Newer version, using a vectorized version of thenew organisation of code in doVertLowPass. itseems to be faster in AltiVec also...
Also includes a compile fix for the new do_a_deblockwhen using AltiVec.
patch by (Romain Dolbeau <dolbeau at irisa dot fr>)...
altivec optimizations and horizontal filter fix by (Romain Dolbeau <dolbeau at irisa dot fr>)
Originally committed as revision 3162 to svn://svn.ffmpeg.org/ffmpeg/trunk