Using scripts/syn123-channelscaling.sh, I see about 15% benefit
for mono/stereo operation when treating it as special cases in
the resampler. Fun fact: Even the higher channel counts profit
a tiny bit (1% to 2%, not sure how significant) from the
separation.
Giving the compiler information about what can be likely values
for those inner loop rounds still makes sense.
git-svn-id: svn://scm.orgis.org/mpg123/trunk@4494 35dc7657-300d-0410-a2e5-dc2837fedb53
