Re: [AD] Allegro 4.2 todos

[Sven Sandberg <svens@xxxxxxxxxx>]

Angelo Mottola wrote:
> I'd have posted the patch here anyway before commiting it this time :P
>
> The current formula is:
>
> ((rand() >> 8) + (rand() & 0xff00))
>
> As previously suggested on this list. I've found the following formula works
> better under OSX though:
>
> ((rand() >> 16) + rand())

It seems there are essentially four types of implementations:

 1. 16-bit numbers, 8 lowest bits not very random.
 2. 16-bit numbers, all bits random enough.
 3. 32-bit numbers, 16 lowest bits not very random.
 4. 32-bit numbers, all bits random enough.

Either bits 8-15 or bits 16-31 are random enough with all these types of 
implementations. Thus, if we need 15 bit random numbers, the following 
might be worth trying:

  ((rand()>>16) ^ (rand()>>8) ^ rand()) & 0x7fff

For type 1 and 2 implementations, the two last terms will give 
randomness, and for type 3 and 4 the first term will give randomness.

> I've tested other shift values between 8 and 16, but the best remains 16. So
> I wonder: is this due to the machine endianess or just OSX rand()
> implementation?

Since the meaning of arithmetic operations does not depend on 
endianness, I think this has to be due to the implementation.

Sven