VIA Releases New X.Org Driver

[phoronix]

Phoronix: VIA Releases New X.Org Driver

Back in April at the Linux Foundation Collaboration Summit, VIA had announced an open-source initiative. However, not all open-source developers have been pleased with these actions by VIA seeing as they have let down the community in the past...

http://www.phoronix.com/vr.php?view=NjY4NA

[stan]

it goes without 3D acceleration ... EXA acceleration ... TV support.

Since 3D and EXA were developed initially by the OpenChrome project, I think VIA plans to provide docs and invite those developers to merge their work into xf86-video-via. Hence no duplication of work To prove the point, check out the DRI/DRM patch that VIA already submitted to the Linux kernel, which includes a 395-line file called via_chrome9_3d_reg.h.

As for TV support, I agree, it would seems silly to put so much work into XV while not providing TV-out modesetting for HTPC setups like MythTV.

It's even admitted VIA has been trying to copy some aspects of AMD's open-source strategy

There's nothing wrong with "copying" a policy that is friendly towards Free Software users. In fact, VIA is going over and beyond AMD in terms of multimedia capability (XV acceleration), especially considering AMD's cozy relationship with the MPAA and dragging their feet on releasing a FOSS driver that accelerates video using the IDCT, UVD and UVD2 blocks.

With this code dump the VIA driver is likely to be the best FOSS multimedia driver ever to have existed.

Back in July though, VIA Technologies had appointed Heralt Welte as its open-source liaison

His name is spelled Harald Welte, and you can get a first-hand account from his blog.

[madman2k]

miachael, could you please stop your Ati fanboyism? Ati had basically the same problems creating the radeonhd drivers; it was duplication of work, since modesetting already worked with radeon, and up to now nobody knows where radeonhd is heading. So they are no better to Via in this respect, they just have Open Source developers blogging about what they are working on...

[bridgman]

madman2k, we needed to write new modesetting code anyways, since the display controller was totally different beginning with 5xx. At the time we decided to write a new driver, even the radeon devs agreed that made sense -- remember that Dave and Alex helped to put together the original plan.

The timing just sucked... Dave and Alex were also surprised how much the randr1.2 work cleaned up the driver - when they removed the pre-RandR support they were able to delete pretty much all of the old code. If I had talked to them in August or September rather than May/June we probably would not have started a new driver.

One thing I have learned over the years is that almost everything is more clear a year *after* the decisions were made

[madman2k]

yeah, I basically know that. Its that Via might be also making decisions based on false assumptions, just like what happened to you. And therefore one should not put them down like Michael did.

[bridgman]

Yeah, it's tricky to get the right message out. I think Michael is trying to remind everyone that there have been false starts before.

Harald seems like a pretty straight shooter though...

With radeon/radeonhd, it wasn't even false assumptions -- we just failed to consider something which hadn't happened yet

[ErikAndren]

Originally Posted by bridgman

Yeah, it's tricky to get the right message out. I think Michael is trying to remind everyone that there have been false starts before.

Harald seems like a pretty straight shooter though...

With radeon/radeonhd, it wasn't even false assumptions -- we just failed to consider something which hadn't happened yet

Bridgman: so why is the code duplication madness with the radeonhd and radeon still continuing until this very day? Why not merge everything into one driver? Isn't this going to happen some day in the future?
Especially as the radeonhd has adopted using atombios these days.
Talk about wasted effort in the long run...
Seems like petty politics in my eyes.

Regarding Via:
I think this is a great move. If via and Harald are really smart they should support the xorg project like intel does, with an active developer group, developing everything in the open, filing bugs in the official xorg bugzilla and announcing their driver roadmaps. This will surely increase a their value, especially in the embedded / low-power market where via needs to show activity in order to stay competitive with the intel atom systems.

[Dieter]

> It's even admitted VIA has been trying to copy some aspects of
> AMD's open-source strategy

Are you implying that this is a bad thing?

> and even X-Video

What about XvMC ?

> but it goes without 3D acceleration.

Certainly we'd like to see this added eventually,
but many applications don't need 3D acceleration.

> The driver also goes without TV support.

This needs to be fixed.

> but it isn't all new code

So?

> Some portions of the code looks like it may be derived from its
> closed-source X.Org driver for VIA hardware.

Closed-source becoming open. Isn't this what we want?

> What isn't clear though at this time is whether they intend to
> develop this driver openly or what features they intend to
> provide with this driver.

Yes, a roadmap would be good.

> What is also a bit disheartening with this code drop is that they
> apparently haven't worked with the community-spawned Unichrome or
> OpenChrome projects.

> The OpenChrome driver is open-source and even has 3D support.

What is missing from OpenChrome? Perhaps we need a chart showing
which features OpenChrome supports and which features xf86-video-via
supports.

As an end-user, I would like to see one FLOSS driver that provides all
the features for a particular GPU. If it makes sense to have different
drivers for different GPUs, that's fine. But I don't want to have to
choose between feature A and feature B. This goes for AMD/ATI as well.

[hmmm]

http://laforge.gnumonks.org/weblog/2...opensource-faq

Harald has updated his blog to answer some of the faqs related to the driver:

"There have been numerous questions regarding the recent open source release of VIA's 2D Xorg driver.

Why did VIA publish yet another driver, rather than improving any of the existing Xorg/openchrome/unichrome drivers?
Because this driver is all but new! It was the base for all the binary-only driver releases that VIA has made (and is still making) for select Linux distributions. So rather than having written a new driver, this is just the disclosure of an existing driver.

One of the commonly asked questions is _why_ not the complete source, including codec acceleration, TV out and 3D was published. I cannot disclose the particular reasons for VIA, sorry. But I can comment on the general reasons on why companies cannot disclose certain source code. As you may have noticed, the situation with regard to the ATI driver e.g. shows certain similarities.... Usually there are some parts of the code, particularly for the 3D driver, which cannot be disclosed due to either

* parts of the source code are under a proprietary license from a 3rd party
* parts of the source code refer to technologies (e.g. macrovision) which are subject to very strong NDA's by the licensor, which in turn prohibit the open documentation or distribution in source code form

Will VIA learn to build a community around that new driver? Will there be mailing lists and a public revision control system?
As of now, this is unlikely. Not because VIA doesn't believe in the community, but rather because the disclose of VIA's source now enables everyone involved to look at all the available drivers. Some consensus has to be found on which driver is best to be used as a base for a future Xorg mainline driver, and then the community and VIA can work together on merging bits from other drivers into that base. Creating VIA's own mailing lists (and community) would lead to more fragmentation, rather than unification.

"

[marshtric]

So does anybody know if they've fixed the problem with Neverwinter Nights that the previous driver had? There's no point me upgrading from 4496 until there is a version which actually works with the only OpenGL program I regularly run, and they have no recognisable message about this in the README...