CUDA 1.0 Mersenne Twister implementation

Accelerated Computing CUDA CUDA Programming and Performance
1

Hi!
I’ve a question about the Mersenne Twister implementation in CUDA 1.0.
Is this an implementation of the original method from 1997:
[url=“Mersenne Twister: A random number generator (since 1997/10)”]http://www.math.sci.hiroshima-u.ac.jp/~m-mat/MT/emt.html[/url]

…or is it based on the newer “SIMD-oriented Fast Mersenne Twister (SFMT)” from 2006?
[url=“SIMD-oriented Fast Mersenne Twister (SFMT)”]http://www.math.sci.hiroshima-u.ac.jp/~m-m...SFMT/index.html[/url]

Thanks,
/Thord

2

It is based on the Dynamic Creator Mersenne Twister.

3

What is the periodicity of the implementation?

4

It’s based on an original implementation (which is included into the package in slightly modified form), running single twister per each CUDA thread. dcmt0.3 library was used to spawn 4096 sets of twister parameters with the period of 2 ** 607 - 1