A SIMD Interpreter for Genetic Programming on GPU Graphics Cards

Mackey-Glass chaotic time series prediction and non-nuclear protein classification show the feasibility of evaluating genetic programming populations on SPMD parallel computing consumer gaming graphics processing units. The C++ framework with a regular disk less Linux KDE desktop equipped with a single leading nVidia GeForce 8800 GTX graphics processing unit card is demonstrated evolving programs at Giga GP operation per second (895 million GPops). The RapidMind general processing on GPU (GPGPU) framework supports evaluating an entire population of a quarter of a million individual programs on a non-trivial problem in 4 seconds. An efficient reverse polish notation (RPN) tree based GP is given. No Leaf Push onto individuals stacks Addition Pop+Pop, Push result Subtraction Pop−Pop, Push result Multiply Pop * Pop, Push result Division Pop/Pop, Push result All programs finished? Yes Result is on top of each stack Figure 1: The SIMD interpreter loops continuously through the whole genetic programming terminal and function sets for everyone in the population. GP individuals select which operations they want as they go past and apply them to their own data and their own stacks. Unwanted results are discarded. When a required instruction is executed by a program, the program moves onto waiting for its next operation. If branches, goto jumps, loops and function calls could be implemented. Boolean and very short integer operations can be implemented by lookup tables. Combining tables for different operations reduces the number of options in the loop and so could make GP faster.