Memory-Aware External Facelist Calculation: A Data-Parallel Atomic Hash Counting Approach

Area of Science:

• Scientific Visualization
• Computational Geometry
• High-Performance Computing

Background:

• Unstructured volumetric meshes are vital for scientific simulations and analyses, particularly in finite element analysis.
• Calculating the external facelist is a fundamental and performance-critical step in scientific visualization for rendering polygonal meshes.
• Existing methods in libraries like VTK and Viskores have performance and memory limitations.

Purpose of the Study:

• To assess the performance and memory constraints of current external facelist calculation algorithms.
• To develop and introduce a novel, memory-aware external facelist calculation algorithm.
• To optimize the external facelist calculation for efficient execution on many-core architectures.

Main Methods:

• Exploration of existing external facelist algorithms in VTK and Viskores.
• Development of a new memory-aware algorithm utilizing an atomic hash counting approach.
• Leveraging Viskores' data-parallel primitive operations for broad architectural compatibility.

Main Results:

• The novel algorithm demonstrates the lowest GPU memory footprint and second-lowest CPU memory footprint among evaluated methods.
• The proposed algorithm achieves the fastest execution performance on both CPU and GPU.
• The new method is integrated into VTK and Viskores as an open-source solution.

Conclusions:

• The developed memory-aware external facelist algorithm significantly improves performance and memory efficiency.
• This advancement benefits scientific visualization by accelerating mesh processing on modern hardware.
• The open-source availability promotes wider adoption and further development in the scientific community.