On Sat, Sep 01, 2001 at 10:50:49AM +0200, Andreas Persson wrote:
> That code was for intel. I'm including some code which I've been told should
> be fairly close to the fastest possible on Athlon. Then again, I don't have
> an Athlon, so I can't really try it. In fact, its movntq loop is
> essentially what AMD recommends in their optimization guide.

Two serious problems have been found:
 - it hangs upon copying less than 64 bytes. I guess you already know
   that.
 - it gives no performance boost with copying 4096 bytes which is the
   size of a page. So your code and mine are not so useful when they
   come to the real world.

Jun-Young

-- 
Bang Jun-Young <bjy@mogua.org>