Bickering Continues About NVIDIA Using DMA-BUF

[christian_frank]

Had you followed closely, you would understand, that AMD does not opensource properly.

They do not provide all specs. This is why performance and functionality are REDUCED.

Also, driver uses AtomBIOS rather than accessing hardware directly (what proprietary does).

And opensource driver lacks efficient VLIW compiler, the original compiler is still used by proprietary catalyst, which is why 3d performance of opensource is reduced.

Releasing *full* specs will help situation *a lot*.
Working together with community on the driver, or funding *adequately* would accelerate the process, so we would see results in this era.

This is how intel does it. The wrong things about intel - not using gallium and not having enough raw power in their hardware.

if the only problem is the hw , why is their win driver faster than the linux one ?

[crazycheese]

if the only problem is the hw , why is their win driver faster than the linux one ?

Suggestions:
- using Intel-compiler to compile the driver.
- patented stuff (opengl4 already)
- cheating and performance/precision cuts
- their windows driver team and linux driver team are not working together, and latter one is much smaller.

Its also not "much" faster, 20% at max.

[pingufunkybeat]

Don't forget app-specific optimizations. Proprietary drivers are full of hacks for speeding up specific games.

OSS developers generally avoid those because they make the code unmaintainable.

[tuke81]

You need to shut up. NOW.

I specifically bought an Optimus laptop to research approaches to solving Optimus on Linux. Why else did I spend so much time on the Bumblebee and Ubuntu Hybrid Graphics team?

I also had expectations that PRIME and dma-buf would be adequate. The situation is arguably worse today since Nvidia is locked out of dma-buf and are making even their high-end GTX chips Optimus-only.

Just one question, Does this matter make it harder to make prime work. In other words could it help if you can reverse-engineer nvidia binary driver while using optimus if nvidia could use dma-buf.

[whitecat]

Don't forget app-specific optimizations. Proprietary drivers are full of hacks for speeding up specific games.

OSS developers generally avoid those because they make the code unmaintainable.

+1, this is certainly and partly why the proprietary blobs are so huge!

[Rigaldo]

Yeah and MAYBE the pixie of open source will become the new CEO of NVidia and open up everything they have.

The only thing, which would improve, when we get more of these people is the number of entries in the bugtrackers.

I like your attitude. Real positive btw. So I take it you have a proposal. One likely to happen.

[crazycheese]

Don't forget app-specific optimizations. Proprietary drivers are full of hacks for speeding up specific games.

Profiles would be VERY GOOD for opensource drivers as well. And you won't be able to cheat - you'll see clearly what hack does what and at what price.
In closed source they are misused however, to claim performance which the card does not give at all. Even profiles are evil, when closed.

[jrch2k8]

My thanks to alan for defending open source, 1000 kudos to him and the other sane developers in the kernel, now for the the tards blaming them !!!!this is not their fault!!!! is a legal issue since in the COPYING files of the kernel is !!!!VERY CLEARLY SPECIFIED THE LICENSE AND RESPECTIVE EXCEPTIONS!!!! and that cannot be changed without the consent of every copyright holder[including the dead ones], so since linux kernel is gpl2 since 2 decades ago and is nearly impossible to change that now they have to adjust their blob or simply don't support optimus on linux and the same applies to fglrx or any other blob.

this time michael is in the wrong since he implies alan cox is stoping nvidia when he is only properly reporting the license issue and stating that in the hypothetical case of a full kernel license change he won't approve it for his code and this position is supported by many other kernel devs that don't consider change the license in their code been good for them[which is their right not your to criticize].

nvidia is trying to get some legal evidence in case of a trial that this devs approved their violations to the license so the can slip throught it undamaged in the eventual case of a litigation for infrigement, this is not a technical issue or a kernel dev being grumpy is just nvidia trying to find a grey area in the license to avoid damages

P.D for the nvidia trolls
* is not possible to change the kernel license in a reasonable timeframe [optimus will be long gone in the past by then], again read the frigging copying
* no change the license in the header probably won't work either read the copying, again read the frigging copying
* alan cox can be hired by nvidia and forced with a gun to accept the patch and still won't be legal, again read the frigging copying
* no not even linus can change the kernel license before you unborn son became a grandpa, again read the frigging copying

[entropy]

How about a pragmatic "solution":

• NVidia will provide documentation to their hardware (similar to what AMD does) and,
  thus, finally shows real support of Linux (and OSS). Additionally, they will support the OSS driver developers.
• In return the use of the kernel API by the proprietary driver will be tolerated.

I know this does not solve the legal issue, hence, let's stick to "pragmatic" solution.

But *I* think this would really solve quite a bit:

• Users of the blob see Optimus support and more (Project Denver related dma-buf stuff , etc.).
  At the same time, they should know this tainted configuration is completely unsupported and even represents
  an enhanced security risk.
• Users that care about OSS drivers and the GPL will significantly benefit from the documentation and enjoy
  support from the kernel developers.

Yes, it sounds like horse-trading and it's full of "goodwill".
But it's exactly what's missing from NVidia until now.

[GT220]

http://www.phoronix.com/scan.php?page=news_item&px=OTAxMw

We tried enabling OpenGL on Linux, and discovered that most Linux drivers are so disastrously buggy (think "crash the X server at the drop of a hat, and paint incorrectly the rest of the time" buggy) that we had to disable it for now. Heck, we?re even disabling WebGL for most Linux drivers, last I checked.

If open source drivers were actually usable or bug free, fact is they're not and it's well proven it's significantly worse and buggy as hell, even with documetation like the Radeon open source drivers. Open source driver Nazis and GPL Nazis are only killing Linux, simple as that. The biggest enemies are not Microsoft or Apple, it's the GPL Nazis themselves.