Cuda Malloc Pitch Doubt on cudaMallocPitch()

Hello All,

Suppose I allocate memory for a 2D array using cudaMallocPitch(). Now say I want to access individual rows of the array. How do I use pitch to correctly access only a part of the array.

Is Array[j*pitch] correct. if j is a variable that has the value 0,1,2,3…

Suppose pitch is 32, In that case the first row is at Array[0], Array[1]… so on
the 2nd row start at Array[32], Array[33]… so on

Am I right?

Thanks in advance.



cudaMallocPitch() returns the pitch in bytes, as it has no notion of the size of the elements in the array. Depending on the size of each element the pitch may not even be an integral multiple of the element size, e.g. if each element is a struct of three ints. In the case where the element size in bytes is a power of two, you can trivially convert byte-pitch to element-pitch and use that in your 2D addressing:

int pitchInElems = (int)(pitch / sizeof(A[0]));

// access row-major A(i,j)

... = A[pitchInElems * i + j];

[Later:] Actually, the documentation for cudaMallocPicth() already shows how to address an element at (row,column) inside a 2D matrix stored in row-major mode and composed of elements of arbitrary type:

T* pElement = (T*)((char*)BaseAddress + Row * pitch) + Column;