Pseudo-Random Number Generators for Vector Processors and Multicore Processors

Abstract

Large scale Monte Carlo applications need a good pseudo-random number generator capable of utilizing both the vector processing capabilities and multiprocessing capabilities of modern computers in order to get the maximum performance. The requirements for such a generator are discussed. New ways of avoiding overlapping subsequences by combining two generators are proposed. Some fundamental philosophical problems in proving independence of random streams are discussed. Remedies for hitherto ignored quantization errors are offered. An open source C++ implementation is provided for a generator that meets these needs.

A Deterministic Portable Parallel Pseudo-Random Number Generator for Pattern-Based Programming of Heterogeneous Parallel Systems

“…He also explains counter-based PRNGs and their suitability for parallelization because they allow skipping any given number of states with constant effort. Fog [12] discusses requirements on PRNGs in parallel computations, but focuses on avoiding overlapping sequences in different threads by combining generators, while L'Ecuyer et al [19] focus on providing independent streams and substreams. Salmon et al [27] focus on output functions for counterbased PRNGs to provide fast skipping of states but still provide good statistical quality.…”