pass a vector from matlab to a c++ program

function res = fun(p,y)
s = call_a_cpp_program_which_operates_on_p
res = s - y;
end

In the above function, 'p' is a vector and 's' as well. I have to pass the vector 'p' to a c++ program to compute 's'. The above function is called several times.

My idea is to first write the vector 'p' in a file, which the c++ program reads in. The c++ program writes its output 's' also in a file, which MATLAB can read in.

Probably this would work, but is there a better approach to transfer the vector 'p' from MATLAB to my c++ program?

[Torsten]

Just out of interest: Which PDE solver do you use as the c++ program in the background ? Some commercial tool like ANSYS ?

[SA-W]

I use an open source library called dealii. Basically, my program is a class. Creating an instance of that class in the main function will call several member functions which eventually compute 's'. I want to store 'p' as a member variable of that class.

[Torsten]

Sounds too complicated for a coupling by direct data transfer. I think it will be a challenge to coordinate data transfer even over written files. One solver makes calculations - the other has to check when the input data file changes ...

But maybe James Tursa will place a comment.

[SA-W]

My idea was to work with simply two files. Whenever the function 'res' is called, the file storing 'p' is overwritten with the 'p' being passed to 'res' and the file storing 's' as well.

Or is it necessary for lsqnonlin to store the 'p's and 's's associated with previous iterations?

[Torsten]

Or is it necessary for lsqnonlin to store the 'p's and 's's associated with previous iterations?

No, but you must keep both solvers in a "wait" state until they get new data from the other solver. Do you already have an idea how to handle this ?

[SA-W]

I do not have deep knowledge about communication between processes.

But is it not enough to keep only MATLAB in a waiting state while the c++ program is running? Because the c++ program is started by MATLAB (probably using the system command).

[Torsten]

Sorry, but I don't know.

[SA-W]

I guess the system command is exactly what I need as for the communication between the processes. In the documentation, it says that MATLAB waits for the executed command to be finished, i.e., until my pde solver exits.

But maybe James wants to add another comment below.

[Torsten]

In such cases I always say:

The proof of the pudding is in the eating (Probieren geht über Studieren).

Answer 1

[James Tursa]

This really depends on what your C++ program does, but the simplest approach is to use a mex routine. You will need a supported C++ compiler installed. An example of a bare bones (i.e., no argument checking) C++ mex routine is as follows:

// File mymex.cpp
// Y = mymex(X)
// where X = a full double variable
// Y = 2.0 * X
#include "mex.h"
void mexFunction(int nlhs, mxArray *plhs[], int nrhs, const mxArray *prhs[])
{
    mwSize i, n;
    double *p;
    plhs[0] = mxDuplicateArray(prhs[0]); // Create the output variable as a deep copy of input
    n = mxGetNumberOfElements(prhs[0]); // Number of elements of input variable
    p = (double *) mxGetData(prhs[0]); // Pointer to output variable data
    for( i=0; i<n; i++ ) {
        p[i] *= 2.0; // Multiply the output elements by 2.0
    }
}

Save this file under the name mymex.cpp, then compile it with the mex command:

    mex mymex.cpp

This routine simply returns the input variable multiplied by 2. E.g.,

    x = 1:5;
    y = mymex(x)

Another approach is to have your C++ code compiled as a library, and then use the loadlibrary( ) and calllib( ) functionality to call the C++ routine.

[SA-W]

Thanks for your answer James. I think I can make it with your suggestions to pass the vector.

Given the discussion with Torsten in the comments above, can you help regarding the communication between MATLAB and my c++ program?