Knowledge Base

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

PSS ID Number: 117903

Article Last Modified on 3/24/2000


The information in this article applies to:


This article was previously published under Q117903

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
				

Additional query words: 1.00 1.00a buglist1.00 fixlist1.00a

Keywords: kbbug KB117903
Technology: kbAudDeveloper kbFORTRANPower100DOS kbFortranSearch kbZNotKeyword3