Why does my C MEX-file, with large locally-scoped variables, generate a segmentation fault in MATLAB?

I have created a C MEX-file that uses large locally-scoped arrays or structures, or uses a recursive algorithm. When I try to run my MEX-file from the MATLAB command line, MATLAB immediately terminates.

A run-time error may be generated before MATLAB crashes, depending both on the current platform and the compiler used to generate the MEX-file.

On a Windows platform using the lcc compiler that is distributed with MATLAB, the following message is generated before MATLAB crashes:

 Runtime error!

 Program D:\Applications\MATLAB704\bin\win32\MATLAB.exe

 The application has requested the runtime to terminate it in an unusual way.  Please contact the application's support team for more information.

On a Windows platform using the Microsoft Visual C++ compiler, the following message is generated and MATLAB does not crash:

 ??? Unknown Windows exception value=c00000fd in MATLAB.

On a Red Hat Linux platform using the gcc compiler, the following message is generated before MATLAB terminates:

 Segmentation fault (core dumped)

Answer 1

[MathWorks Support Team]

This situation is commonly referred to as "stack overflow."

The memory for a variable declared local to a function is automatically allocated (i.e., at run time) by the compiler on the program's call stack. Operating systems impose a maximum size for the call stack, which will vary depending on the operating system. When a program's stack size exceeds this limit, the program will terminate.

The most common causes of stack overflow are infinite or very deep recursion, and large local objects. If a MEX-file function utilizes large data structures or arrays, these should be dynamically allocated in MATLAB's free store (also known as the "heap") using the mxMalloc/mxCalloc functions. Memory allocated dynamically in this way must then be deallocated using the mxFree function.

For more information on these functions, see the C MX-Functions reference page, which can be viewed by copying the following to the MATLAB command prompt:

web([docroot,'/techdoc/apiref/mx-c.html'])