Revision as of 21:17, 12 July 2020 by Siko1056 (talk | contribs) (Add description about the data file.)
(diff) ← Older revision | Latest revision (diff) | Newer revision → (diff)

This page describes an example of how to call liboctave functions from a Fortran program. In the example we will load a single matrix, stored in ASCII format, from a data file. It consists of two steps:

  1. write a C++ function with a C compatible interface and C linkage that reads a variable from an Octave ASCII file
  2. write Fortran code using the "iso_c_binding" intrinsic module to call the C++ function

Data fileEdit

File: data.txt
 1.59797350e-01 5.41307474e-01 1.12127655e-01 2.09249248e-01
 3.22564589e-01 7.94307947e-01 8.25924316e-01 5.38352076e-01
 3.63990736e-01 1.90371212e-02 2.89370865e-01 1.30131246e-01
 6.28360462e-01 1.98831330e-01 6.89539723e-01 6.91062597e-01

The file was generated with

A = rand (4);
save -ascii data.txt A

C++ functionEdit

C++ function to load a single matrix, stored in ASCII format, from a data file.

// Octave header
#include <octave/oct.h>
#include <octave/ls-mat-ascii.h>

// Custom header containing the C compatible interface
#include <octave_file_io.h>

//! Load a single matrix, stored in ASCII format, from a data file.
//! @param file_name name of the data file.
//! @param data pointer to the read-in matrix stored as fortran vector
//!             (column-major order).
//! @param numel number of elements in @p data.

int octave_load (const char* file_name, double** data, int* numel)
  // Define variable to hold the read data.
  octave_value read_data;

  // Read a plain ASCII matrix from data file.
  std::ifstream in_file_stream (file_name, std::ios::binary);
  read_mat_ascii_data (in_file_stream, file_name, read_data);
  in_file_stream.close ();

  // Convert read data to numerical array (matrix).
  NDArray A = read_data.array_value ();

  // Extract number of elements in matrix A.
  *numel = A.numel ();

  // Allocate memory to pointer to returned values.
  *data  = (double*) malloc (A.numel () * sizeof (double));

  // Copy the content of matrix A to data structure Fortran can handle.
  memcpy (*data, A.fortran_vec (), A.numel () * sizeof (double));

  return 0;

Header fileEdit

Header file with C interface to

File: octave_file_io.h

#ifdef __cplusplus
extern "C" {
  int octave_load (const char* filename, double** data, int* numel);

#ifdef __cplusplus


Fortran CodeEdit

Fortran main program to read the plain ASCII matrix with the help of the Octave-C++ function. The read in matrix is printed to the screen.

File: octave_file_io_example.f90
program octave_file_io_example

  use iso_c_binding

  implicit none

     function octave_load (filename, data, numel) bind(c, name="octave_load")

       use iso_c_binding
       implicit none

       integer(c_int) :: octave_load

       character(kind=c_char), intent(in) :: filename(*)

       type(c_ptr), intent(out) :: data
       integer(c_int), intent(out) :: numel

     end function octave_load
  end interface

  integer(c_int) :: res
  type(c_ptr) :: data
  real(c_double), pointer :: fdata(:)
  integer(c_int) :: numel

  res = octave_load (c_char_'data.txt' // c_null_char, data, numel)

  call c_f_pointer (data, fdata, (/numel/))

  write (*,*) fdata

end program octave_file_io_example

Compiling the codeEdit

Generate octave_file_io.o from

mkoctfile -I. -c

Generate octave_file_io_example.exe from octave_file_io_example.f90 including octave_file_io.o.

mkoctfile -I. --link-stand-alone octave_file_io_example.f90 octave_file_io.o -o octave_file_io_example.exe -lgfortran

If you receive errors about missing libraries, make sure your LD_LIBRARY_PATH is set correctly to find all Octave libraries, e.g.

$ ldd ./octave_file_io_example.exe
... => /usr/lib64/ (0x00007fe9eb62b000) => not found => not found

Then find and on your system and type

export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/path/to/lib/octave/6.4.0/