«Life» in Tensor: Implementing Cellular Automata on Graphics Adapters

N.A Shalyapina,M.L Gromov

doi:10.15514/ispras-2019-31(3)-17

Abstract

This paper presents an approach to the description of cellular automata using tensors. This approach allows to attract various frameworks for organizing scientific calculations on high-performance graphics adapter processors, that is, to automatically build parallel software implementations of cellular automata. In our work, we use the TensorFlow framework to organize computations on NVIDIA graphics adapters. As an example cellular automaton we used Conway's Game of Life The effect of the described approach to the cellular automata implementation is estimated experimentally.

Highlights

The use of automata in description of a dynamic systems’ behavior has been known for a long time.The key point of this approach to the description of systems is a representation of the object under study in the form of a discrete automatic device – automaton (State Machine or Transition System).Under the influence of input sequences an automaton changes its state and produces reactions
This paper presents an approach to the description of cellular automata using tensors
We offer an alternative approach for software implementation of cellular automata, which is based on the use of modern graphics adapters

Summary

Introduction

The use of automata in description of a dynamic systems’ behavior has been known for a long time. The number of required calculations and the structure of these calculations suggest the use of modern supercomputers with a large number of cores and supporting large-block parallelism In this case, the cell field of the automaton is divided into separate blocks. After waiting a certain time in the queue, access to the cluster is possible Another difficulty arises: a special skill is needed to write parallel programs in order to organize parallel sections of the program correctly. We offer an alternative approach for software implementation of cellular automata, which is based on the use of modern graphics adapters. Cell states will be represented by the color of the corresponding square (fig. 1c)

Preliminaries

Experimental results

Conclusions