Random number in PASM

Ahle2

lonesock said...

BTW, I found that adding "ror random,random" at the end made for a better distribution, specially if you accidentally start out with a seed of 0. (note: ROR will internally AND the rotation value with 31.) There are still certain bands that seem more likely to come up (with 25,000 values plotted), but to me this version looks a bit better than the original 2-instruction version.
Jonathan

Yeah I realized that too... but it wasn't good for my emulator though because the random generation becomes "too good" compared to a real SID.

BTW, have you any statistics on how good the random generation is?

Post Edited (Ahle2) : 12/17/2009 1:35:00 PM GMT

lonesock

Well, I ran Ent on some resulting data sets, and your original function has better distributions than with the added 3rd instruction.
(www.fourmilab.ch/random/, though it's not overly state of the art...the NIST tests would be better

add/ror:

Entropy = 7.999992 bits per byte.

Optimum compression would reduce the size
of this 40000000 byte file by 0 percent.

Chi square distribution for 40000000 samples is 456.91, and randomly
would exceed this value less than 0.01 percent of the times.

Arithmetic mean value of data bytes is 127.5014 (127.5 = random).
Monte Carlo value for Pi is 3.141500714 (error 0.00 percent).
Serial correlation coefficient is -0.000192 (totally uncorrelated = 0.0).

add/ror/ror:

Entropy = 7.950003 bits per byte.

Optimum compression would reduce the size
of this 40000000 byte file by 0 percent.

Chi square distribution for 40000000 samples is 2905248.42, and randomly
would exceed this value less than 0.01 percent of the times.

Arithmetic mean value of data bytes is 129.4133 (127.5 = random).
Monte Carlo value for Pi is 3.130604713 (error 0.35 percent).
Serial correlation coefficient is -0.005053 (totally uncorrelated = 0.0).

Jonathan

▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
lonesock
Piranha are people too.