[XviD-devel] More MMX improvements

Michael Militzer michael at xvid.org
Fri Jul 29 00:18:03 CEST 2005


Hi all,

Quoting Pascal Massimino <skal at planet-d.net>:

> 
> 	Howdy everybody,
> 
> On Thu, 2005-07-28 at 22:38, Guillaume POIRIER wrote:
> > Hi,
> > 
> > On 7/17/05, carlo.bramix <carlo.bramix at libero.it> wrote:
> > > Hello,
> > > I did more improvements to:
> > > 
> > > \xvid_20050712\xvidcore\src\bitstream\x86_asm\cbp_mmx.c
> > > 
> > > This new version is 6.2% to 7.4% faster if compared to the same source
> into the CVS.
> > > I attached this new patch here.
> > 
> > Your optimization do work for me, but I can't measure any speed-up.
> > You must be optimizing a part of XviD that doesn't account for a large
> > part of the encoding process.
> > 
> > Now, I don't know what the maintainers are gonna do about it, but I'm
> > quite worried about the fact that no one has commented on that patch.
> > :-(
> 
> 	It's on my to-commit list ... when holidays are over :)
> 	Anyway, it's normal and expected you don't see any
> 	speed-up. Cbp calculation is called once per MB. Would
> 	it not be in MMX/SSE, but simply plain C, you wouldn't
> 	notice the difference either. 
> 	*But*, i personally find it very pleasant lot of people
> 	try to optimize this particular function. 

Yes, the reason why not many commented on this optimization (besides the
fact that it's holiday season) is probably that no noticable overall speed-
up can be expected from this patch. As Skal pointed out, the cbp calculation 
code is actually not really speed critical, which however didn't prevent
people from writing many optimizations for it - but that's perfectly fine
and just demonstrates the huge motivation of people to participate in
development...

Michael



More information about the XviD-devel mailing list