Re: gsl_ran_gaussian_ziggurat doesn't work with gsl_rng_slatec and gsl_rng_uni

[Patrick Alken]

Hello,

  I would welcome your help with fixing rng issues. We don't currently
have a GSL developer with expertise in rngs (I myself have never worked
on them). The best approach would be to clone the git repository
(instructions on the GSL webpage), make your changes, and then send me a
git diff when ready. Then I can take a look at things.

Best,
Patrick

On 2/5/21 12:52 AM, camel-cdr--- via Bug reports for the GNU Scientific
Library wrote:
> I've been developing a ziggurat implementation and used 
> https://www.seehuhn.de/pages/ziggurat to benchmark mine against the one used 
> in gsl.
> I ran[timegauss.c,](https://m.seehuhn.de/data/ziggurat/timegauss.c) but it 
> fails with gsl_rng_slatec and gsl_rng_uni.
> Consequently, I checked randist/gausszig.c and gsl_ran_gaussian_ziggurat 
> doesn't support generators with smaller ranges than 16777216.
>
> On a related note: I'd also be interested in improving the performance of 
> gsl_ran_gaussian_ziggurat outside of this bug, but I'm new to this project, 
> and I'm not sure what the procedure would be. (In my current benchmarks my 
> implementation is 1.3 times faster and uses 2/3 off the memory)