Analysis of parallel spatial partitioning algorithms for GPU based DEM

2020 ◽  
Vol 125 ◽  
pp. 103708
Author(s):  
Retief Lubbe ◽  
Wen-Jie Xu ◽  
Daniel N. Wilke ◽  
Patrick Pizette ◽  
Nicolin Govender
2013 ◽  
Author(s):  
Raghuveer Devulapalli ◽  
Mikael Quist ◽  
John Gunnar Carlsson

2006 ◽  
Vol 128 (9) ◽  
pp. 945-952 ◽  
Author(s):  
Sandip Mazumder

Two different algorithms to accelerate ray tracing in surface-to-surface radiation Monte Carlo calculations are investigated. The first algorithm is the well-known binary spatial partitioning (BSP) algorithm, which recursively bisects the computational domain into a set of hierarchically linked boxes that are then made use of to narrow down the number of ray-surface intersection calculations. The second algorithm is the volume-by-volume advancement (VVA) algorithm. This algorithm is new and employs the volumetric mesh to advance the ray through the computational domain until a legitimate intersection point is found. The algorithms are tested for two classical problems, namely an open box, and a box in a box, in both two-dimensional (2D) and three-dimensional (3D) geometries with various mesh sizes. Both algorithms are found to result in orders of magnitude gains in computational efficiency over direct calculations that do not employ any acceleration strategy. For three-dimensional geometries, the VVA algorithm is found to be clearly superior to BSP, particularly for cases with obstructions within the computational domain. For two-dimensional geometries, the VVA algorithm is found to be superior to the BSP algorithm only when obstructions are present and are densely packed.


VLSI Design ◽  
2002 ◽  
Vol 15 (2) ◽  
pp. 485-489
Author(s):  
Youssef Saab

Partitioning is a fundamental problem in the design of VLSI circuits. In recent years, ratio-cut partitioning has received attention due to its tendency to partition circuits into their natural clusters. Node contraction has also been shown to enhance the performance of iterative partitioning algorithms. This paper describes a new simple ratio-cut partitioning algorithm using node contraction. This new algorithm combines iterative improvement with progressive cluster formation. Under suitably mild assumptions, the new algorithm runs in linear time. It is also shown that the new algorithm compares favorably with previous approaches.


2021 ◽  
Vol 2021 (12) ◽  
pp. 124016
Author(s):  
Samuel S Schoenholz ◽  
Ekin D Cubuk

Abstract We introduce JAX MD, a software package for performing differentiable physics simulations with a focus on molecular dynamics. JAX MD includes a number of physics simulation environments, as well as interaction potentials and neural networks that can be integrated into these environments without writing any additional code. Since the simulations themselves are differentiable functions, entire trajectories can be differentiated to perform meta-optimization. These features are built on primitive operations, such as spatial partitioning, that allow simulations to scale to hundreds-of-thousands of particles on a single GPU. These primitives are flexible enough that they can be used to scale up workloads outside of molecular dynamics. We present several examples that highlight the features of JAX MD including: integration of graph neural networks into traditional simulations, meta-optimization through minimization of particle packings, and a multi-agent flocking simulation. JAX MD is available at https://www.github.com/google/jax-md.


Sign in / Sign up

Export Citation Format

Share Document