Date: Wed, 26 Apr 2000 07:07:55 -0400 (EDT)
Message-Id: <200004261107.HAA23684@indy.delorie.com>
From: Eli Zaretskii <eliz AT delorie DOT com>
To: "Dieter Buerssner" <buers AT gmx DOT de>
CC: djgpp-workers AT delorie DOT com
In-reply-to: <200004251736.NAA26252@delorie.com> (buers@gmx.de)
Subject: Re: rand() in libc
References: <200004251246 DOT IAA11271 AT delorie DOT com> (buers AT gmx DOT de) <200004251736 DOT NAA26252 AT delorie DOT com>
Reply-To: djgpp-workers AT delorie DOT com
Errors-To: nobody AT delorie DOT com
X-Mailing-List: djgpp-workers AT delorie DOT com
X-Unsubscribes-To: listserv AT delorie DOT com
Precedence: bulk

> From: "Dieter Buerssner" <buers AT gmx DOT de>
> Date: Tue, 25 Apr 2000 20:41:32 +0200
>
> On 25 Apr 00, at 11:13, DJ Delorie wrote:
>  
> > Do you have URLs for the tests you reference?
>  
> I found http://stat.fsu.edu/~geo/diehard.html in some old files.
> 
> For quick testing, write 3e6 32 bit random numbers in binary to a 
> file. (The 31 bit random numbers of DJGPP must be shifted 1 to the 
> left, because diehard/diequick needs a random msb) Then run diequick. 
> Ignore results about the lsb. Any p-values close to one or close to 
> zero are suspect. P-values very close (say 1e-5) to one or zero are 
> failures.

Assuming that Marsaglia's suite at the above URL is Free Software, it
would be nice to add some or all of it to djtst distro, together with
some driver program that would interpret the results in a way that can
be understood by humans even if they don't know much about this
subject.

That would allow to quickly compare different implementations and/or
small modifications in existing code.  Right now, the RNGs available
in libc do not have *any* test suite.