[x264-devel] add optimised functions (hpel_filter, ssd, ssim)

[Noboru Asai]

Hi,

Thank you for your commit.

2007/11/19, Guillaume POIRIER <poirierg at gmail.com>:

> o pixel_ssd_8x8_altivec()
>
> As far as this once is concerned, I seem to be too sleepy to find a
> set of command-line options that trigger this usage. Would you have
> one that I could use for testing?
>
> Not committed.


 This function is called at

encoder/rdo.c:

int ssd_mb( x264_t *h )
{
    return h->pixf.ssd[PIXEL_16x16]( h->mb.pic.p_fenc[0], FENC_STRIDE,
                                     h->mb.pic.p_fdec[0], FDEC_STRIDE )
*        + h->pixf.ssd[PIXEL_8x8](   h->mb.pic.p_fenc[1], FENC_STRIDE,
                                     h->mb.pic.p_fdec[1], FDEC_STRIDE )
*        + h->pixf.ssd[PIXEL_8x8](   h->mb.pic.p_fenc[2], FENC_STRIDE,
                                     h->mb.pic.p_fdec[2], FDEC_STRIDE );
}

ssd_mb() is called at

o x264_rd_cost_mb() ( m>=6 )
o x264_macroblock_analyse() ( B-Frame, brdo, m>=6 )

so I use the following options.

--bframes 3 --ref 5 -m 7 --b-rdo

BTW, If you're looking for some more stuff to optimize, maybe you
> could have a look at the macro
> ALTIVEC_STORE_SUM_CLIP: the 2 known cases where this macro is used is
> with 8-byte aligned memory or 16-bytes aligned memory, so it should be
> possible to write a specialized version for these two cases, that
> would be a bit faster.
>
> This has been discussed already in the thread:
> http://www.via.ecp.fr/via/ml/x264-devel/2006-09/msg00068.html
>

OK. I'll see them.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mailman.videolan.org/pipermail/x264-devel/attachments/20071119/27507596/attachment.htm