As far, as I'm familiar with math, float is only necessary for resampling as you need to calculate the time interpolation of the waveform (did this already several times). For a linear mixing job, one would simply add the channels (most usecases). Also float could be ommitted, as the Neon should...