Re: [f-cpu] "Tree"

[Juergen Goeritz <goeritz@oekomm.de>]

On Thu, 10 Jan 2002, Yann Guidon wrote:
> > since two years my brain is thinking back and forth how one
> > can integrate the design process of software together with
> > the design of the hardware to begin to build 'real' systems.
> > 
> > There seems to be a need for a more high sophisticated
> > description language that is a superset of hardware and
> > software implementation but without forcing the style to
> > either of them. I would call this a 'system definition
> > language'. This language must also contain some means
> > to simulate the system on its highest level.
> > 
> > If you look at hardware development, there is e.g. VHDL
> > high level description and the netlist/gate level. I am
> > looking for that level above VHDL that can be generated
> > into hardware (e.g. VHDL) and software (e.g. C, C++).
> > 
> > Only with that independent level you are really free to
> > move the border between hardware and software during the
> > development process to achieve a maximum performance of
> > the system in design.
> 
> The highest level i know is an algorithm. This is what
> defines the rest. But there's a trap here ! I don't know
> a way to define an algorithm without defining an
> implementation at the same time. I guess it's a burden
> that is inherited from the "langages". My solution is
> to not use langages and moving the HW/SW border becomes
> easier. Don't underestimate the power of your brain :-)

I think you are not too far apart. And my idea is to rely
on the dataflow of the sytem, not the algorithm itself.
Why? Because to me it seems that a numerical algorithm is
a means to describe a dataflow together with the data
modification (what I would call the functional part). And
most interesting are those algorithms that change their way
of behaving depending on history, data input or output.

Thus I want to separate a system into its dataflow part and
its functional parts (data modification and data storage). 
The dataflow part usually defines the maximum performance
of the system. The functional part adds/subtracts value
to/from the data. The storage part knows about the history.

And when you look into those devices/macros/whatever on
the market you will find that they all rely on dataflow.
It is the basic principle beyond everything else (beside
energy to work - and maybe energy needs to be taken into
account too).

JG

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