FIX: Incorrect Floating-Point Calculations with /Ox and /Op

The information in this article applies to:

• Microsoft FORTRAN PowerStation for MS-DOS, version 1.0

SYMPTOMS

The program produces incorrect floating-point results when it is compiled using the /Ox and /Op optimizations (or /Oxp).

RESOLUTION

Do not use both /Ox and /Op when compiling.

STATUS

Microsoft has confirmed this to be a bug in FORTRAN PowerStation for MS-DOS, version 1.0. This problem was corrected in the FORTRAN PowerStation for MS-DOS maintenance release, version 1.0a.

MORE INFORMATION

FORTRAN PowerStation, version 1.0, can be differentiated from the maintenance release, version 1.0a, by invoking the linker. Typing "link32 | more" from the \F32\BIN directory will show version 2.8 for FORTRAN PowerStation, version 1.0, and will show version 1.0f for the maintenance release, version 1.0a.

The following sample program can be used to demonstrate this problem. The correct result is:

   2.929688E-03
   2.929688E-03

Sample Code

C compile options required: /Oxp

   n1 = 1024
   n2 = 512
   c = (1./n1)+(1./n2)
   write (*,*) c
   write (*,*) (1./n1)+(1./n2)
   end