Random Number Generators: Performance

[Scott Robert Ladd]

Lars Segerlund wrote:
> Would you you mind having a look at the Mersenne twister in
> libgfortran ? Do you think it's up to the job ?

It looks fine to me. Some people here expressed concern about the large 
seed register -- however, *any* long-period generator is going to 
maintain an array of some size.

> I choose the MT since it has a VERY big period and a not too big
> seed, coupled with a performance which is right 'up there'.

The MT's performance is very good, though it can be consistently beaten 
(in terms of random numbers per second) by some of Marsaglia's 
generators. I have tended to favor the MT, but use a couple of different 
generators to ensure that simulations have not been caught in any 
unforeseen traps associate with a given algorithm.

My experience is that people "roll their own" generators, or use 
something like SPRNG, when they are serious about stochastic algorithms. 
We need to provide a solid, journeyman's implementation that can be 
trusted for most applications. The MT is such an implementation, and see 
no reason to replace your code with Andy's weaker algorithm

..Scott

-- 
Scott Robert Ladd
Coyote Gulch Productions (http://www.coyotegulch.com)
Software Invention for High-Performance Computing