I think i know the problem. (i'm
bad i never brought it up before
whip me beat me make me write bad
cheques. =} )

It has to do with the reflection blur.

but it crash dumps if you don't use fastmath
and loops if you do (dunno why just what
happened with me and i never tested the fix
with fast math)

what hapens is in the following snippet of code of reflect:

z = (FRAND() * 2) - 1;
t = FRAND() * M_PI * 2;
r = sqrt(1 - z*z);
x = r * cos(t);
y = r * sin(t);
Make_Vector(Jitter_Offset, x, y, z);


if FRAND ****EVER**** returns something larger then
1 (lets say its really bad and it return 1.01)
z = 1.01 * 2 - 1 = 1.02
r = sqrt( 1 - z*z) = sqrt( 1 - 1.004) 
	= sqrt (- 0.004) = NAN 

so your vector becomes corrupt and either
a test down the line fails and it trys to
acess a pointer that should be null
or a depth intersection test loops.

the fix is in texture.h

is to change 

#define RNDMULTIPLIER ((DBL)0.000030518509476)
cause perl -e 'print 0.000030518509476*(2**15-1);'
1.00000000000009


so i change it to around:
#define RNDMULTIPLIER ((DBL)0.0000305185094759972)
(or round DOWN)
#define RNDMULTIPLIER ((DBL)0.000030518509475)

perl -e 'print 0.0000305185094759972*(2**15-1);'
1

perl -e 'print 0.000030518509475*(2**15-1);'
0.9999999999673251

Carl Bartels wrote:
> 
> Actually, I was using my own compile at the time (under Linux).  I get
> similar results with Marks version, I just hadn't had a chance to play
> with it yet at that point last night.  The difference between the
> behavior of my compile and Marks is that his kindly dumps core while
> mine just sits there all locked up with the image half rendered (kill -9
> is the only way to stop it.) I've found a few more test images that
> it'll do this for, including the chess2 image that comes with the
> official distribution (I figured I'd check to see if the traditional
> images still worked.)
> 
> Mark Gordon wrote:
> >
> > Nathan Kopp wrote:
> > >
> > > Strange.  I just rendered it with no problems.  How widespread is this
> > > problem?  I'll upload the EXE again (the EXE that I just used to
> > > successfully render the scene).
> >
> > Carl is using a Linux version.  I'm not sure if he's using the one he
> > made himself or mine.  Mine is dumping core for me in this file.  I
> > suspect some of the existing code has previously undiagnosed platform
> > dependencies that I haven't chased down yet.  I'll see what I can find
> > out.
> >
> > -Mark Gordon
> 
> --
> Carl Bartels, Department of Chemsitry, Mcgill University, to reply to
> me,
> just kill a and 5 from the email name, Montreal, QC, cAnAdA

-- 
Matthew Corey Brown                                         XenoArch
mcb### [at] xenoarchcom                             http://www.xenoarch.com

Post a reply to this message