Intel Releases New Driver, Kills EXA/DRI1

[phoronix]

Phoronix: Intel Releases New Driver, Kills EXA/DRI1

It was just two weeks ago that Intel released its xf86-video-intel 2.7 driver, but there is already a new test release that will lead up to the release of the xf86-video-intel 2.8 series. Normally it is two or three months before a new test release is available for Intel's quarterly Linux graphics driver update, but that is not the case with their Q3'09 stack. Less than a week ago we shared that Intel is looking to drop EXA support in favor of their UMA Acceleration Architecture (UXA) as well as stripping out compatibility for the older DRI1 infrastructure...

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

[wswartzendruber]

I'm actually all for this. Why maintain outdated legacy code that doesn't run as efficiently as the newer stuff? Getting rid of DRI1 and EXA should also mean more time diverted into maintaining and fixing DRI2 and UXA.

[ivanovic]

I am sure that there are still lots of bugs left in UXA, but probably several of those are not really reported yet. "Forcing" the users to use UXA leads to at least some reporting the bugs which will eventually lead to those bugs being fixed (Devs can only fix what they know is broken!). So yeah, IMO it is a good step to remove the old legacy code even though it will lead to some problems and complaints in the short term.

[ethana2]

Will this driver make it into 9.04 updates?

[Michael]

Will this driver make it into 9.04 updates?

No, Jaunty is tracking 2.6 series... Not even 2.7. So 2.8 will not make it in.

[ethana2]

Ah. Onward to 9.10 then!

[nanonyme]

I am sure that there are still lots of bugs left in UXA, but probably several of those are not really reported yet. "Forcing" the users to use UXA leads to at least some reporting the bugs which will eventually lead to those bugs being fixed (Devs can only fix what they know is broken!).Devs can also only fix what they have time to fix. Perhaps we have to draw the conclusion that Intel developers have too much time on their hands so they're begging for new things to fix. ;)

[ethana2]

Devs can also only fix what they have time to fix. Perhaps we have to draw the conclusion that Intel developers have too much time on their hands so they're begging for new things to fix. ;)

That implies some false things. I'd say more like the opposite of that..

[nanonyme]

That implies some false things. I'd say more like the opposite of that..Well, if you drop the mostly stable acceleration method (EXA) in favour of a raw acceleration method (UXA), what else would it be than trying to ask people to find you new bugs in raw acceleration method? ;) Thus they want people to file more bugs which implies they want more work. Logical so far? ^^

[ethana2]

They're dropping an old, buggy acceleration method for a modern and maintainable acceleration method. The whole point of this is to have less bugs to take care of.

[myxal]

The sooner I get KMS the better. If it's really as good at handling suspend-resume as Intel would have us believe, I hope I can say goodbye to X server dying every 15 suspend-resume cycles, 3D working only inside the top-left 1280x1024 after first resume (yes, some of us enjoy WSXGA+ or higher :) ), VTs not working and who knows what other bugs I haven't run into just because I'm not running a compositing WM.

It's about time Linux graphics stack entered the 21st century. Now if only we got the remaining vendors (notably SiS and nVidia) to provide useful docs...

[Craig73]

I'm finding this driver is running as well as the Jaunty version did for me, except with the whole stack updated GL runs faster for me.

There is an annoying reproducible bug on my 855GM where a 2d or 3d app hangs when exiting (that existing in Juanty) that hopefully will get fixed soon. [I have to figure out how to report it, and how to determine where the issue is in the stack or if I'm responsible for that]

I noticed that Intel has their own automated and manual test suite that builds the latest drivers from GIT. Perhaps more people from here complaining about stability can jump in and help out.

http://intellinuxgraphics.org/testing.html

[luciferin]

The sooner I get KMS the better. If it's really as good at handling suspend-resume as Intel would have us believe, I hope I can say goodbye to X server dying every 15 suspend-resume cycles...

Are you running Ubuntu? I only ask because KMS is supported by most (all?) intel hardware for at least a month now with the most up to date packages (Fedora & Arch are the only distros using them that I know of though). I guess Ubuntu has to wait for the October release.

[drag]

In reality this isn't going to affect users much in a short time frame.

With Fedora 11 being released soon and Ubuntu just released then that means that most users on the Linux desktop are going to be sticking around with the 2.7 series drivers for, at a minimum, another 6-8 months. So they will support DRI1 and the old Xorg EXA model for as long as people continue to use those distributions.

What this release is really saying is "We are planing on dropping non-GEM-based acceleration methods by the end of the year" which is when the majority of desktop oriented Linux distros will be using the 2.8-or-later graphic releases. (Ubuntu 9.10, Fedora 12, etc)

[yoshi314]

a good decision. it might look a bit risky to some, but as long as there are separate stable branches for older X.org users in git - i'm fine with it.

[greg]

A good move, but maybe making UXA the default and dropping only XAA would make for a smoother transition.
Does UXA work well on the other operating systems that support DRI so far (*BSD mostly)?

[wswartzendruber]

Holy crap, good point. This is going to kill BSD/Solaris support as those don't have GEM implementations.

[greg]

Damn, that sucks. Let's just hope they won't completely abandon the older driver branches until there is GEM on *BSD and Solaris.

[drag]

Well if I was a BSD developer my first instinct would be (once I understood what was going on with X) to let the Linux folks and the Intel fellows thrash through the changes and improvements in X until Gallium3D comes along. Then worry about porting Gallium.

Sometimes it's just not worth keeping up with the Joneses.

I get the impression that the majority of the *BSD folks don't really bother much with their system on the desktop. Their main rigs generally are going to be Windows with things like SFU, Cygwin, Putty, and Xming for Unix compatibility. Basically turn Windows into a Unix workstation. If they use their BSD as a desktop it's going to usually be a secondary role.

(Running Unixy stuff on Windows is not really that bad. Seriously.. it isn't. People have put a lot of work into making it work decently. Of course Linux is better, but not everybody has tolerence for it or needs Windows for other reasons.(like that is what IT gave them and they can't install their own OS.)

Either that or use OS X, which is a lot nicer and more affordable then it used to be.



So I wouldn't worry too much about them. Once GEM and DRI2 support stabilizes then it will be much easier for them to get things working well. So the quicker the Intel/Linux folks get their stuff stabilized then the better off everybody will end up being.

[Craig73]

There is an annoying reproducible bug on my 855GM where a 2d or 3d app hangs when exiting (that existing in Juanty) that hopefully will get fixed soon. [I have to figure out how to report it, and how to determine where the issue is in the stack or if I'm responsible for that]


That annoying reproducible bug was actually PulseAudio (weird) ... video/3d/2d.... it all runs great now

Thank-you Intel :-)

I'll see what I can make fail... but they are running wonderfully right now (2.7.99.1 on an 855GM)