CPLEX C++ API in Matlab core dumped using >1 threads

Question

I wrote a C++ Mex file to call the CPLEX API. It works if I just use one thread, but if I use more than one thread I get "segmentation fault (core dumped)".

But the core dump does not happen for every size of matrix, just if A is 64x1024 and B is 63x2014 (or larger, I did not test it for every size)

Before trying with my Mex file, I stored the problem as .sav and .lp file and used the CPLEX interactive mode and it worked well. So there must be some problem in my code, but I don't get it. The Mex function takes 3 arguments (A,b,G).

Hope that anyone could tell me the mistake in my code. Or give me a hint how to figure out what is going wrong.

The linear system should look like:

min sum zp(i)+zm(i) s.t. A x = b and Gx-zp+zm = 0;

My attempted Mex code is:

#include <iostream>
#include <math.h>
#include <vector>
#include <algorithm>

#include "mex.h"
#include "mat.h"
#include "matrix.h"

#include <ilcplex/ilocplex.h>

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

  IloEnv env;
  IloModel model(env);

  int n = mxGetN(prhs[0]);
  int m = mxGetM(prhs[0]);
  int p = mxGetM(prhs[2]);

  IloNumVarArray x = IloNumVarArray(env,n,0,2,IloNumVar::Int);
  IloNumVarArray zp = IloNumVarArray(env,p,0,IloInfinity,IloNumVar::Float);
  IloNumVarArray zm = IloNumVarArray(env,p,0,IloInfinity,IloNumVar::Float);
  x.setNames("x"); zp.setNames("zp"); zm.setNames("zm");
  model.add(x); model.add(zp); model.add(zm);

  //sum_i zp(i)+zm(i)
  IloObjective obj = IloMinimize(env);
  for(int i=0;i<zp.getSize();i++){ 
    obj.setLinearCoef(zp[i],1);
    obj.setLinearCoef(zm[i],1);
  }
  model.add(obj);

  double *A = mxGetPr(prhs[0]);
  double *b = mxGetPr(prhs[1]);
  double *G = mxGetPr(prhs[2]);
  { //Ax=b
    IloRangeArray cs = IloRangeArray(env);
    for(int i=0;i<m;++i){
      IloRange c(env,b[i],b[i]);
      for( int j=0;j<n;++j){
    c.setLinearCoef(x[j],A[j*m+i]);
      }
      cs.add(c);
    }
    model.add(cs);
  }

  { //Gx-zp+zm=0
    IloRangeArray cs = IloRangeArray(env);
    for(int i=0;i<p;++i){
      IloRange c(env,0,0);
      for( int j=0;j<n;++j){
    c.setLinearCoef(x[j],G[j*p+i]);
      }
      c.setLinearCoef(zp[i],-1);
      c.setLinearCoef(zm[i],1);
      cs.add(c);
    }
    model.add(cs);
  }


  IloCplex cpx(model);
  cpx.exportModel("test.lp");
  cpx.exportModel("test.sav");

  IloCplex::ParameterSet cpxParam = cpx.getParameterSet();

  cpxParam.setParam(IloCplex::MIPDisplay,3);
  cpxParam.setParam(IloCplex::TiLim,5);
  cpxParam.setParam(IloCplex::Threads,13);
  cpxParam.setParam(IloCplex::BoolParam::CloneLog,false);

  cpx.setParameterSet(cpxParam);

  cpx.solve();

  env.end();

}