jonathf/matlab2cpp

Matlab2Cpp

matlab2cpp is a semi-automatic tool for converting code from Matlab to C++.

After installing, the matlab2cpp command line executable m2cpp will be available in path that can be used to convert Matlab code.

Note that it is not meant as a complete tool for creating runnable C++ code. For example, the eval-function can not be supported because there is no general way to implement it in C++. Instead the program is a support tool, which aims at speed up the conversion process as much as possible for a user that needs to convert Matlab programs by hand anyway. The software does this by converting the basic structures of the Matlab-program (functions, branches, loops, etc.), adds variable declarations, and for some simple code, do a complete translation. And any problem the program encounters during conversion will be written in a log-file. From there manual conversions can be done by hand.

Currently, the code will not convert the large library collection of functions that Matlab currently possesses. However, there is no reason for the code not to support these features in time. The extension library is easy to extend.

Installation

Installation by running the pip command:

pip install matlab2cpp

The source-to-source parser do not have any requirements beyond having Python installed. However, the generated output does have a few requirements to be compilable. They are as follows.

C++11

Code produces follows the C++11 standard.

armadillo

Armadillo is a linear algebra library for the C++ language. The Armadillo library can be found at `http://arma.sourceforge.net`_. Some functionality in Armadillo rely on a math library like LAPACK, BLAS, OpenBLAS or MKL. When installing Armadillo, it will look for installed math libraries.

If Armadillo is installed, the library can be linked with the link flag -l armadillo. Armadillo can also be linked directly, see the FAQ at the Armadillo webpage for more information.

I believe MKL is the fastest math library and it can be downloaded for free at `https://software.intel.com/en-us/articles/free-mkl`_.

TBB

By inserting pragmas in the code, for loops can be marked by the user. The program can then either insert OpenMP or TBB code to parallelize the for loop. To compile TBB code, the TBB library has to be installed. See :ref:`parallel_flags` for more details.

An illustrating Example

Assuming Linux installation and m2cpp is available in path. Code works analogous in Mac and Windows.

Consider a file example.m with the following content:

function y=f(x)
    y = x+4
end
function g()
    x = [1,2,3]
    f(x)
end

Run conversion on the file:

$ m2cpp example.m

This will create two files: example.m.hpp and example.m.py.

In example.m.hpp, the translated C++ code is placed. It looks as follows:

#include <armadillo>
using namespace arma ;

TYPE f(TYPE x)
{
  TYPE y ;
  y = x+4 ;
  return y ;
}

void g()
{
  TYPE x ;
  x = [1, 2, 3] ;
  f(x) ;
}

Matlab doesn't declare variables explicitly, so m2cpp is unable to complete the translation. To create a full conversion, the variables must be declared. Declarations can be done in the file example.m.py. After the first run, it will look as follows:

# Supplement file
#
# Valid inputs:
#
# uint    int     float   double cx_double
# uvec    ivec    fvec    vec    cx_vec
# urowvec irowvec frowvec rowvec cx_rowvec
# umat    imat    fmat    mat    cx_mat
# ucube   icube   fcube   cube   cx_cube
#
# char    string  struct  structs func_lambda

functions = {
  "f" : {
    "y" : "",
    "x" : "",
  },
  "g" : {
    "x" : "",
  },
}
includes = [
  '#include <armadillo>',
  'using namespace arma ;',
]

In addition to defining includes at the bottom, it is possible to declare variables manually by inserting type names into the respective empty strings. However, some times it is possible to guess some of the variable types from context. To let the software try to guess variable types, run conversion with the -s flag:

$ m2cpp example.m -s

The file example.m.py will then automatically be populated with data types from context:

# ...

functions = {
  "f" : {
    "y" : "irowvec",
    "x" : "irowvec",
  },
  "g" : {
    "x" : "irowvec",
  },
}
includes = [
  '#include <armadillo>',
  'using namespace arma ;',
]

It will not always be successful and some of the types might in some cases be wrong. It is therefore also possible to adjust these values manually at any time.

Having run the conversion with the variables converted, creates a new output for example.m.hpp:

#include <armadillo>
using namespace arma ;

irowvec f(irowvec x)
{
  irowvec y ;
  y = x+4 ;
  return y ;
}

void g()
{
  irowvec x ;
  int _x [] = [1, 2, 3] ;
  x = irowvec(_x, 3, false) ;
  f(x) ;
}

This is valid and runnable C++ code. For such a small example, no manual adjustments were necessary.