Using Reconfigurable Computing to Accelerate Simulation Applications

摘要

A general approach for efficiently generating pseudorandom numbers from continuous uniform and non-uniform probability distribution functions in a reconfigurable computing system is developed. To test the performance of the approach, pseudo-random number generators for a variety of commonly used probability distributions are built in hardware and software to compare performance. The approach is then specifically optimized and implemented to accelerate a pi estimator, a Monte Carlo integrator, and an exact stochastic simulator. It is shown that implementing a pseudo-random number generator in hardware can be an effective starting point for accelerating simulation applications using reconfigurable computing.