Quaternions Revisited - GPU Pro: Advanced Rendering Techniques - page 374

Case Matrix/TBN Quaternion

Measured Value ALUs Memory ALUs Memory

VS → PS VS → PS Pipeline Stage VS PS VS VS PS VS

Normal Mapping 15 12 7 7 13 12 4 5

Instancing (Scale+Rot) 15 - 20 - 21 - 12 -

Skinning 33 - 64 - 71 - 16 -

Tab l e 1. 2. Comparison of 3 × 3 rotation matrices/TBN and quaternions performance (in
ALUs) and memory footprint (vertex attributes (TBN, instance data)/vertex texture
fetch (skinning) size in bytes (VS)/interpolator channels count (VS → PS)).

1.10 Conclusion

In this chapter we tried to fully cover our experience with quaternions. We have
used quaternions throughout the whole engine, significantly reducing memory
and bandwidth costs. There are, however, certain pitfalls, which we described,
that we hope will not prevent quaternions from replacing matrices in modern
engines.

1.11 Acknowledgements

The authors would like to thank Victor Surkov for helping with the illustrations
for this chapter.

Bibliography

[Hejl 04] Jim Hejl. “Hardware Skinning with Quaternions.” In Game Program-
ming Gems 4 , edited by Andrew Kirmse, pp. 487-495. Boston: Cengage
Learning, 2004.

[Lengyel 11] Eric Lengyel. Mathematics for 3D Game Programming and Com-
puter Graphics , Third edition. Boston: Cengage Learning PTR, 2011.

[Malyshau 12] Dzmitry Malyshau. “A Quaternion-Based Rendering Pipeline.”
In GPU Pro 3 , edited by Wolfgang Engel, pp. 265-273. Boca Raton, FL: A
K Peters/CRC Press, 2012.

[Mikkelsen 10] Morten S. Mikkelsen. “Bump Mapping Unparametrized Surfaces
on the GPU.” Journal of Graphics, GPU, and Game Tools 1 (2010), 49-61.

[Mittring 08] Martin Mittring. “Advanced Virtual Texture Topics.” In ACM
SIGGRAPH 2008 Games , pp. 23-51. New York: ACM, 2008.

[Schuler 06] Christian Schuler. “Normal Mapping without Precomputed Tan-
gents.” In ShaderX5: Advanced Rendering Techniques , edited by Wolfgang
Engel, pp. 131-140. Boston: Cengage Learning, 2006.