mex sometimes work, sometimes crash

In the code I try to calculate the norm for each column of a complex matrix. When I test it in MATLAB 2012b with Mat = rand(500,1000)+1i*rand(500,1000); b = matrixnorm(Mat); I can get the correct answer. But when I use it in my other program, MATLAB is killed every time. Could you give me a hint? below is my code. Thanks in advance.

MATLAB code
#include "mex.h"
#include<omp.h>
#include <math.h>
#include <string.h>
#include "complex"
void mexFunction(int nlhs, mxArray *plhs[], int nrhs, const mxArray *prhs[]) 
{ 
  int coreNum = omp_get_num_procs();
  int threadNum=coreNum*4;//coreNum;   
  omp_set_num_threads(threadNum);
  
  if (nrhs != 1)
    mexErrMsgTxt("Wrong number of inputs"); 
  if (nlhs != 1)
    mexErrMsgTxt("Wrong number of outputs");

        const mxArray *mxA = prhs[0];    
  if (mxGetNumberOfDimensions(mxA) != 2)
    mexErrMsgTxt("Invalid input: A");    
  const mwSize *A_dims = mxGetDimensions(mxA);
  mwSize Out_dims[2];
  Out_dims[0] = A_dims[1];
  Out_dims[1] = 1;
  plhs[0] = mxCreateNumericArray(1,(const mwSize *)Out_dims,mxDOUBLE_CLASS,mxREAL);  
  double *out = (double *)mxGetPr(plhs[0]);         
  double *data_imag, *data_real;
  data_imag = mxGetPi(prhs[0]);
  data_real = mxGetPr(prhs[0]);
  
  int idx; double sum;
  #pragma omp parallel for
  for(int i = 0;i<A_dims[1];i++)
  {
    sum = 0;
    for(int j = 0;j<A_dims[0];j++)
    {
       idx = j + i*A_dims[0];
             sum = sum + data_imag[idx]*data_imag[idx]+data_real[idx]*data_real[idx];
    }
      out[i] = sum;    
  }  
}