Re: Anyone need a random number generator?

-----------------------------------<Another 
snip>---------------------------------------

>  
>
> On Wed, Apr 28, 2010 at 4:12 PM, Rick Fochtman <[email protected] 
> <mailto:rfochtman%40ync.net>> wrote:
> >
> > On Wed, Apr 28, 2010 at 3:03 PM, Ron Hudson <[email protected] 
> <mailto:hudson.ra%40gmail.com>> wrote:
> > >
> > > On Wed, Apr 28, 2010 at 12:39 PM, Rick Fochtman <[email protected] 
> <mailto:rfochtman%40ync.net>> wrote:
> > <<deleted>>
> > >> //FORT.SYSIN    DD *
> > >> C RANDOM NUMBER GENERATOR
> > >>       FUNCTION IRND(ISEED)
> > >>       IY = IABS(MOD(1103515245 * ISEED + 12345,2**15))
> > >>       IRND = IY
> > >>       RETURN
> > >>       END
> > <<deleted>>
> > >>
> > >> (Given the math I see there, you might experience integer 
> overflow conditions in the first argument to the MOD function.)
> > >
> > > What can I do? I guess I need to user a 'larger' integer  - What 
> is a good reference (there is a short overview
> > > on Jay Moseley's site but no detail like how to specify more bits...
> > >
> > The assembler equivalent should run (i.e. not overflow the registers),
> > not sure about how this particular compiler creates the machine
> > instructions. You will need to test by calling the routine and
> > logging the results.
> >
> > 
> ------------------------------------<unsnip>----------------------------------
> > Doing it in Assembler will not "use more bits". FORTH uses the full 
> 32 bits of the 370 architecture. Since the value 1,103,515,245 is 
> being multiplied by the ISEED parameter, the resultant value may 
> easily exceed 32 bits. Perhaps even exceeding the 64 bits result of 
> the multiply instruction. Then what?
> >
> > Rick
> Correct. S/370 M or MR multiplies two signed 32 bit operands and
> places the result in a odd-even pair of 32 bit registers. The result
> cannot exceed 64 bits. 7FFFFFFF * 7FFFFFFF will fit.
> Then the S/370 D or DR diviides an odd-even pair of 32 bit registers
> by the 2**15 result and could result in an overflow condition. Since
> he is desiring the MOD result, losing the high order bits of the
> result is of no consequence.
>
--------------------------------<end of "Another 
snip">--------------------------
That's true, with respect to the basic arithmetic. But the FORTRAN error 
handler might not see it that way. :-)

Rick