Hi, I've just finished vectorization of the Adler32 algorithm. this algorithm is part of zlib and also libmhash (in a slightly different implementation but basically the same).

I got a 2-2.5x speed increase from this, I believe I could get a bit more out of it, but I describe the process in more detail in this paper:

http://people.debian.org/~markos/powerp ... ltivec.pdf

Comments, suggestions welcome.

I will look into other parts of zlib, and also the runtime Altivec detection so that it will use the proper version automatically.

Konstantinos

The same algorithm is present in ffmpeg and could enjoy your optimization, pretty nice paper btw =)