[Ffmpeg-devel] [PATCH] h264 *_pixels16_l2_altivec

Michael Niedermayer michaelni
Wed Mar 8 12:47:45 CET 2006


Hi

On Wed, Mar 08, 2006 at 10:12:38AM +0100, Luca Barbato wrote:
> Loren Merritt wrote:
> > On Tue, 7 Mar 2006, Luca Barbato wrote:
> >> Romain Dolbeau wrote:
> >>>
> >>> I think the fact that src_stride2 is hardwired to 16
> >>> should be mentioned above the functions, should it
> >>> be called from someplace that doesn't know about
> >>> the restriction.
> >>
> >> The previous code assumed the same, so I just removed the define and
> >> hardwired it there.
> >>
> >>> Doesn't H264 requires image_stride % 16 == 0 ? Because
> >>> if it can be assumed src_stride1 % 16 == 0, then the
> >>> vec_lvsl can be moved out of the loop.
> >>
> >> Not sure I check this evening
> >>
> >>> (I don't think that src1/src2/dst are always properly
> >>> aligned ? that could solve the performance problem)
> >>
> >> sadly they are all unaligned.
> > 
> > dst and one of the src can be made aligned.
> > 
> 
> about src probably is feasible, but I wonder how you can align dst since
> it could be a buffer provided by uncheckable means (like X), probably
> I'm missing something

we dont support direct rendering into buffers which are missaligned or have
missaligned stride

[...]

-- 
Michael





More information about the ffmpeg-devel mailing list