紀要論文

GPGPUを用いたエージェントシミュレーションの高速化支援

Support for acceleration of agent simulation using GPGPU

jpn

24321192

https://doi.org/10.15002/00021520

法政大学大学院情報科学研究科

法政大学大学院紀要. 情報科学研究科編

13

1

6

2017-03-31

Version of Record

Recently, the number of station buildings and commercial facilities has increased due to redevelopment of cities, emergence of new business etc. In designing these facilities simulations are performed for predicting phenomenon caused by a very number of people aiming for example, to selecting escape routes and promote sales promotion. However, processing by the CPU requires much time for processing. Also, although processing can be performed at high speed by processing using the GPU, it is necessary to acquire knowledge of GPU architecture and special language. In this research, we propose a framework for reducing execution cost in agent simulation. In this framework, the development cost is reduced by processing the simulation with the GPU while concealing the GPGPU-specific commands to the developer. The framework also provides an optimization support for reducing divergence and shorten processing time at runtime. As a confirmation of the usefulness of the framework, we compare the execution time of simulation by CPU and GPU, discuss the description when using this framework, and evaluate this research.

Article

AA12746425