[MPlayer-dev-eng] MS-ADPCM/Stereo Works

[Michael Niedermayer]

Hi

On Friday 28 December 2001 14:27, Arpi wrote:
> Hi,
>
> > > 	There may yet be some opportunity. Is there any efficient way to
> > > rip apart nibbles and interleave them using SIMD instructions? Also,
> >
> > interleave them bytewise and mask and or shift at the end
> > interleave 8 blocks -> n11, n12, n21, n22, n31, n32 ... (4-bit each)
> > after masking (pand) -> n11, n21, n31, ... (8-bit each)
> > or shift&mask (pand, psrlw $4) -> n12, n22, n32, ... (8-bit each)
>
> ehh...
>
> back to the reality :)
:(

>
> is it possible to use MMX for liba52?
> (for non-sse systems)
yes
imdct.c is C or SSE
downmix.c is C or SSE (with tiny bit mmx ...)
resample.c is C or MMX (no SSE)
it might be possible to do the imdct with 32-bit intergers (in mmx) but 
overflows and such could be a problem i guess

>
> are all of those optims sse-only, or may work (mayeb with small mods) on
> 3dnow?
some SSE, some MMX, iam not an 3dnow expert and i have no 3dnow compatile 
chip, but i could try to do it blindly and we'll see if it works ... 

>
> amd k6 users reporetd that libac3 still faster than liba52 :(
> (but i really want to kill libac3 from release)
i see, -> kill k6 users ;)
the k6 has a very slow fpu iirc ...

Michael