Kd-Jump: a path-preserving stackless traversal for faster isosurface raytracing on GPUs

Area of Science:

• Computer Science
• Graphics and Visualization

Background:

• Stackless traversal methods aim to reduce memory usage by replacing stack-based recursion with iterative computation.
• Evaluating the efficacy of these techniques on contemporary hardware, such as CUDA-enabled systems, is crucial for performance optimization.

Purpose of the Study:

• To investigate the applicability and performance of stackless traversal on modern hardware.
• To introduce a novel stackless approach for implicit kd-trees named Kd-Jump.

Main Methods:

• Developed Kd-Jump, a stackless traversal method for implicit kd-trees that uses index-based node traversal.
• Kd-Jump avoids global memory requirements, utilizing a small, fast constant-memory matrix.
• Introduced Hybrid Kd-Jump, combining Kd-Jump with a volume stepper and a dynamic depth threshold.

Main Results:

• Kd-Jump demonstrated a 10-20% performance improvement over stack-based traversal.
• Kd-Jump significantly outperformed the kd-restart method by 100%.
• Hybrid Kd-Jump leverages benefits like empty space removal and efficient texture caching.

Conclusions:

• Stackless traversal, specifically Kd-Jump, is highly effective on modern hardware like CUDA.
• The proposed Hybrid Kd-Jump offers a flexible and efficient approach for real-time rendering and isosurface extraction.