Re: [Fsuk-manchester] Open ARM GPU drivers

[Bob Ham]

On Thu, 2013-01-31 at 18:04 +0000, Dave Love wrote: 
> Michael Dorrington <address@hidden> writes:
> 
> > As you may or may not be aware, the state of free video drivers on the
> > rising ARM architecture, found in mobile computers, is not good.  This
> > includes the Raspberry Pi which has what the makers call a free video
> > driver but really it's merely a shim layer to the real driver¹.
> > Thankfully this is changing due to reverse engineering efforts:
> 
> I'd be interested if anyone knows about the possibility of using the
> drivers for floating point work, presumably via OpenCL.

Non-existant I'd expect.  The leading (only?) free software
implementation of OpenCL is Clover¹ which is still in its relatively
early days.  Most of the work is being done with Radeon chipsets,
particularly the r600g driver for the r600/r700.  There's a status
matrix here:

   http://dri.freedesktop.org/wiki/GalliumCompute

As far as I know, there are three main ARM graphics cores with projects
to develop a free software driver: the Mali, Adreno and PowerVR SGX.
The projects are, respectively, the Lima², Freedreno³ and.. err..
PowerVR⁴ :-) driver projects.  The Lima and Freedreno have some working
code but are still under heavy development and have some way to go to a
stable driver.  The SGX is still being reverse engineered and there is
no code yet.  (I'm actually involved in that project; if anyone can
donate some engineering skills to the effort, please pipe up!)

From what I understand, the Freedreno driver is the only one with a
working Gallium 3D backend, which would be required for Clover.  The
Adreno core is also actually an offshoot of the Radeon line (the name is
an anagram of "Radeon") so it's the most likely candidate to support
OpenCL in future.  I wouldn't hold my breath though; OpenGL support has
a much higher priority.


> If so, what
> sort of performance might Mali 400, for instance, provide?

Good question.  I can't imagine you'd get anywhere near the kind of
performance of a desktop graphics chip due to the power constraints.
That said, having seen some of the 3D games running on modern phones, I
wouldn't be surprised if the performance was above negligible.


> I couldn't
> find much relevant information even for proprietary drivers.

I'm not surprised; the focus of most efforts are intended to support
things like simulations, computational biology and other scientific
computing, often in supercomputers.  In that kind of context it doesn't
make much sense to target low power, performance-constrained chips like
the Mali.

However, there are many applications of parallel processing that OpenCL
opens up so I would expect to see common programs making use of it
before long.  In the proprietary world, I know there is a popular image
editor and a mathematical suite that both make use of OpenCL.

As the level of OpenCL support in applications approaches the level of
OpenGL support in games, I imagine we'll see more effort to provide
implementations for lower spec'ed hardware.  For now though, I'm afraid
I wouldn't expect a Mali to be doing much parallel processing.  (Unless,
of course, you're able to engineer your own floating-point applications
directly for the hardware.  In which case, forget about the Mali and
come and help us with the SGX ;-)

Regards,

Bob


¹ http://people.freedesktop.org/~steckdenis/clover/
² http://limadriver.org/
³ https://gitorious.org/freedreno/pages/Home
⁴ http://powervr.gnu.org.ve/

-- 
Bob Ham <address@hidden>

for (;;) { ++pancakes; }

signature.asc
Description: This is a digitally signed message part