[Libre-soc-dev] idea: get grant from rust foundation to add powerpc/svp64 backend to cranelift

[Jacob Lifshay]

On Tue, Feb 21, 2023, 02:34 Luke Kenneth Casson Leighton <lkcl at lkcl.net>
wrote:

>
>
> On Tuesday, February 21, 2023, Jacob Lifshay via Libre-soc-dev <
> libre-soc-dev at lists.libre-soc.org> wrote:
> > On Sun, Feb 19, 2023, 10:14 Jacob Lifshay <programmerjake at gmail.com>
> wrote:
> >
> >> https://foundation.rust-lang.org/grants/project-grants/
> >>
> >> they're planning on accepting grant requests again later this year.
> >>
> >
> > apparently around may:
>
> please answer publicly so that there is a public audit trail:
> are you intending to disregard the NLnet Memorandums of
> Understanding for which you are listed as a member, which
> require your full-time committment,


i'd like to point out this is iirc exactly what happened when you all
worked on the NGI pointer grant, it wasn't an nlnet grant, imho it's fine
due to achieving goals closely related to the nlnet grants but not strictly
covered by them thereby helping the work covered by the nlnet grants.

by applying for this
> grant despite to people informing you that there are much higher
> priority committments?
>

i haven't really thought through prioritizing, however this seems very
similar to us applying for the grant for cryptographic algorithms even
though we didn't technically need a special fast chacha20 algorithm to be a
good cpu/gpu for our other existing goals.

adding the cranelift backend allows working out the kinks (getting a good
production register allocator in particular) in a codebase that is more
flexible (partially because it's smaller and younger) than gcc or llvm
therefore helping further the cause of getting a working gcc/llvm compiler.
it's also kinda a testing ground for how to design the compiler ir we want
for svp64. both of these benefit the things that nlnet is actually funding
-- getting svp64 up and running.

so, i consider this as part of the process of getting good llvm/gcc svp64
support.

>
> if you perform other Grant work you will be in violation of
> your committment to NLnet.
>

didn't think about that because this seemed just like NGI Pointer which
apparently wasn't a problem.

maybe we should ask michiel what he thinks...imho he'll probably be fine
with it since if the rust foundation pays for it then nlnet can spend the
money nlnet would have needed to spend on it on some other aspect of
libre-soc rather than having to use up lots of nlnet funding starting
getting compilers working for svp64. libre-soc has enough work to go around.

>
> secondary question:
> what NLnet-funded tasks have you worked on in the past 4 months,
> that helps move this project forward?
>

i've been working on the rsa algorithm implementation for:
https://bugs.libre-soc.org/show_bug.cgi?id=942
which is supposed to be a partial subtask (the rsa stuff is, not the
presentation stuff) of:
https://bugs.libre-soc.org/show_bug.cgi?id=773
though i haven't gotten around to updating bugzilla to split out just the
stuff required for rsa as a separate task and adding it as a subtask of 773

the register allocator stuff i'm currently working on is needed for fast
bigint x bigint toom-cook multiplication which is used to construct
montgomery multiplication which is used to construct fast bigint modular
exponentiation which is the core operation of rsa

Jacob