[Tinkerphones] GTA04A5 and News from the Kernel

H. Nikolaus Schaller hns at goldelico.com
Fri Oct 28 08:03:11 CEST 2016


> Am 27.10.2016 um 22:24 schrieb Paul Boddie <paul at boddie.org.uk>:
> Cutting the GTA04 list because it isn't directly relevant, although I'm not 
> really sure of the relevance here anyway...
> On Thursday 27. October 2016 19.43.27 H. Nikolaus Schaller wrote:
>> Now back to the topic of FLOSS. This driver is of
>> course free. But it will need / load firmware binaries
>> into the GPU part of the SoC. These binaries are
>> on one hand some "microkernel" running on the SGX
>> GPU and the ARM libraries for OpenGL etc.
> It was interesting to finally get hold of a programming manual for a 
> completely different SoC - the Ingenic JZ4780 - which is also burdened with a 
> PowerVR implementation (SGX540). Unfortunately, there are no details of how 
> the 3D functionality of the GPU is accessed, in contrast to all the register 
> descriptions for that SoC's 2D functionality (X2D).

Yes they are quite similar. Imagination sells for example a licence for a big
sgx530.vhdl and the SoC vendors (TI, Intel, Ingenic, whoever) combine this with
some arm.vhdl into a single chip.

So I thin an sgx530 is almost the same on all SoCs. And a sgx540 as well, but
differs from the sgx530 in details.

Well, there might be parameters (sort of #define) to control how many shaders
there are, how deep the pipelines are etc. So there might be some small differences
even for the same sgx release.

How they are connected to the CPU is probably predefined by the sgx530.vhdl
as well. Most likely they provide an address and a data bus for a shared RAM.
This can be written by the CPU block with code for "download" and one register
has a special meaning to trigger commands on the GPU side. What will differ
is how this register block is mapped into the address space.

The real problem is that the firmware (microkernel, libSGX) and the chip releases
are so tighlty connected with each other so that an exchange between SoCs is
impossible. So we have a rev121 for the OMAP3530 and a rev125 for the DM3730
and they are not compatible. Except on OpenGL side.

> Given what was said about the Broadcom SoC used in the Raspberry Pi, the 
> claims of "fully free" drivers and yet the continued presence of a non-free 
> microkernel, I can easily imagine the SGX functionality being rather similarly 
> structured, regardless of the larger SoC being ARM-based (like the DM3730) or 
> MIPS-based (like the JZ4780).

>> So thats it for today. And I hope I can bring some
>> news about the GTA04A5 tomorrow.
> I hope it will be good news!
> Paul
> P.S. You mentioned another phone concept on the list a while back that was to 
> use one of the AM series of TI SoCs. Are there any more details on this yet?

It is nothing more than a dream/idea around this module


The key issue is again getting another hardware development project funded if
it is already difficult to finance the current ones.

IMHO there are some things to consider:
* like the DM3730 the AM3358 is "just" a single core Cortex-A8. Which is good
  but the DM3730 is much smaller and is better optimized for handheld operation.
  So the AM335 or OSD335X is just a little easier/cheaper to manufacture and has
  no other benefits or additional functions. I don't even know if it is more
  power efficient.
* Generally, with each Apple/Samsung/LG/Google/Lumia announcement of new smartphones
  it becomes more and more difficult to offer something that is urgently needed
  and nowhere else available at lower price. Except the ability to tinker with it
  and the long-time hard&software support.
* Crowdfunding has become too mature and there are too many platforms competing
  for projects and donators to really work well. Crowdfunding needs to have almost
  finished development.


More information about the Community mailing list