On Thursday 24 February 2005 18:34, Daniel Phillips wrote:
> On Thursday 24 February 2005 17:55, Lourens Veen wrote:
> > 1.m is in [1, 2), so log2(1.m) is in [0, 1). (e - 127) is in [-126,
> > 127], and integer. That means that all we need to know if we want a
> > rounded result, is whether log2(1.m) > 0.5. Which happens if 1.m >
> > 2**0.5. Which happens if m > (sqrt(2) - 1). Or, for a 16-bit m as
> > integer, if m > 27145.
> >
> > So, we need one 16-bit comparator and an incrementor.
> >
> > What did I miss?
>
> It looks great.  In short, floor(log2(f)) is given directly by the
> exponent.  Your observation on rounding is very nice.

Whoops, what you missed is that the trilinear blend needs the fraction.  
However, because only 8 bit precision is needed, this can be a straight 
table lookup.  Or it might not even look that horrible to use the 
mantissa directly.

Regards,

Daniel
_______________________________________________
Open-graphics mailing list
[email protected]
http://lists.duskglow.com/mailman/listinfo/open-graphics
List service provided by Duskglow Consulting, LLC (www.duskglow.com)

Reply via email to