MEX solution based on Amro's, but manually doing an "atoi", also with OpenMP

solution_mex_chappjc_vc12.cpp

// solution_mex_chappjc_vc12.cpp
//  mex -v -largeArrayDims COMPFLAGS="$COMPFLAGS /openmp" ...
//      LINKFLAGS="$LINKFLAGS /openmp" solution_mex_chappjc_vc12.cpp ...
//      -output solution_mex_chappjc_vc12_omp.mexw64

#include "mex.h"
#include <string>
#include <sstream>
#include <algorithm>

extern "C" {
    mxArray *mxCreateSharedDataCopy(const mxArray *pr);
    mxArray *mxCreateUninitNumericMatrix(mwSize m, mwSize n, mxClassID classid, mxComplexity ComplexFlag);
    mxArray *mxFastZeros(mxComplexity ComplexFlag, mwSize m, mwSize n);
}

namespace {

// get i-th string from cell-array of strings
std::string getString(const mxArray *cellstr, const mwIndex idx)
{
    const mxArray *arr = mxGetCell(cellstr, idx);
    if (arr == NULL) mexErrMsgIdAndTxt("mex:err", "null/uninitialized");
    if (!mxIsChar(arr)) mexErrMsgIdAndTxt("mex:err", "not a string");
    char *cstr = mxArrayToString(arr);
    if (cstr == NULL) mexErrMsgIdAndTxt("mex:err", "null");
    std::string str(cstr);
    mxFree(cstr);
    return str;
}

// count of numbers in char-delimited string
mwSize count_numbers(const std::string &s)
{
    return std::count(s.begin(), s.end(), '_') + 1;
}

};

void mexFunction(int nlhs, mxArray *plhs[], int nrhs, const mxArray *prhs[])
{
    // validate inputs
    if (nrhs!=1 || nlhs>1) mexErrMsgIdAndTxt("mex:err", "wrong num args");
    if (!mxIsCell(prhs[0])) mexErrMsgIdAndTxt("mex:err", "not a cell");

    // determine sizes
    const mxArray *cellstr = prhs[0];
    const mwSize n_words = mxGetNumberOfElements(cellstr);
    const mwSize n_nums = (n_words > 0) ? count_numbers(getString(cellstr,0)) : 0;

    // allocate output numeric matrix
    plhs[0] = mxCreateDoubleMatrix(n_nums, n_words, mxREAL);
    mxArray *mxUnmanaged = mxFastZeros(mxREAL,n_nums, n_words);
    plhs[0] = mxCreateSharedDataCopy(mxUnmanaged);
    mxDestroyArray(mxUnmanaged);
    //plhs[0] = mxCreateUninitNumericMatrix(n_nums, n_words, mxDOUBLE_CLASS, mxREAL);
    if (plhs[0] == NULL) mexErrMsgIdAndTxt("mex:err", "null");
    if (n_words == 0 || n_nums == 0) return;
    double *out = mxGetPr(plhs[0]);

    // extract numbers from strings
    #pragma omp parallel for num_threads(4)
    for (mwSignedIndex idx=0; idx<n_words; ++idx) {
        mwIndex i = idx * n_nums;
        std::string str = getString(cellstr, idx);
        
        std::string::const_iterator istr = str.cbegin();
        for (; istr != str.cend(); ++istr) {
            if (*istr < '0' || *istr > '9')
            {
                ++i;
                continue;
            }
                out[i] *= 10;
                out[i] += *istr - '0';
        }
        ++i;
    }
}