The ati petition

Lee Revell rlrevell at
Fri Sep 3 00:10:52 PDT 2004

On Fri, 2004-09-03 at 01:48, Vladimir Dergachev wrote:
> Hmm.. IMO, it is rather ill-conceived. One could achieve more by asking 
> less.

I have to say I agree.

> For example:
>        "The signers of this petition will no longer purchase ATI products 
> until ATI make a workable 3D accelerated 64-bit driver for Linux and a 
> 32-bit that utilizes the full potential of the Raedon chips."
>         Linux (and BSD) is not just about ia32/ia64. We should be asking
> for a way to utilize full graphics and video capabilities on any platform.
>        "In the event that ATI cannot handle the programming of these 
> drivers, we require all modern ATI chips to be open sourced to the Linux 
> community so that drivers can be made. "

Don't hold your breath.  This would be a reasonable request if any 3D
hardware vendors released open source Linux drivers, then someone at ATI
could at least point to another company that this decision made sense
for.  But AFAIK there isn't an example of this.  Management is
conservative, and for very good reason - if they opened their source and
their competitors used this information to gain an advantage, it's
management's ass in a sling.  If you at least have an example you can
point to, it's not as bad, but from ATI's perspective it's a big risk
for nebulous gain.

VIA has open source 3D drivers, but currently the hardware isn't nearly
as good as ATI and Nvidia.

>         It is not necessary to have full source code to the chips 
> themselves to write a driver, rather just a good description on how to 
> *access* the chip. (In case you don't know chips are designed using VHDL 
> or Verilog - programming languages, so designing a chip is somewhat 
> similar to writing a paper in TeX).

Yup, often a single header file with comments is enough.  All you really
need to know is what to write to what register to do what.  Of course
most of my driver experience is with sound cards which are probably a
good bit simpler.  But for most of the cards ALSA supports, you could
write the driver starting with just the header file.

>         More importantly, the petition tone is antagonistic, as it 
> implicitly asks for ATI to surrender to its demands. This is pointless
> as even if this goal were achieved it would have no bearing on a technical 
> issue - getting ATI graphics cards well supported in Linux.

Agreed, this is the worst possible approach.  The only form of democracy
that hardware vendors understand is voting with your wallet.  As long as
there is one vendor that releases source to their 3D drivers, buy that
hardware instead of Nvidia or ATI.  If you actually need good 3D, buy
Nvidia, and recommend other Linux users do the same.

If it turns out that the Linux 3D market is big enough to justify vendor
support, then this will affect them enough to change their behavior.  If
not, then the Linux 3D market just isn't big enough, and your time would
be better spent fixing this by making Linux a better 3D platform.

>         I strongly suggest the writer of the petition to read through (if 
> he or she has not done so already) the following link:
>         Also, it is important to note that Precision Insight (and, in 
> particular, Frank La Monica) had big problems persuading ATI to release 
> specs to allow 3d support for Rage128 and Radeons and undoing the history 
> of angry e-mails demanding documentation and flat refusals to provide it.
>        It would be much better to petition ATI for something to cooperate 
> with - for example a standardized way to encapsulate driver in 
> architecture-indepedent code so they can keep their proprietary stuff, but 
> let open source handle the rest.

This at least has a reasonable chance of succeeding because they can
point to Nvidia as a precedent.  However, I suspect the reason Nvidia
was able to release Linux drivers so easily is because they wrote a HAL
at some point that lets them write the same driver for all platforms,
they just had to port the HAL to Linux.  If ATI's code is not so clean
(not unlikely) then this could be a major undertaking.


More information about the xorg mailing list