[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: Rep:Re: [f-cpu] Are 8 bits SIMD mode usefull ?
I have "lost" the previous mail thread, sorry.
I really don't understand the interret to change the size of the chunk.
Our cpu is natively a vector machine so why not stay with that ! In a
256 bits machine you will load 4 64 integers at a time, that's all.
If we add some "gather" instruction the "SR_SIZE" will not have any
interrest (or only to divide loop counter).
nicO
On Fri, 10 Jan 2003 09:59:41 GMT
"Nicolas Boulay" <nicolas.boulay@ifrance.com> wrote:
> I beleive they are some mix.
>
> Sparc/Mips and i beleive all RISC processor didn't have any 8 bits
> load(nor 16 bits). They just manipulate 32 or 64 bits at the same time
> but still handle C "char".
>
> I am pretty sure that's x86 is the only cpu that could fetch 8 or 16
> bits from the memory (which make the LSU complexe).
>
> The other problem are the usefullness of a 8 bits vector. I know that
> picture use this size a lot but it need so much trick ! (i had
> followed ffmpeg that made a great MMX FFT code, but there are some
> problem of darkness or gain in it). The only use could be the
> convertion between picture type (RGB, BGA, YUV420,YUV420P,YUV422,...)
> but video card will handle it more and more !
>
> ---
> GPU use "128 bits pixel" to enable pixel and virtex shader. Very few
> instructions could be handle (~128) per pixel/vertex. I could imagine
> the udge pipeline that is could be. That's the famous DirectX 9.0
> advantage (from 8 bits to 24 bits float for some part in the >9500 ATI
> card, then (GeoforceFX ?) to full 32 bits float by channel in all the
> 3D pipeline).
>
> nicO
>
> -----Message d'origine-----
> De: Michael Riepe <michael@stud.uni-hannover.de>
> A: f-cpu@seul.org
> Date: 10/01/03
> Objet: Re: [f-cpu] Are 8 bits SIMD mode usefull ?
>
> On Fri, Jan 10, 2003 at 08:37:07PM +0000, nico wrote:
> > Just a stupid question : is 8 bits integer mode usefull ?
>
> Yes!
>
> > "Char" C type are going to be 16 bits (unicode) or even 32. 3D are
> going
> > in floating point even in the GPU. Gimp use 16 bit per channel (in
> > certain mode, even more ?).
>
> What you mean is `wchar_t', not `char'. And even that can be encoded
> in a series of bytes (that's called UTF-8).
>
> 8-bit mode isn't dead. I guess it will never die.
>
> --
> Michael "Tired" Riepe <Michael.Riepe@stud.uni-hannover.de>
> "All I wanna do is have a little fun before I die"
> *************************************************************
> To unsubscribe, send an e-mail to majordomo@seul.org with
> unsubscribe f-cpu in the body. http://f-cpu.seul.org/
> _____________________________________________________________________
> GRAND JEU SMS : Pour gagner un NOKIA 7650, envoyez le mot IF au 61321
> (prix d'un SMS + 0.35 euro). Un SMS vous dira si vous avez gagné.
> Règlement : http://www.ifrance.com/_reloc/sign.sms
>
>
> _____________________________________________________________________
> Envie de discuter en "live" avec vos amis ? Télécharger MSN Messenger
> http://www.ifrance.com/_reloc/m la 1ère messagerie instantanée de
> France
>
> *************************************************************
> To unsubscribe, send an e-mail to majordomo@seul.org with
> unsubscribe f-cpu in the body. http://f-cpu.seul.org/
> _____________________________________________________________________
> Envie de discuter en "live" avec vos amis ? Télécharger MSN Messenger
> http://www.ifrance.com/_reloc/m la 1ère messagerie instantanée de
> France
*************************************************************
To unsubscribe, send an e-mail to majordomo@seul.org with
unsubscribe f-cpu in the body. http://f-cpu.seul.org/