kicad-developers team mailing list archive

[Lorenzo Marcantonio <l.marcantonio@xxxxxxxxxxxx>]

On Wed, Nov 23, 2011 at 10:11:38AM -0600, Dick Hollenbeck wrote:

> So we can go to 64 bit integers, or should I say, "evolve through" 64 integers as part of
> our migration path.  Eventually I think we can get back to 32 bit integers.   There may be
> a quick reward to travel through 64 bit integers temporarily.

Are you sure we actually need 64 bits? Or are you talking about intermediate
calculations?

IIRC the 10 nm unit was good enough to fit in 32 for every 'practical' design
(some thread ago, I don't remember the actual values).  OTOH if you needed to,
say, compute a distance the intermediate squared value would overflow (but then
you would use a double anyway because there isn't a integer sqrt function...)

But since multiplication between coordinates happens often in analytical
geometry I agree that the type for intermediate calculation *could* be an
issue. From a performance point of view there shouldn't be a lot of difference
between a long long and a double, anyway... using a long long for *every*
coordinate type seems overkill to me (unless you specifically target for 64 bit
systems)

-- 
Lorenzo Marcantonio
Logos Srl