Each floating point register is 32 bits long to provide for single
precision IEEE FPS; registers are paired to support double precision.