Full Paper
Cellular Automata based Multi-Stream Parallel Combined Pseudo-Random Number Generators
- 1 National Institute of Technology (India)
- 2 Indian Institute of Engineering Science and Technology (India)
Keywords:
Combined CA-based PRNGMulti-stream PRNGPeriodMaximal EquidistributionEfficiencyjumping-ahead
Abstract
This work introduces multi-stream combined CAs-based PRNGs for the current $32$-bit and $64$-bit GPU architectures that generate pseudo-random numbers in a parallel manner using a multi-threading approach. The period of the PRNGs is close to maximal and is divided into $s$ equal-length non-overlapping streams where all streams generate random numbers in parallel using jumping-ahead technique. We show that these PRNGs are extremely fast, have a small internal state to fit inside the GPUs and achieve uniformity in almost all possible dimensions which is not available in existing multi-stream PRNGs. These features make the proposed PRNGs well suited for high-performance computing.