Approximate Puzzlepiece Compositing

Area of Science:

• Scientific Visualization
• High-Performance Computing (HPC)

Background:

• Adaptive Mesh Refinement (AMR) and unstructured meshes are crucial for efficient large-scale simulations.
• Data partitioning for HPC often results in complex, non-convex boundaries, hindering standard visualization techniques.
• In situ and post hoc visualization face challenges due to the difficulty of re-partitioning complex mesh data.

Purpose of the Study:

• To develop a novel distributed volume rendering and compositing algorithm for AMR and unstructured meshes.
• To enable fast, high-accuracy, in-place rendering of complex simulation data.
• To overcome limitations of standard rendering techniques that require convex and disjoint data partitioning.

Main Methods:

• Introduced Approximate Puzzlepiece Compositing, a distributed volume rendering algorithm.
• Leveraged Moment-Based Ordered-Independent Transparency for scalable, order-independent compositing.
• Designed an algorithm requiring minimal communication and no data re-partitioning.

Main Results:

• Demonstrated fast and high-accuracy in-place rendering of AMR and unstructured meshes.
• Achieved scalable, order-independent compositing with minimal overhead at higher core counts.
• Validated performance and image quality against state-of-the-art sort-last compositing techniques on HPC systems.

Conclusions:

• Approximate Puzzlepiece Compositing offers a scalable, high-performance, and high-quality rendering solution.
• The algorithm is applicable to complex data distributions in large-scale Computational Fluid Dynamics (CFD) simulations.
• Eliminates the need for costly data re-partitioning, facilitating efficient in situ and post hoc analysis.