I haven't heard of anyone porting the exact algorithm to the GPU, but
Chris Dyken (
http://heim.ifi.uio.no/~erikd/index.html) and Gernot
Ziegler (
http://www.mpi-inf.mpg.de/~gziegler/) recently did some very
cool work on a GPU implementation of marching cubes that doesn't
require geometry shaders. It therefore runs on older hardware
(unlike eg: NVIDIA's implementation described in GPU Gems 3 which
requires a GeForce 8), and it outperforms the geometry shader
implementation too.
I didn't investigate deeply enough to say it with confidence, but it
seems likely that the approach above could be used to accelerate
tessellation of the the implicit surfaces defined by ShapeShop
(
http://www.shapeshop3d.com/). The result would be a Teddy-like UI,
but fast fast fast.
A more direct port of Igarashi's algorithm might be possible. We've
known how to compute Voronoi regions on the GPU since before shaders
were around, but I don't know how easy it would be to transform this
into the Delaunay triangulation (the dual of the Voronoi diagram) on
the GPU. Very quick profiling shows that about a third of the
runtime is doing a flood-fill, so we'd have to figure out how to do
that on the GPU too if we want more than a 3x speedup.
Josh
On Nov 17, 2007, at 7:05 AM, David Faught wrote:
> I also wonder how hard it might be to implement Takeo Igarashi's
> sketch plumping method (as modified by Andreas' grad student whose
> name escapes me, and which is used in Croquet's TPainter and
> CCPainter) on the GPU in a few passes ...
>
> On 10/29/07, David Faught <
[hidden email]> wrote:
>> Gee, I wonder if there might be some really spiffy application of
>> OMeta for
>> parsing lots of different 3D file formats?
>>
>>
http://www.cs.ucla.edu/~awarth/papers/dls07.pdf>>
http://www.cs.ucla.edu/~awarth/ometa/>>
Locked
