On 12 Mrz., 16:51, Carl Witty <carl.wi...@gmail.com> wrote:
> On Thu, Mar 12, 2009 at 4:16 AM, David Joyner <wdjoy...@gmail.com> wrote:
>
> > I have an ignorant question: what are the canonical reps of
> > ZZZ/nZZZ in C? (-n/2,n/2]?
> > Is the issue to decide between the interval [0,n-1] as reps of ZZ/nZZ
> > (Python)
> > vs (-n/2,n/2] (C)?
>
> In C, ZZ/nZZ does not have canonical representatives. For example,
> the equivalent of [n%4 for n in [-7 .. 7]] would give:
>
> -3, -2, -1, 0, -3, -2, -1, 0, 1, 2, 3, 0, 1, 2, 3
Hi Carl,
the situation might be worse:
AFAIK, that choice ist *not* defined by the ANSI C standard, but left
complier-dependent. In other words: you might get different answers on
the same system using different C compilers, or even the same C
compiler with different options. C is fun, isn't it?
So the original question turns into:
Do we want to (potentially) sacrifice speed in favor of portability?
If so, we should make the Cython behaviour predictable, i.e. "fixed"
--- favourably identical to the Python behaviour.
And in the Cython build scripts we should check whether we have e.g.
GCC/architecture xyz and set respective options, to not lose speed in
the cases it is not necessary so.
Cheers,
gsw
>
> This is annoying to a mathematician. The big reason in favor of this
> is to align with division: both C and Python give a==(a/b)*b+a%b, but
> Python uses floor division and C uses truncating division.
>
> So why does C use truncating division? I'm not sure what the choice
> point was, but at this point two very important reasons would be:
> because that's what the signed integer division instruction gives you
> on basically all processors, and because that's what the C standard
> has specified for years, so people have written code depending on that
> behavior. Given these facts, it's IMHO unlikely that C will ever
> change.
>
> Carl
--~--~---------~--~----~------------~-------~--~----~
To post to this group, send email to sage-devel@googlegroups.com
To unsubscribe from this group, send email to
sage-devel-unsubscr...@googlegroups.com
For more options, visit this group at http://groups.google.com/group/sage-devel
URLs: http://www.sagemath.org
-~----------~----~----~----~------~----~------~--~---
