Libre-SoC Pursuing New Crypto Primitives & Instructions For OpenPOWER

[phoronix]

Phoronix: Libre-SoC Pursuing New Crypto Primitives & Instructions For OpenPOWER

While Libre-SoC began as "Libre RISC-V" envisioned as a low-power graphics/Vulkan accelerator, it has morphed into a hybrid CPU/GPU design built on OpenPOWER and in very early form seeing test fabrication on a TSMC 180nm process. The latest funding received is now working on adding cryptographic improvements to it and/or the upstream OpenPOWER ISA...

https://www.phoronix.com/scan.php?page=news_item&px=Libre-SoC-Crypto-Project

[OneTimeShot]

Ok.... so they gave up making an open source "GPU design" (think a software renderer compiled for a lightly customised Power CPU), and they've given up making a System on a Chip. Now they're doing open source crypto/blockchain hardware?

They're good at pulling in that tax-payers' money!

[programmerjake]

Ok.... so they gave up making an open source "GPU design" (think a software renderer compiled for a lightly customised Power CPU), and they've given up making a System on a Chip. Now they're doing open source crypto/blockchain hardware?

They're good at pulling in that tax-payers' money!

We've not given up on building a Libre GPU or a SOC, in fact we're currently working on designing special branch instructions specifically for GPU workloads.

[hotaru]

Now they're doing open source crypto/blockchain hardware?

s/doing/talking about/

it doesn't look like they've actually done anything yet, and since they already got paid for it they probably never will.

[Siuoq]

What's a generic galois field alu? No results in google https://www.google.com/search?q="generic+Galois+Field+ALU"

[Siuoq]

Also how can I paste links here?

[lkcl]

Ok.... so they gave up making an open source "GPU design" (think a software renderer compiled for a lightly customised Power CPU), and they've given up making a System on a Chip. Now they're doing open source crypto/blockchain hardware?

the actual focus of the NLnet fund is for improving the internet. (that includes helping you to make bold, brash statements. sigh)

https://nlnet.nl/assure/

i wrote a very specific targetted message to the bitcoin-dev mailing list to invite and attract people there to participate:
https://lists.linuxfoundation.org/pi...st/019326.html

the primary reasons for the NLnet grant application were two-fold:

1. to help with the intermediary goal of creating a single-core ASIC - a gigabit ethernet router ASIC - as a step along the way to an SoC
2. to put in Draft bitmanipulation primitives into the ISA to help with Video and 3D image processing

in short it's multi-faceted and nowhere near as simplistic as you're making it out to be.

[lkcl]

What's a generic galois field alu? No results in google https://www.google.com/search?q="generic+Galois+Field+ALU"

https://libre-soc.org/openpower/sv/bitmanip/

general-purpose Galois Field Arithmetic Logic Unit, which will implement general-purpose Galois Field instructions.

https://en.wikipedia.org/wiki/Finite_field

https://en.wikipedia.org/wiki/Arithmetic_logic_unit

the majority of CPUs, if they have Galois Field instructions at all, will have GF(8) or other fixed-width operations, and some even restrict the modulo, which are only useful under certain very special circumstances.

the purpose of the research that NLnet has kindly agreed to donate towards is to explore creating instructions that are unrestricted in both the width and the modulo field. thus they may be used for any purpose (a general purpose).

most CPUs will not have anything like this, at all. they will instead have hard-coded AES instructions (AES == Rijndael, Rijdael used GF mathematical primitives to *design* the algorithm but with no CPUs having general-purpose GF operations they end up with a special custom macro)

hth.

[lkcl]

it doesn't look like they've actually done anything yet, .

https://git.libre-soc.org/

you are more than welcome to help contribute, and be paid for helping.

this is a massive project: it doesn't happen overnight. there is a huge amount of planning and preparation involved. instructions have to be researched, then designed, then written up. then a simulator designed, then unit tests written. that's several months already.

then once they're proven functional, the HDL has to be designed (hardware version of the simulated version). then that can be tested with the same unit tests. this takes several more months.

then the instruction has to be written up in a form that's suitable for presentation to the ISA Working Group.

then you also have to put that instruction into compilers.

then you have to write accelerated drivers, usually in assembler (ffmpeg) because compilers even with the brains available today are not optimal enough.

and

and

and

and

this is greatly simplified: i have left out a huge number of steps, but are you starting to see why it doesn't "look like" there's anything that you can call "progress"?

all of this is what a standard CPU company does: you just don't see it happen because they present you with a (fait-accomplit) product.

[lkcl]

and since they already got paid for it.

NLnet grants simply do not work that way. unlike other EU Horizon Grants, it is not a Contract. the money is reserved by the E.U., there is a process involving independent audit and review of the proposal given to NLnet, and, on agreement with the EU Horizon Programme, grant recipients must 100% complete agreed milestones BEFORE they can put in a Request for Payment.

NLnet then totals up all RFPs and puts in a corresponding request to the EU Horizon Programme.

we do not receive the money in advance: none of the NLnet Grant recipients do.

this avoids the situation as with most Grants where for example in Academia I have heard horror stories of people blowing the entire Grant budget on expensive equipment then scrabbling around for money to pay for food and other basic living expenses for the duration of the contract. or in one case throwing wild parties.

we actually have to do the actual work that we committed to - and provide reasonable proof of its completion - before we can receive a donation for having done it. past tense.