[XviD-devel] Inlined ASM code again
skal
skal at planet-d.net
Fri Aug 22 10:38:18 CEST 2003
Hi Marco and all,
On Fri, 2003-08-22 at 04:30, Marco Al wrote:
> From: "skal" <skal at planet-d.net>
>
> > But eventually, the only conclusion you may end up
> > with, when inlined gcc gets better than NASM, is...
> > that this latter could have been better written.
>
> You cannot get rid of function call overhead with
> NASM alone, your only option is to move the outer
> loops to assembly too (great option that is :).
>
note that speed up gain from inlined (long) code often
comes from better pairing / optimizing with
the code just above and below the inlined
func, rather than func call overhead removal
(especially on modern CPU, with all those
code caches)...
Also:
IMHO (expressed here: skal.planet-d.net/coding/vademecum.html
as rule #1 :), one should first focus on *not* calling
a func before optimizing it, especially when it come to
weird compile flag fiddling :)
(in a present case, sad16 is a ~100 tick, func.Hence: if you
manage to not call it inappropriately say 10 times in a qcif
pic, you'll have gained your day).
ok, off with this topic. Later!
Skal
More information about the XviD-devel
mailing list