[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: ARM GPU non-free drivers



+++ Luke Kenneth Casson Leighton [2010-11-22 03:56 +0000]:
> On Sun, Nov 21, 2010 at 9:55 PM, Wookey <wookey@wookware.org> wrote:
> 
>  apologies - subtly and tact have never really been my strong points.

Quite. Even amongst other free software peoele not know for their tact
you stick out :-) Do try to think for a moment next time. Avoiding
gratuitously pissing people off can be helpful sometimes, especially
in a case like this where we are trying to persuade them towards a
course of action. 

>  but i actually much much prefer what you've written here, to what i
> have: it's definitely clarifying matters and puts things into
> perspective in a much better way than i ever could!

Softly softly catchee monkey.
 
> > I see strong hints that at least one GPU vendor will provide design
> > info soon so that 3rd party drivers can be written.
> 
>  realllyyy? veery goood.

Well that comes across a little optimisitic actually (I'm an optimist
:-). When I say 'soon' I have no real idea of timescale, I just think
that one of them will crack and open up at some point. The farcical
idea that GPUs are too special to have open specs like other hardware
is rapidly being exposed as a naked emporer. That just leaves a forest
of crappy patents, bought-in 'IP', and worried lawyers to negotiate.
All the engineers involved were persuaded a long time ago.

> > They are too
> > scared to hand over their own code, but seem OK with 3rd-party free
> > drivers being written.
> 
>  could you, privately, let me know which one it is?

Nothing that concrete yet.

Listen to 22:20 to 29:20 in
https://wiki.linaro.org/Releases/1105/PublicPlanReview?action=AttachFile&do=get&target=graphics-11.05.ogg
to get an idea of the current status that someone involved feels able
to talk about publically.

So that's absolutely no promises at this stage, but gives an idea of
what is happening in terms of discussion within and between companies
about how they can open up a bit and maybe follow the ATI or Intel
model of providing GPU ISA specs (for previous-gen hardware), whilst
doing their best not to explain the microarchitecture and thus best
pipeline optimisations. 

We do have the advantage that there are 5(6?) vendors to play off
against each other. We just have to persuade at least one of them that
there is commercial advantage in being the one with free drivers
(because it helps get design wins). The problem here is that their
direct customers are SOC desigers, for whom free drivers is a pretty
minor consideration. The advantages comes further down the line when
you get to the point of trying to get people to use your soc in
linux-based designs - and then there are a lot of considerations to
mix in beyond the GPU library situation when choosing a device. You
and I would put free drivers for all the hardware right at the top of
the list but less dedicated individuals might choose performance or
chip cost first (sometimes due to failing to appreciate the long-term
implications of choosing non-free stuff). 

>  i know that samsung's S3C6410 GPU engine is being reverse-engineered
> from two different angles, the linux angle _and_ the wince angle, but
> i'm seriously seriously trying to find a decent Cortex A8 CPU with a
> _remote_ chance of a free software 3D stack.

Right now your guess is about as good as mine. But the
reverse-engineering option is an important one which we should persue
too, as the laborious plan B. 

Wookey
-- 
Principal hats:  Linaro, Emdebian, Wookware, Balloonboard, ARM
http://wookware.org/


Reply to: