Presentation

<
Fast Sparse Matrix Permutation for Mesh-Based Direct Solvers
SessionNumerical Methods for Geometry Processing
DescriptionWe present a fast sparse matrix permutation algorithm tailored to mesh-based direct solvers. By exploiting mesh structure through patch-based nested dissection, our method reduces permutation overhead while preserving high-quality orderings, leading to substantial end-to-end speedups in sparse Cholesky factorization on both CPUs and GPUs.
Authors
Behrooz Zarebavani
University of Toronto
Ahmed Mahmoud
Massachusetts Institute of Technology (MIT)
Ana Dodik
Massachusetts Institute of Technology (MIT)
Changcheng Yuan
Texas A&M University
Serban Porumbescu
University of California Davis
John Owens
University of California Davis
Maryam Mehri Dehnavi
University of Toronto
NVIDIA
Justin Solomon
Massachusetts Institute of Technology (MIT)
Event Type
Technical Paper
TimeWednesday, 22 July 20263:55pm - 4:05pm PDT
LocationRoom 403 B
Digital Library PDF
Session Time & Location
Sunday, 19 July 20266:00pm - 8:45pm PDTHall K
Wednesday, 22 July 20263:45pm - 5:35pm PDTRoom 403 B
Keywords
Geometry
Registration Categories
Full Conference Supporter
Full Conference
Next PresentationNext PresentationStatistical Gradient Filtering for Geometry Optimization Under Limited Observations
Similar Presentations
Iskra: A System for Inverse Geometry Processing
Schrödinger Bridges on Discretized Geometric Domains
Locality-Aware Automatic Differentiation on the GPU for Mesh-Based Computations