The D_FLOAT type  denotes  a  D_Floating  double-precision  real
  value regardless of the setting of the FLOAT attribute or /FLOAT
  command line qualifier.