Graphics Reference
In-Depth Information
performance is about three times slower (using paged allocation, which is then
faster), making Pre-Open competitive again. This implies that in a non-GLSL
setting the performance ratios are likely to differ.
Finally, the Post-sort techniques could benefit from a smarter sort, bubble-
sort having the advantage of fitting well in the Render pass due to its straight-
forward execution pattern. Using a more complex algorithm would be especially
beneficial for larger depth complexities. However, increasing the number of frag-
ments per-pixel implies increasing the reserved temporary memory. This impedes
performance: for the rendering of Figure 1.3, allocating a temporary array of size
64 gives a Render time of 20ms, while using an array with 256 entries increases
the Render time to 57ms. This is for the exact same rendering: reserving more
memory reduces parallelism. In contrast, the Pre-sort techniques suffer no such
limitations and support early fragment culling.
gpupro5.
1.9 Acknowledgments
We thank NVIDIA for hardware donation as well as Cyril Crassin for discussions
and feedback on GPU programming. This work was funded by ERC ShapeForge
(StG-2012-307877).
Bibliography
[Carpenter 84] Loren Carpenter. “The A-buffer, an Antialiased Hidden Surface
Method.”
SIGGRAPH
18:3 (1984), 103-108.
[Celis et al. 85] Pedro Celis, Per-Ake Larson, and J. Ian Munro. “Robin Hood
Hashing (Preliminary Report).” In
Proceedings of the 25th Annual Sympo-
sium on Foundations of Computer Science
, pp. 281-288. Washington, DC:
IEEE, 1985.
[Crassin 10] Cyril Crassin. “OpenGL 4.0+ A-buffer V2.0: Linked
opengl-40-abuffer-v20-linked-lists-of.html, 2010.
[Garcıa et al. 11] Ismael Garcıa, Sylvain Lefebvre, Samuel Hornus, and Anass
Lasram. “Coherent Parallel Hashing.”
ACM Transactions on Graphics
30:6
(2011), Article no. 161.
[Harris 01] Timothy L. Harris. “A Pragmatic Implementation of Non-blocking
Linked-Lists.” In
Proceedings of the 15th International Conference on Dis-
tributed Computing
,
DISC '01
, pp. 300-314. London: Springer-Verlag, 2001.