Re: New IEEE Language Popularity Ratings

[brbarkstrom@gmail.com]

> > Another example is random number generators.  Knuth's Art of Computer 
> > Programming, Vol. 2, is the basic definitive reference.  It includes ten
> > tests of a random number generator that anybody needing one should be sure
> > have been used.  Park and Miller's article many years ago in CACM provides
> > code for a portable version of an algorithm that they put through the Knuth
> > wringer.  In my own work, I've got an Ada library for generating probability
> > distributions that includes Park and Miller's algorithm, since it includes
> > a test as to whether it's been implemented correctly.  I might trust the 
> > algorithms in MatLab or Mathematica, but this kind of work is not for
> > amateur mathematicians.
> 
> These tests, while useful, are probably badly outdated for modern algorithms on
> 64-bit processors. More modern tests include TestU01's Crush and Big-Crush test
> suites. The Threefry generator is said to be the fastest to pass all the
> Big-Crush tests.
> 
> -- 
> Jeff Carter
> "I've seen projects fail miserably for blindly
> applying the Agile catechism: we're Agile, we
> don't need to stop and think, we just go ahead
> and code!"
> Bertrand Meyer
> 150

The algorithms are mathematical artifacts and they have proofs related
to loss of digital information.  Whether or not the bits in the hardware
are 32 or 64 or 128 bits doesn't matter.  Neither does the age.  Most 
mathematical algorithms are essentially ageless.  Can we get the Big Crush
test suite to identify the specific relations between Knuth's tests and
their tests?  I'll try to find the time to see if I can use the reference
to the RNG in AdaCore.  In the mean time, some examination of the floating
point standard and rounding and truncation error may be appropriate.

Unfortunately, Lawson and Hanson's algorithm doesn't easily translate to
Ada because they rely on FORTRAN's (rather opaque) approach to memory
management within COMMON blocks for dealing with Pivoting.

Bruce B.

Bruce B.