ATi Support on infinityOS

[darkphoenix22]

Sorry, I don't get the connection between driver maturity and good/evil.

NVidia moved to code-shared proprietary drivers for Linux over 10 years ago (not sure of the exact date), while we supported open source driver development or Linux-specific drivers until ~2006, with the first code-shared driver (at least the 3D part) appearing in late 2007. Since then there has been pretty significant process (does anyone disagree ?) and we expect that to continue.

Armin, there is a new 2D acceleration implementation in the works; you've probably seen it mentioned here a couple of times. In the meantime, Felix wrote a "backclear" patch for the X server which replaces the old "no-backfill" patch that distros were previously shipping but which eliminates most of the side-effects that resulted in the no-backfill patch being removed and certain 2D operations becoming very slow on XAA drivers.

I have no problem with proprietary code, as long as it works and it can do its job. I just feel that the open source development model is much faster and better for consumers. As an example, look at the Linux kernel. It had begun to rival the proprietary Unixes in the span of 3 years and all but killed them off in the span of 5.

As Eric Raymond/Linus Torvalds once said: "Given enough eyeballs, all bugs are shallow."

I am happy to see that you guys are working on improving the 2D functionaility of the fglrx drivers. I just feel that switching over development to the open source Radeon drivers would be much better use of your limited time and resources, especially since many of the requested features are already present in the Radeon drivers.

[Armin]

I apologize for greatly understating the work ATi has done on the open source Radeon drivers. However, they still can and should be doing much more.

I agree, there is a lot of work still ahead. Especially there image among the professional users is rather bad. I have been working at three universities (in Germany, Sweden and Finland) and everywhere they use Nvidia graphic cards in the (Linux) workstation segment. AMD/ATI has to catch up in this segment.

An other thing I was wondering (maybe bridgman can anwser this): Does AMD/ATI consider a hybride driver. With this I mean a open source driver stack (KMS, libdrm, xorg-driver) plus a closed source, but highly optimised libGL/ libglx

I guess (or hope) this would avoid a lot of problems for which fglrx is nowadys blaimed (delay in kernel/ xorg support, slow and buggy 2D performance). Just a thought, don't know to much about the problems that could arise with such a constelation, the idea would just combine the good things of both worlds, I suppose.

[darkphoenix22]

An other thing I was wondering (maybe bridgman can anwser this): Does AMD/ATI consider a hybride driver. With this I mean a open source driver stack (KMS, libdrm, xorg-driver) plus a closed source, but highly optimised libGL/ libglx

I guess (or hope) this would avoid a lot of problems for which fglrx is nowadys blaimed (delay in kernel/ xorg support, slow and buggy 2D performance). Just a thought, don't know to much about the problems that could arise with such a constelation, the idea would just combine the good things of both worlds, I suppose.

I would be more than happy with this solution and would unconditionally support it.

[Armin]

Armin, if you aren't using the backclear patch you should really give it a try. There are pre-patched X servers available for most distros, although you have to hunt around to find them sometimes.

Thanks for the hint. :-) I'm actually using a patched xserver (on Thinkpad T500 running Arch 64). Even though I'm already looking forward to the new 2D accel. Also with the 10.4 beta drive I actually don't recognise the slow window-resizing effects any more (also with a non-patched server).

And even though I named a couple of drawbacks of AMD/ATI drives I like to support AMD/ATI's open source efforts and therefore I buy AMD/ATI products. And I have to admit: The fglrx drive increased a lot over the last couple of years. I'm now using it since ~2006 (Ubuntu 06.06 :-) ) Right now I'm really happy with the performance (3D and 2D) and the stability of the fglrx-driver, have not even had one single kernel panic since at least 9 months.

[rohcQaH]

An other thing I was wondering (maybe bridgman can anwser this): Does AMD/ATI consider a hybride driver. With this I mean a open source driver stack (KMS, libdrm, xorg-driver) plus a closed source, but highly optimised libGL/ libglx

In the software world, it's never as easy as assembling Lego bricks. GFX drivers are very complex, every single drawing command is passed through several layers of abstraction until it reaches the hardware.

Those layers communicate over clearly defined APIs. But those APIs are different between fglrx and the OS driver stack. Adding another translation layer in between would probably negate the performance benefit of using fglrx in the first place: there's a translation overhead and no layer can be optimized to work well with the other.


So I doubt that a "hybrid" approach would be faster than a pure OS solution. It might violate the respective licenses of the OS parts, too.

And there's another reason why I'd want a pure OS stack: end-to-end debugging and profiling of gfx applications. Every time a stack trace ends inside some binary libGL.so without debugging symbols, I grow another grey hair.

[darkphoenix22]

It would be good enough for users who want top 2D acceleration but only occasional 3D acceleration.

I honestly just want to be able to view videos on my computer without hassle and maybe occassionally play a game via Wine. Many of my users fit this profile as well.

To be honest with you, nowadays there is increasingly little demand for 3D performance in the desktop sector as mnay end user are switching to consoles for their gaming (I know I only game on the Wii, and I must thank ATi for the graphics chip in that console. It does so much with so little.)

Perhaps there should a desktop ATi driver, with open source 2D capabilites and a closed-source 3D library and fully closed source 3D workstation driver with fully optimised 3D capabilities. The needs of the desktop and the workstation are growing apart with every day, so perhaps having 2 official drivers is the best solution.

This seems like would differ very little from a development persceptive from the situation that exists now at ATi. All that really would be needed is a couple people working on the open source drivers and the glue to bind them to the closed-source 3D library.

[darkphoenix22]

Edit: This seems like it would differ very little from a development perspective from the situation that exists now at ATi. All that really would be needed is a couple *more* people working on the open source drivers and the glue to bind them to the closed-source 3D library.

[Armin]

Ohh don't understand me wrong, I also want a fully open source driver stack! But as bridgman already meantioned a couple of times, the oss-stack will most likely not reach the performance of the closed one.

I guess a lot of people (especially me) would love to switch to the OSS-stack, but unfortunately in some cases (also in mine) 3D performance is still important. I gave the oss-stack already a couple of times a try and I loved it (especially KMS, EAX and Xv), but the 3D performance made me always switching back to fglrx.

It was just meant as a kind of plug-in solution of which AMD/ATI also would benefit, because they "only" have to maintain one linux driver + the sharde code of the libGL. I'm aware of the fact, that one would have to somehow match the two API's (OSS libGL <-> prop. libGL (+ libGLX)), but I would suppose this is still faster than the pure oss-libGL. In the end I don't know and it would anyhow take some time to implement this, so not really a quick solution.

[darkphoenix22]

I would love a completely OSS driver stack as well. However, I will settle for a hybrid driver with awesome 2D performance with as much of the code being OSS as is possible.

[Armin]

Perhaps there should a desktop ATi driver, with open source 2D capabilites and a closed-source 3D library and fully closed source 3D workstation driver with fully optimised 3D capabilities. The needs of the desktop and the workstation are growing apart with every day, so perhaps having 2 official drivers is the best solution.

Actually AMD/ATI is doing almost exactly this: A fully open source "desktop" driver plus a "fully closed source 3D workstation driver". This just takes time ... Funnily u were criticising them for doing that!

Btw, don't underestimate the mesa OpenGL implementation, for a "normal" desktop user the hybrid solution might not even be necessary. I just brought this idea up to replace someday the workstation driver (I guess if this is easy to implement AMD/ATI will do it anyhow, without me mentioning it. )