A Game Developer's Perspective On Linux Driver Quality

[robclark]

This is not an excuse or a justification for how insanely anti-FOSS the company is. AMD can't FOSS Catalyst for similar patent, trademark, and copyright reasons, so they just devoted developers to Mesa. Nothing except Nvidias own vindictiveness, awful legal department, and executive board stops them from doing the same.

To be fair, you really shouldn't paint all of nvidia with the same brush. Groups outside of the desktop-gpu group seem to somewhat better on the FOSS and documentation front. Well, that is still not including the 3d parts, but it is at least as good as any of the other SoC vendors other than broadcom so far. And their mobile folks seem to be doing some good upstream kernel work.

At any rate, I think most of us would rather see documentation rather than code. I could really care less about catalyst src code.

[zanny]

What is =/= meant to mean?

It is the poor man's != when you aren't a programmer.

[anda_skoa]

Also to note, he hasn't tried out A or B's open source drivers.

The scary part of this was the he seemed to be genuinly afraid that vendors A and B would retaliate in some form if he even considered testing with those.

Cheers,
_

[bridgman]

What about Intel for Mac Os X?

My guess was that C#1 was Intel open source for Intel's own gfx hardware, and C#2 was the embedded driver for PowerVR hardware.

This assumes the author was only discussing Linux drivers, which I *think* is the case but not 100% sure.

[GreatEmerald]

According to bridgman, the code for that is going around between developers and the legal department, but on a low priority. If the reason the Catalyst driver will not be open sourced is secret OpenGL ingredients, why don't we have the non-OpenGL parts already? The truth is, no one works on this and we will never see UVD on those chips.

"Low priority" being the key, rather than "legal department". They just don't care as much about it, not that there are some outstanding legal problems with it.

[bridgman]

"Low priority" being the key, rather than "legal department". They just don't care as much about it, not that there are some outstanding legal problems with it.

Interesting because "low priority" is the one thing I don't think I said, other than in the context of having just pushed a lot of new IP out and needing to give the impacted teams time to work on other stuff as well.

And I definitely don't remember saying anything about not caring as much

Also, for the lebenty-million-six-hundred-and-(counts on fingers)sixth time, it's a lot more than legal review.

re: "if the problem is OpenGL secrets why is that holding up other things ?", I don't think I understand the logic behind the question.

The statement was that the main obstacle to opening up Catalyst OpenGL was OpenGL secrets, while (surprisingly enough) the main obstacle to opening up other areas is secrets in *those* areas. There's a beautiful consistency to it all if you step back a bit.

[robclark]

My guess was that C#1 was Intel open source for Intel's own gfx hardware, and C#2 was the embedded driver for PowerVR hardware.

This assumes the author was only discussing Linux drivers, which I *think* is the case but not 100% sure.

I think he was talking about intel windows driver (which maybe shares a codebase w/ osx driver??).. pvr would be 'Vendor D'.. and hopefully game dev's have the good sense to run away from that one :-P

[bridgman]

Huh... you're right. The Phoronix article says "Linux Driver Quality" but the blog is about Windows and Linux. Good catch.

[Awesomeness]

Wouldn't those drivers have to be signed my Microsoft though? I wonder what their stance on freely licensed kernel drivers are. (no sarcasm)

The whole Mesa stack is under a BSD license. Even if MS is against open drivers for Windows, the Windows port could be closed source.

[Awesomeness]

Any existing Non Linux drivers out there using Gallium?
(*BSD was porting some, are those working?)

AFAIK Haiku.