A Variant of Mersenne Twister Suitable for Graphic Processors

Abstract

The author proposes pseudorandom number generators suitable to execute on a graphic processor. They generate pseudorandom numbers in device memory on graphic processors. Each generator uses shared memory on graphic processors as its internal state space, and uses constant memory as a look-up table for a linear transformation. Output formats of the generator are 32-bit integers and single precision floating point numbers obeying the IEEE 754 format. A 64-bit integer version and double precision floating point version are also available. The author also proposes a parameter generator for these generators. The parameter generator provides us with independent streams of pseudorandom numbers, which can be generated in parallel.