Re: Binary-Decimal conversion for extremely large integers

[Thorsten Siebenborn <7_born@xxxxxx>]

Nick Maclaren schrieb:
> Thorsten Siebenborn <7_born@xxxxxx> wrote:
> > I am talking about numbers with thousands or even millions of
> > binary digits.
>
> I don't see the relevance to this mailing list, but still :-)

True. I was hoping that with the advent of higher precision (64-bit
processors, 128-bit quad precision or staggered precision with IEEE
(Priest) which may have much more bits) there are some algorithms
which can be modified to work with arbitrary precision.
I knew that some people posting here are working with
arbitrary-precision libraries like GMP/MPFR and I hoped that
they know the current best approach.
As discussing normal approaches is hard to defend for IEEE issues,
I am retracting the question and discuss it privately. Sorry.

> >  [...]
> > But if you use long division, the division must be splitted
> > into several chunks and the algorithm must correct the bits
> > accordingly which in fact is prohibitively expensive.
> > So expensive that in fact nine runs of my algorithm is still faster
> > than one run with long division; the longer the number, the better
> > the add/shift algorithm.
>
> Grrk.  That doesn't sound like a very good divide instruction.
> It certainly USED to be dire, but that was ages ago.  You might
> try asking Terje Mathisen on comp.arch.arithmetic.

Thank you very much. comp.arch.arithmetic is a good resource;
I have found an interesting papers discussing the different
approaches like Goldschmidt, Byte Division, Restoring Division etc.

Best,
Thorsten