Re: [RFC][PATCH] Faster generic_fls

[Alan Cox]

On Mer, 2003-04-30 at 22:59, Falk Hueffner wrote:
> > ffs(x^(x-1)) == fls(x)
> 
> I don't think so. Maybe you are thinking of ffs(x) == fls(x & -x). So
> you can calculate ffs with fls, but not easily the other way round.

Well I asked my CPU to double check. If I got the code right it thinks that
ffs(i^(i-1))==fls(i) is true for 1->2^32-1

If you think about it it seems to make sense

	x-1 turns the lowest 100..0 sequence into 01......1

	The xor removes unchanged higher bits

	so 
		10100
         -1     10011
	XOR	00111