Closed
Description
It has been discussed if we should consider this RNG alongside or instead of dSFMT.
The RNG we are using now is a Mersenne Twister implementation adapted to use SIMD, but we don't really get anything from SIMD right now because of #6573.
In addition, dSFMT actually consistently fails one of the tests in BigCrush in contrast to the proposed XORShift* algorithm.
Finally, the XORShift* is very simple and can be written in Julia. See some implementations with various perfomance tweaks here
Activity