[libre-riscv-dev] Questions The Libre-RISCV SoC

[lkcl]

ok so as you can see, michiel got back to me, so i propose to reply as
below.  we agreed a cutoff of 48-72 hours, i'd feel more comfortable
if we got back to michiel within under 48, if practical.

given that this could turn into an ongoing discussion (the agreement
of milestones, below), we may have to rethink.  first bit, though:
please do review below, comment *above* (not inline) and let me know
if it's ok.


On Thu, Mar 21, 2019 at 7:32 AM Michiel Leenaars <michiel at nlnet.nl> wrote:
>
> Dear Luke,
>
> just a short clarifying question: while the larger project expects to
> have the outcome of a fully-functioning quad-core 800mhz RISC-V, the
> effort we would fund would basically cover the FPGA pre-implementation?

 yes.  i have a potential client for whom the milestone of achieving
an FPGA implementation is a key trigger for further investment.
NLnet's help would be a welcome financial bridge.

> Or are there additional results included in the project you can commit to?

 we have received an additional offer (separate from the potential
client) of a donation sufficient to cover the cost of MVP
(Multi-Vendor Programme).  if it is practical for us to move forward
to doing the actual ASIC layout using libre VLSI tools, we will start
that as well.

 however it's not something that we can definitively commit to, as the
layout is itself a massive task, as is the task of reaching an FPGA
pre-implementation.

> I assume you are aware that we would pay progressively as milestones are
> reached.

 appreciated.  then we will need to come up with some suitable
milestones, otherwise there would be only the one "end" milestone
(FPGA demo) and it would be rather... challenging.  off the top of my
head i can think of the following tasks:

* an IEEE754 floating point pipeline
* a basic processor core (we need a management / low-power-mode / boot-up core)
* a nmigen pinmux (for peripheral multiplexing)
* a Virtual Memory Management system suitable for hybrid CPU / GPU usage
* a software simulator compliant with the latest SimpleV Vectorisation
Standard proposal
* design and software simulation of some of the 3D custom instructions
needed for the GPU
* Kazan 3D's software SPIR-V to LLVM compiler being operational and
running OpenCL applications (although this is also a massive task so
may need further subdivision)

there are actually a significant number of sub-tasks, many of which
are already underway and/or have been planned in advance: we can track
them using http://bugs.libre-riscv.org if that would work?

http://bugs.libre-riscv.org/buglist.cgi?component=Milestones&list_id=40&product=Libre%20Shakti%20M-Class&resolution=---

> Is Jacob aware of the project proposal?

 yes: as is the whole team. as mentioned earlier we discussed the
proposal and agreed how to respond when you got back to us:
 http://lists.libre-riscv.org/pipermail/libre-riscv-dev/2019-February/000625.html

> How do you see the division of
> funding between him and you, and others?

our Charter requires that everyone is equitably rewarded for their
contributions:
https://libre-riscv.org/charter/discussion/

so we will strike a balance between the individual needs of each team
member and how their contribution, if encouraged and so empowered,
will help move the goal forward.

these are for full-time focus:

* i need around USD $1200-1500 per month for living expenses
* jacob needs around $1000 per month.

we've not yet had anyone else come forward to formally discuss any
financial needs in exchange for an agreed contribution.  we do have
the option of allocating shares (yet to be fully discussed, i am
leaning towards http://slicingpie.com as it is about the only fair
equitable startup calculator available).

i have also contacted the nmigen developers and provisionally offered
them a donation, as we are taking up their time with support
questions, and using code that they spent time and resources
developing, so it is only fair and reasonable to make sure that
they're fairly compensated for their efforts.

would that be something that could be arranged and discussed?  if it
cannot be arranged through NLnet i feel strongly obligated to arrange
a donation to the nmigen developers through some other means.

also, since writing the proposal we have located a suitable low-cost
FPGA developer board (around $200) called the ECP5 that happens to be
compatible with libre-licensed toolchains.  at the appropriate point,
a case could be made to get some of these for the team.  although they
are smaller than the resources needed to run the full FPGA core (for
which I have the Xilinx ZC-706), the ECP5 will manage sub-tasks very
nicely.  and, crucially, will not need payment of FPGA SDK licensing
fees.

l.