Re: lkwt in DragonFly

[David Rhodus <drhodus@xxxxxxxxx>]

I think on the P4 chips you'll find ~150 cycles though as on older 
intel chips
and the AMD4 we are looking at around 12 cycles. Here are the output 
from
some initial tests I ran. You'll notice I've include the memory segment 
type
for the P4 as there was actually a difference between shared and private
mappings, but no change on the AMD64. I've also included the time output
from cpuid because it is another serializing instruction.

-DR

AMD64 - 3200

TIME	ASM
10nS	cmpxchgl
12nS	xchg
38nS	cpuid

P4 - 3.06Ghz

TIME	ASM		MEMORY
34nS	cmpxchgl	shared
38nS	cmpxchgl	private
34nS	xchg		shared
90nS	xchg		private
83nS	cpuid