Streams vs. GBM: The Fight Continues Over NVIDIA's Proposed Wayland Route

[finalzone]

What do you mean "delayed"? Wayland is already here, protocol, server and client library, all are stable, nothing to delay. It's adoption almost here: toolkits. DEs and drivers all support it. AFAIU several distibutives plan to offer wayland server as default by the end of this year. The only thing that can delay is the moment, when it will be possible to drop support of X11. And this moment won't come until DEs will support nvidia driver with wayland (or nvidia will loose their marketshare).

NVidia has already implemented EGLImage/EGLStream on weston, maybe they will implement (or pay someone to implement) same codepath in one or two most popular DEs.
EGLImage/EGLStream are not some NVidia specific extensions, so, maybe, in year or two this codepath will be used on all GPU-accelerated implementations.

The use of "maybe" suggests EGL streams will be left as a bit rot.

[Khrundel]

And who is going to take care of that code after it has been implemented?

Community. This is not something unusual, this year AMD created AMDGPU driver, I think whole point of which is to transfer burden of supporting kernel module and xorg/wayland binding of AMD's proprietary driver to community.

Maintainers have already rejected this approach

We are talking about the free software. Nothing can stop nvidia from forking compositor, nothing can stop distro maintainer from taking this patched version, if it will solve other problems. And later maintainers of GNOME will be forced to merge this branch.
But I think propability of this development is low. My bet - after some grunt GNOME developers will implement this themselves.

and someone from GNOME has called having two code paths "highly undesirable"

They are already have at least two code paths (X11 and Wayland). If two completely different code paths are undesirable, they shouldn't start this wayland thing in a first place. If they want to get rid of X11 in a near future, they have to support the nvidia driver with the wayland, one way or another.

Also this idea has been discredited. None of the other vendors or developers is moving in that direction.

https://lists.freedesktop.org/archives/wayland-devel/2016-May/028818.html

Right now most developers don't like this idea because of two code paths. If there will be already two code paths, they will choose according other matters.
This is IT, many approaches became obsolete before widespread adoption. Maybe GBM - one of these. I believe nvidia developers started this controversy not because they hate humanity, but because EGLStream/EGLImage aproach is better.

[Jedipottsy]

Holy shit, this thread is gold. All the NVIDIA hate going round is outstanding. Pure gold.

Might I remind some of you, that while Linux is a platform for open source evangelists, it is also used in working environments. Pretty much the whole VFX industry use Linux and NVIDIA. I know we do, so I don't care if the driver is open source, closed source or made by bill gates himself, I care that it works, its reliable and that it has great performance. It would be nice if it was open source, but when I'm 8 hours from a deadline I'll bank on reliability over that personal satisfaction of running open source code every single time. And for us, that's nvidia.

Now if what NVIDIA are saying is true, and that eglstreams provides a 10% or more improvement over GBM, while being less abstraction, then that's reason enough to be having this discussion.

Also please remember that while Mesa and the Wayland dev's have years of display protocol experience, no one knows NVIDIA drivers and architecture like NVIDIA engineers.
Also AMD drivers are nothing to shout about on windows, even once they open source their Linux driver I can't imagine it'll be stellar.

[finalzone]

Holy shit, this thread is gold. All the NVIDIA hate going round is outstanding. Pure gold.

The only hate is Nvidia behaviour trying to impose their way

Might I remind some of you, that while Linux is a platform for open source evangelists, it is also used in working environments.....

We are talking about open standard where Nvidia can easily participate and contribute rather then reinventing the wheels like the company usually do. Nvidia had opportunity to preach by example but their greed blind them.

Now if what NVIDIA are saying is true, and that eglstreams provides a 10% or more improvement over GBM, while being less abstraction, then that's reason enough to be having this discussion.

Only claim from Nvidia without backing up. An excuse to avoid contributing in an established standard. Nvidia proposal is rightfully rejected because the company once again tries to keep everything for themselves while having access to the information for a long time and choosing to do nothing.

Also please remember that while Mesa and the Wayland dev's have years of display protocol experience, no one knows NVIDIA drivers and architecture like NVIDIA engineers.
Also AMD drivers are nothing to shout about on windows, even once they open source their Linux driver I can't imagine it'll be stellar.

The reasons is Nvidia want to keep those drivers for themselves (Gamework is a flagrant example of preventing AMD to compete at even playing field) while AMD released then and made them accessible to the public via gpuopen.com thus benefiting everybody.

The entire topic is about Nvidia once again proposing a duplicated effort only benefiting them without adapting to the already established standard.

[pal666]

so is surfaceflinger. let me know when AMD supports it.

amd supports it now. it doesn't need special support, because there is gralloc on top of gbm. so, now you feel yourself in one boat with "fuck you" nvidia

[chithanh]

And who is going to take care of that code after it has been implemented?

Community.

You expect the very community who has just rejected NVidia's code to be willing to maintain it?

This is not something unusual, this year AMD created AMDGPU driver, I think whole point of which is to transfer burden of supporting kernel module and xorg/wayland binding of AMD's proprietary driver to community.

That is not how it works.

And later maintainers of GNOME will be forced to merge this branch.
But I think propability of this development is low. My bet - after some grunt GNOME developers will implement this themselves.

This will boil down to the question whether GNOME needs NVidia more, or NVidia needs GNOME more. And even if the former is the case, GNOME developers have demonstrated in the past that they can stick with their principles and make unpopular decisions.

Holy shit, this thread is gold. All the NVIDIA hate going round is outstanding. Pure gold.

Have we forgotten Linus Torvald's middle finger against NVidia already?

It seems some people are still in denial about the reasons why that happened.

Might I remind some of you, that while Linux is a platform for open source evangelists, it is also used in working environments. Pretty much the whole VFX industry use Linux and NVIDIA. I know we do, so I don't care if the driver is open source, closed source or made by bill gates himself, I care that it works, its reliable and that it has great performance. It would be nice if it was open source, but when I'm 8 hours from a deadline I'll bank on reliability over that personal satisfaction of running open source code every single time. And for us, that's nvidia.

No Linux/Wayland/etc. developer owes you any gratitude for tolerating Linux at your workplace. Linux has become what it is today not because the community focused on getting market share, but because they focused on building the best operating system. And open source is the superior development model for that. You can go use Windows for all I (and many others) care.

[pal666]

It's always funny to read something about "you have to..." from FREE software evangelists

well, it was like "if you have 2 as first argument to addition and expect 4 as result, you have to pass 2 as second argument"
you are FREE to pass anything, but you get correct result only if you pass what you have to
what all of this have to do with free software?

[pal666]

According to a nvidia guy from mailing list, EGLStream variant is faster by 10% on a kepler GPU and can become even faster on future hardware.

it is same bullshit comparison as llvm vs gcc 4.2
nonextistent future version of eglstreams is faster than old version of gbm
how helplessly stupid you have to be to not consider possibility of making future gbm version just as fast?

[pal666]

So, the option for compositor developers are to support wayland and xorg forever or wayland+GBM and wayland+EGLStreams and drop support of xorg in 2-3 years perspective.

again example of idiotic conclusion from you. the option for compositor developers is to support wayland+gbm, x clients will be handled by xwayland forever because some clients will never be upgraded

[pal666]

NVidia has already implemented EGLImage/EGLStream on weston, maybe they will implement (or pay someone to implement) same codepath in one or two most popular DEs.

1) their "implementation" breaks wayland,
2) "maybe" doesn't count. maybe they should go fuck themselves
3) they have to implement same codepath in all des, not just to one or two
4) they have to maintain it afterwards
since nobody is going to believe them on point 4, it stands no chance