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

Re[4]: [f-cpu] "Tree"

On Fri, 11 Jan 2002, Juergen Goeritz wrote:

> On Fri, 11 Jan 2002 nicolas.boulay@ifrance.com wrote:
> > On Thu, 10 Jan 2002, nicO wrote:
> > > Juergen Goeritz a écrit :
> > > > Did you ever program drivers, operating systems and
> > > > really big programs in C?
> > > 
> > > No but somebody else do it (linux, gnome). So what
> > > is the point ?
> > 
> > Oh! Then it's hard to explain. Let me say it this
> > way, the abstraction level is too low.
> > 
> > >>>> So there is "some" millions C lines project. C++
> > are a quite bad design OO language but if you
> > restrict your self to a sub set of it it could be
> > great.
> > SystemC are c++ class to rease the level of
> > abstraction because it's runnable you must be enought
> > precise. 
> > Which kind of concept do you want to be implemented
> > in such language ? (my last year at university was
> > made inside a laboratorie which try to design such
> > kind of language, and todays SystemC are very close
> > to what was espected) 
> The problem is that with C you have an implementation
> already. This means you have already translated from
> the concept you had in mind. You strip information
> about your design when you translate it to C and you
> add 'noise'.
> A good language would resemble the way of thinking of
> mans brain regarding digital design. Therefore C isn't
> a good language. Some problem solutions can be easily
> formed with C, others are hard to implement. Most of
> them are hard to check (time consuming) for correctness.
> Simulation is a means to look at the same issue from
> another point of view - you go to the outside of your
> design and look how it reacts. Testvectors that are
> defined also describe the design functionality. If you
> have a 100% testvector coverage you wouldn't need the
> source of the design any more - the testvectors would
> do because they also completely define your design from
> it's reactions. And they have to because you need them
> to check the final chip.

Oops, forget that writing here about complete definition.
It's only true when you take a complete simulation as
testvectors which is done fairly often.

A 100% coverage should test every path inside the design
to work as expected on LH and HL edges. Normally more
things are checked in parallel in a single teststep to
optimize the time to test. That's what those scanpath
flipflops are for.


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