Update the log2 approximation to accomodate the increased bit precision from bd0610d2...
authorTimothy B. Terriberry <tterribe@xiph.org>
Tue, 16 Mar 2010 03:49:28 +0000 (23:49 -0400)
committerJean-Marc Valin <jean-marc.valin@usherbrooke.ca>
Tue, 16 Mar 2010 04:17:38 +0000 (00:17 -0400)
commitdc8a6eee433ef2cf8866d7bfe051c331d6607a81
tree73a393a41e305035a115f2135d3c1d4a43437d26
parent3a4a463f0a1e17dad8adce1fa846743426e025c8
Update the log2 approximation to accomodate the increased bit precision from bd0610d21b28b76095a49e601df44fb792b96369. This requires another term in the power series to reduce the error to the level of truncation (peak absolute error 0.621785). Also refactor the rounding bias term so that further changes to DB_SHIFT will leave the result unbiased.
libcelt/mathops.h