C Mex crashes if array is too large

Question

0 个投票

Hi all,

I am passing an array to a mex file as follows:

float *var;
var = (float*)mxGetPr(prhs[0]);

The mex file compiles/work perfect when the input is "small" (a few GB for my application), but when the input is larger, it crashes and gives me a segmentation fault. The input is held in memory (loaded into matlab workspace), and furthermore, the mex file shouldn't make a copy of the array (I'm just pointing to the array which is already in memory). Not only that, but I only need to access few elements of the input array at a time.

Any ideas on what could be causing this behavior?

One obvious candidate: I'm not compiling with -largeArrayDims (didn't update my code for 64-bit API). Might this be an issue?

3 个评论
显示 1更早的评论隐藏 1更早的评论

JohnDapper 2016-4-5

Thank you James and Dave,

The problem had to do with using a 32-bit int type index when accessing the array "var" (i.e. x = var(index)). When I cast the index to mwSize (and compile with -largeArrayDims), the problem vanished. I'm assuming this is because the index variable was being assigned a value larger than the maximum allowed with a 32-bit int.

James Tursa 2016-4-5

编辑：James Tursa 2016-4-5

Likely. -largeArrayDims controls whether mwSize is an int or a size_t. If you were using an int for the index and it wrapped around to a negative value then of course you would be pointing to invalid memory hence a crash. Keep in mind, however, that since mwSize is now likely a size_t in your code you need to be careful how you are doing your if-testing since size_t is unsigned. Are you actually "casting" an int to a size_t? I assume you meant typing the variable as a mwSize, not casting.

请先登录，再进行评论。

请先登录，再回答此问题。

Follow Question