[ODE] GIMPACT with doubles

[Pierre Terdiman]

?

I don't see why the "all doubles" version is better for this. As far as I
can see using 3 doubles for the world position and keeping all vertices in
floats (*relative to the mesh position in doubles*) is just as accurate.

It's likely that the meshes are authored (and exported) using floats, so
converting them to doubles doesn't buy you more accuracy. It just adds
noise...

- Pierre

----- Original Message ----- 
From: "Jon Watte (ODE)" <hplus-ode at mindcontrol.org>
To: "Pierre Terdiman" <pierre.terdiman at novodex.com>
Cc: <ode at q12.org>
Sent: Friday, November 03, 2006 7:48 PM
Subject: Re: [ODE] GIMPACT with doubles


> You'll get into pinhole problems where the meshes line up, if you try to
> do world-scale simulation with floating point meshes. There are ways
> around it (adding skirts, for example), but it's annoying -- the
> formulation using doubles is cleaner (although uses more memory, thus
> making it slower).
>
> Cheers,
>
>           / h+
>
>
> Pierre Terdiman wrote:
> > Note that the only place where doubles are actually needed is the
> > world-space position of your objects. If everything else works in
relative
> > space (e.g. collision routines), then the extra accuracy doesn't buy you
> > anything. In particular, storing the mesh vertices in doubles is usually
not
> > needed.
> >
> > - Pierre
> >
>