Accelerating Unstructured Mesh Point Location With RT Cores.

Area of Science:

• Computer Graphics
• GPU Computing
• Scientific Visualization

Background:

• Point location in unstructured meshes is vital for volume rendering and compute tasks.
• Accelerating these queries on GPUs has been challenging.
• Existing methods struggle with general planar and bilinear element meshes.

Purpose of the Study:

• To leverage novel ray tracing hardware on Nvidia RTX GPUs for accelerating point location queries.
• To develop and evaluate techniques for handling general unstructured elements (planar and bilinear faces).
• To improve the performance and robustness of point-in-element tests for GPU architectures.

Main Methods:

• Reformulated point location queries into ray queries.
• Utilized GPU ray tracing cores for accelerated Bounding Volume Hierarchy (BVH) traversal.
• Employed ray-triangle intersections and per-face metadata for efficient point-in-element detection.
• Compared three variants against a CUDA reference method.

Main Results:

• Achieved significant speedups compared to the CUDA reference method.
• Improved unstructured volume renderer performance by up to 4x for tetrahedral meshes.
• Enhanced performance by up to 15x for general bilinear element meshes.
• Outperformed state-of-the-art solutions in speed, robustness, and ease of implementation.

Conclusions:

• The proposed technique effectively utilizes ray tracing hardware for accelerating unstructured mesh point location.
• This approach offers substantial performance gains for critical applications like volume rendering.
• The method provides a robust and easier-to-implement solution compared to previous techniques.