FIX: Precision Problem with COMPLEX*16 and /Op

The information in this article applies to:

• Microsoft FORTRAN PowerStation for MS-DOS, version 1.0

SYMPTOMS

A program, compiled with the /Op option that contains double precision COMPLEX arithmetic may produce only a single precision result.

CAUSE

When optimizing for precision, the compiler always stores intermediate results of equations in memory locations rather than leaving the results on the coprocessor stack. The problem occurs because the COMPLEX*16 intermediate results are incorrectly stored in 8-byte memory locations.

RESOLUTION

Do not compile routines that perform COMPLEX*16 arithmetic with /Op.

STATUS

Microsoft has confirmed this to be a problem in FORTRAN PowerStation version 1.0 for MS-DOS. This problem has been resolved with FORTRAN PowerStation maintenance release version 1.0a for MS-DOS.

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

MORE INFORMATION

The following sample demonstrates the problem under FORTRAN PowerStation 1.0 when compiled with /Op:

Sample Code

c Compile options required: /Op

      COMPLEX*16 A,B,C
      REAL*8     D,E,F
      DATA A / (1.d0,0.d0) /
      DATA B / (.3d0,0.d0) /

      C = A/B
      D = REAL(A)
      E = REAL(B)
      F = D/E
      PRINT *,'C=',C
      PRINT *,'F=',F
      END