How I can Make this code GPU parallel compatible

HI,

I have this code that i want to make it gpu compatiable, I wonder if anyone can tell what should i do , changing the A to gpu arry didnt help . I only want to make parfor loop to run on gpu .

A=permn([0:1:6],3)
N=4
D={}
parfor ii=1:size(A,1);
    D{1,ii}=[]
    for jj=1:size(A,1);
        for kk=1:size(A,1);
      C=DirectSumL(A(ii,:),A(jj,:),A(kk,:)) ;
      if sum(C)==N
       D{1,ii}=[D{1,ii} ; C];
      else 
          D{1,ii}=D{1,ii};
      end
        end
    end
       ii 
    end

[Walter Roberson]

What is DirectSumL ?

Which permn are you using? There are multiple permn in the File Exchange.

Answer 1

[shahabeddin Mostafanazhad aslmarand]

lets just imagin A is a n by m matrix,

the direct sum is this

function A=DirectSumL(A,varargin)
% DIRECTSUM computes the direct sum of matrices
%   A=DirectSum(A1,A2,...,An) computes the direct sum of matrices of
%   arbitrary size (Peter Arbenz, May 30, 1997)

for k=1:length(varargin)
  [n,m]=size(A);
  [o,p]=size(varargin{k});
  A=[A , varargin{k}];
end

[Walter Roberson]

That appears to be a corrupted definition of Peter Arbenz's DirectSum, which is shown at https://www.unige.ch/~gander/Book/Eigen/Matlab/DirectSum.m and which appears to be equivalent to

blkdiag(A, varargin{:})

The version you show pastes rows, and would be equivalent of

horzcat(A, varargin{:})

The result either way is going to be a 2D matrix, and sum() of that would be a vector. Your if would be comparing that vector to the scalar N. Is it correct that you only want the if to succeed in the case where all of the columns each add up to N? If so then that can be made more efficient, as you could eliminate from consideration any column of A that did not add up to N.