Thread: Pointer to 2D array

You're making it slightly too difficult. When you're typedefing like that, pretend like you don't know you have an array:

Code:

typedef float MATRIX4D[4][4];
MATRIX4D oMatrix;
MATRIX4D *poMatrix = &oMatrix;

The syntax is like any simple pointer assignment.

The reason you were getting the error is that poMatrix is literally a pointer to an array of array, whereas &oMatrix[0][0] is a pointer to a float (because oMatrix[0][0] is a float). They point to the same place (in so far as you can say two different types of pointer can point to the same place), but they are different types and thus are not compatible.

Generally speaking, you don't need an explicit pointer to an array like that. Arrays decay into pointers anyway, so if you're trying to simulate a pass by reference, arrays magically do that already.

Save yourself the headaches that is passing actual arrays around, and just use pointers.

Code:

void makematrix( float ***m, size_t rows, size_t columns )
{
    if( m )
    {
        allocate rows
        for each row
            allocate columns
    }
}

...

float **matrix;

...

makematrix( &matrix, 10, 20 );

I find passing actual arrays far more of a headache than it's worth.


Quzah.

Originally Posted by quzah

Save yourself the headaches that is passing actual arrays around, and just use pointers.

Code:

void makematrix( float ***m, size_t rows, size_t columns )
{
    if( m )
    {
        allocate rows
        for each row
            allocate columns
    }
}

...

float **matrix;

...

makematrix( &matrix, 10, 20 );

I find passing actual arrays far more of a headache than it's worth.


Quzah.

Yes, but that causes a whole load of extra overhead:
In terms of data storage since you now need to store 5 pointers to store 16 floats. So you spend 24 bytes [assuming pointers are 32-bit, on a 64-bit machine, it would be 48 bytes] storing 64 bytes of data. That assumes NO overhead on the allocations - which there likely is - each pointer block is usually rounded to some larger address.

On the memory access front, each memory read of the matrix content would incur at least one extra memory read, probably two, where the fixed size array is a single direct read of the data - only a simple multiplication by 4 (or 16) and addition to get the address of each element.

And due to the nature of pointers and compilers inability to understand aliasing, it is not unlikely that the compiler produces less good code from it simply because it doesn't know if the code does something like setting all rows to point to the same data.

--
Mats

Not really. But if you're worried about it, just use a single malloc call, and use one less layer of indirection.

Code:

void makematrix( float **m, size_t rows, size_t cols )
{
    *m = malloc( sizeof **m * rows * cols );
}

Array work is all done by multiplication anyway.


Quzah.

Originally Posted by quzah

Not really. But if you're worried about it, just use a single malloc call, and use one less layer of indirection.

Code:

void makematrix( float **m, size_t rows, size_t cols )
{
    *m = malloc( sizeof **m * rows * cols );
}

Array work is all done by multiplication anyway.


Quzah.

You still get one extra memory access, and now you can't use [row][col] syntax.

--
Mats

How do you figure you're getting an extra memory access?

Code:

ptr[ (row * COL) + col ] = x;

1 pointer accessed. Two if you count the fact that he was wanting to use a pointer to an array, but even with a pointer to an "actual array", you're still ending up with two.


Quzah.