Fast Quadrangular Mass-Spring Systems using Red-Black Ordering
Paper i proceeding, 2018

We introduce a practical iterative solver for mass-spring systems which can be trivially mapped to massively parallel architectures, in particular GPUs.We employ our solver for the interactive animation of virtual cloth and show that it is computationally fast, robust and scalable, making it suitable for real-time graphics applications. Under the assumption that the input system is represented by a quadrangular network of masses connected by springs, we first partition the particles into two independent sets. Then, during the animation, the dynamics of all the particles belonging to each set is computed in parallel. This enables a full Gauss-Seidel iteration in just two parallel steps, leading to an approximated solution of large mass-spring systems in a few milliseconds. We use our solver to accelerate the solution of the popular Projective Dynamics framework, and compare it with other common iterative solvers in the current literature.

Projective Dynamics

Red-Black Gauss-Seidel

Parallel Computing

Position Based Dynamics

Författare

Oskar Nylèn

Deform Dynamics AB

Pontus Pall

Deform Dynamics AB

Marco Fratarcangeli

Chalmers, Data- och informationsteknik, Interaktionsdesign

VRIPHYS 2018 - 14th Workshop on Virtual Reality Interactions and Physical Simulations, VRIPHYS 2018


978-3-03868-059-8 (ISBN)

Workshop on Virtual Reality Interaction and Physical Simulation
Delft, Netherlands,

Interaktiva 3D deformerbara kroppars design

Vetenskapsrådet (VR) (2015-05345), 2016-01-01 -- 2018-12-31.

Styrkeområden

Informations- och kommunikationsteknik

Ämneskategorier

Inbäddad systemteknik

Datavetenskap (datalogi)

Datorsystem

DOI

10.2312/vriphys.20181066

ISBN

9783038680598

Mer information

Senast uppdaterat

2023-03-21