[MPlayer-dev-eng] NEEDS DECISION: liba52 upgrade?

[D Richard Felker III]

On Mon, Feb 10, 2003 at 02:06:48AM +0100, Arpi wrote:
> Hi,
> 
> The liba52 CVS contains algorith-level optimized code (including optional
> integer-only implementation of teh decoder) but with no CPU-specific
> (3dnow/sse/etc) optimizations. Walken (maintainer) said that his C-only
> but algo-level optimized code is/may be faster than our hardcore asm
> optimized version based on the old C code.
> 
> Anyway we agreed on that improvements of both trees should be merged somehow.
> The question: is anyone interested/capable of porting the asm optims to the
> new liba52? Imho it shouldn't be hard, but requires basic asm knowledge and
> knowledge of that code... maybe Michael could help, unless he's too busy
> with libavcodec. But dunno about the 3dnow part, any volunteers?
> 
> Should I replace current liba52 in 'main' mplayer tree by the cvs liba52,
> or clone liba52 as a new tree at mphq?
> It would be easier to commit to liba52 SF project directly but Walken wants
> patches not commits :(

I say go all the way and replace liba52 in CVS. Integer-only should
already make it faster than our asm even without the algo-opts. Also I
tried comparing a52dec performance to mplayer recently, and current
liba52 seemed to do better. So, let's go ahead and put it in place and
then Michael or someone can add back asm if/when he has time.

Rich