[Numpy-discussion] Re: numpy, overflow, inf, ieee, and rich , comparison

[Rainer Deyke]

"Konrad Hinsen" <hinsen at cnrs-orleans.fr> wrote in message
news:m3wvewc4hh.fsf at chinon.cnrs-orleans.fr...
> "Rainer Deyke" <root at rainerdeyke.com> writes:
>
> > Floating points should not be part of the core language.  The correct
way to
> > implement floating points is in terms of integers, preferable within the
> > language (as opposed to a C extension).
>
> If you don't like floating point numbers, you are free not to use
> them. Many of us do need them, know how to deal with them, and need
> them in a usable form, i.e. with decent performance and accessible
> from C modules, which implies using the binary format provided by the
> CPU. Your proposal is completely unrealistic.

Given current CPU design, one might as well take advantage of it, yes.
However, if the CPU was designed to do floating point calculations in terms
of integers, the result would be a cleaner architecture and faster overall
performance (given that integer operations are much more common than
floating point operations).

I am deliberately overstating my position as a counter to the ridiculous
proposition that floating point operations be allowed to sneak into my code
through the division of integers.  Neither is viable at this point, simply
due to the amount of previously existing code.


--
Rainer Deyke (root at rainerdeyke.com)
Shareware computer games           -           http://rainerdeyke.com
"In ihren Reihen zu stehen heisst unter Feinden zu kaempfen" - Abigor