Mesa Developers Discussing Again Whether To Fork Or Drop Non-Gallium3D Drivers

[TemplarGR]

I think that there's a balance between supporting new and legacy hardware. In this case, Intel and AMD employ developers to work on Mesa. It costs both companies money and developer man-hours to support hardware that no longer generates revenue. While I don't think that hardware should be abandoned immediately after a company moves to the next revision, I think a period of support should be defined. I'm not the person who should define the length of this period, but I think 5 years is a good term. If people still want to use that hardware after that period, it should be left to the community to support it.

It is not as simple as specifying a flat date. It depends on many things. Older hardware, if the drivers are well written and mature, does not need much maintenance to be kept up-to-date. The reason Haswell igpus are in danger of losing support is not because they are weak, but because they don't use the gallium infrastructure. I am willing to bet that if Haswell was on Iris, it wouldn't be that hard to keep receiving support, after all the vast majority of fixes for gallium would apply to that arch as well. That was the point of using gallium in the first place.

So it all depends on how hard it is for Mesa devs to keep including older hardware. Gallium ensures easier and extended support. But putting an arbitrary 5 year period of support and then "you are on your own" would be a mistake, now that Moore's law is dead 5 year old hardware can easily keep up with modern hardware in terms of raw power in most cases.

[artivision]

Well people there is no reason for one Linux for all. We should have Legacy and Bleeding Edge versions. For Example a Clear AVX2 only for Vulkan compatible Gpus only and no WineD3D at all.

[s_j_newbury]

The drivers are all open source. Anyone is free to step up and write a new driver or help maintain the code, etc. It's supposed to be an open source community.

I guess people might have time on their hands in the current climate. I'm happy to contribute to bringing the i965g driver up to date, but I'm not the right person to lead the project... I take things on, then wander off... especially if I'm the only one working on it.

So, if anybody is serious, please speak up.

[duby229]

And there was near to no change for these systems in mesa for quite some time. So it's not like you will miss something by using a mesa-legacy driver.

That's just not true. Let me ask you, do you have any experience maintaining out of tree drivers on linux? Let me just assure it's basically and pretty much always impossible. Unless you also keep older kernel, glibc, xserver... The number of serious examples of this is just too numerous to count. How long do you suppose older mesa versions will ontinue to compile successfully? And even if you can get older mesa to compile how long till it just doesn't work anymore? I give it a few releases at the very most optimistic guess. One or two years at maximum and then all those older GPU's are fucked.

If non-gallium drivers get dropped from mesa then users of those GPU's will have no choce but to downgrade to a distro that won't be upgraded past the point of the dropping. And that's literally an insane proposition.

[duby229]

LOL, in your dreams, perhaps. Even the best APU in stores today can barely catch up to a 100$ dGPU (for example, AMD R7 260x) from 2015 in performance. In 2014-15 we had stuff like 7970 and Fury, and no APU will touch those in some time to come... The main issue is VRAM bandwidth, and until this gets solved by some L4 cache or HBM solution, APUs will be left to play simple/old games and display GUIs, that's it.

As for old hardware, the issue is not that the hardware is old, the issue is that Intel were stubborn for years and refused to switch to the gallium infastructure for their drivers until recently. AMD has a gallium driver for even the R300 architecture, which is approaching 2 decades of existence...

I disagree with creating a MESA LTS version. For many reasons. One is that LTS usually means "only bugfixes allowed" but there could be feature work that could apply to Classic drivers in the future, so why call it LTS? And another reason is that we don't need another kernel situation where we have 5-6 different MESA LTS versions, LOL.

I think what MESA should create is a MESA CLASSIC version, and put all the classic MESA drivers there. People with classic MESA drivers will be installing the Classic package, and people with modern gallium drivers will be installing the normal MESA package. Seems clear to me.

Except that won't work. Glibc, xserver and many other components don't have a stable interface and a "classic" mesa will be broken in just a year or two at best.

[duby229]

Well sure, but those systems are not getting that much new attention anyway. The support is pretty good, and that won't change just because the long-term drivers are sitting in a different repo.

If anything, this lessens the chance that the people working on Gallium drivers will inadvertently break something in classic Mesa.

Yes, but things glibc and xserver change frequently. A fork of mesa wouldn't even compile in a year or two at best.

[s_j_newbury]

Yes, but things glibc and xserver change frequently. A fork of mesa wouldn't even compile in a year or two at best.

duby229, this is exactly my concern. I also happen to use the IVB Vulkan driver, but that was never claimed to be officially supported, so it's a slightly different situation. There is no way a "Classic" Mesa will be kept up to date outside of a community effort, which usually boils down to one or two people who really want to keep their system working, until they get bored or upgrade, or something happens to them.. like they get a job! ;-)

Look at my Beignet OpenCL example above, it's maintained outside of Intel by a single Debian developer, who keeps it working on Debian. It's currently broken on Gentoo, so it's even getting dropped there, despite the fact I'm trying to maintain it, but can't get it to successfully build!

[TemplarGR]

Except that won't work. Glibc, xserver and many other components don't have a stable interface and a "classic" mesa will be broken in just a year or two at best.

Not really. Stop spreading FUD.

[maxtertheturtle]

It is not as simple as specifying a flat date. It depends on many things. Older hardware, if the drivers are well written and mature, does not need much maintenance to be kept up-to-date. The reason Haswell igpus are in danger of losing support is not because they are weak, but because they don't use the gallium infrastructure. I am willing to bet that if Haswell was on Iris, it wouldn't be that hard to keep receiving support, after all the vast majority of fixes for gallium would apply to that arch as well. That was the point of using gallium in the first place.

So it all depends on how hard it is for Mesa devs to keep including older hardware. Gallium ensures easier and extended support. But putting an arbitrary 5 year period of support and then "you are on your own" would be a mistake, now that Moore's law is dead 5 year old hardware can easily keep up with modern hardware in terms of raw power in most cases.

I agree that 5 years is too short; however, I think that 20 years is too long. Intel should honestly update Haswell so that it uses the Gallium infrastructure.

Unfortunately, it will never be about the architecture of that hardware. Even as code matures, it still needs to be tested as it's dependencies change. If Intel is no longer selling Haswell iGPUs, there's no incentive to provide support and that hardware will end up needing support from volunteers.

[duby229]

Not really. Stop spreading FUD.

Yes, really. Go grab an old version of mesa right now. Do it. See how far back you can go. Maybe a year or two at best. It would have to be fixed every single time any single one of its dependencies got updated.

Examples include but are not limited to beignet, fglrx, rtl81xx, etc...