Evaluating fucntion by GPU.

Hello, Is that possible to evaluate function written by myself on every element of array at once? Best regards, M. Pydych.

Answer 1

[Walter Roberson]

You can use https://www.mathworks.com/help/distcomp/parallel.gpu.cudakernel.html to invoke relatively arbitrary GPU code .

However, you cannot simply create your own function and call it on a gpu array object . gpuarray methods are only available to the particular functions Mathworks has specially coded for gpuarray: see https://www.mathworks.com/help/distcomp/run-built-in-functions-on-a-gpu.html and note that Mathworks had to enable each of those individually.

[Mateusz Pydych]

So there is no way that I can evaluate my function on multiple elements of array at once? I trought about GPU because computations using CPU are too time consuming,due to ammount of points on which this function have to be evaluated.

[Walter Roberson]

If your function involves only the calls listed in the link above, then it will (if I understand correctly) be compiled to work in parallel on the GPU. Otherwise, to get it to work on the GPU in parallel, you would have to rewrite your code as a .cu and compile it with the CUDA toolkit in order to be able to call it.

[Joss Knight]

Hmm. You can write any gpuArray function you like just by adding it to a directory called @gpuArray that you've put on the MATLAB path. But you only need to do that if you need it to do something special for gpuArrays. If not, you just write a function like normal and pass it a gpuArray.

Answer 2

[Joss Knight]

Your question is extremely confusing. Every function can act on every element of its inputs at once. So if I write a function

   function A = myFunc(A)
       A = A.^2;
   end

then I can pass it any array and it will act on the whole array

   A = gpuArray.rand(1000, 1);
   Asquared = myFunc(A);

In fact, the only way to write a function and get it to act on each element of the input independently is to use gpuArray/arrayfun.