Numerics question: reading floating point constants

nicolas cellier ncellier at ifrance.com
Mon Apr 24 22:44:09 UTC 2006


Le Dimanche 09 Avril 2006 18:23, stéphane ducasse a écrit :
> Do you know a good book or lectures on the problems related to number
> representation in our nice
> computers?
>
> Stef
>

Some more links:

function doing the conversion in C is strtod.
glib implementation use mulitple precision integer arithmetic (mp library).

Google: strtod source documentation
http://www.jhauser.us/arithmetic/SoftFloat.html

This guy has also a TestFloat program written in C for testing conversions
Maybe we can borrow some tests and port in SUnit


Google: Sun Microsystems' Numerical Computation Guide
 http://docs.sun.com/source/806-3568/

Good doc i learned with is available on the net...

Google: IEEE 754 standard
http://en.wikipedia.org/wiki/IEEE_754

Sort introduction. Following the links you will find the two above...

I started programming algorithm M, R and Bellerophon from tony's ref.
But it does not handle denormalized (gradual underflow) nor possible overflow 
(Infinity). Deserve some more work...
It does involve LargeInteger arithmetic anyway because of 52+1 bit 
significand... (Bellerophon use 64 bits). But performance seem not degraded 
vs current implementation (i do not test parsing so far, only the conversion 
part).

Nicolas




More information about the Squeak-dev mailing list