Hi,
Just to let you know that I've added a "raw_bits" branch in git.
It's
still a bit experimental, but the idea is to do a direct encoding of all
the bits that are written explicitly. This bypasses the range coder when
possible and just writes the bits at the end of the stream. There's an
overhead of about one bit, but it should save a bit of complexity and
(more importantly) make the bit-stream more robust to bit errors. It's
still a bit experimental, so please report any bug.
There's also a multiply-free (and divide-free) version of the range
coder I'm considering. You can enable it by replacing rangeenc.c and
rangedec.c by mfrngenc.c and mfrngdec.c in the build. I've had reports
of some problems on 16-bit archs, so be careful with that. I'm again
interested in feedback on that to decide whether the multiply-free algo
is worth it.
Cheers,
Jean-Marc
