C As this below tiny program from Tim Coe, coe@vitsemi.com demonstrates
C the Pentium procesor sometimes returns wrong results when using real
C division. EPS-dependent Fortran version.
C    To Fortran by H. D. Knoble, hdk@psuvm.psu.edu,  11/27/94
C    Penn State Center for Academic Computing

      DOUBLE PRECISION X,Y,Z,FDIV
      REAL EPS
C
C---Define SP EPS for the IBM PC.  EPS is the smallest real number on
C   this architecture such that 1+EPS>1 and 1-EPS<1.
      EPS=5.9604640E-08
      X = 4195835.0D0
      Y = 3145727.0D0
C---Z Theoretically should be a computational (fuzzy) Zero.
      Z = X - FDIV(X,Y) * Y
      WRITE(*,*) 'Residual Z=',Z
      IF (DABS(Z).LT.EPS) THEN
         WRITE(*,*) 'Since Residual Z is essentially Zero, then this'
         WRITE(*,*) 'Pentium Processor DOES NOT have a Divide bug.'
      ELSE
         WRITE(*,*) 'If Z is substantially not Zero, then this'
         WRITE(*,*) 'Pentium Processor DOES HAVE a Divide bug.'
      ENDIF
      STOP
      END
      DOUBLE PRECISION FUNCTION FDIV(X,Y)
      DOUBLE PRECISION X,Y
      FDIV = X/Y
      RETURN
      END