Prev: OSIsoft's PI
Next: Solve system with ode15s
From: James Tursa on 3 Aug 2010 15:02
"Balint Takacs" <b.takacs(a)imperial.ac.uk> wrote in message <i39m02$jou$1(a)fred.mathworks.com>...
> Does anybody know whether mxGetCell() allocates memory, that is, its output mxArray* should be released by mxDestroy() after its use?
>
> The documentation says that it is possible that the function fails because 'Insufficient free heap space to hold the returned cell mxArray'. However, it looks a bit weird and inefficient that a deep copy of the cell element is made upon simply accessing it.

mxGetCell returns the pointer to the original mxArray in the cell, it does *not* allocate memory and create a deep copy of the cell contents. You can verify this with the following mex file:

// cellpr.c
// input is a cell array, prints the address of the first cell
// no output
#include "mex.h"
void mexFunction(int nlhs, mxArray *plhs[], int nrhs, const mxArray *prhs[])
{
mxArray **p;
p = mxGetData(prhs[0]);
mexPrintf("Raw 1st cell address = %p\n",*p);
mexPrintf("mxGetCell 1st address = %p\n",mxGetCell(prhs[0],0));
}

>> mex cellpr.c
>> c = {1.23}
c =
[1.2300]
>> cellpr(c)
Raw 1st cell address = 0B15D0A0
mxGetCell 1st address = 0B15D0A0

You can see that the raw mxArray * address contained in the data area of the cell input prhs[0] is exactly the same as the address returned by the mxGetCell call. No copy is made. The warning in the doc about insufficient heap space causing a failure is probably wrong ... I have no idea why this function would be using any heap space. Same comment would hold for mxGetField and mxGetFieldByNumber for structures. The new mxGetProperty, on the other hand, *does* return a deep copy of the property mxArray and thus uses allocated memory from the heap.


James Tursa
 | 
Pages: 1
Prev: OSIsoft's PI
Next: Solve system with ode15s