Printing Sparse Matrix in mex function in CCS format

Hi,

I am relatively new user of mex and matlab.I have to do some operations on sparse matrices,which also includes printing them in CCS format.Here is the code that I have written.

    #include "mex.h"

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

              double *A;
              int *jc,*ir;
              int m,n,size;
              int i;
              A = mxGetPr(prhs[0]);
              m = mxGetM(prhs[0]);
              n = mxGetN(prhs[0]);

              jc = (int*) mxGetJc(prhs[0]);
              ir = (int*) mxGetIr(prhs[0]);

              size = mxGetNzmax(prhs[0]);
              plhs[0] = mxCreateSparse(m,n,size,mxREAL);
              mexPrintf("JC: ");
              for(i = 0; i < m;i++)
              {
               mexPrintf("%d ",jc[i]);
              }
              mexPrintf("\n");
              mexPrintf("IR:");
              for(i = 0;i < size;i++)
              {
               mexPrintf("%d ",ir[i]);
              }
              mexPrintf("\n");
              mexPrintf("Values:");
              for(i = 0;i < size;i++)
              {
               mexPrintf("%g ",A[i]);
              }
              mexPrintf("\n");
      }

But the output that I am getting is like this.

>> a = sprand(4,4,0.3)

a =

   (1,1)       0.4898
   (3,2)       0.6463
   (4,2)       0.7094
   (2,4)       0.4456
   (4,4)       0.7547

>> com(a)

JC: 0 0 1 0

IR:0 0 2 0 3

Values:0.489764 0.646313 0.709365 0.445586 0.754687

ans =

   All zero sparse: 4-by-4

>> full(a)

ans =

    0.4898         0         0         0
         0         0         0    0.4456
         0    0.6463         0         0
         0    0.7094         0    0.7547

Is this correct? since I dont see the row indices getting printed correctly.Further, is there any in built function to print the sparse matrices in CCS format?

Kindly reply.

Cheers.

Answer 1

[James Tursa]

I'm not sure what your real question is. Do you just want to print a sparse matrix from within a mex function? If so, you can use this function:

void spprint(const mxArray *mx)
{
    mwSize n, nrow;
      mwIndex *ir, *jc;
      mwIndex j, x, y;
      double *pr;

  if( !mxIsSparse(mx) ) return;
      n = mxGetN(mx);
      pr = mxGetPr(mx);
    ir = mxGetIr(mx);
    jc = mxGetJc(mx);

      for( y=0; y<n; y++ ) {
          nrow = jc[y+1] - jc[y];
          for( x=0; x<nrow; x++ ) {
        mexPrintf("   (%d,%d)    %g\n",(*ir++)+1,y+1,*pr++);
          }
      }
  }

Incidentally, you should never recast function pointer return values like this:

                jc = (int*) mxGetJc(prhs[0]);
                ir = (int*) mxGetIr(prhs[0]);

This will not work if mwIndex is not an int. Best to just declare jc and ir as mwIndex pointers (not int pointers) and then don't cast the result of the functions and you will be fine.

Answer 2

[Abhishek]

Hi,

Thanks a ton for your reply. However, I just want to print the matrix as its component arrays, that is IR JC and Val.

Is it possible to print them?

Answer 3

[James Tursa]

To print out the raw array contents:

void spprintraw(const mxArray *mx)
{
    mwSize n;
    mwIndex *ir, *jc;
    mwIndex j, N;
    double *pr;

  if( !mxIsSparse(mx) ) return;
    n = mxGetN(mx);
    pr = mxGetPr(mx);
    ir = mxGetIr(mx);
    jc = mxGetJc(mx);
  N = jc[n];

  mexPrintf("JC:\n");
  for( j=0; j<=n; j++ ) {
    mexPrintf("%d ",jc[j]);
  }
  mexPrintf("\n");

  mexPrintf("IR:\n");
  for( j=0; j<N; j++ ) {
    mexPrintf("%d ",ir[j]);
  }
  mexPrintf("\n");

  mexPrintf("VAL:\n");
  for( j=0; j<N; j++ ) {
    mexPrintf("%g ",pr[j]);
  }
  mexPrintf("\n");
}

Keep the following points in mind:

- mxGetNzmax is useless for this. mxGetNzmax returns the amount of memory that is allocated for the sparse arrays stuff, which is not the same as the number of non-zeros presently in the array. Use jc[mxGetN(prhs[0])] to get the actual number of non-zero values held in the array currently.

- All indexing is 0-based, not 1-based.

- The jc array is the size of number_or_columns+1. It is basically an accumulation array containing the number of non-zero elements in total for all of the previous columns (which is why the last value in the array is the total number of non-zero elements in the array currently). It does not contain column indexes.

Answer 4

[Abhishek]

Thanks a lot.I already wrote something similar but your explanation on size of JC and other caveats for indexing have given me some insights of the CCS format. Thanks a ton again.

[James Tursa]

To put it explicitly:

jc[0] = 0

jc[1] = The number of non-zero elements stored for 1st column

jc[2] = The number of non-zero elements stored for 1st-2nd columns

jc[3] = The number of non-zero elements stored for 1st-3rd columns

:

etc.