Etd

Programming Abstractions for GPU-Accelerated Agent-Based Simulations

Public Deposited

Downloadable Content

open in viewer

Creating accelerated agent-based simulations (ABS) that run on the GPU has historically required knowledge of multiple programming languages and low-level graphics technologies. We propose that developers could instead use a single, high-level programming language that abstracts away many of the complexities associated with GPU acceleration of ABS. To test this we developed Apiary, an ABS framework that transforms dataflow representations of simulation logic into low-level GLSL code through an intermediate representation. We discuss the technical implementation of Apiary, the authoring abstractions we developed for it, and its performance characteristics in three different simulations. Apiary's engine outperforms popular open-source and commercial ABS packages that run on the CPU by multiple orders of magnitude and is accompanied by an authoring environment using a single language for both simulation and representation.

Creator
Contributors
Degree
Unit
Publisher
Identifier
  • etd-105806
Keyword
Advisor
Defense date
Year
  • 2023
Date created
  • 2023-04-27
Resource type
Source
  • etd-105806
Rights statement
Last modified
  • 2024-09-09

Relations

In Collection:

Items

Items

Permanent link to this page: https://digital.wpi.edu/show/8w32r923m