Graphics Reference
In-Depth Information
VI
Compute
Graphics processing units are becoming more capable every year of processing
general purpose calculations. The new console generation is able to execute com-
pute tasks asynchronously. The game industry re-thinks traditional algorithms
to utilize the thread group shared memory or the on-chip cache in better ways
and sees substantial gains in performance.
The Compute section starts with the chapter by Dongsoo Han, “Hair Simula-
tion in TressFX.” It describes the simulation algorithms used in the game
Tomb
Raider
to animate Lara Croft's hair and in the AMD demo
Ruby
. TressFX can
simulate 19,000 strands of hair, consisting of 0.22 million vertices in less than a
millisecond on a high-end GPU.
The next chapter, “Object-Order Ray Tracing for Fully Dynamic Scenes”
by Tobias Zirr, Hauke Rehfeld, and Carsten Dachsbacher, describes the imple-
mentation of a ray tracer that is capable of utilizing the same art assets as a
rasterizer-based engine, which is one of the major obstacles in using ray tracers
in games.
The chapter “Quadtrees on the GPU” by Jonathan Dupuy, Jean-Claude Iehl,
and Pierre Poulin describes a quadtree implementation that is implemented with
linear trees on the GPU. The chapter explains how to update linear trees, which
are a pointer-free alternative to recursive trees on the GPU, and how to render
multiresolution, crack-free surfaces with frustum culling using hardware tessella-
tion and a distance-based LOD selection criterion.
A key feature for simulating rigid bodies on the GPU is the performance of the
constraint solver. Takahiro Harada shows in his chapter “Two-Level Constraint
Solver and Pipelined Local Batching for Rigid Body Simulation on GPUs” an
implementation that uses pipelined batching to run the entire solver on the GPU.
This constraint solver is also implemented in the Bullet Physics Engine.
The last chapter in the series, “Non-separable 2D, 3D, and 4D Filtering with
CUDA” by Anders Eklund and Paul Dufort, covers the ecient implementation
of filtering algorithms tailored to deal with 3D and 4D datasets, which commonly
occur in medical imaging, and 2D datasets.
—Wolfgang Engel
