days time wasted I'm more or less forced back to Windows and am dubious about buying another AMD system/card for graphics since it doesn't seem to really work right even in Windows.

Anyway that's my rant...

A bold comment?

Um, is there an ATI representative listening.

I own an ATI X1650 256 DDR2 PCIe card that performs flawlessly under Vista, but I despise VISTA, and Microsoft, so it's Linux.
BUT! I still cannot play video using the hardware and run compiz at the same time in any linux distribution using this card. It's been on the market for over a year now. What's UP?

Let's sum up the issue, from a user point of view. How you see it may be different but users are the one's spending the money.
Radeon
RadeonHD
ATI
FGLRX
AVIVOWhy do we need 5 different drivers for the same cards?

Why not just spend an afternoon and make ONE work like it should? We know the code is written and that it works because I can change a single word in a configuration file and REBOOT, and video plays.

As a loyal ATI customer, it's unrealistic to keep telling me why it won't work. You people make the cards and you CAN write a driver that WILL drive it under any operating system. You've proven you are capable. Is "intellectual property" so important you'll sacrifice your potential customer base?

Are we human beings living together on a planet as a society of individuals, or are we intellectual enemies in a no-holds-barred fight to the death?

IMHO, Vista was a massive error on the part of microsoft. From this day forward Microsoft goes downhill. It may take a while but it's going to happen. Historically all dishonest corporations self destruct, eventually. and microsoft is no exception.

When money becomes an end, the end of money is near... and revolutions borne of time and disgust leap forth in adoration of future freedoms.

-Tom

[Melcar]

A bold comment?

Um, is there an ATI representative listening.

I own an ATI X1650 256 DDR2 PCIe card that performs flawlessly under Vista, but I despise VISTA, and Microsoft, so it's Linux.
BUT! I still cannot play video using the hardware and run compiz at the same time in any linux distribution using this card. It's been on the market for over a year now. What's UP?

Let's sum up the issue, from a user point of view. How you see it may be different but users are the one's spending the money.
Radeon
RadeonHD
ATI
FGLRX
AVIVOWhy do we need 5 different drivers for the same cards?

Why not just spend an afternoon and make ONE work like it should? We know the code is written and that it works because I can change a single word in a configuration file and REBOOT, and video plays.

As a loyal ATI customer, it's unrealistic to keep telling me why it won't work. You people make the cards and you CAN write a driver that WILL drive it under any operating system. You've proven you are capable. Is "intellectual property" so important you'll sacrifice your potential customer base?

Are we human beings living together on a planet as a society of individuals, or are we intellectual enemies in a no-holds-barred fight to the death?

IMHO, Vista was a massive error on the part of microsoft. From this day forward Microsoft goes downhill. It may take a while but it's going to happen. Historically all dishonest corporations self destruct, eventually. and microsoft is no exception.

When money becomes an end, the end of money is near... and revolutions borne of time and disgust leap forth in adoration of future freedoms.

-Tom



In the Linux world hardly anything is as consolidated as in the Windows world. Personally, I like the fact that we have the option of using several different drivers for our cards.; it gives us (the end user) more choice.

As for not being able to play videos while Compiz is on, make sure you are using plain ol' X to render the videos. It has already been explained that the xserver has a severe limitation when it comes to the whole direct/indirect rendering thing when you throw accelerated desktops in the mix... it's not a driver limitation.

[siggma]

In the Linux world hardly anything is as consolidated as in the Windows world. Personally, I like the fact that we have the option of using several different drivers for our cards.; it gives us (the end user) more choice.

As for not being able to play videos while Compiz is on, make sure you are using plain ol' X to render the videos. It has already been explained that the xserver has a severe limitation when it comes to the whole direct/indirect rendering thing when you throw accelerated desktops in the mix... it's not a driver limitation.

Perhaps, but the only other option is vidix, which doesn't work for r500 chipsets either leaving owners of these cards SOL unless or until someone like the manufacturer steps up and finds a way to make it work. There are plenty of ways this can work without blaming X or rewriting the entire OS. If all we focus on the problem it will never be solved. If we focus on a solution, it will come, quickly.

I'm willing to bet there are people at AMD/ATI with Linux on their desktops that playback video just fine. If I worked there and had an in depth understanding of the hardware, I'd have figured out a way to playback video without all that horrible tearing and crashing. This is not a hardware issue or an Xserver issue, it's a political issue.

Corporations used to fight their battles directly, face to face, with each other. Now they THINK they can offload that battle to the desktop of the user and point the finger. It's an opportunity to make a choice. Honesty choices lead to cooperation and working products. Dishonest choices and unreasonable expectations do not. Microsoft began the practice by encouraging a "cottage" industry of virus scanners and spyware stoppers with it's "push this button to install software" philosophy. That single button solution has now spread to other parts of society and become the very choice referenced above.

Bottom line is, we still don't have a usable solution. We keep hearing explanations but explanations don't play DVD's or my favourite episode of Star Trek.

Please ATI, just put all the code in ONE driver and release to the public on your web site so we can use the hardware we already paid for.

-Tom

[jchapman]

In the Linux world hardly anything is as consolidated as in the Windows world. Personally, I like the fact that we have the option of using several different drivers for our cards.; it gives us (the end user) more choice.



Choice is great. It can also be counter-productive. What use is choice when your choice is between several alternatives, none of which entirely work? Sorry but I'd rather have one solution that actually worked than several that don't. This isn't an issue just with the driver but with open source in general... there are many cases where the user would be much better off if competing camps just dialed down the egos a bit and cooperated on a more limited number of alternatives. What is the point behind developing software if not to give users something that they actually want to use?

I'm all for open source, choice, and all those great things... right up to the point where it means I can't do what I need to do. Not to rant too much here but the whole choice thing carries with it the same whiff I get when someone complains about linux being difficult and is told how they should have known how to do this or that, or that it isn't the fault of linux but manufacturers, which while possibly technically true is a meaningless answer to the user, or whatever... this is a huge problem with linux acceptance.
</p>

[bridgman]

I own an ATI X1650 256 DDR2 PCIe card that performs flawlessly under Vista, but I despise VISTA, and Microsoft, so it's Linux.
BUT! I still cannot play video using the hardware and run compiz at the same time in any linux distribution using this card. It's been on the market for over a year now. What's UP?

As others have posted, you need two things to get video and compiz working together -- the video needs to be generated in a form that a compositor can digest (Xv + TexturedVideo), and the compositor needs to be told to stop trying to draw a desktop in the same space that the video is playing. The second part is the challenge -- Vista does that nicely but I think everyone is still learning all the tricks to make Compiz and KDE play nice and not write over the video.

Let's sum up the issue, from a user point of view. How you see it may be different but users are the one's spending the money.
Radeon
RadeonHD
ATI
FGLRX
AVIVOWhy do we need 5 different drivers for the same cards?

The "ati" driver is just a wrapper around radeon and other drivers. Avivo was created via reverse engineering and was retired when we started supplying 5xx specs, although some of the Avivo code was used in radeon (and possibly radeonhd), so there are really three drivers today -- one proprietary driver developed by ati (fglrx) and two open source drivers(radeon, radeonhd).

The latest radeon code (from git) is worth trying since Alex recently added textured video support and Dave extended that to include R5xx and RS690. The radeonhd driver is mostly modesetting today but will start to pick up the same acceleration code in the next little while.

Why not just spend an afternoon and make ONE work like it should? We know the code is written and that it works because I can change a single word in a configuration file and REBOOT, and video plays.

I assume you're talking about changing OSes here ? If so, the main issue is that we need to either figure out settings that will keep Compiz from stepping on the video window or make changes somewhere in the plumbing so that this happens automatically. It's not really a driver issue AFAIK but it is something we will probably have to work on anyways.

[siggma]

As others have posted, you need two things to get video and compiz working together -- the video needs to be generated in a form that a compositor can digest (Xv + TexturedVideo), and the compositor needs to be told to stop trying to draw a desktop in the same space that the video is playing. The second part is the challenge -- Vista does that nicely but I think everyone is still learning all the tricks to make Compiz and KDE play nice and not write over the video.

Thank you. That helps me understand what's happening.

The "ati" driver is just a wrapper around radeon and other drivers.
--snip--
I assume you're talking about changing OSes here ? Yes, I'm running Ubuntu Gnome at the moment but would like to be able to run Xfce and KDE as well.

If so, the main issue is that we need to either figure out settings that will keep Compiz from stepping on the video window or make changes somewhere in the plumbing so that this happens automatically. It's not really a driver issue AFAIK but it is something we will probably have to work on anyways.Thank you for responding.
I can't seem to find the radeon driver using the same git url as the ati and radeonhd. Id' love to try it. If you, or someone... can post the RADEON git url I'll compile up a copy and see how it works. A url that I think should have been in the announcement article on Feb 24th, or at least a link to a page containing a brief "here's how to try the driver" (hint, hint).

This is the URL I already downloaded and compiled:
git://anongit.freedesktop.org/git/xorg/driver/xf86-video-radeonhd
and there is no source at:
git://anongit.freedesktop.org/git/xorg/driver/xf86-video-radeon

So, unless the freedesktop radeon driver is named radeonHD (Huh?)...and someone is deliberately trying to confuse us all...

-Tom

[bridgman]

Sorry, the "radeon" driver is in the xf86-video-ati git tree. The "ati" driver is just a wrapper around radeon, rage128, and one or two other drivers.

[bulyst]

But things are not perfect on the Windows side either. The problem of a 1080P image only filling 90% of the monitor still happens, and has been reported by more than a few others on the net. Plus there don't seem to be Catalyst drivers for Win2K for the 690/X1250 so I have no way of adjusting anything, in particular the antialiasing or making the display size fit the physical screen (I know, PowerStrip... that's my next try).


I have experienced the same issue of image only filling 90% of my LCDTV in either windows XP Catalyst Driver or Linux Catalyst driver. (Radeonhd and ati opensource fill up the whole screen.) However, once I changed the refresh rate from default 60Hz to 50Hz, the image fills up the whole screen. It worked in both Windows and Linux Catalyst driver.
I noted that the problem only happened when connecting to my sharp LCDTV via HDMI cable.

Hope this help.

-Bruce

[mal13]

-siggma-

Yes, I'm running Ubuntu Gnome at the moment but would like to be able to run Xfce and KDE as well.


In terminal:
Kde= sudo apt-get install kubuntu-desktop
Xfce= sudo apt-get install xubuntu-desktop
or get them from aptitude.

[siggma]

In terminal:
Kde= sudo apt-get install kubuntu-desktop
Xfce= sudo apt-get install xubuntu-desktop

Eww, thanks. I was hoping there was a meta install for them.

[siggma]

Sorry, the "radeon" driver is in the xf86-video-ati git tree. The "ati" driver is just a wrapper around radeon, rage128, and one or two other drivers.

I'll double check but If I recall that driver has no composite support yet, I'm watching Voyager in Vugly Vista at the moment :(. But you knew that. Which brings me back to the original beef.
Are their several different branches at ATI working on Linux dirvers?
Is there a specific reason we have composite support in fglrx but not in the OSS "radeon" driver?
If composite overlapping output is the issue, why not focus on one driver or the other first, then move support to the OSS or SIDEBAND :) driver.
We, including me, all have suggestions, wants, desires, projections, and expectations; it's very easy to be an arms-reach critic but no so easy to try to satisfy the demands of others.Personally I'd rather see the support show up in the fglrx driver first since it's already got reasonable, albeit choppy, AIGLX compositing support and it includes additional tools for building and verifying the driver and it's supporting products. In my mind it also makes the most sense. But I'm not the developer and I don't see what the developer sees. So again, thanks for taking the time to personally respond. Your few words are both reassuring and useful.
-Tom

[bridgman]

I believe that the "radeon" driver has had composite support for a year or so.

[agd5f]

Which brings me back to the original beef.
Are their several different branches at ATI working on Linux dirvers?
Is there a specific reason we have composite support in fglrx but not in the OSS "radeon" driver?
If composite overlapping output is the issue, why not focus on one driver or the other first, then move support to the OSS or SIDEBAND :) driver.
We, including me, all have suggestions, wants, desires, projections, and expectations; it's very easy to be an arms-reach critic but no so easy to try to satisfy the demands of others.

When you say composite I assume you mean support for compiz. The composite extension is available on any recent X server and it can even be accelerated via EXA render acceleration. r1xx-r2xx radeons have full EXA render support. r3xx-r5xx only support limited operations (enough for rotation) right now. Compiz requires OpenGL (via AIGLX). Since there is no open 3D driver yet for r5xx+ hw, there is no way to accelerate it. Once we have a 3D driver, it (compiz) should just work.

[siggma]

When you say composite I assume you mean support for compiz. The composite extension is available on any recent X server and it can even be accelerated via EXA render acceleration. r1xx-r2xx radeons have full EXA render support. r3xx-r5xx only support limited operations (enough for rotation) right now. Compiz requires OpenGL (via AIGLX). Since there is no open 3D driver yet for r5xx+ hw, there is no way to accelerate it. Once we have a 3D driver, it (compiz) should just work.

My ignorance is showing...
Yes, I mean compiz but not just compiz., there are other uses for software and hardware compositing.

I wasn't aware 3d acceleration was missing in the fglrx driver. That explains the horizontal and vertical blocks on the edge of windows when I drag them around, which is only a minor annoyance at this point with the PCIe and a core duo. If that's how it looks without hw, it will be awesome with it.

Until then I'll just have to wait, I have no alternative but to bug people in various forums until I can understand my own ignorance. Then it's not ignorance anymore and I can wait in peace.

No thoughts on vidix?

-Tom

[NaterGator]

Tom,

I believe he was saying there is no r5xx+ acceleration in the opensource drivers (radeonhd). There should be 3d accel in fglrx.

[siggma]

I believe that the "radeon" driver has had composite support for a year or so.

Ok now that I'm using gnome again and the latest ati driver I can say I'm not able to use AIGLX or COMPOSITE with the git references radeon driver, I get the neat little fusion introduction and a white screen with a mouse cursor. I can also see a partial window preview when the mouse is at the bottom but no desktop or panels.

Is there a better list of xorg.conf options for this driver than the "Installing The RadeonHD Driver On Ubuntu" article written last september that specifically says to turn off compositing?

Here's my current xorg.conf as a link (http://www.trbailey.net/xorg.conf.copy)
How do I get the glx extension to load so I can test it?

If you'll point me in an appropriate direction I'll compile a TESTED how to and post it here and a few other key places so google picks it up later tonight.

-Tom

[d2kx]

TexturedVideo with ati/radeon makes my system reboot after 3-5 seconds of watching a video with a R580. Anyone else got this issue? I have it since I first tried the ati/radeon git when TexturedVideo came in and still have it with latest git.

Kernel 2.6.24.3
Xorg 7.3 using Xserver 1.4.1-git
KDE 3.5.9 with Kaffeine 0.8.6 using Xine-lib 1.1.10.1

[siggma]

Tom,

I believe he was saying there is no r5xx+ acceleration in the opensource drivers (radeonhd). There should be 3d accel in fglrx.

Actually, from what I understand the OSS driver is called "ati" from the git source. If you go back a few messages you'll find the link. The radeonhd is not the driver mentioned in the recent "OSS gets r500 support (http://www.phoronix.com/scan.php?page=news_item&px=NjM1Mg)" article, a misunderstanding I'm hoping will be corrected ASAP. A few messages back he stated that "ati" is a wrapper for the "radeon" driver and further states that it does have glx support via AIGLX, just not hardware accelerated 3D support, yet.

As for the fglrx, I don't think it has hardware acceleration for the r500 chipset either, yet, and my comparisons in Vista indicate that to be the case. The Chess game in Vista being a test case, my x1650 PCIe card displays that application smooth as glass, no tearing, no shuddering, not even the slightest hiccup, and rotation seems to be as fast as the keyboard can deliver keystrokes, to the limit of the display refresh (over 60 frames). I was playing HL2 yesterday @ 1280x960 and it's more than awesome fast, not that HL2 is a particularly demanding game but it's an anchor for performance comparisons.

In fact just two posts back he states there is no 3D acceleration in any driver for the r500 chipset.

Unless I've been missing something...
-Tom

[siggma]

TexturedVideo with ati/radeon makes my system reboot after 3-5 seconds of watching a video with a R580. Anyone else got this issue? I have it since I first tried the ati/radeon git when TexturedVideo came in and still have it with latest git.

Kernel 2.6.24.3
Xorg 7.3 using Xserver 1.4.1-git
KDE 3.5.9 with Kaffeine 0.8.6 using Xine-lib 1.1.10.1

First, can you post the git URL for:
Kernel 2.6.24.3 (debian, gentoo?)
Xorg 7.3 using Xserver 1.4.1-git

I'd love to upgrade to the 7.3 Xserver if it's stable.

Mine does something similar but it only plays about three frames. I suspect it's an xorg.conf option, which is the purpose of the previous message request and offer to write a how to, with as many options as I can find.
-tom

[bitnick]

Everybody seems to get different results with this (EXA, textured video, etc).

How about specifying the environment you developers are using so that others can easily test things out with the same settings?

I.e.
1) What software interacts with the the driver? libdrm, mesa, xorg-server, xorg-x11, the kernel, compiz, ...?
2) Which combinations of these have you tested? Which combinations are you using for development?
3) What xorg.conf driver settings are you using?
4) How do you start compiz?

I think this would both help minimize error reports that originates from other packages/dirt behind the wheel, and also (not least) make things clearer for testers.

Sure, when the driver is more mature, it needs to be tested with other versions, but I guess many people will use different versions of the other software none the less, so that might not be an issue.

Just an idea I got.

[d2kx]

First, can you post the git URL for:
Kernel 2.6.24.3 (debian, gentoo?)
Xorg 7.3 using Xserver 1.4.1-git

I'd love to upgrade to the 7.3 Xserver if it's stable.

Mine does something similar but it only plays about three frames. I suspect it's an xorg.conf option, which is the purpose of the previous message request and offer to write a how to, with as many options as I can find.
-tom

I use sidux, based on Debian/sid. Debian/sid has got Xserver 1.4.1-git in its repository. The kernel is from sidux, but the Debian/sid one is always upstream, too. Which distribution are you using? Sometimes the reboot appears without seeing the video (3 frames, maybe), sometimes after 3-5 seconds.

In fact just two posts back he states there is no 3D acceleration in any driver for the r500 chipset.

Unless I've been missing something...
-Tom

fglrx/Catalyst offers full 3D acceleration for R500+.

[siggma]

Everybody seems to get different results with this (EXA, textured video, etc).

How about specifying the environment you developers are using so that others can easily test things out with the same settings?

Maybe you can beat them to it and post a copy of your xorg.conf so maybe I can test it. I have a copy of mine on my server but not all of us have a web server at their disposal.

http://www.trbailey.net/xorg.conf.copy
-Tom

[bridgman]

The fglrx driver has accelerated 3D support for all of our chips from R300 up, including 5xx and 6xx.

Radeon (aka ati) has accelerated 3D support from R100 up to R4xx, plus RS690 which has a 4xx-derived 3D core, but not 5xx or 6xx. The 3D acceleration may still be a bit shaky on the RS480 and RS690.

RadeonHD supports 5xx and 6xx (ie the parts which do not yet have open source 3D acceleration) so does not have 3D acceleration on anything yet.

[siggma]

I use sidux, based on Debian/sid. Debian/sid has got Xserver 1.4.1-git in its repository. The kernel is from sidux, but the Debian/sid one is always upstream, too. Which distribution are you using? Sometimes the reboot appears without seeing the video (3 frames, maybe), sometimes after 3-5 seconds.

I'm using Ubuntu 2.6.22-14-generic i believe so a sidux kernel would probably be a no no for me.

I found xorg, right where it should be on x.org...

[d2kx]

siggma,

your xorg.conf is very ugly. You can't use VideoOverlay and TexturedVideo together, and enabling OpenGLOverlay, too, won't make things better. Also, Textured2D and TexturedXrender are experimental, while all of the options mentioned do not even work with the "ati" driver.

[siggma]

siggma,

your xorg.conf is very ugly. You can't use VideoOverlay and TexturedVideo together, and enabling OpenGLOverlay, too, won't make things better. Also, Textured2D and TexturedXrender are experimental, while all of the options mentioned do not even work with the "ati" driver.

As I figured.

The radeon driver plays video about the same as the fglrx. No big difference, in fact it looks exactly the same to me but I can't really test it, a common problem I presume.

Does the "ati" driver support compiz or not?

My experience is NOT, so unless someone has a configuration that works for an r500 chip card, I'll assume it's not working and that's my TEST report. Since I can't test the output with 3D enabled, I can't test it any more than that. So until we see a 3D composite video playback solution there is nothing more that I can say about this driver.

As for FGLRX:
Do I even need an overlay option in xorg.conf?
Can the driver just detect which options it and the hardware support and enable them as needed?

I've had good luck with only DRI, and the textured 2D and Xrender options enabled so unless I see a better list of options (realizing the driver is still in the works), I'll assume that's the best there is at the moment.

Thank you to all who have posted helpful replies here. I realize you all have other things to do besides chat all day with mua...
-Tom

[bridgman]

AFAIK the "ati" driver supports Compiz on pre-5xx parts but not 5xx/6xx. For your 5xx card, fglrx would be your best bet.

You want TexturedVideo turned on, VideoOverlay off and OpenGLOverlay turned off. Try to use aticonfig since the amdpcsdb entries will over-ride the xorg.conf entries if there is a collision.

[chrisr]

r3xx-r5xx only support limited operations (enough for rotation) right now.

Is there work in progress to improve render acceleration for R300-R500, please? I have enabled EXA with AccelDFS on my dual 2.66 GHz P4 Xeon machine with its 2 GB RAM and Radeon 9550 card, but it still feels more sluggish at times than an old 500 MHz P3 with 384 MB RAM and a Radeon 7000 (also EXA with AccelDFS). This is particularly true when using Firefox.

Something ain't right...

P.S. Both machines are using Fedora 8 with the >= 6.8.0 ATI driver. (The P4 machine is also getting updated with the git version of the driver occasionally.)

[chrisr]

You want TexturedVideo turned on, VideoOverlay off and OpenGLOverlay turned off.

For the xf86-ati driver at least, the CPU usage was higher in all cases with Textured Video than when using the Video Overlay. Is this because it's "early days" for Textured Video in the x86 driver? Or does Textured Video have other advantages beyond hardware acceleration?

[agd5f]

For the xf86-ati driver at least, the CPU usage was higher in all cases with Textured Video than when using the Video Overlay. Is this because it's "early days" for Textured Video in the x86 driver? Or does Textured Video have other advantages beyond hardware acceleration?

The overlay supports planar formats directly and it's colorspace conversion hw may be slightly faster than the texture engine's colorspace conversion hw. It would be interesting to compare a shader-based colorspace conversion routine with the one built into the texture engine. There is room for improvement. The advantage of textured video is that it writes directly to the framebuffer rather than getting overlaid during scan-out so it works with composite and rotation. It also supports multiple ports (more than one video active at once) and videos can span multiple monitors.

[bridgman]

The other big selling point for Textured Video is that the newer chips (R5xx, RS690, R6xx, RS780) don't have the same overlay so Textured Video is the way to go.

Chrisr, I don't know of anyone working on implementing EXA render at the moment but it's just a matter of someone finding the time. We have the 5xx 3D info out and are in final review for the 3xx/4xx 3D registers.

[Tillin9]

Just to add my report, textured video works on my Radeon 9700 Pro, and Radeon 9500, but doesn't work great. The major issues are a) The video quality is much worse when scaling. b) It appears to be slower, as in there is more latency involved.

My guess is the routine for scaling needs work. Overlay video is soft and smooth, while textured video is pixelated and overly sharp. Note, this only really applies for sources scaled at least 1.5x, which sadly is how I watch much of my video (tv tuner gives 720x480). If I'm playing a video at its native resolution, or very close, the quality is the same.

As far as latency, while overlay video does drop frames when lots of other things are going on in the window manager, textured video seems much worse. I get full pauses in the video for second length timescales even if I just move other windows around. If there is nothing else going on for the GPU, framerate is okay.

Anyway... I know its a work in progress, am happy there is support at all, and am sure it will get better.

[ferreira]

Is there work in progress to improve render acceleration for R300-R500, please? I have enabled EXA with AccelDFS on my dual 2.66 GHz P4 Xeon machine with its 2 GB RAM and Radeon 9550 card, but it still feels more sluggish at times than an old 500 MHz P3 with 384 MB RAM and a Radeon 7000 (also EXA with AccelDFS). This is particularly true when using Firefox.

Something ain't right...

P.S. Both machines are using Fedora 8 with the >= 6.8.0 ATI driver. (The P4 machine is also getting updated with the git version of the driver occasionally.)
It was posted somewhere here...
Add MigrationHeuristic "greedy" to your driver options in xorg.conf while you use EXA. Helps alot.

[chrisr]

It was posted somewhere here...
Add MigrationHeuristic "greedy" to your driver options in xorg.conf while you use EXA. Helps alot.

Yes, you're right, it does seem faster. But it also seems to leave "artifacts" behind on my desktop when I move windows around now. It looks like there is a problem repainting the desktop's background.

I am using the xf86-video-ati driver from git.