divergent execution outcomes given by the IVF and PVF compilers

[L__Juan]

I encounter a problem when I try to compile a source code with the following module with two compilers in visual studio 2010: the Intel visual fortran compiler would show a runtime error saying that "error#8169: The specified interface is not declared. [FUNCTION_TEMPLATE] at line 15," while the Portland visual fortran compiler could run it without issuing any warning. Can anyone help me out with figuring out anything wrong?

Thanks, Li

PS. I use this somewhat complicated declaration to implement the Broyden algorithm solving equations system of arbitrary size conditional on the real problem. There are two function types declared in the abstract interface area: The first one is just for describing the equation system of interest; The second one describes the penalty function used for the squared error minimization in the Broyden algorithm.

01 MODULE toolbox

02 IMPLICIT NONE

03 ABSTRACT INTERFACE

04 FUNCTION function_template(x) RESULT(y)

05 IMPLICIT NONE

06 REAL, DIMENSION(:) :: x

07 REAL, DIMENSION(SIZE(x)) :: y

08 END FUNCTION function_template

09

10 FUNCTION penalty_template(x,fvec_p,proc_p) RESULT(y)

11 IMPLICIT NONE

12 REAL, DIMENSION(:) :: x

13 REAL, DIMENSION(:), POINTER :: fvec_p

14 REAL :: y

15 PROCEDURE(function_template), POINTER :: proc_p

16 END FUNCTION penalty_template

17 END INTERFACE

18 CONTAINS

19 ...

20 END MODULE toolbox

[mecej4]

Since you showed only a paraphrase of the code, the following is a guess: in order to use entities defined in the module in an interface, you may need an IMPORT statement.

[Steven_L_Intel1]

The code posted is sufficient to illustrate the problem, though it would have been easier if the line numbers were not included.

mecej4 is correct - inside an interface, no host-associated names are visible by default. You indeed need to add IMPORT following the FUNCTION line in order to make the other interface visible.  I discuss this issue in Domestic or Imported? PGI Fortran has a bug here, or perhaps an extension to the standard that I would not think advisable.