Re: [BENCHMARK] Lmbench 2.5.54-mm2 (impressive improvements)

[Andi Kleen <ak@muc.de>]

On Mon, Jan 06, 2003 at 02:33:28AM +0100, Linus Torvalds wrote:
> > I can think of some things to speed it up more. e.g. replace all the
> > push / pop in SAVE/RESTORE_ALL with sub $frame,%esp ; movl %reg,offset(%esp) 
> > and movl offset(%esp),%reg ; addl $frame,%esp. This way the CPU has 
> > no dependencies between all the load/store options unlike push/pop.
> 
> Last I remember, that only made a difference on Athlons, and Intel CPU's 

I didn't benchmark it, but as a data point ICC 7 generates the movls instead 
of pushes now too, (even though it generates bigger code). In fact it is even more 
aggressive on that than gcc: gcc does it only for more than three or four registers, 
icc does it for two and more.  So I expect it being faster on Intel CPUs - at least on 
the P4 - too.  I doubt they tuned it for Athlons.


-Andi