gsl_ran_gamma

[Brian Gough]

John Lamb writes:
 > I.e. Marsaglia and Tsang is obviously better. There's nothing in either 
 > code that would make it substantially more or less efficient on 
 > different architectures

I'd be interested to see it, and a reference to the original paper (or
link, if it is available online).

 > The obvious downside of doing this is that gsl would generate a 
 > different sequence of random variates

This is not a major problem --- the new generator could have a
different name, e.g. gsl_ran_gamma_mt.

-- 
Brian Gough

Network Theory Ltd,
Commercial support for GSL --- http://www.network-theory.co.uk/gsl/