Re: Interesting pentium-memcpy results

Ingo Molnar (
Wed, 30 Jul 1997 12:51:22 +0200 (MET DST)

On Wed, 30 Jul 1997, Alan Cox wrote:

> > maybe we could just pick out the very case, when the FPU stack is
> > completely empty. This would require a 'fnstcw', here the control word
> > should be all 00000000 for the stack to be empty?
> Sorry to ruin a deeply fascinating scheme but if the user program is doing
> MMX stuff then while FNSAVE does the right thing the other magic may not
> unless there's some extra cleverness going on

hm, ok, in the MMX case we dont want to use an FPU based memcpy(). So the
extra cleverness has to be some checking at bootup?

-- mingo