http://bugs.winehq.org/show_bug.cgi?id=14717
--- Comment #59 from Alexander E. Patrakov patrakov@gmail.com 2010-12-22 06:01:44 CST --- As for bit depth conversion: the original code provides functions for converting any sample format (8, 16, 24 or 32 bits) into any. The new code achieves the same goal by providing the getsample() and putsample() functions that convert anything to and from double and thus achieve the same goal.
There is one bug, though, in their implementation for 24-bit samples: it doesn't take the sign bit into account.
Also, it would be nice to avoid filtering the input through the FIR altogether if no resampling is needed (e.g., 44.1k -> 44.1k).