[FFmpeg-soc] Differences between C, MMX, and MMX2 horizontal filters and xInc values. (was: Differing)

[Ramiro Polla]

What the ...? I can't read the entire message in
http://lists.mplayerhq.hu/pipermail/ffmpeg-soc/2009-July/date.html

And I screwed up the subject.

On Wed, Jul 22, 2009 at 1:58 AM, Ramiro Polla<ramiro.polla at gmail.com> wrote:
> Hi,
>
> In swscale.c, there are some lines that adjust {lum,chr}XInc depending
> on MMX or MMX2 being used for fast bilinear. They were added (still
> with the old name of s_xinc) in this commit:
> http://git.ffmpeg.org/?p=libswscale;a=commitdiff;h=87d2542e33c80731dab64339aebcaebbaa25911e
>
> Where do the magical "20" values come from?
>
> From what I understood and from some tests I've done, this creates
> some rather ugly last lines at the far right of the image. The output
> also differs for C, mmx, and mmx2, and it's not under BITEXACT.
>
> To remove this adjustment and have the same output for all
> configurations, wouldn't it only be necessary to:
> - adjust mmx and mmx2 filters to only work up to the last input pixels
> it can before reading past the end.
> - run the C code on the last pixels if necessary.
>
> This should also remove some +1s being allocated because of the MMX
> filter, and the limit that the MMX2 filter can only work with srcW&31.
>
> Ramiro Polla
>