从C调用返回Fortran函数的数组

Question

I have an already existing Fortran source code that I am working on adding C to, and am having difficulty with passing an array from C into a Fortran function and then receiving another array back.

It compiles without any errors, but when I attempt to run the code, it says that:

"Dimension of array 'p2' (In Fortran Function) has extent 3 instead of -4537421815 (or some other similar number)"

I am not sure what is going on here. I will attach the two codes below.

NOTE : I have removed a lot of the extra variable initialization lines which I think seems unnecessary for finding the problem.

C function:

#include <stdio.h>
#include <stdlib.h>    

extern "C" double* __multiphase_tools_MOD_project(double p1[],double *mydt,int *myi,int *myj,int *myk);

extern "C" void cuda_(int *ptr_band, double *ptr_u, double *ptr_v, double *ptr_w)
{

     double *pt_out;
     double pt_in[3];


     // Loop over the domain and compute fluxes near interfaces
     //=======================================================================================
     //  X FACE
     //=======================================================================================

     for (k = kmin; k <= kmax; k++)
     {
          for (j = jmin; j <= jmax; j++)
          {
               for (i = imin; i <= imax; i++)
               {
                    if (abs(band[i-1][j][k]) <= nband_CFL && abs(band[i][j][k]) <= nband_CFL )
                    {
                         for (int n = 1; n < 10; n++)
                         {
                              pt_in[0] = pt[0][n][1];   pt_in[1] = pt[1][n][1];  pt_in[2] = pt[2][n][1];
                              pt_out = __multiphase_tools_MOD_project(pt_in,&neg_dt_uvw,&i,&j,&k);
                         }
                    }
               }
          }
     }
     return;
}

Fortran Function:

  function project(p1,mydt,myi,myj,myk) result(p2)
    use math
    use iso_c_binding
    implicit none
    real(WP), dimension(3) :: p2
    real(WP), dimension(3), intent(in) :: p1
    real(WP),               intent(in) :: mydt
    integer,                intent(in) :: myi,myj,myk
    real(WP), dimension(3) :: v1,v2,v3,v4
    v1=get_velocity(p1               ,myi,myj,myk)
    v2=get_velocity(p1+0.5_WP*mydt*v1,myi,myj,myk)
    v3=get_velocity(p1+0.5_WP*mydt*v2,myi,myj,myk)
    v4=get_velocity(p1+       mydt*v3,myi,myj,myk)
    p2=p1+mydt/6.0_WP*(v1+2.0_WP*v2+2.0_WP*v3+v4)
    return
  end function project

Answer 1

Fortran functions which return an array actually use a hidden argument and are implemented similarly to subroutines. This is why you can't easily write a compatible C function, because you can't portably determine the hidden argument properties.

For example

function f() result(res)
  real :: res(3)
  res = [1.,2.,3.]
end

looks in the GCC internal code as

f (struct array1_real(kind=4) & __result)
{
}

I suggest to convert the function to a subroutine. That way you can exactly control the argument (parameter) for the array and you can make a C interface:

  subroutine project(p1,p2,mydt,myi,myj,myk) bind(C)
    real(WP), dimension(3), intent(out):: p2
    ...

or you can write a wrapper subroutine which calls your function to maintain the Fortran functionality unchanged.