[FFmpeg-devel] [PATCH] MMX implementation of VC-1 inverse transforms

[Måns Rullgård]

Balatoni Denes <dbalatoni at interware.hu> writes:

> Hi!
>
> Monday 14 January 2008 20:50-kor Ivan Kalvachev ezt ?rta:
>> On Jan 14, 2008 1:05 PM, Michael Niedermayer <michaelni at gmx.at> wrote:
>> > On Sun, Jan 13, 2008 at 05:10:30PM +0100, Christophe GISQUET wrote:
>
>> - Have you seen how other IDCT optimizations work? I may be wrong but
>> vc1 transformations look like IDCT with quite simplified (smaller)
>> coefficients.
>
> Figuring out how the other IDCTs work is not so simple task as you make it 
> look like. For example the simple_idct is the fastest and most accurate, but 
> it is quite involved (some crazy reordering imho), I for one couldn't figure 
> out how it works during the time I spent on it.

Realising how it implements an IDCT is indeed non-trivial.  Writing a
SIMD version is still a fairly straight-forward task, though.

-- 
M?ns Rullg?rd
mans at mansr.com