ASM Community

Performing some experiments with Perlin Noise, I've been looking for a suitable (and reasonably fast) repetitive noise function. As almost all my work was based on the article written by Hugo Elias I've decided not to use the function presented there and to go on my own. RNG algorithm (discussed here) was only one of my achievements during this research. But (for all those, who did not have anything in common with Perlin Noise) what I needed was a "repetitive" pseudo random number generator ("repetitive" - an algorithm which returns always the same "pseudo random" value for a given input value in 1D or for a given pair of input values in 2D). Here's the final function I wrote:

Noise2D		PROC	intX:DWORD, intY:DWORD

	push		EBX
	push		EDX
	mov		EAX,00BB38435h
	mul		intX
	mov		EBX,EAX
	mov		EAX,03619636Bh
	mul		intY
	add		EAX,EBX
	pop		EDX
	pop		EBX

	push		ECX
	mov		CL,AL
	rol		EAX,CL
	mov		CL,AL
	rol		EAX,CL
	mov		CL,AL
	rol		EAX,CL
	pop		ECX
	ret

Noise2D		ENDP

Results are shown in Perlin0.jpg. I've tried with only one ROL, but it seems to be not enough (results, shown in Perlin1.jpg, build a noticeable pattern - what's not the point of the noise function).

Presented function may be easily converted to return floating point values (range 0.0 - 1.0 or any other) - just as in floating point version of RNG algorithm. Hopefully - it will be usefull for one of you ;)

Cheers, Mikael

Nice to hear that it was a bit usefull. But what about my code style? I've always been using tabs instead of spaces (AFAIR) ;) Well - honestly - I'm a little bit pedantic about the code layout (lenght of names, indents, aligments, a.s.o.) - and this could be a real pain in ... sometimes ;)

Cheers.

Seems, that I've lost the point... Did I write "pushEBX" or did I eat a space anywhere?  :sad:

hehe a display error???? (like FF with aarabic and related accents ;)).

Hi, a little new arround here, but a few years ago I wrote a random number generator which uses only modulo and addition (If the limit you choose for your values is 8, 16, 32 bits then the modulo goes away of course).

Problem is I've no long got the code - It was lost in a disk crash a few years ago, but the algorythm goes like this:

1) Seed a 17 value array with pseudo random numbers.
2) Set two pointers pointing to slots 0 and 10
3) When a number is read, read the values at the two pointers, place it at the first pointer, inc pointers (Wrapping round at the end of the array), and return the calculated value.

This produced some very impressive results for me (I was writing a game which required a large number of random values) and it appears to be very, very fast.

YMMV.

Almost (please bear in mind that I am still learning X86 ASM, when I wrote this, it was in 6502 ASM), but the line

   mov  [00],EAX

should be

  mov , EAX

I think.

(I'm at work at the moment, and I don't have my compiler stuff with me, but it looks right)

Oh, and there is no point hardcoding the initial p'rands unless you want predictable p'rands out of this function. You could use a slower function to seed this, read from the time stream, grab a chunk of stack ... something that is not predictable. if you then run the function a random number of times (Say, wait for the user to hit a key), then you get to a pretty non-predictable p'rand set, which stands up to Chi^2 testing reasonably well (At least, it beat the rnd functions back when I first wrote it).

Haqa...

Yes, it should (I've just noticed that you've written "place it at the first pointer" not "place it at the first slot"). And this means, that:


The inc'es shall be placed after the store ;)

Cheers.