A faster uniform generator for 7 bit numbers should be :

rand:
                ld   a,r
                ld   b,a
                ld   a,(old_rnd)
                xor  b
                ld   (old_rnd),a
                ret

IMHO seems even more uniform than the first one posted at the beginning of this tread...
Anyone willing confirm my impression?

Sooooooooooo good!

Anyway, ARTRAG, you should remember to init the R register with a 0 to have numbers between 0 and 127. Something like XOR A ; LD R,A

Well, it generates 7 bit random numbers even if bit 7 stays to 1.

Any idea on how good it is ?

Quote:

Well, it generates 7 bit random numbers even if bit 7 stays to 1. Any idea on how good it is ?

I just incorporated it in my z80 asm random maze generator with mixed results... the mazes look ok (although patterns are visible), but the loops that distribute treasure [making sure there's only 1 treasure per Y tile to avoid too many sprites per row] sometimes gets stuck in an endless (?) loop, waiting for a suitable random number.

Using the high byte (ld A,H) of this function gives very nice results:

SEED: equ rand16 + 1

rand16:
ld BC,0
ld HL,253
xor A
sbc HL,BC
sbc A,B
sbc HL,BC
sbc A,B
ld C,A
sbc HL,BC
jr nc,.end
inc HL
.end
ld (SEED),HL ; self modifying code (seed is hardcoded in opcode)
ret

p/s: this function was ripped of the internet and slightly optimized by me.

speaking of numbers....

http://youtube.com/watch?v=qjOZtWZ56lc

speaking of random numbers....

http://youtube.com/watch?v=qjOZtWZ56lc

I'd like a number 2, please....

Quote:

Quote: Well, it generates 7 bit random numbers even if bit 7 stays to 1. Any idea on how good it is ? I just incorporated it in my z80 asm random maze generator with mixed results... the mazes look ok (although patterns are visible), but the loops that distribute treasure [making sure there's only 1 treasure per Y tile to avoid too many sprites per row] sometimes gets stuck in an endless (?) loop, waiting for a suitable random number. Using the high byte (ld A,H) of this function gives very nice results: SEED: equ rand16 + 1 rand16: ld BC,0 ld HL,253 xor A sbc HL,BC sbc A,B sbc HL,BC sbc A,B ld C,A sbc HL,BC jr nc,.end inc HL .end ld (SEED),HL ; self modifying code (seed is hardcoded in opcode) ret p/s: this function was ripped of the internet and slightly optimized by me.

Thanks!! This is a true random generator.
Actually my code can fall in "loops" (i.e. bad random sequences) when it is called in a loop itself without branches.
This is due to the fact that R counts the number of executed instructions, so when the loop calling the rand generator
has no branches the counter is increased of a fixed quantity and bad sequences can happen.

I call my rand generator after some large switches so in my case I do not incur in big problems.
Anyway I see it is not very random.

Quote:

A faster uniform generator for 7 bit numbers should be : rand: ld a,r ... IMHO seems even more uniform than the first one posted at the beginning of this tread... Anyone willing confirm my impression?

I believe using R register is a nice thing to get a single random value, given that at least one user interaction was required since R register was reset. You can get random (pseudo, of course) sequences using it several times, but I believe they probably will not be uniform.
For uniform random sequences it is better to use a LFSR function, as the original posted one (very well explained by RicBit and the links he provided). It seems weird, but it really works.

I'm sure of that. But this is shorter and seems fine for my needs.

rand:       
                ld   a,r
old_rnd:   
                xor  0
                ld   (old_rnd+1),a
                ret

For the 2K competition size wins on efficiency