[Libre-soc-bugs] [Bug 1155] O(n^2) multiplication REMAP mode(s)

[bugzilla-daemon at libre-soc.org]

https://bugs.libre-soc.org/show_bug.cgi?id=1155

--- Comment #44 from Luke Kenneth Casson Leighton <lkcl at lkcl.net> ---
(In reply to Jacob Lifshay from comment #42)

> I never said the hardware has to,

i know, i apologise, i'm a little jumpy on this topic, things
need to be spelled out explicitly (i am partly writing for future
readers here).

> I've always just said it *can*, not
> *must*. also, it probably doesn't have to match all the instructions, just
> the 4 instructions in the loop and matches on the current SV state.

4 is still far too complex. even just 2 is complex enough.
good to have the discussion but please bear in mind *all*
possible potential implementations have to be considered, not
just one.

ISA design is tedious extreme-patient work.

> actually, while thinking about this, I think we'll want an instruction that
> has the combined functionality of bdnz and svstep.

good thought/idea: already tried it. started to break RISC paradigm.
far too complex both in hardware and in terms of combined functionality.

mostly it was down to number of register hazards exceeding 3-in 2-out
if i recall, but also that branch analysis goes completely sideways.

> maybe one of the branch
> instruction's bits can be used for that for sv.bc?

there are already 512 combined options which is alarming the OPF
(with good reason) so no. the unit tests are going to be a massive
job.

good to have the discussion but can we please keep this bugreport in
scope and on topic? its budget is already quite small

-- 
You are receiving this mail because:
You are on the CC list for the bug.