[Libre-soc-dev] scalar instructions and SVP64

[Luke Kenneth Casson Leighton]

---
crowd-funded eco-conscious hardware: https://www.crowdsupply.com/eoma68

On Wed, Mar 10, 2021 at 3:39 AM Adam Van Ymeren <adam at vany.ca> wrote:

> I can see it being a bit of extra work for compilers to know that accessing high numbered registers requires a bit of extra work but that really doesn't sound unworkable to me.

i would really love that we had the old SVP64 format, where VL and
MAXVL were embedded into 12 bits of the prefix.  that would solve
this... but the trade-off is: we lose *all* of the modes which have
been added instead.

> I can however imagine that adding a special case for VL==0 could result in many more gates/complication for the OoO/scoreboard execution engine.

yes, basically.  no design of any kind could ever be high-performance.

On Wed, Mar 10, 2021 at 3:10 AM Hendrik Boom <hendrik at topoi.pooq.com> wrote:

> uniformity is extremely important for object code generators.
>
> Every exception needs to be programmed around and introduces potential
> code generation bugs.

yes.  this is *another* detrimental consequence of what is best termed
"Complex-V".

there will be permutations that don't "make sense".  it looks on the
face of it like these could, surely, be used "for other purposes"?
and the answer's no, they can't.

l.