[ODE] OPCODE vs. double

[Timothy J. Wood]

--Apple-Mail-2--250306958
Content-Transfer-Encoding: quoted-printable
Content-Type: text/plain;
	charset=ISO-8859-1;
	format=flowed


   No, I understand all that and I'm not using D3D.

   The issue is the logical contradition between ODE and OPCODE.  ODE=20
seems to require double to be highly stable (or D3D's forcing single=20
precision wouldn't be an issue).  OPCODE requires single or it doesn't=20=

work.

   Thus, *neither* precision choice seems workable right now from what=20=

I'm seeing on this list.  Am I missing something?

-tim



On Sunday, July 6, 2003, at 10:44 AM, DjArcas wrote:

> The simplest solution is to create your 3d device with=20
> D3DCREATE_FPU_PRESERVE when your program requires great accuracy, IE a=20=

> physics system. There is a performance hit, at least theoretically -=20=

> I've not seen anything change on my system, and I'm chunking a lot of=20=

> polys at the screen. Changing ODE's accuracy to cope with something=20
> that D3D can be told NOT to do doesn't seem like the best solution,=20
> IMO.
> =A0
> =A0
> ----- Original Message -----
> From: "Timothy J. Wood" <tjw@omnigroup.com>
> To: <ode@q12.org>
> Sent: Sunday, July 06, 2003 6:19 PM
> Subject: [ODE] OPCODE vs. double
>
> >
> >
> > =A0=A0 I'm sort of worried since I've seen people on this list =
mentioned
> > that D3D breaks their simulation (since it gets rid of the extra
> > precision by default), but then OPCODE requires single precision.
> >
> > =A0=A0 So, what's going to give?=A0 Are we just supposed to abandon =
double
> > support in ODE or should OPCODE be made to work with double for when
> > you need precision more than space?=A0 Can ODE be made to lose less
> > precision in floats (seems unlikely)?=A0 Other solutions?
> >
> > -tim
> >
> > _______________________________________________
> > ODE mailing list
> > ODE@q12.org
> > http://q12.org/mailman/listinfo/ode
> >=

--Apple-Mail-2--250306958
Content-Transfer-Encoding: quoted-printable
Content-Type: text/enriched;
	charset=ISO-8859-1



  No, I understand all that and I'm not using D3D.


  The issue is the logical contradition between ODE and OPCODE.  ODE
seems to require double to be highly stable (or D3D's forcing single
precision wouldn't be an issue).  OPCODE requires single or it doesn't
work.


  Thus, *neither* precision choice seems workable right now from what
I'm seeing on this list.  Am I missing something?


-tim




On Sunday, July 6, 2003, at 10:44 AM, DjArcas wrote:


<excerpt><fontfamily><param>Arial</param><smaller>The simplest
solution is to create your 3d device with D3DCREATE_FPU_PRESERVE when
your program requires great accuracy, IE a physics system. There is a
performance hit, at least theoretically - I've not seen anything
change on my system, and I'm chunking a lot of polys at the screen.
Changing ODE's accuracy to cope with something that D3D can be told
NOT to do doesn't seem like the best solution, =
IMO.</smaller></fontfamily><fontfamily><param>Times</param>

=A0

=A0

</fontfamily><fontfamily><param>Arial</param><smaller>----- Original
Message -----</smaller></fontfamily><fontfamily><param>Times</param>

</fontfamily><fontfamily><param>Arial</param><smaller>From: "Timothy
J. Wood"
=
<<<underline><color><param>1999,1999,FFFF</param>tjw@omnigroup.com</color>=
</underline>></smaller></fontfamily><fontfamily><param>Times</param>

</fontfamily><fontfamily><param>Arial</param><smaller>To:
=
<<<underline><color><param>1999,1999,FFFF</param>ode@q12.org</color></unde=
rline>></smaller></fontfamily><fontfamily><param>Times</param>

</fontfamily><fontfamily><param>Arial</param><smaller>Sent: Sunday,
July 06, 2003 6:19 =
PM</smaller></fontfamily><fontfamily><param>Times</param>

</fontfamily><fontfamily><param>Arial</param><smaller>Subject: [ODE]
OPCODE vs. double</smaller></fontfamily><fontfamily><param>Times</param>


</fontfamily><fontfamily><param>Arial</param><smaller>>

>

> =A0=A0 I'm sort of worried since I've seen people on this list =
mentioned

> that D3D breaks their simulation (since it gets rid of the extra

> precision by default), but then OPCODE requires single precision.

>

> =A0=A0 So, what's going to give?=A0 Are we just supposed to abandon =
double

> support in ODE or should OPCODE be made to work with double for when

> you need precision more than space?=A0 Can ODE be made to lose less

> precision in floats (seems unlikely)?=A0 Other solutions?

>

> -tim

>

> _______________________________________________

> ODE mailing list

> <underline><color><param>1999,1999,FFFF</param>ODE@q12.org

</color></underline>>
=
<underline><color><param>1999,1999,FFFF</param>http://q12.org/mailman/list=
info/ode

</color></underline>></smaller></fontfamily></excerpt>=

--Apple-Mail-2--250306958--