[fpu] Rounding Again

["Rudolf Usselmann" <russelmann@hotmail.com>]

>From: Damjan Lampret <damjanlampret@yahoo.com>
>To: Rudolf Usselmann <russelmann@hotmail.com>
>
> > I'm still having a hard time with rounding.
> >
> > I did get the book you recommended. I still don't understand binary
> > rounding. I understand the math, but can't see how the GRS bits are
> > used. It's somehow confuses me. I have read a lot on this subject
> > now,
> > but can't believe that there is no step by step explanation ...
>
>I knew details but forgot (this was a year or so ago). Anyway I know
>that this is used when you shift the result to right (if I remember)
>and you compare shifted result with another shifter result that is
>shifted for one bit more. And GRS bits are the one most right and tell
>you whether you have precise number as it supposed to be after
>normalizing or it is not precise. Anyway something like that. I am
>trying to find pieces of this in my mind. ;-)
>
>I can take a look in Hennessy and Patterson book where FP is also
>explained (I think there is much easier explained since their books are
>targeted to students).

I have that book as well. I still don't get it.
Here is what I got from all the books:

1) You take the operands, increase the precision by
   appending 3 zeros at the right. (So your fractions
   are now 23+3=26 bits wide)
2) Preshift (Normalize).
3) Add
4) Normalize (count leading zeros and shift)
   The shifted result shifted in to GRS, if shifted right
   (where S is sticky), otherwise GRS are zero (and ignored).

Is this correct so far ???
What do I do next ?

>regards,
>Damjan

rudi
________________________________________________________________________
Get Your Private, Free E-mail from MSN Hotmail at http://www.hotmail.com