Graphics Reference
In-Depth Information
Figure 3.5.
Crack-free, multiresolution parametric surfaces produced on the GPU with
our renderer.
Could fractional tessellation be used to produce continuous subdivisions?
If the T-
junctions are removed with the CDLOD morphing function [Strugar 09], any
mode can be used. Our proposed alternative approach does not have restrictions
regarding the inner tessellation levels, but the outer levels must be set to exact
powers of two. Therefore, fractional odd tessellation will not work. The even
mode can be used, though.
Are there any significant reasons to use tessellation rather than the CDLOD mor-
phing function for T-junctions?
The only advantage of tessellation is flexibility,
since arbitrary factors can be used. For some parametric surfaces, tessellation
turned out to be essential because the quadtree we computed was not necessarily
restricted. The phenomenon is visible in Figure 3.5, on the blue portions of the
trefoil knot. If we had used the CDLOD method, a crack would have occurred.
For terrains, the distance-based criterion ensures that the nodes are restricted,
so both solutions are valid.
There are two ways to increase polygon density: either use the GPU tessellator,
or refine the instanced grid. Which approach is best?
This will naturally depend
on the platform. Our renderers provide tools to modify both the grid and the
GPU tessellation values, so that their impact can be thoroughly measured. On
the Radeon 7950, we observed that tessellation levels could be increased up to
a factor of 8 without performance penalties, as long as the average on-screen
polygon size remained reasonable.
