How DLL Export from file Fortran 77

[BABA767]

Hi everyone,
I have some Fortran 77 sources file that I'd like to compile as a DLL so I can call functions from an external GUI to interact with Fortran DLLs.
I compiled the DLL with the latest version of the Intel compiler, and apparently the DLL is generated correctly and called from the external GUI (in C#).
But I've noticed a behavioral issue with the DLL because it doesn't return the expected value of a variable. The problem isn't related to the source file, since compiling the .for file in Visual Studio returns the correct values.

I believe the problem is in the compiler settings or in the directives inserted in the .for file; can anyone help me resolve the issue?
Thanks,
Valerio

[andrew_4619]

I suggest you attach the call and declarations from c# and the header of the Fortran routine in Fortran. Maybe you have a mismatch but we would be guessing.

[BABA767]

Ok I report the extract of my two files .for and .cs:

file .for 77 (two subroutine)

      SUBROUTINE B767AH

      IMPLICIT NONE

.....here there are global vars

      ENTRY  AHYD1()

!DEC$ ATTRIBUTES DLLEXPORT :: AHYD1

      DTPA = 13

...here rest of code

      SUBROUTINE GET_RDEMPRESS(val)

!DEC$ ATTRIBUTES DLLEXPORT :: GET_RDEMPRESS      

      IMPLICIT NONE

      REAL*4   AHP01

       LOGICAL*1  DUM0000001(1022488)

      COMMON   /XRFTEST   /

     & DUM0000001,AHP01

      REAL*4 val

      val = AHP01

      END

import of function from DLL (file c#)

        [DllImport("b767ag.dll", CallingConvention = CallingConvention.Cdecl)]

        public static extern void AHYD1();

        [DllImport("b767ag.dll", CallingConvention = CallingConvention.StdCall)]

        public static extern void GET_RDEMPRESS([Out] float[] x);

When i call cyclically the main routine (at ENTRY point) I read correctly value 13 come from AHP01 by GET_RDEMPRESS but in true that value should increase infact if I run the routine directly from the .for file into Visual code that value increase. So i presume that I'm missing or doing something wrong in declaration or in the compilation. Maybe I've to use ifort instead of ifx as my .for 77 is too older?

I compiled my .dll with these commands:

ifx /c myfile.for ->create the file .obj

ifx /dll myfile.obj /link /out:myfile.dll ->create the file .dll

[BABA767]

I've solved using /Qsave parameter during the compilation, this permits to allocate the local variables into the memory without pass they into the stack.

cheers!

V.

[andrew_4619]

Well that is both good and bad. You haven't fixed it really you have just hidden the problem which is sat there waiting to bite. I would to some builds and runs with the checks for uninitialised variables.

[jimdempseyatthecove]

>>I've solved using /Qsave parameter during the compilation

Do you intend to (later) call your DLL from a multi-threaded process?

If so, then the /Qsave may be problematic.

Jim

[BABA767]

Hi,

yes I'm using multi-thread but each subroutine with /Qsave is called from only one dedicated thread so local variables should not be at risk of being overwritten.

Valerio

[BABA767]

Why might I have hidden problems? The Fortran source is secure and tested, including local variables. Even uninitialized variables will be assigned a value, the important is that the value is not lost.. or not?

[andrew_4619]

Well if you have an issue with mismatched variables on calls or calling conventions or uninitialised variables changing the memory management or layout does not stop corruption it might just mean what is being corrupted is not important but after further code changes.......

[BABA767]

ok that's correct.

Just for info I'm using CallingConvention "Cdecl" C declaration so the caller (my  C# GUI) clear the stack; in the file .for77 use this directive -> !DEC$ ATTRIBUTES DLLEXPORT :: name_of _subroutine

It's correct use this or better STDCALL?  I honestly don't know which one is better to use. I only understood that they must be declared in the same way to correctly manage stack cleaning.

[GVautier]

As far as I know,

with cdecl convention, arguments are pushed to the stack before calling the procedure who doesn't have to know the actual size of arguments to return correctly. That allows variable number of arguments.

with stdcall convention,, it's the opposite so the called procedure use the declared size of arguments to get the return address from the stack. It's why the arguments size is added to the procedure name to avoid mismatches. Optional arguments must be declared and pushed as null pointer if absent to preserve the argument size.

cdecl guaranties that the procedure will return correctly even if arguments number or size is incorrect may be silently.

stdcall will crash if arguments number or size is incorrect.

[BABA767]

Okk thk for your advice..I'll use Cdecl convention..;-)

[MartinSpeare]

>>I've solved using /Qsave parameter during the compilation

or seond sol is this 

SUBROUTINE B767AH

      IMPLICIT NONE

.....here there are global vars

      ENTRY  AHYD1()

!DEC$ ATTRIBUTES DLLEXPORT :: AHYD1

      DTPA = 13

...here rest of code

      SUBROUTINE GET_RDEMPRESS(val)

!DEC$ ATTRIBUTES DLLEXPORT :: GET_RDEMPRESS      

      IMPLICIT NONE

      REAL*4   AHP01

       LOGICAL*1  DUM0000001(1022488)

      COMMON   /XRFTEST   /

     & DUM0000001,AHP01

      REAL*4 val

      val = AHP01

      END