Analysis of Acceleration Structure Parameters and Hybrid Autotuning for Ray Tracing

Area of Science:

• Computer Graphics
• Computational Geometry
• Performance Optimization

Background:

• Optimizing acceleration structures (e.g., Bounding Volume Hierarchies - BVHs) is crucial for efficient ray tracing.
• Parameter interdependencies and context sensitivity complicate manual optimization.
• Previous autotuning methods show promise but can be inefficient for complex parameter spaces.

Purpose of the Study:

• To analyze the behavior and impact of acceleration structure parameters on building and rendering times.
• To develop an efficient autotuning method for ray tracing acceleration structures.
• To introduce a hybrid prediction and online autotuning approach for faster convergence.

Main Methods:

• Detailed analysis of parameter interdependence and context sensitivity (scene, viewpoint).
• Application of autotuning, previously successful on kD-trees, to BVHs.
• Development of a hybrid model-based prediction and online autotuning method.

Main Results:

• Identified parameter interdependence and context sensitivity, enabling targeted optimization.
• Online autotuning achieved up to 11% median speedup over literature configurations.
• The prediction model reached 95% of maximum autotuner speedup while reducing overhead by 90%.

Conclusions:

• Hybrid online autotuning provides an efficient and effective solution for optimizing ray tracing acceleration structures.
• The proposed method enables always-on tuning, significantly improving rendering performance.
• Context-sensitive parameter analysis allows for more focused and efficient optimization strategies.