MATLAB: Vectorised a nested conditional function for gpuarray & arrayfun

gpuoopvectorization

Hi, I have a class which has many arrays and a nested functions.
The goal is I want to make all the arrays as gpuarray and use arrayfun, to be run in GPU, thus I'd like to have them well vectorised. Any thought? Thanks
%%the class
classdef egclass < handle
      properties
          cond = [1 1 1 0 0]; 
          x = [10 20 30 40 50];
          y = [0 0 0 0 0];
          out= [0 0 0 0 0]; 
      end
  end
%%the function 
function doOperation(eg)
      [nR,nC] = size(eg.cond);
      %%need to be vectorised!!!
      for i = 1:nC
          if eg.cond(i) == 0
              doTimes(eg,i);
          end
          if eg.cond(i) == 1  
              doAdd(eg,i);
          end
      end
  end
function doTimes(eg,i)
    eg.out(i) = eg.x(i) * eg.y(i);
end
function doAdd(eg,i)
    eg.out(i) = eg.x(i) + eg.y(i);
end
%%debug.m
clear all; clc;
eg = egclass;
doOperation(eg);

Best Answer

Hi Sendy,
If you're trying to use arrayfun to pass the arrays to the GPU inside the class, this is no problem! Just use this as your member function:
function doOperation(eg)
  gpu_x = gpuArray( eg.x );
  gpu_y = gpuArray( eg.y );
  gpu_cond = gpuArray( eg.cond );
  eg.out = arrayfun( @doAddOrTimes, gpu_cond , gpu_x, gpu_x );
end
where the "doAddOrTimes" function will incorporate the "if" statement. Of course, this function, and any others used on the GPU, will have to be an external function with no class attributes. As far as I'm aware, manipulating classes on the GPU is not possible.
Dan
Related Question