Re: [boost] [metamath] Plans for compile time double

From: Peder Holt (peder.holt_at_[hidden])
Date: 2005-08-31 06:38:34

On 8/30/05, Chris Weed <chrisweed_at_[hidden]> wrote:
> On 8/27/05, Peder Holt <peder.holt_at_[hidden]> wrote:
> > On 8/26/05, Chris Weed <chrisweed_at_[hidden]> wrote:
> > > Hi,
> > > I was wonder what the plans are for the compile time double library, metamath.
> > > Is there a plan to develop compile time versions of the functions in
> > > math.h, such as sin, cos, pow, abs, etc.?
> > > Thanks,
> > > Chris
> > >
> >
> > That was the plan, yes. An implementation of some math functions
> > already exists under
> > http://boost-sandbox.sourceforge.net/vault/ <expaler/mpl_math.zip>
> > I have modified the sine function to work with the compile time double
> > library, but the main problem has been lack of time.
> > I will try to find time to implement more math metafunctions, but help
> > is welcome.
> >
> > Attached are implementations of sine and power (only integral
> > exponents supported for now)
> >
> > Regards,
> > Peder
> >
> 
> Hi,
> I would like to help out with this. Was the plan to implement most of
> the trigonometric functions with taylor expansions or some other
> algorithm?

Great! Thanks.
I'll upload the source to the sandbox cvs.

Cromwell has implemented cosine,sine and sqare root. I have rewritten
his sine implementation to use the _double_double meta-operators.

The idea is to use taylor expansions, but wherever there is another
series that converge faster, this should of course be used.
E.g. The only expansion I have found of pow(x,y) where x and y are
floating point numbers, requires one to calculate the series of ln(x)
and then exp(ln(x)*y). This is very costly.

Regards,
Peder

> Chris
> 
> _______________________________________________
> Unsubscribe & other changes: http://lists.boost.org/mailman/listinfo.cgi/boost
>