Le_Forgeron <lef### [at] freefr> wrote:
> Nothing shoking, povray does not have an integer type, only float or
> double.

But double is way more than enough to handle 9 significant digits.

> Traditional float has 23 bits mantissa, that's about 7 significant digits.
> Traditional double has 52 bits mantissa, that's about 15.
> (expecting about 3 1/3 bits per decimal, or more exactly 10 bits per 3
> digits, as 1024 is 2^10, with an implicit bit in the non degenerated
> representation: float get value in 24, double in 53... yet )

I was going to look up the IEEE specs on that, but you've confirmed what I
thought.

> "Declare" are stored in float (well, IIRC, the type of float in povray
> can be customised via a macro, but you need to compile it yourself).

Why should this be, since POV-Ray uses double internally anyway?

And how is it that 3.6 could handle 9 significant digits with numbers on the
order of 10^4, but not with numbers on the order of 10^8?

Post a reply to this message