Thread Subject:

Mex mxCreateDoubleMatrix() destroys different pointer sporadically

I keep running in to segmentation faults when running this Mex file, but it is sporadic and cannot be predicted. The problem comes from the following bit of code:

for (j =0; j<t->neighbors; j++) {
...
ccv_contour_t* cont = t->letters[j]->contour;
mxArray* S = mxCreateDoubleMatrix(cont->size, 1, mxREAL); //THIS IS THE PROBLEM LINE
}

Leaving out the mxArray* line, everything works perfectly, however when adding in the line causes the variable cont to sometimes be NULL for some iterations of the loop. I'd say this behaviour happens around 50% of the times I run the exact same code.

It must be something to do with the memory allocated, but I can't figure out what is wrong (and I have spent hours on this now...)

"Max " <maxjaderberg@gmail.com> wrote in message <kcf29j$l93$1@newscl01ah.mathworks.com>...
> I keep running in to segmentation faults when running this Mex file, but it is sporadic and cannot be predicted. The problem comes from the following bit of code:
>
> for (j =0; j<t->neighbors; j++) {
> ...
> ccv_contour_t* cont = t->letters[j]->contour;
> mxArray* S = mxCreateDoubleMatrix(cont->size, 1, mxREAL); //THIS IS THE PROBLEM LINE
> }
>
> Leaving out the mxArray* line, everything works perfectly, however when adding in the line causes the variable cont to sometimes be NULL for some iterations of the loop. I'd say this behaviour happens around 50% of the times I run the exact same code.
>
> It must be something to do with the memory allocated, but I can't figure out what is wrong (and I have spent hours on this now...)

Unable to provide any meaningful comments unless we see your code. The small snippet above is pretty much useless to us.

James Tursa

Here is the full code https://gist.github.com/4478220

Basically it seems that when allocating a new large mxArray* S (doesnt matter whether it is a matrix or cell) it seems that it interfers with the memory already assigend to other variables. Also the memory given to S sometimes cannot be written too. So there are two symptoms of this problem,
1. the variable ccv_contour_t* cont is set to zero somehow
2. the memory assigned by S = mxCreateDoubleMatrix(cont_n, 1, mxREAL) cannot be written too (but can be read).

I must stress that this problem occurs around 50% of runs, even though there is no randomness in the code.

Here is the full code https://gist.github.com/4478220

Basically it seems that when allocating a new large mxArray* S (doesnt matter whether it is a matrix or cell) it seems that it interfers with the memory already assigend to other variables. Also the memory given to S sometimes cannot be written too. So there are two symptoms of this problem,
1. the variable ccv_contour_t* cont is set to zero somehow
2. the memory assigned by S = mxCreateDoubleMatrix(cont_n, 1, mxREAL) cannot be written too (but can be read).

I must stress that this problem occurs around 50% of runs, even though there is no randomness in the code.

Sorry for the double reply.

I have solved this problem:

I was using a third party library that does not allocate memory using mxMalloc so the memory was not registered with matlab so conflicted with mx memory. To solve the issue I preallocated memory with matlab (mxMalloc) of more than I knew I needed before using the third party library functions, which has seemed to work.