owensgroup/RXMesh

What it solves

RXMesh provides a high-performance way to process triangle meshes entirely on the GPU, overcoming the limitations of traditional CPU-based processing. It specifically handles meshes of any quality, including non-manifold meshes, and supports both static and dynamic mesh connectivity changes.

How it works

The library uses a compact, generic data structure designed for GPU acceleration. It includes a tightly coupled sparse and dense matrix infrastructure that leverages cuSolver, cuSparse, and cuBlas for geometry processing. Additionally, it implements a GPU-based Automatic Differentiation (AD) system to efficiently compute gradients and Hessians for optimization and simulation tasks.

Who it’s for

It is designed for researchers and developers working in geometry processing, simulation, and inverse design who need high-performance GPU acceleration for mesh-based computations.

Highlights

• GPU-accelerated processing for both static and dynamic meshes.
• Support for non-manifold mesh geometry.
• Integrated sparse and dense matrix operations via NVIDIA CUDA libraries.
• Built-in Automatic Differentiation for computing gradients and Hessians on the GPU.