Deformable registration in D
Don Clugston
dac at nospam.com.au
Fri Aug 3 00:02:53 PDT 2007
Robert Jacques wrote:
>> Exp with finite support, how does this go?
>
> The support of a function is basically the range in which its non-zero.
> See http://en.wikipedia.org/wiki/Support_(mathematics) for more details.
> I’m specifically using it to calculate the Gaussian, which is
> essentially exp(-x) where x > 0 (x is a distance). So I have a few small
> x’s which become relatively large exp(-x) and many large x’s that become
> really tiny exp(-x). By approximating all those large x’s (tiny exp(-x))
> with zero (i.e. making the support finite) I don’t have to calculate as
> many exp functions, which speeds up the computation.
What cutoff are you using? Denormals are only generated for x around -11300 for
80-bit reals, but for floats they'll happen for x between about -85 and -105.
IIRC, denormals are ~100 times slower than normals on x87, so could easily be
performance killers.
More information about the Digitalmars-d-announce
mailing list