CAMA: Contact‐Aware Matrix Assembly with Unified Collision Handling for GPU‐based Cloth Simulation

Abstract

Figure 1: Benchmark Andy: Our GPU-based approach can simulate the clothes dressed on a Kung-Fu boy. The meshes of three cloth pieces are represented by 127K triangles. Our simulator performs all of the computations, including implicit time integration and collision handling, in 2.42s per frame (on average) on an NVIDIA Telsa K40c GPU. Our new parallel algorithms for sparse matrix assembly and collision handling result in significant speedups over prior methods. AbstractWe present a novel GPU-based approach to robustly and efficiently simulate high-resolution and complexly layered cloth.The key component of our formulation is a parallelized matrix assembly algorithm that can quickly build a large and sparse matrix in a compressed format and accurately solve linear systems on GPUs. We also present a fast and integrated solution for parallel collision handling, including collision detection and response computations, which utilizes spatio-temporal coherence. We combine these algorithms as part of a new cloth simulation pipeline that incorporates contact forces into implicit time integration for collision avoidance. The entire pipeline is implemented on GPUs, and we evaluate its performance on complex benchmarks consisting of 100 − 300K triangles. In practice, our system takes a few seconds to simulate one frame of a complex cloth scene, which represents significant speedups over prior CPU and GPU-based cloth simulation systems. † {tang_m, tangle,

Efficient BVH‐based Collision Detection Scheme with Ordering and Restructuring

“…A BVH‐based CD pipeline using the incremental scheme was proposed by [TMLT11,TWT*16]. In their GPU‐streams, the collision pipeline involves a one‐time preprocessing stage and a runtime stage for collision queries.…”

“…We implement two versions of our CD pipeline. The first version is embedded in the ARCSim simulator in [TWT*16]. All primitives are categorized into a deformable BVH and a rigid BVH.…”

“…[LC98] first came up with an incremental scheme that takes advantage of spatial coherence to accelerate CD, which is also known as generalized front tracking [EL01]. It was later used in a GPU‐based streaming algorithm by [TWT*16, TMLT11, DZPW15], which to our knowledge demonstrates the best CD performance among BVH‐based methods so far.…”

Time‐Domain Parallelization for Accelerating Cloth Simulation

“…[NKT15] improved the work from [AVGT12] using Asynchronous Contact Mechanics and reduced the communication by proposing a locality‐aware task assignment, which first scaled more than 16 cores. [TWT*16] implemented a GPU‐based simulation pipeline. Their method has achieved an impressive speedup of 58 times, which is comparable to the performance of our method on a 64‐core cluster.…”

“…On a given set of benchmarks, our method achieves an unprecedented level of scalability in distributed CPU systems when compared to [ZFV04, NKT15]. Its performance gain is also higher than the GPU parallelization [TWT*16], while our approach offers the additional flexibility for coupling with adaptively remeshed cloth simulators. We also verify that given sufficient amount of processors, our method can achieve an average performance as fast as the low‐resolution simulation, while obtaining simulation results similar to ones using high‐resolution meshes.…”