[Arm-netbook] Libre RISC-V -- I mean OpenPower M-Class GPU update

[Luke Kenneth Casson Leighton]

On 1/1/20, Paul Boddie <paul at boddie.org.uk> wrote:
> On 2020-01-01 22:09, Luke Kenneth Casson Leighton wrote:
>>
>> the collaboration that is in place, and which is otherwise
>> successfully taking place, is taking place behind *closed doors*,
>> where we, as Libre Businesses, are told, basically, "sign this
>> agreement which entirely compromises your business model, or go fuck
>> yourself".
>
> I haven't followed what goes on with RISC-V licensing, but it rather
> sounds like there are vested interests who want to retain control and to
> leverage their inherent advantage as developers of the technologies
> concerned.

this resonates with the behaviour i've seen.  it's also that, well,
they just don't care.  i've witnessed around a dozen people, now,
including several professors from highly-regarded world-renowned
universities, just "give up" and walk away from the mailing lists.

on the other hand when an employee from qualcomm innnocently /
accidentally cross-posted a question onto the isa-dev mailing list,
warnings were issued "there's more at stake, here" which basically
implied, "shut the fuck up you moron, we want qualcom's money: your
answers, questions and honesty are causing them to become extremely
nervous about using RISC-V".

then, there is the weird behaviour of SiFive, with the promise of
"fixing the problems of other ISAs" then rushing ahead without proper
consultation and "laying down the law".

> This is a pretty familiar story from the world of standards,
> even Internet standards, where companies effectively "front-run"
> standards by loading them up with descriptions of their own
> technologies, effectively requiring standards implementers to play
> catch-up all the time.

yyyup.  that's precisely and exactly what's happened.

> This latter strategy is not quite the same thing as, say, people being
> gatekeepers to a CPU architecture, but it is part of a more general
> phenomenon of stacking the odds in one's own favour and disadvantaging
> outsiders. A more mundane example would be any one of numerous
> corporate-run Free Software projects that demand copyright or
> comprehensive licensing assignments from contributors, and who make it
> tediously difficult to get code upstream, all because the company's
> needs supposedly override all others.

yes, this is uncalled for.  however, as we've learned from the EOMA68
Certification Mark, having a *Certification Mark* behind a standard, for
the protection and safety of users, turns out to be absolutely essential,
*even if* the source code is GPL'd / Libre-Licensed.

i would not have a problem with the RISC-V Foundation if they actually,
properly, truly understood Trademark Law and their legal obligations
and responsibilties.

>> by complete contrast...
>>
>> the OpenPower Foundation's Director, Hugh Blemings, is someone who has
>> worked with Libre Developers (he himself is one) for over two, nearly
>> three decades.
>>
>> *he* was the one that told *me* that the OpenPower Foundation Members
>> have created a Membership Agreement that is specifically designed to
>> allow Libre Businesses to be "happy" with its terms and conditions.
>
> Maybe ARM's success is a corrupting influence when companies and
> organisations try to monetise hardware architectures, but ARM only got
> into its lucrative position through a combination of good luck and a
> fervour for licensing things,

only because they were desperate!  see below... :)

> the latter only ever coming about as some
> kind of correction for ARM's corporate predecessor's obsession with
> keeping everything proprietary and trying to use such proprietary
> technologies to their own exclusive competitive advantage.

funny story: my friend barry worked for LSI Logic, with norman wilson.
the two of them, how can we put it diplomatically...  LSI's goal was,
um, "to make the customer's design actually work".

undiplomatically: ARM's first processors were non-working piles of f*****g s**t.

after they got the ARM7 functional, barry managed to get ARM their
very first license, ever: with Plessey.  they were so happy, they
offered him a job.  barry turned it down: he would have been employee
number 12, and a very rich man, now :)

but ARM - aka ACORN RISC machines (not Advanced RISC Machines) -
basically had zero cash, and at one point was completely unable to pay
its employees (this is like... early 1990s).  they licensed the ARM11
design to Intel for GBP 100,000, unrestricted, royalty-free, in
exchange for a promise from the team that IBM had bought (the DEC
Alpha developers), would "fix all the problems and give the changes
back".

the DEC Alpha developers - whom Intel themselves didn't know what to
do with, so gave them the PXA Project to do - took one look at the HDL
and went "holy f*** this is s**t" and started again from scratch.
they could do so because they had that GBP 100,000 royalty-free
license from ARM, and the contract wasn't worded carefully enough.

thus, the PXA 2xx series became the world's first superscalar
ARM-compatible architecture... *not* the ARM Cortex A8 as ARM keeps
telling everybody :)

several years later, ARM calls up Intel and says, "um... you've been
selling PXA SoCs for some time now, um... where's the source code?"
and of course they replied, "oh, the ARM11 was so shit we ditched it
and started again.  we never actually made any modifications to the
*ARM11* codebase, check the wording of the contract, bye, have a nice
life".

the irony is that the PXA series was so much more powerful and so much
more power-efficient than the Intel Atom processors available at the
same time, the DEC/Alpha aka PXA team was PROHIBITED from releasing
Intel PXA SoCs running above 600mhz.

of course when Marvell bought the PXA designs (minus the DEC/Alpha
team that developed it), they made absolutely no changes whatsover:
all they did was immediately crank the clock rate up to 1.2ghz and
that's what you find in the OpenRD/Ultimate and the Sheeva Plugs.

ARM's *real* story is littered with embarrassing stumbles from one
architectural mess to another.  they've never *actually* designed a
successful processor in their entire corporate history!

the ARM7 was a mess that LSI Logic had to diplomatically fix

the ARM9 was something that was developed by someone else and offered
to them (i don't know its history).

the ARM11 was developed by a Kiwi who just turned up one day, no
interview, nothing, just sat down at a desk and started working.  when
managers asked him if he actually wanted paying, he said, "no don't
worry about it".

even the Cortex series was bought from a failed (independent)
VC-backed firm that, due to patents and encryption, couldn't work out
how to get THUMB interoperability.

it's a cluster-poop, completely different from the image of the
"highly successful advanced innovative company" that they are today :)

of course *now* they can offer 7-figure US salaries and incentives to
the India Shakti Team lead developer, but they certainly couldn't have
done that even 10 years ago.

> It will be
> interesting to see how the different initiatives (RISC-V, OpenPower,
> MIPS...) evolve to respond to openness concerns and the need to
> cultivate interest in their offerings more generally.

IBM and NXP/Freescale have to be much more careful.  they are patent
heavy-weights, and you've heard the phrase, "do not annoy the 800lb
gorilla".

IBM takes its responsibility as a world-leader extremely seriously,
and they're providing an extremely sneaky royalty-free patent license
for POWER ISA, which says, "you get free unrestricted use of these
patents as long as you don't ever try to assert a patent - ever -
against us, IBM".

*that's* clever :)

l.