Re: [Patch libfortran/28452] random intrinsics for large reals

[Jerry DeLisle <jvdelisle at verizon dot net>]

Steve Kargl wrote:
> On Sun, Jul 30, 2006 at 01:24:55PM -0700, Jerry DeLisle wrote:
> > Steve Kargl wrote:
> > > Jack, see the attached program.  Change the definition of
> > > ep to ep=kind(1.e0_16).  You may want to change n.
>
> BTW, there's an astounding lack of comments.  (It Fortran so
> it's seld documenting ;-).  But, seriously, this is a simple
> equidistant that sorts the numbers into 51 bins, and then
> computes the chi-squared of the sequence.  Refer to Knuth
> to convert chi-squared to a measure of goodness.
>
> > I am curious if anyone has run a spectral test on these or is that 
> > necessary?
>
> What do you have in mind?  Most of the spectral tests I've found
> deal with sequences of random bits (ie., unsigned 32-bit integers).

I was reading Knuth too, and the spectral tests appeared to be a better 
indicator.  We might have to extract bits to do it.  I am more curious then 
anything, especially after reading Knuths comments about the one commonly used 
generator that truly sucked and no one knew it for 20 years or so.

> I thought about writing run-up and run-down tests, but haven't
> gotten very far.