dynamic multi dimensional array alloc on gpu

Question

I can alloc memory (cudaMalloc) on GPU for static multi dimensional arrays, for the ones declared as follows int b[size1][size2][size3][size4][size5]...;. How is it possible to alloc memory (cudaMalloc) for dynamic array on GPU, for example int ***a; (we can also assume higher dimensions), where a has all the sizes distinct? Assuming that a has been allocated of its dimension's sizes on CPU side. Simple example would be appreciated, thanks!

cudaMalloc does not allocate static arrays (it always and exclusively allocates dynamic memory). Don't confuse static objects with statically-known array sizes. — Kerrek SB
– Kerrek SB, Commented Jul 20, 2012 at 22:37

Kerrek SB · Accepted Answer · 2012-07-20 22:35:44Z

1

Use cudaMalloc to allocate memory dynamically. For high-dimensional arrays, just compute the total, flattened size and access the array in strides:

void * p;
cudaError_t e = cudaMalloc(&p, dim1 * dim2 * dim3 /* ... */);
if (e != cudaSuccess) { /* error! */ }

// Access
int * arr = p;
arr[i1 * dim2 * dim3 + i2 * dim3 + i3] = 2; // etc., in strides

(For 2- or 3-dimensional arrays, you may also like to use cudaMalloc3DArray.)

There's also a corresponding host version, cudaMallocHost, which allocates page-locked host memory that's directly accessible by the device.

edited Jul 20, 2012 at 22:35

answered Jul 20, 2012 at 22:27

Kerrek SB

480k96 gold badges904 silver badges1.1k bronze badges

Sign up to request clarification or add additional context in comments.

2 Comments

starter Over a year ago

I can alloc single dimensional arrays, but what about multidimensional ones, for example "int ***v" or even higher "int ****v". thanks!

Kerrek SB Over a year ago

@starter: Compute the size yourself and access the array in strides. I added an example.

Collectives™ on Stack Overflow

dynamic multi dimensional array alloc on gpu

1 Answer 1

2 Comments

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

1 Answer 1

2 Comments

Your Answer

Sign up or log in

Post as a guest

Related