[libre-riscv-dev] PowerISA, NLNet grants

[Luke Kenneth Casson Leighton]

On Sat, Jan 18, 2020 at 8:54 PM Staf Verhaegen <staf at fibraservi.eu> wrote:

> All,
>
> First as announced in other mail I am funded by NLNet to do a first
> 0.18um test tape-out. In the end I will tape out what is decided here
> on the list what will be designed and layouted.
>

fantastic.  much appreciated.


>
> That said, I join the thought that going for the PowerISA would just
> reduce the reach of LibreSOC a lot and I don't see it getting the
> wanted traction in that way.


this was discussed back in october, and outlined in the CS update.  we're
not "only" doing PowerISA.  we're doing user-space RV64GC.  see the last
few paragraphs, here:
https://www.crowdsupply.com/libre-risc-v/m-class/updates/nlnet-grants-approved-power-isa-under-consideration


> AFAICS this is all about the Risc-V trademark where the bylaws are not
> fit for community driven projects.


this is not entirely accurate, it is however a step in the right
direction.  to test and understand this, take the following action:

(1) https://riscv.org/membership-application/ - review the RISC-V
Membership Agreement (in particular section 5)
(2) note the Trademark usage here https://riscv.org/membership-application/
(3) try to find the public archives for the following:
https://riscv.org/mailing-lists/


> It only consider lone hobbyist or
> corporates; not a libre development community.
>

correct.  and we are funded by NLNet, VERY SPECIFICALLY under their PRIVACY
and ENHANCED TRUST programme:
https://nlnet.nl/PET/

we are in the position of not only being a libre community, we are a
*commercial* libre community, and, more than that, we are a commercial
libre community that is expected to operate ENTIRELY transparently, in
order to meet the promises and obligations that we made to NLNet.

how can we meet those promises to be fully transparent at all times, and at
the same time sign an NDA which forces us into secrecy?

In parallel one can do lobbying to get the Risc-V foundation bylaws
> more fit for libre community driven development projects. And with
> lobbying I mean talking off-line with people who are part of the Risc-V
> foundation. I don't think you will achieve this with fighting a war on
> a maillist.


i'm not (is anyone else?)  i am also no longer focussing any energy on the
RISC-V Foundation or interacting with them in any way.


> I have also connection there and will do some talking when
> I meet these people.
>

before doing so can i please ask you to do the following:

(1) decide if you would like to spend the time focussing on this.  you may
find that, once you understand the issues, you do not wish to, and that's
fine
(2) *fully* understand the issues *before* talking to anyone.


> Also the final high volume tape-out could also be done by an entity
> that than conforms with the trademark rules so the chip can be marketed
> as Risc-V.
>

ok, so let's walk through what conformance with the RISC-V Trademark
entails.

(0) firstly, you appreciate - i have to say this - that the standard RV64GC
ISA simply does not make a GPU.  also: RVV also does not make a GPU (it
gets about 25% performance/watt, which is commercially unacceptable).

i have to say this because whilst it may be obvious to experienced people
that a scalar ISA is inadequate for use in a GPU, it may not be obvious to
all readers.

secondly, we are doing a *hybrid* CPU-GPU.  not a "CPU with a full custom
GPU connected via a shared memory bus", the CPU *IS* the GPU.

as a Hybrid CPU-GPU we *NEED* to augment the RISC-V ISA, and we need to
augment it in a massive fashion, well beyond what even RVV does.

therefore, we are *NOT*, repeat *NOT* just in a position of "implementing
and complying with RISC-V Trademarked pre-existing Standards, as set and
defined and dictated by the RISC-V Foundation".

we are in the position of having to innovate - and not just innovate,
innovate in a FULLY transparent fashion.

(1) it entails signing the RISC-V Membership Agreement (which, due to the
conflict of interest with NLNet Funding, and with our business objectives -
yes i said BUSINESS objectives - this is not possible)

(2) to be useful as a MASS-VOLUME hybrid CPU-GPU, we need PUBLIC and
MASSIVE augmentation of the RISC-V ISA.

note that "custom" extension status DOES NOT MAKE THE GRADE.

custom extensions are designed for PRIVATE use.

* the toolchain is PRIVATE and maintained by the proprietary vendor.
* the custom extensions are PRIVATE and maintained by the proprietary vendor
* the custom extensions CONFLICT with other custom extensions and, if two
of them are made public, would cause absolute total chaos.

(3) if through stone-walling the RISC-V Foundation refuses to allocate us a
block of instructions for PUBLIC and OFFICIAL use, this constitutes "delay"
under Trademark Law, and, thanks to the length of time and the consistent
and persistent refusal to acknowledge reasonable in-good-faith requests,
they actually lose both rights of enforcement AND the actual Trademark
itself.

the thing is: even though there is more than enough evidence such that we
would quotes win quotes a legal battle, the destruction and invalidation of
the RISC-V Trademark would be so catastrophic to the entire RISC-V goals
that it would be a "Pyrrhic Victory".

(4) therefore, if we cannot use custom extensions, and the RISC-V
Foundation refuses to include us and refuses to acknowledge the unique
circumstances that they never envisaged would happen (a libre *business*
that has full transparency), it is not only not a good idea for us to
augment RISC-V, it would be IRRESPONSIBLE of us to try.


This is similar to Linux, that also happened to be (mostly) API
> compatible with UNIX but was not called UNIX because of the trademark.
>

as you can see, it's much more involved than that.

l.