High performance genetic programming on GPU

Created by W.Langdon from gp-bibliography.bib Revision:1.4221

  author =       "Denis Robilliard and Virginie Marion and 
                 Cyril Fonlupt",
  title =        "High performance genetic programming on GPU",
  year =         "2009",
  booktitle =    "Proceedings of the 2009 workshop on Bio-inspired
                 algorithms for distributed systems",
  editor =       "Gianluigi Folino and Natalio Krasnogor and 
                 Carlo Mastroianni and Franco Zambonelli",
  pages =        "85--94",
  month =        jun # " 15-19",
  note =         "paper invited for the FGCS special issue",
  publisher =    "ACM",
  address =      "Barcelona, Spain",
  isbn13 =       "978-1-60558-584-0",
  URL =          "http://portal.acm.org/citation.cfm?id=1555284.1555299",
  DOI =          "doi:10.1145/1555284.1555299",
  keywords =     "genetic algorithms, genetic programming, GPU, graphics
                 processing units, parallel processing",
  abstract =     "The availability of low cost powerful parallel
                 graphics cards has stimulated the port of Genetic
                 Programming (GP) on Graphics Processing Units (GPUs).
                 Our work focuses on the possibilities offered by Nvidia
                 G80 GPUs when programmed in the CUDA language. We
                 compare two parallelisation schemes that evaluate
                 several GP programs in parallel. We show that the fine
                 grain distribution of computations over the elementary
                 processors greatly impacts performances. We also
                 present memory and representation optimisations that
                 further enhance computation speed, up to 2.8 billion GP
                 operations per second. The code has been developed with
                 the well known ECJ library.",
  notes =        "BADS 2009 http://bads.icar.cnr.it/

                 ECJ trees replaced by byte per opcode linearised RPN
                 \cite{langdon:2008:eurogp}. nVidia GeForce 8800 GTX
                 reserved for computations (separate card to drive X-11
                 windows monitor display). BlockGP always faster than
                 ThreadGP. 32 threads per CUDA block. Thread divergence
                 due to if: 6-multiplexer, 11-mux, intertwined spirals.
                 Sextic regression (2797 million GPop/s, 144 speed up
                 over 2.6Ghz PC).

                 +-*/ sin cos exp log and or not if iflte. ERCs.
                 Interpreted programs in shared memory (referred to as

Genetic Programming entries for Denis Robilliard Virginie Marion Cyril Fonlupt