Change ec_{enc|dec}_bit_prob to take probabilities in units of 1/65536 instead of...
authorTimothy B. Terriberry <tterribe@xiph.org>
Sun, 30 May 2010 03:02:33 +0000 (23:02 -0400)
committerJean-Marc Valin <jean-marc.valin@usherbrooke.ca>
Sun, 30 May 2010 03:45:28 +0000 (23:45 -0400)
commit43e940620151128e0cea9f05ee9ec58d1d64c93f
treeef48419faabf7f3b4d8505fe2459814a699476cb
parent299747ee24851f2f20a8a466a3a5b58f1b116768
Change ec_{enc|dec}_bit_prob to take probabilities in units of 1/65536 instead of 1/256. This allows them to use a single 16x16->32 multiply instead of a 24x8->32 multiply. Also change the time-frequency resolution flag coding to ensure that "0" is always the most-probable symbol (i.e., that prob("1")<50%), as that's where all the rounding error accumulates.
libcelt/celt.c
libcelt/entdec.h
libcelt/entenc.h
libcelt/rangedec.c
libcelt/rangeenc.c