[libre-riscv-dev] PowerISA, NLNet grants

[Immanuel, Yehowshua U]

I’ve been pretty busy writing a compiler for an ML accelerator and have been loosely following the threads…
That said,

I am wondering why on earth one would wish to switch to PowerISA?

Its a great architecture! ...but it has virtually no software support besides a few niche 
Linux needs for IBMs sever applications…
And without software support - hardware adoption just doesn’t happen.

I did read that Crowdsupply considers RISCV foundation unethical - but why is that?
I think RISCV is restrictive about what they add to the official ISA - but for understandable reasons,
mainly Hennessy and Patterson want to see serious industry adoption for RISCV - and they think the
way to do that is by guaranteeing that RISCV will rarely ever change. This is quite important as x86 has shown with 30 new instructions per year...

Nevertheless, going through RISCV foundation isn’t the only way to get a GPU added to RISCV.
I have some colleagues at Georgia Tech that work with BAR(Bekeley Architecture Research)…
I can tell you showing a working implementation will definitely get BAR’s attention.
And BAR basically more or less controls RISCV - even though this isn’t externally obvious.

That said, from what I’ve heard - BAR’s focus at the moment is on the vector extension - only
because people have complained RISCV doesn’t have one.

All this to say is I think switching to PowerISA would be fighting a losing battle.
BAR has been hard at work with Debian to make sure that 98% of Linux packages work on RISCV,
and the best and brightest in academia are putting much effort into POC RISCV cores that keep getting better and better - this means PowerISA get little attention.
Every year there are new paper’s on architectural exploration for RISCV this and RISCV that, but virtually
nothing on the PowerISA.

RISCV being FOSS means that many Chinese Manufacturers are already pumping out RISCV 
which increases volume and decreases cost. So in the future, it may be quite possible to have a full
LibreSOC with the benefit of Chinese Manufacturers taking the risks of manufacturing off our hands.

RISCV has financial support from Google, Samsung, Nvidia, Micro etc…
Every way I stack it, RISCV will win. Going with PowerISA - well LibreGPU won’t see the light of day.
IBM simply doesn’t have enough clout - and if PowerISA ever makes it - it will always ever be second best…

It certainly possible to see LibreGPU adopted by RISCV foundation - or at least endorsed.
We would just have to be clever about it.

These are just my thoughts - my perspective is mostly informed by academia - but that might be relevant since RISCV came directly out of academia.
Of course I welcome corrections to my perspective.