[FFmpeg-devel] [PATCH 0/5] Scalar product, take 2

[Christophe Gisquet]

Following the previous comments, I have opted for padding of the buffers in
order to use the DSP functions.

The troublesome cases for wma lossless are still untested. Patches 3 and 4
still try to address them, 3 being more secure than 4 (which is the shape of
the best solution in my opinion).

TAK decoder was modified in the same fashion.

Christophe Gisquet (5):
  ra144: use scalarproduct_int16
  wma lossless: reuse scalarproduct_and_madd_int16
  wma lossless: switching back to C dsp function
  wma lossless: pad coeff buffer with 0
  takdec: pad filter coeff buffer for DSP functions

 libavcodec/dsputil.c        |  4 +--
 libavcodec/dsputil.h        |  3 +++
 libavcodec/ra144.c          | 14 ++++------
 libavcodec/ra144.h          |  6 ++++-
 libavcodec/ra144dec.c       |  3 +++
 libavcodec/ra144enc.c       |  6 +++--
 libavcodec/takdec.c         | 33 +++++++-----------------
 libavcodec/wmalosslessdec.c | 63 ++++++++++++++++++++++++---------------------
 8 files changed, 65 insertions(+), 67 deletions(-)

-- 
1.8.0.msysgit.0