Why does is take so long to add elements to an mwArray?

The following code will illustrate this behavior:

#include "matlab.hpp"

#include <stdlib.h>     /* used for EXIT_SUCCESS */

#ifdef GCC

#ifndef EXIT_SUCCESS

#define EXIT_SUCCESS 0

#endif

#endif

int main(void)

{

#ifdef EXCEPTIONS_WORK

    try

#endif

    {

        mwArray a, b, c, x, y, z;

       // x = rand(4,4);

        //y = magic(7);

        //z = eig(x);

for( int i=1; i<30000;i++)

{

x(i)=i;

y(i)=i;

z(i)=i;

//if (i%10000==0) cout<<i<<endl;

}

         // Save (and name) the variables

        save("ex5.mat", "x", x, "y", y, "z", z);

        // Load the named variables

        load("ex5.mat", "x", &a, "y", &b, "z", &c);

        // Check to be sure the variables are equal

        if (tobool(a == x) && tobool(b == y) && tobool(c == z))

        {

            cout << "Success: all variables equal" << endl;

        }

        else 

        {

            cout << "Failure: loaded values not equal to saved values" << endl;

        }

    }

#ifdef EXCEPTIONS_WORK

    catch (mwException &ex)

    {

        cout << ex;

    }

#endif

    return(EXIT_SUCCESS);

}

Testing this example for different values of i yielded the following results. For 300 elements the execution time was .32 seconds; for 3,000 elements it was 2.173 seconds; and for 30,000 elements it was 388 seconds. The execution time seems to be growing exponentially.

Answer 1

[MathWorks Support Team]

The problem is that when you use indexing to expand an array, the library has to reallocate the array before adding the new element. This reallocation takes more time as the array gets larger.

The solution is to preallocate the array before assigning into it. In the previous example, if you add the lines

x = zeros(30000);

y = zeros(30000);

z = zeros(30000);

before the loop, you should see much better performance.

This also applies to vector augmenting in MATLAB; it is not specific to the C/C++ Math Library.