Graphics Reference
In-Depth Information
our paper [Gunther et al. 12], where further results and discussions can be found.
The simulation machinery is partly CUDA-based and partly rasterization-based
to exploit the advantages of both sides. In fact, it is driven by an interplay of
a number of modern GPU features, including GPU ray tracing, dynamic shader
linkage, tessellation, geometry shaders, rasterization-based splatting and trans-
form feedback, which are described in the following.
3.2 Overview
Our simulation of weathering effects is based on material transporting particles—
so-called gammatons, introduced in [Chen et al. 05]. Gammatons are emitted
from sources, such as clouds or roof gutters, to drip and distribute material, as
shown in Figure 3.2. On impact, material is deposited or taken away, and the
gammaton either bounces, floats, or is absorbed (see Figure 3.3). Eventually,
resting material is subject to aging rules, for instance, turning metal into rust or
spreading lichen growth. For this, the resting material is stored in a texture atlas,
which we call
material atlas
; see later in Section 3.3.1 for a detailed discription
of its content.
The simulation pipeline is illustrated in Figure 3.4. The particle movement
and collision detection are done in CUDA, using Nvidia's ray tracing engine Op-
tiX [Parker et al. 10]. Thus, in the
simulation step
a CUDA kernel is called, which
emits and traces gammatons (elaborated in Section 3.3.2). In the following two
stages, the
surface update
and the
gammaton update
, the material is transferred
Figure 3.2.
Patina and dirt being transported by gammatons. A subset of the gamma-
tons are rendered as spheres.
[Appeared in [Gunther et al. 12] and reproduced by kind
permission of the Eurographics Association.]
