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

[Mike Melanson]

On Fri, 28 Dec 2001, Arpi wrote:

> > 	No more msadp32.acm. Enjoy.
> 
> nice work again!
> 
> time to optimize for sse and 3dnow :))))))

	I thought, however briefly, about that. It seems rather 
difficult to use SIMD-type instructions on predictive algorithms. SIMD
instructions are good for parallelizable tasks, right? It's hard to
parallelize ADPCM since f(x) = f(f(x-1)).

	Hmm...but all the nibbles have to be broken apart, which can be
done at the same time. Then there's the fact that all of those PCM samples
have to be clamped, which implies saturation functionality could be
employed. Who knows? This might turn out to be good elementary practice
for advanced instructions.

	Then again, I may not have any idea what I'm talking about. And
believe me, that's entirely possible. Feel free to call me on any of this.

-- 
	-Mike Melanson