[FFmpeg-cvslog] r10453 - trunk/libavcodec/snow.c
Mon Sep 10 05:07:00 CEST 2007
On Sun, Sep 09, 2007 at 06:11:18PM -0400, Rich Felker wrote:
> On Sun, Sep 09, 2007 at 07:12:35PM +0200, Michael Niedermayer wrote:
> > Hi
> > On Sun, Sep 09, 2007 at 05:29:08PM +0200, Guillaume POIRIER wrote:
> > > HI,
> > >
> > > On 9/9/07, michael <subversion at mplayerhq.hu> wrote:
> > >
> > > > argh i really wish id never accepted these slice patches ...
> > >
> > > But aren't slices a good idea in general in non block-based codecs
> > > such as the ones like Snow or Dirac which use wavelets, since they
> > > allow the codec to be more CPU cache-friendly?
> > >
> > > Or am I missing something?
> > yes you optimize code AFTER having a final spec not before
> While I agree in principle, it's also important to know (1) whether
> the optimization will be possible (IIRC old snow bitstream precluded
hmm, i dont remember exactly but i dont think there was a change to the
bitstream due to slices
though there where changes for other things like the 16bit based wavelet
and reordering of the mc code to make it possible to calculate the
halfpel points per image, but here its just possible we dont have that
optimization yet ...
also the slice stuff is quite different compared to other optimizations
as its deeply intermangled into the whole code and that way it often
gets into the way of some experiments ...
> and (2) what benefit the optimization might have, so you
> know if it's worthwhile to make design tradeoffs needed to keep the
> possibility of that optimization open. I think those were the
Michael GnuPG fingerprint: 9FF2128B147EF6730BADF133611EC787040B0FAB
Opposition brings concord. Out of discord comes the fairest harmony.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Size: 189 bytes
Desc: Digital signature
More information about the ffmpeg-cvslog