This is my take on the case order=8. If that's the only case where it is not a multiple of 16, then that's probably not worth it.
The scalarproduct_and_madd_int16 originally required orders that are multiple of 16. wmalossless pro samples sometimes have an order of 8. The patchset first changes wmalossless by having padded buffers, so as to be able to use again 16-wide DSP, then restore the 16-wide constrain in them. Christophe Gisquet (2): wmalossless: reset lms_update lavc/lossless_audiodsp: revert various commits libavcodec/lossless_audiodsp.h | 3 ++- libavcodec/wmalosslessdec.c | 26 ++++++++++++++------------ libavcodec/x86/lossless_audiodsp_init.c | 26 ++------------------------ 3 files changed, 18 insertions(+), 37 deletions(-) -- 1.9.2.msysgit.0 _______________________________________________ ffmpeg-devel mailing list ffmpeg-devel@ffmpeg.org http://ffmpeg.org/mailman/listinfo/ffmpeg-devel