do some 16-bit GLchan computations in floating point (Gerk Huisma)