Amended constant value used by the pseudo-random number generator

[mariosangiorgio]

As I mentioned in my comment in issue #635, I think the constant we're using needs to be amended. The Haskell implementation uses that value:

mkStdGen32 sMaybeNegative = StdGen (s1+1) (s2+1)
    [...]
    s2      = q `mod` 2147483398

stdNext (StdGen s1 s2) = (fromIntegral z', StdGen s1'' s2'')
        [...]
        s2'' = if s2' < 0 then s2' + 2147483399 else s2'

stdSplit std@(StdGen s1 s2)
            [...]
                        new_s2 | s2 == 1          = 2147483398
            [...]

The same value is mentioned in the original paper.

For 32-bit computers, we suggest l= 2, m1 = 2147483563, a1 = 40014, m2 = 2147483399 and a2 =
40692.

I am by no means expert on pseudo-random number generators, so I have no idea why we might have been using a different values. I believe that's a typo but please double check it before accepting this pull request.

[evancz]

Thanks! There is an alternate implementation of PRNG in the pipeline, so I think it makes sense to prefer that route.

[evancz]

Thanks! There is an alternate implementation of PRNG in the pipeline, so I think it makes sense to prefer that route.

[evancz]

Actually, it does make sense to merge this in. Great find, thank you for fixing it in a PR!

[evancz]

Actually, it does make sense to merge this in. Great find, thank you for fixing it in a PR!