gdalgorithms-list

[Algorithms] VIPM With T&L

[Sam K. <sa...@ip...>]

Hi all,

This is bound to have been covered a gazillion times before, but I'm having
trouble searching the archives:
I'm fairly new to VIPM, and have a couple of questions:

a) I've read that its possible to throw the vertices of a model at the 3d
card (So T&L is possible)
And just maintain the indexed lists into that vertex array using collapses
and merges.
Doesn't that mean you have to make the card (hardware) transform all of the
vertices in the highest resolution representation of the model each frame?
even though the current lod might only need say 100.. or am I missing
something?

b) Say you wanted 1000 spaceships on screen sharing a single VIPM model, how
does this work with (a)?
Does the card have to transform all the original (high lod) vertices 1000
times to screen space?

Is it that the hardware T&L only transforms vertices that are pointed to by
the supplied indexed lists? or what

Thanks,

sam
sa...@ip...

Re: [Algorithms] VIPM With T&L

[Jim O. <j.o...@in...>]

> Is it that the hardware T&L only transforms vertices that are pointed to
by
> the supplied indexed lists? or what

That's it :-).

For software T&L you can sort your verts (this is not necessarily specific
to VIPM btw.), so that verts [0...n] represent the first (lowest) lod level,
verts [0...n+m] represents the next lod level and so forth. Using the
dwStartVertex and dwNumVertices parameters in the DIPVB call you can
instruct DX to transform *only*  those verts that fall in the specified
range.

Jim Offerman
Innovade


The Atlantis Project
http://www.theatlantisproject.com

Re: [Algorithms] VIPM With T&L

[Dave E. <eb...@ma...>]

From: "Sam Kuhn" <sa...@ip...>
> a) I've read that its possible to throw the vertices of a model at the 3d
> card (So T&L is possible)
> And just maintain the indexed lists into that vertex array using collapses
> and merges.

The source code for my game engine book has an implementation
of this (actually, the first update does and is available at my web site
for purchasers of the book).  The vertices and indices are ordered so
that as you collapse edges, you just reduce the maximum index for
the vertex array and for the connectivity index array.

> Doesn't that mean you have to make the card (hardware) transform all of
the
> vertices in the highest resolution representation of the model each frame?
> even though the current lod might only need say 100.. or am I missing
> something?

The packed model-space vertices can be cached on the graphics
card.  Each frame you only change the connectivity information and
the model-to-world transforms.  Then let the graphics card do its thing.

> b) Say you wanted 1000 spaceships on screen sharing a single VIPM model,
how
> does this work with (a)?

Yes, both objects share the model-space data that is cached on
the card.  The objects potentially have different LODs and
model-to-world transforms.

--
Dave Eberly
eb...@ma...
http://www.magic-software.com