Changed the structures of r17286 to be of numpy.float32 types instead of float16.
This is to speed up the calculation due to faster type conversions.
Authored by: bugman 2012-07-20
Parent: [r17290]
Child: [r17292]
