|
![](/i/fill.gif) |
On Tue, 11 Feb 2003 06:26:55 +0100, Frank 'Sputnik' Rothfuß
<fr### [at] computermuseum fh-kiel de> wrote:
> Of course floating point calculations can't
> be perfect, but why is 3.1g THREE ORDERS OF MAGNITUDE better than 3.5.?
> I think this is a bug that should be corrected.
That's not a bug, imo, but mentioned floating point accuracy. You can
reproduce the same floating point accuracy error only making the same
calculations over the same values with the same type of FPU. Every changed
line where operations on floats are performed changes floating point accuracy.
Maintaining the same lines in sources only to reproduce the same floating
point accuracy simple leads to not changing sources at all so why release 3.5
at all? The problem in your example is that while screen and objects are
smaller, direction of camera ray has always length 1. The same for normals
after intersection. Performing calculations over values with very different
mantysa (correct word?) always increase floating point accuracy. You should be
happy that you still not found floating point accuracy problem like this:
http://news.povray.org/povray.beta-test.binaries/18705/
ABX
Post a reply to this message
|
![](/i/fill.gif) |