Re: GMP Inputs

[Hans Aberg]

At 08:23 +1000 2001/04/26, Kevin Ryde wrote:
>> - As for the inaccuracy (inverted value of precision) of floating and fixed
>> point numbers, I think that a fairly correct mathematical way would be to
>> use differentials of the operations involved to compute the new inaccuracy.
>> That is, unless the user specifically requests a round-off.
>
>I think mpfr (www.mpfr.org) had ideas for an interval arithmetic along
>those lines, made possible by the exact rounding in the mpfr floats.

There seems to be a balance: On the one hand, one wants to add sufficiently
many digits in order to not decrease the accuracy, on the other hand, one
does not want to add too many digits for the sake of speed.

I saw that Gerardo Ballabio uses the idea to let the LHS side of an
assignment to decide the precision. This gives rise rather complicated
implementation solutions. (But I used the same kind of idea in another
context in the form of a dynamic "double dispatch" implementation.)

One advantage with this precision computation is that one avoid that.

But I how no idea of how practical it would be for your GMP library.

  Hans Aberg
                  * Email: Hans Aberg <mailto:address@hidden>
                  * Home Page: <http://www.matematik.su.se/~haberg/>
                  * AMS member listing: <http://www.ams.org/cml/>