[x264-devel] [PATCH] Speedups of various sorts involving macroblock_encode and RDO.

[Jason Garrett-Glaser]

The 4th speedup was simply changing CABAC copy (which is done on every
single RDO call, even the single-partition i4x4 RDO calls) from
memcpy() to a custom ASM function.

On 32-bit Windows, this cuts the clock cost from ~200 clocks to ~53
clocks.  Compare this to a total cost of ~9000 clocks for
full-macroblock RDO, and I would assume 10-20 times less for
single-partition RDO.

Dark Shikari

On Fri, Mar 14, 2008 at 2:53 AM, Guillaume POIRIER <poirierg at gmail.com> wrote:
> Hello,
>
>
>  On 3/13/08, Jason Garrett-Glaser <darkshikari at gmail.com> wrote:
>  > I apologize, Gmail appears to have destroyed the formatting of my patch.
>  >
>  >  The correct version can be found at http://pastebin.com/f55b55c85 .
>
>  What speed-up did you measure regarding C cabac VS ASM cabac, and
>  overall speed-up? With what encoding parameters?
>
>  Guillaume
>  --
>  A soldier will fight long and hard for a bit of colored ribbon.
>   -- Napoleon Bonaparte
>  _______________________________________________
>  x264-devel mailing list
>  x264-devel at videolan.org
>  http://mailman.videolan.org/listinfo/x264-devel
>