> From: =?iso-8859-1?Q?P=E5l_Halvorsen?= <paalh@unik.no>
>
> Hi!
>
> I have performed an experiment looking at memory copy performance using
> copyin, copyout and memcpy on an i386 machine.
>
> My tests indicate that it is faster to copy data if the size is between
> 1kb and 8 kb. The performance drops dramatically at sizes >=16kb and =<0.5kb.
>
> Why is this so? Is there some optimalization for these sizes?

You now have experimental evidence that...

	1. Damn, Intel really DID put a cache level or two in that chip

	2. There is startup overhead associated with protection checks
	   and setting up the transfer instructions.
:-)

--Ross