Why Wine Developers Don't See Gallium3D D3D9 As An Option

[duby229]

Without forget legal issues, model driver issues (microsoft rights) and other issues like use nvidia adapter name (GTX680) without nvidia authorization

You're wrong. MS lost a lawsuit several years ago that allows alternative implementations of DX legally. Nine is fully legal.

[duby229]

There is a point, and I mentioned it in my talk: Finding out how much impact it has, and using it as a tool to analyze Mesa and Wine performance. With the help of nine it is possible to e.g. compare shaders side by side and find places where the GLSL compiler creates inefficient shader code.

The impact it has is that by not including nine support you're putting everyone that doesn't use nvidia proprietary driver out in the cold. Wine sucks ass on everything but nvidia hardware. And it's wines fault for being so fucking stubborn.

[duby229]

It only works well on nvidia's hardware to begin with, because of choices that wine devs made. And now they are officially making excuses for not supporting functionality that definitely improves non-nvidia configurations.

It's time that the folks in charge of making decisions at wine were removed from the decision making system. It's time for a fork.

[Dorsai!]

That's the driver authors fault. Put blame where it belongs. the whole point in gallium was to develop an architecture that both OSS drivers and proprietary drivers can use.

You are of course right about gallium3d being the ideal situation but I would refrain from putting blame on anyone altogether.

If driver programmers do not want that option for one or another reason, what can we do to stop them?

We could buy less Nvidia cards for their unwillingness to adopt an open gallium 3D driver. But they are still the most powerful cards on linux and work well. So people go on buying them.

Same with Intel. They do not and will not (for the foreseeable future) commit to an gallium 3D driver because they found that pure mesa works best for them. And despite that their driver is still the highest quality open source driver out there. This is one of the Unix design philosophies, sometimes worse is better. Everybody likes "done right" solutions, but they have to work first too.

Maybe all of this will change once the new AMD attempt of a unified driver architecture has become more common. But I wouldn't bet on it.

And as long as the common denominator is OpenGL in Linux application designers will have to use that. But even if we had Gallium 3D drivers we probably wouldn't see much Direct3D ports because Mac will definitely not use it. And most companies will make a Mac port first and then go for an additional Linux port because they were already 80% there with the Mac port anyway. But that also means sticking with OpenGL.

[duby229]

You are of course right about gallium3d being the ideal situation but I would refrain from putting blame on anyone altogether.

If driver programmers do not want that option for one or another reason, what can we do to stop them?

We could buy less Nvidia cards for their unwillingness to adopt an open gallium 3D driver. But they are still the most powerful cards on linux and work well. So people go on buying them.

Same with Intel. They do not and will not (for the foreseeable future) commit to an gallium 3D driver because they found that pure mesa works best for them. And despite that their driver is still the highest quality open source driver out there. This is one of the unix design philosophies, sometimes worse is better. Everybody likes "done right" solutions, but they have to work first too.

Maybe all of this will change once the new AMD attempt of a unified driver architecture has become more common. But I wouldn't bet on it.

And as long as the common denominator is OpenGL in Linux Application designers will have to use that. But even if we had Gallium 3D drivers we probably wouldn't see much Direct3D ports because Mac will definitely not use it. And most companies will make a Mac Port first and then go for an additional Linux port because they were already 80% there with the Mac port anyway. But that also means sticking with OpenGL.

You're missing the point. nvidia doesn't have to open their driver. closed drivers can work just as well on gallium.

EDIT: You're missing a second thing as well, Apple will be making OpenGL a second class citizen soon. That will hurt everyone hoping for mac ports to come to linux.

[Dorsai!]

You're missing the point. nvidia doesn't have to open their driver. closed drivers can work just as well on gallium.

EDIT: You're missing a second thing as well, Apple will be making OpenGL a second class citizen soon. That will hurt everyone hoping for mac ports to come to linux.

Why would I be missing the point? Ok, Closed Source would work to with Gallium, but Nvidia would have to rewrite their complete driver anyway or at least work on nouveau while they have a completely working solution that puts them in front of any other driver on the market right now.

As for apple and its metal, it is still to be seen if they will have success with it on the desktop. My bet is that it stays with iOS, but thats just a guess. If it did happen, there is nothing we could do about it. Nvidia would probably ignore it alltogether or just bolt on its own implementation rather than rewriting their own driver.

Again: I share your hopes and opinions about gallium3d being the cleaner solution, but I do neither expect driver developers nor application developers readily jump on that bandwagon. This will take time and probably even never happen. Realistically speaking. It would be nice though.

[-MacNuke-]

It's time for a fork.

Go ahead and do it... I think the WINE developers could not care less

[duby229]

Go ahead and do it... I think the WINE developers could not care less

Yeah, obviously. At this point it doesn't seem like they care about much at all.

[brent]

I think the justifications are a bit strange and inconsistent. Wine's GL-D3D translator was modeled and optimized to work well with Nvidia's closed-source drivers (because that is what the developers used), so it's not really a surprise it works best with those. Moreover, maybe Nvidia's drivers would also better with native D3D support similar to nine. We just don't know. Also, they're completely ignoring nouveau. And what about CUDA and PhysX? Those are *completely* NVIDIA only yet supported in Wine. And last but not least, wined3d CSMT is not part of Wine master yet, so most users don't benefit from it at all.

[clementl]

Stefan you're free to consider proprietary drivers a viable alternative, but please don't tell people to use Nvidia: it's our choice to not. As a radeonsi user I think it's *STUPID* to not take advantage of a better alternative when it's already available and working. gallium nine is way better than opengl translation and I will keep patching my wine. I would like to say a big THANK YOU to the gallium nine developers, who finally made wine gaming actually *viable* for non-nvidia users.

3) The Nvidia driver shows that a fast OpenGL implementation is possible.

If you read carefully you see he's not recommending or dis-recommending Nvidia's proprietary drivers. The point he's trying to get across is that Wine's Direct3D-OpenGL translation layer is not the bottleneck here. The fact that Nvidia's drivers combined with Wine+CSMT achieve the same performance as running games natively on Windows, shows that Wine's translation layer is adequate. Focus thus should be on improving Mesa's OpenGL performance, which will improve all use-case scenarios.


I honestly hope that something like this will never be needed. I'd rather see one API get all development (OpenGL), than that more complexity is added to service only a small amount of users. On top of that, DX9 is not used much for active development anymore. By the time this driver has decently matured our PC's might be able to run the heavier DX9 applications with translation overhead.