In the parallel program I'm writing, I defined a lot of multidimensional allocatable arrays (actually just 1D, 2D or 3D) which are allocated with negative lower bounds during the execution.
The reason why I did that is that each process handles, in a 2D example, a 14 by 14 A matrix and shares an overlap of 2 layers with the four neighboring processes, so the matrix is allocated with bounds (-1:12,-1:12), where the "internal" part of the matrix corresponds to indices from 1 to 10. (Italic represent choice/input dependency).
The problem is that in doing this I didn't realize that it would have been impossible to rely on this Fortran2003 feature, thus making impossible to call mysub(A(:,i),...) in the caller unit and refer to the dummy counterpart of A in the subroutine using the actual bounds (-1 and 12), since A(:,i) is an expression, and not just a variable.
The problem is solved if I pass the whole array A to the subroutine, as explained in the previous linked answer, but this would force me to write a "bigger" and less universal subroutine, which, on the contrary, is meant to act on one-dimensional subarrays.
Is it possible to pass a subarray of an allocatable array (e.g. A(:,1)) to a subroutine in a way that the subroutine is aware of the bounds of the actual variable?
