First thing you should do is clean up your argument typing to be consistent. You have the following:
mwSignedIndex :: num
:
call make_matrix(num, %val(something))
:
subroutine make_matrix(num, something)
integer :: num, i, j
It is entirely possible that the "mwSignedIndex" type is a 64-bit integer while the default "integer" type is a 32-bit integer ... even on a 64-bit system. Since the make_matrix subroutine is using an implicit interface, the compiler/linker will not catch this potential error. You should always ensure you have the same argument types, so maybe change that declaration in the subroutine to:
mwSignedIndex :: num, i, j
You may need to change that call to DLACPY as well.
Next thing you should do is NEVER use literal constants in Fortran for the arguments to the MATLAB API functions. You can get away with this in C/C++ where arguments are passed by value and you get automatic type promotion from the compiler ... but you can't get away with this in Fortran where you might have a mismatch since arguments are effectively passed by reference. Again, one typical cause is the 64-bit integer vs 32-bit integer thing. So always use the exact type specified in the API signature. In this case, the signature for mxCreateDoubleMatrix is this:
mwPointer mxCreateDoubleMatrix(m, n, ComplexFlag)
mwSize m, n
integer*4 ComplexFlag
So you should change your code to use these exact variable types, and always pass a variable and not a literal constant. So change this:
mwSignedIndex :: num
:
num = int(mxGetScalar(prhs(1)))
plhs(1) = mxCreateDoubleMatrix(num, num, 0)
to something like this:
mwSignedIndex :: num
mwSize m, n
integer*4 ComplexFlag
:
num = int(mxGetScalar(prhs(1)))
m = num
n = num
ComplexFlag = 0
plhs(1) = mxCreateDoubleMatrix(m, n, ComplexFlag)
After making these changes to ensure it is not an argument type mismatch issue, then we can work on the allocate/deallocate issue. You should probably put in a memory allocation failure check in your code for the allocate statement, btw.
