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

Re: [f-cpu] Some answer

hi !

nico wrote:
> Argh ! 92 messages !
this list is very "cyclotimic" :-)

> Define a generic representation of code to write different compiler are
> very hard. The big example is why gcc couldn't produice an output in
> java byte code. JVM are simply a "virtual processor", so it will be nice
> that gcc could compile for this target. But it's a stack machine not a
> register based one. And the semantic of the intermediate representation
> are too poor to be used.

Add to that other parameters, such as the programmer's tastes
(we have a good sample of different opinions in this list)
and some corporate pressures, and the panorama becomes even more fuzzy.

but don't forget that the goal of the project is to make the best CPU
possible, that's all ;-)

> Our register set is definitely 3r 2W, i don't like it too much for
> performance reason but why using it better. One off the main problem are
> exepntion handler. Couldn't it be possible to use the second register to
> output the state without traping (imagine the DIV instruction, the
> second register will say if a div by zero occur or an overflow
> occur,...). If the system need a trap an instruction should verify the
> regiter otherwise it could do nothing. What do you think of it ?

i am not sure to understand what you write...

for IDIV the instruction is "issued" if the zero flags are not set.
it's already in the manual i think. it doesn't use the usual register
read ports but the attribute bits.

for the rest, i don't clearly understand. Maybe if you write in french...

> For the floating point unit, we should provid 128 bits extended floating
> point precision for the day of the register set will double.
if you want to implement it... go ahead... but this has been discussed
earlier, it's not necessary or useful.

> What about written a asm optimiser which could be used as backend for
> every different language compiler ?

we're discussing something like that now.

> VHDL could never be used to write SW. It's too poor !  IO fonction are too poor.
VHDL is only the core of the thing. There are the std-XXXX libraries
and the effort for adding POSIX I/O.
But if i want to use a VHDL-like programming langage, i will no doubt correct
some of the little problems it has. i'm not masochist ;-)

> Nowdays HW world try to find an other way to code to a higher
> level. SystemC are the best canditate (C++ + a lib),
i think it's funny : C is lower-level than VHDL (ADA), and people want
it to be higher level than VHDL. i think that a few (only, but anyway)
people of this list laugh :-)

> but there is other proposal (VHDL-OO, UML-RT, new langage...)
we know that System-C or something like that will emerge.
unfortunately, the companies that promote them are bad at planning for the very long term.

> read you soon !
ans see you if you have time !

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