[MPlayer-dev-eng] [PATCH] (new version) AltiVec: dct64 for mp3lib, IMDCT for liba52, detection code

[Romain Dolbeau]

Daniel Egger wrote:

> We should try to get a statement here. I believe that the line_size
> is always a multiple of 16 for the formats we're interested in; 
> even VirtualDub breaks when this is not the case.

That would be interesting, as many computations could
be moved out-of-loops in altivec (OK. mostly permutations
and useless loads, but still, it would be faster).

> Oooh, you're writing your thesis about AltiVec in mplayer?

Not at all ! AltiVec is just a way to watch movies
faster ;-) But the team works in comp.arch and
compilation, and I wondered if according to (real)
experts it was worth the hassle (to remove the
spurious load). Of course the answer is 'YMMV', but
if we can fix the line_size to a multiple of 16,
then it's likely worth it (as we'll need the load for
either all or no iterations of the loop).

-- 
Romain Dolbeau