[Ffmpeg-cvslog] r6147 - in trunk: ffplay.c libavcodec/avcodec.h libavcodec/dsputil.h libavutil/common.h

Michael Niedermayer michaelni
Fri Sep 1 15:50:03 CEST 2006


Hi

On Fri, Sep 01, 2006 at 09:49:00AM -0400, Rich Felker wrote:
> On Fri, Sep 01, 2006 at 12:56:44AM +0200, Michael Niedermayer wrote:
> > Hi
> > 
> > On Thu, Aug 31, 2006 at 05:45:14PM -0400, Rich Felker wrote:
> > > On Thu, Aug 31, 2006 at 09:21:40PM +0200, Luca Barbato wrote:
> > > > lu_zero wrote:
> > > > [commit of 5 hunks instead of the 2 approved]
> > > > 
> > > > Sigh, I hate that, the other 3 hunks are related to it but not much
> > > > discussed =_=
> > > 
> > > Why was all this crap committed at all?? The code should be fixed to
> > > work correctly when the buffer is not aligned by performing the
> > > initial/final segments with slower code until the pointer is aligned.
> > 
> > that is likely not possible, just think of the case for video, rendering
> > blocks into unaligned memory its not a matter of initial and final cases
> > audio isnt much different, just replace the 8x8 dct by a 1024 point imdct
> > or a 32 point dct/synth filter thingy whch mp* uses
> 
> hmm, but what should the alignment be then? 4? 8? or some
> arch-specific constant? i don't object to alignment requirements but i
> do object to arch-specific requirements on the api. how is a portable
> program supposed to know what alignment to pass in?

require 16byte alignment (the max iam aware of which is needed by some arch)
or
#define the required alignment in a libavcodec or libavutil header

[...]
-- 
Michael     GnuPG fingerprint: 9FF2128B147EF6730BADF133611EC787040B0FAB

In the past you could go to a library and read, borrow or copy any book
Today you'd get arrested for mere telling someone where the library is




More information about the ffmpeg-cvslog mailing list