Rep:Re: [f-cpu] reg. rotation [Was: New suggestion about call convention]

["Nicolas Boulay" <nicolas.boulay@ifrance.com>]

If register windows realy interrest you. Watch Sparc documentation. They
used a rotating register set. Compiler choose to rotate or not. When
overflow/underflow occure, OS interrupt handler dump a (some?) register
window in the current stack frame. (when a function call use the
rotating register set, it use 24*4+n octets, most of the time never
used).

Compiler must be clever to avoid windows overflow. OS always reserve one
windows for interrupt handler (the register rotate at interrupt).

nicO

-----Message d'origine-----
De: Martin Devera <devik@cdi.cz>
A: f-cpu@seul.org
Date: 27/11/02
Objet: Re: [f-cpu] reg. rotation [Was: New suggestion about call
convention]

> > of it unfortunately). Example 1 shows the best case (probably
> > return address could be saved in rotated local reg. too) and you
> > can see that is can be really fast.
>
> If I understand you example correctly, you want to protect parameter
to not
> being transfered to a saved register, right ? But you still do memory
> transfert and an one cycle in the decoder to do this, plus a cycle for
circ
> instruction. I don't see where you expect to have better performance.

eh eh .. Sorry my fault. I recomputed it yesterday and yes you
are right. The only benefit would be fact that memory loads
would not affect saved values whose we saved itself but some "older"
ones which might not be needed soon ...
But on other side these registers would be needed for store by
next sibling subroutine call so that it is nonsence.

> will only use temporary. (I think we need to have some stat about
software on
> RISC CPU. Did somebody know a tool or something for that purpose ?).

Hmm, some time ago I patched gcc to give me interdependency information.
The I found http://sourceforge.net/projects/introspector/.
It might do the job you want.

> >From my point of view, the idea isn't to use all of them, but  to use
them to
> reduce memory operation and to reduce pipeline bubble. And I think,
but
> perhaps I am wrong, that store didn't cost anything (only load
introduce
> bubble, from what I remember about whygee LSU).

this leads me to question, is LSU documented somewhere ? I'd be
interested
to study it but found nothing except some disscussion.

When I tried to write call to validate my assumptions I found that
I sometimes need to move bunch of parameters from s? to a?. Like in:

f(a,b,c) {
 g(a,b,c); g(a,b,c+1); g(a,b,c+2)
}

where you want to save a,b and c as local vars and later move them
to a?. Would be possible to have unconditional double move which
would be 2r2w and moe 2 regs at the time ? Like:

; t0=addr of g
store -8,[sp],s0
store -8,[sp],s1
store -8,[sp],s2
store -8,[sp],ra
dmove a0,a1,s0,s1
move a2,s2
jmp t0,ra
inc s2,a2
dmove s0,s1,a0,a1
move a2,s2
jmp t0,ra
dmove s0,s1,a0,a1
.....

There is enough room in opcode. What do you think ?
I'm not sure with orthogonality of it ...
devik

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

__________________________________________________
Modem offert : 150,92 euros remboursés sur le Pack eXtense de Wanadoo ! 
Haut débit à partir de 30 euros/mois : http://www.ifrance.com/_reloc/w


__________________________________________________
Modem offert : 150,92 euros remboursés sur le Pack eXtense de Wanadoo ! 
Haut débit à partir de 30 euros/mois : http://www.ifrance.com/_reloc/w

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