Mex Coding I

Using Matlab for quite a long time, now Matlab 2014b has already equipped with various programming language support, such as C#, Python 2.7. However, it does not mean an easy job to stir fry them together.

For scientific programming, we consider the most common ones: C++, Python, Java(Scala) and Julia.

The lastest version of Matlab has already provided generic interfaces with C++, Python, Java. However, Python’s support is limited to version 2.7 without some important types, which might be a huge problem for migrating codes from other platforms.

Limitations for Python support:

• Multidimensional arrays
• Structure arrays
• Complex, scalar integers or arrays
• Sparse arrays
• Logical vectors
• Categorical, table, containers.Map, datatime types
• Matlab objects
• Function handle
• meta.class

For Java, reference:

Since Matlab itself runs inside a JVM, it would provide most generic support for Java, with all access of Java API and can call Java object methods within Matlab, the most important one, passing data would be easy.

In return of easy integration, Java method called from Matlab could have very high overhead before each call, much higher than other approaches.

Another thing needs to be mentioned is Object Oriented Programming within Matlab, it is slow compared to other mainstream languages, and the overhead cannot be avoid.

For C/C++, Matlab 2014b has provided two ways for interacting.

• loadlibrary inside Matlab console, then calllib calls function from the library.
• use MEX as bridge.

C-MEX interface as

void mexFunction(int nlhs, mxArray* plhs[], int nrhs, const mxArray* prhs[])

While C++ users may create C++ objects and reuse them later by implementing some wrapper. In that thread, there are tons of potential dangers with memory leaks. The best way found is to define a class inherits the handle class using classdef and also implement constructor and destructor within the class.

A reasonable implementation can be found as mexplus.

However, when object is created and wants running its method with parfor, Matlab needs to copy the object for each worker in the pool. The solution shows that marking the handle property as Transient will work and force calling loadobj.

classdef myclass < handle
    properties(SetAccess=protected, Transient=true)
        cpp_handle
    end
    methods(Static=true)
        function obj = loadobj(a)
            a.cpp_handle = rand(1);
            disp(['Load initialized encoder with handle: ' num2str(a.cpp_handle)]);
            obj = a;
        end
    end
    methods
        function obj = myclass(val)
            obj.cpp_handle = rand(1);
            disp(['Initialized myclass with handle: ' num2str(obj.cpp_handle)]);
        end
        % destructor
        function delete(obj)
            disp(['Deleted myclass with handle: ' num2str(obj.cpp_handle)]);
        end
        % class method
        function amethod(obj)
            disp(['Class method called with handle: ' num2str(obj.cpp_handle)]);
        end
    end
end

Tested below :

cls = myclass();
parfor i = 1:10
    cls.amethod()
end

Result :

Load initialized encoder with handle: 0.096779
Class method called with handle: 0.096779
Class method called with handle: 0.096779
Class method called with handle: 0.096779
Class method called with handle: 0.096779
Load initialized encoder with handle: 0.66182
Class method called with handle: 0.66182
Class method called with handle: 0.66182
Class method called with handle: 0.66182
Class method called with handle: 0.66182
Class method called with handle: 0.096779
Class method called with handle: 0.096779
Deleted myclass with handle: 0.096779
Deleted myclass with handle: 0.66182