[x264-devel] [PATCH] SSE2 motion compensation [from before] + MMX/SSE2/SSSE3 frame_lowres_init

[Jason Garrett-Glaser]

As I posted earlier, there's a small error in the patch; x shouldn't be
re-initted to zero after the first loop.

The x++ is correct, but it'd be better style-wise to use x+=4 for w64 ...
and x+=2 for the w32 loop.

Also note that upon fixing checkasm to use random data for the input, it
appears that the assembly disagrees with the C, so any thoughts there would
be appreciated.

On Fri, Mar 7, 2008 at 3:12 AM, Hannes Domani <ssbssa at yahoo.de> wrote:

> Hello
>
> when looking at the patch i was wondering if the
> following part is what you really intended (since x++
> is called twice):
> > -    if( qpel_idx & 5 ) /* qpel interpolation needed
> */
> > +    int x;
> > +    width = width >> 2;
> > +    for( x = 0; width - x >= 2; x++ )
> >      {
> > -        uint8_t *src2 = src[hpel_ref1[qpel_idx]] +
> offset + ((mvx&3) == 3);
> > -        x264_pixel_avg_wtab_mmxext[i_width>>2](
> > -                dst, *i_dst_stride, src1,
> i_src_stride,
> > -                src2, i_height );
> > -        return dst;
> > +        frame_init_lowres_core_sse2_w32(src_stride,
> dest_stride, height, width, src0, dst0, dsth, dstv,
> dstc );
> > +        src0 += 32;
> > +        dst0 += 16;
> > +        dsth += 16;
> > +        dstv += 16;
> > +        dstc += 16;
> > +        x++;
> >      }
>
>
> regards
> Domani Hannes
>
>
>      Machen Sie Yahoo! zu Ihrer Startseite. Los geht's:
> http://de.yahoo.com/set
>
> _______________________________________________
> x264-devel mailing list
> x264-devel at videolan.org
> http://mailman.videolan.org/listinfo/x264-devel
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mailman.videolan.org/pipermail/x264-devel/attachments/20080307/1e4ca52d/attachment.htm