Future of ATI Linux drivers

[cleric]

Everybody is asking for a bugfree driver. Tomorrow please.

That this is not realistic doesnt matter for most of them.

Anyways:

What will help everyone would be a more open information policy. Is there a reason that ati/amd does not offer plans and intentions (even without dates)? Or is it that ati/amd itself is not sure about what todo next? Hiding information may work in the windows world - but linux is a community that exchanges informations. People who buy a video card will be told by the community to do so (wiki/forum/chat).


That ati/amd is going for oss is a nice direction but it seems that they lack (imho) trust in those steps. Else they would just mark the closed driver deprecated and start using all available (linux) manpower helping out the opensource team, but who knows... maybe that exactly what they are doing - but not even one customer knows about it.

If ati is scared that they cant decode drm stuff. Well bridgeman, in one of your posts you wrote that there are just a "few" linux customers. Believe me: there is not even one who does care about drm when it comes to decoding drm content with his ati/amd gfx card using linux... Not now, not in 2 years.

If you want more customers:

- make a real cut and push the oss driver even more
- discuss a MUST featureset that should run stable
- care more for stability then for speed&features
- tell the customers what your doing
- dont compare windows users/market with the linux one... its something different.

good luck

PS: i dont need to mention that i am not a native english speaker, do i ? Sorry about that.

[yoshi314]

- discuss a MUST featureset that should run stable

that's obvious - 2d, 3d, multi-display support. wide hardware support with multiple platforms (not only x86, x86_64) and systems ( not only linux, but also *bsd and more ). extra "fireworks" like aiglx and friends are secondary.

Believe me: there is not even one who does care about drm when it comes to decoding drm content with his ati/amd gfx card using linux... Not now, not in 2 years.

what about tivo-like device manufacturers?

That ati/amd is going for oss is a nice direction but it seems that they lack (imho) trust in those steps. Else they would just mark the closed driver deprecated and start using all available (linux) manpower helping out the opensource team, but who knows... maybe that exactly what they are doing - but not even one customer knows about it.

how would ati explain it to their customers? fglrx is the only driver that provides fast 3d support for all new cards. of course that will change in time.

remember that it takes time to prepare documentation and release it in a form that is of any use for people outside ati.

my work consists in 50% of writing documentation for commercial software at the company i work for - it takes a lot of time to do it right.

you have to make sure you don't put 3rd party IP in it [which is very time consuming process in case of a gfx card company, since possible legal issues are like a minefield in there], you also have to make sure your documentation covers anything and does not contain errors [which is much faster, but still time consuming].

[bridgman]

Cleric;

The biggest part of our Linux market is workstation, which realistically needs closed drivers today and in the future. Those drivers can also be of interest to some consumer users, particularly if video support improves over time. If we cut off fglrx development we walk away from the workstation business, and that doesn't seem like a great idea. I don't know if you have seen any recent reviews of our workstation products, but it's worth taking a look to see both sides of the driver story.

Having said that, I think we also need an open source driver for all the reasons you already know. It may be possible to do some kind of hybrid in the future but today the open source 3d framework is just not ready to support a high performance driver. If you subscribe to the dri-devel mailing list you can get the play-by-play on this.

I don't believe any of the HW vendors offer much visibility into future plans. This is a very competitive industry and it's no surprise that if company A announces that they will be shipping X on a certain date that company B will make sure they have X+1 on the same day.

I suspect what you're looking for is a "when will I be happy" roadmap; since most of what you need to be happy is fixes for certain apps or certain system configurations that's not a great subject for forecasting. What I can say is that the priorities most of you talk about are the priorities we are working on.

DRM is pretty much a non-issue in our current plans. The only time it comes up is when someone says "why don't you throw everything out and re-architect THIS way", and the possibility of a future OEM / distro call for DRM is one part of the answer to that question. Other than that, all we're trying to do is to keep the DRM we *have* to have for Windows from getting in your way.

I don't think we have an option to not treat Windows and Linux together, since (to be blunt) the Linux market is not big enough on its own to support the kind of products we all enjoy, and so there needs to be a lot of common technology between the two markets. If being open is great in the Linux world, being closed is great for the Windows world, and IT'S THE SAME PRODUCT IN BOTH CASES we really have to think about both markets together for better or for worse.

Finally, if you hadn't said anything I wouldn't have suspected English was not your first language. You're doing great !

Yoshi;

Yep

[cleric]

that's obvious - 2d, 3d, multi-display support. wide hardware support with multiple platforms (not only x86, x86_64) and systems ( not only linux, but also *bsd and more ). extra "fireworks" like aiglx and friends are secondary.

I dont think that they need support many more platforms as there are not really any other platforms supporting PCIe and are used by endusers. New AGP cards are, even on windows, not supported by ati because they were officially not developed by ati. I pesonally don not think that you should buy a 3870 AGP (if there ever is one) because it does not make much sense to buy high priced agp cards. There are several reasons why you only lose buying a new ati agp card.

what about tivo-like device manufacturers?

If they want to make money with that they may feel free to develop their own drivers. If they use DRI2 it should not be too hard to mod the existing ones and develop a interface for such needs. But if you know a manufacturer using linux and BD DRM etc. plz tell me. I am not aware of one.

how would ati explain it to their customers? fglrx is the only driver that provides fast 3d support for all new cards. of course that will change in time.

Remember that many users may be reading this. Ati should not delete the old driver but suspend the main development. I persoanlly cannot use the driver because i need a stable system. Stable: using my desktop environment for days without restarting/lockups etc.

remember that it takes time to prepare documentation and release it in a form that is of any use for people outside ati.

I did not force any timeplan. Free driver with 2D accel and nice overlaysupport would be a nice first step for me.

my work consists in 50% of writing documentation for commercial software at the company i work for - it takes a lot of time to do it right.

Again i never said that they should release it tomorrow. But they should stop wasting resources

you have to make sure you don't put 3rd party IP in it [which is very time consuming process in case of a gfx card company, since possible legal issues are like a minefield in there], you also have to make sure your documentation covers anything and does not contain errors [which is much faster, but still time consuming].

Did i tell you that i dislike 3rd party software IP :P

But your right. Thats something that cannot be helped. But the developer allready coding for ati may use internal docs for oss drivers as long as they can decide what they may use.

[yoshi314]

I dont think that they need support many more platforms as there are not really any other platforms supporting PCIe and are used by endusers.

that's because desktop market is dominated by x86 processors. what if there was a door open for, say, ppc architectures.

or think some time into the future - what if e.g. cell (aka "the ps3 cpu") based pcs will emerge?

[cleric]

bridgman;
The biggest part of our Linux market is workstation

Okay. Something you should change quick. Try it by publishing a nice oss(or even closed) driver :P

bridgman;
I don't believe any of the HW vendors offer much visibility into future plans. This is a very competitive industry and it's no surprise that if company A announces that they will be shipping X on a certain date that company B will make sure they have X+1 on the same day.

Well. Try to be a market leader and not a follower, your opensource ambitions are a great step in that direction. Yes i know writing this is much more easy then doing it

bridgman;
I suspect what you're looking for is a "when will I be happy" roadmap;

Well not really. I did not even post one featurerequest in my first post, did i?

Some posts ago you wrote about "companies will do what people want if they have to" together with music & drm vs video & drm. Your right about that. Many onlinestores are starting to sell non DRM content but "plain" mp3s. As soon as there is a real market for online videos the same will happen or the industrie will really break apart.

My personal featureset has nothing todo with what you should do. It consist of two steps:

a) rock stable 2d acceleration with xv support (i dont care which backend as long as it works) without bugs.

b) same as above and 3d acceleration. also stable.

If you are about to add new features its fine. But please offer a stable driver


So my main objectiv is rock stable. I dont like to say it but the actual oss radeonHD driver is NOT rock stable. Much better then the latest closed one but considering that it has nearly no features its sad that it does not work stable over days.

bridgman;
DRM is pretty much a non-issue in our current plans. The only time it comes up is when someone says "why don't you throw everything out and re-architect THIS way", and the possibility of a future OEM / distro call for DRM is one part of the answer to that question. Other than that, all we're trying to do is to keep the DRM we *have* to have for Windows from getting in your way.

Nice to hear.

bridgman;
I don't think we have an option to not treat Windows and Linux together, since (to be blunt) the Linux market is not big enough on its own to support the kind of products we all enjoy

It does and it will. As far as i know nearly all linux users use nvidia cards. change that. but to do so you need nice drivers

bridgman;
the Windows world, and IT'S THE SAME PRODUCT IN BOTH CASES we really have to think about both markets together for better or for worse.

I dont think that your really do. It seems more like: "ah and dont let us forget our small linux market"

bridgman;
You're doing great !

Thx


PS: I am one of the users that bought a card because AMD is going open source. Thats why i am pretty relaxed My last 14 GFX Cards were all from nvidia but i felt like: hey they are doing the first step. lets reward them even if it wil be a pain in the ass.

[cleric]

or think some time into the future - what if e.g. cell (aka "the ps3 cpu") based pcs will emerge?

Just for the books. If there is a proper oss driver it shouldnt be a problem for new platforms. But just for the records: its not very likely that this will happen in the next 5+ years - after apple changed to x86_64

[Porter]

Just to reiterate one REALLY IMPORTANT part of Cleric's posts above:

a) rock stable 2d acceleration with xv support (i dont care which backend as long as it works) without bugs.

The biggest problem with fglrx right now (in my opinion) is lackluster 2D, and that's probably 90% of the clock cycles that these devices see in the real world, in desktop applications.

Most Linux users aren't gamers, and honestly most of us aren't all that interested in 3D in general as a major daily-use feature. The workstation market certainly is, but you can't (or shouldn't) overlook basic desktop 2D functionality in a dash to first satisfy the workstation market's perceived 3D application demands.

[bridgman]

Just a reminder that xv support on any 5xx or higher part needs 3d first, since the video processing is done in shaders rather than in the overlay hardware. It's not a question of "just turning on the scaler" any more.

[Porter]

Wow, that's interesting. Is doing 2D in shaders faster than in overlay hardware? What's the benefit?

Also, isn't 2D acceleration itself still done in hardware on the R5xx chips? I thought I read that only video overlay was switched to shaders for the R5xx... where in the R6xx chips everything 2D is done in 3D.

Oops.. yep... found it... http://www.phoronix.com/forums/showp...08&postcount=8

So what's the deal with 2D accel on the R5xx chips?