Abstract
Complex systems simulations are well suited to the SIMT paradigm of GPUs, enabling millions of actors to be processed in fractions of a second. At the core of many such simulations, fixed radius near neighbours (FRNN) search provides the actors with spatial awareness of their neighbours. The FRNN search process is frequently the limiting factor of performance, due to the disproportionate level of scattered memory reads demanded by the query stage, leading to FRNN search runtimes exceeding that of simulation logic. In this paper, we propose and evaluate two novel optimisations (Strips and Proportional Bin Width) for improving the performance of uniform spatially partitioned FRNN searches and apply them in combination to demonstrate the impact on the performance of multi-agent simulations. The two approaches aim to reduce latency in search and reduce the amount of data considered (i.e. more efficient searching), respectively. When the two optimisations are combined, the peak obtained speedups observed in a benchmark model are 1.27x and 1.34x in two and three dimensional implementations, respectively. Due to additional non FRNN search computation, the peak speedup obtained when applied to complex system simulations within FLAMEGPU is 1.21x.
Highlights
Fixed Radius Near Neighbours (FRNN) search is central to many complex systems simulations, from molecular dynamics to crowd modelling
The query stage is possible with minimal branch divergence, which is optimal for the cohesive thread execution within vector units of Single Instruction Multiple Threads (SIMT) Graphics Processing Units (GPUs) architectures
This paper presents two independent techniques applicable to FRNN queries on GPUs using the Uniform Spatial Partitioning (USP) data structure
Summary
Fixed Radius Near Neighbours (FRNN) search is central to many complex systems simulations, from molecular dynamics to crowd modelling. Like many GPU algorithms, the query stage of FRNN search is bounded by latency, where maximal hardware utilisation is not achieved by either compute operations or memory transfers This has led researchers to seek out techniques to improve the speed at which FRNN queries can be executed [1, 2, 3, 4]. These two techniques reduce latency within the query stage of FRNN search by optimising code performance and reducing redundant data accesses via the adjustment of bin widths These optimisations are implementation agnostic, suitable for application to a broad range of USP problems. The remainder of this paper is organised as follows: Section 2 provides an overview of available techniques for performing FRNN searches, the technique of USP and prior techniques for its optimisation; Section 3.1 describes the theory and an example implementation of the Strips technique, for the reduction of compute within USP accesses; Section 3.2 describes the theory and an example implementation of the Proportional Bin Width technique, for the reduction of redundant memory accesses within FRNN search; Section 4 details how the technique has been implemented for experimentation; Section 5 explains the benchmark models that have been used for evaluation; Section 6 discusses the results obtained when comparing performance before and after the optimisations have been applied; Section 7 presents the concluding remarks and directions for further research
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a callDisclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.
