Fun with CBM arithmetics, Part III

[Mike]

Hi!

Just type in this small program and look how it wrecks havoc with the built-in random number generator of CBM BASIC:

Code: Select all

1 A=2087*794531
2 B=32768*65536
3 X=A/B:Y=RND(-X)
4 PRINTRND(1):GOTO4

P.S. https://xkcd.com/221/

[chysn]

Woah!

+1 for XKCD

The thing I envy most about the C64 isn't the sound or the sprites or the extra colors. It's the analog noise source in the SID that can act as a reliably-random random number generator. VIC-20's RND is so slow when you need to get a lot of random numbers, like to generate a maze. I've tried linear shift register techniques with limited success.

This only serves to deepen my despair.

[Kweepa]

But LFSR/LCGs are so easy!
There are a number of threads on denial about cheap, well behaved random numbers.

http://sleepingelephant.com/ipw-web/bul ... f=2&t=2304
http://sleepingelephant.com/ipw-web/bul ... f=2&t=2348

Sprites are the one thing I would take out of your list.

[chysn]

Ah! You know where I've been going wrong with shift registers? I've been manipulating, then shifting, and then using the resulting bytes instead of a single bit at a time. Now that I try it as a source of bits, it makes all the difference in the world. Thanks! And that was your code, Mike, so thanks to you, too!

Okay, I'm changing my answer. Sprites are awesome, but I'll take the SID. Which really may not be a change after all.

[Mike]

Some time ago (also @Kweepa) I posted a viable alternative to the built-in RND() function which is based on the QDRand generator published in Numerical Recipes:

Making a random number in Assembly

A bit further down the page, there's a demo program which puts the routine into the USR() function.

...

I found the behaviour of the original RND() function to be suspect already since 2006 (link), where I demonstrated that most of the seeds result in periods of just ~58000 different numbers, even though the seed size would suggest a period of at least 2^32 numbers. Later, I found a seed number that resulted in an even shorter period, just ~700 different numbers.

I then set out to find a possible fixpoint of the generator, and the OP shows the result.

[groepaz]

The thing I envy most about the C64 isn't the sound or the sprites or the extra colors. It's the analog noise source in the SID that can act as a reliably-random random number generator.

there is no such thing in the SID - it's a LFSR

[chysn]

The thing I envy most about the C64 isn't the sound or the sprites or the extra colors. It's the analog noise source in the SID that can act as a reliably-random random number generator.

there is no such thing in the SID - it's a LFSR

Learn something new every day!