Precision difference between Linux and Windows (Fortran77)?

[espenbe]

Consider the following small piece of code (Fortran77) run on 32bit architecture:


PROGRAM TEST
C
IMPLICIT NONE
DOUBLE PRECISION A, B, C
C
A = 319.0D0 / DACOS(-1.0D0)
B = 761.0D0 / 93.0D0
C = A - B
C
WRITE(6, '(3EN35.25)') C, A - B, C - (A - B)
C
END


In this case I would assume "C - (A - B)" would give me "zero" (0.0D0), but it doesn't on Linux. If I compile the same piece of code in Windows, it gives me zero. By the way, if I use Absoft ProFortran compiler I also get zero. Why is this happening? Can I do something with compiler flags? I tried "-O1" and "-O2", but they didn't help me. Simple compiler string was

ifort -O -o test test.f

Result from Linux:
93.3580579937044916505328729E+00 93.3580579937044916505328729E+00 -3.5527136788005009293556213E-15

Result from Windows:
93.3580579937044916505328729E+00 93.3580579937044916505328729E+00 0.0000000000000000000000000E+00

Using the compiler option "-pc64" on Linux did not affect the output. Changing the compiler option to "/Qpc80" on Windows made the output equal the one from Linux. The two physical computers are equal, "only" the operating system differs.

Any idea of what I could do to get consistent results from the two compilers?
Compiler versions:
Windows: 11.0.066
Linux: 11.1.046

I also tried with 10.1.022 for Linux with same result.

Regards

Espen

[Kevin_D_Intel]

Compile with: -assume protect-parens (Linux) or /assume:protect-parens (Windows). The default is noprotect-parens.

I see consistent compiler treatment and resultsacross Linux and Windows.

Linux

$ ifort -V u68415.f90 -o u68415
Intel Fortran Intel 64 Compiler Professional for applications running on Intel 64, Version 11.1 Build 20090827 Package ID: l_cprof_p_11.1.056
Copyright (C) 1985-2009 Intel Corporation. All rights reserved.

Intel Fortran 11.1-2582
GNU ld version 2.17.50.0.6-5.el5 20061020

$ ./u68415
93.3580579937044916505328729E+00 93.3580579937044916505328729E+00 -3.5527136788005009293556213E-15

$ ifort -V u68415.f90 -o u68415 -assume protect-parens
Intel Fortran Intel 64 Compiler Professional for applications running on Intel 64, Version 11.1 Build 20090827 Package ID: l_cprof_p_11.1.056
Copyright (C) 1985-2009 Intel Corporation. All rights reserved.

Intel Fortran 11.1-2582
GNU ld version 2.17.50.0.6-5.el5 20061020

$ ./u68415
93.3580579937044916505328729E+00 93.3580579937044916505328729E+00 0.0000000000000000000000000E+00


Windows

$ ifort u68415.f90
Intel Visual Fortran Intel 64 Compiler Professional for applications running on Intel 64, Version 11.1 Build 20090511 Package ID: w_cprof_p_11.1.035

Copyright (C) 1985-2009 Intel Corporation. All rights reserved.

Microsoft Incremental Linker Version 9.00.21022.08
Copyright (C) Microsoft Corporation. All rights reserved.

-out: u68415.exe
-subsystem:console
u68415.obj

$ .u68415.exe
93.3580579937044916505328729E+00 93.3580579937044916505328729E+00 -3.5527136788005009293556213E-15


$ ifort /assume:protect-parens u68415.f90
Intel Visual Fortran Intel 64 Compiler Professional for applications running on Intel 64, Version 11.1 Build 20090511 Package ID: w_cprof_p_11.1.035

Copyright (C) 1985-2009 Intel Corporation. All rights reserved.

Microsoft Incremental Linker Version 9.00.21022.08
Copyright (C) Microsoft Corporation. All rights reserved.

-out: u68415.exe
-subsystem:console
u68415.obj

$ .u68415.exe
93.3580579937044916505328729E+00 93.3580579937044916505328729E+00 0.0000000000000000000000000E+00

[espenbe]

I can confirm your results and I now get consistent output from Linux and Windows. It is really possible to get a bit lost in all these compilation options...

Thank you so much for the help. It is highly appreciated.

[Kevin_D_Intel]

Quoting - espenbe

It is really possible to get a bit lost in all these compilation options...

Thank you so much for the help. It is highly appreciated.

Indeed it is. There are a lot of them andI cetainly do not remember them all myself.

Glad I could help.