[FFmpeg-cvslog] r21736 - trunk/libavcodec/x86/dsputil_mmx.c

[Michael Niedermayer]

On Fri, Feb 12, 2010 at 05:19:03PM +0000, M?ns Rullg?rd wrote:
> Michael Niedermayer <michaelni at gmx.at> writes:
> 
> > On Thu, Feb 11, 2010 at 03:30:46PM +0000, M?ns Rullg?rd wrote:
> >> Michael Niedermayer <michaelni at gmx.at> writes:
> >> 
> >> > On Wed, Feb 10, 2010 at 05:08:07PM -0500, David Conrad wrote:
> >> >> On Feb 10, 2010, at 4:16 PM, Alex Converse wrote:
> >> >> 
> >> >> > On Tue, Feb 9, 2010 at 9:02 PM, conrad <subversion at mplayerhq.hu> wrote:
> >> >> >> Author: conrad
> >> >> >> Date: Wed Feb 10 03:02:06 2010
> >> >> >> New Revision: 21736
> >> >> >> 
> >> >> >> Log:
> >> >> >> Enable SSE2 (put|avg)_pixels_16_sse2
> >> >> >> 
> >> >> >> SVQ1 chroma has been special-cased aligned to 16-bytes since at least r15466
> >> >> >> Other architectures also assume 16-byte alignment here too but set STRIDE_ALIGN
> >> >> >> to 16.
> >> >> >> 
> >> >> >> Modified:
> >> >> >>   trunk/libavcodec/x86/dsputil_mmx.c
> >> >> >> 
> >> >> > 
> >> >> > This broke vp6f:
> >> >> > http://fate.multimedia.cx/index.php?build_record=178551
> >> >> > http://fate.multimedia.cx/index.php?build_record=178488
> >> >> 
> >> >> Looks like vp5 and vp6 use width 16 pixel functions on chroma as well. I'd really prefer if SSE used 16-byte aligment like other architectures (first patch). Otherwise VP5/6 need to be special-cased as well (second patch), and potentially other codecs that FATE doesn't test, which gets extremely ugly fast.
> >> >
> >> > i prefer the second patch
> >> > otherwise extensive benchmaring is needed
> >> 
> >> Maybe we could make it a codec capability flag or something instead of
> >> an ever-growing list of special cases.
> >
> > yes, but it should go into a non public header, the user app has no need to
> > mess with it nor should we have to bump the ABI if we decide to increase
> > STRIDE_ALIGN one day and remove this
> 
> An app providing its own get_buffer would need to know, methinks...

hmm

maybe we should add avcodec_align_dimensions2() that provides
width/height/lumastride/chromastide alignment values

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

It is not what we do, but why we do it that matters.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 189 bytes
Desc: Digital signature
URL: <http://lists.mplayerhq.hu/pipermail/ffmpeg-cvslog/attachments/20100212/7d805119/attachment.pgp>