Evolving gzip matches Kernel from an nVidia CUDA Template
CREST centre, Department of Computer Science, King’s College, London, WC2R 2LS, UK
Technical report TR-10-02, 5 Feb 2010, Department of Computer Science, King’s College, London
Rather than attempting to evolve a complete program from scratch we demonstrate genetic interface programming by automatically generating a parallel CUDA kernel with identical functionality to existing highly optimised ancient sequential C code. Generic GPGPU nVidia kernel C++ code is converted into a BNF grammar. Strongly typed genetic programming uses the BNF to generate compilable and executable graphics card kernels. Their fitness is given by running the population on a GPU with randomised subsets of training data itself given by running the original code’s test suite.
February 15, 2011 by hgpu
