"Dynamic allocation" and "modules": Is this a bug or a functionnality?

[François-Xavier]

Dear Intel Fortran forum,

I have discovered the following behaviour which can be reproduced with the code attached to this thread.

I have lots of big old programs which uses large amount of static data (and since its windows i am limited to 2GB, even in 64 bits compilation).

I want to do minimal changes to use allocatable arrays.

I have created the following code and i am very suprised that it works:

[fxfortran]      PROGRAM Dynamic_module_example      
               
        IMPLICIT NONE
 
        CALL DYNAMIC_ALLOCATION
        CALL SUBROUTINE1
        CALL SUBROUTINE2
        
        PAUSE
                       
     END PROGRAM Dynamic_module_example[/fxfortran]

[fxfortran]     MODULE arrays
        REAL, ALLOCATABLE :: A(:),B(:),C(:)
     END MODULE[/fxfortran]

[fxfortran]     SUBROUTINE DYNAMIC_ALLOCATION()

     USE arrays
     
     INTEGER*4 I
     
     ALLOCATE (A(1000000),B(1000000),C(1000000))
     
     RETURN
     END SUBROUTINE[/fxfortran]

[fxfortran]    SUBROUTINE SUBROUTINE1()

     USE arrays
     
     INTEGER*4 I
     
     DO I=1,10000
        A(I)=SIN(I*1.0)
        B(I)=COS(I*1.0)
        C(I)=TAN(I*1.0)
     END DO
     RETURN
     END SUBROUTINE[/fxfortran]

[bash]   SUBROUTINE SUBROUTINE2()

     CALL SUBROUTINE3

     RETURN
     END SUBROUTINE[/bash]

<

[bash]     SUBROUTINE SUBROUTINE3()

     USE arrays
     
     INTEGER*4 I
     
     DO I=1,100
        PRINT '(3F12.5)',A(I),B(I),C(I)
     END DO

     RETURN
     END SUBROUTINE[/bash]

Please note that i have read the documentation and haven't found a clear explanation about it.


In fact, i use the MODULE as a COMMON for dynamic variables and i wonder why my variables are correctly maintained in memory.


Is this dangerous, is this a normal behaviour ?Thank you.F.X

[Andrew_Smith]

If you had used the save attribute for the module variables your code should work. But without save it is not required to work. I dont think the compiler is required to deallocate module variables that go out of scope. Perhaps others can confirm.

[François-Xavier]

Hello Andrew,

Thank you for your reply.

i didn't use the "SAVE" attribute, i just used the default compiler option (which does not USE SAVE if my memory serves me right)

[Jugoslav_Dujic]

Quoting Andrew Smith

If you had used the save attribute for the module variables your code should work. But without save it is not required to work. I dont think the compiler is required to deallocate module variables that go out of scope. Perhaps others can confirm.

Theoretically true, but only of academic interest. For all intents and purposes, module variables on modern compilers never go out of scope. Thus, SAVE attribute is required only for standard-compliance (you will get a warning if you turn on standard checking), but not really necessary.

[François-Xavier]

Quoting Jugoslav Dujic

Quoting Andrew Smith

If you had used the save attribute for the module variables your code should work. But without save it is not required to work. I dont think the compiler is required to deallocate module variables that go out of scope. Perhaps others can confirm.

Theoretically true, but only of academic interest. For all intents and purposes, module variables on modern compilers never go out of scope. Thus, SAVE attribute is required only for standard-compliance (you will get a warning if you turn on standard checking), but not really necessary.

Thank you for your response.

To be clear, what conclusion can i understand? Is my testcase a dangerous way to use module variables?

Thank you,

F-Xavier

[Les_Neilson]

Quoting Franois-Xavier

To be clear, what conclusion can i understand? Is my testcase a dangerous way to use module variables?

In practice - no
It is theoretically possible that therecould be a compiler/processer/OS combination that will cause this sort of code to fail because themodule hasgone out of scope. But the de facto situation at the moment is that this will "just work"

Les

[Steven_L_Intel1 · Accepted solution]

Module variables are implicitly SAVEd. Earlier versions of the Fortran standard did not specify this, though I know of no compiler which would behave otherwise. Fortran 2008 now specifies the implied SAVE which everyone applied anyway.

[François-Xavier]

Thank you Steve,

It would be very nice to be clearly explained in MODULE documentation.

F-Xavier