How to use mxCreateNumericArray

I am building a mex function, but am having difficulty creating a vector of integers. My code throws about 10 errors, all of which around the declaration of 'gn' and plhs[0]. I'm just trying to make a xynum long vector of class int16. Why is it so difficult?

void findnegs(double *phi, int gnum, int xynum, int *gn, double *adj, double *newphi, double *dphidtSum)
{
    long   i,g;
    double u,s;




    for (i = 0;  i<xynum; i++)
    {
        s = 0;
        for (g = 0;  g<gnum;  g++)
        {
            u = dphidtSum[g+i*gnum]/phi[g+i*gnum];
            if (newphi[g+i*gnum]<0 & u<s)
            {
                s      =  u;
                gn[i]  =  g;
                adj[i] = -1/u;
            }
        }
    }
}

/* The gateway function that replaces the "main".
*plhs[] - the array of output values
*prhs[] - the array of input values
*/
void mexFunction( int nlhs, mxArray *plhs[], int nrhs, const mxArray *prhs[])
{
    /* DECLARATIONS*/
    double *phi, *newphi, *dphidtSum, *adj;
    int gnum, xynum, *gn;

    /* INPUTS */
    gnum      = mxGetScalar(prhs[0]);
    xynum     = mxGetScalar(prhs[1]);
    phi       = mxGetPr(prhs[2]);
    newphi    = mxGetPr(prhs[3]);
    dphidtSum = mxGetPr(prhs[4]);

    /* OUTPUTS */
    plhs[0]    = mxCreateNumericArray(xynum,1,mxINT16_CLASS); /*Creates a matrix*/
    plhs[1]    = mxCreateDoubleMatrix(xynum,1,mxREAL);        /*Creates a matrix*/
    gn         = mxGetInt16s(plhs[0]);                        /*Set to be the first of the output values*/
    adj        = mxGetPr(plhs[1]);                            /*Set to be the first of the output values*/

    /* CALL ROUTINE */
    findnegs(phi,gnum,xynum,gn,adj,newphi,dphidtSum);
}

[Jan]

How do you compile the code? Which API version are you using? mxGetInt16s belongs to the R2018a API, but and mxGetPr to R12017b. Decide for one of them.

Which error message do you get? Please share them with the readers. This is more efficient thabn letting us guess, what you see.

Answer 1

[James Tursa]

This line is incorrect (wrong function and signature):

plhs[0]    = mxCreateNumericArray(xynum,1,mxINT16_CLASS); /*Creates a matrix*/

It should be this instead:

plhs[0]    = mxCreateNumericMatrix(xynum,1,mxINT16_CLASS,mxREAL); /*Creates a matrix*/

Also, these lines will not work:

    int gnum, xynum, *gn;
        :
    gn         = mxGetInt16s(plhs[0]);                        /*Set to be the first of the output values*/

An int on your system is almost surely a 32-bit integer, not a 16-bit integer. You need to make sure your pointer definition matches exactly what is in memory. E.g., you can probably use a short int for this:

    int gnum, xynum;
    short *gn;
        :
    gn         = mxGetInt16s(plhs[0]);                        /*Set to be the first of the output values*/

That will mean you need to change this signature:

    void findnegs(double *phi, int gnum, int xynum, int *gn, double *adj, double *newphi, double *dphidtSum)

to this instead also:

    void findnegs(double *phi, int gnum, int xynum, short *gn, double *adj, double *newphi, double *dphidtSum)

But this begs the question, are integer calculations in the findnegs function now going to overflow your 16-bit integers in gn? You may need to adjust your code to make sure this isn't a problem.

[Christopher Grose]

Integer overflows will only occur if I use large values of gnum, which I will not.

Thanks for your help. Only the following error remains:

Error using mex
C:\Users\Chris\AppData\Local\Temp\mex_342081054824847_2340\findnegsC.obj:findnegsC.c:(.text+0x11c): undefined reference to `mxGetInt16s'
collect2.exe: error: ld returned 1 exit status

I changed the data type declarations for *gn to short as you suggested. I also changed declarations for g and gnum to short for consistency, since these values are often compared. But maybe that will result in problems when I calculate "g+i*gnum" since i is a long?

[James Tursa]

If you are using R2017b or earlier, or using R2018a or later with the -R2017b option (the default), change this

gn         = mxGetInt16s(plhs[0]);                        /*Set to be the first of the output values*/

to this

gn         = (short *) mxGetData(plhs[0]);                        /*Set to be the first of the output values*/

If you are usint R2018a or later and want to use the new function mxGetInt16s, you will have to include the -R2018a option on your mex command. E.g.

mex findnegsC.c -R2018a

[James Tursa]

" I also changed declarations for g and gnum to short for consistency, since these values are often compared"

Not necessary and in fact wasteful. In C/C++, the language specifies that any integer type shorter than an int will be converted to an int for operations and argument passing ("usual unary and binary conversions"). E.g., with these definitions

short s;
int i;

this expression

s+i

will be compiled as

(int)s + i

And if they are both shorts

short s, i;

then the expression

s+i

will be compiled as

(int)s + (int)i

In general, I would advise avoiding shorter integer types than int unless you have a specific need to use them (like you have very large arrays of integers that fit within the range of a shorter type).

[Christopher Grose]

Thanks for that explanation. 16 bit integers are still useful in this case but that will change some other plans.

Perhaps a final question. The mex function is crashing when called in matlab. I've actually commented out everything in the findnegs function so that nothing is executed but it still fails. It looks to fail around the line

plhs[0]    = mxCreateNumericArray(xynum,1,mxINT16_CLASS,mxREAL); /*Creates a matrix*/

since report shows

Stack Trace (from fault):
[  0] 0x0000000010898add                            bin\win64\pgo\libmx.dll+00035549 matrix::detail::noninlined::mx_array_api::mxCreateNumericArray+00000141

and perhaps it is associated with these warnings during compiling:

Building with 'MinGW64 Compiler (C)'.
C:\Users\Chris\Desktop\New Phase Field\findnegsC.c: In function 'mexFunction':
C:\Users\Chris\Desktop\New Phase Field\findnegsC.c:55:45: warning: passing argument 2 of 'mxCreateNumericArray_730' makes pointer from integer without a cast [-Wint-conversion]
     plhs[0]    = mxCreateNumericArray(xynum,1,mxINT16_CLASS,mxREAL); /*Creates a matrix*/
                                             ^
In file included from C:\Program Files\MATLAB\R2018b/extern/include/mex.h:43:0,
                 from C:\Users\Chris\Desktop\New Phase Field\findnegsC.c:7:
C:\Program Files\MATLAB\R2018b/extern/include/matrix.h:275:30: note: expected 'const mwSize * {aka const long long unsigned int *}' but argument is of type 'int'
 #define mxCreateNumericArray mxCreateNumericArray_730
                              ^
C:\Program Files\MATLAB\R2018b/extern/include/matrix.h:1247:1: note: in expansion of macro 'mxCreateNumericArray'
 mxCreateNumericArray(mwSize ndim, const mwSize *dims, mxClassID classid, mxComplexity flag);
 ^~~~~~~~~~~~~~~~~~~~

MEX completed successfully.

how can this be corrected?

[Christopher Grose]

I was looking at

edit([matlabroot '/extern/examples/refbook/matrixDivide.c']);

and noticed that the syntax for mxCreateNumericArray is not the same as mxCreateDoubleMatrix. I changed the code to:

void mexFunction( int nlhs, mxArray *plhs[], int nrhs, const mxArray *prhs[])
{
    /* DECLARATIONS*/
    double *phi, *newphi, *dphidtSum, *adj;
    int     xynum;
    short   gnum, *gn;
    mwSignedIndex dims[2];

    /* INPUTS */
    gnum       = mxGetScalar(prhs[0]);
    xynum      = mxGetScalar(prhs[1]);
    phi        = mxGetPr(prhs[2]);
    newphi     = mxGetPr(prhs[3]);
    dphidtSum  = mxGetPr(prhs[4]);

    /* OUTPUTS */
    dims[0]    = xynum;
    dims[1]    = 1;
    plhs[0]    = mxCreateNumericArray(2,dims,mxINT16_CLASS,mxREAL); /*Creates a matrix*/
    plhs[1]    = mxCreateDoubleMatrix(xynum,1,mxREAL); /*Creates a matrix*/
//  gn         = mxGetInt16s(plhs[0]);                    /*Set to be the first of the output values*/
    gn         = (short *) mxGetData(plhs[0]);                    /*Set to be the first of the output values*/
    adj        = mxGetPr(plhs[1]);                        /*Set to be the first of the output values*/

    /* CALL ROUTINE */
    findnegs(phi,gnum,xynum,gn,adj,newphi,dphidtSum);
}

which works.

But is there a way to do this without declaring and assigning each level of dims? Can I just put the values in the mxCreateNumericArray line to simplify/save some overhead?

[James Tursa]

You didn't read my initial Answer closely enough. I told you to replace your mxCreateNumericArray call with a mxCreateNumericMatrix call for exactly the reason you specify ... it avoids the dims stuff if you only want a 2D result.

The reason you were still crashing is because you had the wrong arguments for mxCreateNumericArray.