Embedding in Fortran code

Table of Contents

• Introduction
• Fortran Binding Interface

Documentation for linking into a Fortran code base.

Introduction

This section assumes the reader is already familar with developing Fortran codes. The simplist method is to create a C callable function like the C binding exmaple. Then create a Fortran interface for it.

INTERFACE
   SUBROUTINE fortran_callable bind(C, NAME='c_callable_function')
   use, INTRINSIC :: iso_c_binding
   IMPLICIT NONE
   END SUBROUTINE
END INTERFACE

This subroutine can be called like any other Fortran subroutine.

CALL Fortran_Callable

---

Fortran Binding Interface

An alternative is to use the Fortran Language interface. The Fortran binding interface can be enabled as one of the cmake conifgure options. As an example, we will convert the making workflows turorial to use the Fortran language bindings.

SUBROUTINE fortran_binding
USE graph_fortran
use, INTRINSIC :: iso_c_binding
 
IMPLICIT NONE
 
CLASS(graph_context), POINTER :: graph
TYPE(C_PTR)                   :: x
TYPE(C_PTR)                   :: m
TYPE(C_PTR)                   :: b
TYPE(C_PTR)                   :: y
TYPE(C_PTR)                   :: dydx
 
LOGICAL(C_BOOL), PARAMETER :: use_safe_math = .false.
 
graph => graph_double_context(use_safe_math);
 
x = graph%variable(1_c_long, 'x' // c_null_char)
m = graph%constant(0.4_c_double)
b = graph%constant(0.6_c_double)
 
y = graph%add(graph%mul(m, x), b)
dydx = graph%df(y, x);
 
CALL graph%set_variable(x, (/ 1.0_c_double, 2.0_c_double, 3.0_c_double /))
 
CALL graph%set_device_number(0)
CALL graph%add_item((/ graph_ptr(x) /),                                    &
                    (/ graph_ptr(y), graph_ptr(dydx) /))                   &
                    graph_null_array, graph_null_array, c_null_ptr,        &
                    'my_first_kernel' // c_null_char, 3_c_long)
CALL graph%compile
CALL graph%run
CALL graph%print(0, (/ graph_ptr(x), graph_ptr(y), graph_ptr(dydx) /))
CALL graph%print(1, (/ graph_ptr(x), graph_ptr(y), graph_ptr(dydx) /))
CALL graph%print(2, (/ graph_ptr(x), graph_ptr(y), graph_ptr(dydx) /))
 
DEALLOCATE(graph)
END SUBROUTINE

Note

Graphs need to use the graph_fortran::graph_ptr function to get the pointer address of node.

The graph_fortran::graph_null_array allows setting array of nodes arguments to null.