[clean-list] Re: 1.Floating Point Support (clean-list digest)

[RT Happe]

On Thu, 16 Nov 2000, Siegfried Gonzi wrote:

> I do not know what you really mean, but rounding errors are inavoidable (in
> everey language); that is my naive assumption.

Nevertheless you can try to estimate or to minimise rounding errors.
For detailed information, cf. the references in the posts of Jerzy 
Karczmarczuk and myself.  Here are just two illustrations:
1) A simple technique intended to reduce the error accumulation when
summing large unsorted sequences of, say, double precision f. p. numbers
consists in making the accumulator a variable of higher precision, i. e.
extended double precision.  [Alas, few chips support the - optional! -
IEEE754 extended format, among them the Intel 80x86/7 family and
Motorola 68040/60.]
2) Exact adaptive precision arithmetic computes values of arithmetical
expressions within a prescribed (small) error bound.  Evidently, such
things rely on the semantics of arithmetic operations, e. g. as specified
by the IEEE 754 standard.

rthappe