1 of 1

FunMC²: A Filter for Uncertainty Visualization of Marching Cubes

The Science�A team led by Oak Ridge National Laboratory researchers developed a data filter that accelerates uncertainty visualization of marching cubes, a widely used algorithm for understanding scientific data. The FunMC² filter works on multi-core CPUs and on NVIDIA and AMD GPUs.

The team tested this filter on ORNL’s Summit supercomputer and on Crusher, the test system for ORNL’s exascale supercomputer Frontier. The filter has been integrated with the ParaView production tool to support a variety of mesh structures for data visualizations used to show uncertainty in 3D simulations.

The Impact

Accurate interpretation of scientific data requires understanding the inherent degrees of uncertainty in any set of complex scientific data. Most data visualization tools offer only limited and expensive means of depicting uncertainties, hampering analysis.

An ORNL-led team developed a data filter, FunMC², that uses parallel implementation to balance the computational load and overcome obstacles to incorporating isosurface uncertainty in data visualization. The filter supports a variety of approaches to visualization to allow a more detailed depiction of uncertainty and a deeper, more accurate understanding of scientific data.

PI/Facility Lead(s): David Pugmire and Kenneth Moreland, ORNL

ASCR Program/Facility: Scientific Discovery Through Advanced Computing

ASCR PM: Ceren Susut

Funding: ASCR

Publication(s) for this work: Zhe Wang, Tushar M. Athawale, Kenneth Moreland, Jieyang Chen, Chris R. Johnson and David Pugmire, “FunMC²: A Filter for Uncertainty Visualization of Marching Cubes on Multi-Core Devices.” Eurographics Symposium on Parallel Graphics and Visualization, 2023.�DOI: 10.2312/pgv.20231081

The FunMC²filter being used in ParaView to visualize the uncertainty of isosurfaces, or 3D surface points with equal values, in data from a core-collapse supernova simulation.

Credit: ORNL

Team Members: Zhe Wang (ORNL), Tushar M. Athawale (ORNL), Kenneth Moreland (ORNL), Jieyang Chen (University of Alabama), Chris R. Johnson (SCI Institute, University of Utah) and David Pugmire (ORNL)

Talking Points:

A team led by Oak Ridge National Laboratory researchers developed a data filter that accelerates uncertainty visualization of marching cubes, a fundamental and widely used algorithm for building computer graphics, on devices with more than one core, or processor.
The FunMC2 filter, trained on ORNL’s Summit supercomputer and integrated with the ParaView production tool for interactive visualization, supports a variety of mesh structures for data visualizations used to show uncertainty in 3D simulations.
Accurate interpretation of scientific data requires understanding the inherent degrees of uncertainty in any set of complex scientific data. Most data visualization tools offer only limited and often expensive means of depicting these uncertainties, hampering effective analysis.
An ORNL-led team developed a data filter, FunMC2, that uses parallel implementation to balance the computational load and overcome the usual obstacles to incorporating uncertainty in data visualization. The filter supports a variety of approaches to visualization to allow a more detailed depiction of uncertainty and a deeper, more accurate understanding of scientific data.

Metadata:

Name of the associated awarded project: Scientific Discovery through Advanced Computing

PI name(s): David Pugmire and Kenneth Moreland, ORNL

Name of the program manager: Ceren Susut

CITATION: Zhe Wang, Tushar M. Athawale, Kenneth Moreland, Jieyang Chen, Chris R. Johnson and David Pugmire, “FunMC2: A Filter for Uncertainty Visualization of Marching Cubes on Multi-Core Devices.” Eurographics Symposium on Parallel Graphics and Visualization, 2023. * These authors contributed equally to this work.

DOI: 10.2312/pgv.20231081

AWARDS:

REPRODUCIBILITY:

BACKGROUND: The marching cubes algorithm for isosurface visualization is one of the fundamental data analysis techniques, but it does not consider uncertainty in data for surface extraction. Ignoring uncertainty inherent in data, however, may produce misleading visualizations, thereby reducing trustworthiness of analysis. Several recent papers have developed probabilistic versions of the marching cubes algorithm to understand propagation of uncertainty in data into isosurfaces. One of the major roadblocks that stands in the way of integrating these probabilistic algorithms in visualization software (e.g., ParaView and VisIt) is a significant computational overhead caused by carrying uncertainty with data. Such computational overhead can make uncertainty visualization excessively slow and impractical to implement in a visualization software. This project involved the development of a novel parallel algorithm that utilizes the VTK-m library and supercomputing capabilities for fast uncertainty visualization of the marching cubes algorithm. VTK-m is a part of the US Department of Energy’s (DOE’s) Exascale Computing Project (ECP) and provides a platform-portable framework to write scientific visualization algorithms that can utilize shared-memory parallelism. We successfully leveraged the power of VTK-m library to implement parallel algorithms and deployed them on OCLF ORNL GPU computing resources to achieve 5×-9× speedup compared to OpenMP and 295× speedup compared to serial implementation for efficient uncertainty analysis. Such remarkable speedup is a major advancement for making uncertainty visualization usable and accessible to domain users.

The significance of this work is that it addresses the major computational challenge associated with uncertainty visualization of the celebrated marching cubes visualization algorithm through parallel implementation. Additionally, the project showcases the integration of the “FunMC2” filter with the ParaView software which addresses the current limitation of lack of tools for efficiently conveying uncertainty in data to scientists. We demonstrated application of the “FunMC2” filter to efficiently convey uncertainty in isosurfaces extracted from various simulation data including supernova, oceanology, and climate science. The VTK-m implementation enables computing uncertainty in parallel for data sampled on structured as well as unstructured grids. Our parallel uncertainty visualization algorithms visualization will enable domain experts to efficiently gain insight into data uncertainty and make uncertainty-aware decisions in a timely manner.

Acknowledgement of Support:

ASCR