How to Correctly Deal With Pseudorandom Numbers in Manycore Environments - Application to GPU programming with Shoverand

摘要

Stochastic simulations are often sensitive to the source of randomness thatcharacter-izes the statistical quality of their results. Consequently, we needhighly reliable Random Number Generators (RNGs) to feed such applications.Recent developments try to shrink the computa-tion time by relying more andmore General Purpose Graphics Processing Units (GP-GPUs) to speed-up stochasticsimulations. Such devices bring new parallelization possibilities, but theyalso introduce new programming difficulties. Since RNGs are at the base of anystochastic simulation, they also need to be ported to GP-GPU. There is still alack of well-designed implementations of quality-proven RNGs on GP-GPUplatforms. In this paper, we introduce ShoveRand, a frame-work defining commonrules to generate random numbers uniformly on GP-GPU. Our framework is designedto cope with any GPU-enabled development platform and to expose astraightfor-ward interface to users. We also provide an existing RNGimplementation with this framework to demonstrate its efficiency in bothdevelopment and ease of use.