[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [f-cpu] Compagnon cpu


nicO wrote:
> I just had a hard discussion with whygee about a new idea.
> I have speak with a guy who write code for HURD. Compare to Linux,
> context switch is a real problem for microkernel OS (usualy there is 3-4
> task running in the same time). So i have a proposal to speed up such
> thing.
> I propose to introduice a very simple cpu which control the "main" fcpu.
> I think of kind of a simple LEON (wihtout mul and without div) or a fcpu
> without SIMD stuff, without FPU, without mul,...
> What do you thing ?

I have explained my arguments today. To say the truth,
i am not convinced by Nico's advantages.

 - before we make more complex cores, we have to make FC0.
 - I will not stop the ongoing work to include that kind of
   "addition" because (despite your arguments) we have to make
   something that works, after 3 years of work.
 - Adding such stuff like that all the time will only make the
   CPU heavier, more difficult to program efficiently (this
   is an asymetric machine :-/) and it will never be ready at all.
 - Only with emulation (FPGA or dedicated machine) we can gather
   enough "real use" statistics and decide whether it is worth
   it to include a "kernel coprocessor". This "addition" must
   remain a user's choice, not a mandatory thing.
 - P&H's "QA" would say :
   "if you increase the die size by 10%, you need to have more
   than 10% of performance increase." However i don't remember that
   F-CPU was created as a "HURD CPU".
 - Other CPU architectures have showed that if you want to "speedup"
   context switch time too much, you will mess up the architecture
   (register set that explodes and reduces the cycle time for example).

If nico, Cedric or anybody comes with a new working core and wants
to put it besides FC0, fine. The current proposal is however difficult
to scale up (there is no plan yet).

I don't want to take this idea down but it will be difficult to find
a solution for every problem.

> nicO
To unsubscribe, send an e-mail to majordomo@seul.org with
unsubscribe f-cpu       in the body. http://f-cpu.seul.org/