Genetic improvement of GPU software

William B. Langdon; Brian Yee Hong Lam; Marc Modat; Justyna Petke; Mark Harman

doi:10.1007/s10710-016-9273-9

Genetic improvement of GPU software

William B. Langdon^*
, Brian Yee Hong Lam
, Marc Modat
, Justyna Petke
, Mark Harman

^*Corresponding author for this work

Biomedical Engineering Department

UCL University College London
Cambridge University Hospitals NHS Foundation Trust

Research output: Contribution to journal › Article › peer-review

33 Citations (Scopus)

Abstract

We survey genetic improvement (GI) of general purpose computing on graphics cards. We summarise several experiments which demonstrate four themes. Experiments with the gzip program show that genetic programming can automatically port sequential C code to parallel code. Experiments with the StereoCamera program show that GI can upgrade legacy parallel code for new hardware and software. Experiments with NiftyReg and BarraCUDA show that GI can make substantial improvements to current parallel CUDA applications. Finally, experiments with the pknotsRG program show that with semi-automated approaches, enormous speed ups can sometimes be had by growing and grafting new code with genetic programming in combination with human input.

Original language	English
Pages (from-to)	5-44
Number of pages	40
Journal	Genetic Programming and Evolvable Machines
Volume	18
Issue number	1
DOIs	https://doi.org/10.1007/s10710-016-9273-9
Publication status	Published - 1 Mar 2017

Keywords

Dynamic programming
Genetic programming
GGGP
GI-GPGPU
GPGPU
Grammar based genetic programming
Metaprogramming
NVidia CUDA
Parallel computing
SBSE

Access to Document

10.1007/s10710-016-9273-9

Cite this

@article{3d95ffd22b49480e835b00038972c2ab,

title = "Genetic improvement of GPU software",

abstract = "We survey genetic improvement (GI) of general purpose computing on graphics cards. We summarise several experiments which demonstrate four themes. Experiments with the gzip program show that genetic programming can automatically port sequential C code to parallel code. Experiments with the StereoCamera program show that GI can upgrade legacy parallel code for new hardware and software. Experiments with NiftyReg and BarraCUDA show that GI can make substantial improvements to current parallel CUDA applications. Finally, experiments with the pknotsRG program show that with semi-automated approaches, enormous speed ups can sometimes be had by growing and grafting new code with genetic programming in combination with human input.",

keywords = "Dynamic programming, Genetic programming, GGGP, GI-GPGPU, GPGPU, Grammar based genetic programming, Metaprogramming, NVidia CUDA, Parallel computing, SBSE",

author = "Langdon, \{William B.\} and Lam, \{Brian Yee Hong\} and Marc Modat and Justyna Petke and Mark Harman",

year = "2017",

month = mar,

day = "1",

doi = "10.1007/s10710-016-9273-9",

language = "English",

volume = "18",

pages = "5--44",

journal = "Genetic Programming and Evolvable Machines",

issn = "1389-2576",

publisher = "Springer",

number = "1",

}

TY - JOUR

T1 - Genetic improvement of GPU software

AU - Langdon, William B.

AU - Lam, Brian Yee Hong

AU - Modat, Marc

AU - Petke, Justyna

AU - Harman, Mark

PY - 2017/3/1

Y1 - 2017/3/1

N2 - We survey genetic improvement (GI) of general purpose computing on graphics cards. We summarise several experiments which demonstrate four themes. Experiments with the gzip program show that genetic programming can automatically port sequential C code to parallel code. Experiments with the StereoCamera program show that GI can upgrade legacy parallel code for new hardware and software. Experiments with NiftyReg and BarraCUDA show that GI can make substantial improvements to current parallel CUDA applications. Finally, experiments with the pknotsRG program show that with semi-automated approaches, enormous speed ups can sometimes be had by growing and grafting new code with genetic programming in combination with human input.

AB - We survey genetic improvement (GI) of general purpose computing on graphics cards. We summarise several experiments which demonstrate four themes. Experiments with the gzip program show that genetic programming can automatically port sequential C code to parallel code. Experiments with the StereoCamera program show that GI can upgrade legacy parallel code for new hardware and software. Experiments with NiftyReg and BarraCUDA show that GI can make substantial improvements to current parallel CUDA applications. Finally, experiments with the pknotsRG program show that with semi-automated approaches, enormous speed ups can sometimes be had by growing and grafting new code with genetic programming in combination with human input.

KW - Dynamic programming

KW - Genetic programming

KW - GGGP

KW - GI-GPGPU

KW - GPGPU

KW - Grammar based genetic programming

KW - Metaprogramming

KW - NVidia CUDA

KW - Parallel computing

KW - SBSE

UR - https://www.scopus.com/pages/publications/84979619472

U2 - 10.1007/s10710-016-9273-9

DO - 10.1007/s10710-016-9273-9

M3 - Article

AN - SCOPUS:84979619472

SN - 1389-2576

VL - 18

SP - 5

EP - 44

JO - Genetic Programming and Evolvable Machines

JF - Genetic Programming and Evolvable Machines

IS - 1

ER -

Genetic improvement of GPU software

Abstract

Keywords

Access to Document

Other files and links

Fingerprint

Cite this