How can I use the GPU accelerated imdilate function for 3D logical Arrays?

At the moment I am developing a region growing algorithm in which I use the imdilate function to iterative expand the region of interest. From a performance test I noticed that I use the imdilate function numerous times during the process, which took 80% of my computational time. Therefore, I wanted to accelerate the region growing algorithm by using the graphic processing unit (GPU) for the whole process. However, the imdilate function does not support three-dimensional structuring elements for gpuArrays.

So the question is: How can I use the GPU accelerated imdilate function for 3D logical Arrays?

Example code:

A = rand(300,300,250);

B = A < 0.2;

D = imdilate(B, true(3,3,3)); % CPU method

B2 = gpuArray(B);

D2 = imdilate(B2, true(3,3,3)); % GPU method that does not work

D3 = imdilate(imdilate(imdilate(B2, true(3,1,1)), true(1,3,1)), true(1,1,3)); % second GPU method that does not work

Answer 1

[Simão Faria]

D2 = imdilate(B2, true(3,3,3,'gpuArray'));

Both the matrix and the kernel should be gpuArrays