No, it was not "Michal Jaegermann" who wrote this. Folks, can you be
a bit more careful with accreditations?
> Well, it's "taylor" to first order anyway;
That's right. Nobody was contesting that.
> > One cannot be sure without running tests with an actual code in a
> > "real life" situations, but I strongly suspect that a rational
> > approximation of a degree 3 could fare better.
> Digital doesn't seem to think so; there are no divisions in an objdump of
> trigonometric portions of DPML
They may be right. Like I wrote, without careful analysis of a concrete
chip and testing results, both for speed and correctnes, one cannot
tell. On the other hand it doesn't seem likely that a Taylor expansion
around zero gives a good and fast approximation on the whole interval
in question. But maybe it is competitive? I do not know at this
moment but I would be somewhat surprised.
> There's also Numerical Recipes, which is now online.
> The relevant section (in Fortran; sorry)
> section 5.11 onwards
Be careful with Numerical Recipes. This book is not famous for an
always optimal choice of algorithms and detailed numerical analysis.
I even heard some knowledgable people claiming that Numerical Recipes
influence is outright harmful (I would not go to such extremes :-).
I am afraid that we are going somewhat tangentially on Alpha, although
the issue of math libraries is really important.
-- To unsubscribe: send e-mail to email@example.com with 'unsubscribe' as the subject. Do not send it to firstname.lastname@example.org
Copyright © 1995-1997 Red Hat Software. Legal notices