Experiences with Streaming Construction of SAH KD-Trees

Popov, Stefan; Günther, Johannes; Seidel, Hans‐Peter; Slusallek, Philipp

doi:10.1109/rt.2006.280219

Cited by 72 publications

(75 citation statements)

References 9 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…As an acceleration structure, it has been used in a variety of graphics applications, including triangle culling for ray-triangle intersection tests in ray tracing, nearest photon queries in photon mapping, and nearest neighbor search in point cloud modeling and particle-based fluid simulation. Due to its fundamental importance in graphics, fast kd-tree construction has been a subject of much interest in recent years, with several CPU algorithms proposed [Popov et al 2006;Hunt et al 2006;Shevtsov et al 2007]. However, real-time construction of kd-trees on the GPU remains an unsolved problem.…”

Section: Introductionmentioning

confidence: 99%

“…Specifically, our algorithm builds tree nodes in BFS (breadth-first search) order to fully exploit the fine-grained parallelism of modern GPUs at all stages of kd-tree construction. This is an important feature that distinguishes our work from previous parallel kd-tree algorithms including [Popov et al 2006;Shevtsov et al 2007], which resort to DFS (depth-first search) for nodes near the bottom of the kd-tree. Our algorithm builds kd-trees of comparable quality as those constructed by off-line CPU algorithms.…”

Section: Introductionmentioning

confidence: 99%

“…To address this issue, we derive novel schemes for the so-called large and small nodes. A node is deemed as large if the number of triangles in the node is greater than a user-specified threshold; otherwise it is small [Popov et al 2006;Shevtsov et al 2007]. For large nodes at upper tree levels, we use two simple and inexpensive heuristics, median splitting and "empty space maximizing" [Havran 2001;Wald and Havran 2006], to estimate the costs.…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Real-time KD-tree construction on graphics hardware

Zhou

Hou

Wang

et al. 2008

ACM SIGGRAPH Asia 2008 Papers

214

247

View full text Add to dashboard Cite

We present an algorithm for constructing kd-trees on GPUs. This algorithm achieves real-time performance by exploiting the GPU's streaming architecture at all stages of kd-tree construction. Unlike previous parallel kd-tree algorithms, our method builds tree nodes completely in BFS (breadth-first search) order. We also develop a special strategy for large nodes at upper tree levels so as to further exploit the fine-grained parallelism of GPUs. For these nodes, we parallelize the computation over all geometric primitives instead of nodes at each level. Finally, in order to maintain kd-tree quality, we introduce novel schemes for fast evaluation of node split costs.As far as we know, ours is the first real-time kd-tree algorithm on the GPU. The kd-trees built by our algorithm are of comparable quality as those constructed by off-line CPU algorithms. In terms of speed, our algorithm is significantly faster than well-optimized single-core CPU algorithms and competitive with multi-core CPU algorithms. Our algorithm provides a general way for handling dynamic scenes on the GPU. We demonstrate the potential of our algorithm in applications involving dynamic scenes, including GPU ray tracing, interactive photon mapping, and point cloud modeling.

show abstract

Section: Introductionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Real-time KD-tree construction on graphics hardware

Zhou

Hou

Wang

et al. 2008

ACM SIGGRAPH Asia 2008 Papers

214

247

View full text Add to dashboard Cite

show abstract

“…A common way to partition a kD-tree node is to consider split plane candidates at the AABB boundaries of all references in the node. Because this full SAH approach is expensive, binning [Hunt et al 2006;Popov et al 2006;Shevtsov et al 2007] is commonly used to speed up kDtree construction. With binning, split planes are considered only at a fixed number of equidistant positions within the node AABB.…”

Section: Chopped Binningmentioning

confidence: 99%

Spatial splits in bounding volume hierarchies

Stich

Friedrich

Dietrich

2009

Proceedings of the Conference on High Performance Graphics 2009

103

View full text Add to dashboard Cite

Figure 1: Sample scene consisting of roughly 1.9 million triangles (left). Our method (middle) results in a significant reduction of ray shooting costs compared to a regular bounding volume hierarchy (right). The heat views visualize the summed number of traversal steps and primitive intersections for primary rays. AbstractBounding volume hierarchies (BVH) have become a widely used alternative to kD-trees as the acceleration structure of choice in modern ray tracing systems. However, BVHs adapt poorly to nonuniformly tessellated scenes, which leads to increased ray shooting costs. This paper presents a novel and practical BVH construction algorithm, which addresses the issue by utilizing spatial splitting similar to kD-trees. In contrast to previous preprocessing approaches, our method uses the surface area heuristic to control primitive splitting during tree construction. We show that our algorithm produces significantly more efficient hierarchies than other techniques. In addition, user parameters that directly influence splitting are eliminated, making the algorithm easily controllable.

show abstract

“…For each such potential node, we can-using standard binning techniques in the spirit of [8,17]-determine if there is any good split for this node. Starting with a single node that contains all triangles, we can then greedily split the node with the biggest surface area until either no more split-able nodes are available, or until 16 nodes have been generated.…”

Section: Top-down Recursive Splittingmentioning

confidence: 99%

Getting rid of packets - Efficient SIMD single-ray traversal using multi-branching BVHs -

Wald

Benthin

Boulos

2008

2008 IEEE Symposium on Interactive Ray Tracing

View full text Add to dashboard Cite

Figure 1: The BART robots scene (71.7K triangle, 1 quad light), rendered with primary rays only, forced 2-bounce reflections, soft shadows (16 light samples), and a 2-bounce path tracer (16 samples per pixel). Though slower than aggressive packet/frustum techniques for the primary ray case, our single-ray based method is more efficient for the less coherent soft shadows and path traced images. ABSTRACTWhile contemporary approaches to SIMD ray tracing typically rely on traversing packets of coherent rays through a binary data structure, we instead evaluate the alternative of traversing individual rays through a bounding volume hierarchy with a branching factor of 16. Though obviously less efficient than high-performance packet techniques for primary rays, we demonstrate that for less coherent secondary ray distributions this approach is at least competitive with (and often faster than) typical packet traversal techniques.

show abstract

Experiences with Streaming Construction of SAH KD-Trees

Cited by 72 publications

References 9 publications

Real-time KD-tree construction on graphics hardware

Real-time KD-tree construction on graphics hardware

Spatial splits in bounding volume hierarchies

Getting rid of packets - Efficient SIMD single-ray traversal using multi-branching BVHs -

Contact Info

Product

Resources

About