MEX Crashes, pure C++ works
in [Matlab]
|
From: Fabio Pinna on 11 May 2010 10:04 Hi, I've been working for a while on a set of c++ mex file and I recently discovered a bug. For some particular input data the mex file is causing matlab to crash. I saved the data set and tried to reproduce the error just in c++, which didn't happen. This is fairly strange as the mex wrapper is nothing more than an entry point for matlab and all the computations are done inside the (separete) c++ code. This problem is multiplatform Any idea?
From: James Tursa on 11 May 2010 11:03 "Fabio Pinna" <fabio.pinna(a)vki.ac.be-remove.this> wrote in message <hsbo4l$35c$1(a)fred.mathworks.com>... > Hi, > I've been working for a while on a set of c++ mex file and I recently discovered a bug. > For some particular input data the mex file is causing matlab to crash. > I saved the data set and tried to reproduce the error just in c++, which didn't happen. > > This is fairly strange as the mex wrapper is nothing more than an entry point for matlab and all the computations are done inside the (separete) c++ code. > > This problem is multiplatform > > Any idea? We need to see your code. You are likely using an invalid pointer. James Tursa
From: Fabio Pinna on 14 May 2010 08:19 Thank you for your reply! Well the core routines are pretty nasty to understand. Here I post the wrap #include "kdtree2.hpp" #include "mexFunction.h" #include <math.h> #include "boost/multi_array.hpp" #include "boost/random.hpp" #include "mex.h" void mexFunction(int nlhs, mxArray * plhs[], int nrhs, const mxArray * prhs[]){ kdtree2_result_vector result; // chec number of arguments if( nrhs!=3 ) mexErrMsgTxt("This function requires 3 arguments\n"); if( !mxIsNumeric(prhs[0]) ) mexErrMsgTxt("varargin{0} must be a valid kdtree pointer\n"); if( !mxIsNumeric(prhs[1]) ) mexErrMsgTxt("varargin{1} must be a query point\n"); if( !mxIsNumeric(prhs[2]) ) mexErrMsgTxt("varargin{2} must be a scalar integer\n"); // retrieve the tree pointer kdtree2* tree; //retrieve_tree( prhs[0], tree ); double* pointer0; pointer0 = mxGetPr(prhs[0]); long pointer1 = (long) pointer0[0]; // convert it to "KDTree" tree = (kdtree2*) pointer1; // check that I actually received something if( tree == NULL ) mexErrMsgTxt("vararg{1} must be a valid k-D tree pointer\n"); // retrieve the query point vector<float> query(tree->dim,0); // query is only one point! double* data = mxGetPr(prhs[1]); for(int it=0; it < tree->dim; it++){ query[it] = data[it]; } // retrieve the query cardinality int k=0; //retrieve_k( prhs[2], k ); k = mxGetScalar(prhs[2]); tree->n_nearest(query,k,result); // return the indexes plhs[0] = mxCreateDoubleMatrix(k, 1, mxREAL); double* indexes = mxGetPr(plhs[0]); double* output = mxGetPr(plhs[0]); for (int i=0; i < k; i++){ indexes[ i ] = result[i].idx + 1; } } Anyway if the problem is related to a bad pointer it shouldn't even call correctly the routines in other cases. What really happens is the crash inside the c++ routine (only used with matlab but not with pure c++ main) Thank you for your comments
From: Jan Simon on 14 May 2010 10:09 Dear Fabio! Could you be so kind and mention the line where your C++ function crashes? You could ask your debugger or set some "mexPrintf" or "return" statements to locate your problem. Kind regards, Jan
From: Fabio Pinna on 14 May 2010 10:52
Hi Jan, the line is: tree->n_nearest(query,k,result); So it crashes inside the c++ function not while calling it. I could use a "mexPrintf" inside a try-catch block in c++ but this is not allowed in matlab. That's why I first tried to debug the function in C++ with the same dataset and I discoverd that this way it was working flawlessly. Thank you for your support. Fabio "Jan Simon" <matlab.THIS_YEAR(a)nMINUSsimon.de> wrote in message <hsjli0$cvc$1(a)fred.mathworks.com>... > Dear Fabio! > > Could you be so kind and mention the line where your C++ function crashes? > You could ask your debugger or set some "mexPrintf" or "return" statements to locate your problem. > > Kind regards, Jan |