Adopt suggestions as per https://software.intel.com/en-us/articles/copying-accelerated-video-decode-frame-buffers:
- 64kB aligned memory buffer
- Call to mfence only around copy to cache routine.

Use a single instruction to copy non-aligned memory at startup.
Remove compilation warning on mac

This result in a small 1.3% speed increase on an i7-4650U
Signed-off-by: Jean-Baptiste Kempf <jb@videolan.org>